Initial import

112 files changed, 93470 insertions, 0 deletions

diff --git a/fs/ocfs2/Kconfig b/fs/ocfs2/Kconfig
new file mode 100644
index 000000000..77a8de5f7
--- /dev/null
+++ b/fs/ocfs2/Kconfig
@@ -0,0 +1,76 @@
+config OCFS2_FS
+	tristate "OCFS2 file system support"
+	depends on NET && SYSFS && CONFIGFS_FS
+	select JBD2
+	select CRC32
+	select QUOTA
+	select QUOTA_TREE
+	select FS_POSIX_ACL
+	help
+	  OCFS2 is a general purpose extent based shared disk cluster file
+	  system with many similarities to ext3. It supports 64 bit inode
+	  numbers, and has automatically extending metadata groups which may
+	  also make it attractive for non-clustered use.
+
+	  You'll want to install the ocfs2-tools package in order to at least
+	  get "mount.ocfs2".
+
+	  Project web page:    http://oss.oracle.com/projects/ocfs2
+	  Tools web page:      http://oss.oracle.com/projects/ocfs2-tools
+	  OCFS2 mailing lists: http://oss.oracle.com/projects/ocfs2/mailman/
+
+	  For more information on OCFS2, see the file
+	  <file:Documentation/filesystems/ocfs2.txt>.
+
+config OCFS2_FS_O2CB
+	tristate "O2CB Kernelspace Clustering"
+	depends on OCFS2_FS
+	default y
+	help
+	  OCFS2 includes a simple kernelspace clustering package, the OCFS2
+	  Cluster Base.  It only requires a very small userspace component
+	  to configure it. This comes with the standard ocfs2-tools package.
+	  O2CB is limited to maintaining a cluster for OCFS2 file systems.
+	  It cannot manage any other cluster applications.
+
+	  It is always safe to say Y here, as the clustering method is
+	  run-time selectable.
+
+config OCFS2_FS_USERSPACE_CLUSTER
+	tristate "OCFS2 Userspace Clustering"
+	depends on OCFS2_FS && DLM
+	default y
+	help
+	  This option will allow OCFS2 to use userspace clustering services
+	  in conjunction with the DLM in fs/dlm.  If you are using a
+	  userspace cluster manager, say Y here.
+
+	  It is safe to say Y, as the clustering method is run-time
+	  selectable.
+
+config OCFS2_FS_STATS
+	bool "OCFS2 statistics"
+	depends on OCFS2_FS && DEBUG_FS
+	default y
+	help
+	  This option allows some fs statistics to be captured. Enabling
+	  this option may increase the memory consumption.
+
+config OCFS2_DEBUG_MASKLOG
+	bool "OCFS2 logging support"
+	depends on OCFS2_FS
+	default y
+	help
+	  The ocfs2 filesystem has an extensive logging system.  The system
+	  allows selection of events to log via files in /sys/o2cb/logmask/.
+	  This option will enlarge your kernel, but it allows debugging of
+	  ocfs2 filesystem issues.
+
+config OCFS2_DEBUG_FS
+	bool "OCFS2 expensive checks"
+	depends on OCFS2_FS
+	default n
+	help
+	  This option will enable expensive consistency checks. Enable
+	  this option for debugging only as it is likely to decrease
+	  performance of the filesystem.
diff --git a/fs/ocfs2/Makefile b/fs/ocfs2/Makefile
new file mode 100644
index 000000000..ce210d495
--- /dev/null
+++ b/fs/ocfs2/Makefile
@@ -0,0 +1,53 @@
+ccflags-y := -Ifs/ocfs2
+
+ccflags-y += -DCATCH_BH_JBD_RACES
+
+obj-$(CONFIG_OCFS2_FS) += 	\
+	ocfs2.o			\
+	ocfs2_stackglue.o
+
+obj-$(CONFIG_OCFS2_FS_O2CB) += ocfs2_stack_o2cb.o
+obj-$(CONFIG_OCFS2_FS_USERSPACE_CLUSTER) += ocfs2_stack_user.o
+
+ocfs2-objs := \
+	alloc.o 		\
+	aops.o 			\
+	blockcheck.o		\
+	buffer_head_io.o	\
+	dcache.o 		\
+	dir.o 			\
+	dlmglue.o 		\
+	export.o 		\
+	extent_map.o 		\
+	file.o 			\
+	heartbeat.o 		\
+	inode.o 		\
+	ioctl.o 		\
+	journal.o 		\
+	localalloc.o 		\
+	locks.o			\
+	mmap.o 			\
+	namei.o 		\
+	refcounttree.o		\
+	reservations.o		\
+	move_extents.o		\
+	resize.o		\
+	slot_map.o 		\
+	suballoc.o 		\
+	super.o 		\
+	symlink.o 		\
+	sysfile.o 		\
+	uptodate.o		\
+	quota_local.o		\
+	quota_global.o		\
+	xattr.o			\
+	acl.o
+
+ocfs2_stackglue-objs := stackglue.o
+ocfs2_stack_o2cb-objs := stack_o2cb.o
+ocfs2_stack_user-objs := stack_user.o
+
+obj-$(CONFIG_OCFS2_FS) += dlmfs/
+# cluster/ is always needed when OCFS2_FS for masklog support
+obj-$(CONFIG_OCFS2_FS) += cluster/
+obj-$(CONFIG_OCFS2_FS_O2CB) += dlm/
diff --git a/fs/ocfs2/acl.c b/fs/ocfs2/acl.c
new file mode 100644
index 000000000..c58a1bcfd
--- /dev/null
+++ b/fs/ocfs2/acl.c
@@ -0,0 +1,310 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * acl.c
+ *
+ * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
+ *
+ * CREDITS:
+ * Lots of code in this file is copy from linux/fs/ext3/acl.c.
+ * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/string.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+#include "alloc.h"
+#include "dlmglue.h"
+#include "file.h"
+#include "inode.h"
+#include "journal.h"
+#include "ocfs2_fs.h"
+
+#include "xattr.h"
+#include "acl.h"
+
+/*
+ * Convert from xattr value to acl struct.
+ */
+static struct posix_acl *ocfs2_acl_from_xattr(const void *value, size_t size)
+{
+	int n, count;
+	struct posix_acl *acl;
+
+	if (!value)
+		return NULL;
+	if (size < sizeof(struct posix_acl_entry))
+		return ERR_PTR(-EINVAL);
+
+	count = size / sizeof(struct posix_acl_entry);
+
+	acl = posix_acl_alloc(count, GFP_NOFS);
+	if (!acl)
+		return ERR_PTR(-ENOMEM);
+	for (n = 0; n < count; n++) {
+		struct ocfs2_acl_entry *entry =
+			(struct ocfs2_acl_entry *)value;
+
+		acl->a_entries[n].e_tag  = le16_to_cpu(entry->e_tag);
+		acl->a_entries[n].e_perm = le16_to_cpu(entry->e_perm);
+		switch(acl->a_entries[n].e_tag) {
+		case ACL_USER:
+			acl->a_entries[n].e_uid =
+				make_kuid(&init_user_ns,
+					  le32_to_cpu(entry->e_id));
+			break;
+		case ACL_GROUP:
+			acl->a_entries[n].e_gid =
+				make_kgid(&init_user_ns,
+					  le32_to_cpu(entry->e_id));
+			break;
+		default:
+			break;
+		}
+		value += sizeof(struct posix_acl_entry);
+
+	}
+	return acl;
+}
+
+/*
+ * Convert acl struct to xattr value.
+ */
+static void *ocfs2_acl_to_xattr(const struct posix_acl *acl, size_t *size)
+{
+	struct ocfs2_acl_entry *entry = NULL;
+	char *ocfs2_acl;
+	size_t n;
+
+	*size = acl->a_count * sizeof(struct posix_acl_entry);
+
+	ocfs2_acl = kmalloc(*size, GFP_NOFS);
+	if (!ocfs2_acl)
+		return ERR_PTR(-ENOMEM);
+
+	entry = (struct ocfs2_acl_entry *)ocfs2_acl;
+	for (n = 0; n < acl->a_count; n++, entry++) {
+		entry->e_tag  = cpu_to_le16(acl->a_entries[n].e_tag);
+		entry->e_perm = cpu_to_le16(acl->a_entries[n].e_perm);
+		switch(acl->a_entries[n].e_tag) {
+		case ACL_USER:
+			entry->e_id = cpu_to_le32(
+				from_kuid(&init_user_ns,
+					  acl->a_entries[n].e_uid));
+			break;
+		case ACL_GROUP:
+			entry->e_id = cpu_to_le32(
+				from_kgid(&init_user_ns,
+					  acl->a_entries[n].e_gid));
+			break;
+		default:
+			entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
+			break;
+		}
+	}
+	return ocfs2_acl;
+}
+
+static struct posix_acl *ocfs2_get_acl_nolock(struct inode *inode,
+					      int type,
+					      struct buffer_head *di_bh)
+{
+	int name_index;
+	char *value = NULL;
+	struct posix_acl *acl;
+	int retval;
+
+	switch (type) {
+	case ACL_TYPE_ACCESS:
+		name_index = OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS;
+		break;
+	case ACL_TYPE_DEFAULT:
+		name_index = OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
+		break;
+	default:
+		return ERR_PTR(-EINVAL);
+	}
+
+	retval = ocfs2_xattr_get_nolock(inode, di_bh, name_index, "", NULL, 0);
+	if (retval > 0) {
+		value = kmalloc(retval, GFP_NOFS);
+		if (!value)
+			return ERR_PTR(-ENOMEM);
+		retval = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
+						"", value, retval);
+	}
+
+	if (retval > 0)
+		acl = ocfs2_acl_from_xattr(value, retval);
+	else if (retval == -ENODATA || retval == 0)
+		acl = NULL;
+	else
+		acl = ERR_PTR(retval);
+
+	kfree(value);
+
+	return acl;
+}
+
+/*
+ * Helper function to set i_mode in memory and disk. Some call paths
+ * will not have di_bh or a journal handle to pass, in which case it
+ * will create it's own.
+ */
+static int ocfs2_acl_set_mode(struct inode *inode, struct buffer_head *di_bh,
+			      handle_t *handle, umode_t new_mode)
+{
+	int ret, commit_handle = 0;
+	struct ocfs2_dinode *di;
+
+	if (di_bh == NULL) {
+		ret = ocfs2_read_inode_block(inode, &di_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	} else
+		get_bh(di_bh);
+
+	if (handle == NULL) {
+		handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
+					   OCFS2_INODE_UPDATE_CREDITS);
+		if (IS_ERR(handle)) {
+			ret = PTR_ERR(handle);
+			mlog_errno(ret);
+			goto out_brelse;
+		}
+
+		commit_handle = 1;
+	}
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	inode->i_mode = new_mode;
+	inode->i_ctime = CURRENT_TIME;
+	di->i_mode = cpu_to_le16(inode->i_mode);
+	di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
+	di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+	ocfs2_update_inode_fsync_trans(handle, inode, 0);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+out_commit:
+	if (commit_handle)
+		ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
+out_brelse:
+	brelse(di_bh);
+out:
+	return ret;
+}
+
+/*
+ * Set the access or default ACL of an inode.
+ */
+int ocfs2_set_acl(handle_t *handle,
+			 struct inode *inode,
+			 struct buffer_head *di_bh,
+			 int type,
+			 struct posix_acl *acl,
+			 struct ocfs2_alloc_context *meta_ac,
+			 struct ocfs2_alloc_context *data_ac)
+{
+	int name_index;
+	void *value = NULL;
+	size_t size = 0;
+	int ret;
+
+	if (S_ISLNK(inode->i_mode))
+		return -EOPNOTSUPP;
+
+	switch (type) {
+	case ACL_TYPE_ACCESS:
+		name_index = OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS;
+		if (acl) {
+			umode_t mode = inode->i_mode;
+			ret = posix_acl_equiv_mode(acl, &mode);
+			if (ret < 0)
+				return ret;
+
+			if (ret == 0)
+				acl = NULL;
+
+			ret = ocfs2_acl_set_mode(inode, di_bh,
+						 handle, mode);
+			if (ret)
+				return ret;
+		}
+		break;
+	case ACL_TYPE_DEFAULT:
+		name_index = OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
+		if (!S_ISDIR(inode->i_mode))
+			return acl ? -EACCES : 0;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (acl) {
+		value = ocfs2_acl_to_xattr(acl, &size);
+		if (IS_ERR(value))
+			return (int)PTR_ERR(value);
+	}
+
+	if (handle)
+		ret = ocfs2_xattr_set_handle(handle, inode, di_bh, name_index,
+					     "", value, size, 0,
+					     meta_ac, data_ac);
+	else
+		ret = ocfs2_xattr_set(inode, name_index, "", value, size, 0);
+
+	kfree(value);
+
+	return ret;
+}
+
+int ocfs2_iop_set_acl(struct inode *inode, struct posix_acl *acl, int type)
+{
+	return ocfs2_set_acl(NULL, inode, NULL, type, acl, NULL, NULL);
+}
+
+struct posix_acl *ocfs2_iop_get_acl(struct inode *inode, int type)
+{
+	struct ocfs2_super *osb;
+	struct buffer_head *di_bh = NULL;
+	struct posix_acl *acl;
+	int ret = -EAGAIN;
+
+	osb = OCFS2_SB(inode->i_sb);
+	if (!(osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL))
+		return NULL;
+
+	ret = ocfs2_read_inode_block(inode, &di_bh);
+	if (ret < 0)
+		return ERR_PTR(ret);
+
+	acl = ocfs2_get_acl_nolock(inode, type, di_bh);
+
+	brelse(di_bh);
+
+	return acl;
+}
diff --git a/fs/ocfs2/acl.h b/fs/ocfs2/acl.h
new file mode 100644
index 000000000..3fce68d08
--- /dev/null
+++ b/fs/ocfs2/acl.h
@@ -0,0 +1,39 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * acl.h
+ *
+ * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#ifndef OCFS2_ACL_H
+#define OCFS2_ACL_H
+
+#include <linux/posix_acl_xattr.h>
+
+struct ocfs2_acl_entry {
+	__le16 e_tag;
+	__le16 e_perm;
+	__le32 e_id;
+};
+
+struct posix_acl *ocfs2_iop_get_acl(struct inode *inode, int type);
+int ocfs2_iop_set_acl(struct inode *inode, struct posix_acl *acl, int type);
+int ocfs2_set_acl(handle_t *handle,
+			 struct inode *inode,
+			 struct buffer_head *di_bh,
+			 int type,
+			 struct posix_acl *acl,
+			 struct ocfs2_alloc_context *meta_ac,
+			 struct ocfs2_alloc_context *data_ac);
+
+#endif /* OCFS2_ACL_H */
diff --git a/fs/ocfs2/alloc.c b/fs/ocfs2/alloc.c
new file mode 100644
index 000000000..2d7f76e52
--- /dev/null
+++ b/fs/ocfs2/alloc.c
@@ -0,0 +1,7405 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * alloc.c
+ *
+ * Extent allocs and frees
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/swap.h>
+#include <linux/quotaops.h>
+#include <linux/blkdev.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "aops.h"
+#include "blockcheck.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "inode.h"
+#include "journal.h"
+#include "localalloc.h"
+#include "suballoc.h"
+#include "sysfile.h"
+#include "file.h"
+#include "super.h"
+#include "uptodate.h"
+#include "xattr.h"
+#include "refcounttree.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+enum ocfs2_contig_type {
+	CONTIG_NONE = 0,
+	CONTIG_LEFT,
+	CONTIG_RIGHT,
+	CONTIG_LEFTRIGHT,
+};
+
+static enum ocfs2_contig_type
+	ocfs2_extent_rec_contig(struct super_block *sb,
+				struct ocfs2_extent_rec *ext,
+				struct ocfs2_extent_rec *insert_rec);
+/*
+ * Operations for a specific extent tree type.
+ *
+ * To implement an on-disk btree (extent tree) type in ocfs2, add
+ * an ocfs2_extent_tree_operations structure and the matching
+ * ocfs2_init_<thingy>_extent_tree() function.  That's pretty much it
+ * for the allocation portion of the extent tree.
+ */
+struct ocfs2_extent_tree_operations {
+	/*
+	 * last_eb_blk is the block number of the right most leaf extent
+	 * block.  Most on-disk structures containing an extent tree store
+	 * this value for fast access.  The ->eo_set_last_eb_blk() and
+	 * ->eo_get_last_eb_blk() operations access this value.  They are
+	 *  both required.
+	 */
+	void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et,
+				   u64 blkno);
+	u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et);
+
+	/*
+	 * The on-disk structure usually keeps track of how many total
+	 * clusters are stored in this extent tree.  This function updates
+	 * that value.  new_clusters is the delta, and must be
+	 * added to the total.  Required.
+	 */
+	void (*eo_update_clusters)(struct ocfs2_extent_tree *et,
+				   u32 new_clusters);
+
+	/*
+	 * If this extent tree is supported by an extent map, insert
+	 * a record into the map.
+	 */
+	void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et,
+				     struct ocfs2_extent_rec *rec);
+
+	/*
+	 * If this extent tree is supported by an extent map, truncate the
+	 * map to clusters,
+	 */
+	void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et,
+				       u32 clusters);
+
+	/*
+	 * If ->eo_insert_check() exists, it is called before rec is
+	 * inserted into the extent tree.  It is optional.
+	 */
+	int (*eo_insert_check)(struct ocfs2_extent_tree *et,
+			       struct ocfs2_extent_rec *rec);
+	int (*eo_sanity_check)(struct ocfs2_extent_tree *et);
+
+	/*
+	 * --------------------------------------------------------------
+	 * The remaining are internal to ocfs2_extent_tree and don't have
+	 * accessor functions
+	 */
+
+	/*
+	 * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el.
+	 * It is required.
+	 */
+	void (*eo_fill_root_el)(struct ocfs2_extent_tree *et);
+
+	/*
+	 * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if
+	 * it exists.  If it does not, et->et_max_leaf_clusters is set
+	 * to 0 (unlimited).  Optional.
+	 */
+	void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et);
+
+	/*
+	 * ->eo_extent_contig test whether the 2 ocfs2_extent_rec
+	 * are contiguous or not. Optional. Don't need to set it if use
+	 * ocfs2_extent_rec as the tree leaf.
+	 */
+	enum ocfs2_contig_type
+		(*eo_extent_contig)(struct ocfs2_extent_tree *et,
+				    struct ocfs2_extent_rec *ext,
+				    struct ocfs2_extent_rec *insert_rec);
+};
+
+
+/*
+ * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check
+ * in the methods.
+ */
+static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et);
+static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
+					 u64 blkno);
+static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
+					 u32 clusters);
+static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
+					   struct ocfs2_extent_rec *rec);
+static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
+					     u32 clusters);
+static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
+				     struct ocfs2_extent_rec *rec);
+static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et);
+static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et);
+static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = {
+	.eo_set_last_eb_blk	= ocfs2_dinode_set_last_eb_blk,
+	.eo_get_last_eb_blk	= ocfs2_dinode_get_last_eb_blk,
+	.eo_update_clusters	= ocfs2_dinode_update_clusters,
+	.eo_extent_map_insert	= ocfs2_dinode_extent_map_insert,
+	.eo_extent_map_truncate	= ocfs2_dinode_extent_map_truncate,
+	.eo_insert_check	= ocfs2_dinode_insert_check,
+	.eo_sanity_check	= ocfs2_dinode_sanity_check,
+	.eo_fill_root_el	= ocfs2_dinode_fill_root_el,
+};
+
+static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
+					 u64 blkno)
+{
+	struct ocfs2_dinode *di = et->et_object;
+
+	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
+	di->i_last_eb_blk = cpu_to_le64(blkno);
+}
+
+static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_dinode *di = et->et_object;
+
+	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
+	return le64_to_cpu(di->i_last_eb_blk);
+}
+
+static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
+					 u32 clusters)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
+	struct ocfs2_dinode *di = et->et_object;
+
+	le32_add_cpu(&di->i_clusters, clusters);
+	spin_lock(&oi->ip_lock);
+	oi->ip_clusters = le32_to_cpu(di->i_clusters);
+	spin_unlock(&oi->ip_lock);
+}
+
+static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
+					   struct ocfs2_extent_rec *rec)
+{
+	struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
+
+	ocfs2_extent_map_insert_rec(inode, rec);
+}
+
+static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
+					     u32 clusters)
+{
+	struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
+
+	ocfs2_extent_map_trunc(inode, clusters);
+}
+
+static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
+				     struct ocfs2_extent_rec *rec)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
+	struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb);
+
+	BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL);
+	mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
+			(oi->ip_clusters != le32_to_cpu(rec->e_cpos)),
+			"Device %s, asking for sparse allocation: inode %llu, "
+			"cpos %u, clusters %u\n",
+			osb->dev_str,
+			(unsigned long long)oi->ip_blkno,
+			rec->e_cpos, oi->ip_clusters);
+
+	return 0;
+}
+
+static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_dinode *di = et->et_object;
+
+	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
+	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
+
+	return 0;
+}
+
+static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_dinode *di = et->et_object;
+
+	et->et_root_el = &di->id2.i_list;
+}
+
+
+static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_xattr_value_buf *vb = et->et_object;
+
+	et->et_root_el = &vb->vb_xv->xr_list;
+}
+
+static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et,
+					      u64 blkno)
+{
+	struct ocfs2_xattr_value_buf *vb = et->et_object;
+
+	vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno);
+}
+
+static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_xattr_value_buf *vb = et->et_object;
+
+	return le64_to_cpu(vb->vb_xv->xr_last_eb_blk);
+}
+
+static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et,
+					      u32 clusters)
+{
+	struct ocfs2_xattr_value_buf *vb = et->et_object;
+
+	le32_add_cpu(&vb->vb_xv->xr_clusters, clusters);
+}
+
+static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = {
+	.eo_set_last_eb_blk	= ocfs2_xattr_value_set_last_eb_blk,
+	.eo_get_last_eb_blk	= ocfs2_xattr_value_get_last_eb_blk,
+	.eo_update_clusters	= ocfs2_xattr_value_update_clusters,
+	.eo_fill_root_el	= ocfs2_xattr_value_fill_root_el,
+};
+
+static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_xattr_block *xb = et->et_object;
+
+	et->et_root_el = &xb->xb_attrs.xb_root.xt_list;
+}
+
+static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et)
+{
+	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
+	et->et_max_leaf_clusters =
+		ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE);
+}
+
+static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
+					     u64 blkno)
+{
+	struct ocfs2_xattr_block *xb = et->et_object;
+	struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
+
+	xt->xt_last_eb_blk = cpu_to_le64(blkno);
+}
+
+static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_xattr_block *xb = et->et_object;
+	struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
+
+	return le64_to_cpu(xt->xt_last_eb_blk);
+}
+
+static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et,
+					     u32 clusters)
+{
+	struct ocfs2_xattr_block *xb = et->et_object;
+
+	le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters);
+}
+
+static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = {
+	.eo_set_last_eb_blk	= ocfs2_xattr_tree_set_last_eb_blk,
+	.eo_get_last_eb_blk	= ocfs2_xattr_tree_get_last_eb_blk,
+	.eo_update_clusters	= ocfs2_xattr_tree_update_clusters,
+	.eo_fill_root_el	= ocfs2_xattr_tree_fill_root_el,
+	.eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters,
+};
+
+static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et,
+					  u64 blkno)
+{
+	struct ocfs2_dx_root_block *dx_root = et->et_object;
+
+	dx_root->dr_last_eb_blk = cpu_to_le64(blkno);
+}
+
+static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_dx_root_block *dx_root = et->et_object;
+
+	return le64_to_cpu(dx_root->dr_last_eb_blk);
+}
+
+static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et,
+					  u32 clusters)
+{
+	struct ocfs2_dx_root_block *dx_root = et->et_object;
+
+	le32_add_cpu(&dx_root->dr_clusters, clusters);
+}
+
+static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_dx_root_block *dx_root = et->et_object;
+
+	BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root));
+
+	return 0;
+}
+
+static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_dx_root_block *dx_root = et->et_object;
+
+	et->et_root_el = &dx_root->dr_list;
+}
+
+static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = {
+	.eo_set_last_eb_blk	= ocfs2_dx_root_set_last_eb_blk,
+	.eo_get_last_eb_blk	= ocfs2_dx_root_get_last_eb_blk,
+	.eo_update_clusters	= ocfs2_dx_root_update_clusters,
+	.eo_sanity_check	= ocfs2_dx_root_sanity_check,
+	.eo_fill_root_el	= ocfs2_dx_root_fill_root_el,
+};
+
+static void ocfs2_refcount_tree_fill_root_el(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_refcount_block *rb = et->et_object;
+
+	et->et_root_el = &rb->rf_list;
+}
+
+static void ocfs2_refcount_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
+						u64 blkno)
+{
+	struct ocfs2_refcount_block *rb = et->et_object;
+
+	rb->rf_last_eb_blk = cpu_to_le64(blkno);
+}
+
+static u64 ocfs2_refcount_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
+{
+	struct ocfs2_refcount_block *rb = et->et_object;
+
+	return le64_to_cpu(rb->rf_last_eb_blk);
+}
+
+static void ocfs2_refcount_tree_update_clusters(struct ocfs2_extent_tree *et,
+						u32 clusters)
+{
+	struct ocfs2_refcount_block *rb = et->et_object;
+
+	le32_add_cpu(&rb->rf_clusters, clusters);
+}
+
+static enum ocfs2_contig_type
+ocfs2_refcount_tree_extent_contig(struct ocfs2_extent_tree *et,
+				  struct ocfs2_extent_rec *ext,
+				  struct ocfs2_extent_rec *insert_rec)
+{
+	return CONTIG_NONE;
+}
+
+static struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = {
+	.eo_set_last_eb_blk	= ocfs2_refcount_tree_set_last_eb_blk,
+	.eo_get_last_eb_blk	= ocfs2_refcount_tree_get_last_eb_blk,
+	.eo_update_clusters	= ocfs2_refcount_tree_update_clusters,
+	.eo_fill_root_el	= ocfs2_refcount_tree_fill_root_el,
+	.eo_extent_contig	= ocfs2_refcount_tree_extent_contig,
+};
+
+static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *bh,
+				     ocfs2_journal_access_func access,
+				     void *obj,
+				     struct ocfs2_extent_tree_operations *ops)
+{
+	et->et_ops = ops;
+	et->et_root_bh = bh;
+	et->et_ci = ci;
+	et->et_root_journal_access = access;
+	if (!obj)
+		obj = (void *)bh->b_data;
+	et->et_object = obj;
+
+	et->et_ops->eo_fill_root_el(et);
+	if (!et->et_ops->eo_fill_max_leaf_clusters)
+		et->et_max_leaf_clusters = 0;
+	else
+		et->et_ops->eo_fill_max_leaf_clusters(et);
+}
+
+void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
+				   struct ocfs2_caching_info *ci,
+				   struct buffer_head *bh)
+{
+	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_di,
+				 NULL, &ocfs2_dinode_et_ops);
+}
+
+void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
+				       struct ocfs2_caching_info *ci,
+				       struct buffer_head *bh)
+{
+	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_xb,
+				 NULL, &ocfs2_xattr_tree_et_ops);
+}
+
+void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
+					struct ocfs2_caching_info *ci,
+					struct ocfs2_xattr_value_buf *vb)
+{
+	__ocfs2_init_extent_tree(et, ci, vb->vb_bh, vb->vb_access, vb,
+				 &ocfs2_xattr_value_et_ops);
+}
+
+void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
+				    struct ocfs2_caching_info *ci,
+				    struct buffer_head *bh)
+{
+	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_dr,
+				 NULL, &ocfs2_dx_root_et_ops);
+}
+
+void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *bh)
+{
+	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_rb,
+				 NULL, &ocfs2_refcount_tree_et_ops);
+}
+
+static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et,
+					    u64 new_last_eb_blk)
+{
+	et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk);
+}
+
+static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et)
+{
+	return et->et_ops->eo_get_last_eb_blk(et);
+}
+
+static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et,
+					    u32 clusters)
+{
+	et->et_ops->eo_update_clusters(et, clusters);
+}
+
+static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et,
+					      struct ocfs2_extent_rec *rec)
+{
+	if (et->et_ops->eo_extent_map_insert)
+		et->et_ops->eo_extent_map_insert(et, rec);
+}
+
+static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et,
+						u32 clusters)
+{
+	if (et->et_ops->eo_extent_map_truncate)
+		et->et_ops->eo_extent_map_truncate(et, clusters);
+}
+
+static inline int ocfs2_et_root_journal_access(handle_t *handle,
+					       struct ocfs2_extent_tree *et,
+					       int type)
+{
+	return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh,
+					  type);
+}
+
+static inline enum ocfs2_contig_type
+	ocfs2_et_extent_contig(struct ocfs2_extent_tree *et,
+			       struct ocfs2_extent_rec *rec,
+			       struct ocfs2_extent_rec *insert_rec)
+{
+	if (et->et_ops->eo_extent_contig)
+		return et->et_ops->eo_extent_contig(et, rec, insert_rec);
+
+	return ocfs2_extent_rec_contig(
+				ocfs2_metadata_cache_get_super(et->et_ci),
+				rec, insert_rec);
+}
+
+static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et,
+					struct ocfs2_extent_rec *rec)
+{
+	int ret = 0;
+
+	if (et->et_ops->eo_insert_check)
+		ret = et->et_ops->eo_insert_check(et, rec);
+	return ret;
+}
+
+static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et)
+{
+	int ret = 0;
+
+	if (et->et_ops->eo_sanity_check)
+		ret = et->et_ops->eo_sanity_check(et);
+	return ret;
+}
+
+static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
+					 struct ocfs2_extent_block *eb);
+static void ocfs2_adjust_rightmost_records(handle_t *handle,
+					   struct ocfs2_extent_tree *et,
+					   struct ocfs2_path *path,
+					   struct ocfs2_extent_rec *insert_rec);
+/*
+ * Reset the actual path elements so that we can re-use the structure
+ * to build another path. Generally, this involves freeing the buffer
+ * heads.
+ */
+void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root)
+{
+	int i, start = 0, depth = 0;
+	struct ocfs2_path_item *node;
+
+	if (keep_root)
+		start = 1;
+
+	for(i = start; i < path_num_items(path); i++) {
+		node = &path->p_node[i];
+
+		brelse(node->bh);
+		node->bh = NULL;
+		node->el = NULL;
+	}
+
+	/*
+	 * Tree depth may change during truncate, or insert. If we're
+	 * keeping the root extent list, then make sure that our path
+	 * structure reflects the proper depth.
+	 */
+	if (keep_root)
+		depth = le16_to_cpu(path_root_el(path)->l_tree_depth);
+	else
+		path_root_access(path) = NULL;
+
+	path->p_tree_depth = depth;
+}
+
+void ocfs2_free_path(struct ocfs2_path *path)
+{
+	if (path) {
+		ocfs2_reinit_path(path, 0);
+		kfree(path);
+	}
+}
+
+/*
+ * All the elements of src into dest. After this call, src could be freed
+ * without affecting dest.
+ *
+ * Both paths should have the same root. Any non-root elements of dest
+ * will be freed.
+ */
+static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src)
+{
+	int i;
+
+	BUG_ON(path_root_bh(dest) != path_root_bh(src));
+	BUG_ON(path_root_el(dest) != path_root_el(src));
+	BUG_ON(path_root_access(dest) != path_root_access(src));
+
+	ocfs2_reinit_path(dest, 1);
+
+	for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
+		dest->p_node[i].bh = src->p_node[i].bh;
+		dest->p_node[i].el = src->p_node[i].el;
+
+		if (dest->p_node[i].bh)
+			get_bh(dest->p_node[i].bh);
+	}
+}
+
+/*
+ * Make the *dest path the same as src and re-initialize src path to
+ * have a root only.
+ */
+static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src)
+{
+	int i;
+
+	BUG_ON(path_root_bh(dest) != path_root_bh(src));
+	BUG_ON(path_root_access(dest) != path_root_access(src));
+
+	for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
+		brelse(dest->p_node[i].bh);
+
+		dest->p_node[i].bh = src->p_node[i].bh;
+		dest->p_node[i].el = src->p_node[i].el;
+
+		src->p_node[i].bh = NULL;
+		src->p_node[i].el = NULL;
+	}
+}
+
+/*
+ * Insert an extent block at given index.
+ *
+ * This will not take an additional reference on eb_bh.
+ */
+static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index,
+					struct buffer_head *eb_bh)
+{
+	struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data;
+
+	/*
+	 * Right now, no root bh is an extent block, so this helps
+	 * catch code errors with dinode trees. The assertion can be
+	 * safely removed if we ever need to insert extent block
+	 * structures at the root.
+	 */
+	BUG_ON(index == 0);
+
+	path->p_node[index].bh = eb_bh;
+	path->p_node[index].el = &eb->h_list;
+}
+
+static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh,
+					 struct ocfs2_extent_list *root_el,
+					 ocfs2_journal_access_func access)
+{
+	struct ocfs2_path *path;
+
+	BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH);
+
+	path = kzalloc(sizeof(*path), GFP_NOFS);
+	if (path) {
+		path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth);
+		get_bh(root_bh);
+		path_root_bh(path) = root_bh;
+		path_root_el(path) = root_el;
+		path_root_access(path) = access;
+	}
+
+	return path;
+}
+
+struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path)
+{
+	return ocfs2_new_path(path_root_bh(path), path_root_el(path),
+			      path_root_access(path));
+}
+
+struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et)
+{
+	return ocfs2_new_path(et->et_root_bh, et->et_root_el,
+			      et->et_root_journal_access);
+}
+
+/*
+ * Journal the buffer at depth idx.  All idx>0 are extent_blocks,
+ * otherwise it's the root_access function.
+ *
+ * I don't like the way this function's name looks next to
+ * ocfs2_journal_access_path(), but I don't have a better one.
+ */
+int ocfs2_path_bh_journal_access(handle_t *handle,
+				 struct ocfs2_caching_info *ci,
+				 struct ocfs2_path *path,
+				 int idx)
+{
+	ocfs2_journal_access_func access = path_root_access(path);
+
+	if (!access)
+		access = ocfs2_journal_access;
+
+	if (idx)
+		access = ocfs2_journal_access_eb;
+
+	return access(handle, ci, path->p_node[idx].bh,
+		      OCFS2_JOURNAL_ACCESS_WRITE);
+}
+
+/*
+ * Convenience function to journal all components in a path.
+ */
+int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
+			      handle_t *handle,
+			      struct ocfs2_path *path)
+{
+	int i, ret = 0;
+
+	if (!path)
+		goto out;
+
+	for(i = 0; i < path_num_items(path); i++) {
+		ret = ocfs2_path_bh_journal_access(handle, ci, path, i);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+out:
+	return ret;
+}
+
+/*
+ * Return the index of the extent record which contains cluster #v_cluster.
+ * -1 is returned if it was not found.
+ *
+ * Should work fine on interior and exterior nodes.
+ */
+int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster)
+{
+	int ret = -1;
+	int i;
+	struct ocfs2_extent_rec *rec;
+	u32 rec_end, rec_start, clusters;
+
+	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		rec = &el->l_recs[i];
+
+		rec_start = le32_to_cpu(rec->e_cpos);
+		clusters = ocfs2_rec_clusters(el, rec);
+
+		rec_end = rec_start + clusters;
+
+		if (v_cluster >= rec_start && v_cluster < rec_end) {
+			ret = i;
+			break;
+		}
+	}
+
+	return ret;
+}
+
+/*
+ * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
+ * ocfs2_extent_rec_contig only work properly against leaf nodes!
+ */
+static int ocfs2_block_extent_contig(struct super_block *sb,
+				     struct ocfs2_extent_rec *ext,
+				     u64 blkno)
+{
+	u64 blk_end = le64_to_cpu(ext->e_blkno);
+
+	blk_end += ocfs2_clusters_to_blocks(sb,
+				    le16_to_cpu(ext->e_leaf_clusters));
+
+	return blkno == blk_end;
+}
+
+static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
+				  struct ocfs2_extent_rec *right)
+{
+	u32 left_range;
+
+	left_range = le32_to_cpu(left->e_cpos) +
+		le16_to_cpu(left->e_leaf_clusters);
+
+	return (left_range == le32_to_cpu(right->e_cpos));
+}
+
+static enum ocfs2_contig_type
+	ocfs2_extent_rec_contig(struct super_block *sb,
+				struct ocfs2_extent_rec *ext,
+				struct ocfs2_extent_rec *insert_rec)
+{
+	u64 blkno = le64_to_cpu(insert_rec->e_blkno);
+
+	/*
+	 * Refuse to coalesce extent records with different flag
+	 * fields - we don't want to mix unwritten extents with user
+	 * data.
+	 */
+	if (ext->e_flags != insert_rec->e_flags)
+		return CONTIG_NONE;
+
+	if (ocfs2_extents_adjacent(ext, insert_rec) &&
+	    ocfs2_block_extent_contig(sb, ext, blkno))
+			return CONTIG_RIGHT;
+
+	blkno = le64_to_cpu(ext->e_blkno);
+	if (ocfs2_extents_adjacent(insert_rec, ext) &&
+	    ocfs2_block_extent_contig(sb, insert_rec, blkno))
+		return CONTIG_LEFT;
+
+	return CONTIG_NONE;
+}
+
+/*
+ * NOTE: We can have pretty much any combination of contiguousness and
+ * appending.
+ *
+ * The usefulness of APPEND_TAIL is more in that it lets us know that
+ * we'll have to update the path to that leaf.
+ */
+enum ocfs2_append_type {
+	APPEND_NONE = 0,
+	APPEND_TAIL,
+};
+
+enum ocfs2_split_type {
+	SPLIT_NONE = 0,
+	SPLIT_LEFT,
+	SPLIT_RIGHT,
+};
+
+struct ocfs2_insert_type {
+	enum ocfs2_split_type	ins_split;
+	enum ocfs2_append_type	ins_appending;
+	enum ocfs2_contig_type	ins_contig;
+	int			ins_contig_index;
+	int			ins_tree_depth;
+};
+
+struct ocfs2_merge_ctxt {
+	enum ocfs2_contig_type	c_contig_type;
+	int			c_has_empty_extent;
+	int			c_split_covers_rec;
+};
+
+static int ocfs2_validate_extent_block(struct super_block *sb,
+				       struct buffer_head *bh)
+{
+	int rc;
+	struct ocfs2_extent_block *eb =
+		(struct ocfs2_extent_block *)bh->b_data;
+
+	trace_ocfs2_validate_extent_block((unsigned long long)bh->b_blocknr);
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	/*
+	 * If the ecc fails, we return the error but otherwise
+	 * leave the filesystem running.  We know any error is
+	 * local to this block.
+	 */
+	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check);
+	if (rc) {
+		mlog(ML_ERROR, "Checksum failed for extent block %llu\n",
+		     (unsigned long long)bh->b_blocknr);
+		return rc;
+	}
+
+	/*
+	 * Errors after here are fatal.
+	 */
+
+	if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
+		ocfs2_error(sb,
+			    "Extent block #%llu has bad signature %.*s",
+			    (unsigned long long)bh->b_blocknr, 7,
+			    eb->h_signature);
+		return -EINVAL;
+	}
+
+	if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) {
+		ocfs2_error(sb,
+			    "Extent block #%llu has an invalid h_blkno "
+			    "of %llu",
+			    (unsigned long long)bh->b_blocknr,
+			    (unsigned long long)le64_to_cpu(eb->h_blkno));
+		return -EINVAL;
+	}
+
+	if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) {
+		ocfs2_error(sb,
+			    "Extent block #%llu has an invalid "
+			    "h_fs_generation of #%u",
+			    (unsigned long long)bh->b_blocknr,
+			    le32_to_cpu(eb->h_fs_generation));
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
+			    struct buffer_head **bh)
+{
+	int rc;
+	struct buffer_head *tmp = *bh;
+
+	rc = ocfs2_read_block(ci, eb_blkno, &tmp,
+			      ocfs2_validate_extent_block);
+
+	/* If ocfs2_read_block() got us a new bh, pass it up. */
+	if (!rc && !*bh)
+		*bh = tmp;
+
+	return rc;
+}
+
+
+/*
+ * How many free extents have we got before we need more meta data?
+ */
+int ocfs2_num_free_extents(struct ocfs2_super *osb,
+			   struct ocfs2_extent_tree *et)
+{
+	int retval;
+	struct ocfs2_extent_list *el = NULL;
+	struct ocfs2_extent_block *eb;
+	struct buffer_head *eb_bh = NULL;
+	u64 last_eb_blk = 0;
+
+	el = et->et_root_el;
+	last_eb_blk = ocfs2_et_get_last_eb_blk(et);
+
+	if (last_eb_blk) {
+		retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk,
+						 &eb_bh);
+		if (retval < 0) {
+			mlog_errno(retval);
+			goto bail;
+		}
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+	}
+
+	BUG_ON(el->l_tree_depth != 0);
+
+	retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
+bail:
+	brelse(eb_bh);
+
+	trace_ocfs2_num_free_extents(retval);
+	return retval;
+}
+
+/* expects array to already be allocated
+ *
+ * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and
+ * l_count for you
+ */
+static int ocfs2_create_new_meta_bhs(handle_t *handle,
+				     struct ocfs2_extent_tree *et,
+				     int wanted,
+				     struct ocfs2_alloc_context *meta_ac,
+				     struct buffer_head *bhs[])
+{
+	int count, status, i;
+	u16 suballoc_bit_start;
+	u32 num_got;
+	u64 suballoc_loc, first_blkno;
+	struct ocfs2_super *osb =
+		OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
+	struct ocfs2_extent_block *eb;
+
+	count = 0;
+	while (count < wanted) {
+		status = ocfs2_claim_metadata(handle,
+					      meta_ac,
+					      wanted - count,
+					      &suballoc_loc,
+					      &suballoc_bit_start,
+					      &num_got,
+					      &first_blkno);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		for(i = count;  i < (num_got + count); i++) {
+			bhs[i] = sb_getblk(osb->sb, first_blkno);
+			if (bhs[i] == NULL) {
+				status = -ENOMEM;
+				mlog_errno(status);
+				goto bail;
+			}
+			ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]);
+
+			status = ocfs2_journal_access_eb(handle, et->et_ci,
+							 bhs[i],
+							 OCFS2_JOURNAL_ACCESS_CREATE);
+			if (status < 0) {
+				mlog_errno(status);
+				goto bail;
+			}
+
+			memset(bhs[i]->b_data, 0, osb->sb->s_blocksize);
+			eb = (struct ocfs2_extent_block *) bhs[i]->b_data;
+			/* Ok, setup the minimal stuff here. */
+			strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
+			eb->h_blkno = cpu_to_le64(first_blkno);
+			eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
+			eb->h_suballoc_slot =
+				cpu_to_le16(meta_ac->ac_alloc_slot);
+			eb->h_suballoc_loc = cpu_to_le64(suballoc_loc);
+			eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
+			eb->h_list.l_count =
+				cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
+
+			suballoc_bit_start++;
+			first_blkno++;
+
+			/* We'll also be dirtied by the caller, so
+			 * this isn't absolutely necessary. */
+			ocfs2_journal_dirty(handle, bhs[i]);
+		}
+
+		count += num_got;
+	}
+
+	status = 0;
+bail:
+	if (status < 0) {
+		for(i = 0; i < wanted; i++) {
+			brelse(bhs[i]);
+			bhs[i] = NULL;
+		}
+		mlog_errno(status);
+	}
+	return status;
+}
+
+/*
+ * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth().
+ *
+ * Returns the sum of the rightmost extent rec logical offset and
+ * cluster count.
+ *
+ * ocfs2_add_branch() uses this to determine what logical cluster
+ * value should be populated into the leftmost new branch records.
+ *
+ * ocfs2_shift_tree_depth() uses this to determine the # clusters
+ * value for the new topmost tree record.
+ */
+static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list  *el)
+{
+	int i;
+
+	i = le16_to_cpu(el->l_next_free_rec) - 1;
+
+	return le32_to_cpu(el->l_recs[i].e_cpos) +
+		ocfs2_rec_clusters(el, &el->l_recs[i]);
+}
+
+/*
+ * Change range of the branches in the right most path according to the leaf
+ * extent block's rightmost record.
+ */
+static int ocfs2_adjust_rightmost_branch(handle_t *handle,
+					 struct ocfs2_extent_tree *et)
+{
+	int status;
+	struct ocfs2_path *path = NULL;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+
+	path = ocfs2_new_path_from_et(et);
+	if (!path) {
+		status = -ENOMEM;
+		return status;
+	}
+
+	status = ocfs2_find_path(et->et_ci, path, UINT_MAX);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	status = ocfs2_extend_trans(handle, path_num_items(path));
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	status = ocfs2_journal_access_path(et->et_ci, handle, path);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	el = path_leaf_el(path);
+	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec) - 1];
+
+	ocfs2_adjust_rightmost_records(handle, et, path, rec);
+
+out:
+	ocfs2_free_path(path);
+	return status;
+}
+
+/*
+ * Add an entire tree branch to our inode. eb_bh is the extent block
+ * to start at, if we don't want to start the branch at the root
+ * structure.
+ *
+ * last_eb_bh is required as we have to update it's next_leaf pointer
+ * for the new last extent block.
+ *
+ * the new branch will be 'empty' in the sense that every block will
+ * contain a single record with cluster count == 0.
+ */
+static int ocfs2_add_branch(handle_t *handle,
+			    struct ocfs2_extent_tree *et,
+			    struct buffer_head *eb_bh,
+			    struct buffer_head **last_eb_bh,
+			    struct ocfs2_alloc_context *meta_ac)
+{
+	int status, new_blocks, i;
+	u64 next_blkno, new_last_eb_blk;
+	struct buffer_head *bh;
+	struct buffer_head **new_eb_bhs = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list  *eb_el;
+	struct ocfs2_extent_list  *el;
+	u32 new_cpos, root_end;
+
+	BUG_ON(!last_eb_bh || !*last_eb_bh);
+
+	if (eb_bh) {
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+	} else
+		el = et->et_root_el;
+
+	/* we never add a branch to a leaf. */
+	BUG_ON(!el->l_tree_depth);
+
+	new_blocks = le16_to_cpu(el->l_tree_depth);
+
+	eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
+	new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);
+	root_end = ocfs2_sum_rightmost_rec(et->et_root_el);
+
+	/*
+	 * If there is a gap before the root end and the real end
+	 * of the righmost leaf block, we need to remove the gap
+	 * between new_cpos and root_end first so that the tree
+	 * is consistent after we add a new branch(it will start
+	 * from new_cpos).
+	 */
+	if (root_end > new_cpos) {
+		trace_ocfs2_adjust_rightmost_branch(
+			(unsigned long long)
+			ocfs2_metadata_cache_owner(et->et_ci),
+			root_end, new_cpos);
+
+		status = ocfs2_adjust_rightmost_branch(handle, et);
+		if (status) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	/* allocate the number of new eb blocks we need */
+	new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *),
+			     GFP_KERNEL);
+	if (!new_eb_bhs) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_create_new_meta_bhs(handle, et, new_blocks,
+					   meta_ac, new_eb_bhs);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
+	 * linked with the rest of the tree.
+	 * conversly, new_eb_bhs[0] is the new bottommost leaf.
+	 *
+	 * when we leave the loop, new_last_eb_blk will point to the
+	 * newest leaf, and next_blkno will point to the topmost extent
+	 * block. */
+	next_blkno = new_last_eb_blk = 0;
+	for(i = 0; i < new_blocks; i++) {
+		bh = new_eb_bhs[i];
+		eb = (struct ocfs2_extent_block *) bh->b_data;
+		/* ocfs2_create_new_meta_bhs() should create it right! */
+		BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
+		eb_el = &eb->h_list;
+
+		status = ocfs2_journal_access_eb(handle, et->et_ci, bh,
+						 OCFS2_JOURNAL_ACCESS_CREATE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		eb->h_next_leaf_blk = 0;
+		eb_el->l_tree_depth = cpu_to_le16(i);
+		eb_el->l_next_free_rec = cpu_to_le16(1);
+		/*
+		 * This actually counts as an empty extent as
+		 * c_clusters == 0
+		 */
+		eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
+		eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
+		/*
+		 * eb_el isn't always an interior node, but even leaf
+		 * nodes want a zero'd flags and reserved field so
+		 * this gets the whole 32 bits regardless of use.
+		 */
+		eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0);
+		if (!eb_el->l_tree_depth)
+			new_last_eb_blk = le64_to_cpu(eb->h_blkno);
+
+		ocfs2_journal_dirty(handle, bh);
+		next_blkno = le64_to_cpu(eb->h_blkno);
+	}
+
+	/* This is a bit hairy. We want to update up to three blocks
+	 * here without leaving any of them in an inconsistent state
+	 * in case of error. We don't have to worry about
+	 * journal_dirty erroring as it won't unless we've aborted the
+	 * handle (in which case we would never be here) so reserving
+	 * the write with journal_access is all we need to do. */
+	status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	status = ocfs2_et_root_journal_access(handle, et,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	if (eb_bh) {
+		status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh,
+						 OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	/* Link the new branch into the rest of the tree (el will
+	 * either be on the root_bh, or the extent block passed in. */
+	i = le16_to_cpu(el->l_next_free_rec);
+	el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
+	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
+	el->l_recs[i].e_int_clusters = 0;
+	le16_add_cpu(&el->l_next_free_rec, 1);
+
+	/* fe needs a new last extent block pointer, as does the
+	 * next_leaf on the previously last-extent-block. */
+	ocfs2_et_set_last_eb_blk(et, new_last_eb_blk);
+
+	eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
+	eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);
+
+	ocfs2_journal_dirty(handle, *last_eb_bh);
+	ocfs2_journal_dirty(handle, et->et_root_bh);
+	if (eb_bh)
+		ocfs2_journal_dirty(handle, eb_bh);
+
+	/*
+	 * Some callers want to track the rightmost leaf so pass it
+	 * back here.
+	 */
+	brelse(*last_eb_bh);
+	get_bh(new_eb_bhs[0]);
+	*last_eb_bh = new_eb_bhs[0];
+
+	status = 0;
+bail:
+	if (new_eb_bhs) {
+		for (i = 0; i < new_blocks; i++)
+			brelse(new_eb_bhs[i]);
+		kfree(new_eb_bhs);
+	}
+
+	return status;
+}
+
+/*
+ * adds another level to the allocation tree.
+ * returns back the new extent block so you can add a branch to it
+ * after this call.
+ */
+static int ocfs2_shift_tree_depth(handle_t *handle,
+				  struct ocfs2_extent_tree *et,
+				  struct ocfs2_alloc_context *meta_ac,
+				  struct buffer_head **ret_new_eb_bh)
+{
+	int status, i;
+	u32 new_clusters;
+	struct buffer_head *new_eb_bh = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list  *root_el;
+	struct ocfs2_extent_list  *eb_el;
+
+	status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac,
+					   &new_eb_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
+	/* ocfs2_create_new_meta_bhs() should create it right! */
+	BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
+
+	eb_el = &eb->h_list;
+	root_el = et->et_root_el;
+
+	status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh,
+					 OCFS2_JOURNAL_ACCESS_CREATE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/* copy the root extent list data into the new extent block */
+	eb_el->l_tree_depth = root_el->l_tree_depth;
+	eb_el->l_next_free_rec = root_el->l_next_free_rec;
+	for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++)
+		eb_el->l_recs[i] = root_el->l_recs[i];
+
+	ocfs2_journal_dirty(handle, new_eb_bh);
+
+	status = ocfs2_et_root_journal_access(handle, et,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	new_clusters = ocfs2_sum_rightmost_rec(eb_el);
+
+	/* update root_bh now */
+	le16_add_cpu(&root_el->l_tree_depth, 1);
+	root_el->l_recs[0].e_cpos = 0;
+	root_el->l_recs[0].e_blkno = eb->h_blkno;
+	root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
+	for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
+		memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
+	root_el->l_next_free_rec = cpu_to_le16(1);
+
+	/* If this is our 1st tree depth shift, then last_eb_blk
+	 * becomes the allocated extent block */
+	if (root_el->l_tree_depth == cpu_to_le16(1))
+		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
+
+	ocfs2_journal_dirty(handle, et->et_root_bh);
+
+	*ret_new_eb_bh = new_eb_bh;
+	new_eb_bh = NULL;
+	status = 0;
+bail:
+	brelse(new_eb_bh);
+
+	return status;
+}
+
+/*
+ * Should only be called when there is no space left in any of the
+ * leaf nodes. What we want to do is find the lowest tree depth
+ * non-leaf extent block with room for new records. There are three
+ * valid results of this search:
+ *
+ * 1) a lowest extent block is found, then we pass it back in
+ *    *lowest_eb_bh and return '0'
+ *
+ * 2) the search fails to find anything, but the root_el has room. We
+ *    pass NULL back in *lowest_eb_bh, but still return '0'
+ *
+ * 3) the search fails to find anything AND the root_el is full, in
+ *    which case we return > 0
+ *
+ * return status < 0 indicates an error.
+ */
+static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et,
+				    struct buffer_head **target_bh)
+{
+	int status = 0, i;
+	u64 blkno;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list  *el;
+	struct buffer_head *bh = NULL;
+	struct buffer_head *lowest_bh = NULL;
+
+	*target_bh = NULL;
+
+	el = et->et_root_el;
+
+	while(le16_to_cpu(el->l_tree_depth) > 1) {
+		if (le16_to_cpu(el->l_next_free_rec) == 0) {
+			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
+				    "Owner %llu has empty "
+				    "extent list (next_free_rec == 0)",
+				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
+			status = -EIO;
+			goto bail;
+		}
+		i = le16_to_cpu(el->l_next_free_rec) - 1;
+		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
+		if (!blkno) {
+			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
+				    "Owner %llu has extent "
+				    "list where extent # %d has no physical "
+				    "block start",
+				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i);
+			status = -EIO;
+			goto bail;
+		}
+
+		brelse(bh);
+		bh = NULL;
+
+		status = ocfs2_read_extent_block(et->et_ci, blkno, &bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		eb = (struct ocfs2_extent_block *) bh->b_data;
+		el = &eb->h_list;
+
+		if (le16_to_cpu(el->l_next_free_rec) <
+		    le16_to_cpu(el->l_count)) {
+			brelse(lowest_bh);
+			lowest_bh = bh;
+			get_bh(lowest_bh);
+		}
+	}
+
+	/* If we didn't find one and the fe doesn't have any room,
+	 * then return '1' */
+	el = et->et_root_el;
+	if (!lowest_bh && (el->l_next_free_rec == el->l_count))
+		status = 1;
+
+	*target_bh = lowest_bh;
+bail:
+	brelse(bh);
+
+	return status;
+}
+
+/*
+ * Grow a b-tree so that it has more records.
+ *
+ * We might shift the tree depth in which case existing paths should
+ * be considered invalid.
+ *
+ * Tree depth after the grow is returned via *final_depth.
+ *
+ * *last_eb_bh will be updated by ocfs2_add_branch().
+ */
+static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et,
+			   int *final_depth, struct buffer_head **last_eb_bh,
+			   struct ocfs2_alloc_context *meta_ac)
+{
+	int ret, shift;
+	struct ocfs2_extent_list *el = et->et_root_el;
+	int depth = le16_to_cpu(el->l_tree_depth);
+	struct buffer_head *bh = NULL;
+
+	BUG_ON(meta_ac == NULL);
+
+	shift = ocfs2_find_branch_target(et, &bh);
+	if (shift < 0) {
+		ret = shift;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* We traveled all the way to the bottom of the allocation tree
+	 * and didn't find room for any more extents - we need to add
+	 * another tree level */
+	if (shift) {
+		BUG_ON(bh);
+		trace_ocfs2_grow_tree(
+			(unsigned long long)
+			ocfs2_metadata_cache_owner(et->et_ci),
+			depth);
+
+		/* ocfs2_shift_tree_depth will return us a buffer with
+		 * the new extent block (so we can pass that to
+		 * ocfs2_add_branch). */
+		ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+		depth++;
+		if (depth == 1) {
+			/*
+			 * Special case: we have room now if we shifted from
+			 * tree_depth 0, so no more work needs to be done.
+			 *
+			 * We won't be calling add_branch, so pass
+			 * back *last_eb_bh as the new leaf. At depth
+			 * zero, it should always be null so there's
+			 * no reason to brelse.
+			 */
+			BUG_ON(*last_eb_bh);
+			get_bh(bh);
+			*last_eb_bh = bh;
+			goto out;
+		}
+	}
+
+	/* call ocfs2_add_branch to add the final part of the tree with
+	 * the new data. */
+	ret = ocfs2_add_branch(handle, et, bh, last_eb_bh,
+			       meta_ac);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+out:
+	if (final_depth)
+		*final_depth = depth;
+	brelse(bh);
+	return ret;
+}
+
+/*
+ * This function will discard the rightmost extent record.
+ */
+static void ocfs2_shift_records_right(struct ocfs2_extent_list *el)
+{
+	int next_free = le16_to_cpu(el->l_next_free_rec);
+	int count = le16_to_cpu(el->l_count);
+	unsigned int num_bytes;
+
+	BUG_ON(!next_free);
+	/* This will cause us to go off the end of our extent list. */
+	BUG_ON(next_free >= count);
+
+	num_bytes = sizeof(struct ocfs2_extent_rec) * next_free;
+
+	memmove(&el->l_recs[1], &el->l_recs[0], num_bytes);
+}
+
+static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el,
+			      struct ocfs2_extent_rec *insert_rec)
+{
+	int i, insert_index, next_free, has_empty, num_bytes;
+	u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos);
+	struct ocfs2_extent_rec *rec;
+
+	next_free = le16_to_cpu(el->l_next_free_rec);
+	has_empty = ocfs2_is_empty_extent(&el->l_recs[0]);
+
+	BUG_ON(!next_free);
+
+	/* The tree code before us didn't allow enough room in the leaf. */
+	BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
+
+	/*
+	 * The easiest way to approach this is to just remove the
+	 * empty extent and temporarily decrement next_free.
+	 */
+	if (has_empty) {
+		/*
+		 * If next_free was 1 (only an empty extent), this
+		 * loop won't execute, which is fine. We still want
+		 * the decrement above to happen.
+		 */
+		for(i = 0; i < (next_free - 1); i++)
+			el->l_recs[i] = el->l_recs[i+1];
+
+		next_free--;
+	}
+
+	/*
+	 * Figure out what the new record index should be.
+	 */
+	for(i = 0; i < next_free; i++) {
+		rec = &el->l_recs[i];
+
+		if (insert_cpos < le32_to_cpu(rec->e_cpos))
+			break;
+	}
+	insert_index = i;
+
+	trace_ocfs2_rotate_leaf(insert_cpos, insert_index,
+				has_empty, next_free,
+				le16_to_cpu(el->l_count));
+
+	BUG_ON(insert_index < 0);
+	BUG_ON(insert_index >= le16_to_cpu(el->l_count));
+	BUG_ON(insert_index > next_free);
+
+	/*
+	 * No need to memmove if we're just adding to the tail.
+	 */
+	if (insert_index != next_free) {
+		BUG_ON(next_free >= le16_to_cpu(el->l_count));
+
+		num_bytes = next_free - insert_index;
+		num_bytes *= sizeof(struct ocfs2_extent_rec);
+		memmove(&el->l_recs[insert_index + 1],
+			&el->l_recs[insert_index],
+			num_bytes);
+	}
+
+	/*
+	 * Either we had an empty extent, and need to re-increment or
+	 * there was no empty extent on a non full rightmost leaf node,
+	 * in which case we still need to increment.
+	 */
+	next_free++;
+	el->l_next_free_rec = cpu_to_le16(next_free);
+	/*
+	 * Make sure none of the math above just messed up our tree.
+	 */
+	BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count));
+
+	el->l_recs[insert_index] = *insert_rec;
+
+}
+
+static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el)
+{
+	int size, num_recs = le16_to_cpu(el->l_next_free_rec);
+
+	BUG_ON(num_recs == 0);
+
+	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
+		num_recs--;
+		size = num_recs * sizeof(struct ocfs2_extent_rec);
+		memmove(&el->l_recs[0], &el->l_recs[1], size);
+		memset(&el->l_recs[num_recs], 0,
+		       sizeof(struct ocfs2_extent_rec));
+		el->l_next_free_rec = cpu_to_le16(num_recs);
+	}
+}
+
+/*
+ * Create an empty extent record .
+ *
+ * l_next_free_rec may be updated.
+ *
+ * If an empty extent already exists do nothing.
+ */
+static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el)
+{
+	int next_free = le16_to_cpu(el->l_next_free_rec);
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	if (next_free == 0)
+		goto set_and_inc;
+
+	if (ocfs2_is_empty_extent(&el->l_recs[0]))
+		return;
+
+	mlog_bug_on_msg(el->l_count == el->l_next_free_rec,
+			"Asked to create an empty extent in a full list:\n"
+			"count = %u, tree depth = %u",
+			le16_to_cpu(el->l_count),
+			le16_to_cpu(el->l_tree_depth));
+
+	ocfs2_shift_records_right(el);
+
+set_and_inc:
+	le16_add_cpu(&el->l_next_free_rec, 1);
+	memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
+}
+
+/*
+ * For a rotation which involves two leaf nodes, the "root node" is
+ * the lowest level tree node which contains a path to both leafs. This
+ * resulting set of information can be used to form a complete "subtree"
+ *
+ * This function is passed two full paths from the dinode down to a
+ * pair of adjacent leaves. It's task is to figure out which path
+ * index contains the subtree root - this can be the root index itself
+ * in a worst-case rotation.
+ *
+ * The array index of the subtree root is passed back.
+ */
+int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
+			    struct ocfs2_path *left,
+			    struct ocfs2_path *right)
+{
+	int i = 0;
+
+	/*
+	 * Check that the caller passed in two paths from the same tree.
+	 */
+	BUG_ON(path_root_bh(left) != path_root_bh(right));
+
+	do {
+		i++;
+
+		/*
+		 * The caller didn't pass two adjacent paths.
+		 */
+		mlog_bug_on_msg(i > left->p_tree_depth,
+				"Owner %llu, left depth %u, right depth %u\n"
+				"left leaf blk %llu, right leaf blk %llu\n",
+				(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+				left->p_tree_depth, right->p_tree_depth,
+				(unsigned long long)path_leaf_bh(left)->b_blocknr,
+				(unsigned long long)path_leaf_bh(right)->b_blocknr);
+	} while (left->p_node[i].bh->b_blocknr ==
+		 right->p_node[i].bh->b_blocknr);
+
+	return i - 1;
+}
+
+typedef void (path_insert_t)(void *, struct buffer_head *);
+
+/*
+ * Traverse a btree path in search of cpos, starting at root_el.
+ *
+ * This code can be called with a cpos larger than the tree, in which
+ * case it will return the rightmost path.
+ */
+static int __ocfs2_find_path(struct ocfs2_caching_info *ci,
+			     struct ocfs2_extent_list *root_el, u32 cpos,
+			     path_insert_t *func, void *data)
+{
+	int i, ret = 0;
+	u32 range;
+	u64 blkno;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+
+	el = root_el;
+	while (el->l_tree_depth) {
+		if (le16_to_cpu(el->l_next_free_rec) == 0) {
+			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
+				    "Owner %llu has empty extent list at "
+				    "depth %u\n",
+				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
+				    le16_to_cpu(el->l_tree_depth));
+			ret = -EROFS;
+			goto out;
+
+		}
+
+		for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) {
+			rec = &el->l_recs[i];
+
+			/*
+			 * In the case that cpos is off the allocation
+			 * tree, this should just wind up returning the
+			 * rightmost record.
+			 */
+			range = le32_to_cpu(rec->e_cpos) +
+				ocfs2_rec_clusters(el, rec);
+			if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
+			    break;
+		}
+
+		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
+		if (blkno == 0) {
+			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
+				    "Owner %llu has bad blkno in extent list "
+				    "at depth %u (index %d)\n",
+				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
+				    le16_to_cpu(el->l_tree_depth), i);
+			ret = -EROFS;
+			goto out;
+		}
+
+		brelse(bh);
+		bh = NULL;
+		ret = ocfs2_read_extent_block(ci, blkno, &bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) bh->b_data;
+		el = &eb->h_list;
+
+		if (le16_to_cpu(el->l_next_free_rec) >
+		    le16_to_cpu(el->l_count)) {
+			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
+				    "Owner %llu has bad count in extent list "
+				    "at block %llu (next free=%u, count=%u)\n",
+				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
+				    (unsigned long long)bh->b_blocknr,
+				    le16_to_cpu(el->l_next_free_rec),
+				    le16_to_cpu(el->l_count));
+			ret = -EROFS;
+			goto out;
+		}
+
+		if (func)
+			func(data, bh);
+	}
+
+out:
+	/*
+	 * Catch any trailing bh that the loop didn't handle.
+	 */
+	brelse(bh);
+
+	return ret;
+}
+
+/*
+ * Given an initialized path (that is, it has a valid root extent
+ * list), this function will traverse the btree in search of the path
+ * which would contain cpos.
+ *
+ * The path traveled is recorded in the path structure.
+ *
+ * Note that this will not do any comparisons on leaf node extent
+ * records, so it will work fine in the case that we just added a tree
+ * branch.
+ */
+struct find_path_data {
+	int index;
+	struct ocfs2_path *path;
+};
+static void find_path_ins(void *data, struct buffer_head *bh)
+{
+	struct find_path_data *fp = data;
+
+	get_bh(bh);
+	ocfs2_path_insert_eb(fp->path, fp->index, bh);
+	fp->index++;
+}
+int ocfs2_find_path(struct ocfs2_caching_info *ci,
+		    struct ocfs2_path *path, u32 cpos)
+{
+	struct find_path_data data;
+
+	data.index = 1;
+	data.path = path;
+	return __ocfs2_find_path(ci, path_root_el(path), cpos,
+				 find_path_ins, &data);
+}
+
+static void find_leaf_ins(void *data, struct buffer_head *bh)
+{
+	struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data;
+	struct ocfs2_extent_list *el = &eb->h_list;
+	struct buffer_head **ret = data;
+
+	/* We want to retain only the leaf block. */
+	if (le16_to_cpu(el->l_tree_depth) == 0) {
+		get_bh(bh);
+		*ret = bh;
+	}
+}
+/*
+ * Find the leaf block in the tree which would contain cpos. No
+ * checking of the actual leaf is done.
+ *
+ * Some paths want to call this instead of allocating a path structure
+ * and calling ocfs2_find_path().
+ *
+ * This function doesn't handle non btree extent lists.
+ */
+int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
+		    struct ocfs2_extent_list *root_el, u32 cpos,
+		    struct buffer_head **leaf_bh)
+{
+	int ret;
+	struct buffer_head *bh = NULL;
+
+	ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	*leaf_bh = bh;
+out:
+	return ret;
+}
+
+/*
+ * Adjust the adjacent records (left_rec, right_rec) involved in a rotation.
+ *
+ * Basically, we've moved stuff around at the bottom of the tree and
+ * we need to fix up the extent records above the changes to reflect
+ * the new changes.
+ *
+ * left_rec: the record on the left.
+ * left_child_el: is the child list pointed to by left_rec
+ * right_rec: the record to the right of left_rec
+ * right_child_el: is the child list pointed to by right_rec
+ *
+ * By definition, this only works on interior nodes.
+ */
+static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec,
+				  struct ocfs2_extent_list *left_child_el,
+				  struct ocfs2_extent_rec *right_rec,
+				  struct ocfs2_extent_list *right_child_el)
+{
+	u32 left_clusters, right_end;
+
+	/*
+	 * Interior nodes never have holes. Their cpos is the cpos of
+	 * the leftmost record in their child list. Their cluster
+	 * count covers the full theoretical range of their child list
+	 * - the range between their cpos and the cpos of the record
+	 * immediately to their right.
+	 */
+	left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos);
+	if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) {
+		BUG_ON(right_child_el->l_tree_depth);
+		BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1);
+		left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos);
+	}
+	left_clusters -= le32_to_cpu(left_rec->e_cpos);
+	left_rec->e_int_clusters = cpu_to_le32(left_clusters);
+
+	/*
+	 * Calculate the rightmost cluster count boundary before
+	 * moving cpos - we will need to adjust clusters after
+	 * updating e_cpos to keep the same highest cluster count.
+	 */
+	right_end = le32_to_cpu(right_rec->e_cpos);
+	right_end += le32_to_cpu(right_rec->e_int_clusters);
+
+	right_rec->e_cpos = left_rec->e_cpos;
+	le32_add_cpu(&right_rec->e_cpos, left_clusters);
+
+	right_end -= le32_to_cpu(right_rec->e_cpos);
+	right_rec->e_int_clusters = cpu_to_le32(right_end);
+}
+
+/*
+ * Adjust the adjacent root node records involved in a
+ * rotation. left_el_blkno is passed in as a key so that we can easily
+ * find it's index in the root list.
+ */
+static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el,
+				      struct ocfs2_extent_list *left_el,
+				      struct ocfs2_extent_list *right_el,
+				      u64 left_el_blkno)
+{
+	int i;
+
+	BUG_ON(le16_to_cpu(root_el->l_tree_depth) <=
+	       le16_to_cpu(left_el->l_tree_depth));
+
+	for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) {
+		if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno)
+			break;
+	}
+
+	/*
+	 * The path walking code should have never returned a root and
+	 * two paths which are not adjacent.
+	 */
+	BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1));
+
+	ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el,
+				      &root_el->l_recs[i + 1], right_el);
+}
+
+/*
+ * We've changed a leaf block (in right_path) and need to reflect that
+ * change back up the subtree.
+ *
+ * This happens in multiple places:
+ *   - When we've moved an extent record from the left path leaf to the right
+ *     path leaf to make room for an empty extent in the left path leaf.
+ *   - When our insert into the right path leaf is at the leftmost edge
+ *     and requires an update of the path immediately to it's left. This
+ *     can occur at the end of some types of rotation and appending inserts.
+ *   - When we've adjusted the last extent record in the left path leaf and the
+ *     1st extent record in the right path leaf during cross extent block merge.
+ */
+static void ocfs2_complete_edge_insert(handle_t *handle,
+				       struct ocfs2_path *left_path,
+				       struct ocfs2_path *right_path,
+				       int subtree_index)
+{
+	int i, idx;
+	struct ocfs2_extent_list *el, *left_el, *right_el;
+	struct ocfs2_extent_rec *left_rec, *right_rec;
+	struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
+
+	/*
+	 * Update the counts and position values within all the
+	 * interior nodes to reflect the leaf rotation we just did.
+	 *
+	 * The root node is handled below the loop.
+	 *
+	 * We begin the loop with right_el and left_el pointing to the
+	 * leaf lists and work our way up.
+	 *
+	 * NOTE: within this loop, left_el and right_el always refer
+	 * to the *child* lists.
+	 */
+	left_el = path_leaf_el(left_path);
+	right_el = path_leaf_el(right_path);
+	for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) {
+		trace_ocfs2_complete_edge_insert(i);
+
+		/*
+		 * One nice property of knowing that all of these
+		 * nodes are below the root is that we only deal with
+		 * the leftmost right node record and the rightmost
+		 * left node record.
+		 */
+		el = left_path->p_node[i].el;
+		idx = le16_to_cpu(left_el->l_next_free_rec) - 1;
+		left_rec = &el->l_recs[idx];
+
+		el = right_path->p_node[i].el;
+		right_rec = &el->l_recs[0];
+
+		ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec,
+					      right_el);
+
+		ocfs2_journal_dirty(handle, left_path->p_node[i].bh);
+		ocfs2_journal_dirty(handle, right_path->p_node[i].bh);
+
+		/*
+		 * Setup our list pointers now so that the current
+		 * parents become children in the next iteration.
+		 */
+		left_el = left_path->p_node[i].el;
+		right_el = right_path->p_node[i].el;
+	}
+
+	/*
+	 * At the root node, adjust the two adjacent records which
+	 * begin our path to the leaves.
+	 */
+
+	el = left_path->p_node[subtree_index].el;
+	left_el = left_path->p_node[subtree_index + 1].el;
+	right_el = right_path->p_node[subtree_index + 1].el;
+
+	ocfs2_adjust_root_records(el, left_el, right_el,
+				  left_path->p_node[subtree_index + 1].bh->b_blocknr);
+
+	root_bh = left_path->p_node[subtree_index].bh;
+
+	ocfs2_journal_dirty(handle, root_bh);
+}
+
+static int ocfs2_rotate_subtree_right(handle_t *handle,
+				      struct ocfs2_extent_tree *et,
+				      struct ocfs2_path *left_path,
+				      struct ocfs2_path *right_path,
+				      int subtree_index)
+{
+	int ret, i;
+	struct buffer_head *right_leaf_bh;
+	struct buffer_head *left_leaf_bh = NULL;
+	struct buffer_head *root_bh;
+	struct ocfs2_extent_list *right_el, *left_el;
+	struct ocfs2_extent_rec move_rec;
+
+	left_leaf_bh = path_leaf_bh(left_path);
+	left_el = path_leaf_el(left_path);
+
+	if (left_el->l_next_free_rec != left_el->l_count) {
+		ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
+			    "Inode %llu has non-full interior leaf node %llu"
+			    "(next free = %u)",
+			    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+			    (unsigned long long)left_leaf_bh->b_blocknr,
+			    le16_to_cpu(left_el->l_next_free_rec));
+		return -EROFS;
+	}
+
+	/*
+	 * This extent block may already have an empty record, so we
+	 * return early if so.
+	 */
+	if (ocfs2_is_empty_extent(&left_el->l_recs[0]))
+		return 0;
+
+	root_bh = left_path->p_node[subtree_index].bh;
+	BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
+
+	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
+					   subtree_index);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
+		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
+						   right_path, i);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
+						   left_path, i);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	right_leaf_bh = path_leaf_bh(right_path);
+	right_el = path_leaf_el(right_path);
+
+	/* This is a code error, not a disk corruption. */
+	mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails "
+			"because rightmost leaf block %llu is empty\n",
+			(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+			(unsigned long long)right_leaf_bh->b_blocknr);
+
+	ocfs2_create_empty_extent(right_el);
+
+	ocfs2_journal_dirty(handle, right_leaf_bh);
+
+	/* Do the copy now. */
+	i = le16_to_cpu(left_el->l_next_free_rec) - 1;
+	move_rec = left_el->l_recs[i];
+	right_el->l_recs[0] = move_rec;
+
+	/*
+	 * Clear out the record we just copied and shift everything
+	 * over, leaving an empty extent in the left leaf.
+	 *
+	 * We temporarily subtract from next_free_rec so that the
+	 * shift will lose the tail record (which is now defunct).
+	 */
+	le16_add_cpu(&left_el->l_next_free_rec, -1);
+	ocfs2_shift_records_right(left_el);
+	memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
+	le16_add_cpu(&left_el->l_next_free_rec, 1);
+
+	ocfs2_journal_dirty(handle, left_leaf_bh);
+
+	ocfs2_complete_edge_insert(handle, left_path, right_path,
+				   subtree_index);
+
+out:
+	return ret;
+}
+
+/*
+ * Given a full path, determine what cpos value would return us a path
+ * containing the leaf immediately to the left of the current one.
+ *
+ * Will return zero if the path passed in is already the leftmost path.
+ */
+int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
+				  struct ocfs2_path *path, u32 *cpos)
+{
+	int i, j, ret = 0;
+	u64 blkno;
+	struct ocfs2_extent_list *el;
+
+	BUG_ON(path->p_tree_depth == 0);
+
+	*cpos = 0;
+
+	blkno = path_leaf_bh(path)->b_blocknr;
+
+	/* Start at the tree node just above the leaf and work our way up. */
+	i = path->p_tree_depth - 1;
+	while (i >= 0) {
+		el = path->p_node[i].el;
+
+		/*
+		 * Find the extent record just before the one in our
+		 * path.
+		 */
+		for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
+			if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
+				if (j == 0) {
+					if (i == 0) {
+						/*
+						 * We've determined that the
+						 * path specified is already
+						 * the leftmost one - return a
+						 * cpos of zero.
+						 */
+						goto out;
+					}
+					/*
+					 * The leftmost record points to our
+					 * leaf - we need to travel up the
+					 * tree one level.
+					 */
+					goto next_node;
+				}
+
+				*cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos);
+				*cpos = *cpos + ocfs2_rec_clusters(el,
+							   &el->l_recs[j - 1]);
+				*cpos = *cpos - 1;
+				goto out;
+			}
+		}
+
+		/*
+		 * If we got here, we never found a valid node where
+		 * the tree indicated one should be.
+		 */
+		ocfs2_error(sb,
+			    "Invalid extent tree at extent block %llu\n",
+			    (unsigned long long)blkno);
+		ret = -EROFS;
+		goto out;
+
+next_node:
+		blkno = path->p_node[i].bh->b_blocknr;
+		i--;
+	}
+
+out:
+	return ret;
+}
+
+/*
+ * Extend the transaction by enough credits to complete the rotation,
+ * and still leave at least the original number of credits allocated
+ * to this transaction.
+ */
+static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
+					   int op_credits,
+					   struct ocfs2_path *path)
+{
+	int ret = 0;
+	int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
+
+	if (handle->h_buffer_credits < credits)
+		ret = ocfs2_extend_trans(handle,
+					 credits - handle->h_buffer_credits);
+
+	return ret;
+}
+
+/*
+ * Trap the case where we're inserting into the theoretical range past
+ * the _actual_ left leaf range. Otherwise, we'll rotate a record
+ * whose cpos is less than ours into the right leaf.
+ *
+ * It's only necessary to look at the rightmost record of the left
+ * leaf because the logic that calls us should ensure that the
+ * theoretical ranges in the path components above the leaves are
+ * correct.
+ */
+static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
+						 u32 insert_cpos)
+{
+	struct ocfs2_extent_list *left_el;
+	struct ocfs2_extent_rec *rec;
+	int next_free;
+
+	left_el = path_leaf_el(left_path);
+	next_free = le16_to_cpu(left_el->l_next_free_rec);
+	rec = &left_el->l_recs[next_free - 1];
+
+	if (insert_cpos > le32_to_cpu(rec->e_cpos))
+		return 1;
+	return 0;
+}
+
+static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos)
+{
+	int next_free = le16_to_cpu(el->l_next_free_rec);
+	unsigned int range;
+	struct ocfs2_extent_rec *rec;
+
+	if (next_free == 0)
+		return 0;
+
+	rec = &el->l_recs[0];
+	if (ocfs2_is_empty_extent(rec)) {
+		/* Empty list. */
+		if (next_free == 1)
+			return 0;
+		rec = &el->l_recs[1];
+	}
+
+	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+	if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
+		return 1;
+	return 0;
+}
+
+/*
+ * Rotate all the records in a btree right one record, starting at insert_cpos.
+ *
+ * The path to the rightmost leaf should be passed in.
+ *
+ * The array is assumed to be large enough to hold an entire path (tree depth).
+ *
+ * Upon successful return from this function:
+ *
+ * - The 'right_path' array will contain a path to the leaf block
+ *   whose range contains e_cpos.
+ * - That leaf block will have a single empty extent in list index 0.
+ * - In the case that the rotation requires a post-insert update,
+ *   *ret_left_path will contain a valid path which can be passed to
+ *   ocfs2_insert_path().
+ */
+static int ocfs2_rotate_tree_right(handle_t *handle,
+				   struct ocfs2_extent_tree *et,
+				   enum ocfs2_split_type split,
+				   u32 insert_cpos,
+				   struct ocfs2_path *right_path,
+				   struct ocfs2_path **ret_left_path)
+{
+	int ret, start, orig_credits = handle->h_buffer_credits;
+	u32 cpos;
+	struct ocfs2_path *left_path = NULL;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
+
+	*ret_left_path = NULL;
+
+	left_path = ocfs2_new_path_from_path(right_path);
+	if (!left_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_rotate_tree_right(
+		(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+		insert_cpos, cpos);
+
+	/*
+	 * What we want to do here is:
+	 *
+	 * 1) Start with the rightmost path.
+	 *
+	 * 2) Determine a path to the leaf block directly to the left
+	 *    of that leaf.
+	 *
+	 * 3) Determine the 'subtree root' - the lowest level tree node
+	 *    which contains a path to both leaves.
+	 *
+	 * 4) Rotate the subtree.
+	 *
+	 * 5) Find the next subtree by considering the left path to be
+	 *    the new right path.
+	 *
+	 * The check at the top of this while loop also accepts
+	 * insert_cpos == cpos because cpos is only a _theoretical_
+	 * value to get us the left path - insert_cpos might very well
+	 * be filling that hole.
+	 *
+	 * Stop at a cpos of '0' because we either started at the
+	 * leftmost branch (i.e., a tree with one branch and a
+	 * rotation inside of it), or we've gone as far as we can in
+	 * rotating subtrees.
+	 */
+	while (cpos && insert_cpos <= cpos) {
+		trace_ocfs2_rotate_tree_right(
+			(unsigned long long)
+			ocfs2_metadata_cache_owner(et->et_ci),
+			insert_cpos, cpos);
+
+		ret = ocfs2_find_path(et->et_ci, left_path, cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		mlog_bug_on_msg(path_leaf_bh(left_path) ==
+				path_leaf_bh(right_path),
+				"Owner %llu: error during insert of %u "
+				"(left path cpos %u) results in two identical "
+				"paths ending at %llu\n",
+				(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+				insert_cpos, cpos,
+				(unsigned long long)
+				path_leaf_bh(left_path)->b_blocknr);
+
+		if (split == SPLIT_NONE &&
+		    ocfs2_rotate_requires_path_adjustment(left_path,
+							  insert_cpos)) {
+
+			/*
+			 * We've rotated the tree as much as we
+			 * should. The rest is up to
+			 * ocfs2_insert_path() to complete, after the
+			 * record insertion. We indicate this
+			 * situation by returning the left path.
+			 *
+			 * The reason we don't adjust the records here
+			 * before the record insert is that an error
+			 * later might break the rule where a parent
+			 * record e_cpos will reflect the actual
+			 * e_cpos of the 1st nonempty record of the
+			 * child list.
+			 */
+			*ret_left_path = left_path;
+			goto out_ret_path;
+		}
+
+		start = ocfs2_find_subtree_root(et, left_path, right_path);
+
+		trace_ocfs2_rotate_subtree(start,
+			(unsigned long long)
+			right_path->p_node[start].bh->b_blocknr,
+			right_path->p_tree_depth);
+
+		ret = ocfs2_extend_rotate_transaction(handle, start,
+						      orig_credits, right_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_rotate_subtree_right(handle, et, left_path,
+						 right_path, start);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (split != SPLIT_NONE &&
+		    ocfs2_leftmost_rec_contains(path_leaf_el(right_path),
+						insert_cpos)) {
+			/*
+			 * A rotate moves the rightmost left leaf
+			 * record over to the leftmost right leaf
+			 * slot. If we're doing an extent split
+			 * instead of a real insert, then we have to
+			 * check that the extent to be split wasn't
+			 * just moved over. If it was, then we can
+			 * exit here, passing left_path back -
+			 * ocfs2_split_extent() is smart enough to
+			 * search both leaves.
+			 */
+			*ret_left_path = left_path;
+			goto out_ret_path;
+		}
+
+		/*
+		 * There is no need to re-read the next right path
+		 * as we know that it'll be our current left
+		 * path. Optimize by copying values instead.
+		 */
+		ocfs2_mv_path(right_path, left_path);
+
+		ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+out:
+	ocfs2_free_path(left_path);
+
+out_ret_path:
+	return ret;
+}
+
+static int ocfs2_update_edge_lengths(handle_t *handle,
+				     struct ocfs2_extent_tree *et,
+				     int subtree_index, struct ocfs2_path *path)
+{
+	int i, idx, ret;
+	struct ocfs2_extent_rec *rec;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_block *eb;
+	u32 range;
+
+	/*
+	 * In normal tree rotation process, we will never touch the
+	 * tree branch above subtree_index and ocfs2_extend_rotate_transaction
+	 * doesn't reserve the credits for them either.
+	 *
+	 * But we do have a special case here which will update the rightmost
+	 * records for all the bh in the path.
+	 * So we have to allocate extra credits and access them.
+	 */
+	ret = ocfs2_extend_trans(handle, subtree_index);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Path should always be rightmost. */
+	eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
+	BUG_ON(eb->h_next_leaf_blk != 0ULL);
+
+	el = &eb->h_list;
+	BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
+	idx = le16_to_cpu(el->l_next_free_rec) - 1;
+	rec = &el->l_recs[idx];
+	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+
+	for (i = 0; i < path->p_tree_depth; i++) {
+		el = path->p_node[i].el;
+		idx = le16_to_cpu(el->l_next_free_rec) - 1;
+		rec = &el->l_recs[idx];
+
+		rec->e_int_clusters = cpu_to_le32(range);
+		le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
+
+		ocfs2_journal_dirty(handle, path->p_node[i].bh);
+	}
+out:
+	return ret;
+}
+
+static void ocfs2_unlink_path(handle_t *handle,
+			      struct ocfs2_extent_tree *et,
+			      struct ocfs2_cached_dealloc_ctxt *dealloc,
+			      struct ocfs2_path *path, int unlink_start)
+{
+	int ret, i;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *el;
+	struct buffer_head *bh;
+
+	for(i = unlink_start; i < path_num_items(path); i++) {
+		bh = path->p_node[i].bh;
+
+		eb = (struct ocfs2_extent_block *)bh->b_data;
+		/*
+		 * Not all nodes might have had their final count
+		 * decremented by the caller - handle this here.
+		 */
+		el = &eb->h_list;
+		if (le16_to_cpu(el->l_next_free_rec) > 1) {
+			mlog(ML_ERROR,
+			     "Inode %llu, attempted to remove extent block "
+			     "%llu with %u records\n",
+			     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+			     (unsigned long long)le64_to_cpu(eb->h_blkno),
+			     le16_to_cpu(el->l_next_free_rec));
+
+			ocfs2_journal_dirty(handle, bh);
+			ocfs2_remove_from_cache(et->et_ci, bh);
+			continue;
+		}
+
+		el->l_next_free_rec = 0;
+		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
+
+		ocfs2_journal_dirty(handle, bh);
+
+		ret = ocfs2_cache_extent_block_free(dealloc, eb);
+		if (ret)
+			mlog_errno(ret);
+
+		ocfs2_remove_from_cache(et->et_ci, bh);
+	}
+}
+
+static void ocfs2_unlink_subtree(handle_t *handle,
+				 struct ocfs2_extent_tree *et,
+				 struct ocfs2_path *left_path,
+				 struct ocfs2_path *right_path,
+				 int subtree_index,
+				 struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int i;
+	struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
+	struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_block *eb;
+
+	el = path_leaf_el(left_path);
+
+	eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
+
+	for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
+		if (root_el->l_recs[i].e_blkno == eb->h_blkno)
+			break;
+
+	BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
+
+	memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
+	le16_add_cpu(&root_el->l_next_free_rec, -1);
+
+	eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
+	eb->h_next_leaf_blk = 0;
+
+	ocfs2_journal_dirty(handle, root_bh);
+	ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
+
+	ocfs2_unlink_path(handle, et, dealloc, right_path,
+			  subtree_index + 1);
+}
+
+static int ocfs2_rotate_subtree_left(handle_t *handle,
+				     struct ocfs2_extent_tree *et,
+				     struct ocfs2_path *left_path,
+				     struct ocfs2_path *right_path,
+				     int subtree_index,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc,
+				     int *deleted)
+{
+	int ret, i, del_right_subtree = 0, right_has_empty = 0;
+	struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
+	struct ocfs2_extent_list *right_leaf_el, *left_leaf_el;
+	struct ocfs2_extent_block *eb;
+
+	*deleted = 0;
+
+	right_leaf_el = path_leaf_el(right_path);
+	left_leaf_el = path_leaf_el(left_path);
+	root_bh = left_path->p_node[subtree_index].bh;
+	BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
+
+	if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0]))
+		return 0;
+
+	eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
+	if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
+		/*
+		 * It's legal for us to proceed if the right leaf is
+		 * the rightmost one and it has an empty extent. There
+		 * are two cases to handle - whether the leaf will be
+		 * empty after removal or not. If the leaf isn't empty
+		 * then just remove the empty extent up front. The
+		 * next block will handle empty leaves by flagging
+		 * them for unlink.
+		 *
+		 * Non rightmost leaves will throw -EAGAIN and the
+		 * caller can manually move the subtree and retry.
+		 */
+
+		if (eb->h_next_leaf_blk != 0ULL)
+			return -EAGAIN;
+
+		if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
+			ret = ocfs2_journal_access_eb(handle, et->et_ci,
+						      path_leaf_bh(right_path),
+						      OCFS2_JOURNAL_ACCESS_WRITE);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ocfs2_remove_empty_extent(right_leaf_el);
+		} else
+			right_has_empty = 1;
+	}
+
+	if (eb->h_next_leaf_blk == 0ULL &&
+	    le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) {
+		/*
+		 * We have to update i_last_eb_blk during the meta
+		 * data delete.
+		 */
+		ret = ocfs2_et_root_journal_access(handle, et,
+						   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		del_right_subtree = 1;
+	}
+
+	/*
+	 * Getting here with an empty extent in the right path implies
+	 * that it's the rightmost path and will be deleted.
+	 */
+	BUG_ON(right_has_empty && !del_right_subtree);
+
+	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
+					   subtree_index);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
+		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
+						   right_path, i);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
+						   left_path, i);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (!right_has_empty) {
+		/*
+		 * Only do this if we're moving a real
+		 * record. Otherwise, the action is delayed until
+		 * after removal of the right path in which case we
+		 * can do a simple shift to remove the empty extent.
+		 */
+		ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]);
+		memset(&right_leaf_el->l_recs[0], 0,
+		       sizeof(struct ocfs2_extent_rec));
+	}
+	if (eb->h_next_leaf_blk == 0ULL) {
+		/*
+		 * Move recs over to get rid of empty extent, decrease
+		 * next_free. This is allowed to remove the last
+		 * extent in our leaf (setting l_next_free_rec to
+		 * zero) - the delete code below won't care.
+		 */
+		ocfs2_remove_empty_extent(right_leaf_el);
+	}
+
+	ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
+	ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
+
+	if (del_right_subtree) {
+		ocfs2_unlink_subtree(handle, et, left_path, right_path,
+				     subtree_index, dealloc);
+		ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
+						left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
+		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
+
+		/*
+		 * Removal of the extent in the left leaf was skipped
+		 * above so we could delete the right path
+		 * 1st.
+		 */
+		if (right_has_empty)
+			ocfs2_remove_empty_extent(left_leaf_el);
+
+		ocfs2_journal_dirty(handle, et_root_bh);
+
+		*deleted = 1;
+	} else
+		ocfs2_complete_edge_insert(handle, left_path, right_path,
+					   subtree_index);
+
+out:
+	return ret;
+}
+
+/*
+ * Given a full path, determine what cpos value would return us a path
+ * containing the leaf immediately to the right of the current one.
+ *
+ * Will return zero if the path passed in is already the rightmost path.
+ *
+ * This looks similar, but is subtly different to
+ * ocfs2_find_cpos_for_left_leaf().
+ */
+int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
+				   struct ocfs2_path *path, u32 *cpos)
+{
+	int i, j, ret = 0;
+	u64 blkno;
+	struct ocfs2_extent_list *el;
+
+	*cpos = 0;
+
+	if (path->p_tree_depth == 0)
+		return 0;
+
+	blkno = path_leaf_bh(path)->b_blocknr;
+
+	/* Start at the tree node just above the leaf and work our way up. */
+	i = path->p_tree_depth - 1;
+	while (i >= 0) {
+		int next_free;
+
+		el = path->p_node[i].el;
+
+		/*
+		 * Find the extent record just after the one in our
+		 * path.
+		 */
+		next_free = le16_to_cpu(el->l_next_free_rec);
+		for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
+			if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
+				if (j == (next_free - 1)) {
+					if (i == 0) {
+						/*
+						 * We've determined that the
+						 * path specified is already
+						 * the rightmost one - return a
+						 * cpos of zero.
+						 */
+						goto out;
+					}
+					/*
+					 * The rightmost record points to our
+					 * leaf - we need to travel up the
+					 * tree one level.
+					 */
+					goto next_node;
+				}
+
+				*cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos);
+				goto out;
+			}
+		}
+
+		/*
+		 * If we got here, we never found a valid node where
+		 * the tree indicated one should be.
+		 */
+		ocfs2_error(sb,
+			    "Invalid extent tree at extent block %llu\n",
+			    (unsigned long long)blkno);
+		ret = -EROFS;
+		goto out;
+
+next_node:
+		blkno = path->p_node[i].bh->b_blocknr;
+		i--;
+	}
+
+out:
+	return ret;
+}
+
+static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle,
+					    struct ocfs2_extent_tree *et,
+					    struct ocfs2_path *path)
+{
+	int ret;
+	struct buffer_head *bh = path_leaf_bh(path);
+	struct ocfs2_extent_list *el = path_leaf_el(path);
+
+	if (!ocfs2_is_empty_extent(&el->l_recs[0]))
+		return 0;
+
+	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
+					   path_num_items(path) - 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_remove_empty_extent(el);
+	ocfs2_journal_dirty(handle, bh);
+
+out:
+	return ret;
+}
+
+static int __ocfs2_rotate_tree_left(handle_t *handle,
+				    struct ocfs2_extent_tree *et,
+				    int orig_credits,
+				    struct ocfs2_path *path,
+				    struct ocfs2_cached_dealloc_ctxt *dealloc,
+				    struct ocfs2_path **empty_extent_path)
+{
+	int ret, subtree_root, deleted;
+	u32 right_cpos;
+	struct ocfs2_path *left_path = NULL;
+	struct ocfs2_path *right_path = NULL;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
+
+	BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0])));
+
+	*empty_extent_path = NULL;
+
+	ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	left_path = ocfs2_new_path_from_path(path);
+	if (!left_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_cp_path(left_path, path);
+
+	right_path = ocfs2_new_path_from_path(path);
+	if (!right_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	while (right_cpos) {
+		ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		subtree_root = ocfs2_find_subtree_root(et, left_path,
+						       right_path);
+
+		trace_ocfs2_rotate_subtree(subtree_root,
+		     (unsigned long long)
+		     right_path->p_node[subtree_root].bh->b_blocknr,
+		     right_path->p_tree_depth);
+
+		ret = ocfs2_extend_rotate_transaction(handle, subtree_root,
+						      orig_credits, left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * Caller might still want to make changes to the
+		 * tree root, so re-add it to the journal here.
+		 */
+		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
+						   left_path, 0);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_rotate_subtree_left(handle, et, left_path,
+						right_path, subtree_root,
+						dealloc, &deleted);
+		if (ret == -EAGAIN) {
+			/*
+			 * The rotation has to temporarily stop due to
+			 * the right subtree having an empty
+			 * extent. Pass it back to the caller for a
+			 * fixup.
+			 */
+			*empty_extent_path = right_path;
+			right_path = NULL;
+			goto out;
+		}
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * The subtree rotate might have removed records on
+		 * the rightmost edge. If so, then rotation is
+		 * complete.
+		 */
+		if (deleted)
+			break;
+
+		ocfs2_mv_path(left_path, right_path);
+
+		ret = ocfs2_find_cpos_for_right_leaf(sb, left_path,
+						     &right_cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+out:
+	ocfs2_free_path(right_path);
+	ocfs2_free_path(left_path);
+
+	return ret;
+}
+
+static int ocfs2_remove_rightmost_path(handle_t *handle,
+				struct ocfs2_extent_tree *et,
+				struct ocfs2_path *path,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret, subtree_index;
+	u32 cpos;
+	struct ocfs2_path *left_path = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *el;
+
+
+	ret = ocfs2_et_sanity_check(et);
+	if (ret)
+		goto out;
+	/*
+	 * There's two ways we handle this depending on
+	 * whether path is the only existing one.
+	 */
+	ret = ocfs2_extend_rotate_transaction(handle, 0,
+					      handle->h_buffer_credits,
+					      path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
+					    path, &cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (cpos) {
+		/*
+		 * We have a path to the left of this one - it needs
+		 * an update too.
+		 */
+		left_path = ocfs2_new_path_from_path(path);
+		if (!left_path) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_find_path(et->et_ci, left_path, cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		subtree_index = ocfs2_find_subtree_root(et, left_path, path);
+
+		ocfs2_unlink_subtree(handle, et, left_path, path,
+				     subtree_index, dealloc);
+		ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
+						left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
+		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
+	} else {
+		/*
+		 * 'path' is also the leftmost path which
+		 * means it must be the only one. This gets
+		 * handled differently because we want to
+		 * revert the root back to having extents
+		 * in-line.
+		 */
+		ocfs2_unlink_path(handle, et, dealloc, path, 1);
+
+		el = et->et_root_el;
+		el->l_tree_depth = 0;
+		el->l_next_free_rec = 0;
+		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
+
+		ocfs2_et_set_last_eb_blk(et, 0);
+	}
+
+	ocfs2_journal_dirty(handle, path_root_bh(path));
+
+out:
+	ocfs2_free_path(left_path);
+	return ret;
+}
+
+/*
+ * Left rotation of btree records.
+ *
+ * In many ways, this is (unsurprisingly) the opposite of right
+ * rotation. We start at some non-rightmost path containing an empty
+ * extent in the leaf block. The code works its way to the rightmost
+ * path by rotating records to the left in every subtree.
+ *
+ * This is used by any code which reduces the number of extent records
+ * in a leaf. After removal, an empty record should be placed in the
+ * leftmost list position.
+ *
+ * This won't handle a length update of the rightmost path records if
+ * the rightmost tree leaf record is removed so the caller is
+ * responsible for detecting and correcting that.
+ */
+static int ocfs2_rotate_tree_left(handle_t *handle,
+				  struct ocfs2_extent_tree *et,
+				  struct ocfs2_path *path,
+				  struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret, orig_credits = handle->h_buffer_credits;
+	struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *el;
+
+	el = path_leaf_el(path);
+	if (!ocfs2_is_empty_extent(&el->l_recs[0]))
+		return 0;
+
+	if (path->p_tree_depth == 0) {
+rightmost_no_delete:
+		/*
+		 * Inline extents. This is trivially handled, so do
+		 * it up front.
+		 */
+		ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path);
+		if (ret)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Handle rightmost branch now. There's several cases:
+	 *  1) simple rotation leaving records in there. That's trivial.
+	 *  2) rotation requiring a branch delete - there's no more
+	 *     records left. Two cases of this:
+	 *     a) There are branches to the left.
+	 *     b) This is also the leftmost (the only) branch.
+	 *
+	 *  1) is handled via ocfs2_rotate_rightmost_leaf_left()
+	 *  2a) we need the left branch so that we can update it with the unlink
+	 *  2b) we need to bring the root back to inline extents.
+	 */
+
+	eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
+	el = &eb->h_list;
+	if (eb->h_next_leaf_blk == 0) {
+		/*
+		 * This gets a bit tricky if we're going to delete the
+		 * rightmost path. Get the other cases out of the way
+		 * 1st.
+		 */
+		if (le16_to_cpu(el->l_next_free_rec) > 1)
+			goto rightmost_no_delete;
+
+		if (le16_to_cpu(el->l_next_free_rec) == 0) {
+			ret = -EIO;
+			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
+				    "Owner %llu has empty extent block at %llu",
+				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+				    (unsigned long long)le64_to_cpu(eb->h_blkno));
+			goto out;
+		}
+
+		/*
+		 * XXX: The caller can not trust "path" any more after
+		 * this as it will have been deleted. What do we do?
+		 *
+		 * In theory the rotate-for-merge code will never get
+		 * here because it'll always ask for a rotate in a
+		 * nonempty list.
+		 */
+
+		ret = ocfs2_remove_rightmost_path(handle, et, path,
+						  dealloc);
+		if (ret)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Now we can loop, remembering the path we get from -EAGAIN
+	 * and restarting from there.
+	 */
+try_rotate:
+	ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path,
+				       dealloc, &restart_path);
+	if (ret && ret != -EAGAIN) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	while (ret == -EAGAIN) {
+		tmp_path = restart_path;
+		restart_path = NULL;
+
+		ret = __ocfs2_rotate_tree_left(handle, et, orig_credits,
+					       tmp_path, dealloc,
+					       &restart_path);
+		if (ret && ret != -EAGAIN) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ocfs2_free_path(tmp_path);
+		tmp_path = NULL;
+
+		if (ret == 0)
+			goto try_rotate;
+	}
+
+out:
+	ocfs2_free_path(tmp_path);
+	ocfs2_free_path(restart_path);
+	return ret;
+}
+
+static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el,
+				int index)
+{
+	struct ocfs2_extent_rec *rec = &el->l_recs[index];
+	unsigned int size;
+
+	if (rec->e_leaf_clusters == 0) {
+		/*
+		 * We consumed all of the merged-from record. An empty
+		 * extent cannot exist anywhere but the 1st array
+		 * position, so move things over if the merged-from
+		 * record doesn't occupy that position.
+		 *
+		 * This creates a new empty extent so the caller
+		 * should be smart enough to have removed any existing
+		 * ones.
+		 */
+		if (index > 0) {
+			BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
+			size = index * sizeof(struct ocfs2_extent_rec);
+			memmove(&el->l_recs[1], &el->l_recs[0], size);
+		}
+
+		/*
+		 * Always memset - the caller doesn't check whether it
+		 * created an empty extent, so there could be junk in
+		 * the other fields.
+		 */
+		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
+	}
+}
+
+static int ocfs2_get_right_path(struct ocfs2_extent_tree *et,
+				struct ocfs2_path *left_path,
+				struct ocfs2_path **ret_right_path)
+{
+	int ret;
+	u32 right_cpos;
+	struct ocfs2_path *right_path = NULL;
+	struct ocfs2_extent_list *left_el;
+
+	*ret_right_path = NULL;
+
+	/* This function shouldn't be called for non-trees. */
+	BUG_ON(left_path->p_tree_depth == 0);
+
+	left_el = path_leaf_el(left_path);
+	BUG_ON(left_el->l_next_free_rec != left_el->l_count);
+
+	ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
+					     left_path, &right_cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* This function shouldn't be called for the rightmost leaf. */
+	BUG_ON(right_cpos == 0);
+
+	right_path = ocfs2_new_path_from_path(left_path);
+	if (!right_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	*ret_right_path = right_path;
+out:
+	if (ret)
+		ocfs2_free_path(right_path);
+	return ret;
+}
+
+/*
+ * Remove split_rec clusters from the record at index and merge them
+ * onto the beginning of the record "next" to it.
+ * For index < l_count - 1, the next means the extent rec at index + 1.
+ * For index == l_count - 1, the "next" means the 1st extent rec of the
+ * next extent block.
+ */
+static int ocfs2_merge_rec_right(struct ocfs2_path *left_path,
+				 handle_t *handle,
+				 struct ocfs2_extent_tree *et,
+				 struct ocfs2_extent_rec *split_rec,
+				 int index)
+{
+	int ret, next_free, i;
+	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
+	struct ocfs2_extent_rec *left_rec;
+	struct ocfs2_extent_rec *right_rec;
+	struct ocfs2_extent_list *right_el;
+	struct ocfs2_path *right_path = NULL;
+	int subtree_index = 0;
+	struct ocfs2_extent_list *el = path_leaf_el(left_path);
+	struct buffer_head *bh = path_leaf_bh(left_path);
+	struct buffer_head *root_bh = NULL;
+
+	BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
+	left_rec = &el->l_recs[index];
+
+	if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
+	    le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
+		/* we meet with a cross extent block merge. */
+		ret = ocfs2_get_right_path(et, left_path, &right_path);
+		if (ret) {
+			mlog_errno(ret);
+			return ret;
+		}
+
+		right_el = path_leaf_el(right_path);
+		next_free = le16_to_cpu(right_el->l_next_free_rec);
+		BUG_ON(next_free <= 0);
+		right_rec = &right_el->l_recs[0];
+		if (ocfs2_is_empty_extent(right_rec)) {
+			BUG_ON(next_free <= 1);
+			right_rec = &right_el->l_recs[1];
+		}
+
+		BUG_ON(le32_to_cpu(left_rec->e_cpos) +
+		       le16_to_cpu(left_rec->e_leaf_clusters) !=
+		       le32_to_cpu(right_rec->e_cpos));
+
+		subtree_index = ocfs2_find_subtree_root(et, left_path,
+							right_path);
+
+		ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
+						      handle->h_buffer_credits,
+						      right_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		root_bh = left_path->p_node[subtree_index].bh;
+		BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
+
+		ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
+						   subtree_index);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		for (i = subtree_index + 1;
+		     i < path_num_items(right_path); i++) {
+			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
+							   right_path, i);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
+							   left_path, i);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+	} else {
+		BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1);
+		right_rec = &el->l_recs[index + 1];
+	}
+
+	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path,
+					   path_num_items(left_path) - 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters);
+
+	le32_add_cpu(&right_rec->e_cpos, -split_clusters);
+	le64_add_cpu(&right_rec->e_blkno,
+		     -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
+					       split_clusters));
+	le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);
+
+	ocfs2_cleanup_merge(el, index);
+
+	ocfs2_journal_dirty(handle, bh);
+	if (right_path) {
+		ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
+		ocfs2_complete_edge_insert(handle, left_path, right_path,
+					   subtree_index);
+	}
+out:
+	ocfs2_free_path(right_path);
+	return ret;
+}
+
+static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
+			       struct ocfs2_path *right_path,
+			       struct ocfs2_path **ret_left_path)
+{
+	int ret;
+	u32 left_cpos;
+	struct ocfs2_path *left_path = NULL;
+
+	*ret_left_path = NULL;
+
+	/* This function shouldn't be called for non-trees. */
+	BUG_ON(right_path->p_tree_depth == 0);
+
+	ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
+					    right_path, &left_cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* This function shouldn't be called for the leftmost leaf. */
+	BUG_ON(left_cpos == 0);
+
+	left_path = ocfs2_new_path_from_path(right_path);
+	if (!left_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	*ret_left_path = left_path;
+out:
+	if (ret)
+		ocfs2_free_path(left_path);
+	return ret;
+}
+
+/*
+ * Remove split_rec clusters from the record at index and merge them
+ * onto the tail of the record "before" it.
+ * For index > 0, the "before" means the extent rec at index - 1.
+ *
+ * For index == 0, the "before" means the last record of the previous
+ * extent block. And there is also a situation that we may need to
+ * remove the rightmost leaf extent block in the right_path and change
+ * the right path to indicate the new rightmost path.
+ */
+static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
+				handle_t *handle,
+				struct ocfs2_extent_tree *et,
+				struct ocfs2_extent_rec *split_rec,
+				struct ocfs2_cached_dealloc_ctxt *dealloc,
+				int index)
+{
+	int ret, i, subtree_index = 0, has_empty_extent = 0;
+	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
+	struct ocfs2_extent_rec *left_rec;
+	struct ocfs2_extent_rec *right_rec;
+	struct ocfs2_extent_list *el = path_leaf_el(right_path);
+	struct buffer_head *bh = path_leaf_bh(right_path);
+	struct buffer_head *root_bh = NULL;
+	struct ocfs2_path *left_path = NULL;
+	struct ocfs2_extent_list *left_el;
+
+	BUG_ON(index < 0);
+
+	right_rec = &el->l_recs[index];
+	if (index == 0) {
+		/* we meet with a cross extent block merge. */
+		ret = ocfs2_get_left_path(et, right_path, &left_path);
+		if (ret) {
+			mlog_errno(ret);
+			return ret;
+		}
+
+		left_el = path_leaf_el(left_path);
+		BUG_ON(le16_to_cpu(left_el->l_next_free_rec) !=
+		       le16_to_cpu(left_el->l_count));
+
+		left_rec = &left_el->l_recs[
+				le16_to_cpu(left_el->l_next_free_rec) - 1];
+		BUG_ON(le32_to_cpu(left_rec->e_cpos) +
+		       le16_to_cpu(left_rec->e_leaf_clusters) !=
+		       le32_to_cpu(split_rec->e_cpos));
+
+		subtree_index = ocfs2_find_subtree_root(et, left_path,
+							right_path);
+
+		ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
+						      handle->h_buffer_credits,
+						      left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		root_bh = left_path->p_node[subtree_index].bh;
+		BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
+
+		ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
+						   subtree_index);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		for (i = subtree_index + 1;
+		     i < path_num_items(right_path); i++) {
+			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
+							   right_path, i);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
+							   left_path, i);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+	} else {
+		left_rec = &el->l_recs[index - 1];
+		if (ocfs2_is_empty_extent(&el->l_recs[0]))
+			has_empty_extent = 1;
+	}
+
+	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
+					   path_num_items(right_path) - 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (has_empty_extent && index == 1) {
+		/*
+		 * The easy case - we can just plop the record right in.
+		 */
+		*left_rec = *split_rec;
+
+		has_empty_extent = 0;
+	} else
+		le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);
+
+	le32_add_cpu(&right_rec->e_cpos, split_clusters);
+	le64_add_cpu(&right_rec->e_blkno,
+		     ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
+					      split_clusters));
+	le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);
+
+	ocfs2_cleanup_merge(el, index);
+
+	ocfs2_journal_dirty(handle, bh);
+	if (left_path) {
+		ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
+
+		/*
+		 * In the situation that the right_rec is empty and the extent
+		 * block is empty also,  ocfs2_complete_edge_insert can't handle
+		 * it and we need to delete the right extent block.
+		 */
+		if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 &&
+		    le16_to_cpu(el->l_next_free_rec) == 1) {
+
+			ret = ocfs2_remove_rightmost_path(handle, et,
+							  right_path,
+							  dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			/* Now the rightmost extent block has been deleted.
+			 * So we use the new rightmost path.
+			 */
+			ocfs2_mv_path(right_path, left_path);
+			left_path = NULL;
+		} else
+			ocfs2_complete_edge_insert(handle, left_path,
+						   right_path, subtree_index);
+	}
+out:
+	ocfs2_free_path(left_path);
+	return ret;
+}
+
+static int ocfs2_try_to_merge_extent(handle_t *handle,
+				     struct ocfs2_extent_tree *et,
+				     struct ocfs2_path *path,
+				     int split_index,
+				     struct ocfs2_extent_rec *split_rec,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc,
+				     struct ocfs2_merge_ctxt *ctxt)
+{
+	int ret = 0;
+	struct ocfs2_extent_list *el = path_leaf_el(path);
+	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
+
+	BUG_ON(ctxt->c_contig_type == CONTIG_NONE);
+
+	if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) {
+		/*
+		 * The merge code will need to create an empty
+		 * extent to take the place of the newly
+		 * emptied slot. Remove any pre-existing empty
+		 * extents - having more than one in a leaf is
+		 * illegal.
+		 */
+		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		split_index--;
+		rec = &el->l_recs[split_index];
+	}
+
+	if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
+		/*
+		 * Left-right contig implies this.
+		 */
+		BUG_ON(!ctxt->c_split_covers_rec);
+
+		/*
+		 * Since the leftright insert always covers the entire
+		 * extent, this call will delete the insert record
+		 * entirely, resulting in an empty extent record added to
+		 * the extent block.
+		 *
+		 * Since the adding of an empty extent shifts
+		 * everything back to the right, there's no need to
+		 * update split_index here.
+		 *
+		 * When the split_index is zero, we need to merge it to the
+		 * prevoius extent block. It is more efficient and easier
+		 * if we do merge_right first and merge_left later.
+		 */
+		ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
+					    split_index);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * We can only get this from logic error above.
+		 */
+		BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
+
+		/* The merge left us with an empty extent, remove it. */
+		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		rec = &el->l_recs[split_index];
+
+		/*
+		 * Note that we don't pass split_rec here on purpose -
+		 * we've merged it into the rec already.
+		 */
+		ret = ocfs2_merge_rec_left(path, handle, et, rec,
+					   dealloc, split_index);
+
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
+		/*
+		 * Error from this last rotate is not critical, so
+		 * print but don't bubble it up.
+		 */
+		if (ret)
+			mlog_errno(ret);
+		ret = 0;
+	} else {
+		/*
+		 * Merge a record to the left or right.
+		 *
+		 * 'contig_type' is relative to the existing record,
+		 * so for example, if we're "right contig", it's to
+		 * the record on the left (hence the left merge).
+		 */
+		if (ctxt->c_contig_type == CONTIG_RIGHT) {
+			ret = ocfs2_merge_rec_left(path, handle, et,
+						   split_rec, dealloc,
+						   split_index);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		} else {
+			ret = ocfs2_merge_rec_right(path, handle,
+						    et, split_rec,
+						    split_index);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		if (ctxt->c_split_covers_rec) {
+			/*
+			 * The merge may have left an empty extent in
+			 * our leaf. Try to rotate it away.
+			 */
+			ret = ocfs2_rotate_tree_left(handle, et, path,
+						     dealloc);
+			if (ret)
+				mlog_errno(ret);
+			ret = 0;
+		}
+	}
+
+out:
+	return ret;
+}
+
+static void ocfs2_subtract_from_rec(struct super_block *sb,
+				    enum ocfs2_split_type split,
+				    struct ocfs2_extent_rec *rec,
+				    struct ocfs2_extent_rec *split_rec)
+{
+	u64 len_blocks;
+
+	len_blocks = ocfs2_clusters_to_blocks(sb,
+				le16_to_cpu(split_rec->e_leaf_clusters));
+
+	if (split == SPLIT_LEFT) {
+		/*
+		 * Region is on the left edge of the existing
+		 * record.
+		 */
+		le32_add_cpu(&rec->e_cpos,
+			     le16_to_cpu(split_rec->e_leaf_clusters));
+		le64_add_cpu(&rec->e_blkno, len_blocks);
+		le16_add_cpu(&rec->e_leaf_clusters,
+			     -le16_to_cpu(split_rec->e_leaf_clusters));
+	} else {
+		/*
+		 * Region is on the right edge of the existing
+		 * record.
+		 */
+		le16_add_cpu(&rec->e_leaf_clusters,
+			     -le16_to_cpu(split_rec->e_leaf_clusters));
+	}
+}
+
+/*
+ * Do the final bits of extent record insertion at the target leaf
+ * list. If this leaf is part of an allocation tree, it is assumed
+ * that the tree above has been prepared.
+ */
+static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
+				 struct ocfs2_extent_rec *insert_rec,
+				 struct ocfs2_extent_list *el,
+				 struct ocfs2_insert_type *insert)
+{
+	int i = insert->ins_contig_index;
+	unsigned int range;
+	struct ocfs2_extent_rec *rec;
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	if (insert->ins_split != SPLIT_NONE) {
+		i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
+		BUG_ON(i == -1);
+		rec = &el->l_recs[i];
+		ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
+					insert->ins_split, rec,
+					insert_rec);
+		goto rotate;
+	}
+
+	/*
+	 * Contiguous insert - either left or right.
+	 */
+	if (insert->ins_contig != CONTIG_NONE) {
+		rec = &el->l_recs[i];
+		if (insert->ins_contig == CONTIG_LEFT) {
+			rec->e_blkno = insert_rec->e_blkno;
+			rec->e_cpos = insert_rec->e_cpos;
+		}
+		le16_add_cpu(&rec->e_leaf_clusters,
+			     le16_to_cpu(insert_rec->e_leaf_clusters));
+		return;
+	}
+
+	/*
+	 * Handle insert into an empty leaf.
+	 */
+	if (le16_to_cpu(el->l_next_free_rec) == 0 ||
+	    ((le16_to_cpu(el->l_next_free_rec) == 1) &&
+	     ocfs2_is_empty_extent(&el->l_recs[0]))) {
+		el->l_recs[0] = *insert_rec;
+		el->l_next_free_rec = cpu_to_le16(1);
+		return;
+	}
+
+	/*
+	 * Appending insert.
+	 */
+	if (insert->ins_appending == APPEND_TAIL) {
+		i = le16_to_cpu(el->l_next_free_rec) - 1;
+		rec = &el->l_recs[i];
+		range = le32_to_cpu(rec->e_cpos)
+			+ le16_to_cpu(rec->e_leaf_clusters);
+		BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
+
+		mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
+				le16_to_cpu(el->l_count),
+				"owner %llu, depth %u, count %u, next free %u, "
+				"rec.cpos %u, rec.clusters %u, "
+				"insert.cpos %u, insert.clusters %u\n",
+				ocfs2_metadata_cache_owner(et->et_ci),
+				le16_to_cpu(el->l_tree_depth),
+				le16_to_cpu(el->l_count),
+				le16_to_cpu(el->l_next_free_rec),
+				le32_to_cpu(el->l_recs[i].e_cpos),
+				le16_to_cpu(el->l_recs[i].e_leaf_clusters),
+				le32_to_cpu(insert_rec->e_cpos),
+				le16_to_cpu(insert_rec->e_leaf_clusters));
+		i++;
+		el->l_recs[i] = *insert_rec;
+		le16_add_cpu(&el->l_next_free_rec, 1);
+		return;
+	}
+
+rotate:
+	/*
+	 * Ok, we have to rotate.
+	 *
+	 * At this point, it is safe to assume that inserting into an
+	 * empty leaf and appending to a leaf have both been handled
+	 * above.
+	 *
+	 * This leaf needs to have space, either by the empty 1st
+	 * extent record, or by virtue of an l_next_rec < l_count.
+	 */
+	ocfs2_rotate_leaf(el, insert_rec);
+}
+
+static void ocfs2_adjust_rightmost_records(handle_t *handle,
+					   struct ocfs2_extent_tree *et,
+					   struct ocfs2_path *path,
+					   struct ocfs2_extent_rec *insert_rec)
+{
+	int ret, i, next_free;
+	struct buffer_head *bh;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+
+	/*
+	 * Update everything except the leaf block.
+	 */
+	for (i = 0; i < path->p_tree_depth; i++) {
+		bh = path->p_node[i].bh;
+		el = path->p_node[i].el;
+
+		next_free = le16_to_cpu(el->l_next_free_rec);
+		if (next_free == 0) {
+			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
+				    "Owner %llu has a bad extent list",
+				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
+			ret = -EIO;
+			return;
+		}
+
+		rec = &el->l_recs[next_free - 1];
+
+		rec->e_int_clusters = insert_rec->e_cpos;
+		le32_add_cpu(&rec->e_int_clusters,
+			     le16_to_cpu(insert_rec->e_leaf_clusters));
+		le32_add_cpu(&rec->e_int_clusters,
+			     -le32_to_cpu(rec->e_cpos));
+
+		ocfs2_journal_dirty(handle, bh);
+	}
+}
+
+static int ocfs2_append_rec_to_path(handle_t *handle,
+				    struct ocfs2_extent_tree *et,
+				    struct ocfs2_extent_rec *insert_rec,
+				    struct ocfs2_path *right_path,
+				    struct ocfs2_path **ret_left_path)
+{
+	int ret, next_free;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_path *left_path = NULL;
+
+	*ret_left_path = NULL;
+
+	/*
+	 * This shouldn't happen for non-trees. The extent rec cluster
+	 * count manipulation below only works for interior nodes.
+	 */
+	BUG_ON(right_path->p_tree_depth == 0);
+
+	/*
+	 * If our appending insert is at the leftmost edge of a leaf,
+	 * then we might need to update the rightmost records of the
+	 * neighboring path.
+	 */
+	el = path_leaf_el(right_path);
+	next_free = le16_to_cpu(el->l_next_free_rec);
+	if (next_free == 0 ||
+	    (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
+		u32 left_cpos;
+
+		ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
+						    right_path, &left_cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		trace_ocfs2_append_rec_to_path(
+			(unsigned long long)
+			ocfs2_metadata_cache_owner(et->et_ci),
+			le32_to_cpu(insert_rec->e_cpos),
+			left_cpos);
+
+		/*
+		 * No need to worry if the append is already in the
+		 * leftmost leaf.
+		 */
+		if (left_cpos) {
+			left_path = ocfs2_new_path_from_path(right_path);
+			if (!left_path) {
+				ret = -ENOMEM;
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ret = ocfs2_find_path(et->et_ci, left_path,
+					      left_cpos);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			/*
+			 * ocfs2_insert_path() will pass the left_path to the
+			 * journal for us.
+			 */
+		}
+	}
+
+	ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
+
+	*ret_left_path = left_path;
+	ret = 0;
+out:
+	if (ret != 0)
+		ocfs2_free_path(left_path);
+
+	return ret;
+}
+
+static void ocfs2_split_record(struct ocfs2_extent_tree *et,
+			       struct ocfs2_path *left_path,
+			       struct ocfs2_path *right_path,
+			       struct ocfs2_extent_rec *split_rec,
+			       enum ocfs2_split_type split)
+{
+	int index;
+	u32 cpos = le32_to_cpu(split_rec->e_cpos);
+	struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
+	struct ocfs2_extent_rec *rec, *tmprec;
+
+	right_el = path_leaf_el(right_path);
+	if (left_path)
+		left_el = path_leaf_el(left_path);
+
+	el = right_el;
+	insert_el = right_el;
+	index = ocfs2_search_extent_list(el, cpos);
+	if (index != -1) {
+		if (index == 0 && left_path) {
+			BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
+
+			/*
+			 * This typically means that the record
+			 * started in the left path but moved to the
+			 * right as a result of rotation. We either
+			 * move the existing record to the left, or we
+			 * do the later insert there.
+			 *
+			 * In this case, the left path should always
+			 * exist as the rotate code will have passed
+			 * it back for a post-insert update.
+			 */
+
+			if (split == SPLIT_LEFT) {
+				/*
+				 * It's a left split. Since we know
+				 * that the rotate code gave us an
+				 * empty extent in the left path, we
+				 * can just do the insert there.
+				 */
+				insert_el = left_el;
+			} else {
+				/*
+				 * Right split - we have to move the
+				 * existing record over to the left
+				 * leaf. The insert will be into the
+				 * newly created empty extent in the
+				 * right leaf.
+				 */
+				tmprec = &right_el->l_recs[index];
+				ocfs2_rotate_leaf(left_el, tmprec);
+				el = left_el;
+
+				memset(tmprec, 0, sizeof(*tmprec));
+				index = ocfs2_search_extent_list(left_el, cpos);
+				BUG_ON(index == -1);
+			}
+		}
+	} else {
+		BUG_ON(!left_path);
+		BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
+		/*
+		 * Left path is easy - we can just allow the insert to
+		 * happen.
+		 */
+		el = left_el;
+		insert_el = left_el;
+		index = ocfs2_search_extent_list(el, cpos);
+		BUG_ON(index == -1);
+	}
+
+	rec = &el->l_recs[index];
+	ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
+				split, rec, split_rec);
+	ocfs2_rotate_leaf(insert_el, split_rec);
+}
+
+/*
+ * This function only does inserts on an allocation b-tree. For tree
+ * depth = 0, ocfs2_insert_at_leaf() is called directly.
+ *
+ * right_path is the path we want to do the actual insert
+ * in. left_path should only be passed in if we need to update that
+ * portion of the tree after an edge insert.
+ */
+static int ocfs2_insert_path(handle_t *handle,
+			     struct ocfs2_extent_tree *et,
+			     struct ocfs2_path *left_path,
+			     struct ocfs2_path *right_path,
+			     struct ocfs2_extent_rec *insert_rec,
+			     struct ocfs2_insert_type *insert)
+{
+	int ret, subtree_index;
+	struct buffer_head *leaf_bh = path_leaf_bh(right_path);
+
+	if (left_path) {
+		/*
+		 * There's a chance that left_path got passed back to
+		 * us without being accounted for in the
+		 * journal. Extend our transaction here to be sure we
+		 * can change those blocks.
+		 */
+		ret = ocfs2_extend_trans(handle, left_path->p_tree_depth);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	/*
+	 * Pass both paths to the journal. The majority of inserts
+	 * will be touching all components anyway.
+	 */
+	ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (insert->ins_split != SPLIT_NONE) {
+		/*
+		 * We could call ocfs2_insert_at_leaf() for some types
+		 * of splits, but it's easier to just let one separate
+		 * function sort it all out.
+		 */
+		ocfs2_split_record(et, left_path, right_path,
+				   insert_rec, insert->ins_split);
+
+		/*
+		 * Split might have modified either leaf and we don't
+		 * have a guarantee that the later edge insert will
+		 * dirty this for us.
+		 */
+		if (left_path)
+			ocfs2_journal_dirty(handle,
+					    path_leaf_bh(left_path));
+	} else
+		ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
+				     insert);
+
+	ocfs2_journal_dirty(handle, leaf_bh);
+
+	if (left_path) {
+		/*
+		 * The rotate code has indicated that we need to fix
+		 * up portions of the tree after the insert.
+		 *
+		 * XXX: Should we extend the transaction here?
+		 */
+		subtree_index = ocfs2_find_subtree_root(et, left_path,
+							right_path);
+		ocfs2_complete_edge_insert(handle, left_path, right_path,
+					   subtree_index);
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static int ocfs2_do_insert_extent(handle_t *handle,
+				  struct ocfs2_extent_tree *et,
+				  struct ocfs2_extent_rec *insert_rec,
+				  struct ocfs2_insert_type *type)
+{
+	int ret, rotate = 0;
+	u32 cpos;
+	struct ocfs2_path *right_path = NULL;
+	struct ocfs2_path *left_path = NULL;
+	struct ocfs2_extent_list *el;
+
+	el = et->et_root_el;
+
+	ret = ocfs2_et_root_journal_access(handle, et,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (le16_to_cpu(el->l_tree_depth) == 0) {
+		ocfs2_insert_at_leaf(et, insert_rec, el, type);
+		goto out_update_clusters;
+	}
+
+	right_path = ocfs2_new_path_from_et(et);
+	if (!right_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Determine the path to start with. Rotations need the
+	 * rightmost path, everything else can go directly to the
+	 * target leaf.
+	 */
+	cpos = le32_to_cpu(insert_rec->e_cpos);
+	if (type->ins_appending == APPEND_NONE &&
+	    type->ins_contig == CONTIG_NONE) {
+		rotate = 1;
+		cpos = UINT_MAX;
+	}
+
+	ret = ocfs2_find_path(et->et_ci, right_path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Rotations and appends need special treatment - they modify
+	 * parts of the tree's above them.
+	 *
+	 * Both might pass back a path immediate to the left of the
+	 * one being inserted to. This will be cause
+	 * ocfs2_insert_path() to modify the rightmost records of
+	 * left_path to account for an edge insert.
+	 *
+	 * XXX: When modifying this code, keep in mind that an insert
+	 * can wind up skipping both of these two special cases...
+	 */
+	if (rotate) {
+		ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
+					      le32_to_cpu(insert_rec->e_cpos),
+					      right_path, &left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * ocfs2_rotate_tree_right() might have extended the
+		 * transaction without re-journaling our tree root.
+		 */
+		ret = ocfs2_et_root_journal_access(handle, et,
+						   OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	} else if (type->ins_appending == APPEND_TAIL
+		   && type->ins_contig != CONTIG_LEFT) {
+		ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
+					       right_path, &left_path);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_insert_path(handle, et, left_path, right_path,
+				insert_rec, type);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+out_update_clusters:
+	if (type->ins_split == SPLIT_NONE)
+		ocfs2_et_update_clusters(et,
+					 le16_to_cpu(insert_rec->e_leaf_clusters));
+
+	ocfs2_journal_dirty(handle, et->et_root_bh);
+
+out:
+	ocfs2_free_path(left_path);
+	ocfs2_free_path(right_path);
+
+	return ret;
+}
+
+static enum ocfs2_contig_type
+ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
+			       struct ocfs2_path *path,
+			       struct ocfs2_extent_list *el, int index,
+			       struct ocfs2_extent_rec *split_rec)
+{
+	int status;
+	enum ocfs2_contig_type ret = CONTIG_NONE;
+	u32 left_cpos, right_cpos;
+	struct ocfs2_extent_rec *rec = NULL;
+	struct ocfs2_extent_list *new_el;
+	struct ocfs2_path *left_path = NULL, *right_path = NULL;
+	struct buffer_head *bh;
+	struct ocfs2_extent_block *eb;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
+
+	if (index > 0) {
+		rec = &el->l_recs[index - 1];
+	} else if (path->p_tree_depth > 0) {
+		status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
+		if (status)
+			goto exit;
+
+		if (left_cpos != 0) {
+			left_path = ocfs2_new_path_from_path(path);
+			if (!left_path)
+				goto exit;
+
+			status = ocfs2_find_path(et->et_ci, left_path,
+						 left_cpos);
+			if (status)
+				goto free_left_path;
+
+			new_el = path_leaf_el(left_path);
+
+			if (le16_to_cpu(new_el->l_next_free_rec) !=
+			    le16_to_cpu(new_el->l_count)) {
+				bh = path_leaf_bh(left_path);
+				eb = (struct ocfs2_extent_block *)bh->b_data;
+				ocfs2_error(sb,
+					    "Extent block #%llu has an "
+					    "invalid l_next_free_rec of "
+					    "%d.  It should have "
+					    "matched the l_count of %d",
+					    (unsigned long long)le64_to_cpu(eb->h_blkno),
+					    le16_to_cpu(new_el->l_next_free_rec),
+					    le16_to_cpu(new_el->l_count));
+				status = -EINVAL;
+				goto free_left_path;
+			}
+			rec = &new_el->l_recs[
+				le16_to_cpu(new_el->l_next_free_rec) - 1];
+		}
+	}
+
+	/*
+	 * We're careful to check for an empty extent record here -
+	 * the merge code will know what to do if it sees one.
+	 */
+	if (rec) {
+		if (index == 1 && ocfs2_is_empty_extent(rec)) {
+			if (split_rec->e_cpos == el->l_recs[index].e_cpos)
+				ret = CONTIG_RIGHT;
+		} else {
+			ret = ocfs2_et_extent_contig(et, rec, split_rec);
+		}
+	}
+
+	rec = NULL;
+	if (index < (le16_to_cpu(el->l_next_free_rec) - 1))
+		rec = &el->l_recs[index + 1];
+	else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) &&
+		 path->p_tree_depth > 0) {
+		status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
+		if (status)
+			goto free_left_path;
+
+		if (right_cpos == 0)
+			goto free_left_path;
+
+		right_path = ocfs2_new_path_from_path(path);
+		if (!right_path)
+			goto free_left_path;
+
+		status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
+		if (status)
+			goto free_right_path;
+
+		new_el = path_leaf_el(right_path);
+		rec = &new_el->l_recs[0];
+		if (ocfs2_is_empty_extent(rec)) {
+			if (le16_to_cpu(new_el->l_next_free_rec) <= 1) {
+				bh = path_leaf_bh(right_path);
+				eb = (struct ocfs2_extent_block *)bh->b_data;
+				ocfs2_error(sb,
+					    "Extent block #%llu has an "
+					    "invalid l_next_free_rec of %d",
+					    (unsigned long long)le64_to_cpu(eb->h_blkno),
+					    le16_to_cpu(new_el->l_next_free_rec));
+				status = -EINVAL;
+				goto free_right_path;
+			}
+			rec = &new_el->l_recs[1];
+		}
+	}
+
+	if (rec) {
+		enum ocfs2_contig_type contig_type;
+
+		contig_type = ocfs2_et_extent_contig(et, rec, split_rec);
+
+		if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
+			ret = CONTIG_LEFTRIGHT;
+		else if (ret == CONTIG_NONE)
+			ret = contig_type;
+	}
+
+free_right_path:
+	ocfs2_free_path(right_path);
+free_left_path:
+	ocfs2_free_path(left_path);
+exit:
+	return ret;
+}
+
+static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et,
+				     struct ocfs2_insert_type *insert,
+				     struct ocfs2_extent_list *el,
+				     struct ocfs2_extent_rec *insert_rec)
+{
+	int i;
+	enum ocfs2_contig_type contig_type = CONTIG_NONE;
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i],
+						     insert_rec);
+		if (contig_type != CONTIG_NONE) {
+			insert->ins_contig_index = i;
+			break;
+		}
+	}
+	insert->ins_contig = contig_type;
+
+	if (insert->ins_contig != CONTIG_NONE) {
+		struct ocfs2_extent_rec *rec =
+				&el->l_recs[insert->ins_contig_index];
+		unsigned int len = le16_to_cpu(rec->e_leaf_clusters) +
+				   le16_to_cpu(insert_rec->e_leaf_clusters);
+
+		/*
+		 * Caller might want us to limit the size of extents, don't
+		 * calculate contiguousness if we might exceed that limit.
+		 */
+		if (et->et_max_leaf_clusters &&
+		    (len > et->et_max_leaf_clusters))
+			insert->ins_contig = CONTIG_NONE;
+	}
+}
+
+/*
+ * This should only be called against the righmost leaf extent list.
+ *
+ * ocfs2_figure_appending_type() will figure out whether we'll have to
+ * insert at the tail of the rightmost leaf.
+ *
+ * This should also work against the root extent list for tree's with 0
+ * depth. If we consider the root extent list to be the rightmost leaf node
+ * then the logic here makes sense.
+ */
+static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
+					struct ocfs2_extent_list *el,
+					struct ocfs2_extent_rec *insert_rec)
+{
+	int i;
+	u32 cpos = le32_to_cpu(insert_rec->e_cpos);
+	struct ocfs2_extent_rec *rec;
+
+	insert->ins_appending = APPEND_NONE;
+
+	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
+
+	if (!el->l_next_free_rec)
+		goto set_tail_append;
+
+	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
+		/* Were all records empty? */
+		if (le16_to_cpu(el->l_next_free_rec) == 1)
+			goto set_tail_append;
+	}
+
+	i = le16_to_cpu(el->l_next_free_rec) - 1;
+	rec = &el->l_recs[i];
+
+	if (cpos >=
+	    (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
+		goto set_tail_append;
+
+	return;
+
+set_tail_append:
+	insert->ins_appending = APPEND_TAIL;
+}
+
+/*
+ * Helper function called at the beginning of an insert.
+ *
+ * This computes a few things that are commonly used in the process of
+ * inserting into the btree:
+ *   - Whether the new extent is contiguous with an existing one.
+ *   - The current tree depth.
+ *   - Whether the insert is an appending one.
+ *   - The total # of free records in the tree.
+ *
+ * All of the information is stored on the ocfs2_insert_type
+ * structure.
+ */
+static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et,
+				    struct buffer_head **last_eb_bh,
+				    struct ocfs2_extent_rec *insert_rec,
+				    int *free_records,
+				    struct ocfs2_insert_type *insert)
+{
+	int ret;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_path *path = NULL;
+	struct buffer_head *bh = NULL;
+
+	insert->ins_split = SPLIT_NONE;
+
+	el = et->et_root_el;
+	insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
+
+	if (el->l_tree_depth) {
+		/*
+		 * If we have tree depth, we read in the
+		 * rightmost extent block ahead of time as
+		 * ocfs2_figure_insert_type() and ocfs2_add_branch()
+		 * may want it later.
+		 */
+		ret = ocfs2_read_extent_block(et->et_ci,
+					      ocfs2_et_get_last_eb_blk(et),
+					      &bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		eb = (struct ocfs2_extent_block *) bh->b_data;
+		el = &eb->h_list;
+	}
+
+	/*
+	 * Unless we have a contiguous insert, we'll need to know if
+	 * there is room left in our allocation tree for another
+	 * extent record.
+	 *
+	 * XXX: This test is simplistic, we can search for empty
+	 * extent records too.
+	 */
+	*free_records = le16_to_cpu(el->l_count) -
+		le16_to_cpu(el->l_next_free_rec);
+
+	if (!insert->ins_tree_depth) {
+		ocfs2_figure_contig_type(et, insert, el, insert_rec);
+		ocfs2_figure_appending_type(insert, el, insert_rec);
+		return 0;
+	}
+
+	path = ocfs2_new_path_from_et(et);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * In the case that we're inserting past what the tree
+	 * currently accounts for, ocfs2_find_path() will return for
+	 * us the rightmost tree path. This is accounted for below in
+	 * the appending code.
+	 */
+	ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	el = path_leaf_el(path);
+
+	/*
+	 * Now that we have the path, there's two things we want to determine:
+	 * 1) Contiguousness (also set contig_index if this is so)
+	 *
+	 * 2) Are we doing an append? We can trivially break this up
+         *     into two types of appends: simple record append, or a
+         *     rotate inside the tail leaf.
+	 */
+	ocfs2_figure_contig_type(et, insert, el, insert_rec);
+
+	/*
+	 * The insert code isn't quite ready to deal with all cases of
+	 * left contiguousness. Specifically, if it's an insert into
+	 * the 1st record in a leaf, it will require the adjustment of
+	 * cluster count on the last record of the path directly to it's
+	 * left. For now, just catch that case and fool the layers
+	 * above us. This works just fine for tree_depth == 0, which
+	 * is why we allow that above.
+	 */
+	if (insert->ins_contig == CONTIG_LEFT &&
+	    insert->ins_contig_index == 0)
+		insert->ins_contig = CONTIG_NONE;
+
+	/*
+	 * Ok, so we can simply compare against last_eb to figure out
+	 * whether the path doesn't exist. This will only happen in
+	 * the case that we're doing a tail append, so maybe we can
+	 * take advantage of that information somehow.
+	 */
+	if (ocfs2_et_get_last_eb_blk(et) ==
+	    path_leaf_bh(path)->b_blocknr) {
+		/*
+		 * Ok, ocfs2_find_path() returned us the rightmost
+		 * tree path. This might be an appending insert. There are
+		 * two cases:
+		 *    1) We're doing a true append at the tail:
+		 *	-This might even be off the end of the leaf
+		 *    2) We're "appending" by rotating in the tail
+		 */
+		ocfs2_figure_appending_type(insert, el, insert_rec);
+	}
+
+out:
+	ocfs2_free_path(path);
+
+	if (ret == 0)
+		*last_eb_bh = bh;
+	else
+		brelse(bh);
+	return ret;
+}
+
+/*
+ * Insert an extent into a btree.
+ *
+ * The caller needs to update the owning btree's cluster count.
+ */
+int ocfs2_insert_extent(handle_t *handle,
+			struct ocfs2_extent_tree *et,
+			u32 cpos,
+			u64 start_blk,
+			u32 new_clusters,
+			u8 flags,
+			struct ocfs2_alloc_context *meta_ac)
+{
+	int status;
+	int uninitialized_var(free_records);
+	struct buffer_head *last_eb_bh = NULL;
+	struct ocfs2_insert_type insert = {0, };
+	struct ocfs2_extent_rec rec;
+
+	trace_ocfs2_insert_extent_start(
+		(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+		cpos, new_clusters);
+
+	memset(&rec, 0, sizeof(rec));
+	rec.e_cpos = cpu_to_le32(cpos);
+	rec.e_blkno = cpu_to_le64(start_blk);
+	rec.e_leaf_clusters = cpu_to_le16(new_clusters);
+	rec.e_flags = flags;
+	status = ocfs2_et_insert_check(et, &rec);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec,
+					  &free_records, &insert);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	trace_ocfs2_insert_extent(insert.ins_appending, insert.ins_contig,
+				  insert.ins_contig_index, free_records,
+				  insert.ins_tree_depth);
+
+	if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
+		status = ocfs2_grow_tree(handle, et,
+					 &insert.ins_tree_depth, &last_eb_bh,
+					 meta_ac);
+		if (status) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	/* Finally, we can add clusters. This might rotate the tree for us. */
+	status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
+	if (status < 0)
+		mlog_errno(status);
+	else
+		ocfs2_et_extent_map_insert(et, &rec);
+
+bail:
+	brelse(last_eb_bh);
+
+	return status;
+}
+
+/*
+ * Allcate and add clusters into the extent b-tree.
+ * The new clusters(clusters_to_add) will be inserted at logical_offset.
+ * The extent b-tree's root is specified by et, and
+ * it is not limited to the file storage. Any extent tree can use this
+ * function if it implements the proper ocfs2_extent_tree.
+ */
+int ocfs2_add_clusters_in_btree(handle_t *handle,
+				struct ocfs2_extent_tree *et,
+				u32 *logical_offset,
+				u32 clusters_to_add,
+				int mark_unwritten,
+				struct ocfs2_alloc_context *data_ac,
+				struct ocfs2_alloc_context *meta_ac,
+				enum ocfs2_alloc_restarted *reason_ret)
+{
+	int status = 0, err = 0;
+	int need_free = 0;
+	int free_extents;
+	enum ocfs2_alloc_restarted reason = RESTART_NONE;
+	u32 bit_off, num_bits;
+	u64 block;
+	u8 flags = 0;
+	struct ocfs2_super *osb =
+		OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
+
+	BUG_ON(!clusters_to_add);
+
+	if (mark_unwritten)
+		flags = OCFS2_EXT_UNWRITTEN;
+
+	free_extents = ocfs2_num_free_extents(osb, et);
+	if (free_extents < 0) {
+		status = free_extents;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* there are two cases which could cause us to EAGAIN in the
+	 * we-need-more-metadata case:
+	 * 1) we haven't reserved *any*
+	 * 2) we are so fragmented, we've needed to add metadata too
+	 *    many times. */
+	if (!free_extents && !meta_ac) {
+		err = -1;
+		status = -EAGAIN;
+		reason = RESTART_META;
+		goto leave;
+	} else if ((!free_extents)
+		   && (ocfs2_alloc_context_bits_left(meta_ac)
+		       < ocfs2_extend_meta_needed(et->et_root_el))) {
+		err = -2;
+		status = -EAGAIN;
+		reason = RESTART_META;
+		goto leave;
+	}
+
+	status = __ocfs2_claim_clusters(handle, data_ac, 1,
+					clusters_to_add, &bit_off, &num_bits);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto leave;
+	}
+
+	BUG_ON(num_bits > clusters_to_add);
+
+	/* reserve our write early -- insert_extent may update the tree root */
+	status = ocfs2_et_root_journal_access(handle, et,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		need_free = 1;
+		goto bail;
+	}
+
+	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
+	trace_ocfs2_add_clusters_in_btree(
+	     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+	     bit_off, num_bits);
+	status = ocfs2_insert_extent(handle, et, *logical_offset, block,
+				     num_bits, flags, meta_ac);
+	if (status < 0) {
+		mlog_errno(status);
+		need_free = 1;
+		goto bail;
+	}
+
+	ocfs2_journal_dirty(handle, et->et_root_bh);
+
+	clusters_to_add -= num_bits;
+	*logical_offset += num_bits;
+
+	if (clusters_to_add) {
+		err = clusters_to_add;
+		status = -EAGAIN;
+		reason = RESTART_TRANS;
+	}
+
+bail:
+	if (need_free) {
+		if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
+			ocfs2_free_local_alloc_bits(osb, handle, data_ac,
+					bit_off, num_bits);
+		else
+			ocfs2_free_clusters(handle,
+					data_ac->ac_inode,
+					data_ac->ac_bh,
+					ocfs2_clusters_to_blocks(osb->sb, bit_off),
+					num_bits);
+	}
+
+leave:
+	if (reason_ret)
+		*reason_ret = reason;
+	trace_ocfs2_add_clusters_in_btree_ret(status, reason, err);
+	return status;
+}
+
+static void ocfs2_make_right_split_rec(struct super_block *sb,
+				       struct ocfs2_extent_rec *split_rec,
+				       u32 cpos,
+				       struct ocfs2_extent_rec *rec)
+{
+	u32 rec_cpos = le32_to_cpu(rec->e_cpos);
+	u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);
+
+	memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));
+
+	split_rec->e_cpos = cpu_to_le32(cpos);
+	split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);
+
+	split_rec->e_blkno = rec->e_blkno;
+	le64_add_cpu(&split_rec->e_blkno,
+		     ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));
+
+	split_rec->e_flags = rec->e_flags;
+}
+
+static int ocfs2_split_and_insert(handle_t *handle,
+				  struct ocfs2_extent_tree *et,
+				  struct ocfs2_path *path,
+				  struct buffer_head **last_eb_bh,
+				  int split_index,
+				  struct ocfs2_extent_rec *orig_split_rec,
+				  struct ocfs2_alloc_context *meta_ac)
+{
+	int ret = 0, depth;
+	unsigned int insert_range, rec_range, do_leftright = 0;
+	struct ocfs2_extent_rec tmprec;
+	struct ocfs2_extent_list *rightmost_el;
+	struct ocfs2_extent_rec rec;
+	struct ocfs2_extent_rec split_rec = *orig_split_rec;
+	struct ocfs2_insert_type insert;
+	struct ocfs2_extent_block *eb;
+
+leftright:
+	/*
+	 * Store a copy of the record on the stack - it might move
+	 * around as the tree is manipulated below.
+	 */
+	rec = path_leaf_el(path)->l_recs[split_index];
+
+	rightmost_el = et->et_root_el;
+
+	depth = le16_to_cpu(rightmost_el->l_tree_depth);
+	if (depth) {
+		BUG_ON(!(*last_eb_bh));
+		eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
+		rightmost_el = &eb->h_list;
+	}
+
+	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
+	    le16_to_cpu(rightmost_el->l_count)) {
+		ret = ocfs2_grow_tree(handle, et,
+				      &depth, last_eb_bh, meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	memset(&insert, 0, sizeof(struct ocfs2_insert_type));
+	insert.ins_appending = APPEND_NONE;
+	insert.ins_contig = CONTIG_NONE;
+	insert.ins_tree_depth = depth;
+
+	insert_range = le32_to_cpu(split_rec.e_cpos) +
+		le16_to_cpu(split_rec.e_leaf_clusters);
+	rec_range = le32_to_cpu(rec.e_cpos) +
+		le16_to_cpu(rec.e_leaf_clusters);
+
+	if (split_rec.e_cpos == rec.e_cpos) {
+		insert.ins_split = SPLIT_LEFT;
+	} else if (insert_range == rec_range) {
+		insert.ins_split = SPLIT_RIGHT;
+	} else {
+		/*
+		 * Left/right split. We fake this as a right split
+		 * first and then make a second pass as a left split.
+		 */
+		insert.ins_split = SPLIT_RIGHT;
+
+		ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
+					   &tmprec, insert_range, &rec);
+
+		split_rec = tmprec;
+
+		BUG_ON(do_leftright);
+		do_leftright = 1;
+	}
+
+	ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (do_leftright == 1) {
+		u32 cpos;
+		struct ocfs2_extent_list *el;
+
+		do_leftright++;
+		split_rec = *orig_split_rec;
+
+		ocfs2_reinit_path(path, 1);
+
+		cpos = le32_to_cpu(split_rec.e_cpos);
+		ret = ocfs2_find_path(et->et_ci, path, cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		el = path_leaf_el(path);
+		split_index = ocfs2_search_extent_list(el, cpos);
+		if (split_index == -1) {
+			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
+					"Owner %llu has an extent at cpos %u "
+					"which can no longer be found.\n",
+					(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+					cpos);
+			ret = -EROFS;
+			goto out;
+		}
+		goto leftright;
+	}
+out:
+
+	return ret;
+}
+
+static int ocfs2_replace_extent_rec(handle_t *handle,
+				    struct ocfs2_extent_tree *et,
+				    struct ocfs2_path *path,
+				    struct ocfs2_extent_list *el,
+				    int split_index,
+				    struct ocfs2_extent_rec *split_rec)
+{
+	int ret;
+
+	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
+					   path_num_items(path) - 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	el->l_recs[split_index] = *split_rec;
+
+	ocfs2_journal_dirty(handle, path_leaf_bh(path));
+out:
+	return ret;
+}
+
+/*
+ * Split part or all of the extent record at split_index in the leaf
+ * pointed to by path. Merge with the contiguous extent record if needed.
+ *
+ * Care is taken to handle contiguousness so as to not grow the tree.
+ *
+ * meta_ac is not strictly necessary - we only truly need it if growth
+ * of the tree is required. All other cases will degrade into a less
+ * optimal tree layout.
+ *
+ * last_eb_bh should be the rightmost leaf block for any extent
+ * btree. Since a split may grow the tree or a merge might shrink it,
+ * the caller cannot trust the contents of that buffer after this call.
+ *
+ * This code is optimized for readability - several passes might be
+ * made over certain portions of the tree. All of those blocks will
+ * have been brought into cache (and pinned via the journal), so the
+ * extra overhead is not expressed in terms of disk reads.
+ */
+int ocfs2_split_extent(handle_t *handle,
+		       struct ocfs2_extent_tree *et,
+		       struct ocfs2_path *path,
+		       int split_index,
+		       struct ocfs2_extent_rec *split_rec,
+		       struct ocfs2_alloc_context *meta_ac,
+		       struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret = 0;
+	struct ocfs2_extent_list *el = path_leaf_el(path);
+	struct buffer_head *last_eb_bh = NULL;
+	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
+	struct ocfs2_merge_ctxt ctxt;
+	struct ocfs2_extent_list *rightmost_el;
+
+	if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) ||
+	    ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) <
+	     (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) {
+		ret = -EIO;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el,
+							    split_index,
+							    split_rec);
+
+	/*
+	 * The core merge / split code wants to know how much room is
+	 * left in this allocation tree, so we pass the
+	 * rightmost extent list.
+	 */
+	if (path->p_tree_depth) {
+		struct ocfs2_extent_block *eb;
+
+		ret = ocfs2_read_extent_block(et->et_ci,
+					      ocfs2_et_get_last_eb_blk(et),
+					      &last_eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
+		rightmost_el = &eb->h_list;
+	} else
+		rightmost_el = path_root_el(path);
+
+	if (rec->e_cpos == split_rec->e_cpos &&
+	    rec->e_leaf_clusters == split_rec->e_leaf_clusters)
+		ctxt.c_split_covers_rec = 1;
+	else
+		ctxt.c_split_covers_rec = 0;
+
+	ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);
+
+	trace_ocfs2_split_extent(split_index, ctxt.c_contig_type,
+				 ctxt.c_has_empty_extent,
+				 ctxt.c_split_covers_rec);
+
+	if (ctxt.c_contig_type == CONTIG_NONE) {
+		if (ctxt.c_split_covers_rec)
+			ret = ocfs2_replace_extent_rec(handle, et, path, el,
+						       split_index, split_rec);
+		else
+			ret = ocfs2_split_and_insert(handle, et, path,
+						     &last_eb_bh, split_index,
+						     split_rec, meta_ac);
+		if (ret)
+			mlog_errno(ret);
+	} else {
+		ret = ocfs2_try_to_merge_extent(handle, et, path,
+						split_index, split_rec,
+						dealloc, &ctxt);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out:
+	brelse(last_eb_bh);
+	return ret;
+}
+
+/*
+ * Change the flags of the already-existing extent at cpos for len clusters.
+ *
+ * new_flags: the flags we want to set.
+ * clear_flags: the flags we want to clear.
+ * phys: the new physical offset we want this new extent starts from.
+ *
+ * If the existing extent is larger than the request, initiate a
+ * split. An attempt will be made at merging with adjacent extents.
+ *
+ * The caller is responsible for passing down meta_ac if we'll need it.
+ */
+int ocfs2_change_extent_flag(handle_t *handle,
+			     struct ocfs2_extent_tree *et,
+			     u32 cpos, u32 len, u32 phys,
+			     struct ocfs2_alloc_context *meta_ac,
+			     struct ocfs2_cached_dealloc_ctxt *dealloc,
+			     int new_flags, int clear_flags)
+{
+	int ret, index;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
+	u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys);
+	struct ocfs2_extent_rec split_rec;
+	struct ocfs2_path *left_path = NULL;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+
+	left_path = ocfs2_new_path_from_et(et);
+	if (!left_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(et->et_ci, left_path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	el = path_leaf_el(left_path);
+
+	index = ocfs2_search_extent_list(el, cpos);
+	if (index == -1) {
+		ocfs2_error(sb,
+			    "Owner %llu has an extent at cpos %u which can no "
+			    "longer be found.\n",
+			     (unsigned long long)
+			     ocfs2_metadata_cache_owner(et->et_ci), cpos);
+		ret = -EROFS;
+		goto out;
+	}
+
+	ret = -EIO;
+	rec = &el->l_recs[index];
+	if (new_flags && (rec->e_flags & new_flags)) {
+		mlog(ML_ERROR, "Owner %llu tried to set %d flags on an "
+		     "extent that already had them",
+		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+		     new_flags);
+		goto out;
+	}
+
+	if (clear_flags && !(rec->e_flags & clear_flags)) {
+		mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an "
+		     "extent that didn't have them",
+		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+		     clear_flags);
+		goto out;
+	}
+
+	memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
+	split_rec.e_cpos = cpu_to_le32(cpos);
+	split_rec.e_leaf_clusters = cpu_to_le16(len);
+	split_rec.e_blkno = cpu_to_le64(start_blkno);
+	split_rec.e_flags = rec->e_flags;
+	if (new_flags)
+		split_rec.e_flags |= new_flags;
+	if (clear_flags)
+		split_rec.e_flags &= ~clear_flags;
+
+	ret = ocfs2_split_extent(handle, et, left_path,
+				 index, &split_rec, meta_ac,
+				 dealloc);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	ocfs2_free_path(left_path);
+	return ret;
+
+}
+
+/*
+ * Mark the already-existing extent at cpos as written for len clusters.
+ * This removes the unwritten extent flag.
+ *
+ * If the existing extent is larger than the request, initiate a
+ * split. An attempt will be made at merging with adjacent extents.
+ *
+ * The caller is responsible for passing down meta_ac if we'll need it.
+ */
+int ocfs2_mark_extent_written(struct inode *inode,
+			      struct ocfs2_extent_tree *et,
+			      handle_t *handle, u32 cpos, u32 len, u32 phys,
+			      struct ocfs2_alloc_context *meta_ac,
+			      struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+
+	trace_ocfs2_mark_extent_written(
+		(unsigned long long)OCFS2_I(inode)->ip_blkno,
+		cpos, len, phys);
+
+	if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
+		ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents "
+			    "that are being written to, but the feature bit "
+			    "is not set in the super block.",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno);
+		ret = -EROFS;
+		goto out;
+	}
+
+	/*
+	 * XXX: This should be fixed up so that we just re-insert the
+	 * next extent records.
+	 */
+	ocfs2_et_extent_map_truncate(et, 0);
+
+	ret = ocfs2_change_extent_flag(handle, et, cpos,
+				       len, phys, meta_ac, dealloc,
+				       0, OCFS2_EXT_UNWRITTEN);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et,
+			    struct ocfs2_path *path,
+			    int index, u32 new_range,
+			    struct ocfs2_alloc_context *meta_ac)
+{
+	int ret, depth, credits;
+	struct buffer_head *last_eb_bh = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *rightmost_el, *el;
+	struct ocfs2_extent_rec split_rec;
+	struct ocfs2_extent_rec *rec;
+	struct ocfs2_insert_type insert;
+
+	/*
+	 * Setup the record to split before we grow the tree.
+	 */
+	el = path_leaf_el(path);
+	rec = &el->l_recs[index];
+	ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
+				   &split_rec, new_range, rec);
+
+	depth = path->p_tree_depth;
+	if (depth > 0) {
+		ret = ocfs2_read_extent_block(et->et_ci,
+					      ocfs2_et_get_last_eb_blk(et),
+					      &last_eb_bh);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
+		rightmost_el = &eb->h_list;
+	} else
+		rightmost_el = path_leaf_el(path);
+
+	credits = path->p_tree_depth +
+		  ocfs2_extend_meta_needed(et->et_root_el);
+	ret = ocfs2_extend_trans(handle, credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
+	    le16_to_cpu(rightmost_el->l_count)) {
+		ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
+				      meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	memset(&insert, 0, sizeof(struct ocfs2_insert_type));
+	insert.ins_appending = APPEND_NONE;
+	insert.ins_contig = CONTIG_NONE;
+	insert.ins_split = SPLIT_RIGHT;
+	insert.ins_tree_depth = depth;
+
+	ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	brelse(last_eb_bh);
+	return ret;
+}
+
+static int ocfs2_truncate_rec(handle_t *handle,
+			      struct ocfs2_extent_tree *et,
+			      struct ocfs2_path *path, int index,
+			      struct ocfs2_cached_dealloc_ctxt *dealloc,
+			      u32 cpos, u32 len)
+{
+	int ret;
+	u32 left_cpos, rec_range, trunc_range;
+	int wants_rotate = 0, is_rightmost_tree_rec = 0;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
+	struct ocfs2_path *left_path = NULL;
+	struct ocfs2_extent_list *el = path_leaf_el(path);
+	struct ocfs2_extent_rec *rec;
+	struct ocfs2_extent_block *eb;
+
+	if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
+		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		index--;
+	}
+
+	if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
+	    path->p_tree_depth) {
+		/*
+		 * Check whether this is the rightmost tree record. If
+		 * we remove all of this record or part of its right
+		 * edge then an update of the record lengths above it
+		 * will be required.
+		 */
+		eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
+		if (eb->h_next_leaf_blk == 0)
+			is_rightmost_tree_rec = 1;
+	}
+
+	rec = &el->l_recs[index];
+	if (index == 0 && path->p_tree_depth &&
+	    le32_to_cpu(rec->e_cpos) == cpos) {
+		/*
+		 * Changing the leftmost offset (via partial or whole
+		 * record truncate) of an interior (or rightmost) path
+		 * means we have to update the subtree that is formed
+		 * by this leaf and the one to it's left.
+		 *
+		 * There are two cases we can skip:
+		 *   1) Path is the leftmost one in our btree.
+		 *   2) The leaf is rightmost and will be empty after
+		 *      we remove the extent record - the rotate code
+		 *      knows how to update the newly formed edge.
+		 */
+
+		ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
+			left_path = ocfs2_new_path_from_path(path);
+			if (!left_path) {
+				ret = -ENOMEM;
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ret = ocfs2_find_path(et->et_ci, left_path,
+					      left_cpos);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+	}
+
+	ret = ocfs2_extend_rotate_transaction(handle, 0,
+					      handle->h_buffer_credits,
+					      path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+	trunc_range = cpos + len;
+
+	if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
+		int next_free;
+
+		memset(rec, 0, sizeof(*rec));
+		ocfs2_cleanup_merge(el, index);
+		wants_rotate = 1;
+
+		next_free = le16_to_cpu(el->l_next_free_rec);
+		if (is_rightmost_tree_rec && next_free > 1) {
+			/*
+			 * We skip the edge update if this path will
+			 * be deleted by the rotate code.
+			 */
+			rec = &el->l_recs[next_free - 1];
+			ocfs2_adjust_rightmost_records(handle, et, path,
+						       rec);
+		}
+	} else if (le32_to_cpu(rec->e_cpos) == cpos) {
+		/* Remove leftmost portion of the record. */
+		le32_add_cpu(&rec->e_cpos, len);
+		le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
+		le16_add_cpu(&rec->e_leaf_clusters, -len);
+	} else if (rec_range == trunc_range) {
+		/* Remove rightmost portion of the record */
+		le16_add_cpu(&rec->e_leaf_clusters, -len);
+		if (is_rightmost_tree_rec)
+			ocfs2_adjust_rightmost_records(handle, et, path, rec);
+	} else {
+		/* Caller should have trapped this. */
+		mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
+		     "(%u, %u)\n",
+		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+		     le32_to_cpu(rec->e_cpos),
+		     le16_to_cpu(rec->e_leaf_clusters), cpos, len);
+		BUG();
+	}
+
+	if (left_path) {
+		int subtree_index;
+
+		subtree_index = ocfs2_find_subtree_root(et, left_path, path);
+		ocfs2_complete_edge_insert(handle, left_path, path,
+					   subtree_index);
+	}
+
+	ocfs2_journal_dirty(handle, path_leaf_bh(path));
+
+	ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+out:
+	ocfs2_free_path(left_path);
+	return ret;
+}
+
+int ocfs2_remove_extent(handle_t *handle,
+			struct ocfs2_extent_tree *et,
+			u32 cpos, u32 len,
+			struct ocfs2_alloc_context *meta_ac,
+			struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret, index;
+	u32 rec_range, trunc_range;
+	struct ocfs2_extent_rec *rec;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_path *path = NULL;
+
+	/*
+	 * XXX: Why are we truncating to 0 instead of wherever this
+	 * affects us?
+	 */
+	ocfs2_et_extent_map_truncate(et, 0);
+
+	path = ocfs2_new_path_from_et(et);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(et->et_ci, path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	el = path_leaf_el(path);
+	index = ocfs2_search_extent_list(el, cpos);
+	if (index == -1) {
+		ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
+			    "Owner %llu has an extent at cpos %u which can no "
+			    "longer be found.\n",
+			    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+			    cpos);
+		ret = -EROFS;
+		goto out;
+	}
+
+	/*
+	 * We have 3 cases of extent removal:
+	 *   1) Range covers the entire extent rec
+	 *   2) Range begins or ends on one edge of the extent rec
+	 *   3) Range is in the middle of the extent rec (no shared edges)
+	 *
+	 * For case 1 we remove the extent rec and left rotate to
+	 * fill the hole.
+	 *
+	 * For case 2 we just shrink the existing extent rec, with a
+	 * tree update if the shrinking edge is also the edge of an
+	 * extent block.
+	 *
+	 * For case 3 we do a right split to turn the extent rec into
+	 * something case 2 can handle.
+	 */
+	rec = &el->l_recs[index];
+	rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+	trunc_range = cpos + len;
+
+	BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
+
+	trace_ocfs2_remove_extent(
+		(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+		cpos, len, index, le32_to_cpu(rec->e_cpos),
+		ocfs2_rec_clusters(el, rec));
+
+	if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
+		ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
+					 cpos, len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	} else {
+		ret = ocfs2_split_tree(handle, et, path, index,
+				       trunc_range, meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * The split could have manipulated the tree enough to
+		 * move the record location, so we have to look for it again.
+		 */
+		ocfs2_reinit_path(path, 1);
+
+		ret = ocfs2_find_path(et->et_ci, path, cpos);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		el = path_leaf_el(path);
+		index = ocfs2_search_extent_list(el, cpos);
+		if (index == -1) {
+			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
+				    "Owner %llu: split at cpos %u lost record.",
+				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+				    cpos);
+			ret = -EROFS;
+			goto out;
+		}
+
+		/*
+		 * Double check our values here. If anything is fishy,
+		 * it's easier to catch it at the top level.
+		 */
+		rec = &el->l_recs[index];
+		rec_range = le32_to_cpu(rec->e_cpos) +
+			ocfs2_rec_clusters(el, rec);
+		if (rec_range != trunc_range) {
+			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
+				    "Owner %llu: error after split at cpos %u"
+				    "trunc len %u, existing record is (%u,%u)",
+				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
+				    cpos, len, le32_to_cpu(rec->e_cpos),
+				    ocfs2_rec_clusters(el, rec));
+			ret = -EROFS;
+			goto out;
+		}
+
+		ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
+					 cpos, len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+out:
+	ocfs2_free_path(path);
+	return ret;
+}
+
+/*
+ * ocfs2_reserve_blocks_for_rec_trunc() would look basically the
+ * same as ocfs2_lock_alloctors(), except for it accepts a blocks
+ * number to reserve some extra blocks, and it only handles meta
+ * data allocations.
+ *
+ * Currently, only ocfs2_remove_btree_range() uses it for truncating
+ * and punching holes.
+ */
+static int ocfs2_reserve_blocks_for_rec_trunc(struct inode *inode,
+					      struct ocfs2_extent_tree *et,
+					      u32 extents_to_split,
+					      struct ocfs2_alloc_context **ac,
+					      int extra_blocks)
+{
+	int ret = 0, num_free_extents;
+	unsigned int max_recs_needed = 2 * extents_to_split;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	*ac = NULL;
+
+	num_free_extents = ocfs2_num_free_extents(osb, et);
+	if (num_free_extents < 0) {
+		ret = num_free_extents;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (!num_free_extents ||
+	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
+		extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
+
+	if (extra_blocks) {
+		ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, ac);
+		if (ret < 0) {
+			if (ret != -ENOSPC)
+				mlog_errno(ret);
+			goto out;
+		}
+	}
+
+out:
+	if (ret) {
+		if (*ac) {
+			ocfs2_free_alloc_context(*ac);
+			*ac = NULL;
+		}
+	}
+
+	return ret;
+}
+
+int ocfs2_remove_btree_range(struct inode *inode,
+			     struct ocfs2_extent_tree *et,
+			     u32 cpos, u32 phys_cpos, u32 len, int flags,
+			     struct ocfs2_cached_dealloc_ctxt *dealloc,
+			     u64 refcount_loc, bool refcount_tree_locked)
+{
+	int ret, credits = 0, extra_blocks = 0;
+	u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct inode *tl_inode = osb->osb_tl_inode;
+	handle_t *handle;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	struct ocfs2_refcount_tree *ref_tree = NULL;
+
+	if ((flags & OCFS2_EXT_REFCOUNTED) && len) {
+		BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
+			 OCFS2_HAS_REFCOUNT_FL));
+
+		if (!refcount_tree_locked) {
+			ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
+						       &ref_tree, NULL);
+			if (ret) {
+				mlog_errno(ret);
+				goto bail;
+			}
+		}
+
+		ret = ocfs2_prepare_refcount_change_for_del(inode,
+							    refcount_loc,
+							    phys_blkno,
+							    len,
+							    &credits,
+							    &extra_blocks);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto bail;
+		}
+	}
+
+	ret = ocfs2_reserve_blocks_for_rec_trunc(inode, et, 1, &meta_ac,
+						 extra_blocks);
+	if (ret) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	mutex_lock(&tl_inode->i_mutex);
+
+	if (ocfs2_truncate_log_needs_flush(osb)) {
+		ret = __ocfs2_flush_truncate_log(osb);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	handle = ocfs2_start_trans(osb,
+			ocfs2_remove_extent_credits(osb->sb) + credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_et_root_journal_access(handle, et,
+					   OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	dquot_free_space_nodirty(inode,
+				  ocfs2_clusters_to_bytes(inode->i_sb, len));
+
+	ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ocfs2_et_update_clusters(et, -len);
+	ocfs2_update_inode_fsync_trans(handle, inode, 1);
+
+	ocfs2_journal_dirty(handle, et->et_root_bh);
+
+	if (phys_blkno) {
+		if (flags & OCFS2_EXT_REFCOUNTED)
+			ret = ocfs2_decrease_refcount(inode, handle,
+					ocfs2_blocks_to_clusters(osb->sb,
+								 phys_blkno),
+					len, meta_ac,
+					dealloc, 1);
+		else
+			ret = ocfs2_truncate_log_append(osb, handle,
+							phys_blkno, len);
+		if (ret)
+			mlog_errno(ret);
+
+	}
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	mutex_unlock(&tl_inode->i_mutex);
+bail:
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	if (ref_tree)
+		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+
+	return ret;
+}
+
+int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
+{
+	struct buffer_head *tl_bh = osb->osb_tl_bh;
+	struct ocfs2_dinode *di;
+	struct ocfs2_truncate_log *tl;
+
+	di = (struct ocfs2_dinode *) tl_bh->b_data;
+	tl = &di->id2.i_dealloc;
+
+	mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
+			"slot %d, invalid truncate log parameters: used = "
+			"%u, count = %u\n", osb->slot_num,
+			le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
+	return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
+}
+
+static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
+					   unsigned int new_start)
+{
+	unsigned int tail_index;
+	unsigned int current_tail;
+
+	/* No records, nothing to coalesce */
+	if (!le16_to_cpu(tl->tl_used))
+		return 0;
+
+	tail_index = le16_to_cpu(tl->tl_used) - 1;
+	current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
+	current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);
+
+	return current_tail == new_start;
+}
+
+int ocfs2_truncate_log_append(struct ocfs2_super *osb,
+			      handle_t *handle,
+			      u64 start_blk,
+			      unsigned int num_clusters)
+{
+	int status, index;
+	unsigned int start_cluster, tl_count;
+	struct inode *tl_inode = osb->osb_tl_inode;
+	struct buffer_head *tl_bh = osb->osb_tl_bh;
+	struct ocfs2_dinode *di;
+	struct ocfs2_truncate_log *tl;
+
+	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
+
+	start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);
+
+	di = (struct ocfs2_dinode *) tl_bh->b_data;
+
+	/* tl_bh is loaded from ocfs2_truncate_log_init().  It's validated
+	 * by the underlying call to ocfs2_read_inode_block(), so any
+	 * corruption is a code bug */
+	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
+
+	tl = &di->id2.i_dealloc;
+	tl_count = le16_to_cpu(tl->tl_count);
+	mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
+			tl_count == 0,
+			"Truncate record count on #%llu invalid "
+			"wanted %u, actual %u\n",
+			(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
+			ocfs2_truncate_recs_per_inode(osb->sb),
+			le16_to_cpu(tl->tl_count));
+
+	/* Caller should have known to flush before calling us. */
+	index = le16_to_cpu(tl->tl_used);
+	if (index >= tl_count) {
+		status = -ENOSPC;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	trace_ocfs2_truncate_log_append(
+		(unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index,
+		start_cluster, num_clusters);
+	if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
+		/*
+		 * Move index back to the record we are coalescing with.
+		 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
+		 */
+		index--;
+
+		num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
+		trace_ocfs2_truncate_log_append(
+			(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
+			index, le32_to_cpu(tl->tl_recs[index].t_start),
+			num_clusters);
+	} else {
+		tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
+		tl->tl_used = cpu_to_le16(index + 1);
+	}
+	tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);
+
+	ocfs2_journal_dirty(handle, tl_bh);
+
+	osb->truncated_clusters += num_clusters;
+bail:
+	return status;
+}
+
+static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
+					 handle_t *handle,
+					 struct inode *data_alloc_inode,
+					 struct buffer_head *data_alloc_bh)
+{
+	int status = 0;
+	int i;
+	unsigned int num_clusters;
+	u64 start_blk;
+	struct ocfs2_truncate_rec rec;
+	struct ocfs2_dinode *di;
+	struct ocfs2_truncate_log *tl;
+	struct inode *tl_inode = osb->osb_tl_inode;
+	struct buffer_head *tl_bh = osb->osb_tl_bh;
+
+	di = (struct ocfs2_dinode *) tl_bh->b_data;
+	tl = &di->id2.i_dealloc;
+	i = le16_to_cpu(tl->tl_used) - 1;
+	while (i >= 0) {
+		/* Caller has given us at least enough credits to
+		 * update the truncate log dinode */
+		status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
+						 OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		tl->tl_used = cpu_to_le16(i);
+
+		ocfs2_journal_dirty(handle, tl_bh);
+
+		/* TODO: Perhaps we can calculate the bulk of the
+		 * credits up front rather than extending like
+		 * this. */
+		status = ocfs2_extend_trans(handle,
+					    OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		rec = tl->tl_recs[i];
+		start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
+						    le32_to_cpu(rec.t_start));
+		num_clusters = le32_to_cpu(rec.t_clusters);
+
+		/* if start_blk is not set, we ignore the record as
+		 * invalid. */
+		if (start_blk) {
+			trace_ocfs2_replay_truncate_records(
+				(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
+				i, le32_to_cpu(rec.t_start), num_clusters);
+
+			status = ocfs2_free_clusters(handle, data_alloc_inode,
+						     data_alloc_bh, start_blk,
+						     num_clusters);
+			if (status < 0) {
+				mlog_errno(status);
+				goto bail;
+			}
+		}
+		i--;
+	}
+
+	osb->truncated_clusters = 0;
+
+bail:
+	return status;
+}
+
+/* Expects you to already be holding tl_inode->i_mutex */
+int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
+{
+	int status;
+	unsigned int num_to_flush;
+	handle_t *handle;
+	struct inode *tl_inode = osb->osb_tl_inode;
+	struct inode *data_alloc_inode = NULL;
+	struct buffer_head *tl_bh = osb->osb_tl_bh;
+	struct buffer_head *data_alloc_bh = NULL;
+	struct ocfs2_dinode *di;
+	struct ocfs2_truncate_log *tl;
+
+	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
+
+	di = (struct ocfs2_dinode *) tl_bh->b_data;
+
+	/* tl_bh is loaded from ocfs2_truncate_log_init().  It's validated
+	 * by the underlying call to ocfs2_read_inode_block(), so any
+	 * corruption is a code bug */
+	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
+
+	tl = &di->id2.i_dealloc;
+	num_to_flush = le16_to_cpu(tl->tl_used);
+	trace_ocfs2_flush_truncate_log(
+		(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
+		num_to_flush);
+	if (!num_to_flush) {
+		status = 0;
+		goto out;
+	}
+
+	data_alloc_inode = ocfs2_get_system_file_inode(osb,
+						       GLOBAL_BITMAP_SYSTEM_INODE,
+						       OCFS2_INVALID_SLOT);
+	if (!data_alloc_inode) {
+		status = -EINVAL;
+		mlog(ML_ERROR, "Could not get bitmap inode!\n");
+		goto out;
+	}
+
+	mutex_lock(&data_alloc_inode->i_mutex);
+
+	status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_mutex;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out_unlock;
+	}
+
+	status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
+					       data_alloc_bh);
+	if (status < 0)
+		mlog_errno(status);
+
+	ocfs2_commit_trans(osb, handle);
+
+out_unlock:
+	brelse(data_alloc_bh);
+	ocfs2_inode_unlock(data_alloc_inode, 1);
+
+out_mutex:
+	mutex_unlock(&data_alloc_inode->i_mutex);
+	iput(data_alloc_inode);
+
+out:
+	return status;
+}
+
+int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
+{
+	int status;
+	struct inode *tl_inode = osb->osb_tl_inode;
+
+	mutex_lock(&tl_inode->i_mutex);
+	status = __ocfs2_flush_truncate_log(osb);
+	mutex_unlock(&tl_inode->i_mutex);
+
+	return status;
+}
+
+static void ocfs2_truncate_log_worker(struct work_struct *work)
+{
+	int status;
+	struct ocfs2_super *osb =
+		container_of(work, struct ocfs2_super,
+			     osb_truncate_log_wq.work);
+
+	status = ocfs2_flush_truncate_log(osb);
+	if (status < 0)
+		mlog_errno(status);
+	else
+		ocfs2_init_steal_slots(osb);
+}
+
+#define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
+void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
+				       int cancel)
+{
+	if (osb->osb_tl_inode &&
+			atomic_read(&osb->osb_tl_disable) == 0) {
+		/* We want to push off log flushes while truncates are
+		 * still running. */
+		if (cancel)
+			cancel_delayed_work(&osb->osb_truncate_log_wq);
+
+		queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
+				   OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
+	}
+}
+
+static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
+				       int slot_num,
+				       struct inode **tl_inode,
+				       struct buffer_head **tl_bh)
+{
+	int status;
+	struct inode *inode = NULL;
+	struct buffer_head *bh = NULL;
+
+	inode = ocfs2_get_system_file_inode(osb,
+					   TRUNCATE_LOG_SYSTEM_INODE,
+					   slot_num);
+	if (!inode) {
+		status = -EINVAL;
+		mlog(ML_ERROR, "Could not get load truncate log inode!\n");
+		goto bail;
+	}
+
+	status = ocfs2_read_inode_block(inode, &bh);
+	if (status < 0) {
+		iput(inode);
+		mlog_errno(status);
+		goto bail;
+	}
+
+	*tl_inode = inode;
+	*tl_bh    = bh;
+bail:
+	return status;
+}
+
+/* called during the 1st stage of node recovery. we stamp a clean
+ * truncate log and pass back a copy for processing later. if the
+ * truncate log does not require processing, a *tl_copy is set to
+ * NULL. */
+int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
+				      int slot_num,
+				      struct ocfs2_dinode **tl_copy)
+{
+	int status;
+	struct inode *tl_inode = NULL;
+	struct buffer_head *tl_bh = NULL;
+	struct ocfs2_dinode *di;
+	struct ocfs2_truncate_log *tl;
+
+	*tl_copy = NULL;
+
+	trace_ocfs2_begin_truncate_log_recovery(slot_num);
+
+	status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	di = (struct ocfs2_dinode *) tl_bh->b_data;
+
+	/* tl_bh is loaded from ocfs2_get_truncate_log_info().  It's
+	 * validated by the underlying call to ocfs2_read_inode_block(),
+	 * so any corruption is a code bug */
+	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
+
+	tl = &di->id2.i_dealloc;
+	if (le16_to_cpu(tl->tl_used)) {
+		trace_ocfs2_truncate_log_recovery_num(le16_to_cpu(tl->tl_used));
+
+		*tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
+		if (!(*tl_copy)) {
+			status = -ENOMEM;
+			mlog_errno(status);
+			goto bail;
+		}
+
+		/* Assuming the write-out below goes well, this copy
+		 * will be passed back to recovery for processing. */
+		memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);
+
+		/* All we need to do to clear the truncate log is set
+		 * tl_used. */
+		tl->tl_used = 0;
+
+		ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
+		status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+bail:
+	if (tl_inode)
+		iput(tl_inode);
+	brelse(tl_bh);
+
+	if (status < 0 && (*tl_copy)) {
+		kfree(*tl_copy);
+		*tl_copy = NULL;
+		mlog_errno(status);
+	}
+
+	return status;
+}
+
+int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
+					 struct ocfs2_dinode *tl_copy)
+{
+	int status = 0;
+	int i;
+	unsigned int clusters, num_recs, start_cluster;
+	u64 start_blk;
+	handle_t *handle;
+	struct inode *tl_inode = osb->osb_tl_inode;
+	struct ocfs2_truncate_log *tl;
+
+	if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
+		mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
+		return -EINVAL;
+	}
+
+	tl = &tl_copy->id2.i_dealloc;
+	num_recs = le16_to_cpu(tl->tl_used);
+	trace_ocfs2_complete_truncate_log_recovery(
+		(unsigned long long)le64_to_cpu(tl_copy->i_blkno),
+		num_recs);
+
+	mutex_lock(&tl_inode->i_mutex);
+	for(i = 0; i < num_recs; i++) {
+		if (ocfs2_truncate_log_needs_flush(osb)) {
+			status = __ocfs2_flush_truncate_log(osb);
+			if (status < 0) {
+				mlog_errno(status);
+				goto bail_up;
+			}
+		}
+
+		handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
+		if (IS_ERR(handle)) {
+			status = PTR_ERR(handle);
+			mlog_errno(status);
+			goto bail_up;
+		}
+
+		clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
+		start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
+		start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);
+
+		status = ocfs2_truncate_log_append(osb, handle,
+						   start_blk, clusters);
+		ocfs2_commit_trans(osb, handle);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail_up;
+		}
+	}
+
+bail_up:
+	mutex_unlock(&tl_inode->i_mutex);
+
+	return status;
+}
+
+void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
+{
+	int status;
+	struct inode *tl_inode = osb->osb_tl_inode;
+
+	atomic_set(&osb->osb_tl_disable, 1);
+
+	if (tl_inode) {
+		cancel_delayed_work(&osb->osb_truncate_log_wq);
+		flush_workqueue(ocfs2_wq);
+
+		status = ocfs2_flush_truncate_log(osb);
+		if (status < 0)
+			mlog_errno(status);
+
+		brelse(osb->osb_tl_bh);
+		iput(osb->osb_tl_inode);
+	}
+}
+
+int ocfs2_truncate_log_init(struct ocfs2_super *osb)
+{
+	int status;
+	struct inode *tl_inode = NULL;
+	struct buffer_head *tl_bh = NULL;
+
+	status = ocfs2_get_truncate_log_info(osb,
+					     osb->slot_num,
+					     &tl_inode,
+					     &tl_bh);
+	if (status < 0)
+		mlog_errno(status);
+
+	/* ocfs2_truncate_log_shutdown keys on the existence of
+	 * osb->osb_tl_inode so we don't set any of the osb variables
+	 * until we're sure all is well. */
+	INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
+			  ocfs2_truncate_log_worker);
+	atomic_set(&osb->osb_tl_disable, 0);
+	osb->osb_tl_bh    = tl_bh;
+	osb->osb_tl_inode = tl_inode;
+
+	return status;
+}
+
+/*
+ * Delayed de-allocation of suballocator blocks.
+ *
+ * Some sets of block de-allocations might involve multiple suballocator inodes.
+ *
+ * The locking for this can get extremely complicated, especially when
+ * the suballocator inodes to delete from aren't known until deep
+ * within an unrelated codepath.
+ *
+ * ocfs2_extent_block structures are a good example of this - an inode
+ * btree could have been grown by any number of nodes each allocating
+ * out of their own suballoc inode.
+ *
+ * These structures allow the delay of block de-allocation until a
+ * later time, when locking of multiple cluster inodes won't cause
+ * deadlock.
+ */
+
+/*
+ * Describe a single bit freed from a suballocator.  For the block
+ * suballocators, it represents one block.  For the global cluster
+ * allocator, it represents some clusters and free_bit indicates
+ * clusters number.
+ */
+struct ocfs2_cached_block_free {
+	struct ocfs2_cached_block_free		*free_next;
+	u64					free_bg;
+	u64					free_blk;
+	unsigned int				free_bit;
+};
+
+struct ocfs2_per_slot_free_list {
+	struct ocfs2_per_slot_free_list		*f_next_suballocator;
+	int					f_inode_type;
+	int					f_slot;
+	struct ocfs2_cached_block_free		*f_first;
+};
+
+static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
+				    int sysfile_type,
+				    int slot,
+				    struct ocfs2_cached_block_free *head)
+{
+	int ret;
+	u64 bg_blkno;
+	handle_t *handle;
+	struct inode *inode;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_cached_block_free *tmp;
+
+	inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
+	if (!inode) {
+		ret = -EINVAL;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mutex_lock(&inode->i_mutex);
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_mutex;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+
+	while (head) {
+		if (head->free_bg)
+			bg_blkno = head->free_bg;
+		else
+			bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
+							      head->free_bit);
+		trace_ocfs2_free_cached_blocks(
+		     (unsigned long long)head->free_blk, head->free_bit);
+
+		ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
+					       head->free_bit, bg_blkno, 1);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_journal;
+		}
+
+		ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_journal;
+		}
+
+		tmp = head;
+		head = head->free_next;
+		kfree(tmp);
+	}
+
+out_journal:
+	ocfs2_commit_trans(osb, handle);
+
+out_unlock:
+	ocfs2_inode_unlock(inode, 1);
+	brelse(di_bh);
+out_mutex:
+	mutex_unlock(&inode->i_mutex);
+	iput(inode);
+out:
+	while(head) {
+		/* Premature exit may have left some dangling items. */
+		tmp = head;
+		head = head->free_next;
+		kfree(tmp);
+	}
+
+	return ret;
+}
+
+int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
+				u64 blkno, unsigned int bit)
+{
+	int ret = 0;
+	struct ocfs2_cached_block_free *item;
+
+	item = kzalloc(sizeof(*item), GFP_NOFS);
+	if (item == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		return ret;
+	}
+
+	trace_ocfs2_cache_cluster_dealloc((unsigned long long)blkno, bit);
+
+	item->free_blk = blkno;
+	item->free_bit = bit;
+	item->free_next = ctxt->c_global_allocator;
+
+	ctxt->c_global_allocator = item;
+	return ret;
+}
+
+static int ocfs2_free_cached_clusters(struct ocfs2_super *osb,
+				      struct ocfs2_cached_block_free *head)
+{
+	struct ocfs2_cached_block_free *tmp;
+	struct inode *tl_inode = osb->osb_tl_inode;
+	handle_t *handle;
+	int ret = 0;
+
+	mutex_lock(&tl_inode->i_mutex);
+
+	while (head) {
+		if (ocfs2_truncate_log_needs_flush(osb)) {
+			ret = __ocfs2_flush_truncate_log(osb);
+			if (ret < 0) {
+				mlog_errno(ret);
+				break;
+			}
+		}
+
+		handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
+		if (IS_ERR(handle)) {
+			ret = PTR_ERR(handle);
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
+						head->free_bit);
+
+		ocfs2_commit_trans(osb, handle);
+		tmp = head;
+		head = head->free_next;
+		kfree(tmp);
+
+		if (ret < 0) {
+			mlog_errno(ret);
+			break;
+		}
+	}
+
+	mutex_unlock(&tl_inode->i_mutex);
+
+	while (head) {
+		/* Premature exit may have left some dangling items. */
+		tmp = head;
+		head = head->free_next;
+		kfree(tmp);
+	}
+
+	return ret;
+}
+
+int ocfs2_run_deallocs(struct ocfs2_super *osb,
+		       struct ocfs2_cached_dealloc_ctxt *ctxt)
+{
+	int ret = 0, ret2;
+	struct ocfs2_per_slot_free_list *fl;
+
+	if (!ctxt)
+		return 0;
+
+	while (ctxt->c_first_suballocator) {
+		fl = ctxt->c_first_suballocator;
+
+		if (fl->f_first) {
+			trace_ocfs2_run_deallocs(fl->f_inode_type,
+						 fl->f_slot);
+			ret2 = ocfs2_free_cached_blocks(osb,
+							fl->f_inode_type,
+							fl->f_slot,
+							fl->f_first);
+			if (ret2)
+				mlog_errno(ret2);
+			if (!ret)
+				ret = ret2;
+		}
+
+		ctxt->c_first_suballocator = fl->f_next_suballocator;
+		kfree(fl);
+	}
+
+	if (ctxt->c_global_allocator) {
+		ret2 = ocfs2_free_cached_clusters(osb,
+						  ctxt->c_global_allocator);
+		if (ret2)
+			mlog_errno(ret2);
+		if (!ret)
+			ret = ret2;
+
+		ctxt->c_global_allocator = NULL;
+	}
+
+	return ret;
+}
+
+static struct ocfs2_per_slot_free_list *
+ocfs2_find_per_slot_free_list(int type,
+			      int slot,
+			      struct ocfs2_cached_dealloc_ctxt *ctxt)
+{
+	struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
+
+	while (fl) {
+		if (fl->f_inode_type == type && fl->f_slot == slot)
+			return fl;
+
+		fl = fl->f_next_suballocator;
+	}
+
+	fl = kmalloc(sizeof(*fl), GFP_NOFS);
+	if (fl) {
+		fl->f_inode_type = type;
+		fl->f_slot = slot;
+		fl->f_first = NULL;
+		fl->f_next_suballocator = ctxt->c_first_suballocator;
+
+		ctxt->c_first_suballocator = fl;
+	}
+	return fl;
+}
+
+int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
+			      int type, int slot, u64 suballoc,
+			      u64 blkno, unsigned int bit)
+{
+	int ret;
+	struct ocfs2_per_slot_free_list *fl;
+	struct ocfs2_cached_block_free *item;
+
+	fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
+	if (fl == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	item = kzalloc(sizeof(*item), GFP_NOFS);
+	if (item == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_cache_block_dealloc(type, slot,
+					(unsigned long long)suballoc,
+					(unsigned long long)blkno, bit);
+
+	item->free_bg = suballoc;
+	item->free_blk = blkno;
+	item->free_bit = bit;
+	item->free_next = fl->f_first;
+
+	fl->f_first = item;
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
+					 struct ocfs2_extent_block *eb)
+{
+	return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
+					 le16_to_cpu(eb->h_suballoc_slot),
+					 le64_to_cpu(eb->h_suballoc_loc),
+					 le64_to_cpu(eb->h_blkno),
+					 le16_to_cpu(eb->h_suballoc_bit));
+}
+
+static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
+{
+	set_buffer_uptodate(bh);
+	mark_buffer_dirty(bh);
+	return 0;
+}
+
+void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
+			      unsigned int from, unsigned int to,
+			      struct page *page, int zero, u64 *phys)
+{
+	int ret, partial = 0;
+
+	ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0);
+	if (ret)
+		mlog_errno(ret);
+
+	if (zero)
+		zero_user_segment(page, from, to);
+
+	/*
+	 * Need to set the buffers we zero'd into uptodate
+	 * here if they aren't - ocfs2_map_page_blocks()
+	 * might've skipped some
+	 */
+	ret = walk_page_buffers(handle, page_buffers(page),
+				from, to, &partial,
+				ocfs2_zero_func);
+	if (ret < 0)
+		mlog_errno(ret);
+	else if (ocfs2_should_order_data(inode)) {
+		ret = ocfs2_jbd2_file_inode(handle, inode);
+		if (ret < 0)
+			mlog_errno(ret);
+	}
+
+	if (!partial)
+		SetPageUptodate(page);
+
+	flush_dcache_page(page);
+}
+
+static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
+				     loff_t end, struct page **pages,
+				     int numpages, u64 phys, handle_t *handle)
+{
+	int i;
+	struct page *page;
+	unsigned int from, to = PAGE_CACHE_SIZE;
+	struct super_block *sb = inode->i_sb;
+
+	BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
+
+	if (numpages == 0)
+		goto out;
+
+	to = PAGE_CACHE_SIZE;
+	for(i = 0; i < numpages; i++) {
+		page = pages[i];
+
+		from = start & (PAGE_CACHE_SIZE - 1);
+		if ((end >> PAGE_CACHE_SHIFT) == page->index)
+			to = end & (PAGE_CACHE_SIZE - 1);
+
+		BUG_ON(from > PAGE_CACHE_SIZE);
+		BUG_ON(to > PAGE_CACHE_SIZE);
+
+		ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
+					 &phys);
+
+		start = (page->index + 1) << PAGE_CACHE_SHIFT;
+	}
+out:
+	if (pages)
+		ocfs2_unlock_and_free_pages(pages, numpages);
+}
+
+int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
+		     struct page **pages, int *num)
+{
+	int numpages, ret = 0;
+	struct address_space *mapping = inode->i_mapping;
+	unsigned long index;
+	loff_t last_page_bytes;
+
+	BUG_ON(start > end);
+
+	numpages = 0;
+	last_page_bytes = PAGE_ALIGN(end);
+	index = start >> PAGE_CACHE_SHIFT;
+	do {
+		pages[numpages] = find_or_create_page(mapping, index, GFP_NOFS);
+		if (!pages[numpages]) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			goto out;
+		}
+
+		numpages++;
+		index++;
+	} while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
+
+out:
+	if (ret != 0) {
+		if (pages)
+			ocfs2_unlock_and_free_pages(pages, numpages);
+		numpages = 0;
+	}
+
+	*num = numpages;
+
+	return ret;
+}
+
+static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
+				struct page **pages, int *num)
+{
+	struct super_block *sb = inode->i_sb;
+
+	BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
+	       (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);
+
+	return ocfs2_grab_pages(inode, start, end, pages, num);
+}
+
+/*
+ * Zero the area past i_size but still within an allocated
+ * cluster. This avoids exposing nonzero data on subsequent file
+ * extends.
+ *
+ * We need to call this before i_size is updated on the inode because
+ * otherwise block_write_full_page() will skip writeout of pages past
+ * i_size. The new_i_size parameter is passed for this reason.
+ */
+int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
+				  u64 range_start, u64 range_end)
+{
+	int ret = 0, numpages;
+	struct page **pages = NULL;
+	u64 phys;
+	unsigned int ext_flags;
+	struct super_block *sb = inode->i_sb;
+
+	/*
+	 * File systems which don't support sparse files zero on every
+	 * extend.
+	 */
+	if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
+		return 0;
+
+	pages = kcalloc(ocfs2_pages_per_cluster(sb),
+			sizeof(struct page *), GFP_NOFS);
+	if (pages == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (range_start == range_end)
+		goto out;
+
+	ret = ocfs2_extent_map_get_blocks(inode,
+					  range_start >> sb->s_blocksize_bits,
+					  &phys, NULL, &ext_flags);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Tail is a hole, or is marked unwritten. In either case, we
+	 * can count on read and write to return/push zero's.
+	 */
+	if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN)
+		goto out;
+
+	ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
+				   &numpages);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
+				 numpages, phys, handle);
+
+	/*
+	 * Initiate writeout of the pages we zero'd here. We don't
+	 * wait on them - the truncate_inode_pages() call later will
+	 * do that for us.
+	 */
+	ret = filemap_fdatawrite_range(inode->i_mapping, range_start,
+				       range_end - 1);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	kfree(pages);
+
+	return ret;
+}
+
+static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
+					     struct ocfs2_dinode *di)
+{
+	unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
+	unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
+
+	if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
+		memset(&di->id2, 0, blocksize -
+				    offsetof(struct ocfs2_dinode, id2) -
+				    xattrsize);
+	else
+		memset(&di->id2, 0, blocksize -
+				    offsetof(struct ocfs2_dinode, id2));
+}
+
+void ocfs2_dinode_new_extent_list(struct inode *inode,
+				  struct ocfs2_dinode *di)
+{
+	ocfs2_zero_dinode_id2_with_xattr(inode, di);
+	di->id2.i_list.l_tree_depth = 0;
+	di->id2.i_list.l_next_free_rec = 0;
+	di->id2.i_list.l_count = cpu_to_le16(
+		ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
+}
+
+void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_inline_data *idata = &di->id2.i_data;
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	spin_unlock(&oi->ip_lock);
+
+	/*
+	 * We clear the entire i_data structure here so that all
+	 * fields can be properly initialized.
+	 */
+	ocfs2_zero_dinode_id2_with_xattr(inode, di);
+
+	idata->id_count = cpu_to_le16(
+			ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
+}
+
+int ocfs2_convert_inline_data_to_extents(struct inode *inode,
+					 struct buffer_head *di_bh)
+{
+	int ret, i, has_data, num_pages = 0;
+	int need_free = 0;
+	u32 bit_off, num;
+	handle_t *handle;
+	u64 uninitialized_var(block);
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct page **pages = NULL;
+	loff_t end = osb->s_clustersize;
+	struct ocfs2_extent_tree et;
+	int did_quota = 0;
+
+	has_data = i_size_read(inode) ? 1 : 0;
+
+	if (has_data) {
+		pages = kcalloc(ocfs2_pages_per_cluster(osb->sb),
+				sizeof(struct page *), GFP_NOFS);
+		if (pages == NULL) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			return ret;
+		}
+
+		ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto free_pages;
+		}
+	}
+
+	handle = ocfs2_start_trans(osb,
+				   ocfs2_inline_to_extents_credits(osb->sb));
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	if (has_data) {
+		unsigned int page_end;
+		u64 phys;
+
+		ret = dquot_alloc_space_nodirty(inode,
+				       ocfs2_clusters_to_bytes(osb->sb, 1));
+		if (ret)
+			goto out_commit;
+		did_quota = 1;
+
+		data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
+
+		ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
+					   &num);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+
+		/*
+		 * Save two copies, one for insert, and one that can
+		 * be changed by ocfs2_map_and_dirty_page() below.
+		 */
+		block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
+
+		/*
+		 * Non sparse file systems zero on extend, so no need
+		 * to do that now.
+		 */
+		if (!ocfs2_sparse_alloc(osb) &&
+		    PAGE_CACHE_SIZE < osb->s_clustersize)
+			end = PAGE_CACHE_SIZE;
+
+		ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages);
+		if (ret) {
+			mlog_errno(ret);
+			need_free = 1;
+			goto out_commit;
+		}
+
+		/*
+		 * This should populate the 1st page for us and mark
+		 * it up to date.
+		 */
+		ret = ocfs2_read_inline_data(inode, pages[0], di_bh);
+		if (ret) {
+			mlog_errno(ret);
+			need_free = 1;
+			goto out_unlock;
+		}
+
+		page_end = PAGE_CACHE_SIZE;
+		if (PAGE_CACHE_SIZE > osb->s_clustersize)
+			page_end = osb->s_clustersize;
+
+		for (i = 0; i < num_pages; i++)
+			ocfs2_map_and_dirty_page(inode, handle, 0, page_end,
+						 pages[i], i > 0, &phys);
+	}
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	spin_unlock(&oi->ip_lock);
+
+	ocfs2_update_inode_fsync_trans(handle, inode, 1);
+	ocfs2_dinode_new_extent_list(inode, di);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+	if (has_data) {
+		/*
+		 * An error at this point should be extremely rare. If
+		 * this proves to be false, we could always re-build
+		 * the in-inode data from our pages.
+		 */
+		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
+		ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			need_free = 1;
+			goto out_unlock;
+		}
+
+		inode->i_blocks = ocfs2_inode_sector_count(inode);
+	}
+
+out_unlock:
+	if (pages)
+		ocfs2_unlock_and_free_pages(pages, num_pages);
+
+out_commit:
+	if (ret < 0 && did_quota)
+		dquot_free_space_nodirty(inode,
+					  ocfs2_clusters_to_bytes(osb->sb, 1));
+
+	if (need_free) {
+		if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
+			ocfs2_free_local_alloc_bits(osb, handle, data_ac,
+					bit_off, num);
+		else
+			ocfs2_free_clusters(handle,
+					data_ac->ac_inode,
+					data_ac->ac_bh,
+					ocfs2_clusters_to_blocks(osb->sb, bit_off),
+					num);
+	}
+
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+free_pages:
+	kfree(pages);
+	return ret;
+}
+
+/*
+ * It is expected, that by the time you call this function,
+ * inode->i_size and fe->i_size have been adjusted.
+ *
+ * WARNING: This will kfree the truncate context
+ */
+int ocfs2_commit_truncate(struct ocfs2_super *osb,
+			  struct inode *inode,
+			  struct buffer_head *di_bh)
+{
+	int status = 0, i, flags = 0;
+	u32 new_highest_cpos, range, trunc_cpos, trunc_len, phys_cpos, coff;
+	u64 blkno = 0;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+	struct ocfs2_path *path = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_extent_list *root_el = &(di->id2.i_list);
+	u64 refcount_loc = le64_to_cpu(di->i_refcount_loc);
+	struct ocfs2_extent_tree et;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct ocfs2_refcount_tree *ref_tree = NULL;
+
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
+						     i_size_read(inode));
+
+	path = ocfs2_new_path(di_bh, &di->id2.i_list,
+			      ocfs2_journal_access_di);
+	if (!path) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	ocfs2_extent_map_trunc(inode, new_highest_cpos);
+
+start:
+	/*
+	 * Check that we still have allocation to delete.
+	 */
+	if (OCFS2_I(inode)->ip_clusters == 0) {
+		status = 0;
+		goto bail;
+	}
+
+	/*
+	 * Truncate always works against the rightmost tree branch.
+	 */
+	status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	trace_ocfs2_commit_truncate(
+		(unsigned long long)OCFS2_I(inode)->ip_blkno,
+		new_highest_cpos,
+		OCFS2_I(inode)->ip_clusters,
+		path->p_tree_depth);
+
+	/*
+	 * By now, el will point to the extent list on the bottom most
+	 * portion of this tree. Only the tail record is considered in
+	 * each pass.
+	 *
+	 * We handle the following cases, in order:
+	 * - empty extent: delete the remaining branch
+	 * - remove the entire record
+	 * - remove a partial record
+	 * - no record needs to be removed (truncate has completed)
+	 */
+	el = path_leaf_el(path);
+	if (le16_to_cpu(el->l_next_free_rec) == 0) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has empty extent block at %llu\n",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    (unsigned long long)path_leaf_bh(path)->b_blocknr);
+		status = -EROFS;
+		goto bail;
+	}
+
+	i = le16_to_cpu(el->l_next_free_rec) - 1;
+	rec = &el->l_recs[i];
+	flags = rec->e_flags;
+	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+
+	if (i == 0 && ocfs2_is_empty_extent(rec)) {
+		/*
+		 * Lower levels depend on this never happening, but it's best
+		 * to check it up here before changing the tree.
+		*/
+		if (root_el->l_tree_depth && rec->e_int_clusters == 0) {
+			ocfs2_error(inode->i_sb, "Inode %lu has an empty "
+				    "extent record, depth %u\n", inode->i_ino,
+				    le16_to_cpu(root_el->l_tree_depth));
+			status = -EROFS;
+			goto bail;
+		}
+		trunc_cpos = le32_to_cpu(rec->e_cpos);
+		trunc_len = 0;
+		blkno = 0;
+	} else if (le32_to_cpu(rec->e_cpos) >= new_highest_cpos) {
+		/*
+		 * Truncate entire record.
+		 */
+		trunc_cpos = le32_to_cpu(rec->e_cpos);
+		trunc_len = ocfs2_rec_clusters(el, rec);
+		blkno = le64_to_cpu(rec->e_blkno);
+	} else if (range > new_highest_cpos) {
+		/*
+		 * Partial truncate. it also should be
+		 * the last truncate we're doing.
+		 */
+		trunc_cpos = new_highest_cpos;
+		trunc_len = range - new_highest_cpos;
+		coff = new_highest_cpos - le32_to_cpu(rec->e_cpos);
+		blkno = le64_to_cpu(rec->e_blkno) +
+				ocfs2_clusters_to_blocks(inode->i_sb, coff);
+	} else {
+		/*
+		 * Truncate completed, leave happily.
+		 */
+		status = 0;
+		goto bail;
+	}
+
+	phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
+
+	if ((flags & OCFS2_EXT_REFCOUNTED) && trunc_len && !ref_tree) {
+		status = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
+				&ref_tree, NULL);
+		if (status) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	status = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
+					  phys_cpos, trunc_len, flags, &dealloc,
+					  refcount_loc, true);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	ocfs2_reinit_path(path, 1);
+
+	/*
+	 * The check above will catch the case where we've truncated
+	 * away all allocation.
+	 */
+	goto start;
+
+bail:
+	if (ref_tree)
+		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+
+	ocfs2_schedule_truncate_log_flush(osb, 1);
+
+	ocfs2_run_deallocs(osb, &dealloc);
+
+	ocfs2_free_path(path);
+
+	return status;
+}
+
+/*
+ * 'start' is inclusive, 'end' is not.
+ */
+int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
+			  unsigned int start, unsigned int end, int trunc)
+{
+	int ret;
+	unsigned int numbytes;
+	handle_t *handle;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_inline_data *idata = &di->id2.i_data;
+
+	if (end > i_size_read(inode))
+		end = i_size_read(inode);
+
+	BUG_ON(start > end);
+
+	if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
+	    !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) ||
+	    !ocfs2_supports_inline_data(osb)) {
+		ocfs2_error(inode->i_sb,
+			    "Inline data flags for inode %llu don't agree! "
+			    "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    le16_to_cpu(di->i_dyn_features),
+			    OCFS2_I(inode)->ip_dyn_features,
+			    osb->s_feature_incompat);
+		ret = -EROFS;
+		goto out;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	numbytes = end - start;
+	memset(idata->id_data + start, 0, numbytes);
+
+	/*
+	 * No need to worry about the data page here - it's been
+	 * truncated already and inline data doesn't need it for
+	 * pushing zero's to disk, so we'll let readpage pick it up
+	 * later.
+	 */
+	if (trunc) {
+		i_size_write(inode, start);
+		di->i_size = cpu_to_le64(start);
+	}
+
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
+	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
+
+	di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
+	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+
+	ocfs2_update_inode_fsync_trans(handle, inode, 1);
+	ocfs2_journal_dirty(handle, di_bh);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	return ret;
+}
+
+static int ocfs2_trim_extent(struct super_block *sb,
+			     struct ocfs2_group_desc *gd,
+			     u32 start, u32 count)
+{
+	u64 discard, bcount;
+
+	bcount = ocfs2_clusters_to_blocks(sb, count);
+	discard = le64_to_cpu(gd->bg_blkno) +
+			ocfs2_clusters_to_blocks(sb, start);
+
+	trace_ocfs2_trim_extent(sb, (unsigned long long)discard, bcount);
+
+	return sb_issue_discard(sb, discard, bcount, GFP_NOFS, 0);
+}
+
+static int ocfs2_trim_group(struct super_block *sb,
+			    struct ocfs2_group_desc *gd,
+			    u32 start, u32 max, u32 minbits)
+{
+	int ret = 0, count = 0, next;
+	void *bitmap = gd->bg_bitmap;
+
+	if (le16_to_cpu(gd->bg_free_bits_count) < minbits)
+		return 0;
+
+	trace_ocfs2_trim_group((unsigned long long)le64_to_cpu(gd->bg_blkno),
+			       start, max, minbits);
+
+	while (start < max) {
+		start = ocfs2_find_next_zero_bit(bitmap, max, start);
+		if (start >= max)
+			break;
+		next = ocfs2_find_next_bit(bitmap, max, start);
+
+		if ((next - start) >= minbits) {
+			ret = ocfs2_trim_extent(sb, gd,
+						start, next - start);
+			if (ret < 0) {
+				mlog_errno(ret);
+				break;
+			}
+			count += next - start;
+		}
+		start = next + 1;
+
+		if (fatal_signal_pending(current)) {
+			count = -ERESTARTSYS;
+			break;
+		}
+
+		if ((le16_to_cpu(gd->bg_free_bits_count) - count) < minbits)
+			break;
+	}
+
+	if (ret < 0)
+		count = ret;
+
+	return count;
+}
+
+int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range)
+{
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	u64 start, len, trimmed, first_group, last_group, group;
+	int ret, cnt;
+	u32 first_bit, last_bit, minlen;
+	struct buffer_head *main_bm_bh = NULL;
+	struct inode *main_bm_inode = NULL;
+	struct buffer_head *gd_bh = NULL;
+	struct ocfs2_dinode *main_bm;
+	struct ocfs2_group_desc *gd = NULL;
+
+	start = range->start >> osb->s_clustersize_bits;
+	len = range->len >> osb->s_clustersize_bits;
+	minlen = range->minlen >> osb->s_clustersize_bits;
+
+	if (minlen >= osb->bitmap_cpg || range->len < sb->s_blocksize)
+		return -EINVAL;
+
+	main_bm_inode = ocfs2_get_system_file_inode(osb,
+						    GLOBAL_BITMAP_SYSTEM_INODE,
+						    OCFS2_INVALID_SLOT);
+	if (!main_bm_inode) {
+		ret = -EIO;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mutex_lock(&main_bm_inode->i_mutex);
+
+	ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 0);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_mutex;
+	}
+	main_bm = (struct ocfs2_dinode *)main_bm_bh->b_data;
+
+	if (start >= le32_to_cpu(main_bm->i_clusters)) {
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
+	len = range->len >> osb->s_clustersize_bits;
+	if (start + len > le32_to_cpu(main_bm->i_clusters))
+		len = le32_to_cpu(main_bm->i_clusters) - start;
+
+	trace_ocfs2_trim_fs(start, len, minlen);
+
+	/* Determine first and last group to examine based on start and len */
+	first_group = ocfs2_which_cluster_group(main_bm_inode, start);
+	if (first_group == osb->first_cluster_group_blkno)
+		first_bit = start;
+	else
+		first_bit = start - ocfs2_blocks_to_clusters(sb, first_group);
+	last_group = ocfs2_which_cluster_group(main_bm_inode, start + len - 1);
+	last_bit = osb->bitmap_cpg;
+
+	trimmed = 0;
+	for (group = first_group; group <= last_group;) {
+		if (first_bit + len >= osb->bitmap_cpg)
+			last_bit = osb->bitmap_cpg;
+		else
+			last_bit = first_bit + len;
+
+		ret = ocfs2_read_group_descriptor(main_bm_inode,
+						  main_bm, group,
+						  &gd_bh);
+		if (ret < 0) {
+			mlog_errno(ret);
+			break;
+		}
+
+		gd = (struct ocfs2_group_desc *)gd_bh->b_data;
+		cnt = ocfs2_trim_group(sb, gd, first_bit, last_bit, minlen);
+		brelse(gd_bh);
+		gd_bh = NULL;
+		if (cnt < 0) {
+			ret = cnt;
+			mlog_errno(ret);
+			break;
+		}
+
+		trimmed += cnt;
+		len -= osb->bitmap_cpg - first_bit;
+		first_bit = 0;
+		if (group == osb->first_cluster_group_blkno)
+			group = ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
+		else
+			group += ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
+	}
+	range->len = trimmed * sb->s_blocksize;
+out_unlock:
+	ocfs2_inode_unlock(main_bm_inode, 0);
+	brelse(main_bm_bh);
+out_mutex:
+	mutex_unlock(&main_bm_inode->i_mutex);
+	iput(main_bm_inode);
+out:
+	return ret;
+}
diff --git a/fs/ocfs2/alloc.h b/fs/ocfs2/alloc.h
new file mode 100644
index 000000000..fb09b97db
--- /dev/null
+++ b/fs/ocfs2/alloc.h
@@ -0,0 +1,324 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * alloc.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_ALLOC_H
+#define OCFS2_ALLOC_H
+
+
+/*
+ * For xattr tree leaf, we limit the leaf byte size to be 64K.
+ */
+#define OCFS2_MAX_XATTR_TREE_LEAF_SIZE 65536
+
+/*
+ * ocfs2_extent_tree and ocfs2_extent_tree_operations are used to abstract
+ * the b-tree operations in ocfs2. Now all the b-tree operations are not
+ * limited to ocfs2_dinode only. Any data which need to allocate clusters
+ * to store can use b-tree. And it only needs to implement its ocfs2_extent_tree
+ * and operation.
+ *
+ * ocfs2_extent_tree becomes the first-class object for extent tree
+ * manipulation.  Callers of the alloc.c code need to fill it via one of
+ * the ocfs2_init_*_extent_tree() operations below.
+ *
+ * ocfs2_extent_tree contains info for the root of the b-tree, it must have a
+ * root ocfs2_extent_list and a root_bh so that they can be used in the b-tree
+ * functions.  It needs the ocfs2_caching_info structure associated with
+ * I/O on the tree.  With metadata ecc, we now call different journal_access
+ * functions for each type of metadata, so it must have the
+ * root_journal_access function.
+ * ocfs2_extent_tree_operations abstract the normal operations we do for
+ * the root of extent b-tree.
+ */
+struct ocfs2_extent_tree_operations;
+struct ocfs2_extent_tree {
+	struct ocfs2_extent_tree_operations	*et_ops;
+	struct buffer_head			*et_root_bh;
+	struct ocfs2_extent_list		*et_root_el;
+	struct ocfs2_caching_info		*et_ci;
+	ocfs2_journal_access_func		et_root_journal_access;
+	void					*et_object;
+	unsigned int				et_max_leaf_clusters;
+};
+
+/*
+ * ocfs2_init_*_extent_tree() will fill an ocfs2_extent_tree from the
+ * specified object buffer.
+ */
+void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
+				   struct ocfs2_caching_info *ci,
+				   struct buffer_head *bh);
+void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
+				       struct ocfs2_caching_info *ci,
+				       struct buffer_head *bh);
+struct ocfs2_xattr_value_buf;
+void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
+					struct ocfs2_caching_info *ci,
+					struct ocfs2_xattr_value_buf *vb);
+void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
+				    struct ocfs2_caching_info *ci,
+				    struct buffer_head *bh);
+void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *bh);
+
+/*
+ * Read an extent block into *bh.  If *bh is NULL, a bh will be
+ * allocated.  This is a cached read.  The extent block will be validated
+ * with ocfs2_validate_extent_block().
+ */
+int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
+			    struct buffer_head **bh);
+
+struct ocfs2_alloc_context;
+int ocfs2_insert_extent(handle_t *handle,
+			struct ocfs2_extent_tree *et,
+			u32 cpos,
+			u64 start_blk,
+			u32 new_clusters,
+			u8 flags,
+			struct ocfs2_alloc_context *meta_ac);
+
+enum ocfs2_alloc_restarted {
+	RESTART_NONE = 0,
+	RESTART_TRANS,
+	RESTART_META
+};
+int ocfs2_add_clusters_in_btree(handle_t *handle,
+				struct ocfs2_extent_tree *et,
+				u32 *logical_offset,
+				u32 clusters_to_add,
+				int mark_unwritten,
+				struct ocfs2_alloc_context *data_ac,
+				struct ocfs2_alloc_context *meta_ac,
+				enum ocfs2_alloc_restarted *reason_ret);
+struct ocfs2_cached_dealloc_ctxt;
+struct ocfs2_path;
+int ocfs2_split_extent(handle_t *handle,
+		       struct ocfs2_extent_tree *et,
+		       struct ocfs2_path *path,
+		       int split_index,
+		       struct ocfs2_extent_rec *split_rec,
+		       struct ocfs2_alloc_context *meta_ac,
+		       struct ocfs2_cached_dealloc_ctxt *dealloc);
+int ocfs2_mark_extent_written(struct inode *inode,
+			      struct ocfs2_extent_tree *et,
+			      handle_t *handle, u32 cpos, u32 len, u32 phys,
+			      struct ocfs2_alloc_context *meta_ac,
+			      struct ocfs2_cached_dealloc_ctxt *dealloc);
+int ocfs2_change_extent_flag(handle_t *handle,
+			     struct ocfs2_extent_tree *et,
+			     u32 cpos, u32 len, u32 phys,
+			     struct ocfs2_alloc_context *meta_ac,
+			     struct ocfs2_cached_dealloc_ctxt *dealloc,
+			     int new_flags, int clear_flags);
+int ocfs2_remove_extent(handle_t *handle, struct ocfs2_extent_tree *et,
+			u32 cpos, u32 len,
+			struct ocfs2_alloc_context *meta_ac,
+			struct ocfs2_cached_dealloc_ctxt *dealloc);
+int ocfs2_remove_btree_range(struct inode *inode,
+			     struct ocfs2_extent_tree *et,
+			     u32 cpos, u32 phys_cpos, u32 len, int flags,
+			     struct ocfs2_cached_dealloc_ctxt *dealloc,
+			     u64 refcount_loc, bool refcount_tree_locked);
+
+int ocfs2_num_free_extents(struct ocfs2_super *osb,
+			   struct ocfs2_extent_tree *et);
+
+/*
+ * how many new metadata chunks would an allocation need at maximum?
+ *
+ * Please note that the caller must make sure that root_el is the root
+ * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
+ * the result may be wrong.
+ */
+static inline int ocfs2_extend_meta_needed(struct ocfs2_extent_list *root_el)
+{
+	/*
+	 * Rather than do all the work of determining how much we need
+	 * (involves a ton of reads and locks), just ask for the
+	 * maximal limit.  That's a tree depth shift.  So, one block for
+	 * level of the tree (current l_tree_depth), one block for the
+	 * new tree_depth==0 extent_block, and one block at the new
+	 * top-of-the tree.
+	 */
+	return le16_to_cpu(root_el->l_tree_depth) + 2;
+}
+
+void ocfs2_dinode_new_extent_list(struct inode *inode, struct ocfs2_dinode *di);
+void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di);
+int ocfs2_convert_inline_data_to_extents(struct inode *inode,
+					 struct buffer_head *di_bh);
+
+int ocfs2_truncate_log_init(struct ocfs2_super *osb);
+void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb);
+void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
+				       int cancel);
+int ocfs2_flush_truncate_log(struct ocfs2_super *osb);
+int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
+				      int slot_num,
+				      struct ocfs2_dinode **tl_copy);
+int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
+					 struct ocfs2_dinode *tl_copy);
+int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb);
+int ocfs2_truncate_log_append(struct ocfs2_super *osb,
+			      handle_t *handle,
+			      u64 start_blk,
+			      unsigned int num_clusters);
+int __ocfs2_flush_truncate_log(struct ocfs2_super *osb);
+
+/*
+ * Process local structure which describes the block unlinks done
+ * during an operation. This is populated via
+ * ocfs2_cache_block_dealloc().
+ *
+ * ocfs2_run_deallocs() should be called after the potentially
+ * de-allocating routines. No journal handles should be open, and most
+ * locks should have been dropped.
+ */
+struct ocfs2_cached_dealloc_ctxt {
+	struct ocfs2_per_slot_free_list		*c_first_suballocator;
+	struct ocfs2_cached_block_free 		*c_global_allocator;
+};
+static inline void ocfs2_init_dealloc_ctxt(struct ocfs2_cached_dealloc_ctxt *c)
+{
+	c->c_first_suballocator = NULL;
+	c->c_global_allocator = NULL;
+}
+int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
+				u64 blkno, unsigned int bit);
+int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
+			      int type, int slot, u64 suballoc, u64 blkno,
+			      unsigned int bit);
+static inline int ocfs2_dealloc_has_cluster(struct ocfs2_cached_dealloc_ctxt *c)
+{
+	return c->c_global_allocator != NULL;
+}
+int ocfs2_run_deallocs(struct ocfs2_super *osb,
+		       struct ocfs2_cached_dealloc_ctxt *ctxt);
+
+struct ocfs2_truncate_context {
+	struct ocfs2_cached_dealloc_ctxt tc_dealloc;
+	int tc_ext_alloc_locked; /* is it cluster locked? */
+	/* these get destroyed once it's passed to ocfs2_commit_truncate. */
+	struct buffer_head *tc_last_eb_bh;
+};
+
+int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
+				  u64 range_start, u64 range_end);
+int ocfs2_commit_truncate(struct ocfs2_super *osb,
+			  struct inode *inode,
+			  struct buffer_head *di_bh);
+int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
+			  unsigned int start, unsigned int end, int trunc);
+
+int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
+		    struct ocfs2_extent_list *root_el, u32 cpos,
+		    struct buffer_head **leaf_bh);
+int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster);
+
+int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range);
+/*
+ * Helper function to look at the # of clusters in an extent record.
+ */
+static inline unsigned int ocfs2_rec_clusters(struct ocfs2_extent_list *el,
+					      struct ocfs2_extent_rec *rec)
+{
+	/*
+	 * Cluster count in extent records is slightly different
+	 * between interior nodes and leaf nodes. This is to support
+	 * unwritten extents which need a flags field in leaf node
+	 * records, thus shrinking the available space for a clusters
+	 * field.
+	 */
+	if (el->l_tree_depth)
+		return le32_to_cpu(rec->e_int_clusters);
+	else
+		return le16_to_cpu(rec->e_leaf_clusters);
+}
+
+/*
+ * This is only valid for leaf nodes, which are the only ones that can
+ * have empty extents anyway.
+ */
+static inline int ocfs2_is_empty_extent(struct ocfs2_extent_rec *rec)
+{
+	return !rec->e_leaf_clusters;
+}
+
+int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
+		     struct page **pages, int *num);
+void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
+			      unsigned int from, unsigned int to,
+			      struct page *page, int zero, u64 *phys);
+/*
+ * Structures which describe a path through a btree, and functions to
+ * manipulate them.
+ *
+ * The idea here is to be as generic as possible with the tree
+ * manipulation code.
+ */
+struct ocfs2_path_item {
+	struct buffer_head		*bh;
+	struct ocfs2_extent_list	*el;
+};
+
+#define OCFS2_MAX_PATH_DEPTH	5
+
+struct ocfs2_path {
+	int				p_tree_depth;
+	ocfs2_journal_access_func	p_root_access;
+	struct ocfs2_path_item		p_node[OCFS2_MAX_PATH_DEPTH];
+};
+
+#define path_root_bh(_path) ((_path)->p_node[0].bh)
+#define path_root_el(_path) ((_path)->p_node[0].el)
+#define path_root_access(_path)((_path)->p_root_access)
+#define path_leaf_bh(_path) ((_path)->p_node[(_path)->p_tree_depth].bh)
+#define path_leaf_el(_path) ((_path)->p_node[(_path)->p_tree_depth].el)
+#define path_num_items(_path) ((_path)->p_tree_depth + 1)
+
+void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root);
+void ocfs2_free_path(struct ocfs2_path *path);
+int ocfs2_find_path(struct ocfs2_caching_info *ci,
+		    struct ocfs2_path *path,
+		    u32 cpos);
+struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path);
+struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et);
+int ocfs2_path_bh_journal_access(handle_t *handle,
+				 struct ocfs2_caching_info *ci,
+				 struct ocfs2_path *path,
+				 int idx);
+int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
+			      handle_t *handle,
+			      struct ocfs2_path *path);
+int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
+				   struct ocfs2_path *path, u32 *cpos);
+int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
+				  struct ocfs2_path *path, u32 *cpos);
+int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
+			    struct ocfs2_path *left,
+			    struct ocfs2_path *right);
+#endif /* OCFS2_ALLOC_H */
diff --git a/fs/ocfs2/aops.c b/fs/ocfs2/aops.c
new file mode 100644
index 000000000..f906a250d
--- /dev/null
+++ b/fs/ocfs2/aops.c
@@ -0,0 +1,2448 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/pagemap.h>
+#include <asm/byteorder.h>
+#include <linux/swap.h>
+#include <linux/pipe_fs_i.h>
+#include <linux/mpage.h>
+#include <linux/quotaops.h>
+#include <linux/blkdev.h>
+#include <linux/uio.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "aops.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "file.h"
+#include "inode.h"
+#include "journal.h"
+#include "suballoc.h"
+#include "super.h"
+#include "symlink.h"
+#include "refcounttree.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+#include "dir.h"
+#include "namei.h"
+#include "sysfile.h"
+
+static int ocfs2_symlink_get_block(struct inode *inode, sector_t iblock,
+				   struct buffer_head *bh_result, int create)
+{
+	int err = -EIO;
+	int status;
+	struct ocfs2_dinode *fe = NULL;
+	struct buffer_head *bh = NULL;
+	struct buffer_head *buffer_cache_bh = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	void *kaddr;
+
+	trace_ocfs2_symlink_get_block(
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			(unsigned long long)iblock, bh_result, create);
+
+	BUG_ON(ocfs2_inode_is_fast_symlink(inode));
+
+	if ((iblock << inode->i_sb->s_blocksize_bits) > PATH_MAX + 1) {
+		mlog(ML_ERROR, "block offset > PATH_MAX: %llu",
+		     (unsigned long long)iblock);
+		goto bail;
+	}
+
+	status = ocfs2_read_inode_block(inode, &bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	fe = (struct ocfs2_dinode *) bh->b_data;
+
+	if ((u64)iblock >= ocfs2_clusters_to_blocks(inode->i_sb,
+						    le32_to_cpu(fe->i_clusters))) {
+		err = -ENOMEM;
+		mlog(ML_ERROR, "block offset is outside the allocated size: "
+		     "%llu\n", (unsigned long long)iblock);
+		goto bail;
+	}
+
+	/* We don't use the page cache to create symlink data, so if
+	 * need be, copy it over from the buffer cache. */
+	if (!buffer_uptodate(bh_result) && ocfs2_inode_is_new(inode)) {
+		u64 blkno = le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) +
+			    iblock;
+		buffer_cache_bh = sb_getblk(osb->sb, blkno);
+		if (!buffer_cache_bh) {
+			err = -ENOMEM;
+			mlog(ML_ERROR, "couldn't getblock for symlink!\n");
+			goto bail;
+		}
+
+		/* we haven't locked out transactions, so a commit
+		 * could've happened. Since we've got a reference on
+		 * the bh, even if it commits while we're doing the
+		 * copy, the data is still good. */
+		if (buffer_jbd(buffer_cache_bh)
+		    && ocfs2_inode_is_new(inode)) {
+			kaddr = kmap_atomic(bh_result->b_page);
+			if (!kaddr) {
+				mlog(ML_ERROR, "couldn't kmap!\n");
+				goto bail;
+			}
+			memcpy(kaddr + (bh_result->b_size * iblock),
+			       buffer_cache_bh->b_data,
+			       bh_result->b_size);
+			kunmap_atomic(kaddr);
+			set_buffer_uptodate(bh_result);
+		}
+		brelse(buffer_cache_bh);
+	}
+
+	map_bh(bh_result, inode->i_sb,
+	       le64_to_cpu(fe->id2.i_list.l_recs[0].e_blkno) + iblock);
+
+	err = 0;
+
+bail:
+	brelse(bh);
+
+	return err;
+}
+
+int ocfs2_get_block(struct inode *inode, sector_t iblock,
+		    struct buffer_head *bh_result, int create)
+{
+	int err = 0;
+	unsigned int ext_flags;
+	u64 max_blocks = bh_result->b_size >> inode->i_blkbits;
+	u64 p_blkno, count, past_eof;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	trace_ocfs2_get_block((unsigned long long)OCFS2_I(inode)->ip_blkno,
+			      (unsigned long long)iblock, bh_result, create);
+
+	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
+		mlog(ML_NOTICE, "get_block on system inode 0x%p (%lu)\n",
+		     inode, inode->i_ino);
+
+	if (S_ISLNK(inode->i_mode)) {
+		/* this always does I/O for some reason. */
+		err = ocfs2_symlink_get_block(inode, iblock, bh_result, create);
+		goto bail;
+	}
+
+	err = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno, &count,
+					  &ext_flags);
+	if (err) {
+		mlog(ML_ERROR, "Error %d from get_blocks(0x%p, %llu, 1, "
+		     "%llu, NULL)\n", err, inode, (unsigned long long)iblock,
+		     (unsigned long long)p_blkno);
+		goto bail;
+	}
+
+	if (max_blocks < count)
+		count = max_blocks;
+
+	/*
+	 * ocfs2 never allocates in this function - the only time we
+	 * need to use BH_New is when we're extending i_size on a file
+	 * system which doesn't support holes, in which case BH_New
+	 * allows __block_write_begin() to zero.
+	 *
+	 * If we see this on a sparse file system, then a truncate has
+	 * raced us and removed the cluster. In this case, we clear
+	 * the buffers dirty and uptodate bits and let the buffer code
+	 * ignore it as a hole.
+	 */
+	if (create && p_blkno == 0 && ocfs2_sparse_alloc(osb)) {
+		clear_buffer_dirty(bh_result);
+		clear_buffer_uptodate(bh_result);
+		goto bail;
+	}
+
+	/* Treat the unwritten extent as a hole for zeroing purposes. */
+	if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
+		map_bh(bh_result, inode->i_sb, p_blkno);
+
+	bh_result->b_size = count << inode->i_blkbits;
+
+	if (!ocfs2_sparse_alloc(osb)) {
+		if (p_blkno == 0) {
+			err = -EIO;
+			mlog(ML_ERROR,
+			     "iblock = %llu p_blkno = %llu blkno=(%llu)\n",
+			     (unsigned long long)iblock,
+			     (unsigned long long)p_blkno,
+			     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+			mlog(ML_ERROR, "Size %llu, clusters %u\n", (unsigned long long)i_size_read(inode), OCFS2_I(inode)->ip_clusters);
+			dump_stack();
+			goto bail;
+		}
+	}
+
+	past_eof = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
+
+	trace_ocfs2_get_block_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
+				  (unsigned long long)past_eof);
+	if (create && (iblock >= past_eof))
+		set_buffer_new(bh_result);
+
+bail:
+	if (err < 0)
+		err = -EIO;
+
+	return err;
+}
+
+int ocfs2_read_inline_data(struct inode *inode, struct page *page,
+			   struct buffer_head *di_bh)
+{
+	void *kaddr;
+	loff_t size;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	if (!(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL)) {
+		ocfs2_error(inode->i_sb, "Inode %llu lost inline data flag",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno);
+		return -EROFS;
+	}
+
+	size = i_size_read(inode);
+
+	if (size > PAGE_CACHE_SIZE ||
+	    size > ocfs2_max_inline_data_with_xattr(inode->i_sb, di)) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has with inline data has bad size: %Lu",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    (unsigned long long)size);
+		return -EROFS;
+	}
+
+	kaddr = kmap_atomic(page);
+	if (size)
+		memcpy(kaddr, di->id2.i_data.id_data, size);
+	/* Clear the remaining part of the page */
+	memset(kaddr + size, 0, PAGE_CACHE_SIZE - size);
+	flush_dcache_page(page);
+	kunmap_atomic(kaddr);
+
+	SetPageUptodate(page);
+
+	return 0;
+}
+
+static int ocfs2_readpage_inline(struct inode *inode, struct page *page)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+
+	BUG_ON(!PageLocked(page));
+	BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
+
+	ret = ocfs2_read_inode_block(inode, &di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_inline_data(inode, page, di_bh);
+out:
+	unlock_page(page);
+
+	brelse(di_bh);
+	return ret;
+}
+
+static int ocfs2_readpage(struct file *file, struct page *page)
+{
+	struct inode *inode = page->mapping->host;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	loff_t start = (loff_t)page->index << PAGE_CACHE_SHIFT;
+	int ret, unlock = 1;
+
+	trace_ocfs2_readpage((unsigned long long)oi->ip_blkno,
+			     (page ? page->index : 0));
+
+	ret = ocfs2_inode_lock_with_page(inode, NULL, 0, page);
+	if (ret != 0) {
+		if (ret == AOP_TRUNCATED_PAGE)
+			unlock = 0;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
+		/*
+		 * Unlock the page and cycle ip_alloc_sem so that we don't
+		 * busyloop waiting for ip_alloc_sem to unlock
+		 */
+		ret = AOP_TRUNCATED_PAGE;
+		unlock_page(page);
+		unlock = 0;
+		down_read(&oi->ip_alloc_sem);
+		up_read(&oi->ip_alloc_sem);
+		goto out_inode_unlock;
+	}
+
+	/*
+	 * i_size might have just been updated as we grabed the meta lock.  We
+	 * might now be discovering a truncate that hit on another node.
+	 * block_read_full_page->get_block freaks out if it is asked to read
+	 * beyond the end of a file, so we check here.  Callers
+	 * (generic_file_read, vm_ops->fault) are clever enough to check i_size
+	 * and notice that the page they just read isn't needed.
+	 *
+	 * XXX sys_readahead() seems to get that wrong?
+	 */
+	if (start >= i_size_read(inode)) {
+		zero_user(page, 0, PAGE_SIZE);
+		SetPageUptodate(page);
+		ret = 0;
+		goto out_alloc;
+	}
+
+	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		ret = ocfs2_readpage_inline(inode, page);
+	else
+		ret = block_read_full_page(page, ocfs2_get_block);
+	unlock = 0;
+
+out_alloc:
+	up_read(&OCFS2_I(inode)->ip_alloc_sem);
+out_inode_unlock:
+	ocfs2_inode_unlock(inode, 0);
+out:
+	if (unlock)
+		unlock_page(page);
+	return ret;
+}
+
+/*
+ * This is used only for read-ahead. Failures or difficult to handle
+ * situations are safe to ignore.
+ *
+ * Right now, we don't bother with BH_Boundary - in-inode extent lists
+ * are quite large (243 extents on 4k blocks), so most inodes don't
+ * grow out to a tree. If need be, detecting boundary extents could
+ * trivially be added in a future version of ocfs2_get_block().
+ */
+static int ocfs2_readpages(struct file *filp, struct address_space *mapping,
+			   struct list_head *pages, unsigned nr_pages)
+{
+	int ret, err = -EIO;
+	struct inode *inode = mapping->host;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	loff_t start;
+	struct page *last;
+
+	/*
+	 * Use the nonblocking flag for the dlm code to avoid page
+	 * lock inversion, but don't bother with retrying.
+	 */
+	ret = ocfs2_inode_lock_full(inode, NULL, 0, OCFS2_LOCK_NONBLOCK);
+	if (ret)
+		return err;
+
+	if (down_read_trylock(&oi->ip_alloc_sem) == 0) {
+		ocfs2_inode_unlock(inode, 0);
+		return err;
+	}
+
+	/*
+	 * Don't bother with inline-data. There isn't anything
+	 * to read-ahead in that case anyway...
+	 */
+	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		goto out_unlock;
+
+	/*
+	 * Check whether a remote node truncated this file - we just
+	 * drop out in that case as it's not worth handling here.
+	 */
+	last = list_entry(pages->prev, struct page, lru);
+	start = (loff_t)last->index << PAGE_CACHE_SHIFT;
+	if (start >= i_size_read(inode))
+		goto out_unlock;
+
+	err = mpage_readpages(mapping, pages, nr_pages, ocfs2_get_block);
+
+out_unlock:
+	up_read(&oi->ip_alloc_sem);
+	ocfs2_inode_unlock(inode, 0);
+
+	return err;
+}
+
+/* Note: Because we don't support holes, our allocation has
+ * already happened (allocation writes zeros to the file data)
+ * so we don't have to worry about ordered writes in
+ * ocfs2_writepage.
+ *
+ * ->writepage is called during the process of invalidating the page cache
+ * during blocked lock processing.  It can't block on any cluster locks
+ * to during block mapping.  It's relying on the fact that the block
+ * mapping can't have disappeared under the dirty pages that it is
+ * being asked to write back.
+ */
+static int ocfs2_writepage(struct page *page, struct writeback_control *wbc)
+{
+	trace_ocfs2_writepage(
+		(unsigned long long)OCFS2_I(page->mapping->host)->ip_blkno,
+		page->index);
+
+	return block_write_full_page(page, ocfs2_get_block, wbc);
+}
+
+/* Taken from ext3. We don't necessarily need the full blown
+ * functionality yet, but IMHO it's better to cut and paste the whole
+ * thing so we can avoid introducing our own bugs (and easily pick up
+ * their fixes when they happen) --Mark */
+int walk_page_buffers(	handle_t *handle,
+			struct buffer_head *head,
+			unsigned from,
+			unsigned to,
+			int *partial,
+			int (*fn)(	handle_t *handle,
+					struct buffer_head *bh))
+{
+	struct buffer_head *bh;
+	unsigned block_start, block_end;
+	unsigned blocksize = head->b_size;
+	int err, ret = 0;
+	struct buffer_head *next;
+
+	for (	bh = head, block_start = 0;
+		ret == 0 && (bh != head || !block_start);
+	    	block_start = block_end, bh = next)
+	{
+		next = bh->b_this_page;
+		block_end = block_start + blocksize;
+		if (block_end <= from || block_start >= to) {
+			if (partial && !buffer_uptodate(bh))
+				*partial = 1;
+			continue;
+		}
+		err = (*fn)(handle, bh);
+		if (!ret)
+			ret = err;
+	}
+	return ret;
+}
+
+static sector_t ocfs2_bmap(struct address_space *mapping, sector_t block)
+{
+	sector_t status;
+	u64 p_blkno = 0;
+	int err = 0;
+	struct inode *inode = mapping->host;
+
+	trace_ocfs2_bmap((unsigned long long)OCFS2_I(inode)->ip_blkno,
+			 (unsigned long long)block);
+
+	/* We don't need to lock journal system files, since they aren't
+	 * accessed concurrently from multiple nodes.
+	 */
+	if (!INODE_JOURNAL(inode)) {
+		err = ocfs2_inode_lock(inode, NULL, 0);
+		if (err) {
+			if (err != -ENOENT)
+				mlog_errno(err);
+			goto bail;
+		}
+		down_read(&OCFS2_I(inode)->ip_alloc_sem);
+	}
+
+	if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
+		err = ocfs2_extent_map_get_blocks(inode, block, &p_blkno, NULL,
+						  NULL);
+
+	if (!INODE_JOURNAL(inode)) {
+		up_read(&OCFS2_I(inode)->ip_alloc_sem);
+		ocfs2_inode_unlock(inode, 0);
+	}
+
+	if (err) {
+		mlog(ML_ERROR, "get_blocks() failed, block = %llu\n",
+		     (unsigned long long)block);
+		mlog_errno(err);
+		goto bail;
+	}
+
+bail:
+	status = err ? 0 : p_blkno;
+
+	return status;
+}
+
+/*
+ * TODO: Make this into a generic get_blocks function.
+ *
+ * From do_direct_io in direct-io.c:
+ *  "So what we do is to permit the ->get_blocks function to populate
+ *   bh.b_size with the size of IO which is permitted at this offset and
+ *   this i_blkbits."
+ *
+ * This function is called directly from get_more_blocks in direct-io.c.
+ *
+ * called like this: dio->get_blocks(dio->inode, fs_startblk,
+ * 					fs_count, map_bh, dio->rw == WRITE);
+ */
+static int ocfs2_direct_IO_get_blocks(struct inode *inode, sector_t iblock,
+				     struct buffer_head *bh_result, int create)
+{
+	int ret;
+	u32 cpos = 0;
+	int alloc_locked = 0;
+	u64 p_blkno, inode_blocks, contig_blocks;
+	unsigned int ext_flags;
+	unsigned char blocksize_bits = inode->i_sb->s_blocksize_bits;
+	unsigned long max_blocks = bh_result->b_size >> inode->i_blkbits;
+	unsigned long len = bh_result->b_size;
+	unsigned int clusters_to_alloc = 0;
+
+	cpos = ocfs2_blocks_to_clusters(inode->i_sb, iblock);
+
+	/* This function won't even be called if the request isn't all
+	 * nicely aligned and of the right size, so there's no need
+	 * for us to check any of that. */
+
+	inode_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
+
+	/* This figures out the size of the next contiguous block, and
+	 * our logical offset */
+	ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
+					  &contig_blocks, &ext_flags);
+	if (ret) {
+		mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
+		     (unsigned long long)iblock);
+		ret = -EIO;
+		goto bail;
+	}
+
+	/* We should already CoW the refcounted extent in case of create. */
+	BUG_ON(create && (ext_flags & OCFS2_EXT_REFCOUNTED));
+
+	/* allocate blocks if no p_blkno is found, and create == 1 */
+	if (!p_blkno && create) {
+		ret = ocfs2_inode_lock(inode, NULL, 1);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto bail;
+		}
+
+		alloc_locked = 1;
+
+		/* fill hole, allocate blocks can't be larger than the size
+		 * of the hole */
+		clusters_to_alloc = ocfs2_clusters_for_bytes(inode->i_sb, len);
+		if (clusters_to_alloc > contig_blocks)
+			clusters_to_alloc = contig_blocks;
+
+		/* allocate extent and insert them into the extent tree */
+		ret = ocfs2_extend_allocation(inode, cpos,
+				clusters_to_alloc, 0);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto bail;
+		}
+
+		ret = ocfs2_extent_map_get_blocks(inode, iblock, &p_blkno,
+				&contig_blocks, &ext_flags);
+		if (ret < 0) {
+			mlog(ML_ERROR, "get_blocks() failed iblock=%llu\n",
+					(unsigned long long)iblock);
+			ret = -EIO;
+			goto bail;
+		}
+	}
+
+	/*
+	 * get_more_blocks() expects us to describe a hole by clearing
+	 * the mapped bit on bh_result().
+	 *
+	 * Consider an unwritten extent as a hole.
+	 */
+	if (p_blkno && !(ext_flags & OCFS2_EXT_UNWRITTEN))
+		map_bh(bh_result, inode->i_sb, p_blkno);
+	else
+		clear_buffer_mapped(bh_result);
+
+	/* make sure we don't map more than max_blocks blocks here as
+	   that's all the kernel will handle at this point. */
+	if (max_blocks < contig_blocks)
+		contig_blocks = max_blocks;
+	bh_result->b_size = contig_blocks << blocksize_bits;
+bail:
+	if (alloc_locked)
+		ocfs2_inode_unlock(inode, 1);
+	return ret;
+}
+
+/*
+ * ocfs2_dio_end_io is called by the dio core when a dio is finished.  We're
+ * particularly interested in the aio/dio case.  We use the rw_lock DLM lock
+ * to protect io on one node from truncation on another.
+ */
+static void ocfs2_dio_end_io(struct kiocb *iocb,
+			     loff_t offset,
+			     ssize_t bytes,
+			     void *private)
+{
+	struct inode *inode = file_inode(iocb->ki_filp);
+	int level;
+
+	/* this io's submitter should not have unlocked this before we could */
+	BUG_ON(!ocfs2_iocb_is_rw_locked(iocb));
+
+	if (ocfs2_iocb_is_sem_locked(iocb))
+		ocfs2_iocb_clear_sem_locked(iocb);
+
+	if (ocfs2_iocb_is_unaligned_aio(iocb)) {
+		ocfs2_iocb_clear_unaligned_aio(iocb);
+
+		mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio);
+	}
+
+	ocfs2_iocb_clear_rw_locked(iocb);
+
+	level = ocfs2_iocb_rw_locked_level(iocb);
+	ocfs2_rw_unlock(inode, level);
+}
+
+static int ocfs2_releasepage(struct page *page, gfp_t wait)
+{
+	if (!page_has_buffers(page))
+		return 0;
+	return try_to_free_buffers(page);
+}
+
+static int ocfs2_is_overwrite(struct ocfs2_super *osb,
+		struct inode *inode, loff_t offset)
+{
+	int ret = 0;
+	u32 v_cpos = 0;
+	u32 p_cpos = 0;
+	unsigned int num_clusters = 0;
+	unsigned int ext_flags = 0;
+
+	v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset);
+	ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos,
+			&num_clusters, &ext_flags);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN))
+		return 1;
+
+	return 0;
+}
+
+static int ocfs2_direct_IO_zero_extend(struct ocfs2_super *osb,
+		struct inode *inode, loff_t offset,
+		u64 zero_len, int cluster_align)
+{
+	u32 p_cpos = 0;
+	u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, i_size_read(inode));
+	unsigned int num_clusters = 0;
+	unsigned int ext_flags = 0;
+	int ret = 0;
+
+	if (offset <= i_size_read(inode) || cluster_align)
+		return 0;
+
+	ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, &num_clusters,
+			&ext_flags);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
+		u64 s = i_size_read(inode);
+		sector_t sector = (p_cpos << (osb->s_clustersize_bits - 9)) +
+			(do_div(s, osb->s_clustersize) >> 9);
+
+		ret = blkdev_issue_zeroout(osb->sb->s_bdev, sector,
+				zero_len >> 9, GFP_NOFS, false);
+		if (ret < 0)
+			mlog_errno(ret);
+	}
+
+	return ret;
+}
+
+static int ocfs2_direct_IO_extend_no_holes(struct ocfs2_super *osb,
+		struct inode *inode, loff_t offset)
+{
+	u64 zero_start, zero_len, total_zero_len;
+	u32 p_cpos = 0, clusters_to_add;
+	u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, i_size_read(inode));
+	unsigned int num_clusters = 0;
+	unsigned int ext_flags = 0;
+	u32 size_div, offset_div;
+	int ret = 0;
+
+	{
+		u64 o = offset;
+		u64 s = i_size_read(inode);
+
+		offset_div = do_div(o, osb->s_clustersize);
+		size_div = do_div(s, osb->s_clustersize);
+	}
+
+	if (offset <= i_size_read(inode))
+		return 0;
+
+	clusters_to_add = ocfs2_bytes_to_clusters(inode->i_sb, offset) -
+		ocfs2_bytes_to_clusters(inode->i_sb, i_size_read(inode));
+	total_zero_len = offset - i_size_read(inode);
+	if (clusters_to_add)
+		total_zero_len -= offset_div;
+
+	/* Allocate clusters to fill out holes, and this is only needed
+	 * when we add more than one clusters. Otherwise the cluster will
+	 * be allocated during direct IO */
+	if (clusters_to_add > 1) {
+		ret = ocfs2_extend_allocation(inode,
+				OCFS2_I(inode)->ip_clusters,
+				clusters_to_add - 1, 0);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	while (total_zero_len) {
+		ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos, &num_clusters,
+				&ext_flags);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		zero_start = ocfs2_clusters_to_bytes(osb->sb, p_cpos) +
+			size_div;
+		zero_len = ocfs2_clusters_to_bytes(osb->sb, num_clusters) -
+			size_div;
+		zero_len = min(total_zero_len, zero_len);
+
+		if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
+			ret = blkdev_issue_zeroout(osb->sb->s_bdev,
+					zero_start >> 9, zero_len >> 9,
+					GFP_NOFS, false);
+			if (ret < 0) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		total_zero_len -= zero_len;
+		v_cpos += ocfs2_bytes_to_clusters(osb->sb, zero_len + size_div);
+
+		/* Only at first iteration can be cluster not aligned.
+		 * So set size_div to 0 for the rest */
+		size_div = 0;
+	}
+
+out:
+	return ret;
+}
+
+static ssize_t ocfs2_direct_IO_write(struct kiocb *iocb,
+		struct iov_iter *iter,
+		loff_t offset)
+{
+	ssize_t ret = 0;
+	ssize_t written = 0;
+	bool orphaned = false;
+	int is_overwrite = 0;
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file_inode(file)->i_mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *di_bh = NULL;
+	size_t count = iter->count;
+	journal_t *journal = osb->journal->j_journal;
+	u64 zero_len_head, zero_len_tail;
+	int cluster_align_head, cluster_align_tail;
+	loff_t final_size = offset + count;
+	int append_write = offset >= i_size_read(inode) ? 1 : 0;
+	unsigned int num_clusters = 0;
+	unsigned int ext_flags = 0;
+
+	{
+		u64 o = offset;
+		u64 s = i_size_read(inode);
+
+		zero_len_head = do_div(o, 1 << osb->s_clustersize_bits);
+		cluster_align_head = !zero_len_head;
+
+		zero_len_tail = osb->s_clustersize -
+			do_div(s, osb->s_clustersize);
+		if ((offset - i_size_read(inode)) < zero_len_tail)
+			zero_len_tail = offset - i_size_read(inode);
+		cluster_align_tail = !zero_len_tail;
+	}
+
+	/*
+	 * when final_size > inode->i_size, inode->i_size will be
+	 * updated after direct write, so add the inode to orphan
+	 * dir first.
+	 */
+	if (final_size > i_size_read(inode)) {
+		ret = ocfs2_add_inode_to_orphan(osb, inode);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+		orphaned = true;
+	}
+
+	if (append_write) {
+		ret = ocfs2_inode_lock(inode, NULL, 1);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto clean_orphan;
+		}
+
+		/* zeroing out the previously allocated cluster tail
+		 * that but not zeroed */
+		if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+			ret = ocfs2_direct_IO_zero_extend(osb, inode, offset,
+					zero_len_tail, cluster_align_tail);
+		else
+			ret = ocfs2_direct_IO_extend_no_holes(osb, inode,
+					offset);
+		if (ret < 0) {
+			mlog_errno(ret);
+			ocfs2_inode_unlock(inode, 1);
+			goto clean_orphan;
+		}
+
+		is_overwrite = ocfs2_is_overwrite(osb, inode, offset);
+		if (is_overwrite < 0) {
+			mlog_errno(is_overwrite);
+			ocfs2_inode_unlock(inode, 1);
+			goto clean_orphan;
+		}
+
+		ocfs2_inode_unlock(inode, 1);
+	}
+
+	written = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
+				       offset, ocfs2_direct_IO_get_blocks,
+				       ocfs2_dio_end_io, NULL, 0);
+	if (unlikely(written < 0)) {
+		loff_t i_size = i_size_read(inode);
+
+		if (offset + count > i_size) {
+			ret = ocfs2_inode_lock(inode, &di_bh, 1);
+			if (ret < 0) {
+				mlog_errno(ret);
+				goto clean_orphan;
+			}
+
+			if (i_size == i_size_read(inode)) {
+				ret = ocfs2_truncate_file(inode, di_bh,
+						i_size);
+				if (ret < 0) {
+					if (ret != -ENOSPC)
+						mlog_errno(ret);
+
+					ocfs2_inode_unlock(inode, 1);
+					brelse(di_bh);
+					goto clean_orphan;
+				}
+			}
+
+			ocfs2_inode_unlock(inode, 1);
+			brelse(di_bh);
+
+			ret = jbd2_journal_force_commit(journal);
+			if (ret < 0)
+				mlog_errno(ret);
+		}
+	} else if (written > 0 && append_write && !is_overwrite &&
+			!cluster_align_head) {
+		/* zeroing out the allocated cluster head */
+		u32 p_cpos = 0;
+		u32 v_cpos = ocfs2_bytes_to_clusters(osb->sb, offset);
+
+		ret = ocfs2_inode_lock(inode, NULL, 0);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto clean_orphan;
+		}
+
+		ret = ocfs2_get_clusters(inode, v_cpos, &p_cpos,
+				&num_clusters, &ext_flags);
+		if (ret < 0) {
+			mlog_errno(ret);
+			ocfs2_inode_unlock(inode, 0);
+			goto clean_orphan;
+		}
+
+		BUG_ON(!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN));
+
+		ret = blkdev_issue_zeroout(osb->sb->s_bdev,
+				p_cpos << (osb->s_clustersize_bits - 9),
+				zero_len_head >> 9, GFP_NOFS, false);
+		if (ret < 0)
+			mlog_errno(ret);
+
+		ocfs2_inode_unlock(inode, 0);
+	}
+
+clean_orphan:
+	if (orphaned) {
+		int tmp_ret;
+		int update_isize = written > 0 ? 1 : 0;
+		loff_t end = update_isize ? offset + written : 0;
+
+		tmp_ret = ocfs2_del_inode_from_orphan(osb, inode,
+				update_isize, end);
+		if (tmp_ret < 0) {
+			ret = tmp_ret;
+			goto out;
+		}
+
+		tmp_ret = jbd2_journal_force_commit(journal);
+		if (tmp_ret < 0) {
+			ret = tmp_ret;
+			mlog_errno(tmp_ret);
+		}
+	}
+
+out:
+	if (ret >= 0)
+		ret = written;
+	return ret;
+}
+
+static ssize_t ocfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
+			       loff_t offset)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file_inode(file)->i_mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	int full_coherency = !(osb->s_mount_opt &
+			OCFS2_MOUNT_COHERENCY_BUFFERED);
+
+	/*
+	 * Fallback to buffered I/O if we see an inode without
+	 * extents.
+	 */
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		return 0;
+
+	/* Fallback to buffered I/O if we are appending and
+	 * concurrent O_DIRECT writes are allowed.
+	 */
+	if (i_size_read(inode) <= offset && !full_coherency)
+		return 0;
+
+	if (iov_iter_rw(iter) == READ)
+		return __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev,
+					    iter, offset,
+					    ocfs2_direct_IO_get_blocks,
+					    ocfs2_dio_end_io, NULL, 0);
+	else
+		return ocfs2_direct_IO_write(iocb, iter, offset);
+}
+
+static void ocfs2_figure_cluster_boundaries(struct ocfs2_super *osb,
+					    u32 cpos,
+					    unsigned int *start,
+					    unsigned int *end)
+{
+	unsigned int cluster_start = 0, cluster_end = PAGE_CACHE_SIZE;
+
+	if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits)) {
+		unsigned int cpp;
+
+		cpp = 1 << (PAGE_CACHE_SHIFT - osb->s_clustersize_bits);
+
+		cluster_start = cpos % cpp;
+		cluster_start = cluster_start << osb->s_clustersize_bits;
+
+		cluster_end = cluster_start + osb->s_clustersize;
+	}
+
+	BUG_ON(cluster_start > PAGE_SIZE);
+	BUG_ON(cluster_end > PAGE_SIZE);
+
+	if (start)
+		*start = cluster_start;
+	if (end)
+		*end = cluster_end;
+}
+
+/*
+ * 'from' and 'to' are the region in the page to avoid zeroing.
+ *
+ * If pagesize > clustersize, this function will avoid zeroing outside
+ * of the cluster boundary.
+ *
+ * from == to == 0 is code for "zero the entire cluster region"
+ */
+static void ocfs2_clear_page_regions(struct page *page,
+				     struct ocfs2_super *osb, u32 cpos,
+				     unsigned from, unsigned to)
+{
+	void *kaddr;
+	unsigned int cluster_start, cluster_end;
+
+	ocfs2_figure_cluster_boundaries(osb, cpos, &cluster_start, &cluster_end);
+
+	kaddr = kmap_atomic(page);
+
+	if (from || to) {
+		if (from > cluster_start)
+			memset(kaddr + cluster_start, 0, from - cluster_start);
+		if (to < cluster_end)
+			memset(kaddr + to, 0, cluster_end - to);
+	} else {
+		memset(kaddr + cluster_start, 0, cluster_end - cluster_start);
+	}
+
+	kunmap_atomic(kaddr);
+}
+
+/*
+ * Nonsparse file systems fully allocate before we get to the write
+ * code. This prevents ocfs2_write() from tagging the write as an
+ * allocating one, which means ocfs2_map_page_blocks() might try to
+ * read-in the blocks at the tail of our file. Avoid reading them by
+ * testing i_size against each block offset.
+ */
+static int ocfs2_should_read_blk(struct inode *inode, struct page *page,
+				 unsigned int block_start)
+{
+	u64 offset = page_offset(page) + block_start;
+
+	if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+		return 1;
+
+	if (i_size_read(inode) > offset)
+		return 1;
+
+	return 0;
+}
+
+/*
+ * Some of this taken from __block_write_begin(). We already have our
+ * mapping by now though, and the entire write will be allocating or
+ * it won't, so not much need to use BH_New.
+ *
+ * This will also skip zeroing, which is handled externally.
+ */
+int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
+			  struct inode *inode, unsigned int from,
+			  unsigned int to, int new)
+{
+	int ret = 0;
+	struct buffer_head *head, *bh, *wait[2], **wait_bh = wait;
+	unsigned int block_end, block_start;
+	unsigned int bsize = 1 << inode->i_blkbits;
+
+	if (!page_has_buffers(page))
+		create_empty_buffers(page, bsize, 0);
+
+	head = page_buffers(page);
+	for (bh = head, block_start = 0; bh != head || !block_start;
+	     bh = bh->b_this_page, block_start += bsize) {
+		block_end = block_start + bsize;
+
+		clear_buffer_new(bh);
+
+		/*
+		 * Ignore blocks outside of our i/o range -
+		 * they may belong to unallocated clusters.
+		 */
+		if (block_start >= to || block_end <= from) {
+			if (PageUptodate(page))
+				set_buffer_uptodate(bh);
+			continue;
+		}
+
+		/*
+		 * For an allocating write with cluster size >= page
+		 * size, we always write the entire page.
+		 */
+		if (new)
+			set_buffer_new(bh);
+
+		if (!buffer_mapped(bh)) {
+			map_bh(bh, inode->i_sb, *p_blkno);
+			unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
+		}
+
+		if (PageUptodate(page)) {
+			if (!buffer_uptodate(bh))
+				set_buffer_uptodate(bh);
+		} else if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
+			   !buffer_new(bh) &&
+			   ocfs2_should_read_blk(inode, page, block_start) &&
+			   (block_start < from || block_end > to)) {
+			ll_rw_block(READ, 1, &bh);
+			*wait_bh++=bh;
+		}
+
+		*p_blkno = *p_blkno + 1;
+	}
+
+	/*
+	 * If we issued read requests - let them complete.
+	 */
+	while(wait_bh > wait) {
+		wait_on_buffer(*--wait_bh);
+		if (!buffer_uptodate(*wait_bh))
+			ret = -EIO;
+	}
+
+	if (ret == 0 || !new)
+		return ret;
+
+	/*
+	 * If we get -EIO above, zero out any newly allocated blocks
+	 * to avoid exposing stale data.
+	 */
+	bh = head;
+	block_start = 0;
+	do {
+		block_end = block_start + bsize;
+		if (block_end <= from)
+			goto next_bh;
+		if (block_start >= to)
+			break;
+
+		zero_user(page, block_start, bh->b_size);
+		set_buffer_uptodate(bh);
+		mark_buffer_dirty(bh);
+
+next_bh:
+		block_start = block_end;
+		bh = bh->b_this_page;
+	} while (bh != head);
+
+	return ret;
+}
+
+#if (PAGE_CACHE_SIZE >= OCFS2_MAX_CLUSTERSIZE)
+#define OCFS2_MAX_CTXT_PAGES	1
+#else
+#define OCFS2_MAX_CTXT_PAGES	(OCFS2_MAX_CLUSTERSIZE / PAGE_CACHE_SIZE)
+#endif
+
+#define OCFS2_MAX_CLUSTERS_PER_PAGE	(PAGE_CACHE_SIZE / OCFS2_MIN_CLUSTERSIZE)
+
+/*
+ * Describe the state of a single cluster to be written to.
+ */
+struct ocfs2_write_cluster_desc {
+	u32		c_cpos;
+	u32		c_phys;
+	/*
+	 * Give this a unique field because c_phys eventually gets
+	 * filled.
+	 */
+	unsigned	c_new;
+	unsigned	c_unwritten;
+	unsigned	c_needs_zero;
+};
+
+struct ocfs2_write_ctxt {
+	/* Logical cluster position / len of write */
+	u32				w_cpos;
+	u32				w_clen;
+
+	/* First cluster allocated in a nonsparse extend */
+	u32				w_first_new_cpos;
+
+	struct ocfs2_write_cluster_desc	w_desc[OCFS2_MAX_CLUSTERS_PER_PAGE];
+
+	/*
+	 * This is true if page_size > cluster_size.
+	 *
+	 * It triggers a set of special cases during write which might
+	 * have to deal with allocating writes to partial pages.
+	 */
+	unsigned int			w_large_pages;
+
+	/*
+	 * Pages involved in this write.
+	 *
+	 * w_target_page is the page being written to by the user.
+	 *
+	 * w_pages is an array of pages which always contains
+	 * w_target_page, and in the case of an allocating write with
+	 * page_size < cluster size, it will contain zero'd and mapped
+	 * pages adjacent to w_target_page which need to be written
+	 * out in so that future reads from that region will get
+	 * zero's.
+	 */
+	unsigned int			w_num_pages;
+	struct page			*w_pages[OCFS2_MAX_CTXT_PAGES];
+	struct page			*w_target_page;
+
+	/*
+	 * w_target_locked is used for page_mkwrite path indicating no unlocking
+	 * against w_target_page in ocfs2_write_end_nolock.
+	 */
+	unsigned int			w_target_locked:1;
+
+	/*
+	 * ocfs2_write_end() uses this to know what the real range to
+	 * write in the target should be.
+	 */
+	unsigned int			w_target_from;
+	unsigned int			w_target_to;
+
+	/*
+	 * We could use journal_current_handle() but this is cleaner,
+	 * IMHO -Mark
+	 */
+	handle_t			*w_handle;
+
+	struct buffer_head		*w_di_bh;
+
+	struct ocfs2_cached_dealloc_ctxt w_dealloc;
+};
+
+void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages)
+{
+	int i;
+
+	for(i = 0; i < num_pages; i++) {
+		if (pages[i]) {
+			unlock_page(pages[i]);
+			mark_page_accessed(pages[i]);
+			page_cache_release(pages[i]);
+		}
+	}
+}
+
+static void ocfs2_unlock_pages(struct ocfs2_write_ctxt *wc)
+{
+	int i;
+
+	/*
+	 * w_target_locked is only set to true in the page_mkwrite() case.
+	 * The intent is to allow us to lock the target page from write_begin()
+	 * to write_end(). The caller must hold a ref on w_target_page.
+	 */
+	if (wc->w_target_locked) {
+		BUG_ON(!wc->w_target_page);
+		for (i = 0; i < wc->w_num_pages; i++) {
+			if (wc->w_target_page == wc->w_pages[i]) {
+				wc->w_pages[i] = NULL;
+				break;
+			}
+		}
+		mark_page_accessed(wc->w_target_page);
+		page_cache_release(wc->w_target_page);
+	}
+	ocfs2_unlock_and_free_pages(wc->w_pages, wc->w_num_pages);
+}
+
+static void ocfs2_free_write_ctxt(struct ocfs2_write_ctxt *wc)
+{
+	ocfs2_unlock_pages(wc);
+	brelse(wc->w_di_bh);
+	kfree(wc);
+}
+
+static int ocfs2_alloc_write_ctxt(struct ocfs2_write_ctxt **wcp,
+				  struct ocfs2_super *osb, loff_t pos,
+				  unsigned len, struct buffer_head *di_bh)
+{
+	u32 cend;
+	struct ocfs2_write_ctxt *wc;
+
+	wc = kzalloc(sizeof(struct ocfs2_write_ctxt), GFP_NOFS);
+	if (!wc)
+		return -ENOMEM;
+
+	wc->w_cpos = pos >> osb->s_clustersize_bits;
+	wc->w_first_new_cpos = UINT_MAX;
+	cend = (pos + len - 1) >> osb->s_clustersize_bits;
+	wc->w_clen = cend - wc->w_cpos + 1;
+	get_bh(di_bh);
+	wc->w_di_bh = di_bh;
+
+	if (unlikely(PAGE_CACHE_SHIFT > osb->s_clustersize_bits))
+		wc->w_large_pages = 1;
+	else
+		wc->w_large_pages = 0;
+
+	ocfs2_init_dealloc_ctxt(&wc->w_dealloc);
+
+	*wcp = wc;
+
+	return 0;
+}
+
+/*
+ * If a page has any new buffers, zero them out here, and mark them uptodate
+ * and dirty so they'll be written out (in order to prevent uninitialised
+ * block data from leaking). And clear the new bit.
+ */
+static void ocfs2_zero_new_buffers(struct page *page, unsigned from, unsigned to)
+{
+	unsigned int block_start, block_end;
+	struct buffer_head *head, *bh;
+
+	BUG_ON(!PageLocked(page));
+	if (!page_has_buffers(page))
+		return;
+
+	bh = head = page_buffers(page);
+	block_start = 0;
+	do {
+		block_end = block_start + bh->b_size;
+
+		if (buffer_new(bh)) {
+			if (block_end > from && block_start < to) {
+				if (!PageUptodate(page)) {
+					unsigned start, end;
+
+					start = max(from, block_start);
+					end = min(to, block_end);
+
+					zero_user_segment(page, start, end);
+					set_buffer_uptodate(bh);
+				}
+
+				clear_buffer_new(bh);
+				mark_buffer_dirty(bh);
+			}
+		}
+
+		block_start = block_end;
+		bh = bh->b_this_page;
+	} while (bh != head);
+}
+
+/*
+ * Only called when we have a failure during allocating write to write
+ * zero's to the newly allocated region.
+ */
+static void ocfs2_write_failure(struct inode *inode,
+				struct ocfs2_write_ctxt *wc,
+				loff_t user_pos, unsigned user_len)
+{
+	int i;
+	unsigned from = user_pos & (PAGE_CACHE_SIZE - 1),
+		to = user_pos + user_len;
+	struct page *tmppage;
+
+	ocfs2_zero_new_buffers(wc->w_target_page, from, to);
+
+	for(i = 0; i < wc->w_num_pages; i++) {
+		tmppage = wc->w_pages[i];
+
+		if (page_has_buffers(tmppage)) {
+			if (ocfs2_should_order_data(inode))
+				ocfs2_jbd2_file_inode(wc->w_handle, inode);
+
+			block_commit_write(tmppage, from, to);
+		}
+	}
+}
+
+static int ocfs2_prepare_page_for_write(struct inode *inode, u64 *p_blkno,
+					struct ocfs2_write_ctxt *wc,
+					struct page *page, u32 cpos,
+					loff_t user_pos, unsigned user_len,
+					int new)
+{
+	int ret;
+	unsigned int map_from = 0, map_to = 0;
+	unsigned int cluster_start, cluster_end;
+	unsigned int user_data_from = 0, user_data_to = 0;
+
+	ocfs2_figure_cluster_boundaries(OCFS2_SB(inode->i_sb), cpos,
+					&cluster_start, &cluster_end);
+
+	/* treat the write as new if the a hole/lseek spanned across
+	 * the page boundary.
+	 */
+	new = new | ((i_size_read(inode) <= page_offset(page)) &&
+			(page_offset(page) <= user_pos));
+
+	if (page == wc->w_target_page) {
+		map_from = user_pos & (PAGE_CACHE_SIZE - 1);
+		map_to = map_from + user_len;
+
+		if (new)
+			ret = ocfs2_map_page_blocks(page, p_blkno, inode,
+						    cluster_start, cluster_end,
+						    new);
+		else
+			ret = ocfs2_map_page_blocks(page, p_blkno, inode,
+						    map_from, map_to, new);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		user_data_from = map_from;
+		user_data_to = map_to;
+		if (new) {
+			map_from = cluster_start;
+			map_to = cluster_end;
+		}
+	} else {
+		/*
+		 * If we haven't allocated the new page yet, we
+		 * shouldn't be writing it out without copying user
+		 * data. This is likely a math error from the caller.
+		 */
+		BUG_ON(!new);
+
+		map_from = cluster_start;
+		map_to = cluster_end;
+
+		ret = ocfs2_map_page_blocks(page, p_blkno, inode,
+					    cluster_start, cluster_end, new);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	/*
+	 * Parts of newly allocated pages need to be zero'd.
+	 *
+	 * Above, we have also rewritten 'to' and 'from' - as far as
+	 * the rest of the function is concerned, the entire cluster
+	 * range inside of a page needs to be written.
+	 *
+	 * We can skip this if the page is up to date - it's already
+	 * been zero'd from being read in as a hole.
+	 */
+	if (new && !PageUptodate(page))
+		ocfs2_clear_page_regions(page, OCFS2_SB(inode->i_sb),
+					 cpos, user_data_from, user_data_to);
+
+	flush_dcache_page(page);
+
+out:
+	return ret;
+}
+
+/*
+ * This function will only grab one clusters worth of pages.
+ */
+static int ocfs2_grab_pages_for_write(struct address_space *mapping,
+				      struct ocfs2_write_ctxt *wc,
+				      u32 cpos, loff_t user_pos,
+				      unsigned user_len, int new,
+				      struct page *mmap_page)
+{
+	int ret = 0, i;
+	unsigned long start, target_index, end_index, index;
+	struct inode *inode = mapping->host;
+	loff_t last_byte;
+
+	target_index = user_pos >> PAGE_CACHE_SHIFT;
+
+	/*
+	 * Figure out how many pages we'll be manipulating here. For
+	 * non allocating write, we just change the one
+	 * page. Otherwise, we'll need a whole clusters worth.  If we're
+	 * writing past i_size, we only need enough pages to cover the
+	 * last page of the write.
+	 */
+	if (new) {
+		wc->w_num_pages = ocfs2_pages_per_cluster(inode->i_sb);
+		start = ocfs2_align_clusters_to_page_index(inode->i_sb, cpos);
+		/*
+		 * We need the index *past* the last page we could possibly
+		 * touch.  This is the page past the end of the write or
+		 * i_size, whichever is greater.
+		 */
+		last_byte = max(user_pos + user_len, i_size_read(inode));
+		BUG_ON(last_byte < 1);
+		end_index = ((last_byte - 1) >> PAGE_CACHE_SHIFT) + 1;
+		if ((start + wc->w_num_pages) > end_index)
+			wc->w_num_pages = end_index - start;
+	} else {
+		wc->w_num_pages = 1;
+		start = target_index;
+	}
+
+	for(i = 0; i < wc->w_num_pages; i++) {
+		index = start + i;
+
+		if (index == target_index && mmap_page) {
+			/*
+			 * ocfs2_pagemkwrite() is a little different
+			 * and wants us to directly use the page
+			 * passed in.
+			 */
+			lock_page(mmap_page);
+
+			/* Exit and let the caller retry */
+			if (mmap_page->mapping != mapping) {
+				WARN_ON(mmap_page->mapping);
+				unlock_page(mmap_page);
+				ret = -EAGAIN;
+				goto out;
+			}
+
+			page_cache_get(mmap_page);
+			wc->w_pages[i] = mmap_page;
+			wc->w_target_locked = true;
+		} else {
+			wc->w_pages[i] = find_or_create_page(mapping, index,
+							     GFP_NOFS);
+			if (!wc->w_pages[i]) {
+				ret = -ENOMEM;
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+		wait_for_stable_page(wc->w_pages[i]);
+
+		if (index == target_index)
+			wc->w_target_page = wc->w_pages[i];
+	}
+out:
+	if (ret)
+		wc->w_target_locked = false;
+	return ret;
+}
+
+/*
+ * Prepare a single cluster for write one cluster into the file.
+ */
+static int ocfs2_write_cluster(struct address_space *mapping,
+			       u32 phys, unsigned int unwritten,
+			       unsigned int should_zero,
+			       struct ocfs2_alloc_context *data_ac,
+			       struct ocfs2_alloc_context *meta_ac,
+			       struct ocfs2_write_ctxt *wc, u32 cpos,
+			       loff_t user_pos, unsigned user_len)
+{
+	int ret, i, new;
+	u64 v_blkno, p_blkno;
+	struct inode *inode = mapping->host;
+	struct ocfs2_extent_tree et;
+
+	new = phys == 0 ? 1 : 0;
+	if (new) {
+		u32 tmp_pos;
+
+		/*
+		 * This is safe to call with the page locks - it won't take
+		 * any additional semaphores or cluster locks.
+		 */
+		tmp_pos = cpos;
+		ret = ocfs2_add_inode_data(OCFS2_SB(inode->i_sb), inode,
+					   &tmp_pos, 1, 0, wc->w_di_bh,
+					   wc->w_handle, data_ac,
+					   meta_ac, NULL);
+		/*
+		 * This shouldn't happen because we must have already
+		 * calculated the correct meta data allocation required. The
+		 * internal tree allocation code should know how to increase
+		 * transaction credits itself.
+		 *
+		 * If need be, we could handle -EAGAIN for a
+		 * RESTART_TRANS here.
+		 */
+		mlog_bug_on_msg(ret == -EAGAIN,
+				"Inode %llu: EAGAIN return during allocation.\n",
+				(unsigned long long)OCFS2_I(inode)->ip_blkno);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	} else if (unwritten) {
+		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
+					      wc->w_di_bh);
+		ret = ocfs2_mark_extent_written(inode, &et,
+						wc->w_handle, cpos, 1, phys,
+						meta_ac, &wc->w_dealloc);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (should_zero)
+		v_blkno = ocfs2_clusters_to_blocks(inode->i_sb, cpos);
+	else
+		v_blkno = user_pos >> inode->i_sb->s_blocksize_bits;
+
+	/*
+	 * The only reason this should fail is due to an inability to
+	 * find the extent added.
+	 */
+	ret = ocfs2_extent_map_get_blocks(inode, v_blkno, &p_blkno, NULL,
+					  NULL);
+	if (ret < 0) {
+		mlog(ML_ERROR, "Get physical blkno failed for inode %llu, "
+			    "at logical block %llu",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    (unsigned long long)v_blkno);
+		goto out;
+	}
+
+	BUG_ON(p_blkno == 0);
+
+	for(i = 0; i < wc->w_num_pages; i++) {
+		int tmpret;
+
+		tmpret = ocfs2_prepare_page_for_write(inode, &p_blkno, wc,
+						      wc->w_pages[i], cpos,
+						      user_pos, user_len,
+						      should_zero);
+		if (tmpret) {
+			mlog_errno(tmpret);
+			if (ret == 0)
+				ret = tmpret;
+		}
+	}
+
+	/*
+	 * We only have cleanup to do in case of allocating write.
+	 */
+	if (ret && new)
+		ocfs2_write_failure(inode, wc, user_pos, user_len);
+
+out:
+
+	return ret;
+}
+
+static int ocfs2_write_cluster_by_desc(struct address_space *mapping,
+				       struct ocfs2_alloc_context *data_ac,
+				       struct ocfs2_alloc_context *meta_ac,
+				       struct ocfs2_write_ctxt *wc,
+				       loff_t pos, unsigned len)
+{
+	int ret, i;
+	loff_t cluster_off;
+	unsigned int local_len = len;
+	struct ocfs2_write_cluster_desc *desc;
+	struct ocfs2_super *osb = OCFS2_SB(mapping->host->i_sb);
+
+	for (i = 0; i < wc->w_clen; i++) {
+		desc = &wc->w_desc[i];
+
+		/*
+		 * We have to make sure that the total write passed in
+		 * doesn't extend past a single cluster.
+		 */
+		local_len = len;
+		cluster_off = pos & (osb->s_clustersize - 1);
+		if ((cluster_off + local_len) > osb->s_clustersize)
+			local_len = osb->s_clustersize - cluster_off;
+
+		ret = ocfs2_write_cluster(mapping, desc->c_phys,
+					  desc->c_unwritten,
+					  desc->c_needs_zero,
+					  data_ac, meta_ac,
+					  wc, desc->c_cpos, pos, local_len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		len -= local_len;
+		pos += local_len;
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
+/*
+ * ocfs2_write_end() wants to know which parts of the target page it
+ * should complete the write on. It's easiest to compute them ahead of
+ * time when a more complete view of the write is available.
+ */
+static void ocfs2_set_target_boundaries(struct ocfs2_super *osb,
+					struct ocfs2_write_ctxt *wc,
+					loff_t pos, unsigned len, int alloc)
+{
+	struct ocfs2_write_cluster_desc *desc;
+
+	wc->w_target_from = pos & (PAGE_CACHE_SIZE - 1);
+	wc->w_target_to = wc->w_target_from + len;
+
+	if (alloc == 0)
+		return;
+
+	/*
+	 * Allocating write - we may have different boundaries based
+	 * on page size and cluster size.
+	 *
+	 * NOTE: We can no longer compute one value from the other as
+	 * the actual write length and user provided length may be
+	 * different.
+	 */
+
+	if (wc->w_large_pages) {
+		/*
+		 * We only care about the 1st and last cluster within
+		 * our range and whether they should be zero'd or not. Either
+		 * value may be extended out to the start/end of a
+		 * newly allocated cluster.
+		 */
+		desc = &wc->w_desc[0];
+		if (desc->c_needs_zero)
+			ocfs2_figure_cluster_boundaries(osb,
+							desc->c_cpos,
+							&wc->w_target_from,
+							NULL);
+
+		desc = &wc->w_desc[wc->w_clen - 1];
+		if (desc->c_needs_zero)
+			ocfs2_figure_cluster_boundaries(osb,
+							desc->c_cpos,
+							NULL,
+							&wc->w_target_to);
+	} else {
+		wc->w_target_from = 0;
+		wc->w_target_to = PAGE_CACHE_SIZE;
+	}
+}
+
+/*
+ * Populate each single-cluster write descriptor in the write context
+ * with information about the i/o to be done.
+ *
+ * Returns the number of clusters that will have to be allocated, as
+ * well as a worst case estimate of the number of extent records that
+ * would have to be created during a write to an unwritten region.
+ */
+static int ocfs2_populate_write_desc(struct inode *inode,
+				     struct ocfs2_write_ctxt *wc,
+				     unsigned int *clusters_to_alloc,
+				     unsigned int *extents_to_split)
+{
+	int ret;
+	struct ocfs2_write_cluster_desc *desc;
+	unsigned int num_clusters = 0;
+	unsigned int ext_flags = 0;
+	u32 phys = 0;
+	int i;
+
+	*clusters_to_alloc = 0;
+	*extents_to_split = 0;
+
+	for (i = 0; i < wc->w_clen; i++) {
+		desc = &wc->w_desc[i];
+		desc->c_cpos = wc->w_cpos + i;
+
+		if (num_clusters == 0) {
+			/*
+			 * Need to look up the next extent record.
+			 */
+			ret = ocfs2_get_clusters(inode, desc->c_cpos, &phys,
+						 &num_clusters, &ext_flags);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			/* We should already CoW the refcountd extent. */
+			BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
+
+			/*
+			 * Assume worst case - that we're writing in
+			 * the middle of the extent.
+			 *
+			 * We can assume that the write proceeds from
+			 * left to right, in which case the extent
+			 * insert code is smart enough to coalesce the
+			 * next splits into the previous records created.
+			 */
+			if (ext_flags & OCFS2_EXT_UNWRITTEN)
+				*extents_to_split = *extents_to_split + 2;
+		} else if (phys) {
+			/*
+			 * Only increment phys if it doesn't describe
+			 * a hole.
+			 */
+			phys++;
+		}
+
+		/*
+		 * If w_first_new_cpos is < UINT_MAX, we have a non-sparse
+		 * file that got extended.  w_first_new_cpos tells us
+		 * where the newly allocated clusters are so we can
+		 * zero them.
+		 */
+		if (desc->c_cpos >= wc->w_first_new_cpos) {
+			BUG_ON(phys == 0);
+			desc->c_needs_zero = 1;
+		}
+
+		desc->c_phys = phys;
+		if (phys == 0) {
+			desc->c_new = 1;
+			desc->c_needs_zero = 1;
+			*clusters_to_alloc = *clusters_to_alloc + 1;
+		}
+
+		if (ext_flags & OCFS2_EXT_UNWRITTEN) {
+			desc->c_unwritten = 1;
+			desc->c_needs_zero = 1;
+		}
+
+		num_clusters--;
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
+static int ocfs2_write_begin_inline(struct address_space *mapping,
+				    struct inode *inode,
+				    struct ocfs2_write_ctxt *wc)
+{
+	int ret;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct page *page;
+	handle_t *handle;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	page = find_or_create_page(mapping, 0, GFP_NOFS);
+	if (!page) {
+		ocfs2_commit_trans(osb, handle);
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+	/*
+	 * If we don't set w_num_pages then this page won't get unlocked
+	 * and freed on cleanup of the write context.
+	 */
+	wc->w_pages[0] = wc->w_target_page = page;
+	wc->w_num_pages = 1;
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		ocfs2_commit_trans(osb, handle);
+
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL))
+		ocfs2_set_inode_data_inline(inode, di);
+
+	if (!PageUptodate(page)) {
+		ret = ocfs2_read_inline_data(inode, page, wc->w_di_bh);
+		if (ret) {
+			ocfs2_commit_trans(osb, handle);
+
+			goto out;
+		}
+	}
+
+	wc->w_handle = handle;
+out:
+	return ret;
+}
+
+int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size)
+{
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	if (new_size <= le16_to_cpu(di->id2.i_data.id_count))
+		return 1;
+	return 0;
+}
+
+static int ocfs2_try_to_write_inline_data(struct address_space *mapping,
+					  struct inode *inode, loff_t pos,
+					  unsigned len, struct page *mmap_page,
+					  struct ocfs2_write_ctxt *wc)
+{
+	int ret, written = 0;
+	loff_t end = pos + len;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = NULL;
+
+	trace_ocfs2_try_to_write_inline_data((unsigned long long)oi->ip_blkno,
+					     len, (unsigned long long)pos,
+					     oi->ip_dyn_features);
+
+	/*
+	 * Handle inodes which already have inline data 1st.
+	 */
+	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		if (mmap_page == NULL &&
+		    ocfs2_size_fits_inline_data(wc->w_di_bh, end))
+			goto do_inline_write;
+
+		/*
+		 * The write won't fit - we have to give this inode an
+		 * inline extent list now.
+		 */
+		ret = ocfs2_convert_inline_data_to_extents(inode, wc->w_di_bh);
+		if (ret)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Check whether the inode can accept inline data.
+	 */
+	if (oi->ip_clusters != 0 || i_size_read(inode) != 0)
+		return 0;
+
+	/*
+	 * Check whether the write can fit.
+	 */
+	di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
+	if (mmap_page ||
+	    end > ocfs2_max_inline_data_with_xattr(inode->i_sb, di))
+		return 0;
+
+do_inline_write:
+	ret = ocfs2_write_begin_inline(mapping, inode, wc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * This signals to the caller that the data can be written
+	 * inline.
+	 */
+	written = 1;
+out:
+	return written ? written : ret;
+}
+
+/*
+ * This function only does anything for file systems which can't
+ * handle sparse files.
+ *
+ * What we want to do here is fill in any hole between the current end
+ * of allocation and the end of our write. That way the rest of the
+ * write path can treat it as an non-allocating write, which has no
+ * special case code for sparse/nonsparse files.
+ */
+static int ocfs2_expand_nonsparse_inode(struct inode *inode,
+					struct buffer_head *di_bh,
+					loff_t pos, unsigned len,
+					struct ocfs2_write_ctxt *wc)
+{
+	int ret;
+	loff_t newsize = pos + len;
+
+	BUG_ON(ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
+
+	if (newsize <= i_size_read(inode))
+		return 0;
+
+	ret = ocfs2_extend_no_holes(inode, di_bh, newsize, pos);
+	if (ret)
+		mlog_errno(ret);
+
+	wc->w_first_new_cpos =
+		ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
+
+	return ret;
+}
+
+static int ocfs2_zero_tail(struct inode *inode, struct buffer_head *di_bh,
+			   loff_t pos)
+{
+	int ret = 0;
+
+	BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)));
+	if (pos > i_size_read(inode))
+		ret = ocfs2_zero_extend(inode, di_bh, pos);
+
+	return ret;
+}
+
+/*
+ * Try to flush truncate logs if we can free enough clusters from it.
+ * As for return value, "< 0" means error, "0" no space and "1" means
+ * we have freed enough spaces and let the caller try to allocate again.
+ */
+static int ocfs2_try_to_free_truncate_log(struct ocfs2_super *osb,
+					  unsigned int needed)
+{
+	tid_t target;
+	int ret = 0;
+	unsigned int truncated_clusters;
+
+	mutex_lock(&osb->osb_tl_inode->i_mutex);
+	truncated_clusters = osb->truncated_clusters;
+	mutex_unlock(&osb->osb_tl_inode->i_mutex);
+
+	/*
+	 * Check whether we can succeed in allocating if we free
+	 * the truncate log.
+	 */
+	if (truncated_clusters < needed)
+		goto out;
+
+	ret = ocfs2_flush_truncate_log(osb);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (jbd2_journal_start_commit(osb->journal->j_journal, &target)) {
+		jbd2_log_wait_commit(osb->journal->j_journal, target);
+		ret = 1;
+	}
+out:
+	return ret;
+}
+
+int ocfs2_write_begin_nolock(struct file *filp,
+			     struct address_space *mapping,
+			     loff_t pos, unsigned len, unsigned flags,
+			     struct page **pagep, void **fsdata,
+			     struct buffer_head *di_bh, struct page *mmap_page)
+{
+	int ret, cluster_of_pages, credits = OCFS2_INODE_UPDATE_CREDITS;
+	unsigned int clusters_to_alloc, extents_to_split, clusters_need = 0;
+	struct ocfs2_write_ctxt *wc;
+	struct inode *inode = mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di;
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	handle_t *handle;
+	struct ocfs2_extent_tree et;
+	int try_free = 1, ret1;
+
+try_again:
+	ret = ocfs2_alloc_write_ctxt(&wc, osb, pos, len, di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	if (ocfs2_supports_inline_data(osb)) {
+		ret = ocfs2_try_to_write_inline_data(mapping, inode, pos, len,
+						     mmap_page, wc);
+		if (ret == 1) {
+			ret = 0;
+			goto success;
+		}
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (ocfs2_sparse_alloc(osb))
+		ret = ocfs2_zero_tail(inode, di_bh, pos);
+	else
+		ret = ocfs2_expand_nonsparse_inode(inode, di_bh, pos, len,
+						   wc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_check_range_for_refcount(inode, pos, len);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	} else if (ret == 1) {
+		clusters_need = wc->w_clen;
+		ret = ocfs2_refcount_cow(inode, di_bh,
+					 wc->w_cpos, wc->w_clen, UINT_MAX);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_populate_write_desc(inode, wc, &clusters_to_alloc,
+					&extents_to_split);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	clusters_need += clusters_to_alloc;
+
+	di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
+
+	trace_ocfs2_write_begin_nolock(
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			(long long)i_size_read(inode),
+			le32_to_cpu(di->i_clusters),
+			pos, len, flags, mmap_page,
+			clusters_to_alloc, extents_to_split);
+
+	/*
+	 * We set w_target_from, w_target_to here so that
+	 * ocfs2_write_end() knows which range in the target page to
+	 * write out. An allocation requires that we write the entire
+	 * cluster range.
+	 */
+	if (clusters_to_alloc || extents_to_split) {
+		/*
+		 * XXX: We are stretching the limits of
+		 * ocfs2_lock_allocators(). It greatly over-estimates
+		 * the work to be done.
+		 */
+		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode),
+					      wc->w_di_bh);
+		ret = ocfs2_lock_allocators(inode, &et,
+					    clusters_to_alloc, extents_to_split,
+					    &data_ac, &meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (data_ac)
+			data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
+
+		credits = ocfs2_calc_extend_credits(inode->i_sb,
+						    &di->id2.i_list);
+
+	}
+
+	/*
+	 * We have to zero sparse allocated clusters, unwritten extent clusters,
+	 * and non-sparse clusters we just extended.  For non-sparse writes,
+	 * we know zeros will only be needed in the first and/or last cluster.
+	 */
+	if (clusters_to_alloc || extents_to_split ||
+	    (wc->w_clen && (wc->w_desc[0].c_needs_zero ||
+			    wc->w_desc[wc->w_clen - 1].c_needs_zero)))
+		cluster_of_pages = 1;
+	else
+		cluster_of_pages = 0;
+
+	ocfs2_set_target_boundaries(osb, wc, pos, len, cluster_of_pages);
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	wc->w_handle = handle;
+
+	if (clusters_to_alloc) {
+		ret = dquot_alloc_space_nodirty(inode,
+			ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
+		if (ret)
+			goto out_commit;
+	}
+	/*
+	 * We don't want this to fail in ocfs2_write_end(), so do it
+	 * here.
+	 */
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), wc->w_di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_quota;
+	}
+
+	/*
+	 * Fill our page array first. That way we've grabbed enough so
+	 * that we can zero and flush if we error after adding the
+	 * extent.
+	 */
+	ret = ocfs2_grab_pages_for_write(mapping, wc, wc->w_cpos, pos, len,
+					 cluster_of_pages, mmap_page);
+	if (ret && ret != -EAGAIN) {
+		mlog_errno(ret);
+		goto out_quota;
+	}
+
+	/*
+	 * ocfs2_grab_pages_for_write() returns -EAGAIN if it could not lock
+	 * the target page. In this case, we exit with no error and no target
+	 * page. This will trigger the caller, page_mkwrite(), to re-try
+	 * the operation.
+	 */
+	if (ret == -EAGAIN) {
+		BUG_ON(wc->w_target_page);
+		ret = 0;
+		goto out_quota;
+	}
+
+	ret = ocfs2_write_cluster_by_desc(mapping, data_ac, meta_ac, wc, pos,
+					  len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_quota;
+	}
+
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+success:
+	*pagep = wc->w_target_page;
+	*fsdata = wc;
+	return 0;
+out_quota:
+	if (clusters_to_alloc)
+		dquot_free_space(inode,
+			  ocfs2_clusters_to_bytes(osb->sb, clusters_to_alloc));
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	ocfs2_free_write_ctxt(wc);
+
+	if (data_ac) {
+		ocfs2_free_alloc_context(data_ac);
+		data_ac = NULL;
+	}
+	if (meta_ac) {
+		ocfs2_free_alloc_context(meta_ac);
+		meta_ac = NULL;
+	}
+
+	if (ret == -ENOSPC && try_free) {
+		/*
+		 * Try to free some truncate log so that we can have enough
+		 * clusters to allocate.
+		 */
+		try_free = 0;
+
+		ret1 = ocfs2_try_to_free_truncate_log(osb, clusters_need);
+		if (ret1 == 1)
+			goto try_again;
+
+		if (ret1 < 0)
+			mlog_errno(ret1);
+	}
+
+	return ret;
+}
+
+static int ocfs2_write_begin(struct file *file, struct address_space *mapping,
+			     loff_t pos, unsigned len, unsigned flags,
+			     struct page **pagep, void **fsdata)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	struct inode *inode = mapping->host;
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	/*
+	 * Take alloc sem here to prevent concurrent lookups. That way
+	 * the mapping, zeroing and tree manipulation within
+	 * ocfs2_write() will be safe against ->readpage(). This
+	 * should also serve to lock out allocation from a shared
+	 * writeable region.
+	 */
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ret = ocfs2_write_begin_nolock(file, mapping, pos, len, flags, pagep,
+				       fsdata, di_bh, NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_fail;
+	}
+
+	brelse(di_bh);
+
+	return 0;
+
+out_fail:
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	brelse(di_bh);
+	ocfs2_inode_unlock(inode, 1);
+
+	return ret;
+}
+
+static void ocfs2_write_end_inline(struct inode *inode, loff_t pos,
+				   unsigned len, unsigned *copied,
+				   struct ocfs2_dinode *di,
+				   struct ocfs2_write_ctxt *wc)
+{
+	void *kaddr;
+
+	if (unlikely(*copied < len)) {
+		if (!PageUptodate(wc->w_target_page)) {
+			*copied = 0;
+			return;
+		}
+	}
+
+	kaddr = kmap_atomic(wc->w_target_page);
+	memcpy(di->id2.i_data.id_data + pos, kaddr + pos, *copied);
+	kunmap_atomic(kaddr);
+
+	trace_ocfs2_write_end_inline(
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+	     (unsigned long long)pos, *copied,
+	     le16_to_cpu(di->id2.i_data.id_count),
+	     le16_to_cpu(di->i_dyn_features));
+}
+
+int ocfs2_write_end_nolock(struct address_space *mapping,
+			   loff_t pos, unsigned len, unsigned copied,
+			   struct page *page, void *fsdata)
+{
+	int i;
+	unsigned from, to, start = pos & (PAGE_CACHE_SIZE - 1);
+	struct inode *inode = mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_write_ctxt *wc = fsdata;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)wc->w_di_bh->b_data;
+	handle_t *handle = wc->w_handle;
+	struct page *tmppage;
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		ocfs2_write_end_inline(inode, pos, len, &copied, di, wc);
+		goto out_write_size;
+	}
+
+	if (unlikely(copied < len)) {
+		if (!PageUptodate(wc->w_target_page))
+			copied = 0;
+
+		ocfs2_zero_new_buffers(wc->w_target_page, start+copied,
+				       start+len);
+	}
+	flush_dcache_page(wc->w_target_page);
+
+	for(i = 0; i < wc->w_num_pages; i++) {
+		tmppage = wc->w_pages[i];
+
+		if (tmppage == wc->w_target_page) {
+			from = wc->w_target_from;
+			to = wc->w_target_to;
+
+			BUG_ON(from > PAGE_CACHE_SIZE ||
+			       to > PAGE_CACHE_SIZE ||
+			       to < from);
+		} else {
+			/*
+			 * Pages adjacent to the target (if any) imply
+			 * a hole-filling write in which case we want
+			 * to flush their entire range.
+			 */
+			from = 0;
+			to = PAGE_CACHE_SIZE;
+		}
+
+		if (page_has_buffers(tmppage)) {
+			if (ocfs2_should_order_data(inode))
+				ocfs2_jbd2_file_inode(wc->w_handle, inode);
+			block_commit_write(tmppage, from, to);
+		}
+	}
+
+out_write_size:
+	pos += copied;
+	if (pos > i_size_read(inode)) {
+		i_size_write(inode, pos);
+		mark_inode_dirty(inode);
+	}
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
+	di->i_size = cpu_to_le64((u64)i_size_read(inode));
+	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+	di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
+	di->i_mtime_nsec = di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
+	ocfs2_update_inode_fsync_trans(handle, inode, 1);
+	ocfs2_journal_dirty(handle, wc->w_di_bh);
+
+	/* unlock pages before dealloc since it needs acquiring j_trans_barrier
+	 * lock, or it will cause a deadlock since journal commit threads holds
+	 * this lock and will ask for the page lock when flushing the data.
+	 * put it here to preserve the unlock order.
+	 */
+	ocfs2_unlock_pages(wc);
+
+	ocfs2_commit_trans(osb, handle);
+
+	ocfs2_run_deallocs(osb, &wc->w_dealloc);
+
+	brelse(wc->w_di_bh);
+	kfree(wc);
+
+	return copied;
+}
+
+static int ocfs2_write_end(struct file *file, struct address_space *mapping,
+			   loff_t pos, unsigned len, unsigned copied,
+			   struct page *page, void *fsdata)
+{
+	int ret;
+	struct inode *inode = mapping->host;
+
+	ret = ocfs2_write_end_nolock(mapping, pos, len, copied, page, fsdata);
+
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+	ocfs2_inode_unlock(inode, 1);
+
+	return ret;
+}
+
+const struct address_space_operations ocfs2_aops = {
+	.readpage		= ocfs2_readpage,
+	.readpages		= ocfs2_readpages,
+	.writepage		= ocfs2_writepage,
+	.write_begin		= ocfs2_write_begin,
+	.write_end		= ocfs2_write_end,
+	.bmap			= ocfs2_bmap,
+	.direct_IO		= ocfs2_direct_IO,
+	.invalidatepage		= block_invalidatepage,
+	.releasepage		= ocfs2_releasepage,
+	.migratepage		= buffer_migrate_page,
+	.is_partially_uptodate	= block_is_partially_uptodate,
+	.error_remove_page	= generic_error_remove_page,
+};
diff --git a/fs/ocfs2/aops.h b/fs/ocfs2/aops.h
new file mode 100644
index 000000000..dd59599b0
--- /dev/null
+++ b/fs/ocfs2/aops.h
@@ -0,0 +1,105 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2002, 2004, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_AOPS_H
+#define OCFS2_AOPS_H
+
+#include <linux/fs.h>
+
+handle_t *ocfs2_start_walk_page_trans(struct inode *inode,
+							 struct page *page,
+							 unsigned from,
+							 unsigned to);
+
+int ocfs2_map_page_blocks(struct page *page, u64 *p_blkno,
+			  struct inode *inode, unsigned int from,
+			  unsigned int to, int new);
+
+void ocfs2_unlock_and_free_pages(struct page **pages, int num_pages);
+
+int walk_page_buffers(	handle_t *handle,
+			struct buffer_head *head,
+			unsigned from,
+			unsigned to,
+			int *partial,
+			int (*fn)(	handle_t *handle,
+					struct buffer_head *bh));
+
+int ocfs2_write_end_nolock(struct address_space *mapping,
+			   loff_t pos, unsigned len, unsigned copied,
+			   struct page *page, void *fsdata);
+
+int ocfs2_write_begin_nolock(struct file *filp,
+			     struct address_space *mapping,
+			     loff_t pos, unsigned len, unsigned flags,
+			     struct page **pagep, void **fsdata,
+			     struct buffer_head *di_bh, struct page *mmap_page);
+
+int ocfs2_read_inline_data(struct inode *inode, struct page *page,
+			   struct buffer_head *di_bh);
+int ocfs2_size_fits_inline_data(struct buffer_head *di_bh, u64 new_size);
+
+int ocfs2_get_block(struct inode *inode, sector_t iblock,
+		    struct buffer_head *bh_result, int create);
+/* all ocfs2_dio_end_io()'s fault */
+#define ocfs2_iocb_is_rw_locked(iocb) \
+	test_bit(0, (unsigned long *)&iocb->private)
+static inline void ocfs2_iocb_set_rw_locked(struct kiocb *iocb, int level)
+{
+	set_bit(0, (unsigned long *)&iocb->private);
+	if (level)
+		set_bit(1, (unsigned long *)&iocb->private);
+	else
+		clear_bit(1, (unsigned long *)&iocb->private);
+}
+
+/*
+ * Using a named enum representing lock types in terms of #N bit stored in
+ * iocb->private, which is going to be used for communication between
+ * ocfs2_dio_end_io() and ocfs2_file_aio_write/read().
+ */
+enum ocfs2_iocb_lock_bits {
+	OCFS2_IOCB_RW_LOCK = 0,
+	OCFS2_IOCB_RW_LOCK_LEVEL,
+	OCFS2_IOCB_SEM,
+	OCFS2_IOCB_UNALIGNED_IO,
+	OCFS2_IOCB_NUM_LOCKS
+};
+
+#define ocfs2_iocb_clear_rw_locked(iocb) \
+	clear_bit(OCFS2_IOCB_RW_LOCK, (unsigned long *)&iocb->private)
+#define ocfs2_iocb_rw_locked_level(iocb) \
+	test_bit(OCFS2_IOCB_RW_LOCK_LEVEL, (unsigned long *)&iocb->private)
+#define ocfs2_iocb_set_sem_locked(iocb) \
+	set_bit(OCFS2_IOCB_SEM, (unsigned long *)&iocb->private)
+#define ocfs2_iocb_clear_sem_locked(iocb) \
+	clear_bit(OCFS2_IOCB_SEM, (unsigned long *)&iocb->private)
+#define ocfs2_iocb_is_sem_locked(iocb) \
+	test_bit(OCFS2_IOCB_SEM, (unsigned long *)&iocb->private)
+
+#define ocfs2_iocb_set_unaligned_aio(iocb) \
+	set_bit(OCFS2_IOCB_UNALIGNED_IO, (unsigned long *)&iocb->private)
+#define ocfs2_iocb_clear_unaligned_aio(iocb) \
+	clear_bit(OCFS2_IOCB_UNALIGNED_IO, (unsigned long *)&iocb->private)
+#define ocfs2_iocb_is_unaligned_aio(iocb) \
+	test_bit(OCFS2_IOCB_UNALIGNED_IO, (unsigned long *)&iocb->private)
+
+#endif /* OCFS2_FILE_H */
diff --git a/fs/ocfs2/blockcheck.c b/fs/ocfs2/blockcheck.c
new file mode 100644
index 000000000..0725e6054
--- /dev/null
+++ b/fs/ocfs2/blockcheck.c
@@ -0,0 +1,647 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * blockcheck.c
+ *
+ * Checksum and ECC codes for the OCFS2 userspace library.
+ *
+ * Copyright (C) 2006, 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License, version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include <linux/crc32.h>
+#include <linux/buffer_head.h>
+#include <linux/bitops.h>
+#include <linux/debugfs.h>
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <asm/byteorder.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "blockcheck.h"
+
+
+/*
+ * We use the following conventions:
+ *
+ * d = # data bits
+ * p = # parity bits
+ * c = # total code bits (d + p)
+ */
+
+
+/*
+ * Calculate the bit offset in the hamming code buffer based on the bit's
+ * offset in the data buffer.  Since the hamming code reserves all
+ * power-of-two bits for parity, the data bit number and the code bit
+ * number are offset by all the parity bits beforehand.
+ *
+ * Recall that bit numbers in hamming code are 1-based.  This function
+ * takes the 0-based data bit from the caller.
+ *
+ * An example.  Take bit 1 of the data buffer.  1 is a power of two (2^0),
+ * so it's a parity bit.  2 is a power of two (2^1), so it's a parity bit.
+ * 3 is not a power of two.  So bit 1 of the data buffer ends up as bit 3
+ * in the code buffer.
+ *
+ * The caller can pass in *p if it wants to keep track of the most recent
+ * number of parity bits added.  This allows the function to start the
+ * calculation at the last place.
+ */
+static unsigned int calc_code_bit(unsigned int i, unsigned int *p_cache)
+{
+	unsigned int b, p = 0;
+
+	/*
+	 * Data bits are 0-based, but we're talking code bits, which
+	 * are 1-based.
+	 */
+	b = i + 1;
+
+	/* Use the cache if it is there */
+	if (p_cache)
+		p = *p_cache;
+        b += p;
+
+	/*
+	 * For every power of two below our bit number, bump our bit.
+	 *
+	 * We compare with (b + 1) because we have to compare with what b
+	 * would be _if_ it were bumped up by the parity bit.  Capice?
+	 *
+	 * p is set above.
+	 */
+	for (; (1 << p) < (b + 1); p++)
+		b++;
+
+	if (p_cache)
+		*p_cache = p;
+
+	return b;
+}
+
+/*
+ * This is the low level encoder function.  It can be called across
+ * multiple hunks just like the crc32 code.  'd' is the number of bits
+ * _in_this_hunk_.  nr is the bit offset of this hunk.  So, if you had
+ * two 512B buffers, you would do it like so:
+ *
+ * parity = ocfs2_hamming_encode(0, buf1, 512 * 8, 0);
+ * parity = ocfs2_hamming_encode(parity, buf2, 512 * 8, 512 * 8);
+ *
+ * If you just have one buffer, use ocfs2_hamming_encode_block().
+ */
+u32 ocfs2_hamming_encode(u32 parity, void *data, unsigned int d, unsigned int nr)
+{
+	unsigned int i, b, p = 0;
+
+	BUG_ON(!d);
+
+	/*
+	 * b is the hamming code bit number.  Hamming code specifies a
+	 * 1-based array, but C uses 0-based.  So 'i' is for C, and 'b' is
+	 * for the algorithm.
+	 *
+	 * The i++ in the for loop is so that the start offset passed
+	 * to ocfs2_find_next_bit_set() is one greater than the previously
+	 * found bit.
+	 */
+	for (i = 0; (i = ocfs2_find_next_bit(data, d, i)) < d; i++)
+	{
+		/*
+		 * i is the offset in this hunk, nr + i is the total bit
+		 * offset.
+		 */
+		b = calc_code_bit(nr + i, &p);
+
+		/*
+		 * Data bits in the resultant code are checked by
+		 * parity bits that are part of the bit number
+		 * representation.  Huh?
+		 *
+		 * <wikipedia href="http://en.wikipedia.org/wiki/Hamming_code">
+		 * In other words, the parity bit at position 2^k
+		 * checks bits in positions having bit k set in
+		 * their binary representation.  Conversely, for
+		 * instance, bit 13, i.e. 1101(2), is checked by
+		 * bits 1000(2) = 8, 0100(2)=4 and 0001(2) = 1.
+		 * </wikipedia>
+		 *
+		 * Note that 'k' is the _code_ bit number.  'b' in
+		 * our loop.
+		 */
+		parity ^= b;
+	}
+
+	/* While the data buffer was treated as little endian, the
+	 * return value is in host endian. */
+	return parity;
+}
+
+u32 ocfs2_hamming_encode_block(void *data, unsigned int blocksize)
+{
+	return ocfs2_hamming_encode(0, data, blocksize * 8, 0);
+}
+
+/*
+ * Like ocfs2_hamming_encode(), this can handle hunks.  nr is the bit
+ * offset of the current hunk.  If bit to be fixed is not part of the
+ * current hunk, this does nothing.
+ *
+ * If you only have one hunk, use ocfs2_hamming_fix_block().
+ */
+void ocfs2_hamming_fix(void *data, unsigned int d, unsigned int nr,
+		       unsigned int fix)
+{
+	unsigned int i, b;
+
+	BUG_ON(!d);
+
+	/*
+	 * If the bit to fix has an hweight of 1, it's a parity bit.  One
+	 * busted parity bit is its own error.  Nothing to do here.
+	 */
+	if (hweight32(fix) == 1)
+		return;
+
+	/*
+	 * nr + d is the bit right past the data hunk we're looking at.
+	 * If fix after that, nothing to do
+	 */
+	if (fix >= calc_code_bit(nr + d, NULL))
+		return;
+
+	/*
+	 * nr is the offset in the data hunk we're starting at.  Let's
+	 * start b at the offset in the code buffer.  See hamming_encode()
+	 * for a more detailed description of 'b'.
+	 */
+	b = calc_code_bit(nr, NULL);
+	/* If the fix is before this hunk, nothing to do */
+	if (fix < b)
+		return;
+
+	for (i = 0; i < d; i++, b++)
+	{
+		/* Skip past parity bits */
+		while (hweight32(b) == 1)
+			b++;
+
+		/*
+		 * i is the offset in this data hunk.
+		 * nr + i is the offset in the total data buffer.
+		 * b is the offset in the total code buffer.
+		 *
+		 * Thus, when b == fix, bit i in the current hunk needs
+		 * fixing.
+		 */
+		if (b == fix)
+		{
+			if (ocfs2_test_bit(i, data))
+				ocfs2_clear_bit(i, data);
+			else
+				ocfs2_set_bit(i, data);
+			break;
+		}
+	}
+}
+
+void ocfs2_hamming_fix_block(void *data, unsigned int blocksize,
+			     unsigned int fix)
+{
+	ocfs2_hamming_fix(data, blocksize * 8, 0, fix);
+}
+
+
+/*
+ * Debugfs handling.
+ */
+
+#ifdef CONFIG_DEBUG_FS
+
+static int blockcheck_u64_get(void *data, u64 *val)
+{
+	*val = *(u64 *)data;
+	return 0;
+}
+DEFINE_SIMPLE_ATTRIBUTE(blockcheck_fops, blockcheck_u64_get, NULL, "%llu\n");
+
+static struct dentry *blockcheck_debugfs_create(const char *name,
+						struct dentry *parent,
+						u64 *value)
+{
+	return debugfs_create_file(name, S_IFREG | S_IRUSR, parent, value,
+				   &blockcheck_fops);
+}
+
+static void ocfs2_blockcheck_debug_remove(struct ocfs2_blockcheck_stats *stats)
+{
+	if (stats) {
+		debugfs_remove(stats->b_debug_check);
+		stats->b_debug_check = NULL;
+		debugfs_remove(stats->b_debug_failure);
+		stats->b_debug_failure = NULL;
+		debugfs_remove(stats->b_debug_recover);
+		stats->b_debug_recover = NULL;
+		debugfs_remove(stats->b_debug_dir);
+		stats->b_debug_dir = NULL;
+	}
+}
+
+static int ocfs2_blockcheck_debug_install(struct ocfs2_blockcheck_stats *stats,
+					  struct dentry *parent)
+{
+	int rc = -EINVAL;
+
+	if (!stats)
+		goto out;
+
+	stats->b_debug_dir = debugfs_create_dir("blockcheck", parent);
+	if (!stats->b_debug_dir)
+		goto out;
+
+	stats->b_debug_check =
+		blockcheck_debugfs_create("blocks_checked",
+					  stats->b_debug_dir,
+					  &stats->b_check_count);
+
+	stats->b_debug_failure =
+		blockcheck_debugfs_create("checksums_failed",
+					  stats->b_debug_dir,
+					  &stats->b_failure_count);
+
+	stats->b_debug_recover =
+		blockcheck_debugfs_create("ecc_recoveries",
+					  stats->b_debug_dir,
+					  &stats->b_recover_count);
+	if (stats->b_debug_check && stats->b_debug_failure &&
+	    stats->b_debug_recover)
+		rc = 0;
+
+out:
+	if (rc)
+		ocfs2_blockcheck_debug_remove(stats);
+	return rc;
+}
+#else
+static inline int ocfs2_blockcheck_debug_install(struct ocfs2_blockcheck_stats *stats,
+						 struct dentry *parent)
+{
+	return 0;
+}
+
+static inline void ocfs2_blockcheck_debug_remove(struct ocfs2_blockcheck_stats *stats)
+{
+}
+#endif  /* CONFIG_DEBUG_FS */
+
+/* Always-called wrappers for starting and stopping the debugfs files */
+int ocfs2_blockcheck_stats_debugfs_install(struct ocfs2_blockcheck_stats *stats,
+					   struct dentry *parent)
+{
+	return ocfs2_blockcheck_debug_install(stats, parent);
+}
+
+void ocfs2_blockcheck_stats_debugfs_remove(struct ocfs2_blockcheck_stats *stats)
+{
+	ocfs2_blockcheck_debug_remove(stats);
+}
+
+static void ocfs2_blockcheck_inc_check(struct ocfs2_blockcheck_stats *stats)
+{
+	u64 new_count;
+
+	if (!stats)
+		return;
+
+	spin_lock(&stats->b_lock);
+	stats->b_check_count++;
+	new_count = stats->b_check_count;
+	spin_unlock(&stats->b_lock);
+
+	if (!new_count)
+		mlog(ML_NOTICE, "Block check count has wrapped\n");
+}
+
+static void ocfs2_blockcheck_inc_failure(struct ocfs2_blockcheck_stats *stats)
+{
+	u64 new_count;
+
+	if (!stats)
+		return;
+
+	spin_lock(&stats->b_lock);
+	stats->b_failure_count++;
+	new_count = stats->b_failure_count;
+	spin_unlock(&stats->b_lock);
+
+	if (!new_count)
+		mlog(ML_NOTICE, "Checksum failure count has wrapped\n");
+}
+
+static void ocfs2_blockcheck_inc_recover(struct ocfs2_blockcheck_stats *stats)
+{
+	u64 new_count;
+
+	if (!stats)
+		return;
+
+	spin_lock(&stats->b_lock);
+	stats->b_recover_count++;
+	new_count = stats->b_recover_count;
+	spin_unlock(&stats->b_lock);
+
+	if (!new_count)
+		mlog(ML_NOTICE, "ECC recovery count has wrapped\n");
+}
+
+
+
+/*
+ * These are the low-level APIs for using the ocfs2_block_check structure.
+ */
+
+/*
+ * This function generates check information for a block.
+ * data is the block to be checked.  bc is a pointer to the
+ * ocfs2_block_check structure describing the crc32 and the ecc.
+ *
+ * bc should be a pointer inside data, as the function will
+ * take care of zeroing it before calculating the check information.  If
+ * bc does not point inside data, the caller must make sure any inline
+ * ocfs2_block_check structures are zeroed.
+ *
+ * The data buffer must be in on-disk endian (little endian for ocfs2).
+ * bc will be filled with little-endian values and will be ready to go to
+ * disk.
+ */
+void ocfs2_block_check_compute(void *data, size_t blocksize,
+			       struct ocfs2_block_check *bc)
+{
+	u32 crc;
+	u32 ecc;
+
+	memset(bc, 0, sizeof(struct ocfs2_block_check));
+
+	crc = crc32_le(~0, data, blocksize);
+	ecc = ocfs2_hamming_encode_block(data, blocksize);
+
+	/*
+	 * No ecc'd ocfs2 structure is larger than 4K, so ecc will be no
+	 * larger than 16 bits.
+	 */
+	BUG_ON(ecc > USHRT_MAX);
+
+	bc->bc_crc32e = cpu_to_le32(crc);
+	bc->bc_ecc = cpu_to_le16((u16)ecc);
+}
+
+/*
+ * This function validates existing check information.  Like _compute,
+ * the function will take care of zeroing bc before calculating check codes.
+ * If bc is not a pointer inside data, the caller must have zeroed any
+ * inline ocfs2_block_check structures.
+ *
+ * Again, the data passed in should be the on-disk endian.
+ */
+int ocfs2_block_check_validate(void *data, size_t blocksize,
+			       struct ocfs2_block_check *bc,
+			       struct ocfs2_blockcheck_stats *stats)
+{
+	int rc = 0;
+	u32 bc_crc32e;
+	u16 bc_ecc;
+	u32 crc, ecc;
+
+	ocfs2_blockcheck_inc_check(stats);
+
+	bc_crc32e = le32_to_cpu(bc->bc_crc32e);
+	bc_ecc = le16_to_cpu(bc->bc_ecc);
+
+	memset(bc, 0, sizeof(struct ocfs2_block_check));
+
+	/* Fast path - if the crc32 validates, we're good to go */
+	crc = crc32_le(~0, data, blocksize);
+	if (crc == bc_crc32e)
+		goto out;
+
+	ocfs2_blockcheck_inc_failure(stats);
+	mlog(ML_ERROR,
+	     "CRC32 failed: stored: 0x%x, computed 0x%x. Applying ECC.\n",
+	     (unsigned int)bc_crc32e, (unsigned int)crc);
+
+	/* Ok, try ECC fixups */
+	ecc = ocfs2_hamming_encode_block(data, blocksize);
+	ocfs2_hamming_fix_block(data, blocksize, ecc ^ bc_ecc);
+
+	/* And check the crc32 again */
+	crc = crc32_le(~0, data, blocksize);
+	if (crc == bc_crc32e) {
+		ocfs2_blockcheck_inc_recover(stats);
+		goto out;
+	}
+
+	mlog(ML_ERROR, "Fixed CRC32 failed: stored: 0x%x, computed 0x%x\n",
+	     (unsigned int)bc_crc32e, (unsigned int)crc);
+
+	rc = -EIO;
+
+out:
+	bc->bc_crc32e = cpu_to_le32(bc_crc32e);
+	bc->bc_ecc = cpu_to_le16(bc_ecc);
+
+	return rc;
+}
+
+/*
+ * This function generates check information for a list of buffer_heads.
+ * bhs is the blocks to be checked.  bc is a pointer to the
+ * ocfs2_block_check structure describing the crc32 and the ecc.
+ *
+ * bc should be a pointer inside data, as the function will
+ * take care of zeroing it before calculating the check information.  If
+ * bc does not point inside data, the caller must make sure any inline
+ * ocfs2_block_check structures are zeroed.
+ *
+ * The data buffer must be in on-disk endian (little endian for ocfs2).
+ * bc will be filled with little-endian values and will be ready to go to
+ * disk.
+ */
+void ocfs2_block_check_compute_bhs(struct buffer_head **bhs, int nr,
+				   struct ocfs2_block_check *bc)
+{
+	int i;
+	u32 crc, ecc;
+
+	BUG_ON(nr < 0);
+
+	if (!nr)
+		return;
+
+	memset(bc, 0, sizeof(struct ocfs2_block_check));
+
+	for (i = 0, crc = ~0, ecc = 0; i < nr; i++) {
+		crc = crc32_le(crc, bhs[i]->b_data, bhs[i]->b_size);
+		/*
+		 * The number of bits in a buffer is obviously b_size*8.
+		 * The offset of this buffer is b_size*i, so the bit offset
+		 * of this buffer is b_size*8*i.
+		 */
+		ecc = (u16)ocfs2_hamming_encode(ecc, bhs[i]->b_data,
+						bhs[i]->b_size * 8,
+						bhs[i]->b_size * 8 * i);
+	}
+
+	/*
+	 * No ecc'd ocfs2 structure is larger than 4K, so ecc will be no
+	 * larger than 16 bits.
+	 */
+	BUG_ON(ecc > USHRT_MAX);
+
+	bc->bc_crc32e = cpu_to_le32(crc);
+	bc->bc_ecc = cpu_to_le16((u16)ecc);
+}
+
+/*
+ * This function validates existing check information on a list of
+ * buffer_heads.  Like _compute_bhs, the function will take care of
+ * zeroing bc before calculating check codes.  If bc is not a pointer
+ * inside data, the caller must have zeroed any inline
+ * ocfs2_block_check structures.
+ *
+ * Again, the data passed in should be the on-disk endian.
+ */
+int ocfs2_block_check_validate_bhs(struct buffer_head **bhs, int nr,
+				   struct ocfs2_block_check *bc,
+				   struct ocfs2_blockcheck_stats *stats)
+{
+	int i, rc = 0;
+	u32 bc_crc32e;
+	u16 bc_ecc;
+	u32 crc, ecc, fix;
+
+	BUG_ON(nr < 0);
+
+	if (!nr)
+		return 0;
+
+	ocfs2_blockcheck_inc_check(stats);
+
+	bc_crc32e = le32_to_cpu(bc->bc_crc32e);
+	bc_ecc = le16_to_cpu(bc->bc_ecc);
+
+	memset(bc, 0, sizeof(struct ocfs2_block_check));
+
+	/* Fast path - if the crc32 validates, we're good to go */
+	for (i = 0, crc = ~0; i < nr; i++)
+		crc = crc32_le(crc, bhs[i]->b_data, bhs[i]->b_size);
+	if (crc == bc_crc32e)
+		goto out;
+
+	ocfs2_blockcheck_inc_failure(stats);
+	mlog(ML_ERROR,
+	     "CRC32 failed: stored: %u, computed %u.  Applying ECC.\n",
+	     (unsigned int)bc_crc32e, (unsigned int)crc);
+
+	/* Ok, try ECC fixups */
+	for (i = 0, ecc = 0; i < nr; i++) {
+		/*
+		 * The number of bits in a buffer is obviously b_size*8.
+		 * The offset of this buffer is b_size*i, so the bit offset
+		 * of this buffer is b_size*8*i.
+		 */
+		ecc = (u16)ocfs2_hamming_encode(ecc, bhs[i]->b_data,
+						bhs[i]->b_size * 8,
+						bhs[i]->b_size * 8 * i);
+	}
+	fix = ecc ^ bc_ecc;
+	for (i = 0; i < nr; i++) {
+		/*
+		 * Try the fix against each buffer.  It will only affect
+		 * one of them.
+		 */
+		ocfs2_hamming_fix(bhs[i]->b_data, bhs[i]->b_size * 8,
+				  bhs[i]->b_size * 8 * i, fix);
+	}
+
+	/* And check the crc32 again */
+	for (i = 0, crc = ~0; i < nr; i++)
+		crc = crc32_le(crc, bhs[i]->b_data, bhs[i]->b_size);
+	if (crc == bc_crc32e) {
+		ocfs2_blockcheck_inc_recover(stats);
+		goto out;
+	}
+
+	mlog(ML_ERROR, "Fixed CRC32 failed: stored: %u, computed %u\n",
+	     (unsigned int)bc_crc32e, (unsigned int)crc);
+
+	rc = -EIO;
+
+out:
+	bc->bc_crc32e = cpu_to_le32(bc_crc32e);
+	bc->bc_ecc = cpu_to_le16(bc_ecc);
+
+	return rc;
+}
+
+/*
+ * These are the main API.  They check the superblock flag before
+ * calling the underlying operations.
+ *
+ * They expect the buffer(s) to be in disk format.
+ */
+void ocfs2_compute_meta_ecc(struct super_block *sb, void *data,
+			    struct ocfs2_block_check *bc)
+{
+	if (ocfs2_meta_ecc(OCFS2_SB(sb)))
+		ocfs2_block_check_compute(data, sb->s_blocksize, bc);
+}
+
+int ocfs2_validate_meta_ecc(struct super_block *sb, void *data,
+			    struct ocfs2_block_check *bc)
+{
+	int rc = 0;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+
+	if (ocfs2_meta_ecc(osb))
+		rc = ocfs2_block_check_validate(data, sb->s_blocksize, bc,
+						&osb->osb_ecc_stats);
+
+	return rc;
+}
+
+void ocfs2_compute_meta_ecc_bhs(struct super_block *sb,
+				struct buffer_head **bhs, int nr,
+				struct ocfs2_block_check *bc)
+{
+	if (ocfs2_meta_ecc(OCFS2_SB(sb)))
+		ocfs2_block_check_compute_bhs(bhs, nr, bc);
+}
+
+int ocfs2_validate_meta_ecc_bhs(struct super_block *sb,
+				struct buffer_head **bhs, int nr,
+				struct ocfs2_block_check *bc)
+{
+	int rc = 0;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+
+	if (ocfs2_meta_ecc(osb))
+		rc = ocfs2_block_check_validate_bhs(bhs, nr, bc,
+						    &osb->osb_ecc_stats);
+
+	return rc;
+}
+
diff --git a/fs/ocfs2/blockcheck.h b/fs/ocfs2/blockcheck.h
new file mode 100644
index 000000000..d4b69febf
--- /dev/null
+++ b/fs/ocfs2/blockcheck.h
@@ -0,0 +1,107 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * blockcheck.h
+ *
+ * Checksum and ECC codes for the OCFS2 userspace library.
+ *
+ * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License, version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#ifndef OCFS2_BLOCKCHECK_H
+#define OCFS2_BLOCKCHECK_H
+
+
+/* Count errors and error correction from blockcheck.c */
+struct ocfs2_blockcheck_stats {
+	spinlock_t b_lock;
+	u64 b_check_count;	/* Number of blocks we've checked */
+	u64 b_failure_count;	/* Number of failed checksums */
+	u64 b_recover_count;	/* Number of blocks fixed by ecc */
+
+	/*
+	 * debugfs entries, used if this is passed to
+	 * ocfs2_blockcheck_stats_debugfs_install()
+	 */
+	struct dentry *b_debug_dir;	/* Parent of the debugfs  files */
+	struct dentry *b_debug_check;	/* Exposes b_check_count */
+	struct dentry *b_debug_failure;	/* Exposes b_failure_count */
+	struct dentry *b_debug_recover;	/* Exposes b_recover_count */
+};
+
+
+/* High level block API */
+void ocfs2_compute_meta_ecc(struct super_block *sb, void *data,
+			    struct ocfs2_block_check *bc);
+int ocfs2_validate_meta_ecc(struct super_block *sb, void *data,
+			    struct ocfs2_block_check *bc);
+void ocfs2_compute_meta_ecc_bhs(struct super_block *sb,
+				struct buffer_head **bhs, int nr,
+				struct ocfs2_block_check *bc);
+int ocfs2_validate_meta_ecc_bhs(struct super_block *sb,
+				struct buffer_head **bhs, int nr,
+				struct ocfs2_block_check *bc);
+
+/* Lower level API */
+void ocfs2_block_check_compute(void *data, size_t blocksize,
+			       struct ocfs2_block_check *bc);
+int ocfs2_block_check_validate(void *data, size_t blocksize,
+			       struct ocfs2_block_check *bc,
+			       struct ocfs2_blockcheck_stats *stats);
+void ocfs2_block_check_compute_bhs(struct buffer_head **bhs, int nr,
+				   struct ocfs2_block_check *bc);
+int ocfs2_block_check_validate_bhs(struct buffer_head **bhs, int nr,
+				   struct ocfs2_block_check *bc,
+				   struct ocfs2_blockcheck_stats *stats);
+
+/* Debug Initialization */
+int ocfs2_blockcheck_stats_debugfs_install(struct ocfs2_blockcheck_stats *stats,
+					   struct dentry *parent);
+void ocfs2_blockcheck_stats_debugfs_remove(struct ocfs2_blockcheck_stats *stats);
+
+/*
+ * Hamming code functions
+ */
+
+/*
+ * Encoding hamming code parity bits for a buffer.
+ *
+ * This is the low level encoder function.  It can be called across
+ * multiple hunks just like the crc32 code.  'd' is the number of bits
+ * _in_this_hunk_.  nr is the bit offset of this hunk.  So, if you had
+ * two 512B buffers, you would do it like so:
+ *
+ * parity = ocfs2_hamming_encode(0, buf1, 512 * 8, 0);
+ * parity = ocfs2_hamming_encode(parity, buf2, 512 * 8, 512 * 8);
+ *
+ * If you just have one buffer, use ocfs2_hamming_encode_block().
+ */
+u32 ocfs2_hamming_encode(u32 parity, void *data, unsigned int d,
+			 unsigned int nr);
+/*
+ * Fix a buffer with a bit error.  The 'fix' is the original parity
+ * xor'd with the parity calculated now.
+ *
+ * Like ocfs2_hamming_encode(), this can handle hunks.  nr is the bit
+ * offset of the current hunk.  If bit to be fixed is not part of the
+ * current hunk, this does nothing.
+ *
+ * If you only have one buffer, use ocfs2_hamming_fix_block().
+ */
+void ocfs2_hamming_fix(void *data, unsigned int d, unsigned int nr,
+		       unsigned int fix);
+
+/* Convenience wrappers for a single buffer of data */
+extern u32 ocfs2_hamming_encode_block(void *data, unsigned int blocksize);
+extern void ocfs2_hamming_fix_block(void *data, unsigned int blocksize,
+				    unsigned int fix);
+#endif
diff --git a/fs/ocfs2/buffer_head_io.c b/fs/ocfs2/buffer_head_io.c
new file mode 100644
index 000000000..1edcb141f
--- /dev/null
+++ b/fs/ocfs2/buffer_head_io.c
@@ -0,0 +1,427 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * io.c
+ *
+ * Buffer cache handling
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "inode.h"
+#include "journal.h"
+#include "uptodate.h"
+#include "buffer_head_io.h"
+#include "ocfs2_trace.h"
+
+/*
+ * Bits on bh->b_state used by ocfs2.
+ *
+ * These MUST be after the JBD2 bits.  Hence, we use BH_JBDPrivateStart.
+ */
+enum ocfs2_state_bits {
+	BH_NeedsValidate = BH_JBDPrivateStart,
+};
+
+/* Expand the magic b_state functions */
+BUFFER_FNS(NeedsValidate, needs_validate);
+
+int ocfs2_write_block(struct ocfs2_super *osb, struct buffer_head *bh,
+		      struct ocfs2_caching_info *ci)
+{
+	int ret = 0;
+
+	trace_ocfs2_write_block((unsigned long long)bh->b_blocknr, ci);
+
+	BUG_ON(bh->b_blocknr < OCFS2_SUPER_BLOCK_BLKNO);
+	BUG_ON(buffer_jbd(bh));
+
+	/* No need to check for a soft readonly file system here. non
+	 * journalled writes are only ever done on system files which
+	 * can get modified during recovery even if read-only. */
+	if (ocfs2_is_hard_readonly(osb)) {
+		ret = -EROFS;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_metadata_cache_io_lock(ci);
+
+	lock_buffer(bh);
+	set_buffer_uptodate(bh);
+
+	/* remove from dirty list before I/O. */
+	clear_buffer_dirty(bh);
+
+	get_bh(bh); /* for end_buffer_write_sync() */
+	bh->b_end_io = end_buffer_write_sync;
+	submit_bh(WRITE, bh);
+
+	wait_on_buffer(bh);
+
+	if (buffer_uptodate(bh)) {
+		ocfs2_set_buffer_uptodate(ci, bh);
+	} else {
+		/* We don't need to remove the clustered uptodate
+		 * information for this bh as it's not marked locally
+		 * uptodate. */
+		ret = -EIO;
+		mlog_errno(ret);
+	}
+
+	ocfs2_metadata_cache_io_unlock(ci);
+out:
+	return ret;
+}
+
+int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
+			   unsigned int nr, struct buffer_head *bhs[])
+{
+	int status = 0;
+	unsigned int i;
+	struct buffer_head *bh;
+
+	trace_ocfs2_read_blocks_sync((unsigned long long)block, nr);
+
+	if (!nr)
+		goto bail;
+
+	for (i = 0 ; i < nr ; i++) {
+		if (bhs[i] == NULL) {
+			bhs[i] = sb_getblk(osb->sb, block++);
+			if (bhs[i] == NULL) {
+				status = -ENOMEM;
+				mlog_errno(status);
+				goto bail;
+			}
+		}
+		bh = bhs[i];
+
+		if (buffer_jbd(bh)) {
+			trace_ocfs2_read_blocks_sync_jbd(
+					(unsigned long long)bh->b_blocknr);
+			continue;
+		}
+
+		if (buffer_dirty(bh)) {
+			/* This should probably be a BUG, or
+			 * at least return an error. */
+			mlog(ML_ERROR,
+			     "trying to sync read a dirty "
+			     "buffer! (blocknr = %llu), skipping\n",
+			     (unsigned long long)bh->b_blocknr);
+			continue;
+		}
+
+		lock_buffer(bh);
+		if (buffer_jbd(bh)) {
+			mlog(ML_ERROR,
+			     "block %llu had the JBD bit set "
+			     "while I was in lock_buffer!",
+			     (unsigned long long)bh->b_blocknr);
+			BUG();
+		}
+
+		clear_buffer_uptodate(bh);
+		get_bh(bh); /* for end_buffer_read_sync() */
+		bh->b_end_io = end_buffer_read_sync;
+		submit_bh(READ, bh);
+	}
+
+	for (i = nr; i > 0; i--) {
+		bh = bhs[i - 1];
+
+		/* No need to wait on the buffer if it's managed by JBD. */
+		if (!buffer_jbd(bh))
+			wait_on_buffer(bh);
+
+		if (!buffer_uptodate(bh)) {
+			/* Status won't be cleared from here on out,
+			 * so we can safely record this and loop back
+			 * to cleanup the other buffers. */
+			status = -EIO;
+			put_bh(bh);
+			bhs[i - 1] = NULL;
+		}
+	}
+
+bail:
+	return status;
+}
+
+int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
+		      struct buffer_head *bhs[], int flags,
+		      int (*validate)(struct super_block *sb,
+				      struct buffer_head *bh))
+{
+	int status = 0;
+	int i, ignore_cache = 0;
+	struct buffer_head *bh;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+
+	trace_ocfs2_read_blocks_begin(ci, (unsigned long long)block, nr, flags);
+
+	BUG_ON(!ci);
+	BUG_ON((flags & OCFS2_BH_READAHEAD) &&
+	       (flags & OCFS2_BH_IGNORE_CACHE));
+
+	if (bhs == NULL) {
+		status = -EINVAL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	if (nr < 0) {
+		mlog(ML_ERROR, "asked to read %d blocks!\n", nr);
+		status = -EINVAL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	if (nr == 0) {
+		status = 0;
+		goto bail;
+	}
+
+	ocfs2_metadata_cache_io_lock(ci);
+	for (i = 0 ; i < nr ; i++) {
+		if (bhs[i] == NULL) {
+			bhs[i] = sb_getblk(sb, block++);
+			if (bhs[i] == NULL) {
+				ocfs2_metadata_cache_io_unlock(ci);
+				status = -ENOMEM;
+				mlog_errno(status);
+				goto bail;
+			}
+		}
+		bh = bhs[i];
+		ignore_cache = (flags & OCFS2_BH_IGNORE_CACHE);
+
+		/* There are three read-ahead cases here which we need to
+		 * be concerned with. All three assume a buffer has
+		 * previously been submitted with OCFS2_BH_READAHEAD
+		 * and it hasn't yet completed I/O.
+		 *
+		 * 1) The current request is sync to disk. This rarely
+		 *    happens these days, and never when performance
+		 *    matters - the code can just wait on the buffer
+		 *    lock and re-submit.
+		 *
+		 * 2) The current request is cached, but not
+		 *    readahead. ocfs2_buffer_uptodate() will return
+		 *    false anyway, so we'll wind up waiting on the
+		 *    buffer lock to do I/O. We re-check the request
+		 *    with after getting the lock to avoid a re-submit.
+		 *
+		 * 3) The current request is readahead (and so must
+		 *    also be a caching one). We short circuit if the
+		 *    buffer is locked (under I/O) and if it's in the
+		 *    uptodate cache. The re-check from #2 catches the
+		 *    case that the previous read-ahead completes just
+		 *    before our is-it-in-flight check.
+		 */
+
+		if (!ignore_cache && !ocfs2_buffer_uptodate(ci, bh)) {
+			trace_ocfs2_read_blocks_from_disk(
+			     (unsigned long long)bh->b_blocknr,
+			     (unsigned long long)ocfs2_metadata_cache_owner(ci));
+			/* We're using ignore_cache here to say
+			 * "go to disk" */
+			ignore_cache = 1;
+		}
+
+		trace_ocfs2_read_blocks_bh((unsigned long long)bh->b_blocknr,
+			ignore_cache, buffer_jbd(bh), buffer_dirty(bh));
+
+		if (buffer_jbd(bh)) {
+			continue;
+		}
+
+		if (ignore_cache) {
+			if (buffer_dirty(bh)) {
+				/* This should probably be a BUG, or
+				 * at least return an error. */
+				continue;
+			}
+
+			/* A read-ahead request was made - if the
+			 * buffer is already under read-ahead from a
+			 * previously submitted request than we are
+			 * done here. */
+			if ((flags & OCFS2_BH_READAHEAD)
+			    && ocfs2_buffer_read_ahead(ci, bh))
+				continue;
+
+			lock_buffer(bh);
+			if (buffer_jbd(bh)) {
+#ifdef CATCH_BH_JBD_RACES
+				mlog(ML_ERROR, "block %llu had the JBD bit set "
+					       "while I was in lock_buffer!",
+				     (unsigned long long)bh->b_blocknr);
+				BUG();
+#else
+				unlock_buffer(bh);
+				continue;
+#endif
+			}
+
+			/* Re-check ocfs2_buffer_uptodate() as a
+			 * previously read-ahead buffer may have
+			 * completed I/O while we were waiting for the
+			 * buffer lock. */
+			if (!(flags & OCFS2_BH_IGNORE_CACHE)
+			    && !(flags & OCFS2_BH_READAHEAD)
+			    && ocfs2_buffer_uptodate(ci, bh)) {
+				unlock_buffer(bh);
+				continue;
+			}
+
+			clear_buffer_uptodate(bh);
+			get_bh(bh); /* for end_buffer_read_sync() */
+			if (validate)
+				set_buffer_needs_validate(bh);
+			bh->b_end_io = end_buffer_read_sync;
+			submit_bh(READ, bh);
+			continue;
+		}
+	}
+
+	status = 0;
+
+	for (i = (nr - 1); i >= 0; i--) {
+		bh = bhs[i];
+
+		if (!(flags & OCFS2_BH_READAHEAD)) {
+			/* We know this can't have changed as we hold the
+			 * owner sem. Avoid doing any work on the bh if the
+			 * journal has it. */
+			if (!buffer_jbd(bh))
+				wait_on_buffer(bh);
+
+			if (!buffer_uptodate(bh)) {
+				/* Status won't be cleared from here on out,
+				 * so we can safely record this and loop back
+				 * to cleanup the other buffers. Don't need to
+				 * remove the clustered uptodate information
+				 * for this bh as it's not marked locally
+				 * uptodate. */
+				status = -EIO;
+				put_bh(bh);
+				bhs[i] = NULL;
+				continue;
+			}
+
+			if (buffer_needs_validate(bh)) {
+				/* We never set NeedsValidate if the
+				 * buffer was held by the journal, so
+				 * that better not have changed */
+				BUG_ON(buffer_jbd(bh));
+				clear_buffer_needs_validate(bh);
+				status = validate(sb, bh);
+				if (status) {
+					put_bh(bh);
+					bhs[i] = NULL;
+					continue;
+				}
+			}
+		}
+
+		/* Always set the buffer in the cache, even if it was
+		 * a forced read, or read-ahead which hasn't yet
+		 * completed. */
+		ocfs2_set_buffer_uptodate(ci, bh);
+	}
+	ocfs2_metadata_cache_io_unlock(ci);
+
+	trace_ocfs2_read_blocks_end((unsigned long long)block, nr,
+				    flags, ignore_cache);
+
+bail:
+
+	return status;
+}
+
+/* Check whether the blkno is the super block or one of the backups. */
+static void ocfs2_check_super_or_backup(struct super_block *sb,
+					sector_t blkno)
+{
+	int i;
+	u64 backup_blkno;
+
+	if (blkno == OCFS2_SUPER_BLOCK_BLKNO)
+		return;
+
+	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
+		backup_blkno = ocfs2_backup_super_blkno(sb, i);
+		if (backup_blkno == blkno)
+			return;
+	}
+
+	BUG();
+}
+
+/*
+ * Write super block and backups doesn't need to collaborate with journal,
+ * so we don't need to lock ip_io_mutex and ci doesn't need to bea passed
+ * into this function.
+ */
+int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
+				struct buffer_head *bh)
+{
+	int ret = 0;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
+
+	BUG_ON(buffer_jbd(bh));
+	ocfs2_check_super_or_backup(osb->sb, bh->b_blocknr);
+
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb)) {
+		ret = -EROFS;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	lock_buffer(bh);
+	set_buffer_uptodate(bh);
+
+	/* remove from dirty list before I/O. */
+	clear_buffer_dirty(bh);
+
+	get_bh(bh); /* for end_buffer_write_sync() */
+	bh->b_end_io = end_buffer_write_sync;
+	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &di->i_check);
+	submit_bh(WRITE, bh);
+
+	wait_on_buffer(bh);
+
+	if (!buffer_uptodate(bh)) {
+		ret = -EIO;
+		mlog_errno(ret);
+	}
+
+out:
+	return ret;
+}
diff --git a/fs/ocfs2/buffer_head_io.h b/fs/ocfs2/buffer_head_io.h
new file mode 100644
index 000000000..b97bcc6dd
--- /dev/null
+++ b/fs/ocfs2/buffer_head_io.h
@@ -0,0 +1,77 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ocfs2_buffer_head.h
+ *
+ * Buffer cache handling functions defined
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_BUFFER_HEAD_IO_H
+#define OCFS2_BUFFER_HEAD_IO_H
+
+#include <linux/buffer_head.h>
+
+void ocfs2_end_buffer_io_sync(struct buffer_head *bh,
+			     int uptodate);
+
+int ocfs2_write_block(struct ocfs2_super          *osb,
+		      struct buffer_head  *bh,
+		      struct ocfs2_caching_info   *ci);
+int ocfs2_read_blocks_sync(struct ocfs2_super *osb, u64 block,
+			   unsigned int nr, struct buffer_head *bhs[]);
+
+/*
+ * If not NULL, validate() will be called on a buffer that is freshly
+ * read from disk.  It will not be called if the buffer was in cache.
+ * Note that if validate() is being used for this buffer, it needs to
+ * be set even for a READAHEAD call, as it marks the buffer for later
+ * validation.
+ */
+int ocfs2_read_blocks(struct ocfs2_caching_info *ci, u64 block, int nr,
+		      struct buffer_head *bhs[], int flags,
+		      int (*validate)(struct super_block *sb,
+				      struct buffer_head *bh));
+
+int ocfs2_write_super_or_backup(struct ocfs2_super *osb,
+				struct buffer_head *bh);
+
+#define OCFS2_BH_IGNORE_CACHE      1
+#define OCFS2_BH_READAHEAD         8
+
+static inline int ocfs2_read_block(struct ocfs2_caching_info *ci, u64 off,
+				   struct buffer_head **bh,
+				   int (*validate)(struct super_block *sb,
+						   struct buffer_head *bh))
+{
+	int status = 0;
+
+	if (bh == NULL) {
+		printk("ocfs2: bh == NULL\n");
+		status = -EINVAL;
+		goto bail;
+	}
+
+	status = ocfs2_read_blocks(ci, off, 1, bh, 0, validate);
+
+bail:
+	return status;
+}
+
+#endif /* OCFS2_BUFFER_HEAD_IO_H */
diff --git a/fs/ocfs2/cluster/Makefile b/fs/ocfs2/cluster/Makefile
new file mode 100644
index 000000000..1aefc0350
--- /dev/null
+++ b/fs/ocfs2/cluster/Makefile
@@ -0,0 +1,4 @@
+obj-$(CONFIG_OCFS2_FS) += ocfs2_nodemanager.o
+
+ocfs2_nodemanager-objs := heartbeat.o masklog.o sys.o nodemanager.o \
+	quorum.o tcp.o netdebug.o
diff --git a/fs/ocfs2/cluster/heartbeat.c b/fs/ocfs2/cluster/heartbeat.c
new file mode 100644
index 000000000..16eff4572
--- /dev/null
+++ b/fs/ocfs2/cluster/heartbeat.c
@@ -0,0 +1,2678 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/jiffies.h>
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/bio.h>
+#include <linux/blkdev.h>
+#include <linux/delay.h>
+#include <linux/file.h>
+#include <linux/kthread.h>
+#include <linux/configfs.h>
+#include <linux/random.h>
+#include <linux/crc32.h>
+#include <linux/time.h>
+#include <linux/debugfs.h>
+#include <linux/slab.h>
+#include <linux/bitmap.h>
+
+#include "heartbeat.h"
+#include "tcp.h"
+#include "nodemanager.h"
+#include "quorum.h"
+
+#include "masklog.h"
+
+
+/*
+ * The first heartbeat pass had one global thread that would serialize all hb
+ * callback calls.  This global serializing sem should only be removed once
+ * we've made sure that all callees can deal with being called concurrently
+ * from multiple hb region threads.
+ */
+static DECLARE_RWSEM(o2hb_callback_sem);
+
+/*
+ * multiple hb threads are watching multiple regions.  A node is live
+ * whenever any of the threads sees activity from the node in its region.
+ */
+static DEFINE_SPINLOCK(o2hb_live_lock);
+static struct list_head o2hb_live_slots[O2NM_MAX_NODES];
+static unsigned long o2hb_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+static LIST_HEAD(o2hb_node_events);
+static DECLARE_WAIT_QUEUE_HEAD(o2hb_steady_queue);
+
+/*
+ * In global heartbeat, we maintain a series of region bitmaps.
+ * 	- o2hb_region_bitmap allows us to limit the region number to max region.
+ * 	- o2hb_live_region_bitmap tracks live regions (seen steady iterations).
+ * 	- o2hb_quorum_region_bitmap tracks live regions that have seen all nodes
+ * 		heartbeat on it.
+ * 	- o2hb_failed_region_bitmap tracks the regions that have seen io timeouts.
+ */
+static unsigned long o2hb_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
+static unsigned long o2hb_live_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
+static unsigned long o2hb_quorum_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
+static unsigned long o2hb_failed_region_bitmap[BITS_TO_LONGS(O2NM_MAX_REGIONS)];
+
+#define O2HB_DB_TYPE_LIVENODES		0
+#define O2HB_DB_TYPE_LIVEREGIONS	1
+#define O2HB_DB_TYPE_QUORUMREGIONS	2
+#define O2HB_DB_TYPE_FAILEDREGIONS	3
+#define O2HB_DB_TYPE_REGION_LIVENODES	4
+#define O2HB_DB_TYPE_REGION_NUMBER	5
+#define O2HB_DB_TYPE_REGION_ELAPSED_TIME	6
+#define O2HB_DB_TYPE_REGION_PINNED	7
+struct o2hb_debug_buf {
+	int db_type;
+	int db_size;
+	int db_len;
+	void *db_data;
+};
+
+static struct o2hb_debug_buf *o2hb_db_livenodes;
+static struct o2hb_debug_buf *o2hb_db_liveregions;
+static struct o2hb_debug_buf *o2hb_db_quorumregions;
+static struct o2hb_debug_buf *o2hb_db_failedregions;
+
+#define O2HB_DEBUG_DIR			"o2hb"
+#define O2HB_DEBUG_LIVENODES		"livenodes"
+#define O2HB_DEBUG_LIVEREGIONS		"live_regions"
+#define O2HB_DEBUG_QUORUMREGIONS	"quorum_regions"
+#define O2HB_DEBUG_FAILEDREGIONS	"failed_regions"
+#define O2HB_DEBUG_REGION_NUMBER	"num"
+#define O2HB_DEBUG_REGION_ELAPSED_TIME	"elapsed_time_in_ms"
+#define O2HB_DEBUG_REGION_PINNED	"pinned"
+
+static struct dentry *o2hb_debug_dir;
+static struct dentry *o2hb_debug_livenodes;
+static struct dentry *o2hb_debug_liveregions;
+static struct dentry *o2hb_debug_quorumregions;
+static struct dentry *o2hb_debug_failedregions;
+
+static LIST_HEAD(o2hb_all_regions);
+
+static struct o2hb_callback {
+	struct list_head list;
+} o2hb_callbacks[O2HB_NUM_CB];
+
+static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type);
+
+#define O2HB_DEFAULT_BLOCK_BITS       9
+
+enum o2hb_heartbeat_modes {
+	O2HB_HEARTBEAT_LOCAL		= 0,
+	O2HB_HEARTBEAT_GLOBAL,
+	O2HB_HEARTBEAT_NUM_MODES,
+};
+
+char *o2hb_heartbeat_mode_desc[O2HB_HEARTBEAT_NUM_MODES] = {
+		"local",	/* O2HB_HEARTBEAT_LOCAL */
+		"global",	/* O2HB_HEARTBEAT_GLOBAL */
+};
+
+unsigned int o2hb_dead_threshold = O2HB_DEFAULT_DEAD_THRESHOLD;
+unsigned int o2hb_heartbeat_mode = O2HB_HEARTBEAT_LOCAL;
+
+/*
+ * o2hb_dependent_users tracks the number of registered callbacks that depend
+ * on heartbeat. o2net and o2dlm are two entities that register this callback.
+ * However only o2dlm depends on the heartbeat. It does not want the heartbeat
+ * to stop while a dlm domain is still active.
+ */
+unsigned int o2hb_dependent_users;
+
+/*
+ * In global heartbeat mode, all regions are pinned if there are one or more
+ * dependent users and the quorum region count is <= O2HB_PIN_CUT_OFF. All
+ * regions are unpinned if the region count exceeds the cut off or the number
+ * of dependent users falls to zero.
+ */
+#define O2HB_PIN_CUT_OFF		3
+
+/*
+ * In local heartbeat mode, we assume the dlm domain name to be the same as
+ * region uuid. This is true for domains created for the file system but not
+ * necessarily true for userdlm domains. This is a known limitation.
+ *
+ * In global heartbeat mode, we pin/unpin all o2hb regions. This solution
+ * works for both file system and userdlm domains.
+ */
+static int o2hb_region_pin(const char *region_uuid);
+static void o2hb_region_unpin(const char *region_uuid);
+
+/* Only sets a new threshold if there are no active regions.
+ *
+ * No locking or otherwise interesting code is required for reading
+ * o2hb_dead_threshold as it can't change once regions are active and
+ * it's not interesting to anyone until then anyway. */
+static void o2hb_dead_threshold_set(unsigned int threshold)
+{
+	if (threshold > O2HB_MIN_DEAD_THRESHOLD) {
+		spin_lock(&o2hb_live_lock);
+		if (list_empty(&o2hb_all_regions))
+			o2hb_dead_threshold = threshold;
+		spin_unlock(&o2hb_live_lock);
+	}
+}
+
+static int o2hb_global_heartbeat_mode_set(unsigned int hb_mode)
+{
+	int ret = -1;
+
+	if (hb_mode < O2HB_HEARTBEAT_NUM_MODES) {
+		spin_lock(&o2hb_live_lock);
+		if (list_empty(&o2hb_all_regions)) {
+			o2hb_heartbeat_mode = hb_mode;
+			ret = 0;
+		}
+		spin_unlock(&o2hb_live_lock);
+	}
+
+	return ret;
+}
+
+struct o2hb_node_event {
+	struct list_head        hn_item;
+	enum o2hb_callback_type hn_event_type;
+	struct o2nm_node        *hn_node;
+	int                     hn_node_num;
+};
+
+struct o2hb_disk_slot {
+	struct o2hb_disk_heartbeat_block *ds_raw_block;
+	u8			ds_node_num;
+	u64			ds_last_time;
+	u64			ds_last_generation;
+	u16			ds_equal_samples;
+	u16			ds_changed_samples;
+	struct list_head	ds_live_item;
+};
+
+/* each thread owns a region.. when we're asked to tear down the region
+ * we ask the thread to stop, who cleans up the region */
+struct o2hb_region {
+	struct config_item	hr_item;
+
+	struct list_head	hr_all_item;
+	unsigned		hr_unclean_stop:1,
+				hr_aborted_start:1,
+				hr_item_pinned:1,
+				hr_item_dropped:1;
+
+	/* protected by the hr_callback_sem */
+	struct task_struct 	*hr_task;
+
+	unsigned int		hr_blocks;
+	unsigned long long	hr_start_block;
+
+	unsigned int		hr_block_bits;
+	unsigned int		hr_block_bytes;
+
+	unsigned int		hr_slots_per_page;
+	unsigned int		hr_num_pages;
+
+	struct page             **hr_slot_data;
+	struct block_device	*hr_bdev;
+	struct o2hb_disk_slot	*hr_slots;
+
+	/* live node map of this region */
+	unsigned long		hr_live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned int		hr_region_num;
+
+	struct dentry		*hr_debug_dir;
+	struct dentry		*hr_debug_livenodes;
+	struct dentry		*hr_debug_regnum;
+	struct dentry		*hr_debug_elapsed_time;
+	struct dentry		*hr_debug_pinned;
+	struct o2hb_debug_buf	*hr_db_livenodes;
+	struct o2hb_debug_buf	*hr_db_regnum;
+	struct o2hb_debug_buf	*hr_db_elapsed_time;
+	struct o2hb_debug_buf	*hr_db_pinned;
+
+	/* let the person setting up hb wait for it to return until it
+	 * has reached a 'steady' state.  This will be fixed when we have
+	 * a more complete api that doesn't lead to this sort of fragility. */
+	atomic_t		hr_steady_iterations;
+
+	/* terminate o2hb thread if it does not reach steady state
+	 * (hr_steady_iterations == 0) within hr_unsteady_iterations */
+	atomic_t		hr_unsteady_iterations;
+
+	char			hr_dev_name[BDEVNAME_SIZE];
+
+	unsigned int		hr_timeout_ms;
+
+	/* randomized as the region goes up and down so that a node
+	 * recognizes a node going up and down in one iteration */
+	u64			hr_generation;
+
+	struct delayed_work	hr_write_timeout_work;
+	unsigned long		hr_last_timeout_start;
+
+	/* Used during o2hb_check_slot to hold a copy of the block
+	 * being checked because we temporarily have to zero out the
+	 * crc field. */
+	struct o2hb_disk_heartbeat_block *hr_tmp_block;
+};
+
+struct o2hb_bio_wait_ctxt {
+	atomic_t          wc_num_reqs;
+	struct completion wc_io_complete;
+	int               wc_error;
+};
+
+static void o2hb_write_timeout(struct work_struct *work)
+{
+	int failed, quorum;
+	unsigned long flags;
+	struct o2hb_region *reg =
+		container_of(work, struct o2hb_region,
+			     hr_write_timeout_work.work);
+
+	mlog(ML_ERROR, "Heartbeat write timeout to device %s after %u "
+	     "milliseconds\n", reg->hr_dev_name,
+	     jiffies_to_msecs(jiffies - reg->hr_last_timeout_start));
+
+	if (o2hb_global_heartbeat_active()) {
+		spin_lock_irqsave(&o2hb_live_lock, flags);
+		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
+			set_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
+		failed = bitmap_weight(o2hb_failed_region_bitmap,
+					O2NM_MAX_REGIONS);
+		quorum = bitmap_weight(o2hb_quorum_region_bitmap,
+					O2NM_MAX_REGIONS);
+		spin_unlock_irqrestore(&o2hb_live_lock, flags);
+
+		mlog(ML_HEARTBEAT, "Number of regions %d, failed regions %d\n",
+		     quorum, failed);
+
+		/*
+		 * Fence if the number of failed regions >= half the number
+		 * of  quorum regions
+		 */
+		if ((failed << 1) < quorum)
+			return;
+	}
+
+	o2quo_disk_timeout();
+}
+
+static void o2hb_arm_write_timeout(struct o2hb_region *reg)
+{
+	/* Arm writeout only after thread reaches steady state */
+	if (atomic_read(&reg->hr_steady_iterations) != 0)
+		return;
+
+	mlog(ML_HEARTBEAT, "Queue write timeout for %u ms\n",
+	     O2HB_MAX_WRITE_TIMEOUT_MS);
+
+	if (o2hb_global_heartbeat_active()) {
+		spin_lock(&o2hb_live_lock);
+		clear_bit(reg->hr_region_num, o2hb_failed_region_bitmap);
+		spin_unlock(&o2hb_live_lock);
+	}
+	cancel_delayed_work(&reg->hr_write_timeout_work);
+	reg->hr_last_timeout_start = jiffies;
+	schedule_delayed_work(&reg->hr_write_timeout_work,
+			      msecs_to_jiffies(O2HB_MAX_WRITE_TIMEOUT_MS));
+}
+
+static void o2hb_disarm_write_timeout(struct o2hb_region *reg)
+{
+	cancel_delayed_work_sync(&reg->hr_write_timeout_work);
+}
+
+static inline void o2hb_bio_wait_init(struct o2hb_bio_wait_ctxt *wc)
+{
+	atomic_set(&wc->wc_num_reqs, 1);
+	init_completion(&wc->wc_io_complete);
+	wc->wc_error = 0;
+}
+
+/* Used in error paths too */
+static inline void o2hb_bio_wait_dec(struct o2hb_bio_wait_ctxt *wc,
+				     unsigned int num)
+{
+	/* sadly atomic_sub_and_test() isn't available on all platforms.  The
+	 * good news is that the fast path only completes one at a time */
+	while(num--) {
+		if (atomic_dec_and_test(&wc->wc_num_reqs)) {
+			BUG_ON(num > 0);
+			complete(&wc->wc_io_complete);
+		}
+	}
+}
+
+static void o2hb_wait_on_io(struct o2hb_region *reg,
+			    struct o2hb_bio_wait_ctxt *wc)
+{
+	o2hb_bio_wait_dec(wc, 1);
+	wait_for_completion(&wc->wc_io_complete);
+}
+
+static void o2hb_bio_end_io(struct bio *bio,
+			   int error)
+{
+	struct o2hb_bio_wait_ctxt *wc = bio->bi_private;
+
+	if (error) {
+		mlog(ML_ERROR, "IO Error %d\n", error);
+		wc->wc_error = error;
+	}
+
+	o2hb_bio_wait_dec(wc, 1);
+	bio_put(bio);
+}
+
+/* Setup a Bio to cover I/O against num_slots slots starting at
+ * start_slot. */
+static struct bio *o2hb_setup_one_bio(struct o2hb_region *reg,
+				      struct o2hb_bio_wait_ctxt *wc,
+				      unsigned int *current_slot,
+				      unsigned int max_slots)
+{
+	int len, current_page;
+	unsigned int vec_len, vec_start;
+	unsigned int bits = reg->hr_block_bits;
+	unsigned int spp = reg->hr_slots_per_page;
+	unsigned int cs = *current_slot;
+	struct bio *bio;
+	struct page *page;
+
+	/* Testing has shown this allocation to take long enough under
+	 * GFP_KERNEL that the local node can get fenced. It would be
+	 * nicest if we could pre-allocate these bios and avoid this
+	 * all together. */
+	bio = bio_alloc(GFP_ATOMIC, 16);
+	if (!bio) {
+		mlog(ML_ERROR, "Could not alloc slots BIO!\n");
+		bio = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	/* Must put everything in 512 byte sectors for the bio... */
+	bio->bi_iter.bi_sector = (reg->hr_start_block + cs) << (bits - 9);
+	bio->bi_bdev = reg->hr_bdev;
+	bio->bi_private = wc;
+	bio->bi_end_io = o2hb_bio_end_io;
+
+	vec_start = (cs << bits) % PAGE_CACHE_SIZE;
+	while(cs < max_slots) {
+		current_page = cs / spp;
+		page = reg->hr_slot_data[current_page];
+
+		vec_len = min(PAGE_CACHE_SIZE - vec_start,
+			      (max_slots-cs) * (PAGE_CACHE_SIZE/spp) );
+
+		mlog(ML_HB_BIO, "page %d, vec_len = %u, vec_start = %u\n",
+		     current_page, vec_len, vec_start);
+
+		len = bio_add_page(bio, page, vec_len, vec_start);
+		if (len != vec_len) break;
+
+		cs += vec_len / (PAGE_CACHE_SIZE/spp);
+		vec_start = 0;
+	}
+
+bail:
+	*current_slot = cs;
+	return bio;
+}
+
+static int o2hb_read_slots(struct o2hb_region *reg,
+			   unsigned int max_slots)
+{
+	unsigned int current_slot=0;
+	int status;
+	struct o2hb_bio_wait_ctxt wc;
+	struct bio *bio;
+
+	o2hb_bio_wait_init(&wc);
+
+	while(current_slot < max_slots) {
+		bio = o2hb_setup_one_bio(reg, &wc, &current_slot, max_slots);
+		if (IS_ERR(bio)) {
+			status = PTR_ERR(bio);
+			mlog_errno(status);
+			goto bail_and_wait;
+		}
+
+		atomic_inc(&wc.wc_num_reqs);
+		submit_bio(READ, bio);
+	}
+
+	status = 0;
+
+bail_and_wait:
+	o2hb_wait_on_io(reg, &wc);
+	if (wc.wc_error && !status)
+		status = wc.wc_error;
+
+	return status;
+}
+
+static int o2hb_issue_node_write(struct o2hb_region *reg,
+				 struct o2hb_bio_wait_ctxt *write_wc)
+{
+	int status;
+	unsigned int slot;
+	struct bio *bio;
+
+	o2hb_bio_wait_init(write_wc);
+
+	slot = o2nm_this_node();
+
+	bio = o2hb_setup_one_bio(reg, write_wc, &slot, slot+1);
+	if (IS_ERR(bio)) {
+		status = PTR_ERR(bio);
+		mlog_errno(status);
+		goto bail;
+	}
+
+	atomic_inc(&write_wc->wc_num_reqs);
+	submit_bio(WRITE_SYNC, bio);
+
+	status = 0;
+bail:
+	return status;
+}
+
+static u32 o2hb_compute_block_crc_le(struct o2hb_region *reg,
+				     struct o2hb_disk_heartbeat_block *hb_block)
+{
+	__le32 old_cksum;
+	u32 ret;
+
+	/* We want to compute the block crc with a 0 value in the
+	 * hb_cksum field. Save it off here and replace after the
+	 * crc. */
+	old_cksum = hb_block->hb_cksum;
+	hb_block->hb_cksum = 0;
+
+	ret = crc32_le(0, (unsigned char *) hb_block, reg->hr_block_bytes);
+
+	hb_block->hb_cksum = old_cksum;
+
+	return ret;
+}
+
+static void o2hb_dump_slot(struct o2hb_disk_heartbeat_block *hb_block)
+{
+	mlog(ML_ERROR, "Dump slot information: seq = 0x%llx, node = %u, "
+	     "cksum = 0x%x, generation 0x%llx\n",
+	     (long long)le64_to_cpu(hb_block->hb_seq),
+	     hb_block->hb_node, le32_to_cpu(hb_block->hb_cksum),
+	     (long long)le64_to_cpu(hb_block->hb_generation));
+}
+
+static int o2hb_verify_crc(struct o2hb_region *reg,
+			   struct o2hb_disk_heartbeat_block *hb_block)
+{
+	u32 read, computed;
+
+	read = le32_to_cpu(hb_block->hb_cksum);
+	computed = o2hb_compute_block_crc_le(reg, hb_block);
+
+	return read == computed;
+}
+
+/*
+ * Compare the slot data with what we wrote in the last iteration.
+ * If the match fails, print an appropriate error message. This is to
+ * detect errors like... another node hearting on the same slot,
+ * flaky device that is losing writes, etc.
+ * Returns 1 if check succeeds, 0 otherwise.
+ */
+static int o2hb_check_own_slot(struct o2hb_region *reg)
+{
+	struct o2hb_disk_slot *slot;
+	struct o2hb_disk_heartbeat_block *hb_block;
+	char *errstr;
+
+	slot = &reg->hr_slots[o2nm_this_node()];
+	/* Don't check on our 1st timestamp */
+	if (!slot->ds_last_time)
+		return 0;
+
+	hb_block = slot->ds_raw_block;
+	if (le64_to_cpu(hb_block->hb_seq) == slot->ds_last_time &&
+	    le64_to_cpu(hb_block->hb_generation) == slot->ds_last_generation &&
+	    hb_block->hb_node == slot->ds_node_num)
+		return 1;
+
+#define ERRSTR1		"Another node is heartbeating on device"
+#define ERRSTR2		"Heartbeat generation mismatch on device"
+#define ERRSTR3		"Heartbeat sequence mismatch on device"
+
+	if (hb_block->hb_node != slot->ds_node_num)
+		errstr = ERRSTR1;
+	else if (le64_to_cpu(hb_block->hb_generation) !=
+		 slot->ds_last_generation)
+		errstr = ERRSTR2;
+	else
+		errstr = ERRSTR3;
+
+	mlog(ML_ERROR, "%s (%s): expected(%u:0x%llx, 0x%llx), "
+	     "ondisk(%u:0x%llx, 0x%llx)\n", errstr, reg->hr_dev_name,
+	     slot->ds_node_num, (unsigned long long)slot->ds_last_generation,
+	     (unsigned long long)slot->ds_last_time, hb_block->hb_node,
+	     (unsigned long long)le64_to_cpu(hb_block->hb_generation),
+	     (unsigned long long)le64_to_cpu(hb_block->hb_seq));
+
+	return 0;
+}
+
+static inline void o2hb_prepare_block(struct o2hb_region *reg,
+				      u64 generation)
+{
+	int node_num;
+	u64 cputime;
+	struct o2hb_disk_slot *slot;
+	struct o2hb_disk_heartbeat_block *hb_block;
+
+	node_num = o2nm_this_node();
+	slot = &reg->hr_slots[node_num];
+
+	hb_block = (struct o2hb_disk_heartbeat_block *)slot->ds_raw_block;
+	memset(hb_block, 0, reg->hr_block_bytes);
+	/* TODO: time stuff */
+	cputime = CURRENT_TIME.tv_sec;
+	if (!cputime)
+		cputime = 1;
+
+	hb_block->hb_seq = cpu_to_le64(cputime);
+	hb_block->hb_node = node_num;
+	hb_block->hb_generation = cpu_to_le64(generation);
+	hb_block->hb_dead_ms = cpu_to_le32(o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS);
+
+	/* This step must always happen last! */
+	hb_block->hb_cksum = cpu_to_le32(o2hb_compute_block_crc_le(reg,
+								   hb_block));
+
+	mlog(ML_HB_BIO, "our node generation = 0x%llx, cksum = 0x%x\n",
+	     (long long)generation,
+	     le32_to_cpu(hb_block->hb_cksum));
+}
+
+static void o2hb_fire_callbacks(struct o2hb_callback *hbcall,
+				struct o2nm_node *node,
+				int idx)
+{
+	struct o2hb_callback_func *f;
+
+	list_for_each_entry(f, &hbcall->list, hc_item) {
+		mlog(ML_HEARTBEAT, "calling funcs %p\n", f);
+		(f->hc_func)(node, idx, f->hc_data);
+	}
+}
+
+/* Will run the list in order until we process the passed event */
+static void o2hb_run_event_list(struct o2hb_node_event *queued_event)
+{
+	struct o2hb_callback *hbcall;
+	struct o2hb_node_event *event;
+
+	/* Holding callback sem assures we don't alter the callback
+	 * lists when doing this, and serializes ourselves with other
+	 * processes wanting callbacks. */
+	down_write(&o2hb_callback_sem);
+
+	spin_lock(&o2hb_live_lock);
+	while (!list_empty(&o2hb_node_events)
+	       && !list_empty(&queued_event->hn_item)) {
+		event = list_entry(o2hb_node_events.next,
+				   struct o2hb_node_event,
+				   hn_item);
+		list_del_init(&event->hn_item);
+		spin_unlock(&o2hb_live_lock);
+
+		mlog(ML_HEARTBEAT, "Node %s event for %d\n",
+		     event->hn_event_type == O2HB_NODE_UP_CB ? "UP" : "DOWN",
+		     event->hn_node_num);
+
+		hbcall = hbcall_from_type(event->hn_event_type);
+
+		/* We should *never* have gotten on to the list with a
+		 * bad type... This isn't something that we should try
+		 * to recover from. */
+		BUG_ON(IS_ERR(hbcall));
+
+		o2hb_fire_callbacks(hbcall, event->hn_node, event->hn_node_num);
+
+		spin_lock(&o2hb_live_lock);
+	}
+	spin_unlock(&o2hb_live_lock);
+
+	up_write(&o2hb_callback_sem);
+}
+
+static void o2hb_queue_node_event(struct o2hb_node_event *event,
+				  enum o2hb_callback_type type,
+				  struct o2nm_node *node,
+				  int node_num)
+{
+	assert_spin_locked(&o2hb_live_lock);
+
+	BUG_ON((!node) && (type != O2HB_NODE_DOWN_CB));
+
+	event->hn_event_type = type;
+	event->hn_node = node;
+	event->hn_node_num = node_num;
+
+	mlog(ML_HEARTBEAT, "Queue node %s event for node %d\n",
+	     type == O2HB_NODE_UP_CB ? "UP" : "DOWN", node_num);
+
+	list_add_tail(&event->hn_item, &o2hb_node_events);
+}
+
+static void o2hb_shutdown_slot(struct o2hb_disk_slot *slot)
+{
+	struct o2hb_node_event event =
+		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
+	struct o2nm_node *node;
+	int queued = 0;
+
+	node = o2nm_get_node_by_num(slot->ds_node_num);
+	if (!node)
+		return;
+
+	spin_lock(&o2hb_live_lock);
+	if (!list_empty(&slot->ds_live_item)) {
+		mlog(ML_HEARTBEAT, "Shutdown, node %d leaves region\n",
+		     slot->ds_node_num);
+
+		list_del_init(&slot->ds_live_item);
+
+		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
+			clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
+
+			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB, node,
+					      slot->ds_node_num);
+			queued = 1;
+		}
+	}
+	spin_unlock(&o2hb_live_lock);
+
+	if (queued)
+		o2hb_run_event_list(&event);
+
+	o2nm_node_put(node);
+}
+
+static void o2hb_set_quorum_device(struct o2hb_region *reg)
+{
+	if (!o2hb_global_heartbeat_active())
+		return;
+
+	/* Prevent race with o2hb_heartbeat_group_drop_item() */
+	if (kthread_should_stop())
+		return;
+
+	/* Tag region as quorum only after thread reaches steady state */
+	if (atomic_read(&reg->hr_steady_iterations) != 0)
+		return;
+
+	spin_lock(&o2hb_live_lock);
+
+	if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
+		goto unlock;
+
+	/*
+	 * A region can be added to the quorum only when it sees all
+	 * live nodes heartbeat on it. In other words, the region has been
+	 * added to all nodes.
+	 */
+	if (memcmp(reg->hr_live_node_bitmap, o2hb_live_node_bitmap,
+		   sizeof(o2hb_live_node_bitmap)))
+		goto unlock;
+
+	printk(KERN_NOTICE "o2hb: Region %s (%s) is now a quorum device\n",
+	       config_item_name(&reg->hr_item), reg->hr_dev_name);
+
+	set_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
+
+	/*
+	 * If global heartbeat active, unpin all regions if the
+	 * region count > CUT_OFF
+	 */
+	if (bitmap_weight(o2hb_quorum_region_bitmap,
+			   O2NM_MAX_REGIONS) > O2HB_PIN_CUT_OFF)
+		o2hb_region_unpin(NULL);
+unlock:
+	spin_unlock(&o2hb_live_lock);
+}
+
+static int o2hb_check_slot(struct o2hb_region *reg,
+			   struct o2hb_disk_slot *slot)
+{
+	int changed = 0, gen_changed = 0;
+	struct o2hb_node_event event =
+		{ .hn_item = LIST_HEAD_INIT(event.hn_item), };
+	struct o2nm_node *node;
+	struct o2hb_disk_heartbeat_block *hb_block = reg->hr_tmp_block;
+	u64 cputime;
+	unsigned int dead_ms = o2hb_dead_threshold * O2HB_REGION_TIMEOUT_MS;
+	unsigned int slot_dead_ms;
+	int tmp;
+	int queued = 0;
+
+	memcpy(hb_block, slot->ds_raw_block, reg->hr_block_bytes);
+
+	/*
+	 * If a node is no longer configured but is still in the livemap, we
+	 * may need to clear that bit from the livemap.
+	 */
+	node = o2nm_get_node_by_num(slot->ds_node_num);
+	if (!node) {
+		spin_lock(&o2hb_live_lock);
+		tmp = test_bit(slot->ds_node_num, o2hb_live_node_bitmap);
+		spin_unlock(&o2hb_live_lock);
+		if (!tmp)
+			return 0;
+	}
+
+	if (!o2hb_verify_crc(reg, hb_block)) {
+		/* all paths from here will drop o2hb_live_lock for
+		 * us. */
+		spin_lock(&o2hb_live_lock);
+
+		/* Don't print an error on the console in this case -
+		 * a freshly formatted heartbeat area will not have a
+		 * crc set on it. */
+		if (list_empty(&slot->ds_live_item))
+			goto out;
+
+		/* The node is live but pushed out a bad crc. We
+		 * consider it a transient miss but don't populate any
+		 * other values as they may be junk. */
+		mlog(ML_ERROR, "Node %d has written a bad crc to %s\n",
+		     slot->ds_node_num, reg->hr_dev_name);
+		o2hb_dump_slot(hb_block);
+
+		slot->ds_equal_samples++;
+		goto fire_callbacks;
+	}
+
+	/* we don't care if these wrap.. the state transitions below
+	 * clear at the right places */
+	cputime = le64_to_cpu(hb_block->hb_seq);
+	if (slot->ds_last_time != cputime)
+		slot->ds_changed_samples++;
+	else
+		slot->ds_equal_samples++;
+	slot->ds_last_time = cputime;
+
+	/* The node changed heartbeat generations. We assume this to
+	 * mean it dropped off but came back before we timed out. We
+	 * want to consider it down for the time being but don't want
+	 * to lose any changed_samples state we might build up to
+	 * considering it live again. */
+	if (slot->ds_last_generation != le64_to_cpu(hb_block->hb_generation)) {
+		gen_changed = 1;
+		slot->ds_equal_samples = 0;
+		mlog(ML_HEARTBEAT, "Node %d changed generation (0x%llx "
+		     "to 0x%llx)\n", slot->ds_node_num,
+		     (long long)slot->ds_last_generation,
+		     (long long)le64_to_cpu(hb_block->hb_generation));
+	}
+
+	slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
+
+	mlog(ML_HEARTBEAT, "Slot %d gen 0x%llx cksum 0x%x "
+	     "seq %llu last %llu changed %u equal %u\n",
+	     slot->ds_node_num, (long long)slot->ds_last_generation,
+	     le32_to_cpu(hb_block->hb_cksum),
+	     (unsigned long long)le64_to_cpu(hb_block->hb_seq),
+	     (unsigned long long)slot->ds_last_time, slot->ds_changed_samples,
+	     slot->ds_equal_samples);
+
+	spin_lock(&o2hb_live_lock);
+
+fire_callbacks:
+	/* dead nodes only come to life after some number of
+	 * changes at any time during their dead time */
+	if (list_empty(&slot->ds_live_item) &&
+	    slot->ds_changed_samples >= O2HB_LIVE_THRESHOLD) {
+		mlog(ML_HEARTBEAT, "Node %d (id 0x%llx) joined my region\n",
+		     slot->ds_node_num, (long long)slot->ds_last_generation);
+
+		set_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
+
+		/* first on the list generates a callback */
+		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
+			mlog(ML_HEARTBEAT, "o2hb: Add node %d to live nodes "
+			     "bitmap\n", slot->ds_node_num);
+			set_bit(slot->ds_node_num, o2hb_live_node_bitmap);
+
+			o2hb_queue_node_event(&event, O2HB_NODE_UP_CB, node,
+					      slot->ds_node_num);
+
+			changed = 1;
+			queued = 1;
+		}
+
+		list_add_tail(&slot->ds_live_item,
+			      &o2hb_live_slots[slot->ds_node_num]);
+
+		slot->ds_equal_samples = 0;
+
+		/* We want to be sure that all nodes agree on the
+		 * number of milliseconds before a node will be
+		 * considered dead. The self-fencing timeout is
+		 * computed from this value, and a discrepancy might
+		 * result in heartbeat calling a node dead when it
+		 * hasn't self-fenced yet. */
+		slot_dead_ms = le32_to_cpu(hb_block->hb_dead_ms);
+		if (slot_dead_ms && slot_dead_ms != dead_ms) {
+			/* TODO: Perhaps we can fail the region here. */
+			mlog(ML_ERROR, "Node %d on device %s has a dead count "
+			     "of %u ms, but our count is %u ms.\n"
+			     "Please double check your configuration values "
+			     "for 'O2CB_HEARTBEAT_THRESHOLD'\n",
+			     slot->ds_node_num, reg->hr_dev_name, slot_dead_ms,
+			     dead_ms);
+		}
+		goto out;
+	}
+
+	/* if the list is dead, we're done.. */
+	if (list_empty(&slot->ds_live_item))
+		goto out;
+
+	/* live nodes only go dead after enough consequtive missed
+	 * samples..  reset the missed counter whenever we see
+	 * activity */
+	if (slot->ds_equal_samples >= o2hb_dead_threshold || gen_changed) {
+		mlog(ML_HEARTBEAT, "Node %d left my region\n",
+		     slot->ds_node_num);
+
+		clear_bit(slot->ds_node_num, reg->hr_live_node_bitmap);
+
+		/* last off the live_slot generates a callback */
+		list_del_init(&slot->ds_live_item);
+		if (list_empty(&o2hb_live_slots[slot->ds_node_num])) {
+			mlog(ML_HEARTBEAT, "o2hb: Remove node %d from live "
+			     "nodes bitmap\n", slot->ds_node_num);
+			clear_bit(slot->ds_node_num, o2hb_live_node_bitmap);
+
+			/* node can be null */
+			o2hb_queue_node_event(&event, O2HB_NODE_DOWN_CB,
+					      node, slot->ds_node_num);
+
+			changed = 1;
+			queued = 1;
+		}
+
+		/* We don't clear this because the node is still
+		 * actually writing new blocks. */
+		if (!gen_changed)
+			slot->ds_changed_samples = 0;
+		goto out;
+	}
+	if (slot->ds_changed_samples) {
+		slot->ds_changed_samples = 0;
+		slot->ds_equal_samples = 0;
+	}
+out:
+	spin_unlock(&o2hb_live_lock);
+
+	if (queued)
+		o2hb_run_event_list(&event);
+
+	if (node)
+		o2nm_node_put(node);
+	return changed;
+}
+
+static int o2hb_highest_node(unsigned long *nodes, int numbits)
+{
+	return find_last_bit(nodes, numbits);
+}
+
+static int o2hb_do_disk_heartbeat(struct o2hb_region *reg)
+{
+	int i, ret, highest_node;
+	int membership_change = 0, own_slot_ok = 0;
+	unsigned long configured_nodes[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long live_node_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	struct o2hb_bio_wait_ctxt write_wc;
+
+	ret = o2nm_configured_node_map(configured_nodes,
+				       sizeof(configured_nodes));
+	if (ret) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	/*
+	 * If a node is not configured but is in the livemap, we still need
+	 * to read the slot so as to be able to remove it from the livemap.
+	 */
+	o2hb_fill_node_map(live_node_bitmap, sizeof(live_node_bitmap));
+	i = -1;
+	while ((i = find_next_bit(live_node_bitmap,
+				  O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
+		set_bit(i, configured_nodes);
+	}
+
+	highest_node = o2hb_highest_node(configured_nodes, O2NM_MAX_NODES);
+	if (highest_node >= O2NM_MAX_NODES) {
+		mlog(ML_NOTICE, "o2hb: No configured nodes found!\n");
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/* No sense in reading the slots of nodes that don't exist
+	 * yet. Of course, if the node definitions have holes in them
+	 * then we're reading an empty slot anyway... Consider this
+	 * best-effort. */
+	ret = o2hb_read_slots(reg, highest_node + 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	/* With an up to date view of the slots, we can check that no
+	 * other node has been improperly configured to heartbeat in
+	 * our slot. */
+	own_slot_ok = o2hb_check_own_slot(reg);
+
+	/* fill in the proper info for our next heartbeat */
+	o2hb_prepare_block(reg, reg->hr_generation);
+
+	ret = o2hb_issue_node_write(reg, &write_wc);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	i = -1;
+	while((i = find_next_bit(configured_nodes,
+				 O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES) {
+		membership_change |= o2hb_check_slot(reg, &reg->hr_slots[i]);
+	}
+
+	/*
+	 * We have to be sure we've advertised ourselves on disk
+	 * before we can go to steady state.  This ensures that
+	 * people we find in our steady state have seen us.
+	 */
+	o2hb_wait_on_io(reg, &write_wc);
+	if (write_wc.wc_error) {
+		/* Do not re-arm the write timeout on I/O error - we
+		 * can't be sure that the new block ever made it to
+		 * disk */
+		mlog(ML_ERROR, "Write error %d on device \"%s\"\n",
+		     write_wc.wc_error, reg->hr_dev_name);
+		ret = write_wc.wc_error;
+		goto bail;
+	}
+
+	/* Skip disarming the timeout if own slot has stale/bad data */
+	if (own_slot_ok) {
+		o2hb_set_quorum_device(reg);
+		o2hb_arm_write_timeout(reg);
+	}
+
+bail:
+	/* let the person who launched us know when things are steady */
+	if (atomic_read(&reg->hr_steady_iterations) != 0) {
+		if (!ret && own_slot_ok && !membership_change) {
+			if (atomic_dec_and_test(&reg->hr_steady_iterations))
+				wake_up(&o2hb_steady_queue);
+		}
+	}
+
+	if (atomic_read(&reg->hr_steady_iterations) != 0) {
+		if (atomic_dec_and_test(&reg->hr_unsteady_iterations)) {
+			printk(KERN_NOTICE "o2hb: Unable to stabilize "
+			       "heartbeart on region %s (%s)\n",
+			       config_item_name(&reg->hr_item),
+			       reg->hr_dev_name);
+			atomic_set(&reg->hr_steady_iterations, 0);
+			reg->hr_aborted_start = 1;
+			wake_up(&o2hb_steady_queue);
+			ret = -EIO;
+		}
+	}
+
+	return ret;
+}
+
+/* Subtract b from a, storing the result in a. a *must* have a larger
+ * value than b. */
+static void o2hb_tv_subtract(struct timeval *a,
+			     struct timeval *b)
+{
+	/* just return 0 when a is after b */
+	if (a->tv_sec < b->tv_sec ||
+	    (a->tv_sec == b->tv_sec && a->tv_usec < b->tv_usec)) {
+		a->tv_sec = 0;
+		a->tv_usec = 0;
+		return;
+	}
+
+	a->tv_sec -= b->tv_sec;
+	a->tv_usec -= b->tv_usec;
+	while ( a->tv_usec < 0 ) {
+		a->tv_sec--;
+		a->tv_usec += 1000000;
+	}
+}
+
+static unsigned int o2hb_elapsed_msecs(struct timeval *start,
+				       struct timeval *end)
+{
+	struct timeval res = *end;
+
+	o2hb_tv_subtract(&res, start);
+
+	return res.tv_sec * 1000 + res.tv_usec / 1000;
+}
+
+/*
+ * we ride the region ref that the region dir holds.  before the region
+ * dir is removed and drops it ref it will wait to tear down this
+ * thread.
+ */
+static int o2hb_thread(void *data)
+{
+	int i, ret;
+	struct o2hb_region *reg = data;
+	struct o2hb_bio_wait_ctxt write_wc;
+	struct timeval before_hb, after_hb;
+	unsigned int elapsed_msec;
+
+	mlog(ML_HEARTBEAT|ML_KTHREAD, "hb thread running\n");
+
+	set_user_nice(current, MIN_NICE);
+
+	/* Pin node */
+	o2nm_depend_this_node();
+
+	while (!kthread_should_stop() &&
+	       !reg->hr_unclean_stop && !reg->hr_aborted_start) {
+		/* We track the time spent inside
+		 * o2hb_do_disk_heartbeat so that we avoid more than
+		 * hr_timeout_ms between disk writes. On busy systems
+		 * this should result in a heartbeat which is less
+		 * likely to time itself out. */
+		do_gettimeofday(&before_hb);
+
+		ret = o2hb_do_disk_heartbeat(reg);
+
+		do_gettimeofday(&after_hb);
+		elapsed_msec = o2hb_elapsed_msecs(&before_hb, &after_hb);
+
+		mlog(ML_HEARTBEAT,
+		     "start = %lu.%lu, end = %lu.%lu, msec = %u, ret = %d\n",
+		     before_hb.tv_sec, (unsigned long) before_hb.tv_usec,
+		     after_hb.tv_sec, (unsigned long) after_hb.tv_usec,
+		     elapsed_msec, ret);
+
+		if (!kthread_should_stop() &&
+		    elapsed_msec < reg->hr_timeout_ms) {
+			/* the kthread api has blocked signals for us so no
+			 * need to record the return value. */
+			msleep_interruptible(reg->hr_timeout_ms - elapsed_msec);
+		}
+	}
+
+	o2hb_disarm_write_timeout(reg);
+
+	/* unclean stop is only used in very bad situation */
+	for(i = 0; !reg->hr_unclean_stop && i < reg->hr_blocks; i++)
+		o2hb_shutdown_slot(&reg->hr_slots[i]);
+
+	/* Explicit down notification - avoid forcing the other nodes
+	 * to timeout on this region when we could just as easily
+	 * write a clear generation - thus indicating to them that
+	 * this node has left this region.
+	 */
+	if (!reg->hr_unclean_stop && !reg->hr_aborted_start) {
+		o2hb_prepare_block(reg, 0);
+		ret = o2hb_issue_node_write(reg, &write_wc);
+		if (ret == 0)
+			o2hb_wait_on_io(reg, &write_wc);
+		else
+			mlog_errno(ret);
+	}
+
+	/* Unpin node */
+	o2nm_undepend_this_node();
+
+	mlog(ML_HEARTBEAT|ML_KTHREAD, "o2hb thread exiting\n");
+
+	return 0;
+}
+
+#ifdef CONFIG_DEBUG_FS
+static int o2hb_debug_open(struct inode *inode, struct file *file)
+{
+	struct o2hb_debug_buf *db = inode->i_private;
+	struct o2hb_region *reg;
+	unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long lts;
+	char *buf = NULL;
+	int i = -1;
+	int out = 0;
+
+	/* max_nodes should be the largest bitmap we pass here */
+	BUG_ON(sizeof(map) < db->db_size);
+
+	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!buf)
+		goto bail;
+
+	switch (db->db_type) {
+	case O2HB_DB_TYPE_LIVENODES:
+	case O2HB_DB_TYPE_LIVEREGIONS:
+	case O2HB_DB_TYPE_QUORUMREGIONS:
+	case O2HB_DB_TYPE_FAILEDREGIONS:
+		spin_lock(&o2hb_live_lock);
+		memcpy(map, db->db_data, db->db_size);
+		spin_unlock(&o2hb_live_lock);
+		break;
+
+	case O2HB_DB_TYPE_REGION_LIVENODES:
+		spin_lock(&o2hb_live_lock);
+		reg = (struct o2hb_region *)db->db_data;
+		memcpy(map, reg->hr_live_node_bitmap, db->db_size);
+		spin_unlock(&o2hb_live_lock);
+		break;
+
+	case O2HB_DB_TYPE_REGION_NUMBER:
+		reg = (struct o2hb_region *)db->db_data;
+		out += snprintf(buf + out, PAGE_SIZE - out, "%d\n",
+				reg->hr_region_num);
+		goto done;
+
+	case O2HB_DB_TYPE_REGION_ELAPSED_TIME:
+		reg = (struct o2hb_region *)db->db_data;
+		lts = reg->hr_last_timeout_start;
+		/* If 0, it has never been set before */
+		if (lts)
+			lts = jiffies_to_msecs(jiffies - lts);
+		out += snprintf(buf + out, PAGE_SIZE - out, "%lu\n", lts);
+		goto done;
+
+	case O2HB_DB_TYPE_REGION_PINNED:
+		reg = (struct o2hb_region *)db->db_data;
+		out += snprintf(buf + out, PAGE_SIZE - out, "%u\n",
+				!!reg->hr_item_pinned);
+		goto done;
+
+	default:
+		goto done;
+	}
+
+	while ((i = find_next_bit(map, db->db_len, i + 1)) < db->db_len)
+		out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
+	out += snprintf(buf + out, PAGE_SIZE - out, "\n");
+
+done:
+	i_size_write(inode, out);
+
+	file->private_data = buf;
+
+	return 0;
+bail:
+	return -ENOMEM;
+}
+
+static int o2hb_debug_release(struct inode *inode, struct file *file)
+{
+	kfree(file->private_data);
+	return 0;
+}
+
+static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
+				 size_t nbytes, loff_t *ppos)
+{
+	return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
+				       i_size_read(file->f_mapping->host));
+}
+#else
+static int o2hb_debug_open(struct inode *inode, struct file *file)
+{
+	return 0;
+}
+static int o2hb_debug_release(struct inode *inode, struct file *file)
+{
+	return 0;
+}
+static ssize_t o2hb_debug_read(struct file *file, char __user *buf,
+			       size_t nbytes, loff_t *ppos)
+{
+	return 0;
+}
+#endif  /* CONFIG_DEBUG_FS */
+
+static const struct file_operations o2hb_debug_fops = {
+	.open =		o2hb_debug_open,
+	.release =	o2hb_debug_release,
+	.read =		o2hb_debug_read,
+	.llseek =	generic_file_llseek,
+};
+
+void o2hb_exit(void)
+{
+	kfree(o2hb_db_livenodes);
+	kfree(o2hb_db_liveregions);
+	kfree(o2hb_db_quorumregions);
+	kfree(o2hb_db_failedregions);
+	debugfs_remove(o2hb_debug_failedregions);
+	debugfs_remove(o2hb_debug_quorumregions);
+	debugfs_remove(o2hb_debug_liveregions);
+	debugfs_remove(o2hb_debug_livenodes);
+	debugfs_remove(o2hb_debug_dir);
+}
+
+static struct dentry *o2hb_debug_create(const char *name, struct dentry *dir,
+					struct o2hb_debug_buf **db, int db_len,
+					int type, int size, int len, void *data)
+{
+	*db = kmalloc(db_len, GFP_KERNEL);
+	if (!*db)
+		return NULL;
+
+	(*db)->db_type = type;
+	(*db)->db_size = size;
+	(*db)->db_len = len;
+	(*db)->db_data = data;
+
+	return debugfs_create_file(name, S_IFREG|S_IRUSR, dir, *db,
+				   &o2hb_debug_fops);
+}
+
+static int o2hb_debug_init(void)
+{
+	int ret = -ENOMEM;
+
+	o2hb_debug_dir = debugfs_create_dir(O2HB_DEBUG_DIR, NULL);
+	if (!o2hb_debug_dir) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	o2hb_debug_livenodes = o2hb_debug_create(O2HB_DEBUG_LIVENODES,
+						 o2hb_debug_dir,
+						 &o2hb_db_livenodes,
+						 sizeof(*o2hb_db_livenodes),
+						 O2HB_DB_TYPE_LIVENODES,
+						 sizeof(o2hb_live_node_bitmap),
+						 O2NM_MAX_NODES,
+						 o2hb_live_node_bitmap);
+	if (!o2hb_debug_livenodes) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	o2hb_debug_liveregions = o2hb_debug_create(O2HB_DEBUG_LIVEREGIONS,
+						   o2hb_debug_dir,
+						   &o2hb_db_liveregions,
+						   sizeof(*o2hb_db_liveregions),
+						   O2HB_DB_TYPE_LIVEREGIONS,
+						   sizeof(o2hb_live_region_bitmap),
+						   O2NM_MAX_REGIONS,
+						   o2hb_live_region_bitmap);
+	if (!o2hb_debug_liveregions) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	o2hb_debug_quorumregions =
+			o2hb_debug_create(O2HB_DEBUG_QUORUMREGIONS,
+					  o2hb_debug_dir,
+					  &o2hb_db_quorumregions,
+					  sizeof(*o2hb_db_quorumregions),
+					  O2HB_DB_TYPE_QUORUMREGIONS,
+					  sizeof(o2hb_quorum_region_bitmap),
+					  O2NM_MAX_REGIONS,
+					  o2hb_quorum_region_bitmap);
+	if (!o2hb_debug_quorumregions) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	o2hb_debug_failedregions =
+			o2hb_debug_create(O2HB_DEBUG_FAILEDREGIONS,
+					  o2hb_debug_dir,
+					  &o2hb_db_failedregions,
+					  sizeof(*o2hb_db_failedregions),
+					  O2HB_DB_TYPE_FAILEDREGIONS,
+					  sizeof(o2hb_failed_region_bitmap),
+					  O2NM_MAX_REGIONS,
+					  o2hb_failed_region_bitmap);
+	if (!o2hb_debug_failedregions) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	ret = 0;
+bail:
+	if (ret)
+		o2hb_exit();
+
+	return ret;
+}
+
+int o2hb_init(void)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(o2hb_callbacks); i++)
+		INIT_LIST_HEAD(&o2hb_callbacks[i].list);
+
+	for (i = 0; i < ARRAY_SIZE(o2hb_live_slots); i++)
+		INIT_LIST_HEAD(&o2hb_live_slots[i]);
+
+	INIT_LIST_HEAD(&o2hb_node_events);
+
+	memset(o2hb_live_node_bitmap, 0, sizeof(o2hb_live_node_bitmap));
+	memset(o2hb_region_bitmap, 0, sizeof(o2hb_region_bitmap));
+	memset(o2hb_live_region_bitmap, 0, sizeof(o2hb_live_region_bitmap));
+	memset(o2hb_quorum_region_bitmap, 0, sizeof(o2hb_quorum_region_bitmap));
+	memset(o2hb_failed_region_bitmap, 0, sizeof(o2hb_failed_region_bitmap));
+
+	o2hb_dependent_users = 0;
+
+	return o2hb_debug_init();
+}
+
+/* if we're already in a callback then we're already serialized by the sem */
+static void o2hb_fill_node_map_from_callback(unsigned long *map,
+					     unsigned bytes)
+{
+	BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
+
+	memcpy(map, &o2hb_live_node_bitmap, bytes);
+}
+
+/*
+ * get a map of all nodes that are heartbeating in any regions
+ */
+void o2hb_fill_node_map(unsigned long *map, unsigned bytes)
+{
+	/* callers want to serialize this map and callbacks so that they
+	 * can trust that they don't miss nodes coming to the party */
+	down_read(&o2hb_callback_sem);
+	spin_lock(&o2hb_live_lock);
+	o2hb_fill_node_map_from_callback(map, bytes);
+	spin_unlock(&o2hb_live_lock);
+	up_read(&o2hb_callback_sem);
+}
+EXPORT_SYMBOL_GPL(o2hb_fill_node_map);
+
+/*
+ * heartbeat configfs bits.  The heartbeat set is a default set under
+ * the cluster set in nodemanager.c.
+ */
+
+static struct o2hb_region *to_o2hb_region(struct config_item *item)
+{
+	return item ? container_of(item, struct o2hb_region, hr_item) : NULL;
+}
+
+/* drop_item only drops its ref after killing the thread, nothing should
+ * be using the region anymore.  this has to clean up any state that
+ * attributes might have built up. */
+static void o2hb_region_release(struct config_item *item)
+{
+	int i;
+	struct page *page;
+	struct o2hb_region *reg = to_o2hb_region(item);
+
+	mlog(ML_HEARTBEAT, "hb region release (%s)\n", reg->hr_dev_name);
+
+	kfree(reg->hr_tmp_block);
+
+	if (reg->hr_slot_data) {
+		for (i = 0; i < reg->hr_num_pages; i++) {
+			page = reg->hr_slot_data[i];
+			if (page)
+				__free_page(page);
+		}
+		kfree(reg->hr_slot_data);
+	}
+
+	if (reg->hr_bdev)
+		blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
+
+	kfree(reg->hr_slots);
+
+	kfree(reg->hr_db_regnum);
+	kfree(reg->hr_db_livenodes);
+	debugfs_remove(reg->hr_debug_livenodes);
+	debugfs_remove(reg->hr_debug_regnum);
+	debugfs_remove(reg->hr_debug_elapsed_time);
+	debugfs_remove(reg->hr_debug_pinned);
+	debugfs_remove(reg->hr_debug_dir);
+
+	spin_lock(&o2hb_live_lock);
+	list_del(&reg->hr_all_item);
+	spin_unlock(&o2hb_live_lock);
+
+	kfree(reg);
+}
+
+static int o2hb_read_block_input(struct o2hb_region *reg,
+				 const char *page,
+				 size_t count,
+				 unsigned long *ret_bytes,
+				 unsigned int *ret_bits)
+{
+	unsigned long bytes;
+	char *p = (char *)page;
+
+	bytes = simple_strtoul(p, &p, 0);
+	if (!p || (*p && (*p != '\n')))
+		return -EINVAL;
+
+	/* Heartbeat and fs min / max block sizes are the same. */
+	if (bytes > 4096 || bytes < 512)
+		return -ERANGE;
+	if (hweight16(bytes) != 1)
+		return -EINVAL;
+
+	if (ret_bytes)
+		*ret_bytes = bytes;
+	if (ret_bits)
+		*ret_bits = ffs(bytes) - 1;
+
+	return 0;
+}
+
+static ssize_t o2hb_region_block_bytes_read(struct o2hb_region *reg,
+					    char *page)
+{
+	return sprintf(page, "%u\n", reg->hr_block_bytes);
+}
+
+static ssize_t o2hb_region_block_bytes_write(struct o2hb_region *reg,
+					     const char *page,
+					     size_t count)
+{
+	int status;
+	unsigned long block_bytes;
+	unsigned int block_bits;
+
+	if (reg->hr_bdev)
+		return -EINVAL;
+
+	status = o2hb_read_block_input(reg, page, count,
+				       &block_bytes, &block_bits);
+	if (status)
+		return status;
+
+	reg->hr_block_bytes = (unsigned int)block_bytes;
+	reg->hr_block_bits = block_bits;
+
+	return count;
+}
+
+static ssize_t o2hb_region_start_block_read(struct o2hb_region *reg,
+					    char *page)
+{
+	return sprintf(page, "%llu\n", reg->hr_start_block);
+}
+
+static ssize_t o2hb_region_start_block_write(struct o2hb_region *reg,
+					     const char *page,
+					     size_t count)
+{
+	unsigned long long tmp;
+	char *p = (char *)page;
+
+	if (reg->hr_bdev)
+		return -EINVAL;
+
+	tmp = simple_strtoull(p, &p, 0);
+	if (!p || (*p && (*p != '\n')))
+		return -EINVAL;
+
+	reg->hr_start_block = tmp;
+
+	return count;
+}
+
+static ssize_t o2hb_region_blocks_read(struct o2hb_region *reg,
+				       char *page)
+{
+	return sprintf(page, "%d\n", reg->hr_blocks);
+}
+
+static ssize_t o2hb_region_blocks_write(struct o2hb_region *reg,
+					const char *page,
+					size_t count)
+{
+	unsigned long tmp;
+	char *p = (char *)page;
+
+	if (reg->hr_bdev)
+		return -EINVAL;
+
+	tmp = simple_strtoul(p, &p, 0);
+	if (!p || (*p && (*p != '\n')))
+		return -EINVAL;
+
+	if (tmp > O2NM_MAX_NODES || tmp == 0)
+		return -ERANGE;
+
+	reg->hr_blocks = (unsigned int)tmp;
+
+	return count;
+}
+
+static ssize_t o2hb_region_dev_read(struct o2hb_region *reg,
+				    char *page)
+{
+	unsigned int ret = 0;
+
+	if (reg->hr_bdev)
+		ret = sprintf(page, "%s\n", reg->hr_dev_name);
+
+	return ret;
+}
+
+static void o2hb_init_region_params(struct o2hb_region *reg)
+{
+	reg->hr_slots_per_page = PAGE_CACHE_SIZE >> reg->hr_block_bits;
+	reg->hr_timeout_ms = O2HB_REGION_TIMEOUT_MS;
+
+	mlog(ML_HEARTBEAT, "hr_start_block = %llu, hr_blocks = %u\n",
+	     reg->hr_start_block, reg->hr_blocks);
+	mlog(ML_HEARTBEAT, "hr_block_bytes = %u, hr_block_bits = %u\n",
+	     reg->hr_block_bytes, reg->hr_block_bits);
+	mlog(ML_HEARTBEAT, "hr_timeout_ms = %u\n", reg->hr_timeout_ms);
+	mlog(ML_HEARTBEAT, "dead threshold = %u\n", o2hb_dead_threshold);
+}
+
+static int o2hb_map_slot_data(struct o2hb_region *reg)
+{
+	int i, j;
+	unsigned int last_slot;
+	unsigned int spp = reg->hr_slots_per_page;
+	struct page *page;
+	char *raw;
+	struct o2hb_disk_slot *slot;
+
+	reg->hr_tmp_block = kmalloc(reg->hr_block_bytes, GFP_KERNEL);
+	if (reg->hr_tmp_block == NULL) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
+	reg->hr_slots = kcalloc(reg->hr_blocks,
+				sizeof(struct o2hb_disk_slot), GFP_KERNEL);
+	if (reg->hr_slots == NULL) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
+	for(i = 0; i < reg->hr_blocks; i++) {
+		slot = &reg->hr_slots[i];
+		slot->ds_node_num = i;
+		INIT_LIST_HEAD(&slot->ds_live_item);
+		slot->ds_raw_block = NULL;
+	}
+
+	reg->hr_num_pages = (reg->hr_blocks + spp - 1) / spp;
+	mlog(ML_HEARTBEAT, "Going to require %u pages to cover %u blocks "
+			   "at %u blocks per page\n",
+	     reg->hr_num_pages, reg->hr_blocks, spp);
+
+	reg->hr_slot_data = kcalloc(reg->hr_num_pages, sizeof(struct page *),
+				    GFP_KERNEL);
+	if (!reg->hr_slot_data) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
+	for(i = 0; i < reg->hr_num_pages; i++) {
+		page = alloc_page(GFP_KERNEL);
+		if (!page) {
+			mlog_errno(-ENOMEM);
+			return -ENOMEM;
+		}
+
+		reg->hr_slot_data[i] = page;
+
+		last_slot = i * spp;
+		raw = page_address(page);
+		for (j = 0;
+		     (j < spp) && ((j + last_slot) < reg->hr_blocks);
+		     j++) {
+			BUG_ON((j + last_slot) >= reg->hr_blocks);
+
+			slot = &reg->hr_slots[j + last_slot];
+			slot->ds_raw_block =
+				(struct o2hb_disk_heartbeat_block *) raw;
+
+			raw += reg->hr_block_bytes;
+		}
+	}
+
+	return 0;
+}
+
+/* Read in all the slots available and populate the tracking
+ * structures so that we can start with a baseline idea of what's
+ * there. */
+static int o2hb_populate_slot_data(struct o2hb_region *reg)
+{
+	int ret, i;
+	struct o2hb_disk_slot *slot;
+	struct o2hb_disk_heartbeat_block *hb_block;
+
+	ret = o2hb_read_slots(reg, reg->hr_blocks);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* We only want to get an idea of the values initially in each
+	 * slot, so we do no verification - o2hb_check_slot will
+	 * actually determine if each configured slot is valid and
+	 * whether any values have changed. */
+	for(i = 0; i < reg->hr_blocks; i++) {
+		slot = &reg->hr_slots[i];
+		hb_block = (struct o2hb_disk_heartbeat_block *) slot->ds_raw_block;
+
+		/* Only fill the values that o2hb_check_slot uses to
+		 * determine changing slots */
+		slot->ds_last_time = le64_to_cpu(hb_block->hb_seq);
+		slot->ds_last_generation = le64_to_cpu(hb_block->hb_generation);
+	}
+
+out:
+	return ret;
+}
+
+/* this is acting as commit; we set up all of hr_bdev and hr_task or nothing */
+static ssize_t o2hb_region_dev_write(struct o2hb_region *reg,
+				     const char *page,
+				     size_t count)
+{
+	struct task_struct *hb_task;
+	long fd;
+	int sectsize;
+	char *p = (char *)page;
+	struct fd f;
+	struct inode *inode;
+	ssize_t ret = -EINVAL;
+	int live_threshold;
+
+	if (reg->hr_bdev)
+		goto out;
+
+	/* We can't heartbeat without having had our node number
+	 * configured yet. */
+	if (o2nm_this_node() == O2NM_MAX_NODES)
+		goto out;
+
+	fd = simple_strtol(p, &p, 0);
+	if (!p || (*p && (*p != '\n')))
+		goto out;
+
+	if (fd < 0 || fd >= INT_MAX)
+		goto out;
+
+	f = fdget(fd);
+	if (f.file == NULL)
+		goto out;
+
+	if (reg->hr_blocks == 0 || reg->hr_start_block == 0 ||
+	    reg->hr_block_bytes == 0)
+		goto out2;
+
+	inode = igrab(f.file->f_mapping->host);
+	if (inode == NULL)
+		goto out2;
+
+	if (!S_ISBLK(inode->i_mode))
+		goto out3;
+
+	reg->hr_bdev = I_BDEV(f.file->f_mapping->host);
+	ret = blkdev_get(reg->hr_bdev, FMODE_WRITE | FMODE_READ, NULL);
+	if (ret) {
+		reg->hr_bdev = NULL;
+		goto out3;
+	}
+	inode = NULL;
+
+	bdevname(reg->hr_bdev, reg->hr_dev_name);
+
+	sectsize = bdev_logical_block_size(reg->hr_bdev);
+	if (sectsize != reg->hr_block_bytes) {
+		mlog(ML_ERROR,
+		     "blocksize %u incorrect for device, expected %d",
+		     reg->hr_block_bytes, sectsize);
+		ret = -EINVAL;
+		goto out3;
+	}
+
+	o2hb_init_region_params(reg);
+
+	/* Generation of zero is invalid */
+	do {
+		get_random_bytes(&reg->hr_generation,
+				 sizeof(reg->hr_generation));
+	} while (reg->hr_generation == 0);
+
+	ret = o2hb_map_slot_data(reg);
+	if (ret) {
+		mlog_errno(ret);
+		goto out3;
+	}
+
+	ret = o2hb_populate_slot_data(reg);
+	if (ret) {
+		mlog_errno(ret);
+		goto out3;
+	}
+
+	INIT_DELAYED_WORK(&reg->hr_write_timeout_work, o2hb_write_timeout);
+
+	/*
+	 * A node is considered live after it has beat LIVE_THRESHOLD
+	 * times.  We're not steady until we've given them a chance
+	 * _after_ our first read.
+	 * The default threshold is bare minimum so as to limit the delay
+	 * during mounts. For global heartbeat, the threshold doubled for the
+	 * first region.
+	 */
+	live_threshold = O2HB_LIVE_THRESHOLD;
+	if (o2hb_global_heartbeat_active()) {
+		spin_lock(&o2hb_live_lock);
+		if (bitmap_weight(o2hb_region_bitmap, O2NM_MAX_REGIONS) == 1)
+			live_threshold <<= 1;
+		spin_unlock(&o2hb_live_lock);
+	}
+	++live_threshold;
+	atomic_set(&reg->hr_steady_iterations, live_threshold);
+	/* unsteady_iterations is double the steady_iterations */
+	atomic_set(&reg->hr_unsteady_iterations, (live_threshold << 1));
+
+	hb_task = kthread_run(o2hb_thread, reg, "o2hb-%s",
+			      reg->hr_item.ci_name);
+	if (IS_ERR(hb_task)) {
+		ret = PTR_ERR(hb_task);
+		mlog_errno(ret);
+		goto out3;
+	}
+
+	spin_lock(&o2hb_live_lock);
+	reg->hr_task = hb_task;
+	spin_unlock(&o2hb_live_lock);
+
+	ret = wait_event_interruptible(o2hb_steady_queue,
+				atomic_read(&reg->hr_steady_iterations) == 0);
+	if (ret) {
+		atomic_set(&reg->hr_steady_iterations, 0);
+		reg->hr_aborted_start = 1;
+	}
+
+	if (reg->hr_aborted_start) {
+		ret = -EIO;
+		goto out3;
+	}
+
+	/* Ok, we were woken.  Make sure it wasn't by drop_item() */
+	spin_lock(&o2hb_live_lock);
+	hb_task = reg->hr_task;
+	if (o2hb_global_heartbeat_active())
+		set_bit(reg->hr_region_num, o2hb_live_region_bitmap);
+	spin_unlock(&o2hb_live_lock);
+
+	if (hb_task)
+		ret = count;
+	else
+		ret = -EIO;
+
+	if (hb_task && o2hb_global_heartbeat_active())
+		printk(KERN_NOTICE "o2hb: Heartbeat started on region %s (%s)\n",
+		       config_item_name(&reg->hr_item), reg->hr_dev_name);
+
+out3:
+	iput(inode);
+out2:
+	fdput(f);
+out:
+	if (ret < 0) {
+		if (reg->hr_bdev) {
+			blkdev_put(reg->hr_bdev, FMODE_READ|FMODE_WRITE);
+			reg->hr_bdev = NULL;
+		}
+	}
+	return ret;
+}
+
+static ssize_t o2hb_region_pid_read(struct o2hb_region *reg,
+                                      char *page)
+{
+	pid_t pid = 0;
+
+	spin_lock(&o2hb_live_lock);
+	if (reg->hr_task)
+		pid = task_pid_nr(reg->hr_task);
+	spin_unlock(&o2hb_live_lock);
+
+	if (!pid)
+		return 0;
+
+	return sprintf(page, "%u\n", pid);
+}
+
+struct o2hb_region_attribute {
+	struct configfs_attribute attr;
+	ssize_t (*show)(struct o2hb_region *, char *);
+	ssize_t (*store)(struct o2hb_region *, const char *, size_t);
+};
+
+static struct o2hb_region_attribute o2hb_region_attr_block_bytes = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "block_bytes",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2hb_region_block_bytes_read,
+	.store	= o2hb_region_block_bytes_write,
+};
+
+static struct o2hb_region_attribute o2hb_region_attr_start_block = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "start_block",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2hb_region_start_block_read,
+	.store	= o2hb_region_start_block_write,
+};
+
+static struct o2hb_region_attribute o2hb_region_attr_blocks = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "blocks",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2hb_region_blocks_read,
+	.store	= o2hb_region_blocks_write,
+};
+
+static struct o2hb_region_attribute o2hb_region_attr_dev = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "dev",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2hb_region_dev_read,
+	.store	= o2hb_region_dev_write,
+};
+
+static struct o2hb_region_attribute o2hb_region_attr_pid = {
+       .attr   = { .ca_owner = THIS_MODULE,
+                   .ca_name = "pid",
+                   .ca_mode = S_IRUGO | S_IRUSR },
+       .show   = o2hb_region_pid_read,
+};
+
+static struct configfs_attribute *o2hb_region_attrs[] = {
+	&o2hb_region_attr_block_bytes.attr,
+	&o2hb_region_attr_start_block.attr,
+	&o2hb_region_attr_blocks.attr,
+	&o2hb_region_attr_dev.attr,
+	&o2hb_region_attr_pid.attr,
+	NULL,
+};
+
+static ssize_t o2hb_region_show(struct config_item *item,
+				struct configfs_attribute *attr,
+				char *page)
+{
+	struct o2hb_region *reg = to_o2hb_region(item);
+	struct o2hb_region_attribute *o2hb_region_attr =
+		container_of(attr, struct o2hb_region_attribute, attr);
+	ssize_t ret = 0;
+
+	if (o2hb_region_attr->show)
+		ret = o2hb_region_attr->show(reg, page);
+	return ret;
+}
+
+static ssize_t o2hb_region_store(struct config_item *item,
+				 struct configfs_attribute *attr,
+				 const char *page, size_t count)
+{
+	struct o2hb_region *reg = to_o2hb_region(item);
+	struct o2hb_region_attribute *o2hb_region_attr =
+		container_of(attr, struct o2hb_region_attribute, attr);
+	ssize_t ret = -EINVAL;
+
+	if (o2hb_region_attr->store)
+		ret = o2hb_region_attr->store(reg, page, count);
+	return ret;
+}
+
+static struct configfs_item_operations o2hb_region_item_ops = {
+	.release		= o2hb_region_release,
+	.show_attribute		= o2hb_region_show,
+	.store_attribute	= o2hb_region_store,
+};
+
+static struct config_item_type o2hb_region_type = {
+	.ct_item_ops	= &o2hb_region_item_ops,
+	.ct_attrs	= o2hb_region_attrs,
+	.ct_owner	= THIS_MODULE,
+};
+
+/* heartbeat set */
+
+struct o2hb_heartbeat_group {
+	struct config_group hs_group;
+	/* some stuff? */
+};
+
+static struct o2hb_heartbeat_group *to_o2hb_heartbeat_group(struct config_group *group)
+{
+	return group ?
+		container_of(group, struct o2hb_heartbeat_group, hs_group)
+		: NULL;
+}
+
+static int o2hb_debug_region_init(struct o2hb_region *reg, struct dentry *dir)
+{
+	int ret = -ENOMEM;
+
+	reg->hr_debug_dir =
+		debugfs_create_dir(config_item_name(&reg->hr_item), dir);
+	if (!reg->hr_debug_dir) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	reg->hr_debug_livenodes =
+			o2hb_debug_create(O2HB_DEBUG_LIVENODES,
+					  reg->hr_debug_dir,
+					  &(reg->hr_db_livenodes),
+					  sizeof(*(reg->hr_db_livenodes)),
+					  O2HB_DB_TYPE_REGION_LIVENODES,
+					  sizeof(reg->hr_live_node_bitmap),
+					  O2NM_MAX_NODES, reg);
+	if (!reg->hr_debug_livenodes) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	reg->hr_debug_regnum =
+			o2hb_debug_create(O2HB_DEBUG_REGION_NUMBER,
+					  reg->hr_debug_dir,
+					  &(reg->hr_db_regnum),
+					  sizeof(*(reg->hr_db_regnum)),
+					  O2HB_DB_TYPE_REGION_NUMBER,
+					  0, O2NM_MAX_NODES, reg);
+	if (!reg->hr_debug_regnum) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	reg->hr_debug_elapsed_time =
+			o2hb_debug_create(O2HB_DEBUG_REGION_ELAPSED_TIME,
+					  reg->hr_debug_dir,
+					  &(reg->hr_db_elapsed_time),
+					  sizeof(*(reg->hr_db_elapsed_time)),
+					  O2HB_DB_TYPE_REGION_ELAPSED_TIME,
+					  0, 0, reg);
+	if (!reg->hr_debug_elapsed_time) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	reg->hr_debug_pinned =
+			o2hb_debug_create(O2HB_DEBUG_REGION_PINNED,
+					  reg->hr_debug_dir,
+					  &(reg->hr_db_pinned),
+					  sizeof(*(reg->hr_db_pinned)),
+					  O2HB_DB_TYPE_REGION_PINNED,
+					  0, 0, reg);
+	if (!reg->hr_debug_pinned) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	ret = 0;
+bail:
+	return ret;
+}
+
+static struct config_item *o2hb_heartbeat_group_make_item(struct config_group *group,
+							  const char *name)
+{
+	struct o2hb_region *reg = NULL;
+	int ret;
+
+	reg = kzalloc(sizeof(struct o2hb_region), GFP_KERNEL);
+	if (reg == NULL)
+		return ERR_PTR(-ENOMEM);
+
+	if (strlen(name) > O2HB_MAX_REGION_NAME_LEN) {
+		ret = -ENAMETOOLONG;
+		goto free;
+	}
+
+	spin_lock(&o2hb_live_lock);
+	reg->hr_region_num = 0;
+	if (o2hb_global_heartbeat_active()) {
+		reg->hr_region_num = find_first_zero_bit(o2hb_region_bitmap,
+							 O2NM_MAX_REGIONS);
+		if (reg->hr_region_num >= O2NM_MAX_REGIONS) {
+			spin_unlock(&o2hb_live_lock);
+			ret = -EFBIG;
+			goto free;
+		}
+		set_bit(reg->hr_region_num, o2hb_region_bitmap);
+	}
+	list_add_tail(&reg->hr_all_item, &o2hb_all_regions);
+	spin_unlock(&o2hb_live_lock);
+
+	config_item_init_type_name(&reg->hr_item, name, &o2hb_region_type);
+
+	ret = o2hb_debug_region_init(reg, o2hb_debug_dir);
+	if (ret) {
+		config_item_put(&reg->hr_item);
+		goto free;
+	}
+
+	return &reg->hr_item;
+free:
+	kfree(reg);
+	return ERR_PTR(ret);
+}
+
+static void o2hb_heartbeat_group_drop_item(struct config_group *group,
+					   struct config_item *item)
+{
+	struct task_struct *hb_task;
+	struct o2hb_region *reg = to_o2hb_region(item);
+	int quorum_region = 0;
+
+	/* stop the thread when the user removes the region dir */
+	spin_lock(&o2hb_live_lock);
+	hb_task = reg->hr_task;
+	reg->hr_task = NULL;
+	reg->hr_item_dropped = 1;
+	spin_unlock(&o2hb_live_lock);
+
+	if (hb_task)
+		kthread_stop(hb_task);
+
+	if (o2hb_global_heartbeat_active()) {
+		spin_lock(&o2hb_live_lock);
+		clear_bit(reg->hr_region_num, o2hb_region_bitmap);
+		clear_bit(reg->hr_region_num, o2hb_live_region_bitmap);
+		if (test_bit(reg->hr_region_num, o2hb_quorum_region_bitmap))
+			quorum_region = 1;
+		clear_bit(reg->hr_region_num, o2hb_quorum_region_bitmap);
+		spin_unlock(&o2hb_live_lock);
+		printk(KERN_NOTICE "o2hb: Heartbeat %s on region %s (%s)\n",
+		       ((atomic_read(&reg->hr_steady_iterations) == 0) ?
+			"stopped" : "start aborted"), config_item_name(item),
+		       reg->hr_dev_name);
+	}
+
+	/*
+	 * If we're racing a dev_write(), we need to wake them.  They will
+	 * check reg->hr_task
+	 */
+	if (atomic_read(&reg->hr_steady_iterations) != 0) {
+		reg->hr_aborted_start = 1;
+		atomic_set(&reg->hr_steady_iterations, 0);
+		wake_up(&o2hb_steady_queue);
+	}
+
+	config_item_put(item);
+
+	if (!o2hb_global_heartbeat_active() || !quorum_region)
+		return;
+
+	/*
+	 * If global heartbeat active and there are dependent users,
+	 * pin all regions if quorum region count <= CUT_OFF
+	 */
+	spin_lock(&o2hb_live_lock);
+
+	if (!o2hb_dependent_users)
+		goto unlock;
+
+	if (bitmap_weight(o2hb_quorum_region_bitmap,
+			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
+		o2hb_region_pin(NULL);
+
+unlock:
+	spin_unlock(&o2hb_live_lock);
+}
+
+struct o2hb_heartbeat_group_attribute {
+	struct configfs_attribute attr;
+	ssize_t (*show)(struct o2hb_heartbeat_group *, char *);
+	ssize_t (*store)(struct o2hb_heartbeat_group *, const char *, size_t);
+};
+
+static ssize_t o2hb_heartbeat_group_show(struct config_item *item,
+					 struct configfs_attribute *attr,
+					 char *page)
+{
+	struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
+	struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
+		container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
+	ssize_t ret = 0;
+
+	if (o2hb_heartbeat_group_attr->show)
+		ret = o2hb_heartbeat_group_attr->show(reg, page);
+	return ret;
+}
+
+static ssize_t o2hb_heartbeat_group_store(struct config_item *item,
+					  struct configfs_attribute *attr,
+					  const char *page, size_t count)
+{
+	struct o2hb_heartbeat_group *reg = to_o2hb_heartbeat_group(to_config_group(item));
+	struct o2hb_heartbeat_group_attribute *o2hb_heartbeat_group_attr =
+		container_of(attr, struct o2hb_heartbeat_group_attribute, attr);
+	ssize_t ret = -EINVAL;
+
+	if (o2hb_heartbeat_group_attr->store)
+		ret = o2hb_heartbeat_group_attr->store(reg, page, count);
+	return ret;
+}
+
+static ssize_t o2hb_heartbeat_group_threshold_show(struct o2hb_heartbeat_group *group,
+						     char *page)
+{
+	return sprintf(page, "%u\n", o2hb_dead_threshold);
+}
+
+static ssize_t o2hb_heartbeat_group_threshold_store(struct o2hb_heartbeat_group *group,
+						    const char *page,
+						    size_t count)
+{
+	unsigned long tmp;
+	char *p = (char *)page;
+
+	tmp = simple_strtoul(p, &p, 10);
+	if (!p || (*p && (*p != '\n')))
+                return -EINVAL;
+
+	/* this will validate ranges for us. */
+	o2hb_dead_threshold_set((unsigned int) tmp);
+
+	return count;
+}
+
+static
+ssize_t o2hb_heartbeat_group_mode_show(struct o2hb_heartbeat_group *group,
+				       char *page)
+{
+	return sprintf(page, "%s\n",
+		       o2hb_heartbeat_mode_desc[o2hb_heartbeat_mode]);
+}
+
+static
+ssize_t o2hb_heartbeat_group_mode_store(struct o2hb_heartbeat_group *group,
+					const char *page, size_t count)
+{
+	unsigned int i;
+	int ret;
+	size_t len;
+
+	len = (page[count - 1] == '\n') ? count - 1 : count;
+	if (!len)
+		return -EINVAL;
+
+	for (i = 0; i < O2HB_HEARTBEAT_NUM_MODES; ++i) {
+		if (strncasecmp(page, o2hb_heartbeat_mode_desc[i], len))
+			continue;
+
+		ret = o2hb_global_heartbeat_mode_set(i);
+		if (!ret)
+			printk(KERN_NOTICE "o2hb: Heartbeat mode set to %s\n",
+			       o2hb_heartbeat_mode_desc[i]);
+		return count;
+	}
+
+	return -EINVAL;
+
+}
+
+static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_threshold = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "dead_threshold",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2hb_heartbeat_group_threshold_show,
+	.store	= o2hb_heartbeat_group_threshold_store,
+};
+
+static struct o2hb_heartbeat_group_attribute o2hb_heartbeat_group_attr_mode = {
+	.attr   = { .ca_owner = THIS_MODULE,
+		.ca_name = "mode",
+		.ca_mode = S_IRUGO | S_IWUSR },
+	.show   = o2hb_heartbeat_group_mode_show,
+	.store  = o2hb_heartbeat_group_mode_store,
+};
+
+static struct configfs_attribute *o2hb_heartbeat_group_attrs[] = {
+	&o2hb_heartbeat_group_attr_threshold.attr,
+	&o2hb_heartbeat_group_attr_mode.attr,
+	NULL,
+};
+
+static struct configfs_item_operations o2hb_heartbeat_group_item_ops = {
+	.show_attribute		= o2hb_heartbeat_group_show,
+	.store_attribute	= o2hb_heartbeat_group_store,
+};
+
+static struct configfs_group_operations o2hb_heartbeat_group_group_ops = {
+	.make_item	= o2hb_heartbeat_group_make_item,
+	.drop_item	= o2hb_heartbeat_group_drop_item,
+};
+
+static struct config_item_type o2hb_heartbeat_group_type = {
+	.ct_group_ops	= &o2hb_heartbeat_group_group_ops,
+	.ct_item_ops	= &o2hb_heartbeat_group_item_ops,
+	.ct_attrs	= o2hb_heartbeat_group_attrs,
+	.ct_owner	= THIS_MODULE,
+};
+
+/* this is just here to avoid touching group in heartbeat.h which the
+ * entire damn world #includes */
+struct config_group *o2hb_alloc_hb_set(void)
+{
+	struct o2hb_heartbeat_group *hs = NULL;
+	struct config_group *ret = NULL;
+
+	hs = kzalloc(sizeof(struct o2hb_heartbeat_group), GFP_KERNEL);
+	if (hs == NULL)
+		goto out;
+
+	config_group_init_type_name(&hs->hs_group, "heartbeat",
+				    &o2hb_heartbeat_group_type);
+
+	ret = &hs->hs_group;
+out:
+	if (ret == NULL)
+		kfree(hs);
+	return ret;
+}
+
+void o2hb_free_hb_set(struct config_group *group)
+{
+	struct o2hb_heartbeat_group *hs = to_o2hb_heartbeat_group(group);
+	kfree(hs);
+}
+
+/* hb callback registration and issuing */
+
+static struct o2hb_callback *hbcall_from_type(enum o2hb_callback_type type)
+{
+	if (type == O2HB_NUM_CB)
+		return ERR_PTR(-EINVAL);
+
+	return &o2hb_callbacks[type];
+}
+
+void o2hb_setup_callback(struct o2hb_callback_func *hc,
+			 enum o2hb_callback_type type,
+			 o2hb_cb_func *func,
+			 void *data,
+			 int priority)
+{
+	INIT_LIST_HEAD(&hc->hc_item);
+	hc->hc_func = func;
+	hc->hc_data = data;
+	hc->hc_priority = priority;
+	hc->hc_type = type;
+	hc->hc_magic = O2HB_CB_MAGIC;
+}
+EXPORT_SYMBOL_GPL(o2hb_setup_callback);
+
+/*
+ * In local heartbeat mode, region_uuid passed matches the dlm domain name.
+ * In global heartbeat mode, region_uuid passed is NULL.
+ *
+ * In local, we only pin the matching region. In global we pin all the active
+ * regions.
+ */
+static int o2hb_region_pin(const char *region_uuid)
+{
+	int ret = 0, found = 0;
+	struct o2hb_region *reg;
+	char *uuid;
+
+	assert_spin_locked(&o2hb_live_lock);
+
+	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
+		if (reg->hr_item_dropped)
+			continue;
+
+		uuid = config_item_name(&reg->hr_item);
+
+		/* local heartbeat */
+		if (region_uuid) {
+			if (strcmp(region_uuid, uuid))
+				continue;
+			found = 1;
+		}
+
+		if (reg->hr_item_pinned || reg->hr_item_dropped)
+			goto skip_pin;
+
+		/* Ignore ENOENT only for local hb (userdlm domain) */
+		ret = o2nm_depend_item(&reg->hr_item);
+		if (!ret) {
+			mlog(ML_CLUSTER, "Pin region %s\n", uuid);
+			reg->hr_item_pinned = 1;
+		} else {
+			if (ret == -ENOENT && found)
+				ret = 0;
+			else {
+				mlog(ML_ERROR, "Pin region %s fails with %d\n",
+				     uuid, ret);
+				break;
+			}
+		}
+skip_pin:
+		if (found)
+			break;
+	}
+
+	return ret;
+}
+
+/*
+ * In local heartbeat mode, region_uuid passed matches the dlm domain name.
+ * In global heartbeat mode, region_uuid passed is NULL.
+ *
+ * In local, we only unpin the matching region. In global we unpin all the
+ * active regions.
+ */
+static void o2hb_region_unpin(const char *region_uuid)
+{
+	struct o2hb_region *reg;
+	char *uuid;
+	int found = 0;
+
+	assert_spin_locked(&o2hb_live_lock);
+
+	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
+		if (reg->hr_item_dropped)
+			continue;
+
+		uuid = config_item_name(&reg->hr_item);
+		if (region_uuid) {
+			if (strcmp(region_uuid, uuid))
+				continue;
+			found = 1;
+		}
+
+		if (reg->hr_item_pinned) {
+			mlog(ML_CLUSTER, "Unpin region %s\n", uuid);
+			o2nm_undepend_item(&reg->hr_item);
+			reg->hr_item_pinned = 0;
+		}
+		if (found)
+			break;
+	}
+}
+
+static int o2hb_region_inc_user(const char *region_uuid)
+{
+	int ret = 0;
+
+	spin_lock(&o2hb_live_lock);
+
+	/* local heartbeat */
+	if (!o2hb_global_heartbeat_active()) {
+	    ret = o2hb_region_pin(region_uuid);
+	    goto unlock;
+	}
+
+	/*
+	 * if global heartbeat active and this is the first dependent user,
+	 * pin all regions if quorum region count <= CUT_OFF
+	 */
+	o2hb_dependent_users++;
+	if (o2hb_dependent_users > 1)
+		goto unlock;
+
+	if (bitmap_weight(o2hb_quorum_region_bitmap,
+			   O2NM_MAX_REGIONS) <= O2HB_PIN_CUT_OFF)
+		ret = o2hb_region_pin(NULL);
+
+unlock:
+	spin_unlock(&o2hb_live_lock);
+	return ret;
+}
+
+void o2hb_region_dec_user(const char *region_uuid)
+{
+	spin_lock(&o2hb_live_lock);
+
+	/* local heartbeat */
+	if (!o2hb_global_heartbeat_active()) {
+	    o2hb_region_unpin(region_uuid);
+	    goto unlock;
+	}
+
+	/*
+	 * if global heartbeat active and there are no dependent users,
+	 * unpin all quorum regions
+	 */
+	o2hb_dependent_users--;
+	if (!o2hb_dependent_users)
+		o2hb_region_unpin(NULL);
+
+unlock:
+	spin_unlock(&o2hb_live_lock);
+}
+
+int o2hb_register_callback(const char *region_uuid,
+			   struct o2hb_callback_func *hc)
+{
+	struct o2hb_callback_func *f;
+	struct o2hb_callback *hbcall;
+	int ret;
+
+	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
+	BUG_ON(!list_empty(&hc->hc_item));
+
+	hbcall = hbcall_from_type(hc->hc_type);
+	if (IS_ERR(hbcall)) {
+		ret = PTR_ERR(hbcall);
+		goto out;
+	}
+
+	if (region_uuid) {
+		ret = o2hb_region_inc_user(region_uuid);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	down_write(&o2hb_callback_sem);
+
+	list_for_each_entry(f, &hbcall->list, hc_item) {
+		if (hc->hc_priority < f->hc_priority) {
+			list_add_tail(&hc->hc_item, &f->hc_item);
+			break;
+		}
+	}
+	if (list_empty(&hc->hc_item))
+		list_add_tail(&hc->hc_item, &hbcall->list);
+
+	up_write(&o2hb_callback_sem);
+	ret = 0;
+out:
+	mlog(ML_CLUSTER, "returning %d on behalf of %p for funcs %p\n",
+	     ret, __builtin_return_address(0), hc);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(o2hb_register_callback);
+
+void o2hb_unregister_callback(const char *region_uuid,
+			      struct o2hb_callback_func *hc)
+{
+	BUG_ON(hc->hc_magic != O2HB_CB_MAGIC);
+
+	mlog(ML_CLUSTER, "on behalf of %p for funcs %p\n",
+	     __builtin_return_address(0), hc);
+
+	/* XXX Can this happen _with_ a region reference? */
+	if (list_empty(&hc->hc_item))
+		return;
+
+	if (region_uuid)
+		o2hb_region_dec_user(region_uuid);
+
+	down_write(&o2hb_callback_sem);
+
+	list_del_init(&hc->hc_item);
+
+	up_write(&o2hb_callback_sem);
+}
+EXPORT_SYMBOL_GPL(o2hb_unregister_callback);
+
+int o2hb_check_node_heartbeating(u8 node_num)
+{
+	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+
+	o2hb_fill_node_map(testing_map, sizeof(testing_map));
+	if (!test_bit(node_num, testing_map)) {
+		mlog(ML_HEARTBEAT,
+		     "node (%u) does not have heartbeating enabled.\n",
+		     node_num);
+		return 0;
+	}
+
+	return 1;
+}
+EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating);
+
+int o2hb_check_node_heartbeating_no_sem(u8 node_num)
+{
+	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long flags;
+
+	spin_lock_irqsave(&o2hb_live_lock, flags);
+	o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
+	spin_unlock_irqrestore(&o2hb_live_lock, flags);
+	if (!test_bit(node_num, testing_map)) {
+		mlog(ML_HEARTBEAT,
+		     "node (%u) does not have heartbeating enabled.\n",
+		     node_num);
+		return 0;
+	}
+
+	return 1;
+}
+EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_no_sem);
+
+int o2hb_check_node_heartbeating_from_callback(u8 node_num)
+{
+	unsigned long testing_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+
+	o2hb_fill_node_map_from_callback(testing_map, sizeof(testing_map));
+	if (!test_bit(node_num, testing_map)) {
+		mlog(ML_HEARTBEAT,
+		     "node (%u) does not have heartbeating enabled.\n",
+		     node_num);
+		return 0;
+	}
+
+	return 1;
+}
+EXPORT_SYMBOL_GPL(o2hb_check_node_heartbeating_from_callback);
+
+/* Makes sure our local node is configured with a node number, and is
+ * heartbeating. */
+int o2hb_check_local_node_heartbeating(void)
+{
+	u8 node_num;
+
+	/* if this node was set then we have networking */
+	node_num = o2nm_this_node();
+	if (node_num == O2NM_MAX_NODES) {
+		mlog(ML_HEARTBEAT, "this node has not been configured.\n");
+		return 0;
+	}
+
+	return o2hb_check_node_heartbeating(node_num);
+}
+EXPORT_SYMBOL_GPL(o2hb_check_local_node_heartbeating);
+
+/*
+ * this is just a hack until we get the plumbing which flips file systems
+ * read only and drops the hb ref instead of killing the node dead.
+ */
+void o2hb_stop_all_regions(void)
+{
+	struct o2hb_region *reg;
+
+	mlog(ML_ERROR, "stopping heartbeat on all active regions.\n");
+
+	spin_lock(&o2hb_live_lock);
+
+	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item)
+		reg->hr_unclean_stop = 1;
+
+	spin_unlock(&o2hb_live_lock);
+}
+EXPORT_SYMBOL_GPL(o2hb_stop_all_regions);
+
+int o2hb_get_all_regions(char *region_uuids, u8 max_regions)
+{
+	struct o2hb_region *reg;
+	int numregs = 0;
+	char *p;
+
+	spin_lock(&o2hb_live_lock);
+
+	p = region_uuids;
+	list_for_each_entry(reg, &o2hb_all_regions, hr_all_item) {
+		if (reg->hr_item_dropped)
+			continue;
+
+		mlog(0, "Region: %s\n", config_item_name(&reg->hr_item));
+		if (numregs < max_regions) {
+			memcpy(p, config_item_name(&reg->hr_item),
+			       O2HB_MAX_REGION_NAME_LEN);
+			p += O2HB_MAX_REGION_NAME_LEN;
+		}
+		numregs++;
+	}
+
+	spin_unlock(&o2hb_live_lock);
+
+	return numregs;
+}
+EXPORT_SYMBOL_GPL(o2hb_get_all_regions);
+
+int o2hb_global_heartbeat_active(void)
+{
+	return (o2hb_heartbeat_mode == O2HB_HEARTBEAT_GLOBAL);
+}
+EXPORT_SYMBOL(o2hb_global_heartbeat_active);
diff --git a/fs/ocfs2/cluster/heartbeat.h b/fs/ocfs2/cluster/heartbeat.h
new file mode 100644
index 000000000..3ef5137dc
--- /dev/null
+++ b/fs/ocfs2/cluster/heartbeat.h
@@ -0,0 +1,90 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * heartbeat.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef O2CLUSTER_HEARTBEAT_H
+#define O2CLUSTER_HEARTBEAT_H
+
+#include "ocfs2_heartbeat.h"
+
+#define O2HB_REGION_TIMEOUT_MS		2000
+
+#define O2HB_MAX_REGION_NAME_LEN	32
+
+/* number of changes to be seen as live */
+#define O2HB_LIVE_THRESHOLD	   2
+/* number of equal samples to be seen as dead */
+extern unsigned int o2hb_dead_threshold;
+#define O2HB_DEFAULT_DEAD_THRESHOLD	   31
+/* Otherwise MAX_WRITE_TIMEOUT will be zero... */
+#define O2HB_MIN_DEAD_THRESHOLD	  2
+#define O2HB_MAX_WRITE_TIMEOUT_MS (O2HB_REGION_TIMEOUT_MS * (o2hb_dead_threshold - 1))
+
+#define O2HB_CB_MAGIC		0x51d1e4ec
+
+/* callback stuff */
+enum o2hb_callback_type {
+	O2HB_NODE_DOWN_CB = 0,
+	O2HB_NODE_UP_CB,
+	O2HB_NUM_CB
+};
+
+struct o2nm_node;
+typedef void (o2hb_cb_func)(struct o2nm_node *, int, void *);
+
+struct o2hb_callback_func {
+	u32			hc_magic;
+	struct list_head	hc_item;
+	o2hb_cb_func		*hc_func;
+	void			*hc_data;
+	int			hc_priority;
+	enum o2hb_callback_type hc_type;
+};
+
+struct config_group *o2hb_alloc_hb_set(void);
+void o2hb_free_hb_set(struct config_group *group);
+
+void o2hb_setup_callback(struct o2hb_callback_func *hc,
+			 enum o2hb_callback_type type,
+			 o2hb_cb_func *func,
+			 void *data,
+			 int priority);
+int o2hb_register_callback(const char *region_uuid,
+			   struct o2hb_callback_func *hc);
+void o2hb_unregister_callback(const char *region_uuid,
+			      struct o2hb_callback_func *hc);
+void o2hb_fill_node_map(unsigned long *map,
+			unsigned bytes);
+void o2hb_exit(void);
+int o2hb_init(void);
+int o2hb_check_node_heartbeating(u8 node_num);
+int o2hb_check_node_heartbeating_no_sem(u8 node_num);
+int o2hb_check_node_heartbeating_from_callback(u8 node_num);
+int o2hb_check_local_node_heartbeating(void);
+void o2hb_stop_all_regions(void);
+int o2hb_get_all_regions(char *region_uuids, u8 numregions);
+int o2hb_global_heartbeat_active(void);
+
+#endif /* O2CLUSTER_HEARTBEAT_H */
diff --git a/fs/ocfs2/cluster/masklog.c b/fs/ocfs2/cluster/masklog.c
new file mode 100644
index 000000000..af7598bff
--- /dev/null
+++ b/fs/ocfs2/cluster/masklog.c
@@ -0,0 +1,155 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include <linux/string.h>
+#include <asm/uaccess.h>
+
+#include "masklog.h"
+
+struct mlog_bits mlog_and_bits = MLOG_BITS_RHS(MLOG_INITIAL_AND_MASK);
+EXPORT_SYMBOL_GPL(mlog_and_bits);
+struct mlog_bits mlog_not_bits = MLOG_BITS_RHS(0);
+EXPORT_SYMBOL_GPL(mlog_not_bits);
+
+static ssize_t mlog_mask_show(u64 mask, char *buf)
+{
+	char *state;
+
+	if (__mlog_test_u64(mask, mlog_and_bits))
+		state = "allow";
+	else if (__mlog_test_u64(mask, mlog_not_bits))
+		state = "deny";
+	else
+		state = "off";
+
+	return snprintf(buf, PAGE_SIZE, "%s\n", state);
+}
+
+static ssize_t mlog_mask_store(u64 mask, const char *buf, size_t count)
+{
+	if (!strncasecmp(buf, "allow", 5)) {
+		__mlog_set_u64(mask, mlog_and_bits);
+		__mlog_clear_u64(mask, mlog_not_bits);
+	} else if (!strncasecmp(buf, "deny", 4)) {
+		__mlog_set_u64(mask, mlog_not_bits);
+		__mlog_clear_u64(mask, mlog_and_bits);
+	} else if (!strncasecmp(buf, "off", 3)) {
+		__mlog_clear_u64(mask, mlog_not_bits);
+		__mlog_clear_u64(mask, mlog_and_bits);
+	} else
+		return -EINVAL;
+
+	return count;
+}
+
+struct mlog_attribute {
+	struct attribute attr;
+	u64 mask;
+};
+
+#define to_mlog_attr(_attr) container_of(_attr, struct mlog_attribute, attr)
+
+#define define_mask(_name) {			\
+	.attr = {				\
+		.name = #_name,			\
+		.mode = S_IRUGO | S_IWUSR,	\
+	},					\
+	.mask = ML_##_name,			\
+}
+
+static struct mlog_attribute mlog_attrs[MLOG_MAX_BITS] = {
+	define_mask(TCP),
+	define_mask(MSG),
+	define_mask(SOCKET),
+	define_mask(HEARTBEAT),
+	define_mask(HB_BIO),
+	define_mask(DLMFS),
+	define_mask(DLM),
+	define_mask(DLM_DOMAIN),
+	define_mask(DLM_THREAD),
+	define_mask(DLM_MASTER),
+	define_mask(DLM_RECOVERY),
+	define_mask(DLM_GLUE),
+	define_mask(VOTE),
+	define_mask(CONN),
+	define_mask(QUORUM),
+	define_mask(BASTS),
+	define_mask(CLUSTER),
+	define_mask(ERROR),
+	define_mask(NOTICE),
+	define_mask(KTHREAD),
+};
+
+static struct attribute *mlog_attr_ptrs[MLOG_MAX_BITS] = {NULL, };
+
+static ssize_t mlog_show(struct kobject *obj, struct attribute *attr,
+			 char *buf)
+{
+	struct mlog_attribute *mlog_attr = to_mlog_attr(attr);
+
+	return mlog_mask_show(mlog_attr->mask, buf);
+}
+
+static ssize_t mlog_store(struct kobject *obj, struct attribute *attr,
+			  const char *buf, size_t count)
+{
+	struct mlog_attribute *mlog_attr = to_mlog_attr(attr);
+
+	return mlog_mask_store(mlog_attr->mask, buf, count);
+}
+
+static const struct sysfs_ops mlog_attr_ops = {
+	.show  = mlog_show,
+	.store = mlog_store,
+};
+
+static struct kobj_type mlog_ktype = {
+	.default_attrs = mlog_attr_ptrs,
+	.sysfs_ops     = &mlog_attr_ops,
+};
+
+static struct kset mlog_kset = {
+	.kobj   = {.ktype = &mlog_ktype},
+};
+
+int mlog_sys_init(struct kset *o2cb_kset)
+{
+	int i = 0;
+
+	while (mlog_attrs[i].attr.mode) {
+		mlog_attr_ptrs[i] = &mlog_attrs[i].attr;
+		i++;
+	}
+	mlog_attr_ptrs[i] = NULL;
+
+	kobject_set_name(&mlog_kset.kobj, "logmask");
+	mlog_kset.kobj.kset = o2cb_kset;
+	return kset_register(&mlog_kset);
+}
+
+void mlog_sys_shutdown(void)
+{
+	kset_unregister(&mlog_kset);
+}
diff --git a/fs/ocfs2/cluster/masklog.h b/fs/ocfs2/cluster/masklog.h
new file mode 100644
index 000000000..7fdc25a4d
--- /dev/null
+++ b/fs/ocfs2/cluster/masklog.h
@@ -0,0 +1,221 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef O2CLUSTER_MASKLOG_H
+#define O2CLUSTER_MASKLOG_H
+
+/*
+ * For now this is a trivial wrapper around printk() that gives the critical
+ * ability to enable sets of debugging output at run-time.  In the future this
+ * will almost certainly be redirected to relayfs so that it can pay a
+ * substantially lower heisenberg tax.
+ *
+ * Callers associate the message with a bitmask and a global bitmask is
+ * maintained with help from /proc.  If any of the bits match the message is
+ * output.
+ *
+ * We must have efficient bit tests on i386 and it seems gcc still emits crazy
+ * code for the 64bit compare.  It emits very good code for the dual unsigned
+ * long tests, though, completely avoiding tests that can never pass if the
+ * caller gives a constant bitmask that fills one of the longs with all 0s.  So
+ * the desire is to have almost all of the calls decided on by comparing just
+ * one of the longs.  This leads to having infrequently given bits that are
+ * frequently matched in the high bits.
+ *
+ * _ERROR and _NOTICE are used for messages that always go to the console and
+ * have appropriate KERN_ prefixes.  We wrap these in our function instead of
+ * just calling printk() so that this can eventually make its way through
+ * relayfs along with the debugging messages.  Everything else gets KERN_DEBUG.
+ * The inline tests and macro dance give GCC the opportunity to quite cleverly
+ * only emit the appropriage printk() when the caller passes in a constant
+ * mask, as is almost always the case.
+ *
+ * All this bitmask nonsense is managed from the files under
+ * /sys/fs/o2cb/logmask/.  Reading the files gives a straightforward
+ * indication of which bits are allowed (allow) or denied (off/deny).
+ * 	ENTRY deny
+ * 	EXIT deny
+ * 	TCP off
+ * 	MSG off
+ * 	SOCKET off
+ * 	ERROR allow
+ * 	NOTICE allow
+ *
+ * Writing changes the state of a given bit and requires a strictly formatted
+ * single write() call:
+ *
+ * 	write(fd, "allow", 5);
+ *
+ * Echoing allow/deny/off string into the logmask files can flip the bits
+ * on or off as expected; here is the bash script for example:
+ *
+ * log_mask="/sys/fs/o2cb/log_mask"
+ * for node in ENTRY EXIT TCP MSG SOCKET ERROR NOTICE; do
+ *	echo allow >"$log_mask"/"$node"
+ * done
+ *
+ * The debugfs.ocfs2 tool can also flip the bits with the -l option:
+ *
+ * debugfs.ocfs2 -l TCP allow
+ */
+
+/* for task_struct */
+#include <linux/sched.h>
+
+/* bits that are frequently given and infrequently matched in the low word */
+/* NOTE: If you add a flag, you need to also update masklog.c! */
+#define ML_TCP		0x0000000000000001ULL /* net cluster/tcp.c */
+#define ML_MSG		0x0000000000000002ULL /* net network messages */
+#define ML_SOCKET	0x0000000000000004ULL /* net socket lifetime */
+#define ML_HEARTBEAT	0x0000000000000008ULL /* hb all heartbeat tracking */
+#define ML_HB_BIO	0x0000000000000010ULL /* hb io tracing */
+#define ML_DLMFS	0x0000000000000020ULL /* dlm user dlmfs */
+#define ML_DLM		0x0000000000000040ULL /* dlm general debugging */
+#define ML_DLM_DOMAIN	0x0000000000000080ULL /* dlm domain debugging */
+#define ML_DLM_THREAD	0x0000000000000100ULL /* dlm domain thread */
+#define ML_DLM_MASTER	0x0000000000000200ULL /* dlm master functions */
+#define ML_DLM_RECOVERY	0x0000000000000400ULL /* dlm master functions */
+#define ML_DLM_GLUE	0x0000000000000800ULL /* ocfs2 dlm glue layer */
+#define ML_VOTE		0x0000000000001000ULL /* ocfs2 node messaging  */
+#define ML_CONN		0x0000000000002000ULL /* net connection management */
+#define ML_QUORUM	0x0000000000004000ULL /* net connection quorum */
+#define ML_BASTS	0x0000000000008000ULL /* dlmglue asts and basts */
+#define ML_CLUSTER	0x0000000000010000ULL /* cluster stack */
+
+/* bits that are infrequently given and frequently matched in the high word */
+#define ML_ERROR	0x1000000000000000ULL /* sent to KERN_ERR */
+#define ML_NOTICE	0x2000000000000000ULL /* setn to KERN_NOTICE */
+#define ML_KTHREAD	0x4000000000000000ULL /* kernel thread activity */
+
+#define MLOG_INITIAL_AND_MASK (ML_ERROR|ML_NOTICE)
+#ifndef MLOG_MASK_PREFIX
+#define MLOG_MASK_PREFIX 0
+#endif
+
+/*
+ * When logging is disabled, force the bit test to 0 for anything other
+ * than errors and notices, allowing gcc to remove the code completely.
+ * When enabled, allow all masks.
+ */
+#if defined(CONFIG_OCFS2_DEBUG_MASKLOG)
+#define ML_ALLOWED_BITS ~0
+#else
+#define ML_ALLOWED_BITS (ML_ERROR|ML_NOTICE)
+#endif
+
+#define MLOG_MAX_BITS 64
+
+struct mlog_bits {
+	unsigned long words[MLOG_MAX_BITS / BITS_PER_LONG];
+};
+
+extern struct mlog_bits mlog_and_bits, mlog_not_bits;
+
+#if BITS_PER_LONG == 32
+
+#define __mlog_test_u64(mask, bits)			\
+	( (u32)(mask & 0xffffffff) & bits.words[0] || 	\
+	  ((u64)(mask) >> 32) & bits.words[1] )
+#define __mlog_set_u64(mask, bits) do {			\
+	bits.words[0] |= (u32)(mask & 0xffffffff);	\
+       	bits.words[1] |= (u64)(mask) >> 32;		\
+} while (0)
+#define __mlog_clear_u64(mask, bits) do {		\
+	bits.words[0] &= ~((u32)(mask & 0xffffffff));	\
+       	bits.words[1] &= ~((u64)(mask) >> 32);		\
+} while (0)
+#define MLOG_BITS_RHS(mask) {				\
+	{						\
+		[0] = (u32)(mask & 0xffffffff),		\
+		[1] = (u64)(mask) >> 32,		\
+	}						\
+}
+
+#else /* 32bit long above, 64bit long below */
+
+#define __mlog_test_u64(mask, bits)	((mask) & bits.words[0])
+#define __mlog_set_u64(mask, bits) do {		\
+	bits.words[0] |= (mask);		\
+} while (0)
+#define __mlog_clear_u64(mask, bits) do {	\
+	bits.words[0] &= ~(mask);		\
+} while (0)
+#define MLOG_BITS_RHS(mask) { { (mask) } }
+
+#endif
+
+/*
+ * smp_processor_id() "helpfully" screams when called outside preemptible
+ * regions in current kernels.  sles doesn't have the variants that don't
+ * scream.  just do this instead of trying to guess which we're building
+ * against.. *sigh*.
+ */
+#define __mlog_cpu_guess ({		\
+	unsigned long _cpu = get_cpu();	\
+	put_cpu();			\
+	_cpu;				\
+})
+
+/* In the following two macros, the whitespace after the ',' just
+ * before ##args is intentional. Otherwise, gcc 2.95 will eat the
+ * previous token if args expands to nothing.
+ */
+#define __mlog_printk(level, fmt, args...)				\
+	printk(level "(%s,%u,%lu):%s:%d " fmt, current->comm,		\
+	       task_pid_nr(current), __mlog_cpu_guess,			\
+	       __PRETTY_FUNCTION__, __LINE__ , ##args)
+
+#define mlog(mask, fmt, args...) do {					\
+	u64 __m = MLOG_MASK_PREFIX | (mask);				\
+	if ((__m & ML_ALLOWED_BITS) &&					\
+	    __mlog_test_u64(__m, mlog_and_bits) &&			\
+	    !__mlog_test_u64(__m, mlog_not_bits)) {			\
+		if (__m & ML_ERROR)					\
+			__mlog_printk(KERN_ERR, "ERROR: "fmt , ##args);	\
+		else if (__m & ML_NOTICE)				\
+			__mlog_printk(KERN_NOTICE, fmt , ##args);	\
+		else __mlog_printk(KERN_INFO, fmt , ##args);		\
+	}								\
+} while (0)
+
+#define mlog_errno(st) ({						\
+	int _st = (st);							\
+	if (_st != -ERESTARTSYS && _st != -EINTR &&			\
+	    _st != AOP_TRUNCATED_PAGE && _st != -ENOSPC &&		\
+	    _st != -EDQUOT)						\
+		mlog(ML_ERROR, "status = %lld\n", (long long)_st);	\
+	_st;								\
+})
+
+#define mlog_bug_on_msg(cond, fmt, args...) do {			\
+	if (cond) {							\
+		mlog(ML_ERROR, "bug expression: " #cond "\n");		\
+		mlog(ML_ERROR, fmt, ##args);				\
+		BUG();							\
+	}								\
+} while (0)
+
+#include <linux/kobject.h>
+#include <linux/sysfs.h>
+int mlog_sys_init(struct kset *o2cb_subsys);
+void mlog_sys_shutdown(void);
+
+#endif /* O2CLUSTER_MASKLOG_H */
diff --git a/fs/ocfs2/cluster/netdebug.c b/fs/ocfs2/cluster/netdebug.c
new file mode 100644
index 000000000..27d1242c8
--- /dev/null
+++ b/fs/ocfs2/cluster/netdebug.c
@@ -0,0 +1,539 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * netdebug.c
+ *
+ * debug functionality for o2net
+ *
+ * Copyright (C) 2005, 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifdef CONFIG_DEBUG_FS
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/idr.h>
+#include <linux/kref.h>
+#include <linux/seq_file.h>
+#include <linux/debugfs.h>
+
+#include <linux/uaccess.h>
+
+#include "tcp.h"
+#include "nodemanager.h"
+#define MLOG_MASK_PREFIX ML_TCP
+#include "masklog.h"
+
+#include "tcp_internal.h"
+
+#define O2NET_DEBUG_DIR		"o2net"
+#define SC_DEBUG_NAME		"sock_containers"
+#define NST_DEBUG_NAME		"send_tracking"
+#define STATS_DEBUG_NAME	"stats"
+#define NODES_DEBUG_NAME	"connected_nodes"
+
+#define SHOW_SOCK_CONTAINERS	0
+#define SHOW_SOCK_STATS		1
+
+static struct dentry *o2net_dentry;
+static struct dentry *sc_dentry;
+static struct dentry *nst_dentry;
+static struct dentry *stats_dentry;
+static struct dentry *nodes_dentry;
+
+static DEFINE_SPINLOCK(o2net_debug_lock);
+
+static LIST_HEAD(sock_containers);
+static LIST_HEAD(send_tracking);
+
+void o2net_debug_add_nst(struct o2net_send_tracking *nst)
+{
+	spin_lock(&o2net_debug_lock);
+	list_add(&nst->st_net_debug_item, &send_tracking);
+	spin_unlock(&o2net_debug_lock);
+}
+
+void o2net_debug_del_nst(struct o2net_send_tracking *nst)
+{
+	spin_lock(&o2net_debug_lock);
+	if (!list_empty(&nst->st_net_debug_item))
+		list_del_init(&nst->st_net_debug_item);
+	spin_unlock(&o2net_debug_lock);
+}
+
+static struct o2net_send_tracking
+			*next_nst(struct o2net_send_tracking *nst_start)
+{
+	struct o2net_send_tracking *nst, *ret = NULL;
+
+	assert_spin_locked(&o2net_debug_lock);
+
+	list_for_each_entry(nst, &nst_start->st_net_debug_item,
+			    st_net_debug_item) {
+		/* discover the head of the list */
+		if (&nst->st_net_debug_item == &send_tracking)
+			break;
+
+		/* use st_task to detect real nsts in the list */
+		if (nst->st_task != NULL) {
+			ret = nst;
+			break;
+		}
+	}
+
+	return ret;
+}
+
+static void *nst_seq_start(struct seq_file *seq, loff_t *pos)
+{
+	struct o2net_send_tracking *nst, *dummy_nst = seq->private;
+
+	spin_lock(&o2net_debug_lock);
+	nst = next_nst(dummy_nst);
+	spin_unlock(&o2net_debug_lock);
+
+	return nst;
+}
+
+static void *nst_seq_next(struct seq_file *seq, void *v, loff_t *pos)
+{
+	struct o2net_send_tracking *nst, *dummy_nst = seq->private;
+
+	spin_lock(&o2net_debug_lock);
+	nst = next_nst(dummy_nst);
+	list_del_init(&dummy_nst->st_net_debug_item);
+	if (nst)
+		list_add(&dummy_nst->st_net_debug_item,
+			 &nst->st_net_debug_item);
+	spin_unlock(&o2net_debug_lock);
+
+	return nst; /* unused, just needs to be null when done */
+}
+
+static int nst_seq_show(struct seq_file *seq, void *v)
+{
+	struct o2net_send_tracking *nst, *dummy_nst = seq->private;
+	ktime_t now;
+	s64 sock, send, status;
+
+	spin_lock(&o2net_debug_lock);
+	nst = next_nst(dummy_nst);
+	if (!nst)
+		goto out;
+
+	now = ktime_get();
+	sock = ktime_to_us(ktime_sub(now, nst->st_sock_time));
+	send = ktime_to_us(ktime_sub(now, nst->st_send_time));
+	status = ktime_to_us(ktime_sub(now, nst->st_status_time));
+
+	/* get_task_comm isn't exported.  oh well. */
+	seq_printf(seq, "%p:\n"
+		   "  pid:          %lu\n"
+		   "  tgid:         %lu\n"
+		   "  process name: %s\n"
+		   "  node:         %u\n"
+		   "  sc:           %p\n"
+		   "  message id:   %d\n"
+		   "  message type: %u\n"
+		   "  message key:  0x%08x\n"
+		   "  sock acquiry: %lld usecs ago\n"
+		   "  send start:   %lld usecs ago\n"
+		   "  wait start:   %lld usecs ago\n",
+		   nst, (unsigned long)task_pid_nr(nst->st_task),
+		   (unsigned long)nst->st_task->tgid,
+		   nst->st_task->comm, nst->st_node,
+		   nst->st_sc, nst->st_id, nst->st_msg_type,
+		   nst->st_msg_key,
+		   (long long)sock,
+		   (long long)send,
+		   (long long)status);
+
+out:
+	spin_unlock(&o2net_debug_lock);
+
+	return 0;
+}
+
+static void nst_seq_stop(struct seq_file *seq, void *v)
+{
+}
+
+static const struct seq_operations nst_seq_ops = {
+	.start = nst_seq_start,
+	.next = nst_seq_next,
+	.stop = nst_seq_stop,
+	.show = nst_seq_show,
+};
+
+static int nst_fop_open(struct inode *inode, struct file *file)
+{
+	struct o2net_send_tracking *dummy_nst;
+
+	dummy_nst = __seq_open_private(file, &nst_seq_ops, sizeof(*dummy_nst));
+	if (!dummy_nst)
+		return -ENOMEM;
+	o2net_debug_add_nst(dummy_nst);
+
+	return 0;
+}
+
+static int nst_fop_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq = file->private_data;
+	struct o2net_send_tracking *dummy_nst = seq->private;
+
+	o2net_debug_del_nst(dummy_nst);
+	return seq_release_private(inode, file);
+}
+
+static const struct file_operations nst_seq_fops = {
+	.open = nst_fop_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = nst_fop_release,
+};
+
+void o2net_debug_add_sc(struct o2net_sock_container *sc)
+{
+	spin_lock(&o2net_debug_lock);
+	list_add(&sc->sc_net_debug_item, &sock_containers);
+	spin_unlock(&o2net_debug_lock);
+}
+
+void o2net_debug_del_sc(struct o2net_sock_container *sc)
+{
+	spin_lock(&o2net_debug_lock);
+	list_del_init(&sc->sc_net_debug_item);
+	spin_unlock(&o2net_debug_lock);
+}
+
+struct o2net_sock_debug {
+	int dbg_ctxt;
+	struct o2net_sock_container *dbg_sock;
+};
+
+static struct o2net_sock_container
+			*next_sc(struct o2net_sock_container *sc_start)
+{
+	struct o2net_sock_container *sc, *ret = NULL;
+
+	assert_spin_locked(&o2net_debug_lock);
+
+	list_for_each_entry(sc, &sc_start->sc_net_debug_item,
+			    sc_net_debug_item) {
+		/* discover the head of the list miscast as a sc */
+		if (&sc->sc_net_debug_item == &sock_containers)
+			break;
+
+		/* use sc_page to detect real scs in the list */
+		if (sc->sc_page != NULL) {
+			ret = sc;
+			break;
+		}
+	}
+
+	return ret;
+}
+
+static void *sc_seq_start(struct seq_file *seq, loff_t *pos)
+{
+	struct o2net_sock_debug *sd = seq->private;
+	struct o2net_sock_container *sc, *dummy_sc = sd->dbg_sock;
+
+	spin_lock(&o2net_debug_lock);
+	sc = next_sc(dummy_sc);
+	spin_unlock(&o2net_debug_lock);
+
+	return sc;
+}
+
+static void *sc_seq_next(struct seq_file *seq, void *v, loff_t *pos)
+{
+	struct o2net_sock_debug *sd = seq->private;
+	struct o2net_sock_container *sc, *dummy_sc = sd->dbg_sock;
+
+	spin_lock(&o2net_debug_lock);
+	sc = next_sc(dummy_sc);
+	list_del_init(&dummy_sc->sc_net_debug_item);
+	if (sc)
+		list_add(&dummy_sc->sc_net_debug_item, &sc->sc_net_debug_item);
+	spin_unlock(&o2net_debug_lock);
+
+	return sc; /* unused, just needs to be null when done */
+}
+
+#ifdef CONFIG_OCFS2_FS_STATS
+# define sc_send_count(_s)		((_s)->sc_send_count)
+# define sc_recv_count(_s)		((_s)->sc_recv_count)
+# define sc_tv_acquiry_total_ns(_s)	(ktime_to_ns((_s)->sc_tv_acquiry_total))
+# define sc_tv_send_total_ns(_s)	(ktime_to_ns((_s)->sc_tv_send_total))
+# define sc_tv_status_total_ns(_s)	(ktime_to_ns((_s)->sc_tv_status_total))
+# define sc_tv_process_total_ns(_s)	(ktime_to_ns((_s)->sc_tv_process_total))
+#else
+# define sc_send_count(_s)		(0U)
+# define sc_recv_count(_s)		(0U)
+# define sc_tv_acquiry_total_ns(_s)	(0LL)
+# define sc_tv_send_total_ns(_s)	(0LL)
+# define sc_tv_status_total_ns(_s)	(0LL)
+# define sc_tv_process_total_ns(_s)	(0LL)
+#endif
+
+/* So that debugfs.ocfs2 can determine which format is being used */
+#define O2NET_STATS_STR_VERSION		1
+static void sc_show_sock_stats(struct seq_file *seq,
+			       struct o2net_sock_container *sc)
+{
+	if (!sc)
+		return;
+
+	seq_printf(seq, "%d,%u,%lu,%lld,%lld,%lld,%lu,%lld\n", O2NET_STATS_STR_VERSION,
+		   sc->sc_node->nd_num, (unsigned long)sc_send_count(sc),
+		   (long long)sc_tv_acquiry_total_ns(sc),
+		   (long long)sc_tv_send_total_ns(sc),
+		   (long long)sc_tv_status_total_ns(sc),
+		   (unsigned long)sc_recv_count(sc),
+		   (long long)sc_tv_process_total_ns(sc));
+}
+
+static void sc_show_sock_container(struct seq_file *seq,
+				   struct o2net_sock_container *sc)
+{
+	struct inet_sock *inet = NULL;
+	__be32 saddr = 0, daddr = 0;
+	__be16 sport = 0, dport = 0;
+
+	if (!sc)
+		return;
+
+	if (sc->sc_sock) {
+		inet = inet_sk(sc->sc_sock->sk);
+		/* the stack's structs aren't sparse endian clean */
+		saddr = (__force __be32)inet->inet_saddr;
+		daddr = (__force __be32)inet->inet_daddr;
+		sport = (__force __be16)inet->inet_sport;
+		dport = (__force __be16)inet->inet_dport;
+	}
+
+	/* XXX sigh, inet-> doesn't have sparse annotation so any
+	 * use of it here generates a warning with -Wbitwise */
+	seq_printf(seq, "%p:\n"
+		   "  krefs:           %d\n"
+		   "  sock:            %pI4:%u -> "
+				      "%pI4:%u\n"
+		   "  remote node:     %s\n"
+		   "  page off:        %zu\n"
+		   "  handshake ok:    %u\n"
+		   "  timer:           %lld usecs\n"
+		   "  data ready:      %lld usecs\n"
+		   "  advance start:   %lld usecs\n"
+		   "  advance stop:    %lld usecs\n"
+		   "  func start:      %lld usecs\n"
+		   "  func stop:       %lld usecs\n"
+		   "  func key:        0x%08x\n"
+		   "  func type:       %u\n",
+		   sc,
+		   atomic_read(&sc->sc_kref.refcount),
+		   &saddr, inet ? ntohs(sport) : 0,
+		   &daddr, inet ? ntohs(dport) : 0,
+		   sc->sc_node->nd_name,
+		   sc->sc_page_off,
+		   sc->sc_handshake_ok,
+		   (long long)ktime_to_us(sc->sc_tv_timer),
+		   (long long)ktime_to_us(sc->sc_tv_data_ready),
+		   (long long)ktime_to_us(sc->sc_tv_advance_start),
+		   (long long)ktime_to_us(sc->sc_tv_advance_stop),
+		   (long long)ktime_to_us(sc->sc_tv_func_start),
+		   (long long)ktime_to_us(sc->sc_tv_func_stop),
+		   sc->sc_msg_key,
+		   sc->sc_msg_type);
+}
+
+static int sc_seq_show(struct seq_file *seq, void *v)
+{
+	struct o2net_sock_debug *sd = seq->private;
+	struct o2net_sock_container *sc, *dummy_sc = sd->dbg_sock;
+
+	spin_lock(&o2net_debug_lock);
+	sc = next_sc(dummy_sc);
+
+	if (sc) {
+		if (sd->dbg_ctxt == SHOW_SOCK_CONTAINERS)
+			sc_show_sock_container(seq, sc);
+		else
+			sc_show_sock_stats(seq, sc);
+	}
+
+	spin_unlock(&o2net_debug_lock);
+
+	return 0;
+}
+
+static void sc_seq_stop(struct seq_file *seq, void *v)
+{
+}
+
+static const struct seq_operations sc_seq_ops = {
+	.start = sc_seq_start,
+	.next = sc_seq_next,
+	.stop = sc_seq_stop,
+	.show = sc_seq_show,
+};
+
+static int sc_common_open(struct file *file, int ctxt)
+{
+	struct o2net_sock_debug *sd;
+	struct o2net_sock_container *dummy_sc;
+
+	dummy_sc = kzalloc(sizeof(*dummy_sc), GFP_KERNEL);
+	if (!dummy_sc)
+		return -ENOMEM;
+
+	sd = __seq_open_private(file, &sc_seq_ops, sizeof(*sd));
+	if (!sd) {
+		kfree(dummy_sc);
+		return -ENOMEM;
+	}
+
+	sd->dbg_ctxt = ctxt;
+	sd->dbg_sock = dummy_sc;
+
+	o2net_debug_add_sc(dummy_sc);
+
+	return 0;
+}
+
+static int sc_fop_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq = file->private_data;
+	struct o2net_sock_debug *sd = seq->private;
+	struct o2net_sock_container *dummy_sc = sd->dbg_sock;
+
+	o2net_debug_del_sc(dummy_sc);
+	return seq_release_private(inode, file);
+}
+
+static int stats_fop_open(struct inode *inode, struct file *file)
+{
+	return sc_common_open(file, SHOW_SOCK_STATS);
+}
+
+static const struct file_operations stats_seq_fops = {
+	.open = stats_fop_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = sc_fop_release,
+};
+
+static int sc_fop_open(struct inode *inode, struct file *file)
+{
+	return sc_common_open(file, SHOW_SOCK_CONTAINERS);
+}
+
+static const struct file_operations sc_seq_fops = {
+	.open = sc_fop_open,
+	.read = seq_read,
+	.llseek = seq_lseek,
+	.release = sc_fop_release,
+};
+
+static int o2net_fill_bitmap(char *buf, int len)
+{
+	unsigned long map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	int i = -1, out = 0;
+
+	o2net_fill_node_map(map, sizeof(map));
+
+	while ((i = find_next_bit(map, O2NM_MAX_NODES, i + 1)) < O2NM_MAX_NODES)
+		out += snprintf(buf + out, PAGE_SIZE - out, "%d ", i);
+	out += snprintf(buf + out, PAGE_SIZE - out, "\n");
+
+	return out;
+}
+
+static int nodes_fop_open(struct inode *inode, struct file *file)
+{
+	char *buf;
+
+	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
+
+	i_size_write(inode, o2net_fill_bitmap(buf, PAGE_SIZE));
+
+	file->private_data = buf;
+
+	return 0;
+}
+
+static int o2net_debug_release(struct inode *inode, struct file *file)
+{
+	kfree(file->private_data);
+	return 0;
+}
+
+static ssize_t o2net_debug_read(struct file *file, char __user *buf,
+				size_t nbytes, loff_t *ppos)
+{
+	return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
+				       i_size_read(file->f_mapping->host));
+}
+
+static const struct file_operations nodes_fops = {
+	.open		= nodes_fop_open,
+	.release	= o2net_debug_release,
+	.read		= o2net_debug_read,
+	.llseek		= generic_file_llseek,
+};
+
+void o2net_debugfs_exit(void)
+{
+	debugfs_remove(nodes_dentry);
+	debugfs_remove(stats_dentry);
+	debugfs_remove(sc_dentry);
+	debugfs_remove(nst_dentry);
+	debugfs_remove(o2net_dentry);
+}
+
+int o2net_debugfs_init(void)
+{
+	umode_t mode = S_IFREG|S_IRUSR;
+
+	o2net_dentry = debugfs_create_dir(O2NET_DEBUG_DIR, NULL);
+	if (o2net_dentry)
+		nst_dentry = debugfs_create_file(NST_DEBUG_NAME, mode,
+					o2net_dentry, NULL, &nst_seq_fops);
+	if (nst_dentry)
+		sc_dentry = debugfs_create_file(SC_DEBUG_NAME, mode,
+					o2net_dentry, NULL, &sc_seq_fops);
+	if (sc_dentry)
+		stats_dentry = debugfs_create_file(STATS_DEBUG_NAME, mode,
+					o2net_dentry, NULL, &stats_seq_fops);
+	if (stats_dentry)
+		nodes_dentry = debugfs_create_file(NODES_DEBUG_NAME, mode,
+					o2net_dentry, NULL, &nodes_fops);
+	if (nodes_dentry)
+		return 0;
+
+	o2net_debugfs_exit();
+	mlog_errno(-ENOMEM);
+	return -ENOMEM;
+}
+
+#endif	/* CONFIG_DEBUG_FS */
diff --git a/fs/ocfs2/cluster/nodemanager.c b/fs/ocfs2/cluster/nodemanager.c
new file mode 100644
index 000000000..441c84e16
--- /dev/null
+++ b/fs/ocfs2/cluster/nodemanager.c
@@ -0,0 +1,987 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2004, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/slab.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/configfs.h>
+
+#include "tcp.h"
+#include "nodemanager.h"
+#include "heartbeat.h"
+#include "masklog.h"
+#include "sys.h"
+
+/* for now we operate under the assertion that there can be only one
+ * cluster active at a time.  Changing this will require trickling
+ * cluster references throughout where nodes are looked up */
+struct o2nm_cluster *o2nm_single_cluster = NULL;
+
+char *o2nm_fence_method_desc[O2NM_FENCE_METHODS] = {
+		"reset",	/* O2NM_FENCE_RESET */
+		"panic",	/* O2NM_FENCE_PANIC */
+};
+
+struct o2nm_node *o2nm_get_node_by_num(u8 node_num)
+{
+	struct o2nm_node *node = NULL;
+
+	if (node_num >= O2NM_MAX_NODES || o2nm_single_cluster == NULL)
+		goto out;
+
+	read_lock(&o2nm_single_cluster->cl_nodes_lock);
+	node = o2nm_single_cluster->cl_nodes[node_num];
+	if (node)
+		config_item_get(&node->nd_item);
+	read_unlock(&o2nm_single_cluster->cl_nodes_lock);
+out:
+	return node;
+}
+EXPORT_SYMBOL_GPL(o2nm_get_node_by_num);
+
+int o2nm_configured_node_map(unsigned long *map, unsigned bytes)
+{
+	struct o2nm_cluster *cluster = o2nm_single_cluster;
+
+	BUG_ON(bytes < (sizeof(cluster->cl_nodes_bitmap)));
+
+	if (cluster == NULL)
+		return -EINVAL;
+
+	read_lock(&cluster->cl_nodes_lock);
+	memcpy(map, cluster->cl_nodes_bitmap, sizeof(cluster->cl_nodes_bitmap));
+	read_unlock(&cluster->cl_nodes_lock);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(o2nm_configured_node_map);
+
+static struct o2nm_node *o2nm_node_ip_tree_lookup(struct o2nm_cluster *cluster,
+						  __be32 ip_needle,
+						  struct rb_node ***ret_p,
+						  struct rb_node **ret_parent)
+{
+	struct rb_node **p = &cluster->cl_node_ip_tree.rb_node;
+	struct rb_node *parent = NULL;
+	struct o2nm_node *node, *ret = NULL;
+
+	while (*p) {
+		int cmp;
+
+		parent = *p;
+		node = rb_entry(parent, struct o2nm_node, nd_ip_node);
+
+		cmp = memcmp(&ip_needle, &node->nd_ipv4_address,
+				sizeof(ip_needle));
+		if (cmp < 0)
+			p = &(*p)->rb_left;
+		else if (cmp > 0)
+			p = &(*p)->rb_right;
+		else {
+			ret = node;
+			break;
+		}
+	}
+
+	if (ret_p != NULL)
+		*ret_p = p;
+	if (ret_parent != NULL)
+		*ret_parent = parent;
+
+	return ret;
+}
+
+struct o2nm_node *o2nm_get_node_by_ip(__be32 addr)
+{
+	struct o2nm_node *node = NULL;
+	struct o2nm_cluster *cluster = o2nm_single_cluster;
+
+	if (cluster == NULL)
+		goto out;
+
+	read_lock(&cluster->cl_nodes_lock);
+	node = o2nm_node_ip_tree_lookup(cluster, addr, NULL, NULL);
+	if (node)
+		config_item_get(&node->nd_item);
+	read_unlock(&cluster->cl_nodes_lock);
+
+out:
+	return node;
+}
+EXPORT_SYMBOL_GPL(o2nm_get_node_by_ip);
+
+void o2nm_node_put(struct o2nm_node *node)
+{
+	config_item_put(&node->nd_item);
+}
+EXPORT_SYMBOL_GPL(o2nm_node_put);
+
+void o2nm_node_get(struct o2nm_node *node)
+{
+	config_item_get(&node->nd_item);
+}
+EXPORT_SYMBOL_GPL(o2nm_node_get);
+
+u8 o2nm_this_node(void)
+{
+	u8 node_num = O2NM_MAX_NODES;
+
+	if (o2nm_single_cluster && o2nm_single_cluster->cl_has_local)
+		node_num = o2nm_single_cluster->cl_local_node;
+
+	return node_num;
+}
+EXPORT_SYMBOL_GPL(o2nm_this_node);
+
+/* node configfs bits */
+
+static struct o2nm_cluster *to_o2nm_cluster(struct config_item *item)
+{
+	return item ?
+		container_of(to_config_group(item), struct o2nm_cluster,
+			     cl_group)
+		: NULL;
+}
+
+static struct o2nm_node *to_o2nm_node(struct config_item *item)
+{
+	return item ? container_of(item, struct o2nm_node, nd_item) : NULL;
+}
+
+static void o2nm_node_release(struct config_item *item)
+{
+	struct o2nm_node *node = to_o2nm_node(item);
+	kfree(node);
+}
+
+static ssize_t o2nm_node_num_read(struct o2nm_node *node, char *page)
+{
+	return sprintf(page, "%d\n", node->nd_num);
+}
+
+static struct o2nm_cluster *to_o2nm_cluster_from_node(struct o2nm_node *node)
+{
+	/* through the first node_set .parent
+	 * mycluster/nodes/mynode == o2nm_cluster->o2nm_node_group->o2nm_node */
+	return to_o2nm_cluster(node->nd_item.ci_parent->ci_parent);
+}
+
+enum {
+	O2NM_NODE_ATTR_NUM = 0,
+	O2NM_NODE_ATTR_PORT,
+	O2NM_NODE_ATTR_ADDRESS,
+	O2NM_NODE_ATTR_LOCAL,
+};
+
+static ssize_t o2nm_node_num_write(struct o2nm_node *node, const char *page,
+				   size_t count)
+{
+	struct o2nm_cluster *cluster = to_o2nm_cluster_from_node(node);
+	unsigned long tmp;
+	char *p = (char *)page;
+
+	tmp = simple_strtoul(p, &p, 0);
+	if (!p || (*p && (*p != '\n')))
+		return -EINVAL;
+
+	if (tmp >= O2NM_MAX_NODES)
+		return -ERANGE;
+
+	/* once we're in the cl_nodes tree networking can look us up by
+	 * node number and try to use our address and port attributes
+	 * to connect to this node.. make sure that they've been set
+	 * before writing the node attribute? */
+	if (!test_bit(O2NM_NODE_ATTR_ADDRESS, &node->nd_set_attributes) ||
+	    !test_bit(O2NM_NODE_ATTR_PORT, &node->nd_set_attributes))
+		return -EINVAL; /* XXX */
+
+	write_lock(&cluster->cl_nodes_lock);
+	if (cluster->cl_nodes[tmp])
+		p = NULL;
+	else  {
+		cluster->cl_nodes[tmp] = node;
+		node->nd_num = tmp;
+		set_bit(tmp, cluster->cl_nodes_bitmap);
+	}
+	write_unlock(&cluster->cl_nodes_lock);
+	if (p == NULL)
+		return -EEXIST;
+
+	return count;
+}
+static ssize_t o2nm_node_ipv4_port_read(struct o2nm_node *node, char *page)
+{
+	return sprintf(page, "%u\n", ntohs(node->nd_ipv4_port));
+}
+
+static ssize_t o2nm_node_ipv4_port_write(struct o2nm_node *node,
+					 const char *page, size_t count)
+{
+	unsigned long tmp;
+	char *p = (char *)page;
+
+	tmp = simple_strtoul(p, &p, 0);
+	if (!p || (*p && (*p != '\n')))
+		return -EINVAL;
+
+	if (tmp == 0)
+		return -EINVAL;
+	if (tmp >= (u16)-1)
+		return -ERANGE;
+
+	node->nd_ipv4_port = htons(tmp);
+
+	return count;
+}
+
+static ssize_t o2nm_node_ipv4_address_read(struct o2nm_node *node, char *page)
+{
+	return sprintf(page, "%pI4\n", &node->nd_ipv4_address);
+}
+
+static ssize_t o2nm_node_ipv4_address_write(struct o2nm_node *node,
+					    const char *page,
+					    size_t count)
+{
+	struct o2nm_cluster *cluster = to_o2nm_cluster_from_node(node);
+	int ret, i;
+	struct rb_node **p, *parent;
+	unsigned int octets[4];
+	__be32 ipv4_addr = 0;
+
+	ret = sscanf(page, "%3u.%3u.%3u.%3u", &octets[3], &octets[2],
+		     &octets[1], &octets[0]);
+	if (ret != 4)
+		return -EINVAL;
+
+	for (i = 0; i < ARRAY_SIZE(octets); i++) {
+		if (octets[i] > 255)
+			return -ERANGE;
+		be32_add_cpu(&ipv4_addr, octets[i] << (i * 8));
+	}
+
+	ret = 0;
+	write_lock(&cluster->cl_nodes_lock);
+	if (o2nm_node_ip_tree_lookup(cluster, ipv4_addr, &p, &parent))
+		ret = -EEXIST;
+	else {
+		rb_link_node(&node->nd_ip_node, parent, p);
+		rb_insert_color(&node->nd_ip_node, &cluster->cl_node_ip_tree);
+	}
+	write_unlock(&cluster->cl_nodes_lock);
+	if (ret)
+		return ret;
+
+	memcpy(&node->nd_ipv4_address, &ipv4_addr, sizeof(ipv4_addr));
+
+	return count;
+}
+
+static ssize_t o2nm_node_local_read(struct o2nm_node *node, char *page)
+{
+	return sprintf(page, "%d\n", node->nd_local);
+}
+
+static ssize_t o2nm_node_local_write(struct o2nm_node *node, const char *page,
+				     size_t count)
+{
+	struct o2nm_cluster *cluster = to_o2nm_cluster_from_node(node);
+	unsigned long tmp;
+	char *p = (char *)page;
+	ssize_t ret;
+
+	tmp = simple_strtoul(p, &p, 0);
+	if (!p || (*p && (*p != '\n')))
+		return -EINVAL;
+
+	tmp = !!tmp; /* boolean of whether this node wants to be local */
+
+	/* setting local turns on networking rx for now so we require having
+	 * set everything else first */
+	if (!test_bit(O2NM_NODE_ATTR_ADDRESS, &node->nd_set_attributes) ||
+	    !test_bit(O2NM_NODE_ATTR_NUM, &node->nd_set_attributes) ||
+	    !test_bit(O2NM_NODE_ATTR_PORT, &node->nd_set_attributes))
+		return -EINVAL; /* XXX */
+
+	/* the only failure case is trying to set a new local node
+	 * when a different one is already set */
+	if (tmp && tmp == cluster->cl_has_local &&
+	    cluster->cl_local_node != node->nd_num)
+		return -EBUSY;
+
+	/* bring up the rx thread if we're setting the new local node. */
+	if (tmp && !cluster->cl_has_local) {
+		ret = o2net_start_listening(node);
+		if (ret)
+			return ret;
+	}
+
+	if (!tmp && cluster->cl_has_local &&
+	    cluster->cl_local_node == node->nd_num) {
+		o2net_stop_listening(node);
+		cluster->cl_local_node = O2NM_INVALID_NODE_NUM;
+	}
+
+	node->nd_local = tmp;
+	if (node->nd_local) {
+		cluster->cl_has_local = tmp;
+		cluster->cl_local_node = node->nd_num;
+	}
+
+	return count;
+}
+
+struct o2nm_node_attribute {
+	struct configfs_attribute attr;
+	ssize_t (*show)(struct o2nm_node *, char *);
+	ssize_t (*store)(struct o2nm_node *, const char *, size_t);
+};
+
+static struct o2nm_node_attribute o2nm_node_attr_num = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "num",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2nm_node_num_read,
+	.store	= o2nm_node_num_write,
+};
+
+static struct o2nm_node_attribute o2nm_node_attr_ipv4_port = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "ipv4_port",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2nm_node_ipv4_port_read,
+	.store	= o2nm_node_ipv4_port_write,
+};
+
+static struct o2nm_node_attribute o2nm_node_attr_ipv4_address = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "ipv4_address",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2nm_node_ipv4_address_read,
+	.store	= o2nm_node_ipv4_address_write,
+};
+
+static struct o2nm_node_attribute o2nm_node_attr_local = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "local",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2nm_node_local_read,
+	.store	= o2nm_node_local_write,
+};
+
+static struct configfs_attribute *o2nm_node_attrs[] = {
+	[O2NM_NODE_ATTR_NUM] = &o2nm_node_attr_num.attr,
+	[O2NM_NODE_ATTR_PORT] = &o2nm_node_attr_ipv4_port.attr,
+	[O2NM_NODE_ATTR_ADDRESS] = &o2nm_node_attr_ipv4_address.attr,
+	[O2NM_NODE_ATTR_LOCAL] = &o2nm_node_attr_local.attr,
+	NULL,
+};
+
+static int o2nm_attr_index(struct configfs_attribute *attr)
+{
+	int i;
+	for (i = 0; i < ARRAY_SIZE(o2nm_node_attrs); i++) {
+		if (attr == o2nm_node_attrs[i])
+			return i;
+	}
+	BUG();
+	return 0;
+}
+
+static ssize_t o2nm_node_show(struct config_item *item,
+			      struct configfs_attribute *attr,
+			      char *page)
+{
+	struct o2nm_node *node = to_o2nm_node(item);
+	struct o2nm_node_attribute *o2nm_node_attr =
+		container_of(attr, struct o2nm_node_attribute, attr);
+	ssize_t ret = 0;
+
+	if (o2nm_node_attr->show)
+		ret = o2nm_node_attr->show(node, page);
+	return ret;
+}
+
+static ssize_t o2nm_node_store(struct config_item *item,
+			       struct configfs_attribute *attr,
+			       const char *page, size_t count)
+{
+	struct o2nm_node *node = to_o2nm_node(item);
+	struct o2nm_node_attribute *o2nm_node_attr =
+		container_of(attr, struct o2nm_node_attribute, attr);
+	ssize_t ret;
+	int attr_index = o2nm_attr_index(attr);
+
+	if (o2nm_node_attr->store == NULL) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (test_bit(attr_index, &node->nd_set_attributes))
+		return -EBUSY;
+
+	ret = o2nm_node_attr->store(node, page, count);
+	if (ret < count)
+		goto out;
+
+	set_bit(attr_index, &node->nd_set_attributes);
+out:
+	return ret;
+}
+
+static struct configfs_item_operations o2nm_node_item_ops = {
+	.release		= o2nm_node_release,
+	.show_attribute		= o2nm_node_show,
+	.store_attribute	= o2nm_node_store,
+};
+
+static struct config_item_type o2nm_node_type = {
+	.ct_item_ops	= &o2nm_node_item_ops,
+	.ct_attrs	= o2nm_node_attrs,
+	.ct_owner	= THIS_MODULE,
+};
+
+/* node set */
+
+struct o2nm_node_group {
+	struct config_group ns_group;
+	/* some stuff? */
+};
+
+#if 0
+static struct o2nm_node_group *to_o2nm_node_group(struct config_group *group)
+{
+	return group ?
+		container_of(group, struct o2nm_node_group, ns_group)
+		: NULL;
+}
+#endif
+
+struct o2nm_cluster_attribute {
+	struct configfs_attribute attr;
+	ssize_t (*show)(struct o2nm_cluster *, char *);
+	ssize_t (*store)(struct o2nm_cluster *, const char *, size_t);
+};
+
+static ssize_t o2nm_cluster_attr_write(const char *page, ssize_t count,
+                                       unsigned int *val)
+{
+	unsigned long tmp;
+	char *p = (char *)page;
+
+	tmp = simple_strtoul(p, &p, 0);
+	if (!p || (*p && (*p != '\n')))
+		return -EINVAL;
+
+	if (tmp == 0)
+		return -EINVAL;
+	if (tmp >= (u32)-1)
+		return -ERANGE;
+
+	*val = tmp;
+
+	return count;
+}
+
+static ssize_t o2nm_cluster_attr_idle_timeout_ms_read(
+	struct o2nm_cluster *cluster, char *page)
+{
+	return sprintf(page, "%u\n", cluster->cl_idle_timeout_ms);
+}
+
+static ssize_t o2nm_cluster_attr_idle_timeout_ms_write(
+	struct o2nm_cluster *cluster, const char *page, size_t count)
+{
+	ssize_t ret;
+	unsigned int val;
+
+	ret =  o2nm_cluster_attr_write(page, count, &val);
+
+	if (ret > 0) {
+		if (cluster->cl_idle_timeout_ms != val
+			&& o2net_num_connected_peers()) {
+			mlog(ML_NOTICE,
+			     "o2net: cannot change idle timeout after "
+			     "the first peer has agreed to it."
+			     "  %d connected peers\n",
+			     o2net_num_connected_peers());
+			ret = -EINVAL;
+		} else if (val <= cluster->cl_keepalive_delay_ms) {
+			mlog(ML_NOTICE, "o2net: idle timeout must be larger "
+			     "than keepalive delay\n");
+			ret = -EINVAL;
+		} else {
+			cluster->cl_idle_timeout_ms = val;
+		}
+	}
+
+	return ret;
+}
+
+static ssize_t o2nm_cluster_attr_keepalive_delay_ms_read(
+	struct o2nm_cluster *cluster, char *page)
+{
+	return sprintf(page, "%u\n", cluster->cl_keepalive_delay_ms);
+}
+
+static ssize_t o2nm_cluster_attr_keepalive_delay_ms_write(
+	struct o2nm_cluster *cluster, const char *page, size_t count)
+{
+	ssize_t ret;
+	unsigned int val;
+
+	ret =  o2nm_cluster_attr_write(page, count, &val);
+
+	if (ret > 0) {
+		if (cluster->cl_keepalive_delay_ms != val
+		    && o2net_num_connected_peers()) {
+			mlog(ML_NOTICE,
+			     "o2net: cannot change keepalive delay after"
+			     " the first peer has agreed to it."
+			     "  %d connected peers\n",
+			     o2net_num_connected_peers());
+			ret = -EINVAL;
+		} else if (val >= cluster->cl_idle_timeout_ms) {
+			mlog(ML_NOTICE, "o2net: keepalive delay must be "
+			     "smaller than idle timeout\n");
+			ret = -EINVAL;
+		} else {
+			cluster->cl_keepalive_delay_ms = val;
+		}
+	}
+
+	return ret;
+}
+
+static ssize_t o2nm_cluster_attr_reconnect_delay_ms_read(
+	struct o2nm_cluster *cluster, char *page)
+{
+	return sprintf(page, "%u\n", cluster->cl_reconnect_delay_ms);
+}
+
+static ssize_t o2nm_cluster_attr_reconnect_delay_ms_write(
+	struct o2nm_cluster *cluster, const char *page, size_t count)
+{
+	return o2nm_cluster_attr_write(page, count,
+	                               &cluster->cl_reconnect_delay_ms);
+}
+
+static ssize_t o2nm_cluster_attr_fence_method_read(
+	struct o2nm_cluster *cluster, char *page)
+{
+	ssize_t ret = 0;
+
+	if (cluster)
+		ret = sprintf(page, "%s\n",
+			      o2nm_fence_method_desc[cluster->cl_fence_method]);
+	return ret;
+}
+
+static ssize_t o2nm_cluster_attr_fence_method_write(
+	struct o2nm_cluster *cluster, const char *page, size_t count)
+{
+	unsigned int i;
+
+	if (page[count - 1] != '\n')
+		goto bail;
+
+	for (i = 0; i < O2NM_FENCE_METHODS; ++i) {
+		if (count != strlen(o2nm_fence_method_desc[i]) + 1)
+			continue;
+		if (strncasecmp(page, o2nm_fence_method_desc[i], count - 1))
+			continue;
+		if (cluster->cl_fence_method != i) {
+			printk(KERN_INFO "ocfs2: Changing fence method to %s\n",
+			       o2nm_fence_method_desc[i]);
+			cluster->cl_fence_method = i;
+		}
+		return count;
+	}
+
+bail:
+	return -EINVAL;
+}
+
+static struct o2nm_cluster_attribute o2nm_cluster_attr_idle_timeout_ms = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "idle_timeout_ms",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2nm_cluster_attr_idle_timeout_ms_read,
+	.store	= o2nm_cluster_attr_idle_timeout_ms_write,
+};
+
+static struct o2nm_cluster_attribute o2nm_cluster_attr_keepalive_delay_ms = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "keepalive_delay_ms",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2nm_cluster_attr_keepalive_delay_ms_read,
+	.store	= o2nm_cluster_attr_keepalive_delay_ms_write,
+};
+
+static struct o2nm_cluster_attribute o2nm_cluster_attr_reconnect_delay_ms = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "reconnect_delay_ms",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2nm_cluster_attr_reconnect_delay_ms_read,
+	.store	= o2nm_cluster_attr_reconnect_delay_ms_write,
+};
+
+static struct o2nm_cluster_attribute o2nm_cluster_attr_fence_method = {
+	.attr	= { .ca_owner = THIS_MODULE,
+		    .ca_name = "fence_method",
+		    .ca_mode = S_IRUGO | S_IWUSR },
+	.show	= o2nm_cluster_attr_fence_method_read,
+	.store	= o2nm_cluster_attr_fence_method_write,
+};
+
+static struct configfs_attribute *o2nm_cluster_attrs[] = {
+	&o2nm_cluster_attr_idle_timeout_ms.attr,
+	&o2nm_cluster_attr_keepalive_delay_ms.attr,
+	&o2nm_cluster_attr_reconnect_delay_ms.attr,
+	&o2nm_cluster_attr_fence_method.attr,
+	NULL,
+};
+static ssize_t o2nm_cluster_show(struct config_item *item,
+                                 struct configfs_attribute *attr,
+                                 char *page)
+{
+	struct o2nm_cluster *cluster = to_o2nm_cluster(item);
+	struct o2nm_cluster_attribute *o2nm_cluster_attr =
+		container_of(attr, struct o2nm_cluster_attribute, attr);
+	ssize_t ret = 0;
+
+	if (o2nm_cluster_attr->show)
+		ret = o2nm_cluster_attr->show(cluster, page);
+	return ret;
+}
+
+static ssize_t o2nm_cluster_store(struct config_item *item,
+                                  struct configfs_attribute *attr,
+                                  const char *page, size_t count)
+{
+	struct o2nm_cluster *cluster = to_o2nm_cluster(item);
+	struct o2nm_cluster_attribute *o2nm_cluster_attr =
+		container_of(attr, struct o2nm_cluster_attribute, attr);
+	ssize_t ret;
+
+	if (o2nm_cluster_attr->store == NULL) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = o2nm_cluster_attr->store(cluster, page, count);
+	if (ret < count)
+		goto out;
+out:
+	return ret;
+}
+
+static struct config_item *o2nm_node_group_make_item(struct config_group *group,
+						     const char *name)
+{
+	struct o2nm_node *node = NULL;
+
+	if (strlen(name) > O2NM_MAX_NAME_LEN)
+		return ERR_PTR(-ENAMETOOLONG);
+
+	node = kzalloc(sizeof(struct o2nm_node), GFP_KERNEL);
+	if (node == NULL)
+		return ERR_PTR(-ENOMEM);
+
+	strcpy(node->nd_name, name); /* use item.ci_namebuf instead? */
+	config_item_init_type_name(&node->nd_item, name, &o2nm_node_type);
+	spin_lock_init(&node->nd_lock);
+
+	mlog(ML_CLUSTER, "o2nm: Registering node %s\n", name);
+
+	return &node->nd_item;
+}
+
+static void o2nm_node_group_drop_item(struct config_group *group,
+				      struct config_item *item)
+{
+	struct o2nm_node *node = to_o2nm_node(item);
+	struct o2nm_cluster *cluster = to_o2nm_cluster(group->cg_item.ci_parent);
+
+	o2net_disconnect_node(node);
+
+	if (cluster->cl_has_local &&
+	    (cluster->cl_local_node == node->nd_num)) {
+		cluster->cl_has_local = 0;
+		cluster->cl_local_node = O2NM_INVALID_NODE_NUM;
+		o2net_stop_listening(node);
+	}
+
+	/* XXX call into net to stop this node from trading messages */
+
+	write_lock(&cluster->cl_nodes_lock);
+
+	/* XXX sloppy */
+	if (node->nd_ipv4_address)
+		rb_erase(&node->nd_ip_node, &cluster->cl_node_ip_tree);
+
+	/* nd_num might be 0 if the node number hasn't been set.. */
+	if (cluster->cl_nodes[node->nd_num] == node) {
+		cluster->cl_nodes[node->nd_num] = NULL;
+		clear_bit(node->nd_num, cluster->cl_nodes_bitmap);
+	}
+	write_unlock(&cluster->cl_nodes_lock);
+
+	mlog(ML_CLUSTER, "o2nm: Unregistered node %s\n",
+	     config_item_name(&node->nd_item));
+
+	config_item_put(item);
+}
+
+static struct configfs_group_operations o2nm_node_group_group_ops = {
+	.make_item	= o2nm_node_group_make_item,
+	.drop_item	= o2nm_node_group_drop_item,
+};
+
+static struct config_item_type o2nm_node_group_type = {
+	.ct_group_ops	= &o2nm_node_group_group_ops,
+	.ct_owner	= THIS_MODULE,
+};
+
+/* cluster */
+
+static void o2nm_cluster_release(struct config_item *item)
+{
+	struct o2nm_cluster *cluster = to_o2nm_cluster(item);
+
+	kfree(cluster->cl_group.default_groups);
+	kfree(cluster);
+}
+
+static struct configfs_item_operations o2nm_cluster_item_ops = {
+	.release	= o2nm_cluster_release,
+	.show_attribute		= o2nm_cluster_show,
+	.store_attribute	= o2nm_cluster_store,
+};
+
+static struct config_item_type o2nm_cluster_type = {
+	.ct_item_ops	= &o2nm_cluster_item_ops,
+	.ct_attrs	= o2nm_cluster_attrs,
+	.ct_owner	= THIS_MODULE,
+};
+
+/* cluster set */
+
+struct o2nm_cluster_group {
+	struct configfs_subsystem cs_subsys;
+	/* some stuff? */
+};
+
+#if 0
+static struct o2nm_cluster_group *to_o2nm_cluster_group(struct config_group *group)
+{
+	return group ?
+		container_of(to_configfs_subsystem(group), struct o2nm_cluster_group, cs_subsys)
+	       : NULL;
+}
+#endif
+
+static struct config_group *o2nm_cluster_group_make_group(struct config_group *group,
+							  const char *name)
+{
+	struct o2nm_cluster *cluster = NULL;
+	struct o2nm_node_group *ns = NULL;
+	struct config_group *o2hb_group = NULL, *ret = NULL;
+	void *defs = NULL;
+
+	/* this runs under the parent dir's i_mutex; there can be only
+	 * one caller in here at a time */
+	if (o2nm_single_cluster)
+		return ERR_PTR(-ENOSPC);
+
+	cluster = kzalloc(sizeof(struct o2nm_cluster), GFP_KERNEL);
+	ns = kzalloc(sizeof(struct o2nm_node_group), GFP_KERNEL);
+	defs = kcalloc(3, sizeof(struct config_group *), GFP_KERNEL);
+	o2hb_group = o2hb_alloc_hb_set();
+	if (cluster == NULL || ns == NULL || o2hb_group == NULL || defs == NULL)
+		goto out;
+
+	config_group_init_type_name(&cluster->cl_group, name,
+				    &o2nm_cluster_type);
+	config_group_init_type_name(&ns->ns_group, "node",
+				    &o2nm_node_group_type);
+
+	cluster->cl_group.default_groups = defs;
+	cluster->cl_group.default_groups[0] = &ns->ns_group;
+	cluster->cl_group.default_groups[1] = o2hb_group;
+	cluster->cl_group.default_groups[2] = NULL;
+	rwlock_init(&cluster->cl_nodes_lock);
+	cluster->cl_node_ip_tree = RB_ROOT;
+	cluster->cl_reconnect_delay_ms = O2NET_RECONNECT_DELAY_MS_DEFAULT;
+	cluster->cl_idle_timeout_ms    = O2NET_IDLE_TIMEOUT_MS_DEFAULT;
+	cluster->cl_keepalive_delay_ms = O2NET_KEEPALIVE_DELAY_MS_DEFAULT;
+	cluster->cl_fence_method       = O2NM_FENCE_RESET;
+
+	ret = &cluster->cl_group;
+	o2nm_single_cluster = cluster;
+
+out:
+	if (ret == NULL) {
+		kfree(cluster);
+		kfree(ns);
+		o2hb_free_hb_set(o2hb_group);
+		kfree(defs);
+		ret = ERR_PTR(-ENOMEM);
+	}
+
+	return ret;
+}
+
+static void o2nm_cluster_group_drop_item(struct config_group *group, struct config_item *item)
+{
+	struct o2nm_cluster *cluster = to_o2nm_cluster(item);
+	int i;
+	struct config_item *killme;
+
+	BUG_ON(o2nm_single_cluster != cluster);
+	o2nm_single_cluster = NULL;
+
+	for (i = 0; cluster->cl_group.default_groups[i]; i++) {
+		killme = &cluster->cl_group.default_groups[i]->cg_item;
+		cluster->cl_group.default_groups[i] = NULL;
+		config_item_put(killme);
+	}
+
+	config_item_put(item);
+}
+
+static struct configfs_group_operations o2nm_cluster_group_group_ops = {
+	.make_group	= o2nm_cluster_group_make_group,
+	.drop_item	= o2nm_cluster_group_drop_item,
+};
+
+static struct config_item_type o2nm_cluster_group_type = {
+	.ct_group_ops	= &o2nm_cluster_group_group_ops,
+	.ct_owner	= THIS_MODULE,
+};
+
+static struct o2nm_cluster_group o2nm_cluster_group = {
+	.cs_subsys = {
+		.su_group = {
+			.cg_item = {
+				.ci_namebuf = "cluster",
+				.ci_type = &o2nm_cluster_group_type,
+			},
+		},
+	},
+};
+
+int o2nm_depend_item(struct config_item *item)
+{
+	return configfs_depend_item(&o2nm_cluster_group.cs_subsys, item);
+}
+
+void o2nm_undepend_item(struct config_item *item)
+{
+	configfs_undepend_item(&o2nm_cluster_group.cs_subsys, item);
+}
+
+int o2nm_depend_this_node(void)
+{
+	int ret = 0;
+	struct o2nm_node *local_node;
+
+	local_node = o2nm_get_node_by_num(o2nm_this_node());
+	if (!local_node) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = o2nm_depend_item(&local_node->nd_item);
+	o2nm_node_put(local_node);
+
+out:
+	return ret;
+}
+
+void o2nm_undepend_this_node(void)
+{
+	struct o2nm_node *local_node;
+
+	local_node = o2nm_get_node_by_num(o2nm_this_node());
+	BUG_ON(!local_node);
+
+	o2nm_undepend_item(&local_node->nd_item);
+	o2nm_node_put(local_node);
+}
+
+
+static void __exit exit_o2nm(void)
+{
+	/* XXX sync with hb callbacks and shut down hb? */
+	o2net_unregister_hb_callbacks();
+	configfs_unregister_subsystem(&o2nm_cluster_group.cs_subsys);
+	o2cb_sys_shutdown();
+
+	o2net_exit();
+	o2hb_exit();
+}
+
+static int __init init_o2nm(void)
+{
+	int ret = -1;
+
+	ret = o2hb_init();
+	if (ret)
+		goto out;
+
+	ret = o2net_init();
+	if (ret)
+		goto out_o2hb;
+
+	ret = o2net_register_hb_callbacks();
+	if (ret)
+		goto out_o2net;
+
+	config_group_init(&o2nm_cluster_group.cs_subsys.su_group);
+	mutex_init(&o2nm_cluster_group.cs_subsys.su_mutex);
+	ret = configfs_register_subsystem(&o2nm_cluster_group.cs_subsys);
+	if (ret) {
+		printk(KERN_ERR "nodemanager: Registration returned %d\n", ret);
+		goto out_callbacks;
+	}
+
+	ret = o2cb_sys_init();
+	if (!ret)
+		goto out;
+
+	configfs_unregister_subsystem(&o2nm_cluster_group.cs_subsys);
+out_callbacks:
+	o2net_unregister_hb_callbacks();
+out_o2net:
+	o2net_exit();
+out_o2hb:
+	o2hb_exit();
+out:
+	return ret;
+}
+
+MODULE_AUTHOR("Oracle");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("OCFS2 cluster management");
+
+module_init(init_o2nm)
+module_exit(exit_o2nm)
diff --git a/fs/ocfs2/cluster/nodemanager.h b/fs/ocfs2/cluster/nodemanager.h
new file mode 100644
index 000000000..09ea2d388
--- /dev/null
+++ b/fs/ocfs2/cluster/nodemanager.h
@@ -0,0 +1,88 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * nodemanager.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef O2CLUSTER_NODEMANAGER_H
+#define O2CLUSTER_NODEMANAGER_H
+
+#include "ocfs2_nodemanager.h"
+
+/* This totally doesn't belong here. */
+#include <linux/configfs.h>
+#include <linux/rbtree.h>
+
+enum o2nm_fence_method {
+	O2NM_FENCE_RESET	= 0,
+	O2NM_FENCE_PANIC,
+	O2NM_FENCE_METHODS,	/* Number of fence methods */
+};
+
+struct o2nm_node {
+	spinlock_t		nd_lock;
+	struct config_item	nd_item;
+	char			nd_name[O2NM_MAX_NAME_LEN+1]; /* replace? */
+	__u8			nd_num;
+	/* only one address per node, as attributes, for now. */
+	__be32			nd_ipv4_address;
+	__be16			nd_ipv4_port;
+	struct rb_node		nd_ip_node;
+	/* there can be only one local node for now */
+	int			nd_local;
+
+	unsigned long		nd_set_attributes;
+};
+
+struct o2nm_cluster {
+	struct config_group	cl_group;
+	unsigned		cl_has_local:1;
+	u8			cl_local_node;
+	rwlock_t		cl_nodes_lock;
+	struct o2nm_node  	*cl_nodes[O2NM_MAX_NODES];
+	struct rb_root		cl_node_ip_tree;
+	unsigned int		cl_idle_timeout_ms;
+	unsigned int		cl_keepalive_delay_ms;
+	unsigned int		cl_reconnect_delay_ms;
+	enum o2nm_fence_method	cl_fence_method;
+
+	/* this bitmap is part of a hack for disk bitmap.. will go eventually. - zab */
+	unsigned long	cl_nodes_bitmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+};
+
+extern struct o2nm_cluster *o2nm_single_cluster;
+
+u8 o2nm_this_node(void);
+
+int o2nm_configured_node_map(unsigned long *map, unsigned bytes);
+struct o2nm_node *o2nm_get_node_by_num(u8 node_num);
+struct o2nm_node *o2nm_get_node_by_ip(__be32 addr);
+void o2nm_node_get(struct o2nm_node *node);
+void o2nm_node_put(struct o2nm_node *node);
+
+int o2nm_depend_item(struct config_item *item);
+void o2nm_undepend_item(struct config_item *item);
+int o2nm_depend_this_node(void);
+void o2nm_undepend_this_node(void);
+
+#endif /* O2CLUSTER_NODEMANAGER_H */
diff --git a/fs/ocfs2/cluster/ocfs2_heartbeat.h b/fs/ocfs2/cluster/ocfs2_heartbeat.h
new file mode 100644
index 000000000..3f4151da9
--- /dev/null
+++ b/fs/ocfs2/cluster/ocfs2_heartbeat.h
@@ -0,0 +1,38 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ocfs2_heartbeat.h
+ *
+ * On-disk structures for ocfs2_heartbeat
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef _OCFS2_HEARTBEAT_H
+#define _OCFS2_HEARTBEAT_H
+
+struct o2hb_disk_heartbeat_block {
+	__le64 hb_seq;
+	__u8  hb_node;
+	__u8  hb_pad1[3];
+	__le32 hb_cksum;
+	__le64 hb_generation;
+	__le32 hb_dead_ms;
+};
+
+#endif /* _OCFS2_HEARTBEAT_H */
diff --git a/fs/ocfs2/cluster/ocfs2_nodemanager.h b/fs/ocfs2/cluster/ocfs2_nodemanager.h
new file mode 100644
index 000000000..49b594325
--- /dev/null
+++ b/fs/ocfs2/cluster/ocfs2_nodemanager.h
@@ -0,0 +1,45 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ocfs2_nodemanager.h
+ *
+ * Header describing the interface between userspace and the kernel
+ * for the ocfs2_nodemanager module.
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef _OCFS2_NODEMANAGER_H
+#define _OCFS2_NODEMANAGER_H
+
+#define O2NM_API_VERSION	5
+
+#define O2NM_MAX_NODES		255
+#define O2NM_INVALID_NODE_NUM	255
+
+/* host name, group name, cluster name all 64 bytes */
+#define O2NM_MAX_NAME_LEN        64    // __NEW_UTS_LEN
+
+/*
+ * Maximum number of global heartbeat regions allowed.
+ * **CAUTION**  Changing this number will break dlm compatibility.
+ */
+#define O2NM_MAX_REGIONS	32
+
+#endif /* _OCFS2_NODEMANAGER_H */
diff --git a/fs/ocfs2/cluster/quorum.c b/fs/ocfs2/cluster/quorum.c
new file mode 100644
index 000000000..62e8ec619
--- /dev/null
+++ b/fs/ocfs2/cluster/quorum.c
@@ -0,0 +1,340 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ *
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+/* This quorum hack is only here until we transition to some more rational
+ * approach that is driven from userspace.  Honest.  No foolin'.
+ *
+ * Imagine two nodes lose network connectivity to each other but they're still
+ * up and operating in every other way.  Presumably a network timeout indicates
+ * that a node is broken and should be recovered.  They can't both recover each
+ * other and both carry on without serialising their access to the file system.
+ * They need to decide who is authoritative.  Now extend that problem to
+ * arbitrary groups of nodes losing connectivity between each other.
+ *
+ * So we declare that a node which has given up on connecting to a majority
+ * of nodes who are still heartbeating will fence itself.
+ *
+ * There are huge opportunities for races here.  After we give up on a node's
+ * connection we need to wait long enough to give heartbeat an opportunity
+ * to declare the node as truly dead.  We also need to be careful with the
+ * race between when we see a node start heartbeating and when we connect
+ * to it.
+ *
+ * So nodes that are in this transtion put a hold on the quorum decision
+ * with a counter.  As they fall out of this transition they drop the count
+ * and if they're the last, they fire off the decision.
+ */
+#include <linux/kernel.h>
+#include <linux/workqueue.h>
+#include <linux/reboot.h>
+
+#include "heartbeat.h"
+#include "nodemanager.h"
+#define MLOG_MASK_PREFIX ML_QUORUM
+#include "masklog.h"
+#include "quorum.h"
+
+static struct o2quo_state {
+	spinlock_t		qs_lock;
+	struct work_struct	qs_work;
+	int			qs_pending;
+	int			qs_heartbeating;
+	unsigned long		qs_hb_bm[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	int			qs_connected;
+	unsigned long		qs_conn_bm[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	int			qs_holds;
+	unsigned long		qs_hold_bm[BITS_TO_LONGS(O2NM_MAX_NODES)];
+} o2quo_state;
+
+/* this is horribly heavy-handed.  It should instead flip the file
+ * system RO and call some userspace script. */
+static void o2quo_fence_self(void)
+{
+	/* panic spins with interrupts enabled.  with preempt
+	 * threads can still schedule, etc, etc */
+	o2hb_stop_all_regions();
+
+	switch (o2nm_single_cluster->cl_fence_method) {
+	case O2NM_FENCE_PANIC:
+		panic("*** ocfs2 is very sorry to be fencing this system by "
+		      "panicing ***\n");
+		break;
+	default:
+		WARN_ON(o2nm_single_cluster->cl_fence_method >=
+			O2NM_FENCE_METHODS);
+	case O2NM_FENCE_RESET:
+		printk(KERN_ERR "*** ocfs2 is very sorry to be fencing this "
+		       "system by restarting ***\n");
+		emergency_restart();
+		break;
+	};
+}
+
+/* Indicate that a timeout occurred on a hearbeat region write. The
+ * other nodes in the cluster may consider us dead at that time so we
+ * want to "fence" ourselves so that we don't scribble on the disk
+ * after they think they've recovered us. This can't solve all
+ * problems related to writeout after recovery but this hack can at
+ * least close some of those gaps. When we have real fencing, this can
+ * go away as our node would be fenced externally before other nodes
+ * begin recovery. */
+void o2quo_disk_timeout(void)
+{
+	o2quo_fence_self();
+}
+
+static void o2quo_make_decision(struct work_struct *work)
+{
+	int quorum;
+	int lowest_hb, lowest_reachable = 0, fence = 0;
+	struct o2quo_state *qs = &o2quo_state;
+
+	spin_lock(&qs->qs_lock);
+
+	lowest_hb = find_first_bit(qs->qs_hb_bm, O2NM_MAX_NODES);
+	if (lowest_hb != O2NM_MAX_NODES)
+		lowest_reachable = test_bit(lowest_hb, qs->qs_conn_bm);
+
+	mlog(0, "heartbeating: %d, connected: %d, "
+	     "lowest: %d (%sreachable)\n", qs->qs_heartbeating,
+	     qs->qs_connected, lowest_hb, lowest_reachable ? "" : "un");
+
+	if (!test_bit(o2nm_this_node(), qs->qs_hb_bm) ||
+	    qs->qs_heartbeating == 1)
+		goto out;
+
+	if (qs->qs_heartbeating & 1) {
+		/* the odd numbered cluster case is straight forward --
+		 * if we can't talk to the majority we're hosed */
+		quorum = (qs->qs_heartbeating + 1)/2;
+		if (qs->qs_connected < quorum) {
+			mlog(ML_ERROR, "fencing this node because it is "
+			     "only connected to %u nodes and %u is needed "
+			     "to make a quorum out of %u heartbeating nodes\n",
+			     qs->qs_connected, quorum,
+			     qs->qs_heartbeating);
+			fence = 1;
+		}
+	} else {
+		/* the even numbered cluster adds the possibility of each half
+		 * of the cluster being able to talk amongst themselves.. in
+		 * that case we're hosed if we can't talk to the group that has
+		 * the lowest numbered node */
+		quorum = qs->qs_heartbeating / 2;
+		if (qs->qs_connected < quorum) {
+			mlog(ML_ERROR, "fencing this node because it is "
+			     "only connected to %u nodes and %u is needed "
+			     "to make a quorum out of %u heartbeating nodes\n",
+			     qs->qs_connected, quorum,
+			     qs->qs_heartbeating);
+			fence = 1;
+		}
+		else if ((qs->qs_connected == quorum) &&
+			 !lowest_reachable) {
+			mlog(ML_ERROR, "fencing this node because it is "
+			     "connected to a half-quorum of %u out of %u "
+			     "nodes which doesn't include the lowest active "
+			     "node %u\n", quorum, qs->qs_heartbeating,
+			     lowest_hb);
+			fence = 1;
+		}
+	}
+
+out:
+	if (fence) {
+		spin_unlock(&qs->qs_lock);
+		o2quo_fence_self();
+	} else {
+		mlog(ML_NOTICE, "not fencing this node, heartbeating: %d, "
+			"connected: %d, lowest: %d (%sreachable)\n",
+			qs->qs_heartbeating, qs->qs_connected, lowest_hb,
+			lowest_reachable ? "" : "un");
+		spin_unlock(&qs->qs_lock);
+
+	}
+
+}
+
+static void o2quo_set_hold(struct o2quo_state *qs, u8 node)
+{
+	assert_spin_locked(&qs->qs_lock);
+
+	if (!test_and_set_bit(node, qs->qs_hold_bm)) {
+		qs->qs_holds++;
+		mlog_bug_on_msg(qs->qs_holds == O2NM_MAX_NODES,
+			        "node %u\n", node);
+		mlog(0, "node %u, %d total\n", node, qs->qs_holds);
+	}
+}
+
+static void o2quo_clear_hold(struct o2quo_state *qs, u8 node)
+{
+	assert_spin_locked(&qs->qs_lock);
+
+	if (test_and_clear_bit(node, qs->qs_hold_bm)) {
+		mlog(0, "node %u, %d total\n", node, qs->qs_holds - 1);
+		if (--qs->qs_holds == 0) {
+			if (qs->qs_pending) {
+				qs->qs_pending = 0;
+				schedule_work(&qs->qs_work);
+			}
+		}
+		mlog_bug_on_msg(qs->qs_holds < 0, "node %u, holds %d\n",
+				node, qs->qs_holds);
+	}
+}
+
+/* as a node comes up we delay the quorum decision until we know the fate of
+ * the connection.  the hold will be droped in conn_up or hb_down.  it might be
+ * perpetuated by con_err until hb_down.  if we already have a conn, we might
+ * be dropping a hold that conn_up got. */
+void o2quo_hb_up(u8 node)
+{
+	struct o2quo_state *qs = &o2quo_state;
+
+	spin_lock(&qs->qs_lock);
+
+	qs->qs_heartbeating++;
+	mlog_bug_on_msg(qs->qs_heartbeating == O2NM_MAX_NODES,
+		        "node %u\n", node);
+	mlog_bug_on_msg(test_bit(node, qs->qs_hb_bm), "node %u\n", node);
+	set_bit(node, qs->qs_hb_bm);
+
+	mlog(0, "node %u, %d total\n", node, qs->qs_heartbeating);
+
+	if (!test_bit(node, qs->qs_conn_bm))
+		o2quo_set_hold(qs, node);
+	else
+		o2quo_clear_hold(qs, node);
+
+	spin_unlock(&qs->qs_lock);
+}
+
+/* hb going down releases any holds we might have had due to this node from
+ * conn_up, conn_err, or hb_up */
+void o2quo_hb_down(u8 node)
+{
+	struct o2quo_state *qs = &o2quo_state;
+
+	spin_lock(&qs->qs_lock);
+
+	qs->qs_heartbeating--;
+	mlog_bug_on_msg(qs->qs_heartbeating < 0,
+			"node %u, %d heartbeating\n",
+			node, qs->qs_heartbeating);
+	mlog_bug_on_msg(!test_bit(node, qs->qs_hb_bm), "node %u\n", node);
+	clear_bit(node, qs->qs_hb_bm);
+
+	mlog(0, "node %u, %d total\n", node, qs->qs_heartbeating);
+
+	o2quo_clear_hold(qs, node);
+
+	spin_unlock(&qs->qs_lock);
+}
+
+/* this tells us that we've decided that the node is still heartbeating
+ * even though we've lost it's conn.  it must only be called after conn_err
+ * and indicates that we must now make a quorum decision in the future,
+ * though we might be doing so after waiting for holds to drain.  Here
+ * we'll be dropping the hold from conn_err. */
+void o2quo_hb_still_up(u8 node)
+{
+	struct o2quo_state *qs = &o2quo_state;
+
+	spin_lock(&qs->qs_lock);
+
+	mlog(0, "node %u\n", node);
+
+	qs->qs_pending = 1;
+	o2quo_clear_hold(qs, node);
+
+	spin_unlock(&qs->qs_lock);
+}
+
+/* This is analogous to hb_up.  as a node's connection comes up we delay the
+ * quorum decision until we see it heartbeating.  the hold will be droped in
+ * hb_up or hb_down.  it might be perpetuated by con_err until hb_down.  if
+ * it's already heartbeating we might be dropping a hold that conn_up got.
+ * */
+void o2quo_conn_up(u8 node)
+{
+	struct o2quo_state *qs = &o2quo_state;
+
+	spin_lock(&qs->qs_lock);
+
+	qs->qs_connected++;
+	mlog_bug_on_msg(qs->qs_connected == O2NM_MAX_NODES,
+		        "node %u\n", node);
+	mlog_bug_on_msg(test_bit(node, qs->qs_conn_bm), "node %u\n", node);
+	set_bit(node, qs->qs_conn_bm);
+
+	mlog(0, "node %u, %d total\n", node, qs->qs_connected);
+
+	if (!test_bit(node, qs->qs_hb_bm))
+		o2quo_set_hold(qs, node);
+	else
+		o2quo_clear_hold(qs, node);
+
+	spin_unlock(&qs->qs_lock);
+}
+
+/* we've decided that we won't ever be connecting to the node again.  if it's
+ * still heartbeating we grab a hold that will delay decisions until either the
+ * node stops heartbeating from hb_down or the caller decides that the node is
+ * still up and calls still_up */
+void o2quo_conn_err(u8 node)
+{
+	struct o2quo_state *qs = &o2quo_state;
+
+	spin_lock(&qs->qs_lock);
+
+	if (test_bit(node, qs->qs_conn_bm)) {
+		qs->qs_connected--;
+		mlog_bug_on_msg(qs->qs_connected < 0,
+				"node %u, connected %d\n",
+				node, qs->qs_connected);
+
+		clear_bit(node, qs->qs_conn_bm);
+	}
+
+	mlog(0, "node %u, %d total\n", node, qs->qs_connected);
+
+	if (test_bit(node, qs->qs_hb_bm))
+		o2quo_set_hold(qs, node);
+
+	spin_unlock(&qs->qs_lock);
+}
+
+void o2quo_init(void)
+{
+	struct o2quo_state *qs = &o2quo_state;
+
+	spin_lock_init(&qs->qs_lock);
+	INIT_WORK(&qs->qs_work, o2quo_make_decision);
+}
+
+void o2quo_exit(void)
+{
+	struct o2quo_state *qs = &o2quo_state;
+
+	flush_work(&qs->qs_work);
+}
diff --git a/fs/ocfs2/cluster/quorum.h b/fs/ocfs2/cluster/quorum.h
new file mode 100644
index 000000000..6649cc6f6
--- /dev/null
+++ b/fs/ocfs2/cluster/quorum.h
@@ -0,0 +1,36 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef O2CLUSTER_QUORUM_H
+#define O2CLUSTER_QUORUM_H
+
+void o2quo_init(void);
+void o2quo_exit(void);
+
+void o2quo_hb_up(u8 node);
+void o2quo_hb_down(u8 node);
+void o2quo_hb_still_up(u8 node);
+void o2quo_conn_up(u8 node);
+void o2quo_conn_err(u8 node);
+void o2quo_disk_timeout(void);
+
+#endif /* O2CLUSTER_QUORUM_H */
diff --git a/fs/ocfs2/cluster/sys.c b/fs/ocfs2/cluster/sys.c
new file mode 100644
index 000000000..b7f57271d
--- /dev/null
+++ b/fs/ocfs2/cluster/sys.c
@@ -0,0 +1,82 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * sys.c
+ *
+ * OCFS2 cluster sysfs interface
+ *
+ * Copyright (C) 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation,
+ * version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/kobject.h>
+#include <linux/sysfs.h>
+#include <linux/fs.h>
+
+#include "ocfs2_nodemanager.h"
+#include "masklog.h"
+#include "sys.h"
+
+
+static ssize_t version_show(struct kobject *kobj, struct kobj_attribute *attr,
+			    char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "%u\n", O2NM_API_VERSION);
+}
+static struct kobj_attribute attr_version =
+	__ATTR(interface_revision, S_IRUGO, version_show, NULL);
+
+static struct attribute *o2cb_attrs[] = {
+	&attr_version.attr,
+	NULL,
+};
+
+static struct attribute_group o2cb_attr_group = {
+	.attrs = o2cb_attrs,
+};
+
+static struct kset *o2cb_kset;
+
+void o2cb_sys_shutdown(void)
+{
+	mlog_sys_shutdown();
+	kset_unregister(o2cb_kset);
+}
+
+int o2cb_sys_init(void)
+{
+	int ret;
+
+	o2cb_kset = kset_create_and_add("o2cb", NULL, fs_kobj);
+	if (!o2cb_kset)
+		return -ENOMEM;
+
+	ret = sysfs_create_group(&o2cb_kset->kobj, &o2cb_attr_group);
+	if (ret)
+		goto error;
+
+	ret = mlog_sys_init(o2cb_kset);
+	if (ret)
+		goto error;
+	return 0;
+error:
+	kset_unregister(o2cb_kset);
+	return ret;
+}
diff --git a/fs/ocfs2/cluster/sys.h b/fs/ocfs2/cluster/sys.h
new file mode 100644
index 000000000..d66b8ab00
--- /dev/null
+++ b/fs/ocfs2/cluster/sys.h
@@ -0,0 +1,33 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * sys.h
+ *
+ * Function prototypes for o2cb sysfs interface
+ *
+ * Copyright (C) 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation,
+ * version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef O2CLUSTER_SYS_H
+#define O2CLUSTER_SYS_H
+
+void o2cb_sys_shutdown(void);
+int o2cb_sys_init(void);
+
+#endif /* O2CLUSTER_SYS_H */
diff --git a/fs/ocfs2/cluster/tcp.c b/fs/ocfs2/cluster/tcp.c
new file mode 100644
index 000000000..56c403a56
--- /dev/null
+++ b/fs/ocfs2/cluster/tcp.c
@@ -0,0 +1,2219 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ *
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ * ----
+ *
+ * Callers for this were originally written against a very simple synchronus
+ * API.  This implementation reflects those simple callers.  Some day I'm sure
+ * we'll need to move to a more robust posting/callback mechanism.
+ *
+ * Transmit calls pass in kernel virtual addresses and block copying this into
+ * the socket's tx buffers via a usual blocking sendmsg.  They'll block waiting
+ * for a failed socket to timeout.  TX callers can also pass in a poniter to an
+ * 'int' which gets filled with an errno off the wire in response to the
+ * message they send.
+ *
+ * Handlers for unsolicited messages are registered.  Each socket has a page
+ * that incoming data is copied into.  First the header, then the data.
+ * Handlers are called from only one thread with a reference to this per-socket
+ * page.  This page is destroyed after the handler call, so it can't be
+ * referenced beyond the call.  Handlers may block but are discouraged from
+ * doing so.
+ *
+ * Any framing errors (bad magic, large payload lengths) close a connection.
+ *
+ * Our sock_container holds the state we associate with a socket.  It's current
+ * framing state is held there as well as the refcounting we do around when it
+ * is safe to tear down the socket.  The socket is only finally torn down from
+ * the container when the container loses all of its references -- so as long
+ * as you hold a ref on the container you can trust that the socket is valid
+ * for use with kernel socket APIs.
+ *
+ * Connections are initiated between a pair of nodes when the node with the
+ * higher node number gets a heartbeat callback which indicates that the lower
+ * numbered node has started heartbeating.  The lower numbered node is passive
+ * and only accepts the connection if the higher numbered node is heartbeating.
+ */
+
+#include <linux/kernel.h>
+#include <linux/jiffies.h>
+#include <linux/slab.h>
+#include <linux/idr.h>
+#include <linux/kref.h>
+#include <linux/net.h>
+#include <linux/export.h>
+#include <net/tcp.h>
+
+#include <asm/uaccess.h>
+
+#include "heartbeat.h"
+#include "tcp.h"
+#include "nodemanager.h"
+#define MLOG_MASK_PREFIX ML_TCP
+#include "masklog.h"
+#include "quorum.h"
+
+#include "tcp_internal.h"
+
+#define SC_NODEF_FMT "node %s (num %u) at %pI4:%u"
+#define SC_NODEF_ARGS(sc) sc->sc_node->nd_name, sc->sc_node->nd_num,	\
+			  &sc->sc_node->nd_ipv4_address,		\
+			  ntohs(sc->sc_node->nd_ipv4_port)
+
+/*
+ * In the following two log macros, the whitespace after the ',' just
+ * before ##args is intentional. Otherwise, gcc 2.95 will eat the
+ * previous token if args expands to nothing.
+ */
+#define msglog(hdr, fmt, args...) do {					\
+	typeof(hdr) __hdr = (hdr);					\
+	mlog(ML_MSG, "[mag %u len %u typ %u stat %d sys_stat %d "	\
+	     "key %08x num %u] " fmt,					\
+	     be16_to_cpu(__hdr->magic), be16_to_cpu(__hdr->data_len), 	\
+	     be16_to_cpu(__hdr->msg_type), be32_to_cpu(__hdr->status),	\
+	     be32_to_cpu(__hdr->sys_status), be32_to_cpu(__hdr->key),	\
+	     be32_to_cpu(__hdr->msg_num) ,  ##args);			\
+} while (0)
+
+#define sclog(sc, fmt, args...) do {					\
+	typeof(sc) __sc = (sc);						\
+	mlog(ML_SOCKET, "[sc %p refs %d sock %p node %u page %p "	\
+	     "pg_off %zu] " fmt, __sc,					\
+	     atomic_read(&__sc->sc_kref.refcount), __sc->sc_sock,	\
+	    __sc->sc_node->nd_num, __sc->sc_page, __sc->sc_page_off ,	\
+	    ##args);							\
+} while (0)
+
+static DEFINE_RWLOCK(o2net_handler_lock);
+static struct rb_root o2net_handler_tree = RB_ROOT;
+
+static struct o2net_node o2net_nodes[O2NM_MAX_NODES];
+
+/* XXX someday we'll need better accounting */
+static struct socket *o2net_listen_sock;
+
+/*
+ * listen work is only queued by the listening socket callbacks on the
+ * o2net_wq.  teardown detaches the callbacks before destroying the workqueue.
+ * quorum work is queued as sock containers are shutdown.. stop_listening
+ * tears down all the node's sock containers, preventing future shutdowns
+ * and queued quroum work, before canceling delayed quorum work and
+ * destroying the work queue.
+ */
+static struct workqueue_struct *o2net_wq;
+static struct work_struct o2net_listen_work;
+
+static struct o2hb_callback_func o2net_hb_up, o2net_hb_down;
+#define O2NET_HB_PRI 0x1
+
+static struct o2net_handshake *o2net_hand;
+static struct o2net_msg *o2net_keep_req, *o2net_keep_resp;
+
+static int o2net_sys_err_translations[O2NET_ERR_MAX] =
+		{[O2NET_ERR_NONE]	= 0,
+		 [O2NET_ERR_NO_HNDLR]	= -ENOPROTOOPT,
+		 [O2NET_ERR_OVERFLOW]	= -EOVERFLOW,
+		 [O2NET_ERR_DIED]	= -EHOSTDOWN,};
+
+/* can't quite avoid *all* internal declarations :/ */
+static void o2net_sc_connect_completed(struct work_struct *work);
+static void o2net_rx_until_empty(struct work_struct *work);
+static void o2net_shutdown_sc(struct work_struct *work);
+static void o2net_listen_data_ready(struct sock *sk);
+static void o2net_sc_send_keep_req(struct work_struct *work);
+static void o2net_idle_timer(unsigned long data);
+static void o2net_sc_postpone_idle(struct o2net_sock_container *sc);
+static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc);
+
+#ifdef CONFIG_DEBUG_FS
+static void o2net_init_nst(struct o2net_send_tracking *nst, u32 msgtype,
+			   u32 msgkey, struct task_struct *task, u8 node)
+{
+	INIT_LIST_HEAD(&nst->st_net_debug_item);
+	nst->st_task = task;
+	nst->st_msg_type = msgtype;
+	nst->st_msg_key = msgkey;
+	nst->st_node = node;
+}
+
+static inline void o2net_set_nst_sock_time(struct o2net_send_tracking *nst)
+{
+	nst->st_sock_time = ktime_get();
+}
+
+static inline void o2net_set_nst_send_time(struct o2net_send_tracking *nst)
+{
+	nst->st_send_time = ktime_get();
+}
+
+static inline void o2net_set_nst_status_time(struct o2net_send_tracking *nst)
+{
+	nst->st_status_time = ktime_get();
+}
+
+static inline void o2net_set_nst_sock_container(struct o2net_send_tracking *nst,
+						struct o2net_sock_container *sc)
+{
+	nst->st_sc = sc;
+}
+
+static inline void o2net_set_nst_msg_id(struct o2net_send_tracking *nst,
+					u32 msg_id)
+{
+	nst->st_id = msg_id;
+}
+
+static inline void o2net_set_sock_timer(struct o2net_sock_container *sc)
+{
+	sc->sc_tv_timer = ktime_get();
+}
+
+static inline void o2net_set_data_ready_time(struct o2net_sock_container *sc)
+{
+	sc->sc_tv_data_ready = ktime_get();
+}
+
+static inline void o2net_set_advance_start_time(struct o2net_sock_container *sc)
+{
+	sc->sc_tv_advance_start = ktime_get();
+}
+
+static inline void o2net_set_advance_stop_time(struct o2net_sock_container *sc)
+{
+	sc->sc_tv_advance_stop = ktime_get();
+}
+
+static inline void o2net_set_func_start_time(struct o2net_sock_container *sc)
+{
+	sc->sc_tv_func_start = ktime_get();
+}
+
+static inline void o2net_set_func_stop_time(struct o2net_sock_container *sc)
+{
+	sc->sc_tv_func_stop = ktime_get();
+}
+
+#else  /* CONFIG_DEBUG_FS */
+# define o2net_init_nst(a, b, c, d, e)
+# define o2net_set_nst_sock_time(a)
+# define o2net_set_nst_send_time(a)
+# define o2net_set_nst_status_time(a)
+# define o2net_set_nst_sock_container(a, b)
+# define o2net_set_nst_msg_id(a, b)
+# define o2net_set_sock_timer(a)
+# define o2net_set_data_ready_time(a)
+# define o2net_set_advance_start_time(a)
+# define o2net_set_advance_stop_time(a)
+# define o2net_set_func_start_time(a)
+# define o2net_set_func_stop_time(a)
+#endif /* CONFIG_DEBUG_FS */
+
+#ifdef CONFIG_OCFS2_FS_STATS
+static ktime_t o2net_get_func_run_time(struct o2net_sock_container *sc)
+{
+	return ktime_sub(sc->sc_tv_func_stop, sc->sc_tv_func_start);
+}
+
+static void o2net_update_send_stats(struct o2net_send_tracking *nst,
+				    struct o2net_sock_container *sc)
+{
+	sc->sc_tv_status_total = ktime_add(sc->sc_tv_status_total,
+					   ktime_sub(ktime_get(),
+						     nst->st_status_time));
+	sc->sc_tv_send_total = ktime_add(sc->sc_tv_send_total,
+					 ktime_sub(nst->st_status_time,
+						   nst->st_send_time));
+	sc->sc_tv_acquiry_total = ktime_add(sc->sc_tv_acquiry_total,
+					    ktime_sub(nst->st_send_time,
+						      nst->st_sock_time));
+	sc->sc_send_count++;
+}
+
+static void o2net_update_recv_stats(struct o2net_sock_container *sc)
+{
+	sc->sc_tv_process_total = ktime_add(sc->sc_tv_process_total,
+					    o2net_get_func_run_time(sc));
+	sc->sc_recv_count++;
+}
+
+#else
+
+# define o2net_update_send_stats(a, b)
+
+# define o2net_update_recv_stats(sc)
+
+#endif /* CONFIG_OCFS2_FS_STATS */
+
+static inline unsigned int o2net_reconnect_delay(void)
+{
+	return o2nm_single_cluster->cl_reconnect_delay_ms;
+}
+
+static inline unsigned int o2net_keepalive_delay(void)
+{
+	return o2nm_single_cluster->cl_keepalive_delay_ms;
+}
+
+static inline unsigned int o2net_idle_timeout(void)
+{
+	return o2nm_single_cluster->cl_idle_timeout_ms;
+}
+
+static inline int o2net_sys_err_to_errno(enum o2net_system_error err)
+{
+	int trans;
+	BUG_ON(err >= O2NET_ERR_MAX);
+	trans = o2net_sys_err_translations[err];
+
+	/* Just in case we mess up the translation table above */
+	BUG_ON(err != O2NET_ERR_NONE && trans == 0);
+	return trans;
+}
+
+static struct o2net_node * o2net_nn_from_num(u8 node_num)
+{
+	BUG_ON(node_num >= ARRAY_SIZE(o2net_nodes));
+	return &o2net_nodes[node_num];
+}
+
+static u8 o2net_num_from_nn(struct o2net_node *nn)
+{
+	BUG_ON(nn == NULL);
+	return nn - o2net_nodes;
+}
+
+/* ------------------------------------------------------------ */
+
+static int o2net_prep_nsw(struct o2net_node *nn, struct o2net_status_wait *nsw)
+{
+	int ret;
+
+	spin_lock(&nn->nn_lock);
+	ret = idr_alloc(&nn->nn_status_idr, nsw, 0, 0, GFP_ATOMIC);
+	if (ret >= 0) {
+		nsw->ns_id = ret;
+		list_add_tail(&nsw->ns_node_item, &nn->nn_status_list);
+	}
+	spin_unlock(&nn->nn_lock);
+	if (ret < 0)
+		return ret;
+
+	init_waitqueue_head(&nsw->ns_wq);
+	nsw->ns_sys_status = O2NET_ERR_NONE;
+	nsw->ns_status = 0;
+	return 0;
+}
+
+static void o2net_complete_nsw_locked(struct o2net_node *nn,
+				      struct o2net_status_wait *nsw,
+				      enum o2net_system_error sys_status,
+				      s32 status)
+{
+	assert_spin_locked(&nn->nn_lock);
+
+	if (!list_empty(&nsw->ns_node_item)) {
+		list_del_init(&nsw->ns_node_item);
+		nsw->ns_sys_status = sys_status;
+		nsw->ns_status = status;
+		idr_remove(&nn->nn_status_idr, nsw->ns_id);
+		wake_up(&nsw->ns_wq);
+	}
+}
+
+static void o2net_complete_nsw(struct o2net_node *nn,
+			       struct o2net_status_wait *nsw,
+			       u64 id, enum o2net_system_error sys_status,
+			       s32 status)
+{
+	spin_lock(&nn->nn_lock);
+	if (nsw == NULL) {
+		if (id > INT_MAX)
+			goto out;
+
+		nsw = idr_find(&nn->nn_status_idr, id);
+		if (nsw == NULL)
+			goto out;
+	}
+
+	o2net_complete_nsw_locked(nn, nsw, sys_status, status);
+
+out:
+	spin_unlock(&nn->nn_lock);
+	return;
+}
+
+static void o2net_complete_nodes_nsw(struct o2net_node *nn)
+{
+	struct o2net_status_wait *nsw, *tmp;
+	unsigned int num_kills = 0;
+
+	assert_spin_locked(&nn->nn_lock);
+
+	list_for_each_entry_safe(nsw, tmp, &nn->nn_status_list, ns_node_item) {
+		o2net_complete_nsw_locked(nn, nsw, O2NET_ERR_DIED, 0);
+		num_kills++;
+	}
+
+	mlog(0, "completed %d messages for node %u\n", num_kills,
+	     o2net_num_from_nn(nn));
+}
+
+static int o2net_nsw_completed(struct o2net_node *nn,
+			       struct o2net_status_wait *nsw)
+{
+	int completed;
+	spin_lock(&nn->nn_lock);
+	completed = list_empty(&nsw->ns_node_item);
+	spin_unlock(&nn->nn_lock);
+	return completed;
+}
+
+/* ------------------------------------------------------------ */
+
+static void sc_kref_release(struct kref *kref)
+{
+	struct o2net_sock_container *sc = container_of(kref,
+					struct o2net_sock_container, sc_kref);
+	BUG_ON(timer_pending(&sc->sc_idle_timeout));
+
+	sclog(sc, "releasing\n");
+
+	if (sc->sc_sock) {
+		sock_release(sc->sc_sock);
+		sc->sc_sock = NULL;
+	}
+
+	o2nm_undepend_item(&sc->sc_node->nd_item);
+	o2nm_node_put(sc->sc_node);
+	sc->sc_node = NULL;
+
+	o2net_debug_del_sc(sc);
+
+	if (sc->sc_page)
+		__free_page(sc->sc_page);
+	kfree(sc);
+}
+
+static void sc_put(struct o2net_sock_container *sc)
+{
+	sclog(sc, "put\n");
+	kref_put(&sc->sc_kref, sc_kref_release);
+}
+static void sc_get(struct o2net_sock_container *sc)
+{
+	sclog(sc, "get\n");
+	kref_get(&sc->sc_kref);
+}
+static struct o2net_sock_container *sc_alloc(struct o2nm_node *node)
+{
+	struct o2net_sock_container *sc, *ret = NULL;
+	struct page *page = NULL;
+	int status = 0;
+
+	page = alloc_page(GFP_NOFS);
+	sc = kzalloc(sizeof(*sc), GFP_NOFS);
+	if (sc == NULL || page == NULL)
+		goto out;
+
+	kref_init(&sc->sc_kref);
+	o2nm_node_get(node);
+	sc->sc_node = node;
+
+	/* pin the node item of the remote node */
+	status = o2nm_depend_item(&node->nd_item);
+	if (status) {
+		mlog_errno(status);
+		o2nm_node_put(node);
+		goto out;
+	}
+	INIT_WORK(&sc->sc_connect_work, o2net_sc_connect_completed);
+	INIT_WORK(&sc->sc_rx_work, o2net_rx_until_empty);
+	INIT_WORK(&sc->sc_shutdown_work, o2net_shutdown_sc);
+	INIT_DELAYED_WORK(&sc->sc_keepalive_work, o2net_sc_send_keep_req);
+
+	init_timer(&sc->sc_idle_timeout);
+	sc->sc_idle_timeout.function = o2net_idle_timer;
+	sc->sc_idle_timeout.data = (unsigned long)sc;
+
+	sclog(sc, "alloced\n");
+
+	ret = sc;
+	sc->sc_page = page;
+	o2net_debug_add_sc(sc);
+	sc = NULL;
+	page = NULL;
+
+out:
+	if (page)
+		__free_page(page);
+	kfree(sc);
+
+	return ret;
+}
+
+/* ------------------------------------------------------------ */
+
+static void o2net_sc_queue_work(struct o2net_sock_container *sc,
+				struct work_struct *work)
+{
+	sc_get(sc);
+	if (!queue_work(o2net_wq, work))
+		sc_put(sc);
+}
+static void o2net_sc_queue_delayed_work(struct o2net_sock_container *sc,
+					struct delayed_work *work,
+					int delay)
+{
+	sc_get(sc);
+	if (!queue_delayed_work(o2net_wq, work, delay))
+		sc_put(sc);
+}
+static void o2net_sc_cancel_delayed_work(struct o2net_sock_container *sc,
+					 struct delayed_work *work)
+{
+	if (cancel_delayed_work(work))
+		sc_put(sc);
+}
+
+static atomic_t o2net_connected_peers = ATOMIC_INIT(0);
+
+int o2net_num_connected_peers(void)
+{
+	return atomic_read(&o2net_connected_peers);
+}
+
+static void o2net_set_nn_state(struct o2net_node *nn,
+			       struct o2net_sock_container *sc,
+			       unsigned valid, int err)
+{
+	int was_valid = nn->nn_sc_valid;
+	int was_err = nn->nn_persistent_error;
+	struct o2net_sock_container *old_sc = nn->nn_sc;
+
+	assert_spin_locked(&nn->nn_lock);
+
+	if (old_sc && !sc)
+		atomic_dec(&o2net_connected_peers);
+	else if (!old_sc && sc)
+		atomic_inc(&o2net_connected_peers);
+
+	/* the node num comparison and single connect/accept path should stop
+	 * an non-null sc from being overwritten with another */
+	BUG_ON(sc && nn->nn_sc && nn->nn_sc != sc);
+	mlog_bug_on_msg(err && valid, "err %d valid %u\n", err, valid);
+	mlog_bug_on_msg(valid && !sc, "valid %u sc %p\n", valid, sc);
+
+	if (was_valid && !valid && err == 0)
+		err = -ENOTCONN;
+
+	mlog(ML_CONN, "node %u sc: %p -> %p, valid %u -> %u, err %d -> %d\n",
+	     o2net_num_from_nn(nn), nn->nn_sc, sc, nn->nn_sc_valid, valid,
+	     nn->nn_persistent_error, err);
+
+	nn->nn_sc = sc;
+	nn->nn_sc_valid = valid ? 1 : 0;
+	nn->nn_persistent_error = err;
+
+	/* mirrors o2net_tx_can_proceed() */
+	if (nn->nn_persistent_error || nn->nn_sc_valid)
+		wake_up(&nn->nn_sc_wq);
+
+	if (was_valid && !was_err && nn->nn_persistent_error) {
+		o2quo_conn_err(o2net_num_from_nn(nn));
+		queue_delayed_work(o2net_wq, &nn->nn_still_up,
+				   msecs_to_jiffies(O2NET_QUORUM_DELAY_MS));
+	}
+
+	if (was_valid && !valid) {
+		if (old_sc)
+			printk(KERN_NOTICE "o2net: No longer connected to "
+				SC_NODEF_FMT "\n", SC_NODEF_ARGS(old_sc));
+		o2net_complete_nodes_nsw(nn);
+	}
+
+	if (!was_valid && valid) {
+		o2quo_conn_up(o2net_num_from_nn(nn));
+		cancel_delayed_work(&nn->nn_connect_expired);
+		printk(KERN_NOTICE "o2net: %s " SC_NODEF_FMT "\n",
+		       o2nm_this_node() > sc->sc_node->nd_num ?
+		       "Connected to" : "Accepted connection from",
+		       SC_NODEF_ARGS(sc));
+	}
+
+	/* trigger the connecting worker func as long as we're not valid,
+	 * it will back off if it shouldn't connect.  This can be called
+	 * from node config teardown and so needs to be careful about
+	 * the work queue actually being up. */
+	if (!valid && o2net_wq) {
+		unsigned long delay;
+		/* delay if we're within a RECONNECT_DELAY of the
+		 * last attempt */
+		delay = (nn->nn_last_connect_attempt +
+			 msecs_to_jiffies(o2net_reconnect_delay()))
+			- jiffies;
+		if (delay > msecs_to_jiffies(o2net_reconnect_delay()))
+			delay = 0;
+		mlog(ML_CONN, "queueing conn attempt in %lu jiffies\n", delay);
+		queue_delayed_work(o2net_wq, &nn->nn_connect_work, delay);
+
+		/*
+		 * Delay the expired work after idle timeout.
+		 *
+		 * We might have lots of failed connection attempts that run
+		 * through here but we only cancel the connect_expired work when
+		 * a connection attempt succeeds.  So only the first enqueue of
+		 * the connect_expired work will do anything.  The rest will see
+		 * that it's already queued and do nothing.
+		 */
+		delay += msecs_to_jiffies(o2net_idle_timeout());
+		queue_delayed_work(o2net_wq, &nn->nn_connect_expired, delay);
+	}
+
+	/* keep track of the nn's sc ref for the caller */
+	if ((old_sc == NULL) && sc)
+		sc_get(sc);
+	if (old_sc && (old_sc != sc)) {
+		o2net_sc_queue_work(old_sc, &old_sc->sc_shutdown_work);
+		sc_put(old_sc);
+	}
+}
+
+/* see o2net_register_callbacks() */
+static void o2net_data_ready(struct sock *sk)
+{
+	void (*ready)(struct sock *sk);
+
+	read_lock(&sk->sk_callback_lock);
+	if (sk->sk_user_data) {
+		struct o2net_sock_container *sc = sk->sk_user_data;
+		sclog(sc, "data_ready hit\n");
+		o2net_set_data_ready_time(sc);
+		o2net_sc_queue_work(sc, &sc->sc_rx_work);
+		ready = sc->sc_data_ready;
+	} else {
+		ready = sk->sk_data_ready;
+	}
+	read_unlock(&sk->sk_callback_lock);
+
+	ready(sk);
+}
+
+/* see o2net_register_callbacks() */
+static void o2net_state_change(struct sock *sk)
+{
+	void (*state_change)(struct sock *sk);
+	struct o2net_sock_container *sc;
+
+	read_lock(&sk->sk_callback_lock);
+	sc = sk->sk_user_data;
+	if (sc == NULL) {
+		state_change = sk->sk_state_change;
+		goto out;
+	}
+
+	sclog(sc, "state_change to %d\n", sk->sk_state);
+
+	state_change = sc->sc_state_change;
+
+	switch(sk->sk_state) {
+	/* ignore connecting sockets as they make progress */
+	case TCP_SYN_SENT:
+	case TCP_SYN_RECV:
+		break;
+	case TCP_ESTABLISHED:
+		o2net_sc_queue_work(sc, &sc->sc_connect_work);
+		break;
+	default:
+		printk(KERN_INFO "o2net: Connection to " SC_NODEF_FMT
+			" shutdown, state %d\n",
+			SC_NODEF_ARGS(sc), sk->sk_state);
+		o2net_sc_queue_work(sc, &sc->sc_shutdown_work);
+		break;
+	}
+out:
+	read_unlock(&sk->sk_callback_lock);
+	state_change(sk);
+}
+
+/*
+ * we register callbacks so we can queue work on events before calling
+ * the original callbacks.  our callbacks our careful to test user_data
+ * to discover when they've reaced with o2net_unregister_callbacks().
+ */
+static void o2net_register_callbacks(struct sock *sk,
+				     struct o2net_sock_container *sc)
+{
+	write_lock_bh(&sk->sk_callback_lock);
+
+	/* accepted sockets inherit the old listen socket data ready */
+	if (sk->sk_data_ready == o2net_listen_data_ready) {
+		sk->sk_data_ready = sk->sk_user_data;
+		sk->sk_user_data = NULL;
+	}
+
+	BUG_ON(sk->sk_user_data != NULL);
+	sk->sk_user_data = sc;
+	sc_get(sc);
+
+	sc->sc_data_ready = sk->sk_data_ready;
+	sc->sc_state_change = sk->sk_state_change;
+	sk->sk_data_ready = o2net_data_ready;
+	sk->sk_state_change = o2net_state_change;
+
+	mutex_init(&sc->sc_send_lock);
+
+	write_unlock_bh(&sk->sk_callback_lock);
+}
+
+static int o2net_unregister_callbacks(struct sock *sk,
+			           struct o2net_sock_container *sc)
+{
+	int ret = 0;
+
+	write_lock_bh(&sk->sk_callback_lock);
+	if (sk->sk_user_data == sc) {
+		ret = 1;
+		sk->sk_user_data = NULL;
+		sk->sk_data_ready = sc->sc_data_ready;
+		sk->sk_state_change = sc->sc_state_change;
+	}
+	write_unlock_bh(&sk->sk_callback_lock);
+
+	return ret;
+}
+
+/*
+ * this is a little helper that is called by callers who have seen a problem
+ * with an sc and want to detach it from the nn if someone already hasn't beat
+ * them to it.  if an error is given then the shutdown will be persistent
+ * and pending transmits will be canceled.
+ */
+static void o2net_ensure_shutdown(struct o2net_node *nn,
+			           struct o2net_sock_container *sc,
+				   int err)
+{
+	spin_lock(&nn->nn_lock);
+	if (nn->nn_sc == sc)
+		o2net_set_nn_state(nn, NULL, 0, err);
+	spin_unlock(&nn->nn_lock);
+}
+
+/*
+ * This work queue function performs the blocking parts of socket shutdown.  A
+ * few paths lead here.  set_nn_state will trigger this callback if it sees an
+ * sc detached from the nn.  state_change will also trigger this callback
+ * directly when it sees errors.  In that case we need to call set_nn_state
+ * ourselves as state_change couldn't get the nn_lock and call set_nn_state
+ * itself.
+ */
+static void o2net_shutdown_sc(struct work_struct *work)
+{
+	struct o2net_sock_container *sc =
+		container_of(work, struct o2net_sock_container,
+			     sc_shutdown_work);
+	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
+
+	sclog(sc, "shutting down\n");
+
+	/* drop the callbacks ref and call shutdown only once */
+	if (o2net_unregister_callbacks(sc->sc_sock->sk, sc)) {
+		/* we shouldn't flush as we're in the thread, the
+		 * races with pending sc work structs are harmless */
+		del_timer_sync(&sc->sc_idle_timeout);
+		o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work);
+		sc_put(sc);
+		kernel_sock_shutdown(sc->sc_sock, SHUT_RDWR);
+	}
+
+	/* not fatal so failed connects before the other guy has our
+	 * heartbeat can be retried */
+	o2net_ensure_shutdown(nn, sc, 0);
+	sc_put(sc);
+}
+
+/* ------------------------------------------------------------ */
+
+static int o2net_handler_cmp(struct o2net_msg_handler *nmh, u32 msg_type,
+			     u32 key)
+{
+	int ret = memcmp(&nmh->nh_key, &key, sizeof(key));
+
+	if (ret == 0)
+		ret = memcmp(&nmh->nh_msg_type, &msg_type, sizeof(msg_type));
+
+	return ret;
+}
+
+static struct o2net_msg_handler *
+o2net_handler_tree_lookup(u32 msg_type, u32 key, struct rb_node ***ret_p,
+			  struct rb_node **ret_parent)
+{
+	struct rb_node **p = &o2net_handler_tree.rb_node;
+	struct rb_node *parent = NULL;
+	struct o2net_msg_handler *nmh, *ret = NULL;
+	int cmp;
+
+	while (*p) {
+		parent = *p;
+		nmh = rb_entry(parent, struct o2net_msg_handler, nh_node);
+		cmp = o2net_handler_cmp(nmh, msg_type, key);
+
+		if (cmp < 0)
+			p = &(*p)->rb_left;
+		else if (cmp > 0)
+			p = &(*p)->rb_right;
+		else {
+			ret = nmh;
+			break;
+		}
+	}
+
+	if (ret_p != NULL)
+		*ret_p = p;
+	if (ret_parent != NULL)
+		*ret_parent = parent;
+
+	return ret;
+}
+
+static void o2net_handler_kref_release(struct kref *kref)
+{
+	struct o2net_msg_handler *nmh;
+	nmh = container_of(kref, struct o2net_msg_handler, nh_kref);
+
+	kfree(nmh);
+}
+
+static void o2net_handler_put(struct o2net_msg_handler *nmh)
+{
+	kref_put(&nmh->nh_kref, o2net_handler_kref_release);
+}
+
+/* max_len is protection for the handler func.  incoming messages won't
+ * be given to the handler if their payload is longer than the max. */
+int o2net_register_handler(u32 msg_type, u32 key, u32 max_len,
+			   o2net_msg_handler_func *func, void *data,
+			   o2net_post_msg_handler_func *post_func,
+			   struct list_head *unreg_list)
+{
+	struct o2net_msg_handler *nmh = NULL;
+	struct rb_node **p, *parent;
+	int ret = 0;
+
+	if (max_len > O2NET_MAX_PAYLOAD_BYTES) {
+		mlog(0, "max_len for message handler out of range: %u\n",
+			max_len);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (!msg_type) {
+		mlog(0, "no message type provided: %u, %p\n", msg_type, func);
+		ret = -EINVAL;
+		goto out;
+
+	}
+	if (!func) {
+		mlog(0, "no message handler provided: %u, %p\n",
+		       msg_type, func);
+		ret = -EINVAL;
+		goto out;
+	}
+
+       	nmh = kzalloc(sizeof(struct o2net_msg_handler), GFP_NOFS);
+	if (nmh == NULL) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	nmh->nh_func = func;
+	nmh->nh_func_data = data;
+	nmh->nh_post_func = post_func;
+	nmh->nh_msg_type = msg_type;
+	nmh->nh_max_len = max_len;
+	nmh->nh_key = key;
+	/* the tree and list get this ref.. they're both removed in
+	 * unregister when this ref is dropped */
+	kref_init(&nmh->nh_kref);
+	INIT_LIST_HEAD(&nmh->nh_unregister_item);
+
+	write_lock(&o2net_handler_lock);
+	if (o2net_handler_tree_lookup(msg_type, key, &p, &parent))
+		ret = -EEXIST;
+	else {
+	        rb_link_node(&nmh->nh_node, parent, p);
+		rb_insert_color(&nmh->nh_node, &o2net_handler_tree);
+		list_add_tail(&nmh->nh_unregister_item, unreg_list);
+
+		mlog(ML_TCP, "registered handler func %p type %u key %08x\n",
+		     func, msg_type, key);
+		/* we've had some trouble with handlers seemingly vanishing. */
+		mlog_bug_on_msg(o2net_handler_tree_lookup(msg_type, key, &p,
+							  &parent) == NULL,
+			        "couldn't find handler we *just* registered "
+				"for type %u key %08x\n", msg_type, key);
+	}
+	write_unlock(&o2net_handler_lock);
+	if (ret)
+		goto out;
+
+out:
+	if (ret)
+		kfree(nmh);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(o2net_register_handler);
+
+void o2net_unregister_handler_list(struct list_head *list)
+{
+	struct o2net_msg_handler *nmh, *n;
+
+	write_lock(&o2net_handler_lock);
+	list_for_each_entry_safe(nmh, n, list, nh_unregister_item) {
+		mlog(ML_TCP, "unregistering handler func %p type %u key %08x\n",
+		     nmh->nh_func, nmh->nh_msg_type, nmh->nh_key);
+		rb_erase(&nmh->nh_node, &o2net_handler_tree);
+		list_del_init(&nmh->nh_unregister_item);
+		kref_put(&nmh->nh_kref, o2net_handler_kref_release);
+	}
+	write_unlock(&o2net_handler_lock);
+}
+EXPORT_SYMBOL_GPL(o2net_unregister_handler_list);
+
+static struct o2net_msg_handler *o2net_handler_get(u32 msg_type, u32 key)
+{
+	struct o2net_msg_handler *nmh;
+
+	read_lock(&o2net_handler_lock);
+	nmh = o2net_handler_tree_lookup(msg_type, key, NULL, NULL);
+	if (nmh)
+		kref_get(&nmh->nh_kref);
+	read_unlock(&o2net_handler_lock);
+
+	return nmh;
+}
+
+/* ------------------------------------------------------------ */
+
+static int o2net_recv_tcp_msg(struct socket *sock, void *data, size_t len)
+{
+	struct kvec vec = { .iov_len = len, .iov_base = data, };
+	struct msghdr msg = { .msg_flags = MSG_DONTWAIT, };
+	return kernel_recvmsg(sock, &msg, &vec, 1, len, msg.msg_flags);
+}
+
+static int o2net_send_tcp_msg(struct socket *sock, struct kvec *vec,
+			      size_t veclen, size_t total)
+{
+	int ret;
+	struct msghdr msg = {.msg_flags = 0,};
+
+	if (sock == NULL) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	ret = kernel_sendmsg(sock, &msg, vec, veclen, total);
+	if (likely(ret == total))
+		return 0;
+	mlog(ML_ERROR, "sendmsg returned %d instead of %zu\n", ret, total);
+	if (ret >= 0)
+		ret = -EPIPE; /* should be smarter, I bet */
+out:
+	mlog(0, "returning error: %d\n", ret);
+	return ret;
+}
+
+static void o2net_sendpage(struct o2net_sock_container *sc,
+			   void *kmalloced_virt,
+			   size_t size)
+{
+	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
+	ssize_t ret;
+
+	while (1) {
+		mutex_lock(&sc->sc_send_lock);
+		ret = sc->sc_sock->ops->sendpage(sc->sc_sock,
+						 virt_to_page(kmalloced_virt),
+						 (long)kmalloced_virt & ~PAGE_MASK,
+						 size, MSG_DONTWAIT);
+		mutex_unlock(&sc->sc_send_lock);
+		if (ret == size)
+			break;
+		if (ret == (ssize_t)-EAGAIN) {
+			mlog(0, "sendpage of size %zu to " SC_NODEF_FMT
+			     " returned EAGAIN\n", size, SC_NODEF_ARGS(sc));
+			cond_resched();
+			continue;
+		}
+		mlog(ML_ERROR, "sendpage of size %zu to " SC_NODEF_FMT
+		     " failed with %zd\n", size, SC_NODEF_ARGS(sc), ret);
+		o2net_ensure_shutdown(nn, sc, 0);
+		break;
+	}
+}
+
+static void o2net_init_msg(struct o2net_msg *msg, u16 data_len, u16 msg_type, u32 key)
+{
+	memset(msg, 0, sizeof(struct o2net_msg));
+	msg->magic = cpu_to_be16(O2NET_MSG_MAGIC);
+	msg->data_len = cpu_to_be16(data_len);
+	msg->msg_type = cpu_to_be16(msg_type);
+	msg->sys_status = cpu_to_be32(O2NET_ERR_NONE);
+	msg->status = 0;
+	msg->key = cpu_to_be32(key);
+}
+
+static int o2net_tx_can_proceed(struct o2net_node *nn,
+			        struct o2net_sock_container **sc_ret,
+				int *error)
+{
+	int ret = 0;
+
+	spin_lock(&nn->nn_lock);
+	if (nn->nn_persistent_error) {
+		ret = 1;
+		*sc_ret = NULL;
+		*error = nn->nn_persistent_error;
+	} else if (nn->nn_sc_valid) {
+		kref_get(&nn->nn_sc->sc_kref);
+
+		ret = 1;
+		*sc_ret = nn->nn_sc;
+		*error = 0;
+	}
+	spin_unlock(&nn->nn_lock);
+
+	return ret;
+}
+
+/* Get a map of all nodes to which this node is currently connected to */
+void o2net_fill_node_map(unsigned long *map, unsigned bytes)
+{
+	struct o2net_sock_container *sc;
+	int node, ret;
+
+	BUG_ON(bytes < (BITS_TO_LONGS(O2NM_MAX_NODES) * sizeof(unsigned long)));
+
+	memset(map, 0, bytes);
+	for (node = 0; node < O2NM_MAX_NODES; ++node) {
+		if (!o2net_tx_can_proceed(o2net_nn_from_num(node), &sc, &ret))
+			continue;
+		if (!ret) {
+			set_bit(node, map);
+			sc_put(sc);
+		}
+	}
+}
+EXPORT_SYMBOL_GPL(o2net_fill_node_map);
+
+int o2net_send_message_vec(u32 msg_type, u32 key, struct kvec *caller_vec,
+			   size_t caller_veclen, u8 target_node, int *status)
+{
+	int ret = 0;
+	struct o2net_msg *msg = NULL;
+	size_t veclen, caller_bytes = 0;
+	struct kvec *vec = NULL;
+	struct o2net_sock_container *sc = NULL;
+	struct o2net_node *nn = o2net_nn_from_num(target_node);
+	struct o2net_status_wait nsw = {
+		.ns_node_item = LIST_HEAD_INIT(nsw.ns_node_item),
+	};
+	struct o2net_send_tracking nst;
+
+	o2net_init_nst(&nst, msg_type, key, current, target_node);
+
+	if (o2net_wq == NULL) {
+		mlog(0, "attempt to tx without o2netd running\n");
+		ret = -ESRCH;
+		goto out;
+	}
+
+	if (caller_veclen == 0) {
+		mlog(0, "bad kvec array length\n");
+		ret = -EINVAL;
+		goto out;
+	}
+
+	caller_bytes = iov_length((struct iovec *)caller_vec, caller_veclen);
+	if (caller_bytes > O2NET_MAX_PAYLOAD_BYTES) {
+		mlog(0, "total payload len %zu too large\n", caller_bytes);
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (target_node == o2nm_this_node()) {
+		ret = -ELOOP;
+		goto out;
+	}
+
+	o2net_debug_add_nst(&nst);
+
+	o2net_set_nst_sock_time(&nst);
+
+	wait_event(nn->nn_sc_wq, o2net_tx_can_proceed(nn, &sc, &ret));
+	if (ret)
+		goto out;
+
+	o2net_set_nst_sock_container(&nst, sc);
+
+	veclen = caller_veclen + 1;
+	vec = kmalloc(sizeof(struct kvec) * veclen, GFP_ATOMIC);
+	if (vec == NULL) {
+		mlog(0, "failed to %zu element kvec!\n", veclen);
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	msg = kmalloc(sizeof(struct o2net_msg), GFP_ATOMIC);
+	if (!msg) {
+		mlog(0, "failed to allocate a o2net_msg!\n");
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	o2net_init_msg(msg, caller_bytes, msg_type, key);
+
+	vec[0].iov_len = sizeof(struct o2net_msg);
+	vec[0].iov_base = msg;
+	memcpy(&vec[1], caller_vec, caller_veclen * sizeof(struct kvec));
+
+	ret = o2net_prep_nsw(nn, &nsw);
+	if (ret)
+		goto out;
+
+	msg->msg_num = cpu_to_be32(nsw.ns_id);
+	o2net_set_nst_msg_id(&nst, nsw.ns_id);
+
+	o2net_set_nst_send_time(&nst);
+
+	/* finally, convert the message header to network byte-order
+	 * and send */
+	mutex_lock(&sc->sc_send_lock);
+	ret = o2net_send_tcp_msg(sc->sc_sock, vec, veclen,
+				 sizeof(struct o2net_msg) + caller_bytes);
+	mutex_unlock(&sc->sc_send_lock);
+	msglog(msg, "sending returned %d\n", ret);
+	if (ret < 0) {
+		mlog(0, "error returned from o2net_send_tcp_msg=%d\n", ret);
+		goto out;
+	}
+
+	/* wait on other node's handler */
+	o2net_set_nst_status_time(&nst);
+	wait_event(nsw.ns_wq, o2net_nsw_completed(nn, &nsw));
+
+	o2net_update_send_stats(&nst, sc);
+
+	/* Note that we avoid overwriting the callers status return
+	 * variable if a system error was reported on the other
+	 * side. Callers beware. */
+	ret = o2net_sys_err_to_errno(nsw.ns_sys_status);
+	if (status && !ret)
+		*status = nsw.ns_status;
+
+	mlog(0, "woken, returning system status %d, user status %d\n",
+	     ret, nsw.ns_status);
+out:
+	o2net_debug_del_nst(&nst); /* must be before dropping sc and node */
+	if (sc)
+		sc_put(sc);
+	kfree(vec);
+	kfree(msg);
+	o2net_complete_nsw(nn, &nsw, 0, 0, 0);
+	return ret;
+}
+EXPORT_SYMBOL_GPL(o2net_send_message_vec);
+
+int o2net_send_message(u32 msg_type, u32 key, void *data, u32 len,
+		       u8 target_node, int *status)
+{
+	struct kvec vec = {
+		.iov_base = data,
+		.iov_len = len,
+	};
+	return o2net_send_message_vec(msg_type, key, &vec, 1,
+				      target_node, status);
+}
+EXPORT_SYMBOL_GPL(o2net_send_message);
+
+static int o2net_send_status_magic(struct socket *sock, struct o2net_msg *hdr,
+				   enum o2net_system_error syserr, int err)
+{
+	struct kvec vec = {
+		.iov_base = hdr,
+		.iov_len = sizeof(struct o2net_msg),
+	};
+
+	BUG_ON(syserr >= O2NET_ERR_MAX);
+
+	/* leave other fields intact from the incoming message, msg_num
+	 * in particular */
+	hdr->sys_status = cpu_to_be32(syserr);
+	hdr->status = cpu_to_be32(err);
+	hdr->magic = cpu_to_be16(O2NET_MSG_STATUS_MAGIC);  // twiddle the magic
+	hdr->data_len = 0;
+
+	msglog(hdr, "about to send status magic %d\n", err);
+	/* hdr has been in host byteorder this whole time */
+	return o2net_send_tcp_msg(sock, &vec, 1, sizeof(struct o2net_msg));
+}
+
+/* this returns -errno if the header was unknown or too large, etc.
+ * after this is called the buffer us reused for the next message */
+static int o2net_process_message(struct o2net_sock_container *sc,
+				 struct o2net_msg *hdr)
+{
+	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
+	int ret = 0, handler_status;
+	enum  o2net_system_error syserr;
+	struct o2net_msg_handler *nmh = NULL;
+	void *ret_data = NULL;
+
+	msglog(hdr, "processing message\n");
+
+	o2net_sc_postpone_idle(sc);
+
+	switch(be16_to_cpu(hdr->magic)) {
+		case O2NET_MSG_STATUS_MAGIC:
+			/* special type for returning message status */
+			o2net_complete_nsw(nn, NULL,
+					   be32_to_cpu(hdr->msg_num),
+					   be32_to_cpu(hdr->sys_status),
+					   be32_to_cpu(hdr->status));
+			goto out;
+		case O2NET_MSG_KEEP_REQ_MAGIC:
+			o2net_sendpage(sc, o2net_keep_resp,
+				       sizeof(*o2net_keep_resp));
+			goto out;
+		case O2NET_MSG_KEEP_RESP_MAGIC:
+			goto out;
+		case O2NET_MSG_MAGIC:
+			break;
+		default:
+			msglog(hdr, "bad magic\n");
+			ret = -EINVAL;
+			goto out;
+			break;
+	}
+
+	/* find a handler for it */
+	handler_status = 0;
+	nmh = o2net_handler_get(be16_to_cpu(hdr->msg_type),
+				be32_to_cpu(hdr->key));
+	if (!nmh) {
+		mlog(ML_TCP, "couldn't find handler for type %u key %08x\n",
+		     be16_to_cpu(hdr->msg_type), be32_to_cpu(hdr->key));
+		syserr = O2NET_ERR_NO_HNDLR;
+		goto out_respond;
+	}
+
+	syserr = O2NET_ERR_NONE;
+
+	if (be16_to_cpu(hdr->data_len) > nmh->nh_max_len)
+		syserr = O2NET_ERR_OVERFLOW;
+
+	if (syserr != O2NET_ERR_NONE)
+		goto out_respond;
+
+	o2net_set_func_start_time(sc);
+	sc->sc_msg_key = be32_to_cpu(hdr->key);
+	sc->sc_msg_type = be16_to_cpu(hdr->msg_type);
+	handler_status = (nmh->nh_func)(hdr, sizeof(struct o2net_msg) +
+					     be16_to_cpu(hdr->data_len),
+					nmh->nh_func_data, &ret_data);
+	o2net_set_func_stop_time(sc);
+
+	o2net_update_recv_stats(sc);
+
+out_respond:
+	/* this destroys the hdr, so don't use it after this */
+	mutex_lock(&sc->sc_send_lock);
+	ret = o2net_send_status_magic(sc->sc_sock, hdr, syserr,
+				      handler_status);
+	mutex_unlock(&sc->sc_send_lock);
+	hdr = NULL;
+	mlog(0, "sending handler status %d, syserr %d returned %d\n",
+	     handler_status, syserr, ret);
+
+	if (nmh) {
+		BUG_ON(ret_data != NULL && nmh->nh_post_func == NULL);
+		if (nmh->nh_post_func)
+			(nmh->nh_post_func)(handler_status, nmh->nh_func_data,
+					    ret_data);
+	}
+
+out:
+	if (nmh)
+		o2net_handler_put(nmh);
+	return ret;
+}
+
+static int o2net_check_handshake(struct o2net_sock_container *sc)
+{
+	struct o2net_handshake *hand = page_address(sc->sc_page);
+	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
+
+	if (hand->protocol_version != cpu_to_be64(O2NET_PROTOCOL_VERSION)) {
+		printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " Advertised net "
+		       "protocol version %llu but %llu is required. "
+		       "Disconnecting.\n", SC_NODEF_ARGS(sc),
+		       (unsigned long long)be64_to_cpu(hand->protocol_version),
+		       O2NET_PROTOCOL_VERSION);
+
+		/* don't bother reconnecting if its the wrong version. */
+		o2net_ensure_shutdown(nn, sc, -ENOTCONN);
+		return -1;
+	}
+
+	/*
+	 * Ensure timeouts are consistent with other nodes, otherwise
+	 * we can end up with one node thinking that the other must be down,
+	 * but isn't. This can ultimately cause corruption.
+	 */
+	if (be32_to_cpu(hand->o2net_idle_timeout_ms) !=
+				o2net_idle_timeout()) {
+		printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " uses a network "
+		       "idle timeout of %u ms, but we use %u ms locally. "
+		       "Disconnecting.\n", SC_NODEF_ARGS(sc),
+		       be32_to_cpu(hand->o2net_idle_timeout_ms),
+		       o2net_idle_timeout());
+		o2net_ensure_shutdown(nn, sc, -ENOTCONN);
+		return -1;
+	}
+
+	if (be32_to_cpu(hand->o2net_keepalive_delay_ms) !=
+			o2net_keepalive_delay()) {
+		printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " uses a keepalive "
+		       "delay of %u ms, but we use %u ms locally. "
+		       "Disconnecting.\n", SC_NODEF_ARGS(sc),
+		       be32_to_cpu(hand->o2net_keepalive_delay_ms),
+		       o2net_keepalive_delay());
+		o2net_ensure_shutdown(nn, sc, -ENOTCONN);
+		return -1;
+	}
+
+	if (be32_to_cpu(hand->o2hb_heartbeat_timeout_ms) !=
+			O2HB_MAX_WRITE_TIMEOUT_MS) {
+		printk(KERN_NOTICE "o2net: " SC_NODEF_FMT " uses a heartbeat "
+		       "timeout of %u ms, but we use %u ms locally. "
+		       "Disconnecting.\n", SC_NODEF_ARGS(sc),
+		       be32_to_cpu(hand->o2hb_heartbeat_timeout_ms),
+		       O2HB_MAX_WRITE_TIMEOUT_MS);
+		o2net_ensure_shutdown(nn, sc, -ENOTCONN);
+		return -1;
+	}
+
+	sc->sc_handshake_ok = 1;
+
+	spin_lock(&nn->nn_lock);
+	/* set valid and queue the idle timers only if it hasn't been
+	 * shut down already */
+	if (nn->nn_sc == sc) {
+		o2net_sc_reset_idle_timer(sc);
+		atomic_set(&nn->nn_timeout, 0);
+		o2net_set_nn_state(nn, sc, 1, 0);
+	}
+	spin_unlock(&nn->nn_lock);
+
+	/* shift everything up as though it wasn't there */
+	sc->sc_page_off -= sizeof(struct o2net_handshake);
+	if (sc->sc_page_off)
+		memmove(hand, hand + 1, sc->sc_page_off);
+
+	return 0;
+}
+
+/* this demuxes the queued rx bytes into header or payload bits and calls
+ * handlers as each full message is read off the socket.  it returns -error,
+ * == 0 eof, or > 0 for progress made.*/
+static int o2net_advance_rx(struct o2net_sock_container *sc)
+{
+	struct o2net_msg *hdr;
+	int ret = 0;
+	void *data;
+	size_t datalen;
+
+	sclog(sc, "receiving\n");
+	o2net_set_advance_start_time(sc);
+
+	if (unlikely(sc->sc_handshake_ok == 0)) {
+		if(sc->sc_page_off < sizeof(struct o2net_handshake)) {
+			data = page_address(sc->sc_page) + sc->sc_page_off;
+			datalen = sizeof(struct o2net_handshake) - sc->sc_page_off;
+			ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
+			if (ret > 0)
+				sc->sc_page_off += ret;
+		}
+
+		if (sc->sc_page_off == sizeof(struct o2net_handshake)) {
+			o2net_check_handshake(sc);
+			if (unlikely(sc->sc_handshake_ok == 0))
+				ret = -EPROTO;
+		}
+		goto out;
+	}
+
+	/* do we need more header? */
+	if (sc->sc_page_off < sizeof(struct o2net_msg)) {
+		data = page_address(sc->sc_page) + sc->sc_page_off;
+		datalen = sizeof(struct o2net_msg) - sc->sc_page_off;
+		ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
+		if (ret > 0) {
+			sc->sc_page_off += ret;
+			/* only swab incoming here.. we can
+			 * only get here once as we cross from
+			 * being under to over */
+			if (sc->sc_page_off == sizeof(struct o2net_msg)) {
+				hdr = page_address(sc->sc_page);
+				if (be16_to_cpu(hdr->data_len) >
+				    O2NET_MAX_PAYLOAD_BYTES)
+					ret = -EOVERFLOW;
+			}
+		}
+		if (ret <= 0)
+			goto out;
+	}
+
+	if (sc->sc_page_off < sizeof(struct o2net_msg)) {
+		/* oof, still don't have a header */
+		goto out;
+	}
+
+	/* this was swabbed above when we first read it */
+	hdr = page_address(sc->sc_page);
+
+	msglog(hdr, "at page_off %zu\n", sc->sc_page_off);
+
+	/* do we need more payload? */
+	if (sc->sc_page_off - sizeof(struct o2net_msg) < be16_to_cpu(hdr->data_len)) {
+		/* need more payload */
+		data = page_address(sc->sc_page) + sc->sc_page_off;
+		datalen = (sizeof(struct o2net_msg) + be16_to_cpu(hdr->data_len)) -
+			  sc->sc_page_off;
+		ret = o2net_recv_tcp_msg(sc->sc_sock, data, datalen);
+		if (ret > 0)
+			sc->sc_page_off += ret;
+		if (ret <= 0)
+			goto out;
+	}
+
+	if (sc->sc_page_off - sizeof(struct o2net_msg) == be16_to_cpu(hdr->data_len)) {
+		/* we can only get here once, the first time we read
+		 * the payload.. so set ret to progress if the handler
+		 * works out. after calling this the message is toast */
+		ret = o2net_process_message(sc, hdr);
+		if (ret == 0)
+			ret = 1;
+		sc->sc_page_off = 0;
+	}
+
+out:
+	sclog(sc, "ret = %d\n", ret);
+	o2net_set_advance_stop_time(sc);
+	return ret;
+}
+
+/* this work func is triggerd by data ready.  it reads until it can read no
+ * more.  it interprets 0, eof, as fatal.  if data_ready hits while we're doing
+ * our work the work struct will be marked and we'll be called again. */
+static void o2net_rx_until_empty(struct work_struct *work)
+{
+	struct o2net_sock_container *sc =
+		container_of(work, struct o2net_sock_container, sc_rx_work);
+	int ret;
+
+	do {
+		ret = o2net_advance_rx(sc);
+	} while (ret > 0);
+
+	if (ret <= 0 && ret != -EAGAIN) {
+		struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
+		sclog(sc, "saw error %d, closing\n", ret);
+		/* not permanent so read failed handshake can retry */
+		o2net_ensure_shutdown(nn, sc, 0);
+	}
+
+	sc_put(sc);
+}
+
+static int o2net_set_nodelay(struct socket *sock)
+{
+	int ret, val = 1;
+	mm_segment_t oldfs;
+
+	oldfs = get_fs();
+	set_fs(KERNEL_DS);
+
+	/*
+	 * Dear unsuspecting programmer,
+	 *
+	 * Don't use sock_setsockopt() for SOL_TCP.  It doesn't check its level
+	 * argument and assumes SOL_SOCKET so, say, your TCP_NODELAY will
+	 * silently turn into SO_DEBUG.
+	 *
+	 * Yours,
+	 * Keeper of hilariously fragile interfaces.
+	 */
+	ret = sock->ops->setsockopt(sock, SOL_TCP, TCP_NODELAY,
+				    (char __user *)&val, sizeof(val));
+
+	set_fs(oldfs);
+	return ret;
+}
+
+static int o2net_set_usertimeout(struct socket *sock)
+{
+	int user_timeout = O2NET_TCP_USER_TIMEOUT;
+
+	return kernel_setsockopt(sock, SOL_TCP, TCP_USER_TIMEOUT,
+				(char *)&user_timeout, sizeof(user_timeout));
+}
+
+static void o2net_initialize_handshake(void)
+{
+	o2net_hand->o2hb_heartbeat_timeout_ms = cpu_to_be32(
+		O2HB_MAX_WRITE_TIMEOUT_MS);
+	o2net_hand->o2net_idle_timeout_ms = cpu_to_be32(o2net_idle_timeout());
+	o2net_hand->o2net_keepalive_delay_ms = cpu_to_be32(
+		o2net_keepalive_delay());
+	o2net_hand->o2net_reconnect_delay_ms = cpu_to_be32(
+		o2net_reconnect_delay());
+}
+
+/* ------------------------------------------------------------ */
+
+/* called when a connect completes and after a sock is accepted.  the
+ * rx path will see the response and mark the sc valid */
+static void o2net_sc_connect_completed(struct work_struct *work)
+{
+	struct o2net_sock_container *sc =
+		container_of(work, struct o2net_sock_container,
+			     sc_connect_work);
+
+	mlog(ML_MSG, "sc sending handshake with ver %llu id %llx\n",
+              (unsigned long long)O2NET_PROTOCOL_VERSION,
+	      (unsigned long long)be64_to_cpu(o2net_hand->connector_id));
+
+	o2net_initialize_handshake();
+	o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand));
+	sc_put(sc);
+}
+
+/* this is called as a work_struct func. */
+static void o2net_sc_send_keep_req(struct work_struct *work)
+{
+	struct o2net_sock_container *sc =
+		container_of(work, struct o2net_sock_container,
+			     sc_keepalive_work.work);
+
+	o2net_sendpage(sc, o2net_keep_req, sizeof(*o2net_keep_req));
+	sc_put(sc);
+}
+
+/* socket shutdown does a del_timer_sync against this as it tears down.
+ * we can't start this timer until we've got to the point in sc buildup
+ * where shutdown is going to be involved */
+static void o2net_idle_timer(unsigned long data)
+{
+	struct o2net_sock_container *sc = (struct o2net_sock_container *)data;
+	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
+#ifdef CONFIG_DEBUG_FS
+	unsigned long msecs = ktime_to_ms(ktime_get()) -
+		ktime_to_ms(sc->sc_tv_timer);
+#else
+	unsigned long msecs = o2net_idle_timeout();
+#endif
+
+	printk(KERN_NOTICE "o2net: Connection to " SC_NODEF_FMT " has been "
+	       "idle for %lu.%lu secs.\n",
+	       SC_NODEF_ARGS(sc), msecs / 1000, msecs % 1000);
+
+	/* idle timerout happen, don't shutdown the connection, but
+	 * make fence decision. Maybe the connection can recover before
+	 * the decision is made.
+	 */
+	atomic_set(&nn->nn_timeout, 1);
+	o2quo_conn_err(o2net_num_from_nn(nn));
+	queue_delayed_work(o2net_wq, &nn->nn_still_up,
+			msecs_to_jiffies(O2NET_QUORUM_DELAY_MS));
+
+	o2net_sc_reset_idle_timer(sc);
+
+}
+
+static void o2net_sc_reset_idle_timer(struct o2net_sock_container *sc)
+{
+	o2net_sc_cancel_delayed_work(sc, &sc->sc_keepalive_work);
+	o2net_sc_queue_delayed_work(sc, &sc->sc_keepalive_work,
+		      msecs_to_jiffies(o2net_keepalive_delay()));
+	o2net_set_sock_timer(sc);
+	mod_timer(&sc->sc_idle_timeout,
+	       jiffies + msecs_to_jiffies(o2net_idle_timeout()));
+}
+
+static void o2net_sc_postpone_idle(struct o2net_sock_container *sc)
+{
+	struct o2net_node *nn = o2net_nn_from_num(sc->sc_node->nd_num);
+
+	/* clear fence decision since the connection recover from timeout*/
+	if (atomic_read(&nn->nn_timeout)) {
+		o2quo_conn_up(o2net_num_from_nn(nn));
+		cancel_delayed_work(&nn->nn_still_up);
+		atomic_set(&nn->nn_timeout, 0);
+	}
+
+	/* Only push out an existing timer */
+	if (timer_pending(&sc->sc_idle_timeout))
+		o2net_sc_reset_idle_timer(sc);
+}
+
+/* this work func is kicked whenever a path sets the nn state which doesn't
+ * have valid set.  This includes seeing hb come up, losing a connection,
+ * having a connect attempt fail, etc. This centralizes the logic which decides
+ * if a connect attempt should be made or if we should give up and all future
+ * transmit attempts should fail */
+static void o2net_start_connect(struct work_struct *work)
+{
+	struct o2net_node *nn =
+		container_of(work, struct o2net_node, nn_connect_work.work);
+	struct o2net_sock_container *sc = NULL;
+	struct o2nm_node *node = NULL, *mynode = NULL;
+	struct socket *sock = NULL;
+	struct sockaddr_in myaddr = {0, }, remoteaddr = {0, };
+	int ret = 0, stop;
+	unsigned int timeout;
+	unsigned int noio_flag;
+
+	/*
+	 * sock_create allocates the sock with GFP_KERNEL. We must set
+	 * per-process flag PF_MEMALLOC_NOIO so that all allocations done
+	 * by this process are done as if GFP_NOIO was specified. So we
+	 * are not reentering filesystem while doing memory reclaim.
+	 */
+	noio_flag = memalloc_noio_save();
+	/* if we're greater we initiate tx, otherwise we accept */
+	if (o2nm_this_node() <= o2net_num_from_nn(nn))
+		goto out;
+
+	/* watch for racing with tearing a node down */
+	node = o2nm_get_node_by_num(o2net_num_from_nn(nn));
+	if (node == NULL) {
+		ret = 0;
+		goto out;
+	}
+
+	mynode = o2nm_get_node_by_num(o2nm_this_node());
+	if (mynode == NULL) {
+		ret = 0;
+		goto out;
+	}
+
+	spin_lock(&nn->nn_lock);
+	/*
+	 * see if we already have one pending or have given up.
+	 * For nn_timeout, it is set when we close the connection
+	 * because of the idle time out. So it means that we have
+	 * at least connected to that node successfully once,
+	 * now try to connect to it again.
+	 */
+	timeout = atomic_read(&nn->nn_timeout);
+	stop = (nn->nn_sc ||
+		(nn->nn_persistent_error &&
+		(nn->nn_persistent_error != -ENOTCONN || timeout == 0)));
+	spin_unlock(&nn->nn_lock);
+	if (stop)
+		goto out;
+
+	nn->nn_last_connect_attempt = jiffies;
+
+	sc = sc_alloc(node);
+	if (sc == NULL) {
+		mlog(0, "couldn't allocate sc\n");
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
+	if (ret < 0) {
+		mlog(0, "can't create socket: %d\n", ret);
+		goto out;
+	}
+	sc->sc_sock = sock; /* freed by sc_kref_release */
+
+	sock->sk->sk_allocation = GFP_ATOMIC;
+
+	myaddr.sin_family = AF_INET;
+	myaddr.sin_addr.s_addr = mynode->nd_ipv4_address;
+	myaddr.sin_port = htons(0); /* any port */
+
+	ret = sock->ops->bind(sock, (struct sockaddr *)&myaddr,
+			      sizeof(myaddr));
+	if (ret) {
+		mlog(ML_ERROR, "bind failed with %d at address %pI4\n",
+		     ret, &mynode->nd_ipv4_address);
+		goto out;
+	}
+
+	ret = o2net_set_nodelay(sc->sc_sock);
+	if (ret) {
+		mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
+		goto out;
+	}
+
+	ret = o2net_set_usertimeout(sock);
+	if (ret) {
+		mlog(ML_ERROR, "set TCP_USER_TIMEOUT failed with %d\n", ret);
+		goto out;
+	}
+
+	o2net_register_callbacks(sc->sc_sock->sk, sc);
+
+	spin_lock(&nn->nn_lock);
+	/* handshake completion will set nn->nn_sc_valid */
+	o2net_set_nn_state(nn, sc, 0, 0);
+	spin_unlock(&nn->nn_lock);
+
+	remoteaddr.sin_family = AF_INET;
+	remoteaddr.sin_addr.s_addr = node->nd_ipv4_address;
+	remoteaddr.sin_port = node->nd_ipv4_port;
+
+	ret = sc->sc_sock->ops->connect(sc->sc_sock,
+					(struct sockaddr *)&remoteaddr,
+					sizeof(remoteaddr),
+					O_NONBLOCK);
+	if (ret == -EINPROGRESS)
+		ret = 0;
+
+out:
+	if (ret && sc) {
+		printk(KERN_NOTICE "o2net: Connect attempt to " SC_NODEF_FMT
+		       " failed with errno %d\n", SC_NODEF_ARGS(sc), ret);
+		/* 0 err so that another will be queued and attempted
+		 * from set_nn_state */
+		o2net_ensure_shutdown(nn, sc, 0);
+	}
+	if (sc)
+		sc_put(sc);
+	if (node)
+		o2nm_node_put(node);
+	if (mynode)
+		o2nm_node_put(mynode);
+
+	memalloc_noio_restore(noio_flag);
+	return;
+}
+
+static void o2net_connect_expired(struct work_struct *work)
+{
+	struct o2net_node *nn =
+		container_of(work, struct o2net_node, nn_connect_expired.work);
+
+	spin_lock(&nn->nn_lock);
+	if (!nn->nn_sc_valid) {
+		printk(KERN_NOTICE "o2net: No connection established with "
+		       "node %u after %u.%u seconds, check network and"
+		       " cluster configuration.\n",
+		     o2net_num_from_nn(nn),
+		     o2net_idle_timeout() / 1000,
+		     o2net_idle_timeout() % 1000);
+
+		o2net_set_nn_state(nn, NULL, 0, 0);
+	}
+	spin_unlock(&nn->nn_lock);
+}
+
+static void o2net_still_up(struct work_struct *work)
+{
+	struct o2net_node *nn =
+		container_of(work, struct o2net_node, nn_still_up.work);
+
+	o2quo_hb_still_up(o2net_num_from_nn(nn));
+}
+
+/* ------------------------------------------------------------ */
+
+void o2net_disconnect_node(struct o2nm_node *node)
+{
+	struct o2net_node *nn = o2net_nn_from_num(node->nd_num);
+
+	/* don't reconnect until it's heartbeating again */
+	spin_lock(&nn->nn_lock);
+	atomic_set(&nn->nn_timeout, 0);
+	o2net_set_nn_state(nn, NULL, 0, -ENOTCONN);
+	spin_unlock(&nn->nn_lock);
+
+	if (o2net_wq) {
+		cancel_delayed_work(&nn->nn_connect_expired);
+		cancel_delayed_work(&nn->nn_connect_work);
+		cancel_delayed_work(&nn->nn_still_up);
+		flush_workqueue(o2net_wq);
+	}
+}
+
+static void o2net_hb_node_down_cb(struct o2nm_node *node, int node_num,
+				  void *data)
+{
+	o2quo_hb_down(node_num);
+
+	if (!node)
+		return;
+
+	if (node_num != o2nm_this_node())
+		o2net_disconnect_node(node);
+
+	BUG_ON(atomic_read(&o2net_connected_peers) < 0);
+}
+
+static void o2net_hb_node_up_cb(struct o2nm_node *node, int node_num,
+				void *data)
+{
+	struct o2net_node *nn = o2net_nn_from_num(node_num);
+
+	o2quo_hb_up(node_num);
+
+	BUG_ON(!node);
+
+	/* ensure an immediate connect attempt */
+	nn->nn_last_connect_attempt = jiffies -
+		(msecs_to_jiffies(o2net_reconnect_delay()) + 1);
+
+	if (node_num != o2nm_this_node()) {
+		/* believe it or not, accept and node hearbeating testing
+		 * can succeed for this node before we got here.. so
+		 * only use set_nn_state to clear the persistent error
+		 * if that hasn't already happened */
+		spin_lock(&nn->nn_lock);
+		atomic_set(&nn->nn_timeout, 0);
+		if (nn->nn_persistent_error)
+			o2net_set_nn_state(nn, NULL, 0, 0);
+		spin_unlock(&nn->nn_lock);
+	}
+}
+
+void o2net_unregister_hb_callbacks(void)
+{
+	o2hb_unregister_callback(NULL, &o2net_hb_up);
+	o2hb_unregister_callback(NULL, &o2net_hb_down);
+}
+
+int o2net_register_hb_callbacks(void)
+{
+	int ret;
+
+	o2hb_setup_callback(&o2net_hb_down, O2HB_NODE_DOWN_CB,
+			    o2net_hb_node_down_cb, NULL, O2NET_HB_PRI);
+	o2hb_setup_callback(&o2net_hb_up, O2HB_NODE_UP_CB,
+			    o2net_hb_node_up_cb, NULL, O2NET_HB_PRI);
+
+	ret = o2hb_register_callback(NULL, &o2net_hb_up);
+	if (ret == 0)
+		ret = o2hb_register_callback(NULL, &o2net_hb_down);
+
+	if (ret)
+		o2net_unregister_hb_callbacks();
+
+	return ret;
+}
+
+/* ------------------------------------------------------------ */
+
+static int o2net_accept_one(struct socket *sock, int *more)
+{
+	int ret, slen;
+	struct sockaddr_in sin;
+	struct socket *new_sock = NULL;
+	struct o2nm_node *node = NULL;
+	struct o2nm_node *local_node = NULL;
+	struct o2net_sock_container *sc = NULL;
+	struct o2net_node *nn;
+	unsigned int noio_flag;
+
+	/*
+	 * sock_create_lite allocates the sock with GFP_KERNEL. We must set
+	 * per-process flag PF_MEMALLOC_NOIO so that all allocations done
+	 * by this process are done as if GFP_NOIO was specified. So we
+	 * are not reentering filesystem while doing memory reclaim.
+	 */
+	noio_flag = memalloc_noio_save();
+
+	BUG_ON(sock == NULL);
+	*more = 0;
+	ret = sock_create_lite(sock->sk->sk_family, sock->sk->sk_type,
+			       sock->sk->sk_protocol, &new_sock);
+	if (ret)
+		goto out;
+
+	new_sock->type = sock->type;
+	new_sock->ops = sock->ops;
+	ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
+	if (ret < 0)
+		goto out;
+
+	*more = 1;
+	new_sock->sk->sk_allocation = GFP_ATOMIC;
+
+	ret = o2net_set_nodelay(new_sock);
+	if (ret) {
+		mlog(ML_ERROR, "setting TCP_NODELAY failed with %d\n", ret);
+		goto out;
+	}
+
+	ret = o2net_set_usertimeout(new_sock);
+	if (ret) {
+		mlog(ML_ERROR, "set TCP_USER_TIMEOUT failed with %d\n", ret);
+		goto out;
+	}
+
+	slen = sizeof(sin);
+	ret = new_sock->ops->getname(new_sock, (struct sockaddr *) &sin,
+				       &slen, 1);
+	if (ret < 0)
+		goto out;
+
+	node = o2nm_get_node_by_ip(sin.sin_addr.s_addr);
+	if (node == NULL) {
+		printk(KERN_NOTICE "o2net: Attempt to connect from unknown "
+		       "node at %pI4:%d\n", &sin.sin_addr.s_addr,
+		       ntohs(sin.sin_port));
+		ret = -EINVAL;
+		goto out;
+	}
+
+	if (o2nm_this_node() >= node->nd_num) {
+		local_node = o2nm_get_node_by_num(o2nm_this_node());
+		if (local_node)
+			printk(KERN_NOTICE "o2net: Unexpected connect attempt "
+					"seen at node '%s' (%u, %pI4:%d) from "
+					"node '%s' (%u, %pI4:%d)\n",
+					local_node->nd_name, local_node->nd_num,
+					&(local_node->nd_ipv4_address),
+					ntohs(local_node->nd_ipv4_port),
+					node->nd_name,
+					node->nd_num, &sin.sin_addr.s_addr,
+					ntohs(sin.sin_port));
+		ret = -EINVAL;
+		goto out;
+	}
+
+	/* this happens all the time when the other node sees our heartbeat
+	 * and tries to connect before we see their heartbeat */
+	if (!o2hb_check_node_heartbeating_from_callback(node->nd_num)) {
+		mlog(ML_CONN, "attempt to connect from node '%s' at "
+		     "%pI4:%d but it isn't heartbeating\n",
+		     node->nd_name, &sin.sin_addr.s_addr,
+		     ntohs(sin.sin_port));
+		ret = -EINVAL;
+		goto out;
+	}
+
+	nn = o2net_nn_from_num(node->nd_num);
+
+	spin_lock(&nn->nn_lock);
+	if (nn->nn_sc)
+		ret = -EBUSY;
+	else
+		ret = 0;
+	spin_unlock(&nn->nn_lock);
+	if (ret) {
+		printk(KERN_NOTICE "o2net: Attempt to connect from node '%s' "
+		       "at %pI4:%d but it already has an open connection\n",
+		       node->nd_name, &sin.sin_addr.s_addr,
+		       ntohs(sin.sin_port));
+		goto out;
+	}
+
+	sc = sc_alloc(node);
+	if (sc == NULL) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	sc->sc_sock = new_sock;
+	new_sock = NULL;
+
+	spin_lock(&nn->nn_lock);
+	atomic_set(&nn->nn_timeout, 0);
+	o2net_set_nn_state(nn, sc, 0, 0);
+	spin_unlock(&nn->nn_lock);
+
+	o2net_register_callbacks(sc->sc_sock->sk, sc);
+	o2net_sc_queue_work(sc, &sc->sc_rx_work);
+
+	o2net_initialize_handshake();
+	o2net_sendpage(sc, o2net_hand, sizeof(*o2net_hand));
+
+out:
+	if (new_sock)
+		sock_release(new_sock);
+	if (node)
+		o2nm_node_put(node);
+	if (local_node)
+		o2nm_node_put(local_node);
+	if (sc)
+		sc_put(sc);
+
+	memalloc_noio_restore(noio_flag);
+	return ret;
+}
+
+/*
+ * This function is invoked in response to one or more
+ * pending accepts at softIRQ level. We must drain the
+ * entire que before returning.
+ */
+
+static void o2net_accept_many(struct work_struct *work)
+{
+	struct socket *sock = o2net_listen_sock;
+	int	more;
+	int	err;
+
+	/*
+	 * It is critical to note that due to interrupt moderation
+	 * at the network driver level, we can't assume to get a
+	 * softIRQ for every single conn since tcp SYN packets
+	 * can arrive back-to-back, and therefore many pending
+	 * accepts may result in just 1 softIRQ. If we terminate
+	 * the o2net_accept_one() loop upon seeing an err, what happens
+	 * to the rest of the conns in the queue? If no new SYN
+	 * arrives for hours, no softIRQ  will be delivered,
+	 * and the connections will just sit in the queue.
+	 */
+
+	for (;;) {
+		err = o2net_accept_one(sock, &more);
+		if (!more)
+			break;
+		cond_resched();
+	}
+}
+
+static void o2net_listen_data_ready(struct sock *sk)
+{
+	void (*ready)(struct sock *sk);
+
+	read_lock(&sk->sk_callback_lock);
+	ready = sk->sk_user_data;
+	if (ready == NULL) { /* check for teardown race */
+		ready = sk->sk_data_ready;
+		goto out;
+	}
+
+	/* This callback may called twice when a new connection
+	 * is  being established as a child socket inherits everything
+	 * from a parent LISTEN socket, including the data_ready cb of
+	 * the parent. This leads to a hazard. In o2net_accept_one()
+	 * we are still initializing the child socket but have not
+	 * changed the inherited data_ready callback yet when
+	 * data starts arriving.
+	 * We avoid this hazard by checking the state.
+	 * For the listening socket,  the state will be TCP_LISTEN; for the new
+	 * socket, will be  TCP_ESTABLISHED. Also, in this case,
+	 * sk->sk_user_data is not a valid function pointer.
+	 */
+
+	if (sk->sk_state == TCP_LISTEN) {
+		queue_work(o2net_wq, &o2net_listen_work);
+	} else {
+		ready = NULL;
+	}
+
+out:
+	read_unlock(&sk->sk_callback_lock);
+	if (ready != NULL)
+		ready(sk);
+}
+
+static int o2net_open_listening_sock(__be32 addr, __be16 port)
+{
+	struct socket *sock = NULL;
+	int ret;
+	struct sockaddr_in sin = {
+		.sin_family = PF_INET,
+		.sin_addr = { .s_addr = addr },
+		.sin_port = port,
+	};
+
+	ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
+	if (ret < 0) {
+		printk(KERN_ERR "o2net: Error %d while creating socket\n", ret);
+		goto out;
+	}
+
+	sock->sk->sk_allocation = GFP_ATOMIC;
+
+	write_lock_bh(&sock->sk->sk_callback_lock);
+	sock->sk->sk_user_data = sock->sk->sk_data_ready;
+	sock->sk->sk_data_ready = o2net_listen_data_ready;
+	write_unlock_bh(&sock->sk->sk_callback_lock);
+
+	o2net_listen_sock = sock;
+	INIT_WORK(&o2net_listen_work, o2net_accept_many);
+
+	sock->sk->sk_reuse = SK_CAN_REUSE;
+	ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
+	if (ret < 0) {
+		printk(KERN_ERR "o2net: Error %d while binding socket at "
+		       "%pI4:%u\n", ret, &addr, ntohs(port)); 
+		goto out;
+	}
+
+	ret = sock->ops->listen(sock, 64);
+	if (ret < 0)
+		printk(KERN_ERR "o2net: Error %d while listening on %pI4:%u\n",
+		       ret, &addr, ntohs(port));
+
+out:
+	if (ret) {
+		o2net_listen_sock = NULL;
+		if (sock)
+			sock_release(sock);
+	}
+	return ret;
+}
+
+/*
+ * called from node manager when we should bring up our network listening
+ * socket.  node manager handles all the serialization to only call this
+ * once and to match it with o2net_stop_listening().  note,
+ * o2nm_this_node() doesn't work yet as we're being called while it
+ * is being set up.
+ */
+int o2net_start_listening(struct o2nm_node *node)
+{
+	int ret = 0;
+
+	BUG_ON(o2net_wq != NULL);
+	BUG_ON(o2net_listen_sock != NULL);
+
+	mlog(ML_KTHREAD, "starting o2net thread...\n");
+	o2net_wq = create_singlethread_workqueue("o2net");
+	if (o2net_wq == NULL) {
+		mlog(ML_ERROR, "unable to launch o2net thread\n");
+		return -ENOMEM; /* ? */
+	}
+
+	ret = o2net_open_listening_sock(node->nd_ipv4_address,
+					node->nd_ipv4_port);
+	if (ret) {
+		destroy_workqueue(o2net_wq);
+		o2net_wq = NULL;
+	} else
+		o2quo_conn_up(node->nd_num);
+
+	return ret;
+}
+
+/* again, o2nm_this_node() doesn't work here as we're involved in
+ * tearing it down */
+void o2net_stop_listening(struct o2nm_node *node)
+{
+	struct socket *sock = o2net_listen_sock;
+	size_t i;
+
+	BUG_ON(o2net_wq == NULL);
+	BUG_ON(o2net_listen_sock == NULL);
+
+	/* stop the listening socket from generating work */
+	write_lock_bh(&sock->sk->sk_callback_lock);
+	sock->sk->sk_data_ready = sock->sk->sk_user_data;
+	sock->sk->sk_user_data = NULL;
+	write_unlock_bh(&sock->sk->sk_callback_lock);
+
+	for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) {
+		struct o2nm_node *node = o2nm_get_node_by_num(i);
+		if (node) {
+			o2net_disconnect_node(node);
+			o2nm_node_put(node);
+		}
+	}
+
+	/* finish all work and tear down the work queue */
+	mlog(ML_KTHREAD, "waiting for o2net thread to exit....\n");
+	destroy_workqueue(o2net_wq);
+	o2net_wq = NULL;
+
+	sock_release(o2net_listen_sock);
+	o2net_listen_sock = NULL;
+
+	o2quo_conn_err(node->nd_num);
+}
+
+/* ------------------------------------------------------------ */
+
+int o2net_init(void)
+{
+	unsigned long i;
+
+	o2quo_init();
+
+	if (o2net_debugfs_init())
+		goto out;
+
+	o2net_hand = kzalloc(sizeof(struct o2net_handshake), GFP_KERNEL);
+	o2net_keep_req = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL);
+	o2net_keep_resp = kzalloc(sizeof(struct o2net_msg), GFP_KERNEL);
+	if (!o2net_hand || !o2net_keep_req || !o2net_keep_resp)
+		goto out;
+
+	o2net_hand->protocol_version = cpu_to_be64(O2NET_PROTOCOL_VERSION);
+	o2net_hand->connector_id = cpu_to_be64(1);
+
+	o2net_keep_req->magic = cpu_to_be16(O2NET_MSG_KEEP_REQ_MAGIC);
+	o2net_keep_resp->magic = cpu_to_be16(O2NET_MSG_KEEP_RESP_MAGIC);
+
+	for (i = 0; i < ARRAY_SIZE(o2net_nodes); i++) {
+		struct o2net_node *nn = o2net_nn_from_num(i);
+
+		atomic_set(&nn->nn_timeout, 0);
+		spin_lock_init(&nn->nn_lock);
+		INIT_DELAYED_WORK(&nn->nn_connect_work, o2net_start_connect);
+		INIT_DELAYED_WORK(&nn->nn_connect_expired,
+				  o2net_connect_expired);
+		INIT_DELAYED_WORK(&nn->nn_still_up, o2net_still_up);
+		/* until we see hb from a node we'll return einval */
+		nn->nn_persistent_error = -ENOTCONN;
+		init_waitqueue_head(&nn->nn_sc_wq);
+		idr_init(&nn->nn_status_idr);
+		INIT_LIST_HEAD(&nn->nn_status_list);
+	}
+
+	return 0;
+
+out:
+	kfree(o2net_hand);
+	kfree(o2net_keep_req);
+	kfree(o2net_keep_resp);
+
+	o2quo_exit();
+	return -ENOMEM;
+}
+
+void o2net_exit(void)
+{
+	o2quo_exit();
+	kfree(o2net_hand);
+	kfree(o2net_keep_req);
+	kfree(o2net_keep_resp);
+	o2net_debugfs_exit();
+}
diff --git a/fs/ocfs2/cluster/tcp.h b/fs/ocfs2/cluster/tcp.h
new file mode 100644
index 000000000..c571e849f
--- /dev/null
+++ b/fs/ocfs2/cluster/tcp.h
@@ -0,0 +1,155 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * tcp.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef O2CLUSTER_TCP_H
+#define O2CLUSTER_TCP_H
+
+#include <linux/socket.h>
+#ifdef __KERNEL__
+#include <net/sock.h>
+#include <linux/tcp.h>
+#else
+#include <sys/socket.h>
+#endif
+#include <linux/inet.h>
+#include <linux/in.h>
+
+struct o2net_msg
+{
+	__be16 magic;
+	__be16 data_len;
+	__be16 msg_type;
+	__be16 pad1;
+	__be32 sys_status;
+	__be32 status;
+	__be32 key;
+	__be32 msg_num;
+	__u8  buf[0];
+};
+
+typedef int (o2net_msg_handler_func)(struct o2net_msg *msg, u32 len, void *data,
+				     void **ret_data);
+typedef void (o2net_post_msg_handler_func)(int status, void *data,
+					   void *ret_data);
+
+#define O2NET_MAX_PAYLOAD_BYTES  (4096 - sizeof(struct o2net_msg))
+
+/* same as hb delay, we're waiting for another node to recognize our hb */
+#define O2NET_RECONNECT_DELAY_MS_DEFAULT	2000
+
+#define O2NET_KEEPALIVE_DELAY_MS_DEFAULT	2000
+#define O2NET_IDLE_TIMEOUT_MS_DEFAULT		30000
+
+#define O2NET_TCP_USER_TIMEOUT			0x7fffffff
+
+/* TODO: figure this out.... */
+static inline int o2net_link_down(int err, struct socket *sock)
+{
+	if (sock) {
+		if (sock->sk->sk_state != TCP_ESTABLISHED &&
+	    	    sock->sk->sk_state != TCP_CLOSE_WAIT)
+			return 1;
+	}
+
+	if (err >= 0)
+		return 0;
+	switch (err) {
+		/* ????????????????????????? */
+		case -ERESTARTSYS:
+		case -EBADF:
+		/* When the server has died, an ICMP port unreachable
+		 * message prompts ECONNREFUSED. */
+		case -ECONNREFUSED:
+		case -ENOTCONN:
+		case -ECONNRESET:
+		case -EPIPE:
+			return 1;
+	}
+	return 0;
+}
+
+enum {
+	O2NET_DRIVER_UNINITED,
+	O2NET_DRIVER_READY,
+};
+
+int o2net_send_message(u32 msg_type, u32 key, void *data, u32 len,
+		       u8 target_node, int *status);
+int o2net_send_message_vec(u32 msg_type, u32 key, struct kvec *vec,
+			   size_t veclen, u8 target_node, int *status);
+
+int o2net_register_handler(u32 msg_type, u32 key, u32 max_len,
+			   o2net_msg_handler_func *func, void *data,
+			   o2net_post_msg_handler_func *post_func,
+			   struct list_head *unreg_list);
+void o2net_unregister_handler_list(struct list_head *list);
+
+void o2net_fill_node_map(unsigned long *map, unsigned bytes);
+
+struct o2nm_node;
+int o2net_register_hb_callbacks(void);
+void o2net_unregister_hb_callbacks(void);
+int o2net_start_listening(struct o2nm_node *node);
+void o2net_stop_listening(struct o2nm_node *node);
+void o2net_disconnect_node(struct o2nm_node *node);
+int o2net_num_connected_peers(void);
+
+int o2net_init(void);
+void o2net_exit(void);
+
+struct o2net_send_tracking;
+struct o2net_sock_container;
+
+#ifdef CONFIG_DEBUG_FS
+int o2net_debugfs_init(void);
+void o2net_debugfs_exit(void);
+void o2net_debug_add_nst(struct o2net_send_tracking *nst);
+void o2net_debug_del_nst(struct o2net_send_tracking *nst);
+void o2net_debug_add_sc(struct o2net_sock_container *sc);
+void o2net_debug_del_sc(struct o2net_sock_container *sc);
+#else
+static inline int o2net_debugfs_init(void)
+{
+	return 0;
+}
+static inline void o2net_debugfs_exit(void)
+{
+}
+static inline void o2net_debug_add_nst(struct o2net_send_tracking *nst)
+{
+}
+static inline void o2net_debug_del_nst(struct o2net_send_tracking *nst)
+{
+}
+static inline void o2net_debug_add_sc(struct o2net_sock_container *sc)
+{
+}
+static inline void o2net_debug_del_sc(struct o2net_sock_container *sc)
+{
+}
+#endif	/* CONFIG_DEBUG_FS */
+
+#endif /* O2CLUSTER_TCP_H */
diff --git a/fs/ocfs2/cluster/tcp_internal.h b/fs/ocfs2/cluster/tcp_internal.h
new file mode 100644
index 000000000..b95e7df5b
--- /dev/null
+++ b/fs/ocfs2/cluster/tcp_internal.h
@@ -0,0 +1,242 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * Copyright (C) 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef O2CLUSTER_TCP_INTERNAL_H
+#define O2CLUSTER_TCP_INTERNAL_H
+
+#define O2NET_MSG_MAGIC           ((u16)0xfa55)
+#define O2NET_MSG_STATUS_MAGIC    ((u16)0xfa56)
+#define O2NET_MSG_KEEP_REQ_MAGIC  ((u16)0xfa57)
+#define O2NET_MSG_KEEP_RESP_MAGIC ((u16)0xfa58)
+
+/* we're delaying our quorum decision so that heartbeat will have timed
+ * out truly dead nodes by the time we come around to making decisions
+ * on their number */
+#define O2NET_QUORUM_DELAY_MS	((o2hb_dead_threshold + 2) * O2HB_REGION_TIMEOUT_MS)
+
+/*
+ * This version number represents quite a lot, unfortunately.  It not
+ * only represents the raw network message protocol on the wire but also
+ * locking semantics of the file system using the protocol.  It should
+ * be somewhere else, I'm sure, but right now it isn't.
+ *
+ * With version 11, we separate out the filesystem locking portion.  The
+ * filesystem now has a major.minor version it negotiates.  Version 11
+ * introduces this negotiation to the o2dlm protocol, and as such the
+ * version here in tcp_internal.h should not need to be bumped for
+ * filesystem locking changes.
+ *
+ * New in version 11
+ * 	- Negotiation of filesystem locking in the dlm join.
+ *
+ * New in version 10:
+ * 	- Meta/data locks combined
+ *
+ * New in version 9:
+ * 	- All votes removed
+ *
+ * New in version 8:
+ * 	- Replace delete inode votes with a cluster lock
+ *
+ * New in version 7:
+ * 	- DLM join domain includes the live nodemap
+ *
+ * New in version 6:
+ * 	- DLM lockres remote refcount fixes.
+ *
+ * New in version 5:
+ * 	- Network timeout checking protocol
+ *
+ * New in version 4:
+ * 	- Remove i_generation from lock names for better stat performance.
+ *
+ * New in version 3:
+ * 	- Replace dentry votes with a cluster lock
+ *
+ * New in version 2:
+ * 	- full 64 bit i_size in the metadata lock lvbs
+ * 	- introduction of "rw" lock and pushing meta/data locking down
+ */
+#define O2NET_PROTOCOL_VERSION 11ULL
+struct o2net_handshake {
+	__be64	protocol_version;
+	__be64	connector_id;
+	__be32  o2hb_heartbeat_timeout_ms;
+	__be32  o2net_idle_timeout_ms;
+	__be32  o2net_keepalive_delay_ms;
+	__be32  o2net_reconnect_delay_ms;
+};
+
+struct o2net_node {
+	/* this is never called from int/bh */
+	spinlock_t			nn_lock;
+
+	/* set the moment an sc is allocated and a connect is started */
+	struct o2net_sock_container	*nn_sc;
+	/* _valid is only set after the handshake passes and tx can happen */
+	unsigned			nn_sc_valid:1;
+	/* if this is set tx just returns it */
+	int				nn_persistent_error;
+	/* It is only set to 1 after the idle time out. */
+	atomic_t			nn_timeout;
+
+	/* threads waiting for an sc to arrive wait on the wq for generation
+	 * to increase.  it is increased when a connecting socket succeeds
+	 * or fails or when an accepted socket is attached. */
+	wait_queue_head_t		nn_sc_wq;
+
+	struct idr			nn_status_idr;
+	struct list_head		nn_status_list;
+
+	/* connects are attempted from when heartbeat comes up until either hb
+	 * goes down, the node is unconfigured, or a connect succeeds.
+	 * connect_work is queued from set_nn_state both from hb up and from
+	 * itself if a connect attempt fails and so can be self-arming.
+	 * shutdown is careful to first mark the nn such that no connects will
+	 * be attempted before canceling delayed connect work and flushing the
+	 * queue. */
+	struct delayed_work		nn_connect_work;
+	unsigned long			nn_last_connect_attempt;
+
+	/* this is queued as nodes come up and is canceled when a connection is
+	 * established.  this expiring gives up on the node and errors out
+	 * transmits */
+	struct delayed_work		nn_connect_expired;
+
+	/* after we give up on a socket we wait a while before deciding
+	 * that it is still heartbeating and that we should do some
+	 * quorum work */
+	struct delayed_work		nn_still_up;
+};
+
+struct o2net_sock_container {
+	struct kref		sc_kref;
+	/* the next two are valid for the life time of the sc */
+	struct socket		*sc_sock;
+	struct o2nm_node	*sc_node;
+
+	/* all of these sc work structs hold refs on the sc while they are
+	 * queued.  they should not be able to ref a freed sc.  the teardown
+	 * race is with o2net_wq destruction in o2net_stop_listening() */
+
+	/* rx and connect work are generated from socket callbacks.  sc
+	 * shutdown removes the callbacks and then flushes the work queue */
+	struct work_struct	sc_rx_work;
+	struct work_struct	sc_connect_work;
+	/* shutdown work is triggered in two ways.  the simple way is
+	 * for a code path calls ensure_shutdown which gets a lock, removes
+	 * the sc from the nn, and queues the work.  in this case the
+	 * work is single-shot.  the work is also queued from a sock
+	 * callback, though, and in this case the work will find the sc
+	 * still on the nn and will call ensure_shutdown itself.. this
+	 * ends up triggering the shutdown work again, though nothing
+	 * will be done in that second iteration.  so work queue teardown
+	 * has to be careful to remove the sc from the nn before waiting
+	 * on the work queue so that the shutdown work doesn't remove the
+	 * sc and rearm itself.
+	 */
+	struct work_struct	sc_shutdown_work;
+
+	struct timer_list	sc_idle_timeout;
+	struct delayed_work	sc_keepalive_work;
+
+	unsigned		sc_handshake_ok:1;
+
+	struct page 		*sc_page;
+	size_t			sc_page_off;
+
+	/* original handlers for the sockets */
+	void			(*sc_state_change)(struct sock *sk);
+	void			(*sc_data_ready)(struct sock *sk);
+
+	u32			sc_msg_key;
+	u16			sc_msg_type;
+
+#ifdef CONFIG_DEBUG_FS
+	struct list_head        sc_net_debug_item;
+	ktime_t			sc_tv_timer;
+	ktime_t			sc_tv_data_ready;
+	ktime_t			sc_tv_advance_start;
+	ktime_t			sc_tv_advance_stop;
+	ktime_t			sc_tv_func_start;
+	ktime_t			sc_tv_func_stop;
+#endif
+#ifdef CONFIG_OCFS2_FS_STATS
+	ktime_t			sc_tv_acquiry_total;
+	ktime_t			sc_tv_send_total;
+	ktime_t			sc_tv_status_total;
+	u32			sc_send_count;
+	u32			sc_recv_count;
+	ktime_t			sc_tv_process_total;
+#endif
+	struct mutex		sc_send_lock;
+};
+
+struct o2net_msg_handler {
+	struct rb_node		nh_node;
+	u32			nh_max_len;
+	u32			nh_msg_type;
+	u32			nh_key;
+	o2net_msg_handler_func	*nh_func;
+	o2net_msg_handler_func	*nh_func_data;
+	o2net_post_msg_handler_func
+				*nh_post_func;
+	struct kref		nh_kref;
+	struct list_head	nh_unregister_item;
+};
+
+enum o2net_system_error {
+	O2NET_ERR_NONE = 0,
+	O2NET_ERR_NO_HNDLR,
+	O2NET_ERR_OVERFLOW,
+	O2NET_ERR_DIED,
+	O2NET_ERR_MAX
+};
+
+struct o2net_status_wait {
+	enum o2net_system_error	ns_sys_status;
+	s32			ns_status;
+	int			ns_id;
+	wait_queue_head_t	ns_wq;
+	struct list_head	ns_node_item;
+};
+
+#ifdef CONFIG_DEBUG_FS
+/* just for state dumps */
+struct o2net_send_tracking {
+	struct list_head		st_net_debug_item;
+	struct task_struct		*st_task;
+	struct o2net_sock_container	*st_sc;
+	u32				st_id;
+	u32				st_msg_type;
+	u32				st_msg_key;
+	u8				st_node;
+	ktime_t				st_sock_time;
+	ktime_t				st_send_time;
+	ktime_t				st_status_time;
+};
+#else
+struct o2net_send_tracking {
+	u32	dummy;
+};
+#endif	/* CONFIG_DEBUG_FS */
+
+#endif /* O2CLUSTER_TCP_INTERNAL_H */
diff --git a/fs/ocfs2/dcache.c b/fs/ocfs2/dcache.c
new file mode 100644
index 000000000..290373024
--- /dev/null
+++ b/fs/ocfs2/dcache.c
@@ -0,0 +1,474 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dcache.c
+ *
+ * dentry cache handling code
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/namei.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "dcache.h"
+#include "dlmglue.h"
+#include "file.h"
+#include "inode.h"
+#include "ocfs2_trace.h"
+
+void ocfs2_dentry_attach_gen(struct dentry *dentry)
+{
+	unsigned long gen =
+		OCFS2_I(d_inode(dentry->d_parent))->ip_dir_lock_gen;
+	BUG_ON(d_inode(dentry));
+	dentry->d_fsdata = (void *)gen;
+}
+
+
+static int ocfs2_dentry_revalidate(struct dentry *dentry, unsigned int flags)
+{
+	struct inode *inode;
+	int ret = 0;    /* if all else fails, just return false */
+	struct ocfs2_super *osb;
+
+	if (flags & LOOKUP_RCU)
+		return -ECHILD;
+
+	inode = d_inode(dentry);
+	osb = OCFS2_SB(dentry->d_sb);
+
+	trace_ocfs2_dentry_revalidate(dentry, dentry->d_name.len,
+				      dentry->d_name.name);
+
+	/* For a negative dentry -
+	 * check the generation number of the parent and compare with the
+	 * one stored in the inode.
+	 */
+	if (inode == NULL) {
+		unsigned long gen = (unsigned long) dentry->d_fsdata;
+		unsigned long pgen;
+		spin_lock(&dentry->d_lock);
+		pgen = OCFS2_I(d_inode(dentry->d_parent))->ip_dir_lock_gen;
+		spin_unlock(&dentry->d_lock);
+		trace_ocfs2_dentry_revalidate_negative(dentry->d_name.len,
+						       dentry->d_name.name,
+						       pgen, gen);
+		if (gen != pgen)
+			goto bail;
+		goto valid;
+	}
+
+	BUG_ON(!osb);
+
+	if (inode == osb->root_inode || is_bad_inode(inode))
+		goto bail;
+
+	spin_lock(&OCFS2_I(inode)->ip_lock);
+	/* did we or someone else delete this inode? */
+	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
+		spin_unlock(&OCFS2_I(inode)->ip_lock);
+		trace_ocfs2_dentry_revalidate_delete(
+				(unsigned long long)OCFS2_I(inode)->ip_blkno);
+		goto bail;
+	}
+	spin_unlock(&OCFS2_I(inode)->ip_lock);
+
+	/*
+	 * We don't need a cluster lock to test this because once an
+	 * inode nlink hits zero, it never goes back.
+	 */
+	if (inode->i_nlink == 0) {
+		trace_ocfs2_dentry_revalidate_orphaned(
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			S_ISDIR(inode->i_mode));
+		goto bail;
+	}
+
+	/*
+	 * If the last lookup failed to create dentry lock, let us
+	 * redo it.
+	 */
+	if (!dentry->d_fsdata) {
+		trace_ocfs2_dentry_revalidate_nofsdata(
+				(unsigned long long)OCFS2_I(inode)->ip_blkno);
+		goto bail;
+	}
+
+valid:
+	ret = 1;
+
+bail:
+	trace_ocfs2_dentry_revalidate_ret(ret);
+	return ret;
+}
+
+static int ocfs2_match_dentry(struct dentry *dentry,
+			      u64 parent_blkno,
+			      int skip_unhashed)
+{
+	struct inode *parent;
+
+	/*
+	 * ocfs2_lookup() does a d_splice_alias() _before_ attaching
+	 * to the lock data, so we skip those here, otherwise
+	 * ocfs2_dentry_attach_lock() will get its original dentry
+	 * back.
+	 */
+	if (!dentry->d_fsdata)
+		return 0;
+
+	if (!dentry->d_parent)
+		return 0;
+
+	if (skip_unhashed && d_unhashed(dentry))
+		return 0;
+
+	parent = d_inode(dentry->d_parent);
+	/* Negative parent dentry? */
+	if (!parent)
+		return 0;
+
+	/* Name is in a different directory. */
+	if (OCFS2_I(parent)->ip_blkno != parent_blkno)
+		return 0;
+
+	return 1;
+}
+
+/*
+ * Walk the inode alias list, and find a dentry which has a given
+ * parent. ocfs2_dentry_attach_lock() wants to find _any_ alias as it
+ * is looking for a dentry_lock reference. The downconvert thread is
+ * looking to unhash aliases, so we allow it to skip any that already
+ * have that property.
+ */
+struct dentry *ocfs2_find_local_alias(struct inode *inode,
+				      u64 parent_blkno,
+				      int skip_unhashed)
+{
+	struct dentry *dentry;
+
+	spin_lock(&inode->i_lock);
+	hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
+		spin_lock(&dentry->d_lock);
+		if (ocfs2_match_dentry(dentry, parent_blkno, skip_unhashed)) {
+			trace_ocfs2_find_local_alias(dentry->d_name.len,
+						     dentry->d_name.name);
+
+			dget_dlock(dentry);
+			spin_unlock(&dentry->d_lock);
+			spin_unlock(&inode->i_lock);
+			return dentry;
+		}
+		spin_unlock(&dentry->d_lock);
+	}
+	spin_unlock(&inode->i_lock);
+	return NULL;
+}
+
+DEFINE_SPINLOCK(dentry_attach_lock);
+
+/*
+ * Attach this dentry to a cluster lock.
+ *
+ * Dentry locks cover all links in a given directory to a particular
+ * inode. We do this so that ocfs2 can build a lock name which all
+ * nodes in the cluster can agree on at all times. Shoving full names
+ * in the cluster lock won't work due to size restrictions. Covering
+ * links inside of a directory is a good compromise because it still
+ * allows us to use the parent directory lock to synchronize
+ * operations.
+ *
+ * Call this function with the parent dir semaphore and the parent dir
+ * cluster lock held.
+ *
+ * The dir semaphore will protect us from having to worry about
+ * concurrent processes on our node trying to attach a lock at the
+ * same time.
+ *
+ * The dir cluster lock (held at either PR or EX mode) protects us
+ * from unlink and rename on other nodes.
+ *
+ * A dput() can happen asynchronously due to pruning, so we cover
+ * attaching and detaching the dentry lock with a
+ * dentry_attach_lock.
+ *
+ * A node which has done lookup on a name retains a protected read
+ * lock until final dput. If the user requests and unlink or rename,
+ * the protected read is upgraded to an exclusive lock. Other nodes
+ * who have seen the dentry will then be informed that they need to
+ * downgrade their lock, which will involve d_delete on the
+ * dentry. This happens in ocfs2_dentry_convert_worker().
+ */
+int ocfs2_dentry_attach_lock(struct dentry *dentry,
+			     struct inode *inode,
+			     u64 parent_blkno)
+{
+	int ret;
+	struct dentry *alias;
+	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
+
+	trace_ocfs2_dentry_attach_lock(dentry->d_name.len, dentry->d_name.name,
+				       (unsigned long long)parent_blkno, dl);
+
+	/*
+	 * Negative dentry. We ignore these for now.
+	 *
+	 * XXX: Could we can improve ocfs2_dentry_revalidate() by
+	 * tracking these?
+	 */
+	if (!inode)
+		return 0;
+
+	if (d_really_is_negative(dentry) && dentry->d_fsdata) {
+		/* Converting a negative dentry to positive
+		   Clear dentry->d_fsdata */
+		dentry->d_fsdata = dl = NULL;
+	}
+
+	if (dl) {
+		mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
+				" \"%pd\": old parent: %llu, new: %llu\n",
+				dentry,
+				(unsigned long long)parent_blkno,
+				(unsigned long long)dl->dl_parent_blkno);
+		return 0;
+	}
+
+	alias = ocfs2_find_local_alias(inode, parent_blkno, 0);
+	if (alias) {
+		/*
+		 * Great, an alias exists, which means we must have a
+		 * dentry lock already. We can just grab the lock off
+		 * the alias and add it to the list.
+		 *
+		 * We're depending here on the fact that this dentry
+		 * was found and exists in the dcache and so must have
+		 * a reference to the dentry_lock because we can't
+		 * race creates. Final dput() cannot happen on it
+		 * since we have it pinned, so our reference is safe.
+		 */
+		dl = alias->d_fsdata;
+		mlog_bug_on_msg(!dl, "parent %llu, ino %llu\n",
+				(unsigned long long)parent_blkno,
+				(unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+		mlog_bug_on_msg(dl->dl_parent_blkno != parent_blkno,
+				" \"%pd\": old parent: %llu, new: %llu\n",
+				dentry,
+				(unsigned long long)parent_blkno,
+				(unsigned long long)dl->dl_parent_blkno);
+
+		trace_ocfs2_dentry_attach_lock_found(dl->dl_lockres.l_name,
+				(unsigned long long)parent_blkno,
+				(unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+		goto out_attach;
+	}
+
+	/*
+	 * There are no other aliases
+	 */
+	dl = kmalloc(sizeof(*dl), GFP_NOFS);
+	if (!dl) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		return ret;
+	}
+
+	dl->dl_count = 0;
+	/*
+	 * Does this have to happen below, for all attaches, in case
+	 * the struct inode gets blown away by the downconvert thread?
+	 */
+	dl->dl_inode = igrab(inode);
+	dl->dl_parent_blkno = parent_blkno;
+	ocfs2_dentry_lock_res_init(dl, parent_blkno, inode);
+
+out_attach:
+	spin_lock(&dentry_attach_lock);
+	dentry->d_fsdata = dl;
+	dl->dl_count++;
+	spin_unlock(&dentry_attach_lock);
+
+	/*
+	 * This actually gets us our PRMODE level lock. From now on,
+	 * we'll have a notification if one of these names is
+	 * destroyed on another node.
+	 */
+	ret = ocfs2_dentry_lock(dentry, 0);
+	if (!ret)
+		ocfs2_dentry_unlock(dentry, 0);
+	else
+		mlog_errno(ret);
+
+	/*
+	 * In case of error, manually free the allocation and do the iput().
+	 * We need to do this because error here means no d_instantiate(),
+	 * which means iput() will not be called during dput(dentry).
+	 */
+	if (ret < 0 && !alias) {
+		ocfs2_lock_res_free(&dl->dl_lockres);
+		BUG_ON(dl->dl_count != 1);
+		spin_lock(&dentry_attach_lock);
+		dentry->d_fsdata = NULL;
+		spin_unlock(&dentry_attach_lock);
+		kfree(dl);
+		iput(inode);
+	}
+
+	dput(alias);
+
+	return ret;
+}
+
+/*
+ * ocfs2_dentry_iput() and friends.
+ *
+ * At this point, our particular dentry is detached from the inodes
+ * alias list, so there's no way that the locking code can find it.
+ *
+ * The interesting stuff happens when we determine that our lock needs
+ * to go away because this is the last subdir alias in the
+ * system. This function needs to handle a couple things:
+ *
+ * 1) Synchronizing lock shutdown with the downconvert threads. This
+ *    is already handled for us via the lockres release drop function
+ *    called in ocfs2_release_dentry_lock()
+ *
+ * 2) A race may occur when we're doing our lock shutdown and
+ *    another process wants to create a new dentry lock. Right now we
+ *    let them race, which means that for a very short while, this
+ *    node might have two locks on a lock resource. This should be a
+ *    problem though because one of them is in the process of being
+ *    thrown out.
+ */
+static void ocfs2_drop_dentry_lock(struct ocfs2_super *osb,
+				   struct ocfs2_dentry_lock *dl)
+{
+	iput(dl->dl_inode);
+	ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
+	ocfs2_lock_res_free(&dl->dl_lockres);
+	kfree(dl);
+}
+
+void ocfs2_dentry_lock_put(struct ocfs2_super *osb,
+			   struct ocfs2_dentry_lock *dl)
+{
+	int unlock = 0;
+
+	BUG_ON(dl->dl_count == 0);
+
+	spin_lock(&dentry_attach_lock);
+	dl->dl_count--;
+	unlock = !dl->dl_count;
+	spin_unlock(&dentry_attach_lock);
+
+	if (unlock)
+		ocfs2_drop_dentry_lock(osb, dl);
+}
+
+static void ocfs2_dentry_iput(struct dentry *dentry, struct inode *inode)
+{
+	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
+
+	if (!dl) {
+		/*
+		 * No dentry lock is ok if we're disconnected or
+		 * unhashed.
+		 */
+		if (!(dentry->d_flags & DCACHE_DISCONNECTED) &&
+		    !d_unhashed(dentry)) {
+			unsigned long long ino = 0ULL;
+			if (inode)
+				ino = (unsigned long long)OCFS2_I(inode)->ip_blkno;
+			mlog(ML_ERROR, "Dentry is missing cluster lock. "
+			     "inode: %llu, d_flags: 0x%x, d_name: %pd\n",
+			     ino, dentry->d_flags, dentry);
+		}
+
+		goto out;
+	}
+
+	mlog_bug_on_msg(dl->dl_count == 0, "dentry: %pd, count: %u\n",
+			dentry, dl->dl_count);
+
+	ocfs2_dentry_lock_put(OCFS2_SB(dentry->d_sb), dl);
+
+out:
+	iput(inode);
+}
+
+/*
+ * d_move(), but keep the locks in sync.
+ *
+ * When we are done, "dentry" will have the parent dir and name of
+ * "target", which will be thrown away.
+ *
+ * We manually update the lock of "dentry" if need be.
+ *
+ * "target" doesn't have it's dentry lock touched - we allow the later
+ * dput() to handle this for us.
+ *
+ * This is called during ocfs2_rename(), while holding parent
+ * directory locks. The dentries have already been deleted on other
+ * nodes via ocfs2_remote_dentry_delete().
+ *
+ * Normally, the VFS handles the d_move() for the file system, after
+ * the ->rename() callback. OCFS2 wants to handle this internally, so
+ * the new lock can be created atomically with respect to the cluster.
+ */
+void ocfs2_dentry_move(struct dentry *dentry, struct dentry *target,
+		       struct inode *old_dir, struct inode *new_dir)
+{
+	int ret;
+	struct ocfs2_super *osb = OCFS2_SB(old_dir->i_sb);
+	struct inode *inode = d_inode(dentry);
+
+	/*
+	 * Move within the same directory, so the actual lock info won't
+	 * change.
+	 *
+	 * XXX: Is there any advantage to dropping the lock here?
+	 */
+	if (old_dir == new_dir)
+		goto out_move;
+
+	ocfs2_dentry_lock_put(osb, dentry->d_fsdata);
+
+	dentry->d_fsdata = NULL;
+	ret = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(new_dir)->ip_blkno);
+	if (ret)
+		mlog_errno(ret);
+
+out_move:
+	d_move(dentry, target);
+}
+
+const struct dentry_operations ocfs2_dentry_ops = {
+	.d_revalidate		= ocfs2_dentry_revalidate,
+	.d_iput			= ocfs2_dentry_iput,
+};
diff --git a/fs/ocfs2/dcache.h b/fs/ocfs2/dcache.h
new file mode 100644
index 000000000..55f58892b
--- /dev/null
+++ b/fs/ocfs2/dcache.h
@@ -0,0 +1,59 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dcache.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_DCACHE_H
+#define OCFS2_DCACHE_H
+
+extern const struct dentry_operations ocfs2_dentry_ops;
+
+struct ocfs2_dentry_lock {
+	unsigned int		dl_count;
+	u64			dl_parent_blkno;
+
+	/*
+	 * The ocfs2_dentry_lock keeps an inode reference until
+	 * dl_lockres has been destroyed. This is usually done in
+	 * ->d_iput() anyway, so there should be minimal impact.
+	 */
+	struct inode		*dl_inode;
+	struct ocfs2_lock_res	dl_lockres;
+};
+
+int ocfs2_dentry_attach_lock(struct dentry *dentry, struct inode *inode,
+			     u64 parent_blkno);
+
+void ocfs2_dentry_lock_put(struct ocfs2_super *osb,
+			   struct ocfs2_dentry_lock *dl);
+
+struct dentry *ocfs2_find_local_alias(struct inode *inode, u64 parent_blkno,
+				      int skip_unhashed);
+
+void ocfs2_dentry_move(struct dentry *dentry, struct dentry *target,
+		       struct inode *old_dir, struct inode *new_dir);
+
+extern spinlock_t dentry_attach_lock;
+void ocfs2_dentry_attach_gen(struct dentry *dentry);
+
+#endif /* OCFS2_DCACHE_H */
diff --git a/fs/ocfs2/dir.c b/fs/ocfs2/dir.c
new file mode 100644
index 000000000..ccd4dcfc3
--- /dev/null
+++ b/fs/ocfs2/dir.c
@@ -0,0 +1,4501 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dir.c
+ *
+ * Creates, reads, walks and deletes directory-nodes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ *  Portions of this code from linux/fs/ext3/dir.c
+ *
+ *  Copyright (C) 1992, 1993, 1994, 1995
+ *  Remy Card (card@masi.ibp.fr)
+ *  Laboratoire MASI - Institut Blaise pascal
+ *  Universite Pierre et Marie Curie (Paris VI)
+ *
+ *   from
+ *
+ *   linux/fs/minix/dir.c
+ *
+ *   Copyright (C) 1991, 1992 Linus Torvalds
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/quotaops.h>
+#include <linux/sort.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "blockcheck.h"
+#include "dir.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "file.h"
+#include "inode.h"
+#include "journal.h"
+#include "namei.h"
+#include "suballoc.h"
+#include "super.h"
+#include "sysfile.h"
+#include "uptodate.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+#define NAMEI_RA_CHUNKS  2
+#define NAMEI_RA_BLOCKS  4
+#define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
+
+static unsigned char ocfs2_filetype_table[] = {
+	DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
+};
+
+static int ocfs2_do_extend_dir(struct super_block *sb,
+			       handle_t *handle,
+			       struct inode *dir,
+			       struct buffer_head *parent_fe_bh,
+			       struct ocfs2_alloc_context *data_ac,
+			       struct ocfs2_alloc_context *meta_ac,
+			       struct buffer_head **new_bh);
+static int ocfs2_dir_indexed(struct inode *inode);
+
+/*
+ * These are distinct checks because future versions of the file system will
+ * want to have a trailing dirent structure independent of indexing.
+ */
+static int ocfs2_supports_dir_trailer(struct inode *dir)
+{
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+
+	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		return 0;
+
+	return ocfs2_meta_ecc(osb) || ocfs2_dir_indexed(dir);
+}
+
+/*
+ * "new' here refers to the point at which we're creating a new
+ * directory via "mkdir()", but also when we're expanding an inline
+ * directory. In either case, we don't yet have the indexing bit set
+ * on the directory, so the standard checks will fail in when metaecc
+ * is turned off. Only directory-initialization type functions should
+ * use this then. Everything else wants ocfs2_supports_dir_trailer()
+ */
+static int ocfs2_new_dir_wants_trailer(struct inode *dir)
+{
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+
+	return ocfs2_meta_ecc(osb) ||
+		ocfs2_supports_indexed_dirs(osb);
+}
+
+static inline unsigned int ocfs2_dir_trailer_blk_off(struct super_block *sb)
+{
+	return sb->s_blocksize - sizeof(struct ocfs2_dir_block_trailer);
+}
+
+#define ocfs2_trailer_from_bh(_bh, _sb) ((struct ocfs2_dir_block_trailer *) ((_bh)->b_data + ocfs2_dir_trailer_blk_off((_sb))))
+
+/* XXX ocfs2_block_dqtrailer() is similar but not quite - can we make
+ * them more consistent? */
+struct ocfs2_dir_block_trailer *ocfs2_dir_trailer_from_size(int blocksize,
+							    void *data)
+{
+	char *p = data;
+
+	p += blocksize - sizeof(struct ocfs2_dir_block_trailer);
+	return (struct ocfs2_dir_block_trailer *)p;
+}
+
+/*
+ * XXX: This is executed once on every dirent. We should consider optimizing
+ * it.
+ */
+static int ocfs2_skip_dir_trailer(struct inode *dir,
+				  struct ocfs2_dir_entry *de,
+				  unsigned long offset,
+				  unsigned long blklen)
+{
+	unsigned long toff = blklen - sizeof(struct ocfs2_dir_block_trailer);
+
+	if (!ocfs2_supports_dir_trailer(dir))
+		return 0;
+
+	if (offset != toff)
+		return 0;
+
+	return 1;
+}
+
+static void ocfs2_init_dir_trailer(struct inode *inode,
+				   struct buffer_head *bh, u16 rec_len)
+{
+	struct ocfs2_dir_block_trailer *trailer;
+
+	trailer = ocfs2_trailer_from_bh(bh, inode->i_sb);
+	strcpy(trailer->db_signature, OCFS2_DIR_TRAILER_SIGNATURE);
+	trailer->db_compat_rec_len =
+			cpu_to_le16(sizeof(struct ocfs2_dir_block_trailer));
+	trailer->db_parent_dinode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
+	trailer->db_blkno = cpu_to_le64(bh->b_blocknr);
+	trailer->db_free_rec_len = cpu_to_le16(rec_len);
+}
+/*
+ * Link an unindexed block with a dir trailer structure into the index free
+ * list. This function will modify dirdata_bh, but assumes you've already
+ * passed it to the journal.
+ */
+static int ocfs2_dx_dir_link_trailer(struct inode *dir, handle_t *handle,
+				     struct buffer_head *dx_root_bh,
+				     struct buffer_head *dirdata_bh)
+{
+	int ret;
+	struct ocfs2_dx_root_block *dx_root;
+	struct ocfs2_dir_block_trailer *trailer;
+
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	trailer = ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+
+	trailer->db_free_next = dx_root->dr_free_blk;
+	dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
+
+	ocfs2_journal_dirty(handle, dx_root_bh);
+
+out:
+	return ret;
+}
+
+static int ocfs2_free_list_at_root(struct ocfs2_dir_lookup_result *res)
+{
+	return res->dl_prev_leaf_bh == NULL;
+}
+
+void ocfs2_free_dir_lookup_result(struct ocfs2_dir_lookup_result *res)
+{
+	brelse(res->dl_dx_root_bh);
+	brelse(res->dl_leaf_bh);
+	brelse(res->dl_dx_leaf_bh);
+	brelse(res->dl_prev_leaf_bh);
+}
+
+static int ocfs2_dir_indexed(struct inode *inode)
+{
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INDEXED_DIR_FL)
+		return 1;
+	return 0;
+}
+
+static inline int ocfs2_dx_root_inline(struct ocfs2_dx_root_block *dx_root)
+{
+	return dx_root->dr_flags & OCFS2_DX_FLAG_INLINE;
+}
+
+/*
+ * Hashing code adapted from ext3
+ */
+#define DELTA 0x9E3779B9
+
+static void TEA_transform(__u32 buf[4], __u32 const in[])
+{
+	__u32	sum = 0;
+	__u32	b0 = buf[0], b1 = buf[1];
+	__u32	a = in[0], b = in[1], c = in[2], d = in[3];
+	int	n = 16;
+
+	do {
+		sum += DELTA;
+		b0 += ((b1 << 4)+a) ^ (b1+sum) ^ ((b1 >> 5)+b);
+		b1 += ((b0 << 4)+c) ^ (b0+sum) ^ ((b0 >> 5)+d);
+	} while (--n);
+
+	buf[0] += b0;
+	buf[1] += b1;
+}
+
+static void str2hashbuf(const char *msg, int len, __u32 *buf, int num)
+{
+	__u32	pad, val;
+	int	i;
+
+	pad = (__u32)len | ((__u32)len << 8);
+	pad |= pad << 16;
+
+	val = pad;
+	if (len > num*4)
+		len = num * 4;
+	for (i = 0; i < len; i++) {
+		if ((i % 4) == 0)
+			val = pad;
+		val = msg[i] + (val << 8);
+		if ((i % 4) == 3) {
+			*buf++ = val;
+			val = pad;
+			num--;
+		}
+	}
+	if (--num >= 0)
+		*buf++ = val;
+	while (--num >= 0)
+		*buf++ = pad;
+}
+
+static void ocfs2_dx_dir_name_hash(struct inode *dir, const char *name, int len,
+				   struct ocfs2_dx_hinfo *hinfo)
+{
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	const char	*p;
+	__u32		in[8], buf[4];
+
+	/*
+	 * XXX: Is this really necessary, if the index is never looked
+	 * at by readdir? Is a hash value of '0' a bad idea?
+	 */
+	if ((len == 1 && !strncmp(".", name, 1)) ||
+	    (len == 2 && !strncmp("..", name, 2))) {
+		buf[0] = buf[1] = 0;
+		goto out;
+	}
+
+#ifdef OCFS2_DEBUG_DX_DIRS
+	/*
+	 * This makes it very easy to debug indexing problems. We
+	 * should never allow this to be selected without hand editing
+	 * this file though.
+	 */
+	buf[0] = buf[1] = len;
+	goto out;
+#endif
+
+	memcpy(buf, osb->osb_dx_seed, sizeof(buf));
+
+	p = name;
+	while (len > 0) {
+		str2hashbuf(p, len, in, 4);
+		TEA_transform(buf, in);
+		len -= 16;
+		p += 16;
+	}
+
+out:
+	hinfo->major_hash = buf[0];
+	hinfo->minor_hash = buf[1];
+}
+
+/*
+ * bh passed here can be an inode block or a dir data block, depending
+ * on the inode inline data flag.
+ */
+static int ocfs2_check_dir_entry(struct inode * dir,
+				 struct ocfs2_dir_entry * de,
+				 struct buffer_head * bh,
+				 unsigned long offset)
+{
+	const char *error_msg = NULL;
+	const int rlen = le16_to_cpu(de->rec_len);
+
+	if (unlikely(rlen < OCFS2_DIR_REC_LEN(1)))
+		error_msg = "rec_len is smaller than minimal";
+	else if (unlikely(rlen % 4 != 0))
+		error_msg = "rec_len % 4 != 0";
+	else if (unlikely(rlen < OCFS2_DIR_REC_LEN(de->name_len)))
+		error_msg = "rec_len is too small for name_len";
+	else if (unlikely(
+		 ((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize))
+		error_msg = "directory entry across blocks";
+
+	if (unlikely(error_msg != NULL))
+		mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
+		     "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
+		     (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
+		     offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
+		     de->name_len);
+
+	return error_msg == NULL ? 1 : 0;
+}
+
+static inline int ocfs2_match(int len,
+			      const char * const name,
+			      struct ocfs2_dir_entry *de)
+{
+	if (len != de->name_len)
+		return 0;
+	if (!de->inode)
+		return 0;
+	return !memcmp(name, de->name, len);
+}
+
+/*
+ * Returns 0 if not found, -1 on failure, and 1 on success
+ */
+static inline int ocfs2_search_dirblock(struct buffer_head *bh,
+					struct inode *dir,
+					const char *name, int namelen,
+					unsigned long offset,
+					char *first_de,
+					unsigned int bytes,
+					struct ocfs2_dir_entry **res_dir)
+{
+	struct ocfs2_dir_entry *de;
+	char *dlimit, *de_buf;
+	int de_len;
+	int ret = 0;
+
+	de_buf = first_de;
+	dlimit = de_buf + bytes;
+
+	while (de_buf < dlimit) {
+		/* this code is executed quadratically often */
+		/* do minimal checking `by hand' */
+
+		de = (struct ocfs2_dir_entry *) de_buf;
+
+		if (de_buf + namelen <= dlimit &&
+		    ocfs2_match(namelen, name, de)) {
+			/* found a match - just to be sure, do a full check */
+			if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
+				ret = -1;
+				goto bail;
+			}
+			*res_dir = de;
+			ret = 1;
+			goto bail;
+		}
+
+		/* prevent looping on a bad block */
+		de_len = le16_to_cpu(de->rec_len);
+		if (de_len <= 0) {
+			ret = -1;
+			goto bail;
+		}
+
+		de_buf += de_len;
+		offset += de_len;
+	}
+
+bail:
+	trace_ocfs2_search_dirblock(ret);
+	return ret;
+}
+
+static struct buffer_head *ocfs2_find_entry_id(const char *name,
+					       int namelen,
+					       struct inode *dir,
+					       struct ocfs2_dir_entry **res_dir)
+{
+	int ret, found;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di;
+	struct ocfs2_inline_data *data;
+
+	ret = ocfs2_read_inode_block(dir, &di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	data = &di->id2.i_data;
+
+	found = ocfs2_search_dirblock(di_bh, dir, name, namelen, 0,
+				      data->id_data, i_size_read(dir), res_dir);
+	if (found == 1)
+		return di_bh;
+
+	brelse(di_bh);
+out:
+	return NULL;
+}
+
+static int ocfs2_validate_dir_block(struct super_block *sb,
+				    struct buffer_head *bh)
+{
+	int rc;
+	struct ocfs2_dir_block_trailer *trailer =
+		ocfs2_trailer_from_bh(bh, sb);
+
+
+	/*
+	 * We don't validate dirents here, that's handled
+	 * in-place when the code walks them.
+	 */
+	trace_ocfs2_validate_dir_block((unsigned long long)bh->b_blocknr);
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	/*
+	 * If the ecc fails, we return the error but otherwise
+	 * leave the filesystem running.  We know any error is
+	 * local to this block.
+	 *
+	 * Note that we are safe to call this even if the directory
+	 * doesn't have a trailer.  Filesystems without metaecc will do
+	 * nothing, and filesystems with it will have one.
+	 */
+	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &trailer->db_check);
+	if (rc)
+		mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
+		     (unsigned long long)bh->b_blocknr);
+
+	return rc;
+}
+
+/*
+ * Validate a directory trailer.
+ *
+ * We check the trailer here rather than in ocfs2_validate_dir_block()
+ * because that function doesn't have the inode to test.
+ */
+static int ocfs2_check_dir_trailer(struct inode *dir, struct buffer_head *bh)
+{
+	int rc = 0;
+	struct ocfs2_dir_block_trailer *trailer;
+
+	trailer = ocfs2_trailer_from_bh(bh, dir->i_sb);
+	if (!OCFS2_IS_VALID_DIR_TRAILER(trailer)) {
+		rc = -EINVAL;
+		ocfs2_error(dir->i_sb,
+			    "Invalid dirblock #%llu: "
+			    "signature = %.*s\n",
+			    (unsigned long long)bh->b_blocknr, 7,
+			    trailer->db_signature);
+		goto out;
+	}
+	if (le64_to_cpu(trailer->db_blkno) != bh->b_blocknr) {
+		rc = -EINVAL;
+		ocfs2_error(dir->i_sb,
+			    "Directory block #%llu has an invalid "
+			    "db_blkno of %llu",
+			    (unsigned long long)bh->b_blocknr,
+			    (unsigned long long)le64_to_cpu(trailer->db_blkno));
+		goto out;
+	}
+	if (le64_to_cpu(trailer->db_parent_dinode) !=
+	    OCFS2_I(dir)->ip_blkno) {
+		rc = -EINVAL;
+		ocfs2_error(dir->i_sb,
+			    "Directory block #%llu on dinode "
+			    "#%llu has an invalid parent_dinode "
+			    "of %llu",
+			    (unsigned long long)bh->b_blocknr,
+			    (unsigned long long)OCFS2_I(dir)->ip_blkno,
+			    (unsigned long long)le64_to_cpu(trailer->db_blkno));
+		goto out;
+	}
+out:
+	return rc;
+}
+
+/*
+ * This function forces all errors to -EIO for consistency with its
+ * predecessor, ocfs2_bread().  We haven't audited what returning the
+ * real error codes would do to callers.  We log the real codes with
+ * mlog_errno() before we squash them.
+ */
+static int ocfs2_read_dir_block(struct inode *inode, u64 v_block,
+				struct buffer_head **bh, int flags)
+{
+	int rc = 0;
+	struct buffer_head *tmp = *bh;
+
+	rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, flags,
+				    ocfs2_validate_dir_block);
+	if (rc) {
+		mlog_errno(rc);
+		goto out;
+	}
+
+	if (!(flags & OCFS2_BH_READAHEAD) &&
+	    ocfs2_supports_dir_trailer(inode)) {
+		rc = ocfs2_check_dir_trailer(inode, tmp);
+		if (rc) {
+			if (!*bh)
+				brelse(tmp);
+			mlog_errno(rc);
+			goto out;
+		}
+	}
+
+	/* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
+	if (!*bh)
+		*bh = tmp;
+
+out:
+	return rc ? -EIO : 0;
+}
+
+/*
+ * Read the block at 'phys' which belongs to this directory
+ * inode. This function does no virtual->physical block translation -
+ * what's passed in is assumed to be a valid directory block.
+ */
+static int ocfs2_read_dir_block_direct(struct inode *dir, u64 phys,
+				       struct buffer_head **bh)
+{
+	int ret;
+	struct buffer_head *tmp = *bh;
+
+	ret = ocfs2_read_block(INODE_CACHE(dir), phys, &tmp,
+			       ocfs2_validate_dir_block);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (ocfs2_supports_dir_trailer(dir)) {
+		ret = ocfs2_check_dir_trailer(dir, tmp);
+		if (ret) {
+			if (!*bh)
+				brelse(tmp);
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (!ret && !*bh)
+		*bh = tmp;
+out:
+	return ret;
+}
+
+static int ocfs2_validate_dx_root(struct super_block *sb,
+				  struct buffer_head *bh)
+{
+	int ret;
+	struct ocfs2_dx_root_block *dx_root;
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	dx_root = (struct ocfs2_dx_root_block *) bh->b_data;
+
+	ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_root->dr_check);
+	if (ret) {
+		mlog(ML_ERROR,
+		     "Checksum failed for dir index root block %llu\n",
+		     (unsigned long long)bh->b_blocknr);
+		return ret;
+	}
+
+	if (!OCFS2_IS_VALID_DX_ROOT(dx_root)) {
+		ocfs2_error(sb,
+			    "Dir Index Root # %llu has bad signature %.*s",
+			    (unsigned long long)le64_to_cpu(dx_root->dr_blkno),
+			    7, dx_root->dr_signature);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int ocfs2_read_dx_root(struct inode *dir, struct ocfs2_dinode *di,
+			      struct buffer_head **dx_root_bh)
+{
+	int ret;
+	u64 blkno = le64_to_cpu(di->i_dx_root);
+	struct buffer_head *tmp = *dx_root_bh;
+
+	ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
+			       ocfs2_validate_dx_root);
+
+	/* If ocfs2_read_block() got us a new bh, pass it up. */
+	if (!ret && !*dx_root_bh)
+		*dx_root_bh = tmp;
+
+	return ret;
+}
+
+static int ocfs2_validate_dx_leaf(struct super_block *sb,
+				  struct buffer_head *bh)
+{
+	int ret;
+	struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)bh->b_data;
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	ret = ocfs2_validate_meta_ecc(sb, bh->b_data, &dx_leaf->dl_check);
+	if (ret) {
+		mlog(ML_ERROR,
+		     "Checksum failed for dir index leaf block %llu\n",
+		     (unsigned long long)bh->b_blocknr);
+		return ret;
+	}
+
+	if (!OCFS2_IS_VALID_DX_LEAF(dx_leaf)) {
+		ocfs2_error(sb, "Dir Index Leaf has bad signature %.*s",
+			    7, dx_leaf->dl_signature);
+		return -EROFS;
+	}
+
+	return 0;
+}
+
+static int ocfs2_read_dx_leaf(struct inode *dir, u64 blkno,
+			      struct buffer_head **dx_leaf_bh)
+{
+	int ret;
+	struct buffer_head *tmp = *dx_leaf_bh;
+
+	ret = ocfs2_read_block(INODE_CACHE(dir), blkno, &tmp,
+			       ocfs2_validate_dx_leaf);
+
+	/* If ocfs2_read_block() got us a new bh, pass it up. */
+	if (!ret && !*dx_leaf_bh)
+		*dx_leaf_bh = tmp;
+
+	return ret;
+}
+
+/*
+ * Read a series of dx_leaf blocks. This expects all buffer_head
+ * pointers to be NULL on function entry.
+ */
+static int ocfs2_read_dx_leaves(struct inode *dir, u64 start, int num,
+				struct buffer_head **dx_leaf_bhs)
+{
+	int ret;
+
+	ret = ocfs2_read_blocks(INODE_CACHE(dir), start, num, dx_leaf_bhs, 0,
+				ocfs2_validate_dx_leaf);
+	if (ret)
+		mlog_errno(ret);
+
+	return ret;
+}
+
+static struct buffer_head *ocfs2_find_entry_el(const char *name, int namelen,
+					       struct inode *dir,
+					       struct ocfs2_dir_entry **res_dir)
+{
+	struct super_block *sb;
+	struct buffer_head *bh_use[NAMEI_RA_SIZE];
+	struct buffer_head *bh, *ret = NULL;
+	unsigned long start, block, b;
+	int ra_max = 0;		/* Number of bh's in the readahead
+				   buffer, bh_use[] */
+	int ra_ptr = 0;		/* Current index into readahead
+				   buffer */
+	int num = 0;
+	int nblocks, i, err;
+
+	sb = dir->i_sb;
+
+	nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
+	start = OCFS2_I(dir)->ip_dir_start_lookup;
+	if (start >= nblocks)
+		start = 0;
+	block = start;
+
+restart:
+	do {
+		/*
+		 * We deal with the read-ahead logic here.
+		 */
+		if (ra_ptr >= ra_max) {
+			/* Refill the readahead buffer */
+			ra_ptr = 0;
+			b = block;
+			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
+				/*
+				 * Terminate if we reach the end of the
+				 * directory and must wrap, or if our
+				 * search has finished at this block.
+				 */
+				if (b >= nblocks || (num && block == start)) {
+					bh_use[ra_max] = NULL;
+					break;
+				}
+				num++;
+
+				bh = NULL;
+				err = ocfs2_read_dir_block(dir, b++, &bh,
+							   OCFS2_BH_READAHEAD);
+				bh_use[ra_max] = bh;
+			}
+		}
+		if ((bh = bh_use[ra_ptr++]) == NULL)
+			goto next;
+		if (ocfs2_read_dir_block(dir, block, &bh, 0)) {
+			/* read error, skip block & hope for the best.
+			 * ocfs2_read_dir_block() has released the bh. */
+			mlog(ML_ERROR, "reading directory %llu, "
+				    "offset %lu\n",
+				    (unsigned long long)OCFS2_I(dir)->ip_blkno,
+				    block);
+			goto next;
+		}
+		i = ocfs2_search_dirblock(bh, dir, name, namelen,
+					  block << sb->s_blocksize_bits,
+					  bh->b_data, sb->s_blocksize,
+					  res_dir);
+		if (i == 1) {
+			OCFS2_I(dir)->ip_dir_start_lookup = block;
+			ret = bh;
+			goto cleanup_and_exit;
+		} else {
+			brelse(bh);
+			if (i < 0)
+				goto cleanup_and_exit;
+		}
+	next:
+		if (++block >= nblocks)
+			block = 0;
+	} while (block != start);
+
+	/*
+	 * If the directory has grown while we were searching, then
+	 * search the last part of the directory before giving up.
+	 */
+	block = nblocks;
+	nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
+	if (block < nblocks) {
+		start = 0;
+		goto restart;
+	}
+
+cleanup_and_exit:
+	/* Clean up the read-ahead blocks */
+	for (; ra_ptr < ra_max; ra_ptr++)
+		brelse(bh_use[ra_ptr]);
+
+	trace_ocfs2_find_entry_el(ret);
+	return ret;
+}
+
+static int ocfs2_dx_dir_lookup_rec(struct inode *inode,
+				   struct ocfs2_extent_list *el,
+				   u32 major_hash,
+				   u32 *ret_cpos,
+				   u64 *ret_phys_blkno,
+				   unsigned int *ret_clen)
+{
+	int ret = 0, i, found;
+	struct buffer_head *eb_bh = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_rec *rec = NULL;
+
+	if (el->l_tree_depth) {
+		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, major_hash,
+				      &eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+
+		if (el->l_tree_depth) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %lu has non zero tree depth in "
+				    "btree tree block %llu\n", inode->i_ino,
+				    (unsigned long long)eb_bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
+		}
+	}
+
+	found = 0;
+	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
+		rec = &el->l_recs[i];
+
+		if (le32_to_cpu(rec->e_cpos) <= major_hash) {
+			found = 1;
+			break;
+		}
+	}
+
+	if (!found) {
+		ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
+			    "record (%u, %u, 0) in btree", inode->i_ino,
+			    le32_to_cpu(rec->e_cpos),
+			    ocfs2_rec_clusters(el, rec));
+		ret = -EROFS;
+		goto out;
+	}
+
+	if (ret_phys_blkno)
+		*ret_phys_blkno = le64_to_cpu(rec->e_blkno);
+	if (ret_cpos)
+		*ret_cpos = le32_to_cpu(rec->e_cpos);
+	if (ret_clen)
+		*ret_clen = le16_to_cpu(rec->e_leaf_clusters);
+
+out:
+	brelse(eb_bh);
+	return ret;
+}
+
+/*
+ * Returns the block index, from the start of the cluster which this
+ * hash belongs too.
+ */
+static inline unsigned int __ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
+						   u32 minor_hash)
+{
+	return minor_hash & osb->osb_dx_mask;
+}
+
+static inline unsigned int ocfs2_dx_dir_hash_idx(struct ocfs2_super *osb,
+					  struct ocfs2_dx_hinfo *hinfo)
+{
+	return __ocfs2_dx_dir_hash_idx(osb, hinfo->minor_hash);
+}
+
+static int ocfs2_dx_dir_lookup(struct inode *inode,
+			       struct ocfs2_extent_list *el,
+			       struct ocfs2_dx_hinfo *hinfo,
+			       u32 *ret_cpos,
+			       u64 *ret_phys_blkno)
+{
+	int ret = 0;
+	unsigned int cend, uninitialized_var(clen);
+	u32 uninitialized_var(cpos);
+	u64 uninitialized_var(blkno);
+	u32 name_hash = hinfo->major_hash;
+
+	ret = ocfs2_dx_dir_lookup_rec(inode, el, name_hash, &cpos, &blkno,
+				      &clen);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	cend = cpos + clen;
+	if (name_hash >= cend) {
+		/* We want the last cluster */
+		blkno += ocfs2_clusters_to_blocks(inode->i_sb, clen - 1);
+		cpos += clen - 1;
+	} else {
+		blkno += ocfs2_clusters_to_blocks(inode->i_sb,
+						  name_hash - cpos);
+		cpos = name_hash;
+	}
+
+	/*
+	 * We now have the cluster which should hold our entry. To
+	 * find the exact block from the start of the cluster to
+	 * search, we take the lower bits of the hash.
+	 */
+	blkno += ocfs2_dx_dir_hash_idx(OCFS2_SB(inode->i_sb), hinfo);
+
+	if (ret_phys_blkno)
+		*ret_phys_blkno = blkno;
+	if (ret_cpos)
+		*ret_cpos = cpos;
+
+out:
+
+	return ret;
+}
+
+static int ocfs2_dx_dir_search(const char *name, int namelen,
+			       struct inode *dir,
+			       struct ocfs2_dx_root_block *dx_root,
+			       struct ocfs2_dir_lookup_result *res)
+{
+	int ret, i, found;
+	u64 uninitialized_var(phys);
+	struct buffer_head *dx_leaf_bh = NULL;
+	struct ocfs2_dx_leaf *dx_leaf;
+	struct ocfs2_dx_entry *dx_entry = NULL;
+	struct buffer_head *dir_ent_bh = NULL;
+	struct ocfs2_dir_entry *dir_ent = NULL;
+	struct ocfs2_dx_hinfo *hinfo = &res->dl_hinfo;
+	struct ocfs2_extent_list *dr_el;
+	struct ocfs2_dx_entry_list *entry_list;
+
+	ocfs2_dx_dir_name_hash(dir, name, namelen, &res->dl_hinfo);
+
+	if (ocfs2_dx_root_inline(dx_root)) {
+		entry_list = &dx_root->dr_entries;
+		goto search;
+	}
+
+	dr_el = &dx_root->dr_list;
+
+	ret = ocfs2_dx_dir_lookup(dir, dr_el, hinfo, NULL, &phys);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_dx_dir_search((unsigned long long)OCFS2_I(dir)->ip_blkno,
+				  namelen, name, hinfo->major_hash,
+				  hinfo->minor_hash, (unsigned long long)phys);
+
+	ret = ocfs2_read_dx_leaf(dir, phys, &dx_leaf_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	dx_leaf = (struct ocfs2_dx_leaf *) dx_leaf_bh->b_data;
+
+	trace_ocfs2_dx_dir_search_leaf_info(
+			le16_to_cpu(dx_leaf->dl_list.de_num_used),
+			le16_to_cpu(dx_leaf->dl_list.de_count));
+
+	entry_list = &dx_leaf->dl_list;
+
+search:
+	/*
+	 * Empty leaf is legal, so no need to check for that.
+	 */
+	found = 0;
+	for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
+		dx_entry = &entry_list->de_entries[i];
+
+		if (hinfo->major_hash != le32_to_cpu(dx_entry->dx_major_hash)
+		    || hinfo->minor_hash != le32_to_cpu(dx_entry->dx_minor_hash))
+			continue;
+
+		/*
+		 * Search unindexed leaf block now. We're not
+		 * guaranteed to find anything.
+		 */
+		ret = ocfs2_read_dir_block_direct(dir,
+					  le64_to_cpu(dx_entry->dx_dirent_blk),
+					  &dir_ent_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * XXX: We should check the unindexed block here,
+		 * before using it.
+		 */
+
+		found = ocfs2_search_dirblock(dir_ent_bh, dir, name, namelen,
+					      0, dir_ent_bh->b_data,
+					      dir->i_sb->s_blocksize, &dir_ent);
+		if (found == 1)
+			break;
+
+		if (found == -1) {
+			/* This means we found a bad directory entry. */
+			ret = -EIO;
+			mlog_errno(ret);
+			goto out;
+		}
+
+		brelse(dir_ent_bh);
+		dir_ent_bh = NULL;
+	}
+
+	if (found <= 0) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	res->dl_leaf_bh = dir_ent_bh;
+	res->dl_entry = dir_ent;
+	res->dl_dx_leaf_bh = dx_leaf_bh;
+	res->dl_dx_entry = dx_entry;
+
+	ret = 0;
+out:
+	if (ret) {
+		brelse(dx_leaf_bh);
+		brelse(dir_ent_bh);
+	}
+	return ret;
+}
+
+static int ocfs2_find_entry_dx(const char *name, int namelen,
+			       struct inode *dir,
+			       struct ocfs2_dir_lookup_result *lookup)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di;
+	struct buffer_head *dx_root_bh = NULL;
+	struct ocfs2_dx_root_block *dx_root;
+
+	ret = ocfs2_read_inode_block(dir, &di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
+
+	ret = ocfs2_dx_dir_search(name, namelen, dir, dx_root, lookup);
+	if (ret) {
+		if (ret != -ENOENT)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	lookup->dl_dx_root_bh = dx_root_bh;
+	dx_root_bh = NULL;
+out:
+	brelse(di_bh);
+	brelse(dx_root_bh);
+	return ret;
+}
+
+/*
+ * Try to find an entry of the provided name within 'dir'.
+ *
+ * If nothing was found, -ENOENT is returned. Otherwise, zero is
+ * returned and the struct 'res' will contain information useful to
+ * other directory manipulation functions.
+ *
+ * Caller can NOT assume anything about the contents of the
+ * buffer_heads - they are passed back only so that it can be passed
+ * into any one of the manipulation functions (add entry, delete
+ * entry, etc). As an example, bh in the extent directory case is a
+ * data block, in the inline-data case it actually points to an inode,
+ * in the indexed directory case, multiple buffers are involved.
+ */
+int ocfs2_find_entry(const char *name, int namelen,
+		     struct inode *dir, struct ocfs2_dir_lookup_result *lookup)
+{
+	struct buffer_head *bh;
+	struct ocfs2_dir_entry *res_dir = NULL;
+
+	if (ocfs2_dir_indexed(dir))
+		return ocfs2_find_entry_dx(name, namelen, dir, lookup);
+
+	/*
+	 * The unindexed dir code only uses part of the lookup
+	 * structure, so there's no reason to push it down further
+	 * than this.
+	 */
+	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		bh = ocfs2_find_entry_id(name, namelen, dir, &res_dir);
+	else
+		bh = ocfs2_find_entry_el(name, namelen, dir, &res_dir);
+
+	if (bh == NULL)
+		return -ENOENT;
+
+	lookup->dl_leaf_bh = bh;
+	lookup->dl_entry = res_dir;
+	return 0;
+}
+
+/*
+ * Update inode number and type of a previously found directory entry.
+ */
+int ocfs2_update_entry(struct inode *dir, handle_t *handle,
+		       struct ocfs2_dir_lookup_result *res,
+		       struct inode *new_entry_inode)
+{
+	int ret;
+	ocfs2_journal_access_func access = ocfs2_journal_access_db;
+	struct ocfs2_dir_entry *de = res->dl_entry;
+	struct buffer_head *de_bh = res->dl_leaf_bh;
+
+	/*
+	 * The same code works fine for both inline-data and extent
+	 * based directories, so no need to split this up.  The only
+	 * difference is the journal_access function.
+	 */
+
+	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		access = ocfs2_journal_access_di;
+
+	ret = access(handle, INODE_CACHE(dir), de_bh,
+		     OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	de->inode = cpu_to_le64(OCFS2_I(new_entry_inode)->ip_blkno);
+	ocfs2_set_de_type(de, new_entry_inode->i_mode);
+
+	ocfs2_journal_dirty(handle, de_bh);
+
+out:
+	return ret;
+}
+
+/*
+ * __ocfs2_delete_entry deletes a directory entry by merging it with the
+ * previous entry
+ */
+static int __ocfs2_delete_entry(handle_t *handle, struct inode *dir,
+				struct ocfs2_dir_entry *de_del,
+				struct buffer_head *bh, char *first_de,
+				unsigned int bytes)
+{
+	struct ocfs2_dir_entry *de, *pde;
+	int i, status = -ENOENT;
+	ocfs2_journal_access_func access = ocfs2_journal_access_db;
+
+	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		access = ocfs2_journal_access_di;
+
+	i = 0;
+	pde = NULL;
+	de = (struct ocfs2_dir_entry *) first_de;
+	while (i < bytes) {
+		if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
+			status = -EIO;
+			mlog_errno(status);
+			goto bail;
+		}
+		if (de == de_del)  {
+			status = access(handle, INODE_CACHE(dir), bh,
+					OCFS2_JOURNAL_ACCESS_WRITE);
+			if (status < 0) {
+				status = -EIO;
+				mlog_errno(status);
+				goto bail;
+			}
+			if (pde)
+				le16_add_cpu(&pde->rec_len,
+						le16_to_cpu(de->rec_len));
+			de->inode = 0;
+			dir->i_version++;
+			ocfs2_journal_dirty(handle, bh);
+			goto bail;
+		}
+		i += le16_to_cpu(de->rec_len);
+		pde = de;
+		de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
+	}
+bail:
+	return status;
+}
+
+static unsigned int ocfs2_figure_dirent_hole(struct ocfs2_dir_entry *de)
+{
+	unsigned int hole;
+
+	if (le64_to_cpu(de->inode) == 0)
+		hole = le16_to_cpu(de->rec_len);
+	else
+		hole = le16_to_cpu(de->rec_len) -
+			OCFS2_DIR_REC_LEN(de->name_len);
+
+	return hole;
+}
+
+static int ocfs2_find_max_rec_len(struct super_block *sb,
+				  struct buffer_head *dirblock_bh)
+{
+	int size, this_hole, largest_hole = 0;
+	char *trailer, *de_buf, *limit, *start = dirblock_bh->b_data;
+	struct ocfs2_dir_entry *de;
+
+	trailer = (char *)ocfs2_trailer_from_bh(dirblock_bh, sb);
+	size = ocfs2_dir_trailer_blk_off(sb);
+	limit = start + size;
+	de_buf = start;
+	de = (struct ocfs2_dir_entry *)de_buf;
+	do {
+		if (de_buf != trailer) {
+			this_hole = ocfs2_figure_dirent_hole(de);
+			if (this_hole > largest_hole)
+				largest_hole = this_hole;
+		}
+
+		de_buf += le16_to_cpu(de->rec_len);
+		de = (struct ocfs2_dir_entry *)de_buf;
+	} while (de_buf < limit);
+
+	if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
+		return largest_hole;
+	return 0;
+}
+
+static void ocfs2_dx_list_remove_entry(struct ocfs2_dx_entry_list *entry_list,
+				       int index)
+{
+	int num_used = le16_to_cpu(entry_list->de_num_used);
+
+	if (num_used == 1 || index == (num_used - 1))
+		goto clear;
+
+	memmove(&entry_list->de_entries[index],
+		&entry_list->de_entries[index + 1],
+		(num_used - index - 1)*sizeof(struct ocfs2_dx_entry));
+clear:
+	num_used--;
+	memset(&entry_list->de_entries[num_used], 0,
+	       sizeof(struct ocfs2_dx_entry));
+	entry_list->de_num_used = cpu_to_le16(num_used);
+}
+
+static int ocfs2_delete_entry_dx(handle_t *handle, struct inode *dir,
+				 struct ocfs2_dir_lookup_result *lookup)
+{
+	int ret, index, max_rec_len, add_to_free_list = 0;
+	struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
+	struct buffer_head *leaf_bh = lookup->dl_leaf_bh;
+	struct ocfs2_dx_leaf *dx_leaf;
+	struct ocfs2_dx_entry *dx_entry = lookup->dl_dx_entry;
+	struct ocfs2_dir_block_trailer *trailer;
+	struct ocfs2_dx_root_block *dx_root;
+	struct ocfs2_dx_entry_list *entry_list;
+
+	/*
+	 * This function gets a bit messy because we might have to
+	 * modify the root block, regardless of whether the indexed
+	 * entries are stored inline.
+	 */
+
+	/*
+	 * *Only* set 'entry_list' here, based on where we're looking
+	 * for the indexed entries. Later, we might still want to
+	 * journal both blocks, based on free list state.
+	 */
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+	if (ocfs2_dx_root_inline(dx_root)) {
+		entry_list = &dx_root->dr_entries;
+	} else {
+		dx_leaf = (struct ocfs2_dx_leaf *) lookup->dl_dx_leaf_bh->b_data;
+		entry_list = &dx_leaf->dl_list;
+	}
+
+	/* Neither of these are a disk corruption - that should have
+	 * been caught by lookup, before we got here. */
+	BUG_ON(le16_to_cpu(entry_list->de_count) <= 0);
+	BUG_ON(le16_to_cpu(entry_list->de_num_used) <= 0);
+
+	index = (char *)dx_entry - (char *)entry_list->de_entries;
+	index /= sizeof(*dx_entry);
+
+	if (index >= le16_to_cpu(entry_list->de_num_used)) {
+		mlog(ML_ERROR, "Dir %llu: Bad dx_entry ptr idx %d, (%p, %p)\n",
+		     (unsigned long long)OCFS2_I(dir)->ip_blkno, index,
+		     entry_list, dx_entry);
+		return -EIO;
+	}
+
+	/*
+	 * We know that removal of this dirent will leave enough room
+	 * for a new one, so add this block to the free list if it
+	 * isn't already there.
+	 */
+	trailer = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
+	if (trailer->db_free_rec_len == 0)
+		add_to_free_list = 1;
+
+	/*
+	 * Add the block holding our index into the journal before
+	 * removing the unindexed entry. If we get an error return
+	 * from __ocfs2_delete_entry(), then it hasn't removed the
+	 * entry yet. Likewise, successful return means we *must*
+	 * remove the indexed entry.
+	 *
+	 * We're also careful to journal the root tree block here as
+	 * the entry count needs to be updated. Also, we might be
+	 * adding to the start of the free list.
+	 */
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (!ocfs2_dx_root_inline(dx_root)) {
+		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
+					      lookup->dl_dx_leaf_bh,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	trace_ocfs2_delete_entry_dx((unsigned long long)OCFS2_I(dir)->ip_blkno,
+				    index);
+
+	ret = __ocfs2_delete_entry(handle, dir, lookup->dl_entry,
+				   leaf_bh, leaf_bh->b_data, leaf_bh->b_size);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, leaf_bh);
+	trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
+	if (add_to_free_list) {
+		trailer->db_free_next = dx_root->dr_free_blk;
+		dx_root->dr_free_blk = cpu_to_le64(leaf_bh->b_blocknr);
+		ocfs2_journal_dirty(handle, dx_root_bh);
+	}
+
+	/* leaf_bh was journal_accessed for us in __ocfs2_delete_entry */
+	ocfs2_journal_dirty(handle, leaf_bh);
+
+	le32_add_cpu(&dx_root->dr_num_entries, -1);
+	ocfs2_journal_dirty(handle, dx_root_bh);
+
+	ocfs2_dx_list_remove_entry(entry_list, index);
+
+	if (!ocfs2_dx_root_inline(dx_root))
+		ocfs2_journal_dirty(handle, lookup->dl_dx_leaf_bh);
+
+out:
+	return ret;
+}
+
+static inline int ocfs2_delete_entry_id(handle_t *handle,
+					struct inode *dir,
+					struct ocfs2_dir_entry *de_del,
+					struct buffer_head *bh)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di;
+	struct ocfs2_inline_data *data;
+
+	ret = ocfs2_read_inode_block(dir, &di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	data = &di->id2.i_data;
+
+	ret = __ocfs2_delete_entry(handle, dir, de_del, bh, data->id_data,
+				   i_size_read(dir));
+
+	brelse(di_bh);
+out:
+	return ret;
+}
+
+static inline int ocfs2_delete_entry_el(handle_t *handle,
+					struct inode *dir,
+					struct ocfs2_dir_entry *de_del,
+					struct buffer_head *bh)
+{
+	return __ocfs2_delete_entry(handle, dir, de_del, bh, bh->b_data,
+				    bh->b_size);
+}
+
+/*
+ * Delete a directory entry. Hide the details of directory
+ * implementation from the caller.
+ */
+int ocfs2_delete_entry(handle_t *handle,
+		       struct inode *dir,
+		       struct ocfs2_dir_lookup_result *res)
+{
+	if (ocfs2_dir_indexed(dir))
+		return ocfs2_delete_entry_dx(handle, dir, res);
+
+	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		return ocfs2_delete_entry_id(handle, dir, res->dl_entry,
+					     res->dl_leaf_bh);
+
+	return ocfs2_delete_entry_el(handle, dir, res->dl_entry,
+				     res->dl_leaf_bh);
+}
+
+/*
+ * Check whether 'de' has enough room to hold an entry of
+ * 'new_rec_len' bytes.
+ */
+static inline int ocfs2_dirent_would_fit(struct ocfs2_dir_entry *de,
+					 unsigned int new_rec_len)
+{
+	unsigned int de_really_used;
+
+	/* Check whether this is an empty record with enough space */
+	if (le64_to_cpu(de->inode) == 0 &&
+	    le16_to_cpu(de->rec_len) >= new_rec_len)
+		return 1;
+
+	/*
+	 * Record might have free space at the end which we can
+	 * use.
+	 */
+	de_really_used = OCFS2_DIR_REC_LEN(de->name_len);
+	if (le16_to_cpu(de->rec_len) >= (de_really_used + new_rec_len))
+	    return 1;
+
+	return 0;
+}
+
+static void ocfs2_dx_dir_leaf_insert_tail(struct ocfs2_dx_leaf *dx_leaf,
+					  struct ocfs2_dx_entry *dx_new_entry)
+{
+	int i;
+
+	i = le16_to_cpu(dx_leaf->dl_list.de_num_used);
+	dx_leaf->dl_list.de_entries[i] = *dx_new_entry;
+
+	le16_add_cpu(&dx_leaf->dl_list.de_num_used, 1);
+}
+
+static void ocfs2_dx_entry_list_insert(struct ocfs2_dx_entry_list *entry_list,
+				       struct ocfs2_dx_hinfo *hinfo,
+				       u64 dirent_blk)
+{
+	int i;
+	struct ocfs2_dx_entry *dx_entry;
+
+	i = le16_to_cpu(entry_list->de_num_used);
+	dx_entry = &entry_list->de_entries[i];
+
+	memset(dx_entry, 0, sizeof(*dx_entry));
+	dx_entry->dx_major_hash = cpu_to_le32(hinfo->major_hash);
+	dx_entry->dx_minor_hash = cpu_to_le32(hinfo->minor_hash);
+	dx_entry->dx_dirent_blk = cpu_to_le64(dirent_blk);
+
+	le16_add_cpu(&entry_list->de_num_used, 1);
+}
+
+static int __ocfs2_dx_dir_leaf_insert(struct inode *dir, handle_t *handle,
+				      struct ocfs2_dx_hinfo *hinfo,
+				      u64 dirent_blk,
+				      struct buffer_head *dx_leaf_bh)
+{
+	int ret;
+	struct ocfs2_dx_leaf *dx_leaf;
+
+	ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
+	ocfs2_dx_entry_list_insert(&dx_leaf->dl_list, hinfo, dirent_blk);
+	ocfs2_journal_dirty(handle, dx_leaf_bh);
+
+out:
+	return ret;
+}
+
+static void ocfs2_dx_inline_root_insert(struct inode *dir, handle_t *handle,
+					struct ocfs2_dx_hinfo *hinfo,
+					u64 dirent_blk,
+					struct ocfs2_dx_root_block *dx_root)
+{
+	ocfs2_dx_entry_list_insert(&dx_root->dr_entries, hinfo, dirent_blk);
+}
+
+static int ocfs2_dx_dir_insert(struct inode *dir, handle_t *handle,
+			       struct ocfs2_dir_lookup_result *lookup)
+{
+	int ret = 0;
+	struct ocfs2_dx_root_block *dx_root;
+	struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
+
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	dx_root = (struct ocfs2_dx_root_block *)lookup->dl_dx_root_bh->b_data;
+	if (ocfs2_dx_root_inline(dx_root)) {
+		ocfs2_dx_inline_root_insert(dir, handle,
+					    &lookup->dl_hinfo,
+					    lookup->dl_leaf_bh->b_blocknr,
+					    dx_root);
+	} else {
+		ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &lookup->dl_hinfo,
+						 lookup->dl_leaf_bh->b_blocknr,
+						 lookup->dl_dx_leaf_bh);
+		if (ret)
+			goto out;
+	}
+
+	le32_add_cpu(&dx_root->dr_num_entries, 1);
+	ocfs2_journal_dirty(handle, dx_root_bh);
+
+out:
+	return ret;
+}
+
+static void ocfs2_remove_block_from_free_list(struct inode *dir,
+				       handle_t *handle,
+				       struct ocfs2_dir_lookup_result *lookup)
+{
+	struct ocfs2_dir_block_trailer *trailer, *prev;
+	struct ocfs2_dx_root_block *dx_root;
+	struct buffer_head *bh;
+
+	trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
+
+	if (ocfs2_free_list_at_root(lookup)) {
+		bh = lookup->dl_dx_root_bh;
+		dx_root = (struct ocfs2_dx_root_block *)bh->b_data;
+		dx_root->dr_free_blk = trailer->db_free_next;
+	} else {
+		bh = lookup->dl_prev_leaf_bh;
+		prev = ocfs2_trailer_from_bh(bh, dir->i_sb);
+		prev->db_free_next = trailer->db_free_next;
+	}
+
+	trailer->db_free_rec_len = cpu_to_le16(0);
+	trailer->db_free_next = cpu_to_le64(0);
+
+	ocfs2_journal_dirty(handle, bh);
+	ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
+}
+
+/*
+ * This expects that a journal write has been reserved on
+ * lookup->dl_prev_leaf_bh or lookup->dl_dx_root_bh
+ */
+static void ocfs2_recalc_free_list(struct inode *dir, handle_t *handle,
+				   struct ocfs2_dir_lookup_result *lookup)
+{
+	int max_rec_len;
+	struct ocfs2_dir_block_trailer *trailer;
+
+	/* Walk dl_leaf_bh to figure out what the new free rec_len is. */
+	max_rec_len = ocfs2_find_max_rec_len(dir->i_sb, lookup->dl_leaf_bh);
+	if (max_rec_len) {
+		/*
+		 * There's still room in this block, so no need to remove it
+		 * from the free list. In this case, we just want to update
+		 * the rec len accounting.
+		 */
+		trailer = ocfs2_trailer_from_bh(lookup->dl_leaf_bh, dir->i_sb);
+		trailer->db_free_rec_len = cpu_to_le16(max_rec_len);
+		ocfs2_journal_dirty(handle, lookup->dl_leaf_bh);
+	} else {
+		ocfs2_remove_block_from_free_list(dir, handle, lookup);
+	}
+}
+
+/* we don't always have a dentry for what we want to add, so people
+ * like orphan dir can call this instead.
+ *
+ * The lookup context must have been filled from
+ * ocfs2_prepare_dir_for_insert.
+ */
+int __ocfs2_add_entry(handle_t *handle,
+		      struct inode *dir,
+		      const char *name, int namelen,
+		      struct inode *inode, u64 blkno,
+		      struct buffer_head *parent_fe_bh,
+		      struct ocfs2_dir_lookup_result *lookup)
+{
+	unsigned long offset;
+	unsigned short rec_len;
+	struct ocfs2_dir_entry *de, *de1;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_fe_bh->b_data;
+	struct super_block *sb = dir->i_sb;
+	int retval, status;
+	unsigned int size = sb->s_blocksize;
+	struct buffer_head *insert_bh = lookup->dl_leaf_bh;
+	char *data_start = insert_bh->b_data;
+
+	if (!namelen)
+		return -EINVAL;
+
+	if (ocfs2_dir_indexed(dir)) {
+		struct buffer_head *bh;
+
+		/*
+		 * An indexed dir may require that we update the free space
+		 * list. Reserve a write to the previous node in the list so
+		 * that we don't fail later.
+		 *
+		 * XXX: This can be either a dx_root_block, or an unindexed
+		 * directory tree leaf block.
+		 */
+		if (ocfs2_free_list_at_root(lookup)) {
+			bh = lookup->dl_dx_root_bh;
+			retval = ocfs2_journal_access_dr(handle,
+						 INODE_CACHE(dir), bh,
+						 OCFS2_JOURNAL_ACCESS_WRITE);
+		} else {
+			bh = lookup->dl_prev_leaf_bh;
+			retval = ocfs2_journal_access_db(handle,
+						 INODE_CACHE(dir), bh,
+						 OCFS2_JOURNAL_ACCESS_WRITE);
+		}
+		if (retval) {
+			mlog_errno(retval);
+			return retval;
+		}
+	} else if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		data_start = di->id2.i_data.id_data;
+		size = i_size_read(dir);
+
+		BUG_ON(insert_bh != parent_fe_bh);
+	}
+
+	rec_len = OCFS2_DIR_REC_LEN(namelen);
+	offset = 0;
+	de = (struct ocfs2_dir_entry *) data_start;
+	while (1) {
+		BUG_ON((char *)de >= (size + data_start));
+
+		/* These checks should've already been passed by the
+		 * prepare function, but I guess we can leave them
+		 * here anyway. */
+		if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
+			retval = -ENOENT;
+			goto bail;
+		}
+		if (ocfs2_match(namelen, name, de)) {
+			retval = -EEXIST;
+			goto bail;
+		}
+
+		/* We're guaranteed that we should have space, so we
+		 * can't possibly have hit the trailer...right? */
+		mlog_bug_on_msg(ocfs2_skip_dir_trailer(dir, de, offset, size),
+				"Hit dir trailer trying to insert %.*s "
+			        "(namelen %d) into directory %llu.  "
+				"offset is %lu, trailer offset is %d\n",
+				namelen, name, namelen,
+				(unsigned long long)parent_fe_bh->b_blocknr,
+				offset, ocfs2_dir_trailer_blk_off(dir->i_sb));
+
+		if (ocfs2_dirent_would_fit(de, rec_len)) {
+			dir->i_mtime = dir->i_ctime = CURRENT_TIME;
+			retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
+			if (retval < 0) {
+				mlog_errno(retval);
+				goto bail;
+			}
+
+			if (insert_bh == parent_fe_bh)
+				status = ocfs2_journal_access_di(handle,
+								 INODE_CACHE(dir),
+								 insert_bh,
+								 OCFS2_JOURNAL_ACCESS_WRITE);
+			else {
+				status = ocfs2_journal_access_db(handle,
+								 INODE_CACHE(dir),
+								 insert_bh,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+
+				if (ocfs2_dir_indexed(dir)) {
+					status = ocfs2_dx_dir_insert(dir,
+								handle,
+								lookup);
+					if (status) {
+						mlog_errno(status);
+						goto bail;
+					}
+				}
+			}
+
+			/* By now the buffer is marked for journaling */
+			offset += le16_to_cpu(de->rec_len);
+			if (le64_to_cpu(de->inode)) {
+				de1 = (struct ocfs2_dir_entry *)((char *) de +
+					OCFS2_DIR_REC_LEN(de->name_len));
+				de1->rec_len =
+					cpu_to_le16(le16_to_cpu(de->rec_len) -
+					OCFS2_DIR_REC_LEN(de->name_len));
+				de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
+				de = de1;
+			}
+			de->file_type = OCFS2_FT_UNKNOWN;
+			if (blkno) {
+				de->inode = cpu_to_le64(blkno);
+				ocfs2_set_de_type(de, inode->i_mode);
+			} else
+				de->inode = 0;
+			de->name_len = namelen;
+			memcpy(de->name, name, namelen);
+
+			if (ocfs2_dir_indexed(dir))
+				ocfs2_recalc_free_list(dir, handle, lookup);
+
+			dir->i_version++;
+			ocfs2_journal_dirty(handle, insert_bh);
+			retval = 0;
+			goto bail;
+		}
+
+		offset += le16_to_cpu(de->rec_len);
+		de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
+	}
+
+	/* when you think about it, the assert above should prevent us
+	 * from ever getting here. */
+	retval = -ENOSPC;
+bail:
+	if (retval)
+		mlog_errno(retval);
+
+	return retval;
+}
+
+static int ocfs2_dir_foreach_blk_id(struct inode *inode,
+				    u64 *f_version,
+				    struct dir_context *ctx)
+{
+	int ret, i;
+	unsigned long offset = ctx->pos;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di;
+	struct ocfs2_inline_data *data;
+	struct ocfs2_dir_entry *de;
+
+	ret = ocfs2_read_inode_block(inode, &di_bh);
+	if (ret) {
+		mlog(ML_ERROR, "Unable to read inode block for dir %llu\n",
+		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+		goto out;
+	}
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	data = &di->id2.i_data;
+
+	while (ctx->pos < i_size_read(inode)) {
+		/* If the dir block has changed since the last call to
+		 * readdir(2), then we might be pointing to an invalid
+		 * dirent right now.  Scan from the start of the block
+		 * to make sure. */
+		if (*f_version != inode->i_version) {
+			for (i = 0; i < i_size_read(inode) && i < offset; ) {
+				de = (struct ocfs2_dir_entry *)
+					(data->id_data + i);
+				/* It's too expensive to do a full
+				 * dirent test each time round this
+				 * loop, but we do have to test at
+				 * least that it is non-zero.  A
+				 * failure will be detected in the
+				 * dirent test below. */
+				if (le16_to_cpu(de->rec_len) <
+				    OCFS2_DIR_REC_LEN(1))
+					break;
+				i += le16_to_cpu(de->rec_len);
+			}
+			ctx->pos = offset = i;
+			*f_version = inode->i_version;
+		}
+
+		de = (struct ocfs2_dir_entry *) (data->id_data + ctx->pos);
+		if (!ocfs2_check_dir_entry(inode, de, di_bh, ctx->pos)) {
+			/* On error, skip the f_pos to the end. */
+			ctx->pos = i_size_read(inode);
+			break;
+		}
+		offset += le16_to_cpu(de->rec_len);
+		if (le64_to_cpu(de->inode)) {
+			unsigned char d_type = DT_UNKNOWN;
+
+			if (de->file_type < OCFS2_FT_MAX)
+				d_type = ocfs2_filetype_table[de->file_type];
+
+			if (!dir_emit(ctx, de->name, de->name_len,
+				      le64_to_cpu(de->inode), d_type))
+				goto out;
+		}
+		ctx->pos += le16_to_cpu(de->rec_len);
+	}
+out:
+	brelse(di_bh);
+	return 0;
+}
+
+/*
+ * NOTE: This function can be called against unindexed directories,
+ * and indexed ones.
+ */
+static int ocfs2_dir_foreach_blk_el(struct inode *inode,
+				    u64 *f_version,
+				    struct dir_context *ctx,
+				    bool persist)
+{
+	unsigned long offset, blk, last_ra_blk = 0;
+	int i;
+	struct buffer_head * bh, * tmp;
+	struct ocfs2_dir_entry * de;
+	struct super_block * sb = inode->i_sb;
+	unsigned int ra_sectors = 16;
+	int stored = 0;
+
+	bh = NULL;
+
+	offset = ctx->pos & (sb->s_blocksize - 1);
+
+	while (ctx->pos < i_size_read(inode)) {
+		blk = ctx->pos >> sb->s_blocksize_bits;
+		if (ocfs2_read_dir_block(inode, blk, &bh, 0)) {
+			/* Skip the corrupt dirblock and keep trying */
+			ctx->pos += sb->s_blocksize - offset;
+			continue;
+		}
+
+		/* The idea here is to begin with 8k read-ahead and to stay
+		 * 4k ahead of our current position.
+		 *
+		 * TODO: Use the pagecache for this. We just need to
+		 * make sure it's cluster-safe... */
+		if (!last_ra_blk
+		    || (((last_ra_blk - blk) << 9) <= (ra_sectors / 2))) {
+			for (i = ra_sectors >> (sb->s_blocksize_bits - 9);
+			     i > 0; i--) {
+				tmp = NULL;
+				if (!ocfs2_read_dir_block(inode, ++blk, &tmp,
+							  OCFS2_BH_READAHEAD))
+					brelse(tmp);
+			}
+			last_ra_blk = blk;
+			ra_sectors = 8;
+		}
+
+		/* If the dir block has changed since the last call to
+		 * readdir(2), then we might be pointing to an invalid
+		 * dirent right now.  Scan from the start of the block
+		 * to make sure. */
+		if (*f_version != inode->i_version) {
+			for (i = 0; i < sb->s_blocksize && i < offset; ) {
+				de = (struct ocfs2_dir_entry *) (bh->b_data + i);
+				/* It's too expensive to do a full
+				 * dirent test each time round this
+				 * loop, but we do have to test at
+				 * least that it is non-zero.  A
+				 * failure will be detected in the
+				 * dirent test below. */
+				if (le16_to_cpu(de->rec_len) <
+				    OCFS2_DIR_REC_LEN(1))
+					break;
+				i += le16_to_cpu(de->rec_len);
+			}
+			offset = i;
+			ctx->pos = (ctx->pos & ~(sb->s_blocksize - 1))
+				| offset;
+			*f_version = inode->i_version;
+		}
+
+		while (ctx->pos < i_size_read(inode)
+		       && offset < sb->s_blocksize) {
+			de = (struct ocfs2_dir_entry *) (bh->b_data + offset);
+			if (!ocfs2_check_dir_entry(inode, de, bh, offset)) {
+				/* On error, skip the f_pos to the
+				   next block. */
+				ctx->pos = (ctx->pos | (sb->s_blocksize - 1)) + 1;
+				brelse(bh);
+				continue;
+			}
+			if (le64_to_cpu(de->inode)) {
+				unsigned char d_type = DT_UNKNOWN;
+
+				if (de->file_type < OCFS2_FT_MAX)
+					d_type = ocfs2_filetype_table[de->file_type];
+				if (!dir_emit(ctx, de->name,
+						de->name_len,
+						le64_to_cpu(de->inode),
+						d_type)) {
+					brelse(bh);
+					return 0;
+				}
+				stored++;
+			}
+			offset += le16_to_cpu(de->rec_len);
+			ctx->pos += le16_to_cpu(de->rec_len);
+		}
+		offset = 0;
+		brelse(bh);
+		bh = NULL;
+		if (!persist && stored)
+			break;
+	}
+	return 0;
+}
+
+static int ocfs2_dir_foreach_blk(struct inode *inode, u64 *f_version,
+				 struct dir_context *ctx,
+				 bool persist)
+{
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		return ocfs2_dir_foreach_blk_id(inode, f_version, ctx);
+	return ocfs2_dir_foreach_blk_el(inode, f_version, ctx, persist);
+}
+
+/*
+ * This is intended to be called from inside other kernel functions,
+ * so we fake some arguments.
+ */
+int ocfs2_dir_foreach(struct inode *inode, struct dir_context *ctx)
+{
+	u64 version = inode->i_version;
+	ocfs2_dir_foreach_blk(inode, &version, ctx, true);
+	return 0;
+}
+
+/*
+ * ocfs2_readdir()
+ *
+ */
+int ocfs2_readdir(struct file *file, struct dir_context *ctx)
+{
+	int error = 0;
+	struct inode *inode = file_inode(file);
+	int lock_level = 0;
+
+	trace_ocfs2_readdir((unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	error = ocfs2_inode_lock_atime(inode, file->f_path.mnt, &lock_level);
+	if (lock_level && error >= 0) {
+		/* We release EX lock which used to update atime
+		 * and get PR lock again to reduce contention
+		 * on commonly accessed directories. */
+		ocfs2_inode_unlock(inode, 1);
+		lock_level = 0;
+		error = ocfs2_inode_lock(inode, NULL, 0);
+	}
+	if (error < 0) {
+		if (error != -ENOENT)
+			mlog_errno(error);
+		/* we haven't got any yet, so propagate the error. */
+		goto bail_nolock;
+	}
+
+	error = ocfs2_dir_foreach_blk(inode, &file->f_version, ctx, false);
+
+	ocfs2_inode_unlock(inode, lock_level);
+	if (error)
+		mlog_errno(error);
+
+bail_nolock:
+
+	return error;
+}
+
+/*
+ * NOTE: this should always be called with parent dir i_mutex taken.
+ */
+int ocfs2_find_files_on_disk(const char *name,
+			     int namelen,
+			     u64 *blkno,
+			     struct inode *inode,
+			     struct ocfs2_dir_lookup_result *lookup)
+{
+	int status = -ENOENT;
+
+	trace_ocfs2_find_files_on_disk(namelen, name, blkno,
+				(unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	status = ocfs2_find_entry(name, namelen, inode, lookup);
+	if (status)
+		goto leave;
+
+	*blkno = le64_to_cpu(lookup->dl_entry->inode);
+
+	status = 0;
+leave:
+
+	return status;
+}
+
+/*
+ * Convenience function for callers which just want the block number
+ * mapped to a name and don't require the full dirent info, etc.
+ */
+int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
+			       int namelen, u64 *blkno)
+{
+	int ret;
+	struct ocfs2_dir_lookup_result lookup = { NULL, };
+
+	ret = ocfs2_find_files_on_disk(name, namelen, blkno, dir, &lookup);
+	ocfs2_free_dir_lookup_result(&lookup);
+
+	return ret;
+}
+
+/* Check for a name within a directory.
+ *
+ * Return 0 if the name does not exist
+ * Return -EEXIST if the directory contains the name
+ *
+ * Callers should have i_mutex + a cluster lock on dir
+ */
+int ocfs2_check_dir_for_entry(struct inode *dir,
+			      const char *name,
+			      int namelen)
+{
+	int ret = 0;
+	struct ocfs2_dir_lookup_result lookup = { NULL, };
+
+	trace_ocfs2_check_dir_for_entry(
+		(unsigned long long)OCFS2_I(dir)->ip_blkno, namelen, name);
+
+	if (ocfs2_find_entry(name, namelen, dir, &lookup) == 0) {
+		ret = -EEXIST;
+		mlog_errno(ret);
+	}
+
+	ocfs2_free_dir_lookup_result(&lookup);
+
+	return ret;
+}
+
+struct ocfs2_empty_dir_priv {
+	struct dir_context ctx;
+	unsigned seen_dot;
+	unsigned seen_dot_dot;
+	unsigned seen_other;
+	unsigned dx_dir;
+};
+static int ocfs2_empty_dir_filldir(struct dir_context *ctx, const char *name,
+				   int name_len, loff_t pos, u64 ino,
+				   unsigned type)
+{
+	struct ocfs2_empty_dir_priv *p =
+		container_of(ctx, struct ocfs2_empty_dir_priv, ctx);
+
+	/*
+	 * Check the positions of "." and ".." records to be sure
+	 * they're in the correct place.
+	 *
+	 * Indexed directories don't need to proceed past the first
+	 * two entries, so we end the scan after seeing '..'. Despite
+	 * that, we allow the scan to proceed In the event that we
+	 * have a corrupted indexed directory (no dot or dot dot
+	 * entries). This allows us to double check for existing
+	 * entries which might not have been found in the index.
+	 */
+	if (name_len == 1 && !strncmp(".", name, 1) && pos == 0) {
+		p->seen_dot = 1;
+		return 0;
+	}
+
+	if (name_len == 2 && !strncmp("..", name, 2) &&
+	    pos == OCFS2_DIR_REC_LEN(1)) {
+		p->seen_dot_dot = 1;
+
+		if (p->dx_dir && p->seen_dot)
+			return 1;
+
+		return 0;
+	}
+
+	p->seen_other = 1;
+	return 1;
+}
+
+static int ocfs2_empty_dir_dx(struct inode *inode,
+			      struct ocfs2_empty_dir_priv *priv)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	struct buffer_head *dx_root_bh = NULL;
+	struct ocfs2_dinode *di;
+	struct ocfs2_dx_root_block *dx_root;
+
+	priv->dx_dir = 1;
+
+	ret = ocfs2_read_inode_block(inode, &di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	ret = ocfs2_read_dx_root(inode, di, &dx_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+
+	if (le32_to_cpu(dx_root->dr_num_entries) != 2)
+		priv->seen_other = 1;
+
+out:
+	brelse(di_bh);
+	brelse(dx_root_bh);
+	return ret;
+}
+
+/*
+ * routine to check that the specified directory is empty (for rmdir)
+ *
+ * Returns 1 if dir is empty, zero otherwise.
+ *
+ * XXX: This is a performance problem for unindexed directories.
+ */
+int ocfs2_empty_dir(struct inode *inode)
+{
+	int ret;
+	struct ocfs2_empty_dir_priv priv = {
+		.ctx.actor = ocfs2_empty_dir_filldir,
+	};
+
+	if (ocfs2_dir_indexed(inode)) {
+		ret = ocfs2_empty_dir_dx(inode, &priv);
+		if (ret)
+			mlog_errno(ret);
+		/*
+		 * We still run ocfs2_dir_foreach to get the checks
+		 * for "." and "..".
+		 */
+	}
+
+	ret = ocfs2_dir_foreach(inode, &priv.ctx);
+	if (ret)
+		mlog_errno(ret);
+
+	if (!priv.seen_dot || !priv.seen_dot_dot) {
+		mlog(ML_ERROR, "bad directory (dir #%llu) - no `.' or `..'\n",
+		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+		/*
+		 * XXX: Is it really safe to allow an unlink to continue?
+		 */
+		return 1;
+	}
+
+	return !priv.seen_other;
+}
+
+/*
+ * Fills "." and ".." dirents in a new directory block. Returns dirent for
+ * "..", which might be used during creation of a directory with a trailing
+ * header. It is otherwise safe to ignore the return code.
+ */
+static struct ocfs2_dir_entry *ocfs2_fill_initial_dirents(struct inode *inode,
+							  struct inode *parent,
+							  char *start,
+							  unsigned int size)
+{
+	struct ocfs2_dir_entry *de = (struct ocfs2_dir_entry *)start;
+
+	de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
+	de->name_len = 1;
+	de->rec_len =
+		cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
+	strcpy(de->name, ".");
+	ocfs2_set_de_type(de, S_IFDIR);
+
+	de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
+	de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
+	de->rec_len = cpu_to_le16(size - OCFS2_DIR_REC_LEN(1));
+	de->name_len = 2;
+	strcpy(de->name, "..");
+	ocfs2_set_de_type(de, S_IFDIR);
+
+	return de;
+}
+
+/*
+ * This works together with code in ocfs2_mknod_locked() which sets
+ * the inline-data flag and initializes the inline-data section.
+ */
+static int ocfs2_fill_new_dir_id(struct ocfs2_super *osb,
+				 handle_t *handle,
+				 struct inode *parent,
+				 struct inode *inode,
+				 struct buffer_head *di_bh)
+{
+	int ret;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_inline_data *data = &di->id2.i_data;
+	unsigned int size = le16_to_cpu(data->id_count);
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_fill_initial_dirents(inode, parent, data->id_data, size);
+	ocfs2_journal_dirty(handle, di_bh);
+
+	i_size_write(inode, size);
+	set_nlink(inode, 2);
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
+
+	ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
+	if (ret < 0)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+static int ocfs2_fill_new_dir_el(struct ocfs2_super *osb,
+				 handle_t *handle,
+				 struct inode *parent,
+				 struct inode *inode,
+				 struct buffer_head *fe_bh,
+				 struct ocfs2_alloc_context *data_ac,
+				 struct buffer_head **ret_new_bh)
+{
+	int status;
+	unsigned int size = osb->sb->s_blocksize;
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_dir_entry *de;
+
+	if (ocfs2_new_dir_wants_trailer(inode))
+		size = ocfs2_dir_trailer_blk_off(parent->i_sb);
+
+	status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
+				     data_ac, NULL, &new_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
+
+	status = ocfs2_journal_access_db(handle, INODE_CACHE(inode), new_bh,
+					 OCFS2_JOURNAL_ACCESS_CREATE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	memset(new_bh->b_data, 0, osb->sb->s_blocksize);
+
+	de = ocfs2_fill_initial_dirents(inode, parent, new_bh->b_data, size);
+	if (ocfs2_new_dir_wants_trailer(inode)) {
+		int size = le16_to_cpu(de->rec_len);
+
+		/*
+		 * Figure out the size of the hole left over after
+		 * insertion of '.' and '..'. The trailer wants this
+		 * information.
+		 */
+		size -= OCFS2_DIR_REC_LEN(2);
+		size -= sizeof(struct ocfs2_dir_block_trailer);
+
+		ocfs2_init_dir_trailer(inode, new_bh, size);
+	}
+
+	ocfs2_journal_dirty(handle, new_bh);
+
+	i_size_write(inode, inode->i_sb->s_blocksize);
+	set_nlink(inode, 2);
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
+	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = 0;
+	if (ret_new_bh) {
+		*ret_new_bh = new_bh;
+		new_bh = NULL;
+	}
+bail:
+	brelse(new_bh);
+
+	return status;
+}
+
+static int ocfs2_dx_dir_attach_index(struct ocfs2_super *osb,
+				     handle_t *handle, struct inode *dir,
+				     struct buffer_head *di_bh,
+				     struct buffer_head *dirdata_bh,
+				     struct ocfs2_alloc_context *meta_ac,
+				     int dx_inline, u32 num_entries,
+				     struct buffer_head **ret_dx_root_bh)
+{
+	int ret;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
+	u16 dr_suballoc_bit;
+	u64 suballoc_loc, dr_blkno;
+	unsigned int num_bits;
+	struct buffer_head *dx_root_bh = NULL;
+	struct ocfs2_dx_root_block *dx_root;
+	struct ocfs2_dir_block_trailer *trailer =
+		ocfs2_trailer_from_bh(dirdata_bh, dir->i_sb);
+
+	ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
+				   &dr_suballoc_bit, &num_bits, &dr_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_dx_dir_attach_index(
+				(unsigned long long)OCFS2_I(dir)->ip_blkno,
+				(unsigned long long)dr_blkno);
+
+	dx_root_bh = sb_getblk(osb->sb, dr_blkno);
+	if (dx_root_bh == NULL) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dx_root_bh);
+
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+	memset(dx_root, 0, osb->sb->s_blocksize);
+	strcpy(dx_root->dr_signature, OCFS2_DX_ROOT_SIGNATURE);
+	dx_root->dr_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
+	dx_root->dr_suballoc_loc = cpu_to_le64(suballoc_loc);
+	dx_root->dr_suballoc_bit = cpu_to_le16(dr_suballoc_bit);
+	dx_root->dr_fs_generation = cpu_to_le32(osb->fs_generation);
+	dx_root->dr_blkno = cpu_to_le64(dr_blkno);
+	dx_root->dr_dir_blkno = cpu_to_le64(OCFS2_I(dir)->ip_blkno);
+	dx_root->dr_num_entries = cpu_to_le32(num_entries);
+	if (le16_to_cpu(trailer->db_free_rec_len))
+		dx_root->dr_free_blk = cpu_to_le64(dirdata_bh->b_blocknr);
+	else
+		dx_root->dr_free_blk = cpu_to_le64(0);
+
+	if (dx_inline) {
+		dx_root->dr_flags |= OCFS2_DX_FLAG_INLINE;
+		dx_root->dr_entries.de_count =
+			cpu_to_le16(ocfs2_dx_entries_per_root(osb->sb));
+	} else {
+		dx_root->dr_list.l_count =
+			cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
+	}
+	ocfs2_journal_dirty(handle, dx_root_bh);
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	di->i_dx_root = cpu_to_le64(dr_blkno);
+
+	spin_lock(&OCFS2_I(dir)->ip_lock);
+	OCFS2_I(dir)->ip_dyn_features |= OCFS2_INDEXED_DIR_FL;
+	di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
+	spin_unlock(&OCFS2_I(dir)->ip_lock);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+	*ret_dx_root_bh = dx_root_bh;
+	dx_root_bh = NULL;
+
+out:
+	brelse(dx_root_bh);
+	return ret;
+}
+
+static int ocfs2_dx_dir_format_cluster(struct ocfs2_super *osb,
+				       handle_t *handle, struct inode *dir,
+				       struct buffer_head **dx_leaves,
+				       int num_dx_leaves, u64 start_blk)
+{
+	int ret, i;
+	struct ocfs2_dx_leaf *dx_leaf;
+	struct buffer_head *bh;
+
+	for (i = 0; i < num_dx_leaves; i++) {
+		bh = sb_getblk(osb->sb, start_blk + i);
+		if (bh == NULL) {
+			ret = -ENOMEM;
+			goto out;
+		}
+		dx_leaves[i] = bh;
+
+		ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), bh);
+
+		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), bh,
+					      OCFS2_JOURNAL_ACCESS_CREATE);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		dx_leaf = (struct ocfs2_dx_leaf *) bh->b_data;
+
+		memset(dx_leaf, 0, osb->sb->s_blocksize);
+		strcpy(dx_leaf->dl_signature, OCFS2_DX_LEAF_SIGNATURE);
+		dx_leaf->dl_fs_generation = cpu_to_le32(osb->fs_generation);
+		dx_leaf->dl_blkno = cpu_to_le64(bh->b_blocknr);
+		dx_leaf->dl_list.de_count =
+			cpu_to_le16(ocfs2_dx_entries_per_leaf(osb->sb));
+
+		trace_ocfs2_dx_dir_format_cluster(
+				(unsigned long long)OCFS2_I(dir)->ip_blkno,
+				(unsigned long long)bh->b_blocknr,
+				le16_to_cpu(dx_leaf->dl_list.de_count));
+
+		ocfs2_journal_dirty(handle, bh);
+	}
+
+	ret = 0;
+out:
+	return ret;
+}
+
+/*
+ * Allocates and formats a new cluster for use in an indexed dir
+ * leaf. This version will not do the extent insert, so that it can be
+ * used by operations which need careful ordering.
+ */
+static int __ocfs2_dx_dir_new_cluster(struct inode *dir,
+				      u32 cpos, handle_t *handle,
+				      struct ocfs2_alloc_context *data_ac,
+				      struct buffer_head **dx_leaves,
+				      int num_dx_leaves, u64 *ret_phys_blkno)
+{
+	int ret;
+	u32 phys, num;
+	u64 phys_blkno;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+
+	/*
+	 * XXX: For create, this should claim cluster for the index
+	 * *before* the unindexed insert so that we have a better
+	 * chance of contiguousness as the directory grows in number
+	 * of entries.
+	 */
+	ret = __ocfs2_claim_clusters(handle, data_ac, 1, 1, &phys, &num);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Format the new cluster first. That way, we're inserting
+	 * valid data.
+	 */
+	phys_blkno = ocfs2_clusters_to_blocks(osb->sb, phys);
+	ret = ocfs2_dx_dir_format_cluster(osb, handle, dir, dx_leaves,
+					  num_dx_leaves, phys_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	*ret_phys_blkno = phys_blkno;
+out:
+	return ret;
+}
+
+static int ocfs2_dx_dir_new_cluster(struct inode *dir,
+				    struct ocfs2_extent_tree *et,
+				    u32 cpos, handle_t *handle,
+				    struct ocfs2_alloc_context *data_ac,
+				    struct ocfs2_alloc_context *meta_ac,
+				    struct buffer_head **dx_leaves,
+				    int num_dx_leaves)
+{
+	int ret;
+	u64 phys_blkno;
+
+	ret = __ocfs2_dx_dir_new_cluster(dir, cpos, handle, data_ac, dx_leaves,
+					 num_dx_leaves, &phys_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_insert_extent(handle, et, cpos, phys_blkno, 1, 0,
+				  meta_ac);
+	if (ret)
+		mlog_errno(ret);
+out:
+	return ret;
+}
+
+static struct buffer_head **ocfs2_dx_dir_kmalloc_leaves(struct super_block *sb,
+							int *ret_num_leaves)
+{
+	int num_dx_leaves = ocfs2_clusters_to_blocks(sb, 1);
+	struct buffer_head **dx_leaves;
+
+	dx_leaves = kcalloc(num_dx_leaves, sizeof(struct buffer_head *),
+			    GFP_NOFS);
+	if (dx_leaves && ret_num_leaves)
+		*ret_num_leaves = num_dx_leaves;
+
+	return dx_leaves;
+}
+
+static int ocfs2_fill_new_dir_dx(struct ocfs2_super *osb,
+				 handle_t *handle,
+				 struct inode *parent,
+				 struct inode *inode,
+				 struct buffer_head *di_bh,
+				 struct ocfs2_alloc_context *data_ac,
+				 struct ocfs2_alloc_context *meta_ac)
+{
+	int ret;
+	struct buffer_head *leaf_bh = NULL;
+	struct buffer_head *dx_root_bh = NULL;
+	struct ocfs2_dx_hinfo hinfo;
+	struct ocfs2_dx_root_block *dx_root;
+	struct ocfs2_dx_entry_list *entry_list;
+
+	/*
+	 * Our strategy is to create the directory as though it were
+	 * unindexed, then add the index block. This works with very
+	 * little complication since the state of a new directory is a
+	 * very well known quantity.
+	 *
+	 * Essentially, we have two dirents ("." and ".."), in the 1st
+	 * block which need indexing. These are easily inserted into
+	 * the index block.
+	 */
+
+	ret = ocfs2_fill_new_dir_el(osb, handle, parent, inode, di_bh,
+				    data_ac, &leaf_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_dx_dir_attach_index(osb, handle, inode, di_bh, leaf_bh,
+					meta_ac, 1, 2, &dx_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+	entry_list = &dx_root->dr_entries;
+
+	/* Buffer has been journaled for us by ocfs2_dx_dir_attach_index */
+	ocfs2_dx_dir_name_hash(inode, ".", 1, &hinfo);
+	ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
+
+	ocfs2_dx_dir_name_hash(inode, "..", 2, &hinfo);
+	ocfs2_dx_entry_list_insert(entry_list, &hinfo, leaf_bh->b_blocknr);
+
+out:
+	brelse(dx_root_bh);
+	brelse(leaf_bh);
+	return ret;
+}
+
+int ocfs2_fill_new_dir(struct ocfs2_super *osb,
+		       handle_t *handle,
+		       struct inode *parent,
+		       struct inode *inode,
+		       struct buffer_head *fe_bh,
+		       struct ocfs2_alloc_context *data_ac,
+		       struct ocfs2_alloc_context *meta_ac)
+
+{
+	BUG_ON(!ocfs2_supports_inline_data(osb) && data_ac == NULL);
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		return ocfs2_fill_new_dir_id(osb, handle, parent, inode, fe_bh);
+
+	if (ocfs2_supports_indexed_dirs(osb))
+		return ocfs2_fill_new_dir_dx(osb, handle, parent, inode, fe_bh,
+					     data_ac, meta_ac);
+
+	return ocfs2_fill_new_dir_el(osb, handle, parent, inode, fe_bh,
+				     data_ac, NULL);
+}
+
+static int ocfs2_dx_dir_index_block(struct inode *dir,
+				    handle_t *handle,
+				    struct buffer_head **dx_leaves,
+				    int num_dx_leaves,
+				    u32 *num_dx_entries,
+				    struct buffer_head *dirent_bh)
+{
+	int ret = 0, namelen, i;
+	char *de_buf, *limit;
+	struct ocfs2_dir_entry *de;
+	struct buffer_head *dx_leaf_bh;
+	struct ocfs2_dx_hinfo hinfo;
+	u64 dirent_blk = dirent_bh->b_blocknr;
+
+	de_buf = dirent_bh->b_data;
+	limit = de_buf + dir->i_sb->s_blocksize;
+
+	while (de_buf < limit) {
+		de = (struct ocfs2_dir_entry *)de_buf;
+
+		namelen = de->name_len;
+		if (!namelen || !de->inode)
+			goto inc;
+
+		ocfs2_dx_dir_name_hash(dir, de->name, namelen, &hinfo);
+
+		i = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb), &hinfo);
+		dx_leaf_bh = dx_leaves[i];
+
+		ret = __ocfs2_dx_dir_leaf_insert(dir, handle, &hinfo,
+						 dirent_blk, dx_leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		*num_dx_entries = *num_dx_entries + 1;
+
+inc:
+		de_buf += le16_to_cpu(de->rec_len);
+	}
+
+out:
+	return ret;
+}
+
+/*
+ * XXX: This expects dx_root_bh to already be part of the transaction.
+ */
+static void ocfs2_dx_dir_index_root_block(struct inode *dir,
+					 struct buffer_head *dx_root_bh,
+					 struct buffer_head *dirent_bh)
+{
+	char *de_buf, *limit;
+	struct ocfs2_dx_root_block *dx_root;
+	struct ocfs2_dir_entry *de;
+	struct ocfs2_dx_hinfo hinfo;
+	u64 dirent_blk = dirent_bh->b_blocknr;
+
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+
+	de_buf = dirent_bh->b_data;
+	limit = de_buf + dir->i_sb->s_blocksize;
+
+	while (de_buf < limit) {
+		de = (struct ocfs2_dir_entry *)de_buf;
+
+		if (!de->name_len || !de->inode)
+			goto inc;
+
+		ocfs2_dx_dir_name_hash(dir, de->name, de->name_len, &hinfo);
+
+		trace_ocfs2_dx_dir_index_root_block(
+				(unsigned long long)dir->i_ino,
+				hinfo.major_hash, hinfo.minor_hash,
+				de->name_len, de->name,
+				le16_to_cpu(dx_root->dr_entries.de_num_used));
+
+		ocfs2_dx_entry_list_insert(&dx_root->dr_entries, &hinfo,
+					   dirent_blk);
+
+		le32_add_cpu(&dx_root->dr_num_entries, 1);
+inc:
+		de_buf += le16_to_cpu(de->rec_len);
+	}
+}
+
+/*
+ * Count the number of inline directory entries in di_bh and compare
+ * them against the number of entries we can hold in an inline dx root
+ * block.
+ */
+static int ocfs2_new_dx_should_be_inline(struct inode *dir,
+					 struct buffer_head *di_bh)
+{
+	int dirent_count = 0;
+	char *de_buf, *limit;
+	struct ocfs2_dir_entry *de;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	de_buf = di->id2.i_data.id_data;
+	limit = de_buf + i_size_read(dir);
+
+	while (de_buf < limit) {
+		de = (struct ocfs2_dir_entry *)de_buf;
+
+		if (de->name_len && de->inode)
+			dirent_count++;
+
+		de_buf += le16_to_cpu(de->rec_len);
+	}
+
+	/* We are careful to leave room for one extra record. */
+	return dirent_count < ocfs2_dx_entries_per_root(dir->i_sb);
+}
+
+/*
+ * Expand rec_len of the rightmost dirent in a directory block so that it
+ * contains the end of our valid space for dirents. We do this during
+ * expansion from an inline directory to one with extents. The first dir block
+ * in that case is taken from the inline data portion of the inode block.
+ *
+ * This will also return the largest amount of contiguous space for a dirent
+ * in the block. That value is *not* necessarily the last dirent, even after
+ * expansion. The directory indexing code wants this value for free space
+ * accounting. We do this here since we're already walking the entire dir
+ * block.
+ *
+ * We add the dir trailer if this filesystem wants it.
+ */
+static unsigned int ocfs2_expand_last_dirent(char *start, unsigned int old_size,
+					     struct inode *dir)
+{
+	struct super_block *sb = dir->i_sb;
+	struct ocfs2_dir_entry *de;
+	struct ocfs2_dir_entry *prev_de;
+	char *de_buf, *limit;
+	unsigned int new_size = sb->s_blocksize;
+	unsigned int bytes, this_hole;
+	unsigned int largest_hole = 0;
+
+	if (ocfs2_new_dir_wants_trailer(dir))
+		new_size = ocfs2_dir_trailer_blk_off(sb);
+
+	bytes = new_size - old_size;
+
+	limit = start + old_size;
+	de_buf = start;
+	de = (struct ocfs2_dir_entry *)de_buf;
+	do {
+		this_hole = ocfs2_figure_dirent_hole(de);
+		if (this_hole > largest_hole)
+			largest_hole = this_hole;
+
+		prev_de = de;
+		de_buf += le16_to_cpu(de->rec_len);
+		de = (struct ocfs2_dir_entry *)de_buf;
+	} while (de_buf < limit);
+
+	le16_add_cpu(&prev_de->rec_len, bytes);
+
+	/* We need to double check this after modification of the final
+	 * dirent. */
+	this_hole = ocfs2_figure_dirent_hole(prev_de);
+	if (this_hole > largest_hole)
+		largest_hole = this_hole;
+
+	if (largest_hole >= OCFS2_DIR_MIN_REC_LEN)
+		return largest_hole;
+	return 0;
+}
+
+/*
+ * We allocate enough clusters to fulfill "blocks_wanted", but set
+ * i_size to exactly one block. Ocfs2_extend_dir() will handle the
+ * rest automatically for us.
+ *
+ * *first_block_bh is a pointer to the 1st data block allocated to the
+ *  directory.
+ */
+static int ocfs2_expand_inline_dir(struct inode *dir, struct buffer_head *di_bh,
+				   unsigned int blocks_wanted,
+				   struct ocfs2_dir_lookup_result *lookup,
+				   struct buffer_head **first_block_bh)
+{
+	u32 alloc, dx_alloc, bit_off, len, num_dx_entries = 0;
+	struct super_block *sb = dir->i_sb;
+	int ret, i, num_dx_leaves = 0, dx_inline = 0,
+		credits = ocfs2_inline_to_extents_credits(sb);
+	u64 dx_insert_blkno, blkno,
+		bytes = blocks_wanted << sb->s_blocksize_bits;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct ocfs2_inode_info *oi = OCFS2_I(dir);
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	struct buffer_head *dirdata_bh = NULL;
+	struct buffer_head *dx_root_bh = NULL;
+	struct buffer_head **dx_leaves = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	handle_t *handle;
+	struct ocfs2_extent_tree et;
+	struct ocfs2_extent_tree dx_et;
+	int did_quota = 0, bytes_allocated = 0;
+
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir), di_bh);
+
+	alloc = ocfs2_clusters_for_bytes(sb, bytes);
+	dx_alloc = 0;
+
+	down_write(&oi->ip_alloc_sem);
+
+	if (ocfs2_supports_indexed_dirs(osb)) {
+		credits += ocfs2_add_dir_index_credits(sb);
+
+		dx_inline = ocfs2_new_dx_should_be_inline(dir, di_bh);
+		if (!dx_inline) {
+			/* Add one more cluster for an index leaf */
+			dx_alloc++;
+			dx_leaves = ocfs2_dx_dir_kmalloc_leaves(sb,
+								&num_dx_leaves);
+			if (!dx_leaves) {
+				ret = -ENOMEM;
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		/* This gets us the dx_root */
+		ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	/*
+	 * We should never need more than 2 clusters for the unindexed
+	 * tree - maximum dirent size is far less than one block. In
+	 * fact, the only time we'd need more than one cluster is if
+	 * blocksize == clustersize and the dirent won't fit in the
+	 * extra space that the expansion to a single block gives. As
+	 * of today, that only happens on 4k/4k file systems.
+	 */
+	BUG_ON(alloc > 2);
+
+	ret = ocfs2_reserve_clusters(osb, alloc + dx_alloc, &data_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Prepare for worst case allocation scenario of two separate
+	 * extents in the unindexed tree.
+	 */
+	if (alloc == 2)
+		credits += OCFS2_SUBALLOC_ALLOC;
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = dquot_alloc_space_nodirty(dir,
+		ocfs2_clusters_to_bytes(osb->sb, alloc + dx_alloc));
+	if (ret)
+		goto out_commit;
+	did_quota = 1;
+
+	if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
+		/*
+		 * Allocate our index cluster first, to maximize the
+		 * possibility that unindexed leaves grow
+		 * contiguously.
+		 */
+		ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac,
+						 dx_leaves, num_dx_leaves,
+						 &dx_insert_blkno);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+		bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
+	}
+
+	/*
+	 * Try to claim as many clusters as the bitmap can give though
+	 * if we only get one now, that's enough to continue. The rest
+	 * will be claimed after the conversion to extents.
+	 */
+	if (ocfs2_dir_resv_allowed(osb))
+		data_ac->ac_resv = &oi->ip_la_data_resv;
+	ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off, &len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+	bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
+
+	/*
+	 * Operations are carefully ordered so that we set up the new
+	 * data block first. The conversion from inline data to
+	 * extents follows.
+	 */
+	blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
+	dirdata_bh = sb_getblk(sb, blkno);
+	if (!dirdata_bh) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), dirdata_bh);
+
+	ret = ocfs2_journal_access_db(handle, INODE_CACHE(dir), dirdata_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	memcpy(dirdata_bh->b_data, di->id2.i_data.id_data, i_size_read(dir));
+	memset(dirdata_bh->b_data + i_size_read(dir), 0,
+	       sb->s_blocksize - i_size_read(dir));
+	i = ocfs2_expand_last_dirent(dirdata_bh->b_data, i_size_read(dir), dir);
+	if (ocfs2_new_dir_wants_trailer(dir)) {
+		/*
+		 * Prepare the dir trailer up front. It will otherwise look
+		 * like a valid dirent. Even if inserting the index fails
+		 * (unlikely), then all we'll have done is given first dir
+		 * block a small amount of fragmentation.
+		 */
+		ocfs2_init_dir_trailer(dir, dirdata_bh, i);
+	}
+
+	ocfs2_update_inode_fsync_trans(handle, dir, 1);
+	ocfs2_journal_dirty(handle, dirdata_bh);
+
+	if (ocfs2_supports_indexed_dirs(osb) && !dx_inline) {
+		/*
+		 * Dx dirs with an external cluster need to do this up
+		 * front. Inline dx root's get handled later, after
+		 * we've allocated our root block. We get passed back
+		 * a total number of items so that dr_num_entries can
+		 * be correctly set once the dx_root has been
+		 * allocated.
+		 */
+		ret = ocfs2_dx_dir_index_block(dir, handle, dx_leaves,
+					       num_dx_leaves, &num_dx_entries,
+					       dirdata_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+	}
+
+	/*
+	 * Set extent, i_size, etc on the directory. After this, the
+	 * inode should contain the same exact dirents as before and
+	 * be fully accessible from system calls.
+	 *
+	 * We let the later dirent insert modify c/mtime - to the user
+	 * the data hasn't changed.
+	 */
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	spin_unlock(&oi->ip_lock);
+
+	ocfs2_dinode_new_extent_list(dir, di);
+
+	i_size_write(dir, sb->s_blocksize);
+	dir->i_mtime = dir->i_ctime = CURRENT_TIME;
+
+	di->i_size = cpu_to_le64(sb->s_blocksize);
+	di->i_ctime = di->i_mtime = cpu_to_le64(dir->i_ctime.tv_sec);
+	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(dir->i_ctime.tv_nsec);
+	ocfs2_update_inode_fsync_trans(handle, dir, 1);
+
+	/*
+	 * This should never fail as our extent list is empty and all
+	 * related blocks have been journaled already.
+	 */
+	ret = ocfs2_insert_extent(handle, &et, 0, blkno, len,
+				  0, NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	/*
+	 * Set i_blocks after the extent insert for the most up to
+	 * date ip_clusters value.
+	 */
+	dir->i_blocks = ocfs2_inode_sector_count(dir);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+	if (ocfs2_supports_indexed_dirs(osb)) {
+		ret = ocfs2_dx_dir_attach_index(osb, handle, dir, di_bh,
+						dirdata_bh, meta_ac, dx_inline,
+						num_dx_entries, &dx_root_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+
+		if (dx_inline) {
+			ocfs2_dx_dir_index_root_block(dir, dx_root_bh,
+						      dirdata_bh);
+		} else {
+			ocfs2_init_dx_root_extent_tree(&dx_et,
+						       INODE_CACHE(dir),
+						       dx_root_bh);
+			ret = ocfs2_insert_extent(handle, &dx_et, 0,
+						  dx_insert_blkno, 1, 0, NULL);
+			if (ret)
+				mlog_errno(ret);
+		}
+	}
+
+	/*
+	 * We asked for two clusters, but only got one in the 1st
+	 * pass. Claim the 2nd cluster as a separate extent.
+	 */
+	if (alloc > len) {
+		ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
+					   &len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+		blkno = ocfs2_clusters_to_blocks(dir->i_sb, bit_off);
+
+		ret = ocfs2_insert_extent(handle, &et, 1,
+					  blkno, len, 0, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+		bytes_allocated += ocfs2_clusters_to_bytes(dir->i_sb, 1);
+	}
+
+	*first_block_bh = dirdata_bh;
+	dirdata_bh = NULL;
+	if (ocfs2_supports_indexed_dirs(osb)) {
+		unsigned int off;
+
+		if (!dx_inline) {
+			/*
+			 * We need to return the correct block within the
+			 * cluster which should hold our entry.
+			 */
+			off = ocfs2_dx_dir_hash_idx(OCFS2_SB(dir->i_sb),
+						    &lookup->dl_hinfo);
+			get_bh(dx_leaves[off]);
+			lookup->dl_dx_leaf_bh = dx_leaves[off];
+		}
+		lookup->dl_dx_root_bh = dx_root_bh;
+		dx_root_bh = NULL;
+	}
+
+out_commit:
+	if (ret < 0 && did_quota)
+		dquot_free_space_nodirty(dir, bytes_allocated);
+
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	up_write(&oi->ip_alloc_sem);
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	if (dx_leaves) {
+		for (i = 0; i < num_dx_leaves; i++)
+			brelse(dx_leaves[i]);
+		kfree(dx_leaves);
+	}
+
+	brelse(dirdata_bh);
+	brelse(dx_root_bh);
+
+	return ret;
+}
+
+/* returns a bh of the 1st new block in the allocation. */
+static int ocfs2_do_extend_dir(struct super_block *sb,
+			       handle_t *handle,
+			       struct inode *dir,
+			       struct buffer_head *parent_fe_bh,
+			       struct ocfs2_alloc_context *data_ac,
+			       struct ocfs2_alloc_context *meta_ac,
+			       struct buffer_head **new_bh)
+{
+	int status;
+	int extend, did_quota = 0;
+	u64 p_blkno, v_blkno;
+
+	spin_lock(&OCFS2_I(dir)->ip_lock);
+	extend = (i_size_read(dir) == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters));
+	spin_unlock(&OCFS2_I(dir)->ip_lock);
+
+	if (extend) {
+		u32 offset = OCFS2_I(dir)->ip_clusters;
+
+		status = dquot_alloc_space_nodirty(dir,
+					ocfs2_clusters_to_bytes(sb, 1));
+		if (status)
+			goto bail;
+		did_quota = 1;
+
+		status = ocfs2_add_inode_data(OCFS2_SB(sb), dir, &offset,
+					      1, 0, parent_fe_bh, handle,
+					      data_ac, meta_ac, NULL);
+		BUG_ON(status == -EAGAIN);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	v_blkno = ocfs2_blocks_for_bytes(sb, i_size_read(dir));
+	status = ocfs2_extent_map_get_blocks(dir, v_blkno, &p_blkno, NULL, NULL);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	*new_bh = sb_getblk(sb, p_blkno);
+	if (!*new_bh) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+	status = 0;
+bail:
+	if (did_quota && status < 0)
+		dquot_free_space_nodirty(dir, ocfs2_clusters_to_bytes(sb, 1));
+	return status;
+}
+
+/*
+ * Assumes you already have a cluster lock on the directory.
+ *
+ * 'blocks_wanted' is only used if we have an inline directory which
+ * is to be turned into an extent based one. The size of the dirent to
+ * insert might be larger than the space gained by growing to just one
+ * block, so we may have to grow the inode by two blocks in that case.
+ *
+ * If the directory is already indexed, dx_root_bh must be provided.
+ */
+static int ocfs2_extend_dir(struct ocfs2_super *osb,
+			    struct inode *dir,
+			    struct buffer_head *parent_fe_bh,
+			    unsigned int blocks_wanted,
+			    struct ocfs2_dir_lookup_result *lookup,
+			    struct buffer_head **new_de_bh)
+{
+	int status = 0;
+	int credits, num_free_extents, drop_alloc_sem = 0;
+	loff_t dir_i_size;
+	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
+	struct ocfs2_extent_list *el = &fe->id2.i_list;
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	handle_t *handle = NULL;
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_dir_entry * de;
+	struct super_block *sb = osb->sb;
+	struct ocfs2_extent_tree et;
+	struct buffer_head *dx_root_bh = lookup->dl_dx_root_bh;
+
+	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		/*
+		 * This would be a code error as an inline directory should
+		 * never have an index root.
+		 */
+		BUG_ON(dx_root_bh);
+
+		status = ocfs2_expand_inline_dir(dir, parent_fe_bh,
+						 blocks_wanted, lookup,
+						 &new_bh);
+		if (status) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		/* Expansion from inline to an indexed directory will
+		 * have given us this. */
+		dx_root_bh = lookup->dl_dx_root_bh;
+
+		if (blocks_wanted == 1) {
+			/*
+			 * If the new dirent will fit inside the space
+			 * created by pushing out to one block, then
+			 * we can complete the operation
+			 * here. Otherwise we have to expand i_size
+			 * and format the 2nd block below.
+			 */
+			BUG_ON(new_bh == NULL);
+			goto bail_bh;
+		}
+
+		/*
+		 * Get rid of 'new_bh' - we want to format the 2nd
+		 * data block and return that instead.
+		 */
+		brelse(new_bh);
+		new_bh = NULL;
+
+		down_write(&OCFS2_I(dir)->ip_alloc_sem);
+		drop_alloc_sem = 1;
+		dir_i_size = i_size_read(dir);
+		credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
+		goto do_extend;
+	}
+
+	down_write(&OCFS2_I(dir)->ip_alloc_sem);
+	drop_alloc_sem = 1;
+	dir_i_size = i_size_read(dir);
+	trace_ocfs2_extend_dir((unsigned long long)OCFS2_I(dir)->ip_blkno,
+			       dir_i_size);
+
+	/* dir->i_size is always block aligned. */
+	spin_lock(&OCFS2_I(dir)->ip_lock);
+	if (dir_i_size == ocfs2_clusters_to_bytes(sb, OCFS2_I(dir)->ip_clusters)) {
+		spin_unlock(&OCFS2_I(dir)->ip_lock);
+		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(dir),
+					      parent_fe_bh);
+		num_free_extents = ocfs2_num_free_extents(osb, &et);
+		if (num_free_extents < 0) {
+			status = num_free_extents;
+			mlog_errno(status);
+			goto bail;
+		}
+
+		if (!num_free_extents) {
+			status = ocfs2_reserve_new_metadata(osb, el, &meta_ac);
+			if (status < 0) {
+				if (status != -ENOSPC)
+					mlog_errno(status);
+				goto bail;
+			}
+		}
+
+		status = ocfs2_reserve_clusters(osb, 1, &data_ac);
+		if (status < 0) {
+			if (status != -ENOSPC)
+				mlog_errno(status);
+			goto bail;
+		}
+
+		if (ocfs2_dir_resv_allowed(osb))
+			data_ac->ac_resv = &OCFS2_I(dir)->ip_la_data_resv;
+
+		credits = ocfs2_calc_extend_credits(sb, el);
+	} else {
+		spin_unlock(&OCFS2_I(dir)->ip_lock);
+		credits = OCFS2_SIMPLE_DIR_EXTEND_CREDITS;
+	}
+
+do_extend:
+	if (ocfs2_dir_indexed(dir))
+		credits++; /* For attaching the new dirent block to the
+			    * dx_root */
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_do_extend_dir(osb->sb, handle, dir, parent_fe_bh,
+				     data_ac, meta_ac, &new_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(dir), new_bh);
+
+	status = ocfs2_journal_access_db(handle, INODE_CACHE(dir), new_bh,
+					 OCFS2_JOURNAL_ACCESS_CREATE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	memset(new_bh->b_data, 0, sb->s_blocksize);
+
+	de = (struct ocfs2_dir_entry *) new_bh->b_data;
+	de->inode = 0;
+	if (ocfs2_supports_dir_trailer(dir)) {
+		de->rec_len = cpu_to_le16(ocfs2_dir_trailer_blk_off(sb));
+
+		ocfs2_init_dir_trailer(dir, new_bh, le16_to_cpu(de->rec_len));
+
+		if (ocfs2_dir_indexed(dir)) {
+			status = ocfs2_dx_dir_link_trailer(dir, handle,
+							   dx_root_bh, new_bh);
+			if (status) {
+				mlog_errno(status);
+				goto bail;
+			}
+		}
+	} else {
+		de->rec_len = cpu_to_le16(sb->s_blocksize);
+	}
+	ocfs2_update_inode_fsync_trans(handle, dir, 1);
+	ocfs2_journal_dirty(handle, new_bh);
+
+	dir_i_size += dir->i_sb->s_blocksize;
+	i_size_write(dir, dir_i_size);
+	dir->i_blocks = ocfs2_inode_sector_count(dir);
+	status = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+bail_bh:
+	*new_de_bh = new_bh;
+	get_bh(*new_de_bh);
+bail:
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+	if (drop_alloc_sem)
+		up_write(&OCFS2_I(dir)->ip_alloc_sem);
+
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	brelse(new_bh);
+
+	return status;
+}
+
+static int ocfs2_find_dir_space_id(struct inode *dir, struct buffer_head *di_bh,
+				   const char *name, int namelen,
+				   struct buffer_head **ret_de_bh,
+				   unsigned int *blocks_wanted)
+{
+	int ret;
+	struct super_block *sb = dir->i_sb;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_dir_entry *de, *last_de = NULL;
+	char *de_buf, *limit;
+	unsigned long offset = 0;
+	unsigned int rec_len, new_rec_len, free_space = dir->i_sb->s_blocksize;
+
+	/*
+	 * This calculates how many free bytes we'd have in block zero, should
+	 * this function force expansion to an extent tree.
+	 */
+	if (ocfs2_new_dir_wants_trailer(dir))
+		free_space = ocfs2_dir_trailer_blk_off(sb) - i_size_read(dir);
+	else
+		free_space = dir->i_sb->s_blocksize - i_size_read(dir);
+
+	de_buf = di->id2.i_data.id_data;
+	limit = de_buf + i_size_read(dir);
+	rec_len = OCFS2_DIR_REC_LEN(namelen);
+
+	while (de_buf < limit) {
+		de = (struct ocfs2_dir_entry *)de_buf;
+
+		if (!ocfs2_check_dir_entry(dir, de, di_bh, offset)) {
+			ret = -ENOENT;
+			goto out;
+		}
+		if (ocfs2_match(namelen, name, de)) {
+			ret = -EEXIST;
+			goto out;
+		}
+		/*
+		 * No need to check for a trailing dirent record here as
+		 * they're not used for inline dirs.
+		 */
+
+		if (ocfs2_dirent_would_fit(de, rec_len)) {
+			/* Ok, we found a spot. Return this bh and let
+			 * the caller actually fill it in. */
+			*ret_de_bh = di_bh;
+			get_bh(*ret_de_bh);
+			ret = 0;
+			goto out;
+		}
+
+		last_de = de;
+		de_buf += le16_to_cpu(de->rec_len);
+		offset += le16_to_cpu(de->rec_len);
+	}
+
+	/*
+	 * We're going to require expansion of the directory - figure
+	 * out how many blocks we'll need so that a place for the
+	 * dirent can be found.
+	 */
+	*blocks_wanted = 1;
+	new_rec_len = le16_to_cpu(last_de->rec_len) + free_space;
+	if (new_rec_len < (rec_len + OCFS2_DIR_REC_LEN(last_de->name_len)))
+		*blocks_wanted = 2;
+
+	ret = -ENOSPC;
+out:
+	return ret;
+}
+
+static int ocfs2_find_dir_space_el(struct inode *dir, const char *name,
+				   int namelen, struct buffer_head **ret_de_bh)
+{
+	unsigned long offset;
+	struct buffer_head *bh = NULL;
+	unsigned short rec_len;
+	struct ocfs2_dir_entry *de;
+	struct super_block *sb = dir->i_sb;
+	int status;
+	int blocksize = dir->i_sb->s_blocksize;
+
+	status = ocfs2_read_dir_block(dir, 0, &bh, 0);
+	if (status)
+		goto bail;
+
+	rec_len = OCFS2_DIR_REC_LEN(namelen);
+	offset = 0;
+	de = (struct ocfs2_dir_entry *) bh->b_data;
+	while (1) {
+		if ((char *)de >= sb->s_blocksize + bh->b_data) {
+			brelse(bh);
+			bh = NULL;
+
+			if (i_size_read(dir) <= offset) {
+				/*
+				 * Caller will have to expand this
+				 * directory.
+				 */
+				status = -ENOSPC;
+				goto bail;
+			}
+			status = ocfs2_read_dir_block(dir,
+					     offset >> sb->s_blocksize_bits,
+					     &bh, 0);
+			if (status)
+				goto bail;
+
+			/* move to next block */
+			de = (struct ocfs2_dir_entry *) bh->b_data;
+		}
+		if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
+			status = -ENOENT;
+			goto bail;
+		}
+		if (ocfs2_match(namelen, name, de)) {
+			status = -EEXIST;
+			goto bail;
+		}
+
+		if (ocfs2_skip_dir_trailer(dir, de, offset % blocksize,
+					   blocksize))
+			goto next;
+
+		if (ocfs2_dirent_would_fit(de, rec_len)) {
+			/* Ok, we found a spot. Return this bh and let
+			 * the caller actually fill it in. */
+			*ret_de_bh = bh;
+			get_bh(*ret_de_bh);
+			status = 0;
+			goto bail;
+		}
+next:
+		offset += le16_to_cpu(de->rec_len);
+		de = (struct ocfs2_dir_entry *)((char *) de + le16_to_cpu(de->rec_len));
+	}
+
+bail:
+	brelse(bh);
+	if (status)
+		mlog_errno(status);
+
+	return status;
+}
+
+static int dx_leaf_sort_cmp(const void *a, const void *b)
+{
+	const struct ocfs2_dx_entry *entry1 = a;
+	const struct ocfs2_dx_entry *entry2 = b;
+	u32 major_hash1 = le32_to_cpu(entry1->dx_major_hash);
+	u32 major_hash2 = le32_to_cpu(entry2->dx_major_hash);
+	u32 minor_hash1 = le32_to_cpu(entry1->dx_minor_hash);
+	u32 minor_hash2 = le32_to_cpu(entry2->dx_minor_hash);
+
+	if (major_hash1 > major_hash2)
+		return 1;
+	if (major_hash1 < major_hash2)
+		return -1;
+
+	/*
+	 * It is not strictly necessary to sort by minor
+	 */
+	if (minor_hash1 > minor_hash2)
+		return 1;
+	if (minor_hash1 < minor_hash2)
+		return -1;
+	return 0;
+}
+
+static void dx_leaf_sort_swap(void *a, void *b, int size)
+{
+	struct ocfs2_dx_entry *entry1 = a;
+	struct ocfs2_dx_entry *entry2 = b;
+	struct ocfs2_dx_entry tmp;
+
+	BUG_ON(size != sizeof(*entry1));
+
+	tmp = *entry1;
+	*entry1 = *entry2;
+	*entry2 = tmp;
+}
+
+static int ocfs2_dx_leaf_same_major(struct ocfs2_dx_leaf *dx_leaf)
+{
+	struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
+	int i, num = le16_to_cpu(dl_list->de_num_used);
+
+	for (i = 0; i < (num - 1); i++) {
+		if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) !=
+		    le32_to_cpu(dl_list->de_entries[i + 1].dx_major_hash))
+			return 0;
+	}
+
+	return 1;
+}
+
+/*
+ * Find the optimal value to split this leaf on. This expects the leaf
+ * entries to be in sorted order.
+ *
+ * leaf_cpos is the cpos of the leaf we're splitting. insert_hash is
+ * the hash we want to insert.
+ *
+ * This function is only concerned with the major hash - that which
+ * determines which cluster an item belongs to.
+ */
+static int ocfs2_dx_dir_find_leaf_split(struct ocfs2_dx_leaf *dx_leaf,
+					u32 leaf_cpos, u32 insert_hash,
+					u32 *split_hash)
+{
+	struct ocfs2_dx_entry_list *dl_list = &dx_leaf->dl_list;
+	int i, num_used = le16_to_cpu(dl_list->de_num_used);
+	int allsame;
+
+	/*
+	 * There's a couple rare, but nasty corner cases we have to
+	 * check for here. All of them involve a leaf where all value
+	 * have the same hash, which is what we look for first.
+	 *
+	 * Most of the time, all of the above is false, and we simply
+	 * pick the median value for a split.
+	 */
+	allsame = ocfs2_dx_leaf_same_major(dx_leaf);
+	if (allsame) {
+		u32 val = le32_to_cpu(dl_list->de_entries[0].dx_major_hash);
+
+		if (val == insert_hash) {
+			/*
+			 * No matter where we would choose to split,
+			 * the new entry would want to occupy the same
+			 * block as these. Since there's no space left
+			 * in their existing block, we know there
+			 * won't be space after the split.
+			 */
+			return -ENOSPC;
+		}
+
+		if (val == leaf_cpos) {
+			/*
+			 * Because val is the same as leaf_cpos (which
+			 * is the smallest value this leaf can have),
+			 * yet is not equal to insert_hash, then we
+			 * know that insert_hash *must* be larger than
+			 * val (and leaf_cpos). At least cpos+1 in value.
+			 *
+			 * We also know then, that there cannot be an
+			 * adjacent extent (otherwise we'd be looking
+			 * at it). Choosing this value gives us a
+			 * chance to get some contiguousness.
+			 */
+			*split_hash = leaf_cpos + 1;
+			return 0;
+		}
+
+		if (val > insert_hash) {
+			/*
+			 * val can not be the same as insert hash, and
+			 * also must be larger than leaf_cpos. Also,
+			 * we know that there can't be a leaf between
+			 * cpos and val, otherwise the entries with
+			 * hash 'val' would be there.
+			 */
+			*split_hash = val;
+			return 0;
+		}
+
+		*split_hash = insert_hash;
+		return 0;
+	}
+
+	/*
+	 * Since the records are sorted and the checks above
+	 * guaranteed that not all records in this block are the same,
+	 * we simple travel forward, from the median, and pick the 1st
+	 * record whose value is larger than leaf_cpos.
+	 */
+	for (i = (num_used / 2); i < num_used; i++)
+		if (le32_to_cpu(dl_list->de_entries[i].dx_major_hash) >
+		    leaf_cpos)
+			break;
+
+	BUG_ON(i == num_used); /* Should be impossible */
+	*split_hash = le32_to_cpu(dl_list->de_entries[i].dx_major_hash);
+	return 0;
+}
+
+/*
+ * Transfer all entries in orig_dx_leaves whose major hash is equal to or
+ * larger than split_hash into new_dx_leaves. We use a temporary
+ * buffer (tmp_dx_leaf) to make the changes to the original leaf blocks.
+ *
+ * Since the block offset inside a leaf (cluster) is a constant mask
+ * of minor_hash, we can optimize - an item at block offset X within
+ * the original cluster, will be at offset X within the new cluster.
+ */
+static void ocfs2_dx_dir_transfer_leaf(struct inode *dir, u32 split_hash,
+				       handle_t *handle,
+				       struct ocfs2_dx_leaf *tmp_dx_leaf,
+				       struct buffer_head **orig_dx_leaves,
+				       struct buffer_head **new_dx_leaves,
+				       int num_dx_leaves)
+{
+	int i, j, num_used;
+	u32 major_hash;
+	struct ocfs2_dx_leaf *orig_dx_leaf, *new_dx_leaf;
+	struct ocfs2_dx_entry_list *orig_list, *new_list, *tmp_list;
+	struct ocfs2_dx_entry *dx_entry;
+
+	tmp_list = &tmp_dx_leaf->dl_list;
+
+	for (i = 0; i < num_dx_leaves; i++) {
+		orig_dx_leaf = (struct ocfs2_dx_leaf *) orig_dx_leaves[i]->b_data;
+		orig_list = &orig_dx_leaf->dl_list;
+		new_dx_leaf = (struct ocfs2_dx_leaf *) new_dx_leaves[i]->b_data;
+		new_list = &new_dx_leaf->dl_list;
+
+		num_used = le16_to_cpu(orig_list->de_num_used);
+
+		memcpy(tmp_dx_leaf, orig_dx_leaf, dir->i_sb->s_blocksize);
+		tmp_list->de_num_used = cpu_to_le16(0);
+		memset(&tmp_list->de_entries, 0, sizeof(*dx_entry)*num_used);
+
+		for (j = 0; j < num_used; j++) {
+			dx_entry = &orig_list->de_entries[j];
+			major_hash = le32_to_cpu(dx_entry->dx_major_hash);
+			if (major_hash >= split_hash)
+				ocfs2_dx_dir_leaf_insert_tail(new_dx_leaf,
+							      dx_entry);
+			else
+				ocfs2_dx_dir_leaf_insert_tail(tmp_dx_leaf,
+							      dx_entry);
+		}
+		memcpy(orig_dx_leaf, tmp_dx_leaf, dir->i_sb->s_blocksize);
+
+		ocfs2_journal_dirty(handle, orig_dx_leaves[i]);
+		ocfs2_journal_dirty(handle, new_dx_leaves[i]);
+	}
+}
+
+static int ocfs2_dx_dir_rebalance_credits(struct ocfs2_super *osb,
+					  struct ocfs2_dx_root_block *dx_root)
+{
+	int credits = ocfs2_clusters_to_blocks(osb->sb, 2);
+
+	credits += ocfs2_calc_extend_credits(osb->sb, &dx_root->dr_list);
+	credits += ocfs2_quota_trans_credits(osb->sb);
+	return credits;
+}
+
+/*
+ * Find the median value in dx_leaf_bh and allocate a new leaf to move
+ * half our entries into.
+ */
+static int ocfs2_dx_dir_rebalance(struct ocfs2_super *osb, struct inode *dir,
+				  struct buffer_head *dx_root_bh,
+				  struct buffer_head *dx_leaf_bh,
+				  struct ocfs2_dx_hinfo *hinfo, u32 leaf_cpos,
+				  u64 leaf_blkno)
+{
+	struct ocfs2_dx_leaf *dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
+	int credits, ret, i, num_used, did_quota = 0;
+	u32 cpos, split_hash, insert_hash = hinfo->major_hash;
+	u64 orig_leaves_start;
+	int num_dx_leaves;
+	struct buffer_head **orig_dx_leaves = NULL;
+	struct buffer_head **new_dx_leaves = NULL;
+	struct ocfs2_alloc_context *data_ac = NULL, *meta_ac = NULL;
+	struct ocfs2_extent_tree et;
+	handle_t *handle = NULL;
+	struct ocfs2_dx_root_block *dx_root;
+	struct ocfs2_dx_leaf *tmp_dx_leaf = NULL;
+
+	trace_ocfs2_dx_dir_rebalance((unsigned long long)OCFS2_I(dir)->ip_blkno,
+				     (unsigned long long)leaf_blkno,
+				     insert_hash);
+
+	ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
+
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+	/*
+	 * XXX: This is a rather large limit. We should use a more
+	 * realistic value.
+	 */
+	if (le32_to_cpu(dx_root->dr_clusters) == UINT_MAX)
+		return -ENOSPC;
+
+	num_used = le16_to_cpu(dx_leaf->dl_list.de_num_used);
+	if (num_used < le16_to_cpu(dx_leaf->dl_list.de_count)) {
+		mlog(ML_ERROR, "DX Dir: %llu, Asked to rebalance empty leaf: "
+		     "%llu, %d\n", (unsigned long long)OCFS2_I(dir)->ip_blkno,
+		     (unsigned long long)leaf_blkno, num_used);
+		ret = -EIO;
+		goto out;
+	}
+
+	orig_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
+	if (!orig_dx_leaves) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	new_dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, NULL);
+	if (!new_dx_leaves) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_lock_allocators(dir, &et, 1, 0, &data_ac, &meta_ac);
+	if (ret) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	credits = ocfs2_dx_dir_rebalance_credits(osb, dx_root);
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = dquot_alloc_space_nodirty(dir,
+				       ocfs2_clusters_to_bytes(dir->i_sb, 1));
+	if (ret)
+		goto out_commit;
+	did_quota = 1;
+
+	ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir), dx_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	/*
+	 * This block is changing anyway, so we can sort it in place.
+	 */
+	sort(dx_leaf->dl_list.de_entries, num_used,
+	     sizeof(struct ocfs2_dx_entry), dx_leaf_sort_cmp,
+	     dx_leaf_sort_swap);
+
+	ocfs2_journal_dirty(handle, dx_leaf_bh);
+
+	ret = ocfs2_dx_dir_find_leaf_split(dx_leaf, leaf_cpos, insert_hash,
+					   &split_hash);
+	if (ret) {
+		mlog_errno(ret);
+		goto  out_commit;
+	}
+
+	trace_ocfs2_dx_dir_rebalance_split(leaf_cpos, split_hash, insert_hash);
+
+	/*
+	 * We have to carefully order operations here. There are items
+	 * which want to be in the new cluster before insert, but in
+	 * order to put those items in the new cluster, we alter the
+	 * old cluster. A failure to insert gets nasty.
+	 *
+	 * So, start by reserving writes to the old
+	 * cluster. ocfs2_dx_dir_new_cluster will reserve writes on
+	 * the new cluster for us, before inserting it. The insert
+	 * won't happen if there's an error before that. Once the
+	 * insert is done then, we can transfer from one leaf into the
+	 * other without fear of hitting any error.
+	 */
+
+	/*
+	 * The leaf transfer wants some scratch space so that we don't
+	 * wind up doing a bunch of expensive memmove().
+	 */
+	tmp_dx_leaf = kmalloc(osb->sb->s_blocksize, GFP_NOFS);
+	if (!tmp_dx_leaf) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	orig_leaves_start = ocfs2_block_to_cluster_start(dir->i_sb, leaf_blkno);
+	ret = ocfs2_read_dx_leaves(dir, orig_leaves_start, num_dx_leaves,
+				   orig_dx_leaves);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	cpos = split_hash;
+	ret = ocfs2_dx_dir_new_cluster(dir, &et, cpos, handle,
+				       data_ac, meta_ac, new_dx_leaves,
+				       num_dx_leaves);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	for (i = 0; i < num_dx_leaves; i++) {
+		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
+					      orig_dx_leaves[i],
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+
+		ret = ocfs2_journal_access_dl(handle, INODE_CACHE(dir),
+					      new_dx_leaves[i],
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+	}
+
+	ocfs2_dx_dir_transfer_leaf(dir, split_hash, handle, tmp_dx_leaf,
+				   orig_dx_leaves, new_dx_leaves, num_dx_leaves);
+
+out_commit:
+	if (ret < 0 && did_quota)
+		dquot_free_space_nodirty(dir,
+				ocfs2_clusters_to_bytes(dir->i_sb, 1));
+
+	ocfs2_update_inode_fsync_trans(handle, dir, 1);
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	if (orig_dx_leaves || new_dx_leaves) {
+		for (i = 0; i < num_dx_leaves; i++) {
+			if (orig_dx_leaves)
+				brelse(orig_dx_leaves[i]);
+			if (new_dx_leaves)
+				brelse(new_dx_leaves[i]);
+		}
+		kfree(orig_dx_leaves);
+		kfree(new_dx_leaves);
+	}
+
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+
+	kfree(tmp_dx_leaf);
+	return ret;
+}
+
+static int ocfs2_find_dir_space_dx(struct ocfs2_super *osb, struct inode *dir,
+				   struct buffer_head *di_bh,
+				   struct buffer_head *dx_root_bh,
+				   const char *name, int namelen,
+				   struct ocfs2_dir_lookup_result *lookup)
+{
+	int ret, rebalanced = 0;
+	struct ocfs2_dx_root_block *dx_root;
+	struct buffer_head *dx_leaf_bh = NULL;
+	struct ocfs2_dx_leaf *dx_leaf;
+	u64 blkno;
+	u32 leaf_cpos;
+
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+
+restart_search:
+	ret = ocfs2_dx_dir_lookup(dir, &dx_root->dr_list, &lookup->dl_hinfo,
+				  &leaf_cpos, &blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_dx_leaf(dir, blkno, &dx_leaf_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	dx_leaf = (struct ocfs2_dx_leaf *)dx_leaf_bh->b_data;
+
+	if (le16_to_cpu(dx_leaf->dl_list.de_num_used) >=
+	    le16_to_cpu(dx_leaf->dl_list.de_count)) {
+		if (rebalanced) {
+			/*
+			 * Rebalancing should have provided us with
+			 * space in an appropriate leaf.
+			 *
+			 * XXX: Is this an abnormal condition then?
+			 * Should we print a message here?
+			 */
+			ret = -ENOSPC;
+			goto out;
+		}
+
+		ret = ocfs2_dx_dir_rebalance(osb, dir, dx_root_bh, dx_leaf_bh,
+					     &lookup->dl_hinfo, leaf_cpos,
+					     blkno);
+		if (ret) {
+			if (ret != -ENOSPC)
+				mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * Restart the lookup. The rebalance might have
+		 * changed which block our item fits into. Mark our
+		 * progress, so we only execute this once.
+		 */
+		brelse(dx_leaf_bh);
+		dx_leaf_bh = NULL;
+		rebalanced = 1;
+		goto restart_search;
+	}
+
+	lookup->dl_dx_leaf_bh = dx_leaf_bh;
+	dx_leaf_bh = NULL;
+
+out:
+	brelse(dx_leaf_bh);
+	return ret;
+}
+
+static int ocfs2_search_dx_free_list(struct inode *dir,
+				     struct buffer_head *dx_root_bh,
+				     int namelen,
+				     struct ocfs2_dir_lookup_result *lookup)
+{
+	int ret = -ENOSPC;
+	struct buffer_head *leaf_bh = NULL, *prev_leaf_bh = NULL;
+	struct ocfs2_dir_block_trailer *db;
+	u64 next_block;
+	int rec_len = OCFS2_DIR_REC_LEN(namelen);
+	struct ocfs2_dx_root_block *dx_root;
+
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+	next_block = le64_to_cpu(dx_root->dr_free_blk);
+
+	while (next_block) {
+		brelse(prev_leaf_bh);
+		prev_leaf_bh = leaf_bh;
+		leaf_bh = NULL;
+
+		ret = ocfs2_read_dir_block_direct(dir, next_block, &leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		db = ocfs2_trailer_from_bh(leaf_bh, dir->i_sb);
+		if (rec_len <= le16_to_cpu(db->db_free_rec_len)) {
+			lookup->dl_leaf_bh = leaf_bh;
+			lookup->dl_prev_leaf_bh = prev_leaf_bh;
+			leaf_bh = NULL;
+			prev_leaf_bh = NULL;
+			break;
+		}
+
+		next_block = le64_to_cpu(db->db_free_next);
+	}
+
+	if (!next_block)
+		ret = -ENOSPC;
+
+out:
+
+	brelse(leaf_bh);
+	brelse(prev_leaf_bh);
+	return ret;
+}
+
+static int ocfs2_expand_inline_dx_root(struct inode *dir,
+				       struct buffer_head *dx_root_bh)
+{
+	int ret, num_dx_leaves, i, j, did_quota = 0;
+	struct buffer_head **dx_leaves = NULL;
+	struct ocfs2_extent_tree et;
+	u64 insert_blkno;
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	handle_t *handle = NULL;
+	struct ocfs2_dx_root_block *dx_root;
+	struct ocfs2_dx_entry_list *entry_list;
+	struct ocfs2_dx_entry *dx_entry;
+	struct ocfs2_dx_leaf *target_leaf;
+
+	ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	dx_leaves = ocfs2_dx_dir_kmalloc_leaves(osb->sb, &num_dx_leaves);
+	if (!dx_leaves) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_calc_dxi_expand_credits(osb->sb));
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = dquot_alloc_space_nodirty(dir,
+				       ocfs2_clusters_to_bytes(osb->sb, 1));
+	if (ret)
+		goto out_commit;
+	did_quota = 1;
+
+	/*
+	 * We do this up front, before the allocation, so that a
+	 * failure to add the dx_root_bh to the journal won't result
+	 * us losing clusters.
+	 */
+	ret = ocfs2_journal_access_dr(handle, INODE_CACHE(dir), dx_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = __ocfs2_dx_dir_new_cluster(dir, 0, handle, data_ac, dx_leaves,
+					 num_dx_leaves, &insert_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	/*
+	 * Transfer the entries from our dx_root into the appropriate
+	 * block
+	 */
+	dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
+	entry_list = &dx_root->dr_entries;
+
+	for (i = 0; i < le16_to_cpu(entry_list->de_num_used); i++) {
+		dx_entry = &entry_list->de_entries[i];
+
+		j = __ocfs2_dx_dir_hash_idx(osb,
+					    le32_to_cpu(dx_entry->dx_minor_hash));
+		target_leaf = (struct ocfs2_dx_leaf *)dx_leaves[j]->b_data;
+
+		ocfs2_dx_dir_leaf_insert_tail(target_leaf, dx_entry);
+
+		/* Each leaf has been passed to the journal already
+		 * via __ocfs2_dx_dir_new_cluster() */
+	}
+
+	dx_root->dr_flags &= ~OCFS2_DX_FLAG_INLINE;
+	memset(&dx_root->dr_list, 0, osb->sb->s_blocksize -
+	       offsetof(struct ocfs2_dx_root_block, dr_list));
+	dx_root->dr_list.l_count =
+		cpu_to_le16(ocfs2_extent_recs_per_dx_root(osb->sb));
+
+	/* This should never fail considering we start with an empty
+	 * dx_root. */
+	ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
+	ret = ocfs2_insert_extent(handle, &et, 0, insert_blkno, 1, 0, NULL);
+	if (ret)
+		mlog_errno(ret);
+	did_quota = 0;
+
+	ocfs2_update_inode_fsync_trans(handle, dir, 1);
+	ocfs2_journal_dirty(handle, dx_root_bh);
+
+out_commit:
+	if (ret < 0 && did_quota)
+		dquot_free_space_nodirty(dir,
+					  ocfs2_clusters_to_bytes(dir->i_sb, 1));
+
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+
+	if (dx_leaves) {
+		for (i = 0; i < num_dx_leaves; i++)
+			brelse(dx_leaves[i]);
+		kfree(dx_leaves);
+	}
+	return ret;
+}
+
+static int ocfs2_inline_dx_has_space(struct buffer_head *dx_root_bh)
+{
+	struct ocfs2_dx_root_block *dx_root;
+	struct ocfs2_dx_entry_list *entry_list;
+
+	dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
+	entry_list = &dx_root->dr_entries;
+
+	if (le16_to_cpu(entry_list->de_num_used) >=
+	    le16_to_cpu(entry_list->de_count))
+		return -ENOSPC;
+
+	return 0;
+}
+
+static int ocfs2_prepare_dx_dir_for_insert(struct inode *dir,
+					   struct buffer_head *di_bh,
+					   const char *name,
+					   int namelen,
+					   struct ocfs2_dir_lookup_result *lookup)
+{
+	int ret, free_dx_root = 1;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct buffer_head *dx_root_bh = NULL;
+	struct buffer_head *leaf_bh = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_dx_root_block *dx_root;
+
+	ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+	if (le32_to_cpu(dx_root->dr_num_entries) == OCFS2_DX_ENTRIES_MAX) {
+		ret = -ENOSPC;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (ocfs2_dx_root_inline(dx_root)) {
+		ret = ocfs2_inline_dx_has_space(dx_root_bh);
+
+		if (ret == 0)
+			goto search_el;
+
+		/*
+		 * We ran out of room in the root block. Expand it to
+		 * an extent, then allow ocfs2_find_dir_space_dx to do
+		 * the rest.
+		 */
+		ret = ocfs2_expand_inline_dx_root(dir, dx_root_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	/*
+	 * Insert preparation for an indexed directory is split into two
+	 * steps. The call to find_dir_space_dx reserves room in the index for
+	 * an additional item. If we run out of space there, it's a real error
+	 * we can't continue on.
+	 */
+	ret = ocfs2_find_dir_space_dx(osb, dir, di_bh, dx_root_bh, name,
+				      namelen, lookup);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+search_el:
+	/*
+	 * Next, we need to find space in the unindexed tree. This call
+	 * searches using the free space linked list. If the unindexed tree
+	 * lacks sufficient space, we'll expand it below. The expansion code
+	 * is smart enough to add any new blocks to the free space list.
+	 */
+	ret = ocfs2_search_dx_free_list(dir, dx_root_bh, namelen, lookup);
+	if (ret && ret != -ENOSPC) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Do this up here - ocfs2_extend_dir might need the dx_root */
+	lookup->dl_dx_root_bh = dx_root_bh;
+	free_dx_root = 0;
+
+	if (ret == -ENOSPC) {
+		ret = ocfs2_extend_dir(osb, dir, di_bh, 1, lookup, &leaf_bh);
+
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * We make the assumption here that new leaf blocks are added
+		 * to the front of our free list.
+		 */
+		lookup->dl_prev_leaf_bh = NULL;
+		lookup->dl_leaf_bh = leaf_bh;
+	}
+
+out:
+	if (free_dx_root)
+		brelse(dx_root_bh);
+	return ret;
+}
+
+/*
+ * Get a directory ready for insert. Any directory allocation required
+ * happens here. Success returns zero, and enough context in the dir
+ * lookup result that ocfs2_add_entry() will be able complete the task
+ * with minimal performance impact.
+ */
+int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
+				 struct inode *dir,
+				 struct buffer_head *parent_fe_bh,
+				 const char *name,
+				 int namelen,
+				 struct ocfs2_dir_lookup_result *lookup)
+{
+	int ret;
+	unsigned int blocks_wanted = 1;
+	struct buffer_head *bh = NULL;
+
+	trace_ocfs2_prepare_dir_for_insert(
+		(unsigned long long)OCFS2_I(dir)->ip_blkno, namelen);
+
+	if (!namelen) {
+		ret = -EINVAL;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Do this up front to reduce confusion.
+	 *
+	 * The directory might start inline, then be turned into an
+	 * indexed one, in which case we'd need to hash deep inside
+	 * ocfs2_find_dir_space_id(). Since
+	 * ocfs2_prepare_dx_dir_for_insert() also needs this hash
+	 * done, there seems no point in spreading out the calls. We
+	 * can optimize away the case where the file system doesn't
+	 * support indexing.
+	 */
+	if (ocfs2_supports_indexed_dirs(osb))
+		ocfs2_dx_dir_name_hash(dir, name, namelen, &lookup->dl_hinfo);
+
+	if (ocfs2_dir_indexed(dir)) {
+		ret = ocfs2_prepare_dx_dir_for_insert(dir, parent_fe_bh,
+						      name, namelen, lookup);
+		if (ret)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	if (OCFS2_I(dir)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		ret = ocfs2_find_dir_space_id(dir, parent_fe_bh, name,
+					      namelen, &bh, &blocks_wanted);
+	} else
+		ret = ocfs2_find_dir_space_el(dir, name, namelen, &bh);
+
+	if (ret && ret != -ENOSPC) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (ret == -ENOSPC) {
+		/*
+		 * We have to expand the directory to add this name.
+		 */
+		BUG_ON(bh);
+
+		ret = ocfs2_extend_dir(osb, dir, parent_fe_bh, blocks_wanted,
+				       lookup, &bh);
+		if (ret) {
+			if (ret != -ENOSPC)
+				mlog_errno(ret);
+			goto out;
+		}
+
+		BUG_ON(!bh);
+	}
+
+	lookup->dl_leaf_bh = bh;
+	bh = NULL;
+out:
+	brelse(bh);
+	return ret;
+}
+
+static int ocfs2_dx_dir_remove_index(struct inode *dir,
+				     struct buffer_head *di_bh,
+				     struct buffer_head *dx_root_bh)
+{
+	int ret;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_dx_root_block *dx_root;
+	struct inode *dx_alloc_inode = NULL;
+	struct buffer_head *dx_alloc_bh = NULL;
+	handle_t *handle;
+	u64 blk;
+	u16 bit;
+	u64 bg_blkno;
+
+	dx_root = (struct ocfs2_dx_root_block *) dx_root_bh->b_data;
+
+	dx_alloc_inode = ocfs2_get_system_file_inode(osb,
+					EXTENT_ALLOC_SYSTEM_INODE,
+					le16_to_cpu(dx_root->dr_suballoc_slot));
+	if (!dx_alloc_inode) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+	mutex_lock(&dx_alloc_inode->i_mutex);
+
+	ret = ocfs2_inode_lock(dx_alloc_inode, &dx_alloc_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_mutex;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_DX_ROOT_REMOVE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(dir), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	spin_lock(&OCFS2_I(dir)->ip_lock);
+	OCFS2_I(dir)->ip_dyn_features &= ~OCFS2_INDEXED_DIR_FL;
+	di->i_dyn_features = cpu_to_le16(OCFS2_I(dir)->ip_dyn_features);
+	spin_unlock(&OCFS2_I(dir)->ip_lock);
+	di->i_dx_root = cpu_to_le64(0ULL);
+	ocfs2_update_inode_fsync_trans(handle, dir, 1);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+	blk = le64_to_cpu(dx_root->dr_blkno);
+	bit = le16_to_cpu(dx_root->dr_suballoc_bit);
+	if (dx_root->dr_suballoc_loc)
+		bg_blkno = le64_to_cpu(dx_root->dr_suballoc_loc);
+	else
+		bg_blkno = ocfs2_which_suballoc_group(blk, bit);
+	ret = ocfs2_free_suballoc_bits(handle, dx_alloc_inode, dx_alloc_bh,
+				       bit, bg_blkno, 1);
+	if (ret)
+		mlog_errno(ret);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out_unlock:
+	ocfs2_inode_unlock(dx_alloc_inode, 1);
+
+out_mutex:
+	mutex_unlock(&dx_alloc_inode->i_mutex);
+	brelse(dx_alloc_bh);
+out:
+	iput(dx_alloc_inode);
+	return ret;
+}
+
+int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh)
+{
+	int ret;
+	unsigned int uninitialized_var(clen);
+	u32 major_hash = UINT_MAX, p_cpos, uninitialized_var(cpos);
+	u64 uninitialized_var(blkno);
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct buffer_head *dx_root_bh = NULL;
+	struct ocfs2_dx_root_block *dx_root;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct ocfs2_extent_tree et;
+
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	if (!ocfs2_dir_indexed(dir))
+		return 0;
+
+	ret = ocfs2_read_dx_root(dir, di, &dx_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	dx_root = (struct ocfs2_dx_root_block *)dx_root_bh->b_data;
+
+	if (ocfs2_dx_root_inline(dx_root))
+		goto remove_index;
+
+	ocfs2_init_dx_root_extent_tree(&et, INODE_CACHE(dir), dx_root_bh);
+
+	/* XXX: What if dr_clusters is too large? */
+	while (le32_to_cpu(dx_root->dr_clusters)) {
+		ret = ocfs2_dx_dir_lookup_rec(dir, &dx_root->dr_list,
+					      major_hash, &cpos, &blkno, &clen);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		p_cpos = ocfs2_blocks_to_clusters(dir->i_sb, blkno);
+
+		ret = ocfs2_remove_btree_range(dir, &et, cpos, p_cpos, clen, 0,
+					       &dealloc, 0, false);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (cpos == 0)
+			break;
+
+		major_hash = cpos - 1;
+	}
+
+remove_index:
+	ret = ocfs2_dx_dir_remove_index(dir, di_bh, dx_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_remove_from_cache(INODE_CACHE(dir), dx_root_bh);
+out:
+	ocfs2_schedule_truncate_log_flush(osb, 1);
+	ocfs2_run_deallocs(osb, &dealloc);
+
+	brelse(dx_root_bh);
+	return ret;
+}
diff --git a/fs/ocfs2/dir.h b/fs/ocfs2/dir.h
new file mode 100644
index 000000000..3d8639f38
--- /dev/null
+++ b/fs/ocfs2/dir.h
@@ -0,0 +1,116 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dir.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_DIR_H
+#define OCFS2_DIR_H
+
+struct ocfs2_dx_hinfo {
+	u32	major_hash;
+	u32	minor_hash;
+};
+
+struct ocfs2_dir_lookup_result {
+	struct buffer_head		*dl_leaf_bh;	/* Unindexed leaf
+							 * block */
+	struct ocfs2_dir_entry		*dl_entry;	/* Target dirent in
+							 * unindexed leaf */
+
+	struct buffer_head		*dl_dx_root_bh;	/* Root of indexed
+							 * tree */
+
+	struct buffer_head		*dl_dx_leaf_bh;	/* Indexed leaf block */
+	struct ocfs2_dx_entry		*dl_dx_entry;	/* Target dx_entry in
+							 * indexed leaf */
+	struct ocfs2_dx_hinfo		dl_hinfo;	/* Name hash results */
+
+	struct buffer_head		*dl_prev_leaf_bh;/* Previous entry in
+							  * dir free space
+							  * list. NULL if
+							  * previous entry is
+							  * dx root block. */
+};
+
+void ocfs2_free_dir_lookup_result(struct ocfs2_dir_lookup_result *res);
+
+int ocfs2_find_entry(const char *name, int namelen,
+		     struct inode *dir,
+		     struct ocfs2_dir_lookup_result *lookup);
+int ocfs2_delete_entry(handle_t *handle,
+		       struct inode *dir,
+		       struct ocfs2_dir_lookup_result *res);
+int __ocfs2_add_entry(handle_t *handle,
+		      struct inode *dir,
+		      const char *name, int namelen,
+		      struct inode *inode, u64 blkno,
+		      struct buffer_head *parent_fe_bh,
+		      struct ocfs2_dir_lookup_result *lookup);
+static inline int ocfs2_add_entry(handle_t *handle,
+				  struct dentry *dentry,
+				  struct inode *inode, u64 blkno,
+				  struct buffer_head *parent_fe_bh,
+				  struct ocfs2_dir_lookup_result *lookup)
+{
+	return __ocfs2_add_entry(handle, d_inode(dentry->d_parent),
+				 dentry->d_name.name, dentry->d_name.len,
+				 inode, blkno, parent_fe_bh, lookup);
+}
+int ocfs2_update_entry(struct inode *dir, handle_t *handle,
+		       struct ocfs2_dir_lookup_result *res,
+		       struct inode *new_entry_inode);
+
+int ocfs2_check_dir_for_entry(struct inode *dir,
+			      const char *name,
+			      int namelen);
+int ocfs2_empty_dir(struct inode *inode);
+
+int ocfs2_find_files_on_disk(const char *name,
+			     int namelen,
+			     u64 *blkno,
+			     struct inode *inode,
+			     struct ocfs2_dir_lookup_result *res);
+int ocfs2_lookup_ino_from_name(struct inode *dir, const char *name,
+			       int namelen, u64 *blkno);
+int ocfs2_readdir(struct file *file, struct dir_context *ctx);
+int ocfs2_dir_foreach(struct inode *inode, struct dir_context *ctx);
+int ocfs2_prepare_dir_for_insert(struct ocfs2_super *osb,
+				 struct inode *dir,
+				 struct buffer_head *parent_fe_bh,
+				 const char *name,
+				 int namelen,
+				 struct ocfs2_dir_lookup_result *lookup);
+struct ocfs2_alloc_context;
+int ocfs2_fill_new_dir(struct ocfs2_super *osb,
+		       handle_t *handle,
+		       struct inode *parent,
+		       struct inode *inode,
+		       struct buffer_head *fe_bh,
+		       struct ocfs2_alloc_context *data_ac,
+		       struct ocfs2_alloc_context *meta_ac);
+
+int ocfs2_dx_dir_truncate(struct inode *dir, struct buffer_head *di_bh);
+
+struct ocfs2_dir_block_trailer *ocfs2_dir_trailer_from_size(int blocksize,
+							    void *data);
+#endif /* OCFS2_DIR_H */
diff --git a/fs/ocfs2/dlm/Makefile b/fs/ocfs2/dlm/Makefile
new file mode 100644
index 000000000..bd1aab1f4
--- /dev/null
+++ b/fs/ocfs2/dlm/Makefile
@@ -0,0 +1,7 @@
+ccflags-y := -Ifs/ocfs2
+
+obj-$(CONFIG_OCFS2_FS_O2CB) += ocfs2_dlm.o
+
+ocfs2_dlm-objs := dlmdomain.o dlmdebug.o dlmthread.o dlmrecovery.o \
+	dlmmaster.o dlmast.o dlmconvert.o dlmlock.o dlmunlock.o
+
diff --git a/fs/ocfs2/dlm/dlmapi.h b/fs/ocfs2/dlm/dlmapi.h
new file mode 100644
index 000000000..3cfa114aa
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmapi.h
@@ -0,0 +1,220 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmapi.h
+ *
+ * externally exported dlm interfaces
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef DLMAPI_H
+#define DLMAPI_H
+
+struct dlm_lock;
+struct dlm_ctxt;
+
+/* NOTE: changes made to this enum should be reflected in dlmdebug.c */
+enum dlm_status {
+	DLM_NORMAL = 0,           /*  0: request in progress */
+	DLM_GRANTED,              /*  1: request granted */
+	DLM_DENIED,               /*  2: request denied */
+	DLM_DENIED_NOLOCKS,       /*  3: request denied, out of system resources */
+	DLM_WORKING,              /*  4: async request in progress */
+	DLM_BLOCKED,              /*  5: lock request blocked */
+	DLM_BLOCKED_ORPHAN,       /*  6: lock request blocked by a orphan lock*/
+	DLM_DENIED_GRACE_PERIOD,  /*  7: topological change in progress */
+	DLM_SYSERR,               /*  8: system error */
+	DLM_NOSUPPORT,            /*  9: unsupported */
+	DLM_CANCELGRANT,          /* 10: can't cancel convert: already granted */
+	DLM_IVLOCKID,             /* 11: bad lockid */
+	DLM_SYNC,                 /* 12: synchronous request granted */
+	DLM_BADTYPE,              /* 13: bad resource type */
+	DLM_BADRESOURCE,          /* 14: bad resource handle */
+	DLM_MAXHANDLES,           /* 15: no more resource handles */
+	DLM_NOCLINFO,             /* 16: can't contact cluster manager */
+	DLM_NOLOCKMGR,            /* 17: can't contact lock manager */
+	DLM_NOPURGED,             /* 18: can't contact purge daemon */
+	DLM_BADARGS,              /* 19: bad api args */
+	DLM_VOID,                 /* 20: no status */
+	DLM_NOTQUEUED,            /* 21: NOQUEUE was specified and request failed */
+	DLM_IVBUFLEN,             /* 22: invalid resource name length */
+	DLM_CVTUNGRANT,           /* 23: attempted to convert ungranted lock */
+	DLM_BADPARAM,             /* 24: invalid lock mode specified */
+	DLM_VALNOTVALID,          /* 25: value block has been invalidated */
+	DLM_REJECTED,             /* 26: request rejected, unrecognized client */
+	DLM_ABORT,                /* 27: blocked lock request cancelled */
+	DLM_CANCEL,               /* 28: conversion request cancelled */
+	DLM_IVRESHANDLE,          /* 29: invalid resource handle */
+	DLM_DEADLOCK,             /* 30: deadlock recovery refused this request */
+	DLM_DENIED_NOASTS,        /* 31: failed to allocate AST */
+	DLM_FORWARD,              /* 32: request must wait for primary's response */
+	DLM_TIMEOUT,              /* 33: timeout value for lock has expired */
+	DLM_IVGROUPID,            /* 34: invalid group specification */
+	DLM_VERS_CONFLICT,        /* 35: version conflicts prevent request handling */
+	DLM_BAD_DEVICE_PATH,      /* 36: Locks device does not exist or path wrong */
+	DLM_NO_DEVICE_PERMISSION, /* 37: Client has insufficient pers for device */
+	DLM_NO_CONTROL_DEVICE,    /* 38: Cannot set options on opened device */
+
+	DLM_RECOVERING,           /* 39: extension, allows caller to fail a lock
+				     request if it is being recovered */
+	DLM_MIGRATING,            /* 40: extension, allows caller to fail a lock
+				     request if it is being migrated */
+	DLM_MAXSTATS,             /* 41: upper limit for return code validation */
+};
+
+/* for pretty-printing dlm_status error messages */
+const char *dlm_errmsg(enum dlm_status err);
+/* for pretty-printing dlm_status error names */
+const char *dlm_errname(enum dlm_status err);
+
+/* Eventually the DLM will use standard errno values, but in the
+ * meantime this lets us track dlm errors as they bubble up. When we
+ * bring its error reporting into line with the rest of the stack,
+ * these can just be replaced with calls to mlog_errno. */
+#define dlm_error(st) do {						\
+	if ((st) != DLM_RECOVERING &&					\
+	    (st) != DLM_MIGRATING &&					\
+	    (st) != DLM_FORWARD)					\
+		mlog(ML_ERROR, "dlm status = %s\n", dlm_errname((st)));	\
+} while (0)
+
+#define DLM_LKSB_UNUSED1           0x01
+#define DLM_LKSB_PUT_LVB           0x02
+#define DLM_LKSB_GET_LVB           0x04
+#define DLM_LKSB_UNUSED2           0x08
+#define DLM_LKSB_UNUSED3           0x10
+#define DLM_LKSB_UNUSED4           0x20
+#define DLM_LKSB_UNUSED5           0x40
+#define DLM_LKSB_UNUSED6           0x80
+
+#define DLM_LVB_LEN  64
+
+/* Callers are only allowed access to the lvb and status members of
+ * this struct. */
+struct dlm_lockstatus {
+	enum dlm_status status;
+	u32 flags;
+	struct dlm_lock *lockid;
+	char lvb[DLM_LVB_LEN];
+};
+
+/* Valid lock modes. */
+#define LKM_IVMODE      (-1)            /* invalid mode */
+#define LKM_NLMODE      0               /* null lock */
+#define LKM_CRMODE      1               /* concurrent read    unsupported */
+#define LKM_CWMODE      2               /* concurrent write   unsupported */
+#define LKM_PRMODE      3               /* protected read */
+#define LKM_PWMODE      4               /* protected write    unsupported */
+#define LKM_EXMODE      5               /* exclusive */
+#define LKM_MAXMODE     5
+#define LKM_MODEMASK    0xff
+
+/* Flags passed to dlmlock and dlmunlock:
+ * reserved: flags used by the "real" dlm
+ * only a few are supported by this dlm
+ * (U) = unsupported by ocfs2 dlm */
+#define LKM_ORPHAN       0x00000010  /* this lock is orphanable (U) */
+#define LKM_PARENTABLE   0x00000020  /* this lock was orphaned (U) */
+#define LKM_BLOCK        0x00000040  /* blocking lock request (U) */
+#define LKM_LOCAL        0x00000080  /* local lock request */
+#define LKM_VALBLK       0x00000100  /* lock value block request */
+#define LKM_NOQUEUE      0x00000200  /* non blocking request */
+#define LKM_CONVERT      0x00000400  /* conversion request */
+#define LKM_NODLCKWT     0x00000800  /* this lock wont deadlock (U) */
+#define LKM_UNLOCK       0x00001000  /* deallocate this lock */
+#define LKM_CANCEL       0x00002000  /* cancel conversion request */
+#define LKM_DEQALL       0x00004000  /* remove all locks held by proc (U) */
+#define LKM_INVVALBLK    0x00008000  /* invalidate lock value block */
+#define LKM_SYNCSTS      0x00010000  /* return synchronous status if poss (U) */
+#define LKM_TIMEOUT      0x00020000  /* lock request contains timeout (U) */
+#define LKM_SNGLDLCK     0x00040000  /* request can self-deadlock (U) */
+#define LKM_FINDLOCAL    0x00080000  /* find local lock request (U) */
+#define LKM_PROC_OWNED   0x00100000  /* owned by process, not group (U) */
+#define LKM_XID          0x00200000  /* use transaction id for deadlock (U) */
+#define LKM_XID_CONFLICT 0x00400000  /* do not allow lock inheritance (U) */
+#define LKM_FORCE        0x00800000  /* force unlock flag */
+#define LKM_REVVALBLK    0x01000000  /* temporary solution: re-validate
+					lock value block (U) */
+/* unused */
+#define LKM_UNUSED1      0x00000001  /* unused */
+#define LKM_UNUSED2      0x00000002  /* unused */
+#define LKM_UNUSED3      0x00000004  /* unused */
+#define LKM_UNUSED4      0x00000008  /* unused */
+#define LKM_UNUSED5      0x02000000  /* unused */
+#define LKM_UNUSED6      0x04000000  /* unused */
+#define LKM_UNUSED7      0x08000000  /* unused */
+
+/* ocfs2 extensions: internal only
+ * should never be used by caller */
+#define LKM_MIGRATION    0x10000000  /* extension: lockres is to be migrated
+					to another node */
+#define LKM_PUT_LVB      0x20000000  /* extension: lvb is being passed
+					should be applied to lockres */
+#define LKM_GET_LVB      0x40000000  /* extension: lvb should be copied
+					from lockres when lock is granted */
+#define LKM_RECOVERY     0x80000000  /* extension: flag for recovery lock
+					used to avoid recovery rwsem */
+
+
+typedef void (dlm_astlockfunc_t)(void *);
+typedef void (dlm_bastlockfunc_t)(void *, int);
+typedef void (dlm_astunlockfunc_t)(void *, enum dlm_status);
+
+enum dlm_status dlmlock(struct dlm_ctxt *dlm,
+			int mode,
+			struct dlm_lockstatus *lksb,
+			int flags,
+			const char *name,
+			int namelen,
+			dlm_astlockfunc_t *ast,
+			void *data,
+			dlm_bastlockfunc_t *bast);
+
+enum dlm_status dlmunlock(struct dlm_ctxt *dlm,
+			  struct dlm_lockstatus *lksb,
+			  int flags,
+			  dlm_astunlockfunc_t *unlockast,
+			  void *data);
+
+struct dlm_protocol_version {
+	u8 pv_major;
+	u8 pv_minor;
+};
+struct dlm_ctxt * dlm_register_domain(const char *domain, u32 key,
+				      struct dlm_protocol_version *fs_proto);
+
+void dlm_unregister_domain(struct dlm_ctxt *dlm);
+
+void dlm_print_one_lock(struct dlm_lock *lockid);
+
+typedef void (dlm_eviction_func)(int, void *);
+struct dlm_eviction_cb {
+	struct list_head        ec_item;
+	dlm_eviction_func       *ec_func;
+	void                    *ec_data;
+};
+void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
+			   dlm_eviction_func *f,
+			   void *data);
+void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
+			      struct dlm_eviction_cb *cb);
+void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb);
+
+#endif /* DLMAPI_H */
diff --git a/fs/ocfs2/dlm/dlmast.c b/fs/ocfs2/dlm/dlmast.c
new file mode 100644
index 000000000..fd6bbbbd7
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmast.c
@@ -0,0 +1,504 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmast.c
+ *
+ * AST and BAST functionality for local and remote nodes
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/sysctl.h>
+#include <linux/random.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/spinlock.h>
+
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+
+#define MLOG_MASK_PREFIX ML_DLM
+#include "cluster/masklog.h"
+
+static void dlm_update_lvb(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+			   struct dlm_lock *lock);
+static int dlm_should_cancel_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
+
+/* Should be called as an ast gets queued to see if the new
+ * lock level will obsolete a pending bast.
+ * For example, if dlm_thread queued a bast for an EX lock that
+ * was blocking another EX, but before sending the bast the
+ * lock owner downconverted to NL, the bast is now obsolete.
+ * Only the ast should be sent.
+ * This is needed because the lock and convert paths can queue
+ * asts out-of-band (not waiting for dlm_thread) in order to
+ * allow for LKM_NOQUEUE to get immediate responses. */
+static int dlm_should_cancel_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock)
+{
+	assert_spin_locked(&dlm->ast_lock);
+	assert_spin_locked(&lock->spinlock);
+
+	if (lock->ml.highest_blocked == LKM_IVMODE)
+		return 0;
+	BUG_ON(lock->ml.highest_blocked == LKM_NLMODE);
+
+	if (lock->bast_pending &&
+	    list_empty(&lock->bast_list))
+		/* old bast already sent, ok */
+		return 0;
+
+	if (lock->ml.type == LKM_EXMODE)
+		/* EX blocks anything left, any bast still valid */
+		return 0;
+	else if (lock->ml.type == LKM_NLMODE)
+		/* NL blocks nothing, no reason to send any bast, cancel it */
+		return 1;
+	else if (lock->ml.highest_blocked != LKM_EXMODE)
+		/* PR only blocks EX */
+		return 1;
+
+	return 0;
+}
+
+void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock)
+{
+	struct dlm_lock_resource *res;
+
+	BUG_ON(!dlm);
+	BUG_ON(!lock);
+
+	res = lock->lockres;
+
+	assert_spin_locked(&dlm->ast_lock);
+
+	if (!list_empty(&lock->ast_list)) {
+		mlog(ML_ERROR, "%s: res %.*s, lock %u:%llu, "
+		     "AST list not empty, pending %d, newlevel %d\n",
+		     dlm->name, res->lockname.len, res->lockname.name,
+		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+		     lock->ast_pending, lock->ml.type);
+		BUG();
+	}
+	if (lock->ast_pending)
+		mlog(0, "%s: res %.*s, lock %u:%llu, AST getting flushed\n",
+		     dlm->name, res->lockname.len, res->lockname.name,
+		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)));
+
+	/* putting lock on list, add a ref */
+	dlm_lock_get(lock);
+	spin_lock(&lock->spinlock);
+
+	/* check to see if this ast obsoletes the bast */
+	if (dlm_should_cancel_bast(dlm, lock)) {
+		mlog(0, "%s: res %.*s, lock %u:%llu, Cancelling BAST\n",
+		     dlm->name, res->lockname.len, res->lockname.name,
+		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)));
+		lock->bast_pending = 0;
+		list_del_init(&lock->bast_list);
+		lock->ml.highest_blocked = LKM_IVMODE;
+		/* removing lock from list, remove a ref.  guaranteed
+		 * this won't be the last ref because of the get above,
+		 * so res->spinlock will not be taken here */
+		dlm_lock_put(lock);
+		/* free up the reserved bast that we are cancelling.
+		 * guaranteed that this will not be the last reserved
+		 * ast because *both* an ast and a bast were reserved
+		 * to get to this point.  the res->spinlock will not be
+		 * taken here */
+		dlm_lockres_release_ast(dlm, res);
+	}
+	list_add_tail(&lock->ast_list, &dlm->pending_asts);
+	lock->ast_pending = 1;
+	spin_unlock(&lock->spinlock);
+}
+
+void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock)
+{
+	BUG_ON(!dlm);
+	BUG_ON(!lock);
+
+	spin_lock(&dlm->ast_lock);
+	__dlm_queue_ast(dlm, lock);
+	spin_unlock(&dlm->ast_lock);
+}
+
+
+void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock)
+{
+	struct dlm_lock_resource *res;
+
+	BUG_ON(!dlm);
+	BUG_ON(!lock);
+
+	assert_spin_locked(&dlm->ast_lock);
+
+	res = lock->lockres;
+
+	BUG_ON(!list_empty(&lock->bast_list));
+	if (lock->bast_pending)
+		mlog(0, "%s: res %.*s, lock %u:%llu, BAST getting flushed\n",
+		     dlm->name, res->lockname.len, res->lockname.name,
+		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)));
+
+	/* putting lock on list, add a ref */
+	dlm_lock_get(lock);
+	spin_lock(&lock->spinlock);
+	list_add_tail(&lock->bast_list, &dlm->pending_basts);
+	lock->bast_pending = 1;
+	spin_unlock(&lock->spinlock);
+}
+
+void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock)
+{
+	BUG_ON(!dlm);
+	BUG_ON(!lock);
+
+	spin_lock(&dlm->ast_lock);
+	__dlm_queue_bast(dlm, lock);
+	spin_unlock(&dlm->ast_lock);
+}
+
+static void dlm_update_lvb(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+			   struct dlm_lock *lock)
+{
+	struct dlm_lockstatus *lksb = lock->lksb;
+	BUG_ON(!lksb);
+
+	/* only updates if this node masters the lockres */
+	spin_lock(&res->spinlock);
+	if (res->owner == dlm->node_num) {
+		/* check the lksb flags for the direction */
+		if (lksb->flags & DLM_LKSB_GET_LVB) {
+			mlog(0, "getting lvb from lockres for %s node\n",
+				  lock->ml.node == dlm->node_num ? "master" :
+				  "remote");
+			memcpy(lksb->lvb, res->lvb, DLM_LVB_LEN);
+		}
+		/* Do nothing for lvb put requests - they should be done in
+ 		 * place when the lock is downconverted - otherwise we risk
+ 		 * racing gets and puts which could result in old lvb data
+ 		 * being propagated. We leave the put flag set and clear it
+ 		 * here. In the future we might want to clear it at the time
+ 		 * the put is actually done.
+		 */
+	}
+	spin_unlock(&res->spinlock);
+
+	/* reset any lvb flags on the lksb */
+	lksb->flags &= ~(DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB);
+}
+
+void dlm_do_local_ast(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+		      struct dlm_lock *lock)
+{
+	dlm_astlockfunc_t *fn;
+	struct dlm_lockstatus *lksb;
+
+	mlog(0, "%s: res %.*s, lock %u:%llu, Local AST\n", dlm->name,
+	     res->lockname.len, res->lockname.name,
+	     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+	     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)));
+
+	lksb = lock->lksb;
+	fn = lock->ast;
+	BUG_ON(lock->ml.node != dlm->node_num);
+
+	dlm_update_lvb(dlm, res, lock);
+	(*fn)(lock->astdata);
+}
+
+
+int dlm_do_remote_ast(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+		      struct dlm_lock *lock)
+{
+	int ret;
+	struct dlm_lockstatus *lksb;
+	int lksbflags;
+
+	mlog(0, "%s: res %.*s, lock %u:%llu, Remote AST\n", dlm->name,
+	     res->lockname.len, res->lockname.name,
+	     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+	     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)));
+
+	lksb = lock->lksb;
+	BUG_ON(lock->ml.node == dlm->node_num);
+
+	lksbflags = lksb->flags;
+	dlm_update_lvb(dlm, res, lock);
+
+	/* lock request came from another node
+	 * go do the ast over there */
+	ret = dlm_send_proxy_ast(dlm, res, lock, lksbflags);
+	return ret;
+}
+
+void dlm_do_local_bast(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+		       struct dlm_lock *lock, int blocked_type)
+{
+	dlm_bastlockfunc_t *fn = lock->bast;
+
+	BUG_ON(lock->ml.node != dlm->node_num);
+
+	mlog(0, "%s: res %.*s, lock %u:%llu, Local BAST, blocked %d\n",
+	     dlm->name, res->lockname.len, res->lockname.name,
+	     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+	     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+	     blocked_type);
+
+	(*fn)(lock->astdata, blocked_type);
+}
+
+
+
+int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data,
+			  void **ret_data)
+{
+	int ret;
+	unsigned int locklen;
+	struct dlm_ctxt *dlm = data;
+	struct dlm_lock_resource *res = NULL;
+	struct dlm_lock *lock = NULL;
+	struct dlm_proxy_ast *past = (struct dlm_proxy_ast *) msg->buf;
+	char *name;
+	struct list_head *head = NULL;
+	__be64 cookie;
+	u32 flags;
+	u8 node;
+
+	if (!dlm_grab(dlm)) {
+		dlm_error(DLM_REJECTED);
+		return DLM_REJECTED;
+	}
+
+	mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
+			"Domain %s not fully joined!\n", dlm->name);
+
+	name = past->name;
+	locklen = past->namelen;
+	cookie = past->cookie;
+	flags = be32_to_cpu(past->flags);
+	node = past->node_idx;
+
+	if (locklen > DLM_LOCKID_NAME_MAX) {
+		ret = DLM_IVBUFLEN;
+		mlog(ML_ERROR, "Invalid name length (%d) in proxy ast "
+		     "handler!\n", locklen);
+		goto leave;
+	}
+
+	if ((flags & (LKM_PUT_LVB|LKM_GET_LVB)) ==
+	     (LKM_PUT_LVB|LKM_GET_LVB)) {
+		mlog(ML_ERROR, "Both PUT and GET lvb specified, (0x%x)\n",
+		     flags);
+		ret = DLM_BADARGS;
+		goto leave;
+	}
+
+	mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" :
+		  (flags & LKM_GET_LVB ? "get lvb" : "none"));
+
+	mlog(0, "type=%d, blocked_type=%d\n", past->type, past->blocked_type);
+
+	if (past->type != DLM_AST &&
+	    past->type != DLM_BAST) {
+		mlog(ML_ERROR, "Unknown ast type! %d, cookie=%u:%llu"
+		     "name=%.*s, node=%u\n", past->type,
+		     dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
+		     locklen, name, node);
+		ret = DLM_IVLOCKID;
+		goto leave;
+	}
+
+	res = dlm_lookup_lockres(dlm, name, locklen);
+	if (!res) {
+		mlog(0, "Got %sast for unknown lockres! cookie=%u:%llu, "
+		     "name=%.*s, node=%u\n", (past->type == DLM_AST ? "" : "b"),
+		     dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
+		     locklen, name, node);
+		ret = DLM_IVLOCKID;
+		goto leave;
+	}
+
+	/* cannot get a proxy ast message if this node owns it */
+	BUG_ON(res->owner == dlm->node_num);
+
+	mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len,
+	     res->lockname.name);
+
+	spin_lock(&res->spinlock);
+	if (res->state & DLM_LOCK_RES_RECOVERING) {
+		mlog(0, "Responding with DLM_RECOVERING!\n");
+		ret = DLM_RECOVERING;
+		goto unlock_out;
+	}
+	if (res->state & DLM_LOCK_RES_MIGRATING) {
+		mlog(0, "Responding with DLM_MIGRATING!\n");
+		ret = DLM_MIGRATING;
+		goto unlock_out;
+	}
+	/* try convert queue for both ast/bast */
+	head = &res->converting;
+	lock = NULL;
+	list_for_each_entry(lock, head, list) {
+		if (lock->ml.cookie == cookie)
+			goto do_ast;
+	}
+
+	/* if not on convert, try blocked for ast, granted for bast */
+	if (past->type == DLM_AST)
+		head = &res->blocked;
+	else
+		head = &res->granted;
+
+	list_for_each_entry(lock, head, list) {
+		/* if lock is found but unlock is pending ignore the bast */
+		if (lock->ml.cookie == cookie) {
+			if (lock->unlock_pending)
+				break;
+			goto do_ast;
+		}
+	}
+
+	mlog(0, "Got %sast for unknown lock! cookie=%u:%llu, name=%.*s, "
+	     "node=%u\n", past->type == DLM_AST ? "" : "b",
+	     dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
+	     dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
+	     locklen, name, node);
+
+	ret = DLM_NORMAL;
+unlock_out:
+	spin_unlock(&res->spinlock);
+	goto leave;
+
+do_ast:
+	ret = DLM_NORMAL;
+	if (past->type == DLM_AST) {
+		/* do not alter lock refcount.  switching lists. */
+		list_move_tail(&lock->list, &res->granted);
+		mlog(0, "%s: res %.*s, lock %u:%llu, Granted type %d => %d\n",
+		     dlm->name, res->lockname.len, res->lockname.name,
+		     dlm_get_lock_cookie_node(be64_to_cpu(cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(cookie)),
+		     lock->ml.type, lock->ml.convert_type);
+
+		if (lock->ml.convert_type != LKM_IVMODE) {
+			lock->ml.type = lock->ml.convert_type;
+			lock->ml.convert_type = LKM_IVMODE;
+		} else {
+			// should already be there....
+		}
+
+		lock->lksb->status = DLM_NORMAL;
+
+		/* if we requested the lvb, fetch it into our lksb now */
+		if (flags & LKM_GET_LVB) {
+			BUG_ON(!(lock->lksb->flags & DLM_LKSB_GET_LVB));
+			memcpy(lock->lksb->lvb, past->lvb, DLM_LVB_LEN);
+		}
+	}
+	spin_unlock(&res->spinlock);
+
+	if (past->type == DLM_AST)
+		dlm_do_local_ast(dlm, res, lock);
+	else
+		dlm_do_local_bast(dlm, res, lock, past->blocked_type);
+
+leave:
+	if (res)
+		dlm_lockres_put(res);
+
+	dlm_put(dlm);
+	return ret;
+}
+
+
+
+int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+			   struct dlm_lock *lock, int msg_type,
+			   int blocked_type, int flags)
+{
+	int ret = 0;
+	struct dlm_proxy_ast past;
+	struct kvec vec[2];
+	size_t veclen = 1;
+	int status;
+
+	mlog(0, "%s: res %.*s, to %u, type %d, blocked_type %d\n", dlm->name,
+	     res->lockname.len, res->lockname.name, lock->ml.node, msg_type,
+	     blocked_type);
+
+	memset(&past, 0, sizeof(struct dlm_proxy_ast));
+	past.node_idx = dlm->node_num;
+	past.type = msg_type;
+	past.blocked_type = blocked_type;
+	past.namelen = res->lockname.len;
+	memcpy(past.name, res->lockname.name, past.namelen);
+	past.cookie = lock->ml.cookie;
+
+	vec[0].iov_len = sizeof(struct dlm_proxy_ast);
+	vec[0].iov_base = &past;
+	if (flags & DLM_LKSB_GET_LVB) {
+		be32_add_cpu(&past.flags, LKM_GET_LVB);
+		vec[1].iov_len = DLM_LVB_LEN;
+		vec[1].iov_base = lock->lksb->lvb;
+		veclen++;
+	}
+
+	ret = o2net_send_message_vec(DLM_PROXY_AST_MSG, dlm->key, vec, veclen,
+				     lock->ml.node, &status);
+	if (ret < 0)
+		mlog(ML_ERROR, "%s: res %.*s, error %d send AST to node %u\n",
+		     dlm->name, res->lockname.len, res->lockname.name, ret,
+		     lock->ml.node);
+	else {
+		if (status == DLM_RECOVERING) {
+			mlog(ML_ERROR, "sent AST to node %u, it thinks this "
+			     "node is dead!\n", lock->ml.node);
+			BUG();
+		} else if (status == DLM_MIGRATING) {
+			mlog(ML_ERROR, "sent AST to node %u, it returned "
+			     "DLM_MIGRATING!\n", lock->ml.node);
+			BUG();
+		} else if (status != DLM_NORMAL && status != DLM_IVLOCKID) {
+			mlog(ML_ERROR, "AST to node %u returned %d!\n",
+			     lock->ml.node, status);
+			/* ignore it */
+		}
+		ret = 0;
+	}
+	return ret;
+}
diff --git a/fs/ocfs2/dlm/dlmcommon.h b/fs/ocfs2/dlm/dlmcommon.h
new file mode 100644
index 000000000..fae17c640
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmcommon.h
@@ -0,0 +1,1150 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmcommon.h
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef DLMCOMMON_H
+#define DLMCOMMON_H
+
+#include <linux/kref.h>
+
+#define DLM_HB_NODE_DOWN_PRI     (0xf000000)
+#define DLM_HB_NODE_UP_PRI       (0x8000000)
+
+#define DLM_LOCKID_NAME_MAX    32
+
+#define DLM_DOMAIN_NAME_MAX_LEN    255
+#define DLM_LOCK_RES_OWNER_UNKNOWN     O2NM_MAX_NODES
+#define DLM_THREAD_SHUFFLE_INTERVAL    5     // flush everything every 5 passes
+#define DLM_THREAD_MS                  200   // flush at least every 200 ms
+
+#define DLM_HASH_SIZE_DEFAULT	(1 << 17)
+#if DLM_HASH_SIZE_DEFAULT < PAGE_SIZE
+# define DLM_HASH_PAGES		1
+#else
+# define DLM_HASH_PAGES		(DLM_HASH_SIZE_DEFAULT / PAGE_SIZE)
+#endif
+#define DLM_BUCKETS_PER_PAGE	(PAGE_SIZE / sizeof(struct hlist_head))
+#define DLM_HASH_BUCKETS	(DLM_HASH_PAGES * DLM_BUCKETS_PER_PAGE)
+
+/* Intended to make it easier for us to switch out hash functions */
+#define dlm_lockid_hash(_n, _l) full_name_hash(_n, _l)
+
+enum dlm_mle_type {
+	DLM_MLE_BLOCK = 0,
+	DLM_MLE_MASTER = 1,
+	DLM_MLE_MIGRATION = 2,
+	DLM_MLE_NUM_TYPES = 3,
+};
+
+struct dlm_master_list_entry {
+	struct hlist_node master_hash_node;
+	struct list_head hb_events;
+	struct dlm_ctxt *dlm;
+	spinlock_t spinlock;
+	wait_queue_head_t wq;
+	atomic_t woken;
+	struct kref mle_refs;
+	int inuse;
+	unsigned long maybe_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long vote_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long response_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	u8 master;
+	u8 new_master;
+	enum dlm_mle_type type;
+	struct o2hb_callback_func mle_hb_up;
+	struct o2hb_callback_func mle_hb_down;
+	struct dlm_lock_resource *mleres;
+	unsigned char mname[DLM_LOCKID_NAME_MAX];
+	unsigned int mnamelen;
+	unsigned int mnamehash;
+};
+
+enum dlm_ast_type {
+	DLM_AST = 0,
+	DLM_BAST = 1,
+	DLM_ASTUNLOCK = 2,
+};
+
+
+#define LKM_VALID_FLAGS (LKM_VALBLK | LKM_CONVERT | LKM_UNLOCK | \
+			 LKM_CANCEL | LKM_INVVALBLK | LKM_FORCE | \
+			 LKM_RECOVERY | LKM_LOCAL | LKM_NOQUEUE)
+
+#define DLM_RECOVERY_LOCK_NAME       "$RECOVERY"
+#define DLM_RECOVERY_LOCK_NAME_LEN   9
+
+static inline int dlm_is_recovery_lock(const char *lock_name, int name_len)
+{
+	if (name_len == DLM_RECOVERY_LOCK_NAME_LEN &&
+	    memcmp(lock_name, DLM_RECOVERY_LOCK_NAME, name_len)==0)
+		return 1;
+	return 0;
+}
+
+#define DLM_RECO_STATE_ACTIVE    0x0001
+#define DLM_RECO_STATE_FINALIZE  0x0002
+
+struct dlm_recovery_ctxt
+{
+	struct list_head resources;
+	struct list_head node_data;
+	u8  new_master;
+	u8  dead_node;
+	u16 state;
+	unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	wait_queue_head_t event;
+};
+
+enum dlm_ctxt_state {
+	DLM_CTXT_NEW = 0,
+	DLM_CTXT_JOINED = 1,
+	DLM_CTXT_IN_SHUTDOWN = 2,
+	DLM_CTXT_LEAVING = 3,
+};
+
+struct dlm_ctxt
+{
+	struct list_head list;
+	struct hlist_head **lockres_hash;
+	struct list_head dirty_list;
+	struct list_head purge_list;
+	struct list_head pending_asts;
+	struct list_head pending_basts;
+	struct list_head tracking_list;
+	unsigned int purge_count;
+	spinlock_t spinlock;
+	spinlock_t ast_lock;
+	spinlock_t track_lock;
+	char *name;
+	u8 node_num;
+	u32 key;
+	u8  joining_node;
+	wait_queue_head_t dlm_join_events;
+	unsigned long live_nodes_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long exit_domain_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long recovery_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	struct dlm_recovery_ctxt reco;
+	spinlock_t master_lock;
+	struct hlist_head **master_hash;
+	struct list_head mle_hb_events;
+
+	/* these give a really vague idea of the system load */
+	atomic_t mle_tot_count[DLM_MLE_NUM_TYPES];
+	atomic_t mle_cur_count[DLM_MLE_NUM_TYPES];
+	atomic_t res_tot_count;
+	atomic_t res_cur_count;
+
+	struct dlm_debug_ctxt *dlm_debug_ctxt;
+	struct dentry *dlm_debugfs_subroot;
+
+	/* NOTE: Next three are protected by dlm_domain_lock */
+	struct kref dlm_refs;
+	enum dlm_ctxt_state dlm_state;
+	unsigned int num_joins;
+
+	struct o2hb_callback_func dlm_hb_up;
+	struct o2hb_callback_func dlm_hb_down;
+	struct task_struct *dlm_thread_task;
+	struct task_struct *dlm_reco_thread_task;
+	struct workqueue_struct *dlm_worker;
+	wait_queue_head_t dlm_thread_wq;
+	wait_queue_head_t dlm_reco_thread_wq;
+	wait_queue_head_t ast_wq;
+	wait_queue_head_t migration_wq;
+
+	struct work_struct dispatched_work;
+	struct list_head work_list;
+	spinlock_t work_lock;
+	struct list_head dlm_domain_handlers;
+	struct list_head	dlm_eviction_callbacks;
+
+	/* The filesystem specifies this at domain registration.  We
+	 * cache it here to know what to tell other nodes. */
+	struct dlm_protocol_version fs_locking_proto;
+	/* This is the inter-dlm communication version */
+	struct dlm_protocol_version dlm_locking_proto;
+};
+
+static inline struct hlist_head *dlm_lockres_hash(struct dlm_ctxt *dlm, unsigned i)
+{
+	return dlm->lockres_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] + (i % DLM_BUCKETS_PER_PAGE);
+}
+
+static inline struct hlist_head *dlm_master_hash(struct dlm_ctxt *dlm,
+						 unsigned i)
+{
+	return dlm->master_hash[(i / DLM_BUCKETS_PER_PAGE) % DLM_HASH_PAGES] +
+			(i % DLM_BUCKETS_PER_PAGE);
+}
+
+/* these keventd work queue items are for less-frequently
+ * called functions that cannot be directly called from the
+ * net message handlers for some reason, usually because
+ * they need to send net messages of their own. */
+void dlm_dispatch_work(struct work_struct *work);
+
+struct dlm_lock_resource;
+struct dlm_work_item;
+
+typedef void (dlm_workfunc_t)(struct dlm_work_item *, void *);
+
+struct dlm_request_all_locks_priv
+{
+	u8 reco_master;
+	u8 dead_node;
+};
+
+struct dlm_mig_lockres_priv
+{
+	struct dlm_lock_resource *lockres;
+	u8 real_master;
+	u8 extra_ref;
+};
+
+struct dlm_assert_master_priv
+{
+	struct dlm_lock_resource *lockres;
+	u8 request_from;
+	u32 flags;
+	unsigned ignore_higher:1;
+};
+
+struct dlm_deref_lockres_priv
+{
+	struct dlm_lock_resource *deref_res;
+	u8 deref_node;
+};
+
+struct dlm_work_item
+{
+	struct list_head list;
+	dlm_workfunc_t *func;
+	struct dlm_ctxt *dlm;
+	void *data;
+	union {
+		struct dlm_request_all_locks_priv ral;
+		struct dlm_mig_lockres_priv ml;
+		struct dlm_assert_master_priv am;
+		struct dlm_deref_lockres_priv dl;
+	} u;
+};
+
+static inline void dlm_init_work_item(struct dlm_ctxt *dlm,
+				      struct dlm_work_item *i,
+				      dlm_workfunc_t *f, void *data)
+{
+	memset(i, 0, sizeof(*i));
+	i->func = f;
+	INIT_LIST_HEAD(&i->list);
+	i->data = data;
+	i->dlm = dlm;  /* must have already done a dlm_grab on this! */
+}
+
+
+
+static inline void __dlm_set_joining_node(struct dlm_ctxt *dlm,
+					  u8 node)
+{
+	assert_spin_locked(&dlm->spinlock);
+
+	dlm->joining_node = node;
+	wake_up(&dlm->dlm_join_events);
+}
+
+#define DLM_LOCK_RES_UNINITED             0x00000001
+#define DLM_LOCK_RES_RECOVERING           0x00000002
+#define DLM_LOCK_RES_READY                0x00000004
+#define DLM_LOCK_RES_DIRTY                0x00000008
+#define DLM_LOCK_RES_IN_PROGRESS          0x00000010
+#define DLM_LOCK_RES_MIGRATING            0x00000020
+#define DLM_LOCK_RES_DROPPING_REF         0x00000040
+#define DLM_LOCK_RES_BLOCK_DIRTY          0x00001000
+#define DLM_LOCK_RES_SETREF_INPROG        0x00002000
+
+/* max milliseconds to wait to sync up a network failure with a node death */
+#define DLM_NODE_DEATH_WAIT_MAX (5 * 1000)
+
+#define DLM_PURGE_INTERVAL_MS   (8 * 1000)
+
+struct dlm_lock_resource
+{
+	/* WARNING: Please see the comment in dlm_init_lockres before
+	 * adding fields here. */
+	struct hlist_node hash_node;
+	struct qstr lockname;
+	struct kref      refs;
+
+	/*
+	 * Please keep granted, converting, and blocked in this order,
+	 * as some funcs want to iterate over all lists.
+	 *
+	 * All four lists are protected by the hash's reference.
+	 */
+	struct list_head granted;
+	struct list_head converting;
+	struct list_head blocked;
+	struct list_head purge;
+
+	/*
+	 * These two lists require you to hold an additional reference
+	 * while they are on the list.
+	 */
+	struct list_head dirty;
+	struct list_head recovering; // dlm_recovery_ctxt.resources list
+
+	/* Added during init and removed during release */
+	struct list_head tracking;	/* dlm->tracking_list */
+
+	/* unused lock resources have their last_used stamped and are
+	 * put on a list for the dlm thread to run. */
+	unsigned long    last_used;
+
+	struct dlm_ctxt *dlm;
+
+	unsigned migration_pending:1;
+	atomic_t asts_reserved;
+	spinlock_t spinlock;
+	wait_queue_head_t wq;
+	u8  owner;              //node which owns the lock resource, or unknown
+	u16 state;
+	char lvb[DLM_LVB_LEN];
+	unsigned int inflight_locks;
+	unsigned int inflight_assert_workers;
+	unsigned long refmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+};
+
+struct dlm_migratable_lock
+{
+	__be64 cookie;
+
+	/* these 3 are just padding for the in-memory structure, but
+	 * list and flags are actually used when sent over the wire */
+	__be16 pad1;
+	u8 list;  // 0=granted, 1=converting, 2=blocked
+	u8 flags;
+
+	s8 type;
+	s8 convert_type;
+	s8 highest_blocked;
+	u8 node;
+};  // 16 bytes
+
+struct dlm_lock
+{
+	struct dlm_migratable_lock ml;
+
+	struct list_head list;
+	struct list_head ast_list;
+	struct list_head bast_list;
+	struct dlm_lock_resource *lockres;
+	spinlock_t spinlock;
+	struct kref lock_refs;
+
+	// ast and bast must be callable while holding a spinlock!
+	dlm_astlockfunc_t *ast;
+	dlm_bastlockfunc_t *bast;
+	void *astdata;
+	struct dlm_lockstatus *lksb;
+	unsigned ast_pending:1,
+		 bast_pending:1,
+		 convert_pending:1,
+		 lock_pending:1,
+		 cancel_pending:1,
+		 unlock_pending:1,
+		 lksb_kernel_allocated:1;
+};
+
+
+#define DLM_LKSB_UNUSED1           0x01
+#define DLM_LKSB_PUT_LVB           0x02
+#define DLM_LKSB_GET_LVB           0x04
+#define DLM_LKSB_UNUSED2           0x08
+#define DLM_LKSB_UNUSED3           0x10
+#define DLM_LKSB_UNUSED4           0x20
+#define DLM_LKSB_UNUSED5           0x40
+#define DLM_LKSB_UNUSED6           0x80
+
+
+enum dlm_lockres_list {
+	DLM_GRANTED_LIST = 0,
+	DLM_CONVERTING_LIST = 1,
+	DLM_BLOCKED_LIST = 2,
+};
+
+static inline int dlm_lvb_is_empty(char *lvb)
+{
+	int i;
+	for (i=0; i<DLM_LVB_LEN; i++)
+		if (lvb[i])
+			return 0;
+	return 1;
+}
+
+static inline char *dlm_list_in_text(enum dlm_lockres_list idx)
+{
+	if (idx == DLM_GRANTED_LIST)
+		return "granted";
+	else if (idx == DLM_CONVERTING_LIST)
+		return "converting";
+	else if (idx == DLM_BLOCKED_LIST)
+		return "blocked";
+	else
+		return "unknown";
+}
+
+static inline struct list_head *
+dlm_list_idx_to_ptr(struct dlm_lock_resource *res, enum dlm_lockres_list idx)
+{
+	struct list_head *ret = NULL;
+	if (idx == DLM_GRANTED_LIST)
+		ret = &res->granted;
+	else if (idx == DLM_CONVERTING_LIST)
+		ret = &res->converting;
+	else if (idx == DLM_BLOCKED_LIST)
+		ret = &res->blocked;
+	else
+		BUG();
+	return ret;
+}
+
+
+
+
+struct dlm_node_iter
+{
+	unsigned long node_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	int curnode;
+};
+
+
+enum {
+	DLM_MASTER_REQUEST_MSG		= 500,
+	DLM_UNUSED_MSG1			= 501,
+	DLM_ASSERT_MASTER_MSG		= 502,
+	DLM_CREATE_LOCK_MSG		= 503,
+	DLM_CONVERT_LOCK_MSG		= 504,
+	DLM_PROXY_AST_MSG		= 505,
+	DLM_UNLOCK_LOCK_MSG		= 506,
+	DLM_DEREF_LOCKRES_MSG		= 507,
+	DLM_MIGRATE_REQUEST_MSG		= 508,
+	DLM_MIG_LOCKRES_MSG		= 509,
+	DLM_QUERY_JOIN_MSG		= 510,
+	DLM_ASSERT_JOINED_MSG		= 511,
+	DLM_CANCEL_JOIN_MSG		= 512,
+	DLM_EXIT_DOMAIN_MSG		= 513,
+	DLM_MASTER_REQUERY_MSG		= 514,
+	DLM_LOCK_REQUEST_MSG		= 515,
+	DLM_RECO_DATA_DONE_MSG		= 516,
+	DLM_BEGIN_RECO_MSG		= 517,
+	DLM_FINALIZE_RECO_MSG		= 518,
+	DLM_QUERY_REGION		= 519,
+	DLM_QUERY_NODEINFO		= 520,
+	DLM_BEGIN_EXIT_DOMAIN_MSG	= 521,
+};
+
+struct dlm_reco_node_data
+{
+	int state;
+	u8 node_num;
+	struct list_head list;
+};
+
+enum {
+	DLM_RECO_NODE_DATA_DEAD = -1,
+	DLM_RECO_NODE_DATA_INIT = 0,
+	DLM_RECO_NODE_DATA_REQUESTING = 1,
+	DLM_RECO_NODE_DATA_REQUESTED = 2,
+	DLM_RECO_NODE_DATA_RECEIVING = 3,
+	DLM_RECO_NODE_DATA_DONE = 4,
+	DLM_RECO_NODE_DATA_FINALIZE_SENT = 5,
+};
+
+
+enum {
+	DLM_MASTER_RESP_NO = 0,
+	DLM_MASTER_RESP_YES = 1,
+	DLM_MASTER_RESP_MAYBE = 2,
+	DLM_MASTER_RESP_ERROR = 3,
+};
+
+
+struct dlm_master_request
+{
+	u8 node_idx;
+	u8 namelen;
+	__be16 pad1;
+	__be32 flags;
+
+	u8 name[O2NM_MAX_NAME_LEN];
+};
+
+#define DLM_ASSERT_RESPONSE_REASSERT       0x00000001
+#define DLM_ASSERT_RESPONSE_MASTERY_REF    0x00000002
+
+#define DLM_ASSERT_MASTER_MLE_CLEANUP      0x00000001
+#define DLM_ASSERT_MASTER_REQUERY          0x00000002
+#define DLM_ASSERT_MASTER_FINISH_MIGRATION 0x00000004
+struct dlm_assert_master
+{
+	u8 node_idx;
+	u8 namelen;
+	__be16 pad1;
+	__be32 flags;
+
+	u8 name[O2NM_MAX_NAME_LEN];
+};
+
+#define DLM_MIGRATE_RESPONSE_MASTERY_REF   0x00000001
+
+struct dlm_migrate_request
+{
+	u8 master;
+	u8 new_master;
+	u8 namelen;
+	u8 pad1;
+	__be32 pad2;
+	u8 name[O2NM_MAX_NAME_LEN];
+};
+
+struct dlm_master_requery
+{
+	u8 pad1;
+	u8 pad2;
+	u8 node_idx;
+	u8 namelen;
+	__be32 pad3;
+	u8 name[O2NM_MAX_NAME_LEN];
+};
+
+#define DLM_MRES_RECOVERY   0x01
+#define DLM_MRES_MIGRATION  0x02
+#define DLM_MRES_ALL_DONE   0x04
+
+/*
+ * We would like to get one whole lockres into a single network
+ * message whenever possible.  Generally speaking, there will be
+ * at most one dlm_lock on a lockres for each node in the cluster,
+ * plus (infrequently) any additional locks coming in from userdlm.
+ *
+ * struct _dlm_lockres_page
+ * {
+ * 	dlm_migratable_lockres mres;
+ * 	dlm_migratable_lock ml[DLM_MAX_MIGRATABLE_LOCKS];
+ * 	u8 pad[DLM_MIG_LOCKRES_RESERVED];
+ * };
+ *
+ * from ../cluster/tcp.h
+ *    NET_MAX_PAYLOAD_BYTES  (4096 - sizeof(net_msg))
+ *    (roughly 4080 bytes)
+ * and sizeof(dlm_migratable_lockres) = 112 bytes
+ * and sizeof(dlm_migratable_lock) = 16 bytes
+ *
+ * Choosing DLM_MAX_MIGRATABLE_LOCKS=240 and
+ * DLM_MIG_LOCKRES_RESERVED=128 means we have this:
+ *
+ *  (DLM_MAX_MIGRATABLE_LOCKS * sizeof(dlm_migratable_lock)) +
+ *     sizeof(dlm_migratable_lockres) + DLM_MIG_LOCKRES_RESERVED =
+ *        NET_MAX_PAYLOAD_BYTES
+ *  (240 * 16) + 112 + 128 = 4080
+ *
+ * So a lockres would need more than 240 locks before it would
+ * use more than one network packet to recover.  Not too bad.
+ */
+#define DLM_MAX_MIGRATABLE_LOCKS   240
+
+struct dlm_migratable_lockres
+{
+	u8 master;
+	u8 lockname_len;
+	u8 num_locks;    // locks sent in this structure
+	u8 flags;
+	__be32 total_locks; // locks to be sent for this migration cookie
+	__be64 mig_cookie;  // cookie for this lockres migration
+			 // or zero if not needed
+	// 16 bytes
+	u8 lockname[DLM_LOCKID_NAME_MAX];
+	// 48 bytes
+	u8 lvb[DLM_LVB_LEN];
+	// 112 bytes
+	struct dlm_migratable_lock ml[0];  // 16 bytes each, begins at byte 112
+};
+#define DLM_MIG_LOCKRES_MAX_LEN  \
+	(sizeof(struct dlm_migratable_lockres) + \
+	 (sizeof(struct dlm_migratable_lock) * \
+	  DLM_MAX_MIGRATABLE_LOCKS) )
+
+/* from above, 128 bytes
+ * for some undetermined future use */
+#define DLM_MIG_LOCKRES_RESERVED   (NET_MAX_PAYLOAD_BYTES - \
+				    DLM_MIG_LOCKRES_MAX_LEN)
+
+struct dlm_create_lock
+{
+	__be64 cookie;
+
+	__be32 flags;
+	u8 pad1;
+	u8 node_idx;
+	s8 requested_type;
+	u8 namelen;
+
+	u8 name[O2NM_MAX_NAME_LEN];
+};
+
+struct dlm_convert_lock
+{
+	__be64 cookie;
+
+	__be32 flags;
+	u8 pad1;
+	u8 node_idx;
+	s8 requested_type;
+	u8 namelen;
+
+	u8 name[O2NM_MAX_NAME_LEN];
+
+	s8 lvb[0];
+};
+#define DLM_CONVERT_LOCK_MAX_LEN  (sizeof(struct dlm_convert_lock)+DLM_LVB_LEN)
+
+struct dlm_unlock_lock
+{
+	__be64 cookie;
+
+	__be32 flags;
+	__be16 pad1;
+	u8 node_idx;
+	u8 namelen;
+
+	u8 name[O2NM_MAX_NAME_LEN];
+
+	s8 lvb[0];
+};
+#define DLM_UNLOCK_LOCK_MAX_LEN  (sizeof(struct dlm_unlock_lock)+DLM_LVB_LEN)
+
+struct dlm_proxy_ast
+{
+	__be64 cookie;
+
+	__be32 flags;
+	u8 node_idx;
+	u8 type;
+	u8 blocked_type;
+	u8 namelen;
+
+	u8 name[O2NM_MAX_NAME_LEN];
+
+	s8 lvb[0];
+};
+#define DLM_PROXY_AST_MAX_LEN  (sizeof(struct dlm_proxy_ast)+DLM_LVB_LEN)
+
+#define DLM_MOD_KEY (0x666c6172)
+enum dlm_query_join_response_code {
+	JOIN_DISALLOW = 0,
+	JOIN_OK = 1,
+	JOIN_OK_NO_MAP = 2,
+	JOIN_PROTOCOL_MISMATCH = 3,
+};
+
+struct dlm_query_join_packet {
+	u8 code;	/* Response code.  dlm_minor and fs_minor
+			   are only valid if this is JOIN_OK */
+	u8 dlm_minor;	/* The minor version of the protocol the
+			   dlm is speaking. */
+	u8 fs_minor;	/* The minor version of the protocol the
+			   filesystem is speaking. */
+	u8 reserved;
+};
+
+union dlm_query_join_response {
+	__be32 intval;
+	struct dlm_query_join_packet packet;
+};
+
+struct dlm_lock_request
+{
+	u8 node_idx;
+	u8 dead_node;
+	__be16 pad1;
+	__be32 pad2;
+};
+
+struct dlm_reco_data_done
+{
+	u8 node_idx;
+	u8 dead_node;
+	__be16 pad1;
+	__be32 pad2;
+
+	/* unused for now */
+	/* eventually we can use this to attempt
+	 * lvb recovery based on each node's info */
+	u8 reco_lvb[DLM_LVB_LEN];
+};
+
+struct dlm_begin_reco
+{
+	u8 node_idx;
+	u8 dead_node;
+	__be16 pad1;
+	__be32 pad2;
+};
+
+
+#define BITS_PER_BYTE 8
+#define BITS_TO_BYTES(bits) (((bits)+BITS_PER_BYTE-1)/BITS_PER_BYTE)
+
+struct dlm_query_join_request
+{
+	u8 node_idx;
+	u8 pad1[2];
+	u8 name_len;
+	struct dlm_protocol_version dlm_proto;
+	struct dlm_protocol_version fs_proto;
+	u8 domain[O2NM_MAX_NAME_LEN];
+	u8 node_map[BITS_TO_BYTES(O2NM_MAX_NODES)];
+};
+
+struct dlm_assert_joined
+{
+	u8 node_idx;
+	u8 pad1[2];
+	u8 name_len;
+	u8 domain[O2NM_MAX_NAME_LEN];
+};
+
+struct dlm_cancel_join
+{
+	u8 node_idx;
+	u8 pad1[2];
+	u8 name_len;
+	u8 domain[O2NM_MAX_NAME_LEN];
+};
+
+struct dlm_query_region {
+	u8 qr_node;
+	u8 qr_numregions;
+	u8 qr_namelen;
+	u8 pad1;
+	u8 qr_domain[O2NM_MAX_NAME_LEN];
+	u8 qr_regions[O2HB_MAX_REGION_NAME_LEN * O2NM_MAX_REGIONS];
+};
+
+struct dlm_node_info {
+	u8 ni_nodenum;
+	u8 pad1;
+	__be16 ni_ipv4_port;
+	__be32 ni_ipv4_address;
+};
+
+struct dlm_query_nodeinfo {
+	u8 qn_nodenum;
+	u8 qn_numnodes;
+	u8 qn_namelen;
+	u8 pad1;
+	u8 qn_domain[O2NM_MAX_NAME_LEN];
+	struct dlm_node_info qn_nodes[O2NM_MAX_NODES];
+};
+
+struct dlm_exit_domain
+{
+	u8 node_idx;
+	u8 pad1[3];
+};
+
+struct dlm_finalize_reco
+{
+	u8 node_idx;
+	u8 dead_node;
+	u8 flags;
+	u8 pad1;
+	__be32 pad2;
+};
+
+struct dlm_deref_lockres
+{
+	u32 pad1;
+	u16 pad2;
+	u8 node_idx;
+	u8 namelen;
+
+	u8 name[O2NM_MAX_NAME_LEN];
+};
+
+static inline enum dlm_status
+__dlm_lockres_state_to_status(struct dlm_lock_resource *res)
+{
+	enum dlm_status status = DLM_NORMAL;
+
+	assert_spin_locked(&res->spinlock);
+
+	if (res->state & DLM_LOCK_RES_RECOVERING)
+		status = DLM_RECOVERING;
+	else if (res->state & DLM_LOCK_RES_MIGRATING)
+		status = DLM_MIGRATING;
+	else if (res->state & DLM_LOCK_RES_IN_PROGRESS)
+		status = DLM_FORWARD;
+
+	return status;
+}
+
+static inline u8 dlm_get_lock_cookie_node(u64 cookie)
+{
+	u8 ret;
+	cookie >>= 56;
+	ret = (u8)(cookie & 0xffULL);
+	return ret;
+}
+
+static inline unsigned long long dlm_get_lock_cookie_seq(u64 cookie)
+{
+	unsigned long long ret;
+	ret = ((unsigned long long)cookie) & 0x00ffffffffffffffULL;
+	return ret;
+}
+
+struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
+			       struct dlm_lockstatus *lksb);
+void dlm_lock_get(struct dlm_lock *lock);
+void dlm_lock_put(struct dlm_lock *lock);
+
+void dlm_lock_attach_lockres(struct dlm_lock *lock,
+			     struct dlm_lock_resource *res);
+
+int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
+			    void **ret_data);
+int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
+			     void **ret_data);
+int dlm_proxy_ast_handler(struct o2net_msg *msg, u32 len, void *data,
+			  void **ret_data);
+
+void dlm_revert_pending_convert(struct dlm_lock_resource *res,
+				struct dlm_lock *lock);
+void dlm_revert_pending_lock(struct dlm_lock_resource *res,
+			     struct dlm_lock *lock);
+
+int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data,
+			    void **ret_data);
+void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
+			       struct dlm_lock *lock);
+void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
+			       struct dlm_lock *lock);
+
+int dlm_launch_thread(struct dlm_ctxt *dlm);
+void dlm_complete_thread(struct dlm_ctxt *dlm);
+int dlm_launch_recovery_thread(struct dlm_ctxt *dlm);
+void dlm_complete_recovery_thread(struct dlm_ctxt *dlm);
+void dlm_wait_for_recovery(struct dlm_ctxt *dlm);
+void dlm_kick_recovery_thread(struct dlm_ctxt *dlm);
+int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node);
+void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout);
+void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout);
+
+void dlm_put(struct dlm_ctxt *dlm);
+struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm);
+int dlm_domain_fully_joined(struct dlm_ctxt *dlm);
+
+void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
+			      struct dlm_lock_resource *res);
+void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
+			    struct dlm_lock_resource *res);
+static inline void dlm_lockres_get(struct dlm_lock_resource *res)
+{
+	/* This is called on every lookup, so it might be worth
+	 * inlining. */
+	kref_get(&res->refs);
+}
+void dlm_lockres_put(struct dlm_lock_resource *res);
+void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
+void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
+struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
+						     const char *name,
+						     unsigned int len,
+						     unsigned int hash);
+struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
+						const char *name,
+						unsigned int len,
+						unsigned int hash);
+struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
+					      const char *name,
+					      unsigned int len);
+
+int dlm_is_host_down(int errno);
+
+struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
+						 const char *lockid,
+						 int namelen,
+						 int flags);
+struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
+					  const char *name,
+					  unsigned int namelen);
+
+void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm,
+				struct dlm_lock_resource *res, int bit);
+void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *res, int bit);
+
+void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm,
+				   struct dlm_lock_resource *res);
+void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
+				   struct dlm_lock_resource *res);
+
+void __dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm,
+		struct dlm_lock_resource *res);
+
+void dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
+void dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
+void __dlm_queue_ast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
+void __dlm_queue_bast(struct dlm_ctxt *dlm, struct dlm_lock *lock);
+void dlm_do_local_ast(struct dlm_ctxt *dlm,
+		      struct dlm_lock_resource *res,
+		      struct dlm_lock *lock);
+int dlm_do_remote_ast(struct dlm_ctxt *dlm,
+		      struct dlm_lock_resource *res,
+		      struct dlm_lock *lock);
+void dlm_do_local_bast(struct dlm_ctxt *dlm,
+		       struct dlm_lock_resource *res,
+		       struct dlm_lock *lock,
+		       int blocked_type);
+int dlm_send_proxy_ast_msg(struct dlm_ctxt *dlm,
+			   struct dlm_lock_resource *res,
+			   struct dlm_lock *lock,
+			   int msg_type,
+			   int blocked_type, int flags);
+static inline int dlm_send_proxy_bast(struct dlm_ctxt *dlm,
+				      struct dlm_lock_resource *res,
+				      struct dlm_lock *lock,
+				      int blocked_type)
+{
+	return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_BAST,
+				      blocked_type, 0);
+}
+
+static inline int dlm_send_proxy_ast(struct dlm_ctxt *dlm,
+				     struct dlm_lock_resource *res,
+				     struct dlm_lock *lock,
+				     int flags)
+{
+	return dlm_send_proxy_ast_msg(dlm, res, lock, DLM_AST,
+				      0, flags);
+}
+
+void dlm_print_one_lock_resource(struct dlm_lock_resource *res);
+void __dlm_print_one_lock_resource(struct dlm_lock_resource *res);
+
+u8 dlm_nm_this_node(struct dlm_ctxt *dlm);
+void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
+void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
+
+
+int dlm_nm_init(struct dlm_ctxt *dlm);
+int dlm_heartbeat_init(struct dlm_ctxt *dlm);
+void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data);
+void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data);
+
+int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res);
+int dlm_finish_migration(struct dlm_ctxt *dlm,
+			 struct dlm_lock_resource *res,
+			 u8 old_master);
+void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
+			     struct dlm_lock_resource *res);
+void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res);
+
+int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data,
+			       void **ret_data);
+int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data,
+			      void **ret_data);
+void dlm_assert_master_post_handler(int status, void *data, void *ret_data);
+int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
+			      void **ret_data);
+int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data,
+				void **ret_data);
+int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
+			    void **ret_data);
+int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
+			       void **ret_data);
+int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data,
+				  void **ret_data);
+int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
+			       void **ret_data);
+int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data,
+			   void **ret_data);
+int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data,
+			      void **ret_data);
+int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+			  u8 nodenum, u8 *real_master);
+
+
+int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
+			       struct dlm_lock_resource *res,
+			       int ignore_higher,
+			       u8 request_from,
+			       u32 flags);
+
+
+int dlm_send_one_lockres(struct dlm_ctxt *dlm,
+			 struct dlm_lock_resource *res,
+			 struct dlm_migratable_lockres *mres,
+			 u8 send_to,
+			 u8 flags);
+void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
+				       struct dlm_lock_resource *res);
+
+/* will exit holding res->spinlock, but may drop in function */
+void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags);
+void __dlm_wait_on_lockres_flags_set(struct dlm_lock_resource *res, int flags);
+
+/* will exit holding res->spinlock, but may drop in function */
+static inline void __dlm_wait_on_lockres(struct dlm_lock_resource *res)
+{
+	__dlm_wait_on_lockres_flags(res, (DLM_LOCK_RES_IN_PROGRESS|
+				    	  DLM_LOCK_RES_RECOVERING|
+					  DLM_LOCK_RES_MIGRATING));
+}
+
+void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
+void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle);
+
+/* create/destroy slab caches */
+int dlm_init_master_caches(void);
+void dlm_destroy_master_caches(void);
+
+int dlm_init_lock_cache(void);
+void dlm_destroy_lock_cache(void);
+
+int dlm_init_mle_cache(void);
+void dlm_destroy_mle_cache(void);
+
+void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up);
+int dlm_drop_lockres_ref(struct dlm_ctxt *dlm,
+			 struct dlm_lock_resource *res);
+void dlm_clean_master_list(struct dlm_ctxt *dlm,
+			   u8 dead_node);
+void dlm_force_free_mles(struct dlm_ctxt *dlm);
+int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock);
+int __dlm_lockres_has_locks(struct dlm_lock_resource *res);
+int __dlm_lockres_unused(struct dlm_lock_resource *res);
+
+static inline const char * dlm_lock_mode_name(int mode)
+{
+	switch (mode) {
+		case LKM_EXMODE:
+			return "EX";
+		case LKM_PRMODE:
+			return "PR";
+		case LKM_NLMODE:
+			return "NL";
+	}
+	return "UNKNOWN";
+}
+
+
+static inline int dlm_lock_compatible(int existing, int request)
+{
+	/* NO_LOCK compatible with all */
+	if (request == LKM_NLMODE ||
+	    existing == LKM_NLMODE)
+		return 1;
+
+	/* EX incompatible with all non-NO_LOCK */
+	if (request == LKM_EXMODE)
+		return 0;
+
+	/* request must be PR, which is compatible with PR */
+	if (existing == LKM_PRMODE)
+		return 1;
+
+	return 0;
+}
+
+static inline int dlm_lock_on_list(struct list_head *head,
+				   struct dlm_lock *lock)
+{
+	struct dlm_lock *tmplock;
+
+	list_for_each_entry(tmplock, head, list) {
+		if (tmplock == lock)
+			return 1;
+	}
+	return 0;
+}
+
+
+static inline enum dlm_status dlm_err_to_dlm_status(int err)
+{
+	enum dlm_status ret;
+	if (err == -ENOMEM)
+		ret = DLM_SYSERR;
+	else if (err == -ETIMEDOUT || o2net_link_down(err, NULL))
+		ret = DLM_NOLOCKMGR;
+	else if (err == -EINVAL)
+		ret = DLM_BADPARAM;
+	else if (err == -ENAMETOOLONG)
+		ret = DLM_IVBUFLEN;
+	else
+		ret = DLM_BADARGS;
+	return ret;
+}
+
+
+static inline void dlm_node_iter_init(unsigned long *map,
+				      struct dlm_node_iter *iter)
+{
+	memcpy(iter->node_map, map, sizeof(iter->node_map));
+	iter->curnode = -1;
+}
+
+static inline int dlm_node_iter_next(struct dlm_node_iter *iter)
+{
+	int bit;
+	bit = find_next_bit(iter->node_map, O2NM_MAX_NODES, iter->curnode+1);
+	if (bit >= O2NM_MAX_NODES) {
+		iter->curnode = O2NM_MAX_NODES;
+		return -ENOENT;
+	}
+	iter->curnode = bit;
+	return bit;
+}
+
+static inline void dlm_set_lockres_owner(struct dlm_ctxt *dlm,
+					 struct dlm_lock_resource *res,
+					 u8 owner)
+{
+	assert_spin_locked(&res->spinlock);
+
+	res->owner = owner;
+}
+
+static inline void dlm_change_lockres_owner(struct dlm_ctxt *dlm,
+					    struct dlm_lock_resource *res,
+					    u8 owner)
+{
+	assert_spin_locked(&res->spinlock);
+
+	if (owner != res->owner)
+		dlm_set_lockres_owner(dlm, res, owner);
+}
+
+#endif /* DLMCOMMON_H */
diff --git a/fs/ocfs2/dlm/dlmconvert.c b/fs/ocfs2/dlm/dlmconvert.c
new file mode 100644
index 000000000..e36d63ff1
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmconvert.c
@@ -0,0 +1,544 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmconvert.c
+ *
+ * underlying calls for lock conversion
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/sysctl.h>
+#include <linux/random.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/spinlock.h>
+
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+
+#include "dlmconvert.h"
+
+#define MLOG_MASK_PREFIX ML_DLM
+#include "cluster/masklog.h"
+
+/* NOTE: __dlmconvert_master is the only function in here that
+ * needs a spinlock held on entry (res->spinlock) and it is the
+ * only one that holds a lock on exit (res->spinlock).
+ * All other functions in here need no locks and drop all of
+ * the locks that they acquire. */
+static enum dlm_status __dlmconvert_master(struct dlm_ctxt *dlm,
+					   struct dlm_lock_resource *res,
+					   struct dlm_lock *lock, int flags,
+					   int type, int *call_ast,
+					   int *kick_thread);
+static enum dlm_status dlm_send_remote_convert_request(struct dlm_ctxt *dlm,
+					   struct dlm_lock_resource *res,
+					   struct dlm_lock *lock, int flags, int type);
+
+/*
+ * this is only called directly by dlmlock(), and only when the
+ * local node is the owner of the lockres
+ * locking:
+ *   caller needs:  none
+ *   taken:         takes and drops res->spinlock
+ *   held on exit:  none
+ * returns: see __dlmconvert_master
+ */
+enum dlm_status dlmconvert_master(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *res,
+				  struct dlm_lock *lock, int flags, int type)
+{
+	int call_ast = 0, kick_thread = 0;
+	enum dlm_status status;
+
+	spin_lock(&res->spinlock);
+	/* we are not in a network handler, this is fine */
+	__dlm_wait_on_lockres(res);
+	__dlm_lockres_reserve_ast(res);
+	res->state |= DLM_LOCK_RES_IN_PROGRESS;
+
+	status = __dlmconvert_master(dlm, res, lock, flags, type,
+				     &call_ast, &kick_thread);
+
+	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
+	spin_unlock(&res->spinlock);
+	wake_up(&res->wq);
+	if (status != DLM_NORMAL && status != DLM_NOTQUEUED)
+		dlm_error(status);
+
+	/* either queue the ast or release it */
+	if (call_ast)
+		dlm_queue_ast(dlm, lock);
+	else
+		dlm_lockres_release_ast(dlm, res);
+
+	if (kick_thread)
+		dlm_kick_thread(dlm, res);
+
+	return status;
+}
+
+/* performs lock conversion at the lockres master site
+ * locking:
+ *   caller needs:  res->spinlock
+ *   taken:         takes and drops lock->spinlock
+ *   held on exit:  res->spinlock
+ * returns: DLM_NORMAL, DLM_NOTQUEUED, DLM_DENIED
+ *   call_ast: whether ast should be called for this lock
+ *   kick_thread: whether dlm_kick_thread should be called
+ */
+static enum dlm_status __dlmconvert_master(struct dlm_ctxt *dlm,
+					   struct dlm_lock_resource *res,
+					   struct dlm_lock *lock, int flags,
+					   int type, int *call_ast,
+					   int *kick_thread)
+{
+	enum dlm_status status = DLM_NORMAL;
+	struct dlm_lock *tmplock=NULL;
+
+	assert_spin_locked(&res->spinlock);
+
+	mlog(0, "type=%d, convert_type=%d, new convert_type=%d\n",
+	     lock->ml.type, lock->ml.convert_type, type);
+
+	spin_lock(&lock->spinlock);
+
+	/* already converting? */
+	if (lock->ml.convert_type != LKM_IVMODE) {
+		mlog(ML_ERROR, "attempted to convert a lock with a lock "
+		     "conversion pending\n");
+		status = DLM_DENIED;
+		goto unlock_exit;
+	}
+
+	/* must be on grant queue to convert */
+	if (!dlm_lock_on_list(&res->granted, lock)) {
+		mlog(ML_ERROR, "attempted to convert a lock not on grant "
+		     "queue\n");
+		status = DLM_DENIED;
+		goto unlock_exit;
+	}
+
+	if (flags & LKM_VALBLK) {
+		switch (lock->ml.type) {
+			case LKM_EXMODE:
+				/* EX + LKM_VALBLK + convert == set lvb */
+				mlog(0, "will set lvb: converting %s->%s\n",
+				     dlm_lock_mode_name(lock->ml.type),
+				     dlm_lock_mode_name(type));
+				lock->lksb->flags |= DLM_LKSB_PUT_LVB;
+				break;
+			case LKM_PRMODE:
+			case LKM_NLMODE:
+				/* refetch if new level is not NL */
+				if (type > LKM_NLMODE) {
+					mlog(0, "will fetch new value into "
+					     "lvb: converting %s->%s\n",
+					     dlm_lock_mode_name(lock->ml.type),
+					     dlm_lock_mode_name(type));
+					lock->lksb->flags |= DLM_LKSB_GET_LVB;
+				} else {
+					mlog(0, "will NOT fetch new value "
+					     "into lvb: converting %s->%s\n",
+					     dlm_lock_mode_name(lock->ml.type),
+					     dlm_lock_mode_name(type));
+					flags &= ~(LKM_VALBLK);
+				}
+				break;
+		}
+	}
+
+
+	/* in-place downconvert? */
+	if (type <= lock->ml.type)
+		goto grant;
+
+	/* upconvert from here on */
+	status = DLM_NORMAL;
+	list_for_each_entry(tmplock, &res->granted, list) {
+		if (tmplock == lock)
+			continue;
+		if (!dlm_lock_compatible(tmplock->ml.type, type))
+			goto switch_queues;
+	}
+
+	list_for_each_entry(tmplock, &res->converting, list) {
+		if (!dlm_lock_compatible(tmplock->ml.type, type))
+			goto switch_queues;
+		/* existing conversion requests take precedence */
+		if (!dlm_lock_compatible(tmplock->ml.convert_type, type))
+			goto switch_queues;
+	}
+
+	/* fall thru to grant */
+
+grant:
+	mlog(0, "res %.*s, granting %s lock\n", res->lockname.len,
+	     res->lockname.name, dlm_lock_mode_name(type));
+	/* immediately grant the new lock type */
+	lock->lksb->status = DLM_NORMAL;
+	if (lock->ml.node == dlm->node_num)
+		mlog(0, "doing in-place convert for nonlocal lock\n");
+	lock->ml.type = type;
+	if (lock->lksb->flags & DLM_LKSB_PUT_LVB)
+		memcpy(res->lvb, lock->lksb->lvb, DLM_LVB_LEN);
+
+	status = DLM_NORMAL;
+	*call_ast = 1;
+	goto unlock_exit;
+
+switch_queues:
+	if (flags & LKM_NOQUEUE) {
+		mlog(0, "failed to convert NOQUEUE lock %.*s from "
+		     "%d to %d...\n", res->lockname.len, res->lockname.name,
+		     lock->ml.type, type);
+		status = DLM_NOTQUEUED;
+		goto unlock_exit;
+	}
+	mlog(0, "res %.*s, queueing...\n", res->lockname.len,
+	     res->lockname.name);
+
+	lock->ml.convert_type = type;
+	/* do not alter lock refcount.  switching lists. */
+	list_move_tail(&lock->list, &res->converting);
+
+unlock_exit:
+	spin_unlock(&lock->spinlock);
+	if (status == DLM_DENIED) {
+		__dlm_print_one_lock_resource(res);
+	}
+	if (status == DLM_NORMAL)
+		*kick_thread = 1;
+	return status;
+}
+
+void dlm_revert_pending_convert(struct dlm_lock_resource *res,
+				struct dlm_lock *lock)
+{
+	/* do not alter lock refcount.  switching lists. */
+	list_move_tail(&lock->list, &res->granted);
+	lock->ml.convert_type = LKM_IVMODE;
+	lock->lksb->flags &= ~(DLM_LKSB_GET_LVB|DLM_LKSB_PUT_LVB);
+}
+
+/* messages the master site to do lock conversion
+ * locking:
+ *   caller needs:  none
+ *   taken:         takes and drops res->spinlock, uses DLM_LOCK_RES_IN_PROGRESS
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_RECOVERING, status from remote node
+ */
+enum dlm_status dlmconvert_remote(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *res,
+				  struct dlm_lock *lock, int flags, int type)
+{
+	enum dlm_status status;
+
+	mlog(0, "type=%d, convert_type=%d, busy=%d\n", lock->ml.type,
+	     lock->ml.convert_type, res->state & DLM_LOCK_RES_IN_PROGRESS);
+
+	spin_lock(&res->spinlock);
+	if (res->state & DLM_LOCK_RES_RECOVERING) {
+		mlog(0, "bailing out early since res is RECOVERING "
+		     "on secondary queue\n");
+		/* __dlm_print_one_lock_resource(res); */
+		status = DLM_RECOVERING;
+		goto bail;
+	}
+	/* will exit this call with spinlock held */
+	__dlm_wait_on_lockres(res);
+
+	if (lock->ml.convert_type != LKM_IVMODE) {
+		__dlm_print_one_lock_resource(res);
+		mlog(ML_ERROR, "converting a remote lock that is already "
+		     "converting! (cookie=%u:%llu, conv=%d)\n",
+		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+		     lock->ml.convert_type);
+		status = DLM_DENIED;
+		goto bail;
+	}
+	res->state |= DLM_LOCK_RES_IN_PROGRESS;
+	/* move lock to local convert queue */
+	/* do not alter lock refcount.  switching lists. */
+	list_move_tail(&lock->list, &res->converting);
+	lock->convert_pending = 1;
+	lock->ml.convert_type = type;
+
+	if (flags & LKM_VALBLK) {
+		if (lock->ml.type == LKM_EXMODE) {
+			flags |= LKM_PUT_LVB;
+			lock->lksb->flags |= DLM_LKSB_PUT_LVB;
+		} else {
+			if (lock->ml.convert_type == LKM_NLMODE)
+				flags &= ~LKM_VALBLK;
+			else {
+				flags |= LKM_GET_LVB;
+				lock->lksb->flags |= DLM_LKSB_GET_LVB;
+			}
+		}
+	}
+	spin_unlock(&res->spinlock);
+
+	/* no locks held here.
+	 * need to wait for a reply as to whether it got queued or not. */
+	status = dlm_send_remote_convert_request(dlm, res, lock, flags, type);
+
+	spin_lock(&res->spinlock);
+	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
+	lock->convert_pending = 0;
+	/* if it failed, move it back to granted queue */
+	if (status != DLM_NORMAL) {
+		if (status != DLM_NOTQUEUED)
+			dlm_error(status);
+		dlm_revert_pending_convert(res, lock);
+	}
+bail:
+	spin_unlock(&res->spinlock);
+
+	/* TODO: should this be a wake_one? */
+	/* wake up any IN_PROGRESS waiters */
+	wake_up(&res->wq);
+
+	return status;
+}
+
+/* sends DLM_CONVERT_LOCK_MSG to master site
+ * locking:
+ *   caller needs:  none
+ *   taken:         none
+ *   held on exit:  none
+ * returns: DLM_NOLOCKMGR, status from remote node
+ */
+static enum dlm_status dlm_send_remote_convert_request(struct dlm_ctxt *dlm,
+					   struct dlm_lock_resource *res,
+					   struct dlm_lock *lock, int flags, int type)
+{
+	struct dlm_convert_lock convert;
+	int tmpret;
+	enum dlm_status ret;
+	int status = 0;
+	struct kvec vec[2];
+	size_t veclen = 1;
+
+	mlog(0, "%.*s\n", res->lockname.len, res->lockname.name);
+
+	memset(&convert, 0, sizeof(struct dlm_convert_lock));
+	convert.node_idx = dlm->node_num;
+	convert.requested_type = type;
+	convert.cookie = lock->ml.cookie;
+	convert.namelen = res->lockname.len;
+	convert.flags = cpu_to_be32(flags);
+	memcpy(convert.name, res->lockname.name, convert.namelen);
+
+	vec[0].iov_len = sizeof(struct dlm_convert_lock);
+	vec[0].iov_base = &convert;
+
+	if (flags & LKM_PUT_LVB) {
+		/* extra data to send if we are updating lvb */
+		vec[1].iov_len = DLM_LVB_LEN;
+		vec[1].iov_base = lock->lksb->lvb;
+		veclen++;
+	}
+
+	tmpret = o2net_send_message_vec(DLM_CONVERT_LOCK_MSG, dlm->key,
+					vec, veclen, res->owner, &status);
+	if (tmpret >= 0) {
+		// successfully sent and received
+		ret = status;  // this is already a dlm_status
+		if (ret == DLM_RECOVERING) {
+			mlog(0, "node %u returned DLM_RECOVERING from convert "
+			     "message!\n", res->owner);
+		} else if (ret == DLM_MIGRATING) {
+			mlog(0, "node %u returned DLM_MIGRATING from convert "
+			     "message!\n", res->owner);
+		} else if (ret == DLM_FORWARD) {
+			mlog(0, "node %u returned DLM_FORWARD from convert "
+			     "message!\n", res->owner);
+		} else if (ret != DLM_NORMAL && ret != DLM_NOTQUEUED)
+			dlm_error(ret);
+	} else {
+		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
+		     "node %u\n", tmpret, DLM_CONVERT_LOCK_MSG, dlm->key,
+		     res->owner);
+		if (dlm_is_host_down(tmpret)) {
+			/* instead of logging the same network error over
+			 * and over, sleep here and wait for the heartbeat
+			 * to notice the node is dead.  times out after 5s. */
+			dlm_wait_for_node_death(dlm, res->owner,
+						DLM_NODE_DEATH_WAIT_MAX);
+			ret = DLM_RECOVERING;
+			mlog(0, "node %u died so returning DLM_RECOVERING "
+			     "from convert message!\n", res->owner);
+		} else {
+			ret = dlm_err_to_dlm_status(tmpret);
+		}
+	}
+
+	return ret;
+}
+
+/* handler for DLM_CONVERT_LOCK_MSG on master site
+ * locking:
+ *   caller needs:  none
+ *   taken:         takes and drop res->spinlock
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_IVLOCKID, DLM_BADARGS,
+ *          status from __dlmconvert_master
+ */
+int dlm_convert_lock_handler(struct o2net_msg *msg, u32 len, void *data,
+			     void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_convert_lock *cnv = (struct dlm_convert_lock *)msg->buf;
+	struct dlm_lock_resource *res = NULL;
+	struct dlm_lock *lock = NULL;
+	struct dlm_lock *tmp_lock;
+	struct dlm_lockstatus *lksb;
+	enum dlm_status status = DLM_NORMAL;
+	u32 flags;
+	int call_ast = 0, kick_thread = 0, ast_reserved = 0, wake = 0;
+
+	if (!dlm_grab(dlm)) {
+		dlm_error(DLM_REJECTED);
+		return DLM_REJECTED;
+	}
+
+	mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
+			"Domain %s not fully joined!\n", dlm->name);
+
+	if (cnv->namelen > DLM_LOCKID_NAME_MAX) {
+		status = DLM_IVBUFLEN;
+		dlm_error(status);
+		goto leave;
+	}
+
+	flags = be32_to_cpu(cnv->flags);
+
+	if ((flags & (LKM_PUT_LVB|LKM_GET_LVB)) ==
+	     (LKM_PUT_LVB|LKM_GET_LVB)) {
+		mlog(ML_ERROR, "both PUT and GET lvb specified\n");
+		status = DLM_BADARGS;
+		goto leave;
+	}
+
+	mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" :
+	     (flags & LKM_GET_LVB ? "get lvb" : "none"));
+
+	status = DLM_IVLOCKID;
+	res = dlm_lookup_lockres(dlm, cnv->name, cnv->namelen);
+	if (!res) {
+		dlm_error(status);
+		goto leave;
+	}
+
+	spin_lock(&res->spinlock);
+	status = __dlm_lockres_state_to_status(res);
+	if (status != DLM_NORMAL) {
+		spin_unlock(&res->spinlock);
+		dlm_error(status);
+		goto leave;
+	}
+	list_for_each_entry(tmp_lock, &res->granted, list) {
+		if (tmp_lock->ml.cookie == cnv->cookie &&
+		    tmp_lock->ml.node == cnv->node_idx) {
+			lock = tmp_lock;
+			dlm_lock_get(lock);
+			break;
+		}
+	}
+	spin_unlock(&res->spinlock);
+	if (!lock) {
+		status = DLM_IVLOCKID;
+		mlog(ML_ERROR, "did not find lock to convert on grant queue! "
+			       "cookie=%u:%llu\n",
+		     dlm_get_lock_cookie_node(be64_to_cpu(cnv->cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(cnv->cookie)));
+		dlm_print_one_lock_resource(res);
+		goto leave;
+	}
+
+	/* found the lock */
+	lksb = lock->lksb;
+
+	/* see if caller needed to get/put lvb */
+	if (flags & LKM_PUT_LVB) {
+		BUG_ON(lksb->flags & (DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
+		lksb->flags |= DLM_LKSB_PUT_LVB;
+		memcpy(&lksb->lvb[0], &cnv->lvb[0], DLM_LVB_LEN);
+	} else if (flags & LKM_GET_LVB) {
+		BUG_ON(lksb->flags & (DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
+		lksb->flags |= DLM_LKSB_GET_LVB;
+	}
+
+	spin_lock(&res->spinlock);
+	status = __dlm_lockres_state_to_status(res);
+	if (status == DLM_NORMAL) {
+		__dlm_lockres_reserve_ast(res);
+		ast_reserved = 1;
+		res->state |= DLM_LOCK_RES_IN_PROGRESS;
+		status = __dlmconvert_master(dlm, res, lock, flags,
+					     cnv->requested_type,
+					     &call_ast, &kick_thread);
+		res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
+		wake = 1;
+	}
+	spin_unlock(&res->spinlock);
+	if (wake)
+		wake_up(&res->wq);
+
+	if (status != DLM_NORMAL) {
+		if (status != DLM_NOTQUEUED)
+			dlm_error(status);
+		lksb->flags &= ~(DLM_LKSB_GET_LVB|DLM_LKSB_PUT_LVB);
+	}
+
+leave:
+	if (lock)
+		dlm_lock_put(lock);
+
+	/* either queue the ast or release it, if reserved */
+	if (call_ast)
+		dlm_queue_ast(dlm, lock);
+	else if (ast_reserved)
+		dlm_lockres_release_ast(dlm, res);
+
+	if (kick_thread)
+		dlm_kick_thread(dlm, res);
+
+	if (res)
+		dlm_lockres_put(res);
+
+	dlm_put(dlm);
+
+	return status;
+}
diff --git a/fs/ocfs2/dlm/dlmconvert.h b/fs/ocfs2/dlm/dlmconvert.h
new file mode 100644
index 000000000..b2e3677df
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmconvert.h
@@ -0,0 +1,35 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmconvert.h
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef DLMCONVERT_H
+#define DLMCONVERT_H
+
+enum dlm_status dlmconvert_master(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *res,
+				  struct dlm_lock *lock, int flags, int type);
+enum dlm_status dlmconvert_remote(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *res,
+				  struct dlm_lock *lock, int flags, int type);
+
+#endif
diff --git a/fs/ocfs2/dlm/dlmdebug.c b/fs/ocfs2/dlm/dlmdebug.c
new file mode 100644
index 000000000..825136070
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmdebug.c
@@ -0,0 +1,1000 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmdebug.c
+ *
+ * debug functionality for the dlm
+ *
+ * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/sysctl.h>
+#include <linux/spinlock.h>
+#include <linux/debugfs.h>
+#include <linux/export.h>
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+#include "dlmdomain.h"
+#include "dlmdebug.h"
+
+#define MLOG_MASK_PREFIX ML_DLM
+#include "cluster/masklog.h"
+
+static int stringify_lockname(const char *lockname, int locklen, char *buf,
+			      int len);
+
+void dlm_print_one_lock_resource(struct dlm_lock_resource *res)
+{
+	spin_lock(&res->spinlock);
+	__dlm_print_one_lock_resource(res);
+	spin_unlock(&res->spinlock);
+}
+
+static void dlm_print_lockres_refmap(struct dlm_lock_resource *res)
+{
+	int bit;
+	assert_spin_locked(&res->spinlock);
+
+	printk("  refmap nodes: [ ");
+	bit = 0;
+	while (1) {
+		bit = find_next_bit(res->refmap, O2NM_MAX_NODES, bit);
+		if (bit >= O2NM_MAX_NODES)
+			break;
+		printk("%u ", bit);
+		bit++;
+	}
+	printk("], inflight=%u\n", res->inflight_locks);
+}
+
+static void __dlm_print_lock(struct dlm_lock *lock)
+{
+	spin_lock(&lock->spinlock);
+
+	printk("    type=%d, conv=%d, node=%u, cookie=%u:%llu, "
+	       "ref=%u, ast=(empty=%c,pend=%c), bast=(empty=%c,pend=%c), "
+	       "pending=(conv=%c,lock=%c,cancel=%c,unlock=%c)\n",
+	       lock->ml.type, lock->ml.convert_type, lock->ml.node,
+	       dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+	       dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+	       atomic_read(&lock->lock_refs.refcount),
+	       (list_empty(&lock->ast_list) ? 'y' : 'n'),
+	       (lock->ast_pending ? 'y' : 'n'),
+	       (list_empty(&lock->bast_list) ? 'y' : 'n'),
+	       (lock->bast_pending ? 'y' : 'n'),
+	       (lock->convert_pending ? 'y' : 'n'),
+	       (lock->lock_pending ? 'y' : 'n'),
+	       (lock->cancel_pending ? 'y' : 'n'),
+	       (lock->unlock_pending ? 'y' : 'n'));
+
+	spin_unlock(&lock->spinlock);
+}
+
+void __dlm_print_one_lock_resource(struct dlm_lock_resource *res)
+{
+	struct dlm_lock *lock;
+	char buf[DLM_LOCKID_NAME_MAX];
+
+	assert_spin_locked(&res->spinlock);
+
+	stringify_lockname(res->lockname.name, res->lockname.len,
+			   buf, sizeof(buf));
+	printk("lockres: %s, owner=%u, state=%u\n",
+	       buf, res->owner, res->state);
+	printk("  last used: %lu, refcnt: %u, on purge list: %s\n",
+	       res->last_used, atomic_read(&res->refs.refcount),
+	       list_empty(&res->purge) ? "no" : "yes");
+	printk("  on dirty list: %s, on reco list: %s, "
+	       "migrating pending: %s\n",
+	       list_empty(&res->dirty) ? "no" : "yes",
+	       list_empty(&res->recovering) ? "no" : "yes",
+	       res->migration_pending ? "yes" : "no");
+	printk("  inflight locks: %d, asts reserved: %d\n",
+	       res->inflight_locks, atomic_read(&res->asts_reserved));
+	dlm_print_lockres_refmap(res);
+	printk("  granted queue:\n");
+	list_for_each_entry(lock, &res->granted, list) {
+		__dlm_print_lock(lock);
+	}
+	printk("  converting queue:\n");
+	list_for_each_entry(lock, &res->converting, list) {
+		__dlm_print_lock(lock);
+	}
+	printk("  blocked queue:\n");
+	list_for_each_entry(lock, &res->blocked, list) {
+		__dlm_print_lock(lock);
+	}
+}
+
+void dlm_print_one_lock(struct dlm_lock *lockid)
+{
+	dlm_print_one_lock_resource(lockid->lockres);
+}
+EXPORT_SYMBOL_GPL(dlm_print_one_lock);
+
+static const char *dlm_errnames[] = {
+	[DLM_NORMAL] =			"DLM_NORMAL",
+	[DLM_GRANTED] =			"DLM_GRANTED",
+	[DLM_DENIED] =			"DLM_DENIED",
+	[DLM_DENIED_NOLOCKS] =		"DLM_DENIED_NOLOCKS",
+	[DLM_WORKING] =			"DLM_WORKING",
+	[DLM_BLOCKED] =			"DLM_BLOCKED",
+	[DLM_BLOCKED_ORPHAN] =		"DLM_BLOCKED_ORPHAN",
+	[DLM_DENIED_GRACE_PERIOD] =	"DLM_DENIED_GRACE_PERIOD",
+	[DLM_SYSERR] =			"DLM_SYSERR",
+	[DLM_NOSUPPORT] =		"DLM_NOSUPPORT",
+	[DLM_CANCELGRANT] =		"DLM_CANCELGRANT",
+	[DLM_IVLOCKID] =		"DLM_IVLOCKID",
+	[DLM_SYNC] =			"DLM_SYNC",
+	[DLM_BADTYPE] =			"DLM_BADTYPE",
+	[DLM_BADRESOURCE] =		"DLM_BADRESOURCE",
+	[DLM_MAXHANDLES] =		"DLM_MAXHANDLES",
+	[DLM_NOCLINFO] =		"DLM_NOCLINFO",
+	[DLM_NOLOCKMGR] =		"DLM_NOLOCKMGR",
+	[DLM_NOPURGED] =		"DLM_NOPURGED",
+	[DLM_BADARGS] =			"DLM_BADARGS",
+	[DLM_VOID] =			"DLM_VOID",
+	[DLM_NOTQUEUED] =		"DLM_NOTQUEUED",
+	[DLM_IVBUFLEN] =		"DLM_IVBUFLEN",
+	[DLM_CVTUNGRANT] =		"DLM_CVTUNGRANT",
+	[DLM_BADPARAM] =		"DLM_BADPARAM",
+	[DLM_VALNOTVALID] =		"DLM_VALNOTVALID",
+	[DLM_REJECTED] =		"DLM_REJECTED",
+	[DLM_ABORT] =			"DLM_ABORT",
+	[DLM_CANCEL] =			"DLM_CANCEL",
+	[DLM_IVRESHANDLE] =		"DLM_IVRESHANDLE",
+	[DLM_DEADLOCK] =		"DLM_DEADLOCK",
+	[DLM_DENIED_NOASTS] =		"DLM_DENIED_NOASTS",
+	[DLM_FORWARD] =			"DLM_FORWARD",
+	[DLM_TIMEOUT] =			"DLM_TIMEOUT",
+	[DLM_IVGROUPID] =		"DLM_IVGROUPID",
+	[DLM_VERS_CONFLICT] =		"DLM_VERS_CONFLICT",
+	[DLM_BAD_DEVICE_PATH] =		"DLM_BAD_DEVICE_PATH",
+	[DLM_NO_DEVICE_PERMISSION] =	"DLM_NO_DEVICE_PERMISSION",
+	[DLM_NO_CONTROL_DEVICE ] =	"DLM_NO_CONTROL_DEVICE ",
+	[DLM_RECOVERING] =		"DLM_RECOVERING",
+	[DLM_MIGRATING] =		"DLM_MIGRATING",
+	[DLM_MAXSTATS] =		"DLM_MAXSTATS",
+};
+
+static const char *dlm_errmsgs[] = {
+	[DLM_NORMAL] = 			"request in progress",
+	[DLM_GRANTED] = 		"request granted",
+	[DLM_DENIED] = 			"request denied",
+	[DLM_DENIED_NOLOCKS] = 		"request denied, out of system resources",
+	[DLM_WORKING] = 		"async request in progress",
+	[DLM_BLOCKED] = 		"lock request blocked",
+	[DLM_BLOCKED_ORPHAN] = 		"lock request blocked by a orphan lock",
+	[DLM_DENIED_GRACE_PERIOD] = 	"topological change in progress",
+	[DLM_SYSERR] = 			"system error",
+	[DLM_NOSUPPORT] = 		"unsupported",
+	[DLM_CANCELGRANT] = 		"can't cancel convert: already granted",
+	[DLM_IVLOCKID] = 		"bad lockid",
+	[DLM_SYNC] = 			"synchronous request granted",
+	[DLM_BADTYPE] = 		"bad resource type",
+	[DLM_BADRESOURCE] = 		"bad resource handle",
+	[DLM_MAXHANDLES] = 		"no more resource handles",
+	[DLM_NOCLINFO] = 		"can't contact cluster manager",
+	[DLM_NOLOCKMGR] = 		"can't contact lock manager",
+	[DLM_NOPURGED] = 		"can't contact purge daemon",
+	[DLM_BADARGS] = 		"bad api args",
+	[DLM_VOID] = 			"no status",
+	[DLM_NOTQUEUED] = 		"NOQUEUE was specified and request failed",
+	[DLM_IVBUFLEN] = 		"invalid resource name length",
+	[DLM_CVTUNGRANT] = 		"attempted to convert ungranted lock",
+	[DLM_BADPARAM] = 		"invalid lock mode specified",
+	[DLM_VALNOTVALID] = 		"value block has been invalidated",
+	[DLM_REJECTED] = 		"request rejected, unrecognized client",
+	[DLM_ABORT] = 			"blocked lock request cancelled",
+	[DLM_CANCEL] = 			"conversion request cancelled",
+	[DLM_IVRESHANDLE] = 		"invalid resource handle",
+	[DLM_DEADLOCK] = 		"deadlock recovery refused this request",
+	[DLM_DENIED_NOASTS] = 		"failed to allocate AST",
+	[DLM_FORWARD] = 		"request must wait for primary's response",
+	[DLM_TIMEOUT] = 		"timeout value for lock has expired",
+	[DLM_IVGROUPID] = 		"invalid group specification",
+	[DLM_VERS_CONFLICT] = 		"version conflicts prevent request handling",
+	[DLM_BAD_DEVICE_PATH] = 	"Locks device does not exist or path wrong",
+	[DLM_NO_DEVICE_PERMISSION] = 	"Client has insufficient perms for device",
+	[DLM_NO_CONTROL_DEVICE] = 	"Cannot set options on opened device ",
+	[DLM_RECOVERING] = 		"lock resource being recovered",
+	[DLM_MIGRATING] = 		"lock resource being migrated",
+	[DLM_MAXSTATS] = 		"invalid error number",
+};
+
+const char *dlm_errmsg(enum dlm_status err)
+{
+	if (err >= DLM_MAXSTATS || err < 0)
+		return dlm_errmsgs[DLM_MAXSTATS];
+	return dlm_errmsgs[err];
+}
+EXPORT_SYMBOL_GPL(dlm_errmsg);
+
+const char *dlm_errname(enum dlm_status err)
+{
+	if (err >= DLM_MAXSTATS || err < 0)
+		return dlm_errnames[DLM_MAXSTATS];
+	return dlm_errnames[err];
+}
+EXPORT_SYMBOL_GPL(dlm_errname);
+
+/* NOTE: This function converts a lockname into a string. It uses knowledge
+ * of the format of the lockname that should be outside the purview of the dlm.
+ * We are adding only to make dlm debugging slightly easier.
+ *
+ * For more on lockname formats, please refer to dlmglue.c and ocfs2_lockid.h.
+ */
+static int stringify_lockname(const char *lockname, int locklen, char *buf,
+			      int len)
+{
+	int out = 0;
+	__be64 inode_blkno_be;
+
+#define OCFS2_DENTRY_LOCK_INO_START	18
+	if (*lockname == 'N') {
+		memcpy((__be64 *)&inode_blkno_be,
+		       (char *)&lockname[OCFS2_DENTRY_LOCK_INO_START],
+		       sizeof(__be64));
+		out += snprintf(buf + out, len - out, "%.*s%08x",
+				OCFS2_DENTRY_LOCK_INO_START - 1, lockname,
+				(unsigned int)be64_to_cpu(inode_blkno_be));
+	} else
+		out += snprintf(buf + out, len - out, "%.*s",
+				locklen, lockname);
+	return out;
+}
+
+static int stringify_nodemap(unsigned long *nodemap, int maxnodes,
+			     char *buf, int len)
+{
+	int out = 0;
+	int i = -1;
+
+	while ((i = find_next_bit(nodemap, maxnodes, i + 1)) < maxnodes)
+		out += snprintf(buf + out, len - out, "%d ", i);
+
+	return out;
+}
+
+static int dump_mle(struct dlm_master_list_entry *mle, char *buf, int len)
+{
+	int out = 0;
+	char *mle_type;
+
+	if (mle->type == DLM_MLE_BLOCK)
+		mle_type = "BLK";
+	else if (mle->type == DLM_MLE_MASTER)
+		mle_type = "MAS";
+	else
+		mle_type = "MIG";
+
+	out += stringify_lockname(mle->mname, mle->mnamelen, buf + out, len - out);
+	out += snprintf(buf + out, len - out,
+			"\t%3s\tmas=%3u\tnew=%3u\tevt=%1d\tuse=%1d\tref=%3d\n",
+			mle_type, mle->master, mle->new_master,
+			!list_empty(&mle->hb_events),
+			!!mle->inuse,
+			atomic_read(&mle->mle_refs.refcount));
+
+	out += snprintf(buf + out, len - out, "Maybe=");
+	out += stringify_nodemap(mle->maybe_map, O2NM_MAX_NODES,
+				 buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+	out += snprintf(buf + out, len - out, "Vote=");
+	out += stringify_nodemap(mle->vote_map, O2NM_MAX_NODES,
+				 buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+	out += snprintf(buf + out, len - out, "Response=");
+	out += stringify_nodemap(mle->response_map, O2NM_MAX_NODES,
+				 buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+	out += snprintf(buf + out, len - out, "Node=");
+	out += stringify_nodemap(mle->node_map, O2NM_MAX_NODES,
+				 buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+	out += snprintf(buf + out, len - out, "\n");
+
+	return out;
+}
+
+void dlm_print_one_mle(struct dlm_master_list_entry *mle)
+{
+	char *buf;
+
+	buf = (char *) get_zeroed_page(GFP_NOFS);
+	if (buf) {
+		dump_mle(mle, buf, PAGE_SIZE - 1);
+		free_page((unsigned long)buf);
+	}
+}
+
+#ifdef CONFIG_DEBUG_FS
+
+static struct dentry *dlm_debugfs_root;
+
+#define DLM_DEBUGFS_DIR				"o2dlm"
+#define DLM_DEBUGFS_DLM_STATE			"dlm_state"
+#define DLM_DEBUGFS_LOCKING_STATE		"locking_state"
+#define DLM_DEBUGFS_MLE_STATE			"mle_state"
+#define DLM_DEBUGFS_PURGE_LIST			"purge_list"
+
+/* begin - utils funcs */
+static void dlm_debug_free(struct kref *kref)
+{
+	struct dlm_debug_ctxt *dc;
+
+	dc = container_of(kref, struct dlm_debug_ctxt, debug_refcnt);
+
+	kfree(dc);
+}
+
+static void dlm_debug_put(struct dlm_debug_ctxt *dc)
+{
+	if (dc)
+		kref_put(&dc->debug_refcnt, dlm_debug_free);
+}
+
+static void dlm_debug_get(struct dlm_debug_ctxt *dc)
+{
+	kref_get(&dc->debug_refcnt);
+}
+
+static int debug_release(struct inode *inode, struct file *file)
+{
+	free_page((unsigned long)file->private_data);
+	return 0;
+}
+
+static ssize_t debug_read(struct file *file, char __user *buf,
+			  size_t nbytes, loff_t *ppos)
+{
+	return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
+				       i_size_read(file->f_mapping->host));
+}
+/* end - util funcs */
+
+/* begin - purge list funcs */
+static int debug_purgelist_print(struct dlm_ctxt *dlm, char *buf, int len)
+{
+	struct dlm_lock_resource *res;
+	int out = 0;
+	unsigned long total = 0;
+
+	out += snprintf(buf + out, len - out,
+			"Dumping Purgelist for Domain: %s\n", dlm->name);
+
+	spin_lock(&dlm->spinlock);
+	list_for_each_entry(res, &dlm->purge_list, purge) {
+		++total;
+		if (len - out < 100)
+			continue;
+		spin_lock(&res->spinlock);
+		out += stringify_lockname(res->lockname.name,
+					  res->lockname.len,
+					  buf + out, len - out);
+		out += snprintf(buf + out, len - out, "\t%ld\n",
+				(jiffies - res->last_used)/HZ);
+		spin_unlock(&res->spinlock);
+	}
+	spin_unlock(&dlm->spinlock);
+
+	out += snprintf(buf + out, len - out, "Total on list: %lu\n", total);
+
+	return out;
+}
+
+static int debug_purgelist_open(struct inode *inode, struct file *file)
+{
+	struct dlm_ctxt *dlm = inode->i_private;
+	char *buf = NULL;
+
+	buf = (char *) get_zeroed_page(GFP_NOFS);
+	if (!buf)
+		goto bail;
+
+	i_size_write(inode, debug_purgelist_print(dlm, buf, PAGE_SIZE - 1));
+
+	file->private_data = buf;
+
+	return 0;
+bail:
+	return -ENOMEM;
+}
+
+static const struct file_operations debug_purgelist_fops = {
+	.open =		debug_purgelist_open,
+	.release =	debug_release,
+	.read =		debug_read,
+	.llseek =	generic_file_llseek,
+};
+/* end - purge list funcs */
+
+/* begin - debug mle funcs */
+static int debug_mle_print(struct dlm_ctxt *dlm, char *buf, int len)
+{
+	struct dlm_master_list_entry *mle;
+	struct hlist_head *bucket;
+	int i, out = 0;
+	unsigned long total = 0, longest = 0, bucket_count = 0;
+
+	out += snprintf(buf + out, len - out,
+			"Dumping MLEs for Domain: %s\n", dlm->name);
+
+	spin_lock(&dlm->master_lock);
+	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
+		bucket = dlm_master_hash(dlm, i);
+		hlist_for_each_entry(mle, bucket, master_hash_node) {
+			++total;
+			++bucket_count;
+			if (len - out < 200)
+				continue;
+			out += dump_mle(mle, buf + out, len - out);
+		}
+		longest = max(longest, bucket_count);
+		bucket_count = 0;
+	}
+	spin_unlock(&dlm->master_lock);
+
+	out += snprintf(buf + out, len - out,
+			"Total: %lu, Longest: %lu\n", total, longest);
+	return out;
+}
+
+static int debug_mle_open(struct inode *inode, struct file *file)
+{
+	struct dlm_ctxt *dlm = inode->i_private;
+	char *buf = NULL;
+
+	buf = (char *) get_zeroed_page(GFP_NOFS);
+	if (!buf)
+		goto bail;
+
+	i_size_write(inode, debug_mle_print(dlm, buf, PAGE_SIZE - 1));
+
+	file->private_data = buf;
+
+	return 0;
+bail:
+	return -ENOMEM;
+}
+
+static const struct file_operations debug_mle_fops = {
+	.open =		debug_mle_open,
+	.release =	debug_release,
+	.read =		debug_read,
+	.llseek =	generic_file_llseek,
+};
+
+/* end - debug mle funcs */
+
+/* begin - debug lockres funcs */
+static int dump_lock(struct dlm_lock *lock, int list_type, char *buf, int len)
+{
+	int out;
+
+#define DEBUG_LOCK_VERSION	1
+	spin_lock(&lock->spinlock);
+	out = snprintf(buf, len, "LOCK:%d,%d,%d,%d,%d,%d:%lld,%d,%d,%d,%d,%d,"
+		       "%d,%d,%d,%d\n",
+		       DEBUG_LOCK_VERSION,
+		       list_type, lock->ml.type, lock->ml.convert_type,
+		       lock->ml.node,
+		       dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+		       dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+		       !list_empty(&lock->ast_list),
+		       !list_empty(&lock->bast_list),
+		       lock->ast_pending, lock->bast_pending,
+		       lock->convert_pending, lock->lock_pending,
+		       lock->cancel_pending, lock->unlock_pending,
+		       atomic_read(&lock->lock_refs.refcount));
+	spin_unlock(&lock->spinlock);
+
+	return out;
+}
+
+static int dump_lockres(struct dlm_lock_resource *res, char *buf, int len)
+{
+	struct dlm_lock *lock;
+	int i;
+	int out = 0;
+
+	out += snprintf(buf + out, len - out, "NAME:");
+	out += stringify_lockname(res->lockname.name, res->lockname.len,
+				  buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+#define DEBUG_LRES_VERSION	1
+	out += snprintf(buf + out, len - out,
+			"LRES:%d,%d,%d,%ld,%d,%d,%d,%d,%d,%d,%d\n",
+			DEBUG_LRES_VERSION,
+			res->owner, res->state, res->last_used,
+			!list_empty(&res->purge),
+			!list_empty(&res->dirty),
+			!list_empty(&res->recovering),
+			res->inflight_locks, res->migration_pending,
+			atomic_read(&res->asts_reserved),
+			atomic_read(&res->refs.refcount));
+
+	/* refmap */
+	out += snprintf(buf + out, len - out, "RMAP:");
+	out += stringify_nodemap(res->refmap, O2NM_MAX_NODES,
+				 buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+	/* lvb */
+	out += snprintf(buf + out, len - out, "LVBX:");
+	for (i = 0; i < DLM_LVB_LEN; i++)
+		out += snprintf(buf + out, len - out,
+					"%02x", (unsigned char)res->lvb[i]);
+	out += snprintf(buf + out, len - out, "\n");
+
+	/* granted */
+	list_for_each_entry(lock, &res->granted, list)
+		out += dump_lock(lock, 0, buf + out, len - out);
+
+	/* converting */
+	list_for_each_entry(lock, &res->converting, list)
+		out += dump_lock(lock, 1, buf + out, len - out);
+
+	/* blocked */
+	list_for_each_entry(lock, &res->blocked, list)
+		out += dump_lock(lock, 2, buf + out, len - out);
+
+	out += snprintf(buf + out, len - out, "\n");
+
+	return out;
+}
+
+static void *lockres_seq_start(struct seq_file *m, loff_t *pos)
+{
+	struct debug_lockres *dl = m->private;
+	struct dlm_ctxt *dlm = dl->dl_ctxt;
+	struct dlm_lock_resource *oldres = dl->dl_res;
+	struct dlm_lock_resource *res = NULL;
+	struct list_head *track_list;
+
+	spin_lock(&dlm->track_lock);
+	if (oldres)
+		track_list = &oldres->tracking;
+	else {
+		track_list = &dlm->tracking_list;
+		if (list_empty(track_list)) {
+			dl = NULL;
+			spin_unlock(&dlm->track_lock);
+			goto bail;
+		}
+	}
+
+	list_for_each_entry(res, track_list, tracking) {
+		if (&res->tracking == &dlm->tracking_list)
+			res = NULL;
+		else
+			dlm_lockres_get(res);
+		break;
+	}
+	spin_unlock(&dlm->track_lock);
+
+	if (oldres)
+		dlm_lockres_put(oldres);
+
+	dl->dl_res = res;
+
+	if (res) {
+		spin_lock(&res->spinlock);
+		dump_lockres(res, dl->dl_buf, dl->dl_len - 1);
+		spin_unlock(&res->spinlock);
+	} else
+		dl = NULL;
+
+bail:
+	/* passed to seq_show */
+	return dl;
+}
+
+static void lockres_seq_stop(struct seq_file *m, void *v)
+{
+}
+
+static void *lockres_seq_next(struct seq_file *m, void *v, loff_t *pos)
+{
+	return NULL;
+}
+
+static int lockres_seq_show(struct seq_file *s, void *v)
+{
+	struct debug_lockres *dl = (struct debug_lockres *)v;
+
+	seq_printf(s, "%s", dl->dl_buf);
+
+	return 0;
+}
+
+static const struct seq_operations debug_lockres_ops = {
+	.start =	lockres_seq_start,
+	.stop =		lockres_seq_stop,
+	.next =		lockres_seq_next,
+	.show =		lockres_seq_show,
+};
+
+static int debug_lockres_open(struct inode *inode, struct file *file)
+{
+	struct dlm_ctxt *dlm = inode->i_private;
+	struct debug_lockres *dl;
+	void *buf;
+
+	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!buf)
+		goto bail;
+
+	dl = __seq_open_private(file, &debug_lockres_ops, sizeof(*dl));
+	if (!dl)
+		goto bailfree;
+
+	dl->dl_len = PAGE_SIZE;
+	dl->dl_buf = buf;
+
+	dlm_grab(dlm);
+	dl->dl_ctxt = dlm;
+
+	return 0;
+
+bailfree:
+	kfree(buf);
+bail:
+	mlog_errno(-ENOMEM);
+	return -ENOMEM;
+}
+
+static int debug_lockres_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq = file->private_data;
+	struct debug_lockres *dl = (struct debug_lockres *)seq->private;
+
+	if (dl->dl_res)
+		dlm_lockres_put(dl->dl_res);
+	dlm_put(dl->dl_ctxt);
+	kfree(dl->dl_buf);
+	return seq_release_private(inode, file);
+}
+
+static const struct file_operations debug_lockres_fops = {
+	.open =		debug_lockres_open,
+	.release =	debug_lockres_release,
+	.read =		seq_read,
+	.llseek =	seq_lseek,
+};
+/* end - debug lockres funcs */
+
+/* begin - debug state funcs */
+static int debug_state_print(struct dlm_ctxt *dlm, char *buf, int len)
+{
+	int out = 0;
+	struct dlm_reco_node_data *node;
+	char *state;
+	int cur_mles = 0, tot_mles = 0;
+	int i;
+
+	spin_lock(&dlm->spinlock);
+
+	switch (dlm->dlm_state) {
+	case DLM_CTXT_NEW:
+		state = "NEW"; break;
+	case DLM_CTXT_JOINED:
+		state = "JOINED"; break;
+	case DLM_CTXT_IN_SHUTDOWN:
+		state = "SHUTDOWN"; break;
+	case DLM_CTXT_LEAVING:
+		state = "LEAVING"; break;
+	default:
+		state = "UNKNOWN"; break;
+	}
+
+	/* Domain: xxxxxxxxxx  Key: 0xdfbac769 */
+	out += snprintf(buf + out, len - out,
+			"Domain: %s  Key: 0x%08x  Protocol: %d.%d\n",
+			dlm->name, dlm->key, dlm->dlm_locking_proto.pv_major,
+			dlm->dlm_locking_proto.pv_minor);
+
+	/* Thread Pid: xxx  Node: xxx  State: xxxxx */
+	out += snprintf(buf + out, len - out,
+			"Thread Pid: %d  Node: %d  State: %s\n",
+			task_pid_nr(dlm->dlm_thread_task), dlm->node_num, state);
+
+	/* Number of Joins: xxx  Joining Node: xxx */
+	out += snprintf(buf + out, len - out,
+			"Number of Joins: %d  Joining Node: %d\n",
+			dlm->num_joins, dlm->joining_node);
+
+	/* Domain Map: xx xx xx */
+	out += snprintf(buf + out, len - out, "Domain Map: ");
+	out += stringify_nodemap(dlm->domain_map, O2NM_MAX_NODES,
+				 buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+	/* Exit Domain Map: xx xx xx */
+	out += snprintf(buf + out, len - out, "Exit Domain Map: ");
+	out += stringify_nodemap(dlm->exit_domain_map, O2NM_MAX_NODES,
+				 buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+	/* Live Map: xx xx xx */
+	out += snprintf(buf + out, len - out, "Live Map: ");
+	out += stringify_nodemap(dlm->live_nodes_map, O2NM_MAX_NODES,
+				 buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+	/* Lock Resources: xxx (xxx) */
+	out += snprintf(buf + out, len - out,
+			"Lock Resources: %d (%d)\n",
+			atomic_read(&dlm->res_cur_count),
+			atomic_read(&dlm->res_tot_count));
+
+	for (i = 0; i < DLM_MLE_NUM_TYPES; ++i)
+		tot_mles += atomic_read(&dlm->mle_tot_count[i]);
+
+	for (i = 0; i < DLM_MLE_NUM_TYPES; ++i)
+		cur_mles += atomic_read(&dlm->mle_cur_count[i]);
+
+	/* MLEs: xxx (xxx) */
+	out += snprintf(buf + out, len - out,
+			"MLEs: %d (%d)\n", cur_mles, tot_mles);
+
+	/*  Blocking: xxx (xxx) */
+	out += snprintf(buf + out, len - out,
+			"  Blocking: %d (%d)\n",
+			atomic_read(&dlm->mle_cur_count[DLM_MLE_BLOCK]),
+			atomic_read(&dlm->mle_tot_count[DLM_MLE_BLOCK]));
+
+	/*  Mastery: xxx (xxx) */
+	out += snprintf(buf + out, len - out,
+			"  Mastery: %d (%d)\n",
+			atomic_read(&dlm->mle_cur_count[DLM_MLE_MASTER]),
+			atomic_read(&dlm->mle_tot_count[DLM_MLE_MASTER]));
+
+	/*  Migration: xxx (xxx) */
+	out += snprintf(buf + out, len - out,
+			"  Migration: %d (%d)\n",
+			atomic_read(&dlm->mle_cur_count[DLM_MLE_MIGRATION]),
+			atomic_read(&dlm->mle_tot_count[DLM_MLE_MIGRATION]));
+
+	/* Lists: Dirty=Empty  Purge=InUse  PendingASTs=Empty  ... */
+	out += snprintf(buf + out, len - out,
+			"Lists: Dirty=%s  Purge=%s  PendingASTs=%s  "
+			"PendingBASTs=%s\n",
+			(list_empty(&dlm->dirty_list) ? "Empty" : "InUse"),
+			(list_empty(&dlm->purge_list) ? "Empty" : "InUse"),
+			(list_empty(&dlm->pending_asts) ? "Empty" : "InUse"),
+			(list_empty(&dlm->pending_basts) ? "Empty" : "InUse"));
+
+	/* Purge Count: xxx  Refs: xxx */
+	out += snprintf(buf + out, len - out,
+			"Purge Count: %d  Refs: %d\n", dlm->purge_count,
+			atomic_read(&dlm->dlm_refs.refcount));
+
+	/* Dead Node: xxx */
+	out += snprintf(buf + out, len - out,
+			"Dead Node: %d\n", dlm->reco.dead_node);
+
+	/* What about DLM_RECO_STATE_FINALIZE? */
+	if (dlm->reco.state == DLM_RECO_STATE_ACTIVE)
+		state = "ACTIVE";
+	else
+		state = "INACTIVE";
+
+	/* Recovery Pid: xxxx  Master: xxx  State: xxxx */
+	out += snprintf(buf + out, len - out,
+			"Recovery Pid: %d  Master: %d  State: %s\n",
+			task_pid_nr(dlm->dlm_reco_thread_task),
+			dlm->reco.new_master, state);
+
+	/* Recovery Map: xx xx */
+	out += snprintf(buf + out, len - out, "Recovery Map: ");
+	out += stringify_nodemap(dlm->recovery_map, O2NM_MAX_NODES,
+				 buf + out, len - out);
+	out += snprintf(buf + out, len - out, "\n");
+
+	/* Recovery Node State: */
+	out += snprintf(buf + out, len - out, "Recovery Node State:\n");
+	list_for_each_entry(node, &dlm->reco.node_data, list) {
+		switch (node->state) {
+		case DLM_RECO_NODE_DATA_INIT:
+			state = "INIT";
+			break;
+		case DLM_RECO_NODE_DATA_REQUESTING:
+			state = "REQUESTING";
+			break;
+		case DLM_RECO_NODE_DATA_DEAD:
+			state = "DEAD";
+			break;
+		case DLM_RECO_NODE_DATA_RECEIVING:
+			state = "RECEIVING";
+			break;
+		case DLM_RECO_NODE_DATA_REQUESTED:
+			state = "REQUESTED";
+			break;
+		case DLM_RECO_NODE_DATA_DONE:
+			state = "DONE";
+			break;
+		case DLM_RECO_NODE_DATA_FINALIZE_SENT:
+			state = "FINALIZE-SENT";
+			break;
+		default:
+			state = "BAD";
+			break;
+		}
+		out += snprintf(buf + out, len - out, "\t%u - %s\n",
+				node->node_num, state);
+	}
+
+	spin_unlock(&dlm->spinlock);
+
+	return out;
+}
+
+static int debug_state_open(struct inode *inode, struct file *file)
+{
+	struct dlm_ctxt *dlm = inode->i_private;
+	char *buf = NULL;
+
+	buf = (char *) get_zeroed_page(GFP_NOFS);
+	if (!buf)
+		goto bail;
+
+	i_size_write(inode, debug_state_print(dlm, buf, PAGE_SIZE - 1));
+
+	file->private_data = buf;
+
+	return 0;
+bail:
+	return -ENOMEM;
+}
+
+static const struct file_operations debug_state_fops = {
+	.open =		debug_state_open,
+	.release =	debug_release,
+	.read =		debug_read,
+	.llseek =	generic_file_llseek,
+};
+/* end  - debug state funcs */
+
+/* files in subroot */
+int dlm_debug_init(struct dlm_ctxt *dlm)
+{
+	struct dlm_debug_ctxt *dc = dlm->dlm_debug_ctxt;
+
+	/* for dumping dlm_ctxt */
+	dc->debug_state_dentry = debugfs_create_file(DLM_DEBUGFS_DLM_STATE,
+						     S_IFREG|S_IRUSR,
+						     dlm->dlm_debugfs_subroot,
+						     dlm, &debug_state_fops);
+	if (!dc->debug_state_dentry) {
+		mlog_errno(-ENOMEM);
+		goto bail;
+	}
+
+	/* for dumping lockres */
+	dc->debug_lockres_dentry =
+			debugfs_create_file(DLM_DEBUGFS_LOCKING_STATE,
+					    S_IFREG|S_IRUSR,
+					    dlm->dlm_debugfs_subroot,
+					    dlm, &debug_lockres_fops);
+	if (!dc->debug_lockres_dentry) {
+		mlog_errno(-ENOMEM);
+		goto bail;
+	}
+
+	/* for dumping mles */
+	dc->debug_mle_dentry = debugfs_create_file(DLM_DEBUGFS_MLE_STATE,
+						   S_IFREG|S_IRUSR,
+						   dlm->dlm_debugfs_subroot,
+						   dlm, &debug_mle_fops);
+	if (!dc->debug_mle_dentry) {
+		mlog_errno(-ENOMEM);
+		goto bail;
+	}
+
+	/* for dumping lockres on the purge list */
+	dc->debug_purgelist_dentry =
+			debugfs_create_file(DLM_DEBUGFS_PURGE_LIST,
+					    S_IFREG|S_IRUSR,
+					    dlm->dlm_debugfs_subroot,
+					    dlm, &debug_purgelist_fops);
+	if (!dc->debug_purgelist_dentry) {
+		mlog_errno(-ENOMEM);
+		goto bail;
+	}
+
+	dlm_debug_get(dc);
+	return 0;
+
+bail:
+	dlm_debug_shutdown(dlm);
+	return -ENOMEM;
+}
+
+void dlm_debug_shutdown(struct dlm_ctxt *dlm)
+{
+	struct dlm_debug_ctxt *dc = dlm->dlm_debug_ctxt;
+
+	if (dc) {
+		debugfs_remove(dc->debug_purgelist_dentry);
+		debugfs_remove(dc->debug_mle_dentry);
+		debugfs_remove(dc->debug_lockres_dentry);
+		debugfs_remove(dc->debug_state_dentry);
+		dlm_debug_put(dc);
+	}
+}
+
+/* subroot - domain dir */
+int dlm_create_debugfs_subroot(struct dlm_ctxt *dlm)
+{
+	dlm->dlm_debugfs_subroot = debugfs_create_dir(dlm->name,
+						      dlm_debugfs_root);
+	if (!dlm->dlm_debugfs_subroot) {
+		mlog_errno(-ENOMEM);
+		goto bail;
+	}
+
+	dlm->dlm_debug_ctxt = kzalloc(sizeof(struct dlm_debug_ctxt),
+				      GFP_KERNEL);
+	if (!dlm->dlm_debug_ctxt) {
+		mlog_errno(-ENOMEM);
+		goto bail;
+	}
+	kref_init(&dlm->dlm_debug_ctxt->debug_refcnt);
+
+	return 0;
+bail:
+	dlm_destroy_debugfs_subroot(dlm);
+	return -ENOMEM;
+}
+
+void dlm_destroy_debugfs_subroot(struct dlm_ctxt *dlm)
+{
+	debugfs_remove(dlm->dlm_debugfs_subroot);
+}
+
+/* debugfs root */
+int dlm_create_debugfs_root(void)
+{
+	dlm_debugfs_root = debugfs_create_dir(DLM_DEBUGFS_DIR, NULL);
+	if (!dlm_debugfs_root) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+	return 0;
+}
+
+void dlm_destroy_debugfs_root(void)
+{
+	debugfs_remove(dlm_debugfs_root);
+}
+#endif	/* CONFIG_DEBUG_FS */
diff --git a/fs/ocfs2/dlm/dlmdebug.h b/fs/ocfs2/dlm/dlmdebug.h
new file mode 100644
index 000000000..1f27c4812
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmdebug.h
@@ -0,0 +1,81 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmdebug.h
+ *
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef DLMDEBUG_H
+#define DLMDEBUG_H
+
+void dlm_print_one_mle(struct dlm_master_list_entry *mle);
+
+#ifdef CONFIG_DEBUG_FS
+
+struct dlm_debug_ctxt {
+	struct kref debug_refcnt;
+	struct dentry *debug_state_dentry;
+	struct dentry *debug_lockres_dentry;
+	struct dentry *debug_mle_dentry;
+	struct dentry *debug_purgelist_dentry;
+};
+
+struct debug_lockres {
+	int dl_len;
+	char *dl_buf;
+	struct dlm_ctxt *dl_ctxt;
+	struct dlm_lock_resource *dl_res;
+};
+
+int dlm_debug_init(struct dlm_ctxt *dlm);
+void dlm_debug_shutdown(struct dlm_ctxt *dlm);
+
+int dlm_create_debugfs_subroot(struct dlm_ctxt *dlm);
+void dlm_destroy_debugfs_subroot(struct dlm_ctxt *dlm);
+
+int dlm_create_debugfs_root(void);
+void dlm_destroy_debugfs_root(void);
+
+#else
+
+static inline int dlm_debug_init(struct dlm_ctxt *dlm)
+{
+	return 0;
+}
+static inline void dlm_debug_shutdown(struct dlm_ctxt *dlm)
+{
+}
+static inline int dlm_create_debugfs_subroot(struct dlm_ctxt *dlm)
+{
+	return 0;
+}
+static inline void dlm_destroy_debugfs_subroot(struct dlm_ctxt *dlm)
+{
+}
+static inline int dlm_create_debugfs_root(void)
+{
+	return 0;
+}
+static inline void dlm_destroy_debugfs_root(void)
+{
+}
+
+#endif	/* CONFIG_DEBUG_FS */
+#endif	/* DLMDEBUG_H */
diff --git a/fs/ocfs2/dlm/dlmdomain.c b/fs/ocfs2/dlm/dlmdomain.c
new file mode 100644
index 000000000..7df88a6dd
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmdomain.c
@@ -0,0 +1,2375 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmdomain.c
+ *
+ * defines domain join / leave apis
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/spinlock.h>
+#include <linux/delay.h>
+#include <linux/err.h>
+#include <linux/debugfs.h>
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+#include "dlmdomain.h"
+#include "dlmdebug.h"
+
+#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
+#include "cluster/masklog.h"
+
+/*
+ * ocfs2 node maps are array of long int, which limits to send them freely
+ * across the wire due to endianness issues. To workaround this, we convert
+ * long ints to byte arrays. Following 3 routines are helper functions to
+ * set/test/copy bits within those array of bytes
+ */
+static inline void byte_set_bit(u8 nr, u8 map[])
+{
+	map[nr >> 3] |= (1UL << (nr & 7));
+}
+
+static inline int byte_test_bit(u8 nr, u8 map[])
+{
+	return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
+}
+
+static inline void byte_copymap(u8 dmap[], unsigned long smap[],
+			unsigned int sz)
+{
+	unsigned int nn;
+
+	if (!sz)
+		return;
+
+	memset(dmap, 0, ((sz + 7) >> 3));
+	for (nn = 0 ; nn < sz; nn++)
+		if (test_bit(nn, smap))
+			byte_set_bit(nn, dmap);
+}
+
+static void dlm_free_pagevec(void **vec, int pages)
+{
+	while (pages--)
+		free_page((unsigned long)vec[pages]);
+	kfree(vec);
+}
+
+static void **dlm_alloc_pagevec(int pages)
+{
+	void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
+	int i;
+
+	if (!vec)
+		return NULL;
+
+	for (i = 0; i < pages; i++)
+		if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
+			goto out_free;
+
+	mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
+	     pages, (unsigned long)DLM_HASH_PAGES,
+	     (unsigned long)DLM_BUCKETS_PER_PAGE);
+	return vec;
+out_free:
+	dlm_free_pagevec(vec, i);
+	return NULL;
+}
+
+/*
+ *
+ * spinlock lock ordering: if multiple locks are needed, obey this ordering:
+ *    dlm_domain_lock
+ *    struct dlm_ctxt->spinlock
+ *    struct dlm_lock_resource->spinlock
+ *    struct dlm_ctxt->master_lock
+ *    struct dlm_ctxt->ast_lock
+ *    dlm_master_list_entry->spinlock
+ *    dlm_lock->spinlock
+ *
+ */
+
+DEFINE_SPINLOCK(dlm_domain_lock);
+LIST_HEAD(dlm_domains);
+static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
+
+/*
+ * The supported protocol version for DLM communication.  Running domains
+ * will have a negotiated version with the same major number and a minor
+ * number equal or smaller.  The dlm_ctxt->dlm_locking_proto field should
+ * be used to determine what a running domain is actually using.
+ *
+ * New in version 1.1:
+ *	- Message DLM_QUERY_REGION added to support global heartbeat
+ *	- Message DLM_QUERY_NODEINFO added to allow online node removes
+ * New in version 1.2:
+ * 	- Message DLM_BEGIN_EXIT_DOMAIN_MSG added to mark start of exit domain
+ */
+static const struct dlm_protocol_version dlm_protocol = {
+	.pv_major = 1,
+	.pv_minor = 2,
+};
+
+#define DLM_DOMAIN_BACKOFF_MS 200
+
+static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
+				  void **ret_data);
+static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
+				     void **ret_data);
+static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
+				   void **ret_data);
+static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
+				    void *data, void **ret_data);
+static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
+				   void **ret_data);
+static int dlm_protocol_compare(struct dlm_protocol_version *existing,
+				struct dlm_protocol_version *request);
+
+static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
+
+void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
+{
+	if (hlist_unhashed(&res->hash_node))
+		return;
+
+	mlog(0, "%s: Unhash res %.*s\n", dlm->name, res->lockname.len,
+	     res->lockname.name);
+	hlist_del_init(&res->hash_node);
+	dlm_lockres_put(res);
+}
+
+void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
+{
+	struct hlist_head *bucket;
+	struct qstr *q;
+
+	assert_spin_locked(&dlm->spinlock);
+
+	q = &res->lockname;
+	bucket = dlm_lockres_hash(dlm, q->hash);
+
+	/* get a reference for our hashtable */
+	dlm_lockres_get(res);
+
+	hlist_add_head(&res->hash_node, bucket);
+
+	mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len,
+	     res->lockname.name);
+}
+
+struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
+						     const char *name,
+						     unsigned int len,
+						     unsigned int hash)
+{
+	struct hlist_head *bucket;
+	struct dlm_lock_resource *res;
+
+	mlog(0, "%.*s\n", len, name);
+
+	assert_spin_locked(&dlm->spinlock);
+
+	bucket = dlm_lockres_hash(dlm, hash);
+
+	hlist_for_each_entry(res, bucket, hash_node) {
+		if (res->lockname.name[0] != name[0])
+			continue;
+		if (unlikely(res->lockname.len != len))
+			continue;
+		if (memcmp(res->lockname.name + 1, name + 1, len - 1))
+			continue;
+		dlm_lockres_get(res);
+		return res;
+	}
+	return NULL;
+}
+
+/* intended to be called by functions which do not care about lock
+ * resources which are being purged (most net _handler functions).
+ * this will return NULL for any lock resource which is found but
+ * currently in the process of dropping its mastery reference.
+ * use __dlm_lookup_lockres_full when you need the lock resource
+ * regardless (e.g. dlm_get_lock_resource) */
+struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
+						const char *name,
+						unsigned int len,
+						unsigned int hash)
+{
+	struct dlm_lock_resource *res = NULL;
+
+	mlog(0, "%.*s\n", len, name);
+
+	assert_spin_locked(&dlm->spinlock);
+
+	res = __dlm_lookup_lockres_full(dlm, name, len, hash);
+	if (res) {
+		spin_lock(&res->spinlock);
+		if (res->state & DLM_LOCK_RES_DROPPING_REF) {
+			spin_unlock(&res->spinlock);
+			dlm_lockres_put(res);
+			return NULL;
+		}
+		spin_unlock(&res->spinlock);
+	}
+
+	return res;
+}
+
+struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
+				    const char *name,
+				    unsigned int len)
+{
+	struct dlm_lock_resource *res;
+	unsigned int hash = dlm_lockid_hash(name, len);
+
+	spin_lock(&dlm->spinlock);
+	res = __dlm_lookup_lockres(dlm, name, len, hash);
+	spin_unlock(&dlm->spinlock);
+	return res;
+}
+
+static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
+{
+	struct dlm_ctxt *tmp;
+
+	assert_spin_locked(&dlm_domain_lock);
+
+	/* tmp->name here is always NULL terminated,
+	 * but domain may not be! */
+	list_for_each_entry(tmp, &dlm_domains, list) {
+		if (strlen(tmp->name) == len &&
+		    memcmp(tmp->name, domain, len)==0)
+			return tmp;
+	}
+
+	return NULL;
+}
+
+/* For null terminated domain strings ONLY */
+static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
+{
+	assert_spin_locked(&dlm_domain_lock);
+
+	return __dlm_lookup_domain_full(domain, strlen(domain));
+}
+
+
+/* returns true on one of two conditions:
+ * 1) the domain does not exist
+ * 2) the domain exists and it's state is "joined" */
+static int dlm_wait_on_domain_helper(const char *domain)
+{
+	int ret = 0;
+	struct dlm_ctxt *tmp = NULL;
+
+	spin_lock(&dlm_domain_lock);
+
+	tmp = __dlm_lookup_domain(domain);
+	if (!tmp)
+		ret = 1;
+	else if (tmp->dlm_state == DLM_CTXT_JOINED)
+		ret = 1;
+
+	spin_unlock(&dlm_domain_lock);
+	return ret;
+}
+
+static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
+{
+	dlm_destroy_debugfs_subroot(dlm);
+
+	if (dlm->lockres_hash)
+		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
+
+	if (dlm->master_hash)
+		dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
+
+	kfree(dlm->name);
+	kfree(dlm);
+}
+
+/* A little strange - this function will be called while holding
+ * dlm_domain_lock and is expected to be holding it on the way out. We
+ * will however drop and reacquire it multiple times */
+static void dlm_ctxt_release(struct kref *kref)
+{
+	struct dlm_ctxt *dlm;
+
+	dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
+
+	BUG_ON(dlm->num_joins);
+	BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
+
+	/* we may still be in the list if we hit an error during join. */
+	list_del_init(&dlm->list);
+
+	spin_unlock(&dlm_domain_lock);
+
+	mlog(0, "freeing memory from domain %s\n", dlm->name);
+
+	wake_up(&dlm_domain_events);
+
+	dlm_free_ctxt_mem(dlm);
+
+	spin_lock(&dlm_domain_lock);
+}
+
+void dlm_put(struct dlm_ctxt *dlm)
+{
+	spin_lock(&dlm_domain_lock);
+	kref_put(&dlm->dlm_refs, dlm_ctxt_release);
+	spin_unlock(&dlm_domain_lock);
+}
+
+static void __dlm_get(struct dlm_ctxt *dlm)
+{
+	kref_get(&dlm->dlm_refs);
+}
+
+/* given a questionable reference to a dlm object, gets a reference if
+ * it can find it in the list, otherwise returns NULL in which case
+ * you shouldn't trust your pointer. */
+struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
+{
+	struct dlm_ctxt *target;
+	struct dlm_ctxt *ret = NULL;
+
+	spin_lock(&dlm_domain_lock);
+
+	list_for_each_entry(target, &dlm_domains, list) {
+		if (target == dlm) {
+			__dlm_get(target);
+			ret = target;
+			break;
+		}
+	}
+
+	spin_unlock(&dlm_domain_lock);
+
+	return ret;
+}
+
+int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
+{
+	int ret;
+
+	spin_lock(&dlm_domain_lock);
+	ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
+		(dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
+	spin_unlock(&dlm_domain_lock);
+
+	return ret;
+}
+
+static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
+{
+	if (dlm->dlm_worker) {
+		flush_workqueue(dlm->dlm_worker);
+		destroy_workqueue(dlm->dlm_worker);
+		dlm->dlm_worker = NULL;
+	}
+}
+
+static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
+{
+	dlm_unregister_domain_handlers(dlm);
+	dlm_debug_shutdown(dlm);
+	dlm_complete_thread(dlm);
+	dlm_complete_recovery_thread(dlm);
+	dlm_destroy_dlm_worker(dlm);
+
+	/* We've left the domain. Now we can take ourselves out of the
+	 * list and allow the kref stuff to help us free the
+	 * memory. */
+	spin_lock(&dlm_domain_lock);
+	list_del_init(&dlm->list);
+	spin_unlock(&dlm_domain_lock);
+
+	/* Wake up anyone waiting for us to remove this domain */
+	wake_up(&dlm_domain_events);
+}
+
+static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
+{
+	int i, num, n, ret = 0;
+	struct dlm_lock_resource *res;
+	struct hlist_node *iter;
+	struct hlist_head *bucket;
+	int dropped;
+
+	mlog(0, "Migrating locks from domain %s\n", dlm->name);
+
+	num = 0;
+	spin_lock(&dlm->spinlock);
+	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
+redo_bucket:
+		n = 0;
+		bucket = dlm_lockres_hash(dlm, i);
+		iter = bucket->first;
+		while (iter) {
+			n++;
+			res = hlist_entry(iter, struct dlm_lock_resource,
+					  hash_node);
+			dlm_lockres_get(res);
+			/* migrate, if necessary.  this will drop the dlm
+			 * spinlock and retake it if it does migration. */
+			dropped = dlm_empty_lockres(dlm, res);
+
+			spin_lock(&res->spinlock);
+			if (dropped)
+				__dlm_lockres_calc_usage(dlm, res);
+			else
+				iter = res->hash_node.next;
+			spin_unlock(&res->spinlock);
+
+			dlm_lockres_put(res);
+
+			if (dropped) {
+				cond_resched_lock(&dlm->spinlock);
+				goto redo_bucket;
+			}
+		}
+		cond_resched_lock(&dlm->spinlock);
+		num += n;
+	}
+	spin_unlock(&dlm->spinlock);
+	wake_up(&dlm->dlm_thread_wq);
+
+	/* let the dlm thread take care of purging, keep scanning until
+	 * nothing remains in the hash */
+	if (num) {
+		mlog(0, "%s: %d lock resources in hash last pass\n",
+		     dlm->name, num);
+		ret = -EAGAIN;
+	}
+	mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
+	return ret;
+}
+
+static int dlm_no_joining_node(struct dlm_ctxt *dlm)
+{
+	int ret;
+
+	spin_lock(&dlm->spinlock);
+	ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
+	spin_unlock(&dlm->spinlock);
+
+	return ret;
+}
+
+static int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len,
+					 void *data, void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	unsigned int node;
+	struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
+
+	if (!dlm_grab(dlm))
+		return 0;
+
+	node = exit_msg->node_idx;
+	mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node);
+
+	spin_lock(&dlm->spinlock);
+	set_bit(node, dlm->exit_domain_map);
+	spin_unlock(&dlm->spinlock);
+
+	dlm_put(dlm);
+
+	return 0;
+}
+
+static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
+{
+	/* Yikes, a double spinlock! I need domain_lock for the dlm
+	 * state and the dlm spinlock for join state... Sorry! */
+again:
+	spin_lock(&dlm_domain_lock);
+	spin_lock(&dlm->spinlock);
+
+	if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
+		mlog(0, "Node %d is joining, we wait on it.\n",
+			  dlm->joining_node);
+		spin_unlock(&dlm->spinlock);
+		spin_unlock(&dlm_domain_lock);
+
+		wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
+		goto again;
+	}
+
+	dlm->dlm_state = DLM_CTXT_LEAVING;
+	spin_unlock(&dlm->spinlock);
+	spin_unlock(&dlm_domain_lock);
+}
+
+static void __dlm_print_nodes(struct dlm_ctxt *dlm)
+{
+	int node = -1, num = 0;
+
+	assert_spin_locked(&dlm->spinlock);
+
+	printk("( ");
+	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
+				     node + 1)) < O2NM_MAX_NODES) {
+		printk("%d ", node);
+		++num;
+	}
+	printk(") %u nodes\n", num);
+}
+
+static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
+				   void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	unsigned int node;
+	struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
+
+	mlog(0, "%p %u %p", msg, len, data);
+
+	if (!dlm_grab(dlm))
+		return 0;
+
+	node = exit_msg->node_idx;
+
+	spin_lock(&dlm->spinlock);
+	clear_bit(node, dlm->domain_map);
+	clear_bit(node, dlm->exit_domain_map);
+	printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name);
+	__dlm_print_nodes(dlm);
+
+	/* notify anything attached to the heartbeat events */
+	dlm_hb_event_notify_attached(dlm, node, 0);
+
+	spin_unlock(&dlm->spinlock);
+
+	dlm_put(dlm);
+
+	return 0;
+}
+
+static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type,
+				    unsigned int node)
+{
+	int status;
+	struct dlm_exit_domain leave_msg;
+
+	mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name,
+	     msg_type, node);
+
+	memset(&leave_msg, 0, sizeof(leave_msg));
+	leave_msg.node_idx = dlm->node_num;
+
+	status = o2net_send_message(msg_type, dlm->key, &leave_msg,
+				    sizeof(leave_msg), node, NULL);
+	if (status < 0)
+		mlog(ML_ERROR, "Error %d sending domain exit message %u "
+		     "to node %u on domain %s\n", status, msg_type, node,
+		     dlm->name);
+
+	return status;
+}
+
+static void dlm_begin_exit_domain(struct dlm_ctxt *dlm)
+{
+	int node = -1;
+
+	/* Support for begin exit domain was added in 1.2 */
+	if (dlm->dlm_locking_proto.pv_major == 1 &&
+	    dlm->dlm_locking_proto.pv_minor < 2)
+		return;
+
+	/*
+	 * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely
+	 * informational. Meaning if a node does not receive the message,
+	 * so be it.
+	 */
+	spin_lock(&dlm->spinlock);
+	while (1) {
+		node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1);
+		if (node >= O2NM_MAX_NODES)
+			break;
+		if (node == dlm->node_num)
+			continue;
+
+		spin_unlock(&dlm->spinlock);
+		dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node);
+		spin_lock(&dlm->spinlock);
+	}
+	spin_unlock(&dlm->spinlock);
+}
+
+static void dlm_leave_domain(struct dlm_ctxt *dlm)
+{
+	int node, clear_node, status;
+
+	/* At this point we've migrated away all our locks and won't
+	 * accept mastership of new ones. The dlm is responsible for
+	 * almost nothing now. We make sure not to confuse any joining
+	 * nodes and then commence shutdown procedure. */
+
+	spin_lock(&dlm->spinlock);
+	/* Clear ourselves from the domain map */
+	clear_bit(dlm->node_num, dlm->domain_map);
+	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
+				     0)) < O2NM_MAX_NODES) {
+		/* Drop the dlm spinlock. This is safe wrt the domain_map.
+		 * -nodes cannot be added now as the
+		 *   query_join_handlers knows to respond with OK_NO_MAP
+		 * -we catch the right network errors if a node is
+		 *   removed from the map while we're sending him the
+		 *   exit message. */
+		spin_unlock(&dlm->spinlock);
+
+		clear_node = 1;
+
+		status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG,
+						  node);
+		if (status < 0 &&
+		    status != -ENOPROTOOPT &&
+		    status != -ENOTCONN) {
+			mlog(ML_NOTICE, "Error %d sending domain exit message "
+			     "to node %d\n", status, node);
+
+			/* Not sure what to do here but lets sleep for
+			 * a bit in case this was a transient
+			 * error... */
+			msleep(DLM_DOMAIN_BACKOFF_MS);
+			clear_node = 0;
+		}
+
+		spin_lock(&dlm->spinlock);
+		/* If we're not clearing the node bit then we intend
+		 * to loop back around to try again. */
+		if (clear_node)
+			clear_bit(node, dlm->domain_map);
+	}
+	spin_unlock(&dlm->spinlock);
+}
+
+int dlm_shutting_down(struct dlm_ctxt *dlm)
+{
+	int ret = 0;
+
+	spin_lock(&dlm_domain_lock);
+
+	if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
+		ret = 1;
+
+	spin_unlock(&dlm_domain_lock);
+
+	return ret;
+}
+
+void dlm_unregister_domain(struct dlm_ctxt *dlm)
+{
+	int leave = 0;
+	struct dlm_lock_resource *res;
+
+	spin_lock(&dlm_domain_lock);
+	BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
+	BUG_ON(!dlm->num_joins);
+
+	dlm->num_joins--;
+	if (!dlm->num_joins) {
+		/* We mark it "in shutdown" now so new register
+		 * requests wait until we've completely left the
+		 * domain. Don't use DLM_CTXT_LEAVING yet as we still
+		 * want new domain joins to communicate with us at
+		 * least until we've completed migration of our
+		 * resources. */
+		dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
+		leave = 1;
+	}
+	spin_unlock(&dlm_domain_lock);
+
+	if (leave) {
+		mlog(0, "shutting down domain %s\n", dlm->name);
+		dlm_begin_exit_domain(dlm);
+
+		/* We changed dlm state, notify the thread */
+		dlm_kick_thread(dlm, NULL);
+
+		while (dlm_migrate_all_locks(dlm)) {
+			/* Give dlm_thread time to purge the lockres' */
+			msleep(500);
+			mlog(0, "%s: more migration to do\n", dlm->name);
+		}
+
+		/* This list should be empty. If not, print remaining lockres */
+		if (!list_empty(&dlm->tracking_list)) {
+			mlog(ML_ERROR, "Following lockres' are still on the "
+			     "tracking list:\n");
+			list_for_each_entry(res, &dlm->tracking_list, tracking)
+				dlm_print_one_lock_resource(res);
+		}
+
+		dlm_mark_domain_leaving(dlm);
+		dlm_leave_domain(dlm);
+		printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name);
+		dlm_force_free_mles(dlm);
+		dlm_complete_dlm_shutdown(dlm);
+	}
+	dlm_put(dlm);
+}
+EXPORT_SYMBOL_GPL(dlm_unregister_domain);
+
+static int dlm_query_join_proto_check(char *proto_type, int node,
+				      struct dlm_protocol_version *ours,
+				      struct dlm_protocol_version *request)
+{
+	int rc;
+	struct dlm_protocol_version proto = *request;
+
+	if (!dlm_protocol_compare(ours, &proto)) {
+		mlog(0,
+		     "node %u wanted to join with %s locking protocol "
+		     "%u.%u, we respond with %u.%u\n",
+		     node, proto_type,
+		     request->pv_major,
+		     request->pv_minor,
+		     proto.pv_major, proto.pv_minor);
+		request->pv_minor = proto.pv_minor;
+		rc = 0;
+	} else {
+		mlog(ML_NOTICE,
+		     "Node %u wanted to join with %s locking "
+		     "protocol %u.%u, but we have %u.%u, disallowing\n",
+		     node, proto_type,
+		     request->pv_major,
+		     request->pv_minor,
+		     ours->pv_major,
+		     ours->pv_minor);
+		rc = 1;
+	}
+
+	return rc;
+}
+
+/*
+ * struct dlm_query_join_packet is made up of four one-byte fields.  They
+ * are effectively in big-endian order already.  However, little-endian
+ * machines swap them before putting the packet on the wire (because
+ * query_join's response is a status, and that status is treated as a u32
+ * on the wire).  Thus, a big-endian and little-endian machines will treat
+ * this structure differently.
+ *
+ * The solution is to have little-endian machines swap the structure when
+ * converting from the structure to the u32 representation.  This will
+ * result in the structure having the correct format on the wire no matter
+ * the host endian format.
+ */
+static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
+					  u32 *wire)
+{
+	union dlm_query_join_response response;
+
+	response.packet = *packet;
+	*wire = be32_to_cpu(response.intval);
+}
+
+static void dlm_query_join_wire_to_packet(u32 wire,
+					  struct dlm_query_join_packet *packet)
+{
+	union dlm_query_join_response response;
+
+	response.intval = cpu_to_be32(wire);
+	*packet = response.packet;
+}
+
+static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
+				  void **ret_data)
+{
+	struct dlm_query_join_request *query;
+	struct dlm_query_join_packet packet = {
+		.code = JOIN_DISALLOW,
+	};
+	struct dlm_ctxt *dlm = NULL;
+	u32 response;
+	u8 nodenum;
+
+	query = (struct dlm_query_join_request *) msg->buf;
+
+	mlog(0, "node %u wants to join domain %s\n", query->node_idx,
+		  query->domain);
+
+	/*
+	 * If heartbeat doesn't consider the node live, tell it
+	 * to back off and try again.  This gives heartbeat a chance
+	 * to catch up.
+	 */
+	if (!o2hb_check_node_heartbeating_no_sem(query->node_idx)) {
+		mlog(0, "node %u is not in our live map yet\n",
+		     query->node_idx);
+
+		packet.code = JOIN_DISALLOW;
+		goto respond;
+	}
+
+	packet.code = JOIN_OK_NO_MAP;
+
+	spin_lock(&dlm_domain_lock);
+	dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
+	if (!dlm)
+		goto unlock_respond;
+
+	/*
+	 * There is a small window where the joining node may not see the
+	 * node(s) that just left but still part of the cluster. DISALLOW
+	 * join request if joining node has different node map.
+	 */
+	nodenum=0;
+	while (nodenum < O2NM_MAX_NODES) {
+		if (test_bit(nodenum, dlm->domain_map)) {
+			if (!byte_test_bit(nodenum, query->node_map)) {
+				mlog(0, "disallow join as node %u does not "
+				     "have node %u in its nodemap\n",
+				     query->node_idx, nodenum);
+				packet.code = JOIN_DISALLOW;
+				goto unlock_respond;
+			}
+		}
+		nodenum++;
+	}
+
+	/* Once the dlm ctxt is marked as leaving then we don't want
+	 * to be put in someone's domain map.
+	 * Also, explicitly disallow joining at certain troublesome
+	 * times (ie. during recovery). */
+	if (dlm->dlm_state != DLM_CTXT_LEAVING) {
+		int bit = query->node_idx;
+		spin_lock(&dlm->spinlock);
+
+		if (dlm->dlm_state == DLM_CTXT_NEW &&
+		    dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
+			/*If this is a brand new context and we
+			 * haven't started our join process yet, then
+			 * the other node won the race. */
+			packet.code = JOIN_OK_NO_MAP;
+		} else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
+			/* Disallow parallel joins. */
+			packet.code = JOIN_DISALLOW;
+		} else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
+			mlog(0, "node %u trying to join, but recovery "
+			     "is ongoing.\n", bit);
+			packet.code = JOIN_DISALLOW;
+		} else if (test_bit(bit, dlm->recovery_map)) {
+			mlog(0, "node %u trying to join, but it "
+			     "still needs recovery.\n", bit);
+			packet.code = JOIN_DISALLOW;
+		} else if (test_bit(bit, dlm->domain_map)) {
+			mlog(0, "node %u trying to join, but it "
+			     "is still in the domain! needs recovery?\n",
+			     bit);
+			packet.code = JOIN_DISALLOW;
+		} else {
+			/* Alright we're fully a part of this domain
+			 * so we keep some state as to who's joining
+			 * and indicate to him that needs to be fixed
+			 * up. */
+
+			/* Make sure we speak compatible locking protocols.  */
+			if (dlm_query_join_proto_check("DLM", bit,
+						       &dlm->dlm_locking_proto,
+						       &query->dlm_proto)) {
+				packet.code = JOIN_PROTOCOL_MISMATCH;
+			} else if (dlm_query_join_proto_check("fs", bit,
+							      &dlm->fs_locking_proto,
+							      &query->fs_proto)) {
+				packet.code = JOIN_PROTOCOL_MISMATCH;
+			} else {
+				packet.dlm_minor = query->dlm_proto.pv_minor;
+				packet.fs_minor = query->fs_proto.pv_minor;
+				packet.code = JOIN_OK;
+				__dlm_set_joining_node(dlm, query->node_idx);
+			}
+		}
+
+		spin_unlock(&dlm->spinlock);
+	}
+unlock_respond:
+	spin_unlock(&dlm_domain_lock);
+
+respond:
+	mlog(0, "We respond with %u\n", packet.code);
+
+	dlm_query_join_packet_to_wire(&packet, &response);
+	return response;
+}
+
+static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
+				     void **ret_data)
+{
+	struct dlm_assert_joined *assert;
+	struct dlm_ctxt *dlm = NULL;
+
+	assert = (struct dlm_assert_joined *) msg->buf;
+
+	mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
+		  assert->domain);
+
+	spin_lock(&dlm_domain_lock);
+	dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
+	/* XXX should we consider no dlm ctxt an error? */
+	if (dlm) {
+		spin_lock(&dlm->spinlock);
+
+		/* Alright, this node has officially joined our
+		 * domain. Set him in the map and clean up our
+		 * leftover join state. */
+		BUG_ON(dlm->joining_node != assert->node_idx);
+
+		if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
+			mlog(0, "dlm recovery is ongoing, disallow join\n");
+			spin_unlock(&dlm->spinlock);
+			spin_unlock(&dlm_domain_lock);
+			return -EAGAIN;
+		}
+
+		set_bit(assert->node_idx, dlm->domain_map);
+		clear_bit(assert->node_idx, dlm->exit_domain_map);
+		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
+
+		printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ",
+		       assert->node_idx, dlm->name);
+		__dlm_print_nodes(dlm);
+
+		/* notify anything attached to the heartbeat events */
+		dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
+
+		spin_unlock(&dlm->spinlock);
+	}
+	spin_unlock(&dlm_domain_lock);
+
+	return 0;
+}
+
+static int dlm_match_regions(struct dlm_ctxt *dlm,
+			     struct dlm_query_region *qr,
+			     char *local, int locallen)
+{
+	char *remote = qr->qr_regions;
+	char *l, *r;
+	int localnr, i, j, foundit;
+	int status = 0;
+
+	if (!o2hb_global_heartbeat_active()) {
+		if (qr->qr_numregions) {
+			mlog(ML_ERROR, "Domain %s: Joining node %d has global "
+			     "heartbeat enabled but local node %d does not\n",
+			     qr->qr_domain, qr->qr_node, dlm->node_num);
+			status = -EINVAL;
+		}
+		goto bail;
+	}
+
+	if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
+		mlog(ML_ERROR, "Domain %s: Local node %d has global "
+		     "heartbeat enabled but joining node %d does not\n",
+		     qr->qr_domain, dlm->node_num, qr->qr_node);
+		status = -EINVAL;
+		goto bail;
+	}
+
+	r = remote;
+	for (i = 0; i < qr->qr_numregions; ++i) {
+		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
+		r += O2HB_MAX_REGION_NAME_LEN;
+	}
+
+	localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN);
+	localnr = o2hb_get_all_regions(local, (u8)localnr);
+
+	/* compare local regions with remote */
+	l = local;
+	for (i = 0; i < localnr; ++i) {
+		foundit = 0;
+		r = remote;
+		for (j = 0; j <= qr->qr_numregions; ++j) {
+			if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
+				foundit = 1;
+				break;
+			}
+			r += O2HB_MAX_REGION_NAME_LEN;
+		}
+		if (!foundit) {
+			status = -EINVAL;
+			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
+			     "in local node %d but not in joining node %d\n",
+			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
+			     dlm->node_num, qr->qr_node);
+			goto bail;
+		}
+		l += O2HB_MAX_REGION_NAME_LEN;
+	}
+
+	/* compare remote with local regions */
+	r = remote;
+	for (i = 0; i < qr->qr_numregions; ++i) {
+		foundit = 0;
+		l = local;
+		for (j = 0; j < localnr; ++j) {
+			if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
+				foundit = 1;
+				break;
+			}
+			l += O2HB_MAX_REGION_NAME_LEN;
+		}
+		if (!foundit) {
+			status = -EINVAL;
+			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
+			     "in joining node %d but not in local node %d\n",
+			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
+			     qr->qr_node, dlm->node_num);
+			goto bail;
+		}
+		r += O2HB_MAX_REGION_NAME_LEN;
+	}
+
+bail:
+	return status;
+}
+
+static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
+{
+	struct dlm_query_region *qr = NULL;
+	int status, ret = 0, i;
+	char *p;
+
+	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
+		goto bail;
+
+	qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
+	if (!qr) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	qr->qr_node = dlm->node_num;
+	qr->qr_namelen = strlen(dlm->name);
+	memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
+	/* if local hb, the numregions will be zero */
+	if (o2hb_global_heartbeat_active())
+		qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
+							 O2NM_MAX_REGIONS);
+
+	p = qr->qr_regions;
+	for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
+		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
+
+	i = -1;
+	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
+				  i + 1)) < O2NM_MAX_NODES) {
+		if (i == dlm->node_num)
+			continue;
+
+		mlog(0, "Sending regions to node %d\n", i);
+
+		ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
+					 sizeof(struct dlm_query_region),
+					 i, &status);
+		if (ret >= 0)
+			ret = status;
+		if (ret) {
+			mlog(ML_ERROR, "Region mismatch %d, node %d\n",
+			     ret, i);
+			break;
+		}
+	}
+
+bail:
+	kfree(qr);
+	return ret;
+}
+
+static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
+				    void *data, void **ret_data)
+{
+	struct dlm_query_region *qr;
+	struct dlm_ctxt *dlm = NULL;
+	char *local = NULL;
+	int status = 0;
+
+	qr = (struct dlm_query_region *) msg->buf;
+
+	mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
+	     qr->qr_domain);
+
+	/* buffer used in dlm_mast_regions() */
+	local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL);
+	if (!local)
+		return -ENOMEM;
+
+	status = -EINVAL;
+
+	spin_lock(&dlm_domain_lock);
+	dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
+	if (!dlm) {
+		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
+		     "before join domain\n", qr->qr_node, qr->qr_domain);
+		goto out_domain_lock;
+	}
+
+	spin_lock(&dlm->spinlock);
+	if (dlm->joining_node != qr->qr_node) {
+		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
+		     "but joining node is %d\n", qr->qr_node, qr->qr_domain,
+		     dlm->joining_node);
+		goto out_dlm_lock;
+	}
+
+	/* Support for global heartbeat was added in 1.1 */
+	if (dlm->dlm_locking_proto.pv_major == 1 &&
+	    dlm->dlm_locking_proto.pv_minor == 0) {
+		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
+		     "but active dlm protocol is %d.%d\n", qr->qr_node,
+		     qr->qr_domain, dlm->dlm_locking_proto.pv_major,
+		     dlm->dlm_locking_proto.pv_minor);
+		goto out_dlm_lock;
+	}
+
+	status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions));
+
+out_dlm_lock:
+	spin_unlock(&dlm->spinlock);
+
+out_domain_lock:
+	spin_unlock(&dlm_domain_lock);
+
+	kfree(local);
+
+	return status;
+}
+
+static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
+{
+	struct o2nm_node *local;
+	struct dlm_node_info *remote;
+	int i, j;
+	int status = 0;
+
+	for (j = 0; j < qn->qn_numnodes; ++j)
+		mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
+		     &(qn->qn_nodes[j].ni_ipv4_address),
+		     ntohs(qn->qn_nodes[j].ni_ipv4_port));
+
+	for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
+		local = o2nm_get_node_by_num(i);
+		remote = NULL;
+		for (j = 0; j < qn->qn_numnodes; ++j) {
+			if (qn->qn_nodes[j].ni_nodenum == i) {
+				remote = &(qn->qn_nodes[j]);
+				break;
+			}
+		}
+
+		if (!local && !remote)
+			continue;
+
+		if ((local && !remote) || (!local && remote))
+			status = -EINVAL;
+
+		if (!status &&
+		    ((remote->ni_nodenum != local->nd_num) ||
+		     (remote->ni_ipv4_port != local->nd_ipv4_port) ||
+		     (remote->ni_ipv4_address != local->nd_ipv4_address)))
+			status = -EINVAL;
+
+		if (status) {
+			if (remote && !local)
+				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
+				     "registered in joining node %d but not in "
+				     "local node %d\n", qn->qn_domain,
+				     remote->ni_nodenum,
+				     &(remote->ni_ipv4_address),
+				     ntohs(remote->ni_ipv4_port),
+				     qn->qn_nodenum, dlm->node_num);
+			if (local && !remote)
+				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
+				     "registered in local node %d but not in "
+				     "joining node %d\n", qn->qn_domain,
+				     local->nd_num, &(local->nd_ipv4_address),
+				     ntohs(local->nd_ipv4_port),
+				     dlm->node_num, qn->qn_nodenum);
+			BUG_ON((!local && !remote));
+		}
+
+		if (local)
+			o2nm_node_put(local);
+	}
+
+	return status;
+}
+
+static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
+{
+	struct dlm_query_nodeinfo *qn = NULL;
+	struct o2nm_node *node;
+	int ret = 0, status, count, i;
+
+	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
+		goto bail;
+
+	qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
+	if (!qn) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
+		node = o2nm_get_node_by_num(i);
+		if (!node)
+			continue;
+		qn->qn_nodes[count].ni_nodenum = node->nd_num;
+		qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
+		qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
+		mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
+		     &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
+		++count;
+		o2nm_node_put(node);
+	}
+
+	qn->qn_nodenum = dlm->node_num;
+	qn->qn_numnodes = count;
+	qn->qn_namelen = strlen(dlm->name);
+	memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
+
+	i = -1;
+	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
+				  i + 1)) < O2NM_MAX_NODES) {
+		if (i == dlm->node_num)
+			continue;
+
+		mlog(0, "Sending nodeinfo to node %d\n", i);
+
+		ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
+					 qn, sizeof(struct dlm_query_nodeinfo),
+					 i, &status);
+		if (ret >= 0)
+			ret = status;
+		if (ret) {
+			mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
+			break;
+		}
+	}
+
+bail:
+	kfree(qn);
+	return ret;
+}
+
+static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
+				      void *data, void **ret_data)
+{
+	struct dlm_query_nodeinfo *qn;
+	struct dlm_ctxt *dlm = NULL;
+	int locked = 0, status = -EINVAL;
+
+	qn = (struct dlm_query_nodeinfo *) msg->buf;
+
+	mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
+	     qn->qn_domain);
+
+	spin_lock(&dlm_domain_lock);
+	dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
+	if (!dlm) {
+		mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
+		     "join domain\n", qn->qn_nodenum, qn->qn_domain);
+		goto bail;
+	}
+
+	spin_lock(&dlm->spinlock);
+	locked = 1;
+	if (dlm->joining_node != qn->qn_nodenum) {
+		mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
+		     "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
+		     dlm->joining_node);
+		goto bail;
+	}
+
+	/* Support for node query was added in 1.1 */
+	if (dlm->dlm_locking_proto.pv_major == 1 &&
+	    dlm->dlm_locking_proto.pv_minor == 0) {
+		mlog(ML_ERROR, "Node %d queried nodes on domain %s "
+		     "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
+		     qn->qn_domain, dlm->dlm_locking_proto.pv_major,
+		     dlm->dlm_locking_proto.pv_minor);
+		goto bail;
+	}
+
+	status = dlm_match_nodes(dlm, qn);
+
+bail:
+	if (locked)
+		spin_unlock(&dlm->spinlock);
+	spin_unlock(&dlm_domain_lock);
+
+	return status;
+}
+
+static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
+				   void **ret_data)
+{
+	struct dlm_cancel_join *cancel;
+	struct dlm_ctxt *dlm = NULL;
+
+	cancel = (struct dlm_cancel_join *) msg->buf;
+
+	mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
+		  cancel->domain);
+
+	spin_lock(&dlm_domain_lock);
+	dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
+
+	if (dlm) {
+		spin_lock(&dlm->spinlock);
+
+		/* Yikes, this guy wants to cancel his join. No
+		 * problem, we simply cleanup our join state. */
+		BUG_ON(dlm->joining_node != cancel->node_idx);
+		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
+
+		spin_unlock(&dlm->spinlock);
+	}
+	spin_unlock(&dlm_domain_lock);
+
+	return 0;
+}
+
+static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
+				    unsigned int node)
+{
+	int status;
+	struct dlm_cancel_join cancel_msg;
+
+	memset(&cancel_msg, 0, sizeof(cancel_msg));
+	cancel_msg.node_idx = dlm->node_num;
+	cancel_msg.name_len = strlen(dlm->name);
+	memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
+
+	status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
+				    &cancel_msg, sizeof(cancel_msg), node,
+				    NULL);
+	if (status < 0) {
+		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
+		     "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
+		     node);
+		goto bail;
+	}
+
+bail:
+	return status;
+}
+
+/* map_size should be in bytes. */
+static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
+				 unsigned long *node_map,
+				 unsigned int map_size)
+{
+	int status, tmpstat;
+	unsigned int node;
+
+	if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
+			 sizeof(unsigned long))) {
+		mlog(ML_ERROR,
+		     "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
+		     map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
+		return -EINVAL;
+	}
+
+	status = 0;
+	node = -1;
+	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
+				     node + 1)) < O2NM_MAX_NODES) {
+		if (node == dlm->node_num)
+			continue;
+
+		tmpstat = dlm_send_one_join_cancel(dlm, node);
+		if (tmpstat) {
+			mlog(ML_ERROR, "Error return %d cancelling join on "
+			     "node %d\n", tmpstat, node);
+			if (!status)
+				status = tmpstat;
+		}
+	}
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static int dlm_request_join(struct dlm_ctxt *dlm,
+			    int node,
+			    enum dlm_query_join_response_code *response)
+{
+	int status;
+	struct dlm_query_join_request join_msg;
+	struct dlm_query_join_packet packet;
+	u32 join_resp;
+
+	mlog(0, "querying node %d\n", node);
+
+	memset(&join_msg, 0, sizeof(join_msg));
+	join_msg.node_idx = dlm->node_num;
+	join_msg.name_len = strlen(dlm->name);
+	memcpy(join_msg.domain, dlm->name, join_msg.name_len);
+	join_msg.dlm_proto = dlm->dlm_locking_proto;
+	join_msg.fs_proto = dlm->fs_locking_proto;
+
+	/* copy live node map to join message */
+	byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
+
+	status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
+				    sizeof(join_msg), node, &join_resp);
+	if (status < 0 && status != -ENOPROTOOPT) {
+		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
+		     "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
+		     node);
+		goto bail;
+	}
+	dlm_query_join_wire_to_packet(join_resp, &packet);
+
+	/* -ENOPROTOOPT from the net code means the other side isn't
+	    listening for our message type -- that's fine, it means
+	    his dlm isn't up, so we can consider him a 'yes' but not
+	    joined into the domain.  */
+	if (status == -ENOPROTOOPT) {
+		status = 0;
+		*response = JOIN_OK_NO_MAP;
+	} else if (packet.code == JOIN_DISALLOW ||
+		   packet.code == JOIN_OK_NO_MAP) {
+		*response = packet.code;
+	} else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
+		mlog(ML_NOTICE,
+		     "This node requested DLM locking protocol %u.%u and "
+		     "filesystem locking protocol %u.%u.  At least one of "
+		     "the protocol versions on node %d is not compatible, "
+		     "disconnecting\n",
+		     dlm->dlm_locking_proto.pv_major,
+		     dlm->dlm_locking_proto.pv_minor,
+		     dlm->fs_locking_proto.pv_major,
+		     dlm->fs_locking_proto.pv_minor,
+		     node);
+		status = -EPROTO;
+		*response = packet.code;
+	} else if (packet.code == JOIN_OK) {
+		*response = packet.code;
+		/* Use the same locking protocol as the remote node */
+		dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
+		dlm->fs_locking_proto.pv_minor = packet.fs_minor;
+		mlog(0,
+		     "Node %d responds JOIN_OK with DLM locking protocol "
+		     "%u.%u and fs locking protocol %u.%u\n",
+		     node,
+		     dlm->dlm_locking_proto.pv_major,
+		     dlm->dlm_locking_proto.pv_minor,
+		     dlm->fs_locking_proto.pv_major,
+		     dlm->fs_locking_proto.pv_minor);
+	} else {
+		status = -EINVAL;
+		mlog(ML_ERROR, "invalid response %d from node %u\n",
+		     packet.code, node);
+	}
+
+	mlog(0, "status %d, node %d response is %d\n", status, node,
+	     *response);
+
+bail:
+	return status;
+}
+
+static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
+				    unsigned int node)
+{
+	int status;
+	int ret;
+	struct dlm_assert_joined assert_msg;
+
+	mlog(0, "Sending join assert to node %u\n", node);
+
+	memset(&assert_msg, 0, sizeof(assert_msg));
+	assert_msg.node_idx = dlm->node_num;
+	assert_msg.name_len = strlen(dlm->name);
+	memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
+
+	status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
+				    &assert_msg, sizeof(assert_msg), node,
+				    &ret);
+	if (status < 0)
+		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
+		     "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
+		     node);
+	else
+		status = ret;
+
+	return status;
+}
+
+static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
+				  unsigned long *node_map)
+{
+	int status, node, live;
+
+	status = 0;
+	node = -1;
+	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
+				     node + 1)) < O2NM_MAX_NODES) {
+		if (node == dlm->node_num)
+			continue;
+
+		do {
+			/* It is very important that this message be
+			 * received so we spin until either the node
+			 * has died or it gets the message. */
+			status = dlm_send_one_join_assert(dlm, node);
+
+			spin_lock(&dlm->spinlock);
+			live = test_bit(node, dlm->live_nodes_map);
+			spin_unlock(&dlm->spinlock);
+
+			if (status) {
+				mlog(ML_ERROR, "Error return %d asserting "
+				     "join on node %d\n", status, node);
+
+				/* give us some time between errors... */
+				if (live)
+					msleep(DLM_DOMAIN_BACKOFF_MS);
+			}
+		} while (status && live);
+	}
+}
+
+struct domain_join_ctxt {
+	unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
+};
+
+static int dlm_should_restart_join(struct dlm_ctxt *dlm,
+				   struct domain_join_ctxt *ctxt,
+				   enum dlm_query_join_response_code response)
+{
+	int ret;
+
+	if (response == JOIN_DISALLOW) {
+		mlog(0, "Latest response of disallow -- should restart\n");
+		return 1;
+	}
+
+	spin_lock(&dlm->spinlock);
+	/* For now, we restart the process if the node maps have
+	 * changed at all */
+	ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
+		     sizeof(dlm->live_nodes_map));
+	spin_unlock(&dlm->spinlock);
+
+	if (ret)
+		mlog(0, "Node maps changed -- should restart\n");
+
+	return ret;
+}
+
+static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
+{
+	int status = 0, tmpstat, node;
+	struct domain_join_ctxt *ctxt;
+	enum dlm_query_join_response_code response = JOIN_DISALLOW;
+
+	mlog(0, "%p", dlm);
+
+	ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
+	if (!ctxt) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/* group sem locking should work for us here -- we're already
+	 * registered for heartbeat events so filling this should be
+	 * atomic wrt getting those handlers called. */
+	o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
+
+	spin_lock(&dlm->spinlock);
+	memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
+
+	__dlm_set_joining_node(dlm, dlm->node_num);
+
+	spin_unlock(&dlm->spinlock);
+
+	node = -1;
+	while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
+				     node + 1)) < O2NM_MAX_NODES) {
+		if (node == dlm->node_num)
+			continue;
+
+		status = dlm_request_join(dlm, node, &response);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		/* Ok, either we got a response or the node doesn't have a
+		 * dlm up. */
+		if (response == JOIN_OK)
+			set_bit(node, ctxt->yes_resp_map);
+
+		if (dlm_should_restart_join(dlm, ctxt, response)) {
+			status = -EAGAIN;
+			goto bail;
+		}
+	}
+
+	mlog(0, "Yay, done querying nodes!\n");
+
+	/* Yay, everyone agree's we can join the domain. My domain is
+	 * comprised of all nodes who were put in the
+	 * yes_resp_map. Copy that into our domain map and send a join
+	 * assert message to clean up everyone elses state. */
+	spin_lock(&dlm->spinlock);
+	memcpy(dlm->domain_map, ctxt->yes_resp_map,
+	       sizeof(ctxt->yes_resp_map));
+	set_bit(dlm->node_num, dlm->domain_map);
+	spin_unlock(&dlm->spinlock);
+
+	/* Support for global heartbeat and node info was added in 1.1 */
+	if (dlm->dlm_locking_proto.pv_major > 1 ||
+	    dlm->dlm_locking_proto.pv_minor > 0) {
+		status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
+		if (status) {
+			mlog_errno(status);
+			goto bail;
+		}
+		status = dlm_send_regions(dlm, ctxt->yes_resp_map);
+		if (status) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
+
+	/* Joined state *must* be set before the joining node
+	 * information, otherwise the query_join handler may read no
+	 * current joiner but a state of NEW and tell joining nodes
+	 * we're not in the domain. */
+	spin_lock(&dlm_domain_lock);
+	dlm->dlm_state = DLM_CTXT_JOINED;
+	dlm->num_joins++;
+	spin_unlock(&dlm_domain_lock);
+
+bail:
+	spin_lock(&dlm->spinlock);
+	__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
+	if (!status) {
+		printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name);
+		__dlm_print_nodes(dlm);
+	}
+	spin_unlock(&dlm->spinlock);
+
+	if (ctxt) {
+		/* Do we need to send a cancel message to any nodes? */
+		if (status < 0) {
+			tmpstat = dlm_send_join_cancels(dlm,
+							ctxt->yes_resp_map,
+							sizeof(ctxt->yes_resp_map));
+			if (tmpstat < 0)
+				mlog_errno(tmpstat);
+		}
+		kfree(ctxt);
+	}
+
+	mlog(0, "returning %d\n", status);
+	return status;
+}
+
+static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
+{
+	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
+	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
+	o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
+}
+
+static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
+{
+	int status;
+
+	mlog(0, "registering handlers.\n");
+
+	o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
+			    dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
+	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
+	if (status)
+		goto bail;
+
+	o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
+			    dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
+	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
+					sizeof(struct dlm_master_request),
+					dlm_master_request_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
+					sizeof(struct dlm_assert_master),
+					dlm_assert_master_handler,
+					dlm, dlm_assert_master_post_handler,
+					&dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
+					sizeof(struct dlm_create_lock),
+					dlm_create_lock_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
+					DLM_CONVERT_LOCK_MAX_LEN,
+					dlm_convert_lock_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
+					DLM_UNLOCK_LOCK_MAX_LEN,
+					dlm_unlock_lock_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
+					DLM_PROXY_AST_MAX_LEN,
+					dlm_proxy_ast_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
+					sizeof(struct dlm_exit_domain),
+					dlm_exit_domain_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
+					sizeof(struct dlm_deref_lockres),
+					dlm_deref_lockres_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
+					sizeof(struct dlm_migrate_request),
+					dlm_migrate_request_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
+					DLM_MIG_LOCKRES_MAX_LEN,
+					dlm_mig_lockres_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
+					sizeof(struct dlm_master_requery),
+					dlm_master_requery_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
+					sizeof(struct dlm_lock_request),
+					dlm_request_all_locks_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
+					sizeof(struct dlm_reco_data_done),
+					dlm_reco_data_done_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
+					sizeof(struct dlm_begin_reco),
+					dlm_begin_reco_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
+					sizeof(struct dlm_finalize_reco),
+					dlm_finalize_reco_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key,
+					sizeof(struct dlm_exit_domain),
+					dlm_begin_exit_domain_handler,
+					dlm, NULL, &dlm->dlm_domain_handlers);
+	if (status)
+		goto bail;
+
+bail:
+	if (status)
+		dlm_unregister_domain_handlers(dlm);
+
+	return status;
+}
+
+static int dlm_join_domain(struct dlm_ctxt *dlm)
+{
+	int status;
+	unsigned int backoff;
+	unsigned int total_backoff = 0;
+
+	BUG_ON(!dlm);
+
+	mlog(0, "Join domain %s\n", dlm->name);
+
+	status = dlm_register_domain_handlers(dlm);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = dlm_launch_thread(dlm);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = dlm_launch_recovery_thread(dlm);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = dlm_debug_init(dlm);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
+	if (!dlm->dlm_worker) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	do {
+		status = dlm_try_to_join_domain(dlm);
+
+		/* If we're racing another node to the join, then we
+		 * need to back off temporarily and let them
+		 * complete. */
+#define	DLM_JOIN_TIMEOUT_MSECS	90000
+		if (status == -EAGAIN) {
+			if (signal_pending(current)) {
+				status = -ERESTARTSYS;
+				goto bail;
+			}
+
+			if (total_backoff > DLM_JOIN_TIMEOUT_MSECS) {
+				status = -ERESTARTSYS;
+				mlog(ML_NOTICE, "Timed out joining dlm domain "
+				     "%s after %u msecs\n", dlm->name,
+				     total_backoff);
+				goto bail;
+			}
+
+			/*
+			 * <chip> After you!
+			 * <dale> No, after you!
+			 * <chip> I insist!
+			 * <dale> But you first!
+			 * ...
+			 */
+			backoff = (unsigned int)(jiffies & 0x3);
+			backoff *= DLM_DOMAIN_BACKOFF_MS;
+			total_backoff += backoff;
+			mlog(0, "backoff %d\n", backoff);
+			msleep(backoff);
+		}
+	} while (status == -EAGAIN);
+
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = 0;
+bail:
+	wake_up(&dlm_domain_events);
+
+	if (status) {
+		dlm_unregister_domain_handlers(dlm);
+		dlm_debug_shutdown(dlm);
+		dlm_complete_thread(dlm);
+		dlm_complete_recovery_thread(dlm);
+		dlm_destroy_dlm_worker(dlm);
+	}
+
+	return status;
+}
+
+static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
+				u32 key)
+{
+	int i;
+	int ret;
+	struct dlm_ctxt *dlm = NULL;
+
+	dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
+	if (!dlm) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	dlm->name = kstrdup(domain, GFP_KERNEL);
+	if (dlm->name == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
+	if (!dlm->lockres_hash) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	for (i = 0; i < DLM_HASH_BUCKETS; i++)
+		INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
+
+	dlm->master_hash = (struct hlist_head **)
+				dlm_alloc_pagevec(DLM_HASH_PAGES);
+	if (!dlm->master_hash) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	for (i = 0; i < DLM_HASH_BUCKETS; i++)
+		INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
+
+	dlm->key = key;
+	dlm->node_num = o2nm_this_node();
+
+	ret = dlm_create_debugfs_subroot(dlm);
+	if (ret < 0)
+		goto leave;
+
+	spin_lock_init(&dlm->spinlock);
+	spin_lock_init(&dlm->master_lock);
+	spin_lock_init(&dlm->ast_lock);
+	spin_lock_init(&dlm->track_lock);
+	INIT_LIST_HEAD(&dlm->list);
+	INIT_LIST_HEAD(&dlm->dirty_list);
+	INIT_LIST_HEAD(&dlm->reco.resources);
+	INIT_LIST_HEAD(&dlm->reco.node_data);
+	INIT_LIST_HEAD(&dlm->purge_list);
+	INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
+	INIT_LIST_HEAD(&dlm->tracking_list);
+	dlm->reco.state = 0;
+
+	INIT_LIST_HEAD(&dlm->pending_asts);
+	INIT_LIST_HEAD(&dlm->pending_basts);
+
+	mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
+		  dlm->recovery_map, &(dlm->recovery_map[0]));
+
+	memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
+	memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
+	memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
+
+	dlm->dlm_thread_task = NULL;
+	dlm->dlm_reco_thread_task = NULL;
+	dlm->dlm_worker = NULL;
+	init_waitqueue_head(&dlm->dlm_thread_wq);
+	init_waitqueue_head(&dlm->dlm_reco_thread_wq);
+	init_waitqueue_head(&dlm->reco.event);
+	init_waitqueue_head(&dlm->ast_wq);
+	init_waitqueue_head(&dlm->migration_wq);
+	INIT_LIST_HEAD(&dlm->mle_hb_events);
+
+	dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
+	init_waitqueue_head(&dlm->dlm_join_events);
+
+	dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
+	dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
+
+	atomic_set(&dlm->res_tot_count, 0);
+	atomic_set(&dlm->res_cur_count, 0);
+	for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
+		atomic_set(&dlm->mle_tot_count[i], 0);
+		atomic_set(&dlm->mle_cur_count[i], 0);
+	}
+
+	spin_lock_init(&dlm->work_lock);
+	INIT_LIST_HEAD(&dlm->work_list);
+	INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
+
+	kref_init(&dlm->dlm_refs);
+	dlm->dlm_state = DLM_CTXT_NEW;
+
+	INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
+
+	mlog(0, "context init: refcount %u\n",
+		  atomic_read(&dlm->dlm_refs.refcount));
+
+leave:
+	if (ret < 0 && dlm) {
+		if (dlm->master_hash)
+			dlm_free_pagevec((void **)dlm->master_hash,
+					DLM_HASH_PAGES);
+
+		if (dlm->lockres_hash)
+			dlm_free_pagevec((void **)dlm->lockres_hash,
+					DLM_HASH_PAGES);
+
+		kfree(dlm->name);
+		kfree(dlm);
+		dlm = NULL;
+	}
+	return dlm;
+}
+
+/*
+ * Compare a requested locking protocol version against the current one.
+ *
+ * If the major numbers are different, they are incompatible.
+ * If the current minor is greater than the request, they are incompatible.
+ * If the current minor is less than or equal to the request, they are
+ * compatible, and the requester should run at the current minor version.
+ */
+static int dlm_protocol_compare(struct dlm_protocol_version *existing,
+				struct dlm_protocol_version *request)
+{
+	if (existing->pv_major != request->pv_major)
+		return 1;
+
+	if (existing->pv_minor > request->pv_minor)
+		return 1;
+
+	if (existing->pv_minor < request->pv_minor)
+		request->pv_minor = existing->pv_minor;
+
+	return 0;
+}
+
+/*
+ * dlm_register_domain: one-time setup per "domain".
+ *
+ * The filesystem passes in the requested locking version via proto.
+ * If registration was successful, proto will contain the negotiated
+ * locking protocol.
+ */
+struct dlm_ctxt * dlm_register_domain(const char *domain,
+			       u32 key,
+			       struct dlm_protocol_version *fs_proto)
+{
+	int ret;
+	struct dlm_ctxt *dlm = NULL;
+	struct dlm_ctxt *new_ctxt = NULL;
+
+	if (strlen(domain) >= O2NM_MAX_NAME_LEN) {
+		ret = -ENAMETOOLONG;
+		mlog(ML_ERROR, "domain name length too long\n");
+		goto leave;
+	}
+
+	mlog(0, "register called for domain \"%s\"\n", domain);
+
+retry:
+	dlm = NULL;
+	if (signal_pending(current)) {
+		ret = -ERESTARTSYS;
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	spin_lock(&dlm_domain_lock);
+
+	dlm = __dlm_lookup_domain(domain);
+	if (dlm) {
+		if (dlm->dlm_state != DLM_CTXT_JOINED) {
+			spin_unlock(&dlm_domain_lock);
+
+			mlog(0, "This ctxt is not joined yet!\n");
+			wait_event_interruptible(dlm_domain_events,
+						 dlm_wait_on_domain_helper(
+							 domain));
+			goto retry;
+		}
+
+		if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
+			spin_unlock(&dlm_domain_lock);
+			mlog(ML_ERROR,
+			     "Requested locking protocol version is not "
+			     "compatible with already registered domain "
+			     "\"%s\"\n", domain);
+			ret = -EPROTO;
+			goto leave;
+		}
+
+		__dlm_get(dlm);
+		dlm->num_joins++;
+
+		spin_unlock(&dlm_domain_lock);
+
+		ret = 0;
+		goto leave;
+	}
+
+	/* doesn't exist */
+	if (!new_ctxt) {
+		spin_unlock(&dlm_domain_lock);
+
+		new_ctxt = dlm_alloc_ctxt(domain, key);
+		if (new_ctxt)
+			goto retry;
+
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	/* a little variable switch-a-roo here... */
+	dlm = new_ctxt;
+	new_ctxt = NULL;
+
+	/* add the new domain */
+	list_add_tail(&dlm->list, &dlm_domains);
+	spin_unlock(&dlm_domain_lock);
+
+	/*
+	 * Pass the locking protocol version into the join.  If the join
+	 * succeeds, it will have the negotiated protocol set.
+	 */
+	dlm->dlm_locking_proto = dlm_protocol;
+	dlm->fs_locking_proto = *fs_proto;
+
+	ret = dlm_join_domain(dlm);
+	if (ret) {
+		mlog_errno(ret);
+		dlm_put(dlm);
+		goto leave;
+	}
+
+	/* Tell the caller what locking protocol we negotiated */
+	*fs_proto = dlm->fs_locking_proto;
+
+	ret = 0;
+leave:
+	if (new_ctxt)
+		dlm_free_ctxt_mem(new_ctxt);
+
+	if (ret < 0)
+		dlm = ERR_PTR(ret);
+
+	return dlm;
+}
+EXPORT_SYMBOL_GPL(dlm_register_domain);
+
+static LIST_HEAD(dlm_join_handlers);
+
+static void dlm_unregister_net_handlers(void)
+{
+	o2net_unregister_handler_list(&dlm_join_handlers);
+}
+
+static int dlm_register_net_handlers(void)
+{
+	int status = 0;
+
+	status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
+					sizeof(struct dlm_query_join_request),
+					dlm_query_join_handler,
+					NULL, NULL, &dlm_join_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
+					sizeof(struct dlm_assert_joined),
+					dlm_assert_joined_handler,
+					NULL, NULL, &dlm_join_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
+					sizeof(struct dlm_cancel_join),
+					dlm_cancel_join_handler,
+					NULL, NULL, &dlm_join_handlers);
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY,
+					sizeof(struct dlm_query_region),
+					dlm_query_region_handler,
+					NULL, NULL, &dlm_join_handlers);
+
+	if (status)
+		goto bail;
+
+	status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
+					sizeof(struct dlm_query_nodeinfo),
+					dlm_query_nodeinfo_handler,
+					NULL, NULL, &dlm_join_handlers);
+bail:
+	if (status < 0)
+		dlm_unregister_net_handlers();
+
+	return status;
+}
+
+/* Domain eviction callback handling.
+ *
+ * The file system requires notification of node death *before* the
+ * dlm completes it's recovery work, otherwise it may be able to
+ * acquire locks on resources requiring recovery. Since the dlm can
+ * evict a node from it's domain *before* heartbeat fires, a similar
+ * mechanism is required. */
+
+/* Eviction is not expected to happen often, so a per-domain lock is
+ * not necessary. Eviction callbacks are allowed to sleep for short
+ * periods of time. */
+static DECLARE_RWSEM(dlm_callback_sem);
+
+void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
+					int node_num)
+{
+	struct dlm_eviction_cb *cb;
+
+	down_read(&dlm_callback_sem);
+	list_for_each_entry(cb, &dlm->dlm_eviction_callbacks, ec_item) {
+		cb->ec_func(node_num, cb->ec_data);
+	}
+	up_read(&dlm_callback_sem);
+}
+
+void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
+			   dlm_eviction_func *f,
+			   void *data)
+{
+	INIT_LIST_HEAD(&cb->ec_item);
+	cb->ec_func = f;
+	cb->ec_data = data;
+}
+EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
+
+void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
+			      struct dlm_eviction_cb *cb)
+{
+	down_write(&dlm_callback_sem);
+	list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
+	up_write(&dlm_callback_sem);
+}
+EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
+
+void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
+{
+	down_write(&dlm_callback_sem);
+	list_del_init(&cb->ec_item);
+	up_write(&dlm_callback_sem);
+}
+EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
+
+static int __init dlm_init(void)
+{
+	int status;
+
+	status = dlm_init_mle_cache();
+	if (status) {
+		mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
+		goto error;
+	}
+
+	status = dlm_init_master_caches();
+	if (status) {
+		mlog(ML_ERROR, "Could not create o2dlm_lockres and "
+		     "o2dlm_lockname slabcaches\n");
+		goto error;
+	}
+
+	status = dlm_init_lock_cache();
+	if (status) {
+		mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
+		goto error;
+	}
+
+	status = dlm_register_net_handlers();
+	if (status) {
+		mlog(ML_ERROR, "Unable to register network handlers\n");
+		goto error;
+	}
+
+	status = dlm_create_debugfs_root();
+	if (status)
+		goto error;
+
+	return 0;
+error:
+	dlm_unregister_net_handlers();
+	dlm_destroy_lock_cache();
+	dlm_destroy_master_caches();
+	dlm_destroy_mle_cache();
+	return -1;
+}
+
+static void __exit dlm_exit (void)
+{
+	dlm_destroy_debugfs_root();
+	dlm_unregister_net_handlers();
+	dlm_destroy_lock_cache();
+	dlm_destroy_master_caches();
+	dlm_destroy_mle_cache();
+}
+
+MODULE_AUTHOR("Oracle");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("OCFS2 Distributed Lock Management");
+
+module_init(dlm_init);
+module_exit(dlm_exit);
diff --git a/fs/ocfs2/dlm/dlmdomain.h b/fs/ocfs2/dlm/dlmdomain.h
new file mode 100644
index 000000000..fd6122a38
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmdomain.h
@@ -0,0 +1,35 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmdomain.h
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+#ifndef DLMDOMAIN_H
+#define DLMDOMAIN_H
+
+extern spinlock_t dlm_domain_lock;
+extern struct list_head dlm_domains;
+
+int dlm_shutting_down(struct dlm_ctxt *dlm);
+void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
+					int node_num);
+
+#endif
diff --git a/fs/ocfs2/dlm/dlmlock.c b/fs/ocfs2/dlm/dlmlock.c
new file mode 100644
index 000000000..66c2a491f
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmlock.c
@@ -0,0 +1,761 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmlock.c
+ *
+ * underlying calls for lock creation
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/sysctl.h>
+#include <linux/random.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/spinlock.h>
+#include <linux/delay.h>
+
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+
+#include "dlmconvert.h"
+
+#define MLOG_MASK_PREFIX ML_DLM
+#include "cluster/masklog.h"
+
+static struct kmem_cache *dlm_lock_cache;
+
+static DEFINE_SPINLOCK(dlm_cookie_lock);
+static u64 dlm_next_cookie = 1;
+
+static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
+					       struct dlm_lock_resource *res,
+					       struct dlm_lock *lock, int flags);
+static void dlm_init_lock(struct dlm_lock *newlock, int type,
+			  u8 node, u64 cookie);
+static void dlm_lock_release(struct kref *kref);
+static void dlm_lock_detach_lockres(struct dlm_lock *lock);
+
+int dlm_init_lock_cache(void)
+{
+	dlm_lock_cache = kmem_cache_create("o2dlm_lock",
+					   sizeof(struct dlm_lock),
+					   0, SLAB_HWCACHE_ALIGN, NULL);
+	if (dlm_lock_cache == NULL)
+		return -ENOMEM;
+	return 0;
+}
+
+void dlm_destroy_lock_cache(void)
+{
+	if (dlm_lock_cache)
+		kmem_cache_destroy(dlm_lock_cache);
+}
+
+/* Tell us whether we can grant a new lock request.
+ * locking:
+ *   caller needs:  res->spinlock
+ *   taken:         none
+ *   held on exit:  none
+ * returns: 1 if the lock can be granted, 0 otherwise.
+ */
+static int dlm_can_grant_new_lock(struct dlm_lock_resource *res,
+				  struct dlm_lock *lock)
+{
+	struct dlm_lock *tmplock;
+
+	list_for_each_entry(tmplock, &res->granted, list) {
+		if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
+			return 0;
+	}
+
+	list_for_each_entry(tmplock, &res->converting, list) {
+		if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
+			return 0;
+		if (!dlm_lock_compatible(tmplock->ml.convert_type,
+					 lock->ml.type))
+			return 0;
+	}
+
+	return 1;
+}
+
+/* performs lock creation at the lockres master site
+ * locking:
+ *   caller needs:  none
+ *   taken:         takes and drops res->spinlock
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_NOTQUEUED
+ */
+static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
+				      struct dlm_lock_resource *res,
+				      struct dlm_lock *lock, int flags)
+{
+	int call_ast = 0, kick_thread = 0;
+	enum dlm_status status = DLM_NORMAL;
+
+	mlog(0, "type=%d\n", lock->ml.type);
+
+	spin_lock(&res->spinlock);
+	/* if called from dlm_create_lock_handler, need to
+	 * ensure it will not sleep in dlm_wait_on_lockres */
+	status = __dlm_lockres_state_to_status(res);
+	if (status != DLM_NORMAL &&
+	    lock->ml.node != dlm->node_num) {
+		/* erf.  state changed after lock was dropped. */
+		spin_unlock(&res->spinlock);
+		dlm_error(status);
+		return status;
+	}
+	__dlm_wait_on_lockres(res);
+	__dlm_lockres_reserve_ast(res);
+
+	if (dlm_can_grant_new_lock(res, lock)) {
+		mlog(0, "I can grant this lock right away\n");
+		/* got it right away */
+		lock->lksb->status = DLM_NORMAL;
+		status = DLM_NORMAL;
+		dlm_lock_get(lock);
+		list_add_tail(&lock->list, &res->granted);
+
+		/* for the recovery lock, we can't allow the ast
+		 * to be queued since the dlmthread is already
+		 * frozen.  but the recovery lock is always locked
+		 * with LKM_NOQUEUE so we do not need the ast in
+		 * this special case */
+		if (!dlm_is_recovery_lock(res->lockname.name,
+					  res->lockname.len)) {
+			kick_thread = 1;
+			call_ast = 1;
+		} else {
+			mlog(0, "%s: returning DLM_NORMAL to "
+			     "node %u for reco lock\n", dlm->name,
+			     lock->ml.node);
+		}
+	} else {
+		/* for NOQUEUE request, unless we get the
+		 * lock right away, return DLM_NOTQUEUED */
+		if (flags & LKM_NOQUEUE) {
+			status = DLM_NOTQUEUED;
+			if (dlm_is_recovery_lock(res->lockname.name,
+						 res->lockname.len)) {
+				mlog(0, "%s: returning NOTQUEUED to "
+				     "node %u for reco lock\n", dlm->name,
+				     lock->ml.node);
+			}
+		} else {
+			status = DLM_NORMAL;
+			dlm_lock_get(lock);
+			list_add_tail(&lock->list, &res->blocked);
+			kick_thread = 1;
+		}
+	}
+
+	spin_unlock(&res->spinlock);
+	wake_up(&res->wq);
+
+	/* either queue the ast or release it */
+	if (call_ast)
+		dlm_queue_ast(dlm, lock);
+	else
+		dlm_lockres_release_ast(dlm, res);
+
+	dlm_lockres_calc_usage(dlm, res);
+	if (kick_thread)
+		dlm_kick_thread(dlm, res);
+
+	return status;
+}
+
+void dlm_revert_pending_lock(struct dlm_lock_resource *res,
+			     struct dlm_lock *lock)
+{
+	/* remove from local queue if it failed */
+	list_del_init(&lock->list);
+	lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
+}
+
+
+/*
+ * locking:
+ *   caller needs:  none
+ *   taken:         takes and drops res->spinlock
+ *   held on exit:  none
+ * returns: DLM_DENIED, DLM_RECOVERING, or net status
+ */
+static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
+				      struct dlm_lock_resource *res,
+				      struct dlm_lock *lock, int flags)
+{
+	enum dlm_status status = DLM_DENIED;
+	int lockres_changed = 1;
+
+	mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n",
+	     lock->ml.type, res->lockname.len,
+	     res->lockname.name, flags);
+
+	/*
+	 * Wait if resource is getting recovered, remastered, etc.
+	 * If the resource was remastered and new owner is self, then exit.
+	 */
+	spin_lock(&res->spinlock);
+	__dlm_wait_on_lockres(res);
+	if (res->owner == dlm->node_num) {
+		spin_unlock(&res->spinlock);
+		return DLM_RECOVERING;
+	}
+	res->state |= DLM_LOCK_RES_IN_PROGRESS;
+
+	/* add lock to local (secondary) queue */
+	dlm_lock_get(lock);
+	list_add_tail(&lock->list, &res->blocked);
+	lock->lock_pending = 1;
+	spin_unlock(&res->spinlock);
+
+	/* spec seems to say that you will get DLM_NORMAL when the lock
+	 * has been queued, meaning we need to wait for a reply here. */
+	status = dlm_send_remote_lock_request(dlm, res, lock, flags);
+
+	spin_lock(&res->spinlock);
+	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
+	lock->lock_pending = 0;
+	if (status != DLM_NORMAL) {
+		if (status == DLM_RECOVERING &&
+		    dlm_is_recovery_lock(res->lockname.name,
+					 res->lockname.len)) {
+			/* recovery lock was mastered by dead node.
+			 * we need to have calc_usage shoot down this
+			 * lockres and completely remaster it. */
+			mlog(0, "%s: recovery lock was owned by "
+			     "dead node %u, remaster it now.\n",
+			     dlm->name, res->owner);
+		} else if (status != DLM_NOTQUEUED) {
+			/*
+			 * DO NOT call calc_usage, as this would unhash
+			 * the remote lockres before we ever get to use
+			 * it.  treat as if we never made any change to
+			 * the lockres.
+			 */
+			lockres_changed = 0;
+			dlm_error(status);
+		}
+		dlm_revert_pending_lock(res, lock);
+		dlm_lock_put(lock);
+	} else if (dlm_is_recovery_lock(res->lockname.name,
+					res->lockname.len)) {
+		/* special case for the $RECOVERY lock.
+		 * there will never be an AST delivered to put
+		 * this lock on the proper secondary queue
+		 * (granted), so do it manually. */
+		mlog(0, "%s: $RECOVERY lock for this node (%u) is "
+		     "mastered by %u; got lock, manually granting (no ast)\n",
+		     dlm->name, dlm->node_num, res->owner);
+		list_move_tail(&lock->list, &res->granted);
+	}
+	spin_unlock(&res->spinlock);
+
+	if (lockres_changed)
+		dlm_lockres_calc_usage(dlm, res);
+
+	wake_up(&res->wq);
+	return status;
+}
+
+
+/* for remote lock creation.
+ * locking:
+ *   caller needs:  none, but need res->state & DLM_LOCK_RES_IN_PROGRESS
+ *   taken:         none
+ *   held on exit:  none
+ * returns: DLM_NOLOCKMGR, or net status
+ */
+static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
+					       struct dlm_lock_resource *res,
+					       struct dlm_lock *lock, int flags)
+{
+	struct dlm_create_lock create;
+	int tmpret, status = 0;
+	enum dlm_status ret;
+
+	memset(&create, 0, sizeof(create));
+	create.node_idx = dlm->node_num;
+	create.requested_type = lock->ml.type;
+	create.cookie = lock->ml.cookie;
+	create.namelen = res->lockname.len;
+	create.flags = cpu_to_be32(flags);
+	memcpy(create.name, res->lockname.name, create.namelen);
+
+	tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
+				    sizeof(create), res->owner, &status);
+	if (tmpret >= 0) {
+		ret = status;
+		if (ret == DLM_REJECTED) {
+			mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer "
+			     "owned by node %u. That node is coming back up "
+			     "currently.\n", dlm->name, create.namelen,
+			     create.name, res->owner);
+			dlm_print_one_lock_resource(res);
+			BUG();
+		}
+	} else {
+		mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to "
+		     "node %u\n", dlm->name, create.namelen, create.name,
+		     tmpret, res->owner);
+		if (dlm_is_host_down(tmpret))
+			ret = DLM_RECOVERING;
+		else
+			ret = dlm_err_to_dlm_status(tmpret);
+	}
+
+	return ret;
+}
+
+void dlm_lock_get(struct dlm_lock *lock)
+{
+	kref_get(&lock->lock_refs);
+}
+
+void dlm_lock_put(struct dlm_lock *lock)
+{
+	kref_put(&lock->lock_refs, dlm_lock_release);
+}
+
+static void dlm_lock_release(struct kref *kref)
+{
+	struct dlm_lock *lock;
+
+	lock = container_of(kref, struct dlm_lock, lock_refs);
+
+	BUG_ON(!list_empty(&lock->list));
+	BUG_ON(!list_empty(&lock->ast_list));
+	BUG_ON(!list_empty(&lock->bast_list));
+	BUG_ON(lock->ast_pending);
+	BUG_ON(lock->bast_pending);
+
+	dlm_lock_detach_lockres(lock);
+
+	if (lock->lksb_kernel_allocated) {
+		mlog(0, "freeing kernel-allocated lksb\n");
+		kfree(lock->lksb);
+	}
+	kmem_cache_free(dlm_lock_cache, lock);
+}
+
+/* associate a lock with it's lockres, getting a ref on the lockres */
+void dlm_lock_attach_lockres(struct dlm_lock *lock,
+			     struct dlm_lock_resource *res)
+{
+	dlm_lockres_get(res);
+	lock->lockres = res;
+}
+
+/* drop ref on lockres, if there is still one associated with lock */
+static void dlm_lock_detach_lockres(struct dlm_lock *lock)
+{
+	struct dlm_lock_resource *res;
+
+	res = lock->lockres;
+	if (res) {
+		lock->lockres = NULL;
+		mlog(0, "removing lock's lockres reference\n");
+		dlm_lockres_put(res);
+	}
+}
+
+static void dlm_init_lock(struct dlm_lock *newlock, int type,
+			  u8 node, u64 cookie)
+{
+	INIT_LIST_HEAD(&newlock->list);
+	INIT_LIST_HEAD(&newlock->ast_list);
+	INIT_LIST_HEAD(&newlock->bast_list);
+	spin_lock_init(&newlock->spinlock);
+	newlock->ml.type = type;
+	newlock->ml.convert_type = LKM_IVMODE;
+	newlock->ml.highest_blocked = LKM_IVMODE;
+	newlock->ml.node = node;
+	newlock->ml.pad1 = 0;
+	newlock->ml.list = 0;
+	newlock->ml.flags = 0;
+	newlock->ast = NULL;
+	newlock->bast = NULL;
+	newlock->astdata = NULL;
+	newlock->ml.cookie = cpu_to_be64(cookie);
+	newlock->ast_pending = 0;
+	newlock->bast_pending = 0;
+	newlock->convert_pending = 0;
+	newlock->lock_pending = 0;
+	newlock->unlock_pending = 0;
+	newlock->cancel_pending = 0;
+	newlock->lksb_kernel_allocated = 0;
+
+	kref_init(&newlock->lock_refs);
+}
+
+struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
+			       struct dlm_lockstatus *lksb)
+{
+	struct dlm_lock *lock;
+	int kernel_allocated = 0;
+
+	lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
+	if (!lock)
+		return NULL;
+
+	if (!lksb) {
+		/* zero memory only if kernel-allocated */
+		lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
+		if (!lksb) {
+			kmem_cache_free(dlm_lock_cache, lock);
+			return NULL;
+		}
+		kernel_allocated = 1;
+	}
+
+	dlm_init_lock(lock, type, node, cookie);
+	if (kernel_allocated)
+		lock->lksb_kernel_allocated = 1;
+	lock->lksb = lksb;
+	lksb->lockid = lock;
+	return lock;
+}
+
+/* handler for lock creation net message
+ * locking:
+ *   caller needs:  none
+ *   taken:         takes and drops res->spinlock
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
+ */
+int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
+			    void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
+	struct dlm_lock_resource *res = NULL;
+	struct dlm_lock *newlock = NULL;
+	struct dlm_lockstatus *lksb = NULL;
+	enum dlm_status status = DLM_NORMAL;
+	char *name;
+	unsigned int namelen;
+
+	BUG_ON(!dlm);
+
+	if (!dlm_grab(dlm))
+		return DLM_REJECTED;
+
+	name = create->name;
+	namelen = create->namelen;
+	status = DLM_REJECTED;
+	if (!dlm_domain_fully_joined(dlm)) {
+		mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
+		     "sending a create_lock message for lock %.*s!\n",
+		     dlm->name, create->node_idx, namelen, name);
+		dlm_error(status);
+		goto leave;
+	}
+
+	status = DLM_IVBUFLEN;
+	if (namelen > DLM_LOCKID_NAME_MAX) {
+		dlm_error(status);
+		goto leave;
+	}
+
+	status = DLM_SYSERR;
+	newlock = dlm_new_lock(create->requested_type,
+			       create->node_idx,
+			       be64_to_cpu(create->cookie), NULL);
+	if (!newlock) {
+		dlm_error(status);
+		goto leave;
+	}
+
+	lksb = newlock->lksb;
+
+	if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
+		lksb->flags |= DLM_LKSB_GET_LVB;
+		mlog(0, "set DLM_LKSB_GET_LVB flag\n");
+	}
+
+	status = DLM_IVLOCKID;
+	res = dlm_lookup_lockres(dlm, name, namelen);
+	if (!res) {
+		dlm_error(status);
+		goto leave;
+	}
+
+	spin_lock(&res->spinlock);
+	status = __dlm_lockres_state_to_status(res);
+	spin_unlock(&res->spinlock);
+
+	if (status != DLM_NORMAL) {
+		mlog(0, "lockres recovering/migrating/in-progress\n");
+		goto leave;
+	}
+
+	dlm_lock_attach_lockres(newlock, res);
+
+	status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
+leave:
+	if (status != DLM_NORMAL)
+		if (newlock)
+			dlm_lock_put(newlock);
+
+	if (res)
+		dlm_lockres_put(res);
+
+	dlm_put(dlm);
+
+	return status;
+}
+
+
+/* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
+static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
+{
+	u64 tmpnode = node_num;
+
+	/* shift single byte of node num into top 8 bits */
+	tmpnode <<= 56;
+
+	spin_lock(&dlm_cookie_lock);
+	*cookie = (dlm_next_cookie | tmpnode);
+	if (++dlm_next_cookie & 0xff00000000000000ull) {
+		mlog(0, "This node's cookie will now wrap!\n");
+		dlm_next_cookie = 1;
+	}
+	spin_unlock(&dlm_cookie_lock);
+}
+
+enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
+			struct dlm_lockstatus *lksb, int flags,
+			const char *name, int namelen, dlm_astlockfunc_t *ast,
+			void *data, dlm_bastlockfunc_t *bast)
+{
+	enum dlm_status status;
+	struct dlm_lock_resource *res = NULL;
+	struct dlm_lock *lock = NULL;
+	int convert = 0, recovery = 0;
+
+	/* yes this function is a mess.
+	 * TODO: clean this up.  lots of common code in the
+	 *       lock and convert paths, especially in the retry blocks */
+	if (!lksb) {
+		dlm_error(DLM_BADARGS);
+		return DLM_BADARGS;
+	}
+
+	status = DLM_BADPARAM;
+	if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
+		dlm_error(status);
+		goto error;
+	}
+
+	if (flags & ~LKM_VALID_FLAGS) {
+		dlm_error(status);
+		goto error;
+	}
+
+	convert = (flags & LKM_CONVERT);
+	recovery = (flags & LKM_RECOVERY);
+
+	if (recovery &&
+	    (!dlm_is_recovery_lock(name, namelen) || convert) ) {
+		dlm_error(status);
+		goto error;
+	}
+	if (convert && (flags & LKM_LOCAL)) {
+		mlog(ML_ERROR, "strange LOCAL convert request!\n");
+		goto error;
+	}
+
+	if (convert) {
+		/* CONVERT request */
+
+		/* if converting, must pass in a valid dlm_lock */
+		lock = lksb->lockid;
+		if (!lock) {
+			mlog(ML_ERROR, "NULL lock pointer in convert "
+			     "request\n");
+			goto error;
+		}
+
+		res = lock->lockres;
+		if (!res) {
+			mlog(ML_ERROR, "NULL lockres pointer in convert "
+			     "request\n");
+			goto error;
+		}
+		dlm_lockres_get(res);
+
+		/* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
+	 	 * static after the original lock call.  convert requests will
+		 * ensure that everything is the same, or return DLM_BADARGS.
+	 	 * this means that DLM_DENIED_NOASTS will never be returned.
+	 	 */
+		if (lock->lksb != lksb || lock->ast != ast ||
+		    lock->bast != bast || lock->astdata != data) {
+			status = DLM_BADARGS;
+			mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
+			     "astdata=%p\n", lksb, ast, bast, data);
+			mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
+			     "astdata=%p\n", lock->lksb, lock->ast,
+			     lock->bast, lock->astdata);
+			goto error;
+		}
+retry_convert:
+		dlm_wait_for_recovery(dlm);
+
+		if (res->owner == dlm->node_num)
+			status = dlmconvert_master(dlm, res, lock, flags, mode);
+		else
+			status = dlmconvert_remote(dlm, res, lock, flags, mode);
+		if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
+		    status == DLM_FORWARD) {
+			/* for now, see how this works without sleeping
+			 * and just retry right away.  I suspect the reco
+			 * or migration will complete fast enough that
+			 * no waiting will be necessary */
+			mlog(0, "retrying convert with migration/recovery/"
+			     "in-progress\n");
+			msleep(100);
+			goto retry_convert;
+		}
+	} else {
+		u64 tmpcookie;
+
+		/* LOCK request */
+		status = DLM_BADARGS;
+		if (!name) {
+			dlm_error(status);
+			goto error;
+		}
+
+		status = DLM_IVBUFLEN;
+		if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
+			dlm_error(status);
+			goto error;
+		}
+
+		dlm_get_next_cookie(dlm->node_num, &tmpcookie);
+		lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
+		if (!lock) {
+			dlm_error(status);
+			goto error;
+		}
+
+		if (!recovery)
+			dlm_wait_for_recovery(dlm);
+
+		/* find or create the lock resource */
+		res = dlm_get_lock_resource(dlm, name, namelen, flags);
+		if (!res) {
+			status = DLM_IVLOCKID;
+			dlm_error(status);
+			goto error;
+		}
+
+		mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
+		mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
+
+		dlm_lock_attach_lockres(lock, res);
+		lock->ast = ast;
+		lock->bast = bast;
+		lock->astdata = data;
+
+retry_lock:
+		if (flags & LKM_VALBLK) {
+			mlog(0, "LKM_VALBLK passed by caller\n");
+
+			/* LVB requests for non PR, PW or EX locks are
+			 * ignored. */
+			if (mode < LKM_PRMODE)
+				flags &= ~LKM_VALBLK;
+			else {
+				flags |= LKM_GET_LVB;
+				lock->lksb->flags |= DLM_LKSB_GET_LVB;
+			}
+		}
+
+		if (res->owner == dlm->node_num)
+			status = dlmlock_master(dlm, res, lock, flags);
+		else
+			status = dlmlock_remote(dlm, res, lock, flags);
+
+		if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
+		    status == DLM_FORWARD) {
+			msleep(100);
+			if (recovery) {
+				if (status != DLM_RECOVERING)
+					goto retry_lock;
+				/* wait to see the node go down, then
+				 * drop down and allow the lockres to
+				 * get cleaned up.  need to remaster. */
+				dlm_wait_for_node_death(dlm, res->owner,
+						DLM_NODE_DEATH_WAIT_MAX);
+			} else {
+				dlm_wait_for_recovery(dlm);
+				goto retry_lock;
+			}
+		}
+
+		/* Inflight taken in dlm_get_lock_resource() is dropped here */
+		spin_lock(&res->spinlock);
+		dlm_lockres_drop_inflight_ref(dlm, res);
+		spin_unlock(&res->spinlock);
+
+		dlm_lockres_calc_usage(dlm, res);
+		dlm_kick_thread(dlm, res);
+
+		if (status != DLM_NORMAL) {
+			lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
+			if (status != DLM_NOTQUEUED)
+				dlm_error(status);
+			goto error;
+		}
+	}
+
+error:
+	if (status != DLM_NORMAL) {
+		if (lock && !convert)
+			dlm_lock_put(lock);
+		// this is kind of unnecessary
+		lksb->status = status;
+	}
+
+	/* put lockres ref from the convert path
+	 * or from dlm_get_lock_resource */
+	if (res)
+		dlm_lockres_put(res);
+
+	return status;
+}
+EXPORT_SYMBOL_GPL(dlmlock);
diff --git a/fs/ocfs2/dlm/dlmmaster.c b/fs/ocfs2/dlm/dlmmaster.c
new file mode 100644
index 000000000..fdf4b41d0
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmmaster.c
@@ -0,0 +1,3487 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmmod.c
+ *
+ * standalone DLM module
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/sysctl.h>
+#include <linux/random.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/spinlock.h>
+#include <linux/delay.h>
+
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+#include "dlmdomain.h"
+#include "dlmdebug.h"
+
+#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_MASTER)
+#include "cluster/masklog.h"
+
+static void dlm_mle_node_down(struct dlm_ctxt *dlm,
+			      struct dlm_master_list_entry *mle,
+			      struct o2nm_node *node,
+			      int idx);
+static void dlm_mle_node_up(struct dlm_ctxt *dlm,
+			    struct dlm_master_list_entry *mle,
+			    struct o2nm_node *node,
+			    int idx);
+
+static void dlm_assert_master_worker(struct dlm_work_item *item, void *data);
+static int dlm_do_assert_master(struct dlm_ctxt *dlm,
+				struct dlm_lock_resource *res,
+				void *nodemap, u32 flags);
+static void dlm_deref_lockres_worker(struct dlm_work_item *item, void *data);
+
+static inline int dlm_mle_equal(struct dlm_ctxt *dlm,
+				struct dlm_master_list_entry *mle,
+				const char *name,
+				unsigned int namelen)
+{
+	if (dlm != mle->dlm)
+		return 0;
+
+	if (namelen != mle->mnamelen ||
+	    memcmp(name, mle->mname, namelen) != 0)
+		return 0;
+
+	return 1;
+}
+
+static struct kmem_cache *dlm_lockres_cache;
+static struct kmem_cache *dlm_lockname_cache;
+static struct kmem_cache *dlm_mle_cache;
+
+static void dlm_mle_release(struct kref *kref);
+static void dlm_init_mle(struct dlm_master_list_entry *mle,
+			enum dlm_mle_type type,
+			struct dlm_ctxt *dlm,
+			struct dlm_lock_resource *res,
+			const char *name,
+			unsigned int namelen);
+static void dlm_put_mle(struct dlm_master_list_entry *mle);
+static void __dlm_put_mle(struct dlm_master_list_entry *mle);
+static int dlm_find_mle(struct dlm_ctxt *dlm,
+			struct dlm_master_list_entry **mle,
+			char *name, unsigned int namelen);
+
+static int dlm_do_master_request(struct dlm_lock_resource *res,
+				 struct dlm_master_list_entry *mle, int to);
+
+
+static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
+				     struct dlm_lock_resource *res,
+				     struct dlm_master_list_entry *mle,
+				     int *blocked);
+static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
+				    struct dlm_lock_resource *res,
+				    struct dlm_master_list_entry *mle,
+				    int blocked);
+static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
+				 struct dlm_lock_resource *res,
+				 struct dlm_master_list_entry *mle,
+				 struct dlm_master_list_entry **oldmle,
+				 const char *name, unsigned int namelen,
+				 u8 new_master, u8 master);
+
+static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
+				    struct dlm_lock_resource *res);
+static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
+				      struct dlm_lock_resource *res);
+static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
+				       struct dlm_lock_resource *res,
+				       u8 target);
+static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
+				       struct dlm_lock_resource *res);
+
+
+int dlm_is_host_down(int errno)
+{
+	switch (errno) {
+		case -EBADF:
+		case -ECONNREFUSED:
+		case -ENOTCONN:
+		case -ECONNRESET:
+		case -EPIPE:
+		case -EHOSTDOWN:
+		case -EHOSTUNREACH:
+		case -ETIMEDOUT:
+		case -ECONNABORTED:
+		case -ENETDOWN:
+		case -ENETUNREACH:
+		case -ENETRESET:
+		case -ESHUTDOWN:
+		case -ENOPROTOOPT:
+		case -EINVAL:   /* if returned from our tcp code,
+				   this means there is no socket */
+			return 1;
+	}
+	return 0;
+}
+
+
+/*
+ * MASTER LIST FUNCTIONS
+ */
+
+
+/*
+ * regarding master list entries and heartbeat callbacks:
+ *
+ * in order to avoid sleeping and allocation that occurs in
+ * heartbeat, master list entries are simply attached to the
+ * dlm's established heartbeat callbacks.  the mle is attached
+ * when it is created, and since the dlm->spinlock is held at
+ * that time, any heartbeat event will be properly discovered
+ * by the mle.  the mle needs to be detached from the
+ * dlm->mle_hb_events list as soon as heartbeat events are no
+ * longer useful to the mle, and before the mle is freed.
+ *
+ * as a general rule, heartbeat events are no longer needed by
+ * the mle once an "answer" regarding the lock master has been
+ * received.
+ */
+static inline void __dlm_mle_attach_hb_events(struct dlm_ctxt *dlm,
+					      struct dlm_master_list_entry *mle)
+{
+	assert_spin_locked(&dlm->spinlock);
+
+	list_add_tail(&mle->hb_events, &dlm->mle_hb_events);
+}
+
+
+static inline void __dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
+					      struct dlm_master_list_entry *mle)
+{
+	if (!list_empty(&mle->hb_events))
+		list_del_init(&mle->hb_events);
+}
+
+
+static inline void dlm_mle_detach_hb_events(struct dlm_ctxt *dlm,
+					    struct dlm_master_list_entry *mle)
+{
+	spin_lock(&dlm->spinlock);
+	__dlm_mle_detach_hb_events(dlm, mle);
+	spin_unlock(&dlm->spinlock);
+}
+
+static void dlm_get_mle_inuse(struct dlm_master_list_entry *mle)
+{
+	struct dlm_ctxt *dlm;
+	dlm = mle->dlm;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&dlm->master_lock);
+	mle->inuse++;
+	kref_get(&mle->mle_refs);
+}
+
+static void dlm_put_mle_inuse(struct dlm_master_list_entry *mle)
+{
+	struct dlm_ctxt *dlm;
+	dlm = mle->dlm;
+
+	spin_lock(&dlm->spinlock);
+	spin_lock(&dlm->master_lock);
+	mle->inuse--;
+	__dlm_put_mle(mle);
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+
+}
+
+/* remove from list and free */
+static void __dlm_put_mle(struct dlm_master_list_entry *mle)
+{
+	struct dlm_ctxt *dlm;
+	dlm = mle->dlm;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&dlm->master_lock);
+	if (!atomic_read(&mle->mle_refs.refcount)) {
+		/* this may or may not crash, but who cares.
+		 * it's a BUG. */
+		mlog(ML_ERROR, "bad mle: %p\n", mle);
+		dlm_print_one_mle(mle);
+		BUG();
+	} else
+		kref_put(&mle->mle_refs, dlm_mle_release);
+}
+
+
+/* must not have any spinlocks coming in */
+static void dlm_put_mle(struct dlm_master_list_entry *mle)
+{
+	struct dlm_ctxt *dlm;
+	dlm = mle->dlm;
+
+	spin_lock(&dlm->spinlock);
+	spin_lock(&dlm->master_lock);
+	__dlm_put_mle(mle);
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+}
+
+static inline void dlm_get_mle(struct dlm_master_list_entry *mle)
+{
+	kref_get(&mle->mle_refs);
+}
+
+static void dlm_init_mle(struct dlm_master_list_entry *mle,
+			enum dlm_mle_type type,
+			struct dlm_ctxt *dlm,
+			struct dlm_lock_resource *res,
+			const char *name,
+			unsigned int namelen)
+{
+	assert_spin_locked(&dlm->spinlock);
+
+	mle->dlm = dlm;
+	mle->type = type;
+	INIT_HLIST_NODE(&mle->master_hash_node);
+	INIT_LIST_HEAD(&mle->hb_events);
+	memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
+	spin_lock_init(&mle->spinlock);
+	init_waitqueue_head(&mle->wq);
+	atomic_set(&mle->woken, 0);
+	kref_init(&mle->mle_refs);
+	memset(mle->response_map, 0, sizeof(mle->response_map));
+	mle->master = O2NM_MAX_NODES;
+	mle->new_master = O2NM_MAX_NODES;
+	mle->inuse = 0;
+
+	BUG_ON(mle->type != DLM_MLE_BLOCK &&
+	       mle->type != DLM_MLE_MASTER &&
+	       mle->type != DLM_MLE_MIGRATION);
+
+	if (mle->type == DLM_MLE_MASTER) {
+		BUG_ON(!res);
+		mle->mleres = res;
+		memcpy(mle->mname, res->lockname.name, res->lockname.len);
+		mle->mnamelen = res->lockname.len;
+		mle->mnamehash = res->lockname.hash;
+	} else {
+		BUG_ON(!name);
+		mle->mleres = NULL;
+		memcpy(mle->mname, name, namelen);
+		mle->mnamelen = namelen;
+		mle->mnamehash = dlm_lockid_hash(name, namelen);
+	}
+
+	atomic_inc(&dlm->mle_tot_count[mle->type]);
+	atomic_inc(&dlm->mle_cur_count[mle->type]);
+
+	/* copy off the node_map and register hb callbacks on our copy */
+	memcpy(mle->node_map, dlm->domain_map, sizeof(mle->node_map));
+	memcpy(mle->vote_map, dlm->domain_map, sizeof(mle->vote_map));
+	clear_bit(dlm->node_num, mle->vote_map);
+	clear_bit(dlm->node_num, mle->node_map);
+
+	/* attach the mle to the domain node up/down events */
+	__dlm_mle_attach_hb_events(dlm, mle);
+}
+
+void __dlm_unlink_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle)
+{
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&dlm->master_lock);
+
+	if (!hlist_unhashed(&mle->master_hash_node))
+		hlist_del_init(&mle->master_hash_node);
+}
+
+void __dlm_insert_mle(struct dlm_ctxt *dlm, struct dlm_master_list_entry *mle)
+{
+	struct hlist_head *bucket;
+
+	assert_spin_locked(&dlm->master_lock);
+
+	bucket = dlm_master_hash(dlm, mle->mnamehash);
+	hlist_add_head(&mle->master_hash_node, bucket);
+}
+
+/* returns 1 if found, 0 if not */
+static int dlm_find_mle(struct dlm_ctxt *dlm,
+			struct dlm_master_list_entry **mle,
+			char *name, unsigned int namelen)
+{
+	struct dlm_master_list_entry *tmpmle;
+	struct hlist_head *bucket;
+	unsigned int hash;
+
+	assert_spin_locked(&dlm->master_lock);
+
+	hash = dlm_lockid_hash(name, namelen);
+	bucket = dlm_master_hash(dlm, hash);
+	hlist_for_each_entry(tmpmle, bucket, master_hash_node) {
+		if (!dlm_mle_equal(dlm, tmpmle, name, namelen))
+			continue;
+		dlm_get_mle(tmpmle);
+		*mle = tmpmle;
+		return 1;
+	}
+	return 0;
+}
+
+void dlm_hb_event_notify_attached(struct dlm_ctxt *dlm, int idx, int node_up)
+{
+	struct dlm_master_list_entry *mle;
+
+	assert_spin_locked(&dlm->spinlock);
+
+	list_for_each_entry(mle, &dlm->mle_hb_events, hb_events) {
+		if (node_up)
+			dlm_mle_node_up(dlm, mle, NULL, idx);
+		else
+			dlm_mle_node_down(dlm, mle, NULL, idx);
+	}
+}
+
+static void dlm_mle_node_down(struct dlm_ctxt *dlm,
+			      struct dlm_master_list_entry *mle,
+			      struct o2nm_node *node, int idx)
+{
+	spin_lock(&mle->spinlock);
+
+	if (!test_bit(idx, mle->node_map))
+		mlog(0, "node %u already removed from nodemap!\n", idx);
+	else
+		clear_bit(idx, mle->node_map);
+
+	spin_unlock(&mle->spinlock);
+}
+
+static void dlm_mle_node_up(struct dlm_ctxt *dlm,
+			    struct dlm_master_list_entry *mle,
+			    struct o2nm_node *node, int idx)
+{
+	spin_lock(&mle->spinlock);
+
+	if (test_bit(idx, mle->node_map))
+		mlog(0, "node %u already in node map!\n", idx);
+	else
+		set_bit(idx, mle->node_map);
+
+	spin_unlock(&mle->spinlock);
+}
+
+
+int dlm_init_mle_cache(void)
+{
+	dlm_mle_cache = kmem_cache_create("o2dlm_mle",
+					  sizeof(struct dlm_master_list_entry),
+					  0, SLAB_HWCACHE_ALIGN,
+					  NULL);
+	if (dlm_mle_cache == NULL)
+		return -ENOMEM;
+	return 0;
+}
+
+void dlm_destroy_mle_cache(void)
+{
+	if (dlm_mle_cache)
+		kmem_cache_destroy(dlm_mle_cache);
+}
+
+static void dlm_mle_release(struct kref *kref)
+{
+	struct dlm_master_list_entry *mle;
+	struct dlm_ctxt *dlm;
+
+	mle = container_of(kref, struct dlm_master_list_entry, mle_refs);
+	dlm = mle->dlm;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&dlm->master_lock);
+
+	mlog(0, "Releasing mle for %.*s, type %d\n", mle->mnamelen, mle->mname,
+	     mle->type);
+
+	/* remove from list if not already */
+	__dlm_unlink_mle(dlm, mle);
+
+	/* detach the mle from the domain node up/down events */
+	__dlm_mle_detach_hb_events(dlm, mle);
+
+	atomic_dec(&dlm->mle_cur_count[mle->type]);
+
+	/* NOTE: kfree under spinlock here.
+	 * if this is bad, we can move this to a freelist. */
+	kmem_cache_free(dlm_mle_cache, mle);
+}
+
+
+/*
+ * LOCK RESOURCE FUNCTIONS
+ */
+
+int dlm_init_master_caches(void)
+{
+	dlm_lockres_cache = kmem_cache_create("o2dlm_lockres",
+					      sizeof(struct dlm_lock_resource),
+					      0, SLAB_HWCACHE_ALIGN, NULL);
+	if (!dlm_lockres_cache)
+		goto bail;
+
+	dlm_lockname_cache = kmem_cache_create("o2dlm_lockname",
+					       DLM_LOCKID_NAME_MAX, 0,
+					       SLAB_HWCACHE_ALIGN, NULL);
+	if (!dlm_lockname_cache)
+		goto bail;
+
+	return 0;
+bail:
+	dlm_destroy_master_caches();
+	return -ENOMEM;
+}
+
+void dlm_destroy_master_caches(void)
+{
+	if (dlm_lockname_cache) {
+		kmem_cache_destroy(dlm_lockname_cache);
+		dlm_lockname_cache = NULL;
+	}
+
+	if (dlm_lockres_cache) {
+		kmem_cache_destroy(dlm_lockres_cache);
+		dlm_lockres_cache = NULL;
+	}
+}
+
+static void dlm_lockres_release(struct kref *kref)
+{
+	struct dlm_lock_resource *res;
+	struct dlm_ctxt *dlm;
+
+	res = container_of(kref, struct dlm_lock_resource, refs);
+	dlm = res->dlm;
+
+	/* This should not happen -- all lockres' have a name
+	 * associated with them at init time. */
+	BUG_ON(!res->lockname.name);
+
+	mlog(0, "destroying lockres %.*s\n", res->lockname.len,
+	     res->lockname.name);
+
+	spin_lock(&dlm->track_lock);
+	if (!list_empty(&res->tracking))
+		list_del_init(&res->tracking);
+	else {
+		mlog(ML_ERROR, "Resource %.*s not on the Tracking list\n",
+		     res->lockname.len, res->lockname.name);
+		dlm_print_one_lock_resource(res);
+	}
+	spin_unlock(&dlm->track_lock);
+
+	atomic_dec(&dlm->res_cur_count);
+
+	if (!hlist_unhashed(&res->hash_node) ||
+	    !list_empty(&res->granted) ||
+	    !list_empty(&res->converting) ||
+	    !list_empty(&res->blocked) ||
+	    !list_empty(&res->dirty) ||
+	    !list_empty(&res->recovering) ||
+	    !list_empty(&res->purge)) {
+		mlog(ML_ERROR,
+		     "Going to BUG for resource %.*s."
+		     "  We're on a list! [%c%c%c%c%c%c%c]\n",
+		     res->lockname.len, res->lockname.name,
+		     !hlist_unhashed(&res->hash_node) ? 'H' : ' ',
+		     !list_empty(&res->granted) ? 'G' : ' ',
+		     !list_empty(&res->converting) ? 'C' : ' ',
+		     !list_empty(&res->blocked) ? 'B' : ' ',
+		     !list_empty(&res->dirty) ? 'D' : ' ',
+		     !list_empty(&res->recovering) ? 'R' : ' ',
+		     !list_empty(&res->purge) ? 'P' : ' ');
+
+		dlm_print_one_lock_resource(res);
+	}
+
+	/* By the time we're ready to blow this guy away, we shouldn't
+	 * be on any lists. */
+	BUG_ON(!hlist_unhashed(&res->hash_node));
+	BUG_ON(!list_empty(&res->granted));
+	BUG_ON(!list_empty(&res->converting));
+	BUG_ON(!list_empty(&res->blocked));
+	BUG_ON(!list_empty(&res->dirty));
+	BUG_ON(!list_empty(&res->recovering));
+	BUG_ON(!list_empty(&res->purge));
+
+	kmem_cache_free(dlm_lockname_cache, (void *)res->lockname.name);
+
+	kmem_cache_free(dlm_lockres_cache, res);
+}
+
+void dlm_lockres_put(struct dlm_lock_resource *res)
+{
+	kref_put(&res->refs, dlm_lockres_release);
+}
+
+static void dlm_init_lockres(struct dlm_ctxt *dlm,
+			     struct dlm_lock_resource *res,
+			     const char *name, unsigned int namelen)
+{
+	char *qname;
+
+	/* If we memset here, we lose our reference to the kmalloc'd
+	 * res->lockname.name, so be sure to init every field
+	 * correctly! */
+
+	qname = (char *) res->lockname.name;
+	memcpy(qname, name, namelen);
+
+	res->lockname.len = namelen;
+	res->lockname.hash = dlm_lockid_hash(name, namelen);
+
+	init_waitqueue_head(&res->wq);
+	spin_lock_init(&res->spinlock);
+	INIT_HLIST_NODE(&res->hash_node);
+	INIT_LIST_HEAD(&res->granted);
+	INIT_LIST_HEAD(&res->converting);
+	INIT_LIST_HEAD(&res->blocked);
+	INIT_LIST_HEAD(&res->dirty);
+	INIT_LIST_HEAD(&res->recovering);
+	INIT_LIST_HEAD(&res->purge);
+	INIT_LIST_HEAD(&res->tracking);
+	atomic_set(&res->asts_reserved, 0);
+	res->migration_pending = 0;
+	res->inflight_locks = 0;
+	res->inflight_assert_workers = 0;
+
+	res->dlm = dlm;
+
+	kref_init(&res->refs);
+
+	atomic_inc(&dlm->res_tot_count);
+	atomic_inc(&dlm->res_cur_count);
+
+	/* just for consistency */
+	spin_lock(&res->spinlock);
+	dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN);
+	spin_unlock(&res->spinlock);
+
+	res->state = DLM_LOCK_RES_IN_PROGRESS;
+
+	res->last_used = 0;
+
+	spin_lock(&dlm->spinlock);
+	list_add_tail(&res->tracking, &dlm->tracking_list);
+	spin_unlock(&dlm->spinlock);
+
+	memset(res->lvb, 0, DLM_LVB_LEN);
+	memset(res->refmap, 0, sizeof(res->refmap));
+}
+
+struct dlm_lock_resource *dlm_new_lockres(struct dlm_ctxt *dlm,
+				   const char *name,
+				   unsigned int namelen)
+{
+	struct dlm_lock_resource *res = NULL;
+
+	res = kmem_cache_zalloc(dlm_lockres_cache, GFP_NOFS);
+	if (!res)
+		goto error;
+
+	res->lockname.name = kmem_cache_zalloc(dlm_lockname_cache, GFP_NOFS);
+	if (!res->lockname.name)
+		goto error;
+
+	dlm_init_lockres(dlm, res, name, namelen);
+	return res;
+
+error:
+	if (res)
+		kmem_cache_free(dlm_lockres_cache, res);
+	return NULL;
+}
+
+void dlm_lockres_set_refmap_bit(struct dlm_ctxt *dlm,
+				struct dlm_lock_resource *res, int bit)
+{
+	assert_spin_locked(&res->spinlock);
+
+	mlog(0, "res %.*s, set node %u, %ps()\n", res->lockname.len,
+	     res->lockname.name, bit, __builtin_return_address(0));
+
+	set_bit(bit, res->refmap);
+}
+
+void dlm_lockres_clear_refmap_bit(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *res, int bit)
+{
+	assert_spin_locked(&res->spinlock);
+
+	mlog(0, "res %.*s, clr node %u, %ps()\n", res->lockname.len,
+	     res->lockname.name, bit, __builtin_return_address(0));
+
+	clear_bit(bit, res->refmap);
+}
+
+static void __dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
+				   struct dlm_lock_resource *res)
+{
+	res->inflight_locks++;
+
+	mlog(0, "%s: res %.*s, inflight++: now %u, %ps()\n", dlm->name,
+	     res->lockname.len, res->lockname.name, res->inflight_locks,
+	     __builtin_return_address(0));
+}
+
+void dlm_lockres_grab_inflight_ref(struct dlm_ctxt *dlm,
+				   struct dlm_lock_resource *res)
+{
+	assert_spin_locked(&res->spinlock);
+	__dlm_lockres_grab_inflight_ref(dlm, res);
+}
+
+void dlm_lockres_drop_inflight_ref(struct dlm_ctxt *dlm,
+				   struct dlm_lock_resource *res)
+{
+	assert_spin_locked(&res->spinlock);
+
+	BUG_ON(res->inflight_locks == 0);
+
+	res->inflight_locks--;
+
+	mlog(0, "%s: res %.*s, inflight--: now %u, %ps()\n", dlm->name,
+	     res->lockname.len, res->lockname.name, res->inflight_locks,
+	     __builtin_return_address(0));
+
+	wake_up(&res->wq);
+}
+
+void __dlm_lockres_grab_inflight_worker(struct dlm_ctxt *dlm,
+		struct dlm_lock_resource *res)
+{
+	assert_spin_locked(&res->spinlock);
+	res->inflight_assert_workers++;
+	mlog(0, "%s:%.*s: inflight assert worker++: now %u\n",
+			dlm->name, res->lockname.len, res->lockname.name,
+			res->inflight_assert_workers);
+}
+
+static void __dlm_lockres_drop_inflight_worker(struct dlm_ctxt *dlm,
+		struct dlm_lock_resource *res)
+{
+	assert_spin_locked(&res->spinlock);
+	BUG_ON(res->inflight_assert_workers == 0);
+	res->inflight_assert_workers--;
+	mlog(0, "%s:%.*s: inflight assert worker--: now %u\n",
+			dlm->name, res->lockname.len, res->lockname.name,
+			res->inflight_assert_workers);
+}
+
+static void dlm_lockres_drop_inflight_worker(struct dlm_ctxt *dlm,
+		struct dlm_lock_resource *res)
+{
+	spin_lock(&res->spinlock);
+	__dlm_lockres_drop_inflight_worker(dlm, res);
+	spin_unlock(&res->spinlock);
+}
+
+/*
+ * lookup a lock resource by name.
+ * may already exist in the hashtable.
+ * lockid is null terminated
+ *
+ * if not, allocate enough for the lockres and for
+ * the temporary structure used in doing the mastering.
+ *
+ * also, do a lookup in the dlm->master_list to see
+ * if another node has begun mastering the same lock.
+ * if so, there should be a block entry in there
+ * for this name, and we should *not* attempt to master
+ * the lock here.   need to wait around for that node
+ * to assert_master (or die).
+ *
+ */
+struct dlm_lock_resource * dlm_get_lock_resource(struct dlm_ctxt *dlm,
+					  const char *lockid,
+					  int namelen,
+					  int flags)
+{
+	struct dlm_lock_resource *tmpres=NULL, *res=NULL;
+	struct dlm_master_list_entry *mle = NULL;
+	struct dlm_master_list_entry *alloc_mle = NULL;
+	int blocked = 0;
+	int ret, nodenum;
+	struct dlm_node_iter iter;
+	unsigned int hash;
+	int tries = 0;
+	int bit, wait_on_recovery = 0;
+
+	BUG_ON(!lockid);
+
+	hash = dlm_lockid_hash(lockid, namelen);
+
+	mlog(0, "get lockres %s (len %d)\n", lockid, namelen);
+
+lookup:
+	spin_lock(&dlm->spinlock);
+	tmpres = __dlm_lookup_lockres_full(dlm, lockid, namelen, hash);
+	if (tmpres) {
+		spin_unlock(&dlm->spinlock);
+		spin_lock(&tmpres->spinlock);
+
+		/*
+		 * Right after dlm spinlock was released, dlm_thread could have
+		 * purged the lockres. Check if lockres got unhashed. If so
+		 * start over.
+		 */
+		if (hlist_unhashed(&tmpres->hash_node)) {
+			spin_unlock(&tmpres->spinlock);
+			dlm_lockres_put(tmpres);
+			tmpres = NULL;
+			goto lookup;
+		}
+
+		/* Wait on the thread that is mastering the resource */
+		if (tmpres->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
+			__dlm_wait_on_lockres(tmpres);
+			BUG_ON(tmpres->owner == DLM_LOCK_RES_OWNER_UNKNOWN);
+			spin_unlock(&tmpres->spinlock);
+			dlm_lockres_put(tmpres);
+			tmpres = NULL;
+			goto lookup;
+		}
+
+		/* Wait on the resource purge to complete before continuing */
+		if (tmpres->state & DLM_LOCK_RES_DROPPING_REF) {
+			BUG_ON(tmpres->owner == dlm->node_num);
+			__dlm_wait_on_lockres_flags(tmpres,
+						    DLM_LOCK_RES_DROPPING_REF);
+			spin_unlock(&tmpres->spinlock);
+			dlm_lockres_put(tmpres);
+			tmpres = NULL;
+			goto lookup;
+		}
+
+		/* Grab inflight ref to pin the resource */
+		dlm_lockres_grab_inflight_ref(dlm, tmpres);
+
+		spin_unlock(&tmpres->spinlock);
+		if (res)
+			dlm_lockres_put(res);
+		res = tmpres;
+		goto leave;
+	}
+
+	if (!res) {
+		spin_unlock(&dlm->spinlock);
+		mlog(0, "allocating a new resource\n");
+		/* nothing found and we need to allocate one. */
+		alloc_mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
+		if (!alloc_mle)
+			goto leave;
+		res = dlm_new_lockres(dlm, lockid, namelen);
+		if (!res)
+			goto leave;
+		goto lookup;
+	}
+
+	mlog(0, "no lockres found, allocated our own: %p\n", res);
+
+	if (flags & LKM_LOCAL) {
+		/* caller knows it's safe to assume it's not mastered elsewhere
+		 * DONE!  return right away */
+		spin_lock(&res->spinlock);
+		dlm_change_lockres_owner(dlm, res, dlm->node_num);
+		__dlm_insert_lockres(dlm, res);
+		dlm_lockres_grab_inflight_ref(dlm, res);
+		spin_unlock(&res->spinlock);
+		spin_unlock(&dlm->spinlock);
+		/* lockres still marked IN_PROGRESS */
+		goto wake_waiters;
+	}
+
+	/* check master list to see if another node has started mastering it */
+	spin_lock(&dlm->master_lock);
+
+	/* if we found a block, wait for lock to be mastered by another node */
+	blocked = dlm_find_mle(dlm, &mle, (char *)lockid, namelen);
+	if (blocked) {
+		int mig;
+		if (mle->type == DLM_MLE_MASTER) {
+			mlog(ML_ERROR, "master entry for nonexistent lock!\n");
+			BUG();
+		}
+		mig = (mle->type == DLM_MLE_MIGRATION);
+		/* if there is a migration in progress, let the migration
+		 * finish before continuing.  we can wait for the absence
+		 * of the MIGRATION mle: either the migrate finished or
+		 * one of the nodes died and the mle was cleaned up.
+		 * if there is a BLOCK here, but it already has a master
+		 * set, we are too late.  the master does not have a ref
+		 * for us in the refmap.  detach the mle and drop it.
+		 * either way, go back to the top and start over. */
+		if (mig || mle->master != O2NM_MAX_NODES) {
+			BUG_ON(mig && mle->master == dlm->node_num);
+			/* we arrived too late.  the master does not
+			 * have a ref for us. retry. */
+			mlog(0, "%s:%.*s: late on %s\n",
+			     dlm->name, namelen, lockid,
+			     mig ?  "MIGRATION" : "BLOCK");
+			spin_unlock(&dlm->master_lock);
+			spin_unlock(&dlm->spinlock);
+
+			/* master is known, detach */
+			if (!mig)
+				dlm_mle_detach_hb_events(dlm, mle);
+			dlm_put_mle(mle);
+			mle = NULL;
+			/* this is lame, but we can't wait on either
+			 * the mle or lockres waitqueue here */
+			if (mig)
+				msleep(100);
+			goto lookup;
+		}
+	} else {
+		/* go ahead and try to master lock on this node */
+		mle = alloc_mle;
+		/* make sure this does not get freed below */
+		alloc_mle = NULL;
+		dlm_init_mle(mle, DLM_MLE_MASTER, dlm, res, NULL, 0);
+		set_bit(dlm->node_num, mle->maybe_map);
+		__dlm_insert_mle(dlm, mle);
+
+		/* still holding the dlm spinlock, check the recovery map
+		 * to see if there are any nodes that still need to be
+		 * considered.  these will not appear in the mle nodemap
+		 * but they might own this lockres.  wait on them. */
+		bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
+		if (bit < O2NM_MAX_NODES) {
+			mlog(0, "%s: res %.*s, At least one node (%d) "
+			     "to recover before lock mastery can begin\n",
+			     dlm->name, namelen, (char *)lockid, bit);
+			wait_on_recovery = 1;
+		}
+	}
+
+	/* at this point there is either a DLM_MLE_BLOCK or a
+	 * DLM_MLE_MASTER on the master list, so it's safe to add the
+	 * lockres to the hashtable.  anyone who finds the lock will
+	 * still have to wait on the IN_PROGRESS. */
+
+	/* finally add the lockres to its hash bucket */
+	__dlm_insert_lockres(dlm, res);
+
+	/* since this lockres is new it doesn't not require the spinlock */
+	__dlm_lockres_grab_inflight_ref(dlm, res);
+
+	/* get an extra ref on the mle in case this is a BLOCK
+	 * if so, the creator of the BLOCK may try to put the last
+	 * ref at this time in the assert master handler, so we
+	 * need an extra one to keep from a bad ptr deref. */
+	dlm_get_mle_inuse(mle);
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+
+redo_request:
+	while (wait_on_recovery) {
+		/* any cluster changes that occurred after dropping the
+		 * dlm spinlock would be detectable be a change on the mle,
+		 * so we only need to clear out the recovery map once. */
+		if (dlm_is_recovery_lock(lockid, namelen)) {
+			mlog(0, "%s: Recovery map is not empty, but must "
+			     "master $RECOVERY lock now\n", dlm->name);
+			if (!dlm_pre_master_reco_lockres(dlm, res))
+				wait_on_recovery = 0;
+			else {
+				mlog(0, "%s: waiting 500ms for heartbeat state "
+				    "change\n", dlm->name);
+				msleep(500);
+			}
+			continue;
+		}
+
+		dlm_kick_recovery_thread(dlm);
+		msleep(1000);
+		dlm_wait_for_recovery(dlm);
+
+		spin_lock(&dlm->spinlock);
+		bit = find_next_bit(dlm->recovery_map, O2NM_MAX_NODES, 0);
+		if (bit < O2NM_MAX_NODES) {
+			mlog(0, "%s: res %.*s, At least one node (%d) "
+			     "to recover before lock mastery can begin\n",
+			     dlm->name, namelen, (char *)lockid, bit);
+			wait_on_recovery = 1;
+		} else
+			wait_on_recovery = 0;
+		spin_unlock(&dlm->spinlock);
+
+		if (wait_on_recovery)
+			dlm_wait_for_node_recovery(dlm, bit, 10000);
+	}
+
+	/* must wait for lock to be mastered elsewhere */
+	if (blocked)
+		goto wait;
+
+	ret = -EINVAL;
+	dlm_node_iter_init(mle->vote_map, &iter);
+	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
+		ret = dlm_do_master_request(res, mle, nodenum);
+		if (ret < 0)
+			mlog_errno(ret);
+		if (mle->master != O2NM_MAX_NODES) {
+			/* found a master ! */
+			if (mle->master <= nodenum)
+				break;
+			/* if our master request has not reached the master
+			 * yet, keep going until it does.  this is how the
+			 * master will know that asserts are needed back to
+			 * the lower nodes. */
+			mlog(0, "%s: res %.*s, Requests only up to %u but "
+			     "master is %u, keep going\n", dlm->name, namelen,
+			     lockid, nodenum, mle->master);
+		}
+	}
+
+wait:
+	/* keep going until the response map includes all nodes */
+	ret = dlm_wait_for_lock_mastery(dlm, res, mle, &blocked);
+	if (ret < 0) {
+		wait_on_recovery = 1;
+		mlog(0, "%s: res %.*s, Node map changed, redo the master "
+		     "request now, blocked=%d\n", dlm->name, res->lockname.len,
+		     res->lockname.name, blocked);
+		if (++tries > 20) {
+			mlog(ML_ERROR, "%s: res %.*s, Spinning on "
+			     "dlm_wait_for_lock_mastery, blocked = %d\n",
+			     dlm->name, res->lockname.len,
+			     res->lockname.name, blocked);
+			dlm_print_one_lock_resource(res);
+			dlm_print_one_mle(mle);
+			tries = 0;
+		}
+		goto redo_request;
+	}
+
+	mlog(0, "%s: res %.*s, Mastered by %u\n", dlm->name, res->lockname.len,
+	     res->lockname.name, res->owner);
+	/* make sure we never continue without this */
+	BUG_ON(res->owner == O2NM_MAX_NODES);
+
+	/* master is known, detach if not already detached */
+	dlm_mle_detach_hb_events(dlm, mle);
+	dlm_put_mle(mle);
+	/* put the extra ref */
+	dlm_put_mle_inuse(mle);
+
+wake_waiters:
+	spin_lock(&res->spinlock);
+	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
+	spin_unlock(&res->spinlock);
+	wake_up(&res->wq);
+
+leave:
+	/* need to free the unused mle */
+	if (alloc_mle)
+		kmem_cache_free(dlm_mle_cache, alloc_mle);
+
+	return res;
+}
+
+
+#define DLM_MASTERY_TIMEOUT_MS   5000
+
+static int dlm_wait_for_lock_mastery(struct dlm_ctxt *dlm,
+				     struct dlm_lock_resource *res,
+				     struct dlm_master_list_entry *mle,
+				     int *blocked)
+{
+	u8 m;
+	int ret, bit;
+	int map_changed, voting_done;
+	int assert, sleep;
+
+recheck:
+	ret = 0;
+	assert = 0;
+
+	/* check if another node has already become the owner */
+	spin_lock(&res->spinlock);
+	if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
+		mlog(0, "%s:%.*s: owner is suddenly %u\n", dlm->name,
+		     res->lockname.len, res->lockname.name, res->owner);
+		spin_unlock(&res->spinlock);
+		/* this will cause the master to re-assert across
+		 * the whole cluster, freeing up mles */
+		if (res->owner != dlm->node_num) {
+			ret = dlm_do_master_request(res, mle, res->owner);
+			if (ret < 0) {
+				/* give recovery a chance to run */
+				mlog(ML_ERROR, "link to %u went down?: %d\n", res->owner, ret);
+				msleep(500);
+				goto recheck;
+			}
+		}
+		ret = 0;
+		goto leave;
+	}
+	spin_unlock(&res->spinlock);
+
+	spin_lock(&mle->spinlock);
+	m = mle->master;
+	map_changed = (memcmp(mle->vote_map, mle->node_map,
+			      sizeof(mle->vote_map)) != 0);
+	voting_done = (memcmp(mle->vote_map, mle->response_map,
+			     sizeof(mle->vote_map)) == 0);
+
+	/* restart if we hit any errors */
+	if (map_changed) {
+		int b;
+		mlog(0, "%s: %.*s: node map changed, restarting\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
+		ret = dlm_restart_lock_mastery(dlm, res, mle, *blocked);
+		b = (mle->type == DLM_MLE_BLOCK);
+		if ((*blocked && !b) || (!*blocked && b)) {
+			mlog(0, "%s:%.*s: status change: old=%d new=%d\n",
+			     dlm->name, res->lockname.len, res->lockname.name,
+			     *blocked, b);
+			*blocked = b;
+		}
+		spin_unlock(&mle->spinlock);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto leave;
+		}
+		mlog(0, "%s:%.*s: restart lock mastery succeeded, "
+		     "rechecking now\n", dlm->name, res->lockname.len,
+		     res->lockname.name);
+		goto recheck;
+	} else {
+		if (!voting_done) {
+			mlog(0, "map not changed and voting not done "
+			     "for %s:%.*s\n", dlm->name, res->lockname.len,
+			     res->lockname.name);
+		}
+	}
+
+	if (m != O2NM_MAX_NODES) {
+		/* another node has done an assert!
+		 * all done! */
+		sleep = 0;
+	} else {
+		sleep = 1;
+		/* have all nodes responded? */
+		if (voting_done && !*blocked) {
+			bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
+			if (dlm->node_num <= bit) {
+				/* my node number is lowest.
+			 	 * now tell other nodes that I am
+				 * mastering this. */
+				mle->master = dlm->node_num;
+				/* ref was grabbed in get_lock_resource
+				 * will be dropped in dlmlock_master */
+				assert = 1;
+				sleep = 0;
+			}
+			/* if voting is done, but we have not received
+			 * an assert master yet, we must sleep */
+		}
+	}
+
+	spin_unlock(&mle->spinlock);
+
+	/* sleep if we haven't finished voting yet */
+	if (sleep) {
+		unsigned long timeo = msecs_to_jiffies(DLM_MASTERY_TIMEOUT_MS);
+
+		/*
+		if (atomic_read(&mle->mle_refs.refcount) < 2)
+			mlog(ML_ERROR, "mle (%p) refs=%d, name=%.*s\n", mle,
+			atomic_read(&mle->mle_refs.refcount),
+			res->lockname.len, res->lockname.name);
+		*/
+		atomic_set(&mle->woken, 0);
+		(void)wait_event_timeout(mle->wq,
+					 (atomic_read(&mle->woken) == 1),
+					 timeo);
+		if (res->owner == O2NM_MAX_NODES) {
+			mlog(0, "%s:%.*s: waiting again\n", dlm->name,
+			     res->lockname.len, res->lockname.name);
+			goto recheck;
+		}
+		mlog(0, "done waiting, master is %u\n", res->owner);
+		ret = 0;
+		goto leave;
+	}
+
+	ret = 0;   /* done */
+	if (assert) {
+		m = dlm->node_num;
+		mlog(0, "about to master %.*s here, this=%u\n",
+		     res->lockname.len, res->lockname.name, m);
+		ret = dlm_do_assert_master(dlm, res, mle->vote_map, 0);
+		if (ret) {
+			/* This is a failure in the network path,
+			 * not in the response to the assert_master
+			 * (any nonzero response is a BUG on this node).
+			 * Most likely a socket just got disconnected
+			 * due to node death. */
+			mlog_errno(ret);
+		}
+		/* no longer need to restart lock mastery.
+		 * all living nodes have been contacted. */
+		ret = 0;
+	}
+
+	/* set the lockres owner */
+	spin_lock(&res->spinlock);
+	/* mastery reference obtained either during
+	 * assert_master_handler or in get_lock_resource */
+	dlm_change_lockres_owner(dlm, res, m);
+	spin_unlock(&res->spinlock);
+
+leave:
+	return ret;
+}
+
+struct dlm_bitmap_diff_iter
+{
+	int curnode;
+	unsigned long *orig_bm;
+	unsigned long *cur_bm;
+	unsigned long diff_bm[BITS_TO_LONGS(O2NM_MAX_NODES)];
+};
+
+enum dlm_node_state_change
+{
+	NODE_DOWN = -1,
+	NODE_NO_CHANGE = 0,
+	NODE_UP
+};
+
+static void dlm_bitmap_diff_iter_init(struct dlm_bitmap_diff_iter *iter,
+				      unsigned long *orig_bm,
+				      unsigned long *cur_bm)
+{
+	unsigned long p1, p2;
+	int i;
+
+	iter->curnode = -1;
+	iter->orig_bm = orig_bm;
+	iter->cur_bm = cur_bm;
+
+	for (i = 0; i < BITS_TO_LONGS(O2NM_MAX_NODES); i++) {
+       		p1 = *(iter->orig_bm + i);
+	       	p2 = *(iter->cur_bm + i);
+		iter->diff_bm[i] = (p1 & ~p2) | (p2 & ~p1);
+	}
+}
+
+static int dlm_bitmap_diff_iter_next(struct dlm_bitmap_diff_iter *iter,
+				     enum dlm_node_state_change *state)
+{
+	int bit;
+
+	if (iter->curnode >= O2NM_MAX_NODES)
+		return -ENOENT;
+
+	bit = find_next_bit(iter->diff_bm, O2NM_MAX_NODES,
+			    iter->curnode+1);
+	if (bit >= O2NM_MAX_NODES) {
+		iter->curnode = O2NM_MAX_NODES;
+		return -ENOENT;
+	}
+
+	/* if it was there in the original then this node died */
+	if (test_bit(bit, iter->orig_bm))
+		*state = NODE_DOWN;
+	else
+		*state = NODE_UP;
+
+	iter->curnode = bit;
+	return bit;
+}
+
+
+static int dlm_restart_lock_mastery(struct dlm_ctxt *dlm,
+				    struct dlm_lock_resource *res,
+				    struct dlm_master_list_entry *mle,
+				    int blocked)
+{
+	struct dlm_bitmap_diff_iter bdi;
+	enum dlm_node_state_change sc;
+	int node;
+	int ret = 0;
+
+	mlog(0, "something happened such that the "
+	     "master process may need to be restarted!\n");
+
+	assert_spin_locked(&mle->spinlock);
+
+	dlm_bitmap_diff_iter_init(&bdi, mle->vote_map, mle->node_map);
+	node = dlm_bitmap_diff_iter_next(&bdi, &sc);
+	while (node >= 0) {
+		if (sc == NODE_UP) {
+			/* a node came up.  clear any old vote from
+			 * the response map and set it in the vote map
+			 * then restart the mastery. */
+			mlog(ML_NOTICE, "node %d up while restarting\n", node);
+
+			/* redo the master request, but only for the new node */
+			mlog(0, "sending request to new node\n");
+			clear_bit(node, mle->response_map);
+			set_bit(node, mle->vote_map);
+		} else {
+			mlog(ML_ERROR, "node down! %d\n", node);
+			if (blocked) {
+				int lowest = find_next_bit(mle->maybe_map,
+						       O2NM_MAX_NODES, 0);
+
+				/* act like it was never there */
+				clear_bit(node, mle->maybe_map);
+
+			       	if (node == lowest) {
+					mlog(0, "expected master %u died"
+					    " while this node was blocked "
+					    "waiting on it!\n", node);
+					lowest = find_next_bit(mle->maybe_map,
+						       	O2NM_MAX_NODES,
+						       	lowest+1);
+					if (lowest < O2NM_MAX_NODES) {
+						mlog(0, "%s:%.*s:still "
+						     "blocked. waiting on %u "
+						     "now\n", dlm->name,
+						     res->lockname.len,
+						     res->lockname.name,
+						     lowest);
+					} else {
+						/* mle is an MLE_BLOCK, but
+						 * there is now nothing left to
+						 * block on.  we need to return
+						 * all the way back out and try
+						 * again with an MLE_MASTER.
+						 * dlm_do_local_recovery_cleanup
+						 * has already run, so the mle
+						 * refcount is ok */
+						mlog(0, "%s:%.*s: no "
+						     "longer blocking. try to "
+						     "master this here\n",
+						     dlm->name,
+						     res->lockname.len,
+						     res->lockname.name);
+						mle->type = DLM_MLE_MASTER;
+						mle->mleres = res;
+					}
+				}
+			}
+
+			/* now blank out everything, as if we had never
+			 * contacted anyone */
+			memset(mle->maybe_map, 0, sizeof(mle->maybe_map));
+			memset(mle->response_map, 0, sizeof(mle->response_map));
+			/* reset the vote_map to the current node_map */
+			memcpy(mle->vote_map, mle->node_map,
+			       sizeof(mle->node_map));
+			/* put myself into the maybe map */
+			if (mle->type != DLM_MLE_BLOCK)
+				set_bit(dlm->node_num, mle->maybe_map);
+		}
+		ret = -EAGAIN;
+		node = dlm_bitmap_diff_iter_next(&bdi, &sc);
+	}
+	return ret;
+}
+
+
+/*
+ * DLM_MASTER_REQUEST_MSG
+ *
+ * returns: 0 on success,
+ *          -errno on a network error
+ *
+ * on error, the caller should assume the target node is "dead"
+ *
+ */
+
+static int dlm_do_master_request(struct dlm_lock_resource *res,
+				 struct dlm_master_list_entry *mle, int to)
+{
+	struct dlm_ctxt *dlm = mle->dlm;
+	struct dlm_master_request request;
+	int ret, response=0, resend;
+
+	memset(&request, 0, sizeof(request));
+	request.node_idx = dlm->node_num;
+
+	BUG_ON(mle->type == DLM_MLE_MIGRATION);
+
+	request.namelen = (u8)mle->mnamelen;
+	memcpy(request.name, mle->mname, request.namelen);
+
+again:
+	ret = o2net_send_message(DLM_MASTER_REQUEST_MSG, dlm->key, &request,
+				 sizeof(request), to, &response);
+	if (ret < 0)  {
+		if (ret == -ESRCH) {
+			/* should never happen */
+			mlog(ML_ERROR, "TCP stack not ready!\n");
+			BUG();
+		} else if (ret == -EINVAL) {
+			mlog(ML_ERROR, "bad args passed to o2net!\n");
+			BUG();
+		} else if (ret == -ENOMEM) {
+			mlog(ML_ERROR, "out of memory while trying to send "
+			     "network message!  retrying\n");
+			/* this is totally crude */
+			msleep(50);
+			goto again;
+		} else if (!dlm_is_host_down(ret)) {
+			/* not a network error. bad. */
+			mlog_errno(ret);
+			mlog(ML_ERROR, "unhandled error!");
+			BUG();
+		}
+		/* all other errors should be network errors,
+		 * and likely indicate node death */
+		mlog(ML_ERROR, "link to %d went down!\n", to);
+		goto out;
+	}
+
+	ret = 0;
+	resend = 0;
+	spin_lock(&mle->spinlock);
+	switch (response) {
+		case DLM_MASTER_RESP_YES:
+			set_bit(to, mle->response_map);
+			mlog(0, "node %u is the master, response=YES\n", to);
+			mlog(0, "%s:%.*s: master node %u now knows I have a "
+			     "reference\n", dlm->name, res->lockname.len,
+			     res->lockname.name, to);
+			mle->master = to;
+			break;
+		case DLM_MASTER_RESP_NO:
+			mlog(0, "node %u not master, response=NO\n", to);
+			set_bit(to, mle->response_map);
+			break;
+		case DLM_MASTER_RESP_MAYBE:
+			mlog(0, "node %u not master, response=MAYBE\n", to);
+			set_bit(to, mle->response_map);
+			set_bit(to, mle->maybe_map);
+			break;
+		case DLM_MASTER_RESP_ERROR:
+			mlog(0, "node %u hit an error, resending\n", to);
+			resend = 1;
+			response = 0;
+			break;
+		default:
+			mlog(ML_ERROR, "bad response! %u\n", response);
+			BUG();
+	}
+	spin_unlock(&mle->spinlock);
+	if (resend) {
+		/* this is also totally crude */
+		msleep(50);
+		goto again;
+	}
+
+out:
+	return ret;
+}
+
+/*
+ * locks that can be taken here:
+ * dlm->spinlock
+ * res->spinlock
+ * mle->spinlock
+ * dlm->master_list
+ *
+ * if possible, TRIM THIS DOWN!!!
+ */
+int dlm_master_request_handler(struct o2net_msg *msg, u32 len, void *data,
+			       void **ret_data)
+{
+	u8 response = DLM_MASTER_RESP_MAYBE;
+	struct dlm_ctxt *dlm = data;
+	struct dlm_lock_resource *res = NULL;
+	struct dlm_master_request *request = (struct dlm_master_request *) msg->buf;
+	struct dlm_master_list_entry *mle = NULL, *tmpmle = NULL;
+	char *name;
+	unsigned int namelen, hash;
+	int found, ret;
+	int set_maybe;
+	int dispatch_assert = 0;
+
+	if (!dlm_grab(dlm))
+		return DLM_MASTER_RESP_NO;
+
+	if (!dlm_domain_fully_joined(dlm)) {
+		response = DLM_MASTER_RESP_NO;
+		goto send_response;
+	}
+
+	name = request->name;
+	namelen = request->namelen;
+	hash = dlm_lockid_hash(name, namelen);
+
+	if (namelen > DLM_LOCKID_NAME_MAX) {
+		response = DLM_IVBUFLEN;
+		goto send_response;
+	}
+
+way_up_top:
+	spin_lock(&dlm->spinlock);
+	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
+	if (res) {
+		spin_unlock(&dlm->spinlock);
+
+		/* take care of the easy cases up front */
+		spin_lock(&res->spinlock);
+
+		/*
+		 * Right after dlm spinlock was released, dlm_thread could have
+		 * purged the lockres. Check if lockres got unhashed. If so
+		 * start over.
+		 */
+		if (hlist_unhashed(&res->hash_node)) {
+			spin_unlock(&res->spinlock);
+			dlm_lockres_put(res);
+			goto way_up_top;
+		}
+
+		if (res->state & (DLM_LOCK_RES_RECOVERING|
+				  DLM_LOCK_RES_MIGRATING)) {
+			spin_unlock(&res->spinlock);
+			mlog(0, "returning DLM_MASTER_RESP_ERROR since res is "
+			     "being recovered/migrated\n");
+			response = DLM_MASTER_RESP_ERROR;
+			if (mle)
+				kmem_cache_free(dlm_mle_cache, mle);
+			goto send_response;
+		}
+
+		if (res->owner == dlm->node_num) {
+			dlm_lockres_set_refmap_bit(dlm, res, request->node_idx);
+			spin_unlock(&res->spinlock);
+			response = DLM_MASTER_RESP_YES;
+			if (mle)
+				kmem_cache_free(dlm_mle_cache, mle);
+
+			/* this node is the owner.
+			 * there is some extra work that needs to
+			 * happen now.  the requesting node has
+			 * caused all nodes up to this one to
+			 * create mles.  this node now needs to
+			 * go back and clean those up. */
+			dispatch_assert = 1;
+			goto send_response;
+		} else if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
+			spin_unlock(&res->spinlock);
+			// mlog(0, "node %u is the master\n", res->owner);
+			response = DLM_MASTER_RESP_NO;
+			if (mle)
+				kmem_cache_free(dlm_mle_cache, mle);
+			goto send_response;
+		}
+
+		/* ok, there is no owner.  either this node is
+		 * being blocked, or it is actively trying to
+		 * master this lock. */
+		if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
+			mlog(ML_ERROR, "lock with no owner should be "
+			     "in-progress!\n");
+			BUG();
+		}
+
+		// mlog(0, "lockres is in progress...\n");
+		spin_lock(&dlm->master_lock);
+		found = dlm_find_mle(dlm, &tmpmle, name, namelen);
+		if (!found) {
+			mlog(ML_ERROR, "no mle found for this lock!\n");
+			BUG();
+		}
+		set_maybe = 1;
+		spin_lock(&tmpmle->spinlock);
+		if (tmpmle->type == DLM_MLE_BLOCK) {
+			// mlog(0, "this node is waiting for "
+			// "lockres to be mastered\n");
+			response = DLM_MASTER_RESP_NO;
+		} else if (tmpmle->type == DLM_MLE_MIGRATION) {
+			mlog(0, "node %u is master, but trying to migrate to "
+			     "node %u.\n", tmpmle->master, tmpmle->new_master);
+			if (tmpmle->master == dlm->node_num) {
+				mlog(ML_ERROR, "no owner on lockres, but this "
+				     "node is trying to migrate it to %u?!\n",
+				     tmpmle->new_master);
+				BUG();
+			} else {
+				/* the real master can respond on its own */
+				response = DLM_MASTER_RESP_NO;
+			}
+		} else if (tmpmle->master != DLM_LOCK_RES_OWNER_UNKNOWN) {
+			set_maybe = 0;
+			if (tmpmle->master == dlm->node_num) {
+				response = DLM_MASTER_RESP_YES;
+				/* this node will be the owner.
+				 * go back and clean the mles on any
+				 * other nodes */
+				dispatch_assert = 1;
+				dlm_lockres_set_refmap_bit(dlm, res,
+							   request->node_idx);
+			} else
+				response = DLM_MASTER_RESP_NO;
+		} else {
+			// mlog(0, "this node is attempting to "
+			// "master lockres\n");
+			response = DLM_MASTER_RESP_MAYBE;
+		}
+		if (set_maybe)
+			set_bit(request->node_idx, tmpmle->maybe_map);
+		spin_unlock(&tmpmle->spinlock);
+
+		spin_unlock(&dlm->master_lock);
+		spin_unlock(&res->spinlock);
+
+		/* keep the mle attached to heartbeat events */
+		dlm_put_mle(tmpmle);
+		if (mle)
+			kmem_cache_free(dlm_mle_cache, mle);
+		goto send_response;
+	}
+
+	/*
+	 * lockres doesn't exist on this node
+	 * if there is an MLE_BLOCK, return NO
+	 * if there is an MLE_MASTER, return MAYBE
+	 * otherwise, add an MLE_BLOCK, return NO
+	 */
+	spin_lock(&dlm->master_lock);
+	found = dlm_find_mle(dlm, &tmpmle, name, namelen);
+	if (!found) {
+		/* this lockid has never been seen on this node yet */
+		// mlog(0, "no mle found\n");
+		if (!mle) {
+			spin_unlock(&dlm->master_lock);
+			spin_unlock(&dlm->spinlock);
+
+			mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
+			if (!mle) {
+				response = DLM_MASTER_RESP_ERROR;
+				mlog_errno(-ENOMEM);
+				goto send_response;
+			}
+			goto way_up_top;
+		}
+
+		// mlog(0, "this is second time thru, already allocated, "
+		// "add the block.\n");
+		dlm_init_mle(mle, DLM_MLE_BLOCK, dlm, NULL, name, namelen);
+		set_bit(request->node_idx, mle->maybe_map);
+		__dlm_insert_mle(dlm, mle);
+		response = DLM_MASTER_RESP_NO;
+	} else {
+		// mlog(0, "mle was found\n");
+		set_maybe = 1;
+		spin_lock(&tmpmle->spinlock);
+		if (tmpmle->master == dlm->node_num) {
+			mlog(ML_ERROR, "no lockres, but an mle with this node as master!\n");
+			BUG();
+		}
+		if (tmpmle->type == DLM_MLE_BLOCK)
+			response = DLM_MASTER_RESP_NO;
+		else if (tmpmle->type == DLM_MLE_MIGRATION) {
+			mlog(0, "migration mle was found (%u->%u)\n",
+			     tmpmle->master, tmpmle->new_master);
+			/* real master can respond on its own */
+			response = DLM_MASTER_RESP_NO;
+		} else
+			response = DLM_MASTER_RESP_MAYBE;
+		if (set_maybe)
+			set_bit(request->node_idx, tmpmle->maybe_map);
+		spin_unlock(&tmpmle->spinlock);
+	}
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+
+	if (found) {
+		/* keep the mle attached to heartbeat events */
+		dlm_put_mle(tmpmle);
+	}
+send_response:
+	/*
+	 * __dlm_lookup_lockres() grabbed a reference to this lockres.
+	 * The reference is released by dlm_assert_master_worker() under
+	 * the call to dlm_dispatch_assert_master().  If
+	 * dlm_assert_master_worker() isn't called, we drop it here.
+	 */
+	if (dispatch_assert) {
+		if (response != DLM_MASTER_RESP_YES)
+			mlog(ML_ERROR, "invalid response %d\n", response);
+		if (!res) {
+			mlog(ML_ERROR, "bad lockres while trying to assert!\n");
+			BUG();
+		}
+		mlog(0, "%u is the owner of %.*s, cleaning everyone else\n",
+			     dlm->node_num, res->lockname.len, res->lockname.name);
+		spin_lock(&res->spinlock);
+		ret = dlm_dispatch_assert_master(dlm, res, 0, request->node_idx,
+						 DLM_ASSERT_MASTER_MLE_CLEANUP);
+		if (ret < 0) {
+			mlog(ML_ERROR, "failed to dispatch assert master work\n");
+			response = DLM_MASTER_RESP_ERROR;
+			dlm_lockres_put(res);
+		} else
+			__dlm_lockres_grab_inflight_worker(dlm, res);
+		spin_unlock(&res->spinlock);
+	} else {
+		if (res)
+			dlm_lockres_put(res);
+	}
+
+	dlm_put(dlm);
+	return response;
+}
+
+/*
+ * DLM_ASSERT_MASTER_MSG
+ */
+
+
+/*
+ * NOTE: this can be used for debugging
+ * can periodically run all locks owned by this node
+ * and re-assert across the cluster...
+ */
+static int dlm_do_assert_master(struct dlm_ctxt *dlm,
+				struct dlm_lock_resource *res,
+				void *nodemap, u32 flags)
+{
+	struct dlm_assert_master assert;
+	int to, tmpret;
+	struct dlm_node_iter iter;
+	int ret = 0;
+	int reassert;
+	const char *lockname = res->lockname.name;
+	unsigned int namelen = res->lockname.len;
+
+	BUG_ON(namelen > O2NM_MAX_NAME_LEN);
+
+	spin_lock(&res->spinlock);
+	res->state |= DLM_LOCK_RES_SETREF_INPROG;
+	spin_unlock(&res->spinlock);
+
+again:
+	reassert = 0;
+
+	/* note that if this nodemap is empty, it returns 0 */
+	dlm_node_iter_init(nodemap, &iter);
+	while ((to = dlm_node_iter_next(&iter)) >= 0) {
+		int r = 0;
+		struct dlm_master_list_entry *mle = NULL;
+
+		mlog(0, "sending assert master to %d (%.*s)\n", to,
+		     namelen, lockname);
+		memset(&assert, 0, sizeof(assert));
+		assert.node_idx = dlm->node_num;
+		assert.namelen = namelen;
+		memcpy(assert.name, lockname, namelen);
+		assert.flags = cpu_to_be32(flags);
+
+		tmpret = o2net_send_message(DLM_ASSERT_MASTER_MSG, dlm->key,
+					    &assert, sizeof(assert), to, &r);
+		if (tmpret < 0) {
+			mlog(ML_ERROR, "Error %d when sending message %u (key "
+			     "0x%x) to node %u\n", tmpret,
+			     DLM_ASSERT_MASTER_MSG, dlm->key, to);
+			if (!dlm_is_host_down(tmpret)) {
+				mlog(ML_ERROR, "unhandled error=%d!\n", tmpret);
+				BUG();
+			}
+			/* a node died.  finish out the rest of the nodes. */
+			mlog(0, "link to %d went down!\n", to);
+			/* any nonzero status return will do */
+			ret = tmpret;
+			r = 0;
+		} else if (r < 0) {
+			/* ok, something horribly messed.  kill thyself. */
+			mlog(ML_ERROR,"during assert master of %.*s to %u, "
+			     "got %d.\n", namelen, lockname, to, r);
+			spin_lock(&dlm->spinlock);
+			spin_lock(&dlm->master_lock);
+			if (dlm_find_mle(dlm, &mle, (char *)lockname,
+					 namelen)) {
+				dlm_print_one_mle(mle);
+				__dlm_put_mle(mle);
+			}
+			spin_unlock(&dlm->master_lock);
+			spin_unlock(&dlm->spinlock);
+			BUG();
+		}
+
+		if (r & DLM_ASSERT_RESPONSE_REASSERT &&
+		    !(r & DLM_ASSERT_RESPONSE_MASTERY_REF)) {
+				mlog(ML_ERROR, "%.*s: very strange, "
+				     "master MLE but no lockres on %u\n",
+				     namelen, lockname, to);
+		}
+
+		if (r & DLM_ASSERT_RESPONSE_REASSERT) {
+			mlog(0, "%.*s: node %u create mles on other "
+			     "nodes and requests a re-assert\n",
+			     namelen, lockname, to);
+			reassert = 1;
+		}
+		if (r & DLM_ASSERT_RESPONSE_MASTERY_REF) {
+			mlog(0, "%.*s: node %u has a reference to this "
+			     "lockres, set the bit in the refmap\n",
+			     namelen, lockname, to);
+			spin_lock(&res->spinlock);
+			dlm_lockres_set_refmap_bit(dlm, res, to);
+			spin_unlock(&res->spinlock);
+		}
+	}
+
+	if (reassert)
+		goto again;
+
+	spin_lock(&res->spinlock);
+	res->state &= ~DLM_LOCK_RES_SETREF_INPROG;
+	spin_unlock(&res->spinlock);
+	wake_up(&res->wq);
+
+	return ret;
+}
+
+/*
+ * locks that can be taken here:
+ * dlm->spinlock
+ * res->spinlock
+ * mle->spinlock
+ * dlm->master_list
+ *
+ * if possible, TRIM THIS DOWN!!!
+ */
+int dlm_assert_master_handler(struct o2net_msg *msg, u32 len, void *data,
+			      void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_master_list_entry *mle = NULL;
+	struct dlm_assert_master *assert = (struct dlm_assert_master *)msg->buf;
+	struct dlm_lock_resource *res = NULL;
+	char *name;
+	unsigned int namelen, hash;
+	u32 flags;
+	int master_request = 0, have_lockres_ref = 0;
+	int ret = 0;
+
+	if (!dlm_grab(dlm))
+		return 0;
+
+	name = assert->name;
+	namelen = assert->namelen;
+	hash = dlm_lockid_hash(name, namelen);
+	flags = be32_to_cpu(assert->flags);
+
+	if (namelen > DLM_LOCKID_NAME_MAX) {
+		mlog(ML_ERROR, "Invalid name length!");
+		goto done;
+	}
+
+	spin_lock(&dlm->spinlock);
+
+	if (flags)
+		mlog(0, "assert_master with flags: %u\n", flags);
+
+	/* find the MLE */
+	spin_lock(&dlm->master_lock);
+	if (!dlm_find_mle(dlm, &mle, name, namelen)) {
+		/* not an error, could be master just re-asserting */
+		mlog(0, "just got an assert_master from %u, but no "
+		     "MLE for it! (%.*s)\n", assert->node_idx,
+		     namelen, name);
+	} else {
+		int bit = find_next_bit (mle->maybe_map, O2NM_MAX_NODES, 0);
+		if (bit >= O2NM_MAX_NODES) {
+			/* not necessarily an error, though less likely.
+			 * could be master just re-asserting. */
+			mlog(0, "no bits set in the maybe_map, but %u "
+			     "is asserting! (%.*s)\n", assert->node_idx,
+			     namelen, name);
+		} else if (bit != assert->node_idx) {
+			if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
+				mlog(0, "master %u was found, %u should "
+				     "back off\n", assert->node_idx, bit);
+			} else {
+				/* with the fix for bug 569, a higher node
+				 * number winning the mastery will respond
+				 * YES to mastery requests, but this node
+				 * had no way of knowing.  let it pass. */
+				mlog(0, "%u is the lowest node, "
+				     "%u is asserting. (%.*s)  %u must "
+				     "have begun after %u won.\n", bit,
+				     assert->node_idx, namelen, name, bit,
+				     assert->node_idx);
+			}
+		}
+		if (mle->type == DLM_MLE_MIGRATION) {
+			if (flags & DLM_ASSERT_MASTER_MLE_CLEANUP) {
+				mlog(0, "%s:%.*s: got cleanup assert"
+				     " from %u for migration\n",
+				     dlm->name, namelen, name,
+				     assert->node_idx);
+			} else if (!(flags & DLM_ASSERT_MASTER_FINISH_MIGRATION)) {
+				mlog(0, "%s:%.*s: got unrelated assert"
+				     " from %u for migration, ignoring\n",
+				     dlm->name, namelen, name,
+				     assert->node_idx);
+				__dlm_put_mle(mle);
+				spin_unlock(&dlm->master_lock);
+				spin_unlock(&dlm->spinlock);
+				goto done;
+			}
+		}
+	}
+	spin_unlock(&dlm->master_lock);
+
+	/* ok everything checks out with the MLE
+	 * now check to see if there is a lockres */
+	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
+	if (res) {
+		spin_lock(&res->spinlock);
+		if (res->state & DLM_LOCK_RES_RECOVERING)  {
+			mlog(ML_ERROR, "%u asserting but %.*s is "
+			     "RECOVERING!\n", assert->node_idx, namelen, name);
+			goto kill;
+		}
+		if (!mle) {
+			if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN &&
+			    res->owner != assert->node_idx) {
+				mlog(ML_ERROR, "DIE! Mastery assert from %u, "
+				     "but current owner is %u! (%.*s)\n",
+				     assert->node_idx, res->owner, namelen,
+				     name);
+				__dlm_print_one_lock_resource(res);
+				BUG();
+			}
+		} else if (mle->type != DLM_MLE_MIGRATION) {
+			if (res->owner != DLM_LOCK_RES_OWNER_UNKNOWN) {
+				/* owner is just re-asserting */
+				if (res->owner == assert->node_idx) {
+					mlog(0, "owner %u re-asserting on "
+					     "lock %.*s\n", assert->node_idx,
+					     namelen, name);
+					goto ok;
+				}
+				mlog(ML_ERROR, "got assert_master from "
+				     "node %u, but %u is the owner! "
+				     "(%.*s)\n", assert->node_idx,
+				     res->owner, namelen, name);
+				goto kill;
+			}
+			if (!(res->state & DLM_LOCK_RES_IN_PROGRESS)) {
+				mlog(ML_ERROR, "got assert from %u, but lock "
+				     "with no owner should be "
+				     "in-progress! (%.*s)\n",
+				     assert->node_idx,
+				     namelen, name);
+				goto kill;
+			}
+		} else /* mle->type == DLM_MLE_MIGRATION */ {
+			/* should only be getting an assert from new master */
+			if (assert->node_idx != mle->new_master) {
+				mlog(ML_ERROR, "got assert from %u, but "
+				     "new master is %u, and old master "
+				     "was %u (%.*s)\n",
+				     assert->node_idx, mle->new_master,
+				     mle->master, namelen, name);
+				goto kill;
+			}
+
+		}
+ok:
+		spin_unlock(&res->spinlock);
+	}
+
+	// mlog(0, "woo!  got an assert_master from node %u!\n",
+	// 	     assert->node_idx);
+	if (mle) {
+		int extra_ref = 0;
+		int nn = -1;
+		int rr, err = 0;
+
+		spin_lock(&mle->spinlock);
+		if (mle->type == DLM_MLE_BLOCK || mle->type == DLM_MLE_MIGRATION)
+			extra_ref = 1;
+		else {
+			/* MASTER mle: if any bits set in the response map
+			 * then the calling node needs to re-assert to clear
+			 * up nodes that this node contacted */
+			while ((nn = find_next_bit (mle->response_map, O2NM_MAX_NODES,
+						    nn+1)) < O2NM_MAX_NODES) {
+				if (nn != dlm->node_num && nn != assert->node_idx) {
+					master_request = 1;
+					break;
+				}
+			}
+		}
+		mle->master = assert->node_idx;
+		atomic_set(&mle->woken, 1);
+		wake_up(&mle->wq);
+		spin_unlock(&mle->spinlock);
+
+		if (res) {
+			int wake = 0;
+			spin_lock(&res->spinlock);
+			if (mle->type == DLM_MLE_MIGRATION) {
+				mlog(0, "finishing off migration of lockres %.*s, "
+			     		"from %u to %u\n",
+			       		res->lockname.len, res->lockname.name,
+			       		dlm->node_num, mle->new_master);
+				res->state &= ~DLM_LOCK_RES_MIGRATING;
+				wake = 1;
+				dlm_change_lockres_owner(dlm, res, mle->new_master);
+				BUG_ON(res->state & DLM_LOCK_RES_DIRTY);
+			} else {
+				dlm_change_lockres_owner(dlm, res, mle->master);
+			}
+			spin_unlock(&res->spinlock);
+			have_lockres_ref = 1;
+			if (wake)
+				wake_up(&res->wq);
+		}
+
+		/* master is known, detach if not already detached.
+		 * ensures that only one assert_master call will happen
+		 * on this mle. */
+		spin_lock(&dlm->master_lock);
+
+		rr = atomic_read(&mle->mle_refs.refcount);
+		if (mle->inuse > 0) {
+			if (extra_ref && rr < 3)
+				err = 1;
+			else if (!extra_ref && rr < 2)
+				err = 1;
+		} else {
+			if (extra_ref && rr < 2)
+				err = 1;
+			else if (!extra_ref && rr < 1)
+				err = 1;
+		}
+		if (err) {
+			mlog(ML_ERROR, "%s:%.*s: got assert master from %u "
+			     "that will mess up this node, refs=%d, extra=%d, "
+			     "inuse=%d\n", dlm->name, namelen, name,
+			     assert->node_idx, rr, extra_ref, mle->inuse);
+			dlm_print_one_mle(mle);
+		}
+		__dlm_unlink_mle(dlm, mle);
+		__dlm_mle_detach_hb_events(dlm, mle);
+		__dlm_put_mle(mle);
+		if (extra_ref) {
+			/* the assert master message now balances the extra
+		 	 * ref given by the master / migration request message.
+		 	 * if this is the last put, it will be removed
+		 	 * from the list. */
+			__dlm_put_mle(mle);
+		}
+		spin_unlock(&dlm->master_lock);
+	} else if (res) {
+		if (res->owner != assert->node_idx) {
+			mlog(0, "assert_master from %u, but current "
+			     "owner is %u (%.*s), no mle\n", assert->node_idx,
+			     res->owner, namelen, name);
+		}
+	}
+	spin_unlock(&dlm->spinlock);
+
+done:
+	ret = 0;
+	if (res) {
+		spin_lock(&res->spinlock);
+		res->state |= DLM_LOCK_RES_SETREF_INPROG;
+		spin_unlock(&res->spinlock);
+		*ret_data = (void *)res;
+	}
+	dlm_put(dlm);
+	if (master_request) {
+		mlog(0, "need to tell master to reassert\n");
+		/* positive. negative would shoot down the node. */
+		ret |= DLM_ASSERT_RESPONSE_REASSERT;
+		if (!have_lockres_ref) {
+			mlog(ML_ERROR, "strange, got assert from %u, MASTER "
+			     "mle present here for %s:%.*s, but no lockres!\n",
+			     assert->node_idx, dlm->name, namelen, name);
+		}
+	}
+	if (have_lockres_ref) {
+		/* let the master know we have a reference to the lockres */
+		ret |= DLM_ASSERT_RESPONSE_MASTERY_REF;
+		mlog(0, "%s:%.*s: got assert from %u, need a ref\n",
+		     dlm->name, namelen, name, assert->node_idx);
+	}
+	return ret;
+
+kill:
+	/* kill the caller! */
+	mlog(ML_ERROR, "Bad message received from another node.  Dumping state "
+	     "and killing the other node now!  This node is OK and can continue.\n");
+	__dlm_print_one_lock_resource(res);
+	spin_unlock(&res->spinlock);
+	spin_lock(&dlm->master_lock);
+	if (mle)
+		__dlm_put_mle(mle);
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+	*ret_data = (void *)res;
+	dlm_put(dlm);
+	return -EINVAL;
+}
+
+void dlm_assert_master_post_handler(int status, void *data, void *ret_data)
+{
+	struct dlm_lock_resource *res = (struct dlm_lock_resource *)ret_data;
+
+	if (ret_data) {
+		spin_lock(&res->spinlock);
+		res->state &= ~DLM_LOCK_RES_SETREF_INPROG;
+		spin_unlock(&res->spinlock);
+		wake_up(&res->wq);
+		dlm_lockres_put(res);
+	}
+	return;
+}
+
+int dlm_dispatch_assert_master(struct dlm_ctxt *dlm,
+			       struct dlm_lock_resource *res,
+			       int ignore_higher, u8 request_from, u32 flags)
+{
+	struct dlm_work_item *item;
+	item = kzalloc(sizeof(*item), GFP_ATOMIC);
+	if (!item)
+		return -ENOMEM;
+
+
+	/* queue up work for dlm_assert_master_worker */
+	dlm_grab(dlm);  /* get an extra ref for the work item */
+	dlm_init_work_item(dlm, item, dlm_assert_master_worker, NULL);
+	item->u.am.lockres = res; /* already have a ref */
+	/* can optionally ignore node numbers higher than this node */
+	item->u.am.ignore_higher = ignore_higher;
+	item->u.am.request_from = request_from;
+	item->u.am.flags = flags;
+
+	if (ignore_higher)
+		mlog(0, "IGNORE HIGHER: %.*s\n", res->lockname.len,
+		     res->lockname.name);
+
+	spin_lock(&dlm->work_lock);
+	list_add_tail(&item->list, &dlm->work_list);
+	spin_unlock(&dlm->work_lock);
+
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
+	return 0;
+}
+
+static void dlm_assert_master_worker(struct dlm_work_item *item, void *data)
+{
+	struct dlm_ctxt *dlm = data;
+	int ret = 0;
+	struct dlm_lock_resource *res;
+	unsigned long nodemap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	int ignore_higher;
+	int bit;
+	u8 request_from;
+	u32 flags;
+
+	dlm = item->dlm;
+	res = item->u.am.lockres;
+	ignore_higher = item->u.am.ignore_higher;
+	request_from = item->u.am.request_from;
+	flags = item->u.am.flags;
+
+	spin_lock(&dlm->spinlock);
+	memcpy(nodemap, dlm->domain_map, sizeof(nodemap));
+	spin_unlock(&dlm->spinlock);
+
+	clear_bit(dlm->node_num, nodemap);
+	if (ignore_higher) {
+		/* if is this just to clear up mles for nodes below
+		 * this node, do not send the message to the original
+		 * caller or any node number higher than this */
+		clear_bit(request_from, nodemap);
+		bit = dlm->node_num;
+		while (1) {
+			bit = find_next_bit(nodemap, O2NM_MAX_NODES,
+					    bit+1);
+		       	if (bit >= O2NM_MAX_NODES)
+				break;
+			clear_bit(bit, nodemap);
+		}
+	}
+
+	/*
+	 * If we're migrating this lock to someone else, we are no
+	 * longer allowed to assert out own mastery.  OTOH, we need to
+	 * prevent migration from starting while we're still asserting
+	 * our dominance.  The reserved ast delays migration.
+	 */
+	spin_lock(&res->spinlock);
+	if (res->state & DLM_LOCK_RES_MIGRATING) {
+		mlog(0, "Someone asked us to assert mastery, but we're "
+		     "in the middle of migration.  Skipping assert, "
+		     "the new master will handle that.\n");
+		spin_unlock(&res->spinlock);
+		goto put;
+	} else
+		__dlm_lockres_reserve_ast(res);
+	spin_unlock(&res->spinlock);
+
+	/* this call now finishes out the nodemap
+	 * even if one or more nodes die */
+	mlog(0, "worker about to master %.*s here, this=%u\n",
+		     res->lockname.len, res->lockname.name, dlm->node_num);
+	ret = dlm_do_assert_master(dlm, res, nodemap, flags);
+	if (ret < 0) {
+		/* no need to restart, we are done */
+		if (!dlm_is_host_down(ret))
+			mlog_errno(ret);
+	}
+
+	/* Ok, we've asserted ourselves.  Let's let migration start. */
+	dlm_lockres_release_ast(dlm, res);
+
+put:
+	dlm_lockres_drop_inflight_worker(dlm, res);
+
+	dlm_lockres_put(res);
+
+	mlog(0, "finished with dlm_assert_master_worker\n");
+}
+
+/* SPECIAL CASE for the $RECOVERY lock used by the recovery thread.
+ * We cannot wait for node recovery to complete to begin mastering this
+ * lockres because this lockres is used to kick off recovery! ;-)
+ * So, do a pre-check on all living nodes to see if any of those nodes
+ * think that $RECOVERY is currently mastered by a dead node.  If so,
+ * we wait a short time to allow that node to get notified by its own
+ * heartbeat stack, then check again.  All $RECOVERY lock resources
+ * mastered by dead nodes are purged when the hearbeat callback is
+ * fired, so we can know for sure that it is safe to continue once
+ * the node returns a live node or no node.  */
+static int dlm_pre_master_reco_lockres(struct dlm_ctxt *dlm,
+				       struct dlm_lock_resource *res)
+{
+	struct dlm_node_iter iter;
+	int nodenum;
+	int ret = 0;
+	u8 master = DLM_LOCK_RES_OWNER_UNKNOWN;
+
+	spin_lock(&dlm->spinlock);
+	dlm_node_iter_init(dlm->domain_map, &iter);
+	spin_unlock(&dlm->spinlock);
+
+	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
+		/* do not send to self */
+		if (nodenum == dlm->node_num)
+			continue;
+		ret = dlm_do_master_requery(dlm, res, nodenum, &master);
+		if (ret < 0) {
+			mlog_errno(ret);
+			if (!dlm_is_host_down(ret))
+				BUG();
+			/* host is down, so answer for that node would be
+			 * DLM_LOCK_RES_OWNER_UNKNOWN.  continue. */
+			ret = 0;
+		}
+
+		if (master != DLM_LOCK_RES_OWNER_UNKNOWN) {
+			/* check to see if this master is in the recovery map */
+			spin_lock(&dlm->spinlock);
+			if (test_bit(master, dlm->recovery_map)) {
+				mlog(ML_NOTICE, "%s: node %u has not seen "
+				     "node %u go down yet, and thinks the "
+				     "dead node is mastering the recovery "
+				     "lock.  must wait.\n", dlm->name,
+				     nodenum, master);
+				ret = -EAGAIN;
+			}
+			spin_unlock(&dlm->spinlock);
+			mlog(0, "%s: reco lock master is %u\n", dlm->name,
+			     master);
+			break;
+		}
+	}
+	return ret;
+}
+
+/*
+ * DLM_DEREF_LOCKRES_MSG
+ */
+
+int dlm_drop_lockres_ref(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
+{
+	struct dlm_deref_lockres deref;
+	int ret = 0, r;
+	const char *lockname;
+	unsigned int namelen;
+
+	lockname = res->lockname.name;
+	namelen = res->lockname.len;
+	BUG_ON(namelen > O2NM_MAX_NAME_LEN);
+
+	memset(&deref, 0, sizeof(deref));
+	deref.node_idx = dlm->node_num;
+	deref.namelen = namelen;
+	memcpy(deref.name, lockname, namelen);
+
+	ret = o2net_send_message(DLM_DEREF_LOCKRES_MSG, dlm->key,
+				 &deref, sizeof(deref), res->owner, &r);
+	if (ret < 0)
+		mlog(ML_ERROR, "%s: res %.*s, error %d send DEREF to node %u\n",
+		     dlm->name, namelen, lockname, ret, res->owner);
+	else if (r < 0) {
+		/* BAD.  other node says I did not have a ref. */
+		mlog(ML_ERROR, "%s: res %.*s, DEREF to node %u got %d\n",
+		     dlm->name, namelen, lockname, res->owner, r);
+		dlm_print_one_lock_resource(res);
+		BUG();
+	}
+	return ret;
+}
+
+int dlm_deref_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
+			      void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_deref_lockres *deref = (struct dlm_deref_lockres *)msg->buf;
+	struct dlm_lock_resource *res = NULL;
+	char *name;
+	unsigned int namelen;
+	int ret = -EINVAL;
+	u8 node;
+	unsigned int hash;
+	struct dlm_work_item *item;
+	int cleared = 0;
+	int dispatch = 0;
+
+	if (!dlm_grab(dlm))
+		return 0;
+
+	name = deref->name;
+	namelen = deref->namelen;
+	node = deref->node_idx;
+
+	if (namelen > DLM_LOCKID_NAME_MAX) {
+		mlog(ML_ERROR, "Invalid name length!");
+		goto done;
+	}
+	if (deref->node_idx >= O2NM_MAX_NODES) {
+		mlog(ML_ERROR, "Invalid node number: %u\n", node);
+		goto done;
+	}
+
+	hash = dlm_lockid_hash(name, namelen);
+
+	spin_lock(&dlm->spinlock);
+	res = __dlm_lookup_lockres_full(dlm, name, namelen, hash);
+	if (!res) {
+		spin_unlock(&dlm->spinlock);
+		mlog(ML_ERROR, "%s:%.*s: bad lockres name\n",
+		     dlm->name, namelen, name);
+		goto done;
+	}
+	spin_unlock(&dlm->spinlock);
+
+	spin_lock(&res->spinlock);
+	if (res->state & DLM_LOCK_RES_SETREF_INPROG)
+		dispatch = 1;
+	else {
+		BUG_ON(res->state & DLM_LOCK_RES_DROPPING_REF);
+		if (test_bit(node, res->refmap)) {
+			dlm_lockres_clear_refmap_bit(dlm, res, node);
+			cleared = 1;
+		}
+	}
+	spin_unlock(&res->spinlock);
+
+	if (!dispatch) {
+		if (cleared)
+			dlm_lockres_calc_usage(dlm, res);
+		else {
+			mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref "
+		     	"but it is already dropped!\n", dlm->name,
+		     	res->lockname.len, res->lockname.name, node);
+			dlm_print_one_lock_resource(res);
+		}
+		ret = 0;
+		goto done;
+	}
+
+	item = kzalloc(sizeof(*item), GFP_NOFS);
+	if (!item) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto done;
+	}
+
+	dlm_init_work_item(dlm, item, dlm_deref_lockres_worker, NULL);
+	item->u.dl.deref_res = res;
+	item->u.dl.deref_node = node;
+
+	spin_lock(&dlm->work_lock);
+	list_add_tail(&item->list, &dlm->work_list);
+	spin_unlock(&dlm->work_lock);
+
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
+	return 0;
+
+done:
+	if (res)
+		dlm_lockres_put(res);
+	dlm_put(dlm);
+
+	return ret;
+}
+
+static void dlm_deref_lockres_worker(struct dlm_work_item *item, void *data)
+{
+	struct dlm_ctxt *dlm;
+	struct dlm_lock_resource *res;
+	u8 node;
+	u8 cleared = 0;
+
+	dlm = item->dlm;
+	res = item->u.dl.deref_res;
+	node = item->u.dl.deref_node;
+
+	spin_lock(&res->spinlock);
+	BUG_ON(res->state & DLM_LOCK_RES_DROPPING_REF);
+	if (test_bit(node, res->refmap)) {
+		__dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
+		dlm_lockres_clear_refmap_bit(dlm, res, node);
+		cleared = 1;
+	}
+	spin_unlock(&res->spinlock);
+
+	if (cleared) {
+		mlog(0, "%s:%.*s node %u ref dropped in dispatch\n",
+		     dlm->name, res->lockname.len, res->lockname.name, node);
+		dlm_lockres_calc_usage(dlm, res);
+	} else {
+		mlog(ML_ERROR, "%s:%.*s: node %u trying to drop ref "
+		     "but it is already dropped!\n", dlm->name,
+		     res->lockname.len, res->lockname.name, node);
+		dlm_print_one_lock_resource(res);
+	}
+
+	dlm_lockres_put(res);
+}
+
+/*
+ * A migrateable resource is one that is :
+ * 1. locally mastered, and,
+ * 2. zero local locks, and,
+ * 3. one or more non-local locks, or, one or more references
+ * Returns 1 if yes, 0 if not.
+ */
+static int dlm_is_lockres_migrateable(struct dlm_ctxt *dlm,
+				      struct dlm_lock_resource *res)
+{
+	enum dlm_lockres_list idx;
+	int nonlocal = 0, node_ref;
+	struct list_head *queue;
+	struct dlm_lock *lock;
+	u64 cookie;
+
+	assert_spin_locked(&res->spinlock);
+
+	/* delay migration when the lockres is in MIGRATING state */
+	if (res->state & DLM_LOCK_RES_MIGRATING)
+		return 0;
+
+	/* delay migration when the lockres is in RECOCERING state */
+	if (res->state & DLM_LOCK_RES_RECOVERING)
+		return 0;
+
+	if (res->owner != dlm->node_num)
+		return 0;
+
+        for (idx = DLM_GRANTED_LIST; idx <= DLM_BLOCKED_LIST; idx++) {
+		queue = dlm_list_idx_to_ptr(res, idx);
+		list_for_each_entry(lock, queue, list) {
+			if (lock->ml.node != dlm->node_num) {
+				nonlocal++;
+				continue;
+			}
+			cookie = be64_to_cpu(lock->ml.cookie);
+			mlog(0, "%s: Not migrateable res %.*s, lock %u:%llu on "
+			     "%s list\n", dlm->name, res->lockname.len,
+			     res->lockname.name,
+			     dlm_get_lock_cookie_node(cookie),
+			     dlm_get_lock_cookie_seq(cookie),
+			     dlm_list_in_text(idx));
+			return 0;
+		}
+	}
+
+	if (!nonlocal) {
+		node_ref = find_next_bit(res->refmap, O2NM_MAX_NODES, 0);
+		if (node_ref >= O2NM_MAX_NODES)
+			return 0;
+	}
+
+	mlog(0, "%s: res %.*s, Migrateable\n", dlm->name, res->lockname.len,
+	     res->lockname.name);
+
+	return 1;
+}
+
+/*
+ * DLM_MIGRATE_LOCKRES
+ */
+
+
+static int dlm_migrate_lockres(struct dlm_ctxt *dlm,
+			       struct dlm_lock_resource *res, u8 target)
+{
+	struct dlm_master_list_entry *mle = NULL;
+	struct dlm_master_list_entry *oldmle = NULL;
+ 	struct dlm_migratable_lockres *mres = NULL;
+	int ret = 0;
+	const char *name;
+	unsigned int namelen;
+	int mle_added = 0;
+	int wake = 0;
+
+	if (!dlm_grab(dlm))
+		return -EINVAL;
+
+	BUG_ON(target == O2NM_MAX_NODES);
+
+	name = res->lockname.name;
+	namelen = res->lockname.len;
+
+	mlog(0, "%s: Migrating %.*s to node %u\n", dlm->name, namelen, name,
+	     target);
+
+	/* preallocate up front. if this fails, abort */
+	ret = -ENOMEM;
+	mres = (struct dlm_migratable_lockres *) __get_free_page(GFP_NOFS);
+	if (!mres) {
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
+	if (!mle) {
+		mlog_errno(ret);
+		goto leave;
+	}
+	ret = 0;
+
+	/*
+	 * clear any existing master requests and
+	 * add the migration mle to the list
+	 */
+	spin_lock(&dlm->spinlock);
+	spin_lock(&dlm->master_lock);
+	ret = dlm_add_migration_mle(dlm, res, mle, &oldmle, name,
+				    namelen, target, dlm->node_num);
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+
+	if (ret == -EEXIST) {
+		mlog(0, "another process is already migrating it\n");
+		goto fail;
+	}
+	mle_added = 1;
+
+	/*
+	 * set the MIGRATING flag and flush asts
+	 * if we fail after this we need to re-dirty the lockres
+	 */
+	if (dlm_mark_lockres_migrating(dlm, res, target) < 0) {
+		mlog(ML_ERROR, "tried to migrate %.*s to %u, but "
+		     "the target went down.\n", res->lockname.len,
+		     res->lockname.name, target);
+		spin_lock(&res->spinlock);
+		res->state &= ~DLM_LOCK_RES_MIGRATING;
+		wake = 1;
+		spin_unlock(&res->spinlock);
+		ret = -EINVAL;
+	}
+
+fail:
+	if (oldmle) {
+		/* master is known, detach if not already detached */
+		dlm_mle_detach_hb_events(dlm, oldmle);
+		dlm_put_mle(oldmle);
+	}
+
+	if (ret < 0) {
+		if (mle_added) {
+			dlm_mle_detach_hb_events(dlm, mle);
+			dlm_put_mle(mle);
+		} else if (mle) {
+			kmem_cache_free(dlm_mle_cache, mle);
+			mle = NULL;
+		}
+		goto leave;
+	}
+
+	/*
+	 * at this point, we have a migration target, an mle
+	 * in the master list, and the MIGRATING flag set on
+	 * the lockres
+	 */
+
+	/* now that remote nodes are spinning on the MIGRATING flag,
+	 * ensure that all assert_master work is flushed. */
+	flush_workqueue(dlm->dlm_worker);
+
+	/* get an extra reference on the mle.
+	 * otherwise the assert_master from the new
+	 * master will destroy this.
+	 * also, make sure that all callers of dlm_get_mle
+	 * take both dlm->spinlock and dlm->master_lock */
+	spin_lock(&dlm->spinlock);
+	spin_lock(&dlm->master_lock);
+	dlm_get_mle_inuse(mle);
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+
+	/* notify new node and send all lock state */
+	/* call send_one_lockres with migration flag.
+	 * this serves as notice to the target node that a
+	 * migration is starting. */
+	ret = dlm_send_one_lockres(dlm, res, mres, target,
+				   DLM_MRES_MIGRATION);
+
+	if (ret < 0) {
+		mlog(0, "migration to node %u failed with %d\n",
+		     target, ret);
+		/* migration failed, detach and clean up mle */
+		dlm_mle_detach_hb_events(dlm, mle);
+		dlm_put_mle(mle);
+		dlm_put_mle_inuse(mle);
+		spin_lock(&res->spinlock);
+		res->state &= ~DLM_LOCK_RES_MIGRATING;
+		wake = 1;
+		spin_unlock(&res->spinlock);
+		if (dlm_is_host_down(ret))
+			dlm_wait_for_node_death(dlm, target,
+						DLM_NODE_DEATH_WAIT_MAX);
+		goto leave;
+	}
+
+	/* at this point, the target sends a message to all nodes,
+	 * (using dlm_do_migrate_request).  this node is skipped since
+	 * we had to put an mle in the list to begin the process.  this
+	 * node now waits for target to do an assert master.  this node
+	 * will be the last one notified, ensuring that the migration
+	 * is complete everywhere.  if the target dies while this is
+	 * going on, some nodes could potentially see the target as the
+	 * master, so it is important that my recovery finds the migration
+	 * mle and sets the master to UNKNOWN. */
+
+
+	/* wait for new node to assert master */
+	while (1) {
+		ret = wait_event_interruptible_timeout(mle->wq,
+					(atomic_read(&mle->woken) == 1),
+					msecs_to_jiffies(5000));
+
+		if (ret >= 0) {
+		       	if (atomic_read(&mle->woken) == 1 ||
+			    res->owner == target)
+				break;
+
+			mlog(0, "%s:%.*s: timed out during migration\n",
+			     dlm->name, res->lockname.len, res->lockname.name);
+			/* avoid hang during shutdown when migrating lockres
+			 * to a node which also goes down */
+			if (dlm_is_node_dead(dlm, target)) {
+				mlog(0, "%s:%.*s: expected migration "
+				     "target %u is no longer up, restarting\n",
+				     dlm->name, res->lockname.len,
+				     res->lockname.name, target);
+				ret = -EINVAL;
+				/* migration failed, detach and clean up mle */
+				dlm_mle_detach_hb_events(dlm, mle);
+				dlm_put_mle(mle);
+				dlm_put_mle_inuse(mle);
+				spin_lock(&res->spinlock);
+				res->state &= ~DLM_LOCK_RES_MIGRATING;
+				wake = 1;
+				spin_unlock(&res->spinlock);
+				goto leave;
+			}
+		} else
+			mlog(0, "%s:%.*s: caught signal during migration\n",
+			     dlm->name, res->lockname.len, res->lockname.name);
+	}
+
+	/* all done, set the owner, clear the flag */
+	spin_lock(&res->spinlock);
+	dlm_set_lockres_owner(dlm, res, target);
+	res->state &= ~DLM_LOCK_RES_MIGRATING;
+	dlm_remove_nonlocal_locks(dlm, res);
+	spin_unlock(&res->spinlock);
+	wake_up(&res->wq);
+
+	/* master is known, detach if not already detached */
+	dlm_mle_detach_hb_events(dlm, mle);
+	dlm_put_mle_inuse(mle);
+	ret = 0;
+
+	dlm_lockres_calc_usage(dlm, res);
+
+leave:
+	/* re-dirty the lockres if we failed */
+	if (ret < 0)
+		dlm_kick_thread(dlm, res);
+
+	/* wake up waiters if the MIGRATING flag got set
+	 * but migration failed */
+	if (wake)
+		wake_up(&res->wq);
+
+	if (mres)
+		free_page((unsigned long)mres);
+
+	dlm_put(dlm);
+
+	mlog(0, "%s: Migrating %.*s to %u, returns %d\n", dlm->name, namelen,
+	     name, target, ret);
+	return ret;
+}
+
+#define DLM_MIGRATION_RETRY_MS  100
+
+/*
+ * Should be called only after beginning the domain leave process.
+ * There should not be any remaining locks on nonlocal lock resources,
+ * and there should be no local locks left on locally mastered resources.
+ *
+ * Called with the dlm spinlock held, may drop it to do migration, but
+ * will re-acquire before exit.
+ *
+ * Returns: 1 if dlm->spinlock was dropped/retaken, 0 if never dropped
+ */
+int dlm_empty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
+{
+	int ret;
+	int lock_dropped = 0;
+	u8 target = O2NM_MAX_NODES;
+
+	assert_spin_locked(&dlm->spinlock);
+
+	spin_lock(&res->spinlock);
+	if (dlm_is_lockres_migrateable(dlm, res))
+		target = dlm_pick_migration_target(dlm, res);
+	spin_unlock(&res->spinlock);
+
+	if (target == O2NM_MAX_NODES)
+		goto leave;
+
+	/* Wheee! Migrate lockres here! Will sleep so drop spinlock. */
+	spin_unlock(&dlm->spinlock);
+	lock_dropped = 1;
+	ret = dlm_migrate_lockres(dlm, res, target);
+	if (ret)
+		mlog(0, "%s: res %.*s, Migrate to node %u failed with %d\n",
+		     dlm->name, res->lockname.len, res->lockname.name,
+		     target, ret);
+	spin_lock(&dlm->spinlock);
+leave:
+	return lock_dropped;
+}
+
+int dlm_lock_basts_flushed(struct dlm_ctxt *dlm, struct dlm_lock *lock)
+{
+	int ret;
+	spin_lock(&dlm->ast_lock);
+	spin_lock(&lock->spinlock);
+	ret = (list_empty(&lock->bast_list) && !lock->bast_pending);
+	spin_unlock(&lock->spinlock);
+	spin_unlock(&dlm->ast_lock);
+	return ret;
+}
+
+static int dlm_migration_can_proceed(struct dlm_ctxt *dlm,
+				     struct dlm_lock_resource *res,
+				     u8 mig_target)
+{
+	int can_proceed;
+	spin_lock(&res->spinlock);
+	can_proceed = !!(res->state & DLM_LOCK_RES_MIGRATING);
+	spin_unlock(&res->spinlock);
+
+	/* target has died, so make the caller break out of the
+	 * wait_event, but caller must recheck the domain_map */
+	spin_lock(&dlm->spinlock);
+	if (!test_bit(mig_target, dlm->domain_map))
+		can_proceed = 1;
+	spin_unlock(&dlm->spinlock);
+	return can_proceed;
+}
+
+static int dlm_lockres_is_dirty(struct dlm_ctxt *dlm,
+				struct dlm_lock_resource *res)
+{
+	int ret;
+	spin_lock(&res->spinlock);
+	ret = !!(res->state & DLM_LOCK_RES_DIRTY);
+	spin_unlock(&res->spinlock);
+	return ret;
+}
+
+
+static int dlm_mark_lockres_migrating(struct dlm_ctxt *dlm,
+				       struct dlm_lock_resource *res,
+				       u8 target)
+{
+	int ret = 0;
+
+	mlog(0, "dlm_mark_lockres_migrating: %.*s, from %u to %u\n",
+	       res->lockname.len, res->lockname.name, dlm->node_num,
+	       target);
+	/* need to set MIGRATING flag on lockres.  this is done by
+	 * ensuring that all asts have been flushed for this lockres. */
+	spin_lock(&res->spinlock);
+	BUG_ON(res->migration_pending);
+	res->migration_pending = 1;
+	/* strategy is to reserve an extra ast then release
+	 * it below, letting the release do all of the work */
+	__dlm_lockres_reserve_ast(res);
+	spin_unlock(&res->spinlock);
+
+	/* now flush all the pending asts */
+	dlm_kick_thread(dlm, res);
+	/* before waiting on DIRTY, block processes which may
+	 * try to dirty the lockres before MIGRATING is set */
+	spin_lock(&res->spinlock);
+	BUG_ON(res->state & DLM_LOCK_RES_BLOCK_DIRTY);
+	res->state |= DLM_LOCK_RES_BLOCK_DIRTY;
+	spin_unlock(&res->spinlock);
+	/* now wait on any pending asts and the DIRTY state */
+	wait_event(dlm->ast_wq, !dlm_lockres_is_dirty(dlm, res));
+	dlm_lockres_release_ast(dlm, res);
+
+	mlog(0, "about to wait on migration_wq, dirty=%s\n",
+	       res->state & DLM_LOCK_RES_DIRTY ? "yes" : "no");
+	/* if the extra ref we just put was the final one, this
+	 * will pass thru immediately.  otherwise, we need to wait
+	 * for the last ast to finish. */
+again:
+	ret = wait_event_interruptible_timeout(dlm->migration_wq,
+		   dlm_migration_can_proceed(dlm, res, target),
+		   msecs_to_jiffies(1000));
+	if (ret < 0) {
+		mlog(0, "woken again: migrating? %s, dead? %s\n",
+		       res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
+		       test_bit(target, dlm->domain_map) ? "no":"yes");
+	} else {
+		mlog(0, "all is well: migrating? %s, dead? %s\n",
+		       res->state & DLM_LOCK_RES_MIGRATING ? "yes":"no",
+		       test_bit(target, dlm->domain_map) ? "no":"yes");
+	}
+	if (!dlm_migration_can_proceed(dlm, res, target)) {
+		mlog(0, "trying again...\n");
+		goto again;
+	}
+
+	ret = 0;
+	/* did the target go down or die? */
+	spin_lock(&dlm->spinlock);
+	if (!test_bit(target, dlm->domain_map)) {
+		mlog(ML_ERROR, "aha. migration target %u just went down\n",
+		     target);
+		ret = -EHOSTDOWN;
+	}
+	spin_unlock(&dlm->spinlock);
+
+	/*
+	 * if target is down, we need to clear DLM_LOCK_RES_BLOCK_DIRTY for
+	 * another try; otherwise, we are sure the MIGRATING state is there,
+	 * drop the unneded state which blocked threads trying to DIRTY
+	 */
+	spin_lock(&res->spinlock);
+	BUG_ON(!(res->state & DLM_LOCK_RES_BLOCK_DIRTY));
+	res->state &= ~DLM_LOCK_RES_BLOCK_DIRTY;
+	if (!ret)
+		BUG_ON(!(res->state & DLM_LOCK_RES_MIGRATING));
+	spin_unlock(&res->spinlock);
+
+	/*
+	 * at this point:
+	 *
+	 *   o the DLM_LOCK_RES_MIGRATING flag is set if target not down
+	 *   o there are no pending asts on this lockres
+	 *   o all processes trying to reserve an ast on this
+	 *     lockres must wait for the MIGRATING flag to clear
+	 */
+	return ret;
+}
+
+/* last step in the migration process.
+ * original master calls this to free all of the dlm_lock
+ * structures that used to be for other nodes. */
+static void dlm_remove_nonlocal_locks(struct dlm_ctxt *dlm,
+				      struct dlm_lock_resource *res)
+{
+	struct list_head *queue = &res->granted;
+	int i, bit;
+	struct dlm_lock *lock, *next;
+
+	assert_spin_locked(&res->spinlock);
+
+	BUG_ON(res->owner == dlm->node_num);
+
+	for (i=0; i<3; i++) {
+		list_for_each_entry_safe(lock, next, queue, list) {
+			if (lock->ml.node != dlm->node_num) {
+				mlog(0, "putting lock for node %u\n",
+				     lock->ml.node);
+				/* be extra careful */
+				BUG_ON(!list_empty(&lock->ast_list));
+				BUG_ON(!list_empty(&lock->bast_list));
+				BUG_ON(lock->ast_pending);
+				BUG_ON(lock->bast_pending);
+				dlm_lockres_clear_refmap_bit(dlm, res,
+							     lock->ml.node);
+				list_del_init(&lock->list);
+				dlm_lock_put(lock);
+				/* In a normal unlock, we would have added a
+				 * DLM_UNLOCK_FREE_LOCK action. Force it. */
+				dlm_lock_put(lock);
+			}
+		}
+		queue++;
+	}
+	bit = 0;
+	while (1) {
+		bit = find_next_bit(res->refmap, O2NM_MAX_NODES, bit);
+		if (bit >= O2NM_MAX_NODES)
+			break;
+		/* do not clear the local node reference, if there is a
+		 * process holding this, let it drop the ref itself */
+		if (bit != dlm->node_num) {
+			mlog(0, "%s:%.*s: node %u had a ref to this "
+			     "migrating lockres, clearing\n", dlm->name,
+			     res->lockname.len, res->lockname.name, bit);
+			dlm_lockres_clear_refmap_bit(dlm, res, bit);
+		}
+		bit++;
+	}
+}
+
+/*
+ * Pick a node to migrate the lock resource to. This function selects a
+ * potential target based first on the locks and then on refmap. It skips
+ * nodes that are in the process of exiting the domain.
+ */
+static u8 dlm_pick_migration_target(struct dlm_ctxt *dlm,
+				    struct dlm_lock_resource *res)
+{
+	enum dlm_lockres_list idx;
+	struct list_head *queue = &res->granted;
+	struct dlm_lock *lock;
+	int noderef;
+	u8 nodenum = O2NM_MAX_NODES;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&res->spinlock);
+
+	/* Go through all the locks */
+	for (idx = DLM_GRANTED_LIST; idx <= DLM_BLOCKED_LIST; idx++) {
+		queue = dlm_list_idx_to_ptr(res, idx);
+		list_for_each_entry(lock, queue, list) {
+			if (lock->ml.node == dlm->node_num)
+				continue;
+			if (test_bit(lock->ml.node, dlm->exit_domain_map))
+				continue;
+			nodenum = lock->ml.node;
+			goto bail;
+		}
+	}
+
+	/* Go thru the refmap */
+	noderef = -1;
+	while (1) {
+		noderef = find_next_bit(res->refmap, O2NM_MAX_NODES,
+					noderef + 1);
+		if (noderef >= O2NM_MAX_NODES)
+			break;
+		if (noderef == dlm->node_num)
+			continue;
+		if (test_bit(noderef, dlm->exit_domain_map))
+			continue;
+		nodenum = noderef;
+		goto bail;
+	}
+
+bail:
+	return nodenum;
+}
+
+/* this is called by the new master once all lockres
+ * data has been received */
+static int dlm_do_migrate_request(struct dlm_ctxt *dlm,
+				  struct dlm_lock_resource *res,
+				  u8 master, u8 new_master,
+				  struct dlm_node_iter *iter)
+{
+	struct dlm_migrate_request migrate;
+	int ret, skip, status = 0;
+	int nodenum;
+
+	memset(&migrate, 0, sizeof(migrate));
+	migrate.namelen = res->lockname.len;
+	memcpy(migrate.name, res->lockname.name, migrate.namelen);
+	migrate.new_master = new_master;
+	migrate.master = master;
+
+	ret = 0;
+
+	/* send message to all nodes, except the master and myself */
+	while ((nodenum = dlm_node_iter_next(iter)) >= 0) {
+		if (nodenum == master ||
+		    nodenum == new_master)
+			continue;
+
+		/* We could race exit domain. If exited, skip. */
+		spin_lock(&dlm->spinlock);
+		skip = (!test_bit(nodenum, dlm->domain_map));
+		spin_unlock(&dlm->spinlock);
+		if (skip) {
+			clear_bit(nodenum, iter->node_map);
+			continue;
+		}
+
+		ret = o2net_send_message(DLM_MIGRATE_REQUEST_MSG, dlm->key,
+					 &migrate, sizeof(migrate), nodenum,
+					 &status);
+		if (ret < 0) {
+			mlog(ML_ERROR, "%s: res %.*s, Error %d send "
+			     "MIGRATE_REQUEST to node %u\n", dlm->name,
+			     migrate.namelen, migrate.name, ret, nodenum);
+			if (!dlm_is_host_down(ret)) {
+				mlog(ML_ERROR, "unhandled error=%d!\n", ret);
+				BUG();
+			}
+			clear_bit(nodenum, iter->node_map);
+			ret = 0;
+		} else if (status < 0) {
+			mlog(0, "migrate request (node %u) returned %d!\n",
+			     nodenum, status);
+			ret = status;
+		} else if (status == DLM_MIGRATE_RESPONSE_MASTERY_REF) {
+			/* during the migration request we short-circuited
+			 * the mastery of the lockres.  make sure we have
+			 * a mastery ref for nodenum */
+			mlog(0, "%s:%.*s: need ref for node %u\n",
+			     dlm->name, res->lockname.len, res->lockname.name,
+			     nodenum);
+			spin_lock(&res->spinlock);
+			dlm_lockres_set_refmap_bit(dlm, res, nodenum);
+			spin_unlock(&res->spinlock);
+		}
+	}
+
+	if (ret < 0)
+		mlog_errno(ret);
+
+	mlog(0, "returning ret=%d\n", ret);
+	return ret;
+}
+
+
+/* if there is an existing mle for this lockres, we now know who the master is.
+ * (the one who sent us *this* message) we can clear it up right away.
+ * since the process that put the mle on the list still has a reference to it,
+ * we can unhash it now, set the master and wake the process.  as a result,
+ * we will have no mle in the list to start with.  now we can add an mle for
+ * the migration and this should be the only one found for those scanning the
+ * list.  */
+int dlm_migrate_request_handler(struct o2net_msg *msg, u32 len, void *data,
+				void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_lock_resource *res = NULL;
+	struct dlm_migrate_request *migrate = (struct dlm_migrate_request *) msg->buf;
+	struct dlm_master_list_entry *mle = NULL, *oldmle = NULL;
+	const char *name;
+	unsigned int namelen, hash;
+	int ret = 0;
+
+	if (!dlm_grab(dlm))
+		return -EINVAL;
+
+	name = migrate->name;
+	namelen = migrate->namelen;
+	hash = dlm_lockid_hash(name, namelen);
+
+	/* preallocate.. if this fails, abort */
+	mle = kmem_cache_alloc(dlm_mle_cache, GFP_NOFS);
+
+	if (!mle) {
+		ret = -ENOMEM;
+		goto leave;
+	}
+
+	/* check for pre-existing lock */
+	spin_lock(&dlm->spinlock);
+	res = __dlm_lookup_lockres(dlm, name, namelen, hash);
+	if (res) {
+		spin_lock(&res->spinlock);
+		if (res->state & DLM_LOCK_RES_RECOVERING) {
+			/* if all is working ok, this can only mean that we got
+		 	* a migrate request from a node that we now see as
+		 	* dead.  what can we do here?  drop it to the floor? */
+			spin_unlock(&res->spinlock);
+			mlog(ML_ERROR, "Got a migrate request, but the "
+			     "lockres is marked as recovering!");
+			kmem_cache_free(dlm_mle_cache, mle);
+			ret = -EINVAL; /* need a better solution */
+			goto unlock;
+		}
+		res->state |= DLM_LOCK_RES_MIGRATING;
+		spin_unlock(&res->spinlock);
+	}
+
+	spin_lock(&dlm->master_lock);
+	/* ignore status.  only nonzero status would BUG. */
+	ret = dlm_add_migration_mle(dlm, res, mle, &oldmle,
+				    name, namelen,
+				    migrate->new_master,
+				    migrate->master);
+
+	spin_unlock(&dlm->master_lock);
+unlock:
+	spin_unlock(&dlm->spinlock);
+
+	if (oldmle) {
+		/* master is known, detach if not already detached */
+		dlm_mle_detach_hb_events(dlm, oldmle);
+		dlm_put_mle(oldmle);
+	}
+
+	if (res)
+		dlm_lockres_put(res);
+leave:
+	dlm_put(dlm);
+	return ret;
+}
+
+/* must be holding dlm->spinlock and dlm->master_lock
+ * when adding a migration mle, we can clear any other mles
+ * in the master list because we know with certainty that
+ * the master is "master".  so we remove any old mle from
+ * the list after setting it's master field, and then add
+ * the new migration mle.  this way we can hold with the rule
+ * of having only one mle for a given lock name at all times. */
+static int dlm_add_migration_mle(struct dlm_ctxt *dlm,
+				 struct dlm_lock_resource *res,
+				 struct dlm_master_list_entry *mle,
+				 struct dlm_master_list_entry **oldmle,
+				 const char *name, unsigned int namelen,
+				 u8 new_master, u8 master)
+{
+	int found;
+	int ret = 0;
+
+	*oldmle = NULL;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&dlm->master_lock);
+
+	/* caller is responsible for any ref taken here on oldmle */
+	found = dlm_find_mle(dlm, oldmle, (char *)name, namelen);
+	if (found) {
+		struct dlm_master_list_entry *tmp = *oldmle;
+		spin_lock(&tmp->spinlock);
+		if (tmp->type == DLM_MLE_MIGRATION) {
+			if (master == dlm->node_num) {
+				/* ah another process raced me to it */
+				mlog(0, "tried to migrate %.*s, but some "
+				     "process beat me to it\n",
+				     namelen, name);
+				ret = -EEXIST;
+			} else {
+				/* bad.  2 NODES are trying to migrate! */
+				mlog(ML_ERROR, "migration error  mle: "
+				     "master=%u new_master=%u // request: "
+				     "master=%u new_master=%u // "
+				     "lockres=%.*s\n",
+				     tmp->master, tmp->new_master,
+				     master, new_master,
+				     namelen, name);
+				BUG();
+			}
+		} else {
+			/* this is essentially what assert_master does */
+			tmp->master = master;
+			atomic_set(&tmp->woken, 1);
+			wake_up(&tmp->wq);
+			/* remove it so that only one mle will be found */
+			__dlm_unlink_mle(dlm, tmp);
+			__dlm_mle_detach_hb_events(dlm, tmp);
+			if (tmp->type == DLM_MLE_MASTER) {
+				ret = DLM_MIGRATE_RESPONSE_MASTERY_REF;
+				mlog(0, "%s:%.*s: master=%u, newmaster=%u, "
+						"telling master to get ref "
+						"for cleared out mle during "
+						"migration\n", dlm->name,
+						namelen, name, master,
+						new_master);
+			}
+		}
+		spin_unlock(&tmp->spinlock);
+	}
+
+	/* now add a migration mle to the tail of the list */
+	dlm_init_mle(mle, DLM_MLE_MIGRATION, dlm, res, name, namelen);
+	mle->new_master = new_master;
+	/* the new master will be sending an assert master for this.
+	 * at that point we will get the refmap reference */
+	mle->master = master;
+	/* do this for consistency with other mle types */
+	set_bit(new_master, mle->maybe_map);
+	__dlm_insert_mle(dlm, mle);
+
+	return ret;
+}
+
+/*
+ * Sets the owner of the lockres, associated to the mle, to UNKNOWN
+ */
+static struct dlm_lock_resource *dlm_reset_mleres_owner(struct dlm_ctxt *dlm,
+					struct dlm_master_list_entry *mle)
+{
+	struct dlm_lock_resource *res;
+
+	/* Find the lockres associated to the mle and set its owner to UNK */
+	res = __dlm_lookup_lockres(dlm, mle->mname, mle->mnamelen,
+				   mle->mnamehash);
+	if (res) {
+		spin_unlock(&dlm->master_lock);
+
+		/* move lockres onto recovery list */
+		spin_lock(&res->spinlock);
+		dlm_set_lockres_owner(dlm, res, DLM_LOCK_RES_OWNER_UNKNOWN);
+		dlm_move_lockres_to_recovery_list(dlm, res);
+		spin_unlock(&res->spinlock);
+		dlm_lockres_put(res);
+
+		/* about to get rid of mle, detach from heartbeat */
+		__dlm_mle_detach_hb_events(dlm, mle);
+
+		/* dump the mle */
+		spin_lock(&dlm->master_lock);
+		__dlm_put_mle(mle);
+		spin_unlock(&dlm->master_lock);
+	}
+
+	return res;
+}
+
+static void dlm_clean_migration_mle(struct dlm_ctxt *dlm,
+				    struct dlm_master_list_entry *mle)
+{
+	__dlm_mle_detach_hb_events(dlm, mle);
+
+	spin_lock(&mle->spinlock);
+	__dlm_unlink_mle(dlm, mle);
+	atomic_set(&mle->woken, 1);
+	spin_unlock(&mle->spinlock);
+
+	wake_up(&mle->wq);
+}
+
+static void dlm_clean_block_mle(struct dlm_ctxt *dlm,
+				struct dlm_master_list_entry *mle, u8 dead_node)
+{
+	int bit;
+
+	BUG_ON(mle->type != DLM_MLE_BLOCK);
+
+	spin_lock(&mle->spinlock);
+	bit = find_next_bit(mle->maybe_map, O2NM_MAX_NODES, 0);
+	if (bit != dead_node) {
+		mlog(0, "mle found, but dead node %u would not have been "
+		     "master\n", dead_node);
+		spin_unlock(&mle->spinlock);
+	} else {
+		/* Must drop the refcount by one since the assert_master will
+		 * never arrive. This may result in the mle being unlinked and
+		 * freed, but there may still be a process waiting in the
+		 * dlmlock path which is fine. */
+		mlog(0, "node %u was expected master\n", dead_node);
+		atomic_set(&mle->woken, 1);
+		spin_unlock(&mle->spinlock);
+		wake_up(&mle->wq);
+
+		/* Do not need events any longer, so detach from heartbeat */
+		__dlm_mle_detach_hb_events(dlm, mle);
+		__dlm_put_mle(mle);
+	}
+}
+
+void dlm_clean_master_list(struct dlm_ctxt *dlm, u8 dead_node)
+{
+	struct dlm_master_list_entry *mle;
+	struct dlm_lock_resource *res;
+	struct hlist_head *bucket;
+	struct hlist_node *tmp;
+	unsigned int i;
+
+	mlog(0, "dlm=%s, dead node=%u\n", dlm->name, dead_node);
+top:
+	assert_spin_locked(&dlm->spinlock);
+
+	/* clean the master list */
+	spin_lock(&dlm->master_lock);
+	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
+		bucket = dlm_master_hash(dlm, i);
+		hlist_for_each_entry_safe(mle, tmp, bucket, master_hash_node) {
+			BUG_ON(mle->type != DLM_MLE_BLOCK &&
+			       mle->type != DLM_MLE_MASTER &&
+			       mle->type != DLM_MLE_MIGRATION);
+
+			/* MASTER mles are initiated locally. The waiting
+			 * process will notice the node map change shortly.
+			 * Let that happen as normal. */
+			if (mle->type == DLM_MLE_MASTER)
+				continue;
+
+			/* BLOCK mles are initiated by other nodes. Need to
+			 * clean up if the dead node would have been the
+			 * master. */
+			if (mle->type == DLM_MLE_BLOCK) {
+				dlm_clean_block_mle(dlm, mle, dead_node);
+				continue;
+			}
+
+			/* Everything else is a MIGRATION mle */
+
+			/* The rule for MIGRATION mles is that the master
+			 * becomes UNKNOWN if *either* the original or the new
+			 * master dies. All UNKNOWN lockres' are sent to
+			 * whichever node becomes the recovery master. The new
+			 * master is responsible for determining if there is
+			 * still a master for this lockres, or if he needs to
+			 * take over mastery. Either way, this node should
+			 * expect another message to resolve this. */
+
+			if (mle->master != dead_node &&
+			    mle->new_master != dead_node)
+				continue;
+
+			/* If we have reached this point, this mle needs to be
+			 * removed from the list and freed. */
+			dlm_clean_migration_mle(dlm, mle);
+
+			mlog(0, "%s: node %u died during migration from "
+			     "%u to %u!\n", dlm->name, dead_node, mle->master,
+			     mle->new_master);
+
+			/* If we find a lockres associated with the mle, we've
+			 * hit this rare case that messes up our lock ordering.
+			 * If so, we need to drop the master lock so that we can
+			 * take the lockres lock, meaning that we will have to
+			 * restart from the head of list. */
+			res = dlm_reset_mleres_owner(dlm, mle);
+			if (res)
+				/* restart */
+				goto top;
+
+			/* This may be the last reference */
+			__dlm_put_mle(mle);
+		}
+	}
+	spin_unlock(&dlm->master_lock);
+}
+
+int dlm_finish_migration(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+			 u8 old_master)
+{
+	struct dlm_node_iter iter;
+	int ret = 0;
+
+	spin_lock(&dlm->spinlock);
+	dlm_node_iter_init(dlm->domain_map, &iter);
+	clear_bit(old_master, iter.node_map);
+	clear_bit(dlm->node_num, iter.node_map);
+	spin_unlock(&dlm->spinlock);
+
+	/* ownership of the lockres is changing.  account for the
+	 * mastery reference here since old_master will briefly have
+	 * a reference after the migration completes */
+	spin_lock(&res->spinlock);
+	dlm_lockres_set_refmap_bit(dlm, res, old_master);
+	spin_unlock(&res->spinlock);
+
+	mlog(0, "now time to do a migrate request to other nodes\n");
+	ret = dlm_do_migrate_request(dlm, res, old_master,
+				     dlm->node_num, &iter);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	mlog(0, "doing assert master of %.*s to all except the original node\n",
+	     res->lockname.len, res->lockname.name);
+	/* this call now finishes out the nodemap
+	 * even if one or more nodes die */
+	ret = dlm_do_assert_master(dlm, res, iter.node_map,
+				   DLM_ASSERT_MASTER_FINISH_MIGRATION);
+	if (ret < 0) {
+		/* no longer need to retry.  all living nodes contacted. */
+		mlog_errno(ret);
+		ret = 0;
+	}
+
+	memset(iter.node_map, 0, sizeof(iter.node_map));
+	set_bit(old_master, iter.node_map);
+	mlog(0, "doing assert master of %.*s back to %u\n",
+	     res->lockname.len, res->lockname.name, old_master);
+	ret = dlm_do_assert_master(dlm, res, iter.node_map,
+				   DLM_ASSERT_MASTER_FINISH_MIGRATION);
+	if (ret < 0) {
+		mlog(0, "assert master to original master failed "
+		     "with %d.\n", ret);
+		/* the only nonzero status here would be because of
+		 * a dead original node.  we're done. */
+		ret = 0;
+	}
+
+	/* all done, set the owner, clear the flag */
+	spin_lock(&res->spinlock);
+	dlm_set_lockres_owner(dlm, res, dlm->node_num);
+	res->state &= ~DLM_LOCK_RES_MIGRATING;
+	spin_unlock(&res->spinlock);
+	/* re-dirty it on the new master */
+	dlm_kick_thread(dlm, res);
+	wake_up(&res->wq);
+leave:
+	return ret;
+}
+
+/*
+ * LOCKRES AST REFCOUNT
+ * this is integral to migration
+ */
+
+/* for future intent to call an ast, reserve one ahead of time.
+ * this should be called only after waiting on the lockres
+ * with dlm_wait_on_lockres, and while still holding the
+ * spinlock after the call. */
+void __dlm_lockres_reserve_ast(struct dlm_lock_resource *res)
+{
+	assert_spin_locked(&res->spinlock);
+	if (res->state & DLM_LOCK_RES_MIGRATING) {
+		__dlm_print_one_lock_resource(res);
+	}
+	BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
+
+	atomic_inc(&res->asts_reserved);
+}
+
+/*
+ * used to drop the reserved ast, either because it went unused,
+ * or because the ast/bast was actually called.
+ *
+ * also, if there is a pending migration on this lockres,
+ * and this was the last pending ast on the lockres,
+ * atomically set the MIGRATING flag before we drop the lock.
+ * this is how we ensure that migration can proceed with no
+ * asts in progress.  note that it is ok if the state of the
+ * queues is such that a lock should be granted in the future
+ * or that a bast should be fired, because the new master will
+ * shuffle the lists on this lockres as soon as it is migrated.
+ */
+void dlm_lockres_release_ast(struct dlm_ctxt *dlm,
+			     struct dlm_lock_resource *res)
+{
+	if (!atomic_dec_and_lock(&res->asts_reserved, &res->spinlock))
+		return;
+
+	if (!res->migration_pending) {
+		spin_unlock(&res->spinlock);
+		return;
+	}
+
+	BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
+	res->migration_pending = 0;
+	res->state |= DLM_LOCK_RES_MIGRATING;
+	spin_unlock(&res->spinlock);
+	wake_up(&res->wq);
+	wake_up(&dlm->migration_wq);
+}
+
+void dlm_force_free_mles(struct dlm_ctxt *dlm)
+{
+	int i;
+	struct hlist_head *bucket;
+	struct dlm_master_list_entry *mle;
+	struct hlist_node *tmp;
+
+	/*
+	 * We notified all other nodes that we are exiting the domain and
+	 * marked the dlm state to DLM_CTXT_LEAVING. If any mles are still
+	 * around we force free them and wake any processes that are waiting
+	 * on the mles
+	 */
+	spin_lock(&dlm->spinlock);
+	spin_lock(&dlm->master_lock);
+
+	BUG_ON(dlm->dlm_state != DLM_CTXT_LEAVING);
+	BUG_ON((find_next_bit(dlm->domain_map, O2NM_MAX_NODES, 0) < O2NM_MAX_NODES));
+
+	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
+		bucket = dlm_master_hash(dlm, i);
+		hlist_for_each_entry_safe(mle, tmp, bucket, master_hash_node) {
+			if (mle->type != DLM_MLE_BLOCK) {
+				mlog(ML_ERROR, "bad mle: %p\n", mle);
+				dlm_print_one_mle(mle);
+			}
+			atomic_set(&mle->woken, 1);
+			wake_up(&mle->wq);
+
+			__dlm_unlink_mle(dlm, mle);
+			__dlm_mle_detach_hb_events(dlm, mle);
+			__dlm_put_mle(mle);
+		}
+	}
+	spin_unlock(&dlm->master_lock);
+	spin_unlock(&dlm->spinlock);
+}
diff --git a/fs/ocfs2/dlm/dlmrecovery.c b/fs/ocfs2/dlm/dlmrecovery.c
new file mode 100644
index 000000000..ce12e0b1a
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmrecovery.c
@@ -0,0 +1,2935 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmrecovery.c
+ *
+ * recovery stuff
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/sysctl.h>
+#include <linux/random.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/timer.h>
+#include <linux/kthread.h>
+#include <linux/delay.h>
+
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+#include "dlmdomain.h"
+
+#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_RECOVERY)
+#include "cluster/masklog.h"
+
+static void dlm_do_local_recovery_cleanup(struct dlm_ctxt *dlm, u8 dead_node);
+
+static int dlm_recovery_thread(void *data);
+static int dlm_do_recovery(struct dlm_ctxt *dlm);
+
+static int dlm_pick_recovery_master(struct dlm_ctxt *dlm);
+static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node);
+static int dlm_init_recovery_area(struct dlm_ctxt *dlm, u8 dead_node);
+static int dlm_request_all_locks(struct dlm_ctxt *dlm,
+				 u8 request_from, u8 dead_node);
+static void dlm_destroy_recovery_area(struct dlm_ctxt *dlm, u8 dead_node);
+
+static inline int dlm_num_locks_in_lockres(struct dlm_lock_resource *res);
+static void dlm_init_migratable_lockres(struct dlm_migratable_lockres *mres,
+					const char *lockname, int namelen,
+					int total_locks, u64 cookie,
+					u8 flags, u8 master);
+static int dlm_send_mig_lockres_msg(struct dlm_ctxt *dlm,
+				    struct dlm_migratable_lockres *mres,
+				    u8 send_to,
+				    struct dlm_lock_resource *res,
+				    int total_locks);
+static int dlm_process_recovery_data(struct dlm_ctxt *dlm,
+				     struct dlm_lock_resource *res,
+				     struct dlm_migratable_lockres *mres);
+static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm);
+static int dlm_send_all_done_msg(struct dlm_ctxt *dlm,
+				 u8 dead_node, u8 send_to);
+static int dlm_send_begin_reco_message(struct dlm_ctxt *dlm, u8 dead_node);
+static void dlm_move_reco_locks_to_list(struct dlm_ctxt *dlm,
+					struct list_head *list, u8 dead_node);
+static void dlm_finish_local_lockres_recovery(struct dlm_ctxt *dlm,
+					      u8 dead_node, u8 new_master);
+static void dlm_reco_ast(void *astdata);
+static void dlm_reco_bast(void *astdata, int blocked_type);
+static void dlm_reco_unlock_ast(void *astdata, enum dlm_status st);
+static void dlm_request_all_locks_worker(struct dlm_work_item *item,
+					 void *data);
+static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data);
+static int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
+				      struct dlm_lock_resource *res,
+				      u8 *real_master);
+
+static u64 dlm_get_next_mig_cookie(void);
+
+static DEFINE_SPINLOCK(dlm_reco_state_lock);
+static DEFINE_SPINLOCK(dlm_mig_cookie_lock);
+static u64 dlm_mig_cookie = 1;
+
+static u64 dlm_get_next_mig_cookie(void)
+{
+	u64 c;
+	spin_lock(&dlm_mig_cookie_lock);
+	c = dlm_mig_cookie;
+	if (dlm_mig_cookie == (~0ULL))
+		dlm_mig_cookie = 1;
+	else
+		dlm_mig_cookie++;
+	spin_unlock(&dlm_mig_cookie_lock);
+	return c;
+}
+
+static inline void dlm_set_reco_dead_node(struct dlm_ctxt *dlm,
+					  u8 dead_node)
+{
+	assert_spin_locked(&dlm->spinlock);
+	if (dlm->reco.dead_node != dead_node)
+		mlog(0, "%s: changing dead_node from %u to %u\n",
+		     dlm->name, dlm->reco.dead_node, dead_node);
+	dlm->reco.dead_node = dead_node;
+}
+
+static inline void dlm_set_reco_master(struct dlm_ctxt *dlm,
+				       u8 master)
+{
+	assert_spin_locked(&dlm->spinlock);
+	mlog(0, "%s: changing new_master from %u to %u\n",
+	     dlm->name, dlm->reco.new_master, master);
+	dlm->reco.new_master = master;
+}
+
+static inline void __dlm_reset_recovery(struct dlm_ctxt *dlm)
+{
+	assert_spin_locked(&dlm->spinlock);
+	clear_bit(dlm->reco.dead_node, dlm->recovery_map);
+	dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
+	dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM);
+}
+
+static inline void dlm_reset_recovery(struct dlm_ctxt *dlm)
+{
+	spin_lock(&dlm->spinlock);
+	__dlm_reset_recovery(dlm);
+	spin_unlock(&dlm->spinlock);
+}
+
+/* Worker function used during recovery. */
+void dlm_dispatch_work(struct work_struct *work)
+{
+	struct dlm_ctxt *dlm =
+		container_of(work, struct dlm_ctxt, dispatched_work);
+	LIST_HEAD(tmp_list);
+	struct dlm_work_item *item, *next;
+	dlm_workfunc_t *workfunc;
+	int tot=0;
+
+	spin_lock(&dlm->work_lock);
+	list_splice_init(&dlm->work_list, &tmp_list);
+	spin_unlock(&dlm->work_lock);
+
+	list_for_each_entry(item, &tmp_list, list) {
+		tot++;
+	}
+	mlog(0, "%s: work thread has %d work items\n", dlm->name, tot);
+
+	list_for_each_entry_safe(item, next, &tmp_list, list) {
+		workfunc = item->func;
+		list_del_init(&item->list);
+
+		/* already have ref on dlm to avoid having
+		 * it disappear.  just double-check. */
+		BUG_ON(item->dlm != dlm);
+
+		/* this is allowed to sleep and
+		 * call network stuff */
+		workfunc(item, item->data);
+
+		dlm_put(dlm);
+		kfree(item);
+	}
+}
+
+/*
+ * RECOVERY THREAD
+ */
+
+void dlm_kick_recovery_thread(struct dlm_ctxt *dlm)
+{
+	/* wake the recovery thread
+	 * this will wake the reco thread in one of three places
+	 * 1) sleeping with no recovery happening
+	 * 2) sleeping with recovery mastered elsewhere
+	 * 3) recovery mastered here, waiting on reco data */
+
+	wake_up(&dlm->dlm_reco_thread_wq);
+}
+
+/* Launch the recovery thread */
+int dlm_launch_recovery_thread(struct dlm_ctxt *dlm)
+{
+	mlog(0, "starting dlm recovery thread...\n");
+
+	dlm->dlm_reco_thread_task = kthread_run(dlm_recovery_thread, dlm,
+						"dlm_reco_thread");
+	if (IS_ERR(dlm->dlm_reco_thread_task)) {
+		mlog_errno(PTR_ERR(dlm->dlm_reco_thread_task));
+		dlm->dlm_reco_thread_task = NULL;
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+void dlm_complete_recovery_thread(struct dlm_ctxt *dlm)
+{
+	if (dlm->dlm_reco_thread_task) {
+		mlog(0, "waiting for dlm recovery thread to exit\n");
+		kthread_stop(dlm->dlm_reco_thread_task);
+		dlm->dlm_reco_thread_task = NULL;
+	}
+}
+
+
+
+/*
+ * this is lame, but here's how recovery works...
+ * 1) all recovery threads cluster wide will work on recovering
+ *    ONE node at a time
+ * 2) negotiate who will take over all the locks for the dead node.
+ *    thats right... ALL the locks.
+ * 3) once a new master is chosen, everyone scans all locks
+ *    and moves aside those mastered by the dead guy
+ * 4) each of these locks should be locked until recovery is done
+ * 5) the new master collects up all of secondary lock queue info
+ *    one lock at a time, forcing each node to communicate back
+ *    before continuing
+ * 6) each secondary lock queue responds with the full known lock info
+ * 7) once the new master has run all its locks, it sends a ALLDONE!
+ *    message to everyone
+ * 8) upon receiving this message, the secondary queue node unlocks
+ *    and responds to the ALLDONE
+ * 9) once the new master gets responses from everyone, he unlocks
+ *    everything and recovery for this dead node is done
+ *10) go back to 2) while there are still dead nodes
+ *
+ */
+
+static void dlm_print_reco_node_status(struct dlm_ctxt *dlm)
+{
+	struct dlm_reco_node_data *ndata;
+	struct dlm_lock_resource *res;
+
+	mlog(ML_NOTICE, "%s(%d): recovery info, state=%s, dead=%u, master=%u\n",
+	     dlm->name, task_pid_nr(dlm->dlm_reco_thread_task),
+	     dlm->reco.state & DLM_RECO_STATE_ACTIVE ? "ACTIVE" : "inactive",
+	     dlm->reco.dead_node, dlm->reco.new_master);
+
+	list_for_each_entry(ndata, &dlm->reco.node_data, list) {
+		char *st = "unknown";
+		switch (ndata->state) {
+			case DLM_RECO_NODE_DATA_INIT:
+				st = "init";
+				break;
+			case DLM_RECO_NODE_DATA_REQUESTING:
+				st = "requesting";
+				break;
+			case DLM_RECO_NODE_DATA_DEAD:
+				st = "dead";
+				break;
+			case DLM_RECO_NODE_DATA_RECEIVING:
+				st = "receiving";
+				break;
+			case DLM_RECO_NODE_DATA_REQUESTED:
+				st = "requested";
+				break;
+			case DLM_RECO_NODE_DATA_DONE:
+				st = "done";
+				break;
+			case DLM_RECO_NODE_DATA_FINALIZE_SENT:
+				st = "finalize-sent";
+				break;
+			default:
+				st = "bad";
+				break;
+		}
+		mlog(ML_NOTICE, "%s: reco state, node %u, state=%s\n",
+		     dlm->name, ndata->node_num, st);
+	}
+	list_for_each_entry(res, &dlm->reco.resources, recovering) {
+		mlog(ML_NOTICE, "%s: lockres %.*s on recovering list\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
+	}
+}
+
+#define DLM_RECO_THREAD_TIMEOUT_MS (5 * 1000)
+
+static int dlm_recovery_thread(void *data)
+{
+	int status;
+	struct dlm_ctxt *dlm = data;
+	unsigned long timeout = msecs_to_jiffies(DLM_RECO_THREAD_TIMEOUT_MS);
+
+	mlog(0, "dlm thread running for %s...\n", dlm->name);
+
+	while (!kthread_should_stop()) {
+		if (dlm_domain_fully_joined(dlm)) {
+			status = dlm_do_recovery(dlm);
+			if (status == -EAGAIN) {
+				/* do not sleep, recheck immediately. */
+				continue;
+			}
+			if (status < 0)
+				mlog_errno(status);
+		}
+
+		wait_event_interruptible_timeout(dlm->dlm_reco_thread_wq,
+						 kthread_should_stop(),
+						 timeout);
+	}
+
+	mlog(0, "quitting DLM recovery thread\n");
+	return 0;
+}
+
+/* returns true when the recovery master has contacted us */
+static int dlm_reco_master_ready(struct dlm_ctxt *dlm)
+{
+	int ready;
+	spin_lock(&dlm->spinlock);
+	ready = (dlm->reco.new_master != O2NM_INVALID_NODE_NUM);
+	spin_unlock(&dlm->spinlock);
+	return ready;
+}
+
+/* returns true if node is no longer in the domain
+ * could be dead or just not joined */
+int dlm_is_node_dead(struct dlm_ctxt *dlm, u8 node)
+{
+	int dead;
+	spin_lock(&dlm->spinlock);
+	dead = !test_bit(node, dlm->domain_map);
+	spin_unlock(&dlm->spinlock);
+	return dead;
+}
+
+/* returns true if node is no longer in the domain
+ * could be dead or just not joined */
+static int dlm_is_node_recovered(struct dlm_ctxt *dlm, u8 node)
+{
+	int recovered;
+	spin_lock(&dlm->spinlock);
+	recovered = !test_bit(node, dlm->recovery_map);
+	spin_unlock(&dlm->spinlock);
+	return recovered;
+}
+
+
+void dlm_wait_for_node_death(struct dlm_ctxt *dlm, u8 node, int timeout)
+{
+	if (dlm_is_node_dead(dlm, node))
+		return;
+
+	printk(KERN_NOTICE "o2dlm: Waiting on the death of node %u in "
+	       "domain %s\n", node, dlm->name);
+
+	if (timeout)
+		wait_event_timeout(dlm->dlm_reco_thread_wq,
+				   dlm_is_node_dead(dlm, node),
+				   msecs_to_jiffies(timeout));
+	else
+		wait_event(dlm->dlm_reco_thread_wq,
+			   dlm_is_node_dead(dlm, node));
+}
+
+void dlm_wait_for_node_recovery(struct dlm_ctxt *dlm, u8 node, int timeout)
+{
+	if (dlm_is_node_recovered(dlm, node))
+		return;
+
+	printk(KERN_NOTICE "o2dlm: Waiting on the recovery of node %u in "
+	       "domain %s\n", node, dlm->name);
+
+	if (timeout)
+		wait_event_timeout(dlm->dlm_reco_thread_wq,
+				   dlm_is_node_recovered(dlm, node),
+				   msecs_to_jiffies(timeout));
+	else
+		wait_event(dlm->dlm_reco_thread_wq,
+			   dlm_is_node_recovered(dlm, node));
+}
+
+/* callers of the top-level api calls (dlmlock/dlmunlock) should
+ * block on the dlm->reco.event when recovery is in progress.
+ * the dlm recovery thread will set this state when it begins
+ * recovering a dead node (as the new master or not) and clear
+ * the state and wake as soon as all affected lock resources have
+ * been marked with the RECOVERY flag */
+static int dlm_in_recovery(struct dlm_ctxt *dlm)
+{
+	int in_recovery;
+	spin_lock(&dlm->spinlock);
+	in_recovery = !!(dlm->reco.state & DLM_RECO_STATE_ACTIVE);
+	spin_unlock(&dlm->spinlock);
+	return in_recovery;
+}
+
+
+void dlm_wait_for_recovery(struct dlm_ctxt *dlm)
+{
+	if (dlm_in_recovery(dlm)) {
+		mlog(0, "%s: reco thread %d in recovery: "
+		     "state=%d, master=%u, dead=%u\n",
+		     dlm->name, task_pid_nr(dlm->dlm_reco_thread_task),
+		     dlm->reco.state, dlm->reco.new_master,
+		     dlm->reco.dead_node);
+	}
+	wait_event(dlm->reco.event, !dlm_in_recovery(dlm));
+}
+
+static void dlm_begin_recovery(struct dlm_ctxt *dlm)
+{
+	spin_lock(&dlm->spinlock);
+	BUG_ON(dlm->reco.state & DLM_RECO_STATE_ACTIVE);
+	printk(KERN_NOTICE "o2dlm: Begin recovery on domain %s for node %u\n",
+	       dlm->name, dlm->reco.dead_node);
+	dlm->reco.state |= DLM_RECO_STATE_ACTIVE;
+	spin_unlock(&dlm->spinlock);
+}
+
+static void dlm_end_recovery(struct dlm_ctxt *dlm)
+{
+	spin_lock(&dlm->spinlock);
+	BUG_ON(!(dlm->reco.state & DLM_RECO_STATE_ACTIVE));
+	dlm->reco.state &= ~DLM_RECO_STATE_ACTIVE;
+	spin_unlock(&dlm->spinlock);
+	printk(KERN_NOTICE "o2dlm: End recovery on domain %s\n", dlm->name);
+	wake_up(&dlm->reco.event);
+}
+
+static void dlm_print_recovery_master(struct dlm_ctxt *dlm)
+{
+	printk(KERN_NOTICE "o2dlm: Node %u (%s) is the Recovery Master for the "
+	       "dead node %u in domain %s\n", dlm->reco.new_master,
+	       (dlm->node_num == dlm->reco.new_master ? "me" : "he"),
+	       dlm->reco.dead_node, dlm->name);
+}
+
+static int dlm_do_recovery(struct dlm_ctxt *dlm)
+{
+	int status = 0;
+	int ret;
+
+	spin_lock(&dlm->spinlock);
+
+	/* check to see if the new master has died */
+	if (dlm->reco.new_master != O2NM_INVALID_NODE_NUM &&
+	    test_bit(dlm->reco.new_master, dlm->recovery_map)) {
+		mlog(0, "new master %u died while recovering %u!\n",
+		     dlm->reco.new_master, dlm->reco.dead_node);
+		/* unset the new_master, leave dead_node */
+		dlm_set_reco_master(dlm, O2NM_INVALID_NODE_NUM);
+	}
+
+	/* select a target to recover */
+	if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) {
+		int bit;
+
+		bit = find_next_bit (dlm->recovery_map, O2NM_MAX_NODES, 0);
+		if (bit >= O2NM_MAX_NODES || bit < 0)
+			dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
+		else
+			dlm_set_reco_dead_node(dlm, bit);
+	} else if (!test_bit(dlm->reco.dead_node, dlm->recovery_map)) {
+		/* BUG? */
+		mlog(ML_ERROR, "dead_node %u no longer in recovery map!\n",
+		     dlm->reco.dead_node);
+		dlm_set_reco_dead_node(dlm, O2NM_INVALID_NODE_NUM);
+	}
+
+	if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) {
+		// mlog(0, "nothing to recover!  sleeping now!\n");
+		spin_unlock(&dlm->spinlock);
+		/* return to main thread loop and sleep. */
+		return 0;
+	}
+	mlog(0, "%s(%d):recovery thread found node %u in the recovery map!\n",
+	     dlm->name, task_pid_nr(dlm->dlm_reco_thread_task),
+	     dlm->reco.dead_node);
+	spin_unlock(&dlm->spinlock);
+
+	/* take write barrier */
+	/* (stops the list reshuffling thread, proxy ast handling) */
+	dlm_begin_recovery(dlm);
+
+	if (dlm->reco.new_master == dlm->node_num)
+		goto master_here;
+
+	if (dlm->reco.new_master == O2NM_INVALID_NODE_NUM) {
+		/* choose a new master, returns 0 if this node
+		 * is the master, -EEXIST if it's another node.
+		 * this does not return until a new master is chosen
+		 * or recovery completes entirely. */
+		ret = dlm_pick_recovery_master(dlm);
+		if (!ret) {
+			/* already notified everyone.  go. */
+			goto master_here;
+		}
+		mlog(0, "another node will master this recovery session.\n");
+	}
+
+	dlm_print_recovery_master(dlm);
+
+	/* it is safe to start everything back up here
+	 * because all of the dead node's lock resources
+	 * have been marked as in-recovery */
+	dlm_end_recovery(dlm);
+
+	/* sleep out in main dlm_recovery_thread loop. */
+	return 0;
+
+master_here:
+	dlm_print_recovery_master(dlm);
+
+	status = dlm_remaster_locks(dlm, dlm->reco.dead_node);
+	if (status < 0) {
+		/* we should never hit this anymore */
+		mlog(ML_ERROR, "%s: Error %d remastering locks for node %u, "
+		     "retrying.\n", dlm->name, status, dlm->reco.dead_node);
+		/* yield a bit to allow any final network messages
+		 * to get handled on remaining nodes */
+		msleep(100);
+	} else {
+		/* success!  see if any other nodes need recovery */
+		mlog(0, "DONE mastering recovery of %s:%u here(this=%u)!\n",
+		     dlm->name, dlm->reco.dead_node, dlm->node_num);
+		spin_lock(&dlm->spinlock);
+		__dlm_reset_recovery(dlm);
+		dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
+		spin_unlock(&dlm->spinlock);
+	}
+	dlm_end_recovery(dlm);
+
+	/* continue and look for another dead node */
+	return -EAGAIN;
+}
+
+static int dlm_remaster_locks(struct dlm_ctxt *dlm, u8 dead_node)
+{
+	int status = 0;
+	struct dlm_reco_node_data *ndata;
+	int all_nodes_done;
+	int destroy = 0;
+	int pass = 0;
+
+	do {
+		/* we have become recovery master.  there is no escaping
+		 * this, so just keep trying until we get it. */
+		status = dlm_init_recovery_area(dlm, dead_node);
+		if (status < 0) {
+			mlog(ML_ERROR, "%s: failed to alloc recovery area, "
+			     "retrying\n", dlm->name);
+			msleep(1000);
+		}
+	} while (status != 0);
+
+	/* safe to access the node data list without a lock, since this
+	 * process is the only one to change the list */
+	list_for_each_entry(ndata, &dlm->reco.node_data, list) {
+		BUG_ON(ndata->state != DLM_RECO_NODE_DATA_INIT);
+		ndata->state = DLM_RECO_NODE_DATA_REQUESTING;
+
+		mlog(0, "%s: Requesting lock info from node %u\n", dlm->name,
+		     ndata->node_num);
+
+		if (ndata->node_num == dlm->node_num) {
+			ndata->state = DLM_RECO_NODE_DATA_DONE;
+			continue;
+		}
+
+		do {
+			status = dlm_request_all_locks(dlm, ndata->node_num,
+						       dead_node);
+			if (status < 0) {
+				mlog_errno(status);
+				if (dlm_is_host_down(status)) {
+					/* node died, ignore it for recovery */
+					status = 0;
+					ndata->state = DLM_RECO_NODE_DATA_DEAD;
+					/* wait for the domain map to catch up
+					 * with the network state. */
+					wait_event_timeout(dlm->dlm_reco_thread_wq,
+							   dlm_is_node_dead(dlm,
+								ndata->node_num),
+							   msecs_to_jiffies(1000));
+					mlog(0, "waited 1 sec for %u, "
+					     "dead? %s\n", ndata->node_num,
+					     dlm_is_node_dead(dlm, ndata->node_num) ?
+					     "yes" : "no");
+				} else {
+					/* -ENOMEM on the other node */
+					mlog(0, "%s: node %u returned "
+					     "%d during recovery, retrying "
+					     "after a short wait\n",
+					     dlm->name, ndata->node_num,
+					     status);
+					msleep(100);
+				}
+			}
+		} while (status != 0);
+
+		spin_lock(&dlm_reco_state_lock);
+		switch (ndata->state) {
+			case DLM_RECO_NODE_DATA_INIT:
+			case DLM_RECO_NODE_DATA_FINALIZE_SENT:
+			case DLM_RECO_NODE_DATA_REQUESTED:
+				BUG();
+				break;
+			case DLM_RECO_NODE_DATA_DEAD:
+				mlog(0, "node %u died after requesting "
+				     "recovery info for node %u\n",
+				     ndata->node_num, dead_node);
+				/* fine.  don't need this node's info.
+				 * continue without it. */
+				break;
+			case DLM_RECO_NODE_DATA_REQUESTING:
+				ndata->state = DLM_RECO_NODE_DATA_REQUESTED;
+				mlog(0, "now receiving recovery data from "
+				     "node %u for dead node %u\n",
+				     ndata->node_num, dead_node);
+				break;
+			case DLM_RECO_NODE_DATA_RECEIVING:
+				mlog(0, "already receiving recovery data from "
+				     "node %u for dead node %u\n",
+				     ndata->node_num, dead_node);
+				break;
+			case DLM_RECO_NODE_DATA_DONE:
+				mlog(0, "already DONE receiving recovery data "
+				     "from node %u for dead node %u\n",
+				     ndata->node_num, dead_node);
+				break;
+		}
+		spin_unlock(&dlm_reco_state_lock);
+	}
+
+	mlog(0, "%s: Done requesting all lock info\n", dlm->name);
+
+	/* nodes should be sending reco data now
+	 * just need to wait */
+
+	while (1) {
+		/* check all the nodes now to see if we are
+		 * done, or if anyone died */
+		all_nodes_done = 1;
+		spin_lock(&dlm_reco_state_lock);
+		list_for_each_entry(ndata, &dlm->reco.node_data, list) {
+			mlog(0, "checking recovery state of node %u\n",
+			     ndata->node_num);
+			switch (ndata->state) {
+				case DLM_RECO_NODE_DATA_INIT:
+				case DLM_RECO_NODE_DATA_REQUESTING:
+					mlog(ML_ERROR, "bad ndata state for "
+					     "node %u: state=%d\n",
+					     ndata->node_num, ndata->state);
+					BUG();
+					break;
+				case DLM_RECO_NODE_DATA_DEAD:
+					mlog(0, "node %u died after "
+					     "requesting recovery info for "
+					     "node %u\n", ndata->node_num,
+					     dead_node);
+					break;
+				case DLM_RECO_NODE_DATA_RECEIVING:
+				case DLM_RECO_NODE_DATA_REQUESTED:
+					mlog(0, "%s: node %u still in state %s\n",
+					     dlm->name, ndata->node_num,
+					     ndata->state==DLM_RECO_NODE_DATA_RECEIVING ?
+					     "receiving" : "requested");
+					all_nodes_done = 0;
+					break;
+				case DLM_RECO_NODE_DATA_DONE:
+					mlog(0, "%s: node %u state is done\n",
+					     dlm->name, ndata->node_num);
+					break;
+				case DLM_RECO_NODE_DATA_FINALIZE_SENT:
+					mlog(0, "%s: node %u state is finalize\n",
+					     dlm->name, ndata->node_num);
+					break;
+			}
+		}
+		spin_unlock(&dlm_reco_state_lock);
+
+		mlog(0, "pass #%d, all_nodes_done?: %s\n", ++pass,
+		     all_nodes_done?"yes":"no");
+		if (all_nodes_done) {
+			int ret;
+
+			/* Set this flag on recovery master to avoid
+			 * a new recovery for another dead node start
+			 * before the recovery is not done. That may
+			 * cause recovery hung.*/
+			spin_lock(&dlm->spinlock);
+			dlm->reco.state |= DLM_RECO_STATE_FINALIZE;
+			spin_unlock(&dlm->spinlock);
+
+			/* all nodes are now in DLM_RECO_NODE_DATA_DONE state
+	 		 * just send a finalize message to everyone and
+	 		 * clean up */
+			mlog(0, "all nodes are done! send finalize\n");
+			ret = dlm_send_finalize_reco_message(dlm);
+			if (ret < 0)
+				mlog_errno(ret);
+
+			spin_lock(&dlm->spinlock);
+			dlm_finish_local_lockres_recovery(dlm, dead_node,
+							  dlm->node_num);
+			spin_unlock(&dlm->spinlock);
+			mlog(0, "should be done with recovery!\n");
+
+			mlog(0, "finishing recovery of %s at %lu, "
+			     "dead=%u, this=%u, new=%u\n", dlm->name,
+			     jiffies, dlm->reco.dead_node,
+			     dlm->node_num, dlm->reco.new_master);
+			destroy = 1;
+			status = 0;
+			/* rescan everything marked dirty along the way */
+			dlm_kick_thread(dlm, NULL);
+			break;
+		}
+		/* wait to be signalled, with periodic timeout
+		 * to check for node death */
+		wait_event_interruptible_timeout(dlm->dlm_reco_thread_wq,
+					 kthread_should_stop(),
+					 msecs_to_jiffies(DLM_RECO_THREAD_TIMEOUT_MS));
+
+	}
+
+	if (destroy)
+		dlm_destroy_recovery_area(dlm, dead_node);
+
+	return status;
+}
+
+static int dlm_init_recovery_area(struct dlm_ctxt *dlm, u8 dead_node)
+{
+	int num=0;
+	struct dlm_reco_node_data *ndata;
+
+	spin_lock(&dlm->spinlock);
+	memcpy(dlm->reco.node_map, dlm->domain_map, sizeof(dlm->domain_map));
+	/* nodes can only be removed (by dying) after dropping
+	 * this lock, and death will be trapped later, so this should do */
+	spin_unlock(&dlm->spinlock);
+
+	while (1) {
+		num = find_next_bit (dlm->reco.node_map, O2NM_MAX_NODES, num);
+		if (num >= O2NM_MAX_NODES) {
+			break;
+		}
+		BUG_ON(num == dead_node);
+
+		ndata = kzalloc(sizeof(*ndata), GFP_NOFS);
+		if (!ndata) {
+			dlm_destroy_recovery_area(dlm, dead_node);
+			return -ENOMEM;
+		}
+		ndata->node_num = num;
+		ndata->state = DLM_RECO_NODE_DATA_INIT;
+		spin_lock(&dlm_reco_state_lock);
+		list_add_tail(&ndata->list, &dlm->reco.node_data);
+		spin_unlock(&dlm_reco_state_lock);
+		num++;
+	}
+
+	return 0;
+}
+
+static void dlm_destroy_recovery_area(struct dlm_ctxt *dlm, u8 dead_node)
+{
+	struct dlm_reco_node_data *ndata, *next;
+	LIST_HEAD(tmplist);
+
+	spin_lock(&dlm_reco_state_lock);
+	list_splice_init(&dlm->reco.node_data, &tmplist);
+	spin_unlock(&dlm_reco_state_lock);
+
+	list_for_each_entry_safe(ndata, next, &tmplist, list) {
+		list_del_init(&ndata->list);
+		kfree(ndata);
+	}
+}
+
+static int dlm_request_all_locks(struct dlm_ctxt *dlm, u8 request_from,
+				 u8 dead_node)
+{
+	struct dlm_lock_request lr;
+	int ret;
+	int status;
+
+	mlog(0, "\n");
+
+
+	mlog(0, "dlm_request_all_locks: dead node is %u, sending request "
+		  "to %u\n", dead_node, request_from);
+
+	memset(&lr, 0, sizeof(lr));
+	lr.node_idx = dlm->node_num;
+	lr.dead_node = dead_node;
+
+	// send message
+	ret = o2net_send_message(DLM_LOCK_REQUEST_MSG, dlm->key,
+				 &lr, sizeof(lr), request_from, &status);
+
+	/* negative status is handled by caller */
+	if (ret < 0)
+		mlog(ML_ERROR, "%s: Error %d send LOCK_REQUEST to node %u "
+		     "to recover dead node %u\n", dlm->name, ret,
+		     request_from, dead_node);
+	else
+		ret = status;
+	// return from here, then
+	// sleep until all received or error
+	return ret;
+
+}
+
+int dlm_request_all_locks_handler(struct o2net_msg *msg, u32 len, void *data,
+				  void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_lock_request *lr = (struct dlm_lock_request *)msg->buf;
+	char *buf = NULL;
+	struct dlm_work_item *item = NULL;
+
+	if (!dlm_grab(dlm))
+		return -EINVAL;
+
+	if (lr->dead_node != dlm->reco.dead_node) {
+		mlog(ML_ERROR, "%s: node %u sent dead_node=%u, but local "
+		     "dead_node is %u\n", dlm->name, lr->node_idx,
+		     lr->dead_node, dlm->reco.dead_node);
+		dlm_print_reco_node_status(dlm);
+		/* this is a hack */
+		dlm_put(dlm);
+		return -ENOMEM;
+	}
+	BUG_ON(lr->dead_node != dlm->reco.dead_node);
+
+	item = kzalloc(sizeof(*item), GFP_NOFS);
+	if (!item) {
+		dlm_put(dlm);
+		return -ENOMEM;
+	}
+
+	/* this will get freed by dlm_request_all_locks_worker */
+	buf = (char *) __get_free_page(GFP_NOFS);
+	if (!buf) {
+		kfree(item);
+		dlm_put(dlm);
+		return -ENOMEM;
+	}
+
+	/* queue up work for dlm_request_all_locks_worker */
+	dlm_grab(dlm);  /* get an extra ref for the work item */
+	dlm_init_work_item(dlm, item, dlm_request_all_locks_worker, buf);
+	item->u.ral.reco_master = lr->node_idx;
+	item->u.ral.dead_node = lr->dead_node;
+	spin_lock(&dlm->work_lock);
+	list_add_tail(&item->list, &dlm->work_list);
+	spin_unlock(&dlm->work_lock);
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
+
+	dlm_put(dlm);
+	return 0;
+}
+
+static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data)
+{
+	struct dlm_migratable_lockres *mres;
+	struct dlm_lock_resource *res;
+	struct dlm_ctxt *dlm;
+	LIST_HEAD(resources);
+	int ret;
+	u8 dead_node, reco_master;
+	int skip_all_done = 0;
+
+	dlm = item->dlm;
+	dead_node = item->u.ral.dead_node;
+	reco_master = item->u.ral.reco_master;
+	mres = (struct dlm_migratable_lockres *)data;
+
+	mlog(0, "%s: recovery worker started, dead=%u, master=%u\n",
+	     dlm->name, dead_node, reco_master);
+
+	if (dead_node != dlm->reco.dead_node ||
+	    reco_master != dlm->reco.new_master) {
+		/* worker could have been created before the recovery master
+		 * died.  if so, do not continue, but do not error. */
+		if (dlm->reco.new_master == O2NM_INVALID_NODE_NUM) {
+			mlog(ML_NOTICE, "%s: will not send recovery state, "
+			     "recovery master %u died, thread=(dead=%u,mas=%u)"
+			     " current=(dead=%u,mas=%u)\n", dlm->name,
+			     reco_master, dead_node, reco_master,
+			     dlm->reco.dead_node, dlm->reco.new_master);
+		} else {
+			mlog(ML_NOTICE, "%s: reco state invalid: reco(dead=%u, "
+			     "master=%u), request(dead=%u, master=%u)\n",
+			     dlm->name, dlm->reco.dead_node,
+			     dlm->reco.new_master, dead_node, reco_master);
+		}
+		goto leave;
+	}
+
+	/* lock resources should have already been moved to the
+ 	 * dlm->reco.resources list.  now move items from that list
+ 	 * to a temp list if the dead owner matches.  note that the
+	 * whole cluster recovers only one node at a time, so we
+	 * can safely move UNKNOWN lock resources for each recovery
+	 * session. */
+	dlm_move_reco_locks_to_list(dlm, &resources, dead_node);
+
+	/* now we can begin blasting lockreses without the dlm lock */
+
+	/* any errors returned will be due to the new_master dying,
+	 * the dlm_reco_thread should detect this */
+	list_for_each_entry(res, &resources, recovering) {
+		ret = dlm_send_one_lockres(dlm, res, mres, reco_master,
+				   	DLM_MRES_RECOVERY);
+		if (ret < 0) {
+			mlog(ML_ERROR, "%s: node %u went down while sending "
+			     "recovery state for dead node %u, ret=%d\n", dlm->name,
+			     reco_master, dead_node, ret);
+			skip_all_done = 1;
+			break;
+		}
+	}
+
+	/* move the resources back to the list */
+	spin_lock(&dlm->spinlock);
+	list_splice_init(&resources, &dlm->reco.resources);
+	spin_unlock(&dlm->spinlock);
+
+	if (!skip_all_done) {
+		ret = dlm_send_all_done_msg(dlm, dead_node, reco_master);
+		if (ret < 0) {
+			mlog(ML_ERROR, "%s: node %u went down while sending "
+			     "recovery all-done for dead node %u, ret=%d\n",
+			     dlm->name, reco_master, dead_node, ret);
+		}
+	}
+leave:
+	free_page((unsigned long)data);
+}
+
+
+static int dlm_send_all_done_msg(struct dlm_ctxt *dlm, u8 dead_node, u8 send_to)
+{
+	int ret, tmpret;
+	struct dlm_reco_data_done done_msg;
+
+	memset(&done_msg, 0, sizeof(done_msg));
+	done_msg.node_idx = dlm->node_num;
+	done_msg.dead_node = dead_node;
+	mlog(0, "sending DATA DONE message to %u, "
+	     "my node=%u, dead node=%u\n", send_to, done_msg.node_idx,
+	     done_msg.dead_node);
+
+	ret = o2net_send_message(DLM_RECO_DATA_DONE_MSG, dlm->key, &done_msg,
+				 sizeof(done_msg), send_to, &tmpret);
+	if (ret < 0) {
+		mlog(ML_ERROR, "%s: Error %d send RECO_DATA_DONE to node %u "
+		     "to recover dead node %u\n", dlm->name, ret, send_to,
+		     dead_node);
+		if (!dlm_is_host_down(ret)) {
+			BUG();
+		}
+	} else
+		ret = tmpret;
+	return ret;
+}
+
+
+int dlm_reco_data_done_handler(struct o2net_msg *msg, u32 len, void *data,
+			       void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_reco_data_done *done = (struct dlm_reco_data_done *)msg->buf;
+	struct dlm_reco_node_data *ndata = NULL;
+	int ret = -EINVAL;
+
+	if (!dlm_grab(dlm))
+		return -EINVAL;
+
+	mlog(0, "got DATA DONE: dead_node=%u, reco.dead_node=%u, "
+	     "node_idx=%u, this node=%u\n", done->dead_node,
+	     dlm->reco.dead_node, done->node_idx, dlm->node_num);
+
+	mlog_bug_on_msg((done->dead_node != dlm->reco.dead_node),
+			"Got DATA DONE: dead_node=%u, reco.dead_node=%u, "
+			"node_idx=%u, this node=%u\n", done->dead_node,
+			dlm->reco.dead_node, done->node_idx, dlm->node_num);
+
+	spin_lock(&dlm_reco_state_lock);
+	list_for_each_entry(ndata, &dlm->reco.node_data, list) {
+		if (ndata->node_num != done->node_idx)
+			continue;
+
+		switch (ndata->state) {
+			/* should have moved beyond INIT but not to FINALIZE yet */
+			case DLM_RECO_NODE_DATA_INIT:
+			case DLM_RECO_NODE_DATA_DEAD:
+			case DLM_RECO_NODE_DATA_FINALIZE_SENT:
+				mlog(ML_ERROR, "bad ndata state for node %u:"
+				     " state=%d\n", ndata->node_num,
+				     ndata->state);
+				BUG();
+				break;
+			/* these states are possible at this point, anywhere along
+			 * the line of recovery */
+			case DLM_RECO_NODE_DATA_DONE:
+			case DLM_RECO_NODE_DATA_RECEIVING:
+			case DLM_RECO_NODE_DATA_REQUESTED:
+			case DLM_RECO_NODE_DATA_REQUESTING:
+				mlog(0, "node %u is DONE sending "
+					  "recovery data!\n",
+					  ndata->node_num);
+
+				ndata->state = DLM_RECO_NODE_DATA_DONE;
+				ret = 0;
+				break;
+		}
+	}
+	spin_unlock(&dlm_reco_state_lock);
+
+	/* wake the recovery thread, some node is done */
+	if (!ret)
+		dlm_kick_recovery_thread(dlm);
+
+	if (ret < 0)
+		mlog(ML_ERROR, "failed to find recovery node data for node "
+		     "%u\n", done->node_idx);
+	dlm_put(dlm);
+
+	mlog(0, "leaving reco data done handler, ret=%d\n", ret);
+	return ret;
+}
+
+static void dlm_move_reco_locks_to_list(struct dlm_ctxt *dlm,
+					struct list_head *list,
+				       	u8 dead_node)
+{
+	struct dlm_lock_resource *res, *next;
+	struct dlm_lock *lock;
+
+	spin_lock(&dlm->spinlock);
+	list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) {
+		/* always prune any $RECOVERY entries for dead nodes,
+		 * otherwise hangs can occur during later recovery */
+		if (dlm_is_recovery_lock(res->lockname.name,
+					 res->lockname.len)) {
+			spin_lock(&res->spinlock);
+			list_for_each_entry(lock, &res->granted, list) {
+				if (lock->ml.node == dead_node) {
+					mlog(0, "AHA! there was "
+					     "a $RECOVERY lock for dead "
+					     "node %u (%s)!\n",
+					     dead_node, dlm->name);
+					list_del_init(&lock->list);
+					dlm_lock_put(lock);
+					/* Can't schedule DLM_UNLOCK_FREE_LOCK
+					 * - do manually */
+					dlm_lock_put(lock);
+					break;
+				}
+			}
+			spin_unlock(&res->spinlock);
+			continue;
+		}
+
+		if (res->owner == dead_node) {
+			mlog(0, "found lockres owned by dead node while "
+				  "doing recovery for node %u. sending it.\n",
+				  dead_node);
+			list_move_tail(&res->recovering, list);
+		} else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
+			mlog(0, "found UNKNOWN owner while doing recovery "
+				  "for node %u. sending it.\n", dead_node);
+			list_move_tail(&res->recovering, list);
+		}
+	}
+	spin_unlock(&dlm->spinlock);
+}
+
+static inline int dlm_num_locks_in_lockres(struct dlm_lock_resource *res)
+{
+	int total_locks = 0;
+	struct list_head *iter, *queue = &res->granted;
+	int i;
+
+	for (i=0; i<3; i++) {
+		list_for_each(iter, queue)
+			total_locks++;
+		queue++;
+	}
+	return total_locks;
+}
+
+
+static int dlm_send_mig_lockres_msg(struct dlm_ctxt *dlm,
+				      struct dlm_migratable_lockres *mres,
+				      u8 send_to,
+				      struct dlm_lock_resource *res,
+				      int total_locks)
+{
+	u64 mig_cookie = be64_to_cpu(mres->mig_cookie);
+	int mres_total_locks = be32_to_cpu(mres->total_locks);
+	int sz, ret = 0, status = 0;
+	u8 orig_flags = mres->flags,
+	   orig_master = mres->master;
+
+	BUG_ON(mres->num_locks > DLM_MAX_MIGRATABLE_LOCKS);
+	if (!mres->num_locks)
+		return 0;
+
+	sz = sizeof(struct dlm_migratable_lockres) +
+		(mres->num_locks * sizeof(struct dlm_migratable_lock));
+
+	/* add an all-done flag if we reached the last lock */
+	orig_flags = mres->flags;
+	BUG_ON(total_locks > mres_total_locks);
+	if (total_locks == mres_total_locks)
+		mres->flags |= DLM_MRES_ALL_DONE;
+
+	mlog(0, "%s:%.*s: sending mig lockres (%s) to %u\n",
+	     dlm->name, res->lockname.len, res->lockname.name,
+	     orig_flags & DLM_MRES_MIGRATION ? "migration" : "recovery",
+	     send_to);
+
+	/* send it */
+	ret = o2net_send_message(DLM_MIG_LOCKRES_MSG, dlm->key, mres,
+				 sz, send_to, &status);
+	if (ret < 0) {
+		/* XXX: negative status is not handled.
+		 * this will end up killing this node. */
+		mlog(ML_ERROR, "%s: res %.*s, Error %d send MIG_LOCKRES to "
+		     "node %u (%s)\n", dlm->name, mres->lockname_len,
+		     mres->lockname, ret, send_to,
+		     (orig_flags & DLM_MRES_MIGRATION ?
+		      "migration" : "recovery"));
+	} else {
+		/* might get an -ENOMEM back here */
+		ret = status;
+		if (ret < 0) {
+			mlog_errno(ret);
+
+			if (ret == -EFAULT) {
+				mlog(ML_ERROR, "node %u told me to kill "
+				     "myself!\n", send_to);
+				BUG();
+			}
+		}
+	}
+
+	/* zero and reinit the message buffer */
+	dlm_init_migratable_lockres(mres, res->lockname.name,
+				    res->lockname.len, mres_total_locks,
+				    mig_cookie, orig_flags, orig_master);
+	return ret;
+}
+
+static void dlm_init_migratable_lockres(struct dlm_migratable_lockres *mres,
+					const char *lockname, int namelen,
+					int total_locks, u64 cookie,
+					u8 flags, u8 master)
+{
+	/* mres here is one full page */
+	clear_page(mres);
+	mres->lockname_len = namelen;
+	memcpy(mres->lockname, lockname, namelen);
+	mres->num_locks = 0;
+	mres->total_locks = cpu_to_be32(total_locks);
+	mres->mig_cookie = cpu_to_be64(cookie);
+	mres->flags = flags;
+	mres->master = master;
+}
+
+static void dlm_prepare_lvb_for_migration(struct dlm_lock *lock,
+					  struct dlm_migratable_lockres *mres,
+					  int queue)
+{
+	if (!lock->lksb)
+	       return;
+
+	/* Ignore lvb in all locks in the blocked list */
+	if (queue == DLM_BLOCKED_LIST)
+		return;
+
+	/* Only consider lvbs in locks with granted EX or PR lock levels */
+	if (lock->ml.type != LKM_EXMODE && lock->ml.type != LKM_PRMODE)
+		return;
+
+	if (dlm_lvb_is_empty(mres->lvb)) {
+		memcpy(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN);
+		return;
+	}
+
+	/* Ensure the lvb copied for migration matches in other valid locks */
+	if (!memcmp(mres->lvb, lock->lksb->lvb, DLM_LVB_LEN))
+		return;
+
+	mlog(ML_ERROR, "Mismatched lvb in lock cookie=%u:%llu, name=%.*s, "
+	     "node=%u\n",
+	     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+	     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+	     lock->lockres->lockname.len, lock->lockres->lockname.name,
+	     lock->ml.node);
+	dlm_print_one_lock_resource(lock->lockres);
+	BUG();
+}
+
+/* returns 1 if this lock fills the network structure,
+ * 0 otherwise */
+static int dlm_add_lock_to_array(struct dlm_lock *lock,
+				 struct dlm_migratable_lockres *mres, int queue)
+{
+	struct dlm_migratable_lock *ml;
+	int lock_num = mres->num_locks;
+
+	ml = &(mres->ml[lock_num]);
+	ml->cookie = lock->ml.cookie;
+	ml->type = lock->ml.type;
+	ml->convert_type = lock->ml.convert_type;
+	ml->highest_blocked = lock->ml.highest_blocked;
+	ml->list = queue;
+	if (lock->lksb) {
+		ml->flags = lock->lksb->flags;
+		dlm_prepare_lvb_for_migration(lock, mres, queue);
+	}
+	ml->node = lock->ml.node;
+	mres->num_locks++;
+	/* we reached the max, send this network message */
+	if (mres->num_locks == DLM_MAX_MIGRATABLE_LOCKS)
+		return 1;
+	return 0;
+}
+
+static void dlm_add_dummy_lock(struct dlm_ctxt *dlm,
+			       struct dlm_migratable_lockres *mres)
+{
+	struct dlm_lock dummy;
+	memset(&dummy, 0, sizeof(dummy));
+	dummy.ml.cookie = 0;
+	dummy.ml.type = LKM_IVMODE;
+	dummy.ml.convert_type = LKM_IVMODE;
+	dummy.ml.highest_blocked = LKM_IVMODE;
+	dummy.lksb = NULL;
+	dummy.ml.node = dlm->node_num;
+	dlm_add_lock_to_array(&dummy, mres, DLM_BLOCKED_LIST);
+}
+
+static inline int dlm_is_dummy_lock(struct dlm_ctxt *dlm,
+				    struct dlm_migratable_lock *ml,
+				    u8 *nodenum)
+{
+	if (unlikely(ml->cookie == 0 &&
+	    ml->type == LKM_IVMODE &&
+	    ml->convert_type == LKM_IVMODE &&
+	    ml->highest_blocked == LKM_IVMODE &&
+	    ml->list == DLM_BLOCKED_LIST)) {
+		*nodenum = ml->node;
+		return 1;
+	}
+	return 0;
+}
+
+int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+			 struct dlm_migratable_lockres *mres,
+			 u8 send_to, u8 flags)
+{
+	struct list_head *queue;
+	int total_locks, i;
+	u64 mig_cookie = 0;
+	struct dlm_lock *lock;
+	int ret = 0;
+
+	BUG_ON(!(flags & (DLM_MRES_RECOVERY|DLM_MRES_MIGRATION)));
+
+	mlog(0, "sending to %u\n", send_to);
+
+	total_locks = dlm_num_locks_in_lockres(res);
+	if (total_locks > DLM_MAX_MIGRATABLE_LOCKS) {
+		/* rare, but possible */
+		mlog(0, "argh.  lockres has %d locks.  this will "
+			  "require more than one network packet to "
+			  "migrate\n", total_locks);
+		mig_cookie = dlm_get_next_mig_cookie();
+	}
+
+	dlm_init_migratable_lockres(mres, res->lockname.name,
+				    res->lockname.len, total_locks,
+				    mig_cookie, flags, res->owner);
+
+	total_locks = 0;
+	for (i=DLM_GRANTED_LIST; i<=DLM_BLOCKED_LIST; i++) {
+		queue = dlm_list_idx_to_ptr(res, i);
+		list_for_each_entry(lock, queue, list) {
+			/* add another lock. */
+			total_locks++;
+			if (!dlm_add_lock_to_array(lock, mres, i))
+				continue;
+
+			/* this filled the lock message,
+			 * we must send it immediately. */
+			ret = dlm_send_mig_lockres_msg(dlm, mres, send_to,
+						       res, total_locks);
+			if (ret < 0)
+				goto error;
+		}
+	}
+	if (total_locks == 0) {
+		/* send a dummy lock to indicate a mastery reference only */
+		mlog(0, "%s:%.*s: sending dummy lock to %u, %s\n",
+		     dlm->name, res->lockname.len, res->lockname.name,
+		     send_to, flags & DLM_MRES_RECOVERY ? "recovery" :
+		     "migration");
+		dlm_add_dummy_lock(dlm, mres);
+	}
+	/* flush any remaining locks */
+	ret = dlm_send_mig_lockres_msg(dlm, mres, send_to, res, total_locks);
+	if (ret < 0)
+		goto error;
+	return ret;
+
+error:
+	mlog(ML_ERROR, "%s: dlm_send_mig_lockres_msg returned %d\n",
+	     dlm->name, ret);
+	if (!dlm_is_host_down(ret))
+		BUG();
+	mlog(0, "%s: node %u went down while sending %s "
+	     "lockres %.*s\n", dlm->name, send_to,
+	     flags & DLM_MRES_RECOVERY ?  "recovery" : "migration",
+	     res->lockname.len, res->lockname.name);
+	return ret;
+}
+
+
+
+/*
+ * this message will contain no more than one page worth of
+ * recovery data, and it will work on only one lockres.
+ * there may be many locks in this page, and we may need to wait
+ * for additional packets to complete all the locks (rare, but
+ * possible).
+ */
+/*
+ * NOTE: the allocation error cases here are scary
+ * we really cannot afford to fail an alloc in recovery
+ * do we spin?  returning an error only delays the problem really
+ */
+
+int dlm_mig_lockres_handler(struct o2net_msg *msg, u32 len, void *data,
+			    void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_migratable_lockres *mres =
+		(struct dlm_migratable_lockres *)msg->buf;
+	int ret = 0;
+	u8 real_master;
+	u8 extra_refs = 0;
+	char *buf = NULL;
+	struct dlm_work_item *item = NULL;
+	struct dlm_lock_resource *res = NULL;
+
+	if (!dlm_grab(dlm))
+		return -EINVAL;
+
+	BUG_ON(!(mres->flags & (DLM_MRES_RECOVERY|DLM_MRES_MIGRATION)));
+
+	real_master = mres->master;
+	if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
+		/* cannot migrate a lockres with no master */
+		BUG_ON(!(mres->flags & DLM_MRES_RECOVERY));
+	}
+
+	mlog(0, "%s message received from node %u\n",
+		  (mres->flags & DLM_MRES_RECOVERY) ?
+		  "recovery" : "migration", mres->master);
+	if (mres->flags & DLM_MRES_ALL_DONE)
+		mlog(0, "all done flag.  all lockres data received!\n");
+
+	ret = -ENOMEM;
+	buf = kmalloc(be16_to_cpu(msg->data_len), GFP_NOFS);
+	item = kzalloc(sizeof(*item), GFP_NOFS);
+	if (!buf || !item)
+		goto leave;
+
+	/* lookup the lock to see if we have a secondary queue for this
+	 * already...  just add the locks in and this will have its owner
+	 * and RECOVERY flag changed when it completes. */
+	res = dlm_lookup_lockres(dlm, mres->lockname, mres->lockname_len);
+	if (res) {
+	 	/* this will get a ref on res */
+		/* mark it as recovering/migrating and hash it */
+		spin_lock(&res->spinlock);
+		if (mres->flags & DLM_MRES_RECOVERY) {
+			res->state |= DLM_LOCK_RES_RECOVERING;
+		} else {
+			if (res->state & DLM_LOCK_RES_MIGRATING) {
+				/* this is at least the second
+				 * lockres message */
+				mlog(0, "lock %.*s is already migrating\n",
+					  mres->lockname_len,
+					  mres->lockname);
+			} else if (res->state & DLM_LOCK_RES_RECOVERING) {
+				/* caller should BUG */
+				mlog(ML_ERROR, "node is attempting to migrate "
+				     "lock %.*s, but marked as recovering!\n",
+				     mres->lockname_len, mres->lockname);
+				ret = -EFAULT;
+				spin_unlock(&res->spinlock);
+				dlm_lockres_put(res);
+				goto leave;
+			}
+			res->state |= DLM_LOCK_RES_MIGRATING;
+		}
+		spin_unlock(&res->spinlock);
+	} else {
+		/* need to allocate, just like if it was
+		 * mastered here normally  */
+		res = dlm_new_lockres(dlm, mres->lockname, mres->lockname_len);
+		if (!res)
+			goto leave;
+
+		/* to match the ref that we would have gotten if
+		 * dlm_lookup_lockres had succeeded */
+		dlm_lockres_get(res);
+
+		/* mark it as recovering/migrating and hash it */
+		if (mres->flags & DLM_MRES_RECOVERY)
+			res->state |= DLM_LOCK_RES_RECOVERING;
+		else
+			res->state |= DLM_LOCK_RES_MIGRATING;
+
+		spin_lock(&dlm->spinlock);
+		__dlm_insert_lockres(dlm, res);
+		spin_unlock(&dlm->spinlock);
+
+		/* Add an extra ref for this lock-less lockres lest the
+		 * dlm_thread purges it before we get the chance to add
+		 * locks to it */
+		dlm_lockres_get(res);
+
+		/* There are three refs that need to be put.
+		 * 1. Taken above.
+		 * 2. kref_init in dlm_new_lockres()->dlm_init_lockres().
+		 * 3. dlm_lookup_lockres()
+		 * The first one is handled at the end of this function. The
+		 * other two are handled in the worker thread after locks have
+		 * been attached. Yes, we don't wait for purge time to match
+		 * kref_init. The lockres will still have atleast one ref
+		 * added because it is in the hash __dlm_insert_lockres() */
+		extra_refs++;
+
+		/* now that the new lockres is inserted,
+		 * make it usable by other processes */
+		spin_lock(&res->spinlock);
+		res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
+		spin_unlock(&res->spinlock);
+		wake_up(&res->wq);
+	}
+
+	/* at this point we have allocated everything we need,
+	 * and we have a hashed lockres with an extra ref and
+	 * the proper res->state flags. */
+	ret = 0;
+	spin_lock(&res->spinlock);
+	/* drop this either when master requery finds a different master
+	 * or when a lock is added by the recovery worker */
+	dlm_lockres_grab_inflight_ref(dlm, res);
+	if (mres->master == DLM_LOCK_RES_OWNER_UNKNOWN) {
+		/* migration cannot have an unknown master */
+		BUG_ON(!(mres->flags & DLM_MRES_RECOVERY));
+		mlog(0, "recovery has passed me a lockres with an "
+			  "unknown owner.. will need to requery: "
+			  "%.*s\n", mres->lockname_len, mres->lockname);
+	} else {
+		/* take a reference now to pin the lockres, drop it
+		 * when locks are added in the worker */
+		dlm_change_lockres_owner(dlm, res, dlm->node_num);
+	}
+	spin_unlock(&res->spinlock);
+
+	/* queue up work for dlm_mig_lockres_worker */
+	dlm_grab(dlm);  /* get an extra ref for the work item */
+	memcpy(buf, msg->buf, be16_to_cpu(msg->data_len));  /* copy the whole message */
+	dlm_init_work_item(dlm, item, dlm_mig_lockres_worker, buf);
+	item->u.ml.lockres = res; /* already have a ref */
+	item->u.ml.real_master = real_master;
+	item->u.ml.extra_ref = extra_refs;
+	spin_lock(&dlm->work_lock);
+	list_add_tail(&item->list, &dlm->work_list);
+	spin_unlock(&dlm->work_lock);
+	queue_work(dlm->dlm_worker, &dlm->dispatched_work);
+
+leave:
+	/* One extra ref taken needs to be put here */
+	if (extra_refs)
+		dlm_lockres_put(res);
+
+	dlm_put(dlm);
+	if (ret < 0) {
+		kfree(buf);
+		kfree(item);
+		mlog_errno(ret);
+	}
+
+	return ret;
+}
+
+
+static void dlm_mig_lockres_worker(struct dlm_work_item *item, void *data)
+{
+	struct dlm_ctxt *dlm;
+	struct dlm_migratable_lockres *mres;
+	int ret = 0;
+	struct dlm_lock_resource *res;
+	u8 real_master;
+	u8 extra_ref;
+
+	dlm = item->dlm;
+	mres = (struct dlm_migratable_lockres *)data;
+
+	res = item->u.ml.lockres;
+	real_master = item->u.ml.real_master;
+	extra_ref = item->u.ml.extra_ref;
+
+	if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
+		/* this case is super-rare. only occurs if
+		 * node death happens during migration. */
+again:
+		ret = dlm_lockres_master_requery(dlm, res, &real_master);
+		if (ret < 0) {
+			mlog(0, "dlm_lockres_master_requery ret=%d\n",
+				  ret);
+			goto again;
+		}
+		if (real_master == DLM_LOCK_RES_OWNER_UNKNOWN) {
+			mlog(0, "lockres %.*s not claimed.  "
+				   "this node will take it.\n",
+				   res->lockname.len, res->lockname.name);
+		} else {
+			spin_lock(&res->spinlock);
+			dlm_lockres_drop_inflight_ref(dlm, res);
+			spin_unlock(&res->spinlock);
+			mlog(0, "master needs to respond to sender "
+				  "that node %u still owns %.*s\n",
+				  real_master, res->lockname.len,
+				  res->lockname.name);
+			/* cannot touch this lockres */
+			goto leave;
+		}
+	}
+
+	ret = dlm_process_recovery_data(dlm, res, mres);
+	if (ret < 0)
+		mlog(0, "dlm_process_recovery_data returned  %d\n", ret);
+	else
+		mlog(0, "dlm_process_recovery_data succeeded\n");
+
+	if ((mres->flags & (DLM_MRES_MIGRATION|DLM_MRES_ALL_DONE)) ==
+	                   (DLM_MRES_MIGRATION|DLM_MRES_ALL_DONE)) {
+		ret = dlm_finish_migration(dlm, res, mres->master);
+		if (ret < 0)
+			mlog_errno(ret);
+	}
+
+leave:
+	/* See comment in dlm_mig_lockres_handler() */
+	if (res) {
+		if (extra_ref)
+			dlm_lockres_put(res);
+		dlm_lockres_put(res);
+	}
+	kfree(data);
+}
+
+
+
+static int dlm_lockres_master_requery(struct dlm_ctxt *dlm,
+				      struct dlm_lock_resource *res,
+				      u8 *real_master)
+{
+	struct dlm_node_iter iter;
+	int nodenum;
+	int ret = 0;
+
+	*real_master = DLM_LOCK_RES_OWNER_UNKNOWN;
+
+	/* we only reach here if one of the two nodes in a
+	 * migration died while the migration was in progress.
+	 * at this point we need to requery the master.  we
+	 * know that the new_master got as far as creating
+	 * an mle on at least one node, but we do not know
+	 * if any nodes had actually cleared the mle and set
+	 * the master to the new_master.  the old master
+	 * is supposed to set the owner to UNKNOWN in the
+	 * event of a new_master death, so the only possible
+	 * responses that we can get from nodes here are
+	 * that the master is new_master, or that the master
+	 * is UNKNOWN.
+	 * if all nodes come back with UNKNOWN then we know
+	 * the lock needs remastering here.
+	 * if any node comes back with a valid master, check
+	 * to see if that master is the one that we are
+	 * recovering.  if so, then the new_master died and
+	 * we need to remaster this lock.  if not, then the
+	 * new_master survived and that node will respond to
+	 * other nodes about the owner.
+	 * if there is an owner, this node needs to dump this
+	 * lockres and alert the sender that this lockres
+	 * was rejected. */
+	spin_lock(&dlm->spinlock);
+	dlm_node_iter_init(dlm->domain_map, &iter);
+	spin_unlock(&dlm->spinlock);
+
+	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
+		/* do not send to self */
+		if (nodenum == dlm->node_num)
+			continue;
+		ret = dlm_do_master_requery(dlm, res, nodenum, real_master);
+		if (ret < 0) {
+			mlog_errno(ret);
+			if (!dlm_is_host_down(ret))
+				BUG();
+			/* host is down, so answer for that node would be
+			 * DLM_LOCK_RES_OWNER_UNKNOWN.  continue. */
+		}
+		if (*real_master != DLM_LOCK_RES_OWNER_UNKNOWN) {
+			mlog(0, "lock master is %u\n", *real_master);
+			break;
+		}
+	}
+	return ret;
+}
+
+
+int dlm_do_master_requery(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
+			  u8 nodenum, u8 *real_master)
+{
+	int ret = -EINVAL;
+	struct dlm_master_requery req;
+	int status = DLM_LOCK_RES_OWNER_UNKNOWN;
+
+	memset(&req, 0, sizeof(req));
+	req.node_idx = dlm->node_num;
+	req.namelen = res->lockname.len;
+	memcpy(req.name, res->lockname.name, res->lockname.len);
+
+resend:
+	ret = o2net_send_message(DLM_MASTER_REQUERY_MSG, dlm->key,
+				 &req, sizeof(req), nodenum, &status);
+	if (ret < 0)
+		mlog(ML_ERROR, "Error %d when sending message %u (key "
+		     "0x%x) to node %u\n", ret, DLM_MASTER_REQUERY_MSG,
+		     dlm->key, nodenum);
+	else if (status == -ENOMEM) {
+		mlog_errno(status);
+		msleep(50);
+		goto resend;
+	} else {
+		BUG_ON(status < 0);
+		BUG_ON(status > DLM_LOCK_RES_OWNER_UNKNOWN);
+		*real_master = (u8) (status & 0xff);
+		mlog(0, "node %u responded to master requery with %u\n",
+			  nodenum, *real_master);
+		ret = 0;
+	}
+	return ret;
+}
+
+
+/* this function cannot error, so unless the sending
+ * or receiving of the message failed, the owner can
+ * be trusted */
+int dlm_master_requery_handler(struct o2net_msg *msg, u32 len, void *data,
+			       void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_master_requery *req = (struct dlm_master_requery *)msg->buf;
+	struct dlm_lock_resource *res = NULL;
+	unsigned int hash;
+	int master = DLM_LOCK_RES_OWNER_UNKNOWN;
+	u32 flags = DLM_ASSERT_MASTER_REQUERY;
+
+	if (!dlm_grab(dlm)) {
+		/* since the domain has gone away on this
+		 * node, the proper response is UNKNOWN */
+		return master;
+	}
+
+	hash = dlm_lockid_hash(req->name, req->namelen);
+
+	spin_lock(&dlm->spinlock);
+	res = __dlm_lookup_lockres(dlm, req->name, req->namelen, hash);
+	if (res) {
+		spin_lock(&res->spinlock);
+		master = res->owner;
+		if (master == dlm->node_num) {
+			int ret = dlm_dispatch_assert_master(dlm, res,
+							     0, 0, flags);
+			if (ret < 0) {
+				mlog_errno(ret);
+				spin_unlock(&res->spinlock);
+				dlm_lockres_put(res);
+				spin_unlock(&dlm->spinlock);
+				dlm_put(dlm);
+				/* sender will take care of this and retry */
+				return ret;
+			} else
+				__dlm_lockres_grab_inflight_worker(dlm, res);
+			spin_unlock(&res->spinlock);
+		} else {
+			/* put.. incase we are not the master */
+			spin_unlock(&res->spinlock);
+			dlm_lockres_put(res);
+		}
+	}
+	spin_unlock(&dlm->spinlock);
+
+	dlm_put(dlm);
+	return master;
+}
+
+static inline struct list_head *
+dlm_list_num_to_pointer(struct dlm_lock_resource *res, int list_num)
+{
+	struct list_head *ret;
+	BUG_ON(list_num < 0);
+	BUG_ON(list_num > 2);
+	ret = &(res->granted);
+	ret += list_num;
+	return ret;
+}
+/* TODO: do ast flush business
+ * TODO: do MIGRATING and RECOVERING spinning
+ */
+
+/*
+* NOTE about in-flight requests during migration:
+*
+* Before attempting the migrate, the master has marked the lockres as
+* MIGRATING and then flushed all of its pending ASTS.  So any in-flight
+* requests either got queued before the MIGRATING flag got set, in which
+* case the lock data will reflect the change and a return message is on
+* the way, or the request failed to get in before MIGRATING got set.  In
+* this case, the caller will be told to spin and wait for the MIGRATING
+* flag to be dropped, then recheck the master.
+* This holds true for the convert, cancel and unlock cases, and since lvb
+* updates are tied to these same messages, it applies to lvb updates as
+* well.  For the lock case, there is no way a lock can be on the master
+* queue and not be on the secondary queue since the lock is always added
+* locally first.  This means that the new target node will never be sent
+* a lock that he doesn't already have on the list.
+* In total, this means that the local lock is correct and should not be
+* updated to match the one sent by the master.  Any messages sent back
+* from the master before the MIGRATING flag will bring the lock properly
+* up-to-date, and the change will be ordered properly for the waiter.
+* We will *not* attempt to modify the lock underneath the waiter.
+*/
+
+static int dlm_process_recovery_data(struct dlm_ctxt *dlm,
+				     struct dlm_lock_resource *res,
+				     struct dlm_migratable_lockres *mres)
+{
+	struct dlm_migratable_lock *ml;
+	struct list_head *queue, *iter;
+	struct list_head *tmpq = NULL;
+	struct dlm_lock *newlock = NULL;
+	struct dlm_lockstatus *lksb = NULL;
+	int ret = 0;
+	int i, j, bad;
+	struct dlm_lock *lock;
+	u8 from = O2NM_MAX_NODES;
+	unsigned int added = 0;
+	__be64 c;
+
+	mlog(0, "running %d locks for this lockres\n", mres->num_locks);
+	for (i=0; i<mres->num_locks; i++) {
+		ml = &(mres->ml[i]);
+
+		if (dlm_is_dummy_lock(dlm, ml, &from)) {
+			/* placeholder, just need to set the refmap bit */
+			BUG_ON(mres->num_locks != 1);
+			mlog(0, "%s:%.*s: dummy lock for %u\n",
+			     dlm->name, mres->lockname_len, mres->lockname,
+			     from);
+			spin_lock(&res->spinlock);
+			dlm_lockres_set_refmap_bit(dlm, res, from);
+			spin_unlock(&res->spinlock);
+			added++;
+			break;
+		}
+		BUG_ON(ml->highest_blocked != LKM_IVMODE);
+		newlock = NULL;
+		lksb = NULL;
+
+		queue = dlm_list_num_to_pointer(res, ml->list);
+		tmpq = NULL;
+
+		/* if the lock is for the local node it needs to
+		 * be moved to the proper location within the queue.
+		 * do not allocate a new lock structure. */
+		if (ml->node == dlm->node_num) {
+			/* MIGRATION ONLY! */
+			BUG_ON(!(mres->flags & DLM_MRES_MIGRATION));
+
+			lock = NULL;
+			spin_lock(&res->spinlock);
+			for (j = DLM_GRANTED_LIST; j <= DLM_BLOCKED_LIST; j++) {
+				tmpq = dlm_list_idx_to_ptr(res, j);
+				list_for_each(iter, tmpq) {
+					lock = list_entry(iter,
+						  struct dlm_lock, list);
+					if (lock->ml.cookie == ml->cookie)
+						break;
+					lock = NULL;
+				}
+				if (lock)
+					break;
+			}
+
+			/* lock is always created locally first, and
+			 * destroyed locally last.  it must be on the list */
+			if (!lock) {
+				c = ml->cookie;
+				mlog(ML_ERROR, "Could not find local lock "
+					       "with cookie %u:%llu, node %u, "
+					       "list %u, flags 0x%x, type %d, "
+					       "conv %d, highest blocked %d\n",
+				     dlm_get_lock_cookie_node(be64_to_cpu(c)),
+				     dlm_get_lock_cookie_seq(be64_to_cpu(c)),
+				     ml->node, ml->list, ml->flags, ml->type,
+				     ml->convert_type, ml->highest_blocked);
+				__dlm_print_one_lock_resource(res);
+				BUG();
+			}
+
+			if (lock->ml.node != ml->node) {
+				c = lock->ml.cookie;
+				mlog(ML_ERROR, "Mismatched node# in lock "
+				     "cookie %u:%llu, name %.*s, node %u\n",
+				     dlm_get_lock_cookie_node(be64_to_cpu(c)),
+				     dlm_get_lock_cookie_seq(be64_to_cpu(c)),
+				     res->lockname.len, res->lockname.name,
+				     lock->ml.node);
+				c = ml->cookie;
+				mlog(ML_ERROR, "Migrate lock cookie %u:%llu, "
+				     "node %u, list %u, flags 0x%x, type %d, "
+				     "conv %d, highest blocked %d\n",
+				     dlm_get_lock_cookie_node(be64_to_cpu(c)),
+				     dlm_get_lock_cookie_seq(be64_to_cpu(c)),
+				     ml->node, ml->list, ml->flags, ml->type,
+				     ml->convert_type, ml->highest_blocked);
+				__dlm_print_one_lock_resource(res);
+				BUG();
+			}
+
+			if (tmpq != queue) {
+				c = ml->cookie;
+				mlog(0, "Lock cookie %u:%llu was on list %u "
+				     "instead of list %u for %.*s\n",
+				     dlm_get_lock_cookie_node(be64_to_cpu(c)),
+				     dlm_get_lock_cookie_seq(be64_to_cpu(c)),
+				     j, ml->list, res->lockname.len,
+				     res->lockname.name);
+				__dlm_print_one_lock_resource(res);
+				spin_unlock(&res->spinlock);
+				continue;
+			}
+
+			/* see NOTE above about why we do not update
+			 * to match the master here */
+
+			/* move the lock to its proper place */
+			/* do not alter lock refcount.  switching lists. */
+			list_move_tail(&lock->list, queue);
+			spin_unlock(&res->spinlock);
+			added++;
+
+			mlog(0, "just reordered a local lock!\n");
+			continue;
+		}
+
+		/* lock is for another node. */
+		newlock = dlm_new_lock(ml->type, ml->node,
+				       be64_to_cpu(ml->cookie), NULL);
+		if (!newlock) {
+			ret = -ENOMEM;
+			goto leave;
+		}
+		lksb = newlock->lksb;
+		dlm_lock_attach_lockres(newlock, res);
+
+		if (ml->convert_type != LKM_IVMODE) {
+			BUG_ON(queue != &res->converting);
+			newlock->ml.convert_type = ml->convert_type;
+		}
+		lksb->flags |= (ml->flags &
+				(DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB));
+
+		if (ml->type == LKM_NLMODE)
+			goto skip_lvb;
+
+		/*
+		 * If the lock is in the blocked list it can't have a valid lvb,
+		 * so skip it
+		 */
+		if (ml->list == DLM_BLOCKED_LIST)
+			goto skip_lvb;
+
+		if (!dlm_lvb_is_empty(mres->lvb)) {
+			if (lksb->flags & DLM_LKSB_PUT_LVB) {
+				/* other node was trying to update
+				 * lvb when node died.  recreate the
+				 * lksb with the updated lvb. */
+				memcpy(lksb->lvb, mres->lvb, DLM_LVB_LEN);
+				/* the lock resource lvb update must happen
+				 * NOW, before the spinlock is dropped.
+				 * we no longer wait for the AST to update
+				 * the lvb. */
+				memcpy(res->lvb, mres->lvb, DLM_LVB_LEN);
+			} else {
+				/* otherwise, the node is sending its
+				 * most recent valid lvb info */
+				BUG_ON(ml->type != LKM_EXMODE &&
+				       ml->type != LKM_PRMODE);
+				if (!dlm_lvb_is_empty(res->lvb) &&
+ 				    (ml->type == LKM_EXMODE ||
+ 				     memcmp(res->lvb, mres->lvb, DLM_LVB_LEN))) {
+ 					int i;
+ 					mlog(ML_ERROR, "%s:%.*s: received bad "
+ 					     "lvb! type=%d\n", dlm->name,
+ 					     res->lockname.len,
+ 					     res->lockname.name, ml->type);
+ 					printk("lockres lvb=[");
+ 					for (i=0; i<DLM_LVB_LEN; i++)
+ 						printk("%02x", res->lvb[i]);
+ 					printk("]\nmigrated lvb=[");
+ 					for (i=0; i<DLM_LVB_LEN; i++)
+ 						printk("%02x", mres->lvb[i]);
+ 					printk("]\n");
+ 					dlm_print_one_lock_resource(res);
+ 					BUG();
+				}
+				memcpy(res->lvb, mres->lvb, DLM_LVB_LEN);
+			}
+		}
+skip_lvb:
+
+		/* NOTE:
+		 * wrt lock queue ordering and recovery:
+		 *    1. order of locks on granted queue is
+		 *       meaningless.
+		 *    2. order of locks on converting queue is
+		 *       LOST with the node death.  sorry charlie.
+		 *    3. order of locks on the blocked queue is
+		 *       also LOST.
+		 * order of locks does not affect integrity, it
+		 * just means that a lock request may get pushed
+		 * back in line as a result of the node death.
+		 * also note that for a given node the lock order
+		 * for its secondary queue locks is preserved
+		 * relative to each other, but clearly *not*
+		 * preserved relative to locks from other nodes.
+		 */
+		bad = 0;
+		spin_lock(&res->spinlock);
+		list_for_each_entry(lock, queue, list) {
+			if (lock->ml.cookie == ml->cookie) {
+				c = lock->ml.cookie;
+				mlog(ML_ERROR, "%s:%.*s: %u:%llu: lock already "
+				     "exists on this lockres!\n", dlm->name,
+				     res->lockname.len, res->lockname.name,
+				     dlm_get_lock_cookie_node(be64_to_cpu(c)),
+				     dlm_get_lock_cookie_seq(be64_to_cpu(c)));
+
+				mlog(ML_NOTICE, "sent lock: type=%d, conv=%d, "
+				     "node=%u, cookie=%u:%llu, queue=%d\n",
+	      			     ml->type, ml->convert_type, ml->node,
+				     dlm_get_lock_cookie_node(be64_to_cpu(ml->cookie)),
+				     dlm_get_lock_cookie_seq(be64_to_cpu(ml->cookie)),
+				     ml->list);
+
+				__dlm_print_one_lock_resource(res);
+				bad = 1;
+				break;
+			}
+		}
+		if (!bad) {
+			dlm_lock_get(newlock);
+			if (mres->flags & DLM_MRES_RECOVERY &&
+					ml->list == DLM_CONVERTING_LIST &&
+					newlock->ml.type >
+					newlock->ml.convert_type) {
+				/* newlock is doing downconvert, add it to the
+				 * head of converting list */
+				list_add(&newlock->list, queue);
+			} else
+				list_add_tail(&newlock->list, queue);
+			mlog(0, "%s:%.*s: added lock for node %u, "
+			     "setting refmap bit\n", dlm->name,
+			     res->lockname.len, res->lockname.name, ml->node);
+			dlm_lockres_set_refmap_bit(dlm, res, ml->node);
+			added++;
+		}
+		spin_unlock(&res->spinlock);
+	}
+	mlog(0, "done running all the locks\n");
+
+leave:
+	/* balance the ref taken when the work was queued */
+	spin_lock(&res->spinlock);
+	dlm_lockres_drop_inflight_ref(dlm, res);
+	spin_unlock(&res->spinlock);
+
+	if (ret < 0)
+		mlog_errno(ret);
+
+	return ret;
+}
+
+void dlm_move_lockres_to_recovery_list(struct dlm_ctxt *dlm,
+				       struct dlm_lock_resource *res)
+{
+	int i;
+	struct list_head *queue;
+	struct dlm_lock *lock, *next;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&res->spinlock);
+	res->state |= DLM_LOCK_RES_RECOVERING;
+	if (!list_empty(&res->recovering)) {
+		mlog(0,
+		     "Recovering res %s:%.*s, is already on recovery list!\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
+		list_del_init(&res->recovering);
+		dlm_lockres_put(res);
+	}
+	/* We need to hold a reference while on the recovery list */
+	dlm_lockres_get(res);
+	list_add_tail(&res->recovering, &dlm->reco.resources);
+
+	/* find any pending locks and put them back on proper list */
+	for (i=DLM_BLOCKED_LIST; i>=DLM_GRANTED_LIST; i--) {
+		queue = dlm_list_idx_to_ptr(res, i);
+		list_for_each_entry_safe(lock, next, queue, list) {
+			dlm_lock_get(lock);
+			if (lock->convert_pending) {
+				/* move converting lock back to granted */
+				BUG_ON(i != DLM_CONVERTING_LIST);
+				mlog(0, "node died with convert pending "
+				     "on %.*s. move back to granted list.\n",
+				     res->lockname.len, res->lockname.name);
+				dlm_revert_pending_convert(res, lock);
+				lock->convert_pending = 0;
+			} else if (lock->lock_pending) {
+				/* remove pending lock requests completely */
+				BUG_ON(i != DLM_BLOCKED_LIST);
+				mlog(0, "node died with lock pending "
+				     "on %.*s. remove from blocked list and skip.\n",
+				     res->lockname.len, res->lockname.name);
+				/* lock will be floating until ref in
+				 * dlmlock_remote is freed after the network
+				 * call returns.  ok for it to not be on any
+				 * list since no ast can be called
+				 * (the master is dead). */
+				dlm_revert_pending_lock(res, lock);
+				lock->lock_pending = 0;
+			} else if (lock->unlock_pending) {
+				/* if an unlock was in progress, treat as
+				 * if this had completed successfully
+				 * before sending this lock state to the
+				 * new master.  note that the dlm_unlock
+				 * call is still responsible for calling
+				 * the unlockast.  that will happen after
+				 * the network call times out.  for now,
+				 * just move lists to prepare the new
+				 * recovery master.  */
+				BUG_ON(i != DLM_GRANTED_LIST);
+				mlog(0, "node died with unlock pending "
+				     "on %.*s. remove from blocked list and skip.\n",
+				     res->lockname.len, res->lockname.name);
+				dlm_commit_pending_unlock(res, lock);
+				lock->unlock_pending = 0;
+			} else if (lock->cancel_pending) {
+				/* if a cancel was in progress, treat as
+				 * if this had completed successfully
+				 * before sending this lock state to the
+				 * new master */
+				BUG_ON(i != DLM_CONVERTING_LIST);
+				mlog(0, "node died with cancel pending "
+				     "on %.*s. move back to granted list.\n",
+				     res->lockname.len, res->lockname.name);
+				dlm_commit_pending_cancel(res, lock);
+				lock->cancel_pending = 0;
+			}
+			dlm_lock_put(lock);
+		}
+	}
+}
+
+
+
+/* removes all recovered locks from the recovery list.
+ * sets the res->owner to the new master.
+ * unsets the RECOVERY flag and wakes waiters. */
+static void dlm_finish_local_lockres_recovery(struct dlm_ctxt *dlm,
+					      u8 dead_node, u8 new_master)
+{
+	int i;
+	struct hlist_head *bucket;
+	struct dlm_lock_resource *res, *next;
+
+	assert_spin_locked(&dlm->spinlock);
+
+	list_for_each_entry_safe(res, next, &dlm->reco.resources, recovering) {
+		if (res->owner == dead_node) {
+			mlog(0, "%s: res %.*s, Changing owner from %u to %u\n",
+			     dlm->name, res->lockname.len, res->lockname.name,
+			     res->owner, new_master);
+			list_del_init(&res->recovering);
+			spin_lock(&res->spinlock);
+			/* new_master has our reference from
+			 * the lock state sent during recovery */
+			dlm_change_lockres_owner(dlm, res, new_master);
+			res->state &= ~DLM_LOCK_RES_RECOVERING;
+			if (__dlm_lockres_has_locks(res))
+				__dlm_dirty_lockres(dlm, res);
+			spin_unlock(&res->spinlock);
+			wake_up(&res->wq);
+			dlm_lockres_put(res);
+		}
+	}
+
+	/* this will become unnecessary eventually, but
+	 * for now we need to run the whole hash, clear
+	 * the RECOVERING state and set the owner
+	 * if necessary */
+	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
+		bucket = dlm_lockres_hash(dlm, i);
+		hlist_for_each_entry(res, bucket, hash_node) {
+			if (!(res->state & DLM_LOCK_RES_RECOVERING))
+				continue;
+
+			if (res->owner != dead_node &&
+			    res->owner != dlm->node_num)
+				continue;
+
+			if (!list_empty(&res->recovering)) {
+				list_del_init(&res->recovering);
+				dlm_lockres_put(res);
+			}
+
+			/* new_master has our reference from
+			 * the lock state sent during recovery */
+			mlog(0, "%s: res %.*s, Changing owner from %u to %u\n",
+			     dlm->name, res->lockname.len, res->lockname.name,
+			     res->owner, new_master);
+			spin_lock(&res->spinlock);
+			dlm_change_lockres_owner(dlm, res, new_master);
+			res->state &= ~DLM_LOCK_RES_RECOVERING;
+			if (__dlm_lockres_has_locks(res))
+				__dlm_dirty_lockres(dlm, res);
+			spin_unlock(&res->spinlock);
+			wake_up(&res->wq);
+		}
+	}
+}
+
+static inline int dlm_lvb_needs_invalidation(struct dlm_lock *lock, int local)
+{
+	if (local) {
+		if (lock->ml.type != LKM_EXMODE &&
+		    lock->ml.type != LKM_PRMODE)
+			return 1;
+	} else if (lock->ml.type == LKM_EXMODE)
+		return 1;
+	return 0;
+}
+
+static void dlm_revalidate_lvb(struct dlm_ctxt *dlm,
+			       struct dlm_lock_resource *res, u8 dead_node)
+{
+	struct list_head *queue;
+	struct dlm_lock *lock;
+	int blank_lvb = 0, local = 0;
+	int i;
+	u8 search_node;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&res->spinlock);
+
+	if (res->owner == dlm->node_num)
+		/* if this node owned the lockres, and if the dead node
+		 * had an EX when he died, blank out the lvb */
+		search_node = dead_node;
+	else {
+		/* if this is a secondary lockres, and we had no EX or PR
+		 * locks granted, we can no longer trust the lvb */
+		search_node = dlm->node_num;
+		local = 1;  /* check local state for valid lvb */
+	}
+
+	for (i=DLM_GRANTED_LIST; i<=DLM_CONVERTING_LIST; i++) {
+		queue = dlm_list_idx_to_ptr(res, i);
+		list_for_each_entry(lock, queue, list) {
+			if (lock->ml.node == search_node) {
+				if (dlm_lvb_needs_invalidation(lock, local)) {
+					/* zero the lksb lvb and lockres lvb */
+					blank_lvb = 1;
+					memset(lock->lksb->lvb, 0, DLM_LVB_LEN);
+				}
+			}
+		}
+	}
+
+	if (blank_lvb) {
+		mlog(0, "clearing %.*s lvb, dead node %u had EX\n",
+		     res->lockname.len, res->lockname.name, dead_node);
+		memset(res->lvb, 0, DLM_LVB_LEN);
+	}
+}
+
+static void dlm_free_dead_locks(struct dlm_ctxt *dlm,
+				struct dlm_lock_resource *res, u8 dead_node)
+{
+	struct dlm_lock *lock, *next;
+	unsigned int freed = 0;
+
+	/* this node is the lockres master:
+	 * 1) remove any stale locks for the dead node
+	 * 2) if the dead node had an EX when he died, blank out the lvb
+	 */
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&res->spinlock);
+
+	/* We do two dlm_lock_put(). One for removing from list and the other is
+	 * to force the DLM_UNLOCK_FREE_LOCK action so as to free the locks */
+
+	/* TODO: check pending_asts, pending_basts here */
+	list_for_each_entry_safe(lock, next, &res->granted, list) {
+		if (lock->ml.node == dead_node) {
+			list_del_init(&lock->list);
+			dlm_lock_put(lock);
+			/* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+			dlm_lock_put(lock);
+			freed++;
+		}
+	}
+	list_for_each_entry_safe(lock, next, &res->converting, list) {
+		if (lock->ml.node == dead_node) {
+			list_del_init(&lock->list);
+			dlm_lock_put(lock);
+			/* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+			dlm_lock_put(lock);
+			freed++;
+		}
+	}
+	list_for_each_entry_safe(lock, next, &res->blocked, list) {
+		if (lock->ml.node == dead_node) {
+			list_del_init(&lock->list);
+			dlm_lock_put(lock);
+			/* Can't schedule DLM_UNLOCK_FREE_LOCK - do manually */
+			dlm_lock_put(lock);
+			freed++;
+		}
+	}
+
+	if (freed) {
+		mlog(0, "%s:%.*s: freed %u locks for dead node %u, "
+		     "dropping ref from lockres\n", dlm->name,
+		     res->lockname.len, res->lockname.name, freed, dead_node);
+		if(!test_bit(dead_node, res->refmap)) {
+			mlog(ML_ERROR, "%s:%.*s: freed %u locks for dead node %u, "
+			     "but ref was not set\n", dlm->name,
+			     res->lockname.len, res->lockname.name, freed, dead_node);
+			__dlm_print_one_lock_resource(res);
+		}
+		dlm_lockres_clear_refmap_bit(dlm, res, dead_node);
+	} else if (test_bit(dead_node, res->refmap)) {
+		mlog(0, "%s:%.*s: dead node %u had a ref, but had "
+		     "no locks and had not purged before dying\n", dlm->name,
+		     res->lockname.len, res->lockname.name, dead_node);
+		dlm_lockres_clear_refmap_bit(dlm, res, dead_node);
+	}
+
+	/* do not kick thread yet */
+	__dlm_dirty_lockres(dlm, res);
+}
+
+/* if this node is the recovery master, and there are no
+ * locks for a given lockres owned by this node that are in
+ * either PR or EX mode, zero out the lvb before requesting.
+ *
+ */
+
+
+static void dlm_do_local_recovery_cleanup(struct dlm_ctxt *dlm, u8 dead_node)
+{
+	struct dlm_lock_resource *res;
+	int i;
+	struct hlist_head *bucket;
+	struct dlm_lock *lock;
+
+
+	/* purge any stale mles */
+	dlm_clean_master_list(dlm, dead_node);
+
+	/*
+	 * now clean up all lock resources.  there are two rules:
+	 *
+	 * 1) if the dead node was the master, move the lockres
+	 *    to the recovering list.  set the RECOVERING flag.
+	 *    this lockres needs to be cleaned up before it can
+	 *    be used further.
+	 *
+	 * 2) if this node was the master, remove all locks from
+	 *    each of the lockres queues that were owned by the
+	 *    dead node.  once recovery finishes, the dlm thread
+	 *    can be kicked again to see if any ASTs or BASTs
+	 *    need to be fired as a result.
+	 */
+	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
+		bucket = dlm_lockres_hash(dlm, i);
+		hlist_for_each_entry(res, bucket, hash_node) {
+ 			/* always prune any $RECOVERY entries for dead nodes,
+ 			 * otherwise hangs can occur during later recovery */
+			if (dlm_is_recovery_lock(res->lockname.name,
+						 res->lockname.len)) {
+				spin_lock(&res->spinlock);
+				list_for_each_entry(lock, &res->granted, list) {
+					if (lock->ml.node == dead_node) {
+						mlog(0, "AHA! there was "
+						     "a $RECOVERY lock for dead "
+						     "node %u (%s)!\n",
+						     dead_node, dlm->name);
+						list_del_init(&lock->list);
+						dlm_lock_put(lock);
+						/* Can't schedule
+						 * DLM_UNLOCK_FREE_LOCK
+						 * - do manually */
+						dlm_lock_put(lock);
+						break;
+					}
+				}
+				spin_unlock(&res->spinlock);
+				continue;
+			}
+			spin_lock(&res->spinlock);
+			/* zero the lvb if necessary */
+			dlm_revalidate_lvb(dlm, res, dead_node);
+			if (res->owner == dead_node) {
+				if (res->state & DLM_LOCK_RES_DROPPING_REF) {
+					mlog(ML_NOTICE, "%s: res %.*s, Skip "
+					     "recovery as it is being freed\n",
+					     dlm->name, res->lockname.len,
+					     res->lockname.name);
+				} else
+					dlm_move_lockres_to_recovery_list(dlm,
+									  res);
+
+			} else if (res->owner == dlm->node_num) {
+				dlm_free_dead_locks(dlm, res, dead_node);
+				__dlm_lockres_calc_usage(dlm, res);
+			} else if (res->owner == DLM_LOCK_RES_OWNER_UNKNOWN) {
+				if (test_bit(dead_node, res->refmap)) {
+					mlog(0, "%s:%.*s: dead node %u had a ref, but had "
+						"no locks and had not purged before dying\n",
+						dlm->name, res->lockname.len,
+						res->lockname.name, dead_node);
+					dlm_lockres_clear_refmap_bit(dlm, res, dead_node);
+				}
+			}
+			spin_unlock(&res->spinlock);
+		}
+	}
+
+}
+
+static void __dlm_hb_node_down(struct dlm_ctxt *dlm, int idx)
+{
+	assert_spin_locked(&dlm->spinlock);
+
+	if (dlm->reco.new_master == idx) {
+		mlog(0, "%s: recovery master %d just died\n",
+		     dlm->name, idx);
+		if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
+			/* finalize1 was reached, so it is safe to clear
+			 * the new_master and dead_node.  that recovery
+			 * is complete. */
+			mlog(0, "%s: dead master %d had reached "
+			     "finalize1 state, clearing\n", dlm->name, idx);
+			dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
+			__dlm_reset_recovery(dlm);
+		}
+	}
+
+	/* Clean up join state on node death. */
+	if (dlm->joining_node == idx) {
+		mlog(0, "Clearing join state for node %u\n", idx);
+		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
+	}
+
+	/* check to see if the node is already considered dead */
+	if (!test_bit(idx, dlm->live_nodes_map)) {
+		mlog(0, "for domain %s, node %d is already dead. "
+		     "another node likely did recovery already.\n",
+		     dlm->name, idx);
+		return;
+	}
+
+	/* check to see if we do not care about this node */
+	if (!test_bit(idx, dlm->domain_map)) {
+		/* This also catches the case that we get a node down
+		 * but haven't joined the domain yet. */
+		mlog(0, "node %u already removed from domain!\n", idx);
+		return;
+	}
+
+	clear_bit(idx, dlm->live_nodes_map);
+
+	/* make sure local cleanup occurs before the heartbeat events */
+	if (!test_bit(idx, dlm->recovery_map))
+		dlm_do_local_recovery_cleanup(dlm, idx);
+
+	/* notify anything attached to the heartbeat events */
+	dlm_hb_event_notify_attached(dlm, idx, 0);
+
+	mlog(0, "node %u being removed from domain map!\n", idx);
+	clear_bit(idx, dlm->domain_map);
+	clear_bit(idx, dlm->exit_domain_map);
+	/* wake up migration waiters if a node goes down.
+	 * perhaps later we can genericize this for other waiters. */
+	wake_up(&dlm->migration_wq);
+
+	if (test_bit(idx, dlm->recovery_map))
+		mlog(0, "domain %s, node %u already added "
+		     "to recovery map!\n", dlm->name, idx);
+	else
+		set_bit(idx, dlm->recovery_map);
+}
+
+void dlm_hb_node_down_cb(struct o2nm_node *node, int idx, void *data)
+{
+	struct dlm_ctxt *dlm = data;
+
+	if (!dlm_grab(dlm))
+		return;
+
+	/*
+	 * This will notify any dlm users that a node in our domain
+	 * went away without notifying us first.
+	 */
+	if (test_bit(idx, dlm->domain_map))
+		dlm_fire_domain_eviction_callbacks(dlm, idx);
+
+	spin_lock(&dlm->spinlock);
+	__dlm_hb_node_down(dlm, idx);
+	spin_unlock(&dlm->spinlock);
+
+	dlm_put(dlm);
+}
+
+void dlm_hb_node_up_cb(struct o2nm_node *node, int idx, void *data)
+{
+	struct dlm_ctxt *dlm = data;
+
+	if (!dlm_grab(dlm))
+		return;
+
+	spin_lock(&dlm->spinlock);
+	set_bit(idx, dlm->live_nodes_map);
+	/* do NOT notify mle attached to the heartbeat events.
+	 * new nodes are not interesting in mastery until joined. */
+	spin_unlock(&dlm->spinlock);
+
+	dlm_put(dlm);
+}
+
+static void dlm_reco_ast(void *astdata)
+{
+	struct dlm_ctxt *dlm = astdata;
+	mlog(0, "ast for recovery lock fired!, this=%u, dlm=%s\n",
+	     dlm->node_num, dlm->name);
+}
+static void dlm_reco_bast(void *astdata, int blocked_type)
+{
+	struct dlm_ctxt *dlm = astdata;
+	mlog(0, "bast for recovery lock fired!, this=%u, dlm=%s\n",
+	     dlm->node_num, dlm->name);
+}
+static void dlm_reco_unlock_ast(void *astdata, enum dlm_status st)
+{
+	mlog(0, "unlockast for recovery lock fired!\n");
+}
+
+/*
+ * dlm_pick_recovery_master will continually attempt to use
+ * dlmlock() on the special "$RECOVERY" lockres with the
+ * LKM_NOQUEUE flag to get an EX.  every thread that enters
+ * this function on each node racing to become the recovery
+ * master will not stop attempting this until either:
+ * a) this node gets the EX (and becomes the recovery master),
+ * or b) dlm->reco.new_master gets set to some nodenum
+ * != O2NM_INVALID_NODE_NUM (another node will do the reco).
+ * so each time a recovery master is needed, the entire cluster
+ * will sync at this point.  if the new master dies, that will
+ * be detected in dlm_do_recovery */
+static int dlm_pick_recovery_master(struct dlm_ctxt *dlm)
+{
+	enum dlm_status ret;
+	struct dlm_lockstatus lksb;
+	int status = -EINVAL;
+
+	mlog(0, "starting recovery of %s at %lu, dead=%u, this=%u\n",
+	     dlm->name, jiffies, dlm->reco.dead_node, dlm->node_num);
+again:
+	memset(&lksb, 0, sizeof(lksb));
+
+	ret = dlmlock(dlm, LKM_EXMODE, &lksb, LKM_NOQUEUE|LKM_RECOVERY,
+		      DLM_RECOVERY_LOCK_NAME, DLM_RECOVERY_LOCK_NAME_LEN,
+		      dlm_reco_ast, dlm, dlm_reco_bast);
+
+	mlog(0, "%s: dlmlock($RECOVERY) returned %d, lksb=%d\n",
+	     dlm->name, ret, lksb.status);
+
+	if (ret == DLM_NORMAL) {
+		mlog(0, "dlm=%s dlmlock says I got it (this=%u)\n",
+		     dlm->name, dlm->node_num);
+
+		/* got the EX lock.  check to see if another node
+		 * just became the reco master */
+		if (dlm_reco_master_ready(dlm)) {
+			mlog(0, "%s: got reco EX lock, but %u will "
+			     "do the recovery\n", dlm->name,
+			     dlm->reco.new_master);
+			status = -EEXIST;
+		} else {
+			status = 0;
+
+			/* see if recovery was already finished elsewhere */
+			spin_lock(&dlm->spinlock);
+			if (dlm->reco.dead_node == O2NM_INVALID_NODE_NUM) {
+				status = -EINVAL;
+				mlog(0, "%s: got reco EX lock, but "
+				     "node got recovered already\n", dlm->name);
+				if (dlm->reco.new_master != O2NM_INVALID_NODE_NUM) {
+					mlog(ML_ERROR, "%s: new master is %u "
+					     "but no dead node!\n",
+					     dlm->name, dlm->reco.new_master);
+					BUG();
+				}
+			}
+			spin_unlock(&dlm->spinlock);
+		}
+
+		/* if this node has actually become the recovery master,
+		 * set the master and send the messages to begin recovery */
+		if (!status) {
+			mlog(0, "%s: dead=%u, this=%u, sending "
+			     "begin_reco now\n", dlm->name,
+			     dlm->reco.dead_node, dlm->node_num);
+			status = dlm_send_begin_reco_message(dlm,
+				      dlm->reco.dead_node);
+			/* this always succeeds */
+			BUG_ON(status);
+
+			/* set the new_master to this node */
+			spin_lock(&dlm->spinlock);
+			dlm_set_reco_master(dlm, dlm->node_num);
+			spin_unlock(&dlm->spinlock);
+		}
+
+		/* recovery lock is a special case.  ast will not get fired,
+		 * so just go ahead and unlock it. */
+		ret = dlmunlock(dlm, &lksb, 0, dlm_reco_unlock_ast, dlm);
+		if (ret == DLM_DENIED) {
+			mlog(0, "got DLM_DENIED, trying LKM_CANCEL\n");
+			ret = dlmunlock(dlm, &lksb, LKM_CANCEL, dlm_reco_unlock_ast, dlm);
+		}
+		if (ret != DLM_NORMAL) {
+			/* this would really suck. this could only happen
+			 * if there was a network error during the unlock
+			 * because of node death.  this means the unlock
+			 * is actually "done" and the lock structure is
+			 * even freed.  we can continue, but only
+			 * because this specific lock name is special. */
+			mlog(ML_ERROR, "dlmunlock returned %d\n", ret);
+		}
+	} else if (ret == DLM_NOTQUEUED) {
+		mlog(0, "dlm=%s dlmlock says another node got it (this=%u)\n",
+		     dlm->name, dlm->node_num);
+		/* another node is master. wait on
+		 * reco.new_master != O2NM_INVALID_NODE_NUM
+		 * for at most one second */
+		wait_event_timeout(dlm->dlm_reco_thread_wq,
+					 dlm_reco_master_ready(dlm),
+					 msecs_to_jiffies(1000));
+		if (!dlm_reco_master_ready(dlm)) {
+			mlog(0, "%s: reco master taking awhile\n",
+			     dlm->name);
+			goto again;
+		}
+		/* another node has informed this one that it is reco master */
+		mlog(0, "%s: reco master %u is ready to recover %u\n",
+		     dlm->name, dlm->reco.new_master, dlm->reco.dead_node);
+		status = -EEXIST;
+	} else if (ret == DLM_RECOVERING) {
+		mlog(0, "dlm=%s dlmlock says master node died (this=%u)\n",
+		     dlm->name, dlm->node_num);
+		goto again;
+	} else {
+		struct dlm_lock_resource *res;
+
+		/* dlmlock returned something other than NOTQUEUED or NORMAL */
+		mlog(ML_ERROR, "%s: got %s from dlmlock($RECOVERY), "
+		     "lksb.status=%s\n", dlm->name, dlm_errname(ret),
+		     dlm_errname(lksb.status));
+		res = dlm_lookup_lockres(dlm, DLM_RECOVERY_LOCK_NAME,
+					 DLM_RECOVERY_LOCK_NAME_LEN);
+		if (res) {
+			dlm_print_one_lock_resource(res);
+			dlm_lockres_put(res);
+		} else {
+			mlog(ML_ERROR, "recovery lock not found\n");
+		}
+		BUG();
+	}
+
+	return status;
+}
+
+static int dlm_send_begin_reco_message(struct dlm_ctxt *dlm, u8 dead_node)
+{
+	struct dlm_begin_reco br;
+	int ret = 0;
+	struct dlm_node_iter iter;
+	int nodenum;
+	int status;
+
+	mlog(0, "%s: dead node is %u\n", dlm->name, dead_node);
+
+	spin_lock(&dlm->spinlock);
+	dlm_node_iter_init(dlm->domain_map, &iter);
+	spin_unlock(&dlm->spinlock);
+
+	clear_bit(dead_node, iter.node_map);
+
+	memset(&br, 0, sizeof(br));
+	br.node_idx = dlm->node_num;
+	br.dead_node = dead_node;
+
+	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
+		ret = 0;
+		if (nodenum == dead_node) {
+			mlog(0, "not sending begin reco to dead node "
+				  "%u\n", dead_node);
+			continue;
+		}
+		if (nodenum == dlm->node_num) {
+			mlog(0, "not sending begin reco to self\n");
+			continue;
+		}
+retry:
+		ret = -EINVAL;
+		mlog(0, "attempting to send begin reco msg to %d\n",
+			  nodenum);
+		ret = o2net_send_message(DLM_BEGIN_RECO_MSG, dlm->key,
+					 &br, sizeof(br), nodenum, &status);
+		/* negative status is handled ok by caller here */
+		if (ret >= 0)
+			ret = status;
+		if (dlm_is_host_down(ret)) {
+			/* node is down.  not involved in recovery
+			 * so just keep going */
+			mlog(ML_NOTICE, "%s: node %u was down when sending "
+			     "begin reco msg (%d)\n", dlm->name, nodenum, ret);
+			ret = 0;
+		}
+
+		/*
+		 * Prior to commit aad1b15310b9bcd59fa81ab8f2b1513b59553ea8,
+		 * dlm_begin_reco_handler() returned EAGAIN and not -EAGAIN.
+		 * We are handling both for compatibility reasons.
+		 */
+		if (ret == -EAGAIN || ret == EAGAIN) {
+			mlog(0, "%s: trying to start recovery of node "
+			     "%u, but node %u is waiting for last recovery "
+			     "to complete, backoff for a bit\n", dlm->name,
+			     dead_node, nodenum);
+			msleep(100);
+			goto retry;
+		}
+		if (ret < 0) {
+			struct dlm_lock_resource *res;
+
+			/* this is now a serious problem, possibly ENOMEM
+			 * in the network stack.  must retry */
+			mlog_errno(ret);
+			mlog(ML_ERROR, "begin reco of dlm %s to node %u "
+			     "returned %d\n", dlm->name, nodenum, ret);
+			res = dlm_lookup_lockres(dlm, DLM_RECOVERY_LOCK_NAME,
+						 DLM_RECOVERY_LOCK_NAME_LEN);
+			if (res) {
+				dlm_print_one_lock_resource(res);
+				dlm_lockres_put(res);
+			} else {
+				mlog(ML_ERROR, "recovery lock not found\n");
+			}
+			/* sleep for a bit in hopes that we can avoid
+			 * another ENOMEM */
+			msleep(100);
+			goto retry;
+		}
+	}
+
+	return ret;
+}
+
+int dlm_begin_reco_handler(struct o2net_msg *msg, u32 len, void *data,
+			   void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_begin_reco *br = (struct dlm_begin_reco *)msg->buf;
+
+	/* ok to return 0, domain has gone away */
+	if (!dlm_grab(dlm))
+		return 0;
+
+	spin_lock(&dlm->spinlock);
+	if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
+		mlog(0, "%s: node %u wants to recover node %u (%u:%u) "
+		     "but this node is in finalize state, waiting on finalize2\n",
+		     dlm->name, br->node_idx, br->dead_node,
+		     dlm->reco.dead_node, dlm->reco.new_master);
+		spin_unlock(&dlm->spinlock);
+		dlm_put(dlm);
+		return -EAGAIN;
+	}
+	spin_unlock(&dlm->spinlock);
+
+	mlog(0, "%s: node %u wants to recover node %u (%u:%u)\n",
+	     dlm->name, br->node_idx, br->dead_node,
+	     dlm->reco.dead_node, dlm->reco.new_master);
+
+	dlm_fire_domain_eviction_callbacks(dlm, br->dead_node);
+
+	spin_lock(&dlm->spinlock);
+	if (dlm->reco.new_master != O2NM_INVALID_NODE_NUM) {
+		if (test_bit(dlm->reco.new_master, dlm->recovery_map)) {
+			mlog(0, "%s: new_master %u died, changing "
+			     "to %u\n", dlm->name, dlm->reco.new_master,
+			     br->node_idx);
+		} else {
+			mlog(0, "%s: new_master %u NOT DEAD, changing "
+			     "to %u\n", dlm->name, dlm->reco.new_master,
+			     br->node_idx);
+			/* may not have seen the new master as dead yet */
+		}
+	}
+	if (dlm->reco.dead_node != O2NM_INVALID_NODE_NUM) {
+		mlog(ML_NOTICE, "%s: dead_node previously set to %u, "
+		     "node %u changing it to %u\n", dlm->name,
+		     dlm->reco.dead_node, br->node_idx, br->dead_node);
+	}
+	dlm_set_reco_master(dlm, br->node_idx);
+	dlm_set_reco_dead_node(dlm, br->dead_node);
+	if (!test_bit(br->dead_node, dlm->recovery_map)) {
+		mlog(0, "recovery master %u sees %u as dead, but this "
+		     "node has not yet.  marking %u as dead\n",
+		     br->node_idx, br->dead_node, br->dead_node);
+		if (!test_bit(br->dead_node, dlm->domain_map) ||
+		    !test_bit(br->dead_node, dlm->live_nodes_map))
+			mlog(0, "%u not in domain/live_nodes map "
+			     "so setting it in reco map manually\n",
+			     br->dead_node);
+		/* force the recovery cleanup in __dlm_hb_node_down
+		 * both of these will be cleared in a moment */
+		set_bit(br->dead_node, dlm->domain_map);
+		set_bit(br->dead_node, dlm->live_nodes_map);
+		__dlm_hb_node_down(dlm, br->dead_node);
+	}
+	spin_unlock(&dlm->spinlock);
+
+	dlm_kick_recovery_thread(dlm);
+
+	mlog(0, "%s: recovery started by node %u, for %u (%u:%u)\n",
+	     dlm->name, br->node_idx, br->dead_node,
+	     dlm->reco.dead_node, dlm->reco.new_master);
+
+	dlm_put(dlm);
+	return 0;
+}
+
+#define DLM_FINALIZE_STAGE2  0x01
+static int dlm_send_finalize_reco_message(struct dlm_ctxt *dlm)
+{
+	int ret = 0;
+	struct dlm_finalize_reco fr;
+	struct dlm_node_iter iter;
+	int nodenum;
+	int status;
+	int stage = 1;
+
+	mlog(0, "finishing recovery for node %s:%u, "
+	     "stage %d\n", dlm->name, dlm->reco.dead_node, stage);
+
+	spin_lock(&dlm->spinlock);
+	dlm_node_iter_init(dlm->domain_map, &iter);
+	spin_unlock(&dlm->spinlock);
+
+stage2:
+	memset(&fr, 0, sizeof(fr));
+	fr.node_idx = dlm->node_num;
+	fr.dead_node = dlm->reco.dead_node;
+	if (stage == 2)
+		fr.flags |= DLM_FINALIZE_STAGE2;
+
+	while ((nodenum = dlm_node_iter_next(&iter)) >= 0) {
+		if (nodenum == dlm->node_num)
+			continue;
+		ret = o2net_send_message(DLM_FINALIZE_RECO_MSG, dlm->key,
+					 &fr, sizeof(fr), nodenum, &status);
+		if (ret >= 0)
+			ret = status;
+		if (ret < 0) {
+			mlog(ML_ERROR, "Error %d when sending message %u (key "
+			     "0x%x) to node %u\n", ret, DLM_FINALIZE_RECO_MSG,
+			     dlm->key, nodenum);
+			if (dlm_is_host_down(ret)) {
+				/* this has no effect on this recovery
+				 * session, so set the status to zero to
+				 * finish out the last recovery */
+				mlog(ML_ERROR, "node %u went down after this "
+				     "node finished recovery.\n", nodenum);
+				ret = 0;
+				continue;
+			}
+			break;
+		}
+	}
+	if (stage == 1) {
+		/* reset the node_iter back to the top and send finalize2 */
+		iter.curnode = -1;
+		stage = 2;
+		goto stage2;
+	}
+
+	return ret;
+}
+
+int dlm_finalize_reco_handler(struct o2net_msg *msg, u32 len, void *data,
+			      void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_finalize_reco *fr = (struct dlm_finalize_reco *)msg->buf;
+	int stage = 1;
+
+	/* ok to return 0, domain has gone away */
+	if (!dlm_grab(dlm))
+		return 0;
+
+	if (fr->flags & DLM_FINALIZE_STAGE2)
+		stage = 2;
+
+	mlog(0, "%s: node %u finalizing recovery stage%d of "
+	     "node %u (%u:%u)\n", dlm->name, fr->node_idx, stage,
+	     fr->dead_node, dlm->reco.dead_node, dlm->reco.new_master);
+
+	spin_lock(&dlm->spinlock);
+
+	if (dlm->reco.new_master != fr->node_idx) {
+		mlog(ML_ERROR, "node %u sent recovery finalize msg, but node "
+		     "%u is supposed to be the new master, dead=%u\n",
+		     fr->node_idx, dlm->reco.new_master, fr->dead_node);
+		BUG();
+	}
+	if (dlm->reco.dead_node != fr->dead_node) {
+		mlog(ML_ERROR, "node %u sent recovery finalize msg for dead "
+		     "node %u, but node %u is supposed to be dead\n",
+		     fr->node_idx, fr->dead_node, dlm->reco.dead_node);
+		BUG();
+	}
+
+	switch (stage) {
+		case 1:
+			dlm_finish_local_lockres_recovery(dlm, fr->dead_node, fr->node_idx);
+			if (dlm->reco.state & DLM_RECO_STATE_FINALIZE) {
+				mlog(ML_ERROR, "%s: received finalize1 from "
+				     "new master %u for dead node %u, but "
+				     "this node has already received it!\n",
+				     dlm->name, fr->node_idx, fr->dead_node);
+				dlm_print_reco_node_status(dlm);
+				BUG();
+			}
+			dlm->reco.state |= DLM_RECO_STATE_FINALIZE;
+			spin_unlock(&dlm->spinlock);
+			break;
+		case 2:
+			if (!(dlm->reco.state & DLM_RECO_STATE_FINALIZE)) {
+				mlog(ML_ERROR, "%s: received finalize2 from "
+				     "new master %u for dead node %u, but "
+				     "this node did not have finalize1!\n",
+				     dlm->name, fr->node_idx, fr->dead_node);
+				dlm_print_reco_node_status(dlm);
+				BUG();
+			}
+			dlm->reco.state &= ~DLM_RECO_STATE_FINALIZE;
+			__dlm_reset_recovery(dlm);
+			spin_unlock(&dlm->spinlock);
+			dlm_kick_recovery_thread(dlm);
+			break;
+		default:
+			BUG();
+	}
+
+	mlog(0, "%s: recovery done, reco master was %u, dead now %u, master now %u\n",
+	     dlm->name, fr->node_idx, dlm->reco.dead_node, dlm->reco.new_master);
+
+	dlm_put(dlm);
+	return 0;
+}
diff --git a/fs/ocfs2/dlm/dlmthread.c b/fs/ocfs2/dlm/dlmthread.c
new file mode 100644
index 000000000..69aac6f08
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmthread.c
@@ -0,0 +1,756 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmthread.c
+ *
+ * standalone DLM module
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/sysctl.h>
+#include <linux/random.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/timer.h>
+#include <linux/kthread.h>
+#include <linux/delay.h>
+
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+#include "dlmdomain.h"
+
+#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_THREAD)
+#include "cluster/masklog.h"
+
+static int dlm_thread(void *data);
+static void dlm_flush_asts(struct dlm_ctxt *dlm);
+
+#define dlm_lock_is_remote(dlm, lock)     ((lock)->ml.node != (dlm)->node_num)
+
+/* will exit holding res->spinlock, but may drop in function */
+/* waits until flags are cleared on res->state */
+void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags)
+{
+	DECLARE_WAITQUEUE(wait, current);
+
+	assert_spin_locked(&res->spinlock);
+
+	add_wait_queue(&res->wq, &wait);
+repeat:
+	set_current_state(TASK_UNINTERRUPTIBLE);
+	if (res->state & flags) {
+		spin_unlock(&res->spinlock);
+		schedule();
+		spin_lock(&res->spinlock);
+		goto repeat;
+	}
+	remove_wait_queue(&res->wq, &wait);
+	__set_current_state(TASK_RUNNING);
+}
+
+int __dlm_lockres_has_locks(struct dlm_lock_resource *res)
+{
+	if (list_empty(&res->granted) &&
+	    list_empty(&res->converting) &&
+	    list_empty(&res->blocked))
+		return 0;
+	return 1;
+}
+
+/* "unused": the lockres has no locks, is not on the dirty list,
+ * has no inflight locks (in the gap between mastery and acquiring
+ * the first lock), and has no bits in its refmap.
+ * truly ready to be freed. */
+int __dlm_lockres_unused(struct dlm_lock_resource *res)
+{
+	int bit;
+
+	assert_spin_locked(&res->spinlock);
+
+	if (__dlm_lockres_has_locks(res))
+		return 0;
+
+	/* Locks are in the process of being created */
+	if (res->inflight_locks)
+		return 0;
+
+	if (!list_empty(&res->dirty) || res->state & DLM_LOCK_RES_DIRTY)
+		return 0;
+
+	if (res->state & DLM_LOCK_RES_RECOVERING)
+		return 0;
+
+	/* Another node has this resource with this node as the master */
+	bit = find_next_bit(res->refmap, O2NM_MAX_NODES, 0);
+	if (bit < O2NM_MAX_NODES)
+		return 0;
+
+	return 1;
+}
+
+
+/* Call whenever you may have added or deleted something from one of
+ * the lockres queue's. This will figure out whether it belongs on the
+ * unused list or not and does the appropriate thing. */
+void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
+			      struct dlm_lock_resource *res)
+{
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&res->spinlock);
+
+	if (__dlm_lockres_unused(res)){
+		if (list_empty(&res->purge)) {
+			mlog(0, "%s: Adding res %.*s to purge list\n",
+			     dlm->name, res->lockname.len, res->lockname.name);
+
+			res->last_used = jiffies;
+			dlm_lockres_get(res);
+			list_add_tail(&res->purge, &dlm->purge_list);
+			dlm->purge_count++;
+		}
+	} else if (!list_empty(&res->purge)) {
+		mlog(0, "%s: Removing res %.*s from purge list\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
+
+		list_del_init(&res->purge);
+		dlm_lockres_put(res);
+		dlm->purge_count--;
+	}
+}
+
+void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
+			    struct dlm_lock_resource *res)
+{
+	spin_lock(&dlm->spinlock);
+	spin_lock(&res->spinlock);
+
+	__dlm_lockres_calc_usage(dlm, res);
+
+	spin_unlock(&res->spinlock);
+	spin_unlock(&dlm->spinlock);
+}
+
+static void dlm_purge_lockres(struct dlm_ctxt *dlm,
+			     struct dlm_lock_resource *res)
+{
+	int master;
+	int ret = 0;
+
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&res->spinlock);
+
+	master = (res->owner == dlm->node_num);
+
+	mlog(0, "%s: Purging res %.*s, master %d\n", dlm->name,
+	     res->lockname.len, res->lockname.name, master);
+
+	if (!master) {
+		res->state |= DLM_LOCK_RES_DROPPING_REF;
+		/* drop spinlock...  retake below */
+		spin_unlock(&res->spinlock);
+		spin_unlock(&dlm->spinlock);
+
+		spin_lock(&res->spinlock);
+		/* This ensures that clear refmap is sent after the set */
+		__dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
+		spin_unlock(&res->spinlock);
+
+		/* clear our bit from the master's refmap, ignore errors */
+		ret = dlm_drop_lockres_ref(dlm, res);
+		if (ret < 0) {
+			if (!dlm_is_host_down(ret))
+				BUG();
+		}
+		spin_lock(&dlm->spinlock);
+		spin_lock(&res->spinlock);
+	}
+
+	if (!list_empty(&res->purge)) {
+		mlog(0, "%s: Removing res %.*s from purgelist, master %d\n",
+		     dlm->name, res->lockname.len, res->lockname.name, master);
+		list_del_init(&res->purge);
+		dlm_lockres_put(res);
+		dlm->purge_count--;
+	}
+
+	if (!__dlm_lockres_unused(res)) {
+		mlog(ML_ERROR, "%s: res %.*s in use after deref\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
+		__dlm_print_one_lock_resource(res);
+		BUG();
+	}
+
+	__dlm_unhash_lockres(dlm, res);
+
+	/* lockres is not in the hash now.  drop the flag and wake up
+	 * any processes waiting in dlm_get_lock_resource. */
+	if (!master) {
+		res->state &= ~DLM_LOCK_RES_DROPPING_REF;
+		spin_unlock(&res->spinlock);
+		wake_up(&res->wq);
+	} else
+		spin_unlock(&res->spinlock);
+}
+
+static void dlm_run_purge_list(struct dlm_ctxt *dlm,
+			       int purge_now)
+{
+	unsigned int run_max, unused;
+	unsigned long purge_jiffies;
+	struct dlm_lock_resource *lockres;
+
+	spin_lock(&dlm->spinlock);
+	run_max = dlm->purge_count;
+
+	while(run_max && !list_empty(&dlm->purge_list)) {
+		run_max--;
+
+		lockres = list_entry(dlm->purge_list.next,
+				     struct dlm_lock_resource, purge);
+
+		spin_lock(&lockres->spinlock);
+
+		purge_jiffies = lockres->last_used +
+			msecs_to_jiffies(DLM_PURGE_INTERVAL_MS);
+
+		/* Make sure that we want to be processing this guy at
+		 * this time. */
+		if (!purge_now && time_after(purge_jiffies, jiffies)) {
+			/* Since resources are added to the purge list
+			 * in tail order, we can stop at the first
+			 * unpurgable resource -- anyone added after
+			 * him will have a greater last_used value */
+			spin_unlock(&lockres->spinlock);
+			break;
+		}
+
+		/* Status of the lockres *might* change so double
+		 * check. If the lockres is unused, holding the dlm
+		 * spinlock will prevent people from getting and more
+		 * refs on it. */
+		unused = __dlm_lockres_unused(lockres);
+		if (!unused ||
+		    (lockres->state & DLM_LOCK_RES_MIGRATING) ||
+		    (lockres->inflight_assert_workers != 0)) {
+			mlog(0, "%s: res %.*s is in use or being remastered, "
+			     "used %d, state %d, assert master workers %u\n",
+			     dlm->name, lockres->lockname.len,
+			     lockres->lockname.name,
+			     !unused, lockres->state,
+			     lockres->inflight_assert_workers);
+			list_move_tail(&lockres->purge, &dlm->purge_list);
+			spin_unlock(&lockres->spinlock);
+			continue;
+		}
+
+		dlm_lockres_get(lockres);
+
+		dlm_purge_lockres(dlm, lockres);
+
+		dlm_lockres_put(lockres);
+
+		/* Avoid adding any scheduling latencies */
+		cond_resched_lock(&dlm->spinlock);
+	}
+
+	spin_unlock(&dlm->spinlock);
+}
+
+static void dlm_shuffle_lists(struct dlm_ctxt *dlm,
+			      struct dlm_lock_resource *res)
+{
+	struct dlm_lock *lock, *target;
+	int can_grant = 1;
+
+	/*
+	 * Because this function is called with the lockres
+	 * spinlock, and because we know that it is not migrating/
+	 * recovering/in-progress, it is fine to reserve asts and
+	 * basts right before queueing them all throughout
+	 */
+	assert_spin_locked(&dlm->ast_lock);
+	assert_spin_locked(&res->spinlock);
+	BUG_ON((res->state & (DLM_LOCK_RES_MIGRATING|
+			      DLM_LOCK_RES_RECOVERING|
+			      DLM_LOCK_RES_IN_PROGRESS)));
+
+converting:
+	if (list_empty(&res->converting))
+		goto blocked;
+	mlog(0, "%s: res %.*s has locks on the convert queue\n", dlm->name,
+	     res->lockname.len, res->lockname.name);
+
+	target = list_entry(res->converting.next, struct dlm_lock, list);
+	if (target->ml.convert_type == LKM_IVMODE) {
+		mlog(ML_ERROR, "%s: res %.*s converting lock to invalid mode\n",
+		     dlm->name, res->lockname.len, res->lockname.name);
+		BUG();
+	}
+	list_for_each_entry(lock, &res->granted, list) {
+		if (lock==target)
+			continue;
+		if (!dlm_lock_compatible(lock->ml.type,
+					 target->ml.convert_type)) {
+			can_grant = 0;
+			/* queue the BAST if not already */
+			if (lock->ml.highest_blocked == LKM_IVMODE) {
+				__dlm_lockres_reserve_ast(res);
+				__dlm_queue_bast(dlm, lock);
+			}
+			/* update the highest_blocked if needed */
+			if (lock->ml.highest_blocked < target->ml.convert_type)
+				lock->ml.highest_blocked =
+					target->ml.convert_type;
+		}
+	}
+
+	list_for_each_entry(lock, &res->converting, list) {
+		if (lock==target)
+			continue;
+		if (!dlm_lock_compatible(lock->ml.type,
+					 target->ml.convert_type)) {
+			can_grant = 0;
+			if (lock->ml.highest_blocked == LKM_IVMODE) {
+				__dlm_lockres_reserve_ast(res);
+				__dlm_queue_bast(dlm, lock);
+			}
+			if (lock->ml.highest_blocked < target->ml.convert_type)
+				lock->ml.highest_blocked =
+					target->ml.convert_type;
+		}
+	}
+
+	/* we can convert the lock */
+	if (can_grant) {
+		spin_lock(&target->spinlock);
+		BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
+
+		mlog(0, "%s: res %.*s, AST for Converting lock %u:%llu, type "
+		     "%d => %d, node %u\n", dlm->name, res->lockname.len,
+		     res->lockname.name,
+		     dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
+		     target->ml.type,
+		     target->ml.convert_type, target->ml.node);
+
+		target->ml.type = target->ml.convert_type;
+		target->ml.convert_type = LKM_IVMODE;
+		list_move_tail(&target->list, &res->granted);
+
+		BUG_ON(!target->lksb);
+		target->lksb->status = DLM_NORMAL;
+
+		spin_unlock(&target->spinlock);
+
+		__dlm_lockres_reserve_ast(res);
+		__dlm_queue_ast(dlm, target);
+		/* go back and check for more */
+		goto converting;
+	}
+
+blocked:
+	if (list_empty(&res->blocked))
+		goto leave;
+	target = list_entry(res->blocked.next, struct dlm_lock, list);
+
+	list_for_each_entry(lock, &res->granted, list) {
+		if (lock==target)
+			continue;
+		if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
+			can_grant = 0;
+			if (lock->ml.highest_blocked == LKM_IVMODE) {
+				__dlm_lockres_reserve_ast(res);
+				__dlm_queue_bast(dlm, lock);
+			}
+			if (lock->ml.highest_blocked < target->ml.type)
+				lock->ml.highest_blocked = target->ml.type;
+		}
+	}
+
+	list_for_each_entry(lock, &res->converting, list) {
+		if (lock==target)
+			continue;
+		if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
+			can_grant = 0;
+			if (lock->ml.highest_blocked == LKM_IVMODE) {
+				__dlm_lockres_reserve_ast(res);
+				__dlm_queue_bast(dlm, lock);
+			}
+			if (lock->ml.highest_blocked < target->ml.type)
+				lock->ml.highest_blocked = target->ml.type;
+		}
+	}
+
+	/* we can grant the blocked lock (only
+	 * possible if converting list empty) */
+	if (can_grant) {
+		spin_lock(&target->spinlock);
+		BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
+
+		mlog(0, "%s: res %.*s, AST for Blocked lock %u:%llu, type %d, "
+		     "node %u\n", dlm->name, res->lockname.len,
+		     res->lockname.name,
+		     dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
+		     target->ml.type, target->ml.node);
+
+		/* target->ml.type is already correct */
+		list_move_tail(&target->list, &res->granted);
+
+		BUG_ON(!target->lksb);
+		target->lksb->status = DLM_NORMAL;
+
+		spin_unlock(&target->spinlock);
+
+		__dlm_lockres_reserve_ast(res);
+		__dlm_queue_ast(dlm, target);
+		/* go back and check for more */
+		goto converting;
+	}
+
+leave:
+	return;
+}
+
+/* must have NO locks when calling this with res !=NULL * */
+void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
+{
+	if (res) {
+		spin_lock(&dlm->spinlock);
+		spin_lock(&res->spinlock);
+		__dlm_dirty_lockres(dlm, res);
+		spin_unlock(&res->spinlock);
+		spin_unlock(&dlm->spinlock);
+	}
+	wake_up(&dlm->dlm_thread_wq);
+}
+
+void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
+{
+	assert_spin_locked(&dlm->spinlock);
+	assert_spin_locked(&res->spinlock);
+
+	/* don't shuffle secondary queues */
+	if ((res->owner == dlm->node_num)) {
+		if (res->state & (DLM_LOCK_RES_MIGRATING |
+				  DLM_LOCK_RES_BLOCK_DIRTY))
+		    return;
+
+		if (list_empty(&res->dirty)) {
+			/* ref for dirty_list */
+			dlm_lockres_get(res);
+			list_add_tail(&res->dirty, &dlm->dirty_list);
+			res->state |= DLM_LOCK_RES_DIRTY;
+		}
+	}
+
+	mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len,
+	     res->lockname.name);
+}
+
+
+/* Launch the NM thread for the mounted volume */
+int dlm_launch_thread(struct dlm_ctxt *dlm)
+{
+	mlog(0, "Starting dlm_thread...\n");
+
+	dlm->dlm_thread_task = kthread_run(dlm_thread, dlm, "dlm_thread");
+	if (IS_ERR(dlm->dlm_thread_task)) {
+		mlog_errno(PTR_ERR(dlm->dlm_thread_task));
+		dlm->dlm_thread_task = NULL;
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+void dlm_complete_thread(struct dlm_ctxt *dlm)
+{
+	if (dlm->dlm_thread_task) {
+		mlog(ML_KTHREAD, "Waiting for dlm thread to exit\n");
+		kthread_stop(dlm->dlm_thread_task);
+		dlm->dlm_thread_task = NULL;
+	}
+}
+
+static int dlm_dirty_list_empty(struct dlm_ctxt *dlm)
+{
+	int empty;
+
+	spin_lock(&dlm->spinlock);
+	empty = list_empty(&dlm->dirty_list);
+	spin_unlock(&dlm->spinlock);
+
+	return empty;
+}
+
+static void dlm_flush_asts(struct dlm_ctxt *dlm)
+{
+	int ret;
+	struct dlm_lock *lock;
+	struct dlm_lock_resource *res;
+	u8 hi;
+
+	spin_lock(&dlm->ast_lock);
+	while (!list_empty(&dlm->pending_asts)) {
+		lock = list_entry(dlm->pending_asts.next,
+				  struct dlm_lock, ast_list);
+		/* get an extra ref on lock */
+		dlm_lock_get(lock);
+		res = lock->lockres;
+		mlog(0, "%s: res %.*s, Flush AST for lock %u:%llu, type %d, "
+		     "node %u\n", dlm->name, res->lockname.len,
+		     res->lockname.name,
+		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+		     lock->ml.type, lock->ml.node);
+
+		BUG_ON(!lock->ast_pending);
+
+		/* remove from list (including ref) */
+		list_del_init(&lock->ast_list);
+		dlm_lock_put(lock);
+		spin_unlock(&dlm->ast_lock);
+
+		if (lock->ml.node != dlm->node_num) {
+			ret = dlm_do_remote_ast(dlm, res, lock);
+			if (ret < 0)
+				mlog_errno(ret);
+		} else
+			dlm_do_local_ast(dlm, res, lock);
+
+		spin_lock(&dlm->ast_lock);
+
+		/* possible that another ast was queued while
+		 * we were delivering the last one */
+		if (!list_empty(&lock->ast_list)) {
+			mlog(0, "%s: res %.*s, AST queued while flushing last "
+			     "one\n", dlm->name, res->lockname.len,
+			     res->lockname.name);
+		} else
+			lock->ast_pending = 0;
+
+		/* drop the extra ref.
+		 * this may drop it completely. */
+		dlm_lock_put(lock);
+		dlm_lockres_release_ast(dlm, res);
+	}
+
+	while (!list_empty(&dlm->pending_basts)) {
+		lock = list_entry(dlm->pending_basts.next,
+				  struct dlm_lock, bast_list);
+		/* get an extra ref on lock */
+		dlm_lock_get(lock);
+		res = lock->lockres;
+
+		BUG_ON(!lock->bast_pending);
+
+		/* get the highest blocked lock, and reset */
+		spin_lock(&lock->spinlock);
+		BUG_ON(lock->ml.highest_blocked <= LKM_IVMODE);
+		hi = lock->ml.highest_blocked;
+		lock->ml.highest_blocked = LKM_IVMODE;
+		spin_unlock(&lock->spinlock);
+
+		/* remove from list (including ref) */
+		list_del_init(&lock->bast_list);
+		dlm_lock_put(lock);
+		spin_unlock(&dlm->ast_lock);
+
+		mlog(0, "%s: res %.*s, Flush BAST for lock %u:%llu, "
+		     "blocked %d, node %u\n",
+		     dlm->name, res->lockname.len, res->lockname.name,
+		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+		     hi, lock->ml.node);
+
+		if (lock->ml.node != dlm->node_num) {
+			ret = dlm_send_proxy_bast(dlm, res, lock, hi);
+			if (ret < 0)
+				mlog_errno(ret);
+		} else
+			dlm_do_local_bast(dlm, res, lock, hi);
+
+		spin_lock(&dlm->ast_lock);
+
+		/* possible that another bast was queued while
+		 * we were delivering the last one */
+		if (!list_empty(&lock->bast_list)) {
+			mlog(0, "%s: res %.*s, BAST queued while flushing last "
+			     "one\n", dlm->name, res->lockname.len,
+			     res->lockname.name);
+		} else
+			lock->bast_pending = 0;
+
+		/* drop the extra ref.
+		 * this may drop it completely. */
+		dlm_lock_put(lock);
+		dlm_lockres_release_ast(dlm, res);
+	}
+	wake_up(&dlm->ast_wq);
+	spin_unlock(&dlm->ast_lock);
+}
+
+
+#define DLM_THREAD_TIMEOUT_MS (4 * 1000)
+#define DLM_THREAD_MAX_DIRTY  100
+#define DLM_THREAD_MAX_ASTS   10
+
+static int dlm_thread(void *data)
+{
+	struct dlm_lock_resource *res;
+	struct dlm_ctxt *dlm = data;
+	unsigned long timeout = msecs_to_jiffies(DLM_THREAD_TIMEOUT_MS);
+
+	mlog(0, "dlm thread running for %s...\n", dlm->name);
+
+	while (!kthread_should_stop()) {
+		int n = DLM_THREAD_MAX_DIRTY;
+
+		/* dlm_shutting_down is very point-in-time, but that
+		 * doesn't matter as we'll just loop back around if we
+		 * get false on the leading edge of a state
+		 * transition. */
+		dlm_run_purge_list(dlm, dlm_shutting_down(dlm));
+
+		/* We really don't want to hold dlm->spinlock while
+		 * calling dlm_shuffle_lists on each lockres that
+		 * needs to have its queues adjusted and AST/BASTs
+		 * run.  So let's pull each entry off the dirty_list
+		 * and drop dlm->spinlock ASAP.  Once off the list,
+		 * res->spinlock needs to be taken again to protect
+		 * the queues while calling dlm_shuffle_lists.  */
+		spin_lock(&dlm->spinlock);
+		while (!list_empty(&dlm->dirty_list)) {
+			int delay = 0;
+			res = list_entry(dlm->dirty_list.next,
+					 struct dlm_lock_resource, dirty);
+
+			/* peel a lockres off, remove it from the list,
+			 * unset the dirty flag and drop the dlm lock */
+			BUG_ON(!res);
+			dlm_lockres_get(res);
+
+			spin_lock(&res->spinlock);
+			/* We clear the DLM_LOCK_RES_DIRTY state once we shuffle lists below */
+			list_del_init(&res->dirty);
+			spin_unlock(&res->spinlock);
+			spin_unlock(&dlm->spinlock);
+			/* Drop dirty_list ref */
+			dlm_lockres_put(res);
+
+		 	/* lockres can be re-dirtied/re-added to the
+			 * dirty_list in this gap, but that is ok */
+
+			spin_lock(&dlm->ast_lock);
+			spin_lock(&res->spinlock);
+			if (res->owner != dlm->node_num) {
+				__dlm_print_one_lock_resource(res);
+				mlog(ML_ERROR, "%s: inprog %d, mig %d, reco %d,"
+				     " dirty %d\n", dlm->name,
+				     !!(res->state & DLM_LOCK_RES_IN_PROGRESS),
+				     !!(res->state & DLM_LOCK_RES_MIGRATING),
+				     !!(res->state & DLM_LOCK_RES_RECOVERING),
+				     !!(res->state & DLM_LOCK_RES_DIRTY));
+			}
+			BUG_ON(res->owner != dlm->node_num);
+
+			/* it is now ok to move lockreses in these states
+			 * to the dirty list, assuming that they will only be
+			 * dirty for a short while. */
+			BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
+			if (res->state & (DLM_LOCK_RES_IN_PROGRESS |
+					  DLM_LOCK_RES_RECOVERING)) {
+				/* move it to the tail and keep going */
+				res->state &= ~DLM_LOCK_RES_DIRTY;
+				spin_unlock(&res->spinlock);
+				spin_unlock(&dlm->ast_lock);
+				mlog(0, "%s: res %.*s, inprogress, delay list "
+				     "shuffle, state %d\n", dlm->name,
+				     res->lockname.len, res->lockname.name,
+				     res->state);
+				delay = 1;
+				goto in_progress;
+			}
+
+			/* at this point the lockres is not migrating/
+			 * recovering/in-progress.  we have the lockres
+			 * spinlock and do NOT have the dlm lock.
+			 * safe to reserve/queue asts and run the lists. */
+
+			/* called while holding lockres lock */
+			dlm_shuffle_lists(dlm, res);
+			res->state &= ~DLM_LOCK_RES_DIRTY;
+			spin_unlock(&res->spinlock);
+			spin_unlock(&dlm->ast_lock);
+
+			dlm_lockres_calc_usage(dlm, res);
+
+in_progress:
+
+			spin_lock(&dlm->spinlock);
+			/* if the lock was in-progress, stick
+			 * it on the back of the list */
+			if (delay) {
+				spin_lock(&res->spinlock);
+				__dlm_dirty_lockres(dlm, res);
+				spin_unlock(&res->spinlock);
+			}
+			dlm_lockres_put(res);
+
+			/* unlikely, but we may need to give time to
+			 * other tasks */
+			if (!--n) {
+				mlog(0, "%s: Throttling dlm thread\n",
+				     dlm->name);
+				break;
+			}
+		}
+
+		spin_unlock(&dlm->spinlock);
+		dlm_flush_asts(dlm);
+
+		/* yield and continue right away if there is more work to do */
+		if (!n) {
+			cond_resched();
+			continue;
+		}
+
+		wait_event_interruptible_timeout(dlm->dlm_thread_wq,
+						 !dlm_dirty_list_empty(dlm) ||
+						 kthread_should_stop(),
+						 timeout);
+	}
+
+	mlog(0, "quitting DLM thread\n");
+	return 0;
+}
diff --git a/fs/ocfs2/dlm/dlmunlock.c b/fs/ocfs2/dlm/dlmunlock.c
new file mode 100644
index 000000000..2e3c9dbab
--- /dev/null
+++ b/fs/ocfs2/dlm/dlmunlock.c
@@ -0,0 +1,698 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmunlock.c
+ *
+ * underlying calls for unlocking locks
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ */
+
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/sysctl.h>
+#include <linux/random.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/spinlock.h>
+#include <linux/delay.h>
+
+#include "cluster/heartbeat.h"
+#include "cluster/nodemanager.h"
+#include "cluster/tcp.h"
+
+#include "dlmapi.h"
+#include "dlmcommon.h"
+
+#define MLOG_MASK_PREFIX ML_DLM
+#include "cluster/masklog.h"
+
+#define DLM_UNLOCK_FREE_LOCK           0x00000001
+#define DLM_UNLOCK_CALL_AST            0x00000002
+#define DLM_UNLOCK_REMOVE_LOCK         0x00000004
+#define DLM_UNLOCK_REGRANT_LOCK        0x00000008
+#define DLM_UNLOCK_CLEAR_CONVERT_TYPE  0x00000010
+
+
+static enum dlm_status dlm_get_cancel_actions(struct dlm_ctxt *dlm,
+					      struct dlm_lock_resource *res,
+					      struct dlm_lock *lock,
+					      struct dlm_lockstatus *lksb,
+					      int *actions);
+static enum dlm_status dlm_get_unlock_actions(struct dlm_ctxt *dlm,
+					      struct dlm_lock_resource *res,
+					      struct dlm_lock *lock,
+					      struct dlm_lockstatus *lksb,
+					      int *actions);
+
+static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm,
+						 struct dlm_lock_resource *res,
+						 struct dlm_lock *lock,
+						 struct dlm_lockstatus *lksb,
+						 int flags,
+						 u8 owner);
+
+
+/*
+ * according to the spec:
+ * http://opendlm.sourceforge.net/cvsmirror/opendlm/docs/dlmbook_final.pdf
+ *
+ *  flags & LKM_CANCEL != 0: must be converting or blocked
+ *  flags & LKM_CANCEL == 0: must be granted
+ *
+ * So to unlock a converting lock, you must first cancel the
+ * convert (passing LKM_CANCEL in flags), then call the unlock
+ * again (with no LKM_CANCEL in flags).
+ */
+
+
+/*
+ * locking:
+ *   caller needs:  none
+ *   taken:         res->spinlock and lock->spinlock taken and dropped
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_NOLOCKMGR, status from network
+ * all callers should have taken an extra ref on lock coming in
+ */
+static enum dlm_status dlmunlock_common(struct dlm_ctxt *dlm,
+					struct dlm_lock_resource *res,
+					struct dlm_lock *lock,
+					struct dlm_lockstatus *lksb,
+					int flags, int *call_ast,
+					int master_node)
+{
+	enum dlm_status status;
+	int actions = 0;
+	int in_use;
+        u8 owner;
+
+	mlog(0, "master_node = %d, valblk = %d\n", master_node,
+	     flags & LKM_VALBLK);
+
+	if (master_node)
+		BUG_ON(res->owner != dlm->node_num);
+	else
+		BUG_ON(res->owner == dlm->node_num);
+
+	spin_lock(&dlm->ast_lock);
+	/* We want to be sure that we're not freeing a lock
+	 * that still has AST's pending... */
+	in_use = !list_empty(&lock->ast_list);
+	spin_unlock(&dlm->ast_lock);
+	if (in_use && !(flags & LKM_CANCEL)) {
+	       mlog(ML_ERROR, "lockres %.*s: Someone is calling dlmunlock "
+		    "while waiting for an ast!", res->lockname.len,
+		    res->lockname.name);
+		return DLM_BADPARAM;
+	}
+
+	spin_lock(&res->spinlock);
+	if (res->state & DLM_LOCK_RES_IN_PROGRESS) {
+		if (master_node && !(flags & LKM_CANCEL)) {
+			mlog(ML_ERROR, "lockres in progress!\n");
+			spin_unlock(&res->spinlock);
+			return DLM_FORWARD;
+		}
+		/* ok for this to sleep if not in a network handler */
+		__dlm_wait_on_lockres(res);
+		res->state |= DLM_LOCK_RES_IN_PROGRESS;
+	}
+	spin_lock(&lock->spinlock);
+
+	if (res->state & DLM_LOCK_RES_RECOVERING) {
+		status = DLM_RECOVERING;
+		goto leave;
+	}
+
+	if (res->state & DLM_LOCK_RES_MIGRATING) {
+		status = DLM_MIGRATING;
+		goto leave;
+	}
+
+	/* see above for what the spec says about
+	 * LKM_CANCEL and the lock queue state */
+	if (flags & LKM_CANCEL)
+		status = dlm_get_cancel_actions(dlm, res, lock, lksb, &actions);
+	else
+		status = dlm_get_unlock_actions(dlm, res, lock, lksb, &actions);
+
+	if (status != DLM_NORMAL && (status != DLM_CANCELGRANT || !master_node))
+		goto leave;
+
+	/* By now this has been masked out of cancel requests. */
+	if (flags & LKM_VALBLK) {
+		/* make the final update to the lvb */
+		if (master_node)
+			memcpy(res->lvb, lksb->lvb, DLM_LVB_LEN);
+		else
+			flags |= LKM_PUT_LVB; /* let the send function
+					       * handle it. */
+	}
+
+	if (!master_node) {
+		owner = res->owner;
+		/* drop locks and send message */
+		if (flags & LKM_CANCEL)
+			lock->cancel_pending = 1;
+		else
+			lock->unlock_pending = 1;
+		spin_unlock(&lock->spinlock);
+		spin_unlock(&res->spinlock);
+		status = dlm_send_remote_unlock_request(dlm, res, lock, lksb,
+							flags, owner);
+		spin_lock(&res->spinlock);
+		spin_lock(&lock->spinlock);
+		/* if the master told us the lock was already granted,
+		 * let the ast handle all of these actions */
+		if (status == DLM_CANCELGRANT) {
+			actions &= ~(DLM_UNLOCK_REMOVE_LOCK|
+				     DLM_UNLOCK_REGRANT_LOCK|
+				     DLM_UNLOCK_CLEAR_CONVERT_TYPE);
+		} else if (status == DLM_RECOVERING ||
+			   status == DLM_MIGRATING ||
+			   status == DLM_FORWARD ||
+			   status == DLM_NOLOCKMGR
+			   ) {
+			/* must clear the actions because this unlock
+			 * is about to be retried.  cannot free or do
+			 * any list manipulation. */
+			mlog(0, "%s:%.*s: clearing actions, %s\n",
+			     dlm->name, res->lockname.len,
+			     res->lockname.name,
+			     status==DLM_RECOVERING?"recovering":
+			     (status==DLM_MIGRATING?"migrating":
+				(status == DLM_FORWARD ? "forward" :
+						"nolockmanager")));
+			actions = 0;
+		}
+		if (flags & LKM_CANCEL)
+			lock->cancel_pending = 0;
+		else
+			lock->unlock_pending = 0;
+
+	}
+
+	/* get an extra ref on lock.  if we are just switching
+	 * lists here, we dont want the lock to go away. */
+	dlm_lock_get(lock);
+
+	if (actions & DLM_UNLOCK_REMOVE_LOCK) {
+		list_del_init(&lock->list);
+		dlm_lock_put(lock);
+	}
+	if (actions & DLM_UNLOCK_REGRANT_LOCK) {
+		dlm_lock_get(lock);
+		list_add_tail(&lock->list, &res->granted);
+	}
+	if (actions & DLM_UNLOCK_CLEAR_CONVERT_TYPE) {
+		mlog(0, "clearing convert_type at %smaster node\n",
+		     master_node ? "" : "non-");
+		lock->ml.convert_type = LKM_IVMODE;
+	}
+
+	/* remove the extra ref on lock */
+	dlm_lock_put(lock);
+
+leave:
+	res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
+	if (!dlm_lock_on_list(&res->converting, lock))
+		BUG_ON(lock->ml.convert_type != LKM_IVMODE);
+	else
+		BUG_ON(lock->ml.convert_type == LKM_IVMODE);
+	spin_unlock(&lock->spinlock);
+	spin_unlock(&res->spinlock);
+	wake_up(&res->wq);
+
+	/* let the caller's final dlm_lock_put handle the actual kfree */
+	if (actions & DLM_UNLOCK_FREE_LOCK) {
+		/* this should always be coupled with list removal */
+		BUG_ON(!(actions & DLM_UNLOCK_REMOVE_LOCK));
+		mlog(0, "lock %u:%llu should be gone now! refs=%d\n",
+		     dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
+		     atomic_read(&lock->lock_refs.refcount)-1);
+		dlm_lock_put(lock);
+	}
+	if (actions & DLM_UNLOCK_CALL_AST)
+		*call_ast = 1;
+
+	/* if cancel or unlock succeeded, lvb work is done */
+	if (status == DLM_NORMAL)
+		lksb->flags &= ~(DLM_LKSB_PUT_LVB|DLM_LKSB_GET_LVB);
+
+	return status;
+}
+
+void dlm_commit_pending_unlock(struct dlm_lock_resource *res,
+			       struct dlm_lock *lock)
+{
+	/* leave DLM_LKSB_PUT_LVB on the lksb so any final
+	 * update of the lvb will be sent to the new master */
+	list_del_init(&lock->list);
+}
+
+void dlm_commit_pending_cancel(struct dlm_lock_resource *res,
+			       struct dlm_lock *lock)
+{
+	list_move_tail(&lock->list, &res->granted);
+	lock->ml.convert_type = LKM_IVMODE;
+}
+
+
+static inline enum dlm_status dlmunlock_master(struct dlm_ctxt *dlm,
+					  struct dlm_lock_resource *res,
+					  struct dlm_lock *lock,
+					  struct dlm_lockstatus *lksb,
+					  int flags,
+					  int *call_ast)
+{
+	return dlmunlock_common(dlm, res, lock, lksb, flags, call_ast, 1);
+}
+
+static inline enum dlm_status dlmunlock_remote(struct dlm_ctxt *dlm,
+					  struct dlm_lock_resource *res,
+					  struct dlm_lock *lock,
+					  struct dlm_lockstatus *lksb,
+					  int flags, int *call_ast)
+{
+	return dlmunlock_common(dlm, res, lock, lksb, flags, call_ast, 0);
+}
+
+/*
+ * locking:
+ *   caller needs:  none
+ *   taken:         none
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_NOLOCKMGR, status from network
+ */
+static enum dlm_status dlm_send_remote_unlock_request(struct dlm_ctxt *dlm,
+						 struct dlm_lock_resource *res,
+						 struct dlm_lock *lock,
+						 struct dlm_lockstatus *lksb,
+						 int flags,
+						 u8 owner)
+{
+	struct dlm_unlock_lock unlock;
+	int tmpret;
+	enum dlm_status ret;
+	int status = 0;
+	struct kvec vec[2];
+	size_t veclen = 1;
+
+	mlog(0, "%.*s\n", res->lockname.len, res->lockname.name);
+
+	if (owner == dlm->node_num) {
+		/* ended up trying to contact ourself.  this means
+		 * that the lockres had been remote but became local
+		 * via a migration.  just retry it, now as local */
+		mlog(0, "%s:%.*s: this node became the master due to a "
+		     "migration, re-evaluate now\n", dlm->name,
+		     res->lockname.len, res->lockname.name);
+		return DLM_FORWARD;
+	}
+
+	memset(&unlock, 0, sizeof(unlock));
+	unlock.node_idx = dlm->node_num;
+	unlock.flags = cpu_to_be32(flags);
+	unlock.cookie = lock->ml.cookie;
+	unlock.namelen = res->lockname.len;
+	memcpy(unlock.name, res->lockname.name, unlock.namelen);
+
+	vec[0].iov_len = sizeof(struct dlm_unlock_lock);
+	vec[0].iov_base = &unlock;
+
+	if (flags & LKM_PUT_LVB) {
+		/* extra data to send if we are updating lvb */
+		vec[1].iov_len = DLM_LVB_LEN;
+		vec[1].iov_base = lock->lksb->lvb;
+		veclen++;
+	}
+
+	tmpret = o2net_send_message_vec(DLM_UNLOCK_LOCK_MSG, dlm->key,
+					vec, veclen, owner, &status);
+	if (tmpret >= 0) {
+		// successfully sent and received
+		if (status == DLM_FORWARD)
+			mlog(0, "master was in-progress.  retry\n");
+		ret = status;
+	} else {
+		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
+		     "node %u\n", tmpret, DLM_UNLOCK_LOCK_MSG, dlm->key, owner);
+		if (dlm_is_host_down(tmpret)) {
+			/* NOTE: this seems strange, but it is what we want.
+			 * when the master goes down during a cancel or
+			 * unlock, the recovery code completes the operation
+			 * as if the master had not died, then passes the
+			 * updated state to the recovery master.  this thread
+			 * just needs to finish out the operation and call
+			 * the unlockast. */
+			if (dlm_is_node_dead(dlm, owner))
+				ret = DLM_NORMAL;
+			else
+				ret = DLM_NOLOCKMGR;
+		} else {
+			/* something bad.  this will BUG in ocfs2 */
+			ret = dlm_err_to_dlm_status(tmpret);
+		}
+	}
+
+	return ret;
+}
+
+/*
+ * locking:
+ *   caller needs:  none
+ *   taken:         takes and drops res->spinlock
+ *   held on exit:  none
+ * returns: DLM_NORMAL, DLM_BADARGS, DLM_IVLOCKID,
+ *          return value from dlmunlock_master
+ */
+int dlm_unlock_lock_handler(struct o2net_msg *msg, u32 len, void *data,
+			    void **ret_data)
+{
+	struct dlm_ctxt *dlm = data;
+	struct dlm_unlock_lock *unlock = (struct dlm_unlock_lock *)msg->buf;
+	struct dlm_lock_resource *res = NULL;
+	struct dlm_lock *lock = NULL;
+	enum dlm_status status = DLM_NORMAL;
+	int found = 0, i;
+	struct dlm_lockstatus *lksb = NULL;
+	int ignore;
+	u32 flags;
+	struct list_head *queue;
+
+	flags = be32_to_cpu(unlock->flags);
+
+	if (flags & LKM_GET_LVB) {
+		mlog(ML_ERROR, "bad args!  GET_LVB specified on unlock!\n");
+		return DLM_BADARGS;
+	}
+
+	if ((flags & (LKM_PUT_LVB|LKM_CANCEL)) == (LKM_PUT_LVB|LKM_CANCEL)) {
+		mlog(ML_ERROR, "bad args!  cannot modify lvb on a CANCEL "
+		     "request!\n");
+		return DLM_BADARGS;
+	}
+
+	if (unlock->namelen > DLM_LOCKID_NAME_MAX) {
+		mlog(ML_ERROR, "Invalid name length in unlock handler!\n");
+		return DLM_IVBUFLEN;
+	}
+
+	if (!dlm_grab(dlm))
+		return DLM_REJECTED;
+
+	mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
+			"Domain %s not fully joined!\n", dlm->name);
+
+	mlog(0, "lvb: %s\n", flags & LKM_PUT_LVB ? "put lvb" : "none");
+
+	res = dlm_lookup_lockres(dlm, unlock->name, unlock->namelen);
+	if (!res) {
+		/* We assume here that a no lock resource simply means
+		 * it was migrated away and destroyed before the other
+		 * node could detect it. */
+		mlog(0, "returning DLM_FORWARD -- res no longer exists\n");
+		status = DLM_FORWARD;
+		goto not_found;
+	}
+
+	queue=&res->granted;
+	found = 0;
+	spin_lock(&res->spinlock);
+	if (res->state & DLM_LOCK_RES_RECOVERING) {
+		spin_unlock(&res->spinlock);
+		mlog(0, "returning DLM_RECOVERING\n");
+		status = DLM_RECOVERING;
+		goto leave;
+	}
+
+	if (res->state & DLM_LOCK_RES_MIGRATING) {
+		spin_unlock(&res->spinlock);
+		mlog(0, "returning DLM_MIGRATING\n");
+		status = DLM_MIGRATING;
+		goto leave;
+	}
+
+	if (res->owner != dlm->node_num) {
+		spin_unlock(&res->spinlock);
+		mlog(0, "returning DLM_FORWARD -- not master\n");
+		status = DLM_FORWARD;
+		goto leave;
+	}
+
+	for (i=0; i<3; i++) {
+		list_for_each_entry(lock, queue, list) {
+			if (lock->ml.cookie == unlock->cookie &&
+		    	    lock->ml.node == unlock->node_idx) {
+				dlm_lock_get(lock);
+				found = 1;
+				break;
+			}
+		}
+		if (found)
+			break;
+		/* scan granted -> converting -> blocked queues */
+		queue++;
+	}
+	spin_unlock(&res->spinlock);
+	if (!found) {
+		status = DLM_IVLOCKID;
+		goto not_found;
+	}
+
+	/* lock was found on queue */
+	lksb = lock->lksb;
+	if (flags & (LKM_VALBLK|LKM_PUT_LVB) &&
+	    lock->ml.type != LKM_EXMODE)
+		flags &= ~(LKM_VALBLK|LKM_PUT_LVB);
+
+	/* unlockast only called on originating node */
+	if (flags & LKM_PUT_LVB) {
+		lksb->flags |= DLM_LKSB_PUT_LVB;
+		memcpy(&lksb->lvb[0], &unlock->lvb[0], DLM_LVB_LEN);
+	}
+
+	/* if this is in-progress, propagate the DLM_FORWARD
+	 * all the way back out */
+	status = dlmunlock_master(dlm, res, lock, lksb, flags, &ignore);
+	if (status == DLM_FORWARD)
+		mlog(0, "lockres is in progress\n");
+
+	if (flags & LKM_PUT_LVB)
+		lksb->flags &= ~DLM_LKSB_PUT_LVB;
+
+	dlm_lockres_calc_usage(dlm, res);
+	dlm_kick_thread(dlm, res);
+
+not_found:
+	if (!found)
+		mlog(ML_ERROR, "failed to find lock to unlock! "
+			       "cookie=%u:%llu\n",
+		     dlm_get_lock_cookie_node(be64_to_cpu(unlock->cookie)),
+		     dlm_get_lock_cookie_seq(be64_to_cpu(unlock->cookie)));
+	else
+		dlm_lock_put(lock);
+
+leave:
+	if (res)
+		dlm_lockres_put(res);
+
+	dlm_put(dlm);
+
+	return status;
+}
+
+
+static enum dlm_status dlm_get_cancel_actions(struct dlm_ctxt *dlm,
+					      struct dlm_lock_resource *res,
+					      struct dlm_lock *lock,
+					      struct dlm_lockstatus *lksb,
+					      int *actions)
+{
+	enum dlm_status status;
+
+	if (dlm_lock_on_list(&res->blocked, lock)) {
+		/* cancel this outright */
+		status = DLM_NORMAL;
+		*actions = (DLM_UNLOCK_CALL_AST |
+			    DLM_UNLOCK_REMOVE_LOCK);
+	} else if (dlm_lock_on_list(&res->converting, lock)) {
+		/* cancel the request, put back on granted */
+		status = DLM_NORMAL;
+		*actions = (DLM_UNLOCK_CALL_AST |
+			    DLM_UNLOCK_REMOVE_LOCK |
+			    DLM_UNLOCK_REGRANT_LOCK |
+			    DLM_UNLOCK_CLEAR_CONVERT_TYPE);
+	} else if (dlm_lock_on_list(&res->granted, lock)) {
+		/* too late, already granted. */
+		status = DLM_CANCELGRANT;
+		*actions = DLM_UNLOCK_CALL_AST;
+	} else {
+		mlog(ML_ERROR, "lock to cancel is not on any list!\n");
+		status = DLM_IVLOCKID;
+		*actions = 0;
+	}
+	return status;
+}
+
+static enum dlm_status dlm_get_unlock_actions(struct dlm_ctxt *dlm,
+					      struct dlm_lock_resource *res,
+					      struct dlm_lock *lock,
+					      struct dlm_lockstatus *lksb,
+					      int *actions)
+{
+	enum dlm_status status;
+
+	/* unlock request */
+	if (!dlm_lock_on_list(&res->granted, lock)) {
+		status = DLM_DENIED;
+		dlm_error(status);
+		*actions = 0;
+	} else {
+		/* unlock granted lock */
+		status = DLM_NORMAL;
+		*actions = (DLM_UNLOCK_FREE_LOCK |
+			    DLM_UNLOCK_CALL_AST |
+			    DLM_UNLOCK_REMOVE_LOCK);
+	}
+	return status;
+}
+
+/* there seems to be no point in doing this async
+ * since (even for the remote case) there is really
+ * no work to queue up... so just do it and fire the
+ * unlockast by hand when done... */
+enum dlm_status dlmunlock(struct dlm_ctxt *dlm, struct dlm_lockstatus *lksb,
+			  int flags, dlm_astunlockfunc_t *unlockast, void *data)
+{
+	enum dlm_status status;
+	struct dlm_lock_resource *res;
+	struct dlm_lock *lock = NULL;
+	int call_ast, is_master;
+
+	if (!lksb) {
+		dlm_error(DLM_BADARGS);
+		return DLM_BADARGS;
+	}
+
+	if (flags & ~(LKM_CANCEL | LKM_VALBLK | LKM_INVVALBLK)) {
+		dlm_error(DLM_BADPARAM);
+		return DLM_BADPARAM;
+	}
+
+	if ((flags & (LKM_VALBLK | LKM_CANCEL)) == (LKM_VALBLK | LKM_CANCEL)) {
+		mlog(0, "VALBLK given with CANCEL: ignoring VALBLK\n");
+		flags &= ~LKM_VALBLK;
+	}
+
+	if (!lksb->lockid || !lksb->lockid->lockres) {
+		dlm_error(DLM_BADPARAM);
+		return DLM_BADPARAM;
+	}
+
+	lock = lksb->lockid;
+	BUG_ON(!lock);
+	dlm_lock_get(lock);
+
+	res = lock->lockres;
+	BUG_ON(!res);
+	dlm_lockres_get(res);
+retry:
+	call_ast = 0;
+	/* need to retry up here because owner may have changed */
+	mlog(0, "lock=%p res=%p\n", lock, res);
+
+	spin_lock(&res->spinlock);
+	is_master = (res->owner == dlm->node_num);
+	if (flags & LKM_VALBLK && lock->ml.type != LKM_EXMODE)
+		flags &= ~LKM_VALBLK;
+	spin_unlock(&res->spinlock);
+
+	if (is_master) {
+		status = dlmunlock_master(dlm, res, lock, lksb, flags,
+					  &call_ast);
+		mlog(0, "done calling dlmunlock_master: returned %d, "
+		     "call_ast is %d\n", status, call_ast);
+	} else {
+		status = dlmunlock_remote(dlm, res, lock, lksb, flags,
+					  &call_ast);
+		mlog(0, "done calling dlmunlock_remote: returned %d, "
+		     "call_ast is %d\n", status, call_ast);
+	}
+
+	if (status == DLM_RECOVERING ||
+	    status == DLM_MIGRATING ||
+	    status == DLM_FORWARD ||
+	    status == DLM_NOLOCKMGR) {
+
+		/* We want to go away for a tiny bit to allow recovery
+		 * / migration to complete on this resource. I don't
+		 * know of any wait queue we could sleep on as this
+		 * may be happening on another node. Perhaps the
+		 * proper solution is to queue up requests on the
+		 * other end? */
+
+		/* do we want to yield(); ?? */
+		msleep(50);
+
+		mlog(0, "retrying unlock due to pending recovery/"
+		     "migration/in-progress/reconnect\n");
+		goto retry;
+	}
+
+	if (call_ast) {
+		mlog(0, "calling unlockast(%p, %d)\n", data, status);
+		if (is_master) {
+			/* it is possible that there is one last bast
+			 * pending.  make sure it is flushed, then
+			 * call the unlockast.
+			 * not an issue if this is a mastered remotely,
+			 * since this lock has been removed from the
+			 * lockres queues and cannot be found. */
+			dlm_kick_thread(dlm, NULL);
+			wait_event(dlm->ast_wq,
+				   dlm_lock_basts_flushed(dlm, lock));
+		}
+		(*unlockast)(data, status);
+	}
+
+	if (status == DLM_CANCELGRANT)
+		status = DLM_NORMAL;
+
+	if (status == DLM_NORMAL) {
+		mlog(0, "kicking the thread\n");
+		dlm_kick_thread(dlm, res);
+	} else
+		dlm_error(status);
+
+	dlm_lockres_calc_usage(dlm, res);
+	dlm_lockres_put(res);
+	dlm_lock_put(lock);
+
+	mlog(0, "returning status=%d!\n", status);
+	return status;
+}
+EXPORT_SYMBOL_GPL(dlmunlock);
+
diff --git a/fs/ocfs2/dlmfs/Makefile b/fs/ocfs2/dlmfs/Makefile
new file mode 100644
index 000000000..eed3db8c5
--- /dev/null
+++ b/fs/ocfs2/dlmfs/Makefile
@@ -0,0 +1,5 @@
+ccflags-y := -Ifs/ocfs2
+
+obj-$(CONFIG_OCFS2_FS) += ocfs2_dlmfs.o
+
+ocfs2_dlmfs-objs := userdlm.o dlmfs.o
diff --git a/fs/ocfs2/dlmfs/dlmfs.c b/fs/ocfs2/dlmfs/dlmfs.c
new file mode 100644
index 000000000..b5cf27dcb
--- /dev/null
+++ b/fs/ocfs2/dlmfs/dlmfs.c
@@ -0,0 +1,690 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmfs.c
+ *
+ * Code which implements the kernel side of a minimal userspace
+ * interface to our DLM. This file handles the virtual file system
+ * used for communication with userspace. Credit should go to ramfs,
+ * which was a template for the fs side of this module.
+ *
+ * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+/* Simple VFS hooks based on: */
+/*
+ * Resizable simple ram filesystem for Linux.
+ *
+ * Copyright (C) 2000 Linus Torvalds.
+ *               2000 Transmeta Corp.
+ */
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/pagemap.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/string.h>
+#include <linux/backing-dev.h>
+#include <linux/poll.h>
+
+#include <asm/uaccess.h>
+
+#include "stackglue.h"
+#include "userdlm.h"
+
+#define MLOG_MASK_PREFIX ML_DLMFS
+#include "cluster/masklog.h"
+
+
+static const struct super_operations dlmfs_ops;
+static const struct file_operations dlmfs_file_operations;
+static const struct inode_operations dlmfs_dir_inode_operations;
+static const struct inode_operations dlmfs_root_inode_operations;
+static const struct inode_operations dlmfs_file_inode_operations;
+static struct kmem_cache *dlmfs_inode_cache;
+
+struct workqueue_struct *user_dlm_worker;
+
+
+
+/*
+ * These are the ABI capabilities of dlmfs.
+ *
+ * Over time, dlmfs has added some features that were not part of the
+ * initial ABI.  Unfortunately, some of these features are not detectable
+ * via standard usage.  For example, Linux's default poll always returns
+ * POLLIN, so there is no way for a caller of poll(2) to know when dlmfs
+ * added poll support.  Instead, we provide this list of new capabilities.
+ *
+ * Capabilities is a read-only attribute.  We do it as a module parameter
+ * so we can discover it whether dlmfs is built in, loaded, or even not
+ * loaded.
+ *
+ * The ABI features are local to this machine's dlmfs mount.  This is
+ * distinct from the locking protocol, which is concerned with inter-node
+ * interaction.
+ *
+ * Capabilities:
+ * - bast	: POLLIN against the file descriptor of a held lock
+ *		  signifies a bast fired on the lock.
+ */
+#define DLMFS_CAPABILITIES "bast stackglue"
+static int param_set_dlmfs_capabilities(const char *val,
+					struct kernel_param *kp)
+{
+	printk(KERN_ERR "%s: readonly parameter\n", kp->name);
+	return -EINVAL;
+}
+static int param_get_dlmfs_capabilities(char *buffer,
+					struct kernel_param *kp)
+{
+	return strlcpy(buffer, DLMFS_CAPABILITIES,
+		       strlen(DLMFS_CAPABILITIES) + 1);
+}
+module_param_call(capabilities, param_set_dlmfs_capabilities,
+		  param_get_dlmfs_capabilities, NULL, 0444);
+MODULE_PARM_DESC(capabilities, DLMFS_CAPABILITIES);
+
+
+/*
+ * decodes a set of open flags into a valid lock level and a set of flags.
+ * returns < 0 if we have invalid flags
+ * flags which mean something to us:
+ * O_RDONLY -> PRMODE level
+ * O_WRONLY -> EXMODE level
+ *
+ * O_NONBLOCK -> NOQUEUE
+ */
+static int dlmfs_decode_open_flags(int open_flags,
+				   int *level,
+				   int *flags)
+{
+	if (open_flags & (O_WRONLY|O_RDWR))
+		*level = DLM_LOCK_EX;
+	else
+		*level = DLM_LOCK_PR;
+
+	*flags = 0;
+	if (open_flags & O_NONBLOCK)
+		*flags |= DLM_LKF_NOQUEUE;
+
+	return 0;
+}
+
+static int dlmfs_file_open(struct inode *inode,
+			   struct file *file)
+{
+	int status, level, flags;
+	struct dlmfs_filp_private *fp = NULL;
+	struct dlmfs_inode_private *ip;
+
+	if (S_ISDIR(inode->i_mode))
+		BUG();
+
+	mlog(0, "open called on inode %lu, flags 0x%x\n", inode->i_ino,
+		file->f_flags);
+
+	status = dlmfs_decode_open_flags(file->f_flags, &level, &flags);
+	if (status < 0)
+		goto bail;
+
+	/* We don't want to honor O_APPEND at read/write time as it
+	 * doesn't make sense for LVB writes. */
+	file->f_flags &= ~O_APPEND;
+
+	fp = kmalloc(sizeof(*fp), GFP_NOFS);
+	if (!fp) {
+		status = -ENOMEM;
+		goto bail;
+	}
+	fp->fp_lock_level = level;
+
+	ip = DLMFS_I(inode);
+
+	status = user_dlm_cluster_lock(&ip->ip_lockres, level, flags);
+	if (status < 0) {
+		/* this is a strange error to return here but I want
+		 * to be able userspace to be able to distinguish a
+		 * valid lock request from one that simply couldn't be
+		 * granted. */
+		if (flags & DLM_LKF_NOQUEUE && status == -EAGAIN)
+			status = -ETXTBSY;
+		kfree(fp);
+		goto bail;
+	}
+
+	file->private_data = fp;
+bail:
+	return status;
+}
+
+static int dlmfs_file_release(struct inode *inode,
+			      struct file *file)
+{
+	int level, status;
+	struct dlmfs_inode_private *ip = DLMFS_I(inode);
+	struct dlmfs_filp_private *fp = file->private_data;
+
+	if (S_ISDIR(inode->i_mode))
+		BUG();
+
+	mlog(0, "close called on inode %lu\n", inode->i_ino);
+
+	status = 0;
+	if (fp) {
+		level = fp->fp_lock_level;
+		if (level != DLM_LOCK_IV)
+			user_dlm_cluster_unlock(&ip->ip_lockres, level);
+
+		kfree(fp);
+		file->private_data = NULL;
+	}
+
+	return 0;
+}
+
+/*
+ * We do ->setattr() just to override size changes.  Our size is the size
+ * of the LVB and nothing else.
+ */
+static int dlmfs_file_setattr(struct dentry *dentry, struct iattr *attr)
+{
+	int error;
+	struct inode *inode = d_inode(dentry);
+
+	attr->ia_valid &= ~ATTR_SIZE;
+	error = inode_change_ok(inode, attr);
+	if (error)
+		return error;
+
+	setattr_copy(inode, attr);
+	mark_inode_dirty(inode);
+	return 0;
+}
+
+static unsigned int dlmfs_file_poll(struct file *file, poll_table *wait)
+{
+	int event = 0;
+	struct inode *inode = file_inode(file);
+	struct dlmfs_inode_private *ip = DLMFS_I(inode);
+
+	poll_wait(file, &ip->ip_lockres.l_event, wait);
+
+	spin_lock(&ip->ip_lockres.l_lock);
+	if (ip->ip_lockres.l_flags & USER_LOCK_BLOCKED)
+		event = POLLIN | POLLRDNORM;
+	spin_unlock(&ip->ip_lockres.l_lock);
+
+	return event;
+}
+
+static ssize_t dlmfs_file_read(struct file *filp,
+			       char __user *buf,
+			       size_t count,
+			       loff_t *ppos)
+{
+	int bytes_left;
+	ssize_t readlen, got;
+	char *lvb_buf;
+	struct inode *inode = file_inode(filp);
+
+	mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
+		inode->i_ino, count, *ppos);
+
+	if (*ppos >= i_size_read(inode))
+		return 0;
+
+	if (!count)
+		return 0;
+
+	if (!access_ok(VERIFY_WRITE, buf, count))
+		return -EFAULT;
+
+	/* don't read past the lvb */
+	if ((count + *ppos) > i_size_read(inode))
+		readlen = i_size_read(inode) - *ppos;
+	else
+		readlen = count;
+
+	lvb_buf = kmalloc(readlen, GFP_NOFS);
+	if (!lvb_buf)
+		return -ENOMEM;
+
+	got = user_dlm_read_lvb(inode, lvb_buf, readlen);
+	if (got) {
+		BUG_ON(got != readlen);
+		bytes_left = __copy_to_user(buf, lvb_buf, readlen);
+		readlen -= bytes_left;
+	} else
+		readlen = 0;
+
+	kfree(lvb_buf);
+
+	*ppos = *ppos + readlen;
+
+	mlog(0, "read %zd bytes\n", readlen);
+	return readlen;
+}
+
+static ssize_t dlmfs_file_write(struct file *filp,
+				const char __user *buf,
+				size_t count,
+				loff_t *ppos)
+{
+	int bytes_left;
+	ssize_t writelen;
+	char *lvb_buf;
+	struct inode *inode = file_inode(filp);
+
+	mlog(0, "inode %lu, count = %zu, *ppos = %llu\n",
+		inode->i_ino, count, *ppos);
+
+	if (*ppos >= i_size_read(inode))
+		return -ENOSPC;
+
+	if (!count)
+		return 0;
+
+	if (!access_ok(VERIFY_READ, buf, count))
+		return -EFAULT;
+
+	/* don't write past the lvb */
+	if ((count + *ppos) > i_size_read(inode))
+		writelen = i_size_read(inode) - *ppos;
+	else
+		writelen = count - *ppos;
+
+	lvb_buf = kmalloc(writelen, GFP_NOFS);
+	if (!lvb_buf)
+		return -ENOMEM;
+
+	bytes_left = copy_from_user(lvb_buf, buf, writelen);
+	writelen -= bytes_left;
+	if (writelen)
+		user_dlm_write_lvb(inode, lvb_buf, writelen);
+
+	kfree(lvb_buf);
+
+	*ppos = *ppos + writelen;
+	mlog(0, "wrote %zd bytes\n", writelen);
+	return writelen;
+}
+
+static void dlmfs_init_once(void *foo)
+{
+	struct dlmfs_inode_private *ip =
+		(struct dlmfs_inode_private *) foo;
+
+	ip->ip_conn = NULL;
+	ip->ip_parent = NULL;
+
+	inode_init_once(&ip->ip_vfs_inode);
+}
+
+static struct inode *dlmfs_alloc_inode(struct super_block *sb)
+{
+	struct dlmfs_inode_private *ip;
+
+	ip = kmem_cache_alloc(dlmfs_inode_cache, GFP_NOFS);
+	if (!ip)
+		return NULL;
+
+	return &ip->ip_vfs_inode;
+}
+
+static void dlmfs_i_callback(struct rcu_head *head)
+{
+	struct inode *inode = container_of(head, struct inode, i_rcu);
+	kmem_cache_free(dlmfs_inode_cache, DLMFS_I(inode));
+}
+
+static void dlmfs_destroy_inode(struct inode *inode)
+{
+	call_rcu(&inode->i_rcu, dlmfs_i_callback);
+}
+
+static void dlmfs_evict_inode(struct inode *inode)
+{
+	int status;
+	struct dlmfs_inode_private *ip;
+
+	clear_inode(inode);
+
+	mlog(0, "inode %lu\n", inode->i_ino);
+
+	ip = DLMFS_I(inode);
+
+	if (S_ISREG(inode->i_mode)) {
+		status = user_dlm_destroy_lock(&ip->ip_lockres);
+		if (status < 0)
+			mlog_errno(status);
+		iput(ip->ip_parent);
+		goto clear_fields;
+	}
+
+	mlog(0, "we're a directory, ip->ip_conn = 0x%p\n", ip->ip_conn);
+	/* we must be a directory. If required, lets unregister the
+	 * dlm context now. */
+	if (ip->ip_conn)
+		user_dlm_unregister(ip->ip_conn);
+clear_fields:
+	ip->ip_parent = NULL;
+	ip->ip_conn = NULL;
+}
+
+static struct inode *dlmfs_get_root_inode(struct super_block *sb)
+{
+	struct inode *inode = new_inode(sb);
+	umode_t mode = S_IFDIR | 0755;
+
+	if (inode) {
+		inode->i_ino = get_next_ino();
+		inode_init_owner(inode, NULL, mode);
+		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+		inc_nlink(inode);
+
+		inode->i_fop = &simple_dir_operations;
+		inode->i_op = &dlmfs_root_inode_operations;
+	}
+
+	return inode;
+}
+
+static struct inode *dlmfs_get_inode(struct inode *parent,
+				     struct dentry *dentry,
+				     umode_t mode)
+{
+	struct super_block *sb = parent->i_sb;
+	struct inode * inode = new_inode(sb);
+	struct dlmfs_inode_private *ip;
+
+	if (!inode)
+		return NULL;
+
+	inode->i_ino = get_next_ino();
+	inode_init_owner(inode, parent, mode);
+	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+
+	ip = DLMFS_I(inode);
+	ip->ip_conn = DLMFS_I(parent)->ip_conn;
+
+	switch (mode & S_IFMT) {
+	default:
+		/* for now we don't support anything other than
+		 * directories and regular files. */
+		BUG();
+		break;
+	case S_IFREG:
+		inode->i_op = &dlmfs_file_inode_operations;
+		inode->i_fop = &dlmfs_file_operations;
+
+		i_size_write(inode,  DLM_LVB_LEN);
+
+		user_dlm_lock_res_init(&ip->ip_lockres, dentry);
+
+		/* released at clear_inode time, this insures that we
+		 * get to drop the dlm reference on each lock *before*
+		 * we call the unregister code for releasing parent
+		 * directories. */
+		ip->ip_parent = igrab(parent);
+		BUG_ON(!ip->ip_parent);
+		break;
+	case S_IFDIR:
+		inode->i_op = &dlmfs_dir_inode_operations;
+		inode->i_fop = &simple_dir_operations;
+
+		/* directory inodes start off with i_nlink ==
+		 * 2 (for "." entry) */
+		inc_nlink(inode);
+		break;
+	}
+	return inode;
+}
+
+/*
+ * File creation. Allocate an inode, and we're done..
+ */
+/* SMP-safe */
+static int dlmfs_mkdir(struct inode * dir,
+		       struct dentry * dentry,
+		       umode_t mode)
+{
+	int status;
+	struct inode *inode = NULL;
+	struct qstr *domain = &dentry->d_name;
+	struct dlmfs_inode_private *ip;
+	struct ocfs2_cluster_connection *conn;
+
+	mlog(0, "mkdir %.*s\n", domain->len, domain->name);
+
+	/* verify that we have a proper domain */
+	if (domain->len >= GROUP_NAME_MAX) {
+		status = -EINVAL;
+		mlog(ML_ERROR, "invalid domain name for directory.\n");
+		goto bail;
+	}
+
+	inode = dlmfs_get_inode(dir, dentry, mode | S_IFDIR);
+	if (!inode) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	ip = DLMFS_I(inode);
+
+	conn = user_dlm_register(domain);
+	if (IS_ERR(conn)) {
+		status = PTR_ERR(conn);
+		mlog(ML_ERROR, "Error %d could not register domain \"%.*s\"\n",
+		     status, domain->len, domain->name);
+		goto bail;
+	}
+	ip->ip_conn = conn;
+
+	inc_nlink(dir);
+	d_instantiate(dentry, inode);
+	dget(dentry);	/* Extra count - pin the dentry in core */
+
+	status = 0;
+bail:
+	if (status < 0)
+		iput(inode);
+	return status;
+}
+
+static int dlmfs_create(struct inode *dir,
+			struct dentry *dentry,
+			umode_t mode,
+			bool excl)
+{
+	int status = 0;
+	struct inode *inode;
+	struct qstr *name = &dentry->d_name;
+
+	mlog(0, "create %.*s\n", name->len, name->name);
+
+	/* verify name is valid and doesn't contain any dlm reserved
+	 * characters */
+	if (name->len >= USER_DLM_LOCK_ID_MAX_LEN ||
+	    name->name[0] == '$') {
+		status = -EINVAL;
+		mlog(ML_ERROR, "invalid lock name, %.*s\n", name->len,
+		     name->name);
+		goto bail;
+	}
+
+	inode = dlmfs_get_inode(dir, dentry, mode | S_IFREG);
+	if (!inode) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	d_instantiate(dentry, inode);
+	dget(dentry);	/* Extra count - pin the dentry in core */
+bail:
+	return status;
+}
+
+static int dlmfs_unlink(struct inode *dir,
+			struct dentry *dentry)
+{
+	int status;
+	struct inode *inode = d_inode(dentry);
+
+	mlog(0, "unlink inode %lu\n", inode->i_ino);
+
+	/* if there are no current holders, or none that are waiting
+	 * to acquire a lock, this basically destroys our lockres. */
+	status = user_dlm_destroy_lock(&DLMFS_I(inode)->ip_lockres);
+	if (status < 0) {
+		mlog(ML_ERROR, "unlink %pd, error %d from destroy\n",
+		     dentry, status);
+		goto bail;
+	}
+	status = simple_unlink(dir, dentry);
+bail:
+	return status;
+}
+
+static int dlmfs_fill_super(struct super_block * sb,
+			    void * data,
+			    int silent)
+{
+	sb->s_maxbytes = MAX_LFS_FILESIZE;
+	sb->s_blocksize = PAGE_CACHE_SIZE;
+	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
+	sb->s_magic = DLMFS_MAGIC;
+	sb->s_op = &dlmfs_ops;
+	sb->s_root = d_make_root(dlmfs_get_root_inode(sb));
+	if (!sb->s_root)
+		return -ENOMEM;
+	return 0;
+}
+
+static const struct file_operations dlmfs_file_operations = {
+	.open		= dlmfs_file_open,
+	.release	= dlmfs_file_release,
+	.poll		= dlmfs_file_poll,
+	.read		= dlmfs_file_read,
+	.write		= dlmfs_file_write,
+	.llseek		= default_llseek,
+};
+
+static const struct inode_operations dlmfs_dir_inode_operations = {
+	.create		= dlmfs_create,
+	.lookup		= simple_lookup,
+	.unlink		= dlmfs_unlink,
+};
+
+/* this way we can restrict mkdir to only the toplevel of the fs. */
+static const struct inode_operations dlmfs_root_inode_operations = {
+	.lookup		= simple_lookup,
+	.mkdir		= dlmfs_mkdir,
+	.rmdir		= simple_rmdir,
+};
+
+static const struct super_operations dlmfs_ops = {
+	.statfs		= simple_statfs,
+	.alloc_inode	= dlmfs_alloc_inode,
+	.destroy_inode	= dlmfs_destroy_inode,
+	.evict_inode	= dlmfs_evict_inode,
+	.drop_inode	= generic_delete_inode,
+};
+
+static const struct inode_operations dlmfs_file_inode_operations = {
+	.getattr	= simple_getattr,
+	.setattr	= dlmfs_file_setattr,
+};
+
+static struct dentry *dlmfs_mount(struct file_system_type *fs_type,
+	int flags, const char *dev_name, void *data)
+{
+	return mount_nodev(fs_type, flags, data, dlmfs_fill_super);
+}
+
+static struct file_system_type dlmfs_fs_type = {
+	.owner		= THIS_MODULE,
+	.name		= "ocfs2_dlmfs",
+	.mount		= dlmfs_mount,
+	.kill_sb	= kill_litter_super,
+};
+MODULE_ALIAS_FS("ocfs2_dlmfs");
+
+static int __init init_dlmfs_fs(void)
+{
+	int status;
+	int cleanup_inode = 0, cleanup_worker = 0;
+
+	dlmfs_inode_cache = kmem_cache_create("dlmfs_inode_cache",
+				sizeof(struct dlmfs_inode_private),
+				0, (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
+					SLAB_MEM_SPREAD),
+				dlmfs_init_once);
+	if (!dlmfs_inode_cache) {
+		status = -ENOMEM;
+		goto bail;
+	}
+	cleanup_inode = 1;
+
+	user_dlm_worker = create_singlethread_workqueue("user_dlm");
+	if (!user_dlm_worker) {
+		status = -ENOMEM;
+		goto bail;
+	}
+	cleanup_worker = 1;
+
+	user_dlm_set_locking_protocol();
+	status = register_filesystem(&dlmfs_fs_type);
+bail:
+	if (status) {
+		if (cleanup_inode)
+			kmem_cache_destroy(dlmfs_inode_cache);
+		if (cleanup_worker)
+			destroy_workqueue(user_dlm_worker);
+	} else
+		printk("OCFS2 User DLM kernel interface loaded\n");
+	return status;
+}
+
+static void __exit exit_dlmfs_fs(void)
+{
+	unregister_filesystem(&dlmfs_fs_type);
+
+	flush_workqueue(user_dlm_worker);
+	destroy_workqueue(user_dlm_worker);
+
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
+	kmem_cache_destroy(dlmfs_inode_cache);
+
+}
+
+MODULE_AUTHOR("Oracle");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("OCFS2 DLM-Filesystem");
+
+module_init(init_dlmfs_fs)
+module_exit(exit_dlmfs_fs)
diff --git a/fs/ocfs2/dlmfs/userdlm.c b/fs/ocfs2/dlmfs/userdlm.c
new file mode 100644
index 000000000..0499e3fb7
--- /dev/null
+++ b/fs/ocfs2/dlmfs/userdlm.c
@@ -0,0 +1,688 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * userdlm.c
+ *
+ * Code which implements the kernel side of a minimal userspace
+ * interface to our DLM.
+ *
+ * Many of the functions here are pared down versions of dlmglue.c
+ * functions.
+ *
+ * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/signal.h>
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/crc32.h>
+
+#include "ocfs2_lockingver.h"
+#include "stackglue.h"
+#include "userdlm.h"
+
+#define MLOG_MASK_PREFIX ML_DLMFS
+#include "cluster/masklog.h"
+
+
+static inline struct user_lock_res *user_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
+{
+	return container_of(lksb, struct user_lock_res, l_lksb);
+}
+
+static inline int user_check_wait_flag(struct user_lock_res *lockres,
+				       int flag)
+{
+	int ret;
+
+	spin_lock(&lockres->l_lock);
+	ret = lockres->l_flags & flag;
+	spin_unlock(&lockres->l_lock);
+
+	return ret;
+}
+
+static inline void user_wait_on_busy_lock(struct user_lock_res *lockres)
+
+{
+	wait_event(lockres->l_event,
+		   !user_check_wait_flag(lockres, USER_LOCK_BUSY));
+}
+
+static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres)
+
+{
+	wait_event(lockres->l_event,
+		   !user_check_wait_flag(lockres, USER_LOCK_BLOCKED));
+}
+
+/* I heart container_of... */
+static inline struct ocfs2_cluster_connection *
+cluster_connection_from_user_lockres(struct user_lock_res *lockres)
+{
+	struct dlmfs_inode_private *ip;
+
+	ip = container_of(lockres,
+			  struct dlmfs_inode_private,
+			  ip_lockres);
+	return ip->ip_conn;
+}
+
+static struct inode *
+user_dlm_inode_from_user_lockres(struct user_lock_res *lockres)
+{
+	struct dlmfs_inode_private *ip;
+
+	ip = container_of(lockres,
+			  struct dlmfs_inode_private,
+			  ip_lockres);
+	return &ip->ip_vfs_inode;
+}
+
+static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
+{
+	spin_lock(&lockres->l_lock);
+	lockres->l_flags &= ~USER_LOCK_BUSY;
+	spin_unlock(&lockres->l_lock);
+}
+
+#define user_log_dlm_error(_func, _stat, _lockres) do {			\
+	mlog(ML_ERROR, "Dlm error %d while calling %s on "		\
+		"resource %.*s\n", _stat, _func,			\
+		_lockres->l_namelen, _lockres->l_name); 		\
+} while (0)
+
+/* WARNING: This function lives in a world where the only three lock
+ * levels are EX, PR, and NL. It *will* have to be adjusted when more
+ * lock types are added. */
+static inline int user_highest_compat_lock_level(int level)
+{
+	int new_level = DLM_LOCK_EX;
+
+	if (level == DLM_LOCK_EX)
+		new_level = DLM_LOCK_NL;
+	else if (level == DLM_LOCK_PR)
+		new_level = DLM_LOCK_PR;
+	return new_level;
+}
+
+static void user_ast(struct ocfs2_dlm_lksb *lksb)
+{
+	struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
+	int status;
+
+	mlog(ML_BASTS, "AST fired for lockres %.*s, level %d => %d\n",
+	     lockres->l_namelen, lockres->l_name, lockres->l_level,
+	     lockres->l_requested);
+
+	spin_lock(&lockres->l_lock);
+
+	status = ocfs2_dlm_lock_status(&lockres->l_lksb);
+	if (status) {
+		mlog(ML_ERROR, "lksb status value of %u on lockres %.*s\n",
+		     status, lockres->l_namelen, lockres->l_name);
+		spin_unlock(&lockres->l_lock);
+		return;
+	}
+
+	mlog_bug_on_msg(lockres->l_requested == DLM_LOCK_IV,
+			"Lockres %.*s, requested ivmode. flags 0x%x\n",
+			lockres->l_namelen, lockres->l_name, lockres->l_flags);
+
+	/* we're downconverting. */
+	if (lockres->l_requested < lockres->l_level) {
+		if (lockres->l_requested <=
+		    user_highest_compat_lock_level(lockres->l_blocking)) {
+			lockres->l_blocking = DLM_LOCK_NL;
+			lockres->l_flags &= ~USER_LOCK_BLOCKED;
+		}
+	}
+
+	lockres->l_level = lockres->l_requested;
+	lockres->l_requested = DLM_LOCK_IV;
+	lockres->l_flags |= USER_LOCK_ATTACHED;
+	lockres->l_flags &= ~USER_LOCK_BUSY;
+
+	spin_unlock(&lockres->l_lock);
+
+	wake_up(&lockres->l_event);
+}
+
+static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres)
+{
+	struct inode *inode;
+	inode = user_dlm_inode_from_user_lockres(lockres);
+	if (!igrab(inode))
+		BUG();
+}
+
+static void user_dlm_unblock_lock(struct work_struct *work);
+
+static void __user_dlm_queue_lockres(struct user_lock_res *lockres)
+{
+	if (!(lockres->l_flags & USER_LOCK_QUEUED)) {
+		user_dlm_grab_inode_ref(lockres);
+
+		INIT_WORK(&lockres->l_work, user_dlm_unblock_lock);
+
+		queue_work(user_dlm_worker, &lockres->l_work);
+		lockres->l_flags |= USER_LOCK_QUEUED;
+	}
+}
+
+static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres)
+{
+	int queue = 0;
+
+	if (!(lockres->l_flags & USER_LOCK_BLOCKED))
+		return;
+
+	switch (lockres->l_blocking) {
+	case DLM_LOCK_EX:
+		if (!lockres->l_ex_holders && !lockres->l_ro_holders)
+			queue = 1;
+		break;
+	case DLM_LOCK_PR:
+		if (!lockres->l_ex_holders)
+			queue = 1;
+		break;
+	default:
+		BUG();
+	}
+
+	if (queue)
+		__user_dlm_queue_lockres(lockres);
+}
+
+static void user_bast(struct ocfs2_dlm_lksb *lksb, int level)
+{
+	struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
+
+	mlog(ML_BASTS, "BAST fired for lockres %.*s, blocking %d, level %d\n",
+	     lockres->l_namelen, lockres->l_name, level, lockres->l_level);
+
+	spin_lock(&lockres->l_lock);
+	lockres->l_flags |= USER_LOCK_BLOCKED;
+	if (level > lockres->l_blocking)
+		lockres->l_blocking = level;
+
+	__user_dlm_queue_lockres(lockres);
+	spin_unlock(&lockres->l_lock);
+
+	wake_up(&lockres->l_event);
+}
+
+static void user_unlock_ast(struct ocfs2_dlm_lksb *lksb, int status)
+{
+	struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
+
+	mlog(ML_BASTS, "UNLOCK AST fired for lockres %.*s, flags 0x%x\n",
+	     lockres->l_namelen, lockres->l_name, lockres->l_flags);
+
+	if (status)
+		mlog(ML_ERROR, "dlm returns status %d\n", status);
+
+	spin_lock(&lockres->l_lock);
+	/* The teardown flag gets set early during the unlock process,
+	 * so test the cancel flag to make sure that this ast isn't
+	 * for a concurrent cancel. */
+	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN
+	    && !(lockres->l_flags & USER_LOCK_IN_CANCEL)) {
+		lockres->l_level = DLM_LOCK_IV;
+	} else if (status == DLM_CANCELGRANT) {
+		/* We tried to cancel a convert request, but it was
+		 * already granted. Don't clear the busy flag - the
+		 * ast should've done this already. */
+		BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
+		lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
+		goto out_noclear;
+	} else {
+		BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
+		/* Cancel succeeded, we want to re-queue */
+		lockres->l_requested = DLM_LOCK_IV; /* cancel an
+						    * upconvert
+						    * request. */
+		lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
+		/* we want the unblock thread to look at it again
+		 * now. */
+		if (lockres->l_flags & USER_LOCK_BLOCKED)
+			__user_dlm_queue_lockres(lockres);
+	}
+
+	lockres->l_flags &= ~USER_LOCK_BUSY;
+out_noclear:
+	spin_unlock(&lockres->l_lock);
+
+	wake_up(&lockres->l_event);
+}
+
+/*
+ * This is the userdlmfs locking protocol version.
+ *
+ * See fs/ocfs2/dlmglue.c for more details on locking versions.
+ */
+static struct ocfs2_locking_protocol user_dlm_lproto = {
+	.lp_max_version = {
+		.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
+		.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
+	},
+	.lp_lock_ast		= user_ast,
+	.lp_blocking_ast	= user_bast,
+	.lp_unlock_ast		= user_unlock_ast,
+};
+
+static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres)
+{
+	struct inode *inode;
+	inode = user_dlm_inode_from_user_lockres(lockres);
+	iput(inode);
+}
+
+static void user_dlm_unblock_lock(struct work_struct *work)
+{
+	int new_level, status;
+	struct user_lock_res *lockres =
+		container_of(work, struct user_lock_res, l_work);
+	struct ocfs2_cluster_connection *conn =
+		cluster_connection_from_user_lockres(lockres);
+
+	mlog(0, "lockres %.*s\n", lockres->l_namelen, lockres->l_name);
+
+	spin_lock(&lockres->l_lock);
+
+	mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
+			"Lockres %.*s, flags 0x%x\n",
+			lockres->l_namelen, lockres->l_name, lockres->l_flags);
+
+	/* notice that we don't clear USER_LOCK_BLOCKED here. If it's
+	 * set, we want user_ast clear it. */
+	lockres->l_flags &= ~USER_LOCK_QUEUED;
+
+	/* It's valid to get here and no longer be blocked - if we get
+	 * several basts in a row, we might be queued by the first
+	 * one, the unblock thread might run and clear the queued
+	 * flag, and finally we might get another bast which re-queues
+	 * us before our ast for the downconvert is called. */
+	if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
+		mlog(ML_BASTS, "lockres %.*s USER_LOCK_BLOCKED\n",
+		     lockres->l_namelen, lockres->l_name);
+		spin_unlock(&lockres->l_lock);
+		goto drop_ref;
+	}
+
+	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
+		mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_TEARDOWN\n",
+		     lockres->l_namelen, lockres->l_name);
+		spin_unlock(&lockres->l_lock);
+		goto drop_ref;
+	}
+
+	if (lockres->l_flags & USER_LOCK_BUSY) {
+		if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
+			mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_CANCEL\n",
+			     lockres->l_namelen, lockres->l_name);
+			spin_unlock(&lockres->l_lock);
+			goto drop_ref;
+		}
+
+		lockres->l_flags |= USER_LOCK_IN_CANCEL;
+		spin_unlock(&lockres->l_lock);
+
+		status = ocfs2_dlm_unlock(conn, &lockres->l_lksb,
+					  DLM_LKF_CANCEL);
+		if (status)
+			user_log_dlm_error("ocfs2_dlm_unlock", status, lockres);
+		goto drop_ref;
+	}
+
+	/* If there are still incompat holders, we can exit safely
+	 * without worrying about re-queueing this lock as that will
+	 * happen on the last call to user_cluster_unlock. */
+	if ((lockres->l_blocking == DLM_LOCK_EX)
+	    && (lockres->l_ex_holders || lockres->l_ro_holders)) {
+		spin_unlock(&lockres->l_lock);
+		mlog(ML_BASTS, "lockres %.*s, EX/PR Holders %u,%u\n",
+		     lockres->l_namelen, lockres->l_name,
+		     lockres->l_ex_holders, lockres->l_ro_holders);
+		goto drop_ref;
+	}
+
+	if ((lockres->l_blocking == DLM_LOCK_PR)
+	    && lockres->l_ex_holders) {
+		spin_unlock(&lockres->l_lock);
+		mlog(ML_BASTS, "lockres %.*s, EX Holders %u\n",
+		     lockres->l_namelen, lockres->l_name,
+		     lockres->l_ex_holders);
+		goto drop_ref;
+	}
+
+	/* yay, we can downconvert now. */
+	new_level = user_highest_compat_lock_level(lockres->l_blocking);
+	lockres->l_requested = new_level;
+	lockres->l_flags |= USER_LOCK_BUSY;
+	mlog(ML_BASTS, "lockres %.*s, downconvert %d => %d\n",
+	     lockres->l_namelen, lockres->l_name, lockres->l_level, new_level);
+	spin_unlock(&lockres->l_lock);
+
+	/* need lock downconvert request now... */
+	status = ocfs2_dlm_lock(conn, new_level, &lockres->l_lksb,
+				DLM_LKF_CONVERT|DLM_LKF_VALBLK,
+				lockres->l_name,
+				lockres->l_namelen);
+	if (status) {
+		user_log_dlm_error("ocfs2_dlm_lock", status, lockres);
+		user_recover_from_dlm_error(lockres);
+	}
+
+drop_ref:
+	user_dlm_drop_inode_ref(lockres);
+}
+
+static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
+					int level)
+{
+	switch(level) {
+	case DLM_LOCK_EX:
+		lockres->l_ex_holders++;
+		break;
+	case DLM_LOCK_PR:
+		lockres->l_ro_holders++;
+		break;
+	default:
+		BUG();
+	}
+}
+
+/* predict what lock level we'll be dropping down to on behalf
+ * of another node, and return true if the currently wanted
+ * level will be compatible with it. */
+static inline int
+user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
+				  int wanted)
+{
+	BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
+
+	return wanted <= user_highest_compat_lock_level(lockres->l_blocking);
+}
+
+int user_dlm_cluster_lock(struct user_lock_res *lockres,
+			  int level,
+			  int lkm_flags)
+{
+	int status, local_flags;
+	struct ocfs2_cluster_connection *conn =
+		cluster_connection_from_user_lockres(lockres);
+
+	if (level != DLM_LOCK_EX &&
+	    level != DLM_LOCK_PR) {
+		mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
+		     lockres->l_namelen, lockres->l_name);
+		status = -EINVAL;
+		goto bail;
+	}
+
+	mlog(ML_BASTS, "lockres %.*s, level %d, flags = 0x%x\n",
+	     lockres->l_namelen, lockres->l_name, level, lkm_flags);
+
+again:
+	if (signal_pending(current)) {
+		status = -ERESTARTSYS;
+		goto bail;
+	}
+
+	spin_lock(&lockres->l_lock);
+
+	/* We only compare against the currently granted level
+	 * here. If the lock is blocked waiting on a downconvert,
+	 * we'll get caught below. */
+	if ((lockres->l_flags & USER_LOCK_BUSY) &&
+	    (level > lockres->l_level)) {
+		/* is someone sitting in dlm_lock? If so, wait on
+		 * them. */
+		spin_unlock(&lockres->l_lock);
+
+		user_wait_on_busy_lock(lockres);
+		goto again;
+	}
+
+	if ((lockres->l_flags & USER_LOCK_BLOCKED) &&
+	    (!user_may_continue_on_blocked_lock(lockres, level))) {
+		/* is the lock is currently blocked on behalf of
+		 * another node */
+		spin_unlock(&lockres->l_lock);
+
+		user_wait_on_blocked_lock(lockres);
+		goto again;
+	}
+
+	if (level > lockres->l_level) {
+		local_flags = lkm_flags | DLM_LKF_VALBLK;
+		if (lockres->l_level != DLM_LOCK_IV)
+			local_flags |= DLM_LKF_CONVERT;
+
+		lockres->l_requested = level;
+		lockres->l_flags |= USER_LOCK_BUSY;
+		spin_unlock(&lockres->l_lock);
+
+		BUG_ON(level == DLM_LOCK_IV);
+		BUG_ON(level == DLM_LOCK_NL);
+
+		/* call dlm_lock to upgrade lock now */
+		status = ocfs2_dlm_lock(conn, level, &lockres->l_lksb,
+					local_flags, lockres->l_name,
+					lockres->l_namelen);
+		if (status) {
+			if ((lkm_flags & DLM_LKF_NOQUEUE) &&
+			    (status != -EAGAIN))
+				user_log_dlm_error("ocfs2_dlm_lock",
+						   status, lockres);
+			user_recover_from_dlm_error(lockres);
+			goto bail;
+		}
+
+		user_wait_on_busy_lock(lockres);
+		goto again;
+	}
+
+	user_dlm_inc_holders(lockres, level);
+	spin_unlock(&lockres->l_lock);
+
+	status = 0;
+bail:
+	return status;
+}
+
+static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
+					int level)
+{
+	switch(level) {
+	case DLM_LOCK_EX:
+		BUG_ON(!lockres->l_ex_holders);
+		lockres->l_ex_holders--;
+		break;
+	case DLM_LOCK_PR:
+		BUG_ON(!lockres->l_ro_holders);
+		lockres->l_ro_holders--;
+		break;
+	default:
+		BUG();
+	}
+}
+
+void user_dlm_cluster_unlock(struct user_lock_res *lockres,
+			     int level)
+{
+	if (level != DLM_LOCK_EX &&
+	    level != DLM_LOCK_PR) {
+		mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
+		     lockres->l_namelen, lockres->l_name);
+		return;
+	}
+
+	spin_lock(&lockres->l_lock);
+	user_dlm_dec_holders(lockres, level);
+	__user_dlm_cond_queue_lockres(lockres);
+	spin_unlock(&lockres->l_lock);
+}
+
+void user_dlm_write_lvb(struct inode *inode,
+			const char *val,
+			unsigned int len)
+{
+	struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
+	char *lvb;
+
+	BUG_ON(len > DLM_LVB_LEN);
+
+	spin_lock(&lockres->l_lock);
+
+	BUG_ON(lockres->l_level < DLM_LOCK_EX);
+	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+	memcpy(lvb, val, len);
+
+	spin_unlock(&lockres->l_lock);
+}
+
+ssize_t user_dlm_read_lvb(struct inode *inode,
+			  char *val,
+			  unsigned int len)
+{
+	struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
+	char *lvb;
+	ssize_t ret = len;
+
+	BUG_ON(len > DLM_LVB_LEN);
+
+	spin_lock(&lockres->l_lock);
+
+	BUG_ON(lockres->l_level < DLM_LOCK_PR);
+	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)) {
+		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+		memcpy(val, lvb, len);
+	} else
+		ret = 0;
+
+	spin_unlock(&lockres->l_lock);
+	return ret;
+}
+
+void user_dlm_lock_res_init(struct user_lock_res *lockres,
+			    struct dentry *dentry)
+{
+	memset(lockres, 0, sizeof(*lockres));
+
+	spin_lock_init(&lockres->l_lock);
+	init_waitqueue_head(&lockres->l_event);
+	lockres->l_level = DLM_LOCK_IV;
+	lockres->l_requested = DLM_LOCK_IV;
+	lockres->l_blocking = DLM_LOCK_IV;
+
+	/* should have been checked before getting here. */
+	BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);
+
+	memcpy(lockres->l_name,
+	       dentry->d_name.name,
+	       dentry->d_name.len);
+	lockres->l_namelen = dentry->d_name.len;
+}
+
+int user_dlm_destroy_lock(struct user_lock_res *lockres)
+{
+	int status = -EBUSY;
+	struct ocfs2_cluster_connection *conn =
+		cluster_connection_from_user_lockres(lockres);
+
+	mlog(ML_BASTS, "lockres %.*s\n", lockres->l_namelen, lockres->l_name);
+
+	spin_lock(&lockres->l_lock);
+	if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
+		spin_unlock(&lockres->l_lock);
+		return 0;
+	}
+
+	lockres->l_flags |= USER_LOCK_IN_TEARDOWN;
+
+	while (lockres->l_flags & USER_LOCK_BUSY) {
+		spin_unlock(&lockres->l_lock);
+
+		user_wait_on_busy_lock(lockres);
+
+		spin_lock(&lockres->l_lock);
+	}
+
+	if (lockres->l_ro_holders || lockres->l_ex_holders) {
+		spin_unlock(&lockres->l_lock);
+		goto bail;
+	}
+
+	status = 0;
+	if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
+		spin_unlock(&lockres->l_lock);
+		goto bail;
+	}
+
+	lockres->l_flags &= ~USER_LOCK_ATTACHED;
+	lockres->l_flags |= USER_LOCK_BUSY;
+	spin_unlock(&lockres->l_lock);
+
+	status = ocfs2_dlm_unlock(conn, &lockres->l_lksb, DLM_LKF_VALBLK);
+	if (status) {
+		user_log_dlm_error("ocfs2_dlm_unlock", status, lockres);
+		goto bail;
+	}
+
+	user_wait_on_busy_lock(lockres);
+
+	status = 0;
+bail:
+	return status;
+}
+
+static void user_dlm_recovery_handler_noop(int node_num,
+					   void *recovery_data)
+{
+	/* We ignore recovery events */
+	return;
+}
+
+void user_dlm_set_locking_protocol(void)
+{
+	ocfs2_stack_glue_set_max_proto_version(&user_dlm_lproto.lp_max_version);
+}
+
+struct ocfs2_cluster_connection *user_dlm_register(struct qstr *name)
+{
+	int rc;
+	struct ocfs2_cluster_connection *conn;
+
+	rc = ocfs2_cluster_connect_agnostic(name->name, name->len,
+					    &user_dlm_lproto,
+					    user_dlm_recovery_handler_noop,
+					    NULL, &conn);
+	if (rc)
+		mlog_errno(rc);
+
+	return rc ? ERR_PTR(rc) : conn;
+}
+
+void user_dlm_unregister(struct ocfs2_cluster_connection *conn)
+{
+	ocfs2_cluster_disconnect(conn, 0);
+}
diff --git a/fs/ocfs2/dlmfs/userdlm.h b/fs/ocfs2/dlmfs/userdlm.h
new file mode 100644
index 000000000..3b42d7953
--- /dev/null
+++ b/fs/ocfs2/dlmfs/userdlm.h
@@ -0,0 +1,113 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * userdlm.h
+ *
+ * Userspace dlm defines
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+
+#ifndef USERDLM_H
+#define USERDLM_H
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/workqueue.h>
+
+/* user_lock_res->l_flags flags. */
+#define USER_LOCK_ATTACHED      (0x00000001) /* we have initialized
+					       * the lvb */
+#define USER_LOCK_BUSY          (0x00000002) /* we are currently in
+					       * dlm_lock */
+#define USER_LOCK_BLOCKED       (0x00000004) /* blocked waiting to
+					      * downconvert*/
+#define USER_LOCK_IN_TEARDOWN   (0x00000008) /* we're currently
+					      * destroying this
+					      * lock. */
+#define USER_LOCK_QUEUED        (0x00000010) /* lock is on the
+					      * workqueue */
+#define USER_LOCK_IN_CANCEL     (0x00000020)
+
+struct user_lock_res {
+	spinlock_t               l_lock;
+
+	int                      l_flags;
+
+#define USER_DLM_LOCK_ID_MAX_LEN  32
+	char                     l_name[USER_DLM_LOCK_ID_MAX_LEN];
+	int                      l_namelen;
+	int                      l_level;
+	unsigned int             l_ro_holders;
+	unsigned int             l_ex_holders;
+	struct ocfs2_dlm_lksb    l_lksb;
+
+	int                      l_requested;
+	int                      l_blocking;
+
+	wait_queue_head_t        l_event;
+
+	struct work_struct       l_work;
+};
+
+extern struct workqueue_struct *user_dlm_worker;
+
+void user_dlm_lock_res_init(struct user_lock_res *lockres,
+			    struct dentry *dentry);
+int user_dlm_destroy_lock(struct user_lock_res *lockres);
+int user_dlm_cluster_lock(struct user_lock_res *lockres,
+			  int level,
+			  int lkm_flags);
+void user_dlm_cluster_unlock(struct user_lock_res *lockres,
+			     int level);
+void user_dlm_write_lvb(struct inode *inode,
+			const char *val,
+			unsigned int len);
+ssize_t user_dlm_read_lvb(struct inode *inode,
+			  char *val,
+			  unsigned int len);
+struct ocfs2_cluster_connection *user_dlm_register(struct qstr *name);
+void user_dlm_unregister(struct ocfs2_cluster_connection *conn);
+void user_dlm_set_locking_protocol(void);
+
+struct dlmfs_inode_private {
+	struct ocfs2_cluster_connection	*ip_conn;
+
+	struct user_lock_res ip_lockres; /* unused for directories. */
+	struct inode         *ip_parent;
+
+	struct inode         ip_vfs_inode;
+};
+
+static inline struct dlmfs_inode_private *
+DLMFS_I(struct inode *inode)
+{
+        return container_of(inode,
+			    struct dlmfs_inode_private,
+			    ip_vfs_inode);
+}
+
+struct dlmfs_filp_private {
+	int                  fp_lock_level;
+};
+
+#define DLMFS_MAGIC	0x76a9f425
+
+#endif /* USERDLM_H */
diff --git a/fs/ocfs2/dlmglue.c b/fs/ocfs2/dlmglue.c
new file mode 100644
index 000000000..8b23aa2f5
--- /dev/null
+++ b/fs/ocfs2/dlmglue.c
@@ -0,0 +1,4106 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmglue.c
+ *
+ * Code which implements an OCFS2 specific interface to our DLM.
+ *
+ * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/mm.h>
+#include <linux/kthread.h>
+#include <linux/pagemap.h>
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
+#include <linux/time.h>
+#include <linux/quotaops.h>
+
+#define MLOG_MASK_PREFIX ML_DLM_GLUE
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+#include "ocfs2_lockingver.h"
+
+#include "alloc.h"
+#include "dcache.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "file.h"
+#include "heartbeat.h"
+#include "inode.h"
+#include "journal.h"
+#include "stackglue.h"
+#include "slot_map.h"
+#include "super.h"
+#include "uptodate.h"
+#include "quota.h"
+#include "refcounttree.h"
+
+#include "buffer_head_io.h"
+
+struct ocfs2_mask_waiter {
+	struct list_head	mw_item;
+	int			mw_status;
+	struct completion	mw_complete;
+	unsigned long		mw_mask;
+	unsigned long		mw_goal;
+#ifdef CONFIG_OCFS2_FS_STATS
+	ktime_t			mw_lock_start;
+#endif
+};
+
+static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres);
+static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres);
+static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres);
+static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres);
+
+/*
+ * Return value from ->downconvert_worker functions.
+ *
+ * These control the precise actions of ocfs2_unblock_lock()
+ * and ocfs2_process_blocked_lock()
+ *
+ */
+enum ocfs2_unblock_action {
+	UNBLOCK_CONTINUE	= 0, /* Continue downconvert */
+	UNBLOCK_CONTINUE_POST	= 1, /* Continue downconvert, fire
+				      * ->post_unlock callback */
+	UNBLOCK_STOP_POST	= 2, /* Do not downconvert, fire
+				      * ->post_unlock() callback. */
+};
+
+struct ocfs2_unblock_ctl {
+	int requeue;
+	enum ocfs2_unblock_action unblock_action;
+};
+
+/* Lockdep class keys */
+struct lock_class_key lockdep_keys[OCFS2_NUM_LOCK_TYPES];
+
+static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
+					int new_level);
+static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres);
+
+static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
+				     int blocking);
+
+static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
+				       int blocking);
+
+static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
+				     struct ocfs2_lock_res *lockres);
+
+static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres);
+
+static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
+					    int new_level);
+static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
+					 int blocking);
+
+#define mlog_meta_lvb(__level, __lockres) ocfs2_dump_meta_lvb_info(__level, __PRETTY_FUNCTION__, __LINE__, __lockres)
+
+/* This aids in debugging situations where a bad LVB might be involved. */
+static void ocfs2_dump_meta_lvb_info(u64 level,
+				     const char *function,
+				     unsigned int line,
+				     struct ocfs2_lock_res *lockres)
+{
+	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+
+	mlog(level, "LVB information for %s (called from %s:%u):\n",
+	     lockres->l_name, function, line);
+	mlog(level, "version: %u, clusters: %u, generation: 0x%x\n",
+	     lvb->lvb_version, be32_to_cpu(lvb->lvb_iclusters),
+	     be32_to_cpu(lvb->lvb_igeneration));
+	mlog(level, "size: %llu, uid %u, gid %u, mode 0x%x\n",
+	     (unsigned long long)be64_to_cpu(lvb->lvb_isize),
+	     be32_to_cpu(lvb->lvb_iuid), be32_to_cpu(lvb->lvb_igid),
+	     be16_to_cpu(lvb->lvb_imode));
+	mlog(level, "nlink %u, atime_packed 0x%llx, ctime_packed 0x%llx, "
+	     "mtime_packed 0x%llx iattr 0x%x\n", be16_to_cpu(lvb->lvb_inlink),
+	     (long long)be64_to_cpu(lvb->lvb_iatime_packed),
+	     (long long)be64_to_cpu(lvb->lvb_ictime_packed),
+	     (long long)be64_to_cpu(lvb->lvb_imtime_packed),
+	     be32_to_cpu(lvb->lvb_iattr));
+}
+
+
+/*
+ * OCFS2 Lock Resource Operations
+ *
+ * These fine tune the behavior of the generic dlmglue locking infrastructure.
+ *
+ * The most basic of lock types can point ->l_priv to their respective
+ * struct ocfs2_super and allow the default actions to manage things.
+ *
+ * Right now, each lock type also needs to implement an init function,
+ * and trivial lock/unlock wrappers. ocfs2_simple_drop_lockres()
+ * should be called when the lock is no longer needed (i.e., object
+ * destruction time).
+ */
+struct ocfs2_lock_res_ops {
+	/*
+	 * Translate an ocfs2_lock_res * into an ocfs2_super *. Define
+	 * this callback if ->l_priv is not an ocfs2_super pointer
+	 */
+	struct ocfs2_super * (*get_osb)(struct ocfs2_lock_res *);
+
+	/*
+	 * Optionally called in the downconvert thread after a
+	 * successful downconvert. The lockres will not be referenced
+	 * after this callback is called, so it is safe to free
+	 * memory, etc.
+	 *
+	 * The exact semantics of when this is called are controlled
+	 * by ->downconvert_worker()
+	 */
+	void (*post_unlock)(struct ocfs2_super *, struct ocfs2_lock_res *);
+
+	/*
+	 * Allow a lock type to add checks to determine whether it is
+	 * safe to downconvert a lock. Return 0 to re-queue the
+	 * downconvert at a later time, nonzero to continue.
+	 *
+	 * For most locks, the default checks that there are no
+	 * incompatible holders are sufficient.
+	 *
+	 * Called with the lockres spinlock held.
+	 */
+	int (*check_downconvert)(struct ocfs2_lock_res *, int);
+
+	/*
+	 * Allows a lock type to populate the lock value block. This
+	 * is called on downconvert, and when we drop a lock.
+	 *
+	 * Locks that want to use this should set LOCK_TYPE_USES_LVB
+	 * in the flags field.
+	 *
+	 * Called with the lockres spinlock held.
+	 */
+	void (*set_lvb)(struct ocfs2_lock_res *);
+
+	/*
+	 * Called from the downconvert thread when it is determined
+	 * that a lock will be downconverted. This is called without
+	 * any locks held so the function can do work that might
+	 * schedule (syncing out data, etc).
+	 *
+	 * This should return any one of the ocfs2_unblock_action
+	 * values, depending on what it wants the thread to do.
+	 */
+	int (*downconvert_worker)(struct ocfs2_lock_res *, int);
+
+	/*
+	 * LOCK_TYPE_* flags which describe the specific requirements
+	 * of a lock type. Descriptions of each individual flag follow.
+	 */
+	int flags;
+};
+
+/*
+ * Some locks want to "refresh" potentially stale data when a
+ * meaningful (PRMODE or EXMODE) lock level is first obtained. If this
+ * flag is set, the OCFS2_LOCK_NEEDS_REFRESH flag will be set on the
+ * individual lockres l_flags member from the ast function. It is
+ * expected that the locking wrapper will clear the
+ * OCFS2_LOCK_NEEDS_REFRESH flag when done.
+ */
+#define LOCK_TYPE_REQUIRES_REFRESH 0x1
+
+/*
+ * Indicate that a lock type makes use of the lock value block. The
+ * ->set_lvb lock type callback must be defined.
+ */
+#define LOCK_TYPE_USES_LVB		0x2
+
+static struct ocfs2_lock_res_ops ocfs2_inode_rw_lops = {
+	.get_osb	= ocfs2_get_inode_osb,
+	.flags		= 0,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_inode_inode_lops = {
+	.get_osb	= ocfs2_get_inode_osb,
+	.check_downconvert = ocfs2_check_meta_downconvert,
+	.set_lvb	= ocfs2_set_meta_lvb,
+	.downconvert_worker = ocfs2_data_convert_worker,
+	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_super_lops = {
+	.flags		= LOCK_TYPE_REQUIRES_REFRESH,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_rename_lops = {
+	.flags		= 0,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_nfs_sync_lops = {
+	.flags		= 0,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_orphan_scan_lops = {
+	.flags		= LOCK_TYPE_REQUIRES_REFRESH|LOCK_TYPE_USES_LVB,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_dentry_lops = {
+	.get_osb	= ocfs2_get_dentry_osb,
+	.post_unlock	= ocfs2_dentry_post_unlock,
+	.downconvert_worker = ocfs2_dentry_convert_worker,
+	.flags		= 0,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_inode_open_lops = {
+	.get_osb	= ocfs2_get_inode_osb,
+	.flags		= 0,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_flock_lops = {
+	.get_osb	= ocfs2_get_file_osb,
+	.flags		= 0,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_qinfo_lops = {
+	.set_lvb	= ocfs2_set_qinfo_lvb,
+	.get_osb	= ocfs2_get_qinfo_osb,
+	.flags		= LOCK_TYPE_REQUIRES_REFRESH | LOCK_TYPE_USES_LVB,
+};
+
+static struct ocfs2_lock_res_ops ocfs2_refcount_block_lops = {
+	.check_downconvert = ocfs2_check_refcount_downconvert,
+	.downconvert_worker = ocfs2_refcount_convert_worker,
+	.flags		= 0,
+};
+
+static inline int ocfs2_is_inode_lock(struct ocfs2_lock_res *lockres)
+{
+	return lockres->l_type == OCFS2_LOCK_TYPE_META ||
+		lockres->l_type == OCFS2_LOCK_TYPE_RW ||
+		lockres->l_type == OCFS2_LOCK_TYPE_OPEN;
+}
+
+static inline struct ocfs2_lock_res *ocfs2_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
+{
+	return container_of(lksb, struct ocfs2_lock_res, l_lksb);
+}
+
+static inline struct inode *ocfs2_lock_res_inode(struct ocfs2_lock_res *lockres)
+{
+	BUG_ON(!ocfs2_is_inode_lock(lockres));
+
+	return (struct inode *) lockres->l_priv;
+}
+
+static inline struct ocfs2_dentry_lock *ocfs2_lock_res_dl(struct ocfs2_lock_res *lockres)
+{
+	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_DENTRY);
+
+	return (struct ocfs2_dentry_lock *)lockres->l_priv;
+}
+
+static inline struct ocfs2_mem_dqinfo *ocfs2_lock_res_qinfo(struct ocfs2_lock_res *lockres)
+{
+	BUG_ON(lockres->l_type != OCFS2_LOCK_TYPE_QINFO);
+
+	return (struct ocfs2_mem_dqinfo *)lockres->l_priv;
+}
+
+static inline struct ocfs2_refcount_tree *
+ocfs2_lock_res_refcount_tree(struct ocfs2_lock_res *res)
+{
+	return container_of(res, struct ocfs2_refcount_tree, rf_lockres);
+}
+
+static inline struct ocfs2_super *ocfs2_get_lockres_osb(struct ocfs2_lock_res *lockres)
+{
+	if (lockres->l_ops->get_osb)
+		return lockres->l_ops->get_osb(lockres);
+
+	return (struct ocfs2_super *)lockres->l_priv;
+}
+
+static int ocfs2_lock_create(struct ocfs2_super *osb,
+			     struct ocfs2_lock_res *lockres,
+			     int level,
+			     u32 dlm_flags);
+static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
+						     int wanted);
+static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
+				   struct ocfs2_lock_res *lockres,
+				   int level, unsigned long caller_ip);
+static inline void ocfs2_cluster_unlock(struct ocfs2_super *osb,
+					struct ocfs2_lock_res *lockres,
+					int level)
+{
+	__ocfs2_cluster_unlock(osb, lockres, level, _RET_IP_);
+}
+
+static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres);
+static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres);
+static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres);
+static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres, int level);
+static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
+					struct ocfs2_lock_res *lockres);
+static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
+						int convert);
+#define ocfs2_log_dlm_error(_func, _err, _lockres) do {					\
+	if ((_lockres)->l_type != OCFS2_LOCK_TYPE_DENTRY)				\
+		mlog(ML_ERROR, "DLM error %d while calling %s on resource %s\n",	\
+		     _err, _func, _lockres->l_name);					\
+	else										\
+		mlog(ML_ERROR, "DLM error %d while calling %s on resource %.*s%08x\n",	\
+		     _err, _func, OCFS2_DENTRY_LOCK_INO_START - 1, (_lockres)->l_name,	\
+		     (unsigned int)ocfs2_get_dentry_lock_ino(_lockres));		\
+} while (0)
+static int ocfs2_downconvert_thread(void *arg);
+static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
+					struct ocfs2_lock_res *lockres);
+static int ocfs2_inode_lock_update(struct inode *inode,
+				  struct buffer_head **bh);
+static void ocfs2_drop_osb_locks(struct ocfs2_super *osb);
+static inline int ocfs2_highest_compat_lock_level(int level);
+static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
+					      int new_level);
+static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
+				  struct ocfs2_lock_res *lockres,
+				  int new_level,
+				  int lvb,
+				  unsigned int generation);
+static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
+				        struct ocfs2_lock_res *lockres);
+static int ocfs2_cancel_convert(struct ocfs2_super *osb,
+				struct ocfs2_lock_res *lockres);
+
+
+static void ocfs2_build_lock_name(enum ocfs2_lock_type type,
+				  u64 blkno,
+				  u32 generation,
+				  char *name)
+{
+	int len;
+
+	BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
+
+	len = snprintf(name, OCFS2_LOCK_ID_MAX_LEN, "%c%s%016llx%08x",
+		       ocfs2_lock_type_char(type), OCFS2_LOCK_ID_PAD,
+		       (long long)blkno, generation);
+
+	BUG_ON(len != (OCFS2_LOCK_ID_MAX_LEN - 1));
+
+	mlog(0, "built lock resource with name: %s\n", name);
+}
+
+static DEFINE_SPINLOCK(ocfs2_dlm_tracking_lock);
+
+static void ocfs2_add_lockres_tracking(struct ocfs2_lock_res *res,
+				       struct ocfs2_dlm_debug *dlm_debug)
+{
+	mlog(0, "Add tracking for lockres %s\n", res->l_name);
+
+	spin_lock(&ocfs2_dlm_tracking_lock);
+	list_add(&res->l_debug_list, &dlm_debug->d_lockres_tracking);
+	spin_unlock(&ocfs2_dlm_tracking_lock);
+}
+
+static void ocfs2_remove_lockres_tracking(struct ocfs2_lock_res *res)
+{
+	spin_lock(&ocfs2_dlm_tracking_lock);
+	if (!list_empty(&res->l_debug_list))
+		list_del_init(&res->l_debug_list);
+	spin_unlock(&ocfs2_dlm_tracking_lock);
+}
+
+#ifdef CONFIG_OCFS2_FS_STATS
+static void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
+{
+	res->l_lock_refresh = 0;
+	memset(&res->l_lock_prmode, 0, sizeof(struct ocfs2_lock_stats));
+	memset(&res->l_lock_exmode, 0, sizeof(struct ocfs2_lock_stats));
+}
+
+static void ocfs2_update_lock_stats(struct ocfs2_lock_res *res, int level,
+				    struct ocfs2_mask_waiter *mw, int ret)
+{
+	u32 usec;
+	ktime_t kt;
+	struct ocfs2_lock_stats *stats;
+
+	if (level == LKM_PRMODE)
+		stats = &res->l_lock_prmode;
+	else if (level == LKM_EXMODE)
+		stats = &res->l_lock_exmode;
+	else
+		return;
+
+	kt = ktime_sub(ktime_get(), mw->mw_lock_start);
+	usec = ktime_to_us(kt);
+
+	stats->ls_gets++;
+	stats->ls_total += ktime_to_ns(kt);
+	/* overflow */
+	if (unlikely(stats->ls_gets == 0)) {
+		stats->ls_gets++;
+		stats->ls_total = ktime_to_ns(kt);
+	}
+
+	if (stats->ls_max < usec)
+		stats->ls_max = usec;
+
+	if (ret)
+		stats->ls_fail++;
+}
+
+static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
+{
+	lockres->l_lock_refresh++;
+}
+
+static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
+{
+	mw->mw_lock_start = ktime_get();
+}
+#else
+static inline void ocfs2_init_lock_stats(struct ocfs2_lock_res *res)
+{
+}
+static inline void ocfs2_update_lock_stats(struct ocfs2_lock_res *res,
+			   int level, struct ocfs2_mask_waiter *mw, int ret)
+{
+}
+static inline void ocfs2_track_lock_refresh(struct ocfs2_lock_res *lockres)
+{
+}
+static inline void ocfs2_init_start_time(struct ocfs2_mask_waiter *mw)
+{
+}
+#endif
+
+static void ocfs2_lock_res_init_common(struct ocfs2_super *osb,
+				       struct ocfs2_lock_res *res,
+				       enum ocfs2_lock_type type,
+				       struct ocfs2_lock_res_ops *ops,
+				       void *priv)
+{
+	res->l_type          = type;
+	res->l_ops           = ops;
+	res->l_priv          = priv;
+
+	res->l_level         = DLM_LOCK_IV;
+	res->l_requested     = DLM_LOCK_IV;
+	res->l_blocking      = DLM_LOCK_IV;
+	res->l_action        = OCFS2_AST_INVALID;
+	res->l_unlock_action = OCFS2_UNLOCK_INVALID;
+
+	res->l_flags         = OCFS2_LOCK_INITIALIZED;
+
+	ocfs2_add_lockres_tracking(res, osb->osb_dlm_debug);
+
+	ocfs2_init_lock_stats(res);
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+	if (type != OCFS2_LOCK_TYPE_OPEN)
+		lockdep_init_map(&res->l_lockdep_map, ocfs2_lock_type_strings[type],
+				 &lockdep_keys[type], 0);
+	else
+		res->l_lockdep_map.key = NULL;
+#endif
+}
+
+void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res)
+{
+	/* This also clears out the lock status block */
+	memset(res, 0, sizeof(struct ocfs2_lock_res));
+	spin_lock_init(&res->l_lock);
+	init_waitqueue_head(&res->l_event);
+	INIT_LIST_HEAD(&res->l_blocked_list);
+	INIT_LIST_HEAD(&res->l_mask_waiters);
+}
+
+void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
+			       enum ocfs2_lock_type type,
+			       unsigned int generation,
+			       struct inode *inode)
+{
+	struct ocfs2_lock_res_ops *ops;
+
+	switch(type) {
+		case OCFS2_LOCK_TYPE_RW:
+			ops = &ocfs2_inode_rw_lops;
+			break;
+		case OCFS2_LOCK_TYPE_META:
+			ops = &ocfs2_inode_inode_lops;
+			break;
+		case OCFS2_LOCK_TYPE_OPEN:
+			ops = &ocfs2_inode_open_lops;
+			break;
+		default:
+			mlog_bug_on_msg(1, "type: %d\n", type);
+			ops = NULL; /* thanks, gcc */
+			break;
+	};
+
+	ocfs2_build_lock_name(type, OCFS2_I(inode)->ip_blkno,
+			      generation, res->l_name);
+	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), res, type, ops, inode);
+}
+
+static struct ocfs2_super *ocfs2_get_inode_osb(struct ocfs2_lock_res *lockres)
+{
+	struct inode *inode = ocfs2_lock_res_inode(lockres);
+
+	return OCFS2_SB(inode->i_sb);
+}
+
+static struct ocfs2_super *ocfs2_get_qinfo_osb(struct ocfs2_lock_res *lockres)
+{
+	struct ocfs2_mem_dqinfo *info = lockres->l_priv;
+
+	return OCFS2_SB(info->dqi_gi.dqi_sb);
+}
+
+static struct ocfs2_super *ocfs2_get_file_osb(struct ocfs2_lock_res *lockres)
+{
+	struct ocfs2_file_private *fp = lockres->l_priv;
+
+	return OCFS2_SB(fp->fp_file->f_mapping->host->i_sb);
+}
+
+static __u64 ocfs2_get_dentry_lock_ino(struct ocfs2_lock_res *lockres)
+{
+	__be64 inode_blkno_be;
+
+	memcpy(&inode_blkno_be, &lockres->l_name[OCFS2_DENTRY_LOCK_INO_START],
+	       sizeof(__be64));
+
+	return be64_to_cpu(inode_blkno_be);
+}
+
+static struct ocfs2_super *ocfs2_get_dentry_osb(struct ocfs2_lock_res *lockres)
+{
+	struct ocfs2_dentry_lock *dl = lockres->l_priv;
+
+	return OCFS2_SB(dl->dl_inode->i_sb);
+}
+
+void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
+				u64 parent, struct inode *inode)
+{
+	int len;
+	u64 inode_blkno = OCFS2_I(inode)->ip_blkno;
+	__be64 inode_blkno_be = cpu_to_be64(inode_blkno);
+	struct ocfs2_lock_res *lockres = &dl->dl_lockres;
+
+	ocfs2_lock_res_init_once(lockres);
+
+	/*
+	 * Unfortunately, the standard lock naming scheme won't work
+	 * here because we have two 16 byte values to use. Instead,
+	 * we'll stuff the inode number as a binary value. We still
+	 * want error prints to show something without garbling the
+	 * display, so drop a null byte in there before the inode
+	 * number. A future version of OCFS2 will likely use all
+	 * binary lock names. The stringified names have been a
+	 * tremendous aid in debugging, but now that the debugfs
+	 * interface exists, we can mangle things there if need be.
+	 *
+	 * NOTE: We also drop the standard "pad" value (the total lock
+	 * name size stays the same though - the last part is all
+	 * zeros due to the memset in ocfs2_lock_res_init_once()
+	 */
+	len = snprintf(lockres->l_name, OCFS2_DENTRY_LOCK_INO_START,
+		       "%c%016llx",
+		       ocfs2_lock_type_char(OCFS2_LOCK_TYPE_DENTRY),
+		       (long long)parent);
+
+	BUG_ON(len != (OCFS2_DENTRY_LOCK_INO_START - 1));
+
+	memcpy(&lockres->l_name[OCFS2_DENTRY_LOCK_INO_START], &inode_blkno_be,
+	       sizeof(__be64));
+
+	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
+				   OCFS2_LOCK_TYPE_DENTRY, &ocfs2_dentry_lops,
+				   dl);
+}
+
+static void ocfs2_super_lock_res_init(struct ocfs2_lock_res *res,
+				      struct ocfs2_super *osb)
+{
+	/* Superblock lockres doesn't come from a slab so we call init
+	 * once on it manually.  */
+	ocfs2_lock_res_init_once(res);
+	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_SUPER, OCFS2_SUPER_BLOCK_BLKNO,
+			      0, res->l_name);
+	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_SUPER,
+				   &ocfs2_super_lops, osb);
+}
+
+static void ocfs2_rename_lock_res_init(struct ocfs2_lock_res *res,
+				       struct ocfs2_super *osb)
+{
+	/* Rename lockres doesn't come from a slab so we call init
+	 * once on it manually.  */
+	ocfs2_lock_res_init_once(res);
+	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_RENAME, 0, 0, res->l_name);
+	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_RENAME,
+				   &ocfs2_rename_lops, osb);
+}
+
+static void ocfs2_nfs_sync_lock_res_init(struct ocfs2_lock_res *res,
+					 struct ocfs2_super *osb)
+{
+	/* nfs_sync lockres doesn't come from a slab so we call init
+	 * once on it manually.  */
+	ocfs2_lock_res_init_once(res);
+	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_NFS_SYNC, 0, 0, res->l_name);
+	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_NFS_SYNC,
+				   &ocfs2_nfs_sync_lops, osb);
+}
+
+static void ocfs2_orphan_scan_lock_res_init(struct ocfs2_lock_res *res,
+					    struct ocfs2_super *osb)
+{
+	ocfs2_lock_res_init_once(res);
+	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_ORPHAN_SCAN, 0, 0, res->l_name);
+	ocfs2_lock_res_init_common(osb, res, OCFS2_LOCK_TYPE_ORPHAN_SCAN,
+				   &ocfs2_orphan_scan_lops, osb);
+}
+
+void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
+			      struct ocfs2_file_private *fp)
+{
+	struct inode *inode = fp->fp_file->f_mapping->host;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	ocfs2_lock_res_init_once(lockres);
+	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_FLOCK, oi->ip_blkno,
+			      inode->i_generation, lockres->l_name);
+	ocfs2_lock_res_init_common(OCFS2_SB(inode->i_sb), lockres,
+				   OCFS2_LOCK_TYPE_FLOCK, &ocfs2_flock_lops,
+				   fp);
+	lockres->l_flags |= OCFS2_LOCK_NOCACHE;
+}
+
+void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
+			       struct ocfs2_mem_dqinfo *info)
+{
+	ocfs2_lock_res_init_once(lockres);
+	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_QINFO, info->dqi_gi.dqi_type,
+			      0, lockres->l_name);
+	ocfs2_lock_res_init_common(OCFS2_SB(info->dqi_gi.dqi_sb), lockres,
+				   OCFS2_LOCK_TYPE_QINFO, &ocfs2_qinfo_lops,
+				   info);
+}
+
+void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
+				  struct ocfs2_super *osb, u64 ref_blkno,
+				  unsigned int generation)
+{
+	ocfs2_lock_res_init_once(lockres);
+	ocfs2_build_lock_name(OCFS2_LOCK_TYPE_REFCOUNT, ref_blkno,
+			      generation, lockres->l_name);
+	ocfs2_lock_res_init_common(osb, lockres, OCFS2_LOCK_TYPE_REFCOUNT,
+				   &ocfs2_refcount_block_lops, osb);
+}
+
+void ocfs2_lock_res_free(struct ocfs2_lock_res *res)
+{
+	if (!(res->l_flags & OCFS2_LOCK_INITIALIZED))
+		return;
+
+	ocfs2_remove_lockres_tracking(res);
+
+	mlog_bug_on_msg(!list_empty(&res->l_blocked_list),
+			"Lockres %s is on the blocked list\n",
+			res->l_name);
+	mlog_bug_on_msg(!list_empty(&res->l_mask_waiters),
+			"Lockres %s has mask waiters pending\n",
+			res->l_name);
+	mlog_bug_on_msg(spin_is_locked(&res->l_lock),
+			"Lockres %s is locked\n",
+			res->l_name);
+	mlog_bug_on_msg(res->l_ro_holders,
+			"Lockres %s has %u ro holders\n",
+			res->l_name, res->l_ro_holders);
+	mlog_bug_on_msg(res->l_ex_holders,
+			"Lockres %s has %u ex holders\n",
+			res->l_name, res->l_ex_holders);
+
+	/* Need to clear out the lock status block for the dlm */
+	memset(&res->l_lksb, 0, sizeof(res->l_lksb));
+
+	res->l_flags = 0UL;
+}
+
+static inline void ocfs2_inc_holders(struct ocfs2_lock_res *lockres,
+				     int level)
+{
+	BUG_ON(!lockres);
+
+	switch(level) {
+	case DLM_LOCK_EX:
+		lockres->l_ex_holders++;
+		break;
+	case DLM_LOCK_PR:
+		lockres->l_ro_holders++;
+		break;
+	default:
+		BUG();
+	}
+}
+
+static inline void ocfs2_dec_holders(struct ocfs2_lock_res *lockres,
+				     int level)
+{
+	BUG_ON(!lockres);
+
+	switch(level) {
+	case DLM_LOCK_EX:
+		BUG_ON(!lockres->l_ex_holders);
+		lockres->l_ex_holders--;
+		break;
+	case DLM_LOCK_PR:
+		BUG_ON(!lockres->l_ro_holders);
+		lockres->l_ro_holders--;
+		break;
+	default:
+		BUG();
+	}
+}
+
+/* WARNING: This function lives in a world where the only three lock
+ * levels are EX, PR, and NL. It *will* have to be adjusted when more
+ * lock types are added. */
+static inline int ocfs2_highest_compat_lock_level(int level)
+{
+	int new_level = DLM_LOCK_EX;
+
+	if (level == DLM_LOCK_EX)
+		new_level = DLM_LOCK_NL;
+	else if (level == DLM_LOCK_PR)
+		new_level = DLM_LOCK_PR;
+	return new_level;
+}
+
+static void lockres_set_flags(struct ocfs2_lock_res *lockres,
+			      unsigned long newflags)
+{
+	struct ocfs2_mask_waiter *mw, *tmp;
+
+ 	assert_spin_locked(&lockres->l_lock);
+
+	lockres->l_flags = newflags;
+
+	list_for_each_entry_safe(mw, tmp, &lockres->l_mask_waiters, mw_item) {
+		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
+			continue;
+
+		list_del_init(&mw->mw_item);
+		mw->mw_status = 0;
+		complete(&mw->mw_complete);
+	}
+}
+static void lockres_or_flags(struct ocfs2_lock_res *lockres, unsigned long or)
+{
+	lockres_set_flags(lockres, lockres->l_flags | or);
+}
+static void lockres_clear_flags(struct ocfs2_lock_res *lockres,
+				unsigned long clear)
+{
+	lockres_set_flags(lockres, lockres->l_flags & ~clear);
+}
+
+static inline void ocfs2_generic_handle_downconvert_action(struct ocfs2_lock_res *lockres)
+{
+	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
+	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
+	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
+	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
+
+	lockres->l_level = lockres->l_requested;
+	if (lockres->l_level <=
+	    ocfs2_highest_compat_lock_level(lockres->l_blocking)) {
+		lockres->l_blocking = DLM_LOCK_NL;
+		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
+	}
+	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
+}
+
+static inline void ocfs2_generic_handle_convert_action(struct ocfs2_lock_res *lockres)
+{
+	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
+	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_ATTACHED));
+
+	/* Convert from RO to EX doesn't really need anything as our
+	 * information is already up to data. Convert from NL to
+	 * *anything* however should mark ourselves as needing an
+	 * update */
+	if (lockres->l_level == DLM_LOCK_NL &&
+	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
+		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
+
+	lockres->l_level = lockres->l_requested;
+
+	/*
+	 * We set the OCFS2_LOCK_UPCONVERT_FINISHING flag before clearing
+	 * the OCFS2_LOCK_BUSY flag to prevent the dc thread from
+	 * downconverting the lock before the upconvert has fully completed.
+	 * Do not prevent the dc thread from downconverting if NONBLOCK lock
+	 * had already returned.
+	 */
+	if (!(lockres->l_flags & OCFS2_LOCK_NONBLOCK_FINISHED))
+		lockres_or_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
+	else
+		lockres_clear_flags(lockres, OCFS2_LOCK_NONBLOCK_FINISHED);
+
+	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
+}
+
+static inline void ocfs2_generic_handle_attach_action(struct ocfs2_lock_res *lockres)
+{
+	BUG_ON((!(lockres->l_flags & OCFS2_LOCK_BUSY)));
+	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
+
+	if (lockres->l_requested > DLM_LOCK_NL &&
+	    !(lockres->l_flags & OCFS2_LOCK_LOCAL) &&
+	    lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
+		lockres_or_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
+
+	lockres->l_level = lockres->l_requested;
+	lockres_or_flags(lockres, OCFS2_LOCK_ATTACHED);
+	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
+}
+
+static int ocfs2_generic_handle_bast(struct ocfs2_lock_res *lockres,
+				     int level)
+{
+	int needs_downconvert = 0;
+
+	assert_spin_locked(&lockres->l_lock);
+
+	if (level > lockres->l_blocking) {
+		/* only schedule a downconvert if we haven't already scheduled
+		 * one that goes low enough to satisfy the level we're
+		 * blocking.  this also catches the case where we get
+		 * duplicate BASTs */
+		if (ocfs2_highest_compat_lock_level(level) <
+		    ocfs2_highest_compat_lock_level(lockres->l_blocking))
+			needs_downconvert = 1;
+
+		lockres->l_blocking = level;
+	}
+
+	mlog(ML_BASTS, "lockres %s, block %d, level %d, l_block %d, dwn %d\n",
+	     lockres->l_name, level, lockres->l_level, lockres->l_blocking,
+	     needs_downconvert);
+
+	if (needs_downconvert)
+		lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
+	mlog(0, "needs_downconvert = %d\n", needs_downconvert);
+	return needs_downconvert;
+}
+
+/*
+ * OCFS2_LOCK_PENDING and l_pending_gen.
+ *
+ * Why does OCFS2_LOCK_PENDING exist?  To close a race between setting
+ * OCFS2_LOCK_BUSY and calling ocfs2_dlm_lock().  See ocfs2_unblock_lock()
+ * for more details on the race.
+ *
+ * OCFS2_LOCK_PENDING closes the race quite nicely.  However, it introduces
+ * a race on itself.  In o2dlm, we can get the ast before ocfs2_dlm_lock()
+ * returns.  The ast clears OCFS2_LOCK_BUSY, and must therefore clear
+ * OCFS2_LOCK_PENDING at the same time.  When ocfs2_dlm_lock() returns,
+ * the caller is going to try to clear PENDING again.  If nothing else is
+ * happening, __lockres_clear_pending() sees PENDING is unset and does
+ * nothing.
+ *
+ * But what if another path (eg downconvert thread) has just started a
+ * new locking action?  The other path has re-set PENDING.  Our path
+ * cannot clear PENDING, because that will re-open the original race
+ * window.
+ *
+ * [Example]
+ *
+ * ocfs2_meta_lock()
+ *  ocfs2_cluster_lock()
+ *   set BUSY
+ *   set PENDING
+ *   drop l_lock
+ *   ocfs2_dlm_lock()
+ *    ocfs2_locking_ast()		ocfs2_downconvert_thread()
+ *     clear PENDING			 ocfs2_unblock_lock()
+ *					  take_l_lock
+ *					  !BUSY
+ *					  ocfs2_prepare_downconvert()
+ *					   set BUSY
+ *					   set PENDING
+ *					  drop l_lock
+ *   take l_lock
+ *   clear PENDING
+ *   drop l_lock
+ *			<window>
+ *					  ocfs2_dlm_lock()
+ *
+ * So as you can see, we now have a window where l_lock is not held,
+ * PENDING is not set, and ocfs2_dlm_lock() has not been called.
+ *
+ * The core problem is that ocfs2_cluster_lock() has cleared the PENDING
+ * set by ocfs2_prepare_downconvert().  That wasn't nice.
+ *
+ * To solve this we introduce l_pending_gen.  A call to
+ * lockres_clear_pending() will only do so when it is passed a generation
+ * number that matches the lockres.  lockres_set_pending() will return the
+ * current generation number.  When ocfs2_cluster_lock() goes to clear
+ * PENDING, it passes the generation it got from set_pending().  In our
+ * example above, the generation numbers will *not* match.  Thus,
+ * ocfs2_cluster_lock() will not clear the PENDING set by
+ * ocfs2_prepare_downconvert().
+ */
+
+/* Unlocked version for ocfs2_locking_ast() */
+static void __lockres_clear_pending(struct ocfs2_lock_res *lockres,
+				    unsigned int generation,
+				    struct ocfs2_super *osb)
+{
+	assert_spin_locked(&lockres->l_lock);
+
+	/*
+	 * The ast and locking functions can race us here.  The winner
+	 * will clear pending, the loser will not.
+	 */
+	if (!(lockres->l_flags & OCFS2_LOCK_PENDING) ||
+	    (lockres->l_pending_gen != generation))
+		return;
+
+	lockres_clear_flags(lockres, OCFS2_LOCK_PENDING);
+	lockres->l_pending_gen++;
+
+	/*
+	 * The downconvert thread may have skipped us because we
+	 * were PENDING.  Wake it up.
+	 */
+	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
+		ocfs2_wake_downconvert_thread(osb);
+}
+
+/* Locked version for callers of ocfs2_dlm_lock() */
+static void lockres_clear_pending(struct ocfs2_lock_res *lockres,
+				  unsigned int generation,
+				  struct ocfs2_super *osb)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	__lockres_clear_pending(lockres, generation, osb);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+}
+
+static unsigned int lockres_set_pending(struct ocfs2_lock_res *lockres)
+{
+	assert_spin_locked(&lockres->l_lock);
+	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BUSY));
+
+	lockres_or_flags(lockres, OCFS2_LOCK_PENDING);
+
+	return lockres->l_pending_gen;
+}
+
+static void ocfs2_blocking_ast(struct ocfs2_dlm_lksb *lksb, int level)
+{
+	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
+	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
+	int needs_downconvert;
+	unsigned long flags;
+
+	BUG_ON(level <= DLM_LOCK_NL);
+
+	mlog(ML_BASTS, "BAST fired for lockres %s, blocking %d, level %d, "
+	     "type %s\n", lockres->l_name, level, lockres->l_level,
+	     ocfs2_lock_type_string(lockres->l_type));
+
+	/*
+	 * We can skip the bast for locks which don't enable caching -
+	 * they'll be dropped at the earliest possible time anyway.
+	 */
+	if (lockres->l_flags & OCFS2_LOCK_NOCACHE)
+		return;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	needs_downconvert = ocfs2_generic_handle_bast(lockres, level);
+	if (needs_downconvert)
+		ocfs2_schedule_blocked_lock(osb, lockres);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	wake_up(&lockres->l_event);
+
+	ocfs2_wake_downconvert_thread(osb);
+}
+
+static void ocfs2_locking_ast(struct ocfs2_dlm_lksb *lksb)
+{
+	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
+	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
+	unsigned long flags;
+	int status;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+
+	status = ocfs2_dlm_lock_status(&lockres->l_lksb);
+
+	if (status == -EAGAIN) {
+		lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
+		goto out;
+	}
+
+	if (status) {
+		mlog(ML_ERROR, "lockres %s: lksb status value of %d!\n",
+		     lockres->l_name, status);
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		return;
+	}
+
+	mlog(ML_BASTS, "AST fired for lockres %s, action %d, unlock %d, "
+	     "level %d => %d\n", lockres->l_name, lockres->l_action,
+	     lockres->l_unlock_action, lockres->l_level, lockres->l_requested);
+
+	switch(lockres->l_action) {
+	case OCFS2_AST_ATTACH:
+		ocfs2_generic_handle_attach_action(lockres);
+		lockres_clear_flags(lockres, OCFS2_LOCK_LOCAL);
+		break;
+	case OCFS2_AST_CONVERT:
+		ocfs2_generic_handle_convert_action(lockres);
+		break;
+	case OCFS2_AST_DOWNCONVERT:
+		ocfs2_generic_handle_downconvert_action(lockres);
+		break;
+	default:
+		mlog(ML_ERROR, "lockres %s: AST fired with invalid action: %u, "
+		     "flags 0x%lx, unlock: %u\n",
+		     lockres->l_name, lockres->l_action, lockres->l_flags,
+		     lockres->l_unlock_action);
+		BUG();
+	}
+out:
+	/* set it to something invalid so if we get called again we
+	 * can catch it. */
+	lockres->l_action = OCFS2_AST_INVALID;
+
+	/* Did we try to cancel this lock?  Clear that state */
+	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT)
+		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
+
+	/*
+	 * We may have beaten the locking functions here.  We certainly
+	 * know that dlm_lock() has been called :-)
+	 * Because we can't have two lock calls in flight at once, we
+	 * can use lockres->l_pending_gen.
+	 */
+	__lockres_clear_pending(lockres, lockres->l_pending_gen,  osb);
+
+	wake_up(&lockres->l_event);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+}
+
+static void ocfs2_unlock_ast(struct ocfs2_dlm_lksb *lksb, int error)
+{
+	struct ocfs2_lock_res *lockres = ocfs2_lksb_to_lock_res(lksb);
+	unsigned long flags;
+
+	mlog(ML_BASTS, "UNLOCK AST fired for lockres %s, action = %d\n",
+	     lockres->l_name, lockres->l_unlock_action);
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	if (error) {
+		mlog(ML_ERROR, "Dlm passes error %d for lock %s, "
+		     "unlock_action %d\n", error, lockres->l_name,
+		     lockres->l_unlock_action);
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		return;
+	}
+
+	switch(lockres->l_unlock_action) {
+	case OCFS2_UNLOCK_CANCEL_CONVERT:
+		mlog(0, "Cancel convert success for %s\n", lockres->l_name);
+		lockres->l_action = OCFS2_AST_INVALID;
+		/* Downconvert thread may have requeued this lock, we
+		 * need to wake it. */
+		if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
+			ocfs2_wake_downconvert_thread(ocfs2_get_lockres_osb(lockres));
+		break;
+	case OCFS2_UNLOCK_DROP_LOCK:
+		lockres->l_level = DLM_LOCK_IV;
+		break;
+	default:
+		BUG();
+	}
+
+	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
+	lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
+	wake_up(&lockres->l_event);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+}
+
+/*
+ * This is the filesystem locking protocol.  It provides the lock handling
+ * hooks for the underlying DLM.  It has a maximum version number.
+ * The version number allows interoperability with systems running at
+ * the same major number and an equal or smaller minor number.
+ *
+ * Whenever the filesystem does new things with locks (adds or removes a
+ * lock, orders them differently, does different things underneath a lock),
+ * the version must be changed.  The protocol is negotiated when joining
+ * the dlm domain.  A node may join the domain if its major version is
+ * identical to all other nodes and its minor version is greater than
+ * or equal to all other nodes.  When its minor version is greater than
+ * the other nodes, it will run at the minor version specified by the
+ * other nodes.
+ *
+ * If a locking change is made that will not be compatible with older
+ * versions, the major number must be increased and the minor version set
+ * to zero.  If a change merely adds a behavior that can be disabled when
+ * speaking to older versions, the minor version must be increased.  If a
+ * change adds a fully backwards compatible change (eg, LVB changes that
+ * are just ignored by older versions), the version does not need to be
+ * updated.
+ */
+static struct ocfs2_locking_protocol lproto = {
+	.lp_max_version = {
+		.pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
+		.pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
+	},
+	.lp_lock_ast		= ocfs2_locking_ast,
+	.lp_blocking_ast	= ocfs2_blocking_ast,
+	.lp_unlock_ast		= ocfs2_unlock_ast,
+};
+
+void ocfs2_set_locking_protocol(void)
+{
+	ocfs2_stack_glue_set_max_proto_version(&lproto.lp_max_version);
+}
+
+static inline void ocfs2_recover_from_dlm_error(struct ocfs2_lock_res *lockres,
+						int convert)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	lockres_clear_flags(lockres, OCFS2_LOCK_BUSY);
+	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
+	if (convert)
+		lockres->l_action = OCFS2_AST_INVALID;
+	else
+		lockres->l_unlock_action = OCFS2_UNLOCK_INVALID;
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	wake_up(&lockres->l_event);
+}
+
+/* Note: If we detect another process working on the lock (i.e.,
+ * OCFS2_LOCK_BUSY), we'll bail out returning 0. It's up to the caller
+ * to do the right thing in that case.
+ */
+static int ocfs2_lock_create(struct ocfs2_super *osb,
+			     struct ocfs2_lock_res *lockres,
+			     int level,
+			     u32 dlm_flags)
+{
+	int ret = 0;
+	unsigned long flags;
+	unsigned int gen;
+
+	mlog(0, "lock %s, level = %d, flags = %u\n", lockres->l_name, level,
+	     dlm_flags);
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	if ((lockres->l_flags & OCFS2_LOCK_ATTACHED) ||
+	    (lockres->l_flags & OCFS2_LOCK_BUSY)) {
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		goto bail;
+	}
+
+	lockres->l_action = OCFS2_AST_ATTACH;
+	lockres->l_requested = level;
+	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
+	gen = lockres_set_pending(lockres);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	ret = ocfs2_dlm_lock(osb->cconn,
+			     level,
+			     &lockres->l_lksb,
+			     dlm_flags,
+			     lockres->l_name,
+			     OCFS2_LOCK_ID_MAX_LEN - 1);
+	lockres_clear_pending(lockres, gen, osb);
+	if (ret) {
+		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
+		ocfs2_recover_from_dlm_error(lockres, 1);
+	}
+
+	mlog(0, "lock %s, return from ocfs2_dlm_lock\n", lockres->l_name);
+
+bail:
+	return ret;
+}
+
+static inline int ocfs2_check_wait_flag(struct ocfs2_lock_res *lockres,
+					int flag)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	ret = lockres->l_flags & flag;
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	return ret;
+}
+
+static inline void ocfs2_wait_on_busy_lock(struct ocfs2_lock_res *lockres)
+
+{
+	wait_event(lockres->l_event,
+		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_BUSY));
+}
+
+static inline void ocfs2_wait_on_refreshing_lock(struct ocfs2_lock_res *lockres)
+
+{
+	wait_event(lockres->l_event,
+		   !ocfs2_check_wait_flag(lockres, OCFS2_LOCK_REFRESHING));
+}
+
+/* predict what lock level we'll be dropping down to on behalf
+ * of another node, and return true if the currently wanted
+ * level will be compatible with it. */
+static inline int ocfs2_may_continue_on_blocked_lock(struct ocfs2_lock_res *lockres,
+						     int wanted)
+{
+	BUG_ON(!(lockres->l_flags & OCFS2_LOCK_BLOCKED));
+
+	return wanted <= ocfs2_highest_compat_lock_level(lockres->l_blocking);
+}
+
+static void ocfs2_init_mask_waiter(struct ocfs2_mask_waiter *mw)
+{
+	INIT_LIST_HEAD(&mw->mw_item);
+	init_completion(&mw->mw_complete);
+	ocfs2_init_start_time(mw);
+}
+
+static int ocfs2_wait_for_mask(struct ocfs2_mask_waiter *mw)
+{
+	wait_for_completion(&mw->mw_complete);
+	/* Re-arm the completion in case we want to wait on it again */
+	reinit_completion(&mw->mw_complete);
+	return mw->mw_status;
+}
+
+static void lockres_add_mask_waiter(struct ocfs2_lock_res *lockres,
+				    struct ocfs2_mask_waiter *mw,
+				    unsigned long mask,
+				    unsigned long goal)
+{
+	BUG_ON(!list_empty(&mw->mw_item));
+
+	assert_spin_locked(&lockres->l_lock);
+
+	list_add_tail(&mw->mw_item, &lockres->l_mask_waiters);
+	mw->mw_mask = mask;
+	mw->mw_goal = goal;
+}
+
+/* returns 0 if the mw that was removed was already satisfied, -EBUSY
+ * if the mask still hadn't reached its goal */
+static int __lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
+				      struct ocfs2_mask_waiter *mw)
+{
+	int ret = 0;
+
+	assert_spin_locked(&lockres->l_lock);
+	if (!list_empty(&mw->mw_item)) {
+		if ((lockres->l_flags & mw->mw_mask) != mw->mw_goal)
+			ret = -EBUSY;
+
+		list_del_init(&mw->mw_item);
+		init_completion(&mw->mw_complete);
+	}
+
+	return ret;
+}
+
+static int lockres_remove_mask_waiter(struct ocfs2_lock_res *lockres,
+				      struct ocfs2_mask_waiter *mw)
+{
+	unsigned long flags;
+	int ret = 0;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	ret = __lockres_remove_mask_waiter(lockres, mw);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	return ret;
+
+}
+
+static int ocfs2_wait_for_mask_interruptible(struct ocfs2_mask_waiter *mw,
+					     struct ocfs2_lock_res *lockres)
+{
+	int ret;
+
+	ret = wait_for_completion_interruptible(&mw->mw_complete);
+	if (ret)
+		lockres_remove_mask_waiter(lockres, mw);
+	else
+		ret = mw->mw_status;
+	/* Re-arm the completion in case we want to wait on it again */
+	reinit_completion(&mw->mw_complete);
+	return ret;
+}
+
+static int __ocfs2_cluster_lock(struct ocfs2_super *osb,
+				struct ocfs2_lock_res *lockres,
+				int level,
+				u32 lkm_flags,
+				int arg_flags,
+				int l_subclass,
+				unsigned long caller_ip)
+{
+	struct ocfs2_mask_waiter mw;
+	int wait, catch_signals = !(osb->s_mount_opt & OCFS2_MOUNT_NOINTR);
+	int ret = 0; /* gcc doesn't realize wait = 1 guarantees ret is set */
+	unsigned long flags;
+	unsigned int gen;
+	int noqueue_attempted = 0;
+	int dlm_locked = 0;
+
+	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED)) {
+		mlog_errno(-EINVAL);
+		return -EINVAL;
+	}
+
+	ocfs2_init_mask_waiter(&mw);
+
+	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
+		lkm_flags |= DLM_LKF_VALBLK;
+
+again:
+	wait = 0;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+
+	if (catch_signals && signal_pending(current)) {
+		ret = -ERESTARTSYS;
+		goto unlock;
+	}
+
+	mlog_bug_on_msg(lockres->l_flags & OCFS2_LOCK_FREEING,
+			"Cluster lock called on freeing lockres %s! flags "
+			"0x%lx\n", lockres->l_name, lockres->l_flags);
+
+	/* We only compare against the currently granted level
+	 * here. If the lock is blocked waiting on a downconvert,
+	 * we'll get caught below. */
+	if (lockres->l_flags & OCFS2_LOCK_BUSY &&
+	    level > lockres->l_level) {
+		/* is someone sitting in dlm_lock? If so, wait on
+		 * them. */
+		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
+		wait = 1;
+		goto unlock;
+	}
+
+	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING) {
+		/*
+		 * We've upconverted. If the lock now has a level we can
+		 * work with, we take it. If, however, the lock is not at the
+		 * required level, we go thru the full cycle. One way this could
+		 * happen is if a process requesting an upconvert to PR is
+		 * closely followed by another requesting upconvert to an EX.
+		 * If the process requesting EX lands here, we want it to
+		 * continue attempting to upconvert and let the process
+		 * requesting PR take the lock.
+		 * If multiple processes request upconvert to PR, the first one
+		 * here will take the lock. The others will have to go thru the
+		 * OCFS2_LOCK_BLOCKED check to ensure that there is no pending
+		 * downconvert request.
+		 */
+		if (level <= lockres->l_level)
+			goto update_holders;
+	}
+
+	if (lockres->l_flags & OCFS2_LOCK_BLOCKED &&
+	    !ocfs2_may_continue_on_blocked_lock(lockres, level)) {
+		/* is the lock is currently blocked on behalf of
+		 * another node */
+		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BLOCKED, 0);
+		wait = 1;
+		goto unlock;
+	}
+
+	if (level > lockres->l_level) {
+		if (noqueue_attempted > 0) {
+			ret = -EAGAIN;
+			goto unlock;
+		}
+		if (lkm_flags & DLM_LKF_NOQUEUE)
+			noqueue_attempted = 1;
+
+		if (lockres->l_action != OCFS2_AST_INVALID)
+			mlog(ML_ERROR, "lockres %s has action %u pending\n",
+			     lockres->l_name, lockres->l_action);
+
+		if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
+			lockres->l_action = OCFS2_AST_ATTACH;
+			lkm_flags &= ~DLM_LKF_CONVERT;
+		} else {
+			lockres->l_action = OCFS2_AST_CONVERT;
+			lkm_flags |= DLM_LKF_CONVERT;
+		}
+
+		lockres->l_requested = level;
+		lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
+		gen = lockres_set_pending(lockres);
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+		BUG_ON(level == DLM_LOCK_IV);
+		BUG_ON(level == DLM_LOCK_NL);
+
+		mlog(ML_BASTS, "lockres %s, convert from %d to %d\n",
+		     lockres->l_name, lockres->l_level, level);
+
+		/* call dlm_lock to upgrade lock now */
+		ret = ocfs2_dlm_lock(osb->cconn,
+				     level,
+				     &lockres->l_lksb,
+				     lkm_flags,
+				     lockres->l_name,
+				     OCFS2_LOCK_ID_MAX_LEN - 1);
+		lockres_clear_pending(lockres, gen, osb);
+		if (ret) {
+			if (!(lkm_flags & DLM_LKF_NOQUEUE) ||
+			    (ret != -EAGAIN)) {
+				ocfs2_log_dlm_error("ocfs2_dlm_lock",
+						    ret, lockres);
+			}
+			ocfs2_recover_from_dlm_error(lockres, 1);
+			goto out;
+		}
+		dlm_locked = 1;
+
+		mlog(0, "lock %s, successful return from ocfs2_dlm_lock\n",
+		     lockres->l_name);
+
+		/* At this point we've gone inside the dlm and need to
+		 * complete our work regardless. */
+		catch_signals = 0;
+
+		/* wait for busy to clear and carry on */
+		goto again;
+	}
+
+update_holders:
+	/* Ok, if we get here then we're good to go. */
+	ocfs2_inc_holders(lockres, level);
+
+	ret = 0;
+unlock:
+	lockres_clear_flags(lockres, OCFS2_LOCK_UPCONVERT_FINISHING);
+
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+out:
+	/*
+	 * This is helping work around a lock inversion between the page lock
+	 * and dlm locks.  One path holds the page lock while calling aops
+	 * which block acquiring dlm locks.  The voting thread holds dlm
+	 * locks while acquiring page locks while down converting data locks.
+	 * This block is helping an aop path notice the inversion and back
+	 * off to unlock its page lock before trying the dlm lock again.
+	 */
+	if (wait && arg_flags & OCFS2_LOCK_NONBLOCK &&
+	    mw.mw_mask & (OCFS2_LOCK_BUSY|OCFS2_LOCK_BLOCKED)) {
+		wait = 0;
+		spin_lock_irqsave(&lockres->l_lock, flags);
+		if (__lockres_remove_mask_waiter(lockres, &mw)) {
+			if (dlm_locked)
+				lockres_or_flags(lockres,
+					OCFS2_LOCK_NONBLOCK_FINISHED);
+			spin_unlock_irqrestore(&lockres->l_lock, flags);
+			ret = -EAGAIN;
+		} else {
+			spin_unlock_irqrestore(&lockres->l_lock, flags);
+			goto again;
+		}
+	}
+	if (wait) {
+		ret = ocfs2_wait_for_mask(&mw);
+		if (ret == 0)
+			goto again;
+		mlog_errno(ret);
+	}
+	ocfs2_update_lock_stats(lockres, level, &mw, ret);
+
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+	if (!ret && lockres->l_lockdep_map.key != NULL) {
+		if (level == DLM_LOCK_PR)
+			rwsem_acquire_read(&lockres->l_lockdep_map, l_subclass,
+				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
+				caller_ip);
+		else
+			rwsem_acquire(&lockres->l_lockdep_map, l_subclass,
+				!!(arg_flags & OCFS2_META_LOCK_NOQUEUE),
+				caller_ip);
+	}
+#endif
+	return ret;
+}
+
+static inline int ocfs2_cluster_lock(struct ocfs2_super *osb,
+				     struct ocfs2_lock_res *lockres,
+				     int level,
+				     u32 lkm_flags,
+				     int arg_flags)
+{
+	return __ocfs2_cluster_lock(osb, lockres, level, lkm_flags, arg_flags,
+				    0, _RET_IP_);
+}
+
+
+static void __ocfs2_cluster_unlock(struct ocfs2_super *osb,
+				   struct ocfs2_lock_res *lockres,
+				   int level,
+				   unsigned long caller_ip)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	ocfs2_dec_holders(lockres, level);
+	ocfs2_downconvert_on_unlock(osb, lockres);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+	if (lockres->l_lockdep_map.key != NULL)
+		rwsem_release(&lockres->l_lockdep_map, 1, caller_ip);
+#endif
+}
+
+static int ocfs2_create_new_lock(struct ocfs2_super *osb,
+				 struct ocfs2_lock_res *lockres,
+				 int ex,
+				 int local)
+{
+	int level =  ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	unsigned long flags;
+	u32 lkm_flags = local ? DLM_LKF_LOCAL : 0;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	BUG_ON(lockres->l_flags & OCFS2_LOCK_ATTACHED);
+	lockres_or_flags(lockres, OCFS2_LOCK_LOCAL);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	return ocfs2_lock_create(osb, lockres, level, lkm_flags);
+}
+
+/* Grants us an EX lock on the data and metadata resources, skipping
+ * the normal cluster directory lookup. Use this ONLY on newly created
+ * inodes which other nodes can't possibly see, and which haven't been
+ * hashed in the inode hash yet. This can give us a good performance
+ * increase as it'll skip the network broadcast normally associated
+ * with creating a new lock resource. */
+int ocfs2_create_new_inode_locks(struct inode *inode)
+{
+	int ret;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	BUG_ON(!inode);
+	BUG_ON(!ocfs2_inode_is_new(inode));
+
+	mlog(0, "Inode %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	/* NOTE: That we don't increment any of the holder counts, nor
+	 * do we add anything to a journal handle. Since this is
+	 * supposed to be a new inode which the cluster doesn't know
+	 * about yet, there is no need to.  As far as the LVB handling
+	 * is concerned, this is basically like acquiring an EX lock
+	 * on a resource which has an invalid one -- we'll set it
+	 * valid when we release the EX. */
+
+	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_rw_lockres, 1, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	/*
+	 * We don't want to use DLM_LKF_LOCAL on a meta data lock as they
+	 * don't use a generation in their lock names.
+	 */
+	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_inode_lockres, 1, 0);
+	if (ret) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	ret = ocfs2_create_new_lock(osb, &OCFS2_I(inode)->ip_open_lockres, 0, 0);
+	if (ret) {
+		mlog_errno(ret);
+		goto bail;
+	}
+
+bail:
+	return ret;
+}
+
+int ocfs2_rw_lock(struct inode *inode, int write)
+{
+	int status, level;
+	struct ocfs2_lock_res *lockres;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	BUG_ON(!inode);
+
+	mlog(0, "inode %llu take %s RW lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+	     write ? "EXMODE" : "PRMODE");
+
+	if (ocfs2_mount_local(osb))
+		return 0;
+
+	lockres = &OCFS2_I(inode)->ip_rw_lockres;
+
+	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
+
+	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres, level, 0,
+				    0);
+	if (status < 0)
+		mlog_errno(status);
+
+	return status;
+}
+
+void ocfs2_rw_unlock(struct inode *inode, int write)
+{
+	int level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_rw_lockres;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	mlog(0, "inode %llu drop %s RW lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+	     write ? "EXMODE" : "PRMODE");
+
+	if (!ocfs2_mount_local(osb))
+		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
+}
+
+/*
+ * ocfs2_open_lock always get PR mode lock.
+ */
+int ocfs2_open_lock(struct inode *inode)
+{
+	int status = 0;
+	struct ocfs2_lock_res *lockres;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	BUG_ON(!inode);
+
+	mlog(0, "inode %llu take PRMODE open lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
+		goto out;
+
+	lockres = &OCFS2_I(inode)->ip_open_lockres;
+
+	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
+				    DLM_LOCK_PR, 0, 0);
+	if (status < 0)
+		mlog_errno(status);
+
+out:
+	return status;
+}
+
+int ocfs2_try_open_lock(struct inode *inode, int write)
+{
+	int status = 0, level;
+	struct ocfs2_lock_res *lockres;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	BUG_ON(!inode);
+
+	mlog(0, "inode %llu try to take %s open lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+	     write ? "EXMODE" : "PRMODE");
+
+	if (ocfs2_is_hard_readonly(osb)) {
+		if (write)
+			status = -EROFS;
+		goto out;
+	}
+
+	if (ocfs2_mount_local(osb))
+		goto out;
+
+	lockres = &OCFS2_I(inode)->ip_open_lockres;
+
+	level = write ? DLM_LOCK_EX : DLM_LOCK_PR;
+
+	/*
+	 * The file system may already holding a PRMODE/EXMODE open lock.
+	 * Since we pass DLM_LKF_NOQUEUE, the request won't block waiting on
+	 * other nodes and the -EAGAIN will indicate to the caller that
+	 * this inode is still in use.
+	 */
+	status = ocfs2_cluster_lock(OCFS2_SB(inode->i_sb), lockres,
+				    level, DLM_LKF_NOQUEUE, 0);
+
+out:
+	return status;
+}
+
+/*
+ * ocfs2_open_unlock unlock PR and EX mode open locks.
+ */
+void ocfs2_open_unlock(struct inode *inode)
+{
+	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_open_lockres;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	mlog(0, "inode %llu drop open lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	if (ocfs2_mount_local(osb))
+		goto out;
+
+	if(lockres->l_ro_holders)
+		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
+				     DLM_LOCK_PR);
+	if(lockres->l_ex_holders)
+		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres,
+				     DLM_LOCK_EX);
+
+out:
+	return;
+}
+
+static int ocfs2_flock_handle_signal(struct ocfs2_lock_res *lockres,
+				     int level)
+{
+	int ret;
+	struct ocfs2_super *osb = ocfs2_get_lockres_osb(lockres);
+	unsigned long flags;
+	struct ocfs2_mask_waiter mw;
+
+	ocfs2_init_mask_waiter(&mw);
+
+retry_cancel:
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
+		ret = ocfs2_prepare_cancel_convert(osb, lockres);
+		if (ret) {
+			spin_unlock_irqrestore(&lockres->l_lock, flags);
+			ret = ocfs2_cancel_convert(osb, lockres);
+			if (ret < 0) {
+				mlog_errno(ret);
+				goto out;
+			}
+			goto retry_cancel;
+		}
+		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+		ocfs2_wait_for_mask(&mw);
+		goto retry_cancel;
+	}
+
+	ret = -ERESTARTSYS;
+	/*
+	 * We may still have gotten the lock, in which case there's no
+	 * point to restarting the syscall.
+	 */
+	if (lockres->l_level == level)
+		ret = 0;
+
+	mlog(0, "Cancel returning %d. flags: 0x%lx, level: %d, act: %d\n", ret,
+	     lockres->l_flags, lockres->l_level, lockres->l_action);
+
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+out:
+	return ret;
+}
+
+/*
+ * ocfs2_file_lock() and ocfs2_file_unlock() map to a single pair of
+ * flock() calls. The locking approach this requires is sufficiently
+ * different from all other cluster lock types that we implement a
+ * separate path to the "low-level" dlm calls. In particular:
+ *
+ * - No optimization of lock levels is done - we take at exactly
+ *   what's been requested.
+ *
+ * - No lock caching is employed. We immediately downconvert to
+ *   no-lock at unlock time. This also means flock locks never go on
+ *   the blocking list).
+ *
+ * - Since userspace can trivially deadlock itself with flock, we make
+ *   sure to allow cancellation of a misbehaving applications flock()
+ *   request.
+ *
+ * - Access to any flock lockres doesn't require concurrency, so we
+ *   can simplify the code by requiring the caller to guarantee
+ *   serialization of dlmglue flock calls.
+ */
+int ocfs2_file_lock(struct file *file, int ex, int trylock)
+{
+	int ret, level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	unsigned int lkm_flags = trylock ? DLM_LKF_NOQUEUE : 0;
+	unsigned long flags;
+	struct ocfs2_file_private *fp = file->private_data;
+	struct ocfs2_lock_res *lockres = &fp->fp_flock;
+	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
+	struct ocfs2_mask_waiter mw;
+
+	ocfs2_init_mask_waiter(&mw);
+
+	if ((lockres->l_flags & OCFS2_LOCK_BUSY) ||
+	    (lockres->l_level > DLM_LOCK_NL)) {
+		mlog(ML_ERROR,
+		     "File lock \"%s\" has busy or locked state: flags: 0x%lx, "
+		     "level: %u\n", lockres->l_name, lockres->l_flags,
+		     lockres->l_level);
+		return -EINVAL;
+	}
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
+		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+		/*
+		 * Get the lock at NLMODE to start - that way we
+		 * can cancel the upconvert request if need be.
+		 */
+		ret = ocfs2_lock_create(osb, lockres, DLM_LOCK_NL, 0);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_wait_for_mask(&mw);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		spin_lock_irqsave(&lockres->l_lock, flags);
+	}
+
+	lockres->l_action = OCFS2_AST_CONVERT;
+	lkm_flags |= DLM_LKF_CONVERT;
+	lockres->l_requested = level;
+	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
+
+	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	ret = ocfs2_dlm_lock(osb->cconn, level, &lockres->l_lksb, lkm_flags,
+			     lockres->l_name, OCFS2_LOCK_ID_MAX_LEN - 1);
+	if (ret) {
+		if (!trylock || (ret != -EAGAIN)) {
+			ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
+			ret = -EINVAL;
+		}
+
+		ocfs2_recover_from_dlm_error(lockres, 1);
+		lockres_remove_mask_waiter(lockres, &mw);
+		goto out;
+	}
+
+	ret = ocfs2_wait_for_mask_interruptible(&mw, lockres);
+	if (ret == -ERESTARTSYS) {
+		/*
+		 * Userspace can cause deadlock itself with
+		 * flock(). Current behavior locally is to allow the
+		 * deadlock, but abort the system call if a signal is
+		 * received. We follow this example, otherwise a
+		 * poorly written program could sit in kernel until
+		 * reboot.
+		 *
+		 * Handling this is a bit more complicated for Ocfs2
+		 * though. We can't exit this function with an
+		 * outstanding lock request, so a cancel convert is
+		 * required. We intentionally overwrite 'ret' - if the
+		 * cancel fails and the lock was granted, it's easier
+		 * to just bubble success back up to the user.
+		 */
+		ret = ocfs2_flock_handle_signal(lockres, level);
+	} else if (!ret && (level > lockres->l_level)) {
+		/* Trylock failed asynchronously */
+		BUG_ON(!trylock);
+		ret = -EAGAIN;
+	}
+
+out:
+
+	mlog(0, "Lock: \"%s\" ex: %d, trylock: %d, returns: %d\n",
+	     lockres->l_name, ex, trylock, ret);
+	return ret;
+}
+
+void ocfs2_file_unlock(struct file *file)
+{
+	int ret;
+	unsigned int gen;
+	unsigned long flags;
+	struct ocfs2_file_private *fp = file->private_data;
+	struct ocfs2_lock_res *lockres = &fp->fp_flock;
+	struct ocfs2_super *osb = OCFS2_SB(file->f_mapping->host->i_sb);
+	struct ocfs2_mask_waiter mw;
+
+	ocfs2_init_mask_waiter(&mw);
+
+	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED))
+		return;
+
+	if (lockres->l_level == DLM_LOCK_NL)
+		return;
+
+	mlog(0, "Unlock: \"%s\" flags: 0x%lx, level: %d, act: %d\n",
+	     lockres->l_name, lockres->l_flags, lockres->l_level,
+	     lockres->l_action);
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	/*
+	 * Fake a blocking ast for the downconvert code.
+	 */
+	lockres_or_flags(lockres, OCFS2_LOCK_BLOCKED);
+	lockres->l_blocking = DLM_LOCK_EX;
+
+	gen = ocfs2_prepare_downconvert(lockres, DLM_LOCK_NL);
+	lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_BUSY, 0);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	ret = ocfs2_downconvert_lock(osb, lockres, DLM_LOCK_NL, 0, gen);
+	if (ret) {
+		mlog_errno(ret);
+		return;
+	}
+
+	ret = ocfs2_wait_for_mask(&mw);
+	if (ret)
+		mlog_errno(ret);
+}
+
+static void ocfs2_downconvert_on_unlock(struct ocfs2_super *osb,
+					struct ocfs2_lock_res *lockres)
+{
+	int kick = 0;
+
+	/* If we know that another node is waiting on our lock, kick
+	 * the downconvert thread * pre-emptively when we reach a release
+	 * condition. */
+	if (lockres->l_flags & OCFS2_LOCK_BLOCKED) {
+		switch(lockres->l_blocking) {
+		case DLM_LOCK_EX:
+			if (!lockres->l_ex_holders && !lockres->l_ro_holders)
+				kick = 1;
+			break;
+		case DLM_LOCK_PR:
+			if (!lockres->l_ex_holders)
+				kick = 1;
+			break;
+		default:
+			BUG();
+		}
+	}
+
+	if (kick)
+		ocfs2_wake_downconvert_thread(osb);
+}
+
+#define OCFS2_SEC_BITS   34
+#define OCFS2_SEC_SHIFT  (64 - 34)
+#define OCFS2_NSEC_MASK  ((1ULL << OCFS2_SEC_SHIFT) - 1)
+
+/* LVB only has room for 64 bits of time here so we pack it for
+ * now. */
+static u64 ocfs2_pack_timespec(struct timespec *spec)
+{
+	u64 res;
+	u64 sec = spec->tv_sec;
+	u32 nsec = spec->tv_nsec;
+
+	res = (sec << OCFS2_SEC_SHIFT) | (nsec & OCFS2_NSEC_MASK);
+
+	return res;
+}
+
+/* Call this with the lockres locked. I am reasonably sure we don't
+ * need ip_lock in this function as anyone who would be changing those
+ * values is supposed to be blocked in ocfs2_inode_lock right now. */
+static void __ocfs2_stuff_meta_lvb(struct inode *inode)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
+	struct ocfs2_meta_lvb *lvb;
+
+	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+
+	/*
+	 * Invalidate the LVB of a deleted inode - this way other
+	 * nodes are forced to go to disk and discover the new inode
+	 * status.
+	 */
+	if (oi->ip_flags & OCFS2_INODE_DELETED) {
+		lvb->lvb_version = 0;
+		goto out;
+	}
+
+	lvb->lvb_version   = OCFS2_LVB_VERSION;
+	lvb->lvb_isize	   = cpu_to_be64(i_size_read(inode));
+	lvb->lvb_iclusters = cpu_to_be32(oi->ip_clusters);
+	lvb->lvb_iuid      = cpu_to_be32(i_uid_read(inode));
+	lvb->lvb_igid      = cpu_to_be32(i_gid_read(inode));
+	lvb->lvb_imode     = cpu_to_be16(inode->i_mode);
+	lvb->lvb_inlink    = cpu_to_be16(inode->i_nlink);
+	lvb->lvb_iatime_packed  =
+		cpu_to_be64(ocfs2_pack_timespec(&inode->i_atime));
+	lvb->lvb_ictime_packed =
+		cpu_to_be64(ocfs2_pack_timespec(&inode->i_ctime));
+	lvb->lvb_imtime_packed =
+		cpu_to_be64(ocfs2_pack_timespec(&inode->i_mtime));
+	lvb->lvb_iattr    = cpu_to_be32(oi->ip_attr);
+	lvb->lvb_idynfeatures = cpu_to_be16(oi->ip_dyn_features);
+	lvb->lvb_igeneration = cpu_to_be32(inode->i_generation);
+
+out:
+	mlog_meta_lvb(0, lockres);
+}
+
+static void ocfs2_unpack_timespec(struct timespec *spec,
+				  u64 packed_time)
+{
+	spec->tv_sec = packed_time >> OCFS2_SEC_SHIFT;
+	spec->tv_nsec = packed_time & OCFS2_NSEC_MASK;
+}
+
+static void ocfs2_refresh_inode_from_lvb(struct inode *inode)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
+	struct ocfs2_meta_lvb *lvb;
+
+	mlog_meta_lvb(0, lockres);
+
+	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+
+	/* We're safe here without the lockres lock... */
+	spin_lock(&oi->ip_lock);
+	oi->ip_clusters = be32_to_cpu(lvb->lvb_iclusters);
+	i_size_write(inode, be64_to_cpu(lvb->lvb_isize));
+
+	oi->ip_attr = be32_to_cpu(lvb->lvb_iattr);
+	oi->ip_dyn_features = be16_to_cpu(lvb->lvb_idynfeatures);
+	ocfs2_set_inode_flags(inode);
+
+	/* fast-symlinks are a special case */
+	if (S_ISLNK(inode->i_mode) && !oi->ip_clusters)
+		inode->i_blocks = 0;
+	else
+		inode->i_blocks = ocfs2_inode_sector_count(inode);
+
+	i_uid_write(inode, be32_to_cpu(lvb->lvb_iuid));
+	i_gid_write(inode, be32_to_cpu(lvb->lvb_igid));
+	inode->i_mode    = be16_to_cpu(lvb->lvb_imode);
+	set_nlink(inode, be16_to_cpu(lvb->lvb_inlink));
+	ocfs2_unpack_timespec(&inode->i_atime,
+			      be64_to_cpu(lvb->lvb_iatime_packed));
+	ocfs2_unpack_timespec(&inode->i_mtime,
+			      be64_to_cpu(lvb->lvb_imtime_packed));
+	ocfs2_unpack_timespec(&inode->i_ctime,
+			      be64_to_cpu(lvb->lvb_ictime_packed));
+	spin_unlock(&oi->ip_lock);
+}
+
+static inline int ocfs2_meta_lvb_is_trustable(struct inode *inode,
+					      struct ocfs2_lock_res *lockres)
+{
+	struct ocfs2_meta_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+
+	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)
+	    && lvb->lvb_version == OCFS2_LVB_VERSION
+	    && be32_to_cpu(lvb->lvb_igeneration) == inode->i_generation)
+		return 1;
+	return 0;
+}
+
+/* Determine whether a lock resource needs to be refreshed, and
+ * arbitrate who gets to refresh it.
+ *
+ *   0 means no refresh needed.
+ *
+ *   > 0 means you need to refresh this and you MUST call
+ *   ocfs2_complete_lock_res_refresh afterwards. */
+static int ocfs2_should_refresh_lock_res(struct ocfs2_lock_res *lockres)
+{
+	unsigned long flags;
+	int status = 0;
+
+refresh_check:
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	if (!(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH)) {
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		goto bail;
+	}
+
+	if (lockres->l_flags & OCFS2_LOCK_REFRESHING) {
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+		ocfs2_wait_on_refreshing_lock(lockres);
+		goto refresh_check;
+	}
+
+	/* Ok, I'll be the one to refresh this lock. */
+	lockres_or_flags(lockres, OCFS2_LOCK_REFRESHING);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	status = 1;
+bail:
+	mlog(0, "status %d\n", status);
+	return status;
+}
+
+/* If status is non zero, I'll mark it as not being in refresh
+ * anymroe, but i won't clear the needs refresh flag. */
+static inline void ocfs2_complete_lock_res_refresh(struct ocfs2_lock_res *lockres,
+						   int status)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	lockres_clear_flags(lockres, OCFS2_LOCK_REFRESHING);
+	if (!status)
+		lockres_clear_flags(lockres, OCFS2_LOCK_NEEDS_REFRESH);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	wake_up(&lockres->l_event);
+}
+
+/* may or may not return a bh if it went to disk. */
+static int ocfs2_inode_lock_update(struct inode *inode,
+				  struct buffer_head **bh)
+{
+	int status = 0;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_lock_res *lockres = &oi->ip_inode_lockres;
+	struct ocfs2_dinode *fe;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (ocfs2_mount_local(osb))
+		goto bail;
+
+	spin_lock(&oi->ip_lock);
+	if (oi->ip_flags & OCFS2_INODE_DELETED) {
+		mlog(0, "Orphaned inode %llu was deleted while we "
+		     "were waiting on a lock. ip_flags = 0x%x\n",
+		     (unsigned long long)oi->ip_blkno, oi->ip_flags);
+		spin_unlock(&oi->ip_lock);
+		status = -ENOENT;
+		goto bail;
+	}
+	spin_unlock(&oi->ip_lock);
+
+	if (!ocfs2_should_refresh_lock_res(lockres))
+		goto bail;
+
+	/* This will discard any caching information we might have had
+	 * for the inode metadata. */
+	ocfs2_metadata_cache_purge(INODE_CACHE(inode));
+
+	ocfs2_extent_map_trunc(inode, 0);
+
+	if (ocfs2_meta_lvb_is_trustable(inode, lockres)) {
+		mlog(0, "Trusting LVB on inode %llu\n",
+		     (unsigned long long)oi->ip_blkno);
+		ocfs2_refresh_inode_from_lvb(inode);
+	} else {
+		/* Boo, we have to go to disk. */
+		/* read bh, cast, ocfs2_refresh_inode */
+		status = ocfs2_read_inode_block(inode, bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail_refresh;
+		}
+		fe = (struct ocfs2_dinode *) (*bh)->b_data;
+
+		/* This is a good chance to make sure we're not
+		 * locking an invalid object.  ocfs2_read_inode_block()
+		 * already checked that the inode block is sane.
+		 *
+		 * We bug on a stale inode here because we checked
+		 * above whether it was wiped from disk. The wiping
+		 * node provides a guarantee that we receive that
+		 * message and can mark the inode before dropping any
+		 * locks associated with it. */
+		mlog_bug_on_msg(inode->i_generation !=
+				le32_to_cpu(fe->i_generation),
+				"Invalid dinode %llu disk generation: %u "
+				"inode->i_generation: %u\n",
+				(unsigned long long)oi->ip_blkno,
+				le32_to_cpu(fe->i_generation),
+				inode->i_generation);
+		mlog_bug_on_msg(le64_to_cpu(fe->i_dtime) ||
+				!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)),
+				"Stale dinode %llu dtime: %llu flags: 0x%x\n",
+				(unsigned long long)oi->ip_blkno,
+				(unsigned long long)le64_to_cpu(fe->i_dtime),
+				le32_to_cpu(fe->i_flags));
+
+		ocfs2_refresh_inode(inode, fe);
+		ocfs2_track_lock_refresh(lockres);
+	}
+
+	status = 0;
+bail_refresh:
+	ocfs2_complete_lock_res_refresh(lockres, status);
+bail:
+	return status;
+}
+
+static int ocfs2_assign_bh(struct inode *inode,
+			   struct buffer_head **ret_bh,
+			   struct buffer_head *passed_bh)
+{
+	int status;
+
+	if (passed_bh) {
+		/* Ok, the update went to disk for us, use the
+		 * returned bh. */
+		*ret_bh = passed_bh;
+		get_bh(*ret_bh);
+
+		return 0;
+	}
+
+	status = ocfs2_read_inode_block(inode, ret_bh);
+	if (status < 0)
+		mlog_errno(status);
+
+	return status;
+}
+
+/*
+ * returns < 0 error if the callback will never be called, otherwise
+ * the result of the lock will be communicated via the callback.
+ */
+int ocfs2_inode_lock_full_nested(struct inode *inode,
+				 struct buffer_head **ret_bh,
+				 int ex,
+				 int arg_flags,
+				 int subclass)
+{
+	int status, level, acquired;
+	u32 dlm_flags;
+	struct ocfs2_lock_res *lockres = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *local_bh = NULL;
+
+	BUG_ON(!inode);
+
+	mlog(0, "inode %llu, take %s META lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+	     ex ? "EXMODE" : "PRMODE");
+
+	status = 0;
+	acquired = 0;
+	/* We'll allow faking a readonly metadata lock for
+	 * rodevices. */
+	if (ocfs2_is_hard_readonly(osb)) {
+		if (ex)
+			status = -EROFS;
+		goto getbh;
+	}
+
+	if (ocfs2_mount_local(osb))
+		goto local;
+
+	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
+		ocfs2_wait_for_recovery(osb);
+
+	lockres = &OCFS2_I(inode)->ip_inode_lockres;
+	level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	dlm_flags = 0;
+	if (arg_flags & OCFS2_META_LOCK_NOQUEUE)
+		dlm_flags |= DLM_LKF_NOQUEUE;
+
+	status = __ocfs2_cluster_lock(osb, lockres, level, dlm_flags,
+				      arg_flags, subclass, _RET_IP_);
+	if (status < 0) {
+		if (status != -EAGAIN)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	/* Notify the error cleanup path to drop the cluster lock. */
+	acquired = 1;
+
+	/* We wait twice because a node may have died while we were in
+	 * the lower dlm layers. The second time though, we've
+	 * committed to owning this lock so we don't allow signals to
+	 * abort the operation. */
+	if (!(arg_flags & OCFS2_META_LOCK_RECOVERY))
+		ocfs2_wait_for_recovery(osb);
+
+local:
+	/*
+	 * We only see this flag if we're being called from
+	 * ocfs2_read_locked_inode(). It means we're locking an inode
+	 * which hasn't been populated yet, so clear the refresh flag
+	 * and let the caller handle it.
+	 */
+	if (inode->i_state & I_NEW) {
+		status = 0;
+		if (lockres)
+			ocfs2_complete_lock_res_refresh(lockres, 0);
+		goto bail;
+	}
+
+	/* This is fun. The caller may want a bh back, or it may
+	 * not. ocfs2_inode_lock_update definitely wants one in, but
+	 * may or may not read one, depending on what's in the
+	 * LVB. The result of all of this is that we've *only* gone to
+	 * disk if we have to, so the complexity is worthwhile. */
+	status = ocfs2_inode_lock_update(inode, &local_bh);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		goto bail;
+	}
+getbh:
+	if (ret_bh) {
+		status = ocfs2_assign_bh(inode, ret_bh, local_bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+bail:
+	if (status < 0) {
+		if (ret_bh && (*ret_bh)) {
+			brelse(*ret_bh);
+			*ret_bh = NULL;
+		}
+		if (acquired)
+			ocfs2_inode_unlock(inode, ex);
+	}
+
+	if (local_bh)
+		brelse(local_bh);
+
+	return status;
+}
+
+/*
+ * This is working around a lock inversion between tasks acquiring DLM
+ * locks while holding a page lock and the downconvert thread which
+ * blocks dlm lock acquiry while acquiring page locks.
+ *
+ * ** These _with_page variantes are only intended to be called from aop
+ * methods that hold page locks and return a very specific *positive* error
+ * code that aop methods pass up to the VFS -- test for errors with != 0. **
+ *
+ * The DLM is called such that it returns -EAGAIN if it would have
+ * blocked waiting for the downconvert thread.  In that case we unlock
+ * our page so the downconvert thread can make progress.  Once we've
+ * done this we have to return AOP_TRUNCATED_PAGE so the aop method
+ * that called us can bubble that back up into the VFS who will then
+ * immediately retry the aop call.
+ *
+ * We do a blocking lock and immediate unlock before returning, though, so that
+ * the lock has a great chance of being cached on this node by the time the VFS
+ * calls back to retry the aop.    This has a potential to livelock as nodes
+ * ping locks back and forth, but that's a risk we're willing to take to avoid
+ * the lock inversion simply.
+ */
+int ocfs2_inode_lock_with_page(struct inode *inode,
+			      struct buffer_head **ret_bh,
+			      int ex,
+			      struct page *page)
+{
+	int ret;
+
+	ret = ocfs2_inode_lock_full(inode, ret_bh, ex, OCFS2_LOCK_NONBLOCK);
+	if (ret == -EAGAIN) {
+		unlock_page(page);
+		if (ocfs2_inode_lock(inode, ret_bh, ex) == 0)
+			ocfs2_inode_unlock(inode, ex);
+		ret = AOP_TRUNCATED_PAGE;
+	}
+
+	return ret;
+}
+
+int ocfs2_inode_lock_atime(struct inode *inode,
+			  struct vfsmount *vfsmnt,
+			  int *level)
+{
+	int ret;
+
+	ret = ocfs2_inode_lock(inode, NULL, 0);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	/*
+	 * If we should update atime, we will get EX lock,
+	 * otherwise we just get PR lock.
+	 */
+	if (ocfs2_should_update_atime(inode, vfsmnt)) {
+		struct buffer_head *bh = NULL;
+
+		ocfs2_inode_unlock(inode, 0);
+		ret = ocfs2_inode_lock(inode, &bh, 1);
+		if (ret < 0) {
+			mlog_errno(ret);
+			return ret;
+		}
+		*level = 1;
+		if (ocfs2_should_update_atime(inode, vfsmnt))
+			ocfs2_update_inode_atime(inode, bh);
+		if (bh)
+			brelse(bh);
+	} else
+		*level = 0;
+
+	return ret;
+}
+
+void ocfs2_inode_unlock(struct inode *inode,
+		       int ex)
+{
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_lock_res *lockres = &OCFS2_I(inode)->ip_inode_lockres;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	mlog(0, "inode %llu drop %s META lock\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+	     ex ? "EXMODE" : "PRMODE");
+
+	if (!ocfs2_is_hard_readonly(OCFS2_SB(inode->i_sb)) &&
+	    !ocfs2_mount_local(osb))
+		ocfs2_cluster_unlock(OCFS2_SB(inode->i_sb), lockres, level);
+}
+
+int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno)
+{
+	struct ocfs2_lock_res *lockres;
+	struct ocfs2_orphan_scan_lvb *lvb;
+	int status = 0;
+
+	if (ocfs2_is_hard_readonly(osb))
+		return -EROFS;
+
+	if (ocfs2_mount_local(osb))
+		return 0;
+
+	lockres = &osb->osb_orphan_scan.os_lockres;
+	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
+	if (status < 0)
+		return status;
+
+	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
+	    lvb->lvb_version == OCFS2_ORPHAN_LVB_VERSION)
+		*seqno = be32_to_cpu(lvb->lvb_os_seqno);
+	else
+		*seqno = osb->osb_orphan_scan.os_seqno + 1;
+
+	return status;
+}
+
+void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno)
+{
+	struct ocfs2_lock_res *lockres;
+	struct ocfs2_orphan_scan_lvb *lvb;
+
+	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb)) {
+		lockres = &osb->osb_orphan_scan.os_lockres;
+		lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+		lvb->lvb_version = OCFS2_ORPHAN_LVB_VERSION;
+		lvb->lvb_os_seqno = cpu_to_be32(seqno);
+		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
+	}
+}
+
+int ocfs2_super_lock(struct ocfs2_super *osb,
+		     int ex)
+{
+	int status = 0;
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
+
+	if (ocfs2_is_hard_readonly(osb))
+		return -EROFS;
+
+	if (ocfs2_mount_local(osb))
+		goto bail;
+
+	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/* The super block lock path is really in the best position to
+	 * know when resources covered by the lock need to be
+	 * refreshed, so we do it here. Of course, making sense of
+	 * everything is up to the caller :) */
+	status = ocfs2_should_refresh_lock_res(lockres);
+	if (status) {
+		status = ocfs2_refresh_slot_info(osb);
+
+		ocfs2_complete_lock_res_refresh(lockres, status);
+
+		if (status < 0) {
+			ocfs2_cluster_unlock(osb, lockres, level);
+			mlog_errno(status);
+		}
+		ocfs2_track_lock_refresh(lockres);
+	}
+bail:
+	return status;
+}
+
+void ocfs2_super_unlock(struct ocfs2_super *osb,
+			int ex)
+{
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_lock_res *lockres = &osb->osb_super_lockres;
+
+	if (!ocfs2_mount_local(osb))
+		ocfs2_cluster_unlock(osb, lockres, level);
+}
+
+int ocfs2_rename_lock(struct ocfs2_super *osb)
+{
+	int status;
+	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
+
+	if (ocfs2_is_hard_readonly(osb))
+		return -EROFS;
+
+	if (ocfs2_mount_local(osb))
+		return 0;
+
+	status = ocfs2_cluster_lock(osb, lockres, DLM_LOCK_EX, 0, 0);
+	if (status < 0)
+		mlog_errno(status);
+
+	return status;
+}
+
+void ocfs2_rename_unlock(struct ocfs2_super *osb)
+{
+	struct ocfs2_lock_res *lockres = &osb->osb_rename_lockres;
+
+	if (!ocfs2_mount_local(osb))
+		ocfs2_cluster_unlock(osb, lockres, DLM_LOCK_EX);
+}
+
+int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex)
+{
+	int status;
+	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
+
+	if (ocfs2_is_hard_readonly(osb))
+		return -EROFS;
+
+	if (ocfs2_mount_local(osb))
+		return 0;
+
+	status = ocfs2_cluster_lock(osb, lockres, ex ? LKM_EXMODE : LKM_PRMODE,
+				    0, 0);
+	if (status < 0)
+		mlog(ML_ERROR, "lock on nfs sync lock failed %d\n", status);
+
+	return status;
+}
+
+void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex)
+{
+	struct ocfs2_lock_res *lockres = &osb->osb_nfs_sync_lockres;
+
+	if (!ocfs2_mount_local(osb))
+		ocfs2_cluster_unlock(osb, lockres,
+				     ex ? LKM_EXMODE : LKM_PRMODE);
+}
+
+int ocfs2_dentry_lock(struct dentry *dentry, int ex)
+{
+	int ret;
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
+	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
+
+	BUG_ON(!dl);
+
+	if (ocfs2_is_hard_readonly(osb)) {
+		if (ex)
+			return -EROFS;
+		return 0;
+	}
+
+	if (ocfs2_mount_local(osb))
+		return 0;
+
+	ret = ocfs2_cluster_lock(osb, &dl->dl_lockres, level, 0, 0);
+	if (ret < 0)
+		mlog_errno(ret);
+
+	return ret;
+}
+
+void ocfs2_dentry_unlock(struct dentry *dentry, int ex)
+{
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
+	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
+
+	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
+		ocfs2_cluster_unlock(osb, &dl->dl_lockres, level);
+}
+
+/* Reference counting of the dlm debug structure. We want this because
+ * open references on the debug inodes can live on after a mount, so
+ * we can't rely on the ocfs2_super to always exist. */
+static void ocfs2_dlm_debug_free(struct kref *kref)
+{
+	struct ocfs2_dlm_debug *dlm_debug;
+
+	dlm_debug = container_of(kref, struct ocfs2_dlm_debug, d_refcnt);
+
+	kfree(dlm_debug);
+}
+
+void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug)
+{
+	if (dlm_debug)
+		kref_put(&dlm_debug->d_refcnt, ocfs2_dlm_debug_free);
+}
+
+static void ocfs2_get_dlm_debug(struct ocfs2_dlm_debug *debug)
+{
+	kref_get(&debug->d_refcnt);
+}
+
+struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void)
+{
+	struct ocfs2_dlm_debug *dlm_debug;
+
+	dlm_debug = kmalloc(sizeof(struct ocfs2_dlm_debug), GFP_KERNEL);
+	if (!dlm_debug) {
+		mlog_errno(-ENOMEM);
+		goto out;
+	}
+
+	kref_init(&dlm_debug->d_refcnt);
+	INIT_LIST_HEAD(&dlm_debug->d_lockres_tracking);
+	dlm_debug->d_locking_state = NULL;
+out:
+	return dlm_debug;
+}
+
+/* Access to this is arbitrated for us via seq_file->sem. */
+struct ocfs2_dlm_seq_priv {
+	struct ocfs2_dlm_debug *p_dlm_debug;
+	struct ocfs2_lock_res p_iter_res;
+	struct ocfs2_lock_res p_tmp_res;
+};
+
+static struct ocfs2_lock_res *ocfs2_dlm_next_res(struct ocfs2_lock_res *start,
+						 struct ocfs2_dlm_seq_priv *priv)
+{
+	struct ocfs2_lock_res *iter, *ret = NULL;
+	struct ocfs2_dlm_debug *dlm_debug = priv->p_dlm_debug;
+
+	assert_spin_locked(&ocfs2_dlm_tracking_lock);
+
+	list_for_each_entry(iter, &start->l_debug_list, l_debug_list) {
+		/* discover the head of the list */
+		if (&iter->l_debug_list == &dlm_debug->d_lockres_tracking) {
+			mlog(0, "End of list found, %p\n", ret);
+			break;
+		}
+
+		/* We track our "dummy" iteration lockres' by a NULL
+		 * l_ops field. */
+		if (iter->l_ops != NULL) {
+			ret = iter;
+			break;
+		}
+	}
+
+	return ret;
+}
+
+static void *ocfs2_dlm_seq_start(struct seq_file *m, loff_t *pos)
+{
+	struct ocfs2_dlm_seq_priv *priv = m->private;
+	struct ocfs2_lock_res *iter;
+
+	spin_lock(&ocfs2_dlm_tracking_lock);
+	iter = ocfs2_dlm_next_res(&priv->p_iter_res, priv);
+	if (iter) {
+		/* Since lockres' have the lifetime of their container
+		 * (which can be inodes, ocfs2_supers, etc) we want to
+		 * copy this out to a temporary lockres while still
+		 * under the spinlock. Obviously after this we can't
+		 * trust any pointers on the copy returned, but that's
+		 * ok as the information we want isn't typically held
+		 * in them. */
+		priv->p_tmp_res = *iter;
+		iter = &priv->p_tmp_res;
+	}
+	spin_unlock(&ocfs2_dlm_tracking_lock);
+
+	return iter;
+}
+
+static void ocfs2_dlm_seq_stop(struct seq_file *m, void *v)
+{
+}
+
+static void *ocfs2_dlm_seq_next(struct seq_file *m, void *v, loff_t *pos)
+{
+	struct ocfs2_dlm_seq_priv *priv = m->private;
+	struct ocfs2_lock_res *iter = v;
+	struct ocfs2_lock_res *dummy = &priv->p_iter_res;
+
+	spin_lock(&ocfs2_dlm_tracking_lock);
+	iter = ocfs2_dlm_next_res(iter, priv);
+	list_del_init(&dummy->l_debug_list);
+	if (iter) {
+		list_add(&dummy->l_debug_list, &iter->l_debug_list);
+		priv->p_tmp_res = *iter;
+		iter = &priv->p_tmp_res;
+	}
+	spin_unlock(&ocfs2_dlm_tracking_lock);
+
+	return iter;
+}
+
+/*
+ * Version is used by debugfs.ocfs2 to determine the format being used
+ *
+ * New in version 2
+ *	- Lock stats printed
+ * New in version 3
+ *	- Max time in lock stats is in usecs (instead of nsecs)
+ */
+#define OCFS2_DLM_DEBUG_STR_VERSION 3
+static int ocfs2_dlm_seq_show(struct seq_file *m, void *v)
+{
+	int i;
+	char *lvb;
+	struct ocfs2_lock_res *lockres = v;
+
+	if (!lockres)
+		return -EINVAL;
+
+	seq_printf(m, "0x%x\t", OCFS2_DLM_DEBUG_STR_VERSION);
+
+	if (lockres->l_type == OCFS2_LOCK_TYPE_DENTRY)
+		seq_printf(m, "%.*s%08x\t", OCFS2_DENTRY_LOCK_INO_START - 1,
+			   lockres->l_name,
+			   (unsigned int)ocfs2_get_dentry_lock_ino(lockres));
+	else
+		seq_printf(m, "%.*s\t", OCFS2_LOCK_ID_MAX_LEN, lockres->l_name);
+
+	seq_printf(m, "%d\t"
+		   "0x%lx\t"
+		   "0x%x\t"
+		   "0x%x\t"
+		   "%u\t"
+		   "%u\t"
+		   "%d\t"
+		   "%d\t",
+		   lockres->l_level,
+		   lockres->l_flags,
+		   lockres->l_action,
+		   lockres->l_unlock_action,
+		   lockres->l_ro_holders,
+		   lockres->l_ex_holders,
+		   lockres->l_requested,
+		   lockres->l_blocking);
+
+	/* Dump the raw LVB */
+	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+	for(i = 0; i < DLM_LVB_LEN; i++)
+		seq_printf(m, "0x%x\t", lvb[i]);
+
+#ifdef CONFIG_OCFS2_FS_STATS
+# define lock_num_prmode(_l)		((_l)->l_lock_prmode.ls_gets)
+# define lock_num_exmode(_l)		((_l)->l_lock_exmode.ls_gets)
+# define lock_num_prmode_failed(_l)	((_l)->l_lock_prmode.ls_fail)
+# define lock_num_exmode_failed(_l)	((_l)->l_lock_exmode.ls_fail)
+# define lock_total_prmode(_l)		((_l)->l_lock_prmode.ls_total)
+# define lock_total_exmode(_l)		((_l)->l_lock_exmode.ls_total)
+# define lock_max_prmode(_l)		((_l)->l_lock_prmode.ls_max)
+# define lock_max_exmode(_l)		((_l)->l_lock_exmode.ls_max)
+# define lock_refresh(_l)		((_l)->l_lock_refresh)
+#else
+# define lock_num_prmode(_l)		(0)
+# define lock_num_exmode(_l)		(0)
+# define lock_num_prmode_failed(_l)	(0)
+# define lock_num_exmode_failed(_l)	(0)
+# define lock_total_prmode(_l)		(0ULL)
+# define lock_total_exmode(_l)		(0ULL)
+# define lock_max_prmode(_l)		(0)
+# define lock_max_exmode(_l)		(0)
+# define lock_refresh(_l)		(0)
+#endif
+	/* The following seq_print was added in version 2 of this output */
+	seq_printf(m, "%u\t"
+		   "%u\t"
+		   "%u\t"
+		   "%u\t"
+		   "%llu\t"
+		   "%llu\t"
+		   "%u\t"
+		   "%u\t"
+		   "%u\t",
+		   lock_num_prmode(lockres),
+		   lock_num_exmode(lockres),
+		   lock_num_prmode_failed(lockres),
+		   lock_num_exmode_failed(lockres),
+		   lock_total_prmode(lockres),
+		   lock_total_exmode(lockres),
+		   lock_max_prmode(lockres),
+		   lock_max_exmode(lockres),
+		   lock_refresh(lockres));
+
+	/* End the line */
+	seq_printf(m, "\n");
+	return 0;
+}
+
+static const struct seq_operations ocfs2_dlm_seq_ops = {
+	.start =	ocfs2_dlm_seq_start,
+	.stop =		ocfs2_dlm_seq_stop,
+	.next =		ocfs2_dlm_seq_next,
+	.show =		ocfs2_dlm_seq_show,
+};
+
+static int ocfs2_dlm_debug_release(struct inode *inode, struct file *file)
+{
+	struct seq_file *seq = file->private_data;
+	struct ocfs2_dlm_seq_priv *priv = seq->private;
+	struct ocfs2_lock_res *res = &priv->p_iter_res;
+
+	ocfs2_remove_lockres_tracking(res);
+	ocfs2_put_dlm_debug(priv->p_dlm_debug);
+	return seq_release_private(inode, file);
+}
+
+static int ocfs2_dlm_debug_open(struct inode *inode, struct file *file)
+{
+	struct ocfs2_dlm_seq_priv *priv;
+	struct ocfs2_super *osb;
+
+	priv = __seq_open_private(file, &ocfs2_dlm_seq_ops, sizeof(*priv));
+	if (!priv) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
+	osb = inode->i_private;
+	ocfs2_get_dlm_debug(osb->osb_dlm_debug);
+	priv->p_dlm_debug = osb->osb_dlm_debug;
+	INIT_LIST_HEAD(&priv->p_iter_res.l_debug_list);
+
+	ocfs2_add_lockres_tracking(&priv->p_iter_res,
+				   priv->p_dlm_debug);
+
+	return 0;
+}
+
+static const struct file_operations ocfs2_dlm_debug_fops = {
+	.open =		ocfs2_dlm_debug_open,
+	.release =	ocfs2_dlm_debug_release,
+	.read =		seq_read,
+	.llseek =	seq_lseek,
+};
+
+static int ocfs2_dlm_init_debug(struct ocfs2_super *osb)
+{
+	int ret = 0;
+	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
+
+	dlm_debug->d_locking_state = debugfs_create_file("locking_state",
+							 S_IFREG|S_IRUSR,
+							 osb->osb_debug_root,
+							 osb,
+							 &ocfs2_dlm_debug_fops);
+	if (!dlm_debug->d_locking_state) {
+		ret = -EINVAL;
+		mlog(ML_ERROR,
+		     "Unable to create locking state debugfs file.\n");
+		goto out;
+	}
+
+	ocfs2_get_dlm_debug(dlm_debug);
+out:
+	return ret;
+}
+
+static void ocfs2_dlm_shutdown_debug(struct ocfs2_super *osb)
+{
+	struct ocfs2_dlm_debug *dlm_debug = osb->osb_dlm_debug;
+
+	if (dlm_debug) {
+		debugfs_remove(dlm_debug->d_locking_state);
+		ocfs2_put_dlm_debug(dlm_debug);
+	}
+}
+
+int ocfs2_dlm_init(struct ocfs2_super *osb)
+{
+	int status = 0;
+	struct ocfs2_cluster_connection *conn = NULL;
+
+	if (ocfs2_mount_local(osb)) {
+		osb->node_num = 0;
+		goto local;
+	}
+
+	status = ocfs2_dlm_init_debug(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/* launch downconvert thread */
+	osb->dc_task = kthread_run(ocfs2_downconvert_thread, osb, "ocfs2dc");
+	if (IS_ERR(osb->dc_task)) {
+		status = PTR_ERR(osb->dc_task);
+		osb->dc_task = NULL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/* for now, uuid == domain */
+	status = ocfs2_cluster_connect(osb->osb_cluster_stack,
+				       osb->osb_cluster_name,
+				       strlen(osb->osb_cluster_name),
+				       osb->uuid_str,
+				       strlen(osb->uuid_str),
+				       &lproto, ocfs2_do_node_down, osb,
+				       &conn);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_cluster_this_node(conn, &osb->node_num);
+	if (status < 0) {
+		mlog_errno(status);
+		mlog(ML_ERROR,
+		     "could not find this host's node number\n");
+		ocfs2_cluster_disconnect(conn, 0);
+		goto bail;
+	}
+
+local:
+	ocfs2_super_lock_res_init(&osb->osb_super_lockres, osb);
+	ocfs2_rename_lock_res_init(&osb->osb_rename_lockres, osb);
+	ocfs2_nfs_sync_lock_res_init(&osb->osb_nfs_sync_lockres, osb);
+	ocfs2_orphan_scan_lock_res_init(&osb->osb_orphan_scan.os_lockres, osb);
+
+	osb->cconn = conn;
+
+	status = 0;
+bail:
+	if (status < 0) {
+		ocfs2_dlm_shutdown_debug(osb);
+		if (osb->dc_task)
+			kthread_stop(osb->dc_task);
+	}
+
+	return status;
+}
+
+void ocfs2_dlm_shutdown(struct ocfs2_super *osb,
+			int hangup_pending)
+{
+	ocfs2_drop_osb_locks(osb);
+
+	/*
+	 * Now that we have dropped all locks and ocfs2_dismount_volume()
+	 * has disabled recovery, the DLM won't be talking to us.  It's
+	 * safe to tear things down before disconnecting the cluster.
+	 */
+
+	if (osb->dc_task) {
+		kthread_stop(osb->dc_task);
+		osb->dc_task = NULL;
+	}
+
+	ocfs2_lock_res_free(&osb->osb_super_lockres);
+	ocfs2_lock_res_free(&osb->osb_rename_lockres);
+	ocfs2_lock_res_free(&osb->osb_nfs_sync_lockres);
+	ocfs2_lock_res_free(&osb->osb_orphan_scan.os_lockres);
+
+	ocfs2_cluster_disconnect(osb->cconn, hangup_pending);
+	osb->cconn = NULL;
+
+	ocfs2_dlm_shutdown_debug(osb);
+}
+
+static int ocfs2_drop_lock(struct ocfs2_super *osb,
+			   struct ocfs2_lock_res *lockres)
+{
+	int ret;
+	unsigned long flags;
+	u32 lkm_flags = 0;
+
+	/* We didn't get anywhere near actually using this lockres. */
+	if (!(lockres->l_flags & OCFS2_LOCK_INITIALIZED))
+		goto out;
+
+	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB)
+		lkm_flags |= DLM_LKF_VALBLK;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+
+	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_FREEING),
+			"lockres %s, flags 0x%lx\n",
+			lockres->l_name, lockres->l_flags);
+
+	while (lockres->l_flags & OCFS2_LOCK_BUSY) {
+		mlog(0, "waiting on busy lock \"%s\": flags = %lx, action = "
+		     "%u, unlock_action = %u\n",
+		     lockres->l_name, lockres->l_flags, lockres->l_action,
+		     lockres->l_unlock_action);
+
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+		/* XXX: Today we just wait on any busy
+		 * locks... Perhaps we need to cancel converts in the
+		 * future? */
+		ocfs2_wait_on_busy_lock(lockres);
+
+		spin_lock_irqsave(&lockres->l_lock, flags);
+	}
+
+	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
+		if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
+		    lockres->l_level == DLM_LOCK_EX &&
+		    !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
+			lockres->l_ops->set_lvb(lockres);
+	}
+
+	if (lockres->l_flags & OCFS2_LOCK_BUSY)
+		mlog(ML_ERROR, "destroying busy lock: \"%s\"\n",
+		     lockres->l_name);
+	if (lockres->l_flags & OCFS2_LOCK_BLOCKED)
+		mlog(0, "destroying blocked lock: \"%s\"\n", lockres->l_name);
+
+	if (!(lockres->l_flags & OCFS2_LOCK_ATTACHED)) {
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		goto out;
+	}
+
+	lockres_clear_flags(lockres, OCFS2_LOCK_ATTACHED);
+
+	/* make sure we never get here while waiting for an ast to
+	 * fire. */
+	BUG_ON(lockres->l_action != OCFS2_AST_INVALID);
+
+	/* is this necessary? */
+	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
+	lockres->l_unlock_action = OCFS2_UNLOCK_DROP_LOCK;
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	mlog(0, "lock %s\n", lockres->l_name);
+
+	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb, lkm_flags);
+	if (ret) {
+		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
+		mlog(ML_ERROR, "lockres flags: %lu\n", lockres->l_flags);
+		ocfs2_dlm_dump_lksb(&lockres->l_lksb);
+		BUG();
+	}
+	mlog(0, "lock %s, successful return from ocfs2_dlm_unlock\n",
+	     lockres->l_name);
+
+	ocfs2_wait_on_busy_lock(lockres);
+out:
+	return 0;
+}
+
+static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
+				       struct ocfs2_lock_res *lockres);
+
+/* Mark the lockres as being dropped. It will no longer be
+ * queued if blocking, but we still may have to wait on it
+ * being dequeued from the downconvert thread before we can consider
+ * it safe to drop.
+ *
+ * You can *not* attempt to call cluster_lock on this lockres anymore. */
+void ocfs2_mark_lockres_freeing(struct ocfs2_super *osb,
+				struct ocfs2_lock_res *lockres)
+{
+	int status;
+	struct ocfs2_mask_waiter mw;
+	unsigned long flags, flags2;
+
+	ocfs2_init_mask_waiter(&mw);
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	lockres->l_flags |= OCFS2_LOCK_FREEING;
+	if (lockres->l_flags & OCFS2_LOCK_QUEUED && current == osb->dc_task) {
+		/*
+		 * We know the downconvert is queued but not in progress
+		 * because we are the downconvert thread and processing
+		 * different lock. So we can just remove the lock from the
+		 * queue. This is not only an optimization but also a way
+		 * to avoid the following deadlock:
+		 *   ocfs2_dentry_post_unlock()
+		 *     ocfs2_dentry_lock_put()
+		 *       ocfs2_drop_dentry_lock()
+		 *         iput()
+		 *           ocfs2_evict_inode()
+		 *             ocfs2_clear_inode()
+		 *               ocfs2_mark_lockres_freeing()
+		 *                 ... blocks waiting for OCFS2_LOCK_QUEUED
+		 *                 since we are the downconvert thread which
+		 *                 should clear the flag.
+		 */
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		spin_lock_irqsave(&osb->dc_task_lock, flags2);
+		list_del_init(&lockres->l_blocked_list);
+		osb->blocked_lock_count--;
+		spin_unlock_irqrestore(&osb->dc_task_lock, flags2);
+		/*
+		 * Warn if we recurse into another post_unlock call.  Strictly
+		 * speaking it isn't a problem but we need to be careful if
+		 * that happens (stack overflow, deadlocks, ...) so warn if
+		 * ocfs2 grows a path for which this can happen.
+		 */
+		WARN_ON_ONCE(lockres->l_ops->post_unlock);
+		/* Since the lock is freeing we don't do much in the fn below */
+		ocfs2_process_blocked_lock(osb, lockres);
+		return;
+	}
+	while (lockres->l_flags & OCFS2_LOCK_QUEUED) {
+		lockres_add_mask_waiter(lockres, &mw, OCFS2_LOCK_QUEUED, 0);
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+		mlog(0, "Waiting on lockres %s\n", lockres->l_name);
+
+		status = ocfs2_wait_for_mask(&mw);
+		if (status)
+			mlog_errno(status);
+
+		spin_lock_irqsave(&lockres->l_lock, flags);
+	}
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+}
+
+void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
+			       struct ocfs2_lock_res *lockres)
+{
+	int ret;
+
+	ocfs2_mark_lockres_freeing(osb, lockres);
+	ret = ocfs2_drop_lock(osb, lockres);
+	if (ret)
+		mlog_errno(ret);
+}
+
+static void ocfs2_drop_osb_locks(struct ocfs2_super *osb)
+{
+	ocfs2_simple_drop_lockres(osb, &osb->osb_super_lockres);
+	ocfs2_simple_drop_lockres(osb, &osb->osb_rename_lockres);
+	ocfs2_simple_drop_lockres(osb, &osb->osb_nfs_sync_lockres);
+	ocfs2_simple_drop_lockres(osb, &osb->osb_orphan_scan.os_lockres);
+}
+
+int ocfs2_drop_inode_locks(struct inode *inode)
+{
+	int status, err;
+
+	/* No need to call ocfs2_mark_lockres_freeing here -
+	 * ocfs2_clear_inode has done it for us. */
+
+	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
+			      &OCFS2_I(inode)->ip_open_lockres);
+	if (err < 0)
+		mlog_errno(err);
+
+	status = err;
+
+	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
+			      &OCFS2_I(inode)->ip_inode_lockres);
+	if (err < 0)
+		mlog_errno(err);
+	if (err < 0 && !status)
+		status = err;
+
+	err = ocfs2_drop_lock(OCFS2_SB(inode->i_sb),
+			      &OCFS2_I(inode)->ip_rw_lockres);
+	if (err < 0)
+		mlog_errno(err);
+	if (err < 0 && !status)
+		status = err;
+
+	return status;
+}
+
+static unsigned int ocfs2_prepare_downconvert(struct ocfs2_lock_res *lockres,
+					      int new_level)
+{
+	assert_spin_locked(&lockres->l_lock);
+
+	BUG_ON(lockres->l_blocking <= DLM_LOCK_NL);
+
+	if (lockres->l_level <= new_level) {
+		mlog(ML_ERROR, "lockres %s, lvl %d <= %d, blcklst %d, mask %d, "
+		     "type %d, flags 0x%lx, hold %d %d, act %d %d, req %d, "
+		     "block %d, pgen %d\n", lockres->l_name, lockres->l_level,
+		     new_level, list_empty(&lockres->l_blocked_list),
+		     list_empty(&lockres->l_mask_waiters), lockres->l_type,
+		     lockres->l_flags, lockres->l_ro_holders,
+		     lockres->l_ex_holders, lockres->l_action,
+		     lockres->l_unlock_action, lockres->l_requested,
+		     lockres->l_blocking, lockres->l_pending_gen);
+		BUG();
+	}
+
+	mlog(ML_BASTS, "lockres %s, level %d => %d, blocking %d\n",
+	     lockres->l_name, lockres->l_level, new_level, lockres->l_blocking);
+
+	lockres->l_action = OCFS2_AST_DOWNCONVERT;
+	lockres->l_requested = new_level;
+	lockres_or_flags(lockres, OCFS2_LOCK_BUSY);
+	return lockres_set_pending(lockres);
+}
+
+static int ocfs2_downconvert_lock(struct ocfs2_super *osb,
+				  struct ocfs2_lock_res *lockres,
+				  int new_level,
+				  int lvb,
+				  unsigned int generation)
+{
+	int ret;
+	u32 dlm_flags = DLM_LKF_CONVERT;
+
+	mlog(ML_BASTS, "lockres %s, level %d => %d\n", lockres->l_name,
+	     lockres->l_level, new_level);
+
+	if (lvb)
+		dlm_flags |= DLM_LKF_VALBLK;
+
+	ret = ocfs2_dlm_lock(osb->cconn,
+			     new_level,
+			     &lockres->l_lksb,
+			     dlm_flags,
+			     lockres->l_name,
+			     OCFS2_LOCK_ID_MAX_LEN - 1);
+	lockres_clear_pending(lockres, generation, osb);
+	if (ret) {
+		ocfs2_log_dlm_error("ocfs2_dlm_lock", ret, lockres);
+		ocfs2_recover_from_dlm_error(lockres, 1);
+		goto bail;
+	}
+
+	ret = 0;
+bail:
+	return ret;
+}
+
+/* returns 1 when the caller should unlock and call ocfs2_dlm_unlock */
+static int ocfs2_prepare_cancel_convert(struct ocfs2_super *osb,
+				        struct ocfs2_lock_res *lockres)
+{
+	assert_spin_locked(&lockres->l_lock);
+
+	if (lockres->l_unlock_action == OCFS2_UNLOCK_CANCEL_CONVERT) {
+		/* If we're already trying to cancel a lock conversion
+		 * then just drop the spinlock and allow the caller to
+		 * requeue this lock. */
+		mlog(ML_BASTS, "lockres %s, skip convert\n", lockres->l_name);
+		return 0;
+	}
+
+	/* were we in a convert when we got the bast fire? */
+	BUG_ON(lockres->l_action != OCFS2_AST_CONVERT &&
+	       lockres->l_action != OCFS2_AST_DOWNCONVERT);
+	/* set things up for the unlockast to know to just
+	 * clear out the ast_action and unset busy, etc. */
+	lockres->l_unlock_action = OCFS2_UNLOCK_CANCEL_CONVERT;
+
+	mlog_bug_on_msg(!(lockres->l_flags & OCFS2_LOCK_BUSY),
+			"lock %s, invalid flags: 0x%lx\n",
+			lockres->l_name, lockres->l_flags);
+
+	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
+
+	return 1;
+}
+
+static int ocfs2_cancel_convert(struct ocfs2_super *osb,
+				struct ocfs2_lock_res *lockres)
+{
+	int ret;
+
+	ret = ocfs2_dlm_unlock(osb->cconn, &lockres->l_lksb,
+			       DLM_LKF_CANCEL);
+	if (ret) {
+		ocfs2_log_dlm_error("ocfs2_dlm_unlock", ret, lockres);
+		ocfs2_recover_from_dlm_error(lockres, 0);
+	}
+
+	mlog(ML_BASTS, "lockres %s\n", lockres->l_name);
+
+	return ret;
+}
+
+static int ocfs2_unblock_lock(struct ocfs2_super *osb,
+			      struct ocfs2_lock_res *lockres,
+			      struct ocfs2_unblock_ctl *ctl)
+{
+	unsigned long flags;
+	int blocking;
+	int new_level;
+	int level;
+	int ret = 0;
+	int set_lvb = 0;
+	unsigned int gen;
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+
+recheck:
+	/*
+	 * Is it still blocking? If not, we have no more work to do.
+	 */
+	if (!(lockres->l_flags & OCFS2_LOCK_BLOCKED)) {
+		BUG_ON(lockres->l_blocking != DLM_LOCK_NL);
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		ret = 0;
+		goto leave;
+	}
+
+	if (lockres->l_flags & OCFS2_LOCK_BUSY) {
+		/* XXX
+		 * This is a *big* race.  The OCFS2_LOCK_PENDING flag
+		 * exists entirely for one reason - another thread has set
+		 * OCFS2_LOCK_BUSY, but has *NOT* yet called dlm_lock().
+		 *
+		 * If we do ocfs2_cancel_convert() before the other thread
+		 * calls dlm_lock(), our cancel will do nothing.  We will
+		 * get no ast, and we will have no way of knowing the
+		 * cancel failed.  Meanwhile, the other thread will call
+		 * into dlm_lock() and wait...forever.
+		 *
+		 * Why forever?  Because another node has asked for the
+		 * lock first; that's why we're here in unblock_lock().
+		 *
+		 * The solution is OCFS2_LOCK_PENDING.  When PENDING is
+		 * set, we just requeue the unblock.  Only when the other
+		 * thread has called dlm_lock() and cleared PENDING will
+		 * we then cancel their request.
+		 *
+		 * All callers of dlm_lock() must set OCFS2_DLM_PENDING
+		 * at the same time they set OCFS2_DLM_BUSY.  They must
+		 * clear OCFS2_DLM_PENDING after dlm_lock() returns.
+		 */
+		if (lockres->l_flags & OCFS2_LOCK_PENDING) {
+			mlog(ML_BASTS, "lockres %s, ReQ: Pending\n",
+			     lockres->l_name);
+			goto leave_requeue;
+		}
+
+		ctl->requeue = 1;
+		ret = ocfs2_prepare_cancel_convert(osb, lockres);
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		if (ret) {
+			ret = ocfs2_cancel_convert(osb, lockres);
+			if (ret < 0)
+				mlog_errno(ret);
+		}
+		goto leave;
+	}
+
+	/*
+	 * This prevents livelocks. OCFS2_LOCK_UPCONVERT_FINISHING flag is
+	 * set when the ast is received for an upconvert just before the
+	 * OCFS2_LOCK_BUSY flag is cleared. Now if the fs received a bast
+	 * on the heels of the ast, we want to delay the downconvert just
+	 * enough to allow the up requestor to do its task. Because this
+	 * lock is in the blocked queue, the lock will be downconverted
+	 * as soon as the requestor is done with the lock.
+	 */
+	if (lockres->l_flags & OCFS2_LOCK_UPCONVERT_FINISHING)
+		goto leave_requeue;
+
+	/*
+	 * How can we block and yet be at NL?  We were trying to upconvert
+	 * from NL and got canceled.  The code comes back here, and now
+	 * we notice and clear BLOCKING.
+	 */
+	if (lockres->l_level == DLM_LOCK_NL) {
+		BUG_ON(lockres->l_ex_holders || lockres->l_ro_holders);
+		mlog(ML_BASTS, "lockres %s, Aborting dc\n", lockres->l_name);
+		lockres->l_blocking = DLM_LOCK_NL;
+		lockres_clear_flags(lockres, OCFS2_LOCK_BLOCKED);
+		spin_unlock_irqrestore(&lockres->l_lock, flags);
+		goto leave;
+	}
+
+	/* if we're blocking an exclusive and we have *any* holders,
+	 * then requeue. */
+	if ((lockres->l_blocking == DLM_LOCK_EX)
+	    && (lockres->l_ex_holders || lockres->l_ro_holders)) {
+		mlog(ML_BASTS, "lockres %s, ReQ: EX/PR Holders %u,%u\n",
+		     lockres->l_name, lockres->l_ex_holders,
+		     lockres->l_ro_holders);
+		goto leave_requeue;
+	}
+
+	/* If it's a PR we're blocking, then only
+	 * requeue if we've got any EX holders */
+	if (lockres->l_blocking == DLM_LOCK_PR &&
+	    lockres->l_ex_holders) {
+		mlog(ML_BASTS, "lockres %s, ReQ: EX Holders %u\n",
+		     lockres->l_name, lockres->l_ex_holders);
+		goto leave_requeue;
+	}
+
+	/*
+	 * Can we get a lock in this state if the holder counts are
+	 * zero? The meta data unblock code used to check this.
+	 */
+	if ((lockres->l_ops->flags & LOCK_TYPE_REQUIRES_REFRESH)
+	    && (lockres->l_flags & OCFS2_LOCK_REFRESHING)) {
+		mlog(ML_BASTS, "lockres %s, ReQ: Lock Refreshing\n",
+		     lockres->l_name);
+		goto leave_requeue;
+	}
+
+	new_level = ocfs2_highest_compat_lock_level(lockres->l_blocking);
+
+	if (lockres->l_ops->check_downconvert
+	    && !lockres->l_ops->check_downconvert(lockres, new_level)) {
+		mlog(ML_BASTS, "lockres %s, ReQ: Checkpointing\n",
+		     lockres->l_name);
+		goto leave_requeue;
+	}
+
+	/* If we get here, then we know that there are no more
+	 * incompatible holders (and anyone asking for an incompatible
+	 * lock is blocked). We can now downconvert the lock */
+	if (!lockres->l_ops->downconvert_worker)
+		goto downconvert;
+
+	/* Some lockres types want to do a bit of work before
+	 * downconverting a lock. Allow that here. The worker function
+	 * may sleep, so we save off a copy of what we're blocking as
+	 * it may change while we're not holding the spin lock. */
+	blocking = lockres->l_blocking;
+	level = lockres->l_level;
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	ctl->unblock_action = lockres->l_ops->downconvert_worker(lockres, blocking);
+
+	if (ctl->unblock_action == UNBLOCK_STOP_POST) {
+		mlog(ML_BASTS, "lockres %s, UNBLOCK_STOP_POST\n",
+		     lockres->l_name);
+		goto leave;
+	}
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	if ((blocking != lockres->l_blocking) || (level != lockres->l_level)) {
+		/* If this changed underneath us, then we can't drop
+		 * it just yet. */
+		mlog(ML_BASTS, "lockres %s, block=%d:%d, level=%d:%d, "
+		     "Recheck\n", lockres->l_name, blocking,
+		     lockres->l_blocking, level, lockres->l_level);
+		goto recheck;
+	}
+
+downconvert:
+	ctl->requeue = 0;
+
+	if (lockres->l_ops->flags & LOCK_TYPE_USES_LVB) {
+		if (lockres->l_level == DLM_LOCK_EX)
+			set_lvb = 1;
+
+		/*
+		 * We only set the lvb if the lock has been fully
+		 * refreshed - otherwise we risk setting stale
+		 * data. Otherwise, there's no need to actually clear
+		 * out the lvb here as it's value is still valid.
+		 */
+		if (set_lvb && !(lockres->l_flags & OCFS2_LOCK_NEEDS_REFRESH))
+			lockres->l_ops->set_lvb(lockres);
+	}
+
+	gen = ocfs2_prepare_downconvert(lockres, new_level);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+	ret = ocfs2_downconvert_lock(osb, lockres, new_level, set_lvb,
+				     gen);
+
+leave:
+	if (ret)
+		mlog_errno(ret);
+	return ret;
+
+leave_requeue:
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+	ctl->requeue = 1;
+
+	return 0;
+}
+
+static int ocfs2_data_convert_worker(struct ocfs2_lock_res *lockres,
+				     int blocking)
+{
+	struct inode *inode;
+	struct address_space *mapping;
+	struct ocfs2_inode_info *oi;
+
+       	inode = ocfs2_lock_res_inode(lockres);
+	mapping = inode->i_mapping;
+
+	if (S_ISDIR(inode->i_mode)) {
+		oi = OCFS2_I(inode);
+		oi->ip_dir_lock_gen++;
+		mlog(0, "generation: %u\n", oi->ip_dir_lock_gen);
+		goto out;
+	}
+
+	if (!S_ISREG(inode->i_mode))
+		goto out;
+
+	/*
+	 * We need this before the filemap_fdatawrite() so that it can
+	 * transfer the dirty bit from the PTE to the
+	 * page. Unfortunately this means that even for EX->PR
+	 * downconverts, we'll lose our mappings and have to build
+	 * them up again.
+	 */
+	unmap_mapping_range(mapping, 0, 0, 0);
+
+	if (filemap_fdatawrite(mapping)) {
+		mlog(ML_ERROR, "Could not sync inode %llu for downconvert!",
+		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+	}
+	sync_mapping_buffers(mapping);
+	if (blocking == DLM_LOCK_EX) {
+		truncate_inode_pages(mapping, 0);
+	} else {
+		/* We only need to wait on the I/O if we're not also
+		 * truncating pages because truncate_inode_pages waits
+		 * for us above. We don't truncate pages if we're
+		 * blocking anything < EXMODE because we want to keep
+		 * them around in that case. */
+		filemap_fdatawait(mapping);
+	}
+
+out:
+	return UNBLOCK_CONTINUE;
+}
+
+static int ocfs2_ci_checkpointed(struct ocfs2_caching_info *ci,
+				 struct ocfs2_lock_res *lockres,
+				 int new_level)
+{
+	int checkpointed = ocfs2_ci_fully_checkpointed(ci);
+
+	BUG_ON(new_level != DLM_LOCK_NL && new_level != DLM_LOCK_PR);
+	BUG_ON(lockres->l_level != DLM_LOCK_EX && !checkpointed);
+
+	if (checkpointed)
+		return 1;
+
+	ocfs2_start_checkpoint(OCFS2_SB(ocfs2_metadata_cache_get_super(ci)));
+	return 0;
+}
+
+static int ocfs2_check_meta_downconvert(struct ocfs2_lock_res *lockres,
+					int new_level)
+{
+	struct inode *inode = ocfs2_lock_res_inode(lockres);
+
+	return ocfs2_ci_checkpointed(INODE_CACHE(inode), lockres, new_level);
+}
+
+static void ocfs2_set_meta_lvb(struct ocfs2_lock_res *lockres)
+{
+	struct inode *inode = ocfs2_lock_res_inode(lockres);
+
+	__ocfs2_stuff_meta_lvb(inode);
+}
+
+/*
+ * Does the final reference drop on our dentry lock. Right now this
+ * happens in the downconvert thread, but we could choose to simplify the
+ * dlmglue API and push these off to the ocfs2_wq in the future.
+ */
+static void ocfs2_dentry_post_unlock(struct ocfs2_super *osb,
+				     struct ocfs2_lock_res *lockres)
+{
+	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
+	ocfs2_dentry_lock_put(osb, dl);
+}
+
+/*
+ * d_delete() matching dentries before the lock downconvert.
+ *
+ * At this point, any process waiting to destroy the
+ * dentry_lock due to last ref count is stopped by the
+ * OCFS2_LOCK_QUEUED flag.
+ *
+ * We have two potential problems
+ *
+ * 1) If we do the last reference drop on our dentry_lock (via dput)
+ *    we'll wind up in ocfs2_release_dentry_lock(), waiting on
+ *    the downconvert to finish. Instead we take an elevated
+ *    reference and push the drop until after we've completed our
+ *    unblock processing.
+ *
+ * 2) There might be another process with a final reference,
+ *    waiting on us to finish processing. If this is the case, we
+ *    detect it and exit out - there's no more dentries anyway.
+ */
+static int ocfs2_dentry_convert_worker(struct ocfs2_lock_res *lockres,
+				       int blocking)
+{
+	struct ocfs2_dentry_lock *dl = ocfs2_lock_res_dl(lockres);
+	struct ocfs2_inode_info *oi = OCFS2_I(dl->dl_inode);
+	struct dentry *dentry;
+	unsigned long flags;
+	int extra_ref = 0;
+
+	/*
+	 * This node is blocking another node from getting a read
+	 * lock. This happens when we've renamed within a
+	 * directory. We've forced the other nodes to d_delete(), but
+	 * we never actually dropped our lock because it's still
+	 * valid. The downconvert code will retain a PR for this node,
+	 * so there's no further work to do.
+	 */
+	if (blocking == DLM_LOCK_PR)
+		return UNBLOCK_CONTINUE;
+
+	/*
+	 * Mark this inode as potentially orphaned. The code in
+	 * ocfs2_delete_inode() will figure out whether it actually
+	 * needs to be freed or not.
+	 */
+	spin_lock(&oi->ip_lock);
+	oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
+	spin_unlock(&oi->ip_lock);
+
+	/*
+	 * Yuck. We need to make sure however that the check of
+	 * OCFS2_LOCK_FREEING and the extra reference are atomic with
+	 * respect to a reference decrement or the setting of that
+	 * flag.
+	 */
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	spin_lock(&dentry_attach_lock);
+	if (!(lockres->l_flags & OCFS2_LOCK_FREEING)
+	    && dl->dl_count) {
+		dl->dl_count++;
+		extra_ref = 1;
+	}
+	spin_unlock(&dentry_attach_lock);
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	mlog(0, "extra_ref = %d\n", extra_ref);
+
+	/*
+	 * We have a process waiting on us in ocfs2_dentry_iput(),
+	 * which means we can't have any more outstanding
+	 * aliases. There's no need to do any more work.
+	 */
+	if (!extra_ref)
+		return UNBLOCK_CONTINUE;
+
+	spin_lock(&dentry_attach_lock);
+	while (1) {
+		dentry = ocfs2_find_local_alias(dl->dl_inode,
+						dl->dl_parent_blkno, 1);
+		if (!dentry)
+			break;
+		spin_unlock(&dentry_attach_lock);
+
+		if (S_ISDIR(dl->dl_inode->i_mode))
+			shrink_dcache_parent(dentry);
+
+		mlog(0, "d_delete(%pd);\n", dentry);
+
+		/*
+		 * The following dcache calls may do an
+		 * iput(). Normally we don't want that from the
+		 * downconverting thread, but in this case it's ok
+		 * because the requesting node already has an
+		 * exclusive lock on the inode, so it can't be queued
+		 * for a downconvert.
+		 */
+		d_delete(dentry);
+		dput(dentry);
+
+		spin_lock(&dentry_attach_lock);
+	}
+	spin_unlock(&dentry_attach_lock);
+
+	/*
+	 * If we are the last holder of this dentry lock, there is no
+	 * reason to downconvert so skip straight to the unlock.
+	 */
+	if (dl->dl_count == 1)
+		return UNBLOCK_STOP_POST;
+
+	return UNBLOCK_CONTINUE_POST;
+}
+
+static int ocfs2_check_refcount_downconvert(struct ocfs2_lock_res *lockres,
+					    int new_level)
+{
+	struct ocfs2_refcount_tree *tree =
+				ocfs2_lock_res_refcount_tree(lockres);
+
+	return ocfs2_ci_checkpointed(&tree->rf_ci, lockres, new_level);
+}
+
+static int ocfs2_refcount_convert_worker(struct ocfs2_lock_res *lockres,
+					 int blocking)
+{
+	struct ocfs2_refcount_tree *tree =
+				ocfs2_lock_res_refcount_tree(lockres);
+
+	ocfs2_metadata_cache_purge(&tree->rf_ci);
+
+	return UNBLOCK_CONTINUE;
+}
+
+static void ocfs2_set_qinfo_lvb(struct ocfs2_lock_res *lockres)
+{
+	struct ocfs2_qinfo_lvb *lvb;
+	struct ocfs2_mem_dqinfo *oinfo = ocfs2_lock_res_qinfo(lockres);
+	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
+					    oinfo->dqi_gi.dqi_type);
+
+	lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+	lvb->lvb_version = OCFS2_QINFO_LVB_VERSION;
+	lvb->lvb_bgrace = cpu_to_be32(info->dqi_bgrace);
+	lvb->lvb_igrace = cpu_to_be32(info->dqi_igrace);
+	lvb->lvb_syncms = cpu_to_be32(oinfo->dqi_syncms);
+	lvb->lvb_blocks = cpu_to_be32(oinfo->dqi_gi.dqi_blocks);
+	lvb->lvb_free_blk = cpu_to_be32(oinfo->dqi_gi.dqi_free_blk);
+	lvb->lvb_free_entry = cpu_to_be32(oinfo->dqi_gi.dqi_free_entry);
+}
+
+void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex)
+{
+	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
+	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+
+	if (!ocfs2_is_hard_readonly(osb) && !ocfs2_mount_local(osb))
+		ocfs2_cluster_unlock(osb, lockres, level);
+}
+
+static int ocfs2_refresh_qinfo(struct ocfs2_mem_dqinfo *oinfo)
+{
+	struct mem_dqinfo *info = sb_dqinfo(oinfo->dqi_gi.dqi_sb,
+					    oinfo->dqi_gi.dqi_type);
+	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
+	struct ocfs2_qinfo_lvb *lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
+	struct buffer_head *bh = NULL;
+	struct ocfs2_global_disk_dqinfo *gdinfo;
+	int status = 0;
+
+	if (ocfs2_dlm_lvb_valid(&lockres->l_lksb) &&
+	    lvb->lvb_version == OCFS2_QINFO_LVB_VERSION) {
+		info->dqi_bgrace = be32_to_cpu(lvb->lvb_bgrace);
+		info->dqi_igrace = be32_to_cpu(lvb->lvb_igrace);
+		oinfo->dqi_syncms = be32_to_cpu(lvb->lvb_syncms);
+		oinfo->dqi_gi.dqi_blocks = be32_to_cpu(lvb->lvb_blocks);
+		oinfo->dqi_gi.dqi_free_blk = be32_to_cpu(lvb->lvb_free_blk);
+		oinfo->dqi_gi.dqi_free_entry =
+					be32_to_cpu(lvb->lvb_free_entry);
+	} else {
+		status = ocfs2_read_quota_phys_block(oinfo->dqi_gqinode,
+						     oinfo->dqi_giblk, &bh);
+		if (status) {
+			mlog_errno(status);
+			goto bail;
+		}
+		gdinfo = (struct ocfs2_global_disk_dqinfo *)
+					(bh->b_data + OCFS2_GLOBAL_INFO_OFF);
+		info->dqi_bgrace = le32_to_cpu(gdinfo->dqi_bgrace);
+		info->dqi_igrace = le32_to_cpu(gdinfo->dqi_igrace);
+		oinfo->dqi_syncms = le32_to_cpu(gdinfo->dqi_syncms);
+		oinfo->dqi_gi.dqi_blocks = le32_to_cpu(gdinfo->dqi_blocks);
+		oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(gdinfo->dqi_free_blk);
+		oinfo->dqi_gi.dqi_free_entry =
+					le32_to_cpu(gdinfo->dqi_free_entry);
+		brelse(bh);
+		ocfs2_track_lock_refresh(lockres);
+	}
+
+bail:
+	return status;
+}
+
+/* Lock quota info, this function expects at least shared lock on the quota file
+ * so that we can safely refresh quota info from disk. */
+int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex)
+{
+	struct ocfs2_lock_res *lockres = &oinfo->dqi_gqlock;
+	struct ocfs2_super *osb = OCFS2_SB(oinfo->dqi_gi.dqi_sb);
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	int status = 0;
+
+	/* On RO devices, locking really isn't needed... */
+	if (ocfs2_is_hard_readonly(osb)) {
+		if (ex)
+			status = -EROFS;
+		goto bail;
+	}
+	if (ocfs2_mount_local(osb))
+		goto bail;
+
+	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	if (!ocfs2_should_refresh_lock_res(lockres))
+		goto bail;
+	/* OK, we have the lock but we need to refresh the quota info */
+	status = ocfs2_refresh_qinfo(oinfo);
+	if (status)
+		ocfs2_qinfo_unlock(oinfo, ex);
+	ocfs2_complete_lock_res_refresh(lockres, status);
+bail:
+	return status;
+}
+
+int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex)
+{
+	int status;
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
+	struct ocfs2_super *osb = lockres->l_priv;
+
+
+	if (ocfs2_is_hard_readonly(osb))
+		return -EROFS;
+
+	if (ocfs2_mount_local(osb))
+		return 0;
+
+	status = ocfs2_cluster_lock(osb, lockres, level, 0, 0);
+	if (status < 0)
+		mlog_errno(status);
+
+	return status;
+}
+
+void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex)
+{
+	int level = ex ? DLM_LOCK_EX : DLM_LOCK_PR;
+	struct ocfs2_lock_res *lockres = &ref_tree->rf_lockres;
+	struct ocfs2_super *osb = lockres->l_priv;
+
+	if (!ocfs2_mount_local(osb))
+		ocfs2_cluster_unlock(osb, lockres, level);
+}
+
+static void ocfs2_process_blocked_lock(struct ocfs2_super *osb,
+				       struct ocfs2_lock_res *lockres)
+{
+	int status;
+	struct ocfs2_unblock_ctl ctl = {0, 0,};
+	unsigned long flags;
+
+	/* Our reference to the lockres in this function can be
+	 * considered valid until we remove the OCFS2_LOCK_QUEUED
+	 * flag. */
+
+	BUG_ON(!lockres);
+	BUG_ON(!lockres->l_ops);
+
+	mlog(ML_BASTS, "lockres %s blocked\n", lockres->l_name);
+
+	/* Detect whether a lock has been marked as going away while
+	 * the downconvert thread was processing other things. A lock can
+	 * still be marked with OCFS2_LOCK_FREEING after this check,
+	 * but short circuiting here will still save us some
+	 * performance. */
+	spin_lock_irqsave(&lockres->l_lock, flags);
+	if (lockres->l_flags & OCFS2_LOCK_FREEING)
+		goto unqueue;
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	status = ocfs2_unblock_lock(osb, lockres, &ctl);
+	if (status < 0)
+		mlog_errno(status);
+
+	spin_lock_irqsave(&lockres->l_lock, flags);
+unqueue:
+	if (lockres->l_flags & OCFS2_LOCK_FREEING || !ctl.requeue) {
+		lockres_clear_flags(lockres, OCFS2_LOCK_QUEUED);
+	} else
+		ocfs2_schedule_blocked_lock(osb, lockres);
+
+	mlog(ML_BASTS, "lockres %s, requeue = %s.\n", lockres->l_name,
+	     ctl.requeue ? "yes" : "no");
+	spin_unlock_irqrestore(&lockres->l_lock, flags);
+
+	if (ctl.unblock_action != UNBLOCK_CONTINUE
+	    && lockres->l_ops->post_unlock)
+		lockres->l_ops->post_unlock(osb, lockres);
+}
+
+static void ocfs2_schedule_blocked_lock(struct ocfs2_super *osb,
+					struct ocfs2_lock_res *lockres)
+{
+	unsigned long flags;
+
+	assert_spin_locked(&lockres->l_lock);
+
+	if (lockres->l_flags & OCFS2_LOCK_FREEING) {
+		/* Do not schedule a lock for downconvert when it's on
+		 * the way to destruction - any nodes wanting access
+		 * to the resource will get it soon. */
+		mlog(ML_BASTS, "lockres %s won't be scheduled: flags 0x%lx\n",
+		     lockres->l_name, lockres->l_flags);
+		return;
+	}
+
+	lockres_or_flags(lockres, OCFS2_LOCK_QUEUED);
+
+	spin_lock_irqsave(&osb->dc_task_lock, flags);
+	if (list_empty(&lockres->l_blocked_list)) {
+		list_add_tail(&lockres->l_blocked_list,
+			      &osb->blocked_lock_list);
+		osb->blocked_lock_count++;
+	}
+	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
+}
+
+static void ocfs2_downconvert_thread_do_work(struct ocfs2_super *osb)
+{
+	unsigned long processed;
+	unsigned long flags;
+	struct ocfs2_lock_res *lockres;
+
+	spin_lock_irqsave(&osb->dc_task_lock, flags);
+	/* grab this early so we know to try again if a state change and
+	 * wake happens part-way through our work  */
+	osb->dc_work_sequence = osb->dc_wake_sequence;
+
+	processed = osb->blocked_lock_count;
+	while (processed) {
+		BUG_ON(list_empty(&osb->blocked_lock_list));
+
+		lockres = list_entry(osb->blocked_lock_list.next,
+				     struct ocfs2_lock_res, l_blocked_list);
+		list_del_init(&lockres->l_blocked_list);
+		osb->blocked_lock_count--;
+		spin_unlock_irqrestore(&osb->dc_task_lock, flags);
+
+		BUG_ON(!processed);
+		processed--;
+
+		ocfs2_process_blocked_lock(osb, lockres);
+
+		spin_lock_irqsave(&osb->dc_task_lock, flags);
+	}
+	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
+}
+
+static int ocfs2_downconvert_thread_lists_empty(struct ocfs2_super *osb)
+{
+	int empty = 0;
+	unsigned long flags;
+
+	spin_lock_irqsave(&osb->dc_task_lock, flags);
+	if (list_empty(&osb->blocked_lock_list))
+		empty = 1;
+
+	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
+	return empty;
+}
+
+static int ocfs2_downconvert_thread_should_wake(struct ocfs2_super *osb)
+{
+	int should_wake = 0;
+	unsigned long flags;
+
+	spin_lock_irqsave(&osb->dc_task_lock, flags);
+	if (osb->dc_work_sequence != osb->dc_wake_sequence)
+		should_wake = 1;
+	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
+
+	return should_wake;
+}
+
+static int ocfs2_downconvert_thread(void *arg)
+{
+	int status = 0;
+	struct ocfs2_super *osb = arg;
+
+	/* only quit once we've been asked to stop and there is no more
+	 * work available */
+	while (!(kthread_should_stop() &&
+		ocfs2_downconvert_thread_lists_empty(osb))) {
+
+		wait_event_interruptible(osb->dc_event,
+					 ocfs2_downconvert_thread_should_wake(osb) ||
+					 kthread_should_stop());
+
+		mlog(0, "downconvert_thread: awoken\n");
+
+		ocfs2_downconvert_thread_do_work(osb);
+	}
+
+	osb->dc_task = NULL;
+	return status;
+}
+
+void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&osb->dc_task_lock, flags);
+	/* make sure the voting thread gets a swipe at whatever changes
+	 * the caller may have made to the voting state */
+	osb->dc_wake_sequence++;
+	spin_unlock_irqrestore(&osb->dc_task_lock, flags);
+	wake_up(&osb->dc_event);
+}
diff --git a/fs/ocfs2/dlmglue.h b/fs/ocfs2/dlmglue.h
new file mode 100644
index 000000000..d293a22c3
--- /dev/null
+++ b/fs/ocfs2/dlmglue.h
@@ -0,0 +1,173 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * dlmglue.h
+ *
+ * description here
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+
+#ifndef DLMGLUE_H
+#define DLMGLUE_H
+
+#include "dcache.h"
+
+#define OCFS2_LVB_VERSION 5
+
+struct ocfs2_meta_lvb {
+	__u8         lvb_version;
+	__u8         lvb_reserved0;
+	__be16       lvb_idynfeatures;
+	__be32       lvb_iclusters;
+	__be32       lvb_iuid;
+	__be32       lvb_igid;
+	__be64       lvb_iatime_packed;
+	__be64       lvb_ictime_packed;
+	__be64       lvb_imtime_packed;
+	__be64       lvb_isize;
+	__be16       lvb_imode;
+	__be16       lvb_inlink;
+	__be32       lvb_iattr;
+	__be32       lvb_igeneration;
+	__be32       lvb_reserved2;
+};
+
+#define OCFS2_QINFO_LVB_VERSION 1
+
+struct ocfs2_qinfo_lvb {
+	__u8	lvb_version;
+	__u8	lvb_reserved[3];
+	__be32	lvb_bgrace;
+	__be32	lvb_igrace;
+	__be32	lvb_syncms;
+	__be32	lvb_blocks;
+	__be32	lvb_free_blk;
+	__be32	lvb_free_entry;
+};
+
+#define OCFS2_ORPHAN_LVB_VERSION 1
+
+struct ocfs2_orphan_scan_lvb {
+	__u8	lvb_version;
+	__u8	lvb_reserved[3];
+	__be32	lvb_os_seqno;
+};
+
+/* ocfs2_inode_lock_full() 'arg_flags' flags */
+/* don't wait on recovery. */
+#define OCFS2_META_LOCK_RECOVERY	(0x01)
+/* Instruct the dlm not to queue ourselves on the other node. */
+#define OCFS2_META_LOCK_NOQUEUE		(0x02)
+/* don't block waiting for the downconvert thread, instead return -EAGAIN */
+#define OCFS2_LOCK_NONBLOCK		(0x04)
+
+/* Locking subclasses of inode cluster lock */
+enum {
+	OI_LS_NORMAL = 0,
+	OI_LS_PARENT,
+	OI_LS_RENAME1,
+	OI_LS_RENAME2,
+	OI_LS_REFLINK_TARGET,
+};
+
+int ocfs2_dlm_init(struct ocfs2_super *osb);
+void ocfs2_dlm_shutdown(struct ocfs2_super *osb, int hangup_pending);
+void ocfs2_lock_res_init_once(struct ocfs2_lock_res *res);
+void ocfs2_inode_lock_res_init(struct ocfs2_lock_res *res,
+			       enum ocfs2_lock_type type,
+			       unsigned int generation,
+			       struct inode *inode);
+void ocfs2_dentry_lock_res_init(struct ocfs2_dentry_lock *dl,
+				u64 parent, struct inode *inode);
+struct ocfs2_file_private;
+void ocfs2_file_lock_res_init(struct ocfs2_lock_res *lockres,
+			      struct ocfs2_file_private *fp);
+struct ocfs2_mem_dqinfo;
+void ocfs2_qinfo_lock_res_init(struct ocfs2_lock_res *lockres,
+                               struct ocfs2_mem_dqinfo *info);
+void ocfs2_refcount_lock_res_init(struct ocfs2_lock_res *lockres,
+				  struct ocfs2_super *osb, u64 ref_blkno,
+				  unsigned int generation);
+void ocfs2_lock_res_free(struct ocfs2_lock_res *res);
+int ocfs2_create_new_inode_locks(struct inode *inode);
+int ocfs2_drop_inode_locks(struct inode *inode);
+int ocfs2_rw_lock(struct inode *inode, int write);
+void ocfs2_rw_unlock(struct inode *inode, int write);
+int ocfs2_open_lock(struct inode *inode);
+int ocfs2_try_open_lock(struct inode *inode, int write);
+void ocfs2_open_unlock(struct inode *inode);
+int ocfs2_inode_lock_atime(struct inode *inode,
+			  struct vfsmount *vfsmnt,
+			  int *level);
+int ocfs2_inode_lock_full_nested(struct inode *inode,
+			 struct buffer_head **ret_bh,
+			 int ex,
+			 int arg_flags,
+			 int subclass);
+int ocfs2_inode_lock_with_page(struct inode *inode,
+			      struct buffer_head **ret_bh,
+			      int ex,
+			      struct page *page);
+/* Variants without special locking class or flags */
+#define ocfs2_inode_lock_full(i, r, e, f)\
+		ocfs2_inode_lock_full_nested(i, r, e, f, OI_LS_NORMAL)
+#define ocfs2_inode_lock_nested(i, b, e, s)\
+		ocfs2_inode_lock_full_nested(i, b, e, 0, s)
+/* 99% of the time we don't want to supply any additional flags --
+ * those are for very specific cases only. */
+#define ocfs2_inode_lock(i, b, e) ocfs2_inode_lock_full_nested(i, b, e, 0, OI_LS_NORMAL)
+void ocfs2_inode_unlock(struct inode *inode,
+		       int ex);
+int ocfs2_super_lock(struct ocfs2_super *osb,
+		     int ex);
+void ocfs2_super_unlock(struct ocfs2_super *osb,
+			int ex);
+int ocfs2_orphan_scan_lock(struct ocfs2_super *osb, u32 *seqno);
+void ocfs2_orphan_scan_unlock(struct ocfs2_super *osb, u32 seqno);
+
+int ocfs2_rename_lock(struct ocfs2_super *osb);
+void ocfs2_rename_unlock(struct ocfs2_super *osb);
+int ocfs2_nfs_sync_lock(struct ocfs2_super *osb, int ex);
+void ocfs2_nfs_sync_unlock(struct ocfs2_super *osb, int ex);
+int ocfs2_dentry_lock(struct dentry *dentry, int ex);
+void ocfs2_dentry_unlock(struct dentry *dentry, int ex);
+int ocfs2_file_lock(struct file *file, int ex, int trylock);
+void ocfs2_file_unlock(struct file *file);
+int ocfs2_qinfo_lock(struct ocfs2_mem_dqinfo *oinfo, int ex);
+void ocfs2_qinfo_unlock(struct ocfs2_mem_dqinfo *oinfo, int ex);
+struct ocfs2_refcount_tree;
+int ocfs2_refcount_lock(struct ocfs2_refcount_tree *ref_tree, int ex);
+void ocfs2_refcount_unlock(struct ocfs2_refcount_tree *ref_tree, int ex);
+
+
+void ocfs2_mark_lockres_freeing(struct ocfs2_super *osb,
+				struct ocfs2_lock_res *lockres);
+void ocfs2_simple_drop_lockres(struct ocfs2_super *osb,
+			       struct ocfs2_lock_res *lockres);
+
+/* for the downconvert thread */
+void ocfs2_wake_downconvert_thread(struct ocfs2_super *osb);
+
+struct ocfs2_dlm_debug *ocfs2_new_dlm_debug(void);
+void ocfs2_put_dlm_debug(struct ocfs2_dlm_debug *dlm_debug);
+
+/* To set the locking protocol on module initialization */
+void ocfs2_set_locking_protocol(void);
+#endif	/* DLMGLUE_H */
diff --git a/fs/ocfs2/export.c b/fs/ocfs2/export.c
new file mode 100644
index 000000000..827fc9809
--- /dev/null
+++ b/fs/ocfs2/export.c
@@ -0,0 +1,271 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * export.c
+ *
+ * Functions to facilitate NFS exporting
+ *
+ * Copyright (C) 2002, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "dir.h"
+#include "dlmglue.h"
+#include "dcache.h"
+#include "export.h"
+#include "inode.h"
+
+#include "buffer_head_io.h"
+#include "suballoc.h"
+#include "ocfs2_trace.h"
+
+struct ocfs2_inode_handle
+{
+	u64 ih_blkno;
+	u32 ih_generation;
+};
+
+static struct dentry *ocfs2_get_dentry(struct super_block *sb,
+		struct ocfs2_inode_handle *handle)
+{
+	struct inode *inode;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	u64 blkno = handle->ih_blkno;
+	int status, set;
+	struct dentry *result;
+
+	trace_ocfs2_get_dentry_begin(sb, handle, (unsigned long long)blkno);
+
+	if (blkno == 0) {
+		result = ERR_PTR(-ESTALE);
+		goto bail;
+	}
+
+	inode = ocfs2_ilookup(sb, blkno);
+	/*
+	 * If the inode exists in memory, we only need to check it's
+	 * generation number
+	 */
+	if (inode)
+		goto check_gen;
+
+	/*
+	 * This will synchronize us against ocfs2_delete_inode() on
+	 * all nodes
+	 */
+	status = ocfs2_nfs_sync_lock(osb, 1);
+	if (status < 0) {
+		mlog(ML_ERROR, "getting nfs sync lock(EX) failed %d\n", status);
+		goto check_err;
+	}
+
+	status = ocfs2_test_inode_bit(osb, blkno, &set);
+	if (status < 0) {
+		if (status == -EINVAL) {
+			/*
+			 * The blkno NFS gave us doesn't even show up
+			 * as an inode, we return -ESTALE to be
+			 * nice
+			 */
+			status = -ESTALE;
+		} else
+			mlog(ML_ERROR, "test inode bit failed %d\n", status);
+		goto unlock_nfs_sync;
+	}
+
+	trace_ocfs2_get_dentry_test_bit(status, set);
+	/* If the inode allocator bit is clear, this inode must be stale */
+	if (!set) {
+		status = -ESTALE;
+		goto unlock_nfs_sync;
+	}
+
+	inode = ocfs2_iget(osb, blkno, 0, 0);
+
+unlock_nfs_sync:
+	ocfs2_nfs_sync_unlock(osb, 1);
+
+check_err:
+	if (status < 0) {
+		if (status == -ESTALE) {
+			trace_ocfs2_get_dentry_stale((unsigned long long)blkno,
+						     handle->ih_generation);
+		}
+		result = ERR_PTR(status);
+		goto bail;
+	}
+
+	if (IS_ERR(inode)) {
+		mlog_errno(PTR_ERR(inode));
+		result = (void *)inode;
+		goto bail;
+	}
+
+check_gen:
+	if (handle->ih_generation != inode->i_generation) {
+		iput(inode);
+		trace_ocfs2_get_dentry_generation((unsigned long long)blkno,
+						  handle->ih_generation,
+						  inode->i_generation);
+		result = ERR_PTR(-ESTALE);
+		goto bail;
+	}
+
+	result = d_obtain_alias(inode);
+	if (IS_ERR(result))
+		mlog_errno(PTR_ERR(result));
+
+bail:
+	trace_ocfs2_get_dentry_end(result);
+	return result;
+}
+
+static struct dentry *ocfs2_get_parent(struct dentry *child)
+{
+	int status;
+	u64 blkno;
+	struct dentry *parent;
+	struct inode *dir = d_inode(child);
+
+	trace_ocfs2_get_parent(child, child->d_name.len, child->d_name.name,
+			       (unsigned long long)OCFS2_I(dir)->ip_blkno);
+
+	status = ocfs2_inode_lock(dir, NULL, 0);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		parent = ERR_PTR(status);
+		goto bail;
+	}
+
+	status = ocfs2_lookup_ino_from_name(dir, "..", 2, &blkno);
+	if (status < 0) {
+		parent = ERR_PTR(-ENOENT);
+		goto bail_unlock;
+	}
+
+	parent = d_obtain_alias(ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0));
+
+bail_unlock:
+	ocfs2_inode_unlock(dir, 0);
+
+bail:
+	trace_ocfs2_get_parent_end(parent);
+
+	return parent;
+}
+
+static int ocfs2_encode_fh(struct inode *inode, u32 *fh_in, int *max_len,
+			   struct inode *parent)
+{
+	int len = *max_len;
+	int type = 1;
+	u64 blkno;
+	u32 generation;
+	__le32 *fh = (__force __le32 *) fh_in;
+
+#ifdef TRACE_HOOKS_ARE_NOT_BRAINDEAD_IN_YOUR_OPINION
+#error "You go ahead and fix that mess, then.  Somehow"
+	trace_ocfs2_encode_fh_begin(dentry, dentry->d_name.len,
+				    dentry->d_name.name,
+				    fh, len, connectable);
+#endif
+
+	if (parent && (len < 6)) {
+		*max_len = 6;
+		type = FILEID_INVALID;
+		goto bail;
+	} else if (len < 3) {
+		*max_len = 3;
+		type = FILEID_INVALID;
+		goto bail;
+	}
+
+	blkno = OCFS2_I(inode)->ip_blkno;
+	generation = inode->i_generation;
+
+	trace_ocfs2_encode_fh_self((unsigned long long)blkno, generation);
+
+	len = 3;
+	fh[0] = cpu_to_le32((u32)(blkno >> 32));
+	fh[1] = cpu_to_le32((u32)(blkno & 0xffffffff));
+	fh[2] = cpu_to_le32(generation);
+
+	if (parent) {
+		blkno = OCFS2_I(parent)->ip_blkno;
+		generation = parent->i_generation;
+
+		fh[3] = cpu_to_le32((u32)(blkno >> 32));
+		fh[4] = cpu_to_le32((u32)(blkno & 0xffffffff));
+		fh[5] = cpu_to_le32(generation);
+
+		len = 6;
+		type = 2;
+
+		trace_ocfs2_encode_fh_parent((unsigned long long)blkno,
+					     generation);
+	}
+
+	*max_len = len;
+
+bail:
+	trace_ocfs2_encode_fh_type(type);
+	return type;
+}
+
+static struct dentry *ocfs2_fh_to_dentry(struct super_block *sb,
+		struct fid *fid, int fh_len, int fh_type)
+{
+	struct ocfs2_inode_handle handle;
+
+	if (fh_len < 3 || fh_type > 2)
+		return NULL;
+
+	handle.ih_blkno = (u64)le32_to_cpu(fid->raw[0]) << 32;
+	handle.ih_blkno |= (u64)le32_to_cpu(fid->raw[1]);
+	handle.ih_generation = le32_to_cpu(fid->raw[2]);
+	return ocfs2_get_dentry(sb, &handle);
+}
+
+static struct dentry *ocfs2_fh_to_parent(struct super_block *sb,
+		struct fid *fid, int fh_len, int fh_type)
+{
+	struct ocfs2_inode_handle parent;
+
+	if (fh_type != 2 || fh_len < 6)
+		return NULL;
+
+	parent.ih_blkno = (u64)le32_to_cpu(fid->raw[3]) << 32;
+	parent.ih_blkno |= (u64)le32_to_cpu(fid->raw[4]);
+	parent.ih_generation = le32_to_cpu(fid->raw[5]);
+	return ocfs2_get_dentry(sb, &parent);
+}
+
+const struct export_operations ocfs2_export_ops = {
+	.encode_fh	= ocfs2_encode_fh,
+	.fh_to_dentry	= ocfs2_fh_to_dentry,
+	.fh_to_parent	= ocfs2_fh_to_parent,
+	.get_parent	= ocfs2_get_parent,
+};
diff --git a/fs/ocfs2/export.h b/fs/ocfs2/export.h
new file mode 100644
index 000000000..41a738678
--- /dev/null
+++ b/fs/ocfs2/export.h
@@ -0,0 +1,33 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * export.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_EXPORT_H
+#define OCFS2_EXPORT_H
+
+#include <linux/exportfs.h>
+
+extern const struct export_operations ocfs2_export_ops;
+
+#endif /* OCFS2_EXPORT_H */
diff --git a/fs/ocfs2/extent_map.c b/fs/ocfs2/extent_map.c
new file mode 100644
index 000000000..767370b65
--- /dev/null
+++ b/fs/ocfs2/extent_map.c
@@ -0,0 +1,994 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * extent_map.c
+ *
+ * Block/Cluster mapping functions
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License, version 2,  as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+#include <linux/fiemap.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "inode.h"
+#include "super.h"
+#include "symlink.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+/*
+ * The extent caching implementation is intentionally trivial.
+ *
+ * We only cache a small number of extents stored directly on the
+ * inode, so linear order operations are acceptable. If we ever want
+ * to increase the size of the extent map, then these algorithms must
+ * get smarter.
+ */
+
+void ocfs2_extent_map_init(struct inode *inode)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	oi->ip_extent_map.em_num_items = 0;
+	INIT_LIST_HEAD(&oi->ip_extent_map.em_list);
+}
+
+static void __ocfs2_extent_map_lookup(struct ocfs2_extent_map *em,
+				      unsigned int cpos,
+				      struct ocfs2_extent_map_item **ret_emi)
+{
+	unsigned int range;
+	struct ocfs2_extent_map_item *emi;
+
+	*ret_emi = NULL;
+
+	list_for_each_entry(emi, &em->em_list, ei_list) {
+		range = emi->ei_cpos + emi->ei_clusters;
+
+		if (cpos >= emi->ei_cpos && cpos < range) {
+			list_move(&emi->ei_list, &em->em_list);
+
+			*ret_emi = emi;
+			break;
+		}
+	}
+}
+
+static int ocfs2_extent_map_lookup(struct inode *inode, unsigned int cpos,
+				   unsigned int *phys, unsigned int *len,
+				   unsigned int *flags)
+{
+	unsigned int coff;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_extent_map_item *emi;
+
+	spin_lock(&oi->ip_lock);
+
+	__ocfs2_extent_map_lookup(&oi->ip_extent_map, cpos, &emi);
+	if (emi) {
+		coff = cpos - emi->ei_cpos;
+		*phys = emi->ei_phys + coff;
+		if (len)
+			*len = emi->ei_clusters - coff;
+		if (flags)
+			*flags = emi->ei_flags;
+	}
+
+	spin_unlock(&oi->ip_lock);
+
+	if (emi == NULL)
+		return -ENOENT;
+
+	return 0;
+}
+
+/*
+ * Forget about all clusters equal to or greater than cpos.
+ */
+void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cpos)
+{
+	struct ocfs2_extent_map_item *emi, *n;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_extent_map *em = &oi->ip_extent_map;
+	LIST_HEAD(tmp_list);
+	unsigned int range;
+
+	spin_lock(&oi->ip_lock);
+	list_for_each_entry_safe(emi, n, &em->em_list, ei_list) {
+		if (emi->ei_cpos >= cpos) {
+			/* Full truncate of this record. */
+			list_move(&emi->ei_list, &tmp_list);
+			BUG_ON(em->em_num_items == 0);
+			em->em_num_items--;
+			continue;
+		}
+
+		range = emi->ei_cpos + emi->ei_clusters;
+		if (range > cpos) {
+			/* Partial truncate */
+			emi->ei_clusters = cpos - emi->ei_cpos;
+		}
+	}
+	spin_unlock(&oi->ip_lock);
+
+	list_for_each_entry_safe(emi, n, &tmp_list, ei_list) {
+		list_del(&emi->ei_list);
+		kfree(emi);
+	}
+}
+
+/*
+ * Is any part of emi2 contained within emi1
+ */
+static int ocfs2_ei_is_contained(struct ocfs2_extent_map_item *emi1,
+				 struct ocfs2_extent_map_item *emi2)
+{
+	unsigned int range1, range2;
+
+	/*
+	 * Check if logical start of emi2 is inside emi1
+	 */
+	range1 = emi1->ei_cpos + emi1->ei_clusters;
+	if (emi2->ei_cpos >= emi1->ei_cpos && emi2->ei_cpos < range1)
+		return 1;
+
+	/*
+	 * Check if logical end of emi2 is inside emi1
+	 */
+	range2 = emi2->ei_cpos + emi2->ei_clusters;
+	if (range2 > emi1->ei_cpos && range2 <= range1)
+		return 1;
+
+	return 0;
+}
+
+static void ocfs2_copy_emi_fields(struct ocfs2_extent_map_item *dest,
+				  struct ocfs2_extent_map_item *src)
+{
+	dest->ei_cpos = src->ei_cpos;
+	dest->ei_phys = src->ei_phys;
+	dest->ei_clusters = src->ei_clusters;
+	dest->ei_flags = src->ei_flags;
+}
+
+/*
+ * Try to merge emi with ins. Returns 1 if merge succeeds, zero
+ * otherwise.
+ */
+static int ocfs2_try_to_merge_extent_map(struct ocfs2_extent_map_item *emi,
+					 struct ocfs2_extent_map_item *ins)
+{
+	/*
+	 * Handle contiguousness
+	 */
+	if (ins->ei_phys == (emi->ei_phys + emi->ei_clusters) &&
+	    ins->ei_cpos == (emi->ei_cpos + emi->ei_clusters) &&
+	    ins->ei_flags == emi->ei_flags) {
+		emi->ei_clusters += ins->ei_clusters;
+		return 1;
+	} else if ((ins->ei_phys + ins->ei_clusters) == emi->ei_phys &&
+		   (ins->ei_cpos + ins->ei_clusters) == emi->ei_cpos &&
+		   ins->ei_flags == emi->ei_flags) {
+		emi->ei_phys = ins->ei_phys;
+		emi->ei_cpos = ins->ei_cpos;
+		emi->ei_clusters += ins->ei_clusters;
+		return 1;
+	}
+
+	/*
+	 * Overlapping extents - this shouldn't happen unless we've
+	 * split an extent to change it's flags. That is exceedingly
+	 * rare, so there's no sense in trying to optimize it yet.
+	 */
+	if (ocfs2_ei_is_contained(emi, ins) ||
+	    ocfs2_ei_is_contained(ins, emi)) {
+		ocfs2_copy_emi_fields(emi, ins);
+		return 1;
+	}
+
+	/* No merge was possible. */
+	return 0;
+}
+
+/*
+ * In order to reduce complexity on the caller, this insert function
+ * is intentionally liberal in what it will accept.
+ *
+ * The only rule is that the truncate call *must* be used whenever
+ * records have been deleted. This avoids inserting overlapping
+ * records with different physical mappings.
+ */
+void ocfs2_extent_map_insert_rec(struct inode *inode,
+				 struct ocfs2_extent_rec *rec)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_extent_map *em = &oi->ip_extent_map;
+	struct ocfs2_extent_map_item *emi, *new_emi = NULL;
+	struct ocfs2_extent_map_item ins;
+
+	ins.ei_cpos = le32_to_cpu(rec->e_cpos);
+	ins.ei_phys = ocfs2_blocks_to_clusters(inode->i_sb,
+					       le64_to_cpu(rec->e_blkno));
+	ins.ei_clusters = le16_to_cpu(rec->e_leaf_clusters);
+	ins.ei_flags = rec->e_flags;
+
+search:
+	spin_lock(&oi->ip_lock);
+
+	list_for_each_entry(emi, &em->em_list, ei_list) {
+		if (ocfs2_try_to_merge_extent_map(emi, &ins)) {
+			list_move(&emi->ei_list, &em->em_list);
+			spin_unlock(&oi->ip_lock);
+			goto out;
+		}
+	}
+
+	/*
+	 * No item could be merged.
+	 *
+	 * Either allocate and add a new item, or overwrite the last recently
+	 * inserted.
+	 */
+
+	if (em->em_num_items < OCFS2_MAX_EXTENT_MAP_ITEMS) {
+		if (new_emi == NULL) {
+			spin_unlock(&oi->ip_lock);
+
+			new_emi = kmalloc(sizeof(*new_emi), GFP_NOFS);
+			if (new_emi == NULL)
+				goto out;
+
+			goto search;
+		}
+
+		ocfs2_copy_emi_fields(new_emi, &ins);
+		list_add(&new_emi->ei_list, &em->em_list);
+		em->em_num_items++;
+		new_emi = NULL;
+	} else {
+		BUG_ON(list_empty(&em->em_list) || em->em_num_items == 0);
+		emi = list_entry(em->em_list.prev,
+				 struct ocfs2_extent_map_item, ei_list);
+		list_move(&emi->ei_list, &em->em_list);
+		ocfs2_copy_emi_fields(emi, &ins);
+	}
+
+	spin_unlock(&oi->ip_lock);
+
+out:
+	kfree(new_emi);
+}
+
+static int ocfs2_last_eb_is_empty(struct inode *inode,
+				  struct ocfs2_dinode *di)
+{
+	int ret, next_free;
+	u64 last_eb_blk = le64_to_cpu(di->i_last_eb_blk);
+	struct buffer_head *eb_bh = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_list *el;
+
+	ret = ocfs2_read_extent_block(INODE_CACHE(inode), last_eb_blk, &eb_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+	el = &eb->h_list;
+
+	if (el->l_tree_depth) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %lu has non zero tree depth in "
+			    "leaf block %llu\n", inode->i_ino,
+			    (unsigned long long)eb_bh->b_blocknr);
+		ret = -EROFS;
+		goto out;
+	}
+
+	next_free = le16_to_cpu(el->l_next_free_rec);
+
+	if (next_free == 0 ||
+	    (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0])))
+		ret = 1;
+
+out:
+	brelse(eb_bh);
+	return ret;
+}
+
+/*
+ * Return the 1st index within el which contains an extent start
+ * larger than v_cluster.
+ */
+static int ocfs2_search_for_hole_index(struct ocfs2_extent_list *el,
+				       u32 v_cluster)
+{
+	int i;
+	struct ocfs2_extent_rec *rec;
+
+	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		rec = &el->l_recs[i];
+
+		if (v_cluster < le32_to_cpu(rec->e_cpos))
+			break;
+	}
+
+	return i;
+}
+
+/*
+ * Figure out the size of a hole which starts at v_cluster within the given
+ * extent list.
+ *
+ * If there is no more allocation past v_cluster, we return the maximum
+ * cluster size minus v_cluster.
+ *
+ * If we have in-inode extents, then el points to the dinode list and
+ * eb_bh is NULL. Otherwise, eb_bh should point to the extent block
+ * containing el.
+ */
+int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
+			       struct ocfs2_extent_list *el,
+			       struct buffer_head *eb_bh,
+			       u32 v_cluster,
+			       u32 *num_clusters)
+{
+	int ret, i;
+	struct buffer_head *next_eb_bh = NULL;
+	struct ocfs2_extent_block *eb, *next_eb;
+
+	i = ocfs2_search_for_hole_index(el, v_cluster);
+
+	if (i == le16_to_cpu(el->l_next_free_rec) && eb_bh) {
+		eb = (struct ocfs2_extent_block *)eb_bh->b_data;
+
+		/*
+		 * Check the next leaf for any extents.
+		 */
+
+		if (le64_to_cpu(eb->h_next_leaf_blk) == 0ULL)
+			goto no_more_extents;
+
+		ret = ocfs2_read_extent_block(ci,
+					      le64_to_cpu(eb->h_next_leaf_blk),
+					      &next_eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		next_eb = (struct ocfs2_extent_block *)next_eb_bh->b_data;
+		el = &next_eb->h_list;
+		i = ocfs2_search_for_hole_index(el, v_cluster);
+	}
+
+no_more_extents:
+	if (i == le16_to_cpu(el->l_next_free_rec)) {
+		/*
+		 * We're at the end of our existing allocation. Just
+		 * return the maximum number of clusters we could
+		 * possibly allocate.
+		 */
+		*num_clusters = UINT_MAX - v_cluster;
+	} else {
+		*num_clusters = le32_to_cpu(el->l_recs[i].e_cpos) - v_cluster;
+	}
+
+	ret = 0;
+out:
+	brelse(next_eb_bh);
+	return ret;
+}
+
+static int ocfs2_get_clusters_nocache(struct inode *inode,
+				      struct buffer_head *di_bh,
+				      u32 v_cluster, unsigned int *hole_len,
+				      struct ocfs2_extent_rec *ret_rec,
+				      unsigned int *is_last)
+{
+	int i, ret, tree_height, len;
+	struct ocfs2_dinode *di;
+	struct ocfs2_extent_block *uninitialized_var(eb);
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+	struct buffer_head *eb_bh = NULL;
+
+	memset(ret_rec, 0, sizeof(*ret_rec));
+	if (is_last)
+		*is_last = 0;
+
+	di = (struct ocfs2_dinode *) di_bh->b_data;
+	el = &di->id2.i_list;
+	tree_height = le16_to_cpu(el->l_tree_depth);
+
+	if (tree_height > 0) {
+		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
+				      &eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+
+		if (el->l_tree_depth) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %lu has non zero tree depth in "
+				    "leaf block %llu\n", inode->i_ino,
+				    (unsigned long long)eb_bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
+		}
+	}
+
+	i = ocfs2_search_extent_list(el, v_cluster);
+	if (i == -1) {
+		/*
+		 * Holes can be larger than the maximum size of an
+		 * extent, so we return their lengths in a separate
+		 * field.
+		 */
+		if (hole_len) {
+			ret = ocfs2_figure_hole_clusters(INODE_CACHE(inode),
+							 el, eb_bh,
+							 v_cluster, &len);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			*hole_len = len;
+		}
+		goto out_hole;
+	}
+
+	rec = &el->l_recs[i];
+
+	BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
+
+	if (!rec->e_blkno) {
+		ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
+			    "record (%u, %u, 0)", inode->i_ino,
+			    le32_to_cpu(rec->e_cpos),
+			    ocfs2_rec_clusters(el, rec));
+		ret = -EROFS;
+		goto out;
+	}
+
+	*ret_rec = *rec;
+
+	/*
+	 * Checking for last extent is potentially expensive - we
+	 * might have to look at the next leaf over to see if it's
+	 * empty.
+	 *
+	 * The first two checks are to see whether the caller even
+	 * cares for this information, and if the extent is at least
+	 * the last in it's list.
+	 *
+	 * If those hold true, then the extent is last if any of the
+	 * additional conditions hold true:
+	 *  - Extent list is in-inode
+	 *  - Extent list is right-most
+	 *  - Extent list is 2nd to rightmost, with empty right-most
+	 */
+	if (is_last) {
+		if (i == (le16_to_cpu(el->l_next_free_rec) - 1)) {
+			if (tree_height == 0)
+				*is_last = 1;
+			else if (eb->h_blkno == di->i_last_eb_blk)
+				*is_last = 1;
+			else if (eb->h_next_leaf_blk == di->i_last_eb_blk) {
+				ret = ocfs2_last_eb_is_empty(inode, di);
+				if (ret < 0) {
+					mlog_errno(ret);
+					goto out;
+				}
+				if (ret == 1)
+					*is_last = 1;
+			}
+		}
+	}
+
+out_hole:
+	ret = 0;
+out:
+	brelse(eb_bh);
+	return ret;
+}
+
+static void ocfs2_relative_extent_offsets(struct super_block *sb,
+					  u32 v_cluster,
+					  struct ocfs2_extent_rec *rec,
+					  u32 *p_cluster, u32 *num_clusters)
+
+{
+	u32 coff = v_cluster - le32_to_cpu(rec->e_cpos);
+
+	*p_cluster = ocfs2_blocks_to_clusters(sb, le64_to_cpu(rec->e_blkno));
+	*p_cluster = *p_cluster + coff;
+
+	if (num_clusters)
+		*num_clusters = le16_to_cpu(rec->e_leaf_clusters) - coff;
+}
+
+int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
+			     u32 *p_cluster, u32 *num_clusters,
+			     struct ocfs2_extent_list *el,
+			     unsigned int *extent_flags)
+{
+	int ret = 0, i;
+	struct buffer_head *eb_bh = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_rec *rec;
+	u32 coff;
+
+	if (el->l_tree_depth) {
+		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, v_cluster,
+				      &eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+
+		if (el->l_tree_depth) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %lu has non zero tree depth in "
+				    "xattr leaf block %llu\n", inode->i_ino,
+				    (unsigned long long)eb_bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
+		}
+	}
+
+	i = ocfs2_search_extent_list(el, v_cluster);
+	if (i == -1) {
+		ret = -EROFS;
+		mlog_errno(ret);
+		goto out;
+	} else {
+		rec = &el->l_recs[i];
+		BUG_ON(v_cluster < le32_to_cpu(rec->e_cpos));
+
+		if (!rec->e_blkno) {
+			ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
+				    "record (%u, %u, 0) in xattr", inode->i_ino,
+				    le32_to_cpu(rec->e_cpos),
+				    ocfs2_rec_clusters(el, rec));
+			ret = -EROFS;
+			goto out;
+		}
+		coff = v_cluster - le32_to_cpu(rec->e_cpos);
+		*p_cluster = ocfs2_blocks_to_clusters(inode->i_sb,
+						    le64_to_cpu(rec->e_blkno));
+		*p_cluster = *p_cluster + coff;
+		if (num_clusters)
+			*num_clusters = ocfs2_rec_clusters(el, rec) - coff;
+
+		if (extent_flags)
+			*extent_flags = rec->e_flags;
+	}
+out:
+	if (eb_bh)
+		brelse(eb_bh);
+	return ret;
+}
+
+int ocfs2_get_clusters(struct inode *inode, u32 v_cluster,
+		       u32 *p_cluster, u32 *num_clusters,
+		       unsigned int *extent_flags)
+{
+	int ret;
+	unsigned int uninitialized_var(hole_len), flags = 0;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_extent_rec rec;
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		ret = -ERANGE;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_extent_map_lookup(inode, v_cluster, p_cluster,
+				      num_clusters, extent_flags);
+	if (ret == 0)
+		goto out;
+
+	ret = ocfs2_read_inode_block(inode, &di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_get_clusters_nocache(inode, di_bh, v_cluster, &hole_len,
+					 &rec, NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (rec.e_blkno == 0ULL) {
+		/*
+		 * A hole was found. Return some canned values that
+		 * callers can key on. If asked for, num_clusters will
+		 * be populated with the size of the hole.
+		 */
+		*p_cluster = 0;
+		if (num_clusters) {
+			*num_clusters = hole_len;
+		}
+	} else {
+		ocfs2_relative_extent_offsets(inode->i_sb, v_cluster, &rec,
+					      p_cluster, num_clusters);
+		flags = rec.e_flags;
+
+		ocfs2_extent_map_insert_rec(inode, &rec);
+	}
+
+	if (extent_flags)
+		*extent_flags = flags;
+
+out:
+	brelse(di_bh);
+	return ret;
+}
+
+/*
+ * This expects alloc_sem to be held. The allocation cannot change at
+ * all while the map is in the process of being updated.
+ */
+int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
+				u64 *ret_count, unsigned int *extent_flags)
+{
+	int ret;
+	int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
+	u32 cpos, num_clusters, p_cluster;
+	u64 boff = 0;
+
+	cpos = ocfs2_blocks_to_clusters(inode->i_sb, v_blkno);
+
+	ret = ocfs2_get_clusters(inode, cpos, &p_cluster, &num_clusters,
+				 extent_flags);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * p_cluster == 0 indicates a hole.
+	 */
+	if (p_cluster) {
+		boff = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
+		boff += (v_blkno & (u64)(bpc - 1));
+	}
+
+	*p_blkno = boff;
+
+	if (ret_count) {
+		*ret_count = ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
+		*ret_count -= v_blkno & (u64)(bpc - 1);
+	}
+
+out:
+	return ret;
+}
+
+/*
+ * The ocfs2_fiemap_inline() may be a little bit misleading, since
+ * it not only handles the fiemap for inlined files, but also deals
+ * with the fast symlink, cause they have no difference for extent
+ * mapping per se.
+ */
+static int ocfs2_fiemap_inline(struct inode *inode, struct buffer_head *di_bh,
+			       struct fiemap_extent_info *fieinfo,
+			       u64 map_start)
+{
+	int ret;
+	unsigned int id_count;
+	struct ocfs2_dinode *di;
+	u64 phys;
+	u32 flags = FIEMAP_EXTENT_DATA_INLINE|FIEMAP_EXTENT_LAST;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	if (ocfs2_inode_is_fast_symlink(inode))
+		id_count = ocfs2_fast_symlink_chars(inode->i_sb);
+	else
+		id_count = le16_to_cpu(di->id2.i_data.id_count);
+
+	if (map_start < id_count) {
+		phys = oi->ip_blkno << inode->i_sb->s_blocksize_bits;
+		if (ocfs2_inode_is_fast_symlink(inode))
+			phys += offsetof(struct ocfs2_dinode, id2.i_symlink);
+		else
+			phys += offsetof(struct ocfs2_dinode,
+					 id2.i_data.id_data);
+
+		ret = fiemap_fill_next_extent(fieinfo, 0, phys, id_count,
+					      flags);
+		if (ret < 0)
+			return ret;
+	}
+
+	return 0;
+}
+
+#define OCFS2_FIEMAP_FLAGS	(FIEMAP_FLAG_SYNC)
+
+int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
+		 u64 map_start, u64 map_len)
+{
+	int ret, is_last;
+	u32 mapping_end, cpos;
+	unsigned int hole_size;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	u64 len_bytes, phys_bytes, virt_bytes;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_extent_rec rec;
+
+	ret = fiemap_check_flags(fieinfo, OCFS2_FIEMAP_FLAGS);
+	if (ret)
+		return ret;
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 0);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	down_read(&OCFS2_I(inode)->ip_alloc_sem);
+
+	/*
+	 * Handle inline-data and fast symlink separately.
+	 */
+	if ((OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
+	    ocfs2_inode_is_fast_symlink(inode)) {
+		ret = ocfs2_fiemap_inline(inode, di_bh, fieinfo, map_start);
+		goto out_unlock;
+	}
+
+	cpos = map_start >> osb->s_clustersize_bits;
+	mapping_end = ocfs2_clusters_for_bytes(inode->i_sb,
+					       map_start + map_len);
+	is_last = 0;
+	while (cpos < mapping_end && !is_last) {
+		u32 fe_flags;
+
+		ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos,
+						 &hole_size, &rec, &is_last);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_unlock;
+		}
+
+		if (rec.e_blkno == 0ULL) {
+			cpos += hole_size;
+			continue;
+		}
+
+		fe_flags = 0;
+		if (rec.e_flags & OCFS2_EXT_UNWRITTEN)
+			fe_flags |= FIEMAP_EXTENT_UNWRITTEN;
+		if (rec.e_flags & OCFS2_EXT_REFCOUNTED)
+			fe_flags |= FIEMAP_EXTENT_SHARED;
+		if (is_last)
+			fe_flags |= FIEMAP_EXTENT_LAST;
+		len_bytes = (u64)le16_to_cpu(rec.e_leaf_clusters) << osb->s_clustersize_bits;
+		phys_bytes = le64_to_cpu(rec.e_blkno) << osb->sb->s_blocksize_bits;
+		virt_bytes = (u64)le32_to_cpu(rec.e_cpos) << osb->s_clustersize_bits;
+
+		ret = fiemap_fill_next_extent(fieinfo, virt_bytes, phys_bytes,
+					      len_bytes, fe_flags);
+		if (ret)
+			break;
+
+		cpos = le32_to_cpu(rec.e_cpos)+ le16_to_cpu(rec.e_leaf_clusters);
+	}
+
+	if (ret > 0)
+		ret = 0;
+
+out_unlock:
+	brelse(di_bh);
+
+	up_read(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ocfs2_inode_unlock(inode, 0);
+out:
+
+	return ret;
+}
+
+int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int whence)
+{
+	struct inode *inode = file->f_mapping->host;
+	int ret;
+	unsigned int is_last = 0, is_data = 0;
+	u16 cs_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+	u32 cpos, cend, clen, hole_size;
+	u64 extoff, extlen;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_extent_rec rec;
+
+	BUG_ON(whence != SEEK_DATA && whence != SEEK_HOLE);
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 0);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	down_read(&OCFS2_I(inode)->ip_alloc_sem);
+
+	if (*offset >= i_size_read(inode)) {
+		ret = -ENXIO;
+		goto out_unlock;
+	}
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		if (whence == SEEK_HOLE)
+			*offset = i_size_read(inode);
+		goto out_unlock;
+	}
+
+	clen = 0;
+	cpos = *offset >> cs_bits;
+	cend = ocfs2_clusters_for_bytes(inode->i_sb, i_size_read(inode));
+
+	while (cpos < cend && !is_last) {
+		ret = ocfs2_get_clusters_nocache(inode, di_bh, cpos, &hole_size,
+						 &rec, &is_last);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_unlock;
+		}
+
+		extoff = cpos;
+		extoff <<= cs_bits;
+
+		if (rec.e_blkno == 0ULL) {
+			clen = hole_size;
+			is_data = 0;
+		} else {
+			clen = le16_to_cpu(rec.e_leaf_clusters) -
+				(cpos - le32_to_cpu(rec.e_cpos));
+			is_data = (rec.e_flags & OCFS2_EXT_UNWRITTEN) ?  0 : 1;
+		}
+
+		if ((!is_data && whence == SEEK_HOLE) ||
+		    (is_data && whence == SEEK_DATA)) {
+			if (extoff > *offset)
+				*offset = extoff;
+			goto out_unlock;
+		}
+
+		if (!is_last)
+			cpos += clen;
+	}
+
+	if (whence == SEEK_HOLE) {
+		extoff = cpos;
+		extoff <<= cs_bits;
+		extlen = clen;
+		extlen <<=  cs_bits;
+
+		if ((extoff + extlen) > i_size_read(inode))
+			extlen = i_size_read(inode) - extoff;
+		extoff += extlen;
+		if (extoff > *offset)
+			*offset = extoff;
+		goto out_unlock;
+	}
+
+	ret = -ENXIO;
+
+out_unlock:
+
+	brelse(di_bh);
+
+	up_read(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ocfs2_inode_unlock(inode, 0);
+out:
+	return ret;
+}
+
+int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
+			   struct buffer_head *bhs[], int flags,
+			   int (*validate)(struct super_block *sb,
+					   struct buffer_head *bh))
+{
+	int rc = 0;
+	u64 p_block, p_count;
+	int i, count, done = 0;
+
+	trace_ocfs2_read_virt_blocks(
+	     inode, (unsigned long long)v_block, nr, bhs, flags,
+	     validate);
+
+	if (((v_block + nr - 1) << inode->i_sb->s_blocksize_bits) >=
+	    i_size_read(inode)) {
+		BUG_ON(!(flags & OCFS2_BH_READAHEAD));
+		goto out;
+	}
+
+	while (done < nr) {
+		down_read(&OCFS2_I(inode)->ip_alloc_sem);
+		rc = ocfs2_extent_map_get_blocks(inode, v_block + done,
+						 &p_block, &p_count, NULL);
+		up_read(&OCFS2_I(inode)->ip_alloc_sem);
+		if (rc) {
+			mlog_errno(rc);
+			break;
+		}
+
+		if (!p_block) {
+			rc = -EIO;
+			mlog(ML_ERROR,
+			     "Inode #%llu contains a hole at offset %llu\n",
+			     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			     (unsigned long long)(v_block + done) <<
+			     inode->i_sb->s_blocksize_bits);
+			break;
+		}
+
+		count = nr - done;
+		if (p_count < count)
+			count = p_count;
+
+		/*
+		 * If the caller passed us bhs, they should have come
+		 * from a previous readahead call to this function.  Thus,
+		 * they should have the right b_blocknr.
+		 */
+		for (i = 0; i < count; i++) {
+			if (!bhs[done + i])
+				continue;
+			BUG_ON(bhs[done + i]->b_blocknr != (p_block + i));
+		}
+
+		rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, count,
+				       bhs + done, flags, validate);
+		if (rc) {
+			mlog_errno(rc);
+			break;
+		}
+		done += count;
+	}
+
+out:
+	return rc;
+}
+
+
diff --git a/fs/ocfs2/extent_map.h b/fs/ocfs2/extent_map.h
new file mode 100644
index 000000000..67ea57d2f
--- /dev/null
+++ b/fs/ocfs2/extent_map.h
@@ -0,0 +1,92 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * extent_map.h
+ *
+ * In-memory file extent mappings for OCFS2.
+ *
+ * Copyright (C) 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License, version 2,  as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef _EXTENT_MAP_H
+#define _EXTENT_MAP_H
+
+struct ocfs2_extent_map_item {
+	unsigned int			ei_cpos;
+	unsigned int			ei_phys;
+	unsigned int			ei_clusters;
+	unsigned int			ei_flags;
+
+	struct list_head		ei_list;
+};
+
+#define OCFS2_MAX_EXTENT_MAP_ITEMS			3
+struct ocfs2_extent_map {
+	unsigned int			em_num_items;
+	struct list_head		em_list;
+};
+
+void ocfs2_extent_map_init(struct inode *inode);
+void ocfs2_extent_map_trunc(struct inode *inode, unsigned int cluster);
+void ocfs2_extent_map_insert_rec(struct inode *inode,
+				 struct ocfs2_extent_rec *rec);
+
+int ocfs2_get_clusters(struct inode *inode, u32 v_cluster, u32 *p_cluster,
+		       u32 *num_clusters, unsigned int *extent_flags);
+int ocfs2_extent_map_get_blocks(struct inode *inode, u64 v_blkno, u64 *p_blkno,
+				u64 *ret_count, unsigned int *extent_flags);
+
+int ocfs2_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
+		 u64 map_start, u64 map_len);
+
+int ocfs2_seek_data_hole_offset(struct file *file, loff_t *offset, int origin);
+
+int ocfs2_xattr_get_clusters(struct inode *inode, u32 v_cluster,
+			     u32 *p_cluster, u32 *num_clusters,
+			     struct ocfs2_extent_list *el,
+			     unsigned int *extent_flags);
+
+int ocfs2_read_virt_blocks(struct inode *inode, u64 v_block, int nr,
+			   struct buffer_head *bhs[], int flags,
+			   int (*validate)(struct super_block *sb,
+					   struct buffer_head *bh));
+int ocfs2_figure_hole_clusters(struct ocfs2_caching_info *ci,
+			       struct ocfs2_extent_list *el,
+			       struct buffer_head *eb_bh,
+			       u32 v_cluster,
+			       u32 *num_clusters);
+static inline int ocfs2_read_virt_block(struct inode *inode, u64 v_block,
+					struct buffer_head **bh,
+					int (*validate)(struct super_block *sb,
+							struct buffer_head *bh))
+{
+	int status = 0;
+
+	if (bh == NULL) {
+		printk("ocfs2: bh == NULL\n");
+		status = -EINVAL;
+		goto bail;
+	}
+
+	status = ocfs2_read_virt_blocks(inode, v_block, 1, bh, 0, validate);
+
+bail:
+	return status;
+}
+
+
+#endif  /* _EXTENT_MAP_H */
diff --git a/fs/ocfs2/file.c b/fs/ocfs2/file.c
new file mode 100644
index 000000000..d8b670cbd
--- /dev/null
+++ b/fs/ocfs2/file.c
@@ -0,0 +1,2702 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * file.c
+ *
+ * File open, close, extend, truncate
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/capability.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/pagemap.h>
+#include <linux/uio.h>
+#include <linux/sched.h>
+#include <linux/splice.h>
+#include <linux/mount.h>
+#include <linux/writeback.h>
+#include <linux/falloc.h>
+#include <linux/quotaops.h>
+#include <linux/blkdev.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "aops.h"
+#include "dir.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "file.h"
+#include "sysfile.h"
+#include "inode.h"
+#include "ioctl.h"
+#include "journal.h"
+#include "locks.h"
+#include "mmap.h"
+#include "suballoc.h"
+#include "super.h"
+#include "xattr.h"
+#include "acl.h"
+#include "quota.h"
+#include "refcounttree.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+static int ocfs2_init_file_private(struct inode *inode, struct file *file)
+{
+	struct ocfs2_file_private *fp;
+
+	fp = kzalloc(sizeof(struct ocfs2_file_private), GFP_KERNEL);
+	if (!fp)
+		return -ENOMEM;
+
+	fp->fp_file = file;
+	mutex_init(&fp->fp_mutex);
+	ocfs2_file_lock_res_init(&fp->fp_flock, fp);
+	file->private_data = fp;
+
+	return 0;
+}
+
+static void ocfs2_free_file_private(struct inode *inode, struct file *file)
+{
+	struct ocfs2_file_private *fp = file->private_data;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (fp) {
+		ocfs2_simple_drop_lockres(osb, &fp->fp_flock);
+		ocfs2_lock_res_free(&fp->fp_flock);
+		kfree(fp);
+		file->private_data = NULL;
+	}
+}
+
+static int ocfs2_file_open(struct inode *inode, struct file *file)
+{
+	int status;
+	int mode = file->f_flags;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	trace_ocfs2_file_open(inode, file, file->f_path.dentry,
+			      (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			      file->f_path.dentry->d_name.len,
+			      file->f_path.dentry->d_name.name, mode);
+
+	if (file->f_mode & FMODE_WRITE)
+		dquot_initialize(inode);
+
+	spin_lock(&oi->ip_lock);
+
+	/* Check that the inode hasn't been wiped from disk by another
+	 * node. If it hasn't then we're safe as long as we hold the
+	 * spin lock until our increment of open count. */
+	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
+		spin_unlock(&oi->ip_lock);
+
+		status = -ENOENT;
+		goto leave;
+	}
+
+	if (mode & O_DIRECT)
+		oi->ip_flags |= OCFS2_INODE_OPEN_DIRECT;
+
+	oi->ip_open_count++;
+	spin_unlock(&oi->ip_lock);
+
+	status = ocfs2_init_file_private(inode, file);
+	if (status) {
+		/*
+		 * We want to set open count back if we're failing the
+		 * open.
+		 */
+		spin_lock(&oi->ip_lock);
+		oi->ip_open_count--;
+		spin_unlock(&oi->ip_lock);
+	}
+
+leave:
+	return status;
+}
+
+static int ocfs2_file_release(struct inode *inode, struct file *file)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	spin_lock(&oi->ip_lock);
+	if (!--oi->ip_open_count)
+		oi->ip_flags &= ~OCFS2_INODE_OPEN_DIRECT;
+
+	trace_ocfs2_file_release(inode, file, file->f_path.dentry,
+				 oi->ip_blkno,
+				 file->f_path.dentry->d_name.len,
+				 file->f_path.dentry->d_name.name,
+				 oi->ip_open_count);
+	spin_unlock(&oi->ip_lock);
+
+	ocfs2_free_file_private(inode, file);
+
+	return 0;
+}
+
+static int ocfs2_dir_open(struct inode *inode, struct file *file)
+{
+	return ocfs2_init_file_private(inode, file);
+}
+
+static int ocfs2_dir_release(struct inode *inode, struct file *file)
+{
+	ocfs2_free_file_private(inode, file);
+	return 0;
+}
+
+static int ocfs2_sync_file(struct file *file, loff_t start, loff_t end,
+			   int datasync)
+{
+	int err = 0;
+	struct inode *inode = file->f_mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	journal_t *journal = osb->journal->j_journal;
+	int ret;
+	tid_t commit_tid;
+	bool needs_barrier = false;
+
+	trace_ocfs2_sync_file(inode, file, file->f_path.dentry,
+			      OCFS2_I(inode)->ip_blkno,
+			      file->f_path.dentry->d_name.len,
+			      file->f_path.dentry->d_name.name,
+			      (unsigned long long)datasync);
+
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
+		return -EROFS;
+
+	err = filemap_write_and_wait_range(inode->i_mapping, start, end);
+	if (err)
+		return err;
+
+	commit_tid = datasync ? oi->i_datasync_tid : oi->i_sync_tid;
+	if (journal->j_flags & JBD2_BARRIER &&
+	    !jbd2_trans_will_send_data_barrier(journal, commit_tid))
+		needs_barrier = true;
+	err = jbd2_complete_transaction(journal, commit_tid);
+	if (needs_barrier) {
+		ret = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
+		if (!err)
+			err = ret;
+	}
+
+	if (err)
+		mlog_errno(err);
+
+	return (err < 0) ? -EIO : 0;
+}
+
+int ocfs2_should_update_atime(struct inode *inode,
+			      struct vfsmount *vfsmnt)
+{
+	struct timespec now;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
+		return 0;
+
+	if ((inode->i_flags & S_NOATIME) ||
+	    ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode)))
+		return 0;
+
+	/*
+	 * We can be called with no vfsmnt structure - NFSD will
+	 * sometimes do this.
+	 *
+	 * Note that our action here is different than touch_atime() -
+	 * if we can't tell whether this is a noatime mount, then we
+	 * don't know whether to trust the value of s_atime_quantum.
+	 */
+	if (vfsmnt == NULL)
+		return 0;
+
+	if ((vfsmnt->mnt_flags & MNT_NOATIME) ||
+	    ((vfsmnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
+		return 0;
+
+	if (vfsmnt->mnt_flags & MNT_RELATIME) {
+		if ((timespec_compare(&inode->i_atime, &inode->i_mtime) <= 0) ||
+		    (timespec_compare(&inode->i_atime, &inode->i_ctime) <= 0))
+			return 1;
+
+		return 0;
+	}
+
+	now = CURRENT_TIME;
+	if ((now.tv_sec - inode->i_atime.tv_sec <= osb->s_atime_quantum))
+		return 0;
+	else
+		return 1;
+}
+
+int ocfs2_update_inode_atime(struct inode *inode,
+			     struct buffer_head *bh)
+{
+	int ret;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	handle_t *handle;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *) bh->b_data;
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	/*
+	 * Don't use ocfs2_mark_inode_dirty() here as we don't always
+	 * have i_mutex to guard against concurrent changes to other
+	 * inode fields.
+	 */
+	inode->i_atime = CURRENT_TIME;
+	di->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
+	di->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
+	ocfs2_update_inode_fsync_trans(handle, inode, 0);
+	ocfs2_journal_dirty(handle, bh);
+
+out_commit:
+	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
+out:
+	return ret;
+}
+
+int ocfs2_set_inode_size(handle_t *handle,
+				struct inode *inode,
+				struct buffer_head *fe_bh,
+				u64 new_i_size)
+{
+	int status;
+
+	i_size_write(inode, new_i_size);
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
+	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
+
+	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+bail:
+	return status;
+}
+
+int ocfs2_simple_size_update(struct inode *inode,
+			     struct buffer_head *di_bh,
+			     u64 new_i_size)
+{
+	int ret;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	handle_t *handle = NULL;
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_set_inode_size(handle, inode, di_bh,
+				   new_i_size);
+	if (ret < 0)
+		mlog_errno(ret);
+
+	ocfs2_update_inode_fsync_trans(handle, inode, 0);
+	ocfs2_commit_trans(osb, handle);
+out:
+	return ret;
+}
+
+static int ocfs2_cow_file_pos(struct inode *inode,
+			      struct buffer_head *fe_bh,
+			      u64 offset)
+{
+	int status;
+	u32 phys, cpos = offset >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+	unsigned int num_clusters = 0;
+	unsigned int ext_flags = 0;
+
+	/*
+	 * If the new offset is aligned to the range of the cluster, there is
+	 * no space for ocfs2_zero_range_for_truncate to fill, so no need to
+	 * CoW either.
+	 */
+	if ((offset & (OCFS2_SB(inode->i_sb)->s_clustersize - 1)) == 0)
+		return 0;
+
+	status = ocfs2_get_clusters(inode, cpos, &phys,
+				    &num_clusters, &ext_flags);
+	if (status) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
+		goto out;
+
+	return ocfs2_refcount_cow(inode, fe_bh, cpos, 1, cpos+1);
+
+out:
+	return status;
+}
+
+static int ocfs2_orphan_for_truncate(struct ocfs2_super *osb,
+				     struct inode *inode,
+				     struct buffer_head *fe_bh,
+				     u64 new_i_size)
+{
+	int status;
+	handle_t *handle;
+	struct ocfs2_dinode *di;
+	u64 cluster_bytes;
+
+	/*
+	 * We need to CoW the cluster contains the offset if it is reflinked
+	 * since we will call ocfs2_zero_range_for_truncate later which will
+	 * write "0" from offset to the end of the cluster.
+	 */
+	status = ocfs2_cow_file_pos(inode, fe_bh, new_i_size);
+	if (status) {
+		mlog_errno(status);
+		return status;
+	}
+
+	/* TODO: This needs to actually orphan the inode in this
+	 * transaction. */
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out;
+	}
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	/*
+	 * Do this before setting i_size.
+	 */
+	cluster_bytes = ocfs2_align_bytes_to_clusters(inode->i_sb, new_i_size);
+	status = ocfs2_zero_range_for_truncate(inode, handle, new_i_size,
+					       cluster_bytes);
+	if (status) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	i_size_write(inode, new_i_size);
+	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
+
+	di = (struct ocfs2_dinode *) fe_bh->b_data;
+	di->i_size = cpu_to_le64(new_i_size);
+	di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
+	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+	ocfs2_update_inode_fsync_trans(handle, inode, 0);
+
+	ocfs2_journal_dirty(handle, fe_bh);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	return status;
+}
+
+int ocfs2_truncate_file(struct inode *inode,
+			       struct buffer_head *di_bh,
+			       u64 new_i_size)
+{
+	int status = 0;
+	struct ocfs2_dinode *fe = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	/* We trust di_bh because it comes from ocfs2_inode_lock(), which
+	 * already validated it */
+	fe = (struct ocfs2_dinode *) di_bh->b_data;
+
+	trace_ocfs2_truncate_file((unsigned long long)OCFS2_I(inode)->ip_blkno,
+				  (unsigned long long)le64_to_cpu(fe->i_size),
+				  (unsigned long long)new_i_size);
+
+	mlog_bug_on_msg(le64_to_cpu(fe->i_size) != i_size_read(inode),
+			"Inode %llu, inode i_size = %lld != di "
+			"i_size = %llu, i_flags = 0x%x\n",
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			i_size_read(inode),
+			(unsigned long long)le64_to_cpu(fe->i_size),
+			le32_to_cpu(fe->i_flags));
+
+	if (new_i_size > le64_to_cpu(fe->i_size)) {
+		trace_ocfs2_truncate_file_error(
+			(unsigned long long)le64_to_cpu(fe->i_size),
+			(unsigned long long)new_i_size);
+		status = -EINVAL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ocfs2_resv_discard(&osb->osb_la_resmap,
+			   &OCFS2_I(inode)->ip_la_data_resv);
+
+	/*
+	 * The inode lock forced other nodes to sync and drop their
+	 * pages, which (correctly) happens even if we have a truncate
+	 * without allocation change - ocfs2 cluster sizes can be much
+	 * greater than page size, so we have to truncate them
+	 * anyway.
+	 */
+	unmap_mapping_range(inode->i_mapping, new_i_size + PAGE_SIZE - 1, 0, 1);
+	truncate_inode_pages(inode->i_mapping, new_i_size);
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		status = ocfs2_truncate_inline(inode, di_bh, new_i_size,
+					       i_size_read(inode), 1);
+		if (status)
+			mlog_errno(status);
+
+		goto bail_unlock_sem;
+	}
+
+	/* alright, we're going to need to do a full blown alloc size
+	 * change. Orphan the inode so that recovery can complete the
+	 * truncate if necessary. This does the task of marking
+	 * i_size. */
+	status = ocfs2_orphan_for_truncate(osb, inode, di_bh, new_i_size);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_unlock_sem;
+	}
+
+	status = ocfs2_commit_truncate(osb, inode, di_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_unlock_sem;
+	}
+
+	/* TODO: orphan dir cleanup here. */
+bail_unlock_sem:
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+bail:
+	if (!status && OCFS2_I(inode)->ip_clusters == 0)
+		status = ocfs2_try_remove_refcount_tree(inode, di_bh);
+
+	return status;
+}
+
+/*
+ * extend file allocation only here.
+ * we'll update all the disk stuff, and oip->alloc_size
+ *
+ * expect stuff to be locked, a transaction started and enough data /
+ * metadata reservations in the contexts.
+ *
+ * Will return -EAGAIN, and a reason if a restart is needed.
+ * If passed in, *reason will always be set, even in error.
+ */
+int ocfs2_add_inode_data(struct ocfs2_super *osb,
+			 struct inode *inode,
+			 u32 *logical_offset,
+			 u32 clusters_to_add,
+			 int mark_unwritten,
+			 struct buffer_head *fe_bh,
+			 handle_t *handle,
+			 struct ocfs2_alloc_context *data_ac,
+			 struct ocfs2_alloc_context *meta_ac,
+			 enum ocfs2_alloc_restarted *reason_ret)
+{
+	int ret;
+	struct ocfs2_extent_tree et;
+
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), fe_bh);
+	ret = ocfs2_add_clusters_in_btree(handle, &et, logical_offset,
+					  clusters_to_add, mark_unwritten,
+					  data_ac, meta_ac, reason_ret);
+
+	return ret;
+}
+
+static int __ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
+				     u32 clusters_to_add, int mark_unwritten)
+{
+	int status = 0;
+	int restart_func = 0;
+	int credits;
+	u32 prev_clusters;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_dinode *fe = NULL;
+	handle_t *handle = NULL;
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	enum ocfs2_alloc_restarted why = RESTART_NONE;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_extent_tree et;
+	int did_quota = 0;
+
+	/*
+	 * Unwritten extent only exists for file systems which
+	 * support holes.
+	 */
+	BUG_ON(mark_unwritten && !ocfs2_sparse_alloc(osb));
+
+	status = ocfs2_read_inode_block(inode, &bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+	fe = (struct ocfs2_dinode *) bh->b_data;
+
+restart_all:
+	BUG_ON(le32_to_cpu(fe->i_clusters) != OCFS2_I(inode)->ip_clusters);
+
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), bh);
+	status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
+				       &data_ac, &meta_ac);
+	if (status) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	credits = ocfs2_calc_extend_credits(osb->sb, &fe->id2.i_list);
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto leave;
+	}
+
+restarted_transaction:
+	trace_ocfs2_extend_allocation(
+		(unsigned long long)OCFS2_I(inode)->ip_blkno,
+		(unsigned long long)i_size_read(inode),
+		le32_to_cpu(fe->i_clusters), clusters_to_add,
+		why, restart_func);
+
+	status = dquot_alloc_space_nodirty(inode,
+			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
+	if (status)
+		goto leave;
+	did_quota = 1;
+
+	/* reserve a write to the file entry early on - that we if we
+	 * run out of credits in the allocation path, we can still
+	 * update i_size. */
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	prev_clusters = OCFS2_I(inode)->ip_clusters;
+
+	status = ocfs2_add_inode_data(osb,
+				      inode,
+				      &logical_start,
+				      clusters_to_add,
+				      mark_unwritten,
+				      bh,
+				      handle,
+				      data_ac,
+				      meta_ac,
+				      &why);
+	if ((status < 0) && (status != -EAGAIN)) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto leave;
+	}
+	ocfs2_update_inode_fsync_trans(handle, inode, 1);
+	ocfs2_journal_dirty(handle, bh);
+
+	spin_lock(&OCFS2_I(inode)->ip_lock);
+	clusters_to_add -= (OCFS2_I(inode)->ip_clusters - prev_clusters);
+	spin_unlock(&OCFS2_I(inode)->ip_lock);
+	/* Release unused quota reservation */
+	dquot_free_space(inode,
+			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
+	did_quota = 0;
+
+	if (why != RESTART_NONE && clusters_to_add) {
+		if (why == RESTART_META) {
+			restart_func = 1;
+			status = 0;
+		} else {
+			BUG_ON(why != RESTART_TRANS);
+
+			status = ocfs2_allocate_extend_trans(handle, 1);
+			if (status < 0) {
+				/* handle still has to be committed at
+				 * this point. */
+				status = -ENOMEM;
+				mlog_errno(status);
+				goto leave;
+			}
+			goto restarted_transaction;
+		}
+	}
+
+	trace_ocfs2_extend_allocation_end(OCFS2_I(inode)->ip_blkno,
+	     le32_to_cpu(fe->i_clusters),
+	     (unsigned long long)le64_to_cpu(fe->i_size),
+	     OCFS2_I(inode)->ip_clusters,
+	     (unsigned long long)i_size_read(inode));
+
+leave:
+	if (status < 0 && did_quota)
+		dquot_free_space(inode,
+			ocfs2_clusters_to_bytes(osb->sb, clusters_to_add));
+	if (handle) {
+		ocfs2_commit_trans(osb, handle);
+		handle = NULL;
+	}
+	if (data_ac) {
+		ocfs2_free_alloc_context(data_ac);
+		data_ac = NULL;
+	}
+	if (meta_ac) {
+		ocfs2_free_alloc_context(meta_ac);
+		meta_ac = NULL;
+	}
+	if ((!status) && restart_func) {
+		restart_func = 0;
+		goto restart_all;
+	}
+	brelse(bh);
+	bh = NULL;
+
+	return status;
+}
+
+int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
+		u32 clusters_to_add, int mark_unwritten)
+{
+	return __ocfs2_extend_allocation(inode, logical_start,
+			clusters_to_add, mark_unwritten);
+}
+
+/*
+ * While a write will already be ordering the data, a truncate will not.
+ * Thus, we need to explicitly order the zeroed pages.
+ */
+static handle_t *ocfs2_zero_start_ordered_transaction(struct inode *inode,
+						struct buffer_head *di_bh)
+{
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	handle_t *handle = NULL;
+	int ret = 0;
+
+	if (!ocfs2_should_order_data(inode))
+		goto out;
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_jbd2_file_inode(handle, inode);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret)
+		mlog_errno(ret);
+	ocfs2_update_inode_fsync_trans(handle, inode, 1);
+
+out:
+	if (ret) {
+		if (!IS_ERR(handle))
+			ocfs2_commit_trans(osb, handle);
+		handle = ERR_PTR(ret);
+	}
+	return handle;
+}
+
+/* Some parts of this taken from generic_cont_expand, which turned out
+ * to be too fragile to do exactly what we need without us having to
+ * worry about recursive locking in ->write_begin() and ->write_end(). */
+static int ocfs2_write_zero_page(struct inode *inode, u64 abs_from,
+				 u64 abs_to, struct buffer_head *di_bh)
+{
+	struct address_space *mapping = inode->i_mapping;
+	struct page *page;
+	unsigned long index = abs_from >> PAGE_CACHE_SHIFT;
+	handle_t *handle;
+	int ret = 0;
+	unsigned zero_from, zero_to, block_start, block_end;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	BUG_ON(abs_from >= abs_to);
+	BUG_ON(abs_to > (((u64)index + 1) << PAGE_CACHE_SHIFT));
+	BUG_ON(abs_from & (inode->i_blkbits - 1));
+
+	handle = ocfs2_zero_start_ordered_transaction(inode, di_bh);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		goto out;
+	}
+
+	page = find_or_create_page(mapping, index, GFP_NOFS);
+	if (!page) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out_commit_trans;
+	}
+
+	/* Get the offsets within the page that we want to zero */
+	zero_from = abs_from & (PAGE_CACHE_SIZE - 1);
+	zero_to = abs_to & (PAGE_CACHE_SIZE - 1);
+	if (!zero_to)
+		zero_to = PAGE_CACHE_SIZE;
+
+	trace_ocfs2_write_zero_page(
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			(unsigned long long)abs_from,
+			(unsigned long long)abs_to,
+			index, zero_from, zero_to);
+
+	/* We know that zero_from is block aligned */
+	for (block_start = zero_from; block_start < zero_to;
+	     block_start = block_end) {
+		block_end = block_start + (1 << inode->i_blkbits);
+
+		/*
+		 * block_start is block-aligned.  Bump it by one to force
+		 * __block_write_begin and block_commit_write to zero the
+		 * whole block.
+		 */
+		ret = __block_write_begin(page, block_start + 1, 0,
+					  ocfs2_get_block);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out_unlock;
+		}
+
+
+		/* must not update i_size! */
+		ret = block_commit_write(page, block_start + 1,
+					 block_start + 1);
+		if (ret < 0)
+			mlog_errno(ret);
+		else
+			ret = 0;
+	}
+
+	/*
+	 * fs-writeback will release the dirty pages without page lock
+	 * whose offset are over inode size, the release happens at
+	 * block_write_full_page().
+	 */
+	i_size_write(inode, abs_to);
+	inode->i_blocks = ocfs2_inode_sector_count(inode);
+	di->i_size = cpu_to_le64((u64)i_size_read(inode));
+	inode->i_mtime = inode->i_ctime = CURRENT_TIME;
+	di->i_mtime = di->i_ctime = cpu_to_le64(inode->i_mtime.tv_sec);
+	di->i_ctime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
+	di->i_mtime_nsec = di->i_ctime_nsec;
+	if (handle) {
+		ocfs2_journal_dirty(handle, di_bh);
+		ocfs2_update_inode_fsync_trans(handle, inode, 1);
+	}
+
+out_unlock:
+	unlock_page(page);
+	page_cache_release(page);
+out_commit_trans:
+	if (handle)
+		ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
+out:
+	return ret;
+}
+
+/*
+ * Find the next range to zero.  We do this in terms of bytes because
+ * that's what ocfs2_zero_extend() wants, and it is dealing with the
+ * pagecache.  We may return multiple extents.
+ *
+ * zero_start and zero_end are ocfs2_zero_extend()s current idea of what
+ * needs to be zeroed.  range_start and range_end return the next zeroing
+ * range.  A subsequent call should pass the previous range_end as its
+ * zero_start.  If range_end is 0, there's nothing to do.
+ *
+ * Unwritten extents are skipped over.  Refcounted extents are CoWd.
+ */
+static int ocfs2_zero_extend_get_range(struct inode *inode,
+				       struct buffer_head *di_bh,
+				       u64 zero_start, u64 zero_end,
+				       u64 *range_start, u64 *range_end)
+{
+	int rc = 0, needs_cow = 0;
+	u32 p_cpos, zero_clusters = 0;
+	u32 zero_cpos =
+		zero_start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+	u32 last_cpos = ocfs2_clusters_for_bytes(inode->i_sb, zero_end);
+	unsigned int num_clusters = 0;
+	unsigned int ext_flags = 0;
+
+	while (zero_cpos < last_cpos) {
+		rc = ocfs2_get_clusters(inode, zero_cpos, &p_cpos,
+					&num_clusters, &ext_flags);
+		if (rc) {
+			mlog_errno(rc);
+			goto out;
+		}
+
+		if (p_cpos && !(ext_flags & OCFS2_EXT_UNWRITTEN)) {
+			zero_clusters = num_clusters;
+			if (ext_flags & OCFS2_EXT_REFCOUNTED)
+				needs_cow = 1;
+			break;
+		}
+
+		zero_cpos += num_clusters;
+	}
+	if (!zero_clusters) {
+		*range_end = 0;
+		goto out;
+	}
+
+	while ((zero_cpos + zero_clusters) < last_cpos) {
+		rc = ocfs2_get_clusters(inode, zero_cpos + zero_clusters,
+					&p_cpos, &num_clusters,
+					&ext_flags);
+		if (rc) {
+			mlog_errno(rc);
+			goto out;
+		}
+
+		if (!p_cpos || (ext_flags & OCFS2_EXT_UNWRITTEN))
+			break;
+		if (ext_flags & OCFS2_EXT_REFCOUNTED)
+			needs_cow = 1;
+		zero_clusters += num_clusters;
+	}
+	if ((zero_cpos + zero_clusters) > last_cpos)
+		zero_clusters = last_cpos - zero_cpos;
+
+	if (needs_cow) {
+		rc = ocfs2_refcount_cow(inode, di_bh, zero_cpos,
+					zero_clusters, UINT_MAX);
+		if (rc) {
+			mlog_errno(rc);
+			goto out;
+		}
+	}
+
+	*range_start = ocfs2_clusters_to_bytes(inode->i_sb, zero_cpos);
+	*range_end = ocfs2_clusters_to_bytes(inode->i_sb,
+					     zero_cpos + zero_clusters);
+
+out:
+	return rc;
+}
+
+/*
+ * Zero one range returned from ocfs2_zero_extend_get_range().  The caller
+ * has made sure that the entire range needs zeroing.
+ */
+static int ocfs2_zero_extend_range(struct inode *inode, u64 range_start,
+				   u64 range_end, struct buffer_head *di_bh)
+{
+	int rc = 0;
+	u64 next_pos;
+	u64 zero_pos = range_start;
+
+	trace_ocfs2_zero_extend_range(
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			(unsigned long long)range_start,
+			(unsigned long long)range_end);
+	BUG_ON(range_start >= range_end);
+
+	while (zero_pos < range_end) {
+		next_pos = (zero_pos & PAGE_CACHE_MASK) + PAGE_CACHE_SIZE;
+		if (next_pos > range_end)
+			next_pos = range_end;
+		rc = ocfs2_write_zero_page(inode, zero_pos, next_pos, di_bh);
+		if (rc < 0) {
+			mlog_errno(rc);
+			break;
+		}
+		zero_pos = next_pos;
+
+		/*
+		 * Very large extends have the potential to lock up
+		 * the cpu for extended periods of time.
+		 */
+		cond_resched();
+	}
+
+	return rc;
+}
+
+int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
+		      loff_t zero_to_size)
+{
+	int ret = 0;
+	u64 zero_start, range_start = 0, range_end = 0;
+	struct super_block *sb = inode->i_sb;
+
+	zero_start = ocfs2_align_bytes_to_blocks(sb, i_size_read(inode));
+	trace_ocfs2_zero_extend((unsigned long long)OCFS2_I(inode)->ip_blkno,
+				(unsigned long long)zero_start,
+				(unsigned long long)i_size_read(inode));
+	while (zero_start < zero_to_size) {
+		ret = ocfs2_zero_extend_get_range(inode, di_bh, zero_start,
+						  zero_to_size,
+						  &range_start,
+						  &range_end);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+		if (!range_end)
+			break;
+		/* Trim the ends */
+		if (range_start < zero_start)
+			range_start = zero_start;
+		if (range_end > zero_to_size)
+			range_end = zero_to_size;
+
+		ret = ocfs2_zero_extend_range(inode, range_start,
+					      range_end, di_bh);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+		zero_start = range_end;
+	}
+
+	return ret;
+}
+
+int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
+			  u64 new_i_size, u64 zero_to)
+{
+	int ret;
+	u32 clusters_to_add;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	/*
+	 * Only quota files call this without a bh, and they can't be
+	 * refcounted.
+	 */
+	BUG_ON(!di_bh && (oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+	BUG_ON(!di_bh && !(oi->ip_flags & OCFS2_INODE_SYSTEM_FILE));
+
+	clusters_to_add = ocfs2_clusters_for_bytes(inode->i_sb, new_i_size);
+	if (clusters_to_add < oi->ip_clusters)
+		clusters_to_add = 0;
+	else
+		clusters_to_add -= oi->ip_clusters;
+
+	if (clusters_to_add) {
+		ret = __ocfs2_extend_allocation(inode, oi->ip_clusters,
+						clusters_to_add, 0);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	/*
+	 * Call this even if we don't add any clusters to the tree. We
+	 * still need to zero the area between the old i_size and the
+	 * new i_size.
+	 */
+	ret = ocfs2_zero_extend(inode, di_bh, zero_to);
+	if (ret < 0)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+static int ocfs2_extend_file(struct inode *inode,
+			     struct buffer_head *di_bh,
+			     u64 new_i_size)
+{
+	int ret = 0;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	BUG_ON(!di_bh);
+
+	/* setattr sometimes calls us like this. */
+	if (new_i_size == 0)
+		goto out;
+
+	if (i_size_read(inode) == new_i_size)
+		goto out;
+	BUG_ON(new_i_size < i_size_read(inode));
+
+	/*
+	 * The alloc sem blocks people in read/write from reading our
+	 * allocation until we're done changing it. We depend on
+	 * i_mutex to block other extend/truncate calls while we're
+	 * here.  We even have to hold it for sparse files because there
+	 * might be some tail zeroing.
+	 */
+	down_write(&oi->ip_alloc_sem);
+
+	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		/*
+		 * We can optimize small extends by keeping the inodes
+		 * inline data.
+		 */
+		if (ocfs2_size_fits_inline_data(di_bh, new_i_size)) {
+			up_write(&oi->ip_alloc_sem);
+			goto out_update_size;
+		}
+
+		ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
+		if (ret) {
+			up_write(&oi->ip_alloc_sem);
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (ocfs2_sparse_alloc(OCFS2_SB(inode->i_sb)))
+		ret = ocfs2_zero_extend(inode, di_bh, new_i_size);
+	else
+		ret = ocfs2_extend_no_holes(inode, di_bh, new_i_size,
+					    new_i_size);
+
+	up_write(&oi->ip_alloc_sem);
+
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+out_update_size:
+	ret = ocfs2_simple_size_update(inode, di_bh, new_i_size);
+	if (ret < 0)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+int ocfs2_setattr(struct dentry *dentry, struct iattr *attr)
+{
+	int status = 0, size_change;
+	struct inode *inode = d_inode(dentry);
+	struct super_block *sb = inode->i_sb;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	struct buffer_head *bh = NULL;
+	handle_t *handle = NULL;
+	struct dquot *transfer_to[MAXQUOTAS] = { };
+	int qtype;
+
+	trace_ocfs2_setattr(inode, dentry,
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    dentry->d_name.len, dentry->d_name.name,
+			    attr->ia_valid, attr->ia_mode,
+			    from_kuid(&init_user_ns, attr->ia_uid),
+			    from_kgid(&init_user_ns, attr->ia_gid));
+
+	/* ensuring we don't even attempt to truncate a symlink */
+	if (S_ISLNK(inode->i_mode))
+		attr->ia_valid &= ~ATTR_SIZE;
+
+#define OCFS2_VALID_ATTRS (ATTR_ATIME | ATTR_MTIME | ATTR_CTIME | ATTR_SIZE \
+			   | ATTR_GID | ATTR_UID | ATTR_MODE)
+	if (!(attr->ia_valid & OCFS2_VALID_ATTRS))
+		return 0;
+
+	status = inode_change_ok(inode, attr);
+	if (status)
+		return status;
+
+	if (is_quota_modification(inode, attr))
+		dquot_initialize(inode);
+	size_change = S_ISREG(inode->i_mode) && attr->ia_valid & ATTR_SIZE;
+	if (size_change) {
+		status = ocfs2_rw_lock(inode, 1);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	status = ocfs2_inode_lock(inode, &bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		goto bail_unlock_rw;
+	}
+
+	if (size_change) {
+		status = inode_newsize_ok(inode, attr->ia_size);
+		if (status)
+			goto bail_unlock;
+
+		inode_dio_wait(inode);
+
+		if (i_size_read(inode) >= attr->ia_size) {
+			if (ocfs2_should_order_data(inode)) {
+				status = ocfs2_begin_ordered_truncate(inode,
+								      attr->ia_size);
+				if (status)
+					goto bail_unlock;
+			}
+			status = ocfs2_truncate_file(inode, bh, attr->ia_size);
+		} else
+			status = ocfs2_extend_file(inode, bh, attr->ia_size);
+		if (status < 0) {
+			if (status != -ENOSPC)
+				mlog_errno(status);
+			status = -ENOSPC;
+			goto bail_unlock;
+		}
+	}
+
+	if ((attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)) ||
+	    (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid))) {
+		/*
+		 * Gather pointers to quota structures so that allocation /
+		 * freeing of quota structures happens here and not inside
+		 * dquot_transfer() where we have problems with lock ordering
+		 */
+		if (attr->ia_valid & ATTR_UID && !uid_eq(attr->ia_uid, inode->i_uid)
+		    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
+		    OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
+			transfer_to[USRQUOTA] = dqget(sb, make_kqid_uid(attr->ia_uid));
+			if (!transfer_to[USRQUOTA]) {
+				status = -ESRCH;
+				goto bail_unlock;
+			}
+		}
+		if (attr->ia_valid & ATTR_GID && !gid_eq(attr->ia_gid, inode->i_gid)
+		    && OCFS2_HAS_RO_COMPAT_FEATURE(sb,
+		    OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
+			transfer_to[GRPQUOTA] = dqget(sb, make_kqid_gid(attr->ia_gid));
+			if (!transfer_to[GRPQUOTA]) {
+				status = -ESRCH;
+				goto bail_unlock;
+			}
+		}
+		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS +
+					   2 * ocfs2_quota_trans_credits(sb));
+		if (IS_ERR(handle)) {
+			status = PTR_ERR(handle);
+			mlog_errno(status);
+			goto bail_unlock;
+		}
+		status = __dquot_transfer(inode, transfer_to);
+		if (status < 0)
+			goto bail_commit;
+	} else {
+		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+		if (IS_ERR(handle)) {
+			status = PTR_ERR(handle);
+			mlog_errno(status);
+			goto bail_unlock;
+		}
+	}
+
+	setattr_copy(inode, attr);
+	mark_inode_dirty(inode);
+
+	status = ocfs2_mark_inode_dirty(handle, inode, bh);
+	if (status < 0)
+		mlog_errno(status);
+
+bail_commit:
+	ocfs2_commit_trans(osb, handle);
+bail_unlock:
+	ocfs2_inode_unlock(inode, 1);
+bail_unlock_rw:
+	if (size_change)
+		ocfs2_rw_unlock(inode, 1);
+bail:
+	brelse(bh);
+
+	/* Release quota pointers in case we acquired them */
+	for (qtype = 0; qtype < OCFS2_MAXQUOTAS; qtype++)
+		dqput(transfer_to[qtype]);
+
+	if (!status && attr->ia_valid & ATTR_MODE) {
+		status = posix_acl_chmod(inode, inode->i_mode);
+		if (status < 0)
+			mlog_errno(status);
+	}
+
+	return status;
+}
+
+int ocfs2_getattr(struct vfsmount *mnt,
+		  struct dentry *dentry,
+		  struct kstat *stat)
+{
+	struct inode *inode = d_inode(dentry);
+	struct super_block *sb = d_inode(dentry)->i_sb;
+	struct ocfs2_super *osb = sb->s_fs_info;
+	int err;
+
+	err = ocfs2_inode_revalidate(dentry);
+	if (err) {
+		if (err != -ENOENT)
+			mlog_errno(err);
+		goto bail;
+	}
+
+	generic_fillattr(inode, stat);
+
+	/* We set the blksize from the cluster size for performance */
+	stat->blksize = osb->s_clustersize;
+
+bail:
+	return err;
+}
+
+int ocfs2_permission(struct inode *inode, int mask)
+{
+	int ret;
+
+	if (mask & MAY_NOT_BLOCK)
+		return -ECHILD;
+
+	ret = ocfs2_inode_lock(inode, NULL, 0);
+	if (ret) {
+		if (ret != -ENOENT)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	ret = generic_permission(inode, mask);
+
+	ocfs2_inode_unlock(inode, 0);
+out:
+	return ret;
+}
+
+static int __ocfs2_write_remove_suid(struct inode *inode,
+				     struct buffer_head *bh)
+{
+	int ret;
+	handle_t *handle;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di;
+
+	trace_ocfs2_write_remove_suid(
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			inode->i_mode);
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_trans;
+	}
+
+	inode->i_mode &= ~S_ISUID;
+	if ((inode->i_mode & S_ISGID) && (inode->i_mode & S_IXGRP))
+		inode->i_mode &= ~S_ISGID;
+
+	di = (struct ocfs2_dinode *) bh->b_data;
+	di->i_mode = cpu_to_le16(inode->i_mode);
+	ocfs2_update_inode_fsync_trans(handle, inode, 0);
+
+	ocfs2_journal_dirty(handle, bh);
+
+out_trans:
+	ocfs2_commit_trans(osb, handle);
+out:
+	return ret;
+}
+
+/*
+ * Will look for holes and unwritten extents in the range starting at
+ * pos for count bytes (inclusive).
+ */
+static int ocfs2_check_range_for_holes(struct inode *inode, loff_t pos,
+				       size_t count)
+{
+	int ret = 0;
+	unsigned int extent_flags;
+	u32 cpos, clusters, extent_len, phys_cpos;
+	struct super_block *sb = inode->i_sb;
+
+	cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
+	clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
+
+	while (clusters) {
+		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
+					 &extent_flags);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (phys_cpos == 0 || (extent_flags & OCFS2_EXT_UNWRITTEN)) {
+			ret = 1;
+			break;
+		}
+
+		if (extent_len > clusters)
+			extent_len = clusters;
+
+		clusters -= extent_len;
+		cpos += extent_len;
+	}
+out:
+	return ret;
+}
+
+static int ocfs2_write_remove_suid(struct inode *inode)
+{
+	int ret;
+	struct buffer_head *bh = NULL;
+
+	ret = ocfs2_read_inode_block(inode, &bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret =  __ocfs2_write_remove_suid(inode, bh);
+out:
+	brelse(bh);
+	return ret;
+}
+
+/*
+ * Allocate enough extents to cover the region starting at byte offset
+ * start for len bytes. Existing extents are skipped, any extents
+ * added are marked as "unwritten".
+ */
+static int ocfs2_allocate_unwritten_extents(struct inode *inode,
+					    u64 start, u64 len)
+{
+	int ret;
+	u32 cpos, phys_cpos, clusters, alloc_size;
+	u64 end = start + len;
+	struct buffer_head *di_bh = NULL;
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		ret = ocfs2_read_inode_block(inode, &di_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * Nothing to do if the requested reservation range
+		 * fits within the inode.
+		 */
+		if (ocfs2_size_fits_inline_data(di_bh, end))
+			goto out;
+
+		ret = ocfs2_convert_inline_data_to_extents(inode, di_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	/*
+	 * We consider both start and len to be inclusive.
+	 */
+	cpos = start >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+	clusters = ocfs2_clusters_for_bytes(inode->i_sb, start + len);
+	clusters -= cpos;
+
+	while (clusters) {
+		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos,
+					 &alloc_size, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * Hole or existing extent len can be arbitrary, so
+		 * cap it to our own allocation request.
+		 */
+		if (alloc_size > clusters)
+			alloc_size = clusters;
+
+		if (phys_cpos) {
+			/*
+			 * We already have an allocation at this
+			 * region so we can safely skip it.
+			 */
+			goto next;
+		}
+
+		ret = __ocfs2_extend_allocation(inode, cpos, alloc_size, 1);
+		if (ret) {
+			if (ret != -ENOSPC)
+				mlog_errno(ret);
+			goto out;
+		}
+
+next:
+		cpos += alloc_size;
+		clusters -= alloc_size;
+	}
+
+	ret = 0;
+out:
+
+	brelse(di_bh);
+	return ret;
+}
+
+/*
+ * Truncate a byte range, avoiding pages within partial clusters. This
+ * preserves those pages for the zeroing code to write to.
+ */
+static void ocfs2_truncate_cluster_pages(struct inode *inode, u64 byte_start,
+					 u64 byte_len)
+{
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	loff_t start, end;
+	struct address_space *mapping = inode->i_mapping;
+
+	start = (loff_t)ocfs2_align_bytes_to_clusters(inode->i_sb, byte_start);
+	end = byte_start + byte_len;
+	end = end & ~(osb->s_clustersize - 1);
+
+	if (start < end) {
+		unmap_mapping_range(mapping, start, end - start, 0);
+		truncate_inode_pages_range(mapping, start, end - 1);
+	}
+}
+
+static int ocfs2_zero_partial_clusters(struct inode *inode,
+				       u64 start, u64 len)
+{
+	int ret = 0;
+	u64 tmpend, end = start + len;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	unsigned int csize = osb->s_clustersize;
+	handle_t *handle;
+
+	/*
+	 * The "start" and "end" values are NOT necessarily part of
+	 * the range whose allocation is being deleted. Rather, this
+	 * is what the user passed in with the request. We must zero
+	 * partial clusters here. There's no need to worry about
+	 * physical allocation - the zeroing code knows to skip holes.
+	 */
+	trace_ocfs2_zero_partial_clusters(
+		(unsigned long long)OCFS2_I(inode)->ip_blkno,
+		(unsigned long long)start, (unsigned long long)end);
+
+	/*
+	 * If both edges are on a cluster boundary then there's no
+	 * zeroing required as the region is part of the allocation to
+	 * be truncated.
+	 */
+	if ((start & (csize - 1)) == 0 && (end & (csize - 1)) == 0)
+		goto out;
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * We want to get the byte offset of the end of the 1st cluster.
+	 */
+	tmpend = (u64)osb->s_clustersize + (start & ~(osb->s_clustersize - 1));
+	if (tmpend > end)
+		tmpend = end;
+
+	trace_ocfs2_zero_partial_clusters_range1((unsigned long long)start,
+						 (unsigned long long)tmpend);
+
+	ret = ocfs2_zero_range_for_truncate(inode, handle, start, tmpend);
+	if (ret)
+		mlog_errno(ret);
+
+	if (tmpend < end) {
+		/*
+		 * This may make start and end equal, but the zeroing
+		 * code will skip any work in that case so there's no
+		 * need to catch it up here.
+		 */
+		start = end & ~(osb->s_clustersize - 1);
+
+		trace_ocfs2_zero_partial_clusters_range2(
+			(unsigned long long)start, (unsigned long long)end);
+
+		ret = ocfs2_zero_range_for_truncate(inode, handle, start, end);
+		if (ret)
+			mlog_errno(ret);
+	}
+	ocfs2_update_inode_fsync_trans(handle, inode, 1);
+
+	ocfs2_commit_trans(osb, handle);
+out:
+	return ret;
+}
+
+static int ocfs2_find_rec(struct ocfs2_extent_list *el, u32 pos)
+{
+	int i;
+	struct ocfs2_extent_rec *rec = NULL;
+
+	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
+
+		rec = &el->l_recs[i];
+
+		if (le32_to_cpu(rec->e_cpos) < pos)
+			break;
+	}
+
+	return i;
+}
+
+/*
+ * Helper to calculate the punching pos and length in one run, we handle the
+ * following three cases in order:
+ *
+ * - remove the entire record
+ * - remove a partial record
+ * - no record needs to be removed (hole-punching completed)
+*/
+static void ocfs2_calc_trunc_pos(struct inode *inode,
+				 struct ocfs2_extent_list *el,
+				 struct ocfs2_extent_rec *rec,
+				 u32 trunc_start, u32 *trunc_cpos,
+				 u32 *trunc_len, u32 *trunc_end,
+				 u64 *blkno, int *done)
+{
+	int ret = 0;
+	u32 coff, range;
+
+	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
+
+	if (le32_to_cpu(rec->e_cpos) >= trunc_start) {
+		/*
+		 * remove an entire extent record.
+		 */
+		*trunc_cpos = le32_to_cpu(rec->e_cpos);
+		/*
+		 * Skip holes if any.
+		 */
+		if (range < *trunc_end)
+			*trunc_end = range;
+		*trunc_len = *trunc_end - le32_to_cpu(rec->e_cpos);
+		*blkno = le64_to_cpu(rec->e_blkno);
+		*trunc_end = le32_to_cpu(rec->e_cpos);
+	} else if (range > trunc_start) {
+		/*
+		 * remove a partial extent record, which means we're
+		 * removing the last extent record.
+		 */
+		*trunc_cpos = trunc_start;
+		/*
+		 * skip hole if any.
+		 */
+		if (range < *trunc_end)
+			*trunc_end = range;
+		*trunc_len = *trunc_end - trunc_start;
+		coff = trunc_start - le32_to_cpu(rec->e_cpos);
+		*blkno = le64_to_cpu(rec->e_blkno) +
+				ocfs2_clusters_to_blocks(inode->i_sb, coff);
+		*trunc_end = trunc_start;
+	} else {
+		/*
+		 * It may have two following possibilities:
+		 *
+		 * - last record has been removed
+		 * - trunc_start was within a hole
+		 *
+		 * both two cases mean the completion of hole punching.
+		 */
+		ret = 1;
+	}
+
+	*done = ret;
+}
+
+static int ocfs2_remove_inode_range(struct inode *inode,
+				    struct buffer_head *di_bh, u64 byte_start,
+				    u64 byte_len)
+{
+	int ret = 0, flags = 0, done = 0, i;
+	u32 trunc_start, trunc_len, trunc_end, trunc_cpos, phys_cpos;
+	u32 cluster_in_el;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct address_space *mapping = inode->i_mapping;
+	struct ocfs2_extent_tree et;
+	struct ocfs2_path *path = NULL;
+	struct ocfs2_extent_list *el = NULL;
+	struct ocfs2_extent_rec *rec = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	u64 blkno, refcount_loc = le64_to_cpu(di->i_refcount_loc);
+
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	trace_ocfs2_remove_inode_range(
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			(unsigned long long)byte_start,
+			(unsigned long long)byte_len);
+
+	if (byte_len == 0)
+		return 0;
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		ret = ocfs2_truncate_inline(inode, di_bh, byte_start,
+					    byte_start + byte_len, 0);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		/*
+		 * There's no need to get fancy with the page cache
+		 * truncate of an inline-data inode. We're talking
+		 * about less than a page here, which will be cached
+		 * in the dinode buffer anyway.
+		 */
+		unmap_mapping_range(mapping, 0, 0, 0);
+		truncate_inode_pages(mapping, 0);
+		goto out;
+	}
+
+	/*
+	 * For reflinks, we may need to CoW 2 clusters which might be
+	 * partially zero'd later, if hole's start and end offset were
+	 * within one cluster(means is not exactly aligned to clustersize).
+	 */
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
+
+		ret = ocfs2_cow_file_pos(inode, di_bh, byte_start);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_cow_file_pos(inode, di_bh, byte_start + byte_len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	trunc_start = ocfs2_clusters_for_bytes(osb->sb, byte_start);
+	trunc_end = (byte_start + byte_len) >> osb->s_clustersize_bits;
+	cluster_in_el = trunc_end;
+
+	ret = ocfs2_zero_partial_clusters(inode, byte_start, byte_len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	path = ocfs2_new_path_from_et(&et);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	while (trunc_end > trunc_start) {
+
+		ret = ocfs2_find_path(INODE_CACHE(inode), path,
+				      cluster_in_el);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		el = path_leaf_el(path);
+
+		i = ocfs2_find_rec(el, trunc_end);
+		/*
+		 * Need to go to previous extent block.
+		 */
+		if (i < 0) {
+			if (path->p_tree_depth == 0)
+				break;
+
+			ret = ocfs2_find_cpos_for_left_leaf(inode->i_sb,
+							    path,
+							    &cluster_in_el);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			/*
+			 * We've reached the leftmost extent block,
+			 * it's safe to leave.
+			 */
+			if (cluster_in_el == 0)
+				break;
+
+			/*
+			 * The 'pos' searched for previous extent block is
+			 * always one cluster less than actual trunc_end.
+			 */
+			trunc_end = cluster_in_el + 1;
+
+			ocfs2_reinit_path(path, 1);
+
+			continue;
+
+		} else
+			rec = &el->l_recs[i];
+
+		ocfs2_calc_trunc_pos(inode, el, rec, trunc_start, &trunc_cpos,
+				     &trunc_len, &trunc_end, &blkno, &done);
+		if (done)
+			break;
+
+		flags = rec->e_flags;
+		phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
+
+		ret = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
+					       phys_cpos, trunc_len, flags,
+					       &dealloc, refcount_loc, false);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		cluster_in_el = trunc_end;
+
+		ocfs2_reinit_path(path, 1);
+	}
+
+	ocfs2_truncate_cluster_pages(inode, byte_start, byte_len);
+
+out:
+	ocfs2_free_path(path);
+	ocfs2_schedule_truncate_log_flush(osb, 1);
+	ocfs2_run_deallocs(osb, &dealloc);
+
+	return ret;
+}
+
+/*
+ * Parts of this function taken from xfs_change_file_space()
+ */
+static int __ocfs2_change_file_space(struct file *file, struct inode *inode,
+				     loff_t f_pos, unsigned int cmd,
+				     struct ocfs2_space_resv *sr,
+				     int change_size)
+{
+	int ret;
+	s64 llen;
+	loff_t size;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *di_bh = NULL;
+	handle_t *handle;
+	unsigned long long max_off = inode->i_sb->s_maxbytes;
+
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
+		return -EROFS;
+
+	mutex_lock(&inode->i_mutex);
+
+	/*
+	 * This prevents concurrent writes on other nodes
+	 */
+	ret = ocfs2_rw_lock(inode, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_rw_unlock;
+	}
+
+	if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
+		ret = -EPERM;
+		goto out_inode_unlock;
+	}
+
+	switch (sr->l_whence) {
+	case 0: /*SEEK_SET*/
+		break;
+	case 1: /*SEEK_CUR*/
+		sr->l_start += f_pos;
+		break;
+	case 2: /*SEEK_END*/
+		sr->l_start += i_size_read(inode);
+		break;
+	default:
+		ret = -EINVAL;
+		goto out_inode_unlock;
+	}
+	sr->l_whence = 0;
+
+	llen = sr->l_len > 0 ? sr->l_len - 1 : sr->l_len;
+
+	if (sr->l_start < 0
+	    || sr->l_start > max_off
+	    || (sr->l_start + llen) < 0
+	    || (sr->l_start + llen) > max_off) {
+		ret = -EINVAL;
+		goto out_inode_unlock;
+	}
+	size = sr->l_start + sr->l_len;
+
+	if (cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64 ||
+	    cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) {
+		if (sr->l_len <= 0) {
+			ret = -EINVAL;
+			goto out_inode_unlock;
+		}
+	}
+
+	if (file && should_remove_suid(file->f_path.dentry)) {
+		ret = __ocfs2_write_remove_suid(inode, di_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_inode_unlock;
+		}
+	}
+
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+	switch (cmd) {
+	case OCFS2_IOC_RESVSP:
+	case OCFS2_IOC_RESVSP64:
+		/*
+		 * This takes unsigned offsets, but the signed ones we
+		 * pass have been checked against overflow above.
+		 */
+		ret = ocfs2_allocate_unwritten_extents(inode, sr->l_start,
+						       sr->l_len);
+		break;
+	case OCFS2_IOC_UNRESVSP:
+	case OCFS2_IOC_UNRESVSP64:
+		ret = ocfs2_remove_inode_range(inode, di_bh, sr->l_start,
+					       sr->l_len);
+		break;
+	default:
+		ret = -EINVAL;
+	}
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_inode_unlock;
+	}
+
+	/*
+	 * We update c/mtime for these changes
+	 */
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_inode_unlock;
+	}
+
+	if (change_size && i_size_read(inode) < size)
+		i_size_write(inode, size);
+
+	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
+	ret = ocfs2_mark_inode_dirty(handle, inode, di_bh);
+	if (ret < 0)
+		mlog_errno(ret);
+
+	if (file && (file->f_flags & O_SYNC))
+		handle->h_sync = 1;
+
+	ocfs2_commit_trans(osb, handle);
+
+out_inode_unlock:
+	brelse(di_bh);
+	ocfs2_inode_unlock(inode, 1);
+out_rw_unlock:
+	ocfs2_rw_unlock(inode, 1);
+
+out:
+	mutex_unlock(&inode->i_mutex);
+	return ret;
+}
+
+int ocfs2_change_file_space(struct file *file, unsigned int cmd,
+			    struct ocfs2_space_resv *sr)
+{
+	struct inode *inode = file_inode(file);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	int ret;
+
+	if ((cmd == OCFS2_IOC_RESVSP || cmd == OCFS2_IOC_RESVSP64) &&
+	    !ocfs2_writes_unwritten_extents(osb))
+		return -ENOTTY;
+	else if ((cmd == OCFS2_IOC_UNRESVSP || cmd == OCFS2_IOC_UNRESVSP64) &&
+		 !ocfs2_sparse_alloc(osb))
+		return -ENOTTY;
+
+	if (!S_ISREG(inode->i_mode))
+		return -EINVAL;
+
+	if (!(file->f_mode & FMODE_WRITE))
+		return -EBADF;
+
+	ret = mnt_want_write_file(file);
+	if (ret)
+		return ret;
+	ret = __ocfs2_change_file_space(file, inode, file->f_pos, cmd, sr, 0);
+	mnt_drop_write_file(file);
+	return ret;
+}
+
+static long ocfs2_fallocate(struct file *file, int mode, loff_t offset,
+			    loff_t len)
+{
+	struct inode *inode = file_inode(file);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_space_resv sr;
+	int change_size = 1;
+	int cmd = OCFS2_IOC_RESVSP64;
+
+	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
+		return -EOPNOTSUPP;
+	if (!ocfs2_writes_unwritten_extents(osb))
+		return -EOPNOTSUPP;
+
+	if (mode & FALLOC_FL_KEEP_SIZE)
+		change_size = 0;
+
+	if (mode & FALLOC_FL_PUNCH_HOLE)
+		cmd = OCFS2_IOC_UNRESVSP64;
+
+	sr.l_whence = 0;
+	sr.l_start = (s64)offset;
+	sr.l_len = (s64)len;
+
+	return __ocfs2_change_file_space(NULL, inode, offset, cmd, &sr,
+					 change_size);
+}
+
+int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
+				   size_t count)
+{
+	int ret = 0;
+	unsigned int extent_flags;
+	u32 cpos, clusters, extent_len, phys_cpos;
+	struct super_block *sb = inode->i_sb;
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)) ||
+	    !(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) ||
+	    OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		return 0;
+
+	cpos = pos >> OCFS2_SB(sb)->s_clustersize_bits;
+	clusters = ocfs2_clusters_for_bytes(sb, pos + count) - cpos;
+
+	while (clusters) {
+		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &extent_len,
+					 &extent_flags);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (phys_cpos && (extent_flags & OCFS2_EXT_REFCOUNTED)) {
+			ret = 1;
+			break;
+		}
+
+		if (extent_len > clusters)
+			extent_len = clusters;
+
+		clusters -= extent_len;
+		cpos += extent_len;
+	}
+out:
+	return ret;
+}
+
+static int ocfs2_is_io_unaligned(struct inode *inode, size_t count, loff_t pos)
+{
+	int blockmask = inode->i_sb->s_blocksize - 1;
+	loff_t final_size = pos + count;
+
+	if ((pos & blockmask) || (final_size & blockmask))
+		return 1;
+	return 0;
+}
+
+static int ocfs2_prepare_inode_for_refcount(struct inode *inode,
+					    struct file *file,
+					    loff_t pos, size_t count,
+					    int *meta_level)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	u32 cpos = pos >> OCFS2_SB(inode->i_sb)->s_clustersize_bits;
+	u32 clusters =
+		ocfs2_clusters_for_bytes(inode->i_sb, pos + count) - cpos;
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	*meta_level = 1;
+
+	ret = ocfs2_refcount_cow(inode, di_bh, cpos, clusters, UINT_MAX);
+	if (ret)
+		mlog_errno(ret);
+out:
+	brelse(di_bh);
+	return ret;
+}
+
+static int ocfs2_prepare_inode_for_write(struct file *file,
+					 loff_t pos,
+					 size_t count,
+					 int appending,
+					 int *direct_io,
+					 int *has_refcount)
+{
+	int ret = 0, meta_level = 0;
+	struct dentry *dentry = file->f_path.dentry;
+	struct inode *inode = d_inode(dentry);
+	loff_t end;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	int full_coherency = !(osb->s_mount_opt &
+		OCFS2_MOUNT_COHERENCY_BUFFERED);
+
+	/*
+	 * We start with a read level meta lock and only jump to an ex
+	 * if we need to make modifications here.
+	 */
+	for(;;) {
+		ret = ocfs2_inode_lock(inode, NULL, meta_level);
+		if (ret < 0) {
+			meta_level = -1;
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/* Clear suid / sgid if necessary. We do this here
+		 * instead of later in the write path because
+		 * remove_suid() calls ->setattr without any hint that
+		 * we may have already done our cluster locking. Since
+		 * ocfs2_setattr() *must* take cluster locks to
+		 * proceed, this will lead us to recursively lock the
+		 * inode. There's also the dinode i_size state which
+		 * can be lost via setattr during extending writes (we
+		 * set inode->i_size at the end of a write. */
+		if (should_remove_suid(dentry)) {
+			if (meta_level == 0) {
+				ocfs2_inode_unlock(inode, meta_level);
+				meta_level = 1;
+				continue;
+			}
+
+			ret = ocfs2_write_remove_suid(inode);
+			if (ret < 0) {
+				mlog_errno(ret);
+				goto out_unlock;
+			}
+		}
+
+		end = pos + count;
+
+		ret = ocfs2_check_range_for_refcount(inode, pos, count);
+		if (ret == 1) {
+			ocfs2_inode_unlock(inode, meta_level);
+			meta_level = -1;
+
+			ret = ocfs2_prepare_inode_for_refcount(inode,
+							       file,
+							       pos,
+							       count,
+							       &meta_level);
+			if (has_refcount)
+				*has_refcount = 1;
+			if (direct_io)
+				*direct_io = 0;
+		}
+
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out_unlock;
+		}
+
+		/*
+		 * Skip the O_DIRECT checks if we don't need
+		 * them.
+		 */
+		if (!direct_io || !(*direct_io))
+			break;
+
+		/*
+		 * There's no sane way to do direct writes to an inode
+		 * with inline data.
+		 */
+		if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+			*direct_io = 0;
+			break;
+		}
+
+		/*
+		 * Allowing concurrent direct writes means
+		 * i_size changes wouldn't be synchronized, so
+		 * one node could wind up truncating another
+		 * nodes writes.
+		 */
+		if (end > i_size_read(inode) && !full_coherency) {
+			*direct_io = 0;
+			break;
+		}
+
+		/*
+		 * Fallback to old way if the feature bit is not set.
+		 */
+		if (end > i_size_read(inode) &&
+				!ocfs2_supports_append_dio(osb)) {
+			*direct_io = 0;
+			break;
+		}
+
+		/*
+		 * We don't fill holes during direct io, so
+		 * check for them here. If any are found, the
+		 * caller will have to retake some cluster
+		 * locks and initiate the io as buffered.
+		 */
+		ret = ocfs2_check_range_for_holes(inode, pos, count);
+		if (ret == 1) {
+			/*
+			 * Fallback to old way if the feature bit is not set.
+			 * Otherwise try dio first and then complete the rest
+			 * request through buffer io.
+			 */
+			if (!ocfs2_supports_append_dio(osb))
+				*direct_io = 0;
+			ret = 0;
+		} else if (ret < 0)
+			mlog_errno(ret);
+		break;
+	}
+
+out_unlock:
+	trace_ocfs2_prepare_inode_for_write(OCFS2_I(inode)->ip_blkno,
+					    pos, appending, count,
+					    direct_io, has_refcount);
+
+	if (meta_level >= 0)
+		ocfs2_inode_unlock(inode, meta_level);
+
+out:
+	return ret;
+}
+
+static ssize_t ocfs2_file_write_iter(struct kiocb *iocb,
+				    struct iov_iter *from)
+{
+	int direct_io, appending, rw_level, have_alloc_sem  = 0;
+	int can_do_direct, has_refcount = 0;
+	ssize_t written = 0;
+	ssize_t ret;
+	size_t count = iov_iter_count(from), orig_count;
+	loff_t old_size;
+	u32 old_clusters;
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file_inode(file);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	int full_coherency = !(osb->s_mount_opt &
+			       OCFS2_MOUNT_COHERENCY_BUFFERED);
+	int unaligned_dio = 0;
+	int dropped_dio = 0;
+
+	trace_ocfs2_file_aio_write(inode, file, file->f_path.dentry,
+		(unsigned long long)OCFS2_I(inode)->ip_blkno,
+		file->f_path.dentry->d_name.len,
+		file->f_path.dentry->d_name.name,
+		(unsigned int)from->nr_segs);	/* GRRRRR */
+
+	if (count == 0)
+		return 0;
+
+	appending = iocb->ki_flags & IOCB_APPEND ? 1 : 0;
+	direct_io = iocb->ki_flags & IOCB_DIRECT ? 1 : 0;
+
+	mutex_lock(&inode->i_mutex);
+
+	ocfs2_iocb_clear_sem_locked(iocb);
+
+relock:
+	/* to match setattr's i_mutex -> rw_lock ordering */
+	if (direct_io) {
+		have_alloc_sem = 1;
+		/* communicate with ocfs2_dio_end_io */
+		ocfs2_iocb_set_sem_locked(iocb);
+	}
+
+	/*
+	 * Concurrent O_DIRECT writes are allowed with
+	 * mount_option "coherency=buffered".
+	 */
+	rw_level = (!direct_io || full_coherency);
+
+	ret = ocfs2_rw_lock(inode, rw_level);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_sems;
+	}
+
+	/*
+	 * O_DIRECT writes with "coherency=full" need to take EX cluster
+	 * inode_lock to guarantee coherency.
+	 */
+	if (direct_io && full_coherency) {
+		/*
+		 * We need to take and drop the inode lock to force
+		 * other nodes to drop their caches.  Buffered I/O
+		 * already does this in write_begin().
+		 */
+		ret = ocfs2_inode_lock(inode, NULL, 1);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ocfs2_inode_unlock(inode, 1);
+	}
+
+	orig_count = iov_iter_count(from);
+	ret = generic_write_checks(iocb, from);
+	if (ret <= 0) {
+		if (ret)
+			mlog_errno(ret);
+		goto out;
+	}
+	count = ret;
+
+	can_do_direct = direct_io;
+	ret = ocfs2_prepare_inode_for_write(file, iocb->ki_pos, count, appending,
+					    &can_do_direct, &has_refcount);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (direct_io && !is_sync_kiocb(iocb))
+		unaligned_dio = ocfs2_is_io_unaligned(inode, count, iocb->ki_pos);
+
+	/*
+	 * We can't complete the direct I/O as requested, fall back to
+	 * buffered I/O.
+	 */
+	if (direct_io && !can_do_direct) {
+		ocfs2_rw_unlock(inode, rw_level);
+
+		have_alloc_sem = 0;
+		rw_level = -1;
+
+		direct_io = 0;
+		iocb->ki_flags &= ~IOCB_DIRECT;
+		iov_iter_reexpand(from, orig_count);
+		dropped_dio = 1;
+		goto relock;
+	}
+
+	if (unaligned_dio) {
+		/*
+		 * Wait on previous unaligned aio to complete before
+		 * proceeding.
+		 */
+		mutex_lock(&OCFS2_I(inode)->ip_unaligned_aio);
+		/* Mark the iocb as needing an unlock in ocfs2_dio_end_io */
+		ocfs2_iocb_set_unaligned_aio(iocb);
+	}
+
+	/*
+	 * To later detect whether a journal commit for sync writes is
+	 * necessary, we sample i_size, and cluster count here.
+	 */
+	old_size = i_size_read(inode);
+	old_clusters = OCFS2_I(inode)->ip_clusters;
+
+	/* communicate with ocfs2_dio_end_io */
+	ocfs2_iocb_set_rw_locked(iocb, rw_level);
+
+	written = __generic_file_write_iter(iocb, from);
+	/* buffered aio wouldn't have proper lock coverage today */
+	BUG_ON(written == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
+
+	if (unlikely(written <= 0))
+		goto no_sync;
+
+	if (((file->f_flags & O_DSYNC) && !direct_io) ||
+	    IS_SYNC(inode) || dropped_dio) {
+		ret = filemap_fdatawrite_range(file->f_mapping,
+					       iocb->ki_pos - written,
+					       iocb->ki_pos - 1);
+		if (ret < 0)
+			written = ret;
+
+		if (!ret) {
+			ret = jbd2_journal_force_commit(osb->journal->j_journal);
+			if (ret < 0)
+				written = ret;
+		}
+
+		if (!ret)
+			ret = filemap_fdatawait_range(file->f_mapping,
+						      iocb->ki_pos - written,
+						      iocb->ki_pos - 1);
+	}
+
+no_sync:
+	/*
+	 * deep in g_f_a_w_n()->ocfs2_direct_IO we pass in a ocfs2_dio_end_io
+	 * function pointer which is called when o_direct io completes so that
+	 * it can unlock our rw lock.
+	 * Unfortunately there are error cases which call end_io and others
+	 * that don't.  so we don't have to unlock the rw_lock if either an
+	 * async dio is going to do it in the future or an end_io after an
+	 * error has already done it.
+	 */
+	if ((ret == -EIOCBQUEUED) || (!ocfs2_iocb_is_rw_locked(iocb))) {
+		rw_level = -1;
+		have_alloc_sem = 0;
+		unaligned_dio = 0;
+	}
+
+	if (unaligned_dio) {
+		ocfs2_iocb_clear_unaligned_aio(iocb);
+		mutex_unlock(&OCFS2_I(inode)->ip_unaligned_aio);
+	}
+
+out:
+	if (rw_level != -1)
+		ocfs2_rw_unlock(inode, rw_level);
+
+out_sems:
+	if (have_alloc_sem)
+		ocfs2_iocb_clear_sem_locked(iocb);
+
+	mutex_unlock(&inode->i_mutex);
+
+	if (written)
+		ret = written;
+	return ret;
+}
+
+static ssize_t ocfs2_file_splice_read(struct file *in,
+				      loff_t *ppos,
+				      struct pipe_inode_info *pipe,
+				      size_t len,
+				      unsigned int flags)
+{
+	int ret = 0, lock_level = 0;
+	struct inode *inode = file_inode(in);
+
+	trace_ocfs2_file_splice_read(inode, in, in->f_path.dentry,
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			in->f_path.dentry->d_name.len,
+			in->f_path.dentry->d_name.name, len);
+
+	/*
+	 * See the comment in ocfs2_file_read_iter()
+	 */
+	ret = ocfs2_inode_lock_atime(inode, in->f_path.mnt, &lock_level);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto bail;
+	}
+	ocfs2_inode_unlock(inode, lock_level);
+
+	ret = generic_file_splice_read(in, ppos, pipe, len, flags);
+
+bail:
+	return ret;
+}
+
+static ssize_t ocfs2_file_read_iter(struct kiocb *iocb,
+				   struct iov_iter *to)
+{
+	int ret = 0, rw_level = -1, have_alloc_sem = 0, lock_level = 0;
+	struct file *filp = iocb->ki_filp;
+	struct inode *inode = file_inode(filp);
+
+	trace_ocfs2_file_aio_read(inode, filp, filp->f_path.dentry,
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			filp->f_path.dentry->d_name.len,
+			filp->f_path.dentry->d_name.name,
+			to->nr_segs);	/* GRRRRR */
+
+
+	if (!inode) {
+		ret = -EINVAL;
+		mlog_errno(ret);
+		goto bail;
+	}
+
+	ocfs2_iocb_clear_sem_locked(iocb);
+
+	/*
+	 * buffered reads protect themselves in ->readpage().  O_DIRECT reads
+	 * need locks to protect pending reads from racing with truncate.
+	 */
+	if (iocb->ki_flags & IOCB_DIRECT) {
+		have_alloc_sem = 1;
+		ocfs2_iocb_set_sem_locked(iocb);
+
+		ret = ocfs2_rw_lock(inode, 0);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto bail;
+		}
+		rw_level = 0;
+		/* communicate with ocfs2_dio_end_io */
+		ocfs2_iocb_set_rw_locked(iocb, rw_level);
+	}
+
+	/*
+	 * We're fine letting folks race truncates and extending
+	 * writes with read across the cluster, just like they can
+	 * locally. Hence no rw_lock during read.
+	 *
+	 * Take and drop the meta data lock to update inode fields
+	 * like i_size. This allows the checks down below
+	 * generic_file_aio_read() a chance of actually working.
+	 */
+	ret = ocfs2_inode_lock_atime(inode, filp->f_path.mnt, &lock_level);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto bail;
+	}
+	ocfs2_inode_unlock(inode, lock_level);
+
+	ret = generic_file_read_iter(iocb, to);
+	trace_generic_file_aio_read_ret(ret);
+
+	/* buffered aio wouldn't have proper lock coverage today */
+	BUG_ON(ret == -EIOCBQUEUED && !(iocb->ki_flags & IOCB_DIRECT));
+
+	/* see ocfs2_file_write_iter */
+	if (ret == -EIOCBQUEUED || !ocfs2_iocb_is_rw_locked(iocb)) {
+		rw_level = -1;
+		have_alloc_sem = 0;
+	}
+
+bail:
+	if (have_alloc_sem)
+		ocfs2_iocb_clear_sem_locked(iocb);
+
+	if (rw_level != -1)
+		ocfs2_rw_unlock(inode, rw_level);
+
+	return ret;
+}
+
+/* Refer generic_file_llseek_unlocked() */
+static loff_t ocfs2_file_llseek(struct file *file, loff_t offset, int whence)
+{
+	struct inode *inode = file->f_mapping->host;
+	int ret = 0;
+
+	mutex_lock(&inode->i_mutex);
+
+	switch (whence) {
+	case SEEK_SET:
+		break;
+	case SEEK_END:
+		/* SEEK_END requires the OCFS2 inode lock for the file
+		 * because it references the file's size.
+		 */
+		ret = ocfs2_inode_lock(inode, NULL, 0);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+		offset += i_size_read(inode);
+		ocfs2_inode_unlock(inode, 0);
+		break;
+	case SEEK_CUR:
+		if (offset == 0) {
+			offset = file->f_pos;
+			goto out;
+		}
+		offset += file->f_pos;
+		break;
+	case SEEK_DATA:
+	case SEEK_HOLE:
+		ret = ocfs2_seek_data_hole_offset(file, &offset, whence);
+		if (ret)
+			goto out;
+		break;
+	default:
+		ret = -EINVAL;
+		goto out;
+	}
+
+	offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
+
+out:
+	mutex_unlock(&inode->i_mutex);
+	if (ret)
+		return ret;
+	return offset;
+}
+
+const struct inode_operations ocfs2_file_iops = {
+	.setattr	= ocfs2_setattr,
+	.getattr	= ocfs2_getattr,
+	.permission	= ocfs2_permission,
+	.setxattr	= generic_setxattr,
+	.getxattr	= generic_getxattr,
+	.listxattr	= ocfs2_listxattr,
+	.removexattr	= generic_removexattr,
+	.fiemap		= ocfs2_fiemap,
+	.get_acl	= ocfs2_iop_get_acl,
+	.set_acl	= ocfs2_iop_set_acl,
+};
+
+const struct inode_operations ocfs2_special_file_iops = {
+	.setattr	= ocfs2_setattr,
+	.getattr	= ocfs2_getattr,
+	.permission	= ocfs2_permission,
+	.get_acl	= ocfs2_iop_get_acl,
+	.set_acl	= ocfs2_iop_set_acl,
+};
+
+/*
+ * Other than ->lock, keep ocfs2_fops and ocfs2_dops in sync with
+ * ocfs2_fops_no_plocks and ocfs2_dops_no_plocks!
+ */
+const struct file_operations ocfs2_fops = {
+	.llseek		= ocfs2_file_llseek,
+	.mmap		= ocfs2_mmap,
+	.fsync		= ocfs2_sync_file,
+	.release	= ocfs2_file_release,
+	.open		= ocfs2_file_open,
+	.read_iter	= ocfs2_file_read_iter,
+	.write_iter	= ocfs2_file_write_iter,
+	.unlocked_ioctl	= ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl   = ocfs2_compat_ioctl,
+#endif
+	.lock		= ocfs2_lock,
+	.flock		= ocfs2_flock,
+	.splice_read	= ocfs2_file_splice_read,
+	.splice_write	= iter_file_splice_write,
+	.fallocate	= ocfs2_fallocate,
+};
+
+const struct file_operations ocfs2_dops = {
+	.llseek		= generic_file_llseek,
+	.read		= generic_read_dir,
+	.iterate	= ocfs2_readdir,
+	.fsync		= ocfs2_sync_file,
+	.release	= ocfs2_dir_release,
+	.open		= ocfs2_dir_open,
+	.unlocked_ioctl	= ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl   = ocfs2_compat_ioctl,
+#endif
+	.lock		= ocfs2_lock,
+	.flock		= ocfs2_flock,
+};
+
+/*
+ * POSIX-lockless variants of our file_operations.
+ *
+ * These will be used if the underlying cluster stack does not support
+ * posix file locking, if the user passes the "localflocks" mount
+ * option, or if we have a local-only fs.
+ *
+ * ocfs2_flock is in here because all stacks handle UNIX file locks,
+ * so we still want it in the case of no stack support for
+ * plocks. Internally, it will do the right thing when asked to ignore
+ * the cluster.
+ */
+const struct file_operations ocfs2_fops_no_plocks = {
+	.llseek		= ocfs2_file_llseek,
+	.mmap		= ocfs2_mmap,
+	.fsync		= ocfs2_sync_file,
+	.release	= ocfs2_file_release,
+	.open		= ocfs2_file_open,
+	.read_iter	= ocfs2_file_read_iter,
+	.write_iter	= ocfs2_file_write_iter,
+	.unlocked_ioctl	= ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl   = ocfs2_compat_ioctl,
+#endif
+	.flock		= ocfs2_flock,
+	.splice_read	= ocfs2_file_splice_read,
+	.splice_write	= iter_file_splice_write,
+	.fallocate	= ocfs2_fallocate,
+};
+
+const struct file_operations ocfs2_dops_no_plocks = {
+	.llseek		= generic_file_llseek,
+	.read		= generic_read_dir,
+	.iterate	= ocfs2_readdir,
+	.fsync		= ocfs2_sync_file,
+	.release	= ocfs2_dir_release,
+	.open		= ocfs2_dir_open,
+	.unlocked_ioctl	= ocfs2_ioctl,
+#ifdef CONFIG_COMPAT
+	.compat_ioctl   = ocfs2_compat_ioctl,
+#endif
+	.flock		= ocfs2_flock,
+};
diff --git a/fs/ocfs2/file.h b/fs/ocfs2/file.h
new file mode 100644
index 000000000..e8c62f222
--- /dev/null
+++ b/fs/ocfs2/file.h
@@ -0,0 +1,85 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * file.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_FILE_H
+#define OCFS2_FILE_H
+
+extern const struct file_operations ocfs2_fops;
+extern const struct file_operations ocfs2_dops;
+extern const struct file_operations ocfs2_fops_no_plocks;
+extern const struct file_operations ocfs2_dops_no_plocks;
+extern const struct inode_operations ocfs2_file_iops;
+extern const struct inode_operations ocfs2_special_file_iops;
+struct ocfs2_alloc_context;
+enum ocfs2_alloc_restarted;
+
+struct ocfs2_file_private {
+	struct file		*fp_file;
+	struct mutex		fp_mutex;
+	struct ocfs2_lock_res	fp_flock;
+};
+
+int ocfs2_add_inode_data(struct ocfs2_super *osb,
+			 struct inode *inode,
+			 u32 *logical_offset,
+			 u32 clusters_to_add,
+			 int mark_unwritten,
+			 struct buffer_head *fe_bh,
+			 handle_t *handle,
+			 struct ocfs2_alloc_context *data_ac,
+			 struct ocfs2_alloc_context *meta_ac,
+			 enum ocfs2_alloc_restarted *reason_ret);
+int ocfs2_set_inode_size(handle_t *handle,
+		struct inode *inode,
+		struct buffer_head *fe_bh,
+		u64 new_i_size);
+int ocfs2_simple_size_update(struct inode *inode,
+			     struct buffer_head *di_bh,
+			     u64 new_i_size);
+int ocfs2_truncate_file(struct inode *inode,
+		struct buffer_head *di_bh,
+		u64 new_i_size);
+int ocfs2_extend_no_holes(struct inode *inode, struct buffer_head *di_bh,
+			  u64 new_i_size, u64 zero_to);
+int ocfs2_zero_extend(struct inode *inode, struct buffer_head *di_bh,
+		      loff_t zero_to);
+int ocfs2_extend_allocation(struct inode *inode, u32 logical_start,
+		u32 clusters_to_add, int mark_unwritten);
+int ocfs2_setattr(struct dentry *dentry, struct iattr *attr);
+int ocfs2_getattr(struct vfsmount *mnt, struct dentry *dentry,
+		  struct kstat *stat);
+int ocfs2_permission(struct inode *inode, int mask);
+
+int ocfs2_should_update_atime(struct inode *inode,
+			      struct vfsmount *vfsmnt);
+int ocfs2_update_inode_atime(struct inode *inode,
+			     struct buffer_head *bh);
+
+int ocfs2_change_file_space(struct file *file, unsigned int cmd,
+			    struct ocfs2_space_resv *sr);
+
+int ocfs2_check_range_for_refcount(struct inode *inode, loff_t pos,
+				   size_t count);
+#endif /* OCFS2_FILE_H */
diff --git a/fs/ocfs2/heartbeat.c b/fs/ocfs2/heartbeat.c
new file mode 100644
index 000000000..d8208b20d
--- /dev/null
+++ b/fs/ocfs2/heartbeat.c
@@ -0,0 +1,134 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * heartbeat.c
+ *
+ * Register ourselves with the heartbaet service, keep our node maps
+ * up to date, and fire off recovery when needed.
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "heartbeat.h"
+#include "inode.h"
+#include "journal.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+static inline void __ocfs2_node_map_set_bit(struct ocfs2_node_map *map,
+					    int bit);
+static inline void __ocfs2_node_map_clear_bit(struct ocfs2_node_map *map,
+					      int bit);
+
+/* special case -1 for now
+ * TODO: should *really* make sure the calling func never passes -1!!  */
+static void ocfs2_node_map_init(struct ocfs2_node_map *map)
+{
+	map->num_nodes = OCFS2_NODE_MAP_MAX_NODES;
+	memset(map->map, 0, BITS_TO_LONGS(OCFS2_NODE_MAP_MAX_NODES) *
+	       sizeof(unsigned long));
+}
+
+void ocfs2_init_node_maps(struct ocfs2_super *osb)
+{
+	spin_lock_init(&osb->node_map_lock);
+	ocfs2_node_map_init(&osb->osb_recovering_orphan_dirs);
+}
+
+void ocfs2_do_node_down(int node_num, void *data)
+{
+	struct ocfs2_super *osb = data;
+
+	BUG_ON(osb->node_num == node_num);
+
+	trace_ocfs2_do_node_down(node_num);
+
+	if (!osb->cconn) {
+		/*
+		 * No cluster connection means we're not even ready to
+		 * participate yet.  We check the slots after the cluster
+		 * comes up, so we will notice the node death then.  We
+		 * can safely ignore it here.
+		 */
+		return;
+	}
+
+	ocfs2_recovery_thread(osb, node_num);
+}
+
+static inline void __ocfs2_node_map_set_bit(struct ocfs2_node_map *map,
+					    int bit)
+{
+	set_bit(bit, map->map);
+}
+
+void ocfs2_node_map_set_bit(struct ocfs2_super *osb,
+			    struct ocfs2_node_map *map,
+			    int bit)
+{
+	if (bit==-1)
+		return;
+	BUG_ON(bit >= map->num_nodes);
+	spin_lock(&osb->node_map_lock);
+	__ocfs2_node_map_set_bit(map, bit);
+	spin_unlock(&osb->node_map_lock);
+}
+
+static inline void __ocfs2_node_map_clear_bit(struct ocfs2_node_map *map,
+					      int bit)
+{
+	clear_bit(bit, map->map);
+}
+
+void ocfs2_node_map_clear_bit(struct ocfs2_super *osb,
+			      struct ocfs2_node_map *map,
+			      int bit)
+{
+	if (bit==-1)
+		return;
+	BUG_ON(bit >= map->num_nodes);
+	spin_lock(&osb->node_map_lock);
+	__ocfs2_node_map_clear_bit(map, bit);
+	spin_unlock(&osb->node_map_lock);
+}
+
+int ocfs2_node_map_test_bit(struct ocfs2_super *osb,
+			    struct ocfs2_node_map *map,
+			    int bit)
+{
+	int ret;
+	if (bit >= map->num_nodes) {
+		mlog(ML_ERROR, "bit=%d map->num_nodes=%d\n", bit, map->num_nodes);
+		BUG();
+	}
+	spin_lock(&osb->node_map_lock);
+	ret = test_bit(bit, map->map);
+	spin_unlock(&osb->node_map_lock);
+	return ret;
+}
+
diff --git a/fs/ocfs2/heartbeat.h b/fs/ocfs2/heartbeat.h
new file mode 100644
index 000000000..74b9c5dda
--- /dev/null
+++ b/fs/ocfs2/heartbeat.h
@@ -0,0 +1,45 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * heartbeat.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_HEARTBEAT_H
+#define OCFS2_HEARTBEAT_H
+
+void ocfs2_init_node_maps(struct ocfs2_super *osb);
+
+void ocfs2_do_node_down(int node_num, void *data);
+
+/* node map functions - used to keep track of mounted and in-recovery
+ * nodes. */
+void ocfs2_node_map_set_bit(struct ocfs2_super *osb,
+			    struct ocfs2_node_map *map,
+			    int bit);
+void ocfs2_node_map_clear_bit(struct ocfs2_super *osb,
+			      struct ocfs2_node_map *map,
+			      int bit);
+int ocfs2_node_map_test_bit(struct ocfs2_super *osb,
+			    struct ocfs2_node_map *map,
+			    int bit);
+
+#endif /* OCFS2_HEARTBEAT_H */
diff --git a/fs/ocfs2/inode.c b/fs/ocfs2/inode.c
new file mode 100644
index 000000000..b254416dc
--- /dev/null
+++ b/fs/ocfs2/inode.c
@@ -0,0 +1,1460 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * inode.c
+ *
+ * vfs' aops, fops, dops and iops
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+#include <linux/pagemap.h>
+#include <linux/quotaops.h>
+
+#include <asm/byteorder.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "dir.h"
+#include "blockcheck.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "file.h"
+#include "heartbeat.h"
+#include "inode.h"
+#include "journal.h"
+#include "namei.h"
+#include "suballoc.h"
+#include "super.h"
+#include "symlink.h"
+#include "sysfile.h"
+#include "uptodate.h"
+#include "xattr.h"
+#include "refcounttree.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+struct ocfs2_find_inode_args
+{
+	u64		fi_blkno;
+	unsigned long	fi_ino;
+	unsigned int	fi_flags;
+	unsigned int	fi_sysfile_type;
+};
+
+static struct lock_class_key ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES];
+
+static int ocfs2_read_locked_inode(struct inode *inode,
+				   struct ocfs2_find_inode_args *args);
+static int ocfs2_init_locked_inode(struct inode *inode, void *opaque);
+static int ocfs2_find_actor(struct inode *inode, void *opaque);
+static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
+				    struct inode *inode,
+				    struct buffer_head *fe_bh);
+
+void ocfs2_set_inode_flags(struct inode *inode)
+{
+	unsigned int flags = OCFS2_I(inode)->ip_attr;
+
+	inode->i_flags &= ~(S_IMMUTABLE |
+		S_SYNC | S_APPEND | S_NOATIME | S_DIRSYNC);
+
+	if (flags & OCFS2_IMMUTABLE_FL)
+		inode->i_flags |= S_IMMUTABLE;
+
+	if (flags & OCFS2_SYNC_FL)
+		inode->i_flags |= S_SYNC;
+	if (flags & OCFS2_APPEND_FL)
+		inode->i_flags |= S_APPEND;
+	if (flags & OCFS2_NOATIME_FL)
+		inode->i_flags |= S_NOATIME;
+	if (flags & OCFS2_DIRSYNC_FL)
+		inode->i_flags |= S_DIRSYNC;
+}
+
+/* Propagate flags from i_flags to OCFS2_I(inode)->ip_attr */
+void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi)
+{
+	unsigned int flags = oi->vfs_inode.i_flags;
+
+	oi->ip_attr &= ~(OCFS2_SYNC_FL|OCFS2_APPEND_FL|
+			OCFS2_IMMUTABLE_FL|OCFS2_NOATIME_FL|OCFS2_DIRSYNC_FL);
+	if (flags & S_SYNC)
+		oi->ip_attr |= OCFS2_SYNC_FL;
+	if (flags & S_APPEND)
+		oi->ip_attr |= OCFS2_APPEND_FL;
+	if (flags & S_IMMUTABLE)
+		oi->ip_attr |= OCFS2_IMMUTABLE_FL;
+	if (flags & S_NOATIME)
+		oi->ip_attr |= OCFS2_NOATIME_FL;
+	if (flags & S_DIRSYNC)
+		oi->ip_attr |= OCFS2_DIRSYNC_FL;
+}
+
+struct inode *ocfs2_ilookup(struct super_block *sb, u64 blkno)
+{
+	struct ocfs2_find_inode_args args;
+
+	args.fi_blkno = blkno;
+	args.fi_flags = 0;
+	args.fi_ino = ino_from_blkno(sb, blkno);
+	args.fi_sysfile_type = 0;
+
+	return ilookup5(sb, blkno, ocfs2_find_actor, &args);
+}
+struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 blkno, unsigned flags,
+			 int sysfile_type)
+{
+	struct inode *inode = NULL;
+	struct super_block *sb = osb->sb;
+	struct ocfs2_find_inode_args args;
+	journal_t *journal = OCFS2_SB(sb)->journal->j_journal;
+
+	trace_ocfs2_iget_begin((unsigned long long)blkno, flags,
+			       sysfile_type);
+
+	/* Ok. By now we've either got the offsets passed to us by the
+	 * caller, or we just pulled them off the bh. Lets do some
+	 * sanity checks to make sure they're OK. */
+	if (blkno == 0) {
+		inode = ERR_PTR(-EINVAL);
+		mlog_errno(PTR_ERR(inode));
+		goto bail;
+	}
+
+	args.fi_blkno = blkno;
+	args.fi_flags = flags;
+	args.fi_ino = ino_from_blkno(sb, blkno);
+	args.fi_sysfile_type = sysfile_type;
+
+	inode = iget5_locked(sb, args.fi_ino, ocfs2_find_actor,
+			     ocfs2_init_locked_inode, &args);
+	/* inode was *not* in the inode cache. 2.6.x requires
+	 * us to do our own read_inode call and unlock it
+	 * afterwards. */
+	if (inode == NULL) {
+		inode = ERR_PTR(-ENOMEM);
+		mlog_errno(PTR_ERR(inode));
+		goto bail;
+	}
+	trace_ocfs2_iget5_locked(inode->i_state);
+	if (inode->i_state & I_NEW) {
+		ocfs2_read_locked_inode(inode, &args);
+		unlock_new_inode(inode);
+	}
+	if (is_bad_inode(inode)) {
+		iput(inode);
+		inode = ERR_PTR(-ESTALE);
+		goto bail;
+	}
+
+	/*
+	 * Set transaction id's of transactions that have to be committed
+	 * to finish f[data]sync. We set them to currently running transaction
+	 * as we cannot be sure that the inode or some of its metadata isn't
+	 * part of the transaction - the inode could have been reclaimed and
+	 * now it is reread from disk.
+	 */
+	if (journal) {
+		transaction_t *transaction;
+		tid_t tid;
+		struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+		read_lock(&journal->j_state_lock);
+		if (journal->j_running_transaction)
+			transaction = journal->j_running_transaction;
+		else
+			transaction = journal->j_committing_transaction;
+		if (transaction)
+			tid = transaction->t_tid;
+		else
+			tid = journal->j_commit_sequence;
+		read_unlock(&journal->j_state_lock);
+		oi->i_sync_tid = tid;
+		oi->i_datasync_tid = tid;
+	}
+
+bail:
+	if (!IS_ERR(inode)) {
+		trace_ocfs2_iget_end(inode, 
+			(unsigned long long)OCFS2_I(inode)->ip_blkno);
+	}
+
+	return inode;
+}
+
+
+/*
+ * here's how inodes get read from disk:
+ * iget5_locked -> find_actor -> OCFS2_FIND_ACTOR
+ * found? : return the in-memory inode
+ * not found? : get_new_inode -> OCFS2_INIT_LOCKED_INODE
+ */
+
+static int ocfs2_find_actor(struct inode *inode, void *opaque)
+{
+	struct ocfs2_find_inode_args *args = NULL;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	int ret = 0;
+
+	args = opaque;
+
+	mlog_bug_on_msg(!inode, "No inode in find actor!\n");
+
+	trace_ocfs2_find_actor(inode, inode->i_ino, opaque, args->fi_blkno);
+
+	if (oi->ip_blkno != args->fi_blkno)
+		goto bail;
+
+	ret = 1;
+bail:
+	return ret;
+}
+
+/*
+ * initialize the new inode, but don't do anything that would cause
+ * us to sleep.
+ * return 0 on success, 1 on failure
+ */
+static int ocfs2_init_locked_inode(struct inode *inode, void *opaque)
+{
+	struct ocfs2_find_inode_args *args = opaque;
+	static struct lock_class_key ocfs2_quota_ip_alloc_sem_key,
+				     ocfs2_file_ip_alloc_sem_key;
+
+	inode->i_ino = args->fi_ino;
+	OCFS2_I(inode)->ip_blkno = args->fi_blkno;
+	if (args->fi_sysfile_type != 0)
+		lockdep_set_class(&inode->i_mutex,
+			&ocfs2_sysfile_lock_key[args->fi_sysfile_type]);
+	if (args->fi_sysfile_type == USER_QUOTA_SYSTEM_INODE ||
+	    args->fi_sysfile_type == GROUP_QUOTA_SYSTEM_INODE ||
+	    args->fi_sysfile_type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
+	    args->fi_sysfile_type == LOCAL_GROUP_QUOTA_SYSTEM_INODE)
+		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
+				  &ocfs2_quota_ip_alloc_sem_key);
+	else
+		lockdep_set_class(&OCFS2_I(inode)->ip_alloc_sem,
+				  &ocfs2_file_ip_alloc_sem_key);
+
+	return 0;
+}
+
+void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
+			  int create_ino)
+{
+	struct super_block *sb;
+	struct ocfs2_super *osb;
+	int use_plocks = 1;
+
+	sb = inode->i_sb;
+	osb = OCFS2_SB(sb);
+
+	if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
+	    ocfs2_mount_local(osb) || !ocfs2_stack_supports_plocks())
+		use_plocks = 0;
+
+	/*
+	 * These have all been checked by ocfs2_read_inode_block() or set
+	 * by ocfs2_mknod_locked(), so a failure is a code bug.
+	 */
+	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));  /* This means that read_inode
+						cannot create a superblock
+						inode today.  change if
+						that is needed. */
+	BUG_ON(!(fe->i_flags & cpu_to_le32(OCFS2_VALID_FL)));
+	BUG_ON(le32_to_cpu(fe->i_fs_generation) != osb->fs_generation);
+
+
+	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
+	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
+	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
+
+	inode->i_version = 1;
+	inode->i_generation = le32_to_cpu(fe->i_generation);
+	inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
+	inode->i_mode = le16_to_cpu(fe->i_mode);
+	i_uid_write(inode, le32_to_cpu(fe->i_uid));
+	i_gid_write(inode, le32_to_cpu(fe->i_gid));
+
+	/* Fast symlinks will have i_size but no allocated clusters. */
+	if (S_ISLNK(inode->i_mode) && !fe->i_clusters) {
+		inode->i_blocks = 0;
+		inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
+	} else {
+		inode->i_blocks = ocfs2_inode_sector_count(inode);
+		inode->i_mapping->a_ops = &ocfs2_aops;
+	}
+	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
+	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
+	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
+	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
+	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
+	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
+
+	if (OCFS2_I(inode)->ip_blkno != le64_to_cpu(fe->i_blkno))
+		mlog(ML_ERROR,
+		     "ip_blkno %llu != i_blkno %llu!\n",
+		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+		     (unsigned long long)le64_to_cpu(fe->i_blkno));
+
+	set_nlink(inode, ocfs2_read_links_count(fe));
+
+	trace_ocfs2_populate_inode(OCFS2_I(inode)->ip_blkno,
+				   le32_to_cpu(fe->i_flags));
+	if (fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) {
+		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SYSTEM_FILE;
+		inode->i_flags |= S_NOQUOTA;
+	}
+  
+	if (fe->i_flags & cpu_to_le32(OCFS2_LOCAL_ALLOC_FL)) {
+		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
+	} else if (fe->i_flags & cpu_to_le32(OCFS2_BITMAP_FL)) {
+		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_BITMAP;
+	} else if (fe->i_flags & cpu_to_le32(OCFS2_QUOTA_FL)) {
+		inode->i_flags |= S_NOQUOTA;
+	} else if (fe->i_flags & cpu_to_le32(OCFS2_SUPER_BLOCK_FL)) {
+		/* we can't actually hit this as read_inode can't
+		 * handle superblocks today ;-) */
+		BUG();
+	}
+
+	switch (inode->i_mode & S_IFMT) {
+	    case S_IFREG:
+		    if (use_plocks)
+			    inode->i_fop = &ocfs2_fops;
+		    else
+			    inode->i_fop = &ocfs2_fops_no_plocks;
+		    inode->i_op = &ocfs2_file_iops;
+		    i_size_write(inode, le64_to_cpu(fe->i_size));
+		    break;
+	    case S_IFDIR:
+		    inode->i_op = &ocfs2_dir_iops;
+		    if (use_plocks)
+			    inode->i_fop = &ocfs2_dops;
+		    else
+			    inode->i_fop = &ocfs2_dops_no_plocks;
+		    i_size_write(inode, le64_to_cpu(fe->i_size));
+		    OCFS2_I(inode)->ip_dir_lock_gen = 1;
+		    break;
+	    case S_IFLNK:
+		    inode->i_op = &ocfs2_symlink_inode_operations;
+		    i_size_write(inode, le64_to_cpu(fe->i_size));
+		    break;
+	    default:
+		    inode->i_op = &ocfs2_special_file_iops;
+		    init_special_inode(inode, inode->i_mode,
+				       inode->i_rdev);
+		    break;
+	}
+
+	if (create_ino) {
+		inode->i_ino = ino_from_blkno(inode->i_sb,
+			       le64_to_cpu(fe->i_blkno));
+
+		/*
+		 * If we ever want to create system files from kernel,
+		 * the generation argument to
+		 * ocfs2_inode_lock_res_init() will have to change.
+		 */
+		BUG_ON(le32_to_cpu(fe->i_flags) & OCFS2_SYSTEM_FL);
+
+		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
+					  OCFS2_LOCK_TYPE_META, 0, inode);
+
+		ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
+					  OCFS2_LOCK_TYPE_OPEN, 0, inode);
+	}
+
+	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_rw_lockres,
+				  OCFS2_LOCK_TYPE_RW, inode->i_generation,
+				  inode);
+
+	ocfs2_set_inode_flags(inode);
+
+	OCFS2_I(inode)->ip_last_used_slot = 0;
+	OCFS2_I(inode)->ip_last_used_group = 0;
+
+	if (S_ISDIR(inode->i_mode))
+		ocfs2_resv_set_type(&OCFS2_I(inode)->ip_la_data_resv,
+				    OCFS2_RESV_FLAG_DIR);
+}
+
+static int ocfs2_read_locked_inode(struct inode *inode,
+				   struct ocfs2_find_inode_args *args)
+{
+	struct super_block *sb;
+	struct ocfs2_super *osb;
+	struct ocfs2_dinode *fe;
+	struct buffer_head *bh = NULL;
+	int status, can_lock;
+	u32 generation = 0;
+
+	status = -EINVAL;
+	sb = inode->i_sb;
+	osb = OCFS2_SB(sb);
+
+	/*
+	 * To improve performance of cold-cache inode stats, we take
+	 * the cluster lock here if possible.
+	 *
+	 * Generally, OCFS2 never trusts the contents of an inode
+	 * unless it's holding a cluster lock, so taking it here isn't
+	 * a correctness issue as much as it is a performance
+	 * improvement.
+	 *
+	 * There are three times when taking the lock is not a good idea:
+	 *
+	 * 1) During startup, before we have initialized the DLM.
+	 *
+	 * 2) If we are reading certain system files which never get
+	 *    cluster locks (local alloc, truncate log).
+	 *
+	 * 3) If the process doing the iget() is responsible for
+	 *    orphan dir recovery. We're holding the orphan dir lock and
+	 *    can get into a deadlock with another process on another
+	 *    node in ->delete_inode().
+	 *
+	 * #1 and #2 can be simply solved by never taking the lock
+	 * here for system files (which are the only type we read
+	 * during mount). It's a heavier approach, but our main
+	 * concern is user-accessible files anyway.
+	 *
+	 * #3 works itself out because we'll eventually take the
+	 * cluster lock before trusting anything anyway.
+	 */
+	can_lock = !(args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
+		&& !(args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY)
+		&& !ocfs2_mount_local(osb);
+
+	trace_ocfs2_read_locked_inode(
+		(unsigned long long)OCFS2_I(inode)->ip_blkno, can_lock);
+
+	/*
+	 * To maintain backwards compatibility with older versions of
+	 * ocfs2-tools, we still store the generation value for system
+	 * files. The only ones that actually matter to userspace are
+	 * the journals, but it's easier and inexpensive to just flag
+	 * all system files similarly.
+	 */
+	if (args->fi_flags & OCFS2_FI_FLAG_SYSFILE)
+		generation = osb->fs_generation;
+
+	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_inode_lockres,
+				  OCFS2_LOCK_TYPE_META,
+				  generation, inode);
+
+	ocfs2_inode_lock_res_init(&OCFS2_I(inode)->ip_open_lockres,
+				  OCFS2_LOCK_TYPE_OPEN,
+				  0, inode);
+
+	if (can_lock) {
+		status = ocfs2_open_lock(inode);
+		if (status) {
+			make_bad_inode(inode);
+			mlog_errno(status);
+			return status;
+		}
+		status = ocfs2_inode_lock(inode, NULL, 0);
+		if (status) {
+			make_bad_inode(inode);
+			mlog_errno(status);
+			return status;
+		}
+	}
+
+	if (args->fi_flags & OCFS2_FI_FLAG_ORPHAN_RECOVERY) {
+		status = ocfs2_try_open_lock(inode, 0);
+		if (status) {
+			make_bad_inode(inode);
+			return status;
+		}
+	}
+
+	if (can_lock) {
+		status = ocfs2_read_inode_block_full(inode, &bh,
+						     OCFS2_BH_IGNORE_CACHE);
+	} else {
+		status = ocfs2_read_blocks_sync(osb, args->fi_blkno, 1, &bh);
+		/*
+		 * If buffer is in jbd, then its checksum may not have been
+		 * computed as yet.
+		 */
+		if (!status && !buffer_jbd(bh))
+			status = ocfs2_validate_inode_block(osb->sb, bh);
+	}
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = -EINVAL;
+	fe = (struct ocfs2_dinode *) bh->b_data;
+
+	/*
+	 * This is a code bug. Right now the caller needs to
+	 * understand whether it is asking for a system file inode or
+	 * not so the proper lock names can be built.
+	 */
+	mlog_bug_on_msg(!!(fe->i_flags & cpu_to_le32(OCFS2_SYSTEM_FL)) !=
+			!!(args->fi_flags & OCFS2_FI_FLAG_SYSFILE),
+			"Inode %llu: system file state is ambigous\n",
+			(unsigned long long)args->fi_blkno);
+
+	if (S_ISCHR(le16_to_cpu(fe->i_mode)) ||
+	    S_ISBLK(le16_to_cpu(fe->i_mode)))
+		inode->i_rdev = huge_decode_dev(le64_to_cpu(fe->id1.dev1.i_rdev));
+
+	ocfs2_populate_inode(inode, fe, 0);
+
+	BUG_ON(args->fi_blkno != le64_to_cpu(fe->i_blkno));
+
+	status = 0;
+
+bail:
+	if (can_lock)
+		ocfs2_inode_unlock(inode, 0);
+
+	if (status < 0)
+		make_bad_inode(inode);
+
+	brelse(bh);
+
+	return status;
+}
+
+void ocfs2_sync_blockdev(struct super_block *sb)
+{
+	sync_blockdev(sb->s_bdev);
+}
+
+static int ocfs2_truncate_for_delete(struct ocfs2_super *osb,
+				     struct inode *inode,
+				     struct buffer_head *fe_bh)
+{
+	int status = 0;
+	struct ocfs2_dinode *fe;
+	handle_t *handle = NULL;
+
+	fe = (struct ocfs2_dinode *) fe_bh->b_data;
+
+	/*
+	 * This check will also skip truncate of inodes with inline
+	 * data and fast symlinks.
+	 */
+	if (fe->i_clusters) {
+		if (ocfs2_should_order_data(inode))
+			ocfs2_begin_ordered_truncate(inode, 0);
+
+		handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+		if (IS_ERR(handle)) {
+			status = PTR_ERR(handle);
+			handle = NULL;
+			mlog_errno(status);
+			goto out;
+		}
+
+		status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
+						 fe_bh,
+						 OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out;
+		}
+
+		i_size_write(inode, 0);
+
+		status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out;
+		}
+
+		ocfs2_commit_trans(osb, handle);
+		handle = NULL;
+
+		status = ocfs2_commit_truncate(osb, inode, fe_bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out;
+		}
+	}
+
+out:
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+	return status;
+}
+
+static int ocfs2_remove_inode(struct inode *inode,
+			      struct buffer_head *di_bh,
+			      struct inode *orphan_dir_inode,
+			      struct buffer_head *orphan_dir_bh)
+{
+	int status;
+	struct inode *inode_alloc_inode = NULL;
+	struct buffer_head *inode_alloc_bh = NULL;
+	handle_t *handle;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
+
+	inode_alloc_inode =
+		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
+					    le16_to_cpu(di->i_suballoc_slot));
+	if (!inode_alloc_inode) {
+		status = -ENOENT;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	mutex_lock(&inode_alloc_inode->i_mutex);
+	status = ocfs2_inode_lock(inode_alloc_inode, &inode_alloc_bh, 1);
+	if (status < 0) {
+		mutex_unlock(&inode_alloc_inode->i_mutex);
+
+		mlog_errno(status);
+		goto bail;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_DELETE_INODE_CREDITS +
+				   ocfs2_quota_trans_credits(inode->i_sb));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto bail_unlock;
+	}
+
+	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
+		status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
+					  orphan_dir_bh, false);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail_commit;
+		}
+	}
+
+	/* set the inodes dtime */
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_commit;
+	}
+
+	di->i_dtime = cpu_to_le64(CURRENT_TIME.tv_sec);
+	di->i_flags &= cpu_to_le32(~(OCFS2_VALID_FL | OCFS2_ORPHANED_FL));
+	ocfs2_journal_dirty(handle, di_bh);
+
+	ocfs2_remove_from_cache(INODE_CACHE(inode), di_bh);
+	dquot_free_inode(inode);
+
+	status = ocfs2_free_dinode(handle, inode_alloc_inode,
+				   inode_alloc_bh, di);
+	if (status < 0)
+		mlog_errno(status);
+
+bail_commit:
+	ocfs2_commit_trans(osb, handle);
+bail_unlock:
+	ocfs2_inode_unlock(inode_alloc_inode, 1);
+	mutex_unlock(&inode_alloc_inode->i_mutex);
+	brelse(inode_alloc_bh);
+bail:
+	iput(inode_alloc_inode);
+
+	return status;
+}
+
+/*
+ * Serialize with orphan dir recovery. If the process doing
+ * recovery on this orphan dir does an iget() with the dir
+ * i_mutex held, we'll deadlock here. Instead we detect this
+ * and exit early - recovery will wipe this inode for us.
+ */
+static int ocfs2_check_orphan_recovery_state(struct ocfs2_super *osb,
+					     int slot)
+{
+	int ret = 0;
+
+	spin_lock(&osb->osb_lock);
+	if (ocfs2_node_map_test_bit(osb, &osb->osb_recovering_orphan_dirs, slot)) {
+		ret = -EDEADLK;
+		goto out;
+	}
+	/* This signals to the orphan recovery process that it should
+	 * wait for us to handle the wipe. */
+	osb->osb_orphan_wipes[slot]++;
+out:
+	spin_unlock(&osb->osb_lock);
+	trace_ocfs2_check_orphan_recovery_state(slot, ret);
+	return ret;
+}
+
+static void ocfs2_signal_wipe_completion(struct ocfs2_super *osb,
+					 int slot)
+{
+	spin_lock(&osb->osb_lock);
+	osb->osb_orphan_wipes[slot]--;
+	spin_unlock(&osb->osb_lock);
+
+	wake_up(&osb->osb_wipe_event);
+}
+
+static int ocfs2_wipe_inode(struct inode *inode,
+			    struct buffer_head *di_bh)
+{
+	int status, orphaned_slot = -1;
+	struct inode *orphan_dir_inode = NULL;
+	struct buffer_head *orphan_dir_bh = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
+
+	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
+		orphaned_slot = le16_to_cpu(di->i_orphaned_slot);
+
+		status = ocfs2_check_orphan_recovery_state(osb, orphaned_slot);
+		if (status)
+			return status;
+
+		orphan_dir_inode = ocfs2_get_system_file_inode(osb,
+							       ORPHAN_DIR_SYSTEM_INODE,
+							       orphaned_slot);
+		if (!orphan_dir_inode) {
+			status = -ENOENT;
+			mlog_errno(status);
+			goto bail;
+		}
+
+		/* Lock the orphan dir. The lock will be held for the entire
+		 * delete_inode operation. We do this now to avoid races with
+		 * recovery completion on other nodes. */
+		mutex_lock(&orphan_dir_inode->i_mutex);
+		status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
+		if (status < 0) {
+			mutex_unlock(&orphan_dir_inode->i_mutex);
+
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	/* we do this while holding the orphan dir lock because we
+	 * don't want recovery being run from another node to try an
+	 * inode delete underneath us -- this will result in two nodes
+	 * truncating the same file! */
+	status = ocfs2_truncate_for_delete(osb, inode, di_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_unlock_dir;
+	}
+
+	/* Remove any dir index tree */
+	if (S_ISDIR(inode->i_mode)) {
+		status = ocfs2_dx_dir_truncate(inode, di_bh);
+		if (status) {
+			mlog_errno(status);
+			goto bail_unlock_dir;
+		}
+	}
+
+	/*Free extended attribute resources associated with this inode.*/
+	status = ocfs2_xattr_remove(inode, di_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_unlock_dir;
+	}
+
+	status = ocfs2_remove_refcount_tree(inode, di_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_unlock_dir;
+	}
+
+	status = ocfs2_remove_inode(inode, di_bh, orphan_dir_inode,
+				    orphan_dir_bh);
+	if (status < 0)
+		mlog_errno(status);
+
+bail_unlock_dir:
+	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)
+		return status;
+
+	ocfs2_inode_unlock(orphan_dir_inode, 1);
+	mutex_unlock(&orphan_dir_inode->i_mutex);
+	brelse(orphan_dir_bh);
+bail:
+	iput(orphan_dir_inode);
+	ocfs2_signal_wipe_completion(osb, orphaned_slot);
+
+	return status;
+}
+
+/* There is a series of simple checks that should be done before a
+ * trylock is even considered. Encapsulate those in this function. */
+static int ocfs2_inode_is_valid_to_delete(struct inode *inode)
+{
+	int ret = 0;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	trace_ocfs2_inode_is_valid_to_delete(current, osb->dc_task,
+					     (unsigned long long)oi->ip_blkno,
+					     oi->ip_flags);
+
+	/* We shouldn't be getting here for the root directory
+	 * inode.. */
+	if (inode == osb->root_inode) {
+		mlog(ML_ERROR, "Skipping delete of root inode.\n");
+		goto bail;
+	}
+
+	/*
+	 * If we're coming from downconvert_thread we can't go into our own
+	 * voting [hello, deadlock city!] so we cannot delete the inode. But
+	 * since we dropped last inode ref when downconverting dentry lock,
+	 * we cannot have the file open and thus the node doing unlink will
+	 * take care of deleting the inode.
+	 */
+	if (current == osb->dc_task)
+		goto bail;
+
+	spin_lock(&oi->ip_lock);
+	/* OCFS2 *never* deletes system files. This should technically
+	 * never get here as system file inodes should always have a
+	 * positive link count. */
+	if (oi->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
+		mlog(ML_ERROR, "Skipping delete of system file %llu\n",
+		     (unsigned long long)oi->ip_blkno);
+		goto bail_unlock;
+	}
+
+	ret = 1;
+bail_unlock:
+	spin_unlock(&oi->ip_lock);
+bail:
+	return ret;
+}
+
+/* Query the cluster to determine whether we should wipe an inode from
+ * disk or not.
+ *
+ * Requires the inode to have the cluster lock. */
+static int ocfs2_query_inode_wipe(struct inode *inode,
+				  struct buffer_head *di_bh,
+				  int *wipe)
+{
+	int status = 0, reason = 0;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di;
+
+	*wipe = 0;
+
+	trace_ocfs2_query_inode_wipe_begin((unsigned long long)oi->ip_blkno,
+					   inode->i_nlink);
+
+	/* While we were waiting for the cluster lock in
+	 * ocfs2_delete_inode, another node might have asked to delete
+	 * the inode. Recheck our flags to catch this. */
+	if (!ocfs2_inode_is_valid_to_delete(inode)) {
+		reason = 1;
+		goto bail;
+	}
+
+	/* Now that we have an up to date inode, we can double check
+	 * the link count. */
+	if (inode->i_nlink)
+		goto bail;
+
+	/* Do some basic inode verification... */
+	di = (struct ocfs2_dinode *) di_bh->b_data;
+	if (!(di->i_flags & cpu_to_le32(OCFS2_ORPHANED_FL)) &&
+	    !(oi->ip_flags & OCFS2_INODE_SKIP_ORPHAN_DIR)) {
+		/*
+		 * Inodes in the orphan dir must have ORPHANED_FL.  The only
+		 * inodes that come back out of the orphan dir are reflink
+		 * targets. A reflink target may be moved out of the orphan
+		 * dir between the time we scan the directory and the time we
+		 * process it. This would lead to HAS_REFCOUNT_FL being set but
+		 * ORPHANED_FL not.
+		 */
+		if (di->i_dyn_features & cpu_to_le16(OCFS2_HAS_REFCOUNT_FL)) {
+			reason = 2;
+			goto bail;
+		}
+
+		/* for lack of a better error? */
+		status = -EEXIST;
+		mlog(ML_ERROR,
+		     "Inode %llu (on-disk %llu) not orphaned! "
+		     "Disk flags  0x%x, inode flags 0x%x\n",
+		     (unsigned long long)oi->ip_blkno,
+		     (unsigned long long)le64_to_cpu(di->i_blkno),
+		     le32_to_cpu(di->i_flags), oi->ip_flags);
+		goto bail;
+	}
+
+	/* has someone already deleted us?! baaad... */
+	if (di->i_dtime) {
+		status = -EEXIST;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/*
+	 * This is how ocfs2 determines whether an inode is still live
+	 * within the cluster. Every node takes a shared read lock on
+	 * the inode open lock in ocfs2_read_locked_inode(). When we
+	 * get to ->delete_inode(), each node tries to convert it's
+	 * lock to an exclusive. Trylocks are serialized by the inode
+	 * meta data lock. If the upconvert succeeds, we know the inode
+	 * is no longer live and can be deleted.
+	 *
+	 * Though we call this with the meta data lock held, the
+	 * trylock keeps us from ABBA deadlock.
+	 */
+	status = ocfs2_try_open_lock(inode, 1);
+	if (status == -EAGAIN) {
+		status = 0;
+		reason = 3;
+		goto bail;
+	}
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	*wipe = 1;
+	trace_ocfs2_query_inode_wipe_succ(le16_to_cpu(di->i_orphaned_slot));
+
+bail:
+	trace_ocfs2_query_inode_wipe_end(status, reason);
+	return status;
+}
+
+/* Support function for ocfs2_delete_inode. Will help us keep the
+ * inode data in a consistent state for clear_inode. Always truncates
+ * pages, optionally sync's them first. */
+static void ocfs2_cleanup_delete_inode(struct inode *inode,
+				       int sync_data)
+{
+	trace_ocfs2_cleanup_delete_inode(
+		(unsigned long long)OCFS2_I(inode)->ip_blkno, sync_data);
+	if (sync_data)
+		filemap_write_and_wait(inode->i_mapping);
+	truncate_inode_pages_final(&inode->i_data);
+}
+
+static void ocfs2_delete_inode(struct inode *inode)
+{
+	int wipe, status;
+	sigset_t oldset;
+	struct buffer_head *di_bh = NULL;
+
+	trace_ocfs2_delete_inode(inode->i_ino,
+				 (unsigned long long)OCFS2_I(inode)->ip_blkno,
+				 is_bad_inode(inode));
+
+	/* When we fail in read_inode() we mark inode as bad. The second test
+	 * catches the case when inode allocation fails before allocating
+	 * a block for inode. */
+	if (is_bad_inode(inode) || !OCFS2_I(inode)->ip_blkno)
+		goto bail;
+
+	if (!ocfs2_inode_is_valid_to_delete(inode)) {
+		/* It's probably not necessary to truncate_inode_pages
+		 * here but we do it for safety anyway (it will most
+		 * likely be a no-op anyway) */
+		ocfs2_cleanup_delete_inode(inode, 0);
+		goto bail;
+	}
+
+	dquot_initialize(inode);
+
+	/* We want to block signals in delete_inode as the lock and
+	 * messaging paths may return us -ERESTARTSYS. Which would
+	 * cause us to exit early, resulting in inodes being orphaned
+	 * forever. */
+	ocfs2_block_signals(&oldset);
+
+	/*
+	 * Synchronize us against ocfs2_get_dentry. We take this in
+	 * shared mode so that all nodes can still concurrently
+	 * process deletes.
+	 */
+	status = ocfs2_nfs_sync_lock(OCFS2_SB(inode->i_sb), 0);
+	if (status < 0) {
+		mlog(ML_ERROR, "getting nfs sync lock(PR) failed %d\n", status);
+		ocfs2_cleanup_delete_inode(inode, 0);
+		goto bail_unblock;
+	}
+	/* Lock down the inode. This gives us an up to date view of
+	 * it's metadata (for verification), and allows us to
+	 * serialize delete_inode on multiple nodes.
+	 *
+	 * Even though we might be doing a truncate, we don't take the
+	 * allocation lock here as it won't be needed - nobody will
+	 * have the file open.
+	 */
+	status = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		ocfs2_cleanup_delete_inode(inode, 0);
+		goto bail_unlock_nfs_sync;
+	}
+
+	/* Query the cluster. This will be the final decision made
+	 * before we go ahead and wipe the inode. */
+	status = ocfs2_query_inode_wipe(inode, di_bh, &wipe);
+	if (!wipe || status < 0) {
+		/* Error and remote inode busy both mean we won't be
+		 * removing the inode, so they take almost the same
+		 * path. */
+		if (status < 0)
+			mlog_errno(status);
+
+		/* Someone in the cluster has disallowed a wipe of
+		 * this inode, or it was never completely
+		 * orphaned. Write out the pages and exit now. */
+		ocfs2_cleanup_delete_inode(inode, 1);
+		goto bail_unlock_inode;
+	}
+
+	ocfs2_cleanup_delete_inode(inode, 0);
+
+	status = ocfs2_wipe_inode(inode, di_bh);
+	if (status < 0) {
+		if (status != -EDEADLK)
+			mlog_errno(status);
+		goto bail_unlock_inode;
+	}
+
+	/*
+	 * Mark the inode as successfully deleted.
+	 *
+	 * This is important for ocfs2_clear_inode() as it will check
+	 * this flag and skip any checkpointing work
+	 *
+	 * ocfs2_stuff_meta_lvb() also uses this flag to invalidate
+	 * the LVB for other nodes.
+	 */
+	OCFS2_I(inode)->ip_flags |= OCFS2_INODE_DELETED;
+
+bail_unlock_inode:
+	ocfs2_inode_unlock(inode, 1);
+	brelse(di_bh);
+
+bail_unlock_nfs_sync:
+	ocfs2_nfs_sync_unlock(OCFS2_SB(inode->i_sb), 0);
+
+bail_unblock:
+	ocfs2_unblock_signals(&oldset);
+bail:
+	return;
+}
+
+static void ocfs2_clear_inode(struct inode *inode)
+{
+	int status;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	clear_inode(inode);
+	trace_ocfs2_clear_inode((unsigned long long)oi->ip_blkno,
+				inode->i_nlink);
+
+	mlog_bug_on_msg(OCFS2_SB(inode->i_sb) == NULL,
+			"Inode=%lu\n", inode->i_ino);
+
+	dquot_drop(inode);
+
+	/* To preven remote deletes we hold open lock before, now it
+	 * is time to unlock PR and EX open locks. */
+	ocfs2_open_unlock(inode);
+
+	/* Do these before all the other work so that we don't bounce
+	 * the downconvert thread while waiting to destroy the locks. */
+	ocfs2_mark_lockres_freeing(osb, &oi->ip_rw_lockres);
+	ocfs2_mark_lockres_freeing(osb, &oi->ip_inode_lockres);
+	ocfs2_mark_lockres_freeing(osb, &oi->ip_open_lockres);
+
+	ocfs2_resv_discard(&OCFS2_SB(inode->i_sb)->osb_la_resmap,
+			   &oi->ip_la_data_resv);
+	ocfs2_resv_init_once(&oi->ip_la_data_resv);
+
+	/* We very well may get a clear_inode before all an inodes
+	 * metadata has hit disk. Of course, we can't drop any cluster
+	 * locks until the journal has finished with it. The only
+	 * exception here are successfully wiped inodes - their
+	 * metadata can now be considered to be part of the system
+	 * inodes from which it came. */
+	if (!(OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED))
+		ocfs2_checkpoint_inode(inode);
+
+	mlog_bug_on_msg(!list_empty(&oi->ip_io_markers),
+			"Clear inode of %llu, inode has io markers\n",
+			(unsigned long long)oi->ip_blkno);
+
+	ocfs2_extent_map_trunc(inode, 0);
+
+	status = ocfs2_drop_inode_locks(inode);
+	if (status < 0)
+		mlog_errno(status);
+
+	ocfs2_lock_res_free(&oi->ip_rw_lockres);
+	ocfs2_lock_res_free(&oi->ip_inode_lockres);
+	ocfs2_lock_res_free(&oi->ip_open_lockres);
+
+	ocfs2_metadata_cache_exit(INODE_CACHE(inode));
+
+	mlog_bug_on_msg(INODE_CACHE(inode)->ci_num_cached,
+			"Clear inode of %llu, inode has %u cache items\n",
+			(unsigned long long)oi->ip_blkno,
+			INODE_CACHE(inode)->ci_num_cached);
+
+	mlog_bug_on_msg(!(INODE_CACHE(inode)->ci_flags & OCFS2_CACHE_FL_INLINE),
+			"Clear inode of %llu, inode has a bad flag\n",
+			(unsigned long long)oi->ip_blkno);
+
+	mlog_bug_on_msg(spin_is_locked(&oi->ip_lock),
+			"Clear inode of %llu, inode is locked\n",
+			(unsigned long long)oi->ip_blkno);
+
+	mlog_bug_on_msg(!mutex_trylock(&oi->ip_io_mutex),
+			"Clear inode of %llu, io_mutex is locked\n",
+			(unsigned long long)oi->ip_blkno);
+	mutex_unlock(&oi->ip_io_mutex);
+
+	/*
+	 * down_trylock() returns 0, down_write_trylock() returns 1
+	 * kernel 1, world 0
+	 */
+	mlog_bug_on_msg(!down_write_trylock(&oi->ip_alloc_sem),
+			"Clear inode of %llu, alloc_sem is locked\n",
+			(unsigned long long)oi->ip_blkno);
+	up_write(&oi->ip_alloc_sem);
+
+	mlog_bug_on_msg(oi->ip_open_count,
+			"Clear inode of %llu has open count %d\n",
+			(unsigned long long)oi->ip_blkno, oi->ip_open_count);
+
+	/* Clear all other flags. */
+	oi->ip_flags = 0;
+	oi->ip_dir_start_lookup = 0;
+	oi->ip_blkno = 0ULL;
+
+	/*
+	 * ip_jinode is used to track txns against this inode. We ensure that
+	 * the journal is flushed before journal shutdown. Thus it is safe to
+	 * have inodes get cleaned up after journal shutdown.
+	 */
+	jbd2_journal_release_jbd_inode(OCFS2_SB(inode->i_sb)->journal->j_journal,
+				       &oi->ip_jinode);
+}
+
+void ocfs2_evict_inode(struct inode *inode)
+{
+	if (!inode->i_nlink ||
+	    (OCFS2_I(inode)->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)) {
+		ocfs2_delete_inode(inode);
+	} else {
+		truncate_inode_pages_final(&inode->i_data);
+	}
+	ocfs2_clear_inode(inode);
+}
+
+/* Called under inode_lock, with no more references on the
+ * struct inode, so it's safe here to check the flags field
+ * and to manipulate i_nlink without any other locks. */
+int ocfs2_drop_inode(struct inode *inode)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	int res;
+
+	trace_ocfs2_drop_inode((unsigned long long)oi->ip_blkno,
+				inode->i_nlink, oi->ip_flags);
+
+	if (oi->ip_flags & OCFS2_INODE_MAYBE_ORPHANED)
+		res = 1;
+	else
+		res = generic_drop_inode(inode);
+
+	return res;
+}
+
+/*
+ * This is called from our getattr.
+ */
+int ocfs2_inode_revalidate(struct dentry *dentry)
+{
+	struct inode *inode = d_inode(dentry);
+	int status = 0;
+
+	trace_ocfs2_inode_revalidate(inode,
+		inode ? (unsigned long long)OCFS2_I(inode)->ip_blkno : 0ULL,
+		inode ? (unsigned long long)OCFS2_I(inode)->ip_flags : 0);
+
+	if (!inode) {
+		status = -ENOENT;
+		goto bail;
+	}
+
+	spin_lock(&OCFS2_I(inode)->ip_lock);
+	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_DELETED) {
+		spin_unlock(&OCFS2_I(inode)->ip_lock);
+		status = -ENOENT;
+		goto bail;
+	}
+	spin_unlock(&OCFS2_I(inode)->ip_lock);
+
+	/* Let ocfs2_inode_lock do the work of updating our struct
+	 * inode for us. */
+	status = ocfs2_inode_lock(inode, NULL, 0);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		goto bail;
+	}
+	ocfs2_inode_unlock(inode, 0);
+bail:
+	return status;
+}
+
+/*
+ * Updates a disk inode from a
+ * struct inode.
+ * Only takes ip_lock.
+ */
+int ocfs2_mark_inode_dirty(handle_t *handle,
+			   struct inode *inode,
+			   struct buffer_head *bh)
+{
+	int status;
+	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
+
+	trace_ocfs2_mark_inode_dirty((unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	spin_lock(&OCFS2_I(inode)->ip_lock);
+	fe->i_clusters = cpu_to_le32(OCFS2_I(inode)->ip_clusters);
+	ocfs2_get_inode_flags(OCFS2_I(inode));
+	fe->i_attr = cpu_to_le32(OCFS2_I(inode)->ip_attr);
+	fe->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
+	spin_unlock(&OCFS2_I(inode)->ip_lock);
+
+	fe->i_size = cpu_to_le64(i_size_read(inode));
+	ocfs2_set_links_count(fe, inode->i_nlink);
+	fe->i_uid = cpu_to_le32(i_uid_read(inode));
+	fe->i_gid = cpu_to_le32(i_gid_read(inode));
+	fe->i_mode = cpu_to_le16(inode->i_mode);
+	fe->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
+	fe->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
+	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
+	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+	fe->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
+	fe->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
+
+	ocfs2_journal_dirty(handle, bh);
+	ocfs2_update_inode_fsync_trans(handle, inode, 1);
+leave:
+	return status;
+}
+
+/*
+ *
+ * Updates a struct inode from a disk inode.
+ * does no i/o, only takes ip_lock.
+ */
+void ocfs2_refresh_inode(struct inode *inode,
+			 struct ocfs2_dinode *fe)
+{
+	spin_lock(&OCFS2_I(inode)->ip_lock);
+
+	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
+	OCFS2_I(inode)->ip_attr = le32_to_cpu(fe->i_attr);
+	OCFS2_I(inode)->ip_dyn_features = le16_to_cpu(fe->i_dyn_features);
+	ocfs2_set_inode_flags(inode);
+	i_size_write(inode, le64_to_cpu(fe->i_size));
+	set_nlink(inode, ocfs2_read_links_count(fe));
+	i_uid_write(inode, le32_to_cpu(fe->i_uid));
+	i_gid_write(inode, le32_to_cpu(fe->i_gid));
+	inode->i_mode = le16_to_cpu(fe->i_mode);
+	if (S_ISLNK(inode->i_mode) && le32_to_cpu(fe->i_clusters) == 0)
+		inode->i_blocks = 0;
+	else
+		inode->i_blocks = ocfs2_inode_sector_count(inode);
+	inode->i_atime.tv_sec = le64_to_cpu(fe->i_atime);
+	inode->i_atime.tv_nsec = le32_to_cpu(fe->i_atime_nsec);
+	inode->i_mtime.tv_sec = le64_to_cpu(fe->i_mtime);
+	inode->i_mtime.tv_nsec = le32_to_cpu(fe->i_mtime_nsec);
+	inode->i_ctime.tv_sec = le64_to_cpu(fe->i_ctime);
+	inode->i_ctime.tv_nsec = le32_to_cpu(fe->i_ctime_nsec);
+
+	spin_unlock(&OCFS2_I(inode)->ip_lock);
+}
+
+int ocfs2_validate_inode_block(struct super_block *sb,
+			       struct buffer_head *bh)
+{
+	int rc;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
+
+	trace_ocfs2_validate_inode_block((unsigned long long)bh->b_blocknr);
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	/*
+	 * If the ecc fails, we return the error but otherwise
+	 * leave the filesystem running.  We know any error is
+	 * local to this block.
+	 */
+	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &di->i_check);
+	if (rc) {
+		mlog(ML_ERROR, "Checksum failed for dinode %llu\n",
+		     (unsigned long long)bh->b_blocknr);
+		goto bail;
+	}
+
+	/*
+	 * Errors after here are fatal.
+	 */
+
+	rc = -EINVAL;
+
+	if (!OCFS2_IS_VALID_DINODE(di)) {
+		ocfs2_error(sb, "Invalid dinode #%llu: signature = %.*s\n",
+			    (unsigned long long)bh->b_blocknr, 7,
+			    di->i_signature);
+		goto bail;
+	}
+
+	if (le64_to_cpu(di->i_blkno) != bh->b_blocknr) {
+		ocfs2_error(sb, "Invalid dinode #%llu: i_blkno is %llu\n",
+			    (unsigned long long)bh->b_blocknr,
+			    (unsigned long long)le64_to_cpu(di->i_blkno));
+		goto bail;
+	}
+
+	if (!(di->i_flags & cpu_to_le32(OCFS2_VALID_FL))) {
+		ocfs2_error(sb,
+			    "Invalid dinode #%llu: OCFS2_VALID_FL not set\n",
+			    (unsigned long long)bh->b_blocknr);
+		goto bail;
+	}
+
+	if (le32_to_cpu(di->i_fs_generation) !=
+	    OCFS2_SB(sb)->fs_generation) {
+		ocfs2_error(sb,
+			    "Invalid dinode #%llu: fs_generation is %u\n",
+			    (unsigned long long)bh->b_blocknr,
+			    le32_to_cpu(di->i_fs_generation));
+		goto bail;
+	}
+
+	rc = 0;
+
+bail:
+	return rc;
+}
+
+int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
+				int flags)
+{
+	int rc;
+	struct buffer_head *tmp = *bh;
+
+	rc = ocfs2_read_blocks(INODE_CACHE(inode), OCFS2_I(inode)->ip_blkno,
+			       1, &tmp, flags, ocfs2_validate_inode_block);
+
+	/* If ocfs2_read_blocks() got us a new bh, pass it up. */
+	if (!rc && !*bh)
+		*bh = tmp;
+
+	return rc;
+}
+
+int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh)
+{
+	return ocfs2_read_inode_block_full(inode, bh, 0);
+}
+
+
+static u64 ocfs2_inode_cache_owner(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	return oi->ip_blkno;
+}
+
+static struct super_block *ocfs2_inode_cache_get_super(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	return oi->vfs_inode.i_sb;
+}
+
+static void ocfs2_inode_cache_lock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	spin_lock(&oi->ip_lock);
+}
+
+static void ocfs2_inode_cache_unlock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	spin_unlock(&oi->ip_lock);
+}
+
+static void ocfs2_inode_cache_io_lock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	mutex_lock(&oi->ip_io_mutex);
+}
+
+static void ocfs2_inode_cache_io_unlock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_inode_info *oi = cache_info_to_inode(ci);
+
+	mutex_unlock(&oi->ip_io_mutex);
+}
+
+const struct ocfs2_caching_operations ocfs2_inode_caching_ops = {
+	.co_owner		= ocfs2_inode_cache_owner,
+	.co_get_super		= ocfs2_inode_cache_get_super,
+	.co_cache_lock		= ocfs2_inode_cache_lock,
+	.co_cache_unlock	= ocfs2_inode_cache_unlock,
+	.co_io_lock		= ocfs2_inode_cache_io_lock,
+	.co_io_unlock		= ocfs2_inode_cache_io_unlock,
+};
+
diff --git a/fs/ocfs2/inode.h b/fs/ocfs2/inode.h
new file mode 100644
index 000000000..5e86b247c
--- /dev/null
+++ b/fs/ocfs2/inode.h
@@ -0,0 +1,190 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * inode.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_INODE_H
+#define OCFS2_INODE_H
+
+#include "extent_map.h"
+
+/* OCFS2 Inode Private Data */
+struct ocfs2_inode_info
+{
+	u64			ip_blkno;
+
+	struct ocfs2_lock_res		ip_rw_lockres;
+	struct ocfs2_lock_res		ip_inode_lockres;
+	struct ocfs2_lock_res		ip_open_lockres;
+
+	/* protects allocation changes on this inode. */
+	struct rw_semaphore		ip_alloc_sem;
+
+	/* protects extended attribute changes on this inode */
+	struct rw_semaphore		ip_xattr_sem;
+
+	/* Number of outstanding AIO's which are not page aligned */
+	struct mutex			ip_unaligned_aio;
+
+	/* These fields are protected by ip_lock */
+	spinlock_t			ip_lock;
+	u32				ip_open_count;
+	struct list_head		ip_io_markers;
+	u32				ip_clusters;
+
+	u16				ip_dyn_features;
+	struct mutex			ip_io_mutex;
+	u32				ip_flags; /* see below */
+	u32				ip_attr; /* inode attributes */
+
+	/* protected by recovery_lock. */
+	struct inode			*ip_next_orphan;
+
+	struct ocfs2_caching_info	ip_metadata_cache;
+	struct ocfs2_extent_map		ip_extent_map;
+	struct inode			vfs_inode;
+	struct jbd2_inode		ip_jinode;
+
+	u32				ip_dir_start_lookup;
+
+	/* Only valid if the inode is the dir. */
+	u32				ip_last_used_slot;
+	u64				ip_last_used_group;
+	u32				ip_dir_lock_gen;
+
+	struct ocfs2_alloc_reservation	ip_la_data_resv;
+
+	/*
+	 * Transactions that contain inode's metadata needed to complete
+	 * fsync and fdatasync, respectively.
+	 */
+	tid_t i_sync_tid;
+	tid_t i_datasync_tid;
+
+	wait_queue_head_t append_dio_wq;
+
+	struct dquot *i_dquot[MAXQUOTAS];
+};
+
+/*
+ * Flags for the ip_flags field
+ */
+/* System file inodes  */
+#define OCFS2_INODE_SYSTEM_FILE		0x00000001
+#define OCFS2_INODE_JOURNAL		0x00000002
+#define OCFS2_INODE_BITMAP		0x00000004
+/* This inode has been wiped from disk */
+#define OCFS2_INODE_DELETED		0x00000008
+/* Has the inode been orphaned on another node?
+ *
+ * This hints to ocfs2_drop_inode that it should clear i_nlink before
+ * continuing.
+ *
+ * We *only* set this on unlink vote from another node. If the inode
+ * was locally orphaned, then we're sure of the state and don't need
+ * to twiddle i_nlink later - it's either zero or not depending on
+ * whether our unlink succeeded. Otherwise we got this from a node
+ * whose intention was to orphan the inode, however he may have
+ * crashed, failed etc, so we let ocfs2_drop_inode zero the value and
+ * rely on ocfs2_delete_inode to sort things out under the proper
+ * cluster locks.
+ */
+#define OCFS2_INODE_MAYBE_ORPHANED	0x00000010
+/* Does someone have the file open O_DIRECT */
+#define OCFS2_INODE_OPEN_DIRECT		0x00000020
+/* Tell the inode wipe code it's not in orphan dir */
+#define OCFS2_INODE_SKIP_ORPHAN_DIR     0x00000040
+
+static inline struct ocfs2_inode_info *OCFS2_I(struct inode *inode)
+{
+	return container_of(inode, struct ocfs2_inode_info, vfs_inode);
+}
+
+#define INODE_JOURNAL(i) (OCFS2_I(i)->ip_flags & OCFS2_INODE_JOURNAL)
+#define SET_INODE_JOURNAL(i) (OCFS2_I(i)->ip_flags |= OCFS2_INODE_JOURNAL)
+
+extern struct kmem_cache *ocfs2_inode_cache;
+
+extern const struct address_space_operations ocfs2_aops;
+extern const struct ocfs2_caching_operations ocfs2_inode_caching_ops;
+
+static inline struct ocfs2_caching_info *INODE_CACHE(struct inode *inode)
+{
+	return &OCFS2_I(inode)->ip_metadata_cache;
+}
+
+void ocfs2_evict_inode(struct inode *inode);
+int ocfs2_drop_inode(struct inode *inode);
+
+/* Flags for ocfs2_iget() */
+#define OCFS2_FI_FLAG_SYSFILE		0x1
+#define OCFS2_FI_FLAG_ORPHAN_RECOVERY	0x2
+struct inode *ocfs2_ilookup(struct super_block *sb, u64 feoff);
+struct inode *ocfs2_iget(struct ocfs2_super *osb, u64 feoff, unsigned flags,
+			 int sysfile_type);
+int ocfs2_inode_init_private(struct inode *inode);
+int ocfs2_inode_revalidate(struct dentry *dentry);
+void ocfs2_populate_inode(struct inode *inode, struct ocfs2_dinode *fe,
+			  int create_ino);
+void ocfs2_read_inode(struct inode *inode);
+void ocfs2_read_inode2(struct inode *inode, void *opaque);
+ssize_t ocfs2_rw_direct(int rw, struct file *filp, char *buf,
+			size_t size, loff_t *offp);
+void ocfs2_sync_blockdev(struct super_block *sb);
+void ocfs2_refresh_inode(struct inode *inode,
+			 struct ocfs2_dinode *fe);
+int ocfs2_mark_inode_dirty(handle_t *handle,
+			   struct inode *inode,
+			   struct buffer_head *bh);
+struct buffer_head *ocfs2_bread(struct inode *inode,
+				int block, int *err, int reada);
+
+void ocfs2_set_inode_flags(struct inode *inode);
+void ocfs2_get_inode_flags(struct ocfs2_inode_info *oi);
+
+static inline blkcnt_t ocfs2_inode_sector_count(struct inode *inode)
+{
+	int c_to_s_bits = OCFS2_SB(inode->i_sb)->s_clustersize_bits - 9;
+
+	return (blkcnt_t)OCFS2_I(inode)->ip_clusters << c_to_s_bits;
+}
+
+/* Validate that a bh contains a valid inode */
+int ocfs2_validate_inode_block(struct super_block *sb,
+			       struct buffer_head *bh);
+/*
+ * Read an inode block into *bh.  If *bh is NULL, a bh will be allocated.
+ * This is a cached read.  The inode will be validated with
+ * ocfs2_validate_inode_block().
+ */
+int ocfs2_read_inode_block(struct inode *inode, struct buffer_head **bh);
+/* The same, but can be passed OCFS2_BH_* flags */
+int ocfs2_read_inode_block_full(struct inode *inode, struct buffer_head **bh,
+				int flags);
+
+static inline struct ocfs2_inode_info *cache_info_to_inode(struct ocfs2_caching_info *ci)
+{
+	return container_of(ci, struct ocfs2_inode_info, ip_metadata_cache);
+}
+
+#endif /* OCFS2_INODE_H */
diff --git a/fs/ocfs2/ioctl.c b/fs/ocfs2/ioctl.c
new file mode 100644
index 000000000..53e6c40ed
--- /dev/null
+++ b/fs/ocfs2/ioctl.c
@@ -0,0 +1,1005 @@
+/*
+ * linux/fs/ocfs2/ioctl.c
+ *
+ * Copyright (C) 2006 Herbert Poetzl
+ * adapted from Remy Card's ext2/ioctl.c
+ */
+
+#include <linux/fs.h>
+#include <linux/mount.h>
+#include <linux/blkdev.h>
+#include <linux/compat.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+#include "alloc.h"
+#include "dlmglue.h"
+#include "file.h"
+#include "inode.h"
+#include "journal.h"
+
+#include "ocfs2_fs.h"
+#include "ioctl.h"
+#include "resize.h"
+#include "refcounttree.h"
+#include "sysfile.h"
+#include "dir.h"
+#include "buffer_head_io.h"
+#include "suballoc.h"
+#include "move_extents.h"
+
+#define o2info_from_user(a, b)	\
+		copy_from_user(&(a), (b), sizeof(a))
+#define o2info_to_user(a, b)	\
+		copy_to_user((typeof(a) __user *)b, &(a), sizeof(a))
+
+/*
+ * This is just a best-effort to tell userspace that this request
+ * caused the error.
+ */
+static inline void o2info_set_request_error(struct ocfs2_info_request *kreq,
+					struct ocfs2_info_request __user *req)
+{
+	kreq->ir_flags |= OCFS2_INFO_FL_ERROR;
+	(void)put_user(kreq->ir_flags, (__u32 __user *)&(req->ir_flags));
+}
+
+static inline void o2info_set_request_filled(struct ocfs2_info_request *req)
+{
+	req->ir_flags |= OCFS2_INFO_FL_FILLED;
+}
+
+static inline void o2info_clear_request_filled(struct ocfs2_info_request *req)
+{
+	req->ir_flags &= ~OCFS2_INFO_FL_FILLED;
+}
+
+static inline int o2info_coherent(struct ocfs2_info_request *req)
+{
+	return (!(req->ir_flags & OCFS2_INFO_FL_NON_COHERENT));
+}
+
+static int ocfs2_get_inode_attr(struct inode *inode, unsigned *flags)
+{
+	int status;
+
+	status = ocfs2_inode_lock(inode, NULL, 0);
+	if (status < 0) {
+		mlog_errno(status);
+		return status;
+	}
+	ocfs2_get_inode_flags(OCFS2_I(inode));
+	*flags = OCFS2_I(inode)->ip_attr;
+	ocfs2_inode_unlock(inode, 0);
+
+	return status;
+}
+
+static int ocfs2_set_inode_attr(struct inode *inode, unsigned flags,
+				unsigned mask)
+{
+	struct ocfs2_inode_info *ocfs2_inode = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	handle_t *handle = NULL;
+	struct buffer_head *bh = NULL;
+	unsigned oldflags;
+	int status;
+
+	mutex_lock(&inode->i_mutex);
+
+	status = ocfs2_inode_lock(inode, &bh, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = -EACCES;
+	if (!inode_owner_or_capable(inode))
+		goto bail_unlock;
+
+	if (!S_ISDIR(inode->i_mode))
+		flags &= ~OCFS2_DIRSYNC_FL;
+
+	oldflags = ocfs2_inode->ip_attr;
+	flags = flags & mask;
+	flags |= oldflags & ~mask;
+
+	/*
+	 * The IMMUTABLE and APPEND_ONLY flags can only be changed by
+	 * the relevant capability.
+	 */
+	status = -EPERM;
+	if ((oldflags & OCFS2_IMMUTABLE_FL) || ((flags ^ oldflags) &
+		(OCFS2_APPEND_FL | OCFS2_IMMUTABLE_FL))) {
+		if (!capable(CAP_LINUX_IMMUTABLE))
+			goto bail_unlock;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto bail_unlock;
+	}
+
+	ocfs2_inode->ip_attr = flags;
+	ocfs2_set_inode_flags(inode);
+
+	status = ocfs2_mark_inode_dirty(handle, inode, bh);
+	if (status < 0)
+		mlog_errno(status);
+
+	ocfs2_commit_trans(osb, handle);
+
+bail_unlock:
+	ocfs2_inode_unlock(inode, 1);
+bail:
+	mutex_unlock(&inode->i_mutex);
+
+	brelse(bh);
+
+	return status;
+}
+
+static int ocfs2_info_handle_blocksize(struct inode *inode,
+				       struct ocfs2_info_request __user *req)
+{
+	struct ocfs2_info_blocksize oib;
+
+	if (o2info_from_user(oib, req))
+		return -EFAULT;
+
+	oib.ib_blocksize = inode->i_sb->s_blocksize;
+
+	o2info_set_request_filled(&oib.ib_req);
+
+	if (o2info_to_user(oib, req))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int ocfs2_info_handle_clustersize(struct inode *inode,
+					 struct ocfs2_info_request __user *req)
+{
+	struct ocfs2_info_clustersize oic;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (o2info_from_user(oic, req))
+		return -EFAULT;
+
+	oic.ic_clustersize = osb->s_clustersize;
+
+	o2info_set_request_filled(&oic.ic_req);
+
+	if (o2info_to_user(oic, req))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int ocfs2_info_handle_maxslots(struct inode *inode,
+				      struct ocfs2_info_request __user *req)
+{
+	struct ocfs2_info_maxslots oim;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (o2info_from_user(oim, req))
+		return -EFAULT;
+
+	oim.im_max_slots = osb->max_slots;
+
+	o2info_set_request_filled(&oim.im_req);
+
+	if (o2info_to_user(oim, req))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int ocfs2_info_handle_label(struct inode *inode,
+				   struct ocfs2_info_request __user *req)
+{
+	struct ocfs2_info_label oil;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (o2info_from_user(oil, req))
+		return -EFAULT;
+
+	memcpy(oil.il_label, osb->vol_label, OCFS2_MAX_VOL_LABEL_LEN);
+
+	o2info_set_request_filled(&oil.il_req);
+
+	if (o2info_to_user(oil, req))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int ocfs2_info_handle_uuid(struct inode *inode,
+				  struct ocfs2_info_request __user *req)
+{
+	struct ocfs2_info_uuid oiu;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (o2info_from_user(oiu, req))
+		return -EFAULT;
+
+	memcpy(oiu.iu_uuid_str, osb->uuid_str, OCFS2_TEXT_UUID_LEN + 1);
+
+	o2info_set_request_filled(&oiu.iu_req);
+
+	if (o2info_to_user(oiu, req))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int ocfs2_info_handle_fs_features(struct inode *inode,
+					 struct ocfs2_info_request __user *req)
+{
+	struct ocfs2_info_fs_features oif;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (o2info_from_user(oif, req))
+		return -EFAULT;
+
+	oif.if_compat_features = osb->s_feature_compat;
+	oif.if_incompat_features = osb->s_feature_incompat;
+	oif.if_ro_compat_features = osb->s_feature_ro_compat;
+
+	o2info_set_request_filled(&oif.if_req);
+
+	if (o2info_to_user(oif, req))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int ocfs2_info_handle_journal_size(struct inode *inode,
+					  struct ocfs2_info_request __user *req)
+{
+	struct ocfs2_info_journal_size oij;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (o2info_from_user(oij, req))
+		return -EFAULT;
+
+	oij.ij_journal_size = i_size_read(osb->journal->j_inode);
+
+	o2info_set_request_filled(&oij.ij_req);
+
+	if (o2info_to_user(oij, req))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int ocfs2_info_scan_inode_alloc(struct ocfs2_super *osb,
+				       struct inode *inode_alloc, u64 blkno,
+				       struct ocfs2_info_freeinode *fi,
+				       u32 slot)
+{
+	int status = 0, unlock = 0;
+
+	struct buffer_head *bh = NULL;
+	struct ocfs2_dinode *dinode_alloc = NULL;
+
+	if (inode_alloc)
+		mutex_lock(&inode_alloc->i_mutex);
+
+	if (o2info_coherent(&fi->ifi_req)) {
+		status = ocfs2_inode_lock(inode_alloc, &bh, 0);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+		unlock = 1;
+	} else {
+		status = ocfs2_read_blocks_sync(osb, blkno, 1, &bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	dinode_alloc = (struct ocfs2_dinode *)bh->b_data;
+
+	fi->ifi_stat[slot].lfi_total =
+		le32_to_cpu(dinode_alloc->id1.bitmap1.i_total);
+	fi->ifi_stat[slot].lfi_free =
+		le32_to_cpu(dinode_alloc->id1.bitmap1.i_total) -
+		le32_to_cpu(dinode_alloc->id1.bitmap1.i_used);
+
+bail:
+	if (unlock)
+		ocfs2_inode_unlock(inode_alloc, 0);
+
+	if (inode_alloc)
+		mutex_unlock(&inode_alloc->i_mutex);
+
+	brelse(bh);
+
+	return status;
+}
+
+static int ocfs2_info_handle_freeinode(struct inode *inode,
+				       struct ocfs2_info_request __user *req)
+{
+	u32 i;
+	u64 blkno = -1;
+	char namebuf[40];
+	int status, type = INODE_ALLOC_SYSTEM_INODE;
+	struct ocfs2_info_freeinode *oifi = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct inode *inode_alloc = NULL;
+
+	oifi = kzalloc(sizeof(struct ocfs2_info_freeinode), GFP_KERNEL);
+	if (!oifi) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto out_err;
+	}
+
+	if (o2info_from_user(*oifi, req)) {
+		status = -EFAULT;
+		goto out_free;
+	}
+
+	oifi->ifi_slotnum = osb->max_slots;
+
+	for (i = 0; i < oifi->ifi_slotnum; i++) {
+		if (o2info_coherent(&oifi->ifi_req)) {
+			inode_alloc = ocfs2_get_system_file_inode(osb, type, i);
+			if (!inode_alloc) {
+				mlog(ML_ERROR, "unable to get alloc inode in "
+				    "slot %u\n", i);
+				status = -EIO;
+				goto bail;
+			}
+		} else {
+			ocfs2_sprintf_system_inode_name(namebuf,
+							sizeof(namebuf),
+							type, i);
+			status = ocfs2_lookup_ino_from_name(osb->sys_root_inode,
+							    namebuf,
+							    strlen(namebuf),
+							    &blkno);
+			if (status < 0) {
+				status = -ENOENT;
+				goto bail;
+			}
+		}
+
+		status = ocfs2_info_scan_inode_alloc(osb, inode_alloc, blkno, oifi, i);
+
+		iput(inode_alloc);
+		inode_alloc = NULL;
+
+		if (status < 0)
+			goto bail;
+	}
+
+	o2info_set_request_filled(&oifi->ifi_req);
+
+	if (o2info_to_user(*oifi, req)) {
+		status = -EFAULT;
+		goto out_free;
+	}
+
+	status = 0;
+bail:
+	if (status)
+		o2info_set_request_error(&oifi->ifi_req, req);
+out_free:
+	kfree(oifi);
+out_err:
+	return status;
+}
+
+static void o2ffg_update_histogram(struct ocfs2_info_free_chunk_list *hist,
+				   unsigned int chunksize)
+{
+	int index;
+
+	index = __ilog2_u32(chunksize);
+	if (index >= OCFS2_INFO_MAX_HIST)
+		index = OCFS2_INFO_MAX_HIST - 1;
+
+	hist->fc_chunks[index]++;
+	hist->fc_clusters[index] += chunksize;
+}
+
+static void o2ffg_update_stats(struct ocfs2_info_freefrag_stats *stats,
+			       unsigned int chunksize)
+{
+	if (chunksize > stats->ffs_max)
+		stats->ffs_max = chunksize;
+
+	if (chunksize < stats->ffs_min)
+		stats->ffs_min = chunksize;
+
+	stats->ffs_avg += chunksize;
+	stats->ffs_free_chunks_real++;
+}
+
+static void ocfs2_info_update_ffg(struct ocfs2_info_freefrag *ffg,
+				  unsigned int chunksize)
+{
+	o2ffg_update_histogram(&(ffg->iff_ffs.ffs_fc_hist), chunksize);
+	o2ffg_update_stats(&(ffg->iff_ffs), chunksize);
+}
+
+static int ocfs2_info_freefrag_scan_chain(struct ocfs2_super *osb,
+					  struct inode *gb_inode,
+					  struct ocfs2_dinode *gb_dinode,
+					  struct ocfs2_chain_rec *rec,
+					  struct ocfs2_info_freefrag *ffg,
+					  u32 chunks_in_group)
+{
+	int status = 0, used;
+	u64 blkno;
+
+	struct buffer_head *bh = NULL;
+	struct ocfs2_group_desc *bg = NULL;
+
+	unsigned int max_bits, num_clusters;
+	unsigned int offset = 0, cluster, chunk;
+	unsigned int chunk_free, last_chunksize = 0;
+
+	if (!le32_to_cpu(rec->c_free))
+		goto bail;
+
+	do {
+		if (!bg)
+			blkno = le64_to_cpu(rec->c_blkno);
+		else
+			blkno = le64_to_cpu(bg->bg_next_group);
+
+		if (bh) {
+			brelse(bh);
+			bh = NULL;
+		}
+
+		if (o2info_coherent(&ffg->iff_req))
+			status = ocfs2_read_group_descriptor(gb_inode,
+							     gb_dinode,
+							     blkno, &bh);
+		else
+			status = ocfs2_read_blocks_sync(osb, blkno, 1, &bh);
+
+		if (status < 0) {
+			mlog(ML_ERROR, "Can't read the group descriptor # "
+			     "%llu from device.", (unsigned long long)blkno);
+			status = -EIO;
+			goto bail;
+		}
+
+		bg = (struct ocfs2_group_desc *)bh->b_data;
+
+		if (!le16_to_cpu(bg->bg_free_bits_count))
+			continue;
+
+		max_bits = le16_to_cpu(bg->bg_bits);
+		offset = 0;
+
+		for (chunk = 0; chunk < chunks_in_group; chunk++) {
+			/*
+			 * last chunk may be not an entire one.
+			 */
+			if ((offset + ffg->iff_chunksize) > max_bits)
+				num_clusters = max_bits - offset;
+			else
+				num_clusters = ffg->iff_chunksize;
+
+			chunk_free = 0;
+			for (cluster = 0; cluster < num_clusters; cluster++) {
+				used = ocfs2_test_bit(offset,
+						(unsigned long *)bg->bg_bitmap);
+				/*
+				 * - chunk_free counts free clusters in #N chunk.
+				 * - last_chunksize records the size(in) clusters
+				 *   for the last real free chunk being counted.
+				 */
+				if (!used) {
+					last_chunksize++;
+					chunk_free++;
+				}
+
+				if (used && last_chunksize) {
+					ocfs2_info_update_ffg(ffg,
+							      last_chunksize);
+					last_chunksize = 0;
+				}
+
+				offset++;
+			}
+
+			if (chunk_free == ffg->iff_chunksize)
+				ffg->iff_ffs.ffs_free_chunks++;
+		}
+
+		/*
+		 * need to update the info for last free chunk.
+		 */
+		if (last_chunksize)
+			ocfs2_info_update_ffg(ffg, last_chunksize);
+
+	} while (le64_to_cpu(bg->bg_next_group));
+
+bail:
+	brelse(bh);
+
+	return status;
+}
+
+static int ocfs2_info_freefrag_scan_bitmap(struct ocfs2_super *osb,
+					   struct inode *gb_inode, u64 blkno,
+					   struct ocfs2_info_freefrag *ffg)
+{
+	u32 chunks_in_group;
+	int status = 0, unlock = 0, i;
+
+	struct buffer_head *bh = NULL;
+	struct ocfs2_chain_list *cl = NULL;
+	struct ocfs2_chain_rec *rec = NULL;
+	struct ocfs2_dinode *gb_dinode = NULL;
+
+	if (gb_inode)
+		mutex_lock(&gb_inode->i_mutex);
+
+	if (o2info_coherent(&ffg->iff_req)) {
+		status = ocfs2_inode_lock(gb_inode, &bh, 0);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+		unlock = 1;
+	} else {
+		status = ocfs2_read_blocks_sync(osb, blkno, 1, &bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	gb_dinode = (struct ocfs2_dinode *)bh->b_data;
+	cl = &(gb_dinode->id2.i_chain);
+
+	/*
+	 * Chunksize(in) clusters from userspace should be
+	 * less than clusters in a group.
+	 */
+	if (ffg->iff_chunksize > le16_to_cpu(cl->cl_cpg)) {
+		status = -EINVAL;
+		goto bail;
+	}
+
+	memset(&ffg->iff_ffs, 0, sizeof(struct ocfs2_info_freefrag_stats));
+
+	ffg->iff_ffs.ffs_min = ~0U;
+	ffg->iff_ffs.ffs_clusters =
+			le32_to_cpu(gb_dinode->id1.bitmap1.i_total);
+	ffg->iff_ffs.ffs_free_clusters = ffg->iff_ffs.ffs_clusters -
+			le32_to_cpu(gb_dinode->id1.bitmap1.i_used);
+
+	chunks_in_group = le16_to_cpu(cl->cl_cpg) / ffg->iff_chunksize + 1;
+
+	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i++) {
+		rec = &(cl->cl_recs[i]);
+		status = ocfs2_info_freefrag_scan_chain(osb, gb_inode,
+							gb_dinode,
+							rec, ffg,
+							chunks_in_group);
+		if (status)
+			goto bail;
+	}
+
+	if (ffg->iff_ffs.ffs_free_chunks_real)
+		ffg->iff_ffs.ffs_avg = (ffg->iff_ffs.ffs_avg /
+					ffg->iff_ffs.ffs_free_chunks_real);
+bail:
+	if (unlock)
+		ocfs2_inode_unlock(gb_inode, 0);
+
+	if (gb_inode)
+		mutex_unlock(&gb_inode->i_mutex);
+
+	if (gb_inode)
+		iput(gb_inode);
+
+	brelse(bh);
+
+	return status;
+}
+
+static int ocfs2_info_handle_freefrag(struct inode *inode,
+				      struct ocfs2_info_request __user *req)
+{
+	u64 blkno = -1;
+	char namebuf[40];
+	int status, type = GLOBAL_BITMAP_SYSTEM_INODE;
+
+	struct ocfs2_info_freefrag *oiff;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct inode *gb_inode = NULL;
+
+	oiff = kzalloc(sizeof(struct ocfs2_info_freefrag), GFP_KERNEL);
+	if (!oiff) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto out_err;
+	}
+
+	if (o2info_from_user(*oiff, req)) {
+		status = -EFAULT;
+		goto out_free;
+	}
+	/*
+	 * chunksize from userspace should be power of 2.
+	 */
+	if ((oiff->iff_chunksize & (oiff->iff_chunksize - 1)) ||
+	    (!oiff->iff_chunksize)) {
+		status = -EINVAL;
+		goto bail;
+	}
+
+	if (o2info_coherent(&oiff->iff_req)) {
+		gb_inode = ocfs2_get_system_file_inode(osb, type,
+						       OCFS2_INVALID_SLOT);
+		if (!gb_inode) {
+			mlog(ML_ERROR, "unable to get global_bitmap inode\n");
+			status = -EIO;
+			goto bail;
+		}
+	} else {
+		ocfs2_sprintf_system_inode_name(namebuf, sizeof(namebuf), type,
+						OCFS2_INVALID_SLOT);
+		status = ocfs2_lookup_ino_from_name(osb->sys_root_inode,
+						    namebuf,
+						    strlen(namebuf),
+						    &blkno);
+		if (status < 0) {
+			status = -ENOENT;
+			goto bail;
+		}
+	}
+
+	status = ocfs2_info_freefrag_scan_bitmap(osb, gb_inode, blkno, oiff);
+	if (status < 0)
+		goto bail;
+
+	o2info_set_request_filled(&oiff->iff_req);
+
+	if (o2info_to_user(*oiff, req)) {
+		status = -EFAULT;
+		goto out_free;
+	}
+
+	status = 0;
+bail:
+	if (status)
+		o2info_set_request_error(&oiff->iff_req, req);
+out_free:
+	kfree(oiff);
+out_err:
+	return status;
+}
+
+static int ocfs2_info_handle_unknown(struct inode *inode,
+				     struct ocfs2_info_request __user *req)
+{
+	struct ocfs2_info_request oir;
+
+	if (o2info_from_user(oir, req))
+		return -EFAULT;
+
+	o2info_clear_request_filled(&oir);
+
+	if (o2info_to_user(oir, req))
+		return -EFAULT;
+
+	return 0;
+}
+
+/*
+ * Validate and distinguish OCFS2_IOC_INFO requests.
+ *
+ * - validate the magic number.
+ * - distinguish different requests.
+ * - validate size of different requests.
+ */
+static int ocfs2_info_handle_request(struct inode *inode,
+				     struct ocfs2_info_request __user *req)
+{
+	int status = -EFAULT;
+	struct ocfs2_info_request oir;
+
+	if (o2info_from_user(oir, req))
+		goto bail;
+
+	status = -EINVAL;
+	if (oir.ir_magic != OCFS2_INFO_MAGIC)
+		goto bail;
+
+	switch (oir.ir_code) {
+	case OCFS2_INFO_BLOCKSIZE:
+		if (oir.ir_size == sizeof(struct ocfs2_info_blocksize))
+			status = ocfs2_info_handle_blocksize(inode, req);
+		break;
+	case OCFS2_INFO_CLUSTERSIZE:
+		if (oir.ir_size == sizeof(struct ocfs2_info_clustersize))
+			status = ocfs2_info_handle_clustersize(inode, req);
+		break;
+	case OCFS2_INFO_MAXSLOTS:
+		if (oir.ir_size == sizeof(struct ocfs2_info_maxslots))
+			status = ocfs2_info_handle_maxslots(inode, req);
+		break;
+	case OCFS2_INFO_LABEL:
+		if (oir.ir_size == sizeof(struct ocfs2_info_label))
+			status = ocfs2_info_handle_label(inode, req);
+		break;
+	case OCFS2_INFO_UUID:
+		if (oir.ir_size == sizeof(struct ocfs2_info_uuid))
+			status = ocfs2_info_handle_uuid(inode, req);
+		break;
+	case OCFS2_INFO_FS_FEATURES:
+		if (oir.ir_size == sizeof(struct ocfs2_info_fs_features))
+			status = ocfs2_info_handle_fs_features(inode, req);
+		break;
+	case OCFS2_INFO_JOURNAL_SIZE:
+		if (oir.ir_size == sizeof(struct ocfs2_info_journal_size))
+			status = ocfs2_info_handle_journal_size(inode, req);
+		break;
+	case OCFS2_INFO_FREEINODE:
+		if (oir.ir_size == sizeof(struct ocfs2_info_freeinode))
+			status = ocfs2_info_handle_freeinode(inode, req);
+		break;
+	case OCFS2_INFO_FREEFRAG:
+		if (oir.ir_size == sizeof(struct ocfs2_info_freefrag))
+			status = ocfs2_info_handle_freefrag(inode, req);
+		break;
+	default:
+		status = ocfs2_info_handle_unknown(inode, req);
+		break;
+	}
+
+bail:
+	return status;
+}
+
+static int ocfs2_get_request_ptr(struct ocfs2_info *info, int idx,
+				 u64 *req_addr, int compat_flag)
+{
+	int status = -EFAULT;
+	u64 __user *bp = NULL;
+
+	if (compat_flag) {
+#ifdef CONFIG_COMPAT
+		/*
+		 * pointer bp stores the base address of a pointers array,
+		 * which collects all addresses of separate request.
+		 */
+		bp = (u64 __user *)(unsigned long)compat_ptr(info->oi_requests);
+#else
+		BUG();
+#endif
+	} else
+		bp = (u64 __user *)(unsigned long)(info->oi_requests);
+
+	if (o2info_from_user(*req_addr, bp + idx))
+		goto bail;
+
+	status = 0;
+bail:
+	return status;
+}
+
+/*
+ * OCFS2_IOC_INFO handles an array of requests passed from userspace.
+ *
+ * ocfs2_info_handle() recevies a large info aggregation, grab and
+ * validate the request count from header, then break it into small
+ * pieces, later specific handlers can handle them one by one.
+ *
+ * Idea here is to make each separate request small enough to ensure
+ * a better backward&forward compatibility, since a small piece of
+ * request will be less likely to be broken if disk layout get changed.
+ */
+static int ocfs2_info_handle(struct inode *inode, struct ocfs2_info *info,
+			     int compat_flag)
+{
+	int i, status = 0;
+	u64 req_addr;
+	struct ocfs2_info_request __user *reqp;
+
+	if ((info->oi_count > OCFS2_INFO_MAX_REQUEST) ||
+	    (!info->oi_requests)) {
+		status = -EINVAL;
+		goto bail;
+	}
+
+	for (i = 0; i < info->oi_count; i++) {
+
+		status = ocfs2_get_request_ptr(info, i, &req_addr, compat_flag);
+		if (status)
+			break;
+
+		reqp = (struct ocfs2_info_request __user *)(unsigned long)req_addr;
+		if (!reqp) {
+			status = -EINVAL;
+			goto bail;
+		}
+
+		status = ocfs2_info_handle_request(inode, reqp);
+		if (status)
+			break;
+	}
+
+bail:
+	return status;
+}
+
+long ocfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
+{
+	struct inode *inode = file_inode(filp);
+	unsigned int flags;
+	int new_clusters;
+	int status;
+	struct ocfs2_space_resv sr;
+	struct ocfs2_new_group_input input;
+	struct reflink_arguments args;
+	const char __user *old_path;
+	const char __user *new_path;
+	bool preserve;
+	struct ocfs2_info info;
+	void __user *argp = (void __user *)arg;
+
+	switch (cmd) {
+	case OCFS2_IOC_GETFLAGS:
+		status = ocfs2_get_inode_attr(inode, &flags);
+		if (status < 0)
+			return status;
+
+		flags &= OCFS2_FL_VISIBLE;
+		return put_user(flags, (int __user *) arg);
+	case OCFS2_IOC_SETFLAGS:
+		if (get_user(flags, (int __user *) arg))
+			return -EFAULT;
+
+		status = mnt_want_write_file(filp);
+		if (status)
+			return status;
+		status = ocfs2_set_inode_attr(inode, flags,
+			OCFS2_FL_MODIFIABLE);
+		mnt_drop_write_file(filp);
+		return status;
+	case OCFS2_IOC_RESVSP:
+	case OCFS2_IOC_RESVSP64:
+	case OCFS2_IOC_UNRESVSP:
+	case OCFS2_IOC_UNRESVSP64:
+		if (copy_from_user(&sr, (int __user *) arg, sizeof(sr)))
+			return -EFAULT;
+
+		return ocfs2_change_file_space(filp, cmd, &sr);
+	case OCFS2_IOC_GROUP_EXTEND:
+		if (!capable(CAP_SYS_RESOURCE))
+			return -EPERM;
+
+		if (get_user(new_clusters, (int __user *)arg))
+			return -EFAULT;
+
+		status = mnt_want_write_file(filp);
+		if (status)
+			return status;
+		status = ocfs2_group_extend(inode, new_clusters);
+		mnt_drop_write_file(filp);
+		return status;
+	case OCFS2_IOC_GROUP_ADD:
+	case OCFS2_IOC_GROUP_ADD64:
+		if (!capable(CAP_SYS_RESOURCE))
+			return -EPERM;
+
+		if (copy_from_user(&input, (int __user *) arg, sizeof(input)))
+			return -EFAULT;
+
+		status = mnt_want_write_file(filp);
+		if (status)
+			return status;
+		status = ocfs2_group_add(inode, &input);
+		mnt_drop_write_file(filp);
+		return status;
+	case OCFS2_IOC_REFLINK:
+		if (copy_from_user(&args, argp, sizeof(args)))
+			return -EFAULT;
+		old_path = (const char __user *)(unsigned long)args.old_path;
+		new_path = (const char __user *)(unsigned long)args.new_path;
+		preserve = (args.preserve != 0);
+
+		return ocfs2_reflink_ioctl(inode, old_path, new_path, preserve);
+	case OCFS2_IOC_INFO:
+		if (copy_from_user(&info, argp, sizeof(struct ocfs2_info)))
+			return -EFAULT;
+
+		return ocfs2_info_handle(inode, &info, 0);
+	case FITRIM:
+	{
+		struct super_block *sb = inode->i_sb;
+		struct request_queue *q = bdev_get_queue(sb->s_bdev);
+		struct fstrim_range range;
+		int ret = 0;
+
+		if (!capable(CAP_SYS_ADMIN))
+			return -EPERM;
+
+		if (!blk_queue_discard(q))
+			return -EOPNOTSUPP;
+
+		if (copy_from_user(&range, argp, sizeof(range)))
+			return -EFAULT;
+
+		range.minlen = max_t(u64, q->limits.discard_granularity,
+				     range.minlen);
+		ret = ocfs2_trim_fs(sb, &range);
+		if (ret < 0)
+			return ret;
+
+		if (copy_to_user(argp, &range, sizeof(range)))
+			return -EFAULT;
+
+		return 0;
+	}
+	case OCFS2_IOC_MOVE_EXT:
+		return ocfs2_ioctl_move_extents(filp, argp);
+	default:
+		return -ENOTTY;
+	}
+}
+
+#ifdef CONFIG_COMPAT
+long ocfs2_compat_ioctl(struct file *file, unsigned cmd, unsigned long arg)
+{
+	bool preserve;
+	struct reflink_arguments args;
+	struct inode *inode = file_inode(file);
+	struct ocfs2_info info;
+	void __user *argp = (void __user *)arg;
+
+	switch (cmd) {
+	case OCFS2_IOC32_GETFLAGS:
+		cmd = OCFS2_IOC_GETFLAGS;
+		break;
+	case OCFS2_IOC32_SETFLAGS:
+		cmd = OCFS2_IOC_SETFLAGS;
+		break;
+	case OCFS2_IOC_RESVSP:
+	case OCFS2_IOC_RESVSP64:
+	case OCFS2_IOC_UNRESVSP:
+	case OCFS2_IOC_UNRESVSP64:
+	case OCFS2_IOC_GROUP_EXTEND:
+	case OCFS2_IOC_GROUP_ADD:
+	case OCFS2_IOC_GROUP_ADD64:
+	case FITRIM:
+		break;
+	case OCFS2_IOC_REFLINK:
+		if (copy_from_user(&args, argp, sizeof(args)))
+			return -EFAULT;
+		preserve = (args.preserve != 0);
+
+		return ocfs2_reflink_ioctl(inode, compat_ptr(args.old_path),
+					   compat_ptr(args.new_path), preserve);
+	case OCFS2_IOC_INFO:
+		if (copy_from_user(&info, argp, sizeof(struct ocfs2_info)))
+			return -EFAULT;
+
+		return ocfs2_info_handle(inode, &info, 1);
+	case OCFS2_IOC_MOVE_EXT:
+		break;
+	default:
+		return -ENOIOCTLCMD;
+	}
+
+	return ocfs2_ioctl(file, cmd, arg);
+}
+#endif
diff --git a/fs/ocfs2/ioctl.h b/fs/ocfs2/ioctl.h
new file mode 100644
index 000000000..0cd5323bd
--- /dev/null
+++ b/fs/ocfs2/ioctl.h
@@ -0,0 +1,16 @@
+/*
+ * ioctl.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2006 Herbert Poetzl
+ *
+ */
+
+#ifndef OCFS2_IOCTL_PROTO_H
+#define OCFS2_IOCTL_PROTO_H
+
+long ocfs2_ioctl(struct file *filp, unsigned int cmd, unsigned long arg);
+long ocfs2_compat_ioctl(struct file *file, unsigned cmd, unsigned long arg);
+
+#endif /* OCFS2_IOCTL_PROTO_H */
diff --git a/fs/ocfs2/journal.c b/fs/ocfs2/journal.c
new file mode 100644
index 000000000..ff5319282
--- /dev/null
+++ b/fs/ocfs2/journal.c
@@ -0,0 +1,2323 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * journal.c
+ *
+ * Defines functions of journalling api
+ *
+ * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/kthread.h>
+#include <linux/time.h>
+#include <linux/random.h>
+#include <linux/delay.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "blockcheck.h"
+#include "dir.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "heartbeat.h"
+#include "inode.h"
+#include "journal.h"
+#include "localalloc.h"
+#include "slot_map.h"
+#include "super.h"
+#include "sysfile.h"
+#include "uptodate.h"
+#include "quota.h"
+#include "file.h"
+#include "namei.h"
+
+#include "buffer_head_io.h"
+#include "ocfs2_trace.h"
+
+DEFINE_SPINLOCK(trans_inc_lock);
+
+#define ORPHAN_SCAN_SCHEDULE_TIMEOUT 300000
+
+static int ocfs2_force_read_journal(struct inode *inode);
+static int ocfs2_recover_node(struct ocfs2_super *osb,
+			      int node_num, int slot_num);
+static int __ocfs2_recovery_thread(void *arg);
+static int ocfs2_commit_cache(struct ocfs2_super *osb);
+static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota);
+static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
+				      int dirty, int replayed);
+static int ocfs2_trylock_journal(struct ocfs2_super *osb,
+				 int slot_num);
+static int ocfs2_recover_orphans(struct ocfs2_super *osb,
+				 int slot,
+				 enum ocfs2_orphan_reco_type orphan_reco_type);
+static int ocfs2_commit_thread(void *arg);
+static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
+					    int slot_num,
+					    struct ocfs2_dinode *la_dinode,
+					    struct ocfs2_dinode *tl_dinode,
+					    struct ocfs2_quota_recovery *qrec,
+					    enum ocfs2_orphan_reco_type orphan_reco_type);
+
+static inline int ocfs2_wait_on_mount(struct ocfs2_super *osb)
+{
+	return __ocfs2_wait_on_mount(osb, 0);
+}
+
+static inline int ocfs2_wait_on_quotas(struct ocfs2_super *osb)
+{
+	return __ocfs2_wait_on_mount(osb, 1);
+}
+
+/*
+ * This replay_map is to track online/offline slots, so we could recover
+ * offline slots during recovery and mount
+ */
+
+enum ocfs2_replay_state {
+	REPLAY_UNNEEDED = 0,	/* Replay is not needed, so ignore this map */
+	REPLAY_NEEDED, 		/* Replay slots marked in rm_replay_slots */
+	REPLAY_DONE 		/* Replay was already queued */
+};
+
+struct ocfs2_replay_map {
+	unsigned int rm_slots;
+	enum ocfs2_replay_state rm_state;
+	unsigned char rm_replay_slots[0];
+};
+
+void ocfs2_replay_map_set_state(struct ocfs2_super *osb, int state)
+{
+	if (!osb->replay_map)
+		return;
+
+	/* If we've already queued the replay, we don't have any more to do */
+	if (osb->replay_map->rm_state == REPLAY_DONE)
+		return;
+
+	osb->replay_map->rm_state = state;
+}
+
+int ocfs2_compute_replay_slots(struct ocfs2_super *osb)
+{
+	struct ocfs2_replay_map *replay_map;
+	int i, node_num;
+
+	/* If replay map is already set, we don't do it again */
+	if (osb->replay_map)
+		return 0;
+
+	replay_map = kzalloc(sizeof(struct ocfs2_replay_map) +
+			     (osb->max_slots * sizeof(char)), GFP_KERNEL);
+
+	if (!replay_map) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
+	spin_lock(&osb->osb_lock);
+
+	replay_map->rm_slots = osb->max_slots;
+	replay_map->rm_state = REPLAY_UNNEEDED;
+
+	/* set rm_replay_slots for offline slot(s) */
+	for (i = 0; i < replay_map->rm_slots; i++) {
+		if (ocfs2_slot_to_node_num_locked(osb, i, &node_num) == -ENOENT)
+			replay_map->rm_replay_slots[i] = 1;
+	}
+
+	osb->replay_map = replay_map;
+	spin_unlock(&osb->osb_lock);
+	return 0;
+}
+
+void ocfs2_queue_replay_slots(struct ocfs2_super *osb,
+		enum ocfs2_orphan_reco_type orphan_reco_type)
+{
+	struct ocfs2_replay_map *replay_map = osb->replay_map;
+	int i;
+
+	if (!replay_map)
+		return;
+
+	if (replay_map->rm_state != REPLAY_NEEDED)
+		return;
+
+	for (i = 0; i < replay_map->rm_slots; i++)
+		if (replay_map->rm_replay_slots[i])
+			ocfs2_queue_recovery_completion(osb->journal, i, NULL,
+							NULL, NULL,
+							orphan_reco_type);
+	replay_map->rm_state = REPLAY_DONE;
+}
+
+void ocfs2_free_replay_slots(struct ocfs2_super *osb)
+{
+	struct ocfs2_replay_map *replay_map = osb->replay_map;
+
+	if (!osb->replay_map)
+		return;
+
+	kfree(replay_map);
+	osb->replay_map = NULL;
+}
+
+int ocfs2_recovery_init(struct ocfs2_super *osb)
+{
+	struct ocfs2_recovery_map *rm;
+
+	mutex_init(&osb->recovery_lock);
+	osb->disable_recovery = 0;
+	osb->recovery_thread_task = NULL;
+	init_waitqueue_head(&osb->recovery_event);
+
+	rm = kzalloc(sizeof(struct ocfs2_recovery_map) +
+		     osb->max_slots * sizeof(unsigned int),
+		     GFP_KERNEL);
+	if (!rm) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
+	rm->rm_entries = (unsigned int *)((char *)rm +
+					  sizeof(struct ocfs2_recovery_map));
+	osb->recovery_map = rm;
+
+	return 0;
+}
+
+/* we can't grab the goofy sem lock from inside wait_event, so we use
+ * memory barriers to make sure that we'll see the null task before
+ * being woken up */
+static int ocfs2_recovery_thread_running(struct ocfs2_super *osb)
+{
+	mb();
+	return osb->recovery_thread_task != NULL;
+}
+
+void ocfs2_recovery_exit(struct ocfs2_super *osb)
+{
+	struct ocfs2_recovery_map *rm;
+
+	/* disable any new recovery threads and wait for any currently
+	 * running ones to exit. Do this before setting the vol_state. */
+	mutex_lock(&osb->recovery_lock);
+	osb->disable_recovery = 1;
+	mutex_unlock(&osb->recovery_lock);
+	wait_event(osb->recovery_event, !ocfs2_recovery_thread_running(osb));
+
+	/* At this point, we know that no more recovery threads can be
+	 * launched, so wait for any recovery completion work to
+	 * complete. */
+	flush_workqueue(ocfs2_wq);
+
+	/*
+	 * Now that recovery is shut down, and the osb is about to be
+	 * freed,  the osb_lock is not taken here.
+	 */
+	rm = osb->recovery_map;
+	/* XXX: Should we bug if there are dirty entries? */
+
+	kfree(rm);
+}
+
+static int __ocfs2_recovery_map_test(struct ocfs2_super *osb,
+				     unsigned int node_num)
+{
+	int i;
+	struct ocfs2_recovery_map *rm = osb->recovery_map;
+
+	assert_spin_locked(&osb->osb_lock);
+
+	for (i = 0; i < rm->rm_used; i++) {
+		if (rm->rm_entries[i] == node_num)
+			return 1;
+	}
+
+	return 0;
+}
+
+/* Behaves like test-and-set.  Returns the previous value */
+static int ocfs2_recovery_map_set(struct ocfs2_super *osb,
+				  unsigned int node_num)
+{
+	struct ocfs2_recovery_map *rm = osb->recovery_map;
+
+	spin_lock(&osb->osb_lock);
+	if (__ocfs2_recovery_map_test(osb, node_num)) {
+		spin_unlock(&osb->osb_lock);
+		return 1;
+	}
+
+	/* XXX: Can this be exploited? Not from o2dlm... */
+	BUG_ON(rm->rm_used >= osb->max_slots);
+
+	rm->rm_entries[rm->rm_used] = node_num;
+	rm->rm_used++;
+	spin_unlock(&osb->osb_lock);
+
+	return 0;
+}
+
+static void ocfs2_recovery_map_clear(struct ocfs2_super *osb,
+				     unsigned int node_num)
+{
+	int i;
+	struct ocfs2_recovery_map *rm = osb->recovery_map;
+
+	spin_lock(&osb->osb_lock);
+
+	for (i = 0; i < rm->rm_used; i++) {
+		if (rm->rm_entries[i] == node_num)
+			break;
+	}
+
+	if (i < rm->rm_used) {
+		/* XXX: be careful with the pointer math */
+		memmove(&(rm->rm_entries[i]), &(rm->rm_entries[i + 1]),
+			(rm->rm_used - i - 1) * sizeof(unsigned int));
+		rm->rm_used--;
+	}
+
+	spin_unlock(&osb->osb_lock);
+}
+
+static int ocfs2_commit_cache(struct ocfs2_super *osb)
+{
+	int status = 0;
+	unsigned int flushed;
+	struct ocfs2_journal *journal = NULL;
+
+	journal = osb->journal;
+
+	/* Flush all pending commits and checkpoint the journal. */
+	down_write(&journal->j_trans_barrier);
+
+	flushed = atomic_read(&journal->j_num_trans);
+	trace_ocfs2_commit_cache_begin(flushed);
+	if (flushed == 0) {
+		up_write(&journal->j_trans_barrier);
+		goto finally;
+	}
+
+	jbd2_journal_lock_updates(journal->j_journal);
+	status = jbd2_journal_flush(journal->j_journal);
+	jbd2_journal_unlock_updates(journal->j_journal);
+	if (status < 0) {
+		up_write(&journal->j_trans_barrier);
+		mlog_errno(status);
+		goto finally;
+	}
+
+	ocfs2_inc_trans_id(journal);
+
+	flushed = atomic_read(&journal->j_num_trans);
+	atomic_set(&journal->j_num_trans, 0);
+	up_write(&journal->j_trans_barrier);
+
+	trace_ocfs2_commit_cache_end(journal->j_trans_id, flushed);
+
+	ocfs2_wake_downconvert_thread(osb);
+	wake_up(&journal->j_checkpointed);
+finally:
+	return status;
+}
+
+handle_t *ocfs2_start_trans(struct ocfs2_super *osb, int max_buffs)
+{
+	journal_t *journal = osb->journal->j_journal;
+	handle_t *handle;
+
+	BUG_ON(!osb || !osb->journal->j_journal);
+
+	if (ocfs2_is_hard_readonly(osb))
+		return ERR_PTR(-EROFS);
+
+	BUG_ON(osb->journal->j_state == OCFS2_JOURNAL_FREE);
+	BUG_ON(max_buffs <= 0);
+
+	/* Nested transaction? Just return the handle... */
+	if (journal_current_handle())
+		return jbd2_journal_start(journal, max_buffs);
+
+	sb_start_intwrite(osb->sb);
+
+	down_read(&osb->journal->j_trans_barrier);
+
+	handle = jbd2_journal_start(journal, max_buffs);
+	if (IS_ERR(handle)) {
+		up_read(&osb->journal->j_trans_barrier);
+		sb_end_intwrite(osb->sb);
+
+		mlog_errno(PTR_ERR(handle));
+
+		if (is_journal_aborted(journal)) {
+			ocfs2_abort(osb->sb, "Detected aborted journal");
+			handle = ERR_PTR(-EROFS);
+		}
+	} else {
+		if (!ocfs2_mount_local(osb))
+			atomic_inc(&(osb->journal->j_num_trans));
+	}
+
+	return handle;
+}
+
+int ocfs2_commit_trans(struct ocfs2_super *osb,
+		       handle_t *handle)
+{
+	int ret, nested;
+	struct ocfs2_journal *journal = osb->journal;
+
+	BUG_ON(!handle);
+
+	nested = handle->h_ref > 1;
+	ret = jbd2_journal_stop(handle);
+	if (ret < 0)
+		mlog_errno(ret);
+
+	if (!nested) {
+		up_read(&journal->j_trans_barrier);
+		sb_end_intwrite(osb->sb);
+	}
+
+	return ret;
+}
+
+/*
+ * 'nblocks' is what you want to add to the current transaction.
+ *
+ * This might call jbd2_journal_restart() which will commit dirty buffers
+ * and then restart the transaction. Before calling
+ * ocfs2_extend_trans(), any changed blocks should have been
+ * dirtied. After calling it, all blocks which need to be changed must
+ * go through another set of journal_access/journal_dirty calls.
+ *
+ * WARNING: This will not release any semaphores or disk locks taken
+ * during the transaction, so make sure they were taken *before*
+ * start_trans or we'll have ordering deadlocks.
+ *
+ * WARNING2: Note that we do *not* drop j_trans_barrier here. This is
+ * good because transaction ids haven't yet been recorded on the
+ * cluster locks associated with this handle.
+ */
+int ocfs2_extend_trans(handle_t *handle, int nblocks)
+{
+	int status, old_nblocks;
+
+	BUG_ON(!handle);
+	BUG_ON(nblocks < 0);
+
+	if (!nblocks)
+		return 0;
+
+	old_nblocks = handle->h_buffer_credits;
+
+	trace_ocfs2_extend_trans(old_nblocks, nblocks);
+
+#ifdef CONFIG_OCFS2_DEBUG_FS
+	status = 1;
+#else
+	status = jbd2_journal_extend(handle, nblocks);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+#endif
+
+	if (status > 0) {
+		trace_ocfs2_extend_trans_restart(old_nblocks + nblocks);
+		status = jbd2_journal_restart(handle,
+					      old_nblocks + nblocks);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	status = 0;
+bail:
+	return status;
+}
+
+/*
+ * If we have fewer than thresh credits, extend by OCFS2_MAX_TRANS_DATA.
+ * If that fails, restart the transaction & regain write access for the
+ * buffer head which is used for metadata modifications.
+ * Taken from Ext4: extend_or_restart_transaction()
+ */
+int ocfs2_allocate_extend_trans(handle_t *handle, int thresh)
+{
+	int status, old_nblks;
+
+	BUG_ON(!handle);
+
+	old_nblks = handle->h_buffer_credits;
+	trace_ocfs2_allocate_extend_trans(old_nblks, thresh);
+
+	if (old_nblks < thresh)
+		return 0;
+
+	status = jbd2_journal_extend(handle, OCFS2_MAX_TRANS_DATA);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	if (status > 0) {
+		status = jbd2_journal_restart(handle, OCFS2_MAX_TRANS_DATA);
+		if (status < 0)
+			mlog_errno(status);
+	}
+
+bail:
+	return status;
+}
+
+
+struct ocfs2_triggers {
+	struct jbd2_buffer_trigger_type	ot_triggers;
+	int				ot_offset;
+};
+
+static inline struct ocfs2_triggers *to_ocfs2_trigger(struct jbd2_buffer_trigger_type *triggers)
+{
+	return container_of(triggers, struct ocfs2_triggers, ot_triggers);
+}
+
+static void ocfs2_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
+				 struct buffer_head *bh,
+				 void *data, size_t size)
+{
+	struct ocfs2_triggers *ot = to_ocfs2_trigger(triggers);
+
+	/*
+	 * We aren't guaranteed to have the superblock here, so we
+	 * must unconditionally compute the ecc data.
+	 * __ocfs2_journal_access() will only set the triggers if
+	 * metaecc is enabled.
+	 */
+	ocfs2_block_check_compute(data, size, data + ot->ot_offset);
+}
+
+/*
+ * Quota blocks have their own trigger because the struct ocfs2_block_check
+ * offset depends on the blocksize.
+ */
+static void ocfs2_dq_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
+				 struct buffer_head *bh,
+				 void *data, size_t size)
+{
+	struct ocfs2_disk_dqtrailer *dqt =
+		ocfs2_block_dqtrailer(size, data);
+
+	/*
+	 * We aren't guaranteed to have the superblock here, so we
+	 * must unconditionally compute the ecc data.
+	 * __ocfs2_journal_access() will only set the triggers if
+	 * metaecc is enabled.
+	 */
+	ocfs2_block_check_compute(data, size, &dqt->dq_check);
+}
+
+/*
+ * Directory blocks also have their own trigger because the
+ * struct ocfs2_block_check offset depends on the blocksize.
+ */
+static void ocfs2_db_frozen_trigger(struct jbd2_buffer_trigger_type *triggers,
+				 struct buffer_head *bh,
+				 void *data, size_t size)
+{
+	struct ocfs2_dir_block_trailer *trailer =
+		ocfs2_dir_trailer_from_size(size, data);
+
+	/*
+	 * We aren't guaranteed to have the superblock here, so we
+	 * must unconditionally compute the ecc data.
+	 * __ocfs2_journal_access() will only set the triggers if
+	 * metaecc is enabled.
+	 */
+	ocfs2_block_check_compute(data, size, &trailer->db_check);
+}
+
+static void ocfs2_abort_trigger(struct jbd2_buffer_trigger_type *triggers,
+				struct buffer_head *bh)
+{
+	mlog(ML_ERROR,
+	     "ocfs2_abort_trigger called by JBD2.  bh = 0x%lx, "
+	     "bh->b_blocknr = %llu\n",
+	     (unsigned long)bh,
+	     (unsigned long long)bh->b_blocknr);
+
+	/* We aren't guaranteed to have the superblock here - but if we
+	 * don't, it'll just crash. */
+	ocfs2_error(bh->b_assoc_map->host->i_sb,
+		    "JBD2 has aborted our journal, ocfs2 cannot continue\n");
+}
+
+static struct ocfs2_triggers di_triggers = {
+	.ot_triggers = {
+		.t_frozen = ocfs2_frozen_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+	.ot_offset	= offsetof(struct ocfs2_dinode, i_check),
+};
+
+static struct ocfs2_triggers eb_triggers = {
+	.ot_triggers = {
+		.t_frozen = ocfs2_frozen_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+	.ot_offset	= offsetof(struct ocfs2_extent_block, h_check),
+};
+
+static struct ocfs2_triggers rb_triggers = {
+	.ot_triggers = {
+		.t_frozen = ocfs2_frozen_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+	.ot_offset	= offsetof(struct ocfs2_refcount_block, rf_check),
+};
+
+static struct ocfs2_triggers gd_triggers = {
+	.ot_triggers = {
+		.t_frozen = ocfs2_frozen_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+	.ot_offset	= offsetof(struct ocfs2_group_desc, bg_check),
+};
+
+static struct ocfs2_triggers db_triggers = {
+	.ot_triggers = {
+		.t_frozen = ocfs2_db_frozen_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+};
+
+static struct ocfs2_triggers xb_triggers = {
+	.ot_triggers = {
+		.t_frozen = ocfs2_frozen_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+	.ot_offset	= offsetof(struct ocfs2_xattr_block, xb_check),
+};
+
+static struct ocfs2_triggers dq_triggers = {
+	.ot_triggers = {
+		.t_frozen = ocfs2_dq_frozen_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+};
+
+static struct ocfs2_triggers dr_triggers = {
+	.ot_triggers = {
+		.t_frozen = ocfs2_frozen_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+	.ot_offset	= offsetof(struct ocfs2_dx_root_block, dr_check),
+};
+
+static struct ocfs2_triggers dl_triggers = {
+	.ot_triggers = {
+		.t_frozen = ocfs2_frozen_trigger,
+		.t_abort = ocfs2_abort_trigger,
+	},
+	.ot_offset	= offsetof(struct ocfs2_dx_leaf, dl_check),
+};
+
+static int __ocfs2_journal_access(handle_t *handle,
+				  struct ocfs2_caching_info *ci,
+				  struct buffer_head *bh,
+				  struct ocfs2_triggers *triggers,
+				  int type)
+{
+	int status;
+	struct ocfs2_super *osb =
+		OCFS2_SB(ocfs2_metadata_cache_get_super(ci));
+
+	BUG_ON(!ci || !ci->ci_ops);
+	BUG_ON(!handle);
+	BUG_ON(!bh);
+
+	trace_ocfs2_journal_access(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long)bh->b_blocknr, type, bh->b_size);
+
+	/* we can safely remove this assertion after testing. */
+	if (!buffer_uptodate(bh)) {
+		mlog(ML_ERROR, "giving me a buffer that's not uptodate!\n");
+		mlog(ML_ERROR, "b_blocknr=%llu\n",
+		     (unsigned long long)bh->b_blocknr);
+		BUG();
+	}
+
+	/* Set the current transaction information on the ci so
+	 * that the locking code knows whether it can drop it's locks
+	 * on this ci or not. We're protected from the commit
+	 * thread updating the current transaction id until
+	 * ocfs2_commit_trans() because ocfs2_start_trans() took
+	 * j_trans_barrier for us. */
+	ocfs2_set_ci_lock_trans(osb->journal, ci);
+
+	ocfs2_metadata_cache_io_lock(ci);
+	switch (type) {
+	case OCFS2_JOURNAL_ACCESS_CREATE:
+	case OCFS2_JOURNAL_ACCESS_WRITE:
+		status = jbd2_journal_get_write_access(handle, bh);
+		break;
+
+	case OCFS2_JOURNAL_ACCESS_UNDO:
+		status = jbd2_journal_get_undo_access(handle, bh);
+		break;
+
+	default:
+		status = -EINVAL;
+		mlog(ML_ERROR, "Unknown access type!\n");
+	}
+	if (!status && ocfs2_meta_ecc(osb) && triggers)
+		jbd2_journal_set_triggers(bh, &triggers->ot_triggers);
+	ocfs2_metadata_cache_io_unlock(ci);
+
+	if (status < 0)
+		mlog(ML_ERROR, "Error %d getting %d access to buffer!\n",
+		     status, type);
+
+	return status;
+}
+
+int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &di_triggers, type);
+}
+
+int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &eb_triggers, type);
+}
+
+int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &rb_triggers,
+				      type);
+}
+
+int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &gd_triggers, type);
+}
+
+int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &db_triggers, type);
+}
+
+int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &xb_triggers, type);
+}
+
+int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &dq_triggers, type);
+}
+
+int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &dr_triggers, type);
+}
+
+int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, &dl_triggers, type);
+}
+
+int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
+			 struct buffer_head *bh, int type)
+{
+	return __ocfs2_journal_access(handle, ci, bh, NULL, type);
+}
+
+void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh)
+{
+	int status;
+
+	trace_ocfs2_journal_dirty((unsigned long long)bh->b_blocknr);
+
+	status = jbd2_journal_dirty_metadata(handle, bh);
+	BUG_ON(status);
+}
+
+#define OCFS2_DEFAULT_COMMIT_INTERVAL	(HZ * JBD2_DEFAULT_MAX_COMMIT_AGE)
+
+void ocfs2_set_journal_params(struct ocfs2_super *osb)
+{
+	journal_t *journal = osb->journal->j_journal;
+	unsigned long commit_interval = OCFS2_DEFAULT_COMMIT_INTERVAL;
+
+	if (osb->osb_commit_interval)
+		commit_interval = osb->osb_commit_interval;
+
+	write_lock(&journal->j_state_lock);
+	journal->j_commit_interval = commit_interval;
+	if (osb->s_mount_opt & OCFS2_MOUNT_BARRIER)
+		journal->j_flags |= JBD2_BARRIER;
+	else
+		journal->j_flags &= ~JBD2_BARRIER;
+	write_unlock(&journal->j_state_lock);
+}
+
+int ocfs2_journal_init(struct ocfs2_journal *journal, int *dirty)
+{
+	int status = -1;
+	struct inode *inode = NULL; /* the journal inode */
+	journal_t *j_journal = NULL;
+	struct ocfs2_dinode *di = NULL;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_super *osb;
+	int inode_lock = 0;
+
+	BUG_ON(!journal);
+
+	osb = journal->j_osb;
+
+	/* already have the inode for our journal */
+	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
+					    osb->slot_num);
+	if (inode == NULL) {
+		status = -EACCES;
+		mlog_errno(status);
+		goto done;
+	}
+	if (is_bad_inode(inode)) {
+		mlog(ML_ERROR, "access error (bad inode)\n");
+		iput(inode);
+		inode = NULL;
+		status = -EACCES;
+		goto done;
+	}
+
+	SET_INODE_JOURNAL(inode);
+	OCFS2_I(inode)->ip_open_count++;
+
+	/* Skip recovery waits here - journal inode metadata never
+	 * changes in a live cluster so it can be considered an
+	 * exception to the rule. */
+	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
+	if (status < 0) {
+		if (status != -ERESTARTSYS)
+			mlog(ML_ERROR, "Could not get lock on journal!\n");
+		goto done;
+	}
+
+	inode_lock = 1;
+	di = (struct ocfs2_dinode *)bh->b_data;
+
+	if (i_size_read(inode) <  OCFS2_MIN_JOURNAL_SIZE) {
+		mlog(ML_ERROR, "Journal file size (%lld) is too small!\n",
+		     i_size_read(inode));
+		status = -EINVAL;
+		goto done;
+	}
+
+	trace_ocfs2_journal_init(i_size_read(inode),
+				 (unsigned long long)inode->i_blocks,
+				 OCFS2_I(inode)->ip_clusters);
+
+	/* call the kernels journal init function now */
+	j_journal = jbd2_journal_init_inode(inode);
+	if (j_journal == NULL) {
+		mlog(ML_ERROR, "Linux journal layer error\n");
+		status = -EINVAL;
+		goto done;
+	}
+
+	trace_ocfs2_journal_init_maxlen(j_journal->j_maxlen);
+
+	*dirty = (le32_to_cpu(di->id1.journal1.ij_flags) &
+		  OCFS2_JOURNAL_DIRTY_FL);
+
+	journal->j_journal = j_journal;
+	journal->j_inode = inode;
+	journal->j_bh = bh;
+
+	ocfs2_set_journal_params(osb);
+
+	journal->j_state = OCFS2_JOURNAL_LOADED;
+
+	status = 0;
+done:
+	if (status < 0) {
+		if (inode_lock)
+			ocfs2_inode_unlock(inode, 1);
+		brelse(bh);
+		if (inode) {
+			OCFS2_I(inode)->ip_open_count--;
+			iput(inode);
+		}
+	}
+
+	return status;
+}
+
+static void ocfs2_bump_recovery_generation(struct ocfs2_dinode *di)
+{
+	le32_add_cpu(&(di->id1.journal1.ij_recovery_generation), 1);
+}
+
+static u32 ocfs2_get_recovery_generation(struct ocfs2_dinode *di)
+{
+	return le32_to_cpu(di->id1.journal1.ij_recovery_generation);
+}
+
+static int ocfs2_journal_toggle_dirty(struct ocfs2_super *osb,
+				      int dirty, int replayed)
+{
+	int status;
+	unsigned int flags;
+	struct ocfs2_journal *journal = osb->journal;
+	struct buffer_head *bh = journal->j_bh;
+	struct ocfs2_dinode *fe;
+
+	fe = (struct ocfs2_dinode *)bh->b_data;
+
+	/* The journal bh on the osb always comes from ocfs2_journal_init()
+	 * and was validated there inside ocfs2_inode_lock_full().  It's a
+	 * code bug if we mess it up. */
+	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
+
+	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
+	if (dirty)
+		flags |= OCFS2_JOURNAL_DIRTY_FL;
+	else
+		flags &= ~OCFS2_JOURNAL_DIRTY_FL;
+	fe->id1.journal1.ij_flags = cpu_to_le32(flags);
+
+	if (replayed)
+		ocfs2_bump_recovery_generation(fe);
+
+	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
+	status = ocfs2_write_block(osb, bh, INODE_CACHE(journal->j_inode));
+	if (status < 0)
+		mlog_errno(status);
+
+	return status;
+}
+
+/*
+ * If the journal has been kmalloc'd it needs to be freed after this
+ * call.
+ */
+void ocfs2_journal_shutdown(struct ocfs2_super *osb)
+{
+	struct ocfs2_journal *journal = NULL;
+	int status = 0;
+	struct inode *inode = NULL;
+	int num_running_trans = 0;
+
+	BUG_ON(!osb);
+
+	journal = osb->journal;
+	if (!journal)
+		goto done;
+
+	inode = journal->j_inode;
+
+	if (journal->j_state != OCFS2_JOURNAL_LOADED)
+		goto done;
+
+	/* need to inc inode use count - jbd2_journal_destroy will iput. */
+	if (!igrab(inode))
+		BUG();
+
+	num_running_trans = atomic_read(&(osb->journal->j_num_trans));
+	trace_ocfs2_journal_shutdown(num_running_trans);
+
+	/* Do a commit_cache here. It will flush our journal, *and*
+	 * release any locks that are still held.
+	 * set the SHUTDOWN flag and release the trans lock.
+	 * the commit thread will take the trans lock for us below. */
+	journal->j_state = OCFS2_JOURNAL_IN_SHUTDOWN;
+
+	/* The OCFS2_JOURNAL_IN_SHUTDOWN will signal to commit_cache to not
+	 * drop the trans_lock (which we want to hold until we
+	 * completely destroy the journal. */
+	if (osb->commit_task) {
+		/* Wait for the commit thread */
+		trace_ocfs2_journal_shutdown_wait(osb->commit_task);
+		kthread_stop(osb->commit_task);
+		osb->commit_task = NULL;
+	}
+
+	BUG_ON(atomic_read(&(osb->journal->j_num_trans)) != 0);
+
+	if (ocfs2_mount_local(osb)) {
+		jbd2_journal_lock_updates(journal->j_journal);
+		status = jbd2_journal_flush(journal->j_journal);
+		jbd2_journal_unlock_updates(journal->j_journal);
+		if (status < 0)
+			mlog_errno(status);
+	}
+
+	if (status == 0) {
+		/*
+		 * Do not toggle if flush was unsuccessful otherwise
+		 * will leave dirty metadata in a "clean" journal
+		 */
+		status = ocfs2_journal_toggle_dirty(osb, 0, 0);
+		if (status < 0)
+			mlog_errno(status);
+	}
+
+	/* Shutdown the kernel journal system */
+	jbd2_journal_destroy(journal->j_journal);
+	journal->j_journal = NULL;
+
+	OCFS2_I(inode)->ip_open_count--;
+
+	/* unlock our journal */
+	ocfs2_inode_unlock(inode, 1);
+
+	brelse(journal->j_bh);
+	journal->j_bh = NULL;
+
+	journal->j_state = OCFS2_JOURNAL_FREE;
+
+//	up_write(&journal->j_trans_barrier);
+done:
+	if (inode)
+		iput(inode);
+}
+
+static void ocfs2_clear_journal_error(struct super_block *sb,
+				      journal_t *journal,
+				      int slot)
+{
+	int olderr;
+
+	olderr = jbd2_journal_errno(journal);
+	if (olderr) {
+		mlog(ML_ERROR, "File system error %d recorded in "
+		     "journal %u.\n", olderr, slot);
+		mlog(ML_ERROR, "File system on device %s needs checking.\n",
+		     sb->s_id);
+
+		jbd2_journal_ack_err(journal);
+		jbd2_journal_clear_err(journal);
+	}
+}
+
+int ocfs2_journal_load(struct ocfs2_journal *journal, int local, int replayed)
+{
+	int status = 0;
+	struct ocfs2_super *osb;
+
+	BUG_ON(!journal);
+
+	osb = journal->j_osb;
+
+	status = jbd2_journal_load(journal->j_journal);
+	if (status < 0) {
+		mlog(ML_ERROR, "Failed to load journal!\n");
+		goto done;
+	}
+
+	ocfs2_clear_journal_error(osb->sb, journal->j_journal, osb->slot_num);
+
+	status = ocfs2_journal_toggle_dirty(osb, 1, replayed);
+	if (status < 0) {
+		mlog_errno(status);
+		goto done;
+	}
+
+	/* Launch the commit thread */
+	if (!local) {
+		osb->commit_task = kthread_run(ocfs2_commit_thread, osb,
+					       "ocfs2cmt");
+		if (IS_ERR(osb->commit_task)) {
+			status = PTR_ERR(osb->commit_task);
+			osb->commit_task = NULL;
+			mlog(ML_ERROR, "unable to launch ocfs2commit thread, "
+			     "error=%d", status);
+			goto done;
+		}
+	} else
+		osb->commit_task = NULL;
+
+done:
+	return status;
+}
+
+
+/* 'full' flag tells us whether we clear out all blocks or if we just
+ * mark the journal clean */
+int ocfs2_journal_wipe(struct ocfs2_journal *journal, int full)
+{
+	int status;
+
+	BUG_ON(!journal);
+
+	status = jbd2_journal_wipe(journal->j_journal, full);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_journal_toggle_dirty(journal->j_osb, 0, 0);
+	if (status < 0)
+		mlog_errno(status);
+
+bail:
+	return status;
+}
+
+static int ocfs2_recovery_completed(struct ocfs2_super *osb)
+{
+	int empty;
+	struct ocfs2_recovery_map *rm = osb->recovery_map;
+
+	spin_lock(&osb->osb_lock);
+	empty = (rm->rm_used == 0);
+	spin_unlock(&osb->osb_lock);
+
+	return empty;
+}
+
+void ocfs2_wait_for_recovery(struct ocfs2_super *osb)
+{
+	wait_event(osb->recovery_event, ocfs2_recovery_completed(osb));
+}
+
+/*
+ * JBD Might read a cached version of another nodes journal file. We
+ * don't want this as this file changes often and we get no
+ * notification on those changes. The only way to be sure that we've
+ * got the most up to date version of those blocks then is to force
+ * read them off disk. Just searching through the buffer cache won't
+ * work as there may be pages backing this file which are still marked
+ * up to date. We know things can't change on this file underneath us
+ * as we have the lock by now :)
+ */
+static int ocfs2_force_read_journal(struct inode *inode)
+{
+	int status = 0;
+	int i;
+	u64 v_blkno, p_blkno, p_blocks, num_blocks;
+#define CONCURRENT_JOURNAL_FILL 32ULL
+	struct buffer_head *bhs[CONCURRENT_JOURNAL_FILL];
+
+	memset(bhs, 0, sizeof(struct buffer_head *) * CONCURRENT_JOURNAL_FILL);
+
+	num_blocks = ocfs2_blocks_for_bytes(inode->i_sb, i_size_read(inode));
+	v_blkno = 0;
+	while (v_blkno < num_blocks) {
+		status = ocfs2_extent_map_get_blocks(inode, v_blkno,
+						     &p_blkno, &p_blocks, NULL);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		if (p_blocks > CONCURRENT_JOURNAL_FILL)
+			p_blocks = CONCURRENT_JOURNAL_FILL;
+
+		/* We are reading journal data which should not
+		 * be put in the uptodate cache */
+		status = ocfs2_read_blocks_sync(OCFS2_SB(inode->i_sb),
+						p_blkno, p_blocks, bhs);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		for(i = 0; i < p_blocks; i++) {
+			brelse(bhs[i]);
+			bhs[i] = NULL;
+		}
+
+		v_blkno += p_blocks;
+	}
+
+bail:
+	for(i = 0; i < CONCURRENT_JOURNAL_FILL; i++)
+		brelse(bhs[i]);
+	return status;
+}
+
+struct ocfs2_la_recovery_item {
+	struct list_head	lri_list;
+	int			lri_slot;
+	struct ocfs2_dinode	*lri_la_dinode;
+	struct ocfs2_dinode	*lri_tl_dinode;
+	struct ocfs2_quota_recovery *lri_qrec;
+	enum ocfs2_orphan_reco_type  lri_orphan_reco_type;
+};
+
+/* Does the second half of the recovery process. By this point, the
+ * node is marked clean and can actually be considered recovered,
+ * hence it's no longer in the recovery map, but there's still some
+ * cleanup we can do which shouldn't happen within the recovery thread
+ * as locking in that context becomes very difficult if we are to take
+ * recovering nodes into account.
+ *
+ * NOTE: This function can and will sleep on recovery of other nodes
+ * during cluster locking, just like any other ocfs2 process.
+ */
+void ocfs2_complete_recovery(struct work_struct *work)
+{
+	int ret = 0;
+	struct ocfs2_journal *journal =
+		container_of(work, struct ocfs2_journal, j_recovery_work);
+	struct ocfs2_super *osb = journal->j_osb;
+	struct ocfs2_dinode *la_dinode, *tl_dinode;
+	struct ocfs2_la_recovery_item *item, *n;
+	struct ocfs2_quota_recovery *qrec;
+	enum ocfs2_orphan_reco_type orphan_reco_type;
+	LIST_HEAD(tmp_la_list);
+
+	trace_ocfs2_complete_recovery(
+		(unsigned long long)OCFS2_I(journal->j_inode)->ip_blkno);
+
+	spin_lock(&journal->j_lock);
+	list_splice_init(&journal->j_la_cleanups, &tmp_la_list);
+	spin_unlock(&journal->j_lock);
+
+	list_for_each_entry_safe(item, n, &tmp_la_list, lri_list) {
+		list_del_init(&item->lri_list);
+
+		ocfs2_wait_on_quotas(osb);
+
+		la_dinode = item->lri_la_dinode;
+		tl_dinode = item->lri_tl_dinode;
+		qrec = item->lri_qrec;
+		orphan_reco_type = item->lri_orphan_reco_type;
+
+		trace_ocfs2_complete_recovery_slot(item->lri_slot,
+			la_dinode ? le64_to_cpu(la_dinode->i_blkno) : 0,
+			tl_dinode ? le64_to_cpu(tl_dinode->i_blkno) : 0,
+			qrec);
+
+		if (la_dinode) {
+			ret = ocfs2_complete_local_alloc_recovery(osb,
+								  la_dinode);
+			if (ret < 0)
+				mlog_errno(ret);
+
+			kfree(la_dinode);
+		}
+
+		if (tl_dinode) {
+			ret = ocfs2_complete_truncate_log_recovery(osb,
+								   tl_dinode);
+			if (ret < 0)
+				mlog_errno(ret);
+
+			kfree(tl_dinode);
+		}
+
+		ret = ocfs2_recover_orphans(osb, item->lri_slot,
+				orphan_reco_type);
+		if (ret < 0)
+			mlog_errno(ret);
+
+		if (qrec) {
+			ret = ocfs2_finish_quota_recovery(osb, qrec,
+							  item->lri_slot);
+			if (ret < 0)
+				mlog_errno(ret);
+			/* Recovery info is already freed now */
+		}
+
+		kfree(item);
+	}
+
+	trace_ocfs2_complete_recovery_end(ret);
+}
+
+/* NOTE: This function always eats your references to la_dinode and
+ * tl_dinode, either manually on error, or by passing them to
+ * ocfs2_complete_recovery */
+static void ocfs2_queue_recovery_completion(struct ocfs2_journal *journal,
+					    int slot_num,
+					    struct ocfs2_dinode *la_dinode,
+					    struct ocfs2_dinode *tl_dinode,
+					    struct ocfs2_quota_recovery *qrec,
+					    enum ocfs2_orphan_reco_type orphan_reco_type)
+{
+	struct ocfs2_la_recovery_item *item;
+
+	item = kmalloc(sizeof(struct ocfs2_la_recovery_item), GFP_NOFS);
+	if (!item) {
+		/* Though we wish to avoid it, we are in fact safe in
+		 * skipping local alloc cleanup as fsck.ocfs2 is more
+		 * than capable of reclaiming unused space. */
+		kfree(la_dinode);
+		kfree(tl_dinode);
+
+		if (qrec)
+			ocfs2_free_quota_recovery(qrec);
+
+		mlog_errno(-ENOMEM);
+		return;
+	}
+
+	INIT_LIST_HEAD(&item->lri_list);
+	item->lri_la_dinode = la_dinode;
+	item->lri_slot = slot_num;
+	item->lri_tl_dinode = tl_dinode;
+	item->lri_qrec = qrec;
+	item->lri_orphan_reco_type = orphan_reco_type;
+
+	spin_lock(&journal->j_lock);
+	list_add_tail(&item->lri_list, &journal->j_la_cleanups);
+	queue_work(ocfs2_wq, &journal->j_recovery_work);
+	spin_unlock(&journal->j_lock);
+}
+
+/* Called by the mount code to queue recovery the last part of
+ * recovery for it's own and offline slot(s). */
+void ocfs2_complete_mount_recovery(struct ocfs2_super *osb)
+{
+	struct ocfs2_journal *journal = osb->journal;
+
+	if (ocfs2_is_hard_readonly(osb))
+		return;
+
+	/* No need to queue up our truncate_log as regular cleanup will catch
+	 * that */
+	ocfs2_queue_recovery_completion(journal, osb->slot_num,
+					osb->local_alloc_copy, NULL, NULL,
+					ORPHAN_NEED_TRUNCATE);
+	ocfs2_schedule_truncate_log_flush(osb, 0);
+
+	osb->local_alloc_copy = NULL;
+	osb->dirty = 0;
+
+	/* queue to recover orphan slots for all offline slots */
+	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
+	ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
+	ocfs2_free_replay_slots(osb);
+}
+
+void ocfs2_complete_quota_recovery(struct ocfs2_super *osb)
+{
+	if (osb->quota_rec) {
+		ocfs2_queue_recovery_completion(osb->journal,
+						osb->slot_num,
+						NULL,
+						NULL,
+						osb->quota_rec,
+						ORPHAN_NEED_TRUNCATE);
+		osb->quota_rec = NULL;
+	}
+}
+
+static int __ocfs2_recovery_thread(void *arg)
+{
+	int status, node_num, slot_num;
+	struct ocfs2_super *osb = arg;
+	struct ocfs2_recovery_map *rm = osb->recovery_map;
+	int *rm_quota = NULL;
+	int rm_quota_used = 0, i;
+	struct ocfs2_quota_recovery *qrec;
+
+	status = ocfs2_wait_on_mount(osb);
+	if (status < 0) {
+		goto bail;
+	}
+
+	rm_quota = kzalloc(osb->max_slots * sizeof(int), GFP_NOFS);
+	if (!rm_quota) {
+		status = -ENOMEM;
+		goto bail;
+	}
+restart:
+	status = ocfs2_super_lock(osb, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_compute_replay_slots(osb);
+	if (status < 0)
+		mlog_errno(status);
+
+	/* queue recovery for our own slot */
+	ocfs2_queue_recovery_completion(osb->journal, osb->slot_num, NULL,
+					NULL, NULL, ORPHAN_NO_NEED_TRUNCATE);
+
+	spin_lock(&osb->osb_lock);
+	while (rm->rm_used) {
+		/* It's always safe to remove entry zero, as we won't
+		 * clear it until ocfs2_recover_node() has succeeded. */
+		node_num = rm->rm_entries[0];
+		spin_unlock(&osb->osb_lock);
+		slot_num = ocfs2_node_num_to_slot(osb, node_num);
+		trace_ocfs2_recovery_thread_node(node_num, slot_num);
+		if (slot_num == -ENOENT) {
+			status = 0;
+			goto skip_recovery;
+		}
+
+		/* It is a bit subtle with quota recovery. We cannot do it
+		 * immediately because we have to obtain cluster locks from
+		 * quota files and we also don't want to just skip it because
+		 * then quota usage would be out of sync until some node takes
+		 * the slot. So we remember which nodes need quota recovery
+		 * and when everything else is done, we recover quotas. */
+		for (i = 0; i < rm_quota_used && rm_quota[i] != slot_num; i++);
+		if (i == rm_quota_used)
+			rm_quota[rm_quota_used++] = slot_num;
+
+		status = ocfs2_recover_node(osb, node_num, slot_num);
+skip_recovery:
+		if (!status) {
+			ocfs2_recovery_map_clear(osb, node_num);
+		} else {
+			mlog(ML_ERROR,
+			     "Error %d recovering node %d on device (%u,%u)!\n",
+			     status, node_num,
+			     MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
+			mlog(ML_ERROR, "Volume requires unmount.\n");
+		}
+
+		spin_lock(&osb->osb_lock);
+	}
+	spin_unlock(&osb->osb_lock);
+	trace_ocfs2_recovery_thread_end(status);
+
+	/* Refresh all journal recovery generations from disk */
+	status = ocfs2_check_journals_nolocks(osb);
+	status = (status == -EROFS) ? 0 : status;
+	if (status < 0)
+		mlog_errno(status);
+
+	/* Now it is right time to recover quotas... We have to do this under
+	 * superblock lock so that no one can start using the slot (and crash)
+	 * before we recover it */
+	for (i = 0; i < rm_quota_used; i++) {
+		qrec = ocfs2_begin_quota_recovery(osb, rm_quota[i]);
+		if (IS_ERR(qrec)) {
+			status = PTR_ERR(qrec);
+			mlog_errno(status);
+			continue;
+		}
+		ocfs2_queue_recovery_completion(osb->journal, rm_quota[i],
+						NULL, NULL, qrec,
+						ORPHAN_NEED_TRUNCATE);
+	}
+
+	ocfs2_super_unlock(osb, 1);
+
+	/* queue recovery for offline slots */
+	ocfs2_queue_replay_slots(osb, ORPHAN_NEED_TRUNCATE);
+
+bail:
+	mutex_lock(&osb->recovery_lock);
+	if (!status && !ocfs2_recovery_completed(osb)) {
+		mutex_unlock(&osb->recovery_lock);
+		goto restart;
+	}
+
+	ocfs2_free_replay_slots(osb);
+	osb->recovery_thread_task = NULL;
+	mb(); /* sync with ocfs2_recovery_thread_running */
+	wake_up(&osb->recovery_event);
+
+	mutex_unlock(&osb->recovery_lock);
+
+	kfree(rm_quota);
+
+	/* no one is callint kthread_stop() for us so the kthread() api
+	 * requires that we call do_exit().  And it isn't exported, but
+	 * complete_and_exit() seems to be a minimal wrapper around it. */
+	complete_and_exit(NULL, status);
+}
+
+void ocfs2_recovery_thread(struct ocfs2_super *osb, int node_num)
+{
+	mutex_lock(&osb->recovery_lock);
+
+	trace_ocfs2_recovery_thread(node_num, osb->node_num,
+		osb->disable_recovery, osb->recovery_thread_task,
+		osb->disable_recovery ?
+		-1 : ocfs2_recovery_map_set(osb, node_num));
+
+	if (osb->disable_recovery)
+		goto out;
+
+	if (osb->recovery_thread_task)
+		goto out;
+
+	osb->recovery_thread_task =  kthread_run(__ocfs2_recovery_thread, osb,
+						 "ocfs2rec");
+	if (IS_ERR(osb->recovery_thread_task)) {
+		mlog_errno((int)PTR_ERR(osb->recovery_thread_task));
+		osb->recovery_thread_task = NULL;
+	}
+
+out:
+	mutex_unlock(&osb->recovery_lock);
+	wake_up(&osb->recovery_event);
+}
+
+static int ocfs2_read_journal_inode(struct ocfs2_super *osb,
+				    int slot_num,
+				    struct buffer_head **bh,
+				    struct inode **ret_inode)
+{
+	int status = -EACCES;
+	struct inode *inode = NULL;
+
+	BUG_ON(slot_num >= osb->max_slots);
+
+	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
+					    slot_num);
+	if (!inode || is_bad_inode(inode)) {
+		mlog_errno(status);
+		goto bail;
+	}
+	SET_INODE_JOURNAL(inode);
+
+	status = ocfs2_read_inode_block_full(inode, bh, OCFS2_BH_IGNORE_CACHE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = 0;
+
+bail:
+	if (inode) {
+		if (status || !ret_inode)
+			iput(inode);
+		else
+			*ret_inode = inode;
+	}
+	return status;
+}
+
+/* Does the actual journal replay and marks the journal inode as
+ * clean. Will only replay if the journal inode is marked dirty. */
+static int ocfs2_replay_journal(struct ocfs2_super *osb,
+				int node_num,
+				int slot_num)
+{
+	int status;
+	int got_lock = 0;
+	unsigned int flags;
+	struct inode *inode = NULL;
+	struct ocfs2_dinode *fe;
+	journal_t *journal = NULL;
+	struct buffer_head *bh = NULL;
+	u32 slot_reco_gen;
+
+	status = ocfs2_read_journal_inode(osb, slot_num, &bh, &inode);
+	if (status) {
+		mlog_errno(status);
+		goto done;
+	}
+
+	fe = (struct ocfs2_dinode *)bh->b_data;
+	slot_reco_gen = ocfs2_get_recovery_generation(fe);
+	brelse(bh);
+	bh = NULL;
+
+	/*
+	 * As the fs recovery is asynchronous, there is a small chance that
+	 * another node mounted (and recovered) the slot before the recovery
+	 * thread could get the lock. To handle that, we dirty read the journal
+	 * inode for that slot to get the recovery generation. If it is
+	 * different than what we expected, the slot has been recovered.
+	 * If not, it needs recovery.
+	 */
+	if (osb->slot_recovery_generations[slot_num] != slot_reco_gen) {
+		trace_ocfs2_replay_journal_recovered(slot_num,
+		     osb->slot_recovery_generations[slot_num], slot_reco_gen);
+		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
+		status = -EBUSY;
+		goto done;
+	}
+
+	/* Continue with recovery as the journal has not yet been recovered */
+
+	status = ocfs2_inode_lock_full(inode, &bh, 1, OCFS2_META_LOCK_RECOVERY);
+	if (status < 0) {
+		trace_ocfs2_replay_journal_lock_err(status);
+		if (status != -ERESTARTSYS)
+			mlog(ML_ERROR, "Could not lock journal!\n");
+		goto done;
+	}
+	got_lock = 1;
+
+	fe = (struct ocfs2_dinode *) bh->b_data;
+
+	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
+	slot_reco_gen = ocfs2_get_recovery_generation(fe);
+
+	if (!(flags & OCFS2_JOURNAL_DIRTY_FL)) {
+		trace_ocfs2_replay_journal_skip(node_num);
+		/* Refresh recovery generation for the slot */
+		osb->slot_recovery_generations[slot_num] = slot_reco_gen;
+		goto done;
+	}
+
+	/* we need to run complete recovery for offline orphan slots */
+	ocfs2_replay_map_set_state(osb, REPLAY_NEEDED);
+
+	printk(KERN_NOTICE "ocfs2: Begin replay journal (node %d, slot %d) on "\
+	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
+	       MINOR(osb->sb->s_dev));
+
+	OCFS2_I(inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
+
+	status = ocfs2_force_read_journal(inode);
+	if (status < 0) {
+		mlog_errno(status);
+		goto done;
+	}
+
+	journal = jbd2_journal_init_inode(inode);
+	if (journal == NULL) {
+		mlog(ML_ERROR, "Linux journal layer error\n");
+		status = -EIO;
+		goto done;
+	}
+
+	status = jbd2_journal_load(journal);
+	if (status < 0) {
+		mlog_errno(status);
+		if (!igrab(inode))
+			BUG();
+		jbd2_journal_destroy(journal);
+		goto done;
+	}
+
+	ocfs2_clear_journal_error(osb->sb, journal, slot_num);
+
+	/* wipe the journal */
+	jbd2_journal_lock_updates(journal);
+	status = jbd2_journal_flush(journal);
+	jbd2_journal_unlock_updates(journal);
+	if (status < 0)
+		mlog_errno(status);
+
+	/* This will mark the node clean */
+	flags = le32_to_cpu(fe->id1.journal1.ij_flags);
+	flags &= ~OCFS2_JOURNAL_DIRTY_FL;
+	fe->id1.journal1.ij_flags = cpu_to_le32(flags);
+
+	/* Increment recovery generation to indicate successful recovery */
+	ocfs2_bump_recovery_generation(fe);
+	osb->slot_recovery_generations[slot_num] =
+					ocfs2_get_recovery_generation(fe);
+
+	ocfs2_compute_meta_ecc(osb->sb, bh->b_data, &fe->i_check);
+	status = ocfs2_write_block(osb, bh, INODE_CACHE(inode));
+	if (status < 0)
+		mlog_errno(status);
+
+	if (!igrab(inode))
+		BUG();
+
+	jbd2_journal_destroy(journal);
+
+	printk(KERN_NOTICE "ocfs2: End replay journal (node %d, slot %d) on "\
+	       "device (%u,%u)\n", node_num, slot_num, MAJOR(osb->sb->s_dev),
+	       MINOR(osb->sb->s_dev));
+done:
+	/* drop the lock on this nodes journal */
+	if (got_lock)
+		ocfs2_inode_unlock(inode, 1);
+
+	if (inode)
+		iput(inode);
+
+	brelse(bh);
+
+	return status;
+}
+
+/*
+ * Do the most important parts of node recovery:
+ *  - Replay it's journal
+ *  - Stamp a clean local allocator file
+ *  - Stamp a clean truncate log
+ *  - Mark the node clean
+ *
+ * If this function completes without error, a node in OCFS2 can be
+ * said to have been safely recovered. As a result, failure during the
+ * second part of a nodes recovery process (local alloc recovery) is
+ * far less concerning.
+ */
+static int ocfs2_recover_node(struct ocfs2_super *osb,
+			      int node_num, int slot_num)
+{
+	int status = 0;
+	struct ocfs2_dinode *la_copy = NULL;
+	struct ocfs2_dinode *tl_copy = NULL;
+
+	trace_ocfs2_recover_node(node_num, slot_num, osb->node_num);
+
+	/* Should not ever be called to recover ourselves -- in that
+	 * case we should've called ocfs2_journal_load instead. */
+	BUG_ON(osb->node_num == node_num);
+
+	status = ocfs2_replay_journal(osb, node_num, slot_num);
+	if (status < 0) {
+		if (status == -EBUSY) {
+			trace_ocfs2_recover_node_skip(slot_num, node_num);
+			status = 0;
+			goto done;
+		}
+		mlog_errno(status);
+		goto done;
+	}
+
+	/* Stamp a clean local alloc file AFTER recovering the journal... */
+	status = ocfs2_begin_local_alloc_recovery(osb, slot_num, &la_copy);
+	if (status < 0) {
+		mlog_errno(status);
+		goto done;
+	}
+
+	/* An error from begin_truncate_log_recovery is not
+	 * serious enough to warrant halting the rest of
+	 * recovery. */
+	status = ocfs2_begin_truncate_log_recovery(osb, slot_num, &tl_copy);
+	if (status < 0)
+		mlog_errno(status);
+
+	/* Likewise, this would be a strange but ultimately not so
+	 * harmful place to get an error... */
+	status = ocfs2_clear_slot(osb, slot_num);
+	if (status < 0)
+		mlog_errno(status);
+
+	/* This will kfree the memory pointed to by la_copy and tl_copy */
+	ocfs2_queue_recovery_completion(osb->journal, slot_num, la_copy,
+					tl_copy, NULL, ORPHAN_NEED_TRUNCATE);
+
+	status = 0;
+done:
+
+	return status;
+}
+
+/* Test node liveness by trylocking his journal. If we get the lock,
+ * we drop it here. Return 0 if we got the lock, -EAGAIN if node is
+ * still alive (we couldn't get the lock) and < 0 on error. */
+static int ocfs2_trylock_journal(struct ocfs2_super *osb,
+				 int slot_num)
+{
+	int status, flags;
+	struct inode *inode = NULL;
+
+	inode = ocfs2_get_system_file_inode(osb, JOURNAL_SYSTEM_INODE,
+					    slot_num);
+	if (inode == NULL) {
+		mlog(ML_ERROR, "access error\n");
+		status = -EACCES;
+		goto bail;
+	}
+	if (is_bad_inode(inode)) {
+		mlog(ML_ERROR, "access error (bad inode)\n");
+		iput(inode);
+		inode = NULL;
+		status = -EACCES;
+		goto bail;
+	}
+	SET_INODE_JOURNAL(inode);
+
+	flags = OCFS2_META_LOCK_RECOVERY | OCFS2_META_LOCK_NOQUEUE;
+	status = ocfs2_inode_lock_full(inode, NULL, 1, flags);
+	if (status < 0) {
+		if (status != -EAGAIN)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	ocfs2_inode_unlock(inode, 1);
+bail:
+	if (inode)
+		iput(inode);
+
+	return status;
+}
+
+/* Call this underneath ocfs2_super_lock. It also assumes that the
+ * slot info struct has been updated from disk. */
+int ocfs2_mark_dead_nodes(struct ocfs2_super *osb)
+{
+	unsigned int node_num;
+	int status, i;
+	u32 gen;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_dinode *di;
+
+	/* This is called with the super block cluster lock, so we
+	 * know that the slot map can't change underneath us. */
+
+	for (i = 0; i < osb->max_slots; i++) {
+		/* Read journal inode to get the recovery generation */
+		status = ocfs2_read_journal_inode(osb, i, &bh, NULL);
+		if (status) {
+			mlog_errno(status);
+			goto bail;
+		}
+		di = (struct ocfs2_dinode *)bh->b_data;
+		gen = ocfs2_get_recovery_generation(di);
+		brelse(bh);
+		bh = NULL;
+
+		spin_lock(&osb->osb_lock);
+		osb->slot_recovery_generations[i] = gen;
+
+		trace_ocfs2_mark_dead_nodes(i,
+					    osb->slot_recovery_generations[i]);
+
+		if (i == osb->slot_num) {
+			spin_unlock(&osb->osb_lock);
+			continue;
+		}
+
+		status = ocfs2_slot_to_node_num_locked(osb, i, &node_num);
+		if (status == -ENOENT) {
+			spin_unlock(&osb->osb_lock);
+			continue;
+		}
+
+		if (__ocfs2_recovery_map_test(osb, node_num)) {
+			spin_unlock(&osb->osb_lock);
+			continue;
+		}
+		spin_unlock(&osb->osb_lock);
+
+		/* Ok, we have a slot occupied by another node which
+		 * is not in the recovery map. We trylock his journal
+		 * file here to test if he's alive. */
+		status = ocfs2_trylock_journal(osb, i);
+		if (!status) {
+			/* Since we're called from mount, we know that
+			 * the recovery thread can't race us on
+			 * setting / checking the recovery bits. */
+			ocfs2_recovery_thread(osb, node_num);
+		} else if ((status < 0) && (status != -EAGAIN)) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	status = 0;
+bail:
+	return status;
+}
+
+/*
+ * Scan timer should get fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT. Add some
+ * randomness to the timeout to minimize multple nodes firing the timer at the
+ * same time.
+ */
+static inline unsigned long ocfs2_orphan_scan_timeout(void)
+{
+	unsigned long time;
+
+	get_random_bytes(&time, sizeof(time));
+	time = ORPHAN_SCAN_SCHEDULE_TIMEOUT + (time % 5000);
+	return msecs_to_jiffies(time);
+}
+
+/*
+ * ocfs2_queue_orphan_scan calls ocfs2_queue_recovery_completion for
+ * every slot, queuing a recovery of the slot on the ocfs2_wq thread. This
+ * is done to catch any orphans that are left over in orphan directories.
+ *
+ * It scans all slots, even ones that are in use. It does so to handle the
+ * case described below:
+ *
+ *   Node 1 has an inode it was using. The dentry went away due to memory
+ *   pressure.  Node 1 closes the inode, but it's on the free list. The node
+ *   has the open lock.
+ *   Node 2 unlinks the inode. It grabs the dentry lock to notify others,
+ *   but node 1 has no dentry and doesn't get the message. It trylocks the
+ *   open lock, sees that another node has a PR, and does nothing.
+ *   Later node 2 runs its orphan dir. It igets the inode, trylocks the
+ *   open lock, sees the PR still, and does nothing.
+ *   Basically, we have to trigger an orphan iput on node 1. The only way
+ *   for this to happen is if node 1 runs node 2's orphan dir.
+ *
+ * ocfs2_queue_orphan_scan gets called every ORPHAN_SCAN_SCHEDULE_TIMEOUT
+ * seconds.  It gets an EX lock on os_lockres and checks sequence number
+ * stored in LVB. If the sequence number has changed, it means some other
+ * node has done the scan.  This node skips the scan and tracks the
+ * sequence number.  If the sequence number didn't change, it means a scan
+ * hasn't happened.  The node queues a scan and increments the
+ * sequence number in the LVB.
+ */
+void ocfs2_queue_orphan_scan(struct ocfs2_super *osb)
+{
+	struct ocfs2_orphan_scan *os;
+	int status, i;
+	u32 seqno = 0;
+
+	os = &osb->osb_orphan_scan;
+
+	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
+		goto out;
+
+	trace_ocfs2_queue_orphan_scan_begin(os->os_count, os->os_seqno,
+					    atomic_read(&os->os_state));
+
+	status = ocfs2_orphan_scan_lock(osb, &seqno);
+	if (status < 0) {
+		if (status != -EAGAIN)
+			mlog_errno(status);
+		goto out;
+	}
+
+	/* Do no queue the tasks if the volume is being umounted */
+	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
+		goto unlock;
+
+	if (os->os_seqno != seqno) {
+		os->os_seqno = seqno;
+		goto unlock;
+	}
+
+	for (i = 0; i < osb->max_slots; i++)
+		ocfs2_queue_recovery_completion(osb->journal, i, NULL, NULL,
+						NULL, ORPHAN_NO_NEED_TRUNCATE);
+	/*
+	 * We queued a recovery on orphan slots, increment the sequence
+	 * number and update LVB so other node will skip the scan for a while
+	 */
+	seqno++;
+	os->os_count++;
+	os->os_scantime = CURRENT_TIME;
+unlock:
+	ocfs2_orphan_scan_unlock(osb, seqno);
+out:
+	trace_ocfs2_queue_orphan_scan_end(os->os_count, os->os_seqno,
+					  atomic_read(&os->os_state));
+	return;
+}
+
+/* Worker task that gets fired every ORPHAN_SCAN_SCHEDULE_TIMEOUT millsec */
+void ocfs2_orphan_scan_work(struct work_struct *work)
+{
+	struct ocfs2_orphan_scan *os;
+	struct ocfs2_super *osb;
+
+	os = container_of(work, struct ocfs2_orphan_scan,
+			  os_orphan_scan_work.work);
+	osb = os->os_osb;
+
+	mutex_lock(&os->os_lock);
+	ocfs2_queue_orphan_scan(osb);
+	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE)
+		queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
+				      ocfs2_orphan_scan_timeout());
+	mutex_unlock(&os->os_lock);
+}
+
+void ocfs2_orphan_scan_stop(struct ocfs2_super *osb)
+{
+	struct ocfs2_orphan_scan *os;
+
+	os = &osb->osb_orphan_scan;
+	if (atomic_read(&os->os_state) == ORPHAN_SCAN_ACTIVE) {
+		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
+		mutex_lock(&os->os_lock);
+		cancel_delayed_work(&os->os_orphan_scan_work);
+		mutex_unlock(&os->os_lock);
+	}
+}
+
+void ocfs2_orphan_scan_init(struct ocfs2_super *osb)
+{
+	struct ocfs2_orphan_scan *os;
+
+	os = &osb->osb_orphan_scan;
+	os->os_osb = osb;
+	os->os_count = 0;
+	os->os_seqno = 0;
+	mutex_init(&os->os_lock);
+	INIT_DELAYED_WORK(&os->os_orphan_scan_work, ocfs2_orphan_scan_work);
+}
+
+void ocfs2_orphan_scan_start(struct ocfs2_super *osb)
+{
+	struct ocfs2_orphan_scan *os;
+
+	os = &osb->osb_orphan_scan;
+	os->os_scantime = CURRENT_TIME;
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_mount_local(osb))
+		atomic_set(&os->os_state, ORPHAN_SCAN_INACTIVE);
+	else {
+		atomic_set(&os->os_state, ORPHAN_SCAN_ACTIVE);
+		queue_delayed_work(ocfs2_wq, &os->os_orphan_scan_work,
+				   ocfs2_orphan_scan_timeout());
+	}
+}
+
+struct ocfs2_orphan_filldir_priv {
+	struct dir_context	ctx;
+	struct inode		*head;
+	struct ocfs2_super	*osb;
+};
+
+static int ocfs2_orphan_filldir(struct dir_context *ctx, const char *name,
+				int name_len, loff_t pos, u64 ino,
+				unsigned type)
+{
+	struct ocfs2_orphan_filldir_priv *p =
+		container_of(ctx, struct ocfs2_orphan_filldir_priv, ctx);
+	struct inode *iter;
+
+	if (name_len == 1 && !strncmp(".", name, 1))
+		return 0;
+	if (name_len == 2 && !strncmp("..", name, 2))
+		return 0;
+
+	/* Skip bad inodes so that recovery can continue */
+	iter = ocfs2_iget(p->osb, ino,
+			  OCFS2_FI_FLAG_ORPHAN_RECOVERY, 0);
+	if (IS_ERR(iter))
+		return 0;
+
+	/* Skip inodes which are already added to recover list, since dio may
+	 * happen concurrently with unlink/rename */
+	if (OCFS2_I(iter)->ip_next_orphan) {
+		iput(iter);
+		return 0;
+	}
+
+	trace_ocfs2_orphan_filldir((unsigned long long)OCFS2_I(iter)->ip_blkno);
+	/* No locking is required for the next_orphan queue as there
+	 * is only ever a single process doing orphan recovery. */
+	OCFS2_I(iter)->ip_next_orphan = p->head;
+	p->head = iter;
+
+	return 0;
+}
+
+static int ocfs2_queue_orphans(struct ocfs2_super *osb,
+			       int slot,
+			       struct inode **head)
+{
+	int status;
+	struct inode *orphan_dir_inode = NULL;
+	struct ocfs2_orphan_filldir_priv priv = {
+		.ctx.actor = ocfs2_orphan_filldir,
+		.osb = osb,
+		.head = *head
+	};
+
+	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
+						       ORPHAN_DIR_SYSTEM_INODE,
+						       slot);
+	if  (!orphan_dir_inode) {
+		status = -ENOENT;
+		mlog_errno(status);
+		return status;
+	}
+
+	mutex_lock(&orphan_dir_inode->i_mutex);
+	status = ocfs2_inode_lock(orphan_dir_inode, NULL, 0);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	status = ocfs2_dir_foreach(orphan_dir_inode, &priv.ctx);
+	if (status) {
+		mlog_errno(status);
+		goto out_cluster;
+	}
+
+	*head = priv.head;
+
+out_cluster:
+	ocfs2_inode_unlock(orphan_dir_inode, 0);
+out:
+	mutex_unlock(&orphan_dir_inode->i_mutex);
+	iput(orphan_dir_inode);
+	return status;
+}
+
+static int ocfs2_orphan_recovery_can_continue(struct ocfs2_super *osb,
+					      int slot)
+{
+	int ret;
+
+	spin_lock(&osb->osb_lock);
+	ret = !osb->osb_orphan_wipes[slot];
+	spin_unlock(&osb->osb_lock);
+	return ret;
+}
+
+static void ocfs2_mark_recovering_orphan_dir(struct ocfs2_super *osb,
+					     int slot)
+{
+	spin_lock(&osb->osb_lock);
+	/* Mark ourselves such that new processes in delete_inode()
+	 * know to quit early. */
+	ocfs2_node_map_set_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
+	while (osb->osb_orphan_wipes[slot]) {
+		/* If any processes are already in the middle of an
+		 * orphan wipe on this dir, then we need to wait for
+		 * them. */
+		spin_unlock(&osb->osb_lock);
+		wait_event_interruptible(osb->osb_wipe_event,
+					 ocfs2_orphan_recovery_can_continue(osb, slot));
+		spin_lock(&osb->osb_lock);
+	}
+	spin_unlock(&osb->osb_lock);
+}
+
+static void ocfs2_clear_recovering_orphan_dir(struct ocfs2_super *osb,
+					      int slot)
+{
+	ocfs2_node_map_clear_bit(osb, &osb->osb_recovering_orphan_dirs, slot);
+}
+
+/*
+ * Orphan recovery. Each mounted node has it's own orphan dir which we
+ * must run during recovery. Our strategy here is to build a list of
+ * the inodes in the orphan dir and iget/iput them. The VFS does
+ * (most) of the rest of the work.
+ *
+ * Orphan recovery can happen at any time, not just mount so we have a
+ * couple of extra considerations.
+ *
+ * - We grab as many inodes as we can under the orphan dir lock -
+ *   doing iget() outside the orphan dir risks getting a reference on
+ *   an invalid inode.
+ * - We must be sure not to deadlock with other processes on the
+ *   system wanting to run delete_inode(). This can happen when they go
+ *   to lock the orphan dir and the orphan recovery process attempts to
+ *   iget() inside the orphan dir lock. This can be avoided by
+ *   advertising our state to ocfs2_delete_inode().
+ */
+static int ocfs2_recover_orphans(struct ocfs2_super *osb,
+				 int slot,
+				 enum ocfs2_orphan_reco_type orphan_reco_type)
+{
+	int ret = 0;
+	struct inode *inode = NULL;
+	struct inode *iter;
+	struct ocfs2_inode_info *oi;
+
+	trace_ocfs2_recover_orphans(slot);
+
+	ocfs2_mark_recovering_orphan_dir(osb, slot);
+	ret = ocfs2_queue_orphans(osb, slot, &inode);
+	ocfs2_clear_recovering_orphan_dir(osb, slot);
+
+	/* Error here should be noted, but we want to continue with as
+	 * many queued inodes as we've got. */
+	if (ret)
+		mlog_errno(ret);
+
+	while (inode) {
+		oi = OCFS2_I(inode);
+		trace_ocfs2_recover_orphans_iput(
+					(unsigned long long)oi->ip_blkno);
+
+		iter = oi->ip_next_orphan;
+		oi->ip_next_orphan = NULL;
+
+		/*
+		 * We need to take and drop the inode lock to
+		 * force read inode from disk.
+		 */
+		ret = ocfs2_inode_lock(inode, NULL, 0);
+		if (ret) {
+			mlog_errno(ret);
+			goto next;
+		}
+		ocfs2_inode_unlock(inode, 0);
+
+		if (inode->i_nlink == 0) {
+			spin_lock(&oi->ip_lock);
+			/* Set the proper information to get us going into
+			 * ocfs2_delete_inode. */
+			oi->ip_flags |= OCFS2_INODE_MAYBE_ORPHANED;
+			spin_unlock(&oi->ip_lock);
+		} else if (orphan_reco_type == ORPHAN_NEED_TRUNCATE) {
+			struct buffer_head *di_bh = NULL;
+
+			ret = ocfs2_rw_lock(inode, 1);
+			if (ret) {
+				mlog_errno(ret);
+				goto next;
+			}
+
+			ret = ocfs2_inode_lock(inode, &di_bh, 1);
+			if (ret < 0) {
+				ocfs2_rw_unlock(inode, 1);
+				mlog_errno(ret);
+				goto next;
+			}
+
+			ret = ocfs2_truncate_file(inode, di_bh,
+					i_size_read(inode));
+			ocfs2_inode_unlock(inode, 1);
+			ocfs2_rw_unlock(inode, 1);
+			brelse(di_bh);
+			if (ret < 0) {
+				if (ret != -ENOSPC)
+					mlog_errno(ret);
+				goto next;
+			}
+
+			ret = ocfs2_del_inode_from_orphan(osb, inode, 0, 0);
+			if (ret)
+				mlog_errno(ret);
+
+			wake_up(&OCFS2_I(inode)->append_dio_wq);
+		} /* else if ORPHAN_NO_NEED_TRUNCATE, do nothing */
+
+next:
+		iput(inode);
+
+		inode = iter;
+	}
+
+	return ret;
+}
+
+static int __ocfs2_wait_on_mount(struct ocfs2_super *osb, int quota)
+{
+	/* This check is good because ocfs2 will wait on our recovery
+	 * thread before changing it to something other than MOUNTED
+	 * or DISABLED. */
+	wait_event(osb->osb_mount_event,
+		  (!quota && atomic_read(&osb->vol_state) == VOLUME_MOUNTED) ||
+		   atomic_read(&osb->vol_state) == VOLUME_MOUNTED_QUOTAS ||
+		   atomic_read(&osb->vol_state) == VOLUME_DISABLED);
+
+	/* If there's an error on mount, then we may never get to the
+	 * MOUNTED flag, but this is set right before
+	 * dismount_volume() so we can trust it. */
+	if (atomic_read(&osb->vol_state) == VOLUME_DISABLED) {
+		trace_ocfs2_wait_on_mount(VOLUME_DISABLED);
+		mlog(0, "mount error, exiting!\n");
+		return -EBUSY;
+	}
+
+	return 0;
+}
+
+static int ocfs2_commit_thread(void *arg)
+{
+	int status;
+	struct ocfs2_super *osb = arg;
+	struct ocfs2_journal *journal = osb->journal;
+
+	/* we can trust j_num_trans here because _should_stop() is only set in
+	 * shutdown and nobody other than ourselves should be able to start
+	 * transactions.  committing on shutdown might take a few iterations
+	 * as final transactions put deleted inodes on the list */
+	while (!(kthread_should_stop() &&
+		 atomic_read(&journal->j_num_trans) == 0)) {
+
+		wait_event_interruptible(osb->checkpoint_event,
+					 atomic_read(&journal->j_num_trans)
+					 || kthread_should_stop());
+
+		status = ocfs2_commit_cache(osb);
+		if (status < 0) {
+			static unsigned long abort_warn_time;
+
+			/* Warn about this once per minute */
+			if (printk_timed_ratelimit(&abort_warn_time, 60*HZ))
+				mlog(ML_ERROR, "status = %d, journal is "
+						"already aborted.\n", status);
+			/*
+			 * After ocfs2_commit_cache() fails, j_num_trans has a
+			 * non-zero value.  Sleep here to avoid a busy-wait
+			 * loop.
+			 */
+			msleep_interruptible(1000);
+		}
+
+		if (kthread_should_stop() && atomic_read(&journal->j_num_trans)){
+			mlog(ML_KTHREAD,
+			     "commit_thread: %u transactions pending on "
+			     "shutdown\n",
+			     atomic_read(&journal->j_num_trans));
+		}
+	}
+
+	return 0;
+}
+
+/* Reads all the journal inodes without taking any cluster locks. Used
+ * for hard readonly access to determine whether any journal requires
+ * recovery. Also used to refresh the recovery generation numbers after
+ * a journal has been recovered by another node.
+ */
+int ocfs2_check_journals_nolocks(struct ocfs2_super *osb)
+{
+	int ret = 0;
+	unsigned int slot;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di;
+	int journal_dirty = 0;
+
+	for(slot = 0; slot < osb->max_slots; slot++) {
+		ret = ocfs2_read_journal_inode(osb, slot, &di_bh, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		di = (struct ocfs2_dinode *) di_bh->b_data;
+
+		osb->slot_recovery_generations[slot] =
+					ocfs2_get_recovery_generation(di);
+
+		if (le32_to_cpu(di->id1.journal1.ij_flags) &
+		    OCFS2_JOURNAL_DIRTY_FL)
+			journal_dirty = 1;
+
+		brelse(di_bh);
+		di_bh = NULL;
+	}
+
+out:
+	if (journal_dirty)
+		ret = -EROFS;
+	return ret;
+}
diff --git a/fs/ocfs2/journal.h b/fs/ocfs2/journal.h
new file mode 100644
index 000000000..f4cd3c3e9
--- /dev/null
+++ b/fs/ocfs2/journal.h
@@ -0,0 +1,645 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * journal.h
+ *
+ * Defines journalling api and structures.
+ *
+ * Copyright (C) 2003, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_JOURNAL_H
+#define OCFS2_JOURNAL_H
+
+#include <linux/fs.h>
+#include <linux/jbd2.h>
+
+enum ocfs2_journal_state {
+	OCFS2_JOURNAL_FREE = 0,
+	OCFS2_JOURNAL_LOADED,
+	OCFS2_JOURNAL_IN_SHUTDOWN,
+};
+
+struct ocfs2_super;
+struct ocfs2_dinode;
+
+/*
+ * The recovery_list is a simple linked list of node numbers to recover.
+ * It is protected by the recovery_lock.
+ */
+
+struct ocfs2_recovery_map {
+	unsigned int rm_used;
+	unsigned int *rm_entries;
+};
+
+
+struct ocfs2_journal {
+	enum ocfs2_journal_state   j_state;    /* Journals current state   */
+
+	journal_t                 *j_journal; /* The kernels journal type */
+	struct inode              *j_inode;   /* Kernel inode pointing to
+					       * this journal             */
+	struct ocfs2_super        *j_osb;     /* pointer to the super
+					       * block for the node
+					       * we're currently
+					       * running on -- not
+					       * necessarily the super
+					       * block from the node
+					       * which we usually run
+					       * from (recovery,
+					       * etc)                     */
+	struct buffer_head        *j_bh;      /* Journal disk inode block */
+	atomic_t                  j_num_trans; /* Number of transactions
+					        * currently in the system. */
+	spinlock_t                j_lock;
+	unsigned long             j_trans_id;
+	struct rw_semaphore       j_trans_barrier;
+	wait_queue_head_t         j_checkpointed;
+
+	/* both fields protected by j_lock*/
+	struct list_head          j_la_cleanups;
+	struct work_struct        j_recovery_work;
+};
+
+extern spinlock_t trans_inc_lock;
+
+/* wrap j_trans_id so we never have it equal to zero. */
+static inline unsigned long ocfs2_inc_trans_id(struct ocfs2_journal *j)
+{
+	unsigned long old_id;
+	spin_lock(&trans_inc_lock);
+	old_id = j->j_trans_id++;
+	if (unlikely(!j->j_trans_id))
+		j->j_trans_id = 1;
+	spin_unlock(&trans_inc_lock);
+	return old_id;
+}
+
+static inline void ocfs2_set_ci_lock_trans(struct ocfs2_journal *journal,
+					   struct ocfs2_caching_info *ci)
+{
+	spin_lock(&trans_inc_lock);
+	ci->ci_last_trans = journal->j_trans_id;
+	spin_unlock(&trans_inc_lock);
+}
+
+/* Used to figure out whether it's safe to drop a metadata lock on an
+ * cached object. Returns true if all the object's changes have been
+ * checkpointed to disk. You should be holding the spinlock on the
+ * metadata lock while calling this to be sure that nobody can take
+ * the lock and put it on another transaction. */
+static inline int ocfs2_ci_fully_checkpointed(struct ocfs2_caching_info *ci)
+{
+	int ret;
+	struct ocfs2_journal *journal =
+		OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
+
+	spin_lock(&trans_inc_lock);
+	ret = time_after(journal->j_trans_id, ci->ci_last_trans);
+	spin_unlock(&trans_inc_lock);
+	return ret;
+}
+
+/* convenience function to check if an object backed by struct
+ * ocfs2_caching_info  is still new (has never hit disk) Will do you a
+ * favor and set created_trans = 0 when you've
+ * been checkpointed.  returns '1' if the ci is still new. */
+static inline int ocfs2_ci_is_new(struct ocfs2_caching_info *ci)
+{
+	int ret;
+	struct ocfs2_journal *journal =
+		OCFS2_SB(ocfs2_metadata_cache_get_super(ci))->journal;
+
+	spin_lock(&trans_inc_lock);
+	ret = !(time_after(journal->j_trans_id, ci->ci_created_trans));
+	if (!ret)
+		ci->ci_created_trans = 0;
+	spin_unlock(&trans_inc_lock);
+	return ret;
+}
+
+/* Wrapper for inodes so we can check system files */
+static inline int ocfs2_inode_is_new(struct inode *inode)
+{
+	/* System files are never "new" as they're written out by
+	 * mkfs. This helps us early during mount, before we have the
+	 * journal open and j_trans_id could be junk. */
+	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE)
+		return 0;
+
+	return ocfs2_ci_is_new(INODE_CACHE(inode));
+}
+
+static inline void ocfs2_ci_set_new(struct ocfs2_super *osb,
+				    struct ocfs2_caching_info *ci)
+{
+	spin_lock(&trans_inc_lock);
+	ci->ci_created_trans = osb->journal->j_trans_id;
+	spin_unlock(&trans_inc_lock);
+}
+
+/* Exported only for the journal struct init code in super.c. Do not call. */
+void ocfs2_orphan_scan_init(struct ocfs2_super *osb);
+void ocfs2_orphan_scan_start(struct ocfs2_super *osb);
+void ocfs2_orphan_scan_stop(struct ocfs2_super *osb);
+void ocfs2_orphan_scan_exit(struct ocfs2_super *osb);
+
+void ocfs2_complete_recovery(struct work_struct *work);
+void ocfs2_wait_for_recovery(struct ocfs2_super *osb);
+
+int ocfs2_recovery_init(struct ocfs2_super *osb);
+void ocfs2_recovery_exit(struct ocfs2_super *osb);
+
+int ocfs2_compute_replay_slots(struct ocfs2_super *osb);
+/*
+ *  Journal Control:
+ *  Initialize, Load, Shutdown, Wipe a journal.
+ *
+ *  ocfs2_journal_init     - Initialize journal structures in the OSB.
+ *  ocfs2_journal_load     - Load the given journal off disk. Replay it if
+ *                          there's transactions still in there.
+ *  ocfs2_journal_shutdown - Shutdown a journal, this will flush all
+ *                          uncommitted, uncheckpointed transactions.
+ *  ocfs2_journal_wipe     - Wipe transactions from a journal. Optionally
+ *                          zero out each block.
+ *  ocfs2_recovery_thread  - Perform recovery on a node. osb is our own osb.
+ *  ocfs2_mark_dead_nodes - Start recovery on nodes we won't get a heartbeat
+ *                          event on.
+ *  ocfs2_start_checkpoint - Kick the commit thread to do a checkpoint.
+ */
+void   ocfs2_set_journal_params(struct ocfs2_super *osb);
+int    ocfs2_journal_init(struct ocfs2_journal *journal,
+			  int *dirty);
+void   ocfs2_journal_shutdown(struct ocfs2_super *osb);
+int    ocfs2_journal_wipe(struct ocfs2_journal *journal,
+			  int full);
+int    ocfs2_journal_load(struct ocfs2_journal *journal, int local,
+			  int replayed);
+int    ocfs2_check_journals_nolocks(struct ocfs2_super *osb);
+void   ocfs2_recovery_thread(struct ocfs2_super *osb,
+			     int node_num);
+int    ocfs2_mark_dead_nodes(struct ocfs2_super *osb);
+void   ocfs2_complete_mount_recovery(struct ocfs2_super *osb);
+void ocfs2_complete_quota_recovery(struct ocfs2_super *osb);
+
+static inline void ocfs2_start_checkpoint(struct ocfs2_super *osb)
+{
+	wake_up(&osb->checkpoint_event);
+}
+
+static inline void ocfs2_checkpoint_inode(struct inode *inode)
+{
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (ocfs2_mount_local(osb))
+		return;
+
+	if (!ocfs2_ci_fully_checkpointed(INODE_CACHE(inode))) {
+		/* WARNING: This only kicks off a single
+		 * checkpoint. If someone races you and adds more
+		 * metadata to the journal, you won't know, and will
+		 * wind up waiting *a lot* longer than necessary. Right
+		 * now we only use this in clear_inode so that's
+		 * OK. */
+		ocfs2_start_checkpoint(osb);
+
+		wait_event(osb->journal->j_checkpointed,
+			   ocfs2_ci_fully_checkpointed(INODE_CACHE(inode)));
+	}
+}
+
+/*
+ *  Transaction Handling:
+ *  Manage the lifetime of a transaction handle.
+ *
+ *  ocfs2_start_trans      - Begin a transaction. Give it an upper estimate of
+ *                          the number of blocks that will be changed during
+ *                          this handle.
+ *  ocfs2_commit_trans - Complete a handle. It might return -EIO if
+ *                       the journal was aborted. The majority of paths don't
+ *                       check the return value as an error there comes too
+ *                       late to do anything (and will be picked up in a
+ *                       later transaction).
+ *  ocfs2_extend_trans     - Extend a handle by nblocks credits. This may
+ *                          commit the handle to disk in the process, but will
+ *                          not release any locks taken during the transaction.
+ *  ocfs2_journal_access* - Notify the handle that we want to journal this
+ *                          buffer. Will have to call ocfs2_journal_dirty once
+ *                          we've actually dirtied it. Type is one of . or .
+ *                          Always call the specific flavor of
+ *                          ocfs2_journal_access_*() unless you intend to
+ *                          manage the checksum by hand.
+ *  ocfs2_journal_dirty    - Mark a journalled buffer as having dirty data.
+ *  ocfs2_jbd2_file_inode  - Mark an inode so that its data goes out before
+ *                           the current handle commits.
+ */
+
+/* You must always start_trans with a number of buffs > 0, but it's
+ * perfectly legal to go through an entire transaction without having
+ * dirtied any buffers. */
+handle_t		    *ocfs2_start_trans(struct ocfs2_super *osb,
+					       int max_buffs);
+int			     ocfs2_commit_trans(struct ocfs2_super *osb,
+						handle_t *handle);
+int			     ocfs2_extend_trans(handle_t *handle, int nblocks);
+int			     ocfs2_allocate_extend_trans(handle_t *handle,
+						int thresh);
+
+/*
+ * Define an arbitrary limit for the amount of data we will anticipate
+ * writing to any given transaction.  For unbounded transactions such as
+ * fallocate(2) we can write more than this, but we always
+ * start off at the maximum transaction size and grow the transaction
+ * optimistically as we go.
+ */
+#define OCFS2_MAX_TRANS_DATA	64U
+
+/*
+ * Create access is for when we get a newly created buffer and we're
+ * not gonna read it off disk, but rather fill it ourselves.  Right
+ * now, we don't do anything special with this (it turns into a write
+ * request), but this is a good placeholder in case we do...
+ *
+ * Write access is for when we read a block off disk and are going to
+ * modify it. This way the journalling layer knows it may need to make
+ * a copy of that block (if it's part of another, uncommitted
+ * transaction) before we do so.
+ */
+#define OCFS2_JOURNAL_ACCESS_CREATE 0
+#define OCFS2_JOURNAL_ACCESS_WRITE  1
+#define OCFS2_JOURNAL_ACCESS_UNDO   2
+
+
+/* ocfs2_inode */
+int ocfs2_journal_access_di(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* ocfs2_extent_block */
+int ocfs2_journal_access_eb(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* ocfs2_refcount_block */
+int ocfs2_journal_access_rb(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* ocfs2_group_desc */
+int ocfs2_journal_access_gd(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* ocfs2_xattr_block */
+int ocfs2_journal_access_xb(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* quota blocks */
+int ocfs2_journal_access_dq(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* dirblock */
+int ocfs2_journal_access_db(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* ocfs2_dx_root_block */
+int ocfs2_journal_access_dr(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* ocfs2_dx_leaf */
+int ocfs2_journal_access_dl(handle_t *handle, struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh, int type);
+/* Anything that has no ecc */
+int ocfs2_journal_access(handle_t *handle, struct ocfs2_caching_info *ci,
+			 struct buffer_head *bh, int type);
+
+/*
+ * A word about the journal_access/journal_dirty "dance". It is
+ * entirely legal to journal_access a buffer more than once (as long
+ * as the access type is the same -- I'm not sure what will happen if
+ * access type is different but this should never happen anyway) It is
+ * also legal to journal_dirty a buffer more than once. In fact, you
+ * can even journal_access a buffer after you've done a
+ * journal_access/journal_dirty pair. The only thing you cannot do
+ * however, is journal_dirty a buffer which you haven't yet passed to
+ * journal_access at least once.
+ *
+ * That said, 99% of the time this doesn't matter and this is what the
+ * path looks like:
+ *
+ *	<read a bh>
+ *	ocfs2_journal_access(handle, bh,	OCFS2_JOURNAL_ACCESS_WRITE);
+ *	<modify the bh>
+ * 	ocfs2_journal_dirty(handle, bh);
+ */
+void ocfs2_journal_dirty(handle_t *handle, struct buffer_head *bh);
+
+/*
+ *  Credit Macros:
+ *  Convenience macros to calculate number of credits needed.
+ *
+ *  For convenience sake, I have a set of macros here which calculate
+ *  the *maximum* number of sectors which will be changed for various
+ *  metadata updates.
+ */
+
+/* simple file updates like chmod, etc. */
+#define OCFS2_INODE_UPDATE_CREDITS 1
+
+/* extended attribute block update */
+#define OCFS2_XATTR_BLOCK_UPDATE_CREDITS 1
+
+/* Update of a single quota block */
+#define OCFS2_QUOTA_BLOCK_UPDATE_CREDITS 1
+
+/* global quotafile inode update, data block */
+#define OCFS2_QINFO_WRITE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + \
+				   OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
+
+#define OCFS2_LOCAL_QINFO_WRITE_CREDITS OCFS2_QUOTA_BLOCK_UPDATE_CREDITS
+/*
+ * The two writes below can accidentally see global info dirty due
+ * to set_info() quotactl so make them prepared for the writes.
+ */
+/* quota data block, global info */
+/* Write to local quota file */
+#define OCFS2_QWRITE_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
+			      OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
+
+/* global quota data block, local quota data block, global quota inode,
+ * global quota info */
+#define OCFS2_QSYNC_CREDITS (OCFS2_QINFO_WRITE_CREDITS + \
+			     2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS)
+
+static inline int ocfs2_quota_trans_credits(struct super_block *sb)
+{
+	int credits = 0;
+
+	if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA))
+		credits += OCFS2_QWRITE_CREDITS;
+	if (OCFS2_HAS_RO_COMPAT_FEATURE(sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA))
+		credits += OCFS2_QWRITE_CREDITS;
+	return credits;
+}
+
+/* group extend. inode update and last group update. */
+#define OCFS2_GROUP_EXTEND_CREDITS	(OCFS2_INODE_UPDATE_CREDITS + 1)
+
+/* group add. inode update and the new group update. */
+#define OCFS2_GROUP_ADD_CREDITS	(OCFS2_INODE_UPDATE_CREDITS + 1)
+
+/* get one bit out of a suballocator: dinode + group descriptor +
+ * prev. group desc. if we relink. */
+#define OCFS2_SUBALLOC_ALLOC (3)
+
+static inline int ocfs2_inline_to_extents_credits(struct super_block *sb)
+{
+	return OCFS2_SUBALLOC_ALLOC + OCFS2_INODE_UPDATE_CREDITS +
+	       ocfs2_quota_trans_credits(sb);
+}
+
+/* dinode + group descriptor update. We don't relink on free yet. */
+#define OCFS2_SUBALLOC_FREE  (2)
+
+#define OCFS2_TRUNCATE_LOG_UPDATE OCFS2_INODE_UPDATE_CREDITS
+#define OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC (OCFS2_SUBALLOC_FREE 		      \
+					 + OCFS2_TRUNCATE_LOG_UPDATE)
+
+static inline int ocfs2_remove_extent_credits(struct super_block *sb)
+{
+	return OCFS2_TRUNCATE_LOG_UPDATE + OCFS2_INODE_UPDATE_CREDITS +
+	       ocfs2_quota_trans_credits(sb);
+}
+
+/* data block for new dir/symlink, allocation of directory block, dx_root
+ * update for free list */
+#define OCFS2_DIR_LINK_ADDITIONAL_CREDITS (1 + OCFS2_SUBALLOC_ALLOC + 1)
+
+static inline int ocfs2_add_dir_index_credits(struct super_block *sb)
+{
+	/* 1 block for index, 2 allocs (data, metadata), 1 clusters
+	 * worth of blocks for initial extent. */
+	return 1 + 2 * OCFS2_SUBALLOC_ALLOC +
+		ocfs2_clusters_to_blocks(sb, 1);
+}
+
+/* parent fe, parent block, new file entry, index leaf, inode alloc fe, inode
+ * alloc group descriptor + mkdir/symlink blocks + dir blocks + xattr
+ * blocks + quota update */
+static inline int ocfs2_mknod_credits(struct super_block *sb, int is_dir,
+				      int xattr_credits)
+{
+	int dir_credits = OCFS2_DIR_LINK_ADDITIONAL_CREDITS;
+
+	if (is_dir)
+		dir_credits += ocfs2_add_dir_index_credits(sb);
+
+	return 4 + OCFS2_SUBALLOC_ALLOC + dir_credits + xattr_credits +
+	       ocfs2_quota_trans_credits(sb);
+}
+
+/* local alloc metadata change + main bitmap updates */
+#define OCFS2_WINDOW_MOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS                 \
+				  + OCFS2_SUBALLOC_ALLOC + OCFS2_SUBALLOC_FREE)
+
+/* used when we don't need an allocation change for a dir extend. One
+ * for the dinode, one for the new block. */
+#define OCFS2_SIMPLE_DIR_EXTEND_CREDITS (2)
+
+/* file update (nlink, etc) + directory mtime/ctime + dir entry block + quota
+ * update on dir + index leaf + dx root update for free list +
+ * previous dirblock update in the free list */
+static inline int ocfs2_link_credits(struct super_block *sb)
+{
+	return 2*OCFS2_INODE_UPDATE_CREDITS + 4 +
+	       ocfs2_quota_trans_credits(sb);
+}
+
+/* inode + dir inode (if we unlink a dir), + dir entry block + orphan
+ * dir inode link + dir inode index leaf + dir index root */
+static inline int ocfs2_unlink_credits(struct super_block *sb)
+{
+	/* The quota update from ocfs2_link_credits is unused here... */
+	return 2 * OCFS2_INODE_UPDATE_CREDITS + 3 + ocfs2_link_credits(sb);
+}
+
+/* dinode + orphan dir dinode + inode alloc dinode + orphan dir entry +
+ * inode alloc group descriptor + orphan dir index root +
+ * orphan dir index leaf */
+#define OCFS2_DELETE_INODE_CREDITS (3 * OCFS2_INODE_UPDATE_CREDITS + 4)
+
+/* dinode + orphan dir dinode + extent tree leaf block + orphan dir entry +
+ * orphan dir index root + orphan dir index leaf */
+#define OCFS2_INODE_ADD_TO_ORPHAN_CREDITS  (2 * OCFS2_INODE_UPDATE_CREDITS + 4)
+#define OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS  OCFS2_INODE_ADD_TO_ORPHAN_CREDITS
+
+/* dinode update, old dir dinode update, new dir dinode update, old
+ * dir dir entry, new dir dir entry, dir entry update for renaming
+ * directory + target unlink + 3 x dir index leaves */
+static inline int ocfs2_rename_credits(struct super_block *sb)
+{
+	return 3 * OCFS2_INODE_UPDATE_CREDITS + 6 + ocfs2_unlink_credits(sb);
+}
+
+/* global bitmap dinode, group desc., relinked group,
+ * suballocator dinode, group desc., relinked group,
+ * dinode, xattr block */
+#define OCFS2_XATTR_BLOCK_CREATE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + \
+					  + OCFS2_INODE_UPDATE_CREDITS \
+					  + OCFS2_XATTR_BLOCK_UPDATE_CREDITS)
+
+/* inode update, removal of dx root block from allocator */
+#define OCFS2_DX_ROOT_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS +	\
+				      OCFS2_SUBALLOC_FREE)
+
+static inline int ocfs2_calc_dxi_expand_credits(struct super_block *sb)
+{
+	int credits = 1 + OCFS2_SUBALLOC_ALLOC;
+
+	credits += ocfs2_clusters_to_blocks(sb, 1);
+	credits += ocfs2_quota_trans_credits(sb);
+
+	return credits;
+}
+
+/* inode update, new refcount block and its allocation credits. */
+#define OCFS2_REFCOUNT_TREE_CREATE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1 \
+					    + OCFS2_SUBALLOC_ALLOC)
+
+/* inode and the refcount block update. */
+#define OCFS2_REFCOUNT_TREE_SET_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
+
+/*
+ * inode and the refcount block update.
+ * It doesn't include the credits for sub alloc change.
+ * So if we need to free the bit, OCFS2_SUBALLOC_FREE needs to be added.
+ */
+#define OCFS2_REFCOUNT_TREE_REMOVE_CREDITS (OCFS2_INODE_UPDATE_CREDITS + 1)
+
+/* 2 metadata alloc, 2 new blocks and root refcount block */
+#define OCFS2_EXPAND_REFCOUNT_TREE_CREDITS (OCFS2_SUBALLOC_ALLOC * 2 + 3)
+
+/*
+ * Please note that the caller must make sure that root_el is the root
+ * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
+ * the result may be wrong.
+ */
+static inline int ocfs2_calc_extend_credits(struct super_block *sb,
+					    struct ocfs2_extent_list *root_el)
+{
+	int bitmap_blocks, sysfile_bitmap_blocks, extent_blocks;
+
+	/* bitmap dinode, group desc. + relinked group. */
+	bitmap_blocks = OCFS2_SUBALLOC_ALLOC;
+
+	/* we might need to shift tree depth so lets assume an
+	 * absolute worst case of complete fragmentation.  Even with
+	 * that, we only need one update for the dinode, and then
+	 * however many metadata chunks needed * a remaining suballoc
+	 * alloc. */
+	sysfile_bitmap_blocks = 1 +
+		(OCFS2_SUBALLOC_ALLOC - 1) * ocfs2_extend_meta_needed(root_el);
+
+	/* this does not include *new* metadata blocks, which are
+	 * accounted for in sysfile_bitmap_blocks. root_el +
+	 * prev. last_eb_blk + blocks along edge of tree.
+	 * calc_symlink_credits passes because we just need 1
+	 * credit for the dinode there. */
+	extent_blocks = 1 + 1 + le16_to_cpu(root_el->l_tree_depth);
+
+	return bitmap_blocks + sysfile_bitmap_blocks + extent_blocks +
+	       ocfs2_quota_trans_credits(sb);
+}
+
+static inline int ocfs2_calc_symlink_credits(struct super_block *sb)
+{
+	int blocks = ocfs2_mknod_credits(sb, 0, 0);
+
+	/* links can be longer than one block so we may update many
+	 * within our single allocated extent. */
+	blocks += ocfs2_clusters_to_blocks(sb, 1);
+
+	return blocks + ocfs2_quota_trans_credits(sb);
+}
+
+static inline int ocfs2_calc_group_alloc_credits(struct super_block *sb,
+						 unsigned int cpg)
+{
+	int blocks;
+	int bitmap_blocks = OCFS2_SUBALLOC_ALLOC + 1;
+	/* parent inode update + new block group header + bitmap inode update
+	   + bitmap blocks affected */
+	blocks = 1 + 1 + 1 + bitmap_blocks;
+	return blocks;
+}
+
+/*
+ * Allocating a discontiguous block group requires the credits from
+ * ocfs2_calc_group_alloc_credits() as well as enough credits to fill
+ * the group descriptor's extent list.  The caller already has started
+ * the transaction with ocfs2_calc_group_alloc_credits().  They extend
+ * it with these credits.
+ */
+static inline int ocfs2_calc_bg_discontig_credits(struct super_block *sb)
+{
+	return ocfs2_extent_recs_per_gd(sb);
+}
+
+static inline int ocfs2_calc_tree_trunc_credits(struct super_block *sb,
+						unsigned int clusters_to_del,
+						struct ocfs2_dinode *fe,
+						struct ocfs2_extent_list *last_el)
+{
+ 	/* for dinode + all headers in this pass + update to next leaf */
+	u16 next_free = le16_to_cpu(last_el->l_next_free_rec);
+	u16 tree_depth = le16_to_cpu(fe->id2.i_list.l_tree_depth);
+	int credits = 1 + tree_depth + 1;
+	int i;
+
+	i = next_free - 1;
+	BUG_ON(i < 0);
+
+	/* We may be deleting metadata blocks, so metadata alloc dinode +
+	   one desc. block for each possible delete. */
+	if (tree_depth && next_free == 1 &&
+	    ocfs2_rec_clusters(last_el, &last_el->l_recs[i]) == clusters_to_del)
+		credits += 1 + tree_depth;
+
+	/* update to the truncate log. */
+	credits += OCFS2_TRUNCATE_LOG_UPDATE;
+
+	credits += ocfs2_quota_trans_credits(sb);
+
+	return credits;
+}
+
+static inline int ocfs2_jbd2_file_inode(handle_t *handle, struct inode *inode)
+{
+	return jbd2_journal_file_inode(handle, &OCFS2_I(inode)->ip_jinode);
+}
+
+static inline int ocfs2_begin_ordered_truncate(struct inode *inode,
+					       loff_t new_size)
+{
+	return jbd2_journal_begin_ordered_truncate(
+				OCFS2_SB(inode->i_sb)->journal->j_journal,
+				&OCFS2_I(inode)->ip_jinode,
+				new_size);
+}
+
+static inline void ocfs2_update_inode_fsync_trans(handle_t *handle,
+						  struct inode *inode,
+						  int datasync)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	oi->i_sync_tid = handle->h_transaction->t_tid;
+	if (datasync)
+		oi->i_datasync_tid = handle->h_transaction->t_tid;
+}
+
+#endif /* OCFS2_JOURNAL_H */
diff --git a/fs/ocfs2/localalloc.c b/fs/ocfs2/localalloc.c
new file mode 100644
index 000000000..857bbbcd3
--- /dev/null
+++ b/fs/ocfs2/localalloc.c
@@ -0,0 +1,1343 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * localalloc.c
+ *
+ * Node local data allocation
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/bitops.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "blockcheck.h"
+#include "dlmglue.h"
+#include "inode.h"
+#include "journal.h"
+#include "localalloc.h"
+#include "suballoc.h"
+#include "super.h"
+#include "sysfile.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+#define OCFS2_LOCAL_ALLOC(dinode)	(&((dinode)->id2.i_lab))
+
+static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc);
+
+static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
+					     struct ocfs2_dinode *alloc,
+					     u32 *numbits,
+					     struct ocfs2_alloc_reservation *resv);
+
+static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc);
+
+static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
+				    handle_t *handle,
+				    struct ocfs2_dinode *alloc,
+				    struct inode *main_bm_inode,
+				    struct buffer_head *main_bm_bh);
+
+static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
+						struct ocfs2_alloc_context **ac,
+						struct inode **bitmap_inode,
+						struct buffer_head **bitmap_bh);
+
+static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
+					handle_t *handle,
+					struct ocfs2_alloc_context *ac);
+
+static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
+					  struct inode *local_alloc_inode);
+
+/*
+ * ocfs2_la_default_mb() - determine a default size, in megabytes of
+ * the local alloc.
+ *
+ * Generally, we'd like to pick as large a local alloc as
+ * possible. Performance on large workloads tends to scale
+ * proportionally to la size. In addition to that, the reservations
+ * code functions more efficiently as it can reserve more windows for
+ * write.
+ *
+ * Some things work against us when trying to choose a large local alloc:
+ *
+ * - We need to ensure our sizing is picked to leave enough space in
+ *   group descriptors for other allocations (such as block groups,
+ *   etc). Picking default sizes which are a multiple of 4 could help
+ *   - block groups are allocated in 2mb and 4mb chunks.
+ *
+ * - Likewise, we don't want to starve other nodes of bits on small
+ *   file systems. This can easily be taken care of by limiting our
+ *   default to a reasonable size (256M) on larger cluster sizes.
+ *
+ * - Some file systems can't support very large sizes - 4k and 8k in
+ *   particular are limited to less than 128 and 256 megabytes respectively.
+ *
+ * The following reference table shows group descriptor and local
+ * alloc maximums at various cluster sizes (4k blocksize)
+ *
+ * csize: 4K	group: 126M	la: 121M
+ * csize: 8K	group: 252M	la: 243M
+ * csize: 16K	group: 504M	la: 486M
+ * csize: 32K	group: 1008M	la: 972M
+ * csize: 64K	group: 2016M	la: 1944M
+ * csize: 128K	group: 4032M	la: 3888M
+ * csize: 256K	group: 8064M	la: 7776M
+ * csize: 512K	group: 16128M	la: 15552M
+ * csize: 1024K	group: 32256M	la: 31104M
+ */
+#define	OCFS2_LA_MAX_DEFAULT_MB	256
+#define	OCFS2_LA_OLD_DEFAULT	8
+unsigned int ocfs2_la_default_mb(struct ocfs2_super *osb)
+{
+	unsigned int la_mb;
+	unsigned int gd_mb;
+	unsigned int la_max_mb;
+	unsigned int megs_per_slot;
+	struct super_block *sb = osb->sb;
+
+	gd_mb = ocfs2_clusters_to_megabytes(osb->sb,
+		8 * ocfs2_group_bitmap_size(sb, 0, osb->s_feature_incompat));
+
+	/*
+	 * This takes care of files systems with very small group
+	 * descriptors - 512 byte blocksize at cluster sizes lower
+	 * than 16K and also 1k blocksize with 4k cluster size.
+	 */
+	if ((sb->s_blocksize == 512 && osb->s_clustersize <= 8192)
+	    || (sb->s_blocksize == 1024 && osb->s_clustersize == 4096))
+		return OCFS2_LA_OLD_DEFAULT;
+
+	/*
+	 * Leave enough room for some block groups and make the final
+	 * value we work from a multiple of 4.
+	 */
+	gd_mb -= 16;
+	gd_mb &= 0xFFFFFFFB;
+
+	la_mb = gd_mb;
+
+	/*
+	 * Keep window sizes down to a reasonable default
+	 */
+	if (la_mb > OCFS2_LA_MAX_DEFAULT_MB) {
+		/*
+		 * Some clustersize / blocksize combinations will have
+		 * given us a larger than OCFS2_LA_MAX_DEFAULT_MB
+		 * default size, but get poor distribution when
+		 * limited to exactly 256 megabytes.
+		 *
+		 * As an example, 16K clustersize at 4K blocksize
+		 * gives us a cluster group size of 504M. Paring the
+		 * local alloc size down to 256 however, would give us
+		 * only one window and around 200MB left in the
+		 * cluster group. Instead, find the first size below
+		 * 256 which would give us an even distribution.
+		 *
+		 * Larger cluster group sizes actually work out pretty
+		 * well when pared to 256, so we don't have to do this
+		 * for any group that fits more than two
+		 * OCFS2_LA_MAX_DEFAULT_MB windows.
+		 */
+		if (gd_mb > (2 * OCFS2_LA_MAX_DEFAULT_MB))
+			la_mb = 256;
+		else {
+			unsigned int gd_mult = gd_mb;
+
+			while (gd_mult > 256)
+				gd_mult = gd_mult >> 1;
+
+			la_mb = gd_mult;
+		}
+	}
+
+	megs_per_slot = osb->osb_clusters_at_boot / osb->max_slots;
+	megs_per_slot = ocfs2_clusters_to_megabytes(osb->sb, megs_per_slot);
+	/* Too many nodes, too few disk clusters. */
+	if (megs_per_slot < la_mb)
+		la_mb = megs_per_slot;
+
+	/* We can't store more bits than we can in a block. */
+	la_max_mb = ocfs2_clusters_to_megabytes(osb->sb,
+						ocfs2_local_alloc_size(sb) * 8);
+	if (la_mb > la_max_mb)
+		la_mb = la_max_mb;
+
+	return la_mb;
+}
+
+void ocfs2_la_set_sizes(struct ocfs2_super *osb, int requested_mb)
+{
+	struct super_block *sb = osb->sb;
+	unsigned int la_default_mb = ocfs2_la_default_mb(osb);
+	unsigned int la_max_mb;
+
+	la_max_mb = ocfs2_clusters_to_megabytes(sb,
+						ocfs2_local_alloc_size(sb) * 8);
+
+	trace_ocfs2_la_set_sizes(requested_mb, la_max_mb, la_default_mb);
+
+	if (requested_mb == -1) {
+		/* No user request - use defaults */
+		osb->local_alloc_default_bits =
+			ocfs2_megabytes_to_clusters(sb, la_default_mb);
+	} else if (requested_mb > la_max_mb) {
+		/* Request is too big, we give the maximum available */
+		osb->local_alloc_default_bits =
+			ocfs2_megabytes_to_clusters(sb, la_max_mb);
+	} else {
+		osb->local_alloc_default_bits =
+			ocfs2_megabytes_to_clusters(sb, requested_mb);
+	}
+
+	osb->local_alloc_bits = osb->local_alloc_default_bits;
+}
+
+static inline int ocfs2_la_state_enabled(struct ocfs2_super *osb)
+{
+	return (osb->local_alloc_state == OCFS2_LA_THROTTLED ||
+		osb->local_alloc_state == OCFS2_LA_ENABLED);
+}
+
+void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb,
+				      unsigned int num_clusters)
+{
+	spin_lock(&osb->osb_lock);
+	if (osb->local_alloc_state == OCFS2_LA_DISABLED ||
+	    osb->local_alloc_state == OCFS2_LA_THROTTLED)
+		if (num_clusters >= osb->local_alloc_default_bits) {
+			cancel_delayed_work(&osb->la_enable_wq);
+			osb->local_alloc_state = OCFS2_LA_ENABLED;
+		}
+	spin_unlock(&osb->osb_lock);
+}
+
+void ocfs2_la_enable_worker(struct work_struct *work)
+{
+	struct ocfs2_super *osb =
+		container_of(work, struct ocfs2_super,
+			     la_enable_wq.work);
+	spin_lock(&osb->osb_lock);
+	osb->local_alloc_state = OCFS2_LA_ENABLED;
+	spin_unlock(&osb->osb_lock);
+}
+
+/*
+ * Tell us whether a given allocation should use the local alloc
+ * file. Otherwise, it has to go to the main bitmap.
+ *
+ * This function does semi-dirty reads of local alloc size and state!
+ * This is ok however, as the values are re-checked once under mutex.
+ */
+int ocfs2_alloc_should_use_local(struct ocfs2_super *osb, u64 bits)
+{
+	int ret = 0;
+	int la_bits;
+
+	spin_lock(&osb->osb_lock);
+	la_bits = osb->local_alloc_bits;
+
+	if (!ocfs2_la_state_enabled(osb))
+		goto bail;
+
+	/* la_bits should be at least twice the size (in clusters) of
+	 * a new block group. We want to be sure block group
+	 * allocations go through the local alloc, so allow an
+	 * allocation to take up to half the bitmap. */
+	if (bits > (la_bits / 2))
+		goto bail;
+
+	ret = 1;
+bail:
+	trace_ocfs2_alloc_should_use_local(
+	     (unsigned long long)bits, osb->local_alloc_state, la_bits, ret);
+	spin_unlock(&osb->osb_lock);
+	return ret;
+}
+
+int ocfs2_load_local_alloc(struct ocfs2_super *osb)
+{
+	int status = 0;
+	struct ocfs2_dinode *alloc = NULL;
+	struct buffer_head *alloc_bh = NULL;
+	u32 num_used;
+	struct inode *inode = NULL;
+	struct ocfs2_local_alloc *la;
+
+	if (osb->local_alloc_bits == 0)
+		goto bail;
+
+	if (osb->local_alloc_bits >= osb->bitmap_cpg) {
+		mlog(ML_NOTICE, "Requested local alloc window %d is larger "
+		     "than max possible %u. Using defaults.\n",
+		     osb->local_alloc_bits, (osb->bitmap_cpg - 1));
+		osb->local_alloc_bits =
+			ocfs2_megabytes_to_clusters(osb->sb,
+						    ocfs2_la_default_mb(osb));
+	}
+
+	/* read the alloc off disk */
+	inode = ocfs2_get_system_file_inode(osb, LOCAL_ALLOC_SYSTEM_INODE,
+					    osb->slot_num);
+	if (!inode) {
+		status = -EINVAL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_read_inode_block_full(inode, &alloc_bh,
+					     OCFS2_BH_IGNORE_CACHE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
+	la = OCFS2_LOCAL_ALLOC(alloc);
+
+	if (!(le32_to_cpu(alloc->i_flags) &
+	    (OCFS2_LOCAL_ALLOC_FL|OCFS2_BITMAP_FL))) {
+		mlog(ML_ERROR, "Invalid local alloc inode, %llu\n",
+		     (unsigned long long)OCFS2_I(inode)->ip_blkno);
+		status = -EINVAL;
+		goto bail;
+	}
+
+	if ((la->la_size == 0) ||
+	    (le16_to_cpu(la->la_size) > ocfs2_local_alloc_size(inode->i_sb))) {
+		mlog(ML_ERROR, "Local alloc size is invalid (la_size = %u)\n",
+		     le16_to_cpu(la->la_size));
+		status = -EINVAL;
+		goto bail;
+	}
+
+	/* do a little verification. */
+	num_used = ocfs2_local_alloc_count_bits(alloc);
+
+	/* hopefully the local alloc has always been recovered before
+	 * we load it. */
+	if (num_used
+	    || alloc->id1.bitmap1.i_used
+	    || alloc->id1.bitmap1.i_total
+	    || la->la_bm_off)
+		mlog(ML_ERROR, "Local alloc hasn't been recovered!\n"
+		     "found = %u, set = %u, taken = %u, off = %u\n",
+		     num_used, le32_to_cpu(alloc->id1.bitmap1.i_used),
+		     le32_to_cpu(alloc->id1.bitmap1.i_total),
+		     OCFS2_LOCAL_ALLOC(alloc)->la_bm_off);
+
+	osb->local_alloc_bh = alloc_bh;
+	osb->local_alloc_state = OCFS2_LA_ENABLED;
+
+bail:
+	if (status < 0)
+		brelse(alloc_bh);
+	if (inode)
+		iput(inode);
+
+	trace_ocfs2_load_local_alloc(osb->local_alloc_bits);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/*
+ * return any unused bits to the bitmap and write out a clean
+ * local_alloc.
+ *
+ * local_alloc_bh is optional. If not passed, we will simply use the
+ * one off osb. If you do pass it however, be warned that it *will* be
+ * returned brelse'd and NULL'd out.*/
+void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb)
+{
+	int status;
+	handle_t *handle;
+	struct inode *local_alloc_inode = NULL;
+	struct buffer_head *bh = NULL;
+	struct buffer_head *main_bm_bh = NULL;
+	struct inode *main_bm_inode = NULL;
+	struct ocfs2_dinode *alloc_copy = NULL;
+	struct ocfs2_dinode *alloc = NULL;
+
+	cancel_delayed_work(&osb->la_enable_wq);
+	flush_workqueue(ocfs2_wq);
+
+	if (osb->local_alloc_state == OCFS2_LA_UNUSED)
+		goto out;
+
+	local_alloc_inode =
+		ocfs2_get_system_file_inode(osb,
+					    LOCAL_ALLOC_SYSTEM_INODE,
+					    osb->slot_num);
+	if (!local_alloc_inode) {
+		status = -ENOENT;
+		mlog_errno(status);
+		goto out;
+	}
+
+	osb->local_alloc_state = OCFS2_LA_DISABLED;
+
+	ocfs2_resmap_uninit(&osb->osb_la_resmap);
+
+	main_bm_inode = ocfs2_get_system_file_inode(osb,
+						    GLOBAL_BITMAP_SYSTEM_INODE,
+						    OCFS2_INVALID_SLOT);
+	if (!main_bm_inode) {
+		status = -EINVAL;
+		mlog_errno(status);
+		goto out;
+	}
+
+	mutex_lock(&main_bm_inode->i_mutex);
+
+	status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_mutex;
+	}
+
+	/* WINDOW_MOVE_CREDITS is a bit heavy... */
+	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
+	if (IS_ERR(handle)) {
+		mlog_errno(PTR_ERR(handle));
+		handle = NULL;
+		goto out_unlock;
+	}
+
+	bh = osb->local_alloc_bh;
+	alloc = (struct ocfs2_dinode *) bh->b_data;
+
+	alloc_copy = kmalloc(bh->b_size, GFP_NOFS);
+	if (!alloc_copy) {
+		status = -ENOMEM;
+		goto out_commit;
+	}
+	memcpy(alloc_copy, alloc, bh->b_size);
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(local_alloc_inode),
+					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	ocfs2_clear_local_alloc(alloc);
+	ocfs2_journal_dirty(handle, bh);
+
+	brelse(bh);
+	osb->local_alloc_bh = NULL;
+	osb->local_alloc_state = OCFS2_LA_UNUSED;
+
+	status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
+					  main_bm_inode, main_bm_bh);
+	if (status < 0)
+		mlog_errno(status);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out_unlock:
+	brelse(main_bm_bh);
+
+	ocfs2_inode_unlock(main_bm_inode, 1);
+
+out_mutex:
+	mutex_unlock(&main_bm_inode->i_mutex);
+	iput(main_bm_inode);
+
+out:
+	if (local_alloc_inode)
+		iput(local_alloc_inode);
+
+	kfree(alloc_copy);
+}
+
+/*
+ * We want to free the bitmap bits outside of any recovery context as
+ * we'll need a cluster lock to do so, but we must clear the local
+ * alloc before giving up the recovered nodes journal. To solve this,
+ * we kmalloc a copy of the local alloc before it's change for the
+ * caller to process with ocfs2_complete_local_alloc_recovery
+ */
+int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
+				     int slot_num,
+				     struct ocfs2_dinode **alloc_copy)
+{
+	int status = 0;
+	struct buffer_head *alloc_bh = NULL;
+	struct inode *inode = NULL;
+	struct ocfs2_dinode *alloc;
+
+	trace_ocfs2_begin_local_alloc_recovery(slot_num);
+
+	*alloc_copy = NULL;
+
+	inode = ocfs2_get_system_file_inode(osb,
+					    LOCAL_ALLOC_SYSTEM_INODE,
+					    slot_num);
+	if (!inode) {
+		status = -EINVAL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	mutex_lock(&inode->i_mutex);
+
+	status = ocfs2_read_inode_block_full(inode, &alloc_bh,
+					     OCFS2_BH_IGNORE_CACHE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	*alloc_copy = kmalloc(alloc_bh->b_size, GFP_KERNEL);
+	if (!(*alloc_copy)) {
+		status = -ENOMEM;
+		goto bail;
+	}
+	memcpy((*alloc_copy), alloc_bh->b_data, alloc_bh->b_size);
+
+	alloc = (struct ocfs2_dinode *) alloc_bh->b_data;
+	ocfs2_clear_local_alloc(alloc);
+
+	ocfs2_compute_meta_ecc(osb->sb, alloc_bh->b_data, &alloc->i_check);
+	status = ocfs2_write_block(osb, alloc_bh, INODE_CACHE(inode));
+	if (status < 0)
+		mlog_errno(status);
+
+bail:
+	if (status < 0) {
+		kfree(*alloc_copy);
+		*alloc_copy = NULL;
+	}
+
+	brelse(alloc_bh);
+
+	if (inode) {
+		mutex_unlock(&inode->i_mutex);
+		iput(inode);
+	}
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/*
+ * Step 2: By now, we've completed the journal recovery, we've stamped
+ * a clean local alloc on disk and dropped the node out of the
+ * recovery map. Dlm locks will no longer stall, so lets clear out the
+ * main bitmap.
+ */
+int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
+					struct ocfs2_dinode *alloc)
+{
+	int status;
+	handle_t *handle;
+	struct buffer_head *main_bm_bh = NULL;
+	struct inode *main_bm_inode;
+
+	main_bm_inode = ocfs2_get_system_file_inode(osb,
+						    GLOBAL_BITMAP_SYSTEM_INODE,
+						    OCFS2_INVALID_SLOT);
+	if (!main_bm_inode) {
+		status = -EINVAL;
+		mlog_errno(status);
+		goto out;
+	}
+
+	mutex_lock(&main_bm_inode->i_mutex);
+
+	status = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_mutex;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto out_unlock;
+	}
+
+	/* we want the bitmap change to be recorded on disk asap */
+	handle->h_sync = 1;
+
+	status = ocfs2_sync_local_to_main(osb, handle, alloc,
+					  main_bm_inode, main_bm_bh);
+	if (status < 0)
+		mlog_errno(status);
+
+	ocfs2_commit_trans(osb, handle);
+
+out_unlock:
+	ocfs2_inode_unlock(main_bm_inode, 1);
+
+out_mutex:
+	mutex_unlock(&main_bm_inode->i_mutex);
+
+	brelse(main_bm_bh);
+
+	iput(main_bm_inode);
+
+out:
+	if (!status)
+		ocfs2_init_steal_slots(osb);
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/*
+ * make sure we've got at least bits_wanted contiguous bits in the
+ * local alloc. You lose them when you drop i_mutex.
+ *
+ * We will add ourselves to the transaction passed in, but may start
+ * our own in order to shift windows.
+ */
+int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
+				   u32 bits_wanted,
+				   struct ocfs2_alloc_context *ac)
+{
+	int status;
+	struct ocfs2_dinode *alloc;
+	struct inode *local_alloc_inode;
+	unsigned int free_bits;
+
+	BUG_ON(!ac);
+
+	local_alloc_inode =
+		ocfs2_get_system_file_inode(osb,
+					    LOCAL_ALLOC_SYSTEM_INODE,
+					    osb->slot_num);
+	if (!local_alloc_inode) {
+		status = -ENOENT;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	mutex_lock(&local_alloc_inode->i_mutex);
+
+	/*
+	 * We must double check state and allocator bits because
+	 * another process may have changed them while holding i_mutex.
+	 */
+	spin_lock(&osb->osb_lock);
+	if (!ocfs2_la_state_enabled(osb) ||
+	    (bits_wanted > osb->local_alloc_bits)) {
+		spin_unlock(&osb->osb_lock);
+		status = -ENOSPC;
+		goto bail;
+	}
+	spin_unlock(&osb->osb_lock);
+
+	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
+
+#ifdef CONFIG_OCFS2_DEBUG_FS
+	if (le32_to_cpu(alloc->id1.bitmap1.i_used) !=
+	    ocfs2_local_alloc_count_bits(alloc)) {
+		ocfs2_error(osb->sb, "local alloc inode %llu says it has "
+			    "%u used bits, but a count shows %u",
+			    (unsigned long long)le64_to_cpu(alloc->i_blkno),
+			    le32_to_cpu(alloc->id1.bitmap1.i_used),
+			    ocfs2_local_alloc_count_bits(alloc));
+		status = -EIO;
+		goto bail;
+	}
+#endif
+
+	free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
+		le32_to_cpu(alloc->id1.bitmap1.i_used);
+	if (bits_wanted > free_bits) {
+		/* uhoh, window change time. */
+		status =
+			ocfs2_local_alloc_slide_window(osb, local_alloc_inode);
+		if (status < 0) {
+			if (status != -ENOSPC)
+				mlog_errno(status);
+			goto bail;
+		}
+
+		/*
+		 * Under certain conditions, the window slide code
+		 * might have reduced the number of bits available or
+		 * disabled the the local alloc entirely. Re-check
+		 * here and return -ENOSPC if necessary.
+		 */
+		status = -ENOSPC;
+		if (!ocfs2_la_state_enabled(osb))
+			goto bail;
+
+		free_bits = le32_to_cpu(alloc->id1.bitmap1.i_total) -
+			le32_to_cpu(alloc->id1.bitmap1.i_used);
+		if (bits_wanted > free_bits)
+			goto bail;
+	}
+
+	ac->ac_inode = local_alloc_inode;
+	/* We should never use localalloc from another slot */
+	ac->ac_alloc_slot = osb->slot_num;
+	ac->ac_which = OCFS2_AC_USE_LOCAL;
+	get_bh(osb->local_alloc_bh);
+	ac->ac_bh = osb->local_alloc_bh;
+	status = 0;
+bail:
+	if (status < 0 && local_alloc_inode) {
+		mutex_unlock(&local_alloc_inode->i_mutex);
+		iput(local_alloc_inode);
+	}
+
+	trace_ocfs2_reserve_local_alloc_bits(
+		(unsigned long long)ac->ac_max_block,
+		bits_wanted, osb->slot_num, status);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
+				 handle_t *handle,
+				 struct ocfs2_alloc_context *ac,
+				 u32 bits_wanted,
+				 u32 *bit_off,
+				 u32 *num_bits)
+{
+	int status, start;
+	struct inode *local_alloc_inode;
+	void *bitmap;
+	struct ocfs2_dinode *alloc;
+	struct ocfs2_local_alloc *la;
+
+	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
+
+	local_alloc_inode = ac->ac_inode;
+	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
+	la = OCFS2_LOCAL_ALLOC(alloc);
+
+	start = ocfs2_local_alloc_find_clear_bits(osb, alloc, &bits_wanted,
+						  ac->ac_resv);
+	if (start == -1) {
+		/* TODO: Shouldn't we just BUG here? */
+		status = -ENOSPC;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	bitmap = la->la_bitmap;
+	*bit_off = le32_to_cpu(la->la_bm_off) + start;
+	*num_bits = bits_wanted;
+
+	status = ocfs2_journal_access_di(handle,
+					 INODE_CACHE(local_alloc_inode),
+					 osb->local_alloc_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	ocfs2_resmap_claimed_bits(&osb->osb_la_resmap, ac->ac_resv, start,
+				  bits_wanted);
+
+	while(bits_wanted--)
+		ocfs2_set_bit(start++, bitmap);
+
+	le32_add_cpu(&alloc->id1.bitmap1.i_used, *num_bits);
+	ocfs2_journal_dirty(handle, osb->local_alloc_bh);
+
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+int ocfs2_free_local_alloc_bits(struct ocfs2_super *osb,
+				handle_t *handle,
+				struct ocfs2_alloc_context *ac,
+				u32 bit_off,
+				u32 num_bits)
+{
+	int status, start;
+	u32 clear_bits;
+	struct inode *local_alloc_inode;
+	void *bitmap;
+	struct ocfs2_dinode *alloc;
+	struct ocfs2_local_alloc *la;
+
+	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL);
+
+	local_alloc_inode = ac->ac_inode;
+	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
+	la = OCFS2_LOCAL_ALLOC(alloc);
+
+	bitmap = la->la_bitmap;
+	start = bit_off - le32_to_cpu(la->la_bm_off);
+	clear_bits = num_bits;
+
+	status = ocfs2_journal_access_di(handle,
+			INODE_CACHE(local_alloc_inode),
+			osb->local_alloc_bh,
+			OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	while (clear_bits--)
+		ocfs2_clear_bit(start++, bitmap);
+
+	le32_add_cpu(&alloc->id1.bitmap1.i_used, -num_bits);
+	ocfs2_journal_dirty(handle, osb->local_alloc_bh);
+
+bail:
+	return status;
+}
+
+static u32 ocfs2_local_alloc_count_bits(struct ocfs2_dinode *alloc)
+{
+	u32 count;
+	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
+
+	count = memweight(la->la_bitmap, le16_to_cpu(la->la_size));
+
+	trace_ocfs2_local_alloc_count_bits(count);
+	return count;
+}
+
+static int ocfs2_local_alloc_find_clear_bits(struct ocfs2_super *osb,
+				     struct ocfs2_dinode *alloc,
+				     u32 *numbits,
+				     struct ocfs2_alloc_reservation *resv)
+{
+	int numfound = 0, bitoff, left, startoff, lastzero;
+	int local_resv = 0;
+	struct ocfs2_alloc_reservation r;
+	void *bitmap = NULL;
+	struct ocfs2_reservation_map *resmap = &osb->osb_la_resmap;
+
+	if (!alloc->id1.bitmap1.i_total) {
+		bitoff = -1;
+		goto bail;
+	}
+
+	if (!resv) {
+		local_resv = 1;
+		ocfs2_resv_init_once(&r);
+		ocfs2_resv_set_type(&r, OCFS2_RESV_FLAG_TMP);
+		resv = &r;
+	}
+
+	numfound = *numbits;
+	if (ocfs2_resmap_resv_bits(resmap, resv, &bitoff, &numfound) == 0) {
+		if (numfound < *numbits)
+			*numbits = numfound;
+		goto bail;
+	}
+
+	/*
+	 * Code error. While reservations are enabled, local
+	 * allocation should _always_ go through them.
+	 */
+	BUG_ON(osb->osb_resv_level != 0);
+
+	/*
+	 * Reservations are disabled. Handle this the old way.
+	 */
+
+	bitmap = OCFS2_LOCAL_ALLOC(alloc)->la_bitmap;
+
+	numfound = bitoff = startoff = 0;
+	lastzero = -1;
+	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
+	while ((bitoff = ocfs2_find_next_zero_bit(bitmap, left, startoff)) != -1) {
+		if (bitoff == left) {
+			/* mlog(0, "bitoff (%d) == left", bitoff); */
+			break;
+		}
+		/* mlog(0, "Found a zero: bitoff = %d, startoff = %d, "
+		   "numfound = %d\n", bitoff, startoff, numfound);*/
+
+		/* Ok, we found a zero bit... is it contig. or do we
+		 * start over?*/
+		if (bitoff == startoff) {
+			/* we found a zero */
+			numfound++;
+			startoff++;
+		} else {
+			/* got a zero after some ones */
+			numfound = 1;
+			startoff = bitoff+1;
+		}
+		/* we got everything we needed */
+		if (numfound == *numbits) {
+			/* mlog(0, "Found it all!\n"); */
+			break;
+		}
+	}
+
+	trace_ocfs2_local_alloc_find_clear_bits_search_bitmap(bitoff, numfound);
+
+	if (numfound == *numbits)
+		bitoff = startoff - numfound;
+	else
+		bitoff = -1;
+
+bail:
+	if (local_resv)
+		ocfs2_resv_discard(resmap, resv);
+
+	trace_ocfs2_local_alloc_find_clear_bits(*numbits,
+		le32_to_cpu(alloc->id1.bitmap1.i_total),
+		bitoff, numfound);
+
+	return bitoff;
+}
+
+static void ocfs2_clear_local_alloc(struct ocfs2_dinode *alloc)
+{
+	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
+	int i;
+
+	alloc->id1.bitmap1.i_total = 0;
+	alloc->id1.bitmap1.i_used = 0;
+	la->la_bm_off = 0;
+	for(i = 0; i < le16_to_cpu(la->la_size); i++)
+		la->la_bitmap[i] = 0;
+}
+
+#if 0
+/* turn this on and uncomment below to aid debugging window shifts. */
+static void ocfs2_verify_zero_bits(unsigned long *bitmap,
+				   unsigned int start,
+				   unsigned int count)
+{
+	unsigned int tmp = count;
+	while(tmp--) {
+		if (ocfs2_test_bit(start + tmp, bitmap)) {
+			printk("ocfs2_verify_zero_bits: start = %u, count = "
+			       "%u\n", start, count);
+			printk("ocfs2_verify_zero_bits: bit %u is set!",
+			       start + tmp);
+			BUG();
+		}
+	}
+}
+#endif
+
+/*
+ * sync the local alloc to main bitmap.
+ *
+ * assumes you've already locked the main bitmap -- the bitmap inode
+ * passed is used for caching.
+ */
+static int ocfs2_sync_local_to_main(struct ocfs2_super *osb,
+				    handle_t *handle,
+				    struct ocfs2_dinode *alloc,
+				    struct inode *main_bm_inode,
+				    struct buffer_head *main_bm_bh)
+{
+	int status = 0;
+	int bit_off, left, count, start;
+	u64 la_start_blk;
+	u64 blkno;
+	void *bitmap;
+	struct ocfs2_local_alloc *la = OCFS2_LOCAL_ALLOC(alloc);
+
+	trace_ocfs2_sync_local_to_main(
+	     le32_to_cpu(alloc->id1.bitmap1.i_total),
+	     le32_to_cpu(alloc->id1.bitmap1.i_used));
+
+	if (!alloc->id1.bitmap1.i_total) {
+		goto bail;
+	}
+
+	if (le32_to_cpu(alloc->id1.bitmap1.i_used) ==
+	    le32_to_cpu(alloc->id1.bitmap1.i_total)) {
+		goto bail;
+	}
+
+	la_start_blk = ocfs2_clusters_to_blocks(osb->sb,
+						le32_to_cpu(la->la_bm_off));
+	bitmap = la->la_bitmap;
+	start = count = bit_off = 0;
+	left = le32_to_cpu(alloc->id1.bitmap1.i_total);
+
+	while ((bit_off = ocfs2_find_next_zero_bit(bitmap, left, start))
+	       != -1) {
+		if ((bit_off < left) && (bit_off == start)) {
+			count++;
+			start++;
+			continue;
+		}
+		if (count) {
+			blkno = la_start_blk +
+				ocfs2_clusters_to_blocks(osb->sb,
+							 start - count);
+
+			trace_ocfs2_sync_local_to_main_free(
+			     count, start - count,
+			     (unsigned long long)la_start_blk,
+			     (unsigned long long)blkno);
+
+			status = ocfs2_release_clusters(handle,
+							main_bm_inode,
+							main_bm_bh, blkno,
+							count);
+			if (status < 0) {
+				mlog_errno(status);
+				goto bail;
+			}
+		}
+		if (bit_off >= left)
+			break;
+		count = 1;
+		start = bit_off + 1;
+	}
+
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+enum ocfs2_la_event {
+	OCFS2_LA_EVENT_SLIDE,		/* Normal window slide. */
+	OCFS2_LA_EVENT_FRAGMENTED,	/* The global bitmap has
+					 * enough bits theoretically
+					 * free, but a contiguous
+					 * allocation could not be
+					 * found. */
+	OCFS2_LA_EVENT_ENOSPC,		/* Global bitmap doesn't have
+					 * enough bits free to satisfy
+					 * our request. */
+};
+#define OCFS2_LA_ENABLE_INTERVAL (30 * HZ)
+/*
+ * Given an event, calculate the size of our next local alloc window.
+ *
+ * This should always be called under i_mutex of the local alloc inode
+ * so that local alloc disabling doesn't race with processes trying to
+ * use the allocator.
+ *
+ * Returns the state which the local alloc was left in. This value can
+ * be ignored by some paths.
+ */
+static int ocfs2_recalc_la_window(struct ocfs2_super *osb,
+				  enum ocfs2_la_event event)
+{
+	unsigned int bits;
+	int state;
+
+	spin_lock(&osb->osb_lock);
+	if (osb->local_alloc_state == OCFS2_LA_DISABLED) {
+		WARN_ON_ONCE(osb->local_alloc_state == OCFS2_LA_DISABLED);
+		goto out_unlock;
+	}
+
+	/*
+	 * ENOSPC and fragmentation are treated similarly for now.
+	 */
+	if (event == OCFS2_LA_EVENT_ENOSPC ||
+	    event == OCFS2_LA_EVENT_FRAGMENTED) {
+		/*
+		 * We ran out of contiguous space in the primary
+		 * bitmap. Drastically reduce the number of bits used
+		 * by local alloc until we have to disable it.
+		 */
+		bits = osb->local_alloc_bits >> 1;
+		if (bits > ocfs2_megabytes_to_clusters(osb->sb, 1)) {
+			/*
+			 * By setting state to THROTTLED, we'll keep
+			 * the number of local alloc bits used down
+			 * until an event occurs which would give us
+			 * reason to assume the bitmap situation might
+			 * have changed.
+			 */
+			osb->local_alloc_state = OCFS2_LA_THROTTLED;
+			osb->local_alloc_bits = bits;
+		} else {
+			osb->local_alloc_state = OCFS2_LA_DISABLED;
+		}
+		queue_delayed_work(ocfs2_wq, &osb->la_enable_wq,
+				   OCFS2_LA_ENABLE_INTERVAL);
+		goto out_unlock;
+	}
+
+	/*
+	 * Don't increase the size of the local alloc window until we
+	 * know we might be able to fulfill the request. Otherwise, we
+	 * risk bouncing around the global bitmap during periods of
+	 * low space.
+	 */
+	if (osb->local_alloc_state != OCFS2_LA_THROTTLED)
+		osb->local_alloc_bits = osb->local_alloc_default_bits;
+
+out_unlock:
+	state = osb->local_alloc_state;
+	spin_unlock(&osb->osb_lock);
+
+	return state;
+}
+
+static int ocfs2_local_alloc_reserve_for_window(struct ocfs2_super *osb,
+						struct ocfs2_alloc_context **ac,
+						struct inode **bitmap_inode,
+						struct buffer_head **bitmap_bh)
+{
+	int status;
+
+	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
+	if (!(*ac)) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+retry_enospc:
+	(*ac)->ac_bits_wanted = osb->local_alloc_bits;
+	status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
+	if (status == -ENOSPC) {
+		if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_ENOSPC) ==
+		    OCFS2_LA_DISABLED)
+			goto bail;
+
+		ocfs2_free_ac_resource(*ac);
+		memset(*ac, 0, sizeof(struct ocfs2_alloc_context));
+		goto retry_enospc;
+	}
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	*bitmap_inode = (*ac)->ac_inode;
+	igrab(*bitmap_inode);
+	*bitmap_bh = (*ac)->ac_bh;
+	get_bh(*bitmap_bh);
+	status = 0;
+bail:
+	if ((status < 0) && *ac) {
+		ocfs2_free_alloc_context(*ac);
+		*ac = NULL;
+	}
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/*
+ * pass it the bitmap lock in lock_bh if you have it.
+ */
+static int ocfs2_local_alloc_new_window(struct ocfs2_super *osb,
+					handle_t *handle,
+					struct ocfs2_alloc_context *ac)
+{
+	int status = 0;
+	u32 cluster_off, cluster_count;
+	struct ocfs2_dinode *alloc = NULL;
+	struct ocfs2_local_alloc *la;
+
+	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
+	la = OCFS2_LOCAL_ALLOC(alloc);
+
+	trace_ocfs2_local_alloc_new_window(
+		le32_to_cpu(alloc->id1.bitmap1.i_total),
+		osb->local_alloc_bits);
+
+	/* Instruct the allocation code to try the most recently used
+	 * cluster group. We'll re-record the group used this pass
+	 * below. */
+	ac->ac_last_group = osb->la_last_gd;
+
+	/* we used the generic suballoc reserve function, but we set
+	 * everything up nicely, so there's no reason why we can't use
+	 * the more specific cluster api to claim bits. */
+	status = ocfs2_claim_clusters(handle, ac, osb->local_alloc_bits,
+				      &cluster_off, &cluster_count);
+	if (status == -ENOSPC) {
+retry_enospc:
+		/*
+		 * Note: We could also try syncing the journal here to
+		 * allow use of any free bits which the current
+		 * transaction can't give us access to. --Mark
+		 */
+		if (ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_FRAGMENTED) ==
+		    OCFS2_LA_DISABLED)
+			goto bail;
+
+		ac->ac_bits_wanted = osb->local_alloc_bits;
+		status = ocfs2_claim_clusters(handle, ac,
+					      osb->local_alloc_bits,
+					      &cluster_off,
+					      &cluster_count);
+		if (status == -ENOSPC)
+			goto retry_enospc;
+		/*
+		 * We only shrunk the *minimum* number of in our
+		 * request - it's entirely possible that the allocator
+		 * might give us more than we asked for.
+		 */
+		if (status == 0) {
+			spin_lock(&osb->osb_lock);
+			osb->local_alloc_bits = cluster_count;
+			spin_unlock(&osb->osb_lock);
+		}
+	}
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	osb->la_last_gd = ac->ac_last_group;
+
+	la->la_bm_off = cpu_to_le32(cluster_off);
+	alloc->id1.bitmap1.i_total = cpu_to_le32(cluster_count);
+	/* just in case... In the future when we find space ourselves,
+	 * we don't have to get all contiguous -- but we'll have to
+	 * set all previously used bits in bitmap and update
+	 * la_bits_set before setting the bits in the main bitmap. */
+	alloc->id1.bitmap1.i_used = 0;
+	memset(OCFS2_LOCAL_ALLOC(alloc)->la_bitmap, 0,
+	       le16_to_cpu(la->la_size));
+
+	ocfs2_resmap_restart(&osb->osb_la_resmap, cluster_count,
+			     OCFS2_LOCAL_ALLOC(alloc)->la_bitmap);
+
+	trace_ocfs2_local_alloc_new_window_result(
+		OCFS2_LOCAL_ALLOC(alloc)->la_bm_off,
+		le32_to_cpu(alloc->id1.bitmap1.i_total));
+
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/* Note that we do *NOT* lock the local alloc inode here as
+ * it's been locked already for us. */
+static int ocfs2_local_alloc_slide_window(struct ocfs2_super *osb,
+					  struct inode *local_alloc_inode)
+{
+	int status = 0;
+	struct buffer_head *main_bm_bh = NULL;
+	struct inode *main_bm_inode = NULL;
+	handle_t *handle = NULL;
+	struct ocfs2_dinode *alloc;
+	struct ocfs2_dinode *alloc_copy = NULL;
+	struct ocfs2_alloc_context *ac = NULL;
+
+	ocfs2_recalc_la_window(osb, OCFS2_LA_EVENT_SLIDE);
+
+	/* This will lock the main bitmap for us. */
+	status = ocfs2_local_alloc_reserve_for_window(osb,
+						      &ac,
+						      &main_bm_inode,
+						      &main_bm_bh);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_WINDOW_MOVE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	alloc = (struct ocfs2_dinode *) osb->local_alloc_bh->b_data;
+
+	/* We want to clear the local alloc before doing anything
+	 * else, so that if we error later during this operation,
+	 * local alloc shutdown won't try to double free main bitmap
+	 * bits. Make a copy so the sync function knows which bits to
+	 * free. */
+	alloc_copy = kmalloc(osb->local_alloc_bh->b_size, GFP_NOFS);
+	if (!alloc_copy) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+	memcpy(alloc_copy, alloc, osb->local_alloc_bh->b_size);
+
+	status = ocfs2_journal_access_di(handle,
+					 INODE_CACHE(local_alloc_inode),
+					 osb->local_alloc_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	ocfs2_clear_local_alloc(alloc);
+	ocfs2_journal_dirty(handle, osb->local_alloc_bh);
+
+	status = ocfs2_sync_local_to_main(osb, handle, alloc_copy,
+					  main_bm_inode, main_bm_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_local_alloc_new_window(osb, handle, ac);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	atomic_inc(&osb->alloc_stats.moves);
+
+bail:
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+
+	brelse(main_bm_bh);
+
+	if (main_bm_inode)
+		iput(main_bm_inode);
+
+	kfree(alloc_copy);
+
+	if (ac)
+		ocfs2_free_alloc_context(ac);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
diff --git a/fs/ocfs2/localalloc.h b/fs/ocfs2/localalloc.h
new file mode 100644
index 000000000..44a7d1fb2
--- /dev/null
+++ b/fs/ocfs2/localalloc.h
@@ -0,0 +1,68 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * localalloc.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_LOCALALLOC_H
+#define OCFS2_LOCALALLOC_H
+
+int ocfs2_load_local_alloc(struct ocfs2_super *osb);
+
+void ocfs2_shutdown_local_alloc(struct ocfs2_super *osb);
+
+void ocfs2_la_set_sizes(struct ocfs2_super *osb, int requested_mb);
+unsigned int ocfs2_la_default_mb(struct ocfs2_super *osb);
+
+int ocfs2_begin_local_alloc_recovery(struct ocfs2_super *osb,
+				     int node_num,
+				     struct ocfs2_dinode **alloc_copy);
+
+int ocfs2_complete_local_alloc_recovery(struct ocfs2_super *osb,
+					struct ocfs2_dinode *alloc);
+
+int ocfs2_alloc_should_use_local(struct ocfs2_super *osb,
+				 u64 bits);
+
+struct ocfs2_alloc_context;
+int ocfs2_reserve_local_alloc_bits(struct ocfs2_super *osb,
+				   u32 bits_wanted,
+				   struct ocfs2_alloc_context *ac);
+
+int ocfs2_claim_local_alloc_bits(struct ocfs2_super *osb,
+				 handle_t *handle,
+				 struct ocfs2_alloc_context *ac,
+				 u32 bits_wanted,
+				 u32 *bit_off,
+				 u32 *num_bits);
+
+int ocfs2_free_local_alloc_bits(struct ocfs2_super *osb,
+				handle_t *handle,
+				struct ocfs2_alloc_context *ac,
+				u32 bit_off,
+				u32 num_bits);
+
+void ocfs2_local_alloc_seen_free_bits(struct ocfs2_super *osb,
+				      unsigned int num_clusters);
+void ocfs2_la_enable_worker(struct work_struct *work);
+
+#endif /* OCFS2_LOCALALLOC_H */
diff --git a/fs/ocfs2/locks.c b/fs/ocfs2/locks.c
new file mode 100644
index 000000000..6b6d092b0
--- /dev/null
+++ b/fs/ocfs2/locks.c
@@ -0,0 +1,141 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * locks.c
+ *
+ * Userspace file locking support
+ *
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/fcntl.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "dlmglue.h"
+#include "file.h"
+#include "inode.h"
+#include "locks.h"
+
+static int ocfs2_do_flock(struct file *file, struct inode *inode,
+			  int cmd, struct file_lock *fl)
+{
+	int ret = 0, level = 0, trylock = 0;
+	struct ocfs2_file_private *fp = file->private_data;
+	struct ocfs2_lock_res *lockres = &fp->fp_flock;
+
+	if (fl->fl_type == F_WRLCK)
+		level = 1;
+	if (!IS_SETLKW(cmd))
+		trylock = 1;
+
+	mutex_lock(&fp->fp_mutex);
+
+	if (lockres->l_flags & OCFS2_LOCK_ATTACHED &&
+	    lockres->l_level > LKM_NLMODE) {
+		int old_level = 0;
+
+		if (lockres->l_level == LKM_EXMODE)
+			old_level = 1;
+
+		if (level == old_level)
+			goto out;
+
+		/*
+		 * Converting an existing lock is not guaranteed to be
+		 * atomic, so we can get away with simply unlocking
+		 * here and allowing the lock code to try at the new
+		 * level.
+		 */
+
+		flock_lock_file_wait(file,
+				     &(struct file_lock){.fl_type = F_UNLCK});
+
+		ocfs2_file_unlock(file);
+	}
+
+	ret = ocfs2_file_lock(file, level, trylock);
+	if (ret) {
+		if (ret == -EAGAIN && trylock)
+			ret = -EWOULDBLOCK;
+		else
+			mlog_errno(ret);
+		goto out;
+	}
+
+	ret = flock_lock_file_wait(file, fl);
+	if (ret)
+		ocfs2_file_unlock(file);
+
+out:
+	mutex_unlock(&fp->fp_mutex);
+
+	return ret;
+}
+
+static int ocfs2_do_funlock(struct file *file, int cmd, struct file_lock *fl)
+{
+	int ret;
+	struct ocfs2_file_private *fp = file->private_data;
+
+	mutex_lock(&fp->fp_mutex);
+	ocfs2_file_unlock(file);
+	ret = flock_lock_file_wait(file, fl);
+	mutex_unlock(&fp->fp_mutex);
+
+	return ret;
+}
+
+/*
+ * Overall flow of ocfs2_flock() was influenced by gfs2_flock().
+ */
+int ocfs2_flock(struct file *file, int cmd, struct file_lock *fl)
+{
+	struct inode *inode = file->f_mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (!(fl->fl_flags & FL_FLOCK))
+		return -ENOLCK;
+	if (__mandatory_lock(inode))
+		return -ENOLCK;
+
+	if ((osb->s_mount_opt & OCFS2_MOUNT_LOCALFLOCKS) ||
+	    ocfs2_mount_local(osb))
+		return flock_lock_file_wait(file, fl);
+
+	if (fl->fl_type == F_UNLCK)
+		return ocfs2_do_funlock(file, cmd, fl);
+	else
+		return ocfs2_do_flock(file, inode, cmd, fl);
+}
+
+int ocfs2_lock(struct file *file, int cmd, struct file_lock *fl)
+{
+	struct inode *inode = file->f_mapping->host;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (!(fl->fl_flags & FL_POSIX))
+		return -ENOLCK;
+	if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
+		return -ENOLCK;
+
+	return ocfs2_plock(osb->cconn, OCFS2_I(inode)->ip_blkno, file, cmd, fl);
+}
diff --git a/fs/ocfs2/locks.h b/fs/ocfs2/locks.h
new file mode 100644
index 000000000..496d488b2
--- /dev/null
+++ b/fs/ocfs2/locks.h
@@ -0,0 +1,32 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * locks.h
+ *
+ * Function prototypes for Userspace file locking support
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_LOCKS_H
+#define OCFS2_LOCKS_H
+
+int ocfs2_flock(struct file *file, int cmd, struct file_lock *fl);
+int ocfs2_lock(struct file *file, int cmd, struct file_lock *fl);
+
+#endif /* OCFS2_LOCKS_H */
diff --git a/fs/ocfs2/mmap.c b/fs/ocfs2/mmap.c
new file mode 100644
index 000000000..9581d190f
--- /dev/null
+++ b/fs/ocfs2/mmap.c
@@ -0,0 +1,193 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * mmap.c
+ *
+ * Code to deal with the mess that is clustered mmap.
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+#include <linux/pagemap.h>
+#include <linux/uio.h>
+#include <linux/signal.h>
+#include <linux/rbtree.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "aops.h"
+#include "dlmglue.h"
+#include "file.h"
+#include "inode.h"
+#include "mmap.h"
+#include "super.h"
+#include "ocfs2_trace.h"
+
+
+static int ocfs2_fault(struct vm_area_struct *area, struct vm_fault *vmf)
+{
+	sigset_t oldset;
+	int ret;
+
+	ocfs2_block_signals(&oldset);
+	ret = filemap_fault(area, vmf);
+	ocfs2_unblock_signals(&oldset);
+
+	trace_ocfs2_fault(OCFS2_I(area->vm_file->f_mapping->host)->ip_blkno,
+			  area, vmf->page, vmf->pgoff);
+	return ret;
+}
+
+static int __ocfs2_page_mkwrite(struct file *file, struct buffer_head *di_bh,
+				struct page *page)
+{
+	int ret = VM_FAULT_NOPAGE;
+	struct inode *inode = file_inode(file);
+	struct address_space *mapping = inode->i_mapping;
+	loff_t pos = page_offset(page);
+	unsigned int len = PAGE_CACHE_SIZE;
+	pgoff_t last_index;
+	struct page *locked_page = NULL;
+	void *fsdata;
+	loff_t size = i_size_read(inode);
+
+	last_index = (size - 1) >> PAGE_CACHE_SHIFT;
+
+	/*
+	 * There are cases that lead to the page no longer bebongs to the
+	 * mapping.
+	 * 1) pagecache truncates locally due to memory pressure.
+	 * 2) pagecache truncates when another is taking EX lock against 
+	 * inode lock. see ocfs2_data_convert_worker.
+	 * 
+	 * The i_size check doesn't catch the case where nodes truncated and
+	 * then re-extended the file. We'll re-check the page mapping after
+	 * taking the page lock inside of ocfs2_write_begin_nolock().
+	 *
+	 * Let VM retry with these cases.
+	 */
+	if ((page->mapping != inode->i_mapping) ||
+	    (!PageUptodate(page)) ||
+	    (page_offset(page) >= size))
+		goto out;
+
+	/*
+	 * Call ocfs2_write_begin() and ocfs2_write_end() to take
+	 * advantage of the allocation code there. We pass a write
+	 * length of the whole page (chopped to i_size) to make sure
+	 * the whole thing is allocated.
+	 *
+	 * Since we know the page is up to date, we don't have to
+	 * worry about ocfs2_write_begin() skipping some buffer reads
+	 * because the "write" would invalidate their data.
+	 */
+	if (page->index == last_index)
+		len = ((size - 1) & ~PAGE_CACHE_MASK) + 1;
+
+	ret = ocfs2_write_begin_nolock(file, mapping, pos, len, 0, &locked_page,
+				       &fsdata, di_bh, page);
+	if (ret) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		if (ret == -ENOMEM)
+			ret = VM_FAULT_OOM;
+		else
+			ret = VM_FAULT_SIGBUS;
+		goto out;
+	}
+
+	if (!locked_page) {
+		ret = VM_FAULT_NOPAGE;
+		goto out;
+	}
+	ret = ocfs2_write_end_nolock(mapping, pos, len, len, locked_page,
+				     fsdata);
+	BUG_ON(ret != len);
+	ret = VM_FAULT_LOCKED;
+out:
+	return ret;
+}
+
+static int ocfs2_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
+{
+	struct page *page = vmf->page;
+	struct inode *inode = file_inode(vma->vm_file);
+	struct buffer_head *di_bh = NULL;
+	sigset_t oldset;
+	int ret;
+
+	sb_start_pagefault(inode->i_sb);
+	ocfs2_block_signals(&oldset);
+
+	/*
+	 * The cluster locks taken will block a truncate from another
+	 * node. Taking the data lock will also ensure that we don't
+	 * attempt page truncation as part of a downconvert.
+	 */
+	ret = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * The alloc sem should be enough to serialize with
+	 * ocfs2_truncate_file() changing i_size as well as any thread
+	 * modifying the inode btree.
+	 */
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	ret = __ocfs2_page_mkwrite(vma->vm_file, di_bh, page);
+
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	brelse(di_bh);
+	ocfs2_inode_unlock(inode, 1);
+
+out:
+	ocfs2_unblock_signals(&oldset);
+	sb_end_pagefault(inode->i_sb);
+	return ret;
+}
+
+static const struct vm_operations_struct ocfs2_file_vm_ops = {
+	.fault		= ocfs2_fault,
+	.page_mkwrite	= ocfs2_page_mkwrite,
+};
+
+int ocfs2_mmap(struct file *file, struct vm_area_struct *vma)
+{
+	int ret = 0, lock_level = 0;
+
+	ret = ocfs2_inode_lock_atime(file_inode(file),
+				    file->f_path.mnt, &lock_level);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+	ocfs2_inode_unlock(file_inode(file), lock_level);
+out:
+	vma->vm_ops = &ocfs2_file_vm_ops;
+	return 0;
+}
+
diff --git a/fs/ocfs2/mmap.h b/fs/ocfs2/mmap.h
new file mode 100644
index 000000000..1274ee0f1
--- /dev/null
+++ b/fs/ocfs2/mmap.h
@@ -0,0 +1,6 @@
+#ifndef OCFS2_MMAP_H
+#define OCFS2_MMAP_H
+
+int ocfs2_mmap(struct file *file, struct vm_area_struct *vma);
+
+#endif  /* OCFS2_MMAP_H */
diff --git a/fs/ocfs2/move_extents.c b/fs/ocfs2/move_extents.c
new file mode 100644
index 000000000..56a768d06
--- /dev/null
+++ b/fs/ocfs2/move_extents.c
@@ -0,0 +1,1075 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * move_extents.c
+ *
+ * Copyright (C) 2011 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/mount.h>
+#include <linux/swap.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+#include "ocfs2_ioctl.h"
+
+#include "alloc.h"
+#include "aops.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "inode.h"
+#include "journal.h"
+#include "suballoc.h"
+#include "uptodate.h"
+#include "super.h"
+#include "dir.h"
+#include "buffer_head_io.h"
+#include "sysfile.h"
+#include "refcounttree.h"
+#include "move_extents.h"
+
+struct ocfs2_move_extents_context {
+	struct inode *inode;
+	struct file *file;
+	int auto_defrag;
+	int partial;
+	int credits;
+	u32 new_phys_cpos;
+	u32 clusters_moved;
+	u64 refcount_loc;
+	struct ocfs2_move_extents *range;
+	struct ocfs2_extent_tree et;
+	struct ocfs2_alloc_context *meta_ac;
+	struct ocfs2_alloc_context *data_ac;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+};
+
+static int __ocfs2_move_extent(handle_t *handle,
+			       struct ocfs2_move_extents_context *context,
+			       u32 cpos, u32 len, u32 p_cpos, u32 new_p_cpos,
+			       int ext_flags)
+{
+	int ret = 0, index;
+	struct inode *inode = context->inode;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_extent_rec *rec, replace_rec;
+	struct ocfs2_path *path = NULL;
+	struct ocfs2_extent_list *el;
+	u64 ino = ocfs2_metadata_cache_owner(context->et.et_ci);
+	u64 old_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cpos);
+
+	ret = ocfs2_duplicate_clusters_by_page(handle, inode, cpos,
+					       p_cpos, new_p_cpos, len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	memset(&replace_rec, 0, sizeof(replace_rec));
+	replace_rec.e_cpos = cpu_to_le32(cpos);
+	replace_rec.e_leaf_clusters = cpu_to_le16(len);
+	replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(inode->i_sb,
+								   new_p_cpos));
+
+	path = ocfs2_new_path_from_et(&context->et);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(INODE_CACHE(inode), path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	el = path_leaf_el(path);
+
+	index = ocfs2_search_extent_list(el, cpos);
+	if (index == -1) {
+		ocfs2_error(inode->i_sb,
+			    "Inode %llu has an extent at cpos %u which can no "
+			    "longer be found.\n",
+			    (unsigned long long)ino, cpos);
+		ret = -EROFS;
+		goto out;
+	}
+
+	rec = &el->l_recs[index];
+
+	BUG_ON(ext_flags != rec->e_flags);
+	/*
+	 * after moving/defraging to new location, the extent is not going
+	 * to be refcounted anymore.
+	 */
+	replace_rec.e_flags = ext_flags & ~OCFS2_EXT_REFCOUNTED;
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
+				      context->et.et_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_split_extent(handle, &context->et, path, index,
+				 &replace_rec, context->meta_ac,
+				 &context->dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_journal_dirty(handle, context->et.et_root_bh);
+
+	context->new_phys_cpos = new_p_cpos;
+
+	/*
+	 * need I to append truncate log for old clusters?
+	 */
+	if (old_blkno) {
+		if (ext_flags & OCFS2_EXT_REFCOUNTED)
+			ret = ocfs2_decrease_refcount(inode, handle,
+					ocfs2_blocks_to_clusters(osb->sb,
+								 old_blkno),
+					len, context->meta_ac,
+					&context->dealloc, 1);
+		else
+			ret = ocfs2_truncate_log_append(osb, handle,
+							old_blkno, len);
+	}
+
+	ocfs2_update_inode_fsync_trans(handle, inode, 0);
+out:
+	ocfs2_free_path(path);
+	return ret;
+}
+
+/*
+ * lock allocators, and reserving appropriate number of bits for
+ * meta blocks and data clusters.
+ *
+ * in some cases, we don't need to reserve clusters, just let data_ac
+ * be NULL.
+ */
+static int ocfs2_lock_allocators_move_extents(struct inode *inode,
+					struct ocfs2_extent_tree *et,
+					u32 clusters_to_move,
+					u32 extents_to_split,
+					struct ocfs2_alloc_context **meta_ac,
+					struct ocfs2_alloc_context **data_ac,
+					int extra_blocks,
+					int *credits)
+{
+	int ret, num_free_extents;
+	unsigned int max_recs_needed = 2 * extents_to_split + clusters_to_move;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	num_free_extents = ocfs2_num_free_extents(osb, et);
+	if (num_free_extents < 0) {
+		ret = num_free_extents;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (!num_free_extents ||
+	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
+		extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
+
+	ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (data_ac) {
+		ret = ocfs2_reserve_clusters(osb, clusters_to_move, data_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	*credits += ocfs2_calc_extend_credits(osb->sb, et->et_root_el);
+
+	mlog(0, "reserve metadata_blocks: %d, data_clusters: %u, credits: %d\n",
+	     extra_blocks, clusters_to_move, *credits);
+out:
+	if (ret) {
+		if (*meta_ac) {
+			ocfs2_free_alloc_context(*meta_ac);
+			*meta_ac = NULL;
+		}
+	}
+
+	return ret;
+}
+
+/*
+ * Using one journal handle to guarantee the data consistency in case
+ * crash happens anywhere.
+ *
+ *  XXX: defrag can end up with finishing partial extent as requested,
+ * due to not enough contiguous clusters can be found in allocator.
+ */
+static int ocfs2_defrag_extent(struct ocfs2_move_extents_context *context,
+			       u32 cpos, u32 phys_cpos, u32 *len, int ext_flags)
+{
+	int ret, credits = 0, extra_blocks = 0, partial = context->partial;
+	handle_t *handle;
+	struct inode *inode = context->inode;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct inode *tl_inode = osb->osb_tl_inode;
+	struct ocfs2_refcount_tree *ref_tree = NULL;
+	u32 new_phys_cpos, new_len;
+	u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
+
+	if ((ext_flags & OCFS2_EXT_REFCOUNTED) && *len) {
+
+		BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
+			 OCFS2_HAS_REFCOUNT_FL));
+
+		BUG_ON(!context->refcount_loc);
+
+		ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1,
+					       &ref_tree, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			return ret;
+		}
+
+		ret = ocfs2_prepare_refcount_change_for_del(inode,
+							context->refcount_loc,
+							phys_blkno,
+							*len,
+							&credits,
+							&extra_blocks);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, *len, 1,
+						 &context->meta_ac,
+						 &context->data_ac,
+						 extra_blocks, &credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * should be using allocation reservation strategy there?
+	 *
+	 * if (context->data_ac)
+	 *	context->data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
+	 */
+
+	mutex_lock(&tl_inode->i_mutex);
+
+	if (ocfs2_truncate_log_needs_flush(osb)) {
+		ret = __ocfs2_flush_truncate_log(osb);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out_unlock_mutex;
+		}
+	}
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_unlock_mutex;
+	}
+
+	ret = __ocfs2_claim_clusters(handle, context->data_ac, 1, *len,
+				     &new_phys_cpos, &new_len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	/*
+	 * allowing partial extent moving is kind of 'pros and cons', it makes
+	 * whole defragmentation less likely to fail, on the contrary, the bad
+	 * thing is it may make the fs even more fragmented after moving, let
+	 * userspace make a good decision here.
+	 */
+	if (new_len != *len) {
+		mlog(0, "len_claimed: %u, len: %u\n", new_len, *len);
+		if (!partial) {
+			context->range->me_flags &= ~OCFS2_MOVE_EXT_FL_COMPLETE;
+			ret = -ENOSPC;
+			goto out_commit;
+		}
+	}
+
+	mlog(0, "cpos: %u, phys_cpos: %u, new_phys_cpos: %u\n", cpos,
+	     phys_cpos, new_phys_cpos);
+
+	ret = __ocfs2_move_extent(handle, context, cpos, new_len, phys_cpos,
+				  new_phys_cpos, ext_flags);
+	if (ret)
+		mlog_errno(ret);
+
+	if (partial && (new_len != *len))
+		*len = new_len;
+
+	/*
+	 * Here we should write the new page out first if we are
+	 * in write-back mode.
+	 */
+	ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, *len);
+	if (ret)
+		mlog_errno(ret);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out_unlock_mutex:
+	mutex_unlock(&tl_inode->i_mutex);
+
+	if (context->data_ac) {
+		ocfs2_free_alloc_context(context->data_ac);
+		context->data_ac = NULL;
+	}
+
+	if (context->meta_ac) {
+		ocfs2_free_alloc_context(context->meta_ac);
+		context->meta_ac = NULL;
+	}
+
+out:
+	if (ref_tree)
+		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+
+	return ret;
+}
+
+/*
+ * find the victim alloc group, where #blkno fits.
+ */
+static int ocfs2_find_victim_alloc_group(struct inode *inode,
+					 u64 vict_blkno,
+					 int type, int slot,
+					 int *vict_bit,
+					 struct buffer_head **ret_bh)
+{
+	int ret, i, bits_per_unit = 0;
+	u64 blkno;
+	char namebuf[40];
+
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *ac_bh = NULL, *gd_bh = NULL;
+	struct ocfs2_chain_list *cl;
+	struct ocfs2_chain_rec *rec;
+	struct ocfs2_dinode *ac_dinode;
+	struct ocfs2_group_desc *bg;
+
+	ocfs2_sprintf_system_inode_name(namebuf, sizeof(namebuf), type, slot);
+	ret = ocfs2_lookup_ino_from_name(osb->sys_root_inode, namebuf,
+					 strlen(namebuf), &blkno);
+	if (ret) {
+		ret = -ENOENT;
+		goto out;
+	}
+
+	ret = ocfs2_read_blocks_sync(osb, blkno, 1, &ac_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ac_dinode = (struct ocfs2_dinode *)ac_bh->b_data;
+	cl = &(ac_dinode->id2.i_chain);
+	rec = &(cl->cl_recs[0]);
+
+	if (type == GLOBAL_BITMAP_SYSTEM_INODE)
+		bits_per_unit = osb->s_clustersize_bits -
+					inode->i_sb->s_blocksize_bits;
+	/*
+	 * 'vict_blkno' was out of the valid range.
+	 */
+	if ((vict_blkno < le64_to_cpu(rec->c_blkno)) ||
+	    (vict_blkno >= ((u64)le32_to_cpu(ac_dinode->id1.bitmap1.i_total) <<
+				bits_per_unit))) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i++) {
+
+		rec = &(cl->cl_recs[i]);
+		if (!rec)
+			continue;
+
+		bg = NULL;
+
+		do {
+			if (!bg)
+				blkno = le64_to_cpu(rec->c_blkno);
+			else
+				blkno = le64_to_cpu(bg->bg_next_group);
+
+			if (gd_bh) {
+				brelse(gd_bh);
+				gd_bh = NULL;
+			}
+
+			ret = ocfs2_read_blocks_sync(osb, blkno, 1, &gd_bh);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			bg = (struct ocfs2_group_desc *)gd_bh->b_data;
+
+			if (vict_blkno < (le64_to_cpu(bg->bg_blkno) +
+						le16_to_cpu(bg->bg_bits))) {
+
+				*ret_bh = gd_bh;
+				*vict_bit = (vict_blkno - blkno) >>
+							bits_per_unit;
+				mlog(0, "find the victim group: #%llu, "
+				     "total_bits: %u, vict_bit: %u\n",
+				     blkno, le16_to_cpu(bg->bg_bits),
+				     *vict_bit);
+				goto out;
+			}
+
+		} while (le64_to_cpu(bg->bg_next_group));
+	}
+
+	ret = -EINVAL;
+out:
+	brelse(ac_bh);
+
+	/*
+	 * caller has to release the gd_bh properly.
+	 */
+	return ret;
+}
+
+/*
+ * XXX: helper to validate and adjust moving goal.
+ */
+static int ocfs2_validate_and_adjust_move_goal(struct inode *inode,
+					       struct ocfs2_move_extents *range)
+{
+	int ret, goal_bit = 0;
+
+	struct buffer_head *gd_bh = NULL;
+	struct ocfs2_group_desc *bg;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	int c_to_b = 1 << (osb->s_clustersize_bits -
+					inode->i_sb->s_blocksize_bits);
+
+	/*
+	 * make goal become cluster aligned.
+	 */
+	range->me_goal = ocfs2_block_to_cluster_start(inode->i_sb,
+						      range->me_goal);
+	/*
+	 * validate goal sits within global_bitmap, and return the victim
+	 * group desc
+	 */
+	ret = ocfs2_find_victim_alloc_group(inode, range->me_goal,
+					    GLOBAL_BITMAP_SYSTEM_INODE,
+					    OCFS2_INVALID_SLOT,
+					    &goal_bit, &gd_bh);
+	if (ret)
+		goto out;
+
+	bg = (struct ocfs2_group_desc *)gd_bh->b_data;
+
+	/*
+	 * moving goal is not allowd to start with a group desc blok(#0 blk)
+	 * let's compromise to the latter cluster.
+	 */
+	if (range->me_goal == le64_to_cpu(bg->bg_blkno))
+		range->me_goal += c_to_b;
+
+	/*
+	 * movement is not gonna cross two groups.
+	 */
+	if ((le16_to_cpu(bg->bg_bits) - goal_bit) * osb->s_clustersize <
+								range->me_len) {
+		ret = -EINVAL;
+		goto out;
+	}
+	/*
+	 * more exact validations/adjustments will be performed later during
+	 * moving operation for each extent range.
+	 */
+	mlog(0, "extents get ready to be moved to #%llu block\n",
+	     range->me_goal);
+
+out:
+	brelse(gd_bh);
+
+	return ret;
+}
+
+static void ocfs2_probe_alloc_group(struct inode *inode, struct buffer_head *bh,
+				    int *goal_bit, u32 move_len, u32 max_hop,
+				    u32 *phys_cpos)
+{
+	int i, used, last_free_bits = 0, base_bit = *goal_bit;
+	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
+	u32 base_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
+						 le64_to_cpu(gd->bg_blkno));
+
+	for (i = base_bit; i < le16_to_cpu(gd->bg_bits); i++) {
+
+		used = ocfs2_test_bit(i, (unsigned long *)gd->bg_bitmap);
+		if (used) {
+			/*
+			 * we even tried searching the free chunk by jumping
+			 * a 'max_hop' distance, but still failed.
+			 */
+			if ((i - base_bit) > max_hop) {
+				*phys_cpos = 0;
+				break;
+			}
+
+			if (last_free_bits)
+				last_free_bits = 0;
+
+			continue;
+		} else
+			last_free_bits++;
+
+		if (last_free_bits == move_len) {
+			*goal_bit = i;
+			*phys_cpos = base_cpos + i;
+			break;
+		}
+	}
+
+	mlog(0, "found phys_cpos: %u to fit the wanted moving.\n", *phys_cpos);
+}
+
+static int ocfs2_move_extent(struct ocfs2_move_extents_context *context,
+			     u32 cpos, u32 phys_cpos, u32 *new_phys_cpos,
+			     u32 len, int ext_flags)
+{
+	int ret, credits = 0, extra_blocks = 0, goal_bit = 0;
+	handle_t *handle;
+	struct inode *inode = context->inode;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct inode *tl_inode = osb->osb_tl_inode;
+	struct inode *gb_inode = NULL;
+	struct buffer_head *gb_bh = NULL;
+	struct buffer_head *gd_bh = NULL;
+	struct ocfs2_group_desc *gd;
+	struct ocfs2_refcount_tree *ref_tree = NULL;
+	u32 move_max_hop = ocfs2_blocks_to_clusters(inode->i_sb,
+						    context->range->me_threshold);
+	u64 phys_blkno, new_phys_blkno;
+
+	phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
+
+	if ((ext_flags & OCFS2_EXT_REFCOUNTED) && len) {
+
+		BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
+			 OCFS2_HAS_REFCOUNT_FL));
+
+		BUG_ON(!context->refcount_loc);
+
+		ret = ocfs2_lock_refcount_tree(osb, context->refcount_loc, 1,
+					       &ref_tree, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			return ret;
+		}
+
+		ret = ocfs2_prepare_refcount_change_for_del(inode,
+							context->refcount_loc,
+							phys_blkno,
+							len,
+							&credits,
+							&extra_blocks);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_lock_allocators_move_extents(inode, &context->et, len, 1,
+						 &context->meta_ac,
+						 NULL, extra_blocks, &credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * need to count 2 extra credits for global_bitmap inode and
+	 * group descriptor.
+	 */
+	credits += OCFS2_INODE_UPDATE_CREDITS + 1;
+
+	/*
+	 * ocfs2_move_extent() didn't reserve any clusters in lock_allocators()
+	 * logic, while we still need to lock the global_bitmap.
+	 */
+	gb_inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
+					       OCFS2_INVALID_SLOT);
+	if (!gb_inode) {
+		mlog(ML_ERROR, "unable to get global_bitmap inode\n");
+		ret = -EIO;
+		goto out;
+	}
+
+	mutex_lock(&gb_inode->i_mutex);
+
+	ret = ocfs2_inode_lock(gb_inode, &gb_bh, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock_gb_mutex;
+	}
+
+	mutex_lock(&tl_inode->i_mutex);
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_unlock_tl_inode;
+	}
+
+	new_phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, *new_phys_cpos);
+	ret = ocfs2_find_victim_alloc_group(inode, new_phys_blkno,
+					    GLOBAL_BITMAP_SYSTEM_INODE,
+					    OCFS2_INVALID_SLOT,
+					    &goal_bit, &gd_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	/*
+	 * probe the victim cluster group to find a proper
+	 * region to fit wanted movement, it even will perfrom
+	 * a best-effort attempt by compromising to a threshold
+	 * around the goal.
+	 */
+	ocfs2_probe_alloc_group(inode, gd_bh, &goal_bit, len, move_max_hop,
+				new_phys_cpos);
+	if (!*new_phys_cpos) {
+		ret = -ENOSPC;
+		goto out_commit;
+	}
+
+	ret = __ocfs2_move_extent(handle, context, cpos, len, phys_cpos,
+				  *new_phys_cpos, ext_flags);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	gd = (struct ocfs2_group_desc *)gd_bh->b_data;
+	ret = ocfs2_alloc_dinode_update_counts(gb_inode, handle, gb_bh, len,
+					       le16_to_cpu(gd->bg_chain));
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_block_group_set_bits(handle, gb_inode, gd, gd_bh,
+					 goal_bit, len);
+	if (ret) {
+		ocfs2_rollback_alloc_dinode_counts(gb_inode, gb_bh, len,
+					       le16_to_cpu(gd->bg_chain));
+		mlog_errno(ret);
+	}
+
+	/*
+	 * Here we should write the new page out first if we are
+	 * in write-back mode.
+	 */
+	ret = ocfs2_cow_sync_writeback(inode->i_sb, context->inode, cpos, len);
+	if (ret)
+		mlog_errno(ret);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+	brelse(gd_bh);
+
+out_unlock_tl_inode:
+	mutex_unlock(&tl_inode->i_mutex);
+
+	ocfs2_inode_unlock(gb_inode, 1);
+out_unlock_gb_mutex:
+	mutex_unlock(&gb_inode->i_mutex);
+	brelse(gb_bh);
+	iput(gb_inode);
+
+out:
+	if (context->meta_ac) {
+		ocfs2_free_alloc_context(context->meta_ac);
+		context->meta_ac = NULL;
+	}
+
+	if (ref_tree)
+		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+
+	return ret;
+}
+
+/*
+ * Helper to calculate the defraging length in one run according to threshold.
+ */
+static void ocfs2_calc_extent_defrag_len(u32 *alloc_size, u32 *len_defraged,
+					 u32 threshold, int *skip)
+{
+	if ((*alloc_size + *len_defraged) < threshold) {
+		/*
+		 * proceed defragmentation until we meet the thresh
+		 */
+		*len_defraged += *alloc_size;
+	} else if (*len_defraged == 0) {
+		/*
+		 * XXX: skip a large extent.
+		 */
+		*skip = 1;
+	} else {
+		/*
+		 * split this extent to coalesce with former pieces as
+		 * to reach the threshold.
+		 *
+		 * we're done here with one cycle of defragmentation
+		 * in a size of 'thresh', resetting 'len_defraged'
+		 * forces a new defragmentation.
+		 */
+		*alloc_size = threshold - *len_defraged;
+		*len_defraged = 0;
+	}
+}
+
+static int __ocfs2_move_extents_range(struct buffer_head *di_bh,
+				struct ocfs2_move_extents_context *context)
+{
+	int ret = 0, flags, do_defrag, skip = 0;
+	u32 cpos, phys_cpos, move_start, len_to_move, alloc_size;
+	u32 len_defraged = 0, defrag_thresh = 0, new_phys_cpos = 0;
+
+	struct inode *inode = context->inode;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_move_extents *range = context->range;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if ((i_size_read(inode) == 0) || (range->me_len == 0))
+		return 0;
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		return 0;
+
+	context->refcount_loc = le64_to_cpu(di->i_refcount_loc);
+
+	ocfs2_init_dinode_extent_tree(&context->et, INODE_CACHE(inode), di_bh);
+	ocfs2_init_dealloc_ctxt(&context->dealloc);
+
+	/*
+	 * TO-DO XXX:
+	 *
+	 * - xattr extents.
+	 */
+
+	do_defrag = context->auto_defrag;
+
+	/*
+	 * extents moving happens in unit of clusters, for the sake
+	 * of simplicity, we may ignore two clusters where 'byte_start'
+	 * and 'byte_start + len' were within.
+	 */
+	move_start = ocfs2_clusters_for_bytes(osb->sb, range->me_start);
+	len_to_move = (range->me_start + range->me_len) >>
+						osb->s_clustersize_bits;
+	if (len_to_move >= move_start)
+		len_to_move -= move_start;
+	else
+		len_to_move = 0;
+
+	if (do_defrag) {
+		defrag_thresh = range->me_threshold >> osb->s_clustersize_bits;
+		if (defrag_thresh <= 1)
+			goto done;
+	} else
+		new_phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb,
+							 range->me_goal);
+
+	mlog(0, "Inode: %llu, start: %llu, len: %llu, cstart: %u, clen: %u, "
+	     "thresh: %u\n",
+	     (unsigned long long)OCFS2_I(inode)->ip_blkno,
+	     (unsigned long long)range->me_start,
+	     (unsigned long long)range->me_len,
+	     move_start, len_to_move, defrag_thresh);
+
+	cpos = move_start;
+	while (len_to_move) {
+		ret = ocfs2_get_clusters(inode, cpos, &phys_cpos, &alloc_size,
+					 &flags);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (alloc_size > len_to_move)
+			alloc_size = len_to_move;
+
+		/*
+		 * XXX: how to deal with a hole:
+		 *
+		 * - skip the hole of course
+		 * - force a new defragmentation
+		 */
+		if (!phys_cpos) {
+			if (do_defrag)
+				len_defraged = 0;
+
+			goto next;
+		}
+
+		if (do_defrag) {
+			ocfs2_calc_extent_defrag_len(&alloc_size, &len_defraged,
+						     defrag_thresh, &skip);
+			/*
+			 * skip large extents
+			 */
+			if (skip) {
+				skip = 0;
+				goto next;
+			}
+
+			mlog(0, "#Defrag: cpos: %u, phys_cpos: %u, "
+			     "alloc_size: %u, len_defraged: %u\n",
+			     cpos, phys_cpos, alloc_size, len_defraged);
+
+			ret = ocfs2_defrag_extent(context, cpos, phys_cpos,
+						  &alloc_size, flags);
+		} else {
+			ret = ocfs2_move_extent(context, cpos, phys_cpos,
+						&new_phys_cpos, alloc_size,
+						flags);
+
+			new_phys_cpos += alloc_size;
+		}
+
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		context->clusters_moved += alloc_size;
+next:
+		cpos += alloc_size;
+		len_to_move -= alloc_size;
+	}
+
+done:
+	range->me_flags |= OCFS2_MOVE_EXT_FL_COMPLETE;
+
+out:
+	range->me_moved_len = ocfs2_clusters_to_bytes(osb->sb,
+						      context->clusters_moved);
+	range->me_new_offset = ocfs2_clusters_to_bytes(osb->sb,
+						       context->new_phys_cpos);
+
+	ocfs2_schedule_truncate_log_flush(osb, 1);
+	ocfs2_run_deallocs(osb, &context->dealloc);
+
+	return ret;
+}
+
+static int ocfs2_move_extents(struct ocfs2_move_extents_context *context)
+{
+	int status;
+	handle_t *handle;
+	struct inode *inode = context->inode;
+	struct ocfs2_dinode *di;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
+		return -EROFS;
+
+	mutex_lock(&inode->i_mutex);
+
+	/*
+	 * This prevents concurrent writes from other nodes
+	 */
+	status = ocfs2_rw_lock(inode, 1);
+	if (status) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	status = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (status) {
+		mlog_errno(status);
+		goto out_rw_unlock;
+	}
+
+	/*
+	 * rememer ip_xattr_sem also needs to be held if necessary
+	 */
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+
+	status = __ocfs2_move_extents_range(di_bh, context);
+
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+	if (status) {
+		mlog_errno(status);
+		goto out_inode_unlock;
+	}
+
+	/*
+	 * We update ctime for these changes
+	 */
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out_inode_unlock;
+	}
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	inode->i_ctime = CURRENT_TIME;
+	di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
+	di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+	ocfs2_update_inode_fsync_trans(handle, inode, 0);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out_inode_unlock:
+	brelse(di_bh);
+	ocfs2_inode_unlock(inode, 1);
+out_rw_unlock:
+	ocfs2_rw_unlock(inode, 1);
+out:
+	mutex_unlock(&inode->i_mutex);
+
+	return status;
+}
+
+int ocfs2_ioctl_move_extents(struct file *filp, void __user *argp)
+{
+	int status;
+
+	struct inode *inode = file_inode(filp);
+	struct ocfs2_move_extents range;
+	struct ocfs2_move_extents_context *context;
+
+	if (!argp)
+		return -EINVAL;
+
+	status = mnt_want_write_file(filp);
+	if (status)
+		return status;
+
+	if ((!S_ISREG(inode->i_mode)) || !(filp->f_mode & FMODE_WRITE)) {
+		status = -EPERM;
+		goto out_drop;
+	}
+
+	if (inode->i_flags & (S_IMMUTABLE|S_APPEND)) {
+		status = -EPERM;
+		goto out_drop;
+	}
+
+	context = kzalloc(sizeof(struct ocfs2_move_extents_context), GFP_NOFS);
+	if (!context) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto out_drop;
+	}
+
+	context->inode = inode;
+	context->file = filp;
+
+	if (copy_from_user(&range, argp, sizeof(range))) {
+		status = -EFAULT;
+		goto out_free;
+	}
+
+	if (range.me_start > i_size_read(inode)) {
+		status = -EINVAL;
+		goto out_free;
+	}
+
+	if (range.me_start + range.me_len > i_size_read(inode))
+			range.me_len = i_size_read(inode) - range.me_start;
+
+	context->range = &range;
+
+	if (range.me_flags & OCFS2_MOVE_EXT_FL_AUTO_DEFRAG) {
+		context->auto_defrag = 1;
+		/*
+		 * ok, the default theshold for the defragmentation
+		 * is 1M, since our maximum clustersize was 1M also.
+		 * any thought?
+		 */
+		if (!range.me_threshold)
+			range.me_threshold = 1024 * 1024;
+
+		if (range.me_threshold > i_size_read(inode))
+			range.me_threshold = i_size_read(inode);
+
+		if (range.me_flags & OCFS2_MOVE_EXT_FL_PART_DEFRAG)
+			context->partial = 1;
+	} else {
+		/*
+		 * first best-effort attempt to validate and adjust the goal
+		 * (physical address in block), while it can't guarantee later
+		 * operation can succeed all the time since global_bitmap may
+		 * change a bit over time.
+		 */
+
+		status = ocfs2_validate_and_adjust_move_goal(inode, &range);
+		if (status)
+			goto out_copy;
+	}
+
+	status = ocfs2_move_extents(context);
+	if (status)
+		mlog_errno(status);
+out_copy:
+	/*
+	 * movement/defragmentation may end up being partially completed,
+	 * that's the reason why we need to return userspace the finished
+	 * length and new_offset even if failure happens somewhere.
+	 */
+	if (copy_to_user(argp, &range, sizeof(range)))
+		status = -EFAULT;
+
+out_free:
+	kfree(context);
+out_drop:
+	mnt_drop_write_file(filp);
+
+	return status;
+}
diff --git a/fs/ocfs2/move_extents.h b/fs/ocfs2/move_extents.h
new file mode 100644
index 000000000..4e143e811
--- /dev/null
+++ b/fs/ocfs2/move_extents.h
@@ -0,0 +1,22 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * move_extents.h
+ *
+ * Copyright (C) 2011 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+#ifndef OCFS2_MOVE_EXTENTS_H
+#define OCFS2_MOVE_EXTENTS_H
+
+int ocfs2_ioctl_move_extents(struct file *filp,  void __user *argp);
+
+#endif /* OCFS2_MOVE_EXTENTS_H */
diff --git a/fs/ocfs2/namei.c b/fs/ocfs2/namei.c
new file mode 100644
index 000000000..176fe6afd
--- /dev/null
+++ b/fs/ocfs2/namei.c
@@ -0,0 +1,2921 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * namei.c
+ *
+ * Create and rename file, directory, symlinks
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ *  Portions of this code from linux/fs/ext3/dir.c
+ *
+ *  Copyright (C) 1992, 1993, 1994, 1995
+ *  Remy Card (card@masi.ibp.fr)
+ *  Laboratoire MASI - Institut Blaise pascal
+ *  Universite Pierre et Marie Curie (Paris VI)
+ *
+ *   from
+ *
+ *   linux/fs/minix/dir.c
+ *
+ *   Copyright (C) 1991, 1992 Linux Torvalds
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/quotaops.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "dcache.h"
+#include "dir.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "file.h"
+#include "inode.h"
+#include "journal.h"
+#include "namei.h"
+#include "suballoc.h"
+#include "super.h"
+#include "symlink.h"
+#include "sysfile.h"
+#include "uptodate.h"
+#include "xattr.h"
+#include "acl.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+static int ocfs2_mknod_locked(struct ocfs2_super *osb,
+			      struct inode *dir,
+			      struct inode *inode,
+			      dev_t dev,
+			      struct buffer_head **new_fe_bh,
+			      struct buffer_head *parent_fe_bh,
+			      handle_t *handle,
+			      struct ocfs2_alloc_context *inode_ac);
+
+static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
+				    struct inode **ret_orphan_dir,
+				    u64 blkno,
+				    char *name,
+				    struct ocfs2_dir_lookup_result *lookup,
+				    bool dio);
+
+static int ocfs2_orphan_add(struct ocfs2_super *osb,
+			    handle_t *handle,
+			    struct inode *inode,
+			    struct buffer_head *fe_bh,
+			    char *name,
+			    struct ocfs2_dir_lookup_result *lookup,
+			    struct inode *orphan_dir_inode,
+			    bool dio);
+
+static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
+				     handle_t *handle,
+				     struct inode *inode,
+				     const char *symname);
+
+static int ocfs2_double_lock(struct ocfs2_super *osb,
+			     struct buffer_head **bh1,
+			     struct inode *inode1,
+			     struct buffer_head **bh2,
+			     struct inode *inode2,
+			     int rename);
+
+static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
+/* An orphan dir name is an 8 byte value, printed as a hex string */
+#define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
+#define OCFS2_DIO_ORPHAN_PREFIX "dio-"
+#define OCFS2_DIO_ORPHAN_PREFIX_LEN 4
+
+static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
+				   unsigned int flags)
+{
+	int status;
+	u64 blkno;
+	struct inode *inode = NULL;
+	struct dentry *ret;
+	struct ocfs2_inode_info *oi;
+
+	trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
+			   dentry->d_name.name,
+			   (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
+
+	if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
+		ret = ERR_PTR(-ENAMETOOLONG);
+		goto bail;
+	}
+
+	status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		ret = ERR_PTR(status);
+		goto bail;
+	}
+
+	status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
+					    dentry->d_name.len, &blkno);
+	if (status < 0)
+		goto bail_add;
+
+	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
+	if (IS_ERR(inode)) {
+		ret = ERR_PTR(-EACCES);
+		goto bail_unlock;
+	}
+
+	oi = OCFS2_I(inode);
+	/* Clear any orphaned state... If we were able to look up the
+	 * inode from a directory, it certainly can't be orphaned. We
+	 * might have the bad state from a node which intended to
+	 * orphan this inode but crashed before it could commit the
+	 * unlink. */
+	spin_lock(&oi->ip_lock);
+	oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
+	spin_unlock(&oi->ip_lock);
+
+bail_add:
+	ret = d_splice_alias(inode, dentry);
+
+	if (inode) {
+		/*
+		 * If d_splice_alias() finds a DCACHE_DISCONNECTED
+		 * dentry, it will d_move() it on top of ourse. The
+		 * return value will indicate this however, so in
+		 * those cases, we switch them around for the locking
+		 * code.
+		 *
+		 * NOTE: This dentry already has ->d_op set from
+		 * ocfs2_get_parent() and ocfs2_get_dentry()
+		 */
+		if (!IS_ERR_OR_NULL(ret))
+			dentry = ret;
+
+		status = ocfs2_dentry_attach_lock(dentry, inode,
+						  OCFS2_I(dir)->ip_blkno);
+		if (status) {
+			mlog_errno(status);
+			ret = ERR_PTR(status);
+			goto bail_unlock;
+		}
+	} else
+		ocfs2_dentry_attach_gen(dentry);
+
+bail_unlock:
+	/* Don't drop the cluster lock until *after* the d_add --
+	 * unlink on another node will message us to remove that
+	 * dentry under this lock so otherwise we can race this with
+	 * the downconvert thread and have a stale dentry. */
+	ocfs2_inode_unlock(dir, 0);
+
+bail:
+
+	trace_ocfs2_lookup_ret(ret);
+
+	return ret;
+}
+
+static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
+{
+	struct inode *inode;
+
+	inode = new_inode(dir->i_sb);
+	if (!inode) {
+		mlog(ML_ERROR, "new_inode failed!\n");
+		return NULL;
+	}
+
+	/* populate as many fields early on as possible - many of
+	 * these are used by the support functions here and in
+	 * callers. */
+	if (S_ISDIR(mode))
+		set_nlink(inode, 2);
+	inode_init_owner(inode, dir, mode);
+	dquot_initialize(inode);
+	return inode;
+}
+
+static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
+		struct dentry *dentry, struct inode *inode)
+{
+	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
+
+	ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
+	ocfs2_lock_res_free(&dl->dl_lockres);
+	BUG_ON(dl->dl_count != 1);
+	spin_lock(&dentry_attach_lock);
+	dentry->d_fsdata = NULL;
+	spin_unlock(&dentry_attach_lock);
+	kfree(dl);
+	iput(inode);
+}
+
+static int ocfs2_mknod(struct inode *dir,
+		       struct dentry *dentry,
+		       umode_t mode,
+		       dev_t dev)
+{
+	int status = 0;
+	struct buffer_head *parent_fe_bh = NULL;
+	handle_t *handle = NULL;
+	struct ocfs2_super *osb;
+	struct ocfs2_dinode *dirfe;
+	struct buffer_head *new_fe_bh = NULL;
+	struct inode *inode = NULL;
+	struct ocfs2_alloc_context *inode_ac = NULL;
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	int want_clusters = 0;
+	int want_meta = 0;
+	int xattr_credits = 0;
+	struct ocfs2_security_xattr_info si = {
+		.enable = 1,
+	};
+	int did_quota_inode = 0;
+	struct ocfs2_dir_lookup_result lookup = { NULL, };
+	sigset_t oldset;
+	int did_block_signals = 0;
+	struct posix_acl *default_acl = NULL, *acl = NULL;
+	struct ocfs2_dentry_lock *dl = NULL;
+
+	trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
+			  (unsigned long long)OCFS2_I(dir)->ip_blkno,
+			  (unsigned long)dev, mode);
+
+	dquot_initialize(dir);
+
+	/* get our super block */
+	osb = OCFS2_SB(dir->i_sb);
+
+	status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		return status;
+	}
+
+	if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
+		status = -EMLINK;
+		goto leave;
+	}
+
+	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
+	if (!ocfs2_read_links_count(dirfe)) {
+		/* can't make a file in a deleted directory. */
+		status = -ENOENT;
+		goto leave;
+	}
+
+	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
+					   dentry->d_name.len);
+	if (status)
+		goto leave;
+
+	/* get a spot inside the dir. */
+	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
+					      dentry->d_name.name,
+					      dentry->d_name.len, &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* reserve an inode spot */
+	status = ocfs2_reserve_new_inode(osb, &inode_ac);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto leave;
+	}
+
+	inode = ocfs2_get_init_inode(dir, mode);
+	if (!inode) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* get security xattr */
+	status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
+	if (status) {
+		if (status == -EOPNOTSUPP)
+			si.enable = 0;
+		else {
+			mlog_errno(status);
+			goto leave;
+		}
+	}
+
+	/* calculate meta data/clusters for setting security and acl xattr */
+	status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
+				       &si, &want_clusters,
+				       &xattr_credits, &want_meta);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* Reserve a cluster if creating an extent based directory. */
+	if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
+		want_clusters += 1;
+
+		/* Dir indexing requires extra space as well */
+		if (ocfs2_supports_indexed_dirs(osb))
+			want_meta++;
+	}
+
+	status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto leave;
+	}
+
+	status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto leave;
+	}
+
+	status = posix_acl_create(dir, &mode, &default_acl, &acl);
+	if (status) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
+							    S_ISDIR(mode),
+							    xattr_credits));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* Starting to change things, restart is no longer possible. */
+	ocfs2_block_signals(&oldset);
+	did_block_signals = 1;
+
+	status = dquot_alloc_inode(inode);
+	if (status)
+		goto leave;
+	did_quota_inode = 1;
+
+	/* do the real work now. */
+	status = ocfs2_mknod_locked(osb, dir, inode, dev,
+				    &new_fe_bh, parent_fe_bh, handle,
+				    inode_ac);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	if (S_ISDIR(mode)) {
+		status = ocfs2_fill_new_dir(osb, handle, dir, inode,
+					    new_fe_bh, data_ac, meta_ac);
+		if (status < 0) {
+			mlog_errno(status);
+			goto leave;
+		}
+
+		status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
+						 parent_fe_bh,
+						 OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto leave;
+		}
+		ocfs2_add_links_count(dirfe, 1);
+		ocfs2_journal_dirty(handle, parent_fe_bh);
+		inc_nlink(dir);
+	}
+
+	if (default_acl) {
+		status = ocfs2_set_acl(handle, inode, new_fe_bh,
+				       ACL_TYPE_DEFAULT, default_acl,
+				       meta_ac, data_ac);
+	}
+	if (!status && acl) {
+		status = ocfs2_set_acl(handle, inode, new_fe_bh,
+				       ACL_TYPE_ACCESS, acl,
+				       meta_ac, data_ac);
+	}
+
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	if (si.enable) {
+		status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
+						 meta_ac, data_ac);
+		if (status < 0) {
+			mlog_errno(status);
+			goto leave;
+		}
+	}
+
+	/*
+	 * Do this before adding the entry to the directory. We add
+	 * also set d_op after success so that ->d_iput() will cleanup
+	 * the dentry lock even if ocfs2_add_entry() fails below.
+	 */
+	status = ocfs2_dentry_attach_lock(dentry, inode,
+					  OCFS2_I(dir)->ip_blkno);
+	if (status) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	dl = dentry->d_fsdata;
+
+	status = ocfs2_add_entry(handle, dentry, inode,
+				 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
+				 &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	insert_inode_hash(inode);
+	d_instantiate(dentry, inode);
+	status = 0;
+leave:
+	if (default_acl)
+		posix_acl_release(default_acl);
+	if (acl)
+		posix_acl_release(acl);
+	if (status < 0 && did_quota_inode)
+		dquot_free_inode(inode);
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+
+	ocfs2_inode_unlock(dir, 1);
+	if (did_block_signals)
+		ocfs2_unblock_signals(&oldset);
+
+	brelse(new_fe_bh);
+	brelse(parent_fe_bh);
+	kfree(si.value);
+
+	ocfs2_free_dir_lookup_result(&lookup);
+
+	if (inode_ac)
+		ocfs2_free_alloc_context(inode_ac);
+
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	/*
+	 * We should call iput after the i_mutex of the bitmap been
+	 * unlocked in ocfs2_free_alloc_context, or the
+	 * ocfs2_delete_inode will mutex_lock again.
+	 */
+	if ((status < 0) && inode) {
+		if (dl)
+			ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
+
+		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
+		clear_nlink(inode);
+		iput(inode);
+	}
+
+	if (status)
+		mlog_errno(status);
+
+	return status;
+}
+
+static int __ocfs2_mknod_locked(struct inode *dir,
+				struct inode *inode,
+				dev_t dev,
+				struct buffer_head **new_fe_bh,
+				struct buffer_head *parent_fe_bh,
+				handle_t *handle,
+				struct ocfs2_alloc_context *inode_ac,
+				u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
+{
+	int status = 0;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct ocfs2_dinode *fe = NULL;
+	struct ocfs2_extent_list *fel;
+	u16 feat;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+
+	*new_fe_bh = NULL;
+
+	/* populate as many fields early on as possible - many of
+	 * these are used by the support functions here and in
+	 * callers. */
+	inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
+	OCFS2_I(inode)->ip_blkno = fe_blkno;
+	spin_lock(&osb->osb_lock);
+	inode->i_generation = osb->s_next_generation++;
+	spin_unlock(&osb->osb_lock);
+
+	*new_fe_bh = sb_getblk(osb->sb, fe_blkno);
+	if (!*new_fe_bh) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto leave;
+	}
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
+					 *new_fe_bh,
+					 OCFS2_JOURNAL_ACCESS_CREATE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
+	memset(fe, 0, osb->sb->s_blocksize);
+
+	fe->i_generation = cpu_to_le32(inode->i_generation);
+	fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
+	fe->i_blkno = cpu_to_le64(fe_blkno);
+	fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
+	fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
+	fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
+	fe->i_uid = cpu_to_le32(i_uid_read(inode));
+	fe->i_gid = cpu_to_le32(i_gid_read(inode));
+	fe->i_mode = cpu_to_le16(inode->i_mode);
+	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
+		fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
+
+	ocfs2_set_links_count(fe, inode->i_nlink);
+
+	fe->i_last_eb_blk = 0;
+	strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
+	fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
+	fe->i_atime = fe->i_ctime = fe->i_mtime =
+		cpu_to_le64(CURRENT_TIME.tv_sec);
+	fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
+		cpu_to_le32(CURRENT_TIME.tv_nsec);
+	fe->i_dtime = 0;
+
+	/*
+	 * If supported, directories start with inline data. If inline
+	 * isn't supported, but indexing is, we start them as indexed.
+	 */
+	feat = le16_to_cpu(fe->i_dyn_features);
+	if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
+		fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
+
+		fe->id2.i_data.id_count = cpu_to_le16(
+				ocfs2_max_inline_data_with_xattr(osb->sb, fe));
+	} else {
+		fel = &fe->id2.i_list;
+		fel->l_tree_depth = 0;
+		fel->l_next_free_rec = 0;
+		fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
+	}
+
+	ocfs2_journal_dirty(handle, *new_fe_bh);
+
+	ocfs2_populate_inode(inode, fe, 1);
+	ocfs2_ci_set_new(osb, INODE_CACHE(inode));
+	if (!ocfs2_mount_local(osb)) {
+		status = ocfs2_create_new_inode_locks(inode);
+		if (status < 0)
+			mlog_errno(status);
+	}
+
+	oi->i_sync_tid = handle->h_transaction->t_tid;
+	oi->i_datasync_tid = handle->h_transaction->t_tid;
+
+leave:
+	if (status < 0) {
+		if (*new_fe_bh) {
+			brelse(*new_fe_bh);
+			*new_fe_bh = NULL;
+		}
+	}
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static int ocfs2_mknod_locked(struct ocfs2_super *osb,
+			      struct inode *dir,
+			      struct inode *inode,
+			      dev_t dev,
+			      struct buffer_head **new_fe_bh,
+			      struct buffer_head *parent_fe_bh,
+			      handle_t *handle,
+			      struct ocfs2_alloc_context *inode_ac)
+{
+	int status = 0;
+	u64 suballoc_loc, fe_blkno = 0;
+	u16 suballoc_bit;
+
+	*new_fe_bh = NULL;
+
+	status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
+				       inode_ac, &suballoc_loc,
+				       &suballoc_bit, &fe_blkno);
+	if (status < 0) {
+		mlog_errno(status);
+		return status;
+	}
+
+	return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
+				    parent_fe_bh, handle, inode_ac,
+				    fe_blkno, suballoc_loc, suballoc_bit);
+}
+
+static int ocfs2_mkdir(struct inode *dir,
+		       struct dentry *dentry,
+		       umode_t mode)
+{
+	int ret;
+
+	trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
+			  OCFS2_I(dir)->ip_blkno, mode);
+	ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
+	if (ret)
+		mlog_errno(ret);
+
+	return ret;
+}
+
+static int ocfs2_create(struct inode *dir,
+			struct dentry *dentry,
+			umode_t mode,
+			bool excl)
+{
+	int ret;
+
+	trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
+			   (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
+	ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
+	if (ret)
+		mlog_errno(ret);
+
+	return ret;
+}
+
+static int ocfs2_link(struct dentry *old_dentry,
+		      struct inode *dir,
+		      struct dentry *dentry)
+{
+	handle_t *handle;
+	struct inode *inode = d_inode(old_dentry);
+	struct inode *old_dir = d_inode(old_dentry->d_parent);
+	int err;
+	struct buffer_head *fe_bh = NULL;
+	struct buffer_head *old_dir_bh = NULL;
+	struct buffer_head *parent_fe_bh = NULL;
+	struct ocfs2_dinode *fe = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct ocfs2_dir_lookup_result lookup = { NULL, };
+	sigset_t oldset;
+	u64 old_de_ino;
+
+	trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
+			 old_dentry->d_name.len, old_dentry->d_name.name,
+			 dentry->d_name.len, dentry->d_name.name);
+
+	if (S_ISDIR(inode->i_mode))
+		return -EPERM;
+
+	dquot_initialize(dir);
+
+	err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
+			&parent_fe_bh, dir, 0);
+	if (err < 0) {
+		if (err != -ENOENT)
+			mlog_errno(err);
+		return err;
+	}
+
+	/* make sure both dirs have bhs
+	 * get an extra ref on old_dir_bh if old==new */
+	if (!parent_fe_bh) {
+		if (old_dir_bh) {
+			parent_fe_bh = old_dir_bh;
+			get_bh(parent_fe_bh);
+		} else {
+			mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
+			err = -EIO;
+			goto out;
+		}
+	}
+
+	if (!dir->i_nlink) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
+			old_dentry->d_name.len, &old_de_ino);
+	if (err) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	/*
+	 * Check whether another node removed the source inode while we
+	 * were in the vfs.
+	 */
+	if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
+		err = -ENOENT;
+		goto out;
+	}
+
+	err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
+					dentry->d_name.len);
+	if (err)
+		goto out;
+
+	err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
+					   dentry->d_name.name,
+					   dentry->d_name.len, &lookup);
+	if (err < 0) {
+		mlog_errno(err);
+		goto out;
+	}
+
+	err = ocfs2_inode_lock(inode, &fe_bh, 1);
+	if (err < 0) {
+		if (err != -ENOENT)
+			mlog_errno(err);
+		goto out;
+	}
+
+	fe = (struct ocfs2_dinode *) fe_bh->b_data;
+	if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
+		err = -EMLINK;
+		goto out_unlock_inode;
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
+	if (IS_ERR(handle)) {
+		err = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(err);
+		goto out_unlock_inode;
+	}
+
+	/* Starting to change things, restart is no longer possible. */
+	ocfs2_block_signals(&oldset);
+
+	err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (err < 0) {
+		mlog_errno(err);
+		goto out_commit;
+	}
+
+	inc_nlink(inode);
+	inode->i_ctime = CURRENT_TIME;
+	ocfs2_set_links_count(fe, inode->i_nlink);
+	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
+	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+	ocfs2_journal_dirty(handle, fe_bh);
+
+	err = ocfs2_add_entry(handle, dentry, inode,
+			      OCFS2_I(inode)->ip_blkno,
+			      parent_fe_bh, &lookup);
+	if (err) {
+		ocfs2_add_links_count(fe, -1);
+		drop_nlink(inode);
+		mlog_errno(err);
+		goto out_commit;
+	}
+
+	err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
+	if (err) {
+		mlog_errno(err);
+		goto out_commit;
+	}
+
+	ihold(inode);
+	d_instantiate(dentry, inode);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+	ocfs2_unblock_signals(&oldset);
+out_unlock_inode:
+	ocfs2_inode_unlock(inode, 1);
+
+out:
+	ocfs2_double_unlock(old_dir, dir);
+
+	brelse(fe_bh);
+	brelse(parent_fe_bh);
+	brelse(old_dir_bh);
+
+	ocfs2_free_dir_lookup_result(&lookup);
+
+	if (err)
+		mlog_errno(err);
+
+	return err;
+}
+
+/*
+ * Takes and drops an exclusive lock on the given dentry. This will
+ * force other nodes to drop it.
+ */
+static int ocfs2_remote_dentry_delete(struct dentry *dentry)
+{
+	int ret;
+
+	ret = ocfs2_dentry_lock(dentry, 1);
+	if (ret)
+		mlog_errno(ret);
+	else
+		ocfs2_dentry_unlock(dentry, 1);
+
+	return ret;
+}
+
+static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
+{
+	if (S_ISDIR(inode->i_mode)) {
+		if (inode->i_nlink == 2)
+			return 1;
+		return 0;
+	}
+
+	if (inode->i_nlink == 1)
+		return 1;
+	return 0;
+}
+
+static int ocfs2_unlink(struct inode *dir,
+			struct dentry *dentry)
+{
+	int status;
+	int child_locked = 0;
+	bool is_unlinkable = false;
+	struct inode *inode = d_inode(dentry);
+	struct inode *orphan_dir = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	u64 blkno;
+	struct ocfs2_dinode *fe = NULL;
+	struct buffer_head *fe_bh = NULL;
+	struct buffer_head *parent_node_bh = NULL;
+	handle_t *handle = NULL;
+	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
+	struct ocfs2_dir_lookup_result lookup = { NULL, };
+	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
+
+	trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
+			   dentry->d_name.name,
+			   (unsigned long long)OCFS2_I(dir)->ip_blkno,
+			   (unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	dquot_initialize(dir);
+
+	BUG_ON(d_inode(dentry->d_parent) != dir);
+
+	if (inode == osb->root_inode)
+		return -EPERM;
+
+	status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
+					 OI_LS_PARENT);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		return status;
+	}
+
+	status = ocfs2_find_files_on_disk(dentry->d_name.name,
+					  dentry->d_name.len, &blkno, dir,
+					  &lookup);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		goto leave;
+	}
+
+	if (OCFS2_I(inode)->ip_blkno != blkno) {
+		status = -ENOENT;
+
+		trace_ocfs2_unlink_noent(
+				(unsigned long long)OCFS2_I(inode)->ip_blkno,
+				(unsigned long long)blkno,
+				OCFS2_I(inode)->ip_flags);
+		goto leave;
+	}
+
+	status = ocfs2_inode_lock(inode, &fe_bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		goto leave;
+	}
+	child_locked = 1;
+
+	if (S_ISDIR(inode->i_mode)) {
+		if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
+			status = -ENOTEMPTY;
+			goto leave;
+		}
+	}
+
+	status = ocfs2_remote_dentry_delete(dentry);
+	if (status < 0) {
+		/* This remote delete should succeed under all normal
+		 * circumstances. */
+		mlog_errno(status);
+		goto leave;
+	}
+
+	if (ocfs2_inode_is_unlinkable(inode)) {
+		status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
+						  OCFS2_I(inode)->ip_blkno,
+						  orphan_name, &orphan_insert,
+						  false);
+		if (status < 0) {
+			mlog_errno(status);
+			goto leave;
+		}
+		is_unlinkable = true;
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	fe = (struct ocfs2_dinode *) fe_bh->b_data;
+
+	/* delete the name from the parent dir */
+	status = ocfs2_delete_entry(handle, dir, &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	if (S_ISDIR(inode->i_mode))
+		drop_nlink(inode);
+	drop_nlink(inode);
+	ocfs2_set_links_count(fe, inode->i_nlink);
+	ocfs2_journal_dirty(handle, fe_bh);
+
+	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
+	if (S_ISDIR(inode->i_mode))
+		drop_nlink(dir);
+
+	status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		if (S_ISDIR(inode->i_mode))
+			inc_nlink(dir);
+		goto leave;
+	}
+
+	if (is_unlinkable) {
+		status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
+				orphan_name, &orphan_insert, orphan_dir, false);
+		if (status < 0)
+			mlog_errno(status);
+	}
+
+leave:
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+
+	if (child_locked)
+		ocfs2_inode_unlock(inode, 1);
+
+	ocfs2_inode_unlock(dir, 1);
+
+	if (orphan_dir) {
+		/* This was locked for us in ocfs2_prepare_orphan_dir() */
+		ocfs2_inode_unlock(orphan_dir, 1);
+		mutex_unlock(&orphan_dir->i_mutex);
+		iput(orphan_dir);
+	}
+
+	brelse(fe_bh);
+	brelse(parent_node_bh);
+
+	ocfs2_free_dir_lookup_result(&orphan_insert);
+	ocfs2_free_dir_lookup_result(&lookup);
+
+	if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
+		mlog_errno(status);
+
+	return status;
+}
+
+static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
+		u64 src_inode_no, u64 dest_inode_no)
+{
+	int ret = 0, i = 0;
+	u64 parent_inode_no = 0;
+	u64 child_inode_no = src_inode_no;
+	struct inode *child_inode;
+
+#define MAX_LOOKUP_TIMES 32
+	while (1) {
+		child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
+		if (IS_ERR(child_inode)) {
+			ret = PTR_ERR(child_inode);
+			break;
+		}
+
+		ret = ocfs2_inode_lock(child_inode, NULL, 0);
+		if (ret < 0) {
+			iput(child_inode);
+			if (ret != -ENOENT)
+				mlog_errno(ret);
+			break;
+		}
+
+		ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
+				&parent_inode_no);
+		ocfs2_inode_unlock(child_inode, 0);
+		iput(child_inode);
+		if (ret < 0) {
+			ret = -ENOENT;
+			break;
+		}
+
+		if (parent_inode_no == dest_inode_no) {
+			ret = 1;
+			break;
+		}
+
+		if (parent_inode_no == osb->root_inode->i_ino) {
+			ret = 0;
+			break;
+		}
+
+		child_inode_no = parent_inode_no;
+
+		if (++i >= MAX_LOOKUP_TIMES) {
+			mlog(ML_NOTICE, "max lookup times reached, filesystem "
+					"may have nested directories, "
+					"src inode: %llu, dest inode: %llu.\n",
+					(unsigned long long)src_inode_no,
+					(unsigned long long)dest_inode_no);
+			ret = 0;
+			break;
+		}
+	}
+
+	return ret;
+}
+
+/*
+ * The only place this should be used is rename and link!
+ * if they have the same id, then the 1st one is the only one locked.
+ */
+static int ocfs2_double_lock(struct ocfs2_super *osb,
+			     struct buffer_head **bh1,
+			     struct inode *inode1,
+			     struct buffer_head **bh2,
+			     struct inode *inode2,
+			     int rename)
+{
+	int status;
+	int inode1_is_ancestor, inode2_is_ancestor;
+	struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
+	struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
+	struct buffer_head **tmpbh;
+	struct inode *tmpinode;
+
+	trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
+				(unsigned long long)oi2->ip_blkno);
+
+	if (*bh1)
+		*bh1 = NULL;
+	if (*bh2)
+		*bh2 = NULL;
+
+	/* we always want to lock the one with the lower lockid first.
+	 * and if they are nested, we lock ancestor first */
+	if (oi1->ip_blkno != oi2->ip_blkno) {
+		inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
+				oi1->ip_blkno);
+		if (inode1_is_ancestor < 0) {
+			status = inode1_is_ancestor;
+			goto bail;
+		}
+
+		inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
+				oi2->ip_blkno);
+		if (inode2_is_ancestor < 0) {
+			status = inode2_is_ancestor;
+			goto bail;
+		}
+
+		if ((inode1_is_ancestor == 1) ||
+				(oi1->ip_blkno < oi2->ip_blkno &&
+				inode2_is_ancestor == 0)) {
+			/* switch id1 and id2 around */
+			tmpbh = bh2;
+			bh2 = bh1;
+			bh1 = tmpbh;
+
+			tmpinode = inode2;
+			inode2 = inode1;
+			inode1 = tmpinode;
+		}
+		/* lock id2 */
+		status = ocfs2_inode_lock_nested(inode2, bh2, 1,
+				rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
+		if (status < 0) {
+			if (status != -ENOENT)
+				mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	/* lock id1 */
+	status = ocfs2_inode_lock_nested(inode1, bh1, 1,
+			rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
+	if (status < 0) {
+		/*
+		 * An error return must mean that no cluster locks
+		 * were held on function exit.
+		 */
+		if (oi1->ip_blkno != oi2->ip_blkno) {
+			ocfs2_inode_unlock(inode2, 1);
+			brelse(*bh2);
+			*bh2 = NULL;
+		}
+
+		if (status != -ENOENT)
+			mlog_errno(status);
+	}
+
+	trace_ocfs2_double_lock_end(
+			(unsigned long long)OCFS2_I(inode1)->ip_blkno,
+			(unsigned long long)OCFS2_I(inode2)->ip_blkno);
+
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
+{
+	ocfs2_inode_unlock(inode1, 1);
+
+	if (inode1 != inode2)
+		ocfs2_inode_unlock(inode2, 1);
+}
+
+static int ocfs2_rename(struct inode *old_dir,
+			struct dentry *old_dentry,
+			struct inode *new_dir,
+			struct dentry *new_dentry)
+{
+	int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
+	int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
+	struct inode *old_inode = d_inode(old_dentry);
+	struct inode *new_inode = d_inode(new_dentry);
+	struct inode *orphan_dir = NULL;
+	struct ocfs2_dinode *newfe = NULL;
+	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
+	struct buffer_head *newfe_bh = NULL;
+	struct buffer_head *old_inode_bh = NULL;
+	struct ocfs2_super *osb = NULL;
+	u64 newfe_blkno, old_de_ino;
+	handle_t *handle = NULL;
+	struct buffer_head *old_dir_bh = NULL;
+	struct buffer_head *new_dir_bh = NULL;
+	u32 old_dir_nlink = old_dir->i_nlink;
+	struct ocfs2_dinode *old_di;
+	struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
+	struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
+	struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
+	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
+	struct ocfs2_dir_lookup_result target_insert = { NULL, };
+	bool should_add_orphan = false;
+
+	/* At some point it might be nice to break this function up a
+	 * bit. */
+
+	trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
+			   old_dentry->d_name.len, old_dentry->d_name.name,
+			   new_dentry->d_name.len, new_dentry->d_name.name);
+
+	dquot_initialize(old_dir);
+	dquot_initialize(new_dir);
+
+	osb = OCFS2_SB(old_dir->i_sb);
+
+	if (new_inode) {
+		if (!igrab(new_inode))
+			BUG();
+	}
+
+	/* Assume a directory hierarchy thusly:
+	 * a/b/c
+	 * a/d
+	 * a,b,c, and d are all directories.
+	 *
+	 * from cwd of 'a' on both nodes:
+	 * node1: mv b/c d
+	 * node2: mv d   b/c
+	 *
+	 * And that's why, just like the VFS, we need a file system
+	 * rename lock. */
+	if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
+		status = ocfs2_rename_lock(osb);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+		rename_lock = 1;
+
+		/* here we cannot guarantee the inodes haven't just been
+		 * changed, so check if they are nested again */
+		status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
+				old_inode->i_ino);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		} else if (status == 1) {
+			status = -EPERM;
+			trace_ocfs2_rename_not_permitted(
+					(unsigned long long)old_inode->i_ino,
+					(unsigned long long)new_dir->i_ino);
+			goto bail;
+		}
+	}
+
+	/* if old and new are the same, this'll just do one lock. */
+	status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
+				   &new_dir_bh, new_dir, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	parents_locked = 1;
+
+	/* make sure both dirs have bhs
+	 * get an extra ref on old_dir_bh if old==new */
+	if (!new_dir_bh) {
+		if (old_dir_bh) {
+			new_dir_bh = old_dir_bh;
+			get_bh(new_dir_bh);
+		} else {
+			mlog(ML_ERROR, "no old_dir_bh!\n");
+			status = -EIO;
+			goto bail;
+		}
+	}
+
+	/*
+	 * Aside from allowing a meta data update, the locking here
+	 * also ensures that the downconvert thread on other nodes
+	 * won't have to concurrently downconvert the inode and the
+	 * dentry locks.
+	 */
+	status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
+					 OI_LS_PARENT);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		goto bail;
+	}
+	old_child_locked = 1;
+
+	status = ocfs2_remote_dentry_delete(old_dentry);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	if (S_ISDIR(old_inode->i_mode)) {
+		u64 old_inode_parent;
+
+		update_dot_dot = 1;
+		status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
+						  old_inode,
+						  &old_inode_dot_dot_res);
+		if (status) {
+			status = -EIO;
+			goto bail;
+		}
+
+		if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
+			status = -EIO;
+			goto bail;
+		}
+
+		if (!new_inode && new_dir != old_dir &&
+		    new_dir->i_nlink >= ocfs2_link_max(osb)) {
+			status = -EMLINK;
+			goto bail;
+		}
+	}
+
+	status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
+					    old_dentry->d_name.len,
+					    &old_de_ino);
+	if (status) {
+		status = -ENOENT;
+		goto bail;
+	}
+
+	/*
+	 *  Check for inode number is _not_ due to possible IO errors.
+	 *  We might rmdir the source, keep it as pwd of some process
+	 *  and merrily kill the link to whatever was created under the
+	 *  same name. Goodbye sticky bit ;-<
+	 */
+	if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
+		status = -ENOENT;
+		goto bail;
+	}
+
+	/* check if the target already exists (in which case we need
+	 * to delete it */
+	status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
+					  new_dentry->d_name.len,
+					  &newfe_blkno, new_dir,
+					  &target_lookup_res);
+	/* The only error we allow here is -ENOENT because the new
+	 * file not existing is perfectly valid. */
+	if ((status < 0) && (status != -ENOENT)) {
+		/* If we cannot find the file specified we should just */
+		/* return the error... */
+		mlog_errno(status);
+		goto bail;
+	}
+	if (status == 0)
+		target_exists = 1;
+
+	if (!target_exists && new_inode) {
+		/*
+		 * Target was unlinked by another node while we were
+		 * waiting to get to ocfs2_rename(). There isn't
+		 * anything we can do here to help the situation, so
+		 * bubble up the appropriate error.
+		 */
+		status = -ENOENT;
+		goto bail;
+	}
+
+	/* In case we need to overwrite an existing file, we blow it
+	 * away first */
+	if (target_exists) {
+		/* VFS didn't think there existed an inode here, but
+		 * someone else in the cluster must have raced our
+		 * rename to create one. Today we error cleanly, in
+		 * the future we should consider calling iget to build
+		 * a new struct inode for this entry. */
+		if (!new_inode) {
+			status = -EACCES;
+
+			trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
+						new_dentry->d_name.name);
+			goto bail;
+		}
+
+		if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
+			status = -EACCES;
+
+			trace_ocfs2_rename_disagree(
+			     (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
+			     (unsigned long long)newfe_blkno,
+			     OCFS2_I(new_inode)->ip_flags);
+			goto bail;
+		}
+
+		status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
+		if (status < 0) {
+			if (status != -ENOENT)
+				mlog_errno(status);
+			goto bail;
+		}
+		new_child_locked = 1;
+
+		status = ocfs2_remote_dentry_delete(new_dentry);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
+
+		trace_ocfs2_rename_over_existing(
+		     (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
+		     (unsigned long long)newfe_bh->b_blocknr : 0ULL);
+
+		if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
+			status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
+						OCFS2_I(new_inode)->ip_blkno,
+						orphan_name, &orphan_insert,
+						false);
+			if (status < 0) {
+				mlog_errno(status);
+				goto bail;
+			}
+			should_add_orphan = true;
+		}
+	} else {
+		BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
+
+		status = ocfs2_check_dir_for_entry(new_dir,
+						   new_dentry->d_name.name,
+						   new_dentry->d_name.len);
+		if (status)
+			goto bail;
+
+		status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
+						      new_dentry->d_name.name,
+						      new_dentry->d_name.len,
+						      &target_insert);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	if (target_exists) {
+		if (S_ISDIR(new_inode->i_mode)) {
+			if (new_inode->i_nlink != 2 ||
+			    !ocfs2_empty_dir(new_inode)) {
+				status = -ENOTEMPTY;
+				goto bail;
+			}
+		}
+		status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
+						 newfe_bh,
+						 OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		/* change the dirent to point to the correct inode */
+		status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
+					    old_inode);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+		new_dir->i_version++;
+
+		if (S_ISDIR(new_inode->i_mode))
+			ocfs2_set_links_count(newfe, 0);
+		else
+			ocfs2_add_links_count(newfe, -1);
+		ocfs2_journal_dirty(handle, newfe_bh);
+		if (should_add_orphan) {
+			status = ocfs2_orphan_add(osb, handle, new_inode,
+					newfe_bh, orphan_name,
+					&orphan_insert, orphan_dir, false);
+			if (status < 0) {
+				mlog_errno(status);
+				goto bail;
+			}
+		}
+	} else {
+		/* if the name was not found in new_dir, add it now */
+		status = ocfs2_add_entry(handle, new_dentry, old_inode,
+					 OCFS2_I(old_inode)->ip_blkno,
+					 new_dir_bh, &target_insert);
+	}
+
+	old_inode->i_ctime = CURRENT_TIME;
+	mark_inode_dirty(old_inode);
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
+					 old_inode_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status >= 0) {
+		old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
+
+		old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
+		old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
+		ocfs2_journal_dirty(handle, old_inode_bh);
+	} else
+		mlog_errno(status);
+
+	/*
+	 * Now that the name has been added to new_dir, remove the old name.
+	 *
+	 * We don't keep any directory entry context around until now
+	 * because the insert might have changed the type of directory
+	 * we're dealing with.
+	 */
+	status = ocfs2_find_entry(old_dentry->d_name.name,
+				  old_dentry->d_name.len, old_dir,
+				  &old_entry_lookup);
+	if (status)
+		goto bail;
+
+	status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	if (new_inode) {
+		drop_nlink(new_inode);
+		new_inode->i_ctime = CURRENT_TIME;
+	}
+	old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
+
+	if (update_dot_dot) {
+		status = ocfs2_update_entry(old_inode, handle,
+					    &old_inode_dot_dot_res, new_dir);
+		drop_nlink(old_dir);
+		if (new_inode) {
+			drop_nlink(new_inode);
+		} else {
+			inc_nlink(new_dir);
+			mark_inode_dirty(new_dir);
+		}
+	}
+	mark_inode_dirty(old_dir);
+	ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
+	if (new_inode) {
+		mark_inode_dirty(new_inode);
+		ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
+	}
+
+	if (old_dir != new_dir) {
+		/* Keep the same times on both directories.*/
+		new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
+
+		/*
+		 * This will also pick up the i_nlink change from the
+		 * block above.
+		 */
+		ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
+	}
+
+	if (old_dir_nlink != old_dir->i_nlink) {
+		if (!old_dir_bh) {
+			mlog(ML_ERROR, "need to change nlink for old dir "
+			     "%llu from %d to %d but bh is NULL!\n",
+			     (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
+			     (int)old_dir_nlink, old_dir->i_nlink);
+		} else {
+			struct ocfs2_dinode *fe;
+			status = ocfs2_journal_access_di(handle,
+							 INODE_CACHE(old_dir),
+							 old_dir_bh,
+							 OCFS2_JOURNAL_ACCESS_WRITE);
+			fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
+			ocfs2_set_links_count(fe, old_dir->i_nlink);
+			ocfs2_journal_dirty(handle, old_dir_bh);
+		}
+	}
+	ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
+	status = 0;
+bail:
+	if (rename_lock)
+		ocfs2_rename_unlock(osb);
+
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+
+	if (parents_locked)
+		ocfs2_double_unlock(old_dir, new_dir);
+
+	if (old_child_locked)
+		ocfs2_inode_unlock(old_inode, 1);
+
+	if (new_child_locked)
+		ocfs2_inode_unlock(new_inode, 1);
+
+	if (orphan_dir) {
+		/* This was locked for us in ocfs2_prepare_orphan_dir() */
+		ocfs2_inode_unlock(orphan_dir, 1);
+		mutex_unlock(&orphan_dir->i_mutex);
+		iput(orphan_dir);
+	}
+
+	if (new_inode)
+		sync_mapping_buffers(old_inode->i_mapping);
+
+	if (new_inode)
+		iput(new_inode);
+
+	ocfs2_free_dir_lookup_result(&target_lookup_res);
+	ocfs2_free_dir_lookup_result(&old_entry_lookup);
+	ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
+	ocfs2_free_dir_lookup_result(&orphan_insert);
+	ocfs2_free_dir_lookup_result(&target_insert);
+
+	brelse(newfe_bh);
+	brelse(old_inode_bh);
+	brelse(old_dir_bh);
+	brelse(new_dir_bh);
+
+	if (status)
+		mlog_errno(status);
+
+	return status;
+}
+
+/*
+ * we expect i_size = strlen(symname). Copy symname into the file
+ * data, including the null terminator.
+ */
+static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
+				     handle_t *handle,
+				     struct inode *inode,
+				     const char *symname)
+{
+	struct buffer_head **bhs = NULL;
+	const char *c;
+	struct super_block *sb = osb->sb;
+	u64 p_blkno, p_blocks;
+	int virtual, blocks, status, i, bytes_left;
+
+	bytes_left = i_size_read(inode) + 1;
+	/* we can't trust i_blocks because we're actually going to
+	 * write i_size + 1 bytes. */
+	blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
+
+	trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
+					i_size_read(inode), blocks);
+
+	/* Sanity check -- make sure we're going to fit. */
+	if (bytes_left >
+	    ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
+		status = -EIO;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
+	if (!bhs) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
+					     NULL);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/* links can never be larger than one cluster so we know this
+	 * is all going to be contiguous, but do a sanity check
+	 * anyway. */
+	if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
+		status = -EIO;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	virtual = 0;
+	while(bytes_left > 0) {
+		c = &symname[virtual * sb->s_blocksize];
+
+		bhs[virtual] = sb_getblk(sb, p_blkno);
+		if (!bhs[virtual]) {
+			status = -ENOMEM;
+			mlog_errno(status);
+			goto bail;
+		}
+		ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
+					      bhs[virtual]);
+
+		status = ocfs2_journal_access(handle, INODE_CACHE(inode),
+					      bhs[virtual],
+					      OCFS2_JOURNAL_ACCESS_CREATE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
+
+		memcpy(bhs[virtual]->b_data, c,
+		       (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
+		       bytes_left);
+
+		ocfs2_journal_dirty(handle, bhs[virtual]);
+
+		virtual++;
+		p_blkno++;
+		bytes_left -= sb->s_blocksize;
+	}
+
+	status = 0;
+bail:
+
+	if (bhs) {
+		for(i = 0; i < blocks; i++)
+			brelse(bhs[i]);
+		kfree(bhs);
+	}
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static int ocfs2_symlink(struct inode *dir,
+			 struct dentry *dentry,
+			 const char *symname)
+{
+	int status, l, credits;
+	u64 newsize;
+	struct ocfs2_super *osb = NULL;
+	struct inode *inode = NULL;
+	struct super_block *sb;
+	struct buffer_head *new_fe_bh = NULL;
+	struct buffer_head *parent_fe_bh = NULL;
+	struct ocfs2_dinode *fe = NULL;
+	struct ocfs2_dinode *dirfe;
+	handle_t *handle = NULL;
+	struct ocfs2_alloc_context *inode_ac = NULL;
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct ocfs2_alloc_context *xattr_ac = NULL;
+	int want_clusters = 0;
+	int xattr_credits = 0;
+	struct ocfs2_security_xattr_info si = {
+		.enable = 1,
+	};
+	int did_quota = 0, did_quota_inode = 0;
+	struct ocfs2_dir_lookup_result lookup = { NULL, };
+	sigset_t oldset;
+	int did_block_signals = 0;
+	struct ocfs2_dentry_lock *dl = NULL;
+
+	trace_ocfs2_symlink_begin(dir, dentry, symname,
+				  dentry->d_name.len, dentry->d_name.name);
+
+	dquot_initialize(dir);
+
+	sb = dir->i_sb;
+	osb = OCFS2_SB(sb);
+
+	l = strlen(symname) + 1;
+
+	credits = ocfs2_calc_symlink_credits(sb);
+
+	/* lock the parent directory */
+	status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		return status;
+	}
+
+	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
+	if (!ocfs2_read_links_count(dirfe)) {
+		/* can't make a file in a deleted directory. */
+		status = -ENOENT;
+		goto bail;
+	}
+
+	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
+					   dentry->d_name.len);
+	if (status)
+		goto bail;
+
+	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
+					      dentry->d_name.name,
+					      dentry->d_name.len, &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_reserve_new_inode(osb, &inode_ac);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
+	if (!inode) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/* get security xattr */
+	status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
+	if (status) {
+		if (status == -EOPNOTSUPP)
+			si.enable = 0;
+		else {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	/* calculate meta data/clusters for setting security xattr */
+	if (si.enable) {
+		status = ocfs2_calc_security_init(dir, &si, &want_clusters,
+						  &xattr_credits, &xattr_ac);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	/* don't reserve bitmap space for fast symlinks. */
+	if (l > ocfs2_fast_symlink_chars(sb))
+		want_clusters += 1;
+
+	status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	handle = ocfs2_start_trans(osb, credits + xattr_credits);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/* Starting to change things, restart is no longer possible. */
+	ocfs2_block_signals(&oldset);
+	did_block_signals = 1;
+
+	status = dquot_alloc_inode(inode);
+	if (status)
+		goto bail;
+	did_quota_inode = 1;
+
+	trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
+				   dentry->d_name.name,
+				   (unsigned long long)OCFS2_I(dir)->ip_blkno,
+				   inode->i_mode);
+
+	status = ocfs2_mknod_locked(osb, dir, inode,
+				    0, &new_fe_bh, parent_fe_bh, handle,
+				    inode_ac);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
+	inode->i_rdev = 0;
+	newsize = l - 1;
+	inode->i_op = &ocfs2_symlink_inode_operations;
+	if (l > ocfs2_fast_symlink_chars(sb)) {
+		u32 offset = 0;
+
+		status = dquot_alloc_space_nodirty(inode,
+		    ocfs2_clusters_to_bytes(osb->sb, 1));
+		if (status)
+			goto bail;
+		did_quota = 1;
+		inode->i_mapping->a_ops = &ocfs2_aops;
+		status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
+					      new_fe_bh,
+					      handle, data_ac, NULL,
+					      NULL);
+		if (status < 0) {
+			if (status != -ENOSPC && status != -EINTR) {
+				mlog(ML_ERROR,
+				     "Failed to extend file to %llu\n",
+				     (unsigned long long)newsize);
+				mlog_errno(status);
+				status = -ENOSPC;
+			}
+			goto bail;
+		}
+		i_size_write(inode, newsize);
+		inode->i_blocks = ocfs2_inode_sector_count(inode);
+	} else {
+		inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
+		memcpy((char *) fe->id2.i_symlink, symname, l);
+		i_size_write(inode, newsize);
+		inode->i_blocks = 0;
+	}
+
+	status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	if (!ocfs2_inode_is_fast_symlink(inode)) {
+		status = ocfs2_create_symlink_data(osb, handle, inode,
+						   symname);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	if (si.enable) {
+		status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
+						 xattr_ac, data_ac);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	/*
+	 * Do this before adding the entry to the directory. We add
+	 * also set d_op after success so that ->d_iput() will cleanup
+	 * the dentry lock even if ocfs2_add_entry() fails below.
+	 */
+	status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	dl = dentry->d_fsdata;
+
+	status = ocfs2_add_entry(handle, dentry, inode,
+				 le64_to_cpu(fe->i_blkno), parent_fe_bh,
+				 &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	insert_inode_hash(inode);
+	d_instantiate(dentry, inode);
+bail:
+	if (status < 0 && did_quota)
+		dquot_free_space_nodirty(inode,
+					ocfs2_clusters_to_bytes(osb->sb, 1));
+	if (status < 0 && did_quota_inode)
+		dquot_free_inode(inode);
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+
+	ocfs2_inode_unlock(dir, 1);
+	if (did_block_signals)
+		ocfs2_unblock_signals(&oldset);
+
+	brelse(new_fe_bh);
+	brelse(parent_fe_bh);
+	kfree(si.value);
+	ocfs2_free_dir_lookup_result(&lookup);
+	if (inode_ac)
+		ocfs2_free_alloc_context(inode_ac);
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+	if (xattr_ac)
+		ocfs2_free_alloc_context(xattr_ac);
+	if ((status < 0) && inode) {
+		if (dl)
+			ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
+
+		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
+		clear_nlink(inode);
+		iput(inode);
+	}
+
+	if (status)
+		mlog_errno(status);
+
+	return status;
+}
+
+static int ocfs2_blkno_stringify(u64 blkno, char *name)
+{
+	int status, namelen;
+
+	namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
+			   (long long)blkno);
+	if (namelen <= 0) {
+		if (namelen)
+			status = namelen;
+		else
+			status = -EINVAL;
+		mlog_errno(status);
+		goto bail;
+	}
+	if (namelen != OCFS2_ORPHAN_NAMELEN) {
+		status = -EINVAL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	trace_ocfs2_blkno_stringify(blkno, name, namelen);
+
+	status = 0;
+bail:
+	if (status < 0)
+		mlog_errno(status);
+	return status;
+}
+
+static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
+					struct inode **ret_orphan_dir,
+					struct buffer_head **ret_orphan_dir_bh)
+{
+	struct inode *orphan_dir_inode;
+	struct buffer_head *orphan_dir_bh = NULL;
+	int ret = 0;
+
+	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
+						       ORPHAN_DIR_SYSTEM_INODE,
+						       osb->slot_num);
+	if (!orphan_dir_inode) {
+		ret = -ENOENT;
+		mlog_errno(ret);
+		return ret;
+	}
+
+	mutex_lock(&orphan_dir_inode->i_mutex);
+
+	ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
+	if (ret < 0) {
+		mutex_unlock(&orphan_dir_inode->i_mutex);
+		iput(orphan_dir_inode);
+
+		mlog_errno(ret);
+		return ret;
+	}
+
+	*ret_orphan_dir = orphan_dir_inode;
+	*ret_orphan_dir_bh = orphan_dir_bh;
+
+	return 0;
+}
+
+static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
+				      struct buffer_head *orphan_dir_bh,
+				      u64 blkno,
+				      char *name,
+				      struct ocfs2_dir_lookup_result *lookup,
+				      bool dio)
+{
+	int ret;
+	struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
+	int namelen = dio ?
+			(OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
+			OCFS2_ORPHAN_NAMELEN;
+
+	if (dio) {
+		ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
+				OCFS2_DIO_ORPHAN_PREFIX);
+		if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
+			ret = -EINVAL;
+			mlog_errno(ret);
+			return ret;
+		}
+
+		ret = ocfs2_blkno_stringify(blkno,
+				name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
+	} else
+		ret = ocfs2_blkno_stringify(blkno, name);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
+					   orphan_dir_bh, name,
+					   namelen, lookup);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	return 0;
+}
+
+/**
+ * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
+ * insertion of an orphan.
+ * @osb: ocfs2 file system
+ * @ret_orphan_dir: Orphan dir inode - returned locked!
+ * @blkno: Actual block number of the inode to be inserted into orphan dir.
+ * @lookup: dir lookup result, to be passed back into functions like
+ *          ocfs2_orphan_add
+ *
+ * Returns zero on success and the ret_orphan_dir, name and lookup
+ * fields will be populated.
+ *
+ * Returns non-zero on failure. 
+ */
+static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
+				    struct inode **ret_orphan_dir,
+				    u64 blkno,
+				    char *name,
+				    struct ocfs2_dir_lookup_result *lookup,
+				    bool dio)
+{
+	struct inode *orphan_dir_inode = NULL;
+	struct buffer_head *orphan_dir_bh = NULL;
+	int ret = 0;
+
+	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
+					   &orphan_dir_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
+					 blkno, name, lookup, dio);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	*ret_orphan_dir = orphan_dir_inode;
+
+out:
+	brelse(orphan_dir_bh);
+
+	if (ret) {
+		ocfs2_inode_unlock(orphan_dir_inode, 1);
+		mutex_unlock(&orphan_dir_inode->i_mutex);
+		iput(orphan_dir_inode);
+	}
+
+	if (ret)
+		mlog_errno(ret);
+	return ret;
+}
+
+static int ocfs2_orphan_add(struct ocfs2_super *osb,
+			    handle_t *handle,
+			    struct inode *inode,
+			    struct buffer_head *fe_bh,
+			    char *name,
+			    struct ocfs2_dir_lookup_result *lookup,
+			    struct inode *orphan_dir_inode,
+			    bool dio)
+{
+	struct buffer_head *orphan_dir_bh = NULL;
+	int status = 0;
+	struct ocfs2_dinode *orphan_fe;
+	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
+	int namelen = dio ?
+			(OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
+			OCFS2_ORPHAN_NAMELEN;
+
+	trace_ocfs2_orphan_add_begin(
+				(unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	status = ocfs2_journal_access_di(handle,
+					 INODE_CACHE(orphan_dir_inode),
+					 orphan_dir_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/*
+	 * We're going to journal the change of i_flags and i_orphaned_slot.
+	 * It's safe anyway, though some callers may duplicate the journaling.
+	 * Journaling within the func just make the logic look more
+	 * straightforward.
+	 */
+	status = ocfs2_journal_access_di(handle,
+					 INODE_CACHE(inode),
+					 fe_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* we're a cluster, and nlink can change on disk from
+	 * underneath us... */
+	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
+	if (S_ISDIR(inode->i_mode))
+		ocfs2_add_links_count(orphan_fe, 1);
+	set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
+	ocfs2_journal_dirty(handle, orphan_dir_bh);
+
+	status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
+				   namelen, inode,
+				   OCFS2_I(inode)->ip_blkno,
+				   orphan_dir_bh, lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto rollback;
+	}
+
+	if (dio) {
+		/* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
+		 * slot.
+		 */
+		fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
+		fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
+	} else {
+		fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
+		OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
+
+		/* Record which orphan dir our inode now resides
+		 * in. delete_inode will use this to determine which orphan
+		 * dir to lock. */
+		fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
+	}
+
+	ocfs2_journal_dirty(handle, fe_bh);
+
+	trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
+				   osb->slot_num);
+
+rollback:
+	if (status < 0) {
+		if (S_ISDIR(inode->i_mode))
+			ocfs2_add_links_count(orphan_fe, -1);
+		set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
+	}
+
+leave:
+	brelse(orphan_dir_bh);
+
+	return status;
+}
+
+/* unlike orphan_add, we expect the orphan dir to already be locked here. */
+int ocfs2_orphan_del(struct ocfs2_super *osb,
+		     handle_t *handle,
+		     struct inode *orphan_dir_inode,
+		     struct inode *inode,
+		     struct buffer_head *orphan_dir_bh,
+		     bool dio)
+{
+	const int namelen = OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN;
+	char name[namelen + 1];
+	struct ocfs2_dinode *orphan_fe;
+	int status = 0;
+	struct ocfs2_dir_lookup_result lookup = { NULL, };
+
+	if (dio) {
+		status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
+				OCFS2_DIO_ORPHAN_PREFIX);
+		if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
+			status = -EINVAL;
+			mlog_errno(status);
+			return status;
+		}
+
+		status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
+				name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
+	} else
+		status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	trace_ocfs2_orphan_del(
+	     (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
+	     name, strlen(name));
+
+	/* find it's spot in the orphan directory */
+	status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
+				  &lookup);
+	if (status) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* remove it from the orphan directory */
+	status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	status = ocfs2_journal_access_di(handle,
+					 INODE_CACHE(orphan_dir_inode),
+					 orphan_dir_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* do the i_nlink dance! :) */
+	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
+	if (S_ISDIR(inode->i_mode))
+		ocfs2_add_links_count(orphan_fe, -1);
+	set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
+	ocfs2_journal_dirty(handle, orphan_dir_bh);
+
+leave:
+	ocfs2_free_dir_lookup_result(&lookup);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/**
+ * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
+ * allocated file. This is different from the typical 'add to orphan dir'
+ * operation in that the inode does not yet exist. This is a problem because
+ * the orphan dir stringifies the inode block number to come up with it's
+ * dirent. Obviously if the inode does not yet exist we have a chicken and egg
+ * problem. This function works around it by calling deeper into the orphan
+ * and suballoc code than other callers. Use this only by necessity.
+ * @dir: The directory which this inode will ultimately wind up under - not the
+ * orphan dir!
+ * @dir_bh: buffer_head the @dir inode block
+ * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
+ * with the string to be used for orphan dirent. Pass back to the orphan dir
+ * code.
+ * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
+ * dir code.
+ * @ret_di_blkno: block number where the new inode will be allocated.
+ * @orphan_insert: Dir insert context to be passed back into orphan dir code.
+ * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
+ *
+ * Returns zero on success and the ret_orphan_dir, name and lookup
+ * fields will be populated.
+ *
+ * Returns non-zero on failure. 
+ */
+static int ocfs2_prep_new_orphaned_file(struct inode *dir,
+					struct buffer_head *dir_bh,
+					char *orphan_name,
+					struct inode **ret_orphan_dir,
+					u64 *ret_di_blkno,
+					struct ocfs2_dir_lookup_result *orphan_insert,
+					struct ocfs2_alloc_context **ret_inode_ac)
+{
+	int ret;
+	u64 di_blkno;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct inode *orphan_dir = NULL;
+	struct buffer_head *orphan_dir_bh = NULL;
+	struct ocfs2_alloc_context *inode_ac = NULL;
+
+	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	/* reserve an inode spot */
+	ret = ocfs2_reserve_new_inode(osb, &inode_ac);
+	if (ret < 0) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
+				       &di_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
+					 di_blkno, orphan_name, orphan_insert,
+					 false);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+out:
+	if (ret == 0) {
+		*ret_orphan_dir = orphan_dir;
+		*ret_di_blkno = di_blkno;
+		*ret_inode_ac = inode_ac;
+		/*
+		 * orphan_name and orphan_insert are already up to
+		 * date via prepare_orphan_dir
+		 */
+	} else {
+		/* Unroll reserve_new_inode* */
+		if (inode_ac)
+			ocfs2_free_alloc_context(inode_ac);
+
+		/* Unroll orphan dir locking */
+		mutex_unlock(&orphan_dir->i_mutex);
+		ocfs2_inode_unlock(orphan_dir, 1);
+		iput(orphan_dir);
+	}
+
+	brelse(orphan_dir_bh);
+
+	return ret;
+}
+
+int ocfs2_create_inode_in_orphan(struct inode *dir,
+				 int mode,
+				 struct inode **new_inode)
+{
+	int status, did_quota_inode = 0;
+	struct inode *inode = NULL;
+	struct inode *orphan_dir = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct ocfs2_dinode *di = NULL;
+	handle_t *handle = NULL;
+	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
+	struct buffer_head *parent_di_bh = NULL;
+	struct buffer_head *new_di_bh = NULL;
+	struct ocfs2_alloc_context *inode_ac = NULL;
+	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
+	u64 uninitialized_var(di_blkno), suballoc_loc;
+	u16 suballoc_bit;
+
+	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		return status;
+	}
+
+	status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
+					      orphan_name, &orphan_dir,
+					      &di_blkno, &orphan_insert, &inode_ac);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto leave;
+	}
+
+	inode = ocfs2_get_init_inode(dir, mode);
+	if (!inode) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	status = dquot_alloc_inode(inode);
+	if (status)
+		goto leave;
+	did_quota_inode = 1;
+
+	status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
+					      &suballoc_loc,
+					      &suballoc_bit, di_blkno);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	clear_nlink(inode);
+	/* do the real work now. */
+	status = __ocfs2_mknod_locked(dir, inode,
+				      0, &new_di_bh, parent_di_bh, handle,
+				      inode_ac, di_blkno, suballoc_loc,
+				      suballoc_bit);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	di = (struct ocfs2_dinode *)new_di_bh->b_data;
+	status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
+				  &orphan_insert, orphan_dir, false);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* get open lock so that only nodes can't remove it from orphan dir. */
+	status = ocfs2_open_lock(inode);
+	if (status < 0)
+		mlog_errno(status);
+
+	insert_inode_hash(inode);
+leave:
+	if (status < 0 && did_quota_inode)
+		dquot_free_inode(inode);
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+
+	if (orphan_dir) {
+		/* This was locked for us in ocfs2_prepare_orphan_dir() */
+		ocfs2_inode_unlock(orphan_dir, 1);
+		mutex_unlock(&orphan_dir->i_mutex);
+		iput(orphan_dir);
+	}
+
+	if ((status < 0) && inode) {
+		clear_nlink(inode);
+		iput(inode);
+	}
+
+	if (inode_ac)
+		ocfs2_free_alloc_context(inode_ac);
+
+	brelse(new_di_bh);
+
+	if (!status)
+		*new_inode = inode;
+
+	ocfs2_free_dir_lookup_result(&orphan_insert);
+
+	ocfs2_inode_unlock(dir, 1);
+	brelse(parent_di_bh);
+	return status;
+}
+
+static int ocfs2_dio_orphan_recovered(struct inode *inode)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di = NULL;
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return 0;
+	}
+
+	di = (struct ocfs2_dinode *) di_bh->b_data;
+	ret = !(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL));
+	ocfs2_inode_unlock(inode, 1);
+	brelse(di_bh);
+
+	return ret;
+}
+
+#define OCFS2_DIO_ORPHANED_FL_CHECK_INTERVAL 10000
+int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
+	struct inode *inode)
+{
+	char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
+	struct inode *orphan_dir_inode = NULL;
+	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
+	struct buffer_head *di_bh = NULL;
+	int status = 0;
+	handle_t *handle = NULL;
+	struct ocfs2_dinode *di = NULL;
+
+restart:
+	status = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	di = (struct ocfs2_dinode *) di_bh->b_data;
+	/*
+	 * Another append dio crashed?
+	 * If so, wait for recovery first.
+	 */
+	if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
+		ocfs2_inode_unlock(inode, 1);
+		brelse(di_bh);
+		wait_event_interruptible_timeout(OCFS2_I(inode)->append_dio_wq,
+				ocfs2_dio_orphan_recovered(inode),
+				msecs_to_jiffies(OCFS2_DIO_ORPHANED_FL_CHECK_INTERVAL));
+		goto restart;
+	}
+
+	status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
+			OCFS2_I(inode)->ip_blkno,
+			orphan_name,
+			&orphan_insert,
+			true);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_unlock_inode;
+	}
+
+	handle = ocfs2_start_trans(osb,
+			OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		goto bail_unlock_orphan;
+	}
+
+	status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
+			&orphan_insert, orphan_dir_inode, true);
+	if (status)
+		mlog_errno(status);
+
+	ocfs2_commit_trans(osb, handle);
+
+bail_unlock_orphan:
+	ocfs2_inode_unlock(orphan_dir_inode, 1);
+	mutex_unlock(&orphan_dir_inode->i_mutex);
+	iput(orphan_dir_inode);
+
+	ocfs2_free_dir_lookup_result(&orphan_insert);
+
+bail_unlock_inode:
+	ocfs2_inode_unlock(inode, 1);
+	brelse(di_bh);
+
+bail:
+	return status;
+}
+
+int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
+		struct inode *inode, int update_isize,
+		loff_t end)
+{
+	struct inode *orphan_dir_inode = NULL;
+	struct buffer_head *orphan_dir_bh = NULL;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di = NULL;
+	handle_t *handle = NULL;
+	int status = 0;
+
+	status = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	di = (struct ocfs2_dinode *) di_bh->b_data;
+
+	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
+			ORPHAN_DIR_SYSTEM_INODE,
+			le16_to_cpu(di->i_dio_orphaned_slot));
+	if (!orphan_dir_inode) {
+		status = -ENOENT;
+		mlog_errno(status);
+		goto bail_unlock_inode;
+	}
+
+	mutex_lock(&orphan_dir_inode->i_mutex);
+	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
+	if (status < 0) {
+		mutex_unlock(&orphan_dir_inode->i_mutex);
+		iput(orphan_dir_inode);
+		mlog_errno(status);
+		goto bail_unlock_inode;
+	}
+
+	handle = ocfs2_start_trans(osb,
+			OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		goto bail_unlock_orphan;
+	}
+
+	BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
+
+	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
+				inode, orphan_dir_bh, true);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_commit;
+	}
+
+	status = ocfs2_journal_access_di(handle,
+			INODE_CACHE(inode),
+			di_bh,
+			OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail_commit;
+	}
+
+	di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
+	di->i_dio_orphaned_slot = 0;
+
+	if (update_isize) {
+		status = ocfs2_set_inode_size(handle, inode, di_bh, end);
+		if (status)
+			mlog_errno(status);
+	} else
+		ocfs2_journal_dirty(handle, di_bh);
+
+bail_commit:
+	ocfs2_commit_trans(osb, handle);
+
+bail_unlock_orphan:
+	ocfs2_inode_unlock(orphan_dir_inode, 1);
+	mutex_unlock(&orphan_dir_inode->i_mutex);
+	brelse(orphan_dir_bh);
+	iput(orphan_dir_inode);
+
+bail_unlock_inode:
+	ocfs2_inode_unlock(inode, 1);
+	brelse(di_bh);
+
+bail:
+	return status;
+}
+
+int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
+				   struct inode *inode,
+				   struct dentry *dentry)
+{
+	int status = 0;
+	struct buffer_head *parent_di_bh = NULL;
+	handle_t *handle = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	struct ocfs2_dinode *dir_di, *di;
+	struct inode *orphan_dir_inode = NULL;
+	struct buffer_head *orphan_dir_bh = NULL;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dir_lookup_result lookup = { NULL, };
+
+	trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
+				dentry->d_name.len, dentry->d_name.name,
+				(unsigned long long)OCFS2_I(dir)->ip_blkno,
+				(unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
+	if (status < 0) {
+		if (status != -ENOENT)
+			mlog_errno(status);
+		return status;
+	}
+
+	dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
+	if (!dir_di->i_links_count) {
+		/* can't make a file in a deleted directory. */
+		status = -ENOENT;
+		goto leave;
+	}
+
+	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
+					   dentry->d_name.len);
+	if (status)
+		goto leave;
+
+	/* get a spot inside the dir. */
+	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
+					      dentry->d_name.name,
+					      dentry->d_name.len, &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
+						       ORPHAN_DIR_SYSTEM_INODE,
+						       osb->slot_num);
+	if (!orphan_dir_inode) {
+		status = -ENOENT;
+		mlog_errno(status);
+		goto leave;
+	}
+
+	mutex_lock(&orphan_dir_inode->i_mutex);
+
+	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		mutex_unlock(&orphan_dir_inode->i_mutex);
+		iput(orphan_dir_inode);
+		goto leave;
+	}
+
+	status = ocfs2_read_inode_block(inode, &di_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto orphan_unlock;
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto orphan_unlock;
+	}
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
+					 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
+				  orphan_dir_bh, false);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
+	di->i_orphaned_slot = 0;
+	set_nlink(inode, 1);
+	ocfs2_set_links_count(di, inode->i_nlink);
+	ocfs2_update_inode_fsync_trans(handle, inode, 1);
+	ocfs2_journal_dirty(handle, di_bh);
+
+	status = ocfs2_add_entry(handle, dentry, inode,
+				 OCFS2_I(inode)->ip_blkno, parent_di_bh,
+				 &lookup);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	status = ocfs2_dentry_attach_lock(dentry, inode,
+					  OCFS2_I(dir)->ip_blkno);
+	if (status) {
+		mlog_errno(status);
+		goto out_commit;
+	}
+
+	d_instantiate(dentry, inode);
+	status = 0;
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+orphan_unlock:
+	ocfs2_inode_unlock(orphan_dir_inode, 1);
+	mutex_unlock(&orphan_dir_inode->i_mutex);
+	iput(orphan_dir_inode);
+leave:
+
+	ocfs2_inode_unlock(dir, 1);
+
+	brelse(di_bh);
+	brelse(parent_di_bh);
+	brelse(orphan_dir_bh);
+
+	ocfs2_free_dir_lookup_result(&lookup);
+
+	if (status)
+		mlog_errno(status);
+
+	return status;
+}
+
+const struct inode_operations ocfs2_dir_iops = {
+	.create		= ocfs2_create,
+	.lookup		= ocfs2_lookup,
+	.link		= ocfs2_link,
+	.unlink		= ocfs2_unlink,
+	.rmdir		= ocfs2_unlink,
+	.symlink	= ocfs2_symlink,
+	.mkdir		= ocfs2_mkdir,
+	.mknod		= ocfs2_mknod,
+	.rename		= ocfs2_rename,
+	.setattr	= ocfs2_setattr,
+	.getattr	= ocfs2_getattr,
+	.permission	= ocfs2_permission,
+	.setxattr	= generic_setxattr,
+	.getxattr	= generic_getxattr,
+	.listxattr	= ocfs2_listxattr,
+	.removexattr	= generic_removexattr,
+	.fiemap         = ocfs2_fiemap,
+	.get_acl	= ocfs2_iop_get_acl,
+	.set_acl	= ocfs2_iop_set_acl,
+};
diff --git a/fs/ocfs2/namei.h b/fs/ocfs2/namei.h
new file mode 100644
index 000000000..5ddecce17
--- /dev/null
+++ b/fs/ocfs2/namei.h
@@ -0,0 +1,51 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * namei.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_NAMEI_H
+#define OCFS2_NAMEI_H
+
+extern const struct inode_operations ocfs2_dir_iops;
+
+struct dentry *ocfs2_get_parent(struct dentry *child);
+
+int ocfs2_orphan_del(struct ocfs2_super *osb,
+		     handle_t *handle,
+		     struct inode *orphan_dir_inode,
+		     struct inode *inode,
+		     struct buffer_head *orphan_dir_bh,
+		     bool dio);
+int ocfs2_create_inode_in_orphan(struct inode *dir,
+				 int mode,
+				 struct inode **new_inode);
+int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
+		struct inode *inode);
+int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
+		struct inode *inode, int update_isize,
+		loff_t end);
+int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
+				   struct inode *new_inode,
+				   struct dentry *new_dentry);
+
+#endif /* OCFS2_NAMEI_H */
diff --git a/fs/ocfs2/ocfs1_fs_compat.h b/fs/ocfs2/ocfs1_fs_compat.h
new file mode 100644
index 000000000..dfb313bda
--- /dev/null
+++ b/fs/ocfs2/ocfs1_fs_compat.h
@@ -0,0 +1,109 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ocfs1_fs_compat.h
+ *
+ * OCFS1 volume header definitions.  OCFS2 creates valid but unmountable
+ * OCFS1 volume headers on the first two sectors of an OCFS2 volume.
+ * This allows an OCFS1 volume to see the partition and cleanly fail to
+ * mount it.
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License, version 2,  as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef _OCFS1_FS_COMPAT_H
+#define _OCFS1_FS_COMPAT_H
+
+#define OCFS1_MAX_VOL_SIGNATURE_LEN          128
+#define OCFS1_MAX_MOUNT_POINT_LEN            128
+#define OCFS1_MAX_VOL_ID_LENGTH               16
+#define OCFS1_MAX_VOL_LABEL_LEN               64
+#define OCFS1_MAX_CLUSTER_NAME_LEN            64
+
+#define OCFS1_MAJOR_VERSION              (2)
+#define OCFS1_MINOR_VERSION              (0)
+#define OCFS1_VOLUME_SIGNATURE		 "OracleCFS"
+
+/*
+ * OCFS1 superblock.  Lives at sector 0.
+ */
+struct ocfs1_vol_disk_hdr
+{
+/*00*/	__u32 minor_version;
+	__u32 major_version;
+/*08*/	__u8 signature[OCFS1_MAX_VOL_SIGNATURE_LEN];
+/*88*/	__u8 mount_point[OCFS1_MAX_MOUNT_POINT_LEN];
+/*108*/	__u64 serial_num;
+/*110*/	__u64 device_size;
+	__u64 start_off;
+/*120*/	__u64 bitmap_off;
+	__u64 publ_off;
+/*130*/	__u64 vote_off;
+	__u64 root_bitmap_off;
+/*140*/	__u64 data_start_off;
+	__u64 root_bitmap_size;
+/*150*/	__u64 root_off;
+	__u64 root_size;
+/*160*/	__u64 cluster_size;
+	__u64 num_nodes;
+/*170*/	__u64 num_clusters;
+	__u64 dir_node_size;
+/*180*/	__u64 file_node_size;
+	__u64 internal_off;
+/*190*/	__u64 node_cfg_off;
+	__u64 node_cfg_size;
+/*1A0*/	__u64 new_cfg_off;
+	__u32 prot_bits;
+	__s32 excl_mount;
+/*1B0*/
+};
+
+
+struct ocfs1_disk_lock
+{
+/*00*/	__u32 curr_master;
+	__u8 file_lock;
+	__u8 compat_pad[3];  /* Not in original definition.  Used to
+				make the already existing alignment
+				explicit */
+	__u64 last_write_time;
+/*10*/	__u64 last_read_time;
+	__u32 writer_node_num;
+	__u32 reader_node_num;
+/*20*/	__u64 oin_node_map;
+	__u64 dlock_seq_num;
+/*30*/
+};
+
+/*
+ * OCFS1 volume label.  Lives at sector 1.
+ */
+struct ocfs1_vol_label
+{
+/*00*/	struct ocfs1_disk_lock disk_lock;
+/*30*/	__u8 label[OCFS1_MAX_VOL_LABEL_LEN];
+/*70*/	__u16 label_len;
+/*72*/	__u8 vol_id[OCFS1_MAX_VOL_ID_LENGTH];
+/*82*/	__u16 vol_id_len;
+/*84*/	__u8 cluster_name[OCFS1_MAX_CLUSTER_NAME_LEN];
+/*A4*/	__u16 cluster_name_len;
+/*A6*/
+};
+
+
+#endif /* _OCFS1_FS_COMPAT_H */
+
diff --git a/fs/ocfs2/ocfs2.h b/fs/ocfs2/ocfs2.h
new file mode 100644
index 000000000..460c6c37e
--- /dev/null
+++ b/fs/ocfs2/ocfs2.h
@@ -0,0 +1,919 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ocfs2.h
+ *
+ * Defines macros and structures used in OCFS2
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_H
+#define OCFS2_H
+
+#include <linux/spinlock.h>
+#include <linux/sched.h>
+#include <linux/wait.h>
+#include <linux/list.h>
+#include <linux/llist.h>
+#include <linux/rbtree.h>
+#include <linux/workqueue.h>
+#include <linux/kref.h>
+#include <linux/mutex.h>
+#include <linux/lockdep.h>
+#include <linux/jbd2.h>
+
+/* For union ocfs2_dlm_lksb */
+#include "stackglue.h"
+
+#include "ocfs2_fs.h"
+#include "ocfs2_lockid.h"
+#include "ocfs2_ioctl.h"
+
+/* For struct ocfs2_blockcheck_stats */
+#include "blockcheck.h"
+
+#include "reservations.h"
+
+/* Caching of metadata buffers */
+
+/* Most user visible OCFS2 inodes will have very few pieces of
+ * metadata, but larger files (including bitmaps, etc) must be taken
+ * into account when designing an access scheme. We allow a small
+ * amount of inlined blocks to be stored on an array and grow the
+ * structure into a rb tree when necessary. */
+#define OCFS2_CACHE_INFO_MAX_ARRAY 2
+
+/* Flags for ocfs2_caching_info */
+
+enum ocfs2_caching_info_flags {
+	/* Indicates that the metadata cache is using the inline array */
+	OCFS2_CACHE_FL_INLINE	= 1<<1,
+};
+
+struct ocfs2_caching_operations;
+struct ocfs2_caching_info {
+	/*
+	 * The parent structure provides the locks, but because the
+	 * parent structure can differ, it provides locking operations
+	 * to struct ocfs2_caching_info.
+	 */
+	const struct ocfs2_caching_operations *ci_ops;
+
+	/* next two are protected by trans_inc_lock */
+	/* which transaction were we created on? Zero if none. */
+	unsigned long		ci_created_trans;
+	/* last transaction we were a part of. */
+	unsigned long		ci_last_trans;
+
+	/* Cache structures */
+	unsigned int		ci_flags;
+	unsigned int		ci_num_cached;
+	union {
+	sector_t	ci_array[OCFS2_CACHE_INFO_MAX_ARRAY];
+		struct rb_root	ci_tree;
+	} ci_cache;
+};
+/*
+ * Need this prototype here instead of in uptodate.h because journal.h
+ * uses it.
+ */
+struct super_block *ocfs2_metadata_cache_get_super(struct ocfs2_caching_info *ci);
+
+/* this limits us to 256 nodes
+ * if we need more, we can do a kmalloc for the map */
+#define OCFS2_NODE_MAP_MAX_NODES    256
+struct ocfs2_node_map {
+	u16 num_nodes;
+	unsigned long map[BITS_TO_LONGS(OCFS2_NODE_MAP_MAX_NODES)];
+};
+
+enum ocfs2_ast_action {
+	OCFS2_AST_INVALID = 0,
+	OCFS2_AST_ATTACH,
+	OCFS2_AST_CONVERT,
+	OCFS2_AST_DOWNCONVERT,
+};
+
+/* actions for an unlockast function to take. */
+enum ocfs2_unlock_action {
+	OCFS2_UNLOCK_INVALID = 0,
+	OCFS2_UNLOCK_CANCEL_CONVERT,
+	OCFS2_UNLOCK_DROP_LOCK,
+};
+
+/* ocfs2_lock_res->l_flags flags. */
+#define OCFS2_LOCK_ATTACHED      (0x00000001) /* we have initialized
+					       * the lvb */
+#define OCFS2_LOCK_BUSY          (0x00000002) /* we are currently in
+					       * dlm_lock */
+#define OCFS2_LOCK_BLOCKED       (0x00000004) /* blocked waiting to
+					       * downconvert*/
+#define OCFS2_LOCK_LOCAL         (0x00000008) /* newly created inode */
+#define OCFS2_LOCK_NEEDS_REFRESH (0x00000010)
+#define OCFS2_LOCK_REFRESHING    (0x00000020)
+#define OCFS2_LOCK_INITIALIZED   (0x00000040) /* track initialization
+					       * for shutdown paths */
+#define OCFS2_LOCK_FREEING       (0x00000080) /* help dlmglue track
+					       * when to skip queueing
+					       * a lock because it's
+					       * about to be
+					       * dropped. */
+#define OCFS2_LOCK_QUEUED        (0x00000100) /* queued for downconvert */
+#define OCFS2_LOCK_NOCACHE       (0x00000200) /* don't use a holder count */
+#define OCFS2_LOCK_PENDING       (0x00000400) /* This lockres is pending a
+						 call to dlm_lock.  Only
+						 exists with BUSY set. */
+#define OCFS2_LOCK_UPCONVERT_FINISHING (0x00000800) /* blocks the dc thread
+						     * from downconverting
+						     * before the upconvert
+						     * has completed */
+
+#define OCFS2_LOCK_NONBLOCK_FINISHED (0x00001000) /* NONBLOCK cluster
+						   * lock has already
+						   * returned, do not block
+						   * dc thread from
+						   * downconverting */
+
+struct ocfs2_lock_res_ops;
+
+typedef void (*ocfs2_lock_callback)(int status, unsigned long data);
+
+#ifdef CONFIG_OCFS2_FS_STATS
+struct ocfs2_lock_stats {
+	u64		ls_total;	/* Total wait in NSEC */
+	u32		ls_gets;	/* Num acquires */
+	u32		ls_fail;	/* Num failed acquires */
+
+	/* Storing max wait in usecs saves 24 bytes per inode */
+	u32		ls_max;		/* Max wait in USEC */
+};
+#endif
+
+struct ocfs2_lock_res {
+	void                    *l_priv;
+	struct ocfs2_lock_res_ops *l_ops;
+
+
+	struct list_head         l_blocked_list;
+	struct list_head         l_mask_waiters;
+
+	unsigned long		 l_flags;
+	char                     l_name[OCFS2_LOCK_ID_MAX_LEN];
+	unsigned int             l_ro_holders;
+	unsigned int             l_ex_holders;
+	signed char		 l_level;
+	signed char		 l_requested;
+	signed char		 l_blocking;
+
+	/* Data packed - type enum ocfs2_lock_type */
+	unsigned char            l_type;
+
+	/* used from AST/BAST funcs. */
+	/* Data packed - enum type ocfs2_ast_action */
+	unsigned char            l_action;
+	/* Data packed - enum type ocfs2_unlock_action */
+	unsigned char            l_unlock_action;
+	unsigned int             l_pending_gen;
+
+	spinlock_t               l_lock;
+
+	struct ocfs2_dlm_lksb    l_lksb;
+
+	wait_queue_head_t        l_event;
+
+	struct list_head         l_debug_list;
+
+#ifdef CONFIG_OCFS2_FS_STATS
+	struct ocfs2_lock_stats  l_lock_prmode;		/* PR mode stats */
+	u32                      l_lock_refresh;	/* Disk refreshes */
+	struct ocfs2_lock_stats  l_lock_exmode;		/* EX mode stats */
+#endif
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+	struct lockdep_map	 l_lockdep_map;
+#endif
+};
+
+enum ocfs2_orphan_reco_type {
+	ORPHAN_NO_NEED_TRUNCATE = 0,
+	ORPHAN_NEED_TRUNCATE,
+};
+
+enum ocfs2_orphan_scan_state {
+	ORPHAN_SCAN_ACTIVE,
+	ORPHAN_SCAN_INACTIVE
+};
+
+struct ocfs2_orphan_scan {
+	struct mutex 		os_lock;
+	struct ocfs2_super 	*os_osb;
+	struct ocfs2_lock_res 	os_lockres;     /* lock to synchronize scans */
+	struct delayed_work 	os_orphan_scan_work;
+	struct timespec		os_scantime;  /* time this node ran the scan */
+	u32			os_count;      /* tracks node specific scans */
+	u32  			os_seqno;       /* tracks cluster wide scans */
+	atomic_t		os_state;              /* ACTIVE or INACTIVE */
+};
+
+struct ocfs2_dlm_debug {
+	struct kref d_refcnt;
+	struct dentry *d_locking_state;
+	struct list_head d_lockres_tracking;
+};
+
+enum ocfs2_vol_state
+{
+	VOLUME_INIT = 0,
+	VOLUME_MOUNTED,
+	VOLUME_MOUNTED_QUOTAS,
+	VOLUME_DISMOUNTED,
+	VOLUME_DISABLED
+};
+
+struct ocfs2_alloc_stats
+{
+	atomic_t moves;
+	atomic_t local_data;
+	atomic_t bitmap_data;
+	atomic_t bg_allocs;
+	atomic_t bg_extends;
+};
+
+enum ocfs2_local_alloc_state
+{
+	OCFS2_LA_UNUSED = 0,	/* Local alloc will never be used for
+				 * this mountpoint. */
+	OCFS2_LA_ENABLED,	/* Local alloc is in use. */
+	OCFS2_LA_THROTTLED,	/* Local alloc is in use, but number
+				 * of bits has been reduced. */
+	OCFS2_LA_DISABLED	/* Local alloc has temporarily been
+				 * disabled. */
+};
+
+enum ocfs2_mount_options
+{
+	OCFS2_MOUNT_HB_LOCAL = 1 << 0, /* Local heartbeat */
+	OCFS2_MOUNT_BARRIER = 1 << 1,	/* Use block barriers */
+	OCFS2_MOUNT_NOINTR  = 1 << 2,   /* Don't catch signals */
+	OCFS2_MOUNT_ERRORS_PANIC = 1 << 3, /* Panic on errors */
+	OCFS2_MOUNT_DATA_WRITEBACK = 1 << 4, /* No data ordering */
+	OCFS2_MOUNT_LOCALFLOCKS = 1 << 5, /* No cluster aware user file locks */
+	OCFS2_MOUNT_NOUSERXATTR = 1 << 6, /* No user xattr */
+	OCFS2_MOUNT_INODE64 = 1 << 7,	/* Allow inode numbers > 2^32 */
+	OCFS2_MOUNT_POSIX_ACL = 1 << 8,	/* Force POSIX access control lists */
+	OCFS2_MOUNT_NO_POSIX_ACL = 1 << 9,	/* Disable POSIX access
+						   control lists */
+	OCFS2_MOUNT_USRQUOTA = 1 << 10, /* We support user quotas */
+	OCFS2_MOUNT_GRPQUOTA = 1 << 11, /* We support group quotas */
+	OCFS2_MOUNT_COHERENCY_BUFFERED = 1 << 12, /* Allow concurrent O_DIRECT
+						     writes */
+	OCFS2_MOUNT_HB_NONE = 1 << 13, /* No heartbeat */
+	OCFS2_MOUNT_HB_GLOBAL = 1 << 14, /* Global heartbeat */
+
+	OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT = 1 << 15,  /* Journal Async Commit */
+};
+
+#define OCFS2_OSB_SOFT_RO	0x0001
+#define OCFS2_OSB_HARD_RO	0x0002
+#define OCFS2_OSB_ERROR_FS	0x0004
+#define OCFS2_DEFAULT_ATIME_QUANTUM	60
+
+struct ocfs2_journal;
+struct ocfs2_slot_info;
+struct ocfs2_recovery_map;
+struct ocfs2_replay_map;
+struct ocfs2_quota_recovery;
+struct ocfs2_super
+{
+	struct task_struct *commit_task;
+	struct super_block *sb;
+	struct inode *root_inode;
+	struct inode *sys_root_inode;
+	struct inode *global_system_inodes[NUM_GLOBAL_SYSTEM_INODES];
+	struct inode **local_system_inodes;
+
+	struct ocfs2_slot_info *slot_info;
+
+	u32 *slot_recovery_generations;
+
+	spinlock_t node_map_lock;
+
+	u64 root_blkno;
+	u64 system_dir_blkno;
+	u64 bitmap_blkno;
+	u32 bitmap_cpg;
+	u8 *uuid;
+	char *uuid_str;
+	u32 uuid_hash;
+	u8 *vol_label;
+	u64 first_cluster_group_blkno;
+	u32 fs_generation;
+
+	u32 s_feature_compat;
+	u32 s_feature_incompat;
+	u32 s_feature_ro_compat;
+
+	/* Protects s_next_generation, osb_flags and s_inode_steal_slot.
+	 * Could protect more on osb as it's very short lived.
+	 */
+	spinlock_t osb_lock;
+	u32 s_next_generation;
+	unsigned long osb_flags;
+	s16 s_inode_steal_slot;
+	s16 s_meta_steal_slot;
+	atomic_t s_num_inodes_stolen;
+	atomic_t s_num_meta_stolen;
+
+	unsigned long s_mount_opt;
+	unsigned int s_atime_quantum;
+
+	unsigned int max_slots;
+	unsigned int node_num;
+	int slot_num;
+	int preferred_slot;
+	int s_sectsize_bits;
+	int s_clustersize;
+	int s_clustersize_bits;
+	unsigned int s_xattr_inline_size;
+
+	atomic_t vol_state;
+	struct mutex recovery_lock;
+	struct ocfs2_recovery_map *recovery_map;
+	struct ocfs2_replay_map *replay_map;
+	struct task_struct *recovery_thread_task;
+	int disable_recovery;
+	wait_queue_head_t checkpoint_event;
+	struct ocfs2_journal *journal;
+	unsigned long osb_commit_interval;
+
+	struct delayed_work		la_enable_wq;
+
+	/*
+	 * Must hold local alloc i_mutex and osb->osb_lock to change
+	 * local_alloc_bits. Reads can be done under either lock.
+	 */
+	unsigned int local_alloc_bits;
+	unsigned int local_alloc_default_bits;
+	/* osb_clusters_at_boot can become stale! Do not trust it to
+	 * be up to date. */
+	unsigned int osb_clusters_at_boot;
+
+	enum ocfs2_local_alloc_state local_alloc_state; /* protected
+							 * by osb_lock */
+
+	struct buffer_head *local_alloc_bh;
+
+	u64 la_last_gd;
+
+	struct ocfs2_reservation_map	osb_la_resmap;
+
+	unsigned int	osb_resv_level;
+	unsigned int	osb_dir_resv_level;
+
+	/* Next three fields are for local node slot recovery during
+	 * mount. */
+	int dirty;
+	struct ocfs2_dinode *local_alloc_copy;
+	struct ocfs2_quota_recovery *quota_rec;
+
+	struct ocfs2_blockcheck_stats osb_ecc_stats;
+	struct ocfs2_alloc_stats alloc_stats;
+	char dev_str[20];		/* "major,minor" of the device */
+
+	u8 osb_stackflags;
+
+	char osb_cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
+	char osb_cluster_name[OCFS2_CLUSTER_NAME_LEN + 1];
+	struct ocfs2_cluster_connection *cconn;
+	struct ocfs2_lock_res osb_super_lockres;
+	struct ocfs2_lock_res osb_rename_lockres;
+	struct ocfs2_lock_res osb_nfs_sync_lockres;
+	struct ocfs2_dlm_debug *osb_dlm_debug;
+
+	struct dentry *osb_debug_root;
+	struct dentry *osb_ctxt;
+
+	wait_queue_head_t recovery_event;
+
+	spinlock_t dc_task_lock;
+	struct task_struct *dc_task;
+	wait_queue_head_t dc_event;
+	unsigned long dc_wake_sequence;
+	unsigned long dc_work_sequence;
+
+	/*
+	 * Any thread can add locks to the list, but the downconvert
+	 * thread is the only one allowed to remove locks. Any change
+	 * to this rule requires updating
+	 * ocfs2_downconvert_thread_do_work().
+	 */
+	struct list_head blocked_lock_list;
+	unsigned long blocked_lock_count;
+
+	/* List of dquot structures to drop last reference to */
+	struct llist_head dquot_drop_list;
+	struct work_struct dquot_drop_work;
+
+	wait_queue_head_t		osb_mount_event;
+
+	/* Truncate log info */
+	struct inode			*osb_tl_inode;
+	struct buffer_head		*osb_tl_bh;
+	struct delayed_work		osb_truncate_log_wq;
+	atomic_t			osb_tl_disable;
+	/*
+	 * How many clusters in our truncate log.
+	 * It must be protected by osb_tl_inode->i_mutex.
+	 */
+	unsigned int truncated_clusters;
+
+	struct ocfs2_node_map		osb_recovering_orphan_dirs;
+	unsigned int			*osb_orphan_wipes;
+	wait_queue_head_t		osb_wipe_event;
+
+	struct ocfs2_orphan_scan	osb_orphan_scan;
+
+	/* used to protect metaecc calculation check of xattr. */
+	spinlock_t osb_xattr_lock;
+
+	unsigned int			osb_dx_mask;
+	u32				osb_dx_seed[4];
+
+	/* the group we used to allocate inodes. */
+	u64				osb_inode_alloc_group;
+
+	/* rb tree root for refcount lock. */
+	struct rb_root	osb_rf_lock_tree;
+	struct ocfs2_refcount_tree *osb_ref_tree_lru;
+
+	struct mutex system_file_mutex;
+};
+
+#define OCFS2_SB(sb)	    ((struct ocfs2_super *)(sb)->s_fs_info)
+
+/* Useful typedef for passing around journal access functions */
+typedef int (*ocfs2_journal_access_func)(handle_t *handle,
+					 struct ocfs2_caching_info *ci,
+					 struct buffer_head *bh, int type);
+
+static inline int ocfs2_should_order_data(struct inode *inode)
+{
+	if (!S_ISREG(inode->i_mode))
+		return 0;
+	if (OCFS2_SB(inode->i_sb)->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)
+		return 0;
+	return 1;
+}
+
+static inline int ocfs2_sparse_alloc(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_SPARSE_ALLOC)
+		return 1;
+	return 0;
+}
+
+static inline int ocfs2_writes_unwritten_extents(struct ocfs2_super *osb)
+{
+	/*
+	 * Support for sparse files is a pre-requisite
+	 */
+	if (!ocfs2_sparse_alloc(osb))
+		return 0;
+
+	if (osb->s_feature_ro_compat & OCFS2_FEATURE_RO_COMPAT_UNWRITTEN)
+		return 1;
+	return 0;
+}
+
+static inline int ocfs2_supports_append_dio(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_APPEND_DIO)
+		return 1;
+	return 0;
+}
+
+
+static inline int ocfs2_supports_inline_data(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_INLINE_DATA)
+		return 1;
+	return 0;
+}
+
+static inline int ocfs2_supports_xattr(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_XATTR)
+		return 1;
+	return 0;
+}
+
+static inline int ocfs2_meta_ecc(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_META_ECC)
+		return 1;
+	return 0;
+}
+
+static inline int ocfs2_supports_indexed_dirs(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_INDEXED_DIRS)
+		return 1;
+	return 0;
+}
+
+static inline int ocfs2_supports_discontig_bg(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_DISCONTIG_BG)
+		return 1;
+	return 0;
+}
+
+static inline unsigned int ocfs2_link_max(struct ocfs2_super *osb)
+{
+	if (ocfs2_supports_indexed_dirs(osb))
+		return OCFS2_DX_LINK_MAX;
+	return OCFS2_LINK_MAX;
+}
+
+static inline unsigned int ocfs2_read_links_count(struct ocfs2_dinode *di)
+{
+	u32 nlink = le16_to_cpu(di->i_links_count);
+	u32 hi = le16_to_cpu(di->i_links_count_hi);
+
+	if (di->i_dyn_features & cpu_to_le16(OCFS2_INDEXED_DIR_FL))
+		nlink |= (hi << OCFS2_LINKS_HI_SHIFT);
+
+	return nlink;
+}
+
+static inline void ocfs2_set_links_count(struct ocfs2_dinode *di, u32 nlink)
+{
+	u16 lo, hi;
+
+	lo = nlink;
+	hi = nlink >> OCFS2_LINKS_HI_SHIFT;
+
+	di->i_links_count = cpu_to_le16(lo);
+	di->i_links_count_hi = cpu_to_le16(hi);
+}
+
+static inline void ocfs2_add_links_count(struct ocfs2_dinode *di, int n)
+{
+	u32 links = ocfs2_read_links_count(di);
+
+	links += n;
+
+	ocfs2_set_links_count(di, links);
+}
+
+static inline int ocfs2_refcount_tree(struct ocfs2_super *osb)
+{
+	if (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_REFCOUNT_TREE)
+		return 1;
+	return 0;
+}
+
+/* set / clear functions because cluster events can make these happen
+ * in parallel so we want the transitions to be atomic. this also
+ * means that any future flags osb_flags must be protected by spinlock
+ * too! */
+static inline void ocfs2_set_osb_flag(struct ocfs2_super *osb,
+				      unsigned long flag)
+{
+	spin_lock(&osb->osb_lock);
+	osb->osb_flags |= flag;
+	spin_unlock(&osb->osb_lock);
+}
+
+static inline void ocfs2_set_ro_flag(struct ocfs2_super *osb,
+				     int hard)
+{
+	spin_lock(&osb->osb_lock);
+	osb->osb_flags &= ~(OCFS2_OSB_SOFT_RO|OCFS2_OSB_HARD_RO);
+	if (hard)
+		osb->osb_flags |= OCFS2_OSB_HARD_RO;
+	else
+		osb->osb_flags |= OCFS2_OSB_SOFT_RO;
+	spin_unlock(&osb->osb_lock);
+}
+
+static inline int ocfs2_is_hard_readonly(struct ocfs2_super *osb)
+{
+	int ret;
+
+	spin_lock(&osb->osb_lock);
+	ret = osb->osb_flags & OCFS2_OSB_HARD_RO;
+	spin_unlock(&osb->osb_lock);
+
+	return ret;
+}
+
+static inline int ocfs2_is_soft_readonly(struct ocfs2_super *osb)
+{
+	int ret;
+
+	spin_lock(&osb->osb_lock);
+	ret = osb->osb_flags & OCFS2_OSB_SOFT_RO;
+	spin_unlock(&osb->osb_lock);
+
+	return ret;
+}
+
+static inline int ocfs2_clusterinfo_valid(struct ocfs2_super *osb)
+{
+	return (osb->s_feature_incompat &
+		(OCFS2_FEATURE_INCOMPAT_USERSPACE_STACK |
+		 OCFS2_FEATURE_INCOMPAT_CLUSTERINFO));
+}
+
+static inline int ocfs2_userspace_stack(struct ocfs2_super *osb)
+{
+	if (ocfs2_clusterinfo_valid(osb) &&
+	    memcmp(osb->osb_cluster_stack, OCFS2_CLASSIC_CLUSTER_STACK,
+		   OCFS2_STACK_LABEL_LEN))
+		return 1;
+	return 0;
+}
+
+static inline int ocfs2_o2cb_stack(struct ocfs2_super *osb)
+{
+	if (ocfs2_clusterinfo_valid(osb) &&
+	    !memcmp(osb->osb_cluster_stack, OCFS2_CLASSIC_CLUSTER_STACK,
+		   OCFS2_STACK_LABEL_LEN))
+		return 1;
+	return 0;
+}
+
+static inline int ocfs2_cluster_o2cb_global_heartbeat(struct ocfs2_super *osb)
+{
+	return ocfs2_o2cb_stack(osb) &&
+		(osb->osb_stackflags & OCFS2_CLUSTER_O2CB_GLOBAL_HEARTBEAT);
+}
+
+static inline int ocfs2_mount_local(struct ocfs2_super *osb)
+{
+	return (osb->s_feature_incompat & OCFS2_FEATURE_INCOMPAT_LOCAL_MOUNT);
+}
+
+static inline int ocfs2_uses_extended_slot_map(struct ocfs2_super *osb)
+{
+	return (osb->s_feature_incompat &
+		OCFS2_FEATURE_INCOMPAT_EXTENDED_SLOT_MAP);
+}
+
+
+#define OCFS2_IS_VALID_DINODE(ptr)					\
+	(!strcmp((ptr)->i_signature, OCFS2_INODE_SIGNATURE))
+
+#define OCFS2_IS_VALID_EXTENT_BLOCK(ptr)				\
+	(!strcmp((ptr)->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE))
+
+#define OCFS2_IS_VALID_GROUP_DESC(ptr)					\
+	(!strcmp((ptr)->bg_signature, OCFS2_GROUP_DESC_SIGNATURE))
+
+
+#define OCFS2_IS_VALID_XATTR_BLOCK(ptr)					\
+	(!strcmp((ptr)->xb_signature, OCFS2_XATTR_BLOCK_SIGNATURE))
+
+#define OCFS2_IS_VALID_DIR_TRAILER(ptr)					\
+	(!strcmp((ptr)->db_signature, OCFS2_DIR_TRAILER_SIGNATURE))
+
+#define OCFS2_IS_VALID_DX_ROOT(ptr)					\
+	(!strcmp((ptr)->dr_signature, OCFS2_DX_ROOT_SIGNATURE))
+
+#define OCFS2_IS_VALID_DX_LEAF(ptr)					\
+	(!strcmp((ptr)->dl_signature, OCFS2_DX_LEAF_SIGNATURE))
+
+#define OCFS2_IS_VALID_REFCOUNT_BLOCK(ptr)				\
+	(!strcmp((ptr)->rf_signature, OCFS2_REFCOUNT_BLOCK_SIGNATURE))
+
+static inline unsigned long ino_from_blkno(struct super_block *sb,
+					   u64 blkno)
+{
+	return (unsigned long)(blkno & (u64)ULONG_MAX);
+}
+
+static inline u64 ocfs2_clusters_to_blocks(struct super_block *sb,
+					   u32 clusters)
+{
+	int c_to_b_bits = OCFS2_SB(sb)->s_clustersize_bits -
+		sb->s_blocksize_bits;
+
+	return (u64)clusters << c_to_b_bits;
+}
+
+static inline u32 ocfs2_blocks_to_clusters(struct super_block *sb,
+					   u64 blocks)
+{
+	int b_to_c_bits = OCFS2_SB(sb)->s_clustersize_bits -
+		sb->s_blocksize_bits;
+
+	return (u32)(blocks >> b_to_c_bits);
+}
+
+static inline unsigned int ocfs2_clusters_for_bytes(struct super_block *sb,
+						    u64 bytes)
+{
+	int cl_bits = OCFS2_SB(sb)->s_clustersize_bits;
+	unsigned int clusters;
+
+	bytes += OCFS2_SB(sb)->s_clustersize - 1;
+	/* OCFS2 just cannot have enough clusters to overflow this */
+	clusters = (unsigned int)(bytes >> cl_bits);
+
+	return clusters;
+}
+
+static inline unsigned int ocfs2_bytes_to_clusters(struct super_block *sb,
+		u64 bytes)
+{
+	int cl_bits = OCFS2_SB(sb)->s_clustersize_bits;
+	unsigned int clusters;
+
+	clusters = (unsigned int)(bytes >> cl_bits);
+	return clusters;
+}
+
+static inline u64 ocfs2_blocks_for_bytes(struct super_block *sb,
+					 u64 bytes)
+{
+	bytes += sb->s_blocksize - 1;
+	return bytes >> sb->s_blocksize_bits;
+}
+
+static inline u64 ocfs2_clusters_to_bytes(struct super_block *sb,
+					  u32 clusters)
+{
+	return (u64)clusters << OCFS2_SB(sb)->s_clustersize_bits;
+}
+
+static inline u64 ocfs2_block_to_cluster_start(struct super_block *sb,
+					       u64 blocks)
+{
+	int bits = OCFS2_SB(sb)->s_clustersize_bits - sb->s_blocksize_bits;
+	unsigned int clusters;
+
+	clusters = ocfs2_blocks_to_clusters(sb, blocks);
+	return (u64)clusters << bits;
+}
+
+static inline u64 ocfs2_align_bytes_to_clusters(struct super_block *sb,
+						u64 bytes)
+{
+	int cl_bits = OCFS2_SB(sb)->s_clustersize_bits;
+	unsigned int clusters;
+
+	clusters = ocfs2_clusters_for_bytes(sb, bytes);
+	return (u64)clusters << cl_bits;
+}
+
+static inline u64 ocfs2_align_bytes_to_blocks(struct super_block *sb,
+					      u64 bytes)
+{
+	u64 blocks;
+
+        blocks = ocfs2_blocks_for_bytes(sb, bytes);
+	return blocks << sb->s_blocksize_bits;
+}
+
+static inline unsigned long ocfs2_align_bytes_to_sectors(u64 bytes)
+{
+	return (unsigned long)((bytes + 511) >> 9);
+}
+
+static inline unsigned int ocfs2_page_index_to_clusters(struct super_block *sb,
+							unsigned long pg_index)
+{
+	u32 clusters = pg_index;
+	unsigned int cbits = OCFS2_SB(sb)->s_clustersize_bits;
+
+	if (unlikely(PAGE_CACHE_SHIFT > cbits))
+		clusters = pg_index << (PAGE_CACHE_SHIFT - cbits);
+	else if (PAGE_CACHE_SHIFT < cbits)
+		clusters = pg_index >> (cbits - PAGE_CACHE_SHIFT);
+
+	return clusters;
+}
+
+/*
+ * Find the 1st page index which covers the given clusters.
+ */
+static inline pgoff_t ocfs2_align_clusters_to_page_index(struct super_block *sb,
+							u32 clusters)
+{
+	unsigned int cbits = OCFS2_SB(sb)->s_clustersize_bits;
+        pgoff_t index = clusters;
+
+	if (PAGE_CACHE_SHIFT > cbits) {
+		index = (pgoff_t)clusters >> (PAGE_CACHE_SHIFT - cbits);
+	} else if (PAGE_CACHE_SHIFT < cbits) {
+		index = (pgoff_t)clusters << (cbits - PAGE_CACHE_SHIFT);
+	}
+
+	return index;
+}
+
+static inline unsigned int ocfs2_pages_per_cluster(struct super_block *sb)
+{
+	unsigned int cbits = OCFS2_SB(sb)->s_clustersize_bits;
+	unsigned int pages_per_cluster = 1;
+
+	if (PAGE_CACHE_SHIFT < cbits)
+		pages_per_cluster = 1 << (cbits - PAGE_CACHE_SHIFT);
+
+	return pages_per_cluster;
+}
+
+static inline unsigned int ocfs2_megabytes_to_clusters(struct super_block *sb,
+						       unsigned int megs)
+{
+	BUILD_BUG_ON(OCFS2_MAX_CLUSTERSIZE > 1048576);
+
+	return megs << (20 - OCFS2_SB(sb)->s_clustersize_bits);
+}
+
+static inline unsigned int ocfs2_clusters_to_megabytes(struct super_block *sb,
+						       unsigned int clusters)
+{
+	return clusters >> (20 - OCFS2_SB(sb)->s_clustersize_bits);
+}
+
+static inline void _ocfs2_set_bit(unsigned int bit, unsigned long *bitmap)
+{
+	__set_bit_le(bit, bitmap);
+}
+#define ocfs2_set_bit(bit, addr) _ocfs2_set_bit((bit), (unsigned long *)(addr))
+
+static inline void _ocfs2_clear_bit(unsigned int bit, unsigned long *bitmap)
+{
+	__clear_bit_le(bit, bitmap);
+}
+#define ocfs2_clear_bit(bit, addr) _ocfs2_clear_bit((bit), (unsigned long *)(addr))
+
+#define ocfs2_test_bit test_bit_le
+#define ocfs2_find_next_zero_bit find_next_zero_bit_le
+#define ocfs2_find_next_bit find_next_bit_le
+
+static inline void *correct_addr_and_bit_unaligned(int *bit, void *addr)
+{
+#if BITS_PER_LONG == 64
+	*bit += ((unsigned long) addr & 7UL) << 3;
+	addr = (void *) ((unsigned long) addr & ~7UL);
+#elif BITS_PER_LONG == 32
+	*bit += ((unsigned long) addr & 3UL) << 3;
+	addr = (void *) ((unsigned long) addr & ~3UL);
+#else
+#error "how many bits you are?!"
+#endif
+	return addr;
+}
+
+static inline void ocfs2_set_bit_unaligned(int bit, void *bitmap)
+{
+	bitmap = correct_addr_and_bit_unaligned(&bit, bitmap);
+	ocfs2_set_bit(bit, bitmap);
+}
+
+static inline void ocfs2_clear_bit_unaligned(int bit, void *bitmap)
+{
+	bitmap = correct_addr_and_bit_unaligned(&bit, bitmap);
+	ocfs2_clear_bit(bit, bitmap);
+}
+
+static inline int ocfs2_test_bit_unaligned(int bit, void *bitmap)
+{
+	bitmap = correct_addr_and_bit_unaligned(&bit, bitmap);
+	return ocfs2_test_bit(bit, bitmap);
+}
+
+static inline int ocfs2_find_next_zero_bit_unaligned(void *bitmap, int max,
+							int start)
+{
+	int fix = 0, ret, tmpmax;
+	bitmap = correct_addr_and_bit_unaligned(&fix, bitmap);
+	tmpmax = max + fix;
+	start += fix;
+
+	ret = ocfs2_find_next_zero_bit(bitmap, tmpmax, start) - fix;
+	if (ret > max)
+		return max;
+	return ret;
+}
+
+#endif  /* OCFS2_H */
+
diff --git a/fs/ocfs2/ocfs2_fs.h b/fs/ocfs2/ocfs2_fs.h
new file mode 100644
index 000000000..db64ce2d4
--- /dev/null
+++ b/fs/ocfs2/ocfs2_fs.h
@@ -0,0 +1,1651 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ocfs2_fs.h
+ *
+ * On-disk structures for OCFS2.
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License, version 2,  as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef _OCFS2_FS_H
+#define _OCFS2_FS_H
+
+/* Version */
+#define OCFS2_MAJOR_REV_LEVEL		0
+#define OCFS2_MINOR_REV_LEVEL          	90
+
+/*
+ * An OCFS2 volume starts this way:
+ * Sector 0: Valid ocfs1_vol_disk_hdr that cleanly fails to mount OCFS.
+ * Sector 1: Valid ocfs1_vol_label that cleanly fails to mount OCFS.
+ * Block OCFS2_SUPER_BLOCK_BLKNO: OCFS2 superblock.
+ *
+ * All other structures are found from the superblock information.
+ *
+ * OCFS2_SUPER_BLOCK_BLKNO is in blocks, not sectors.  eg, for a
+ * blocksize of 2K, it is 4096 bytes into disk.
+ */
+#define OCFS2_SUPER_BLOCK_BLKNO		2
+
+/*
+ * Cluster size limits. The maximum is kept arbitrarily at 1 MB, and could
+ * grow if needed.
+ */
+#define OCFS2_MIN_CLUSTERSIZE		4096
+#define OCFS2_MAX_CLUSTERSIZE		1048576
+
+/*
+ * Blocks cannot be bigger than clusters, so the maximum blocksize is the
+ * minimum cluster size.
+ */
+#define OCFS2_MIN_BLOCKSIZE		512
+#define OCFS2_MAX_BLOCKSIZE		OCFS2_MIN_CLUSTERSIZE
+
+/* Filesystem magic number */
+#define OCFS2_SUPER_MAGIC		0x7461636f
+
+/* Object signatures */
+#define OCFS2_SUPER_BLOCK_SIGNATURE	"OCFSV2"
+#define OCFS2_INODE_SIGNATURE		"INODE01"
+#define OCFS2_EXTENT_BLOCK_SIGNATURE	"EXBLK01"
+#define OCFS2_GROUP_DESC_SIGNATURE      "GROUP01"
+#define OCFS2_XATTR_BLOCK_SIGNATURE	"XATTR01"
+#define OCFS2_DIR_TRAILER_SIGNATURE	"DIRTRL1"
+#define OCFS2_DX_ROOT_SIGNATURE		"DXDIR01"
+#define OCFS2_DX_LEAF_SIGNATURE		"DXLEAF1"
+#define OCFS2_REFCOUNT_BLOCK_SIGNATURE	"REFCNT1"
+
+/* Compatibility flags */
+#define OCFS2_HAS_COMPAT_FEATURE(sb,mask)			\
+	( OCFS2_SB(sb)->s_feature_compat & (mask) )
+#define OCFS2_HAS_RO_COMPAT_FEATURE(sb,mask)			\
+	( OCFS2_SB(sb)->s_feature_ro_compat & (mask) )
+#define OCFS2_HAS_INCOMPAT_FEATURE(sb,mask)			\
+	( OCFS2_SB(sb)->s_feature_incompat & (mask) )
+#define OCFS2_SET_COMPAT_FEATURE(sb,mask)			\
+	OCFS2_SB(sb)->s_feature_compat |= (mask)
+#define OCFS2_SET_RO_COMPAT_FEATURE(sb,mask)			\
+	OCFS2_SB(sb)->s_feature_ro_compat |= (mask)
+#define OCFS2_SET_INCOMPAT_FEATURE(sb,mask)			\
+	OCFS2_SB(sb)->s_feature_incompat |= (mask)
+#define OCFS2_CLEAR_COMPAT_FEATURE(sb,mask)			\
+	OCFS2_SB(sb)->s_feature_compat &= ~(mask)
+#define OCFS2_CLEAR_RO_COMPAT_FEATURE(sb,mask)			\
+	OCFS2_SB(sb)->s_feature_ro_compat &= ~(mask)
+#define OCFS2_CLEAR_INCOMPAT_FEATURE(sb,mask)			\
+	OCFS2_SB(sb)->s_feature_incompat &= ~(mask)
+
+#define OCFS2_FEATURE_COMPAT_SUPP	(OCFS2_FEATURE_COMPAT_BACKUP_SB	\
+					 | OCFS2_FEATURE_COMPAT_JBD2_SB)
+#define OCFS2_FEATURE_INCOMPAT_SUPP	(OCFS2_FEATURE_INCOMPAT_LOCAL_MOUNT \
+					 | OCFS2_FEATURE_INCOMPAT_SPARSE_ALLOC \
+					 | OCFS2_FEATURE_INCOMPAT_INLINE_DATA \
+					 | OCFS2_FEATURE_INCOMPAT_EXTENDED_SLOT_MAP \
+					 | OCFS2_FEATURE_INCOMPAT_USERSPACE_STACK \
+					 | OCFS2_FEATURE_INCOMPAT_XATTR \
+					 | OCFS2_FEATURE_INCOMPAT_META_ECC \
+					 | OCFS2_FEATURE_INCOMPAT_INDEXED_DIRS \
+					 | OCFS2_FEATURE_INCOMPAT_REFCOUNT_TREE \
+					 | OCFS2_FEATURE_INCOMPAT_DISCONTIG_BG	\
+					 | OCFS2_FEATURE_INCOMPAT_CLUSTERINFO \
+					 | OCFS2_FEATURE_INCOMPAT_APPEND_DIO)
+#define OCFS2_FEATURE_RO_COMPAT_SUPP	(OCFS2_FEATURE_RO_COMPAT_UNWRITTEN \
+					 | OCFS2_FEATURE_RO_COMPAT_USRQUOTA \
+					 | OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
+
+/*
+ * Heartbeat-only devices are missing journals and other files.  The
+ * filesystem driver can't load them, but the library can.  Never put
+ * this in OCFS2_FEATURE_INCOMPAT_SUPP, *ever*.
+ */
+#define OCFS2_FEATURE_INCOMPAT_HEARTBEAT_DEV	0x0002
+
+/*
+ * tunefs sets this incompat flag before starting the resize and clears it
+ * at the end. This flag protects users from inadvertently mounting the fs
+ * after an aborted run without fsck-ing.
+ */
+#define OCFS2_FEATURE_INCOMPAT_RESIZE_INPROG    0x0004
+
+/* Used to denote a non-clustered volume */
+#define OCFS2_FEATURE_INCOMPAT_LOCAL_MOUNT	0x0008
+
+/* Support for sparse allocation in b-trees */
+#define OCFS2_FEATURE_INCOMPAT_SPARSE_ALLOC	0x0010
+
+/*
+ * Tunefs sets this incompat flag before starting an operation which
+ * would require cleanup on abort. This is done to protect users from
+ * inadvertently mounting the fs after an aborted run without
+ * fsck-ing.
+ *
+ * s_tunefs_flags on the super block describes precisely which
+ * operations were in progress.
+ */
+#define OCFS2_FEATURE_INCOMPAT_TUNEFS_INPROG	0x0020
+
+/* Support for data packed into inode blocks */
+#define OCFS2_FEATURE_INCOMPAT_INLINE_DATA	0x0040
+
+/*
+ * Support for alternate, userspace cluster stacks.  If set, the superblock
+ * field s_cluster_info contains a tag for the alternate stack in use as
+ * well as the name of the cluster being joined.
+ * mount.ocfs2 must pass in a matching stack name.
+ *
+ * If not set, the classic stack will be used.  This is compatbile with
+ * all older versions.
+ */
+#define OCFS2_FEATURE_INCOMPAT_USERSPACE_STACK	0x0080
+
+/* Support for the extended slot map */
+#define OCFS2_FEATURE_INCOMPAT_EXTENDED_SLOT_MAP 0x100
+
+/* Support for extended attributes */
+#define OCFS2_FEATURE_INCOMPAT_XATTR		0x0200
+
+/* Support for indexed directores */
+#define OCFS2_FEATURE_INCOMPAT_INDEXED_DIRS	0x0400
+
+/* Metadata checksum and error correction */
+#define OCFS2_FEATURE_INCOMPAT_META_ECC		0x0800
+
+/* Refcount tree support */
+#define OCFS2_FEATURE_INCOMPAT_REFCOUNT_TREE	0x1000
+
+/* Discontigous block groups */
+#define OCFS2_FEATURE_INCOMPAT_DISCONTIG_BG	0x2000
+
+/*
+ * Incompat bit to indicate useable clusterinfo with stackflags for all
+ * cluster stacks (userspace adnd o2cb). If this bit is set,
+ * INCOMPAT_USERSPACE_STACK becomes superfluous and thus should not be set.
+ */
+#define OCFS2_FEATURE_INCOMPAT_CLUSTERINFO	0x4000
+
+/*
+ * Append Direct IO support
+ */
+#define OCFS2_FEATURE_INCOMPAT_APPEND_DIO	0x8000
+
+/*
+ * backup superblock flag is used to indicate that this volume
+ * has backup superblocks.
+ */
+#define OCFS2_FEATURE_COMPAT_BACKUP_SB		0x0001
+
+/*
+ * The filesystem will correctly handle journal feature bits.
+ */
+#define OCFS2_FEATURE_COMPAT_JBD2_SB		0x0002
+
+/*
+ * Unwritten extents support.
+ */
+#define OCFS2_FEATURE_RO_COMPAT_UNWRITTEN	0x0001
+
+/*
+ * Maintain quota information for this filesystem
+ */
+#define OCFS2_FEATURE_RO_COMPAT_USRQUOTA	0x0002
+#define OCFS2_FEATURE_RO_COMPAT_GRPQUOTA	0x0004
+
+
+/* The byte offset of the first backup block will be 1G.
+ * The following will be 4G, 16G, 64G, 256G and 1T.
+ */
+#define OCFS2_BACKUP_SB_START			1 << 30
+
+/* the max backup superblock nums */
+#define OCFS2_MAX_BACKUP_SUPERBLOCKS	6
+
+/*
+ * Flags on ocfs2_super_block.s_tunefs_flags
+ */
+#define OCFS2_TUNEFS_INPROG_REMOVE_SLOT		0x0001	/* Removing slots */
+
+/*
+ * Flags on ocfs2_dinode.i_flags
+ */
+#define OCFS2_VALID_FL		(0x00000001)	/* Inode is valid */
+#define OCFS2_UNUSED2_FL	(0x00000002)
+#define OCFS2_ORPHANED_FL	(0x00000004)	/* On the orphan list */
+#define OCFS2_UNUSED3_FL	(0x00000008)
+/* System inode flags */
+#define OCFS2_SYSTEM_FL		(0x00000010)	/* System inode */
+#define OCFS2_SUPER_BLOCK_FL	(0x00000020)	/* Super block */
+#define OCFS2_LOCAL_ALLOC_FL	(0x00000040)	/* Slot local alloc bitmap */
+#define OCFS2_BITMAP_FL		(0x00000080)	/* Allocation bitmap */
+#define OCFS2_JOURNAL_FL	(0x00000100)	/* Slot local journal */
+#define OCFS2_HEARTBEAT_FL	(0x00000200)	/* Heartbeat area */
+#define OCFS2_CHAIN_FL		(0x00000400)	/* Chain allocator */
+#define OCFS2_DEALLOC_FL	(0x00000800)	/* Truncate log */
+#define OCFS2_QUOTA_FL		(0x00001000)	/* Quota file */
+#define OCFS2_DIO_ORPHANED_FL	(0X00002000)	/* On the orphan list especially
+						 * for dio */
+
+/*
+ * Flags on ocfs2_dinode.i_dyn_features
+ *
+ * These can change much more often than i_flags. When adding flags,
+ * keep in mind that i_dyn_features is only 16 bits wide.
+ */
+#define OCFS2_INLINE_DATA_FL	(0x0001)	/* Data stored in inode block */
+#define OCFS2_HAS_XATTR_FL	(0x0002)
+#define OCFS2_INLINE_XATTR_FL	(0x0004)
+#define OCFS2_INDEXED_DIR_FL	(0x0008)
+#define OCFS2_HAS_REFCOUNT_FL   (0x0010)
+
+/* Inode attributes, keep in sync with EXT2 */
+#define OCFS2_SECRM_FL			FS_SECRM_FL	/* Secure deletion */
+#define OCFS2_UNRM_FL			FS_UNRM_FL	/* Undelete */
+#define OCFS2_COMPR_FL			FS_COMPR_FL	/* Compress file */
+#define OCFS2_SYNC_FL			FS_SYNC_FL	/* Synchronous updates */
+#define OCFS2_IMMUTABLE_FL		FS_IMMUTABLE_FL	/* Immutable file */
+#define OCFS2_APPEND_FL			FS_APPEND_FL	/* writes to file may only append */
+#define OCFS2_NODUMP_FL			FS_NODUMP_FL	/* do not dump file */
+#define OCFS2_NOATIME_FL		FS_NOATIME_FL	/* do not update atime */
+/* Reserved for compression usage... */
+#define OCFS2_DIRTY_FL			FS_DIRTY_FL
+#define OCFS2_COMPRBLK_FL		FS_COMPRBLK_FL	/* One or more compressed clusters */
+#define OCFS2_NOCOMP_FL			FS_NOCOMP_FL	/* Don't compress */
+#define OCFS2_ECOMPR_FL			FS_ECOMPR_FL	/* Compression error */
+/* End compression flags --- maybe not all used */
+#define OCFS2_BTREE_FL			FS_BTREE_FL	/* btree format dir */
+#define OCFS2_INDEX_FL			FS_INDEX_FL	/* hash-indexed directory */
+#define OCFS2_IMAGIC_FL			FS_IMAGIC_FL	/* AFS directory */
+#define OCFS2_JOURNAL_DATA_FL		FS_JOURNAL_DATA_FL /* Reserved for ext3 */
+#define OCFS2_NOTAIL_FL			FS_NOTAIL_FL	/* file tail should not be merged */
+#define OCFS2_DIRSYNC_FL		FS_DIRSYNC_FL	/* dirsync behaviour (directories only) */
+#define OCFS2_TOPDIR_FL			FS_TOPDIR_FL	/* Top of directory hierarchies*/
+#define OCFS2_RESERVED_FL		FS_RESERVED_FL	/* reserved for ext2 lib */
+
+#define OCFS2_FL_VISIBLE		FS_FL_USER_VISIBLE	/* User visible flags */
+#define OCFS2_FL_MODIFIABLE		FS_FL_USER_MODIFIABLE	/* User modifiable flags */
+
+/*
+ * Extent record flags (e_node.leaf.flags)
+ */
+#define OCFS2_EXT_UNWRITTEN		(0x01)	/* Extent is allocated but
+						 * unwritten */
+#define OCFS2_EXT_REFCOUNTED		(0x02)  /* Extent is reference
+						 * counted in an associated
+						 * refcount tree */
+
+/*
+ * Journal Flags (ocfs2_dinode.id1.journal1.i_flags)
+ */
+#define OCFS2_JOURNAL_DIRTY_FL	(0x00000001)	/* Journal needs recovery */
+
+/*
+ * superblock s_state flags
+ */
+#define OCFS2_ERROR_FS		(0x00000001)	/* FS saw errors */
+
+/* Limit of space in ocfs2_dir_entry */
+#define OCFS2_MAX_FILENAME_LEN		255
+
+/* Maximum slots on an ocfs2 file system */
+#define OCFS2_MAX_SLOTS			255
+
+/* Slot map indicator for an empty slot */
+#define OCFS2_INVALID_SLOT		-1
+
+#define OCFS2_VOL_UUID_LEN		16
+#define OCFS2_MAX_VOL_LABEL_LEN		64
+
+/* The cluster stack fields */
+#define OCFS2_STACK_LABEL_LEN		4
+#define OCFS2_CLUSTER_NAME_LEN		16
+
+/* Classic (historically speaking) cluster stack */
+#define OCFS2_CLASSIC_CLUSTER_STACK	"o2cb"
+
+/* Journal limits (in bytes) */
+#define OCFS2_MIN_JOURNAL_SIZE		(4 * 1024 * 1024)
+
+/*
+ * Inline extended attribute size (in bytes)
+ * The value chosen should be aligned to 16 byte boundaries.
+ */
+#define OCFS2_MIN_XATTR_INLINE_SIZE     256
+
+/*
+ * Cluster info flags (ocfs2_cluster_info.ci_stackflags)
+ */
+#define OCFS2_CLUSTER_O2CB_GLOBAL_HEARTBEAT	(0x01)
+
+struct ocfs2_system_inode_info {
+	char	*si_name;
+	int	si_iflags;
+	int	si_mode;
+};
+
+/* System file index */
+enum {
+	BAD_BLOCK_SYSTEM_INODE = 0,
+	GLOBAL_INODE_ALLOC_SYSTEM_INODE,
+	SLOT_MAP_SYSTEM_INODE,
+#define OCFS2_FIRST_ONLINE_SYSTEM_INODE SLOT_MAP_SYSTEM_INODE
+	HEARTBEAT_SYSTEM_INODE,
+	GLOBAL_BITMAP_SYSTEM_INODE,
+	USER_QUOTA_SYSTEM_INODE,
+	GROUP_QUOTA_SYSTEM_INODE,
+#define OCFS2_LAST_GLOBAL_SYSTEM_INODE GROUP_QUOTA_SYSTEM_INODE
+#define OCFS2_FIRST_LOCAL_SYSTEM_INODE ORPHAN_DIR_SYSTEM_INODE
+	ORPHAN_DIR_SYSTEM_INODE,
+	EXTENT_ALLOC_SYSTEM_INODE,
+	INODE_ALLOC_SYSTEM_INODE,
+	JOURNAL_SYSTEM_INODE,
+	LOCAL_ALLOC_SYSTEM_INODE,
+	TRUNCATE_LOG_SYSTEM_INODE,
+	LOCAL_USER_QUOTA_SYSTEM_INODE,
+	LOCAL_GROUP_QUOTA_SYSTEM_INODE,
+#define OCFS2_LAST_LOCAL_SYSTEM_INODE LOCAL_GROUP_QUOTA_SYSTEM_INODE
+	NUM_SYSTEM_INODES
+};
+#define NUM_GLOBAL_SYSTEM_INODES OCFS2_FIRST_LOCAL_SYSTEM_INODE
+#define NUM_LOCAL_SYSTEM_INODES	\
+		(NUM_SYSTEM_INODES - OCFS2_FIRST_LOCAL_SYSTEM_INODE)
+
+static struct ocfs2_system_inode_info ocfs2_system_inodes[NUM_SYSTEM_INODES] = {
+	/* Global system inodes (single copy) */
+	/* The first two are only used from userspace mfks/tunefs */
+	[BAD_BLOCK_SYSTEM_INODE]		= { "bad_blocks", 0, S_IFREG | 0644 },
+	[GLOBAL_INODE_ALLOC_SYSTEM_INODE] 	= { "global_inode_alloc", OCFS2_BITMAP_FL | OCFS2_CHAIN_FL, S_IFREG | 0644 },
+
+	/* These are used by the running filesystem */
+	[SLOT_MAP_SYSTEM_INODE]			= { "slot_map", 0, S_IFREG | 0644 },
+	[HEARTBEAT_SYSTEM_INODE]		= { "heartbeat", OCFS2_HEARTBEAT_FL, S_IFREG | 0644 },
+	[GLOBAL_BITMAP_SYSTEM_INODE]		= { "global_bitmap", 0, S_IFREG | 0644 },
+	[USER_QUOTA_SYSTEM_INODE]		= { "aquota.user", OCFS2_QUOTA_FL, S_IFREG | 0644 },
+	[GROUP_QUOTA_SYSTEM_INODE]		= { "aquota.group", OCFS2_QUOTA_FL, S_IFREG | 0644 },
+
+	/* Slot-specific system inodes (one copy per slot) */
+	[ORPHAN_DIR_SYSTEM_INODE]		= { "orphan_dir:%04d", 0, S_IFDIR | 0755 },
+	[EXTENT_ALLOC_SYSTEM_INODE]		= { "extent_alloc:%04d", OCFS2_BITMAP_FL | OCFS2_CHAIN_FL, S_IFREG | 0644 },
+	[INODE_ALLOC_SYSTEM_INODE]		= { "inode_alloc:%04d", OCFS2_BITMAP_FL | OCFS2_CHAIN_FL, S_IFREG | 0644 },
+	[JOURNAL_SYSTEM_INODE]			= { "journal:%04d", OCFS2_JOURNAL_FL, S_IFREG | 0644 },
+	[LOCAL_ALLOC_SYSTEM_INODE]		= { "local_alloc:%04d", OCFS2_BITMAP_FL | OCFS2_LOCAL_ALLOC_FL, S_IFREG | 0644 },
+	[TRUNCATE_LOG_SYSTEM_INODE]		= { "truncate_log:%04d", OCFS2_DEALLOC_FL, S_IFREG | 0644 },
+	[LOCAL_USER_QUOTA_SYSTEM_INODE]		= { "aquota.user:%04d", OCFS2_QUOTA_FL, S_IFREG | 0644 },
+	[LOCAL_GROUP_QUOTA_SYSTEM_INODE]	= { "aquota.group:%04d", OCFS2_QUOTA_FL, S_IFREG | 0644 },
+};
+
+/* Parameter passed from mount.ocfs2 to module */
+#define OCFS2_HB_NONE			"heartbeat=none"
+#define OCFS2_HB_LOCAL			"heartbeat=local"
+#define OCFS2_HB_GLOBAL			"heartbeat=global"
+
+/*
+ * OCFS2 directory file types.  Only the low 3 bits are used.  The
+ * other bits are reserved for now.
+ */
+#define OCFS2_FT_UNKNOWN	0
+#define OCFS2_FT_REG_FILE	1
+#define OCFS2_FT_DIR		2
+#define OCFS2_FT_CHRDEV		3
+#define OCFS2_FT_BLKDEV		4
+#define OCFS2_FT_FIFO		5
+#define OCFS2_FT_SOCK		6
+#define OCFS2_FT_SYMLINK	7
+
+#define OCFS2_FT_MAX		8
+
+/*
+ * OCFS2_DIR_PAD defines the directory entries boundaries
+ *
+ * NOTE: It must be a multiple of 4
+ */
+#define OCFS2_DIR_PAD			4
+#define OCFS2_DIR_ROUND			(OCFS2_DIR_PAD - 1)
+#define OCFS2_DIR_MEMBER_LEN 		offsetof(struct ocfs2_dir_entry, name)
+#define OCFS2_DIR_REC_LEN(name_len)	(((name_len) + OCFS2_DIR_MEMBER_LEN + \
+                                          OCFS2_DIR_ROUND) & \
+					 ~OCFS2_DIR_ROUND)
+#define OCFS2_DIR_MIN_REC_LEN	OCFS2_DIR_REC_LEN(1)
+
+#define OCFS2_LINK_MAX		32000
+#define	OCFS2_DX_LINK_MAX	((1U << 31) - 1U)
+#define	OCFS2_LINKS_HI_SHIFT	16
+#define	OCFS2_DX_ENTRIES_MAX	(0xffffffffU)
+
+#define S_SHIFT			12
+static unsigned char ocfs2_type_by_mode[S_IFMT >> S_SHIFT] = {
+	[S_IFREG >> S_SHIFT]  = OCFS2_FT_REG_FILE,
+	[S_IFDIR >> S_SHIFT]  = OCFS2_FT_DIR,
+	[S_IFCHR >> S_SHIFT]  = OCFS2_FT_CHRDEV,
+	[S_IFBLK >> S_SHIFT]  = OCFS2_FT_BLKDEV,
+	[S_IFIFO >> S_SHIFT]  = OCFS2_FT_FIFO,
+	[S_IFSOCK >> S_SHIFT] = OCFS2_FT_SOCK,
+	[S_IFLNK >> S_SHIFT]  = OCFS2_FT_SYMLINK,
+};
+
+
+/*
+ * Convenience casts
+ */
+#define OCFS2_RAW_SB(dinode)		(&((dinode)->id2.i_super))
+
+/*
+ * Block checking structure.  This is used in metadata to validate the
+ * contents.  If OCFS2_FEATURE_INCOMPAT_META_ECC is not set, it is all
+ * zeros.
+ */
+struct ocfs2_block_check {
+/*00*/	__le32 bc_crc32e;	/* 802.3 Ethernet II CRC32 */
+	__le16 bc_ecc;		/* Single-error-correction parity vector.
+				   This is a simple Hamming code dependent
+				   on the blocksize.  OCFS2's maximum
+				   blocksize, 4K, requires 16 parity bits,
+				   so we fit in __le16. */
+	__le16 bc_reserved1;
+/*08*/
+};
+
+/*
+ * On disk extent record for OCFS2
+ * It describes a range of clusters on disk.
+ *
+ * Length fields are divided into interior and leaf node versions.
+ * This leaves room for a flags field (OCFS2_EXT_*) in the leaf nodes.
+ */
+struct ocfs2_extent_rec {
+/*00*/	__le32 e_cpos;		/* Offset into the file, in clusters */
+	union {
+		__le32 e_int_clusters; /* Clusters covered by all children */
+		struct {
+			__le16 e_leaf_clusters; /* Clusters covered by this
+						   extent */
+			__u8 e_reserved1;
+			__u8 e_flags; /* Extent flags */
+		};
+	};
+	__le64 e_blkno;		/* Physical disk offset, in blocks */
+/*10*/
+};
+
+struct ocfs2_chain_rec {
+	__le32 c_free;	/* Number of free bits in this chain. */
+	__le32 c_total;	/* Number of total bits in this chain */
+	__le64 c_blkno;	/* Physical disk offset (blocks) of 1st group */
+};
+
+struct ocfs2_truncate_rec {
+	__le32 t_start;		/* 1st cluster in this log */
+	__le32 t_clusters;	/* Number of total clusters covered */
+};
+
+/*
+ * On disk extent list for OCFS2 (node in the tree).  Note that this
+ * is contained inside ocfs2_dinode or ocfs2_extent_block, so the
+ * offsets are relative to ocfs2_dinode.id2.i_list or
+ * ocfs2_extent_block.h_list, respectively.
+ */
+struct ocfs2_extent_list {
+/*00*/	__le16 l_tree_depth;		/* Extent tree depth from this
+					   point.  0 means data extents
+					   hang directly off this
+					   header (a leaf)
+					   NOTE: The high 8 bits cannot be
+					   used - tree_depth is never that big.
+					*/
+	__le16 l_count;			/* Number of extent records */
+	__le16 l_next_free_rec;		/* Next unused extent slot */
+	__le16 l_reserved1;
+	__le64 l_reserved2;		/* Pad to
+					   sizeof(ocfs2_extent_rec) */
+/*10*/	struct ocfs2_extent_rec l_recs[0];	/* Extent records */
+};
+
+/*
+ * On disk allocation chain list for OCFS2.  Note that this is
+ * contained inside ocfs2_dinode, so the offsets are relative to
+ * ocfs2_dinode.id2.i_chain.
+ */
+struct ocfs2_chain_list {
+/*00*/	__le16 cl_cpg;			/* Clusters per Block Group */
+	__le16 cl_bpc;			/* Bits per cluster */
+	__le16 cl_count;		/* Total chains in this list */
+	__le16 cl_next_free_rec;	/* Next unused chain slot */
+	__le64 cl_reserved1;
+/*10*/	struct ocfs2_chain_rec cl_recs[0];	/* Chain records */
+};
+
+/*
+ * On disk deallocation log for OCFS2.  Note that this is
+ * contained inside ocfs2_dinode, so the offsets are relative to
+ * ocfs2_dinode.id2.i_dealloc.
+ */
+struct ocfs2_truncate_log {
+/*00*/	__le16 tl_count;		/* Total records in this log */
+	__le16 tl_used;			/* Number of records in use */
+	__le32 tl_reserved1;
+/*08*/	struct ocfs2_truncate_rec tl_recs[0];	/* Truncate records */
+};
+
+/*
+ * On disk extent block (indirect block) for OCFS2
+ */
+struct ocfs2_extent_block
+{
+/*00*/	__u8 h_signature[8];		/* Signature for verification */
+	struct ocfs2_block_check h_check;	/* Error checking */
+/*10*/	__le16 h_suballoc_slot;		/* Slot suballocator this
+					   extent_header belongs to */
+	__le16 h_suballoc_bit;		/* Bit offset in suballocator
+					   block group */
+	__le32 h_fs_generation;		/* Must match super block */
+	__le64 h_blkno;			/* Offset on disk, in blocks */
+/*20*/	__le64 h_suballoc_loc;		/* Suballocator block group this
+					   eb belongs to.  Only valid
+					   if allocated from a
+					   discontiguous block group */
+	__le64 h_next_leaf_blk;		/* Offset on disk, in blocks,
+					   of next leaf header pointing
+					   to data */
+/*30*/	struct ocfs2_extent_list h_list;	/* Extent record list */
+/* Actual on-disk size is one block */
+};
+
+/*
+ * On disk slot map for OCFS2.  This defines the contents of the "slot_map"
+ * system file.  A slot is valid if it contains a node number >= 0.  The
+ * value -1 (0xFFFF) is OCFS2_INVALID_SLOT.  This marks a slot empty.
+ */
+struct ocfs2_slot_map {
+/*00*/	__le16 sm_slots[0];
+/*
+ * Actual on-disk size is one block.  OCFS2_MAX_SLOTS is 255,
+ * 255 * sizeof(__le16) == 512B, within the 512B block minimum blocksize.
+ */
+};
+
+struct ocfs2_extended_slot {
+/*00*/	__u8	es_valid;
+	__u8	es_reserved1[3];
+	__le32	es_node_num;
+/*10*/
+};
+
+/*
+ * The extended slot map, used when OCFS2_FEATURE_INCOMPAT_EXTENDED_SLOT_MAP
+ * is set.  It separates out the valid marker from the node number, and
+ * has room to grow.  Unlike the old slot map, this format is defined by
+ * i_size.
+ */
+struct ocfs2_slot_map_extended {
+/*00*/	struct ocfs2_extended_slot se_slots[0];
+/*
+ * Actual size is i_size of the slot_map system file.  It should
+ * match s_max_slots * sizeof(struct ocfs2_extended_slot)
+ */
+};
+
+/*
+ * ci_stackflags is only valid if the incompat bit
+ * OCFS2_FEATURE_INCOMPAT_CLUSTERINFO is set.
+ */
+struct ocfs2_cluster_info {
+/*00*/	__u8   ci_stack[OCFS2_STACK_LABEL_LEN];
+	union {
+		__le32 ci_reserved;
+		struct {
+			__u8 ci_stackflags;
+			__u8 ci_reserved1;
+			__u8 ci_reserved2;
+			__u8 ci_reserved3;
+		};
+	};
+/*08*/	__u8   ci_cluster[OCFS2_CLUSTER_NAME_LEN];
+/*18*/
+};
+
+/*
+ * On disk superblock for OCFS2
+ * Note that it is contained inside an ocfs2_dinode, so all offsets
+ * are relative to the start of ocfs2_dinode.id2.
+ */
+struct ocfs2_super_block {
+/*00*/	__le16 s_major_rev_level;
+	__le16 s_minor_rev_level;
+	__le16 s_mnt_count;
+	__le16 s_max_mnt_count;
+	__le16 s_state;			/* File system state */
+	__le16 s_errors;			/* Behaviour when detecting errors */
+	__le32 s_checkinterval;		/* Max time between checks */
+/*10*/	__le64 s_lastcheck;		/* Time of last check */
+	__le32 s_creator_os;		/* OS */
+	__le32 s_feature_compat;		/* Compatible feature set */
+/*20*/	__le32 s_feature_incompat;	/* Incompatible feature set */
+	__le32 s_feature_ro_compat;	/* Readonly-compatible feature set */
+	__le64 s_root_blkno;		/* Offset, in blocks, of root directory
+					   dinode */
+/*30*/	__le64 s_system_dir_blkno;	/* Offset, in blocks, of system
+					   directory dinode */
+	__le32 s_blocksize_bits;		/* Blocksize for this fs */
+	__le32 s_clustersize_bits;	/* Clustersize for this fs */
+/*40*/	__le16 s_max_slots;		/* Max number of simultaneous mounts
+					   before tunefs required */
+	__le16 s_tunefs_flag;
+	__le32 s_uuid_hash;		/* hash value of uuid */
+	__le64 s_first_cluster_group;	/* Block offset of 1st cluster
+					 * group header */
+/*50*/	__u8  s_label[OCFS2_MAX_VOL_LABEL_LEN];	/* Label for mounting, etc. */
+/*90*/	__u8  s_uuid[OCFS2_VOL_UUID_LEN];	/* 128-bit uuid */
+/*A0*/  struct ocfs2_cluster_info s_cluster_info; /* Only valid if either
+						     userspace or clusterinfo
+						     INCOMPAT flag set. */
+/*B8*/	__le16 s_xattr_inline_size;	/* extended attribute inline size
+					   for this fs*/
+	__le16 s_reserved0;
+	__le32 s_dx_seed[3];		/* seed[0-2] for dx dir hash.
+					 * s_uuid_hash serves as seed[3]. */
+/*C0*/  __le64 s_reserved2[15];		/* Fill out superblock */
+/*140*/
+
+	/*
+	 * NOTE: As stated above, all offsets are relative to
+	 * ocfs2_dinode.id2, which is at 0xC0 in the inode.
+	 * 0xC0 + 0x140 = 0x200 or 512 bytes.  A superblock must fit within
+	 * our smallest blocksize, which is 512 bytes.  To ensure this,
+	 * we reserve the space in s_reserved2.  Anything past s_reserved2
+	 * will not be available on the smallest blocksize.
+	 */
+};
+
+/*
+ * Local allocation bitmap for OCFS2 slots
+ * Note that it exists inside an ocfs2_dinode, so all offsets are
+ * relative to the start of ocfs2_dinode.id2.
+ */
+struct ocfs2_local_alloc
+{
+/*00*/	__le32 la_bm_off;	/* Starting bit offset in main bitmap */
+	__le16 la_size;		/* Size of included bitmap, in bytes */
+	__le16 la_reserved1;
+	__le64 la_reserved2;
+/*10*/	__u8   la_bitmap[0];
+};
+
+/*
+ * Data-in-inode header. This is only used if i_dyn_features has
+ * OCFS2_INLINE_DATA_FL set.
+ */
+struct ocfs2_inline_data
+{
+/*00*/	__le16	id_count;	/* Number of bytes that can be used
+				 * for data, starting at id_data */
+	__le16	id_reserved0;
+	__le32	id_reserved1;
+	__u8	id_data[0];	/* Start of user data */
+};
+
+/*
+ * On disk inode for OCFS2
+ */
+struct ocfs2_dinode {
+/*00*/	__u8 i_signature[8];		/* Signature for validation */
+	__le32 i_generation;		/* Generation number */
+	__le16 i_suballoc_slot;		/* Slot suballocator this inode
+					   belongs to */
+	__le16 i_suballoc_bit;		/* Bit offset in suballocator
+					   block group */
+/*10*/	__le16 i_links_count_hi;	/* High 16 bits of links count */
+	__le16 i_xattr_inline_size;
+	__le32 i_clusters;		/* Cluster count */
+	__le32 i_uid;			/* Owner UID */
+	__le32 i_gid;			/* Owning GID */
+/*20*/	__le64 i_size;			/* Size in bytes */
+	__le16 i_mode;			/* File mode */
+	__le16 i_links_count;		/* Links count */
+	__le32 i_flags;			/* File flags */
+/*30*/	__le64 i_atime;			/* Access time */
+	__le64 i_ctime;			/* Creation time */
+/*40*/	__le64 i_mtime;			/* Modification time */
+	__le64 i_dtime;			/* Deletion time */
+/*50*/	__le64 i_blkno;			/* Offset on disk, in blocks */
+	__le64 i_last_eb_blk;		/* Pointer to last extent
+					   block */
+/*60*/	__le32 i_fs_generation;		/* Generation per fs-instance */
+	__le32 i_atime_nsec;
+	__le32 i_ctime_nsec;
+	__le32 i_mtime_nsec;
+/*70*/	__le32 i_attr;
+	__le16 i_orphaned_slot;		/* Only valid when OCFS2_ORPHANED_FL
+					   was set in i_flags */
+	__le16 i_dyn_features;
+	__le64 i_xattr_loc;
+/*80*/	struct ocfs2_block_check i_check;	/* Error checking */
+/*88*/	__le64 i_dx_root;		/* Pointer to dir index root block */
+/*90*/	__le64 i_refcount_loc;
+	__le64 i_suballoc_loc;		/* Suballocator block group this
+					   inode belongs to.  Only valid
+					   if allocated from a
+					   discontiguous block group */
+/*A0*/	__le16 i_dio_orphaned_slot;	/* only used for append dio write */
+	__le16 i_reserved1[3];
+	__le64 i_reserved2[2];
+/*B8*/	union {
+		__le64 i_pad1;		/* Generic way to refer to this
+					   64bit union */
+		struct {
+			__le64 i_rdev;	/* Device number */
+		} dev1;
+		struct {		/* Info for bitmap system
+					   inodes */
+			__le32 i_used;	/* Bits (ie, clusters) used  */
+			__le32 i_total;	/* Total bits (clusters)
+					   available */
+		} bitmap1;
+		struct {		/* Info for journal system
+					   inodes */
+			__le32 ij_flags;	/* Mounted, version, etc. */
+			__le32 ij_recovery_generation; /* Incremented when the
+							  journal is recovered
+							  after an unclean
+							  shutdown */
+		} journal1;
+	} id1;				/* Inode type dependent 1 */
+/*C0*/	union {
+		struct ocfs2_super_block	i_super;
+		struct ocfs2_local_alloc	i_lab;
+		struct ocfs2_chain_list		i_chain;
+		struct ocfs2_extent_list	i_list;
+		struct ocfs2_truncate_log	i_dealloc;
+		struct ocfs2_inline_data	i_data;
+		__u8               		i_symlink[0];
+	} id2;
+/* Actual on-disk size is one block */
+};
+
+/*
+ * On-disk directory entry structure for OCFS2
+ *
+ * Packed as this structure could be accessed unaligned on 64-bit platforms
+ */
+struct ocfs2_dir_entry {
+/*00*/	__le64   inode;                  /* Inode number */
+	__le16   rec_len;                /* Directory entry length */
+	__u8    name_len;               /* Name length */
+	__u8    file_type;
+/*0C*/	char    name[OCFS2_MAX_FILENAME_LEN];   /* File name */
+/* Actual on-disk length specified by rec_len */
+} __attribute__ ((packed));
+
+/*
+ * Per-block record for the unindexed directory btree. This is carefully
+ * crafted so that the rec_len and name_len records of an ocfs2_dir_entry are
+ * mirrored. That way, the directory manipulation code needs a minimal amount
+ * of update.
+ *
+ * NOTE: Keep this structure aligned to a multiple of 4 bytes.
+ */
+struct ocfs2_dir_block_trailer {
+/*00*/	__le64		db_compat_inode;	/* Always zero. Was inode */
+
+	__le16		db_compat_rec_len;	/* Backwards compatible with
+						 * ocfs2_dir_entry. */
+	__u8		db_compat_name_len;	/* Always zero. Was name_len */
+	__u8		db_reserved0;
+	__le16		db_reserved1;
+	__le16		db_free_rec_len;	/* Size of largest empty hole
+						 * in this block. (unused) */
+/*10*/	__u8		db_signature[8];	/* Signature for verification */
+	__le64		db_reserved2;
+	__le64		db_free_next;		/* Next block in list (unused) */
+/*20*/	__le64		db_blkno;		/* Offset on disk, in blocks */
+	__le64		db_parent_dinode;	/* dinode which owns me, in
+						   blocks */
+/*30*/	struct ocfs2_block_check db_check;	/* Error checking */
+/*40*/
+};
+
+ /*
+ * A directory entry in the indexed tree. We don't store the full name here,
+ * but instead provide a pointer to the full dirent in the unindexed tree.
+ *
+ * We also store name_len here so as to reduce the number of leaf blocks we
+ * need to search in case of collisions.
+ */
+struct ocfs2_dx_entry {
+	__le32		dx_major_hash;	/* Used to find logical
+					 * cluster in index */
+	__le32		dx_minor_hash;	/* Lower bits used to find
+					 * block in cluster */
+	__le64		dx_dirent_blk;	/* Physical block in unindexed
+					 * tree holding this dirent. */
+};
+
+struct ocfs2_dx_entry_list {
+	__le32		de_reserved;
+	__le16		de_count;	/* Maximum number of entries
+					 * possible in de_entries */
+	__le16		de_num_used;	/* Current number of
+					 * de_entries entries */
+	struct	ocfs2_dx_entry		de_entries[0];	/* Indexed dir entries
+							 * in a packed array of
+							 * length de_num_used */
+};
+
+#define OCFS2_DX_FLAG_INLINE	0x01
+
+/*
+ * A directory indexing block. Each indexed directory has one of these,
+ * pointed to by ocfs2_dinode.
+ *
+ * This block stores an indexed btree root, and a set of free space
+ * start-of-list pointers.
+ */
+struct ocfs2_dx_root_block {
+	__u8		dr_signature[8];	/* Signature for verification */
+	struct ocfs2_block_check dr_check;	/* Error checking */
+	__le16		dr_suballoc_slot;	/* Slot suballocator this
+						 * block belongs to. */
+	__le16		dr_suballoc_bit;	/* Bit offset in suballocator
+						 * block group */
+	__le32		dr_fs_generation;	/* Must match super block */
+	__le64		dr_blkno;		/* Offset on disk, in blocks */
+	__le64		dr_last_eb_blk;		/* Pointer to last
+						 * extent block */
+	__le32		dr_clusters;		/* Clusters allocated
+						 * to the indexed tree. */
+	__u8		dr_flags;		/* OCFS2_DX_FLAG_* flags */
+	__u8		dr_reserved0;
+	__le16		dr_reserved1;
+	__le64		dr_dir_blkno;		/* Pointer to parent inode */
+	__le32		dr_num_entries;		/* Total number of
+						 * names stored in
+						 * this directory.*/
+	__le32		dr_reserved2;
+	__le64		dr_free_blk;		/* Pointer to head of free
+						 * unindexed block list. */
+	__le64		dr_suballoc_loc;	/* Suballocator block group
+						   this root belongs to.
+						   Only valid if allocated
+						   from a discontiguous
+						   block group */
+	__le64		dr_reserved3[14];
+	union {
+		struct ocfs2_extent_list dr_list; /* Keep this aligned to 128
+						   * bits for maximum space
+						   * efficiency. */
+		struct ocfs2_dx_entry_list dr_entries; /* In-root-block list of
+							* entries. We grow out
+							* to extents if this
+							* gets too big. */
+	};
+};
+
+/*
+ * The header of a leaf block in the indexed tree.
+ */
+struct ocfs2_dx_leaf {
+	__u8		dl_signature[8];/* Signature for verification */
+	struct ocfs2_block_check dl_check;	/* Error checking */
+	__le64		dl_blkno;	/* Offset on disk, in blocks */
+	__le32		dl_fs_generation;/* Must match super block */
+	__le32		dl_reserved0;
+	__le64		dl_reserved1;
+	struct ocfs2_dx_entry_list	dl_list;
+};
+
+/*
+ * Largest bitmap for a block (suballocator) group in bytes.  This limit
+ * does not affect cluster groups (global allocator).  Cluster group
+ * bitmaps run to the end of the block.
+ */
+#define OCFS2_MAX_BG_BITMAP_SIZE	256
+
+/*
+ * On disk allocator group structure for OCFS2
+ */
+struct ocfs2_group_desc
+{
+/*00*/	__u8    bg_signature[8];        /* Signature for validation */
+	__le16   bg_size;                /* Size of included bitmap in
+					   bytes. */
+	__le16   bg_bits;                /* Bits represented by this
+					   group. */
+	__le16	bg_free_bits_count;     /* Free bits count */
+	__le16   bg_chain;               /* What chain I am in. */
+/*10*/	__le32   bg_generation;
+	__le32	bg_reserved1;
+	__le64   bg_next_group;          /* Next group in my list, in
+					   blocks */
+/*20*/	__le64   bg_parent_dinode;       /* dinode which owns me, in
+					   blocks */
+	__le64   bg_blkno;               /* Offset on disk, in blocks */
+/*30*/	struct ocfs2_block_check bg_check;	/* Error checking */
+	__le64   bg_reserved2;
+/*40*/	union {
+		__u8    bg_bitmap[0];
+		struct {
+			/*
+			 * Block groups may be discontiguous when
+			 * OCFS2_FEATURE_INCOMPAT_DISCONTIG_BG is set.
+			 * The extents of a discontigous block group are
+			 * stored in bg_list.  It is a flat list.
+			 * l_tree_depth must always be zero.  A
+			 * discontiguous group is signified by a non-zero
+			 * bg_list->l_next_free_rec.  Only block groups
+			 * can be discontiguous; Cluster groups cannot.
+			 * We've never made a block group with more than
+			 * 2048 blocks (256 bytes of bg_bitmap).  This
+			 * codifies that limit so that we can fit bg_list.
+			 * bg_size of a discontiguous block group will
+			 * be 256 to match bg_bitmap_filler.
+			 */
+			__u8 bg_bitmap_filler[OCFS2_MAX_BG_BITMAP_SIZE];
+/*140*/			struct ocfs2_extent_list bg_list;
+		};
+	};
+/* Actual on-disk size is one block */
+};
+
+struct ocfs2_refcount_rec {
+/*00*/	__le64 r_cpos;		/* Physical offset, in clusters */
+	__le32 r_clusters;	/* Clusters covered by this extent */
+	__le32 r_refcount;	/* Reference count of this extent */
+/*10*/
+};
+#define OCFS2_32BIT_POS_MASK		(0xffffffffULL)
+
+#define OCFS2_REFCOUNT_LEAF_FL          (0x00000001)
+#define OCFS2_REFCOUNT_TREE_FL          (0x00000002)
+
+struct ocfs2_refcount_list {
+/*00*/	__le16 rl_count;	/* Maximum number of entries possible
+				   in rl_records */
+	__le16 rl_used;		/* Current number of used records */
+	__le32 rl_reserved2;
+	__le64 rl_reserved1;	/* Pad to sizeof(ocfs2_refcount_record) */
+/*10*/	struct ocfs2_refcount_rec rl_recs[0];	/* Refcount records */
+};
+
+
+struct ocfs2_refcount_block {
+/*00*/	__u8 rf_signature[8];		/* Signature for verification */
+	__le16 rf_suballoc_slot;	/* Slot suballocator this block
+					   belongs to */
+	__le16 rf_suballoc_bit;		/* Bit offset in suballocator
+					   block group */
+	__le32 rf_fs_generation;	/* Must match superblock */
+/*10*/	__le64 rf_blkno;		/* Offset on disk, in blocks */
+	__le64 rf_parent;		/* Parent block, only valid if
+					   OCFS2_REFCOUNT_LEAF_FL is set in
+					   rf_flags */
+/*20*/	struct ocfs2_block_check rf_check;	/* Error checking */
+	__le64 rf_last_eb_blk;		/* Pointer to last extent block */
+/*30*/	__le32 rf_count;		/* Number of inodes sharing this
+					   refcount tree */
+	__le32 rf_flags;		/* See the flags above */
+	__le32 rf_clusters;		/* clusters covered by refcount tree. */
+	__le32 rf_cpos;			/* cluster offset in refcount tree.*/
+/*40*/	__le32 rf_generation;		/* generation number. all be the same
+					 * for the same refcount tree. */
+	__le32 rf_reserved0;
+	__le64 rf_suballoc_loc;		/* Suballocator block group this
+					   refcount block belongs to. Only
+					   valid if allocated from a
+					   discontiguous block group */
+/*50*/	__le64 rf_reserved1[6];
+/*80*/	union {
+		struct ocfs2_refcount_list rf_records;  /* List of refcount
+							  records */
+		struct ocfs2_extent_list rf_list;	/* Extent record list,
+							only valid if
+							OCFS2_REFCOUNT_TREE_FL
+							is set in rf_flags */
+	};
+/* Actual on-disk size is one block */
+};
+
+/*
+ * On disk extended attribute structure for OCFS2.
+ */
+
+/*
+ * ocfs2_xattr_entry indicates one extend attribute.
+ *
+ * Note that it can be stored in inode, one block or one xattr bucket.
+ */
+struct ocfs2_xattr_entry {
+	__le32	xe_name_hash;    /* hash value of xattr prefix+suffix. */
+	__le16	xe_name_offset;  /* byte offset from the 1st entry in the
+				    local xattr storage(inode, xattr block or
+				    xattr bucket). */
+	__u8	xe_name_len;	 /* xattr name len, doesn't include prefix. */
+	__u8	xe_type;         /* the low 7 bits indicate the name prefix
+				  * type and the highest bit indicates whether
+				  * the EA is stored in the local storage. */
+	__le64	xe_value_size;	 /* real xattr value length. */
+};
+
+/*
+ * On disk structure for xattr header.
+ *
+ * One ocfs2_xattr_header describes how many ocfs2_xattr_entry records in
+ * the local xattr storage.
+ */
+struct ocfs2_xattr_header {
+	__le16	xh_count;                       /* contains the count of how
+						   many records are in the
+						   local xattr storage. */
+	__le16	xh_free_start;                  /* current offset for storing
+						   xattr. */
+	__le16	xh_name_value_len;              /* total length of name/value
+						   length in this bucket. */
+	__le16	xh_num_buckets;                 /* Number of xattr buckets
+						   in this extent record,
+						   only valid in the first
+						   bucket. */
+	struct ocfs2_block_check xh_check;	/* Error checking
+						   (Note, this is only
+						    used for xattr
+						    buckets.  A block uses
+						    xb_check and sets
+						    this field to zero.) */
+	struct ocfs2_xattr_entry xh_entries[0]; /* xattr entry list. */
+};
+
+/*
+ * On disk structure for xattr value root.
+ *
+ * When an xattr's value is large enough, it is stored in an external
+ * b-tree like file data.  The xattr value root points to this structure.
+ */
+struct ocfs2_xattr_value_root {
+/*00*/	__le32	xr_clusters;              /* clusters covered by xattr value. */
+	__le32	xr_reserved0;
+	__le64	xr_last_eb_blk;           /* Pointer to last extent block */
+/*10*/	struct ocfs2_extent_list xr_list; /* Extent record list */
+};
+
+/*
+ * On disk structure for xattr tree root.
+ *
+ * It is used when there are too many extended attributes for one file. These
+ * attributes will be organized and stored in an indexed-btree.
+ */
+struct ocfs2_xattr_tree_root {
+/*00*/	__le32	xt_clusters;              /* clusters covered by xattr. */
+	__le32	xt_reserved0;
+	__le64	xt_last_eb_blk;           /* Pointer to last extent block */
+/*10*/	struct ocfs2_extent_list xt_list; /* Extent record list */
+};
+
+#define OCFS2_XATTR_INDEXED	0x1
+#define OCFS2_HASH_SHIFT	5
+#define OCFS2_XATTR_ROUND	3
+#define OCFS2_XATTR_SIZE(size)	(((size) + OCFS2_XATTR_ROUND) & \
+				~(OCFS2_XATTR_ROUND))
+
+#define OCFS2_XATTR_BUCKET_SIZE			4096
+#define OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET 	(OCFS2_XATTR_BUCKET_SIZE \
+						 / OCFS2_MIN_BLOCKSIZE)
+
+/*
+ * On disk structure for xattr block.
+ */
+struct ocfs2_xattr_block {
+/*00*/	__u8	xb_signature[8];     /* Signature for verification */
+	__le16	xb_suballoc_slot;    /* Slot suballocator this
+					block belongs to. */
+	__le16	xb_suballoc_bit;     /* Bit offset in suballocator
+					block group */
+	__le32	xb_fs_generation;    /* Must match super block */
+/*10*/	__le64	xb_blkno;            /* Offset on disk, in blocks */
+	struct ocfs2_block_check xb_check;	/* Error checking */
+/*20*/	__le16	xb_flags;            /* Indicates whether this block contains
+					real xattr or a xattr tree. */
+	__le16	xb_reserved0;
+	__le32  xb_reserved1;
+	__le64	xb_suballoc_loc;	/* Suballocator block group this
+					   xattr block belongs to. Only
+					   valid if allocated from a
+					   discontiguous block group */
+/*30*/	union {
+		struct ocfs2_xattr_header xb_header; /* xattr header if this
+							block contains xattr */
+		struct ocfs2_xattr_tree_root xb_root;/* xattr tree root if this
+							block cotains xattr
+							tree. */
+	} xb_attrs;
+};
+
+#define OCFS2_XATTR_ENTRY_LOCAL		0x80
+#define OCFS2_XATTR_TYPE_MASK		0x7F
+static inline void ocfs2_xattr_set_local(struct ocfs2_xattr_entry *xe,
+					 int local)
+{
+	if (local)
+		xe->xe_type |= OCFS2_XATTR_ENTRY_LOCAL;
+	else
+		xe->xe_type &= ~OCFS2_XATTR_ENTRY_LOCAL;
+}
+
+static inline int ocfs2_xattr_is_local(struct ocfs2_xattr_entry *xe)
+{
+	return xe->xe_type & OCFS2_XATTR_ENTRY_LOCAL;
+}
+
+static inline void ocfs2_xattr_set_type(struct ocfs2_xattr_entry *xe, int type)
+{
+	xe->xe_type |= type & OCFS2_XATTR_TYPE_MASK;
+}
+
+static inline int ocfs2_xattr_get_type(struct ocfs2_xattr_entry *xe)
+{
+	return xe->xe_type & OCFS2_XATTR_TYPE_MASK;
+}
+
+/*
+ *  On disk structures for global quota file
+ */
+
+/* Magic numbers and known versions for global quota files */
+#define OCFS2_GLOBAL_QMAGICS {\
+	0x0cf52470, /* USRQUOTA */ \
+	0x0cf52471  /* GRPQUOTA */ \
+}
+
+#define OCFS2_GLOBAL_QVERSIONS {\
+	0, \
+	0, \
+}
+
+
+/* Each block of each quota file has a certain fixed number of bytes reserved
+ * for OCFS2 internal use at its end. OCFS2 can use it for things like
+ * checksums, etc. */
+#define OCFS2_QBLK_RESERVED_SPACE 8
+
+/* Generic header of all quota files */
+struct ocfs2_disk_dqheader {
+	__le32 dqh_magic;	/* Magic number identifying file */
+	__le32 dqh_version;	/* Quota format version */
+};
+
+#define OCFS2_GLOBAL_INFO_OFF (sizeof(struct ocfs2_disk_dqheader))
+
+/* Information header of global quota file (immediately follows the generic
+ * header) */
+struct ocfs2_global_disk_dqinfo {
+/*00*/	__le32 dqi_bgrace;	/* Grace time for space softlimit excess */
+	__le32 dqi_igrace;	/* Grace time for inode softlimit excess */
+	__le32 dqi_syncms;	/* Time after which we sync local changes to
+				 * global quota file */
+	__le32 dqi_blocks;	/* Number of blocks in quota file */
+/*10*/	__le32 dqi_free_blk;	/* First free block in quota file */
+	__le32 dqi_free_entry;	/* First block with free dquot entry in quota
+				 * file */
+};
+
+/* Structure with global user / group information. We reserve some space
+ * for future use. */
+struct ocfs2_global_disk_dqblk {
+/*00*/	__le32 dqb_id;          /* ID the structure belongs to */
+	__le32 dqb_use_count;   /* Number of nodes having reference to this structure */
+	__le64 dqb_ihardlimit;  /* absolute limit on allocated inodes */
+/*10*/	__le64 dqb_isoftlimit;  /* preferred inode limit */
+	__le64 dqb_curinodes;   /* current # allocated inodes */
+/*20*/	__le64 dqb_bhardlimit;  /* absolute limit on disk space */
+	__le64 dqb_bsoftlimit;  /* preferred limit on disk space */
+/*30*/	__le64 dqb_curspace;    /* current space occupied */
+	__le64 dqb_btime;       /* time limit for excessive disk use */
+/*40*/	__le64 dqb_itime;       /* time limit for excessive inode use */
+	__le64 dqb_pad1;
+/*50*/	__le64 dqb_pad2;
+};
+
+/*
+ *  On-disk structures for local quota file
+ */
+
+/* Magic numbers and known versions for local quota files */
+#define OCFS2_LOCAL_QMAGICS {\
+	0x0cf524c0, /* USRQUOTA */ \
+	0x0cf524c1  /* GRPQUOTA */ \
+}
+
+#define OCFS2_LOCAL_QVERSIONS {\
+	0, \
+	0, \
+}
+
+/* Quota flags in dqinfo header */
+#define OLQF_CLEAN	0x0001	/* Quota file is empty (this should be after\
+				 * quota has been cleanly turned off) */
+
+#define OCFS2_LOCAL_INFO_OFF (sizeof(struct ocfs2_disk_dqheader))
+
+/* Information header of local quota file (immediately follows the generic
+ * header) */
+struct ocfs2_local_disk_dqinfo {
+	__le32 dqi_flags;	/* Flags for quota file */
+	__le32 dqi_chunks;	/* Number of chunks of quota structures
+				 * with a bitmap */
+	__le32 dqi_blocks;	/* Number of blocks allocated for quota file */
+};
+
+/* Header of one chunk of a quota file */
+struct ocfs2_local_disk_chunk {
+	__le32 dqc_free;	/* Number of free entries in the bitmap */
+	__u8 dqc_bitmap[0];	/* Bitmap of entries in the corresponding
+				 * chunk of quota file */
+};
+
+/* One entry in local quota file */
+struct ocfs2_local_disk_dqblk {
+/*00*/	__le64 dqb_id;		/* id this quota applies to */
+	__le64 dqb_spacemod;	/* Change in the amount of used space */
+/*10*/	__le64 dqb_inodemod;	/* Change in the amount of used inodes */
+};
+
+
+/*
+ * The quota trailer lives at the end of each quota block.
+ */
+
+struct ocfs2_disk_dqtrailer {
+/*00*/	struct ocfs2_block_check dq_check;	/* Error checking */
+/*08*/	/* Cannot be larger than OCFS2_QBLK_RESERVED_SPACE */
+};
+
+static inline struct ocfs2_disk_dqtrailer *ocfs2_block_dqtrailer(int blocksize,
+								 void *buf)
+{
+	char *ptr = buf;
+	ptr += blocksize - OCFS2_QBLK_RESERVED_SPACE;
+
+	return (struct ocfs2_disk_dqtrailer *)ptr;
+}
+
+#ifdef __KERNEL__
+static inline int ocfs2_fast_symlink_chars(struct super_block *sb)
+{
+	return  sb->s_blocksize -
+		 offsetof(struct ocfs2_dinode, id2.i_symlink);
+}
+
+static inline int ocfs2_max_inline_data_with_xattr(struct super_block *sb,
+						   struct ocfs2_dinode *di)
+{
+	unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
+
+	if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
+		return sb->s_blocksize -
+			offsetof(struct ocfs2_dinode, id2.i_data.id_data) -
+			xattrsize;
+	else
+		return sb->s_blocksize -
+			offsetof(struct ocfs2_dinode, id2.i_data.id_data);
+}
+
+static inline int ocfs2_extent_recs_per_inode(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_dinode, id2.i_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline int ocfs2_extent_recs_per_inode_with_xattr(
+						struct super_block *sb,
+						struct ocfs2_dinode *di)
+{
+	int size;
+	unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
+
+	if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
+		size = sb->s_blocksize -
+			offsetof(struct ocfs2_dinode, id2.i_list.l_recs) -
+			xattrsize;
+	else
+		size = sb->s_blocksize -
+			offsetof(struct ocfs2_dinode, id2.i_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline int ocfs2_extent_recs_per_dx_root(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_dx_root_block, dr_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline int ocfs2_chain_recs_per_inode(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_dinode, id2.i_chain.cl_recs);
+
+	return size / sizeof(struct ocfs2_chain_rec);
+}
+
+static inline u16 ocfs2_extent_recs_per_eb(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_extent_block, h_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline u16 ocfs2_extent_recs_per_gd(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_group_desc, bg_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline int ocfs2_dx_entries_per_leaf(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_dx_leaf, dl_list.de_entries);
+
+	return size / sizeof(struct ocfs2_dx_entry);
+}
+
+static inline int ocfs2_dx_entries_per_root(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_dx_root_block, dr_entries.de_entries);
+
+	return size / sizeof(struct ocfs2_dx_entry);
+}
+
+static inline u16 ocfs2_local_alloc_size(struct super_block *sb)
+{
+	u16 size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_dinode, id2.i_lab.la_bitmap);
+
+	return size;
+}
+
+static inline int ocfs2_group_bitmap_size(struct super_block *sb,
+					  int suballocator,
+					  u32 feature_incompat)
+{
+	int size = sb->s_blocksize -
+		offsetof(struct ocfs2_group_desc, bg_bitmap);
+
+	/*
+	 * The cluster allocator uses the entire block.  Suballocators have
+	 * never used more than OCFS2_MAX_BG_BITMAP_SIZE.  Unfortunately, older
+	 * code expects bg_size set to the maximum.  Thus we must keep
+	 * bg_size as-is unless discontig_bg is enabled.
+	 */
+	if (suballocator &&
+	    (feature_incompat & OCFS2_FEATURE_INCOMPAT_DISCONTIG_BG))
+		size = OCFS2_MAX_BG_BITMAP_SIZE;
+
+	return size;
+}
+
+static inline int ocfs2_truncate_recs_per_inode(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_dinode, id2.i_dealloc.tl_recs);
+
+	return size / sizeof(struct ocfs2_truncate_rec);
+}
+
+static inline u64 ocfs2_backup_super_blkno(struct super_block *sb, int index)
+{
+	u64 offset = OCFS2_BACKUP_SB_START;
+
+	if (index >= 0 && index < OCFS2_MAX_BACKUP_SUPERBLOCKS) {
+		offset <<= (2 * index);
+		offset >>= sb->s_blocksize_bits;
+		return offset;
+	}
+
+	return 0;
+
+}
+
+static inline u16 ocfs2_xattr_recs_per_xb(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_xattr_block,
+			 xb_attrs.xb_root.xt_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline u16 ocfs2_extent_recs_per_rb(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_refcount_block, rf_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline u16 ocfs2_refcount_recs_per_rb(struct super_block *sb)
+{
+	int size;
+
+	size = sb->s_blocksize -
+		offsetof(struct ocfs2_refcount_block, rf_records.rl_recs);
+
+	return size / sizeof(struct ocfs2_refcount_rec);
+}
+
+static inline u32
+ocfs2_get_ref_rec_low_cpos(const struct ocfs2_refcount_rec *rec)
+{
+	return le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
+}
+#else
+static inline int ocfs2_fast_symlink_chars(int blocksize)
+{
+	return blocksize - offsetof(struct ocfs2_dinode, id2.i_symlink);
+}
+
+static inline int ocfs2_max_inline_data_with_xattr(int blocksize,
+						   struct ocfs2_dinode *di)
+{
+	if (di && (di->i_dyn_features & OCFS2_INLINE_XATTR_FL))
+		return blocksize -
+			offsetof(struct ocfs2_dinode, id2.i_data.id_data) -
+			di->i_xattr_inline_size;
+	else
+		return blocksize -
+			offsetof(struct ocfs2_dinode, id2.i_data.id_data);
+}
+
+static inline int ocfs2_extent_recs_per_inode(int blocksize)
+{
+	int size;
+
+	size = blocksize -
+		offsetof(struct ocfs2_dinode, id2.i_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline int ocfs2_chain_recs_per_inode(int blocksize)
+{
+	int size;
+
+	size = blocksize -
+		offsetof(struct ocfs2_dinode, id2.i_chain.cl_recs);
+
+	return size / sizeof(struct ocfs2_chain_rec);
+}
+
+static inline int ocfs2_extent_recs_per_eb(int blocksize)
+{
+	int size;
+
+	size = blocksize -
+		offsetof(struct ocfs2_extent_block, h_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline int ocfs2_extent_recs_per_gd(int blocksize)
+{
+	int size;
+
+	size = blocksize -
+		offsetof(struct ocfs2_group_desc, bg_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+
+static inline int ocfs2_local_alloc_size(int blocksize)
+{
+	int size;
+
+	size = blocksize -
+		offsetof(struct ocfs2_dinode, id2.i_lab.la_bitmap);
+
+	return size;
+}
+
+static inline int ocfs2_group_bitmap_size(int blocksize,
+					  int suballocator,
+					  uint32_t feature_incompat)
+{
+	int size = sb->s_blocksize -
+		offsetof(struct ocfs2_group_desc, bg_bitmap);
+
+	/*
+	 * The cluster allocator uses the entire block.  Suballocators have
+	 * never used more than OCFS2_MAX_BG_BITMAP_SIZE.  Unfortunately, older
+	 * code expects bg_size set to the maximum.  Thus we must keep
+	 * bg_size as-is unless discontig_bg is enabled.
+	 */
+	if (suballocator &&
+	    (feature_incompat & OCFS2_FEATURE_INCOMPAT_DISCONTIG_BG))
+		size = OCFS2_MAX_BG_BITMAP_SIZE;
+
+	return size;
+}
+
+static inline int ocfs2_truncate_recs_per_inode(int blocksize)
+{
+	int size;
+
+	size = blocksize -
+		offsetof(struct ocfs2_dinode, id2.i_dealloc.tl_recs);
+
+	return size / sizeof(struct ocfs2_truncate_rec);
+}
+
+static inline uint64_t ocfs2_backup_super_blkno(int blocksize, int index)
+{
+	uint64_t offset = OCFS2_BACKUP_SB_START;
+
+	if (index >= 0 && index < OCFS2_MAX_BACKUP_SUPERBLOCKS) {
+		offset <<= (2 * index);
+		offset /= blocksize;
+		return offset;
+	}
+
+	return 0;
+}
+
+static inline int ocfs2_xattr_recs_per_xb(int blocksize)
+{
+	int size;
+
+	size = blocksize -
+		offsetof(struct ocfs2_xattr_block,
+			 xb_attrs.xb_root.xt_list.l_recs);
+
+	return size / sizeof(struct ocfs2_extent_rec);
+}
+#endif  /* __KERNEL__ */
+
+
+static inline int ocfs2_system_inode_is_global(int type)
+{
+	return ((type >= 0) &&
+		(type <= OCFS2_LAST_GLOBAL_SYSTEM_INODE));
+}
+
+static inline int ocfs2_sprintf_system_inode_name(char *buf, int len,
+						  int type, int slot)
+{
+	int chars;
+
+        /*
+         * Global system inodes can only have one copy.  Everything
+         * after OCFS2_LAST_GLOBAL_SYSTEM_INODE in the system inode
+         * list has a copy per slot.
+         */
+	if (type <= OCFS2_LAST_GLOBAL_SYSTEM_INODE)
+		chars = snprintf(buf, len, "%s",
+				 ocfs2_system_inodes[type].si_name);
+	else
+		chars = snprintf(buf, len,
+				 ocfs2_system_inodes[type].si_name,
+				 slot);
+
+	return chars;
+}
+
+static inline void ocfs2_set_de_type(struct ocfs2_dir_entry *de,
+				    umode_t mode)
+{
+	de->file_type = ocfs2_type_by_mode[(mode & S_IFMT)>>S_SHIFT];
+}
+
+static inline int ocfs2_gd_is_discontig(struct ocfs2_group_desc *gd)
+{
+	if ((offsetof(struct ocfs2_group_desc, bg_bitmap) +
+	     le16_to_cpu(gd->bg_size)) !=
+	    offsetof(struct ocfs2_group_desc, bg_list))
+		return 0;
+	/*
+	 * Only valid to check l_next_free_rec if
+	 * bg_bitmap + bg_size == bg_list.
+	 */
+	if (!gd->bg_list.l_next_free_rec)
+		return 0;
+	return 1;
+}
+#endif  /* _OCFS2_FS_H */
+
diff --git a/fs/ocfs2/ocfs2_ioctl.h b/fs/ocfs2/ocfs2_ioctl.h
new file mode 100644
index 000000000..5b27ff1fa
--- /dev/null
+++ b/fs/ocfs2/ocfs2_ioctl.h
@@ -0,0 +1,242 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ocfs2_ioctl.h
+ *
+ * Defines OCFS2 ioctls.
+ *
+ * Copyright (C) 2010 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License, version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#ifndef OCFS2_IOCTL_H
+#define OCFS2_IOCTL_H
+
+/*
+ * ioctl commands
+ */
+#define OCFS2_IOC_GETFLAGS	FS_IOC_GETFLAGS
+#define OCFS2_IOC_SETFLAGS	FS_IOC_SETFLAGS
+#define OCFS2_IOC32_GETFLAGS	FS_IOC32_GETFLAGS
+#define OCFS2_IOC32_SETFLAGS	FS_IOC32_SETFLAGS
+
+/*
+ * Space reservation / allocation / free ioctls and argument structure
+ * are designed to be compatible with XFS.
+ *
+ * ALLOCSP* and FREESP* are not and will never be supported, but are
+ * included here for completeness.
+ */
+struct ocfs2_space_resv {
+	__s16		l_type;
+	__s16		l_whence;
+	__s64		l_start;
+	__s64		l_len;		/* len == 0 means until end of file */
+	__s32		l_sysid;
+	__u32		l_pid;
+	__s32		l_pad[4];	/* reserve area			    */
+};
+
+#define OCFS2_IOC_ALLOCSP		_IOW ('X', 10, struct ocfs2_space_resv)
+#define OCFS2_IOC_FREESP		_IOW ('X', 11, struct ocfs2_space_resv)
+#define OCFS2_IOC_RESVSP		_IOW ('X', 40, struct ocfs2_space_resv)
+#define OCFS2_IOC_UNRESVSP	_IOW ('X', 41, struct ocfs2_space_resv)
+#define OCFS2_IOC_ALLOCSP64	_IOW ('X', 36, struct ocfs2_space_resv)
+#define OCFS2_IOC_FREESP64	_IOW ('X', 37, struct ocfs2_space_resv)
+#define OCFS2_IOC_RESVSP64	_IOW ('X', 42, struct ocfs2_space_resv)
+#define OCFS2_IOC_UNRESVSP64	_IOW ('X', 43, struct ocfs2_space_resv)
+
+/* Used to pass group descriptor data when online resize is done */
+struct ocfs2_new_group_input {
+	__u64 group;		/* Group descriptor's blkno. */
+	__u32 clusters;		/* Total number of clusters in this group */
+	__u32 frees;		/* Total free clusters in this group */
+	__u16 chain;		/* Chain for this group */
+	__u16 reserved1;
+	__u32 reserved2;
+};
+
+#define OCFS2_IOC_GROUP_EXTEND	_IOW('o', 1, int)
+#define OCFS2_IOC_GROUP_ADD	_IOW('o', 2,struct ocfs2_new_group_input)
+#define OCFS2_IOC_GROUP_ADD64	_IOW('o', 3,struct ocfs2_new_group_input)
+
+/* Used to pass 2 file names to reflink. */
+struct reflink_arguments {
+	__u64 old_path;
+	__u64 new_path;
+	__u64 preserve;
+};
+#define OCFS2_IOC_REFLINK	_IOW('o', 4, struct reflink_arguments)
+
+/* Following definitions dedicated for ocfs2_info_request ioctls. */
+#define OCFS2_INFO_MAX_REQUEST		(50)
+#define OCFS2_TEXT_UUID_LEN		(OCFS2_VOL_UUID_LEN * 2)
+
+/* Magic number of all requests */
+#define OCFS2_INFO_MAGIC		(0x4F32494E)
+
+/*
+ * Always try to separate info request into small pieces to
+ * guarantee the backward&forward compatibility.
+ */
+struct ocfs2_info {
+	__u64 oi_requests;	/* Array of __u64 pointers to requests */
+	__u32 oi_count;		/* Number of requests in info_requests */
+	__u32 oi_pad;
+};
+
+struct ocfs2_info_request {
+/*00*/	__u32 ir_magic;	/* Magic number */
+	__u32 ir_code;	/* Info request code */
+	__u32 ir_size;	/* Size of request */
+	__u32 ir_flags;	/* Request flags */
+/*10*/			/* Request specific fields */
+};
+
+struct ocfs2_info_clustersize {
+	struct ocfs2_info_request ic_req;
+	__u32 ic_clustersize;
+	__u32 ic_pad;
+};
+
+struct ocfs2_info_blocksize {
+	struct ocfs2_info_request ib_req;
+	__u32 ib_blocksize;
+	__u32 ib_pad;
+};
+
+struct ocfs2_info_maxslots {
+	struct ocfs2_info_request im_req;
+	__u32 im_max_slots;
+	__u32 im_pad;
+};
+
+struct ocfs2_info_label {
+	struct ocfs2_info_request il_req;
+	__u8	il_label[OCFS2_MAX_VOL_LABEL_LEN];
+} __attribute__ ((packed));
+
+struct ocfs2_info_uuid {
+	struct ocfs2_info_request iu_req;
+	__u8	iu_uuid_str[OCFS2_TEXT_UUID_LEN + 1];
+} __attribute__ ((packed));
+
+struct ocfs2_info_fs_features {
+	struct ocfs2_info_request if_req;
+	__u32 if_compat_features;
+	__u32 if_incompat_features;
+	__u32 if_ro_compat_features;
+	__u32 if_pad;
+};
+
+struct ocfs2_info_journal_size {
+	struct ocfs2_info_request ij_req;
+	__u64 ij_journal_size;
+};
+
+struct ocfs2_info_freeinode {
+	struct ocfs2_info_request ifi_req;
+	struct ocfs2_info_local_freeinode {
+		__u64 lfi_total;
+		__u64 lfi_free;
+	} ifi_stat[OCFS2_MAX_SLOTS];
+	__u32 ifi_slotnum; /* out */
+	__u32 ifi_pad;
+};
+
+#define OCFS2_INFO_MAX_HIST     (32)
+
+struct ocfs2_info_freefrag {
+	struct ocfs2_info_request iff_req;
+	struct ocfs2_info_freefrag_stats { /* (out) */
+		struct ocfs2_info_free_chunk_list {
+			__u32 fc_chunks[OCFS2_INFO_MAX_HIST];
+			__u32 fc_clusters[OCFS2_INFO_MAX_HIST];
+		} ffs_fc_hist;
+		__u32 ffs_clusters;
+		__u32 ffs_free_clusters;
+		__u32 ffs_free_chunks;
+		__u32 ffs_free_chunks_real;
+		__u32 ffs_min; /* Minimum free chunksize in clusters */
+		__u32 ffs_max;
+		__u32 ffs_avg;
+		__u32 ffs_pad;
+	} iff_ffs;
+	__u32 iff_chunksize; /* chunksize in clusters(in) */
+	__u32 iff_pad;
+};
+
+/* Codes for ocfs2_info_request */
+enum ocfs2_info_type {
+	OCFS2_INFO_CLUSTERSIZE = 1,
+	OCFS2_INFO_BLOCKSIZE,
+	OCFS2_INFO_MAXSLOTS,
+	OCFS2_INFO_LABEL,
+	OCFS2_INFO_UUID,
+	OCFS2_INFO_FS_FEATURES,
+	OCFS2_INFO_JOURNAL_SIZE,
+	OCFS2_INFO_FREEINODE,
+	OCFS2_INFO_FREEFRAG,
+	OCFS2_INFO_NUM_TYPES
+};
+
+/* Flags for struct ocfs2_info_request */
+/* Filled by the caller */
+#define OCFS2_INFO_FL_NON_COHERENT	(0x00000001)	/* Cluster coherency not
+							   required. This is a hint.
+							   It is up to ocfs2 whether
+							   the request can be fulfilled
+							   without locking. */
+/* Filled by ocfs2 */
+#define OCFS2_INFO_FL_FILLED		(0x40000000)	/* Filesystem understood
+							   this request and
+							   filled in the answer */
+
+#define OCFS2_INFO_FL_ERROR		(0x80000000)	/* Error happened during
+							   request handling. */
+
+#define OCFS2_IOC_INFO		_IOR('o', 5, struct ocfs2_info)
+
+struct ocfs2_move_extents {
+/* All values are in bytes */
+	/* in */
+	__u64 me_start;		/* Virtual start in the file to move */
+	__u64 me_len;		/* Length of the extents to be moved */
+	__u64 me_goal;		/* Physical offset of the goal,
+				   it's in block unit */
+	__u64 me_threshold;	/* Maximum distance from goal or threshold
+				   for auto defragmentation */
+	__u64 me_flags;		/* Flags for the operation:
+				 * - auto defragmentation.
+				 * - refcount,xattr cases.
+				 */
+	/* out */
+	__u64 me_moved_len;	/* Moved/defraged length */
+	__u64 me_new_offset;	/* Resulting physical location */
+	__u32 me_reserved[2];	/* Reserved for futhure */
+};
+
+#define OCFS2_MOVE_EXT_FL_AUTO_DEFRAG	(0x00000001)	/* Kernel manages to
+							   claim new clusters
+							   as the goal place
+							   for extents moving */
+#define OCFS2_MOVE_EXT_FL_PART_DEFRAG	(0x00000002)	/* Allow partial extent
+							   moving, is to make
+							   movement less likely
+							   to fail, may make fs
+							   even more fragmented */
+#define OCFS2_MOVE_EXT_FL_COMPLETE	(0x00000004)	/* Move or defragmenation
+							   completely gets done.
+							 */
+
+#define OCFS2_IOC_MOVE_EXT	_IOW('o', 6, struct ocfs2_move_extents)
+
+#endif /* OCFS2_IOCTL_H */
diff --git a/fs/ocfs2/ocfs2_lockid.h b/fs/ocfs2/ocfs2_lockid.h
new file mode 100644
index 000000000..d277aabf5
--- /dev/null
+++ b/fs/ocfs2/ocfs2_lockid.h
@@ -0,0 +1,128 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ocfs2_lockid.h
+ *
+ * Defines OCFS2 lockid bits.
+ *
+ * Copyright (C) 2002, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_LOCKID_H
+#define OCFS2_LOCKID_H
+
+/* lock ids are made up in the following manner:
+ * name[0]     --> type
+ * name[1-6]   --> 6 pad characters, reserved for now
+ * name[7-22]  --> block number, expressed in hex as 16 chars
+ * name[23-30] --> i_generation, expressed in hex 8 chars
+ * name[31]    --> '\0' */
+#define OCFS2_LOCK_ID_MAX_LEN  32
+#define OCFS2_LOCK_ID_PAD "000000"
+
+#define OCFS2_DENTRY_LOCK_INO_START 18
+
+enum ocfs2_lock_type {
+	OCFS2_LOCK_TYPE_META = 0,
+	OCFS2_LOCK_TYPE_DATA,
+	OCFS2_LOCK_TYPE_SUPER,
+	OCFS2_LOCK_TYPE_RENAME,
+	OCFS2_LOCK_TYPE_RW,
+	OCFS2_LOCK_TYPE_DENTRY,
+	OCFS2_LOCK_TYPE_OPEN,
+	OCFS2_LOCK_TYPE_FLOCK,
+	OCFS2_LOCK_TYPE_QINFO,
+	OCFS2_LOCK_TYPE_NFS_SYNC,
+	OCFS2_LOCK_TYPE_ORPHAN_SCAN,
+	OCFS2_LOCK_TYPE_REFCOUNT,
+	OCFS2_NUM_LOCK_TYPES
+};
+
+static inline char ocfs2_lock_type_char(enum ocfs2_lock_type type)
+{
+	char c;
+	switch (type) {
+		case OCFS2_LOCK_TYPE_META:
+			c = 'M';
+			break;
+		case OCFS2_LOCK_TYPE_DATA:
+			c = 'D';
+			break;
+		case OCFS2_LOCK_TYPE_SUPER:
+			c = 'S';
+			break;
+		case OCFS2_LOCK_TYPE_RENAME:
+			c = 'R';
+			break;
+		case OCFS2_LOCK_TYPE_RW:
+			c = 'W';
+			break;
+		case OCFS2_LOCK_TYPE_DENTRY:
+			c = 'N';
+			break;
+		case OCFS2_LOCK_TYPE_OPEN:
+			c = 'O';
+			break;
+		case OCFS2_LOCK_TYPE_FLOCK:
+			c = 'F';
+			break;
+		case OCFS2_LOCK_TYPE_QINFO:
+			c = 'Q';
+			break;
+		case OCFS2_LOCK_TYPE_NFS_SYNC:
+			c = 'Y';
+			break;
+		case OCFS2_LOCK_TYPE_ORPHAN_SCAN:
+			c = 'P';
+			break;
+		case OCFS2_LOCK_TYPE_REFCOUNT:
+			c = 'T';
+			break;
+		default:
+			c = '\0';
+	}
+
+	return c;
+}
+
+static char *ocfs2_lock_type_strings[] = {
+	[OCFS2_LOCK_TYPE_META] = "Meta",
+	[OCFS2_LOCK_TYPE_DATA] = "Data",
+	[OCFS2_LOCK_TYPE_SUPER] = "Super",
+	[OCFS2_LOCK_TYPE_RENAME] = "Rename",
+	/* Need to differntiate from [R]ename.. serializing writes is the
+	 * important job it does, anyway. */
+	[OCFS2_LOCK_TYPE_RW] = "Write/Read",
+	[OCFS2_LOCK_TYPE_DENTRY] = "Dentry",
+	[OCFS2_LOCK_TYPE_OPEN] = "Open",
+	[OCFS2_LOCK_TYPE_FLOCK] = "Flock",
+	[OCFS2_LOCK_TYPE_QINFO] = "Quota",
+	[OCFS2_LOCK_TYPE_NFS_SYNC] = "NFSSync",
+	[OCFS2_LOCK_TYPE_ORPHAN_SCAN] = "OrphanScan",
+	[OCFS2_LOCK_TYPE_REFCOUNT] = "Refcount",
+};
+
+static inline const char *ocfs2_lock_type_string(enum ocfs2_lock_type type)
+{
+#ifdef __KERNEL__
+	BUG_ON(type >= OCFS2_NUM_LOCK_TYPES);
+#endif
+	return ocfs2_lock_type_strings[type];
+}
+
+#endif  /* OCFS2_LOCKID_H */
diff --git a/fs/ocfs2/ocfs2_lockingver.h b/fs/ocfs2/ocfs2_lockingver.h
new file mode 100644
index 000000000..2e45c8d2e
--- /dev/null
+++ b/fs/ocfs2/ocfs2_lockingver.h
@@ -0,0 +1,32 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * ocfs2_lockingver.h
+ *
+ * Defines OCFS2 Locking version values.
+ *
+ * Copyright (C) 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License, version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#ifndef OCFS2_LOCKINGVER_H
+#define OCFS2_LOCKINGVER_H
+
+/*
+ * The protocol version for ocfs2 cluster locking.  See dlmglue.c for
+ * more details.
+ *
+ * 1.0 - Initial locking version from ocfs2 1.4.
+ */
+#define OCFS2_LOCKING_PROTOCOL_MAJOR 1
+#define OCFS2_LOCKING_PROTOCOL_MINOR 0
+
+#endif  /* OCFS2_LOCKINGVER_H */
diff --git a/fs/ocfs2/ocfs2_trace.h b/fs/ocfs2/ocfs2_trace.h
new file mode 100644
index 000000000..6cb019b7c
--- /dev/null
+++ b/fs/ocfs2/ocfs2_trace.h
@@ -0,0 +1,2768 @@
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM ocfs2
+
+#if !defined(_TRACE_OCFS2_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_OCFS2_H
+
+#include <linux/tracepoint.h>
+
+DECLARE_EVENT_CLASS(ocfs2__int,
+	TP_PROTO(int num),
+	TP_ARGS(num),
+	TP_STRUCT__entry(
+		__field(int, num)
+	),
+	TP_fast_assign(
+		__entry->num = num;
+	),
+	TP_printk("%d", __entry->num)
+);
+
+#define DEFINE_OCFS2_INT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__int, name,	\
+	TP_PROTO(int num),	\
+	TP_ARGS(num))
+
+DECLARE_EVENT_CLASS(ocfs2__uint,
+	TP_PROTO(unsigned int num),
+	TP_ARGS(num),
+	TP_STRUCT__entry(
+		__field(	unsigned int,	num		)
+	),
+	TP_fast_assign(
+		__entry->num	= 	num;
+	),
+	TP_printk("%u", __entry->num)
+);
+
+#define DEFINE_OCFS2_UINT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__uint, name,	\
+	TP_PROTO(unsigned int num),	\
+	TP_ARGS(num))
+
+DECLARE_EVENT_CLASS(ocfs2__ull,
+	TP_PROTO(unsigned long long blkno),
+	TP_ARGS(blkno),
+	TP_STRUCT__entry(
+		__field(unsigned long long, blkno)
+	),
+	TP_fast_assign(
+		__entry->blkno = blkno;
+	),
+	TP_printk("%llu", __entry->blkno)
+);
+
+#define DEFINE_OCFS2_ULL_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull, name,	\
+	TP_PROTO(unsigned long long num),	\
+	TP_ARGS(num))
+
+DECLARE_EVENT_CLASS(ocfs2__pointer,
+	TP_PROTO(void *pointer),
+	TP_ARGS(pointer),
+	TP_STRUCT__entry(
+		__field(void *, pointer)
+	),
+	TP_fast_assign(
+		__entry->pointer = pointer;
+	),
+	TP_printk("%p", __entry->pointer)
+);
+
+#define DEFINE_OCFS2_POINTER_EVENT(name)	\
+DEFINE_EVENT(ocfs2__pointer, name,	\
+	TP_PROTO(void *pointer),	\
+	TP_ARGS(pointer))
+
+DECLARE_EVENT_CLASS(ocfs2__string,
+	TP_PROTO(const char *name),
+	TP_ARGS(name),
+	TP_STRUCT__entry(
+		__string(name,name)
+	),
+	TP_fast_assign(
+		__assign_str(name, name);
+	),
+	TP_printk("%s", __get_str(name))
+);
+
+#define DEFINE_OCFS2_STRING_EVENT(name)	\
+DEFINE_EVENT(ocfs2__string, name,	\
+	TP_PROTO(const char *name),	\
+	TP_ARGS(name))
+
+DECLARE_EVENT_CLASS(ocfs2__int_int,
+	TP_PROTO(int value1, int value2),
+	TP_ARGS(value1, value2),
+	TP_STRUCT__entry(
+		__field(int, value1)
+		__field(int, value2)
+	),
+	TP_fast_assign(
+		__entry->value1	= value1;
+		__entry->value2	= value2;
+	),
+	TP_printk("%d %d", __entry->value1, __entry->value2)
+);
+
+#define DEFINE_OCFS2_INT_INT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__int_int, name,	\
+	TP_PROTO(int val1, int val2),	\
+	TP_ARGS(val1, val2))
+
+DECLARE_EVENT_CLASS(ocfs2__uint_int,
+	TP_PROTO(unsigned int value1, int value2),
+	TP_ARGS(value1, value2),
+	TP_STRUCT__entry(
+		__field(unsigned int, value1)
+		__field(int, value2)
+	),
+	TP_fast_assign(
+		__entry->value1	= value1;
+		__entry->value2	= value2;
+	),
+	TP_printk("%u %d", __entry->value1, __entry->value2)
+);
+
+#define DEFINE_OCFS2_UINT_INT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__uint_int, name,	\
+	TP_PROTO(unsigned int val1, int val2),	\
+	TP_ARGS(val1, val2))
+
+DECLARE_EVENT_CLASS(ocfs2__uint_uint,
+	TP_PROTO(unsigned int value1, unsigned int value2),
+	TP_ARGS(value1, value2),
+	TP_STRUCT__entry(
+		__field(unsigned int, value1)
+		__field(unsigned int, value2)
+	),
+	TP_fast_assign(
+		__entry->value1 = value1;
+		__entry->value2 = value2;
+	),
+	TP_printk("%u %u", __entry->value1, __entry->value2)
+);
+
+#define DEFINE_OCFS2_UINT_UINT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__uint_uint, name,	\
+	TP_PROTO(unsigned int val1, unsigned int val2),	\
+	TP_ARGS(val1, val2))
+
+DECLARE_EVENT_CLASS(ocfs2__ull_uint,
+	TP_PROTO(unsigned long long value1, unsigned int value2),
+	TP_ARGS(value1, value2),
+	TP_STRUCT__entry(
+		__field(unsigned long long, value1)
+		__field(unsigned int, value2)
+	),
+	TP_fast_assign(
+		__entry->value1 = value1;
+		__entry->value2 = value2;
+	),
+	TP_printk("%llu %u", __entry->value1, __entry->value2)
+);
+
+#define DEFINE_OCFS2_ULL_UINT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull_uint, name,	\
+	TP_PROTO(unsigned long long val1, unsigned int val2),	\
+	TP_ARGS(val1, val2))
+
+DECLARE_EVENT_CLASS(ocfs2__ull_int,
+	TP_PROTO(unsigned long long value1, int value2),
+	TP_ARGS(value1, value2),
+	TP_STRUCT__entry(
+		__field(unsigned long long, value1)
+		__field(int, value2)
+	),
+	TP_fast_assign(
+		__entry->value1	= value1;
+		__entry->value2	= value2;
+	),
+	TP_printk("%llu %d", __entry->value1, __entry->value2)
+);
+
+#define DEFINE_OCFS2_ULL_INT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull_int, name,	\
+	TP_PROTO(unsigned long long val1, int val2),	\
+	TP_ARGS(val1, val2))
+
+DECLARE_EVENT_CLASS(ocfs2__ull_ull,
+	TP_PROTO(unsigned long long value1, unsigned long long value2),
+	TP_ARGS(value1, value2),
+	TP_STRUCT__entry(
+		__field(unsigned long long, value1)
+		__field(unsigned long long, value2)
+	),
+	TP_fast_assign(
+		__entry->value1 = value1;
+		__entry->value2 = value2;
+	),
+	TP_printk("%llu %llu", __entry->value1, __entry->value2)
+);
+
+#define DEFINE_OCFS2_ULL_ULL_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull_ull, name,	\
+	TP_PROTO(unsigned long long val1, unsigned long long val2),	\
+	TP_ARGS(val1, val2))
+
+DECLARE_EVENT_CLASS(ocfs2__ull_ull_uint,
+	TP_PROTO(unsigned long long value1,
+		 unsigned long long value2, unsigned int value3),
+	TP_ARGS(value1, value2, value3),
+	TP_STRUCT__entry(
+		__field(unsigned long long, value1)
+		__field(unsigned long long, value2)
+		__field(unsigned int, value3)
+	),
+	TP_fast_assign(
+		__entry->value1 = value1;
+		__entry->value2 = value2;
+		__entry->value3 = value3;
+	),
+	TP_printk("%llu %llu %u",
+		  __entry->value1, __entry->value2, __entry->value3)
+);
+
+#define DEFINE_OCFS2_ULL_ULL_UINT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull_ull_uint, name,	\
+	TP_PROTO(unsigned long long val1,	\
+		 unsigned long long val2, unsigned int val3),	\
+	TP_ARGS(val1, val2, val3))
+
+DECLARE_EVENT_CLASS(ocfs2__ull_uint_uint,
+	TP_PROTO(unsigned long long value1,
+		 unsigned int value2, unsigned int value3),
+	TP_ARGS(value1, value2, value3),
+	TP_STRUCT__entry(
+		__field(unsigned long long, value1)
+		__field(unsigned int, value2)
+		__field(unsigned int, value3)
+	),
+	TP_fast_assign(
+		__entry->value1 = value1;
+		__entry->value2 = value2;
+		__entry->value3	= value3;
+	),
+	TP_printk("%llu %u %u", __entry->value1,
+		  __entry->value2, __entry->value3)
+);
+
+#define DEFINE_OCFS2_ULL_UINT_UINT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull_uint_uint, name,	\
+	TP_PROTO(unsigned long long val1,	\
+		 unsigned int val2, unsigned int val3),	\
+	TP_ARGS(val1, val2, val3))
+
+DECLARE_EVENT_CLASS(ocfs2__uint_uint_uint,
+	TP_PROTO(unsigned int value1, unsigned int value2,
+		 unsigned int value3),
+	TP_ARGS(value1, value2, value3),
+	TP_STRUCT__entry(
+		__field(	unsigned int,	value1		)
+		__field(	unsigned int,	value2		)
+		__field(	unsigned int,	value3		)
+	),
+	TP_fast_assign(
+		__entry->value1	= 	value1;
+		__entry->value2	= 	value2;
+		__entry->value3	= 	value3;
+	),
+	TP_printk("%u %u %u", __entry->value1, __entry->value2, __entry->value3)
+);
+
+#define DEFINE_OCFS2_UINT_UINT_UINT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__uint_uint_uint, name,	\
+	TP_PROTO(unsigned int value1, unsigned int value2,	\
+		 unsigned int value3),	\
+	TP_ARGS(value1, value2, value3))
+
+DECLARE_EVENT_CLASS(ocfs2__ull_ull_ull,
+	TP_PROTO(unsigned long long value1,
+		 unsigned long long value2, unsigned long long value3),
+	TP_ARGS(value1, value2, value3),
+	TP_STRUCT__entry(
+		__field(unsigned long long, value1)
+		__field(unsigned long long, value2)
+		__field(unsigned long long, value3)
+	),
+	TP_fast_assign(
+		__entry->value1 = value1;
+		__entry->value2 = value2;
+		__entry->value3 = value3;
+	),
+	TP_printk("%llu %llu %llu",
+		  __entry->value1, __entry->value2, __entry->value3)
+);
+
+#define DEFINE_OCFS2_ULL_ULL_ULL_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull_ull_ull, name,	\
+	TP_PROTO(unsigned long long value1, unsigned long long value2,	\
+		 unsigned long long value3),	\
+	TP_ARGS(value1, value2, value3))
+
+DECLARE_EVENT_CLASS(ocfs2__ull_int_int_int,
+	TP_PROTO(unsigned long long ull, int value1, int value2, int value3),
+	TP_ARGS(ull, value1, value2, value3),
+	TP_STRUCT__entry(
+		__field(	unsigned long long,	ull	)
+		__field(	int,	value1			)
+		__field(	int,	value2			)
+		__field(	int,	value3			)
+	),
+	TP_fast_assign(
+		__entry->ull		= ull;
+		__entry->value1		= value1;
+		__entry->value2		= value2;
+		__entry->value3		= value3;
+	),
+	TP_printk("%llu %d %d %d",
+		  __entry->ull, __entry->value1,
+		  __entry->value2, __entry->value3)
+);
+
+#define DEFINE_OCFS2_ULL_INT_INT_INT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull_int_int_int, name,	\
+	TP_PROTO(unsigned long long ull, int value1,	\
+		 int value2, int value3),	\
+	TP_ARGS(ull, value1, value2, value3))
+
+DECLARE_EVENT_CLASS(ocfs2__ull_uint_uint_uint,
+	TP_PROTO(unsigned long long ull, unsigned int value1,
+		 unsigned int value2, unsigned int value3),
+	TP_ARGS(ull, value1, value2, value3),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ull)
+		__field(unsigned int, value1)
+		__field(unsigned int, value2)
+		__field(unsigned int, value3)
+	),
+	TP_fast_assign(
+		__entry->ull = ull;
+		__entry->value1 = value1;
+		__entry->value2	= value2;
+		__entry->value3	= value3;
+	),
+	TP_printk("%llu %u %u %u",
+		  __entry->ull, __entry->value1,
+		  __entry->value2, __entry->value3)
+);
+
+#define DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull_uint_uint_uint, name,	\
+	TP_PROTO(unsigned long long ull, unsigned int value1,	\
+		 unsigned int value2, unsigned int value3),	\
+	TP_ARGS(ull, value1, value2, value3))
+
+DECLARE_EVENT_CLASS(ocfs2__ull_ull_uint_uint,
+	TP_PROTO(unsigned long long value1, unsigned long long value2,
+		 unsigned int value3, unsigned int value4),
+	TP_ARGS(value1, value2, value3, value4),
+	TP_STRUCT__entry(
+		__field(unsigned long long, value1)
+		__field(unsigned long long, value2)
+		__field(unsigned int, value3)
+		__field(unsigned int, value4)
+	),
+	TP_fast_assign(
+		__entry->value1 = value1;
+		__entry->value2 = value2;
+		__entry->value3 = value3;
+		__entry->value4 = value4;
+	),
+	TP_printk("%llu %llu %u %u",
+		  __entry->value1, __entry->value2,
+		  __entry->value3, __entry->value4)
+);
+
+#define DEFINE_OCFS2_ULL_ULL_UINT_UINT_EVENT(name)	\
+DEFINE_EVENT(ocfs2__ull_ull_uint_uint, name,	\
+	TP_PROTO(unsigned long long ull, unsigned long long ull1,	\
+		 unsigned int value2, unsigned int value3),	\
+	TP_ARGS(ull, ull1, value2, value3))
+
+/* Trace events for fs/ocfs2/alloc.c. */
+DECLARE_EVENT_CLASS(ocfs2__btree_ops,
+	TP_PROTO(unsigned long long owner,\
+		 unsigned int value1, unsigned int value2),
+	TP_ARGS(owner, value1, value2),
+	TP_STRUCT__entry(
+		__field(unsigned long long, owner)
+		__field(unsigned int, value1)
+		__field(unsigned int, value2)
+	),
+	TP_fast_assign(
+		__entry->owner = owner;
+		__entry->value1 = value1;
+		__entry->value2	= value2;
+	),
+	TP_printk("%llu %u %u",
+		  __entry->owner, __entry->value1, __entry->value2)
+);
+
+#define DEFINE_OCFS2_BTREE_EVENT(name)	\
+DEFINE_EVENT(ocfs2__btree_ops, name,	\
+	TP_PROTO(unsigned long long owner,	\
+		 unsigned int value1, unsigned int value2),	\
+	TP_ARGS(owner, value1, value2))
+
+DEFINE_OCFS2_BTREE_EVENT(ocfs2_adjust_rightmost_branch);
+
+DEFINE_OCFS2_BTREE_EVENT(ocfs2_rotate_tree_right);
+
+DEFINE_OCFS2_BTREE_EVENT(ocfs2_append_rec_to_path);
+
+DEFINE_OCFS2_BTREE_EVENT(ocfs2_insert_extent_start);
+
+DEFINE_OCFS2_BTREE_EVENT(ocfs2_add_clusters_in_btree);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_num_free_extents);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_complete_edge_insert);
+
+TRACE_EVENT(ocfs2_grow_tree,
+	TP_PROTO(unsigned long long owner, int depth),
+	TP_ARGS(owner, depth),
+	TP_STRUCT__entry(
+		__field(unsigned long long, owner)
+		__field(int, depth)
+	),
+	TP_fast_assign(
+		__entry->owner = owner;
+		__entry->depth = depth;
+	),
+	TP_printk("%llu %d", __entry->owner, __entry->depth)
+);
+
+TRACE_EVENT(ocfs2_rotate_subtree,
+	TP_PROTO(int subtree_root, unsigned long long blkno,
+		 int depth),
+	TP_ARGS(subtree_root, blkno, depth),
+	TP_STRUCT__entry(
+		__field(int, subtree_root)
+		__field(unsigned long long, blkno)
+		__field(int, depth)
+	),
+	TP_fast_assign(
+		__entry->subtree_root = subtree_root;
+		__entry->blkno = blkno;
+		__entry->depth = depth;
+	),
+	TP_printk("%d %llu %d", __entry->subtree_root,
+		  __entry->blkno, __entry->depth)
+);
+
+TRACE_EVENT(ocfs2_insert_extent,
+	TP_PROTO(unsigned int ins_appending, unsigned int ins_contig,
+		 int ins_contig_index, int free_records, int ins_tree_depth),
+	TP_ARGS(ins_appending, ins_contig, ins_contig_index, free_records,
+		ins_tree_depth),
+	TP_STRUCT__entry(
+		__field(unsigned int, ins_appending)
+		__field(unsigned int, ins_contig)
+		__field(int, ins_contig_index)
+		__field(int, free_records)
+		__field(int, ins_tree_depth)
+	),
+	TP_fast_assign(
+		__entry->ins_appending = ins_appending;
+		__entry->ins_contig = ins_contig;
+		__entry->ins_contig_index = ins_contig_index;
+		__entry->free_records = free_records;
+		__entry->ins_tree_depth = ins_tree_depth;
+	),
+	TP_printk("%u %u %d %d %d",
+		  __entry->ins_appending, __entry->ins_contig,
+		  __entry->ins_contig_index, __entry->free_records,
+		  __entry->ins_tree_depth)
+);
+
+TRACE_EVENT(ocfs2_split_extent,
+	TP_PROTO(int split_index, unsigned int c_contig_type,
+		 unsigned int c_has_empty_extent,
+		 unsigned int c_split_covers_rec),
+	TP_ARGS(split_index, c_contig_type,
+		c_has_empty_extent, c_split_covers_rec),
+	TP_STRUCT__entry(
+		__field(int, split_index)
+		__field(unsigned int, c_contig_type)
+		__field(unsigned int, c_has_empty_extent)
+		__field(unsigned int, c_split_covers_rec)
+	),
+	TP_fast_assign(
+		__entry->split_index = split_index;
+		__entry->c_contig_type = c_contig_type;
+		__entry->c_has_empty_extent = c_has_empty_extent;
+		__entry->c_split_covers_rec = c_split_covers_rec;
+	),
+	TP_printk("%d %u %u %u", __entry->split_index, __entry->c_contig_type,
+		  __entry->c_has_empty_extent, __entry->c_split_covers_rec)
+);
+
+TRACE_EVENT(ocfs2_remove_extent,
+	TP_PROTO(unsigned long long owner, unsigned int cpos,
+		 unsigned int len, int index,
+		 unsigned int e_cpos, unsigned int clusters),
+	TP_ARGS(owner, cpos, len, index, e_cpos, clusters),
+	TP_STRUCT__entry(
+		__field(unsigned long long, owner)
+		__field(unsigned int, cpos)
+		__field(unsigned int, len)
+		__field(int, index)
+		__field(unsigned int, e_cpos)
+		__field(unsigned int, clusters)
+	),
+	TP_fast_assign(
+		__entry->owner = owner;
+		__entry->cpos = cpos;
+		__entry->len = len;
+		__entry->index = index;
+		__entry->e_cpos = e_cpos;
+		__entry->clusters = clusters;
+	),
+	TP_printk("%llu %u %u %d %u %u",
+		  __entry->owner, __entry->cpos, __entry->len, __entry->index,
+		  __entry->e_cpos, __entry->clusters)
+);
+
+TRACE_EVENT(ocfs2_commit_truncate,
+	TP_PROTO(unsigned long long ino, unsigned int new_cpos,
+		 unsigned int clusters, unsigned int depth),
+	TP_ARGS(ino, new_cpos, clusters, depth),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned int, new_cpos)
+		__field(unsigned int, clusters)
+		__field(unsigned int, depth)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->new_cpos = new_cpos;
+		__entry->clusters = clusters;
+		__entry->depth = depth;
+	),
+	TP_printk("%llu %u %u %u",
+		  __entry->ino, __entry->new_cpos,
+		  __entry->clusters, __entry->depth)
+);
+
+TRACE_EVENT(ocfs2_validate_extent_block,
+	TP_PROTO(unsigned long long blkno),
+	TP_ARGS(blkno),
+	TP_STRUCT__entry(
+		__field(unsigned long long, blkno)
+	),
+	TP_fast_assign(
+		__entry->blkno = blkno;
+	),
+	TP_printk("%llu ", __entry->blkno)
+);
+
+TRACE_EVENT(ocfs2_rotate_leaf,
+	TP_PROTO(unsigned int insert_cpos, int insert_index,
+		 int has_empty, int next_free,
+		 unsigned int l_count),
+	TP_ARGS(insert_cpos, insert_index, has_empty,
+		next_free, l_count),
+	TP_STRUCT__entry(
+		__field(unsigned int, insert_cpos)
+		__field(int, insert_index)
+		__field(int, has_empty)
+		__field(int, next_free)
+		__field(unsigned int, l_count)
+	),
+	TP_fast_assign(
+		__entry->insert_cpos = insert_cpos;
+		__entry->insert_index = insert_index;
+		__entry->has_empty = has_empty;
+		__entry->next_free = next_free;
+		__entry->l_count = l_count;
+	),
+	TP_printk("%u %d %d %d %u", __entry->insert_cpos,
+		  __entry->insert_index, __entry->has_empty,
+		  __entry->next_free, __entry->l_count)
+);
+
+TRACE_EVENT(ocfs2_add_clusters_in_btree_ret,
+	TP_PROTO(int status, int reason, int err),
+	TP_ARGS(status, reason, err),
+	TP_STRUCT__entry(
+		__field(int, status)
+		__field(int, reason)
+		__field(int, err)
+	),
+	TP_fast_assign(
+		__entry->status = status;
+		__entry->reason = reason;
+		__entry->err = err;
+	),
+	TP_printk("%d %d %d", __entry->status,
+		  __entry->reason, __entry->err)
+);
+
+TRACE_EVENT(ocfs2_mark_extent_written,
+	TP_PROTO(unsigned long long owner, unsigned int cpos,
+		 unsigned int len, unsigned int phys),
+	TP_ARGS(owner, cpos, len, phys),
+	TP_STRUCT__entry(
+		__field(unsigned long long, owner)
+		__field(unsigned int, cpos)
+		__field(unsigned int, len)
+		__field(unsigned int, phys)
+	),
+	TP_fast_assign(
+		__entry->owner = owner;
+		__entry->cpos = cpos;
+		__entry->len = len;
+		__entry->phys = phys;
+	),
+	TP_printk("%llu %u %u %u",
+		  __entry->owner, __entry->cpos,
+		  __entry->len, __entry->phys)
+);
+
+DECLARE_EVENT_CLASS(ocfs2__truncate_log_ops,
+	TP_PROTO(unsigned long long blkno, int index,
+		 unsigned int start, unsigned int num),
+	TP_ARGS(blkno, index, start, num),
+	TP_STRUCT__entry(
+		__field(unsigned long long, blkno)
+		__field(int, index)
+		__field(unsigned int, start)
+		__field(unsigned int, num)
+	),
+	TP_fast_assign(
+		__entry->blkno = blkno;
+		__entry->index = index;
+		__entry->start = start;
+		__entry->num = num;
+	),
+	TP_printk("%llu %d %u %u",
+		  __entry->blkno, __entry->index,
+		  __entry->start, __entry->num)
+);
+
+#define DEFINE_OCFS2_TRUNCATE_LOG_OPS_EVENT(name)	\
+DEFINE_EVENT(ocfs2__truncate_log_ops, name,	\
+	TP_PROTO(unsigned long long blkno, int index,	\
+		 unsigned int start, unsigned int num),	\
+	TP_ARGS(blkno, index, start, num))
+
+DEFINE_OCFS2_TRUNCATE_LOG_OPS_EVENT(ocfs2_truncate_log_append);
+
+DEFINE_OCFS2_TRUNCATE_LOG_OPS_EVENT(ocfs2_replay_truncate_records);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_flush_truncate_log);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_begin_truncate_log_recovery);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_truncate_log_recovery_num);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_complete_truncate_log_recovery);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_free_cached_blocks);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_cache_cluster_dealloc);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_run_deallocs);
+
+TRACE_EVENT(ocfs2_cache_block_dealloc,
+	TP_PROTO(int type, int slot, unsigned long long suballoc,
+		 unsigned long long blkno, unsigned int bit),
+	TP_ARGS(type, slot, suballoc, blkno, bit),
+	TP_STRUCT__entry(
+		__field(int, type)
+		__field(int, slot)
+		__field(unsigned long long, suballoc)
+		__field(unsigned long long, blkno)
+		__field(unsigned int, bit)
+	),
+	TP_fast_assign(
+		__entry->type = type;
+		__entry->slot = slot;
+		__entry->suballoc = suballoc;
+		__entry->blkno = blkno;
+		__entry->bit = bit;
+	),
+	TP_printk("%d %d %llu %llu %u",
+		  __entry->type, __entry->slot, __entry->suballoc,
+		  __entry->blkno, __entry->bit)
+);
+
+TRACE_EVENT(ocfs2_trim_extent,
+	TP_PROTO(struct super_block *sb, unsigned long long blk,
+		 unsigned long long count),
+	TP_ARGS(sb, blk, count),
+	TP_STRUCT__entry(
+		__field(int, dev_major)
+		__field(int, dev_minor)
+		__field(unsigned long long, blk)
+		__field(__u64,	count)
+	),
+	TP_fast_assign(
+		__entry->dev_major = MAJOR(sb->s_dev);
+		__entry->dev_minor = MINOR(sb->s_dev);
+		__entry->blk = blk;
+		__entry->count = count;
+	),
+	TP_printk("%d %d %llu %llu",
+		  __entry->dev_major, __entry->dev_minor,
+		  __entry->blk, __entry->count)
+);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_trim_group);
+
+DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_trim_fs);
+
+/* End of trace events for fs/ocfs2/alloc.c. */
+
+/* Trace events for fs/ocfs2/localalloc.c. */
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_la_set_sizes);
+
+DEFINE_OCFS2_ULL_INT_INT_INT_EVENT(ocfs2_alloc_should_use_local);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_load_local_alloc);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_begin_local_alloc_recovery);
+
+DEFINE_OCFS2_ULL_INT_INT_INT_EVENT(ocfs2_reserve_local_alloc_bits);
+
+DEFINE_OCFS2_UINT_EVENT(ocfs2_local_alloc_count_bits);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_local_alloc_find_clear_bits_search_bitmap);
+
+DEFINE_OCFS2_ULL_INT_INT_INT_EVENT(ocfs2_local_alloc_find_clear_bits);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_sync_local_to_main);
+
+TRACE_EVENT(ocfs2_sync_local_to_main_free,
+	TP_PROTO(int count, int bit, unsigned long long start_blk,
+		 unsigned long long blkno),
+	TP_ARGS(count, bit, start_blk, blkno),
+	TP_STRUCT__entry(
+		__field(int, count)
+		__field(int, bit)
+		__field(unsigned long long, start_blk)
+		__field(unsigned long long, blkno)
+	),
+	TP_fast_assign(
+		__entry->count = count;
+		__entry->bit = bit;
+		__entry->start_blk = start_blk;
+		__entry->blkno = blkno;
+	),
+	TP_printk("%d %d %llu %llu",
+		  __entry->count, __entry->bit, __entry->start_blk,
+		  __entry->blkno)
+);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_local_alloc_new_window);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_local_alloc_new_window_result);
+
+/* End of trace events for fs/ocfs2/localalloc.c. */
+
+/* Trace events for fs/ocfs2/resize.c. */
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_update_last_group_and_inode);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_group_extend);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_group_add);
+
+/* End of trace events for fs/ocfs2/resize.c. */
+
+/* Trace events for fs/ocfs2/suballoc.c. */
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_validate_group_descriptor);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_block_group_alloc_contig);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_block_group_alloc_discontig);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_block_group_alloc);
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_reserve_suballoc_bits_nospc);
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_reserve_suballoc_bits_no_new_group);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_reserve_new_inode_new_group);
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_block_group_set_bits);
+
+TRACE_EVENT(ocfs2_relink_block_group,
+	TP_PROTO(unsigned long long i_blkno, unsigned int chain,
+		 unsigned long long bg_blkno,
+		 unsigned long long prev_blkno),
+	TP_ARGS(i_blkno, chain, bg_blkno, prev_blkno),
+	TP_STRUCT__entry(
+		__field(unsigned long long, i_blkno)
+		__field(unsigned int, chain)
+		__field(unsigned long long, bg_blkno)
+		__field(unsigned long long, prev_blkno)
+	),
+	TP_fast_assign(
+		__entry->i_blkno = i_blkno;
+		__entry->chain = chain;
+		__entry->bg_blkno = bg_blkno;
+		__entry->prev_blkno = prev_blkno;
+	),
+	TP_printk("%llu %u %llu %llu",
+		  __entry->i_blkno, __entry->chain, __entry->bg_blkno,
+		  __entry->prev_blkno)
+);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_cluster_group_search_wrong_max_bits);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_cluster_group_search_max_block);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_block_group_search_max_block);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_search_chain_begin);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_search_chain_succ);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_search_chain_end);
+
+DEFINE_OCFS2_UINT_EVENT(ocfs2_claim_suballoc_bits);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_claim_new_inode_at_loc);
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_block_group_clear_bits);
+
+TRACE_EVENT(ocfs2_free_suballoc_bits,
+	TP_PROTO(unsigned long long inode, unsigned long long group,
+		 unsigned int start_bit, unsigned int count),
+	TP_ARGS(inode, group, start_bit, count),
+	TP_STRUCT__entry(
+		__field(unsigned long long, inode)
+		__field(unsigned long long, group)
+		__field(unsigned int, start_bit)
+		__field(unsigned int, count)
+	),
+	TP_fast_assign(
+		__entry->inode = inode;
+		__entry->group = group;
+		__entry->start_bit = start_bit;
+		__entry->count = count;
+	),
+	TP_printk("%llu %llu %u %u", __entry->inode, __entry->group,
+		  __entry->start_bit, __entry->count)
+);
+
+TRACE_EVENT(ocfs2_free_clusters,
+	TP_PROTO(unsigned long long bg_blkno, unsigned long long start_blk,
+		 unsigned int start_bit, unsigned int count),
+	TP_ARGS(bg_blkno, start_blk, start_bit, count),
+	TP_STRUCT__entry(
+		__field(unsigned long long, bg_blkno)
+		__field(unsigned long long, start_blk)
+		__field(unsigned int, start_bit)
+		__field(unsigned int, count)
+	),
+	TP_fast_assign(
+		__entry->bg_blkno = bg_blkno;
+		__entry->start_blk = start_blk;
+		__entry->start_bit = start_bit;
+		__entry->count = count;
+	),
+	TP_printk("%llu %llu %u %u", __entry->bg_blkno, __entry->start_blk,
+		  __entry->start_bit, __entry->count)
+);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_get_suballoc_slot_bit);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_test_suballoc_bit);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_test_inode_bit);
+
+/* End of trace events for fs/ocfs2/suballoc.c. */
+
+/* Trace events for fs/ocfs2/refcounttree.c. */
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_validate_refcount_block);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_purge_refcount_trees);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_create_refcount_tree);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_create_refcount_tree_blkno);
+
+DEFINE_OCFS2_ULL_INT_INT_INT_EVENT(ocfs2_change_refcount_rec);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_expand_inline_ref_root);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_divide_leaf_refcount_block);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_new_leaf_refcount_block);
+
+DECLARE_EVENT_CLASS(ocfs2__refcount_tree_ops,
+	TP_PROTO(unsigned long long blkno, int index,
+		 unsigned long long cpos,
+		 unsigned int clusters, unsigned int refcount),
+	TP_ARGS(blkno, index, cpos, clusters, refcount),
+	TP_STRUCT__entry(
+		__field(unsigned long long, blkno)
+		__field(int, index)
+		__field(unsigned long long, cpos)
+		__field(unsigned int, clusters)
+		__field(unsigned int, refcount)
+	),
+	TP_fast_assign(
+		__entry->blkno = blkno;
+		__entry->index = index;
+		__entry->cpos = cpos;
+		__entry->clusters = clusters;
+		__entry->refcount = refcount;
+	),
+	TP_printk("%llu %d %llu %u %u", __entry->blkno, __entry->index,
+		  __entry->cpos, __entry->clusters, __entry->refcount)
+);
+
+#define DEFINE_OCFS2_REFCOUNT_TREE_OPS_EVENT(name)	\
+DEFINE_EVENT(ocfs2__refcount_tree_ops, name,		\
+	TP_PROTO(unsigned long long blkno, int index,	\
+		 unsigned long long cpos,		\
+		 unsigned int count, unsigned int refcount),	\
+	TP_ARGS(blkno, index, cpos, count, refcount))
+
+DEFINE_OCFS2_REFCOUNT_TREE_OPS_EVENT(ocfs2_insert_refcount_rec);
+
+TRACE_EVENT(ocfs2_split_refcount_rec,
+	TP_PROTO(unsigned long long cpos,
+		 unsigned int clusters, unsigned int refcount,
+		 unsigned long long split_cpos,
+		 unsigned int split_clusters, unsigned int split_refcount),
+	TP_ARGS(cpos, clusters, refcount,
+		split_cpos, split_clusters, split_refcount),
+	TP_STRUCT__entry(
+		__field(unsigned long long, cpos)
+		__field(unsigned int, clusters)
+		__field(unsigned int, refcount)
+		__field(unsigned long long, split_cpos)
+		__field(unsigned int, split_clusters)
+		__field(unsigned int, split_refcount)
+	),
+	TP_fast_assign(
+		__entry->cpos = cpos;
+		__entry->clusters = clusters;
+		__entry->refcount = refcount;
+		__entry->split_cpos = split_cpos;
+		__entry->split_clusters = split_clusters;
+		__entry->split_refcount	= split_refcount;
+	),
+	TP_printk("%llu %u %u %llu %u %u",
+		  __entry->cpos, __entry->clusters, __entry->refcount,
+		  __entry->split_cpos, __entry->split_clusters,
+		  __entry->split_refcount)
+);
+
+DEFINE_OCFS2_REFCOUNT_TREE_OPS_EVENT(ocfs2_split_refcount_rec_insert);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_increase_refcount_begin);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_increase_refcount_change);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_increase_refcount_insert);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_increase_refcount_split);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_remove_refcount_extent);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_restore_refcount_block);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_decrease_refcount_rec);
+
+TRACE_EVENT(ocfs2_decrease_refcount,
+	TP_PROTO(unsigned long long owner,
+		 unsigned long long cpos,
+		 unsigned int len, int delete),
+	TP_ARGS(owner, cpos, len, delete),
+	TP_STRUCT__entry(
+		__field(unsigned long long, owner)
+		__field(unsigned long long, cpos)
+		__field(unsigned int, len)
+		__field(int, delete)
+	),
+	TP_fast_assign(
+		__entry->owner = owner;
+		__entry->cpos = cpos;
+		__entry->len = len;
+		__entry->delete = delete;
+	),
+	TP_printk("%llu %llu %u %d",
+		  __entry->owner, __entry->cpos, __entry->len, __entry->delete)
+);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_mark_extent_refcounted);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_calc_refcount_meta_credits);
+
+TRACE_EVENT(ocfs2_calc_refcount_meta_credits_iterate,
+	TP_PROTO(int recs_add, unsigned long long cpos,
+		 unsigned int clusters, unsigned long long r_cpos,
+		 unsigned int r_clusters, unsigned int refcount, int index),
+	TP_ARGS(recs_add, cpos, clusters, r_cpos, r_clusters, refcount, index),
+	TP_STRUCT__entry(
+		__field(int, recs_add)
+		__field(unsigned long long, cpos)
+		__field(unsigned int, clusters)
+		__field(unsigned long long, r_cpos)
+		__field(unsigned int, r_clusters)
+		__field(unsigned int, refcount)
+		__field(int, index)
+	),
+	TP_fast_assign(
+		__entry->recs_add = recs_add;
+		__entry->cpos = cpos;
+		__entry->clusters = clusters;
+		__entry->r_cpos = r_cpos;
+		__entry->r_clusters = r_clusters;
+		__entry->refcount = refcount;
+		__entry->index = index;
+	),
+	TP_printk("%d %llu %u %llu %u %u %d",
+		  __entry->recs_add, __entry->cpos, __entry->clusters,
+		  __entry->r_cpos, __entry->r_clusters,
+		  __entry->refcount, __entry->index)
+);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_add_refcount_flag);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_prepare_refcount_change_for_del);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_lock_refcount_allocators);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_duplicate_clusters_by_page);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_duplicate_clusters_by_jbd);
+
+TRACE_EVENT(ocfs2_clear_ext_refcount,
+	TP_PROTO(unsigned long long ino, unsigned int cpos,
+		 unsigned int len, unsigned int p_cluster,
+		 unsigned int ext_flags),
+	TP_ARGS(ino, cpos, len, p_cluster, ext_flags),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned int, cpos)
+		__field(unsigned int, len)
+		__field(unsigned int, p_cluster)
+		__field(unsigned int, ext_flags)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->cpos = cpos;
+		__entry->len = len;
+		__entry->p_cluster = p_cluster;
+		__entry->ext_flags = ext_flags;
+	),
+	TP_printk("%llu %u %u %u %u",
+		  __entry->ino, __entry->cpos, __entry->len,
+		  __entry->p_cluster, __entry->ext_flags)
+);
+
+TRACE_EVENT(ocfs2_replace_clusters,
+	TP_PROTO(unsigned long long ino, unsigned int cpos,
+		 unsigned int old, unsigned int new, unsigned int len,
+		 unsigned int ext_flags),
+	TP_ARGS(ino, cpos, old, new, len, ext_flags),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned int, cpos)
+		__field(unsigned int, old)
+		__field(unsigned int, new)
+		__field(unsigned int, len)
+		__field(unsigned int, ext_flags)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->cpos = cpos;
+		__entry->old = old;
+		__entry->new = new;
+		__entry->len = len;
+		__entry->ext_flags = ext_flags;
+	),
+	TP_printk("%llu %u %u %u %u %u",
+		  __entry->ino, __entry->cpos, __entry->old, __entry->new,
+		  __entry->len, __entry->ext_flags)
+);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_make_clusters_writable);
+
+TRACE_EVENT(ocfs2_refcount_cow_hunk,
+	TP_PROTO(unsigned long long ino, unsigned int cpos,
+		 unsigned int write_len, unsigned int max_cpos,
+		 unsigned int cow_start, unsigned int cow_len),
+	TP_ARGS(ino, cpos, write_len, max_cpos, cow_start, cow_len),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned int, cpos)
+		__field(unsigned int, write_len)
+		__field(unsigned int, max_cpos)
+		__field(unsigned int, cow_start)
+		__field(unsigned int, cow_len)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->cpos = cpos;
+		__entry->write_len = write_len;
+		__entry->max_cpos = max_cpos;
+		__entry->cow_start = cow_start;
+		__entry->cow_len = cow_len;
+	),
+	TP_printk("%llu %u %u %u %u %u",
+		  __entry->ino, __entry->cpos, __entry->write_len,
+		  __entry->max_cpos, __entry->cow_start, __entry->cow_len)
+);
+
+/* End of trace events for fs/ocfs2/refcounttree.c. */
+
+/* Trace events for fs/ocfs2/aops.c. */
+
+DECLARE_EVENT_CLASS(ocfs2__get_block,
+	TP_PROTO(unsigned long long ino, unsigned long long iblock,
+		 void *bh_result, int create),
+	TP_ARGS(ino, iblock, bh_result, create),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned long long, iblock)
+		__field(void *, bh_result)
+		__field(int, create)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->iblock = iblock;
+		__entry->bh_result = bh_result;
+		__entry->create = create;
+	),
+	TP_printk("%llu %llu %p %d",
+		  __entry->ino, __entry->iblock,
+		  __entry->bh_result, __entry->create)
+);
+
+#define DEFINE_OCFS2_GET_BLOCK_EVENT(name)	\
+DEFINE_EVENT(ocfs2__get_block, name,	\
+	TP_PROTO(unsigned long long ino, unsigned long long iblock,	\
+		 void *bh_result, int create),	\
+	TP_ARGS(ino, iblock, bh_result, create))
+
+DEFINE_OCFS2_GET_BLOCK_EVENT(ocfs2_symlink_get_block);
+
+DEFINE_OCFS2_GET_BLOCK_EVENT(ocfs2_get_block);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_get_block_end);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_readpage);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_writepage);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_bmap);
+
+TRACE_EVENT(ocfs2_try_to_write_inline_data,
+	TP_PROTO(unsigned long long ino, unsigned int len,
+		 unsigned long long pos, unsigned int flags),
+	TP_ARGS(ino, len, pos, flags),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned int, len)
+		__field(unsigned long long, pos)
+		__field(unsigned int, flags)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->len = len;
+		__entry->pos = pos;
+		__entry->flags = flags;
+	),
+	TP_printk("%llu %u %llu 0x%x",
+		  __entry->ino, __entry->len, __entry->pos, __entry->flags)
+);
+
+TRACE_EVENT(ocfs2_write_begin_nolock,
+	TP_PROTO(unsigned long long ino,
+		 long long i_size, unsigned int i_clusters,
+		 unsigned long long pos, unsigned int len,
+		 unsigned int flags, void *page,
+		 unsigned int clusters, unsigned int extents_to_split),
+	TP_ARGS(ino, i_size, i_clusters, pos, len, flags,
+		page, clusters, extents_to_split),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(long long, i_size)
+		__field(unsigned int, i_clusters)
+		__field(unsigned long long, pos)
+		__field(unsigned int, len)
+		__field(unsigned int, flags)
+		__field(void *, page)
+		__field(unsigned int, clusters)
+		__field(unsigned int, extents_to_split)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->i_size = i_size;
+		__entry->i_clusters = i_clusters;
+		__entry->pos = pos;
+		__entry->len = len;
+		__entry->flags = flags;
+		__entry->page = page;
+		__entry->clusters = clusters;
+		__entry->extents_to_split = extents_to_split;
+	),
+	TP_printk("%llu %lld %u %llu %u %u %p %u %u",
+		  __entry->ino, __entry->i_size, __entry->i_clusters,
+		  __entry->pos, __entry->len,
+		  __entry->flags, __entry->page, __entry->clusters,
+		  __entry->extents_to_split)
+);
+
+TRACE_EVENT(ocfs2_write_end_inline,
+	TP_PROTO(unsigned long long ino,
+		 unsigned long long pos, unsigned int copied,
+		 unsigned int id_count, unsigned int features),
+	TP_ARGS(ino, pos, copied, id_count, features),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned long long, pos)
+		__field(unsigned int, copied)
+		__field(unsigned int, id_count)
+		__field(unsigned int, features)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->pos = pos;
+		__entry->copied = copied;
+		__entry->id_count = id_count;
+		__entry->features = features;
+	),
+	TP_printk("%llu %llu %u %u %u",
+		  __entry->ino, __entry->pos, __entry->copied,
+		  __entry->id_count, __entry->features)
+);
+
+/* End of trace events for fs/ocfs2/aops.c. */
+
+/* Trace events for fs/ocfs2/mmap.c. */
+
+TRACE_EVENT(ocfs2_fault,
+	TP_PROTO(unsigned long long ino,
+		 void *area, void *page, unsigned long pgoff),
+	TP_ARGS(ino, area, page, pgoff),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(void *, area)
+		__field(void *, page)
+		__field(unsigned long, pgoff)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->area = area;
+		__entry->page = page;
+		__entry->pgoff = pgoff;
+	),
+	TP_printk("%llu %p %p %lu",
+		  __entry->ino, __entry->area, __entry->page, __entry->pgoff)
+);
+
+/* End of trace events for fs/ocfs2/mmap.c. */
+
+/* Trace events for fs/ocfs2/file.c. */
+
+DECLARE_EVENT_CLASS(ocfs2__file_ops,
+	TP_PROTO(void *inode, void *file, void *dentry,
+		 unsigned long long ino,
+		 unsigned int d_len, const unsigned char *d_name,
+		 unsigned long long para),
+	TP_ARGS(inode, file, dentry, ino, d_len, d_name, para),
+	TP_STRUCT__entry(
+		__field(void *, inode)
+		__field(void *, file)
+		__field(void *, dentry)
+		__field(unsigned long long, ino)
+		__field(unsigned int, d_len)
+		__string(d_name, d_name)
+		__field(unsigned long long, para)
+	),
+	TP_fast_assign(
+		__entry->inode = inode;
+		__entry->file = file;
+		__entry->dentry = dentry;
+		__entry->ino = ino;
+		__entry->d_len = d_len;
+		__assign_str(d_name, d_name);
+		__entry->para = para;
+	),
+	TP_printk("%p %p %p %llu %llu %.*s", __entry->inode, __entry->file,
+		  __entry->dentry, __entry->ino, __entry->para,
+		  __entry->d_len, __get_str(d_name))
+);
+
+#define DEFINE_OCFS2_FILE_OPS(name)				\
+DEFINE_EVENT(ocfs2__file_ops, name,				\
+TP_PROTO(void *inode, void *file, void *dentry,			\
+	 unsigned long long ino,				\
+	 unsigned int d_len, const unsigned char *d_name,	\
+	 unsigned long long mode),				\
+	TP_ARGS(inode, file, dentry, ino, d_len, d_name, mode))
+
+DEFINE_OCFS2_FILE_OPS(ocfs2_file_open);
+
+DEFINE_OCFS2_FILE_OPS(ocfs2_file_release);
+
+DEFINE_OCFS2_FILE_OPS(ocfs2_sync_file);
+
+DEFINE_OCFS2_FILE_OPS(ocfs2_file_aio_write);
+
+DEFINE_OCFS2_FILE_OPS(ocfs2_file_splice_write);
+
+DEFINE_OCFS2_FILE_OPS(ocfs2_file_splice_read);
+
+DEFINE_OCFS2_FILE_OPS(ocfs2_file_aio_read);
+
+DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_truncate_file);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_truncate_file_error);
+
+TRACE_EVENT(ocfs2_extend_allocation,
+	TP_PROTO(unsigned long long ip_blkno, unsigned long long size,
+		 unsigned int clusters, unsigned int clusters_to_add,
+		 int why, int restart_func),
+	TP_ARGS(ip_blkno, size, clusters, clusters_to_add, why, restart_func),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ip_blkno)
+		__field(unsigned long long, size)
+		__field(unsigned int, clusters)
+		__field(unsigned int, clusters_to_add)
+		__field(int, why)
+		__field(int, restart_func)
+	),
+	TP_fast_assign(
+		__entry->ip_blkno = ip_blkno;
+		__entry->size = size;
+		__entry->clusters = clusters;
+		__entry->clusters_to_add = clusters_to_add;
+		__entry->why = why;
+		__entry->restart_func = restart_func;
+	),
+	TP_printk("%llu %llu %u %u %d %d",
+		  __entry->ip_blkno, __entry->size, __entry->clusters,
+		  __entry->clusters_to_add, __entry->why, __entry->restart_func)
+);
+
+TRACE_EVENT(ocfs2_extend_allocation_end,
+	TP_PROTO(unsigned long long ino,
+		 unsigned int di_clusters, unsigned long long di_size,
+		 unsigned int ip_clusters, unsigned long long i_size),
+	TP_ARGS(ino, di_clusters, di_size, ip_clusters, i_size),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned int, di_clusters)
+		__field(unsigned long long, di_size)
+		__field(unsigned int, ip_clusters)
+		__field(unsigned long long, i_size)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->di_clusters = di_clusters;
+		__entry->di_size = di_size;
+		__entry->ip_clusters = ip_clusters;
+		__entry->i_size = i_size;
+	),
+	TP_printk("%llu %u %llu %u %llu", __entry->ino, __entry->di_clusters,
+		  __entry->di_size, __entry->ip_clusters, __entry->i_size)
+);
+
+TRACE_EVENT(ocfs2_write_zero_page,
+	TP_PROTO(unsigned long long ino,
+		 unsigned long long abs_from, unsigned long long abs_to,
+		 unsigned long index, unsigned int zero_from,
+		 unsigned int zero_to),
+	TP_ARGS(ino, abs_from, abs_to, index, zero_from, zero_to),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned long long, abs_from)
+		__field(unsigned long long, abs_to)
+		__field(unsigned long, index)
+		__field(unsigned int, zero_from)
+		__field(unsigned int, zero_to)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->abs_from = abs_from;
+		__entry->abs_to = abs_to;
+		__entry->index = index;
+		__entry->zero_from = zero_from;
+		__entry->zero_to = zero_to;
+	),
+	TP_printk("%llu %llu %llu %lu %u %u", __entry->ino,
+		  __entry->abs_from, __entry->abs_to,
+		  __entry->index, __entry->zero_from, __entry->zero_to)
+);
+
+DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_zero_extend_range);
+
+DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_zero_extend);
+
+TRACE_EVENT(ocfs2_setattr,
+	TP_PROTO(void *inode, void *dentry,
+		 unsigned long long ino,
+		 unsigned int d_len, const unsigned char *d_name,
+		 unsigned int ia_valid, unsigned int ia_mode,
+		 unsigned int ia_uid, unsigned int ia_gid),
+	TP_ARGS(inode, dentry, ino, d_len, d_name,
+		ia_valid, ia_mode, ia_uid, ia_gid),
+	TP_STRUCT__entry(
+		__field(void *, inode)
+		__field(void *, dentry)
+		__field(unsigned long long, ino)
+		__field(unsigned int, d_len)
+		__string(d_name, d_name)
+		__field(unsigned int, ia_valid)
+		__field(unsigned int, ia_mode)
+		__field(unsigned int, ia_uid)
+		__field(unsigned int, ia_gid)
+	),
+	TP_fast_assign(
+		__entry->inode = inode;
+		__entry->dentry = dentry;
+		__entry->ino = ino;
+		__entry->d_len = d_len;
+		__assign_str(d_name, d_name);
+		__entry->ia_valid = ia_valid;
+		__entry->ia_mode = ia_mode;
+		__entry->ia_uid = ia_uid;
+		__entry->ia_gid = ia_gid;
+	),
+	TP_printk("%p %p %llu %.*s %u %u %u %u", __entry->inode,
+		  __entry->dentry, __entry->ino, __entry->d_len,
+		  __get_str(d_name), __entry->ia_valid, __entry->ia_mode,
+		  __entry->ia_uid, __entry->ia_gid)
+);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_write_remove_suid);
+
+DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_zero_partial_clusters);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_zero_partial_clusters_range1);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_zero_partial_clusters_range2);
+
+DEFINE_OCFS2_ULL_ULL_ULL_EVENT(ocfs2_remove_inode_range);
+
+TRACE_EVENT(ocfs2_prepare_inode_for_write,
+	TP_PROTO(unsigned long long ino, unsigned long long saved_pos,
+		 int appending, unsigned long count,
+		 int *direct_io, int *has_refcount),
+	TP_ARGS(ino, saved_pos, appending, count, direct_io, has_refcount),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned long long, saved_pos)
+		__field(int, appending)
+		__field(unsigned long, count)
+		__field(int, direct_io)
+		__field(int, has_refcount)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->saved_pos = saved_pos;
+		__entry->appending = appending;
+		__entry->count = count;
+		__entry->direct_io = direct_io ? *direct_io : -1;
+		__entry->has_refcount = has_refcount ? *has_refcount : -1;
+	),
+	TP_printk("%llu %llu %d %lu %d %d", __entry->ino,
+		  __entry->saved_pos, __entry->appending, __entry->count,
+		  __entry->direct_io, __entry->has_refcount)
+);
+
+DEFINE_OCFS2_INT_EVENT(generic_file_aio_read_ret);
+
+/* End of trace events for fs/ocfs2/file.c. */
+
+/* Trace events for fs/ocfs2/inode.c. */
+
+TRACE_EVENT(ocfs2_iget_begin,
+	TP_PROTO(unsigned long long ino, unsigned int flags, int sysfile_type),
+	TP_ARGS(ino, flags, sysfile_type),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(unsigned int, flags)
+		__field(int, sysfile_type)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->flags = flags;
+		__entry->sysfile_type = sysfile_type;
+	),
+	TP_printk("%llu %u %d", __entry->ino,
+		  __entry->flags, __entry->sysfile_type)
+);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_iget5_locked);
+
+TRACE_EVENT(ocfs2_iget_end,
+	TP_PROTO(void *inode, unsigned long long ino),
+	TP_ARGS(inode, ino),
+	TP_STRUCT__entry(
+		__field(void *, inode)
+		__field(unsigned long long, ino)
+	),
+	TP_fast_assign(
+		__entry->inode = inode;
+		__entry->ino = ino;
+	),
+	TP_printk("%p %llu", __entry->inode, __entry->ino)
+);
+
+TRACE_EVENT(ocfs2_find_actor,
+	TP_PROTO(void *inode, unsigned long long ino,
+		 void *args,  unsigned long long fi_blkno),
+	TP_ARGS(inode, ino, args, fi_blkno),
+	TP_STRUCT__entry(
+		__field(void *, inode)
+		__field(unsigned long long, ino)
+		__field(void *, args)
+		__field(unsigned long long, fi_blkno)
+	),
+	TP_fast_assign(
+		__entry->inode = inode;
+		__entry->ino = ino;
+		__entry->args = args;
+		__entry->fi_blkno = fi_blkno;
+	),
+	TP_printk("%p %llu %p %llu", __entry->inode, __entry->ino,
+		  __entry->args, __entry->fi_blkno)
+);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_populate_inode);
+
+DEFINE_OCFS2_ULL_INT_EVENT(ocfs2_read_locked_inode);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_check_orphan_recovery_state);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_validate_inode_block);
+
+TRACE_EVENT(ocfs2_inode_is_valid_to_delete,
+	TP_PROTO(void *task, void *dc_task, unsigned long long ino,
+		 unsigned int flags),
+	TP_ARGS(task, dc_task, ino, flags),
+	TP_STRUCT__entry(
+		__field(void *, task)
+		__field(void *, dc_task)
+		__field(unsigned long long, ino)
+		__field(unsigned int, flags)
+	),
+	TP_fast_assign(
+		__entry->task = task;
+		__entry->dc_task = dc_task;
+		__entry->ino = ino;
+		__entry->flags = flags;
+	),
+	TP_printk("%p %p %llu %u", __entry->task, __entry->dc_task,
+		  __entry->ino, __entry->flags)
+);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_query_inode_wipe_begin);
+
+DEFINE_OCFS2_UINT_EVENT(ocfs2_query_inode_wipe_succ);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_query_inode_wipe_end);
+
+DEFINE_OCFS2_ULL_INT_EVENT(ocfs2_cleanup_delete_inode);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_delete_inode);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_clear_inode);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_drop_inode);
+
+TRACE_EVENT(ocfs2_inode_revalidate,
+	TP_PROTO(void *inode, unsigned long long ino,
+		 unsigned int flags),
+	TP_ARGS(inode, ino, flags),
+	TP_STRUCT__entry(
+		__field(void *, inode)
+		__field(unsigned long long, ino)
+		__field(unsigned int, flags)
+	),
+	TP_fast_assign(
+		__entry->inode = inode;
+		__entry->ino = ino;
+		__entry->flags = flags;
+	),
+	TP_printk("%p %llu %u", __entry->inode, __entry->ino, __entry->flags)
+);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_mark_inode_dirty);
+
+/* End of trace events for fs/ocfs2/inode.c. */
+
+/* Trace events for fs/ocfs2/extent_map.c. */
+
+TRACE_EVENT(ocfs2_read_virt_blocks,
+	TP_PROTO(void *inode, unsigned long long vblock, int nr,
+		 void *bhs, unsigned int flags, void *validate),
+	TP_ARGS(inode, vblock, nr, bhs, flags, validate),
+	TP_STRUCT__entry(
+		__field(void *, inode)
+		__field(unsigned long long, vblock)
+		__field(int, nr)
+		__field(void *, bhs)
+		__field(unsigned int, flags)
+		__field(void *, validate)
+	),
+	TP_fast_assign(
+		__entry->inode = inode;
+		__entry->vblock = vblock;
+		__entry->nr = nr;
+		__entry->bhs = bhs;
+		__entry->flags = flags;
+		__entry->validate = validate;
+	),
+	TP_printk("%p %llu %d %p %x %p", __entry->inode, __entry->vblock,
+		  __entry->nr, __entry->bhs, __entry->flags, __entry->validate)
+);
+
+/* End of trace events for fs/ocfs2/extent_map.c. */
+
+/* Trace events for fs/ocfs2/slot_map.c. */
+
+DEFINE_OCFS2_UINT_EVENT(ocfs2_refresh_slot_info);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_map_slot_buffers);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_map_slot_buffers_block);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_find_slot);
+
+/* End of trace events for fs/ocfs2/slot_map.c. */
+
+/* Trace events for fs/ocfs2/heartbeat.c. */
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_do_node_down);
+
+/* End of trace events for fs/ocfs2/heartbeat.c. */
+
+/* Trace events for fs/ocfs2/super.c. */
+
+TRACE_EVENT(ocfs2_remount,
+	TP_PROTO(unsigned long s_flags, unsigned long osb_flags, int flags),
+	TP_ARGS(s_flags, osb_flags, flags),
+	TP_STRUCT__entry(
+		__field(unsigned long, s_flags)
+		__field(unsigned long, osb_flags)
+		__field(int, flags)
+	),
+	TP_fast_assign(
+		__entry->s_flags = s_flags;
+		__entry->osb_flags = osb_flags;
+		__entry->flags = flags;
+	),
+	TP_printk("%lu %lu %d", __entry->s_flags,
+		  __entry->osb_flags, __entry->flags)
+);
+
+TRACE_EVENT(ocfs2_fill_super,
+	TP_PROTO(void *sb, void *data, int silent),
+	TP_ARGS(sb, data, silent),
+	TP_STRUCT__entry(
+		__field(void *, sb)
+		__field(void *, data)
+		__field(int, silent)
+	),
+	TP_fast_assign(
+		__entry->sb = sb;
+		__entry->data = data;
+		__entry->silent = silent;
+	),
+	TP_printk("%p %p %d", __entry->sb,
+		  __entry->data, __entry->silent)
+);
+
+TRACE_EVENT(ocfs2_parse_options,
+	TP_PROTO(int is_remount, char *options),
+	TP_ARGS(is_remount, options),
+	TP_STRUCT__entry(
+		__field(int, is_remount)
+		__string(options, options)
+	),
+	TP_fast_assign(
+		__entry->is_remount = is_remount;
+		__assign_str(options, options);
+	),
+	TP_printk("%d %s", __entry->is_remount, __get_str(options))
+);
+
+DEFINE_OCFS2_POINTER_EVENT(ocfs2_put_super);
+
+TRACE_EVENT(ocfs2_statfs,
+	TP_PROTO(void *sb, void *buf),
+	TP_ARGS(sb, buf),
+	TP_STRUCT__entry(
+		__field(void *, sb)
+		__field(void *, buf)
+	),
+	TP_fast_assign(
+		__entry->sb = sb;
+		__entry->buf = buf;
+	),
+	TP_printk("%p %p", __entry->sb, __entry->buf)
+);
+
+DEFINE_OCFS2_POINTER_EVENT(ocfs2_dismount_volume);
+
+TRACE_EVENT(ocfs2_initialize_super,
+	TP_PROTO(char *label, char *uuid_str, unsigned long long root_dir,
+		 unsigned long long system_dir, int cluster_bits),
+	TP_ARGS(label, uuid_str, root_dir, system_dir, cluster_bits),
+	TP_STRUCT__entry(
+		__string(label, label)
+		__string(uuid_str, uuid_str)
+		__field(unsigned long long, root_dir)
+		__field(unsigned long long, system_dir)
+		__field(int, cluster_bits)
+	),
+	TP_fast_assign(
+		__assign_str(label, label);
+		__assign_str(uuid_str, uuid_str);
+		__entry->root_dir = root_dir;
+		__entry->system_dir = system_dir;
+		__entry->cluster_bits = cluster_bits;
+	),
+	TP_printk("%s %s %llu %llu %d", __get_str(label), __get_str(uuid_str),
+		  __entry->root_dir, __entry->system_dir, __entry->cluster_bits)
+);
+
+/* End of trace events for fs/ocfs2/super.c. */
+
+/* Trace events for fs/ocfs2/xattr.c. */
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_validate_xattr_block);
+
+DEFINE_OCFS2_UINT_EVENT(ocfs2_xattr_extend_allocation);
+
+TRACE_EVENT(ocfs2_init_xattr_set_ctxt,
+	TP_PROTO(const char *name, int meta, int clusters, int credits),
+	TP_ARGS(name, meta, clusters, credits),
+	TP_STRUCT__entry(
+		__string(name, name)
+		__field(int, meta)
+		__field(int, clusters)
+		__field(int, credits)
+	),
+	TP_fast_assign(
+		__assign_str(name, name);
+		__entry->meta = meta;
+		__entry->clusters = clusters;
+		__entry->credits = credits;
+	),
+	TP_printk("%s %d %d %d", __get_str(name), __entry->meta,
+		  __entry->clusters, __entry->credits)
+);
+
+DECLARE_EVENT_CLASS(ocfs2__xattr_find,
+	TP_PROTO(unsigned long long ino, const char *name, int name_index,
+		 unsigned int hash, unsigned long long location,
+		 int xe_index),
+	TP_ARGS(ino, name, name_index, hash, location, xe_index),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__string(name, name)
+		__field(int, name_index)
+		__field(unsigned int, hash)
+		__field(unsigned long long, location)
+		__field(int, xe_index)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__assign_str(name, name);
+		__entry->name_index = name_index;
+		__entry->hash = hash;
+		__entry->location = location;
+		__entry->xe_index = xe_index;
+	),
+	TP_printk("%llu %s %d %u %llu %d", __entry->ino, __get_str(name),
+		  __entry->name_index, __entry->hash, __entry->location,
+		  __entry->xe_index)
+);
+
+#define DEFINE_OCFS2_XATTR_FIND_EVENT(name)					\
+DEFINE_EVENT(ocfs2__xattr_find, name,					\
+TP_PROTO(unsigned long long ino, const char *name, int name_index,	\
+	 unsigned int hash, unsigned long long bucket,			\
+	 int xe_index),							\
+	TP_ARGS(ino, name, name_index, hash, bucket, xe_index))
+
+DEFINE_OCFS2_XATTR_FIND_EVENT(ocfs2_xattr_bucket_find);
+
+DEFINE_OCFS2_XATTR_FIND_EVENT(ocfs2_xattr_index_block_find);
+
+DEFINE_OCFS2_XATTR_FIND_EVENT(ocfs2_xattr_index_block_find_rec);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_iterate_xattr_buckets);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_iterate_xattr_bucket);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_cp_xattr_block_to_bucket_begin);
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_cp_xattr_block_to_bucket_end);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_xattr_create_index_block_begin);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_xattr_create_index_block);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_defrag_xattr_bucket);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_mv_xattr_bucket_cross_cluster);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_divide_xattr_bucket_begin);
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_divide_xattr_bucket_move);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_cp_xattr_bucket);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_mv_xattr_buckets);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_adjust_xattr_cross_cluster);
+
+DEFINE_OCFS2_ULL_ULL_UINT_UINT_EVENT(ocfs2_add_new_xattr_cluster_begin);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_add_new_xattr_cluster);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_add_new_xattr_cluster_insert);
+
+DEFINE_OCFS2_ULL_ULL_UINT_UINT_EVENT(ocfs2_extend_xattr_bucket);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_add_new_xattr_bucket);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_xattr_bucket_value_truncate);
+
+DEFINE_OCFS2_ULL_ULL_UINT_UINT_EVENT(ocfs2_rm_xattr_cluster);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_reflink_xattr_header);
+
+DEFINE_OCFS2_ULL_INT_EVENT(ocfs2_create_empty_xattr_block);
+
+DEFINE_OCFS2_STRING_EVENT(ocfs2_xattr_set_entry_bucket);
+
+DEFINE_OCFS2_STRING_EVENT(ocfs2_xattr_set_entry_index_block);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_xattr_bucket_value_refcount);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_reflink_xattr_buckets);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_reflink_xattr_rec);
+
+/* End of trace events for fs/ocfs2/xattr.c. */
+
+/* Trace events for fs/ocfs2/reservations.c. */
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_resv_insert);
+
+DEFINE_OCFS2_ULL_UINT_UINT_UINT_EVENT(ocfs2_resmap_find_free_bits_begin);
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_resmap_find_free_bits_end);
+
+TRACE_EVENT(ocfs2_resv_find_window_begin,
+	TP_PROTO(unsigned int r_start, unsigned int r_end, unsigned int goal,
+		 unsigned int wanted, int empty_root),
+	TP_ARGS(r_start, r_end, goal, wanted, empty_root),
+	TP_STRUCT__entry(
+		__field(unsigned int, r_start)
+		__field(unsigned int, r_end)
+		__field(unsigned int, goal)
+		__field(unsigned int, wanted)
+		__field(int, empty_root)
+	),
+	TP_fast_assign(
+		__entry->r_start = r_start;
+		__entry->r_end = r_end;
+		__entry->goal = goal;
+		__entry->wanted = wanted;
+		__entry->empty_root = empty_root;
+	),
+	TP_printk("%u %u %u %u %d", __entry->r_start, __entry->r_end,
+		  __entry->goal, __entry->wanted, __entry->empty_root)
+);
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_resv_find_window_prev);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_resv_find_window_next);
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_cannibalize_resv_begin);
+
+TRACE_EVENT(ocfs2_cannibalize_resv_end,
+	TP_PROTO(unsigned int start, unsigned int end, unsigned int len,
+		 unsigned int last_start, unsigned int last_len),
+	TP_ARGS(start, end, len, last_start, last_len),
+	TP_STRUCT__entry(
+		__field(unsigned int, start)
+		__field(unsigned int, end)
+		__field(unsigned int, len)
+		__field(unsigned int, last_start)
+		__field(unsigned int, last_len)
+	),
+	TP_fast_assign(
+		__entry->start = start;
+		__entry->end = end;
+		__entry->len = len;
+		__entry->last_start = last_start;
+		__entry->last_len = last_len;
+	),
+	TP_printk("%u %u %u %u %u", __entry->start, __entry->end,
+		  __entry->len, __entry->last_start, __entry->last_len)
+);
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_resmap_resv_bits);
+
+TRACE_EVENT(ocfs2_resmap_claimed_bits_begin,
+	TP_PROTO(unsigned int cstart, unsigned int cend, unsigned int clen,
+		 unsigned int r_start, unsigned int r_end, unsigned int r_len,
+		 unsigned int last_start, unsigned int last_len),
+	TP_ARGS(cstart, cend, clen, r_start, r_end,
+		r_len, last_start, last_len),
+	TP_STRUCT__entry(
+		__field(unsigned int, cstart)
+		__field(unsigned int, cend)
+		__field(unsigned int, clen)
+		__field(unsigned int, r_start)
+		__field(unsigned int, r_end)
+		__field(unsigned int, r_len)
+		__field(unsigned int, last_start)
+		__field(unsigned int, last_len)
+	),
+	TP_fast_assign(
+		__entry->cstart = cstart;
+		__entry->cend = cend;
+		__entry->clen = clen;
+		__entry->r_start = r_start;
+		__entry->r_end = r_end;
+		__entry->r_len = r_len;
+		__entry->last_start = last_start;
+		__entry->last_len = last_len;
+	),
+	TP_printk("%u %u %u %u %u %u %u %u",
+		  __entry->cstart, __entry->cend, __entry->clen,
+		  __entry->r_start, __entry->r_end, __entry->r_len,
+		  __entry->last_start, __entry->last_len)
+);
+
+TRACE_EVENT(ocfs2_resmap_claimed_bits_end,
+	TP_PROTO(unsigned int start, unsigned int end, unsigned int len,
+		 unsigned int last_start, unsigned int last_len),
+	TP_ARGS(start, end, len, last_start, last_len),
+	TP_STRUCT__entry(
+		__field(unsigned int, start)
+		__field(unsigned int, end)
+		__field(unsigned int, len)
+		__field(unsigned int, last_start)
+		__field(unsigned int, last_len)
+	),
+	TP_fast_assign(
+		__entry->start = start;
+		__entry->end = end;
+		__entry->len = len;
+		__entry->last_start = last_start;
+		__entry->last_len = last_len;
+	),
+	TP_printk("%u %u %u %u %u", __entry->start, __entry->end,
+		  __entry->len, __entry->last_start, __entry->last_len)
+);
+
+/* End of trace events for fs/ocfs2/reservations.c. */
+
+/* Trace events for fs/ocfs2/quota_local.c. */
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_recover_local_quota_file);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_finish_quota_recovery);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(olq_set_dquot);
+
+/* End of trace events for fs/ocfs2/quota_local.c. */
+
+/* Trace events for fs/ocfs2/quota_global.c. */
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_validate_quota_block);
+
+TRACE_EVENT(ocfs2_sync_dquot,
+	TP_PROTO(unsigned int dq_id, long long dqb_curspace,
+		 long long spacechange, long long curinodes,
+		 long long inodechange),
+	TP_ARGS(dq_id, dqb_curspace, spacechange, curinodes, inodechange),
+	TP_STRUCT__entry(
+		__field(unsigned int, dq_id)
+		__field(long long, dqb_curspace)
+		__field(long long, spacechange)
+		__field(long long, curinodes)
+		__field(long long, inodechange)
+	),
+	TP_fast_assign(
+		__entry->dq_id = dq_id;
+		__entry->dqb_curspace = dqb_curspace;
+		__entry->spacechange = spacechange;
+		__entry->curinodes = curinodes;
+		__entry->inodechange = inodechange;
+	),
+	TP_printk("%u %lld %lld %lld %lld", __entry->dq_id,
+		  __entry->dqb_curspace, __entry->spacechange,
+		  __entry->curinodes, __entry->inodechange)
+);
+
+TRACE_EVENT(ocfs2_sync_dquot_helper,
+	TP_PROTO(unsigned int dq_id, unsigned int dq_type, unsigned long type,
+		 const char *s_id),
+	TP_ARGS(dq_id, dq_type, type, s_id),
+
+	TP_STRUCT__entry(
+		__field(unsigned int, dq_id)
+		__field(unsigned int, dq_type)
+		__field(unsigned long, type)
+		__string(s_id, s_id)
+	),
+	TP_fast_assign(
+		__entry->dq_id = dq_id;
+		__entry->dq_type = dq_type;
+		__entry->type = type;
+		__assign_str(s_id, s_id);
+	),
+	TP_printk("%u %u %lu %s", __entry->dq_id, __entry->dq_type,
+		  __entry->type, __get_str(s_id))
+);
+
+DEFINE_OCFS2_UINT_INT_EVENT(ocfs2_write_dquot);
+
+DEFINE_OCFS2_UINT_INT_EVENT(ocfs2_release_dquot);
+
+DEFINE_OCFS2_UINT_INT_EVENT(ocfs2_acquire_dquot);
+
+DEFINE_OCFS2_UINT_INT_EVENT(ocfs2_mark_dquot_dirty);
+
+/* End of trace events for fs/ocfs2/quota_global.c. */
+
+/* Trace events for fs/ocfs2/dir.c. */
+DEFINE_OCFS2_INT_EVENT(ocfs2_search_dirblock);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_validate_dir_block);
+
+DEFINE_OCFS2_POINTER_EVENT(ocfs2_find_entry_el);
+
+TRACE_EVENT(ocfs2_dx_dir_search,
+	TP_PROTO(unsigned long long ino, int namelen, const char *name,
+		 unsigned int major_hash, unsigned int minor_hash,
+		 unsigned long long blkno),
+	TP_ARGS(ino, namelen, name, major_hash, minor_hash, blkno),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(int, namelen)
+		__string(name, name)
+		__field(unsigned int, major_hash)
+		__field(unsigned int,minor_hash)
+		__field(unsigned long long, blkno)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->namelen = namelen;
+		__assign_str(name, name);
+		__entry->major_hash = major_hash;
+		__entry->minor_hash = minor_hash;
+		__entry->blkno = blkno;
+	),
+	TP_printk("%llu %.*s %u %u %llu", __entry->ino,
+		   __entry->namelen, __get_str(name),
+		  __entry->major_hash, __entry->minor_hash, __entry->blkno)
+);
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_dx_dir_search_leaf_info);
+
+DEFINE_OCFS2_ULL_INT_EVENT(ocfs2_delete_entry_dx);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_readdir);
+
+TRACE_EVENT(ocfs2_find_files_on_disk,
+	TP_PROTO(int namelen, const char *name, void *blkno,
+		 unsigned long long dir),
+	TP_ARGS(namelen, name, blkno, dir),
+	TP_STRUCT__entry(
+		__field(int, namelen)
+		__string(name, name)
+		__field(void *, blkno)
+		__field(unsigned long long, dir)
+	),
+	TP_fast_assign(
+		__entry->namelen = namelen;
+		__assign_str(name, name);
+		__entry->blkno = blkno;
+		__entry->dir = dir;
+	),
+	TP_printk("%.*s %p %llu", __entry->namelen, __get_str(name),
+		  __entry->blkno, __entry->dir)
+);
+
+TRACE_EVENT(ocfs2_check_dir_for_entry,
+	TP_PROTO(unsigned long long dir, int namelen, const char *name),
+	TP_ARGS(dir, namelen, name),
+	TP_STRUCT__entry(
+		__field(unsigned long long, dir)
+		__field(int, namelen)
+		__string(name, name)
+	),
+	TP_fast_assign(
+		__entry->dir = dir;
+		__entry->namelen = namelen;
+		__assign_str(name, name);
+	),
+	TP_printk("%llu %.*s", __entry->dir,
+		  __entry->namelen, __get_str(name))
+);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_dx_dir_attach_index);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_dx_dir_format_cluster);
+
+TRACE_EVENT(ocfs2_dx_dir_index_root_block,
+	TP_PROTO(unsigned long long dir,
+		 unsigned int major_hash, unsigned int minor_hash,
+		 int namelen, const char *name, unsigned int num_used),
+	TP_ARGS(dir, major_hash, minor_hash, namelen, name, num_used),
+	TP_STRUCT__entry(
+		__field(unsigned long long, dir)
+		__field(unsigned int, major_hash)
+		__field(unsigned int, minor_hash)
+		__field(int, namelen)
+		__string(name, name)
+		__field(unsigned int, num_used)
+	),
+	TP_fast_assign(
+		__entry->dir = dir;
+		__entry->major_hash = major_hash;
+		__entry->minor_hash = minor_hash;
+		__entry->namelen = namelen;
+		__assign_str(name, name);
+		__entry->num_used = num_used;
+	),
+	TP_printk("%llu %x %x %.*s %u", __entry->dir,
+		  __entry->major_hash, __entry->minor_hash,
+		   __entry->namelen, __get_str(name), __entry->num_used)
+);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_extend_dir);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_dx_dir_rebalance);
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_dx_dir_rebalance_split);
+
+DEFINE_OCFS2_ULL_INT_EVENT(ocfs2_prepare_dir_for_insert);
+
+/* End of trace events for fs/ocfs2/dir.c. */
+
+/* Trace events for fs/ocfs2/namei.c. */
+
+DECLARE_EVENT_CLASS(ocfs2__dentry_ops,
+	TP_PROTO(void *dir, void *dentry, int name_len, const char *name,
+		 unsigned long long dir_blkno, unsigned long long extra),
+	TP_ARGS(dir, dentry, name_len, name, dir_blkno, extra),
+	TP_STRUCT__entry(
+		__field(void *, dir)
+		__field(void *, dentry)
+		__field(int, name_len)
+		__string(name, name)
+		__field(unsigned long long, dir_blkno)
+		__field(unsigned long long, extra)
+	),
+	TP_fast_assign(
+		__entry->dir = dir;
+		__entry->dentry = dentry;
+		__entry->name_len = name_len;
+		__assign_str(name, name);
+		__entry->dir_blkno = dir_blkno;
+		__entry->extra = extra;
+	),
+	TP_printk("%p %p %.*s %llu %llu", __entry->dir, __entry->dentry,
+		  __entry->name_len, __get_str(name),
+		  __entry->dir_blkno, __entry->extra)
+);
+
+#define DEFINE_OCFS2_DENTRY_OPS(name)					\
+DEFINE_EVENT(ocfs2__dentry_ops, name,					\
+TP_PROTO(void *dir, void *dentry, int name_len, const char *name,	\
+	 unsigned long long dir_blkno, unsigned long long extra),	\
+	TP_ARGS(dir, dentry, name_len, name, dir_blkno, extra))
+
+DEFINE_OCFS2_DENTRY_OPS(ocfs2_lookup);
+
+DEFINE_OCFS2_DENTRY_OPS(ocfs2_mkdir);
+
+DEFINE_OCFS2_DENTRY_OPS(ocfs2_create);
+
+DEFINE_OCFS2_DENTRY_OPS(ocfs2_unlink);
+
+DEFINE_OCFS2_DENTRY_OPS(ocfs2_symlink_create);
+
+DEFINE_OCFS2_DENTRY_OPS(ocfs2_mv_orphaned_inode_to_new);
+
+DEFINE_OCFS2_POINTER_EVENT(ocfs2_lookup_ret);
+
+TRACE_EVENT(ocfs2_mknod,
+	TP_PROTO(void *dir, void *dentry, int name_len, const char *name,
+		 unsigned long long dir_blkno, unsigned long dev, int mode),
+	TP_ARGS(dir, dentry, name_len, name, dir_blkno, dev, mode),
+	TP_STRUCT__entry(
+		__field(void *, dir)
+		__field(void *, dentry)
+		__field(int, name_len)
+		__string(name, name)
+		__field(unsigned long long, dir_blkno)
+		__field(unsigned long, dev)
+		__field(int, mode)
+	),
+	TP_fast_assign(
+		__entry->dir = dir;
+		__entry->dentry = dentry;
+		__entry->name_len = name_len;
+		__assign_str(name, name);
+		__entry->dir_blkno = dir_blkno;
+		__entry->dev = dev;
+		__entry->mode = mode;
+	),
+	TP_printk("%p %p %.*s %llu %lu %d", __entry->dir, __entry->dentry,
+		  __entry->name_len, __get_str(name),
+		  __entry->dir_blkno, __entry->dev, __entry->mode)
+);
+
+TRACE_EVENT(ocfs2_link,
+	TP_PROTO(unsigned long long ino, int old_len, const char *old_name,
+		 int name_len, const char *name),
+	TP_ARGS(ino, old_len, old_name, name_len, name),
+	TP_STRUCT__entry(
+		__field(unsigned long long, ino)
+		__field(int, old_len)
+		__string(old_name, old_name)
+		__field(int, name_len)
+		__string(name, name)
+	),
+	TP_fast_assign(
+		__entry->ino = ino;
+		__entry->old_len = old_len;
+		__assign_str(old_name, old_name);
+		__entry->name_len = name_len;
+		__assign_str(name, name);
+	),
+	TP_printk("%llu %.*s %.*s", __entry->ino,
+		  __entry->old_len, __get_str(old_name),
+		  __entry->name_len, __get_str(name))
+);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_unlink_noent);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_double_lock);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_double_lock_end);
+
+TRACE_EVENT(ocfs2_rename,
+	TP_PROTO(void *old_dir, void *old_dentry,
+		 void *new_dir, void *new_dentry,
+		 int old_len, const char *old_name,
+		 int new_len, const char *new_name),
+	TP_ARGS(old_dir, old_dentry, new_dir, new_dentry,
+		old_len, old_name, new_len, new_name),
+	TP_STRUCT__entry(
+		__field(void *, old_dir)
+		__field(void *, old_dentry)
+		__field(void *, new_dir)
+		__field(void *, new_dentry)
+		__field(int, old_len)
+		__string(old_name, old_name)
+		__field(int, new_len)
+		__string(new_name, new_name)
+	),
+	TP_fast_assign(
+		__entry->old_dir = old_dir;
+		__entry->old_dentry = old_dentry;
+		__entry->new_dir = new_dir;
+		__entry->new_dentry = new_dentry;
+		__entry->old_len = old_len;
+		__assign_str(old_name, old_name);
+		__entry->new_len = new_len;
+		__assign_str(new_name, new_name);
+	),
+	TP_printk("%p %p %p %p %.*s %.*s",
+		  __entry->old_dir, __entry->old_dentry,
+		  __entry->new_dir, __entry->new_dentry,
+		  __entry->old_len, __get_str(old_name),
+		  __entry->new_len, __get_str(new_name))
+);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_rename_not_permitted);
+
+TRACE_EVENT(ocfs2_rename_target_exists,
+	TP_PROTO(int new_len, const char *new_name),
+	TP_ARGS(new_len, new_name),
+	TP_STRUCT__entry(
+		__field(int, new_len)
+		__string(new_name, new_name)
+	),
+	TP_fast_assign(
+		__entry->new_len = new_len;
+		__assign_str(new_name, new_name);
+	),
+	TP_printk("%.*s", __entry->new_len, __get_str(new_name))
+);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_rename_disagree);
+
+TRACE_EVENT(ocfs2_rename_over_existing,
+	TP_PROTO(unsigned long long new_blkno, void *new_bh,
+		 unsigned long long newdi_blkno),
+	TP_ARGS(new_blkno, new_bh, newdi_blkno),
+	TP_STRUCT__entry(
+		__field(unsigned long long, new_blkno)
+		__field(void *, new_bh)
+		__field(unsigned long long, newdi_blkno)
+	),
+	TP_fast_assign(
+		__entry->new_blkno = new_blkno;
+		__entry->new_bh = new_bh;
+		__entry->newdi_blkno = newdi_blkno;
+	),
+	TP_printk("%llu %p %llu", __entry->new_blkno, __entry->new_bh,
+		  __entry->newdi_blkno)
+);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_create_symlink_data);
+
+TRACE_EVENT(ocfs2_symlink_begin,
+	TP_PROTO(void *dir, void *dentry, const char *symname,
+		 int len, const char *name),
+	TP_ARGS(dir, dentry, symname, len, name),
+	TP_STRUCT__entry(
+		__field(void *, dir)
+		__field(void *, dentry)
+		__field(const char *, symname)
+		__field(int, len)
+		__string(name, name)
+	),
+	TP_fast_assign(
+		__entry->dir = dir;
+		__entry->dentry = dentry;
+		__entry->symname = symname;
+		__entry->len = len;
+		__assign_str(name, name);
+	),
+	TP_printk("%p %p %s %.*s", __entry->dir, __entry->dentry,
+		  __entry->symname, __entry->len, __get_str(name))
+);
+
+TRACE_EVENT(ocfs2_blkno_stringify,
+	TP_PROTO(unsigned long long blkno, const char *name, int namelen),
+	TP_ARGS(blkno, name, namelen),
+	TP_STRUCT__entry(
+		__field(unsigned long long, blkno)
+		__string(name, name)
+		__field(int, namelen)
+	),
+	TP_fast_assign(
+		__entry->blkno = blkno;
+		__assign_str(name, name);
+		__entry->namelen = namelen;
+	),
+	TP_printk("%llu %s %d", __entry->blkno, __get_str(name),
+		  __entry->namelen)
+);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_orphan_add_begin);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_orphan_add_end);
+
+TRACE_EVENT(ocfs2_orphan_del,
+	TP_PROTO(unsigned long long dir, const char *name, int namelen),
+	TP_ARGS(dir, name, namelen),
+	TP_STRUCT__entry(
+		__field(unsigned long long, dir)
+		__string(name, name)
+		__field(int, namelen)
+	),
+	TP_fast_assign(
+		__entry->dir = dir;
+		__assign_str(name, name);
+		__entry->namelen = namelen;
+	),
+	TP_printk("%llu %s %d", __entry->dir, __get_str(name),
+		  __entry->namelen)
+);
+
+/* End of trace events for fs/ocfs2/namei.c. */
+
+/* Trace events for fs/ocfs2/dcache.c. */
+
+TRACE_EVENT(ocfs2_dentry_revalidate,
+	TP_PROTO(void *dentry, int len, const char *name),
+	TP_ARGS(dentry, len, name),
+	TP_STRUCT__entry(
+		__field(void *, dentry)
+		__field(int, len)
+		__string(name, name)
+	),
+	TP_fast_assign(
+		__entry->dentry = dentry;
+		__entry->len = len;
+		__assign_str(name, name);
+	),
+	TP_printk("%p %.*s", __entry->dentry, __entry->len, __get_str(name))
+);
+
+TRACE_EVENT(ocfs2_dentry_revalidate_negative,
+	TP_PROTO(int len, const char *name, unsigned long pgen,
+		 unsigned long gen),
+	TP_ARGS(len, name, pgen, gen),
+	TP_STRUCT__entry(
+		__field(int, len)
+		__string(name, name)
+		__field(unsigned long, pgen)
+		__field(unsigned long, gen)
+	),
+	TP_fast_assign(
+		__entry->len = len;
+		__assign_str(name, name);
+		__entry->pgen = pgen;
+		__entry->gen = gen;
+	),
+	TP_printk("%.*s %lu %lu", __entry->len, __get_str(name),
+		  __entry->pgen, __entry->gen)
+);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_dentry_revalidate_delete);
+
+DEFINE_OCFS2_ULL_INT_EVENT(ocfs2_dentry_revalidate_orphaned);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_dentry_revalidate_nofsdata);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_dentry_revalidate_ret);
+
+TRACE_EVENT(ocfs2_find_local_alias,
+	TP_PROTO(int len, const char *name),
+	TP_ARGS(len, name),
+	TP_STRUCT__entry(
+		__field(int, len)
+		__string(name, name)
+	),
+	TP_fast_assign(
+		__entry->len = len;
+		__assign_str(name, name);
+	),
+	TP_printk("%.*s", __entry->len, __get_str(name))
+);
+
+TRACE_EVENT(ocfs2_dentry_attach_lock,
+	TP_PROTO(int len, const char *name,
+		 unsigned long long parent, void *fsdata),
+	TP_ARGS(len, name, parent, fsdata),
+	TP_STRUCT__entry(
+		__field(int, len)
+		__string(name, name)
+		__field(unsigned long long, parent)
+		__field(void *, fsdata)
+	),
+	TP_fast_assign(
+		__entry->len = len;
+		__assign_str(name, name);
+		__entry->parent = parent;
+		__entry->fsdata = fsdata;
+	),
+	TP_printk("%.*s %llu %p", __entry->len, __get_str(name),
+		  __entry->parent, __entry->fsdata)
+);
+
+TRACE_EVENT(ocfs2_dentry_attach_lock_found,
+	TP_PROTO(const char *name, unsigned long long parent,
+		 unsigned long long ino),
+	TP_ARGS(name, parent, ino),
+	TP_STRUCT__entry(
+		__string(name, name)
+		__field(unsigned long long, parent)
+		__field(unsigned long long, ino)
+	),
+	TP_fast_assign(
+		__assign_str(name, name);
+		__entry->parent = parent;
+		__entry->ino = ino;
+	),
+	TP_printk("%s %llu %llu", __get_str(name), __entry->parent, __entry->ino)
+);
+/* End of trace events for fs/ocfs2/dcache.c. */
+
+/* Trace events for fs/ocfs2/export.c. */
+
+TRACE_EVENT(ocfs2_get_dentry_begin,
+	TP_PROTO(void *sb, void *handle, unsigned long long blkno),
+	TP_ARGS(sb, handle, blkno),
+	TP_STRUCT__entry(
+		__field(void *, sb)
+		__field(void *, handle)
+		__field(unsigned long long, blkno)
+	),
+	TP_fast_assign(
+		__entry->sb = sb;
+		__entry->handle = handle;
+		__entry->blkno = blkno;
+	),
+	TP_printk("%p %p %llu", __entry->sb, __entry->handle, __entry->blkno)
+);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_get_dentry_test_bit);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_get_dentry_stale);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_get_dentry_generation);
+
+DEFINE_OCFS2_POINTER_EVENT(ocfs2_get_dentry_end);
+
+TRACE_EVENT(ocfs2_get_parent,
+	TP_PROTO(void *child, int len, const char *name,
+		 unsigned long long ino),
+	TP_ARGS(child, len, name, ino),
+	TP_STRUCT__entry(
+		__field(void *,	child)
+		__field(int, len)
+		__string(name, name)
+		__field(unsigned long long, ino)
+	),
+	TP_fast_assign(
+		__entry->child = child;
+		__entry->len = len;
+		__assign_str(name, name);
+		__entry->ino = ino;
+	),
+	TP_printk("%p %.*s %llu", __entry->child, __entry->len,
+		  __get_str(name), __entry->ino)
+);
+
+DEFINE_OCFS2_POINTER_EVENT(ocfs2_get_parent_end);
+
+TRACE_EVENT(ocfs2_encode_fh_begin,
+	TP_PROTO(void *dentry, int name_len, const char *name,
+		 void *fh, int len, int connectable),
+	TP_ARGS(dentry, name_len, name, fh, len, connectable),
+	TP_STRUCT__entry(
+		__field(void *, dentry)
+		__field(int, name_len)
+		__string(name, name)
+		__field(void *, fh)
+		__field(int, len)
+		__field(int, connectable)
+	),
+	TP_fast_assign(
+		__entry->dentry = dentry;
+		__entry->name_len = name_len;
+		__assign_str(name, name);
+		__entry->fh = fh;
+		__entry->len = len;
+		__entry->connectable = connectable;
+	),
+	TP_printk("%p %.*s %p %d %d", __entry->dentry, __entry->name_len,
+		  __get_str(name), __entry->fh, __entry->len,
+		  __entry->connectable)
+);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_encode_fh_self);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_encode_fh_parent);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_encode_fh_type);
+
+/* End of trace events for fs/ocfs2/export.c. */
+
+/* Trace events for fs/ocfs2/journal.c. */
+
+DEFINE_OCFS2_UINT_EVENT(ocfs2_commit_cache_begin);
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_commit_cache_end);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_extend_trans);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_extend_trans_restart);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_allocate_extend_trans);
+
+DEFINE_OCFS2_ULL_ULL_UINT_UINT_EVENT(ocfs2_journal_access);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_journal_dirty);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_journal_init);
+
+DEFINE_OCFS2_UINT_EVENT(ocfs2_journal_init_maxlen);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_journal_shutdown);
+
+DEFINE_OCFS2_POINTER_EVENT(ocfs2_journal_shutdown_wait);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_complete_recovery);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_complete_recovery_end);
+
+TRACE_EVENT(ocfs2_complete_recovery_slot,
+	TP_PROTO(int slot, unsigned long long la_ino,
+		 unsigned long long tl_ino, void *qrec),
+	TP_ARGS(slot, la_ino, tl_ino, qrec),
+	TP_STRUCT__entry(
+		__field(int, slot)
+		__field(unsigned long long, la_ino)
+		__field(unsigned long long, tl_ino)
+		__field(void *, qrec)
+	),
+	TP_fast_assign(
+		__entry->slot = slot;
+		__entry->la_ino = la_ino;
+		__entry->tl_ino = tl_ino;
+		__entry->qrec = qrec;
+	),
+	TP_printk("%d %llu %llu %p", __entry->slot, __entry->la_ino,
+		  __entry->tl_ino, __entry->qrec)
+);
+
+DEFINE_OCFS2_INT_INT_EVENT(ocfs2_recovery_thread_node);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_recovery_thread_end);
+
+TRACE_EVENT(ocfs2_recovery_thread,
+	TP_PROTO(int node_num, int osb_node_num, int disable,
+		 void *recovery_thread, int map_set),
+	TP_ARGS(node_num, osb_node_num, disable, recovery_thread, map_set),
+	TP_STRUCT__entry(
+		__field(int, node_num)
+		__field(int, osb_node_num)
+		__field(int,disable)
+		__field(void *, recovery_thread)
+		__field(int,map_set)
+	),
+	TP_fast_assign(
+		__entry->node_num = node_num;
+		__entry->osb_node_num = osb_node_num;
+		__entry->disable = disable;
+		__entry->recovery_thread = recovery_thread;
+		__entry->map_set = map_set;
+	),
+	TP_printk("%d %d %d %p %d", __entry->node_num,
+		   __entry->osb_node_num, __entry->disable,
+		   __entry->recovery_thread, __entry->map_set)
+);
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_replay_journal_recovered);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_replay_journal_lock_err);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_replay_journal_skip);
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_recover_node);
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_recover_node_skip);
+
+DEFINE_OCFS2_UINT_UINT_EVENT(ocfs2_mark_dead_nodes);
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_queue_orphan_scan_begin);
+
+DEFINE_OCFS2_UINT_UINT_UINT_EVENT(ocfs2_queue_orphan_scan_end);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_orphan_filldir);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_recover_orphans);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_recover_orphans_iput);
+
+DEFINE_OCFS2_INT_EVENT(ocfs2_wait_on_mount);
+
+/* End of trace events for fs/ocfs2/journal.c. */
+
+/* Trace events for fs/ocfs2/buffer_head_io.c. */
+
+DEFINE_OCFS2_ULL_UINT_EVENT(ocfs2_read_blocks_sync);
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_read_blocks_sync_jbd);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_read_blocks_from_disk);
+
+DEFINE_OCFS2_ULL_INT_INT_INT_EVENT(ocfs2_read_blocks_bh);
+
+DEFINE_OCFS2_ULL_INT_INT_INT_EVENT(ocfs2_read_blocks_end);
+
+TRACE_EVENT(ocfs2_write_block,
+	TP_PROTO(unsigned long long block, void *ci),
+	TP_ARGS(block, ci),
+	TP_STRUCT__entry(
+		__field(unsigned long long, block)
+		__field(void *, ci)
+	),
+	TP_fast_assign(
+		__entry->block = block;
+		__entry->ci = ci;
+	),
+	TP_printk("%llu %p", __entry->block, __entry->ci)
+);
+
+TRACE_EVENT(ocfs2_read_blocks_begin,
+	TP_PROTO(void *ci, unsigned long long block,
+		 unsigned int nr, int flags),
+	TP_ARGS(ci, block, nr, flags),
+	TP_STRUCT__entry(
+		__field(void *, ci)
+		__field(unsigned long long, block)
+		__field(unsigned int, nr)
+		__field(int, flags)
+	),
+	TP_fast_assign(
+		__entry->ci = ci;
+		__entry->block = block;
+		__entry->nr = nr;
+		__entry->flags = flags;
+	),
+	TP_printk("%p %llu %u %d", __entry->ci, __entry->block,
+		  __entry->nr, __entry->flags)
+);
+
+/* End of trace events for fs/ocfs2/buffer_head_io.c. */
+
+/* Trace events for fs/ocfs2/uptodate.c. */
+
+DEFINE_OCFS2_ULL_EVENT(ocfs2_purge_copied_metadata_tree);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_metadata_cache_purge);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_buffer_cached_begin);
+
+TRACE_EVENT(ocfs2_buffer_cached_end,
+	TP_PROTO(int index, void *item),
+	TP_ARGS(index, item),
+	TP_STRUCT__entry(
+		__field(int, index)
+		__field(void *, item)
+	),
+	TP_fast_assign(
+		__entry->index = index;
+		__entry->item = item;
+	),
+	TP_printk("%d %p", __entry->index, __entry->item)
+);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_append_cache_array);
+
+DEFINE_OCFS2_ULL_ULL_UINT_EVENT(ocfs2_insert_cache_tree);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_expand_cache);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_set_buffer_uptodate);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_set_buffer_uptodate_begin);
+
+DEFINE_OCFS2_ULL_UINT_UINT_EVENT(ocfs2_remove_metadata_array);
+
+DEFINE_OCFS2_ULL_ULL_EVENT(ocfs2_remove_metadata_tree);
+
+DEFINE_OCFS2_ULL_ULL_UINT_UINT_EVENT(ocfs2_remove_block_from_cache);
+
+/* End of trace events for fs/ocfs2/uptodate.c. */
+#endif /* _TRACE_OCFS2_H */
+
+/* This part must be outside protection */
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+#define TRACE_INCLUDE_FILE ocfs2_trace
+#include <trace/define_trace.h>
diff --git a/fs/ocfs2/quota.h b/fs/ocfs2/quota.h
new file mode 100644
index 000000000..b6d51333a
--- /dev/null
+++ b/fs/ocfs2/quota.h
@@ -0,0 +1,123 @@
+/*
+ * quota.h for OCFS2
+ *
+ * On disk quota structures for local and global quota file, in-memory
+ * structures.
+ *
+ */
+
+#ifndef _OCFS2_QUOTA_H
+#define _OCFS2_QUOTA_H
+
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/quota.h>
+#include <linux/list.h>
+#include <linux/dqblk_qtree.h>
+
+#include "ocfs2.h"
+
+/* Number of quota types we support */
+#define OCFS2_MAXQUOTAS 2
+
+/*
+ * In-memory structures
+ */
+struct ocfs2_dquot {
+	struct dquot dq_dquot;	/* Generic VFS dquot */
+	loff_t dq_local_off;	/* Offset in the local quota file */
+	u64 dq_local_phys_blk;	/* Physical block carrying quota structure */
+	struct ocfs2_quota_chunk *dq_chunk;	/* Chunk dquot is in */
+	unsigned int dq_use_count;	/* Number of nodes having reference to this entry in global quota file */
+	s64 dq_origspace;	/* Last globally synced space usage */
+	s64 dq_originodes;	/* Last globally synced inode usage */
+	struct llist_node list;	/* Member of list of dquots to drop */
+};
+
+/* Description of one chunk to recover in memory */
+struct ocfs2_recovery_chunk {
+	struct list_head rc_list;	/* List of chunks */
+	int rc_chunk;			/* Chunk number */
+	unsigned long *rc_bitmap;	/* Bitmap of entries to recover */
+};
+
+struct ocfs2_quota_recovery {
+	struct list_head r_list[OCFS2_MAXQUOTAS];	/* List of chunks to recover */
+};
+
+/* In-memory structure with quota header information */
+struct ocfs2_mem_dqinfo {
+	unsigned int dqi_type;		/* Quota type this structure describes */
+	unsigned int dqi_flags;		/* Flags OLQF_* */
+	unsigned int dqi_chunks;	/* Number of chunks in local quota file */
+	unsigned int dqi_blocks;	/* Number of blocks allocated for local quota file */
+	unsigned int dqi_syncms;	/* How often should we sync with other nodes */
+	struct list_head dqi_chunk;	/* List of chunks */
+	struct inode *dqi_gqinode;	/* Global quota file inode */
+	struct ocfs2_lock_res dqi_gqlock;	/* Lock protecting quota information structure */
+	struct buffer_head *dqi_gqi_bh;	/* Buffer head with global quota file inode - set only if inode lock is obtained */
+	int dqi_gqi_count;		/* Number of holders of dqi_gqi_bh */
+	u64 dqi_giblk;			/* Number of block with global information header */
+	struct buffer_head *dqi_lqi_bh;	/* Buffer head with local quota file inode */
+	struct buffer_head *dqi_libh;	/* Buffer with local information header */
+	struct qtree_mem_dqinfo dqi_gi;	/* Info about global file */
+	struct delayed_work dqi_sync_work;	/* Work for syncing dquots */
+	struct ocfs2_quota_recovery *dqi_rec;	/* Pointer to recovery
+						 * information, in case we
+						 * enable quotas on file
+						 * needing it */
+};
+
+static inline struct ocfs2_dquot *OCFS2_DQUOT(struct dquot *dquot)
+{
+	return container_of(dquot, struct ocfs2_dquot, dq_dquot);
+}
+
+struct ocfs2_quota_chunk {
+	struct list_head qc_chunk;	/* List of quotafile chunks */
+	int qc_num;			/* Number of quota chunk */
+	struct buffer_head *qc_headerbh;	/* Buffer head with chunk header */
+};
+
+extern struct kmem_cache *ocfs2_dquot_cachep;
+extern struct kmem_cache *ocfs2_qf_chunk_cachep;
+
+extern struct qtree_fmt_operations ocfs2_global_ops;
+
+struct ocfs2_quota_recovery *ocfs2_begin_quota_recovery(
+				struct ocfs2_super *osb, int slot_num);
+int ocfs2_finish_quota_recovery(struct ocfs2_super *osb,
+				struct ocfs2_quota_recovery *rec,
+				int slot_num);
+void ocfs2_free_quota_recovery(struct ocfs2_quota_recovery *rec);
+ssize_t ocfs2_quota_read(struct super_block *sb, int type, char *data,
+			 size_t len, loff_t off);
+ssize_t ocfs2_quota_write(struct super_block *sb, int type,
+			  const char *data, size_t len, loff_t off);
+int ocfs2_global_read_info(struct super_block *sb, int type);
+int ocfs2_global_write_info(struct super_block *sb, int type);
+int ocfs2_global_read_dquot(struct dquot *dquot);
+int __ocfs2_sync_dquot(struct dquot *dquot, int freeing);
+static inline int ocfs2_sync_dquot(struct dquot *dquot)
+{
+	return __ocfs2_sync_dquot(dquot, 0);
+}
+static inline int ocfs2_global_release_dquot(struct dquot *dquot)
+{
+	return __ocfs2_sync_dquot(dquot, 1);
+}
+
+int ocfs2_lock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex);
+void ocfs2_unlock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex);
+int ocfs2_validate_quota_block(struct super_block *sb, struct buffer_head *bh);
+int ocfs2_read_quota_phys_block(struct inode *inode, u64 p_block,
+				struct buffer_head **bh);
+int ocfs2_create_local_dquot(struct dquot *dquot);
+int ocfs2_local_release_dquot(handle_t *handle, struct dquot *dquot);
+int ocfs2_local_write_dquot(struct dquot *dquot);
+void ocfs2_drop_dquot_refs(struct work_struct *work);
+
+extern const struct dquot_operations ocfs2_quota_operations;
+extern struct quota_format_type ocfs2_quota_format;
+
+#endif /* _OCFS2_QUOTA_H */
diff --git a/fs/ocfs2/quota_global.c b/fs/ocfs2/quota_global.c
new file mode 100644
index 000000000..c93d67220
--- /dev/null
+++ b/fs/ocfs2/quota_global.c
@@ -0,0 +1,971 @@
+/*
+ *  Implementation of operations over global quota file
+ */
+#include <linux/spinlock.h>
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include <linux/quota.h>
+#include <linux/quotaops.h>
+#include <linux/dqblk_qtree.h>
+#include <linux/jiffies.h>
+#include <linux/writeback.h>
+#include <linux/workqueue.h>
+#include <linux/llist.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2_fs.h"
+#include "ocfs2.h"
+#include "alloc.h"
+#include "blockcheck.h"
+#include "inode.h"
+#include "journal.h"
+#include "file.h"
+#include "sysfile.h"
+#include "dlmglue.h"
+#include "uptodate.h"
+#include "super.h"
+#include "buffer_head_io.h"
+#include "quota.h"
+#include "ocfs2_trace.h"
+
+/*
+ * Locking of quotas with OCFS2 is rather complex. Here are rules that
+ * should be obeyed by all the functions:
+ * - any write of quota structure (either to local or global file) is protected
+ *   by dqio_mutex or dquot->dq_lock.
+ * - any modification of global quota file holds inode cluster lock, i_mutex,
+ *   and ip_alloc_sem of the global quota file (achieved by
+ *   ocfs2_lock_global_qf). It also has to hold qinfo_lock.
+ * - an allocation of new blocks for local quota file is protected by
+ *   its ip_alloc_sem
+ *
+ * A rough sketch of locking dependencies (lf = local file, gf = global file):
+ * Normal filesystem operation:
+ *   start_trans -> dqio_mutex -> write to lf
+ * Syncing of local and global file:
+ *   ocfs2_lock_global_qf -> start_trans -> dqio_mutex -> qinfo_lock ->
+ *     write to gf
+ *						       -> write to lf
+ * Acquire dquot for the first time:
+ *   dq_lock -> ocfs2_lock_global_qf -> qinfo_lock -> read from gf
+ *				     -> alloc space for gf
+ *				     -> start_trans -> qinfo_lock -> write to gf
+ *	     -> ip_alloc_sem of lf -> alloc space for lf
+ *	     -> write to lf
+ * Release last reference to dquot:
+ *   dq_lock -> ocfs2_lock_global_qf -> start_trans -> qinfo_lock -> write to gf
+ *	     -> write to lf
+ * Note that all the above operations also hold the inode cluster lock of lf.
+ * Recovery:
+ *   inode cluster lock of recovered lf
+ *     -> read bitmaps -> ip_alloc_sem of lf
+ *     -> ocfs2_lock_global_qf -> start_trans -> dqio_mutex -> qinfo_lock ->
+ *        write to gf
+ */
+
+static void qsync_work_fn(struct work_struct *work);
+
+static void ocfs2_global_disk2memdqb(struct dquot *dquot, void *dp)
+{
+	struct ocfs2_global_disk_dqblk *d = dp;
+	struct mem_dqblk *m = &dquot->dq_dqb;
+
+	/* Update from disk only entries not set by the admin */
+	if (!test_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags)) {
+		m->dqb_ihardlimit = le64_to_cpu(d->dqb_ihardlimit);
+		m->dqb_isoftlimit = le64_to_cpu(d->dqb_isoftlimit);
+	}
+	if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
+		m->dqb_curinodes = le64_to_cpu(d->dqb_curinodes);
+	if (!test_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags)) {
+		m->dqb_bhardlimit = le64_to_cpu(d->dqb_bhardlimit);
+		m->dqb_bsoftlimit = le64_to_cpu(d->dqb_bsoftlimit);
+	}
+	if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
+		m->dqb_curspace = le64_to_cpu(d->dqb_curspace);
+	if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags))
+		m->dqb_btime = le64_to_cpu(d->dqb_btime);
+	if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags))
+		m->dqb_itime = le64_to_cpu(d->dqb_itime);
+	OCFS2_DQUOT(dquot)->dq_use_count = le32_to_cpu(d->dqb_use_count);
+}
+
+static void ocfs2_global_mem2diskdqb(void *dp, struct dquot *dquot)
+{
+	struct ocfs2_global_disk_dqblk *d = dp;
+	struct mem_dqblk *m = &dquot->dq_dqb;
+
+	d->dqb_id = cpu_to_le32(from_kqid(&init_user_ns, dquot->dq_id));
+	d->dqb_use_count = cpu_to_le32(OCFS2_DQUOT(dquot)->dq_use_count);
+	d->dqb_ihardlimit = cpu_to_le64(m->dqb_ihardlimit);
+	d->dqb_isoftlimit = cpu_to_le64(m->dqb_isoftlimit);
+	d->dqb_curinodes = cpu_to_le64(m->dqb_curinodes);
+	d->dqb_bhardlimit = cpu_to_le64(m->dqb_bhardlimit);
+	d->dqb_bsoftlimit = cpu_to_le64(m->dqb_bsoftlimit);
+	d->dqb_curspace = cpu_to_le64(m->dqb_curspace);
+	d->dqb_btime = cpu_to_le64(m->dqb_btime);
+	d->dqb_itime = cpu_to_le64(m->dqb_itime);
+	d->dqb_pad1 = d->dqb_pad2 = 0;
+}
+
+static int ocfs2_global_is_id(void *dp, struct dquot *dquot)
+{
+	struct ocfs2_global_disk_dqblk *d = dp;
+	struct ocfs2_mem_dqinfo *oinfo =
+			sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
+
+	if (qtree_entry_unused(&oinfo->dqi_gi, dp))
+		return 0;
+
+	return qid_eq(make_kqid(&init_user_ns, dquot->dq_id.type,
+				le32_to_cpu(d->dqb_id)),
+		      dquot->dq_id);
+}
+
+struct qtree_fmt_operations ocfs2_global_ops = {
+	.mem2disk_dqblk = ocfs2_global_mem2diskdqb,
+	.disk2mem_dqblk = ocfs2_global_disk2memdqb,
+	.is_id = ocfs2_global_is_id,
+};
+
+int ocfs2_validate_quota_block(struct super_block *sb, struct buffer_head *bh)
+{
+	struct ocfs2_disk_dqtrailer *dqt =
+		ocfs2_block_dqtrailer(sb->s_blocksize, bh->b_data);
+
+	trace_ocfs2_validate_quota_block((unsigned long long)bh->b_blocknr);
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	/*
+	 * If the ecc fails, we return the error but otherwise
+	 * leave the filesystem running.  We know any error is
+	 * local to this block.
+	 */
+	return ocfs2_validate_meta_ecc(sb, bh->b_data, &dqt->dq_check);
+}
+
+int ocfs2_read_quota_phys_block(struct inode *inode, u64 p_block,
+				struct buffer_head **bhp)
+{
+	int rc;
+
+	*bhp = NULL;
+	rc = ocfs2_read_blocks(INODE_CACHE(inode), p_block, 1, bhp, 0,
+			       ocfs2_validate_quota_block);
+	if (rc)
+		mlog_errno(rc);
+	return rc;
+}
+
+/* Read data from global quotafile - avoid pagecache and such because we cannot
+ * afford acquiring the locks... We use quota cluster lock to serialize
+ * operations. Caller is responsible for acquiring it. */
+ssize_t ocfs2_quota_read(struct super_block *sb, int type, char *data,
+			 size_t len, loff_t off)
+{
+	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
+	struct inode *gqinode = oinfo->dqi_gqinode;
+	loff_t i_size = i_size_read(gqinode);
+	int offset = off & (sb->s_blocksize - 1);
+	sector_t blk = off >> sb->s_blocksize_bits;
+	int err = 0;
+	struct buffer_head *bh;
+	size_t toread, tocopy;
+	u64 pblock = 0, pcount = 0;
+
+	if (off > i_size)
+		return 0;
+	if (off + len > i_size)
+		len = i_size - off;
+	toread = len;
+	while (toread > 0) {
+		tocopy = min_t(size_t, (sb->s_blocksize - offset), toread);
+		if (!pcount) {
+			err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock,
+							  &pcount, NULL);
+			if (err) {
+				mlog_errno(err);
+				return err;
+			}
+		} else {
+			pcount--;
+			pblock++;
+		}
+		bh = NULL;
+		err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
+		if (err) {
+			mlog_errno(err);
+			return err;
+		}
+		memcpy(data, bh->b_data + offset, tocopy);
+		brelse(bh);
+		offset = 0;
+		toread -= tocopy;
+		data += tocopy;
+		blk++;
+	}
+	return len;
+}
+
+/* Write to quotafile (we know the transaction is already started and has
+ * enough credits) */
+ssize_t ocfs2_quota_write(struct super_block *sb, int type,
+			  const char *data, size_t len, loff_t off)
+{
+	struct mem_dqinfo *info = sb_dqinfo(sb, type);
+	struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
+	struct inode *gqinode = oinfo->dqi_gqinode;
+	int offset = off & (sb->s_blocksize - 1);
+	sector_t blk = off >> sb->s_blocksize_bits;
+	int err = 0, new = 0, ja_type;
+	struct buffer_head *bh = NULL;
+	handle_t *handle = journal_current_handle();
+	u64 pblock, pcount;
+
+	if (!handle) {
+		mlog(ML_ERROR, "Quota write (off=%llu, len=%llu) cancelled "
+		     "because transaction was not started.\n",
+		     (unsigned long long)off, (unsigned long long)len);
+		return -EIO;
+	}
+	if (len > sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset) {
+		WARN_ON(1);
+		len = sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE - offset;
+	}
+
+	if (i_size_read(gqinode) < off + len) {
+		loff_t rounded_end =
+				ocfs2_align_bytes_to_blocks(sb, off + len);
+
+		/* Space is already allocated in ocfs2_acquire_dquot() */
+		err = ocfs2_simple_size_update(gqinode,
+					       oinfo->dqi_gqi_bh,
+					       rounded_end);
+		if (err < 0)
+			goto out;
+		new = 1;
+	}
+	err = ocfs2_extent_map_get_blocks(gqinode, blk, &pblock, &pcount, NULL);
+	if (err) {
+		mlog_errno(err);
+		goto out;
+	}
+	/* Not rewriting whole block? */
+	if ((offset || len < sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE) &&
+	    !new) {
+		err = ocfs2_read_quota_phys_block(gqinode, pblock, &bh);
+		ja_type = OCFS2_JOURNAL_ACCESS_WRITE;
+	} else {
+		bh = sb_getblk(sb, pblock);
+		if (!bh)
+			err = -ENOMEM;
+		ja_type = OCFS2_JOURNAL_ACCESS_CREATE;
+	}
+	if (err) {
+		mlog_errno(err);
+		goto out;
+	}
+	lock_buffer(bh);
+	if (new)
+		memset(bh->b_data, 0, sb->s_blocksize);
+	memcpy(bh->b_data + offset, data, len);
+	flush_dcache_page(bh->b_page);
+	set_buffer_uptodate(bh);
+	unlock_buffer(bh);
+	ocfs2_set_buffer_uptodate(INODE_CACHE(gqinode), bh);
+	err = ocfs2_journal_access_dq(handle, INODE_CACHE(gqinode), bh,
+				      ja_type);
+	if (err < 0) {
+		brelse(bh);
+		goto out;
+	}
+	ocfs2_journal_dirty(handle, bh);
+	brelse(bh);
+out:
+	if (err) {
+		mlog_errno(err);
+		return err;
+	}
+	gqinode->i_version++;
+	ocfs2_mark_inode_dirty(handle, gqinode, oinfo->dqi_gqi_bh);
+	return len;
+}
+
+int ocfs2_lock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
+{
+	int status;
+	struct buffer_head *bh = NULL;
+
+	status = ocfs2_inode_lock(oinfo->dqi_gqinode, &bh, ex);
+	if (status < 0)
+		return status;
+	spin_lock(&dq_data_lock);
+	if (!oinfo->dqi_gqi_count++)
+		oinfo->dqi_gqi_bh = bh;
+	else
+		WARN_ON(bh != oinfo->dqi_gqi_bh);
+	spin_unlock(&dq_data_lock);
+	if (ex) {
+		mutex_lock(&oinfo->dqi_gqinode->i_mutex);
+		down_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
+	} else {
+		down_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
+	}
+	return 0;
+}
+
+void ocfs2_unlock_global_qf(struct ocfs2_mem_dqinfo *oinfo, int ex)
+{
+	if (ex) {
+		up_write(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
+		mutex_unlock(&oinfo->dqi_gqinode->i_mutex);
+	} else {
+		up_read(&OCFS2_I(oinfo->dqi_gqinode)->ip_alloc_sem);
+	}
+	ocfs2_inode_unlock(oinfo->dqi_gqinode, ex);
+	brelse(oinfo->dqi_gqi_bh);
+	spin_lock(&dq_data_lock);
+	if (!--oinfo->dqi_gqi_count)
+		oinfo->dqi_gqi_bh = NULL;
+	spin_unlock(&dq_data_lock);
+}
+
+/* Read information header from global quota file */
+int ocfs2_global_read_info(struct super_block *sb, int type)
+{
+	struct inode *gqinode = NULL;
+	unsigned int ino[OCFS2_MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
+					      GROUP_QUOTA_SYSTEM_INODE };
+	struct ocfs2_global_disk_dqinfo dinfo;
+	struct mem_dqinfo *info = sb_dqinfo(sb, type);
+	struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
+	u64 pcount;
+	int status;
+
+	/* Read global header */
+	gqinode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
+			OCFS2_INVALID_SLOT);
+	if (!gqinode) {
+		mlog(ML_ERROR, "failed to get global quota inode (type=%d)\n",
+			type);
+		status = -EINVAL;
+		goto out_err;
+	}
+	oinfo->dqi_gi.dqi_sb = sb;
+	oinfo->dqi_gi.dqi_type = type;
+	ocfs2_qinfo_lock_res_init(&oinfo->dqi_gqlock, oinfo);
+	oinfo->dqi_gi.dqi_entry_size = sizeof(struct ocfs2_global_disk_dqblk);
+	oinfo->dqi_gi.dqi_ops = &ocfs2_global_ops;
+	oinfo->dqi_gqi_bh = NULL;
+	oinfo->dqi_gqi_count = 0;
+	oinfo->dqi_gqinode = gqinode;
+	status = ocfs2_lock_global_qf(oinfo, 0);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_err;
+	}
+
+	status = ocfs2_extent_map_get_blocks(gqinode, 0, &oinfo->dqi_giblk,
+					     &pcount, NULL);
+	if (status < 0)
+		goto out_unlock;
+
+	status = ocfs2_qinfo_lock(oinfo, 0);
+	if (status < 0)
+		goto out_unlock;
+	status = sb->s_op->quota_read(sb, type, (char *)&dinfo,
+				      sizeof(struct ocfs2_global_disk_dqinfo),
+				      OCFS2_GLOBAL_INFO_OFF);
+	ocfs2_qinfo_unlock(oinfo, 0);
+	ocfs2_unlock_global_qf(oinfo, 0);
+	if (status != sizeof(struct ocfs2_global_disk_dqinfo)) {
+		mlog(ML_ERROR, "Cannot read global quota info (%d).\n",
+		     status);
+		if (status >= 0)
+			status = -EIO;
+		mlog_errno(status);
+		goto out_err;
+	}
+	info->dqi_bgrace = le32_to_cpu(dinfo.dqi_bgrace);
+	info->dqi_igrace = le32_to_cpu(dinfo.dqi_igrace);
+	oinfo->dqi_syncms = le32_to_cpu(dinfo.dqi_syncms);
+	oinfo->dqi_gi.dqi_blocks = le32_to_cpu(dinfo.dqi_blocks);
+	oinfo->dqi_gi.dqi_free_blk = le32_to_cpu(dinfo.dqi_free_blk);
+	oinfo->dqi_gi.dqi_free_entry = le32_to_cpu(dinfo.dqi_free_entry);
+	oinfo->dqi_gi.dqi_blocksize_bits = sb->s_blocksize_bits;
+	oinfo->dqi_gi.dqi_usable_bs = sb->s_blocksize -
+						OCFS2_QBLK_RESERVED_SPACE;
+	oinfo->dqi_gi.dqi_qtree_depth = qtree_depth(&oinfo->dqi_gi);
+	INIT_DELAYED_WORK(&oinfo->dqi_sync_work, qsync_work_fn);
+	schedule_delayed_work(&oinfo->dqi_sync_work,
+			      msecs_to_jiffies(oinfo->dqi_syncms));
+
+out_err:
+	return status;
+out_unlock:
+	ocfs2_unlock_global_qf(oinfo, 0);
+	mlog_errno(status);
+	goto out_err;
+}
+
+/* Write information to global quota file. Expects exlusive lock on quota
+ * file inode and quota info */
+static int __ocfs2_global_write_info(struct super_block *sb, int type)
+{
+	struct mem_dqinfo *info = sb_dqinfo(sb, type);
+	struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
+	struct ocfs2_global_disk_dqinfo dinfo;
+	ssize_t size;
+
+	spin_lock(&dq_data_lock);
+	info->dqi_flags &= ~DQF_INFO_DIRTY;
+	dinfo.dqi_bgrace = cpu_to_le32(info->dqi_bgrace);
+	dinfo.dqi_igrace = cpu_to_le32(info->dqi_igrace);
+	spin_unlock(&dq_data_lock);
+	dinfo.dqi_syncms = cpu_to_le32(oinfo->dqi_syncms);
+	dinfo.dqi_blocks = cpu_to_le32(oinfo->dqi_gi.dqi_blocks);
+	dinfo.dqi_free_blk = cpu_to_le32(oinfo->dqi_gi.dqi_free_blk);
+	dinfo.dqi_free_entry = cpu_to_le32(oinfo->dqi_gi.dqi_free_entry);
+	size = sb->s_op->quota_write(sb, type, (char *)&dinfo,
+				     sizeof(struct ocfs2_global_disk_dqinfo),
+				     OCFS2_GLOBAL_INFO_OFF);
+	if (size != sizeof(struct ocfs2_global_disk_dqinfo)) {
+		mlog(ML_ERROR, "Cannot write global quota info structure\n");
+		if (size >= 0)
+			size = -EIO;
+		return size;
+	}
+	return 0;
+}
+
+int ocfs2_global_write_info(struct super_block *sb, int type)
+{
+	int err;
+	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
+
+	err = ocfs2_qinfo_lock(info, 1);
+	if (err < 0)
+		return err;
+	err = __ocfs2_global_write_info(sb, type);
+	ocfs2_qinfo_unlock(info, 1);
+	return err;
+}
+
+static int ocfs2_global_qinit_alloc(struct super_block *sb, int type)
+{
+	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
+
+	/*
+	 * We may need to allocate tree blocks and a leaf block but not the
+	 * root block
+	 */
+	return oinfo->dqi_gi.dqi_qtree_depth;
+}
+
+static int ocfs2_calc_global_qinit_credits(struct super_block *sb, int type)
+{
+	/* We modify all the allocated blocks, tree root, info block and
+	 * the inode */
+	return (ocfs2_global_qinit_alloc(sb, type) + 2) *
+			OCFS2_QUOTA_BLOCK_UPDATE_CREDITS + 1;
+}
+
+/* Sync local information about quota modifications with global quota file.
+ * Caller must have started the transaction and obtained exclusive lock for
+ * global quota file inode */
+int __ocfs2_sync_dquot(struct dquot *dquot, int freeing)
+{
+	int err, err2;
+	struct super_block *sb = dquot->dq_sb;
+	int type = dquot->dq_id.type;
+	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
+	struct ocfs2_global_disk_dqblk dqblk;
+	s64 spacechange, inodechange;
+	time_t olditime, oldbtime;
+
+	err = sb->s_op->quota_read(sb, type, (char *)&dqblk,
+				   sizeof(struct ocfs2_global_disk_dqblk),
+				   dquot->dq_off);
+	if (err != sizeof(struct ocfs2_global_disk_dqblk)) {
+		if (err >= 0) {
+			mlog(ML_ERROR, "Short read from global quota file "
+				       "(%u read)\n", err);
+			err = -EIO;
+		}
+		goto out;
+	}
+
+	/* Update space and inode usage. Get also other information from
+	 * global quota file so that we don't overwrite any changes there.
+	 * We are */
+	spin_lock(&dq_data_lock);
+	spacechange = dquot->dq_dqb.dqb_curspace -
+					OCFS2_DQUOT(dquot)->dq_origspace;
+	inodechange = dquot->dq_dqb.dqb_curinodes -
+					OCFS2_DQUOT(dquot)->dq_originodes;
+	olditime = dquot->dq_dqb.dqb_itime;
+	oldbtime = dquot->dq_dqb.dqb_btime;
+	ocfs2_global_disk2memdqb(dquot, &dqblk);
+	trace_ocfs2_sync_dquot(from_kqid(&init_user_ns, dquot->dq_id),
+			       dquot->dq_dqb.dqb_curspace,
+			       (long long)spacechange,
+			       dquot->dq_dqb.dqb_curinodes,
+			       (long long)inodechange);
+	if (!test_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags))
+		dquot->dq_dqb.dqb_curspace += spacechange;
+	if (!test_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags))
+		dquot->dq_dqb.dqb_curinodes += inodechange;
+	/* Set properly space grace time... */
+	if (dquot->dq_dqb.dqb_bsoftlimit &&
+	    dquot->dq_dqb.dqb_curspace > dquot->dq_dqb.dqb_bsoftlimit) {
+		if (!test_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags) &&
+		    oldbtime > 0) {
+			if (dquot->dq_dqb.dqb_btime > 0)
+				dquot->dq_dqb.dqb_btime =
+					min(dquot->dq_dqb.dqb_btime, oldbtime);
+			else
+				dquot->dq_dqb.dqb_btime = oldbtime;
+		}
+	} else {
+		dquot->dq_dqb.dqb_btime = 0;
+		clear_bit(DQ_BLKS_B, &dquot->dq_flags);
+	}
+	/* Set properly inode grace time... */
+	if (dquot->dq_dqb.dqb_isoftlimit &&
+	    dquot->dq_dqb.dqb_curinodes > dquot->dq_dqb.dqb_isoftlimit) {
+		if (!test_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags) &&
+		    olditime > 0) {
+			if (dquot->dq_dqb.dqb_itime > 0)
+				dquot->dq_dqb.dqb_itime =
+					min(dquot->dq_dqb.dqb_itime, olditime);
+			else
+				dquot->dq_dqb.dqb_itime = olditime;
+		}
+	} else {
+		dquot->dq_dqb.dqb_itime = 0;
+		clear_bit(DQ_INODES_B, &dquot->dq_flags);
+	}
+	/* All information is properly updated, clear the flags */
+	__clear_bit(DQ_LASTSET_B + QIF_SPACE_B, &dquot->dq_flags);
+	__clear_bit(DQ_LASTSET_B + QIF_INODES_B, &dquot->dq_flags);
+	__clear_bit(DQ_LASTSET_B + QIF_BLIMITS_B, &dquot->dq_flags);
+	__clear_bit(DQ_LASTSET_B + QIF_ILIMITS_B, &dquot->dq_flags);
+	__clear_bit(DQ_LASTSET_B + QIF_BTIME_B, &dquot->dq_flags);
+	__clear_bit(DQ_LASTSET_B + QIF_ITIME_B, &dquot->dq_flags);
+	OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
+	OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
+	spin_unlock(&dq_data_lock);
+	err = ocfs2_qinfo_lock(info, freeing);
+	if (err < 0) {
+		mlog(ML_ERROR, "Failed to lock quota info, losing quota write"
+			       " (type=%d, id=%u)\n", dquot->dq_id.type,
+			       (unsigned)from_kqid(&init_user_ns, dquot->dq_id));
+		goto out;
+	}
+	if (freeing)
+		OCFS2_DQUOT(dquot)->dq_use_count--;
+	err = qtree_write_dquot(&info->dqi_gi, dquot);
+	if (err < 0)
+		goto out_qlock;
+	if (freeing && !OCFS2_DQUOT(dquot)->dq_use_count) {
+		err = qtree_release_dquot(&info->dqi_gi, dquot);
+		if (info_dirty(sb_dqinfo(sb, type))) {
+			err2 = __ocfs2_global_write_info(sb, type);
+			if (!err)
+				err = err2;
+		}
+	}
+out_qlock:
+	ocfs2_qinfo_unlock(info, freeing);
+out:
+	if (err < 0)
+		mlog_errno(err);
+	return err;
+}
+
+/*
+ *  Functions for periodic syncing of dquots with global file
+ */
+static int ocfs2_sync_dquot_helper(struct dquot *dquot, unsigned long type)
+{
+	handle_t *handle;
+	struct super_block *sb = dquot->dq_sb;
+	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	int status = 0;
+
+	trace_ocfs2_sync_dquot_helper(from_kqid(&init_user_ns, dquot->dq_id),
+				      dquot->dq_id.type,
+				      type, sb->s_id);
+	if (type != dquot->dq_id.type)
+		goto out;
+	status = ocfs2_lock_global_qf(oinfo, 1);
+	if (status < 0)
+		goto out;
+
+	handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out_ilock;
+	}
+	mutex_lock(&sb_dqopt(sb)->dqio_mutex);
+	status = ocfs2_sync_dquot(dquot);
+	if (status < 0)
+		mlog_errno(status);
+	/* We have to write local structure as well... */
+	status = ocfs2_local_write_dquot(dquot);
+	if (status < 0)
+		mlog_errno(status);
+	mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
+	ocfs2_commit_trans(osb, handle);
+out_ilock:
+	ocfs2_unlock_global_qf(oinfo, 1);
+out:
+	return status;
+}
+
+static void qsync_work_fn(struct work_struct *work)
+{
+	struct ocfs2_mem_dqinfo *oinfo = container_of(work,
+						      struct ocfs2_mem_dqinfo,
+						      dqi_sync_work.work);
+	struct super_block *sb = oinfo->dqi_gqinode->i_sb;
+
+	dquot_scan_active(sb, ocfs2_sync_dquot_helper, oinfo->dqi_type);
+	schedule_delayed_work(&oinfo->dqi_sync_work,
+			      msecs_to_jiffies(oinfo->dqi_syncms));
+}
+
+/*
+ *  Wrappers for generic quota functions
+ */
+
+static int ocfs2_write_dquot(struct dquot *dquot)
+{
+	handle_t *handle;
+	struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
+	int status = 0;
+
+	trace_ocfs2_write_dquot(from_kqid(&init_user_ns, dquot->dq_id),
+				dquot->dq_id.type);
+
+	handle = ocfs2_start_trans(osb, OCFS2_QWRITE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out;
+	}
+	mutex_lock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
+	status = ocfs2_local_write_dquot(dquot);
+	mutex_unlock(&sb_dqopt(dquot->dq_sb)->dqio_mutex);
+	ocfs2_commit_trans(osb, handle);
+out:
+	return status;
+}
+
+static int ocfs2_calc_qdel_credits(struct super_block *sb, int type)
+{
+	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
+	/*
+	 * We modify tree, leaf block, global info, local chunk header,
+	 * global and local inode; OCFS2_QINFO_WRITE_CREDITS already
+	 * accounts for inode update
+	 */
+	return (oinfo->dqi_gi.dqi_qtree_depth + 2) *
+	       OCFS2_QUOTA_BLOCK_UPDATE_CREDITS +
+	       OCFS2_QINFO_WRITE_CREDITS +
+	       OCFS2_INODE_UPDATE_CREDITS;
+}
+
+void ocfs2_drop_dquot_refs(struct work_struct *work)
+{
+	struct ocfs2_super *osb = container_of(work, struct ocfs2_super,
+					       dquot_drop_work);
+	struct llist_node *list;
+	struct ocfs2_dquot *odquot, *next_odquot;
+
+	list = llist_del_all(&osb->dquot_drop_list);
+	llist_for_each_entry_safe(odquot, next_odquot, list, list) {
+		/* Drop the reference we acquired in ocfs2_dquot_release() */
+		dqput(&odquot->dq_dquot);
+	}
+}
+
+/*
+ * Called when the last reference to dquot is dropped. If we are called from
+ * downconvert thread, we cannot do all the handling here because grabbing
+ * quota lock could deadlock (the node holding the quota lock could need some
+ * other cluster lock to proceed but with blocked downconvert thread we cannot
+ * release any lock).
+ */
+static int ocfs2_release_dquot(struct dquot *dquot)
+{
+	handle_t *handle;
+	struct ocfs2_mem_dqinfo *oinfo =
+			sb_dqinfo(dquot->dq_sb, dquot->dq_id.type)->dqi_priv;
+	struct ocfs2_super *osb = OCFS2_SB(dquot->dq_sb);
+	int status = 0;
+
+	trace_ocfs2_release_dquot(from_kqid(&init_user_ns, dquot->dq_id),
+				  dquot->dq_id.type);
+
+	mutex_lock(&dquot->dq_lock);
+	/* Check whether we are not racing with some other dqget() */
+	if (atomic_read(&dquot->dq_count) > 1)
+		goto out;
+	/* Running from downconvert thread? Postpone quota processing to wq */
+	if (current == osb->dc_task) {
+		/*
+		 * Grab our own reference to dquot and queue it for delayed
+		 * dropping.  Quota code rechecks after calling
+		 * ->release_dquot() and won't free dquot structure.
+		 */
+		dqgrab(dquot);
+		/* First entry on list -> queue work */
+		if (llist_add(&OCFS2_DQUOT(dquot)->list, &osb->dquot_drop_list))
+			queue_work(ocfs2_wq, &osb->dquot_drop_work);
+		goto out;
+	}
+	status = ocfs2_lock_global_qf(oinfo, 1);
+	if (status < 0)
+		goto out;
+	handle = ocfs2_start_trans(osb,
+		ocfs2_calc_qdel_credits(dquot->dq_sb, dquot->dq_id.type));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out_ilock;
+	}
+
+	status = ocfs2_global_release_dquot(dquot);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_trans;
+	}
+	status = ocfs2_local_release_dquot(handle, dquot);
+	/*
+	 * If we fail here, we cannot do much as global structure is
+	 * already released. So just complain...
+	 */
+	if (status < 0)
+		mlog_errno(status);
+	/*
+	 * Clear dq_off so that we search for the structure in quota file next
+	 * time we acquire it. The structure might be deleted and reallocated
+	 * elsewhere by another node while our dquot structure is on freelist.
+	 */
+	dquot->dq_off = 0;
+	clear_bit(DQ_ACTIVE_B, &dquot->dq_flags);
+out_trans:
+	ocfs2_commit_trans(osb, handle);
+out_ilock:
+	ocfs2_unlock_global_qf(oinfo, 1);
+out:
+	mutex_unlock(&dquot->dq_lock);
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/*
+ * Read global dquot structure from disk or create it if it does
+ * not exist. Also update use count of the global structure and
+ * create structure in node-local quota file.
+ */
+static int ocfs2_acquire_dquot(struct dquot *dquot)
+{
+	int status = 0, err;
+	int ex = 0;
+	struct super_block *sb = dquot->dq_sb;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	int type = dquot->dq_id.type;
+	struct ocfs2_mem_dqinfo *info = sb_dqinfo(sb, type)->dqi_priv;
+	struct inode *gqinode = info->dqi_gqinode;
+	int need_alloc = ocfs2_global_qinit_alloc(sb, type);
+	handle_t *handle;
+
+	trace_ocfs2_acquire_dquot(from_kqid(&init_user_ns, dquot->dq_id),
+				  type);
+	mutex_lock(&dquot->dq_lock);
+	/*
+	 * We need an exclusive lock, because we're going to update use count
+	 * and instantiate possibly new dquot structure
+	 */
+	status = ocfs2_lock_global_qf(info, 1);
+	if (status < 0)
+		goto out;
+	status = ocfs2_qinfo_lock(info, 0);
+	if (status < 0)
+		goto out_dq;
+	/*
+	 * We always want to read dquot structure from disk because we don't
+	 * know what happened with it while it was on freelist.
+	 */
+	status = qtree_read_dquot(&info->dqi_gi, dquot);
+	ocfs2_qinfo_unlock(info, 0);
+	if (status < 0)
+		goto out_dq;
+
+	OCFS2_DQUOT(dquot)->dq_use_count++;
+	OCFS2_DQUOT(dquot)->dq_origspace = dquot->dq_dqb.dqb_curspace;
+	OCFS2_DQUOT(dquot)->dq_originodes = dquot->dq_dqb.dqb_curinodes;
+	if (!dquot->dq_off) {	/* No real quota entry? */
+		ex = 1;
+		/*
+		 * Add blocks to quota file before we start a transaction since
+		 * locking allocators ranks above a transaction start
+		 */
+		WARN_ON(journal_current_handle());
+		status = ocfs2_extend_no_holes(gqinode, NULL,
+			i_size_read(gqinode) + (need_alloc << sb->s_blocksize_bits),
+			i_size_read(gqinode));
+		if (status < 0)
+			goto out_dq;
+	}
+
+	handle = ocfs2_start_trans(osb,
+				   ocfs2_calc_global_qinit_credits(sb, type));
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		goto out_dq;
+	}
+	status = ocfs2_qinfo_lock(info, ex);
+	if (status < 0)
+		goto out_trans;
+	status = qtree_write_dquot(&info->dqi_gi, dquot);
+	if (ex && info_dirty(sb_dqinfo(sb, type))) {
+		err = __ocfs2_global_write_info(sb, type);
+		if (!status)
+			status = err;
+	}
+	ocfs2_qinfo_unlock(info, ex);
+out_trans:
+	ocfs2_commit_trans(osb, handle);
+out_dq:
+	ocfs2_unlock_global_qf(info, 1);
+	if (status < 0)
+		goto out;
+
+	status = ocfs2_create_local_dquot(dquot);
+	if (status < 0)
+		goto out;
+	set_bit(DQ_ACTIVE_B, &dquot->dq_flags);
+out:
+	mutex_unlock(&dquot->dq_lock);
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static int ocfs2_mark_dquot_dirty(struct dquot *dquot)
+{
+	unsigned long mask = (1 << (DQ_LASTSET_B + QIF_ILIMITS_B)) |
+			     (1 << (DQ_LASTSET_B + QIF_BLIMITS_B)) |
+			     (1 << (DQ_LASTSET_B + QIF_INODES_B)) |
+			     (1 << (DQ_LASTSET_B + QIF_SPACE_B)) |
+			     (1 << (DQ_LASTSET_B + QIF_BTIME_B)) |
+			     (1 << (DQ_LASTSET_B + QIF_ITIME_B));
+	int sync = 0;
+	int status;
+	struct super_block *sb = dquot->dq_sb;
+	int type = dquot->dq_id.type;
+	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
+	handle_t *handle;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+
+	trace_ocfs2_mark_dquot_dirty(from_kqid(&init_user_ns, dquot->dq_id),
+				     type);
+
+	/* In case user set some limits, sync dquot immediately to global
+	 * quota file so that information propagates quicker */
+	spin_lock(&dq_data_lock);
+	if (dquot->dq_flags & mask)
+		sync = 1;
+	spin_unlock(&dq_data_lock);
+	/* This is a slight hack but we can't afford getting global quota
+	 * lock if we already have a transaction started. */
+	if (!sync || journal_current_handle()) {
+		status = ocfs2_write_dquot(dquot);
+		goto out;
+	}
+	status = ocfs2_lock_global_qf(oinfo, 1);
+	if (status < 0)
+		goto out;
+	handle = ocfs2_start_trans(osb, OCFS2_QSYNC_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out_ilock;
+	}
+	mutex_lock(&sb_dqopt(sb)->dqio_mutex);
+	status = ocfs2_sync_dquot(dquot);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_dlock;
+	}
+	/* Now write updated local dquot structure */
+	status = ocfs2_local_write_dquot(dquot);
+out_dlock:
+	mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
+	ocfs2_commit_trans(osb, handle);
+out_ilock:
+	ocfs2_unlock_global_qf(oinfo, 1);
+out:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/* This should happen only after set_dqinfo(). */
+static int ocfs2_write_info(struct super_block *sb, int type)
+{
+	handle_t *handle;
+	int status = 0;
+	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
+
+	status = ocfs2_lock_global_qf(oinfo, 1);
+	if (status < 0)
+		goto out;
+	handle = ocfs2_start_trans(OCFS2_SB(sb), OCFS2_QINFO_WRITE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out_ilock;
+	}
+	status = dquot_commit_info(sb, type);
+	ocfs2_commit_trans(OCFS2_SB(sb), handle);
+out_ilock:
+	ocfs2_unlock_global_qf(oinfo, 1);
+out:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static struct dquot *ocfs2_alloc_dquot(struct super_block *sb, int type)
+{
+	struct ocfs2_dquot *dquot =
+				kmem_cache_zalloc(ocfs2_dquot_cachep, GFP_NOFS);
+
+	if (!dquot)
+		return NULL;
+	return &dquot->dq_dquot;
+}
+
+static void ocfs2_destroy_dquot(struct dquot *dquot)
+{
+	kmem_cache_free(ocfs2_dquot_cachep, dquot);
+}
+
+const struct dquot_operations ocfs2_quota_operations = {
+	/* We never make dquot dirty so .write_dquot is never called */
+	.acquire_dquot	= ocfs2_acquire_dquot,
+	.release_dquot	= ocfs2_release_dquot,
+	.mark_dirty	= ocfs2_mark_dquot_dirty,
+	.write_info	= ocfs2_write_info,
+	.alloc_dquot	= ocfs2_alloc_dquot,
+	.destroy_dquot	= ocfs2_destroy_dquot,
+};
diff --git a/fs/ocfs2/quota_local.c b/fs/ocfs2/quota_local.c
new file mode 100644
index 000000000..3d0b63d34
--- /dev/null
+++ b/fs/ocfs2/quota_local.c
@@ -0,0 +1,1315 @@
+/*
+ *  Implementation of operations over local quota file
+ */
+
+#include <linux/fs.h>
+#include <linux/slab.h>
+#include <linux/quota.h>
+#include <linux/quotaops.h>
+#include <linux/module.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2_fs.h"
+#include "ocfs2.h"
+#include "inode.h"
+#include "alloc.h"
+#include "file.h"
+#include "buffer_head_io.h"
+#include "journal.h"
+#include "sysfile.h"
+#include "dlmglue.h"
+#include "quota.h"
+#include "uptodate.h"
+#include "super.h"
+#include "ocfs2_trace.h"
+
+/* Number of local quota structures per block */
+static inline unsigned int ol_quota_entries_per_block(struct super_block *sb)
+{
+	return ((sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE) /
+		sizeof(struct ocfs2_local_disk_dqblk));
+}
+
+/* Number of blocks with entries in one chunk */
+static inline unsigned int ol_chunk_blocks(struct super_block *sb)
+{
+	return ((sb->s_blocksize - sizeof(struct ocfs2_local_disk_chunk) -
+		 OCFS2_QBLK_RESERVED_SPACE) << 3) /
+	       ol_quota_entries_per_block(sb);
+}
+
+/* Number of entries in a chunk bitmap */
+static unsigned int ol_chunk_entries(struct super_block *sb)
+{
+	return ol_chunk_blocks(sb) * ol_quota_entries_per_block(sb);
+}
+
+/* Offset of the chunk in quota file */
+static unsigned int ol_quota_chunk_block(struct super_block *sb, int c)
+{
+	/* 1 block for local quota file info, 1 block per chunk for chunk info */
+	return 1 + (ol_chunk_blocks(sb) + 1) * c;
+}
+
+static unsigned int ol_dqblk_block(struct super_block *sb, int c, int off)
+{
+	int epb = ol_quota_entries_per_block(sb);
+
+	return ol_quota_chunk_block(sb, c) + 1 + off / epb;
+}
+
+static unsigned int ol_dqblk_block_off(struct super_block *sb, int c, int off)
+{
+	int epb = ol_quota_entries_per_block(sb);
+
+	return (off % epb) * sizeof(struct ocfs2_local_disk_dqblk);
+}
+
+/* Offset of the dquot structure in the quota file */
+static loff_t ol_dqblk_off(struct super_block *sb, int c, int off)
+{
+	return (ol_dqblk_block(sb, c, off) << sb->s_blocksize_bits) +
+	       ol_dqblk_block_off(sb, c, off);
+}
+
+static inline unsigned int ol_dqblk_block_offset(struct super_block *sb, loff_t off)
+{
+	return off & ((1 << sb->s_blocksize_bits) - 1);
+}
+
+/* Compute offset in the chunk of a structure with the given offset */
+static int ol_dqblk_chunk_off(struct super_block *sb, int c, loff_t off)
+{
+	int epb = ol_quota_entries_per_block(sb);
+
+	return ((off >> sb->s_blocksize_bits) -
+			ol_quota_chunk_block(sb, c) - 1) * epb
+	       + ((unsigned int)(off & ((1 << sb->s_blocksize_bits) - 1))) /
+		 sizeof(struct ocfs2_local_disk_dqblk);
+}
+
+/* Write bufferhead into the fs */
+static int ocfs2_modify_bh(struct inode *inode, struct buffer_head *bh,
+		void (*modify)(struct buffer_head *, void *), void *private)
+{
+	struct super_block *sb = inode->i_sb;
+	handle_t *handle;
+	int status;
+
+	handle = ocfs2_start_trans(OCFS2_SB(sb),
+				   OCFS2_QUOTA_BLOCK_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		return status;
+	}
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(inode), bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		ocfs2_commit_trans(OCFS2_SB(sb), handle);
+		return status;
+	}
+	lock_buffer(bh);
+	modify(bh, private);
+	unlock_buffer(bh);
+	ocfs2_journal_dirty(handle, bh);
+
+	status = ocfs2_commit_trans(OCFS2_SB(sb), handle);
+	if (status < 0) {
+		mlog_errno(status);
+		return status;
+	}
+	return 0;
+}
+
+/*
+ * Read quota block from a given logical offset.
+ *
+ * This function acquires ip_alloc_sem and thus it must not be called with a
+ * transaction started.
+ */
+static int ocfs2_read_quota_block(struct inode *inode, u64 v_block,
+				  struct buffer_head **bh)
+{
+	int rc = 0;
+	struct buffer_head *tmp = *bh;
+
+	if (i_size_read(inode) >> inode->i_sb->s_blocksize_bits <= v_block) {
+		ocfs2_error(inode->i_sb,
+			    "Quota file %llu is probably corrupted! Requested "
+			    "to read block %Lu but file has size only %Lu\n",
+			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
+			    (unsigned long long)v_block,
+			    (unsigned long long)i_size_read(inode));
+		return -EIO;
+	}
+	rc = ocfs2_read_virt_blocks(inode, v_block, 1, &tmp, 0,
+				    ocfs2_validate_quota_block);
+	if (rc)
+		mlog_errno(rc);
+
+	/* If ocfs2_read_virt_blocks() got us a new bh, pass it up. */
+	if (!rc && !*bh)
+		*bh = tmp;
+
+	return rc;
+}
+
+/* Check whether we understand format of quota files */
+static int ocfs2_local_check_quota_file(struct super_block *sb, int type)
+{
+	unsigned int lmagics[OCFS2_MAXQUOTAS] = OCFS2_LOCAL_QMAGICS;
+	unsigned int lversions[OCFS2_MAXQUOTAS] = OCFS2_LOCAL_QVERSIONS;
+	unsigned int gmagics[OCFS2_MAXQUOTAS] = OCFS2_GLOBAL_QMAGICS;
+	unsigned int gversions[OCFS2_MAXQUOTAS] = OCFS2_GLOBAL_QVERSIONS;
+	unsigned int ino[OCFS2_MAXQUOTAS] = { USER_QUOTA_SYSTEM_INODE,
+					      GROUP_QUOTA_SYSTEM_INODE };
+	struct buffer_head *bh = NULL;
+	struct inode *linode = sb_dqopt(sb)->files[type];
+	struct inode *ginode = NULL;
+	struct ocfs2_disk_dqheader *dqhead;
+	int status, ret = 0;
+
+	/* First check whether we understand local quota file */
+	status = ocfs2_read_quota_block(linode, 0, &bh);
+	if (status) {
+		mlog_errno(status);
+		mlog(ML_ERROR, "failed to read quota file header (type=%d)\n",
+			type);
+		goto out_err;
+	}
+	dqhead = (struct ocfs2_disk_dqheader *)(bh->b_data);
+	if (le32_to_cpu(dqhead->dqh_magic) != lmagics[type]) {
+		mlog(ML_ERROR, "quota file magic does not match (%u != %u),"
+			" type=%d\n", le32_to_cpu(dqhead->dqh_magic),
+			lmagics[type], type);
+		goto out_err;
+	}
+	if (le32_to_cpu(dqhead->dqh_version) != lversions[type]) {
+		mlog(ML_ERROR, "quota file version does not match (%u != %u),"
+			" type=%d\n", le32_to_cpu(dqhead->dqh_version),
+			lversions[type], type);
+		goto out_err;
+	}
+	brelse(bh);
+	bh = NULL;
+
+	/* Next check whether we understand global quota file */
+	ginode = ocfs2_get_system_file_inode(OCFS2_SB(sb), ino[type],
+						OCFS2_INVALID_SLOT);
+	if (!ginode) {
+		mlog(ML_ERROR, "cannot get global quota file inode "
+				"(type=%d)\n", type);
+		goto out_err;
+	}
+	/* Since the header is read only, we don't care about locking */
+	status = ocfs2_read_quota_block(ginode, 0, &bh);
+	if (status) {
+		mlog_errno(status);
+		mlog(ML_ERROR, "failed to read global quota file header "
+				"(type=%d)\n", type);
+		goto out_err;
+	}
+	dqhead = (struct ocfs2_disk_dqheader *)(bh->b_data);
+	if (le32_to_cpu(dqhead->dqh_magic) != gmagics[type]) {
+		mlog(ML_ERROR, "global quota file magic does not match "
+			"(%u != %u), type=%d\n",
+			le32_to_cpu(dqhead->dqh_magic), gmagics[type], type);
+		goto out_err;
+	}
+	if (le32_to_cpu(dqhead->dqh_version) != gversions[type]) {
+		mlog(ML_ERROR, "global quota file version does not match "
+			"(%u != %u), type=%d\n",
+			le32_to_cpu(dqhead->dqh_version), gversions[type],
+			type);
+		goto out_err;
+	}
+
+	ret = 1;
+out_err:
+	brelse(bh);
+	iput(ginode);
+	return ret;
+}
+
+/* Release given list of quota file chunks */
+static void ocfs2_release_local_quota_bitmaps(struct list_head *head)
+{
+	struct ocfs2_quota_chunk *pos, *next;
+
+	list_for_each_entry_safe(pos, next, head, qc_chunk) {
+		list_del(&pos->qc_chunk);
+		brelse(pos->qc_headerbh);
+		kmem_cache_free(ocfs2_qf_chunk_cachep, pos);
+	}
+}
+
+/* Load quota bitmaps into memory */
+static int ocfs2_load_local_quota_bitmaps(struct inode *inode,
+			struct ocfs2_local_disk_dqinfo *ldinfo,
+			struct list_head *head)
+{
+	struct ocfs2_quota_chunk *newchunk;
+	int i, status;
+
+	INIT_LIST_HEAD(head);
+	for (i = 0; i < le32_to_cpu(ldinfo->dqi_chunks); i++) {
+		newchunk = kmem_cache_alloc(ocfs2_qf_chunk_cachep, GFP_NOFS);
+		if (!newchunk) {
+			ocfs2_release_local_quota_bitmaps(head);
+			return -ENOMEM;
+		}
+		newchunk->qc_num = i;
+		newchunk->qc_headerbh = NULL;
+		status = ocfs2_read_quota_block(inode,
+				ol_quota_chunk_block(inode->i_sb, i),
+				&newchunk->qc_headerbh);
+		if (status) {
+			mlog_errno(status);
+			kmem_cache_free(ocfs2_qf_chunk_cachep, newchunk);
+			ocfs2_release_local_quota_bitmaps(head);
+			return status;
+		}
+		list_add_tail(&newchunk->qc_chunk, head);
+	}
+	return 0;
+}
+
+static void olq_update_info(struct buffer_head *bh, void *private)
+{
+	struct mem_dqinfo *info = private;
+	struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
+	struct ocfs2_local_disk_dqinfo *ldinfo;
+
+	ldinfo = (struct ocfs2_local_disk_dqinfo *)(bh->b_data +
+						OCFS2_LOCAL_INFO_OFF);
+	spin_lock(&dq_data_lock);
+	ldinfo->dqi_flags = cpu_to_le32(oinfo->dqi_flags);
+	ldinfo->dqi_chunks = cpu_to_le32(oinfo->dqi_chunks);
+	ldinfo->dqi_blocks = cpu_to_le32(oinfo->dqi_blocks);
+	spin_unlock(&dq_data_lock);
+}
+
+static int ocfs2_add_recovery_chunk(struct super_block *sb,
+				    struct ocfs2_local_disk_chunk *dchunk,
+				    int chunk,
+				    struct list_head *head)
+{
+	struct ocfs2_recovery_chunk *rc;
+
+	rc = kmalloc(sizeof(struct ocfs2_recovery_chunk), GFP_NOFS);
+	if (!rc)
+		return -ENOMEM;
+	rc->rc_chunk = chunk;
+	rc->rc_bitmap = kmalloc(sb->s_blocksize, GFP_NOFS);
+	if (!rc->rc_bitmap) {
+		kfree(rc);
+		return -ENOMEM;
+	}
+	memcpy(rc->rc_bitmap, dchunk->dqc_bitmap,
+	       (ol_chunk_entries(sb) + 7) >> 3);
+	list_add_tail(&rc->rc_list, head);
+	return 0;
+}
+
+static void free_recovery_list(struct list_head *head)
+{
+	struct ocfs2_recovery_chunk *next;
+	struct ocfs2_recovery_chunk *rchunk;
+
+	list_for_each_entry_safe(rchunk, next, head, rc_list) {
+		list_del(&rchunk->rc_list);
+		kfree(rchunk->rc_bitmap);
+		kfree(rchunk);
+	}
+}
+
+void ocfs2_free_quota_recovery(struct ocfs2_quota_recovery *rec)
+{
+	int type;
+
+	for (type = 0; type < OCFS2_MAXQUOTAS; type++)
+		free_recovery_list(&(rec->r_list[type]));
+	kfree(rec);
+}
+
+/* Load entries in our quota file we have to recover*/
+static int ocfs2_recovery_load_quota(struct inode *lqinode,
+				     struct ocfs2_local_disk_dqinfo *ldinfo,
+				     int type,
+				     struct list_head *head)
+{
+	struct super_block *sb = lqinode->i_sb;
+	struct buffer_head *hbh;
+	struct ocfs2_local_disk_chunk *dchunk;
+	int i, chunks = le32_to_cpu(ldinfo->dqi_chunks);
+	int status = 0;
+
+	for (i = 0; i < chunks; i++) {
+		hbh = NULL;
+		status = ocfs2_read_quota_block(lqinode,
+						ol_quota_chunk_block(sb, i),
+						&hbh);
+		if (status) {
+			mlog_errno(status);
+			break;
+		}
+		dchunk = (struct ocfs2_local_disk_chunk *)hbh->b_data;
+		if (le32_to_cpu(dchunk->dqc_free) < ol_chunk_entries(sb))
+			status = ocfs2_add_recovery_chunk(sb, dchunk, i, head);
+		brelse(hbh);
+		if (status < 0)
+			break;
+	}
+	if (status < 0)
+		free_recovery_list(head);
+	return status;
+}
+
+static struct ocfs2_quota_recovery *ocfs2_alloc_quota_recovery(void)
+{
+	int type;
+	struct ocfs2_quota_recovery *rec;
+
+	rec = kmalloc(sizeof(struct ocfs2_quota_recovery), GFP_NOFS);
+	if (!rec)
+		return NULL;
+	for (type = 0; type < OCFS2_MAXQUOTAS; type++)
+		INIT_LIST_HEAD(&(rec->r_list[type]));
+	return rec;
+}
+
+/* Load information we need for quota recovery into memory */
+struct ocfs2_quota_recovery *ocfs2_begin_quota_recovery(
+						struct ocfs2_super *osb,
+						int slot_num)
+{
+	unsigned int feature[OCFS2_MAXQUOTAS] = {
+					OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
+					OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
+	unsigned int ino[OCFS2_MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
+					      LOCAL_GROUP_QUOTA_SYSTEM_INODE };
+	struct super_block *sb = osb->sb;
+	struct ocfs2_local_disk_dqinfo *ldinfo;
+	struct inode *lqinode;
+	struct buffer_head *bh;
+	int type;
+	int status = 0;
+	struct ocfs2_quota_recovery *rec;
+
+	printk(KERN_NOTICE "ocfs2: Beginning quota recovery on device (%s) for "
+	       "slot %u\n", osb->dev_str, slot_num);
+
+	rec = ocfs2_alloc_quota_recovery();
+	if (!rec)
+		return ERR_PTR(-ENOMEM);
+	/* First init... */
+
+	for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
+		if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
+			continue;
+		/* At this point, journal of the slot is already replayed so
+		 * we can trust metadata and data of the quota file */
+		lqinode = ocfs2_get_system_file_inode(osb, ino[type], slot_num);
+		if (!lqinode) {
+			status = -ENOENT;
+			goto out;
+		}
+		status = ocfs2_inode_lock_full(lqinode, NULL, 1,
+					       OCFS2_META_LOCK_RECOVERY);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out_put;
+		}
+		/* Now read local header */
+		bh = NULL;
+		status = ocfs2_read_quota_block(lqinode, 0, &bh);
+		if (status) {
+			mlog_errno(status);
+			mlog(ML_ERROR, "failed to read quota file info header "
+				"(slot=%d type=%d)\n", slot_num, type);
+			goto out_lock;
+		}
+		ldinfo = (struct ocfs2_local_disk_dqinfo *)(bh->b_data +
+							OCFS2_LOCAL_INFO_OFF);
+		status = ocfs2_recovery_load_quota(lqinode, ldinfo, type,
+						   &rec->r_list[type]);
+		brelse(bh);
+out_lock:
+		ocfs2_inode_unlock(lqinode, 1);
+out_put:
+		iput(lqinode);
+		if (status < 0)
+			break;
+	}
+out:
+	if (status < 0) {
+		ocfs2_free_quota_recovery(rec);
+		rec = ERR_PTR(status);
+	}
+	return rec;
+}
+
+/* Sync changes in local quota file into global quota file and
+ * reinitialize local quota file.
+ * The function expects local quota file to be already locked and
+ * dqonoff_mutex locked. */
+static int ocfs2_recover_local_quota_file(struct inode *lqinode,
+					  int type,
+					  struct ocfs2_quota_recovery *rec)
+{
+	struct super_block *sb = lqinode->i_sb;
+	struct ocfs2_mem_dqinfo *oinfo = sb_dqinfo(sb, type)->dqi_priv;
+	struct ocfs2_local_disk_chunk *dchunk;
+	struct ocfs2_local_disk_dqblk *dqblk;
+	struct dquot *dquot;
+	handle_t *handle;
+	struct buffer_head *hbh = NULL, *qbh = NULL;
+	int status = 0;
+	int bit, chunk;
+	struct ocfs2_recovery_chunk *rchunk, *next;
+	qsize_t spacechange, inodechange;
+
+	trace_ocfs2_recover_local_quota_file((unsigned long)lqinode->i_ino, type);
+
+	list_for_each_entry_safe(rchunk, next, &(rec->r_list[type]), rc_list) {
+		chunk = rchunk->rc_chunk;
+		hbh = NULL;
+		status = ocfs2_read_quota_block(lqinode,
+						ol_quota_chunk_block(sb, chunk),
+						&hbh);
+		if (status) {
+			mlog_errno(status);
+			break;
+		}
+		dchunk = (struct ocfs2_local_disk_chunk *)hbh->b_data;
+		for_each_set_bit(bit, rchunk->rc_bitmap, ol_chunk_entries(sb)) {
+			qbh = NULL;
+			status = ocfs2_read_quota_block(lqinode,
+						ol_dqblk_block(sb, chunk, bit),
+						&qbh);
+			if (status) {
+				mlog_errno(status);
+				break;
+			}
+			dqblk = (struct ocfs2_local_disk_dqblk *)(qbh->b_data +
+				ol_dqblk_block_off(sb, chunk, bit));
+			dquot = dqget(sb,
+				      make_kqid(&init_user_ns, type,
+						le64_to_cpu(dqblk->dqb_id)));
+			if (!dquot) {
+				status = -EIO;
+				mlog(ML_ERROR, "Failed to get quota structure "
+				     "for id %u, type %d. Cannot finish quota "
+				     "file recovery.\n",
+				     (unsigned)le64_to_cpu(dqblk->dqb_id),
+				     type);
+				goto out_put_bh;
+			}
+			status = ocfs2_lock_global_qf(oinfo, 1);
+			if (status < 0) {
+				mlog_errno(status);
+				goto out_put_dquot;
+			}
+
+			handle = ocfs2_start_trans(OCFS2_SB(sb),
+						   OCFS2_QSYNC_CREDITS);
+			if (IS_ERR(handle)) {
+				status = PTR_ERR(handle);
+				mlog_errno(status);
+				goto out_drop_lock;
+			}
+			mutex_lock(&sb_dqopt(sb)->dqio_mutex);
+			spin_lock(&dq_data_lock);
+			/* Add usage from quota entry into quota changes
+			 * of our node. Auxiliary variables are important
+			 * due to signedness */
+			spacechange = le64_to_cpu(dqblk->dqb_spacemod);
+			inodechange = le64_to_cpu(dqblk->dqb_inodemod);
+			dquot->dq_dqb.dqb_curspace += spacechange;
+			dquot->dq_dqb.dqb_curinodes += inodechange;
+			spin_unlock(&dq_data_lock);
+			/* We want to drop reference held by the crashed
+			 * node. Since we have our own reference we know
+			 * global structure actually won't be freed. */
+			status = ocfs2_global_release_dquot(dquot);
+			if (status < 0) {
+				mlog_errno(status);
+				goto out_commit;
+			}
+			/* Release local quota file entry */
+			status = ocfs2_journal_access_dq(handle,
+					INODE_CACHE(lqinode),
+					qbh, OCFS2_JOURNAL_ACCESS_WRITE);
+			if (status < 0) {
+				mlog_errno(status);
+				goto out_commit;
+			}
+			lock_buffer(qbh);
+			WARN_ON(!ocfs2_test_bit_unaligned(bit, dchunk->dqc_bitmap));
+			ocfs2_clear_bit_unaligned(bit, dchunk->dqc_bitmap);
+			le32_add_cpu(&dchunk->dqc_free, 1);
+			unlock_buffer(qbh);
+			ocfs2_journal_dirty(handle, qbh);
+out_commit:
+			mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
+			ocfs2_commit_trans(OCFS2_SB(sb), handle);
+out_drop_lock:
+			ocfs2_unlock_global_qf(oinfo, 1);
+out_put_dquot:
+			dqput(dquot);
+out_put_bh:
+			brelse(qbh);
+			if (status < 0)
+				break;
+		}
+		brelse(hbh);
+		list_del(&rchunk->rc_list);
+		kfree(rchunk->rc_bitmap);
+		kfree(rchunk);
+		if (status < 0)
+			break;
+	}
+	if (status < 0)
+		free_recovery_list(&(rec->r_list[type]));
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/* Recover local quota files for given node different from us */
+int ocfs2_finish_quota_recovery(struct ocfs2_super *osb,
+				struct ocfs2_quota_recovery *rec,
+				int slot_num)
+{
+	unsigned int ino[OCFS2_MAXQUOTAS] = { LOCAL_USER_QUOTA_SYSTEM_INODE,
+					      LOCAL_GROUP_QUOTA_SYSTEM_INODE };
+	struct super_block *sb = osb->sb;
+	struct ocfs2_local_disk_dqinfo *ldinfo;
+	struct buffer_head *bh;
+	handle_t *handle;
+	int type;
+	int status = 0;
+	struct inode *lqinode;
+	unsigned int flags;
+
+	printk(KERN_NOTICE "ocfs2: Finishing quota recovery on device (%s) for "
+	       "slot %u\n", osb->dev_str, slot_num);
+
+	mutex_lock(&sb_dqopt(sb)->dqonoff_mutex);
+	for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
+		if (list_empty(&(rec->r_list[type])))
+			continue;
+		trace_ocfs2_finish_quota_recovery(slot_num);
+		lqinode = ocfs2_get_system_file_inode(osb, ino[type], slot_num);
+		if (!lqinode) {
+			status = -ENOENT;
+			goto out;
+		}
+		status = ocfs2_inode_lock_full(lqinode, NULL, 1,
+						       OCFS2_META_LOCK_NOQUEUE);
+		/* Someone else is holding the lock? Then he must be
+		 * doing the recovery. Just skip the file... */
+		if (status == -EAGAIN) {
+			printk(KERN_NOTICE "ocfs2: Skipping quota recovery on "
+			       "device (%s) for slot %d because quota file is "
+			       "locked.\n", osb->dev_str, slot_num);
+			status = 0;
+			goto out_put;
+		} else if (status < 0) {
+			mlog_errno(status);
+			goto out_put;
+		}
+		/* Now read local header */
+		bh = NULL;
+		status = ocfs2_read_quota_block(lqinode, 0, &bh);
+		if (status) {
+			mlog_errno(status);
+			mlog(ML_ERROR, "failed to read quota file info header "
+				"(slot=%d type=%d)\n", slot_num, type);
+			goto out_lock;
+		}
+		ldinfo = (struct ocfs2_local_disk_dqinfo *)(bh->b_data +
+							OCFS2_LOCAL_INFO_OFF);
+		/* Is recovery still needed? */
+		flags = le32_to_cpu(ldinfo->dqi_flags);
+		if (!(flags & OLQF_CLEAN))
+			status = ocfs2_recover_local_quota_file(lqinode,
+								type,
+								rec);
+		/* We don't want to mark file as clean when it is actually
+		 * active */
+		if (slot_num == osb->slot_num)
+			goto out_bh;
+		/* Mark quota file as clean if we are recovering quota file of
+		 * some other node. */
+		handle = ocfs2_start_trans(osb,
+					   OCFS2_LOCAL_QINFO_WRITE_CREDITS);
+		if (IS_ERR(handle)) {
+			status = PTR_ERR(handle);
+			mlog_errno(status);
+			goto out_bh;
+		}
+		status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode),
+						 bh,
+						 OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out_trans;
+		}
+		lock_buffer(bh);
+		ldinfo->dqi_flags = cpu_to_le32(flags | OLQF_CLEAN);
+		unlock_buffer(bh);
+		ocfs2_journal_dirty(handle, bh);
+out_trans:
+		ocfs2_commit_trans(osb, handle);
+out_bh:
+		brelse(bh);
+out_lock:
+		ocfs2_inode_unlock(lqinode, 1);
+out_put:
+		iput(lqinode);
+		if (status < 0)
+			break;
+	}
+out:
+	mutex_unlock(&sb_dqopt(sb)->dqonoff_mutex);
+	kfree(rec);
+	return status;
+}
+
+/* Read information header from quota file */
+static int ocfs2_local_read_info(struct super_block *sb, int type)
+{
+	struct ocfs2_local_disk_dqinfo *ldinfo;
+	struct mem_dqinfo *info = sb_dqinfo(sb, type);
+	struct ocfs2_mem_dqinfo *oinfo;
+	struct inode *lqinode = sb_dqopt(sb)->files[type];
+	int status;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_quota_recovery *rec;
+	int locked = 0;
+
+	/* We don't need the lock and we have to acquire quota file locks
+	 * which will later depend on this lock */
+	mutex_unlock(&sb_dqopt(sb)->dqio_mutex);
+	info->dqi_max_spc_limit = 0x7fffffffffffffffLL;
+	info->dqi_max_ino_limit = 0x7fffffffffffffffLL;
+	oinfo = kmalloc(sizeof(struct ocfs2_mem_dqinfo), GFP_NOFS);
+	if (!oinfo) {
+		mlog(ML_ERROR, "failed to allocate memory for ocfs2 quota"
+			       " info.");
+		goto out_err;
+	}
+	info->dqi_priv = oinfo;
+	oinfo->dqi_type = type;
+	INIT_LIST_HEAD(&oinfo->dqi_chunk);
+	oinfo->dqi_rec = NULL;
+	oinfo->dqi_lqi_bh = NULL;
+	oinfo->dqi_libh = NULL;
+
+	status = ocfs2_global_read_info(sb, type);
+	if (status < 0)
+		goto out_err;
+
+	status = ocfs2_inode_lock(lqinode, &oinfo->dqi_lqi_bh, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_err;
+	}
+	locked = 1;
+
+	/* Now read local header */
+	status = ocfs2_read_quota_block(lqinode, 0, &bh);
+	if (status) {
+		mlog_errno(status);
+		mlog(ML_ERROR, "failed to read quota file info header "
+			"(type=%d)\n", type);
+		goto out_err;
+	}
+	ldinfo = (struct ocfs2_local_disk_dqinfo *)(bh->b_data +
+						OCFS2_LOCAL_INFO_OFF);
+	oinfo->dqi_flags = le32_to_cpu(ldinfo->dqi_flags);
+	oinfo->dqi_chunks = le32_to_cpu(ldinfo->dqi_chunks);
+	oinfo->dqi_blocks = le32_to_cpu(ldinfo->dqi_blocks);
+	oinfo->dqi_libh = bh;
+
+	/* We crashed when using local quota file? */
+	if (!(oinfo->dqi_flags & OLQF_CLEAN)) {
+		rec = OCFS2_SB(sb)->quota_rec;
+		if (!rec) {
+			rec = ocfs2_alloc_quota_recovery();
+			if (!rec) {
+				status = -ENOMEM;
+				mlog_errno(status);
+				goto out_err;
+			}
+			OCFS2_SB(sb)->quota_rec = rec;
+		}
+
+		status = ocfs2_recovery_load_quota(lqinode, ldinfo, type,
+                                                   &rec->r_list[type]);
+		if (status < 0) {
+			mlog_errno(status);
+			goto out_err;
+		}
+	}
+
+	status = ocfs2_load_local_quota_bitmaps(lqinode,
+						ldinfo,
+						&oinfo->dqi_chunk);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_err;
+	}
+
+	/* Now mark quota file as used */
+	oinfo->dqi_flags &= ~OLQF_CLEAN;
+	status = ocfs2_modify_bh(lqinode, bh, olq_update_info, info);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_err;
+	}
+
+	mutex_lock(&sb_dqopt(sb)->dqio_mutex);
+	return 0;
+out_err:
+	if (oinfo) {
+		iput(oinfo->dqi_gqinode);
+		ocfs2_simple_drop_lockres(OCFS2_SB(sb), &oinfo->dqi_gqlock);
+		ocfs2_lock_res_free(&oinfo->dqi_gqlock);
+		brelse(oinfo->dqi_lqi_bh);
+		if (locked)
+			ocfs2_inode_unlock(lqinode, 1);
+		ocfs2_release_local_quota_bitmaps(&oinfo->dqi_chunk);
+		kfree(oinfo);
+	}
+	brelse(bh);
+	mutex_lock(&sb_dqopt(sb)->dqio_mutex);
+	return -1;
+}
+
+/* Write local info to quota file */
+static int ocfs2_local_write_info(struct super_block *sb, int type)
+{
+	struct mem_dqinfo *info = sb_dqinfo(sb, type);
+	struct buffer_head *bh = ((struct ocfs2_mem_dqinfo *)info->dqi_priv)
+						->dqi_libh;
+	int status;
+
+	status = ocfs2_modify_bh(sb_dqopt(sb)->files[type], bh, olq_update_info,
+				 info);
+	if (status < 0) {
+		mlog_errno(status);
+		return -1;
+	}
+
+	return 0;
+}
+
+/* Release info from memory */
+static int ocfs2_local_free_info(struct super_block *sb, int type)
+{
+	struct mem_dqinfo *info = sb_dqinfo(sb, type);
+	struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
+	struct ocfs2_quota_chunk *chunk;
+	struct ocfs2_local_disk_chunk *dchunk;
+	int mark_clean = 1, len;
+	int status;
+
+	iput(oinfo->dqi_gqinode);
+	ocfs2_simple_drop_lockres(OCFS2_SB(sb), &oinfo->dqi_gqlock);
+	ocfs2_lock_res_free(&oinfo->dqi_gqlock);
+	list_for_each_entry(chunk, &oinfo->dqi_chunk, qc_chunk) {
+		dchunk = (struct ocfs2_local_disk_chunk *)
+					(chunk->qc_headerbh->b_data);
+		if (chunk->qc_num < oinfo->dqi_chunks - 1) {
+			len = ol_chunk_entries(sb);
+		} else {
+			len = (oinfo->dqi_blocks -
+			       ol_quota_chunk_block(sb, chunk->qc_num) - 1)
+			      * ol_quota_entries_per_block(sb);
+		}
+		/* Not all entries free? Bug! */
+		if (le32_to_cpu(dchunk->dqc_free) != len) {
+			mlog(ML_ERROR, "releasing quota file with used "
+					"entries (type=%d)\n", type);
+			mark_clean = 0;
+		}
+	}
+	ocfs2_release_local_quota_bitmaps(&oinfo->dqi_chunk);
+
+	/* dqonoff_mutex protects us against racing with recovery thread... */
+	if (oinfo->dqi_rec) {
+		ocfs2_free_quota_recovery(oinfo->dqi_rec);
+		mark_clean = 0;
+	}
+
+	if (!mark_clean)
+		goto out;
+
+	/* Mark local file as clean */
+	oinfo->dqi_flags |= OLQF_CLEAN;
+	status = ocfs2_modify_bh(sb_dqopt(sb)->files[type],
+				 oinfo->dqi_libh,
+				 olq_update_info,
+				 info);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+out:
+	ocfs2_inode_unlock(sb_dqopt(sb)->files[type], 1);
+	brelse(oinfo->dqi_libh);
+	brelse(oinfo->dqi_lqi_bh);
+	kfree(oinfo);
+	return 0;
+}
+
+static void olq_set_dquot(struct buffer_head *bh, void *private)
+{
+	struct ocfs2_dquot *od = private;
+	struct ocfs2_local_disk_dqblk *dqblk;
+	struct super_block *sb = od->dq_dquot.dq_sb;
+
+	dqblk = (struct ocfs2_local_disk_dqblk *)(bh->b_data
+		+ ol_dqblk_block_offset(sb, od->dq_local_off));
+
+	dqblk->dqb_id = cpu_to_le64(from_kqid(&init_user_ns,
+					      od->dq_dquot.dq_id));
+	spin_lock(&dq_data_lock);
+	dqblk->dqb_spacemod = cpu_to_le64(od->dq_dquot.dq_dqb.dqb_curspace -
+					  od->dq_origspace);
+	dqblk->dqb_inodemod = cpu_to_le64(od->dq_dquot.dq_dqb.dqb_curinodes -
+					  od->dq_originodes);
+	spin_unlock(&dq_data_lock);
+	trace_olq_set_dquot(
+		(unsigned long long)le64_to_cpu(dqblk->dqb_spacemod),
+		(unsigned long long)le64_to_cpu(dqblk->dqb_inodemod),
+		from_kqid(&init_user_ns, od->dq_dquot.dq_id));
+}
+
+/* Write dquot to local quota file */
+int ocfs2_local_write_dquot(struct dquot *dquot)
+{
+	struct super_block *sb = dquot->dq_sb;
+	struct ocfs2_dquot *od = OCFS2_DQUOT(dquot);
+	struct buffer_head *bh;
+	struct inode *lqinode = sb_dqopt(sb)->files[dquot->dq_id.type];
+	int status;
+
+	status = ocfs2_read_quota_phys_block(lqinode, od->dq_local_phys_blk,
+					     &bh);
+	if (status) {
+		mlog_errno(status);
+		goto out;
+	}
+	status = ocfs2_modify_bh(lqinode, bh, olq_set_dquot, od);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+out:
+	brelse(bh);
+	return status;
+}
+
+/* Find free entry in local quota file */
+static struct ocfs2_quota_chunk *ocfs2_find_free_entry(struct super_block *sb,
+						       int type,
+						       int *offset)
+{
+	struct mem_dqinfo *info = sb_dqinfo(sb, type);
+	struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
+	struct ocfs2_quota_chunk *chunk;
+	struct ocfs2_local_disk_chunk *dchunk;
+	int found = 0, len;
+
+	list_for_each_entry(chunk, &oinfo->dqi_chunk, qc_chunk) {
+		dchunk = (struct ocfs2_local_disk_chunk *)
+						chunk->qc_headerbh->b_data;
+		if (le32_to_cpu(dchunk->dqc_free) > 0) {
+			found = 1;
+			break;
+		}
+	}
+	if (!found)
+		return NULL;
+
+	if (chunk->qc_num < oinfo->dqi_chunks - 1) {
+		len = ol_chunk_entries(sb);
+	} else {
+		len = (oinfo->dqi_blocks -
+		       ol_quota_chunk_block(sb, chunk->qc_num) - 1)
+		      * ol_quota_entries_per_block(sb);
+	}
+
+	found = ocfs2_find_next_zero_bit_unaligned(dchunk->dqc_bitmap, len, 0);
+	/* We failed? */
+	if (found == len) {
+		mlog(ML_ERROR, "Did not find empty entry in chunk %d with %u"
+		     " entries free (type=%d)\n", chunk->qc_num,
+		     le32_to_cpu(dchunk->dqc_free), type);
+		return ERR_PTR(-EIO);
+	}
+	*offset = found;
+	return chunk;
+}
+
+/* Add new chunk to the local quota file */
+static struct ocfs2_quota_chunk *ocfs2_local_quota_add_chunk(
+							struct super_block *sb,
+							int type,
+							int *offset)
+{
+	struct mem_dqinfo *info = sb_dqinfo(sb, type);
+	struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
+	struct inode *lqinode = sb_dqopt(sb)->files[type];
+	struct ocfs2_quota_chunk *chunk = NULL;
+	struct ocfs2_local_disk_chunk *dchunk;
+	int status;
+	handle_t *handle;
+	struct buffer_head *bh = NULL, *dbh = NULL;
+	u64 p_blkno;
+
+	/* We are protected by dqio_sem so no locking needed */
+	status = ocfs2_extend_no_holes(lqinode, NULL,
+				       i_size_read(lqinode) + 2 * sb->s_blocksize,
+				       i_size_read(lqinode));
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+	status = ocfs2_simple_size_update(lqinode, oinfo->dqi_lqi_bh,
+					  i_size_read(lqinode) + 2 * sb->s_blocksize);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	chunk = kmem_cache_alloc(ocfs2_qf_chunk_cachep, GFP_NOFS);
+	if (!chunk) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto out;
+	}
+	/* Local quota info and two new blocks we initialize */
+	handle = ocfs2_start_trans(OCFS2_SB(sb),
+			OCFS2_LOCAL_QINFO_WRITE_CREDITS +
+			2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out;
+	}
+
+	/* Initialize chunk header */
+	status = ocfs2_extent_map_get_blocks(lqinode, oinfo->dqi_blocks,
+					     &p_blkno, NULL, NULL);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_trans;
+	}
+	bh = sb_getblk(sb, p_blkno);
+	if (!bh) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto out_trans;
+	}
+	dchunk = (struct ocfs2_local_disk_chunk *)bh->b_data;
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(lqinode), bh);
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode), bh,
+					 OCFS2_JOURNAL_ACCESS_CREATE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_trans;
+	}
+	lock_buffer(bh);
+	dchunk->dqc_free = cpu_to_le32(ol_quota_entries_per_block(sb));
+	memset(dchunk->dqc_bitmap, 0,
+	       sb->s_blocksize - sizeof(struct ocfs2_local_disk_chunk) -
+	       OCFS2_QBLK_RESERVED_SPACE);
+	unlock_buffer(bh);
+	ocfs2_journal_dirty(handle, bh);
+
+	/* Initialize new block with structures */
+	status = ocfs2_extent_map_get_blocks(lqinode, oinfo->dqi_blocks + 1,
+					     &p_blkno, NULL, NULL);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_trans;
+	}
+	dbh = sb_getblk(sb, p_blkno);
+	if (!dbh) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto out_trans;
+	}
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(lqinode), dbh);
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode), dbh,
+					 OCFS2_JOURNAL_ACCESS_CREATE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_trans;
+	}
+	lock_buffer(dbh);
+	memset(dbh->b_data, 0, sb->s_blocksize - OCFS2_QBLK_RESERVED_SPACE);
+	unlock_buffer(dbh);
+	ocfs2_journal_dirty(handle, dbh);
+
+	/* Update local quotafile info */
+	oinfo->dqi_blocks += 2;
+	oinfo->dqi_chunks++;
+	status = ocfs2_local_write_info(sb, type);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_trans;
+	}
+	status = ocfs2_commit_trans(OCFS2_SB(sb), handle);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	list_add_tail(&chunk->qc_chunk, &oinfo->dqi_chunk);
+	chunk->qc_num = list_entry(chunk->qc_chunk.prev,
+				   struct ocfs2_quota_chunk,
+				   qc_chunk)->qc_num + 1;
+	chunk->qc_headerbh = bh;
+	*offset = 0;
+	return chunk;
+out_trans:
+	ocfs2_commit_trans(OCFS2_SB(sb), handle);
+out:
+	brelse(bh);
+	brelse(dbh);
+	kmem_cache_free(ocfs2_qf_chunk_cachep, chunk);
+	return ERR_PTR(status);
+}
+
+/* Find free entry in local quota file */
+static struct ocfs2_quota_chunk *ocfs2_extend_local_quota_file(
+						       struct super_block *sb,
+						       int type,
+						       int *offset)
+{
+	struct mem_dqinfo *info = sb_dqinfo(sb, type);
+	struct ocfs2_mem_dqinfo *oinfo = info->dqi_priv;
+	struct ocfs2_quota_chunk *chunk;
+	struct inode *lqinode = sb_dqopt(sb)->files[type];
+	struct ocfs2_local_disk_chunk *dchunk;
+	int epb = ol_quota_entries_per_block(sb);
+	unsigned int chunk_blocks;
+	struct buffer_head *bh;
+	u64 p_blkno;
+	int status;
+	handle_t *handle;
+
+	if (list_empty(&oinfo->dqi_chunk))
+		return ocfs2_local_quota_add_chunk(sb, type, offset);
+	/* Is the last chunk full? */
+	chunk = list_entry(oinfo->dqi_chunk.prev,
+			struct ocfs2_quota_chunk, qc_chunk);
+	chunk_blocks = oinfo->dqi_blocks -
+			ol_quota_chunk_block(sb, chunk->qc_num) - 1;
+	if (ol_chunk_blocks(sb) == chunk_blocks)
+		return ocfs2_local_quota_add_chunk(sb, type, offset);
+
+	/* We are protected by dqio_sem so no locking needed */
+	status = ocfs2_extend_no_holes(lqinode, NULL,
+				       i_size_read(lqinode) + sb->s_blocksize,
+				       i_size_read(lqinode));
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+	status = ocfs2_simple_size_update(lqinode, oinfo->dqi_lqi_bh,
+					  i_size_read(lqinode) + sb->s_blocksize);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	/* Get buffer from the just added block */
+	status = ocfs2_extent_map_get_blocks(lqinode, oinfo->dqi_blocks,
+					     &p_blkno, NULL, NULL);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+	bh = sb_getblk(sb, p_blkno);
+	if (!bh) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto out;
+	}
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(lqinode), bh);
+
+	/* Local quota info, chunk header and the new block we initialize */
+	handle = ocfs2_start_trans(OCFS2_SB(sb),
+			OCFS2_LOCAL_QINFO_WRITE_CREDITS +
+			2 * OCFS2_QUOTA_BLOCK_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		mlog_errno(status);
+		goto out;
+	}
+	/* Zero created block */
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode), bh,
+				 OCFS2_JOURNAL_ACCESS_CREATE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_trans;
+	}
+	lock_buffer(bh);
+	memset(bh->b_data, 0, sb->s_blocksize);
+	unlock_buffer(bh);
+	ocfs2_journal_dirty(handle, bh);
+
+	/* Update chunk header */
+	status = ocfs2_journal_access_dq(handle, INODE_CACHE(lqinode),
+					 chunk->qc_headerbh,
+				 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_trans;
+	}
+
+	dchunk = (struct ocfs2_local_disk_chunk *)chunk->qc_headerbh->b_data;
+	lock_buffer(chunk->qc_headerbh);
+	le32_add_cpu(&dchunk->dqc_free, ol_quota_entries_per_block(sb));
+	unlock_buffer(chunk->qc_headerbh);
+	ocfs2_journal_dirty(handle, chunk->qc_headerbh);
+
+	/* Update file header */
+	oinfo->dqi_blocks++;
+	status = ocfs2_local_write_info(sb, type);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out_trans;
+	}
+
+	status = ocfs2_commit_trans(OCFS2_SB(sb), handle);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+	*offset = chunk_blocks * epb;
+	return chunk;
+out_trans:
+	ocfs2_commit_trans(OCFS2_SB(sb), handle);
+out:
+	return ERR_PTR(status);
+}
+
+static void olq_alloc_dquot(struct buffer_head *bh, void *private)
+{
+	int *offset = private;
+	struct ocfs2_local_disk_chunk *dchunk;
+
+	dchunk = (struct ocfs2_local_disk_chunk *)bh->b_data;
+	ocfs2_set_bit_unaligned(*offset, dchunk->dqc_bitmap);
+	le32_add_cpu(&dchunk->dqc_free, -1);
+}
+
+/* Create dquot in the local file for given id */
+int ocfs2_create_local_dquot(struct dquot *dquot)
+{
+	struct super_block *sb = dquot->dq_sb;
+	int type = dquot->dq_id.type;
+	struct inode *lqinode = sb_dqopt(sb)->files[type];
+	struct ocfs2_quota_chunk *chunk;
+	struct ocfs2_dquot *od = OCFS2_DQUOT(dquot);
+	int offset;
+	int status;
+	u64 pcount;
+
+	down_write(&OCFS2_I(lqinode)->ip_alloc_sem);
+	chunk = ocfs2_find_free_entry(sb, type, &offset);
+	if (!chunk) {
+		chunk = ocfs2_extend_local_quota_file(sb, type, &offset);
+		if (IS_ERR(chunk)) {
+			status = PTR_ERR(chunk);
+			goto out;
+		}
+	} else if (IS_ERR(chunk)) {
+		status = PTR_ERR(chunk);
+		goto out;
+	}
+	od->dq_local_off = ol_dqblk_off(sb, chunk->qc_num, offset);
+	od->dq_chunk = chunk;
+	status = ocfs2_extent_map_get_blocks(lqinode,
+				     ol_dqblk_block(sb, chunk->qc_num, offset),
+				     &od->dq_local_phys_blk,
+				     &pcount,
+				     NULL);
+
+	/* Initialize dquot structure on disk */
+	status = ocfs2_local_write_dquot(dquot);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	/* Mark structure as allocated */
+	status = ocfs2_modify_bh(lqinode, chunk->qc_headerbh, olq_alloc_dquot,
+				 &offset);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+out:
+	up_write(&OCFS2_I(lqinode)->ip_alloc_sem);
+	return status;
+}
+
+/*
+ * Release dquot structure from local quota file. ocfs2_release_dquot() has
+ * already started a transaction and written all changes to global quota file
+ */
+int ocfs2_local_release_dquot(handle_t *handle, struct dquot *dquot)
+{
+	int status;
+	int type = dquot->dq_id.type;
+	struct ocfs2_dquot *od = OCFS2_DQUOT(dquot);
+	struct super_block *sb = dquot->dq_sb;
+	struct ocfs2_local_disk_chunk *dchunk;
+	int offset;
+
+	status = ocfs2_journal_access_dq(handle,
+			INODE_CACHE(sb_dqopt(sb)->files[type]),
+			od->dq_chunk->qc_headerbh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+	offset = ol_dqblk_chunk_off(sb, od->dq_chunk->qc_num,
+					     od->dq_local_off);
+	dchunk = (struct ocfs2_local_disk_chunk *)
+			(od->dq_chunk->qc_headerbh->b_data);
+	/* Mark structure as freed */
+	lock_buffer(od->dq_chunk->qc_headerbh);
+	ocfs2_clear_bit_unaligned(offset, dchunk->dqc_bitmap);
+	le32_add_cpu(&dchunk->dqc_free, 1);
+	unlock_buffer(od->dq_chunk->qc_headerbh);
+	ocfs2_journal_dirty(handle, od->dq_chunk->qc_headerbh);
+
+out:
+	return status;
+}
+
+static const struct quota_format_ops ocfs2_format_ops = {
+	.check_quota_file	= ocfs2_local_check_quota_file,
+	.read_file_info		= ocfs2_local_read_info,
+	.write_file_info	= ocfs2_global_write_info,
+	.free_file_info		= ocfs2_local_free_info,
+};
+
+struct quota_format_type ocfs2_quota_format = {
+	.qf_fmt_id = QFMT_OCFS2,
+	.qf_ops = &ocfs2_format_ops,
+	.qf_owner = THIS_MODULE
+};
diff --git a/fs/ocfs2/refcounttree.c b/fs/ocfs2/refcounttree.c
new file mode 100644
index 000000000..d8c6af101
--- /dev/null
+++ b/fs/ocfs2/refcounttree.c
@@ -0,0 +1,4474 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * refcounttree.c
+ *
+ * Copyright (C) 2009 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/sort.h>
+#include <cluster/masklog.h>
+#include "ocfs2.h"
+#include "inode.h"
+#include "alloc.h"
+#include "suballoc.h"
+#include "journal.h"
+#include "uptodate.h"
+#include "super.h"
+#include "buffer_head_io.h"
+#include "blockcheck.h"
+#include "refcounttree.h"
+#include "sysfile.h"
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "aops.h"
+#include "xattr.h"
+#include "namei.h"
+#include "ocfs2_trace.h"
+
+#include <linux/bio.h>
+#include <linux/blkdev.h>
+#include <linux/slab.h>
+#include <linux/writeback.h>
+#include <linux/pagevec.h>
+#include <linux/swap.h>
+#include <linux/security.h>
+#include <linux/fsnotify.h>
+#include <linux/quotaops.h>
+#include <linux/namei.h>
+#include <linux/mount.h>
+#include <linux/posix_acl.h>
+
+struct ocfs2_cow_context {
+	struct inode *inode;
+	u32 cow_start;
+	u32 cow_len;
+	struct ocfs2_extent_tree data_et;
+	struct ocfs2_refcount_tree *ref_tree;
+	struct buffer_head *ref_root_bh;
+	struct ocfs2_alloc_context *meta_ac;
+	struct ocfs2_alloc_context *data_ac;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	void *cow_object;
+	struct ocfs2_post_refcount *post_refcount;
+	int extra_credits;
+	int (*get_clusters)(struct ocfs2_cow_context *context,
+			    u32 v_cluster, u32 *p_cluster,
+			    u32 *num_clusters,
+			    unsigned int *extent_flags);
+	int (*cow_duplicate_clusters)(handle_t *handle,
+				      struct inode *inode,
+				      u32 cpos, u32 old_cluster,
+				      u32 new_cluster, u32 new_len);
+};
+
+static inline struct ocfs2_refcount_tree *
+cache_info_to_refcount(struct ocfs2_caching_info *ci)
+{
+	return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
+}
+
+static int ocfs2_validate_refcount_block(struct super_block *sb,
+					 struct buffer_head *bh)
+{
+	int rc;
+	struct ocfs2_refcount_block *rb =
+		(struct ocfs2_refcount_block *)bh->b_data;
+
+	trace_ocfs2_validate_refcount_block((unsigned long long)bh->b_blocknr);
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	/*
+	 * If the ecc fails, we return the error but otherwise
+	 * leave the filesystem running.  We know any error is
+	 * local to this block.
+	 */
+	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
+	if (rc) {
+		mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
+		     (unsigned long long)bh->b_blocknr);
+		return rc;
+	}
+
+
+	if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
+		ocfs2_error(sb,
+			    "Refcount block #%llu has bad signature %.*s",
+			    (unsigned long long)bh->b_blocknr, 7,
+			    rb->rf_signature);
+		return -EINVAL;
+	}
+
+	if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
+		ocfs2_error(sb,
+			    "Refcount block #%llu has an invalid rf_blkno "
+			    "of %llu",
+			    (unsigned long long)bh->b_blocknr,
+			    (unsigned long long)le64_to_cpu(rb->rf_blkno));
+		return -EINVAL;
+	}
+
+	if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
+		ocfs2_error(sb,
+			    "Refcount block #%llu has an invalid "
+			    "rf_fs_generation of #%u",
+			    (unsigned long long)bh->b_blocknr,
+			    le32_to_cpu(rb->rf_fs_generation));
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
+				     u64 rb_blkno,
+				     struct buffer_head **bh)
+{
+	int rc;
+	struct buffer_head *tmp = *bh;
+
+	rc = ocfs2_read_block(ci, rb_blkno, &tmp,
+			      ocfs2_validate_refcount_block);
+
+	/* If ocfs2_read_block() got us a new bh, pass it up. */
+	if (!rc && !*bh)
+		*bh = tmp;
+
+	return rc;
+}
+
+static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	return rf->rf_blkno;
+}
+
+static struct super_block *
+ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	return rf->rf_sb;
+}
+
+static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	spin_lock(&rf->rf_lock);
+}
+
+static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	spin_unlock(&rf->rf_lock);
+}
+
+static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	mutex_lock(&rf->rf_io_mutex);
+}
+
+static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
+{
+	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
+
+	mutex_unlock(&rf->rf_io_mutex);
+}
+
+static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
+	.co_owner		= ocfs2_refcount_cache_owner,
+	.co_get_super		= ocfs2_refcount_cache_get_super,
+	.co_cache_lock		= ocfs2_refcount_cache_lock,
+	.co_cache_unlock	= ocfs2_refcount_cache_unlock,
+	.co_io_lock		= ocfs2_refcount_cache_io_lock,
+	.co_io_unlock		= ocfs2_refcount_cache_io_unlock,
+};
+
+static struct ocfs2_refcount_tree *
+ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
+{
+	struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
+	struct ocfs2_refcount_tree *tree = NULL;
+
+	while (n) {
+		tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
+
+		if (blkno < tree->rf_blkno)
+			n = n->rb_left;
+		else if (blkno > tree->rf_blkno)
+			n = n->rb_right;
+		else
+			return tree;
+	}
+
+	return NULL;
+}
+
+/* osb_lock is already locked. */
+static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
+				       struct ocfs2_refcount_tree *new)
+{
+	u64 rf_blkno = new->rf_blkno;
+	struct rb_node *parent = NULL;
+	struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
+	struct ocfs2_refcount_tree *tmp;
+
+	while (*p) {
+		parent = *p;
+
+		tmp = rb_entry(parent, struct ocfs2_refcount_tree,
+			       rf_node);
+
+		if (rf_blkno < tmp->rf_blkno)
+			p = &(*p)->rb_left;
+		else if (rf_blkno > tmp->rf_blkno)
+			p = &(*p)->rb_right;
+		else {
+			/* This should never happen! */
+			mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
+			     (unsigned long long)rf_blkno);
+			BUG();
+		}
+	}
+
+	rb_link_node(&new->rf_node, parent, p);
+	rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
+}
+
+static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
+{
+	ocfs2_metadata_cache_exit(&tree->rf_ci);
+	ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
+	ocfs2_lock_res_free(&tree->rf_lockres);
+	kfree(tree);
+}
+
+static inline void
+ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
+					struct ocfs2_refcount_tree *tree)
+{
+	rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
+	if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
+		osb->osb_ref_tree_lru = NULL;
+}
+
+static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
+					struct ocfs2_refcount_tree *tree)
+{
+	spin_lock(&osb->osb_lock);
+	ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
+	spin_unlock(&osb->osb_lock);
+}
+
+static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
+{
+	struct ocfs2_refcount_tree *tree =
+		container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
+
+	ocfs2_free_refcount_tree(tree);
+}
+
+static inline void
+ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
+{
+	kref_get(&tree->rf_getcnt);
+}
+
+static inline void
+ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
+{
+	kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
+}
+
+static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
+					       struct super_block *sb)
+{
+	ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
+	mutex_init(&new->rf_io_mutex);
+	new->rf_sb = sb;
+	spin_lock_init(&new->rf_lock);
+}
+
+static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
+					struct ocfs2_refcount_tree *new,
+					u64 rf_blkno, u32 generation)
+{
+	init_rwsem(&new->rf_sem);
+	ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
+				     rf_blkno, generation);
+}
+
+static struct ocfs2_refcount_tree*
+ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
+{
+	struct ocfs2_refcount_tree *new;
+
+	new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
+	if (!new)
+		return NULL;
+
+	new->rf_blkno = rf_blkno;
+	kref_init(&new->rf_getcnt);
+	ocfs2_init_refcount_tree_ci(new, osb->sb);
+
+	return new;
+}
+
+static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
+				   struct ocfs2_refcount_tree **ret_tree)
+{
+	int ret = 0;
+	struct ocfs2_refcount_tree *tree, *new = NULL;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_block *ref_rb;
+
+	spin_lock(&osb->osb_lock);
+	if (osb->osb_ref_tree_lru &&
+	    osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
+		tree = osb->osb_ref_tree_lru;
+	else
+		tree = ocfs2_find_refcount_tree(osb, rf_blkno);
+	if (tree)
+		goto out;
+
+	spin_unlock(&osb->osb_lock);
+
+	new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
+	if (!new) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		return ret;
+	}
+	/*
+	 * We need the generation to create the refcount tree lock and since
+	 * it isn't changed during the tree modification, we are safe here to
+	 * read without protection.
+	 * We also have to purge the cache after we create the lock since the
+	 * refcount block may have the stale data. It can only be trusted when
+	 * we hold the refcount lock.
+	 */
+	ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		ocfs2_metadata_cache_exit(&new->rf_ci);
+		kfree(new);
+		return ret;
+	}
+
+	ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
+	ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
+				      new->rf_generation);
+	ocfs2_metadata_cache_purge(&new->rf_ci);
+
+	spin_lock(&osb->osb_lock);
+	tree = ocfs2_find_refcount_tree(osb, rf_blkno);
+	if (tree)
+		goto out;
+
+	ocfs2_insert_refcount_tree(osb, new);
+
+	tree = new;
+	new = NULL;
+
+out:
+	*ret_tree = tree;
+
+	osb->osb_ref_tree_lru = tree;
+
+	spin_unlock(&osb->osb_lock);
+
+	if (new)
+		ocfs2_free_refcount_tree(new);
+
+	brelse(ref_root_bh);
+	return ret;
+}
+
+static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di;
+
+	ret = ocfs2_read_inode_block(inode, &di_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+	*ref_blkno = le64_to_cpu(di->i_refcount_loc);
+	brelse(di_bh);
+out:
+	return ret;
+}
+
+static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
+				      struct ocfs2_refcount_tree *tree, int rw)
+{
+	int ret;
+
+	ret = ocfs2_refcount_lock(tree, rw);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (rw)
+		down_write(&tree->rf_sem);
+	else
+		down_read(&tree->rf_sem);
+
+out:
+	return ret;
+}
+
+/*
+ * Lock the refcount tree pointed by ref_blkno and return the tree.
+ * In most case, we lock the tree and read the refcount block.
+ * So read it here if the caller really needs it.
+ *
+ * If the tree has been re-created by other node, it will free the
+ * old one and re-create it.
+ */
+int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
+			     u64 ref_blkno, int rw,
+			     struct ocfs2_refcount_tree **ret_tree,
+			     struct buffer_head **ref_bh)
+{
+	int ret, delete_tree = 0;
+	struct ocfs2_refcount_tree *tree = NULL;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_block *rb;
+
+again:
+	ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	ocfs2_refcount_tree_get(tree);
+
+	ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
+	if (ret) {
+		mlog_errno(ret);
+		ocfs2_refcount_tree_put(tree);
+		goto out;
+	}
+
+	ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
+					&ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		ocfs2_unlock_refcount_tree(osb, tree, rw);
+		ocfs2_refcount_tree_put(tree);
+		goto out;
+	}
+
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	/*
+	 * If the refcount block has been freed and re-created, we may need
+	 * to recreate the refcount tree also.
+	 *
+	 * Here we just remove the tree from the rb-tree, and the last
+	 * kref holder will unlock and delete this refcount_tree.
+	 * Then we goto "again" and ocfs2_get_refcount_tree will create
+	 * the new refcount tree for us.
+	 */
+	if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
+		if (!tree->rf_removed) {
+			ocfs2_erase_refcount_tree_from_list(osb, tree);
+			tree->rf_removed = 1;
+			delete_tree = 1;
+		}
+
+		ocfs2_unlock_refcount_tree(osb, tree, rw);
+		/*
+		 * We get an extra reference when we create the refcount
+		 * tree, so another put will destroy it.
+		 */
+		if (delete_tree)
+			ocfs2_refcount_tree_put(tree);
+		brelse(ref_root_bh);
+		ref_root_bh = NULL;
+		goto again;
+	}
+
+	*ret_tree = tree;
+	if (ref_bh) {
+		*ref_bh = ref_root_bh;
+		ref_root_bh = NULL;
+	}
+out:
+	brelse(ref_root_bh);
+	return ret;
+}
+
+void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
+				struct ocfs2_refcount_tree *tree, int rw)
+{
+	if (rw)
+		up_write(&tree->rf_sem);
+	else
+		up_read(&tree->rf_sem);
+
+	ocfs2_refcount_unlock(tree, rw);
+	ocfs2_refcount_tree_put(tree);
+}
+
+void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
+{
+	struct rb_node *node;
+	struct ocfs2_refcount_tree *tree;
+	struct rb_root *root = &osb->osb_rf_lock_tree;
+
+	while ((node = rb_last(root)) != NULL) {
+		tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
+
+		trace_ocfs2_purge_refcount_trees(
+				(unsigned long long) tree->rf_blkno);
+
+		rb_erase(&tree->rf_node, root);
+		ocfs2_free_refcount_tree(tree);
+	}
+}
+
+/*
+ * Create a refcount tree for an inode.
+ * We take for granted that the inode is already locked.
+ */
+static int ocfs2_create_refcount_tree(struct inode *inode,
+				      struct buffer_head *di_bh)
+{
+	int ret;
+	handle_t *handle = NULL;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
+	u16 suballoc_bit_start;
+	u32 num_got;
+	u64 suballoc_loc, first_blkno;
+
+	BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
+
+	trace_ocfs2_create_refcount_tree(
+		(unsigned long long)OCFS2_I(inode)->ip_blkno);
+
+	ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
+				   &suballoc_bit_start, &num_got,
+				   &first_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
+	if (!new_tree) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	new_bh = sb_getblk(inode->i_sb, first_blkno);
+	if (!new_bh) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out_commit;
+	}
+	ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
+
+	ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	/* Initialize ocfs2_refcount_block. */
+	rb = (struct ocfs2_refcount_block *)new_bh->b_data;
+	memset(rb, 0, inode->i_sb->s_blocksize);
+	strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
+	rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
+	rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
+	rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
+	rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
+	rb->rf_blkno = cpu_to_le64(first_blkno);
+	rb->rf_count = cpu_to_le32(1);
+	rb->rf_records.rl_count =
+			cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
+	spin_lock(&osb->osb_lock);
+	rb->rf_generation = osb->s_next_generation++;
+	spin_unlock(&osb->osb_lock);
+
+	ocfs2_journal_dirty(handle, new_bh);
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	di->i_refcount_loc = cpu_to_le64(first_blkno);
+	spin_unlock(&oi->ip_lock);
+
+	trace_ocfs2_create_refcount_tree_blkno((unsigned long long)first_blkno);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+	/*
+	 * We have to init the tree lock here since it will use
+	 * the generation number to create it.
+	 */
+	new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
+	ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
+				      new_tree->rf_generation);
+
+	spin_lock(&osb->osb_lock);
+	tree = ocfs2_find_refcount_tree(osb, first_blkno);
+
+	/*
+	 * We've just created a new refcount tree in this block.  If
+	 * we found a refcount tree on the ocfs2_super, it must be
+	 * one we just deleted.  We free the old tree before
+	 * inserting the new tree.
+	 */
+	BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
+	if (tree)
+		ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
+	ocfs2_insert_refcount_tree(osb, new_tree);
+	spin_unlock(&osb->osb_lock);
+	new_tree = NULL;
+	if (tree)
+		ocfs2_refcount_tree_put(tree);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	if (new_tree) {
+		ocfs2_metadata_cache_exit(&new_tree->rf_ci);
+		kfree(new_tree);
+	}
+
+	brelse(new_bh);
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	return ret;
+}
+
+static int ocfs2_set_refcount_tree(struct inode *inode,
+				   struct buffer_head *di_bh,
+				   u64 refcount_loc)
+{
+	int ret;
+	handle_t *handle = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_refcount_tree *ref_tree;
+
+	BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
+
+	ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
+				       &ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	le32_add_cpu(&rb->rf_count, 1);
+
+	ocfs2_journal_dirty(handle, ref_root_bh);
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	di->i_refcount_loc = cpu_to_le64(refcount_loc);
+	spin_unlock(&oi->ip_lock);
+	ocfs2_journal_dirty(handle, di_bh);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	brelse(ref_root_bh);
+
+	return ret;
+}
+
+int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
+{
+	int ret, delete_tree = 0;
+	handle_t *handle = NULL;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_refcount_block *rb;
+	struct inode *alloc_inode = NULL;
+	struct buffer_head *alloc_bh = NULL;
+	struct buffer_head *blk_bh = NULL;
+	struct ocfs2_refcount_tree *ref_tree;
+	int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
+	u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
+	u16 bit = 0;
+
+	if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
+		return 0;
+
+	BUG_ON(!ref_blkno);
+	ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
+
+	/*
+	 * If we are the last user, we need to free the block.
+	 * So lock the allocator ahead.
+	 */
+	if (le32_to_cpu(rb->rf_count) == 1) {
+		blk = le64_to_cpu(rb->rf_blkno);
+		bit = le16_to_cpu(rb->rf_suballoc_bit);
+		if (rb->rf_suballoc_loc)
+			bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
+		else
+			bg_blkno = ocfs2_which_suballoc_group(blk, bit);
+
+		alloc_inode = ocfs2_get_system_file_inode(osb,
+					EXTENT_ALLOC_SYSTEM_INODE,
+					le16_to_cpu(rb->rf_suballoc_slot));
+		if (!alloc_inode) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			goto out;
+		}
+		mutex_lock(&alloc_inode->i_mutex);
+
+		ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_mutex;
+		}
+
+		credits += OCFS2_SUBALLOC_FREE;
+	}
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	di->i_refcount_loc = 0;
+	spin_unlock(&oi->ip_lock);
+	ocfs2_journal_dirty(handle, di_bh);
+
+	le32_add_cpu(&rb->rf_count , -1);
+	ocfs2_journal_dirty(handle, blk_bh);
+
+	if (!rb->rf_count) {
+		delete_tree = 1;
+		ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
+		ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
+					       alloc_bh, bit, bg_blkno, 1);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out_unlock:
+	if (alloc_inode) {
+		ocfs2_inode_unlock(alloc_inode, 1);
+		brelse(alloc_bh);
+	}
+out_mutex:
+	if (alloc_inode) {
+		mutex_unlock(&alloc_inode->i_mutex);
+		iput(alloc_inode);
+	}
+out:
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	if (delete_tree)
+		ocfs2_refcount_tree_put(ref_tree);
+	brelse(blk_bh);
+
+	return ret;
+}
+
+static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
+					  struct buffer_head *ref_leaf_bh,
+					  u64 cpos, unsigned int len,
+					  struct ocfs2_refcount_rec *ret_rec,
+					  int *index)
+{
+	int i = 0;
+	struct ocfs2_refcount_block *rb =
+		(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_rec *rec = NULL;
+
+	for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
+		rec = &rb->rf_records.rl_recs[i];
+
+		if (le64_to_cpu(rec->r_cpos) +
+		    le32_to_cpu(rec->r_clusters) <= cpos)
+			continue;
+		else if (le64_to_cpu(rec->r_cpos) > cpos)
+			break;
+
+		/* ok, cpos fail in this rec. Just return. */
+		if (ret_rec)
+			*ret_rec = *rec;
+		goto out;
+	}
+
+	if (ret_rec) {
+		/* We meet with a hole here, so fake the rec. */
+		ret_rec->r_cpos = cpu_to_le64(cpos);
+		ret_rec->r_refcount = 0;
+		if (i < le16_to_cpu(rb->rf_records.rl_used) &&
+		    le64_to_cpu(rec->r_cpos) < cpos + len)
+			ret_rec->r_clusters =
+				cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
+		else
+			ret_rec->r_clusters = cpu_to_le32(len);
+	}
+
+out:
+	*index = i;
+}
+
+/*
+ * Try to remove refcount tree. The mechanism is:
+ * 1) Check whether i_clusters == 0, if no, exit.
+ * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
+ * 3) Check whether we have inline xattr stored outside, if yes, exit.
+ * 4) Remove the tree.
+ */
+int ocfs2_try_remove_refcount_tree(struct inode *inode,
+				   struct buffer_head *di_bh)
+{
+	int ret;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	down_write(&oi->ip_xattr_sem);
+	down_write(&oi->ip_alloc_sem);
+
+	if (oi->ip_clusters)
+		goto out;
+
+	if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
+		goto out;
+
+	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
+	    ocfs2_has_inline_xattr_value_outside(inode, di))
+		goto out;
+
+	ret = ocfs2_remove_refcount_tree(inode, di_bh);
+	if (ret)
+		mlog_errno(ret);
+out:
+	up_write(&oi->ip_alloc_sem);
+	up_write(&oi->ip_xattr_sem);
+	return 0;
+}
+
+/*
+ * Find the end range for a leaf refcount block indicated by
+ * el->l_recs[index].e_blkno.
+ */
+static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
+				       struct buffer_head *ref_root_bh,
+				       struct ocfs2_extent_block *eb,
+				       struct ocfs2_extent_list *el,
+				       int index,  u32 *cpos_end)
+{
+	int ret, i, subtree_root;
+	u32 cpos;
+	u64 blkno;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct ocfs2_path *left_path = NULL, *right_path = NULL;
+	struct ocfs2_extent_tree et;
+	struct ocfs2_extent_list *tmp_el;
+
+	if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
+		/*
+		 * We have a extent rec after index, so just use the e_cpos
+		 * of the next extent rec.
+		 */
+		*cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
+		return 0;
+	}
+
+	if (!eb || (eb && !eb->h_next_leaf_blk)) {
+		/*
+		 * We are the last extent rec, so any high cpos should
+		 * be stored in this leaf refcount block.
+		 */
+		*cpos_end = UINT_MAX;
+		return 0;
+	}
+
+	/*
+	 * If the extent block isn't the last one, we have to find
+	 * the subtree root between this extent block and the next
+	 * leaf extent block and get the corresponding e_cpos from
+	 * the subroot. Otherwise we may corrupt the b-tree.
+	 */
+	ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
+
+	left_path = ocfs2_new_path_from_et(&et);
+	if (!left_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
+	ret = ocfs2_find_path(ci, left_path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	right_path = ocfs2_new_path_from_path(left_path);
+	if (!right_path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(ci, right_path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	subtree_root = ocfs2_find_subtree_root(&et, left_path,
+					       right_path);
+
+	tmp_el = left_path->p_node[subtree_root].el;
+	blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
+	for (i = 0; i < le16_to_cpu(tmp_el->l_next_free_rec); i++) {
+		if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
+			*cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
+			break;
+		}
+	}
+
+	BUG_ON(i == le16_to_cpu(tmp_el->l_next_free_rec));
+
+out:
+	ocfs2_free_path(left_path);
+	ocfs2_free_path(right_path);
+	return ret;
+}
+
+/*
+ * Given a cpos and len, try to find the refcount record which contains cpos.
+ * 1. If cpos can be found in one refcount record, return the record.
+ * 2. If cpos can't be found, return a fake record which start from cpos
+ *    and end at a small value between cpos+len and start of the next record.
+ *    This fake record has r_refcount = 0.
+ */
+static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
+				  struct buffer_head *ref_root_bh,
+				  u64 cpos, unsigned int len,
+				  struct ocfs2_refcount_rec *ret_rec,
+				  int *index,
+				  struct buffer_head **ret_bh)
+{
+	int ret = 0, i, found;
+	u32 low_cpos, uninitialized_var(cpos_end);
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec = NULL;
+	struct ocfs2_extent_block *eb = NULL;
+	struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+	if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
+		ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
+					      ret_rec, index);
+		*ret_bh = ref_root_bh;
+		get_bh(ref_root_bh);
+		return 0;
+	}
+
+	el = &rb->rf_list;
+	low_cpos = cpos & OCFS2_32BIT_POS_MASK;
+
+	if (el->l_tree_depth) {
+		ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+
+		if (el->l_tree_depth) {
+			ocfs2_error(sb,
+			"refcount tree %llu has non zero tree "
+			"depth in leaf btree tree block %llu\n",
+			(unsigned long long)ocfs2_metadata_cache_owner(ci),
+			(unsigned long long)eb_bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
+		}
+	}
+
+	found = 0;
+	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
+		rec = &el->l_recs[i];
+
+		if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
+			found = 1;
+			break;
+		}
+	}
+
+	if (found) {
+		ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
+						  eb, el, i, &cpos_end);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (cpos_end < low_cpos + len)
+			len = cpos_end - low_cpos;
+	}
+
+	ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
+					&ref_leaf_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
+				      ret_rec, index);
+	*ret_bh = ref_leaf_bh;
+out:
+	brelse(eb_bh);
+	return ret;
+}
+
+enum ocfs2_ref_rec_contig {
+	REF_CONTIG_NONE = 0,
+	REF_CONTIG_LEFT,
+	REF_CONTIG_RIGHT,
+	REF_CONTIG_LEFTRIGHT,
+};
+
+static enum ocfs2_ref_rec_contig
+	ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
+				    int index)
+{
+	if ((rb->rf_records.rl_recs[index].r_refcount ==
+	    rb->rf_records.rl_recs[index + 1].r_refcount) &&
+	    (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
+	    le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
+	    le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
+		return REF_CONTIG_RIGHT;
+
+	return REF_CONTIG_NONE;
+}
+
+static enum ocfs2_ref_rec_contig
+	ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
+				  int index)
+{
+	enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
+
+	if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
+		ret = ocfs2_refcount_rec_adjacent(rb, index);
+
+	if (index > 0) {
+		enum ocfs2_ref_rec_contig tmp;
+
+		tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
+
+		if (tmp == REF_CONTIG_RIGHT) {
+			if (ret == REF_CONTIG_RIGHT)
+				ret = REF_CONTIG_LEFTRIGHT;
+			else
+				ret = REF_CONTIG_LEFT;
+		}
+	}
+
+	return ret;
+}
+
+static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
+					   int index)
+{
+	BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
+	       rb->rf_records.rl_recs[index+1].r_refcount);
+
+	le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
+		     le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
+
+	if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
+		memmove(&rb->rf_records.rl_recs[index + 1],
+			&rb->rf_records.rl_recs[index + 2],
+			sizeof(struct ocfs2_refcount_rec) *
+			(le16_to_cpu(rb->rf_records.rl_used) - index - 2));
+
+	memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
+	       0, sizeof(struct ocfs2_refcount_rec));
+	le16_add_cpu(&rb->rf_records.rl_used, -1);
+}
+
+/*
+ * Merge the refcount rec if we are contiguous with the adjacent recs.
+ */
+static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
+				     int index)
+{
+	enum ocfs2_ref_rec_contig contig =
+				ocfs2_refcount_rec_contig(rb, index);
+
+	if (contig == REF_CONTIG_NONE)
+		return;
+
+	if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
+		BUG_ON(index == 0);
+		index--;
+	}
+
+	ocfs2_rotate_refcount_rec_left(rb, index);
+
+	if (contig == REF_CONTIG_LEFTRIGHT)
+		ocfs2_rotate_refcount_rec_left(rb, index);
+}
+
+/*
+ * Change the refcount indexed by "index" in ref_bh.
+ * If refcount reaches 0, remove it.
+ */
+static int ocfs2_change_refcount_rec(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_leaf_bh,
+				     int index, int merge, int change)
+{
+	int ret;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_list *rl = &rb->rf_records;
+	struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_change_refcount_rec(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		index, le32_to_cpu(rec->r_refcount), change);
+	le32_add_cpu(&rec->r_refcount, change);
+
+	if (!rec->r_refcount) {
+		if (index != le16_to_cpu(rl->rl_used) - 1) {
+			memmove(rec, rec + 1,
+				(le16_to_cpu(rl->rl_used) - index - 1) *
+				sizeof(struct ocfs2_refcount_rec));
+			memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
+			       0, sizeof(struct ocfs2_refcount_rec));
+		}
+
+		le16_add_cpu(&rl->rl_used, -1);
+	} else if (merge)
+		ocfs2_refcount_rec_merge(rb, index);
+
+	ocfs2_journal_dirty(handle, ref_leaf_bh);
+out:
+	return ret;
+}
+
+static int ocfs2_expand_inline_ref_root(handle_t *handle,
+					struct ocfs2_caching_info *ci,
+					struct buffer_head *ref_root_bh,
+					struct buffer_head **ref_leaf_bh,
+					struct ocfs2_alloc_context *meta_ac)
+{
+	int ret;
+	u16 suballoc_bit_start;
+	u32 num_got;
+	u64 suballoc_loc, blkno;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_refcount_block *new_rb;
+	struct ocfs2_refcount_block *root_rb =
+			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
+				   &suballoc_bit_start, &num_got,
+				   &blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	new_bh = sb_getblk(sb, blkno);
+	if (new_bh == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+	ocfs2_set_new_buffer_uptodate(ci, new_bh);
+
+	ret = ocfs2_journal_access_rb(handle, ci, new_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Initialize ocfs2_refcount_block.
+	 * It should contain the same information as the old root.
+	 * so just memcpy it and change the corresponding field.
+	 */
+	memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
+
+	new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
+	new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
+	new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
+	new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
+	new_rb->rf_blkno = cpu_to_le64(blkno);
+	new_rb->rf_cpos = cpu_to_le32(0);
+	new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
+	new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
+	ocfs2_journal_dirty(handle, new_bh);
+
+	/* Now change the root. */
+	memset(&root_rb->rf_list, 0, sb->s_blocksize -
+	       offsetof(struct ocfs2_refcount_block, rf_list));
+	root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
+	root_rb->rf_clusters = cpu_to_le32(1);
+	root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
+	root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
+	root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
+	root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
+
+	ocfs2_journal_dirty(handle, ref_root_bh);
+
+	trace_ocfs2_expand_inline_ref_root((unsigned long long)blkno,
+		le16_to_cpu(new_rb->rf_records.rl_used));
+
+	*ref_leaf_bh = new_bh;
+	new_bh = NULL;
+out:
+	brelse(new_bh);
+	return ret;
+}
+
+static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
+					   struct ocfs2_refcount_rec *next)
+{
+	if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
+		ocfs2_get_ref_rec_low_cpos(next))
+		return 1;
+
+	return 0;
+}
+
+static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
+{
+	const struct ocfs2_refcount_rec *l = a, *r = b;
+	u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
+	u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
+
+	if (l_cpos > r_cpos)
+		return 1;
+	if (l_cpos < r_cpos)
+		return -1;
+	return 0;
+}
+
+static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
+{
+	const struct ocfs2_refcount_rec *l = a, *r = b;
+	u64 l_cpos = le64_to_cpu(l->r_cpos);
+	u64 r_cpos = le64_to_cpu(r->r_cpos);
+
+	if (l_cpos > r_cpos)
+		return 1;
+	if (l_cpos < r_cpos)
+		return -1;
+	return 0;
+}
+
+static void swap_refcount_rec(void *a, void *b, int size)
+{
+	struct ocfs2_refcount_rec *l = a, *r = b, tmp;
+
+	tmp = *l;
+	*l = *r;
+	*r = tmp;
+}
+
+/*
+ * The refcount cpos are ordered by their 64bit cpos,
+ * But we will use the low 32 bit to be the e_cpos in the b-tree.
+ * So we need to make sure that this pos isn't intersected with others.
+ *
+ * Note: The refcount block is already sorted by their low 32 bit cpos,
+ *       So just try the middle pos first, and we will exit when we find
+ *       the good position.
+ */
+static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
+					 u32 *split_pos, int *split_index)
+{
+	int num_used = le16_to_cpu(rl->rl_used);
+	int delta, middle = num_used / 2;
+
+	for (delta = 0; delta < middle; delta++) {
+		/* Let's check delta earlier than middle */
+		if (ocfs2_refcount_rec_no_intersect(
+					&rl->rl_recs[middle - delta - 1],
+					&rl->rl_recs[middle - delta])) {
+			*split_index = middle - delta;
+			break;
+		}
+
+		/* For even counts, don't walk off the end */
+		if ((middle + delta + 1) == num_used)
+			continue;
+
+		/* Now try delta past middle */
+		if (ocfs2_refcount_rec_no_intersect(
+					&rl->rl_recs[middle + delta],
+					&rl->rl_recs[middle + delta + 1])) {
+			*split_index = middle + delta + 1;
+			break;
+		}
+	}
+
+	if (delta >= middle)
+		return -ENOSPC;
+
+	*split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
+	return 0;
+}
+
+static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
+					    struct buffer_head *new_bh,
+					    u32 *split_cpos)
+{
+	int split_index = 0, num_moved, ret;
+	u32 cpos = 0;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_list *rl = &rb->rf_records;
+	struct ocfs2_refcount_block *new_rb =
+			(struct ocfs2_refcount_block *)new_bh->b_data;
+	struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
+
+	trace_ocfs2_divide_leaf_refcount_block(
+		(unsigned long long)ref_leaf_bh->b_blocknr,
+		le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
+
+	/*
+	 * XXX: Improvement later.
+	 * If we know all the high 32 bit cpos is the same, no need to sort.
+	 *
+	 * In order to make the whole process safe, we do:
+	 * 1. sort the entries by their low 32 bit cpos first so that we can
+	 *    find the split cpos easily.
+	 * 2. call ocfs2_insert_extent to insert the new refcount block.
+	 * 3. move the refcount rec to the new block.
+	 * 4. sort the entries by their 64 bit cpos.
+	 * 5. dirty the new_rb and rb.
+	 */
+	sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
+	     sizeof(struct ocfs2_refcount_rec),
+	     cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
+
+	ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	new_rb->rf_cpos = cpu_to_le32(cpos);
+
+	/* move refcount records starting from split_index to the new block. */
+	num_moved = le16_to_cpu(rl->rl_used) - split_index;
+	memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
+	       num_moved * sizeof(struct ocfs2_refcount_rec));
+
+	/*ok, remove the entries we just moved over to the other block. */
+	memset(&rl->rl_recs[split_index], 0,
+	       num_moved * sizeof(struct ocfs2_refcount_rec));
+
+	/* change old and new rl_used accordingly. */
+	le16_add_cpu(&rl->rl_used, -num_moved);
+	new_rl->rl_used = cpu_to_le16(num_moved);
+
+	sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
+	     sizeof(struct ocfs2_refcount_rec),
+	     cmp_refcount_rec_by_cpos, swap_refcount_rec);
+
+	sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
+	     sizeof(struct ocfs2_refcount_rec),
+	     cmp_refcount_rec_by_cpos, swap_refcount_rec);
+
+	*split_cpos = cpos;
+	return 0;
+}
+
+static int ocfs2_new_leaf_refcount_block(handle_t *handle,
+					 struct ocfs2_caching_info *ci,
+					 struct buffer_head *ref_root_bh,
+					 struct buffer_head *ref_leaf_bh,
+					 struct ocfs2_alloc_context *meta_ac)
+{
+	int ret;
+	u16 suballoc_bit_start;
+	u32 num_got, new_cpos;
+	u64 suballoc_loc, blkno;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct ocfs2_refcount_block *root_rb =
+			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_refcount_block *new_rb;
+	struct ocfs2_extent_tree ref_et;
+
+	BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
+				   &suballoc_bit_start, &num_got,
+				   &blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	new_bh = sb_getblk(sb, blkno);
+	if (new_bh == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+	ocfs2_set_new_buffer_uptodate(ci, new_bh);
+
+	ret = ocfs2_journal_access_rb(handle, ci, new_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Initialize ocfs2_refcount_block. */
+	new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
+	memset(new_rb, 0, sb->s_blocksize);
+	strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
+	new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
+	new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
+	new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
+	new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
+	new_rb->rf_blkno = cpu_to_le64(blkno);
+	new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
+	new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
+	new_rb->rf_records.rl_count =
+				cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
+	new_rb->rf_generation = root_rb->rf_generation;
+
+	ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_journal_dirty(handle, ref_leaf_bh);
+	ocfs2_journal_dirty(handle, new_bh);
+
+	ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
+
+	trace_ocfs2_new_leaf_refcount_block(
+			(unsigned long long)new_bh->b_blocknr, new_cpos);
+
+	/* Insert the new leaf block with the specific offset cpos. */
+	ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
+				  1, 0, meta_ac);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	brelse(new_bh);
+	return ret;
+}
+
+static int ocfs2_expand_refcount_tree(handle_t *handle,
+				      struct ocfs2_caching_info *ci,
+				      struct buffer_head *ref_root_bh,
+				      struct buffer_head *ref_leaf_bh,
+				      struct ocfs2_alloc_context *meta_ac)
+{
+	int ret;
+	struct buffer_head *expand_bh = NULL;
+
+	if (ref_root_bh == ref_leaf_bh) {
+		/*
+		 * the old root bh hasn't been expanded to a b-tree,
+		 * so expand it first.
+		 */
+		ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
+						   &expand_bh, meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	} else {
+		expand_bh = ref_leaf_bh;
+		get_bh(expand_bh);
+	}
+
+
+	/* Now add a new refcount block into the tree.*/
+	ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
+					    expand_bh, meta_ac);
+	if (ret)
+		mlog_errno(ret);
+out:
+	brelse(expand_bh);
+	return ret;
+}
+
+/*
+ * Adjust the extent rec in b-tree representing ref_leaf_bh.
+ *
+ * Only called when we have inserted a new refcount rec at index 0
+ * which means ocfs2_extent_rec.e_cpos may need some change.
+ */
+static int ocfs2_adjust_refcount_rec(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_root_bh,
+				     struct buffer_head *ref_leaf_bh,
+				     struct ocfs2_refcount_rec *rec)
+{
+	int ret = 0, i;
+	u32 new_cpos, old_cpos;
+	struct ocfs2_path *path = NULL;
+	struct ocfs2_extent_tree et;
+	struct ocfs2_refcount_block *rb =
+		(struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	struct ocfs2_extent_list *el;
+
+	if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
+		goto out;
+
+	rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	old_cpos = le32_to_cpu(rb->rf_cpos);
+	new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
+	if (old_cpos <= new_cpos)
+		goto out;
+
+	ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
+
+	path = ocfs2_new_path_from_et(&et);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(ci, path, old_cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * 2 more credits, one for the leaf refcount block, one for
+	 * the extent block contains the extent rec.
+	 */
+	ret = ocfs2_extend_trans(handle, 2);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* change the leaf extent block first. */
+	el = path_leaf_el(path);
+
+	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
+		if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
+			break;
+
+	BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
+
+	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
+
+	/* change the r_cpos in the leaf block. */
+	rb->rf_cpos = cpu_to_le32(new_cpos);
+
+	ocfs2_journal_dirty(handle, path_leaf_bh(path));
+	ocfs2_journal_dirty(handle, ref_leaf_bh);
+
+out:
+	ocfs2_free_path(path);
+	return ret;
+}
+
+static int ocfs2_insert_refcount_rec(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_root_bh,
+				     struct buffer_head *ref_leaf_bh,
+				     struct ocfs2_refcount_rec *rec,
+				     int index, int merge,
+				     struct ocfs2_alloc_context *meta_ac)
+{
+	int ret;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_list *rf_list = &rb->rf_records;
+	struct buffer_head *new_bh = NULL;
+
+	BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
+
+	if (rf_list->rl_used == rf_list->rl_count) {
+		u64 cpos = le64_to_cpu(rec->r_cpos);
+		u32 len = le32_to_cpu(rec->r_clusters);
+
+		ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
+						 ref_leaf_bh, meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, len, NULL, &index,
+					     &new_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ref_leaf_bh = new_bh;
+		rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+		rf_list = &rb->rf_records;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (index < le16_to_cpu(rf_list->rl_used))
+		memmove(&rf_list->rl_recs[index + 1],
+			&rf_list->rl_recs[index],
+			(le16_to_cpu(rf_list->rl_used) - index) *
+			 sizeof(struct ocfs2_refcount_rec));
+
+	trace_ocfs2_insert_refcount_rec(
+		(unsigned long long)ref_leaf_bh->b_blocknr, index,
+		(unsigned long long)le64_to_cpu(rec->r_cpos),
+		le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
+
+	rf_list->rl_recs[index] = *rec;
+
+	le16_add_cpu(&rf_list->rl_used, 1);
+
+	if (merge)
+		ocfs2_refcount_rec_merge(rb, index);
+
+	ocfs2_journal_dirty(handle, ref_leaf_bh);
+
+	if (index == 0) {
+		ret = ocfs2_adjust_refcount_rec(handle, ci,
+						ref_root_bh,
+						ref_leaf_bh, rec);
+		if (ret)
+			mlog_errno(ret);
+	}
+out:
+	brelse(new_bh);
+	return ret;
+}
+
+/*
+ * Split the refcount_rec indexed by "index" in ref_leaf_bh.
+ * This is much simple than our b-tree code.
+ * split_rec is the new refcount rec we want to insert.
+ * If split_rec->r_refcount > 0, we are changing the refcount(in case we
+ * increase refcount or decrease a refcount to non-zero).
+ * If split_rec->r_refcount == 0, we are punching a hole in current refcount
+ * rec( in case we decrease a refcount to zero).
+ */
+static int ocfs2_split_refcount_rec(handle_t *handle,
+				    struct ocfs2_caching_info *ci,
+				    struct buffer_head *ref_root_bh,
+				    struct buffer_head *ref_leaf_bh,
+				    struct ocfs2_refcount_rec *split_rec,
+				    int index, int merge,
+				    struct ocfs2_alloc_context *meta_ac,
+				    struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret, recs_need;
+	u32 len;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_list *rf_list = &rb->rf_records;
+	struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
+	struct ocfs2_refcount_rec *tail_rec = NULL;
+	struct buffer_head *new_bh = NULL;
+
+	BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
+
+	trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
+		le32_to_cpu(orig_rec->r_clusters),
+		le32_to_cpu(orig_rec->r_refcount),
+		le64_to_cpu(split_rec->r_cpos),
+		le32_to_cpu(split_rec->r_clusters),
+		le32_to_cpu(split_rec->r_refcount));
+
+	/*
+	 * If we just need to split the header or tail clusters,
+	 * no more recs are needed, just split is OK.
+	 * Otherwise we at least need one new recs.
+	 */
+	if (!split_rec->r_refcount &&
+	    (split_rec->r_cpos == orig_rec->r_cpos ||
+	     le64_to_cpu(split_rec->r_cpos) +
+	     le32_to_cpu(split_rec->r_clusters) ==
+	     le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
+		recs_need = 0;
+	else
+		recs_need = 1;
+
+	/*
+	 * We need one more rec if we split in the middle and the new rec have
+	 * some refcount in it.
+	 */
+	if (split_rec->r_refcount &&
+	    (split_rec->r_cpos != orig_rec->r_cpos &&
+	     le64_to_cpu(split_rec->r_cpos) +
+	     le32_to_cpu(split_rec->r_clusters) !=
+	     le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
+		recs_need++;
+
+	/* If the leaf block don't have enough record, expand it. */
+	if (le16_to_cpu(rf_list->rl_used) + recs_need >
+					 le16_to_cpu(rf_list->rl_count)) {
+		struct ocfs2_refcount_rec tmp_rec;
+		u64 cpos = le64_to_cpu(orig_rec->r_cpos);
+		len = le32_to_cpu(orig_rec->r_clusters);
+		ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
+						 ref_leaf_bh, meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * We have to re-get it since now cpos may be moved to
+		 * another leaf block.
+		 */
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, len, &tmp_rec, &index,
+					     &new_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		ref_leaf_bh = new_bh;
+		rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+		rf_list = &rb->rf_records;
+		orig_rec = &rf_list->rl_recs[index];
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * We have calculated out how many new records we need and store
+	 * in recs_need, so spare enough space first by moving the records
+	 * after "index" to the end.
+	 */
+	if (index != le16_to_cpu(rf_list->rl_used) - 1)
+		memmove(&rf_list->rl_recs[index + 1 + recs_need],
+			&rf_list->rl_recs[index + 1],
+			(le16_to_cpu(rf_list->rl_used) - index - 1) *
+			 sizeof(struct ocfs2_refcount_rec));
+
+	len = (le64_to_cpu(orig_rec->r_cpos) +
+	      le32_to_cpu(orig_rec->r_clusters)) -
+	      (le64_to_cpu(split_rec->r_cpos) +
+	      le32_to_cpu(split_rec->r_clusters));
+
+	/*
+	 * If we have "len", the we will split in the tail and move it
+	 * to the end of the space we have just spared.
+	 */
+	if (len) {
+		tail_rec = &rf_list->rl_recs[index + recs_need];
+
+		memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
+		le64_add_cpu(&tail_rec->r_cpos,
+			     le32_to_cpu(tail_rec->r_clusters) - len);
+		tail_rec->r_clusters = cpu_to_le32(len);
+	}
+
+	/*
+	 * If the split pos isn't the same as the original one, we need to
+	 * split in the head.
+	 *
+	 * Note: We have the chance that split_rec.r_refcount = 0,
+	 * recs_need = 0 and len > 0, which means we just cut the head from
+	 * the orig_rec and in that case we have done some modification in
+	 * orig_rec above, so the check for r_cpos is faked.
+	 */
+	if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
+		len = le64_to_cpu(split_rec->r_cpos) -
+		      le64_to_cpu(orig_rec->r_cpos);
+		orig_rec->r_clusters = cpu_to_le32(len);
+		index++;
+	}
+
+	le16_add_cpu(&rf_list->rl_used, recs_need);
+
+	if (split_rec->r_refcount) {
+		rf_list->rl_recs[index] = *split_rec;
+		trace_ocfs2_split_refcount_rec_insert(
+			(unsigned long long)ref_leaf_bh->b_blocknr, index,
+			(unsigned long long)le64_to_cpu(split_rec->r_cpos),
+			le32_to_cpu(split_rec->r_clusters),
+			le32_to_cpu(split_rec->r_refcount));
+
+		if (merge)
+			ocfs2_refcount_rec_merge(rb, index);
+	}
+
+	ocfs2_journal_dirty(handle, ref_leaf_bh);
+
+out:
+	brelse(new_bh);
+	return ret;
+}
+
+static int __ocfs2_increase_refcount(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_root_bh,
+				     u64 cpos, u32 len, int merge,
+				     struct ocfs2_alloc_context *meta_ac,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret = 0, index;
+	struct buffer_head *ref_leaf_bh = NULL;
+	struct ocfs2_refcount_rec rec;
+	unsigned int set_len = 0;
+
+	trace_ocfs2_increase_refcount_begin(
+	     (unsigned long long)ocfs2_metadata_cache_owner(ci),
+	     (unsigned long long)cpos, len);
+
+	while (len) {
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, len, &rec, &index,
+					     &ref_leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		set_len = le32_to_cpu(rec.r_clusters);
+
+		/*
+		 * Here we may meet with 3 situations:
+		 *
+		 * 1. If we find an already existing record, and the length
+		 *    is the same, cool, we just need to increase the r_refcount
+		 *    and it is OK.
+		 * 2. If we find a hole, just insert it with r_refcount = 1.
+		 * 3. If we are in the middle of one extent record, split
+		 *    it.
+		 */
+		if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
+		    set_len <= len) {
+			trace_ocfs2_increase_refcount_change(
+				(unsigned long long)cpos, set_len,
+				le32_to_cpu(rec.r_refcount));
+			ret = ocfs2_change_refcount_rec(handle, ci,
+							ref_leaf_bh, index,
+							merge, 1);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		} else if (!rec.r_refcount) {
+			rec.r_refcount = cpu_to_le32(1);
+
+			trace_ocfs2_increase_refcount_insert(
+			     (unsigned long long)le64_to_cpu(rec.r_cpos),
+			     set_len);
+			ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
+							ref_leaf_bh,
+							&rec, index,
+							merge, meta_ac);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		} else  {
+			set_len = min((u64)(cpos + len),
+				      le64_to_cpu(rec.r_cpos) + set_len) - cpos;
+			rec.r_cpos = cpu_to_le64(cpos);
+			rec.r_clusters = cpu_to_le32(set_len);
+			le32_add_cpu(&rec.r_refcount, 1);
+
+			trace_ocfs2_increase_refcount_split(
+			     (unsigned long long)le64_to_cpu(rec.r_cpos),
+			     set_len, le32_to_cpu(rec.r_refcount));
+			ret = ocfs2_split_refcount_rec(handle, ci,
+						       ref_root_bh, ref_leaf_bh,
+						       &rec, index, merge,
+						       meta_ac, dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		cpos += set_len;
+		len -= set_len;
+		brelse(ref_leaf_bh);
+		ref_leaf_bh = NULL;
+	}
+
+out:
+	brelse(ref_leaf_bh);
+	return ret;
+}
+
+static int ocfs2_remove_refcount_extent(handle_t *handle,
+				struct ocfs2_caching_info *ci,
+				struct buffer_head *ref_root_bh,
+				struct buffer_head *ref_leaf_bh,
+				struct ocfs2_alloc_context *meta_ac,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_extent_tree et;
+
+	BUG_ON(rb->rf_records.rl_used);
+
+	trace_ocfs2_remove_refcount_extent(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long)ref_leaf_bh->b_blocknr,
+		le32_to_cpu(rb->rf_cpos));
+
+	ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
+	ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
+				  1, meta_ac, dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_remove_from_cache(ci, ref_leaf_bh);
+
+	/*
+	 * add the freed block to the dealloc so that it will be freed
+	 * when we run dealloc.
+	 */
+	ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
+					le16_to_cpu(rb->rf_suballoc_slot),
+					le64_to_cpu(rb->rf_suballoc_loc),
+					le64_to_cpu(rb->rf_blkno),
+					le16_to_cpu(rb->rf_suballoc_bit));
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+	le32_add_cpu(&rb->rf_clusters, -1);
+
+	/*
+	 * check whether we need to restore the root refcount block if
+	 * there is no leaf extent block at atll.
+	 */
+	if (!rb->rf_list.l_next_free_rec) {
+		BUG_ON(rb->rf_clusters);
+
+		trace_ocfs2_restore_refcount_block(
+		     (unsigned long long)ref_root_bh->b_blocknr);
+
+		rb->rf_flags = 0;
+		rb->rf_parent = 0;
+		rb->rf_cpos = 0;
+		memset(&rb->rf_records, 0, sb->s_blocksize -
+		       offsetof(struct ocfs2_refcount_block, rf_records));
+		rb->rf_records.rl_count =
+				cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
+	}
+
+	ocfs2_journal_dirty(handle, ref_root_bh);
+
+out:
+	return ret;
+}
+
+int ocfs2_increase_refcount(handle_t *handle,
+			    struct ocfs2_caching_info *ci,
+			    struct buffer_head *ref_root_bh,
+			    u64 cpos, u32 len,
+			    struct ocfs2_alloc_context *meta_ac,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
+					 cpos, len, 1,
+					 meta_ac, dealloc);
+}
+
+static int ocfs2_decrease_refcount_rec(handle_t *handle,
+				struct ocfs2_caching_info *ci,
+				struct buffer_head *ref_root_bh,
+				struct buffer_head *ref_leaf_bh,
+				int index, u64 cpos, unsigned int len,
+				struct ocfs2_alloc_context *meta_ac,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+	struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
+
+	BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
+	BUG_ON(cpos + len >
+	       le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
+
+	trace_ocfs2_decrease_refcount_rec(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long)cpos, len);
+
+	if (cpos == le64_to_cpu(rec->r_cpos) &&
+	    len == le32_to_cpu(rec->r_clusters))
+		ret = ocfs2_change_refcount_rec(handle, ci,
+						ref_leaf_bh, index, 1, -1);
+	else {
+		struct ocfs2_refcount_rec split = *rec;
+		split.r_cpos = cpu_to_le64(cpos);
+		split.r_clusters = cpu_to_le32(len);
+
+		le32_add_cpu(&split.r_refcount, -1);
+
+		ret = ocfs2_split_refcount_rec(handle, ci,
+					       ref_root_bh, ref_leaf_bh,
+					       &split, index, 1,
+					       meta_ac, dealloc);
+	}
+
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Remove the leaf refcount block if it contains no refcount record. */
+	if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
+		ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
+						   ref_leaf_bh, meta_ac,
+						   dealloc);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out:
+	return ret;
+}
+
+static int __ocfs2_decrease_refcount(handle_t *handle,
+				     struct ocfs2_caching_info *ci,
+				     struct buffer_head *ref_root_bh,
+				     u64 cpos, u32 len,
+				     struct ocfs2_alloc_context *meta_ac,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc,
+				     int delete)
+{
+	int ret = 0, index = 0;
+	struct ocfs2_refcount_rec rec;
+	unsigned int r_count = 0, r_len;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	struct buffer_head *ref_leaf_bh = NULL;
+
+	trace_ocfs2_decrease_refcount(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long)cpos, len, delete);
+
+	while (len) {
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, len, &rec, &index,
+					     &ref_leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		r_count = le32_to_cpu(rec.r_refcount);
+		BUG_ON(r_count == 0);
+		if (!delete)
+			BUG_ON(r_count > 1);
+
+		r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
+			      le32_to_cpu(rec.r_clusters)) - cpos;
+
+		ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
+						  ref_leaf_bh, index,
+						  cpos, r_len,
+						  meta_ac, dealloc);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
+			ret = ocfs2_cache_cluster_dealloc(dealloc,
+					  ocfs2_clusters_to_blocks(sb, cpos),
+							  r_len);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		cpos += r_len;
+		len -= r_len;
+		brelse(ref_leaf_bh);
+		ref_leaf_bh = NULL;
+	}
+
+out:
+	brelse(ref_leaf_bh);
+	return ret;
+}
+
+/* Caller must hold refcount tree lock. */
+int ocfs2_decrease_refcount(struct inode *inode,
+			    handle_t *handle, u32 cpos, u32 len,
+			    struct ocfs2_alloc_context *meta_ac,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc,
+			    int delete)
+{
+	int ret;
+	u64 ref_blkno;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_tree *tree;
+
+	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	ret = ocfs2_get_refcount_block(inode, &ref_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
+					&ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
+					cpos, len, meta_ac, dealloc, delete);
+	if (ret)
+		mlog_errno(ret);
+out:
+	brelse(ref_root_bh);
+	return ret;
+}
+
+/*
+ * Mark the already-existing extent at cpos as refcounted for len clusters.
+ * This adds the refcount extent flag.
+ *
+ * If the existing extent is larger than the request, initiate a
+ * split. An attempt will be made at merging with adjacent extents.
+ *
+ * The caller is responsible for passing down meta_ac if we'll need it.
+ */
+static int ocfs2_mark_extent_refcounted(struct inode *inode,
+				struct ocfs2_extent_tree *et,
+				handle_t *handle, u32 cpos,
+				u32 len, u32 phys,
+				struct ocfs2_alloc_context *meta_ac,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+
+	trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
+					   cpos, len, phys);
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
+		ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
+			    "tree, but the feature bit is not set in the "
+			    "super block.", inode->i_ino);
+		ret = -EROFS;
+		goto out;
+	}
+
+	ret = ocfs2_change_extent_flag(handle, et, cpos,
+				       len, phys, meta_ac, dealloc,
+				       OCFS2_EXT_REFCOUNTED, 0);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+/*
+ * Given some contiguous physical clusters, calculate what we need
+ * for modifying their refcount.
+ */
+static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
+					    struct ocfs2_caching_info *ci,
+					    struct buffer_head *ref_root_bh,
+					    u64 start_cpos,
+					    u32 clusters,
+					    int *meta_add,
+					    int *credits)
+{
+	int ret = 0, index, ref_blocks = 0, recs_add = 0;
+	u64 cpos = start_cpos;
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_refcount_rec rec;
+	struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
+	u32 len;
+
+	while (clusters) {
+		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
+					     cpos, clusters, &rec,
+					     &index, &ref_leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (ref_leaf_bh != prev_bh) {
+			/*
+			 * Now we encounter a new leaf block, so calculate
+			 * whether we need to extend the old leaf.
+			 */
+			if (prev_bh) {
+				rb = (struct ocfs2_refcount_block *)
+							prev_bh->b_data;
+
+				if (le16_to_cpu(rb->rf_records.rl_used) +
+				    recs_add >
+				    le16_to_cpu(rb->rf_records.rl_count))
+					ref_blocks++;
+			}
+
+			recs_add = 0;
+			*credits += 1;
+			brelse(prev_bh);
+			prev_bh = ref_leaf_bh;
+			get_bh(prev_bh);
+		}
+
+		trace_ocfs2_calc_refcount_meta_credits_iterate(
+				recs_add, (unsigned long long)cpos, clusters,
+				(unsigned long long)le64_to_cpu(rec.r_cpos),
+				le32_to_cpu(rec.r_clusters),
+				le32_to_cpu(rec.r_refcount), index);
+
+		len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
+			  le32_to_cpu(rec.r_clusters)) - cpos;
+		/*
+		 * We record all the records which will be inserted to the
+		 * same refcount block, so that we can tell exactly whether
+		 * we need a new refcount block or not.
+		 *
+		 * If we will insert a new one, this is easy and only happens
+		 * during adding refcounted flag to the extent, so we don't
+		 * have a chance of spliting. We just need one record.
+		 *
+		 * If the refcount rec already exists, that would be a little
+		 * complicated. we may have to:
+		 * 1) split at the beginning if the start pos isn't aligned.
+		 *    we need 1 more record in this case.
+		 * 2) split int the end if the end pos isn't aligned.
+		 *    we need 1 more record in this case.
+		 * 3) split in the middle because of file system fragmentation.
+		 *    we need 2 more records in this case(we can't detect this
+		 *    beforehand, so always think of the worst case).
+		 */
+		if (rec.r_refcount) {
+			recs_add += 2;
+			/* Check whether we need a split at the beginning. */
+			if (cpos == start_cpos &&
+			    cpos != le64_to_cpu(rec.r_cpos))
+				recs_add++;
+
+			/* Check whether we need a split in the end. */
+			if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
+			    le32_to_cpu(rec.r_clusters))
+				recs_add++;
+		} else
+			recs_add++;
+
+		brelse(ref_leaf_bh);
+		ref_leaf_bh = NULL;
+		clusters -= len;
+		cpos += len;
+	}
+
+	if (prev_bh) {
+		rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
+
+		if (le16_to_cpu(rb->rf_records.rl_used) + recs_add >
+		    le16_to_cpu(rb->rf_records.rl_count))
+			ref_blocks++;
+
+		*credits += 1;
+	}
+
+	if (!ref_blocks)
+		goto out;
+
+	*meta_add += ref_blocks;
+	*credits += ref_blocks;
+
+	/*
+	 * So we may need ref_blocks to insert into the tree.
+	 * That also means we need to change the b-tree and add that number
+	 * of records since we never merge them.
+	 * We need one more block for expansion since the new created leaf
+	 * block is also full and needs split.
+	 */
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
+		struct ocfs2_extent_tree et;
+
+		ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
+		*meta_add += ocfs2_extend_meta_needed(et.et_root_el);
+		*credits += ocfs2_calc_extend_credits(sb,
+						      et.et_root_el);
+	} else {
+		*credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
+		*meta_add += 1;
+	}
+
+out:
+
+	trace_ocfs2_calc_refcount_meta_credits(
+		(unsigned long long)start_cpos, clusters,
+		*meta_add, *credits);
+	brelse(ref_leaf_bh);
+	brelse(prev_bh);
+	return ret;
+}
+
+/*
+ * For refcount tree, we will decrease some contiguous clusters
+ * refcount count, so just go through it to see how many blocks
+ * we gonna touch and whether we need to create new blocks.
+ *
+ * Normally the refcount blocks store these refcount should be
+ * contiguous also, so that we can get the number easily.
+ * We will at most add split 2 refcount records and 2 more
+ * refcount blocks, so just check it in a rough way.
+ *
+ * Caller must hold refcount tree lock.
+ */
+int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
+					  u64 refcount_loc,
+					  u64 phys_blkno,
+					  u32 clusters,
+					  int *credits,
+					  int *ref_blocks)
+{
+	int ret;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_tree *tree;
+	u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
+		ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
+			    "tree, but the feature bit is not set in the "
+			    "super block.", inode->i_ino);
+		ret = -EROFS;
+		goto out;
+	}
+
+	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
+				      refcount_loc, &tree);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
+					&ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
+					       &tree->rf_ci,
+					       ref_root_bh,
+					       start_cpos, clusters,
+					       ref_blocks, credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
+
+out:
+	brelse(ref_root_bh);
+	return ret;
+}
+
+#define	MAX_CONTIG_BYTES	1048576
+
+static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
+{
+	return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
+}
+
+static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
+{
+	return ~(ocfs2_cow_contig_clusters(sb) - 1);
+}
+
+/*
+ * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
+ * find an offset (start + (n * contig_clusters)) that is closest to cpos
+ * while still being less than or equal to it.
+ *
+ * The goal is to break the extent at a multiple of contig_clusters.
+ */
+static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
+						 unsigned int start,
+						 unsigned int cpos)
+{
+	BUG_ON(start > cpos);
+
+	return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
+}
+
+/*
+ * Given a cluster count of len, pad it out so that it is a multiple
+ * of contig_clusters.
+ */
+static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
+						  unsigned int len)
+{
+	unsigned int padded =
+		(len + (ocfs2_cow_contig_clusters(sb) - 1)) &
+		ocfs2_cow_contig_mask(sb);
+
+	/* Did we wrap? */
+	if (padded < len)
+		padded = UINT_MAX;
+
+	return padded;
+}
+
+/*
+ * Calculate out the start and number of virtual clusters we need to to CoW.
+ *
+ * cpos is vitual start cluster position we want to do CoW in a
+ * file and write_len is the cluster length.
+ * max_cpos is the place where we want to stop CoW intentionally.
+ *
+ * Normal we will start CoW from the beginning of extent record cotaining cpos.
+ * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
+ * get good I/O from the resulting extent tree.
+ */
+static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
+					   struct ocfs2_extent_list *el,
+					   u32 cpos,
+					   u32 write_len,
+					   u32 max_cpos,
+					   u32 *cow_start,
+					   u32 *cow_len)
+{
+	int ret = 0;
+	int tree_height = le16_to_cpu(el->l_tree_depth), i;
+	struct buffer_head *eb_bh = NULL;
+	struct ocfs2_extent_block *eb = NULL;
+	struct ocfs2_extent_rec *rec;
+	unsigned int want_clusters, rec_end = 0;
+	int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
+	int leaf_clusters;
+
+	BUG_ON(cpos + write_len > max_cpos);
+
+	if (tree_height > 0) {
+		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+
+		if (el->l_tree_depth) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %lu has non zero tree depth in "
+				    "leaf block %llu\n", inode->i_ino,
+				    (unsigned long long)eb_bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
+		}
+	}
+
+	*cow_len = 0;
+	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		rec = &el->l_recs[i];
+
+		if (ocfs2_is_empty_extent(rec)) {
+			mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
+					"index %d\n", inode->i_ino, i);
+			continue;
+		}
+
+		if (le32_to_cpu(rec->e_cpos) +
+		    le16_to_cpu(rec->e_leaf_clusters) <= cpos)
+			continue;
+
+		if (*cow_len == 0) {
+			/*
+			 * We should find a refcounted record in the
+			 * first pass.
+			 */
+			BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
+			*cow_start = le32_to_cpu(rec->e_cpos);
+		}
+
+		/*
+		 * If we encounter a hole, a non-refcounted record or
+		 * pass the max_cpos, stop the search.
+		 */
+		if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
+		    (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
+		    (max_cpos <= le32_to_cpu(rec->e_cpos)))
+			break;
+
+		leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
+		rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
+		if (rec_end > max_cpos) {
+			rec_end = max_cpos;
+			leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
+		}
+
+		/*
+		 * How many clusters do we actually need from
+		 * this extent?  First we see how many we actually
+		 * need to complete the write.  If that's smaller
+		 * than contig_clusters, we try for contig_clusters.
+		 */
+		if (!*cow_len)
+			want_clusters = write_len;
+		else
+			want_clusters = (cpos + write_len) -
+				(*cow_start + *cow_len);
+		if (want_clusters < contig_clusters)
+			want_clusters = contig_clusters;
+
+		/*
+		 * If the write does not cover the whole extent, we
+		 * need to calculate how we're going to split the extent.
+		 * We try to do it on contig_clusters boundaries.
+		 *
+		 * Any extent smaller than contig_clusters will be
+		 * CoWed in its entirety.
+		 */
+		if (leaf_clusters <= contig_clusters)
+			*cow_len += leaf_clusters;
+		else if (*cow_len || (*cow_start == cpos)) {
+			/*
+			 * This extent needs to be CoW'd from its
+			 * beginning, so all we have to do is compute
+			 * how many clusters to grab.  We align
+			 * want_clusters to the edge of contig_clusters
+			 * to get better I/O.
+			 */
+			want_clusters = ocfs2_cow_align_length(inode->i_sb,
+							       want_clusters);
+
+			if (leaf_clusters < want_clusters)
+				*cow_len += leaf_clusters;
+			else
+				*cow_len += want_clusters;
+		} else if ((*cow_start + contig_clusters) >=
+			   (cpos + write_len)) {
+			/*
+			 * Breaking off contig_clusters at the front
+			 * of the extent will cover our write.  That's
+			 * easy.
+			 */
+			*cow_len = contig_clusters;
+		} else if ((rec_end - cpos) <= contig_clusters) {
+			/*
+			 * Breaking off contig_clusters at the tail of
+			 * this extent will cover cpos.
+			 */
+			*cow_start = rec_end - contig_clusters;
+			*cow_len = contig_clusters;
+		} else if ((rec_end - cpos) <= want_clusters) {
+			/*
+			 * While we can't fit the entire write in this
+			 * extent, we know that the write goes from cpos
+			 * to the end of the extent.  Break that off.
+			 * We try to break it at some multiple of
+			 * contig_clusters from the front of the extent.
+			 * Failing that (ie, cpos is within
+			 * contig_clusters of the front), we'll CoW the
+			 * entire extent.
+			 */
+			*cow_start = ocfs2_cow_align_start(inode->i_sb,
+							   *cow_start, cpos);
+			*cow_len = rec_end - *cow_start;
+		} else {
+			/*
+			 * Ok, the entire write lives in the middle of
+			 * this extent.  Let's try to slice the extent up
+			 * nicely.  Optimally, our CoW region starts at
+			 * m*contig_clusters from the beginning of the
+			 * extent and goes for n*contig_clusters,
+			 * covering the entire write.
+			 */
+			*cow_start = ocfs2_cow_align_start(inode->i_sb,
+							   *cow_start, cpos);
+
+			want_clusters = (cpos + write_len) - *cow_start;
+			want_clusters = ocfs2_cow_align_length(inode->i_sb,
+							       want_clusters);
+			if (*cow_start + want_clusters <= rec_end)
+				*cow_len = want_clusters;
+			else
+				*cow_len = rec_end - *cow_start;
+		}
+
+		/* Have we covered our entire write yet? */
+		if ((*cow_start + *cow_len) >= (cpos + write_len))
+			break;
+
+		/*
+		 * If we reach the end of the extent block and don't get enough
+		 * clusters, continue with the next extent block if possible.
+		 */
+		if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
+		    eb && eb->h_next_leaf_blk) {
+			brelse(eb_bh);
+			eb_bh = NULL;
+
+			ret = ocfs2_read_extent_block(INODE_CACHE(inode),
+					       le64_to_cpu(eb->h_next_leaf_blk),
+					       &eb_bh);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+			el = &eb->h_list;
+			i = -1;
+		}
+	}
+
+out:
+	brelse(eb_bh);
+	return ret;
+}
+
+/*
+ * Prepare meta_ac, data_ac and calculate credits when we want to add some
+ * num_clusters in data_tree "et" and change the refcount for the old
+ * clusters(starting form p_cluster) in the refcount tree.
+ *
+ * Note:
+ * 1. since we may split the old tree, so we at most will need num_clusters + 2
+ *    more new leaf records.
+ * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
+ *    just give data_ac = NULL.
+ */
+static int ocfs2_lock_refcount_allocators(struct super_block *sb,
+					u32 p_cluster, u32 num_clusters,
+					struct ocfs2_extent_tree *et,
+					struct ocfs2_caching_info *ref_ci,
+					struct buffer_head *ref_root_bh,
+					struct ocfs2_alloc_context **meta_ac,
+					struct ocfs2_alloc_context **data_ac,
+					int *credits)
+{
+	int ret = 0, meta_add = 0;
+	int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
+
+	if (num_free_extents < 0) {
+		ret = num_free_extents;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (num_free_extents < num_clusters + 2)
+		meta_add =
+			ocfs2_extend_meta_needed(et->et_root_el);
+
+	*credits += ocfs2_calc_extend_credits(sb, et->et_root_el);
+
+	ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
+					       p_cluster, num_clusters,
+					       &meta_add, credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
+	ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
+						meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (data_ac) {
+		ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
+					     data_ac);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out:
+	if (ret) {
+		if (*meta_ac) {
+			ocfs2_free_alloc_context(*meta_ac);
+			*meta_ac = NULL;
+		}
+	}
+
+	return ret;
+}
+
+static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
+{
+	BUG_ON(buffer_dirty(bh));
+
+	clear_buffer_mapped(bh);
+
+	return 0;
+}
+
+int ocfs2_duplicate_clusters_by_page(handle_t *handle,
+				     struct inode *inode,
+				     u32 cpos, u32 old_cluster,
+				     u32 new_cluster, u32 new_len)
+{
+	int ret = 0, partial;
+	struct super_block *sb = inode->i_sb;
+	u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
+	struct page *page;
+	pgoff_t page_index;
+	unsigned int from, to, readahead_pages;
+	loff_t offset, end, map_end;
+	struct address_space *mapping = inode->i_mapping;
+
+	trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
+					       new_cluster, new_len);
+
+	readahead_pages =
+		(ocfs2_cow_contig_clusters(sb) <<
+		 OCFS2_SB(sb)->s_clustersize_bits) >> PAGE_CACHE_SHIFT;
+	offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
+	end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
+	/*
+	 * We only duplicate pages until we reach the page contains i_size - 1.
+	 * So trim 'end' to i_size.
+	 */
+	if (end > i_size_read(inode))
+		end = i_size_read(inode);
+
+	while (offset < end) {
+		page_index = offset >> PAGE_CACHE_SHIFT;
+		map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
+		if (map_end > end)
+			map_end = end;
+
+		/* from, to is the offset within the page. */
+		from = offset & (PAGE_CACHE_SIZE - 1);
+		to = PAGE_CACHE_SIZE;
+		if (map_end & (PAGE_CACHE_SIZE - 1))
+			to = map_end & (PAGE_CACHE_SIZE - 1);
+
+		page = find_or_create_page(mapping, page_index, GFP_NOFS);
+		if (!page) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			break;
+		}
+
+		/*
+		 * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
+		 * can't be dirtied before we CoW it out.
+		 */
+		if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
+			BUG_ON(PageDirty(page));
+
+		if (!PageUptodate(page)) {
+			ret = block_read_full_page(page, ocfs2_get_block);
+			if (ret) {
+				mlog_errno(ret);
+				goto unlock;
+			}
+			lock_page(page);
+		}
+
+		if (page_has_buffers(page)) {
+			ret = walk_page_buffers(handle, page_buffers(page),
+						from, to, &partial,
+						ocfs2_clear_cow_buffer);
+			if (ret) {
+				mlog_errno(ret);
+				goto unlock;
+			}
+		}
+
+		ocfs2_map_and_dirty_page(inode,
+					 handle, from, to,
+					 page, 0, &new_block);
+		mark_page_accessed(page);
+unlock:
+		unlock_page(page);
+		page_cache_release(page);
+		page = NULL;
+		offset = map_end;
+		if (ret)
+			break;
+	}
+
+	return ret;
+}
+
+int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
+				    struct inode *inode,
+				    u32 cpos, u32 old_cluster,
+				    u32 new_cluster, u32 new_len)
+{
+	int ret = 0;
+	struct super_block *sb = inode->i_sb;
+	struct ocfs2_caching_info *ci = INODE_CACHE(inode);
+	int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
+	u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
+	u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	struct buffer_head *old_bh = NULL;
+	struct buffer_head *new_bh = NULL;
+
+	trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
+					       new_cluster, new_len);
+
+	for (i = 0; i < blocks; i++, old_block++, new_block++) {
+		new_bh = sb_getblk(osb->sb, new_block);
+		if (new_bh == NULL) {
+			ret = -ENOMEM;
+			mlog_errno(ret);
+			break;
+		}
+
+		ocfs2_set_new_buffer_uptodate(ci, new_bh);
+
+		ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = ocfs2_journal_access(handle, ci, new_bh,
+					   OCFS2_JOURNAL_ACCESS_CREATE);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
+		ocfs2_journal_dirty(handle, new_bh);
+
+		brelse(new_bh);
+		brelse(old_bh);
+		new_bh = NULL;
+		old_bh = NULL;
+	}
+
+	brelse(new_bh);
+	brelse(old_bh);
+	return ret;
+}
+
+static int ocfs2_clear_ext_refcount(handle_t *handle,
+				    struct ocfs2_extent_tree *et,
+				    u32 cpos, u32 p_cluster, u32 len,
+				    unsigned int ext_flags,
+				    struct ocfs2_alloc_context *meta_ac,
+				    struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret, index;
+	struct ocfs2_extent_rec replace_rec;
+	struct ocfs2_path *path = NULL;
+	struct ocfs2_extent_list *el;
+	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
+	u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
+
+	trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
+				       cpos, len, p_cluster, ext_flags);
+
+	memset(&replace_rec, 0, sizeof(replace_rec));
+	replace_rec.e_cpos = cpu_to_le32(cpos);
+	replace_rec.e_leaf_clusters = cpu_to_le16(len);
+	replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
+								   p_cluster));
+	replace_rec.e_flags = ext_flags;
+	replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
+
+	path = ocfs2_new_path_from_et(et);
+	if (!path) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_find_path(et->et_ci, path, cpos);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	el = path_leaf_el(path);
+
+	index = ocfs2_search_extent_list(el, cpos);
+	if (index == -1) {
+		ocfs2_error(sb,
+			    "Inode %llu has an extent at cpos %u which can no "
+			    "longer be found.\n",
+			    (unsigned long long)ino, cpos);
+		ret = -EROFS;
+		goto out;
+	}
+
+	ret = ocfs2_split_extent(handle, et, path, index,
+				 &replace_rec, meta_ac, dealloc);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	ocfs2_free_path(path);
+	return ret;
+}
+
+static int ocfs2_replace_clusters(handle_t *handle,
+				  struct ocfs2_cow_context *context,
+				  u32 cpos, u32 old,
+				  u32 new, u32 len,
+				  unsigned int ext_flags)
+{
+	int ret;
+	struct ocfs2_caching_info *ci = context->data_et.et_ci;
+	u64 ino = ocfs2_metadata_cache_owner(ci);
+
+	trace_ocfs2_replace_clusters((unsigned long long)ino,
+				     cpos, old, new, len, ext_flags);
+
+	/*If the old clusters is unwritten, no need to duplicate. */
+	if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
+		ret = context->cow_duplicate_clusters(handle, context->inode,
+						      cpos, old, new, len);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
+				       cpos, new, len, ext_flags,
+				       context->meta_ac, &context->dealloc);
+	if (ret)
+		mlog_errno(ret);
+out:
+	return ret;
+}
+
+int ocfs2_cow_sync_writeback(struct super_block *sb,
+			     struct inode *inode,
+			     u32 cpos, u32 num_clusters)
+{
+	int ret = 0;
+	loff_t offset, end, map_end;
+	pgoff_t page_index;
+	struct page *page;
+
+	if (ocfs2_should_order_data(inode))
+		return 0;
+
+	offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
+	end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
+
+	ret = filemap_fdatawrite_range(inode->i_mapping,
+				       offset, end - 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	while (offset < end) {
+		page_index = offset >> PAGE_CACHE_SHIFT;
+		map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
+		if (map_end > end)
+			map_end = end;
+
+		page = find_or_create_page(inode->i_mapping,
+					   page_index, GFP_NOFS);
+		BUG_ON(!page);
+
+		wait_on_page_writeback(page);
+		if (PageError(page)) {
+			ret = -EIO;
+			mlog_errno(ret);
+		} else
+			mark_page_accessed(page);
+
+		unlock_page(page);
+		page_cache_release(page);
+		page = NULL;
+		offset = map_end;
+		if (ret)
+			break;
+	}
+
+	return ret;
+}
+
+static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
+				 u32 v_cluster, u32 *p_cluster,
+				 u32 *num_clusters,
+				 unsigned int *extent_flags)
+{
+	return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
+				  num_clusters, extent_flags);
+}
+
+static int ocfs2_make_clusters_writable(struct super_block *sb,
+					struct ocfs2_cow_context *context,
+					u32 cpos, u32 p_cluster,
+					u32 num_clusters, unsigned int e_flags)
+{
+	int ret, delete, index, credits =  0;
+	u32 new_bit, new_len, orig_num_clusters;
+	unsigned int set_len;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	handle_t *handle;
+	struct buffer_head *ref_leaf_bh = NULL;
+	struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
+	struct ocfs2_refcount_rec rec;
+
+	trace_ocfs2_make_clusters_writable(cpos, p_cluster,
+					   num_clusters, e_flags);
+
+	ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
+					     &context->data_et,
+					     ref_ci,
+					     context->ref_root_bh,
+					     &context->meta_ac,
+					     &context->data_ac, &credits);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	if (context->post_refcount)
+		credits += context->post_refcount->credits;
+
+	credits += context->extra_credits;
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	orig_num_clusters = num_clusters;
+
+	while (num_clusters) {
+		ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
+					     p_cluster, num_clusters,
+					     &rec, &index, &ref_leaf_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+
+		BUG_ON(!rec.r_refcount);
+		set_len = min((u64)p_cluster + num_clusters,
+			      le64_to_cpu(rec.r_cpos) +
+			      le32_to_cpu(rec.r_clusters)) - p_cluster;
+
+		/*
+		 * There are many different situation here.
+		 * 1. If refcount == 1, remove the flag and don't COW.
+		 * 2. If refcount > 1, allocate clusters.
+		 *    Here we may not allocate r_len once at a time, so continue
+		 *    until we reach num_clusters.
+		 */
+		if (le32_to_cpu(rec.r_refcount) == 1) {
+			delete = 0;
+			ret = ocfs2_clear_ext_refcount(handle,
+						       &context->data_et,
+						       cpos, p_cluster,
+						       set_len, e_flags,
+						       context->meta_ac,
+						       &context->dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out_commit;
+			}
+		} else {
+			delete = 1;
+
+			ret = __ocfs2_claim_clusters(handle,
+						     context->data_ac,
+						     1, set_len,
+						     &new_bit, &new_len);
+			if (ret) {
+				mlog_errno(ret);
+				goto out_commit;
+			}
+
+			ret = ocfs2_replace_clusters(handle, context,
+						     cpos, p_cluster, new_bit,
+						     new_len, e_flags);
+			if (ret) {
+				mlog_errno(ret);
+				goto out_commit;
+			}
+			set_len = new_len;
+		}
+
+		ret = __ocfs2_decrease_refcount(handle, ref_ci,
+						context->ref_root_bh,
+						p_cluster, set_len,
+						context->meta_ac,
+						&context->dealloc, delete);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+
+		cpos += set_len;
+		p_cluster += set_len;
+		num_clusters -= set_len;
+		brelse(ref_leaf_bh);
+		ref_leaf_bh = NULL;
+	}
+
+	/* handle any post_cow action. */
+	if (context->post_refcount && context->post_refcount->func) {
+		ret = context->post_refcount->func(context->inode, handle,
+						context->post_refcount->para);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+	}
+
+	/*
+	 * Here we should write the new page out first if we are
+	 * in write-back mode.
+	 */
+	if (context->get_clusters == ocfs2_di_get_clusters) {
+		ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
+					       orig_num_clusters);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	if (context->data_ac) {
+		ocfs2_free_alloc_context(context->data_ac);
+		context->data_ac = NULL;
+	}
+	if (context->meta_ac) {
+		ocfs2_free_alloc_context(context->meta_ac);
+		context->meta_ac = NULL;
+	}
+	brelse(ref_leaf_bh);
+
+	return ret;
+}
+
+static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
+{
+	int ret = 0;
+	struct inode *inode = context->inode;
+	u32 cow_start = context->cow_start, cow_len = context->cow_len;
+	u32 p_cluster, num_clusters;
+	unsigned int ext_flags;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
+		ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
+			    "tree, but the feature bit is not set in the "
+			    "super block.", inode->i_ino);
+		return -EROFS;
+	}
+
+	ocfs2_init_dealloc_ctxt(&context->dealloc);
+
+	while (cow_len) {
+		ret = context->get_clusters(context, cow_start, &p_cluster,
+					    &num_clusters, &ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
+
+		if (cow_len < num_clusters)
+			num_clusters = cow_len;
+
+		ret = ocfs2_make_clusters_writable(inode->i_sb, context,
+						   cow_start, p_cluster,
+						   num_clusters, ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		cow_len -= num_clusters;
+		cow_start += num_clusters;
+	}
+
+	if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
+		ocfs2_schedule_truncate_log_flush(osb, 1);
+		ocfs2_run_deallocs(osb, &context->dealloc);
+	}
+
+	return ret;
+}
+
+/*
+ * Starting at cpos, try to CoW write_len clusters.  Don't CoW
+ * past max_cpos.  This will stop when it runs into a hole or an
+ * unrefcounted extent.
+ */
+static int ocfs2_refcount_cow_hunk(struct inode *inode,
+				   struct buffer_head *di_bh,
+				   u32 cpos, u32 write_len, u32 max_cpos)
+{
+	int ret;
+	u32 cow_start = 0, cow_len = 0;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_refcount_tree *ref_tree;
+	struct ocfs2_cow_context *context = NULL;
+
+	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
+					      cpos, write_len, max_cpos,
+					      &cow_start, &cow_len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_refcount_cow_hunk(OCFS2_I(inode)->ip_blkno,
+				      cpos, write_len, max_cpos,
+				      cow_start, cow_len);
+
+	BUG_ON(cow_len == 0);
+
+	context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
+	if (!context) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
+				       1, &ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	context->inode = inode;
+	context->cow_start = cow_start;
+	context->cow_len = cow_len;
+	context->ref_tree = ref_tree;
+	context->ref_root_bh = ref_root_bh;
+	context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
+	context->get_clusters = ocfs2_di_get_clusters;
+
+	ocfs2_init_dinode_extent_tree(&context->data_et,
+				      INODE_CACHE(inode), di_bh);
+
+	ret = ocfs2_replace_cow(context);
+	if (ret)
+		mlog_errno(ret);
+
+	/*
+	 * truncate the extent map here since no matter whether we meet with
+	 * any error during the action, we shouldn't trust cached extent map
+	 * any more.
+	 */
+	ocfs2_extent_map_trunc(inode, cow_start);
+
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	brelse(ref_root_bh);
+out:
+	kfree(context);
+	return ret;
+}
+
+/*
+ * CoW any and all clusters between cpos and cpos+write_len.
+ * Don't CoW past max_cpos.  If this returns successfully, all
+ * clusters between cpos and cpos+write_len are safe to modify.
+ */
+int ocfs2_refcount_cow(struct inode *inode,
+		       struct buffer_head *di_bh,
+		       u32 cpos, u32 write_len, u32 max_cpos)
+{
+	int ret = 0;
+	u32 p_cluster, num_clusters;
+	unsigned int ext_flags;
+
+	while (write_len) {
+		ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
+					 &num_clusters, &ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		if (write_len < num_clusters)
+			num_clusters = write_len;
+
+		if (ext_flags & OCFS2_EXT_REFCOUNTED) {
+			ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
+						      num_clusters, max_cpos);
+			if (ret) {
+				mlog_errno(ret);
+				break;
+			}
+		}
+
+		write_len -= num_clusters;
+		cpos += num_clusters;
+	}
+
+	return ret;
+}
+
+static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
+					  u32 v_cluster, u32 *p_cluster,
+					  u32 *num_clusters,
+					  unsigned int *extent_flags)
+{
+	struct inode *inode = context->inode;
+	struct ocfs2_xattr_value_root *xv = context->cow_object;
+
+	return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
+					num_clusters, &xv->xr_list,
+					extent_flags);
+}
+
+/*
+ * Given a xattr value root, calculate the most meta/credits we need for
+ * refcount tree change if we truncate it to 0.
+ */
+int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
+				       struct ocfs2_caching_info *ref_ci,
+				       struct buffer_head *ref_root_bh,
+				       struct ocfs2_xattr_value_root *xv,
+				       int *meta_add, int *credits)
+{
+	int ret = 0, index, ref_blocks = 0;
+	u32 p_cluster, num_clusters;
+	u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_refcount_rec rec;
+	struct buffer_head *ref_leaf_bh = NULL;
+
+	while (cpos < clusters) {
+		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
+					       &num_clusters, &xv->xr_list,
+					       NULL);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		cpos += num_clusters;
+
+		while (num_clusters) {
+			ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
+						     p_cluster, num_clusters,
+						     &rec, &index,
+						     &ref_leaf_bh);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			BUG_ON(!rec.r_refcount);
+
+			rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
+
+			/*
+			 * We really don't know whether the other clusters is in
+			 * this refcount block or not, so just take the worst
+			 * case that all the clusters are in this block and each
+			 * one will split a refcount rec, so totally we need
+			 * clusters * 2 new refcount rec.
+			 */
+			if (le16_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
+			    le16_to_cpu(rb->rf_records.rl_count))
+				ref_blocks++;
+
+			*credits += 1;
+			brelse(ref_leaf_bh);
+			ref_leaf_bh = NULL;
+
+			if (num_clusters <= le32_to_cpu(rec.r_clusters))
+				break;
+			else
+				num_clusters -= le32_to_cpu(rec.r_clusters);
+			p_cluster += num_clusters;
+		}
+	}
+
+	*meta_add += ref_blocks;
+	if (!ref_blocks)
+		goto out;
+
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
+		*credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
+	else {
+		struct ocfs2_extent_tree et;
+
+		ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
+		*credits += ocfs2_calc_extend_credits(inode->i_sb,
+						      et.et_root_el);
+	}
+
+out:
+	brelse(ref_leaf_bh);
+	return ret;
+}
+
+/*
+ * Do CoW for xattr.
+ */
+int ocfs2_refcount_cow_xattr(struct inode *inode,
+			     struct ocfs2_dinode *di,
+			     struct ocfs2_xattr_value_buf *vb,
+			     struct ocfs2_refcount_tree *ref_tree,
+			     struct buffer_head *ref_root_bh,
+			     u32 cpos, u32 write_len,
+			     struct ocfs2_post_refcount *post)
+{
+	int ret;
+	struct ocfs2_xattr_value_root *xv = vb->vb_xv;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_cow_context *context = NULL;
+	u32 cow_start, cow_len;
+
+	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
+
+	ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
+					      cpos, write_len, UINT_MAX,
+					      &cow_start, &cow_len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	BUG_ON(cow_len == 0);
+
+	context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
+	if (!context) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	context->inode = inode;
+	context->cow_start = cow_start;
+	context->cow_len = cow_len;
+	context->ref_tree = ref_tree;
+	context->ref_root_bh = ref_root_bh;
+	context->cow_object = xv;
+
+	context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
+	/* We need the extra credits for duplicate_clusters by jbd. */
+	context->extra_credits =
+		ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
+	context->get_clusters = ocfs2_xattr_value_get_clusters;
+	context->post_refcount = post;
+
+	ocfs2_init_xattr_value_extent_tree(&context->data_et,
+					   INODE_CACHE(inode), vb);
+
+	ret = ocfs2_replace_cow(context);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	kfree(context);
+	return ret;
+}
+
+/*
+ * Insert a new extent into refcount tree and mark a extent rec
+ * as refcounted in the dinode tree.
+ */
+int ocfs2_add_refcount_flag(struct inode *inode,
+			    struct ocfs2_extent_tree *data_et,
+			    struct ocfs2_caching_info *ref_ci,
+			    struct buffer_head *ref_root_bh,
+			    u32 cpos, u32 p_cluster, u32 num_clusters,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc,
+			    struct ocfs2_post_refcount *post)
+{
+	int ret;
+	handle_t *handle;
+	int credits = 1, ref_blocks = 0;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_alloc_context *meta_ac = NULL;
+
+	ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
+					       ref_ci, ref_root_bh,
+					       p_cluster, num_clusters,
+					       &ref_blocks, &credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_add_refcount_flag(ref_blocks, credits);
+
+	if (ref_blocks) {
+		ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
+							ref_blocks, &meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (post)
+		credits += post->credits;
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
+					   cpos, num_clusters, p_cluster,
+					   meta_ac, dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
+					p_cluster, num_clusters, 0,
+					meta_ac, dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	if (post && post->func) {
+		ret = post->func(inode, handle, post->para);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+	return ret;
+}
+
+static int ocfs2_change_ctime(struct inode *inode,
+			      struct buffer_head *di_bh)
+{
+	int ret;
+	handle_t *handle;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
+				   OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	inode->i_ctime = CURRENT_TIME;
+	di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
+	di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+
+	ocfs2_journal_dirty(handle, di_bh);
+
+out_commit:
+	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
+out:
+	return ret;
+}
+
+static int ocfs2_attach_refcount_tree(struct inode *inode,
+				      struct buffer_head *di_bh)
+{
+	int ret, data_changed = 0;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_refcount_tree *ref_tree;
+	unsigned int ext_flags;
+	loff_t size;
+	u32 cpos, num_clusters, clusters, p_cluster;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct ocfs2_extent_tree di_et;
+
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
+		ret = ocfs2_create_refcount_tree(inode, di_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	BUG_ON(!di->i_refcount_loc);
+	ret = ocfs2_lock_refcount_tree(osb,
+				       le64_to_cpu(di->i_refcount_loc), 1,
+				       &ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
+		goto attach_xattr;
+
+	ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
+
+	size = i_size_read(inode);
+	clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
+
+	cpos = 0;
+	while (cpos < clusters) {
+		ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
+					 &num_clusters, &ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			goto unlock;
+		}
+		if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
+			ret = ocfs2_add_refcount_flag(inode, &di_et,
+						      &ref_tree->rf_ci,
+						      ref_root_bh, cpos,
+						      p_cluster, num_clusters,
+						      &dealloc, NULL);
+			if (ret) {
+				mlog_errno(ret);
+				goto unlock;
+			}
+
+			data_changed = 1;
+		}
+		cpos += num_clusters;
+	}
+
+attach_xattr:
+	if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
+		ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
+						       &ref_tree->rf_ci,
+						       ref_root_bh,
+						       &dealloc);
+		if (ret) {
+			mlog_errno(ret);
+			goto unlock;
+		}
+	}
+
+	if (data_changed) {
+		ret = ocfs2_change_ctime(inode, di_bh);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+unlock:
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	brelse(ref_root_bh);
+
+	if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
+		ocfs2_schedule_truncate_log_flush(osb, 1);
+		ocfs2_run_deallocs(osb, &dealloc);
+	}
+out:
+	/*
+	 * Empty the extent map so that we may get the right extent
+	 * record from the disk.
+	 */
+	ocfs2_extent_map_trunc(inode, 0);
+
+	return ret;
+}
+
+static int ocfs2_add_refcounted_extent(struct inode *inode,
+				   struct ocfs2_extent_tree *et,
+				   struct ocfs2_caching_info *ref_ci,
+				   struct buffer_head *ref_root_bh,
+				   u32 cpos, u32 p_cluster, u32 num_clusters,
+				   unsigned int ext_flags,
+				   struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret;
+	handle_t *handle;
+	int credits = 0;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_alloc_context *meta_ac = NULL;
+
+	ret = ocfs2_lock_refcount_allocators(inode->i_sb,
+					     p_cluster, num_clusters,
+					     et, ref_ci,
+					     ref_root_bh, &meta_ac,
+					     NULL, &credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_insert_extent(handle, et, cpos,
+			ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
+			num_clusters, ext_flags, meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
+				      p_cluster, num_clusters,
+				      meta_ac, dealloc);
+	if (ret)
+		mlog_errno(ret);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+	return ret;
+}
+
+static int ocfs2_duplicate_inline_data(struct inode *s_inode,
+				       struct buffer_head *s_bh,
+				       struct inode *t_inode,
+				       struct buffer_head *t_bh)
+{
+	int ret;
+	handle_t *handle;
+	struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
+	struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
+	struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
+
+	BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
+
+	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
+	memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
+	       le16_to_cpu(s_di->id2.i_data.id_count));
+	spin_lock(&OCFS2_I(t_inode)->ip_lock);
+	OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
+	t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
+	spin_unlock(&OCFS2_I(t_inode)->ip_lock);
+
+	ocfs2_journal_dirty(handle, t_bh);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	return ret;
+}
+
+static int ocfs2_duplicate_extent_list(struct inode *s_inode,
+				struct inode *t_inode,
+				struct buffer_head *t_bh,
+				struct ocfs2_caching_info *ref_ci,
+				struct buffer_head *ref_root_bh,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret = 0;
+	u32 p_cluster, num_clusters, clusters, cpos;
+	loff_t size;
+	unsigned int ext_flags;
+	struct ocfs2_extent_tree et;
+
+	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
+
+	size = i_size_read(s_inode);
+	clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
+
+	cpos = 0;
+	while (cpos < clusters) {
+		ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
+					 &num_clusters, &ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		if (p_cluster) {
+			ret = ocfs2_add_refcounted_extent(t_inode, &et,
+							  ref_ci, ref_root_bh,
+							  cpos, p_cluster,
+							  num_clusters,
+							  ext_flags,
+							  dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+		}
+
+		cpos += num_clusters;
+	}
+
+out:
+	return ret;
+}
+
+/*
+ * change the new file's attributes to the src.
+ *
+ * reflink creates a snapshot of a file, that means the attributes
+ * must be identical except for three exceptions - nlink, ino, and ctime.
+ */
+static int ocfs2_complete_reflink(struct inode *s_inode,
+				  struct buffer_head *s_bh,
+				  struct inode *t_inode,
+				  struct buffer_head *t_bh,
+				  bool preserve)
+{
+	int ret;
+	handle_t *handle;
+	struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
+	loff_t size = i_size_read(s_inode);
+
+	handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
+				   OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		return ret;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	spin_lock(&OCFS2_I(t_inode)->ip_lock);
+	OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
+	OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
+	OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
+	spin_unlock(&OCFS2_I(t_inode)->ip_lock);
+	i_size_write(t_inode, size);
+	t_inode->i_blocks = s_inode->i_blocks;
+
+	di->i_xattr_inline_size = s_di->i_xattr_inline_size;
+	di->i_clusters = s_di->i_clusters;
+	di->i_size = s_di->i_size;
+	di->i_dyn_features = s_di->i_dyn_features;
+	di->i_attr = s_di->i_attr;
+
+	if (preserve) {
+		t_inode->i_uid = s_inode->i_uid;
+		t_inode->i_gid = s_inode->i_gid;
+		t_inode->i_mode = s_inode->i_mode;
+		di->i_uid = s_di->i_uid;
+		di->i_gid = s_di->i_gid;
+		di->i_mode = s_di->i_mode;
+
+		/*
+		 * update time.
+		 * we want mtime to appear identical to the source and
+		 * update ctime.
+		 */
+		t_inode->i_ctime = CURRENT_TIME;
+
+		di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
+		di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
+
+		t_inode->i_mtime = s_inode->i_mtime;
+		di->i_mtime = s_di->i_mtime;
+		di->i_mtime_nsec = s_di->i_mtime_nsec;
+	}
+
+	ocfs2_journal_dirty(handle, t_bh);
+
+out_commit:
+	ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
+	return ret;
+}
+
+static int ocfs2_create_reflink_node(struct inode *s_inode,
+				     struct buffer_head *s_bh,
+				     struct inode *t_inode,
+				     struct buffer_head *t_bh,
+				     bool preserve)
+{
+	int ret;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
+	struct ocfs2_refcount_block *rb;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
+	struct ocfs2_refcount_tree *ref_tree;
+
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	ret = ocfs2_set_refcount_tree(t_inode, t_bh,
+				      le64_to_cpu(di->i_refcount_loc));
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
+						  t_inode, t_bh);
+		if (ret)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
+				       1, &ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+	ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
+					  &ref_tree->rf_ci, ref_root_bh,
+					  &dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock_refcount;
+	}
+
+out_unlock_refcount:
+	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	brelse(ref_root_bh);
+out:
+	if (ocfs2_dealloc_has_cluster(&dealloc)) {
+		ocfs2_schedule_truncate_log_flush(osb, 1);
+		ocfs2_run_deallocs(osb, &dealloc);
+	}
+
+	return ret;
+}
+
+static int __ocfs2_reflink(struct dentry *old_dentry,
+			   struct buffer_head *old_bh,
+			   struct inode *new_inode,
+			   bool preserve)
+{
+	int ret;
+	struct inode *inode = d_inode(old_dentry);
+	struct buffer_head *new_bh = NULL;
+
+	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
+		ret = -EINVAL;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = filemap_fdatawrite(inode->i_mapping);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_attach_refcount_tree(inode, old_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mutex_lock_nested(&new_inode->i_mutex, I_MUTEX_CHILD);
+	ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
+				      OI_LS_REFLINK_TARGET);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+
+	ret = ocfs2_create_reflink_node(inode, old_bh,
+					new_inode, new_bh, preserve);
+	if (ret) {
+		mlog_errno(ret);
+		goto inode_unlock;
+	}
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
+		ret = ocfs2_reflink_xattrs(inode, old_bh,
+					   new_inode, new_bh,
+					   preserve);
+		if (ret) {
+			mlog_errno(ret);
+			goto inode_unlock;
+		}
+	}
+
+	ret = ocfs2_complete_reflink(inode, old_bh,
+				     new_inode, new_bh, preserve);
+	if (ret)
+		mlog_errno(ret);
+
+inode_unlock:
+	ocfs2_inode_unlock(new_inode, 1);
+	brelse(new_bh);
+out_unlock:
+	mutex_unlock(&new_inode->i_mutex);
+out:
+	if (!ret) {
+		ret = filemap_fdatawait(inode->i_mapping);
+		if (ret)
+			mlog_errno(ret);
+	}
+	return ret;
+}
+
+static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
+			 struct dentry *new_dentry, bool preserve)
+{
+	int error;
+	struct inode *inode = d_inode(old_dentry);
+	struct buffer_head *old_bh = NULL;
+	struct inode *new_orphan_inode = NULL;
+	struct posix_acl *default_acl, *acl;
+	umode_t mode;
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
+		return -EOPNOTSUPP;
+
+	mode = inode->i_mode;
+	error = posix_acl_create(dir, &mode, &default_acl, &acl);
+	if (error) {
+		mlog_errno(error);
+		return error;
+	}
+
+	error = ocfs2_create_inode_in_orphan(dir, mode,
+					     &new_orphan_inode);
+	if (error) {
+		mlog_errno(error);
+		goto out;
+	}
+
+	error = ocfs2_rw_lock(inode, 1);
+	if (error) {
+		mlog_errno(error);
+		goto out;
+	}
+
+	error = ocfs2_inode_lock(inode, &old_bh, 1);
+	if (error) {
+		mlog_errno(error);
+		ocfs2_rw_unlock(inode, 1);
+		goto out;
+	}
+
+	down_write(&OCFS2_I(inode)->ip_xattr_sem);
+	down_write(&OCFS2_I(inode)->ip_alloc_sem);
+	error = __ocfs2_reflink(old_dentry, old_bh,
+				new_orphan_inode, preserve);
+	up_write(&OCFS2_I(inode)->ip_alloc_sem);
+	up_write(&OCFS2_I(inode)->ip_xattr_sem);
+
+	ocfs2_inode_unlock(inode, 1);
+	ocfs2_rw_unlock(inode, 1);
+	brelse(old_bh);
+
+	if (error) {
+		mlog_errno(error);
+		goto out;
+	}
+
+	/* If the security isn't preserved, we need to re-initialize them. */
+	if (!preserve) {
+		error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
+						    &new_dentry->d_name,
+						    default_acl, acl);
+		if (error)
+			mlog_errno(error);
+	}
+out:
+	if (default_acl)
+		posix_acl_release(default_acl);
+	if (acl)
+		posix_acl_release(acl);
+	if (!error) {
+		error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
+						       new_dentry);
+		if (error)
+			mlog_errno(error);
+	}
+
+	if (new_orphan_inode) {
+		/*
+		 * We need to open_unlock the inode no matter whether we
+		 * succeed or not, so that other nodes can delete it later.
+		 */
+		ocfs2_open_unlock(new_orphan_inode);
+		if (error)
+			iput(new_orphan_inode);
+	}
+
+	return error;
+}
+
+/*
+ * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
+ * sys_reflink().  This will go away when vfs_reflink() exists in
+ * fs/namei.c.
+ */
+
+/* copied from may_create in VFS. */
+static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
+{
+	if (d_really_is_positive(child))
+		return -EEXIST;
+	if (IS_DEADDIR(dir))
+		return -ENOENT;
+	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
+}
+
+/**
+ * ocfs2_vfs_reflink - Create a reference-counted link
+ *
+ * @old_dentry:        source dentry + inode
+ * @dir:       directory to create the target
+ * @new_dentry:        target dentry
+ * @preserve:  if true, preserve all file attributes
+ */
+static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
+			     struct dentry *new_dentry, bool preserve)
+{
+	struct inode *inode = d_inode(old_dentry);
+	int error;
+
+	if (!inode)
+		return -ENOENT;
+
+	error = ocfs2_may_create(dir, new_dentry);
+	if (error)
+		return error;
+
+	if (dir->i_sb != inode->i_sb)
+		return -EXDEV;
+
+	/*
+	 * A reflink to an append-only or immutable file cannot be created.
+	 */
+	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
+		return -EPERM;
+
+	/* Only regular files can be reflinked. */
+	if (!S_ISREG(inode->i_mode))
+		return -EPERM;
+
+	/*
+	 * If the caller wants to preserve ownership, they require the
+	 * rights to do so.
+	 */
+	if (preserve) {
+		if (!uid_eq(current_fsuid(), inode->i_uid) && !capable(CAP_CHOWN))
+			return -EPERM;
+		if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
+			return -EPERM;
+	}
+
+	/*
+	 * If the caller is modifying any aspect of the attributes, they
+	 * are not creating a snapshot.  They need read permission on the
+	 * file.
+	 */
+	if (!preserve) {
+		error = inode_permission(inode, MAY_READ);
+		if (error)
+			return error;
+	}
+
+	mutex_lock(&inode->i_mutex);
+	dquot_initialize(dir);
+	error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
+	mutex_unlock(&inode->i_mutex);
+	if (!error)
+		fsnotify_create(dir, new_dentry);
+	return error;
+}
+/*
+ * Most codes are copied from sys_linkat.
+ */
+int ocfs2_reflink_ioctl(struct inode *inode,
+			const char __user *oldname,
+			const char __user *newname,
+			bool preserve)
+{
+	struct dentry *new_dentry;
+	struct path old_path, new_path;
+	int error;
+
+	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
+		return -EOPNOTSUPP;
+
+	error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
+	if (error) {
+		mlog_errno(error);
+		return error;
+	}
+
+	new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
+	error = PTR_ERR(new_dentry);
+	if (IS_ERR(new_dentry)) {
+		mlog_errno(error);
+		goto out;
+	}
+
+	error = -EXDEV;
+	if (old_path.mnt != new_path.mnt) {
+		mlog_errno(error);
+		goto out_dput;
+	}
+
+	error = ocfs2_vfs_reflink(old_path.dentry,
+				  d_inode(new_path.dentry),
+				  new_dentry, preserve);
+out_dput:
+	done_path_create(&new_path, new_dentry);
+out:
+	path_put(&old_path);
+
+	return error;
+}
diff --git a/fs/ocfs2/refcounttree.h b/fs/ocfs2/refcounttree.h
new file mode 100644
index 000000000..6422bbcdb
--- /dev/null
+++ b/fs/ocfs2/refcounttree.h
@@ -0,0 +1,118 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * refcounttree.h
+ *
+ * Copyright (C) 2009 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+#ifndef OCFS2_REFCOUNTTREE_H
+#define OCFS2_REFCOUNTTREE_H
+
+struct ocfs2_refcount_tree {
+	struct rb_node rf_node;
+	u64 rf_blkno;
+	u32 rf_generation;
+	struct kref rf_getcnt;
+	struct rw_semaphore rf_sem;
+	struct ocfs2_lock_res rf_lockres;
+	int rf_removed;
+
+	/* the following 4 fields are used by caching_info. */
+	spinlock_t rf_lock;
+	struct ocfs2_caching_info rf_ci;
+	struct mutex rf_io_mutex;
+	struct super_block *rf_sb;
+};
+
+void ocfs2_purge_refcount_trees(struct ocfs2_super *osb);
+int ocfs2_lock_refcount_tree(struct ocfs2_super *osb, u64 ref_blkno, int rw,
+			     struct ocfs2_refcount_tree **tree,
+			     struct buffer_head **ref_bh);
+void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
+				struct ocfs2_refcount_tree *tree,
+				int rw);
+
+int ocfs2_decrease_refcount(struct inode *inode,
+			    handle_t *handle, u32 cpos, u32 len,
+			    struct ocfs2_alloc_context *meta_ac,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc,
+			    int delete);
+int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
+					  u64 refcount_loc,
+					  u64 phys_blkno,
+					  u32 clusters,
+					  int *credits,
+					  int *ref_blocks);
+int ocfs2_refcount_cow(struct inode *inode,
+		       struct buffer_head *di_bh,
+		       u32 cpos, u32 write_len, u32 max_cpos);
+
+typedef int (ocfs2_post_refcount_func)(struct inode *inode,
+				       handle_t *handle,
+				       void *para);
+/*
+ * Some refcount caller need to do more work after we modify the data b-tree
+ * during refcount operation(including CoW and add refcount flag), and make the
+ * transaction complete. So it must give us this structure so that we can do it
+ * within our transaction.
+ *
+ */
+struct ocfs2_post_refcount {
+	int credits;			/* credits it need for journal. */
+	ocfs2_post_refcount_func *func;	/* real function. */
+	void *para;
+};
+
+int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
+				       struct ocfs2_caching_info *ref_ci,
+				       struct buffer_head *ref_root_bh,
+				       struct ocfs2_xattr_value_root *xv,
+				       int *meta_add, int *credits);
+int ocfs2_refcount_cow_xattr(struct inode *inode,
+			     struct ocfs2_dinode *di,
+			     struct ocfs2_xattr_value_buf *vb,
+			     struct ocfs2_refcount_tree *ref_tree,
+			     struct buffer_head *ref_root_bh,
+			     u32 cpos, u32 write_len,
+			     struct ocfs2_post_refcount *post);
+int ocfs2_duplicate_clusters_by_page(handle_t *handle,
+				     struct inode *inode,
+				     u32 cpos, u32 old_cluster,
+				     u32 new_cluster, u32 new_len);
+int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
+				    struct inode *inode,
+				    u32 cpos, u32 old_cluster,
+				    u32 new_cluster, u32 new_len);
+int ocfs2_cow_sync_writeback(struct super_block *sb,
+			     struct inode *inode,
+			     u32 cpos, u32 num_clusters);
+int ocfs2_add_refcount_flag(struct inode *inode,
+			    struct ocfs2_extent_tree *data_et,
+			    struct ocfs2_caching_info *ref_ci,
+			    struct buffer_head *ref_root_bh,
+			    u32 cpos, u32 p_cluster, u32 num_clusters,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc,
+			    struct ocfs2_post_refcount *post);
+int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh);
+int ocfs2_try_remove_refcount_tree(struct inode *inode,
+				   struct buffer_head *di_bh);
+int ocfs2_increase_refcount(handle_t *handle,
+			    struct ocfs2_caching_info *ci,
+			    struct buffer_head *ref_root_bh,
+			    u64 cpos, u32 len,
+			    struct ocfs2_alloc_context *meta_ac,
+			    struct ocfs2_cached_dealloc_ctxt *dealloc);
+int ocfs2_reflink_ioctl(struct inode *inode,
+			const char __user *oldname,
+			const char __user *newname,
+			bool preserve);
+#endif /* OCFS2_REFCOUNTTREE_H */
diff --git a/fs/ocfs2/reservations.c b/fs/ocfs2/reservations.c
new file mode 100644
index 000000000..6a348b029
--- /dev/null
+++ b/fs/ocfs2/reservations.c
@@ -0,0 +1,839 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * reservations.c
+ *
+ * Allocation reservations implementation
+ *
+ * Some code borrowed from fs/ext3/balloc.c and is:
+ *
+ * Copyright (C) 1992, 1993, 1994, 1995
+ * Remy Card (card@masi.ibp.fr)
+ * Laboratoire MASI - Institut Blaise Pascal
+ * Universite Pierre et Marie Curie (Paris VI)
+ *
+ * The rest is copyright (C) 2010 Novell.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+#include <linux/bitops.h>
+#include <linux/list.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+#include "ocfs2_trace.h"
+
+#ifdef CONFIG_OCFS2_DEBUG_FS
+#define OCFS2_CHECK_RESERVATIONS
+#endif
+
+static DEFINE_SPINLOCK(resv_lock);
+
+#define	OCFS2_MIN_RESV_WINDOW_BITS	8
+#define	OCFS2_MAX_RESV_WINDOW_BITS	1024
+
+int ocfs2_dir_resv_allowed(struct ocfs2_super *osb)
+{
+	return (osb->osb_resv_level && osb->osb_dir_resv_level);
+}
+
+static unsigned int ocfs2_resv_window_bits(struct ocfs2_reservation_map *resmap,
+					   struct ocfs2_alloc_reservation *resv)
+{
+	struct ocfs2_super *osb = resmap->m_osb;
+	unsigned int bits;
+
+	if (!(resv->r_flags & OCFS2_RESV_FLAG_DIR)) {
+		/* 8, 16, 32, 64, 128, 256, 512, 1024 */
+		bits = 4 << osb->osb_resv_level;
+	} else {
+		bits = 4 << osb->osb_dir_resv_level;
+	}
+	return bits;
+}
+
+static inline unsigned int ocfs2_resv_end(struct ocfs2_alloc_reservation *resv)
+{
+	if (resv->r_len)
+		return resv->r_start + resv->r_len - 1;
+	return resv->r_start;
+}
+
+static inline int ocfs2_resv_empty(struct ocfs2_alloc_reservation *resv)
+{
+	return !!(resv->r_len == 0);
+}
+
+static inline int ocfs2_resmap_disabled(struct ocfs2_reservation_map *resmap)
+{
+	if (resmap->m_osb->osb_resv_level == 0)
+		return 1;
+	return 0;
+}
+
+static void ocfs2_dump_resv(struct ocfs2_reservation_map *resmap)
+{
+	struct ocfs2_super *osb = resmap->m_osb;
+	struct rb_node *node;
+	struct ocfs2_alloc_reservation *resv;
+	int i = 0;
+
+	mlog(ML_NOTICE, "Dumping resmap for device %s. Bitmap length: %u\n",
+	     osb->dev_str, resmap->m_bitmap_len);
+
+	node = rb_first(&resmap->m_reservations);
+	while (node) {
+		resv = rb_entry(node, struct ocfs2_alloc_reservation, r_node);
+
+		mlog(ML_NOTICE, "start: %u\tend: %u\tlen: %u\tlast_start: %u"
+		     "\tlast_len: %u\n", resv->r_start,
+		     ocfs2_resv_end(resv), resv->r_len, resv->r_last_start,
+		     resv->r_last_len);
+
+		node = rb_next(node);
+		i++;
+	}
+
+	mlog(ML_NOTICE, "%d reservations found. LRU follows\n", i);
+
+	i = 0;
+	list_for_each_entry(resv, &resmap->m_lru, r_lru) {
+		mlog(ML_NOTICE, "LRU(%d) start: %u\tend: %u\tlen: %u\t"
+		     "last_start: %u\tlast_len: %u\n", i, resv->r_start,
+		     ocfs2_resv_end(resv), resv->r_len, resv->r_last_start,
+		     resv->r_last_len);
+
+		i++;
+	}
+}
+
+#ifdef OCFS2_CHECK_RESERVATIONS
+static int ocfs2_validate_resmap_bits(struct ocfs2_reservation_map *resmap,
+				      int i,
+				      struct ocfs2_alloc_reservation *resv)
+{
+	char *disk_bitmap = resmap->m_disk_bitmap;
+	unsigned int start = resv->r_start;
+	unsigned int end = ocfs2_resv_end(resv);
+
+	while (start <= end) {
+		if (ocfs2_test_bit(start, disk_bitmap)) {
+			mlog(ML_ERROR,
+			     "reservation %d covers an allocated area "
+			     "starting at bit %u!\n", i, start);
+			return 1;
+		}
+
+		start++;
+	}
+	return 0;
+}
+
+static void ocfs2_check_resmap(struct ocfs2_reservation_map *resmap)
+{
+	unsigned int off = 0;
+	int i = 0;
+	struct rb_node *node;
+	struct ocfs2_alloc_reservation *resv;
+
+	node = rb_first(&resmap->m_reservations);
+	while (node) {
+		resv = rb_entry(node, struct ocfs2_alloc_reservation, r_node);
+
+		if (i > 0 && resv->r_start <= off) {
+			mlog(ML_ERROR, "reservation %d has bad start off!\n",
+			     i);
+			goto bad;
+		}
+
+		if (resv->r_len == 0) {
+			mlog(ML_ERROR, "reservation %d has no length!\n",
+			     i);
+			goto bad;
+		}
+
+		if (resv->r_start > ocfs2_resv_end(resv)) {
+			mlog(ML_ERROR, "reservation %d has invalid range!\n",
+			     i);
+			goto bad;
+		}
+
+		if (ocfs2_resv_end(resv) >= resmap->m_bitmap_len) {
+			mlog(ML_ERROR, "reservation %d extends past bitmap!\n",
+			     i);
+			goto bad;
+		}
+
+		if (ocfs2_validate_resmap_bits(resmap, i, resv))
+			goto bad;
+
+		off = ocfs2_resv_end(resv);
+		node = rb_next(node);
+
+		i++;
+	}
+	return;
+
+bad:
+	ocfs2_dump_resv(resmap);
+	BUG();
+}
+#else
+static inline void ocfs2_check_resmap(struct ocfs2_reservation_map *resmap)
+{
+
+}
+#endif
+
+void ocfs2_resv_init_once(struct ocfs2_alloc_reservation *resv)
+{
+	memset(resv, 0, sizeof(*resv));
+	INIT_LIST_HEAD(&resv->r_lru);
+}
+
+void ocfs2_resv_set_type(struct ocfs2_alloc_reservation *resv,
+			 unsigned int flags)
+{
+	BUG_ON(flags & ~OCFS2_RESV_TYPES);
+
+	resv->r_flags |= flags;
+}
+
+int ocfs2_resmap_init(struct ocfs2_super *osb,
+		      struct ocfs2_reservation_map *resmap)
+{
+	memset(resmap, 0, sizeof(*resmap));
+
+	resmap->m_osb = osb;
+	resmap->m_reservations = RB_ROOT;
+	/* m_bitmap_len is initialized to zero by the above memset. */
+	INIT_LIST_HEAD(&resmap->m_lru);
+
+	return 0;
+}
+
+static void ocfs2_resv_mark_lru(struct ocfs2_reservation_map *resmap,
+				struct ocfs2_alloc_reservation *resv)
+{
+	assert_spin_locked(&resv_lock);
+
+	if (!list_empty(&resv->r_lru))
+		list_del_init(&resv->r_lru);
+
+	list_add_tail(&resv->r_lru, &resmap->m_lru);
+}
+
+static void __ocfs2_resv_trunc(struct ocfs2_alloc_reservation *resv)
+{
+	resv->r_len = 0;
+	resv->r_start = 0;
+}
+
+static void ocfs2_resv_remove(struct ocfs2_reservation_map *resmap,
+			      struct ocfs2_alloc_reservation *resv)
+{
+	if (resv->r_flags & OCFS2_RESV_FLAG_INUSE) {
+		list_del_init(&resv->r_lru);
+		rb_erase(&resv->r_node, &resmap->m_reservations);
+		resv->r_flags &= ~OCFS2_RESV_FLAG_INUSE;
+	}
+}
+
+static void __ocfs2_resv_discard(struct ocfs2_reservation_map *resmap,
+				 struct ocfs2_alloc_reservation *resv)
+{
+	assert_spin_locked(&resv_lock);
+
+	__ocfs2_resv_trunc(resv);
+	/*
+	 * last_len and last_start no longer make sense if
+	 * we're changing the range of our allocations.
+	 */
+	resv->r_last_len = resv->r_last_start = 0;
+
+	ocfs2_resv_remove(resmap, resv);
+}
+
+/* does nothing if 'resv' is null */
+void ocfs2_resv_discard(struct ocfs2_reservation_map *resmap,
+			struct ocfs2_alloc_reservation *resv)
+{
+	if (resv) {
+		spin_lock(&resv_lock);
+		__ocfs2_resv_discard(resmap, resv);
+		spin_unlock(&resv_lock);
+	}
+}
+
+static void ocfs2_resmap_clear_all_resv(struct ocfs2_reservation_map *resmap)
+{
+	struct rb_node *node;
+	struct ocfs2_alloc_reservation *resv;
+
+	assert_spin_locked(&resv_lock);
+
+	while ((node = rb_last(&resmap->m_reservations)) != NULL) {
+		resv = rb_entry(node, struct ocfs2_alloc_reservation, r_node);
+
+		__ocfs2_resv_discard(resmap, resv);
+	}
+}
+
+void ocfs2_resmap_restart(struct ocfs2_reservation_map *resmap,
+			  unsigned int clen, char *disk_bitmap)
+{
+	if (ocfs2_resmap_disabled(resmap))
+		return;
+
+	spin_lock(&resv_lock);
+
+	ocfs2_resmap_clear_all_resv(resmap);
+	resmap->m_bitmap_len = clen;
+	resmap->m_disk_bitmap = disk_bitmap;
+
+	spin_unlock(&resv_lock);
+}
+
+void ocfs2_resmap_uninit(struct ocfs2_reservation_map *resmap)
+{
+	/* Does nothing for now. Keep this around for API symmetry */
+}
+
+static void ocfs2_resv_insert(struct ocfs2_reservation_map *resmap,
+			      struct ocfs2_alloc_reservation *new)
+{
+	struct rb_root *root = &resmap->m_reservations;
+	struct rb_node *parent = NULL;
+	struct rb_node **p = &root->rb_node;
+	struct ocfs2_alloc_reservation *tmp;
+
+	assert_spin_locked(&resv_lock);
+
+	trace_ocfs2_resv_insert(new->r_start, new->r_len);
+
+	while (*p) {
+		parent = *p;
+
+		tmp = rb_entry(parent, struct ocfs2_alloc_reservation, r_node);
+
+		if (new->r_start < tmp->r_start) {
+			p = &(*p)->rb_left;
+
+			/*
+			 * This is a good place to check for
+			 * overlapping reservations.
+			 */
+			BUG_ON(ocfs2_resv_end(new) >= tmp->r_start);
+		} else if (new->r_start > ocfs2_resv_end(tmp)) {
+			p = &(*p)->rb_right;
+		} else {
+			/* This should never happen! */
+			mlog(ML_ERROR, "Duplicate reservation window!\n");
+			BUG();
+		}
+	}
+
+	rb_link_node(&new->r_node, parent, p);
+	rb_insert_color(&new->r_node, root);
+	new->r_flags |= OCFS2_RESV_FLAG_INUSE;
+
+	ocfs2_resv_mark_lru(resmap, new);
+
+	ocfs2_check_resmap(resmap);
+}
+
+/**
+ * ocfs2_find_resv_lhs() - find the window which contains goal
+ * @resmap: reservation map to search
+ * @goal: which bit to search for
+ *
+ * If a window containing that goal is not found, we return the window
+ * which comes before goal. Returns NULL on empty rbtree or no window
+ * before goal.
+ */
+static struct ocfs2_alloc_reservation *
+ocfs2_find_resv_lhs(struct ocfs2_reservation_map *resmap, unsigned int goal)
+{
+	struct ocfs2_alloc_reservation *resv = NULL;
+	struct ocfs2_alloc_reservation *prev_resv = NULL;
+	struct rb_node *node = resmap->m_reservations.rb_node;
+
+	assert_spin_locked(&resv_lock);
+
+	if (!node)
+		return NULL;
+
+	node = rb_first(&resmap->m_reservations);
+	while (node) {
+		resv = rb_entry(node, struct ocfs2_alloc_reservation, r_node);
+
+		if (resv->r_start <= goal && ocfs2_resv_end(resv) >= goal)
+			break;
+
+		/* Check if we overshot the reservation just before goal? */
+		if (resv->r_start > goal) {
+			resv = prev_resv;
+			break;
+		}
+
+		prev_resv = resv;
+		node = rb_next(node);
+	}
+
+	return resv;
+}
+
+/*
+ * We are given a range within the bitmap, which corresponds to a gap
+ * inside the reservations tree (search_start, search_len). The range
+ * can be anything from the whole bitmap, to a gap between
+ * reservations.
+ *
+ * The start value of *rstart is insignificant.
+ *
+ * This function searches the bitmap range starting at search_start
+ * with length search_len for a set of contiguous free bits. We try
+ * to find up to 'wanted' bits, but can sometimes return less.
+ *
+ * Returns the length of allocation, 0 if no free bits are found.
+ *
+ * *cstart and *clen will also be populated with the result.
+ */
+static int ocfs2_resmap_find_free_bits(struct ocfs2_reservation_map *resmap,
+				       unsigned int wanted,
+				       unsigned int search_start,
+				       unsigned int search_len,
+				       unsigned int *rstart,
+				       unsigned int *rlen)
+{
+	void *bitmap = resmap->m_disk_bitmap;
+	unsigned int best_start, best_len = 0;
+	int offset, start, found;
+
+	trace_ocfs2_resmap_find_free_bits_begin(search_start, search_len,
+						wanted, resmap->m_bitmap_len);
+
+	found = best_start = best_len = 0;
+
+	start = search_start;
+	while ((offset = ocfs2_find_next_zero_bit(bitmap, resmap->m_bitmap_len,
+						 start)) != -1) {
+		/* Search reached end of the region */
+		if (offset >= (search_start + search_len))
+			break;
+
+		if (offset == start) {
+			/* we found a zero */
+			found++;
+			/* move start to the next bit to test */
+			start++;
+		} else {
+			/* got a zero after some ones */
+			found = 1;
+			start = offset + 1;
+		}
+		if (found > best_len) {
+			best_len = found;
+			best_start = start - found;
+		}
+
+		if (found >= wanted)
+			break;
+	}
+
+	if (best_len == 0)
+		return 0;
+
+	if (best_len >= wanted)
+		best_len = wanted;
+
+	*rlen = best_len;
+	*rstart = best_start;
+
+	trace_ocfs2_resmap_find_free_bits_end(best_start, best_len);
+
+	return *rlen;
+}
+
+static void __ocfs2_resv_find_window(struct ocfs2_reservation_map *resmap,
+				     struct ocfs2_alloc_reservation *resv,
+				     unsigned int goal, unsigned int wanted)
+{
+	struct rb_root *root = &resmap->m_reservations;
+	unsigned int gap_start, gap_end, gap_len;
+	struct ocfs2_alloc_reservation *prev_resv, *next_resv;
+	struct rb_node *prev, *next;
+	unsigned int cstart, clen;
+	unsigned int best_start = 0, best_len = 0;
+
+	/*
+	 * Nasty cases to consider:
+	 *
+	 * - rbtree is empty
+	 * - our window should be first in all reservations
+	 * - our window should be last in all reservations
+	 * - need to make sure we don't go past end of bitmap
+	 */
+	trace_ocfs2_resv_find_window_begin(resv->r_start, ocfs2_resv_end(resv),
+					   goal, wanted, RB_EMPTY_ROOT(root));
+
+	assert_spin_locked(&resv_lock);
+
+	if (RB_EMPTY_ROOT(root)) {
+		/*
+		 * Easiest case - empty tree. We can just take
+		 * whatever window of free bits we want.
+		 */
+		clen = ocfs2_resmap_find_free_bits(resmap, wanted, goal,
+						   resmap->m_bitmap_len - goal,
+						   &cstart, &clen);
+
+		/*
+		 * This should never happen - the local alloc window
+		 * will always have free bits when we're called.
+		 */
+		BUG_ON(goal == 0 && clen == 0);
+
+		if (clen == 0)
+			return;
+
+		resv->r_start = cstart;
+		resv->r_len = clen;
+
+		ocfs2_resv_insert(resmap, resv);
+		return;
+	}
+
+	prev_resv = ocfs2_find_resv_lhs(resmap, goal);
+
+	if (prev_resv == NULL) {
+		/*
+		 * A NULL here means that the search code couldn't
+		 * find a window that starts before goal.
+		 *
+		 * However, we can take the first window after goal,
+		 * which is also by definition, the leftmost window in
+		 * the entire tree. If we can find free bits in the
+		 * gap between goal and the LHS window, then the
+		 * reservation can safely be placed there.
+		 *
+		 * Otherwise we fall back to a linear search, checking
+		 * the gaps in between windows for a place to
+		 * allocate.
+		 */
+
+		next = rb_first(root);
+		next_resv = rb_entry(next, struct ocfs2_alloc_reservation,
+				     r_node);
+
+		/*
+		 * The search should never return such a window. (see
+		 * comment above
+		 */
+		if (next_resv->r_start <= goal) {
+			mlog(ML_ERROR, "goal: %u next_resv: start %u len %u\n",
+			     goal, next_resv->r_start, next_resv->r_len);
+			ocfs2_dump_resv(resmap);
+			BUG();
+		}
+
+		clen = ocfs2_resmap_find_free_bits(resmap, wanted, goal,
+						   next_resv->r_start - goal,
+						   &cstart, &clen);
+		if (clen) {
+			best_len = clen;
+			best_start = cstart;
+			if (best_len == wanted)
+				goto out_insert;
+		}
+
+		prev_resv = next_resv;
+		next_resv = NULL;
+	}
+
+	trace_ocfs2_resv_find_window_prev(prev_resv->r_start,
+					  ocfs2_resv_end(prev_resv));
+
+	prev = &prev_resv->r_node;
+
+	/* Now we do a linear search for a window, starting at 'prev_rsv' */
+	while (1) {
+		next = rb_next(prev);
+		if (next) {
+			next_resv = rb_entry(next,
+					     struct ocfs2_alloc_reservation,
+					     r_node);
+
+			gap_start = ocfs2_resv_end(prev_resv) + 1;
+			gap_end = next_resv->r_start - 1;
+			gap_len = gap_end - gap_start + 1;
+		} else {
+			/*
+			 * We're at the rightmost edge of the
+			 * tree. See if a reservation between this
+			 * window and the end of the bitmap will work.
+			 */
+			gap_start = ocfs2_resv_end(prev_resv) + 1;
+			gap_len = resmap->m_bitmap_len - gap_start;
+			gap_end = resmap->m_bitmap_len - 1;
+		}
+
+		trace_ocfs2_resv_find_window_next(next ? next_resv->r_start: -1,
+					next ? ocfs2_resv_end(next_resv) : -1);
+		/*
+		 * No need to check this gap if we have already found
+		 * a larger region of free bits.
+		 */
+		if (gap_len <= best_len)
+			goto next_resv;
+
+		clen = ocfs2_resmap_find_free_bits(resmap, wanted, gap_start,
+						   gap_len, &cstart, &clen);
+		if (clen == wanted) {
+			best_len = clen;
+			best_start = cstart;
+			goto out_insert;
+		} else if (clen > best_len) {
+			best_len = clen;
+			best_start = cstart;
+		}
+
+next_resv:
+		if (!next)
+			break;
+
+		prev = next;
+		prev_resv = rb_entry(prev, struct ocfs2_alloc_reservation,
+				     r_node);
+	}
+
+out_insert:
+	if (best_len) {
+		resv->r_start = best_start;
+		resv->r_len = best_len;
+		ocfs2_resv_insert(resmap, resv);
+	}
+}
+
+static void ocfs2_cannibalize_resv(struct ocfs2_reservation_map *resmap,
+				   struct ocfs2_alloc_reservation *resv,
+				   unsigned int wanted)
+{
+	struct ocfs2_alloc_reservation *lru_resv;
+	int tmpwindow = !!(resv->r_flags & OCFS2_RESV_FLAG_TMP);
+	unsigned int min_bits;
+
+	if (!tmpwindow)
+		min_bits = ocfs2_resv_window_bits(resmap, resv) >> 1;
+	else
+		min_bits = wanted; /* We at know the temp window will use all
+				    * of these bits */
+
+	/*
+	 * Take the first reservation off the LRU as our 'target'. We
+	 * don't try to be smart about it. There might be a case for
+	 * searching based on size but I don't have enough data to be
+	 * sure. --Mark (3/16/2010)
+	 */
+	lru_resv = list_first_entry(&resmap->m_lru,
+				    struct ocfs2_alloc_reservation, r_lru);
+
+	trace_ocfs2_cannibalize_resv_begin(lru_resv->r_start,
+					   lru_resv->r_len,
+					   ocfs2_resv_end(lru_resv));
+
+	/*
+	 * Cannibalize (some or all) of the target reservation and
+	 * feed it to the current window.
+	 */
+	if (lru_resv->r_len <= min_bits) {
+		/*
+		 * Discard completely if size is less than or equal to a
+		 * reasonable threshold - 50% of window bits for non temporary
+		 * windows.
+		 */
+		resv->r_start = lru_resv->r_start;
+		resv->r_len = lru_resv->r_len;
+
+		__ocfs2_resv_discard(resmap, lru_resv);
+	} else {
+		unsigned int shrink;
+		if (tmpwindow)
+			shrink = min_bits;
+		else
+			shrink = lru_resv->r_len / 2;
+
+		lru_resv->r_len -= shrink;
+
+		resv->r_start = ocfs2_resv_end(lru_resv) + 1;
+		resv->r_len = shrink;
+	}
+
+	trace_ocfs2_cannibalize_resv_end(resv->r_start, ocfs2_resv_end(resv),
+					 resv->r_len, resv->r_last_start,
+					 resv->r_last_len);
+
+	ocfs2_resv_insert(resmap, resv);
+}
+
+static void ocfs2_resv_find_window(struct ocfs2_reservation_map *resmap,
+				   struct ocfs2_alloc_reservation *resv,
+				   unsigned int wanted)
+{
+	unsigned int goal = 0;
+
+	BUG_ON(!ocfs2_resv_empty(resv));
+
+	/*
+	 * Begin by trying to get a window as close to the previous
+	 * one as possible. Using the most recent allocation as a
+	 * start goal makes sense.
+	 */
+	if (resv->r_last_len) {
+		goal = resv->r_last_start + resv->r_last_len;
+		if (goal >= resmap->m_bitmap_len)
+			goal = 0;
+	}
+
+	__ocfs2_resv_find_window(resmap, resv, goal, wanted);
+
+	/* Search from last alloc didn't work, try once more from beginning. */
+	if (ocfs2_resv_empty(resv) && goal != 0)
+		__ocfs2_resv_find_window(resmap, resv, 0, wanted);
+
+	if (ocfs2_resv_empty(resv)) {
+		/*
+		 * Still empty? Pull oldest one off the LRU, remove it from
+		 * tree, put this one in it's place.
+		 */
+		ocfs2_cannibalize_resv(resmap, resv, wanted);
+	}
+
+	BUG_ON(ocfs2_resv_empty(resv));
+}
+
+int ocfs2_resmap_resv_bits(struct ocfs2_reservation_map *resmap,
+			   struct ocfs2_alloc_reservation *resv,
+			   int *cstart, int *clen)
+{
+	if (resv == NULL || ocfs2_resmap_disabled(resmap))
+		return -ENOSPC;
+
+	spin_lock(&resv_lock);
+
+	if (ocfs2_resv_empty(resv)) {
+		/*
+		 * We don't want to over-allocate for temporary
+		 * windows. Otherwise, we run the risk of fragmenting the
+		 * allocation space.
+		 */
+		unsigned int wanted = ocfs2_resv_window_bits(resmap, resv);
+
+		if ((resv->r_flags & OCFS2_RESV_FLAG_TMP) || wanted < *clen)
+			wanted = *clen;
+
+		/*
+		 * Try to get a window here. If it works, we must fall
+		 * through and test the bitmap . This avoids some
+		 * ping-ponging of windows due to non-reserved space
+		 * being allocation before we initialize a window for
+		 * that inode.
+		 */
+		ocfs2_resv_find_window(resmap, resv, wanted);
+		trace_ocfs2_resmap_resv_bits(resv->r_start, resv->r_len);
+	}
+
+	BUG_ON(ocfs2_resv_empty(resv));
+
+	*cstart = resv->r_start;
+	*clen = resv->r_len;
+
+	spin_unlock(&resv_lock);
+	return 0;
+}
+
+static void
+	ocfs2_adjust_resv_from_alloc(struct ocfs2_reservation_map *resmap,
+				     struct ocfs2_alloc_reservation *resv,
+				     unsigned int start, unsigned int end)
+{
+	unsigned int rhs = 0;
+	unsigned int old_end = ocfs2_resv_end(resv);
+
+	BUG_ON(start != resv->r_start || old_end < end);
+
+	/*
+	 * Completely used? We can remove it then.
+	 */
+	if (old_end == end) {
+		__ocfs2_resv_discard(resmap, resv);
+		return;
+	}
+
+	rhs = old_end - end;
+
+	/*
+	 * This should have been trapped above.
+	 */
+	BUG_ON(rhs == 0);
+
+	resv->r_start = end + 1;
+	resv->r_len = old_end - resv->r_start + 1;
+}
+
+void ocfs2_resmap_claimed_bits(struct ocfs2_reservation_map *resmap,
+			       struct ocfs2_alloc_reservation *resv,
+			       u32 cstart, u32 clen)
+{
+	unsigned int cend = cstart + clen - 1;
+
+	if (resmap == NULL || ocfs2_resmap_disabled(resmap))
+		return;
+
+	if (resv == NULL)
+		return;
+
+	BUG_ON(cstart != resv->r_start);
+
+	spin_lock(&resv_lock);
+
+	trace_ocfs2_resmap_claimed_bits_begin(cstart, cend, clen, resv->r_start,
+					      ocfs2_resv_end(resv), resv->r_len,
+					      resv->r_last_start,
+					      resv->r_last_len);
+
+	BUG_ON(cstart < resv->r_start);
+	BUG_ON(cstart > ocfs2_resv_end(resv));
+	BUG_ON(cend > ocfs2_resv_end(resv));
+
+	ocfs2_adjust_resv_from_alloc(resmap, resv, cstart, cend);
+	resv->r_last_start = cstart;
+	resv->r_last_len = clen;
+
+	/*
+	 * May have been discarded above from
+	 * ocfs2_adjust_resv_from_alloc().
+	 */
+	if (!ocfs2_resv_empty(resv))
+		ocfs2_resv_mark_lru(resmap, resv);
+
+	trace_ocfs2_resmap_claimed_bits_end(resv->r_start, ocfs2_resv_end(resv),
+					    resv->r_len, resv->r_last_start,
+					    resv->r_last_len);
+
+	ocfs2_check_resmap(resmap);
+
+	spin_unlock(&resv_lock);
+}
diff --git a/fs/ocfs2/reservations.h b/fs/ocfs2/reservations.h
new file mode 100644
index 000000000..42c2b804f
--- /dev/null
+++ b/fs/ocfs2/reservations.h
@@ -0,0 +1,159 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * reservations.h
+ *
+ * Allocation reservations function prototypes and structures.
+ *
+ * Copyright (C) 2010 Novell.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#ifndef	OCFS2_RESERVATIONS_H
+#define	OCFS2_RESERVATIONS_H
+
+#include <linux/rbtree.h>
+
+#define OCFS2_DEFAULT_RESV_LEVEL	2
+#define OCFS2_MAX_RESV_LEVEL	9
+#define OCFS2_MIN_RESV_LEVEL	0
+
+struct ocfs2_alloc_reservation {
+	struct rb_node	r_node;
+
+	unsigned int	r_start;	/* Beginning of current window */
+	unsigned int	r_len;		/* Length of the window */
+
+	unsigned int	r_last_len;	/* Length of most recent alloc */
+	unsigned int	r_last_start;	/* Start of most recent alloc */
+	struct list_head	r_lru;	/* LRU list head */
+
+	unsigned int	r_flags;
+};
+
+#define	OCFS2_RESV_FLAG_INUSE	0x01	/* Set when r_node is part of a btree */
+#define	OCFS2_RESV_FLAG_TMP	0x02	/* Temporary reservation, will be
+					 * destroyed immedately after use */
+#define	OCFS2_RESV_FLAG_DIR	0x04	/* Reservation is for an unindexed
+					 * directory btree */
+
+struct ocfs2_reservation_map {
+	struct rb_root		m_reservations;
+	char			*m_disk_bitmap;
+
+	struct ocfs2_super	*m_osb;
+
+	/* The following are not initialized to meaningful values until a disk
+	 * bitmap is provided. */
+	u32			m_bitmap_len;	/* Number of valid
+						 * bits available */
+
+	struct list_head	m_lru;		/* LRU of reservations
+						 * structures. */
+
+};
+
+void ocfs2_resv_init_once(struct ocfs2_alloc_reservation *resv);
+
+#define OCFS2_RESV_TYPES	(OCFS2_RESV_FLAG_TMP|OCFS2_RESV_FLAG_DIR)
+void ocfs2_resv_set_type(struct ocfs2_alloc_reservation *resv,
+			 unsigned int flags);
+
+int ocfs2_dir_resv_allowed(struct ocfs2_super *osb);
+
+/**
+ * ocfs2_resv_discard() - truncate a reservation
+ * @resmap:
+ * @resv: the reservation to truncate.
+ *
+ * After this function is called, the reservation will be empty, and
+ * unlinked from the rbtree.
+ */
+void ocfs2_resv_discard(struct ocfs2_reservation_map *resmap,
+			struct ocfs2_alloc_reservation *resv);
+
+
+/**
+ * ocfs2_resmap_init() - Initialize fields of a reservations bitmap
+ * @resmap: struct ocfs2_reservation_map to initialize
+ * @obj: unused for now
+ * @ops: unused for now
+ * @max_bitmap_bytes: Maximum size of the bitmap (typically blocksize)
+ *
+ * Only possible return value other than '0' is -ENOMEM for failure to
+ * allocation mirror bitmap.
+ */
+int ocfs2_resmap_init(struct ocfs2_super *osb,
+		      struct ocfs2_reservation_map *resmap);
+
+/**
+ * ocfs2_resmap_restart() - "restart" a reservation bitmap
+ * @resmap: reservations bitmap
+ * @clen: Number of valid bits in the bitmap
+ * @disk_bitmap: the disk bitmap this resmap should refer to.
+ *
+ * Re-initialize the parameters of a reservation bitmap. This is
+ * useful for local alloc window slides.
+ *
+ * This function will call ocfs2_trunc_resv against all existing
+ * reservations. A future version will recalculate existing
+ * reservations based on the new bitmap.
+ */
+void ocfs2_resmap_restart(struct ocfs2_reservation_map *resmap,
+			  unsigned int clen, char *disk_bitmap);
+
+/**
+ * ocfs2_resmap_uninit() - uninitialize a reservation bitmap structure
+ * @resmap: the struct ocfs2_reservation_map to uninitialize
+ */
+void ocfs2_resmap_uninit(struct ocfs2_reservation_map *resmap);
+
+/**
+ * ocfs2_resmap_resv_bits() - Return still-valid reservation bits
+ * @resmap: reservations bitmap
+ * @resv: reservation to base search from
+ * @cstart: start of proposed allocation
+ * @clen: length (in clusters) of proposed allocation
+ *
+ * Using the reservation data from resv, this function will compare
+ * resmap and resmap->m_disk_bitmap to determine what part (if any) of
+ * the reservation window is still clear to use. If resv is empty,
+ * this function will try to allocate a window for it.
+ *
+ * On success, zero is returned and the valid allocation area is set in cstart
+ * and clen.
+ *
+ * Returns -ENOSPC if reservations are disabled.
+ */
+int ocfs2_resmap_resv_bits(struct ocfs2_reservation_map *resmap,
+			   struct ocfs2_alloc_reservation *resv,
+			   int *cstart, int *clen);
+
+/**
+ * ocfs2_resmap_claimed_bits() - Tell the reservation code that bits were used.
+ * @resmap: reservations bitmap
+ * @resv: optional reservation to recalulate based on new bitmap
+ * @cstart: start of allocation in clusters
+ * @clen: end of allocation in clusters.
+ *
+ * Tell the reservation code that bits were used to fulfill allocation in
+ * resmap. The bits don't have to have been part of any existing
+ * reservation. But we must always call this function when bits are claimed.
+ * Internally, the reservations code will use this information to mark the
+ * reservations bitmap. If resv is passed, it's next allocation window will be
+ * calculated. It also expects that 'cstart' is the same as we passed back
+ * from ocfs2_resmap_resv_bits().
+ */
+void ocfs2_resmap_claimed_bits(struct ocfs2_reservation_map *resmap,
+			       struct ocfs2_alloc_reservation *resv,
+			       u32 cstart, u32 clen);
+
+#endif	/* OCFS2_RESERVATIONS_H */
diff --git a/fs/ocfs2/resize.c b/fs/ocfs2/resize.c
new file mode 100644
index 000000000..d5da6f624
--- /dev/null
+++ b/fs/ocfs2/resize.c
@@ -0,0 +1,589 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * resize.c
+ *
+ * volume resize.
+ * Inspired by ext3/resize.c.
+ *
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "dlmglue.h"
+#include "inode.h"
+#include "journal.h"
+#include "super.h"
+#include "sysfile.h"
+#include "uptodate.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+#include "suballoc.h"
+#include "resize.h"
+
+/*
+ * Check whether there are new backup superblocks exist
+ * in the last group. If there are some, mark them or clear
+ * them in the bitmap.
+ *
+ * Return how many backups we find in the last group.
+ */
+static u16 ocfs2_calc_new_backup_super(struct inode *inode,
+				       struct ocfs2_group_desc *gd,
+				       u16 cl_cpg,
+				       int set)
+{
+	int i;
+	u16 backups = 0;
+	u32 cluster;
+	u64 blkno, gd_blkno, lgd_blkno = le64_to_cpu(gd->bg_blkno);
+
+	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
+		blkno = ocfs2_backup_super_blkno(inode->i_sb, i);
+		cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
+
+		gd_blkno = ocfs2_which_cluster_group(inode, cluster);
+		if (gd_blkno < lgd_blkno)
+			continue;
+		else if (gd_blkno > lgd_blkno)
+			break;
+
+		if (set)
+			ocfs2_set_bit(cluster % cl_cpg,
+				      (unsigned long *)gd->bg_bitmap);
+		else
+			ocfs2_clear_bit(cluster % cl_cpg,
+					(unsigned long *)gd->bg_bitmap);
+		backups++;
+	}
+
+	return backups;
+}
+
+static int ocfs2_update_last_group_and_inode(handle_t *handle,
+					     struct inode *bm_inode,
+					     struct buffer_head *bm_bh,
+					     struct buffer_head *group_bh,
+					     u32 first_new_cluster,
+					     int new_clusters)
+{
+	int ret = 0;
+	struct ocfs2_super *osb = OCFS2_SB(bm_inode->i_sb);
+	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bm_bh->b_data;
+	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
+	struct ocfs2_chain_rec *cr;
+	struct ocfs2_group_desc *group;
+	u16 chain, num_bits, backups = 0;
+	u16 cl_bpc = le16_to_cpu(cl->cl_bpc);
+	u16 cl_cpg = le16_to_cpu(cl->cl_cpg);
+
+	trace_ocfs2_update_last_group_and_inode(new_clusters,
+						first_new_cluster);
+
+	ret = ocfs2_journal_access_gd(handle, INODE_CACHE(bm_inode),
+				      group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	group = (struct ocfs2_group_desc *)group_bh->b_data;
+
+	/* update the group first. */
+	num_bits = new_clusters * cl_bpc;
+	le16_add_cpu(&group->bg_bits, num_bits);
+	le16_add_cpu(&group->bg_free_bits_count, num_bits);
+
+	/*
+	 * check whether there are some new backup superblocks exist in
+	 * this group and update the group bitmap accordingly.
+	 */
+	if (OCFS2_HAS_COMPAT_FEATURE(osb->sb,
+				     OCFS2_FEATURE_COMPAT_BACKUP_SB)) {
+		backups = ocfs2_calc_new_backup_super(bm_inode,
+						     group,
+						     cl_cpg, 1);
+		le16_add_cpu(&group->bg_free_bits_count, -1 * backups);
+	}
+
+	ocfs2_journal_dirty(handle, group_bh);
+
+	/* update the inode accordingly. */
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(bm_inode), bm_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_rollback;
+	}
+
+	chain = le16_to_cpu(group->bg_chain);
+	cr = (&cl->cl_recs[chain]);
+	le32_add_cpu(&cr->c_total, num_bits);
+	le32_add_cpu(&cr->c_free, num_bits);
+	le32_add_cpu(&fe->id1.bitmap1.i_total, num_bits);
+	le32_add_cpu(&fe->i_clusters, new_clusters);
+
+	if (backups) {
+		le32_add_cpu(&cr->c_free, -1 * backups);
+		le32_add_cpu(&fe->id1.bitmap1.i_used, backups);
+	}
+
+	spin_lock(&OCFS2_I(bm_inode)->ip_lock);
+	OCFS2_I(bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
+	le64_add_cpu(&fe->i_size, (u64)new_clusters << osb->s_clustersize_bits);
+	spin_unlock(&OCFS2_I(bm_inode)->ip_lock);
+	i_size_write(bm_inode, le64_to_cpu(fe->i_size));
+
+	ocfs2_journal_dirty(handle, bm_bh);
+
+out_rollback:
+	if (ret < 0) {
+		ocfs2_calc_new_backup_super(bm_inode,
+					    group,
+					    cl_cpg, 0);
+		le16_add_cpu(&group->bg_free_bits_count, backups);
+		le16_add_cpu(&group->bg_bits, -1 * num_bits);
+		le16_add_cpu(&group->bg_free_bits_count, -1 * num_bits);
+	}
+out:
+	if (ret)
+		mlog_errno(ret);
+	return ret;
+}
+
+static int update_backups(struct inode * inode, u32 clusters, char *data)
+{
+	int i, ret = 0;
+	u32 cluster;
+	u64 blkno;
+	struct buffer_head *backup = NULL;
+	struct ocfs2_dinode *backup_di = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	/* calculate the real backups we need to update. */
+	for (i = 0; i < OCFS2_MAX_BACKUP_SUPERBLOCKS; i++) {
+		blkno = ocfs2_backup_super_blkno(inode->i_sb, i);
+		cluster = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
+		if (cluster > clusters)
+			break;
+
+		ret = ocfs2_read_blocks_sync(osb, blkno, 1, &backup);
+		if (ret < 0) {
+			mlog_errno(ret);
+			break;
+		}
+
+		memcpy(backup->b_data, data, inode->i_sb->s_blocksize);
+
+		backup_di = (struct ocfs2_dinode *)backup->b_data;
+		backup_di->i_blkno = cpu_to_le64(blkno);
+
+		ret = ocfs2_write_super_or_backup(osb, backup);
+		brelse(backup);
+		backup = NULL;
+		if (ret < 0) {
+			mlog_errno(ret);
+			break;
+		}
+	}
+
+	return ret;
+}
+
+static void ocfs2_update_super_and_backups(struct inode *inode,
+					   int new_clusters)
+{
+	int ret;
+	u32 clusters = 0;
+	struct buffer_head *super_bh = NULL;
+	struct ocfs2_dinode *super_di = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	/*
+	 * update the superblock last.
+	 * It doesn't matter if the write failed.
+	 */
+	ret = ocfs2_read_blocks_sync(osb, OCFS2_SUPER_BLOCK_BLKNO, 1,
+				     &super_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	super_di = (struct ocfs2_dinode *)super_bh->b_data;
+	le32_add_cpu(&super_di->i_clusters, new_clusters);
+	clusters = le32_to_cpu(super_di->i_clusters);
+
+	ret = ocfs2_write_super_or_backup(osb, super_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (OCFS2_HAS_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_COMPAT_BACKUP_SB))
+		ret = update_backups(inode, clusters, super_bh->b_data);
+
+out:
+	brelse(super_bh);
+	if (ret)
+		printk(KERN_WARNING "ocfs2: Failed to update super blocks on %s"
+			" during fs resize. This condition is not fatal,"
+			" but fsck.ocfs2 should be run to fix it\n",
+			osb->dev_str);
+	return;
+}
+
+/*
+ * Extend the filesystem to the new number of clusters specified.  This entry
+ * point is only used to extend the current filesystem to the end of the last
+ * existing group.
+ */
+int ocfs2_group_extend(struct inode * inode, int new_clusters)
+{
+	int ret;
+	handle_t *handle;
+	struct buffer_head *main_bm_bh = NULL;
+	struct buffer_head *group_bh = NULL;
+	struct inode *main_bm_inode = NULL;
+	struct ocfs2_dinode *fe = NULL;
+	struct ocfs2_group_desc *group = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	u16 cl_bpc;
+	u32 first_new_cluster;
+	u64 lgd_blkno;
+
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
+		return -EROFS;
+
+	if (new_clusters < 0)
+		return -EINVAL;
+	else if (new_clusters == 0)
+		return 0;
+
+	main_bm_inode = ocfs2_get_system_file_inode(osb,
+						    GLOBAL_BITMAP_SYSTEM_INODE,
+						    OCFS2_INVALID_SLOT);
+	if (!main_bm_inode) {
+		ret = -EINVAL;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mutex_lock(&main_bm_inode->i_mutex);
+
+	ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_mutex;
+	}
+
+	fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
+
+	/* main_bm_bh is validated by inode read inside ocfs2_inode_lock(),
+	 * so any corruption is a code bug. */
+	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
+
+	if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
+		ocfs2_group_bitmap_size(osb->sb, 0,
+					osb->s_feature_incompat) * 8) {
+		mlog(ML_ERROR, "The disk is too old and small. "
+		     "Force to do offline resize.");
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
+	first_new_cluster = le32_to_cpu(fe->i_clusters);
+	lgd_blkno = ocfs2_which_cluster_group(main_bm_inode,
+					      first_new_cluster - 1);
+
+	ret = ocfs2_read_group_descriptor(main_bm_inode, fe, lgd_blkno,
+					  &group_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+	group = (struct ocfs2_group_desc *)group_bh->b_data;
+
+	cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
+	if (le16_to_cpu(group->bg_bits) / cl_bpc + new_clusters >
+		le16_to_cpu(fe->id2.i_chain.cl_cpg)) {
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
+
+	trace_ocfs2_group_extend(
+	     (unsigned long long)le64_to_cpu(group->bg_blkno), new_clusters);
+
+	handle = ocfs2_start_trans(osb, OCFS2_GROUP_EXTEND_CREDITS);
+	if (IS_ERR(handle)) {
+		mlog_errno(PTR_ERR(handle));
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
+	/* update the last group descriptor and inode. */
+	ret = ocfs2_update_last_group_and_inode(handle, main_bm_inode,
+						main_bm_bh, group_bh,
+						first_new_cluster,
+						new_clusters);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ocfs2_update_super_and_backups(main_bm_inode, new_clusters);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out_unlock:
+	brelse(group_bh);
+	brelse(main_bm_bh);
+
+	ocfs2_inode_unlock(main_bm_inode, 1);
+
+out_mutex:
+	mutex_unlock(&main_bm_inode->i_mutex);
+	iput(main_bm_inode);
+
+out:
+	return ret;
+}
+
+static int ocfs2_check_new_group(struct inode *inode,
+				 struct ocfs2_dinode *di,
+				 struct ocfs2_new_group_input *input,
+				 struct buffer_head *group_bh)
+{
+	int ret;
+	struct ocfs2_group_desc *gd =
+		(struct ocfs2_group_desc *)group_bh->b_data;
+	u16 cl_bpc = le16_to_cpu(di->id2.i_chain.cl_bpc);
+
+	ret = ocfs2_check_group_descriptor(inode->i_sb, di, group_bh);
+	if (ret)
+		goto out;
+
+	ret = -EINVAL;
+	if (le16_to_cpu(gd->bg_chain) != input->chain)
+		mlog(ML_ERROR, "Group descriptor # %llu has bad chain %u "
+		     "while input has %u set.\n",
+		     (unsigned long long)le64_to_cpu(gd->bg_blkno),
+		     le16_to_cpu(gd->bg_chain), input->chain);
+	else if (le16_to_cpu(gd->bg_bits) != input->clusters * cl_bpc)
+		mlog(ML_ERROR, "Group descriptor # %llu has bit count %u but "
+		     "input has %u clusters set\n",
+		     (unsigned long long)le64_to_cpu(gd->bg_blkno),
+		     le16_to_cpu(gd->bg_bits), input->clusters);
+	else if (le16_to_cpu(gd->bg_free_bits_count) != input->frees * cl_bpc)
+		mlog(ML_ERROR, "Group descriptor # %llu has free bit count %u "
+		     "but it should have %u set\n",
+		     (unsigned long long)le64_to_cpu(gd->bg_blkno),
+		     le16_to_cpu(gd->bg_bits),
+		     input->frees * cl_bpc);
+	else
+		ret = 0;
+
+out:
+	return ret;
+}
+
+static int ocfs2_verify_group_and_input(struct inode *inode,
+					struct ocfs2_dinode *di,
+					struct ocfs2_new_group_input *input,
+					struct buffer_head *group_bh)
+{
+	u16 cl_count = le16_to_cpu(di->id2.i_chain.cl_count);
+	u16 cl_cpg = le16_to_cpu(di->id2.i_chain.cl_cpg);
+	u16 next_free = le16_to_cpu(di->id2.i_chain.cl_next_free_rec);
+	u32 cluster = ocfs2_blocks_to_clusters(inode->i_sb, input->group);
+	u32 total_clusters = le32_to_cpu(di->i_clusters);
+	int ret = -EINVAL;
+
+	if (cluster < total_clusters)
+		mlog(ML_ERROR, "add a group which is in the current volume.\n");
+	else if (input->chain >= cl_count)
+		mlog(ML_ERROR, "input chain exceeds the limit.\n");
+	else if (next_free != cl_count && next_free != input->chain)
+		mlog(ML_ERROR,
+		     "the add group should be in chain %u\n", next_free);
+	else if (total_clusters + input->clusters < total_clusters)
+		mlog(ML_ERROR, "add group's clusters overflow.\n");
+	else if (input->clusters > cl_cpg)
+		mlog(ML_ERROR, "the cluster exceeds the maximum of a group\n");
+	else if (input->frees > input->clusters)
+		mlog(ML_ERROR, "the free cluster exceeds the total clusters\n");
+	else if (total_clusters % cl_cpg != 0)
+		mlog(ML_ERROR,
+		     "the last group isn't full. Use group extend first.\n");
+	else if (input->group != ocfs2_which_cluster_group(inode, cluster))
+		mlog(ML_ERROR, "group blkno is invalid\n");
+	else if ((ret = ocfs2_check_new_group(inode, di, input, group_bh)))
+		mlog(ML_ERROR, "group descriptor check failed.\n");
+	else
+		ret = 0;
+
+	return ret;
+}
+
+/* Add a new group descriptor to global_bitmap. */
+int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input)
+{
+	int ret;
+	handle_t *handle;
+	struct buffer_head *main_bm_bh = NULL;
+	struct inode *main_bm_inode = NULL;
+	struct ocfs2_dinode *fe = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct buffer_head *group_bh = NULL;
+	struct ocfs2_group_desc *group = NULL;
+	struct ocfs2_chain_list *cl;
+	struct ocfs2_chain_rec *cr;
+	u16 cl_bpc;
+	u64 bg_ptr;
+
+	if (ocfs2_is_hard_readonly(osb) || ocfs2_is_soft_readonly(osb))
+		return -EROFS;
+
+	main_bm_inode = ocfs2_get_system_file_inode(osb,
+						    GLOBAL_BITMAP_SYSTEM_INODE,
+						    OCFS2_INVALID_SLOT);
+	if (!main_bm_inode) {
+		ret = -EINVAL;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	mutex_lock(&main_bm_inode->i_mutex);
+
+	ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_mutex;
+	}
+
+	fe = (struct ocfs2_dinode *)main_bm_bh->b_data;
+
+	if (le16_to_cpu(fe->id2.i_chain.cl_cpg) !=
+		ocfs2_group_bitmap_size(osb->sb, 0,
+					osb->s_feature_incompat) * 8) {
+		mlog(ML_ERROR, "The disk is too old and small."
+		     " Force to do offline resize.");
+		ret = -EINVAL;
+		goto out_unlock;
+	}
+
+	ret = ocfs2_read_blocks_sync(osb, input->group, 1, &group_bh);
+	if (ret < 0) {
+		mlog(ML_ERROR, "Can't read the group descriptor # %llu "
+		     "from the device.", (unsigned long long)input->group);
+		goto out_unlock;
+	}
+
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), group_bh);
+
+	ret = ocfs2_verify_group_and_input(main_bm_inode, fe, input, group_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_free_group_bh;
+	}
+
+	trace_ocfs2_group_add((unsigned long long)input->group,
+			       input->chain, input->clusters, input->frees);
+
+	handle = ocfs2_start_trans(osb, OCFS2_GROUP_ADD_CREDITS);
+	if (IS_ERR(handle)) {
+		mlog_errno(PTR_ERR(handle));
+		ret = -EINVAL;
+		goto out_free_group_bh;
+	}
+
+	cl_bpc = le16_to_cpu(fe->id2.i_chain.cl_bpc);
+	cl = &fe->id2.i_chain;
+	cr = &cl->cl_recs[input->chain];
+
+	ret = ocfs2_journal_access_gd(handle, INODE_CACHE(main_bm_inode),
+				      group_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	group = (struct ocfs2_group_desc *)group_bh->b_data;
+	bg_ptr = le64_to_cpu(group->bg_next_group);
+	group->bg_next_group = cr->c_blkno;
+	ocfs2_journal_dirty(handle, group_bh);
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(main_bm_inode),
+				      main_bm_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		group->bg_next_group = cpu_to_le64(bg_ptr);
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	if (input->chain == le16_to_cpu(cl->cl_next_free_rec)) {
+		le16_add_cpu(&cl->cl_next_free_rec, 1);
+		memset(cr, 0, sizeof(struct ocfs2_chain_rec));
+	}
+
+	cr->c_blkno = cpu_to_le64(input->group);
+	le32_add_cpu(&cr->c_total, input->clusters * cl_bpc);
+	le32_add_cpu(&cr->c_free, input->frees * cl_bpc);
+
+	le32_add_cpu(&fe->id1.bitmap1.i_total, input->clusters *cl_bpc);
+	le32_add_cpu(&fe->id1.bitmap1.i_used,
+		     (input->clusters - input->frees) * cl_bpc);
+	le32_add_cpu(&fe->i_clusters, input->clusters);
+
+	ocfs2_journal_dirty(handle, main_bm_bh);
+
+	spin_lock(&OCFS2_I(main_bm_inode)->ip_lock);
+	OCFS2_I(main_bm_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
+	le64_add_cpu(&fe->i_size, (u64)input->clusters << osb->s_clustersize_bits);
+	spin_unlock(&OCFS2_I(main_bm_inode)->ip_lock);
+	i_size_write(main_bm_inode, le64_to_cpu(fe->i_size));
+
+	ocfs2_update_super_and_backups(main_bm_inode, input->clusters);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out_free_group_bh:
+	brelse(group_bh);
+
+out_unlock:
+	brelse(main_bm_bh);
+
+	ocfs2_inode_unlock(main_bm_inode, 1);
+
+out_mutex:
+	mutex_unlock(&main_bm_inode->i_mutex);
+	iput(main_bm_inode);
+
+out:
+	return ret;
+}
diff --git a/fs/ocfs2/resize.h b/fs/ocfs2/resize.h
new file mode 100644
index 000000000..f38841abf
--- /dev/null
+++ b/fs/ocfs2/resize.h
@@ -0,0 +1,32 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * resize.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_RESIZE_H
+#define OCFS2_RESIZE_H
+
+int ocfs2_group_extend(struct inode * inode, int new_clusters);
+int ocfs2_group_add(struct inode *inode, struct ocfs2_new_group_input *input);
+
+#endif /* OCFS2_RESIZE_H */
diff --git a/fs/ocfs2/slot_map.c b/fs/ocfs2/slot_map.c
new file mode 100644
index 000000000..e78a203d4
--- /dev/null
+++ b/fs/ocfs2/slot_map.c
@@ -0,0 +1,538 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * slot_map.c
+ *
+ *
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "dlmglue.h"
+#include "extent_map.h"
+#include "heartbeat.h"
+#include "inode.h"
+#include "slot_map.h"
+#include "super.h"
+#include "sysfile.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+
+struct ocfs2_slot {
+	int sl_valid;
+	unsigned int sl_node_num;
+};
+
+struct ocfs2_slot_info {
+	int si_extended;
+	int si_slots_per_block;
+	struct inode *si_inode;
+	unsigned int si_blocks;
+	struct buffer_head **si_bh;
+	unsigned int si_num_slots;
+	struct ocfs2_slot *si_slots;
+};
+
+
+static int __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si,
+				    unsigned int node_num);
+
+static void ocfs2_invalidate_slot(struct ocfs2_slot_info *si,
+				  int slot_num)
+{
+	BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots));
+	si->si_slots[slot_num].sl_valid = 0;
+}
+
+static void ocfs2_set_slot(struct ocfs2_slot_info *si,
+			   int slot_num, unsigned int node_num)
+{
+	BUG_ON((slot_num < 0) || (slot_num >= si->si_num_slots));
+
+	si->si_slots[slot_num].sl_valid = 1;
+	si->si_slots[slot_num].sl_node_num = node_num;
+}
+
+/* This version is for the extended slot map */
+static void ocfs2_update_slot_info_extended(struct ocfs2_slot_info *si)
+{
+	int b, i, slotno;
+	struct ocfs2_slot_map_extended *se;
+
+	slotno = 0;
+	for (b = 0; b < si->si_blocks; b++) {
+		se = (struct ocfs2_slot_map_extended *)si->si_bh[b]->b_data;
+		for (i = 0;
+		     (i < si->si_slots_per_block) &&
+		     (slotno < si->si_num_slots);
+		     i++, slotno++) {
+			if (se->se_slots[i].es_valid)
+				ocfs2_set_slot(si, slotno,
+					       le32_to_cpu(se->se_slots[i].es_node_num));
+			else
+				ocfs2_invalidate_slot(si, slotno);
+		}
+	}
+}
+
+/*
+ * Post the slot information on disk into our slot_info struct.
+ * Must be protected by osb_lock.
+ */
+static void ocfs2_update_slot_info_old(struct ocfs2_slot_info *si)
+{
+	int i;
+	struct ocfs2_slot_map *sm;
+
+	sm = (struct ocfs2_slot_map *)si->si_bh[0]->b_data;
+
+	for (i = 0; i < si->si_num_slots; i++) {
+		if (le16_to_cpu(sm->sm_slots[i]) == (u16)OCFS2_INVALID_SLOT)
+			ocfs2_invalidate_slot(si, i);
+		else
+			ocfs2_set_slot(si, i, le16_to_cpu(sm->sm_slots[i]));
+	}
+}
+
+static void ocfs2_update_slot_info(struct ocfs2_slot_info *si)
+{
+	/*
+	 * The slot data will have been refreshed when ocfs2_super_lock
+	 * was taken.
+	 */
+	if (si->si_extended)
+		ocfs2_update_slot_info_extended(si);
+	else
+		ocfs2_update_slot_info_old(si);
+}
+
+int ocfs2_refresh_slot_info(struct ocfs2_super *osb)
+{
+	int ret;
+	struct ocfs2_slot_info *si = osb->slot_info;
+
+	if (si == NULL)
+		return 0;
+
+	BUG_ON(si->si_blocks == 0);
+	BUG_ON(si->si_bh == NULL);
+
+	trace_ocfs2_refresh_slot_info(si->si_blocks);
+
+	/*
+	 * We pass -1 as blocknr because we expect all of si->si_bh to
+	 * be !NULL.  Thus, ocfs2_read_blocks() will ignore blocknr.  If
+	 * this is not true, the read of -1 (UINT64_MAX) will fail.
+	 */
+	ret = ocfs2_read_blocks(INODE_CACHE(si->si_inode), -1, si->si_blocks,
+				si->si_bh, OCFS2_BH_IGNORE_CACHE, NULL);
+	if (ret == 0) {
+		spin_lock(&osb->osb_lock);
+		ocfs2_update_slot_info(si);
+		spin_unlock(&osb->osb_lock);
+	}
+
+	return ret;
+}
+
+/* post the our slot info stuff into it's destination bh and write it
+ * out. */
+static void ocfs2_update_disk_slot_extended(struct ocfs2_slot_info *si,
+					    int slot_num,
+					    struct buffer_head **bh)
+{
+	int blkind = slot_num / si->si_slots_per_block;
+	int slotno = slot_num % si->si_slots_per_block;
+	struct ocfs2_slot_map_extended *se;
+
+	BUG_ON(blkind >= si->si_blocks);
+
+	se = (struct ocfs2_slot_map_extended *)si->si_bh[blkind]->b_data;
+	se->se_slots[slotno].es_valid = si->si_slots[slot_num].sl_valid;
+	if (si->si_slots[slot_num].sl_valid)
+		se->se_slots[slotno].es_node_num =
+			cpu_to_le32(si->si_slots[slot_num].sl_node_num);
+	*bh = si->si_bh[blkind];
+}
+
+static void ocfs2_update_disk_slot_old(struct ocfs2_slot_info *si,
+				       int slot_num,
+				       struct buffer_head **bh)
+{
+	int i;
+	struct ocfs2_slot_map *sm;
+
+	sm = (struct ocfs2_slot_map *)si->si_bh[0]->b_data;
+	for (i = 0; i < si->si_num_slots; i++) {
+		if (si->si_slots[i].sl_valid)
+			sm->sm_slots[i] =
+				cpu_to_le16(si->si_slots[i].sl_node_num);
+		else
+			sm->sm_slots[i] = cpu_to_le16(OCFS2_INVALID_SLOT);
+	}
+	*bh = si->si_bh[0];
+}
+
+static int ocfs2_update_disk_slot(struct ocfs2_super *osb,
+				  struct ocfs2_slot_info *si,
+				  int slot_num)
+{
+	int status;
+	struct buffer_head *bh;
+
+	spin_lock(&osb->osb_lock);
+	if (si->si_extended)
+		ocfs2_update_disk_slot_extended(si, slot_num, &bh);
+	else
+		ocfs2_update_disk_slot_old(si, slot_num, &bh);
+	spin_unlock(&osb->osb_lock);
+
+	status = ocfs2_write_block(osb, bh, INODE_CACHE(si->si_inode));
+	if (status < 0)
+		mlog_errno(status);
+
+	return status;
+}
+
+/*
+ * Calculate how many bytes are needed by the slot map.  Returns
+ * an error if the slot map file is too small.
+ */
+static int ocfs2_slot_map_physical_size(struct ocfs2_super *osb,
+					struct inode *inode,
+					unsigned long long *bytes)
+{
+	unsigned long long bytes_needed;
+
+	if (ocfs2_uses_extended_slot_map(osb)) {
+		bytes_needed = osb->max_slots *
+			sizeof(struct ocfs2_extended_slot);
+	} else {
+		bytes_needed = osb->max_slots * sizeof(__le16);
+	}
+	if (bytes_needed > i_size_read(inode)) {
+		mlog(ML_ERROR,
+		     "Slot map file is too small!  (size %llu, needed %llu)\n",
+		     i_size_read(inode), bytes_needed);
+		return -ENOSPC;
+	}
+
+	*bytes = bytes_needed;
+	return 0;
+}
+
+/* try to find global node in the slot info. Returns -ENOENT
+ * if nothing is found. */
+static int __ocfs2_node_num_to_slot(struct ocfs2_slot_info *si,
+				    unsigned int node_num)
+{
+	int i, ret = -ENOENT;
+
+	for(i = 0; i < si->si_num_slots; i++) {
+		if (si->si_slots[i].sl_valid &&
+		    (node_num == si->si_slots[i].sl_node_num)) {
+			ret = i;
+			break;
+		}
+	}
+
+	return ret;
+}
+
+static int __ocfs2_find_empty_slot(struct ocfs2_slot_info *si,
+				   int preferred)
+{
+	int i, ret = -ENOSPC;
+
+	if ((preferred >= 0) && (preferred < si->si_num_slots)) {
+		if (!si->si_slots[preferred].sl_valid) {
+			ret = preferred;
+			goto out;
+		}
+	}
+
+	for(i = 0; i < si->si_num_slots; i++) {
+		if (!si->si_slots[i].sl_valid) {
+			ret = i;
+			break;
+		}
+	}
+out:
+	return ret;
+}
+
+int ocfs2_node_num_to_slot(struct ocfs2_super *osb, unsigned int node_num)
+{
+	int slot;
+	struct ocfs2_slot_info *si = osb->slot_info;
+
+	spin_lock(&osb->osb_lock);
+	slot = __ocfs2_node_num_to_slot(si, node_num);
+	spin_unlock(&osb->osb_lock);
+
+	return slot;
+}
+
+int ocfs2_slot_to_node_num_locked(struct ocfs2_super *osb, int slot_num,
+				  unsigned int *node_num)
+{
+	struct ocfs2_slot_info *si = osb->slot_info;
+
+	assert_spin_locked(&osb->osb_lock);
+
+	BUG_ON(slot_num < 0);
+	BUG_ON(slot_num >= osb->max_slots);
+
+	if (!si->si_slots[slot_num].sl_valid)
+		return -ENOENT;
+
+	*node_num = si->si_slots[slot_num].sl_node_num;
+	return 0;
+}
+
+static void __ocfs2_free_slot_info(struct ocfs2_slot_info *si)
+{
+	unsigned int i;
+
+	if (si == NULL)
+		return;
+
+	if (si->si_inode)
+		iput(si->si_inode);
+	if (si->si_bh) {
+		for (i = 0; i < si->si_blocks; i++) {
+			if (si->si_bh[i]) {
+				brelse(si->si_bh[i]);
+				si->si_bh[i] = NULL;
+			}
+		}
+		kfree(si->si_bh);
+	}
+
+	kfree(si);
+}
+
+int ocfs2_clear_slot(struct ocfs2_super *osb, int slot_num)
+{
+	struct ocfs2_slot_info *si = osb->slot_info;
+
+	if (si == NULL)
+		return 0;
+
+	spin_lock(&osb->osb_lock);
+	ocfs2_invalidate_slot(si, slot_num);
+	spin_unlock(&osb->osb_lock);
+
+	return ocfs2_update_disk_slot(osb, osb->slot_info, slot_num);
+}
+
+static int ocfs2_map_slot_buffers(struct ocfs2_super *osb,
+				  struct ocfs2_slot_info *si)
+{
+	int status = 0;
+	u64 blkno;
+	unsigned long long blocks, bytes = 0;
+	unsigned int i;
+	struct buffer_head *bh;
+
+	status = ocfs2_slot_map_physical_size(osb, si->si_inode, &bytes);
+	if (status)
+		goto bail;
+
+	blocks = ocfs2_blocks_for_bytes(si->si_inode->i_sb, bytes);
+	BUG_ON(blocks > UINT_MAX);
+	si->si_blocks = blocks;
+	if (!si->si_blocks)
+		goto bail;
+
+	if (si->si_extended)
+		si->si_slots_per_block =
+			(osb->sb->s_blocksize /
+			 sizeof(struct ocfs2_extended_slot));
+	else
+		si->si_slots_per_block = osb->sb->s_blocksize / sizeof(__le16);
+
+	/* The size checks above should ensure this */
+	BUG_ON((osb->max_slots / si->si_slots_per_block) > blocks);
+
+	trace_ocfs2_map_slot_buffers(bytes, si->si_blocks);
+
+	si->si_bh = kcalloc(si->si_blocks, sizeof(struct buffer_head *),
+			    GFP_KERNEL);
+	if (!si->si_bh) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	for (i = 0; i < si->si_blocks; i++) {
+		status = ocfs2_extent_map_get_blocks(si->si_inode, i,
+						     &blkno, NULL, NULL);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		trace_ocfs2_map_slot_buffers_block((unsigned long long)blkno, i);
+
+		bh = NULL;  /* Acquire a fresh bh */
+		status = ocfs2_read_blocks(INODE_CACHE(si->si_inode), blkno,
+					   1, &bh, OCFS2_BH_IGNORE_CACHE, NULL);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+
+		si->si_bh[i] = bh;
+	}
+
+bail:
+	return status;
+}
+
+int ocfs2_init_slot_info(struct ocfs2_super *osb)
+{
+	int status;
+	struct inode *inode = NULL;
+	struct ocfs2_slot_info *si;
+
+	si = kzalloc(sizeof(struct ocfs2_slot_info) +
+		     (sizeof(struct ocfs2_slot) * osb->max_slots),
+		     GFP_KERNEL);
+	if (!si) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		return status;
+	}
+
+	si->si_extended = ocfs2_uses_extended_slot_map(osb);
+	si->si_num_slots = osb->max_slots;
+	si->si_slots = (struct ocfs2_slot *)((char *)si +
+					     sizeof(struct ocfs2_slot_info));
+
+	inode = ocfs2_get_system_file_inode(osb, SLOT_MAP_SYSTEM_INODE,
+					    OCFS2_INVALID_SLOT);
+	if (!inode) {
+		status = -EINVAL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	si->si_inode = inode;
+	status = ocfs2_map_slot_buffers(osb, si);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	osb->slot_info = (struct ocfs2_slot_info *)si;
+bail:
+	if (status < 0)
+		__ocfs2_free_slot_info(si);
+
+	return status;
+}
+
+void ocfs2_free_slot_info(struct ocfs2_super *osb)
+{
+	struct ocfs2_slot_info *si = osb->slot_info;
+
+	osb->slot_info = NULL;
+	__ocfs2_free_slot_info(si);
+}
+
+int ocfs2_find_slot(struct ocfs2_super *osb)
+{
+	int status;
+	int slot;
+	struct ocfs2_slot_info *si;
+
+	si = osb->slot_info;
+
+	spin_lock(&osb->osb_lock);
+	ocfs2_update_slot_info(si);
+
+	/* search for ourselves first and take the slot if it already
+	 * exists. Perhaps we need to mark this in a variable for our
+	 * own journal recovery? Possibly not, though we certainly
+	 * need to warn to the user */
+	slot = __ocfs2_node_num_to_slot(si, osb->node_num);
+	if (slot < 0) {
+		/* if no slot yet, then just take 1st available
+		 * one. */
+		slot = __ocfs2_find_empty_slot(si, osb->preferred_slot);
+		if (slot < 0) {
+			spin_unlock(&osb->osb_lock);
+			mlog(ML_ERROR, "no free slots available!\n");
+			status = -EINVAL;
+			goto bail;
+		}
+	} else
+		printk(KERN_INFO "ocfs2: Slot %d on device (%s) was already "
+		       "allocated to this node!\n", slot, osb->dev_str);
+
+	ocfs2_set_slot(si, slot, osb->node_num);
+	osb->slot_num = slot;
+	spin_unlock(&osb->osb_lock);
+
+	trace_ocfs2_find_slot(osb->slot_num);
+
+	status = ocfs2_update_disk_slot(osb, si, osb->slot_num);
+	if (status < 0)
+		mlog_errno(status);
+
+bail:
+	return status;
+}
+
+void ocfs2_put_slot(struct ocfs2_super *osb)
+{
+	int status, slot_num;
+	struct ocfs2_slot_info *si = osb->slot_info;
+
+	if (!si)
+		return;
+
+	spin_lock(&osb->osb_lock);
+	ocfs2_update_slot_info(si);
+
+	slot_num = osb->slot_num;
+	ocfs2_invalidate_slot(si, osb->slot_num);
+	osb->slot_num = OCFS2_INVALID_SLOT;
+	spin_unlock(&osb->osb_lock);
+
+	status = ocfs2_update_disk_slot(osb, si, slot_num);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+bail:
+	ocfs2_free_slot_info(osb);
+}
+
diff --git a/fs/ocfs2/slot_map.h b/fs/ocfs2/slot_map.h
new file mode 100644
index 000000000..601c95fd7
--- /dev/null
+++ b/fs/ocfs2/slot_map.h
@@ -0,0 +1,44 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * slotmap.h
+ *
+ * description here
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+
+#ifndef SLOTMAP_H
+#define SLOTMAP_H
+
+int ocfs2_init_slot_info(struct ocfs2_super *osb);
+void ocfs2_free_slot_info(struct ocfs2_super *osb);
+
+int ocfs2_find_slot(struct ocfs2_super *osb);
+void ocfs2_put_slot(struct ocfs2_super *osb);
+
+int ocfs2_refresh_slot_info(struct ocfs2_super *osb);
+
+int ocfs2_node_num_to_slot(struct ocfs2_super *osb, unsigned int node_num);
+int ocfs2_slot_to_node_num_locked(struct ocfs2_super *osb, int slot_num,
+				  unsigned int *node_num);
+
+int ocfs2_clear_slot(struct ocfs2_super *osb, int slot_num);
+
+#endif
diff --git a/fs/ocfs2/stack_o2cb.c b/fs/ocfs2/stack_o2cb.c
new file mode 100644
index 000000000..220cae7bb
--- /dev/null
+++ b/fs/ocfs2/stack_o2cb.c
@@ -0,0 +1,449 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * stack_o2cb.c
+ *
+ * Code which interfaces ocfs2 with the o2cb stack.
+ *
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation, version 2.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/kernel.h>
+#include <linux/crc32.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+
+/* Needed for AOP_TRUNCATED_PAGE in mlog_errno() */
+#include <linux/fs.h>
+
+#include "cluster/masklog.h"
+#include "cluster/nodemanager.h"
+#include "cluster/heartbeat.h"
+#include "cluster/tcp.h"
+
+#include "stackglue.h"
+
+struct o2dlm_private {
+	struct dlm_eviction_cb op_eviction_cb;
+};
+
+static struct ocfs2_stack_plugin o2cb_stack;
+
+/* These should be identical */
+#if (DLM_LOCK_IV != LKM_IVMODE)
+# error Lock modes do not match
+#endif
+#if (DLM_LOCK_NL != LKM_NLMODE)
+# error Lock modes do not match
+#endif
+#if (DLM_LOCK_CR != LKM_CRMODE)
+# error Lock modes do not match
+#endif
+#if (DLM_LOCK_CW != LKM_CWMODE)
+# error Lock modes do not match
+#endif
+#if (DLM_LOCK_PR != LKM_PRMODE)
+# error Lock modes do not match
+#endif
+#if (DLM_LOCK_PW != LKM_PWMODE)
+# error Lock modes do not match
+#endif
+#if (DLM_LOCK_EX != LKM_EXMODE)
+# error Lock modes do not match
+#endif
+static inline int mode_to_o2dlm(int mode)
+{
+	BUG_ON(mode > LKM_MAXMODE);
+
+	return mode;
+}
+
+#define map_flag(_generic, _o2dlm)		\
+	if (flags & (_generic)) {		\
+		flags &= ~(_generic);		\
+		o2dlm_flags |= (_o2dlm);	\
+	}
+static int flags_to_o2dlm(u32 flags)
+{
+	int o2dlm_flags = 0;
+
+	map_flag(DLM_LKF_NOQUEUE, LKM_NOQUEUE);
+	map_flag(DLM_LKF_CANCEL, LKM_CANCEL);
+	map_flag(DLM_LKF_CONVERT, LKM_CONVERT);
+	map_flag(DLM_LKF_VALBLK, LKM_VALBLK);
+	map_flag(DLM_LKF_IVVALBLK, LKM_INVVALBLK);
+	map_flag(DLM_LKF_ORPHAN, LKM_ORPHAN);
+	map_flag(DLM_LKF_FORCEUNLOCK, LKM_FORCE);
+	map_flag(DLM_LKF_TIMEOUT, LKM_TIMEOUT);
+	map_flag(DLM_LKF_LOCAL, LKM_LOCAL);
+
+	/* map_flag() should have cleared every flag passed in */
+	BUG_ON(flags != 0);
+
+	return o2dlm_flags;
+}
+#undef map_flag
+
+/*
+ * Map an o2dlm status to standard errno values.
+ *
+ * o2dlm only uses a handful of these, and returns even fewer to the
+ * caller. Still, we try to assign sane values to each error.
+ *
+ * The following value pairs have special meanings to dlmglue, thus
+ * the right hand side needs to stay unique - never duplicate the
+ * mapping elsewhere in the table!
+ *
+ * DLM_NORMAL:		0
+ * DLM_NOTQUEUED:	-EAGAIN
+ * DLM_CANCELGRANT:	-EBUSY
+ * DLM_CANCEL:		-DLM_ECANCEL
+ */
+/* Keep in sync with dlmapi.h */
+static int status_map[] = {
+	[DLM_NORMAL]			= 0,		/* Success */
+	[DLM_GRANTED]			= -EINVAL,
+	[DLM_DENIED]			= -EACCES,
+	[DLM_DENIED_NOLOCKS]		= -EACCES,
+	[DLM_WORKING]			= -EACCES,
+	[DLM_BLOCKED]			= -EINVAL,
+	[DLM_BLOCKED_ORPHAN]		= -EINVAL,
+	[DLM_DENIED_GRACE_PERIOD]	= -EACCES,
+	[DLM_SYSERR]			= -ENOMEM,	/* It is what it is */
+	[DLM_NOSUPPORT]			= -EPROTO,
+	[DLM_CANCELGRANT]		= -EBUSY,	/* Cancel after grant */
+	[DLM_IVLOCKID]			= -EINVAL,
+	[DLM_SYNC]			= -EINVAL,
+	[DLM_BADTYPE]			= -EINVAL,
+	[DLM_BADRESOURCE]		= -EINVAL,
+	[DLM_MAXHANDLES]		= -ENOMEM,
+	[DLM_NOCLINFO]			= -EINVAL,
+	[DLM_NOLOCKMGR]			= -EINVAL,
+	[DLM_NOPURGED]			= -EINVAL,
+	[DLM_BADARGS]			= -EINVAL,
+	[DLM_VOID]			= -EINVAL,
+	[DLM_NOTQUEUED]			= -EAGAIN,	/* Trylock failed */
+	[DLM_IVBUFLEN]			= -EINVAL,
+	[DLM_CVTUNGRANT]		= -EPERM,
+	[DLM_BADPARAM]			= -EINVAL,
+	[DLM_VALNOTVALID]		= -EINVAL,
+	[DLM_REJECTED]			= -EPERM,
+	[DLM_ABORT]			= -EINVAL,
+	[DLM_CANCEL]			= -DLM_ECANCEL,	/* Successful cancel */
+	[DLM_IVRESHANDLE]		= -EINVAL,
+	[DLM_DEADLOCK]			= -EDEADLK,
+	[DLM_DENIED_NOASTS]		= -EINVAL,
+	[DLM_FORWARD]			= -EINVAL,
+	[DLM_TIMEOUT]			= -ETIMEDOUT,
+	[DLM_IVGROUPID]			= -EINVAL,
+	[DLM_VERS_CONFLICT]		= -EOPNOTSUPP,
+	[DLM_BAD_DEVICE_PATH]		= -ENOENT,
+	[DLM_NO_DEVICE_PERMISSION]	= -EPERM,
+	[DLM_NO_CONTROL_DEVICE]		= -ENOENT,
+	[DLM_RECOVERING]		= -ENOTCONN,
+	[DLM_MIGRATING]			= -ERESTART,
+	[DLM_MAXSTATS]			= -EINVAL,
+};
+
+static int dlm_status_to_errno(enum dlm_status status)
+{
+	BUG_ON(status < 0 || status >= ARRAY_SIZE(status_map));
+
+	return status_map[status];
+}
+
+static void o2dlm_lock_ast_wrapper(void *astarg)
+{
+	struct ocfs2_dlm_lksb *lksb = astarg;
+
+	lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
+}
+
+static void o2dlm_blocking_ast_wrapper(void *astarg, int level)
+{
+	struct ocfs2_dlm_lksb *lksb = astarg;
+
+	lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
+}
+
+static void o2dlm_unlock_ast_wrapper(void *astarg, enum dlm_status status)
+{
+	struct ocfs2_dlm_lksb *lksb = astarg;
+	int error = dlm_status_to_errno(status);
+
+	/*
+	 * In o2dlm, you can get both the lock_ast() for the lock being
+	 * granted and the unlock_ast() for the CANCEL failing.  A
+	 * successful cancel sends DLM_NORMAL here.  If the
+	 * lock grant happened before the cancel arrived, you get
+	 * DLM_CANCELGRANT.
+	 *
+	 * There's no need for the double-ast.  If we see DLM_CANCELGRANT,
+	 * we just ignore it.  We expect the lock_ast() to handle the
+	 * granted lock.
+	 */
+	if (status == DLM_CANCELGRANT)
+		return;
+
+	lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, error);
+}
+
+static int o2cb_dlm_lock(struct ocfs2_cluster_connection *conn,
+			 int mode,
+			 struct ocfs2_dlm_lksb *lksb,
+			 u32 flags,
+			 void *name,
+			 unsigned int namelen)
+{
+	enum dlm_status status;
+	int o2dlm_mode = mode_to_o2dlm(mode);
+	int o2dlm_flags = flags_to_o2dlm(flags);
+	int ret;
+
+	status = dlmlock(conn->cc_lockspace, o2dlm_mode, &lksb->lksb_o2dlm,
+			 o2dlm_flags, name, namelen,
+			 o2dlm_lock_ast_wrapper, lksb,
+			 o2dlm_blocking_ast_wrapper);
+	ret = dlm_status_to_errno(status);
+	return ret;
+}
+
+static int o2cb_dlm_unlock(struct ocfs2_cluster_connection *conn,
+			   struct ocfs2_dlm_lksb *lksb,
+			   u32 flags)
+{
+	enum dlm_status status;
+	int o2dlm_flags = flags_to_o2dlm(flags);
+	int ret;
+
+	status = dlmunlock(conn->cc_lockspace, &lksb->lksb_o2dlm,
+			   o2dlm_flags, o2dlm_unlock_ast_wrapper, lksb);
+	ret = dlm_status_to_errno(status);
+	return ret;
+}
+
+static int o2cb_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
+{
+	return dlm_status_to_errno(lksb->lksb_o2dlm.status);
+}
+
+/*
+ * o2dlm aways has a "valid" LVB. If the dlm loses track of the LVB
+ * contents, it will zero out the LVB.  Thus the caller can always trust
+ * the contents.
+ */
+static int o2cb_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
+{
+	return 1;
+}
+
+static void *o2cb_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
+{
+	return (void *)(lksb->lksb_o2dlm.lvb);
+}
+
+static void o2cb_dump_lksb(struct ocfs2_dlm_lksb *lksb)
+{
+	dlm_print_one_lock(lksb->lksb_o2dlm.lockid);
+}
+
+/*
+ * Check if this node is heartbeating and is connected to all other
+ * heartbeating nodes.
+ */
+static int o2cb_cluster_check(void)
+{
+	u8 node_num;
+	int i;
+	unsigned long hbmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+	unsigned long netmap[BITS_TO_LONGS(O2NM_MAX_NODES)];
+
+	node_num = o2nm_this_node();
+	if (node_num == O2NM_MAX_NODES) {
+		printk(KERN_ERR "o2cb: This node has not been configured.\n");
+		return -EINVAL;
+	}
+
+	/*
+	 * o2dlm expects o2net sockets to be created. If not, then
+	 * dlm_join_domain() fails with a stack of errors which are both cryptic
+	 * and incomplete. The idea here is to detect upfront whether we have
+	 * managed to connect to all nodes or not. If not, then list the nodes
+	 * to allow the user to check the configuration (incorrect IP, firewall,
+	 * etc.) Yes, this is racy. But its not the end of the world.
+	 */
+#define	O2CB_MAP_STABILIZE_COUNT	60
+	for (i = 0; i < O2CB_MAP_STABILIZE_COUNT; ++i) {
+		o2hb_fill_node_map(hbmap, sizeof(hbmap));
+		if (!test_bit(node_num, hbmap)) {
+			printk(KERN_ERR "o2cb: %s heartbeat has not been "
+			       "started.\n", (o2hb_global_heartbeat_active() ?
+					      "Global" : "Local"));
+			return -EINVAL;
+		}
+		o2net_fill_node_map(netmap, sizeof(netmap));
+		/* Force set the current node to allow easy compare */
+		set_bit(node_num, netmap);
+		if (!memcmp(hbmap, netmap, sizeof(hbmap)))
+			return 0;
+		if (i < O2CB_MAP_STABILIZE_COUNT - 1)
+			msleep(1000);
+	}
+
+	printk(KERN_ERR "o2cb: This node could not connect to nodes:");
+	i = -1;
+	while ((i = find_next_bit(hbmap, O2NM_MAX_NODES,
+				  i + 1)) < O2NM_MAX_NODES) {
+		if (!test_bit(i, netmap))
+			printk(" %u", i);
+	}
+	printk(".\n");
+
+	return -ENOTCONN;
+}
+
+/*
+ * Called from the dlm when it's about to evict a node. This is how the
+ * classic stack signals node death.
+ */
+static void o2dlm_eviction_cb(int node_num, void *data)
+{
+	struct ocfs2_cluster_connection *conn = data;
+
+	printk(KERN_NOTICE "o2cb: o2dlm has evicted node %d from domain %.*s\n",
+	       node_num, conn->cc_namelen, conn->cc_name);
+
+	conn->cc_recovery_handler(node_num, conn->cc_recovery_data);
+}
+
+static int o2cb_cluster_connect(struct ocfs2_cluster_connection *conn)
+{
+	int rc = 0;
+	u32 dlm_key;
+	struct dlm_ctxt *dlm;
+	struct o2dlm_private *priv;
+	struct dlm_protocol_version fs_version;
+
+	BUG_ON(conn == NULL);
+	BUG_ON(conn->cc_proto == NULL);
+
+	/* Ensure cluster stack is up and all nodes are connected */
+	rc = o2cb_cluster_check();
+	if (rc) {
+		printk(KERN_ERR "o2cb: Cluster check failed. Fix errors "
+		       "before retrying.\n");
+		goto out;
+	}
+
+	priv = kzalloc(sizeof(struct o2dlm_private), GFP_KERNEL);
+	if (!priv) {
+		rc = -ENOMEM;
+		goto out_free;
+	}
+
+	/* This just fills the structure in.  It is safe to pass conn. */
+	dlm_setup_eviction_cb(&priv->op_eviction_cb, o2dlm_eviction_cb,
+			      conn);
+
+	conn->cc_private = priv;
+
+	/* used by the dlm code to make message headers unique, each
+	 * node in this domain must agree on this. */
+	dlm_key = crc32_le(0, conn->cc_name, conn->cc_namelen);
+	fs_version.pv_major = conn->cc_version.pv_major;
+	fs_version.pv_minor = conn->cc_version.pv_minor;
+
+	dlm = dlm_register_domain(conn->cc_name, dlm_key, &fs_version);
+	if (IS_ERR(dlm)) {
+		rc = PTR_ERR(dlm);
+		mlog_errno(rc);
+		goto out_free;
+	}
+
+	conn->cc_version.pv_major = fs_version.pv_major;
+	conn->cc_version.pv_minor = fs_version.pv_minor;
+	conn->cc_lockspace = dlm;
+
+	dlm_register_eviction_cb(dlm, &priv->op_eviction_cb);
+
+out_free:
+	if (rc)
+		kfree(conn->cc_private);
+
+out:
+	return rc;
+}
+
+static int o2cb_cluster_disconnect(struct ocfs2_cluster_connection *conn)
+{
+	struct dlm_ctxt *dlm = conn->cc_lockspace;
+	struct o2dlm_private *priv = conn->cc_private;
+
+	dlm_unregister_eviction_cb(&priv->op_eviction_cb);
+	conn->cc_private = NULL;
+	kfree(priv);
+
+	dlm_unregister_domain(dlm);
+	conn->cc_lockspace = NULL;
+
+	return 0;
+}
+
+static int o2cb_cluster_this_node(struct ocfs2_cluster_connection *conn,
+				  unsigned int *node)
+{
+	int node_num;
+
+	node_num = o2nm_this_node();
+	if (node_num == O2NM_INVALID_NODE_NUM)
+		return -ENOENT;
+
+	if (node_num >= O2NM_MAX_NODES)
+		return -EOVERFLOW;
+
+	*node = node_num;
+	return 0;
+}
+
+static struct ocfs2_stack_operations o2cb_stack_ops = {
+	.connect	= o2cb_cluster_connect,
+	.disconnect	= o2cb_cluster_disconnect,
+	.this_node	= o2cb_cluster_this_node,
+	.dlm_lock	= o2cb_dlm_lock,
+	.dlm_unlock	= o2cb_dlm_unlock,
+	.lock_status	= o2cb_dlm_lock_status,
+	.lvb_valid	= o2cb_dlm_lvb_valid,
+	.lock_lvb	= o2cb_dlm_lvb,
+	.dump_lksb	= o2cb_dump_lksb,
+};
+
+static struct ocfs2_stack_plugin o2cb_stack = {
+	.sp_name	= "o2cb",
+	.sp_ops		= &o2cb_stack_ops,
+	.sp_owner	= THIS_MODULE,
+};
+
+static int __init o2cb_stack_init(void)
+{
+	return ocfs2_stack_glue_register(&o2cb_stack);
+}
+
+static void __exit o2cb_stack_exit(void)
+{
+	ocfs2_stack_glue_unregister(&o2cb_stack);
+}
+
+MODULE_AUTHOR("Oracle");
+MODULE_DESCRIPTION("ocfs2 driver for the classic o2cb stack");
+MODULE_LICENSE("GPL");
+module_init(o2cb_stack_init);
+module_exit(o2cb_stack_exit);
diff --git a/fs/ocfs2/stack_user.c b/fs/ocfs2/stack_user.c
new file mode 100644
index 000000000..2768eb1da
--- /dev/null
+++ b/fs/ocfs2/stack_user.c
@@ -0,0 +1,1134 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * stack_user.c
+ *
+ * Code which interfaces ocfs2 with fs/dlm and a userspace stack.
+ *
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation, version 2.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/miscdevice.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/reboot.h>
+#include <linux/sched.h>
+#include <asm/uaccess.h>
+
+#include "stackglue.h"
+
+#include <linux/dlm_plock.h>
+
+/*
+ * The control protocol starts with a handshake.  Until the handshake
+ * is complete, the control device will fail all write(2)s.
+ *
+ * The handshake is simple.  First, the client reads until EOF.  Each line
+ * of output is a supported protocol tag.  All protocol tags are a single
+ * character followed by a two hex digit version number.  Currently the
+ * only things supported is T01, for "Text-base version 0x01".  Next, the
+ * client writes the version they would like to use, including the newline.
+ * Thus, the protocol tag is 'T01\n'.  If the version tag written is
+ * unknown, -EINVAL is returned.  Once the negotiation is complete, the
+ * client can start sending messages.
+ *
+ * The T01 protocol has three messages.  First is the "SETN" message.
+ * It has the following syntax:
+ *
+ *  SETN<space><8-char-hex-nodenum><newline>
+ *
+ * This is 14 characters.
+ *
+ * The "SETN" message must be the first message following the protocol.
+ * It tells ocfs2_control the local node number.
+ *
+ * Next comes the "SETV" message.  It has the following syntax:
+ *
+ *  SETV<space><2-char-hex-major><space><2-char-hex-minor><newline>
+ *
+ * This is 11 characters.
+ *
+ * The "SETV" message sets the filesystem locking protocol version as
+ * negotiated by the client.  The client negotiates based on the maximum
+ * version advertised in /sys/fs/ocfs2/max_locking_protocol.  The major
+ * number from the "SETV" message must match
+ * ocfs2_user_plugin.sp_max_proto.pv_major, and the minor number
+ * must be less than or equal to ...sp_max_version.pv_minor.
+ *
+ * Once this information has been set, mounts will be allowed.  From this
+ * point on, the "DOWN" message can be sent for node down notification.
+ * It has the following syntax:
+ *
+ *  DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline>
+ *
+ * eg:
+ *
+ *  DOWN 632A924FDD844190BDA93C0DF6B94899 00000001\n
+ *
+ * This is 47 characters.
+ */
+
+/*
+ * Whether or not the client has done the handshake.
+ * For now, we have just one protocol version.
+ */
+#define OCFS2_CONTROL_PROTO			"T01\n"
+#define OCFS2_CONTROL_PROTO_LEN			4
+
+/* Handshake states */
+#define OCFS2_CONTROL_HANDSHAKE_INVALID		(0)
+#define OCFS2_CONTROL_HANDSHAKE_READ		(1)
+#define OCFS2_CONTROL_HANDSHAKE_PROTOCOL	(2)
+#define OCFS2_CONTROL_HANDSHAKE_VALID		(3)
+
+/* Messages */
+#define OCFS2_CONTROL_MESSAGE_OP_LEN		4
+#define OCFS2_CONTROL_MESSAGE_SETNODE_OP	"SETN"
+#define OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN	14
+#define OCFS2_CONTROL_MESSAGE_SETVERSION_OP	"SETV"
+#define OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN	11
+#define OCFS2_CONTROL_MESSAGE_DOWN_OP		"DOWN"
+#define OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN	47
+#define OCFS2_TEXT_UUID_LEN			32
+#define OCFS2_CONTROL_MESSAGE_VERNUM_LEN	2
+#define OCFS2_CONTROL_MESSAGE_NODENUM_LEN	8
+#define VERSION_LOCK				"version_lock"
+
+enum ocfs2_connection_type {
+	WITH_CONTROLD,
+	NO_CONTROLD
+};
+
+/*
+ * ocfs2_live_connection is refcounted because the filesystem and
+ * miscdevice sides can detach in different order.  Let's just be safe.
+ */
+struct ocfs2_live_connection {
+	struct list_head		oc_list;
+	struct ocfs2_cluster_connection	*oc_conn;
+	enum ocfs2_connection_type	oc_type;
+	atomic_t                        oc_this_node;
+	int                             oc_our_slot;
+	struct dlm_lksb                 oc_version_lksb;
+	char                            oc_lvb[DLM_LVB_LEN];
+	struct completion               oc_sync_wait;
+	wait_queue_head_t		oc_wait;
+};
+
+struct ocfs2_control_private {
+	struct list_head op_list;
+	int op_state;
+	int op_this_node;
+	struct ocfs2_protocol_version op_proto;
+};
+
+/* SETN<space><8-char-hex-nodenum><newline> */
+struct ocfs2_control_message_setn {
+	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
+	char	space;
+	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
+	char	newline;
+};
+
+/* SETV<space><2-char-hex-major><space><2-char-hex-minor><newline> */
+struct ocfs2_control_message_setv {
+	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
+	char	space1;
+	char	major[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
+	char	space2;
+	char	minor[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
+	char	newline;
+};
+
+/* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */
+struct ocfs2_control_message_down {
+	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
+	char	space1;
+	char	uuid[OCFS2_TEXT_UUID_LEN];
+	char	space2;
+	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
+	char	newline;
+};
+
+union ocfs2_control_message {
+	char					tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
+	struct ocfs2_control_message_setn	u_setn;
+	struct ocfs2_control_message_setv	u_setv;
+	struct ocfs2_control_message_down	u_down;
+};
+
+static struct ocfs2_stack_plugin ocfs2_user_plugin;
+
+static atomic_t ocfs2_control_opened;
+static int ocfs2_control_this_node = -1;
+static struct ocfs2_protocol_version running_proto;
+
+static LIST_HEAD(ocfs2_live_connection_list);
+static LIST_HEAD(ocfs2_control_private_list);
+static DEFINE_MUTEX(ocfs2_control_lock);
+
+static inline void ocfs2_control_set_handshake_state(struct file *file,
+						     int state)
+{
+	struct ocfs2_control_private *p = file->private_data;
+	p->op_state = state;
+}
+
+static inline int ocfs2_control_get_handshake_state(struct file *file)
+{
+	struct ocfs2_control_private *p = file->private_data;
+	return p->op_state;
+}
+
+static struct ocfs2_live_connection *ocfs2_connection_find(const char *name)
+{
+	size_t len = strlen(name);
+	struct ocfs2_live_connection *c;
+
+	BUG_ON(!mutex_is_locked(&ocfs2_control_lock));
+
+	list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) {
+		if ((c->oc_conn->cc_namelen == len) &&
+		    !strncmp(c->oc_conn->cc_name, name, len))
+			return c;
+	}
+
+	return NULL;
+}
+
+/*
+ * ocfs2_live_connection structures are created underneath the ocfs2
+ * mount path.  Since the VFS prevents multiple calls to
+ * fill_super(), we can't get dupes here.
+ */
+static int ocfs2_live_connection_attach(struct ocfs2_cluster_connection *conn,
+				     struct ocfs2_live_connection *c)
+{
+	int rc = 0;
+
+	mutex_lock(&ocfs2_control_lock);
+	c->oc_conn = conn;
+
+	if ((c->oc_type == NO_CONTROLD) || atomic_read(&ocfs2_control_opened))
+		list_add(&c->oc_list, &ocfs2_live_connection_list);
+	else {
+		printk(KERN_ERR
+		       "ocfs2: Userspace control daemon is not present\n");
+		rc = -ESRCH;
+	}
+
+	mutex_unlock(&ocfs2_control_lock);
+	return rc;
+}
+
+/*
+ * This function disconnects the cluster connection from ocfs2_control.
+ * Afterwards, userspace can't affect the cluster connection.
+ */
+static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c)
+{
+	mutex_lock(&ocfs2_control_lock);
+	list_del_init(&c->oc_list);
+	c->oc_conn = NULL;
+	mutex_unlock(&ocfs2_control_lock);
+
+	kfree(c);
+}
+
+static int ocfs2_control_cfu(void *target, size_t target_len,
+			     const char __user *buf, size_t count)
+{
+	/* The T01 expects write(2) calls to have exactly one command */
+	if ((count != target_len) ||
+	    (count > sizeof(union ocfs2_control_message)))
+		return -EINVAL;
+
+	if (copy_from_user(target, buf, target_len))
+		return -EFAULT;
+
+	return 0;
+}
+
+static ssize_t ocfs2_control_validate_protocol(struct file *file,
+					       const char __user *buf,
+					       size_t count)
+{
+	ssize_t ret;
+	char kbuf[OCFS2_CONTROL_PROTO_LEN];
+
+	ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN,
+				buf, count);
+	if (ret)
+		return ret;
+
+	if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN))
+		return -EINVAL;
+
+	ocfs2_control_set_handshake_state(file,
+					  OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
+
+	return count;
+}
+
+static void ocfs2_control_send_down(const char *uuid,
+				    int nodenum)
+{
+	struct ocfs2_live_connection *c;
+
+	mutex_lock(&ocfs2_control_lock);
+
+	c = ocfs2_connection_find(uuid);
+	if (c) {
+		BUG_ON(c->oc_conn == NULL);
+		c->oc_conn->cc_recovery_handler(nodenum,
+						c->oc_conn->cc_recovery_data);
+	}
+
+	mutex_unlock(&ocfs2_control_lock);
+}
+
+/*
+ * Called whenever configuration elements are sent to /dev/ocfs2_control.
+ * If all configuration elements are present, try to set the global
+ * values.  If there is a problem, return an error.  Skip any missing
+ * elements, and only bump ocfs2_control_opened when we have all elements
+ * and are successful.
+ */
+static int ocfs2_control_install_private(struct file *file)
+{
+	int rc = 0;
+	int set_p = 1;
+	struct ocfs2_control_private *p = file->private_data;
+
+	BUG_ON(p->op_state != OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
+
+	mutex_lock(&ocfs2_control_lock);
+
+	if (p->op_this_node < 0) {
+		set_p = 0;
+	} else if ((ocfs2_control_this_node >= 0) &&
+		   (ocfs2_control_this_node != p->op_this_node)) {
+		rc = -EINVAL;
+		goto out_unlock;
+	}
+
+	if (!p->op_proto.pv_major) {
+		set_p = 0;
+	} else if (!list_empty(&ocfs2_live_connection_list) &&
+		   ((running_proto.pv_major != p->op_proto.pv_major) ||
+		    (running_proto.pv_minor != p->op_proto.pv_minor))) {
+		rc = -EINVAL;
+		goto out_unlock;
+	}
+
+	if (set_p) {
+		ocfs2_control_this_node = p->op_this_node;
+		running_proto.pv_major = p->op_proto.pv_major;
+		running_proto.pv_minor = p->op_proto.pv_minor;
+	}
+
+out_unlock:
+	mutex_unlock(&ocfs2_control_lock);
+
+	if (!rc && set_p) {
+		/* We set the global values successfully */
+		atomic_inc(&ocfs2_control_opened);
+		ocfs2_control_set_handshake_state(file,
+					OCFS2_CONTROL_HANDSHAKE_VALID);
+	}
+
+	return rc;
+}
+
+static int ocfs2_control_get_this_node(void)
+{
+	int rc;
+
+	mutex_lock(&ocfs2_control_lock);
+	if (ocfs2_control_this_node < 0)
+		rc = -EINVAL;
+	else
+		rc = ocfs2_control_this_node;
+	mutex_unlock(&ocfs2_control_lock);
+
+	return rc;
+}
+
+static int ocfs2_control_do_setnode_msg(struct file *file,
+					struct ocfs2_control_message_setn *msg)
+{
+	long nodenum;
+	char *ptr = NULL;
+	struct ocfs2_control_private *p = file->private_data;
+
+	if (ocfs2_control_get_handshake_state(file) !=
+	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
+		return -EINVAL;
+
+	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
+		    OCFS2_CONTROL_MESSAGE_OP_LEN))
+		return -EINVAL;
+
+	if ((msg->space != ' ') || (msg->newline != '\n'))
+		return -EINVAL;
+	msg->space = msg->newline = '\0';
+
+	nodenum = simple_strtol(msg->nodestr, &ptr, 16);
+	if (!ptr || *ptr)
+		return -EINVAL;
+
+	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
+	    (nodenum > INT_MAX) || (nodenum < 0))
+		return -ERANGE;
+	p->op_this_node = nodenum;
+
+	return ocfs2_control_install_private(file);
+}
+
+static int ocfs2_control_do_setversion_msg(struct file *file,
+					   struct ocfs2_control_message_setv *msg)
+ {
+	long major, minor;
+	char *ptr = NULL;
+	struct ocfs2_control_private *p = file->private_data;
+	struct ocfs2_protocol_version *max =
+		&ocfs2_user_plugin.sp_max_proto;
+
+	if (ocfs2_control_get_handshake_state(file) !=
+	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
+		return -EINVAL;
+
+	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
+		    OCFS2_CONTROL_MESSAGE_OP_LEN))
+		return -EINVAL;
+
+	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
+	    (msg->newline != '\n'))
+		return -EINVAL;
+	msg->space1 = msg->space2 = msg->newline = '\0';
+
+	major = simple_strtol(msg->major, &ptr, 16);
+	if (!ptr || *ptr)
+		return -EINVAL;
+	minor = simple_strtol(msg->minor, &ptr, 16);
+	if (!ptr || *ptr)
+		return -EINVAL;
+
+	/*
+	 * The major must be between 1 and 255, inclusive.  The minor
+	 * must be between 0 and 255, inclusive.  The version passed in
+	 * must be within the maximum version supported by the filesystem.
+	 */
+	if ((major == LONG_MIN) || (major == LONG_MAX) ||
+	    (major > (u8)-1) || (major < 1))
+		return -ERANGE;
+	if ((minor == LONG_MIN) || (minor == LONG_MAX) ||
+	    (minor > (u8)-1) || (minor < 0))
+		return -ERANGE;
+	if ((major != max->pv_major) ||
+	    (minor > max->pv_minor))
+		return -EINVAL;
+
+	p->op_proto.pv_major = major;
+	p->op_proto.pv_minor = minor;
+
+	return ocfs2_control_install_private(file);
+}
+
+static int ocfs2_control_do_down_msg(struct file *file,
+				     struct ocfs2_control_message_down *msg)
+{
+	long nodenum;
+	char *p = NULL;
+
+	if (ocfs2_control_get_handshake_state(file) !=
+	    OCFS2_CONTROL_HANDSHAKE_VALID)
+		return -EINVAL;
+
+	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
+		    OCFS2_CONTROL_MESSAGE_OP_LEN))
+		return -EINVAL;
+
+	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
+	    (msg->newline != '\n'))
+		return -EINVAL;
+	msg->space1 = msg->space2 = msg->newline = '\0';
+
+	nodenum = simple_strtol(msg->nodestr, &p, 16);
+	if (!p || *p)
+		return -EINVAL;
+
+	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
+	    (nodenum > INT_MAX) || (nodenum < 0))
+		return -ERANGE;
+
+	ocfs2_control_send_down(msg->uuid, nodenum);
+
+	return 0;
+}
+
+static ssize_t ocfs2_control_message(struct file *file,
+				     const char __user *buf,
+				     size_t count)
+{
+	ssize_t ret;
+	union ocfs2_control_message msg;
+
+	/* Try to catch padding issues */
+	WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) !=
+		(sizeof(msg.u_down.tag) + sizeof(msg.u_down.space1)));
+
+	memset(&msg, 0, sizeof(union ocfs2_control_message));
+	ret = ocfs2_control_cfu(&msg, count, buf, count);
+	if (ret)
+		goto out;
+
+	if ((count == OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN) &&
+	    !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
+		     OCFS2_CONTROL_MESSAGE_OP_LEN))
+		ret = ocfs2_control_do_setnode_msg(file, &msg.u_setn);
+	else if ((count == OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN) &&
+		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
+			  OCFS2_CONTROL_MESSAGE_OP_LEN))
+		ret = ocfs2_control_do_setversion_msg(file, &msg.u_setv);
+	else if ((count == OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN) &&
+		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
+			  OCFS2_CONTROL_MESSAGE_OP_LEN))
+		ret = ocfs2_control_do_down_msg(file, &msg.u_down);
+	else
+		ret = -EINVAL;
+
+out:
+	return ret ? ret : count;
+}
+
+static ssize_t ocfs2_control_write(struct file *file,
+				   const char __user *buf,
+				   size_t count,
+				   loff_t *ppos)
+{
+	ssize_t ret;
+
+	switch (ocfs2_control_get_handshake_state(file)) {
+		case OCFS2_CONTROL_HANDSHAKE_INVALID:
+			ret = -EINVAL;
+			break;
+
+		case OCFS2_CONTROL_HANDSHAKE_READ:
+			ret = ocfs2_control_validate_protocol(file, buf,
+							      count);
+			break;
+
+		case OCFS2_CONTROL_HANDSHAKE_PROTOCOL:
+		case OCFS2_CONTROL_HANDSHAKE_VALID:
+			ret = ocfs2_control_message(file, buf, count);
+			break;
+
+		default:
+			BUG();
+			ret = -EIO;
+			break;
+	}
+
+	return ret;
+}
+
+/*
+ * This is a naive version.  If we ever have a new protocol, we'll expand
+ * it.  Probably using seq_file.
+ */
+static ssize_t ocfs2_control_read(struct file *file,
+				  char __user *buf,
+				  size_t count,
+				  loff_t *ppos)
+{
+	ssize_t ret;
+
+	ret = simple_read_from_buffer(buf, count, ppos,
+			OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN);
+
+	/* Have we read the whole protocol list? */
+	if (ret > 0 && *ppos >= OCFS2_CONTROL_PROTO_LEN)
+		ocfs2_control_set_handshake_state(file,
+						  OCFS2_CONTROL_HANDSHAKE_READ);
+
+	return ret;
+}
+
+static int ocfs2_control_release(struct inode *inode, struct file *file)
+{
+	struct ocfs2_control_private *p = file->private_data;
+
+	mutex_lock(&ocfs2_control_lock);
+
+	if (ocfs2_control_get_handshake_state(file) !=
+	    OCFS2_CONTROL_HANDSHAKE_VALID)
+		goto out;
+
+	if (atomic_dec_and_test(&ocfs2_control_opened)) {
+		if (!list_empty(&ocfs2_live_connection_list)) {
+			/* XXX: Do bad things! */
+			printk(KERN_ERR
+			       "ocfs2: Unexpected release of ocfs2_control!\n"
+			       "       Loss of cluster connection requires "
+			       "an emergency restart!\n");
+			emergency_restart();
+		}
+		/*
+		 * Last valid close clears the node number and resets
+		 * the locking protocol version
+		 */
+		ocfs2_control_this_node = -1;
+		running_proto.pv_major = 0;
+		running_proto.pv_minor = 0;
+	}
+
+out:
+	list_del_init(&p->op_list);
+	file->private_data = NULL;
+
+	mutex_unlock(&ocfs2_control_lock);
+
+	kfree(p);
+
+	return 0;
+}
+
+static int ocfs2_control_open(struct inode *inode, struct file *file)
+{
+	struct ocfs2_control_private *p;
+
+	p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL);
+	if (!p)
+		return -ENOMEM;
+	p->op_this_node = -1;
+
+	mutex_lock(&ocfs2_control_lock);
+	file->private_data = p;
+	list_add(&p->op_list, &ocfs2_control_private_list);
+	mutex_unlock(&ocfs2_control_lock);
+
+	return 0;
+}
+
+static const struct file_operations ocfs2_control_fops = {
+	.open    = ocfs2_control_open,
+	.release = ocfs2_control_release,
+	.read    = ocfs2_control_read,
+	.write   = ocfs2_control_write,
+	.owner   = THIS_MODULE,
+	.llseek  = default_llseek,
+};
+
+static struct miscdevice ocfs2_control_device = {
+	.minor		= MISC_DYNAMIC_MINOR,
+	.name		= "ocfs2_control",
+	.fops		= &ocfs2_control_fops,
+};
+
+static int ocfs2_control_init(void)
+{
+	int rc;
+
+	atomic_set(&ocfs2_control_opened, 0);
+
+	rc = misc_register(&ocfs2_control_device);
+	if (rc)
+		printk(KERN_ERR
+		       "ocfs2: Unable to register ocfs2_control device "
+		       "(errno %d)\n",
+		       -rc);
+
+	return rc;
+}
+
+static void ocfs2_control_exit(void)
+{
+	int rc;
+
+	rc = misc_deregister(&ocfs2_control_device);
+	if (rc)
+		printk(KERN_ERR
+		       "ocfs2: Unable to deregister ocfs2_control device "
+		       "(errno %d)\n",
+		       -rc);
+}
+
+static void fsdlm_lock_ast_wrapper(void *astarg)
+{
+	struct ocfs2_dlm_lksb *lksb = astarg;
+	int status = lksb->lksb_fsdlm.sb_status;
+
+	/*
+	 * For now we're punting on the issue of other non-standard errors
+	 * where we can't tell if the unlock_ast or lock_ast should be called.
+	 * The main "other error" that's possible is EINVAL which means the
+	 * function was called with invalid args, which shouldn't be possible
+	 * since the caller here is under our control.  Other non-standard
+	 * errors probably fall into the same category, or otherwise are fatal
+	 * which means we can't carry on anyway.
+	 */
+
+	if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
+		lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, 0);
+	else
+		lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
+}
+
+static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
+{
+	struct ocfs2_dlm_lksb *lksb = astarg;
+
+	lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
+}
+
+static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
+			 int mode,
+			 struct ocfs2_dlm_lksb *lksb,
+			 u32 flags,
+			 void *name,
+			 unsigned int namelen)
+{
+	int ret;
+
+	if (!lksb->lksb_fsdlm.sb_lvbptr)
+		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
+					     sizeof(struct dlm_lksb);
+
+	ret = dlm_lock(conn->cc_lockspace, mode, &lksb->lksb_fsdlm,
+		       flags|DLM_LKF_NODLCKWT, name, namelen, 0,
+		       fsdlm_lock_ast_wrapper, lksb,
+		       fsdlm_blocking_ast_wrapper);
+	return ret;
+}
+
+static int user_dlm_unlock(struct ocfs2_cluster_connection *conn,
+			   struct ocfs2_dlm_lksb *lksb,
+			   u32 flags)
+{
+	int ret;
+
+	ret = dlm_unlock(conn->cc_lockspace, lksb->lksb_fsdlm.sb_lkid,
+			 flags, &lksb->lksb_fsdlm, lksb);
+	return ret;
+}
+
+static int user_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
+{
+	return lksb->lksb_fsdlm.sb_status;
+}
+
+static int user_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
+{
+	int invalid = lksb->lksb_fsdlm.sb_flags & DLM_SBF_VALNOTVALID;
+
+	return !invalid;
+}
+
+static void *user_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
+{
+	if (!lksb->lksb_fsdlm.sb_lvbptr)
+		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
+					     sizeof(struct dlm_lksb);
+	return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
+}
+
+static void user_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
+{
+}
+
+static int user_plock(struct ocfs2_cluster_connection *conn,
+		      u64 ino,
+		      struct file *file,
+		      int cmd,
+		      struct file_lock *fl)
+{
+	/*
+	 * This more or less just demuxes the plock request into any
+	 * one of three dlm calls.
+	 *
+	 * Internally, fs/dlm will pass these to a misc device, which
+	 * a userspace daemon will read and write to.
+	 *
+	 * For now, cancel requests (which happen internally only),
+	 * are turned into unlocks. Most of this function taken from
+	 * gfs2_lock.
+	 */
+
+	if (cmd == F_CANCELLK) {
+		cmd = F_SETLK;
+		fl->fl_type = F_UNLCK;
+	}
+
+	if (IS_GETLK(cmd))
+		return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
+	else if (fl->fl_type == F_UNLCK)
+		return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
+	else
+		return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
+}
+
+/*
+ * Compare a requested locking protocol version against the current one.
+ *
+ * If the major numbers are different, they are incompatible.
+ * If the current minor is greater than the request, they are incompatible.
+ * If the current minor is less than or equal to the request, they are
+ * compatible, and the requester should run at the current minor version.
+ */
+static int fs_protocol_compare(struct ocfs2_protocol_version *existing,
+			       struct ocfs2_protocol_version *request)
+{
+	if (existing->pv_major != request->pv_major)
+		return 1;
+
+	if (existing->pv_minor > request->pv_minor)
+		return 1;
+
+	if (existing->pv_minor < request->pv_minor)
+		request->pv_minor = existing->pv_minor;
+
+	return 0;
+}
+
+static void lvb_to_version(char *lvb, struct ocfs2_protocol_version *ver)
+{
+	struct ocfs2_protocol_version *pv =
+		(struct ocfs2_protocol_version *)lvb;
+	/*
+	 * ocfs2_protocol_version has two u8 variables, so we don't
+	 * need any endian conversion.
+	 */
+	ver->pv_major = pv->pv_major;
+	ver->pv_minor = pv->pv_minor;
+}
+
+static void version_to_lvb(struct ocfs2_protocol_version *ver, char *lvb)
+{
+	struct ocfs2_protocol_version *pv =
+		(struct ocfs2_protocol_version *)lvb;
+	/*
+	 * ocfs2_protocol_version has two u8 variables, so we don't
+	 * need any endian conversion.
+	 */
+	pv->pv_major = ver->pv_major;
+	pv->pv_minor = ver->pv_minor;
+}
+
+static void sync_wait_cb(void *arg)
+{
+	struct ocfs2_cluster_connection *conn = arg;
+	struct ocfs2_live_connection *lc = conn->cc_private;
+	complete(&lc->oc_sync_wait);
+}
+
+static int sync_unlock(struct ocfs2_cluster_connection *conn,
+		struct dlm_lksb *lksb, char *name)
+{
+	int error;
+	struct ocfs2_live_connection *lc = conn->cc_private;
+
+	error = dlm_unlock(conn->cc_lockspace, lksb->sb_lkid, 0, lksb, conn);
+	if (error) {
+		printk(KERN_ERR "%s lkid %x error %d\n",
+				name, lksb->sb_lkid, error);
+		return error;
+	}
+
+	wait_for_completion(&lc->oc_sync_wait);
+
+	if (lksb->sb_status != -DLM_EUNLOCK) {
+		printk(KERN_ERR "%s lkid %x status %d\n",
+				name, lksb->sb_lkid, lksb->sb_status);
+		return -1;
+	}
+	return 0;
+}
+
+static int sync_lock(struct ocfs2_cluster_connection *conn,
+		int mode, uint32_t flags,
+		struct dlm_lksb *lksb, char *name)
+{
+	int error, status;
+	struct ocfs2_live_connection *lc = conn->cc_private;
+
+	error = dlm_lock(conn->cc_lockspace, mode, lksb, flags,
+			name, strlen(name),
+			0, sync_wait_cb, conn, NULL);
+	if (error) {
+		printk(KERN_ERR "%s lkid %x flags %x mode %d error %d\n",
+				name, lksb->sb_lkid, flags, mode, error);
+		return error;
+	}
+
+	wait_for_completion(&lc->oc_sync_wait);
+
+	status = lksb->sb_status;
+
+	if (status && status != -EAGAIN) {
+		printk(KERN_ERR "%s lkid %x flags %x mode %d status %d\n",
+				name, lksb->sb_lkid, flags, mode, status);
+	}
+
+	return status;
+}
+
+
+static int version_lock(struct ocfs2_cluster_connection *conn, int mode,
+		int flags)
+{
+	struct ocfs2_live_connection *lc = conn->cc_private;
+	return sync_lock(conn, mode, flags,
+			&lc->oc_version_lksb, VERSION_LOCK);
+}
+
+static int version_unlock(struct ocfs2_cluster_connection *conn)
+{
+	struct ocfs2_live_connection *lc = conn->cc_private;
+	return sync_unlock(conn, &lc->oc_version_lksb, VERSION_LOCK);
+}
+
+/* get_protocol_version()
+ *
+ * To exchange ocfs2 versioning, we use the LVB of the version dlm lock.
+ * The algorithm is:
+ * 1. Attempt to take the lock in EX mode (non-blocking).
+ * 2. If successful (which means it is the first mount), write the
+ *    version number and downconvert to PR lock.
+ * 3. If unsuccessful (returns -EAGAIN), read the version from the LVB after
+ *    taking the PR lock.
+ */
+
+static int get_protocol_version(struct ocfs2_cluster_connection *conn)
+{
+	int ret;
+	struct ocfs2_live_connection *lc = conn->cc_private;
+	struct ocfs2_protocol_version pv;
+
+	running_proto.pv_major =
+		ocfs2_user_plugin.sp_max_proto.pv_major;
+	running_proto.pv_minor =
+		ocfs2_user_plugin.sp_max_proto.pv_minor;
+
+	lc->oc_version_lksb.sb_lvbptr = lc->oc_lvb;
+	ret = version_lock(conn, DLM_LOCK_EX,
+			DLM_LKF_VALBLK|DLM_LKF_NOQUEUE);
+	if (!ret) {
+		conn->cc_version.pv_major = running_proto.pv_major;
+		conn->cc_version.pv_minor = running_proto.pv_minor;
+		version_to_lvb(&running_proto, lc->oc_lvb);
+		version_lock(conn, DLM_LOCK_PR, DLM_LKF_CONVERT|DLM_LKF_VALBLK);
+	} else if (ret == -EAGAIN) {
+		ret = version_lock(conn, DLM_LOCK_PR, DLM_LKF_VALBLK);
+		if (ret)
+			goto out;
+		lvb_to_version(lc->oc_lvb, &pv);
+
+		if ((pv.pv_major != running_proto.pv_major) ||
+				(pv.pv_minor > running_proto.pv_minor)) {
+			ret = -EINVAL;
+			goto out;
+		}
+
+		conn->cc_version.pv_major = pv.pv_major;
+		conn->cc_version.pv_minor = pv.pv_minor;
+	}
+out:
+	return ret;
+}
+
+static void user_recover_prep(void *arg)
+{
+}
+
+static void user_recover_slot(void *arg, struct dlm_slot *slot)
+{
+	struct ocfs2_cluster_connection *conn = arg;
+	printk(KERN_INFO "ocfs2: Node %d/%d down. Initiating recovery.\n",
+			slot->nodeid, slot->slot);
+	conn->cc_recovery_handler(slot->nodeid, conn->cc_recovery_data);
+
+}
+
+static void user_recover_done(void *arg, struct dlm_slot *slots,
+		int num_slots, int our_slot,
+		uint32_t generation)
+{
+	struct ocfs2_cluster_connection *conn = arg;
+	struct ocfs2_live_connection *lc = conn->cc_private;
+	int i;
+
+	for (i = 0; i < num_slots; i++)
+		if (slots[i].slot == our_slot) {
+			atomic_set(&lc->oc_this_node, slots[i].nodeid);
+			break;
+		}
+
+	lc->oc_our_slot = our_slot;
+	wake_up(&lc->oc_wait);
+}
+
+static const struct dlm_lockspace_ops ocfs2_ls_ops = {
+	.recover_prep = user_recover_prep,
+	.recover_slot = user_recover_slot,
+	.recover_done = user_recover_done,
+};
+
+static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
+{
+	version_unlock(conn);
+	dlm_release_lockspace(conn->cc_lockspace, 2);
+	conn->cc_lockspace = NULL;
+	ocfs2_live_connection_drop(conn->cc_private);
+	conn->cc_private = NULL;
+	return 0;
+}
+
+static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
+{
+	dlm_lockspace_t *fsdlm;
+	struct ocfs2_live_connection *lc;
+	int rc, ops_rv;
+
+	BUG_ON(conn == NULL);
+
+	lc = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL);
+	if (!lc)
+		return -ENOMEM;
+
+	init_waitqueue_head(&lc->oc_wait);
+	init_completion(&lc->oc_sync_wait);
+	atomic_set(&lc->oc_this_node, 0);
+	conn->cc_private = lc;
+	lc->oc_type = NO_CONTROLD;
+
+	rc = dlm_new_lockspace(conn->cc_name, conn->cc_cluster_name,
+			       DLM_LSFL_FS, DLM_LVB_LEN,
+			       &ocfs2_ls_ops, conn, &ops_rv, &fsdlm);
+	if (rc)
+		goto out;
+
+	if (ops_rv == -EOPNOTSUPP) {
+		lc->oc_type = WITH_CONTROLD;
+		printk(KERN_NOTICE "ocfs2: You seem to be using an older "
+				"version of dlm_controld and/or ocfs2-tools."
+				" Please consider upgrading.\n");
+	} else if (ops_rv) {
+		rc = ops_rv;
+		goto out;
+	}
+	conn->cc_lockspace = fsdlm;
+
+	rc = ocfs2_live_connection_attach(conn, lc);
+	if (rc)
+		goto out;
+
+	if (lc->oc_type == NO_CONTROLD) {
+		rc = get_protocol_version(conn);
+		if (rc) {
+			printk(KERN_ERR "ocfs2: Could not determine"
+					" locking version\n");
+			user_cluster_disconnect(conn);
+			goto out;
+		}
+		wait_event(lc->oc_wait, (atomic_read(&lc->oc_this_node) > 0));
+	}
+
+	/*
+	 * running_proto must have been set before we allowed any mounts
+	 * to proceed.
+	 */
+	if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
+		printk(KERN_ERR
+		       "Unable to mount with fs locking protocol version "
+		       "%u.%u because negotiated protocol is %u.%u\n",
+		       conn->cc_version.pv_major, conn->cc_version.pv_minor,
+		       running_proto.pv_major, running_proto.pv_minor);
+		rc = -EPROTO;
+		ocfs2_live_connection_drop(lc);
+		lc = NULL;
+	}
+
+out:
+	if (rc)
+		kfree(lc);
+	return rc;
+}
+
+
+static int user_cluster_this_node(struct ocfs2_cluster_connection *conn,
+				  unsigned int *this_node)
+{
+	int rc;
+	struct ocfs2_live_connection *lc = conn->cc_private;
+
+	if (lc->oc_type == WITH_CONTROLD)
+		rc = ocfs2_control_get_this_node();
+	else if (lc->oc_type == NO_CONTROLD)
+		rc = atomic_read(&lc->oc_this_node);
+	else
+		rc = -EINVAL;
+
+	if (rc < 0)
+		return rc;
+
+	*this_node = rc;
+	return 0;
+}
+
+static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
+	.connect	= user_cluster_connect,
+	.disconnect	= user_cluster_disconnect,
+	.this_node	= user_cluster_this_node,
+	.dlm_lock	= user_dlm_lock,
+	.dlm_unlock	= user_dlm_unlock,
+	.lock_status	= user_dlm_lock_status,
+	.lvb_valid	= user_dlm_lvb_valid,
+	.lock_lvb	= user_dlm_lvb,
+	.plock		= user_plock,
+	.dump_lksb	= user_dlm_dump_lksb,
+};
+
+static struct ocfs2_stack_plugin ocfs2_user_plugin = {
+	.sp_name	= "user",
+	.sp_ops		= &ocfs2_user_plugin_ops,
+	.sp_owner	= THIS_MODULE,
+};
+
+
+static int __init ocfs2_user_plugin_init(void)
+{
+	int rc;
+
+	rc = ocfs2_control_init();
+	if (!rc) {
+		rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
+		if (rc)
+			ocfs2_control_exit();
+	}
+
+	return rc;
+}
+
+static void __exit ocfs2_user_plugin_exit(void)
+{
+	ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
+	ocfs2_control_exit();
+}
+
+MODULE_AUTHOR("Oracle");
+MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
+MODULE_LICENSE("GPL");
+module_init(ocfs2_user_plugin_init);
+module_exit(ocfs2_user_plugin_exit);
diff --git a/fs/ocfs2/stackglue.c b/fs/ocfs2/stackglue.c
new file mode 100644
index 000000000..5d965e83b
--- /dev/null
+++ b/fs/ocfs2/stackglue.c
@@ -0,0 +1,748 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * stackglue.c
+ *
+ * Code which implements an OCFS2 specific interface to underlying
+ * cluster stacks.
+ *
+ * Copyright (C) 2007, 2009 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation, version 2.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/kmod.h>
+#include <linux/fs.h>
+#include <linux/kobject.h>
+#include <linux/sysfs.h>
+#include <linux/sysctl.h>
+
+#include "ocfs2_fs.h"
+
+#include "stackglue.h"
+
+#define OCFS2_STACK_PLUGIN_O2CB		"o2cb"
+#define OCFS2_STACK_PLUGIN_USER		"user"
+#define OCFS2_MAX_HB_CTL_PATH		256
+
+static struct ocfs2_protocol_version locking_max_version;
+static DEFINE_SPINLOCK(ocfs2_stack_lock);
+static LIST_HEAD(ocfs2_stack_list);
+static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
+static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
+
+/*
+ * The stack currently in use.  If not null, active_stack->sp_count > 0,
+ * the module is pinned, and the locking protocol cannot be changed.
+ */
+static struct ocfs2_stack_plugin *active_stack;
+
+static struct ocfs2_stack_plugin *ocfs2_stack_lookup(const char *name)
+{
+	struct ocfs2_stack_plugin *p;
+
+	assert_spin_locked(&ocfs2_stack_lock);
+
+	list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
+		if (!strcmp(p->sp_name, name))
+			return p;
+	}
+
+	return NULL;
+}
+
+static int ocfs2_stack_driver_request(const char *stack_name,
+				      const char *plugin_name)
+{
+	int rc;
+	struct ocfs2_stack_plugin *p;
+
+	spin_lock(&ocfs2_stack_lock);
+
+	/*
+	 * If the stack passed by the filesystem isn't the selected one,
+	 * we can't continue.
+	 */
+	if (strcmp(stack_name, cluster_stack_name)) {
+		rc = -EBUSY;
+		goto out;
+	}
+
+	if (active_stack) {
+		/*
+		 * If the active stack isn't the one we want, it cannot
+		 * be selected right now.
+		 */
+		if (!strcmp(active_stack->sp_name, plugin_name))
+			rc = 0;
+		else
+			rc = -EBUSY;
+		goto out;
+	}
+
+	p = ocfs2_stack_lookup(plugin_name);
+	if (!p || !try_module_get(p->sp_owner)) {
+		rc = -ENOENT;
+		goto out;
+	}
+
+	active_stack = p;
+	rc = 0;
+
+out:
+	/* If we found it, pin it */
+	if (!rc)
+		active_stack->sp_count++;
+
+	spin_unlock(&ocfs2_stack_lock);
+	return rc;
+}
+
+/*
+ * This function looks up the appropriate stack and makes it active.  If
+ * there is no stack, it tries to load it.  It will fail if the stack still
+ * cannot be found.  It will also fail if a different stack is in use.
+ */
+static int ocfs2_stack_driver_get(const char *stack_name)
+{
+	int rc;
+	char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
+
+	/*
+	 * Classic stack does not pass in a stack name.  This is
+	 * compatible with older tools as well.
+	 */
+	if (!stack_name || !*stack_name)
+		stack_name = OCFS2_STACK_PLUGIN_O2CB;
+
+	if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
+		printk(KERN_ERR
+		       "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
+		       stack_name);
+		return -EINVAL;
+	}
+
+	/* Anything that isn't the classic stack is a user stack */
+	if (strcmp(stack_name, OCFS2_STACK_PLUGIN_O2CB))
+		plugin_name = OCFS2_STACK_PLUGIN_USER;
+
+	rc = ocfs2_stack_driver_request(stack_name, plugin_name);
+	if (rc == -ENOENT) {
+		request_module("ocfs2_stack_%s", plugin_name);
+		rc = ocfs2_stack_driver_request(stack_name, plugin_name);
+	}
+
+	if (rc == -ENOENT) {
+		printk(KERN_ERR
+		       "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
+		       plugin_name);
+	} else if (rc == -EBUSY) {
+		printk(KERN_ERR
+		       "ocfs2: A different cluster stack is in use\n");
+	}
+
+	return rc;
+}
+
+static void ocfs2_stack_driver_put(void)
+{
+	spin_lock(&ocfs2_stack_lock);
+	BUG_ON(active_stack == NULL);
+	BUG_ON(active_stack->sp_count == 0);
+
+	active_stack->sp_count--;
+	if (!active_stack->sp_count) {
+		module_put(active_stack->sp_owner);
+		active_stack = NULL;
+	}
+	spin_unlock(&ocfs2_stack_lock);
+}
+
+int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
+{
+	int rc;
+
+	spin_lock(&ocfs2_stack_lock);
+	if (!ocfs2_stack_lookup(plugin->sp_name)) {
+		plugin->sp_count = 0;
+		plugin->sp_max_proto = locking_max_version;
+		list_add(&plugin->sp_list, &ocfs2_stack_list);
+		printk(KERN_INFO "ocfs2: Registered cluster interface %s\n",
+		       plugin->sp_name);
+		rc = 0;
+	} else {
+		printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
+		       plugin->sp_name);
+		rc = -EEXIST;
+	}
+	spin_unlock(&ocfs2_stack_lock);
+
+	return rc;
+}
+EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
+
+void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
+{
+	struct ocfs2_stack_plugin *p;
+
+	spin_lock(&ocfs2_stack_lock);
+	p = ocfs2_stack_lookup(plugin->sp_name);
+	if (p) {
+		BUG_ON(p != plugin);
+		BUG_ON(plugin == active_stack);
+		BUG_ON(plugin->sp_count != 0);
+		list_del_init(&plugin->sp_list);
+		printk(KERN_INFO "ocfs2: Unregistered cluster interface %s\n",
+		       plugin->sp_name);
+	} else {
+		printk(KERN_ERR "Stack \"%s\" is not registered\n",
+		       plugin->sp_name);
+	}
+	spin_unlock(&ocfs2_stack_lock);
+}
+EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
+
+void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_proto)
+{
+	struct ocfs2_stack_plugin *p;
+
+	spin_lock(&ocfs2_stack_lock);
+	if (memcmp(max_proto, &locking_max_version,
+		   sizeof(struct ocfs2_protocol_version))) {
+		BUG_ON(locking_max_version.pv_major != 0);
+
+		locking_max_version = *max_proto;
+		list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
+			p->sp_max_proto = locking_max_version;
+		}
+	}
+	spin_unlock(&ocfs2_stack_lock);
+}
+EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version);
+
+
+/*
+ * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take no argument
+ * for the ast and bast functions.  They will pass the lksb to the ast
+ * and bast.  The caller can wrap the lksb with their own structure to
+ * get more information.
+ */
+int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
+		   int mode,
+		   struct ocfs2_dlm_lksb *lksb,
+		   u32 flags,
+		   void *name,
+		   unsigned int namelen)
+{
+	if (!lksb->lksb_conn)
+		lksb->lksb_conn = conn;
+	else
+		BUG_ON(lksb->lksb_conn != conn);
+	return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
+					      name, namelen);
+}
+EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
+
+int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
+		     struct ocfs2_dlm_lksb *lksb,
+		     u32 flags)
+{
+	BUG_ON(lksb->lksb_conn == NULL);
+
+	return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
+}
+EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
+
+int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
+{
+	return active_stack->sp_ops->lock_status(lksb);
+}
+EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
+
+int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
+{
+	return active_stack->sp_ops->lvb_valid(lksb);
+}
+EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
+
+void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
+{
+	return active_stack->sp_ops->lock_lvb(lksb);
+}
+EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
+
+void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
+{
+	active_stack->sp_ops->dump_lksb(lksb);
+}
+EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
+
+int ocfs2_stack_supports_plocks(void)
+{
+	return active_stack && active_stack->sp_ops->plock;
+}
+EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
+
+/*
+ * ocfs2_plock() can only be safely called if
+ * ocfs2_stack_supports_plocks() returned true
+ */
+int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
+		struct file *file, int cmd, struct file_lock *fl)
+{
+	WARN_ON_ONCE(active_stack->sp_ops->plock == NULL);
+	if (active_stack->sp_ops->plock)
+		return active_stack->sp_ops->plock(conn, ino, file, cmd, fl);
+	return -EOPNOTSUPP;
+}
+EXPORT_SYMBOL_GPL(ocfs2_plock);
+
+int ocfs2_cluster_connect(const char *stack_name,
+			  const char *cluster_name,
+			  int cluster_name_len,
+			  const char *group,
+			  int grouplen,
+			  struct ocfs2_locking_protocol *lproto,
+			  void (*recovery_handler)(int node_num,
+						   void *recovery_data),
+			  void *recovery_data,
+			  struct ocfs2_cluster_connection **conn)
+{
+	int rc = 0;
+	struct ocfs2_cluster_connection *new_conn;
+
+	BUG_ON(group == NULL);
+	BUG_ON(conn == NULL);
+	BUG_ON(recovery_handler == NULL);
+
+	if (grouplen > GROUP_NAME_MAX) {
+		rc = -EINVAL;
+		goto out;
+	}
+
+	if (memcmp(&lproto->lp_max_version, &locking_max_version,
+		   sizeof(struct ocfs2_protocol_version))) {
+		rc = -EINVAL;
+		goto out;
+	}
+
+	new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
+			   GFP_KERNEL);
+	if (!new_conn) {
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	strlcpy(new_conn->cc_name, group, GROUP_NAME_MAX + 1);
+	new_conn->cc_namelen = grouplen;
+	if (cluster_name_len)
+		strlcpy(new_conn->cc_cluster_name, cluster_name,
+			CLUSTER_NAME_MAX + 1);
+	new_conn->cc_cluster_name_len = cluster_name_len;
+	new_conn->cc_recovery_handler = recovery_handler;
+	new_conn->cc_recovery_data = recovery_data;
+
+	new_conn->cc_proto = lproto;
+	/* Start the new connection at our maximum compatibility level */
+	new_conn->cc_version = lproto->lp_max_version;
+
+	/* This will pin the stack driver if successful */
+	rc = ocfs2_stack_driver_get(stack_name);
+	if (rc)
+		goto out_free;
+
+	rc = active_stack->sp_ops->connect(new_conn);
+	if (rc) {
+		ocfs2_stack_driver_put();
+		goto out_free;
+	}
+
+	*conn = new_conn;
+
+out_free:
+	if (rc)
+		kfree(new_conn);
+
+out:
+	return rc;
+}
+EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
+
+/* The caller will ensure all nodes have the same cluster stack */
+int ocfs2_cluster_connect_agnostic(const char *group,
+				   int grouplen,
+				   struct ocfs2_locking_protocol *lproto,
+				   void (*recovery_handler)(int node_num,
+							    void *recovery_data),
+				   void *recovery_data,
+				   struct ocfs2_cluster_connection **conn)
+{
+	char *stack_name = NULL;
+
+	if (cluster_stack_name[0])
+		stack_name = cluster_stack_name;
+	return ocfs2_cluster_connect(stack_name, NULL, 0, group, grouplen,
+				     lproto, recovery_handler, recovery_data,
+				     conn);
+}
+EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic);
+
+/* If hangup_pending is 0, the stack driver will be dropped */
+int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
+			     int hangup_pending)
+{
+	int ret;
+
+	BUG_ON(conn == NULL);
+
+	ret = active_stack->sp_ops->disconnect(conn);
+
+	/* XXX Should we free it anyway? */
+	if (!ret) {
+		kfree(conn);
+		if (!hangup_pending)
+			ocfs2_stack_driver_put();
+	}
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
+
+/*
+ * Leave the group for this filesystem.  This is executed by a userspace
+ * program (stored in ocfs2_hb_ctl_path).
+ */
+static void ocfs2_leave_group(const char *group)
+{
+	int ret;
+	char *argv[5], *envp[3];
+
+	argv[0] = ocfs2_hb_ctl_path;
+	argv[1] = "-K";
+	argv[2] = "-u";
+	argv[3] = (char *)group;
+	argv[4] = NULL;
+
+	/* minimal command environment taken from cpu_run_sbin_hotplug */
+	envp[0] = "HOME=/";
+	envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
+	envp[2] = NULL;
+
+	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
+	if (ret < 0) {
+		printk(KERN_ERR
+		       "ocfs2: Error %d running user helper "
+		       "\"%s %s %s %s\"\n",
+		       ret, argv[0], argv[1], argv[2], argv[3]);
+	}
+}
+
+/*
+ * Hangup is a required post-umount.  ocfs2-tools software expects the
+ * filesystem to call "ocfs2_hb_ctl" during unmount.  This happens
+ * regardless of whether the DLM got started, so we can't do it
+ * in ocfs2_cluster_disconnect().  The ocfs2_leave_group() function does
+ * the actual work.
+ */
+void ocfs2_cluster_hangup(const char *group, int grouplen)
+{
+	BUG_ON(group == NULL);
+	BUG_ON(group[grouplen] != '\0');
+
+	ocfs2_leave_group(group);
+
+	/* cluster_disconnect() was called with hangup_pending==1 */
+	ocfs2_stack_driver_put();
+}
+EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
+
+int ocfs2_cluster_this_node(struct ocfs2_cluster_connection *conn,
+			    unsigned int *node)
+{
+	return active_stack->sp_ops->this_node(conn, node);
+}
+EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
+
+
+/*
+ * Sysfs bits
+ */
+
+static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
+					       struct kobj_attribute *attr,
+					       char *buf)
+{
+	ssize_t ret = 0;
+
+	spin_lock(&ocfs2_stack_lock);
+	if (locking_max_version.pv_major)
+		ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
+			       locking_max_version.pv_major,
+			       locking_max_version.pv_minor);
+	spin_unlock(&ocfs2_stack_lock);
+
+	return ret;
+}
+
+static struct kobj_attribute ocfs2_attr_max_locking_protocol =
+	__ATTR(max_locking_protocol, S_IRUGO,
+	       ocfs2_max_locking_protocol_show, NULL);
+
+static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
+						 struct kobj_attribute *attr,
+						 char *buf)
+{
+	ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
+	struct ocfs2_stack_plugin *p;
+
+	spin_lock(&ocfs2_stack_lock);
+	list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
+		ret = snprintf(buf, remain, "%s\n",
+			       p->sp_name);
+		if (ret < 0) {
+			total = ret;
+			break;
+		}
+		if (ret == remain) {
+			/* snprintf() didn't fit */
+			total = -E2BIG;
+			break;
+		}
+		total += ret;
+		remain -= ret;
+	}
+	spin_unlock(&ocfs2_stack_lock);
+
+	return total;
+}
+
+static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
+	__ATTR(loaded_cluster_plugins, S_IRUGO,
+	       ocfs2_loaded_cluster_plugins_show, NULL);
+
+static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
+						struct kobj_attribute *attr,
+						char *buf)
+{
+	ssize_t ret = 0;
+
+	spin_lock(&ocfs2_stack_lock);
+	if (active_stack) {
+		ret = snprintf(buf, PAGE_SIZE, "%s\n",
+			       active_stack->sp_name);
+		if (ret == PAGE_SIZE)
+			ret = -E2BIG;
+	}
+	spin_unlock(&ocfs2_stack_lock);
+
+	return ret;
+}
+
+static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
+	__ATTR(active_cluster_plugin, S_IRUGO,
+	       ocfs2_active_cluster_plugin_show, NULL);
+
+static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
+					struct kobj_attribute *attr,
+					char *buf)
+{
+	ssize_t ret;
+	spin_lock(&ocfs2_stack_lock);
+	ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
+	spin_unlock(&ocfs2_stack_lock);
+
+	return ret;
+}
+
+static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
+					 struct kobj_attribute *attr,
+					 const char *buf, size_t count)
+{
+	size_t len = count;
+	ssize_t ret;
+
+	if (len == 0)
+		return len;
+
+	if (buf[len - 1] == '\n')
+		len--;
+
+	if ((len != OCFS2_STACK_LABEL_LEN) ||
+	    (strnlen(buf, len) != len))
+		return -EINVAL;
+
+	spin_lock(&ocfs2_stack_lock);
+	if (active_stack) {
+		if (!strncmp(buf, cluster_stack_name, len))
+			ret = count;
+		else
+			ret = -EBUSY;
+	} else {
+		memcpy(cluster_stack_name, buf, len);
+		ret = count;
+	}
+	spin_unlock(&ocfs2_stack_lock);
+
+	return ret;
+}
+
+
+static struct kobj_attribute ocfs2_attr_cluster_stack =
+	__ATTR(cluster_stack, S_IRUGO | S_IWUSR,
+	       ocfs2_cluster_stack_show,
+	       ocfs2_cluster_stack_store);
+
+
+
+static ssize_t ocfs2_dlm_recover_show(struct kobject *kobj,
+					struct kobj_attribute *attr,
+					char *buf)
+{
+	return snprintf(buf, PAGE_SIZE, "1\n");
+}
+
+static struct kobj_attribute ocfs2_attr_dlm_recover_support =
+	__ATTR(dlm_recover_callback_support, S_IRUGO,
+	       ocfs2_dlm_recover_show, NULL);
+
+static struct attribute *ocfs2_attrs[] = {
+	&ocfs2_attr_max_locking_protocol.attr,
+	&ocfs2_attr_loaded_cluster_plugins.attr,
+	&ocfs2_attr_active_cluster_plugin.attr,
+	&ocfs2_attr_cluster_stack.attr,
+	&ocfs2_attr_dlm_recover_support.attr,
+	NULL,
+};
+
+static struct attribute_group ocfs2_attr_group = {
+	.attrs = ocfs2_attrs,
+};
+
+static struct kset *ocfs2_kset;
+
+static void ocfs2_sysfs_exit(void)
+{
+	kset_unregister(ocfs2_kset);
+}
+
+static int ocfs2_sysfs_init(void)
+{
+	int ret;
+
+	ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
+	if (!ocfs2_kset)
+		return -ENOMEM;
+
+	ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
+	if (ret)
+		goto error;
+
+	return 0;
+
+error:
+	kset_unregister(ocfs2_kset);
+	return ret;
+}
+
+/*
+ * Sysctl bits
+ *
+ * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path.  The 'nm' doesn't
+ * make as much sense in a multiple cluster stack world, but it's safer
+ * and easier to preserve the name.
+ */
+
+#define FS_OCFS2_NM		1
+
+static struct ctl_table ocfs2_nm_table[] = {
+	{
+		.procname	= "hb_ctl_path",
+		.data		= ocfs2_hb_ctl_path,
+		.maxlen		= OCFS2_MAX_HB_CTL_PATH,
+		.mode		= 0644,
+		.proc_handler	= proc_dostring,
+	},
+	{ }
+};
+
+static struct ctl_table ocfs2_mod_table[] = {
+	{
+		.procname	= "nm",
+		.data		= NULL,
+		.maxlen		= 0,
+		.mode		= 0555,
+		.child		= ocfs2_nm_table
+	},
+	{ }
+};
+
+static struct ctl_table ocfs2_kern_table[] = {
+	{
+		.procname	= "ocfs2",
+		.data		= NULL,
+		.maxlen		= 0,
+		.mode		= 0555,
+		.child		= ocfs2_mod_table
+	},
+	{ }
+};
+
+static struct ctl_table ocfs2_root_table[] = {
+	{
+		.procname	= "fs",
+		.data		= NULL,
+		.maxlen		= 0,
+		.mode		= 0555,
+		.child		= ocfs2_kern_table
+	},
+	{ }
+};
+
+static struct ctl_table_header *ocfs2_table_header;
+
+
+/*
+ * Initialization
+ */
+
+static int __init ocfs2_stack_glue_init(void)
+{
+	strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
+
+	ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
+	if (!ocfs2_table_header) {
+		printk(KERN_ERR
+		       "ocfs2 stack glue: unable to register sysctl\n");
+		return -ENOMEM; /* or something. */
+	}
+
+	return ocfs2_sysfs_init();
+}
+
+static void __exit ocfs2_stack_glue_exit(void)
+{
+	memset(&locking_max_version, 0,
+	       sizeof(struct ocfs2_protocol_version));
+	locking_max_version.pv_major = 0;
+	locking_max_version.pv_minor = 0;
+	ocfs2_sysfs_exit();
+	if (ocfs2_table_header)
+		unregister_sysctl_table(ocfs2_table_header);
+}
+
+MODULE_AUTHOR("Oracle");
+MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
+MODULE_LICENSE("GPL");
+module_init(ocfs2_stack_glue_init);
+module_exit(ocfs2_stack_glue_exit);
diff --git a/fs/ocfs2/stackglue.h b/fs/ocfs2/stackglue.h
new file mode 100644
index 000000000..66334a30c
--- /dev/null
+++ b/fs/ocfs2/stackglue.h
@@ -0,0 +1,301 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * stackglue.h
+ *
+ * Glue to the underlying cluster stack.
+ *
+ * Copyright (C) 2007 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation, version 2.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+
+#ifndef STACKGLUE_H
+#define STACKGLUE_H
+
+#include <linux/types.h>
+#include <linux/list.h>
+#include <linux/dlmconstants.h>
+
+#include "dlm/dlmapi.h"
+#include <linux/dlm.h>
+
+/* Needed for plock-related prototypes */
+struct file;
+struct file_lock;
+
+/*
+ * dlmconstants.h does not have a LOCAL flag.  We hope to remove it
+ * some day, but right now we need it.  Let's fake it.  This value is larger
+ * than any flag in dlmconstants.h.
+ */
+#define DLM_LKF_LOCAL		0x00100000
+
+/*
+ * This shadows DLM_LOCKSPACE_LEN in fs/dlm/dlm_internal.h.  That probably
+ * wants to be in a public header.
+ */
+#define GROUP_NAME_MAX		64
+
+/* This shadows  OCFS2_CLUSTER_NAME_LEN */
+#define CLUSTER_NAME_MAX	16
+
+
+/*
+ * ocfs2_protocol_version changes when ocfs2 does something different in
+ * its inter-node behavior.  See dlmglue.c for more information.
+ */
+struct ocfs2_protocol_version {
+	u8 pv_major;
+	u8 pv_minor;
+};
+
+/*
+ * The dlm_lockstatus struct includes lvb space, but the dlm_lksb struct only
+ * has a pointer to separately allocated lvb space.  This struct exists only to
+ * include in the lksb union to make space for a combined dlm_lksb and lvb.
+ */
+struct fsdlm_lksb_plus_lvb {
+	struct dlm_lksb lksb;
+	char lvb[DLM_LVB_LEN];
+};
+
+/*
+ * A union of all lock status structures.  We define it here so that the
+ * size of the union is known.  Lock status structures are embedded in
+ * ocfs2 inodes.
+ */
+struct ocfs2_cluster_connection;
+struct ocfs2_dlm_lksb {
+	 union {
+		 struct dlm_lockstatus lksb_o2dlm;
+		 struct dlm_lksb lksb_fsdlm;
+		 struct fsdlm_lksb_plus_lvb padding;
+	 };
+	 struct ocfs2_cluster_connection *lksb_conn;
+};
+
+/*
+ * The ocfs2_locking_protocol defines the handlers called on ocfs2's behalf.
+ */
+struct ocfs2_locking_protocol {
+	struct ocfs2_protocol_version lp_max_version;
+	void (*lp_lock_ast)(struct ocfs2_dlm_lksb *lksb);
+	void (*lp_blocking_ast)(struct ocfs2_dlm_lksb *lksb, int level);
+	void (*lp_unlock_ast)(struct ocfs2_dlm_lksb *lksb, int error);
+};
+
+
+/*
+ * A cluster connection.  Mostly opaque to ocfs2, the connection holds
+ * state for the underlying stack.  ocfs2 does use cc_version to determine
+ * locking compatibility.
+ */
+struct ocfs2_cluster_connection {
+	char cc_name[GROUP_NAME_MAX + 1];
+	int cc_namelen;
+	char cc_cluster_name[CLUSTER_NAME_MAX + 1];
+	int cc_cluster_name_len;
+	struct ocfs2_protocol_version cc_version;
+	struct ocfs2_locking_protocol *cc_proto;
+	void (*cc_recovery_handler)(int node_num, void *recovery_data);
+	void *cc_recovery_data;
+	void *cc_lockspace;
+	void *cc_private;
+};
+
+/*
+ * Each cluster stack implements the stack operations structure.  Not used
+ * in the ocfs2 code, the stackglue code translates generic cluster calls
+ * into stack operations.
+ */
+struct ocfs2_stack_operations {
+	/*
+	 * The fs code calls ocfs2_cluster_connect() to attach a new
+	 * filesystem to the cluster stack.  The ->connect() op is passed
+	 * an ocfs2_cluster_connection with the name and recovery field
+	 * filled in.
+	 *
+	 * The stack must set up any notification mechanisms and create
+	 * the filesystem lockspace in the DLM.  The lockspace should be
+	 * stored on cc_lockspace.  Any other information can be stored on
+	 * cc_private.
+	 *
+	 * ->connect() must not return until it is guaranteed that
+	 *
+	 *  - Node down notifications for the filesystem will be received
+	 *    and passed to conn->cc_recovery_handler().
+	 *  - Locking requests for the filesystem will be processed.
+	 */
+	int (*connect)(struct ocfs2_cluster_connection *conn);
+
+	/*
+	 * The fs code calls ocfs2_cluster_disconnect() when a filesystem
+	 * no longer needs cluster services.  All DLM locks have been
+	 * dropped, and recovery notification is being ignored by the
+	 * fs code.  The stack must disengage from the DLM and discontinue
+	 * recovery notification.
+	 *
+	 * Once ->disconnect() has returned, the connection structure will
+	 * be freed.  Thus, a stack must not return from ->disconnect()
+	 * until it will no longer reference the conn pointer.
+	 *
+	 * Once this call returns, the stack glue will be dropping this
+	 * connection's reference on the module.
+	 */
+	int (*disconnect)(struct ocfs2_cluster_connection *conn);
+
+	/*
+	 * ->this_node() returns the cluster's unique identifier for the
+	 * local node.
+	 */
+	int (*this_node)(struct ocfs2_cluster_connection *conn,
+			 unsigned int *node);
+
+	/*
+	 * Call the underlying dlm lock function.  The ->dlm_lock()
+	 * callback should convert the flags and mode as appropriate.
+	 *
+	 * ast and bast functions are not part of the call because the
+	 * stack will likely want to wrap ast and bast calls before passing
+	 * them to stack->sp_proto.  There is no astarg.  The lksb will
+	 * be passed back to the ast and bast functions.  The caller can
+	 * use this to find their object.
+	 */
+	int (*dlm_lock)(struct ocfs2_cluster_connection *conn,
+			int mode,
+			struct ocfs2_dlm_lksb *lksb,
+			u32 flags,
+			void *name,
+			unsigned int namelen);
+
+	/*
+	 * Call the underlying dlm unlock function.  The ->dlm_unlock()
+	 * function should convert the flags as appropriate.
+	 *
+	 * The unlock ast is not passed, as the stack will want to wrap
+	 * it before calling stack->sp_proto->lp_unlock_ast().  There is
+	 * no astarg.  The lksb will be passed back to the unlock ast
+	 * function.  The caller can use this to find their object.
+	 */
+	int (*dlm_unlock)(struct ocfs2_cluster_connection *conn,
+			  struct ocfs2_dlm_lksb *lksb,
+			  u32 flags);
+
+	/*
+	 * Return the status of the current lock status block.  The fs
+	 * code should never dereference the union.  The ->lock_status()
+	 * callback pulls out the stack-specific lksb, converts the status
+	 * to a proper errno, and returns it.
+	 */
+	int (*lock_status)(struct ocfs2_dlm_lksb *lksb);
+
+	/*
+	 * Return non-zero if the LVB is valid.
+	 */
+	int (*lvb_valid)(struct ocfs2_dlm_lksb *lksb);
+
+	/*
+	 * Pull the lvb pointer off of the stack-specific lksb.
+	 */
+	void *(*lock_lvb)(struct ocfs2_dlm_lksb *lksb);
+
+	/*
+	 * Cluster-aware posix locks
+	 *
+	 * This is NULL for stacks which do not support posix locks.
+	 */
+	int (*plock)(struct ocfs2_cluster_connection *conn,
+		     u64 ino,
+		     struct file *file,
+		     int cmd,
+		     struct file_lock *fl);
+
+	/*
+	 * This is an optoinal debugging hook.  If provided, the
+	 * stack can dump debugging information about this lock.
+	 */
+	void (*dump_lksb)(struct ocfs2_dlm_lksb *lksb);
+};
+
+/*
+ * Each stack plugin must describe itself by registering a
+ * ocfs2_stack_plugin structure.  This is only seen by stackglue and the
+ * stack driver.
+ */
+struct ocfs2_stack_plugin {
+	char *sp_name;
+	struct ocfs2_stack_operations *sp_ops;
+	struct module *sp_owner;
+
+	/* These are managed by the stackglue code. */
+	struct list_head sp_list;
+	unsigned int sp_count;
+	struct ocfs2_protocol_version sp_max_proto;
+};
+
+
+/* Used by the filesystem */
+int ocfs2_cluster_connect(const char *stack_name,
+			  const char *cluster_name,
+			  int cluster_name_len,
+			  const char *group,
+			  int grouplen,
+			  struct ocfs2_locking_protocol *lproto,
+			  void (*recovery_handler)(int node_num,
+						   void *recovery_data),
+			  void *recovery_data,
+			  struct ocfs2_cluster_connection **conn);
+/*
+ * Used by callers that don't store their stack name.  They must ensure
+ * all nodes have the same stack.
+ */
+int ocfs2_cluster_connect_agnostic(const char *group,
+				   int grouplen,
+				   struct ocfs2_locking_protocol *lproto,
+				   void (*recovery_handler)(int node_num,
+							    void *recovery_data),
+				   void *recovery_data,
+				   struct ocfs2_cluster_connection **conn);
+int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
+			     int hangup_pending);
+void ocfs2_cluster_hangup(const char *group, int grouplen);
+int ocfs2_cluster_this_node(struct ocfs2_cluster_connection *conn,
+			    unsigned int *node);
+
+struct ocfs2_lock_res;
+int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
+		   int mode,
+		   struct ocfs2_dlm_lksb *lksb,
+		   u32 flags,
+		   void *name,
+		   unsigned int namelen);
+int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
+		     struct ocfs2_dlm_lksb *lksb,
+		     u32 flags);
+
+int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb);
+int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb);
+void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb);
+void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb);
+
+int ocfs2_stack_supports_plocks(void);
+int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
+		struct file *file, int cmd, struct file_lock *fl);
+
+void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_proto);
+
+
+/* Used by stack plugins */
+int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin);
+void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin);
+
+#endif  /* STACKGLUE_H */
diff --git a/fs/ocfs2/suballoc.c b/fs/ocfs2/suballoc.c
new file mode 100644
index 000000000..447902963
--- /dev/null
+++ b/fs/ocfs2/suballoc.c
@@ -0,0 +1,2919 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * suballoc.c
+ *
+ * metadata alloc and free
+ * Inspired by ext3 block groups.
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "blockcheck.h"
+#include "dlmglue.h"
+#include "inode.h"
+#include "journal.h"
+#include "localalloc.h"
+#include "suballoc.h"
+#include "super.h"
+#include "sysfile.h"
+#include "uptodate.h"
+#include "ocfs2_trace.h"
+
+#include "buffer_head_io.h"
+
+#define NOT_ALLOC_NEW_GROUP		0
+#define ALLOC_NEW_GROUP			0x1
+#define ALLOC_GROUPS_FROM_GLOBAL	0x2
+
+#define OCFS2_MAX_TO_STEAL		1024
+
+struct ocfs2_suballoc_result {
+	u64		sr_bg_blkno;	/* The bg we allocated from.  Set
+					   to 0 when a block group is
+					   contiguous. */
+	u64		sr_bg_stable_blkno; /*
+					     * Doesn't change, always
+					     * set to target block
+					     * group descriptor
+					     * block.
+					     */
+	u64		sr_blkno;	/* The first allocated block */
+	unsigned int	sr_bit_offset;	/* The bit in the bg */
+	unsigned int	sr_bits;	/* How many bits we claimed */
+};
+
+static u64 ocfs2_group_from_res(struct ocfs2_suballoc_result *res)
+{
+	if (res->sr_blkno == 0)
+		return 0;
+
+	if (res->sr_bg_blkno)
+		return res->sr_bg_blkno;
+
+	return ocfs2_which_suballoc_group(res->sr_blkno, res->sr_bit_offset);
+}
+
+static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg);
+static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe);
+static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl);
+static int ocfs2_block_group_fill(handle_t *handle,
+				  struct inode *alloc_inode,
+				  struct buffer_head *bg_bh,
+				  u64 group_blkno,
+				  unsigned int group_clusters,
+				  u16 my_chain,
+				  struct ocfs2_chain_list *cl);
+static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
+				   struct inode *alloc_inode,
+				   struct buffer_head *bh,
+				   u64 max_block,
+				   u64 *last_alloc_group,
+				   int flags);
+
+static int ocfs2_cluster_group_search(struct inode *inode,
+				      struct buffer_head *group_bh,
+				      u32 bits_wanted, u32 min_bits,
+				      u64 max_block,
+				      struct ocfs2_suballoc_result *res);
+static int ocfs2_block_group_search(struct inode *inode,
+				    struct buffer_head *group_bh,
+				    u32 bits_wanted, u32 min_bits,
+				    u64 max_block,
+				    struct ocfs2_suballoc_result *res);
+static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
+				     handle_t *handle,
+				     u32 bits_wanted,
+				     u32 min_bits,
+				     struct ocfs2_suballoc_result *res);
+static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
+					 int nr);
+static int ocfs2_relink_block_group(handle_t *handle,
+				    struct inode *alloc_inode,
+				    struct buffer_head *fe_bh,
+				    struct buffer_head *bg_bh,
+				    struct buffer_head *prev_bg_bh,
+				    u16 chain);
+static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
+						     u32 wanted);
+static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
+						   u64 bg_blkno,
+						   u16 bg_bit_off);
+static inline void ocfs2_block_to_cluster_group(struct inode *inode,
+						u64 data_blkno,
+						u64 *bg_blkno,
+						u16 *bg_bit_off);
+static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
+					     u32 bits_wanted, u64 max_block,
+					     int flags,
+					     struct ocfs2_alloc_context **ac);
+
+void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac)
+{
+	struct inode *inode = ac->ac_inode;
+
+	if (inode) {
+		if (ac->ac_which != OCFS2_AC_USE_LOCAL)
+			ocfs2_inode_unlock(inode, 1);
+
+		mutex_unlock(&inode->i_mutex);
+
+		iput(inode);
+		ac->ac_inode = NULL;
+	}
+	brelse(ac->ac_bh);
+	ac->ac_bh = NULL;
+	ac->ac_resv = NULL;
+	if (ac->ac_find_loc_priv) {
+		kfree(ac->ac_find_loc_priv);
+		ac->ac_find_loc_priv = NULL;
+	}
+}
+
+void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac)
+{
+	ocfs2_free_ac_resource(ac);
+	kfree(ac);
+}
+
+static u32 ocfs2_bits_per_group(struct ocfs2_chain_list *cl)
+{
+	return (u32)le16_to_cpu(cl->cl_cpg) * (u32)le16_to_cpu(cl->cl_bpc);
+}
+
+#define do_error(fmt, ...)						\
+	do{								\
+		if (resize)					\
+			mlog(ML_ERROR, fmt "\n", ##__VA_ARGS__);	\
+		else							\
+			ocfs2_error(sb, fmt, ##__VA_ARGS__);		\
+	} while (0)
+
+static int ocfs2_validate_gd_self(struct super_block *sb,
+				  struct buffer_head *bh,
+				  int resize)
+{
+	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
+
+	if (!OCFS2_IS_VALID_GROUP_DESC(gd)) {
+		do_error("Group descriptor #%llu has bad signature %.*s",
+			 (unsigned long long)bh->b_blocknr, 7,
+			 gd->bg_signature);
+		return -EINVAL;
+	}
+
+	if (le64_to_cpu(gd->bg_blkno) != bh->b_blocknr) {
+		do_error("Group descriptor #%llu has an invalid bg_blkno "
+			 "of %llu",
+			 (unsigned long long)bh->b_blocknr,
+			 (unsigned long long)le64_to_cpu(gd->bg_blkno));
+		return -EINVAL;
+	}
+
+	if (le32_to_cpu(gd->bg_generation) != OCFS2_SB(sb)->fs_generation) {
+		do_error("Group descriptor #%llu has an invalid "
+			 "fs_generation of #%u",
+			 (unsigned long long)bh->b_blocknr,
+			 le32_to_cpu(gd->bg_generation));
+		return -EINVAL;
+	}
+
+	if (le16_to_cpu(gd->bg_free_bits_count) > le16_to_cpu(gd->bg_bits)) {
+		do_error("Group descriptor #%llu has bit count %u but "
+			 "claims that %u are free",
+			 (unsigned long long)bh->b_blocknr,
+			 le16_to_cpu(gd->bg_bits),
+			 le16_to_cpu(gd->bg_free_bits_count));
+		return -EINVAL;
+	}
+
+	if (le16_to_cpu(gd->bg_bits) > (8 * le16_to_cpu(gd->bg_size))) {
+		do_error("Group descriptor #%llu has bit count %u but "
+			 "max bitmap bits of %u",
+			 (unsigned long long)bh->b_blocknr,
+			 le16_to_cpu(gd->bg_bits),
+			 8 * le16_to_cpu(gd->bg_size));
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int ocfs2_validate_gd_parent(struct super_block *sb,
+				    struct ocfs2_dinode *di,
+				    struct buffer_head *bh,
+				    int resize)
+{
+	unsigned int max_bits;
+	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
+
+	if (di->i_blkno != gd->bg_parent_dinode) {
+		do_error("Group descriptor #%llu has bad parent "
+			 "pointer (%llu, expected %llu)",
+			 (unsigned long long)bh->b_blocknr,
+			 (unsigned long long)le64_to_cpu(gd->bg_parent_dinode),
+			 (unsigned long long)le64_to_cpu(di->i_blkno));
+		return -EINVAL;
+	}
+
+	max_bits = le16_to_cpu(di->id2.i_chain.cl_cpg) * le16_to_cpu(di->id2.i_chain.cl_bpc);
+	if (le16_to_cpu(gd->bg_bits) > max_bits) {
+		do_error("Group descriptor #%llu has bit count of %u",
+			 (unsigned long long)bh->b_blocknr,
+			 le16_to_cpu(gd->bg_bits));
+		return -EINVAL;
+	}
+
+	/* In resize, we may meet the case bg_chain == cl_next_free_rec. */
+	if ((le16_to_cpu(gd->bg_chain) >
+	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) ||
+	    ((le16_to_cpu(gd->bg_chain) ==
+	     le16_to_cpu(di->id2.i_chain.cl_next_free_rec)) && !resize)) {
+		do_error("Group descriptor #%llu has bad chain %u",
+			 (unsigned long long)bh->b_blocknr,
+			 le16_to_cpu(gd->bg_chain));
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+#undef do_error
+
+/*
+ * This version only prints errors.  It does not fail the filesystem, and
+ * exists only for resize.
+ */
+int ocfs2_check_group_descriptor(struct super_block *sb,
+				 struct ocfs2_dinode *di,
+				 struct buffer_head *bh)
+{
+	int rc;
+	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	/*
+	 * If the ecc fails, we return the error but otherwise
+	 * leave the filesystem running.  We know any error is
+	 * local to this block.
+	 */
+	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
+	if (rc) {
+		mlog(ML_ERROR,
+		     "Checksum failed for group descriptor %llu\n",
+		     (unsigned long long)bh->b_blocknr);
+	} else
+		rc = ocfs2_validate_gd_self(sb, bh, 1);
+	if (!rc)
+		rc = ocfs2_validate_gd_parent(sb, di, bh, 1);
+
+	return rc;
+}
+
+static int ocfs2_validate_group_descriptor(struct super_block *sb,
+					   struct buffer_head *bh)
+{
+	int rc;
+	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *)bh->b_data;
+
+	trace_ocfs2_validate_group_descriptor(
+					(unsigned long long)bh->b_blocknr);
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	/*
+	 * If the ecc fails, we return the error but otherwise
+	 * leave the filesystem running.  We know any error is
+	 * local to this block.
+	 */
+	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &gd->bg_check);
+	if (rc)
+		return rc;
+
+	/*
+	 * Errors after here are fatal.
+	 */
+
+	return ocfs2_validate_gd_self(sb, bh, 0);
+}
+
+int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
+				u64 gd_blkno, struct buffer_head **bh)
+{
+	int rc;
+	struct buffer_head *tmp = *bh;
+
+	rc = ocfs2_read_block(INODE_CACHE(inode), gd_blkno, &tmp,
+			      ocfs2_validate_group_descriptor);
+	if (rc)
+		goto out;
+
+	rc = ocfs2_validate_gd_parent(inode->i_sb, di, tmp, 0);
+	if (rc) {
+		brelse(tmp);
+		goto out;
+	}
+
+	/* If ocfs2_read_block() got us a new bh, pass it up. */
+	if (!*bh)
+		*bh = tmp;
+
+out:
+	return rc;
+}
+
+static void ocfs2_bg_discontig_add_extent(struct ocfs2_super *osb,
+					  struct ocfs2_group_desc *bg,
+					  struct ocfs2_chain_list *cl,
+					  u64 p_blkno, unsigned int clusters)
+{
+	struct ocfs2_extent_list *el = &bg->bg_list;
+	struct ocfs2_extent_rec *rec;
+
+	BUG_ON(!ocfs2_supports_discontig_bg(osb));
+	if (!el->l_next_free_rec)
+		el->l_count = cpu_to_le16(ocfs2_extent_recs_per_gd(osb->sb));
+	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec)];
+	rec->e_blkno = cpu_to_le64(p_blkno);
+	rec->e_cpos = cpu_to_le32(le16_to_cpu(bg->bg_bits) /
+				  le16_to_cpu(cl->cl_bpc));
+	rec->e_leaf_clusters = cpu_to_le16(clusters);
+	le16_add_cpu(&bg->bg_bits, clusters * le16_to_cpu(cl->cl_bpc));
+	le16_add_cpu(&bg->bg_free_bits_count,
+		     clusters * le16_to_cpu(cl->cl_bpc));
+	le16_add_cpu(&el->l_next_free_rec, 1);
+}
+
+static int ocfs2_block_group_fill(handle_t *handle,
+				  struct inode *alloc_inode,
+				  struct buffer_head *bg_bh,
+				  u64 group_blkno,
+				  unsigned int group_clusters,
+				  u16 my_chain,
+				  struct ocfs2_chain_list *cl)
+{
+	int status = 0;
+	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
+	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
+	struct super_block * sb = alloc_inode->i_sb;
+
+	if (((unsigned long long) bg_bh->b_blocknr) != group_blkno) {
+		ocfs2_error(alloc_inode->i_sb, "group block (%llu) != "
+			    "b_blocknr (%llu)",
+			    (unsigned long long)group_blkno,
+			    (unsigned long long) bg_bh->b_blocknr);
+		status = -EIO;
+		goto bail;
+	}
+
+	status = ocfs2_journal_access_gd(handle,
+					 INODE_CACHE(alloc_inode),
+					 bg_bh,
+					 OCFS2_JOURNAL_ACCESS_CREATE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	memset(bg, 0, sb->s_blocksize);
+	strcpy(bg->bg_signature, OCFS2_GROUP_DESC_SIGNATURE);
+	bg->bg_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
+	bg->bg_size = cpu_to_le16(ocfs2_group_bitmap_size(sb, 1,
+						osb->s_feature_incompat));
+	bg->bg_chain = cpu_to_le16(my_chain);
+	bg->bg_next_group = cl->cl_recs[my_chain].c_blkno;
+	bg->bg_parent_dinode = cpu_to_le64(OCFS2_I(alloc_inode)->ip_blkno);
+	bg->bg_blkno = cpu_to_le64(group_blkno);
+	if (group_clusters == le16_to_cpu(cl->cl_cpg))
+		bg->bg_bits = cpu_to_le16(ocfs2_bits_per_group(cl));
+	else
+		ocfs2_bg_discontig_add_extent(osb, bg, cl, group_blkno,
+					      group_clusters);
+
+	/* set the 1st bit in the bitmap to account for the descriptor block */
+	ocfs2_set_bit(0, (unsigned long *)bg->bg_bitmap);
+	bg->bg_free_bits_count = cpu_to_le16(le16_to_cpu(bg->bg_bits) - 1);
+
+	ocfs2_journal_dirty(handle, bg_bh);
+
+	/* There is no need to zero out or otherwise initialize the
+	 * other blocks in a group - All valid FS metadata in a block
+	 * group stores the superblock fs_generation value at
+	 * allocation time. */
+
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static inline u16 ocfs2_find_smallest_chain(struct ocfs2_chain_list *cl)
+{
+	u16 curr, best;
+
+	best = curr = 0;
+	while (curr < le16_to_cpu(cl->cl_count)) {
+		if (le32_to_cpu(cl->cl_recs[best].c_total) >
+		    le32_to_cpu(cl->cl_recs[curr].c_total))
+			best = curr;
+		curr++;
+	}
+	return best;
+}
+
+static struct buffer_head *
+ocfs2_block_group_alloc_contig(struct ocfs2_super *osb, handle_t *handle,
+			       struct inode *alloc_inode,
+			       struct ocfs2_alloc_context *ac,
+			       struct ocfs2_chain_list *cl)
+{
+	int status;
+	u32 bit_off, num_bits;
+	u64 bg_blkno;
+	struct buffer_head *bg_bh;
+	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
+
+	status = ocfs2_claim_clusters(handle, ac,
+				      le16_to_cpu(cl->cl_cpg), &bit_off,
+				      &num_bits);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	/* setup the group */
+	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
+	trace_ocfs2_block_group_alloc_contig(
+	     (unsigned long long)bg_blkno, alloc_rec);
+
+	bg_bh = sb_getblk(osb->sb, bg_blkno);
+	if (!bg_bh) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
+
+	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
+					bg_blkno, num_bits, alloc_rec, cl);
+	if (status < 0) {
+		brelse(bg_bh);
+		mlog_errno(status);
+	}
+
+bail:
+	return status ? ERR_PTR(status) : bg_bh;
+}
+
+static int ocfs2_block_group_claim_bits(struct ocfs2_super *osb,
+					handle_t *handle,
+					struct ocfs2_alloc_context *ac,
+					unsigned int min_bits,
+					u32 *bit_off, u32 *num_bits)
+{
+	int status = 0;
+
+	while (min_bits) {
+		status = ocfs2_claim_clusters(handle, ac, min_bits,
+					      bit_off, num_bits);
+		if (status != -ENOSPC)
+			break;
+
+		min_bits >>= 1;
+	}
+
+	return status;
+}
+
+static int ocfs2_block_group_grow_discontig(handle_t *handle,
+					    struct inode *alloc_inode,
+					    struct buffer_head *bg_bh,
+					    struct ocfs2_alloc_context *ac,
+					    struct ocfs2_chain_list *cl,
+					    unsigned int min_bits)
+{
+	int status;
+	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
+	struct ocfs2_group_desc *bg =
+		(struct ocfs2_group_desc *)bg_bh->b_data;
+	unsigned int needed = le16_to_cpu(cl->cl_cpg) -
+			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
+	u32 p_cpos, clusters;
+	u64 p_blkno;
+	struct ocfs2_extent_list *el = &bg->bg_list;
+
+	status = ocfs2_journal_access_gd(handle,
+					 INODE_CACHE(alloc_inode),
+					 bg_bh,
+					 OCFS2_JOURNAL_ACCESS_CREATE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	while ((needed > 0) && (le16_to_cpu(el->l_next_free_rec) <
+				le16_to_cpu(el->l_count))) {
+		if (min_bits > needed)
+			min_bits = needed;
+		status = ocfs2_block_group_claim_bits(osb, handle, ac,
+						      min_bits, &p_cpos,
+						      &clusters);
+		if (status < 0) {
+			if (status != -ENOSPC)
+				mlog_errno(status);
+			goto bail;
+		}
+		p_blkno = ocfs2_clusters_to_blocks(osb->sb, p_cpos);
+		ocfs2_bg_discontig_add_extent(osb, bg, cl, p_blkno,
+					      clusters);
+
+		min_bits = clusters;
+		needed = le16_to_cpu(cl->cl_cpg) -
+			 le16_to_cpu(bg->bg_bits) / le16_to_cpu(cl->cl_bpc);
+	}
+
+	if (needed > 0) {
+		/*
+		 * We have used up all the extent rec but can't fill up
+		 * the cpg. So bail out.
+		 */
+		status = -ENOSPC;
+		goto bail;
+	}
+
+	ocfs2_journal_dirty(handle, bg_bh);
+
+bail:
+	return status;
+}
+
+static void ocfs2_bg_alloc_cleanup(handle_t *handle,
+				   struct ocfs2_alloc_context *cluster_ac,
+				   struct inode *alloc_inode,
+				   struct buffer_head *bg_bh)
+{
+	int i, ret;
+	struct ocfs2_group_desc *bg;
+	struct ocfs2_extent_list *el;
+	struct ocfs2_extent_rec *rec;
+
+	if (!bg_bh)
+		return;
+
+	bg = (struct ocfs2_group_desc *)bg_bh->b_data;
+	el = &bg->bg_list;
+	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
+		rec = &el->l_recs[i];
+		ret = ocfs2_free_clusters(handle, cluster_ac->ac_inode,
+					  cluster_ac->ac_bh,
+					  le64_to_cpu(rec->e_blkno),
+					  le16_to_cpu(rec->e_leaf_clusters));
+		if (ret)
+			mlog_errno(ret);
+		/* Try all the clusters to free */
+	}
+
+	ocfs2_remove_from_cache(INODE_CACHE(alloc_inode), bg_bh);
+	brelse(bg_bh);
+}
+
+static struct buffer_head *
+ocfs2_block_group_alloc_discontig(handle_t *handle,
+				  struct inode *alloc_inode,
+				  struct ocfs2_alloc_context *ac,
+				  struct ocfs2_chain_list *cl)
+{
+	int status;
+	u32 bit_off, num_bits;
+	u64 bg_blkno;
+	unsigned int min_bits = le16_to_cpu(cl->cl_cpg) >> 1;
+	struct buffer_head *bg_bh = NULL;
+	unsigned int alloc_rec = ocfs2_find_smallest_chain(cl);
+	struct ocfs2_super *osb = OCFS2_SB(alloc_inode->i_sb);
+
+	if (!ocfs2_supports_discontig_bg(osb)) {
+		status = -ENOSPC;
+		goto bail;
+	}
+
+	status = ocfs2_extend_trans(handle,
+				    ocfs2_calc_bg_discontig_credits(osb->sb));
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/*
+	 * We're going to be grabbing from multiple cluster groups.
+	 * We don't have enough credits to relink them all, and the
+	 * cluster groups will be staying in cache for the duration of
+	 * this operation.
+	 */
+	ac->ac_disable_chain_relink = 1;
+
+	/* Claim the first region */
+	status = ocfs2_block_group_claim_bits(osb, handle, ac, min_bits,
+					      &bit_off, &num_bits);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+	min_bits = num_bits;
+
+	/* setup the group */
+	bg_blkno = ocfs2_clusters_to_blocks(osb->sb, bit_off);
+	trace_ocfs2_block_group_alloc_discontig(
+				(unsigned long long)bg_blkno, alloc_rec);
+
+	bg_bh = sb_getblk(osb->sb, bg_blkno);
+	if (!bg_bh) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(alloc_inode), bg_bh);
+
+	status = ocfs2_block_group_fill(handle, alloc_inode, bg_bh,
+					bg_blkno, num_bits, alloc_rec, cl);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_block_group_grow_discontig(handle, alloc_inode,
+						  bg_bh, ac, cl, min_bits);
+	if (status)
+		mlog_errno(status);
+
+bail:
+	if (status)
+		ocfs2_bg_alloc_cleanup(handle, ac, alloc_inode, bg_bh);
+	return status ? ERR_PTR(status) : bg_bh;
+}
+
+/*
+ * We expect the block group allocator to already be locked.
+ */
+static int ocfs2_block_group_alloc(struct ocfs2_super *osb,
+				   struct inode *alloc_inode,
+				   struct buffer_head *bh,
+				   u64 max_block,
+				   u64 *last_alloc_group,
+				   int flags)
+{
+	int status, credits;
+	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) bh->b_data;
+	struct ocfs2_chain_list *cl;
+	struct ocfs2_alloc_context *ac = NULL;
+	handle_t *handle = NULL;
+	u16 alloc_rec;
+	struct buffer_head *bg_bh = NULL;
+	struct ocfs2_group_desc *bg;
+
+	BUG_ON(ocfs2_is_cluster_bitmap(alloc_inode));
+
+	cl = &fe->id2.i_chain;
+	status = ocfs2_reserve_clusters_with_limit(osb,
+						   le16_to_cpu(cl->cl_cpg),
+						   max_block, flags, &ac);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	credits = ocfs2_calc_group_alloc_credits(osb->sb,
+						 le16_to_cpu(cl->cl_cpg));
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		status = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	if (last_alloc_group && *last_alloc_group != 0) {
+		trace_ocfs2_block_group_alloc(
+				(unsigned long long)*last_alloc_group);
+		ac->ac_last_group = *last_alloc_group;
+	}
+
+	bg_bh = ocfs2_block_group_alloc_contig(osb, handle, alloc_inode,
+					       ac, cl);
+	if (IS_ERR(bg_bh) && (PTR_ERR(bg_bh) == -ENOSPC))
+		bg_bh = ocfs2_block_group_alloc_discontig(handle,
+							  alloc_inode,
+							  ac, cl);
+	if (IS_ERR(bg_bh)) {
+		status = PTR_ERR(bg_bh);
+		bg_bh = NULL;
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
+					 bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	alloc_rec = le16_to_cpu(bg->bg_chain);
+	le32_add_cpu(&cl->cl_recs[alloc_rec].c_free,
+		     le16_to_cpu(bg->bg_free_bits_count));
+	le32_add_cpu(&cl->cl_recs[alloc_rec].c_total,
+		     le16_to_cpu(bg->bg_bits));
+	cl->cl_recs[alloc_rec].c_blkno = bg->bg_blkno;
+	if (le16_to_cpu(cl->cl_next_free_rec) < le16_to_cpu(cl->cl_count))
+		le16_add_cpu(&cl->cl_next_free_rec, 1);
+
+	le32_add_cpu(&fe->id1.bitmap1.i_used, le16_to_cpu(bg->bg_bits) -
+					le16_to_cpu(bg->bg_free_bits_count));
+	le32_add_cpu(&fe->id1.bitmap1.i_total, le16_to_cpu(bg->bg_bits));
+	le32_add_cpu(&fe->i_clusters, le16_to_cpu(cl->cl_cpg));
+
+	ocfs2_journal_dirty(handle, bh);
+
+	spin_lock(&OCFS2_I(alloc_inode)->ip_lock);
+	OCFS2_I(alloc_inode)->ip_clusters = le32_to_cpu(fe->i_clusters);
+	fe->i_size = cpu_to_le64(ocfs2_clusters_to_bytes(alloc_inode->i_sb,
+					     le32_to_cpu(fe->i_clusters)));
+	spin_unlock(&OCFS2_I(alloc_inode)->ip_lock);
+	i_size_write(alloc_inode, le64_to_cpu(fe->i_size));
+	alloc_inode->i_blocks = ocfs2_inode_sector_count(alloc_inode);
+	ocfs2_update_inode_fsync_trans(handle, alloc_inode, 0);
+
+	status = 0;
+
+	/* save the new last alloc group so that the caller can cache it. */
+	if (last_alloc_group)
+		*last_alloc_group = ac->ac_last_group;
+
+bail:
+	if (handle)
+		ocfs2_commit_trans(osb, handle);
+
+	if (ac)
+		ocfs2_free_alloc_context(ac);
+
+	brelse(bg_bh);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static int ocfs2_reserve_suballoc_bits(struct ocfs2_super *osb,
+				       struct ocfs2_alloc_context *ac,
+				       int type,
+				       u32 slot,
+				       u64 *last_alloc_group,
+				       int flags)
+{
+	int status;
+	u32 bits_wanted = ac->ac_bits_wanted;
+	struct inode *alloc_inode;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_dinode *fe;
+	u32 free_bits;
+
+	alloc_inode = ocfs2_get_system_file_inode(osb, type, slot);
+	if (!alloc_inode) {
+		mlog_errno(-EINVAL);
+		return -EINVAL;
+	}
+
+	mutex_lock(&alloc_inode->i_mutex);
+
+	status = ocfs2_inode_lock(alloc_inode, &bh, 1);
+	if (status < 0) {
+		mutex_unlock(&alloc_inode->i_mutex);
+		iput(alloc_inode);
+
+		mlog_errno(status);
+		return status;
+	}
+
+	ac->ac_inode = alloc_inode;
+	ac->ac_alloc_slot = slot;
+
+	fe = (struct ocfs2_dinode *) bh->b_data;
+
+	/* The bh was validated by the inode read inside
+	 * ocfs2_inode_lock().  Any corruption is a code bug. */
+	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
+
+	if (!(fe->i_flags & cpu_to_le32(OCFS2_CHAIN_FL))) {
+		ocfs2_error(alloc_inode->i_sb, "Invalid chain allocator %llu",
+			    (unsigned long long)le64_to_cpu(fe->i_blkno));
+		status = -EIO;
+		goto bail;
+	}
+
+	free_bits = le32_to_cpu(fe->id1.bitmap1.i_total) -
+		le32_to_cpu(fe->id1.bitmap1.i_used);
+
+	if (bits_wanted > free_bits) {
+		/* cluster bitmap never grows */
+		if (ocfs2_is_cluster_bitmap(alloc_inode)) {
+			trace_ocfs2_reserve_suballoc_bits_nospc(bits_wanted,
+								free_bits);
+			status = -ENOSPC;
+			goto bail;
+		}
+
+		if (!(flags & ALLOC_NEW_GROUP)) {
+			trace_ocfs2_reserve_suballoc_bits_no_new_group(
+						slot, bits_wanted, free_bits);
+			status = -ENOSPC;
+			goto bail;
+		}
+
+		status = ocfs2_block_group_alloc(osb, alloc_inode, bh,
+						 ac->ac_max_block,
+						 last_alloc_group, flags);
+		if (status < 0) {
+			if (status != -ENOSPC)
+				mlog_errno(status);
+			goto bail;
+		}
+		atomic_inc(&osb->alloc_stats.bg_extends);
+
+		/* You should never ask for this much metadata */
+		BUG_ON(bits_wanted >
+		       (le32_to_cpu(fe->id1.bitmap1.i_total)
+			- le32_to_cpu(fe->id1.bitmap1.i_used)));
+	}
+
+	get_bh(bh);
+	ac->ac_bh = bh;
+bail:
+	brelse(bh);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static void ocfs2_init_inode_steal_slot(struct ocfs2_super *osb)
+{
+	spin_lock(&osb->osb_lock);
+	osb->s_inode_steal_slot = OCFS2_INVALID_SLOT;
+	spin_unlock(&osb->osb_lock);
+	atomic_set(&osb->s_num_inodes_stolen, 0);
+}
+
+static void ocfs2_init_meta_steal_slot(struct ocfs2_super *osb)
+{
+	spin_lock(&osb->osb_lock);
+	osb->s_meta_steal_slot = OCFS2_INVALID_SLOT;
+	spin_unlock(&osb->osb_lock);
+	atomic_set(&osb->s_num_meta_stolen, 0);
+}
+
+void ocfs2_init_steal_slots(struct ocfs2_super *osb)
+{
+	ocfs2_init_inode_steal_slot(osb);
+	ocfs2_init_meta_steal_slot(osb);
+}
+
+static void __ocfs2_set_steal_slot(struct ocfs2_super *osb, int slot, int type)
+{
+	spin_lock(&osb->osb_lock);
+	if (type == INODE_ALLOC_SYSTEM_INODE)
+		osb->s_inode_steal_slot = slot;
+	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
+		osb->s_meta_steal_slot = slot;
+	spin_unlock(&osb->osb_lock);
+}
+
+static int __ocfs2_get_steal_slot(struct ocfs2_super *osb, int type)
+{
+	int slot = OCFS2_INVALID_SLOT;
+
+	spin_lock(&osb->osb_lock);
+	if (type == INODE_ALLOC_SYSTEM_INODE)
+		slot = osb->s_inode_steal_slot;
+	else if (type == EXTENT_ALLOC_SYSTEM_INODE)
+		slot = osb->s_meta_steal_slot;
+	spin_unlock(&osb->osb_lock);
+
+	return slot;
+}
+
+static int ocfs2_get_inode_steal_slot(struct ocfs2_super *osb)
+{
+	return __ocfs2_get_steal_slot(osb, INODE_ALLOC_SYSTEM_INODE);
+}
+
+static int ocfs2_get_meta_steal_slot(struct ocfs2_super *osb)
+{
+	return __ocfs2_get_steal_slot(osb, EXTENT_ALLOC_SYSTEM_INODE);
+}
+
+static int ocfs2_steal_resource(struct ocfs2_super *osb,
+				struct ocfs2_alloc_context *ac,
+				int type)
+{
+	int i, status = -ENOSPC;
+	int slot = __ocfs2_get_steal_slot(osb, type);
+
+	/* Start to steal resource from the first slot after ours. */
+	if (slot == OCFS2_INVALID_SLOT)
+		slot = osb->slot_num + 1;
+
+	for (i = 0; i < osb->max_slots; i++, slot++) {
+		if (slot == osb->max_slots)
+			slot = 0;
+
+		if (slot == osb->slot_num)
+			continue;
+
+		status = ocfs2_reserve_suballoc_bits(osb, ac,
+						     type,
+						     (u32)slot, NULL,
+						     NOT_ALLOC_NEW_GROUP);
+		if (status >= 0) {
+			__ocfs2_set_steal_slot(osb, slot, type);
+			break;
+		}
+
+		ocfs2_free_ac_resource(ac);
+	}
+
+	return status;
+}
+
+static int ocfs2_steal_inode(struct ocfs2_super *osb,
+			     struct ocfs2_alloc_context *ac)
+{
+	return ocfs2_steal_resource(osb, ac, INODE_ALLOC_SYSTEM_INODE);
+}
+
+static int ocfs2_steal_meta(struct ocfs2_super *osb,
+			    struct ocfs2_alloc_context *ac)
+{
+	return ocfs2_steal_resource(osb, ac, EXTENT_ALLOC_SYSTEM_INODE);
+}
+
+int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
+				      int blocks,
+				      struct ocfs2_alloc_context **ac)
+{
+	int status;
+	int slot = ocfs2_get_meta_steal_slot(osb);
+
+	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
+	if (!(*ac)) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	(*ac)->ac_bits_wanted = blocks;
+	(*ac)->ac_which = OCFS2_AC_USE_META;
+	(*ac)->ac_group_search = ocfs2_block_group_search;
+
+	if (slot != OCFS2_INVALID_SLOT &&
+		atomic_read(&osb->s_num_meta_stolen) < OCFS2_MAX_TO_STEAL)
+		goto extent_steal;
+
+	atomic_set(&osb->s_num_meta_stolen, 0);
+	status = ocfs2_reserve_suballoc_bits(osb, (*ac),
+					     EXTENT_ALLOC_SYSTEM_INODE,
+					     (u32)osb->slot_num, NULL,
+					     ALLOC_GROUPS_FROM_GLOBAL|ALLOC_NEW_GROUP);
+
+
+	if (status >= 0) {
+		status = 0;
+		if (slot != OCFS2_INVALID_SLOT)
+			ocfs2_init_meta_steal_slot(osb);
+		goto bail;
+	} else if (status < 0 && status != -ENOSPC) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	ocfs2_free_ac_resource(*ac);
+
+extent_steal:
+	status = ocfs2_steal_meta(osb, *ac);
+	atomic_inc(&osb->s_num_meta_stolen);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	status = 0;
+bail:
+	if ((status < 0) && *ac) {
+		ocfs2_free_alloc_context(*ac);
+		*ac = NULL;
+	}
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
+			       struct ocfs2_extent_list *root_el,
+			       struct ocfs2_alloc_context **ac)
+{
+	return ocfs2_reserve_new_metadata_blocks(osb,
+					ocfs2_extend_meta_needed(root_el),
+					ac);
+}
+
+int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
+			    struct ocfs2_alloc_context **ac)
+{
+	int status;
+	int slot = ocfs2_get_inode_steal_slot(osb);
+	u64 alloc_group;
+
+	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
+	if (!(*ac)) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	(*ac)->ac_bits_wanted = 1;
+	(*ac)->ac_which = OCFS2_AC_USE_INODE;
+
+	(*ac)->ac_group_search = ocfs2_block_group_search;
+
+	/*
+	 * stat(2) can't handle i_ino > 32bits, so we tell the
+	 * lower levels not to allocate us a block group past that
+	 * limit.  The 'inode64' mount option avoids this behavior.
+	 */
+	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64))
+		(*ac)->ac_max_block = (u32)~0U;
+
+	/*
+	 * slot is set when we successfully steal inode from other nodes.
+	 * It is reset in 3 places:
+	 * 1. when we flush the truncate log
+	 * 2. when we complete local alloc recovery.
+	 * 3. when we successfully allocate from our own slot.
+	 * After it is set, we will go on stealing inodes until we find the
+	 * need to check our slots to see whether there is some space for us.
+	 */
+	if (slot != OCFS2_INVALID_SLOT &&
+	    atomic_read(&osb->s_num_inodes_stolen) < OCFS2_MAX_TO_STEAL)
+		goto inode_steal;
+
+	atomic_set(&osb->s_num_inodes_stolen, 0);
+	alloc_group = osb->osb_inode_alloc_group;
+	status = ocfs2_reserve_suballoc_bits(osb, *ac,
+					     INODE_ALLOC_SYSTEM_INODE,
+					     (u32)osb->slot_num,
+					     &alloc_group,
+					     ALLOC_NEW_GROUP |
+					     ALLOC_GROUPS_FROM_GLOBAL);
+	if (status >= 0) {
+		status = 0;
+
+		spin_lock(&osb->osb_lock);
+		osb->osb_inode_alloc_group = alloc_group;
+		spin_unlock(&osb->osb_lock);
+		trace_ocfs2_reserve_new_inode_new_group(
+			(unsigned long long)alloc_group);
+
+		/*
+		 * Some inodes must be freed by us, so try to allocate
+		 * from our own next time.
+		 */
+		if (slot != OCFS2_INVALID_SLOT)
+			ocfs2_init_inode_steal_slot(osb);
+		goto bail;
+	} else if (status < 0 && status != -ENOSPC) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	ocfs2_free_ac_resource(*ac);
+
+inode_steal:
+	status = ocfs2_steal_inode(osb, *ac);
+	atomic_inc(&osb->s_num_inodes_stolen);
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	status = 0;
+bail:
+	if ((status < 0) && *ac) {
+		ocfs2_free_alloc_context(*ac);
+		*ac = NULL;
+	}
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/* local alloc code has to do the same thing, so rather than do this
+ * twice.. */
+int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
+				      struct ocfs2_alloc_context *ac)
+{
+	int status;
+
+	ac->ac_which = OCFS2_AC_USE_MAIN;
+	ac->ac_group_search = ocfs2_cluster_group_search;
+
+	status = ocfs2_reserve_suballoc_bits(osb, ac,
+					     GLOBAL_BITMAP_SYSTEM_INODE,
+					     OCFS2_INVALID_SLOT, NULL,
+					     ALLOC_NEW_GROUP);
+	if (status < 0 && status != -ENOSPC) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+bail:
+	return status;
+}
+
+/* Callers don't need to care which bitmap (local alloc or main) to
+ * use so we figure it out for them, but unfortunately this clutters
+ * things a bit. */
+static int ocfs2_reserve_clusters_with_limit(struct ocfs2_super *osb,
+					     u32 bits_wanted, u64 max_block,
+					     int flags,
+					     struct ocfs2_alloc_context **ac)
+{
+	int status;
+
+	*ac = kzalloc(sizeof(struct ocfs2_alloc_context), GFP_KERNEL);
+	if (!(*ac)) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	(*ac)->ac_bits_wanted = bits_wanted;
+	(*ac)->ac_max_block = max_block;
+
+	status = -ENOSPC;
+	if (!(flags & ALLOC_GROUPS_FROM_GLOBAL) &&
+	    ocfs2_alloc_should_use_local(osb, bits_wanted)) {
+		status = ocfs2_reserve_local_alloc_bits(osb,
+							bits_wanted,
+							*ac);
+		if ((status < 0) && (status != -ENOSPC)) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	if (status == -ENOSPC) {
+		status = ocfs2_reserve_cluster_bitmap_bits(osb, *ac);
+		if (status < 0) {
+			if (status != -ENOSPC)
+				mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	status = 0;
+bail:
+	if ((status < 0) && *ac) {
+		ocfs2_free_alloc_context(*ac);
+		*ac = NULL;
+	}
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+int ocfs2_reserve_clusters(struct ocfs2_super *osb,
+			   u32 bits_wanted,
+			   struct ocfs2_alloc_context **ac)
+{
+	return ocfs2_reserve_clusters_with_limit(osb, bits_wanted, 0,
+						 ALLOC_NEW_GROUP, ac);
+}
+
+/*
+ * More or less lifted from ext3. I'll leave their description below:
+ *
+ * "For ext3 allocations, we must not reuse any blocks which are
+ * allocated in the bitmap buffer's "last committed data" copy.  This
+ * prevents deletes from freeing up the page for reuse until we have
+ * committed the delete transaction.
+ *
+ * If we didn't do this, then deleting something and reallocating it as
+ * data would allow the old block to be overwritten before the
+ * transaction committed (because we force data to disk before commit).
+ * This would lead to corruption if we crashed between overwriting the
+ * data and committing the delete.
+ *
+ * @@@ We may want to make this allocation behaviour conditional on
+ * data-writes at some point, and disable it for metadata allocations or
+ * sync-data inodes."
+ *
+ * Note: OCFS2 already does this differently for metadata vs data
+ * allocations, as those bitmaps are separate and undo access is never
+ * called on a metadata group descriptor.
+ */
+static int ocfs2_test_bg_bit_allocatable(struct buffer_head *bg_bh,
+					 int nr)
+{
+	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
+	int ret;
+
+	if (ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap))
+		return 0;
+
+	if (!buffer_jbd(bg_bh))
+		return 1;
+
+	jbd_lock_bh_state(bg_bh);
+	bg = (struct ocfs2_group_desc *) bh2jh(bg_bh)->b_committed_data;
+	if (bg)
+		ret = !ocfs2_test_bit(nr, (unsigned long *)bg->bg_bitmap);
+	else
+		ret = 1;
+	jbd_unlock_bh_state(bg_bh);
+
+	return ret;
+}
+
+static int ocfs2_block_group_find_clear_bits(struct ocfs2_super *osb,
+					     struct buffer_head *bg_bh,
+					     unsigned int bits_wanted,
+					     unsigned int total_bits,
+					     struct ocfs2_suballoc_result *res)
+{
+	void *bitmap;
+	u16 best_offset, best_size;
+	int offset, start, found, status = 0;
+	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
+
+	/* Callers got this descriptor from
+	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
+	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
+
+	found = start = best_offset = best_size = 0;
+	bitmap = bg->bg_bitmap;
+
+	while((offset = ocfs2_find_next_zero_bit(bitmap, total_bits, start)) != -1) {
+		if (offset == total_bits)
+			break;
+
+		if (!ocfs2_test_bg_bit_allocatable(bg_bh, offset)) {
+			/* We found a zero, but we can't use it as it
+			 * hasn't been put to disk yet! */
+			found = 0;
+			start = offset + 1;
+		} else if (offset == start) {
+			/* we found a zero */
+			found++;
+			/* move start to the next bit to test */
+			start++;
+		} else {
+			/* got a zero after some ones */
+			found = 1;
+			start = offset + 1;
+		}
+		if (found > best_size) {
+			best_size = found;
+			best_offset = start - found;
+		}
+		/* we got everything we needed */
+		if (found == bits_wanted) {
+			/* mlog(0, "Found it all!\n"); */
+			break;
+		}
+	}
+
+	if (best_size) {
+		res->sr_bit_offset = best_offset;
+		res->sr_bits = best_size;
+	} else {
+		status = -ENOSPC;
+		/* No error log here -- see the comment above
+		 * ocfs2_test_bg_bit_allocatable */
+	}
+
+	return status;
+}
+
+int ocfs2_block_group_set_bits(handle_t *handle,
+					     struct inode *alloc_inode,
+					     struct ocfs2_group_desc *bg,
+					     struct buffer_head *group_bh,
+					     unsigned int bit_off,
+					     unsigned int num_bits)
+{
+	int status;
+	void *bitmap = bg->bg_bitmap;
+	int journal_type = OCFS2_JOURNAL_ACCESS_WRITE;
+
+	/* All callers get the descriptor via
+	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
+	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
+	BUG_ON(le16_to_cpu(bg->bg_free_bits_count) < num_bits);
+
+	trace_ocfs2_block_group_set_bits(bit_off, num_bits);
+
+	if (ocfs2_is_cluster_bitmap(alloc_inode))
+		journal_type = OCFS2_JOURNAL_ACCESS_UNDO;
+
+	status = ocfs2_journal_access_gd(handle,
+					 INODE_CACHE(alloc_inode),
+					 group_bh,
+					 journal_type);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	le16_add_cpu(&bg->bg_free_bits_count, -num_bits);
+	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
+		ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
+			    " count %u but claims %u are freed. num_bits %d",
+			    (unsigned long long)le64_to_cpu(bg->bg_blkno),
+			    le16_to_cpu(bg->bg_bits),
+			    le16_to_cpu(bg->bg_free_bits_count), num_bits);
+		return -EROFS;
+	}
+	while(num_bits--)
+		ocfs2_set_bit(bit_off++, bitmap);
+
+	ocfs2_journal_dirty(handle, group_bh);
+
+bail:
+	return status;
+}
+
+/* find the one with the most empty bits */
+static inline u16 ocfs2_find_victim_chain(struct ocfs2_chain_list *cl)
+{
+	u16 curr, best;
+
+	BUG_ON(!cl->cl_next_free_rec);
+
+	best = curr = 0;
+	while (curr < le16_to_cpu(cl->cl_next_free_rec)) {
+		if (le32_to_cpu(cl->cl_recs[curr].c_free) >
+		    le32_to_cpu(cl->cl_recs[best].c_free))
+			best = curr;
+		curr++;
+	}
+
+	BUG_ON(best >= le16_to_cpu(cl->cl_next_free_rec));
+	return best;
+}
+
+static int ocfs2_relink_block_group(handle_t *handle,
+				    struct inode *alloc_inode,
+				    struct buffer_head *fe_bh,
+				    struct buffer_head *bg_bh,
+				    struct buffer_head *prev_bg_bh,
+				    u16 chain)
+{
+	int status;
+	/* there is a really tiny chance the journal calls could fail,
+	 * but we wouldn't want inconsistent blocks in *any* case. */
+	u64 bg_ptr, prev_bg_ptr;
+	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
+	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) bg_bh->b_data;
+	struct ocfs2_group_desc *prev_bg = (struct ocfs2_group_desc *) prev_bg_bh->b_data;
+
+	/* The caller got these descriptors from
+	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
+	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
+	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(prev_bg));
+
+	trace_ocfs2_relink_block_group(
+		(unsigned long long)le64_to_cpu(fe->i_blkno), chain,
+		(unsigned long long)le64_to_cpu(bg->bg_blkno),
+		(unsigned long long)le64_to_cpu(prev_bg->bg_blkno));
+
+	bg_ptr = le64_to_cpu(bg->bg_next_group);
+	prev_bg_ptr = le64_to_cpu(prev_bg->bg_next_group);
+
+	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
+					 prev_bg_bh,
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0)
+		goto out;
+
+	prev_bg->bg_next_group = bg->bg_next_group;
+	ocfs2_journal_dirty(handle, prev_bg_bh);
+
+	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
+					 bg_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0)
+		goto out_rollback_prev_bg;
+
+	bg->bg_next_group = fe->id2.i_chain.cl_recs[chain].c_blkno;
+	ocfs2_journal_dirty(handle, bg_bh);
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
+					 fe_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0)
+		goto out_rollback_bg;
+
+	fe->id2.i_chain.cl_recs[chain].c_blkno = bg->bg_blkno;
+	ocfs2_journal_dirty(handle, fe_bh);
+
+out:
+	if (status < 0)
+		mlog_errno(status);
+	return status;
+
+out_rollback_bg:
+	bg->bg_next_group = cpu_to_le64(bg_ptr);
+out_rollback_prev_bg:
+	prev_bg->bg_next_group = cpu_to_le64(prev_bg_ptr);
+	goto out;
+}
+
+static inline int ocfs2_block_group_reasonably_empty(struct ocfs2_group_desc *bg,
+						     u32 wanted)
+{
+	return le16_to_cpu(bg->bg_free_bits_count) > wanted;
+}
+
+/* return 0 on success, -ENOSPC to keep searching and any other < 0
+ * value on error. */
+static int ocfs2_cluster_group_search(struct inode *inode,
+				      struct buffer_head *group_bh,
+				      u32 bits_wanted, u32 min_bits,
+				      u64 max_block,
+				      struct ocfs2_suballoc_result *res)
+{
+	int search = -ENOSPC;
+	int ret;
+	u64 blkoff;
+	struct ocfs2_group_desc *gd = (struct ocfs2_group_desc *) group_bh->b_data;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	unsigned int max_bits, gd_cluster_off;
+
+	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
+
+	if (gd->bg_free_bits_count) {
+		max_bits = le16_to_cpu(gd->bg_bits);
+
+		/* Tail groups in cluster bitmaps which aren't cpg
+		 * aligned are prone to partial extension by a failed
+		 * fs resize. If the file system resize never got to
+		 * update the dinode cluster count, then we don't want
+		 * to trust any clusters past it, regardless of what
+		 * the group descriptor says. */
+		gd_cluster_off = ocfs2_blocks_to_clusters(inode->i_sb,
+							  le64_to_cpu(gd->bg_blkno));
+		if ((gd_cluster_off + max_bits) >
+		    OCFS2_I(inode)->ip_clusters) {
+			max_bits = OCFS2_I(inode)->ip_clusters - gd_cluster_off;
+			trace_ocfs2_cluster_group_search_wrong_max_bits(
+				(unsigned long long)le64_to_cpu(gd->bg_blkno),
+				le16_to_cpu(gd->bg_bits),
+				OCFS2_I(inode)->ip_clusters, max_bits);
+		}
+
+		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
+							group_bh, bits_wanted,
+							max_bits, res);
+		if (ret)
+			return ret;
+
+		if (max_block) {
+			blkoff = ocfs2_clusters_to_blocks(inode->i_sb,
+							  gd_cluster_off +
+							  res->sr_bit_offset +
+							  res->sr_bits);
+			trace_ocfs2_cluster_group_search_max_block(
+				(unsigned long long)blkoff,
+				(unsigned long long)max_block);
+			if (blkoff > max_block)
+				return -ENOSPC;
+		}
+
+		/* ocfs2_block_group_find_clear_bits() might
+		 * return success, but we still want to return
+		 * -ENOSPC unless it found the minimum number
+		 * of bits. */
+		if (min_bits <= res->sr_bits)
+			search = 0; /* success */
+		else if (res->sr_bits) {
+			/*
+			 * Don't show bits which we'll be returning
+			 * for allocation to the local alloc bitmap.
+			 */
+			ocfs2_local_alloc_seen_free_bits(osb, res->sr_bits);
+		}
+	}
+
+	return search;
+}
+
+static int ocfs2_block_group_search(struct inode *inode,
+				    struct buffer_head *group_bh,
+				    u32 bits_wanted, u32 min_bits,
+				    u64 max_block,
+				    struct ocfs2_suballoc_result *res)
+{
+	int ret = -ENOSPC;
+	u64 blkoff;
+	struct ocfs2_group_desc *bg = (struct ocfs2_group_desc *) group_bh->b_data;
+
+	BUG_ON(min_bits != 1);
+	BUG_ON(ocfs2_is_cluster_bitmap(inode));
+
+	if (bg->bg_free_bits_count) {
+		ret = ocfs2_block_group_find_clear_bits(OCFS2_SB(inode->i_sb),
+							group_bh, bits_wanted,
+							le16_to_cpu(bg->bg_bits),
+							res);
+		if (!ret && max_block) {
+			blkoff = le64_to_cpu(bg->bg_blkno) +
+				res->sr_bit_offset + res->sr_bits;
+			trace_ocfs2_block_group_search_max_block(
+				(unsigned long long)blkoff,
+				(unsigned long long)max_block);
+			if (blkoff > max_block)
+				ret = -ENOSPC;
+		}
+	}
+
+	return ret;
+}
+
+int ocfs2_alloc_dinode_update_counts(struct inode *inode,
+				       handle_t *handle,
+				       struct buffer_head *di_bh,
+				       u32 num_bits,
+				       u16 chain)
+{
+	int ret;
+	u32 tmp_used;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
+	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &di->id2.i_chain;
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
+	di->id1.bitmap1.i_used = cpu_to_le32(num_bits + tmp_used);
+	le32_add_cpu(&cl->cl_recs[chain].c_free, -num_bits);
+	ocfs2_journal_dirty(handle, di_bh);
+
+out:
+	return ret;
+}
+
+void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
+				       struct buffer_head *di_bh,
+				       u32 num_bits,
+				       u16 chain)
+{
+	u32 tmp_used;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *) di_bh->b_data;
+	struct ocfs2_chain_list *cl;
+
+	cl = (struct ocfs2_chain_list *)&di->id2.i_chain;
+	tmp_used = le32_to_cpu(di->id1.bitmap1.i_used);
+	di->id1.bitmap1.i_used = cpu_to_le32(tmp_used - num_bits);
+	le32_add_cpu(&cl->cl_recs[chain].c_free, num_bits);
+}
+
+static int ocfs2_bg_discontig_fix_by_rec(struct ocfs2_suballoc_result *res,
+					 struct ocfs2_extent_rec *rec,
+					 struct ocfs2_chain_list *cl)
+{
+	unsigned int bpc = le16_to_cpu(cl->cl_bpc);
+	unsigned int bitoff = le32_to_cpu(rec->e_cpos) * bpc;
+	unsigned int bitcount = le16_to_cpu(rec->e_leaf_clusters) * bpc;
+
+	if (res->sr_bit_offset < bitoff)
+		return 0;
+	if (res->sr_bit_offset >= (bitoff + bitcount))
+		return 0;
+	res->sr_blkno = le64_to_cpu(rec->e_blkno) +
+		(res->sr_bit_offset - bitoff);
+	if ((res->sr_bit_offset + res->sr_bits) > (bitoff + bitcount))
+		res->sr_bits = (bitoff + bitcount) - res->sr_bit_offset;
+	return 1;
+}
+
+static void ocfs2_bg_discontig_fix_result(struct ocfs2_alloc_context *ac,
+					  struct ocfs2_group_desc *bg,
+					  struct ocfs2_suballoc_result *res)
+{
+	int i;
+	u64 bg_blkno = res->sr_bg_blkno;  /* Save off */
+	struct ocfs2_extent_rec *rec;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
+	struct ocfs2_chain_list *cl = &di->id2.i_chain;
+
+	if (ocfs2_is_cluster_bitmap(ac->ac_inode)) {
+		res->sr_blkno = 0;
+		return;
+	}
+
+	res->sr_blkno = res->sr_bg_blkno + res->sr_bit_offset;
+	res->sr_bg_blkno = 0;  /* Clear it for contig block groups */
+	if (!ocfs2_supports_discontig_bg(OCFS2_SB(ac->ac_inode->i_sb)) ||
+	    !bg->bg_list.l_next_free_rec)
+		return;
+
+	for (i = 0; i < le16_to_cpu(bg->bg_list.l_next_free_rec); i++) {
+		rec = &bg->bg_list.l_recs[i];
+		if (ocfs2_bg_discontig_fix_by_rec(res, rec, cl)) {
+			res->sr_bg_blkno = bg_blkno;  /* Restore */
+			break;
+		}
+	}
+}
+
+static int ocfs2_search_one_group(struct ocfs2_alloc_context *ac,
+				  handle_t *handle,
+				  u32 bits_wanted,
+				  u32 min_bits,
+				  struct ocfs2_suballoc_result *res,
+				  u16 *bits_left)
+{
+	int ret;
+	struct buffer_head *group_bh = NULL;
+	struct ocfs2_group_desc *gd;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)ac->ac_bh->b_data;
+	struct inode *alloc_inode = ac->ac_inode;
+
+	ret = ocfs2_read_group_descriptor(alloc_inode, di,
+					  res->sr_bg_blkno, &group_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	gd = (struct ocfs2_group_desc *) group_bh->b_data;
+	ret = ac->ac_group_search(alloc_inode, group_bh, bits_wanted, min_bits,
+				  ac->ac_max_block, res);
+	if (ret < 0) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	if (!ret)
+		ocfs2_bg_discontig_fix_result(ac, gd, res);
+
+	/*
+	 * sr_bg_blkno might have been changed by
+	 * ocfs2_bg_discontig_fix_result
+	 */
+	res->sr_bg_stable_blkno = group_bh->b_blocknr;
+
+	if (ac->ac_find_loc_only)
+		goto out_loc_only;
+
+	ret = ocfs2_alloc_dinode_update_counts(alloc_inode, handle, ac->ac_bh,
+					       res->sr_bits,
+					       le16_to_cpu(gd->bg_chain));
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_block_group_set_bits(handle, alloc_inode, gd, group_bh,
+					 res->sr_bit_offset, res->sr_bits);
+	if (ret < 0) {
+		ocfs2_rollback_alloc_dinode_counts(alloc_inode, ac->ac_bh,
+					       res->sr_bits,
+					       le16_to_cpu(gd->bg_chain));
+		mlog_errno(ret);
+	}
+
+out_loc_only:
+	*bits_left = le16_to_cpu(gd->bg_free_bits_count);
+
+out:
+	brelse(group_bh);
+
+	return ret;
+}
+
+static int ocfs2_search_chain(struct ocfs2_alloc_context *ac,
+			      handle_t *handle,
+			      u32 bits_wanted,
+			      u32 min_bits,
+			      struct ocfs2_suballoc_result *res,
+			      u16 *bits_left)
+{
+	int status;
+	u16 chain;
+	u64 next_group;
+	struct inode *alloc_inode = ac->ac_inode;
+	struct buffer_head *group_bh = NULL;
+	struct buffer_head *prev_group_bh = NULL;
+	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
+	struct ocfs2_chain_list *cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
+	struct ocfs2_group_desc *bg;
+
+	chain = ac->ac_chain;
+	trace_ocfs2_search_chain_begin(
+		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
+		bits_wanted, chain);
+
+	status = ocfs2_read_group_descriptor(alloc_inode, fe,
+					     le64_to_cpu(cl->cl_recs[chain].c_blkno),
+					     &group_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	bg = (struct ocfs2_group_desc *) group_bh->b_data;
+
+	status = -ENOSPC;
+	/* for now, the chain search is a bit simplistic. We just use
+	 * the 1st group with any empty bits. */
+	while ((status = ac->ac_group_search(alloc_inode, group_bh,
+					     bits_wanted, min_bits,
+					     ac->ac_max_block,
+					     res)) == -ENOSPC) {
+		if (!bg->bg_next_group)
+			break;
+
+		brelse(prev_group_bh);
+		prev_group_bh = NULL;
+
+		next_group = le64_to_cpu(bg->bg_next_group);
+		prev_group_bh = group_bh;
+		group_bh = NULL;
+		status = ocfs2_read_group_descriptor(alloc_inode, fe,
+						     next_group, &group_bh);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+		bg = (struct ocfs2_group_desc *) group_bh->b_data;
+	}
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	trace_ocfs2_search_chain_succ(
+		(unsigned long long)le64_to_cpu(bg->bg_blkno), res->sr_bits);
+
+	res->sr_bg_blkno = le64_to_cpu(bg->bg_blkno);
+
+	BUG_ON(res->sr_bits == 0);
+	if (!status)
+		ocfs2_bg_discontig_fix_result(ac, bg, res);
+
+	/*
+	 * sr_bg_blkno might have been changed by
+	 * ocfs2_bg_discontig_fix_result
+	 */
+	res->sr_bg_stable_blkno = group_bh->b_blocknr;
+
+	/*
+	 * Keep track of previous block descriptor read. When
+	 * we find a target, if we have read more than X
+	 * number of descriptors, and the target is reasonably
+	 * empty, relink him to top of his chain.
+	 *
+	 * We've read 0 extra blocks and only send one more to
+	 * the transaction, yet the next guy to search has a
+	 * much easier time.
+	 *
+	 * Do this *after* figuring out how many bits we're taking out
+	 * of our target group.
+	 */
+	if (!ac->ac_disable_chain_relink &&
+	    (prev_group_bh) &&
+	    (ocfs2_block_group_reasonably_empty(bg, res->sr_bits))) {
+		status = ocfs2_relink_block_group(handle, alloc_inode,
+						  ac->ac_bh, group_bh,
+						  prev_group_bh, chain);
+		if (status < 0) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	if (ac->ac_find_loc_only)
+		goto out_loc_only;
+
+	status = ocfs2_alloc_dinode_update_counts(alloc_inode, handle,
+						  ac->ac_bh, res->sr_bits,
+						  chain);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_block_group_set_bits(handle,
+					    alloc_inode,
+					    bg,
+					    group_bh,
+					    res->sr_bit_offset,
+					    res->sr_bits);
+	if (status < 0) {
+		ocfs2_rollback_alloc_dinode_counts(alloc_inode,
+					ac->ac_bh, res->sr_bits, chain);
+		mlog_errno(status);
+		goto bail;
+	}
+
+	trace_ocfs2_search_chain_end(
+			(unsigned long long)le64_to_cpu(fe->i_blkno),
+			res->sr_bits);
+
+out_loc_only:
+	*bits_left = le16_to_cpu(bg->bg_free_bits_count);
+bail:
+	brelse(group_bh);
+	brelse(prev_group_bh);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/* will give out up to bits_wanted contiguous bits. */
+static int ocfs2_claim_suballoc_bits(struct ocfs2_alloc_context *ac,
+				     handle_t *handle,
+				     u32 bits_wanted,
+				     u32 min_bits,
+				     struct ocfs2_suballoc_result *res)
+{
+	int status;
+	u16 victim, i;
+	u16 bits_left = 0;
+	u64 hint = ac->ac_last_group;
+	struct ocfs2_chain_list *cl;
+	struct ocfs2_dinode *fe;
+
+	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
+	BUG_ON(bits_wanted > (ac->ac_bits_wanted - ac->ac_bits_given));
+	BUG_ON(!ac->ac_bh);
+
+	fe = (struct ocfs2_dinode *) ac->ac_bh->b_data;
+
+	/* The bh was validated by the inode read during
+	 * ocfs2_reserve_suballoc_bits().  Any corruption is a code bug. */
+	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
+
+	if (le32_to_cpu(fe->id1.bitmap1.i_used) >=
+	    le32_to_cpu(fe->id1.bitmap1.i_total)) {
+		ocfs2_error(ac->ac_inode->i_sb,
+			    "Chain allocator dinode %llu has %u used "
+			    "bits but only %u total.",
+			    (unsigned long long)le64_to_cpu(fe->i_blkno),
+			    le32_to_cpu(fe->id1.bitmap1.i_used),
+			    le32_to_cpu(fe->id1.bitmap1.i_total));
+		status = -EIO;
+		goto bail;
+	}
+
+	res->sr_bg_blkno = hint;
+	if (res->sr_bg_blkno) {
+		/* Attempt to short-circuit the usual search mechanism
+		 * by jumping straight to the most recently used
+		 * allocation group. This helps us maintain some
+		 * contiguousness across allocations. */
+		status = ocfs2_search_one_group(ac, handle, bits_wanted,
+						min_bits, res, &bits_left);
+		if (!status)
+			goto set_hint;
+		if (status < 0 && status != -ENOSPC) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+	cl = (struct ocfs2_chain_list *) &fe->id2.i_chain;
+
+	victim = ocfs2_find_victim_chain(cl);
+	ac->ac_chain = victim;
+
+	status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
+				    res, &bits_left);
+	if (!status) {
+		hint = ocfs2_group_from_res(res);
+		goto set_hint;
+	}
+	if (status < 0 && status != -ENOSPC) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	trace_ocfs2_claim_suballoc_bits(victim);
+
+	/* If we didn't pick a good victim, then just default to
+	 * searching each chain in order. Don't allow chain relinking
+	 * because we only calculate enough journal credits for one
+	 * relink per alloc. */
+	ac->ac_disable_chain_relink = 1;
+	for (i = 0; i < le16_to_cpu(cl->cl_next_free_rec); i ++) {
+		if (i == victim)
+			continue;
+		if (!cl->cl_recs[i].c_free)
+			continue;
+
+		ac->ac_chain = i;
+		status = ocfs2_search_chain(ac, handle, bits_wanted, min_bits,
+					    res, &bits_left);
+		if (!status) {
+			hint = ocfs2_group_from_res(res);
+			break;
+		}
+		if (status < 0 && status != -ENOSPC) {
+			mlog_errno(status);
+			goto bail;
+		}
+	}
+
+set_hint:
+	if (status != -ENOSPC) {
+		/* If the next search of this group is not likely to
+		 * yield a suitable extent, then we reset the last
+		 * group hint so as to not waste a disk read */
+		if (bits_left < min_bits)
+			ac->ac_last_group = 0;
+		else
+			ac->ac_last_group = hint;
+	}
+
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+int ocfs2_claim_metadata(handle_t *handle,
+			 struct ocfs2_alloc_context *ac,
+			 u32 bits_wanted,
+			 u64 *suballoc_loc,
+			 u16 *suballoc_bit_start,
+			 unsigned int *num_bits,
+			 u64 *blkno_start)
+{
+	int status;
+	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
+
+	BUG_ON(!ac);
+	BUG_ON(ac->ac_bits_wanted < (ac->ac_bits_given + bits_wanted));
+	BUG_ON(ac->ac_which != OCFS2_AC_USE_META);
+
+	status = ocfs2_claim_suballoc_bits(ac,
+					   handle,
+					   bits_wanted,
+					   1,
+					   &res);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
+
+	*suballoc_loc = res.sr_bg_blkno;
+	*suballoc_bit_start = res.sr_bit_offset;
+	*blkno_start = res.sr_blkno;
+	ac->ac_bits_given += res.sr_bits;
+	*num_bits = res.sr_bits;
+	status = 0;
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static void ocfs2_init_inode_ac_group(struct inode *dir,
+				      struct buffer_head *parent_di_bh,
+				      struct ocfs2_alloc_context *ac)
+{
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)parent_di_bh->b_data;
+	/*
+	 * Try to allocate inodes from some specific group.
+	 *
+	 * If the parent dir has recorded the last group used in allocation,
+	 * cool, use it. Otherwise if we try to allocate new inode from the
+	 * same slot the parent dir belongs to, use the same chunk.
+	 *
+	 * We are very careful here to avoid the mistake of setting
+	 * ac_last_group to a group descriptor from a different (unlocked) slot.
+	 */
+	if (OCFS2_I(dir)->ip_last_used_group &&
+	    OCFS2_I(dir)->ip_last_used_slot == ac->ac_alloc_slot)
+		ac->ac_last_group = OCFS2_I(dir)->ip_last_used_group;
+	else if (le16_to_cpu(di->i_suballoc_slot) == ac->ac_alloc_slot) {
+		if (di->i_suballoc_loc)
+			ac->ac_last_group = le64_to_cpu(di->i_suballoc_loc);
+		else
+			ac->ac_last_group = ocfs2_which_suballoc_group(
+					le64_to_cpu(di->i_blkno),
+					le16_to_cpu(di->i_suballoc_bit));
+	}
+}
+
+static inline void ocfs2_save_inode_ac_group(struct inode *dir,
+					     struct ocfs2_alloc_context *ac)
+{
+	OCFS2_I(dir)->ip_last_used_group = ac->ac_last_group;
+	OCFS2_I(dir)->ip_last_used_slot = ac->ac_alloc_slot;
+}
+
+int ocfs2_find_new_inode_loc(struct inode *dir,
+			     struct buffer_head *parent_fe_bh,
+			     struct ocfs2_alloc_context *ac,
+			     u64 *fe_blkno)
+{
+	int ret;
+	handle_t *handle = NULL;
+	struct ocfs2_suballoc_result *res;
+
+	BUG_ON(!ac);
+	BUG_ON(ac->ac_bits_given != 0);
+	BUG_ON(ac->ac_bits_wanted != 1);
+	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
+
+	res = kzalloc(sizeof(*res), GFP_NOFS);
+	if (res == NULL) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
+
+	/*
+	 * The handle started here is for chain relink. Alternatively,
+	 * we could just disable relink for these calls.
+	 */
+	handle = ocfs2_start_trans(OCFS2_SB(dir->i_sb), OCFS2_SUBALLOC_ALLOC);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		handle = NULL;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * This will instruct ocfs2_claim_suballoc_bits and
+	 * ocfs2_search_one_group to search but save actual allocation
+	 * for later.
+	 */
+	ac->ac_find_loc_only = 1;
+
+	ret = ocfs2_claim_suballoc_bits(ac, handle, 1, 1, res);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ac->ac_find_loc_priv = res;
+	*fe_blkno = res->sr_blkno;
+	ocfs2_update_inode_fsync_trans(handle, dir, 0);
+out:
+	if (handle)
+		ocfs2_commit_trans(OCFS2_SB(dir->i_sb), handle);
+
+	if (ret)
+		kfree(res);
+
+	return ret;
+}
+
+int ocfs2_claim_new_inode_at_loc(handle_t *handle,
+				 struct inode *dir,
+				 struct ocfs2_alloc_context *ac,
+				 u64 *suballoc_loc,
+				 u16 *suballoc_bit,
+				 u64 di_blkno)
+{
+	int ret;
+	u16 chain;
+	struct ocfs2_suballoc_result *res = ac->ac_find_loc_priv;
+	struct buffer_head *bg_bh = NULL;
+	struct ocfs2_group_desc *bg;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *) ac->ac_bh->b_data;
+
+	/*
+	 * Since di_blkno is being passed back in, we check for any
+	 * inconsistencies which may have happened between
+	 * calls. These are code bugs as di_blkno is not expected to
+	 * change once returned from ocfs2_find_new_inode_loc()
+	 */
+	BUG_ON(res->sr_blkno != di_blkno);
+
+	ret = ocfs2_read_group_descriptor(ac->ac_inode, di,
+					  res->sr_bg_stable_blkno, &bg_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	bg = (struct ocfs2_group_desc *) bg_bh->b_data;
+	chain = le16_to_cpu(bg->bg_chain);
+
+	ret = ocfs2_alloc_dinode_update_counts(ac->ac_inode, handle,
+					       ac->ac_bh, res->sr_bits,
+					       chain);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_block_group_set_bits(handle,
+					 ac->ac_inode,
+					 bg,
+					 bg_bh,
+					 res->sr_bit_offset,
+					 res->sr_bits);
+	if (ret < 0) {
+		ocfs2_rollback_alloc_dinode_counts(ac->ac_inode,
+					       ac->ac_bh, res->sr_bits, chain);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_claim_new_inode_at_loc((unsigned long long)di_blkno,
+					   res->sr_bits);
+
+	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
+
+	BUG_ON(res->sr_bits != 1);
+
+	*suballoc_loc = res->sr_bg_blkno;
+	*suballoc_bit = res->sr_bit_offset;
+	ac->ac_bits_given++;
+	ocfs2_save_inode_ac_group(dir, ac);
+
+out:
+	brelse(bg_bh);
+
+	return ret;
+}
+
+int ocfs2_claim_new_inode(handle_t *handle,
+			  struct inode *dir,
+			  struct buffer_head *parent_fe_bh,
+			  struct ocfs2_alloc_context *ac,
+			  u64 *suballoc_loc,
+			  u16 *suballoc_bit,
+			  u64 *fe_blkno)
+{
+	int status;
+	struct ocfs2_suballoc_result res;
+
+	BUG_ON(!ac);
+	BUG_ON(ac->ac_bits_given != 0);
+	BUG_ON(ac->ac_bits_wanted != 1);
+	BUG_ON(ac->ac_which != OCFS2_AC_USE_INODE);
+
+	ocfs2_init_inode_ac_group(dir, parent_fe_bh, ac);
+
+	status = ocfs2_claim_suballoc_bits(ac,
+					   handle,
+					   1,
+					   1,
+					   &res);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	atomic_inc(&OCFS2_SB(ac->ac_inode->i_sb)->alloc_stats.bg_allocs);
+
+	BUG_ON(res.sr_bits != 1);
+
+	*suballoc_loc = res.sr_bg_blkno;
+	*suballoc_bit = res.sr_bit_offset;
+	*fe_blkno = res.sr_blkno;
+	ac->ac_bits_given++;
+	ocfs2_save_inode_ac_group(dir, ac);
+	status = 0;
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/* translate a group desc. blkno and it's bitmap offset into
+ * disk cluster offset. */
+static inline u32 ocfs2_desc_bitmap_to_cluster_off(struct inode *inode,
+						   u64 bg_blkno,
+						   u16 bg_bit_off)
+{
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	u32 cluster = 0;
+
+	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
+
+	if (bg_blkno != osb->first_cluster_group_blkno)
+		cluster = ocfs2_blocks_to_clusters(inode->i_sb, bg_blkno);
+	cluster += (u32) bg_bit_off;
+	return cluster;
+}
+
+/* given a cluster offset, calculate which block group it belongs to
+ * and return that block offset. */
+u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster)
+{
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	u32 group_no;
+
+	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
+
+	group_no = cluster / osb->bitmap_cpg;
+	if (!group_no)
+		return osb->first_cluster_group_blkno;
+	return ocfs2_clusters_to_blocks(inode->i_sb,
+					group_no * osb->bitmap_cpg);
+}
+
+/* given the block number of a cluster start, calculate which cluster
+ * group and descriptor bitmap offset that corresponds to. */
+static inline void ocfs2_block_to_cluster_group(struct inode *inode,
+						u64 data_blkno,
+						u64 *bg_blkno,
+						u16 *bg_bit_off)
+{
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	u32 data_cluster = ocfs2_blocks_to_clusters(osb->sb, data_blkno);
+
+	BUG_ON(!ocfs2_is_cluster_bitmap(inode));
+
+	*bg_blkno = ocfs2_which_cluster_group(inode,
+					      data_cluster);
+
+	if (*bg_blkno == osb->first_cluster_group_blkno)
+		*bg_bit_off = (u16) data_cluster;
+	else
+		*bg_bit_off = (u16) ocfs2_blocks_to_clusters(osb->sb,
+							     data_blkno - *bg_blkno);
+}
+
+/*
+ * min_bits - minimum contiguous chunk from this total allocation we
+ * can handle. set to what we asked for originally for a full
+ * contig. allocation, set to '1' to indicate we can deal with extents
+ * of any size.
+ */
+int __ocfs2_claim_clusters(handle_t *handle,
+			   struct ocfs2_alloc_context *ac,
+			   u32 min_clusters,
+			   u32 max_clusters,
+			   u32 *cluster_start,
+			   u32 *num_clusters)
+{
+	int status;
+	unsigned int bits_wanted = max_clusters;
+	struct ocfs2_suballoc_result res = { .sr_blkno = 0, };
+	struct ocfs2_super *osb = OCFS2_SB(ac->ac_inode->i_sb);
+
+	BUG_ON(ac->ac_bits_given >= ac->ac_bits_wanted);
+
+	BUG_ON(ac->ac_which != OCFS2_AC_USE_LOCAL
+	       && ac->ac_which != OCFS2_AC_USE_MAIN);
+
+	if (ac->ac_which == OCFS2_AC_USE_LOCAL) {
+		WARN_ON(min_clusters > 1);
+
+		status = ocfs2_claim_local_alloc_bits(osb,
+						      handle,
+						      ac,
+						      bits_wanted,
+						      cluster_start,
+						      num_clusters);
+		if (!status)
+			atomic_inc(&osb->alloc_stats.local_data);
+	} else {
+		if (min_clusters > (osb->bitmap_cpg - 1)) {
+			/* The only paths asking for contiguousness
+			 * should know about this already. */
+			mlog(ML_ERROR, "minimum allocation requested %u exceeds "
+			     "group bitmap size %u!\n", min_clusters,
+			     osb->bitmap_cpg);
+			status = -ENOSPC;
+			goto bail;
+		}
+		/* clamp the current request down to a realistic size. */
+		if (bits_wanted > (osb->bitmap_cpg - 1))
+			bits_wanted = osb->bitmap_cpg - 1;
+
+		status = ocfs2_claim_suballoc_bits(ac,
+						   handle,
+						   bits_wanted,
+						   min_clusters,
+						   &res);
+		if (!status) {
+			BUG_ON(res.sr_blkno); /* cluster alloc can't set */
+			*cluster_start =
+				ocfs2_desc_bitmap_to_cluster_off(ac->ac_inode,
+								 res.sr_bg_blkno,
+								 res.sr_bit_offset);
+			atomic_inc(&osb->alloc_stats.bitmap_data);
+			*num_clusters = res.sr_bits;
+		}
+	}
+	if (status < 0) {
+		if (status != -ENOSPC)
+			mlog_errno(status);
+		goto bail;
+	}
+
+	ac->ac_bits_given += *num_clusters;
+
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+int ocfs2_claim_clusters(handle_t *handle,
+			 struct ocfs2_alloc_context *ac,
+			 u32 min_clusters,
+			 u32 *cluster_start,
+			 u32 *num_clusters)
+{
+	unsigned int bits_wanted = ac->ac_bits_wanted - ac->ac_bits_given;
+
+	return __ocfs2_claim_clusters(handle, ac, min_clusters,
+				      bits_wanted, cluster_start, num_clusters);
+}
+
+static int ocfs2_block_group_clear_bits(handle_t *handle,
+					struct inode *alloc_inode,
+					struct ocfs2_group_desc *bg,
+					struct buffer_head *group_bh,
+					unsigned int bit_off,
+					unsigned int num_bits,
+					void (*undo_fn)(unsigned int bit,
+							unsigned long *bmap))
+{
+	int status;
+	unsigned int tmp;
+	struct ocfs2_group_desc *undo_bg = NULL;
+
+	/* The caller got this descriptor from
+	 * ocfs2_read_group_descriptor().  Any corruption is a code bug. */
+	BUG_ON(!OCFS2_IS_VALID_GROUP_DESC(bg));
+
+	trace_ocfs2_block_group_clear_bits(bit_off, num_bits);
+
+	BUG_ON(undo_fn && !ocfs2_is_cluster_bitmap(alloc_inode));
+	status = ocfs2_journal_access_gd(handle, INODE_CACHE(alloc_inode),
+					 group_bh,
+					 undo_fn ?
+					 OCFS2_JOURNAL_ACCESS_UNDO :
+					 OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	if (undo_fn) {
+		jbd_lock_bh_state(group_bh);
+		undo_bg = (struct ocfs2_group_desc *)
+					bh2jh(group_bh)->b_committed_data;
+		BUG_ON(!undo_bg);
+	}
+
+	tmp = num_bits;
+	while(tmp--) {
+		ocfs2_clear_bit((bit_off + tmp),
+				(unsigned long *) bg->bg_bitmap);
+		if (undo_fn)
+			undo_fn(bit_off + tmp,
+				(unsigned long *) undo_bg->bg_bitmap);
+	}
+	le16_add_cpu(&bg->bg_free_bits_count, num_bits);
+	if (le16_to_cpu(bg->bg_free_bits_count) > le16_to_cpu(bg->bg_bits)) {
+		ocfs2_error(alloc_inode->i_sb, "Group descriptor # %llu has bit"
+			    " count %u but claims %u are freed. num_bits %d",
+			    (unsigned long long)le64_to_cpu(bg->bg_blkno),
+			    le16_to_cpu(bg->bg_bits),
+			    le16_to_cpu(bg->bg_free_bits_count), num_bits);
+		return -EROFS;
+	}
+
+	if (undo_fn)
+		jbd_unlock_bh_state(group_bh);
+
+	ocfs2_journal_dirty(handle, group_bh);
+bail:
+	return status;
+}
+
+/*
+ * expects the suballoc inode to already be locked.
+ */
+static int _ocfs2_free_suballoc_bits(handle_t *handle,
+				     struct inode *alloc_inode,
+				     struct buffer_head *alloc_bh,
+				     unsigned int start_bit,
+				     u64 bg_blkno,
+				     unsigned int count,
+				     void (*undo_fn)(unsigned int bit,
+						     unsigned long *bitmap))
+{
+	int status = 0;
+	u32 tmp_used;
+	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) alloc_bh->b_data;
+	struct ocfs2_chain_list *cl = &fe->id2.i_chain;
+	struct buffer_head *group_bh = NULL;
+	struct ocfs2_group_desc *group;
+
+	/* The alloc_bh comes from ocfs2_free_dinode() or
+	 * ocfs2_free_clusters().  The callers have all locked the
+	 * allocator and gotten alloc_bh from the lock call.  This
+	 * validates the dinode buffer.  Any corruption that has happened
+	 * is a code bug. */
+	BUG_ON(!OCFS2_IS_VALID_DINODE(fe));
+	BUG_ON((count + start_bit) > ocfs2_bits_per_group(cl));
+
+	trace_ocfs2_free_suballoc_bits(
+		(unsigned long long)OCFS2_I(alloc_inode)->ip_blkno,
+		(unsigned long long)bg_blkno,
+		start_bit, count);
+
+	status = ocfs2_read_group_descriptor(alloc_inode, fe, bg_blkno,
+					     &group_bh);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	group = (struct ocfs2_group_desc *) group_bh->b_data;
+
+	BUG_ON((count + start_bit) > le16_to_cpu(group->bg_bits));
+
+	status = ocfs2_block_group_clear_bits(handle, alloc_inode,
+					      group, group_bh,
+					      start_bit, count, undo_fn);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	status = ocfs2_journal_access_di(handle, INODE_CACHE(alloc_inode),
+					 alloc_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (status < 0) {
+		mlog_errno(status);
+		ocfs2_block_group_set_bits(handle, alloc_inode, group, group_bh,
+				start_bit, count);
+		goto bail;
+	}
+
+	le32_add_cpu(&cl->cl_recs[le16_to_cpu(group->bg_chain)].c_free,
+		     count);
+	tmp_used = le32_to_cpu(fe->id1.bitmap1.i_used);
+	fe->id1.bitmap1.i_used = cpu_to_le32(tmp_used - count);
+	ocfs2_journal_dirty(handle, alloc_bh);
+
+bail:
+	brelse(group_bh);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+int ocfs2_free_suballoc_bits(handle_t *handle,
+			     struct inode *alloc_inode,
+			     struct buffer_head *alloc_bh,
+			     unsigned int start_bit,
+			     u64 bg_blkno,
+			     unsigned int count)
+{
+	return _ocfs2_free_suballoc_bits(handle, alloc_inode, alloc_bh,
+					 start_bit, bg_blkno, count, NULL);
+}
+
+int ocfs2_free_dinode(handle_t *handle,
+		      struct inode *inode_alloc_inode,
+		      struct buffer_head *inode_alloc_bh,
+		      struct ocfs2_dinode *di)
+{
+	u64 blk = le64_to_cpu(di->i_blkno);
+	u16 bit = le16_to_cpu(di->i_suballoc_bit);
+	u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
+
+	if (di->i_suballoc_loc)
+		bg_blkno = le64_to_cpu(di->i_suballoc_loc);
+	return ocfs2_free_suballoc_bits(handle, inode_alloc_inode,
+					inode_alloc_bh, bit, bg_blkno, 1);
+}
+
+static int _ocfs2_free_clusters(handle_t *handle,
+				struct inode *bitmap_inode,
+				struct buffer_head *bitmap_bh,
+				u64 start_blk,
+				unsigned int num_clusters,
+				void (*undo_fn)(unsigned int bit,
+						unsigned long *bitmap))
+{
+	int status;
+	u16 bg_start_bit;
+	u64 bg_blkno;
+	struct ocfs2_dinode *fe;
+
+	/* You can't ever have a contiguous set of clusters
+	 * bigger than a block group bitmap so we never have to worry
+	 * about looping on them.
+	 * This is expensive. We can safely remove once this stuff has
+	 * gotten tested really well. */
+	BUG_ON(start_blk != ocfs2_clusters_to_blocks(bitmap_inode->i_sb, ocfs2_blocks_to_clusters(bitmap_inode->i_sb, start_blk)));
+
+	fe = (struct ocfs2_dinode *) bitmap_bh->b_data;
+
+	ocfs2_block_to_cluster_group(bitmap_inode, start_blk, &bg_blkno,
+				     &bg_start_bit);
+
+	trace_ocfs2_free_clusters((unsigned long long)bg_blkno,
+			(unsigned long long)start_blk,
+			bg_start_bit, num_clusters);
+
+	status = _ocfs2_free_suballoc_bits(handle, bitmap_inode, bitmap_bh,
+					   bg_start_bit, bg_blkno,
+					   num_clusters, undo_fn);
+	if (status < 0) {
+		mlog_errno(status);
+		goto out;
+	}
+
+	ocfs2_local_alloc_seen_free_bits(OCFS2_SB(bitmap_inode->i_sb),
+					 num_clusters);
+
+out:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+int ocfs2_free_clusters(handle_t *handle,
+			struct inode *bitmap_inode,
+			struct buffer_head *bitmap_bh,
+			u64 start_blk,
+			unsigned int num_clusters)
+{
+	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
+				    start_blk, num_clusters,
+				    _ocfs2_set_bit);
+}
+
+/*
+ * Give never-used clusters back to the global bitmap.  We don't need
+ * to protect these bits in the undo buffer.
+ */
+int ocfs2_release_clusters(handle_t *handle,
+			   struct inode *bitmap_inode,
+			   struct buffer_head *bitmap_bh,
+			   u64 start_blk,
+			   unsigned int num_clusters)
+{
+	return _ocfs2_free_clusters(handle, bitmap_inode, bitmap_bh,
+				    start_blk, num_clusters,
+				    _ocfs2_clear_bit);
+}
+
+static inline void ocfs2_debug_bg(struct ocfs2_group_desc *bg)
+{
+	printk("Block Group:\n");
+	printk("bg_signature:       %s\n", bg->bg_signature);
+	printk("bg_size:            %u\n", bg->bg_size);
+	printk("bg_bits:            %u\n", bg->bg_bits);
+	printk("bg_free_bits_count: %u\n", bg->bg_free_bits_count);
+	printk("bg_chain:           %u\n", bg->bg_chain);
+	printk("bg_generation:      %u\n", le32_to_cpu(bg->bg_generation));
+	printk("bg_next_group:      %llu\n",
+	       (unsigned long long)bg->bg_next_group);
+	printk("bg_parent_dinode:   %llu\n",
+	       (unsigned long long)bg->bg_parent_dinode);
+	printk("bg_blkno:           %llu\n",
+	       (unsigned long long)bg->bg_blkno);
+}
+
+static inline void ocfs2_debug_suballoc_inode(struct ocfs2_dinode *fe)
+{
+	int i;
+
+	printk("Suballoc Inode %llu:\n", (unsigned long long)fe->i_blkno);
+	printk("i_signature:                  %s\n", fe->i_signature);
+	printk("i_size:                       %llu\n",
+	       (unsigned long long)fe->i_size);
+	printk("i_clusters:                   %u\n", fe->i_clusters);
+	printk("i_generation:                 %u\n",
+	       le32_to_cpu(fe->i_generation));
+	printk("id1.bitmap1.i_used:           %u\n",
+	       le32_to_cpu(fe->id1.bitmap1.i_used));
+	printk("id1.bitmap1.i_total:          %u\n",
+	       le32_to_cpu(fe->id1.bitmap1.i_total));
+	printk("id2.i_chain.cl_cpg:           %u\n", fe->id2.i_chain.cl_cpg);
+	printk("id2.i_chain.cl_bpc:           %u\n", fe->id2.i_chain.cl_bpc);
+	printk("id2.i_chain.cl_count:         %u\n", fe->id2.i_chain.cl_count);
+	printk("id2.i_chain.cl_next_free_rec: %u\n",
+	       fe->id2.i_chain.cl_next_free_rec);
+	for(i = 0; i < fe->id2.i_chain.cl_next_free_rec; i++) {
+		printk("fe->id2.i_chain.cl_recs[%d].c_free:  %u\n", i,
+		       fe->id2.i_chain.cl_recs[i].c_free);
+		printk("fe->id2.i_chain.cl_recs[%d].c_total: %u\n", i,
+		       fe->id2.i_chain.cl_recs[i].c_total);
+		printk("fe->id2.i_chain.cl_recs[%d].c_blkno: %llu\n", i,
+		       (unsigned long long)fe->id2.i_chain.cl_recs[i].c_blkno);
+	}
+}
+
+/*
+ * For a given allocation, determine which allocators will need to be
+ * accessed, and lock them, reserving the appropriate number of bits.
+ *
+ * Sparse file systems call this from ocfs2_write_begin_nolock()
+ * and ocfs2_allocate_unwritten_extents().
+ *
+ * File systems which don't support holes call this from
+ * ocfs2_extend_allocation().
+ */
+int ocfs2_lock_allocators(struct inode *inode,
+			  struct ocfs2_extent_tree *et,
+			  u32 clusters_to_add, u32 extents_to_split,
+			  struct ocfs2_alloc_context **data_ac,
+			  struct ocfs2_alloc_context **meta_ac)
+{
+	int ret = 0, num_free_extents;
+	unsigned int max_recs_needed = clusters_to_add + 2 * extents_to_split;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	*meta_ac = NULL;
+	if (data_ac)
+		*data_ac = NULL;
+
+	BUG_ON(clusters_to_add != 0 && data_ac == NULL);
+
+	num_free_extents = ocfs2_num_free_extents(osb, et);
+	if (num_free_extents < 0) {
+		ret = num_free_extents;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Sparse allocation file systems need to be more conservative
+	 * with reserving room for expansion - the actual allocation
+	 * happens while we've got a journal handle open so re-taking
+	 * a cluster lock (because we ran out of room for another
+	 * extent) will violate ordering rules.
+	 *
+	 * Most of the time we'll only be seeing this 1 cluster at a time
+	 * anyway.
+	 *
+	 * Always lock for any unwritten extents - we might want to
+	 * add blocks during a split.
+	 */
+	if (!num_free_extents ||
+	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed)) {
+		ret = ocfs2_reserve_new_metadata(osb, et->et_root_el, meta_ac);
+		if (ret < 0) {
+			if (ret != -ENOSPC)
+				mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (clusters_to_add == 0)
+		goto out;
+
+	ret = ocfs2_reserve_clusters(osb, clusters_to_add, data_ac);
+	if (ret < 0) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		goto out;
+	}
+
+out:
+	if (ret) {
+		if (*meta_ac) {
+			ocfs2_free_alloc_context(*meta_ac);
+			*meta_ac = NULL;
+		}
+
+		/*
+		 * We cannot have an error and a non null *data_ac.
+		 */
+	}
+
+	return ret;
+}
+
+/*
+ * Read the inode specified by blkno to get suballoc_slot and
+ * suballoc_bit.
+ */
+static int ocfs2_get_suballoc_slot_bit(struct ocfs2_super *osb, u64 blkno,
+				       u16 *suballoc_slot, u64 *group_blkno,
+				       u16 *suballoc_bit)
+{
+	int status;
+	struct buffer_head *inode_bh = NULL;
+	struct ocfs2_dinode *inode_fe;
+
+	trace_ocfs2_get_suballoc_slot_bit((unsigned long long)blkno);
+
+	/* dirty read disk */
+	status = ocfs2_read_blocks_sync(osb, blkno, 1, &inode_bh);
+	if (status < 0) {
+		mlog(ML_ERROR, "read block %llu failed %d\n",
+		     (unsigned long long)blkno, status);
+		goto bail;
+	}
+
+	inode_fe = (struct ocfs2_dinode *) inode_bh->b_data;
+	if (!OCFS2_IS_VALID_DINODE(inode_fe)) {
+		mlog(ML_ERROR, "invalid inode %llu requested\n",
+		     (unsigned long long)blkno);
+		status = -EINVAL;
+		goto bail;
+	}
+
+	if (le16_to_cpu(inode_fe->i_suballoc_slot) != (u16)OCFS2_INVALID_SLOT &&
+	    (u32)le16_to_cpu(inode_fe->i_suballoc_slot) > osb->max_slots - 1) {
+		mlog(ML_ERROR, "inode %llu has invalid suballoc slot %u\n",
+		     (unsigned long long)blkno,
+		     (u32)le16_to_cpu(inode_fe->i_suballoc_slot));
+		status = -EINVAL;
+		goto bail;
+	}
+
+	if (suballoc_slot)
+		*suballoc_slot = le16_to_cpu(inode_fe->i_suballoc_slot);
+	if (suballoc_bit)
+		*suballoc_bit = le16_to_cpu(inode_fe->i_suballoc_bit);
+	if (group_blkno)
+		*group_blkno = le64_to_cpu(inode_fe->i_suballoc_loc);
+
+bail:
+	brelse(inode_bh);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/*
+ * test whether bit is SET in allocator bitmap or not.  on success, 0
+ * is returned and *res is 1 for SET; 0 otherwise.  when fails, errno
+ * is returned and *res is meaningless.  Call this after you have
+ * cluster locked against suballoc, or you may get a result based on
+ * non-up2date contents
+ */
+static int ocfs2_test_suballoc_bit(struct ocfs2_super *osb,
+				   struct inode *suballoc,
+				   struct buffer_head *alloc_bh,
+				   u64 group_blkno, u64 blkno,
+				   u16 bit, int *res)
+{
+	struct ocfs2_dinode *alloc_di;
+	struct ocfs2_group_desc *group;
+	struct buffer_head *group_bh = NULL;
+	u64 bg_blkno;
+	int status;
+
+	trace_ocfs2_test_suballoc_bit((unsigned long long)blkno,
+				      (unsigned int)bit);
+
+	alloc_di = (struct ocfs2_dinode *)alloc_bh->b_data;
+	if ((bit + 1) > ocfs2_bits_per_group(&alloc_di->id2.i_chain)) {
+		mlog(ML_ERROR, "suballoc bit %u out of range of %u\n",
+		     (unsigned int)bit,
+		     ocfs2_bits_per_group(&alloc_di->id2.i_chain));
+		status = -EINVAL;
+		goto bail;
+	}
+
+	bg_blkno = group_blkno ? group_blkno :
+		   ocfs2_which_suballoc_group(blkno, bit);
+	status = ocfs2_read_group_descriptor(suballoc, alloc_di, bg_blkno,
+					     &group_bh);
+	if (status < 0) {
+		mlog(ML_ERROR, "read group %llu failed %d\n",
+		     (unsigned long long)bg_blkno, status);
+		goto bail;
+	}
+
+	group = (struct ocfs2_group_desc *) group_bh->b_data;
+	*res = ocfs2_test_bit(bit, (unsigned long *)group->bg_bitmap);
+
+bail:
+	brelse(group_bh);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/*
+ * Test if the bit representing this inode (blkno) is set in the
+ * suballocator.
+ *
+ * On success, 0 is returned and *res is 1 for SET; 0 otherwise.
+ *
+ * In the event of failure, a negative value is returned and *res is
+ * meaningless.
+ *
+ * Callers must make sure to hold nfs_sync_lock to prevent
+ * ocfs2_delete_inode() on another node from accessing the same
+ * suballocator concurrently.
+ */
+int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res)
+{
+	int status;
+	u64 group_blkno = 0;
+	u16 suballoc_bit = 0, suballoc_slot = 0;
+	struct inode *inode_alloc_inode;
+	struct buffer_head *alloc_bh = NULL;
+
+	trace_ocfs2_test_inode_bit((unsigned long long)blkno);
+
+	status = ocfs2_get_suballoc_slot_bit(osb, blkno, &suballoc_slot,
+					     &group_blkno, &suballoc_bit);
+	if (status < 0) {
+		mlog(ML_ERROR, "get alloc slot and bit failed %d\n", status);
+		goto bail;
+	}
+
+	inode_alloc_inode =
+		ocfs2_get_system_file_inode(osb, INODE_ALLOC_SYSTEM_INODE,
+					    suballoc_slot);
+	if (!inode_alloc_inode) {
+		/* the error code could be inaccurate, but we are not able to
+		 * get the correct one. */
+		status = -EINVAL;
+		mlog(ML_ERROR, "unable to get alloc inode in slot %u\n",
+		     (u32)suballoc_slot);
+		goto bail;
+	}
+
+	mutex_lock(&inode_alloc_inode->i_mutex);
+	status = ocfs2_inode_lock(inode_alloc_inode, &alloc_bh, 0);
+	if (status < 0) {
+		mutex_unlock(&inode_alloc_inode->i_mutex);
+		iput(inode_alloc_inode);
+		mlog(ML_ERROR, "lock on alloc inode on slot %u failed %d\n",
+		     (u32)suballoc_slot, status);
+		goto bail;
+	}
+
+	status = ocfs2_test_suballoc_bit(osb, inode_alloc_inode, alloc_bh,
+					 group_blkno, blkno, suballoc_bit, res);
+	if (status < 0)
+		mlog(ML_ERROR, "test suballoc bit failed %d\n", status);
+
+	ocfs2_inode_unlock(inode_alloc_inode, 0);
+	mutex_unlock(&inode_alloc_inode->i_mutex);
+
+	iput(inode_alloc_inode);
+	brelse(alloc_bh);
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
diff --git a/fs/ocfs2/suballoc.h b/fs/ocfs2/suballoc.h
new file mode 100644
index 000000000..2d2501767
--- /dev/null
+++ b/fs/ocfs2/suballoc.h
@@ -0,0 +1,237 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * suballoc.h
+ *
+ * Defines sub allocator api
+ *
+ * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef _CHAINALLOC_H_
+#define _CHAINALLOC_H_
+
+struct ocfs2_suballoc_result;
+typedef int (group_search_t)(struct inode *,
+			     struct buffer_head *,
+			     u32,			/* bits_wanted */
+			     u32,			/* min_bits */
+			     u64,			/* max_block */
+			     struct ocfs2_suballoc_result *);
+							/* found bits */
+
+struct ocfs2_alloc_context {
+	struct inode *ac_inode;    /* which bitmap are we allocating from? */
+	struct buffer_head *ac_bh; /* file entry bh */
+	u32    ac_alloc_slot;   /* which slot are we allocating from? */
+	u32    ac_bits_wanted;
+	u32    ac_bits_given;
+#define OCFS2_AC_USE_LOCAL 1
+#define OCFS2_AC_USE_MAIN  2
+#define OCFS2_AC_USE_INODE 3
+#define OCFS2_AC_USE_META  4
+	u32    ac_which;
+
+	/* these are used by the chain search */
+	u16    ac_chain;
+	int    ac_disable_chain_relink;
+	group_search_t *ac_group_search;
+
+	u64    ac_last_group;
+	u64    ac_max_block;  /* Highest block number to allocate. 0 is
+				 is the same as ~0 - unlimited */
+
+	int    ac_find_loc_only;  /* hack for reflink operation ordering */
+	struct ocfs2_suballoc_result *ac_find_loc_priv; /* */
+
+	struct ocfs2_alloc_reservation	*ac_resv;
+};
+
+void ocfs2_init_steal_slots(struct ocfs2_super *osb);
+void ocfs2_free_alloc_context(struct ocfs2_alloc_context *ac);
+static inline int ocfs2_alloc_context_bits_left(struct ocfs2_alloc_context *ac)
+{
+	return ac->ac_bits_wanted - ac->ac_bits_given;
+}
+
+/*
+ * Please note that the caller must make sure that root_el is the root
+ * of extent tree. So for an inode, it should be &fe->id2.i_list. Otherwise
+ * the result may be wrong.
+ */
+int ocfs2_reserve_new_metadata(struct ocfs2_super *osb,
+			       struct ocfs2_extent_list *root_el,
+			       struct ocfs2_alloc_context **ac);
+int ocfs2_reserve_new_metadata_blocks(struct ocfs2_super *osb,
+				      int blocks,
+				      struct ocfs2_alloc_context **ac);
+int ocfs2_reserve_new_inode(struct ocfs2_super *osb,
+			    struct ocfs2_alloc_context **ac);
+int ocfs2_reserve_clusters(struct ocfs2_super *osb,
+			   u32 bits_wanted,
+			   struct ocfs2_alloc_context **ac);
+
+int ocfs2_alloc_dinode_update_counts(struct inode *inode,
+			 handle_t *handle,
+			 struct buffer_head *di_bh,
+			 u32 num_bits,
+			 u16 chain);
+void ocfs2_rollback_alloc_dinode_counts(struct inode *inode,
+			 struct buffer_head *di_bh,
+			 u32 num_bits,
+			 u16 chain);
+int ocfs2_block_group_set_bits(handle_t *handle,
+			 struct inode *alloc_inode,
+			 struct ocfs2_group_desc *bg,
+			 struct buffer_head *group_bh,
+			 unsigned int bit_off,
+			 unsigned int num_bits);
+
+int ocfs2_claim_metadata(handle_t *handle,
+			 struct ocfs2_alloc_context *ac,
+			 u32 bits_wanted,
+			 u64 *suballoc_loc,
+			 u16 *suballoc_bit_start,
+			 u32 *num_bits,
+			 u64 *blkno_start);
+int ocfs2_claim_new_inode(handle_t *handle,
+			  struct inode *dir,
+			  struct buffer_head *parent_fe_bh,
+			  struct ocfs2_alloc_context *ac,
+			  u64 *suballoc_loc,
+			  u16 *suballoc_bit,
+			  u64 *fe_blkno);
+int ocfs2_claim_clusters(handle_t *handle,
+			 struct ocfs2_alloc_context *ac,
+			 u32 min_clusters,
+			 u32 *cluster_start,
+			 u32 *num_clusters);
+/*
+ * Use this variant of ocfs2_claim_clusters to specify a maxiumum
+ * number of clusters smaller than the allocation reserved.
+ */
+int __ocfs2_claim_clusters(handle_t *handle,
+			   struct ocfs2_alloc_context *ac,
+			   u32 min_clusters,
+			   u32 max_clusters,
+			   u32 *cluster_start,
+			   u32 *num_clusters);
+
+int ocfs2_free_suballoc_bits(handle_t *handle,
+			     struct inode *alloc_inode,
+			     struct buffer_head *alloc_bh,
+			     unsigned int start_bit,
+			     u64 bg_blkno,
+			     unsigned int count);
+int ocfs2_free_dinode(handle_t *handle,
+		      struct inode *inode_alloc_inode,
+		      struct buffer_head *inode_alloc_bh,
+		      struct ocfs2_dinode *di);
+int ocfs2_free_clusters(handle_t *handle,
+			struct inode *bitmap_inode,
+			struct buffer_head *bitmap_bh,
+			u64 start_blk,
+			unsigned int num_clusters);
+int ocfs2_release_clusters(handle_t *handle,
+			   struct inode *bitmap_inode,
+			   struct buffer_head *bitmap_bh,
+			   u64 start_blk,
+			   unsigned int num_clusters);
+
+static inline u64 ocfs2_which_suballoc_group(u64 block, unsigned int bit)
+{
+	u64 group = block - (u64) bit;
+
+	return group;
+}
+
+static inline u32 ocfs2_cluster_from_desc(struct ocfs2_super *osb,
+					  u64 bg_blkno)
+{
+	/* This should work for all block group descriptors as only
+	 * the 1st group descriptor of the cluster bitmap is
+	 * different. */
+
+	if (bg_blkno == osb->first_cluster_group_blkno)
+		return 0;
+
+	/* the rest of the block groups are located at the beginning
+	 * of their 1st cluster, so a direct translation just
+	 * works. */
+	return ocfs2_blocks_to_clusters(osb->sb, bg_blkno);
+}
+
+static inline int ocfs2_is_cluster_bitmap(struct inode *inode)
+{
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	return osb->bitmap_blkno == OCFS2_I(inode)->ip_blkno;
+}
+
+/* This is for local alloc ONLY. Others should use the task-specific
+ * apis above. */
+int ocfs2_reserve_cluster_bitmap_bits(struct ocfs2_super *osb,
+				      struct ocfs2_alloc_context *ac);
+void ocfs2_free_ac_resource(struct ocfs2_alloc_context *ac);
+
+/* given a cluster offset, calculate which block group it belongs to
+ * and return that block offset. */
+u64 ocfs2_which_cluster_group(struct inode *inode, u32 cluster);
+
+/*
+ * By default, ocfs2_read_group_descriptor() calls ocfs2_error() when it
+ * finds a problem.  A caller that wants to check a group descriptor
+ * without going readonly should read the block with ocfs2_read_block[s]()
+ * and then checking it with this function.  This is only resize, really.
+ * Everyone else should be using ocfs2_read_group_descriptor().
+ */
+int ocfs2_check_group_descriptor(struct super_block *sb,
+				 struct ocfs2_dinode *di,
+				 struct buffer_head *bh);
+/*
+ * Read a group descriptor block into *bh.  If *bh is NULL, a bh will be
+ * allocated.  This is a cached read.  The descriptor will be validated with
+ * ocfs2_validate_group_descriptor().
+ */
+int ocfs2_read_group_descriptor(struct inode *inode, struct ocfs2_dinode *di,
+				u64 gd_blkno, struct buffer_head **bh);
+
+int ocfs2_lock_allocators(struct inode *inode, struct ocfs2_extent_tree *et,
+			  u32 clusters_to_add, u32 extents_to_split,
+			  struct ocfs2_alloc_context **data_ac,
+			  struct ocfs2_alloc_context **meta_ac);
+
+int ocfs2_test_inode_bit(struct ocfs2_super *osb, u64 blkno, int *res);
+
+
+
+/*
+ * The following two interfaces are for ocfs2_create_inode_in_orphan().
+ */
+int ocfs2_find_new_inode_loc(struct inode *dir,
+			     struct buffer_head *parent_fe_bh,
+			     struct ocfs2_alloc_context *ac,
+			     u64 *fe_blkno);
+
+int ocfs2_claim_new_inode_at_loc(handle_t *handle,
+				 struct inode *dir,
+				 struct ocfs2_alloc_context *ac,
+				 u64 *suballoc_loc,
+				 u16 *suballoc_bit,
+				 u64 di_blkno);
+
+#endif /* _CHAINALLOC_H_ */
diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
new file mode 100644
index 000000000..403c5660b
--- /dev/null
+++ b/fs/ocfs2/super.c
@@ -0,0 +1,2644 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * super.c
+ *
+ * load/unload driver, mount/dismount volumes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/init.h>
+#include <linux/random.h>
+#include <linux/statfs.h>
+#include <linux/moduleparam.h>
+#include <linux/blkdev.h>
+#include <linux/socket.h>
+#include <linux/inet.h>
+#include <linux/parser.h>
+#include <linux/crc32.h>
+#include <linux/debugfs.h>
+#include <linux/mount.h>
+#include <linux/seq_file.h>
+#include <linux/quotaops.h>
+#include <linux/cleancache.h>
+
+#define CREATE_TRACE_POINTS
+#include "ocfs2_trace.h"
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+/* this should be the only file to include a version 1 header */
+#include "ocfs1_fs_compat.h"
+
+#include "alloc.h"
+#include "aops.h"
+#include "blockcheck.h"
+#include "dlmglue.h"
+#include "export.h"
+#include "extent_map.h"
+#include "heartbeat.h"
+#include "inode.h"
+#include "journal.h"
+#include "localalloc.h"
+#include "namei.h"
+#include "slot_map.h"
+#include "super.h"
+#include "sysfile.h"
+#include "uptodate.h"
+#include "xattr.h"
+#include "quota.h"
+#include "refcounttree.h"
+#include "suballoc.h"
+
+#include "buffer_head_io.h"
+
+static struct kmem_cache *ocfs2_inode_cachep;
+struct kmem_cache *ocfs2_dquot_cachep;
+struct kmem_cache *ocfs2_qf_chunk_cachep;
+
+/* OCFS2 needs to schedule several different types of work which
+ * require cluster locking, disk I/O, recovery waits, etc. Since these
+ * types of work tend to be heavy we avoid using the kernel events
+ * workqueue and schedule on our own. */
+struct workqueue_struct *ocfs2_wq = NULL;
+
+static struct dentry *ocfs2_debugfs_root;
+
+MODULE_AUTHOR("Oracle");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("OCFS2 cluster file system");
+
+struct mount_options
+{
+	unsigned long	commit_interval;
+	unsigned long	mount_opt;
+	unsigned int	atime_quantum;
+	signed short	slot;
+	int		localalloc_opt;
+	unsigned int	resv_level;
+	int		dir_resv_level;
+	char		cluster_stack[OCFS2_STACK_LABEL_LEN + 1];
+};
+
+static int ocfs2_parse_options(struct super_block *sb, char *options,
+			       struct mount_options *mopt,
+			       int is_remount);
+static int ocfs2_check_set_options(struct super_block *sb,
+				   struct mount_options *options);
+static int ocfs2_show_options(struct seq_file *s, struct dentry *root);
+static void ocfs2_put_super(struct super_block *sb);
+static int ocfs2_mount_volume(struct super_block *sb);
+static int ocfs2_remount(struct super_block *sb, int *flags, char *data);
+static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err);
+static int ocfs2_initialize_mem_caches(void);
+static void ocfs2_free_mem_caches(void);
+static void ocfs2_delete_osb(struct ocfs2_super *osb);
+
+static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf);
+
+static int ocfs2_sync_fs(struct super_block *sb, int wait);
+
+static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb);
+static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb);
+static void ocfs2_release_system_inodes(struct ocfs2_super *osb);
+static int ocfs2_check_volume(struct ocfs2_super *osb);
+static int ocfs2_verify_volume(struct ocfs2_dinode *di,
+			       struct buffer_head *bh,
+			       u32 sectsize,
+			       struct ocfs2_blockcheck_stats *stats);
+static int ocfs2_initialize_super(struct super_block *sb,
+				  struct buffer_head *bh,
+				  int sector_size,
+				  struct ocfs2_blockcheck_stats *stats);
+static int ocfs2_get_sector(struct super_block *sb,
+			    struct buffer_head **bh,
+			    int block,
+			    int sect_size);
+static struct inode *ocfs2_alloc_inode(struct super_block *sb);
+static void ocfs2_destroy_inode(struct inode *inode);
+static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend);
+static int ocfs2_enable_quotas(struct ocfs2_super *osb);
+static void ocfs2_disable_quotas(struct ocfs2_super *osb);
+
+static struct dquot **ocfs2_get_dquots(struct inode *inode)
+{
+	return OCFS2_I(inode)->i_dquot;
+}
+
+static const struct super_operations ocfs2_sops = {
+	.statfs		= ocfs2_statfs,
+	.alloc_inode	= ocfs2_alloc_inode,
+	.destroy_inode	= ocfs2_destroy_inode,
+	.drop_inode	= ocfs2_drop_inode,
+	.evict_inode	= ocfs2_evict_inode,
+	.sync_fs	= ocfs2_sync_fs,
+	.put_super	= ocfs2_put_super,
+	.remount_fs	= ocfs2_remount,
+	.show_options   = ocfs2_show_options,
+	.quota_read	= ocfs2_quota_read,
+	.quota_write	= ocfs2_quota_write,
+	.get_dquots	= ocfs2_get_dquots,
+};
+
+enum {
+	Opt_barrier,
+	Opt_err_panic,
+	Opt_err_ro,
+	Opt_intr,
+	Opt_nointr,
+	Opt_hb_none,
+	Opt_hb_local,
+	Opt_hb_global,
+	Opt_data_ordered,
+	Opt_data_writeback,
+	Opt_atime_quantum,
+	Opt_slot,
+	Opt_commit,
+	Opt_localalloc,
+	Opt_localflocks,
+	Opt_stack,
+	Opt_user_xattr,
+	Opt_nouser_xattr,
+	Opt_inode64,
+	Opt_acl,
+	Opt_noacl,
+	Opt_usrquota,
+	Opt_grpquota,
+	Opt_coherency_buffered,
+	Opt_coherency_full,
+	Opt_resv_level,
+	Opt_dir_resv_level,
+	Opt_journal_async_commit,
+	Opt_err,
+};
+
+static const match_table_t tokens = {
+	{Opt_barrier, "barrier=%u"},
+	{Opt_err_panic, "errors=panic"},
+	{Opt_err_ro, "errors=remount-ro"},
+	{Opt_intr, "intr"},
+	{Opt_nointr, "nointr"},
+	{Opt_hb_none, OCFS2_HB_NONE},
+	{Opt_hb_local, OCFS2_HB_LOCAL},
+	{Opt_hb_global, OCFS2_HB_GLOBAL},
+	{Opt_data_ordered, "data=ordered"},
+	{Opt_data_writeback, "data=writeback"},
+	{Opt_atime_quantum, "atime_quantum=%u"},
+	{Opt_slot, "preferred_slot=%u"},
+	{Opt_commit, "commit=%u"},
+	{Opt_localalloc, "localalloc=%d"},
+	{Opt_localflocks, "localflocks"},
+	{Opt_stack, "cluster_stack=%s"},
+	{Opt_user_xattr, "user_xattr"},
+	{Opt_nouser_xattr, "nouser_xattr"},
+	{Opt_inode64, "inode64"},
+	{Opt_acl, "acl"},
+	{Opt_noacl, "noacl"},
+	{Opt_usrquota, "usrquota"},
+	{Opt_grpquota, "grpquota"},
+	{Opt_coherency_buffered, "coherency=buffered"},
+	{Opt_coherency_full, "coherency=full"},
+	{Opt_resv_level, "resv_level=%u"},
+	{Opt_dir_resv_level, "dir_resv_level=%u"},
+	{Opt_journal_async_commit, "journal_async_commit"},
+	{Opt_err, NULL}
+};
+
+#ifdef CONFIG_DEBUG_FS
+static int ocfs2_osb_dump(struct ocfs2_super *osb, char *buf, int len)
+{
+	struct ocfs2_cluster_connection *cconn = osb->cconn;
+	struct ocfs2_recovery_map *rm = osb->recovery_map;
+	struct ocfs2_orphan_scan *os = &osb->osb_orphan_scan;
+	int i, out = 0;
+
+	out += snprintf(buf + out, len - out,
+			"%10s => Id: %-s  Uuid: %-s  Gen: 0x%X  Label: %-s\n",
+			"Device", osb->dev_str, osb->uuid_str,
+			osb->fs_generation, osb->vol_label);
+
+	out += snprintf(buf + out, len - out,
+			"%10s => State: %d  Flags: 0x%lX\n", "Volume",
+			atomic_read(&osb->vol_state), osb->osb_flags);
+
+	out += snprintf(buf + out, len - out,
+			"%10s => Block: %lu  Cluster: %d\n", "Sizes",
+			osb->sb->s_blocksize, osb->s_clustersize);
+
+	out += snprintf(buf + out, len - out,
+			"%10s => Compat: 0x%X  Incompat: 0x%X  "
+			"ROcompat: 0x%X\n",
+			"Features", osb->s_feature_compat,
+			osb->s_feature_incompat, osb->s_feature_ro_compat);
+
+	out += snprintf(buf + out, len - out,
+			"%10s => Opts: 0x%lX  AtimeQuanta: %u\n", "Mount",
+			osb->s_mount_opt, osb->s_atime_quantum);
+
+	if (cconn) {
+		out += snprintf(buf + out, len - out,
+				"%10s => Stack: %s  Name: %*s  "
+				"Version: %d.%d\n", "Cluster",
+				(*osb->osb_cluster_stack == '\0' ?
+				 "o2cb" : osb->osb_cluster_stack),
+				cconn->cc_namelen, cconn->cc_name,
+				cconn->cc_version.pv_major,
+				cconn->cc_version.pv_minor);
+	}
+
+	spin_lock(&osb->dc_task_lock);
+	out += snprintf(buf + out, len - out,
+			"%10s => Pid: %d  Count: %lu  WakeSeq: %lu  "
+			"WorkSeq: %lu\n", "DownCnvt",
+			(osb->dc_task ?  task_pid_nr(osb->dc_task) : -1),
+			osb->blocked_lock_count, osb->dc_wake_sequence,
+			osb->dc_work_sequence);
+	spin_unlock(&osb->dc_task_lock);
+
+	spin_lock(&osb->osb_lock);
+	out += snprintf(buf + out, len - out, "%10s => Pid: %d  Nodes:",
+			"Recovery",
+			(osb->recovery_thread_task ?
+			 task_pid_nr(osb->recovery_thread_task) : -1));
+	if (rm->rm_used == 0)
+		out += snprintf(buf + out, len - out, " None\n");
+	else {
+		for (i = 0; i < rm->rm_used; i++)
+			out += snprintf(buf + out, len - out, " %d",
+					rm->rm_entries[i]);
+		out += snprintf(buf + out, len - out, "\n");
+	}
+	spin_unlock(&osb->osb_lock);
+
+	out += snprintf(buf + out, len - out,
+			"%10s => Pid: %d  Interval: %lu\n", "Commit",
+			(osb->commit_task ? task_pid_nr(osb->commit_task) : -1),
+			osb->osb_commit_interval);
+
+	out += snprintf(buf + out, len - out,
+			"%10s => State: %d  TxnId: %lu  NumTxns: %d\n",
+			"Journal", osb->journal->j_state,
+			osb->journal->j_trans_id,
+			atomic_read(&osb->journal->j_num_trans));
+
+	out += snprintf(buf + out, len - out,
+			"%10s => GlobalAllocs: %d  LocalAllocs: %d  "
+			"SubAllocs: %d  LAWinMoves: %d  SAExtends: %d\n",
+			"Stats",
+			atomic_read(&osb->alloc_stats.bitmap_data),
+			atomic_read(&osb->alloc_stats.local_data),
+			atomic_read(&osb->alloc_stats.bg_allocs),
+			atomic_read(&osb->alloc_stats.moves),
+			atomic_read(&osb->alloc_stats.bg_extends));
+
+	out += snprintf(buf + out, len - out,
+			"%10s => State: %u  Descriptor: %llu  Size: %u bits  "
+			"Default: %u bits\n",
+			"LocalAlloc", osb->local_alloc_state,
+			(unsigned long long)osb->la_last_gd,
+			osb->local_alloc_bits, osb->local_alloc_default_bits);
+
+	spin_lock(&osb->osb_lock);
+	out += snprintf(buf + out, len - out,
+			"%10s => InodeSlot: %d  StolenInodes: %d, "
+			"MetaSlot: %d  StolenMeta: %d\n", "Steal",
+			osb->s_inode_steal_slot,
+			atomic_read(&osb->s_num_inodes_stolen),
+			osb->s_meta_steal_slot,
+			atomic_read(&osb->s_num_meta_stolen));
+	spin_unlock(&osb->osb_lock);
+
+	out += snprintf(buf + out, len - out, "OrphanScan => ");
+	out += snprintf(buf + out, len - out, "Local: %u  Global: %u ",
+			os->os_count, os->os_seqno);
+	out += snprintf(buf + out, len - out, " Last Scan: ");
+	if (atomic_read(&os->os_state) == ORPHAN_SCAN_INACTIVE)
+		out += snprintf(buf + out, len - out, "Disabled\n");
+	else
+		out += snprintf(buf + out, len - out, "%lu seconds ago\n",
+				(get_seconds() - os->os_scantime.tv_sec));
+
+	out += snprintf(buf + out, len - out, "%10s => %3s  %10s\n",
+			"Slots", "Num", "RecoGen");
+	for (i = 0; i < osb->max_slots; ++i) {
+		out += snprintf(buf + out, len - out,
+				"%10s  %c %3d  %10d\n",
+				" ",
+				(i == osb->slot_num ? '*' : ' '),
+				i, osb->slot_recovery_generations[i]);
+	}
+
+	return out;
+}
+
+static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
+{
+	struct ocfs2_super *osb = inode->i_private;
+	char *buf = NULL;
+
+	buf = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!buf)
+		goto bail;
+
+	i_size_write(inode, ocfs2_osb_dump(osb, buf, PAGE_SIZE));
+
+	file->private_data = buf;
+
+	return 0;
+bail:
+	return -ENOMEM;
+}
+
+static int ocfs2_debug_release(struct inode *inode, struct file *file)
+{
+	kfree(file->private_data);
+	return 0;
+}
+
+static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
+				size_t nbytes, loff_t *ppos)
+{
+	return simple_read_from_buffer(buf, nbytes, ppos, file->private_data,
+				       i_size_read(file->f_mapping->host));
+}
+#else
+static int ocfs2_osb_debug_open(struct inode *inode, struct file *file)
+{
+	return 0;
+}
+static int ocfs2_debug_release(struct inode *inode, struct file *file)
+{
+	return 0;
+}
+static ssize_t ocfs2_debug_read(struct file *file, char __user *buf,
+				size_t nbytes, loff_t *ppos)
+{
+	return 0;
+}
+#endif	/* CONFIG_DEBUG_FS */
+
+static const struct file_operations ocfs2_osb_debug_fops = {
+	.open =		ocfs2_osb_debug_open,
+	.release =	ocfs2_debug_release,
+	.read =		ocfs2_debug_read,
+	.llseek =	generic_file_llseek,
+};
+
+static int ocfs2_sync_fs(struct super_block *sb, int wait)
+{
+	int status;
+	tid_t target;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+
+	if (ocfs2_is_hard_readonly(osb))
+		return -EROFS;
+
+	if (wait) {
+		status = ocfs2_flush_truncate_log(osb);
+		if (status < 0)
+			mlog_errno(status);
+	} else {
+		ocfs2_schedule_truncate_log_flush(osb, 0);
+	}
+
+	if (jbd2_journal_start_commit(OCFS2_SB(sb)->journal->j_journal,
+				      &target)) {
+		if (wait)
+			jbd2_log_wait_commit(OCFS2_SB(sb)->journal->j_journal,
+					     target);
+	}
+	return 0;
+}
+
+static int ocfs2_need_system_inode(struct ocfs2_super *osb, int ino)
+{
+	if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_USRQUOTA)
+	    && (ino == USER_QUOTA_SYSTEM_INODE
+		|| ino == LOCAL_USER_QUOTA_SYSTEM_INODE))
+		return 0;
+	if (!OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)
+	    && (ino == GROUP_QUOTA_SYSTEM_INODE
+		|| ino == LOCAL_GROUP_QUOTA_SYSTEM_INODE))
+		return 0;
+	return 1;
+}
+
+static int ocfs2_init_global_system_inodes(struct ocfs2_super *osb)
+{
+	struct inode *new = NULL;
+	int status = 0;
+	int i;
+
+	new = ocfs2_iget(osb, osb->root_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
+	if (IS_ERR(new)) {
+		status = PTR_ERR(new);
+		mlog_errno(status);
+		goto bail;
+	}
+	osb->root_inode = new;
+
+	new = ocfs2_iget(osb, osb->system_dir_blkno, OCFS2_FI_FLAG_SYSFILE, 0);
+	if (IS_ERR(new)) {
+		status = PTR_ERR(new);
+		mlog_errno(status);
+		goto bail;
+	}
+	osb->sys_root_inode = new;
+
+	for (i = OCFS2_FIRST_ONLINE_SYSTEM_INODE;
+	     i <= OCFS2_LAST_GLOBAL_SYSTEM_INODE; i++) {
+		if (!ocfs2_need_system_inode(osb, i))
+			continue;
+		new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
+		if (!new) {
+			ocfs2_release_system_inodes(osb);
+			status = -EINVAL;
+			mlog_errno(status);
+			/* FIXME: Should ERROR_RO_FS */
+			mlog(ML_ERROR, "Unable to load system inode %d, "
+			     "possibly corrupt fs?", i);
+			goto bail;
+		}
+		// the array now has one ref, so drop this one
+		iput(new);
+	}
+
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static int ocfs2_init_local_system_inodes(struct ocfs2_super *osb)
+{
+	struct inode *new = NULL;
+	int status = 0;
+	int i;
+
+	for (i = OCFS2_LAST_GLOBAL_SYSTEM_INODE + 1;
+	     i < NUM_SYSTEM_INODES;
+	     i++) {
+		if (!ocfs2_need_system_inode(osb, i))
+			continue;
+		new = ocfs2_get_system_file_inode(osb, i, osb->slot_num);
+		if (!new) {
+			ocfs2_release_system_inodes(osb);
+			status = -EINVAL;
+			mlog(ML_ERROR, "status=%d, sysfile=%d, slot=%d\n",
+			     status, i, osb->slot_num);
+			goto bail;
+		}
+		/* the array now has one ref, so drop this one */
+		iput(new);
+	}
+
+bail:
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static void ocfs2_release_system_inodes(struct ocfs2_super *osb)
+{
+	int i;
+	struct inode *inode;
+
+	for (i = 0; i < NUM_GLOBAL_SYSTEM_INODES; i++) {
+		inode = osb->global_system_inodes[i];
+		if (inode) {
+			iput(inode);
+			osb->global_system_inodes[i] = NULL;
+		}
+	}
+
+	inode = osb->sys_root_inode;
+	if (inode) {
+		iput(inode);
+		osb->sys_root_inode = NULL;
+	}
+
+	inode = osb->root_inode;
+	if (inode) {
+		iput(inode);
+		osb->root_inode = NULL;
+	}
+
+	if (!osb->local_system_inodes)
+		return;
+
+	for (i = 0; i < NUM_LOCAL_SYSTEM_INODES * osb->max_slots; i++) {
+		if (osb->local_system_inodes[i]) {
+			iput(osb->local_system_inodes[i]);
+			osb->local_system_inodes[i] = NULL;
+		}
+	}
+
+	kfree(osb->local_system_inodes);
+	osb->local_system_inodes = NULL;
+}
+
+/* We're allocating fs objects, use GFP_NOFS */
+static struct inode *ocfs2_alloc_inode(struct super_block *sb)
+{
+	struct ocfs2_inode_info *oi;
+
+	oi = kmem_cache_alloc(ocfs2_inode_cachep, GFP_NOFS);
+	if (!oi)
+		return NULL;
+
+	oi->i_sync_tid = 0;
+	oi->i_datasync_tid = 0;
+	memset(&oi->i_dquot, 0, sizeof(oi->i_dquot));
+
+	jbd2_journal_init_jbd_inode(&oi->ip_jinode, &oi->vfs_inode);
+	return &oi->vfs_inode;
+}
+
+static void ocfs2_i_callback(struct rcu_head *head)
+{
+	struct inode *inode = container_of(head, struct inode, i_rcu);
+	kmem_cache_free(ocfs2_inode_cachep, OCFS2_I(inode));
+}
+
+static void ocfs2_destroy_inode(struct inode *inode)
+{
+	call_rcu(&inode->i_rcu, ocfs2_i_callback);
+}
+
+static unsigned long long ocfs2_max_file_offset(unsigned int bbits,
+						unsigned int cbits)
+{
+	unsigned int bytes = 1 << cbits;
+	unsigned int trim = bytes;
+	unsigned int bitshift = 32;
+
+	/*
+	 * i_size and all block offsets in ocfs2 are always 64 bits
+	 * wide. i_clusters is 32 bits, in cluster-sized units. So on
+	 * 64 bit platforms, cluster size will be the limiting factor.
+	 */
+
+#if BITS_PER_LONG == 32
+# if defined(CONFIG_LBDAF)
+	BUILD_BUG_ON(sizeof(sector_t) != 8);
+	/*
+	 * We might be limited by page cache size.
+	 */
+	if (bytes > PAGE_CACHE_SIZE) {
+		bytes = PAGE_CACHE_SIZE;
+		trim = 1;
+		/*
+		 * Shift by 31 here so that we don't get larger than
+		 * MAX_LFS_FILESIZE
+		 */
+		bitshift = 31;
+	}
+# else
+	/*
+	 * We are limited by the size of sector_t. Use block size, as
+	 * that's what we expose to the VFS.
+	 */
+	bytes = 1 << bbits;
+	trim = 1;
+	bitshift = 31;
+# endif
+#endif
+
+	/*
+	 * Trim by a whole cluster when we can actually approach the
+	 * on-disk limits. Otherwise we can overflow i_clusters when
+	 * an extent start is at the max offset.
+	 */
+	return (((unsigned long long)bytes) << bitshift) - trim;
+}
+
+static int ocfs2_remount(struct super_block *sb, int *flags, char *data)
+{
+	int incompat_features;
+	int ret = 0;
+	struct mount_options parsed_options;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+	u32 tmp;
+
+	sync_filesystem(sb);
+
+	if (!ocfs2_parse_options(sb, data, &parsed_options, 1) ||
+	    !ocfs2_check_set_options(sb, &parsed_options)) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	tmp = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL |
+		OCFS2_MOUNT_HB_NONE;
+	if ((osb->s_mount_opt & tmp) != (parsed_options.mount_opt & tmp)) {
+		ret = -EINVAL;
+		mlog(ML_ERROR, "Cannot change heartbeat mode on remount\n");
+		goto out;
+	}
+
+	if ((osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK) !=
+	    (parsed_options.mount_opt & OCFS2_MOUNT_DATA_WRITEBACK)) {
+		ret = -EINVAL;
+		mlog(ML_ERROR, "Cannot change data mode on remount\n");
+		goto out;
+	}
+
+	/* Probably don't want this on remount; it might
+	 * mess with other nodes */
+	if (!(osb->s_mount_opt & OCFS2_MOUNT_INODE64) &&
+	    (parsed_options.mount_opt & OCFS2_MOUNT_INODE64)) {
+		ret = -EINVAL;
+		mlog(ML_ERROR, "Cannot enable inode64 on remount\n");
+		goto out;
+	}
+
+	/* We're going to/from readonly mode. */
+	if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
+		/* Disable quota accounting before remounting RO */
+		if (*flags & MS_RDONLY) {
+			ret = ocfs2_susp_quotas(osb, 0);
+			if (ret < 0)
+				goto out;
+		}
+		/* Lock here so the check of HARD_RO and the potential
+		 * setting of SOFT_RO is atomic. */
+		spin_lock(&osb->osb_lock);
+		if (osb->osb_flags & OCFS2_OSB_HARD_RO) {
+			mlog(ML_ERROR, "Remount on readonly device is forbidden.\n");
+			ret = -EROFS;
+			goto unlock_osb;
+		}
+
+		if (*flags & MS_RDONLY) {
+			sb->s_flags |= MS_RDONLY;
+			osb->osb_flags |= OCFS2_OSB_SOFT_RO;
+		} else {
+			if (osb->osb_flags & OCFS2_OSB_ERROR_FS) {
+				mlog(ML_ERROR, "Cannot remount RDWR "
+				     "filesystem due to previous errors.\n");
+				ret = -EROFS;
+				goto unlock_osb;
+			}
+			incompat_features = OCFS2_HAS_RO_COMPAT_FEATURE(sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP);
+			if (incompat_features) {
+				mlog(ML_ERROR, "Cannot remount RDWR because "
+				     "of unsupported optional features "
+				     "(%x).\n", incompat_features);
+				ret = -EINVAL;
+				goto unlock_osb;
+			}
+			sb->s_flags &= ~MS_RDONLY;
+			osb->osb_flags &= ~OCFS2_OSB_SOFT_RO;
+		}
+		trace_ocfs2_remount(sb->s_flags, osb->osb_flags, *flags);
+unlock_osb:
+		spin_unlock(&osb->osb_lock);
+		/* Enable quota accounting after remounting RW */
+		if (!ret && !(*flags & MS_RDONLY)) {
+			if (sb_any_quota_suspended(sb))
+				ret = ocfs2_susp_quotas(osb, 1);
+			else
+				ret = ocfs2_enable_quotas(osb);
+			if (ret < 0) {
+				/* Return back changes... */
+				spin_lock(&osb->osb_lock);
+				sb->s_flags |= MS_RDONLY;
+				osb->osb_flags |= OCFS2_OSB_SOFT_RO;
+				spin_unlock(&osb->osb_lock);
+				goto out;
+			}
+		}
+	}
+
+	if (!ret) {
+		/* Only save off the new mount options in case of a successful
+		 * remount. */
+		osb->s_mount_opt = parsed_options.mount_opt;
+		osb->s_atime_quantum = parsed_options.atime_quantum;
+		osb->preferred_slot = parsed_options.slot;
+		if (parsed_options.commit_interval)
+			osb->osb_commit_interval = parsed_options.commit_interval;
+
+		if (!ocfs2_is_hard_readonly(osb))
+			ocfs2_set_journal_params(osb);
+
+		sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
+			((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ?
+							MS_POSIXACL : 0);
+	}
+out:
+	return ret;
+}
+
+static int ocfs2_sb_probe(struct super_block *sb,
+			  struct buffer_head **bh,
+			  int *sector_size,
+			  struct ocfs2_blockcheck_stats *stats)
+{
+	int status, tmpstat;
+	struct ocfs1_vol_disk_hdr *hdr;
+	struct ocfs2_dinode *di;
+	int blksize;
+
+	*bh = NULL;
+
+	/* may be > 512 */
+	*sector_size = bdev_logical_block_size(sb->s_bdev);
+	if (*sector_size > OCFS2_MAX_BLOCKSIZE) {
+		mlog(ML_ERROR, "Hardware sector size too large: %d (max=%d)\n",
+		     *sector_size, OCFS2_MAX_BLOCKSIZE);
+		status = -EINVAL;
+		goto bail;
+	}
+
+	/* Can this really happen? */
+	if (*sector_size < OCFS2_MIN_BLOCKSIZE)
+		*sector_size = OCFS2_MIN_BLOCKSIZE;
+
+	/* check block zero for old format */
+	status = ocfs2_get_sector(sb, bh, 0, *sector_size);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	hdr = (struct ocfs1_vol_disk_hdr *) (*bh)->b_data;
+	if (hdr->major_version == OCFS1_MAJOR_VERSION) {
+		mlog(ML_ERROR, "incompatible version: %u.%u\n",
+		     hdr->major_version, hdr->minor_version);
+		status = -EINVAL;
+	}
+	if (memcmp(hdr->signature, OCFS1_VOLUME_SIGNATURE,
+		   strlen(OCFS1_VOLUME_SIGNATURE)) == 0) {
+		mlog(ML_ERROR, "incompatible volume signature: %8s\n",
+		     hdr->signature);
+		status = -EINVAL;
+	}
+	brelse(*bh);
+	*bh = NULL;
+	if (status < 0) {
+		mlog(ML_ERROR, "This is an ocfs v1 filesystem which must be "
+		     "upgraded before mounting with ocfs v2\n");
+		goto bail;
+	}
+
+	/*
+	 * Now check at magic offset for 512, 1024, 2048, 4096
+	 * blocksizes.  4096 is the maximum blocksize because it is
+	 * the minimum clustersize.
+	 */
+	status = -EINVAL;
+	for (blksize = *sector_size;
+	     blksize <= OCFS2_MAX_BLOCKSIZE;
+	     blksize <<= 1) {
+		tmpstat = ocfs2_get_sector(sb, bh,
+					   OCFS2_SUPER_BLOCK_BLKNO,
+					   blksize);
+		if (tmpstat < 0) {
+			status = tmpstat;
+			mlog_errno(status);
+			break;
+		}
+		di = (struct ocfs2_dinode *) (*bh)->b_data;
+		memset(stats, 0, sizeof(struct ocfs2_blockcheck_stats));
+		spin_lock_init(&stats->b_lock);
+		tmpstat = ocfs2_verify_volume(di, *bh, blksize, stats);
+		if (tmpstat < 0) {
+			brelse(*bh);
+			*bh = NULL;
+		}
+		if (tmpstat != -EAGAIN) {
+			status = tmpstat;
+			break;
+		}
+	}
+
+bail:
+	return status;
+}
+
+static int ocfs2_verify_heartbeat(struct ocfs2_super *osb)
+{
+	u32 hb_enabled = OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL;
+
+	if (osb->s_mount_opt & hb_enabled) {
+		if (ocfs2_mount_local(osb)) {
+			mlog(ML_ERROR, "Cannot heartbeat on a locally "
+			     "mounted device.\n");
+			return -EINVAL;
+		}
+		if (ocfs2_userspace_stack(osb)) {
+			mlog(ML_ERROR, "Userspace stack expected, but "
+			     "o2cb heartbeat arguments passed to mount\n");
+			return -EINVAL;
+		}
+		if (((osb->s_mount_opt & OCFS2_MOUNT_HB_GLOBAL) &&
+		     !ocfs2_cluster_o2cb_global_heartbeat(osb)) ||
+		    ((osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) &&
+		     ocfs2_cluster_o2cb_global_heartbeat(osb))) {
+			mlog(ML_ERROR, "Mismatching o2cb heartbeat modes\n");
+			return -EINVAL;
+		}
+	}
+
+	if (!(osb->s_mount_opt & hb_enabled)) {
+		if (!ocfs2_mount_local(osb) && !ocfs2_is_hard_readonly(osb) &&
+		    !ocfs2_userspace_stack(osb)) {
+			mlog(ML_ERROR, "Heartbeat has to be started to mount "
+			     "a read-write clustered device.\n");
+			return -EINVAL;
+		}
+	}
+
+	return 0;
+}
+
+/*
+ * If we're using a userspace stack, mount should have passed
+ * a name that matches the disk.  If not, mount should not
+ * have passed a stack.
+ */
+static int ocfs2_verify_userspace_stack(struct ocfs2_super *osb,
+					struct mount_options *mopt)
+{
+	if (!ocfs2_userspace_stack(osb) && mopt->cluster_stack[0]) {
+		mlog(ML_ERROR,
+		     "cluster stack passed to mount, but this filesystem "
+		     "does not support it\n");
+		return -EINVAL;
+	}
+
+	if (ocfs2_userspace_stack(osb) &&
+	    strncmp(osb->osb_cluster_stack, mopt->cluster_stack,
+		    OCFS2_STACK_LABEL_LEN)) {
+		mlog(ML_ERROR,
+		     "cluster stack passed to mount (\"%s\") does not "
+		     "match the filesystem (\"%s\")\n",
+		     mopt->cluster_stack,
+		     osb->osb_cluster_stack);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int ocfs2_susp_quotas(struct ocfs2_super *osb, int unsuspend)
+{
+	int type;
+	struct super_block *sb = osb->sb;
+	unsigned int feature[OCFS2_MAXQUOTAS] = {
+					OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
+					OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
+	int status = 0;
+
+	for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
+		if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
+			continue;
+		if (unsuspend)
+			status = dquot_resume(sb, type);
+		else {
+			struct ocfs2_mem_dqinfo *oinfo;
+
+			/* Cancel periodic syncing before suspending */
+			oinfo = sb_dqinfo(sb, type)->dqi_priv;
+			cancel_delayed_work_sync(&oinfo->dqi_sync_work);
+			status = dquot_suspend(sb, type);
+		}
+		if (status < 0)
+			break;
+	}
+	if (status < 0)
+		mlog(ML_ERROR, "Failed to suspend/unsuspend quotas on "
+		     "remount (error = %d).\n", status);
+	return status;
+}
+
+static int ocfs2_enable_quotas(struct ocfs2_super *osb)
+{
+	struct inode *inode[OCFS2_MAXQUOTAS] = { NULL, NULL };
+	struct super_block *sb = osb->sb;
+	unsigned int feature[OCFS2_MAXQUOTAS] = {
+					OCFS2_FEATURE_RO_COMPAT_USRQUOTA,
+					OCFS2_FEATURE_RO_COMPAT_GRPQUOTA};
+	unsigned int ino[OCFS2_MAXQUOTAS] = {
+					LOCAL_USER_QUOTA_SYSTEM_INODE,
+					LOCAL_GROUP_QUOTA_SYSTEM_INODE };
+	int status;
+	int type;
+
+	sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE | DQUOT_NEGATIVE_USAGE;
+	for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
+		if (!OCFS2_HAS_RO_COMPAT_FEATURE(sb, feature[type]))
+			continue;
+		inode[type] = ocfs2_get_system_file_inode(osb, ino[type],
+							osb->slot_num);
+		if (!inode[type]) {
+			status = -ENOENT;
+			goto out_quota_off;
+		}
+		status = dquot_enable(inode[type], type, QFMT_OCFS2,
+				      DQUOT_USAGE_ENABLED);
+		if (status < 0)
+			goto out_quota_off;
+	}
+
+	for (type = 0; type < OCFS2_MAXQUOTAS; type++)
+		iput(inode[type]);
+	return 0;
+out_quota_off:
+	ocfs2_disable_quotas(osb);
+	for (type = 0; type < OCFS2_MAXQUOTAS; type++)
+		iput(inode[type]);
+	mlog_errno(status);
+	return status;
+}
+
+static void ocfs2_disable_quotas(struct ocfs2_super *osb)
+{
+	int type;
+	struct inode *inode;
+	struct super_block *sb = osb->sb;
+	struct ocfs2_mem_dqinfo *oinfo;
+
+	/* We mostly ignore errors in this function because there's not much
+	 * we can do when we see them */
+	for (type = 0; type < OCFS2_MAXQUOTAS; type++) {
+		if (!sb_has_quota_loaded(sb, type))
+			continue;
+		/* Cancel periodic syncing before we grab dqonoff_mutex */
+		oinfo = sb_dqinfo(sb, type)->dqi_priv;
+		cancel_delayed_work_sync(&oinfo->dqi_sync_work);
+		inode = igrab(sb->s_dquot.files[type]);
+		/* Turn off quotas. This will remove all dquot structures from
+		 * memory and so they will be automatically synced to global
+		 * quota files */
+		dquot_disable(sb, type, DQUOT_USAGE_ENABLED |
+					DQUOT_LIMITS_ENABLED);
+		if (!inode)
+			continue;
+		iput(inode);
+	}
+}
+
+static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
+{
+	struct dentry *root;
+	int status, sector_size;
+	struct mount_options parsed_options;
+	struct inode *inode = NULL;
+	struct ocfs2_super *osb = NULL;
+	struct buffer_head *bh = NULL;
+	char nodestr[12];
+	struct ocfs2_blockcheck_stats stats;
+
+	trace_ocfs2_fill_super(sb, data, silent);
+
+	if (!ocfs2_parse_options(sb, data, &parsed_options, 0)) {
+		status = -EINVAL;
+		goto read_super_error;
+	}
+
+	/* probe for superblock */
+	status = ocfs2_sb_probe(sb, &bh, &sector_size, &stats);
+	if (status < 0) {
+		mlog(ML_ERROR, "superblock probe failed!\n");
+		goto read_super_error;
+	}
+
+	status = ocfs2_initialize_super(sb, bh, sector_size, &stats);
+	osb = OCFS2_SB(sb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto read_super_error;
+	}
+	brelse(bh);
+	bh = NULL;
+
+	if (!ocfs2_check_set_options(sb, &parsed_options)) {
+		status = -EINVAL;
+		goto read_super_error;
+	}
+	osb->s_mount_opt = parsed_options.mount_opt;
+	osb->s_atime_quantum = parsed_options.atime_quantum;
+	osb->preferred_slot = parsed_options.slot;
+	osb->osb_commit_interval = parsed_options.commit_interval;
+
+	ocfs2_la_set_sizes(osb, parsed_options.localalloc_opt);
+	osb->osb_resv_level = parsed_options.resv_level;
+	osb->osb_dir_resv_level = parsed_options.resv_level;
+	if (parsed_options.dir_resv_level == -1)
+		osb->osb_dir_resv_level = parsed_options.resv_level;
+	else
+		osb->osb_dir_resv_level = parsed_options.dir_resv_level;
+
+	status = ocfs2_verify_userspace_stack(osb, &parsed_options);
+	if (status)
+		goto read_super_error;
+
+	sb->s_magic = OCFS2_SUPER_MAGIC;
+
+	sb->s_flags = (sb->s_flags & ~(MS_POSIXACL | MS_NOSEC)) |
+		((osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) ? MS_POSIXACL : 0);
+
+	/* Hard readonly mode only if: bdev_read_only, MS_RDONLY,
+	 * heartbeat=none */
+	if (bdev_read_only(sb->s_bdev)) {
+		if (!(sb->s_flags & MS_RDONLY)) {
+			status = -EACCES;
+			mlog(ML_ERROR, "Readonly device detected but readonly "
+			     "mount was not specified.\n");
+			goto read_super_error;
+		}
+
+		/* You should not be able to start a local heartbeat
+		 * on a readonly device. */
+		if (osb->s_mount_opt & OCFS2_MOUNT_HB_LOCAL) {
+			status = -EROFS;
+			mlog(ML_ERROR, "Local heartbeat specified on readonly "
+			     "device.\n");
+			goto read_super_error;
+		}
+
+		status = ocfs2_check_journals_nolocks(osb);
+		if (status < 0) {
+			if (status == -EROFS)
+				mlog(ML_ERROR, "Recovery required on readonly "
+				     "file system, but write access is "
+				     "unavailable.\n");
+			else
+				mlog_errno(status);
+			goto read_super_error;
+		}
+
+		ocfs2_set_ro_flag(osb, 1);
+
+		printk(KERN_NOTICE "ocfs2: Readonly device (%s) detected. "
+		       "Cluster services will not be used for this mount. "
+		       "Recovery will be skipped.\n", osb->dev_str);
+	}
+
+	if (!ocfs2_is_hard_readonly(osb)) {
+		if (sb->s_flags & MS_RDONLY)
+			ocfs2_set_ro_flag(osb, 0);
+	}
+
+	status = ocfs2_verify_heartbeat(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto read_super_error;
+	}
+
+	osb->osb_debug_root = debugfs_create_dir(osb->uuid_str,
+						 ocfs2_debugfs_root);
+	if (!osb->osb_debug_root) {
+		status = -EINVAL;
+		mlog(ML_ERROR, "Unable to create per-mount debugfs root.\n");
+		goto read_super_error;
+	}
+
+	osb->osb_ctxt = debugfs_create_file("fs_state", S_IFREG|S_IRUSR,
+					    osb->osb_debug_root,
+					    osb,
+					    &ocfs2_osb_debug_fops);
+	if (!osb->osb_ctxt) {
+		status = -EINVAL;
+		mlog_errno(status);
+		goto read_super_error;
+	}
+
+	if (ocfs2_meta_ecc(osb)) {
+		status = ocfs2_blockcheck_stats_debugfs_install(
+						&osb->osb_ecc_stats,
+						osb->osb_debug_root);
+		if (status) {
+			mlog(ML_ERROR,
+			     "Unable to create blockcheck statistics "
+			     "files\n");
+			goto read_super_error;
+		}
+	}
+
+	status = ocfs2_mount_volume(sb);
+	if (status < 0)
+		goto read_super_error;
+
+	if (osb->root_inode)
+		inode = igrab(osb->root_inode);
+
+	if (!inode) {
+		status = -EIO;
+		mlog_errno(status);
+		goto read_super_error;
+	}
+
+	root = d_make_root(inode);
+	if (!root) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto read_super_error;
+	}
+
+	sb->s_root = root;
+
+	ocfs2_complete_mount_recovery(osb);
+
+	if (ocfs2_mount_local(osb))
+		snprintf(nodestr, sizeof(nodestr), "local");
+	else
+		snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
+
+	printk(KERN_INFO "ocfs2: Mounting device (%s) on (node %s, slot %d) "
+	       "with %s data mode.\n",
+	       osb->dev_str, nodestr, osb->slot_num,
+	       osb->s_mount_opt & OCFS2_MOUNT_DATA_WRITEBACK ? "writeback" :
+	       "ordered");
+
+	atomic_set(&osb->vol_state, VOLUME_MOUNTED);
+	wake_up(&osb->osb_mount_event);
+
+	/* Now we can initialize quotas because we can afford to wait
+	 * for cluster locks recovery now. That also means that truncation
+	 * log recovery can happen but that waits for proper quota setup */
+	if (!(sb->s_flags & MS_RDONLY)) {
+		status = ocfs2_enable_quotas(osb);
+		if (status < 0) {
+			/* We have to err-out specially here because
+			 * s_root is already set */
+			mlog_errno(status);
+			atomic_set(&osb->vol_state, VOLUME_DISABLED);
+			wake_up(&osb->osb_mount_event);
+			return status;
+		}
+	}
+
+	ocfs2_complete_quota_recovery(osb);
+
+	/* Now we wake up again for processes waiting for quotas */
+	atomic_set(&osb->vol_state, VOLUME_MOUNTED_QUOTAS);
+	wake_up(&osb->osb_mount_event);
+
+	/* Start this when the mount is almost sure of being successful */
+	ocfs2_orphan_scan_start(osb);
+
+	return status;
+
+read_super_error:
+	brelse(bh);
+
+	if (osb) {
+		atomic_set(&osb->vol_state, VOLUME_DISABLED);
+		wake_up(&osb->osb_mount_event);
+		ocfs2_dismount_volume(sb, 1);
+	}
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+static struct dentry *ocfs2_mount(struct file_system_type *fs_type,
+			int flags,
+			const char *dev_name,
+			void *data)
+{
+	return mount_bdev(fs_type, flags, dev_name, data, ocfs2_fill_super);
+}
+
+static struct file_system_type ocfs2_fs_type = {
+	.owner          = THIS_MODULE,
+	.name           = "ocfs2",
+	.mount          = ocfs2_mount,
+	.kill_sb        = kill_block_super,
+	.fs_flags       = FS_REQUIRES_DEV|FS_RENAME_DOES_D_MOVE,
+	.next           = NULL
+};
+MODULE_ALIAS_FS("ocfs2");
+
+static int ocfs2_check_set_options(struct super_block *sb,
+				   struct mount_options *options)
+{
+	if (options->mount_opt & OCFS2_MOUNT_USRQUOTA &&
+	    !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
+					 OCFS2_FEATURE_RO_COMPAT_USRQUOTA)) {
+		mlog(ML_ERROR, "User quotas were requested, but this "
+		     "filesystem does not have the feature enabled.\n");
+		return 0;
+	}
+	if (options->mount_opt & OCFS2_MOUNT_GRPQUOTA &&
+	    !OCFS2_HAS_RO_COMPAT_FEATURE(sb,
+					 OCFS2_FEATURE_RO_COMPAT_GRPQUOTA)) {
+		mlog(ML_ERROR, "Group quotas were requested, but this "
+		     "filesystem does not have the feature enabled.\n");
+		return 0;
+	}
+	if (options->mount_opt & OCFS2_MOUNT_POSIX_ACL &&
+	    !OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR)) {
+		mlog(ML_ERROR, "ACL support requested but extended attributes "
+		     "feature is not enabled\n");
+		return 0;
+	}
+	/* No ACL setting specified? Use XATTR feature... */
+	if (!(options->mount_opt & (OCFS2_MOUNT_POSIX_ACL |
+				    OCFS2_MOUNT_NO_POSIX_ACL))) {
+		if (OCFS2_HAS_INCOMPAT_FEATURE(sb, OCFS2_FEATURE_INCOMPAT_XATTR))
+			options->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
+		else
+			options->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
+	}
+	return 1;
+}
+
+static int ocfs2_parse_options(struct super_block *sb,
+			       char *options,
+			       struct mount_options *mopt,
+			       int is_remount)
+{
+	int status, user_stack = 0;
+	char *p;
+	u32 tmp;
+
+	trace_ocfs2_parse_options(is_remount, options ? options : "(none)");
+
+	mopt->commit_interval = 0;
+	mopt->mount_opt = OCFS2_MOUNT_NOINTR;
+	mopt->atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
+	mopt->slot = OCFS2_INVALID_SLOT;
+	mopt->localalloc_opt = -1;
+	mopt->cluster_stack[0] = '\0';
+	mopt->resv_level = OCFS2_DEFAULT_RESV_LEVEL;
+	mopt->dir_resv_level = -1;
+
+	if (!options) {
+		status = 1;
+		goto bail;
+	}
+
+	while ((p = strsep(&options, ",")) != NULL) {
+		int token, option;
+		substring_t args[MAX_OPT_ARGS];
+
+		if (!*p)
+			continue;
+
+		token = match_token(p, tokens, args);
+		switch (token) {
+		case Opt_hb_local:
+			mopt->mount_opt |= OCFS2_MOUNT_HB_LOCAL;
+			break;
+		case Opt_hb_none:
+			mopt->mount_opt |= OCFS2_MOUNT_HB_NONE;
+			break;
+		case Opt_hb_global:
+			mopt->mount_opt |= OCFS2_MOUNT_HB_GLOBAL;
+			break;
+		case Opt_barrier:
+			if (match_int(&args[0], &option)) {
+				status = 0;
+				goto bail;
+			}
+			if (option)
+				mopt->mount_opt |= OCFS2_MOUNT_BARRIER;
+			else
+				mopt->mount_opt &= ~OCFS2_MOUNT_BARRIER;
+			break;
+		case Opt_intr:
+			mopt->mount_opt &= ~OCFS2_MOUNT_NOINTR;
+			break;
+		case Opt_nointr:
+			mopt->mount_opt |= OCFS2_MOUNT_NOINTR;
+			break;
+		case Opt_err_panic:
+			mopt->mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
+			break;
+		case Opt_err_ro:
+			mopt->mount_opt &= ~OCFS2_MOUNT_ERRORS_PANIC;
+			break;
+		case Opt_data_ordered:
+			mopt->mount_opt &= ~OCFS2_MOUNT_DATA_WRITEBACK;
+			break;
+		case Opt_data_writeback:
+			mopt->mount_opt |= OCFS2_MOUNT_DATA_WRITEBACK;
+			break;
+		case Opt_user_xattr:
+			mopt->mount_opt &= ~OCFS2_MOUNT_NOUSERXATTR;
+			break;
+		case Opt_nouser_xattr:
+			mopt->mount_opt |= OCFS2_MOUNT_NOUSERXATTR;
+			break;
+		case Opt_atime_quantum:
+			if (match_int(&args[0], &option)) {
+				status = 0;
+				goto bail;
+			}
+			if (option >= 0)
+				mopt->atime_quantum = option;
+			break;
+		case Opt_slot:
+			option = 0;
+			if (match_int(&args[0], &option)) {
+				status = 0;
+				goto bail;
+			}
+			if (option)
+				mopt->slot = (s16)option;
+			break;
+		case Opt_commit:
+			option = 0;
+			if (match_int(&args[0], &option)) {
+				status = 0;
+				goto bail;
+			}
+			if (option < 0)
+				return 0;
+			if (option == 0)
+				option = JBD2_DEFAULT_MAX_COMMIT_AGE;
+			mopt->commit_interval = HZ * option;
+			break;
+		case Opt_localalloc:
+			option = 0;
+			if (match_int(&args[0], &option)) {
+				status = 0;
+				goto bail;
+			}
+			if (option >= 0)
+				mopt->localalloc_opt = option;
+			break;
+		case Opt_localflocks:
+			/*
+			 * Changing this during remount could race
+			 * flock() requests, or "unbalance" existing
+			 * ones (e.g., a lock is taken in one mode but
+			 * dropped in the other). If users care enough
+			 * to flip locking modes during remount, we
+			 * could add a "local" flag to individual
+			 * flock structures for proper tracking of
+			 * state.
+			 */
+			if (!is_remount)
+				mopt->mount_opt |= OCFS2_MOUNT_LOCALFLOCKS;
+			break;
+		case Opt_stack:
+			/* Check both that the option we were passed
+			 * is of the right length and that it is a proper
+			 * string of the right length.
+			 */
+			if (((args[0].to - args[0].from) !=
+			     OCFS2_STACK_LABEL_LEN) ||
+			    (strnlen(args[0].from,
+				     OCFS2_STACK_LABEL_LEN) !=
+			     OCFS2_STACK_LABEL_LEN)) {
+				mlog(ML_ERROR,
+				     "Invalid cluster_stack option\n");
+				status = 0;
+				goto bail;
+			}
+			memcpy(mopt->cluster_stack, args[0].from,
+			       OCFS2_STACK_LABEL_LEN);
+			mopt->cluster_stack[OCFS2_STACK_LABEL_LEN] = '\0';
+			/*
+			 * Open code the memcmp here as we don't have
+			 * an osb to pass to
+			 * ocfs2_userspace_stack().
+			 */
+			if (memcmp(mopt->cluster_stack,
+				   OCFS2_CLASSIC_CLUSTER_STACK,
+				   OCFS2_STACK_LABEL_LEN))
+				user_stack = 1;
+			break;
+		case Opt_inode64:
+			mopt->mount_opt |= OCFS2_MOUNT_INODE64;
+			break;
+		case Opt_usrquota:
+			mopt->mount_opt |= OCFS2_MOUNT_USRQUOTA;
+			break;
+		case Opt_grpquota:
+			mopt->mount_opt |= OCFS2_MOUNT_GRPQUOTA;
+			break;
+		case Opt_coherency_buffered:
+			mopt->mount_opt |= OCFS2_MOUNT_COHERENCY_BUFFERED;
+			break;
+		case Opt_coherency_full:
+			mopt->mount_opt &= ~OCFS2_MOUNT_COHERENCY_BUFFERED;
+			break;
+		case Opt_acl:
+			mopt->mount_opt |= OCFS2_MOUNT_POSIX_ACL;
+			mopt->mount_opt &= ~OCFS2_MOUNT_NO_POSIX_ACL;
+			break;
+		case Opt_noacl:
+			mopt->mount_opt |= OCFS2_MOUNT_NO_POSIX_ACL;
+			mopt->mount_opt &= ~OCFS2_MOUNT_POSIX_ACL;
+			break;
+		case Opt_resv_level:
+			if (is_remount)
+				break;
+			if (match_int(&args[0], &option)) {
+				status = 0;
+				goto bail;
+			}
+			if (option >= OCFS2_MIN_RESV_LEVEL &&
+			    option < OCFS2_MAX_RESV_LEVEL)
+				mopt->resv_level = option;
+			break;
+		case Opt_dir_resv_level:
+			if (is_remount)
+				break;
+			if (match_int(&args[0], &option)) {
+				status = 0;
+				goto bail;
+			}
+			if (option >= OCFS2_MIN_RESV_LEVEL &&
+			    option < OCFS2_MAX_RESV_LEVEL)
+				mopt->dir_resv_level = option;
+			break;
+		case Opt_journal_async_commit:
+			mopt->mount_opt |= OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT;
+			break;
+		default:
+			mlog(ML_ERROR,
+			     "Unrecognized mount option \"%s\" "
+			     "or missing value\n", p);
+			status = 0;
+			goto bail;
+		}
+	}
+
+	if (user_stack == 0) {
+		/* Ensure only one heartbeat mode */
+		tmp = mopt->mount_opt & (OCFS2_MOUNT_HB_LOCAL |
+					 OCFS2_MOUNT_HB_GLOBAL |
+					 OCFS2_MOUNT_HB_NONE);
+		if (hweight32(tmp) != 1) {
+			mlog(ML_ERROR, "Invalid heartbeat mount options\n");
+			status = 0;
+			goto bail;
+		}
+	}
+
+	status = 1;
+
+bail:
+	return status;
+}
+
+static int ocfs2_show_options(struct seq_file *s, struct dentry *root)
+{
+	struct ocfs2_super *osb = OCFS2_SB(root->d_sb);
+	unsigned long opts = osb->s_mount_opt;
+	unsigned int local_alloc_megs;
+
+	if (opts & (OCFS2_MOUNT_HB_LOCAL | OCFS2_MOUNT_HB_GLOBAL)) {
+		seq_printf(s, ",_netdev");
+		if (opts & OCFS2_MOUNT_HB_LOCAL)
+			seq_printf(s, ",%s", OCFS2_HB_LOCAL);
+		else
+			seq_printf(s, ",%s", OCFS2_HB_GLOBAL);
+	} else
+		seq_printf(s, ",%s", OCFS2_HB_NONE);
+
+	if (opts & OCFS2_MOUNT_NOINTR)
+		seq_printf(s, ",nointr");
+
+	if (opts & OCFS2_MOUNT_DATA_WRITEBACK)
+		seq_printf(s, ",data=writeback");
+	else
+		seq_printf(s, ",data=ordered");
+
+	if (opts & OCFS2_MOUNT_BARRIER)
+		seq_printf(s, ",barrier=1");
+
+	if (opts & OCFS2_MOUNT_ERRORS_PANIC)
+		seq_printf(s, ",errors=panic");
+	else
+		seq_printf(s, ",errors=remount-ro");
+
+	if (osb->preferred_slot != OCFS2_INVALID_SLOT)
+		seq_printf(s, ",preferred_slot=%d", osb->preferred_slot);
+
+	seq_printf(s, ",atime_quantum=%u", osb->s_atime_quantum);
+
+	if (osb->osb_commit_interval)
+		seq_printf(s, ",commit=%u",
+			   (unsigned) (osb->osb_commit_interval / HZ));
+
+	local_alloc_megs = osb->local_alloc_bits >> (20 - osb->s_clustersize_bits);
+	if (local_alloc_megs != ocfs2_la_default_mb(osb))
+		seq_printf(s, ",localalloc=%d", local_alloc_megs);
+
+	if (opts & OCFS2_MOUNT_LOCALFLOCKS)
+		seq_printf(s, ",localflocks,");
+
+	if (osb->osb_cluster_stack[0])
+		seq_printf(s, ",cluster_stack=%.*s", OCFS2_STACK_LABEL_LEN,
+			   osb->osb_cluster_stack);
+	if (opts & OCFS2_MOUNT_USRQUOTA)
+		seq_printf(s, ",usrquota");
+	if (opts & OCFS2_MOUNT_GRPQUOTA)
+		seq_printf(s, ",grpquota");
+
+	if (opts & OCFS2_MOUNT_COHERENCY_BUFFERED)
+		seq_printf(s, ",coherency=buffered");
+	else
+		seq_printf(s, ",coherency=full");
+
+	if (opts & OCFS2_MOUNT_NOUSERXATTR)
+		seq_printf(s, ",nouser_xattr");
+	else
+		seq_printf(s, ",user_xattr");
+
+	if (opts & OCFS2_MOUNT_INODE64)
+		seq_printf(s, ",inode64");
+
+	if (opts & OCFS2_MOUNT_POSIX_ACL)
+		seq_printf(s, ",acl");
+	else
+		seq_printf(s, ",noacl");
+
+	if (osb->osb_resv_level != OCFS2_DEFAULT_RESV_LEVEL)
+		seq_printf(s, ",resv_level=%d", osb->osb_resv_level);
+
+	if (osb->osb_dir_resv_level != osb->osb_resv_level)
+		seq_printf(s, ",dir_resv_level=%d", osb->osb_resv_level);
+
+	if (opts & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
+		seq_printf(s, ",journal_async_commit");
+
+	return 0;
+}
+
+static int __init ocfs2_init(void)
+{
+	int status;
+
+	status = init_ocfs2_uptodate_cache();
+	if (status < 0)
+		goto out1;
+
+	status = ocfs2_initialize_mem_caches();
+	if (status < 0)
+		goto out2;
+
+	ocfs2_wq = create_singlethread_workqueue("ocfs2_wq");
+	if (!ocfs2_wq) {
+		status = -ENOMEM;
+		goto out3;
+	}
+
+	ocfs2_debugfs_root = debugfs_create_dir("ocfs2", NULL);
+	if (!ocfs2_debugfs_root) {
+		status = -ENOMEM;
+		mlog(ML_ERROR, "Unable to create ocfs2 debugfs root.\n");
+		goto out4;
+	}
+
+	ocfs2_set_locking_protocol();
+
+	status = register_quota_format(&ocfs2_quota_format);
+	if (status < 0)
+		goto out4;
+	status = register_filesystem(&ocfs2_fs_type);
+	if (!status)
+		return 0;
+
+	unregister_quota_format(&ocfs2_quota_format);
+out4:
+	destroy_workqueue(ocfs2_wq);
+	debugfs_remove(ocfs2_debugfs_root);
+out3:
+	ocfs2_free_mem_caches();
+out2:
+	exit_ocfs2_uptodate_cache();
+out1:
+	mlog_errno(status);
+	return status;
+}
+
+static void __exit ocfs2_exit(void)
+{
+	if (ocfs2_wq) {
+		flush_workqueue(ocfs2_wq);
+		destroy_workqueue(ocfs2_wq);
+	}
+
+	unregister_quota_format(&ocfs2_quota_format);
+
+	debugfs_remove(ocfs2_debugfs_root);
+
+	ocfs2_free_mem_caches();
+
+	unregister_filesystem(&ocfs2_fs_type);
+
+	exit_ocfs2_uptodate_cache();
+}
+
+static void ocfs2_put_super(struct super_block *sb)
+{
+	trace_ocfs2_put_super(sb);
+
+	ocfs2_sync_blockdev(sb);
+	ocfs2_dismount_volume(sb, 0);
+}
+
+static int ocfs2_statfs(struct dentry *dentry, struct kstatfs *buf)
+{
+	struct ocfs2_super *osb;
+	u32 numbits, freebits;
+	int status;
+	struct ocfs2_dinode *bm_lock;
+	struct buffer_head *bh = NULL;
+	struct inode *inode = NULL;
+
+	trace_ocfs2_statfs(dentry->d_sb, buf);
+
+	osb = OCFS2_SB(dentry->d_sb);
+
+	inode = ocfs2_get_system_file_inode(osb,
+					    GLOBAL_BITMAP_SYSTEM_INODE,
+					    OCFS2_INVALID_SLOT);
+	if (!inode) {
+		mlog(ML_ERROR, "failed to get bitmap inode\n");
+		status = -EIO;
+		goto bail;
+	}
+
+	status = ocfs2_inode_lock(inode, &bh, 0);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	bm_lock = (struct ocfs2_dinode *) bh->b_data;
+
+	numbits = le32_to_cpu(bm_lock->id1.bitmap1.i_total);
+	freebits = numbits - le32_to_cpu(bm_lock->id1.bitmap1.i_used);
+
+	buf->f_type = OCFS2_SUPER_MAGIC;
+	buf->f_bsize = dentry->d_sb->s_blocksize;
+	buf->f_namelen = OCFS2_MAX_FILENAME_LEN;
+	buf->f_blocks = ((sector_t) numbits) *
+			(osb->s_clustersize >> osb->sb->s_blocksize_bits);
+	buf->f_bfree = ((sector_t) freebits) *
+		       (osb->s_clustersize >> osb->sb->s_blocksize_bits);
+	buf->f_bavail = buf->f_bfree;
+	buf->f_files = numbits;
+	buf->f_ffree = freebits;
+	buf->f_fsid.val[0] = crc32_le(0, osb->uuid_str, OCFS2_VOL_UUID_LEN)
+				& 0xFFFFFFFFUL;
+	buf->f_fsid.val[1] = crc32_le(0, osb->uuid_str + OCFS2_VOL_UUID_LEN,
+				OCFS2_VOL_UUID_LEN) & 0xFFFFFFFFUL;
+
+	brelse(bh);
+
+	ocfs2_inode_unlock(inode, 0);
+	status = 0;
+bail:
+	if (inode)
+		iput(inode);
+
+	if (status)
+		mlog_errno(status);
+
+	return status;
+}
+
+static void ocfs2_inode_init_once(void *data)
+{
+	struct ocfs2_inode_info *oi = data;
+
+	oi->ip_flags = 0;
+	oi->ip_open_count = 0;
+	spin_lock_init(&oi->ip_lock);
+	ocfs2_extent_map_init(&oi->vfs_inode);
+	INIT_LIST_HEAD(&oi->ip_io_markers);
+	oi->ip_dir_start_lookup = 0;
+	mutex_init(&oi->ip_unaligned_aio);
+	init_rwsem(&oi->ip_alloc_sem);
+	init_rwsem(&oi->ip_xattr_sem);
+	mutex_init(&oi->ip_io_mutex);
+
+	oi->ip_blkno = 0ULL;
+	oi->ip_clusters = 0;
+
+	ocfs2_resv_init_once(&oi->ip_la_data_resv);
+
+	ocfs2_lock_res_init_once(&oi->ip_rw_lockres);
+	ocfs2_lock_res_init_once(&oi->ip_inode_lockres);
+	ocfs2_lock_res_init_once(&oi->ip_open_lockres);
+
+	init_waitqueue_head(&oi->append_dio_wq);
+
+	ocfs2_metadata_cache_init(INODE_CACHE(&oi->vfs_inode),
+				  &ocfs2_inode_caching_ops);
+
+	inode_init_once(&oi->vfs_inode);
+}
+
+static int ocfs2_initialize_mem_caches(void)
+{
+	ocfs2_inode_cachep = kmem_cache_create("ocfs2_inode_cache",
+				       sizeof(struct ocfs2_inode_info),
+				       0,
+				       (SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
+						SLAB_MEM_SPREAD),
+				       ocfs2_inode_init_once);
+	ocfs2_dquot_cachep = kmem_cache_create("ocfs2_dquot_cache",
+					sizeof(struct ocfs2_dquot),
+					0,
+					(SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT|
+						SLAB_MEM_SPREAD),
+					NULL);
+	ocfs2_qf_chunk_cachep = kmem_cache_create("ocfs2_qf_chunk_cache",
+					sizeof(struct ocfs2_quota_chunk),
+					0,
+					(SLAB_RECLAIM_ACCOUNT|SLAB_MEM_SPREAD),
+					NULL);
+	if (!ocfs2_inode_cachep || !ocfs2_dquot_cachep ||
+	    !ocfs2_qf_chunk_cachep) {
+		if (ocfs2_inode_cachep)
+			kmem_cache_destroy(ocfs2_inode_cachep);
+		if (ocfs2_dquot_cachep)
+			kmem_cache_destroy(ocfs2_dquot_cachep);
+		if (ocfs2_qf_chunk_cachep)
+			kmem_cache_destroy(ocfs2_qf_chunk_cachep);
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+static void ocfs2_free_mem_caches(void)
+{
+	/*
+	 * Make sure all delayed rcu free inodes are flushed before we
+	 * destroy cache.
+	 */
+	rcu_barrier();
+	if (ocfs2_inode_cachep)
+		kmem_cache_destroy(ocfs2_inode_cachep);
+	ocfs2_inode_cachep = NULL;
+
+	if (ocfs2_dquot_cachep)
+		kmem_cache_destroy(ocfs2_dquot_cachep);
+	ocfs2_dquot_cachep = NULL;
+
+	if (ocfs2_qf_chunk_cachep)
+		kmem_cache_destroy(ocfs2_qf_chunk_cachep);
+	ocfs2_qf_chunk_cachep = NULL;
+}
+
+static int ocfs2_get_sector(struct super_block *sb,
+			    struct buffer_head **bh,
+			    int block,
+			    int sect_size)
+{
+	if (!sb_set_blocksize(sb, sect_size)) {
+		mlog(ML_ERROR, "unable to set blocksize\n");
+		return -EIO;
+	}
+
+	*bh = sb_getblk(sb, block);
+	if (!*bh) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+	lock_buffer(*bh);
+	if (!buffer_dirty(*bh))
+		clear_buffer_uptodate(*bh);
+	unlock_buffer(*bh);
+	ll_rw_block(READ, 1, bh);
+	wait_on_buffer(*bh);
+	if (!buffer_uptodate(*bh)) {
+		mlog_errno(-EIO);
+		brelse(*bh);
+		*bh = NULL;
+		return -EIO;
+	}
+
+	return 0;
+}
+
+static int ocfs2_mount_volume(struct super_block *sb)
+{
+	int status = 0;
+	int unlock_super = 0;
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+
+	if (ocfs2_is_hard_readonly(osb))
+		goto leave;
+
+	status = ocfs2_dlm_init(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	status = ocfs2_super_lock(osb, 1);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+	unlock_super = 1;
+
+	/* This will load up the node map and add ourselves to it. */
+	status = ocfs2_find_slot(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	/* load all node-local system inodes */
+	status = ocfs2_init_local_system_inodes(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	status = ocfs2_check_volume(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto leave;
+	}
+
+	status = ocfs2_truncate_log_init(osb);
+	if (status < 0)
+		mlog_errno(status);
+
+leave:
+	if (unlock_super)
+		ocfs2_super_unlock(osb, 1);
+
+	return status;
+}
+
+static void ocfs2_dismount_volume(struct super_block *sb, int mnt_err)
+{
+	int tmp, hangup_needed = 0;
+	struct ocfs2_super *osb = NULL;
+	char nodestr[12];
+
+	trace_ocfs2_dismount_volume(sb);
+
+	BUG_ON(!sb);
+	osb = OCFS2_SB(sb);
+	BUG_ON(!osb);
+
+	debugfs_remove(osb->osb_ctxt);
+
+	/* Orphan scan should be stopped as early as possible */
+	ocfs2_orphan_scan_stop(osb);
+
+	ocfs2_disable_quotas(osb);
+
+	/* All dquots should be freed by now */
+	WARN_ON(!llist_empty(&osb->dquot_drop_list));
+	/* Wait for worker to be done with the work structure in osb */
+	cancel_work_sync(&osb->dquot_drop_work);
+
+	ocfs2_shutdown_local_alloc(osb);
+
+	ocfs2_truncate_log_shutdown(osb);
+
+	/* This will disable recovery and flush any recovery work. */
+	ocfs2_recovery_exit(osb);
+
+	ocfs2_journal_shutdown(osb);
+
+	ocfs2_sync_blockdev(sb);
+
+	ocfs2_purge_refcount_trees(osb);
+
+	/* No cluster connection means we've failed during mount, so skip
+	 * all the steps which depended on that to complete. */
+	if (osb->cconn) {
+		tmp = ocfs2_super_lock(osb, 1);
+		if (tmp < 0) {
+			mlog_errno(tmp);
+			return;
+		}
+	}
+
+	if (osb->slot_num != OCFS2_INVALID_SLOT)
+		ocfs2_put_slot(osb);
+
+	if (osb->cconn)
+		ocfs2_super_unlock(osb, 1);
+
+	ocfs2_release_system_inodes(osb);
+
+	/*
+	 * If we're dismounting due to mount error, mount.ocfs2 will clean
+	 * up heartbeat.  If we're a local mount, there is no heartbeat.
+	 * If we failed before we got a uuid_str yet, we can't stop
+	 * heartbeat.  Otherwise, do it.
+	 */
+	if (!mnt_err && !ocfs2_mount_local(osb) && osb->uuid_str &&
+	    !ocfs2_is_hard_readonly(osb))
+		hangup_needed = 1;
+
+	if (osb->cconn)
+		ocfs2_dlm_shutdown(osb, hangup_needed);
+
+	ocfs2_blockcheck_stats_debugfs_remove(&osb->osb_ecc_stats);
+	debugfs_remove(osb->osb_debug_root);
+
+	if (hangup_needed)
+		ocfs2_cluster_hangup(osb->uuid_str, strlen(osb->uuid_str));
+
+	atomic_set(&osb->vol_state, VOLUME_DISMOUNTED);
+
+	if (ocfs2_mount_local(osb))
+		snprintf(nodestr, sizeof(nodestr), "local");
+	else
+		snprintf(nodestr, sizeof(nodestr), "%u", osb->node_num);
+
+	printk(KERN_INFO "ocfs2: Unmounting device (%s) on (node %s)\n",
+	       osb->dev_str, nodestr);
+
+	ocfs2_delete_osb(osb);
+	kfree(osb);
+	sb->s_dev = 0;
+	sb->s_fs_info = NULL;
+}
+
+static int ocfs2_setup_osb_uuid(struct ocfs2_super *osb, const unsigned char *uuid,
+				unsigned uuid_bytes)
+{
+	int i, ret;
+	char *ptr;
+
+	BUG_ON(uuid_bytes != OCFS2_VOL_UUID_LEN);
+
+	osb->uuid_str = kzalloc(OCFS2_VOL_UUID_LEN * 2 + 1, GFP_KERNEL);
+	if (osb->uuid_str == NULL)
+		return -ENOMEM;
+
+	for (i = 0, ptr = osb->uuid_str; i < OCFS2_VOL_UUID_LEN; i++) {
+		/* print with null */
+		ret = snprintf(ptr, 3, "%02X", uuid[i]);
+		if (ret != 2) /* drop super cleans up */
+			return -EINVAL;
+		/* then only advance past the last char */
+		ptr += 2;
+	}
+
+	return 0;
+}
+
+/* Make sure entire volume is addressable by our journal.  Requires
+   osb_clusters_at_boot to be valid and for the journal to have been
+   initialized by ocfs2_journal_init(). */
+static int ocfs2_journal_addressable(struct ocfs2_super *osb)
+{
+	int status = 0;
+	u64 max_block =
+		ocfs2_clusters_to_blocks(osb->sb,
+					 osb->osb_clusters_at_boot) - 1;
+
+	/* 32-bit block number is always OK. */
+	if (max_block <= (u32)~0ULL)
+		goto out;
+
+	/* Volume is "huge", so see if our journal is new enough to
+	   support it. */
+	if (!(OCFS2_HAS_COMPAT_FEATURE(osb->sb,
+				       OCFS2_FEATURE_COMPAT_JBD2_SB) &&
+	      jbd2_journal_check_used_features(osb->journal->j_journal, 0, 0,
+					       JBD2_FEATURE_INCOMPAT_64BIT))) {
+		mlog(ML_ERROR, "The journal cannot address the entire volume. "
+		     "Enable the 'block64' journal option with tunefs.ocfs2");
+		status = -EFBIG;
+		goto out;
+	}
+
+ out:
+	return status;
+}
+
+static int ocfs2_initialize_super(struct super_block *sb,
+				  struct buffer_head *bh,
+				  int sector_size,
+				  struct ocfs2_blockcheck_stats *stats)
+{
+	int status;
+	int i, cbits, bbits;
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
+	struct inode *inode = NULL;
+	struct ocfs2_journal *journal;
+	struct ocfs2_super *osb;
+	u64 total_blocks;
+
+	osb = kzalloc(sizeof(struct ocfs2_super), GFP_KERNEL);
+	if (!osb) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	sb->s_fs_info = osb;
+	sb->s_op = &ocfs2_sops;
+	sb->s_d_op = &ocfs2_dentry_ops;
+	sb->s_export_op = &ocfs2_export_ops;
+	sb->s_qcop = &dquot_quotactl_sysfile_ops;
+	sb->dq_op = &ocfs2_quota_operations;
+	sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
+	sb->s_xattr = ocfs2_xattr_handlers;
+	sb->s_time_gran = 1;
+	sb->s_flags |= MS_NOATIME;
+	/* this is needed to support O_LARGEFILE */
+	cbits = le32_to_cpu(di->id2.i_super.s_clustersize_bits);
+	bbits = le32_to_cpu(di->id2.i_super.s_blocksize_bits);
+	sb->s_maxbytes = ocfs2_max_file_offset(bbits, cbits);
+	memcpy(sb->s_uuid, di->id2.i_super.s_uuid,
+	       sizeof(di->id2.i_super.s_uuid));
+
+	osb->osb_dx_mask = (1 << (cbits - bbits)) - 1;
+
+	for (i = 0; i < 3; i++)
+		osb->osb_dx_seed[i] = le32_to_cpu(di->id2.i_super.s_dx_seed[i]);
+	osb->osb_dx_seed[3] = le32_to_cpu(di->id2.i_super.s_uuid_hash);
+
+	osb->sb = sb;
+	/* Save off for ocfs2_rw_direct */
+	osb->s_sectsize_bits = blksize_bits(sector_size);
+	BUG_ON(!osb->s_sectsize_bits);
+
+	spin_lock_init(&osb->dc_task_lock);
+	init_waitqueue_head(&osb->dc_event);
+	osb->dc_work_sequence = 0;
+	osb->dc_wake_sequence = 0;
+	INIT_LIST_HEAD(&osb->blocked_lock_list);
+	osb->blocked_lock_count = 0;
+	spin_lock_init(&osb->osb_lock);
+	spin_lock_init(&osb->osb_xattr_lock);
+	ocfs2_init_steal_slots(osb);
+
+	mutex_init(&osb->system_file_mutex);
+
+	atomic_set(&osb->alloc_stats.moves, 0);
+	atomic_set(&osb->alloc_stats.local_data, 0);
+	atomic_set(&osb->alloc_stats.bitmap_data, 0);
+	atomic_set(&osb->alloc_stats.bg_allocs, 0);
+	atomic_set(&osb->alloc_stats.bg_extends, 0);
+
+	/* Copy the blockcheck stats from the superblock probe */
+	osb->osb_ecc_stats = *stats;
+
+	ocfs2_init_node_maps(osb);
+
+	snprintf(osb->dev_str, sizeof(osb->dev_str), "%u,%u",
+		 MAJOR(osb->sb->s_dev), MINOR(osb->sb->s_dev));
+
+	osb->max_slots = le16_to_cpu(di->id2.i_super.s_max_slots);
+	if (osb->max_slots > OCFS2_MAX_SLOTS || osb->max_slots == 0) {
+		mlog(ML_ERROR, "Invalid number of node slots (%u)\n",
+		     osb->max_slots);
+		status = -EINVAL;
+		goto bail;
+	}
+
+	ocfs2_orphan_scan_init(osb);
+
+	status = ocfs2_recovery_init(osb);
+	if (status) {
+		mlog(ML_ERROR, "Unable to initialize recovery state\n");
+		mlog_errno(status);
+		goto bail;
+	}
+
+	init_waitqueue_head(&osb->checkpoint_event);
+
+	osb->s_atime_quantum = OCFS2_DEFAULT_ATIME_QUANTUM;
+
+	osb->slot_num = OCFS2_INVALID_SLOT;
+
+	osb->s_xattr_inline_size = le16_to_cpu(
+					di->id2.i_super.s_xattr_inline_size);
+
+	osb->local_alloc_state = OCFS2_LA_UNUSED;
+	osb->local_alloc_bh = NULL;
+	INIT_DELAYED_WORK(&osb->la_enable_wq, ocfs2_la_enable_worker);
+
+	init_waitqueue_head(&osb->osb_mount_event);
+
+	status = ocfs2_resmap_init(osb, &osb->osb_la_resmap);
+	if (status) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	osb->vol_label = kmalloc(OCFS2_MAX_VOL_LABEL_LEN, GFP_KERNEL);
+	if (!osb->vol_label) {
+		mlog(ML_ERROR, "unable to alloc vol label\n");
+		status = -ENOMEM;
+		goto bail;
+	}
+
+	osb->slot_recovery_generations =
+		kcalloc(osb->max_slots, sizeof(*osb->slot_recovery_generations),
+			GFP_KERNEL);
+	if (!osb->slot_recovery_generations) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	init_waitqueue_head(&osb->osb_wipe_event);
+	osb->osb_orphan_wipes = kcalloc(osb->max_slots,
+					sizeof(*osb->osb_orphan_wipes),
+					GFP_KERNEL);
+	if (!osb->osb_orphan_wipes) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	osb->osb_rf_lock_tree = RB_ROOT;
+
+	osb->s_feature_compat =
+		le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_compat);
+	osb->s_feature_ro_compat =
+		le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_ro_compat);
+	osb->s_feature_incompat =
+		le32_to_cpu(OCFS2_RAW_SB(di)->s_feature_incompat);
+
+	if ((i = OCFS2_HAS_INCOMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_INCOMPAT_SUPP))) {
+		mlog(ML_ERROR, "couldn't mount because of unsupported "
+		     "optional features (%x).\n", i);
+		status = -EINVAL;
+		goto bail;
+	}
+	if (!(osb->sb->s_flags & MS_RDONLY) &&
+	    (i = OCFS2_HAS_RO_COMPAT_FEATURE(osb->sb, ~OCFS2_FEATURE_RO_COMPAT_SUPP))) {
+		mlog(ML_ERROR, "couldn't mount RDWR because of "
+		     "unsupported optional features (%x).\n", i);
+		status = -EINVAL;
+		goto bail;
+	}
+
+	if (ocfs2_clusterinfo_valid(osb)) {
+		osb->osb_stackflags =
+			OCFS2_RAW_SB(di)->s_cluster_info.ci_stackflags;
+		strlcpy(osb->osb_cluster_stack,
+		       OCFS2_RAW_SB(di)->s_cluster_info.ci_stack,
+		       OCFS2_STACK_LABEL_LEN + 1);
+		if (strlen(osb->osb_cluster_stack) != OCFS2_STACK_LABEL_LEN) {
+			mlog(ML_ERROR,
+			     "couldn't mount because of an invalid "
+			     "cluster stack label (%s) \n",
+			     osb->osb_cluster_stack);
+			status = -EINVAL;
+			goto bail;
+		}
+		strlcpy(osb->osb_cluster_name,
+			OCFS2_RAW_SB(di)->s_cluster_info.ci_cluster,
+			OCFS2_CLUSTER_NAME_LEN + 1);
+	} else {
+		/* The empty string is identical with classic tools that
+		 * don't know about s_cluster_info. */
+		osb->osb_cluster_stack[0] = '\0';
+	}
+
+	get_random_bytes(&osb->s_next_generation, sizeof(u32));
+
+	/* FIXME
+	 * This should be done in ocfs2_journal_init(), but unknown
+	 * ordering issues will cause the filesystem to crash.
+	 * If anyone wants to figure out what part of the code
+	 * refers to osb->journal before ocfs2_journal_init() is run,
+	 * be my guest.
+	 */
+	/* initialize our journal structure */
+
+	journal = kzalloc(sizeof(struct ocfs2_journal), GFP_KERNEL);
+	if (!journal) {
+		mlog(ML_ERROR, "unable to alloc journal\n");
+		status = -ENOMEM;
+		goto bail;
+	}
+	osb->journal = journal;
+	journal->j_osb = osb;
+
+	atomic_set(&journal->j_num_trans, 0);
+	init_rwsem(&journal->j_trans_barrier);
+	init_waitqueue_head(&journal->j_checkpointed);
+	spin_lock_init(&journal->j_lock);
+	journal->j_trans_id = (unsigned long) 1;
+	INIT_LIST_HEAD(&journal->j_la_cleanups);
+	INIT_WORK(&journal->j_recovery_work, ocfs2_complete_recovery);
+	journal->j_state = OCFS2_JOURNAL_FREE;
+
+	INIT_WORK(&osb->dquot_drop_work, ocfs2_drop_dquot_refs);
+	init_llist_head(&osb->dquot_drop_list);
+
+	/* get some pseudo constants for clustersize bits */
+	osb->s_clustersize_bits =
+		le32_to_cpu(di->id2.i_super.s_clustersize_bits);
+	osb->s_clustersize = 1 << osb->s_clustersize_bits;
+
+	if (osb->s_clustersize < OCFS2_MIN_CLUSTERSIZE ||
+	    osb->s_clustersize > OCFS2_MAX_CLUSTERSIZE) {
+		mlog(ML_ERROR, "Volume has invalid cluster size (%d)\n",
+		     osb->s_clustersize);
+		status = -EINVAL;
+		goto bail;
+	}
+
+	total_blocks = ocfs2_clusters_to_blocks(osb->sb,
+						le32_to_cpu(di->i_clusters));
+
+	status = generic_check_addressable(osb->sb->s_blocksize_bits,
+					   total_blocks);
+	if (status) {
+		mlog(ML_ERROR, "Volume too large "
+		     "to mount safely on this system");
+		status = -EFBIG;
+		goto bail;
+	}
+
+	if (ocfs2_setup_osb_uuid(osb, di->id2.i_super.s_uuid,
+				 sizeof(di->id2.i_super.s_uuid))) {
+		mlog(ML_ERROR, "Out of memory trying to setup our uuid.\n");
+		status = -ENOMEM;
+		goto bail;
+	}
+
+	strlcpy(osb->vol_label, di->id2.i_super.s_label,
+		OCFS2_MAX_VOL_LABEL_LEN);
+	osb->root_blkno = le64_to_cpu(di->id2.i_super.s_root_blkno);
+	osb->system_dir_blkno = le64_to_cpu(di->id2.i_super.s_system_dir_blkno);
+	osb->first_cluster_group_blkno =
+		le64_to_cpu(di->id2.i_super.s_first_cluster_group);
+	osb->fs_generation = le32_to_cpu(di->i_fs_generation);
+	osb->uuid_hash = le32_to_cpu(di->id2.i_super.s_uuid_hash);
+	trace_ocfs2_initialize_super(osb->vol_label, osb->uuid_str,
+				     (unsigned long long)osb->root_blkno,
+				     (unsigned long long)osb->system_dir_blkno,
+				     osb->s_clustersize_bits);
+
+	osb->osb_dlm_debug = ocfs2_new_dlm_debug();
+	if (!osb->osb_dlm_debug) {
+		status = -ENOMEM;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	atomic_set(&osb->vol_state, VOLUME_INIT);
+
+	/* load root, system_dir, and all global system inodes */
+	status = ocfs2_init_global_system_inodes(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+
+	/*
+	 * global bitmap
+	 */
+	inode = ocfs2_get_system_file_inode(osb, GLOBAL_BITMAP_SYSTEM_INODE,
+					    OCFS2_INVALID_SLOT);
+	if (!inode) {
+		status = -EINVAL;
+		mlog_errno(status);
+		goto bail;
+	}
+
+	osb->bitmap_blkno = OCFS2_I(inode)->ip_blkno;
+	osb->osb_clusters_at_boot = OCFS2_I(inode)->ip_clusters;
+	iput(inode);
+
+	osb->bitmap_cpg = ocfs2_group_bitmap_size(sb, 0,
+				 osb->s_feature_incompat) * 8;
+
+	status = ocfs2_init_slot_info(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto bail;
+	}
+	cleancache_init_shared_fs(sb);
+
+bail:
+	return status;
+}
+
+/*
+ * will return: -EAGAIN if it is ok to keep searching for superblocks
+ *              -EINVAL if there is a bad superblock
+ *              0 on success
+ */
+static int ocfs2_verify_volume(struct ocfs2_dinode *di,
+			       struct buffer_head *bh,
+			       u32 blksz,
+			       struct ocfs2_blockcheck_stats *stats)
+{
+	int status = -EAGAIN;
+
+	if (memcmp(di->i_signature, OCFS2_SUPER_BLOCK_SIGNATURE,
+		   strlen(OCFS2_SUPER_BLOCK_SIGNATURE)) == 0) {
+		/* We have to do a raw check of the feature here */
+		if (le32_to_cpu(di->id2.i_super.s_feature_incompat) &
+		    OCFS2_FEATURE_INCOMPAT_META_ECC) {
+			status = ocfs2_block_check_validate(bh->b_data,
+							    bh->b_size,
+							    &di->i_check,
+							    stats);
+			if (status)
+				goto out;
+		}
+		status = -EINVAL;
+		if ((1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits)) != blksz) {
+			mlog(ML_ERROR, "found superblock with incorrect block "
+			     "size: found %u, should be %u\n",
+			     1 << le32_to_cpu(di->id2.i_super.s_blocksize_bits),
+			       blksz);
+		} else if (le16_to_cpu(di->id2.i_super.s_major_rev_level) !=
+			   OCFS2_MAJOR_REV_LEVEL ||
+			   le16_to_cpu(di->id2.i_super.s_minor_rev_level) !=
+			   OCFS2_MINOR_REV_LEVEL) {
+			mlog(ML_ERROR, "found superblock with bad version: "
+			     "found %u.%u, should be %u.%u\n",
+			     le16_to_cpu(di->id2.i_super.s_major_rev_level),
+			     le16_to_cpu(di->id2.i_super.s_minor_rev_level),
+			     OCFS2_MAJOR_REV_LEVEL,
+			     OCFS2_MINOR_REV_LEVEL);
+		} else if (bh->b_blocknr != le64_to_cpu(di->i_blkno)) {
+			mlog(ML_ERROR, "bad block number on superblock: "
+			     "found %llu, should be %llu\n",
+			     (unsigned long long)le64_to_cpu(di->i_blkno),
+			     (unsigned long long)bh->b_blocknr);
+		} else if (le32_to_cpu(di->id2.i_super.s_clustersize_bits) < 12 ||
+			    le32_to_cpu(di->id2.i_super.s_clustersize_bits) > 20) {
+			mlog(ML_ERROR, "bad cluster size found: %u\n",
+			     1 << le32_to_cpu(di->id2.i_super.s_clustersize_bits));
+		} else if (!le64_to_cpu(di->id2.i_super.s_root_blkno)) {
+			mlog(ML_ERROR, "bad root_blkno: 0\n");
+		} else if (!le64_to_cpu(di->id2.i_super.s_system_dir_blkno)) {
+			mlog(ML_ERROR, "bad system_dir_blkno: 0\n");
+		} else if (le16_to_cpu(di->id2.i_super.s_max_slots) > OCFS2_MAX_SLOTS) {
+			mlog(ML_ERROR,
+			     "Superblock slots found greater than file system "
+			     "maximum: found %u, max %u\n",
+			     le16_to_cpu(di->id2.i_super.s_max_slots),
+			     OCFS2_MAX_SLOTS);
+		} else {
+			/* found it! */
+			status = 0;
+		}
+	}
+
+out:
+	if (status && status != -EAGAIN)
+		mlog_errno(status);
+	return status;
+}
+
+static int ocfs2_check_volume(struct ocfs2_super *osb)
+{
+	int status;
+	int dirty;
+	int local;
+	struct ocfs2_dinode *local_alloc = NULL; /* only used if we
+						  * recover
+						  * ourselves. */
+
+	/* Init our journal object. */
+	status = ocfs2_journal_init(osb->journal, &dirty);
+	if (status < 0) {
+		mlog(ML_ERROR, "Could not initialize journal!\n");
+		goto finally;
+	}
+
+	/* Now that journal has been initialized, check to make sure
+	   entire volume is addressable. */
+	status = ocfs2_journal_addressable(osb);
+	if (status)
+		goto finally;
+
+	/* If the journal was unmounted cleanly then we don't want to
+	 * recover anything. Otherwise, journal_load will do that
+	 * dirty work for us :) */
+	if (!dirty) {
+		status = ocfs2_journal_wipe(osb->journal, 0);
+		if (status < 0) {
+			mlog_errno(status);
+			goto finally;
+		}
+	} else {
+		printk(KERN_NOTICE "ocfs2: File system on device (%s) was not "
+		       "unmounted cleanly, recovering it.\n", osb->dev_str);
+	}
+
+	local = ocfs2_mount_local(osb);
+
+	/* will play back anything left in the journal. */
+	status = ocfs2_journal_load(osb->journal, local, dirty);
+	if (status < 0) {
+		mlog(ML_ERROR, "ocfs2 journal load failed! %d\n", status);
+		goto finally;
+	}
+
+	if (osb->s_mount_opt & OCFS2_MOUNT_JOURNAL_ASYNC_COMMIT)
+		jbd2_journal_set_features(osb->journal->j_journal,
+				JBD2_FEATURE_COMPAT_CHECKSUM, 0,
+				JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
+	else
+		jbd2_journal_clear_features(osb->journal->j_journal,
+				JBD2_FEATURE_COMPAT_CHECKSUM, 0,
+				JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
+
+	if (dirty) {
+		/* recover my local alloc if we didn't unmount cleanly. */
+		status = ocfs2_begin_local_alloc_recovery(osb,
+							  osb->slot_num,
+							  &local_alloc);
+		if (status < 0) {
+			mlog_errno(status);
+			goto finally;
+		}
+		/* we complete the recovery process after we've marked
+		 * ourselves as mounted. */
+	}
+
+	status = ocfs2_load_local_alloc(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto finally;
+	}
+
+	if (dirty) {
+		/* Recovery will be completed after we've mounted the
+		 * rest of the volume. */
+		osb->dirty = 1;
+		osb->local_alloc_copy = local_alloc;
+		local_alloc = NULL;
+	}
+
+	/* go through each journal, trylock it and if you get the
+	 * lock, and it's marked as dirty, set the bit in the recover
+	 * map and launch a recovery thread for it. */
+	status = ocfs2_mark_dead_nodes(osb);
+	if (status < 0) {
+		mlog_errno(status);
+		goto finally;
+	}
+
+	status = ocfs2_compute_replay_slots(osb);
+	if (status < 0)
+		mlog_errno(status);
+
+finally:
+	kfree(local_alloc);
+
+	if (status)
+		mlog_errno(status);
+	return status;
+}
+
+/*
+ * The routine gets called from dismount or close whenever a dismount on
+ * volume is requested and the osb open count becomes 1.
+ * It will remove the osb from the global list and also free up all the
+ * initialized resources and fileobject.
+ */
+static void ocfs2_delete_osb(struct ocfs2_super *osb)
+{
+	/* This function assumes that the caller has the main osb resource */
+
+	ocfs2_free_slot_info(osb);
+
+	kfree(osb->osb_orphan_wipes);
+	kfree(osb->slot_recovery_generations);
+	/* FIXME
+	 * This belongs in journal shutdown, but because we have to
+	 * allocate osb->journal at the start of ocfs2_initialize_osb(),
+	 * we free it here.
+	 */
+	kfree(osb->journal);
+	kfree(osb->local_alloc_copy);
+	kfree(osb->uuid_str);
+	kfree(osb->vol_label);
+	ocfs2_put_dlm_debug(osb->osb_dlm_debug);
+	memset(osb, 0, sizeof(struct ocfs2_super));
+}
+
+/* Put OCFS2 into a readonly state, or (if the user specifies it),
+ * panic(). We do not support continue-on-error operation. */
+static void ocfs2_handle_error(struct super_block *sb)
+{
+	struct ocfs2_super *osb = OCFS2_SB(sb);
+
+	if (osb->s_mount_opt & OCFS2_MOUNT_ERRORS_PANIC)
+		panic("OCFS2: (device %s): panic forced after error\n",
+		      sb->s_id);
+
+	ocfs2_set_osb_flag(osb, OCFS2_OSB_ERROR_FS);
+
+	if (sb->s_flags & MS_RDONLY &&
+	    (ocfs2_is_soft_readonly(osb) ||
+	     ocfs2_is_hard_readonly(osb)))
+		return;
+
+	printk(KERN_CRIT "File system is now read-only due to the potential "
+	       "of on-disk corruption. Please run fsck.ocfs2 once the file "
+	       "system is unmounted.\n");
+	sb->s_flags |= MS_RDONLY;
+	ocfs2_set_ro_flag(osb, 0);
+}
+
+void __ocfs2_error(struct super_block *sb, const char *function,
+		  const char *fmt, ...)
+{
+	struct va_format vaf;
+	va_list args;
+
+	va_start(args, fmt);
+	vaf.fmt = fmt;
+	vaf.va = &args;
+
+	/* Not using mlog here because we want to show the actual
+	 * function the error came from. */
+	printk(KERN_CRIT "OCFS2: ERROR (device %s): %s: %pV\n",
+	       sb->s_id, function, &vaf);
+
+	va_end(args);
+
+	ocfs2_handle_error(sb);
+}
+
+/* Handle critical errors. This is intentionally more drastic than
+ * ocfs2_handle_error, so we only use for things like journal errors,
+ * etc. */
+void __ocfs2_abort(struct super_block *sb, const char *function,
+		   const char *fmt, ...)
+{
+	struct va_format vaf;
+	va_list args;
+
+	va_start(args, fmt);
+
+	vaf.fmt = fmt;
+	vaf.va = &args;
+
+	printk(KERN_CRIT "OCFS2: abort (device %s): %s: %pV\n",
+	       sb->s_id, function, &vaf);
+
+	va_end(args);
+
+	/* We don't have the cluster support yet to go straight to
+	 * hard readonly in here. Until then, we want to keep
+	 * ocfs2_abort() so that we can at least mark critical
+	 * errors.
+	 *
+	 * TODO: This should abort the journal and alert other nodes
+	 * that our slot needs recovery. */
+
+	/* Force a panic(). This stinks, but it's better than letting
+	 * things continue without having a proper hard readonly
+	 * here. */
+	if (!ocfs2_mount_local(OCFS2_SB(sb)))
+		OCFS2_SB(sb)->s_mount_opt |= OCFS2_MOUNT_ERRORS_PANIC;
+	ocfs2_handle_error(sb);
+}
+
+/*
+ * Void signal blockers, because in-kernel sigprocmask() only fails
+ * when SIG_* is wrong.
+ */
+void ocfs2_block_signals(sigset_t *oldset)
+{
+	int rc;
+	sigset_t blocked;
+
+	sigfillset(&blocked);
+	rc = sigprocmask(SIG_BLOCK, &blocked, oldset);
+	BUG_ON(rc);
+}
+
+void ocfs2_unblock_signals(sigset_t *oldset)
+{
+	int rc = sigprocmask(SIG_SETMASK, oldset, NULL);
+	BUG_ON(rc);
+}
+
+module_init(ocfs2_init);
+module_exit(ocfs2_exit);
diff --git a/fs/ocfs2/super.h b/fs/ocfs2/super.h
new file mode 100644
index 000000000..74ff74cf7
--- /dev/null
+++ b/fs/ocfs2/super.h
@@ -0,0 +1,53 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * super.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_SUPER_H
+#define OCFS2_SUPER_H
+
+extern struct workqueue_struct *ocfs2_wq;
+
+int ocfs2_publish_get_mount_state(struct ocfs2_super *osb,
+				  int node_num);
+
+__printf(3, 4)
+void __ocfs2_error(struct super_block *sb, const char *function,
+		   const char *fmt, ...);
+
+#define ocfs2_error(sb, fmt, args...) __ocfs2_error(sb, __PRETTY_FUNCTION__, fmt, ##args)
+
+__printf(3, 4)
+void __ocfs2_abort(struct super_block *sb, const char *function,
+		   const char *fmt, ...);
+
+#define ocfs2_abort(sb, fmt, args...) __ocfs2_abort(sb, __PRETTY_FUNCTION__, fmt, ##args)
+
+/*
+ * Void signal blockers, because in-kernel sigprocmask() only fails
+ * when SIG_* is wrong.
+ */
+void ocfs2_block_signals(sigset_t *oldset);
+void ocfs2_unblock_signals(sigset_t *oldset);
+
+#endif /* OCFS2_SUPER_H */
diff --git a/fs/ocfs2/symlink.c b/fs/ocfs2/symlink.c
new file mode 100644
index 000000000..66edce7ec
--- /dev/null
+++ b/fs/ocfs2/symlink.c
@@ -0,0 +1,100 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ *  linux/cluster/ssi/cfs/symlink.c
+ *
+ *	This program is free software; you can redistribute it and/or
+ *	modify it under the terms of the GNU General Public License as
+ *	published by the Free Software Foundation; either version 2 of
+ *	the License, or (at your option) any later version.
+ *
+ *	This program is distributed in the hope that it will be useful,
+ *	but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *	MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE
+ *	or NON INFRINGEMENT.  See the GNU General Public License for more
+ *	details.
+ *
+ * 	You should have received a copy of the GNU General Public License
+ * 	along with this program; if not, write to the Free Software
+ * 	Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ *
+ *	Questions/Comments/Bugfixes to ssic-linux-devel@lists.sourceforge.net
+ *
+ *  Copyright (C) 1992  Rick Sladkey
+ *
+ *  Optimization changes Copyright (C) 1994 Florian La Roche
+ *
+ *  Jun 7 1999, cache symlink lookups in the page cache.  -DaveM
+ *
+ *  Portions Copyright (C) 2001 Compaq Computer Corporation
+ *
+ *  ocfs2 symlink handling code.
+ *
+ *  Copyright (C) 2004, 2005 Oracle.
+ *
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/pagemap.h>
+#include <linux/namei.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "file.h"
+#include "inode.h"
+#include "journal.h"
+#include "symlink.h"
+#include "xattr.h"
+
+#include "buffer_head_io.h"
+
+
+static int ocfs2_fast_symlink_readpage(struct file *unused, struct page *page)
+{
+	struct inode *inode = page->mapping->host;
+	struct buffer_head *bh = NULL;
+	int status = ocfs2_read_inode_block(inode, &bh);
+	struct ocfs2_dinode *fe;
+	const char *link;
+	void *kaddr;
+	size_t len;
+
+	if (status < 0) {
+		mlog_errno(status);
+		return status;
+	}
+
+	fe = (struct ocfs2_dinode *) bh->b_data;
+	link = (char *) fe->id2.i_symlink;
+	/* will be less than a page size */
+	len = strnlen(link, ocfs2_fast_symlink_chars(inode->i_sb));
+	kaddr = kmap_atomic(page);
+	memcpy(kaddr, link, len + 1);
+	kunmap_atomic(kaddr);
+	SetPageUptodate(page);
+	unlock_page(page);
+	brelse(bh);
+	return 0;
+}
+
+const struct address_space_operations ocfs2_fast_symlink_aops = {
+	.readpage		= ocfs2_fast_symlink_readpage,
+};
+
+const struct inode_operations ocfs2_symlink_inode_operations = {
+	.readlink	= generic_readlink,
+	.follow_link	= page_follow_link_light,
+	.put_link	= page_put_link,
+	.getattr	= ocfs2_getattr,
+	.setattr	= ocfs2_setattr,
+	.setxattr	= generic_setxattr,
+	.getxattr	= generic_getxattr,
+	.listxattr	= ocfs2_listxattr,
+	.removexattr	= generic_removexattr,
+	.fiemap		= ocfs2_fiemap,
+};
diff --git a/fs/ocfs2/symlink.h b/fs/ocfs2/symlink.h
new file mode 100644
index 000000000..71ee4245e
--- /dev/null
+++ b/fs/ocfs2/symlink.h
@@ -0,0 +1,42 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * symlink.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_SYMLINK_H
+#define OCFS2_SYMLINK_H
+
+extern const struct inode_operations ocfs2_symlink_inode_operations;
+extern const struct address_space_operations ocfs2_fast_symlink_aops;
+
+/*
+ * Test whether an inode is a fast symlink.
+ */
+static inline int ocfs2_inode_is_fast_symlink(struct inode *inode)
+{
+	return (S_ISLNK(inode->i_mode) &&
+		inode->i_blocks == 0);
+}
+
+
+#endif /* OCFS2_SYMLINK_H */
diff --git a/fs/ocfs2/sysfile.c b/fs/ocfs2/sysfile.c
new file mode 100644
index 000000000..af155c183
--- /dev/null
+++ b/fs/ocfs2/sysfile.c
@@ -0,0 +1,182 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * sysfile.c
+ *
+ * Initialize, read, write, etc. system files.
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/highmem.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "alloc.h"
+#include "dir.h"
+#include "inode.h"
+#include "journal.h"
+#include "sysfile.h"
+
+#include "buffer_head_io.h"
+
+static struct inode * _ocfs2_get_system_file_inode(struct ocfs2_super *osb,
+						   int type,
+						   u32 slot);
+
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+static struct lock_class_key ocfs2_sysfile_cluster_lock_key[NUM_SYSTEM_INODES];
+#endif
+
+static inline int is_global_system_inode(int type)
+{
+	return type >= OCFS2_FIRST_ONLINE_SYSTEM_INODE &&
+		type <= OCFS2_LAST_GLOBAL_SYSTEM_INODE;
+}
+
+static struct inode **get_local_system_inode(struct ocfs2_super *osb,
+					     int type,
+					     u32 slot)
+{
+	int index;
+	struct inode **local_system_inodes, **free = NULL;
+
+	BUG_ON(slot == OCFS2_INVALID_SLOT);
+	BUG_ON(type < OCFS2_FIRST_LOCAL_SYSTEM_INODE ||
+	       type > OCFS2_LAST_LOCAL_SYSTEM_INODE);
+
+	spin_lock(&osb->osb_lock);
+	local_system_inodes = osb->local_system_inodes;
+	spin_unlock(&osb->osb_lock);
+
+	if (unlikely(!local_system_inodes)) {
+		local_system_inodes = kzalloc(sizeof(struct inode *) *
+					      NUM_LOCAL_SYSTEM_INODES *
+					      osb->max_slots,
+					      GFP_NOFS);
+		if (!local_system_inodes) {
+			mlog_errno(-ENOMEM);
+			/*
+			 * return NULL here so that ocfs2_get_sytem_file_inodes
+			 * will try to create an inode and use it. We will try
+			 * to initialize local_system_inodes next time.
+			 */
+			return NULL;
+		}
+
+		spin_lock(&osb->osb_lock);
+		if (osb->local_system_inodes) {
+			/* Someone has initialized it for us. */
+			free = local_system_inodes;
+			local_system_inodes = osb->local_system_inodes;
+		} else
+			osb->local_system_inodes = local_system_inodes;
+		spin_unlock(&osb->osb_lock);
+		kfree(free);
+	}
+
+	index = (slot * NUM_LOCAL_SYSTEM_INODES) +
+		(type - OCFS2_FIRST_LOCAL_SYSTEM_INODE);
+
+	return &local_system_inodes[index];
+}
+
+struct inode *ocfs2_get_system_file_inode(struct ocfs2_super *osb,
+					  int type,
+					  u32 slot)
+{
+	struct inode *inode = NULL;
+	struct inode **arr = NULL;
+
+	/* avoid the lookup if cached in local system file array */
+	if (is_global_system_inode(type)) {
+		arr = &(osb->global_system_inodes[type]);
+	} else
+		arr = get_local_system_inode(osb, type, slot);
+
+	mutex_lock(&osb->system_file_mutex);
+	if (arr && ((inode = *arr) != NULL)) {
+		/* get a ref in addition to the array ref */
+		inode = igrab(inode);
+		mutex_unlock(&osb->system_file_mutex);
+		BUG_ON(!inode);
+
+		return inode;
+	}
+
+	/* this gets one ref thru iget */
+	inode = _ocfs2_get_system_file_inode(osb, type, slot);
+
+	/* add one more if putting into array for first time */
+	if (arr && inode) {
+		*arr = igrab(inode);
+		BUG_ON(!*arr);
+	}
+	mutex_unlock(&osb->system_file_mutex);
+	return inode;
+}
+
+static struct inode * _ocfs2_get_system_file_inode(struct ocfs2_super *osb,
+						   int type,
+						   u32 slot)
+{
+	char namebuf[40];
+	struct inode *inode = NULL;
+	u64 blkno;
+	int status = 0;
+
+	ocfs2_sprintf_system_inode_name(namebuf,
+					sizeof(namebuf),
+					type, slot);
+
+	status = ocfs2_lookup_ino_from_name(osb->sys_root_inode, namebuf,
+					    strlen(namebuf), &blkno);
+	if (status < 0) {
+		goto bail;
+	}
+
+	inode = ocfs2_iget(osb, blkno, OCFS2_FI_FLAG_SYSFILE, type);
+	if (IS_ERR(inode)) {
+		mlog_errno(PTR_ERR(inode));
+		inode = NULL;
+		goto bail;
+	}
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+	if (type == LOCAL_USER_QUOTA_SYSTEM_INODE ||
+	    type == LOCAL_GROUP_QUOTA_SYSTEM_INODE ||
+	    type == JOURNAL_SYSTEM_INODE) {
+		/* Ignore inode lock on these inodes as the lock does not
+		 * really belong to any process and lockdep cannot handle
+		 * that */
+		OCFS2_I(inode)->ip_inode_lockres.l_lockdep_map.key = NULL;
+	} else {
+		lockdep_init_map(&OCFS2_I(inode)->ip_inode_lockres.
+								l_lockdep_map,
+				 ocfs2_system_inodes[type].si_name,
+				 &ocfs2_sysfile_cluster_lock_key[type], 0);
+	}
+#endif
+bail:
+
+	return inode;
+}
+
diff --git a/fs/ocfs2/sysfile.h b/fs/ocfs2/sysfile.h
new file mode 100644
index 000000000..cc9ea661f
--- /dev/null
+++ b/fs/ocfs2/sysfile.h
@@ -0,0 +1,33 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * sysfile.h
+ *
+ * Function prototypes
+ *
+ * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_SYSFILE_H
+#define OCFS2_SYSFILE_H
+
+struct inode * ocfs2_get_system_file_inode(struct ocfs2_super *osb,
+					   int type,
+					   u32 slot);
+
+#endif /* OCFS2_SYSFILE_H */
diff --git a/fs/ocfs2/uptodate.c b/fs/ocfs2/uptodate.c
new file mode 100644
index 000000000..82e17b076
--- /dev/null
+++ b/fs/ocfs2/uptodate.c
@@ -0,0 +1,638 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * uptodate.c
+ *
+ * Tracking the up-to-date-ness of a local buffer_head with respect to
+ * the cluster.
+ *
+ * Copyright (C) 2002, 2004, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ *
+ * Standard buffer head caching flags (uptodate, etc) are insufficient
+ * in a clustered environment - a buffer may be marked up to date on
+ * our local node but could have been modified by another cluster
+ * member. As a result an additional (and performant) caching scheme
+ * is required. A further requirement is that we consume as little
+ * memory as possible - we never pin buffer_head structures in order
+ * to cache them.
+ *
+ * We track the existence of up to date buffers on the inodes which
+ * are associated with them. Because we don't want to pin
+ * buffer_heads, this is only a (strong) hint and several other checks
+ * are made in the I/O path to ensure that we don't use a stale or
+ * invalid buffer without going to disk:
+ *	- buffer_jbd is used liberally - if a bh is in the journal on
+ *	  this node then it *must* be up to date.
+ *	- the standard buffer_uptodate() macro is used to detect buffers
+ *	  which may be invalid (even if we have an up to date tracking
+ * 	  item for them)
+ *
+ * For a full understanding of how this code works together, one
+ * should read the callers in dlmglue.c, the I/O functions in
+ * buffer_head_io.c and ocfs2_journal_access in journal.c
+ */
+
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/buffer_head.h>
+#include <linux/rbtree.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+
+#include "inode.h"
+#include "uptodate.h"
+#include "ocfs2_trace.h"
+
+struct ocfs2_meta_cache_item {
+	struct rb_node	c_node;
+	sector_t	c_block;
+};
+
+static struct kmem_cache *ocfs2_uptodate_cachep;
+
+u64 ocfs2_metadata_cache_owner(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	return ci->ci_ops->co_owner(ci);
+}
+
+struct super_block *ocfs2_metadata_cache_get_super(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	return ci->ci_ops->co_get_super(ci);
+}
+
+static void ocfs2_metadata_cache_lock(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ci->ci_ops->co_cache_lock(ci);
+}
+
+static void ocfs2_metadata_cache_unlock(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ci->ci_ops->co_cache_unlock(ci);
+}
+
+void ocfs2_metadata_cache_io_lock(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ci->ci_ops->co_io_lock(ci);
+}
+
+void ocfs2_metadata_cache_io_unlock(struct ocfs2_caching_info *ci)
+{
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ci->ci_ops->co_io_unlock(ci);
+}
+
+
+static void ocfs2_metadata_cache_reset(struct ocfs2_caching_info *ci,
+				       int clear)
+{
+	ci->ci_flags |= OCFS2_CACHE_FL_INLINE;
+	ci->ci_num_cached = 0;
+
+	if (clear) {
+		ci->ci_created_trans = 0;
+		ci->ci_last_trans = 0;
+	}
+}
+
+void ocfs2_metadata_cache_init(struct ocfs2_caching_info *ci,
+			       const struct ocfs2_caching_operations *ops)
+{
+	BUG_ON(!ops);
+
+	ci->ci_ops = ops;
+	ocfs2_metadata_cache_reset(ci, 1);
+}
+
+void ocfs2_metadata_cache_exit(struct ocfs2_caching_info *ci)
+{
+	ocfs2_metadata_cache_purge(ci);
+	ocfs2_metadata_cache_reset(ci, 1);
+}
+
+
+/* No lock taken here as 'root' is not expected to be visible to other
+ * processes. */
+static unsigned int ocfs2_purge_copied_metadata_tree(struct rb_root *root)
+{
+	unsigned int purged = 0;
+	struct rb_node *node;
+	struct ocfs2_meta_cache_item *item;
+
+	while ((node = rb_last(root)) != NULL) {
+		item = rb_entry(node, struct ocfs2_meta_cache_item, c_node);
+
+		trace_ocfs2_purge_copied_metadata_tree(
+					(unsigned long long) item->c_block);
+
+		rb_erase(&item->c_node, root);
+		kmem_cache_free(ocfs2_uptodate_cachep, item);
+
+		purged++;
+	}
+	return purged;
+}
+
+/* Called from locking and called from ocfs2_clear_inode. Dump the
+ * cache for a given inode.
+ *
+ * This function is a few more lines longer than necessary due to some
+ * accounting done here, but I think it's worth tracking down those
+ * bugs sooner -- Mark */
+void ocfs2_metadata_cache_purge(struct ocfs2_caching_info *ci)
+{
+	unsigned int tree, to_purge, purged;
+	struct rb_root root = RB_ROOT;
+
+	BUG_ON(!ci || !ci->ci_ops);
+
+	ocfs2_metadata_cache_lock(ci);
+	tree = !(ci->ci_flags & OCFS2_CACHE_FL_INLINE);
+	to_purge = ci->ci_num_cached;
+
+	trace_ocfs2_metadata_cache_purge(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		to_purge, tree);
+
+	/* If we're a tree, save off the root so that we can safely
+	 * initialize the cache. We do the work to free tree members
+	 * without the spinlock. */
+	if (tree)
+		root = ci->ci_cache.ci_tree;
+
+	ocfs2_metadata_cache_reset(ci, 0);
+	ocfs2_metadata_cache_unlock(ci);
+
+	purged = ocfs2_purge_copied_metadata_tree(&root);
+	/* If possible, track the number wiped so that we can more
+	 * easily detect counting errors. Unfortunately, this is only
+	 * meaningful for trees. */
+	if (tree && purged != to_purge)
+		mlog(ML_ERROR, "Owner %llu, count = %u, purged = %u\n",
+		     (unsigned long long)ocfs2_metadata_cache_owner(ci),
+		     to_purge, purged);
+}
+
+/* Returns the index in the cache array, -1 if not found.
+ * Requires ip_lock. */
+static int ocfs2_search_cache_array(struct ocfs2_caching_info *ci,
+				    sector_t item)
+{
+	int i;
+
+	for (i = 0; i < ci->ci_num_cached; i++) {
+		if (item == ci->ci_cache.ci_array[i])
+			return i;
+	}
+
+	return -1;
+}
+
+/* Returns the cache item if found, otherwise NULL.
+ * Requires ip_lock. */
+static struct ocfs2_meta_cache_item *
+ocfs2_search_cache_tree(struct ocfs2_caching_info *ci,
+			sector_t block)
+{
+	struct rb_node * n = ci->ci_cache.ci_tree.rb_node;
+	struct ocfs2_meta_cache_item *item = NULL;
+
+	while (n) {
+		item = rb_entry(n, struct ocfs2_meta_cache_item, c_node);
+
+		if (block < item->c_block)
+			n = n->rb_left;
+		else if (block > item->c_block)
+			n = n->rb_right;
+		else
+			return item;
+	}
+
+	return NULL;
+}
+
+static int ocfs2_buffer_cached(struct ocfs2_caching_info *ci,
+			       struct buffer_head *bh)
+{
+	int index = -1;
+	struct ocfs2_meta_cache_item *item = NULL;
+
+	ocfs2_metadata_cache_lock(ci);
+
+	trace_ocfs2_buffer_cached_begin(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long) bh->b_blocknr,
+		!!(ci->ci_flags & OCFS2_CACHE_FL_INLINE));
+
+	if (ci->ci_flags & OCFS2_CACHE_FL_INLINE)
+		index = ocfs2_search_cache_array(ci, bh->b_blocknr);
+	else
+		item = ocfs2_search_cache_tree(ci, bh->b_blocknr);
+
+	ocfs2_metadata_cache_unlock(ci);
+
+	trace_ocfs2_buffer_cached_end(index, item);
+
+	return (index != -1) || (item != NULL);
+}
+
+/* Warning: even if it returns true, this does *not* guarantee that
+ * the block is stored in our inode metadata cache.
+ *
+ * This can be called under lock_buffer()
+ */
+int ocfs2_buffer_uptodate(struct ocfs2_caching_info *ci,
+			  struct buffer_head *bh)
+{
+	/* Doesn't matter if the bh is in our cache or not -- if it's
+	 * not marked uptodate then we know it can't have correct
+	 * data. */
+	if (!buffer_uptodate(bh))
+		return 0;
+
+	/* OCFS2 does not allow multiple nodes to be changing the same
+	 * block at the same time. */
+	if (buffer_jbd(bh))
+		return 1;
+
+	/* Ok, locally the buffer is marked as up to date, now search
+	 * our cache to see if we can trust that. */
+	return ocfs2_buffer_cached(ci, bh);
+}
+
+/*
+ * Determine whether a buffer is currently out on a read-ahead request.
+ * ci_io_sem should be held to serialize submitters with the logic here.
+ */
+int ocfs2_buffer_read_ahead(struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh)
+{
+	return buffer_locked(bh) && ocfs2_buffer_cached(ci, bh);
+}
+
+/* Requires ip_lock */
+static void ocfs2_append_cache_array(struct ocfs2_caching_info *ci,
+				     sector_t block)
+{
+	BUG_ON(ci->ci_num_cached >= OCFS2_CACHE_INFO_MAX_ARRAY);
+
+	trace_ocfs2_append_cache_array(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long)block, ci->ci_num_cached);
+
+	ci->ci_cache.ci_array[ci->ci_num_cached] = block;
+	ci->ci_num_cached++;
+}
+
+/* By now the caller should have checked that the item does *not*
+ * exist in the tree.
+ * Requires ip_lock. */
+static void __ocfs2_insert_cache_tree(struct ocfs2_caching_info *ci,
+				      struct ocfs2_meta_cache_item *new)
+{
+	sector_t block = new->c_block;
+	struct rb_node *parent = NULL;
+	struct rb_node **p = &ci->ci_cache.ci_tree.rb_node;
+	struct ocfs2_meta_cache_item *tmp;
+
+	trace_ocfs2_insert_cache_tree(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long)block, ci->ci_num_cached);
+
+	while(*p) {
+		parent = *p;
+
+		tmp = rb_entry(parent, struct ocfs2_meta_cache_item, c_node);
+
+		if (block < tmp->c_block)
+			p = &(*p)->rb_left;
+		else if (block > tmp->c_block)
+			p = &(*p)->rb_right;
+		else {
+			/* This should never happen! */
+			mlog(ML_ERROR, "Duplicate block %llu cached!\n",
+			     (unsigned long long) block);
+			BUG();
+		}
+	}
+
+	rb_link_node(&new->c_node, parent, p);
+	rb_insert_color(&new->c_node, &ci->ci_cache.ci_tree);
+	ci->ci_num_cached++;
+}
+
+/* co_cache_lock() must be held */
+static inline int ocfs2_insert_can_use_array(struct ocfs2_caching_info *ci)
+{
+	return (ci->ci_flags & OCFS2_CACHE_FL_INLINE) &&
+		(ci->ci_num_cached < OCFS2_CACHE_INFO_MAX_ARRAY);
+}
+
+/* tree should be exactly OCFS2_CACHE_INFO_MAX_ARRAY wide. NULL the
+ * pointers in tree after we use them - this allows caller to detect
+ * when to free in case of error.
+ *
+ * The co_cache_lock() must be held. */
+static void ocfs2_expand_cache(struct ocfs2_caching_info *ci,
+			       struct ocfs2_meta_cache_item **tree)
+{
+	int i;
+
+	mlog_bug_on_msg(ci->ci_num_cached != OCFS2_CACHE_INFO_MAX_ARRAY,
+			"Owner %llu, num cached = %u, should be %u\n",
+			(unsigned long long)ocfs2_metadata_cache_owner(ci),
+			ci->ci_num_cached, OCFS2_CACHE_INFO_MAX_ARRAY);
+	mlog_bug_on_msg(!(ci->ci_flags & OCFS2_CACHE_FL_INLINE),
+			"Owner %llu not marked as inline anymore!\n",
+			(unsigned long long)ocfs2_metadata_cache_owner(ci));
+
+	/* Be careful to initialize the tree members *first* because
+	 * once the ci_tree is used, the array is junk... */
+	for (i = 0; i < OCFS2_CACHE_INFO_MAX_ARRAY; i++)
+		tree[i]->c_block = ci->ci_cache.ci_array[i];
+
+	ci->ci_flags &= ~OCFS2_CACHE_FL_INLINE;
+	ci->ci_cache.ci_tree = RB_ROOT;
+	/* this will be set again by __ocfs2_insert_cache_tree */
+	ci->ci_num_cached = 0;
+
+	for (i = 0; i < OCFS2_CACHE_INFO_MAX_ARRAY; i++) {
+		__ocfs2_insert_cache_tree(ci, tree[i]);
+		tree[i] = NULL;
+	}
+
+	trace_ocfs2_expand_cache(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		ci->ci_flags, ci->ci_num_cached);
+}
+
+/* Slow path function - memory allocation is necessary. See the
+ * comment above ocfs2_set_buffer_uptodate for more information. */
+static void __ocfs2_set_buffer_uptodate(struct ocfs2_caching_info *ci,
+					sector_t block,
+					int expand_tree)
+{
+	int i;
+	struct ocfs2_meta_cache_item *new = NULL;
+	struct ocfs2_meta_cache_item *tree[OCFS2_CACHE_INFO_MAX_ARRAY] =
+		{ NULL, };
+
+	trace_ocfs2_set_buffer_uptodate(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long)block, expand_tree);
+
+	new = kmem_cache_alloc(ocfs2_uptodate_cachep, GFP_NOFS);
+	if (!new) {
+		mlog_errno(-ENOMEM);
+		return;
+	}
+	new->c_block = block;
+
+	if (expand_tree) {
+		/* Do *not* allocate an array here - the removal code
+		 * has no way of tracking that. */
+		for (i = 0; i < OCFS2_CACHE_INFO_MAX_ARRAY; i++) {
+			tree[i] = kmem_cache_alloc(ocfs2_uptodate_cachep,
+						   GFP_NOFS);
+			if (!tree[i]) {
+				mlog_errno(-ENOMEM);
+				goto out_free;
+			}
+
+			/* These are initialized in ocfs2_expand_cache! */
+		}
+	}
+
+	ocfs2_metadata_cache_lock(ci);
+	if (ocfs2_insert_can_use_array(ci)) {
+		/* Ok, items were removed from the cache in between
+		 * locks. Detect this and revert back to the fast path */
+		ocfs2_append_cache_array(ci, block);
+		ocfs2_metadata_cache_unlock(ci);
+		goto out_free;
+	}
+
+	if (expand_tree)
+		ocfs2_expand_cache(ci, tree);
+
+	__ocfs2_insert_cache_tree(ci, new);
+	ocfs2_metadata_cache_unlock(ci);
+
+	new = NULL;
+out_free:
+	if (new)
+		kmem_cache_free(ocfs2_uptodate_cachep, new);
+
+	/* If these were used, then ocfs2_expand_cache re-set them to
+	 * NULL for us. */
+	if (tree[0]) {
+		for (i = 0; i < OCFS2_CACHE_INFO_MAX_ARRAY; i++)
+			if (tree[i])
+				kmem_cache_free(ocfs2_uptodate_cachep,
+						tree[i]);
+	}
+}
+
+/* Item insertion is guarded by co_io_lock(), so the insertion path takes
+ * advantage of this by not rechecking for a duplicate insert during
+ * the slow case. Additionally, if the cache needs to be bumped up to
+ * a tree, the code will not recheck after acquiring the lock --
+ * multiple paths cannot be expanding to a tree at the same time.
+ *
+ * The slow path takes into account that items can be removed
+ * (including the whole tree wiped and reset) when this process it out
+ * allocating memory. In those cases, it reverts back to the fast
+ * path.
+ *
+ * Note that this function may actually fail to insert the block if
+ * memory cannot be allocated. This is not fatal however (but may
+ * result in a performance penalty)
+ *
+ * Readahead buffers can be passed in here before the I/O request is
+ * completed.
+ */
+void ocfs2_set_buffer_uptodate(struct ocfs2_caching_info *ci,
+			       struct buffer_head *bh)
+{
+	int expand;
+
+	/* The block may very well exist in our cache already, so avoid
+	 * doing any more work in that case. */
+	if (ocfs2_buffer_cached(ci, bh))
+		return;
+
+	trace_ocfs2_set_buffer_uptodate_begin(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long)bh->b_blocknr);
+
+	/* No need to recheck under spinlock - insertion is guarded by
+	 * co_io_lock() */
+	ocfs2_metadata_cache_lock(ci);
+	if (ocfs2_insert_can_use_array(ci)) {
+		/* Fast case - it's an array and there's a free
+		 * spot. */
+		ocfs2_append_cache_array(ci, bh->b_blocknr);
+		ocfs2_metadata_cache_unlock(ci);
+		return;
+	}
+
+	expand = 0;
+	if (ci->ci_flags & OCFS2_CACHE_FL_INLINE) {
+		/* We need to bump things up to a tree. */
+		expand = 1;
+	}
+	ocfs2_metadata_cache_unlock(ci);
+
+	__ocfs2_set_buffer_uptodate(ci, bh->b_blocknr, expand);
+}
+
+/* Called against a newly allocated buffer. Most likely nobody should
+ * be able to read this sort of metadata while it's still being
+ * allocated, but this is careful to take co_io_lock() anyway. */
+void ocfs2_set_new_buffer_uptodate(struct ocfs2_caching_info *ci,
+				   struct buffer_head *bh)
+{
+	/* This should definitely *not* exist in our cache */
+	BUG_ON(ocfs2_buffer_cached(ci, bh));
+
+	set_buffer_uptodate(bh);
+
+	ocfs2_metadata_cache_io_lock(ci);
+	ocfs2_set_buffer_uptodate(ci, bh);
+	ocfs2_metadata_cache_io_unlock(ci);
+}
+
+/* Requires ip_lock. */
+static void ocfs2_remove_metadata_array(struct ocfs2_caching_info *ci,
+					int index)
+{
+	sector_t *array = ci->ci_cache.ci_array;
+	int bytes;
+
+	BUG_ON(index < 0 || index >= OCFS2_CACHE_INFO_MAX_ARRAY);
+	BUG_ON(index >= ci->ci_num_cached);
+	BUG_ON(!ci->ci_num_cached);
+
+	trace_ocfs2_remove_metadata_array(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		index, ci->ci_num_cached);
+
+	ci->ci_num_cached--;
+
+	/* don't need to copy if the array is now empty, or if we
+	 * removed at the tail */
+	if (ci->ci_num_cached && index < ci->ci_num_cached) {
+		bytes = sizeof(sector_t) * (ci->ci_num_cached - index);
+		memmove(&array[index], &array[index + 1], bytes);
+	}
+}
+
+/* Requires ip_lock. */
+static void ocfs2_remove_metadata_tree(struct ocfs2_caching_info *ci,
+				       struct ocfs2_meta_cache_item *item)
+{
+	trace_ocfs2_remove_metadata_tree(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long)item->c_block);
+
+	rb_erase(&item->c_node, &ci->ci_cache.ci_tree);
+	ci->ci_num_cached--;
+}
+
+static void ocfs2_remove_block_from_cache(struct ocfs2_caching_info *ci,
+					  sector_t block)
+{
+	int index;
+	struct ocfs2_meta_cache_item *item = NULL;
+
+	ocfs2_metadata_cache_lock(ci);
+	trace_ocfs2_remove_block_from_cache(
+		(unsigned long long)ocfs2_metadata_cache_owner(ci),
+		(unsigned long long) block, ci->ci_num_cached,
+		ci->ci_flags);
+
+	if (ci->ci_flags & OCFS2_CACHE_FL_INLINE) {
+		index = ocfs2_search_cache_array(ci, block);
+		if (index != -1)
+			ocfs2_remove_metadata_array(ci, index);
+	} else {
+		item = ocfs2_search_cache_tree(ci, block);
+		if (item)
+			ocfs2_remove_metadata_tree(ci, item);
+	}
+	ocfs2_metadata_cache_unlock(ci);
+
+	if (item)
+		kmem_cache_free(ocfs2_uptodate_cachep, item);
+}
+
+/*
+ * Called when we remove a chunk of metadata from an inode. We don't
+ * bother reverting things to an inlined array in the case of a remove
+ * which moves us back under the limit.
+ */
+void ocfs2_remove_from_cache(struct ocfs2_caching_info *ci,
+			     struct buffer_head *bh)
+{
+	sector_t block = bh->b_blocknr;
+
+	ocfs2_remove_block_from_cache(ci, block);
+}
+
+/* Called when we remove xattr clusters from an inode. */
+void ocfs2_remove_xattr_clusters_from_cache(struct ocfs2_caching_info *ci,
+					    sector_t block,
+					    u32 c_len)
+{
+	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
+	unsigned int i, b_len = ocfs2_clusters_to_blocks(sb, 1) * c_len;
+
+	for (i = 0; i < b_len; i++, block++)
+		ocfs2_remove_block_from_cache(ci, block);
+}
+
+int __init init_ocfs2_uptodate_cache(void)
+{
+	ocfs2_uptodate_cachep = kmem_cache_create("ocfs2_uptodate",
+				  sizeof(struct ocfs2_meta_cache_item),
+				  0, SLAB_HWCACHE_ALIGN, NULL);
+	if (!ocfs2_uptodate_cachep)
+		return -ENOMEM;
+
+	return 0;
+}
+
+void exit_ocfs2_uptodate_cache(void)
+{
+	if (ocfs2_uptodate_cachep)
+		kmem_cache_destroy(ocfs2_uptodate_cachep);
+}
diff --git a/fs/ocfs2/uptodate.h b/fs/ocfs2/uptodate.h
new file mode 100644
index 000000000..0d826fe2d
--- /dev/null
+++ b/fs/ocfs2/uptodate.h
@@ -0,0 +1,84 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * uptodate.h
+ *
+ * Cluster uptodate tracking
+ *
+ * Copyright (C) 2002, 2004, 2005 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License as published by the Free Software Foundation; either
+ * version 2 of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef OCFS2_UPTODATE_H
+#define OCFS2_UPTODATE_H
+
+/*
+ * The caching code relies on locking provided by the user of
+ * struct ocfs2_caching_info.  These operations connect that up.
+ */
+struct ocfs2_caching_operations {
+	/*
+	 * A u64 representing the owning structure.  Usually this
+	 * is the block number (i_blkno or whatnot).  This is used so
+	 * that caching log messages can identify the owning structure.
+	 */
+	u64	(*co_owner)(struct ocfs2_caching_info *ci);
+
+	/* The superblock is needed during I/O. */
+	struct super_block *(*co_get_super)(struct ocfs2_caching_info *ci);
+	/*
+	 * Lock and unlock the caching data.  These will not sleep, and
+	 * should probably be spinlocks.
+	 */
+	void	(*co_cache_lock)(struct ocfs2_caching_info *ci);
+	void	(*co_cache_unlock)(struct ocfs2_caching_info *ci);
+
+	/*
+	 * Lock and unlock for disk I/O.  These will sleep, and should
+	 * be mutexes.
+	 */
+	void	(*co_io_lock)(struct ocfs2_caching_info *ci);
+	void	(*co_io_unlock)(struct ocfs2_caching_info *ci);
+};
+
+int __init init_ocfs2_uptodate_cache(void);
+void exit_ocfs2_uptodate_cache(void);
+
+void ocfs2_metadata_cache_init(struct ocfs2_caching_info *ci,
+			       const struct ocfs2_caching_operations *ops);
+void ocfs2_metadata_cache_purge(struct ocfs2_caching_info *ci);
+void ocfs2_metadata_cache_exit(struct ocfs2_caching_info *ci);
+
+u64 ocfs2_metadata_cache_owner(struct ocfs2_caching_info *ci);
+void ocfs2_metadata_cache_io_lock(struct ocfs2_caching_info *ci);
+void ocfs2_metadata_cache_io_unlock(struct ocfs2_caching_info *ci);
+
+int ocfs2_buffer_uptodate(struct ocfs2_caching_info *ci,
+			  struct buffer_head *bh);
+void ocfs2_set_buffer_uptodate(struct ocfs2_caching_info *ci,
+			       struct buffer_head *bh);
+void ocfs2_set_new_buffer_uptodate(struct ocfs2_caching_info *ci,
+				   struct buffer_head *bh);
+void ocfs2_remove_from_cache(struct ocfs2_caching_info *ci,
+			     struct buffer_head *bh);
+void ocfs2_remove_xattr_clusters_from_cache(struct ocfs2_caching_info *ci,
+					    sector_t block,
+					    u32 c_len);
+int ocfs2_buffer_read_ahead(struct ocfs2_caching_info *ci,
+			    struct buffer_head *bh);
+
+#endif /* OCFS2_UPTODATE_H */
diff --git a/fs/ocfs2/xattr.c b/fs/ocfs2/xattr.c
new file mode 100644
index 000000000..d03bfbf3d
--- /dev/null
+++ b/fs/ocfs2/xattr.c
@@ -0,0 +1,7425 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * xattr.c
+ *
+ * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
+ *
+ * CREDITS:
+ * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
+ * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#include <linux/capability.h>
+#include <linux/fs.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/pagemap.h>
+#include <linux/uio.h>
+#include <linux/sched.h>
+#include <linux/splice.h>
+#include <linux/mount.h>
+#include <linux/writeback.h>
+#include <linux/falloc.h>
+#include <linux/sort.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/security.h>
+
+#include <cluster/masklog.h>
+
+#include "ocfs2.h"
+#include "alloc.h"
+#include "blockcheck.h"
+#include "dlmglue.h"
+#include "file.h"
+#include "symlink.h"
+#include "sysfile.h"
+#include "inode.h"
+#include "journal.h"
+#include "ocfs2_fs.h"
+#include "suballoc.h"
+#include "uptodate.h"
+#include "buffer_head_io.h"
+#include "super.h"
+#include "xattr.h"
+#include "refcounttree.h"
+#include "acl.h"
+#include "ocfs2_trace.h"
+
+struct ocfs2_xattr_def_value_root {
+	struct ocfs2_xattr_value_root	xv;
+	struct ocfs2_extent_rec		er;
+};
+
+struct ocfs2_xattr_bucket {
+	/* The inode these xattrs are associated with */
+	struct inode *bu_inode;
+
+	/* The actual buffers that make up the bucket */
+	struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
+
+	/* How many blocks make up one bucket for this filesystem */
+	int bu_blocks;
+};
+
+struct ocfs2_xattr_set_ctxt {
+	handle_t *handle;
+	struct ocfs2_alloc_context *meta_ac;
+	struct ocfs2_alloc_context *data_ac;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	int set_abort;
+};
+
+#define OCFS2_XATTR_ROOT_SIZE	(sizeof(struct ocfs2_xattr_def_value_root))
+#define OCFS2_XATTR_INLINE_SIZE	80
+#define OCFS2_XATTR_HEADER_GAP	4
+#define OCFS2_XATTR_FREE_IN_IBODY	(OCFS2_MIN_XATTR_INLINE_SIZE \
+					 - sizeof(struct ocfs2_xattr_header) \
+					 - OCFS2_XATTR_HEADER_GAP)
+#define OCFS2_XATTR_FREE_IN_BLOCK(ptr)	((ptr)->i_sb->s_blocksize \
+					 - sizeof(struct ocfs2_xattr_block) \
+					 - sizeof(struct ocfs2_xattr_header) \
+					 - OCFS2_XATTR_HEADER_GAP)
+
+static struct ocfs2_xattr_def_value_root def_xv = {
+	.xv.xr_list.l_count = cpu_to_le16(1),
+};
+
+const struct xattr_handler *ocfs2_xattr_handlers[] = {
+	&ocfs2_xattr_user_handler,
+	&posix_acl_access_xattr_handler,
+	&posix_acl_default_xattr_handler,
+	&ocfs2_xattr_trusted_handler,
+	&ocfs2_xattr_security_handler,
+	NULL
+};
+
+static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
+	[OCFS2_XATTR_INDEX_USER]	= &ocfs2_xattr_user_handler,
+	[OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
+					= &posix_acl_access_xattr_handler,
+	[OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
+					= &posix_acl_default_xattr_handler,
+	[OCFS2_XATTR_INDEX_TRUSTED]	= &ocfs2_xattr_trusted_handler,
+	[OCFS2_XATTR_INDEX_SECURITY]	= &ocfs2_xattr_security_handler,
+};
+
+struct ocfs2_xattr_info {
+	int		xi_name_index;
+	const char	*xi_name;
+	int		xi_name_len;
+	const void	*xi_value;
+	size_t		xi_value_len;
+};
+
+struct ocfs2_xattr_search {
+	struct buffer_head *inode_bh;
+	/*
+	 * xattr_bh point to the block buffer head which has extended attribute
+	 * when extended attribute in inode, xattr_bh is equal to inode_bh.
+	 */
+	struct buffer_head *xattr_bh;
+	struct ocfs2_xattr_header *header;
+	struct ocfs2_xattr_bucket *bucket;
+	void *base;
+	void *end;
+	struct ocfs2_xattr_entry *here;
+	int not_found;
+};
+
+/* Operations on struct ocfs2_xa_entry */
+struct ocfs2_xa_loc;
+struct ocfs2_xa_loc_operations {
+	/*
+	 * Journal functions
+	 */
+	int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
+				  int type);
+	void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
+
+	/*
+	 * Return a pointer to the appropriate buffer in loc->xl_storage
+	 * at the given offset from loc->xl_header.
+	 */
+	void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
+
+	/* Can we reuse the existing entry for the new value? */
+	int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
+			     struct ocfs2_xattr_info *xi);
+
+	/* How much space is needed for the new value? */
+	int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
+			       struct ocfs2_xattr_info *xi);
+
+	/*
+	 * Return the offset of the first name+value pair.  This is
+	 * the start of our downward-filling free space.
+	 */
+	int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
+
+	/*
+	 * Remove the name+value at this location.  Do whatever is
+	 * appropriate with the remaining name+value pairs.
+	 */
+	void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
+
+	/* Fill xl_entry with a new entry */
+	void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
+
+	/* Add name+value storage to an entry */
+	void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
+
+	/*
+	 * Initialize the value buf's access and bh fields for this entry.
+	 * ocfs2_xa_fill_value_buf() will handle the xv pointer.
+	 */
+	void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
+				   struct ocfs2_xattr_value_buf *vb);
+};
+
+/*
+ * Describes an xattr entry location.  This is a memory structure
+ * tracking the on-disk structure.
+ */
+struct ocfs2_xa_loc {
+	/* This xattr belongs to this inode */
+	struct inode *xl_inode;
+
+	/* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
+	struct ocfs2_xattr_header *xl_header;
+
+	/* Bytes from xl_header to the end of the storage */
+	int xl_size;
+
+	/*
+	 * The ocfs2_xattr_entry this location describes.  If this is
+	 * NULL, this location describes the on-disk structure where it
+	 * would have been.
+	 */
+	struct ocfs2_xattr_entry *xl_entry;
+
+	/*
+	 * Internal housekeeping
+	 */
+
+	/* Buffer(s) containing this entry */
+	void *xl_storage;
+
+	/* Operations on the storage backing this location */
+	const struct ocfs2_xa_loc_operations *xl_ops;
+};
+
+/*
+ * Convenience functions to calculate how much space is needed for a
+ * given name+value pair
+ */
+static int namevalue_size(int name_len, uint64_t value_len)
+{
+	if (value_len > OCFS2_XATTR_INLINE_SIZE)
+		return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
+	else
+		return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
+}
+
+static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
+{
+	return namevalue_size(xi->xi_name_len, xi->xi_value_len);
+}
+
+static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
+{
+	u64 value_len = le64_to_cpu(xe->xe_value_size);
+
+	BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
+	       ocfs2_xattr_is_local(xe));
+	return namevalue_size(xe->xe_name_len, value_len);
+}
+
+
+static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
+					     struct ocfs2_xattr_header *xh,
+					     int index,
+					     int *block_off,
+					     int *new_offset);
+
+static int ocfs2_xattr_block_find(struct inode *inode,
+				  int name_index,
+				  const char *name,
+				  struct ocfs2_xattr_search *xs);
+static int ocfs2_xattr_index_block_find(struct inode *inode,
+					struct buffer_head *root_bh,
+					int name_index,
+					const char *name,
+					struct ocfs2_xattr_search *xs);
+
+static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
+					struct buffer_head *blk_bh,
+					char *buffer,
+					size_t buffer_size);
+
+static int ocfs2_xattr_create_index_block(struct inode *inode,
+					  struct ocfs2_xattr_search *xs,
+					  struct ocfs2_xattr_set_ctxt *ctxt);
+
+static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
+					     struct ocfs2_xattr_info *xi,
+					     struct ocfs2_xattr_search *xs,
+					     struct ocfs2_xattr_set_ctxt *ctxt);
+
+typedef int (xattr_tree_rec_func)(struct inode *inode,
+				  struct buffer_head *root_bh,
+				  u64 blkno, u32 cpos, u32 len, void *para);
+static int ocfs2_iterate_xattr_index_block(struct inode *inode,
+					   struct buffer_head *root_bh,
+					   xattr_tree_rec_func *rec_func,
+					   void *para);
+static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
+					struct ocfs2_xattr_bucket *bucket,
+					void *para);
+static int ocfs2_rm_xattr_cluster(struct inode *inode,
+				  struct buffer_head *root_bh,
+				  u64 blkno,
+				  u32 cpos,
+				  u32 len,
+				  void *para);
+
+static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
+				  u64 src_blk, u64 last_blk, u64 to_blk,
+				  unsigned int start_bucket,
+				  u32 *first_hash);
+static int ocfs2_prepare_refcount_xattr(struct inode *inode,
+					struct ocfs2_dinode *di,
+					struct ocfs2_xattr_info *xi,
+					struct ocfs2_xattr_search *xis,
+					struct ocfs2_xattr_search *xbs,
+					struct ocfs2_refcount_tree **ref_tree,
+					int *meta_need,
+					int *credits);
+static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
+					   struct ocfs2_xattr_bucket *bucket,
+					   int offset,
+					   struct ocfs2_xattr_value_root **xv,
+					   struct buffer_head **bh);
+
+static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
+{
+	return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
+}
+
+static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
+{
+	return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
+}
+
+#define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
+#define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
+#define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
+
+static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
+{
+	struct ocfs2_xattr_bucket *bucket;
+	int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+
+	BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
+
+	bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
+	if (bucket) {
+		bucket->bu_inode = inode;
+		bucket->bu_blocks = blks;
+	}
+
+	return bucket;
+}
+
+static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
+{
+	int i;
+
+	for (i = 0; i < bucket->bu_blocks; i++) {
+		brelse(bucket->bu_bhs[i]);
+		bucket->bu_bhs[i] = NULL;
+	}
+}
+
+static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
+{
+	if (bucket) {
+		ocfs2_xattr_bucket_relse(bucket);
+		bucket->bu_inode = NULL;
+		kfree(bucket);
+	}
+}
+
+/*
+ * A bucket that has never been written to disk doesn't need to be
+ * read.  We just need the buffer_heads.  Don't call this for
+ * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
+ * them fully.
+ */
+static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
+				   u64 xb_blkno, int new)
+{
+	int i, rc = 0;
+
+	for (i = 0; i < bucket->bu_blocks; i++) {
+		bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
+					      xb_blkno + i);
+		if (!bucket->bu_bhs[i]) {
+			rc = -ENOMEM;
+			mlog_errno(rc);
+			break;
+		}
+
+		if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
+					   bucket->bu_bhs[i])) {
+			if (new)
+				ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
+							      bucket->bu_bhs[i]);
+			else {
+				set_buffer_uptodate(bucket->bu_bhs[i]);
+				ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
+							  bucket->bu_bhs[i]);
+			}
+		}
+	}
+
+	if (rc)
+		ocfs2_xattr_bucket_relse(bucket);
+	return rc;
+}
+
+/* Read the xattr bucket at xb_blkno */
+static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
+				   u64 xb_blkno)
+{
+	int rc;
+
+	rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
+			       bucket->bu_blocks, bucket->bu_bhs, 0,
+			       NULL);
+	if (!rc) {
+		spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
+		rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
+						 bucket->bu_bhs,
+						 bucket->bu_blocks,
+						 &bucket_xh(bucket)->xh_check);
+		spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
+		if (rc)
+			mlog_errno(rc);
+	}
+
+	if (rc)
+		ocfs2_xattr_bucket_relse(bucket);
+	return rc;
+}
+
+static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
+					     struct ocfs2_xattr_bucket *bucket,
+					     int type)
+{
+	int i, rc = 0;
+
+	for (i = 0; i < bucket->bu_blocks; i++) {
+		rc = ocfs2_journal_access(handle,
+					  INODE_CACHE(bucket->bu_inode),
+					  bucket->bu_bhs[i], type);
+		if (rc) {
+			mlog_errno(rc);
+			break;
+		}
+	}
+
+	return rc;
+}
+
+static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
+					     struct ocfs2_xattr_bucket *bucket)
+{
+	int i;
+
+	spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
+	ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
+				   bucket->bu_bhs, bucket->bu_blocks,
+				   &bucket_xh(bucket)->xh_check);
+	spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
+
+	for (i = 0; i < bucket->bu_blocks; i++)
+		ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
+}
+
+static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
+					 struct ocfs2_xattr_bucket *src)
+{
+	int i;
+	int blocksize = src->bu_inode->i_sb->s_blocksize;
+
+	BUG_ON(dest->bu_blocks != src->bu_blocks);
+	BUG_ON(dest->bu_inode != src->bu_inode);
+
+	for (i = 0; i < src->bu_blocks; i++) {
+		memcpy(bucket_block(dest, i), bucket_block(src, i),
+		       blocksize);
+	}
+}
+
+static int ocfs2_validate_xattr_block(struct super_block *sb,
+				      struct buffer_head *bh)
+{
+	int rc;
+	struct ocfs2_xattr_block *xb =
+		(struct ocfs2_xattr_block *)bh->b_data;
+
+	trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
+
+	BUG_ON(!buffer_uptodate(bh));
+
+	/*
+	 * If the ecc fails, we return the error but otherwise
+	 * leave the filesystem running.  We know any error is
+	 * local to this block.
+	 */
+	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
+	if (rc)
+		return rc;
+
+	/*
+	 * Errors after here are fatal
+	 */
+
+	if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
+		ocfs2_error(sb,
+			    "Extended attribute block #%llu has bad "
+			    "signature %.*s",
+			    (unsigned long long)bh->b_blocknr, 7,
+			    xb->xb_signature);
+		return -EINVAL;
+	}
+
+	if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
+		ocfs2_error(sb,
+			    "Extended attribute block #%llu has an "
+			    "invalid xb_blkno of %llu",
+			    (unsigned long long)bh->b_blocknr,
+			    (unsigned long long)le64_to_cpu(xb->xb_blkno));
+		return -EINVAL;
+	}
+
+	if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
+		ocfs2_error(sb,
+			    "Extended attribute block #%llu has an invalid "
+			    "xb_fs_generation of #%u",
+			    (unsigned long long)bh->b_blocknr,
+			    le32_to_cpu(xb->xb_fs_generation));
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
+				  struct buffer_head **bh)
+{
+	int rc;
+	struct buffer_head *tmp = *bh;
+
+	rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
+			      ocfs2_validate_xattr_block);
+
+	/* If ocfs2_read_block() got us a new bh, pass it up. */
+	if (!rc && !*bh)
+		*bh = tmp;
+
+	return rc;
+}
+
+static inline const char *ocfs2_xattr_prefix(int name_index)
+{
+	const struct xattr_handler *handler = NULL;
+
+	if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
+		handler = ocfs2_xattr_handler_map[name_index];
+
+	return handler ? handler->prefix : NULL;
+}
+
+static u32 ocfs2_xattr_name_hash(struct inode *inode,
+				 const char *name,
+				 int name_len)
+{
+	/* Get hash value of uuid from super block */
+	u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
+	int i;
+
+	/* hash extended attribute name */
+	for (i = 0; i < name_len; i++) {
+		hash = (hash << OCFS2_HASH_SHIFT) ^
+		       (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
+		       *name++;
+	}
+
+	return hash;
+}
+
+static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
+{
+	return namevalue_size(name_len, value_len) +
+		sizeof(struct ocfs2_xattr_entry);
+}
+
+static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
+{
+	return namevalue_size_xi(xi) +
+		sizeof(struct ocfs2_xattr_entry);
+}
+
+static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
+{
+	return namevalue_size_xe(xe) +
+		sizeof(struct ocfs2_xattr_entry);
+}
+
+int ocfs2_calc_security_init(struct inode *dir,
+			     struct ocfs2_security_xattr_info *si,
+			     int *want_clusters,
+			     int *xattr_credits,
+			     struct ocfs2_alloc_context **xattr_ac)
+{
+	int ret = 0;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
+						 si->value_len);
+
+	/*
+	 * The max space of security xattr taken inline is
+	 * 256(name) + 80(value) + 16(entry) = 352 bytes,
+	 * So reserve one metadata block for it is ok.
+	 */
+	if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
+	    s_size > OCFS2_XATTR_FREE_IN_IBODY) {
+		ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
+		if (ret) {
+			mlog_errno(ret);
+			return ret;
+		}
+		*xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
+	}
+
+	/* reserve clusters for xattr value which will be set in B tree*/
+	if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
+		int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
+							    si->value_len);
+
+		*xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
+							   new_clusters);
+		*want_clusters += new_clusters;
+	}
+	return ret;
+}
+
+int ocfs2_calc_xattr_init(struct inode *dir,
+			  struct buffer_head *dir_bh,
+			  umode_t mode,
+			  struct ocfs2_security_xattr_info *si,
+			  int *want_clusters,
+			  int *xattr_credits,
+			  int *want_meta)
+{
+	int ret = 0;
+	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
+	int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
+
+	if (si->enable)
+		s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
+						     si->value_len);
+
+	if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
+		acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
+					OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
+					"", NULL, 0);
+		if (acl_len > 0) {
+			a_size = ocfs2_xattr_entry_real_size(0, acl_len);
+			if (S_ISDIR(mode))
+				a_size <<= 1;
+		} else if (acl_len != 0 && acl_len != -ENODATA) {
+			mlog_errno(ret);
+			return ret;
+		}
+	}
+
+	if (!(s_size + a_size))
+		return ret;
+
+	/*
+	 * The max space of security xattr taken inline is
+	 * 256(name) + 80(value) + 16(entry) = 352 bytes,
+	 * The max space of acl xattr taken inline is
+	 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
+	 * when blocksize = 512, may reserve one more cluser for
+	 * xattr bucket, otherwise reserve one metadata block
+	 * for them is ok.
+	 * If this is a new directory with inline data,
+	 * we choose to reserve the entire inline area for
+	 * directory contents and force an external xattr block.
+	 */
+	if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
+	    (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
+	    (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
+		*want_meta = *want_meta + 1;
+		*xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
+	}
+
+	if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
+	    (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
+		*want_clusters += 1;
+		*xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
+	}
+
+	/*
+	 * reserve credits and clusters for xattrs which has large value
+	 * and have to be set outside
+	 */
+	if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
+		new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
+							si->value_len);
+		*xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
+							   new_clusters);
+		*want_clusters += new_clusters;
+	}
+	if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
+	    acl_len > OCFS2_XATTR_INLINE_SIZE) {
+		/* for directory, it has DEFAULT and ACCESS two types of acls */
+		new_clusters = (S_ISDIR(mode) ? 2 : 1) *
+				ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
+		*xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
+							   new_clusters);
+		*want_clusters += new_clusters;
+	}
+
+	return ret;
+}
+
+static int ocfs2_xattr_extend_allocation(struct inode *inode,
+					 u32 clusters_to_add,
+					 struct ocfs2_xattr_value_buf *vb,
+					 struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int status = 0, credits;
+	handle_t *handle = ctxt->handle;
+	enum ocfs2_alloc_restarted why;
+	u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
+	struct ocfs2_extent_tree et;
+
+	ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
+
+	while (clusters_to_add) {
+		trace_ocfs2_xattr_extend_allocation(clusters_to_add);
+
+		status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
+				       OCFS2_JOURNAL_ACCESS_WRITE);
+		if (status < 0) {
+			mlog_errno(status);
+			break;
+		}
+
+		prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
+		status = ocfs2_add_clusters_in_btree(handle,
+						     &et,
+						     &logical_start,
+						     clusters_to_add,
+						     0,
+						     ctxt->data_ac,
+						     ctxt->meta_ac,
+						     &why);
+		if ((status < 0) && (status != -EAGAIN)) {
+			if (status != -ENOSPC)
+				mlog_errno(status);
+			break;
+		}
+
+		ocfs2_journal_dirty(handle, vb->vb_bh);
+
+		clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
+					 prev_clusters;
+
+		if (why != RESTART_NONE && clusters_to_add) {
+			/*
+			 * We can only fail in case the alloc file doesn't give
+			 * up enough clusters.
+			 */
+			BUG_ON(why == RESTART_META);
+
+			credits = ocfs2_calc_extend_credits(inode->i_sb,
+							    &vb->vb_xv->xr_list);
+			status = ocfs2_extend_trans(handle, credits);
+			if (status < 0) {
+				status = -ENOMEM;
+				mlog_errno(status);
+				break;
+			}
+		}
+	}
+
+	return status;
+}
+
+static int __ocfs2_remove_xattr_range(struct inode *inode,
+				      struct ocfs2_xattr_value_buf *vb,
+				      u32 cpos, u32 phys_cpos, u32 len,
+				      unsigned int ext_flags,
+				      struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret;
+	u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
+	handle_t *handle = ctxt->handle;
+	struct ocfs2_extent_tree et;
+
+	ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
+
+	ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
+			    OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
+				  &ctxt->dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
+	ocfs2_journal_dirty(handle, vb->vb_bh);
+
+	if (ext_flags & OCFS2_EXT_REFCOUNTED)
+		ret = ocfs2_decrease_refcount(inode, handle,
+					ocfs2_blocks_to_clusters(inode->i_sb,
+								 phys_blkno),
+					len, ctxt->meta_ac, &ctxt->dealloc, 1);
+	else
+		ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
+						  phys_blkno, len);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+static int ocfs2_xattr_shrink_size(struct inode *inode,
+				   u32 old_clusters,
+				   u32 new_clusters,
+				   struct ocfs2_xattr_value_buf *vb,
+				   struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret = 0;
+	unsigned int ext_flags;
+	u32 trunc_len, cpos, phys_cpos, alloc_size;
+	u64 block;
+
+	if (old_clusters <= new_clusters)
+		return 0;
+
+	cpos = new_clusters;
+	trunc_len = old_clusters - new_clusters;
+	while (trunc_len) {
+		ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
+					       &alloc_size,
+					       &vb->vb_xv->xr_list, &ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		if (alloc_size > trunc_len)
+			alloc_size = trunc_len;
+
+		ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
+						 phys_cpos, alloc_size,
+						 ext_flags, ctxt);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
+		ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
+						       block, alloc_size);
+		cpos += alloc_size;
+		trunc_len -= alloc_size;
+	}
+
+out:
+	return ret;
+}
+
+static int ocfs2_xattr_value_truncate(struct inode *inode,
+				      struct ocfs2_xattr_value_buf *vb,
+				      int len,
+				      struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret;
+	u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
+	u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
+
+	if (new_clusters == old_clusters)
+		return 0;
+
+	if (new_clusters > old_clusters)
+		ret = ocfs2_xattr_extend_allocation(inode,
+						    new_clusters - old_clusters,
+						    vb, ctxt);
+	else
+		ret = ocfs2_xattr_shrink_size(inode,
+					      old_clusters, new_clusters,
+					      vb, ctxt);
+
+	return ret;
+}
+
+static int ocfs2_xattr_list_entry(char *buffer, size_t size,
+				  size_t *result, const char *prefix,
+				  const char *name, int name_len)
+{
+	char *p = buffer + *result;
+	int prefix_len = strlen(prefix);
+	int total_len = prefix_len + name_len + 1;
+
+	*result += total_len;
+
+	/* we are just looking for how big our buffer needs to be */
+	if (!size)
+		return 0;
+
+	if (*result > size)
+		return -ERANGE;
+
+	memcpy(p, prefix, prefix_len);
+	memcpy(p + prefix_len, name, name_len);
+	p[prefix_len + name_len] = '\0';
+
+	return 0;
+}
+
+static int ocfs2_xattr_list_entries(struct inode *inode,
+				    struct ocfs2_xattr_header *header,
+				    char *buffer, size_t buffer_size)
+{
+	size_t result = 0;
+	int i, type, ret;
+	const char *prefix, *name;
+
+	for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
+		struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
+		type = ocfs2_xattr_get_type(entry);
+		prefix = ocfs2_xattr_prefix(type);
+
+		if (prefix) {
+			name = (const char *)header +
+				le16_to_cpu(entry->xe_name_offset);
+
+			ret = ocfs2_xattr_list_entry(buffer, buffer_size,
+						     &result, prefix, name,
+						     entry->xe_name_len);
+			if (ret)
+				return ret;
+		}
+	}
+
+	return result;
+}
+
+int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
+					 struct ocfs2_dinode *di)
+{
+	struct ocfs2_xattr_header *xh;
+	int i;
+
+	xh = (struct ocfs2_xattr_header *)
+		 ((void *)di + inode->i_sb->s_blocksize -
+		 le16_to_cpu(di->i_xattr_inline_size));
+
+	for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
+		if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
+			return 1;
+
+	return 0;
+}
+
+static int ocfs2_xattr_ibody_list(struct inode *inode,
+				  struct ocfs2_dinode *di,
+				  char *buffer,
+				  size_t buffer_size)
+{
+	struct ocfs2_xattr_header *header = NULL;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	int ret = 0;
+
+	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
+		return ret;
+
+	header = (struct ocfs2_xattr_header *)
+		 ((void *)di + inode->i_sb->s_blocksize -
+		 le16_to_cpu(di->i_xattr_inline_size));
+
+	ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
+
+	return ret;
+}
+
+static int ocfs2_xattr_block_list(struct inode *inode,
+				  struct ocfs2_dinode *di,
+				  char *buffer,
+				  size_t buffer_size)
+{
+	struct buffer_head *blk_bh = NULL;
+	struct ocfs2_xattr_block *xb;
+	int ret = 0;
+
+	if (!di->i_xattr_loc)
+		return ret;
+
+	ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
+				     &blk_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
+	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
+		struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
+		ret = ocfs2_xattr_list_entries(inode, header,
+					       buffer, buffer_size);
+	} else
+		ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
+						   buffer, buffer_size);
+
+	brelse(blk_bh);
+
+	return ret;
+}
+
+ssize_t ocfs2_listxattr(struct dentry *dentry,
+			char *buffer,
+			size_t size)
+{
+	int ret = 0, i_ret = 0, b_ret = 0;
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di = NULL;
+	struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
+
+	if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
+		return -EOPNOTSUPP;
+
+	if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
+		return ret;
+
+	ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	down_read(&oi->ip_xattr_sem);
+	i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
+	if (i_ret < 0)
+		b_ret = 0;
+	else {
+		if (buffer) {
+			buffer += i_ret;
+			size -= i_ret;
+		}
+		b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
+					       buffer, size);
+		if (b_ret < 0)
+			i_ret = 0;
+	}
+	up_read(&oi->ip_xattr_sem);
+	ocfs2_inode_unlock(d_inode(dentry), 0);
+
+	brelse(di_bh);
+
+	return i_ret + b_ret;
+}
+
+static int ocfs2_xattr_find_entry(int name_index,
+				  const char *name,
+				  struct ocfs2_xattr_search *xs)
+{
+	struct ocfs2_xattr_entry *entry;
+	size_t name_len;
+	int i, cmp = 1;
+
+	if (name == NULL)
+		return -EINVAL;
+
+	name_len = strlen(name);
+	entry = xs->here;
+	for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
+		cmp = name_index - ocfs2_xattr_get_type(entry);
+		if (!cmp)
+			cmp = name_len - entry->xe_name_len;
+		if (!cmp)
+			cmp = memcmp(name, (xs->base +
+				     le16_to_cpu(entry->xe_name_offset)),
+				     name_len);
+		if (cmp == 0)
+			break;
+		entry += 1;
+	}
+	xs->here = entry;
+
+	return cmp ? -ENODATA : 0;
+}
+
+static int ocfs2_xattr_get_value_outside(struct inode *inode,
+					 struct ocfs2_xattr_value_root *xv,
+					 void *buffer,
+					 size_t len)
+{
+	u32 cpos, p_cluster, num_clusters, bpc, clusters;
+	u64 blkno;
+	int i, ret = 0;
+	size_t cplen, blocksize;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_extent_list *el;
+
+	el = &xv->xr_list;
+	clusters = le32_to_cpu(xv->xr_clusters);
+	bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
+	blocksize = inode->i_sb->s_blocksize;
+
+	cpos = 0;
+	while (cpos < clusters) {
+		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
+					       &num_clusters, el, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
+		/* Copy ocfs2_xattr_value */
+		for (i = 0; i < num_clusters * bpc; i++, blkno++) {
+			ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
+					       &bh, NULL);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			cplen = len >= blocksize ? blocksize : len;
+			memcpy(buffer, bh->b_data, cplen);
+			len -= cplen;
+			buffer += cplen;
+
+			brelse(bh);
+			bh = NULL;
+			if (len == 0)
+				break;
+		}
+		cpos += num_clusters;
+	}
+out:
+	return ret;
+}
+
+static int ocfs2_xattr_ibody_get(struct inode *inode,
+				 int name_index,
+				 const char *name,
+				 void *buffer,
+				 size_t buffer_size,
+				 struct ocfs2_xattr_search *xs)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
+	struct ocfs2_xattr_value_root *xv;
+	size_t size;
+	int ret = 0;
+
+	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
+		return -ENODATA;
+
+	xs->end = (void *)di + inode->i_sb->s_blocksize;
+	xs->header = (struct ocfs2_xattr_header *)
+			(xs->end - le16_to_cpu(di->i_xattr_inline_size));
+	xs->base = (void *)xs->header;
+	xs->here = xs->header->xh_entries;
+
+	ret = ocfs2_xattr_find_entry(name_index, name, xs);
+	if (ret)
+		return ret;
+	size = le64_to_cpu(xs->here->xe_value_size);
+	if (buffer) {
+		if (size > buffer_size)
+			return -ERANGE;
+		if (ocfs2_xattr_is_local(xs->here)) {
+			memcpy(buffer, (void *)xs->base +
+			       le16_to_cpu(xs->here->xe_name_offset) +
+			       OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
+		} else {
+			xv = (struct ocfs2_xattr_value_root *)
+				(xs->base + le16_to_cpu(
+				 xs->here->xe_name_offset) +
+				OCFS2_XATTR_SIZE(xs->here->xe_name_len));
+			ret = ocfs2_xattr_get_value_outside(inode, xv,
+							    buffer, size);
+			if (ret < 0) {
+				mlog_errno(ret);
+				return ret;
+			}
+		}
+	}
+
+	return size;
+}
+
+static int ocfs2_xattr_block_get(struct inode *inode,
+				 int name_index,
+				 const char *name,
+				 void *buffer,
+				 size_t buffer_size,
+				 struct ocfs2_xattr_search *xs)
+{
+	struct ocfs2_xattr_block *xb;
+	struct ocfs2_xattr_value_root *xv;
+	size_t size;
+	int ret = -ENODATA, name_offset, name_len, i;
+	int uninitialized_var(block_off);
+
+	xs->bucket = ocfs2_xattr_bucket_new(inode);
+	if (!xs->bucket) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto cleanup;
+	}
+
+	ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
+	if (ret) {
+		mlog_errno(ret);
+		goto cleanup;
+	}
+
+	if (xs->not_found) {
+		ret = -ENODATA;
+		goto cleanup;
+	}
+
+	xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
+	size = le64_to_cpu(xs->here->xe_value_size);
+	if (buffer) {
+		ret = -ERANGE;
+		if (size > buffer_size)
+			goto cleanup;
+
+		name_offset = le16_to_cpu(xs->here->xe_name_offset);
+		name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
+		i = xs->here - xs->header->xh_entries;
+
+		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
+			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
+								bucket_xh(xs->bucket),
+								i,
+								&block_off,
+								&name_offset);
+			if (ret) {
+				mlog_errno(ret);
+				goto cleanup;
+			}
+			xs->base = bucket_block(xs->bucket, block_off);
+		}
+		if (ocfs2_xattr_is_local(xs->here)) {
+			memcpy(buffer, (void *)xs->base +
+			       name_offset + name_len, size);
+		} else {
+			xv = (struct ocfs2_xattr_value_root *)
+				(xs->base + name_offset + name_len);
+			ret = ocfs2_xattr_get_value_outside(inode, xv,
+							    buffer, size);
+			if (ret < 0) {
+				mlog_errno(ret);
+				goto cleanup;
+			}
+		}
+	}
+	ret = size;
+cleanup:
+	ocfs2_xattr_bucket_free(xs->bucket);
+
+	brelse(xs->xattr_bh);
+	xs->xattr_bh = NULL;
+	return ret;
+}
+
+int ocfs2_xattr_get_nolock(struct inode *inode,
+			   struct buffer_head *di_bh,
+			   int name_index,
+			   const char *name,
+			   void *buffer,
+			   size_t buffer_size)
+{
+	int ret;
+	struct ocfs2_dinode *di = NULL;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_xattr_search xis = {
+		.not_found = -ENODATA,
+	};
+	struct ocfs2_xattr_search xbs = {
+		.not_found = -ENODATA,
+	};
+
+	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
+		return -EOPNOTSUPP;
+
+	if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
+		return -ENODATA;
+
+	xis.inode_bh = xbs.inode_bh = di_bh;
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
+				    buffer_size, &xis);
+	if (ret == -ENODATA && di->i_xattr_loc)
+		ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
+					    buffer_size, &xbs);
+
+	return ret;
+}
+
+/* ocfs2_xattr_get()
+ *
+ * Copy an extended attribute into the buffer provided.
+ * Buffer is NULL to compute the size of buffer required.
+ */
+static int ocfs2_xattr_get(struct inode *inode,
+			   int name_index,
+			   const char *name,
+			   void *buffer,
+			   size_t buffer_size)
+{
+	int ret;
+	struct buffer_head *di_bh = NULL;
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 0);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+	down_read(&OCFS2_I(inode)->ip_xattr_sem);
+	ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
+				     name, buffer, buffer_size);
+	up_read(&OCFS2_I(inode)->ip_xattr_sem);
+
+	ocfs2_inode_unlock(inode, 0);
+
+	brelse(di_bh);
+
+	return ret;
+}
+
+static int __ocfs2_xattr_set_value_outside(struct inode *inode,
+					   handle_t *handle,
+					   struct ocfs2_xattr_value_buf *vb,
+					   const void *value,
+					   int value_len)
+{
+	int ret = 0, i, cp_len;
+	u16 blocksize = inode->i_sb->s_blocksize;
+	u32 p_cluster, num_clusters;
+	u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
+	u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
+	u64 blkno;
+	struct buffer_head *bh = NULL;
+	unsigned int ext_flags;
+	struct ocfs2_xattr_value_root *xv = vb->vb_xv;
+
+	BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
+
+	while (cpos < clusters) {
+		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
+					       &num_clusters, &xv->xr_list,
+					       &ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
+
+		blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
+
+		for (i = 0; i < num_clusters * bpc; i++, blkno++) {
+			ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
+					       &bh, NULL);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ret = ocfs2_journal_access(handle,
+						   INODE_CACHE(inode),
+						   bh,
+						   OCFS2_JOURNAL_ACCESS_WRITE);
+			if (ret < 0) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			cp_len = value_len > blocksize ? blocksize : value_len;
+			memcpy(bh->b_data, value, cp_len);
+			value_len -= cp_len;
+			value += cp_len;
+			if (cp_len < blocksize)
+				memset(bh->b_data + cp_len, 0,
+				       blocksize - cp_len);
+
+			ocfs2_journal_dirty(handle, bh);
+			brelse(bh);
+			bh = NULL;
+
+			/*
+			 * XXX: do we need to empty all the following
+			 * blocks in this cluster?
+			 */
+			if (!value_len)
+				break;
+		}
+		cpos += num_clusters;
+	}
+out:
+	brelse(bh);
+
+	return ret;
+}
+
+static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
+				       int num_entries)
+{
+	int free_space;
+
+	if (!needed_space)
+		return 0;
+
+	free_space = free_start -
+		sizeof(struct ocfs2_xattr_header) -
+		(num_entries * sizeof(struct ocfs2_xattr_entry)) -
+		OCFS2_XATTR_HEADER_GAP;
+	if (free_space < 0)
+		return -EIO;
+	if (free_space < needed_space)
+		return -ENOSPC;
+
+	return 0;
+}
+
+static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
+				   int type)
+{
+	return loc->xl_ops->xlo_journal_access(handle, loc, type);
+}
+
+static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
+{
+	loc->xl_ops->xlo_journal_dirty(handle, loc);
+}
+
+/* Give a pointer into the storage for the given offset */
+static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
+{
+	BUG_ON(offset >= loc->xl_size);
+	return loc->xl_ops->xlo_offset_pointer(loc, offset);
+}
+
+/*
+ * Wipe the name+value pair and allow the storage to reclaim it.  This
+ * must be followed by either removal of the entry or a call to
+ * ocfs2_xa_add_namevalue().
+ */
+static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
+{
+	loc->xl_ops->xlo_wipe_namevalue(loc);
+}
+
+/*
+ * Find lowest offset to a name+value pair.  This is the start of our
+ * downward-growing free space.
+ */
+static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
+{
+	return loc->xl_ops->xlo_get_free_start(loc);
+}
+
+/* Can we reuse loc->xl_entry for xi? */
+static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
+				    struct ocfs2_xattr_info *xi)
+{
+	return loc->xl_ops->xlo_can_reuse(loc, xi);
+}
+
+/* How much free space is needed to set the new value */
+static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
+				struct ocfs2_xattr_info *xi)
+{
+	return loc->xl_ops->xlo_check_space(loc, xi);
+}
+
+static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
+{
+	loc->xl_ops->xlo_add_entry(loc, name_hash);
+	loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
+	/*
+	 * We can't leave the new entry's xe_name_offset at zero or
+	 * add_namevalue() will go nuts.  We set it to the size of our
+	 * storage so that it can never be less than any other entry.
+	 */
+	loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
+}
+
+static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
+				   struct ocfs2_xattr_info *xi)
+{
+	int size = namevalue_size_xi(xi);
+	int nameval_offset;
+	char *nameval_buf;
+
+	loc->xl_ops->xlo_add_namevalue(loc, size);
+	loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
+	loc->xl_entry->xe_name_len = xi->xi_name_len;
+	ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
+	ocfs2_xattr_set_local(loc->xl_entry,
+			      xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
+
+	nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
+	nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
+	memset(nameval_buf, 0, size);
+	memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
+}
+
+static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
+				    struct ocfs2_xattr_value_buf *vb)
+{
+	int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
+	int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
+
+	/* Value bufs are for value trees */
+	BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
+	BUG_ON(namevalue_size_xe(loc->xl_entry) !=
+	       (name_size + OCFS2_XATTR_ROOT_SIZE));
+
+	loc->xl_ops->xlo_fill_value_buf(loc, vb);
+	vb->vb_xv =
+		(struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
+							nameval_offset +
+							name_size);
+}
+
+static int ocfs2_xa_block_journal_access(handle_t *handle,
+					 struct ocfs2_xa_loc *loc, int type)
+{
+	struct buffer_head *bh = loc->xl_storage;
+	ocfs2_journal_access_func access;
+
+	if (loc->xl_size == (bh->b_size -
+			     offsetof(struct ocfs2_xattr_block,
+				      xb_attrs.xb_header)))
+		access = ocfs2_journal_access_xb;
+	else
+		access = ocfs2_journal_access_di;
+	return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
+}
+
+static void ocfs2_xa_block_journal_dirty(handle_t *handle,
+					 struct ocfs2_xa_loc *loc)
+{
+	struct buffer_head *bh = loc->xl_storage;
+
+	ocfs2_journal_dirty(handle, bh);
+}
+
+static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
+					   int offset)
+{
+	return (char *)loc->xl_header + offset;
+}
+
+static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
+				    struct ocfs2_xattr_info *xi)
+{
+	/*
+	 * Block storage is strict.  If the sizes aren't exact, we will
+	 * remove the old one and reinsert the new.
+	 */
+	return namevalue_size_xe(loc->xl_entry) ==
+		namevalue_size_xi(xi);
+}
+
+static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
+{
+	struct ocfs2_xattr_header *xh = loc->xl_header;
+	int i, count = le16_to_cpu(xh->xh_count);
+	int offset, free_start = loc->xl_size;
+
+	for (i = 0; i < count; i++) {
+		offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
+		if (offset < free_start)
+			free_start = offset;
+	}
+
+	return free_start;
+}
+
+static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
+				      struct ocfs2_xattr_info *xi)
+{
+	int count = le16_to_cpu(loc->xl_header->xh_count);
+	int free_start = ocfs2_xa_get_free_start(loc);
+	int needed_space = ocfs2_xi_entry_usage(xi);
+
+	/*
+	 * Block storage will reclaim the original entry before inserting
+	 * the new value, so we only need the difference.  If the new
+	 * entry is smaller than the old one, we don't need anything.
+	 */
+	if (loc->xl_entry) {
+		/* Don't need space if we're reusing! */
+		if (ocfs2_xa_can_reuse_entry(loc, xi))
+			needed_space = 0;
+		else
+			needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
+	}
+	if (needed_space < 0)
+		needed_space = 0;
+	return ocfs2_xa_check_space_helper(needed_space, free_start, count);
+}
+
+/*
+ * Block storage for xattrs keeps the name+value pairs compacted.  When
+ * we remove one, we have to shift any that preceded it towards the end.
+ */
+static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
+{
+	int i, offset;
+	int namevalue_offset, first_namevalue_offset, namevalue_size;
+	struct ocfs2_xattr_entry *entry = loc->xl_entry;
+	struct ocfs2_xattr_header *xh = loc->xl_header;
+	int count = le16_to_cpu(xh->xh_count);
+
+	namevalue_offset = le16_to_cpu(entry->xe_name_offset);
+	namevalue_size = namevalue_size_xe(entry);
+	first_namevalue_offset = ocfs2_xa_get_free_start(loc);
+
+	/* Shift the name+value pairs */
+	memmove((char *)xh + first_namevalue_offset + namevalue_size,
+		(char *)xh + first_namevalue_offset,
+		namevalue_offset - first_namevalue_offset);
+	memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
+
+	/* Now tell xh->xh_entries about it */
+	for (i = 0; i < count; i++) {
+		offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
+		if (offset <= namevalue_offset)
+			le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
+				     namevalue_size);
+	}
+
+	/*
+	 * Note that we don't update xh_free_start or xh_name_value_len
+	 * because they're not used in block-stored xattrs.
+	 */
+}
+
+static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
+{
+	int count = le16_to_cpu(loc->xl_header->xh_count);
+	loc->xl_entry = &(loc->xl_header->xh_entries[count]);
+	le16_add_cpu(&loc->xl_header->xh_count, 1);
+	memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
+}
+
+static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
+{
+	int free_start = ocfs2_xa_get_free_start(loc);
+
+	loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
+}
+
+static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
+					  struct ocfs2_xattr_value_buf *vb)
+{
+	struct buffer_head *bh = loc->xl_storage;
+
+	if (loc->xl_size == (bh->b_size -
+			     offsetof(struct ocfs2_xattr_block,
+				      xb_attrs.xb_header)))
+		vb->vb_access = ocfs2_journal_access_xb;
+	else
+		vb->vb_access = ocfs2_journal_access_di;
+	vb->vb_bh = bh;
+}
+
+/*
+ * Operations for xattrs stored in blocks.  This includes inline inode
+ * storage and unindexed ocfs2_xattr_blocks.
+ */
+static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
+	.xlo_journal_access	= ocfs2_xa_block_journal_access,
+	.xlo_journal_dirty	= ocfs2_xa_block_journal_dirty,
+	.xlo_offset_pointer	= ocfs2_xa_block_offset_pointer,
+	.xlo_check_space	= ocfs2_xa_block_check_space,
+	.xlo_can_reuse		= ocfs2_xa_block_can_reuse,
+	.xlo_get_free_start	= ocfs2_xa_block_get_free_start,
+	.xlo_wipe_namevalue	= ocfs2_xa_block_wipe_namevalue,
+	.xlo_add_entry		= ocfs2_xa_block_add_entry,
+	.xlo_add_namevalue	= ocfs2_xa_block_add_namevalue,
+	.xlo_fill_value_buf	= ocfs2_xa_block_fill_value_buf,
+};
+
+static int ocfs2_xa_bucket_journal_access(handle_t *handle,
+					  struct ocfs2_xa_loc *loc, int type)
+{
+	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
+
+	return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
+}
+
+static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
+					  struct ocfs2_xa_loc *loc)
+{
+	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
+
+	ocfs2_xattr_bucket_journal_dirty(handle, bucket);
+}
+
+static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
+					    int offset)
+{
+	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
+	int block, block_offset;
+
+	/* The header is at the front of the bucket */
+	block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
+	block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
+
+	return bucket_block(bucket, block) + block_offset;
+}
+
+static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
+				     struct ocfs2_xattr_info *xi)
+{
+	return namevalue_size_xe(loc->xl_entry) >=
+		namevalue_size_xi(xi);
+}
+
+static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
+{
+	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
+	return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
+}
+
+static int ocfs2_bucket_align_free_start(struct super_block *sb,
+					 int free_start, int size)
+{
+	/*
+	 * We need to make sure that the name+value pair fits within
+	 * one block.
+	 */
+	if (((free_start - size) >> sb->s_blocksize_bits) !=
+	    ((free_start - 1) >> sb->s_blocksize_bits))
+		free_start -= free_start % sb->s_blocksize;
+
+	return free_start;
+}
+
+static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
+				       struct ocfs2_xattr_info *xi)
+{
+	int rc;
+	int count = le16_to_cpu(loc->xl_header->xh_count);
+	int free_start = ocfs2_xa_get_free_start(loc);
+	int needed_space = ocfs2_xi_entry_usage(xi);
+	int size = namevalue_size_xi(xi);
+	struct super_block *sb = loc->xl_inode->i_sb;
+
+	/*
+	 * Bucket storage does not reclaim name+value pairs it cannot
+	 * reuse.  They live as holes until the bucket fills, and then
+	 * the bucket is defragmented.  However, the bucket can reclaim
+	 * the ocfs2_xattr_entry.
+	 */
+	if (loc->xl_entry) {
+		/* Don't need space if we're reusing! */
+		if (ocfs2_xa_can_reuse_entry(loc, xi))
+			needed_space = 0;
+		else
+			needed_space -= sizeof(struct ocfs2_xattr_entry);
+	}
+	BUG_ON(needed_space < 0);
+
+	if (free_start < size) {
+		if (needed_space)
+			return -ENOSPC;
+	} else {
+		/*
+		 * First we check if it would fit in the first place.
+		 * Below, we align the free start to a block.  This may
+		 * slide us below the minimum gap.  By checking unaligned
+		 * first, we avoid that error.
+		 */
+		rc = ocfs2_xa_check_space_helper(needed_space, free_start,
+						 count);
+		if (rc)
+			return rc;
+		free_start = ocfs2_bucket_align_free_start(sb, free_start,
+							   size);
+	}
+	return ocfs2_xa_check_space_helper(needed_space, free_start, count);
+}
+
+static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
+{
+	le16_add_cpu(&loc->xl_header->xh_name_value_len,
+		     -namevalue_size_xe(loc->xl_entry));
+}
+
+static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
+{
+	struct ocfs2_xattr_header *xh = loc->xl_header;
+	int count = le16_to_cpu(xh->xh_count);
+	int low = 0, high = count - 1, tmp;
+	struct ocfs2_xattr_entry *tmp_xe;
+
+	/*
+	 * We keep buckets sorted by name_hash, so we need to find
+	 * our insert place.
+	 */
+	while (low <= high && count) {
+		tmp = (low + high) / 2;
+		tmp_xe = &xh->xh_entries[tmp];
+
+		if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
+			low = tmp + 1;
+		else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
+			high = tmp - 1;
+		else {
+			low = tmp;
+			break;
+		}
+	}
+
+	if (low != count)
+		memmove(&xh->xh_entries[low + 1],
+			&xh->xh_entries[low],
+			((count - low) * sizeof(struct ocfs2_xattr_entry)));
+
+	le16_add_cpu(&xh->xh_count, 1);
+	loc->xl_entry = &xh->xh_entries[low];
+	memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
+}
+
+static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
+{
+	int free_start = ocfs2_xa_get_free_start(loc);
+	struct ocfs2_xattr_header *xh = loc->xl_header;
+	struct super_block *sb = loc->xl_inode->i_sb;
+	int nameval_offset;
+
+	free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
+	nameval_offset = free_start - size;
+	loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
+	xh->xh_free_start = cpu_to_le16(nameval_offset);
+	le16_add_cpu(&xh->xh_name_value_len, size);
+
+}
+
+static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
+					   struct ocfs2_xattr_value_buf *vb)
+{
+	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
+	struct super_block *sb = loc->xl_inode->i_sb;
+	int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
+	int size = namevalue_size_xe(loc->xl_entry);
+	int block_offset = nameval_offset >> sb->s_blocksize_bits;
+
+	/* Values are not allowed to straddle block boundaries */
+	BUG_ON(block_offset !=
+	       ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
+	/* We expect the bucket to be filled in */
+	BUG_ON(!bucket->bu_bhs[block_offset]);
+
+	vb->vb_access = ocfs2_journal_access;
+	vb->vb_bh = bucket->bu_bhs[block_offset];
+}
+
+/* Operations for xattrs stored in buckets. */
+static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
+	.xlo_journal_access	= ocfs2_xa_bucket_journal_access,
+	.xlo_journal_dirty	= ocfs2_xa_bucket_journal_dirty,
+	.xlo_offset_pointer	= ocfs2_xa_bucket_offset_pointer,
+	.xlo_check_space	= ocfs2_xa_bucket_check_space,
+	.xlo_can_reuse		= ocfs2_xa_bucket_can_reuse,
+	.xlo_get_free_start	= ocfs2_xa_bucket_get_free_start,
+	.xlo_wipe_namevalue	= ocfs2_xa_bucket_wipe_namevalue,
+	.xlo_add_entry		= ocfs2_xa_bucket_add_entry,
+	.xlo_add_namevalue	= ocfs2_xa_bucket_add_namevalue,
+	.xlo_fill_value_buf	= ocfs2_xa_bucket_fill_value_buf,
+};
+
+static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
+{
+	struct ocfs2_xattr_value_buf vb;
+
+	if (ocfs2_xattr_is_local(loc->xl_entry))
+		return 0;
+
+	ocfs2_xa_fill_value_buf(loc, &vb);
+	return le32_to_cpu(vb.vb_xv->xr_clusters);
+}
+
+static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
+				   struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int trunc_rc, access_rc;
+	struct ocfs2_xattr_value_buf vb;
+
+	ocfs2_xa_fill_value_buf(loc, &vb);
+	trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
+					      ctxt);
+
+	/*
+	 * The caller of ocfs2_xa_value_truncate() has already called
+	 * ocfs2_xa_journal_access on the loc.  However, The truncate code
+	 * calls ocfs2_extend_trans().  This may commit the previous
+	 * transaction and open a new one.  If this is a bucket, truncate
+	 * could leave only vb->vb_bh set up for journaling.  Meanwhile,
+	 * the caller is expecting to dirty the entire bucket.  So we must
+	 * reset the journal work.  We do this even if truncate has failed,
+	 * as it could have failed after committing the extend.
+	 */
+	access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
+					    OCFS2_JOURNAL_ACCESS_WRITE);
+
+	/* Errors in truncate take precedence */
+	return trunc_rc ? trunc_rc : access_rc;
+}
+
+static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
+{
+	int index, count;
+	struct ocfs2_xattr_header *xh = loc->xl_header;
+	struct ocfs2_xattr_entry *entry = loc->xl_entry;
+
+	ocfs2_xa_wipe_namevalue(loc);
+	loc->xl_entry = NULL;
+
+	le16_add_cpu(&xh->xh_count, -1);
+	count = le16_to_cpu(xh->xh_count);
+
+	/*
+	 * Only zero out the entry if there are more remaining.  This is
+	 * important for an empty bucket, as it keeps track of the
+	 * bucket's hash value.  It doesn't hurt empty block storage.
+	 */
+	if (count) {
+		index = ((char *)entry - (char *)&xh->xh_entries) /
+			sizeof(struct ocfs2_xattr_entry);
+		memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
+			(count - index) * sizeof(struct ocfs2_xattr_entry));
+		memset(&xh->xh_entries[count], 0,
+		       sizeof(struct ocfs2_xattr_entry));
+	}
+}
+
+/*
+ * If we have a problem adjusting the size of an external value during
+ * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
+ * in an intermediate state.  For example, the value may be partially
+ * truncated.
+ *
+ * If the value tree hasn't changed, the extend/truncate went nowhere.
+ * We have nothing to do.  The caller can treat it as a straight error.
+ *
+ * If the value tree got partially truncated, we now have a corrupted
+ * extended attribute.  We're going to wipe its entry and leak the
+ * clusters.  Better to leak some storage than leave a corrupt entry.
+ *
+ * If the value tree grew, it obviously didn't grow enough for the
+ * new entry.  We're not going to try and reclaim those clusters either.
+ * If there was already an external value there (orig_clusters != 0),
+ * the new clusters are attached safely and we can just leave the old
+ * value in place.  If there was no external value there, we remove
+ * the entry.
+ *
+ * This way, the xattr block we store in the journal will be consistent.
+ * If the size change broke because of the journal, no changes will hit
+ * disk anyway.
+ */
+static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
+					    const char *what,
+					    unsigned int orig_clusters)
+{
+	unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
+	char *nameval_buf = ocfs2_xa_offset_pointer(loc,
+				le16_to_cpu(loc->xl_entry->xe_name_offset));
+
+	if (new_clusters < orig_clusters) {
+		mlog(ML_ERROR,
+		     "Partial truncate while %s xattr %.*s.  Leaking "
+		     "%u clusters and removing the entry\n",
+		     what, loc->xl_entry->xe_name_len, nameval_buf,
+		     orig_clusters - new_clusters);
+		ocfs2_xa_remove_entry(loc);
+	} else if (!orig_clusters) {
+		mlog(ML_ERROR,
+		     "Unable to allocate an external value for xattr "
+		     "%.*s safely.  Leaking %u clusters and removing the "
+		     "entry\n",
+		     loc->xl_entry->xe_name_len, nameval_buf,
+		     new_clusters - orig_clusters);
+		ocfs2_xa_remove_entry(loc);
+	} else if (new_clusters > orig_clusters)
+		mlog(ML_ERROR,
+		     "Unable to grow xattr %.*s safely.  %u new clusters "
+		     "have been added, but the value will not be "
+		     "modified\n",
+		     loc->xl_entry->xe_name_len, nameval_buf,
+		     new_clusters - orig_clusters);
+}
+
+static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
+			   struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int rc = 0;
+	unsigned int orig_clusters;
+
+	if (!ocfs2_xattr_is_local(loc->xl_entry)) {
+		orig_clusters = ocfs2_xa_value_clusters(loc);
+		rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
+		if (rc) {
+			mlog_errno(rc);
+			/*
+			 * Since this is remove, we can return 0 if
+			 * ocfs2_xa_cleanup_value_truncate() is going to
+			 * wipe the entry anyway.  So we check the
+			 * cluster count as well.
+			 */
+			if (orig_clusters != ocfs2_xa_value_clusters(loc))
+				rc = 0;
+			ocfs2_xa_cleanup_value_truncate(loc, "removing",
+							orig_clusters);
+			if (rc)
+				goto out;
+		}
+	}
+
+	ocfs2_xa_remove_entry(loc);
+
+out:
+	return rc;
+}
+
+static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
+{
+	int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
+	char *nameval_buf;
+
+	nameval_buf = ocfs2_xa_offset_pointer(loc,
+				le16_to_cpu(loc->xl_entry->xe_name_offset));
+	memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
+}
+
+/*
+ * Take an existing entry and make it ready for the new value.  This
+ * won't allocate space, but it may free space.  It should be ready for
+ * ocfs2_xa_prepare_entry() to finish the work.
+ */
+static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
+				struct ocfs2_xattr_info *xi,
+				struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int rc = 0;
+	int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
+	unsigned int orig_clusters;
+	char *nameval_buf;
+	int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
+	int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
+
+	BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
+	       name_size);
+
+	nameval_buf = ocfs2_xa_offset_pointer(loc,
+				le16_to_cpu(loc->xl_entry->xe_name_offset));
+	if (xe_local) {
+		memset(nameval_buf + name_size, 0,
+		       namevalue_size_xe(loc->xl_entry) - name_size);
+		if (!xi_local)
+			ocfs2_xa_install_value_root(loc);
+	} else {
+		orig_clusters = ocfs2_xa_value_clusters(loc);
+		if (xi_local) {
+			rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
+			if (rc < 0)
+				mlog_errno(rc);
+			else
+				memset(nameval_buf + name_size, 0,
+				       namevalue_size_xe(loc->xl_entry) -
+				       name_size);
+		} else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
+			   xi->xi_value_len) {
+			rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
+						     ctxt);
+			if (rc < 0)
+				mlog_errno(rc);
+		}
+
+		if (rc) {
+			ocfs2_xa_cleanup_value_truncate(loc, "reusing",
+							orig_clusters);
+			goto out;
+		}
+	}
+
+	loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
+	ocfs2_xattr_set_local(loc->xl_entry, xi_local);
+
+out:
+	return rc;
+}
+
+/*
+ * Prepares loc->xl_entry to receive the new xattr.  This includes
+ * properly setting up the name+value pair region.  If loc->xl_entry
+ * already exists, it will take care of modifying it appropriately.
+ *
+ * Note that this modifies the data.  You did journal_access already,
+ * right?
+ */
+static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
+				  struct ocfs2_xattr_info *xi,
+				  u32 name_hash,
+				  struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int rc = 0;
+	unsigned int orig_clusters;
+	__le64 orig_value_size = 0;
+
+	rc = ocfs2_xa_check_space(loc, xi);
+	if (rc)
+		goto out;
+
+	if (loc->xl_entry) {
+		if (ocfs2_xa_can_reuse_entry(loc, xi)) {
+			orig_value_size = loc->xl_entry->xe_value_size;
+			rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
+			if (rc)
+				goto out;
+			goto alloc_value;
+		}
+
+		if (!ocfs2_xattr_is_local(loc->xl_entry)) {
+			orig_clusters = ocfs2_xa_value_clusters(loc);
+			rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
+			if (rc) {
+				mlog_errno(rc);
+				ocfs2_xa_cleanup_value_truncate(loc,
+								"overwriting",
+								orig_clusters);
+				goto out;
+			}
+		}
+		ocfs2_xa_wipe_namevalue(loc);
+	} else
+		ocfs2_xa_add_entry(loc, name_hash);
+
+	/*
+	 * If we get here, we have a blank entry.  Fill it.  We grow our
+	 * name+value pair back from the end.
+	 */
+	ocfs2_xa_add_namevalue(loc, xi);
+	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
+		ocfs2_xa_install_value_root(loc);
+
+alloc_value:
+	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
+		orig_clusters = ocfs2_xa_value_clusters(loc);
+		rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
+		if (rc < 0) {
+			ctxt->set_abort = 1;
+			ocfs2_xa_cleanup_value_truncate(loc, "growing",
+							orig_clusters);
+			/*
+			 * If we were growing an existing value,
+			 * ocfs2_xa_cleanup_value_truncate() won't remove
+			 * the entry. We need to restore the original value
+			 * size.
+			 */
+			if (loc->xl_entry) {
+				BUG_ON(!orig_value_size);
+				loc->xl_entry->xe_value_size = orig_value_size;
+			}
+			mlog_errno(rc);
+		}
+	}
+
+out:
+	return rc;
+}
+
+/*
+ * Store the value portion of the name+value pair.  This will skip
+ * values that are stored externally.  Their tree roots were set up
+ * by ocfs2_xa_prepare_entry().
+ */
+static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
+				struct ocfs2_xattr_info *xi,
+				struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int rc = 0;
+	int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
+	int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
+	char *nameval_buf;
+	struct ocfs2_xattr_value_buf vb;
+
+	nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
+	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
+		ocfs2_xa_fill_value_buf(loc, &vb);
+		rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
+						     ctxt->handle, &vb,
+						     xi->xi_value,
+						     xi->xi_value_len);
+	} else
+		memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
+
+	return rc;
+}
+
+static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
+			struct ocfs2_xattr_info *xi,
+			struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret;
+	u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
+					      xi->xi_name_len);
+
+	ret = ocfs2_xa_journal_access(ctxt->handle, loc,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * From here on out, everything is going to modify the buffer a
+	 * little.  Errors are going to leave the xattr header in a
+	 * sane state.  Thus, even with errors we dirty the sucker.
+	 */
+
+	/* Don't worry, we are never called with !xi_value and !xl_entry */
+	if (!xi->xi_value) {
+		ret = ocfs2_xa_remove(loc, ctxt);
+		goto out_dirty;
+	}
+
+	ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
+	if (ret) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		goto out_dirty;
+	}
+
+	ret = ocfs2_xa_store_value(loc, xi, ctxt);
+	if (ret)
+		mlog_errno(ret);
+
+out_dirty:
+	ocfs2_xa_journal_dirty(ctxt->handle, loc);
+
+out:
+	return ret;
+}
+
+static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
+				     struct inode *inode,
+				     struct buffer_head *bh,
+				     struct ocfs2_xattr_entry *entry)
+{
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
+
+	BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
+
+	loc->xl_inode = inode;
+	loc->xl_ops = &ocfs2_xa_block_loc_ops;
+	loc->xl_storage = bh;
+	loc->xl_entry = entry;
+	loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
+	loc->xl_header =
+		(struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
+					      loc->xl_size);
+}
+
+static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
+					  struct inode *inode,
+					  struct buffer_head *bh,
+					  struct ocfs2_xattr_entry *entry)
+{
+	struct ocfs2_xattr_block *xb =
+		(struct ocfs2_xattr_block *)bh->b_data;
+
+	BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
+
+	loc->xl_inode = inode;
+	loc->xl_ops = &ocfs2_xa_block_loc_ops;
+	loc->xl_storage = bh;
+	loc->xl_header = &(xb->xb_attrs.xb_header);
+	loc->xl_entry = entry;
+	loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
+					     xb_attrs.xb_header);
+}
+
+static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
+					   struct ocfs2_xattr_bucket *bucket,
+					   struct ocfs2_xattr_entry *entry)
+{
+	loc->xl_inode = bucket->bu_inode;
+	loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
+	loc->xl_storage = bucket;
+	loc->xl_header = bucket_xh(bucket);
+	loc->xl_entry = entry;
+	loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
+}
+
+/*
+ * In xattr remove, if it is stored outside and refcounted, we may have
+ * the chance to split the refcount tree. So need the allocators.
+ */
+static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
+					struct ocfs2_xattr_value_root *xv,
+					struct ocfs2_caching_info *ref_ci,
+					struct buffer_head *ref_root_bh,
+					struct ocfs2_alloc_context **meta_ac,
+					int *ref_credits)
+{
+	int ret, meta_add = 0;
+	u32 p_cluster, num_clusters;
+	unsigned int ext_flags;
+
+	*ref_credits = 0;
+	ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
+				       &num_clusters,
+				       &xv->xr_list,
+				       &ext_flags);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
+		goto out;
+
+	ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
+						 ref_root_bh, xv,
+						 &meta_add, ref_credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
+						meta_add, meta_ac);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+static int ocfs2_remove_value_outside(struct inode*inode,
+				      struct ocfs2_xattr_value_buf *vb,
+				      struct ocfs2_xattr_header *header,
+				      struct ocfs2_caching_info *ref_ci,
+				      struct buffer_head *ref_root_bh)
+{
+	int ret = 0, i, ref_credits;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
+	void *val;
+
+	ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
+
+	for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
+		struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
+
+		if (ocfs2_xattr_is_local(entry))
+			continue;
+
+		val = (void *)header +
+			le16_to_cpu(entry->xe_name_offset);
+		vb->vb_xv = (struct ocfs2_xattr_value_root *)
+			(val + OCFS2_XATTR_SIZE(entry->xe_name_len));
+
+		ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
+							 ref_ci, ref_root_bh,
+							 &ctxt.meta_ac,
+							 &ref_credits);
+
+		ctxt.handle = ocfs2_start_trans(osb, ref_credits +
+					ocfs2_remove_extent_credits(osb->sb));
+		if (IS_ERR(ctxt.handle)) {
+			ret = PTR_ERR(ctxt.handle);
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
+
+		ocfs2_commit_trans(osb, ctxt.handle);
+		if (ctxt.meta_ac) {
+			ocfs2_free_alloc_context(ctxt.meta_ac);
+			ctxt.meta_ac = NULL;
+		}
+
+		if (ret < 0) {
+			mlog_errno(ret);
+			break;
+		}
+
+	}
+
+	if (ctxt.meta_ac)
+		ocfs2_free_alloc_context(ctxt.meta_ac);
+	ocfs2_schedule_truncate_log_flush(osb, 1);
+	ocfs2_run_deallocs(osb, &ctxt.dealloc);
+	return ret;
+}
+
+static int ocfs2_xattr_ibody_remove(struct inode *inode,
+				    struct buffer_head *di_bh,
+				    struct ocfs2_caching_info *ref_ci,
+				    struct buffer_head *ref_root_bh)
+{
+
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_xattr_header *header;
+	int ret;
+	struct ocfs2_xattr_value_buf vb = {
+		.vb_bh = di_bh,
+		.vb_access = ocfs2_journal_access_di,
+	};
+
+	header = (struct ocfs2_xattr_header *)
+		 ((void *)di + inode->i_sb->s_blocksize -
+		 le16_to_cpu(di->i_xattr_inline_size));
+
+	ret = ocfs2_remove_value_outside(inode, &vb, header,
+					 ref_ci, ref_root_bh);
+
+	return ret;
+}
+
+struct ocfs2_rm_xattr_bucket_para {
+	struct ocfs2_caching_info *ref_ci;
+	struct buffer_head *ref_root_bh;
+};
+
+static int ocfs2_xattr_block_remove(struct inode *inode,
+				    struct buffer_head *blk_bh,
+				    struct ocfs2_caching_info *ref_ci,
+				    struct buffer_head *ref_root_bh)
+{
+	struct ocfs2_xattr_block *xb;
+	int ret = 0;
+	struct ocfs2_xattr_value_buf vb = {
+		.vb_bh = blk_bh,
+		.vb_access = ocfs2_journal_access_xb,
+	};
+	struct ocfs2_rm_xattr_bucket_para args = {
+		.ref_ci = ref_ci,
+		.ref_root_bh = ref_root_bh,
+	};
+
+	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
+	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
+		struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
+		ret = ocfs2_remove_value_outside(inode, &vb, header,
+						 ref_ci, ref_root_bh);
+	} else
+		ret = ocfs2_iterate_xattr_index_block(inode,
+						blk_bh,
+						ocfs2_rm_xattr_cluster,
+						&args);
+
+	return ret;
+}
+
+static int ocfs2_xattr_free_block(struct inode *inode,
+				  u64 block,
+				  struct ocfs2_caching_info *ref_ci,
+				  struct buffer_head *ref_root_bh)
+{
+	struct inode *xb_alloc_inode;
+	struct buffer_head *xb_alloc_bh = NULL;
+	struct buffer_head *blk_bh = NULL;
+	struct ocfs2_xattr_block *xb;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	handle_t *handle;
+	int ret = 0;
+	u64 blk, bg_blkno;
+	u16 bit;
+
+	ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
+	blk = le64_to_cpu(xb->xb_blkno);
+	bit = le16_to_cpu(xb->xb_suballoc_bit);
+	if (xb->xb_suballoc_loc)
+		bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
+	else
+		bg_blkno = ocfs2_which_suballoc_group(blk, bit);
+
+	xb_alloc_inode = ocfs2_get_system_file_inode(osb,
+				EXTENT_ALLOC_SYSTEM_INODE,
+				le16_to_cpu(xb->xb_suballoc_slot));
+	if (!xb_alloc_inode) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+	mutex_lock(&xb_alloc_inode->i_mutex);
+
+	ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_mutex;
+	}
+
+	handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+
+	ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
+				       bit, bg_blkno, 1);
+	if (ret < 0)
+		mlog_errno(ret);
+
+	ocfs2_commit_trans(osb, handle);
+out_unlock:
+	ocfs2_inode_unlock(xb_alloc_inode, 1);
+	brelse(xb_alloc_bh);
+out_mutex:
+	mutex_unlock(&xb_alloc_inode->i_mutex);
+	iput(xb_alloc_inode);
+out:
+	brelse(blk_bh);
+	return ret;
+}
+
+/*
+ * ocfs2_xattr_remove()
+ *
+ * Free extended attribute resources associated with this inode.
+ */
+int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_refcount_tree *ref_tree = NULL;
+	struct buffer_head *ref_root_bh = NULL;
+	struct ocfs2_caching_info *ref_ci = NULL;
+	handle_t *handle;
+	int ret;
+
+	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
+		return 0;
+
+	if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
+		return 0;
+
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
+		ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
+					       le64_to_cpu(di->i_refcount_loc),
+					       1, &ref_tree, &ref_root_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		ref_ci = &ref_tree->rf_ci;
+
+	}
+
+	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
+		ret = ocfs2_xattr_ibody_remove(inode, di_bh,
+					       ref_ci, ref_root_bh);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (di->i_xattr_loc) {
+		ret = ocfs2_xattr_free_block(inode,
+					     le64_to_cpu(di->i_xattr_loc),
+					     ref_ci, ref_root_bh);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
+				   OCFS2_INODE_UPDATE_CREDITS);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	di->i_xattr_loc = 0;
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	spin_unlock(&oi->ip_lock);
+	ocfs2_update_inode_fsync_trans(handle, inode, 0);
+
+	ocfs2_journal_dirty(handle, di_bh);
+out_commit:
+	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
+out:
+	if (ref_tree)
+		ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
+	brelse(ref_root_bh);
+	return ret;
+}
+
+static int ocfs2_xattr_has_space_inline(struct inode *inode,
+					struct ocfs2_dinode *di)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
+	int free;
+
+	if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
+		return 0;
+
+	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		struct ocfs2_inline_data *idata = &di->id2.i_data;
+		free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
+	} else if (ocfs2_inode_is_fast_symlink(inode)) {
+		free = ocfs2_fast_symlink_chars(inode->i_sb) -
+			le64_to_cpu(di->i_size);
+	} else {
+		struct ocfs2_extent_list *el = &di->id2.i_list;
+		free = (le16_to_cpu(el->l_count) -
+			le16_to_cpu(el->l_next_free_rec)) *
+			sizeof(struct ocfs2_extent_rec);
+	}
+	if (free >= xattrsize)
+		return 1;
+
+	return 0;
+}
+
+/*
+ * ocfs2_xattr_ibody_find()
+ *
+ * Find extended attribute in inode block and
+ * fill search info into struct ocfs2_xattr_search.
+ */
+static int ocfs2_xattr_ibody_find(struct inode *inode,
+				  int name_index,
+				  const char *name,
+				  struct ocfs2_xattr_search *xs)
+{
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
+	int ret;
+	int has_space = 0;
+
+	if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
+		return 0;
+
+	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
+		down_read(&oi->ip_alloc_sem);
+		has_space = ocfs2_xattr_has_space_inline(inode, di);
+		up_read(&oi->ip_alloc_sem);
+		if (!has_space)
+			return 0;
+	}
+
+	xs->xattr_bh = xs->inode_bh;
+	xs->end = (void *)di + inode->i_sb->s_blocksize;
+	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
+		xs->header = (struct ocfs2_xattr_header *)
+			(xs->end - le16_to_cpu(di->i_xattr_inline_size));
+	else
+		xs->header = (struct ocfs2_xattr_header *)
+			(xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
+	xs->base = (void *)xs->header;
+	xs->here = xs->header->xh_entries;
+
+	/* Find the named attribute. */
+	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
+		ret = ocfs2_xattr_find_entry(name_index, name, xs);
+		if (ret && ret != -ENODATA)
+			return ret;
+		xs->not_found = ret;
+	}
+
+	return 0;
+}
+
+static int ocfs2_xattr_ibody_init(struct inode *inode,
+				  struct buffer_head *di_bh,
+				  struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	unsigned int xattrsize = osb->s_xattr_inline_size;
+
+	if (!ocfs2_xattr_has_space_inline(inode, di)) {
+		ret = -ENOSPC;
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Adjust extent record count or inline data size
+	 * to reserve space for extended attribute.
+	 */
+	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
+		struct ocfs2_inline_data *idata = &di->id2.i_data;
+		le16_add_cpu(&idata->id_count, -xattrsize);
+	} else if (!(ocfs2_inode_is_fast_symlink(inode))) {
+		struct ocfs2_extent_list *el = &di->id2.i_list;
+		le16_add_cpu(&el->l_count, -(xattrsize /
+					     sizeof(struct ocfs2_extent_rec)));
+	}
+	di->i_xattr_inline_size = cpu_to_le16(xattrsize);
+
+	spin_lock(&oi->ip_lock);
+	oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
+	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
+	spin_unlock(&oi->ip_lock);
+
+	ocfs2_journal_dirty(ctxt->handle, di_bh);
+
+out:
+	return ret;
+}
+
+/*
+ * ocfs2_xattr_ibody_set()
+ *
+ * Set, replace or remove an extended attribute into inode block.
+ *
+ */
+static int ocfs2_xattr_ibody_set(struct inode *inode,
+				 struct ocfs2_xattr_info *xi,
+				 struct ocfs2_xattr_search *xs,
+				 struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_xa_loc loc;
+
+	if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
+		return -ENOSPC;
+
+	down_write(&oi->ip_alloc_sem);
+	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
+		ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
+		if (ret) {
+			if (ret != -ENOSPC)
+				mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
+				 xs->not_found ? NULL : xs->here);
+	ret = ocfs2_xa_set(&loc, xi, ctxt);
+	if (ret) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		goto out;
+	}
+	xs->here = loc.xl_entry;
+
+out:
+	up_write(&oi->ip_alloc_sem);
+
+	return ret;
+}
+
+/*
+ * ocfs2_xattr_block_find()
+ *
+ * Find extended attribute in external block and
+ * fill search info into struct ocfs2_xattr_search.
+ */
+static int ocfs2_xattr_block_find(struct inode *inode,
+				  int name_index,
+				  const char *name,
+				  struct ocfs2_xattr_search *xs)
+{
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
+	struct buffer_head *blk_bh = NULL;
+	struct ocfs2_xattr_block *xb;
+	int ret = 0;
+
+	if (!di->i_xattr_loc)
+		return ret;
+
+	ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
+				     &blk_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	xs->xattr_bh = blk_bh;
+	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
+
+	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
+		xs->header = &xb->xb_attrs.xb_header;
+		xs->base = (void *)xs->header;
+		xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
+		xs->here = xs->header->xh_entries;
+
+		ret = ocfs2_xattr_find_entry(name_index, name, xs);
+	} else
+		ret = ocfs2_xattr_index_block_find(inode, blk_bh,
+						   name_index,
+						   name, xs);
+
+	if (ret && ret != -ENODATA) {
+		xs->xattr_bh = NULL;
+		goto cleanup;
+	}
+	xs->not_found = ret;
+	return 0;
+cleanup:
+	brelse(blk_bh);
+
+	return ret;
+}
+
+static int ocfs2_create_xattr_block(struct inode *inode,
+				    struct buffer_head *inode_bh,
+				    struct ocfs2_xattr_set_ctxt *ctxt,
+				    int indexed,
+				    struct buffer_head **ret_bh)
+{
+	int ret;
+	u16 suballoc_bit_start;
+	u32 num_got;
+	u64 suballoc_loc, first_blkno;
+	struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_xattr_block *xblk;
+
+	ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
+				      inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto end;
+	}
+
+	ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
+				   &suballoc_loc, &suballoc_bit_start,
+				   &num_got, &first_blkno);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto end;
+	}
+
+	new_bh = sb_getblk(inode->i_sb, first_blkno);
+	if (!new_bh) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto end;
+	}
+
+	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
+
+	ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
+				      new_bh,
+				      OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto end;
+	}
+
+	/* Initialize ocfs2_xattr_block */
+	xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
+	memset(xblk, 0, inode->i_sb->s_blocksize);
+	strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
+	xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
+	xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
+	xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
+	xblk->xb_fs_generation =
+		cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
+	xblk->xb_blkno = cpu_to_le64(first_blkno);
+	if (indexed) {
+		struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
+		xr->xt_clusters = cpu_to_le32(1);
+		xr->xt_last_eb_blk = 0;
+		xr->xt_list.l_tree_depth = 0;
+		xr->xt_list.l_count = cpu_to_le16(
+					ocfs2_xattr_recs_per_xb(inode->i_sb));
+		xr->xt_list.l_next_free_rec = cpu_to_le16(1);
+		xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
+	}
+	ocfs2_journal_dirty(ctxt->handle, new_bh);
+
+	/* Add it to the inode */
+	di->i_xattr_loc = cpu_to_le64(first_blkno);
+
+	spin_lock(&OCFS2_I(inode)->ip_lock);
+	OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
+	di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
+	spin_unlock(&OCFS2_I(inode)->ip_lock);
+
+	ocfs2_journal_dirty(ctxt->handle, inode_bh);
+
+	*ret_bh = new_bh;
+	new_bh = NULL;
+
+end:
+	brelse(new_bh);
+	return ret;
+}
+
+/*
+ * ocfs2_xattr_block_set()
+ *
+ * Set, replace or remove an extended attribute into external block.
+ *
+ */
+static int ocfs2_xattr_block_set(struct inode *inode,
+				 struct ocfs2_xattr_info *xi,
+				 struct ocfs2_xattr_search *xs,
+				 struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	struct buffer_head *new_bh = NULL;
+	struct ocfs2_xattr_block *xblk = NULL;
+	int ret;
+	struct ocfs2_xa_loc loc;
+
+	if (!xs->xattr_bh) {
+		ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
+					       0, &new_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto end;
+		}
+
+		xs->xattr_bh = new_bh;
+		xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
+		xs->header = &xblk->xb_attrs.xb_header;
+		xs->base = (void *)xs->header;
+		xs->end = (void *)xblk + inode->i_sb->s_blocksize;
+		xs->here = xs->header->xh_entries;
+	} else
+		xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
+
+	if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
+		ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
+					      xs->not_found ? NULL : xs->here);
+
+		ret = ocfs2_xa_set(&loc, xi, ctxt);
+		if (!ret)
+			xs->here = loc.xl_entry;
+		else if ((ret != -ENOSPC) || ctxt->set_abort)
+			goto end;
+		else {
+			ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
+			if (ret)
+				goto end;
+		}
+	}
+
+	if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
+		ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
+
+end:
+	return ret;
+}
+
+/* Check whether the new xattr can be inserted into the inode. */
+static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
+				       struct ocfs2_xattr_info *xi,
+				       struct ocfs2_xattr_search *xs)
+{
+	struct ocfs2_xattr_entry *last;
+	int free, i;
+	size_t min_offs = xs->end - xs->base;
+
+	if (!xs->header)
+		return 0;
+
+	last = xs->header->xh_entries;
+
+	for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
+		size_t offs = le16_to_cpu(last->xe_name_offset);
+		if (offs < min_offs)
+			min_offs = offs;
+		last += 1;
+	}
+
+	free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
+	if (free < 0)
+		return 0;
+
+	BUG_ON(!xs->not_found);
+
+	if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
+		return 1;
+
+	return 0;
+}
+
+static int ocfs2_calc_xattr_set_need(struct inode *inode,
+				     struct ocfs2_dinode *di,
+				     struct ocfs2_xattr_info *xi,
+				     struct ocfs2_xattr_search *xis,
+				     struct ocfs2_xattr_search *xbs,
+				     int *clusters_need,
+				     int *meta_need,
+				     int *credits_need)
+{
+	int ret = 0, old_in_xb = 0;
+	int clusters_add = 0, meta_add = 0, credits = 0;
+	struct buffer_head *bh = NULL;
+	struct ocfs2_xattr_block *xb = NULL;
+	struct ocfs2_xattr_entry *xe = NULL;
+	struct ocfs2_xattr_value_root *xv = NULL;
+	char *base = NULL;
+	int name_offset, name_len = 0;
+	u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
+						    xi->xi_value_len);
+	u64 value_size;
+
+	/*
+	 * Calculate the clusters we need to write.
+	 * No matter whether we replace an old one or add a new one,
+	 * we need this for writing.
+	 */
+	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
+		credits += new_clusters *
+			   ocfs2_clusters_to_blocks(inode->i_sb, 1);
+
+	if (xis->not_found && xbs->not_found) {
+		credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+
+		if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
+			clusters_add += new_clusters;
+			credits += ocfs2_calc_extend_credits(inode->i_sb,
+							&def_xv.xv.xr_list);
+		}
+
+		goto meta_guess;
+	}
+
+	if (!xis->not_found) {
+		xe = xis->here;
+		name_offset = le16_to_cpu(xe->xe_name_offset);
+		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
+		base = xis->base;
+		credits += OCFS2_INODE_UPDATE_CREDITS;
+	} else {
+		int i, block_off = 0;
+		xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
+		xe = xbs->here;
+		name_offset = le16_to_cpu(xe->xe_name_offset);
+		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
+		i = xbs->here - xbs->header->xh_entries;
+		old_in_xb = 1;
+
+		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
+			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
+							bucket_xh(xbs->bucket),
+							i, &block_off,
+							&name_offset);
+			base = bucket_block(xbs->bucket, block_off);
+			credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+		} else {
+			base = xbs->base;
+			credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
+		}
+	}
+
+	/*
+	 * delete a xattr doesn't need metadata and cluster allocation.
+	 * so just calculate the credits and return.
+	 *
+	 * The credits for removing the value tree will be extended
+	 * by ocfs2_remove_extent itself.
+	 */
+	if (!xi->xi_value) {
+		if (!ocfs2_xattr_is_local(xe))
+			credits += ocfs2_remove_extent_credits(inode->i_sb);
+
+		goto out;
+	}
+
+	/* do cluster allocation guess first. */
+	value_size = le64_to_cpu(xe->xe_value_size);
+
+	if (old_in_xb) {
+		/*
+		 * In xattr set, we always try to set the xe in inode first,
+		 * so if it can be inserted into inode successfully, the old
+		 * one will be removed from the xattr block, and this xattr
+		 * will be inserted into inode as a new xattr in inode.
+		 */
+		if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
+			clusters_add += new_clusters;
+			credits += ocfs2_remove_extent_credits(inode->i_sb) +
+				    OCFS2_INODE_UPDATE_CREDITS;
+			if (!ocfs2_xattr_is_local(xe))
+				credits += ocfs2_calc_extend_credits(
+							inode->i_sb,
+							&def_xv.xv.xr_list);
+			goto out;
+		}
+	}
+
+	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
+		/* the new values will be stored outside. */
+		u32 old_clusters = 0;
+
+		if (!ocfs2_xattr_is_local(xe)) {
+			old_clusters =	ocfs2_clusters_for_bytes(inode->i_sb,
+								 value_size);
+			xv = (struct ocfs2_xattr_value_root *)
+			     (base + name_offset + name_len);
+			value_size = OCFS2_XATTR_ROOT_SIZE;
+		} else
+			xv = &def_xv.xv;
+
+		if (old_clusters >= new_clusters) {
+			credits += ocfs2_remove_extent_credits(inode->i_sb);
+			goto out;
+		} else {
+			meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
+			clusters_add += new_clusters - old_clusters;
+			credits += ocfs2_calc_extend_credits(inode->i_sb,
+							     &xv->xr_list);
+			if (value_size >= OCFS2_XATTR_ROOT_SIZE)
+				goto out;
+		}
+	} else {
+		/*
+		 * Now the new value will be stored inside. So if the new
+		 * value is smaller than the size of value root or the old
+		 * value, we don't need any allocation, otherwise we have
+		 * to guess metadata allocation.
+		 */
+		if ((ocfs2_xattr_is_local(xe) &&
+		     (value_size >= xi->xi_value_len)) ||
+		    (!ocfs2_xattr_is_local(xe) &&
+		     OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
+			goto out;
+	}
+
+meta_guess:
+	/* calculate metadata allocation. */
+	if (di->i_xattr_loc) {
+		if (!xbs->xattr_bh) {
+			ret = ocfs2_read_xattr_block(inode,
+						     le64_to_cpu(di->i_xattr_loc),
+						     &bh);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			xb = (struct ocfs2_xattr_block *)bh->b_data;
+		} else
+			xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
+
+		/*
+		 * If there is already an xattr tree, good, we can calculate
+		 * like other b-trees. Otherwise we may have the chance of
+		 * create a tree, the credit calculation is borrowed from
+		 * ocfs2_calc_extend_credits with root_el = NULL. And the
+		 * new tree will be cluster based, so no meta is needed.
+		 */
+		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
+			struct ocfs2_extent_list *el =
+				 &xb->xb_attrs.xb_root.xt_list;
+			meta_add += ocfs2_extend_meta_needed(el);
+			credits += ocfs2_calc_extend_credits(inode->i_sb,
+							     el);
+		} else
+			credits += OCFS2_SUBALLOC_ALLOC + 1;
+
+		/*
+		 * This cluster will be used either for new bucket or for
+		 * new xattr block.
+		 * If the cluster size is the same as the bucket size, one
+		 * more is needed since we may need to extend the bucket
+		 * also.
+		 */
+		clusters_add += 1;
+		credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+		if (OCFS2_XATTR_BUCKET_SIZE ==
+			OCFS2_SB(inode->i_sb)->s_clustersize) {
+			credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+			clusters_add += 1;
+		}
+	} else {
+		credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
+		if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
+			struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
+			meta_add += ocfs2_extend_meta_needed(el);
+			credits += ocfs2_calc_extend_credits(inode->i_sb,
+							     el);
+		} else {
+			meta_add += 1;
+		}
+	}
+out:
+	if (clusters_need)
+		*clusters_need = clusters_add;
+	if (meta_need)
+		*meta_need = meta_add;
+	if (credits_need)
+		*credits_need = credits;
+	brelse(bh);
+	return ret;
+}
+
+static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
+				     struct ocfs2_dinode *di,
+				     struct ocfs2_xattr_info *xi,
+				     struct ocfs2_xattr_search *xis,
+				     struct ocfs2_xattr_search *xbs,
+				     struct ocfs2_xattr_set_ctxt *ctxt,
+				     int extra_meta,
+				     int *credits)
+{
+	int clusters_add, meta_add, ret;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+
+	memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
+
+	ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
+
+	ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
+					&clusters_add, &meta_add, credits);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	meta_add += extra_meta;
+	trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
+					clusters_add, *credits);
+
+	if (meta_add) {
+		ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
+							&ctxt->meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (clusters_add) {
+		ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
+		if (ret)
+			mlog_errno(ret);
+	}
+out:
+	if (ret) {
+		if (ctxt->meta_ac) {
+			ocfs2_free_alloc_context(ctxt->meta_ac);
+			ctxt->meta_ac = NULL;
+		}
+
+		/*
+		 * We cannot have an error and a non null ctxt->data_ac.
+		 */
+	}
+
+	return ret;
+}
+
+static int __ocfs2_xattr_set_handle(struct inode *inode,
+				    struct ocfs2_dinode *di,
+				    struct ocfs2_xattr_info *xi,
+				    struct ocfs2_xattr_search *xis,
+				    struct ocfs2_xattr_search *xbs,
+				    struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret = 0, credits, old_found;
+
+	if (!xi->xi_value) {
+		/* Remove existing extended attribute */
+		if (!xis->not_found)
+			ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
+		else if (!xbs->not_found)
+			ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
+	} else {
+		/* We always try to set extended attribute into inode first*/
+		ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
+		if (!ret && !xbs->not_found) {
+			/*
+			 * If succeed and that extended attribute existing in
+			 * external block, then we will remove it.
+			 */
+			xi->xi_value = NULL;
+			xi->xi_value_len = 0;
+
+			old_found = xis->not_found;
+			xis->not_found = -ENODATA;
+			ret = ocfs2_calc_xattr_set_need(inode,
+							di,
+							xi,
+							xis,
+							xbs,
+							NULL,
+							NULL,
+							&credits);
+			xis->not_found = old_found;
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			ret = ocfs2_extend_trans(ctxt->handle, credits);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+			ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
+		} else if ((ret == -ENOSPC) && !ctxt->set_abort) {
+			if (di->i_xattr_loc && !xbs->xattr_bh) {
+				ret = ocfs2_xattr_block_find(inode,
+							     xi->xi_name_index,
+							     xi->xi_name, xbs);
+				if (ret)
+					goto out;
+
+				old_found = xis->not_found;
+				xis->not_found = -ENODATA;
+				ret = ocfs2_calc_xattr_set_need(inode,
+								di,
+								xi,
+								xis,
+								xbs,
+								NULL,
+								NULL,
+								&credits);
+				xis->not_found = old_found;
+				if (ret) {
+					mlog_errno(ret);
+					goto out;
+				}
+
+				ret = ocfs2_extend_trans(ctxt->handle, credits);
+				if (ret) {
+					mlog_errno(ret);
+					goto out;
+				}
+			}
+			/*
+			 * If no space in inode, we will set extended attribute
+			 * into external block.
+			 */
+			ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
+			if (ret)
+				goto out;
+			if (!xis->not_found) {
+				/*
+				 * If succeed and that extended attribute
+				 * existing in inode, we will remove it.
+				 */
+				xi->xi_value = NULL;
+				xi->xi_value_len = 0;
+				xbs->not_found = -ENODATA;
+				ret = ocfs2_calc_xattr_set_need(inode,
+								di,
+								xi,
+								xis,
+								xbs,
+								NULL,
+								NULL,
+								&credits);
+				if (ret) {
+					mlog_errno(ret);
+					goto out;
+				}
+
+				ret = ocfs2_extend_trans(ctxt->handle, credits);
+				if (ret) {
+					mlog_errno(ret);
+					goto out;
+				}
+				ret = ocfs2_xattr_ibody_set(inode, xi,
+							    xis, ctxt);
+			}
+		}
+	}
+
+	if (!ret) {
+		/* Update inode ctime. */
+		ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
+					      xis->inode_bh,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		inode->i_ctime = CURRENT_TIME;
+		di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
+		di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
+		ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
+	}
+out:
+	return ret;
+}
+
+/*
+ * This function only called duing creating inode
+ * for init security/acl xattrs of the new inode.
+ * All transanction credits have been reserved in mknod.
+ */
+int ocfs2_xattr_set_handle(handle_t *handle,
+			   struct inode *inode,
+			   struct buffer_head *di_bh,
+			   int name_index,
+			   const char *name,
+			   const void *value,
+			   size_t value_len,
+			   int flags,
+			   struct ocfs2_alloc_context *meta_ac,
+			   struct ocfs2_alloc_context *data_ac)
+{
+	struct ocfs2_dinode *di;
+	int ret;
+
+	struct ocfs2_xattr_info xi = {
+		.xi_name_index = name_index,
+		.xi_name = name,
+		.xi_name_len = strlen(name),
+		.xi_value = value,
+		.xi_value_len = value_len,
+	};
+
+	struct ocfs2_xattr_search xis = {
+		.not_found = -ENODATA,
+	};
+
+	struct ocfs2_xattr_search xbs = {
+		.not_found = -ENODATA,
+	};
+
+	struct ocfs2_xattr_set_ctxt ctxt = {
+		.handle = handle,
+		.meta_ac = meta_ac,
+		.data_ac = data_ac,
+	};
+
+	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
+		return -EOPNOTSUPP;
+
+	/*
+	 * In extreme situation, may need xattr bucket when
+	 * block size is too small. And we have already reserved
+	 * the credits for bucket in mknod.
+	 */
+	if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
+		xbs.bucket = ocfs2_xattr_bucket_new(inode);
+		if (!xbs.bucket) {
+			mlog_errno(-ENOMEM);
+			return -ENOMEM;
+		}
+	}
+
+	xis.inode_bh = xbs.inode_bh = di_bh;
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	down_write(&OCFS2_I(inode)->ip_xattr_sem);
+
+	ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
+	if (ret)
+		goto cleanup;
+	if (xis.not_found) {
+		ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
+		if (ret)
+			goto cleanup;
+	}
+
+	ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
+
+cleanup:
+	up_write(&OCFS2_I(inode)->ip_xattr_sem);
+	brelse(xbs.xattr_bh);
+	ocfs2_xattr_bucket_free(xbs.bucket);
+
+	return ret;
+}
+
+/*
+ * ocfs2_xattr_set()
+ *
+ * Set, replace or remove an extended attribute for this inode.
+ * value is NULL to remove an existing extended attribute, else either
+ * create or replace an extended attribute.
+ */
+int ocfs2_xattr_set(struct inode *inode,
+		    int name_index,
+		    const char *name,
+		    const void *value,
+		    size_t value_len,
+		    int flags)
+{
+	struct buffer_head *di_bh = NULL;
+	struct ocfs2_dinode *di;
+	int ret, credits, ref_meta = 0, ref_credits = 0;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct inode *tl_inode = osb->osb_tl_inode;
+	struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
+	struct ocfs2_refcount_tree *ref_tree = NULL;
+
+	struct ocfs2_xattr_info xi = {
+		.xi_name_index = name_index,
+		.xi_name = name,
+		.xi_name_len = strlen(name),
+		.xi_value = value,
+		.xi_value_len = value_len,
+	};
+
+	struct ocfs2_xattr_search xis = {
+		.not_found = -ENODATA,
+	};
+
+	struct ocfs2_xattr_search xbs = {
+		.not_found = -ENODATA,
+	};
+
+	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
+		return -EOPNOTSUPP;
+
+	/*
+	 * Only xbs will be used on indexed trees.  xis doesn't need a
+	 * bucket.
+	 */
+	xbs.bucket = ocfs2_xattr_bucket_new(inode);
+	if (!xbs.bucket) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
+	ret = ocfs2_inode_lock(inode, &di_bh, 1);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto cleanup_nolock;
+	}
+	xis.inode_bh = xbs.inode_bh = di_bh;
+	di = (struct ocfs2_dinode *)di_bh->b_data;
+
+	down_write(&OCFS2_I(inode)->ip_xattr_sem);
+	/*
+	 * Scan inode and external block to find the same name
+	 * extended attribute and collect search information.
+	 */
+	ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
+	if (ret)
+		goto cleanup;
+	if (xis.not_found) {
+		ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
+		if (ret)
+			goto cleanup;
+	}
+
+	if (xis.not_found && xbs.not_found) {
+		ret = -ENODATA;
+		if (flags & XATTR_REPLACE)
+			goto cleanup;
+		ret = 0;
+		if (!value)
+			goto cleanup;
+	} else {
+		ret = -EEXIST;
+		if (flags & XATTR_CREATE)
+			goto cleanup;
+	}
+
+	/* Check whether the value is refcounted and do some preparation. */
+	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
+	    (!xis.not_found || !xbs.not_found)) {
+		ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
+						   &xis, &xbs, &ref_tree,
+						   &ref_meta, &ref_credits);
+		if (ret) {
+			mlog_errno(ret);
+			goto cleanup;
+		}
+	}
+
+	mutex_lock(&tl_inode->i_mutex);
+
+	if (ocfs2_truncate_log_needs_flush(osb)) {
+		ret = __ocfs2_flush_truncate_log(osb);
+		if (ret < 0) {
+			mutex_unlock(&tl_inode->i_mutex);
+			mlog_errno(ret);
+			goto cleanup;
+		}
+	}
+	mutex_unlock(&tl_inode->i_mutex);
+
+	ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
+					&xbs, &ctxt, ref_meta, &credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto cleanup;
+	}
+
+	/* we need to update inode's ctime field, so add credit for it. */
+	credits += OCFS2_INODE_UPDATE_CREDITS;
+	ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
+	if (IS_ERR(ctxt.handle)) {
+		ret = PTR_ERR(ctxt.handle);
+		mlog_errno(ret);
+		goto out_free_ac;
+	}
+
+	ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
+	ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
+
+	ocfs2_commit_trans(osb, ctxt.handle);
+
+out_free_ac:
+	if (ctxt.data_ac)
+		ocfs2_free_alloc_context(ctxt.data_ac);
+	if (ctxt.meta_ac)
+		ocfs2_free_alloc_context(ctxt.meta_ac);
+	if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
+		ocfs2_schedule_truncate_log_flush(osb, 1);
+	ocfs2_run_deallocs(osb, &ctxt.dealloc);
+
+cleanup:
+	if (ref_tree)
+		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
+	up_write(&OCFS2_I(inode)->ip_xattr_sem);
+	if (!value && !ret) {
+		ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
+		if (ret)
+			mlog_errno(ret);
+	}
+	ocfs2_inode_unlock(inode, 1);
+cleanup_nolock:
+	brelse(di_bh);
+	brelse(xbs.xattr_bh);
+	ocfs2_xattr_bucket_free(xbs.bucket);
+
+	return ret;
+}
+
+/*
+ * Find the xattr extent rec which may contains name_hash.
+ * e_cpos will be the first name hash of the xattr rec.
+ * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
+ */
+static int ocfs2_xattr_get_rec(struct inode *inode,
+			       u32 name_hash,
+			       u64 *p_blkno,
+			       u32 *e_cpos,
+			       u32 *num_clusters,
+			       struct ocfs2_extent_list *el)
+{
+	int ret = 0, i;
+	struct buffer_head *eb_bh = NULL;
+	struct ocfs2_extent_block *eb;
+	struct ocfs2_extent_rec *rec = NULL;
+	u64 e_blkno = 0;
+
+	if (el->l_tree_depth) {
+		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
+				      &eb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
+		el = &eb->h_list;
+
+		if (el->l_tree_depth) {
+			ocfs2_error(inode->i_sb,
+				    "Inode %lu has non zero tree depth in "
+				    "xattr tree block %llu\n", inode->i_ino,
+				    (unsigned long long)eb_bh->b_blocknr);
+			ret = -EROFS;
+			goto out;
+		}
+	}
+
+	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
+		rec = &el->l_recs[i];
+
+		if (le32_to_cpu(rec->e_cpos) <= name_hash) {
+			e_blkno = le64_to_cpu(rec->e_blkno);
+			break;
+		}
+	}
+
+	if (!e_blkno) {
+		ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
+			    "record (%u, %u, 0) in xattr", inode->i_ino,
+			    le32_to_cpu(rec->e_cpos),
+			    ocfs2_rec_clusters(el, rec));
+		ret = -EROFS;
+		goto out;
+	}
+
+	*p_blkno = le64_to_cpu(rec->e_blkno);
+	*num_clusters = le16_to_cpu(rec->e_leaf_clusters);
+	if (e_cpos)
+		*e_cpos = le32_to_cpu(rec->e_cpos);
+out:
+	brelse(eb_bh);
+	return ret;
+}
+
+typedef int (xattr_bucket_func)(struct inode *inode,
+				struct ocfs2_xattr_bucket *bucket,
+				void *para);
+
+static int ocfs2_find_xe_in_bucket(struct inode *inode,
+				   struct ocfs2_xattr_bucket *bucket,
+				   int name_index,
+				   const char *name,
+				   u32 name_hash,
+				   u16 *xe_index,
+				   int *found)
+{
+	int i, ret = 0, cmp = 1, block_off, new_offset;
+	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
+	size_t name_len = strlen(name);
+	struct ocfs2_xattr_entry *xe = NULL;
+	char *xe_name;
+
+	/*
+	 * We don't use binary search in the bucket because there
+	 * may be multiple entries with the same name hash.
+	 */
+	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
+		xe = &xh->xh_entries[i];
+
+		if (name_hash > le32_to_cpu(xe->xe_name_hash))
+			continue;
+		else if (name_hash < le32_to_cpu(xe->xe_name_hash))
+			break;
+
+		cmp = name_index - ocfs2_xattr_get_type(xe);
+		if (!cmp)
+			cmp = name_len - xe->xe_name_len;
+		if (cmp)
+			continue;
+
+		ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
+							xh,
+							i,
+							&block_off,
+							&new_offset);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+
+		xe_name = bucket_block(bucket, block_off) + new_offset;
+		if (!memcmp(name, xe_name, name_len)) {
+			*xe_index = i;
+			*found = 1;
+			ret = 0;
+			break;
+		}
+	}
+
+	return ret;
+}
+
+/*
+ * Find the specified xattr entry in a series of buckets.
+ * This series start from p_blkno and last for num_clusters.
+ * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
+ * the num of the valid buckets.
+ *
+ * Return the buffer_head this xattr should reside in. And if the xattr's
+ * hash is in the gap of 2 buckets, return the lower bucket.
+ */
+static int ocfs2_xattr_bucket_find(struct inode *inode,
+				   int name_index,
+				   const char *name,
+				   u32 name_hash,
+				   u64 p_blkno,
+				   u32 first_hash,
+				   u32 num_clusters,
+				   struct ocfs2_xattr_search *xs)
+{
+	int ret, found = 0;
+	struct ocfs2_xattr_header *xh = NULL;
+	struct ocfs2_xattr_entry *xe = NULL;
+	u16 index = 0;
+	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+	int low_bucket = 0, bucket, high_bucket;
+	struct ocfs2_xattr_bucket *search;
+	u32 last_hash;
+	u64 blkno, lower_blkno = 0;
+
+	search = ocfs2_xattr_bucket_new(inode);
+	if (!search) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_xattr_bucket(search, p_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	xh = bucket_xh(search);
+	high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
+	while (low_bucket <= high_bucket) {
+		ocfs2_xattr_bucket_relse(search);
+
+		bucket = (low_bucket + high_bucket) / 2;
+		blkno = p_blkno + bucket * blk_per_bucket;
+		ret = ocfs2_read_xattr_bucket(search, blkno);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		xh = bucket_xh(search);
+		xe = &xh->xh_entries[0];
+		if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
+			high_bucket = bucket - 1;
+			continue;
+		}
+
+		/*
+		 * Check whether the hash of the last entry in our
+		 * bucket is larger than the search one. for an empty
+		 * bucket, the last one is also the first one.
+		 */
+		if (xh->xh_count)
+			xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
+
+		last_hash = le32_to_cpu(xe->xe_name_hash);
+
+		/* record lower_blkno which may be the insert place. */
+		lower_blkno = blkno;
+
+		if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
+			low_bucket = bucket + 1;
+			continue;
+		}
+
+		/* the searched xattr should reside in this bucket if exists. */
+		ret = ocfs2_find_xe_in_bucket(inode, search,
+					      name_index, name, name_hash,
+					      &index, &found);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		break;
+	}
+
+	/*
+	 * Record the bucket we have found.
+	 * When the xattr's hash value is in the gap of 2 buckets, we will
+	 * always set it to the previous bucket.
+	 */
+	if (!lower_blkno)
+		lower_blkno = p_blkno;
+
+	/* This should be in cache - we just read it during the search */
+	ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	xs->header = bucket_xh(xs->bucket);
+	xs->base = bucket_block(xs->bucket, 0);
+	xs->end = xs->base + inode->i_sb->s_blocksize;
+
+	if (found) {
+		xs->here = &xs->header->xh_entries[index];
+		trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
+			name, name_index, name_hash,
+			(unsigned long long)bucket_blkno(xs->bucket),
+			index);
+	} else
+		ret = -ENODATA;
+
+out:
+	ocfs2_xattr_bucket_free(search);
+	return ret;
+}
+
+static int ocfs2_xattr_index_block_find(struct inode *inode,
+					struct buffer_head *root_bh,
+					int name_index,
+					const char *name,
+					struct ocfs2_xattr_search *xs)
+{
+	int ret;
+	struct ocfs2_xattr_block *xb =
+			(struct ocfs2_xattr_block *)root_bh->b_data;
+	struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
+	struct ocfs2_extent_list *el = &xb_root->xt_list;
+	u64 p_blkno = 0;
+	u32 first_hash, num_clusters = 0;
+	u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
+
+	if (le16_to_cpu(el->l_next_free_rec) == 0)
+		return -ENODATA;
+
+	trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
+					name, name_index, name_hash,
+					(unsigned long long)root_bh->b_blocknr,
+					-1);
+
+	ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
+				  &num_clusters, el);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
+
+	trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
+					name, name_index, first_hash,
+					(unsigned long long)p_blkno,
+					num_clusters);
+
+	ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
+				      p_blkno, first_hash, num_clusters, xs);
+
+out:
+	return ret;
+}
+
+static int ocfs2_iterate_xattr_buckets(struct inode *inode,
+				       u64 blkno,
+				       u32 clusters,
+				       xattr_bucket_func *func,
+				       void *para)
+{
+	int i, ret = 0;
+	u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
+	u32 num_buckets = clusters * bpc;
+	struct ocfs2_xattr_bucket *bucket;
+
+	bucket = ocfs2_xattr_bucket_new(inode);
+	if (!bucket) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
+	trace_ocfs2_iterate_xattr_buckets(
+		(unsigned long long)OCFS2_I(inode)->ip_blkno,
+		(unsigned long long)blkno, clusters);
+
+	for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
+		ret = ocfs2_read_xattr_bucket(bucket, blkno);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		/*
+		 * The real bucket num in this series of blocks is stored
+		 * in the 1st bucket.
+		 */
+		if (i == 0)
+			num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
+
+		trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
+		     le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
+		if (func) {
+			ret = func(inode, bucket, para);
+			if (ret && ret != -ERANGE)
+				mlog_errno(ret);
+			/* Fall through to bucket_relse() */
+		}
+
+		ocfs2_xattr_bucket_relse(bucket);
+		if (ret)
+			break;
+	}
+
+	ocfs2_xattr_bucket_free(bucket);
+	return ret;
+}
+
+struct ocfs2_xattr_tree_list {
+	char *buffer;
+	size_t buffer_size;
+	size_t result;
+};
+
+static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
+					     struct ocfs2_xattr_header *xh,
+					     int index,
+					     int *block_off,
+					     int *new_offset)
+{
+	u16 name_offset;
+
+	if (index < 0 || index >= le16_to_cpu(xh->xh_count))
+		return -EINVAL;
+
+	name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
+
+	*block_off = name_offset >> sb->s_blocksize_bits;
+	*new_offset = name_offset % sb->s_blocksize;
+
+	return 0;
+}
+
+static int ocfs2_list_xattr_bucket(struct inode *inode,
+				   struct ocfs2_xattr_bucket *bucket,
+				   void *para)
+{
+	int ret = 0, type;
+	struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
+	int i, block_off, new_offset;
+	const char *prefix, *name;
+
+	for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
+		struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
+		type = ocfs2_xattr_get_type(entry);
+		prefix = ocfs2_xattr_prefix(type);
+
+		if (prefix) {
+			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
+								bucket_xh(bucket),
+								i,
+								&block_off,
+								&new_offset);
+			if (ret)
+				break;
+
+			name = (const char *)bucket_block(bucket, block_off) +
+				new_offset;
+			ret = ocfs2_xattr_list_entry(xl->buffer,
+						     xl->buffer_size,
+						     &xl->result,
+						     prefix, name,
+						     entry->xe_name_len);
+			if (ret)
+				break;
+		}
+	}
+
+	return ret;
+}
+
+static int ocfs2_iterate_xattr_index_block(struct inode *inode,
+					   struct buffer_head *blk_bh,
+					   xattr_tree_rec_func *rec_func,
+					   void *para)
+{
+	struct ocfs2_xattr_block *xb =
+			(struct ocfs2_xattr_block *)blk_bh->b_data;
+	struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
+	int ret = 0;
+	u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
+	u64 p_blkno = 0;
+
+	if (!el->l_next_free_rec || !rec_func)
+		return 0;
+
+	while (name_hash > 0) {
+		ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
+					  &e_cpos, &num_clusters, el);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
+			       num_clusters, para);
+		if (ret) {
+			if (ret != -ERANGE)
+				mlog_errno(ret);
+			break;
+		}
+
+		if (e_cpos == 0)
+			break;
+
+		name_hash = e_cpos - 1;
+	}
+
+	return ret;
+
+}
+
+static int ocfs2_list_xattr_tree_rec(struct inode *inode,
+				     struct buffer_head *root_bh,
+				     u64 blkno, u32 cpos, u32 len, void *para)
+{
+	return ocfs2_iterate_xattr_buckets(inode, blkno, len,
+					   ocfs2_list_xattr_bucket, para);
+}
+
+static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
+					     struct buffer_head *blk_bh,
+					     char *buffer,
+					     size_t buffer_size)
+{
+	int ret;
+	struct ocfs2_xattr_tree_list xl = {
+		.buffer = buffer,
+		.buffer_size = buffer_size,
+		.result = 0,
+	};
+
+	ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
+					      ocfs2_list_xattr_tree_rec, &xl);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = xl.result;
+out:
+	return ret;
+}
+
+static int cmp_xe(const void *a, const void *b)
+{
+	const struct ocfs2_xattr_entry *l = a, *r = b;
+	u32 l_hash = le32_to_cpu(l->xe_name_hash);
+	u32 r_hash = le32_to_cpu(r->xe_name_hash);
+
+	if (l_hash > r_hash)
+		return 1;
+	if (l_hash < r_hash)
+		return -1;
+	return 0;
+}
+
+static void swap_xe(void *a, void *b, int size)
+{
+	struct ocfs2_xattr_entry *l = a, *r = b, tmp;
+
+	tmp = *l;
+	memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
+	memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
+}
+
+/*
+ * When the ocfs2_xattr_block is filled up, new bucket will be created
+ * and all the xattr entries will be moved to the new bucket.
+ * The header goes at the start of the bucket, and the names+values are
+ * filled from the end.  This is why *target starts as the last buffer.
+ * Note: we need to sort the entries since they are not saved in order
+ * in the ocfs2_xattr_block.
+ */
+static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
+					   struct buffer_head *xb_bh,
+					   struct ocfs2_xattr_bucket *bucket)
+{
+	int i, blocksize = inode->i_sb->s_blocksize;
+	int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+	u16 offset, size, off_change;
+	struct ocfs2_xattr_entry *xe;
+	struct ocfs2_xattr_block *xb =
+				(struct ocfs2_xattr_block *)xb_bh->b_data;
+	struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
+	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
+	u16 count = le16_to_cpu(xb_xh->xh_count);
+	char *src = xb_bh->b_data;
+	char *target = bucket_block(bucket, blks - 1);
+
+	trace_ocfs2_cp_xattr_block_to_bucket_begin(
+				(unsigned long long)xb_bh->b_blocknr,
+				(unsigned long long)bucket_blkno(bucket));
+
+	for (i = 0; i < blks; i++)
+		memset(bucket_block(bucket, i), 0, blocksize);
+
+	/*
+	 * Since the xe_name_offset is based on ocfs2_xattr_header,
+	 * there is a offset change corresponding to the change of
+	 * ocfs2_xattr_header's position.
+	 */
+	off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
+	xe = &xb_xh->xh_entries[count - 1];
+	offset = le16_to_cpu(xe->xe_name_offset) + off_change;
+	size = blocksize - offset;
+
+	/* copy all the names and values. */
+	memcpy(target + offset, src + offset, size);
+
+	/* Init new header now. */
+	xh->xh_count = xb_xh->xh_count;
+	xh->xh_num_buckets = cpu_to_le16(1);
+	xh->xh_name_value_len = cpu_to_le16(size);
+	xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
+
+	/* copy all the entries. */
+	target = bucket_block(bucket, 0);
+	offset = offsetof(struct ocfs2_xattr_header, xh_entries);
+	size = count * sizeof(struct ocfs2_xattr_entry);
+	memcpy(target + offset, (char *)xb_xh + offset, size);
+
+	/* Change the xe offset for all the xe because of the move. */
+	off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
+		 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
+	for (i = 0; i < count; i++)
+		le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
+
+	trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
+
+	sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
+	     cmp_xe, swap_xe);
+}
+
+/*
+ * After we move xattr from block to index btree, we have to
+ * update ocfs2_xattr_search to the new xe and base.
+ *
+ * When the entry is in xattr block, xattr_bh indicates the storage place.
+ * While if the entry is in index b-tree, "bucket" indicates the
+ * real place of the xattr.
+ */
+static void ocfs2_xattr_update_xattr_search(struct inode *inode,
+					    struct ocfs2_xattr_search *xs,
+					    struct buffer_head *old_bh)
+{
+	char *buf = old_bh->b_data;
+	struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
+	struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
+	int i;
+
+	xs->header = bucket_xh(xs->bucket);
+	xs->base = bucket_block(xs->bucket, 0);
+	xs->end = xs->base + inode->i_sb->s_blocksize;
+
+	if (xs->not_found)
+		return;
+
+	i = xs->here - old_xh->xh_entries;
+	xs->here = &xs->header->xh_entries[i];
+}
+
+static int ocfs2_xattr_create_index_block(struct inode *inode,
+					  struct ocfs2_xattr_search *xs,
+					  struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret;
+	u32 bit_off, len;
+	u64 blkno;
+	handle_t *handle = ctxt->handle;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct buffer_head *xb_bh = xs->xattr_bh;
+	struct ocfs2_xattr_block *xb =
+			(struct ocfs2_xattr_block *)xb_bh->b_data;
+	struct ocfs2_xattr_tree_root *xr;
+	u16 xb_flags = le16_to_cpu(xb->xb_flags);
+
+	trace_ocfs2_xattr_create_index_block_begin(
+				(unsigned long long)xb_bh->b_blocknr);
+
+	BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
+	BUG_ON(!xs->bucket);
+
+	/*
+	 * XXX:
+	 * We can use this lock for now, and maybe move to a dedicated mutex
+	 * if performance becomes a problem later.
+	 */
+	down_write(&oi->ip_alloc_sem);
+
+	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
+				     1, 1, &bit_off, &len);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * The bucket may spread in many blocks, and
+	 * we will only touch the 1st block and the last block
+	 * in the whole bucket(one for entry and one for data).
+	 */
+	blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
+
+	trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
+
+	ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
+						OCFS2_JOURNAL_ACCESS_CREATE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
+	ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
+
+	ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
+
+	/* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
+	memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
+	       offsetof(struct ocfs2_xattr_block, xb_attrs));
+
+	xr = &xb->xb_attrs.xb_root;
+	xr->xt_clusters = cpu_to_le32(1);
+	xr->xt_last_eb_blk = 0;
+	xr->xt_list.l_tree_depth = 0;
+	xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
+	xr->xt_list.l_next_free_rec = cpu_to_le16(1);
+
+	xr->xt_list.l_recs[0].e_cpos = 0;
+	xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
+	xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
+
+	xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
+
+	ocfs2_journal_dirty(handle, xb_bh);
+
+out:
+	up_write(&oi->ip_alloc_sem);
+
+	return ret;
+}
+
+static int cmp_xe_offset(const void *a, const void *b)
+{
+	const struct ocfs2_xattr_entry *l = a, *r = b;
+	u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
+	u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
+
+	if (l_name_offset < r_name_offset)
+		return 1;
+	if (l_name_offset > r_name_offset)
+		return -1;
+	return 0;
+}
+
+/*
+ * defrag a xattr bucket if we find that the bucket has some
+ * holes beteen name/value pairs.
+ * We will move all the name/value pairs to the end of the bucket
+ * so that we can spare some space for insertion.
+ */
+static int ocfs2_defrag_xattr_bucket(struct inode *inode,
+				     handle_t *handle,
+				     struct ocfs2_xattr_bucket *bucket)
+{
+	int ret, i;
+	size_t end, offset, len;
+	struct ocfs2_xattr_header *xh;
+	char *entries, *buf, *bucket_buf = NULL;
+	u64 blkno = bucket_blkno(bucket);
+	u16 xh_free_start;
+	size_t blocksize = inode->i_sb->s_blocksize;
+	struct ocfs2_xattr_entry *xe;
+
+	/*
+	 * In order to make the operation more efficient and generic,
+	 * we copy all the blocks into a contiguous memory and do the
+	 * defragment there, so if anything is error, we will not touch
+	 * the real block.
+	 */
+	bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
+	if (!bucket_buf) {
+		ret = -EIO;
+		goto out;
+	}
+
+	buf = bucket_buf;
+	for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
+		memcpy(buf, bucket_block(bucket, i), blocksize);
+
+	ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
+						OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	xh = (struct ocfs2_xattr_header *)bucket_buf;
+	entries = (char *)xh->xh_entries;
+	xh_free_start = le16_to_cpu(xh->xh_free_start);
+
+	trace_ocfs2_defrag_xattr_bucket(
+	     (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
+	     xh_free_start, le16_to_cpu(xh->xh_name_value_len));
+
+	/*
+	 * sort all the entries by their offset.
+	 * the largest will be the first, so that we can
+	 * move them to the end one by one.
+	 */
+	sort(entries, le16_to_cpu(xh->xh_count),
+	     sizeof(struct ocfs2_xattr_entry),
+	     cmp_xe_offset, swap_xe);
+
+	/* Move all name/values to the end of the bucket. */
+	xe = xh->xh_entries;
+	end = OCFS2_XATTR_BUCKET_SIZE;
+	for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
+		offset = le16_to_cpu(xe->xe_name_offset);
+		len = namevalue_size_xe(xe);
+
+		/*
+		 * We must make sure that the name/value pair
+		 * exist in the same block. So adjust end to
+		 * the previous block end if needed.
+		 */
+		if (((end - len) / blocksize !=
+			(end - 1) / blocksize))
+			end = end - end % blocksize;
+
+		if (end > offset + len) {
+			memmove(bucket_buf + end - len,
+				bucket_buf + offset, len);
+			xe->xe_name_offset = cpu_to_le16(end - len);
+		}
+
+		mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
+				"bucket %llu\n", (unsigned long long)blkno);
+
+		end -= len;
+	}
+
+	mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
+			"bucket %llu\n", (unsigned long long)blkno);
+
+	if (xh_free_start == end)
+		goto out;
+
+	memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
+	xh->xh_free_start = cpu_to_le16(end);
+
+	/* sort the entries by their name_hash. */
+	sort(entries, le16_to_cpu(xh->xh_count),
+	     sizeof(struct ocfs2_xattr_entry),
+	     cmp_xe, swap_xe);
+
+	buf = bucket_buf;
+	for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
+		memcpy(bucket_block(bucket, i), buf, blocksize);
+	ocfs2_xattr_bucket_journal_dirty(handle, bucket);
+
+out:
+	kfree(bucket_buf);
+	return ret;
+}
+
+/*
+ * prev_blkno points to the start of an existing extent.  new_blkno
+ * points to a newly allocated extent.  Because we know each of our
+ * clusters contains more than bucket, we can easily split one cluster
+ * at a bucket boundary.  So we take the last cluster of the existing
+ * extent and split it down the middle.  We move the last half of the
+ * buckets in the last cluster of the existing extent over to the new
+ * extent.
+ *
+ * first_bh is the buffer at prev_blkno so we can update the existing
+ * extent's bucket count.  header_bh is the bucket were we were hoping
+ * to insert our xattr.  If the bucket move places the target in the new
+ * extent, we'll update first_bh and header_bh after modifying the old
+ * extent.
+ *
+ * first_hash will be set as the 1st xe's name_hash in the new extent.
+ */
+static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
+					       handle_t *handle,
+					       struct ocfs2_xattr_bucket *first,
+					       struct ocfs2_xattr_bucket *target,
+					       u64 new_blkno,
+					       u32 num_clusters,
+					       u32 *first_hash)
+{
+	int ret;
+	struct super_block *sb = inode->i_sb;
+	int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
+	int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
+	int to_move = num_buckets / 2;
+	u64 src_blkno;
+	u64 last_cluster_blkno = bucket_blkno(first) +
+		((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
+
+	BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
+	BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
+
+	trace_ocfs2_mv_xattr_bucket_cross_cluster(
+				(unsigned long long)last_cluster_blkno,
+				(unsigned long long)new_blkno);
+
+	ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
+				     last_cluster_blkno, new_blkno,
+				     to_move, first_hash);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* This is the first bucket that got moved */
+	src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
+
+	/*
+	 * If the target bucket was part of the moved buckets, we need to
+	 * update first and target.
+	 */
+	if (bucket_blkno(target) >= src_blkno) {
+		/* Find the block for the new target bucket */
+		src_blkno = new_blkno +
+			(bucket_blkno(target) - src_blkno);
+
+		ocfs2_xattr_bucket_relse(first);
+		ocfs2_xattr_bucket_relse(target);
+
+		/*
+		 * These shouldn't fail - the buffers are in the
+		 * journal from ocfs2_cp_xattr_bucket().
+		 */
+		ret = ocfs2_read_xattr_bucket(first, new_blkno);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+		ret = ocfs2_read_xattr_bucket(target, src_blkno);
+		if (ret)
+			mlog_errno(ret);
+
+	}
+
+out:
+	return ret;
+}
+
+/*
+ * Find the suitable pos when we divide a bucket into 2.
+ * We have to make sure the xattrs with the same hash value exist
+ * in the same bucket.
+ *
+ * If this ocfs2_xattr_header covers more than one hash value, find a
+ * place where the hash value changes.  Try to find the most even split.
+ * The most common case is that all entries have different hash values,
+ * and the first check we make will find a place to split.
+ */
+static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
+{
+	struct ocfs2_xattr_entry *entries = xh->xh_entries;
+	int count = le16_to_cpu(xh->xh_count);
+	int delta, middle = count / 2;
+
+	/*
+	 * We start at the middle.  Each step gets farther away in both
+	 * directions.  We therefore hit the change in hash value
+	 * nearest to the middle.  Note that this loop does not execute for
+	 * count < 2.
+	 */
+	for (delta = 0; delta < middle; delta++) {
+		/* Let's check delta earlier than middle */
+		if (cmp_xe(&entries[middle - delta - 1],
+			   &entries[middle - delta]))
+			return middle - delta;
+
+		/* For even counts, don't walk off the end */
+		if ((middle + delta + 1) == count)
+			continue;
+
+		/* Now try delta past middle */
+		if (cmp_xe(&entries[middle + delta],
+			   &entries[middle + delta + 1]))
+			return middle + delta + 1;
+	}
+
+	/* Every entry had the same hash */
+	return count;
+}
+
+/*
+ * Move some xattrs in old bucket(blk) to new bucket(new_blk).
+ * first_hash will record the 1st hash of the new bucket.
+ *
+ * Normally half of the xattrs will be moved.  But we have to make
+ * sure that the xattrs with the same hash value are stored in the
+ * same bucket. If all the xattrs in this bucket have the same hash
+ * value, the new bucket will be initialized as an empty one and the
+ * first_hash will be initialized as (hash_value+1).
+ */
+static int ocfs2_divide_xattr_bucket(struct inode *inode,
+				    handle_t *handle,
+				    u64 blk,
+				    u64 new_blk,
+				    u32 *first_hash,
+				    int new_bucket_head)
+{
+	int ret, i;
+	int count, start, len, name_value_len = 0, name_offset = 0;
+	struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
+	struct ocfs2_xattr_header *xh;
+	struct ocfs2_xattr_entry *xe;
+	int blocksize = inode->i_sb->s_blocksize;
+
+	trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
+					      (unsigned long long)new_blk);
+
+	s_bucket = ocfs2_xattr_bucket_new(inode);
+	t_bucket = ocfs2_xattr_bucket_new(inode);
+	if (!s_bucket || !t_bucket) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_xattr_bucket(s_bucket, blk);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
+						OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
+	 * there's no need to read it.
+	 */
+	ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * Hey, if we're overwriting t_bucket, what difference does
+	 * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
+	 * same part of ocfs2_cp_xattr_bucket().
+	 */
+	ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
+						new_bucket_head ?
+						OCFS2_JOURNAL_ACCESS_CREATE :
+						OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	xh = bucket_xh(s_bucket);
+	count = le16_to_cpu(xh->xh_count);
+	start = ocfs2_xattr_find_divide_pos(xh);
+
+	if (start == count) {
+		xe = &xh->xh_entries[start-1];
+
+		/*
+		 * initialized a new empty bucket here.
+		 * The hash value is set as one larger than
+		 * that of the last entry in the previous bucket.
+		 */
+		for (i = 0; i < t_bucket->bu_blocks; i++)
+			memset(bucket_block(t_bucket, i), 0, blocksize);
+
+		xh = bucket_xh(t_bucket);
+		xh->xh_free_start = cpu_to_le16(blocksize);
+		xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
+		le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
+
+		goto set_num_buckets;
+	}
+
+	/* copy the whole bucket to the new first. */
+	ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
+
+	/* update the new bucket. */
+	xh = bucket_xh(t_bucket);
+
+	/*
+	 * Calculate the total name/value len and xh_free_start for
+	 * the old bucket first.
+	 */
+	name_offset = OCFS2_XATTR_BUCKET_SIZE;
+	name_value_len = 0;
+	for (i = 0; i < start; i++) {
+		xe = &xh->xh_entries[i];
+		name_value_len += namevalue_size_xe(xe);
+		if (le16_to_cpu(xe->xe_name_offset) < name_offset)
+			name_offset = le16_to_cpu(xe->xe_name_offset);
+	}
+
+	/*
+	 * Now begin the modification to the new bucket.
+	 *
+	 * In the new bucket, We just move the xattr entry to the beginning
+	 * and don't touch the name/value. So there will be some holes in the
+	 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
+	 * called.
+	 */
+	xe = &xh->xh_entries[start];
+	len = sizeof(struct ocfs2_xattr_entry) * (count - start);
+	trace_ocfs2_divide_xattr_bucket_move(len,
+			(int)((char *)xe - (char *)xh),
+			(int)((char *)xh->xh_entries - (char *)xh));
+	memmove((char *)xh->xh_entries, (char *)xe, len);
+	xe = &xh->xh_entries[count - start];
+	len = sizeof(struct ocfs2_xattr_entry) * start;
+	memset((char *)xe, 0, len);
+
+	le16_add_cpu(&xh->xh_count, -start);
+	le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
+
+	/* Calculate xh_free_start for the new bucket. */
+	xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
+	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
+		xe = &xh->xh_entries[i];
+		if (le16_to_cpu(xe->xe_name_offset) <
+		    le16_to_cpu(xh->xh_free_start))
+			xh->xh_free_start = xe->xe_name_offset;
+	}
+
+set_num_buckets:
+	/* set xh->xh_num_buckets for the new xh. */
+	if (new_bucket_head)
+		xh->xh_num_buckets = cpu_to_le16(1);
+	else
+		xh->xh_num_buckets = 0;
+
+	ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
+
+	/* store the first_hash of the new bucket. */
+	if (first_hash)
+		*first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
+
+	/*
+	 * Now only update the 1st block of the old bucket.  If we
+	 * just added a new empty bucket, there is no need to modify
+	 * it.
+	 */
+	if (start == count)
+		goto out;
+
+	xh = bucket_xh(s_bucket);
+	memset(&xh->xh_entries[start], 0,
+	       sizeof(struct ocfs2_xattr_entry) * (count - start));
+	xh->xh_count = cpu_to_le16(start);
+	xh->xh_free_start = cpu_to_le16(name_offset);
+	xh->xh_name_value_len = cpu_to_le16(name_value_len);
+
+	ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
+
+out:
+	ocfs2_xattr_bucket_free(s_bucket);
+	ocfs2_xattr_bucket_free(t_bucket);
+
+	return ret;
+}
+
+/*
+ * Copy xattr from one bucket to another bucket.
+ *
+ * The caller must make sure that the journal transaction
+ * has enough space for journaling.
+ */
+static int ocfs2_cp_xattr_bucket(struct inode *inode,
+				 handle_t *handle,
+				 u64 s_blkno,
+				 u64 t_blkno,
+				 int t_is_new)
+{
+	int ret;
+	struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
+
+	BUG_ON(s_blkno == t_blkno);
+
+	trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
+				    (unsigned long long)t_blkno,
+				    t_is_new);
+
+	s_bucket = ocfs2_xattr_bucket_new(inode);
+	t_bucket = ocfs2_xattr_bucket_new(inode);
+	if (!s_bucket || !t_bucket) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
+	if (ret)
+		goto out;
+
+	/*
+	 * Even if !t_is_new, we're overwriting t_bucket.  Thus,
+	 * there's no need to read it.
+	 */
+	ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
+	if (ret)
+		goto out;
+
+	/*
+	 * Hey, if we're overwriting t_bucket, what difference does
+	 * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
+	 * cluster to fill, we came here from
+	 * ocfs2_mv_xattr_buckets(), and it is really new -
+	 * ACCESS_CREATE is required.  But we also might have moved data
+	 * out of t_bucket before extending back into it.
+	 * ocfs2_add_new_xattr_bucket() can do this - its call to
+	 * ocfs2_add_new_xattr_cluster() may have created a new extent
+	 * and copied out the end of the old extent.  Then it re-extends
+	 * the old extent back to create space for new xattrs.  That's
+	 * how we get here, and the bucket isn't really new.
+	 */
+	ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
+						t_is_new ?
+						OCFS2_JOURNAL_ACCESS_CREATE :
+						OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret)
+		goto out;
+
+	ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
+	ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
+
+out:
+	ocfs2_xattr_bucket_free(t_bucket);
+	ocfs2_xattr_bucket_free(s_bucket);
+
+	return ret;
+}
+
+/*
+ * src_blk points to the start of an existing extent.  last_blk points to
+ * last cluster in that extent.  to_blk points to a newly allocated
+ * extent.  We copy the buckets from the cluster at last_blk to the new
+ * extent.  If start_bucket is non-zero, we skip that many buckets before
+ * we start copying.  The new extent's xh_num_buckets gets set to the
+ * number of buckets we copied.  The old extent's xh_num_buckets shrinks
+ * by the same amount.
+ */
+static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
+				  u64 src_blk, u64 last_blk, u64 to_blk,
+				  unsigned int start_bucket,
+				  u32 *first_hash)
+{
+	int i, ret, credits;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+	int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
+	struct ocfs2_xattr_bucket *old_first, *new_first;
+
+	trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
+				     (unsigned long long)to_blk);
+
+	BUG_ON(start_bucket >= num_buckets);
+	if (start_bucket) {
+		num_buckets -= start_bucket;
+		last_blk += (start_bucket * blks_per_bucket);
+	}
+
+	/* The first bucket of the original extent */
+	old_first = ocfs2_xattr_bucket_new(inode);
+	/* The first bucket of the new extent */
+	new_first = ocfs2_xattr_bucket_new(inode);
+	if (!old_first || !new_first) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_xattr_bucket(old_first, src_blk);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * We need to update the first bucket of the old extent and all
+	 * the buckets going to the new extent.
+	 */
+	credits = ((num_buckets + 1) * blks_per_bucket);
+	ret = ocfs2_extend_trans(handle, credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
+						OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	for (i = 0; i < num_buckets; i++) {
+		ret = ocfs2_cp_xattr_bucket(inode, handle,
+					    last_blk + (i * blks_per_bucket),
+					    to_blk + (i * blks_per_bucket),
+					    1);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	/*
+	 * Get the new bucket ready before we dirty anything
+	 * (This actually shouldn't fail, because we already dirtied
+	 * it once in ocfs2_cp_xattr_bucket()).
+	 */
+	ret = ocfs2_read_xattr_bucket(new_first, to_blk);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
+						OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Now update the headers */
+	le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
+	ocfs2_xattr_bucket_journal_dirty(handle, old_first);
+
+	bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
+	ocfs2_xattr_bucket_journal_dirty(handle, new_first);
+
+	if (first_hash)
+		*first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
+
+out:
+	ocfs2_xattr_bucket_free(new_first);
+	ocfs2_xattr_bucket_free(old_first);
+	return ret;
+}
+
+/*
+ * Move some xattrs in this cluster to the new cluster.
+ * This function should only be called when bucket size == cluster size.
+ * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
+ */
+static int ocfs2_divide_xattr_cluster(struct inode *inode,
+				      handle_t *handle,
+				      u64 prev_blk,
+				      u64 new_blk,
+				      u32 *first_hash)
+{
+	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+	int ret, credits = 2 * blk_per_bucket;
+
+	BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
+
+	ret = ocfs2_extend_trans(handle, credits);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	/* Move half of the xattr in start_blk to the next bucket. */
+	return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
+					  new_blk, first_hash, 1);
+}
+
+/*
+ * Move some xattrs from the old cluster to the new one since they are not
+ * contiguous in ocfs2 xattr tree.
+ *
+ * new_blk starts a new separate cluster, and we will move some xattrs from
+ * prev_blk to it. v_start will be set as the first name hash value in this
+ * new cluster so that it can be used as e_cpos during tree insertion and
+ * don't collide with our original b-tree operations. first_bh and header_bh
+ * will also be updated since they will be used in ocfs2_extend_xattr_bucket
+ * to extend the insert bucket.
+ *
+ * The problem is how much xattr should we move to the new one and when should
+ * we update first_bh and header_bh?
+ * 1. If cluster size > bucket size, that means the previous cluster has more
+ *    than 1 bucket, so just move half nums of bucket into the new cluster and
+ *    update the first_bh and header_bh if the insert bucket has been moved
+ *    to the new cluster.
+ * 2. If cluster_size == bucket_size:
+ *    a) If the previous extent rec has more than one cluster and the insert
+ *       place isn't in the last cluster, copy the entire last cluster to the
+ *       new one. This time, we don't need to upate the first_bh and header_bh
+ *       since they will not be moved into the new cluster.
+ *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
+ *       the new one. And we set the extend flag to zero if the insert place is
+ *       moved into the new allocated cluster since no extend is needed.
+ */
+static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
+					    handle_t *handle,
+					    struct ocfs2_xattr_bucket *first,
+					    struct ocfs2_xattr_bucket *target,
+					    u64 new_blk,
+					    u32 prev_clusters,
+					    u32 *v_start,
+					    int *extend)
+{
+	int ret;
+
+	trace_ocfs2_adjust_xattr_cross_cluster(
+			(unsigned long long)bucket_blkno(first),
+			(unsigned long long)new_blk, prev_clusters);
+
+	if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
+		ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
+							  handle,
+							  first, target,
+							  new_blk,
+							  prev_clusters,
+							  v_start);
+		if (ret)
+			mlog_errno(ret);
+	} else {
+		/* The start of the last cluster in the first extent */
+		u64 last_blk = bucket_blkno(first) +
+			((prev_clusters - 1) *
+			 ocfs2_clusters_to_blocks(inode->i_sb, 1));
+
+		if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
+			ret = ocfs2_mv_xattr_buckets(inode, handle,
+						     bucket_blkno(first),
+						     last_blk, new_blk, 0,
+						     v_start);
+			if (ret)
+				mlog_errno(ret);
+		} else {
+			ret = ocfs2_divide_xattr_cluster(inode, handle,
+							 last_blk, new_blk,
+							 v_start);
+			if (ret)
+				mlog_errno(ret);
+
+			if ((bucket_blkno(target) == last_blk) && extend)
+				*extend = 0;
+		}
+	}
+
+	return ret;
+}
+
+/*
+ * Add a new cluster for xattr storage.
+ *
+ * If the new cluster is contiguous with the previous one, it will be
+ * appended to the same extent record, and num_clusters will be updated.
+ * If not, we will insert a new extent for it and move some xattrs in
+ * the last cluster into the new allocated one.
+ * We also need to limit the maximum size of a btree leaf, otherwise we'll
+ * lose the benefits of hashing because we'll have to search large leaves.
+ * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
+ * if it's bigger).
+ *
+ * first_bh is the first block of the previous extent rec and header_bh
+ * indicates the bucket we will insert the new xattrs. They will be updated
+ * when the header_bh is moved into the new cluster.
+ */
+static int ocfs2_add_new_xattr_cluster(struct inode *inode,
+				       struct buffer_head *root_bh,
+				       struct ocfs2_xattr_bucket *first,
+				       struct ocfs2_xattr_bucket *target,
+				       u32 *num_clusters,
+				       u32 prev_cpos,
+				       int *extend,
+				       struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret;
+	u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
+	u32 prev_clusters = *num_clusters;
+	u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
+	u64 block;
+	handle_t *handle = ctxt->handle;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_extent_tree et;
+
+	trace_ocfs2_add_new_xattr_cluster_begin(
+		(unsigned long long)OCFS2_I(inode)->ip_blkno,
+		(unsigned long long)bucket_blkno(first),
+		prev_cpos, prev_clusters);
+
+	ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
+
+	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
+				     clusters_to_add, &bit_off, &num_bits);
+	if (ret < 0) {
+		if (ret != -ENOSPC)
+			mlog_errno(ret);
+		goto leave;
+	}
+
+	BUG_ON(num_bits > clusters_to_add);
+
+	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
+	trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
+
+	if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
+	    (prev_clusters + num_bits) << osb->s_clustersize_bits <=
+	     OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
+		/*
+		 * If this cluster is contiguous with the old one and
+		 * adding this new cluster, we don't surpass the limit of
+		 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
+		 * initialized and used like other buckets in the previous
+		 * cluster.
+		 * So add it as a contiguous one. The caller will handle
+		 * its init process.
+		 */
+		v_start = prev_cpos + prev_clusters;
+		*num_clusters = prev_clusters + num_bits;
+	} else {
+		ret = ocfs2_adjust_xattr_cross_cluster(inode,
+						       handle,
+						       first,
+						       target,
+						       block,
+						       prev_clusters,
+						       &v_start,
+						       extend);
+		if (ret) {
+			mlog_errno(ret);
+			goto leave;
+		}
+	}
+
+	trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
+						 v_start, num_bits);
+	ret = ocfs2_insert_extent(handle, &et, v_start, block,
+				  num_bits, 0, ctxt->meta_ac);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	ocfs2_journal_dirty(handle, root_bh);
+
+leave:
+	return ret;
+}
+
+/*
+ * We are given an extent.  'first' is the bucket at the very front of
+ * the extent.  The extent has space for an additional bucket past
+ * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
+ * of the target bucket.  We wish to shift every bucket past the target
+ * down one, filling in that additional space.  When we get back to the
+ * target, we split the target between itself and the now-empty bucket
+ * at target+1 (aka, target_blkno + blks_per_bucket).
+ */
+static int ocfs2_extend_xattr_bucket(struct inode *inode,
+				     handle_t *handle,
+				     struct ocfs2_xattr_bucket *first,
+				     u64 target_blk,
+				     u32 num_clusters)
+{
+	int ret, credits;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+	u64 end_blk;
+	u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
+
+	trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
+					(unsigned long long)bucket_blkno(first),
+					num_clusters, new_bucket);
+
+	/* The extent must have room for an additional bucket */
+	BUG_ON(new_bucket >=
+	       (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
+
+	/* end_blk points to the last existing bucket */
+	end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
+
+	/*
+	 * end_blk is the start of the last existing bucket.
+	 * Thus, (end_blk - target_blk) covers the target bucket and
+	 * every bucket after it up to, but not including, the last
+	 * existing bucket.  Then we add the last existing bucket, the
+	 * new bucket, and the first bucket (3 * blk_per_bucket).
+	 */
+	credits = (end_blk - target_blk) + (3 * blk_per_bucket);
+	ret = ocfs2_extend_trans(handle, credits);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_xattr_bucket_journal_access(handle, first,
+						OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	while (end_blk != target_blk) {
+		ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
+					    end_blk + blk_per_bucket, 0);
+		if (ret)
+			goto out;
+		end_blk -= blk_per_bucket;
+	}
+
+	/* Move half of the xattr in target_blkno to the next bucket. */
+	ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
+					target_blk + blk_per_bucket, NULL, 0);
+
+	le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
+	ocfs2_xattr_bucket_journal_dirty(handle, first);
+
+out:
+	return ret;
+}
+
+/*
+ * Add new xattr bucket in an extent record and adjust the buckets
+ * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
+ * bucket we want to insert into.
+ *
+ * In the easy case, we will move all the buckets after target down by
+ * one. Half of target's xattrs will be moved to the next bucket.
+ *
+ * If current cluster is full, we'll allocate a new one.  This may not
+ * be contiguous.  The underlying calls will make sure that there is
+ * space for the insert, shifting buckets around if necessary.
+ * 'target' may be moved by those calls.
+ */
+static int ocfs2_add_new_xattr_bucket(struct inode *inode,
+				      struct buffer_head *xb_bh,
+				      struct ocfs2_xattr_bucket *target,
+				      struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	struct ocfs2_xattr_block *xb =
+			(struct ocfs2_xattr_block *)xb_bh->b_data;
+	struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
+	struct ocfs2_extent_list *el = &xb_root->xt_list;
+	u32 name_hash =
+		le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	int ret, num_buckets, extend = 1;
+	u64 p_blkno;
+	u32 e_cpos, num_clusters;
+	/* The bucket at the front of the extent */
+	struct ocfs2_xattr_bucket *first;
+
+	trace_ocfs2_add_new_xattr_bucket(
+				(unsigned long long)bucket_blkno(target));
+
+	/* The first bucket of the original extent */
+	first = ocfs2_xattr_bucket_new(inode);
+	if (!first) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
+				  &num_clusters, el);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_read_xattr_bucket(first, p_blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
+	if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
+		/*
+		 * This can move first+target if the target bucket moves
+		 * to the new extent.
+		 */
+		ret = ocfs2_add_new_xattr_cluster(inode,
+						  xb_bh,
+						  first,
+						  target,
+						  &num_clusters,
+						  e_cpos,
+						  &extend,
+						  ctxt);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (extend) {
+		ret = ocfs2_extend_xattr_bucket(inode,
+						ctxt->handle,
+						first,
+						bucket_blkno(target),
+						num_clusters);
+		if (ret)
+			mlog_errno(ret);
+	}
+
+out:
+	ocfs2_xattr_bucket_free(first);
+
+	return ret;
+}
+
+/*
+ * Truncate the specified xe_off entry in xattr bucket.
+ * bucket is indicated by header_bh and len is the new length.
+ * Both the ocfs2_xattr_value_root and the entry will be updated here.
+ *
+ * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
+ */
+static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
+					     struct ocfs2_xattr_bucket *bucket,
+					     int xe_off,
+					     int len,
+					     struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret, offset;
+	u64 value_blk;
+	struct ocfs2_xattr_entry *xe;
+	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
+	size_t blocksize = inode->i_sb->s_blocksize;
+	struct ocfs2_xattr_value_buf vb = {
+		.vb_access = ocfs2_journal_access,
+	};
+
+	xe = &xh->xh_entries[xe_off];
+
+	BUG_ON(!xe || ocfs2_xattr_is_local(xe));
+
+	offset = le16_to_cpu(xe->xe_name_offset) +
+		 OCFS2_XATTR_SIZE(xe->xe_name_len);
+
+	value_blk = offset / blocksize;
+
+	/* We don't allow ocfs2_xattr_value to be stored in different block. */
+	BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
+
+	vb.vb_bh = bucket->bu_bhs[value_blk];
+	BUG_ON(!vb.vb_bh);
+
+	vb.vb_xv = (struct ocfs2_xattr_value_root *)
+		(vb.vb_bh->b_data + offset % blocksize);
+
+	/*
+	 * From here on out we have to dirty the bucket.  The generic
+	 * value calls only modify one of the bucket's bhs, but we need
+	 * to send the bucket at once.  So if they error, they *could* have
+	 * modified something.  We have to assume they did, and dirty
+	 * the whole bucket.  This leaves us in a consistent state.
+	 */
+	trace_ocfs2_xattr_bucket_value_truncate(
+			(unsigned long long)bucket_blkno(bucket), xe_off, len);
+	ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
+						OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	xe->xe_value_size = cpu_to_le64(len);
+
+	ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
+
+out:
+	return ret;
+}
+
+static int ocfs2_rm_xattr_cluster(struct inode *inode,
+				  struct buffer_head *root_bh,
+				  u64 blkno,
+				  u32 cpos,
+				  u32 len,
+				  void *para)
+{
+	int ret;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct inode *tl_inode = osb->osb_tl_inode;
+	handle_t *handle;
+	struct ocfs2_xattr_block *xb =
+			(struct ocfs2_xattr_block *)root_bh->b_data;
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct ocfs2_extent_tree et;
+
+	ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
+					  ocfs2_delete_xattr_in_bucket, para);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
+
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	trace_ocfs2_rm_xattr_cluster(
+			(unsigned long long)OCFS2_I(inode)->ip_blkno,
+			(unsigned long long)blkno, cpos, len);
+
+	ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
+					       len);
+
+	ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	mutex_lock(&tl_inode->i_mutex);
+
+	if (ocfs2_truncate_log_needs_flush(osb)) {
+		ret = __ocfs2_flush_truncate_log(osb);
+		if (ret < 0) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
+	if (IS_ERR(handle)) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
+				      OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
+				  &dealloc);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
+	ocfs2_journal_dirty(handle, root_bh);
+
+	ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
+	if (ret)
+		mlog_errno(ret);
+	ocfs2_update_inode_fsync_trans(handle, inode, 0);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+out:
+	ocfs2_schedule_truncate_log_flush(osb, 1);
+
+	mutex_unlock(&tl_inode->i_mutex);
+
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+
+	ocfs2_run_deallocs(osb, &dealloc);
+
+	return ret;
+}
+
+/*
+ * check whether the xattr bucket is filled up with the same hash value.
+ * If we want to insert the xattr with the same hash, return -ENOSPC.
+ * If we want to insert a xattr with different hash value, go ahead
+ * and ocfs2_divide_xattr_bucket will handle this.
+ */
+static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
+					      struct ocfs2_xattr_bucket *bucket,
+					      const char *name)
+{
+	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
+	u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
+
+	if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
+		return 0;
+
+	if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
+	    xh->xh_entries[0].xe_name_hash) {
+		mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
+		     "hash = %u\n",
+		     (unsigned long long)bucket_blkno(bucket),
+		     le32_to_cpu(xh->xh_entries[0].xe_name_hash));
+		return -ENOSPC;
+	}
+
+	return 0;
+}
+
+/*
+ * Try to set the entry in the current bucket.  If we fail, the caller
+ * will handle getting us another bucket.
+ */
+static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
+					struct ocfs2_xattr_info *xi,
+					struct ocfs2_xattr_search *xs,
+					struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret;
+	struct ocfs2_xa_loc loc;
+
+	trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
+
+	ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
+				       xs->not_found ? NULL : xs->here);
+	ret = ocfs2_xa_set(&loc, xi, ctxt);
+	if (!ret) {
+		xs->here = loc.xl_entry;
+		goto out;
+	}
+	if (ret != -ENOSPC) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Ok, we need space.  Let's try defragmenting the bucket. */
+	ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
+					xs->bucket);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_xa_set(&loc, xi, ctxt);
+	if (!ret) {
+		xs->here = loc.xl_entry;
+		goto out;
+	}
+	if (ret != -ENOSPC)
+		mlog_errno(ret);
+
+
+out:
+	return ret;
+}
+
+static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
+					     struct ocfs2_xattr_info *xi,
+					     struct ocfs2_xattr_search *xs,
+					     struct ocfs2_xattr_set_ctxt *ctxt)
+{
+	int ret;
+
+	trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
+
+	ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
+	if (!ret)
+		goto out;
+	if (ret != -ENOSPC) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/* Ack, need more space.  Let's try to get another bucket! */
+
+	/*
+	 * We do not allow for overlapping ranges between buckets. And
+	 * the maximum number of collisions we will allow for then is
+	 * one bucket's worth, so check it here whether we need to
+	 * add a new bucket for the insert.
+	 */
+	ret = ocfs2_check_xattr_bucket_collision(inode,
+						 xs->bucket,
+						 xi->xi_name);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_add_new_xattr_bucket(inode,
+					 xs->xattr_bh,
+					 xs->bucket,
+					 ctxt);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * ocfs2_add_new_xattr_bucket() will have updated
+	 * xs->bucket if it moved, but it will not have updated
+	 * any of the other search fields.  Thus, we drop it and
+	 * re-search.  Everything should be cached, so it'll be
+	 * quick.
+	 */
+	ocfs2_xattr_bucket_relse(xs->bucket);
+	ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
+					   xi->xi_name_index,
+					   xi->xi_name, xs);
+	if (ret && ret != -ENODATA)
+		goto out;
+	xs->not_found = ret;
+
+	/* Ok, we have a new bucket, let's try again */
+	ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
+	if (ret && (ret != -ENOSPC))
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
+					struct ocfs2_xattr_bucket *bucket,
+					void *para)
+{
+	int ret = 0, ref_credits;
+	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
+	u16 i;
+	struct ocfs2_xattr_entry *xe;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
+	int credits = ocfs2_remove_extent_credits(osb->sb) +
+		ocfs2_blocks_per_xattr_bucket(inode->i_sb);
+	struct ocfs2_xattr_value_root *xv;
+	struct ocfs2_rm_xattr_bucket_para *args =
+			(struct ocfs2_rm_xattr_bucket_para *)para;
+
+	ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
+
+	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
+		xe = &xh->xh_entries[i];
+		if (ocfs2_xattr_is_local(xe))
+			continue;
+
+		ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
+						      i, &xv, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
+							 args->ref_ci,
+							 args->ref_root_bh,
+							 &ctxt.meta_ac,
+							 &ref_credits);
+
+		ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
+		if (IS_ERR(ctxt.handle)) {
+			ret = PTR_ERR(ctxt.handle);
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
+							i, 0, &ctxt);
+
+		ocfs2_commit_trans(osb, ctxt.handle);
+		if (ctxt.meta_ac) {
+			ocfs2_free_alloc_context(ctxt.meta_ac);
+			ctxt.meta_ac = NULL;
+		}
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+	}
+
+	if (ctxt.meta_ac)
+		ocfs2_free_alloc_context(ctxt.meta_ac);
+	ocfs2_schedule_truncate_log_flush(osb, 1);
+	ocfs2_run_deallocs(osb, &ctxt.dealloc);
+	return ret;
+}
+
+/*
+ * Whenever we modify a xattr value root in the bucket(e.g, CoW
+ * or change the extent record flag), we need to recalculate
+ * the metaecc for the whole bucket. So it is done here.
+ *
+ * Note:
+ * We have to give the extra credits for the caller.
+ */
+static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
+					    handle_t *handle,
+					    void *para)
+{
+	int ret;
+	struct ocfs2_xattr_bucket *bucket =
+			(struct ocfs2_xattr_bucket *)para;
+
+	ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
+						OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	ocfs2_xattr_bucket_journal_dirty(handle, bucket);
+
+	return 0;
+}
+
+/*
+ * Special action we need if the xattr value is refcounted.
+ *
+ * 1. If the xattr is refcounted, lock the tree.
+ * 2. CoW the xattr if we are setting the new value and the value
+ *    will be stored outside.
+ * 3. In other case, decrease_refcount will work for us, so just
+ *    lock the refcount tree, calculate the meta and credits is OK.
+ *
+ * We have to do CoW before ocfs2_init_xattr_set_ctxt since
+ * currently CoW is a completed transaction, while this function
+ * will also lock the allocators and let us deadlock. So we will
+ * CoW the whole xattr value.
+ */
+static int ocfs2_prepare_refcount_xattr(struct inode *inode,
+					struct ocfs2_dinode *di,
+					struct ocfs2_xattr_info *xi,
+					struct ocfs2_xattr_search *xis,
+					struct ocfs2_xattr_search *xbs,
+					struct ocfs2_refcount_tree **ref_tree,
+					int *meta_add,
+					int *credits)
+{
+	int ret = 0;
+	struct ocfs2_xattr_block *xb;
+	struct ocfs2_xattr_entry *xe;
+	char *base;
+	u32 p_cluster, num_clusters;
+	unsigned int ext_flags;
+	int name_offset, name_len;
+	struct ocfs2_xattr_value_buf vb;
+	struct ocfs2_xattr_bucket *bucket = NULL;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_post_refcount refcount;
+	struct ocfs2_post_refcount *p = NULL;
+	struct buffer_head *ref_root_bh = NULL;
+
+	if (!xis->not_found) {
+		xe = xis->here;
+		name_offset = le16_to_cpu(xe->xe_name_offset);
+		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
+		base = xis->base;
+		vb.vb_bh = xis->inode_bh;
+		vb.vb_access = ocfs2_journal_access_di;
+	} else {
+		int i, block_off = 0;
+		xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
+		xe = xbs->here;
+		name_offset = le16_to_cpu(xe->xe_name_offset);
+		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
+		i = xbs->here - xbs->header->xh_entries;
+
+		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
+			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
+							bucket_xh(xbs->bucket),
+							i, &block_off,
+							&name_offset);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+			base = bucket_block(xbs->bucket, block_off);
+			vb.vb_bh = xbs->bucket->bu_bhs[block_off];
+			vb.vb_access = ocfs2_journal_access;
+
+			if (ocfs2_meta_ecc(osb)) {
+				/*create parameters for ocfs2_post_refcount. */
+				bucket = xbs->bucket;
+				refcount.credits = bucket->bu_blocks;
+				refcount.para = bucket;
+				refcount.func =
+					ocfs2_xattr_bucket_post_refcount;
+				p = &refcount;
+			}
+		} else {
+			base = xbs->base;
+			vb.vb_bh = xbs->xattr_bh;
+			vb.vb_access = ocfs2_journal_access_xb;
+		}
+	}
+
+	if (ocfs2_xattr_is_local(xe))
+		goto out;
+
+	vb.vb_xv = (struct ocfs2_xattr_value_root *)
+				(base + name_offset + name_len);
+
+	ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
+				       &num_clusters, &vb.vb_xv->xr_list,
+				       &ext_flags);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * We just need to check the 1st extent record, since we always
+	 * CoW the whole xattr. So there shouldn't be a xattr with
+	 * some REFCOUNT extent recs after the 1st one.
+	 */
+	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
+		goto out;
+
+	ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
+				       1, ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * If we are deleting the xattr or the new size will be stored inside,
+	 * cool, leave it there, the xattr truncate process will remove them
+	 * for us(it still needs the refcount tree lock and the meta, credits).
+	 * And the worse case is that every cluster truncate will split the
+	 * refcount tree, and make the original extent become 3. So we will need
+	 * 2 * cluster more extent recs at most.
+	 */
+	if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
+
+		ret = ocfs2_refcounted_xattr_delete_need(inode,
+							 &(*ref_tree)->rf_ci,
+							 ref_root_bh, vb.vb_xv,
+							 meta_add, credits);
+		if (ret)
+			mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
+				       *ref_tree, ref_root_bh, 0,
+				       le32_to_cpu(vb.vb_xv->xr_clusters), p);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	brelse(ref_root_bh);
+	return ret;
+}
+
+/*
+ * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
+ * The physical clusters will be added to refcount tree.
+ */
+static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
+				struct ocfs2_xattr_value_root *xv,
+				struct ocfs2_extent_tree *value_et,
+				struct ocfs2_caching_info *ref_ci,
+				struct buffer_head *ref_root_bh,
+				struct ocfs2_cached_dealloc_ctxt *dealloc,
+				struct ocfs2_post_refcount *refcount)
+{
+	int ret = 0;
+	u32 clusters = le32_to_cpu(xv->xr_clusters);
+	u32 cpos, p_cluster, num_clusters;
+	struct ocfs2_extent_list *el = &xv->xr_list;
+	unsigned int ext_flags;
+
+	cpos = 0;
+	while (cpos < clusters) {
+		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
+					       &num_clusters, el, &ext_flags);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		cpos += num_clusters;
+		if ((ext_flags & OCFS2_EXT_REFCOUNTED))
+			continue;
+
+		BUG_ON(!p_cluster);
+
+		ret = ocfs2_add_refcount_flag(inode, value_et,
+					      ref_ci, ref_root_bh,
+					      cpos - num_clusters,
+					      p_cluster, num_clusters,
+					      dealloc, refcount);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+	}
+
+	return ret;
+}
+
+/*
+ * Given a normal ocfs2_xattr_header, refcount all the entries which
+ * have value stored outside.
+ * Used for xattrs stored in inode and ocfs2_xattr_block.
+ */
+static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
+				struct ocfs2_xattr_value_buf *vb,
+				struct ocfs2_xattr_header *header,
+				struct ocfs2_caching_info *ref_ci,
+				struct buffer_head *ref_root_bh,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+
+	struct ocfs2_xattr_entry *xe;
+	struct ocfs2_xattr_value_root *xv;
+	struct ocfs2_extent_tree et;
+	int i, ret = 0;
+
+	for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
+		xe = &header->xh_entries[i];
+
+		if (ocfs2_xattr_is_local(xe))
+			continue;
+
+		xv = (struct ocfs2_xattr_value_root *)((void *)header +
+			le16_to_cpu(xe->xe_name_offset) +
+			OCFS2_XATTR_SIZE(xe->xe_name_len));
+
+		vb->vb_xv = xv;
+		ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
+
+		ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
+							ref_ci, ref_root_bh,
+							dealloc, NULL);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+	}
+
+	return ret;
+}
+
+static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
+				struct buffer_head *fe_bh,
+				struct ocfs2_caching_info *ref_ci,
+				struct buffer_head *ref_root_bh,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
+	struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
+				(fe_bh->b_data + inode->i_sb->s_blocksize -
+				le16_to_cpu(di->i_xattr_inline_size));
+	struct ocfs2_xattr_value_buf vb = {
+		.vb_bh = fe_bh,
+		.vb_access = ocfs2_journal_access_di,
+	};
+
+	return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
+						  ref_ci, ref_root_bh, dealloc);
+}
+
+struct ocfs2_xattr_tree_value_refcount_para {
+	struct ocfs2_caching_info *ref_ci;
+	struct buffer_head *ref_root_bh;
+	struct ocfs2_cached_dealloc_ctxt *dealloc;
+};
+
+static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
+					   struct ocfs2_xattr_bucket *bucket,
+					   int offset,
+					   struct ocfs2_xattr_value_root **xv,
+					   struct buffer_head **bh)
+{
+	int ret, block_off, name_offset;
+	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
+	struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
+	void *base;
+
+	ret = ocfs2_xattr_bucket_get_name_value(sb,
+						bucket_xh(bucket),
+						offset,
+						&block_off,
+						&name_offset);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	base = bucket_block(bucket, block_off);
+
+	*xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
+			 OCFS2_XATTR_SIZE(xe->xe_name_len));
+
+	if (bh)
+		*bh = bucket->bu_bhs[block_off];
+out:
+	return ret;
+}
+
+/*
+ * For a given xattr bucket, refcount all the entries which
+ * have value stored outside.
+ */
+static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
+					     struct ocfs2_xattr_bucket *bucket,
+					     void *para)
+{
+	int i, ret = 0;
+	struct ocfs2_extent_tree et;
+	struct ocfs2_xattr_tree_value_refcount_para *ref =
+			(struct ocfs2_xattr_tree_value_refcount_para *)para;
+	struct ocfs2_xattr_header *xh =
+			(struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
+	struct ocfs2_xattr_entry *xe;
+	struct ocfs2_xattr_value_buf vb = {
+		.vb_access = ocfs2_journal_access,
+	};
+	struct ocfs2_post_refcount refcount = {
+		.credits = bucket->bu_blocks,
+		.para = bucket,
+		.func = ocfs2_xattr_bucket_post_refcount,
+	};
+	struct ocfs2_post_refcount *p = NULL;
+
+	/* We only need post_refcount if we support metaecc. */
+	if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
+		p = &refcount;
+
+	trace_ocfs2_xattr_bucket_value_refcount(
+				(unsigned long long)bucket_blkno(bucket),
+				le16_to_cpu(xh->xh_count));
+	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
+		xe = &xh->xh_entries[i];
+
+		if (ocfs2_xattr_is_local(xe))
+			continue;
+
+		ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
+						      &vb.vb_xv, &vb.vb_bh);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		ocfs2_init_xattr_value_extent_tree(&et,
+						   INODE_CACHE(inode), &vb);
+
+		ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
+							&et, ref->ref_ci,
+							ref->ref_root_bh,
+							ref->dealloc, p);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+	}
+
+	return ret;
+
+}
+
+static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
+				     struct buffer_head *root_bh,
+				     u64 blkno, u32 cpos, u32 len, void *para)
+{
+	return ocfs2_iterate_xattr_buckets(inode, blkno, len,
+					   ocfs2_xattr_bucket_value_refcount,
+					   para);
+}
+
+static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
+				struct buffer_head *blk_bh,
+				struct ocfs2_caching_info *ref_ci,
+				struct buffer_head *ref_root_bh,
+				struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret = 0;
+	struct ocfs2_xattr_block *xb =
+				(struct ocfs2_xattr_block *)blk_bh->b_data;
+
+	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
+		struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
+		struct ocfs2_xattr_value_buf vb = {
+			.vb_bh = blk_bh,
+			.vb_access = ocfs2_journal_access_xb,
+		};
+
+		ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
+							 ref_ci, ref_root_bh,
+							 dealloc);
+	} else {
+		struct ocfs2_xattr_tree_value_refcount_para para = {
+			.ref_ci = ref_ci,
+			.ref_root_bh = ref_root_bh,
+			.dealloc = dealloc,
+		};
+
+		ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
+						ocfs2_refcount_xattr_tree_rec,
+						&para);
+	}
+
+	return ret;
+}
+
+int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
+				     struct buffer_head *fe_bh,
+				     struct ocfs2_caching_info *ref_ci,
+				     struct buffer_head *ref_root_bh,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc)
+{
+	int ret = 0;
+	struct ocfs2_inode_info *oi = OCFS2_I(inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
+	struct buffer_head *blk_bh = NULL;
+
+	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
+		ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
+							 ref_ci, ref_root_bh,
+							 dealloc);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (!di->i_xattr_loc)
+		goto out;
+
+	ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
+				     &blk_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
+						ref_root_bh, dealloc);
+	if (ret)
+		mlog_errno(ret);
+
+	brelse(blk_bh);
+out:
+
+	return ret;
+}
+
+typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
+/*
+ * Store the information we need in xattr reflink.
+ * old_bh and new_bh are inode bh for the old and new inode.
+ */
+struct ocfs2_xattr_reflink {
+	struct inode *old_inode;
+	struct inode *new_inode;
+	struct buffer_head *old_bh;
+	struct buffer_head *new_bh;
+	struct ocfs2_caching_info *ref_ci;
+	struct buffer_head *ref_root_bh;
+	struct ocfs2_cached_dealloc_ctxt *dealloc;
+	should_xattr_reflinked *xattr_reflinked;
+};
+
+/*
+ * Given a xattr header and xe offset,
+ * return the proper xv and the corresponding bh.
+ * xattr in inode, block and xattr tree have different implementaions.
+ */
+typedef int (get_xattr_value_root)(struct super_block *sb,
+				   struct buffer_head *bh,
+				   struct ocfs2_xattr_header *xh,
+				   int offset,
+				   struct ocfs2_xattr_value_root **xv,
+				   struct buffer_head **ret_bh,
+				   void *para);
+
+/*
+ * Calculate all the xattr value root metadata stored in this xattr header and
+ * credits we need if we create them from the scratch.
+ * We use get_xattr_value_root so that all types of xattr container can use it.
+ */
+static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
+					     struct buffer_head *bh,
+					     struct ocfs2_xattr_header *xh,
+					     int *metas, int *credits,
+					     int *num_recs,
+					     get_xattr_value_root *func,
+					     void *para)
+{
+	int i, ret = 0;
+	struct ocfs2_xattr_value_root *xv;
+	struct ocfs2_xattr_entry *xe;
+
+	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
+		xe = &xh->xh_entries[i];
+		if (ocfs2_xattr_is_local(xe))
+			continue;
+
+		ret = func(sb, bh, xh, i, &xv, NULL, para);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		*metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
+			  le16_to_cpu(xv->xr_list.l_next_free_rec);
+
+		*credits += ocfs2_calc_extend_credits(sb,
+						&def_xv.xv.xr_list);
+
+		/*
+		 * If the value is a tree with depth > 1, We don't go deep
+		 * to the extent block, so just calculate a maximum record num.
+		 */
+		if (!xv->xr_list.l_tree_depth)
+			*num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
+		else
+			*num_recs += ocfs2_clusters_for_bytes(sb,
+							      XATTR_SIZE_MAX);
+	}
+
+	return ret;
+}
+
+/* Used by xattr inode and block to return the right xv and buffer_head. */
+static int ocfs2_get_xattr_value_root(struct super_block *sb,
+				      struct buffer_head *bh,
+				      struct ocfs2_xattr_header *xh,
+				      int offset,
+				      struct ocfs2_xattr_value_root **xv,
+				      struct buffer_head **ret_bh,
+				      void *para)
+{
+	struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
+
+	*xv = (struct ocfs2_xattr_value_root *)((void *)xh +
+		le16_to_cpu(xe->xe_name_offset) +
+		OCFS2_XATTR_SIZE(xe->xe_name_len));
+
+	if (ret_bh)
+		*ret_bh = bh;
+
+	return 0;
+}
+
+/*
+ * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
+ * It is only used for inline xattr and xattr block.
+ */
+static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
+					struct ocfs2_xattr_header *xh,
+					struct buffer_head *ref_root_bh,
+					int *credits,
+					struct ocfs2_alloc_context **meta_ac)
+{
+	int ret, meta_add = 0, num_recs = 0;
+	struct ocfs2_refcount_block *rb =
+			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
+
+	*credits = 0;
+
+	ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
+						&meta_add, credits, &num_recs,
+						ocfs2_get_xattr_value_root,
+						NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	/*
+	 * We need to add/modify num_recs in refcount tree, so just calculate
+	 * an approximate number we need for refcount tree change.
+	 * Sometimes we need to split the tree, and after split,  half recs
+	 * will be moved to the new block, and a new block can only provide
+	 * half number of recs. So we multiple new blocks by 2.
+	 */
+	num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
+	meta_add += num_recs;
+	*credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
+	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
+		*credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
+			    le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
+	else
+		*credits += 1;
+
+	ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	return ret;
+}
+
+/*
+ * Given a xattr header, reflink all the xattrs in this container.
+ * It can be used for inode, block and bucket.
+ *
+ * NOTE:
+ * Before we call this function, the caller has memcpy the xattr in
+ * old_xh to the new_xh.
+ *
+ * If args.xattr_reflinked is set, call it to decide whether the xe should
+ * be reflinked or not. If not, remove it from the new xattr header.
+ */
+static int ocfs2_reflink_xattr_header(handle_t *handle,
+				      struct ocfs2_xattr_reflink *args,
+				      struct buffer_head *old_bh,
+				      struct ocfs2_xattr_header *xh,
+				      struct buffer_head *new_bh,
+				      struct ocfs2_xattr_header *new_xh,
+				      struct ocfs2_xattr_value_buf *vb,
+				      struct ocfs2_alloc_context *meta_ac,
+				      get_xattr_value_root *func,
+				      void *para)
+{
+	int ret = 0, i, j;
+	struct super_block *sb = args->old_inode->i_sb;
+	struct buffer_head *value_bh;
+	struct ocfs2_xattr_entry *xe, *last;
+	struct ocfs2_xattr_value_root *xv, *new_xv;
+	struct ocfs2_extent_tree data_et;
+	u32 clusters, cpos, p_cluster, num_clusters;
+	unsigned int ext_flags = 0;
+
+	trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
+					 le16_to_cpu(xh->xh_count));
+
+	last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
+	for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
+		xe = &xh->xh_entries[i];
+
+		if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
+			xe = &new_xh->xh_entries[j];
+
+			le16_add_cpu(&new_xh->xh_count, -1);
+			if (new_xh->xh_count) {
+				memmove(xe, xe + 1,
+					(void *)last - (void *)xe);
+				memset(last, 0,
+				       sizeof(struct ocfs2_xattr_entry));
+			}
+
+			/*
+			 * We don't want j to increase in the next round since
+			 * it is already moved ahead.
+			 */
+			j--;
+			continue;
+		}
+
+		if (ocfs2_xattr_is_local(xe))
+			continue;
+
+		ret = func(sb, old_bh, xh, i, &xv, NULL, para);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		/*
+		 * For the xattr which has l_tree_depth = 0, all the extent
+		 * recs have already be copied to the new xh with the
+		 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
+		 * increase the refount count int the refcount tree.
+		 *
+		 * For the xattr which has l_tree_depth > 0, we need
+		 * to initialize it to the empty default value root,
+		 * and then insert the extents one by one.
+		 */
+		if (xv->xr_list.l_tree_depth) {
+			memcpy(new_xv, &def_xv, sizeof(def_xv));
+			vb->vb_xv = new_xv;
+			vb->vb_bh = value_bh;
+			ocfs2_init_xattr_value_extent_tree(&data_et,
+					INODE_CACHE(args->new_inode), vb);
+		}
+
+		clusters = le32_to_cpu(xv->xr_clusters);
+		cpos = 0;
+		while (cpos < clusters) {
+			ret = ocfs2_xattr_get_clusters(args->old_inode,
+						       cpos,
+						       &p_cluster,
+						       &num_clusters,
+						       &xv->xr_list,
+						       &ext_flags);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			BUG_ON(!p_cluster);
+
+			if (xv->xr_list.l_tree_depth) {
+				ret = ocfs2_insert_extent(handle,
+						&data_et, cpos,
+						ocfs2_clusters_to_blocks(
+							args->old_inode->i_sb,
+							p_cluster),
+						num_clusters, ext_flags,
+						meta_ac);
+				if (ret) {
+					mlog_errno(ret);
+					goto out;
+				}
+			}
+
+			ret = ocfs2_increase_refcount(handle, args->ref_ci,
+						      args->ref_root_bh,
+						      p_cluster, num_clusters,
+						      meta_ac, args->dealloc);
+			if (ret) {
+				mlog_errno(ret);
+				goto out;
+			}
+
+			cpos += num_clusters;
+		}
+	}
+
+out:
+	return ret;
+}
+
+static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
+{
+	int ret = 0, credits = 0;
+	handle_t *handle;
+	struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
+	int inline_size = le16_to_cpu(di->i_xattr_inline_size);
+	int header_off = osb->sb->s_blocksize - inline_size;
+	struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
+					(args->old_bh->b_data + header_off);
+	struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
+					(args->new_bh->b_data + header_off);
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	struct ocfs2_inode_info *new_oi;
+	struct ocfs2_dinode *new_di;
+	struct ocfs2_xattr_value_buf vb = {
+		.vb_bh = args->new_bh,
+		.vb_access = ocfs2_journal_access_di,
+	};
+
+	ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
+						  &credits, &meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
+				      args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	memcpy(args->new_bh->b_data + header_off,
+	       args->old_bh->b_data + header_off, inline_size);
+
+	new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
+	new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
+
+	ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
+					 args->new_bh, new_xh, &vb, meta_ac,
+					 ocfs2_get_xattr_value_root, NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	new_oi = OCFS2_I(args->new_inode);
+	/*
+	 * Adjust extent record count to reserve space for extended attribute.
+	 * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
+	 */
+	if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
+	    !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
+		struct ocfs2_extent_list *el = &new_di->id2.i_list;
+		le16_add_cpu(&el->l_count, -(inline_size /
+					sizeof(struct ocfs2_extent_rec)));
+	}
+	spin_lock(&new_oi->ip_lock);
+	new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
+	new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
+	spin_unlock(&new_oi->ip_lock);
+
+	ocfs2_journal_dirty(handle, args->new_bh);
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+	return ret;
+}
+
+static int ocfs2_create_empty_xattr_block(struct inode *inode,
+					  struct buffer_head *fe_bh,
+					  struct buffer_head **ret_bh,
+					  int indexed)
+{
+	int ret;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_xattr_set_ctxt ctxt;
+
+	memset(&ctxt, 0, sizeof(ctxt));
+	ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
+	if (ret < 0) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
+	if (IS_ERR(ctxt.handle)) {
+		ret = PTR_ERR(ctxt.handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	trace_ocfs2_create_empty_xattr_block(
+				(unsigned long long)fe_bh->b_blocknr, indexed);
+	ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
+				       ret_bh);
+	if (ret)
+		mlog_errno(ret);
+
+	ocfs2_commit_trans(osb, ctxt.handle);
+out:
+	ocfs2_free_alloc_context(ctxt.meta_ac);
+	return ret;
+}
+
+static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
+				     struct buffer_head *blk_bh,
+				     struct buffer_head *new_blk_bh)
+{
+	int ret = 0, credits = 0;
+	handle_t *handle;
+	struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
+	struct ocfs2_dinode *new_di;
+	struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
+	int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
+	struct ocfs2_xattr_block *xb =
+			(struct ocfs2_xattr_block *)blk_bh->b_data;
+	struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
+	struct ocfs2_xattr_block *new_xb =
+			(struct ocfs2_xattr_block *)new_blk_bh->b_data;
+	struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
+	struct ocfs2_alloc_context *meta_ac;
+	struct ocfs2_xattr_value_buf vb = {
+		.vb_bh = new_blk_bh,
+		.vb_access = ocfs2_journal_access_xb,
+	};
+
+	ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
+						  &credits, &meta_ac);
+	if (ret) {
+		mlog_errno(ret);
+		return ret;
+	}
+
+	/* One more credits in case we need to add xattr flags in new inode. */
+	handle = ocfs2_start_trans(osb, credits + 1);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
+		ret = ocfs2_journal_access_di(handle,
+					      INODE_CACHE(args->new_inode),
+					      args->new_bh,
+					      OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_commit;
+		}
+	}
+
+	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
+				      new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
+	       osb->sb->s_blocksize - header_off);
+
+	ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
+					 new_blk_bh, new_xh, &vb, meta_ac,
+					 ocfs2_get_xattr_value_root, NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto out_commit;
+	}
+
+	ocfs2_journal_dirty(handle, new_blk_bh);
+
+	if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
+		new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
+		spin_lock(&new_oi->ip_lock);
+		new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
+		new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
+		spin_unlock(&new_oi->ip_lock);
+
+		ocfs2_journal_dirty(handle, args->new_bh);
+	}
+
+out_commit:
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	ocfs2_free_alloc_context(meta_ac);
+	return ret;
+}
+
+struct ocfs2_reflink_xattr_tree_args {
+	struct ocfs2_xattr_reflink *reflink;
+	struct buffer_head *old_blk_bh;
+	struct buffer_head *new_blk_bh;
+	struct ocfs2_xattr_bucket *old_bucket;
+	struct ocfs2_xattr_bucket *new_bucket;
+};
+
+/*
+ * NOTE:
+ * We have to handle the case that both old bucket and new bucket
+ * will call this function to get the right ret_bh.
+ * So The caller must give us the right bh.
+ */
+static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
+					struct buffer_head *bh,
+					struct ocfs2_xattr_header *xh,
+					int offset,
+					struct ocfs2_xattr_value_root **xv,
+					struct buffer_head **ret_bh,
+					void *para)
+{
+	struct ocfs2_reflink_xattr_tree_args *args =
+			(struct ocfs2_reflink_xattr_tree_args *)para;
+	struct ocfs2_xattr_bucket *bucket;
+
+	if (bh == args->old_bucket->bu_bhs[0])
+		bucket = args->old_bucket;
+	else
+		bucket = args->new_bucket;
+
+	return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
+					       xv, ret_bh);
+}
+
+struct ocfs2_value_tree_metas {
+	int num_metas;
+	int credits;
+	int num_recs;
+};
+
+static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
+					struct buffer_head *bh,
+					struct ocfs2_xattr_header *xh,
+					int offset,
+					struct ocfs2_xattr_value_root **xv,
+					struct buffer_head **ret_bh,
+					void *para)
+{
+	struct ocfs2_xattr_bucket *bucket =
+				(struct ocfs2_xattr_bucket *)para;
+
+	return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
+					       xv, ret_bh);
+}
+
+static int ocfs2_calc_value_tree_metas(struct inode *inode,
+				      struct ocfs2_xattr_bucket *bucket,
+				      void *para)
+{
+	struct ocfs2_value_tree_metas *metas =
+			(struct ocfs2_value_tree_metas *)para;
+	struct ocfs2_xattr_header *xh =
+			(struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
+
+	/* Add the credits for this bucket first. */
+	metas->credits += bucket->bu_blocks;
+	return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
+					xh, &metas->num_metas,
+					&metas->credits, &metas->num_recs,
+					ocfs2_value_tree_metas_in_bucket,
+					bucket);
+}
+
+/*
+ * Given a xattr extent rec starting from blkno and having len clusters,
+ * iterate all the buckets calculate how much metadata we need for reflinking
+ * all the ocfs2_xattr_value_root and lock the allocators accordingly.
+ */
+static int ocfs2_lock_reflink_xattr_rec_allocators(
+				struct ocfs2_reflink_xattr_tree_args *args,
+				struct ocfs2_extent_tree *xt_et,
+				u64 blkno, u32 len, int *credits,
+				struct ocfs2_alloc_context **meta_ac,
+				struct ocfs2_alloc_context **data_ac)
+{
+	int ret, num_free_extents;
+	struct ocfs2_value_tree_metas metas;
+	struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
+	struct ocfs2_refcount_block *rb;
+
+	memset(&metas, 0, sizeof(metas));
+
+	ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
+					  ocfs2_calc_value_tree_metas, &metas);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	*credits = metas.credits;
+
+	/*
+	 * Calculate we need for refcount tree change.
+	 *
+	 * We need to add/modify num_recs in refcount tree, so just calculate
+	 * an approximate number we need for refcount tree change.
+	 * Sometimes we need to split the tree, and after split,  half recs
+	 * will be moved to the new block, and a new block can only provide
+	 * half number of recs. So we multiple new blocks by 2.
+	 * In the end, we have to add credits for modifying the already
+	 * existed refcount block.
+	 */
+	rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
+	metas.num_recs =
+		(metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
+		 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
+	metas.num_metas += metas.num_recs;
+	*credits += metas.num_recs +
+		    metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
+	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
+		*credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
+			    le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
+	else
+		*credits += 1;
+
+	/* count in the xattr tree change. */
+	num_free_extents = ocfs2_num_free_extents(osb, xt_et);
+	if (num_free_extents < 0) {
+		ret = num_free_extents;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (num_free_extents < len)
+		metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
+
+	*credits += ocfs2_calc_extend_credits(osb->sb,
+					      xt_et->et_root_el);
+
+	if (metas.num_metas) {
+		ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
+							meta_ac);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+	}
+
+	if (len) {
+		ret = ocfs2_reserve_clusters(osb, len, data_ac);
+		if (ret)
+			mlog_errno(ret);
+	}
+out:
+	if (ret) {
+		if (*meta_ac) {
+			ocfs2_free_alloc_context(*meta_ac);
+			*meta_ac = NULL;
+		}
+	}
+
+	return ret;
+}
+
+static int ocfs2_reflink_xattr_bucket(handle_t *handle,
+				u64 blkno, u64 new_blkno, u32 clusters,
+				u32 *cpos, int num_buckets,
+				struct ocfs2_alloc_context *meta_ac,
+				struct ocfs2_alloc_context *data_ac,
+				struct ocfs2_reflink_xattr_tree_args *args)
+{
+	int i, j, ret = 0;
+	struct super_block *sb = args->reflink->old_inode->i_sb;
+	int bpb = args->old_bucket->bu_blocks;
+	struct ocfs2_xattr_value_buf vb = {
+		.vb_access = ocfs2_journal_access,
+	};
+
+	for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
+		ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		ret = ocfs2_xattr_bucket_journal_access(handle,
+						args->new_bucket,
+						OCFS2_JOURNAL_ACCESS_CREATE);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		for (j = 0; j < bpb; j++)
+			memcpy(bucket_block(args->new_bucket, j),
+			       bucket_block(args->old_bucket, j),
+			       sb->s_blocksize);
+
+		/*
+		 * Record the start cpos so that we can use it to initialize
+		 * our xattr tree we also set the xh_num_bucket for the new
+		 * bucket.
+		 */
+		if (i == 0) {
+			*cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
+					    xh_entries[0].xe_name_hash);
+			bucket_xh(args->new_bucket)->xh_num_buckets =
+				cpu_to_le16(num_buckets);
+		}
+
+		ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
+
+		ret = ocfs2_reflink_xattr_header(handle, args->reflink,
+					args->old_bucket->bu_bhs[0],
+					bucket_xh(args->old_bucket),
+					args->new_bucket->bu_bhs[0],
+					bucket_xh(args->new_bucket),
+					&vb, meta_ac,
+					ocfs2_get_reflink_xattr_value_root,
+					args);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		/*
+		 * Re-access and dirty the bucket to calculate metaecc.
+		 * Because we may extend the transaction in reflink_xattr_header
+		 * which will let the already accessed block gone.
+		 */
+		ret = ocfs2_xattr_bucket_journal_access(handle,
+						args->new_bucket,
+						OCFS2_JOURNAL_ACCESS_WRITE);
+		if (ret) {
+			mlog_errno(ret);
+			break;
+		}
+
+		ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
+
+		ocfs2_xattr_bucket_relse(args->old_bucket);
+		ocfs2_xattr_bucket_relse(args->new_bucket);
+	}
+
+	ocfs2_xattr_bucket_relse(args->old_bucket);
+	ocfs2_xattr_bucket_relse(args->new_bucket);
+	return ret;
+}
+
+static int ocfs2_reflink_xattr_buckets(handle_t *handle,
+				struct inode *inode,
+				struct ocfs2_reflink_xattr_tree_args *args,
+				struct ocfs2_extent_tree *et,
+				struct ocfs2_alloc_context *meta_ac,
+				struct ocfs2_alloc_context *data_ac,
+				u64 blkno, u32 cpos, u32 len)
+{
+	int ret, first_inserted = 0;
+	u32 p_cluster, num_clusters, reflink_cpos = 0;
+	u64 new_blkno;
+	unsigned int num_buckets, reflink_buckets;
+	unsigned int bpc =
+		ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
+
+	ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+	num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
+	ocfs2_xattr_bucket_relse(args->old_bucket);
+
+	while (len && num_buckets) {
+		ret = ocfs2_claim_clusters(handle, data_ac,
+					   1, &p_cluster, &num_clusters);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
+		reflink_buckets = min(num_buckets, bpc * num_clusters);
+
+		ret = ocfs2_reflink_xattr_bucket(handle, blkno,
+						 new_blkno, num_clusters,
+						 &reflink_cpos, reflink_buckets,
+						 meta_ac, data_ac, args);
+		if (ret) {
+			mlog_errno(ret);
+			goto out;
+		}
+
+		/*
+		 * For the 1st allocated cluster, we make it use the same cpos
+		 * so that the xattr tree looks the same as the original one
+		 * in the most case.
+		 */
+		if (!first_inserted) {
+			reflink_cpos = cpos;
+			first_inserted = 1;
+		}
+		ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
+					  num_clusters, 0, meta_ac);
+		if (ret)
+			mlog_errno(ret);
+
+		trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
+						  num_clusters, reflink_cpos);
+
+		len -= num_clusters;
+		blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
+		num_buckets -= reflink_buckets;
+	}
+out:
+	return ret;
+}
+
+/*
+ * Create the same xattr extent record in the new inode's xattr tree.
+ */
+static int ocfs2_reflink_xattr_rec(struct inode *inode,
+				   struct buffer_head *root_bh,
+				   u64 blkno,
+				   u32 cpos,
+				   u32 len,
+				   void *para)
+{
+	int ret, credits = 0;
+	handle_t *handle;
+	struct ocfs2_reflink_xattr_tree_args *args =
+			(struct ocfs2_reflink_xattr_tree_args *)para;
+	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
+	struct ocfs2_alloc_context *meta_ac = NULL;
+	struct ocfs2_alloc_context *data_ac = NULL;
+	struct ocfs2_extent_tree et;
+
+	trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
+
+	ocfs2_init_xattr_tree_extent_tree(&et,
+					  INODE_CACHE(args->reflink->new_inode),
+					  args->new_blk_bh);
+
+	ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
+						      len, &credits,
+						      &meta_ac, &data_ac);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	handle = ocfs2_start_trans(osb, credits);
+	if (IS_ERR(handle)) {
+		ret = PTR_ERR(handle);
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
+					  meta_ac, data_ac,
+					  blkno, cpos, len);
+	if (ret)
+		mlog_errno(ret);
+
+	ocfs2_commit_trans(osb, handle);
+
+out:
+	if (meta_ac)
+		ocfs2_free_alloc_context(meta_ac);
+	if (data_ac)
+		ocfs2_free_alloc_context(data_ac);
+	return ret;
+}
+
+/*
+ * Create reflinked xattr buckets.
+ * We will add bucket one by one, and refcount all the xattrs in the bucket
+ * if they are stored outside.
+ */
+static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
+				    struct buffer_head *blk_bh,
+				    struct buffer_head *new_blk_bh)
+{
+	int ret;
+	struct ocfs2_reflink_xattr_tree_args para;
+
+	memset(&para, 0, sizeof(para));
+	para.reflink = args;
+	para.old_blk_bh = blk_bh;
+	para.new_blk_bh = new_blk_bh;
+
+	para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
+	if (!para.old_bucket) {
+		mlog_errno(-ENOMEM);
+		return -ENOMEM;
+	}
+
+	para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
+	if (!para.new_bucket) {
+		ret = -ENOMEM;
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
+					      ocfs2_reflink_xattr_rec,
+					      &para);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	ocfs2_xattr_bucket_free(para.old_bucket);
+	ocfs2_xattr_bucket_free(para.new_bucket);
+	return ret;
+}
+
+static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
+					struct buffer_head *blk_bh)
+{
+	int ret, indexed = 0;
+	struct buffer_head *new_blk_bh = NULL;
+	struct ocfs2_xattr_block *xb =
+			(struct ocfs2_xattr_block *)blk_bh->b_data;
+
+
+	if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
+		indexed = 1;
+
+	ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
+					     &new_blk_bh, indexed);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	if (!indexed)
+		ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
+	else
+		ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
+	if (ret)
+		mlog_errno(ret);
+
+out:
+	brelse(new_blk_bh);
+	return ret;
+}
+
+static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
+{
+	int type = ocfs2_xattr_get_type(xe);
+
+	return type != OCFS2_XATTR_INDEX_SECURITY &&
+	       type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
+	       type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
+}
+
+int ocfs2_reflink_xattrs(struct inode *old_inode,
+			 struct buffer_head *old_bh,
+			 struct inode *new_inode,
+			 struct buffer_head *new_bh,
+			 bool preserve_security)
+{
+	int ret;
+	struct ocfs2_xattr_reflink args;
+	struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
+	struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
+	struct buffer_head *blk_bh = NULL;
+	struct ocfs2_cached_dealloc_ctxt dealloc;
+	struct ocfs2_refcount_tree *ref_tree;
+	struct buffer_head *ref_root_bh = NULL;
+
+	ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
+				       le64_to_cpu(di->i_refcount_loc),
+				       1, &ref_tree, &ref_root_bh);
+	if (ret) {
+		mlog_errno(ret);
+		goto out;
+	}
+
+	ocfs2_init_dealloc_ctxt(&dealloc);
+
+	args.old_inode = old_inode;
+	args.new_inode = new_inode;
+	args.old_bh = old_bh;
+	args.new_bh = new_bh;
+	args.ref_ci = &ref_tree->rf_ci;
+	args.ref_root_bh = ref_root_bh;
+	args.dealloc = &dealloc;
+	if (preserve_security)
+		args.xattr_reflinked = NULL;
+	else
+		args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
+
+	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
+		ret = ocfs2_reflink_xattr_inline(&args);
+		if (ret) {
+			mlog_errno(ret);
+			goto out_unlock;
+		}
+	}
+
+	if (!di->i_xattr_loc)
+		goto out_unlock;
+
+	ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
+				     &blk_bh);
+	if (ret < 0) {
+		mlog_errno(ret);
+		goto out_unlock;
+	}
+
+	ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
+	if (ret)
+		mlog_errno(ret);
+
+	brelse(blk_bh);
+
+out_unlock:
+	ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
+				   ref_tree, 1);
+	brelse(ref_root_bh);
+
+	if (ocfs2_dealloc_has_cluster(&dealloc)) {
+		ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
+		ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
+	}
+
+out:
+	return ret;
+}
+
+/*
+ * Initialize security and acl for a already created inode.
+ * Used for reflink a non-preserve-security file.
+ *
+ * It uses common api like ocfs2_xattr_set, so the caller
+ * must not hold any lock expect i_mutex.
+ */
+int ocfs2_init_security_and_acl(struct inode *dir,
+				struct inode *inode,
+				const struct qstr *qstr,
+				struct posix_acl *default_acl,
+				struct posix_acl *acl)
+{
+	struct buffer_head *dir_bh = NULL;
+	int ret = 0;
+
+	ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
+	if (ret) {
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	ret = ocfs2_inode_lock(dir, &dir_bh, 0);
+	if (ret) {
+		mlog_errno(ret);
+		goto leave;
+	}
+
+	if (!ret && default_acl)
+		ret = ocfs2_iop_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
+	if (!ret && acl)
+		ret = ocfs2_iop_set_acl(inode, acl, ACL_TYPE_ACCESS);
+
+	ocfs2_inode_unlock(dir, 0);
+	brelse(dir_bh);
+leave:
+	return ret;
+}
+/*
+ * 'security' attributes support
+ */
+static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
+					size_t list_size, const char *name,
+					size_t name_len, int type)
+{
+	const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
+	const size_t total_len = prefix_len + name_len + 1;
+
+	if (list && total_len <= list_size) {
+		memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
+		memcpy(list + prefix_len, name, name_len);
+		list[prefix_len + name_len] = '\0';
+	}
+	return total_len;
+}
+
+static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
+				    void *buffer, size_t size, int type)
+{
+	if (strcmp(name, "") == 0)
+		return -EINVAL;
+	return ocfs2_xattr_get(d_inode(dentry), OCFS2_XATTR_INDEX_SECURITY,
+			       name, buffer, size);
+}
+
+static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
+		const void *value, size_t size, int flags, int type)
+{
+	if (strcmp(name, "") == 0)
+		return -EINVAL;
+
+	return ocfs2_xattr_set(d_inode(dentry), OCFS2_XATTR_INDEX_SECURITY,
+			       name, value, size, flags);
+}
+
+int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
+		     void *fs_info)
+{
+	const struct xattr *xattr;
+	int err = 0;
+
+	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
+		err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
+				      xattr->name, xattr->value,
+				      xattr->value_len, XATTR_CREATE);
+		if (err)
+			break;
+	}
+	return err;
+}
+
+int ocfs2_init_security_get(struct inode *inode,
+			    struct inode *dir,
+			    const struct qstr *qstr,
+			    struct ocfs2_security_xattr_info *si)
+{
+	/* check whether ocfs2 support feature xattr */
+	if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
+		return -EOPNOTSUPP;
+	if (si)
+		return security_old_inode_init_security(inode, dir, qstr,
+							&si->name, &si->value,
+							&si->value_len);
+
+	return security_inode_init_security(inode, dir, qstr,
+					    &ocfs2_initxattrs, NULL);
+}
+
+int ocfs2_init_security_set(handle_t *handle,
+			    struct inode *inode,
+			    struct buffer_head *di_bh,
+			    struct ocfs2_security_xattr_info *si,
+			    struct ocfs2_alloc_context *xattr_ac,
+			    struct ocfs2_alloc_context *data_ac)
+{
+	return ocfs2_xattr_set_handle(handle, inode, di_bh,
+				     OCFS2_XATTR_INDEX_SECURITY,
+				     si->name, si->value, si->value_len, 0,
+				     xattr_ac, data_ac);
+}
+
+const struct xattr_handler ocfs2_xattr_security_handler = {
+	.prefix	= XATTR_SECURITY_PREFIX,
+	.list	= ocfs2_xattr_security_list,
+	.get	= ocfs2_xattr_security_get,
+	.set	= ocfs2_xattr_security_set,
+};
+
+/*
+ * 'trusted' attributes support
+ */
+static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
+				       size_t list_size, const char *name,
+				       size_t name_len, int type)
+{
+	const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
+	const size_t total_len = prefix_len + name_len + 1;
+
+	if (list && total_len <= list_size) {
+		memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
+		memcpy(list + prefix_len, name, name_len);
+		list[prefix_len + name_len] = '\0';
+	}
+	return total_len;
+}
+
+static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
+		void *buffer, size_t size, int type)
+{
+	if (strcmp(name, "") == 0)
+		return -EINVAL;
+	return ocfs2_xattr_get(d_inode(dentry), OCFS2_XATTR_INDEX_TRUSTED,
+			       name, buffer, size);
+}
+
+static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
+		const void *value, size_t size, int flags, int type)
+{
+	if (strcmp(name, "") == 0)
+		return -EINVAL;
+
+	return ocfs2_xattr_set(d_inode(dentry), OCFS2_XATTR_INDEX_TRUSTED,
+			       name, value, size, flags);
+}
+
+const struct xattr_handler ocfs2_xattr_trusted_handler = {
+	.prefix	= XATTR_TRUSTED_PREFIX,
+	.list	= ocfs2_xattr_trusted_list,
+	.get	= ocfs2_xattr_trusted_get,
+	.set	= ocfs2_xattr_trusted_set,
+};
+
+/*
+ * 'user' attributes support
+ */
+static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
+				    size_t list_size, const char *name,
+				    size_t name_len, int type)
+{
+	const size_t prefix_len = XATTR_USER_PREFIX_LEN;
+	const size_t total_len = prefix_len + name_len + 1;
+	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
+
+	if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
+		return 0;
+
+	if (list && total_len <= list_size) {
+		memcpy(list, XATTR_USER_PREFIX, prefix_len);
+		memcpy(list + prefix_len, name, name_len);
+		list[prefix_len + name_len] = '\0';
+	}
+	return total_len;
+}
+
+static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
+		void *buffer, size_t size, int type)
+{
+	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
+
+	if (strcmp(name, "") == 0)
+		return -EINVAL;
+	if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
+		return -EOPNOTSUPP;
+	return ocfs2_xattr_get(d_inode(dentry), OCFS2_XATTR_INDEX_USER, name,
+			       buffer, size);
+}
+
+static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
+		const void *value, size_t size, int flags, int type)
+{
+	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
+
+	if (strcmp(name, "") == 0)
+		return -EINVAL;
+	if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
+		return -EOPNOTSUPP;
+
+	return ocfs2_xattr_set(d_inode(dentry), OCFS2_XATTR_INDEX_USER,
+			       name, value, size, flags);
+}
+
+const struct xattr_handler ocfs2_xattr_user_handler = {
+	.prefix	= XATTR_USER_PREFIX,
+	.list	= ocfs2_xattr_user_list,
+	.get	= ocfs2_xattr_user_get,
+	.set	= ocfs2_xattr_user_set,
+};
diff --git a/fs/ocfs2/xattr.h b/fs/ocfs2/xattr.h
new file mode 100644
index 000000000..f10d5b93c
--- /dev/null
+++ b/fs/ocfs2/xattr.h
@@ -0,0 +1,100 @@
+/* -*- mode: c; c-basic-offset: 8; -*-
+ * vim: noexpandtab sw=8 ts=8 sts=0:
+ *
+ * xattr.h
+ *
+ * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+
+#ifndef OCFS2_XATTR_H
+#define OCFS2_XATTR_H
+
+#include <linux/init.h>
+#include <linux/xattr.h>
+
+enum ocfs2_xattr_type {
+	OCFS2_XATTR_INDEX_USER = 1,
+	OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS,
+	OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
+	OCFS2_XATTR_INDEX_TRUSTED,
+	OCFS2_XATTR_INDEX_SECURITY,
+	OCFS2_XATTR_MAX
+};
+
+struct ocfs2_security_xattr_info {
+	int enable;
+	const char *name;
+	void *value;
+	size_t value_len;
+};
+
+extern const struct xattr_handler ocfs2_xattr_user_handler;
+extern const struct xattr_handler ocfs2_xattr_trusted_handler;
+extern const struct xattr_handler ocfs2_xattr_security_handler;
+extern const struct xattr_handler *ocfs2_xattr_handlers[];
+
+ssize_t ocfs2_listxattr(struct dentry *, char *, size_t);
+int ocfs2_xattr_get_nolock(struct inode *, struct buffer_head *, int,
+			   const char *, void *, size_t);
+int ocfs2_xattr_set(struct inode *, int, const char *, const void *,
+		    size_t, int);
+int ocfs2_xattr_set_handle(handle_t *, struct inode *, struct buffer_head *,
+			   int, const char *, const void *, size_t, int,
+			   struct ocfs2_alloc_context *,
+			   struct ocfs2_alloc_context *);
+int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
+					 struct ocfs2_dinode *di);
+int ocfs2_xattr_remove(struct inode *, struct buffer_head *);
+int ocfs2_init_security_get(struct inode *, struct inode *,
+			    const struct qstr *,
+			    struct ocfs2_security_xattr_info *);
+int ocfs2_init_security_set(handle_t *, struct inode *,
+			    struct buffer_head *,
+			    struct ocfs2_security_xattr_info *,
+			    struct ocfs2_alloc_context *,
+			    struct ocfs2_alloc_context *);
+int ocfs2_calc_security_init(struct inode *,
+			     struct ocfs2_security_xattr_info *,
+			     int *, int *, struct ocfs2_alloc_context **);
+int ocfs2_calc_xattr_init(struct inode *, struct buffer_head *,
+			  umode_t, struct ocfs2_security_xattr_info *,
+			  int *, int *, int *);
+
+/*
+ * xattrs can live inside an inode, as part of an external xattr block,
+ * or inside an xattr bucket, which is the leaf of a tree rooted in an
+ * xattr block.  Some of the xattr calls, especially the value setting
+ * functions, want to treat each of these locations as equal.  Let's wrap
+ * them in a structure that we can pass around instead of raw buffer_heads.
+ */
+struct ocfs2_xattr_value_buf {
+	struct buffer_head		*vb_bh;
+	ocfs2_journal_access_func	vb_access;
+	struct ocfs2_xattr_value_root	*vb_xv;
+};
+
+int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
+				     struct buffer_head *fe_bh,
+				     struct ocfs2_caching_info *ref_ci,
+				     struct buffer_head *ref_root_bh,
+				     struct ocfs2_cached_dealloc_ctxt *dealloc);
+int ocfs2_reflink_xattrs(struct inode *old_inode,
+			 struct buffer_head *old_bh,
+			 struct inode *new_inode,
+			 struct buffer_head *new_bh,
+			 bool preserve_security);
+int ocfs2_init_security_and_acl(struct inode *dir,
+				struct inode *inode,
+				const struct qstr *qstr,
+				struct posix_acl *default_acl,
+				struct posix_acl *acl);
+#endif /* OCFS2_XATTR_H */