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"
29 * - subvol delete -> delete when ref goes to 0? delete limits also?
33 * - copy also limits on subvol creation
36 * - performance benchmarks
37 * - check all ioctl parameters
41 * Helpers to access qgroup reservation
43 * Callers should ensure the lock context and type are valid
46 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
51 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
52 ret += qgroup->rsv.values[i];
57 #ifdef CONFIG_BTRFS_DEBUG
58 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
60 if (type == BTRFS_QGROUP_RSV_DATA)
62 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
63 return "meta_pertrans";
64 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
65 return "meta_prealloc";
70 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
71 struct btrfs_qgroup *qgroup, u64 num_bytes,
72 enum btrfs_qgroup_rsv_type type)
74 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
75 qgroup->rsv.values[type] += num_bytes;
78 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
79 struct btrfs_qgroup *qgroup, u64 num_bytes,
80 enum btrfs_qgroup_rsv_type type)
82 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
83 if (qgroup->rsv.values[type] >= num_bytes) {
84 qgroup->rsv.values[type] -= num_bytes;
87 #ifdef CONFIG_BTRFS_DEBUG
89 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
90 qgroup->qgroupid, qgroup_rsv_type_str(type),
91 qgroup->rsv.values[type], num_bytes);
93 qgroup->rsv.values[type] = 0;
96 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
97 struct btrfs_qgroup *dest,
98 struct btrfs_qgroup *src)
102 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
103 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
106 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
107 struct btrfs_qgroup *dest,
108 struct btrfs_qgroup *src)
112 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
113 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
116 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
119 if (qg->old_refcnt < seq)
120 qg->old_refcnt = seq;
121 qg->old_refcnt += mod;
124 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
127 if (qg->new_refcnt < seq)
128 qg->new_refcnt = seq;
129 qg->new_refcnt += mod;
132 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
134 if (qg->old_refcnt < seq)
136 return qg->old_refcnt - seq;
139 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
141 if (qg->new_refcnt < seq)
143 return qg->new_refcnt - seq;
147 * glue structure to represent the relations between qgroups.
149 struct btrfs_qgroup_list {
150 struct list_head next_group;
151 struct list_head next_member;
152 struct btrfs_qgroup *group;
153 struct btrfs_qgroup *member;
156 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
158 return (u64)(uintptr_t)qg;
161 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
163 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
167 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
169 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
171 /* must be called with qgroup_ioctl_lock held */
172 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
175 struct rb_node *n = fs_info->qgroup_tree.rb_node;
176 struct btrfs_qgroup *qgroup;
179 qgroup = rb_entry(n, struct btrfs_qgroup, node);
180 if (qgroup->qgroupid < qgroupid)
182 else if (qgroup->qgroupid > qgroupid)
190 /* must be called with qgroup_lock held */
191 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
194 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
195 struct rb_node *parent = NULL;
196 struct btrfs_qgroup *qgroup;
200 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
202 if (qgroup->qgroupid < qgroupid)
204 else if (qgroup->qgroupid > qgroupid)
210 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
212 return ERR_PTR(-ENOMEM);
214 qgroup->qgroupid = qgroupid;
215 INIT_LIST_HEAD(&qgroup->groups);
216 INIT_LIST_HEAD(&qgroup->members);
217 INIT_LIST_HEAD(&qgroup->dirty);
219 rb_link_node(&qgroup->node, parent, p);
220 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
225 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
226 struct btrfs_qgroup *qgroup)
228 struct btrfs_qgroup_list *list;
230 list_del(&qgroup->dirty);
231 while (!list_empty(&qgroup->groups)) {
232 list = list_first_entry(&qgroup->groups,
233 struct btrfs_qgroup_list, next_group);
234 list_del(&list->next_group);
235 list_del(&list->next_member);
239 while (!list_empty(&qgroup->members)) {
240 list = list_first_entry(&qgroup->members,
241 struct btrfs_qgroup_list, next_member);
242 list_del(&list->next_group);
243 list_del(&list->next_member);
248 /* must be called with qgroup_lock held */
249 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
251 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
256 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
257 __del_qgroup_rb(fs_info, qgroup);
261 /* must be called with qgroup_lock held */
262 static int add_relation_rb(struct btrfs_fs_info *fs_info,
263 u64 memberid, u64 parentid)
265 struct btrfs_qgroup *member;
266 struct btrfs_qgroup *parent;
267 struct btrfs_qgroup_list *list;
269 member = find_qgroup_rb(fs_info, memberid);
270 parent = find_qgroup_rb(fs_info, parentid);
271 if (!member || !parent)
274 list = kzalloc(sizeof(*list), GFP_ATOMIC);
278 list->group = parent;
279 list->member = member;
280 list_add_tail(&list->next_group, &member->groups);
281 list_add_tail(&list->next_member, &parent->members);
286 /* must be called with qgroup_lock held */
287 static int del_relation_rb(struct btrfs_fs_info *fs_info,
288 u64 memberid, u64 parentid)
290 struct btrfs_qgroup *member;
291 struct btrfs_qgroup *parent;
292 struct btrfs_qgroup_list *list;
294 member = find_qgroup_rb(fs_info, memberid);
295 parent = find_qgroup_rb(fs_info, parentid);
296 if (!member || !parent)
299 list_for_each_entry(list, &member->groups, next_group) {
300 if (list->group == parent) {
301 list_del(&list->next_group);
302 list_del(&list->next_member);
310 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
311 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
314 struct btrfs_qgroup *qgroup;
316 qgroup = find_qgroup_rb(fs_info, qgroupid);
319 if (qgroup->rfer != rfer || qgroup->excl != excl)
326 * The full config is read in one go, only called from open_ctree()
327 * It doesn't use any locking, as at this point we're still single-threaded
329 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
331 struct btrfs_key key;
332 struct btrfs_key found_key;
333 struct btrfs_root *quota_root = fs_info->quota_root;
334 struct btrfs_path *path = NULL;
335 struct extent_buffer *l;
339 u64 rescan_progress = 0;
341 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
344 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
345 if (!fs_info->qgroup_ulist) {
350 path = btrfs_alloc_path();
356 ret = btrfs_sysfs_add_qgroups(fs_info);
359 /* default this to quota off, in case no status key is found */
360 fs_info->qgroup_flags = 0;
363 * pass 1: read status, all qgroup infos and limits
368 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
373 struct btrfs_qgroup *qgroup;
375 slot = path->slots[0];
377 btrfs_item_key_to_cpu(l, &found_key, slot);
379 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
380 struct btrfs_qgroup_status_item *ptr;
382 ptr = btrfs_item_ptr(l, slot,
383 struct btrfs_qgroup_status_item);
385 if (btrfs_qgroup_status_version(l, ptr) !=
386 BTRFS_QGROUP_STATUS_VERSION) {
388 "old qgroup version, quota disabled");
391 if (btrfs_qgroup_status_generation(l, ptr) !=
392 fs_info->generation) {
393 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
395 "qgroup generation mismatch, marked as inconsistent");
397 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
399 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
403 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
404 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
407 qgroup = find_qgroup_rb(fs_info, found_key.offset);
408 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
409 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
410 btrfs_err(fs_info, "inconsistent qgroup config");
411 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
414 qgroup = add_qgroup_rb(fs_info, found_key.offset);
415 if (IS_ERR(qgroup)) {
416 ret = PTR_ERR(qgroup);
420 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
424 switch (found_key.type) {
425 case BTRFS_QGROUP_INFO_KEY: {
426 struct btrfs_qgroup_info_item *ptr;
428 ptr = btrfs_item_ptr(l, slot,
429 struct btrfs_qgroup_info_item);
430 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
431 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
432 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
433 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
434 /* generation currently unused */
437 case BTRFS_QGROUP_LIMIT_KEY: {
438 struct btrfs_qgroup_limit_item *ptr;
440 ptr = btrfs_item_ptr(l, slot,
441 struct btrfs_qgroup_limit_item);
442 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
443 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
444 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
445 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
446 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
451 ret = btrfs_next_item(quota_root, path);
457 btrfs_release_path(path);
460 * pass 2: read all qgroup relations
463 key.type = BTRFS_QGROUP_RELATION_KEY;
465 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
469 slot = path->slots[0];
471 btrfs_item_key_to_cpu(l, &found_key, slot);
473 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
476 if (found_key.objectid > found_key.offset) {
477 /* parent <- member, not needed to build config */
478 /* FIXME should we omit the key completely? */
482 ret = add_relation_rb(fs_info, found_key.objectid,
484 if (ret == -ENOENT) {
486 "orphan qgroup relation 0x%llx->0x%llx",
487 found_key.objectid, found_key.offset);
488 ret = 0; /* ignore the error */
493 ret = btrfs_next_item(quota_root, path);
500 btrfs_free_path(path);
501 fs_info->qgroup_flags |= flags;
502 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
503 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
504 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
506 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
509 ulist_free(fs_info->qgroup_ulist);
510 fs_info->qgroup_ulist = NULL;
511 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
512 btrfs_sysfs_del_qgroups(fs_info);
515 return ret < 0 ? ret : 0;
519 * Called in close_ctree() when quota is still enabled. This verifies we don't
520 * leak some reserved space.
522 * Return false if no reserved space is left.
523 * Return true if some reserved space is leaked.
525 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
527 struct rb_node *node;
530 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
533 * Since we're unmounting, there is no race and no need to grab qgroup
534 * lock. And here we don't go post-order to provide a more user
535 * friendly sorted result.
537 for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
538 struct btrfs_qgroup *qgroup;
541 qgroup = rb_entry(node, struct btrfs_qgroup, node);
542 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
543 if (qgroup->rsv.values[i]) {
546 "qgroup %hu/%llu has unreleased space, type %d rsv %llu",
547 btrfs_qgroup_level(qgroup->qgroupid),
548 btrfs_qgroup_subvolid(qgroup->qgroupid),
549 i, qgroup->rsv.values[i]);
557 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
558 * first two are in single-threaded paths.And for the third one, we have set
559 * quota_root to be null with qgroup_lock held before, so it is safe to clean
560 * up the in-memory structures without qgroup_lock held.
562 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
565 struct btrfs_qgroup *qgroup;
567 while ((n = rb_first(&fs_info->qgroup_tree))) {
568 qgroup = rb_entry(n, struct btrfs_qgroup, node);
569 rb_erase(n, &fs_info->qgroup_tree);
570 __del_qgroup_rb(fs_info, qgroup);
571 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
575 * We call btrfs_free_qgroup_config() when unmounting
576 * filesystem and disabling quota, so we set qgroup_ulist
577 * to be null here to avoid double free.
579 ulist_free(fs_info->qgroup_ulist);
580 fs_info->qgroup_ulist = NULL;
581 btrfs_sysfs_del_qgroups(fs_info);
584 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
588 struct btrfs_root *quota_root = trans->fs_info->quota_root;
589 struct btrfs_path *path;
590 struct btrfs_key key;
592 path = btrfs_alloc_path();
597 key.type = BTRFS_QGROUP_RELATION_KEY;
600 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
602 btrfs_mark_buffer_dirty(path->nodes[0]);
604 btrfs_free_path(path);
608 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
612 struct btrfs_root *quota_root = trans->fs_info->quota_root;
613 struct btrfs_path *path;
614 struct btrfs_key key;
616 path = btrfs_alloc_path();
621 key.type = BTRFS_QGROUP_RELATION_KEY;
624 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
633 ret = btrfs_del_item(trans, quota_root, path);
635 btrfs_free_path(path);
639 static int add_qgroup_item(struct btrfs_trans_handle *trans,
640 struct btrfs_root *quota_root, u64 qgroupid)
643 struct btrfs_path *path;
644 struct btrfs_qgroup_info_item *qgroup_info;
645 struct btrfs_qgroup_limit_item *qgroup_limit;
646 struct extent_buffer *leaf;
647 struct btrfs_key key;
649 if (btrfs_is_testing(quota_root->fs_info))
652 path = btrfs_alloc_path();
657 key.type = BTRFS_QGROUP_INFO_KEY;
658 key.offset = qgroupid;
661 * Avoid a transaction abort by catching -EEXIST here. In that
662 * case, we proceed by re-initializing the existing structure
666 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
667 sizeof(*qgroup_info));
668 if (ret && ret != -EEXIST)
671 leaf = path->nodes[0];
672 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
673 struct btrfs_qgroup_info_item);
674 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
675 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
676 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
677 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
678 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
680 btrfs_mark_buffer_dirty(leaf);
682 btrfs_release_path(path);
684 key.type = BTRFS_QGROUP_LIMIT_KEY;
685 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
686 sizeof(*qgroup_limit));
687 if (ret && ret != -EEXIST)
690 leaf = path->nodes[0];
691 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
692 struct btrfs_qgroup_limit_item);
693 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
694 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
695 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
696 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
697 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
699 btrfs_mark_buffer_dirty(leaf);
703 btrfs_free_path(path);
707 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
710 struct btrfs_root *quota_root = trans->fs_info->quota_root;
711 struct btrfs_path *path;
712 struct btrfs_key key;
714 path = btrfs_alloc_path();
719 key.type = BTRFS_QGROUP_INFO_KEY;
720 key.offset = qgroupid;
721 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
730 ret = btrfs_del_item(trans, quota_root, path);
734 btrfs_release_path(path);
736 key.type = BTRFS_QGROUP_LIMIT_KEY;
737 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
746 ret = btrfs_del_item(trans, quota_root, path);
749 btrfs_free_path(path);
753 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
754 struct btrfs_qgroup *qgroup)
756 struct btrfs_root *quota_root = trans->fs_info->quota_root;
757 struct btrfs_path *path;
758 struct btrfs_key key;
759 struct extent_buffer *l;
760 struct btrfs_qgroup_limit_item *qgroup_limit;
765 key.type = BTRFS_QGROUP_LIMIT_KEY;
766 key.offset = qgroup->qgroupid;
768 path = btrfs_alloc_path();
772 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
780 slot = path->slots[0];
781 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
782 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
783 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
784 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
785 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
786 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
788 btrfs_mark_buffer_dirty(l);
791 btrfs_free_path(path);
795 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
796 struct btrfs_qgroup *qgroup)
798 struct btrfs_fs_info *fs_info = trans->fs_info;
799 struct btrfs_root *quota_root = fs_info->quota_root;
800 struct btrfs_path *path;
801 struct btrfs_key key;
802 struct extent_buffer *l;
803 struct btrfs_qgroup_info_item *qgroup_info;
807 if (btrfs_is_testing(fs_info))
811 key.type = BTRFS_QGROUP_INFO_KEY;
812 key.offset = qgroup->qgroupid;
814 path = btrfs_alloc_path();
818 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
826 slot = path->slots[0];
827 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
828 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
829 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
830 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
831 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
832 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
834 btrfs_mark_buffer_dirty(l);
837 btrfs_free_path(path);
841 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
843 struct btrfs_fs_info *fs_info = trans->fs_info;
844 struct btrfs_root *quota_root = fs_info->quota_root;
845 struct btrfs_path *path;
846 struct btrfs_key key;
847 struct extent_buffer *l;
848 struct btrfs_qgroup_status_item *ptr;
853 key.type = BTRFS_QGROUP_STATUS_KEY;
856 path = btrfs_alloc_path();
860 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
868 slot = path->slots[0];
869 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
870 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
871 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
872 btrfs_set_qgroup_status_rescan(l, ptr,
873 fs_info->qgroup_rescan_progress.objectid);
875 btrfs_mark_buffer_dirty(l);
878 btrfs_free_path(path);
883 * called with qgroup_lock held
885 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
886 struct btrfs_root *root)
888 struct btrfs_path *path;
889 struct btrfs_key key;
890 struct extent_buffer *leaf = NULL;
894 path = btrfs_alloc_path();
903 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
906 leaf = path->nodes[0];
907 nr = btrfs_header_nritems(leaf);
911 * delete the leaf one by one
912 * since the whole tree is going
916 ret = btrfs_del_items(trans, root, path, 0, nr);
920 btrfs_release_path(path);
924 btrfs_free_path(path);
928 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
930 struct btrfs_root *quota_root;
931 struct btrfs_root *tree_root = fs_info->tree_root;
932 struct btrfs_path *path = NULL;
933 struct btrfs_qgroup_status_item *ptr;
934 struct extent_buffer *leaf;
935 struct btrfs_key key;
936 struct btrfs_key found_key;
937 struct btrfs_qgroup *qgroup = NULL;
938 struct btrfs_trans_handle *trans = NULL;
939 struct ulist *ulist = NULL;
944 * We need to have subvol_sem write locked, to prevent races between
945 * concurrent tasks trying to enable quotas, because we will unlock
946 * and relock qgroup_ioctl_lock before setting fs_info->quota_root
947 * and before setting BTRFS_FS_QUOTA_ENABLED.
949 lockdep_assert_held_write(&fs_info->subvol_sem);
951 mutex_lock(&fs_info->qgroup_ioctl_lock);
952 if (fs_info->quota_root)
955 ulist = ulist_alloc(GFP_KERNEL);
961 ret = btrfs_sysfs_add_qgroups(fs_info);
966 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
967 * avoid lock acquisition inversion problems (reported by lockdep) between
968 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
969 * start a transaction.
970 * After we started the transaction lock qgroup_ioctl_lock again and
971 * check if someone else created the quota root in the meanwhile. If so,
972 * just return success and release the transaction handle.
974 * Also we don't need to worry about someone else calling
975 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
976 * that function returns 0 (success) when the sysfs entries already exist.
978 mutex_unlock(&fs_info->qgroup_ioctl_lock);
981 * 1 for quota root item
982 * 1 for BTRFS_QGROUP_STATUS item
984 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
985 * per subvolume. However those are not currently reserved since it
986 * would be a lot of overkill.
988 trans = btrfs_start_transaction(tree_root, 2);
990 mutex_lock(&fs_info->qgroup_ioctl_lock);
992 ret = PTR_ERR(trans);
997 if (fs_info->quota_root)
1000 fs_info->qgroup_ulist = ulist;
1004 * initially create the quota tree
1006 quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
1007 if (IS_ERR(quota_root)) {
1008 ret = PTR_ERR(quota_root);
1009 btrfs_abort_transaction(trans, ret);
1013 path = btrfs_alloc_path();
1016 btrfs_abort_transaction(trans, ret);
1021 key.type = BTRFS_QGROUP_STATUS_KEY;
1024 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1027 btrfs_abort_transaction(trans, ret);
1031 leaf = path->nodes[0];
1032 ptr = btrfs_item_ptr(leaf, path->slots[0],
1033 struct btrfs_qgroup_status_item);
1034 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1035 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1036 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
1037 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1038 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
1039 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1041 btrfs_mark_buffer_dirty(leaf);
1044 key.type = BTRFS_ROOT_REF_KEY;
1047 btrfs_release_path(path);
1048 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1052 btrfs_abort_transaction(trans, ret);
1057 slot = path->slots[0];
1058 leaf = path->nodes[0];
1059 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1061 if (found_key.type == BTRFS_ROOT_REF_KEY) {
1063 /* Release locks on tree_root before we access quota_root */
1064 btrfs_release_path(path);
1066 ret = add_qgroup_item(trans, quota_root,
1069 btrfs_abort_transaction(trans, ret);
1073 qgroup = add_qgroup_rb(fs_info, found_key.offset);
1074 if (IS_ERR(qgroup)) {
1075 ret = PTR_ERR(qgroup);
1076 btrfs_abort_transaction(trans, ret);
1079 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1081 btrfs_abort_transaction(trans, ret);
1084 ret = btrfs_search_slot_for_read(tree_root, &found_key,
1087 btrfs_abort_transaction(trans, ret);
1092 * Shouldn't happen, but in case it does we
1093 * don't need to do the btrfs_next_item, just
1099 ret = btrfs_next_item(tree_root, path);
1101 btrfs_abort_transaction(trans, ret);
1109 btrfs_release_path(path);
1110 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1112 btrfs_abort_transaction(trans, ret);
1116 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1117 if (IS_ERR(qgroup)) {
1118 ret = PTR_ERR(qgroup);
1119 btrfs_abort_transaction(trans, ret);
1122 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1124 btrfs_abort_transaction(trans, ret);
1128 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1130 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
1131 * a deadlock with tasks concurrently doing other qgroup operations, such
1132 * adding/removing qgroups or adding/deleting qgroup relations for example,
1133 * because all qgroup operations first start or join a transaction and then
1134 * lock the qgroup_ioctl_lock mutex.
1135 * We are safe from a concurrent task trying to enable quotas, by calling
1136 * this function, since we are serialized by fs_info->subvol_sem.
1138 ret = btrfs_commit_transaction(trans);
1140 mutex_lock(&fs_info->qgroup_ioctl_lock);
1145 * Set quota enabled flag after committing the transaction, to avoid
1146 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1149 spin_lock(&fs_info->qgroup_lock);
1150 fs_info->quota_root = quota_root;
1151 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1152 spin_unlock(&fs_info->qgroup_lock);
1154 ret = qgroup_rescan_init(fs_info, 0, 1);
1156 qgroup_rescan_zero_tracking(fs_info);
1157 fs_info->qgroup_rescan_running = true;
1158 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1159 &fs_info->qgroup_rescan_work);
1162 * We have set both BTRFS_FS_QUOTA_ENABLED and
1163 * BTRFS_QGROUP_STATUS_FLAG_ON, so we can only fail with
1164 * -EINPROGRESS. That can happen because someone started the
1165 * rescan worker by calling quota rescan ioctl before we
1166 * attempted to initialize the rescan worker. Failure due to
1167 * quotas disabled in the meanwhile is not possible, because
1168 * we are holding a write lock on fs_info->subvol_sem, which
1169 * is also acquired when disabling quotas.
1170 * Ignore such error, and any other error would need to undo
1171 * everything we did in the transaction we just committed.
1173 ASSERT(ret == -EINPROGRESS);
1178 btrfs_free_path(path);
1181 btrfs_put_root(quota_root);
1184 ulist_free(fs_info->qgroup_ulist);
1185 fs_info->qgroup_ulist = NULL;
1186 btrfs_sysfs_del_qgroups(fs_info);
1188 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1190 btrfs_end_transaction(trans);
1192 ret = btrfs_end_transaction(trans);
1197 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1199 struct btrfs_root *quota_root;
1200 struct btrfs_trans_handle *trans = NULL;
1204 * We need to have subvol_sem write locked, to prevent races between
1205 * concurrent tasks trying to disable quotas, because we will unlock
1206 * and relock qgroup_ioctl_lock across BTRFS_FS_QUOTA_ENABLED changes.
1208 lockdep_assert_held_write(&fs_info->subvol_sem);
1210 mutex_lock(&fs_info->qgroup_ioctl_lock);
1211 if (!fs_info->quota_root)
1215 * Unlock the qgroup_ioctl_lock mutex before waiting for the rescan worker to
1216 * complete. Otherwise we can deadlock because btrfs_remove_qgroup() needs
1217 * to lock that mutex while holding a transaction handle and the rescan
1218 * worker needs to commit a transaction.
1220 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1223 * Request qgroup rescan worker to complete and wait for it. This wait
1224 * must be done before transaction start for quota disable since it may
1225 * deadlock with transaction by the qgroup rescan worker.
1227 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1228 btrfs_qgroup_wait_for_completion(fs_info, false);
1231 * 1 For the root item
1233 * We should also reserve enough items for the quota tree deletion in
1234 * btrfs_clean_quota_tree but this is not done.
1236 * Also, we must always start a transaction without holding the mutex
1237 * qgroup_ioctl_lock, see btrfs_quota_enable().
1239 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1241 mutex_lock(&fs_info->qgroup_ioctl_lock);
1242 if (IS_ERR(trans)) {
1243 ret = PTR_ERR(trans);
1245 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1249 if (!fs_info->quota_root)
1252 spin_lock(&fs_info->qgroup_lock);
1253 quota_root = fs_info->quota_root;
1254 fs_info->quota_root = NULL;
1255 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1256 spin_unlock(&fs_info->qgroup_lock);
1258 btrfs_free_qgroup_config(fs_info);
1260 ret = btrfs_clean_quota_tree(trans, quota_root);
1262 btrfs_abort_transaction(trans, ret);
1266 ret = btrfs_del_root(trans, "a_root->root_key);
1268 btrfs_abort_transaction(trans, ret);
1272 list_del("a_root->dirty_list);
1274 btrfs_tree_lock(quota_root->node);
1275 btrfs_clean_tree_block(quota_root->node);
1276 btrfs_tree_unlock(quota_root->node);
1277 btrfs_free_tree_block(trans, btrfs_root_id(quota_root),
1278 quota_root->node, 0, 1);
1280 btrfs_put_root(quota_root);
1283 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1285 btrfs_end_transaction(trans);
1287 ret = btrfs_end_transaction(trans);
1292 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1293 struct btrfs_qgroup *qgroup)
1295 if (list_empty(&qgroup->dirty))
1296 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1300 * The easy accounting, we're updating qgroup relationship whose child qgroup
1301 * only has exclusive extents.
1303 * In this case, all exclusive extents will also be exclusive for parent, so
1304 * excl/rfer just get added/removed.
1306 * So is qgroup reservation space, which should also be added/removed to
1308 * Or when child tries to release reservation space, parent will underflow its
1309 * reservation (for relationship adding case).
1311 * Caller should hold fs_info->qgroup_lock.
1313 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1314 struct ulist *tmp, u64 ref_root,
1315 struct btrfs_qgroup *src, int sign)
1317 struct btrfs_qgroup *qgroup;
1318 struct btrfs_qgroup_list *glist;
1319 struct ulist_node *unode;
1320 struct ulist_iterator uiter;
1321 u64 num_bytes = src->excl;
1324 qgroup = find_qgroup_rb(fs_info, ref_root);
1328 qgroup->rfer += sign * num_bytes;
1329 qgroup->rfer_cmpr += sign * num_bytes;
1331 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1332 qgroup->excl += sign * num_bytes;
1333 qgroup->excl_cmpr += sign * num_bytes;
1336 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1338 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1340 qgroup_dirty(fs_info, qgroup);
1342 /* Get all of the parent groups that contain this qgroup */
1343 list_for_each_entry(glist, &qgroup->groups, next_group) {
1344 ret = ulist_add(tmp, glist->group->qgroupid,
1345 qgroup_to_aux(glist->group), GFP_ATOMIC);
1350 /* Iterate all of the parents and adjust their reference counts */
1351 ULIST_ITER_INIT(&uiter);
1352 while ((unode = ulist_next(tmp, &uiter))) {
1353 qgroup = unode_aux_to_qgroup(unode);
1354 qgroup->rfer += sign * num_bytes;
1355 qgroup->rfer_cmpr += sign * num_bytes;
1356 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1357 qgroup->excl += sign * num_bytes;
1359 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1361 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1362 qgroup->excl_cmpr += sign * num_bytes;
1363 qgroup_dirty(fs_info, qgroup);
1365 /* Add any parents of the parents */
1366 list_for_each_entry(glist, &qgroup->groups, next_group) {
1367 ret = ulist_add(tmp, glist->group->qgroupid,
1368 qgroup_to_aux(glist->group), GFP_ATOMIC);
1380 * Quick path for updating qgroup with only excl refs.
1382 * In that case, just update all parent will be enough.
1383 * Or we needs to do a full rescan.
1384 * Caller should also hold fs_info->qgroup_lock.
1386 * Return 0 for quick update, return >0 for need to full rescan
1387 * and mark INCONSISTENT flag.
1388 * Return < 0 for other error.
1390 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1391 struct ulist *tmp, u64 src, u64 dst,
1394 struct btrfs_qgroup *qgroup;
1398 qgroup = find_qgroup_rb(fs_info, src);
1401 if (qgroup->excl == qgroup->rfer) {
1403 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1412 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1416 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1419 struct btrfs_fs_info *fs_info = trans->fs_info;
1420 struct btrfs_qgroup *parent;
1421 struct btrfs_qgroup *member;
1422 struct btrfs_qgroup_list *list;
1424 unsigned int nofs_flag;
1427 /* Check the level of src and dst first */
1428 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1431 /* We hold a transaction handle open, must do a NOFS allocation. */
1432 nofs_flag = memalloc_nofs_save();
1433 tmp = ulist_alloc(GFP_KERNEL);
1434 memalloc_nofs_restore(nofs_flag);
1438 mutex_lock(&fs_info->qgroup_ioctl_lock);
1439 if (!fs_info->quota_root) {
1443 member = find_qgroup_rb(fs_info, src);
1444 parent = find_qgroup_rb(fs_info, dst);
1445 if (!member || !parent) {
1450 /* check if such qgroup relation exist firstly */
1451 list_for_each_entry(list, &member->groups, next_group) {
1452 if (list->group == parent) {
1458 ret = add_qgroup_relation_item(trans, src, dst);
1462 ret = add_qgroup_relation_item(trans, dst, src);
1464 del_qgroup_relation_item(trans, src, dst);
1468 spin_lock(&fs_info->qgroup_lock);
1469 ret = add_relation_rb(fs_info, src, dst);
1471 spin_unlock(&fs_info->qgroup_lock);
1474 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1475 spin_unlock(&fs_info->qgroup_lock);
1477 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1482 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1485 struct btrfs_fs_info *fs_info = trans->fs_info;
1486 struct btrfs_qgroup *parent;
1487 struct btrfs_qgroup *member;
1488 struct btrfs_qgroup_list *list;
1491 unsigned int nofs_flag;
1495 /* We hold a transaction handle open, must do a NOFS allocation. */
1496 nofs_flag = memalloc_nofs_save();
1497 tmp = ulist_alloc(GFP_KERNEL);
1498 memalloc_nofs_restore(nofs_flag);
1502 if (!fs_info->quota_root) {
1507 member = find_qgroup_rb(fs_info, src);
1508 parent = find_qgroup_rb(fs_info, dst);
1510 * The parent/member pair doesn't exist, then try to delete the dead
1511 * relation items only.
1513 if (!member || !parent)
1516 /* check if such qgroup relation exist firstly */
1517 list_for_each_entry(list, &member->groups, next_group) {
1518 if (list->group == parent) {
1525 ret = del_qgroup_relation_item(trans, src, dst);
1526 if (ret < 0 && ret != -ENOENT)
1528 ret2 = del_qgroup_relation_item(trans, dst, src);
1529 if (ret2 < 0 && ret2 != -ENOENT)
1532 /* At least one deletion succeeded, return 0 */
1537 spin_lock(&fs_info->qgroup_lock);
1538 del_relation_rb(fs_info, src, dst);
1539 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1540 spin_unlock(&fs_info->qgroup_lock);
1547 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1550 struct btrfs_fs_info *fs_info = trans->fs_info;
1553 mutex_lock(&fs_info->qgroup_ioctl_lock);
1554 ret = __del_qgroup_relation(trans, src, dst);
1555 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1560 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1562 struct btrfs_fs_info *fs_info = trans->fs_info;
1563 struct btrfs_root *quota_root;
1564 struct btrfs_qgroup *qgroup;
1567 mutex_lock(&fs_info->qgroup_ioctl_lock);
1568 if (!fs_info->quota_root) {
1572 quota_root = fs_info->quota_root;
1573 qgroup = find_qgroup_rb(fs_info, qgroupid);
1579 ret = add_qgroup_item(trans, quota_root, qgroupid);
1583 spin_lock(&fs_info->qgroup_lock);
1584 qgroup = add_qgroup_rb(fs_info, qgroupid);
1585 spin_unlock(&fs_info->qgroup_lock);
1587 if (IS_ERR(qgroup)) {
1588 ret = PTR_ERR(qgroup);
1591 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1593 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1597 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1599 struct btrfs_fs_info *fs_info = trans->fs_info;
1600 struct btrfs_qgroup *qgroup;
1601 struct btrfs_qgroup_list *list;
1604 mutex_lock(&fs_info->qgroup_ioctl_lock);
1605 if (!fs_info->quota_root) {
1610 qgroup = find_qgroup_rb(fs_info, qgroupid);
1616 /* Check if there are no children of this qgroup */
1617 if (!list_empty(&qgroup->members)) {
1622 ret = del_qgroup_item(trans, qgroupid);
1623 if (ret && ret != -ENOENT)
1626 while (!list_empty(&qgroup->groups)) {
1627 list = list_first_entry(&qgroup->groups,
1628 struct btrfs_qgroup_list, next_group);
1629 ret = __del_qgroup_relation(trans, qgroupid,
1630 list->group->qgroupid);
1635 spin_lock(&fs_info->qgroup_lock);
1636 del_qgroup_rb(fs_info, qgroupid);
1637 spin_unlock(&fs_info->qgroup_lock);
1640 * Remove the qgroup from sysfs now without holding the qgroup_lock
1641 * spinlock, since the sysfs_remove_group() function needs to take
1642 * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1644 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
1647 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1651 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1652 struct btrfs_qgroup_limit *limit)
1654 struct btrfs_fs_info *fs_info = trans->fs_info;
1655 struct btrfs_qgroup *qgroup;
1657 /* Sometimes we would want to clear the limit on this qgroup.
1658 * To meet this requirement, we treat the -1 as a special value
1659 * which tell kernel to clear the limit on this qgroup.
1661 const u64 CLEAR_VALUE = -1;
1663 mutex_lock(&fs_info->qgroup_ioctl_lock);
1664 if (!fs_info->quota_root) {
1669 qgroup = find_qgroup_rb(fs_info, qgroupid);
1675 spin_lock(&fs_info->qgroup_lock);
1676 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1677 if (limit->max_rfer == CLEAR_VALUE) {
1678 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1679 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1680 qgroup->max_rfer = 0;
1682 qgroup->max_rfer = limit->max_rfer;
1685 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1686 if (limit->max_excl == CLEAR_VALUE) {
1687 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1688 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1689 qgroup->max_excl = 0;
1691 qgroup->max_excl = limit->max_excl;
1694 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1695 if (limit->rsv_rfer == CLEAR_VALUE) {
1696 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1697 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1698 qgroup->rsv_rfer = 0;
1700 qgroup->rsv_rfer = limit->rsv_rfer;
1703 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1704 if (limit->rsv_excl == CLEAR_VALUE) {
1705 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1706 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1707 qgroup->rsv_excl = 0;
1709 qgroup->rsv_excl = limit->rsv_excl;
1712 qgroup->lim_flags |= limit->flags;
1714 spin_unlock(&fs_info->qgroup_lock);
1716 ret = update_qgroup_limit_item(trans, qgroup);
1718 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1719 btrfs_info(fs_info, "unable to update quota limit for %llu",
1724 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1728 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1729 struct btrfs_delayed_ref_root *delayed_refs,
1730 struct btrfs_qgroup_extent_record *record)
1732 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1733 struct rb_node *parent_node = NULL;
1734 struct btrfs_qgroup_extent_record *entry;
1735 u64 bytenr = record->bytenr;
1737 lockdep_assert_held(&delayed_refs->lock);
1738 trace_btrfs_qgroup_trace_extent(fs_info, record);
1742 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1744 if (bytenr < entry->bytenr) {
1746 } else if (bytenr > entry->bytenr) {
1747 p = &(*p)->rb_right;
1749 if (record->data_rsv && !entry->data_rsv) {
1750 entry->data_rsv = record->data_rsv;
1751 entry->data_rsv_refroot =
1752 record->data_rsv_refroot;
1758 rb_link_node(&record->node, parent_node, p);
1759 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1763 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
1764 struct btrfs_qgroup_extent_record *qrecord)
1766 struct ulist *old_root;
1767 u64 bytenr = qrecord->bytenr;
1771 * We are always called in a context where we are already holding a
1772 * transaction handle. Often we are called when adding a data delayed
1773 * reference from btrfs_truncate_inode_items() (truncating or unlinking),
1774 * in which case we will be holding a write lock on extent buffer from a
1775 * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
1776 * acquire fs_info->commit_root_sem, because that is a higher level lock
1777 * that must be acquired before locking any extent buffers.
1779 * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
1780 * but we can't pass it a non-NULL transaction handle, because otherwise
1781 * it would not use commit roots and would lock extent buffers, causing
1782 * a deadlock if it ends up trying to read lock the same extent buffer
1783 * that was previously write locked at btrfs_truncate_inode_items().
1785 * So pass a NULL transaction handle to btrfs_find_all_roots() and
1786 * explicitly tell it to not acquire the commit_root_sem - if we are
1787 * holding a transaction handle we don't need its protection.
1789 ASSERT(trans != NULL);
1791 ret = btrfs_find_all_roots(NULL, trans->fs_info, bytenr, 0, &old_root,
1794 trans->fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1795 btrfs_warn(trans->fs_info,
1796 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1802 * Here we don't need to get the lock of
1803 * trans->transaction->delayed_refs, since inserted qrecord won't
1804 * be deleted, only qrecord->node may be modified (new qrecord insert)
1806 * So modifying qrecord->old_roots is safe here
1808 qrecord->old_roots = old_root;
1812 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1813 u64 num_bytes, gfp_t gfp_flag)
1815 struct btrfs_fs_info *fs_info = trans->fs_info;
1816 struct btrfs_qgroup_extent_record *record;
1817 struct btrfs_delayed_ref_root *delayed_refs;
1820 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1821 || bytenr == 0 || num_bytes == 0)
1823 record = kzalloc(sizeof(*record), gfp_flag);
1827 delayed_refs = &trans->transaction->delayed_refs;
1828 record->bytenr = bytenr;
1829 record->num_bytes = num_bytes;
1830 record->old_roots = NULL;
1832 spin_lock(&delayed_refs->lock);
1833 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1834 spin_unlock(&delayed_refs->lock);
1839 return btrfs_qgroup_trace_extent_post(trans, record);
1842 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1843 struct extent_buffer *eb)
1845 struct btrfs_fs_info *fs_info = trans->fs_info;
1846 int nr = btrfs_header_nritems(eb);
1847 int i, extent_type, ret;
1848 struct btrfs_key key;
1849 struct btrfs_file_extent_item *fi;
1850 u64 bytenr, num_bytes;
1852 /* We can be called directly from walk_up_proc() */
1853 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1856 for (i = 0; i < nr; i++) {
1857 btrfs_item_key_to_cpu(eb, &key, i);
1859 if (key.type != BTRFS_EXTENT_DATA_KEY)
1862 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1863 /* filter out non qgroup-accountable extents */
1864 extent_type = btrfs_file_extent_type(eb, fi);
1866 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1869 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1873 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1875 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1885 * Walk up the tree from the bottom, freeing leaves and any interior
1886 * nodes which have had all slots visited. If a node (leaf or
1887 * interior) is freed, the node above it will have it's slot
1888 * incremented. The root node will never be freed.
1890 * At the end of this function, we should have a path which has all
1891 * slots incremented to the next position for a search. If we need to
1892 * read a new node it will be NULL and the node above it will have the
1893 * correct slot selected for a later read.
1895 * If we increment the root nodes slot counter past the number of
1896 * elements, 1 is returned to signal completion of the search.
1898 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1902 struct extent_buffer *eb;
1904 if (root_level == 0)
1907 while (level <= root_level) {
1908 eb = path->nodes[level];
1909 nr = btrfs_header_nritems(eb);
1910 path->slots[level]++;
1911 slot = path->slots[level];
1912 if (slot >= nr || level == 0) {
1914 * Don't free the root - we will detect this
1915 * condition after our loop and return a
1916 * positive value for caller to stop walking the tree.
1918 if (level != root_level) {
1919 btrfs_tree_unlock_rw(eb, path->locks[level]);
1920 path->locks[level] = 0;
1922 free_extent_buffer(eb);
1923 path->nodes[level] = NULL;
1924 path->slots[level] = 0;
1928 * We have a valid slot to walk back down
1929 * from. Stop here so caller can process these
1938 eb = path->nodes[root_level];
1939 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1946 * Helper function to trace a subtree tree block swap.
1948 * The swap will happen in highest tree block, but there may be a lot of
1949 * tree blocks involved.
1952 * OO = Old tree blocks
1953 * NN = New tree blocks allocated during balance
1955 * File tree (257) Reloc tree for 257
1958 * L1 OO OO (a) OO NN (a)
1960 * L0 OO OO OO OO OO OO NN NN
1963 * When calling qgroup_trace_extent_swap(), we will pass:
1965 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1969 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1970 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1972 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1974 * 1) Tree search from @src_eb
1975 * It should acts as a simplified btrfs_search_slot().
1976 * The key for search can be extracted from @dst_path->nodes[dst_level]
1979 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1980 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1981 * They should be marked during previous (@dst_level = 1) iteration.
1983 * 3) Mark file extents in leaves dirty
1984 * We don't have good way to pick out new file extents only.
1985 * So we still follow the old method by scanning all file extents in
1988 * This function can free us from keeping two paths, thus later we only need
1989 * to care about how to iterate all new tree blocks in reloc tree.
1991 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1992 struct extent_buffer *src_eb,
1993 struct btrfs_path *dst_path,
1994 int dst_level, int root_level,
1997 struct btrfs_key key;
1998 struct btrfs_path *src_path;
1999 struct btrfs_fs_info *fs_info = trans->fs_info;
2000 u32 nodesize = fs_info->nodesize;
2001 int cur_level = root_level;
2004 BUG_ON(dst_level > root_level);
2005 /* Level mismatch */
2006 if (btrfs_header_level(src_eb) != root_level)
2009 src_path = btrfs_alloc_path();
2016 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2018 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2021 atomic_inc(&src_eb->refs);
2022 src_path->nodes[root_level] = src_eb;
2023 src_path->slots[root_level] = dst_path->slots[root_level];
2024 src_path->locks[root_level] = 0;
2026 /* A simplified version of btrfs_search_slot() */
2027 while (cur_level >= dst_level) {
2028 struct btrfs_key src_key;
2029 struct btrfs_key dst_key;
2031 if (src_path->nodes[cur_level] == NULL) {
2032 struct extent_buffer *eb;
2035 eb = src_path->nodes[cur_level + 1];
2036 parent_slot = src_path->slots[cur_level + 1];
2038 eb = btrfs_read_node_slot(eb, parent_slot);
2044 src_path->nodes[cur_level] = eb;
2046 btrfs_tree_read_lock(eb);
2047 src_path->locks[cur_level] = BTRFS_READ_LOCK;
2050 src_path->slots[cur_level] = dst_path->slots[cur_level];
2052 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
2053 &dst_key, dst_path->slots[cur_level]);
2054 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
2055 &src_key, src_path->slots[cur_level]);
2057 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
2058 &dst_key, dst_path->slots[cur_level]);
2059 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
2060 &src_key, src_path->slots[cur_level]);
2062 /* Content mismatch, something went wrong */
2063 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
2071 * Now both @dst_path and @src_path have been populated, record the tree
2072 * blocks for qgroup accounting.
2074 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2075 nodesize, GFP_NOFS);
2078 ret = btrfs_qgroup_trace_extent(trans,
2079 dst_path->nodes[dst_level]->start,
2080 nodesize, GFP_NOFS);
2084 /* Record leaf file extents */
2085 if (dst_level == 0 && trace_leaf) {
2086 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2089 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2092 btrfs_free_path(src_path);
2097 * Helper function to do recursive generation-aware depth-first search, to
2098 * locate all new tree blocks in a subtree of reloc tree.
2100 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2109 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2113 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2114 * above tree blocks along with their counter parts in file tree.
2115 * While during search, old tree blocks OO(c) will be skipped as tree block swap
2116 * won't affect OO(c).
2118 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2119 struct extent_buffer *src_eb,
2120 struct btrfs_path *dst_path,
2121 int cur_level, int root_level,
2122 u64 last_snapshot, bool trace_leaf)
2124 struct btrfs_fs_info *fs_info = trans->fs_info;
2125 struct extent_buffer *eb;
2126 bool need_cleanup = false;
2130 /* Level sanity check */
2131 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2132 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2133 root_level < cur_level) {
2134 btrfs_err_rl(fs_info,
2135 "%s: bad levels, cur_level=%d root_level=%d",
2136 __func__, cur_level, root_level);
2140 /* Read the tree block if needed */
2141 if (dst_path->nodes[cur_level] == NULL) {
2146 * dst_path->nodes[root_level] must be initialized before
2147 * calling this function.
2149 if (cur_level == root_level) {
2150 btrfs_err_rl(fs_info,
2151 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2152 __func__, root_level, root_level, cur_level);
2157 * We need to get child blockptr/gen from parent before we can
2160 eb = dst_path->nodes[cur_level + 1];
2161 parent_slot = dst_path->slots[cur_level + 1];
2162 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2164 /* This node is old, no need to trace */
2165 if (child_gen < last_snapshot)
2168 eb = btrfs_read_node_slot(eb, parent_slot);
2174 dst_path->nodes[cur_level] = eb;
2175 dst_path->slots[cur_level] = 0;
2177 btrfs_tree_read_lock(eb);
2178 dst_path->locks[cur_level] = BTRFS_READ_LOCK;
2179 need_cleanup = true;
2182 /* Now record this tree block and its counter part for qgroups */
2183 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2184 root_level, trace_leaf);
2188 eb = dst_path->nodes[cur_level];
2190 if (cur_level > 0) {
2191 /* Iterate all child tree blocks */
2192 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2193 /* Skip old tree blocks as they won't be swapped */
2194 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2196 dst_path->slots[cur_level] = i;
2198 /* Recursive call (at most 7 times) */
2199 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2200 dst_path, cur_level - 1, root_level,
2201 last_snapshot, trace_leaf);
2210 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2211 dst_path->locks[cur_level]);
2212 free_extent_buffer(dst_path->nodes[cur_level]);
2213 dst_path->nodes[cur_level] = NULL;
2214 dst_path->slots[cur_level] = 0;
2215 dst_path->locks[cur_level] = 0;
2221 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2222 struct extent_buffer *src_eb,
2223 struct extent_buffer *dst_eb,
2224 u64 last_snapshot, bool trace_leaf)
2226 struct btrfs_fs_info *fs_info = trans->fs_info;
2227 struct btrfs_path *dst_path = NULL;
2231 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2234 /* Wrong parameter order */
2235 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2236 btrfs_err_rl(fs_info,
2237 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2238 btrfs_header_generation(src_eb),
2239 btrfs_header_generation(dst_eb));
2243 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2248 level = btrfs_header_level(dst_eb);
2249 dst_path = btrfs_alloc_path();
2255 atomic_inc(&dst_eb->refs);
2256 dst_path->nodes[level] = dst_eb;
2257 dst_path->slots[level] = 0;
2258 dst_path->locks[level] = 0;
2260 /* Do the generation aware breadth-first search */
2261 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2262 level, last_snapshot, trace_leaf);
2268 btrfs_free_path(dst_path);
2270 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2274 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2275 struct extent_buffer *root_eb,
2276 u64 root_gen, int root_level)
2278 struct btrfs_fs_info *fs_info = trans->fs_info;
2281 struct extent_buffer *eb = root_eb;
2282 struct btrfs_path *path = NULL;
2284 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2285 BUG_ON(root_eb == NULL);
2287 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2290 if (!extent_buffer_uptodate(root_eb)) {
2291 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2296 if (root_level == 0) {
2297 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2301 path = btrfs_alloc_path();
2306 * Walk down the tree. Missing extent blocks are filled in as
2307 * we go. Metadata is accounted every time we read a new
2310 * When we reach a leaf, we account for file extent items in it,
2311 * walk back up the tree (adjusting slot pointers as we go)
2312 * and restart the search process.
2314 atomic_inc(&root_eb->refs); /* For path */
2315 path->nodes[root_level] = root_eb;
2316 path->slots[root_level] = 0;
2317 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2320 while (level >= 0) {
2321 if (path->nodes[level] == NULL) {
2326 * We need to get child blockptr from parent before we
2329 eb = path->nodes[level + 1];
2330 parent_slot = path->slots[level + 1];
2331 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2333 eb = btrfs_read_node_slot(eb, parent_slot);
2339 path->nodes[level] = eb;
2340 path->slots[level] = 0;
2342 btrfs_tree_read_lock(eb);
2343 path->locks[level] = BTRFS_READ_LOCK;
2345 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2353 ret = btrfs_qgroup_trace_leaf_items(trans,
2354 path->nodes[level]);
2358 /* Nonzero return here means we completed our search */
2359 ret = adjust_slots_upwards(path, root_level);
2363 /* Restart search with new slots */
2372 btrfs_free_path(path);
2377 #define UPDATE_NEW 0
2378 #define UPDATE_OLD 1
2380 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2382 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2383 struct ulist *roots, struct ulist *tmp,
2384 struct ulist *qgroups, u64 seq, int update_old)
2386 struct ulist_node *unode;
2387 struct ulist_iterator uiter;
2388 struct ulist_node *tmp_unode;
2389 struct ulist_iterator tmp_uiter;
2390 struct btrfs_qgroup *qg;
2395 ULIST_ITER_INIT(&uiter);
2396 while ((unode = ulist_next(roots, &uiter))) {
2397 qg = find_qgroup_rb(fs_info, unode->val);
2402 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2406 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2409 ULIST_ITER_INIT(&tmp_uiter);
2410 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2411 struct btrfs_qgroup_list *glist;
2413 qg = unode_aux_to_qgroup(tmp_unode);
2415 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2417 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2418 list_for_each_entry(glist, &qg->groups, next_group) {
2419 ret = ulist_add(qgroups, glist->group->qgroupid,
2420 qgroup_to_aux(glist->group),
2424 ret = ulist_add(tmp, glist->group->qgroupid,
2425 qgroup_to_aux(glist->group),
2436 * Update qgroup rfer/excl counters.
2437 * Rfer update is easy, codes can explain themselves.
2439 * Excl update is tricky, the update is split into 2 parts.
2440 * Part 1: Possible exclusive <-> sharing detect:
2442 * -------------------------------------
2444 * -------------------------------------
2446 * -------------------------------------
2449 * A: cur_old_roots < nr_old_roots (not exclusive before)
2450 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2451 * B: cur_new_roots < nr_new_roots (not exclusive now)
2452 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2455 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2456 * *: Definitely not changed. **: Possible unchanged.
2458 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2460 * To make the logic clear, we first use condition A and B to split
2461 * combination into 4 results.
2463 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2464 * only on variant maybe 0.
2466 * Lastly, check result **, since there are 2 variants maybe 0, split them
2468 * But this time we don't need to consider other things, the codes and logic
2469 * is easy to understand now.
2471 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2472 struct ulist *qgroups,
2475 u64 num_bytes, u64 seq)
2477 struct ulist_node *unode;
2478 struct ulist_iterator uiter;
2479 struct btrfs_qgroup *qg;
2480 u64 cur_new_count, cur_old_count;
2482 ULIST_ITER_INIT(&uiter);
2483 while ((unode = ulist_next(qgroups, &uiter))) {
2486 qg = unode_aux_to_qgroup(unode);
2487 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2488 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2490 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2493 /* Rfer update part */
2494 if (cur_old_count == 0 && cur_new_count > 0) {
2495 qg->rfer += num_bytes;
2496 qg->rfer_cmpr += num_bytes;
2499 if (cur_old_count > 0 && cur_new_count == 0) {
2500 qg->rfer -= num_bytes;
2501 qg->rfer_cmpr -= num_bytes;
2505 /* Excl update part */
2506 /* Exclusive/none -> shared case */
2507 if (cur_old_count == nr_old_roots &&
2508 cur_new_count < nr_new_roots) {
2509 /* Exclusive -> shared */
2510 if (cur_old_count != 0) {
2511 qg->excl -= num_bytes;
2512 qg->excl_cmpr -= num_bytes;
2517 /* Shared -> exclusive/none case */
2518 if (cur_old_count < nr_old_roots &&
2519 cur_new_count == nr_new_roots) {
2520 /* Shared->exclusive */
2521 if (cur_new_count != 0) {
2522 qg->excl += num_bytes;
2523 qg->excl_cmpr += num_bytes;
2528 /* Exclusive/none -> exclusive/none case */
2529 if (cur_old_count == nr_old_roots &&
2530 cur_new_count == nr_new_roots) {
2531 if (cur_old_count == 0) {
2532 /* None -> exclusive/none */
2534 if (cur_new_count != 0) {
2535 /* None -> exclusive */
2536 qg->excl += num_bytes;
2537 qg->excl_cmpr += num_bytes;
2540 /* None -> none, nothing changed */
2542 /* Exclusive -> exclusive/none */
2544 if (cur_new_count == 0) {
2545 /* Exclusive -> none */
2546 qg->excl -= num_bytes;
2547 qg->excl_cmpr -= num_bytes;
2550 /* Exclusive -> exclusive, nothing changed */
2555 qgroup_dirty(fs_info, qg);
2561 * Check if the @roots potentially is a list of fs tree roots
2563 * Return 0 for definitely not a fs/subvol tree roots ulist
2564 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2567 static int maybe_fs_roots(struct ulist *roots)
2569 struct ulist_node *unode;
2570 struct ulist_iterator uiter;
2572 /* Empty one, still possible for fs roots */
2573 if (!roots || roots->nnodes == 0)
2576 ULIST_ITER_INIT(&uiter);
2577 unode = ulist_next(roots, &uiter);
2582 * If it contains fs tree roots, then it must belong to fs/subvol
2584 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2586 return is_fstree(unode->val);
2589 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2590 u64 num_bytes, struct ulist *old_roots,
2591 struct ulist *new_roots)
2593 struct btrfs_fs_info *fs_info = trans->fs_info;
2594 struct ulist *qgroups = NULL;
2595 struct ulist *tmp = NULL;
2597 u64 nr_new_roots = 0;
2598 u64 nr_old_roots = 0;
2602 * If quotas get disabled meanwhile, the resources need to be freed and
2603 * we can't just exit here.
2605 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2609 if (!maybe_fs_roots(new_roots))
2611 nr_new_roots = new_roots->nnodes;
2614 if (!maybe_fs_roots(old_roots))
2616 nr_old_roots = old_roots->nnodes;
2619 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2620 if (nr_old_roots == 0 && nr_new_roots == 0)
2623 BUG_ON(!fs_info->quota_root);
2625 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2626 num_bytes, nr_old_roots, nr_new_roots);
2628 qgroups = ulist_alloc(GFP_NOFS);
2633 tmp = ulist_alloc(GFP_NOFS);
2639 mutex_lock(&fs_info->qgroup_rescan_lock);
2640 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2641 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2642 mutex_unlock(&fs_info->qgroup_rescan_lock);
2647 mutex_unlock(&fs_info->qgroup_rescan_lock);
2649 spin_lock(&fs_info->qgroup_lock);
2650 seq = fs_info->qgroup_seq;
2652 /* Update old refcnts using old_roots */
2653 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2658 /* Update new refcnts using new_roots */
2659 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2664 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2668 * Bump qgroup_seq to avoid seq overlap
2670 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2672 spin_unlock(&fs_info->qgroup_lock);
2675 ulist_free(qgroups);
2676 ulist_free(old_roots);
2677 ulist_free(new_roots);
2681 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2683 struct btrfs_fs_info *fs_info = trans->fs_info;
2684 struct btrfs_qgroup_extent_record *record;
2685 struct btrfs_delayed_ref_root *delayed_refs;
2686 struct ulist *new_roots = NULL;
2687 struct rb_node *node;
2688 u64 num_dirty_extents = 0;
2692 delayed_refs = &trans->transaction->delayed_refs;
2693 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2694 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2695 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2698 num_dirty_extents++;
2699 trace_btrfs_qgroup_account_extents(fs_info, record);
2703 * Old roots should be searched when inserting qgroup
2706 if (WARN_ON(!record->old_roots)) {
2707 /* Search commit root to find old_roots */
2708 ret = btrfs_find_all_roots(NULL, fs_info,
2710 &record->old_roots, false);
2715 /* Free the reserved data space */
2716 btrfs_qgroup_free_refroot(fs_info,
2717 record->data_rsv_refroot,
2719 BTRFS_QGROUP_RSV_DATA);
2721 * Use BTRFS_SEQ_LAST as time_seq to do special search,
2722 * which doesn't lock tree or delayed_refs and search
2723 * current root. It's safe inside commit_transaction().
2725 ret = btrfs_find_all_roots(trans, fs_info,
2726 record->bytenr, BTRFS_SEQ_LAST, &new_roots, false);
2729 if (qgroup_to_skip) {
2730 ulist_del(new_roots, qgroup_to_skip, 0);
2731 ulist_del(record->old_roots, qgroup_to_skip,
2734 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2738 record->old_roots = NULL;
2742 ulist_free(record->old_roots);
2743 ulist_free(new_roots);
2745 rb_erase(node, &delayed_refs->dirty_extent_root);
2749 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2755 * called from commit_transaction. Writes all changed qgroups to disk.
2757 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2759 struct btrfs_fs_info *fs_info = trans->fs_info;
2762 if (!fs_info->quota_root)
2765 spin_lock(&fs_info->qgroup_lock);
2766 while (!list_empty(&fs_info->dirty_qgroups)) {
2767 struct btrfs_qgroup *qgroup;
2768 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2769 struct btrfs_qgroup, dirty);
2770 list_del_init(&qgroup->dirty);
2771 spin_unlock(&fs_info->qgroup_lock);
2772 ret = update_qgroup_info_item(trans, qgroup);
2774 fs_info->qgroup_flags |=
2775 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2776 ret = update_qgroup_limit_item(trans, qgroup);
2778 fs_info->qgroup_flags |=
2779 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2780 spin_lock(&fs_info->qgroup_lock);
2782 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2783 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2785 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2786 spin_unlock(&fs_info->qgroup_lock);
2788 ret = update_qgroup_status_item(trans);
2790 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2796 * Copy the accounting information between qgroups. This is necessary
2797 * when a snapshot or a subvolume is created. Throwing an error will
2798 * cause a transaction abort so we take extra care here to only error
2799 * when a readonly fs is a reasonable outcome.
2801 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2802 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2807 bool committing = false;
2808 struct btrfs_fs_info *fs_info = trans->fs_info;
2809 struct btrfs_root *quota_root;
2810 struct btrfs_qgroup *srcgroup;
2811 struct btrfs_qgroup *dstgroup;
2812 bool need_rescan = false;
2817 * There are only two callers of this function.
2819 * One in create_subvol() in the ioctl context, which needs to hold
2820 * the qgroup_ioctl_lock.
2822 * The other one in create_pending_snapshot() where no other qgroup
2823 * code can modify the fs as they all need to either start a new trans
2824 * or hold a trans handler, thus we don't need to hold
2825 * qgroup_ioctl_lock.
2826 * This would avoid long and complex lock chain and make lockdep happy.
2828 spin_lock(&fs_info->trans_lock);
2829 if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2831 spin_unlock(&fs_info->trans_lock);
2834 mutex_lock(&fs_info->qgroup_ioctl_lock);
2835 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2838 quota_root = fs_info->quota_root;
2845 i_qgroups = (u64 *)(inherit + 1);
2846 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2847 2 * inherit->num_excl_copies;
2848 for (i = 0; i < nums; ++i) {
2849 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2852 * Zero out invalid groups so we can ignore
2856 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2864 * create a tracking group for the subvol itself
2866 ret = add_qgroup_item(trans, quota_root, objectid);
2871 * add qgroup to all inherited groups
2874 i_qgroups = (u64 *)(inherit + 1);
2875 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2876 if (*i_qgroups == 0)
2878 ret = add_qgroup_relation_item(trans, objectid,
2880 if (ret && ret != -EEXIST)
2882 ret = add_qgroup_relation_item(trans, *i_qgroups,
2884 if (ret && ret != -EEXIST)
2891 spin_lock(&fs_info->qgroup_lock);
2893 dstgroup = add_qgroup_rb(fs_info, objectid);
2894 if (IS_ERR(dstgroup)) {
2895 ret = PTR_ERR(dstgroup);
2899 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2900 dstgroup->lim_flags = inherit->lim.flags;
2901 dstgroup->max_rfer = inherit->lim.max_rfer;
2902 dstgroup->max_excl = inherit->lim.max_excl;
2903 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2904 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2906 qgroup_dirty(fs_info, dstgroup);
2910 srcgroup = find_qgroup_rb(fs_info, srcid);
2915 * We call inherit after we clone the root in order to make sure
2916 * our counts don't go crazy, so at this point the only
2917 * difference between the two roots should be the root node.
2919 level_size = fs_info->nodesize;
2920 dstgroup->rfer = srcgroup->rfer;
2921 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2922 dstgroup->excl = level_size;
2923 dstgroup->excl_cmpr = level_size;
2924 srcgroup->excl = level_size;
2925 srcgroup->excl_cmpr = level_size;
2927 /* inherit the limit info */
2928 dstgroup->lim_flags = srcgroup->lim_flags;
2929 dstgroup->max_rfer = srcgroup->max_rfer;
2930 dstgroup->max_excl = srcgroup->max_excl;
2931 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2932 dstgroup->rsv_excl = srcgroup->rsv_excl;
2934 qgroup_dirty(fs_info, dstgroup);
2935 qgroup_dirty(fs_info, srcgroup);
2941 i_qgroups = (u64 *)(inherit + 1);
2942 for (i = 0; i < inherit->num_qgroups; ++i) {
2944 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2951 * If we're doing a snapshot, and adding the snapshot to a new
2952 * qgroup, the numbers are guaranteed to be incorrect.
2958 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2959 struct btrfs_qgroup *src;
2960 struct btrfs_qgroup *dst;
2962 if (!i_qgroups[0] || !i_qgroups[1])
2965 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2966 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2973 dst->rfer = src->rfer - level_size;
2974 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2976 /* Manually tweaking numbers certainly needs a rescan */
2979 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2980 struct btrfs_qgroup *src;
2981 struct btrfs_qgroup *dst;
2983 if (!i_qgroups[0] || !i_qgroups[1])
2986 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2987 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2994 dst->excl = src->excl + level_size;
2995 dst->excl_cmpr = src->excl_cmpr + level_size;
3000 spin_unlock(&fs_info->qgroup_lock);
3002 ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
3005 mutex_unlock(&fs_info->qgroup_ioctl_lock);
3007 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3011 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
3013 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
3014 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
3017 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
3018 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
3024 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
3025 enum btrfs_qgroup_rsv_type type)
3027 struct btrfs_qgroup *qgroup;
3028 struct btrfs_fs_info *fs_info = root->fs_info;
3029 u64 ref_root = root->root_key.objectid;
3031 struct ulist_node *unode;
3032 struct ulist_iterator uiter;
3034 if (!is_fstree(ref_root))
3040 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
3041 capable(CAP_SYS_RESOURCE))
3044 spin_lock(&fs_info->qgroup_lock);
3045 if (!fs_info->quota_root)
3048 qgroup = find_qgroup_rb(fs_info, ref_root);
3053 * in a first step, we check all affected qgroups if any limits would
3056 ulist_reinit(fs_info->qgroup_ulist);
3057 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3058 qgroup_to_aux(qgroup), GFP_ATOMIC);
3061 ULIST_ITER_INIT(&uiter);
3062 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3063 struct btrfs_qgroup *qg;
3064 struct btrfs_qgroup_list *glist;
3066 qg = unode_aux_to_qgroup(unode);
3068 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
3073 list_for_each_entry(glist, &qg->groups, next_group) {
3074 ret = ulist_add(fs_info->qgroup_ulist,
3075 glist->group->qgroupid,
3076 qgroup_to_aux(glist->group), GFP_ATOMIC);
3083 * no limits exceeded, now record the reservation into all qgroups
3085 ULIST_ITER_INIT(&uiter);
3086 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3087 struct btrfs_qgroup *qg;
3089 qg = unode_aux_to_qgroup(unode);
3091 qgroup_rsv_add(fs_info, qg, num_bytes, type);
3095 spin_unlock(&fs_info->qgroup_lock);
3100 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
3103 * Will handle all higher level qgroup too.
3105 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3106 * This special case is only used for META_PERTRANS type.
3108 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3109 u64 ref_root, u64 num_bytes,
3110 enum btrfs_qgroup_rsv_type type)
3112 struct btrfs_qgroup *qgroup;
3113 struct ulist_node *unode;
3114 struct ulist_iterator uiter;
3117 if (!is_fstree(ref_root))
3123 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3124 WARN(1, "%s: Invalid type to free", __func__);
3127 spin_lock(&fs_info->qgroup_lock);
3129 if (!fs_info->quota_root)
3132 qgroup = find_qgroup_rb(fs_info, ref_root);
3136 if (num_bytes == (u64)-1)
3138 * We're freeing all pertrans rsv, get reserved value from
3139 * level 0 qgroup as real num_bytes to free.
3141 num_bytes = qgroup->rsv.values[type];
3143 ulist_reinit(fs_info->qgroup_ulist);
3144 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3145 qgroup_to_aux(qgroup), GFP_ATOMIC);
3148 ULIST_ITER_INIT(&uiter);
3149 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3150 struct btrfs_qgroup *qg;
3151 struct btrfs_qgroup_list *glist;
3153 qg = unode_aux_to_qgroup(unode);
3155 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3157 list_for_each_entry(glist, &qg->groups, next_group) {
3158 ret = ulist_add(fs_info->qgroup_ulist,
3159 glist->group->qgroupid,
3160 qgroup_to_aux(glist->group), GFP_ATOMIC);
3167 spin_unlock(&fs_info->qgroup_lock);
3171 * Check if the leaf is the last leaf. Which means all node pointers
3172 * are at their last position.
3174 static bool is_last_leaf(struct btrfs_path *path)
3178 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3179 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3186 * returns < 0 on error, 0 when more leafs are to be scanned.
3187 * returns 1 when done.
3189 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3190 struct btrfs_path *path)
3192 struct btrfs_fs_info *fs_info = trans->fs_info;
3193 struct btrfs_key found;
3194 struct extent_buffer *scratch_leaf = NULL;
3195 struct ulist *roots = NULL;
3201 mutex_lock(&fs_info->qgroup_rescan_lock);
3202 ret = btrfs_search_slot_for_read(fs_info->extent_root,
3203 &fs_info->qgroup_rescan_progress,
3206 btrfs_debug(fs_info,
3207 "current progress key (%llu %u %llu), search_slot ret %d",
3208 fs_info->qgroup_rescan_progress.objectid,
3209 fs_info->qgroup_rescan_progress.type,
3210 fs_info->qgroup_rescan_progress.offset, ret);
3214 * The rescan is about to end, we will not be scanning any
3215 * further blocks. We cannot unset the RESCAN flag here, because
3216 * we want to commit the transaction if everything went well.
3217 * To make the live accounting work in this phase, we set our
3218 * scan progress pointer such that every real extent objectid
3221 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3222 btrfs_release_path(path);
3223 mutex_unlock(&fs_info->qgroup_rescan_lock);
3226 done = is_last_leaf(path);
3228 btrfs_item_key_to_cpu(path->nodes[0], &found,
3229 btrfs_header_nritems(path->nodes[0]) - 1);
3230 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3232 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3233 if (!scratch_leaf) {
3235 mutex_unlock(&fs_info->qgroup_rescan_lock);
3238 slot = path->slots[0];
3239 btrfs_release_path(path);
3240 mutex_unlock(&fs_info->qgroup_rescan_lock);
3242 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3243 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3244 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3245 found.type != BTRFS_METADATA_ITEM_KEY)
3247 if (found.type == BTRFS_METADATA_ITEM_KEY)
3248 num_bytes = fs_info->nodesize;
3250 num_bytes = found.offset;
3252 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3256 /* For rescan, just pass old_roots as NULL */
3257 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3258 num_bytes, NULL, roots);
3264 free_extent_buffer(scratch_leaf);
3268 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3273 static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
3275 return btrfs_fs_closing(fs_info) ||
3276 test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state) ||
3277 !test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
3280 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3282 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3283 qgroup_rescan_work);
3284 struct btrfs_path *path;
3285 struct btrfs_trans_handle *trans = NULL;
3288 bool stopped = false;
3290 path = btrfs_alloc_path();
3294 * Rescan should only search for commit root, and any later difference
3295 * should be recorded by qgroup
3297 path->search_commit_root = 1;
3298 path->skip_locking = 1;
3301 while (!err && !(stopped = rescan_should_stop(fs_info))) {
3302 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3303 if (IS_ERR(trans)) {
3304 err = PTR_ERR(trans);
3308 err = qgroup_rescan_leaf(trans, path);
3311 btrfs_commit_transaction(trans);
3313 btrfs_end_transaction(trans);
3317 btrfs_free_path(path);
3319 mutex_lock(&fs_info->qgroup_rescan_lock);
3321 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3322 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3323 } else if (err < 0 || stopped) {
3324 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3326 mutex_unlock(&fs_info->qgroup_rescan_lock);
3329 * only update status, since the previous part has already updated the
3332 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3333 if (IS_ERR(trans)) {
3334 err = PTR_ERR(trans);
3337 "fail to start transaction for status update: %d",
3341 mutex_lock(&fs_info->qgroup_rescan_lock);
3343 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3345 ret = update_qgroup_status_item(trans);
3348 btrfs_err(fs_info, "fail to update qgroup status: %d",
3352 fs_info->qgroup_rescan_running = false;
3353 complete_all(&fs_info->qgroup_rescan_completion);
3354 mutex_unlock(&fs_info->qgroup_rescan_lock);
3359 btrfs_end_transaction(trans);
3362 btrfs_info(fs_info, "qgroup scan paused");
3363 } else if (err >= 0) {
3364 btrfs_info(fs_info, "qgroup scan completed%s",
3365 err > 0 ? " (inconsistency flag cleared)" : "");
3367 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3372 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3373 * memory required for the rescan context.
3376 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3382 /* we're resuming qgroup rescan at mount time */
3383 if (!(fs_info->qgroup_flags &
3384 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3386 "qgroup rescan init failed, qgroup rescan is not queued");
3388 } else if (!(fs_info->qgroup_flags &
3389 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3391 "qgroup rescan init failed, qgroup is not enabled");
3399 mutex_lock(&fs_info->qgroup_rescan_lock);
3402 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3404 "qgroup rescan is already in progress");
3406 } else if (!(fs_info->qgroup_flags &
3407 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3409 "qgroup rescan init failed, qgroup is not enabled");
3411 } else if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3412 /* Quota disable is in progress */
3417 mutex_unlock(&fs_info->qgroup_rescan_lock);
3420 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3423 memset(&fs_info->qgroup_rescan_progress, 0,
3424 sizeof(fs_info->qgroup_rescan_progress));
3425 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3426 init_completion(&fs_info->qgroup_rescan_completion);
3427 mutex_unlock(&fs_info->qgroup_rescan_lock);
3429 btrfs_init_work(&fs_info->qgroup_rescan_work,
3430 btrfs_qgroup_rescan_worker, NULL, NULL);
3435 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3438 struct btrfs_qgroup *qgroup;
3440 spin_lock(&fs_info->qgroup_lock);
3441 /* clear all current qgroup tracking information */
3442 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3443 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3445 qgroup->rfer_cmpr = 0;
3447 qgroup->excl_cmpr = 0;
3448 qgroup_dirty(fs_info, qgroup);
3450 spin_unlock(&fs_info->qgroup_lock);
3454 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3457 struct btrfs_trans_handle *trans;
3459 ret = qgroup_rescan_init(fs_info, 0, 1);
3464 * We have set the rescan_progress to 0, which means no more
3465 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3466 * However, btrfs_qgroup_account_ref may be right after its call
3467 * to btrfs_find_all_roots, in which case it would still do the
3469 * To solve this, we're committing the transaction, which will
3470 * ensure we run all delayed refs and only after that, we are
3471 * going to clear all tracking information for a clean start.
3474 trans = btrfs_join_transaction(fs_info->fs_root);
3475 if (IS_ERR(trans)) {
3476 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3477 return PTR_ERR(trans);
3479 ret = btrfs_commit_transaction(trans);
3481 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3485 qgroup_rescan_zero_tracking(fs_info);
3487 mutex_lock(&fs_info->qgroup_rescan_lock);
3488 fs_info->qgroup_rescan_running = true;
3489 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3490 &fs_info->qgroup_rescan_work);
3491 mutex_unlock(&fs_info->qgroup_rescan_lock);
3496 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3502 mutex_lock(&fs_info->qgroup_rescan_lock);
3503 running = fs_info->qgroup_rescan_running;
3504 mutex_unlock(&fs_info->qgroup_rescan_lock);
3510 ret = wait_for_completion_interruptible(
3511 &fs_info->qgroup_rescan_completion);
3513 wait_for_completion(&fs_info->qgroup_rescan_completion);
3519 * this is only called from open_ctree where we're still single threaded, thus
3520 * locking is omitted here.
3523 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3525 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3526 mutex_lock(&fs_info->qgroup_rescan_lock);
3527 fs_info->qgroup_rescan_running = true;
3528 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3529 &fs_info->qgroup_rescan_work);
3530 mutex_unlock(&fs_info->qgroup_rescan_lock);
3534 #define rbtree_iterate_from_safe(node, next, start) \
3535 for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3537 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3538 struct extent_changeset *reserved, u64 start,
3541 struct rb_node *node;
3542 struct rb_node *next;
3543 struct ulist_node *entry;
3546 node = reserved->range_changed.root.rb_node;
3550 entry = rb_entry(node, struct ulist_node, rb_node);
3551 if (entry->val < start)
3552 node = node->rb_right;
3554 node = node->rb_left;
3557 if (entry->val > start && rb_prev(&entry->rb_node))
3558 entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
3561 rbtree_iterate_from_safe(node, next, &entry->rb_node) {
3567 entry = rb_entry(node, struct ulist_node, rb_node);
3568 entry_start = entry->val;
3569 entry_end = entry->aux;
3570 entry_len = entry_end - entry_start + 1;
3572 if (entry_start >= start + len)
3574 if (entry_start + entry_len <= start)
3577 * Now the entry is in [start, start + len), revert the
3578 * EXTENT_QGROUP_RESERVED bit.
3580 clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
3581 entry_end, EXTENT_QGROUP_RESERVED);
3582 if (!ret && clear_ret < 0)
3585 ulist_del(&reserved->range_changed, entry->val, entry->aux);
3586 if (likely(reserved->bytes_changed >= entry_len)) {
3587 reserved->bytes_changed -= entry_len;
3590 reserved->bytes_changed = 0;
3598 * Try to free some space for qgroup.
3600 * For qgroup, there are only 3 ways to free qgroup space:
3601 * - Flush nodatacow write
3602 * Any nodatacow write will free its reserved data space at run_delalloc_range().
3603 * In theory, we should only flush nodatacow inodes, but it's not yet
3604 * possible, so we need to flush the whole root.
3606 * - Wait for ordered extents
3607 * When ordered extents are finished, their reserved metadata is finally
3608 * converted to per_trans status, which can be freed by later commit
3611 * - Commit transaction
3612 * This would free the meta_per_trans space.
3613 * In theory this shouldn't provide much space, but any more qgroup space
3616 static int try_flush_qgroup(struct btrfs_root *root)
3618 struct btrfs_trans_handle *trans;
3621 /* Can't hold an open transaction or we run the risk of deadlocking. */
3622 ASSERT(current->journal_info == NULL);
3623 if (WARN_ON(current->journal_info))
3627 * We don't want to run flush again and again, so if there is a running
3628 * one, we won't try to start a new flush, but exit directly.
3630 if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
3631 wait_event(root->qgroup_flush_wait,
3632 !test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
3636 ret = btrfs_start_delalloc_snapshot(root, true);
3639 btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
3641 trans = btrfs_join_transaction(root);
3642 if (IS_ERR(trans)) {
3643 ret = PTR_ERR(trans);
3647 ret = btrfs_commit_transaction(trans);
3649 clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
3650 wake_up(&root->qgroup_flush_wait);
3654 static int qgroup_reserve_data(struct btrfs_inode *inode,
3655 struct extent_changeset **reserved_ret, u64 start,
3658 struct btrfs_root *root = inode->root;
3659 struct extent_changeset *reserved;
3660 bool new_reserved = false;
3665 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3666 !is_fstree(root->root_key.objectid) || len == 0)
3669 /* @reserved parameter is mandatory for qgroup */
3670 if (WARN_ON(!reserved_ret))
3672 if (!*reserved_ret) {
3673 new_reserved = true;
3674 *reserved_ret = extent_changeset_alloc();
3678 reserved = *reserved_ret;
3679 /* Record already reserved space */
3680 orig_reserved = reserved->bytes_changed;
3681 ret = set_record_extent_bits(&inode->io_tree, start,
3682 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3684 /* Newly reserved space */
3685 to_reserve = reserved->bytes_changed - orig_reserved;
3686 trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
3687 to_reserve, QGROUP_RESERVE);
3690 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3697 qgroup_unreserve_range(inode, reserved, start, len);
3700 extent_changeset_free(reserved);
3701 *reserved_ret = NULL;
3707 * Reserve qgroup space for range [start, start + len).
3709 * This function will either reserve space from related qgroups or do nothing
3710 * if the range is already reserved.
3712 * Return 0 for successful reservation
3713 * Return <0 for error (including -EQUOT)
3715 * NOTE: This function may sleep for memory allocation, dirty page flushing and
3716 * commit transaction. So caller should not hold any dirty page locked.
3718 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
3719 struct extent_changeset **reserved_ret, u64 start,
3724 ret = qgroup_reserve_data(inode, reserved_ret, start, len);
3725 if (ret <= 0 && ret != -EDQUOT)
3728 ret = try_flush_qgroup(inode->root);
3731 return qgroup_reserve_data(inode, reserved_ret, start, len);
3734 /* Free ranges specified by @reserved, normally in error path */
3735 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
3736 struct extent_changeset *reserved, u64 start, u64 len)
3738 struct btrfs_root *root = inode->root;
3739 struct ulist_node *unode;
3740 struct ulist_iterator uiter;
3741 struct extent_changeset changeset;
3745 extent_changeset_init(&changeset);
3746 len = round_up(start + len, root->fs_info->sectorsize);
3747 start = round_down(start, root->fs_info->sectorsize);
3749 ULIST_ITER_INIT(&uiter);
3750 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3751 u64 range_start = unode->val;
3752 /* unode->aux is the inclusive end */
3753 u64 range_len = unode->aux - range_start + 1;
3757 extent_changeset_release(&changeset);
3759 /* Only free range in range [start, start + len) */
3760 if (range_start >= start + len ||
3761 range_start + range_len <= start)
3763 free_start = max(range_start, start);
3764 free_len = min(start + len, range_start + range_len) -
3767 * TODO: To also modify reserved->ranges_reserved to reflect
3770 * However as long as we free qgroup reserved according to
3771 * EXTENT_QGROUP_RESERVED, we won't double free.
3772 * So not need to rush.
3774 ret = clear_record_extent_bits(&inode->io_tree, free_start,
3775 free_start + free_len - 1,
3776 EXTENT_QGROUP_RESERVED, &changeset);
3779 freed += changeset.bytes_changed;
3781 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3782 BTRFS_QGROUP_RSV_DATA);
3785 extent_changeset_release(&changeset);
3789 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
3790 struct extent_changeset *reserved, u64 start, u64 len,
3793 struct extent_changeset changeset;
3794 int trace_op = QGROUP_RELEASE;
3797 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &inode->root->fs_info->flags))
3800 /* In release case, we shouldn't have @reserved */
3801 WARN_ON(!free && reserved);
3802 if (free && reserved)
3803 return qgroup_free_reserved_data(inode, reserved, start, len);
3804 extent_changeset_init(&changeset);
3805 ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
3806 EXTENT_QGROUP_RESERVED, &changeset);
3811 trace_op = QGROUP_FREE;
3812 trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
3813 changeset.bytes_changed, trace_op);
3815 btrfs_qgroup_free_refroot(inode->root->fs_info,
3816 inode->root->root_key.objectid,
3817 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3818 ret = changeset.bytes_changed;
3820 extent_changeset_release(&changeset);
3825 * Free a reserved space range from io_tree and related qgroups
3827 * Should be called when a range of pages get invalidated before reaching disk.
3828 * Or for error cleanup case.
3829 * if @reserved is given, only reserved range in [@start, @start + @len) will
3832 * For data written to disk, use btrfs_qgroup_release_data().
3834 * NOTE: This function may sleep for memory allocation.
3836 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
3837 struct extent_changeset *reserved, u64 start, u64 len)
3839 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3843 * Release a reserved space range from io_tree only.
3845 * Should be called when a range of pages get written to disk and corresponding
3846 * FILE_EXTENT is inserted into corresponding root.
3848 * Since new qgroup accounting framework will only update qgroup numbers at
3849 * commit_transaction() time, its reserved space shouldn't be freed from
3852 * But we should release the range from io_tree, to allow further write to be
3855 * NOTE: This function may sleep for memory allocation.
3857 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len)
3859 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3862 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3863 enum btrfs_qgroup_rsv_type type)
3865 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3866 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3871 spin_lock(&root->qgroup_meta_rsv_lock);
3872 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3873 root->qgroup_meta_rsv_prealloc += num_bytes;
3875 root->qgroup_meta_rsv_pertrans += num_bytes;
3876 spin_unlock(&root->qgroup_meta_rsv_lock);
3879 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3880 enum btrfs_qgroup_rsv_type type)
3882 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3883 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3888 spin_lock(&root->qgroup_meta_rsv_lock);
3889 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3890 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3892 root->qgroup_meta_rsv_prealloc -= num_bytes;
3894 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3896 root->qgroup_meta_rsv_pertrans -= num_bytes;
3898 spin_unlock(&root->qgroup_meta_rsv_lock);
3902 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3903 enum btrfs_qgroup_rsv_type type, bool enforce)
3905 struct btrfs_fs_info *fs_info = root->fs_info;
3908 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3909 !is_fstree(root->root_key.objectid) || num_bytes == 0)
3912 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3913 trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3914 ret = qgroup_reserve(root, num_bytes, enforce, type);
3918 * Record what we have reserved into root.
3920 * To avoid quota disabled->enabled underflow.
3921 * In that case, we may try to free space we haven't reserved
3922 * (since quota was disabled), so record what we reserved into root.
3923 * And ensure later release won't underflow this number.
3925 add_root_meta_rsv(root, num_bytes, type);
3929 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3930 enum btrfs_qgroup_rsv_type type, bool enforce)
3934 ret = btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
3935 if (ret <= 0 && ret != -EDQUOT)
3938 ret = try_flush_qgroup(root);
3941 return btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
3944 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3946 struct btrfs_fs_info *fs_info = root->fs_info;
3948 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3949 !is_fstree(root->root_key.objectid))
3952 /* TODO: Update trace point to handle such free */
3953 trace_qgroup_meta_free_all_pertrans(root);
3954 /* Special value -1 means to free all reserved space */
3955 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3956 BTRFS_QGROUP_RSV_META_PERTRANS);
3959 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3960 enum btrfs_qgroup_rsv_type type)
3962 struct btrfs_fs_info *fs_info = root->fs_info;
3964 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3965 !is_fstree(root->root_key.objectid))
3969 * reservation for META_PREALLOC can happen before quota is enabled,
3970 * which can lead to underflow.
3971 * Here ensure we will only free what we really have reserved.
3973 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3974 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3975 trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3976 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3980 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3983 struct btrfs_qgroup *qgroup;
3984 struct ulist_node *unode;
3985 struct ulist_iterator uiter;
3990 if (!fs_info->quota_root)
3993 spin_lock(&fs_info->qgroup_lock);
3994 qgroup = find_qgroup_rb(fs_info, ref_root);
3997 ulist_reinit(fs_info->qgroup_ulist);
3998 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3999 qgroup_to_aux(qgroup), GFP_ATOMIC);
4002 ULIST_ITER_INIT(&uiter);
4003 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
4004 struct btrfs_qgroup *qg;
4005 struct btrfs_qgroup_list *glist;
4007 qg = unode_aux_to_qgroup(unode);
4009 qgroup_rsv_release(fs_info, qg, num_bytes,
4010 BTRFS_QGROUP_RSV_META_PREALLOC);
4011 qgroup_rsv_add(fs_info, qg, num_bytes,
4012 BTRFS_QGROUP_RSV_META_PERTRANS);
4013 list_for_each_entry(glist, &qg->groups, next_group) {
4014 ret = ulist_add(fs_info->qgroup_ulist,
4015 glist->group->qgroupid,
4016 qgroup_to_aux(glist->group), GFP_ATOMIC);
4022 spin_unlock(&fs_info->qgroup_lock);
4025 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
4027 struct btrfs_fs_info *fs_info = root->fs_info;
4029 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
4030 !is_fstree(root->root_key.objectid))
4032 /* Same as btrfs_qgroup_free_meta_prealloc() */
4033 num_bytes = sub_root_meta_rsv(root, num_bytes,
4034 BTRFS_QGROUP_RSV_META_PREALLOC);
4035 trace_qgroup_meta_convert(root, num_bytes);
4036 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
4040 * Check qgroup reserved space leaking, normally at destroy inode
4043 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
4045 struct extent_changeset changeset;
4046 struct ulist_node *unode;
4047 struct ulist_iterator iter;
4050 extent_changeset_init(&changeset);
4051 ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4052 EXTENT_QGROUP_RESERVED, &changeset);
4055 if (WARN_ON(changeset.bytes_changed)) {
4056 ULIST_ITER_INIT(&iter);
4057 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4058 btrfs_warn(inode->root->fs_info,
4059 "leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4060 btrfs_ino(inode), unode->val, unode->aux);
4062 btrfs_qgroup_free_refroot(inode->root->fs_info,
4063 inode->root->root_key.objectid,
4064 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4067 extent_changeset_release(&changeset);
4070 void btrfs_qgroup_init_swapped_blocks(
4071 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4075 spin_lock_init(&swapped_blocks->lock);
4076 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4077 swapped_blocks->blocks[i] = RB_ROOT;
4078 swapped_blocks->swapped = false;
4082 * Delete all swapped blocks record of @root.
4083 * Every record here means we skipped a full subtree scan for qgroup.
4085 * Gets called when committing one transaction.
4087 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4089 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4092 swapped_blocks = &root->swapped_blocks;
4094 spin_lock(&swapped_blocks->lock);
4095 if (!swapped_blocks->swapped)
4097 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4098 struct rb_root *cur_root = &swapped_blocks->blocks[i];
4099 struct btrfs_qgroup_swapped_block *entry;
4100 struct btrfs_qgroup_swapped_block *next;
4102 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4105 swapped_blocks->blocks[i] = RB_ROOT;
4107 swapped_blocks->swapped = false;
4109 spin_unlock(&swapped_blocks->lock);
4113 * Add subtree roots record into @subvol_root.
4115 * @subvol_root: tree root of the subvolume tree get swapped
4116 * @bg: block group under balance
4117 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
4118 * @reloc_parent/slot: pointer to the subtree root in reloc tree
4119 * BOTH POINTERS ARE BEFORE TREE SWAP
4120 * @last_snapshot: last snapshot generation of the subvolume tree
4122 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4123 struct btrfs_root *subvol_root,
4124 struct btrfs_block_group *bg,
4125 struct extent_buffer *subvol_parent, int subvol_slot,
4126 struct extent_buffer *reloc_parent, int reloc_slot,
4129 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4130 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4131 struct btrfs_qgroup_swapped_block *block;
4132 struct rb_node **cur;
4133 struct rb_node *parent = NULL;
4134 int level = btrfs_header_level(subvol_parent) - 1;
4137 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4140 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4141 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4142 btrfs_err_rl(fs_info,
4143 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4145 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4146 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4150 block = kmalloc(sizeof(*block), GFP_NOFS);
4157 * @reloc_parent/slot is still before swap, while @block is going to
4158 * record the bytenr after swap, so we do the swap here.
4160 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4161 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4163 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4164 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4166 block->last_snapshot = last_snapshot;
4167 block->level = level;
4170 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4171 * no one else can modify tree blocks thus we qgroup will not change
4172 * no matter the value of trace_leaf.
4174 if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4175 block->trace_leaf = true;
4177 block->trace_leaf = false;
4178 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4180 /* Insert @block into @blocks */
4181 spin_lock(&blocks->lock);
4182 cur = &blocks->blocks[level].rb_node;
4184 struct btrfs_qgroup_swapped_block *entry;
4187 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4190 if (entry->subvol_bytenr < block->subvol_bytenr) {
4191 cur = &(*cur)->rb_left;
4192 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
4193 cur = &(*cur)->rb_right;
4195 if (entry->subvol_generation !=
4196 block->subvol_generation ||
4197 entry->reloc_bytenr != block->reloc_bytenr ||
4198 entry->reloc_generation !=
4199 block->reloc_generation) {
4201 * Duplicated but mismatch entry found.
4204 * Marking qgroup inconsistent should be enough
4207 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4214 rb_link_node(&block->node, parent, cur);
4215 rb_insert_color(&block->node, &blocks->blocks[level]);
4216 blocks->swapped = true;
4218 spin_unlock(&blocks->lock);
4221 fs_info->qgroup_flags |=
4222 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4227 * Check if the tree block is a subtree root, and if so do the needed
4228 * delayed subtree trace for qgroup.
4230 * This is called during btrfs_cow_block().
4232 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4233 struct btrfs_root *root,
4234 struct extent_buffer *subvol_eb)
4236 struct btrfs_fs_info *fs_info = root->fs_info;
4237 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4238 struct btrfs_qgroup_swapped_block *block;
4239 struct extent_buffer *reloc_eb = NULL;
4240 struct rb_node *node;
4242 bool swapped = false;
4243 int level = btrfs_header_level(subvol_eb);
4247 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
4249 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4252 spin_lock(&blocks->lock);
4253 if (!blocks->swapped) {
4254 spin_unlock(&blocks->lock);
4257 node = blocks->blocks[level].rb_node;
4260 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4261 if (block->subvol_bytenr < subvol_eb->start) {
4262 node = node->rb_left;
4263 } else if (block->subvol_bytenr > subvol_eb->start) {
4264 node = node->rb_right;
4271 spin_unlock(&blocks->lock);
4274 /* Found one, remove it from @blocks first and update blocks->swapped */
4275 rb_erase(&block->node, &blocks->blocks[level]);
4276 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4277 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4282 blocks->swapped = swapped;
4283 spin_unlock(&blocks->lock);
4285 /* Read out reloc subtree root */
4286 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr, 0,
4287 block->reloc_generation, block->level,
4289 if (IS_ERR(reloc_eb)) {
4290 ret = PTR_ERR(reloc_eb);
4294 if (!extent_buffer_uptodate(reloc_eb)) {
4299 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4300 block->last_snapshot, block->trace_leaf);
4303 free_extent_buffer(reloc_eb);
4306 btrfs_err_rl(fs_info,
4307 "failed to account subtree at bytenr %llu: %d",
4308 subvol_eb->start, ret);
4309 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
4314 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4316 struct btrfs_qgroup_extent_record *entry;
4317 struct btrfs_qgroup_extent_record *next;
4318 struct rb_root *root;
4320 root = &trans->delayed_refs.dirty_extent_root;
4321 rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4322 ulist_free(entry->old_roots);