2 * Copyright (C) 2011 STRATO. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/rbtree.h>
24 #include <linux/slab.h>
25 #include <linux/workqueue.h>
26 #include <linux/btrfs.h>
29 #include "transaction.h"
34 #include "extent_io.h"
37 * - subvol delete -> delete when ref goes to 0? delete limits also?
41 * - copy also limits on subvol creation
43 * - caches fuer ulists
44 * - performance benchmarks
45 * - check all ioctl parameters
49 * one struct for each qgroup, organized in fs_info->qgroup_tree.
57 u64 rfer; /* referenced */
58 u64 rfer_cmpr; /* referenced compressed */
59 u64 excl; /* exclusive */
60 u64 excl_cmpr; /* exclusive compressed */
65 u64 lim_flags; /* which limits are set */
72 * reservation tracking
79 struct list_head groups; /* groups this group is member of */
80 struct list_head members; /* groups that are members of this group */
81 struct list_head dirty; /* dirty groups */
82 struct rb_node node; /* tree of qgroups */
85 * temp variables for accounting operations
92 * glue structure to represent the relations between qgroups.
94 struct btrfs_qgroup_list {
95 struct list_head next_group;
96 struct list_head next_member;
97 struct btrfs_qgroup *group;
98 struct btrfs_qgroup *member;
101 struct qgroup_rescan {
102 struct btrfs_work work;
103 struct btrfs_fs_info *fs_info;
106 static void qgroup_rescan_start(struct btrfs_fs_info *fs_info,
107 struct qgroup_rescan *qscan);
109 /* must be called with qgroup_ioctl_lock held */
110 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
113 struct rb_node *n = fs_info->qgroup_tree.rb_node;
114 struct btrfs_qgroup *qgroup;
117 qgroup = rb_entry(n, struct btrfs_qgroup, node);
118 if (qgroup->qgroupid < qgroupid)
120 else if (qgroup->qgroupid > qgroupid)
128 /* must be called with qgroup_lock held */
129 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
132 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
133 struct rb_node *parent = NULL;
134 struct btrfs_qgroup *qgroup;
138 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
140 if (qgroup->qgroupid < qgroupid)
142 else if (qgroup->qgroupid > qgroupid)
148 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
150 return ERR_PTR(-ENOMEM);
152 qgroup->qgroupid = qgroupid;
153 INIT_LIST_HEAD(&qgroup->groups);
154 INIT_LIST_HEAD(&qgroup->members);
155 INIT_LIST_HEAD(&qgroup->dirty);
157 rb_link_node(&qgroup->node, parent, p);
158 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
163 /* must be called with qgroup_lock held */
164 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
166 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
167 struct btrfs_qgroup_list *list;
172 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
173 list_del(&qgroup->dirty);
175 while (!list_empty(&qgroup->groups)) {
176 list = list_first_entry(&qgroup->groups,
177 struct btrfs_qgroup_list, next_group);
178 list_del(&list->next_group);
179 list_del(&list->next_member);
183 while (!list_empty(&qgroup->members)) {
184 list = list_first_entry(&qgroup->members,
185 struct btrfs_qgroup_list, next_member);
186 list_del(&list->next_group);
187 list_del(&list->next_member);
195 /* must be called with qgroup_lock held */
196 static int add_relation_rb(struct btrfs_fs_info *fs_info,
197 u64 memberid, u64 parentid)
199 struct btrfs_qgroup *member;
200 struct btrfs_qgroup *parent;
201 struct btrfs_qgroup_list *list;
203 member = find_qgroup_rb(fs_info, memberid);
204 parent = find_qgroup_rb(fs_info, parentid);
205 if (!member || !parent)
208 list = kzalloc(sizeof(*list), GFP_ATOMIC);
212 list->group = parent;
213 list->member = member;
214 list_add_tail(&list->next_group, &member->groups);
215 list_add_tail(&list->next_member, &parent->members);
220 /* must be called with qgroup_lock held */
221 static int del_relation_rb(struct btrfs_fs_info *fs_info,
222 u64 memberid, u64 parentid)
224 struct btrfs_qgroup *member;
225 struct btrfs_qgroup *parent;
226 struct btrfs_qgroup_list *list;
228 member = find_qgroup_rb(fs_info, memberid);
229 parent = find_qgroup_rb(fs_info, parentid);
230 if (!member || !parent)
233 list_for_each_entry(list, &member->groups, next_group) {
234 if (list->group == parent) {
235 list_del(&list->next_group);
236 list_del(&list->next_member);
245 * The full config is read in one go, only called from open_ctree()
246 * It doesn't use any locking, as at this point we're still single-threaded
248 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
250 struct btrfs_key key;
251 struct btrfs_key found_key;
252 struct btrfs_root *quota_root = fs_info->quota_root;
253 struct btrfs_path *path = NULL;
254 struct extent_buffer *l;
259 if (!fs_info->quota_enabled)
262 path = btrfs_alloc_path();
268 /* default this to quota off, in case no status key is found */
269 fs_info->qgroup_flags = 0;
272 * pass 1: read status, all qgroup infos and limits
277 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
282 struct btrfs_qgroup *qgroup;
284 slot = path->slots[0];
286 btrfs_item_key_to_cpu(l, &found_key, slot);
288 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
289 struct btrfs_qgroup_status_item *ptr;
291 ptr = btrfs_item_ptr(l, slot,
292 struct btrfs_qgroup_status_item);
294 if (btrfs_qgroup_status_version(l, ptr) !=
295 BTRFS_QGROUP_STATUS_VERSION) {
297 "btrfs: old qgroup version, quota disabled\n");
300 if (btrfs_qgroup_status_generation(l, ptr) !=
301 fs_info->generation) {
302 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
304 "btrfs: qgroup generation mismatch, "
305 "marked as inconsistent\n");
307 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
309 fs_info->qgroup_rescan_progress.objectid =
310 btrfs_qgroup_status_rescan(l, ptr);
311 if (fs_info->qgroup_flags &
312 BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
313 struct qgroup_rescan *qscan =
314 kmalloc(sizeof(*qscan), GFP_NOFS);
319 fs_info->qgroup_rescan_progress.type = 0;
320 fs_info->qgroup_rescan_progress.offset = 0;
321 qgroup_rescan_start(fs_info, qscan);
326 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
327 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
330 qgroup = find_qgroup_rb(fs_info, found_key.offset);
331 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
332 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
333 printk(KERN_ERR "btrfs: inconsitent qgroup config\n");
334 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
337 qgroup = add_qgroup_rb(fs_info, found_key.offset);
338 if (IS_ERR(qgroup)) {
339 ret = PTR_ERR(qgroup);
343 switch (found_key.type) {
344 case BTRFS_QGROUP_INFO_KEY: {
345 struct btrfs_qgroup_info_item *ptr;
347 ptr = btrfs_item_ptr(l, slot,
348 struct btrfs_qgroup_info_item);
349 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
350 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
351 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
352 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
353 /* generation currently unused */
356 case BTRFS_QGROUP_LIMIT_KEY: {
357 struct btrfs_qgroup_limit_item *ptr;
359 ptr = btrfs_item_ptr(l, slot,
360 struct btrfs_qgroup_limit_item);
361 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
362 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
363 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
364 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
365 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
370 ret = btrfs_next_item(quota_root, path);
376 btrfs_release_path(path);
379 * pass 2: read all qgroup relations
382 key.type = BTRFS_QGROUP_RELATION_KEY;
384 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
388 slot = path->slots[0];
390 btrfs_item_key_to_cpu(l, &found_key, slot);
392 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
395 if (found_key.objectid > found_key.offset) {
396 /* parent <- member, not needed to build config */
397 /* FIXME should we omit the key completely? */
401 ret = add_relation_rb(fs_info, found_key.objectid,
403 if (ret == -ENOENT) {
405 "btrfs: orphan qgroup relation 0x%llx->0x%llx\n",
406 (unsigned long long)found_key.objectid,
407 (unsigned long long)found_key.offset);
408 ret = 0; /* ignore the error */
413 ret = btrfs_next_item(quota_root, path);
420 fs_info->qgroup_flags |= flags;
421 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
422 fs_info->quota_enabled = 0;
423 fs_info->pending_quota_state = 0;
425 btrfs_free_path(path);
427 return ret < 0 ? ret : 0;
431 * This is only called from close_ctree() or open_ctree(), both in single-
432 * treaded paths. Clean up the in-memory structures. No locking needed.
434 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
437 struct btrfs_qgroup *qgroup;
438 struct btrfs_qgroup_list *list;
440 while ((n = rb_first(&fs_info->qgroup_tree))) {
441 qgroup = rb_entry(n, struct btrfs_qgroup, node);
442 rb_erase(n, &fs_info->qgroup_tree);
444 while (!list_empty(&qgroup->groups)) {
445 list = list_first_entry(&qgroup->groups,
446 struct btrfs_qgroup_list,
448 list_del(&list->next_group);
449 list_del(&list->next_member);
453 while (!list_empty(&qgroup->members)) {
454 list = list_first_entry(&qgroup->members,
455 struct btrfs_qgroup_list,
457 list_del(&list->next_group);
458 list_del(&list->next_member);
465 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
466 struct btrfs_root *quota_root,
470 struct btrfs_path *path;
471 struct btrfs_key key;
473 path = btrfs_alloc_path();
478 key.type = BTRFS_QGROUP_RELATION_KEY;
481 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
483 btrfs_mark_buffer_dirty(path->nodes[0]);
485 btrfs_free_path(path);
489 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
490 struct btrfs_root *quota_root,
494 struct btrfs_path *path;
495 struct btrfs_key key;
497 path = btrfs_alloc_path();
502 key.type = BTRFS_QGROUP_RELATION_KEY;
505 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
514 ret = btrfs_del_item(trans, quota_root, path);
516 btrfs_free_path(path);
520 static int add_qgroup_item(struct btrfs_trans_handle *trans,
521 struct btrfs_root *quota_root, u64 qgroupid)
524 struct btrfs_path *path;
525 struct btrfs_qgroup_info_item *qgroup_info;
526 struct btrfs_qgroup_limit_item *qgroup_limit;
527 struct extent_buffer *leaf;
528 struct btrfs_key key;
530 path = btrfs_alloc_path();
535 key.type = BTRFS_QGROUP_INFO_KEY;
536 key.offset = qgroupid;
538 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
539 sizeof(*qgroup_info));
543 leaf = path->nodes[0];
544 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
545 struct btrfs_qgroup_info_item);
546 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
547 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
548 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
549 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
550 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
552 btrfs_mark_buffer_dirty(leaf);
554 btrfs_release_path(path);
556 key.type = BTRFS_QGROUP_LIMIT_KEY;
557 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
558 sizeof(*qgroup_limit));
562 leaf = path->nodes[0];
563 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
564 struct btrfs_qgroup_limit_item);
565 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
566 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
567 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
568 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
569 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
571 btrfs_mark_buffer_dirty(leaf);
575 btrfs_free_path(path);
579 static int del_qgroup_item(struct btrfs_trans_handle *trans,
580 struct btrfs_root *quota_root, u64 qgroupid)
583 struct btrfs_path *path;
584 struct btrfs_key key;
586 path = btrfs_alloc_path();
591 key.type = BTRFS_QGROUP_INFO_KEY;
592 key.offset = qgroupid;
593 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
602 ret = btrfs_del_item(trans, quota_root, path);
606 btrfs_release_path(path);
608 key.type = BTRFS_QGROUP_LIMIT_KEY;
609 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
618 ret = btrfs_del_item(trans, quota_root, path);
621 btrfs_free_path(path);
625 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
626 struct btrfs_root *root, u64 qgroupid,
627 u64 flags, u64 max_rfer, u64 max_excl,
628 u64 rsv_rfer, u64 rsv_excl)
630 struct btrfs_path *path;
631 struct btrfs_key key;
632 struct extent_buffer *l;
633 struct btrfs_qgroup_limit_item *qgroup_limit;
638 key.type = BTRFS_QGROUP_LIMIT_KEY;
639 key.offset = qgroupid;
641 path = btrfs_alloc_path();
645 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
653 slot = path->slots[0];
654 qgroup_limit = btrfs_item_ptr(l, path->slots[0],
655 struct btrfs_qgroup_limit_item);
656 btrfs_set_qgroup_limit_flags(l, qgroup_limit, flags);
657 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, max_rfer);
658 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, max_excl);
659 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, rsv_rfer);
660 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, rsv_excl);
662 btrfs_mark_buffer_dirty(l);
665 btrfs_free_path(path);
669 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
670 struct btrfs_root *root,
671 struct btrfs_qgroup *qgroup)
673 struct btrfs_path *path;
674 struct btrfs_key key;
675 struct extent_buffer *l;
676 struct btrfs_qgroup_info_item *qgroup_info;
681 key.type = BTRFS_QGROUP_INFO_KEY;
682 key.offset = qgroup->qgroupid;
684 path = btrfs_alloc_path();
688 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
696 slot = path->slots[0];
697 qgroup_info = btrfs_item_ptr(l, path->slots[0],
698 struct btrfs_qgroup_info_item);
699 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
700 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
701 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
702 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
703 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
705 btrfs_mark_buffer_dirty(l);
708 btrfs_free_path(path);
712 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
713 struct btrfs_fs_info *fs_info,
714 struct btrfs_root *root)
716 struct btrfs_path *path;
717 struct btrfs_key key;
718 struct extent_buffer *l;
719 struct btrfs_qgroup_status_item *ptr;
724 key.type = BTRFS_QGROUP_STATUS_KEY;
727 path = btrfs_alloc_path();
731 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
739 slot = path->slots[0];
740 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
741 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
742 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
743 btrfs_set_qgroup_status_rescan(l, ptr,
744 fs_info->qgroup_rescan_progress.objectid);
746 btrfs_mark_buffer_dirty(l);
749 btrfs_free_path(path);
754 * called with qgroup_lock held
756 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
757 struct btrfs_root *root)
759 struct btrfs_path *path;
760 struct btrfs_key key;
761 struct extent_buffer *leaf = NULL;
765 path = btrfs_alloc_path();
769 path->leave_spinning = 1;
776 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
779 leaf = path->nodes[0];
780 nr = btrfs_header_nritems(leaf);
784 * delete the leaf one by one
785 * since the whole tree is going
789 ret = btrfs_del_items(trans, root, path, 0, nr);
793 btrfs_release_path(path);
797 root->fs_info->pending_quota_state = 0;
798 btrfs_free_path(path);
802 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
803 struct btrfs_fs_info *fs_info)
805 struct btrfs_root *quota_root;
806 struct btrfs_root *tree_root = fs_info->tree_root;
807 struct btrfs_path *path = NULL;
808 struct btrfs_qgroup_status_item *ptr;
809 struct extent_buffer *leaf;
810 struct btrfs_key key;
811 struct btrfs_key found_key;
812 struct btrfs_qgroup *qgroup = NULL;
816 mutex_lock(&fs_info->qgroup_ioctl_lock);
817 if (fs_info->quota_root) {
818 fs_info->pending_quota_state = 1;
823 * initially create the quota tree
825 quota_root = btrfs_create_tree(trans, fs_info,
826 BTRFS_QUOTA_TREE_OBJECTID);
827 if (IS_ERR(quota_root)) {
828 ret = PTR_ERR(quota_root);
832 path = btrfs_alloc_path();
839 key.type = BTRFS_QGROUP_STATUS_KEY;
842 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
847 leaf = path->nodes[0];
848 ptr = btrfs_item_ptr(leaf, path->slots[0],
849 struct btrfs_qgroup_status_item);
850 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
851 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
852 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
853 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
854 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
855 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
857 btrfs_mark_buffer_dirty(leaf);
860 key.type = BTRFS_ROOT_REF_KEY;
863 btrfs_release_path(path);
864 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
872 slot = path->slots[0];
873 leaf = path->nodes[0];
874 btrfs_item_key_to_cpu(leaf, &found_key, slot);
876 if (found_key.type == BTRFS_ROOT_REF_KEY) {
877 ret = add_qgroup_item(trans, quota_root,
882 qgroup = add_qgroup_rb(fs_info, found_key.offset);
883 if (IS_ERR(qgroup)) {
884 ret = PTR_ERR(qgroup);
888 ret = btrfs_next_item(tree_root, path);
896 btrfs_release_path(path);
897 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
901 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
902 if (IS_ERR(qgroup)) {
903 ret = PTR_ERR(qgroup);
906 spin_lock(&fs_info->qgroup_lock);
907 fs_info->quota_root = quota_root;
908 fs_info->pending_quota_state = 1;
909 spin_unlock(&fs_info->qgroup_lock);
911 btrfs_free_path(path);
914 free_extent_buffer(quota_root->node);
915 free_extent_buffer(quota_root->commit_root);
919 mutex_unlock(&fs_info->qgroup_ioctl_lock);
923 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
924 struct btrfs_fs_info *fs_info)
926 struct btrfs_root *tree_root = fs_info->tree_root;
927 struct btrfs_root *quota_root;
930 mutex_lock(&fs_info->qgroup_ioctl_lock);
931 if (!fs_info->quota_root)
933 spin_lock(&fs_info->qgroup_lock);
934 fs_info->quota_enabled = 0;
935 fs_info->pending_quota_state = 0;
936 quota_root = fs_info->quota_root;
937 fs_info->quota_root = NULL;
938 btrfs_free_qgroup_config(fs_info);
939 spin_unlock(&fs_info->qgroup_lock);
946 ret = btrfs_clean_quota_tree(trans, quota_root);
950 ret = btrfs_del_root(trans, tree_root, "a_root->root_key);
954 list_del("a_root->dirty_list);
956 btrfs_tree_lock(quota_root->node);
957 clean_tree_block(trans, tree_root, quota_root->node);
958 btrfs_tree_unlock(quota_root->node);
959 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
961 free_extent_buffer(quota_root->node);
962 free_extent_buffer(quota_root->commit_root);
965 mutex_unlock(&fs_info->qgroup_ioctl_lock);
969 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
970 struct btrfs_qgroup *qgroup)
972 if (list_empty(&qgroup->dirty))
973 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
976 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
977 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
979 struct btrfs_root *quota_root;
980 struct btrfs_qgroup *parent;
981 struct btrfs_qgroup *member;
982 struct btrfs_qgroup_list *list;
985 mutex_lock(&fs_info->qgroup_ioctl_lock);
986 quota_root = fs_info->quota_root;
991 member = find_qgroup_rb(fs_info, src);
992 parent = find_qgroup_rb(fs_info, dst);
993 if (!member || !parent) {
998 /* check if such qgroup relation exist firstly */
999 list_for_each_entry(list, &member->groups, next_group) {
1000 if (list->group == parent) {
1006 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1010 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1012 del_qgroup_relation_item(trans, quota_root, src, dst);
1016 spin_lock(&fs_info->qgroup_lock);
1017 ret = add_relation_rb(quota_root->fs_info, src, dst);
1018 spin_unlock(&fs_info->qgroup_lock);
1020 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1024 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1025 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1027 struct btrfs_root *quota_root;
1028 struct btrfs_qgroup *parent;
1029 struct btrfs_qgroup *member;
1030 struct btrfs_qgroup_list *list;
1034 mutex_lock(&fs_info->qgroup_ioctl_lock);
1035 quota_root = fs_info->quota_root;
1041 member = find_qgroup_rb(fs_info, src);
1042 parent = find_qgroup_rb(fs_info, dst);
1043 if (!member || !parent) {
1048 /* check if such qgroup relation exist firstly */
1049 list_for_each_entry(list, &member->groups, next_group) {
1050 if (list->group == parent)
1056 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1057 err = del_qgroup_relation_item(trans, quota_root, dst, src);
1061 spin_lock(&fs_info->qgroup_lock);
1062 del_relation_rb(fs_info, src, dst);
1063 spin_unlock(&fs_info->qgroup_lock);
1065 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1069 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1070 struct btrfs_fs_info *fs_info, u64 qgroupid, char *name)
1072 struct btrfs_root *quota_root;
1073 struct btrfs_qgroup *qgroup;
1076 mutex_lock(&fs_info->qgroup_ioctl_lock);
1077 quota_root = fs_info->quota_root;
1082 qgroup = find_qgroup_rb(fs_info, qgroupid);
1088 ret = add_qgroup_item(trans, quota_root, qgroupid);
1092 spin_lock(&fs_info->qgroup_lock);
1093 qgroup = add_qgroup_rb(fs_info, qgroupid);
1094 spin_unlock(&fs_info->qgroup_lock);
1097 ret = PTR_ERR(qgroup);
1099 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1103 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1104 struct btrfs_fs_info *fs_info, u64 qgroupid)
1106 struct btrfs_root *quota_root;
1107 struct btrfs_qgroup *qgroup;
1110 mutex_lock(&fs_info->qgroup_ioctl_lock);
1111 quota_root = fs_info->quota_root;
1117 qgroup = find_qgroup_rb(fs_info, qgroupid);
1122 /* check if there are no relations to this qgroup */
1123 if (!list_empty(&qgroup->groups) ||
1124 !list_empty(&qgroup->members)) {
1129 ret = del_qgroup_item(trans, quota_root, qgroupid);
1131 spin_lock(&fs_info->qgroup_lock);
1132 del_qgroup_rb(quota_root->fs_info, qgroupid);
1133 spin_unlock(&fs_info->qgroup_lock);
1135 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1139 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1140 struct btrfs_fs_info *fs_info, u64 qgroupid,
1141 struct btrfs_qgroup_limit *limit)
1143 struct btrfs_root *quota_root;
1144 struct btrfs_qgroup *qgroup;
1147 mutex_lock(&fs_info->qgroup_ioctl_lock);
1148 quota_root = fs_info->quota_root;
1154 qgroup = find_qgroup_rb(fs_info, qgroupid);
1159 ret = update_qgroup_limit_item(trans, quota_root, qgroupid,
1160 limit->flags, limit->max_rfer,
1161 limit->max_excl, limit->rsv_rfer,
1164 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1165 printk(KERN_INFO "unable to update quota limit for %llu\n",
1166 (unsigned long long)qgroupid);
1169 spin_lock(&fs_info->qgroup_lock);
1170 qgroup->lim_flags = limit->flags;
1171 qgroup->max_rfer = limit->max_rfer;
1172 qgroup->max_excl = limit->max_excl;
1173 qgroup->rsv_rfer = limit->rsv_rfer;
1174 qgroup->rsv_excl = limit->rsv_excl;
1175 spin_unlock(&fs_info->qgroup_lock);
1177 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1182 * btrfs_qgroup_record_ref is called when the ref is added or deleted. it puts
1183 * the modification into a list that's later used by btrfs_end_transaction to
1184 * pass the recorded modifications on to btrfs_qgroup_account_ref.
1186 int btrfs_qgroup_record_ref(struct btrfs_trans_handle *trans,
1187 struct btrfs_delayed_ref_node *node,
1188 struct btrfs_delayed_extent_op *extent_op)
1190 struct qgroup_update *u;
1192 BUG_ON(!trans->delayed_ref_elem.seq);
1193 u = kmalloc(sizeof(*u), GFP_NOFS);
1198 u->extent_op = extent_op;
1199 list_add_tail(&u->list, &trans->qgroup_ref_list);
1204 static int qgroup_account_ref_step1(struct btrfs_fs_info *fs_info,
1205 struct ulist *roots, struct ulist *tmp,
1208 struct ulist_node *unode;
1209 struct ulist_iterator uiter;
1210 struct ulist_node *tmp_unode;
1211 struct ulist_iterator tmp_uiter;
1212 struct btrfs_qgroup *qg;
1215 ULIST_ITER_INIT(&uiter);
1216 while ((unode = ulist_next(roots, &uiter))) {
1217 qg = find_qgroup_rb(fs_info, unode->val);
1222 /* XXX id not needed */
1223 ret = ulist_add(tmp, qg->qgroupid,
1224 (u64)(uintptr_t)qg, GFP_ATOMIC);
1227 ULIST_ITER_INIT(&tmp_uiter);
1228 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1229 struct btrfs_qgroup_list *glist;
1231 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1232 if (qg->refcnt < seq)
1233 qg->refcnt = seq + 1;
1237 list_for_each_entry(glist, &qg->groups, next_group) {
1238 ret = ulist_add(tmp, glist->group->qgroupid,
1239 (u64)(uintptr_t)glist->group,
1250 static int qgroup_account_ref_step2(struct btrfs_fs_info *fs_info,
1251 struct ulist *roots, struct ulist *tmp,
1252 u64 seq, int sgn, u64 num_bytes,
1253 struct btrfs_qgroup *qgroup)
1255 struct ulist_node *unode;
1256 struct ulist_iterator uiter;
1257 struct btrfs_qgroup *qg;
1258 struct btrfs_qgroup_list *glist;
1262 ret = ulist_add(tmp, qgroup->qgroupid, (uintptr_t)qgroup, GFP_ATOMIC);
1266 ULIST_ITER_INIT(&uiter);
1267 while ((unode = ulist_next(tmp, &uiter))) {
1268 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1269 if (qg->refcnt < seq) {
1270 /* not visited by step 1 */
1271 qg->rfer += sgn * num_bytes;
1272 qg->rfer_cmpr += sgn * num_bytes;
1273 if (roots->nnodes == 0) {
1274 qg->excl += sgn * num_bytes;
1275 qg->excl_cmpr += sgn * num_bytes;
1277 qgroup_dirty(fs_info, qg);
1279 WARN_ON(qg->tag >= seq);
1282 list_for_each_entry(glist, &qg->groups, next_group) {
1283 ret = ulist_add(tmp, glist->group->qgroupid,
1284 (uintptr_t)glist->group, GFP_ATOMIC);
1293 static int qgroup_account_ref_step3(struct btrfs_fs_info *fs_info,
1294 struct ulist *roots, struct ulist *tmp,
1295 u64 seq, int sgn, u64 num_bytes)
1297 struct ulist_node *unode;
1298 struct ulist_iterator uiter;
1299 struct btrfs_qgroup *qg;
1300 struct ulist_node *tmp_unode;
1301 struct ulist_iterator tmp_uiter;
1304 ULIST_ITER_INIT(&uiter);
1305 while ((unode = ulist_next(roots, &uiter))) {
1306 qg = find_qgroup_rb(fs_info, unode->val);
1311 ret = ulist_add(tmp, qg->qgroupid, (uintptr_t)qg, GFP_ATOMIC);
1315 ULIST_ITER_INIT(&tmp_uiter);
1316 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1317 struct btrfs_qgroup_list *glist;
1319 qg = (struct btrfs_qgroup *)(uintptr_t)tmp_unode->aux;
1323 if (qg->refcnt - seq == roots->nnodes) {
1324 qg->excl -= sgn * num_bytes;
1325 qg->excl_cmpr -= sgn * num_bytes;
1326 qgroup_dirty(fs_info, qg);
1329 list_for_each_entry(glist, &qg->groups, next_group) {
1330 ret = ulist_add(tmp, glist->group->qgroupid,
1331 (uintptr_t)glist->group,
1343 * btrfs_qgroup_account_ref is called for every ref that is added to or deleted
1344 * from the fs. First, all roots referencing the extent are searched, and
1345 * then the space is accounted accordingly to the different roots. The
1346 * accounting algorithm works in 3 steps documented inline.
1348 int btrfs_qgroup_account_ref(struct btrfs_trans_handle *trans,
1349 struct btrfs_fs_info *fs_info,
1350 struct btrfs_delayed_ref_node *node,
1351 struct btrfs_delayed_extent_op *extent_op)
1353 struct btrfs_key ins;
1354 struct btrfs_root *quota_root;
1356 struct btrfs_qgroup *qgroup;
1357 struct ulist *roots = NULL;
1358 struct ulist *tmp = NULL;
1363 if (!fs_info->quota_enabled)
1366 BUG_ON(!fs_info->quota_root);
1368 ins.objectid = node->bytenr;
1369 ins.offset = node->num_bytes;
1370 ins.type = BTRFS_EXTENT_ITEM_KEY;
1372 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
1373 node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
1374 struct btrfs_delayed_tree_ref *ref;
1375 ref = btrfs_delayed_node_to_tree_ref(node);
1376 ref_root = ref->root;
1377 } else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
1378 node->type == BTRFS_SHARED_DATA_REF_KEY) {
1379 struct btrfs_delayed_data_ref *ref;
1380 ref = btrfs_delayed_node_to_data_ref(node);
1381 ref_root = ref->root;
1386 if (!is_fstree(ref_root)) {
1388 * non-fs-trees are not being accounted
1393 switch (node->action) {
1394 case BTRFS_ADD_DELAYED_REF:
1395 case BTRFS_ADD_DELAYED_EXTENT:
1397 seq = btrfs_tree_mod_seq_prev(node->seq);
1399 case BTRFS_DROP_DELAYED_REF:
1403 case BTRFS_UPDATE_DELAYED_HEAD:
1409 mutex_lock(&fs_info->qgroup_rescan_lock);
1410 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1411 if (fs_info->qgroup_rescan_progress.objectid <= node->bytenr) {
1412 mutex_unlock(&fs_info->qgroup_rescan_lock);
1416 mutex_unlock(&fs_info->qgroup_rescan_lock);
1419 * the delayed ref sequence number we pass depends on the direction of
1420 * the operation. for add operations, we pass
1421 * tree_mod_log_prev_seq(node->seq) to skip
1422 * the delayed ref's current sequence number, because we need the state
1423 * of the tree before the add operation. for delete operations, we pass
1424 * (node->seq) to include the delayed ref's current sequence number,
1425 * because we need the state of the tree after the delete operation.
1427 ret = btrfs_find_all_roots(trans, fs_info, node->bytenr, seq, &roots);
1431 mutex_lock(&fs_info->qgroup_rescan_lock);
1432 spin_lock(&fs_info->qgroup_lock);
1433 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1434 if (fs_info->qgroup_rescan_progress.objectid <= node->bytenr) {
1440 quota_root = fs_info->quota_root;
1444 qgroup = find_qgroup_rb(fs_info, ref_root);
1449 * step 1: for each old ref, visit all nodes once and inc refcnt
1451 tmp = ulist_alloc(GFP_ATOMIC);
1456 seq = fs_info->qgroup_seq;
1457 fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1459 ret = qgroup_account_ref_step1(fs_info, roots, tmp, seq);
1464 * step 2: walk from the new root
1466 ret = qgroup_account_ref_step2(fs_info, roots, tmp, seq, sgn,
1467 node->num_bytes, qgroup);
1472 * step 3: walk again from old refs
1474 ret = qgroup_account_ref_step3(fs_info, roots, tmp, seq, sgn,
1480 spin_unlock(&fs_info->qgroup_lock);
1481 mutex_unlock(&fs_info->qgroup_rescan_lock);
1489 * called from commit_transaction. Writes all changed qgroups to disk.
1491 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1492 struct btrfs_fs_info *fs_info)
1494 struct btrfs_root *quota_root = fs_info->quota_root;
1496 int start_rescan_worker = 0;
1501 if (!fs_info->quota_enabled && fs_info->pending_quota_state)
1502 start_rescan_worker = 1;
1504 fs_info->quota_enabled = fs_info->pending_quota_state;
1506 spin_lock(&fs_info->qgroup_lock);
1507 while (!list_empty(&fs_info->dirty_qgroups)) {
1508 struct btrfs_qgroup *qgroup;
1509 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1510 struct btrfs_qgroup, dirty);
1511 list_del_init(&qgroup->dirty);
1512 spin_unlock(&fs_info->qgroup_lock);
1513 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1515 fs_info->qgroup_flags |=
1516 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1517 spin_lock(&fs_info->qgroup_lock);
1519 if (fs_info->quota_enabled)
1520 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1522 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1523 spin_unlock(&fs_info->qgroup_lock);
1525 ret = update_qgroup_status_item(trans, fs_info, quota_root);
1527 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1529 if (!ret && start_rescan_worker) {
1530 ret = btrfs_qgroup_rescan(fs_info);
1532 pr_err("btrfs: start rescan quota failed: %d\n", ret);
1542 * copy the acounting information between qgroups. This is necessary when a
1543 * snapshot or a subvolume is created
1545 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1546 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1547 struct btrfs_qgroup_inherit *inherit)
1552 struct btrfs_root *quota_root = fs_info->quota_root;
1553 struct btrfs_qgroup *srcgroup;
1554 struct btrfs_qgroup *dstgroup;
1558 mutex_lock(&fs_info->qgroup_ioctl_lock);
1559 if (!fs_info->quota_enabled)
1568 i_qgroups = (u64 *)(inherit + 1);
1569 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1570 2 * inherit->num_excl_copies;
1571 for (i = 0; i < nums; ++i) {
1572 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
1582 * create a tracking group for the subvol itself
1584 ret = add_qgroup_item(trans, quota_root, objectid);
1588 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1589 ret = update_qgroup_limit_item(trans, quota_root, objectid,
1591 inherit->lim.max_rfer,
1592 inherit->lim.max_excl,
1593 inherit->lim.rsv_rfer,
1594 inherit->lim.rsv_excl);
1600 struct btrfs_root *srcroot;
1601 struct btrfs_key srckey;
1604 srckey.objectid = srcid;
1605 srckey.type = BTRFS_ROOT_ITEM_KEY;
1606 srckey.offset = (u64)-1;
1607 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1608 if (IS_ERR(srcroot)) {
1609 ret = PTR_ERR(srcroot);
1614 srcroot_level = btrfs_header_level(srcroot->node);
1615 level_size = btrfs_level_size(srcroot, srcroot_level);
1620 * add qgroup to all inherited groups
1623 i_qgroups = (u64 *)(inherit + 1);
1624 for (i = 0; i < inherit->num_qgroups; ++i) {
1625 ret = add_qgroup_relation_item(trans, quota_root,
1626 objectid, *i_qgroups);
1629 ret = add_qgroup_relation_item(trans, quota_root,
1630 *i_qgroups, objectid);
1638 spin_lock(&fs_info->qgroup_lock);
1640 dstgroup = add_qgroup_rb(fs_info, objectid);
1641 if (IS_ERR(dstgroup)) {
1642 ret = PTR_ERR(dstgroup);
1647 srcgroup = find_qgroup_rb(fs_info, srcid);
1650 dstgroup->rfer = srcgroup->rfer - level_size;
1651 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr - level_size;
1652 srcgroup->excl = level_size;
1653 srcgroup->excl_cmpr = level_size;
1654 qgroup_dirty(fs_info, dstgroup);
1655 qgroup_dirty(fs_info, srcgroup);
1661 i_qgroups = (u64 *)(inherit + 1);
1662 for (i = 0; i < inherit->num_qgroups; ++i) {
1663 ret = add_relation_rb(quota_root->fs_info, objectid,
1670 for (i = 0; i < inherit->num_ref_copies; ++i) {
1671 struct btrfs_qgroup *src;
1672 struct btrfs_qgroup *dst;
1674 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1675 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1682 dst->rfer = src->rfer - level_size;
1683 dst->rfer_cmpr = src->rfer_cmpr - level_size;
1686 for (i = 0; i < inherit->num_excl_copies; ++i) {
1687 struct btrfs_qgroup *src;
1688 struct btrfs_qgroup *dst;
1690 src = find_qgroup_rb(fs_info, i_qgroups[0]);
1691 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
1698 dst->excl = src->excl + level_size;
1699 dst->excl_cmpr = src->excl_cmpr + level_size;
1704 spin_unlock(&fs_info->qgroup_lock);
1706 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1711 * reserve some space for a qgroup and all its parents. The reservation takes
1712 * place with start_transaction or dealloc_reserve, similar to ENOSPC
1713 * accounting. If not enough space is available, EDQUOT is returned.
1714 * We assume that the requested space is new for all qgroups.
1716 int btrfs_qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
1718 struct btrfs_root *quota_root;
1719 struct btrfs_qgroup *qgroup;
1720 struct btrfs_fs_info *fs_info = root->fs_info;
1721 u64 ref_root = root->root_key.objectid;
1723 struct ulist *ulist = NULL;
1724 struct ulist_node *unode;
1725 struct ulist_iterator uiter;
1727 if (!is_fstree(ref_root))
1733 spin_lock(&fs_info->qgroup_lock);
1734 quota_root = fs_info->quota_root;
1738 qgroup = find_qgroup_rb(fs_info, ref_root);
1743 * in a first step, we check all affected qgroups if any limits would
1746 ulist = ulist_alloc(GFP_ATOMIC);
1751 ret = ulist_add(ulist, qgroup->qgroupid,
1752 (uintptr_t)qgroup, GFP_ATOMIC);
1755 ULIST_ITER_INIT(&uiter);
1756 while ((unode = ulist_next(ulist, &uiter))) {
1757 struct btrfs_qgroup *qg;
1758 struct btrfs_qgroup_list *glist;
1760 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1762 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
1763 qg->reserved + (s64)qg->rfer + num_bytes >
1769 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
1770 qg->reserved + (s64)qg->excl + num_bytes >
1776 list_for_each_entry(glist, &qg->groups, next_group) {
1777 ret = ulist_add(ulist, glist->group->qgroupid,
1778 (uintptr_t)glist->group, GFP_ATOMIC);
1785 * no limits exceeded, now record the reservation into all qgroups
1787 ULIST_ITER_INIT(&uiter);
1788 while ((unode = ulist_next(ulist, &uiter))) {
1789 struct btrfs_qgroup *qg;
1791 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1793 qg->reserved += num_bytes;
1797 spin_unlock(&fs_info->qgroup_lock);
1803 void btrfs_qgroup_free(struct btrfs_root *root, u64 num_bytes)
1805 struct btrfs_root *quota_root;
1806 struct btrfs_qgroup *qgroup;
1807 struct btrfs_fs_info *fs_info = root->fs_info;
1808 struct ulist *ulist = NULL;
1809 struct ulist_node *unode;
1810 struct ulist_iterator uiter;
1811 u64 ref_root = root->root_key.objectid;
1814 if (!is_fstree(ref_root))
1820 spin_lock(&fs_info->qgroup_lock);
1822 quota_root = fs_info->quota_root;
1826 qgroup = find_qgroup_rb(fs_info, ref_root);
1830 ulist = ulist_alloc(GFP_ATOMIC);
1832 btrfs_std_error(fs_info, -ENOMEM);
1835 ret = ulist_add(ulist, qgroup->qgroupid,
1836 (uintptr_t)qgroup, GFP_ATOMIC);
1839 ULIST_ITER_INIT(&uiter);
1840 while ((unode = ulist_next(ulist, &uiter))) {
1841 struct btrfs_qgroup *qg;
1842 struct btrfs_qgroup_list *glist;
1844 qg = (struct btrfs_qgroup *)(uintptr_t)unode->aux;
1846 qg->reserved -= num_bytes;
1848 list_for_each_entry(glist, &qg->groups, next_group) {
1849 ret = ulist_add(ulist, glist->group->qgroupid,
1850 (uintptr_t)glist->group, GFP_ATOMIC);
1857 spin_unlock(&fs_info->qgroup_lock);
1861 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
1863 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
1865 pr_err("btrfs: qgroups not uptodate in trans handle %p: list is%s empty, seq is %#x.%x\n",
1866 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
1867 (u32)(trans->delayed_ref_elem.seq >> 32),
1868 (u32)trans->delayed_ref_elem.seq);
1873 * returns < 0 on error, 0 when more leafs are to be scanned.
1874 * returns 1 when done, 2 when done and FLAG_INCONSISTENT was cleared.
1877 qgroup_rescan_leaf(struct qgroup_rescan *qscan, struct btrfs_path *path,
1878 struct btrfs_trans_handle *trans, struct ulist *tmp,
1879 struct extent_buffer *scratch_leaf)
1881 struct btrfs_key found;
1882 struct btrfs_fs_info *fs_info = qscan->fs_info;
1883 struct ulist *roots = NULL;
1884 struct ulist_node *unode;
1885 struct ulist_iterator uiter;
1886 struct seq_list tree_mod_seq_elem = {};
1891 path->leave_spinning = 1;
1892 mutex_lock(&fs_info->qgroup_rescan_lock);
1893 ret = btrfs_search_slot_for_read(fs_info->extent_root,
1894 &fs_info->qgroup_rescan_progress,
1897 pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
1898 (unsigned long long)fs_info->qgroup_rescan_progress.objectid,
1899 fs_info->qgroup_rescan_progress.type,
1900 (unsigned long long)fs_info->qgroup_rescan_progress.offset,
1905 * The rescan is about to end, we will not be scanning any
1906 * further blocks. We cannot unset the RESCAN flag here, because
1907 * we want to commit the transaction if everything went well.
1908 * To make the live accounting work in this phase, we set our
1909 * scan progress pointer such that every real extent objectid
1912 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
1913 btrfs_release_path(path);
1914 mutex_unlock(&fs_info->qgroup_rescan_lock);
1918 btrfs_item_key_to_cpu(path->nodes[0], &found,
1919 btrfs_header_nritems(path->nodes[0]) - 1);
1920 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
1922 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1923 memcpy(scratch_leaf, path->nodes[0], sizeof(*scratch_leaf));
1924 slot = path->slots[0];
1925 btrfs_release_path(path);
1926 mutex_unlock(&fs_info->qgroup_rescan_lock);
1928 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
1929 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
1930 if (found.type != BTRFS_EXTENT_ITEM_KEY)
1932 ret = btrfs_find_all_roots(trans, fs_info, found.objectid,
1933 tree_mod_seq_elem.seq, &roots);
1936 spin_lock(&fs_info->qgroup_lock);
1937 seq = fs_info->qgroup_seq;
1938 fs_info->qgroup_seq += roots->nnodes + 1; /* max refcnt */
1940 ret = qgroup_account_ref_step1(fs_info, roots, tmp, seq);
1942 spin_unlock(&fs_info->qgroup_lock);
1948 * step2 of btrfs_qgroup_account_ref works from a single root,
1949 * we're doing all at once here.
1952 ULIST_ITER_INIT(&uiter);
1953 while ((unode = ulist_next(roots, &uiter))) {
1954 struct btrfs_qgroup *qg;
1956 qg = find_qgroup_rb(fs_info, unode->val);
1960 ret = ulist_add(tmp, qg->qgroupid, (uintptr_t)qg,
1963 spin_unlock(&fs_info->qgroup_lock);
1969 /* this loop is similar to step 2 of btrfs_qgroup_account_ref */
1970 ULIST_ITER_INIT(&uiter);
1971 while ((unode = ulist_next(tmp, &uiter))) {
1972 struct btrfs_qgroup *qg;
1973 struct btrfs_qgroup_list *glist;
1975 qg = (struct btrfs_qgroup *)(uintptr_t) unode->aux;
1976 qg->rfer += found.offset;
1977 qg->rfer_cmpr += found.offset;
1978 WARN_ON(qg->tag >= seq);
1979 if (qg->refcnt - seq == roots->nnodes) {
1980 qg->excl += found.offset;
1981 qg->excl_cmpr += found.offset;
1983 qgroup_dirty(fs_info, qg);
1985 list_for_each_entry(glist, &qg->groups, next_group) {
1986 ret = ulist_add(tmp, glist->group->qgroupid,
1987 (uintptr_t)glist->group,
1990 spin_unlock(&fs_info->qgroup_lock);
1997 spin_unlock(&fs_info->qgroup_lock);
2003 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2008 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2010 struct qgroup_rescan *qscan = container_of(work, struct qgroup_rescan,
2012 struct btrfs_path *path;
2013 struct btrfs_trans_handle *trans = NULL;
2014 struct btrfs_fs_info *fs_info = qscan->fs_info;
2015 struct ulist *tmp = NULL;
2016 struct extent_buffer *scratch_leaf = NULL;
2019 path = btrfs_alloc_path();
2022 tmp = ulist_alloc(GFP_NOFS);
2025 scratch_leaf = kmalloc(sizeof(*scratch_leaf), GFP_NOFS);
2031 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2032 if (IS_ERR(trans)) {
2033 err = PTR_ERR(trans);
2036 if (!fs_info->quota_enabled) {
2039 err = qgroup_rescan_leaf(qscan, path, trans,
2043 btrfs_commit_transaction(trans, fs_info->fs_root);
2045 btrfs_end_transaction(trans, fs_info->fs_root);
2049 kfree(scratch_leaf);
2051 btrfs_free_path(path);
2054 mutex_lock(&fs_info->qgroup_rescan_lock);
2055 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2058 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2059 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2060 } else if (err < 0) {
2061 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2063 mutex_unlock(&fs_info->qgroup_rescan_lock);
2066 pr_info("btrfs: qgroup scan completed%s\n",
2067 err == 2 ? " (inconsistency flag cleared)" : "");
2069 pr_err("btrfs: qgroup scan failed with %d\n", err);
2074 qgroup_rescan_start(struct btrfs_fs_info *fs_info, struct qgroup_rescan *qscan)
2076 memset(&qscan->work, 0, sizeof(qscan->work));
2077 qscan->work.func = btrfs_qgroup_rescan_worker;
2078 qscan->fs_info = fs_info;
2080 pr_info("btrfs: qgroup scan started\n");
2081 btrfs_queue_worker(&fs_info->qgroup_rescan_workers, &qscan->work);
2085 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2089 struct btrfs_qgroup *qgroup;
2090 struct qgroup_rescan *qscan = kmalloc(sizeof(*qscan), GFP_NOFS);
2095 mutex_lock(&fs_info->qgroup_rescan_lock);
2096 spin_lock(&fs_info->qgroup_lock);
2097 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2099 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2102 spin_unlock(&fs_info->qgroup_lock);
2103 mutex_unlock(&fs_info->qgroup_rescan_lock);
2108 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2109 memset(&fs_info->qgroup_rescan_progress, 0,
2110 sizeof(fs_info->qgroup_rescan_progress));
2112 /* clear all current qgroup tracking information */
2113 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2114 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2116 qgroup->rfer_cmpr = 0;
2118 qgroup->excl_cmpr = 0;
2120 spin_unlock(&fs_info->qgroup_lock);
2121 mutex_unlock(&fs_info->qgroup_rescan_lock);
2123 qgroup_rescan_start(fs_info, qscan);