/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/

/***
  This file is part of systemd.

  Copyright 2014 Lennart Poettering

  systemd is free software; you can redistribute it and/or modify it
  under the terms of the GNU Lesser General Public License as published by
  the Free Software Foundation; either version 2.1 of the License, or
  (at your option) any later version.

  systemd 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
  Lesser General Public License for more details.

  You should have received a copy of the GNU Lesser General Public License
  along with systemd; If not, see <http://www.gnu.org/licenses/>.
***/

#include <stdlib.h>
#include <sys/vfs.h>
#include <sys/stat.h>

#ifdef HAVE_LINUX_BTRFS_H
#include <linux/btrfs.h>
#endif

#include "missing.h"
#include "util.h"
#include "path-util.h"
#include "macro.h"
#include "copy.h"
#include "selinux-util.h"
#include "smack-util.h"
#include "fileio.h"
#include "btrfs-ctree.h"
#include "btrfs-util.h"

/* WARNING: Be careful with file system ioctls! When we get an fd, we
 * need to make sure it either refers to only a regular file or
 * directory, or that it is located on btrfs, before invoking any
 * btrfs ioctls. The ioctl numbers are reused by some device drivers
 * (such as DRM), and hence might have bad effects when invoked on
 * device nodes (that reference drivers) rather than fds to normal
 * files or directories. */

static int validate_subvolume_name(const char *name) {

        if (!filename_is_valid(name))
                return -EINVAL;

        if (strlen(name) > BTRFS_SUBVOL_NAME_MAX)
                return -E2BIG;

        return 0;
}

static int open_parent(const char *path, int flags) {
        _cleanup_free_ char *parent = NULL;
        int r, fd;

        assert(path);

        r = path_get_parent(path, &parent);
        if (r < 0)
                return r;

        fd = open(parent, flags);
        if (fd < 0)
                return -errno;

        return fd;
}

static int extract_subvolume_name(const char *path, const char **subvolume) {
        const char *fn;
        int r;

        assert(path);
        assert(subvolume);

        fn = basename(path);

        r = validate_subvolume_name(fn);
        if (r < 0)
                return r;

        *subvolume = fn;
        return 0;
}

int btrfs_is_filesystem(int fd) {
        struct statfs sfs;

        assert(fd >= 0);

        if (fstatfs(fd, &sfs) < 0)
                return -errno;

        return F_TYPE_EQUAL(sfs.f_type, BTRFS_SUPER_MAGIC);
}

int btrfs_is_subvol(int fd) {
        struct stat st;

        assert(fd >= 0);

        /* On btrfs subvolumes always have the inode 256 */

        if (fstat(fd, &st) < 0)
                return -errno;

        if (!S_ISDIR(st.st_mode) || st.st_ino != 256)
                return 0;

        return btrfs_is_filesystem(fd);
}

int btrfs_subvol_make(const char *path) {
        struct btrfs_ioctl_vol_args args = {};
        _cleanup_close_ int fd = -1;
        const char *subvolume;
        int r;

        assert(path);

        r = extract_subvolume_name(path, &subvolume);
        if (r < 0)
                return r;

        fd = open_parent(path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
        if (fd < 0)
                return fd;

        strncpy(args.name, subvolume, sizeof(args.name)-1);

        if (ioctl(fd, BTRFS_IOC_SUBVOL_CREATE, &args) < 0)
                return -errno;

        return 0;
}

int btrfs_subvol_make_label(const char *path) {
        int r;

        assert(path);

        r = mac_selinux_create_file_prepare(path, S_IFDIR);
        if (r < 0)
                return r;

        r = btrfs_subvol_make(path);
        mac_selinux_create_file_clear();

        if (r < 0)
                return r;

        return mac_smack_fix(path, false, false);
}

int btrfs_subvol_set_read_only_fd(int fd, bool b) {
        uint64_t flags, nflags;
        struct stat st;

        assert(fd >= 0);

        if (fstat(fd, &st) < 0)
                return -errno;

        if (!S_ISDIR(st.st_mode) || st.st_ino != 256)
                return -EINVAL;

        if (ioctl(fd, BTRFS_IOC_SUBVOL_GETFLAGS, &flags) < 0)
                return -errno;

        if (b)
                nflags = flags | BTRFS_SUBVOL_RDONLY;
        else
                nflags = flags & ~BTRFS_SUBVOL_RDONLY;

        if (flags == nflags)
                return 0;

        if (ioctl(fd, BTRFS_IOC_SUBVOL_SETFLAGS, &nflags) < 0)
                return -errno;

        return 0;
}

int btrfs_subvol_set_read_only(const char *path, bool b) {
        _cleanup_close_ int fd = -1;

        fd = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
        if (fd < 0)
                return -errno;

        return btrfs_subvol_set_read_only_fd(fd, b);
}

int btrfs_subvol_get_read_only_fd(int fd) {
        uint64_t flags;
        struct stat st;

        assert(fd >= 0);

        if (fstat(fd, &st) < 0)
                return -errno;

        if (!S_ISDIR(st.st_mode) || st.st_ino != 256)
                return -EINVAL;

        if (ioctl(fd, BTRFS_IOC_SUBVOL_GETFLAGS, &flags) < 0)
                return -errno;

        return !!(flags & BTRFS_SUBVOL_RDONLY);
}

int btrfs_reflink(int infd, int outfd) {
        struct stat st;
        int r;

        assert(infd >= 0);
        assert(outfd >= 0);

        /* Make sure we invoke the ioctl on a regular file, so that no
         * device driver accidentally gets it. */

        if (fstat(outfd, &st) < 0)
                return -errno;

        if (!S_ISREG(st.st_mode))
                return -EINVAL;

        r = ioctl(outfd, BTRFS_IOC_CLONE, infd);
        if (r < 0)
                return -errno;

        return 0;
}

int btrfs_clone_range(int infd, uint64_t in_offset, int outfd, uint64_t out_offset, uint64_t sz) {
        struct btrfs_ioctl_clone_range_args args = {
                .src_fd = infd,
                .src_offset = in_offset,
                .src_length = sz,
                .dest_offset = out_offset,
        };
        struct stat st;
        int r;

        assert(infd >= 0);
        assert(outfd >= 0);
        assert(sz > 0);

        if (fstat(outfd, &st) < 0)
                return -errno;

        if (!S_ISREG(st.st_mode))
                return -EINVAL;

        r = ioctl(outfd, BTRFS_IOC_CLONE_RANGE, &args);
        if (r < 0)
                return -errno;

        return 0;
}

int btrfs_get_block_device_fd(int fd, dev_t *dev) {
        struct btrfs_ioctl_fs_info_args fsi = {};
        uint64_t id;
        int r;

        assert(fd >= 0);
        assert(dev);

        r = btrfs_is_filesystem(fd);
        if (r < 0)
                return r;
        if (!r)
                return -ENOTTY;

        if (ioctl(fd, BTRFS_IOC_FS_INFO, &fsi) < 0)
                return -errno;

        /* We won't do this for btrfs RAID */
        if (fsi.num_devices != 1)
                return 0;

        for (id = 1; id <= fsi.max_id; id++) {
                struct btrfs_ioctl_dev_info_args di = {
                        .devid = id,
                };
                struct stat st;

                if (ioctl(fd, BTRFS_IOC_DEV_INFO, &di) < 0) {
                        if (errno == ENODEV)
                                continue;

                        return -errno;
                }

                if (stat((char*) di.path, &st) < 0)
                        return -errno;

                if (!S_ISBLK(st.st_mode))
                        return -ENODEV;

                if (major(st.st_rdev) == 0)
                        return -ENODEV;

                *dev = st.st_rdev;
                return 1;
        }

        return -ENODEV;
}

int btrfs_get_block_device(const char *path, dev_t *dev) {
        _cleanup_close_ int fd = -1;

        assert(path);
        assert(dev);

        fd = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC);
        if (fd < 0)
                return -errno;

        return btrfs_get_block_device_fd(fd, dev);
}

int btrfs_subvol_get_id_fd(int fd, uint64_t *ret) {
        struct btrfs_ioctl_ino_lookup_args args = {
                .objectid = BTRFS_FIRST_FREE_OBJECTID
        };
        int r;

        assert(fd >= 0);
        assert(ret);

        r = btrfs_is_filesystem(fd);
        if (r < 0)
                return r;
        if (!r)
                return -ENOTTY;

        if (ioctl(fd, BTRFS_IOC_INO_LOOKUP, &args) < 0)
                return -errno;

        *ret = args.treeid;
        return 0;
}

int btrfs_subvol_get_id(int fd, const char *subvol, uint64_t *ret) {
        _cleanup_close_ int subvol_fd = -1;

        assert(fd >= 0);
        assert(ret);

        subvol_fd = openat(fd, subvol, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
        if (subvol_fd < 0)
                return -errno;

        return btrfs_subvol_get_id_fd(subvol_fd, ret);
}

static bool btrfs_ioctl_search_args_inc(struct btrfs_ioctl_search_args *args) {
        assert(args);

        /* the objectid, type, offset together make up the btrfs key,
         * which is considered a single 136byte integer when
         * comparing. This call increases the counter by one, dealing
         * with the overflow between the overflows */

        if (args->key.min_offset < (uint64_t) -1) {
                args->key.min_offset++;
                return true;
        }

        if (args->key.min_type < (uint8_t) -1) {
                args->key.min_type++;
                args->key.min_offset = 0;
                return true;
        }

        if (args->key.min_objectid < (uint64_t) -1) {
                args->key.min_objectid++;
                args->key.min_offset = 0;
                args->key.min_type = 0;
                return true;
        }

        return 0;
}

static void btrfs_ioctl_search_args_set(struct btrfs_ioctl_search_args *args, const struct btrfs_ioctl_search_header *h) {
        assert(args);
        assert(h);

        args->key.min_objectid = h->objectid;
        args->key.min_type = h->type;
        args->key.min_offset = h->offset;
}

static int btrfs_ioctl_search_args_compare(const struct btrfs_ioctl_search_args *args) {
        assert(args);

        /* Compare min and max */

        if (args->key.min_objectid < args->key.max_objectid)
                return -1;
        if (args->key.min_objectid > args->key.max_objectid)
                return 1;

        if (args->key.min_type < args->key.max_type)
                return -1;
        if (args->key.min_type > args->key.max_type)
                return 1;

        if (args->key.min_offset < args->key.max_offset)
                return -1;
        if (args->key.min_offset > args->key.max_offset)
                return 1;

        return 0;
}

#define FOREACH_BTRFS_IOCTL_SEARCH_HEADER(i, sh, args)                  \
        for ((i) = 0,                                                   \
             (sh) = (const struct btrfs_ioctl_search_header*) (args).buf; \
             (i) < (args).key.nr_items;                                 \
             (i)++,                                                     \
             (sh) = (const struct btrfs_ioctl_search_header*) ((uint8_t*) (sh) + sizeof(struct btrfs_ioctl_search_header) + (sh)->len))

#define BTRFS_IOCTL_SEARCH_HEADER_BODY(sh)                              \
        ((void*) ((uint8_t*) sh + sizeof(struct btrfs_ioctl_search_header)))

int btrfs_subvol_get_info_fd(int fd, BtrfsSubvolInfo *ret) {
        struct btrfs_ioctl_search_args args = {
                /* Tree of tree roots */
                .key.tree_id = BTRFS_ROOT_TREE_OBJECTID,

                /* Look precisely for the subvolume items */
                .key.min_type = BTRFS_ROOT_ITEM_KEY,
                .key.max_type = BTRFS_ROOT_ITEM_KEY,

                .key.min_offset = 0,
                .key.max_offset = (uint64_t) -1,

                /* No restrictions on the other components */
                .key.min_transid = 0,
                .key.max_transid = (uint64_t) -1,
        };

        uint64_t subvol_id;
        bool found = false;
        int r;

        assert(fd >= 0);
        assert(ret);

        r = btrfs_subvol_get_id_fd(fd, &subvol_id);
        if (r < 0)
                return r;

        args.key.min_objectid = args.key.max_objectid = subvol_id;

        while (btrfs_ioctl_search_args_compare(&args) <= 0) {
                const struct btrfs_ioctl_search_header *sh;
                unsigned i;

                args.key.nr_items = 256;
                if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0)
                        return -errno;

                if (args.key.nr_items <= 0)
                        break;

                FOREACH_BTRFS_IOCTL_SEARCH_HEADER(i, sh, args) {

                        const struct btrfs_root_item *ri;

                        /* Make sure we start the next search at least from this entry */
                        btrfs_ioctl_search_args_set(&args, sh);

                        if (sh->objectid != subvol_id)
                                continue;
                        if (sh->type != BTRFS_ROOT_ITEM_KEY)
                                continue;

                        /* Older versions of the struct lacked the otime setting */
                        if (sh->len < offsetof(struct btrfs_root_item, otime) + sizeof(struct btrfs_timespec))
                                continue;

                        ri = BTRFS_IOCTL_SEARCH_HEADER_BODY(sh);

                        ret->otime = (usec_t) le64toh(ri->otime.sec) * USEC_PER_SEC +
                                (usec_t) le32toh(ri->otime.nsec) / NSEC_PER_USEC;

                        ret->subvol_id = subvol_id;
                        ret->read_only = !!(le64toh(ri->flags) & BTRFS_ROOT_SUBVOL_RDONLY);

                        assert_cc(sizeof(ri->uuid) == sizeof(ret->uuid));
                        memcpy(&ret->uuid, ri->uuid, sizeof(ret->uuid));
                        memcpy(&ret->parent_uuid, ri->parent_uuid, sizeof(ret->parent_uuid));

                        found = true;
                        goto finish;
                }

                /* Increase search key by one, to read the next item, if we can. */
                if (!btrfs_ioctl_search_args_inc(&args))
                        break;
        }

finish:
        if (!found)
                return -ENODATA;

        return 0;
}

int btrfs_subvol_get_quota_fd(int fd, BtrfsQuotaInfo *ret) {

        struct btrfs_ioctl_search_args args = {
                /* Tree of quota items */
                .key.tree_id = BTRFS_QUOTA_TREE_OBJECTID,

                /* The object ID is always 0 */
                .key.min_objectid = 0,
                .key.max_objectid = 0,

                /* Look precisely for the quota items */
                .key.min_type = BTRFS_QGROUP_STATUS_KEY,
                .key.max_type = BTRFS_QGROUP_LIMIT_KEY,

                /* No restrictions on the other components */
                .key.min_transid = 0,
                .key.max_transid = (uint64_t) -1,
        };

        uint64_t subvol_id;
        bool found_info = false, found_limit = false;
        int r;

        assert(fd >= 0);
        assert(ret);

        r = btrfs_subvol_get_id_fd(fd, &subvol_id);
        if (r < 0)
                return r;

        args.key.min_offset = args.key.max_offset = subvol_id;

        while (btrfs_ioctl_search_args_compare(&args) <= 0) {
                const struct btrfs_ioctl_search_header *sh;
                unsigned i;

                args.key.nr_items = 256;
                if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0)
                        return -errno;

                if (args.key.nr_items <= 0)
                        break;

                FOREACH_BTRFS_IOCTL_SEARCH_HEADER(i, sh, args) {

                        /* Make sure we start the next search at least from this entry */
                        btrfs_ioctl_search_args_set(&args, sh);

                        if (sh->objectid != 0)
                                continue;
                        if (sh->offset != subvol_id)
                                continue;

                        if (sh->type == BTRFS_QGROUP_INFO_KEY) {
                                const struct btrfs_qgroup_info_item *qii = BTRFS_IOCTL_SEARCH_HEADER_BODY(sh);

                                ret->referenced = le64toh(qii->rfer);
                                ret->exclusive = le64toh(qii->excl);

                                found_info = true;

                        } else if (sh->type == BTRFS_QGROUP_LIMIT_KEY) {
                                const struct btrfs_qgroup_limit_item *qli = BTRFS_IOCTL_SEARCH_HEADER_BODY(sh);

                                ret->referenced_max = le64toh(qli->max_rfer);
                                ret->exclusive_max = le64toh(qli->max_excl);

                                if (ret->referenced_max == 0)
                                        ret->referenced_max = (uint64_t) -1;
                                if (ret->exclusive_max == 0)
                                        ret->exclusive_max = (uint64_t) -1;

                                found_limit = true;
                        }

                        if (found_info && found_limit)
                                goto finish;
                }

                /* Increase search key by one, to read the next item, if we can. */
                if (!btrfs_ioctl_search_args_inc(&args))
                        break;
        }

finish:
        if (!found_limit && !found_info)
                return -ENODATA;

        if (!found_info) {
                ret->referenced = (uint64_t) -1;
                ret->exclusive = (uint64_t) -1;
        }

        if (!found_limit) {
                ret->referenced_max = (uint64_t) -1;
                ret->exclusive_max = (uint64_t) -1;
        }

        return 0;
}

int btrfs_defrag_fd(int fd) {
        struct stat st;

        assert(fd >= 0);

        if (fstat(fd, &st) < 0)
                return -errno;

        if (!S_ISREG(st.st_mode))
                return -EINVAL;

        if (ioctl(fd, BTRFS_IOC_DEFRAG, NULL) < 0)
                return -errno;

        return 0;
}

int btrfs_defrag(const char *p) {
        _cleanup_close_ int fd = -1;

        fd = open(p, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
        if (fd < 0)
                return -errno;

        return btrfs_defrag_fd(fd);
}

int btrfs_quota_enable_fd(int fd, bool b) {
        struct btrfs_ioctl_quota_ctl_args args = {
                .cmd = b ? BTRFS_QUOTA_CTL_ENABLE : BTRFS_QUOTA_CTL_DISABLE,
        };
        int r;

        assert(fd >= 0);

        r = btrfs_is_filesystem(fd);
        if (r < 0)
                return r;
        if (!r)
                return -ENOTTY;

        if (ioctl(fd, BTRFS_IOC_QUOTA_CTL, &args) < 0)
                return -errno;

        return 0;
}

int btrfs_quota_enable(const char *path, bool b) {
        _cleanup_close_ int fd = -1;

        fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
        if (fd < 0)
                return -errno;

        return btrfs_quota_enable_fd(fd, b);
}

int btrfs_quota_limit_fd(int fd, uint64_t referenced_max) {
        struct btrfs_ioctl_qgroup_limit_args args = {
                .lim.max_rfer =
                        referenced_max == (uint64_t) -1 ? 0 :
                        referenced_max == 0 ? 1 : referenced_max,
                .lim.flags = BTRFS_QGROUP_LIMIT_MAX_RFER,
        };
        int r;

        assert(fd >= 0);

        r = btrfs_is_filesystem(fd);
        if (r < 0)
                return r;
        if (!r)
                return -ENOTTY;

        if (ioctl(fd, BTRFS_IOC_QGROUP_LIMIT, &args) < 0)
                return -errno;

        return 0;
}

int btrfs_quota_limit(const char *path, uint64_t referenced_max) {
        _cleanup_close_ int fd = -1;

        fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW);
        if (fd < 0)
                return -errno;

        return btrfs_quota_limit_fd(fd, referenced_max);
}

int btrfs_resize_loopback_fd(int fd, uint64_t new_size, bool grow_only) {
        struct btrfs_ioctl_vol_args args = {};
        _cleanup_free_ char *p = NULL, *loop = NULL, *backing = NULL;
        _cleanup_close_ int loop_fd = -1, backing_fd = -1;
        struct stat st;
        dev_t dev = 0;
        int r;

        /* btrfs cannot handle file systems < 16M, hence use this as minimum */
        if (new_size < 16*1024*1024)
                new_size = 16*1024*1024;

        r = btrfs_get_block_device_fd(fd, &dev);
        if (r < 0)
                return r;
        if (r == 0)
                return -ENODEV;

        if (asprintf(&p, "/sys/dev/block/%u:%u/loop/backing_file", major(dev), minor(dev)) < 0)
                return -ENOMEM;
        r = read_one_line_file(p, &backing);
        if (r == -ENOENT)
                return -ENODEV;
        if (r < 0)
                return r;
        if (isempty(backing) || !path_is_absolute(backing))
                return -ENODEV;

        backing_fd = open(backing, O_RDWR|O_CLOEXEC|O_NOCTTY);
        if (backing_fd < 0)
                return -errno;

        if (fstat(backing_fd, &st) < 0)
                return -errno;
        if (!S_ISREG(st.st_mode))
                return -ENODEV;

        if (new_size == (uint64_t) st.st_size)
                return 0;

        if (grow_only && new_size < (uint64_t) st.st_size)
                return -EINVAL;

        if (asprintf(&loop, "/dev/block/%u:%u", major(dev), minor(dev)) < 0)
                return -ENOMEM;
        loop_fd = open(loop, O_RDWR|O_CLOEXEC|O_NOCTTY);
        if (loop_fd < 0)
                return -errno;

        if (snprintf(args.name, sizeof(args.name), "%" PRIu64, new_size) >= (int) sizeof(args.name))
                return -EINVAL;

        if (new_size < (uint64_t) st.st_size) {
                /* Decrease size: first decrease btrfs size, then shorten loopback */
                if (ioctl(fd, BTRFS_IOC_RESIZE, &args) < 0)
                        return -errno;
        }

        if (ftruncate(backing_fd, new_size) < 0)
                return -errno;

        if (ioctl(loop_fd, LOOP_SET_CAPACITY, 0) < 0)
                return -errno;

        if (new_size > (uint64_t) st.st_size) {
                /* Increase size: first enlarge loopback, then increase btrfs size */
                if (ioctl(fd, BTRFS_IOC_RESIZE, &args) < 0)
                        return -errno;
        }

        /* Make sure the free disk space is correctly updated for both file systems */
        (void) fsync(fd);
        (void) fsync(backing_fd);

        return 1;
}

int btrfs_resize_loopback(const char *p, uint64_t new_size, bool grow_only) {
        _cleanup_close_ int fd = -1;

        fd = open(p, O_RDONLY|O_NOCTTY|O_CLOEXEC);
        if (fd < 0)
                return -errno;

        return btrfs_resize_loopback_fd(fd, new_size, grow_only);
}

static int subvol_remove_children(int fd, const char *subvolume, uint64_t subvol_id, bool recursive) {
        struct btrfs_ioctl_search_args args = {
                .key.tree_id = BTRFS_ROOT_TREE_OBJECTID,

                .key.min_objectid = BTRFS_FIRST_FREE_OBJECTID,
                .key.max_objectid = BTRFS_LAST_FREE_OBJECTID,

                .key.min_type = BTRFS_ROOT_BACKREF_KEY,
                .key.max_type = BTRFS_ROOT_BACKREF_KEY,

                .key.min_transid = 0,
                .key.max_transid = (uint64_t) -1,
        };

        struct btrfs_ioctl_vol_args vol_args = {};
        _cleanup_close_ int subvol_fd = -1;
        struct stat st;
        bool made_writable = false;
        int r;

        assert(fd >= 0);
        assert(subvolume);

        if (fstat(fd, &st) < 0)
                return -errno;

        if (!S_ISDIR(st.st_mode))
                return -EINVAL;

        /* First, try to remove the subvolume. If it happens to be
         * already empty, this will just work. */
        strncpy(vol_args.name, subvolume, sizeof(vol_args.name)-1);
        if (ioctl(fd, BTRFS_IOC_SNAP_DESTROY, &vol_args) >= 0)
                return 0;
        if (!recursive || errno != ENOTEMPTY)
                return -errno;

        /* OK, the subvolume is not empty, let's look for child
         * subvolumes, and remove them, first */
        subvol_fd = openat(fd, subvolume, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
        if (subvol_fd < 0)
                return -errno;

        if (subvol_id == 0) {
                r = btrfs_subvol_get_id_fd(subvol_fd, &subvol_id);
                if (r < 0)
                        return r;
        }

        args.key.min_offset = args.key.max_offset = subvol_id;

        while (btrfs_ioctl_search_args_compare(&args) <= 0) {
                const struct btrfs_ioctl_search_header *sh;
                unsigned i;

                args.key.nr_items = 256;
                if (ioctl(fd, BTRFS_IOC_TREE_SEARCH, &args) < 0)
                        return -errno;

                if (args.key.nr_items <= 0)
                        break;

                FOREACH_BTRFS_IOCTL_SEARCH_HEADER(i, sh, args) {
                        _cleanup_free_ char *p = NULL;
                        const struct btrfs_root_ref *ref;
                        struct btrfs_ioctl_ino_lookup_args ino_args;

                        btrfs_ioctl_search_args_set(&args, sh);

                        if (sh->type != BTRFS_ROOT_BACKREF_KEY)
                                continue;
                        if (sh->offset != subvol_id)
                                continue;

                        ref = BTRFS_IOCTL_SEARCH_HEADER_BODY(sh);

                        p = strndup((char*) ref + sizeof(struct btrfs_root_ref), le64toh(ref->name_len));
                        if (!p)
                                return -ENOMEM;

                        zero(ino_args);
                        ino_args.treeid = subvol_id;
                        ino_args.objectid = htole64(ref->dirid);

                        if (ioctl(fd, BTRFS_IOC_INO_LOOKUP, &ino_args) < 0)
                                return -errno;

                        if (!made_writable) {
                                r = btrfs_subvol_set_read_only_fd(subvol_fd, false);
                                if (r < 0)
                                        return r;

                                made_writable = true;
                        }

                        if (isempty(ino_args.name))
                                /* Subvolume is in the top-level
                                 * directory of the subvolume. */
                                r = subvol_remove_children(subvol_fd, p, sh->objectid, recursive);
                        else {
                                _cleanup_close_ int child_fd = -1;

                                /* Subvolume is somewhere further down,
                                 * hence we need to open the
                                 * containing directory first */

                                child_fd = openat(subvol_fd, ino_args.name, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
                                if (child_fd < 0)
                                        return -errno;

                                r = subvol_remove_children(child_fd, p, sh->objectid, recursive);
                        }
                        if (r < 0)
                                return r;
                }

                /* Increase search key by one, to read the next item, if we can. */
                if (!btrfs_ioctl_search_args_inc(&args))
                        break;
        }

        /* OK, the child subvolumes should all be gone now, let's try
         * again to remove the subvolume */
        if (ioctl(fd, BTRFS_IOC_SNAP_DESTROY, &vol_args) < 0)
                return -errno;

        return 0;
}

int btrfs_subvol_remove(const char *path, bool recursive) {
        _cleanup_close_ int fd = -1;
        const char *subvolume;
        int r;

        assert(path);

        r = extract_subvolume_name(path, &subvolume);
        if (r < 0)
                return r;

        fd = open_parent(path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
        if (fd < 0)
                return fd;

        return subvol_remove_children(fd, subvolume, 0, recursive);
}

int btrfs_subvol_remove_fd(int fd, const char *subvolume, bool recursive) {
        return subvol_remove_children(fd, subvolume, 0, recursive);
}

static int subvol_snapshot_children(int old_fd, int new_fd, const char *subvolume, uint64_t old_subvol_id, BtrfsSnapshotFlags flags) {

        struct btrfs_ioctl_search_args args = {
                .key.tree_id = BTRFS_ROOT_TREE_OBJECTID,

                .key.min_objectid = BTRFS_FIRST_FREE_OBJECTID,
                .key.max_objectid = BTRFS_LAST_FREE_OBJECTID,

                .key.min_type = BTRFS_ROOT_BACKREF_KEY,
                .key.max_type = BTRFS_ROOT_BACKREF_KEY,

                .key.min_transid = 0,
                .key.max_transid = (uint64_t) -1,
        };

        struct btrfs_ioctl_vol_args_v2 vol_args = {
                .flags = flags & BTRFS_SNAPSHOT_READ_ONLY ? BTRFS_SUBVOL_RDONLY : 0,
                .fd = old_fd,
        };
        _cleanup_close_ int subvolume_fd = -1;
        uint64_t new_subvol_id;
        int r;

        assert(old_fd >= 0);
        assert(new_fd >= 0);
        assert(subvolume);

        strncpy(vol_args.name, subvolume, sizeof(vol_args.name)-1);
        vol_args.fd = old_fd;

        if (ioctl(new_fd, BTRFS_IOC_SNAP_CREATE_V2, &vol_args) < 0)
                return -errno;

        if (!(flags & BTRFS_SNAPSHOT_RECURSIVE))
                return 0;

        if (old_subvol_id == 0) {
                r = btrfs_subvol_get_id_fd(old_fd, &old_subvol_id);
                if (r < 0)
                        return r;
        }

        r = btrfs_subvol_get_id(new_fd, vol_args.name, &new_subvol_id);
        if (r < 0)
                return r;

        args.key.min_offset = args.key.max_offset = old_subvol_id;

        while (btrfs_ioctl_search_args_compare(&args) <= 0) {
                const struct btrfs_ioctl_search_header *sh;
                unsigned i;

                args.key.nr_items = 256;
                if (ioctl(old_fd, BTRFS_IOC_TREE_SEARCH, &args) < 0)
                        return -errno;

                if (args.key.nr_items <= 0)
                        break;

                FOREACH_BTRFS_IOCTL_SEARCH_HEADER(i, sh, args) {
                        _cleanup_free_ char *p = NULL, *c = NULL, *np = NULL;
                        struct btrfs_ioctl_ino_lookup_args ino_args;
                        const struct btrfs_root_ref *ref;
                        _cleanup_close_ int old_child_fd = -1, new_child_fd = -1;

                        btrfs_ioctl_search_args_set(&args, sh);

                        if (sh->type != BTRFS_ROOT_BACKREF_KEY)
                                continue;

                        /* Avoid finding the source subvolume a second
                         * time */
                        if (sh->offset != old_subvol_id)
                                continue;

                        /* Avoid running into loops if the new
                         * subvolume is below the old one. */
                        if (sh->objectid == new_subvol_id)
                                continue;

                        ref = BTRFS_IOCTL_SEARCH_HEADER_BODY(sh);
                        p = strndup((char*) ref + sizeof(struct btrfs_root_ref), le64toh(ref->name_len));
                        if (!p)
                                return -ENOMEM;

                        zero(ino_args);
                        ino_args.treeid = old_subvol_id;
                        ino_args.objectid = htole64(ref->dirid);

                        if (ioctl(old_fd, BTRFS_IOC_INO_LOOKUP, &ino_args) < 0)
                                return -errno;

                        /* The kernel returns an empty name if the
                         * subvolume is in the top-level directory,
                         * and otherwise appends a slash, so that we
                         * can just concatenate easily here, without
                         * adding a slash. */
                        c = strappend(ino_args.name, p);
                        if (!c)
                                return -ENOMEM;

                        old_child_fd = openat(old_fd, c, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
                        if (old_child_fd < 0)
                                return -errno;

                        np = strjoin(subvolume, "/", ino_args.name, NULL);
                        if (!np)
                                return -ENOMEM;

                        new_child_fd = openat(new_fd, np, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
                        if (new_child_fd < 0)
                                return -errno;

                        if (flags & BTRFS_SNAPSHOT_READ_ONLY) {
                                /* If the snapshot is read-only we
                                 * need to mark it writable
                                 * temporarily, to put the subsnapshot
                                 * into place. */

                                if (subvolume_fd < 0) {
                                        subvolume_fd = openat(new_fd, subvolume, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
                                        if (subvolume_fd < 0)
                                                return -errno;
                                }

                                r = btrfs_subvol_set_read_only_fd(subvolume_fd, false);
                                if (r < 0)
                                        return r;
                        }

                        /* When btrfs clones the subvolumes, child
                         * subvolumes appear as empty directories. Remove
                         * them, so that we can create a new snapshot
                         * in their place */
                        if (unlinkat(new_child_fd, p, AT_REMOVEDIR) < 0) {
                                int k = -errno;

                                if (flags & BTRFS_SNAPSHOT_READ_ONLY)
                                        (void) btrfs_subvol_set_read_only_fd(subvolume_fd, true);

                                return k;
                        }

                        r = subvol_snapshot_children(old_child_fd, new_child_fd, p, sh->objectid, flags & ~BTRFS_SNAPSHOT_FALLBACK_COPY);

                        /* Restore the readonly flag */
                        if (flags & BTRFS_SNAPSHOT_READ_ONLY) {
                                int k;

                                k = btrfs_subvol_set_read_only_fd(subvolume_fd, true);
                                if (r >= 0 && k < 0)
                                        return k;
                        }

                        if (r < 0)
                                return r;
                }

                /* Increase search key by one, to read the next item, if we can. */
                if (!btrfs_ioctl_search_args_inc(&args))
                        break;
        }

        return 0;
}

int btrfs_subvol_snapshot_fd(int old_fd, const char *new_path, BtrfsSnapshotFlags flags) {
        _cleanup_close_ int new_fd = -1;
        const char *subvolume;
        int r;

        assert(old_fd >= 0);
        assert(new_path);

        r = btrfs_is_subvol(old_fd);
        if (r < 0)
                return r;
        if (r == 0) {
                if (!(flags & BTRFS_SNAPSHOT_FALLBACK_COPY))
                        return -EISDIR;

                r = btrfs_subvol_make(new_path);
                if (r < 0)
                        return r;

                r = copy_directory_fd(old_fd, new_path, true);
                if (r < 0) {
                        btrfs_subvol_remove(new_path, false);
                        return r;
                }

                if (flags & BTRFS_SNAPSHOT_READ_ONLY) {
                        r = btrfs_subvol_set_read_only(new_path, true);
                        if (r < 0) {
                                btrfs_subvol_remove(new_path, false);
                                return r;
                        }
                }

                return 0;
        }

        r = extract_subvolume_name(new_path, &subvolume);
        if (r < 0)
                return r;

        new_fd = open_parent(new_path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
        if (new_fd < 0)
                return new_fd;

        return subvol_snapshot_children(old_fd, new_fd, subvolume, 0, flags);
}

int btrfs_subvol_snapshot(const char *old_path, const char *new_path, BtrfsSnapshotFlags flags) {
        _cleanup_close_ int old_fd = -1;

        assert(old_path);
        assert(new_path);

        old_fd = open(old_path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY);
        if (old_fd < 0)
                return -errno;

        return btrfs_subvol_snapshot_fd(old_fd, new_path, flags);
}