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_NOFS);
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_NOFS);
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(trans, 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);
1042 * The easy accounting, if we are adding/removing the only ref for an extent
1043 * then this qgroup and all of the parent qgroups get their reference and
1044 * exclusive counts adjusted.
1046 * Caller should hold fs_info->qgroup_lock.
1048 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1049 struct ulist *tmp, u64 ref_root,
1050 u64 num_bytes, int sign)
1052 struct btrfs_qgroup *qgroup;
1053 struct btrfs_qgroup_list *glist;
1054 struct ulist_node *unode;
1055 struct ulist_iterator uiter;
1058 qgroup = find_qgroup_rb(fs_info, ref_root);
1062 qgroup->rfer += sign * num_bytes;
1063 qgroup->rfer_cmpr += sign * num_bytes;
1065 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1066 qgroup->excl += sign * num_bytes;
1067 qgroup->excl_cmpr += sign * num_bytes;
1069 qgroup->reserved -= num_bytes;
1071 qgroup_dirty(fs_info, qgroup);
1073 /* Get all of the parent groups that contain this qgroup */
1074 list_for_each_entry(glist, &qgroup->groups, next_group) {
1075 ret = ulist_add(tmp, glist->group->qgroupid,
1076 qgroup_to_aux(glist->group), GFP_ATOMIC);
1081 /* Iterate all of the parents and adjust their reference counts */
1082 ULIST_ITER_INIT(&uiter);
1083 while ((unode = ulist_next(tmp, &uiter))) {
1084 qgroup = unode_aux_to_qgroup(unode);
1085 qgroup->rfer += sign * num_bytes;
1086 qgroup->rfer_cmpr += sign * num_bytes;
1087 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1088 qgroup->excl += sign * num_bytes;
1090 qgroup->reserved -= num_bytes;
1091 qgroup->excl_cmpr += sign * num_bytes;
1092 qgroup_dirty(fs_info, qgroup);
1094 /* Add any parents of the parents */
1095 list_for_each_entry(glist, &qgroup->groups, next_group) {
1096 ret = ulist_add(tmp, glist->group->qgroupid,
1097 qgroup_to_aux(glist->group), GFP_ATOMIC);
1109 * Quick path for updating qgroup with only excl refs.
1111 * In that case, just update all parent will be enough.
1112 * Or we needs to do a full rescan.
1113 * Caller should also hold fs_info->qgroup_lock.
1115 * Return 0 for quick update, return >0 for need to full rescan
1116 * and mark INCONSISTENT flag.
1117 * Return < 0 for other error.
1119 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1120 struct ulist *tmp, u64 src, u64 dst,
1123 struct btrfs_qgroup *qgroup;
1127 qgroup = find_qgroup_rb(fs_info, src);
1130 if (qgroup->excl == qgroup->rfer) {
1132 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1133 qgroup->excl, sign);
1141 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1145 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1146 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1148 struct btrfs_root *quota_root;
1149 struct btrfs_qgroup *parent;
1150 struct btrfs_qgroup *member;
1151 struct btrfs_qgroup_list *list;
1155 /* Check the level of src and dst first */
1156 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1159 tmp = ulist_alloc(GFP_NOFS);
1163 mutex_lock(&fs_info->qgroup_ioctl_lock);
1164 quota_root = fs_info->quota_root;
1169 member = find_qgroup_rb(fs_info, src);
1170 parent = find_qgroup_rb(fs_info, dst);
1171 if (!member || !parent) {
1176 /* check if such qgroup relation exist firstly */
1177 list_for_each_entry(list, &member->groups, next_group) {
1178 if (list->group == parent) {
1184 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1188 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1190 del_qgroup_relation_item(trans, quota_root, src, dst);
1194 spin_lock(&fs_info->qgroup_lock);
1195 ret = add_relation_rb(fs_info, src, dst);
1197 spin_unlock(&fs_info->qgroup_lock);
1200 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1201 spin_unlock(&fs_info->qgroup_lock);
1203 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1208 int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1209 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1211 struct btrfs_root *quota_root;
1212 struct btrfs_qgroup *parent;
1213 struct btrfs_qgroup *member;
1214 struct btrfs_qgroup_list *list;
1219 tmp = ulist_alloc(GFP_NOFS);
1223 quota_root = fs_info->quota_root;
1229 member = find_qgroup_rb(fs_info, src);
1230 parent = find_qgroup_rb(fs_info, dst);
1231 if (!member || !parent) {
1236 /* check if such qgroup relation exist firstly */
1237 list_for_each_entry(list, &member->groups, next_group) {
1238 if (list->group == parent)
1244 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1245 err = del_qgroup_relation_item(trans, quota_root, dst, src);
1249 spin_lock(&fs_info->qgroup_lock);
1250 del_relation_rb(fs_info, src, dst);
1251 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1252 spin_unlock(&fs_info->qgroup_lock);
1258 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1259 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1263 mutex_lock(&fs_info->qgroup_ioctl_lock);
1264 ret = __del_qgroup_relation(trans, fs_info, src, dst);
1265 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1270 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1271 struct btrfs_fs_info *fs_info, u64 qgroupid)
1273 struct btrfs_root *quota_root;
1274 struct btrfs_qgroup *qgroup;
1277 mutex_lock(&fs_info->qgroup_ioctl_lock);
1278 quota_root = fs_info->quota_root;
1283 qgroup = find_qgroup_rb(fs_info, qgroupid);
1289 ret = add_qgroup_item(trans, quota_root, qgroupid);
1293 spin_lock(&fs_info->qgroup_lock);
1294 qgroup = add_qgroup_rb(fs_info, qgroupid);
1295 spin_unlock(&fs_info->qgroup_lock);
1298 ret = PTR_ERR(qgroup);
1300 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1304 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1305 struct btrfs_fs_info *fs_info, u64 qgroupid)
1307 struct btrfs_root *quota_root;
1308 struct btrfs_qgroup *qgroup;
1309 struct btrfs_qgroup_list *list;
1312 mutex_lock(&fs_info->qgroup_ioctl_lock);
1313 quota_root = fs_info->quota_root;
1319 qgroup = find_qgroup_rb(fs_info, qgroupid);
1324 /* check if there are no children of this qgroup */
1325 if (!list_empty(&qgroup->members)) {
1330 ret = del_qgroup_item(trans, quota_root, qgroupid);
1332 while (!list_empty(&qgroup->groups)) {
1333 list = list_first_entry(&qgroup->groups,
1334 struct btrfs_qgroup_list, next_group);
1335 ret = __del_qgroup_relation(trans, fs_info,
1337 list->group->qgroupid);
1342 spin_lock(&fs_info->qgroup_lock);
1343 del_qgroup_rb(fs_info, qgroupid);
1344 spin_unlock(&fs_info->qgroup_lock);
1346 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1350 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1351 struct btrfs_fs_info *fs_info, u64 qgroupid,
1352 struct btrfs_qgroup_limit *limit)
1354 struct btrfs_root *quota_root;
1355 struct btrfs_qgroup *qgroup;
1357 /* Sometimes we would want to clear the limit on this qgroup.
1358 * To meet this requirement, we treat the -1 as a special value
1359 * which tell kernel to clear the limit on this qgroup.
1361 const u64 CLEAR_VALUE = -1;
1363 mutex_lock(&fs_info->qgroup_ioctl_lock);
1364 quota_root = fs_info->quota_root;
1370 qgroup = find_qgroup_rb(fs_info, qgroupid);
1376 spin_lock(&fs_info->qgroup_lock);
1377 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1378 if (limit->max_rfer == CLEAR_VALUE) {
1379 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1380 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1381 qgroup->max_rfer = 0;
1383 qgroup->max_rfer = limit->max_rfer;
1386 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1387 if (limit->max_excl == CLEAR_VALUE) {
1388 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1389 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1390 qgroup->max_excl = 0;
1392 qgroup->max_excl = limit->max_excl;
1395 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1396 if (limit->rsv_rfer == CLEAR_VALUE) {
1397 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1398 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1399 qgroup->rsv_rfer = 0;
1401 qgroup->rsv_rfer = limit->rsv_rfer;
1404 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1405 if (limit->rsv_excl == CLEAR_VALUE) {
1406 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1407 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1408 qgroup->rsv_excl = 0;
1410 qgroup->rsv_excl = limit->rsv_excl;
1413 qgroup->lim_flags |= limit->flags;
1415 spin_unlock(&fs_info->qgroup_lock);
1417 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1419 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1420 btrfs_info(fs_info, "unable to update quota limit for %llu",
1425 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1429 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
1430 struct btrfs_fs_info *fs_info)
1432 struct btrfs_qgroup_extent_record *record;
1433 struct btrfs_delayed_ref_root *delayed_refs;
1434 struct rb_node *node;
1438 delayed_refs = &trans->transaction->delayed_refs;
1439 qgroup_to_skip = delayed_refs->qgroup_to_skip;
1442 * No need to do lock, since this function will only be called in
1443 * btrfs_commit_transaction().
1445 node = rb_first(&delayed_refs->dirty_extent_root);
1447 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1449 ret = btrfs_find_all_roots(NULL, fs_info, record->bytenr, 0,
1450 &record->old_roots);
1454 ulist_del(record->old_roots, qgroup_to_skip, 0);
1455 node = rb_next(node);
1460 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1461 struct btrfs_delayed_ref_root *delayed_refs,
1462 struct btrfs_qgroup_extent_record *record)
1464 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1465 struct rb_node *parent_node = NULL;
1466 struct btrfs_qgroup_extent_record *entry;
1467 u64 bytenr = record->bytenr;
1469 assert_spin_locked(&delayed_refs->lock);
1470 trace_btrfs_qgroup_trace_extent(fs_info, record);
1474 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1476 if (bytenr < entry->bytenr)
1478 else if (bytenr > entry->bytenr)
1479 p = &(*p)->rb_right;
1484 rb_link_node(&record->node, parent_node, p);
1485 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1489 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1490 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1493 struct btrfs_qgroup_extent_record *record;
1494 struct btrfs_delayed_ref_root *delayed_refs;
1497 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1498 || bytenr == 0 || num_bytes == 0)
1500 if (WARN_ON(trans == NULL))
1502 record = kmalloc(sizeof(*record), gfp_flag);
1506 delayed_refs = &trans->transaction->delayed_refs;
1507 record->bytenr = bytenr;
1508 record->num_bytes = num_bytes;
1509 record->old_roots = NULL;
1511 spin_lock(&delayed_refs->lock);
1512 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1513 spin_unlock(&delayed_refs->lock);
1519 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1520 struct btrfs_fs_info *fs_info,
1521 struct extent_buffer *eb)
1523 int nr = btrfs_header_nritems(eb);
1524 int i, extent_type, ret;
1525 struct btrfs_key key;
1526 struct btrfs_file_extent_item *fi;
1527 u64 bytenr, num_bytes;
1529 /* We can be called directly from walk_up_proc() */
1530 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1533 for (i = 0; i < nr; i++) {
1534 btrfs_item_key_to_cpu(eb, &key, i);
1536 if (key.type != BTRFS_EXTENT_DATA_KEY)
1539 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1540 /* filter out non qgroup-accountable extents */
1541 extent_type = btrfs_file_extent_type(eb, fi);
1543 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1546 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1550 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1552 ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1553 num_bytes, GFP_NOFS);
1561 * Walk up the tree from the bottom, freeing leaves and any interior
1562 * nodes which have had all slots visited. If a node (leaf or
1563 * interior) is freed, the node above it will have it's slot
1564 * incremented. The root node will never be freed.
1566 * At the end of this function, we should have a path which has all
1567 * slots incremented to the next position for a search. If we need to
1568 * read a new node it will be NULL and the node above it will have the
1569 * correct slot selected for a later read.
1571 * If we increment the root nodes slot counter past the number of
1572 * elements, 1 is returned to signal completion of the search.
1574 static int adjust_slots_upwards(struct btrfs_root *root,
1575 struct btrfs_path *path, int root_level)
1579 struct extent_buffer *eb;
1581 if (root_level == 0)
1584 while (level <= root_level) {
1585 eb = path->nodes[level];
1586 nr = btrfs_header_nritems(eb);
1587 path->slots[level]++;
1588 slot = path->slots[level];
1589 if (slot >= nr || level == 0) {
1591 * Don't free the root - we will detect this
1592 * condition after our loop and return a
1593 * positive value for caller to stop walking the tree.
1595 if (level != root_level) {
1596 btrfs_tree_unlock_rw(eb, path->locks[level]);
1597 path->locks[level] = 0;
1599 free_extent_buffer(eb);
1600 path->nodes[level] = NULL;
1601 path->slots[level] = 0;
1605 * We have a valid slot to walk back down
1606 * from. Stop here so caller can process these
1615 eb = path->nodes[root_level];
1616 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1622 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1623 struct btrfs_root *root,
1624 struct extent_buffer *root_eb,
1625 u64 root_gen, int root_level)
1627 struct btrfs_fs_info *fs_info = root->fs_info;
1630 struct extent_buffer *eb = root_eb;
1631 struct btrfs_path *path = NULL;
1633 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
1634 BUG_ON(root_eb == NULL);
1636 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1639 if (!extent_buffer_uptodate(root_eb)) {
1640 ret = btrfs_read_buffer(root_eb, root_gen);
1645 if (root_level == 0) {
1646 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1650 path = btrfs_alloc_path();
1655 * Walk down the tree. Missing extent blocks are filled in as
1656 * we go. Metadata is accounted every time we read a new
1659 * When we reach a leaf, we account for file extent items in it,
1660 * walk back up the tree (adjusting slot pointers as we go)
1661 * and restart the search process.
1663 extent_buffer_get(root_eb); /* For path */
1664 path->nodes[root_level] = root_eb;
1665 path->slots[root_level] = 0;
1666 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1669 while (level >= 0) {
1670 if (path->nodes[level] == NULL) {
1676 * We need to get child blockptr/gen from parent before
1679 eb = path->nodes[level + 1];
1680 parent_slot = path->slots[level + 1];
1681 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1682 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1684 eb = read_tree_block(fs_info, child_bytenr, child_gen);
1688 } else if (!extent_buffer_uptodate(eb)) {
1689 free_extent_buffer(eb);
1694 path->nodes[level] = eb;
1695 path->slots[level] = 0;
1697 btrfs_tree_read_lock(eb);
1698 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1699 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1701 ret = btrfs_qgroup_trace_extent(trans, fs_info,
1710 ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1711 path->nodes[level]);
1715 /* Nonzero return here means we completed our search */
1716 ret = adjust_slots_upwards(root, path, root_level);
1720 /* Restart search with new slots */
1729 btrfs_free_path(path);
1734 #define UPDATE_NEW 0
1735 #define UPDATE_OLD 1
1737 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1739 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1740 struct ulist *roots, struct ulist *tmp,
1741 struct ulist *qgroups, u64 seq, int update_old)
1743 struct ulist_node *unode;
1744 struct ulist_iterator uiter;
1745 struct ulist_node *tmp_unode;
1746 struct ulist_iterator tmp_uiter;
1747 struct btrfs_qgroup *qg;
1752 ULIST_ITER_INIT(&uiter);
1753 while ((unode = ulist_next(roots, &uiter))) {
1754 qg = find_qgroup_rb(fs_info, unode->val);
1759 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1763 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1766 ULIST_ITER_INIT(&tmp_uiter);
1767 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1768 struct btrfs_qgroup_list *glist;
1770 qg = unode_aux_to_qgroup(tmp_unode);
1772 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1774 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1775 list_for_each_entry(glist, &qg->groups, next_group) {
1776 ret = ulist_add(qgroups, glist->group->qgroupid,
1777 qgroup_to_aux(glist->group),
1781 ret = ulist_add(tmp, glist->group->qgroupid,
1782 qgroup_to_aux(glist->group),
1793 * Update qgroup rfer/excl counters.
1794 * Rfer update is easy, codes can explain themselves.
1796 * Excl update is tricky, the update is split into 2 part.
1797 * Part 1: Possible exclusive <-> sharing detect:
1799 * -------------------------------------
1801 * -------------------------------------
1803 * -------------------------------------
1806 * A: cur_old_roots < nr_old_roots (not exclusive before)
1807 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1808 * B: cur_new_roots < nr_new_roots (not exclusive now)
1809 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
1812 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1813 * *: Definitely not changed. **: Possible unchanged.
1815 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1817 * To make the logic clear, we first use condition A and B to split
1818 * combination into 4 results.
1820 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1821 * only on variant maybe 0.
1823 * Lastly, check result **, since there are 2 variants maybe 0, split them
1825 * But this time we don't need to consider other things, the codes and logic
1826 * is easy to understand now.
1828 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1829 struct ulist *qgroups,
1832 u64 num_bytes, u64 seq)
1834 struct ulist_node *unode;
1835 struct ulist_iterator uiter;
1836 struct btrfs_qgroup *qg;
1837 u64 cur_new_count, cur_old_count;
1839 ULIST_ITER_INIT(&uiter);
1840 while ((unode = ulist_next(qgroups, &uiter))) {
1843 qg = unode_aux_to_qgroup(unode);
1844 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1845 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1847 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1848 cur_old_count, cur_new_count);
1850 /* Rfer update part */
1851 if (cur_old_count == 0 && cur_new_count > 0) {
1852 qg->rfer += num_bytes;
1853 qg->rfer_cmpr += num_bytes;
1856 if (cur_old_count > 0 && cur_new_count == 0) {
1857 qg->rfer -= num_bytes;
1858 qg->rfer_cmpr -= num_bytes;
1862 /* Excl update part */
1863 /* Exclusive/none -> shared case */
1864 if (cur_old_count == nr_old_roots &&
1865 cur_new_count < nr_new_roots) {
1866 /* Exclusive -> shared */
1867 if (cur_old_count != 0) {
1868 qg->excl -= num_bytes;
1869 qg->excl_cmpr -= num_bytes;
1874 /* Shared -> exclusive/none case */
1875 if (cur_old_count < nr_old_roots &&
1876 cur_new_count == nr_new_roots) {
1877 /* Shared->exclusive */
1878 if (cur_new_count != 0) {
1879 qg->excl += num_bytes;
1880 qg->excl_cmpr += num_bytes;
1885 /* Exclusive/none -> exclusive/none case */
1886 if (cur_old_count == nr_old_roots &&
1887 cur_new_count == nr_new_roots) {
1888 if (cur_old_count == 0) {
1889 /* None -> exclusive/none */
1891 if (cur_new_count != 0) {
1892 /* None -> exclusive */
1893 qg->excl += num_bytes;
1894 qg->excl_cmpr += num_bytes;
1897 /* None -> none, nothing changed */
1899 /* Exclusive -> exclusive/none */
1901 if (cur_new_count == 0) {
1902 /* Exclusive -> none */
1903 qg->excl -= num_bytes;
1904 qg->excl_cmpr -= num_bytes;
1907 /* Exclusive -> exclusive, nothing changed */
1912 qgroup_dirty(fs_info, qg);
1918 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1919 struct btrfs_fs_info *fs_info,
1920 u64 bytenr, u64 num_bytes,
1921 struct ulist *old_roots, struct ulist *new_roots)
1923 struct ulist *qgroups = NULL;
1924 struct ulist *tmp = NULL;
1926 u64 nr_new_roots = 0;
1927 u64 nr_old_roots = 0;
1931 nr_new_roots = new_roots->nnodes;
1933 nr_old_roots = old_roots->nnodes;
1935 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1937 BUG_ON(!fs_info->quota_root);
1939 trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1940 nr_old_roots, nr_new_roots);
1942 qgroups = ulist_alloc(GFP_NOFS);
1947 tmp = ulist_alloc(GFP_NOFS);
1953 mutex_lock(&fs_info->qgroup_rescan_lock);
1954 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1955 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1956 mutex_unlock(&fs_info->qgroup_rescan_lock);
1961 mutex_unlock(&fs_info->qgroup_rescan_lock);
1963 spin_lock(&fs_info->qgroup_lock);
1964 seq = fs_info->qgroup_seq;
1966 /* Update old refcnts using old_roots */
1967 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1972 /* Update new refcnts using new_roots */
1973 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1978 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1982 * Bump qgroup_seq to avoid seq overlap
1984 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
1986 spin_unlock(&fs_info->qgroup_lock);
1989 ulist_free(qgroups);
1990 ulist_free(old_roots);
1991 ulist_free(new_roots);
1995 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
1996 struct btrfs_fs_info *fs_info)
1998 struct btrfs_qgroup_extent_record *record;
1999 struct btrfs_delayed_ref_root *delayed_refs;
2000 struct ulist *new_roots = NULL;
2001 struct rb_node *node;
2005 delayed_refs = &trans->transaction->delayed_refs;
2006 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2007 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2008 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2011 trace_btrfs_qgroup_account_extents(fs_info, record);
2015 * Use (u64)-1 as time_seq to do special search, which
2016 * doesn't lock tree or delayed_refs and search current
2017 * root. It's safe inside commit_transaction().
2019 ret = btrfs_find_all_roots(trans, fs_info,
2020 record->bytenr, (u64)-1, &new_roots);
2024 ulist_del(new_roots, qgroup_to_skip, 0);
2025 ret = btrfs_qgroup_account_extent(trans, fs_info,
2026 record->bytenr, record->num_bytes,
2027 record->old_roots, new_roots);
2028 record->old_roots = NULL;
2032 ulist_free(record->old_roots);
2033 ulist_free(new_roots);
2035 rb_erase(node, &delayed_refs->dirty_extent_root);
2043 * called from commit_transaction. Writes all changed qgroups to disk.
2045 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2046 struct btrfs_fs_info *fs_info)
2048 struct btrfs_root *quota_root = fs_info->quota_root;
2050 int start_rescan_worker = 0;
2055 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) &&
2056 test_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2057 start_rescan_worker = 1;
2059 if (test_and_clear_bit(BTRFS_FS_QUOTA_ENABLING, &fs_info->flags))
2060 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2061 if (test_and_clear_bit(BTRFS_FS_QUOTA_DISABLING, &fs_info->flags))
2062 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
2064 spin_lock(&fs_info->qgroup_lock);
2065 while (!list_empty(&fs_info->dirty_qgroups)) {
2066 struct btrfs_qgroup *qgroup;
2067 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2068 struct btrfs_qgroup, dirty);
2069 list_del_init(&qgroup->dirty);
2070 spin_unlock(&fs_info->qgroup_lock);
2071 ret = update_qgroup_info_item(trans, quota_root, qgroup);
2073 fs_info->qgroup_flags |=
2074 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2075 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2077 fs_info->qgroup_flags |=
2078 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2079 spin_lock(&fs_info->qgroup_lock);
2081 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2082 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2084 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2085 spin_unlock(&fs_info->qgroup_lock);
2087 ret = update_qgroup_status_item(trans, fs_info, quota_root);
2089 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2091 if (!ret && start_rescan_worker) {
2092 ret = qgroup_rescan_init(fs_info, 0, 1);
2094 qgroup_rescan_zero_tracking(fs_info);
2095 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2096 &fs_info->qgroup_rescan_work);
2107 * Copy the accounting information between qgroups. This is necessary
2108 * when a snapshot or a subvolume is created. Throwing an error will
2109 * cause a transaction abort so we take extra care here to only error
2110 * when a readonly fs is a reasonable outcome.
2112 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2113 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2114 struct btrfs_qgroup_inherit *inherit)
2119 struct btrfs_root *quota_root = fs_info->quota_root;
2120 struct btrfs_qgroup *srcgroup;
2121 struct btrfs_qgroup *dstgroup;
2125 mutex_lock(&fs_info->qgroup_ioctl_lock);
2126 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2135 i_qgroups = (u64 *)(inherit + 1);
2136 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2137 2 * inherit->num_excl_copies;
2138 for (i = 0; i < nums; ++i) {
2139 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2142 * Zero out invalid groups so we can ignore
2146 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2154 * create a tracking group for the subvol itself
2156 ret = add_qgroup_item(trans, quota_root, objectid);
2161 struct btrfs_root *srcroot;
2162 struct btrfs_key srckey;
2164 srckey.objectid = srcid;
2165 srckey.type = BTRFS_ROOT_ITEM_KEY;
2166 srckey.offset = (u64)-1;
2167 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2168 if (IS_ERR(srcroot)) {
2169 ret = PTR_ERR(srcroot);
2174 level_size = fs_info->nodesize;
2179 * add qgroup to all inherited groups
2182 i_qgroups = (u64 *)(inherit + 1);
2183 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2184 if (*i_qgroups == 0)
2186 ret = add_qgroup_relation_item(trans, quota_root,
2187 objectid, *i_qgroups);
2188 if (ret && ret != -EEXIST)
2190 ret = add_qgroup_relation_item(trans, quota_root,
2191 *i_qgroups, objectid);
2192 if (ret && ret != -EEXIST)
2199 spin_lock(&fs_info->qgroup_lock);
2201 dstgroup = add_qgroup_rb(fs_info, objectid);
2202 if (IS_ERR(dstgroup)) {
2203 ret = PTR_ERR(dstgroup);
2207 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2208 dstgroup->lim_flags = inherit->lim.flags;
2209 dstgroup->max_rfer = inherit->lim.max_rfer;
2210 dstgroup->max_excl = inherit->lim.max_excl;
2211 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2212 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2214 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2216 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2218 "unable to update quota limit for %llu",
2219 dstgroup->qgroupid);
2225 srcgroup = find_qgroup_rb(fs_info, srcid);
2230 * We call inherit after we clone the root in order to make sure
2231 * our counts don't go crazy, so at this point the only
2232 * difference between the two roots should be the root node.
2234 dstgroup->rfer = srcgroup->rfer;
2235 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2236 dstgroup->excl = level_size;
2237 dstgroup->excl_cmpr = level_size;
2238 srcgroup->excl = level_size;
2239 srcgroup->excl_cmpr = level_size;
2241 /* inherit the limit info */
2242 dstgroup->lim_flags = srcgroup->lim_flags;
2243 dstgroup->max_rfer = srcgroup->max_rfer;
2244 dstgroup->max_excl = srcgroup->max_excl;
2245 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2246 dstgroup->rsv_excl = srcgroup->rsv_excl;
2248 qgroup_dirty(fs_info, dstgroup);
2249 qgroup_dirty(fs_info, srcgroup);
2255 i_qgroups = (u64 *)(inherit + 1);
2256 for (i = 0; i < inherit->num_qgroups; ++i) {
2258 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2265 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2266 struct btrfs_qgroup *src;
2267 struct btrfs_qgroup *dst;
2269 if (!i_qgroups[0] || !i_qgroups[1])
2272 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2273 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2280 dst->rfer = src->rfer - level_size;
2281 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2283 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2284 struct btrfs_qgroup *src;
2285 struct btrfs_qgroup *dst;
2287 if (!i_qgroups[0] || !i_qgroups[1])
2290 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2291 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2298 dst->excl = src->excl + level_size;
2299 dst->excl_cmpr = src->excl_cmpr + level_size;
2303 spin_unlock(&fs_info->qgroup_lock);
2305 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2309 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
2311 struct btrfs_root *quota_root;
2312 struct btrfs_qgroup *qgroup;
2313 struct btrfs_fs_info *fs_info = root->fs_info;
2314 u64 ref_root = root->root_key.objectid;
2316 struct ulist_node *unode;
2317 struct ulist_iterator uiter;
2319 if (!is_fstree(ref_root))
2325 spin_lock(&fs_info->qgroup_lock);
2326 quota_root = fs_info->quota_root;
2330 qgroup = find_qgroup_rb(fs_info, ref_root);
2335 * in a first step, we check all affected qgroups if any limits would
2338 ulist_reinit(fs_info->qgroup_ulist);
2339 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2340 (uintptr_t)qgroup, GFP_ATOMIC);
2343 ULIST_ITER_INIT(&uiter);
2344 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2345 struct btrfs_qgroup *qg;
2346 struct btrfs_qgroup_list *glist;
2348 qg = unode_aux_to_qgroup(unode);
2350 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2351 qg->reserved + (s64)qg->rfer + num_bytes >
2357 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2358 qg->reserved + (s64)qg->excl + num_bytes >
2364 list_for_each_entry(glist, &qg->groups, next_group) {
2365 ret = ulist_add(fs_info->qgroup_ulist,
2366 glist->group->qgroupid,
2367 (uintptr_t)glist->group, GFP_ATOMIC);
2374 * no limits exceeded, now record the reservation into all qgroups
2376 ULIST_ITER_INIT(&uiter);
2377 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2378 struct btrfs_qgroup *qg;
2380 qg = unode_aux_to_qgroup(unode);
2382 qg->reserved += num_bytes;
2386 spin_unlock(&fs_info->qgroup_lock);
2390 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2391 u64 ref_root, u64 num_bytes)
2393 struct btrfs_root *quota_root;
2394 struct btrfs_qgroup *qgroup;
2395 struct ulist_node *unode;
2396 struct ulist_iterator uiter;
2399 if (!is_fstree(ref_root))
2405 spin_lock(&fs_info->qgroup_lock);
2407 quota_root = fs_info->quota_root;
2411 qgroup = find_qgroup_rb(fs_info, ref_root);
2415 ulist_reinit(fs_info->qgroup_ulist);
2416 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2417 (uintptr_t)qgroup, GFP_ATOMIC);
2420 ULIST_ITER_INIT(&uiter);
2421 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2422 struct btrfs_qgroup *qg;
2423 struct btrfs_qgroup_list *glist;
2425 qg = unode_aux_to_qgroup(unode);
2427 qg->reserved -= num_bytes;
2429 list_for_each_entry(glist, &qg->groups, next_group) {
2430 ret = ulist_add(fs_info->qgroup_ulist,
2431 glist->group->qgroupid,
2432 (uintptr_t)glist->group, GFP_ATOMIC);
2439 spin_unlock(&fs_info->qgroup_lock);
2442 static inline void qgroup_free(struct btrfs_root *root, u64 num_bytes)
2444 return btrfs_qgroup_free_refroot(root->fs_info, root->objectid,
2447 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2449 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2451 btrfs_err(trans->fs_info,
2452 "qgroups not uptodate in trans handle %p: list is%s empty, seq is %#x.%x",
2453 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
2454 (u32)(trans->delayed_ref_elem.seq >> 32),
2455 (u32)trans->delayed_ref_elem.seq);
2460 * returns < 0 on error, 0 when more leafs are to be scanned.
2461 * returns 1 when done.
2464 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2465 struct btrfs_trans_handle *trans)
2467 struct btrfs_key found;
2468 struct extent_buffer *scratch_leaf = NULL;
2469 struct ulist *roots = NULL;
2470 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2475 mutex_lock(&fs_info->qgroup_rescan_lock);
2476 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2477 &fs_info->qgroup_rescan_progress,
2480 btrfs_debug(fs_info,
2481 "current progress key (%llu %u %llu), search_slot ret %d",
2482 fs_info->qgroup_rescan_progress.objectid,
2483 fs_info->qgroup_rescan_progress.type,
2484 fs_info->qgroup_rescan_progress.offset, ret);
2488 * The rescan is about to end, we will not be scanning any
2489 * further blocks. We cannot unset the RESCAN flag here, because
2490 * we want to commit the transaction if everything went well.
2491 * To make the live accounting work in this phase, we set our
2492 * scan progress pointer such that every real extent objectid
2495 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2496 btrfs_release_path(path);
2497 mutex_unlock(&fs_info->qgroup_rescan_lock);
2501 btrfs_item_key_to_cpu(path->nodes[0], &found,
2502 btrfs_header_nritems(path->nodes[0]) - 1);
2503 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2505 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2506 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2507 if (!scratch_leaf) {
2509 mutex_unlock(&fs_info->qgroup_rescan_lock);
2512 extent_buffer_get(scratch_leaf);
2513 btrfs_tree_read_lock(scratch_leaf);
2514 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2515 slot = path->slots[0];
2516 btrfs_release_path(path);
2517 mutex_unlock(&fs_info->qgroup_rescan_lock);
2519 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2520 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2521 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2522 found.type != BTRFS_METADATA_ITEM_KEY)
2524 if (found.type == BTRFS_METADATA_ITEM_KEY)
2525 num_bytes = fs_info->nodesize;
2527 num_bytes = found.offset;
2529 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2533 /* For rescan, just pass old_roots as NULL */
2534 ret = btrfs_qgroup_account_extent(trans, fs_info,
2535 found.objectid, num_bytes, NULL, roots);
2541 btrfs_tree_read_unlock_blocking(scratch_leaf);
2542 free_extent_buffer(scratch_leaf);
2544 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2549 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2551 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2552 qgroup_rescan_work);
2553 struct btrfs_path *path;
2554 struct btrfs_trans_handle *trans = NULL;
2558 path = btrfs_alloc_path();
2563 while (!err && !btrfs_fs_closing(fs_info)) {
2564 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2565 if (IS_ERR(trans)) {
2566 err = PTR_ERR(trans);
2569 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2572 err = qgroup_rescan_leaf(fs_info, path, trans);
2575 btrfs_commit_transaction(trans);
2577 btrfs_end_transaction(trans);
2581 btrfs_free_path(path);
2583 mutex_lock(&fs_info->qgroup_rescan_lock);
2584 if (!btrfs_fs_closing(fs_info))
2585 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2588 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2589 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2590 } else if (err < 0) {
2591 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2593 mutex_unlock(&fs_info->qgroup_rescan_lock);
2596 * only update status, since the previous part has already updated the
2599 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2600 if (IS_ERR(trans)) {
2601 err = PTR_ERR(trans);
2603 "fail to start transaction for status update: %d\n",
2607 ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2610 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2612 btrfs_end_transaction(trans);
2614 if (btrfs_fs_closing(fs_info)) {
2615 btrfs_info(fs_info, "qgroup scan paused");
2616 } else if (err >= 0) {
2617 btrfs_info(fs_info, "qgroup scan completed%s",
2618 err > 0 ? " (inconsistency flag cleared)" : "");
2620 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2624 mutex_lock(&fs_info->qgroup_rescan_lock);
2625 fs_info->qgroup_rescan_running = false;
2626 mutex_unlock(&fs_info->qgroup_rescan_lock);
2627 complete_all(&fs_info->qgroup_rescan_completion);
2631 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2632 * memory required for the rescan context.
2635 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2641 (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2642 !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2647 mutex_lock(&fs_info->qgroup_rescan_lock);
2648 spin_lock(&fs_info->qgroup_lock);
2651 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2653 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2657 spin_unlock(&fs_info->qgroup_lock);
2658 mutex_unlock(&fs_info->qgroup_rescan_lock);
2661 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2664 memset(&fs_info->qgroup_rescan_progress, 0,
2665 sizeof(fs_info->qgroup_rescan_progress));
2666 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2667 init_completion(&fs_info->qgroup_rescan_completion);
2668 fs_info->qgroup_rescan_running = true;
2670 spin_unlock(&fs_info->qgroup_lock);
2671 mutex_unlock(&fs_info->qgroup_rescan_lock);
2673 memset(&fs_info->qgroup_rescan_work, 0,
2674 sizeof(fs_info->qgroup_rescan_work));
2675 btrfs_init_work(&fs_info->qgroup_rescan_work,
2676 btrfs_qgroup_rescan_helper,
2677 btrfs_qgroup_rescan_worker, NULL, NULL);
2681 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2689 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2692 struct btrfs_qgroup *qgroup;
2694 spin_lock(&fs_info->qgroup_lock);
2695 /* clear all current qgroup tracking information */
2696 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2697 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2699 qgroup->rfer_cmpr = 0;
2701 qgroup->excl_cmpr = 0;
2703 spin_unlock(&fs_info->qgroup_lock);
2707 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2710 struct btrfs_trans_handle *trans;
2712 ret = qgroup_rescan_init(fs_info, 0, 1);
2717 * We have set the rescan_progress to 0, which means no more
2718 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2719 * However, btrfs_qgroup_account_ref may be right after its call
2720 * to btrfs_find_all_roots, in which case it would still do the
2722 * To solve this, we're committing the transaction, which will
2723 * ensure we run all delayed refs and only after that, we are
2724 * going to clear all tracking information for a clean start.
2727 trans = btrfs_join_transaction(fs_info->fs_root);
2728 if (IS_ERR(trans)) {
2729 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2730 return PTR_ERR(trans);
2732 ret = btrfs_commit_transaction(trans);
2734 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2738 qgroup_rescan_zero_tracking(fs_info);
2740 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2741 &fs_info->qgroup_rescan_work);
2746 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2752 mutex_lock(&fs_info->qgroup_rescan_lock);
2753 spin_lock(&fs_info->qgroup_lock);
2754 running = fs_info->qgroup_rescan_running;
2755 spin_unlock(&fs_info->qgroup_lock);
2756 mutex_unlock(&fs_info->qgroup_rescan_lock);
2762 ret = wait_for_completion_interruptible(
2763 &fs_info->qgroup_rescan_completion);
2765 wait_for_completion(&fs_info->qgroup_rescan_completion);
2771 * this is only called from open_ctree where we're still single threaded, thus
2772 * locking is omitted here.
2775 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2777 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2778 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2779 &fs_info->qgroup_rescan_work);
2783 * Reserve qgroup space for range [start, start + len).
2785 * This function will either reserve space from related qgroups or doing
2786 * nothing if the range is already reserved.
2788 * Return 0 for successful reserve
2789 * Return <0 for error (including -EQUOT)
2791 * NOTE: this function may sleep for memory allocation.
2793 int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len)
2795 struct btrfs_root *root = BTRFS_I(inode)->root;
2796 struct extent_changeset changeset;
2797 struct ulist_node *unode;
2798 struct ulist_iterator uiter;
2801 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2802 !is_fstree(root->objectid) || len == 0)
2805 changeset.bytes_changed = 0;
2806 changeset.range_changed = ulist_alloc(GFP_NOFS);
2807 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2808 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2809 trace_btrfs_qgroup_reserve_data(inode, start, len,
2810 changeset.bytes_changed,
2814 ret = qgroup_reserve(root, changeset.bytes_changed);
2818 ulist_free(changeset.range_changed);
2822 /* cleanup already reserved ranges */
2823 ULIST_ITER_INIT(&uiter);
2824 while ((unode = ulist_next(changeset.range_changed, &uiter)))
2825 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2826 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2828 ulist_free(changeset.range_changed);
2832 static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len,
2835 struct extent_changeset changeset;
2836 int trace_op = QGROUP_RELEASE;
2839 changeset.bytes_changed = 0;
2840 changeset.range_changed = ulist_alloc(GFP_NOFS);
2841 if (!changeset.range_changed)
2844 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2845 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2850 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2851 trace_op = QGROUP_FREE;
2853 trace_btrfs_qgroup_release_data(inode, start, len,
2854 changeset.bytes_changed, trace_op);
2856 ulist_free(changeset.range_changed);
2861 * Free a reserved space range from io_tree and related qgroups
2863 * Should be called when a range of pages get invalidated before reaching disk.
2864 * Or for error cleanup case.
2866 * For data written to disk, use btrfs_qgroup_release_data().
2868 * NOTE: This function may sleep for memory allocation.
2870 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len)
2872 return __btrfs_qgroup_release_data(inode, start, len, 1);
2876 * Release a reserved space range from io_tree only.
2878 * Should be called when a range of pages get written to disk and corresponding
2879 * FILE_EXTENT is inserted into corresponding root.
2881 * Since new qgroup accounting framework will only update qgroup numbers at
2882 * commit_transaction() time, its reserved space shouldn't be freed from
2885 * But we should release the range from io_tree, to allow further write to be
2888 * NOTE: This function may sleep for memory allocation.
2890 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
2892 return __btrfs_qgroup_release_data(inode, start, len, 0);
2895 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes)
2897 struct btrfs_fs_info *fs_info = root->fs_info;
2900 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2901 !is_fstree(root->objectid) || num_bytes == 0)
2904 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
2905 ret = qgroup_reserve(root, num_bytes);
2908 atomic_add(num_bytes, &root->qgroup_meta_rsv);
2912 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
2914 struct btrfs_fs_info *fs_info = root->fs_info;
2917 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2918 !is_fstree(root->objectid))
2921 reserved = atomic_xchg(&root->qgroup_meta_rsv, 0);
2924 qgroup_free(root, reserved);
2927 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
2929 struct btrfs_fs_info *fs_info = root->fs_info;
2931 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2932 !is_fstree(root->objectid))
2935 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
2936 WARN_ON(atomic_read(&root->qgroup_meta_rsv) < num_bytes);
2937 atomic_sub(num_bytes, &root->qgroup_meta_rsv);
2938 qgroup_free(root, num_bytes);
2942 * Check qgroup reserved space leaking, normally at destroy inode
2945 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
2947 struct extent_changeset changeset;
2948 struct ulist_node *unode;
2949 struct ulist_iterator iter;
2952 changeset.bytes_changed = 0;
2953 changeset.range_changed = ulist_alloc(GFP_NOFS);
2954 if (WARN_ON(!changeset.range_changed))
2957 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
2958 EXTENT_QGROUP_RESERVED, &changeset);
2961 if (WARN_ON(changeset.bytes_changed)) {
2962 ULIST_ITER_INIT(&iter);
2963 while ((unode = ulist_next(changeset.range_changed, &iter))) {
2964 btrfs_warn(BTRFS_I(inode)->root->fs_info,
2965 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
2966 inode->i_ino, unode->val, unode->aux);
2968 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2970 ulist_free(changeset.range_changed);