diff options
Diffstat (limited to 'src/basic/btrfs-util.c')
-rw-r--r-- | src/basic/btrfs-util.c | 971 |
1 files changed, 908 insertions, 63 deletions
diff --git a/src/basic/btrfs-util.c b/src/basic/btrfs-util.c index 074deeccda..4c90bc0c80 100644 --- a/src/basic/btrfs-util.c +++ b/src/basic/btrfs-util.c @@ -20,23 +20,26 @@ ***/ #include <stdlib.h> -#include <sys/vfs.h> #include <sys/stat.h> - +#include <sys/vfs.h> #ifdef HAVE_LINUX_BTRFS_H #include <linux/btrfs.h> #endif +#include "alloc-util.h" +#include "btrfs-ctree.h" +#include "btrfs-util.h" +#include "copy.h" +#include "fd-util.h" +#include "fileio.h" +#include "macro.h" #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" +#include "stat-util.h" +#include "string-util.h" +#include "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 @@ -59,13 +62,13 @@ static int validate_subvolume_name(const char *name) { static int open_parent(const char *path, int flags) { _cleanup_free_ char *parent = NULL; - int r, fd; + int fd; assert(path); - r = path_get_parent(path, &parent); - if (r < 0) - return r; + parent = dirname_malloc(path); + if (!parent) + return -ENOMEM; fd = open(parent, flags); if (fd < 0) @@ -436,7 +439,7 @@ static int btrfs_ioctl_search_args_compare(const struct btrfs_ioctl_search_args #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) { +int btrfs_subvol_get_info_fd(int fd, uint64_t subvol_id, BtrfsSubvolInfo *ret) { struct btrfs_ioctl_search_args args = { /* Tree of tree roots */ .key.tree_id = BTRFS_ROOT_TREE_OBJECTID, @@ -453,16 +456,23 @@ int btrfs_subvol_get_info_fd(int fd, BtrfsSubvolInfo *ret) { .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; + if (subvol_id == 0) { + r = btrfs_subvol_get_id_fd(fd, &subvol_id); + if (r < 0) + return r; + } else { + r = btrfs_is_filesystem(fd); + if (r < 0) + return r; + if (!r) + return -ENOTTY; + } args.key.min_objectid = args.key.max_objectid = subvol_id; @@ -521,7 +531,7 @@ finish: return 0; } -int btrfs_subvol_get_quota_fd(int fd, BtrfsQuotaInfo *ret) { +int btrfs_qgroup_get_quota_fd(int fd, uint64_t qgroupid, BtrfsQuotaInfo *ret) { struct btrfs_ioctl_search_args args = { /* Tree of quota items */ @@ -540,18 +550,25 @@ int btrfs_subvol_get_quota_fd(int fd, BtrfsQuotaInfo *ret) { .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; + if (qgroupid == 0) { + r = btrfs_subvol_get_id_fd(fd, &qgroupid); + if (r < 0) + return r; + } else { + r = btrfs_is_filesystem(fd); + if (r < 0) + return r; + if (!r) + return -ENOTTY; + } - args.key.min_offset = args.key.max_offset = subvol_id; + args.key.min_offset = args.key.max_offset = qgroupid; while (btrfs_ioctl_search_args_compare(&args) <= 0) { const struct btrfs_ioctl_search_header *sh; @@ -571,7 +588,7 @@ int btrfs_subvol_get_quota_fd(int fd, BtrfsQuotaInfo *ret) { if (sh->objectid != 0) continue; - if (sh->offset != subvol_id) + if (sh->offset != qgroupid) continue; if (sh->type == BTRFS_QGROUP_INFO_KEY) { @@ -585,12 +602,14 @@ int btrfs_subvol_get_quota_fd(int fd, BtrfsQuotaInfo *ret) { } 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) + if (le64toh(qli->flags) & BTRFS_QGROUP_LIMIT_MAX_RFER) + ret->referenced_max = le64toh(qli->max_rfer); + else ret->referenced_max = (uint64_t) -1; - if (ret->exclusive_max == 0) + + if (le64toh(qli->flags) & BTRFS_QGROUP_LIMIT_MAX_EXCL) + ret->exclusive_max = le64toh(qli->max_excl); + else ret->exclusive_max = (uint64_t) -1; found_limit = true; @@ -622,6 +641,109 @@ finish: return 0; } +int btrfs_qgroup_get_quota(const char *path, uint64_t qgroupid, BtrfsQuotaInfo *ret) { + _cleanup_close_ int fd = -1; + + fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); + if (fd < 0) + return -errno; + + return btrfs_qgroup_get_quota_fd(fd, qgroupid, ret); +} + +int btrfs_subvol_find_subtree_qgroup(int fd, uint64_t subvol_id, uint64_t *ret) { + uint64_t level, lowest = (uint64_t) -1, lowest_qgroupid = 0; + _cleanup_free_ uint64_t *qgroups = NULL; + int r, n, i; + + assert(fd >= 0); + assert(ret); + + /* This finds the "subtree" qgroup for a specific + * subvolume. This only works for subvolumes that have been + * prepared with btrfs_subvol_auto_qgroup_fd() with + * insert_intermediary_qgroup=true (or equivalent). For others + * it will return the leaf qgroup instead. The two cases may + * be distuingished via the return value, which is 1 in case + * an appropriate "subtree" qgroup was found, and 0 + * otherwise. */ + + if (subvol_id == 0) { + r = btrfs_subvol_get_id_fd(fd, &subvol_id); + if (r < 0) + return r; + } + + r = btrfs_qgroupid_split(subvol_id, &level, NULL); + if (r < 0) + return r; + if (level != 0) /* Input must be a leaf qgroup */ + return -EINVAL; + + n = btrfs_qgroup_find_parents(fd, subvol_id, &qgroups); + if (n < 0) + return n; + + for (i = 0; i < n; i++) { + uint64_t id; + + r = btrfs_qgroupid_split(qgroups[i], &level, &id); + if (r < 0) + return r; + + if (id != subvol_id) + continue; + + if (lowest == (uint64_t) -1 || level < lowest) { + lowest_qgroupid = qgroups[i]; + lowest = level; + } + } + + if (lowest == (uint64_t) -1) { + /* No suitable higher-level qgroup found, let's return + * the leaf qgroup instead, and indicate that with the + * return value. */ + + *ret = subvol_id; + return 0; + } + + *ret = lowest_qgroupid; + return 1; +} + +int btrfs_subvol_get_subtree_quota_fd(int fd, uint64_t subvol_id, BtrfsQuotaInfo *ret) { + uint64_t qgroupid; + int r; + + assert(fd >= 0); + assert(ret); + + /* This determines the quota data of the qgroup with the + * lowest level, that shares the id part with the specified + * subvolume. This is useful for determining the quota data + * for entire subvolume subtrees, as long as the subtrees have + * been set up with btrfs_qgroup_subvol_auto_fd() or in a + * compatible way */ + + r = btrfs_subvol_find_subtree_qgroup(fd, subvol_id, &qgroupid); + if (r < 0) + return r; + + return btrfs_qgroup_get_quota_fd(fd, qgroupid, ret); +} + +int btrfs_subvol_get_subtree_quota(const char *path, uint64_t subvol_id, BtrfsQuotaInfo *ret) { + _cleanup_close_ int fd = -1; + + fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NOFOLLOW); + if (fd < 0) + return -errno; + + return btrfs_subvol_get_subtree_quota_fd(fd, subvol_id, ret); +} + int btrfs_defrag_fd(int fd) { struct stat st; @@ -679,37 +801,79 @@ int btrfs_quota_enable(const char *path, bool b) { return btrfs_quota_enable_fd(fd, b); } -int btrfs_quota_limit_fd(int fd, uint64_t referenced_max) { +int btrfs_qgroup_set_limit_fd(int fd, uint64_t qgroupid, 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.max_rfer = referenced_max, .lim.flags = BTRFS_QGROUP_LIMIT_MAX_RFER, }; + unsigned c; int r; assert(fd >= 0); - r = btrfs_is_filesystem(fd); - if (r < 0) - return r; - if (!r) - return -ENOTTY; + if (qgroupid == 0) { + r = btrfs_subvol_get_id_fd(fd, &qgroupid); + if (r < 0) + return r; + } else { + 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; + args.qgroupid = qgroupid; + + for (c = 0;; c++) { + if (ioctl(fd, BTRFS_IOC_QGROUP_LIMIT, &args) < 0) { + + if (errno == EBUSY && c < 10) { + (void) btrfs_quota_scan_wait(fd); + continue; + } + + return -errno; + } + + break; + } return 0; } -int btrfs_quota_limit(const char *path, uint64_t referenced_max) { +int btrfs_qgroup_set_limit(const char *path, uint64_t qgroupid, 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_qgroup_set_limit_fd(fd, qgroupid, referenced_max); +} + +int btrfs_subvol_set_subtree_quota_limit_fd(int fd, uint64_t subvol_id, uint64_t referenced_max) { + uint64_t qgroupid; + int r; + + assert(fd >= 0); + + r = btrfs_subvol_find_subtree_qgroup(fd, subvol_id, &qgroupid); + if (r < 0) + return r; + + return btrfs_qgroup_set_limit_fd(fd, qgroupid, referenced_max); +} + +int btrfs_subvol_set_subtree_quota_limit(const char *path, uint64_t subvol_id, 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); + return btrfs_subvol_set_subtree_quota_limit_fd(fd, subvol_id, referenced_max); } int btrfs_resize_loopback_fd(int fd, uint64_t new_size, bool grow_only) { @@ -799,7 +963,188 @@ int btrfs_resize_loopback(const char *p, uint64_t new_size, bool grow_only) { 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) { +int btrfs_qgroupid_make(uint64_t level, uint64_t id, uint64_t *ret) { + assert(ret); + + if (level >= (UINT64_C(1) << (64 - BTRFS_QGROUP_LEVEL_SHIFT))) + return -EINVAL; + + if (id >= (UINT64_C(1) << BTRFS_QGROUP_LEVEL_SHIFT)) + return -EINVAL; + + *ret = (level << BTRFS_QGROUP_LEVEL_SHIFT) | id; + return 0; +} + +int btrfs_qgroupid_split(uint64_t qgroupid, uint64_t *level, uint64_t *id) { + assert(level || id); + + if (level) + *level = qgroupid >> BTRFS_QGROUP_LEVEL_SHIFT; + + if (id) + *id = qgroupid & ((UINT64_C(1) << BTRFS_QGROUP_LEVEL_SHIFT) - 1); + + return 0; +} + +static int qgroup_create_or_destroy(int fd, bool b, uint64_t qgroupid) { + + struct btrfs_ioctl_qgroup_create_args args = { + .create = b, + .qgroupid = qgroupid, + }; + unsigned c; + int r; + + r = btrfs_is_filesystem(fd); + if (r < 0) + return r; + if (r == 0) + return -ENOTTY; + + for (c = 0;; c++) { + if (ioctl(fd, BTRFS_IOC_QGROUP_CREATE, &args) < 0) { + + if (errno == EBUSY && c < 10) { + (void) btrfs_quota_scan_wait(fd); + continue; + } + + return -errno; + } + + break; + } + + return 0; +} + +int btrfs_qgroup_create(int fd, uint64_t qgroupid) { + return qgroup_create_or_destroy(fd, true, qgroupid); +} + +int btrfs_qgroup_destroy(int fd, uint64_t qgroupid) { + return qgroup_create_or_destroy(fd, false, qgroupid); +} + +int btrfs_qgroup_destroy_recursive(int fd, uint64_t qgroupid) { + _cleanup_free_ uint64_t *qgroups = NULL; + uint64_t subvol_id; + int i, n, r; + + /* Destroys the specified qgroup, but unassigns it from all + * its parents first. Also, it recursively destroys all + * qgroups it is assgined to that have the same id part of the + * qgroupid as the specified group. */ + + r = btrfs_qgroupid_split(qgroupid, NULL, &subvol_id); + if (r < 0) + return r; + + n = btrfs_qgroup_find_parents(fd, qgroupid, &qgroups); + if (n < 0) + return n; + + for (i = 0; i < n; i++) { + uint64_t id; + + r = btrfs_qgroupid_split(qgroups[i], NULL, &id); + if (r < 0) + return r; + + r = btrfs_qgroup_unassign(fd, qgroupid, qgroups[i]); + if (r < 0) + return r; + + if (id != subvol_id) + continue; + + /* The parent qgroupid shares the same id part with + * us? If so, destroy it too. */ + + (void) btrfs_qgroup_destroy_recursive(fd, qgroups[i]); + } + + return btrfs_qgroup_destroy(fd, qgroupid); +} + +int btrfs_quota_scan_start(int fd) { + struct btrfs_ioctl_quota_rescan_args args = {}; + + assert(fd >= 0); + + if (ioctl(fd, BTRFS_IOC_QUOTA_RESCAN, &args) < 0) + return -errno; + + return 0; +} + +int btrfs_quota_scan_wait(int fd) { + assert(fd >= 0); + + if (ioctl(fd, BTRFS_IOC_QUOTA_RESCAN_WAIT) < 0) + return -errno; + + return 0; +} + +int btrfs_quota_scan_ongoing(int fd) { + struct btrfs_ioctl_quota_rescan_args args = {}; + + assert(fd >= 0); + + if (ioctl(fd, BTRFS_IOC_QUOTA_RESCAN_STATUS, &args) < 0) + return -errno; + + return !!args.flags; +} + +static int qgroup_assign_or_unassign(int fd, bool b, uint64_t child, uint64_t parent) { + struct btrfs_ioctl_qgroup_assign_args args = { + .assign = b, + .src = child, + .dst = parent, + }; + unsigned c; + int r; + + r = btrfs_is_filesystem(fd); + if (r < 0) + return r; + if (r == 0) + return -ENOTTY; + + for (c = 0;; c++) { + r = ioctl(fd, BTRFS_IOC_QGROUP_ASSIGN, &args); + if (r < 0) { + if (errno == EBUSY && c < 10) { + (void) btrfs_quota_scan_wait(fd); + continue; + } + + return -errno; + } + + if (r == 0) + return 0; + + /* If the return value is > 0, we need to request a rescan */ + + (void) btrfs_quota_scan_start(fd); + return 1; + } +} + +int btrfs_qgroup_assign(int fd, uint64_t child, uint64_t parent) { + return qgroup_assign_or_unassign(fd, true, child, parent); +} + +int btrfs_qgroup_unassign(int fd, uint64_t child, uint64_t parent) { + return qgroup_assign_or_unassign(fd, false, child, parent); +} + +static int subvol_remove_children(int fd, const char *subvolume, uint64_t subvol_id, BtrfsRemoveFlags flags) { struct btrfs_ioctl_search_args args = { .key.tree_id = BTRFS_ROOT_TREE_OBJECTID, @@ -828,16 +1173,6 @@ static int subvol_remove_children(int fd, const char *subvolume, uint64_t subvol 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; @@ -848,6 +1183,19 @@ static int subvol_remove_children(int fd, const char *subvolume, uint64_t subvol return r; } + /* 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) { + (void) btrfs_qgroup_destroy_recursive(fd, subvol_id); /* for the leaf subvolumes, the qgroup id is identical to the subvol id */ + return 0; + } + if (!(flags & BTRFS_REMOVE_RECURSIVE) || errno != ENOTEMPTY) + return -errno; + + /* OK, the subvolume is not empty, let's look for child + * subvolumes, and remove them, first */ + args.key.min_offset = args.key.max_offset = subvol_id; while (btrfs_ioctl_search_args_compare(&args) <= 0) { @@ -897,7 +1245,7 @@ static int subvol_remove_children(int fd, const char *subvolume, uint64_t subvol 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); + r = subvol_remove_children(subvol_fd, p, sh->objectid, flags); else { _cleanup_close_ int child_fd = -1; @@ -909,7 +1257,7 @@ static int subvol_remove_children(int fd, const char *subvolume, uint64_t subvol if (child_fd < 0) return -errno; - r = subvol_remove_children(child_fd, p, sh->objectid, recursive); + r = subvol_remove_children(child_fd, p, sh->objectid, flags); } if (r < 0) return r; @@ -925,10 +1273,11 @@ static int subvol_remove_children(int fd, const char *subvolume, uint64_t subvol if (ioctl(fd, BTRFS_IOC_SNAP_DESTROY, &vol_args) < 0) return -errno; + (void) btrfs_qgroup_destroy_recursive(fd, subvol_id); return 0; } -int btrfs_subvol_remove(const char *path, bool recursive) { +int btrfs_subvol_remove(const char *path, BtrfsRemoveFlags flags) { _cleanup_close_ int fd = -1; const char *subvolume; int r; @@ -943,11 +1292,194 @@ int btrfs_subvol_remove(const char *path, bool recursive) { if (fd < 0) return fd; - return subvol_remove_children(fd, subvolume, 0, recursive); + return subvol_remove_children(fd, subvolume, 0, flags); +} + +int btrfs_subvol_remove_fd(int fd, const char *subvolume, BtrfsRemoveFlags flags) { + return subvol_remove_children(fd, subvolume, 0, flags); +} + +int btrfs_qgroup_copy_limits(int fd, uint64_t old_qgroupid, uint64_t new_qgroupid) { + + 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_LIMIT_KEY, + .key.max_type = BTRFS_QGROUP_LIMIT_KEY, + + /* For our qgroup */ + .key.min_offset = old_qgroupid, + .key.max_offset = old_qgroupid, + + /* No restrictions on the other components */ + .key.min_transid = 0, + .key.max_transid = (uint64_t) -1, + }; + + int r; + + r = btrfs_is_filesystem(fd); + if (r < 0) + return r; + if (!r) + return -ENOTTY; + + 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_qgroup_limit_item *qli = BTRFS_IOCTL_SEARCH_HEADER_BODY(sh); + struct btrfs_ioctl_qgroup_limit_args qargs; + unsigned c; + + /* 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->type != BTRFS_QGROUP_LIMIT_KEY) + continue; + if (sh->offset != old_qgroupid) + continue; + + /* We found the entry, now copy things over. */ + + qargs = (struct btrfs_ioctl_qgroup_limit_args) { + .qgroupid = new_qgroupid, + + .lim.max_rfer = le64toh(qli->max_rfer), + .lim.max_excl = le64toh(qli->max_excl), + .lim.rsv_rfer = le64toh(qli->rsv_rfer), + .lim.rsv_excl = le64toh(qli->rsv_excl), + + .lim.flags = le64toh(qli->flags) & (BTRFS_QGROUP_LIMIT_MAX_RFER| + BTRFS_QGROUP_LIMIT_MAX_EXCL| + BTRFS_QGROUP_LIMIT_RSV_RFER| + BTRFS_QGROUP_LIMIT_RSV_EXCL), + }; + + for (c = 0;; c++) { + if (ioctl(fd, BTRFS_IOC_QGROUP_LIMIT, &qargs) < 0) { + if (errno == EBUSY && c < 10) { + (void) btrfs_quota_scan_wait(fd); + continue; + } + return -errno; + } + + break; + } + + return 1; + } + + /* Increase search key by one, to read the next item, if we can. */ + if (!btrfs_ioctl_search_args_inc(&args)) + break; + } + + return 0; +} + +static int copy_quota_hierarchy(int fd, uint64_t old_subvol_id, uint64_t new_subvol_id) { + _cleanup_free_ uint64_t *old_qgroups = NULL, *old_parent_qgroups = NULL; + bool copy_from_parent = false, insert_intermediary_qgroup = false; + int n_old_qgroups, n_old_parent_qgroups, r, i; + uint64_t old_parent_id; + + assert(fd >= 0); + + /* Copies a reduced form of quota information from the old to + * the new subvolume. */ + + n_old_qgroups = btrfs_qgroup_find_parents(fd, old_subvol_id, &old_qgroups); + if (n_old_qgroups <= 0) /* Nothing to copy */ + return n_old_qgroups; + + r = btrfs_subvol_get_parent(fd, old_subvol_id, &old_parent_id); + if (r < 0) + return r; + + n_old_parent_qgroups = btrfs_qgroup_find_parents(fd, old_parent_id, &old_parent_qgroups); + if (n_old_parent_qgroups < 0) + return n_old_parent_qgroups; + + for (i = 0; i < n_old_qgroups; i++) { + uint64_t id; + int j; + + r = btrfs_qgroupid_split(old_qgroups[i], NULL, &id); + if (r < 0) + return r; + + if (id == old_subvol_id) { + /* The old subvolume was member of a qgroup + * that had the same id, but a different level + * as it self. Let's set up something similar + * in the destination. */ + insert_intermediary_qgroup = true; + break; + } + + for (j = 0; j < n_old_parent_qgroups; j++) + if (old_parent_qgroups[j] == old_qgroups[i]) { + /* The old subvolume shared a common + * parent qgroup with its parent + * subvolume. Let's set up something + * similar in the destination. */ + copy_from_parent = true; + } + } + + if (!insert_intermediary_qgroup && !copy_from_parent) + return 0; + + return btrfs_subvol_auto_qgroup_fd(fd, new_subvol_id, insert_intermediary_qgroup); } -int btrfs_subvol_remove_fd(int fd, const char *subvolume, bool recursive) { - return subvol_remove_children(fd, subvolume, 0, recursive); +static int copy_subtree_quota_limits(int fd, uint64_t old_subvol, uint64_t new_subvol) { + uint64_t old_subtree_qgroup, new_subtree_qgroup; + bool changed; + int r; + + /* First copy the leaf limits */ + r = btrfs_qgroup_copy_limits(fd, old_subvol, new_subvol); + if (r < 0) + return r; + changed = r > 0; + + /* Then, try to copy the subtree limits, if there are any. */ + r = btrfs_subvol_find_subtree_qgroup(fd, old_subvol, &old_subtree_qgroup); + if (r < 0) + return r; + if (r == 0) + return changed; + + r = btrfs_subvol_find_subtree_qgroup(fd, new_subvol, &new_subtree_qgroup); + if (r < 0) + return r; + if (r == 0) + return changed; + + r = btrfs_qgroup_copy_limits(fd, old_subtree_qgroup, new_subtree_qgroup); + if (r != 0) + return r; + + return changed; } static int subvol_snapshot_children(int old_fd, int new_fd, const char *subvolume, uint64_t old_subvol_id, BtrfsSnapshotFlags flags) { @@ -978,12 +1510,12 @@ static int subvol_snapshot_children(int old_fd, int new_fd, const char *subvolum 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)) + if (!(flags & BTRFS_SNAPSHOT_RECURSIVE) && + !(flags & BTRFS_SNAPSHOT_QUOTA)) return 0; if (old_subvol_id == 0) { @@ -996,6 +1528,17 @@ static int subvol_snapshot_children(int old_fd, int new_fd, const char *subvolum if (r < 0) return r; + if (flags & BTRFS_SNAPSHOT_QUOTA) + (void) copy_quota_hierarchy(new_fd, old_subvol_id, new_subvol_id); + + if (!(flags & BTRFS_SNAPSHOT_RECURSIVE)) { + + if (flags & BTRFS_SNAPSHOT_QUOTA) + (void) copy_subtree_quota_limits(new_fd, old_subvol_id, new_subvol_id); + + return 0; + } + args.key.min_offset = args.key.max_offset = old_subvol_id; while (btrfs_ioctl_search_args_compare(&args) <= 0) { @@ -1113,6 +1656,9 @@ static int subvol_snapshot_children(int old_fd, int new_fd, const char *subvolum break; } + if (flags & BTRFS_SNAPSHOT_QUOTA) + (void) copy_subtree_quota_limits(new_fd, old_subvol_id, new_subvol_id); + return 0; } @@ -1137,14 +1683,14 @@ int btrfs_subvol_snapshot_fd(int old_fd, const char *new_path, BtrfsSnapshotFlag r = copy_directory_fd(old_fd, new_path, true); if (r < 0) { - btrfs_subvol_remove(new_path, false); + (void) btrfs_subvol_remove(new_path, BTRFS_REMOVE_QUOTA); 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); + (void) btrfs_subvol_remove(new_path, BTRFS_REMOVE_QUOTA); return r; } } @@ -1175,3 +1721,302 @@ int btrfs_subvol_snapshot(const char *old_path, const char *new_path, BtrfsSnaps return btrfs_subvol_snapshot_fd(old_fd, new_path, flags); } + +int btrfs_qgroup_find_parents(int fd, uint64_t qgroupid, uint64_t **ret) { + + struct btrfs_ioctl_search_args args = { + /* Tree of quota items */ + .key.tree_id = BTRFS_QUOTA_TREE_OBJECTID, + + /* Look precisely for the quota relation items */ + .key.min_type = BTRFS_QGROUP_RELATION_KEY, + .key.max_type = BTRFS_QGROUP_RELATION_KEY, + + /* No restrictions on the other components */ + .key.min_offset = 0, + .key.max_offset = (uint64_t) -1, + + .key.min_transid = 0, + .key.max_transid = (uint64_t) -1, + }; + + _cleanup_free_ uint64_t *items = NULL; + size_t n_items = 0, n_allocated = 0; + int r; + + assert(fd >= 0); + assert(ret); + + if (qgroupid == 0) { + r = btrfs_subvol_get_id_fd(fd, &qgroupid); + if (r < 0) + return r; + } else { + r = btrfs_is_filesystem(fd); + if (r < 0) + return r; + if (!r) + return -ENOTTY; + } + + args.key.min_objectid = args.key.max_objectid = qgroupid; + + 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->type != BTRFS_QGROUP_RELATION_KEY) + continue; + if (sh->offset < sh->objectid) + continue; + if (sh->objectid != qgroupid) + continue; + + if (!GREEDY_REALLOC(items, n_allocated, n_items+1)) + return -ENOMEM; + + items[n_items++] = sh->offset; + } + + /* Increase search key by one, to read the next item, if we can. */ + if (!btrfs_ioctl_search_args_inc(&args)) + break; + } + + if (n_items <= 0) { + *ret = NULL; + return 0; + } + + *ret = items; + items = NULL; + + return (int) n_items; +} + +int btrfs_subvol_auto_qgroup_fd(int fd, uint64_t subvol_id, bool insert_intermediary_qgroup) { + _cleanup_free_ uint64_t *qgroups = NULL; + uint64_t parent_subvol; + bool changed = false; + int n = 0, r; + + assert(fd >= 0); + + /* + * Sets up the specified subvolume's qgroup automatically in + * one of two ways: + * + * If insert_intermediary_qgroup is false, the subvolume's + * leaf qgroup will be assigned to the same parent qgroups as + * the subvolume's parent subvolume. + * + * If insert_intermediary_qgroup is true a new intermediary + * higher-level qgroup is created, with a higher level number, + * but reusing the id of the subvolume. The level number is + * picked as one smaller than the lowest level qgroup the + * parent subvolume is a member of. If the parent subvolume's + * leaf qgroup is assigned to no higher-level qgroup a new + * qgroup of level 255 is created instead. Either way, the new + * qgroup is then assigned to the parent's higher-level + * qgroup, and the subvolume itself is assigned to it. + * + * If the subvolume is already assigned to a higher level + * qgroup, no operation is executed. + * + * Effectively this means: regardless if + * insert_intermediary_qgroup is true or not, after this + * function is invoked the subvolume will be accounted within + * the same qgroups as the parent. However, if it is true, it + * will also get its own higher-level qgroup, which may in + * turn be used by subvolumes created beneath this subvolume + * later on. + * + * This hence defines a simple default qgroup setup for + * subvolumes, as long as this function is invoked on each + * created subvolume: each subvolume is always accounting + * together with its immediate parents. Optionally, if + * insert_intermediary_qgroup is true, it will also get a + * qgroup that then includes all its own child subvolumes. + */ + + if (subvol_id == 0) { + r = btrfs_is_subvol(fd); + if (r < 0) + return r; + if (!r) + return -ENOTTY; + + r = btrfs_subvol_get_id_fd(fd, &subvol_id); + if (r < 0) + return r; + } + + n = btrfs_qgroup_find_parents(fd, subvol_id, &qgroups); + if (n < 0) + return n; + if (n > 0) /* already parent qgroups set up, let's bail */ + return 0; + + r = btrfs_subvol_get_parent(fd, subvol_id, &parent_subvol); + if (r < 0) + return r; + + qgroups = mfree(qgroups); + n = btrfs_qgroup_find_parents(fd, parent_subvol, &qgroups); + if (n < 0) + return n; + + if (insert_intermediary_qgroup) { + uint64_t lowest = 256, new_qgroupid; + bool created = false; + int i; + + /* Determine the lowest qgroup that the parent + * subvolume is assigned to. */ + + for (i = 0; i < n; i++) { + uint64_t level; + + r = btrfs_qgroupid_split(qgroups[i], &level, NULL); + if (r < 0) + return r; + + if (level < lowest) + lowest = level; + } + + if (lowest <= 1) /* There are no levels left we could use insert an intermediary qgroup at */ + return -EBUSY; + + r = btrfs_qgroupid_make(lowest - 1, subvol_id, &new_qgroupid); + if (r < 0) + return r; + + /* Create the new intermediary group, unless it already exists */ + r = btrfs_qgroup_create(fd, new_qgroupid); + if (r < 0 && r != -EEXIST) + return r; + if (r >= 0) + changed = created = true; + + for (i = 0; i < n; i++) { + r = btrfs_qgroup_assign(fd, new_qgroupid, qgroups[i]); + if (r < 0 && r != -EEXIST) { + if (created) + (void) btrfs_qgroup_destroy_recursive(fd, new_qgroupid); + + return r; + } + if (r >= 0) + changed = true; + } + + r = btrfs_qgroup_assign(fd, subvol_id, new_qgroupid); + if (r < 0 && r != -EEXIST) { + if (created) + (void) btrfs_qgroup_destroy_recursive(fd, new_qgroupid); + return r; + } + if (r >= 0) + changed = true; + + } else { + int i; + + /* Assign our subvolume to all the same qgroups as the parent */ + + for (i = 0; i < n; i++) { + r = btrfs_qgroup_assign(fd, subvol_id, qgroups[i]); + if (r < 0 && r != -EEXIST) + return r; + if (r >= 0) + changed = true; + } + } + + return changed; +} + +int btrfs_subvol_auto_qgroup(const char *path, uint64_t subvol_id, bool create_intermediary_qgroup) { + _cleanup_close_ int fd = -1; + + fd = open(path, O_RDONLY|O_NOCTTY|O_CLOEXEC|O_DIRECTORY); + if (fd < 0) + return -errno; + + return btrfs_subvol_auto_qgroup_fd(fd, subvol_id, create_intermediary_qgroup); +} + +int btrfs_subvol_get_parent(int fd, uint64_t subvol_id, uint64_t *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_BACKREF_KEY, + .key.max_type = BTRFS_ROOT_BACKREF_KEY, + + /* No restrictions on the other components */ + .key.min_offset = 0, + .key.max_offset = (uint64_t) -1, + + .key.min_transid = 0, + .key.max_transid = (uint64_t) -1, + }; + int r; + + assert(fd >= 0); + assert(ret); + + if (subvol_id == 0) { + r = btrfs_subvol_get_id_fd(fd, &subvol_id); + if (r < 0) + return r; + } else { + r = btrfs_is_filesystem(fd); + if (r < 0) + return r; + if (!r) + return -ENOTTY; + } + + 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) { + + if (sh->type != BTRFS_ROOT_BACKREF_KEY) + continue; + if (sh->objectid != subvol_id) + continue; + + *ret = sh->offset; + return 0; + } + } + + return -ENXIO; +} |