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"
39 * - subvol delete -> delete when ref goes to 0? delete limits also?
43 * - copy also limits on subvol creation
45 * - caches fuer ulists
46 * - performance benchmarks
47 * - check all ioctl parameters
51 * one struct for each qgroup, organized in fs_info->qgroup_tree.
59 u64 rfer; /* referenced */
60 u64 rfer_cmpr; /* referenced compressed */
61 u64 excl; /* exclusive */
62 u64 excl_cmpr; /* exclusive compressed */
67 u64 lim_flags; /* which limits are set */
74 * reservation tracking
81 struct list_head groups; /* groups this group is member of */
82 struct list_head members; /* groups that are members of this group */
83 struct list_head dirty; /* dirty groups */
84 struct rb_node node; /* tree of qgroups */
87 * temp variables for accounting operations
88 * Refer to qgroup_shared_accounting() for details.
94 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
97 if (qg->old_refcnt < seq)
99 qg->old_refcnt += mod;
102 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
105 if (qg->new_refcnt < seq)
106 qg->new_refcnt = seq;
107 qg->new_refcnt += mod;
110 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
112 if (qg->old_refcnt < seq)
114 return qg->old_refcnt - seq;
117 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
119 if (qg->new_refcnt < seq)
121 return qg->new_refcnt - seq;
125 * glue structure to represent the relations between qgroups.
127 struct btrfs_qgroup_list {
128 struct list_head next_group;
129 struct list_head next_member;
130 struct btrfs_qgroup *group;
131 struct btrfs_qgroup *member;
134 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
136 return (u64)(uintptr_t)qg;
139 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
141 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
145 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
147 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
149 /* must be called with qgroup_ioctl_lock held */
150 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
153 struct rb_node *n = fs_info->qgroup_tree.rb_node;
154 struct btrfs_qgroup *qgroup;
157 qgroup = rb_entry(n, struct btrfs_qgroup, node);
158 if (qgroup->qgroupid < qgroupid)
160 else if (qgroup->qgroupid > qgroupid)
168 /* must be called with qgroup_lock held */
169 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
172 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
173 struct rb_node *parent = NULL;
174 struct btrfs_qgroup *qgroup;
178 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
180 if (qgroup->qgroupid < qgroupid)
182 else if (qgroup->qgroupid > qgroupid)
188 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
190 return ERR_PTR(-ENOMEM);
192 qgroup->qgroupid = qgroupid;
193 INIT_LIST_HEAD(&qgroup->groups);
194 INIT_LIST_HEAD(&qgroup->members);
195 INIT_LIST_HEAD(&qgroup->dirty);
197 rb_link_node(&qgroup->node, parent, p);
198 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
203 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
205 struct btrfs_qgroup_list *list;
207 list_del(&qgroup->dirty);
208 while (!list_empty(&qgroup->groups)) {
209 list = list_first_entry(&qgroup->groups,
210 struct btrfs_qgroup_list, next_group);
211 list_del(&list->next_group);
212 list_del(&list->next_member);
216 while (!list_empty(&qgroup->members)) {
217 list = list_first_entry(&qgroup->members,
218 struct btrfs_qgroup_list, next_member);
219 list_del(&list->next_group);
220 list_del(&list->next_member);
226 /* must be called with qgroup_lock held */
227 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
229 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
234 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
235 __del_qgroup_rb(qgroup);
239 /* must be called with qgroup_lock held */
240 static int add_relation_rb(struct btrfs_fs_info *fs_info,
241 u64 memberid, u64 parentid)
243 struct btrfs_qgroup *member;
244 struct btrfs_qgroup *parent;
245 struct btrfs_qgroup_list *list;
247 member = find_qgroup_rb(fs_info, memberid);
248 parent = find_qgroup_rb(fs_info, parentid);
249 if (!member || !parent)
252 list = kzalloc(sizeof(*list), GFP_ATOMIC);
256 list->group = parent;
257 list->member = member;
258 list_add_tail(&list->next_group, &member->groups);
259 list_add_tail(&list->next_member, &parent->members);
264 /* must be called with qgroup_lock held */
265 static int del_relation_rb(struct btrfs_fs_info *fs_info,
266 u64 memberid, u64 parentid)
268 struct btrfs_qgroup *member;
269 struct btrfs_qgroup *parent;
270 struct btrfs_qgroup_list *list;
272 member = find_qgroup_rb(fs_info, memberid);
273 parent = find_qgroup_rb(fs_info, parentid);
274 if (!member || !parent)
277 list_for_each_entry(list, &member->groups, next_group) {
278 if (list->group == parent) {
279 list_del(&list->next_group);
280 list_del(&list->next_member);
288 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
289 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
292 struct btrfs_qgroup *qgroup;
294 qgroup = find_qgroup_rb(fs_info, qgroupid);
297 if (qgroup->rfer != rfer || qgroup->excl != excl)
304 * The full config is read in one go, only called from open_ctree()
305 * It doesn't use any locking, as at this point we're still single-threaded
307 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
309 struct btrfs_key key;
310 struct btrfs_key found_key;
311 struct btrfs_root *quota_root = fs_info->quota_root;
312 struct btrfs_path *path = NULL;
313 struct extent_buffer *l;
317 u64 rescan_progress = 0;
319 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
322 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
323 if (!fs_info->qgroup_ulist) {
328 path = btrfs_alloc_path();
334 /* default this to quota off, in case no status key is found */
335 fs_info->qgroup_flags = 0;
338 * pass 1: read status, all qgroup infos and limits
343 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
348 struct btrfs_qgroup *qgroup;
350 slot = path->slots[0];
352 btrfs_item_key_to_cpu(l, &found_key, slot);
354 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
355 struct btrfs_qgroup_status_item *ptr;
357 ptr = btrfs_item_ptr(l, slot,
358 struct btrfs_qgroup_status_item);
360 if (btrfs_qgroup_status_version(l, ptr) !=
361 BTRFS_QGROUP_STATUS_VERSION) {
363 "old qgroup version, quota disabled");
366 if (btrfs_qgroup_status_generation(l, ptr) !=
367 fs_info->generation) {
368 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
370 "qgroup generation mismatch, marked as inconsistent");
372 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
374 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
378 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
379 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
382 qgroup = find_qgroup_rb(fs_info, found_key.offset);
383 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
384 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
385 btrfs_err(fs_info, "inconsistent qgroup config");
386 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
389 qgroup = add_qgroup_rb(fs_info, found_key.offset);
390 if (IS_ERR(qgroup)) {
391 ret = PTR_ERR(qgroup);
395 switch (found_key.type) {
396 case BTRFS_QGROUP_INFO_KEY: {
397 struct btrfs_qgroup_info_item *ptr;
399 ptr = btrfs_item_ptr(l, slot,
400 struct btrfs_qgroup_info_item);
401 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
402 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
403 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
404 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
405 /* generation currently unused */
408 case BTRFS_QGROUP_LIMIT_KEY: {
409 struct btrfs_qgroup_limit_item *ptr;
411 ptr = btrfs_item_ptr(l, slot,
412 struct btrfs_qgroup_limit_item);
413 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
414 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
415 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
416 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
417 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
422 ret = btrfs_next_item(quota_root, path);
428 btrfs_release_path(path);
431 * pass 2: read all qgroup relations
434 key.type = BTRFS_QGROUP_RELATION_KEY;
436 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
440 slot = path->slots[0];
442 btrfs_item_key_to_cpu(l, &found_key, slot);
444 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
447 if (found_key.objectid > found_key.offset) {
448 /* parent <- member, not needed to build config */
449 /* FIXME should we omit the key completely? */
453 ret = add_relation_rb(fs_info, found_key.objectid,
455 if (ret == -ENOENT) {
457 "orphan qgroup relation 0x%llx->0x%llx",
458 found_key.objectid, found_key.offset);
459 ret = 0; /* ignore the error */
464 ret = btrfs_next_item(quota_root, path);
471 fs_info->qgroup_flags |= flags;
472 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
473 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
474 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
476 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
477 btrfs_free_path(path);
480 ulist_free(fs_info->qgroup_ulist);
481 fs_info->qgroup_ulist = NULL;
482 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
485 return ret < 0 ? ret : 0;
489 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
490 * first two are in single-threaded paths.And for the third one, we have set
491 * quota_root to be null with qgroup_lock held before, so it is safe to clean
492 * up the in-memory structures without qgroup_lock held.
494 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
497 struct btrfs_qgroup *qgroup;
499 while ((n = rb_first(&fs_info->qgroup_tree))) {
500 qgroup = rb_entry(n, struct btrfs_qgroup, node);
501 rb_erase(n, &fs_info->qgroup_tree);
502 __del_qgroup_rb(qgroup);
505 * we call btrfs_free_qgroup_config() when umounting
506 * filesystem and disabling quota, so we set qgroup_ulist
507 * to be null here to avoid double free.
509 ulist_free(fs_info->qgroup_ulist);
510 fs_info->qgroup_ulist = NULL;
513 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
514 struct btrfs_root *quota_root,
518 struct btrfs_path *path;
519 struct btrfs_key key;
521 path = btrfs_alloc_path();
526 key.type = BTRFS_QGROUP_RELATION_KEY;
529 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
531 btrfs_mark_buffer_dirty(path->nodes[0]);
533 btrfs_free_path(path);
537 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
538 struct btrfs_root *quota_root,
542 struct btrfs_path *path;
543 struct btrfs_key key;
545 path = btrfs_alloc_path();
550 key.type = BTRFS_QGROUP_RELATION_KEY;
553 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
562 ret = btrfs_del_item(trans, quota_root, path);
564 btrfs_free_path(path);
568 static int add_qgroup_item(struct btrfs_trans_handle *trans,
569 struct btrfs_root *quota_root, u64 qgroupid)
572 struct btrfs_path *path;
573 struct btrfs_qgroup_info_item *qgroup_info;
574 struct btrfs_qgroup_limit_item *qgroup_limit;
575 struct extent_buffer *leaf;
576 struct btrfs_key key;
578 if (btrfs_is_testing(quota_root->fs_info))
581 path = btrfs_alloc_path();
586 key.type = BTRFS_QGROUP_INFO_KEY;
587 key.offset = qgroupid;
590 * Avoid a transaction abort by catching -EEXIST here. In that
591 * case, we proceed by re-initializing the existing structure
595 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
596 sizeof(*qgroup_info));
597 if (ret && ret != -EEXIST)
600 leaf = path->nodes[0];
601 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
602 struct btrfs_qgroup_info_item);
603 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
604 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
605 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
606 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
607 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
609 btrfs_mark_buffer_dirty(leaf);
611 btrfs_release_path(path);
613 key.type = BTRFS_QGROUP_LIMIT_KEY;
614 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
615 sizeof(*qgroup_limit));
616 if (ret && ret != -EEXIST)
619 leaf = path->nodes[0];
620 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
621 struct btrfs_qgroup_limit_item);
622 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
623 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
624 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
625 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
626 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
628 btrfs_mark_buffer_dirty(leaf);
632 btrfs_free_path(path);
636 static int del_qgroup_item(struct btrfs_trans_handle *trans,
637 struct btrfs_root *quota_root, u64 qgroupid)
640 struct btrfs_path *path;
641 struct btrfs_key key;
643 path = btrfs_alloc_path();
648 key.type = BTRFS_QGROUP_INFO_KEY;
649 key.offset = qgroupid;
650 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
659 ret = btrfs_del_item(trans, quota_root, path);
663 btrfs_release_path(path);
665 key.type = BTRFS_QGROUP_LIMIT_KEY;
666 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
675 ret = btrfs_del_item(trans, quota_root, path);
678 btrfs_free_path(path);
682 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
683 struct btrfs_root *root,
684 struct btrfs_qgroup *qgroup)
686 struct btrfs_path *path;
687 struct btrfs_key key;
688 struct extent_buffer *l;
689 struct btrfs_qgroup_limit_item *qgroup_limit;
694 key.type = BTRFS_QGROUP_LIMIT_KEY;
695 key.offset = qgroup->qgroupid;
697 path = btrfs_alloc_path();
701 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
709 slot = path->slots[0];
710 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
711 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
712 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
713 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
714 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
715 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
717 btrfs_mark_buffer_dirty(l);
720 btrfs_free_path(path);
724 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
725 struct btrfs_root *root,
726 struct btrfs_qgroup *qgroup)
728 struct btrfs_path *path;
729 struct btrfs_key key;
730 struct extent_buffer *l;
731 struct btrfs_qgroup_info_item *qgroup_info;
735 if (btrfs_is_testing(root->fs_info))
739 key.type = BTRFS_QGROUP_INFO_KEY;
740 key.offset = qgroup->qgroupid;
742 path = btrfs_alloc_path();
746 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
754 slot = path->slots[0];
755 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
756 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
757 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
758 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
759 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
760 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
762 btrfs_mark_buffer_dirty(l);
765 btrfs_free_path(path);
769 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
770 struct btrfs_fs_info *fs_info,
771 struct btrfs_root *root)
773 struct btrfs_path *path;
774 struct btrfs_key key;
775 struct extent_buffer *l;
776 struct btrfs_qgroup_status_item *ptr;
781 key.type = BTRFS_QGROUP_STATUS_KEY;
784 path = btrfs_alloc_path();
788 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
796 slot = path->slots[0];
797 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
798 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
799 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
800 btrfs_set_qgroup_status_rescan(l, ptr,
801 fs_info->qgroup_rescan_progress.objectid);
803 btrfs_mark_buffer_dirty(l);
806 btrfs_free_path(path);
811 * called with qgroup_lock held
813 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
814 struct btrfs_root *root)
816 struct btrfs_path *path;
817 struct btrfs_key key;
818 struct extent_buffer *leaf = NULL;
822 path = btrfs_alloc_path();
826 path->leave_spinning = 1;
833 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
836 leaf = path->nodes[0];
837 nr = btrfs_header_nritems(leaf);
841 * delete the leaf one by one
842 * since the whole tree is going
846 ret = btrfs_del_items(trans, root, path, 0, nr);
850 btrfs_release_path(path);
854 set_bit(BTRFS_FS_QUOTA_DISABLING, &root->fs_info->flags);
855 btrfs_free_path(path);
859 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
860 struct btrfs_fs_info *fs_info)
862 struct btrfs_root *quota_root;
863 struct btrfs_root *tree_root = fs_info->tree_root;
864 struct btrfs_path *path = NULL;
865 struct btrfs_qgroup_status_item *ptr;
866 struct extent_buffer *leaf;
867 struct btrfs_key key;
868 struct btrfs_key found_key;
869 struct btrfs_qgroup *qgroup = NULL;
873 mutex_lock(&fs_info->qgroup_ioctl_lock);
874 if (fs_info->quota_root) {
875 set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
879 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
880 if (!fs_info->qgroup_ulist) {
886 * initially create the quota tree
888 quota_root = btrfs_create_tree(trans, fs_info,
889 BTRFS_QUOTA_TREE_OBJECTID);
890 if (IS_ERR(quota_root)) {
891 ret = PTR_ERR(quota_root);
895 path = btrfs_alloc_path();
902 key.type = BTRFS_QGROUP_STATUS_KEY;
905 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
910 leaf = path->nodes[0];
911 ptr = btrfs_item_ptr(leaf, path->slots[0],
912 struct btrfs_qgroup_status_item);
913 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
914 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
915 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
916 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
917 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
918 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
920 btrfs_mark_buffer_dirty(leaf);
923 key.type = BTRFS_ROOT_REF_KEY;
926 btrfs_release_path(path);
927 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
935 slot = path->slots[0];
936 leaf = path->nodes[0];
937 btrfs_item_key_to_cpu(leaf, &found_key, slot);
939 if (found_key.type == BTRFS_ROOT_REF_KEY) {
940 ret = add_qgroup_item(trans, quota_root,
945 qgroup = add_qgroup_rb(fs_info, found_key.offset);
946 if (IS_ERR(qgroup)) {
947 ret = PTR_ERR(qgroup);
951 ret = btrfs_next_item(tree_root, path);
959 btrfs_release_path(path);
960 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
964 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
965 if (IS_ERR(qgroup)) {
966 ret = PTR_ERR(qgroup);
969 spin_lock(&fs_info->qgroup_lock);
970 fs_info->quota_root = quota_root;
971 set_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags);
972 spin_unlock(&fs_info->qgroup_lock);
974 btrfs_free_path(path);
977 free_extent_buffer(quota_root->node);
978 free_extent_buffer(quota_root->commit_root);
983 ulist_free(fs_info->qgroup_ulist);
984 fs_info->qgroup_ulist = NULL;
986 mutex_unlock(&fs_info->qgroup_ioctl_lock);
990 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
991 struct btrfs_fs_info *fs_info)
993 struct btrfs_root *tree_root = fs_info->tree_root;
994 struct btrfs_root *quota_root;
997 mutex_lock(&fs_info->qgroup_ioctl_lock);
998 if (!fs_info->quota_root)
1000 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1001 set_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags);
1002 btrfs_qgroup_wait_for_completion(fs_info, false);
1003 spin_lock(&fs_info->qgroup_lock);
1004 quota_root = fs_info->quota_root;
1005 fs_info->quota_root = NULL;
1006 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1007 spin_unlock(&fs_info->qgroup_lock);
1009 btrfs_free_qgroup_config(fs_info);
1011 ret = btrfs_clean_quota_tree(trans, quota_root);
1015 ret = btrfs_del_root(trans, tree_root, "a_root->root_key);
1019 list_del("a_root->dirty_list);
1021 btrfs_tree_lock(quota_root->node);
1022 clean_tree_block(fs_info, quota_root->node);
1023 btrfs_tree_unlock(quota_root->node);
1024 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1026 free_extent_buffer(quota_root->node);
1027 free_extent_buffer(quota_root->commit_root);
1030 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1034 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1035 struct btrfs_qgroup *qgroup)
1037 if (list_empty(&qgroup->dirty))
1038 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1041 static void report_reserved_underflow(struct btrfs_fs_info *fs_info,
1042 struct btrfs_qgroup *qgroup,
1045 #ifdef CONFIG_BTRFS_DEBUG
1046 WARN_ON(qgroup->reserved < num_bytes);
1047 btrfs_debug(fs_info,
1048 "qgroup %llu reserved space underflow, have: %llu, to free: %llu",
1049 qgroup->qgroupid, qgroup->reserved, num_bytes);
1051 qgroup->reserved = 0;
1054 * The easy accounting, if we are adding/removing the only ref for an extent
1055 * then this qgroup and all of the parent qgroups get their reference and
1056 * exclusive counts adjusted.
1058 * Caller should hold fs_info->qgroup_lock.
1060 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1061 struct ulist *tmp, u64 ref_root,
1062 u64 num_bytes, int sign)
1064 struct btrfs_qgroup *qgroup;
1065 struct btrfs_qgroup_list *glist;
1066 struct ulist_node *unode;
1067 struct ulist_iterator uiter;
1070 qgroup = find_qgroup_rb(fs_info, ref_root);
1074 qgroup->rfer += sign * num_bytes;
1075 qgroup->rfer_cmpr += sign * num_bytes;
1077 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1078 qgroup->excl += sign * num_bytes;
1079 qgroup->excl_cmpr += sign * num_bytes;
1081 if (qgroup->reserved < num_bytes)
1082 report_reserved_underflow(fs_info, qgroup, num_bytes);
1084 qgroup->reserved -= num_bytes;
1087 qgroup_dirty(fs_info, qgroup);
1089 /* Get all of the parent groups that contain this qgroup */
1090 list_for_each_entry(glist, &qgroup->groups, next_group) {
1091 ret = ulist_add(tmp, glist->group->qgroupid,
1092 qgroup_to_aux(glist->group), GFP_ATOMIC);
1097 /* Iterate all of the parents and adjust their reference counts */
1098 ULIST_ITER_INIT(&uiter);
1099 while ((unode = ulist_next(tmp, &uiter))) {
1100 qgroup = unode_aux_to_qgroup(unode);
1101 qgroup->rfer += sign * num_bytes;
1102 qgroup->rfer_cmpr += sign * num_bytes;
1103 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1104 qgroup->excl += sign * num_bytes;
1106 if (qgroup->reserved < num_bytes)
1107 report_reserved_underflow(fs_info, qgroup,
1110 qgroup->reserved -= num_bytes;
1112 qgroup->excl_cmpr += sign * num_bytes;
1113 qgroup_dirty(fs_info, qgroup);
1115 /* Add any parents of the parents */
1116 list_for_each_entry(glist, &qgroup->groups, next_group) {
1117 ret = ulist_add(tmp, glist->group->qgroupid,
1118 qgroup_to_aux(glist->group), GFP_ATOMIC);
1130 * Quick path for updating qgroup with only excl refs.
1132 * In that case, just update all parent will be enough.
1133 * Or we needs to do a full rescan.
1134 * Caller should also hold fs_info->qgroup_lock.
1136 * Return 0 for quick update, return >0 for need to full rescan
1137 * and mark INCONSISTENT flag.
1138 * Return < 0 for other error.
1140 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1141 struct ulist *tmp, u64 src, u64 dst,
1144 struct btrfs_qgroup *qgroup;
1148 qgroup = find_qgroup_rb(fs_info, src);
1151 if (qgroup->excl == qgroup->rfer) {
1153 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1154 qgroup->excl, sign);
1162 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1166 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1167 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1169 struct btrfs_root *quota_root;
1170 struct btrfs_qgroup *parent;
1171 struct btrfs_qgroup *member;
1172 struct btrfs_qgroup_list *list;
1176 /* Check the level of src and dst first */
1177 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1180 tmp = ulist_alloc(GFP_KERNEL);
1184 mutex_lock(&fs_info->qgroup_ioctl_lock);
1185 quota_root = fs_info->quota_root;
1190 member = find_qgroup_rb(fs_info, src);
1191 parent = find_qgroup_rb(fs_info, dst);
1192 if (!member || !parent) {
1197 /* check if such qgroup relation exist firstly */
1198 list_for_each_entry(list, &member->groups, next_group) {
1199 if (list->group == parent) {
1205 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1209 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1211 del_qgroup_relation_item(trans, quota_root, src, dst);
1215 spin_lock(&fs_info->qgroup_lock);
1216 ret = add_relation_rb(fs_info, src, dst);
1218 spin_unlock(&fs_info->qgroup_lock);
1221 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1222 spin_unlock(&fs_info->qgroup_lock);
1224 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1229 static int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1230 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1232 struct btrfs_root *quota_root;
1233 struct btrfs_qgroup *parent;
1234 struct btrfs_qgroup *member;
1235 struct btrfs_qgroup_list *list;
1240 tmp = ulist_alloc(GFP_KERNEL);
1244 quota_root = fs_info->quota_root;
1250 member = find_qgroup_rb(fs_info, src);
1251 parent = find_qgroup_rb(fs_info, dst);
1252 if (!member || !parent) {
1257 /* check if such qgroup relation exist firstly */
1258 list_for_each_entry(list, &member->groups, next_group) {
1259 if (list->group == parent)
1265 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1266 err = del_qgroup_relation_item(trans, quota_root, dst, src);
1270 spin_lock(&fs_info->qgroup_lock);
1271 del_relation_rb(fs_info, src, dst);
1272 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1273 spin_unlock(&fs_info->qgroup_lock);
1279 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1280 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1284 mutex_lock(&fs_info->qgroup_ioctl_lock);
1285 ret = __del_qgroup_relation(trans, fs_info, src, dst);
1286 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1291 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1292 struct btrfs_fs_info *fs_info, u64 qgroupid)
1294 struct btrfs_root *quota_root;
1295 struct btrfs_qgroup *qgroup;
1298 mutex_lock(&fs_info->qgroup_ioctl_lock);
1299 quota_root = fs_info->quota_root;
1304 qgroup = find_qgroup_rb(fs_info, qgroupid);
1310 ret = add_qgroup_item(trans, quota_root, qgroupid);
1314 spin_lock(&fs_info->qgroup_lock);
1315 qgroup = add_qgroup_rb(fs_info, qgroupid);
1316 spin_unlock(&fs_info->qgroup_lock);
1319 ret = PTR_ERR(qgroup);
1321 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1325 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1326 struct btrfs_fs_info *fs_info, u64 qgroupid)
1328 struct btrfs_root *quota_root;
1329 struct btrfs_qgroup *qgroup;
1330 struct btrfs_qgroup_list *list;
1333 mutex_lock(&fs_info->qgroup_ioctl_lock);
1334 quota_root = fs_info->quota_root;
1340 qgroup = find_qgroup_rb(fs_info, qgroupid);
1345 /* check if there are no children of this qgroup */
1346 if (!list_empty(&qgroup->members)) {
1351 ret = del_qgroup_item(trans, quota_root, qgroupid);
1353 while (!list_empty(&qgroup->groups)) {
1354 list = list_first_entry(&qgroup->groups,
1355 struct btrfs_qgroup_list, next_group);
1356 ret = __del_qgroup_relation(trans, fs_info,
1358 list->group->qgroupid);
1363 spin_lock(&fs_info->qgroup_lock);
1364 del_qgroup_rb(fs_info, qgroupid);
1365 spin_unlock(&fs_info->qgroup_lock);
1367 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1371 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1372 struct btrfs_fs_info *fs_info, u64 qgroupid,
1373 struct btrfs_qgroup_limit *limit)
1375 struct btrfs_root *quota_root;
1376 struct btrfs_qgroup *qgroup;
1378 /* Sometimes we would want to clear the limit on this qgroup.
1379 * To meet this requirement, we treat the -1 as a special value
1380 * which tell kernel to clear the limit on this qgroup.
1382 const u64 CLEAR_VALUE = -1;
1384 mutex_lock(&fs_info->qgroup_ioctl_lock);
1385 quota_root = fs_info->quota_root;
1391 qgroup = find_qgroup_rb(fs_info, qgroupid);
1397 spin_lock(&fs_info->qgroup_lock);
1398 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1399 if (limit->max_rfer == CLEAR_VALUE) {
1400 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1401 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1402 qgroup->max_rfer = 0;
1404 qgroup->max_rfer = limit->max_rfer;
1407 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1408 if (limit->max_excl == CLEAR_VALUE) {
1409 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1410 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1411 qgroup->max_excl = 0;
1413 qgroup->max_excl = limit->max_excl;
1416 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1417 if (limit->rsv_rfer == CLEAR_VALUE) {
1418 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1419 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1420 qgroup->rsv_rfer = 0;
1422 qgroup->rsv_rfer = limit->rsv_rfer;
1425 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1426 if (limit->rsv_excl == CLEAR_VALUE) {
1427 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1428 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1429 qgroup->rsv_excl = 0;
1431 qgroup->rsv_excl = limit->rsv_excl;
1434 qgroup->lim_flags |= limit->flags;
1436 spin_unlock(&fs_info->qgroup_lock);
1438 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1440 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1441 btrfs_info(fs_info, "unable to update quota limit for %llu",
1446 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1450 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
1451 struct btrfs_fs_info *fs_info)
1453 struct btrfs_qgroup_extent_record *record;
1454 struct btrfs_delayed_ref_root *delayed_refs;
1455 struct rb_node *node;
1459 delayed_refs = &trans->transaction->delayed_refs;
1460 qgroup_to_skip = delayed_refs->qgroup_to_skip;
1463 * No need to do lock, since this function will only be called in
1464 * btrfs_commit_transaction().
1466 node = rb_first(&delayed_refs->dirty_extent_root);
1468 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1470 if (WARN_ON(!record->old_roots))
1471 ret = btrfs_find_all_roots(NULL, fs_info,
1472 record->bytenr, 0, &record->old_roots);
1476 ulist_del(record->old_roots, qgroup_to_skip, 0);
1477 node = rb_next(node);
1482 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1483 struct btrfs_delayed_ref_root *delayed_refs,
1484 struct btrfs_qgroup_extent_record *record)
1486 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1487 struct rb_node *parent_node = NULL;
1488 struct btrfs_qgroup_extent_record *entry;
1489 u64 bytenr = record->bytenr;
1491 assert_spin_locked(&delayed_refs->lock);
1492 trace_btrfs_qgroup_trace_extent(fs_info, record);
1496 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1498 if (bytenr < entry->bytenr)
1500 else if (bytenr > entry->bytenr)
1501 p = &(*p)->rb_right;
1506 rb_link_node(&record->node, parent_node, p);
1507 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1511 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1512 struct btrfs_qgroup_extent_record *qrecord)
1514 struct ulist *old_root;
1515 u64 bytenr = qrecord->bytenr;
1518 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root);
1523 * Here we don't need to get the lock of
1524 * trans->transaction->delayed_refs, since inserted qrecord won't
1525 * be deleted, only qrecord->node may be modified (new qrecord insert)
1527 * So modifying qrecord->old_roots is safe here
1529 qrecord->old_roots = old_root;
1533 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1534 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1537 struct btrfs_qgroup_extent_record *record;
1538 struct btrfs_delayed_ref_root *delayed_refs;
1541 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1542 || bytenr == 0 || num_bytes == 0)
1544 if (WARN_ON(trans == NULL))
1546 record = kmalloc(sizeof(*record), gfp_flag);
1550 delayed_refs = &trans->transaction->delayed_refs;
1551 record->bytenr = bytenr;
1552 record->num_bytes = num_bytes;
1553 record->old_roots = NULL;
1555 spin_lock(&delayed_refs->lock);
1556 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1557 spin_unlock(&delayed_refs->lock);
1562 return btrfs_qgroup_trace_extent_post(fs_info, record);
1565 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1566 struct btrfs_fs_info *fs_info,
1567 struct extent_buffer *eb)
1569 int nr = btrfs_header_nritems(eb);
1570 int i, extent_type, ret;
1571 struct btrfs_key key;
1572 struct btrfs_file_extent_item *fi;
1573 u64 bytenr, num_bytes;
1575 /* We can be called directly from walk_up_proc() */
1576 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1579 for (i = 0; i < nr; i++) {
1580 btrfs_item_key_to_cpu(eb, &key, i);
1582 if (key.type != BTRFS_EXTENT_DATA_KEY)
1585 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1586 /* filter out non qgroup-accountable extents */
1587 extent_type = btrfs_file_extent_type(eb, fi);
1589 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1592 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1596 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1598 ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1599 num_bytes, GFP_NOFS);
1607 * Walk up the tree from the bottom, freeing leaves and any interior
1608 * nodes which have had all slots visited. If a node (leaf or
1609 * interior) is freed, the node above it will have it's slot
1610 * incremented. The root node will never be freed.
1612 * At the end of this function, we should have a path which has all
1613 * slots incremented to the next position for a search. If we need to
1614 * read a new node it will be NULL and the node above it will have the
1615 * correct slot selected for a later read.
1617 * If we increment the root nodes slot counter past the number of
1618 * elements, 1 is returned to signal completion of the search.
1620 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1624 struct extent_buffer *eb;
1626 if (root_level == 0)
1629 while (level <= root_level) {
1630 eb = path->nodes[level];
1631 nr = btrfs_header_nritems(eb);
1632 path->slots[level]++;
1633 slot = path->slots[level];
1634 if (slot >= nr || level == 0) {
1636 * Don't free the root - we will detect this
1637 * condition after our loop and return a
1638 * positive value for caller to stop walking the tree.
1640 if (level != root_level) {
1641 btrfs_tree_unlock_rw(eb, path->locks[level]);
1642 path->locks[level] = 0;
1644 free_extent_buffer(eb);
1645 path->nodes[level] = NULL;
1646 path->slots[level] = 0;
1650 * We have a valid slot to walk back down
1651 * from. Stop here so caller can process these
1660 eb = path->nodes[root_level];
1661 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1667 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1668 struct btrfs_root *root,
1669 struct extent_buffer *root_eb,
1670 u64 root_gen, int root_level)
1672 struct btrfs_fs_info *fs_info = root->fs_info;
1675 struct extent_buffer *eb = root_eb;
1676 struct btrfs_path *path = NULL;
1678 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
1679 BUG_ON(root_eb == NULL);
1681 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1684 if (!extent_buffer_uptodate(root_eb)) {
1685 ret = btrfs_read_buffer(root_eb, root_gen);
1690 if (root_level == 0) {
1691 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1695 path = btrfs_alloc_path();
1700 * Walk down the tree. Missing extent blocks are filled in as
1701 * we go. Metadata is accounted every time we read a new
1704 * When we reach a leaf, we account for file extent items in it,
1705 * walk back up the tree (adjusting slot pointers as we go)
1706 * and restart the search process.
1708 extent_buffer_get(root_eb); /* For path */
1709 path->nodes[root_level] = root_eb;
1710 path->slots[root_level] = 0;
1711 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1714 while (level >= 0) {
1715 if (path->nodes[level] == NULL) {
1721 * We need to get child blockptr/gen from parent before
1724 eb = path->nodes[level + 1];
1725 parent_slot = path->slots[level + 1];
1726 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1727 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1729 eb = read_tree_block(fs_info, child_bytenr, child_gen);
1733 } else if (!extent_buffer_uptodate(eb)) {
1734 free_extent_buffer(eb);
1739 path->nodes[level] = eb;
1740 path->slots[level] = 0;
1742 btrfs_tree_read_lock(eb);
1743 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1744 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1746 ret = btrfs_qgroup_trace_extent(trans, fs_info,
1755 ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1756 path->nodes[level]);
1760 /* Nonzero return here means we completed our search */
1761 ret = adjust_slots_upwards(path, root_level);
1765 /* Restart search with new slots */
1774 btrfs_free_path(path);
1779 #define UPDATE_NEW 0
1780 #define UPDATE_OLD 1
1782 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1784 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1785 struct ulist *roots, struct ulist *tmp,
1786 struct ulist *qgroups, u64 seq, int update_old)
1788 struct ulist_node *unode;
1789 struct ulist_iterator uiter;
1790 struct ulist_node *tmp_unode;
1791 struct ulist_iterator tmp_uiter;
1792 struct btrfs_qgroup *qg;
1797 ULIST_ITER_INIT(&uiter);
1798 while ((unode = ulist_next(roots, &uiter))) {
1799 qg = find_qgroup_rb(fs_info, unode->val);
1804 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1808 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1811 ULIST_ITER_INIT(&tmp_uiter);
1812 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1813 struct btrfs_qgroup_list *glist;
1815 qg = unode_aux_to_qgroup(tmp_unode);
1817 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1819 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1820 list_for_each_entry(glist, &qg->groups, next_group) {
1821 ret = ulist_add(qgroups, glist->group->qgroupid,
1822 qgroup_to_aux(glist->group),
1826 ret = ulist_add(tmp, glist->group->qgroupid,
1827 qgroup_to_aux(glist->group),
1838 * Update qgroup rfer/excl counters.
1839 * Rfer update is easy, codes can explain themselves.
1841 * Excl update is tricky, the update is split into 2 part.
1842 * Part 1: Possible exclusive <-> sharing detect:
1844 * -------------------------------------
1846 * -------------------------------------
1848 * -------------------------------------
1851 * A: cur_old_roots < nr_old_roots (not exclusive before)
1852 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1853 * B: cur_new_roots < nr_new_roots (not exclusive now)
1854 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
1857 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1858 * *: Definitely not changed. **: Possible unchanged.
1860 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1862 * To make the logic clear, we first use condition A and B to split
1863 * combination into 4 results.
1865 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1866 * only on variant maybe 0.
1868 * Lastly, check result **, since there are 2 variants maybe 0, split them
1870 * But this time we don't need to consider other things, the codes and logic
1871 * is easy to understand now.
1873 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1874 struct ulist *qgroups,
1877 u64 num_bytes, u64 seq)
1879 struct ulist_node *unode;
1880 struct ulist_iterator uiter;
1881 struct btrfs_qgroup *qg;
1882 u64 cur_new_count, cur_old_count;
1884 ULIST_ITER_INIT(&uiter);
1885 while ((unode = ulist_next(qgroups, &uiter))) {
1888 qg = unode_aux_to_qgroup(unode);
1889 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1890 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1892 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1893 cur_old_count, cur_new_count);
1895 /* Rfer update part */
1896 if (cur_old_count == 0 && cur_new_count > 0) {
1897 qg->rfer += num_bytes;
1898 qg->rfer_cmpr += num_bytes;
1901 if (cur_old_count > 0 && cur_new_count == 0) {
1902 qg->rfer -= num_bytes;
1903 qg->rfer_cmpr -= num_bytes;
1907 /* Excl update part */
1908 /* Exclusive/none -> shared case */
1909 if (cur_old_count == nr_old_roots &&
1910 cur_new_count < nr_new_roots) {
1911 /* Exclusive -> shared */
1912 if (cur_old_count != 0) {
1913 qg->excl -= num_bytes;
1914 qg->excl_cmpr -= num_bytes;
1919 /* Shared -> exclusive/none case */
1920 if (cur_old_count < nr_old_roots &&
1921 cur_new_count == nr_new_roots) {
1922 /* Shared->exclusive */
1923 if (cur_new_count != 0) {
1924 qg->excl += num_bytes;
1925 qg->excl_cmpr += num_bytes;
1930 /* Exclusive/none -> exclusive/none case */
1931 if (cur_old_count == nr_old_roots &&
1932 cur_new_count == nr_new_roots) {
1933 if (cur_old_count == 0) {
1934 /* None -> exclusive/none */
1936 if (cur_new_count != 0) {
1937 /* None -> exclusive */
1938 qg->excl += num_bytes;
1939 qg->excl_cmpr += num_bytes;
1942 /* None -> none, nothing changed */
1944 /* Exclusive -> exclusive/none */
1946 if (cur_new_count == 0) {
1947 /* Exclusive -> none */
1948 qg->excl -= num_bytes;
1949 qg->excl_cmpr -= num_bytes;
1952 /* Exclusive -> exclusive, nothing changed */
1957 qgroup_dirty(fs_info, qg);
1963 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1964 struct btrfs_fs_info *fs_info,
1965 u64 bytenr, u64 num_bytes,
1966 struct ulist *old_roots, struct ulist *new_roots)
1968 struct ulist *qgroups = NULL;
1969 struct ulist *tmp = NULL;
1971 u64 nr_new_roots = 0;
1972 u64 nr_old_roots = 0;
1975 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1979 nr_new_roots = new_roots->nnodes;
1981 nr_old_roots = old_roots->nnodes;
1983 BUG_ON(!fs_info->quota_root);
1985 trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1986 nr_old_roots, nr_new_roots);
1988 qgroups = ulist_alloc(GFP_NOFS);
1993 tmp = ulist_alloc(GFP_NOFS);
1999 mutex_lock(&fs_info->qgroup_rescan_lock);
2000 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2001 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2002 mutex_unlock(&fs_info->qgroup_rescan_lock);
2007 mutex_unlock(&fs_info->qgroup_rescan_lock);
2009 spin_lock(&fs_info->qgroup_lock);
2010 seq = fs_info->qgroup_seq;
2012 /* Update old refcnts using old_roots */
2013 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2018 /* Update new refcnts using new_roots */
2019 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2024 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2028 * Bump qgroup_seq to avoid seq overlap
2030 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2032 spin_unlock(&fs_info->qgroup_lock);
2035 ulist_free(qgroups);
2036 ulist_free(old_roots);
2037 ulist_free(new_roots);
2041 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
2042 struct btrfs_fs_info *fs_info)
2044 struct btrfs_qgroup_extent_record *record;
2045 struct btrfs_delayed_ref_root *delayed_refs;
2046 struct ulist *new_roots = NULL;
2047 struct rb_node *node;
2051 delayed_refs = &trans->transaction->delayed_refs;
2052 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2053 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2054 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2057 trace_btrfs_qgroup_account_extents(fs_info, record);
2061 * Use (u64)-1 as time_seq to do special search, which
2062 * doesn't lock tree or delayed_refs and search current
2063 * root. It's safe inside commit_transaction().
2065 ret = btrfs_find_all_roots(trans, fs_info,
2066 record->bytenr, (u64)-1, &new_roots);
2070 ulist_del(new_roots, qgroup_to_skip, 0);
2071 ret = btrfs_qgroup_account_extent(trans, fs_info,
2072 record->bytenr, record->num_bytes,
2073 record->old_roots, new_roots);
2074 record->old_roots = NULL;
2078 ulist_free(record->old_roots);
2079 ulist_free(new_roots);
2081 rb_erase(node, &delayed_refs->dirty_extent_root);
2089 * called from commit_transaction. Writes all changed qgroups to disk.
2091 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2092 struct btrfs_fs_info *fs_info)
2094 struct btrfs_root *quota_root = fs_info->quota_root;
2096 int start_rescan_worker = 0;
2101 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
2102 test_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2103 start_rescan_worker = 1;
2105 if (test_and_clear_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2106 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2107 if (test_and_clear_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags))
2108 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2110 spin_lock(&fs_info->qgroup_lock);
2111 while (!list_empty(&fs_info->dirty_qgroups)) {
2112 struct btrfs_qgroup *qgroup;
2113 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2114 struct btrfs_qgroup, dirty);
2115 list_del_init(&qgroup->dirty);
2116 spin_unlock(&fs_info->qgroup_lock);
2117 ret = update_qgroup_info_item(trans, quota_root, qgroup);
2119 fs_info->qgroup_flags |=
2120 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2121 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2123 fs_info->qgroup_flags |=
2124 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2125 spin_lock(&fs_info->qgroup_lock);
2127 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2128 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2130 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2131 spin_unlock(&fs_info->qgroup_lock);
2133 ret = update_qgroup_status_item(trans, fs_info, quota_root);
2135 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2137 if (!ret && start_rescan_worker) {
2138 ret = qgroup_rescan_init(fs_info, 0, 1);
2140 qgroup_rescan_zero_tracking(fs_info);
2141 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2142 &fs_info->qgroup_rescan_work);
2153 * Copy the accounting information between qgroups. This is necessary
2154 * when a snapshot or a subvolume is created. Throwing an error will
2155 * cause a transaction abort so we take extra care here to only error
2156 * when a readonly fs is a reasonable outcome.
2158 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2159 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2160 struct btrfs_qgroup_inherit *inherit)
2165 struct btrfs_root *quota_root = fs_info->quota_root;
2166 struct btrfs_qgroup *srcgroup;
2167 struct btrfs_qgroup *dstgroup;
2171 mutex_lock(&fs_info->qgroup_ioctl_lock);
2172 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2181 i_qgroups = (u64 *)(inherit + 1);
2182 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2183 2 * inherit->num_excl_copies;
2184 for (i = 0; i < nums; ++i) {
2185 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2188 * Zero out invalid groups so we can ignore
2192 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2200 * create a tracking group for the subvol itself
2202 ret = add_qgroup_item(trans, quota_root, objectid);
2207 struct btrfs_root *srcroot;
2208 struct btrfs_key srckey;
2210 srckey.objectid = srcid;
2211 srckey.type = BTRFS_ROOT_ITEM_KEY;
2212 srckey.offset = (u64)-1;
2213 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2214 if (IS_ERR(srcroot)) {
2215 ret = PTR_ERR(srcroot);
2219 level_size = fs_info->nodesize;
2223 * add qgroup to all inherited groups
2226 i_qgroups = (u64 *)(inherit + 1);
2227 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2228 if (*i_qgroups == 0)
2230 ret = add_qgroup_relation_item(trans, quota_root,
2231 objectid, *i_qgroups);
2232 if (ret && ret != -EEXIST)
2234 ret = add_qgroup_relation_item(trans, quota_root,
2235 *i_qgroups, objectid);
2236 if (ret && ret != -EEXIST)
2243 spin_lock(&fs_info->qgroup_lock);
2245 dstgroup = add_qgroup_rb(fs_info, objectid);
2246 if (IS_ERR(dstgroup)) {
2247 ret = PTR_ERR(dstgroup);
2251 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2252 dstgroup->lim_flags = inherit->lim.flags;
2253 dstgroup->max_rfer = inherit->lim.max_rfer;
2254 dstgroup->max_excl = inherit->lim.max_excl;
2255 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2256 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2258 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2260 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2262 "unable to update quota limit for %llu",
2263 dstgroup->qgroupid);
2269 srcgroup = find_qgroup_rb(fs_info, srcid);
2274 * We call inherit after we clone the root in order to make sure
2275 * our counts don't go crazy, so at this point the only
2276 * difference between the two roots should be the root node.
2278 dstgroup->rfer = srcgroup->rfer;
2279 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2280 dstgroup->excl = level_size;
2281 dstgroup->excl_cmpr = level_size;
2282 srcgroup->excl = level_size;
2283 srcgroup->excl_cmpr = level_size;
2285 /* inherit the limit info */
2286 dstgroup->lim_flags = srcgroup->lim_flags;
2287 dstgroup->max_rfer = srcgroup->max_rfer;
2288 dstgroup->max_excl = srcgroup->max_excl;
2289 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2290 dstgroup->rsv_excl = srcgroup->rsv_excl;
2292 qgroup_dirty(fs_info, dstgroup);
2293 qgroup_dirty(fs_info, srcgroup);
2299 i_qgroups = (u64 *)(inherit + 1);
2300 for (i = 0; i < inherit->num_qgroups; ++i) {
2302 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2309 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2310 struct btrfs_qgroup *src;
2311 struct btrfs_qgroup *dst;
2313 if (!i_qgroups[0] || !i_qgroups[1])
2316 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2317 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2324 dst->rfer = src->rfer - level_size;
2325 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2327 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2328 struct btrfs_qgroup *src;
2329 struct btrfs_qgroup *dst;
2331 if (!i_qgroups[0] || !i_qgroups[1])
2334 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2335 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2342 dst->excl = src->excl + level_size;
2343 dst->excl_cmpr = src->excl_cmpr + level_size;
2347 spin_unlock(&fs_info->qgroup_lock);
2349 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2353 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2355 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2356 qg->reserved + (s64)qg->rfer + num_bytes > qg->max_rfer)
2359 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2360 qg->reserved + (s64)qg->excl + num_bytes > qg->max_excl)
2366 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce)
2368 struct btrfs_root *quota_root;
2369 struct btrfs_qgroup *qgroup;
2370 struct btrfs_fs_info *fs_info = root->fs_info;
2371 u64 ref_root = root->root_key.objectid;
2373 struct ulist_node *unode;
2374 struct ulist_iterator uiter;
2376 if (!is_fstree(ref_root))
2382 spin_lock(&fs_info->qgroup_lock);
2383 quota_root = fs_info->quota_root;
2387 qgroup = find_qgroup_rb(fs_info, ref_root);
2392 * in a first step, we check all affected qgroups if any limits would
2395 ulist_reinit(fs_info->qgroup_ulist);
2396 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2397 (uintptr_t)qgroup, GFP_ATOMIC);
2400 ULIST_ITER_INIT(&uiter);
2401 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2402 struct btrfs_qgroup *qg;
2403 struct btrfs_qgroup_list *glist;
2405 qg = unode_aux_to_qgroup(unode);
2407 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
2412 list_for_each_entry(glist, &qg->groups, next_group) {
2413 ret = ulist_add(fs_info->qgroup_ulist,
2414 glist->group->qgroupid,
2415 (uintptr_t)glist->group, GFP_ATOMIC);
2422 * no limits exceeded, now record the reservation into all qgroups
2424 ULIST_ITER_INIT(&uiter);
2425 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2426 struct btrfs_qgroup *qg;
2428 qg = unode_aux_to_qgroup(unode);
2430 qg->reserved += num_bytes;
2434 spin_unlock(&fs_info->qgroup_lock);
2438 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2439 u64 ref_root, u64 num_bytes)
2441 struct btrfs_root *quota_root;
2442 struct btrfs_qgroup *qgroup;
2443 struct ulist_node *unode;
2444 struct ulist_iterator uiter;
2447 if (!is_fstree(ref_root))
2453 spin_lock(&fs_info->qgroup_lock);
2455 quota_root = fs_info->quota_root;
2459 qgroup = find_qgroup_rb(fs_info, ref_root);
2463 ulist_reinit(fs_info->qgroup_ulist);
2464 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2465 (uintptr_t)qgroup, GFP_ATOMIC);
2468 ULIST_ITER_INIT(&uiter);
2469 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2470 struct btrfs_qgroup *qg;
2471 struct btrfs_qgroup_list *glist;
2473 qg = unode_aux_to_qgroup(unode);
2475 if (qg->reserved < num_bytes)
2476 report_reserved_underflow(fs_info, qg, num_bytes);
2478 qg->reserved -= num_bytes;
2480 list_for_each_entry(glist, &qg->groups, next_group) {
2481 ret = ulist_add(fs_info->qgroup_ulist,
2482 glist->group->qgroupid,
2483 (uintptr_t)glist->group, GFP_ATOMIC);
2490 spin_unlock(&fs_info->qgroup_lock);
2493 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2495 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2497 btrfs_err(trans->fs_info,
2498 "qgroups not uptodate in trans handle %p: list is%s empty, seq is %#x.%x",
2499 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
2500 (u32)(trans->delayed_ref_elem.seq >> 32),
2501 (u32)trans->delayed_ref_elem.seq);
2506 * returns < 0 on error, 0 when more leafs are to be scanned.
2507 * returns 1 when done.
2510 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2511 struct btrfs_trans_handle *trans)
2513 struct btrfs_key found;
2514 struct extent_buffer *scratch_leaf = NULL;
2515 struct ulist *roots = NULL;
2516 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2521 mutex_lock(&fs_info->qgroup_rescan_lock);
2522 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2523 &fs_info->qgroup_rescan_progress,
2526 btrfs_debug(fs_info,
2527 "current progress key (%llu %u %llu), search_slot ret %d",
2528 fs_info->qgroup_rescan_progress.objectid,
2529 fs_info->qgroup_rescan_progress.type,
2530 fs_info->qgroup_rescan_progress.offset, ret);
2534 * The rescan is about to end, we will not be scanning any
2535 * further blocks. We cannot unset the RESCAN flag here, because
2536 * we want to commit the transaction if everything went well.
2537 * To make the live accounting work in this phase, we set our
2538 * scan progress pointer such that every real extent objectid
2541 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2542 btrfs_release_path(path);
2543 mutex_unlock(&fs_info->qgroup_rescan_lock);
2547 btrfs_item_key_to_cpu(path->nodes[0], &found,
2548 btrfs_header_nritems(path->nodes[0]) - 1);
2549 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2551 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2552 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2553 if (!scratch_leaf) {
2555 mutex_unlock(&fs_info->qgroup_rescan_lock);
2558 extent_buffer_get(scratch_leaf);
2559 btrfs_tree_read_lock(scratch_leaf);
2560 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2561 slot = path->slots[0];
2562 btrfs_release_path(path);
2563 mutex_unlock(&fs_info->qgroup_rescan_lock);
2565 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2566 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2567 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2568 found.type != BTRFS_METADATA_ITEM_KEY)
2570 if (found.type == BTRFS_METADATA_ITEM_KEY)
2571 num_bytes = fs_info->nodesize;
2573 num_bytes = found.offset;
2575 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2579 /* For rescan, just pass old_roots as NULL */
2580 ret = btrfs_qgroup_account_extent(trans, fs_info,
2581 found.objectid, num_bytes, NULL, roots);
2587 btrfs_tree_read_unlock_blocking(scratch_leaf);
2588 free_extent_buffer(scratch_leaf);
2590 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2595 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2597 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2598 qgroup_rescan_work);
2599 struct btrfs_path *path;
2600 struct btrfs_trans_handle *trans = NULL;
2604 path = btrfs_alloc_path();
2609 while (!err && !btrfs_fs_closing(fs_info)) {
2610 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2611 if (IS_ERR(trans)) {
2612 err = PTR_ERR(trans);
2615 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2618 err = qgroup_rescan_leaf(fs_info, path, trans);
2621 btrfs_commit_transaction(trans);
2623 btrfs_end_transaction(trans);
2627 btrfs_free_path(path);
2629 mutex_lock(&fs_info->qgroup_rescan_lock);
2630 if (!btrfs_fs_closing(fs_info))
2631 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2634 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2635 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2636 } else if (err < 0) {
2637 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2639 mutex_unlock(&fs_info->qgroup_rescan_lock);
2642 * only update status, since the previous part has already updated the
2645 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2646 if (IS_ERR(trans)) {
2647 err = PTR_ERR(trans);
2649 "fail to start transaction for status update: %d\n",
2653 ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2656 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2658 btrfs_end_transaction(trans);
2660 if (btrfs_fs_closing(fs_info)) {
2661 btrfs_info(fs_info, "qgroup scan paused");
2662 } else if (err >= 0) {
2663 btrfs_info(fs_info, "qgroup scan completed%s",
2664 err > 0 ? " (inconsistency flag cleared)" : "");
2666 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2670 mutex_lock(&fs_info->qgroup_rescan_lock);
2671 fs_info->qgroup_rescan_running = false;
2672 mutex_unlock(&fs_info->qgroup_rescan_lock);
2673 complete_all(&fs_info->qgroup_rescan_completion);
2677 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2678 * memory required for the rescan context.
2681 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2687 (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2688 !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2693 mutex_lock(&fs_info->qgroup_rescan_lock);
2694 spin_lock(&fs_info->qgroup_lock);
2697 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2699 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2703 spin_unlock(&fs_info->qgroup_lock);
2704 mutex_unlock(&fs_info->qgroup_rescan_lock);
2707 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2710 memset(&fs_info->qgroup_rescan_progress, 0,
2711 sizeof(fs_info->qgroup_rescan_progress));
2712 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2713 init_completion(&fs_info->qgroup_rescan_completion);
2714 fs_info->qgroup_rescan_running = true;
2716 spin_unlock(&fs_info->qgroup_lock);
2717 mutex_unlock(&fs_info->qgroup_rescan_lock);
2719 memset(&fs_info->qgroup_rescan_work, 0,
2720 sizeof(fs_info->qgroup_rescan_work));
2721 btrfs_init_work(&fs_info->qgroup_rescan_work,
2722 btrfs_qgroup_rescan_helper,
2723 btrfs_qgroup_rescan_worker, NULL, NULL);
2727 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2735 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2738 struct btrfs_qgroup *qgroup;
2740 spin_lock(&fs_info->qgroup_lock);
2741 /* clear all current qgroup tracking information */
2742 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2743 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2745 qgroup->rfer_cmpr = 0;
2747 qgroup->excl_cmpr = 0;
2749 spin_unlock(&fs_info->qgroup_lock);
2753 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2756 struct btrfs_trans_handle *trans;
2758 ret = qgroup_rescan_init(fs_info, 0, 1);
2763 * We have set the rescan_progress to 0, which means no more
2764 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2765 * However, btrfs_qgroup_account_ref may be right after its call
2766 * to btrfs_find_all_roots, in which case it would still do the
2768 * To solve this, we're committing the transaction, which will
2769 * ensure we run all delayed refs and only after that, we are
2770 * going to clear all tracking information for a clean start.
2773 trans = btrfs_join_transaction(fs_info->fs_root);
2774 if (IS_ERR(trans)) {
2775 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2776 return PTR_ERR(trans);
2778 ret = btrfs_commit_transaction(trans);
2780 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2784 qgroup_rescan_zero_tracking(fs_info);
2786 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2787 &fs_info->qgroup_rescan_work);
2792 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2798 mutex_lock(&fs_info->qgroup_rescan_lock);
2799 spin_lock(&fs_info->qgroup_lock);
2800 running = fs_info->qgroup_rescan_running;
2801 spin_unlock(&fs_info->qgroup_lock);
2802 mutex_unlock(&fs_info->qgroup_rescan_lock);
2808 ret = wait_for_completion_interruptible(
2809 &fs_info->qgroup_rescan_completion);
2811 wait_for_completion(&fs_info->qgroup_rescan_completion);
2817 * this is only called from open_ctree where we're still single threaded, thus
2818 * locking is omitted here.
2821 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2823 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2824 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2825 &fs_info->qgroup_rescan_work);
2829 * Reserve qgroup space for range [start, start + len).
2831 * This function will either reserve space from related qgroups or doing
2832 * nothing if the range is already reserved.
2834 * Return 0 for successful reserve
2835 * Return <0 for error (including -EQUOT)
2837 * NOTE: this function may sleep for memory allocation.
2839 int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len)
2841 struct btrfs_root *root = BTRFS_I(inode)->root;
2842 struct extent_changeset changeset;
2843 struct ulist_node *unode;
2844 struct ulist_iterator uiter;
2847 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2848 !is_fstree(root->objectid) || len == 0)
2851 changeset.bytes_changed = 0;
2852 ulist_init(&changeset.range_changed);
2853 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2854 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2855 trace_btrfs_qgroup_reserve_data(inode, start, len,
2856 changeset.bytes_changed,
2860 ret = qgroup_reserve(root, changeset.bytes_changed, true);
2864 ulist_release(&changeset.range_changed);
2868 /* cleanup already reserved ranges */
2869 ULIST_ITER_INIT(&uiter);
2870 while ((unode = ulist_next(&changeset.range_changed, &uiter)))
2871 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2872 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2874 ulist_release(&changeset.range_changed);
2878 static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len,
2881 struct extent_changeset changeset;
2882 int trace_op = QGROUP_RELEASE;
2885 changeset.bytes_changed = 0;
2886 ulist_init(&changeset.range_changed);
2887 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2888 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2893 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
2894 BTRFS_I(inode)->root->objectid,
2895 changeset.bytes_changed);
2896 trace_op = QGROUP_FREE;
2898 trace_btrfs_qgroup_release_data(inode, start, len,
2899 changeset.bytes_changed, trace_op);
2901 ulist_release(&changeset.range_changed);
2906 * Free a reserved space range from io_tree and related qgroups
2908 * Should be called when a range of pages get invalidated before reaching disk.
2909 * Or for error cleanup case.
2911 * For data written to disk, use btrfs_qgroup_release_data().
2913 * NOTE: This function may sleep for memory allocation.
2915 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len)
2917 return __btrfs_qgroup_release_data(inode, start, len, 1);
2921 * Release a reserved space range from io_tree only.
2923 * Should be called when a range of pages get written to disk and corresponding
2924 * FILE_EXTENT is inserted into corresponding root.
2926 * Since new qgroup accounting framework will only update qgroup numbers at
2927 * commit_transaction() time, its reserved space shouldn't be freed from
2930 * But we should release the range from io_tree, to allow further write to be
2933 * NOTE: This function may sleep for memory allocation.
2935 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
2937 return __btrfs_qgroup_release_data(inode, start, len, 0);
2940 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
2943 struct btrfs_fs_info *fs_info = root->fs_info;
2946 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2947 !is_fstree(root->objectid) || num_bytes == 0)
2950 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
2951 ret = qgroup_reserve(root, num_bytes, enforce);
2954 atomic64_add(num_bytes, &root->qgroup_meta_rsv);
2958 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
2960 struct btrfs_fs_info *fs_info = root->fs_info;
2963 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2964 !is_fstree(root->objectid))
2967 reserved = atomic64_xchg(&root->qgroup_meta_rsv, 0);
2970 btrfs_qgroup_free_refroot(fs_info, root->objectid, reserved);
2973 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
2975 struct btrfs_fs_info *fs_info = root->fs_info;
2977 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2978 !is_fstree(root->objectid))
2981 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
2982 WARN_ON(atomic64_read(&root->qgroup_meta_rsv) < num_bytes);
2983 atomic64_sub(num_bytes, &root->qgroup_meta_rsv);
2984 btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes);
2988 * Check qgroup reserved space leaking, normally at destroy inode
2991 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
2993 struct extent_changeset changeset;
2994 struct ulist_node *unode;
2995 struct ulist_iterator iter;
2998 changeset.bytes_changed = 0;
2999 ulist_init(&changeset.range_changed);
3000 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3001 EXTENT_QGROUP_RESERVED, &changeset);
3004 if (WARN_ON(changeset.bytes_changed)) {
3005 ULIST_ITER_INIT(&iter);
3006 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3007 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3008 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3009 inode->i_ino, unode->val, unode->aux);
3011 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3012 BTRFS_I(inode)->root->objectid,
3013 changeset.bytes_changed);
3016 ulist_release(&changeset.range_changed);