1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2011 STRATO. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sched/mm.h>
17 #include "transaction.h"
22 #include "extent_io.h"
24 #include "block-group.h"
26 #include "tree-mod-log.h"
28 #include "accessors.h"
29 #include "extent-tree.h"
30 #include "root-tree.h"
31 #include "tree-checker.h"
34 * Helpers to access qgroup reservation
36 * Callers should ensure the lock context and type are valid
39 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
44 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
45 ret += qgroup->rsv.values[i];
50 #ifdef CONFIG_BTRFS_DEBUG
51 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
53 if (type == BTRFS_QGROUP_RSV_DATA)
55 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
56 return "meta_pertrans";
57 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
58 return "meta_prealloc";
63 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
64 struct btrfs_qgroup *qgroup, u64 num_bytes,
65 enum btrfs_qgroup_rsv_type type)
67 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
68 qgroup->rsv.values[type] += num_bytes;
71 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
72 struct btrfs_qgroup *qgroup, u64 num_bytes,
73 enum btrfs_qgroup_rsv_type type)
75 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
76 if (qgroup->rsv.values[type] >= num_bytes) {
77 qgroup->rsv.values[type] -= num_bytes;
80 #ifdef CONFIG_BTRFS_DEBUG
82 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
83 qgroup->qgroupid, qgroup_rsv_type_str(type),
84 qgroup->rsv.values[type], num_bytes);
86 qgroup->rsv.values[type] = 0;
89 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
90 struct btrfs_qgroup *dest,
91 struct btrfs_qgroup *src)
95 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
96 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
99 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
100 struct btrfs_qgroup *dest,
101 struct btrfs_qgroup *src)
105 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
106 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
109 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
112 if (qg->old_refcnt < seq)
113 qg->old_refcnt = seq;
114 qg->old_refcnt += mod;
117 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
120 if (qg->new_refcnt < seq)
121 qg->new_refcnt = seq;
122 qg->new_refcnt += mod;
125 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
127 if (qg->old_refcnt < seq)
129 return qg->old_refcnt - seq;
132 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
134 if (qg->new_refcnt < seq)
136 return qg->new_refcnt - seq;
140 * glue structure to represent the relations between qgroups.
142 struct btrfs_qgroup_list {
143 struct list_head next_group;
144 struct list_head next_member;
145 struct btrfs_qgroup *group;
146 struct btrfs_qgroup *member;
149 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
151 return (u64)(uintptr_t)qg;
154 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
156 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
160 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
162 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
164 /* must be called with qgroup_ioctl_lock held */
165 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
168 struct rb_node *n = fs_info->qgroup_tree.rb_node;
169 struct btrfs_qgroup *qgroup;
172 qgroup = rb_entry(n, struct btrfs_qgroup, node);
173 if (qgroup->qgroupid < qgroupid)
175 else if (qgroup->qgroupid > qgroupid)
183 /* must be called with qgroup_lock held */
184 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
187 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
188 struct rb_node *parent = NULL;
189 struct btrfs_qgroup *qgroup;
193 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
195 if (qgroup->qgroupid < qgroupid)
197 else if (qgroup->qgroupid > qgroupid)
203 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
205 return ERR_PTR(-ENOMEM);
207 qgroup->qgroupid = qgroupid;
208 INIT_LIST_HEAD(&qgroup->groups);
209 INIT_LIST_HEAD(&qgroup->members);
210 INIT_LIST_HEAD(&qgroup->dirty);
212 rb_link_node(&qgroup->node, parent, p);
213 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
218 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
219 struct btrfs_qgroup *qgroup)
221 struct btrfs_qgroup_list *list;
223 list_del(&qgroup->dirty);
224 while (!list_empty(&qgroup->groups)) {
225 list = list_first_entry(&qgroup->groups,
226 struct btrfs_qgroup_list, next_group);
227 list_del(&list->next_group);
228 list_del(&list->next_member);
232 while (!list_empty(&qgroup->members)) {
233 list = list_first_entry(&qgroup->members,
234 struct btrfs_qgroup_list, next_member);
235 list_del(&list->next_group);
236 list_del(&list->next_member);
241 /* must be called with qgroup_lock held */
242 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
244 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
249 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
250 __del_qgroup_rb(fs_info, qgroup);
255 * Add relation specified by two qgroups.
257 * Must be called with qgroup_lock held.
259 * Return: 0 on success
260 * -ENOENT if one of the qgroups is NULL
263 static int __add_relation_rb(struct btrfs_qgroup *member, struct btrfs_qgroup *parent)
265 struct btrfs_qgroup_list *list;
267 if (!member || !parent)
270 list = kzalloc(sizeof(*list), GFP_ATOMIC);
274 list->group = parent;
275 list->member = member;
276 list_add_tail(&list->next_group, &member->groups);
277 list_add_tail(&list->next_member, &parent->members);
283 * Add relation specified by two qgroup ids.
285 * Must be called with qgroup_lock held.
287 * Return: 0 on success
288 * -ENOENT if one of the ids does not exist
291 static int add_relation_rb(struct btrfs_fs_info *fs_info, u64 memberid, u64 parentid)
293 struct btrfs_qgroup *member;
294 struct btrfs_qgroup *parent;
296 member = find_qgroup_rb(fs_info, memberid);
297 parent = find_qgroup_rb(fs_info, parentid);
299 return __add_relation_rb(member, parent);
302 /* Must be called with qgroup_lock held */
303 static int del_relation_rb(struct btrfs_fs_info *fs_info,
304 u64 memberid, u64 parentid)
306 struct btrfs_qgroup *member;
307 struct btrfs_qgroup *parent;
308 struct btrfs_qgroup_list *list;
310 member = find_qgroup_rb(fs_info, memberid);
311 parent = find_qgroup_rb(fs_info, parentid);
312 if (!member || !parent)
315 list_for_each_entry(list, &member->groups, next_group) {
316 if (list->group == parent) {
317 list_del(&list->next_group);
318 list_del(&list->next_member);
326 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
327 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
330 struct btrfs_qgroup *qgroup;
332 qgroup = find_qgroup_rb(fs_info, qgroupid);
335 if (qgroup->rfer != rfer || qgroup->excl != excl)
341 static void qgroup_mark_inconsistent(struct btrfs_fs_info *fs_info)
343 fs_info->qgroup_flags |= (BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT |
344 BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
345 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
349 * The full config is read in one go, only called from open_ctree()
350 * It doesn't use any locking, as at this point we're still single-threaded
352 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
354 struct btrfs_key key;
355 struct btrfs_key found_key;
356 struct btrfs_root *quota_root = fs_info->quota_root;
357 struct btrfs_path *path = NULL;
358 struct extent_buffer *l;
362 u64 rescan_progress = 0;
364 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
367 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
368 if (!fs_info->qgroup_ulist) {
373 path = btrfs_alloc_path();
379 ret = btrfs_sysfs_add_qgroups(fs_info);
382 /* default this to quota off, in case no status key is found */
383 fs_info->qgroup_flags = 0;
386 * pass 1: read status, all qgroup infos and limits
391 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
396 struct btrfs_qgroup *qgroup;
398 slot = path->slots[0];
400 btrfs_item_key_to_cpu(l, &found_key, slot);
402 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
403 struct btrfs_qgroup_status_item *ptr;
405 ptr = btrfs_item_ptr(l, slot,
406 struct btrfs_qgroup_status_item);
408 if (btrfs_qgroup_status_version(l, ptr) !=
409 BTRFS_QGROUP_STATUS_VERSION) {
411 "old qgroup version, quota disabled");
414 if (btrfs_qgroup_status_generation(l, ptr) !=
415 fs_info->generation) {
416 qgroup_mark_inconsistent(fs_info);
418 "qgroup generation mismatch, marked as inconsistent");
420 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
422 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
426 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
427 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
430 qgroup = find_qgroup_rb(fs_info, found_key.offset);
431 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
432 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
433 btrfs_err(fs_info, "inconsistent qgroup config");
434 qgroup_mark_inconsistent(fs_info);
437 qgroup = add_qgroup_rb(fs_info, found_key.offset);
438 if (IS_ERR(qgroup)) {
439 ret = PTR_ERR(qgroup);
443 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
447 switch (found_key.type) {
448 case BTRFS_QGROUP_INFO_KEY: {
449 struct btrfs_qgroup_info_item *ptr;
451 ptr = btrfs_item_ptr(l, slot,
452 struct btrfs_qgroup_info_item);
453 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
454 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
455 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
456 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
457 /* generation currently unused */
460 case BTRFS_QGROUP_LIMIT_KEY: {
461 struct btrfs_qgroup_limit_item *ptr;
463 ptr = btrfs_item_ptr(l, slot,
464 struct btrfs_qgroup_limit_item);
465 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
466 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
467 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
468 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
469 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
474 ret = btrfs_next_item(quota_root, path);
480 btrfs_release_path(path);
483 * pass 2: read all qgroup relations
486 key.type = BTRFS_QGROUP_RELATION_KEY;
488 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
492 slot = path->slots[0];
494 btrfs_item_key_to_cpu(l, &found_key, slot);
496 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
499 if (found_key.objectid > found_key.offset) {
500 /* parent <- member, not needed to build config */
501 /* FIXME should we omit the key completely? */
505 ret = add_relation_rb(fs_info, found_key.objectid,
507 if (ret == -ENOENT) {
509 "orphan qgroup relation 0x%llx->0x%llx",
510 found_key.objectid, found_key.offset);
511 ret = 0; /* ignore the error */
516 ret = btrfs_next_item(quota_root, path);
523 btrfs_free_path(path);
524 fs_info->qgroup_flags |= flags;
525 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
526 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
527 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
529 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
532 ulist_free(fs_info->qgroup_ulist);
533 fs_info->qgroup_ulist = NULL;
534 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
535 btrfs_sysfs_del_qgroups(fs_info);
538 return ret < 0 ? ret : 0;
542 * Called in close_ctree() when quota is still enabled. This verifies we don't
543 * leak some reserved space.
545 * Return false if no reserved space is left.
546 * Return true if some reserved space is leaked.
548 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
550 struct rb_node *node;
553 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
556 * Since we're unmounting, there is no race and no need to grab qgroup
557 * lock. And here we don't go post-order to provide a more user
558 * friendly sorted result.
560 for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
561 struct btrfs_qgroup *qgroup;
564 qgroup = rb_entry(node, struct btrfs_qgroup, node);
565 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
566 if (qgroup->rsv.values[i]) {
569 "qgroup %hu/%llu has unreleased space, type %d rsv %llu",
570 btrfs_qgroup_level(qgroup->qgroupid),
571 btrfs_qgroup_subvolid(qgroup->qgroupid),
572 i, qgroup->rsv.values[i]);
580 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
581 * first two are in single-threaded paths.And for the third one, we have set
582 * quota_root to be null with qgroup_lock held before, so it is safe to clean
583 * up the in-memory structures without qgroup_lock held.
585 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
588 struct btrfs_qgroup *qgroup;
590 while ((n = rb_first(&fs_info->qgroup_tree))) {
591 qgroup = rb_entry(n, struct btrfs_qgroup, node);
592 rb_erase(n, &fs_info->qgroup_tree);
593 __del_qgroup_rb(fs_info, qgroup);
594 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
598 * We call btrfs_free_qgroup_config() when unmounting
599 * filesystem and disabling quota, so we set qgroup_ulist
600 * to be null here to avoid double free.
602 ulist_free(fs_info->qgroup_ulist);
603 fs_info->qgroup_ulist = NULL;
604 btrfs_sysfs_del_qgroups(fs_info);
607 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
611 struct btrfs_root *quota_root = trans->fs_info->quota_root;
612 struct btrfs_path *path;
613 struct btrfs_key key;
615 path = btrfs_alloc_path();
620 key.type = BTRFS_QGROUP_RELATION_KEY;
623 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
625 btrfs_mark_buffer_dirty(path->nodes[0]);
627 btrfs_free_path(path);
631 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
635 struct btrfs_root *quota_root = trans->fs_info->quota_root;
636 struct btrfs_path *path;
637 struct btrfs_key key;
639 path = btrfs_alloc_path();
644 key.type = BTRFS_QGROUP_RELATION_KEY;
647 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
656 ret = btrfs_del_item(trans, quota_root, path);
658 btrfs_free_path(path);
662 static int add_qgroup_item(struct btrfs_trans_handle *trans,
663 struct btrfs_root *quota_root, u64 qgroupid)
666 struct btrfs_path *path;
667 struct btrfs_qgroup_info_item *qgroup_info;
668 struct btrfs_qgroup_limit_item *qgroup_limit;
669 struct extent_buffer *leaf;
670 struct btrfs_key key;
672 if (btrfs_is_testing(quota_root->fs_info))
675 path = btrfs_alloc_path();
680 key.type = BTRFS_QGROUP_INFO_KEY;
681 key.offset = qgroupid;
684 * Avoid a transaction abort by catching -EEXIST here. In that
685 * case, we proceed by re-initializing the existing structure
689 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
690 sizeof(*qgroup_info));
691 if (ret && ret != -EEXIST)
694 leaf = path->nodes[0];
695 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
696 struct btrfs_qgroup_info_item);
697 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
698 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
699 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
700 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
701 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
703 btrfs_mark_buffer_dirty(leaf);
705 btrfs_release_path(path);
707 key.type = BTRFS_QGROUP_LIMIT_KEY;
708 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
709 sizeof(*qgroup_limit));
710 if (ret && ret != -EEXIST)
713 leaf = path->nodes[0];
714 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
715 struct btrfs_qgroup_limit_item);
716 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
717 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
718 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
719 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
720 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
722 btrfs_mark_buffer_dirty(leaf);
726 btrfs_free_path(path);
730 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
733 struct btrfs_root *quota_root = trans->fs_info->quota_root;
734 struct btrfs_path *path;
735 struct btrfs_key key;
737 path = btrfs_alloc_path();
742 key.type = BTRFS_QGROUP_INFO_KEY;
743 key.offset = qgroupid;
744 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
753 ret = btrfs_del_item(trans, quota_root, path);
757 btrfs_release_path(path);
759 key.type = BTRFS_QGROUP_LIMIT_KEY;
760 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
769 ret = btrfs_del_item(trans, quota_root, path);
772 btrfs_free_path(path);
776 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
777 struct btrfs_qgroup *qgroup)
779 struct btrfs_root *quota_root = trans->fs_info->quota_root;
780 struct btrfs_path *path;
781 struct btrfs_key key;
782 struct extent_buffer *l;
783 struct btrfs_qgroup_limit_item *qgroup_limit;
788 key.type = BTRFS_QGROUP_LIMIT_KEY;
789 key.offset = qgroup->qgroupid;
791 path = btrfs_alloc_path();
795 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
803 slot = path->slots[0];
804 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
805 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
806 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
807 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
808 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
809 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
811 btrfs_mark_buffer_dirty(l);
814 btrfs_free_path(path);
818 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
819 struct btrfs_qgroup *qgroup)
821 struct btrfs_fs_info *fs_info = trans->fs_info;
822 struct btrfs_root *quota_root = fs_info->quota_root;
823 struct btrfs_path *path;
824 struct btrfs_key key;
825 struct extent_buffer *l;
826 struct btrfs_qgroup_info_item *qgroup_info;
830 if (btrfs_is_testing(fs_info))
834 key.type = BTRFS_QGROUP_INFO_KEY;
835 key.offset = qgroup->qgroupid;
837 path = btrfs_alloc_path();
841 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
849 slot = path->slots[0];
850 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
851 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
852 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
853 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
854 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
855 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
857 btrfs_mark_buffer_dirty(l);
860 btrfs_free_path(path);
864 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
866 struct btrfs_fs_info *fs_info = trans->fs_info;
867 struct btrfs_root *quota_root = fs_info->quota_root;
868 struct btrfs_path *path;
869 struct btrfs_key key;
870 struct extent_buffer *l;
871 struct btrfs_qgroup_status_item *ptr;
876 key.type = BTRFS_QGROUP_STATUS_KEY;
879 path = btrfs_alloc_path();
883 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
891 slot = path->slots[0];
892 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
893 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags &
894 BTRFS_QGROUP_STATUS_FLAGS_MASK);
895 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
896 btrfs_set_qgroup_status_rescan(l, ptr,
897 fs_info->qgroup_rescan_progress.objectid);
899 btrfs_mark_buffer_dirty(l);
902 btrfs_free_path(path);
907 * called with qgroup_lock held
909 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
910 struct btrfs_root *root)
912 struct btrfs_path *path;
913 struct btrfs_key key;
914 struct extent_buffer *leaf = NULL;
918 path = btrfs_alloc_path();
927 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
930 leaf = path->nodes[0];
931 nr = btrfs_header_nritems(leaf);
935 * delete the leaf one by one
936 * since the whole tree is going
940 ret = btrfs_del_items(trans, root, path, 0, nr);
944 btrfs_release_path(path);
948 btrfs_free_path(path);
952 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
954 struct btrfs_root *quota_root;
955 struct btrfs_root *tree_root = fs_info->tree_root;
956 struct btrfs_path *path = NULL;
957 struct btrfs_qgroup_status_item *ptr;
958 struct extent_buffer *leaf;
959 struct btrfs_key key;
960 struct btrfs_key found_key;
961 struct btrfs_qgroup *qgroup = NULL;
962 struct btrfs_trans_handle *trans = NULL;
963 struct ulist *ulist = NULL;
968 * We need to have subvol_sem write locked, to prevent races between
969 * concurrent tasks trying to enable quotas, because we will unlock
970 * and relock qgroup_ioctl_lock before setting fs_info->quota_root
971 * and before setting BTRFS_FS_QUOTA_ENABLED.
973 lockdep_assert_held_write(&fs_info->subvol_sem);
975 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
977 "qgroups are currently unsupported in extent tree v2");
981 mutex_lock(&fs_info->qgroup_ioctl_lock);
982 if (fs_info->quota_root)
985 ulist = ulist_alloc(GFP_KERNEL);
991 ret = btrfs_sysfs_add_qgroups(fs_info);
996 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
997 * avoid lock acquisition inversion problems (reported by lockdep) between
998 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
999 * start a transaction.
1000 * After we started the transaction lock qgroup_ioctl_lock again and
1001 * check if someone else created the quota root in the meanwhile. If so,
1002 * just return success and release the transaction handle.
1004 * Also we don't need to worry about someone else calling
1005 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
1006 * that function returns 0 (success) when the sysfs entries already exist.
1008 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1011 * 1 for quota root item
1012 * 1 for BTRFS_QGROUP_STATUS item
1014 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
1015 * per subvolume. However those are not currently reserved since it
1016 * would be a lot of overkill.
1018 trans = btrfs_start_transaction(tree_root, 2);
1020 mutex_lock(&fs_info->qgroup_ioctl_lock);
1021 if (IS_ERR(trans)) {
1022 ret = PTR_ERR(trans);
1027 if (fs_info->quota_root)
1030 fs_info->qgroup_ulist = ulist;
1034 * initially create the quota tree
1036 quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
1037 if (IS_ERR(quota_root)) {
1038 ret = PTR_ERR(quota_root);
1039 btrfs_abort_transaction(trans, ret);
1043 path = btrfs_alloc_path();
1046 btrfs_abort_transaction(trans, ret);
1051 key.type = BTRFS_QGROUP_STATUS_KEY;
1054 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1057 btrfs_abort_transaction(trans, ret);
1061 leaf = path->nodes[0];
1062 ptr = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_qgroup_status_item);
1064 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1065 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1066 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1067 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1068 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
1069 BTRFS_QGROUP_STATUS_FLAGS_MASK);
1070 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1072 btrfs_mark_buffer_dirty(leaf);
1075 key.type = BTRFS_ROOT_REF_KEY;
1078 btrfs_release_path(path);
1079 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1083 btrfs_abort_transaction(trans, ret);
1088 slot = path->slots[0];
1089 leaf = path->nodes[0];
1090 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1092 if (found_key.type == BTRFS_ROOT_REF_KEY) {
1094 /* Release locks on tree_root before we access quota_root */
1095 btrfs_release_path(path);
1097 ret = add_qgroup_item(trans, quota_root,
1100 btrfs_abort_transaction(trans, ret);
1104 qgroup = add_qgroup_rb(fs_info, found_key.offset);
1105 if (IS_ERR(qgroup)) {
1106 ret = PTR_ERR(qgroup);
1107 btrfs_abort_transaction(trans, ret);
1110 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1112 btrfs_abort_transaction(trans, ret);
1115 ret = btrfs_search_slot_for_read(tree_root, &found_key,
1118 btrfs_abort_transaction(trans, ret);
1123 * Shouldn't happen, but in case it does we
1124 * don't need to do the btrfs_next_item, just
1130 ret = btrfs_next_item(tree_root, path);
1132 btrfs_abort_transaction(trans, ret);
1140 btrfs_release_path(path);
1141 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1143 btrfs_abort_transaction(trans, ret);
1147 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1148 if (IS_ERR(qgroup)) {
1149 ret = PTR_ERR(qgroup);
1150 btrfs_abort_transaction(trans, ret);
1153 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1155 btrfs_abort_transaction(trans, ret);
1159 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1161 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
1162 * a deadlock with tasks concurrently doing other qgroup operations, such
1163 * adding/removing qgroups or adding/deleting qgroup relations for example,
1164 * because all qgroup operations first start or join a transaction and then
1165 * lock the qgroup_ioctl_lock mutex.
1166 * We are safe from a concurrent task trying to enable quotas, by calling
1167 * this function, since we are serialized by fs_info->subvol_sem.
1169 ret = btrfs_commit_transaction(trans);
1171 mutex_lock(&fs_info->qgroup_ioctl_lock);
1176 * Set quota enabled flag after committing the transaction, to avoid
1177 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1180 spin_lock(&fs_info->qgroup_lock);
1181 fs_info->quota_root = quota_root;
1182 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1183 spin_unlock(&fs_info->qgroup_lock);
1185 ret = qgroup_rescan_init(fs_info, 0, 1);
1187 qgroup_rescan_zero_tracking(fs_info);
1188 fs_info->qgroup_rescan_running = true;
1189 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1190 &fs_info->qgroup_rescan_work);
1193 * We have set both BTRFS_FS_QUOTA_ENABLED and
1194 * BTRFS_QGROUP_STATUS_FLAG_ON, so we can only fail with
1195 * -EINPROGRESS. That can happen because someone started the
1196 * rescan worker by calling quota rescan ioctl before we
1197 * attempted to initialize the rescan worker. Failure due to
1198 * quotas disabled in the meanwhile is not possible, because
1199 * we are holding a write lock on fs_info->subvol_sem, which
1200 * is also acquired when disabling quotas.
1201 * Ignore such error, and any other error would need to undo
1202 * everything we did in the transaction we just committed.
1204 ASSERT(ret == -EINPROGRESS);
1209 btrfs_free_path(path);
1212 btrfs_put_root(quota_root);
1215 ulist_free(fs_info->qgroup_ulist);
1216 fs_info->qgroup_ulist = NULL;
1217 btrfs_sysfs_del_qgroups(fs_info);
1219 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1221 btrfs_end_transaction(trans);
1223 ret = btrfs_end_transaction(trans);
1228 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1230 struct btrfs_root *quota_root;
1231 struct btrfs_trans_handle *trans = NULL;
1235 * We need to have subvol_sem write locked, to prevent races between
1236 * concurrent tasks trying to disable quotas, because we will unlock
1237 * and relock qgroup_ioctl_lock across BTRFS_FS_QUOTA_ENABLED changes.
1239 lockdep_assert_held_write(&fs_info->subvol_sem);
1241 mutex_lock(&fs_info->qgroup_ioctl_lock);
1242 if (!fs_info->quota_root)
1246 * Unlock the qgroup_ioctl_lock mutex before waiting for the rescan worker to
1247 * complete. Otherwise we can deadlock because btrfs_remove_qgroup() needs
1248 * to lock that mutex while holding a transaction handle and the rescan
1249 * worker needs to commit a transaction.
1251 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1254 * Request qgroup rescan worker to complete and wait for it. This wait
1255 * must be done before transaction start for quota disable since it may
1256 * deadlock with transaction by the qgroup rescan worker.
1258 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1259 btrfs_qgroup_wait_for_completion(fs_info, false);
1262 * 1 For the root item
1264 * We should also reserve enough items for the quota tree deletion in
1265 * btrfs_clean_quota_tree but this is not done.
1267 * Also, we must always start a transaction without holding the mutex
1268 * qgroup_ioctl_lock, see btrfs_quota_enable().
1270 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1272 mutex_lock(&fs_info->qgroup_ioctl_lock);
1273 if (IS_ERR(trans)) {
1274 ret = PTR_ERR(trans);
1276 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1280 if (!fs_info->quota_root)
1283 spin_lock(&fs_info->qgroup_lock);
1284 quota_root = fs_info->quota_root;
1285 fs_info->quota_root = NULL;
1286 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1287 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1288 spin_unlock(&fs_info->qgroup_lock);
1290 btrfs_free_qgroup_config(fs_info);
1292 ret = btrfs_clean_quota_tree(trans, quota_root);
1294 btrfs_abort_transaction(trans, ret);
1298 ret = btrfs_del_root(trans, "a_root->root_key);
1300 btrfs_abort_transaction(trans, ret);
1304 list_del("a_root->dirty_list);
1306 btrfs_tree_lock(quota_root->node);
1307 btrfs_clean_tree_block(quota_root->node);
1308 btrfs_tree_unlock(quota_root->node);
1309 btrfs_free_tree_block(trans, btrfs_root_id(quota_root),
1310 quota_root->node, 0, 1);
1312 btrfs_put_root(quota_root);
1315 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1317 btrfs_end_transaction(trans);
1319 ret = btrfs_end_transaction(trans);
1324 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1325 struct btrfs_qgroup *qgroup)
1327 if (list_empty(&qgroup->dirty))
1328 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1332 * The easy accounting, we're updating qgroup relationship whose child qgroup
1333 * only has exclusive extents.
1335 * In this case, all exclusive extents will also be exclusive for parent, so
1336 * excl/rfer just get added/removed.
1338 * So is qgroup reservation space, which should also be added/removed to
1340 * Or when child tries to release reservation space, parent will underflow its
1341 * reservation (for relationship adding case).
1343 * Caller should hold fs_info->qgroup_lock.
1345 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1346 struct ulist *tmp, u64 ref_root,
1347 struct btrfs_qgroup *src, int sign)
1349 struct btrfs_qgroup *qgroup;
1350 struct btrfs_qgroup_list *glist;
1351 struct ulist_node *unode;
1352 struct ulist_iterator uiter;
1353 u64 num_bytes = src->excl;
1356 qgroup = find_qgroup_rb(fs_info, ref_root);
1360 qgroup->rfer += sign * num_bytes;
1361 qgroup->rfer_cmpr += sign * num_bytes;
1363 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1364 qgroup->excl += sign * num_bytes;
1365 qgroup->excl_cmpr += sign * num_bytes;
1368 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1370 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1372 qgroup_dirty(fs_info, qgroup);
1374 /* Get all of the parent groups that contain this qgroup */
1375 list_for_each_entry(glist, &qgroup->groups, next_group) {
1376 ret = ulist_add(tmp, glist->group->qgroupid,
1377 qgroup_to_aux(glist->group), GFP_ATOMIC);
1382 /* Iterate all of the parents and adjust their reference counts */
1383 ULIST_ITER_INIT(&uiter);
1384 while ((unode = ulist_next(tmp, &uiter))) {
1385 qgroup = unode_aux_to_qgroup(unode);
1386 qgroup->rfer += sign * num_bytes;
1387 qgroup->rfer_cmpr += sign * num_bytes;
1388 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1389 qgroup->excl += sign * num_bytes;
1391 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1393 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1394 qgroup->excl_cmpr += sign * num_bytes;
1395 qgroup_dirty(fs_info, qgroup);
1397 /* Add any parents of the parents */
1398 list_for_each_entry(glist, &qgroup->groups, next_group) {
1399 ret = ulist_add(tmp, glist->group->qgroupid,
1400 qgroup_to_aux(glist->group), GFP_ATOMIC);
1412 * Quick path for updating qgroup with only excl refs.
1414 * In that case, just update all parent will be enough.
1415 * Or we needs to do a full rescan.
1416 * Caller should also hold fs_info->qgroup_lock.
1418 * Return 0 for quick update, return >0 for need to full rescan
1419 * and mark INCONSISTENT flag.
1420 * Return < 0 for other error.
1422 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1423 struct ulist *tmp, u64 src, u64 dst,
1426 struct btrfs_qgroup *qgroup;
1430 qgroup = find_qgroup_rb(fs_info, src);
1433 if (qgroup->excl == qgroup->rfer) {
1435 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1444 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1448 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1451 struct btrfs_fs_info *fs_info = trans->fs_info;
1452 struct btrfs_qgroup *parent;
1453 struct btrfs_qgroup *member;
1454 struct btrfs_qgroup_list *list;
1456 unsigned int nofs_flag;
1459 /* Check the level of src and dst first */
1460 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1463 /* We hold a transaction handle open, must do a NOFS allocation. */
1464 nofs_flag = memalloc_nofs_save();
1465 tmp = ulist_alloc(GFP_KERNEL);
1466 memalloc_nofs_restore(nofs_flag);
1470 mutex_lock(&fs_info->qgroup_ioctl_lock);
1471 if (!fs_info->quota_root) {
1475 member = find_qgroup_rb(fs_info, src);
1476 parent = find_qgroup_rb(fs_info, dst);
1477 if (!member || !parent) {
1482 /* check if such qgroup relation exist firstly */
1483 list_for_each_entry(list, &member->groups, next_group) {
1484 if (list->group == parent) {
1490 ret = add_qgroup_relation_item(trans, src, dst);
1494 ret = add_qgroup_relation_item(trans, dst, src);
1496 del_qgroup_relation_item(trans, src, dst);
1500 spin_lock(&fs_info->qgroup_lock);
1501 ret = __add_relation_rb(member, parent);
1503 spin_unlock(&fs_info->qgroup_lock);
1506 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1507 spin_unlock(&fs_info->qgroup_lock);
1509 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1514 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1517 struct btrfs_fs_info *fs_info = trans->fs_info;
1518 struct btrfs_qgroup *parent;
1519 struct btrfs_qgroup *member;
1520 struct btrfs_qgroup_list *list;
1523 unsigned int nofs_flag;
1527 /* We hold a transaction handle open, must do a NOFS allocation. */
1528 nofs_flag = memalloc_nofs_save();
1529 tmp = ulist_alloc(GFP_KERNEL);
1530 memalloc_nofs_restore(nofs_flag);
1534 if (!fs_info->quota_root) {
1539 member = find_qgroup_rb(fs_info, src);
1540 parent = find_qgroup_rb(fs_info, dst);
1542 * The parent/member pair doesn't exist, then try to delete the dead
1543 * relation items only.
1545 if (!member || !parent)
1548 /* check if such qgroup relation exist firstly */
1549 list_for_each_entry(list, &member->groups, next_group) {
1550 if (list->group == parent) {
1557 ret = del_qgroup_relation_item(trans, src, dst);
1558 if (ret < 0 && ret != -ENOENT)
1560 ret2 = del_qgroup_relation_item(trans, dst, src);
1561 if (ret2 < 0 && ret2 != -ENOENT)
1564 /* At least one deletion succeeded, return 0 */
1569 spin_lock(&fs_info->qgroup_lock);
1570 del_relation_rb(fs_info, src, dst);
1571 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1572 spin_unlock(&fs_info->qgroup_lock);
1579 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1582 struct btrfs_fs_info *fs_info = trans->fs_info;
1585 mutex_lock(&fs_info->qgroup_ioctl_lock);
1586 ret = __del_qgroup_relation(trans, src, dst);
1587 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1592 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1594 struct btrfs_fs_info *fs_info = trans->fs_info;
1595 struct btrfs_root *quota_root;
1596 struct btrfs_qgroup *qgroup;
1599 mutex_lock(&fs_info->qgroup_ioctl_lock);
1600 if (!fs_info->quota_root) {
1604 quota_root = fs_info->quota_root;
1605 qgroup = find_qgroup_rb(fs_info, qgroupid);
1611 ret = add_qgroup_item(trans, quota_root, qgroupid);
1615 spin_lock(&fs_info->qgroup_lock);
1616 qgroup = add_qgroup_rb(fs_info, qgroupid);
1617 spin_unlock(&fs_info->qgroup_lock);
1619 if (IS_ERR(qgroup)) {
1620 ret = PTR_ERR(qgroup);
1623 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1625 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1629 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1631 struct btrfs_fs_info *fs_info = trans->fs_info;
1632 struct btrfs_qgroup *qgroup;
1633 struct btrfs_qgroup_list *list;
1636 mutex_lock(&fs_info->qgroup_ioctl_lock);
1637 if (!fs_info->quota_root) {
1642 qgroup = find_qgroup_rb(fs_info, qgroupid);
1648 /* Check if there are no children of this qgroup */
1649 if (!list_empty(&qgroup->members)) {
1654 ret = del_qgroup_item(trans, qgroupid);
1655 if (ret && ret != -ENOENT)
1658 while (!list_empty(&qgroup->groups)) {
1659 list = list_first_entry(&qgroup->groups,
1660 struct btrfs_qgroup_list, next_group);
1661 ret = __del_qgroup_relation(trans, qgroupid,
1662 list->group->qgroupid);
1667 spin_lock(&fs_info->qgroup_lock);
1668 del_qgroup_rb(fs_info, qgroupid);
1669 spin_unlock(&fs_info->qgroup_lock);
1672 * Remove the qgroup from sysfs now without holding the qgroup_lock
1673 * spinlock, since the sysfs_remove_group() function needs to take
1674 * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1676 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
1679 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1683 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1684 struct btrfs_qgroup_limit *limit)
1686 struct btrfs_fs_info *fs_info = trans->fs_info;
1687 struct btrfs_qgroup *qgroup;
1689 /* Sometimes we would want to clear the limit on this qgroup.
1690 * To meet this requirement, we treat the -1 as a special value
1691 * which tell kernel to clear the limit on this qgroup.
1693 const u64 CLEAR_VALUE = -1;
1695 mutex_lock(&fs_info->qgroup_ioctl_lock);
1696 if (!fs_info->quota_root) {
1701 qgroup = find_qgroup_rb(fs_info, qgroupid);
1707 spin_lock(&fs_info->qgroup_lock);
1708 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1709 if (limit->max_rfer == CLEAR_VALUE) {
1710 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1711 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1712 qgroup->max_rfer = 0;
1714 qgroup->max_rfer = limit->max_rfer;
1717 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1718 if (limit->max_excl == CLEAR_VALUE) {
1719 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1720 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1721 qgroup->max_excl = 0;
1723 qgroup->max_excl = limit->max_excl;
1726 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1727 if (limit->rsv_rfer == CLEAR_VALUE) {
1728 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1729 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1730 qgroup->rsv_rfer = 0;
1732 qgroup->rsv_rfer = limit->rsv_rfer;
1735 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1736 if (limit->rsv_excl == CLEAR_VALUE) {
1737 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1738 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1739 qgroup->rsv_excl = 0;
1741 qgroup->rsv_excl = limit->rsv_excl;
1744 qgroup->lim_flags |= limit->flags;
1746 spin_unlock(&fs_info->qgroup_lock);
1748 ret = update_qgroup_limit_item(trans, qgroup);
1750 qgroup_mark_inconsistent(fs_info);
1751 btrfs_info(fs_info, "unable to update quota limit for %llu",
1756 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1760 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1761 struct btrfs_delayed_ref_root *delayed_refs,
1762 struct btrfs_qgroup_extent_record *record)
1764 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1765 struct rb_node *parent_node = NULL;
1766 struct btrfs_qgroup_extent_record *entry;
1767 u64 bytenr = record->bytenr;
1769 lockdep_assert_held(&delayed_refs->lock);
1770 trace_btrfs_qgroup_trace_extent(fs_info, record);
1774 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1776 if (bytenr < entry->bytenr) {
1778 } else if (bytenr > entry->bytenr) {
1779 p = &(*p)->rb_right;
1781 if (record->data_rsv && !entry->data_rsv) {
1782 entry->data_rsv = record->data_rsv;
1783 entry->data_rsv_refroot =
1784 record->data_rsv_refroot;
1790 rb_link_node(&record->node, parent_node, p);
1791 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1795 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
1796 struct btrfs_qgroup_extent_record *qrecord)
1798 struct btrfs_backref_walk_ctx ctx = { 0 };
1802 * We are always called in a context where we are already holding a
1803 * transaction handle. Often we are called when adding a data delayed
1804 * reference from btrfs_truncate_inode_items() (truncating or unlinking),
1805 * in which case we will be holding a write lock on extent buffer from a
1806 * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
1807 * acquire fs_info->commit_root_sem, because that is a higher level lock
1808 * that must be acquired before locking any extent buffers.
1810 * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
1811 * but we can't pass it a non-NULL transaction handle, because otherwise
1812 * it would not use commit roots and would lock extent buffers, causing
1813 * a deadlock if it ends up trying to read lock the same extent buffer
1814 * that was previously write locked at btrfs_truncate_inode_items().
1816 * So pass a NULL transaction handle to btrfs_find_all_roots() and
1817 * explicitly tell it to not acquire the commit_root_sem - if we are
1818 * holding a transaction handle we don't need its protection.
1820 ASSERT(trans != NULL);
1822 if (trans->fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
1825 ctx.bytenr = qrecord->bytenr;
1826 ctx.fs_info = trans->fs_info;
1828 ret = btrfs_find_all_roots(&ctx, true);
1830 qgroup_mark_inconsistent(trans->fs_info);
1831 btrfs_warn(trans->fs_info,
1832 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1838 * Here we don't need to get the lock of
1839 * trans->transaction->delayed_refs, since inserted qrecord won't
1840 * be deleted, only qrecord->node may be modified (new qrecord insert)
1842 * So modifying qrecord->old_roots is safe here
1844 qrecord->old_roots = ctx.roots;
1848 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1851 struct btrfs_fs_info *fs_info = trans->fs_info;
1852 struct btrfs_qgroup_extent_record *record;
1853 struct btrfs_delayed_ref_root *delayed_refs;
1856 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1857 || bytenr == 0 || num_bytes == 0)
1859 record = kzalloc(sizeof(*record), GFP_NOFS);
1863 delayed_refs = &trans->transaction->delayed_refs;
1864 record->bytenr = bytenr;
1865 record->num_bytes = num_bytes;
1866 record->old_roots = NULL;
1868 spin_lock(&delayed_refs->lock);
1869 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1870 spin_unlock(&delayed_refs->lock);
1875 return btrfs_qgroup_trace_extent_post(trans, record);
1878 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1879 struct extent_buffer *eb)
1881 struct btrfs_fs_info *fs_info = trans->fs_info;
1882 int nr = btrfs_header_nritems(eb);
1883 int i, extent_type, ret;
1884 struct btrfs_key key;
1885 struct btrfs_file_extent_item *fi;
1886 u64 bytenr, num_bytes;
1888 /* We can be called directly from walk_up_proc() */
1889 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1892 for (i = 0; i < nr; i++) {
1893 btrfs_item_key_to_cpu(eb, &key, i);
1895 if (key.type != BTRFS_EXTENT_DATA_KEY)
1898 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1899 /* filter out non qgroup-accountable extents */
1900 extent_type = btrfs_file_extent_type(eb, fi);
1902 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1905 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1909 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1911 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes);
1920 * Walk up the tree from the bottom, freeing leaves and any interior
1921 * nodes which have had all slots visited. If a node (leaf or
1922 * interior) is freed, the node above it will have it's slot
1923 * incremented. The root node will never be freed.
1925 * At the end of this function, we should have a path which has all
1926 * slots incremented to the next position for a search. If we need to
1927 * read a new node it will be NULL and the node above it will have the
1928 * correct slot selected for a later read.
1930 * If we increment the root nodes slot counter past the number of
1931 * elements, 1 is returned to signal completion of the search.
1933 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1937 struct extent_buffer *eb;
1939 if (root_level == 0)
1942 while (level <= root_level) {
1943 eb = path->nodes[level];
1944 nr = btrfs_header_nritems(eb);
1945 path->slots[level]++;
1946 slot = path->slots[level];
1947 if (slot >= nr || level == 0) {
1949 * Don't free the root - we will detect this
1950 * condition after our loop and return a
1951 * positive value for caller to stop walking the tree.
1953 if (level != root_level) {
1954 btrfs_tree_unlock_rw(eb, path->locks[level]);
1955 path->locks[level] = 0;
1957 free_extent_buffer(eb);
1958 path->nodes[level] = NULL;
1959 path->slots[level] = 0;
1963 * We have a valid slot to walk back down
1964 * from. Stop here so caller can process these
1973 eb = path->nodes[root_level];
1974 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1981 * Helper function to trace a subtree tree block swap.
1983 * The swap will happen in highest tree block, but there may be a lot of
1984 * tree blocks involved.
1987 * OO = Old tree blocks
1988 * NN = New tree blocks allocated during balance
1990 * File tree (257) Reloc tree for 257
1993 * L1 OO OO (a) OO NN (a)
1995 * L0 OO OO OO OO OO OO NN NN
1998 * When calling qgroup_trace_extent_swap(), we will pass:
2000 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
2004 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
2005 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
2007 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
2009 * 1) Tree search from @src_eb
2010 * It should acts as a simplified btrfs_search_slot().
2011 * The key for search can be extracted from @dst_path->nodes[dst_level]
2014 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
2015 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
2016 * They should be marked during previous (@dst_level = 1) iteration.
2018 * 3) Mark file extents in leaves dirty
2019 * We don't have good way to pick out new file extents only.
2020 * So we still follow the old method by scanning all file extents in
2023 * This function can free us from keeping two paths, thus later we only need
2024 * to care about how to iterate all new tree blocks in reloc tree.
2026 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
2027 struct extent_buffer *src_eb,
2028 struct btrfs_path *dst_path,
2029 int dst_level, int root_level,
2032 struct btrfs_key key;
2033 struct btrfs_path *src_path;
2034 struct btrfs_fs_info *fs_info = trans->fs_info;
2035 u32 nodesize = fs_info->nodesize;
2036 int cur_level = root_level;
2039 BUG_ON(dst_level > root_level);
2040 /* Level mismatch */
2041 if (btrfs_header_level(src_eb) != root_level)
2044 src_path = btrfs_alloc_path();
2051 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2053 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2056 atomic_inc(&src_eb->refs);
2057 src_path->nodes[root_level] = src_eb;
2058 src_path->slots[root_level] = dst_path->slots[root_level];
2059 src_path->locks[root_level] = 0;
2061 /* A simplified version of btrfs_search_slot() */
2062 while (cur_level >= dst_level) {
2063 struct btrfs_key src_key;
2064 struct btrfs_key dst_key;
2066 if (src_path->nodes[cur_level] == NULL) {
2067 struct extent_buffer *eb;
2070 eb = src_path->nodes[cur_level + 1];
2071 parent_slot = src_path->slots[cur_level + 1];
2073 eb = btrfs_read_node_slot(eb, parent_slot);
2079 src_path->nodes[cur_level] = eb;
2081 btrfs_tree_read_lock(eb);
2082 src_path->locks[cur_level] = BTRFS_READ_LOCK;
2085 src_path->slots[cur_level] = dst_path->slots[cur_level];
2087 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
2088 &dst_key, dst_path->slots[cur_level]);
2089 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
2090 &src_key, src_path->slots[cur_level]);
2092 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
2093 &dst_key, dst_path->slots[cur_level]);
2094 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
2095 &src_key, src_path->slots[cur_level]);
2097 /* Content mismatch, something went wrong */
2098 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
2106 * Now both @dst_path and @src_path have been populated, record the tree
2107 * blocks for qgroup accounting.
2109 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2113 ret = btrfs_qgroup_trace_extent(trans, dst_path->nodes[dst_level]->start,
2118 /* Record leaf file extents */
2119 if (dst_level == 0 && trace_leaf) {
2120 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2123 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2126 btrfs_free_path(src_path);
2131 * Helper function to do recursive generation-aware depth-first search, to
2132 * locate all new tree blocks in a subtree of reloc tree.
2134 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2143 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2147 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2148 * above tree blocks along with their counter parts in file tree.
2149 * While during search, old tree blocks OO(c) will be skipped as tree block swap
2150 * won't affect OO(c).
2152 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2153 struct extent_buffer *src_eb,
2154 struct btrfs_path *dst_path,
2155 int cur_level, int root_level,
2156 u64 last_snapshot, bool trace_leaf)
2158 struct btrfs_fs_info *fs_info = trans->fs_info;
2159 struct extent_buffer *eb;
2160 bool need_cleanup = false;
2164 /* Level sanity check */
2165 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2166 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2167 root_level < cur_level) {
2168 btrfs_err_rl(fs_info,
2169 "%s: bad levels, cur_level=%d root_level=%d",
2170 __func__, cur_level, root_level);
2174 /* Read the tree block if needed */
2175 if (dst_path->nodes[cur_level] == NULL) {
2180 * dst_path->nodes[root_level] must be initialized before
2181 * calling this function.
2183 if (cur_level == root_level) {
2184 btrfs_err_rl(fs_info,
2185 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2186 __func__, root_level, root_level, cur_level);
2191 * We need to get child blockptr/gen from parent before we can
2194 eb = dst_path->nodes[cur_level + 1];
2195 parent_slot = dst_path->slots[cur_level + 1];
2196 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2198 /* This node is old, no need to trace */
2199 if (child_gen < last_snapshot)
2202 eb = btrfs_read_node_slot(eb, parent_slot);
2208 dst_path->nodes[cur_level] = eb;
2209 dst_path->slots[cur_level] = 0;
2211 btrfs_tree_read_lock(eb);
2212 dst_path->locks[cur_level] = BTRFS_READ_LOCK;
2213 need_cleanup = true;
2216 /* Now record this tree block and its counter part for qgroups */
2217 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2218 root_level, trace_leaf);
2222 eb = dst_path->nodes[cur_level];
2224 if (cur_level > 0) {
2225 /* Iterate all child tree blocks */
2226 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2227 /* Skip old tree blocks as they won't be swapped */
2228 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2230 dst_path->slots[cur_level] = i;
2232 /* Recursive call (at most 7 times) */
2233 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2234 dst_path, cur_level - 1, root_level,
2235 last_snapshot, trace_leaf);
2244 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2245 dst_path->locks[cur_level]);
2246 free_extent_buffer(dst_path->nodes[cur_level]);
2247 dst_path->nodes[cur_level] = NULL;
2248 dst_path->slots[cur_level] = 0;
2249 dst_path->locks[cur_level] = 0;
2255 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2256 struct extent_buffer *src_eb,
2257 struct extent_buffer *dst_eb,
2258 u64 last_snapshot, bool trace_leaf)
2260 struct btrfs_fs_info *fs_info = trans->fs_info;
2261 struct btrfs_path *dst_path = NULL;
2265 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2268 /* Wrong parameter order */
2269 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2270 btrfs_err_rl(fs_info,
2271 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2272 btrfs_header_generation(src_eb),
2273 btrfs_header_generation(dst_eb));
2277 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2282 level = btrfs_header_level(dst_eb);
2283 dst_path = btrfs_alloc_path();
2289 atomic_inc(&dst_eb->refs);
2290 dst_path->nodes[level] = dst_eb;
2291 dst_path->slots[level] = 0;
2292 dst_path->locks[level] = 0;
2294 /* Do the generation aware breadth-first search */
2295 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2296 level, last_snapshot, trace_leaf);
2302 btrfs_free_path(dst_path);
2304 qgroup_mark_inconsistent(fs_info);
2308 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2309 struct extent_buffer *root_eb,
2310 u64 root_gen, int root_level)
2312 struct btrfs_fs_info *fs_info = trans->fs_info;
2315 u8 drop_subptree_thres;
2316 struct extent_buffer *eb = root_eb;
2317 struct btrfs_path *path = NULL;
2319 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2320 BUG_ON(root_eb == NULL);
2322 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2325 spin_lock(&fs_info->qgroup_lock);
2326 drop_subptree_thres = fs_info->qgroup_drop_subtree_thres;
2327 spin_unlock(&fs_info->qgroup_lock);
2330 * This function only gets called for snapshot drop, if we hit a high
2331 * node here, it means we are going to change ownership for quite a lot
2332 * of extents, which will greatly slow down btrfs_commit_transaction().
2334 * So here if we find a high tree here, we just skip the accounting and
2335 * mark qgroup inconsistent.
2337 if (root_level >= drop_subptree_thres) {
2338 qgroup_mark_inconsistent(fs_info);
2342 if (!extent_buffer_uptodate(root_eb)) {
2343 struct btrfs_tree_parent_check check = {
2344 .has_first_key = false,
2345 .transid = root_gen,
2349 ret = btrfs_read_extent_buffer(root_eb, &check);
2354 if (root_level == 0) {
2355 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2359 path = btrfs_alloc_path();
2364 * Walk down the tree. Missing extent blocks are filled in as
2365 * we go. Metadata is accounted every time we read a new
2368 * When we reach a leaf, we account for file extent items in it,
2369 * walk back up the tree (adjusting slot pointers as we go)
2370 * and restart the search process.
2372 atomic_inc(&root_eb->refs); /* For path */
2373 path->nodes[root_level] = root_eb;
2374 path->slots[root_level] = 0;
2375 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2378 while (level >= 0) {
2379 if (path->nodes[level] == NULL) {
2384 * We need to get child blockptr from parent before we
2387 eb = path->nodes[level + 1];
2388 parent_slot = path->slots[level + 1];
2389 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2391 eb = btrfs_read_node_slot(eb, parent_slot);
2397 path->nodes[level] = eb;
2398 path->slots[level] = 0;
2400 btrfs_tree_read_lock(eb);
2401 path->locks[level] = BTRFS_READ_LOCK;
2403 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2410 ret = btrfs_qgroup_trace_leaf_items(trans,
2411 path->nodes[level]);
2415 /* Nonzero return here means we completed our search */
2416 ret = adjust_slots_upwards(path, root_level);
2420 /* Restart search with new slots */
2429 btrfs_free_path(path);
2434 #define UPDATE_NEW 0
2435 #define UPDATE_OLD 1
2437 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2439 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2440 struct ulist *roots, struct ulist *tmp,
2441 struct ulist *qgroups, u64 seq, int update_old)
2443 struct ulist_node *unode;
2444 struct ulist_iterator uiter;
2445 struct ulist_node *tmp_unode;
2446 struct ulist_iterator tmp_uiter;
2447 struct btrfs_qgroup *qg;
2452 ULIST_ITER_INIT(&uiter);
2453 while ((unode = ulist_next(roots, &uiter))) {
2454 qg = find_qgroup_rb(fs_info, unode->val);
2459 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2463 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2466 ULIST_ITER_INIT(&tmp_uiter);
2467 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2468 struct btrfs_qgroup_list *glist;
2470 qg = unode_aux_to_qgroup(tmp_unode);
2472 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2474 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2475 list_for_each_entry(glist, &qg->groups, next_group) {
2476 ret = ulist_add(qgroups, glist->group->qgroupid,
2477 qgroup_to_aux(glist->group),
2481 ret = ulist_add(tmp, glist->group->qgroupid,
2482 qgroup_to_aux(glist->group),
2493 * Update qgroup rfer/excl counters.
2494 * Rfer update is easy, codes can explain themselves.
2496 * Excl update is tricky, the update is split into 2 parts.
2497 * Part 1: Possible exclusive <-> sharing detect:
2499 * -------------------------------------
2501 * -------------------------------------
2503 * -------------------------------------
2506 * A: cur_old_roots < nr_old_roots (not exclusive before)
2507 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2508 * B: cur_new_roots < nr_new_roots (not exclusive now)
2509 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2512 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2513 * *: Definitely not changed. **: Possible unchanged.
2515 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2517 * To make the logic clear, we first use condition A and B to split
2518 * combination into 4 results.
2520 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2521 * only on variant maybe 0.
2523 * Lastly, check result **, since there are 2 variants maybe 0, split them
2525 * But this time we don't need to consider other things, the codes and logic
2526 * is easy to understand now.
2528 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2529 struct ulist *qgroups,
2532 u64 num_bytes, u64 seq)
2534 struct ulist_node *unode;
2535 struct ulist_iterator uiter;
2536 struct btrfs_qgroup *qg;
2537 u64 cur_new_count, cur_old_count;
2539 ULIST_ITER_INIT(&uiter);
2540 while ((unode = ulist_next(qgroups, &uiter))) {
2543 qg = unode_aux_to_qgroup(unode);
2544 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2545 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2547 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2550 /* Rfer update part */
2551 if (cur_old_count == 0 && cur_new_count > 0) {
2552 qg->rfer += num_bytes;
2553 qg->rfer_cmpr += num_bytes;
2556 if (cur_old_count > 0 && cur_new_count == 0) {
2557 qg->rfer -= num_bytes;
2558 qg->rfer_cmpr -= num_bytes;
2562 /* Excl update part */
2563 /* Exclusive/none -> shared case */
2564 if (cur_old_count == nr_old_roots &&
2565 cur_new_count < nr_new_roots) {
2566 /* Exclusive -> shared */
2567 if (cur_old_count != 0) {
2568 qg->excl -= num_bytes;
2569 qg->excl_cmpr -= num_bytes;
2574 /* Shared -> exclusive/none case */
2575 if (cur_old_count < nr_old_roots &&
2576 cur_new_count == nr_new_roots) {
2577 /* Shared->exclusive */
2578 if (cur_new_count != 0) {
2579 qg->excl += num_bytes;
2580 qg->excl_cmpr += num_bytes;
2585 /* Exclusive/none -> exclusive/none case */
2586 if (cur_old_count == nr_old_roots &&
2587 cur_new_count == nr_new_roots) {
2588 if (cur_old_count == 0) {
2589 /* None -> exclusive/none */
2591 if (cur_new_count != 0) {
2592 /* None -> exclusive */
2593 qg->excl += num_bytes;
2594 qg->excl_cmpr += num_bytes;
2597 /* None -> none, nothing changed */
2599 /* Exclusive -> exclusive/none */
2601 if (cur_new_count == 0) {
2602 /* Exclusive -> none */
2603 qg->excl -= num_bytes;
2604 qg->excl_cmpr -= num_bytes;
2607 /* Exclusive -> exclusive, nothing changed */
2612 qgroup_dirty(fs_info, qg);
2618 * Check if the @roots potentially is a list of fs tree roots
2620 * Return 0 for definitely not a fs/subvol tree roots ulist
2621 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2624 static int maybe_fs_roots(struct ulist *roots)
2626 struct ulist_node *unode;
2627 struct ulist_iterator uiter;
2629 /* Empty one, still possible for fs roots */
2630 if (!roots || roots->nnodes == 0)
2633 ULIST_ITER_INIT(&uiter);
2634 unode = ulist_next(roots, &uiter);
2639 * If it contains fs tree roots, then it must belong to fs/subvol
2641 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2643 return is_fstree(unode->val);
2646 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2647 u64 num_bytes, struct ulist *old_roots,
2648 struct ulist *new_roots)
2650 struct btrfs_fs_info *fs_info = trans->fs_info;
2651 struct ulist *qgroups = NULL;
2652 struct ulist *tmp = NULL;
2654 u64 nr_new_roots = 0;
2655 u64 nr_old_roots = 0;
2659 * If quotas get disabled meanwhile, the resources need to be freed and
2660 * we can't just exit here.
2662 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
2663 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
2667 if (!maybe_fs_roots(new_roots))
2669 nr_new_roots = new_roots->nnodes;
2672 if (!maybe_fs_roots(old_roots))
2674 nr_old_roots = old_roots->nnodes;
2677 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2678 if (nr_old_roots == 0 && nr_new_roots == 0)
2681 BUG_ON(!fs_info->quota_root);
2683 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2684 num_bytes, nr_old_roots, nr_new_roots);
2686 qgroups = ulist_alloc(GFP_NOFS);
2691 tmp = ulist_alloc(GFP_NOFS);
2697 mutex_lock(&fs_info->qgroup_rescan_lock);
2698 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2699 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2700 mutex_unlock(&fs_info->qgroup_rescan_lock);
2705 mutex_unlock(&fs_info->qgroup_rescan_lock);
2707 spin_lock(&fs_info->qgroup_lock);
2708 seq = fs_info->qgroup_seq;
2710 /* Update old refcnts using old_roots */
2711 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2716 /* Update new refcnts using new_roots */
2717 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2722 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2726 * Bump qgroup_seq to avoid seq overlap
2728 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2730 spin_unlock(&fs_info->qgroup_lock);
2733 ulist_free(qgroups);
2734 ulist_free(old_roots);
2735 ulist_free(new_roots);
2739 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2741 struct btrfs_fs_info *fs_info = trans->fs_info;
2742 struct btrfs_qgroup_extent_record *record;
2743 struct btrfs_delayed_ref_root *delayed_refs;
2744 struct ulist *new_roots = NULL;
2745 struct rb_node *node;
2746 u64 num_dirty_extents = 0;
2750 delayed_refs = &trans->transaction->delayed_refs;
2751 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2752 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2753 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2756 num_dirty_extents++;
2757 trace_btrfs_qgroup_account_extents(fs_info, record);
2759 if (!ret && !(fs_info->qgroup_flags &
2760 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)) {
2761 struct btrfs_backref_walk_ctx ctx = { 0 };
2763 ctx.bytenr = record->bytenr;
2764 ctx.fs_info = fs_info;
2767 * Old roots should be searched when inserting qgroup
2770 * But for INCONSISTENT (NO_ACCOUNTING) -> rescan case,
2771 * we may have some record inserted during
2772 * NO_ACCOUNTING (thus no old_roots populated), but
2773 * later we start rescan, which clears NO_ACCOUNTING,
2774 * leaving some inserted records without old_roots
2777 * Those cases are rare and should not cause too much
2778 * time spent during commit_transaction().
2780 if (!record->old_roots) {
2781 /* Search commit root to find old_roots */
2782 ret = btrfs_find_all_roots(&ctx, false);
2785 record->old_roots = ctx.roots;
2789 /* Free the reserved data space */
2790 btrfs_qgroup_free_refroot(fs_info,
2791 record->data_rsv_refroot,
2793 BTRFS_QGROUP_RSV_DATA);
2795 * Use BTRFS_SEQ_LAST as time_seq to do special search,
2796 * which doesn't lock tree or delayed_refs and search
2797 * current root. It's safe inside commit_transaction().
2800 ctx.time_seq = BTRFS_SEQ_LAST;
2801 ret = btrfs_find_all_roots(&ctx, false);
2804 new_roots = ctx.roots;
2805 if (qgroup_to_skip) {
2806 ulist_del(new_roots, qgroup_to_skip, 0);
2807 ulist_del(record->old_roots, qgroup_to_skip,
2810 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2814 record->old_roots = NULL;
2818 ulist_free(record->old_roots);
2819 ulist_free(new_roots);
2821 rb_erase(node, &delayed_refs->dirty_extent_root);
2825 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2831 * called from commit_transaction. Writes all changed qgroups to disk.
2833 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2835 struct btrfs_fs_info *fs_info = trans->fs_info;
2838 if (!fs_info->quota_root)
2841 spin_lock(&fs_info->qgroup_lock);
2842 while (!list_empty(&fs_info->dirty_qgroups)) {
2843 struct btrfs_qgroup *qgroup;
2844 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2845 struct btrfs_qgroup, dirty);
2846 list_del_init(&qgroup->dirty);
2847 spin_unlock(&fs_info->qgroup_lock);
2848 ret = update_qgroup_info_item(trans, qgroup);
2850 qgroup_mark_inconsistent(fs_info);
2851 ret = update_qgroup_limit_item(trans, qgroup);
2853 qgroup_mark_inconsistent(fs_info);
2854 spin_lock(&fs_info->qgroup_lock);
2856 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2857 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2859 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2860 spin_unlock(&fs_info->qgroup_lock);
2862 ret = update_qgroup_status_item(trans);
2864 qgroup_mark_inconsistent(fs_info);
2870 * Copy the accounting information between qgroups. This is necessary
2871 * when a snapshot or a subvolume is created. Throwing an error will
2872 * cause a transaction abort so we take extra care here to only error
2873 * when a readonly fs is a reasonable outcome.
2875 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2876 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2881 bool committing = false;
2882 struct btrfs_fs_info *fs_info = trans->fs_info;
2883 struct btrfs_root *quota_root;
2884 struct btrfs_qgroup *srcgroup;
2885 struct btrfs_qgroup *dstgroup;
2886 bool need_rescan = false;
2891 * There are only two callers of this function.
2893 * One in create_subvol() in the ioctl context, which needs to hold
2894 * the qgroup_ioctl_lock.
2896 * The other one in create_pending_snapshot() where no other qgroup
2897 * code can modify the fs as they all need to either start a new trans
2898 * or hold a trans handler, thus we don't need to hold
2899 * qgroup_ioctl_lock.
2900 * This would avoid long and complex lock chain and make lockdep happy.
2902 spin_lock(&fs_info->trans_lock);
2903 if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2905 spin_unlock(&fs_info->trans_lock);
2908 mutex_lock(&fs_info->qgroup_ioctl_lock);
2909 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2912 quota_root = fs_info->quota_root;
2919 i_qgroups = (u64 *)(inherit + 1);
2920 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2921 2 * inherit->num_excl_copies;
2922 for (i = 0; i < nums; ++i) {
2923 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2926 * Zero out invalid groups so we can ignore
2930 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2938 * create a tracking group for the subvol itself
2940 ret = add_qgroup_item(trans, quota_root, objectid);
2945 * add qgroup to all inherited groups
2948 i_qgroups = (u64 *)(inherit + 1);
2949 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2950 if (*i_qgroups == 0)
2952 ret = add_qgroup_relation_item(trans, objectid,
2954 if (ret && ret != -EEXIST)
2956 ret = add_qgroup_relation_item(trans, *i_qgroups,
2958 if (ret && ret != -EEXIST)
2965 spin_lock(&fs_info->qgroup_lock);
2967 dstgroup = add_qgroup_rb(fs_info, objectid);
2968 if (IS_ERR(dstgroup)) {
2969 ret = PTR_ERR(dstgroup);
2973 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2974 dstgroup->lim_flags = inherit->lim.flags;
2975 dstgroup->max_rfer = inherit->lim.max_rfer;
2976 dstgroup->max_excl = inherit->lim.max_excl;
2977 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2978 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2980 qgroup_dirty(fs_info, dstgroup);
2984 srcgroup = find_qgroup_rb(fs_info, srcid);
2989 * We call inherit after we clone the root in order to make sure
2990 * our counts don't go crazy, so at this point the only
2991 * difference between the two roots should be the root node.
2993 level_size = fs_info->nodesize;
2994 dstgroup->rfer = srcgroup->rfer;
2995 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2996 dstgroup->excl = level_size;
2997 dstgroup->excl_cmpr = level_size;
2998 srcgroup->excl = level_size;
2999 srcgroup->excl_cmpr = level_size;
3001 /* inherit the limit info */
3002 dstgroup->lim_flags = srcgroup->lim_flags;
3003 dstgroup->max_rfer = srcgroup->max_rfer;
3004 dstgroup->max_excl = srcgroup->max_excl;
3005 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
3006 dstgroup->rsv_excl = srcgroup->rsv_excl;
3008 qgroup_dirty(fs_info, dstgroup);
3009 qgroup_dirty(fs_info, srcgroup);
3015 i_qgroups = (u64 *)(inherit + 1);
3016 for (i = 0; i < inherit->num_qgroups; ++i) {
3018 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
3025 * If we're doing a snapshot, and adding the snapshot to a new
3026 * qgroup, the numbers are guaranteed to be incorrect.
3032 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
3033 struct btrfs_qgroup *src;
3034 struct btrfs_qgroup *dst;
3036 if (!i_qgroups[0] || !i_qgroups[1])
3039 src = find_qgroup_rb(fs_info, i_qgroups[0]);
3040 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3047 dst->rfer = src->rfer - level_size;
3048 dst->rfer_cmpr = src->rfer_cmpr - level_size;
3050 /* Manually tweaking numbers certainly needs a rescan */
3053 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
3054 struct btrfs_qgroup *src;
3055 struct btrfs_qgroup *dst;
3057 if (!i_qgroups[0] || !i_qgroups[1])
3060 src = find_qgroup_rb(fs_info, i_qgroups[0]);
3061 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3068 dst->excl = src->excl + level_size;
3069 dst->excl_cmpr = src->excl_cmpr + level_size;
3074 spin_unlock(&fs_info->qgroup_lock);
3076 ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
3079 mutex_unlock(&fs_info->qgroup_ioctl_lock);
3081 qgroup_mark_inconsistent(fs_info);
3085 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
3087 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
3088 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
3091 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
3092 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
3098 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
3099 enum btrfs_qgroup_rsv_type type)
3101 struct btrfs_qgroup *qgroup;
3102 struct btrfs_fs_info *fs_info = root->fs_info;
3103 u64 ref_root = root->root_key.objectid;
3105 struct ulist_node *unode;
3106 struct ulist_iterator uiter;
3108 if (!is_fstree(ref_root))
3114 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
3115 capable(CAP_SYS_RESOURCE))
3118 spin_lock(&fs_info->qgroup_lock);
3119 if (!fs_info->quota_root)
3122 qgroup = find_qgroup_rb(fs_info, ref_root);
3127 * in a first step, we check all affected qgroups if any limits would
3130 ulist_reinit(fs_info->qgroup_ulist);
3131 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3132 qgroup_to_aux(qgroup), GFP_ATOMIC);
3135 ULIST_ITER_INIT(&uiter);
3136 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3137 struct btrfs_qgroup *qg;
3138 struct btrfs_qgroup_list *glist;
3140 qg = unode_aux_to_qgroup(unode);
3142 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
3147 list_for_each_entry(glist, &qg->groups, next_group) {
3148 ret = ulist_add(fs_info->qgroup_ulist,
3149 glist->group->qgroupid,
3150 qgroup_to_aux(glist->group), GFP_ATOMIC);
3157 * no limits exceeded, now record the reservation into all qgroups
3159 ULIST_ITER_INIT(&uiter);
3160 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3161 struct btrfs_qgroup *qg;
3163 qg = unode_aux_to_qgroup(unode);
3165 qgroup_rsv_add(fs_info, qg, num_bytes, type);
3169 spin_unlock(&fs_info->qgroup_lock);
3174 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
3177 * Will handle all higher level qgroup too.
3179 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3180 * This special case is only used for META_PERTRANS type.
3182 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3183 u64 ref_root, u64 num_bytes,
3184 enum btrfs_qgroup_rsv_type type)
3186 struct btrfs_qgroup *qgroup;
3187 struct ulist_node *unode;
3188 struct ulist_iterator uiter;
3191 if (!is_fstree(ref_root))
3197 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3198 WARN(1, "%s: Invalid type to free", __func__);
3201 spin_lock(&fs_info->qgroup_lock);
3203 if (!fs_info->quota_root)
3206 qgroup = find_qgroup_rb(fs_info, ref_root);
3210 if (num_bytes == (u64)-1)
3212 * We're freeing all pertrans rsv, get reserved value from
3213 * level 0 qgroup as real num_bytes to free.
3215 num_bytes = qgroup->rsv.values[type];
3217 ulist_reinit(fs_info->qgroup_ulist);
3218 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3219 qgroup_to_aux(qgroup), GFP_ATOMIC);
3222 ULIST_ITER_INIT(&uiter);
3223 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3224 struct btrfs_qgroup *qg;
3225 struct btrfs_qgroup_list *glist;
3227 qg = unode_aux_to_qgroup(unode);
3229 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3231 list_for_each_entry(glist, &qg->groups, next_group) {
3232 ret = ulist_add(fs_info->qgroup_ulist,
3233 glist->group->qgroupid,
3234 qgroup_to_aux(glist->group), GFP_ATOMIC);
3241 spin_unlock(&fs_info->qgroup_lock);
3245 * Check if the leaf is the last leaf. Which means all node pointers
3246 * are at their last position.
3248 static bool is_last_leaf(struct btrfs_path *path)
3252 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3253 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3260 * returns < 0 on error, 0 when more leafs are to be scanned.
3261 * returns 1 when done.
3263 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3264 struct btrfs_path *path)
3266 struct btrfs_fs_info *fs_info = trans->fs_info;
3267 struct btrfs_root *extent_root;
3268 struct btrfs_key found;
3269 struct extent_buffer *scratch_leaf = NULL;
3275 mutex_lock(&fs_info->qgroup_rescan_lock);
3276 extent_root = btrfs_extent_root(fs_info,
3277 fs_info->qgroup_rescan_progress.objectid);
3278 ret = btrfs_search_slot_for_read(extent_root,
3279 &fs_info->qgroup_rescan_progress,
3282 btrfs_debug(fs_info,
3283 "current progress key (%llu %u %llu), search_slot ret %d",
3284 fs_info->qgroup_rescan_progress.objectid,
3285 fs_info->qgroup_rescan_progress.type,
3286 fs_info->qgroup_rescan_progress.offset, ret);
3290 * The rescan is about to end, we will not be scanning any
3291 * further blocks. We cannot unset the RESCAN flag here, because
3292 * we want to commit the transaction if everything went well.
3293 * To make the live accounting work in this phase, we set our
3294 * scan progress pointer such that every real extent objectid
3297 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3298 btrfs_release_path(path);
3299 mutex_unlock(&fs_info->qgroup_rescan_lock);
3302 done = is_last_leaf(path);
3304 btrfs_item_key_to_cpu(path->nodes[0], &found,
3305 btrfs_header_nritems(path->nodes[0]) - 1);
3306 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3308 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3309 if (!scratch_leaf) {
3311 mutex_unlock(&fs_info->qgroup_rescan_lock);
3314 slot = path->slots[0];
3315 btrfs_release_path(path);
3316 mutex_unlock(&fs_info->qgroup_rescan_lock);
3318 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3319 struct btrfs_backref_walk_ctx ctx = { 0 };
3321 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3322 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3323 found.type != BTRFS_METADATA_ITEM_KEY)
3325 if (found.type == BTRFS_METADATA_ITEM_KEY)
3326 num_bytes = fs_info->nodesize;
3328 num_bytes = found.offset;
3330 ctx.bytenr = found.objectid;
3331 ctx.fs_info = fs_info;
3333 ret = btrfs_find_all_roots(&ctx, false);
3336 /* For rescan, just pass old_roots as NULL */
3337 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3338 num_bytes, NULL, ctx.roots);
3344 free_extent_buffer(scratch_leaf);
3348 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3353 static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
3355 return btrfs_fs_closing(fs_info) ||
3356 test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state) ||
3357 !test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3358 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
3361 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3363 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3364 qgroup_rescan_work);
3365 struct btrfs_path *path;
3366 struct btrfs_trans_handle *trans = NULL;
3369 bool stopped = false;
3370 bool did_leaf_rescans = false;
3372 path = btrfs_alloc_path();
3376 * Rescan should only search for commit root, and any later difference
3377 * should be recorded by qgroup
3379 path->search_commit_root = 1;
3380 path->skip_locking = 1;
3383 while (!err && !(stopped = rescan_should_stop(fs_info))) {
3384 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3385 if (IS_ERR(trans)) {
3386 err = PTR_ERR(trans);
3390 err = qgroup_rescan_leaf(trans, path);
3391 did_leaf_rescans = true;
3394 btrfs_commit_transaction(trans);
3396 btrfs_end_transaction(trans);
3400 btrfs_free_path(path);
3402 mutex_lock(&fs_info->qgroup_rescan_lock);
3404 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3405 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3406 } else if (err < 0 || stopped) {
3407 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3409 mutex_unlock(&fs_info->qgroup_rescan_lock);
3412 * Only update status, since the previous part has already updated the
3413 * qgroup info, and only if we did any actual work. This also prevents
3414 * race with a concurrent quota disable, which has already set
3415 * fs_info->quota_root to NULL and cleared BTRFS_FS_QUOTA_ENABLED at
3416 * btrfs_quota_disable().
3418 if (did_leaf_rescans) {
3419 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3420 if (IS_ERR(trans)) {
3421 err = PTR_ERR(trans);
3424 "fail to start transaction for status update: %d",
3431 mutex_lock(&fs_info->qgroup_rescan_lock);
3433 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN)
3434 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3436 ret = update_qgroup_status_item(trans);
3439 btrfs_err(fs_info, "fail to update qgroup status: %d",
3443 fs_info->qgroup_rescan_running = false;
3444 fs_info->qgroup_flags &= ~BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
3445 complete_all(&fs_info->qgroup_rescan_completion);
3446 mutex_unlock(&fs_info->qgroup_rescan_lock);
3451 btrfs_end_transaction(trans);
3454 btrfs_info(fs_info, "qgroup scan paused");
3455 } else if (fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN) {
3456 btrfs_info(fs_info, "qgroup scan cancelled");
3457 } else if (err >= 0) {
3458 btrfs_info(fs_info, "qgroup scan completed%s",
3459 err > 0 ? " (inconsistency flag cleared)" : "");
3461 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3466 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3467 * memory required for the rescan context.
3470 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3476 /* we're resuming qgroup rescan at mount time */
3477 if (!(fs_info->qgroup_flags &
3478 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3480 "qgroup rescan init failed, qgroup rescan is not queued");
3482 } else if (!(fs_info->qgroup_flags &
3483 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3485 "qgroup rescan init failed, qgroup is not enabled");
3493 mutex_lock(&fs_info->qgroup_rescan_lock);
3496 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3498 "qgroup rescan is already in progress");
3500 } else if (!(fs_info->qgroup_flags &
3501 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3503 "qgroup rescan init failed, qgroup is not enabled");
3505 } else if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3506 /* Quota disable is in progress */
3511 mutex_unlock(&fs_info->qgroup_rescan_lock);
3514 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3517 memset(&fs_info->qgroup_rescan_progress, 0,
3518 sizeof(fs_info->qgroup_rescan_progress));
3519 fs_info->qgroup_flags &= ~(BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
3520 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
3521 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3522 init_completion(&fs_info->qgroup_rescan_completion);
3523 mutex_unlock(&fs_info->qgroup_rescan_lock);
3525 btrfs_init_work(&fs_info->qgroup_rescan_work,
3526 btrfs_qgroup_rescan_worker, NULL, NULL);
3531 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3534 struct btrfs_qgroup *qgroup;
3536 spin_lock(&fs_info->qgroup_lock);
3537 /* clear all current qgroup tracking information */
3538 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3539 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3541 qgroup->rfer_cmpr = 0;
3543 qgroup->excl_cmpr = 0;
3544 qgroup_dirty(fs_info, qgroup);
3546 spin_unlock(&fs_info->qgroup_lock);
3550 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3553 struct btrfs_trans_handle *trans;
3555 ret = qgroup_rescan_init(fs_info, 0, 1);
3560 * We have set the rescan_progress to 0, which means no more
3561 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3562 * However, btrfs_qgroup_account_ref may be right after its call
3563 * to btrfs_find_all_roots, in which case it would still do the
3565 * To solve this, we're committing the transaction, which will
3566 * ensure we run all delayed refs and only after that, we are
3567 * going to clear all tracking information for a clean start.
3570 trans = btrfs_join_transaction(fs_info->fs_root);
3571 if (IS_ERR(trans)) {
3572 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3573 return PTR_ERR(trans);
3575 ret = btrfs_commit_transaction(trans);
3577 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3581 qgroup_rescan_zero_tracking(fs_info);
3583 mutex_lock(&fs_info->qgroup_rescan_lock);
3584 fs_info->qgroup_rescan_running = true;
3585 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3586 &fs_info->qgroup_rescan_work);
3587 mutex_unlock(&fs_info->qgroup_rescan_lock);
3592 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3598 mutex_lock(&fs_info->qgroup_rescan_lock);
3599 running = fs_info->qgroup_rescan_running;
3600 mutex_unlock(&fs_info->qgroup_rescan_lock);
3606 ret = wait_for_completion_interruptible(
3607 &fs_info->qgroup_rescan_completion);
3609 wait_for_completion(&fs_info->qgroup_rescan_completion);
3615 * this is only called from open_ctree where we're still single threaded, thus
3616 * locking is omitted here.
3619 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3621 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3622 mutex_lock(&fs_info->qgroup_rescan_lock);
3623 fs_info->qgroup_rescan_running = true;
3624 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3625 &fs_info->qgroup_rescan_work);
3626 mutex_unlock(&fs_info->qgroup_rescan_lock);
3630 #define rbtree_iterate_from_safe(node, next, start) \
3631 for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3633 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3634 struct extent_changeset *reserved, u64 start,
3637 struct rb_node *node;
3638 struct rb_node *next;
3639 struct ulist_node *entry;
3642 node = reserved->range_changed.root.rb_node;
3646 entry = rb_entry(node, struct ulist_node, rb_node);
3647 if (entry->val < start)
3648 node = node->rb_right;
3650 node = node->rb_left;
3653 if (entry->val > start && rb_prev(&entry->rb_node))
3654 entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3657 rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3663 entry = rb_entry(node, struct ulist_node, rb_node);
3664 entry_start = entry->val;
3665 entry_end = entry->aux;
3666 entry_len = entry_end - entry_start + 1;
3668 if (entry_start >= start + len)
3670 if (entry_start + entry_len <= start)
3673 * Now the entry is in [start, start + len), revert the
3674 * EXTENT_QGROUP_RESERVED bit.
3676 clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3677 entry_end, EXTENT_QGROUP_RESERVED);
3678 if (!ret && clear_ret < 0)
3681 ulist_del(&reserved->range_changed, entry->val, entry->aux);
3682 if (likely(reserved->bytes_changed >= entry_len)) {
3683 reserved->bytes_changed -= entry_len;
3686 reserved->bytes_changed = 0;
3694 * Try to free some space for qgroup.
3696 * For qgroup, there are only 3 ways to free qgroup space:
3697 * - Flush nodatacow write
3698 * Any nodatacow write will free its reserved data space at run_delalloc_range().
3699 * In theory, we should only flush nodatacow inodes, but it's not yet
3700 * possible, so we need to flush the whole root.
3702 * - Wait for ordered extents
3703 * When ordered extents are finished, their reserved metadata is finally
3704 * converted to per_trans status, which can be freed by later commit
3707 * - Commit transaction
3708 * This would free the meta_per_trans space.
3709 * In theory this shouldn't provide much space, but any more qgroup space
3712 static int try_flush_qgroup(struct btrfs_root *root)
3714 struct btrfs_trans_handle *trans;
3717 /* Can't hold an open transaction or we run the risk of deadlocking. */
3718 ASSERT(current->journal_info == NULL);
3719 if (WARN_ON(current->journal_info))
3723 * We don't want to run flush again and again, so if there is a running
3724 * one, we won't try to start a new flush, but exit directly.
3726 if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3727 wait_event(root->qgroup_flush_wait,
3728 !test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3732 ret = btrfs_start_delalloc_snapshot(root, true);
3735 btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3737 trans = btrfs_join_transaction(root);
3738 if (IS_ERR(trans)) {
3739 ret = PTR_ERR(trans);
3743 ret = btrfs_commit_transaction(trans);
3745 clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3746 wake_up(&root->qgroup_flush_wait);
3750 static int qgroup_reserve_data(struct btrfs_inode *inode,
3751 struct extent_changeset **reserved_ret, u64 start,
3754 struct btrfs_root *root = inode->root;
3755 struct extent_changeset *reserved;
3756 bool new_reserved = false;
3761 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3762 !is_fstree(root->root_key.objectid) || len == 0)
3765 /* @reserved parameter is mandatory for qgroup */
3766 if (WARN_ON(!reserved_ret))
3768 if (!*reserved_ret) {
3769 new_reserved = true;
3770 *reserved_ret = extent_changeset_alloc();
3774 reserved = *reserved_ret;
3775 /* Record already reserved space */
3776 orig_reserved = reserved->bytes_changed;
3777 ret = set_record_extent_bits(&inode->io_tree, start,
3778 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3780 /* Newly reserved space */
3781 to_reserve = reserved->bytes_changed - orig_reserved;
3782 trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3783 to_reserve, QGROUP_RESERVE);
3786 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3793 qgroup_unreserve_range(inode, reserved, start, len);
3796 extent_changeset_free(reserved);
3797 *reserved_ret = NULL;
3803 * Reserve qgroup space for range [start, start + len).
3805 * This function will either reserve space from related qgroups or do nothing
3806 * if the range is already reserved.
3808 * Return 0 for successful reservation
3809 * Return <0 for error (including -EQUOT)
3811 * NOTE: This function may sleep for memory allocation, dirty page flushing and
3812 * commit transaction. So caller should not hold any dirty page locked.
3814 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3815 struct extent_changeset **reserved_ret, u64 start,
3820 ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3821 if (ret <= 0 && ret != -EDQUOT)
3824 ret = try_flush_qgroup(inode->root);
3827 return qgroup_reserve_data(inode, reserved_ret, start, len);
3830 /* Free ranges specified by @reserved, normally in error path */
3831 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3832 struct extent_changeset *reserved, u64 start, u64 len)
3834 struct btrfs_root *root = inode->root;
3835 struct ulist_node *unode;
3836 struct ulist_iterator uiter;
3837 struct extent_changeset changeset;
3841 extent_changeset_init(&changeset);
3842 len = round_up(start + len, root->fs_info->sectorsize);
3843 start = round_down(start, root->fs_info->sectorsize);
3845 ULIST_ITER_INIT(&uiter);
3846 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3847 u64 range_start = unode->val;
3848 /* unode->aux is the inclusive end */
3849 u64 range_len = unode->aux - range_start + 1;
3853 extent_changeset_release(&changeset);
3855 /* Only free range in range [start, start + len) */
3856 if (range_start >= start + len ||
3857 range_start + range_len <= start)
3859 free_start = max(range_start, start);
3860 free_len = min(start + len, range_start + range_len) -
3863 * TODO: To also modify reserved->ranges_reserved to reflect
3866 * However as long as we free qgroup reserved according to
3867 * EXTENT_QGROUP_RESERVED, we won't double free.
3868 * So not need to rush.
3870 ret = clear_record_extent_bits(&inode->io_tree, free_start,
3871 free_start + free_len - 1,
3872 EXTENT_QGROUP_RESERVED, &changeset);
3875 freed += changeset.bytes_changed;
3877 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3878 BTRFS_QGROUP_RSV_DATA);
3881 extent_changeset_release(&changeset);
3885 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3886 struct extent_changeset *reserved, u64 start, u64 len,
3889 struct extent_changeset changeset;
3890 int trace_op = QGROUP_RELEASE;
3893 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3896 /* In release case, we shouldn't have @reserved */
3897 WARN_ON(!free && reserved);
3898 if (free && reserved)
3899 return qgroup_free_reserved_data(inode, reserved, start, len);
3900 extent_changeset_init(&changeset);
3901 ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3902 EXTENT_QGROUP_RESERVED, &changeset);
3907 trace_op = QGROUP_FREE;
3908 trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3909 changeset.bytes_changed, trace_op);
3911 btrfs_qgroup_free_refroot(inode->root->fs_info,
3912 inode->root->root_key.objectid,
3913 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3914 ret = changeset.bytes_changed;
3916 extent_changeset_release(&changeset);
3921 * Free a reserved space range from io_tree and related qgroups
3923 * Should be called when a range of pages get invalidated before reaching disk.
3924 * Or for error cleanup case.
3925 * if @reserved is given, only reserved range in [@start, @start + @len) will
3928 * For data written to disk, use btrfs_qgroup_release_data().
3930 * NOTE: This function may sleep for memory allocation.
3932 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3933 struct extent_changeset *reserved, u64 start, u64 len)
3935 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3939 * Release a reserved space range from io_tree only.
3941 * Should be called when a range of pages get written to disk and corresponding
3942 * FILE_EXTENT is inserted into corresponding root.
3944 * Since new qgroup accounting framework will only update qgroup numbers at
3945 * commit_transaction() time, its reserved space shouldn't be freed from
3948 * But we should release the range from io_tree, to allow further write to be
3951 * NOTE: This function may sleep for memory allocation.
3953 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3955 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3958 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3959 enum btrfs_qgroup_rsv_type type)
3961 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3962 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3967 spin_lock(&root->qgroup_meta_rsv_lock);
3968 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3969 root->qgroup_meta_rsv_prealloc += num_bytes;
3971 root->qgroup_meta_rsv_pertrans += num_bytes;
3972 spin_unlock(&root->qgroup_meta_rsv_lock);
3975 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3976 enum btrfs_qgroup_rsv_type type)
3978 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3979 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3984 spin_lock(&root->qgroup_meta_rsv_lock);
3985 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3986 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3988 root->qgroup_meta_rsv_prealloc -= num_bytes;
3990 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3992 root->qgroup_meta_rsv_pertrans -= num_bytes;
3994 spin_unlock(&root->qgroup_meta_rsv_lock);
3998 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3999 enum btrfs_qgroup_rsv_type type, bool enforce)
4001 struct btrfs_fs_info *fs_info = root->fs_info;
4004 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4005 !is_fstree(root->root_key.objectid) || num_bytes == 0)
4008 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4009 trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
4010 ret = qgroup_reserve(root, num_bytes, enforce, type);
4014 * Record what we have reserved into root.
4016 * To avoid quota disabled->enabled underflow.
4017 * In that case, we may try to free space we haven't reserved
4018 * (since quota was disabled), so record what we reserved into root.
4019 * And ensure later release won't underflow this number.
4021 add_root_meta_rsv(root, num_bytes, type);
4025 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
4026 enum btrfs_qgroup_rsv_type type, bool enforce,
4031 ret = btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4032 if ((ret <= 0 && ret != -EDQUOT) || noflush)
4035 ret = try_flush_qgroup(root);
4038 return btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4041 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
4043 struct btrfs_fs_info *fs_info = root->fs_info;
4045 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4046 !is_fstree(root->root_key.objectid))
4049 /* TODO: Update trace point to handle such free */
4050 trace_qgroup_meta_free_all_pertrans(root);
4051 /* Special value -1 means to free all reserved space */
4052 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
4053 BTRFS_QGROUP_RSV_META_PERTRANS);
4056 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
4057 enum btrfs_qgroup_rsv_type type)
4059 struct btrfs_fs_info *fs_info = root->fs_info;
4061 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4062 !is_fstree(root->root_key.objectid))
4066 * reservation for META_PREALLOC can happen before quota is enabled,
4067 * which can lead to underflow.
4068 * Here ensure we will only free what we really have reserved.
4070 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
4071 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4072 trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
4073 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
4077 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
4080 struct btrfs_qgroup *qgroup;
4081 struct ulist_node *unode;
4082 struct ulist_iterator uiter;
4087 if (!fs_info->quota_root)
4090 spin_lock(&fs_info->qgroup_lock);
4091 qgroup = find_qgroup_rb(fs_info, ref_root);
4094 ulist_reinit(fs_info->qgroup_ulist);
4095 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
4096 qgroup_to_aux(qgroup), GFP_ATOMIC);
4099 ULIST_ITER_INIT(&uiter);
4100 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
4101 struct btrfs_qgroup *qg;
4102 struct btrfs_qgroup_list *glist;
4104 qg = unode_aux_to_qgroup(unode);
4106 qgroup_rsv_release(fs_info, qg, num_bytes,
4107 BTRFS_QGROUP_RSV_META_PREALLOC);
4108 qgroup_rsv_add(fs_info, qg, num_bytes,
4109 BTRFS_QGROUP_RSV_META_PERTRANS);
4110 list_for_each_entry(glist, &qg->groups, next_group) {
4111 ret = ulist_add(fs_info->qgroup_ulist,
4112 glist->group->qgroupid,
4113 qgroup_to_aux(glist->group), GFP_ATOMIC);
4119 spin_unlock(&fs_info->qgroup_lock);
4122 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
4124 struct btrfs_fs_info *fs_info = root->fs_info;
4126 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4127 !is_fstree(root->root_key.objectid))
4129 /* Same as btrfs_qgroup_free_meta_prealloc() */
4130 num_bytes = sub_root_meta_rsv(root, num_bytes,
4131 BTRFS_QGROUP_RSV_META_PREALLOC);
4132 trace_qgroup_meta_convert(root, num_bytes);
4133 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
4137 * Check qgroup reserved space leaking, normally at destroy inode
4140 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
4142 struct extent_changeset changeset;
4143 struct ulist_node *unode;
4144 struct ulist_iterator iter;
4147 extent_changeset_init(&changeset);
4148 ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4149 EXTENT_QGROUP_RESERVED, &changeset);
4152 if (WARN_ON(changeset.bytes_changed)) {
4153 ULIST_ITER_INIT(&iter);
4154 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4155 btrfs_warn(inode->root->fs_info,
4156 "leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4157 btrfs_ino(inode), unode->val, unode->aux);
4159 btrfs_qgroup_free_refroot(inode->root->fs_info,
4160 inode->root->root_key.objectid,
4161 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4164 extent_changeset_release(&changeset);
4167 void btrfs_qgroup_init_swapped_blocks(
4168 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4172 spin_lock_init(&swapped_blocks->lock);
4173 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4174 swapped_blocks->blocks[i] = RB_ROOT;
4175 swapped_blocks->swapped = false;
4179 * Delete all swapped blocks record of @root.
4180 * Every record here means we skipped a full subtree scan for qgroup.
4182 * Gets called when committing one transaction.
4184 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4186 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4189 swapped_blocks = &root->swapped_blocks;
4191 spin_lock(&swapped_blocks->lock);
4192 if (!swapped_blocks->swapped)
4194 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4195 struct rb_root *cur_root = &swapped_blocks->blocks[i];
4196 struct btrfs_qgroup_swapped_block *entry;
4197 struct btrfs_qgroup_swapped_block *next;
4199 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4202 swapped_blocks->blocks[i] = RB_ROOT;
4204 swapped_blocks->swapped = false;
4206 spin_unlock(&swapped_blocks->lock);
4210 * Add subtree roots record into @subvol_root.
4212 * @subvol_root: tree root of the subvolume tree get swapped
4213 * @bg: block group under balance
4214 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
4215 * @reloc_parent/slot: pointer to the subtree root in reloc tree
4216 * BOTH POINTERS ARE BEFORE TREE SWAP
4217 * @last_snapshot: last snapshot generation of the subvolume tree
4219 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4220 struct btrfs_root *subvol_root,
4221 struct btrfs_block_group *bg,
4222 struct extent_buffer *subvol_parent, int subvol_slot,
4223 struct extent_buffer *reloc_parent, int reloc_slot,
4226 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4227 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4228 struct btrfs_qgroup_swapped_block *block;
4229 struct rb_node **cur;
4230 struct rb_node *parent = NULL;
4231 int level = btrfs_header_level(subvol_parent) - 1;
4234 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4237 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4238 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4239 btrfs_err_rl(fs_info,
4240 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4242 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4243 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4247 block = kmalloc(sizeof(*block), GFP_NOFS);
4254 * @reloc_parent/slot is still before swap, while @block is going to
4255 * record the bytenr after swap, so we do the swap here.
4257 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4258 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4260 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4261 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4263 block->last_snapshot = last_snapshot;
4264 block->level = level;
4267 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4268 * no one else can modify tree blocks thus we qgroup will not change
4269 * no matter the value of trace_leaf.
4271 if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4272 block->trace_leaf = true;
4274 block->trace_leaf = false;
4275 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4277 /* Insert @block into @blocks */
4278 spin_lock(&blocks->lock);
4279 cur = &blocks->blocks[level].rb_node;
4281 struct btrfs_qgroup_swapped_block *entry;
4284 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4287 if (entry->subvol_bytenr < block->subvol_bytenr) {
4288 cur = &(*cur)->rb_left;
4289 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
4290 cur = &(*cur)->rb_right;
4292 if (entry->subvol_generation !=
4293 block->subvol_generation ||
4294 entry->reloc_bytenr != block->reloc_bytenr ||
4295 entry->reloc_generation !=
4296 block->reloc_generation) {
4298 * Duplicated but mismatch entry found.
4301 * Marking qgroup inconsistent should be enough
4304 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4311 rb_link_node(&block->node, parent, cur);
4312 rb_insert_color(&block->node, &blocks->blocks[level]);
4313 blocks->swapped = true;
4315 spin_unlock(&blocks->lock);
4318 qgroup_mark_inconsistent(fs_info);
4323 * Check if the tree block is a subtree root, and if so do the needed
4324 * delayed subtree trace for qgroup.
4326 * This is called during btrfs_cow_block().
4328 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4329 struct btrfs_root *root,
4330 struct extent_buffer *subvol_eb)
4332 struct btrfs_fs_info *fs_info = root->fs_info;
4333 struct btrfs_tree_parent_check check = { 0 };
4334 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4335 struct btrfs_qgroup_swapped_block *block;
4336 struct extent_buffer *reloc_eb = NULL;
4337 struct rb_node *node;
4339 bool swapped = false;
4340 int level = btrfs_header_level(subvol_eb);
4344 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4346 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4349 spin_lock(&blocks->lock);
4350 if (!blocks->swapped) {
4351 spin_unlock(&blocks->lock);
4354 node = blocks->blocks[level].rb_node;
4357 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4358 if (block->subvol_bytenr < subvol_eb->start) {
4359 node = node->rb_left;
4360 } else if (block->subvol_bytenr > subvol_eb->start) {
4361 node = node->rb_right;
4368 spin_unlock(&blocks->lock);
4371 /* Found one, remove it from @blocks first and update blocks->swapped */
4372 rb_erase(&block->node, &blocks->blocks[level]);
4373 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4374 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4379 blocks->swapped = swapped;
4380 spin_unlock(&blocks->lock);
4382 check.level = block->level;
4383 check.transid = block->reloc_generation;
4384 check.has_first_key = true;
4385 memcpy(&check.first_key, &block->first_key, sizeof(check.first_key));
4387 /* Read out reloc subtree root */
4388 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr, &check);
4389 if (IS_ERR(reloc_eb)) {
4390 ret = PTR_ERR(reloc_eb);
4394 if (!extent_buffer_uptodate(reloc_eb)) {
4399 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4400 block->last_snapshot, block->trace_leaf);
4403 free_extent_buffer(reloc_eb);
4406 btrfs_err_rl(fs_info,
4407 "failed to account subtree at bytenr %llu: %d",
4408 subvol_eb->start, ret);
4409 qgroup_mark_inconsistent(fs_info);
4414 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4416 struct btrfs_qgroup_extent_record *entry;
4417 struct btrfs_qgroup_extent_record *next;
4418 struct rb_root *root;
4420 root = &trans->delayed_refs.dirty_extent_root;
4421 rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4422 ulist_free(entry->old_roots);