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_accouting() 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 #define ptr_to_u64(x) ((u64)(uintptr_t)x)
135 #define u64_to_ptr(x) ((struct btrfs_qgroup *)(uintptr_t)x)
138 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
140 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
142 /* must be called with qgroup_ioctl_lock held */
143 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
146 struct rb_node *n = fs_info->qgroup_tree.rb_node;
147 struct btrfs_qgroup *qgroup;
150 qgroup = rb_entry(n, struct btrfs_qgroup, node);
151 if (qgroup->qgroupid < qgroupid)
153 else if (qgroup->qgroupid > qgroupid)
161 /* must be called with qgroup_lock held */
162 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
165 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
166 struct rb_node *parent = NULL;
167 struct btrfs_qgroup *qgroup;
171 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
173 if (qgroup->qgroupid < qgroupid)
175 else if (qgroup->qgroupid > qgroupid)
181 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
183 return ERR_PTR(-ENOMEM);
185 qgroup->qgroupid = qgroupid;
186 INIT_LIST_HEAD(&qgroup->groups);
187 INIT_LIST_HEAD(&qgroup->members);
188 INIT_LIST_HEAD(&qgroup->dirty);
190 rb_link_node(&qgroup->node, parent, p);
191 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
196 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
198 struct btrfs_qgroup_list *list;
200 list_del(&qgroup->dirty);
201 while (!list_empty(&qgroup->groups)) {
202 list = list_first_entry(&qgroup->groups,
203 struct btrfs_qgroup_list, next_group);
204 list_del(&list->next_group);
205 list_del(&list->next_member);
209 while (!list_empty(&qgroup->members)) {
210 list = list_first_entry(&qgroup->members,
211 struct btrfs_qgroup_list, next_member);
212 list_del(&list->next_group);
213 list_del(&list->next_member);
219 /* must be called with qgroup_lock held */
220 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
222 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
227 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
228 __del_qgroup_rb(qgroup);
232 /* must be called with qgroup_lock held */
233 static int add_relation_rb(struct btrfs_fs_info *fs_info,
234 u64 memberid, u64 parentid)
236 struct btrfs_qgroup *member;
237 struct btrfs_qgroup *parent;
238 struct btrfs_qgroup_list *list;
240 member = find_qgroup_rb(fs_info, memberid);
241 parent = find_qgroup_rb(fs_info, parentid);
242 if (!member || !parent)
245 list = kzalloc(sizeof(*list), GFP_ATOMIC);
249 list->group = parent;
250 list->member = member;
251 list_add_tail(&list->next_group, &member->groups);
252 list_add_tail(&list->next_member, &parent->members);
257 /* must be called with qgroup_lock held */
258 static int del_relation_rb(struct btrfs_fs_info *fs_info,
259 u64 memberid, u64 parentid)
261 struct btrfs_qgroup *member;
262 struct btrfs_qgroup *parent;
263 struct btrfs_qgroup_list *list;
265 member = find_qgroup_rb(fs_info, memberid);
266 parent = find_qgroup_rb(fs_info, parentid);
267 if (!member || !parent)
270 list_for_each_entry(list, &member->groups, next_group) {
271 if (list->group == parent) {
272 list_del(&list->next_group);
273 list_del(&list->next_member);
281 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
282 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
285 struct btrfs_qgroup *qgroup;
287 qgroup = find_qgroup_rb(fs_info, qgroupid);
290 if (qgroup->rfer != rfer || qgroup->excl != excl)
297 * The full config is read in one go, only called from open_ctree()
298 * It doesn't use any locking, as at this point we're still single-threaded
300 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
302 struct btrfs_key key;
303 struct btrfs_key found_key;
304 struct btrfs_root *quota_root = fs_info->quota_root;
305 struct btrfs_path *path = NULL;
306 struct extent_buffer *l;
310 u64 rescan_progress = 0;
312 if (!fs_info->quota_enabled)
315 fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
316 if (!fs_info->qgroup_ulist) {
321 path = btrfs_alloc_path();
327 /* default this to quota off, in case no status key is found */
328 fs_info->qgroup_flags = 0;
331 * pass 1: read status, all qgroup infos and limits
336 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
341 struct btrfs_qgroup *qgroup;
343 slot = path->slots[0];
345 btrfs_item_key_to_cpu(l, &found_key, slot);
347 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
348 struct btrfs_qgroup_status_item *ptr;
350 ptr = btrfs_item_ptr(l, slot,
351 struct btrfs_qgroup_status_item);
353 if (btrfs_qgroup_status_version(l, ptr) !=
354 BTRFS_QGROUP_STATUS_VERSION) {
356 "old qgroup version, quota disabled");
359 if (btrfs_qgroup_status_generation(l, ptr) !=
360 fs_info->generation) {
361 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
363 "qgroup generation mismatch, "
364 "marked as inconsistent");
366 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
368 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
372 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
373 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
376 qgroup = find_qgroup_rb(fs_info, found_key.offset);
377 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
378 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
379 btrfs_err(fs_info, "inconsistent qgroup config");
380 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
383 qgroup = add_qgroup_rb(fs_info, found_key.offset);
384 if (IS_ERR(qgroup)) {
385 ret = PTR_ERR(qgroup);
389 switch (found_key.type) {
390 case BTRFS_QGROUP_INFO_KEY: {
391 struct btrfs_qgroup_info_item *ptr;
393 ptr = btrfs_item_ptr(l, slot,
394 struct btrfs_qgroup_info_item);
395 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
396 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
397 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
398 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
399 /* generation currently unused */
402 case BTRFS_QGROUP_LIMIT_KEY: {
403 struct btrfs_qgroup_limit_item *ptr;
405 ptr = btrfs_item_ptr(l, slot,
406 struct btrfs_qgroup_limit_item);
407 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
408 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
409 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
410 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
411 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
416 ret = btrfs_next_item(quota_root, path);
422 btrfs_release_path(path);
425 * pass 2: read all qgroup relations
428 key.type = BTRFS_QGROUP_RELATION_KEY;
430 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
434 slot = path->slots[0];
436 btrfs_item_key_to_cpu(l, &found_key, slot);
438 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
441 if (found_key.objectid > found_key.offset) {
442 /* parent <- member, not needed to build config */
443 /* FIXME should we omit the key completely? */
447 ret = add_relation_rb(fs_info, found_key.objectid,
449 if (ret == -ENOENT) {
451 "orphan qgroup relation 0x%llx->0x%llx",
452 found_key.objectid, found_key.offset);
453 ret = 0; /* ignore the error */
458 ret = btrfs_next_item(quota_root, path);
465 fs_info->qgroup_flags |= flags;
466 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)) {
467 fs_info->quota_enabled = 0;
468 fs_info->pending_quota_state = 0;
469 } else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
471 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
473 btrfs_free_path(path);
476 ulist_free(fs_info->qgroup_ulist);
477 fs_info->qgroup_ulist = NULL;
478 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
481 return ret < 0 ? ret : 0;
485 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
486 * first two are in single-threaded paths.And for the third one, we have set
487 * quota_root to be null with qgroup_lock held before, so it is safe to clean
488 * up the in-memory structures without qgroup_lock held.
490 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
493 struct btrfs_qgroup *qgroup;
495 while ((n = rb_first(&fs_info->qgroup_tree))) {
496 qgroup = rb_entry(n, struct btrfs_qgroup, node);
497 rb_erase(n, &fs_info->qgroup_tree);
498 __del_qgroup_rb(qgroup);
501 * we call btrfs_free_qgroup_config() when umounting
502 * filesystem and disabling quota, so we set qgroup_ulit
503 * to be null here to avoid double free.
505 ulist_free(fs_info->qgroup_ulist);
506 fs_info->qgroup_ulist = NULL;
509 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
510 struct btrfs_root *quota_root,
514 struct btrfs_path *path;
515 struct btrfs_key key;
517 path = btrfs_alloc_path();
522 key.type = BTRFS_QGROUP_RELATION_KEY;
525 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
527 btrfs_mark_buffer_dirty(path->nodes[0]);
529 btrfs_free_path(path);
533 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
534 struct btrfs_root *quota_root,
538 struct btrfs_path *path;
539 struct btrfs_key key;
541 path = btrfs_alloc_path();
546 key.type = BTRFS_QGROUP_RELATION_KEY;
549 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
558 ret = btrfs_del_item(trans, quota_root, path);
560 btrfs_free_path(path);
564 static int add_qgroup_item(struct btrfs_trans_handle *trans,
565 struct btrfs_root *quota_root, u64 qgroupid)
568 struct btrfs_path *path;
569 struct btrfs_qgroup_info_item *qgroup_info;
570 struct btrfs_qgroup_limit_item *qgroup_limit;
571 struct extent_buffer *leaf;
572 struct btrfs_key key;
574 if (btrfs_test_is_dummy_root(quota_root))
577 path = btrfs_alloc_path();
582 key.type = BTRFS_QGROUP_INFO_KEY;
583 key.offset = qgroupid;
586 * Avoid a transaction abort by catching -EEXIST here. In that
587 * case, we proceed by re-initializing the existing structure
591 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
592 sizeof(*qgroup_info));
593 if (ret && ret != -EEXIST)
596 leaf = path->nodes[0];
597 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
598 struct btrfs_qgroup_info_item);
599 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
600 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
601 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
602 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
603 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
605 btrfs_mark_buffer_dirty(leaf);
607 btrfs_release_path(path);
609 key.type = BTRFS_QGROUP_LIMIT_KEY;
610 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
611 sizeof(*qgroup_limit));
612 if (ret && ret != -EEXIST)
615 leaf = path->nodes[0];
616 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
617 struct btrfs_qgroup_limit_item);
618 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
619 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
620 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
621 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
622 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
624 btrfs_mark_buffer_dirty(leaf);
628 btrfs_free_path(path);
632 static int del_qgroup_item(struct btrfs_trans_handle *trans,
633 struct btrfs_root *quota_root, u64 qgroupid)
636 struct btrfs_path *path;
637 struct btrfs_key key;
639 path = btrfs_alloc_path();
644 key.type = BTRFS_QGROUP_INFO_KEY;
645 key.offset = qgroupid;
646 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
655 ret = btrfs_del_item(trans, quota_root, path);
659 btrfs_release_path(path);
661 key.type = BTRFS_QGROUP_LIMIT_KEY;
662 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
671 ret = btrfs_del_item(trans, quota_root, path);
674 btrfs_free_path(path);
678 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
679 struct btrfs_root *root,
680 struct btrfs_qgroup *qgroup)
682 struct btrfs_path *path;
683 struct btrfs_key key;
684 struct extent_buffer *l;
685 struct btrfs_qgroup_limit_item *qgroup_limit;
690 key.type = BTRFS_QGROUP_LIMIT_KEY;
691 key.offset = qgroup->qgroupid;
693 path = btrfs_alloc_path();
697 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
705 slot = path->slots[0];
706 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
707 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
708 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
709 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
710 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
711 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
713 btrfs_mark_buffer_dirty(l);
716 btrfs_free_path(path);
720 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
721 struct btrfs_root *root,
722 struct btrfs_qgroup *qgroup)
724 struct btrfs_path *path;
725 struct btrfs_key key;
726 struct extent_buffer *l;
727 struct btrfs_qgroup_info_item *qgroup_info;
731 if (btrfs_test_is_dummy_root(root))
735 key.type = BTRFS_QGROUP_INFO_KEY;
736 key.offset = qgroup->qgroupid;
738 path = btrfs_alloc_path();
742 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
750 slot = path->slots[0];
751 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
752 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
753 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
754 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
755 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
756 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
758 btrfs_mark_buffer_dirty(l);
761 btrfs_free_path(path);
765 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
766 struct btrfs_fs_info *fs_info,
767 struct btrfs_root *root)
769 struct btrfs_path *path;
770 struct btrfs_key key;
771 struct extent_buffer *l;
772 struct btrfs_qgroup_status_item *ptr;
777 key.type = BTRFS_QGROUP_STATUS_KEY;
780 path = btrfs_alloc_path();
784 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
792 slot = path->slots[0];
793 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
794 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
795 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
796 btrfs_set_qgroup_status_rescan(l, ptr,
797 fs_info->qgroup_rescan_progress.objectid);
799 btrfs_mark_buffer_dirty(l);
802 btrfs_free_path(path);
807 * called with qgroup_lock held
809 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
810 struct btrfs_root *root)
812 struct btrfs_path *path;
813 struct btrfs_key key;
814 struct extent_buffer *leaf = NULL;
818 path = btrfs_alloc_path();
822 path->leave_spinning = 1;
829 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
832 leaf = path->nodes[0];
833 nr = btrfs_header_nritems(leaf);
837 * delete the leaf one by one
838 * since the whole tree is going
842 ret = btrfs_del_items(trans, root, path, 0, nr);
846 btrfs_release_path(path);
850 root->fs_info->pending_quota_state = 0;
851 btrfs_free_path(path);
855 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
856 struct btrfs_fs_info *fs_info)
858 struct btrfs_root *quota_root;
859 struct btrfs_root *tree_root = fs_info->tree_root;
860 struct btrfs_path *path = NULL;
861 struct btrfs_qgroup_status_item *ptr;
862 struct extent_buffer *leaf;
863 struct btrfs_key key;
864 struct btrfs_key found_key;
865 struct btrfs_qgroup *qgroup = NULL;
869 mutex_lock(&fs_info->qgroup_ioctl_lock);
870 if (fs_info->quota_root) {
871 fs_info->pending_quota_state = 1;
875 fs_info->qgroup_ulist = ulist_alloc(GFP_NOFS);
876 if (!fs_info->qgroup_ulist) {
882 * initially create the quota tree
884 quota_root = btrfs_create_tree(trans, fs_info,
885 BTRFS_QUOTA_TREE_OBJECTID);
886 if (IS_ERR(quota_root)) {
887 ret = PTR_ERR(quota_root);
891 path = btrfs_alloc_path();
898 key.type = BTRFS_QGROUP_STATUS_KEY;
901 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
906 leaf = path->nodes[0];
907 ptr = btrfs_item_ptr(leaf, path->slots[0],
908 struct btrfs_qgroup_status_item);
909 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
910 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
911 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
912 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
913 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
914 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
916 btrfs_mark_buffer_dirty(leaf);
919 key.type = BTRFS_ROOT_REF_KEY;
922 btrfs_release_path(path);
923 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
931 slot = path->slots[0];
932 leaf = path->nodes[0];
933 btrfs_item_key_to_cpu(leaf, &found_key, slot);
935 if (found_key.type == BTRFS_ROOT_REF_KEY) {
936 ret = add_qgroup_item(trans, quota_root,
941 qgroup = add_qgroup_rb(fs_info, found_key.offset);
942 if (IS_ERR(qgroup)) {
943 ret = PTR_ERR(qgroup);
947 ret = btrfs_next_item(tree_root, path);
955 btrfs_release_path(path);
956 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
960 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
961 if (IS_ERR(qgroup)) {
962 ret = PTR_ERR(qgroup);
965 spin_lock(&fs_info->qgroup_lock);
966 fs_info->quota_root = quota_root;
967 fs_info->pending_quota_state = 1;
968 spin_unlock(&fs_info->qgroup_lock);
970 btrfs_free_path(path);
973 free_extent_buffer(quota_root->node);
974 free_extent_buffer(quota_root->commit_root);
979 ulist_free(fs_info->qgroup_ulist);
980 fs_info->qgroup_ulist = NULL;
982 mutex_unlock(&fs_info->qgroup_ioctl_lock);
986 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
987 struct btrfs_fs_info *fs_info)
989 struct btrfs_root *tree_root = fs_info->tree_root;
990 struct btrfs_root *quota_root;
993 mutex_lock(&fs_info->qgroup_ioctl_lock);
994 if (!fs_info->quota_root)
996 fs_info->quota_enabled = 0;
997 fs_info->pending_quota_state = 0;
998 btrfs_qgroup_wait_for_completion(fs_info, false);
999 spin_lock(&fs_info->qgroup_lock);
1000 quota_root = fs_info->quota_root;
1001 fs_info->quota_root = NULL;
1002 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1003 spin_unlock(&fs_info->qgroup_lock);
1005 btrfs_free_qgroup_config(fs_info);
1007 ret = btrfs_clean_quota_tree(trans, quota_root);
1011 ret = btrfs_del_root(trans, tree_root, "a_root->root_key);
1015 list_del("a_root->dirty_list);
1017 btrfs_tree_lock(quota_root->node);
1018 clean_tree_block(trans, tree_root->fs_info, quota_root->node);
1019 btrfs_tree_unlock(quota_root->node);
1020 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1022 free_extent_buffer(quota_root->node);
1023 free_extent_buffer(quota_root->commit_root);
1026 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1030 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1031 struct btrfs_qgroup *qgroup)
1033 if (list_empty(&qgroup->dirty))
1034 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1038 * The easy accounting, if we are adding/removing the only ref for an extent
1039 * then this qgroup and all of the parent qgroups get their refrence and
1040 * exclusive counts adjusted.
1042 * Caller should hold fs_info->qgroup_lock.
1044 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1045 struct ulist *tmp, u64 ref_root,
1046 u64 num_bytes, int sign)
1048 struct btrfs_qgroup *qgroup;
1049 struct btrfs_qgroup_list *glist;
1050 struct ulist_node *unode;
1051 struct ulist_iterator uiter;
1054 qgroup = find_qgroup_rb(fs_info, ref_root);
1058 qgroup->rfer += sign * num_bytes;
1059 qgroup->rfer_cmpr += sign * num_bytes;
1061 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1062 qgroup->excl += sign * num_bytes;
1063 qgroup->excl_cmpr += sign * num_bytes;
1065 qgroup->reserved -= num_bytes;
1067 qgroup_dirty(fs_info, qgroup);
1069 /* Get all of the parent groups that contain this qgroup */
1070 list_for_each_entry(glist, &qgroup->groups, next_group) {
1071 ret = ulist_add(tmp, glist->group->qgroupid,
1072 ptr_to_u64(glist->group), GFP_ATOMIC);
1077 /* Iterate all of the parents and adjust their reference counts */
1078 ULIST_ITER_INIT(&uiter);
1079 while ((unode = ulist_next(tmp, &uiter))) {
1080 qgroup = u64_to_ptr(unode->aux);
1081 qgroup->rfer += sign * num_bytes;
1082 qgroup->rfer_cmpr += sign * num_bytes;
1083 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1084 qgroup->excl += sign * num_bytes;
1086 qgroup->reserved -= num_bytes;
1087 qgroup->excl_cmpr += sign * num_bytes;
1088 qgroup_dirty(fs_info, qgroup);
1090 /* Add any parents of the parents */
1091 list_for_each_entry(glist, &qgroup->groups, next_group) {
1092 ret = ulist_add(tmp, glist->group->qgroupid,
1093 ptr_to_u64(glist->group), GFP_ATOMIC);
1105 * Quick path for updating qgroup with only excl refs.
1107 * In that case, just update all parent will be enough.
1108 * Or we needs to do a full rescan.
1109 * Caller should also hold fs_info->qgroup_lock.
1111 * Return 0 for quick update, return >0 for need to full rescan
1112 * and mark INCONSISTENT flag.
1113 * Return < 0 for other error.
1115 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1116 struct ulist *tmp, u64 src, u64 dst,
1119 struct btrfs_qgroup *qgroup;
1123 qgroup = find_qgroup_rb(fs_info, src);
1126 if (qgroup->excl == qgroup->rfer) {
1128 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1129 qgroup->excl, sign);
1137 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1141 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1142 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1144 struct btrfs_root *quota_root;
1145 struct btrfs_qgroup *parent;
1146 struct btrfs_qgroup *member;
1147 struct btrfs_qgroup_list *list;
1151 /* Check the level of src and dst first */
1152 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1155 tmp = ulist_alloc(GFP_NOFS);
1159 mutex_lock(&fs_info->qgroup_ioctl_lock);
1160 quota_root = fs_info->quota_root;
1165 member = find_qgroup_rb(fs_info, src);
1166 parent = find_qgroup_rb(fs_info, dst);
1167 if (!member || !parent) {
1172 /* check if such qgroup relation exist firstly */
1173 list_for_each_entry(list, &member->groups, next_group) {
1174 if (list->group == parent) {
1180 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1184 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1186 del_qgroup_relation_item(trans, quota_root, src, dst);
1190 spin_lock(&fs_info->qgroup_lock);
1191 ret = add_relation_rb(quota_root->fs_info, src, dst);
1193 spin_unlock(&fs_info->qgroup_lock);
1196 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1197 spin_unlock(&fs_info->qgroup_lock);
1199 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1204 int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1205 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1207 struct btrfs_root *quota_root;
1208 struct btrfs_qgroup *parent;
1209 struct btrfs_qgroup *member;
1210 struct btrfs_qgroup_list *list;
1215 tmp = ulist_alloc(GFP_NOFS);
1219 quota_root = fs_info->quota_root;
1225 member = find_qgroup_rb(fs_info, src);
1226 parent = find_qgroup_rb(fs_info, dst);
1227 if (!member || !parent) {
1232 /* check if such qgroup relation exist firstly */
1233 list_for_each_entry(list, &member->groups, next_group) {
1234 if (list->group == parent)
1240 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1241 err = del_qgroup_relation_item(trans, quota_root, dst, src);
1245 spin_lock(&fs_info->qgroup_lock);
1246 del_relation_rb(fs_info, src, dst);
1247 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1248 spin_unlock(&fs_info->qgroup_lock);
1254 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1255 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1259 mutex_lock(&fs_info->qgroup_ioctl_lock);
1260 ret = __del_qgroup_relation(trans, fs_info, src, dst);
1261 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1266 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1267 struct btrfs_fs_info *fs_info, u64 qgroupid)
1269 struct btrfs_root *quota_root;
1270 struct btrfs_qgroup *qgroup;
1273 mutex_lock(&fs_info->qgroup_ioctl_lock);
1274 quota_root = fs_info->quota_root;
1279 qgroup = find_qgroup_rb(fs_info, qgroupid);
1285 ret = add_qgroup_item(trans, quota_root, qgroupid);
1289 spin_lock(&fs_info->qgroup_lock);
1290 qgroup = add_qgroup_rb(fs_info, qgroupid);
1291 spin_unlock(&fs_info->qgroup_lock);
1294 ret = PTR_ERR(qgroup);
1296 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1300 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1301 struct btrfs_fs_info *fs_info, u64 qgroupid)
1303 struct btrfs_root *quota_root;
1304 struct btrfs_qgroup *qgroup;
1305 struct btrfs_qgroup_list *list;
1308 mutex_lock(&fs_info->qgroup_ioctl_lock);
1309 quota_root = fs_info->quota_root;
1315 qgroup = find_qgroup_rb(fs_info, qgroupid);
1320 /* check if there are no children of this qgroup */
1321 if (!list_empty(&qgroup->members)) {
1326 ret = del_qgroup_item(trans, quota_root, qgroupid);
1328 while (!list_empty(&qgroup->groups)) {
1329 list = list_first_entry(&qgroup->groups,
1330 struct btrfs_qgroup_list, next_group);
1331 ret = __del_qgroup_relation(trans, fs_info,
1333 list->group->qgroupid);
1338 spin_lock(&fs_info->qgroup_lock);
1339 del_qgroup_rb(quota_root->fs_info, qgroupid);
1340 spin_unlock(&fs_info->qgroup_lock);
1342 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1346 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1347 struct btrfs_fs_info *fs_info, u64 qgroupid,
1348 struct btrfs_qgroup_limit *limit)
1350 struct btrfs_root *quota_root;
1351 struct btrfs_qgroup *qgroup;
1353 /* Sometimes we would want to clear the limit on this qgroup.
1354 * To meet this requirement, we treat the -1 as a special value
1355 * which tell kernel to clear the limit on this qgroup.
1357 const u64 CLEAR_VALUE = -1;
1359 mutex_lock(&fs_info->qgroup_ioctl_lock);
1360 quota_root = fs_info->quota_root;
1366 qgroup = find_qgroup_rb(fs_info, qgroupid);
1372 spin_lock(&fs_info->qgroup_lock);
1373 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1374 if (limit->max_rfer == CLEAR_VALUE) {
1375 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1376 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1377 qgroup->max_rfer = 0;
1379 qgroup->max_rfer = limit->max_rfer;
1382 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1383 if (limit->max_excl == CLEAR_VALUE) {
1384 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1385 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1386 qgroup->max_excl = 0;
1388 qgroup->max_excl = limit->max_excl;
1391 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1392 if (limit->rsv_rfer == CLEAR_VALUE) {
1393 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1394 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1395 qgroup->rsv_rfer = 0;
1397 qgroup->rsv_rfer = limit->rsv_rfer;
1400 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1401 if (limit->rsv_excl == CLEAR_VALUE) {
1402 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1403 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1404 qgroup->rsv_excl = 0;
1406 qgroup->rsv_excl = limit->rsv_excl;
1409 qgroup->lim_flags |= limit->flags;
1411 spin_unlock(&fs_info->qgroup_lock);
1413 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1415 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1416 btrfs_info(fs_info, "unable to update quota limit for %llu",
1421 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1425 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
1426 struct btrfs_fs_info *fs_info)
1428 struct btrfs_qgroup_extent_record *record;
1429 struct btrfs_delayed_ref_root *delayed_refs;
1430 struct rb_node *node;
1434 delayed_refs = &trans->transaction->delayed_refs;
1435 qgroup_to_skip = delayed_refs->qgroup_to_skip;
1438 * No need to do lock, since this function will only be called in
1439 * btrfs_commmit_transaction().
1441 node = rb_first(&delayed_refs->dirty_extent_root);
1443 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1445 ret = btrfs_find_all_roots(NULL, fs_info, record->bytenr, 0,
1446 &record->old_roots);
1450 ulist_del(record->old_roots, qgroup_to_skip, 0);
1451 node = rb_next(node);
1456 struct btrfs_qgroup_extent_record
1457 *btrfs_qgroup_insert_dirty_extent(struct btrfs_delayed_ref_root *delayed_refs,
1458 struct btrfs_qgroup_extent_record *record)
1460 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1461 struct rb_node *parent_node = NULL;
1462 struct btrfs_qgroup_extent_record *entry;
1463 u64 bytenr = record->bytenr;
1465 assert_spin_locked(&delayed_refs->lock);
1469 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1471 if (bytenr < entry->bytenr)
1473 else if (bytenr > entry->bytenr)
1474 p = &(*p)->rb_right;
1479 rb_link_node(&record->node, parent_node, p);
1480 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1484 #define UPDATE_NEW 0
1485 #define UPDATE_OLD 1
1487 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1489 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1490 struct ulist *roots, struct ulist *tmp,
1491 struct ulist *qgroups, u64 seq, int update_old)
1493 struct ulist_node *unode;
1494 struct ulist_iterator uiter;
1495 struct ulist_node *tmp_unode;
1496 struct ulist_iterator tmp_uiter;
1497 struct btrfs_qgroup *qg;
1502 ULIST_ITER_INIT(&uiter);
1503 while ((unode = ulist_next(roots, &uiter))) {
1504 qg = find_qgroup_rb(fs_info, unode->val);
1509 ret = ulist_add(qgroups, qg->qgroupid, ptr_to_u64(qg),
1513 ret = ulist_add(tmp, qg->qgroupid, ptr_to_u64(qg), GFP_ATOMIC);
1516 ULIST_ITER_INIT(&tmp_uiter);
1517 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1518 struct btrfs_qgroup_list *glist;
1520 qg = u64_to_ptr(tmp_unode->aux);
1522 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1524 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1525 list_for_each_entry(glist, &qg->groups, next_group) {
1526 ret = ulist_add(qgroups, glist->group->qgroupid,
1527 ptr_to_u64(glist->group),
1531 ret = ulist_add(tmp, glist->group->qgroupid,
1532 ptr_to_u64(glist->group),
1543 * Update qgroup rfer/excl counters.
1544 * Rfer update is easy, codes can explain themselves.
1546 * Excl update is tricky, the update is split into 2 part.
1547 * Part 1: Possible exclusive <-> sharing detect:
1549 * -------------------------------------
1551 * -------------------------------------
1553 * -------------------------------------
1556 * A: cur_old_roots < nr_old_roots (not exclusive before)
1557 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1558 * B: cur_new_roots < nr_new_roots (not exclusive now)
1559 * !B: cur_new_roots == nr_new_roots (possible exclsuive now)
1562 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1563 * *: Definitely not changed. **: Possible unchanged.
1565 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1567 * To make the logic clear, we first use condition A and B to split
1568 * combination into 4 results.
1570 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1571 * only on variant maybe 0.
1573 * Lastly, check result **, since there are 2 variants maybe 0, split them
1575 * But this time we don't need to consider other things, the codes and logic
1576 * is easy to understand now.
1578 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1579 struct ulist *qgroups,
1582 u64 num_bytes, u64 seq)
1584 struct ulist_node *unode;
1585 struct ulist_iterator uiter;
1586 struct btrfs_qgroup *qg;
1587 u64 cur_new_count, cur_old_count;
1589 ULIST_ITER_INIT(&uiter);
1590 while ((unode = ulist_next(qgroups, &uiter))) {
1593 qg = u64_to_ptr(unode->aux);
1594 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1595 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1597 /* Rfer update part */
1598 if (cur_old_count == 0 && cur_new_count > 0) {
1599 qg->rfer += num_bytes;
1600 qg->rfer_cmpr += num_bytes;
1603 if (cur_old_count > 0 && cur_new_count == 0) {
1604 qg->rfer -= num_bytes;
1605 qg->rfer_cmpr -= num_bytes;
1609 /* Excl update part */
1610 /* Exclusive/none -> shared case */
1611 if (cur_old_count == nr_old_roots &&
1612 cur_new_count < nr_new_roots) {
1613 /* Exclusive -> shared */
1614 if (cur_old_count != 0) {
1615 qg->excl -= num_bytes;
1616 qg->excl_cmpr -= num_bytes;
1621 /* Shared -> exclusive/none case */
1622 if (cur_old_count < nr_old_roots &&
1623 cur_new_count == nr_new_roots) {
1624 /* Shared->exclusive */
1625 if (cur_new_count != 0) {
1626 qg->excl += num_bytes;
1627 qg->excl_cmpr += num_bytes;
1632 /* Exclusive/none -> exclusive/none case */
1633 if (cur_old_count == nr_old_roots &&
1634 cur_new_count == nr_new_roots) {
1635 if (cur_old_count == 0) {
1636 /* None -> exclusive/none */
1638 if (cur_new_count != 0) {
1639 /* None -> exclusive */
1640 qg->excl += num_bytes;
1641 qg->excl_cmpr += num_bytes;
1644 /* None -> none, nothing changed */
1646 /* Exclusive -> exclusive/none */
1648 if (cur_new_count == 0) {
1649 /* Exclusive -> none */
1650 qg->excl -= num_bytes;
1651 qg->excl_cmpr -= num_bytes;
1654 /* Exclusive -> exclusive, nothing changed */
1659 qgroup_dirty(fs_info, qg);
1665 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1666 struct btrfs_fs_info *fs_info,
1667 u64 bytenr, u64 num_bytes,
1668 struct ulist *old_roots, struct ulist *new_roots)
1670 struct ulist *qgroups = NULL;
1671 struct ulist *tmp = NULL;
1673 u64 nr_new_roots = 0;
1674 u64 nr_old_roots = 0;
1678 nr_new_roots = new_roots->nnodes;
1680 nr_old_roots = old_roots->nnodes;
1682 if (!fs_info->quota_enabled)
1684 BUG_ON(!fs_info->quota_root);
1686 qgroups = ulist_alloc(GFP_NOFS);
1691 tmp = ulist_alloc(GFP_NOFS);
1697 mutex_lock(&fs_info->qgroup_rescan_lock);
1698 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1699 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1700 mutex_unlock(&fs_info->qgroup_rescan_lock);
1705 mutex_unlock(&fs_info->qgroup_rescan_lock);
1707 spin_lock(&fs_info->qgroup_lock);
1708 seq = fs_info->qgroup_seq;
1710 /* Update old refcnts using old_roots */
1711 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1716 /* Update new refcnts using new_roots */
1717 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1722 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1726 * Bump qgroup_seq to avoid seq overlap
1728 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
1730 spin_unlock(&fs_info->qgroup_lock);
1733 ulist_free(qgroups);
1734 ulist_free(old_roots);
1735 ulist_free(new_roots);
1739 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
1740 struct btrfs_fs_info *fs_info)
1742 struct btrfs_qgroup_extent_record *record;
1743 struct btrfs_delayed_ref_root *delayed_refs;
1744 struct ulist *new_roots = NULL;
1745 struct rb_node *node;
1749 delayed_refs = &trans->transaction->delayed_refs;
1750 qgroup_to_skip = delayed_refs->qgroup_to_skip;
1751 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
1752 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1757 * Use (u64)-1 as time_seq to do special search, which
1758 * doesn't lock tree or delayed_refs and search current
1759 * root. It's safe inside commit_transaction().
1761 ret = btrfs_find_all_roots(trans, fs_info,
1762 record->bytenr, (u64)-1, &new_roots);
1766 ulist_del(new_roots, qgroup_to_skip, 0);
1767 ret = btrfs_qgroup_account_extent(trans, fs_info,
1768 record->bytenr, record->num_bytes,
1769 record->old_roots, new_roots);
1770 record->old_roots = NULL;
1774 ulist_free(record->old_roots);
1775 ulist_free(new_roots);
1777 rb_erase(node, &delayed_refs->dirty_extent_root);
1785 * called from commit_transaction. Writes all changed qgroups to disk.
1787 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1788 struct btrfs_fs_info *fs_info)
1790 struct btrfs_root *quota_root = fs_info->quota_root;
1792 int start_rescan_worker = 0;
1797 if (!fs_info->quota_enabled && fs_info->pending_quota_state)
1798 start_rescan_worker = 1;
1800 fs_info->quota_enabled = fs_info->pending_quota_state;
1802 spin_lock(&fs_info->qgroup_lock);
1803 while (!list_empty(&fs_info->dirty_qgroups)) {
1804 struct btrfs_qgroup *qgroup;
1805 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1806 struct btrfs_qgroup, dirty);
1807 list_del_init(&qgroup->dirty);
1808 spin_unlock(&fs_info->qgroup_lock);
1809 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1811 fs_info->qgroup_flags |=
1812 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1813 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1815 fs_info->qgroup_flags |=
1816 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1817 spin_lock(&fs_info->qgroup_lock);
1819 if (fs_info->quota_enabled)
1820 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1822 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1823 spin_unlock(&fs_info->qgroup_lock);
1825 ret = update_qgroup_status_item(trans, fs_info, quota_root);
1827 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1829 if (!ret && start_rescan_worker) {
1830 ret = qgroup_rescan_init(fs_info, 0, 1);
1832 qgroup_rescan_zero_tracking(fs_info);
1833 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1834 &fs_info->qgroup_rescan_work);
1845 * copy the acounting information between qgroups. This is necessary when a
1846 * snapshot or a subvolume is created
1848 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1849 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1850 struct btrfs_qgroup_inherit *inherit)
1855 struct btrfs_root *quota_root = fs_info->quota_root;
1856 struct btrfs_qgroup *srcgroup;
1857 struct btrfs_qgroup *dstgroup;
1861 mutex_lock(&fs_info->qgroup_ioctl_lock);
1862 if (!fs_info->quota_enabled)
1871 i_qgroups = (u64 *)(inherit + 1);
1872 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1873 2 * inherit->num_excl_copies;
1874 for (i = 0; i < nums; ++i) {
1875 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
1881 if ((srcgroup->qgroupid >> 48) <= (objectid >> 48)) {
1890 * create a tracking group for the subvol itself
1892 ret = add_qgroup_item(trans, quota_root, objectid);
1897 struct btrfs_root *srcroot;
1898 struct btrfs_key srckey;
1900 srckey.objectid = srcid;
1901 srckey.type = BTRFS_ROOT_ITEM_KEY;
1902 srckey.offset = (u64)-1;
1903 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1904 if (IS_ERR(srcroot)) {
1905 ret = PTR_ERR(srcroot);
1910 level_size = srcroot->nodesize;
1915 * add qgroup to all inherited groups
1918 i_qgroups = (u64 *)(inherit + 1);
1919 for (i = 0; i < inherit->num_qgroups; ++i) {
1920 ret = add_qgroup_relation_item(trans, quota_root,
1921 objectid, *i_qgroups);
1924 ret = add_qgroup_relation_item(trans, quota_root,
1925 *i_qgroups, objectid);
1933 spin_lock(&fs_info->qgroup_lock);
1935 dstgroup = add_qgroup_rb(fs_info, objectid);
1936 if (IS_ERR(dstgroup)) {
1937 ret = PTR_ERR(dstgroup);
1941 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1942 dstgroup->lim_flags = inherit->lim.flags;
1943 dstgroup->max_rfer = inherit->lim.max_rfer;
1944 dstgroup->max_excl = inherit->lim.max_excl;
1945 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
1946 dstgroup->rsv_excl = inherit->lim.rsv_excl;
1948 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
1950 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1951 btrfs_info(fs_info, "unable to update quota limit for %llu",
1952 dstgroup->qgroupid);
1958 srcgroup = find_qgroup_rb(fs_info, srcid);
1963 * We call inherit after we clone the root in order to make sure
1964 * our counts don't go crazy, so at this point the only
1965 * difference between the two roots should be the root node.
1967 dstgroup->rfer = srcgroup->rfer;
1968 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
1969 dstgroup->excl = level_size;
1970 dstgroup->excl_cmpr = level_size;
1971 srcgroup->excl = level_size;
1972 srcgroup->excl_cmpr = level_size;
1974 /* inherit the limit info */
1975 dstgroup->lim_flags = srcgroup->lim_flags;
1976 dstgroup->max_rfer = srcgroup->max_rfer;
1977 dstgroup->max_excl = srcgroup->max_excl;
1978 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
1979 dstgroup->rsv_excl = srcgroup->rsv_excl;
1981 qgroup_dirty(fs_info, dstgroup);
1982 qgroup_dirty(fs_info, srcgroup);
1988 i_qgroups = (u64 *)(inherit + 1);
1989 for (i = 0; i < inherit->num_qgroups; ++i) {
1990 ret = add_relation_rb(quota_root->fs_info, objectid,
1997 for (i = 0; i < inherit->num_ref_copies; ++i) {
1998 struct btrfs_qgroup *src;
1999 struct btrfs_qgroup *dst;
2001 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2002 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2009 dst->rfer = src->rfer - level_size;
2010 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2013 for (i = 0; i < inherit->num_excl_copies; ++i) {
2014 struct btrfs_qgroup *src;
2015 struct btrfs_qgroup *dst;
2017 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2018 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2025 dst->excl = src->excl + level_size;
2026 dst->excl_cmpr = src->excl_cmpr + level_size;
2031 spin_unlock(&fs_info->qgroup_lock);
2033 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2037 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
2039 struct btrfs_root *quota_root;
2040 struct btrfs_qgroup *qgroup;
2041 struct btrfs_fs_info *fs_info = root->fs_info;
2042 u64 ref_root = root->root_key.objectid;
2044 struct ulist_node *unode;
2045 struct ulist_iterator uiter;
2047 if (!is_fstree(ref_root))
2053 spin_lock(&fs_info->qgroup_lock);
2054 quota_root = fs_info->quota_root;
2058 qgroup = find_qgroup_rb(fs_info, ref_root);
2063 * in a first step, we check all affected qgroups if any limits would
2066 ulist_reinit(fs_info->qgroup_ulist);
2067 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2068 (uintptr_t)qgroup, GFP_ATOMIC);
2071 ULIST_ITER_INIT(&uiter);
2072 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2073 struct btrfs_qgroup *qg;
2074 struct btrfs_qgroup_list *glist;
2076 qg = u64_to_ptr(unode->aux);
2078 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2079 qg->reserved + (s64)qg->rfer + num_bytes >
2085 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2086 qg->reserved + (s64)qg->excl + num_bytes >
2092 list_for_each_entry(glist, &qg->groups, next_group) {
2093 ret = ulist_add(fs_info->qgroup_ulist,
2094 glist->group->qgroupid,
2095 (uintptr_t)glist->group, GFP_ATOMIC);
2102 * no limits exceeded, now record the reservation into all qgroups
2104 ULIST_ITER_INIT(&uiter);
2105 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2106 struct btrfs_qgroup *qg;
2108 qg = u64_to_ptr(unode->aux);
2110 qg->reserved += num_bytes;
2114 spin_unlock(&fs_info->qgroup_lock);
2118 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2119 u64 ref_root, u64 num_bytes)
2121 struct btrfs_root *quota_root;
2122 struct btrfs_qgroup *qgroup;
2123 struct ulist_node *unode;
2124 struct ulist_iterator uiter;
2127 if (!is_fstree(ref_root))
2133 spin_lock(&fs_info->qgroup_lock);
2135 quota_root = fs_info->quota_root;
2139 qgroup = find_qgroup_rb(fs_info, ref_root);
2143 ulist_reinit(fs_info->qgroup_ulist);
2144 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2145 (uintptr_t)qgroup, GFP_ATOMIC);
2148 ULIST_ITER_INIT(&uiter);
2149 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2150 struct btrfs_qgroup *qg;
2151 struct btrfs_qgroup_list *glist;
2153 qg = u64_to_ptr(unode->aux);
2155 qg->reserved -= num_bytes;
2157 list_for_each_entry(glist, &qg->groups, next_group) {
2158 ret = ulist_add(fs_info->qgroup_ulist,
2159 glist->group->qgroupid,
2160 (uintptr_t)glist->group, GFP_ATOMIC);
2167 spin_unlock(&fs_info->qgroup_lock);
2170 static inline void qgroup_free(struct btrfs_root *root, u64 num_bytes)
2172 return btrfs_qgroup_free_refroot(root->fs_info, root->objectid,
2175 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2177 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2179 btrfs_err(trans->root->fs_info,
2180 "qgroups not uptodate in trans handle %p: list is%s empty, "
2182 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
2183 (u32)(trans->delayed_ref_elem.seq >> 32),
2184 (u32)trans->delayed_ref_elem.seq);
2189 * Check if the leaf is the last leaf. Which means all node pointers
2190 * are at their last position.
2192 static bool is_last_leaf(struct btrfs_path *path)
2196 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2197 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
2204 * returns < 0 on error, 0 when more leafs are to be scanned.
2205 * returns 1 when done.
2208 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2209 struct btrfs_trans_handle *trans)
2211 struct btrfs_key found;
2212 struct extent_buffer *scratch_leaf = NULL;
2213 struct ulist *roots = NULL;
2214 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2220 mutex_lock(&fs_info->qgroup_rescan_lock);
2221 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2222 &fs_info->qgroup_rescan_progress,
2225 pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
2226 fs_info->qgroup_rescan_progress.objectid,
2227 fs_info->qgroup_rescan_progress.type,
2228 fs_info->qgroup_rescan_progress.offset, ret);
2232 * The rescan is about to end, we will not be scanning any
2233 * further blocks. We cannot unset the RESCAN flag here, because
2234 * we want to commit the transaction if everything went well.
2235 * To make the live accounting work in this phase, we set our
2236 * scan progress pointer such that every real extent objectid
2239 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2240 btrfs_release_path(path);
2241 mutex_unlock(&fs_info->qgroup_rescan_lock);
2244 done = is_last_leaf(path);
2246 btrfs_item_key_to_cpu(path->nodes[0], &found,
2247 btrfs_header_nritems(path->nodes[0]) - 1);
2248 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2250 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2251 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2252 if (!scratch_leaf) {
2254 mutex_unlock(&fs_info->qgroup_rescan_lock);
2257 extent_buffer_get(scratch_leaf);
2258 btrfs_tree_read_lock(scratch_leaf);
2259 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2260 slot = path->slots[0];
2261 btrfs_release_path(path);
2262 mutex_unlock(&fs_info->qgroup_rescan_lock);
2264 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2265 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2266 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2267 found.type != BTRFS_METADATA_ITEM_KEY)
2269 if (found.type == BTRFS_METADATA_ITEM_KEY)
2270 num_bytes = fs_info->extent_root->nodesize;
2272 num_bytes = found.offset;
2274 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2278 /* For rescan, just pass old_roots as NULL */
2279 ret = btrfs_qgroup_account_extent(trans, fs_info,
2280 found.objectid, num_bytes, NULL, roots);
2286 btrfs_tree_read_unlock_blocking(scratch_leaf);
2287 free_extent_buffer(scratch_leaf);
2289 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2296 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2298 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2299 qgroup_rescan_work);
2300 struct btrfs_path *path;
2301 struct btrfs_trans_handle *trans = NULL;
2305 path = btrfs_alloc_path();
2310 while (!err && !btrfs_fs_closing(fs_info)) {
2311 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2312 if (IS_ERR(trans)) {
2313 err = PTR_ERR(trans);
2316 if (!fs_info->quota_enabled) {
2319 err = qgroup_rescan_leaf(fs_info, path, trans);
2322 btrfs_commit_transaction(trans, fs_info->fs_root);
2324 btrfs_end_transaction(trans, fs_info->fs_root);
2328 btrfs_free_path(path);
2330 mutex_lock(&fs_info->qgroup_rescan_lock);
2331 if (!btrfs_fs_closing(fs_info))
2332 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2335 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2336 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2337 } else if (err < 0) {
2338 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2340 mutex_unlock(&fs_info->qgroup_rescan_lock);
2343 * only update status, since the previous part has alreay updated the
2346 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2347 if (IS_ERR(trans)) {
2348 err = PTR_ERR(trans);
2350 "fail to start transaction for status update: %d\n",
2354 ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2357 btrfs_err(fs_info, "fail to update qgroup status: %d\n", err);
2359 btrfs_end_transaction(trans, fs_info->quota_root);
2361 if (btrfs_fs_closing(fs_info)) {
2362 btrfs_info(fs_info, "qgroup scan paused");
2363 } else if (err >= 0) {
2364 btrfs_info(fs_info, "qgroup scan completed%s",
2365 err > 0 ? " (inconsistency flag cleared)" : "");
2367 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2371 mutex_lock(&fs_info->qgroup_rescan_lock);
2372 fs_info->qgroup_rescan_running = false;
2373 mutex_unlock(&fs_info->qgroup_rescan_lock);
2374 complete_all(&fs_info->qgroup_rescan_completion);
2378 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2379 * memory required for the rescan context.
2382 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2388 (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2389 !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2394 mutex_lock(&fs_info->qgroup_rescan_lock);
2395 spin_lock(&fs_info->qgroup_lock);
2398 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2400 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2404 spin_unlock(&fs_info->qgroup_lock);
2405 mutex_unlock(&fs_info->qgroup_rescan_lock);
2408 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2411 memset(&fs_info->qgroup_rescan_progress, 0,
2412 sizeof(fs_info->qgroup_rescan_progress));
2413 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2414 init_completion(&fs_info->qgroup_rescan_completion);
2415 fs_info->qgroup_rescan_running = true;
2417 spin_unlock(&fs_info->qgroup_lock);
2418 mutex_unlock(&fs_info->qgroup_rescan_lock);
2420 memset(&fs_info->qgroup_rescan_work, 0,
2421 sizeof(fs_info->qgroup_rescan_work));
2422 btrfs_init_work(&fs_info->qgroup_rescan_work,
2423 btrfs_qgroup_rescan_helper,
2424 btrfs_qgroup_rescan_worker, NULL, NULL);
2428 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2436 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2439 struct btrfs_qgroup *qgroup;
2441 spin_lock(&fs_info->qgroup_lock);
2442 /* clear all current qgroup tracking information */
2443 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2444 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2446 qgroup->rfer_cmpr = 0;
2448 qgroup->excl_cmpr = 0;
2449 qgroup_dirty(fs_info, qgroup);
2451 spin_unlock(&fs_info->qgroup_lock);
2455 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2458 struct btrfs_trans_handle *trans;
2460 ret = qgroup_rescan_init(fs_info, 0, 1);
2465 * We have set the rescan_progress to 0, which means no more
2466 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2467 * However, btrfs_qgroup_account_ref may be right after its call
2468 * to btrfs_find_all_roots, in which case it would still do the
2470 * To solve this, we're committing the transaction, which will
2471 * ensure we run all delayed refs and only after that, we are
2472 * going to clear all tracking information for a clean start.
2475 trans = btrfs_join_transaction(fs_info->fs_root);
2476 if (IS_ERR(trans)) {
2477 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2478 return PTR_ERR(trans);
2480 ret = btrfs_commit_transaction(trans, fs_info->fs_root);
2482 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2486 qgroup_rescan_zero_tracking(fs_info);
2488 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2489 &fs_info->qgroup_rescan_work);
2494 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2500 mutex_lock(&fs_info->qgroup_rescan_lock);
2501 spin_lock(&fs_info->qgroup_lock);
2502 running = fs_info->qgroup_rescan_running;
2503 spin_unlock(&fs_info->qgroup_lock);
2504 mutex_unlock(&fs_info->qgroup_rescan_lock);
2510 ret = wait_for_completion_interruptible(
2511 &fs_info->qgroup_rescan_completion);
2513 wait_for_completion(&fs_info->qgroup_rescan_completion);
2519 * this is only called from open_ctree where we're still single threaded, thus
2520 * locking is omitted here.
2523 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2525 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2526 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2527 &fs_info->qgroup_rescan_work);
2531 * Reserve qgroup space for range [start, start + len).
2533 * This function will either reserve space from related qgroups or doing
2534 * nothing if the range is already reserved.
2536 * Return 0 for successful reserve
2537 * Return <0 for error (including -EQUOT)
2539 * NOTE: this function may sleep for memory allocation.
2541 int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len)
2543 struct btrfs_root *root = BTRFS_I(inode)->root;
2544 struct extent_changeset changeset;
2545 struct ulist_node *unode;
2546 struct ulist_iterator uiter;
2549 if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) ||
2553 changeset.bytes_changed = 0;
2554 changeset.range_changed = ulist_alloc(GFP_NOFS);
2555 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2556 start + len -1, EXTENT_QGROUP_RESERVED, GFP_NOFS,
2558 trace_btrfs_qgroup_reserve_data(inode, start, len,
2559 changeset.bytes_changed,
2563 ret = qgroup_reserve(root, changeset.bytes_changed);
2567 ulist_free(changeset.range_changed);
2571 /* cleanup already reserved ranges */
2572 ULIST_ITER_INIT(&uiter);
2573 while ((unode = ulist_next(changeset.range_changed, &uiter)))
2574 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2575 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2577 ulist_free(changeset.range_changed);
2581 static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len,
2584 struct extent_changeset changeset;
2585 int trace_op = QGROUP_RELEASE;
2588 changeset.bytes_changed = 0;
2589 changeset.range_changed = ulist_alloc(GFP_NOFS);
2590 if (!changeset.range_changed)
2593 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2594 start + len -1, EXTENT_QGROUP_RESERVED, GFP_NOFS,
2600 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2601 trace_op = QGROUP_FREE;
2603 trace_btrfs_qgroup_release_data(inode, start, len,
2604 changeset.bytes_changed, trace_op);
2606 ulist_free(changeset.range_changed);
2611 * Free a reserved space range from io_tree and related qgroups
2613 * Should be called when a range of pages get invalidated before reaching disk.
2614 * Or for error cleanup case.
2616 * For data written to disk, use btrfs_qgroup_release_data().
2618 * NOTE: This function may sleep for memory allocation.
2620 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len)
2622 return __btrfs_qgroup_release_data(inode, start, len, 1);
2626 * Release a reserved space range from io_tree only.
2628 * Should be called when a range of pages get written to disk and corresponding
2629 * FILE_EXTENT is inserted into corresponding root.
2631 * Since new qgroup accounting framework will only update qgroup numbers at
2632 * commit_transaction() time, its reserved space shouldn't be freed from
2635 * But we should release the range from io_tree, to allow further write to be
2638 * NOTE: This function may sleep for memory allocation.
2640 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
2642 return __btrfs_qgroup_release_data(inode, start, len, 0);
2645 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes)
2649 if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) ||
2653 BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
2654 ret = qgroup_reserve(root, num_bytes);
2657 atomic_add(num_bytes, &root->qgroup_meta_rsv);
2661 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
2665 if (!root->fs_info->quota_enabled || !is_fstree(root->objectid))
2668 reserved = atomic_xchg(&root->qgroup_meta_rsv, 0);
2671 qgroup_free(root, reserved);
2674 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
2676 if (!root->fs_info->quota_enabled || !is_fstree(root->objectid))
2679 BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
2680 WARN_ON(atomic_read(&root->qgroup_meta_rsv) < num_bytes);
2681 atomic_sub(num_bytes, &root->qgroup_meta_rsv);
2682 qgroup_free(root, num_bytes);
2686 * Check qgroup reserved space leaking, normally at destory inode
2689 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
2691 struct extent_changeset changeset;
2692 struct ulist_node *unode;
2693 struct ulist_iterator iter;
2696 changeset.bytes_changed = 0;
2697 changeset.range_changed = ulist_alloc(GFP_NOFS);
2698 if (WARN_ON(!changeset.range_changed))
2701 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
2702 EXTENT_QGROUP_RESERVED, GFP_NOFS, &changeset);
2705 if (WARN_ON(changeset.bytes_changed)) {
2706 ULIST_ITER_INIT(&iter);
2707 while ((unode = ulist_next(changeset.range_changed, &iter))) {
2708 btrfs_warn(BTRFS_I(inode)->root->fs_info,
2709 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
2710 inode->i_ino, unode->val, unode->aux);
2712 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2714 ulist_free(changeset.range_changed);