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
50 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
53 if (qg->old_refcnt < seq)
55 qg->old_refcnt += mod;
58 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
61 if (qg->new_refcnt < seq)
63 qg->new_refcnt += mod;
66 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
68 if (qg->old_refcnt < seq)
70 return qg->old_refcnt - seq;
73 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
75 if (qg->new_refcnt < seq)
77 return qg->new_refcnt - seq;
81 * glue structure to represent the relations between qgroups.
83 struct btrfs_qgroup_list {
84 struct list_head next_group;
85 struct list_head next_member;
86 struct btrfs_qgroup *group;
87 struct btrfs_qgroup *member;
90 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
92 return (u64)(uintptr_t)qg;
95 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
97 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
101 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
103 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
105 /* must be called with qgroup_ioctl_lock held */
106 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
109 struct rb_node *n = fs_info->qgroup_tree.rb_node;
110 struct btrfs_qgroup *qgroup;
113 qgroup = rb_entry(n, struct btrfs_qgroup, node);
114 if (qgroup->qgroupid < qgroupid)
116 else if (qgroup->qgroupid > qgroupid)
124 /* must be called with qgroup_lock held */
125 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
128 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
129 struct rb_node *parent = NULL;
130 struct btrfs_qgroup *qgroup;
134 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
136 if (qgroup->qgroupid < qgroupid)
138 else if (qgroup->qgroupid > qgroupid)
144 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
146 return ERR_PTR(-ENOMEM);
148 qgroup->qgroupid = qgroupid;
149 INIT_LIST_HEAD(&qgroup->groups);
150 INIT_LIST_HEAD(&qgroup->members);
151 INIT_LIST_HEAD(&qgroup->dirty);
153 rb_link_node(&qgroup->node, parent, p);
154 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
159 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
161 struct btrfs_qgroup_list *list;
163 list_del(&qgroup->dirty);
164 while (!list_empty(&qgroup->groups)) {
165 list = list_first_entry(&qgroup->groups,
166 struct btrfs_qgroup_list, next_group);
167 list_del(&list->next_group);
168 list_del(&list->next_member);
172 while (!list_empty(&qgroup->members)) {
173 list = list_first_entry(&qgroup->members,
174 struct btrfs_qgroup_list, next_member);
175 list_del(&list->next_group);
176 list_del(&list->next_member);
182 /* must be called with qgroup_lock held */
183 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
185 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
190 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
191 __del_qgroup_rb(qgroup);
195 /* must be called with qgroup_lock held */
196 static int add_relation_rb(struct btrfs_fs_info *fs_info,
197 u64 memberid, u64 parentid)
199 struct btrfs_qgroup *member;
200 struct btrfs_qgroup *parent;
201 struct btrfs_qgroup_list *list;
203 member = find_qgroup_rb(fs_info, memberid);
204 parent = find_qgroup_rb(fs_info, parentid);
205 if (!member || !parent)
208 list = kzalloc(sizeof(*list), GFP_ATOMIC);
212 list->group = parent;
213 list->member = member;
214 list_add_tail(&list->next_group, &member->groups);
215 list_add_tail(&list->next_member, &parent->members);
220 /* must be called with qgroup_lock held */
221 static int del_relation_rb(struct btrfs_fs_info *fs_info,
222 u64 memberid, u64 parentid)
224 struct btrfs_qgroup *member;
225 struct btrfs_qgroup *parent;
226 struct btrfs_qgroup_list *list;
228 member = find_qgroup_rb(fs_info, memberid);
229 parent = find_qgroup_rb(fs_info, parentid);
230 if (!member || !parent)
233 list_for_each_entry(list, &member->groups, next_group) {
234 if (list->group == parent) {
235 list_del(&list->next_group);
236 list_del(&list->next_member);
244 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
245 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
248 struct btrfs_qgroup *qgroup;
250 qgroup = find_qgroup_rb(fs_info, qgroupid);
253 if (qgroup->rfer != rfer || qgroup->excl != excl)
260 * The full config is read in one go, only called from open_ctree()
261 * It doesn't use any locking, as at this point we're still single-threaded
263 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
265 struct btrfs_key key;
266 struct btrfs_key found_key;
267 struct btrfs_root *quota_root = fs_info->quota_root;
268 struct btrfs_path *path = NULL;
269 struct extent_buffer *l;
273 u64 rescan_progress = 0;
275 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
278 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
279 if (!fs_info->qgroup_ulist) {
284 path = btrfs_alloc_path();
290 /* default this to quota off, in case no status key is found */
291 fs_info->qgroup_flags = 0;
294 * pass 1: read status, all qgroup infos and limits
299 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
304 struct btrfs_qgroup *qgroup;
306 slot = path->slots[0];
308 btrfs_item_key_to_cpu(l, &found_key, slot);
310 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
311 struct btrfs_qgroup_status_item *ptr;
313 ptr = btrfs_item_ptr(l, slot,
314 struct btrfs_qgroup_status_item);
316 if (btrfs_qgroup_status_version(l, ptr) !=
317 BTRFS_QGROUP_STATUS_VERSION) {
319 "old qgroup version, quota disabled");
322 if (btrfs_qgroup_status_generation(l, ptr) !=
323 fs_info->generation) {
324 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
326 "qgroup generation mismatch, marked as inconsistent");
328 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
330 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
334 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
335 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
338 qgroup = find_qgroup_rb(fs_info, found_key.offset);
339 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
340 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
341 btrfs_err(fs_info, "inconsistent qgroup config");
342 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
345 qgroup = add_qgroup_rb(fs_info, found_key.offset);
346 if (IS_ERR(qgroup)) {
347 ret = PTR_ERR(qgroup);
351 switch (found_key.type) {
352 case BTRFS_QGROUP_INFO_KEY: {
353 struct btrfs_qgroup_info_item *ptr;
355 ptr = btrfs_item_ptr(l, slot,
356 struct btrfs_qgroup_info_item);
357 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
358 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
359 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
360 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
361 /* generation currently unused */
364 case BTRFS_QGROUP_LIMIT_KEY: {
365 struct btrfs_qgroup_limit_item *ptr;
367 ptr = btrfs_item_ptr(l, slot,
368 struct btrfs_qgroup_limit_item);
369 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
370 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
371 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
372 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
373 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
378 ret = btrfs_next_item(quota_root, path);
384 btrfs_release_path(path);
387 * pass 2: read all qgroup relations
390 key.type = BTRFS_QGROUP_RELATION_KEY;
392 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
396 slot = path->slots[0];
398 btrfs_item_key_to_cpu(l, &found_key, slot);
400 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
403 if (found_key.objectid > found_key.offset) {
404 /* parent <- member, not needed to build config */
405 /* FIXME should we omit the key completely? */
409 ret = add_relation_rb(fs_info, found_key.objectid,
411 if (ret == -ENOENT) {
413 "orphan qgroup relation 0x%llx->0x%llx",
414 found_key.objectid, found_key.offset);
415 ret = 0; /* ignore the error */
420 ret = btrfs_next_item(quota_root, path);
427 fs_info->qgroup_flags |= flags;
428 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
429 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
430 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
432 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
433 btrfs_free_path(path);
436 ulist_free(fs_info->qgroup_ulist);
437 fs_info->qgroup_ulist = NULL;
438 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
441 return ret < 0 ? ret : 0;
445 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
446 * first two are in single-threaded paths.And for the third one, we have set
447 * quota_root to be null with qgroup_lock held before, so it is safe to clean
448 * up the in-memory structures without qgroup_lock held.
450 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
453 struct btrfs_qgroup *qgroup;
455 while ((n = rb_first(&fs_info->qgroup_tree))) {
456 qgroup = rb_entry(n, struct btrfs_qgroup, node);
457 rb_erase(n, &fs_info->qgroup_tree);
458 __del_qgroup_rb(qgroup);
461 * we call btrfs_free_qgroup_config() when umounting
462 * filesystem and disabling quota, so we set qgroup_ulist
463 * to be null here to avoid double free.
465 ulist_free(fs_info->qgroup_ulist);
466 fs_info->qgroup_ulist = NULL;
469 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
470 struct btrfs_root *quota_root,
474 struct btrfs_path *path;
475 struct btrfs_key key;
477 path = btrfs_alloc_path();
482 key.type = BTRFS_QGROUP_RELATION_KEY;
485 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
487 btrfs_mark_buffer_dirty(path->nodes[0]);
489 btrfs_free_path(path);
493 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
494 struct btrfs_root *quota_root,
498 struct btrfs_path *path;
499 struct btrfs_key key;
501 path = btrfs_alloc_path();
506 key.type = BTRFS_QGROUP_RELATION_KEY;
509 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
518 ret = btrfs_del_item(trans, quota_root, path);
520 btrfs_free_path(path);
524 static int add_qgroup_item(struct btrfs_trans_handle *trans,
525 struct btrfs_root *quota_root, u64 qgroupid)
528 struct btrfs_path *path;
529 struct btrfs_qgroup_info_item *qgroup_info;
530 struct btrfs_qgroup_limit_item *qgroup_limit;
531 struct extent_buffer *leaf;
532 struct btrfs_key key;
534 if (btrfs_is_testing(quota_root->fs_info))
537 path = btrfs_alloc_path();
542 key.type = BTRFS_QGROUP_INFO_KEY;
543 key.offset = qgroupid;
546 * Avoid a transaction abort by catching -EEXIST here. In that
547 * case, we proceed by re-initializing the existing structure
551 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
552 sizeof(*qgroup_info));
553 if (ret && ret != -EEXIST)
556 leaf = path->nodes[0];
557 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
558 struct btrfs_qgroup_info_item);
559 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
560 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
561 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
562 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
563 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
565 btrfs_mark_buffer_dirty(leaf);
567 btrfs_release_path(path);
569 key.type = BTRFS_QGROUP_LIMIT_KEY;
570 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
571 sizeof(*qgroup_limit));
572 if (ret && ret != -EEXIST)
575 leaf = path->nodes[0];
576 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
577 struct btrfs_qgroup_limit_item);
578 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
579 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
580 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
581 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
582 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
584 btrfs_mark_buffer_dirty(leaf);
588 btrfs_free_path(path);
592 static int del_qgroup_item(struct btrfs_trans_handle *trans,
593 struct btrfs_root *quota_root, u64 qgroupid)
596 struct btrfs_path *path;
597 struct btrfs_key key;
599 path = btrfs_alloc_path();
604 key.type = BTRFS_QGROUP_INFO_KEY;
605 key.offset = qgroupid;
606 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
615 ret = btrfs_del_item(trans, quota_root, path);
619 btrfs_release_path(path);
621 key.type = BTRFS_QGROUP_LIMIT_KEY;
622 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
631 ret = btrfs_del_item(trans, quota_root, path);
634 btrfs_free_path(path);
638 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
639 struct btrfs_root *root,
640 struct btrfs_qgroup *qgroup)
642 struct btrfs_path *path;
643 struct btrfs_key key;
644 struct extent_buffer *l;
645 struct btrfs_qgroup_limit_item *qgroup_limit;
650 key.type = BTRFS_QGROUP_LIMIT_KEY;
651 key.offset = qgroup->qgroupid;
653 path = btrfs_alloc_path();
657 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
665 slot = path->slots[0];
666 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
667 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
668 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
669 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
670 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
671 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
673 btrfs_mark_buffer_dirty(l);
676 btrfs_free_path(path);
680 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
681 struct btrfs_root *root,
682 struct btrfs_qgroup *qgroup)
684 struct btrfs_path *path;
685 struct btrfs_key key;
686 struct extent_buffer *l;
687 struct btrfs_qgroup_info_item *qgroup_info;
691 if (btrfs_is_testing(root->fs_info))
695 key.type = BTRFS_QGROUP_INFO_KEY;
696 key.offset = qgroup->qgroupid;
698 path = btrfs_alloc_path();
702 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
710 slot = path->slots[0];
711 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
712 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
713 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
714 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
715 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
716 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
718 btrfs_mark_buffer_dirty(l);
721 btrfs_free_path(path);
725 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
726 struct btrfs_fs_info *fs_info,
727 struct btrfs_root *root)
729 struct btrfs_path *path;
730 struct btrfs_key key;
731 struct extent_buffer *l;
732 struct btrfs_qgroup_status_item *ptr;
737 key.type = BTRFS_QGROUP_STATUS_KEY;
740 path = btrfs_alloc_path();
744 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
752 slot = path->slots[0];
753 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
754 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
755 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
756 btrfs_set_qgroup_status_rescan(l, ptr,
757 fs_info->qgroup_rescan_progress.objectid);
759 btrfs_mark_buffer_dirty(l);
762 btrfs_free_path(path);
767 * called with qgroup_lock held
769 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
770 struct btrfs_root *root)
772 struct btrfs_path *path;
773 struct btrfs_key key;
774 struct extent_buffer *leaf = NULL;
778 path = btrfs_alloc_path();
782 path->leave_spinning = 1;
789 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
792 leaf = path->nodes[0];
793 nr = btrfs_header_nritems(leaf);
797 * delete the leaf one by one
798 * since the whole tree is going
802 ret = btrfs_del_items(trans, root, path, 0, nr);
806 btrfs_release_path(path);
810 btrfs_free_path(path);
814 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
815 struct btrfs_fs_info *fs_info)
817 struct btrfs_root *quota_root;
818 struct btrfs_root *tree_root = fs_info->tree_root;
819 struct btrfs_path *path = NULL;
820 struct btrfs_qgroup_status_item *ptr;
821 struct extent_buffer *leaf;
822 struct btrfs_key key;
823 struct btrfs_key found_key;
824 struct btrfs_qgroup *qgroup = NULL;
828 mutex_lock(&fs_info->qgroup_ioctl_lock);
829 if (fs_info->quota_root)
832 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
833 if (!fs_info->qgroup_ulist) {
839 * initially create the quota tree
841 quota_root = btrfs_create_tree(trans, fs_info,
842 BTRFS_QUOTA_TREE_OBJECTID);
843 if (IS_ERR(quota_root)) {
844 ret = PTR_ERR(quota_root);
848 path = btrfs_alloc_path();
855 key.type = BTRFS_QGROUP_STATUS_KEY;
858 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
863 leaf = path->nodes[0];
864 ptr = btrfs_item_ptr(leaf, path->slots[0],
865 struct btrfs_qgroup_status_item);
866 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
867 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
868 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
869 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
870 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
871 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
873 btrfs_mark_buffer_dirty(leaf);
876 key.type = BTRFS_ROOT_REF_KEY;
879 btrfs_release_path(path);
880 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
888 slot = path->slots[0];
889 leaf = path->nodes[0];
890 btrfs_item_key_to_cpu(leaf, &found_key, slot);
892 if (found_key.type == BTRFS_ROOT_REF_KEY) {
893 ret = add_qgroup_item(trans, quota_root,
898 qgroup = add_qgroup_rb(fs_info, found_key.offset);
899 if (IS_ERR(qgroup)) {
900 ret = PTR_ERR(qgroup);
904 ret = btrfs_next_item(tree_root, path);
912 btrfs_release_path(path);
913 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
917 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
918 if (IS_ERR(qgroup)) {
919 ret = PTR_ERR(qgroup);
922 spin_lock(&fs_info->qgroup_lock);
923 fs_info->quota_root = quota_root;
924 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
925 spin_unlock(&fs_info->qgroup_lock);
926 ret = qgroup_rescan_init(fs_info, 0, 1);
928 qgroup_rescan_zero_tracking(fs_info);
929 btrfs_queue_work(fs_info->qgroup_rescan_workers,
930 &fs_info->qgroup_rescan_work);
934 btrfs_free_path(path);
937 free_extent_buffer(quota_root->node);
938 free_extent_buffer(quota_root->commit_root);
943 ulist_free(fs_info->qgroup_ulist);
944 fs_info->qgroup_ulist = NULL;
946 mutex_unlock(&fs_info->qgroup_ioctl_lock);
950 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
951 struct btrfs_fs_info *fs_info)
953 struct btrfs_root *quota_root;
956 mutex_lock(&fs_info->qgroup_ioctl_lock);
957 if (!fs_info->quota_root)
959 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
960 btrfs_qgroup_wait_for_completion(fs_info, false);
961 spin_lock(&fs_info->qgroup_lock);
962 quota_root = fs_info->quota_root;
963 fs_info->quota_root = NULL;
964 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
965 spin_unlock(&fs_info->qgroup_lock);
967 btrfs_free_qgroup_config(fs_info);
969 ret = btrfs_clean_quota_tree(trans, quota_root);
973 ret = btrfs_del_root(trans, fs_info, "a_root->root_key);
977 list_del("a_root->dirty_list);
979 btrfs_tree_lock(quota_root->node);
980 clean_tree_block(fs_info, quota_root->node);
981 btrfs_tree_unlock(quota_root->node);
982 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
984 free_extent_buffer(quota_root->node);
985 free_extent_buffer(quota_root->commit_root);
988 mutex_unlock(&fs_info->qgroup_ioctl_lock);
992 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
993 struct btrfs_qgroup *qgroup)
995 if (list_empty(&qgroup->dirty))
996 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
999 static void report_reserved_underflow(struct btrfs_fs_info *fs_info,
1000 struct btrfs_qgroup *qgroup,
1003 #ifdef CONFIG_BTRFS_DEBUG
1004 WARN_ON(qgroup->reserved < num_bytes);
1005 btrfs_debug(fs_info,
1006 "qgroup %llu reserved space underflow, have: %llu, to free: %llu",
1007 qgroup->qgroupid, qgroup->reserved, num_bytes);
1009 qgroup->reserved = 0;
1012 * The easy accounting, if we are adding/removing the only ref for an extent
1013 * then this qgroup and all of the parent qgroups get their reference and
1014 * exclusive counts adjusted.
1016 * Caller should hold fs_info->qgroup_lock.
1018 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1019 struct ulist *tmp, u64 ref_root,
1020 u64 num_bytes, int sign)
1022 struct btrfs_qgroup *qgroup;
1023 struct btrfs_qgroup_list *glist;
1024 struct ulist_node *unode;
1025 struct ulist_iterator uiter;
1028 qgroup = find_qgroup_rb(fs_info, ref_root);
1032 qgroup->rfer += sign * num_bytes;
1033 qgroup->rfer_cmpr += sign * num_bytes;
1035 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1036 qgroup->excl += sign * num_bytes;
1037 qgroup->excl_cmpr += sign * num_bytes;
1039 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes);
1040 if (qgroup->reserved < num_bytes)
1041 report_reserved_underflow(fs_info, qgroup, num_bytes);
1043 qgroup->reserved -= num_bytes;
1046 qgroup_dirty(fs_info, qgroup);
1048 /* Get all of the parent groups that contain this qgroup */
1049 list_for_each_entry(glist, &qgroup->groups, next_group) {
1050 ret = ulist_add(tmp, glist->group->qgroupid,
1051 qgroup_to_aux(glist->group), GFP_ATOMIC);
1056 /* Iterate all of the parents and adjust their reference counts */
1057 ULIST_ITER_INIT(&uiter);
1058 while ((unode = ulist_next(tmp, &uiter))) {
1059 qgroup = unode_aux_to_qgroup(unode);
1060 qgroup->rfer += sign * num_bytes;
1061 qgroup->rfer_cmpr += sign * num_bytes;
1062 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1063 qgroup->excl += sign * num_bytes;
1065 trace_qgroup_update_reserve(fs_info, qgroup,
1067 if (qgroup->reserved < num_bytes)
1068 report_reserved_underflow(fs_info, qgroup,
1071 qgroup->reserved -= num_bytes;
1073 qgroup->excl_cmpr += sign * num_bytes;
1074 qgroup_dirty(fs_info, qgroup);
1076 /* Add any parents of the parents */
1077 list_for_each_entry(glist, &qgroup->groups, next_group) {
1078 ret = ulist_add(tmp, glist->group->qgroupid,
1079 qgroup_to_aux(glist->group), GFP_ATOMIC);
1091 * Quick path for updating qgroup with only excl refs.
1093 * In that case, just update all parent will be enough.
1094 * Or we needs to do a full rescan.
1095 * Caller should also hold fs_info->qgroup_lock.
1097 * Return 0 for quick update, return >0 for need to full rescan
1098 * and mark INCONSISTENT flag.
1099 * Return < 0 for other error.
1101 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1102 struct ulist *tmp, u64 src, u64 dst,
1105 struct btrfs_qgroup *qgroup;
1109 qgroup = find_qgroup_rb(fs_info, src);
1112 if (qgroup->excl == qgroup->rfer) {
1114 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1115 qgroup->excl, sign);
1123 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1127 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1128 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1130 struct btrfs_root *quota_root;
1131 struct btrfs_qgroup *parent;
1132 struct btrfs_qgroup *member;
1133 struct btrfs_qgroup_list *list;
1137 /* Check the level of src and dst first */
1138 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1141 tmp = ulist_alloc(GFP_KERNEL);
1145 mutex_lock(&fs_info->qgroup_ioctl_lock);
1146 quota_root = fs_info->quota_root;
1151 member = find_qgroup_rb(fs_info, src);
1152 parent = find_qgroup_rb(fs_info, dst);
1153 if (!member || !parent) {
1158 /* check if such qgroup relation exist firstly */
1159 list_for_each_entry(list, &member->groups, next_group) {
1160 if (list->group == parent) {
1166 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1170 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1172 del_qgroup_relation_item(trans, quota_root, src, dst);
1176 spin_lock(&fs_info->qgroup_lock);
1177 ret = add_relation_rb(fs_info, src, dst);
1179 spin_unlock(&fs_info->qgroup_lock);
1182 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1183 spin_unlock(&fs_info->qgroup_lock);
1185 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1190 static int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1191 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1193 struct btrfs_root *quota_root;
1194 struct btrfs_qgroup *parent;
1195 struct btrfs_qgroup *member;
1196 struct btrfs_qgroup_list *list;
1201 tmp = ulist_alloc(GFP_KERNEL);
1205 quota_root = fs_info->quota_root;
1211 member = find_qgroup_rb(fs_info, src);
1212 parent = find_qgroup_rb(fs_info, dst);
1213 if (!member || !parent) {
1218 /* check if such qgroup relation exist firstly */
1219 list_for_each_entry(list, &member->groups, next_group) {
1220 if (list->group == parent)
1226 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1227 err = del_qgroup_relation_item(trans, quota_root, dst, src);
1231 spin_lock(&fs_info->qgroup_lock);
1232 del_relation_rb(fs_info, src, dst);
1233 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1234 spin_unlock(&fs_info->qgroup_lock);
1240 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1241 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1245 mutex_lock(&fs_info->qgroup_ioctl_lock);
1246 ret = __del_qgroup_relation(trans, fs_info, src, dst);
1247 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1252 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1253 struct btrfs_fs_info *fs_info, u64 qgroupid)
1255 struct btrfs_root *quota_root;
1256 struct btrfs_qgroup *qgroup;
1259 mutex_lock(&fs_info->qgroup_ioctl_lock);
1260 quota_root = fs_info->quota_root;
1265 qgroup = find_qgroup_rb(fs_info, qgroupid);
1271 ret = add_qgroup_item(trans, quota_root, qgroupid);
1275 spin_lock(&fs_info->qgroup_lock);
1276 qgroup = add_qgroup_rb(fs_info, qgroupid);
1277 spin_unlock(&fs_info->qgroup_lock);
1280 ret = PTR_ERR(qgroup);
1282 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1286 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1287 struct btrfs_fs_info *fs_info, u64 qgroupid)
1289 struct btrfs_root *quota_root;
1290 struct btrfs_qgroup *qgroup;
1291 struct btrfs_qgroup_list *list;
1294 mutex_lock(&fs_info->qgroup_ioctl_lock);
1295 quota_root = fs_info->quota_root;
1301 qgroup = find_qgroup_rb(fs_info, qgroupid);
1306 /* check if there are no children of this qgroup */
1307 if (!list_empty(&qgroup->members)) {
1312 ret = del_qgroup_item(trans, quota_root, qgroupid);
1313 if (ret && ret != -ENOENT)
1316 while (!list_empty(&qgroup->groups)) {
1317 list = list_first_entry(&qgroup->groups,
1318 struct btrfs_qgroup_list, next_group);
1319 ret = __del_qgroup_relation(trans, fs_info,
1321 list->group->qgroupid);
1326 spin_lock(&fs_info->qgroup_lock);
1327 del_qgroup_rb(fs_info, qgroupid);
1328 spin_unlock(&fs_info->qgroup_lock);
1330 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1334 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1335 struct btrfs_fs_info *fs_info, u64 qgroupid,
1336 struct btrfs_qgroup_limit *limit)
1338 struct btrfs_root *quota_root;
1339 struct btrfs_qgroup *qgroup;
1341 /* Sometimes we would want to clear the limit on this qgroup.
1342 * To meet this requirement, we treat the -1 as a special value
1343 * which tell kernel to clear the limit on this qgroup.
1345 const u64 CLEAR_VALUE = -1;
1347 mutex_lock(&fs_info->qgroup_ioctl_lock);
1348 quota_root = fs_info->quota_root;
1354 qgroup = find_qgroup_rb(fs_info, qgroupid);
1360 spin_lock(&fs_info->qgroup_lock);
1361 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1362 if (limit->max_rfer == CLEAR_VALUE) {
1363 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1364 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1365 qgroup->max_rfer = 0;
1367 qgroup->max_rfer = limit->max_rfer;
1370 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1371 if (limit->max_excl == CLEAR_VALUE) {
1372 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1373 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1374 qgroup->max_excl = 0;
1376 qgroup->max_excl = limit->max_excl;
1379 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1380 if (limit->rsv_rfer == CLEAR_VALUE) {
1381 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1382 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1383 qgroup->rsv_rfer = 0;
1385 qgroup->rsv_rfer = limit->rsv_rfer;
1388 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1389 if (limit->rsv_excl == CLEAR_VALUE) {
1390 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1391 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1392 qgroup->rsv_excl = 0;
1394 qgroup->rsv_excl = limit->rsv_excl;
1397 qgroup->lim_flags |= limit->flags;
1399 spin_unlock(&fs_info->qgroup_lock);
1401 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1403 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1404 btrfs_info(fs_info, "unable to update quota limit for %llu",
1409 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1413 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1414 struct btrfs_delayed_ref_root *delayed_refs,
1415 struct btrfs_qgroup_extent_record *record)
1417 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1418 struct rb_node *parent_node = NULL;
1419 struct btrfs_qgroup_extent_record *entry;
1420 u64 bytenr = record->bytenr;
1422 assert_spin_locked(&delayed_refs->lock);
1423 trace_btrfs_qgroup_trace_extent(fs_info, record);
1427 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1429 if (bytenr < entry->bytenr)
1431 else if (bytenr > entry->bytenr)
1432 p = &(*p)->rb_right;
1437 rb_link_node(&record->node, parent_node, p);
1438 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1442 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1443 struct btrfs_qgroup_extent_record *qrecord)
1445 struct ulist *old_root;
1446 u64 bytenr = qrecord->bytenr;
1449 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1451 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1453 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1459 * Here we don't need to get the lock of
1460 * trans->transaction->delayed_refs, since inserted qrecord won't
1461 * be deleted, only qrecord->node may be modified (new qrecord insert)
1463 * So modifying qrecord->old_roots is safe here
1465 qrecord->old_roots = old_root;
1469 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1470 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1473 struct btrfs_qgroup_extent_record *record;
1474 struct btrfs_delayed_ref_root *delayed_refs;
1477 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1478 || bytenr == 0 || num_bytes == 0)
1480 if (WARN_ON(trans == NULL))
1482 record = kmalloc(sizeof(*record), gfp_flag);
1486 delayed_refs = &trans->transaction->delayed_refs;
1487 record->bytenr = bytenr;
1488 record->num_bytes = num_bytes;
1489 record->old_roots = NULL;
1491 spin_lock(&delayed_refs->lock);
1492 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1493 spin_unlock(&delayed_refs->lock);
1498 return btrfs_qgroup_trace_extent_post(fs_info, record);
1501 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1502 struct btrfs_fs_info *fs_info,
1503 struct extent_buffer *eb)
1505 int nr = btrfs_header_nritems(eb);
1506 int i, extent_type, ret;
1507 struct btrfs_key key;
1508 struct btrfs_file_extent_item *fi;
1509 u64 bytenr, num_bytes;
1511 /* We can be called directly from walk_up_proc() */
1512 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1515 for (i = 0; i < nr; i++) {
1516 btrfs_item_key_to_cpu(eb, &key, i);
1518 if (key.type != BTRFS_EXTENT_DATA_KEY)
1521 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1522 /* filter out non qgroup-accountable extents */
1523 extent_type = btrfs_file_extent_type(eb, fi);
1525 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1528 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1532 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1534 ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1535 num_bytes, GFP_NOFS);
1544 * Walk up the tree from the bottom, freeing leaves and any interior
1545 * nodes which have had all slots visited. If a node (leaf or
1546 * interior) is freed, the node above it will have it's slot
1547 * incremented. The root node will never be freed.
1549 * At the end of this function, we should have a path which has all
1550 * slots incremented to the next position for a search. If we need to
1551 * read a new node it will be NULL and the node above it will have the
1552 * correct slot selected for a later read.
1554 * If we increment the root nodes slot counter past the number of
1555 * elements, 1 is returned to signal completion of the search.
1557 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1561 struct extent_buffer *eb;
1563 if (root_level == 0)
1566 while (level <= root_level) {
1567 eb = path->nodes[level];
1568 nr = btrfs_header_nritems(eb);
1569 path->slots[level]++;
1570 slot = path->slots[level];
1571 if (slot >= nr || level == 0) {
1573 * Don't free the root - we will detect this
1574 * condition after our loop and return a
1575 * positive value for caller to stop walking the tree.
1577 if (level != root_level) {
1578 btrfs_tree_unlock_rw(eb, path->locks[level]);
1579 path->locks[level] = 0;
1581 free_extent_buffer(eb);
1582 path->nodes[level] = NULL;
1583 path->slots[level] = 0;
1587 * We have a valid slot to walk back down
1588 * from. Stop here so caller can process these
1597 eb = path->nodes[root_level];
1598 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1604 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1605 struct btrfs_root *root,
1606 struct extent_buffer *root_eb,
1607 u64 root_gen, int root_level)
1609 struct btrfs_fs_info *fs_info = root->fs_info;
1612 struct extent_buffer *eb = root_eb;
1613 struct btrfs_path *path = NULL;
1615 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1616 BUG_ON(root_eb == NULL);
1618 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1621 if (!extent_buffer_uptodate(root_eb)) {
1622 ret = btrfs_read_buffer(root_eb, root_gen);
1627 if (root_level == 0) {
1628 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1632 path = btrfs_alloc_path();
1637 * Walk down the tree. Missing extent blocks are filled in as
1638 * we go. Metadata is accounted every time we read a new
1641 * When we reach a leaf, we account for file extent items in it,
1642 * walk back up the tree (adjusting slot pointers as we go)
1643 * and restart the search process.
1645 extent_buffer_get(root_eb); /* For path */
1646 path->nodes[root_level] = root_eb;
1647 path->slots[root_level] = 0;
1648 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1651 while (level >= 0) {
1652 if (path->nodes[level] == NULL) {
1658 * We need to get child blockptr/gen from parent before
1661 eb = path->nodes[level + 1];
1662 parent_slot = path->slots[level + 1];
1663 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1664 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1666 eb = read_tree_block(fs_info, child_bytenr, child_gen);
1670 } else if (!extent_buffer_uptodate(eb)) {
1671 free_extent_buffer(eb);
1676 path->nodes[level] = eb;
1677 path->slots[level] = 0;
1679 btrfs_tree_read_lock(eb);
1680 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1681 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1683 ret = btrfs_qgroup_trace_extent(trans, fs_info,
1692 ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1693 path->nodes[level]);
1697 /* Nonzero return here means we completed our search */
1698 ret = adjust_slots_upwards(path, root_level);
1702 /* Restart search with new slots */
1711 btrfs_free_path(path);
1716 #define UPDATE_NEW 0
1717 #define UPDATE_OLD 1
1719 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1721 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1722 struct ulist *roots, struct ulist *tmp,
1723 struct ulist *qgroups, u64 seq, int update_old)
1725 struct ulist_node *unode;
1726 struct ulist_iterator uiter;
1727 struct ulist_node *tmp_unode;
1728 struct ulist_iterator tmp_uiter;
1729 struct btrfs_qgroup *qg;
1734 ULIST_ITER_INIT(&uiter);
1735 while ((unode = ulist_next(roots, &uiter))) {
1736 qg = find_qgroup_rb(fs_info, unode->val);
1741 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1745 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1748 ULIST_ITER_INIT(&tmp_uiter);
1749 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1750 struct btrfs_qgroup_list *glist;
1752 qg = unode_aux_to_qgroup(tmp_unode);
1754 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1756 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1757 list_for_each_entry(glist, &qg->groups, next_group) {
1758 ret = ulist_add(qgroups, glist->group->qgroupid,
1759 qgroup_to_aux(glist->group),
1763 ret = ulist_add(tmp, glist->group->qgroupid,
1764 qgroup_to_aux(glist->group),
1775 * Update qgroup rfer/excl counters.
1776 * Rfer update is easy, codes can explain themselves.
1778 * Excl update is tricky, the update is split into 2 part.
1779 * Part 1: Possible exclusive <-> sharing detect:
1781 * -------------------------------------
1783 * -------------------------------------
1785 * -------------------------------------
1788 * A: cur_old_roots < nr_old_roots (not exclusive before)
1789 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1790 * B: cur_new_roots < nr_new_roots (not exclusive now)
1791 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
1794 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1795 * *: Definitely not changed. **: Possible unchanged.
1797 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1799 * To make the logic clear, we first use condition A and B to split
1800 * combination into 4 results.
1802 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1803 * only on variant maybe 0.
1805 * Lastly, check result **, since there are 2 variants maybe 0, split them
1807 * But this time we don't need to consider other things, the codes and logic
1808 * is easy to understand now.
1810 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1811 struct ulist *qgroups,
1814 u64 num_bytes, u64 seq)
1816 struct ulist_node *unode;
1817 struct ulist_iterator uiter;
1818 struct btrfs_qgroup *qg;
1819 u64 cur_new_count, cur_old_count;
1821 ULIST_ITER_INIT(&uiter);
1822 while ((unode = ulist_next(qgroups, &uiter))) {
1825 qg = unode_aux_to_qgroup(unode);
1826 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1827 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1829 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1830 cur_old_count, cur_new_count);
1832 /* Rfer update part */
1833 if (cur_old_count == 0 && cur_new_count > 0) {
1834 qg->rfer += num_bytes;
1835 qg->rfer_cmpr += num_bytes;
1838 if (cur_old_count > 0 && cur_new_count == 0) {
1839 qg->rfer -= num_bytes;
1840 qg->rfer_cmpr -= num_bytes;
1844 /* Excl update part */
1845 /* Exclusive/none -> shared case */
1846 if (cur_old_count == nr_old_roots &&
1847 cur_new_count < nr_new_roots) {
1848 /* Exclusive -> shared */
1849 if (cur_old_count != 0) {
1850 qg->excl -= num_bytes;
1851 qg->excl_cmpr -= num_bytes;
1856 /* Shared -> exclusive/none case */
1857 if (cur_old_count < nr_old_roots &&
1858 cur_new_count == nr_new_roots) {
1859 /* Shared->exclusive */
1860 if (cur_new_count != 0) {
1861 qg->excl += num_bytes;
1862 qg->excl_cmpr += num_bytes;
1867 /* Exclusive/none -> exclusive/none case */
1868 if (cur_old_count == nr_old_roots &&
1869 cur_new_count == nr_new_roots) {
1870 if (cur_old_count == 0) {
1871 /* None -> exclusive/none */
1873 if (cur_new_count != 0) {
1874 /* None -> exclusive */
1875 qg->excl += num_bytes;
1876 qg->excl_cmpr += num_bytes;
1879 /* None -> none, nothing changed */
1881 /* Exclusive -> exclusive/none */
1883 if (cur_new_count == 0) {
1884 /* Exclusive -> none */
1885 qg->excl -= num_bytes;
1886 qg->excl_cmpr -= num_bytes;
1889 /* Exclusive -> exclusive, nothing changed */
1894 qgroup_dirty(fs_info, qg);
1900 * Check if the @roots potentially is a list of fs tree roots
1902 * Return 0 for definitely not a fs/subvol tree roots ulist
1903 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
1906 static int maybe_fs_roots(struct ulist *roots)
1908 struct ulist_node *unode;
1909 struct ulist_iterator uiter;
1911 /* Empty one, still possible for fs roots */
1912 if (!roots || roots->nnodes == 0)
1915 ULIST_ITER_INIT(&uiter);
1916 unode = ulist_next(roots, &uiter);
1921 * If it contains fs tree roots, then it must belong to fs/subvol
1923 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
1925 return is_fstree(unode->val);
1929 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1930 struct btrfs_fs_info *fs_info,
1931 u64 bytenr, u64 num_bytes,
1932 struct ulist *old_roots, struct ulist *new_roots)
1934 struct ulist *qgroups = NULL;
1935 struct ulist *tmp = NULL;
1937 u64 nr_new_roots = 0;
1938 u64 nr_old_roots = 0;
1941 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1945 if (!maybe_fs_roots(new_roots))
1947 nr_new_roots = new_roots->nnodes;
1950 if (!maybe_fs_roots(old_roots))
1952 nr_old_roots = old_roots->nnodes;
1955 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
1956 if (nr_old_roots == 0 && nr_new_roots == 0)
1959 BUG_ON(!fs_info->quota_root);
1961 trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
1962 nr_old_roots, nr_new_roots);
1964 qgroups = ulist_alloc(GFP_NOFS);
1969 tmp = ulist_alloc(GFP_NOFS);
1975 mutex_lock(&fs_info->qgroup_rescan_lock);
1976 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
1977 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
1978 mutex_unlock(&fs_info->qgroup_rescan_lock);
1983 mutex_unlock(&fs_info->qgroup_rescan_lock);
1985 spin_lock(&fs_info->qgroup_lock);
1986 seq = fs_info->qgroup_seq;
1988 /* Update old refcnts using old_roots */
1989 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
1994 /* Update new refcnts using new_roots */
1995 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2000 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2004 * Bump qgroup_seq to avoid seq overlap
2006 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2008 spin_unlock(&fs_info->qgroup_lock);
2011 ulist_free(qgroups);
2012 ulist_free(old_roots);
2013 ulist_free(new_roots);
2017 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2019 struct btrfs_fs_info *fs_info = trans->fs_info;
2020 struct btrfs_qgroup_extent_record *record;
2021 struct btrfs_delayed_ref_root *delayed_refs;
2022 struct ulist *new_roots = NULL;
2023 struct rb_node *node;
2027 delayed_refs = &trans->transaction->delayed_refs;
2028 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2029 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2030 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2033 trace_btrfs_qgroup_account_extents(fs_info, record);
2037 * Old roots should be searched when inserting qgroup
2040 if (WARN_ON(!record->old_roots)) {
2041 /* Search commit root to find old_roots */
2042 ret = btrfs_find_all_roots(NULL, fs_info,
2044 &record->old_roots, false);
2050 * Use SEQ_LAST as time_seq to do special search, which
2051 * doesn't lock tree or delayed_refs and search current
2052 * root. It's safe inside commit_transaction().
2054 ret = btrfs_find_all_roots(trans, fs_info,
2055 record->bytenr, SEQ_LAST, &new_roots, false);
2058 if (qgroup_to_skip) {
2059 ulist_del(new_roots, qgroup_to_skip, 0);
2060 ulist_del(record->old_roots, qgroup_to_skip,
2063 ret = btrfs_qgroup_account_extent(trans, fs_info,
2064 record->bytenr, record->num_bytes,
2065 record->old_roots, new_roots);
2066 record->old_roots = NULL;
2070 ulist_free(record->old_roots);
2071 ulist_free(new_roots);
2073 rb_erase(node, &delayed_refs->dirty_extent_root);
2081 * called from commit_transaction. Writes all changed qgroups to disk.
2083 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2084 struct btrfs_fs_info *fs_info)
2086 struct btrfs_root *quota_root = fs_info->quota_root;
2092 spin_lock(&fs_info->qgroup_lock);
2093 while (!list_empty(&fs_info->dirty_qgroups)) {
2094 struct btrfs_qgroup *qgroup;
2095 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2096 struct btrfs_qgroup, dirty);
2097 list_del_init(&qgroup->dirty);
2098 spin_unlock(&fs_info->qgroup_lock);
2099 ret = update_qgroup_info_item(trans, quota_root, qgroup);
2101 fs_info->qgroup_flags |=
2102 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2103 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2105 fs_info->qgroup_flags |=
2106 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2107 spin_lock(&fs_info->qgroup_lock);
2109 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2110 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2112 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2113 spin_unlock(&fs_info->qgroup_lock);
2115 ret = update_qgroup_status_item(trans, fs_info, quota_root);
2117 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2123 * Copy the accounting information between qgroups. This is necessary
2124 * when a snapshot or a subvolume is created. Throwing an error will
2125 * cause a transaction abort so we take extra care here to only error
2126 * when a readonly fs is a reasonable outcome.
2128 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2129 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2130 struct btrfs_qgroup_inherit *inherit)
2135 struct btrfs_root *quota_root = fs_info->quota_root;
2136 struct btrfs_qgroup *srcgroup;
2137 struct btrfs_qgroup *dstgroup;
2141 mutex_lock(&fs_info->qgroup_ioctl_lock);
2142 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2151 i_qgroups = (u64 *)(inherit + 1);
2152 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2153 2 * inherit->num_excl_copies;
2154 for (i = 0; i < nums; ++i) {
2155 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2158 * Zero out invalid groups so we can ignore
2162 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2170 * create a tracking group for the subvol itself
2172 ret = add_qgroup_item(trans, quota_root, objectid);
2177 struct btrfs_root *srcroot;
2178 struct btrfs_key srckey;
2180 srckey.objectid = srcid;
2181 srckey.type = BTRFS_ROOT_ITEM_KEY;
2182 srckey.offset = (u64)-1;
2183 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2184 if (IS_ERR(srcroot)) {
2185 ret = PTR_ERR(srcroot);
2189 level_size = fs_info->nodesize;
2193 * add qgroup to all inherited groups
2196 i_qgroups = (u64 *)(inherit + 1);
2197 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2198 if (*i_qgroups == 0)
2200 ret = add_qgroup_relation_item(trans, quota_root,
2201 objectid, *i_qgroups);
2202 if (ret && ret != -EEXIST)
2204 ret = add_qgroup_relation_item(trans, quota_root,
2205 *i_qgroups, objectid);
2206 if (ret && ret != -EEXIST)
2213 spin_lock(&fs_info->qgroup_lock);
2215 dstgroup = add_qgroup_rb(fs_info, objectid);
2216 if (IS_ERR(dstgroup)) {
2217 ret = PTR_ERR(dstgroup);
2221 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2222 dstgroup->lim_flags = inherit->lim.flags;
2223 dstgroup->max_rfer = inherit->lim.max_rfer;
2224 dstgroup->max_excl = inherit->lim.max_excl;
2225 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2226 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2228 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2230 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2232 "unable to update quota limit for %llu",
2233 dstgroup->qgroupid);
2239 srcgroup = find_qgroup_rb(fs_info, srcid);
2244 * We call inherit after we clone the root in order to make sure
2245 * our counts don't go crazy, so at this point the only
2246 * difference between the two roots should be the root node.
2248 dstgroup->rfer = srcgroup->rfer;
2249 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2250 dstgroup->excl = level_size;
2251 dstgroup->excl_cmpr = level_size;
2252 srcgroup->excl = level_size;
2253 srcgroup->excl_cmpr = level_size;
2255 /* inherit the limit info */
2256 dstgroup->lim_flags = srcgroup->lim_flags;
2257 dstgroup->max_rfer = srcgroup->max_rfer;
2258 dstgroup->max_excl = srcgroup->max_excl;
2259 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2260 dstgroup->rsv_excl = srcgroup->rsv_excl;
2262 qgroup_dirty(fs_info, dstgroup);
2263 qgroup_dirty(fs_info, srcgroup);
2269 i_qgroups = (u64 *)(inherit + 1);
2270 for (i = 0; i < inherit->num_qgroups; ++i) {
2272 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2279 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2280 struct btrfs_qgroup *src;
2281 struct btrfs_qgroup *dst;
2283 if (!i_qgroups[0] || !i_qgroups[1])
2286 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2287 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2294 dst->rfer = src->rfer - level_size;
2295 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2297 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2298 struct btrfs_qgroup *src;
2299 struct btrfs_qgroup *dst;
2301 if (!i_qgroups[0] || !i_qgroups[1])
2304 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2305 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2312 dst->excl = src->excl + level_size;
2313 dst->excl_cmpr = src->excl_cmpr + level_size;
2317 spin_unlock(&fs_info->qgroup_lock);
2319 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2323 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2325 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2326 qg->reserved + (s64)qg->rfer + num_bytes > qg->max_rfer)
2329 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2330 qg->reserved + (s64)qg->excl + num_bytes > qg->max_excl)
2336 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce)
2338 struct btrfs_root *quota_root;
2339 struct btrfs_qgroup *qgroup;
2340 struct btrfs_fs_info *fs_info = root->fs_info;
2341 u64 ref_root = root->root_key.objectid;
2344 struct ulist_node *unode;
2345 struct ulist_iterator uiter;
2347 if (!is_fstree(ref_root))
2353 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2354 capable(CAP_SYS_RESOURCE))
2358 spin_lock(&fs_info->qgroup_lock);
2359 quota_root = fs_info->quota_root;
2363 qgroup = find_qgroup_rb(fs_info, ref_root);
2368 * in a first step, we check all affected qgroups if any limits would
2371 ulist_reinit(fs_info->qgroup_ulist);
2372 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2373 (uintptr_t)qgroup, GFP_ATOMIC);
2376 ULIST_ITER_INIT(&uiter);
2377 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2378 struct btrfs_qgroup *qg;
2379 struct btrfs_qgroup_list *glist;
2381 qg = unode_aux_to_qgroup(unode);
2383 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
2385 * Commit the tree and retry, since we may have
2386 * deletions which would free up space.
2388 if (!retried && qg->reserved > 0) {
2389 struct btrfs_trans_handle *trans;
2391 spin_unlock(&fs_info->qgroup_lock);
2392 ret = btrfs_start_delalloc_inodes(root, 0);
2395 btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
2396 trans = btrfs_join_transaction(root);
2398 return PTR_ERR(trans);
2399 ret = btrfs_commit_transaction(trans);
2409 list_for_each_entry(glist, &qg->groups, next_group) {
2410 ret = ulist_add(fs_info->qgroup_ulist,
2411 glist->group->qgroupid,
2412 (uintptr_t)glist->group, GFP_ATOMIC);
2419 * no limits exceeded, now record the reservation into all qgroups
2421 ULIST_ITER_INIT(&uiter);
2422 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2423 struct btrfs_qgroup *qg;
2425 qg = unode_aux_to_qgroup(unode);
2427 trace_qgroup_update_reserve(fs_info, qg, num_bytes);
2428 qg->reserved += num_bytes;
2432 spin_unlock(&fs_info->qgroup_lock);
2436 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2437 u64 ref_root, u64 num_bytes)
2439 struct btrfs_root *quota_root;
2440 struct btrfs_qgroup *qgroup;
2441 struct ulist_node *unode;
2442 struct ulist_iterator uiter;
2445 if (!is_fstree(ref_root))
2451 spin_lock(&fs_info->qgroup_lock);
2453 quota_root = fs_info->quota_root;
2457 qgroup = find_qgroup_rb(fs_info, ref_root);
2461 ulist_reinit(fs_info->qgroup_ulist);
2462 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2463 (uintptr_t)qgroup, GFP_ATOMIC);
2466 ULIST_ITER_INIT(&uiter);
2467 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2468 struct btrfs_qgroup *qg;
2469 struct btrfs_qgroup_list *glist;
2471 qg = unode_aux_to_qgroup(unode);
2473 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes);
2474 if (qg->reserved < num_bytes)
2475 report_reserved_underflow(fs_info, qg, num_bytes);
2477 qg->reserved -= num_bytes;
2479 list_for_each_entry(glist, &qg->groups, next_group) {
2480 ret = ulist_add(fs_info->qgroup_ulist,
2481 glist->group->qgroupid,
2482 (uintptr_t)glist->group, GFP_ATOMIC);
2489 spin_unlock(&fs_info->qgroup_lock);
2493 * returns < 0 on error, 0 when more leafs are to be scanned.
2494 * returns 1 when done.
2497 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2498 struct btrfs_trans_handle *trans)
2500 struct btrfs_key found;
2501 struct extent_buffer *scratch_leaf = NULL;
2502 struct ulist *roots = NULL;
2503 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2508 mutex_lock(&fs_info->qgroup_rescan_lock);
2509 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2510 &fs_info->qgroup_rescan_progress,
2513 btrfs_debug(fs_info,
2514 "current progress key (%llu %u %llu), search_slot ret %d",
2515 fs_info->qgroup_rescan_progress.objectid,
2516 fs_info->qgroup_rescan_progress.type,
2517 fs_info->qgroup_rescan_progress.offset, ret);
2521 * The rescan is about to end, we will not be scanning any
2522 * further blocks. We cannot unset the RESCAN flag here, because
2523 * we want to commit the transaction if everything went well.
2524 * To make the live accounting work in this phase, we set our
2525 * scan progress pointer such that every real extent objectid
2528 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2529 btrfs_release_path(path);
2530 mutex_unlock(&fs_info->qgroup_rescan_lock);
2534 btrfs_item_key_to_cpu(path->nodes[0], &found,
2535 btrfs_header_nritems(path->nodes[0]) - 1);
2536 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2538 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2539 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2540 if (!scratch_leaf) {
2542 mutex_unlock(&fs_info->qgroup_rescan_lock);
2545 extent_buffer_get(scratch_leaf);
2546 btrfs_tree_read_lock(scratch_leaf);
2547 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2548 slot = path->slots[0];
2549 btrfs_release_path(path);
2550 mutex_unlock(&fs_info->qgroup_rescan_lock);
2552 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2553 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2554 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2555 found.type != BTRFS_METADATA_ITEM_KEY)
2557 if (found.type == BTRFS_METADATA_ITEM_KEY)
2558 num_bytes = fs_info->nodesize;
2560 num_bytes = found.offset;
2562 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2566 /* For rescan, just pass old_roots as NULL */
2567 ret = btrfs_qgroup_account_extent(trans, fs_info,
2568 found.objectid, num_bytes, NULL, roots);
2574 btrfs_tree_read_unlock_blocking(scratch_leaf);
2575 free_extent_buffer(scratch_leaf);
2577 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2582 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2584 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2585 qgroup_rescan_work);
2586 struct btrfs_path *path;
2587 struct btrfs_trans_handle *trans = NULL;
2591 path = btrfs_alloc_path();
2596 while (!err && !btrfs_fs_closing(fs_info)) {
2597 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2598 if (IS_ERR(trans)) {
2599 err = PTR_ERR(trans);
2602 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2605 err = qgroup_rescan_leaf(fs_info, path, trans);
2608 btrfs_commit_transaction(trans);
2610 btrfs_end_transaction(trans);
2614 btrfs_free_path(path);
2616 mutex_lock(&fs_info->qgroup_rescan_lock);
2617 if (!btrfs_fs_closing(fs_info))
2618 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2621 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2622 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2623 } else if (err < 0) {
2624 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2626 mutex_unlock(&fs_info->qgroup_rescan_lock);
2629 * only update status, since the previous part has already updated the
2632 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2633 if (IS_ERR(trans)) {
2634 err = PTR_ERR(trans);
2636 "fail to start transaction for status update: %d",
2640 ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2643 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2645 btrfs_end_transaction(trans);
2647 if (btrfs_fs_closing(fs_info)) {
2648 btrfs_info(fs_info, "qgroup scan paused");
2649 } else if (err >= 0) {
2650 btrfs_info(fs_info, "qgroup scan completed%s",
2651 err > 0 ? " (inconsistency flag cleared)" : "");
2653 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2657 mutex_lock(&fs_info->qgroup_rescan_lock);
2658 fs_info->qgroup_rescan_running = false;
2659 mutex_unlock(&fs_info->qgroup_rescan_lock);
2660 complete_all(&fs_info->qgroup_rescan_completion);
2664 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2665 * memory required for the rescan context.
2668 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2674 (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2675 !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2680 mutex_lock(&fs_info->qgroup_rescan_lock);
2681 spin_lock(&fs_info->qgroup_lock);
2684 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2686 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2690 spin_unlock(&fs_info->qgroup_lock);
2691 mutex_unlock(&fs_info->qgroup_rescan_lock);
2694 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2697 memset(&fs_info->qgroup_rescan_progress, 0,
2698 sizeof(fs_info->qgroup_rescan_progress));
2699 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2700 init_completion(&fs_info->qgroup_rescan_completion);
2701 fs_info->qgroup_rescan_running = true;
2703 spin_unlock(&fs_info->qgroup_lock);
2704 mutex_unlock(&fs_info->qgroup_rescan_lock);
2706 memset(&fs_info->qgroup_rescan_work, 0,
2707 sizeof(fs_info->qgroup_rescan_work));
2708 btrfs_init_work(&fs_info->qgroup_rescan_work,
2709 btrfs_qgroup_rescan_helper,
2710 btrfs_qgroup_rescan_worker, NULL, NULL);
2714 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2722 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2725 struct btrfs_qgroup *qgroup;
2727 spin_lock(&fs_info->qgroup_lock);
2728 /* clear all current qgroup tracking information */
2729 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2730 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2732 qgroup->rfer_cmpr = 0;
2734 qgroup->excl_cmpr = 0;
2736 spin_unlock(&fs_info->qgroup_lock);
2740 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2743 struct btrfs_trans_handle *trans;
2745 ret = qgroup_rescan_init(fs_info, 0, 1);
2750 * We have set the rescan_progress to 0, which means no more
2751 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2752 * However, btrfs_qgroup_account_ref may be right after its call
2753 * to btrfs_find_all_roots, in which case it would still do the
2755 * To solve this, we're committing the transaction, which will
2756 * ensure we run all delayed refs and only after that, we are
2757 * going to clear all tracking information for a clean start.
2760 trans = btrfs_join_transaction(fs_info->fs_root);
2761 if (IS_ERR(trans)) {
2762 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2763 return PTR_ERR(trans);
2765 ret = btrfs_commit_transaction(trans);
2767 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2771 qgroup_rescan_zero_tracking(fs_info);
2773 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2774 &fs_info->qgroup_rescan_work);
2779 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2785 mutex_lock(&fs_info->qgroup_rescan_lock);
2786 spin_lock(&fs_info->qgroup_lock);
2787 running = fs_info->qgroup_rescan_running;
2788 spin_unlock(&fs_info->qgroup_lock);
2789 mutex_unlock(&fs_info->qgroup_rescan_lock);
2795 ret = wait_for_completion_interruptible(
2796 &fs_info->qgroup_rescan_completion);
2798 wait_for_completion(&fs_info->qgroup_rescan_completion);
2804 * this is only called from open_ctree where we're still single threaded, thus
2805 * locking is omitted here.
2808 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2810 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2811 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2812 &fs_info->qgroup_rescan_work);
2816 * Reserve qgroup space for range [start, start + len).
2818 * This function will either reserve space from related qgroups or doing
2819 * nothing if the range is already reserved.
2821 * Return 0 for successful reserve
2822 * Return <0 for error (including -EQUOT)
2824 * NOTE: this function may sleep for memory allocation.
2825 * if btrfs_qgroup_reserve_data() is called multiple times with
2826 * same @reserved, caller must ensure when error happens it's OK
2827 * to free *ALL* reserved space.
2829 int btrfs_qgroup_reserve_data(struct inode *inode,
2830 struct extent_changeset **reserved_ret, u64 start,
2833 struct btrfs_root *root = BTRFS_I(inode)->root;
2834 struct ulist_node *unode;
2835 struct ulist_iterator uiter;
2836 struct extent_changeset *reserved;
2841 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2842 !is_fstree(root->objectid) || len == 0)
2845 /* @reserved parameter is mandatory for qgroup */
2846 if (WARN_ON(!reserved_ret))
2848 if (!*reserved_ret) {
2849 *reserved_ret = extent_changeset_alloc();
2853 reserved = *reserved_ret;
2854 /* Record already reserved space */
2855 orig_reserved = reserved->bytes_changed;
2856 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2857 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
2859 /* Newly reserved space */
2860 to_reserve = reserved->bytes_changed - orig_reserved;
2861 trace_btrfs_qgroup_reserve_data(inode, start, len,
2862 to_reserve, QGROUP_RESERVE);
2865 ret = qgroup_reserve(root, to_reserve, true);
2872 /* cleanup *ALL* already reserved ranges */
2873 ULIST_ITER_INIT(&uiter);
2874 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
2875 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2876 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
2877 extent_changeset_release(reserved);
2881 /* Free ranges specified by @reserved, normally in error path */
2882 static int qgroup_free_reserved_data(struct inode *inode,
2883 struct extent_changeset *reserved, u64 start, u64 len)
2885 struct btrfs_root *root = BTRFS_I(inode)->root;
2886 struct ulist_node *unode;
2887 struct ulist_iterator uiter;
2888 struct extent_changeset changeset;
2892 extent_changeset_init(&changeset);
2893 len = round_up(start + len, root->fs_info->sectorsize);
2894 start = round_down(start, root->fs_info->sectorsize);
2896 ULIST_ITER_INIT(&uiter);
2897 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
2898 u64 range_start = unode->val;
2899 /* unode->aux is the inclusive end */
2900 u64 range_len = unode->aux - range_start + 1;
2904 extent_changeset_release(&changeset);
2906 /* Only free range in range [start, start + len) */
2907 if (range_start >= start + len ||
2908 range_start + range_len <= start)
2910 free_start = max(range_start, start);
2911 free_len = min(start + len, range_start + range_len) -
2914 * TODO: To also modify reserved->ranges_reserved to reflect
2917 * However as long as we free qgroup reserved according to
2918 * EXTENT_QGROUP_RESERVED, we won't double free.
2919 * So not need to rush.
2921 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
2922 free_start, free_start + free_len - 1,
2923 EXTENT_QGROUP_RESERVED, &changeset);
2926 freed += changeset.bytes_changed;
2928 btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed);
2931 extent_changeset_release(&changeset);
2935 static int __btrfs_qgroup_release_data(struct inode *inode,
2936 struct extent_changeset *reserved, u64 start, u64 len,
2939 struct extent_changeset changeset;
2940 int trace_op = QGROUP_RELEASE;
2943 /* In release case, we shouldn't have @reserved */
2944 WARN_ON(!free && reserved);
2945 if (free && reserved)
2946 return qgroup_free_reserved_data(inode, reserved, start, len);
2947 extent_changeset_init(&changeset);
2948 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2949 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
2954 trace_op = QGROUP_FREE;
2955 trace_btrfs_qgroup_release_data(inode, start, len,
2956 changeset.bytes_changed, trace_op);
2958 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
2959 BTRFS_I(inode)->root->objectid,
2960 changeset.bytes_changed);
2961 ret = changeset.bytes_changed;
2963 extent_changeset_release(&changeset);
2968 * Free a reserved space range from io_tree and related qgroups
2970 * Should be called when a range of pages get invalidated before reaching disk.
2971 * Or for error cleanup case.
2972 * if @reserved is given, only reserved range in [@start, @start + @len) will
2975 * For data written to disk, use btrfs_qgroup_release_data().
2977 * NOTE: This function may sleep for memory allocation.
2979 int btrfs_qgroup_free_data(struct inode *inode,
2980 struct extent_changeset *reserved, u64 start, u64 len)
2982 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
2986 * Release a reserved space range from io_tree only.
2988 * Should be called when a range of pages get written to disk and corresponding
2989 * FILE_EXTENT is inserted into corresponding root.
2991 * Since new qgroup accounting framework will only update qgroup numbers at
2992 * commit_transaction() time, its reserved space shouldn't be freed from
2995 * But we should release the range from io_tree, to allow further write to be
2998 * NOTE: This function may sleep for memory allocation.
3000 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3002 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3005 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3008 struct btrfs_fs_info *fs_info = root->fs_info;
3011 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3012 !is_fstree(root->objectid) || num_bytes == 0)
3015 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3016 trace_qgroup_meta_reserve(root, (s64)num_bytes);
3017 ret = qgroup_reserve(root, num_bytes, enforce);
3020 atomic64_add(num_bytes, &root->qgroup_meta_rsv);
3024 void btrfs_qgroup_free_meta_all(struct btrfs_root *root)
3026 struct btrfs_fs_info *fs_info = root->fs_info;
3029 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3030 !is_fstree(root->objectid))
3033 reserved = atomic64_xchg(&root->qgroup_meta_rsv, 0);
3036 trace_qgroup_meta_reserve(root, -(s64)reserved);
3037 btrfs_qgroup_free_refroot(fs_info, root->objectid, reserved);
3040 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes)
3042 struct btrfs_fs_info *fs_info = root->fs_info;
3044 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3045 !is_fstree(root->objectid))
3048 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3049 WARN_ON(atomic64_read(&root->qgroup_meta_rsv) < num_bytes);
3050 atomic64_sub(num_bytes, &root->qgroup_meta_rsv);
3051 trace_qgroup_meta_reserve(root, -(s64)num_bytes);
3052 btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes);
3056 * Check qgroup reserved space leaking, normally at destroy inode
3059 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3061 struct extent_changeset changeset;
3062 struct ulist_node *unode;
3063 struct ulist_iterator iter;
3066 extent_changeset_init(&changeset);
3067 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3068 EXTENT_QGROUP_RESERVED, &changeset);
3071 if (WARN_ON(changeset.bytes_changed)) {
3072 ULIST_ITER_INIT(&iter);
3073 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3074 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3075 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3076 inode->i_ino, unode->val, unode->aux);
3078 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3079 BTRFS_I(inode)->root->objectid,
3080 changeset.bytes_changed);
3083 extent_changeset_release(&changeset);