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 #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_ulist
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_is_testing(quota_root->fs_info))
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_is_testing(root->fs_info))
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 reference 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_commit_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 int btrfs_qgroup_insert_dirty_extent_nolock(struct btrfs_fs_info *fs_info,
1457 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);
1466 trace_btrfs_qgroup_insert_dirty_extent(fs_info, record);
1470 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1472 if (bytenr < entry->bytenr)
1474 else if (bytenr > entry->bytenr)
1475 p = &(*p)->rb_right;
1480 rb_link_node(&record->node, parent_node, p);
1481 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1485 int btrfs_qgroup_insert_dirty_extent(struct btrfs_trans_handle *trans,
1486 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1489 struct btrfs_qgroup_extent_record *record;
1490 struct btrfs_delayed_ref_root *delayed_refs;
1493 if (!fs_info->quota_enabled || bytenr == 0 || num_bytes == 0)
1495 if (WARN_ON(trans == NULL))
1497 record = kmalloc(sizeof(*record), gfp_flag);
1501 delayed_refs = &trans->transaction->delayed_refs;
1502 record->bytenr = bytenr;
1503 record->num_bytes = num_bytes;
1504 record->old_roots = NULL;
1506 spin_lock(&delayed_refs->lock);
1507 ret = btrfs_qgroup_insert_dirty_extent_nolock(fs_info, delayed_refs,
1509 spin_unlock(&delayed_refs->lock);
1515 #define UPDATE_NEW 0
1516 #define UPDATE_OLD 1
1518 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1520 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1521 struct ulist *roots, struct ulist *tmp,
1522 struct ulist *qgroups, u64 seq, int update_old)
1524 struct ulist_node *unode;
1525 struct ulist_iterator uiter;
1526 struct ulist_node *tmp_unode;
1527 struct ulist_iterator tmp_uiter;
1528 struct btrfs_qgroup *qg;
1533 ULIST_ITER_INIT(&uiter);
1534 while ((unode = ulist_next(roots, &uiter))) {
1535 qg = find_qgroup_rb(fs_info, unode->val);
1540 ret = ulist_add(qgroups, qg->qgroupid, ptr_to_u64(qg),
1544 ret = ulist_add(tmp, qg->qgroupid, ptr_to_u64(qg), GFP_ATOMIC);
1547 ULIST_ITER_INIT(&tmp_uiter);
1548 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1549 struct btrfs_qgroup_list *glist;
1551 qg = u64_to_ptr(tmp_unode->aux);
1553 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1555 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1556 list_for_each_entry(glist, &qg->groups, next_group) {
1557 ret = ulist_add(qgroups, glist->group->qgroupid,
1558 ptr_to_u64(glist->group),
1562 ret = ulist_add(tmp, glist->group->qgroupid,
1563 ptr_to_u64(glist->group),
1574 * Update qgroup rfer/excl counters.
1575 * Rfer update is easy, codes can explain themselves.
1577 * Excl update is tricky, the update is split into 2 part.
1578 * Part 1: Possible exclusive <-> sharing detect:
1580 * -------------------------------------
1582 * -------------------------------------
1584 * -------------------------------------
1587 * A: cur_old_roots < nr_old_roots (not exclusive before)
1588 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1589 * B: cur_new_roots < nr_new_roots (not exclusive now)
1590 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
1593 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1594 * *: Definitely not changed. **: Possible unchanged.
1596 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1598 * To make the logic clear, we first use condition A and B to split
1599 * combination into 4 results.
1601 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1602 * only on variant maybe 0.
1604 * Lastly, check result **, since there are 2 variants maybe 0, split them
1606 * But this time we don't need to consider other things, the codes and logic
1607 * is easy to understand now.
1609 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1610 struct ulist *qgroups,
1613 u64 num_bytes, u64 seq)
1615 struct ulist_node *unode;
1616 struct ulist_iterator uiter;
1617 struct btrfs_qgroup *qg;
1618 u64 cur_new_count, cur_old_count;
1620 ULIST_ITER_INIT(&uiter);
1621 while ((unode = ulist_next(qgroups, &uiter))) {
1624 qg = u64_to_ptr(unode->aux);
1625 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1626 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1628 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1629 cur_old_count, cur_new_count);
1631 /* Rfer update part */
1632 if (cur_old_count == 0 && cur_new_count > 0) {
1633 qg->rfer += num_bytes;
1634 qg->rfer_cmpr += num_bytes;
1637 if (cur_old_count > 0 && cur_new_count == 0) {
1638 qg->rfer -= num_bytes;
1639 qg->rfer_cmpr -= num_bytes;
1643 /* Excl update part */
1644 /* Exclusive/none -> shared case */
1645 if (cur_old_count == nr_old_roots &&
1646 cur_new_count < nr_new_roots) {
1647 /* Exclusive -> shared */
1648 if (cur_old_count != 0) {
1649 qg->excl -= num_bytes;
1650 qg->excl_cmpr -= num_bytes;
1655 /* Shared -> exclusive/none case */
1656 if (cur_old_count < nr_old_roots &&
1657 cur_new_count == nr_new_roots) {
1658 /* Shared->exclusive */
1659 if (cur_new_count != 0) {
1660 qg->excl += num_bytes;
1661 qg->excl_cmpr += num_bytes;
1666 /* Exclusive/none -> exclusive/none case */
1667 if (cur_old_count == nr_old_roots &&
1668 cur_new_count == nr_new_roots) {
1669 if (cur_old_count == 0) {
1670 /* None -> exclusive/none */
1672 if (cur_new_count != 0) {
1673 /* None -> exclusive */
1674 qg->excl += num_bytes;
1675 qg->excl_cmpr += num_bytes;
1678 /* None -> none, nothing changed */
1680 /* Exclusive -> exclusive/none */
1682 if (cur_new_count == 0) {
1683 /* Exclusive -> none */
1684 qg->excl -= num_bytes;
1685 qg->excl_cmpr -= num_bytes;
1688 /* Exclusive -> exclusive, nothing changed */
1693 qgroup_dirty(fs_info, qg);
1699 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1700 struct btrfs_fs_info *fs_info,
1701 u64 bytenr, u64 num_bytes,
1702 struct ulist *old_roots, struct ulist *new_roots)
1704 struct ulist *qgroups = NULL;
1705 struct ulist *tmp = NULL;
1707 u64 nr_new_roots = 0;
1708 u64 nr_old_roots = 0;
1712 nr_new_roots = new_roots->nnodes;
1714 nr_old_roots = old_roots->nnodes;
1716 if (!fs_info->quota_enabled)
1718 BUG_ON(!fs_info->quota_root);
1720 trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1721 nr_old_roots, nr_new_roots);
1723 qgroups = ulist_alloc(GFP_NOFS);
1728 tmp = ulist_alloc(GFP_NOFS);
1734 mutex_lock(&fs_info->qgroup_rescan_lock);
1735 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1736 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1737 mutex_unlock(&fs_info->qgroup_rescan_lock);
1742 mutex_unlock(&fs_info->qgroup_rescan_lock);
1744 spin_lock(&fs_info->qgroup_lock);
1745 seq = fs_info->qgroup_seq;
1747 /* Update old refcnts using old_roots */
1748 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1753 /* Update new refcnts using new_roots */
1754 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
1759 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
1763 * Bump qgroup_seq to avoid seq overlap
1765 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
1767 spin_unlock(&fs_info->qgroup_lock);
1770 ulist_free(qgroups);
1771 ulist_free(old_roots);
1772 ulist_free(new_roots);
1776 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans,
1777 struct btrfs_fs_info *fs_info)
1779 struct btrfs_qgroup_extent_record *record;
1780 struct btrfs_delayed_ref_root *delayed_refs;
1781 struct ulist *new_roots = NULL;
1782 struct rb_node *node;
1786 delayed_refs = &trans->transaction->delayed_refs;
1787 qgroup_to_skip = delayed_refs->qgroup_to_skip;
1788 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
1789 record = rb_entry(node, struct btrfs_qgroup_extent_record,
1792 trace_btrfs_qgroup_account_extents(fs_info, record);
1796 * Use (u64)-1 as time_seq to do special search, which
1797 * doesn't lock tree or delayed_refs and search current
1798 * root. It's safe inside commit_transaction().
1800 ret = btrfs_find_all_roots(trans, fs_info,
1801 record->bytenr, (u64)-1, &new_roots);
1805 ulist_del(new_roots, qgroup_to_skip, 0);
1806 ret = btrfs_qgroup_account_extent(trans, fs_info,
1807 record->bytenr, record->num_bytes,
1808 record->old_roots, new_roots);
1809 record->old_roots = NULL;
1813 ulist_free(record->old_roots);
1814 ulist_free(new_roots);
1816 rb_erase(node, &delayed_refs->dirty_extent_root);
1824 * called from commit_transaction. Writes all changed qgroups to disk.
1826 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
1827 struct btrfs_fs_info *fs_info)
1829 struct btrfs_root *quota_root = fs_info->quota_root;
1831 int start_rescan_worker = 0;
1836 if (!fs_info->quota_enabled && fs_info->pending_quota_state)
1837 start_rescan_worker = 1;
1839 fs_info->quota_enabled = fs_info->pending_quota_state;
1841 spin_lock(&fs_info->qgroup_lock);
1842 while (!list_empty(&fs_info->dirty_qgroups)) {
1843 struct btrfs_qgroup *qgroup;
1844 qgroup = list_first_entry(&fs_info->dirty_qgroups,
1845 struct btrfs_qgroup, dirty);
1846 list_del_init(&qgroup->dirty);
1847 spin_unlock(&fs_info->qgroup_lock);
1848 ret = update_qgroup_info_item(trans, quota_root, qgroup);
1850 fs_info->qgroup_flags |=
1851 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1852 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1854 fs_info->qgroup_flags |=
1855 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1856 spin_lock(&fs_info->qgroup_lock);
1858 if (fs_info->quota_enabled)
1859 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
1861 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1862 spin_unlock(&fs_info->qgroup_lock);
1864 ret = update_qgroup_status_item(trans, fs_info, quota_root);
1866 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1868 if (!ret && start_rescan_worker) {
1869 ret = qgroup_rescan_init(fs_info, 0, 1);
1871 qgroup_rescan_zero_tracking(fs_info);
1872 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1873 &fs_info->qgroup_rescan_work);
1884 * Copy the accounting information between qgroups. This is necessary
1885 * when a snapshot or a subvolume is created. Throwing an error will
1886 * cause a transaction abort so we take extra care here to only error
1887 * when a readonly fs is a reasonable outcome.
1889 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
1890 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
1891 struct btrfs_qgroup_inherit *inherit)
1896 struct btrfs_root *quota_root = fs_info->quota_root;
1897 struct btrfs_qgroup *srcgroup;
1898 struct btrfs_qgroup *dstgroup;
1902 mutex_lock(&fs_info->qgroup_ioctl_lock);
1903 if (!fs_info->quota_enabled)
1912 i_qgroups = (u64 *)(inherit + 1);
1913 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
1914 2 * inherit->num_excl_copies;
1915 for (i = 0; i < nums; ++i) {
1916 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
1919 * Zero out invalid groups so we can ignore
1923 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
1931 * create a tracking group for the subvol itself
1933 ret = add_qgroup_item(trans, quota_root, objectid);
1938 struct btrfs_root *srcroot;
1939 struct btrfs_key srckey;
1941 srckey.objectid = srcid;
1942 srckey.type = BTRFS_ROOT_ITEM_KEY;
1943 srckey.offset = (u64)-1;
1944 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
1945 if (IS_ERR(srcroot)) {
1946 ret = PTR_ERR(srcroot);
1951 level_size = srcroot->nodesize;
1956 * add qgroup to all inherited groups
1959 i_qgroups = (u64 *)(inherit + 1);
1960 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
1961 if (*i_qgroups == 0)
1963 ret = add_qgroup_relation_item(trans, quota_root,
1964 objectid, *i_qgroups);
1965 if (ret && ret != -EEXIST)
1967 ret = add_qgroup_relation_item(trans, quota_root,
1968 *i_qgroups, objectid);
1969 if (ret && ret != -EEXIST)
1976 spin_lock(&fs_info->qgroup_lock);
1978 dstgroup = add_qgroup_rb(fs_info, objectid);
1979 if (IS_ERR(dstgroup)) {
1980 ret = PTR_ERR(dstgroup);
1984 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
1985 dstgroup->lim_flags = inherit->lim.flags;
1986 dstgroup->max_rfer = inherit->lim.max_rfer;
1987 dstgroup->max_excl = inherit->lim.max_excl;
1988 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
1989 dstgroup->rsv_excl = inherit->lim.rsv_excl;
1991 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
1993 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1994 btrfs_info(fs_info, "unable to update quota limit for %llu",
1995 dstgroup->qgroupid);
2001 srcgroup = find_qgroup_rb(fs_info, srcid);
2006 * We call inherit after we clone the root in order to make sure
2007 * our counts don't go crazy, so at this point the only
2008 * difference between the two roots should be the root node.
2010 dstgroup->rfer = srcgroup->rfer;
2011 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2012 dstgroup->excl = level_size;
2013 dstgroup->excl_cmpr = level_size;
2014 srcgroup->excl = level_size;
2015 srcgroup->excl_cmpr = level_size;
2017 /* inherit the limit info */
2018 dstgroup->lim_flags = srcgroup->lim_flags;
2019 dstgroup->max_rfer = srcgroup->max_rfer;
2020 dstgroup->max_excl = srcgroup->max_excl;
2021 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2022 dstgroup->rsv_excl = srcgroup->rsv_excl;
2024 qgroup_dirty(fs_info, dstgroup);
2025 qgroup_dirty(fs_info, srcgroup);
2031 i_qgroups = (u64 *)(inherit + 1);
2032 for (i = 0; i < inherit->num_qgroups; ++i) {
2034 ret = add_relation_rb(quota_root->fs_info, objectid,
2042 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2043 struct btrfs_qgroup *src;
2044 struct btrfs_qgroup *dst;
2046 if (!i_qgroups[0] || !i_qgroups[1])
2049 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2050 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2057 dst->rfer = src->rfer - level_size;
2058 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2060 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2061 struct btrfs_qgroup *src;
2062 struct btrfs_qgroup *dst;
2064 if (!i_qgroups[0] || !i_qgroups[1])
2067 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2068 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2075 dst->excl = src->excl + level_size;
2076 dst->excl_cmpr = src->excl_cmpr + level_size;
2080 spin_unlock(&fs_info->qgroup_lock);
2082 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2086 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes)
2088 struct btrfs_root *quota_root;
2089 struct btrfs_qgroup *qgroup;
2090 struct btrfs_fs_info *fs_info = root->fs_info;
2091 u64 ref_root = root->root_key.objectid;
2093 struct ulist_node *unode;
2094 struct ulist_iterator uiter;
2096 if (!is_fstree(ref_root))
2102 spin_lock(&fs_info->qgroup_lock);
2103 quota_root = fs_info->quota_root;
2107 qgroup = find_qgroup_rb(fs_info, ref_root);
2112 * in a first step, we check all affected qgroups if any limits would
2115 ulist_reinit(fs_info->qgroup_ulist);
2116 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2117 (uintptr_t)qgroup, GFP_ATOMIC);
2120 ULIST_ITER_INIT(&uiter);
2121 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2122 struct btrfs_qgroup *qg;
2123 struct btrfs_qgroup_list *glist;
2125 qg = u64_to_ptr(unode->aux);
2127 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2128 qg->reserved + (s64)qg->rfer + num_bytes >
2134 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2135 qg->reserved + (s64)qg->excl + num_bytes >
2141 list_for_each_entry(glist, &qg->groups, next_group) {
2142 ret = ulist_add(fs_info->qgroup_ulist,
2143 glist->group->qgroupid,
2144 (uintptr_t)glist->group, GFP_ATOMIC);
2151 * no limits exceeded, now record the reservation into all qgroups
2153 ULIST_ITER_INIT(&uiter);
2154 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2155 struct btrfs_qgroup *qg;
2157 qg = u64_to_ptr(unode->aux);
2159 qg->reserved += num_bytes;
2163 spin_unlock(&fs_info->qgroup_lock);
2167 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2168 u64 ref_root, u64 num_bytes)
2170 struct btrfs_root *quota_root;
2171 struct btrfs_qgroup *qgroup;
2172 struct ulist_node *unode;
2173 struct ulist_iterator uiter;
2176 if (!is_fstree(ref_root))
2182 spin_lock(&fs_info->qgroup_lock);
2184 quota_root = fs_info->quota_root;
2188 qgroup = find_qgroup_rb(fs_info, ref_root);
2192 ulist_reinit(fs_info->qgroup_ulist);
2193 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2194 (uintptr_t)qgroup, GFP_ATOMIC);
2197 ULIST_ITER_INIT(&uiter);
2198 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2199 struct btrfs_qgroup *qg;
2200 struct btrfs_qgroup_list *glist;
2202 qg = u64_to_ptr(unode->aux);
2204 qg->reserved -= num_bytes;
2206 list_for_each_entry(glist, &qg->groups, next_group) {
2207 ret = ulist_add(fs_info->qgroup_ulist,
2208 glist->group->qgroupid,
2209 (uintptr_t)glist->group, GFP_ATOMIC);
2216 spin_unlock(&fs_info->qgroup_lock);
2219 static inline void qgroup_free(struct btrfs_root *root, u64 num_bytes)
2221 return btrfs_qgroup_free_refroot(root->fs_info, root->objectid,
2224 void assert_qgroups_uptodate(struct btrfs_trans_handle *trans)
2226 if (list_empty(&trans->qgroup_ref_list) && !trans->delayed_ref_elem.seq)
2228 btrfs_err(trans->fs_info,
2229 "qgroups not uptodate in trans handle %p: list is%s empty, "
2231 trans, list_empty(&trans->qgroup_ref_list) ? "" : " not",
2232 (u32)(trans->delayed_ref_elem.seq >> 32),
2233 (u32)trans->delayed_ref_elem.seq);
2238 * returns < 0 on error, 0 when more leafs are to be scanned.
2239 * returns 1 when done.
2242 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2243 struct btrfs_trans_handle *trans)
2245 struct btrfs_key found;
2246 struct extent_buffer *scratch_leaf = NULL;
2247 struct ulist *roots = NULL;
2248 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2253 mutex_lock(&fs_info->qgroup_rescan_lock);
2254 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2255 &fs_info->qgroup_rescan_progress,
2258 pr_debug("current progress key (%llu %u %llu), search_slot ret %d\n",
2259 fs_info->qgroup_rescan_progress.objectid,
2260 fs_info->qgroup_rescan_progress.type,
2261 fs_info->qgroup_rescan_progress.offset, ret);
2265 * The rescan is about to end, we will not be scanning any
2266 * further blocks. We cannot unset the RESCAN flag here, because
2267 * we want to commit the transaction if everything went well.
2268 * To make the live accounting work in this phase, we set our
2269 * scan progress pointer such that every real extent objectid
2272 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2273 btrfs_release_path(path);
2274 mutex_unlock(&fs_info->qgroup_rescan_lock);
2278 btrfs_item_key_to_cpu(path->nodes[0], &found,
2279 btrfs_header_nritems(path->nodes[0]) - 1);
2280 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2282 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2283 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2284 if (!scratch_leaf) {
2286 mutex_unlock(&fs_info->qgroup_rescan_lock);
2289 extent_buffer_get(scratch_leaf);
2290 btrfs_tree_read_lock(scratch_leaf);
2291 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2292 slot = path->slots[0];
2293 btrfs_release_path(path);
2294 mutex_unlock(&fs_info->qgroup_rescan_lock);
2296 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2297 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2298 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2299 found.type != BTRFS_METADATA_ITEM_KEY)
2301 if (found.type == BTRFS_METADATA_ITEM_KEY)
2302 num_bytes = fs_info->extent_root->nodesize;
2304 num_bytes = found.offset;
2306 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2310 /* For rescan, just pass old_roots as NULL */
2311 ret = btrfs_qgroup_account_extent(trans, fs_info,
2312 found.objectid, num_bytes, NULL, roots);
2318 btrfs_tree_read_unlock_blocking(scratch_leaf);
2319 free_extent_buffer(scratch_leaf);
2321 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2326 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2328 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2329 qgroup_rescan_work);
2330 struct btrfs_path *path;
2331 struct btrfs_trans_handle *trans = NULL;
2335 mutex_lock(&fs_info->qgroup_rescan_lock);
2336 fs_info->qgroup_rescan_running = true;
2337 mutex_unlock(&fs_info->qgroup_rescan_lock);
2339 path = btrfs_alloc_path();
2344 while (!err && !btrfs_fs_closing(fs_info)) {
2345 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2346 if (IS_ERR(trans)) {
2347 err = PTR_ERR(trans);
2350 if (!fs_info->quota_enabled) {
2353 err = qgroup_rescan_leaf(fs_info, path, trans);
2356 btrfs_commit_transaction(trans, fs_info->fs_root);
2358 btrfs_end_transaction(trans, fs_info->fs_root);
2362 btrfs_free_path(path);
2364 mutex_lock(&fs_info->qgroup_rescan_lock);
2365 if (!btrfs_fs_closing(fs_info))
2366 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2369 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2370 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2371 } else if (err < 0) {
2372 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2374 mutex_unlock(&fs_info->qgroup_rescan_lock);
2377 * only update status, since the previous part has already updated the
2380 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2381 if (IS_ERR(trans)) {
2382 err = PTR_ERR(trans);
2384 "fail to start transaction for status update: %d\n",
2388 ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2391 btrfs_err(fs_info, "fail to update qgroup status: %d\n", err);
2393 btrfs_end_transaction(trans, fs_info->quota_root);
2395 if (btrfs_fs_closing(fs_info)) {
2396 btrfs_info(fs_info, "qgroup scan paused");
2397 } else if (err >= 0) {
2398 btrfs_info(fs_info, "qgroup scan completed%s",
2399 err > 0 ? " (inconsistency flag cleared)" : "");
2401 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2405 mutex_lock(&fs_info->qgroup_rescan_lock);
2406 fs_info->qgroup_rescan_running = false;
2407 mutex_unlock(&fs_info->qgroup_rescan_lock);
2408 complete_all(&fs_info->qgroup_rescan_completion);
2412 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2413 * memory required for the rescan context.
2416 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2422 (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2423 !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2428 mutex_lock(&fs_info->qgroup_rescan_lock);
2429 spin_lock(&fs_info->qgroup_lock);
2432 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2434 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2438 spin_unlock(&fs_info->qgroup_lock);
2439 mutex_unlock(&fs_info->qgroup_rescan_lock);
2442 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2445 memset(&fs_info->qgroup_rescan_progress, 0,
2446 sizeof(fs_info->qgroup_rescan_progress));
2447 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2448 init_completion(&fs_info->qgroup_rescan_completion);
2450 spin_unlock(&fs_info->qgroup_lock);
2451 mutex_unlock(&fs_info->qgroup_rescan_lock);
2453 memset(&fs_info->qgroup_rescan_work, 0,
2454 sizeof(fs_info->qgroup_rescan_work));
2455 btrfs_init_work(&fs_info->qgroup_rescan_work,
2456 btrfs_qgroup_rescan_helper,
2457 btrfs_qgroup_rescan_worker, NULL, NULL);
2461 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2469 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2472 struct btrfs_qgroup *qgroup;
2474 spin_lock(&fs_info->qgroup_lock);
2475 /* clear all current qgroup tracking information */
2476 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2477 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2479 qgroup->rfer_cmpr = 0;
2481 qgroup->excl_cmpr = 0;
2483 spin_unlock(&fs_info->qgroup_lock);
2487 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2490 struct btrfs_trans_handle *trans;
2492 ret = qgroup_rescan_init(fs_info, 0, 1);
2497 * We have set the rescan_progress to 0, which means no more
2498 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2499 * However, btrfs_qgroup_account_ref may be right after its call
2500 * to btrfs_find_all_roots, in which case it would still do the
2502 * To solve this, we're committing the transaction, which will
2503 * ensure we run all delayed refs and only after that, we are
2504 * going to clear all tracking information for a clean start.
2507 trans = btrfs_join_transaction(fs_info->fs_root);
2508 if (IS_ERR(trans)) {
2509 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2510 return PTR_ERR(trans);
2512 ret = btrfs_commit_transaction(trans, fs_info->fs_root);
2514 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2518 qgroup_rescan_zero_tracking(fs_info);
2520 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2521 &fs_info->qgroup_rescan_work);
2526 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2532 mutex_lock(&fs_info->qgroup_rescan_lock);
2533 spin_lock(&fs_info->qgroup_lock);
2534 running = fs_info->qgroup_rescan_running;
2535 spin_unlock(&fs_info->qgroup_lock);
2536 mutex_unlock(&fs_info->qgroup_rescan_lock);
2542 ret = wait_for_completion_interruptible(
2543 &fs_info->qgroup_rescan_completion);
2545 wait_for_completion(&fs_info->qgroup_rescan_completion);
2551 * this is only called from open_ctree where we're still single threaded, thus
2552 * locking is omitted here.
2555 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2557 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2558 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2559 &fs_info->qgroup_rescan_work);
2563 * Reserve qgroup space for range [start, start + len).
2565 * This function will either reserve space from related qgroups or doing
2566 * nothing if the range is already reserved.
2568 * Return 0 for successful reserve
2569 * Return <0 for error (including -EQUOT)
2571 * NOTE: this function may sleep for memory allocation.
2573 int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len)
2575 struct btrfs_root *root = BTRFS_I(inode)->root;
2576 struct extent_changeset changeset;
2577 struct ulist_node *unode;
2578 struct ulist_iterator uiter;
2581 if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) ||
2585 changeset.bytes_changed = 0;
2586 changeset.range_changed = ulist_alloc(GFP_NOFS);
2587 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2588 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2589 trace_btrfs_qgroup_reserve_data(inode, start, len,
2590 changeset.bytes_changed,
2594 ret = qgroup_reserve(root, changeset.bytes_changed);
2598 ulist_free(changeset.range_changed);
2602 /* cleanup already reserved ranges */
2603 ULIST_ITER_INIT(&uiter);
2604 while ((unode = ulist_next(changeset.range_changed, &uiter)))
2605 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2606 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL,
2608 ulist_free(changeset.range_changed);
2612 static int __btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len,
2615 struct extent_changeset changeset;
2616 int trace_op = QGROUP_RELEASE;
2619 changeset.bytes_changed = 0;
2620 changeset.range_changed = ulist_alloc(GFP_NOFS);
2621 if (!changeset.range_changed)
2624 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2625 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2630 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2631 trace_op = QGROUP_FREE;
2633 trace_btrfs_qgroup_release_data(inode, start, len,
2634 changeset.bytes_changed, trace_op);
2636 ulist_free(changeset.range_changed);
2641 * Free a reserved space range from io_tree and related qgroups
2643 * Should be called when a range of pages get invalidated before reaching disk.
2644 * Or for error cleanup case.
2646 * For data written to disk, use btrfs_qgroup_release_data().
2648 * NOTE: This function may sleep for memory allocation.
2650 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len)
2652 return __btrfs_qgroup_release_data(inode, start, len, 1);
2656 * Release a reserved space range from io_tree only.
2658 * Should be called when a range of pages get written to disk and corresponding
2659 * FILE_EXTENT is inserted into corresponding root.
2661 * Since new qgroup accounting framework will only update qgroup numbers at
2662 * commit_transaction() time, its reserved space shouldn't be freed from
2665 * But we should release the range from io_tree, to allow further write to be
2668 * NOTE: This function may sleep for memory allocation.
2670 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
2672 return __btrfs_qgroup_release_data(inode, start, len, 0);
2675 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes)
2679 if (!root->fs_info->quota_enabled || !is_fstree(root->objectid) ||
2683 BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
2684 ret = qgroup_reserve(root, num_bytes);
2687 atomic_add(num_bytes, &root->qgroup_meta_rsv);
2691 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
2695 if (!root->fs_info->quota_enabled || !is_fstree(root->objectid))
2698 reserved = atomic_xchg(&root->qgroup_meta_rsv, 0);
2701 qgroup_free(root, reserved);
2704 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
2706 if (!root->fs_info->quota_enabled || !is_fstree(root->objectid))
2709 BUG_ON(num_bytes != round_down(num_bytes, root->nodesize));
2710 WARN_ON(atomic_read(&root->qgroup_meta_rsv) < num_bytes);
2711 atomic_sub(num_bytes, &root->qgroup_meta_rsv);
2712 qgroup_free(root, num_bytes);
2716 * Check qgroup reserved space leaking, normally at destroy inode
2719 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
2721 struct extent_changeset changeset;
2722 struct ulist_node *unode;
2723 struct ulist_iterator iter;
2726 changeset.bytes_changed = 0;
2727 changeset.range_changed = ulist_alloc(GFP_NOFS);
2728 if (WARN_ON(!changeset.range_changed))
2731 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
2732 EXTENT_QGROUP_RESERVED, &changeset);
2735 if (WARN_ON(changeset.bytes_changed)) {
2736 ULIST_ITER_INIT(&iter);
2737 while ((unode = ulist_next(changeset.range_changed, &iter))) {
2738 btrfs_warn(BTRFS_I(inode)->root->fs_info,
2739 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
2740 inode->i_ino, unode->val, unode->aux);
2742 qgroup_free(BTRFS_I(inode)->root, changeset.bytes_changed);
2744 ulist_free(changeset.range_changed);