1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2011 STRATO. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sched/mm.h>
17 #include "transaction.h"
22 #include "extent_io.h"
24 #include "block-group.h"
26 #include "tree-mod-log.h"
28 #include "accessors.h"
29 #include "extent-tree.h"
30 #include "root-tree.h"
31 #include "tree-checker.h"
33 enum btrfs_qgroup_mode btrfs_qgroup_mode(struct btrfs_fs_info *fs_info)
35 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
36 return BTRFS_QGROUP_MODE_DISABLED;
37 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_SIMPLE_MODE)
38 return BTRFS_QGROUP_MODE_SIMPLE;
39 return BTRFS_QGROUP_MODE_FULL;
42 bool btrfs_qgroup_enabled(struct btrfs_fs_info *fs_info)
44 return btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_DISABLED;
47 bool btrfs_qgroup_full_accounting(struct btrfs_fs_info *fs_info)
49 return btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL;
53 * Helpers to access qgroup reservation
55 * Callers should ensure the lock context and type are valid
58 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
63 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
64 ret += qgroup->rsv.values[i];
69 #ifdef CONFIG_BTRFS_DEBUG
70 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
72 if (type == BTRFS_QGROUP_RSV_DATA)
74 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
75 return "meta_pertrans";
76 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
77 return "meta_prealloc";
82 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
83 struct btrfs_qgroup *qgroup, u64 num_bytes,
84 enum btrfs_qgroup_rsv_type type)
86 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
87 qgroup->rsv.values[type] += num_bytes;
90 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
91 struct btrfs_qgroup *qgroup, u64 num_bytes,
92 enum btrfs_qgroup_rsv_type type)
94 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
95 if (qgroup->rsv.values[type] >= num_bytes) {
96 qgroup->rsv.values[type] -= num_bytes;
99 #ifdef CONFIG_BTRFS_DEBUG
101 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
102 qgroup->qgroupid, qgroup_rsv_type_str(type),
103 qgroup->rsv.values[type], num_bytes);
105 qgroup->rsv.values[type] = 0;
108 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
109 struct btrfs_qgroup *dest,
110 struct btrfs_qgroup *src)
114 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
115 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
118 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
119 struct btrfs_qgroup *dest,
120 struct btrfs_qgroup *src)
124 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
125 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
128 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
131 if (qg->old_refcnt < seq)
132 qg->old_refcnt = seq;
133 qg->old_refcnt += mod;
136 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
139 if (qg->new_refcnt < seq)
140 qg->new_refcnt = seq;
141 qg->new_refcnt += mod;
144 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
146 if (qg->old_refcnt < seq)
148 return qg->old_refcnt - seq;
151 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
153 if (qg->new_refcnt < seq)
155 return qg->new_refcnt - seq;
159 * glue structure to represent the relations between qgroups.
161 struct btrfs_qgroup_list {
162 struct list_head next_group;
163 struct list_head next_member;
164 struct btrfs_qgroup *group;
165 struct btrfs_qgroup *member;
169 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
171 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
173 /* must be called with qgroup_ioctl_lock held */
174 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
177 struct rb_node *n = fs_info->qgroup_tree.rb_node;
178 struct btrfs_qgroup *qgroup;
181 qgroup = rb_entry(n, struct btrfs_qgroup, node);
182 if (qgroup->qgroupid < qgroupid)
184 else if (qgroup->qgroupid > qgroupid)
193 * Add qgroup to the filesystem's qgroup tree.
195 * Must be called with qgroup_lock held and @prealloc preallocated.
197 * The control on the lifespan of @prealloc would be transferred to this
198 * function, thus caller should no longer touch @prealloc.
200 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
201 struct btrfs_qgroup *prealloc,
204 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
205 struct rb_node *parent = NULL;
206 struct btrfs_qgroup *qgroup;
208 /* Caller must have pre-allocated @prealloc. */
213 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
215 if (qgroup->qgroupid < qgroupid) {
217 } else if (qgroup->qgroupid > qgroupid) {
226 qgroup->qgroupid = qgroupid;
227 INIT_LIST_HEAD(&qgroup->groups);
228 INIT_LIST_HEAD(&qgroup->members);
229 INIT_LIST_HEAD(&qgroup->dirty);
230 INIT_LIST_HEAD(&qgroup->iterator);
231 INIT_LIST_HEAD(&qgroup->nested_iterator);
233 rb_link_node(&qgroup->node, parent, p);
234 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
239 static void __del_qgroup_rb(struct btrfs_fs_info *fs_info,
240 struct btrfs_qgroup *qgroup)
242 struct btrfs_qgroup_list *list;
244 list_del(&qgroup->dirty);
245 while (!list_empty(&qgroup->groups)) {
246 list = list_first_entry(&qgroup->groups,
247 struct btrfs_qgroup_list, next_group);
248 list_del(&list->next_group);
249 list_del(&list->next_member);
253 while (!list_empty(&qgroup->members)) {
254 list = list_first_entry(&qgroup->members,
255 struct btrfs_qgroup_list, next_member);
256 list_del(&list->next_group);
257 list_del(&list->next_member);
262 /* must be called with qgroup_lock held */
263 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
265 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
270 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
271 __del_qgroup_rb(fs_info, qgroup);
276 * Add relation specified by two qgroups.
278 * Must be called with qgroup_lock held, the ownership of @prealloc is
279 * transferred to this function and caller should not touch it anymore.
281 * Return: 0 on success
282 * -ENOENT if one of the qgroups is NULL
285 static int __add_relation_rb(struct btrfs_qgroup_list *prealloc,
286 struct btrfs_qgroup *member,
287 struct btrfs_qgroup *parent)
289 if (!member || !parent) {
294 prealloc->group = parent;
295 prealloc->member = member;
296 list_add_tail(&prealloc->next_group, &member->groups);
297 list_add_tail(&prealloc->next_member, &parent->members);
303 * Add relation specified by two qgroup ids.
305 * Must be called with qgroup_lock held.
307 * Return: 0 on success
308 * -ENOENT if one of the ids does not exist
311 static int add_relation_rb(struct btrfs_fs_info *fs_info,
312 struct btrfs_qgroup_list *prealloc,
313 u64 memberid, u64 parentid)
315 struct btrfs_qgroup *member;
316 struct btrfs_qgroup *parent;
318 member = find_qgroup_rb(fs_info, memberid);
319 parent = find_qgroup_rb(fs_info, parentid);
321 return __add_relation_rb(prealloc, member, parent);
324 /* Must be called with qgroup_lock held */
325 static int del_relation_rb(struct btrfs_fs_info *fs_info,
326 u64 memberid, u64 parentid)
328 struct btrfs_qgroup *member;
329 struct btrfs_qgroup *parent;
330 struct btrfs_qgroup_list *list;
332 member = find_qgroup_rb(fs_info, memberid);
333 parent = find_qgroup_rb(fs_info, parentid);
334 if (!member || !parent)
337 list_for_each_entry(list, &member->groups, next_group) {
338 if (list->group == parent) {
339 list_del(&list->next_group);
340 list_del(&list->next_member);
348 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
349 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
352 struct btrfs_qgroup *qgroup;
354 qgroup = find_qgroup_rb(fs_info, qgroupid);
357 if (qgroup->rfer != rfer || qgroup->excl != excl)
363 static void qgroup_mark_inconsistent(struct btrfs_fs_info *fs_info)
365 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
367 fs_info->qgroup_flags |= (BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT |
368 BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
369 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
372 static void qgroup_read_enable_gen(struct btrfs_fs_info *fs_info,
373 struct extent_buffer *leaf, int slot,
374 struct btrfs_qgroup_status_item *ptr)
376 ASSERT(btrfs_fs_incompat(fs_info, SIMPLE_QUOTA));
377 ASSERT(btrfs_item_size(leaf, slot) >= sizeof(*ptr));
378 fs_info->qgroup_enable_gen = btrfs_qgroup_status_enable_gen(leaf, ptr);
382 * The full config is read in one go, only called from open_ctree()
383 * It doesn't use any locking, as at this point we're still single-threaded
385 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
387 struct btrfs_key key;
388 struct btrfs_key found_key;
389 struct btrfs_root *quota_root = fs_info->quota_root;
390 struct btrfs_path *path = NULL;
391 struct extent_buffer *l;
395 u64 rescan_progress = 0;
397 if (!fs_info->quota_root)
400 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
401 if (!fs_info->qgroup_ulist) {
406 path = btrfs_alloc_path();
412 ret = btrfs_sysfs_add_qgroups(fs_info);
415 /* default this to quota off, in case no status key is found */
416 fs_info->qgroup_flags = 0;
419 * pass 1: read status, all qgroup infos and limits
424 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
429 struct btrfs_qgroup *qgroup;
431 slot = path->slots[0];
433 btrfs_item_key_to_cpu(l, &found_key, slot);
435 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
436 struct btrfs_qgroup_status_item *ptr;
438 ptr = btrfs_item_ptr(l, slot,
439 struct btrfs_qgroup_status_item);
441 if (btrfs_qgroup_status_version(l, ptr) !=
442 BTRFS_QGROUP_STATUS_VERSION) {
444 "old qgroup version, quota disabled");
447 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l, ptr);
448 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_SIMPLE_MODE) {
449 qgroup_read_enable_gen(fs_info, l, slot, ptr);
450 } else if (btrfs_qgroup_status_generation(l, ptr) != fs_info->generation) {
451 qgroup_mark_inconsistent(fs_info);
453 "qgroup generation mismatch, marked as inconsistent");
455 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
459 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
460 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
463 qgroup = find_qgroup_rb(fs_info, found_key.offset);
464 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
465 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
466 btrfs_err(fs_info, "inconsistent qgroup config");
467 qgroup_mark_inconsistent(fs_info);
470 struct btrfs_qgroup *prealloc;
472 prealloc = kzalloc(sizeof(*prealloc), GFP_KERNEL);
477 qgroup = add_qgroup_rb(fs_info, prealloc, found_key.offset);
479 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
483 switch (found_key.type) {
484 case BTRFS_QGROUP_INFO_KEY: {
485 struct btrfs_qgroup_info_item *ptr;
487 ptr = btrfs_item_ptr(l, slot,
488 struct btrfs_qgroup_info_item);
489 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
490 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
491 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
492 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
493 /* generation currently unused */
496 case BTRFS_QGROUP_LIMIT_KEY: {
497 struct btrfs_qgroup_limit_item *ptr;
499 ptr = btrfs_item_ptr(l, slot,
500 struct btrfs_qgroup_limit_item);
501 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
502 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
503 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
504 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
505 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
510 ret = btrfs_next_item(quota_root, path);
516 btrfs_release_path(path);
519 * pass 2: read all qgroup relations
522 key.type = BTRFS_QGROUP_RELATION_KEY;
524 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
528 struct btrfs_qgroup_list *list = NULL;
530 slot = path->slots[0];
532 btrfs_item_key_to_cpu(l, &found_key, slot);
534 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
537 if (found_key.objectid > found_key.offset) {
538 /* parent <- member, not needed to build config */
539 /* FIXME should we omit the key completely? */
543 list = kzalloc(sizeof(*list), GFP_KERNEL);
548 ret = add_relation_rb(fs_info, list, found_key.objectid,
551 if (ret == -ENOENT) {
553 "orphan qgroup relation 0x%llx->0x%llx",
554 found_key.objectid, found_key.offset);
555 ret = 0; /* ignore the error */
560 ret = btrfs_next_item(quota_root, path);
567 btrfs_free_path(path);
568 fs_info->qgroup_flags |= flags;
570 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON)
571 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
572 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
573 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
575 ulist_free(fs_info->qgroup_ulist);
576 fs_info->qgroup_ulist = NULL;
577 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
578 btrfs_sysfs_del_qgroups(fs_info);
581 return ret < 0 ? ret : 0;
585 * Called in close_ctree() when quota is still enabled. This verifies we don't
586 * leak some reserved space.
588 * Return false if no reserved space is left.
589 * Return true if some reserved space is leaked.
591 bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info)
593 struct rb_node *node;
596 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED)
599 * Since we're unmounting, there is no race and no need to grab qgroup
600 * lock. And here we don't go post-order to provide a more user
601 * friendly sorted result.
603 for (node = rb_first(&fs_info->qgroup_tree); node; node = rb_next(node)) {
604 struct btrfs_qgroup *qgroup;
607 qgroup = rb_entry(node, struct btrfs_qgroup, node);
608 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++) {
609 if (qgroup->rsv.values[i]) {
612 "qgroup %hu/%llu has unreleased space, type %d rsv %llu",
613 btrfs_qgroup_level(qgroup->qgroupid),
614 btrfs_qgroup_subvolid(qgroup->qgroupid),
615 i, qgroup->rsv.values[i]);
623 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
624 * first two are in single-threaded paths.And for the third one, we have set
625 * quota_root to be null with qgroup_lock held before, so it is safe to clean
626 * up the in-memory structures without qgroup_lock held.
628 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
631 struct btrfs_qgroup *qgroup;
633 while ((n = rb_first(&fs_info->qgroup_tree))) {
634 qgroup = rb_entry(n, struct btrfs_qgroup, node);
635 rb_erase(n, &fs_info->qgroup_tree);
636 __del_qgroup_rb(fs_info, qgroup);
637 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
641 * We call btrfs_free_qgroup_config() when unmounting
642 * filesystem and disabling quota, so we set qgroup_ulist
643 * to be null here to avoid double free.
645 ulist_free(fs_info->qgroup_ulist);
646 fs_info->qgroup_ulist = NULL;
647 btrfs_sysfs_del_qgroups(fs_info);
650 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
654 struct btrfs_root *quota_root = trans->fs_info->quota_root;
655 struct btrfs_path *path;
656 struct btrfs_key key;
658 path = btrfs_alloc_path();
663 key.type = BTRFS_QGROUP_RELATION_KEY;
666 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
668 btrfs_mark_buffer_dirty(trans, path->nodes[0]);
670 btrfs_free_path(path);
674 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
678 struct btrfs_root *quota_root = trans->fs_info->quota_root;
679 struct btrfs_path *path;
680 struct btrfs_key key;
682 path = btrfs_alloc_path();
687 key.type = BTRFS_QGROUP_RELATION_KEY;
690 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
699 ret = btrfs_del_item(trans, quota_root, path);
701 btrfs_free_path(path);
705 static int add_qgroup_item(struct btrfs_trans_handle *trans,
706 struct btrfs_root *quota_root, u64 qgroupid)
709 struct btrfs_path *path;
710 struct btrfs_qgroup_info_item *qgroup_info;
711 struct btrfs_qgroup_limit_item *qgroup_limit;
712 struct extent_buffer *leaf;
713 struct btrfs_key key;
715 if (btrfs_is_testing(quota_root->fs_info))
718 path = btrfs_alloc_path();
723 key.type = BTRFS_QGROUP_INFO_KEY;
724 key.offset = qgroupid;
727 * Avoid a transaction abort by catching -EEXIST here. In that
728 * case, we proceed by re-initializing the existing structure
732 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
733 sizeof(*qgroup_info));
734 if (ret && ret != -EEXIST)
737 leaf = path->nodes[0];
738 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
739 struct btrfs_qgroup_info_item);
740 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
741 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
742 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
743 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
744 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
746 btrfs_mark_buffer_dirty(trans, leaf);
748 btrfs_release_path(path);
750 key.type = BTRFS_QGROUP_LIMIT_KEY;
751 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
752 sizeof(*qgroup_limit));
753 if (ret && ret != -EEXIST)
756 leaf = path->nodes[0];
757 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
758 struct btrfs_qgroup_limit_item);
759 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
760 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
761 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
762 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
763 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
765 btrfs_mark_buffer_dirty(trans, leaf);
769 btrfs_free_path(path);
773 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
776 struct btrfs_root *quota_root = trans->fs_info->quota_root;
777 struct btrfs_path *path;
778 struct btrfs_key key;
780 path = btrfs_alloc_path();
785 key.type = BTRFS_QGROUP_INFO_KEY;
786 key.offset = qgroupid;
787 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
796 ret = btrfs_del_item(trans, quota_root, path);
800 btrfs_release_path(path);
802 key.type = BTRFS_QGROUP_LIMIT_KEY;
803 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
812 ret = btrfs_del_item(trans, quota_root, path);
815 btrfs_free_path(path);
819 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
820 struct btrfs_qgroup *qgroup)
822 struct btrfs_root *quota_root = trans->fs_info->quota_root;
823 struct btrfs_path *path;
824 struct btrfs_key key;
825 struct extent_buffer *l;
826 struct btrfs_qgroup_limit_item *qgroup_limit;
831 key.type = BTRFS_QGROUP_LIMIT_KEY;
832 key.offset = qgroup->qgroupid;
834 path = btrfs_alloc_path();
838 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
846 slot = path->slots[0];
847 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
848 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
849 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
850 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
851 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
852 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
854 btrfs_mark_buffer_dirty(trans, l);
857 btrfs_free_path(path);
861 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
862 struct btrfs_qgroup *qgroup)
864 struct btrfs_fs_info *fs_info = trans->fs_info;
865 struct btrfs_root *quota_root = fs_info->quota_root;
866 struct btrfs_path *path;
867 struct btrfs_key key;
868 struct extent_buffer *l;
869 struct btrfs_qgroup_info_item *qgroup_info;
873 if (btrfs_is_testing(fs_info))
877 key.type = BTRFS_QGROUP_INFO_KEY;
878 key.offset = qgroup->qgroupid;
880 path = btrfs_alloc_path();
884 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
892 slot = path->slots[0];
893 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
894 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
895 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
896 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
897 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
898 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
900 btrfs_mark_buffer_dirty(trans, l);
903 btrfs_free_path(path);
907 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
909 struct btrfs_fs_info *fs_info = trans->fs_info;
910 struct btrfs_root *quota_root = fs_info->quota_root;
911 struct btrfs_path *path;
912 struct btrfs_key key;
913 struct extent_buffer *l;
914 struct btrfs_qgroup_status_item *ptr;
919 key.type = BTRFS_QGROUP_STATUS_KEY;
922 path = btrfs_alloc_path();
926 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
934 slot = path->slots[0];
935 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
936 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags &
937 BTRFS_QGROUP_STATUS_FLAGS_MASK);
938 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
939 btrfs_set_qgroup_status_rescan(l, ptr,
940 fs_info->qgroup_rescan_progress.objectid);
942 btrfs_mark_buffer_dirty(trans, l);
945 btrfs_free_path(path);
950 * called with qgroup_lock held
952 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
953 struct btrfs_root *root)
955 struct btrfs_path *path;
956 struct btrfs_key key;
957 struct extent_buffer *leaf = NULL;
961 path = btrfs_alloc_path();
970 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
973 leaf = path->nodes[0];
974 nr = btrfs_header_nritems(leaf);
978 * delete the leaf one by one
979 * since the whole tree is going
983 ret = btrfs_del_items(trans, root, path, 0, nr);
987 btrfs_release_path(path);
991 btrfs_free_path(path);
995 int btrfs_quota_enable(struct btrfs_fs_info *fs_info,
996 struct btrfs_ioctl_quota_ctl_args *quota_ctl_args)
998 struct btrfs_root *quota_root;
999 struct btrfs_root *tree_root = fs_info->tree_root;
1000 struct btrfs_path *path = NULL;
1001 struct btrfs_qgroup_status_item *ptr;
1002 struct extent_buffer *leaf;
1003 struct btrfs_key key;
1004 struct btrfs_key found_key;
1005 struct btrfs_qgroup *qgroup = NULL;
1006 struct btrfs_qgroup *prealloc = NULL;
1007 struct btrfs_trans_handle *trans = NULL;
1008 struct ulist *ulist = NULL;
1009 const bool simple = (quota_ctl_args->cmd == BTRFS_QUOTA_CTL_ENABLE_SIMPLE_QUOTA);
1014 * We need to have subvol_sem write locked, to prevent races between
1015 * concurrent tasks trying to enable quotas, because we will unlock
1016 * and relock qgroup_ioctl_lock before setting fs_info->quota_root
1017 * and before setting BTRFS_FS_QUOTA_ENABLED.
1019 lockdep_assert_held_write(&fs_info->subvol_sem);
1021 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
1023 "qgroups are currently unsupported in extent tree v2");
1027 mutex_lock(&fs_info->qgroup_ioctl_lock);
1028 if (fs_info->quota_root)
1031 ulist = ulist_alloc(GFP_KERNEL);
1037 ret = btrfs_sysfs_add_qgroups(fs_info);
1042 * Unlock qgroup_ioctl_lock before starting the transaction. This is to
1043 * avoid lock acquisition inversion problems (reported by lockdep) between
1044 * qgroup_ioctl_lock and the vfs freeze semaphores, acquired when we
1045 * start a transaction.
1046 * After we started the transaction lock qgroup_ioctl_lock again and
1047 * check if someone else created the quota root in the meanwhile. If so,
1048 * just return success and release the transaction handle.
1050 * Also we don't need to worry about someone else calling
1051 * btrfs_sysfs_add_qgroups() after we unlock and getting an error because
1052 * that function returns 0 (success) when the sysfs entries already exist.
1054 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1057 * 1 for quota root item
1058 * 1 for BTRFS_QGROUP_STATUS item
1060 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
1061 * per subvolume. However those are not currently reserved since it
1062 * would be a lot of overkill.
1064 trans = btrfs_start_transaction(tree_root, 2);
1066 mutex_lock(&fs_info->qgroup_ioctl_lock);
1067 if (IS_ERR(trans)) {
1068 ret = PTR_ERR(trans);
1073 if (fs_info->quota_root)
1076 fs_info->qgroup_ulist = ulist;
1080 * initially create the quota tree
1082 quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
1083 if (IS_ERR(quota_root)) {
1084 ret = PTR_ERR(quota_root);
1085 btrfs_abort_transaction(trans, ret);
1089 path = btrfs_alloc_path();
1092 btrfs_abort_transaction(trans, ret);
1097 key.type = BTRFS_QGROUP_STATUS_KEY;
1100 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
1103 btrfs_abort_transaction(trans, ret);
1107 leaf = path->nodes[0];
1108 ptr = btrfs_item_ptr(leaf, path->slots[0],
1109 struct btrfs_qgroup_status_item);
1110 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
1111 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
1112 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON;
1114 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_SIMPLE_MODE;
1115 btrfs_set_qgroup_status_enable_gen(leaf, ptr, trans->transid);
1117 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1119 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags &
1120 BTRFS_QGROUP_STATUS_FLAGS_MASK);
1121 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
1123 btrfs_mark_buffer_dirty(trans, leaf);
1126 key.type = BTRFS_ROOT_REF_KEY;
1129 btrfs_release_path(path);
1130 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
1134 btrfs_abort_transaction(trans, ret);
1139 slot = path->slots[0];
1140 leaf = path->nodes[0];
1141 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1143 if (found_key.type == BTRFS_ROOT_REF_KEY) {
1145 /* Release locks on tree_root before we access quota_root */
1146 btrfs_release_path(path);
1148 /* We should not have a stray @prealloc pointer. */
1149 ASSERT(prealloc == NULL);
1150 prealloc = kzalloc(sizeof(*prealloc), GFP_NOFS);
1153 btrfs_abort_transaction(trans, ret);
1157 ret = add_qgroup_item(trans, quota_root,
1160 btrfs_abort_transaction(trans, ret);
1164 qgroup = add_qgroup_rb(fs_info, prealloc, found_key.offset);
1166 if (IS_ERR(qgroup)) {
1167 ret = PTR_ERR(qgroup);
1168 btrfs_abort_transaction(trans, ret);
1171 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1173 btrfs_abort_transaction(trans, ret);
1176 ret = btrfs_search_slot_for_read(tree_root, &found_key,
1179 btrfs_abort_transaction(trans, ret);
1184 * Shouldn't happen, but in case it does we
1185 * don't need to do the btrfs_next_item, just
1191 ret = btrfs_next_item(tree_root, path);
1193 btrfs_abort_transaction(trans, ret);
1201 btrfs_release_path(path);
1202 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1204 btrfs_abort_transaction(trans, ret);
1208 ASSERT(prealloc == NULL);
1209 prealloc = kzalloc(sizeof(*prealloc), GFP_NOFS);
1214 qgroup = add_qgroup_rb(fs_info, prealloc, BTRFS_FS_TREE_OBJECTID);
1216 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1218 btrfs_abort_transaction(trans, ret);
1222 fs_info->qgroup_enable_gen = trans->transid;
1224 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1226 * Commit the transaction while not holding qgroup_ioctl_lock, to avoid
1227 * a deadlock with tasks concurrently doing other qgroup operations, such
1228 * adding/removing qgroups or adding/deleting qgroup relations for example,
1229 * because all qgroup operations first start or join a transaction and then
1230 * lock the qgroup_ioctl_lock mutex.
1231 * We are safe from a concurrent task trying to enable quotas, by calling
1232 * this function, since we are serialized by fs_info->subvol_sem.
1234 ret = btrfs_commit_transaction(trans);
1236 mutex_lock(&fs_info->qgroup_ioctl_lock);
1241 * Set quota enabled flag after committing the transaction, to avoid
1242 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1245 spin_lock(&fs_info->qgroup_lock);
1246 fs_info->quota_root = quota_root;
1247 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1249 btrfs_set_fs_incompat(fs_info, SIMPLE_QUOTA);
1250 spin_unlock(&fs_info->qgroup_lock);
1252 /* Skip rescan for simple qgroups. */
1253 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
1256 ret = qgroup_rescan_init(fs_info, 0, 1);
1258 qgroup_rescan_zero_tracking(fs_info);
1259 fs_info->qgroup_rescan_running = true;
1260 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1261 &fs_info->qgroup_rescan_work);
1264 * We have set both BTRFS_FS_QUOTA_ENABLED and
1265 * BTRFS_QGROUP_STATUS_FLAG_ON, so we can only fail with
1266 * -EINPROGRESS. That can happen because someone started the
1267 * rescan worker by calling quota rescan ioctl before we
1268 * attempted to initialize the rescan worker. Failure due to
1269 * quotas disabled in the meanwhile is not possible, because
1270 * we are holding a write lock on fs_info->subvol_sem, which
1271 * is also acquired when disabling quotas.
1272 * Ignore such error, and any other error would need to undo
1273 * everything we did in the transaction we just committed.
1275 ASSERT(ret == -EINPROGRESS);
1280 btrfs_free_path(path);
1283 btrfs_put_root(quota_root);
1286 ulist_free(fs_info->qgroup_ulist);
1287 fs_info->qgroup_ulist = NULL;
1288 btrfs_sysfs_del_qgroups(fs_info);
1290 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1292 btrfs_end_transaction(trans);
1294 ret = btrfs_end_transaction(trans);
1301 * It is possible to have outstanding ordered extents which reserved bytes
1302 * before we disabled. We need to fully flush delalloc, ordered extents, and a
1303 * commit to ensure that we don't leak such reservations, only to have them
1304 * come back if we re-enable.
1306 * - enable simple quotas
1308 * - release it, store rsv_bytes in OE
1310 * - enable simple quotas (qgroup rsv are all 0)
1312 * - run delayed refs
1313 * - free rsv_bytes, resulting in miscounting or even underflow
1315 static int flush_reservations(struct btrfs_fs_info *fs_info)
1317 struct btrfs_trans_handle *trans;
1320 ret = btrfs_start_delalloc_roots(fs_info, LONG_MAX, false);
1323 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1);
1324 trans = btrfs_join_transaction(fs_info->tree_root);
1326 return PTR_ERR(trans);
1327 ret = btrfs_commit_transaction(trans);
1332 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1334 struct btrfs_root *quota_root;
1335 struct btrfs_trans_handle *trans = NULL;
1339 * We need to have subvol_sem write locked to prevent races with
1340 * snapshot creation.
1342 lockdep_assert_held_write(&fs_info->subvol_sem);
1345 * Relocation will mess with backrefs, so make sure we have the
1346 * cleaner_mutex held to protect us from relocate.
1348 lockdep_assert_held(&fs_info->cleaner_mutex);
1350 mutex_lock(&fs_info->qgroup_ioctl_lock);
1351 if (!fs_info->quota_root)
1355 * Unlock the qgroup_ioctl_lock mutex before waiting for the rescan worker to
1356 * complete. Otherwise we can deadlock because btrfs_remove_qgroup() needs
1357 * to lock that mutex while holding a transaction handle and the rescan
1358 * worker needs to commit a transaction.
1360 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1363 * Request qgroup rescan worker to complete and wait for it. This wait
1364 * must be done before transaction start for quota disable since it may
1365 * deadlock with transaction by the qgroup rescan worker.
1367 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1368 btrfs_qgroup_wait_for_completion(fs_info, false);
1371 * We have nothing held here and no trans handle, just return the error
1374 ret = flush_reservations(fs_info);
1379 * 1 For the root item
1381 * We should also reserve enough items for the quota tree deletion in
1382 * btrfs_clean_quota_tree but this is not done.
1384 * Also, we must always start a transaction without holding the mutex
1385 * qgroup_ioctl_lock, see btrfs_quota_enable().
1387 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1389 mutex_lock(&fs_info->qgroup_ioctl_lock);
1390 if (IS_ERR(trans)) {
1391 ret = PTR_ERR(trans);
1393 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1397 if (!fs_info->quota_root)
1400 spin_lock(&fs_info->qgroup_lock);
1401 quota_root = fs_info->quota_root;
1402 fs_info->quota_root = NULL;
1403 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1404 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_SIMPLE_MODE;
1405 fs_info->qgroup_drop_subtree_thres = BTRFS_MAX_LEVEL;
1406 spin_unlock(&fs_info->qgroup_lock);
1408 btrfs_free_qgroup_config(fs_info);
1410 ret = btrfs_clean_quota_tree(trans, quota_root);
1412 btrfs_abort_transaction(trans, ret);
1416 ret = btrfs_del_root(trans, "a_root->root_key);
1418 btrfs_abort_transaction(trans, ret);
1422 spin_lock(&fs_info->trans_lock);
1423 list_del("a_root->dirty_list);
1424 spin_unlock(&fs_info->trans_lock);
1426 btrfs_tree_lock(quota_root->node);
1427 btrfs_clear_buffer_dirty(trans, quota_root->node);
1428 btrfs_tree_unlock(quota_root->node);
1429 btrfs_free_tree_block(trans, btrfs_root_id(quota_root),
1430 quota_root->node, 0, 1);
1432 btrfs_put_root(quota_root);
1435 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1437 btrfs_end_transaction(trans);
1439 ret = btrfs_commit_transaction(trans);
1443 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1444 struct btrfs_qgroup *qgroup)
1446 if (list_empty(&qgroup->dirty))
1447 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1450 static void qgroup_iterator_add(struct list_head *head, struct btrfs_qgroup *qgroup)
1452 if (!list_empty(&qgroup->iterator))
1455 list_add_tail(&qgroup->iterator, head);
1458 static void qgroup_iterator_clean(struct list_head *head)
1460 while (!list_empty(head)) {
1461 struct btrfs_qgroup *qgroup;
1463 qgroup = list_first_entry(head, struct btrfs_qgroup, iterator);
1464 list_del_init(&qgroup->iterator);
1469 * The easy accounting, we're updating qgroup relationship whose child qgroup
1470 * only has exclusive extents.
1472 * In this case, all exclusive extents will also be exclusive for parent, so
1473 * excl/rfer just get added/removed.
1475 * So is qgroup reservation space, which should also be added/removed to
1477 * Or when child tries to release reservation space, parent will underflow its
1478 * reservation (for relationship adding case).
1480 * Caller should hold fs_info->qgroup_lock.
1482 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info, u64 ref_root,
1483 struct btrfs_qgroup *src, int sign)
1485 struct btrfs_qgroup *qgroup;
1486 struct btrfs_qgroup *cur;
1487 LIST_HEAD(qgroup_list);
1488 u64 num_bytes = src->excl;
1491 qgroup = find_qgroup_rb(fs_info, ref_root);
1495 qgroup_iterator_add(&qgroup_list, qgroup);
1496 list_for_each_entry(cur, &qgroup_list, iterator) {
1497 struct btrfs_qgroup_list *glist;
1499 qgroup->rfer += sign * num_bytes;
1500 qgroup->rfer_cmpr += sign * num_bytes;
1502 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1503 qgroup->excl += sign * num_bytes;
1504 qgroup->excl_cmpr += sign * num_bytes;
1507 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1509 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1510 qgroup_dirty(fs_info, qgroup);
1512 /* Append parent qgroups to @qgroup_list. */
1513 list_for_each_entry(glist, &qgroup->groups, next_group)
1514 qgroup_iterator_add(&qgroup_list, glist->group);
1518 qgroup_iterator_clean(&qgroup_list);
1524 * Quick path for updating qgroup with only excl refs.
1526 * In that case, just update all parent will be enough.
1527 * Or we needs to do a full rescan.
1528 * Caller should also hold fs_info->qgroup_lock.
1530 * Return 0 for quick update, return >0 for need to full rescan
1531 * and mark INCONSISTENT flag.
1532 * Return < 0 for other error.
1534 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1535 u64 src, u64 dst, int sign)
1537 struct btrfs_qgroup *qgroup;
1541 qgroup = find_qgroup_rb(fs_info, src);
1544 if (qgroup->excl == qgroup->rfer) {
1546 err = __qgroup_excl_accounting(fs_info, dst, qgroup, sign);
1554 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1558 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src, u64 dst)
1560 struct btrfs_fs_info *fs_info = trans->fs_info;
1561 struct btrfs_qgroup *parent;
1562 struct btrfs_qgroup *member;
1563 struct btrfs_qgroup_list *list;
1564 struct btrfs_qgroup_list *prealloc = NULL;
1567 /* Check the level of src and dst first */
1568 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1571 mutex_lock(&fs_info->qgroup_ioctl_lock);
1572 if (!fs_info->quota_root) {
1576 member = find_qgroup_rb(fs_info, src);
1577 parent = find_qgroup_rb(fs_info, dst);
1578 if (!member || !parent) {
1583 /* check if such qgroup relation exist firstly */
1584 list_for_each_entry(list, &member->groups, next_group) {
1585 if (list->group == parent) {
1591 prealloc = kzalloc(sizeof(*list), GFP_NOFS);
1596 ret = add_qgroup_relation_item(trans, src, dst);
1600 ret = add_qgroup_relation_item(trans, dst, src);
1602 del_qgroup_relation_item(trans, src, dst);
1606 spin_lock(&fs_info->qgroup_lock);
1607 ret = __add_relation_rb(prealloc, member, parent);
1610 spin_unlock(&fs_info->qgroup_lock);
1613 ret = quick_update_accounting(fs_info, src, dst, 1);
1614 spin_unlock(&fs_info->qgroup_lock);
1617 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1621 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1624 struct btrfs_fs_info *fs_info = trans->fs_info;
1625 struct btrfs_qgroup *parent;
1626 struct btrfs_qgroup *member;
1627 struct btrfs_qgroup_list *list;
1632 if (!fs_info->quota_root) {
1637 member = find_qgroup_rb(fs_info, src);
1638 parent = find_qgroup_rb(fs_info, dst);
1640 * The parent/member pair doesn't exist, then try to delete the dead
1641 * relation items only.
1643 if (!member || !parent)
1646 /* check if such qgroup relation exist firstly */
1647 list_for_each_entry(list, &member->groups, next_group) {
1648 if (list->group == parent) {
1655 ret = del_qgroup_relation_item(trans, src, dst);
1656 if (ret < 0 && ret != -ENOENT)
1658 ret2 = del_qgroup_relation_item(trans, dst, src);
1659 if (ret2 < 0 && ret2 != -ENOENT)
1662 /* At least one deletion succeeded, return 0 */
1667 spin_lock(&fs_info->qgroup_lock);
1668 del_relation_rb(fs_info, src, dst);
1669 ret = quick_update_accounting(fs_info, src, dst, -1);
1670 spin_unlock(&fs_info->qgroup_lock);
1676 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1679 struct btrfs_fs_info *fs_info = trans->fs_info;
1682 mutex_lock(&fs_info->qgroup_ioctl_lock);
1683 ret = __del_qgroup_relation(trans, src, dst);
1684 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1689 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1691 struct btrfs_fs_info *fs_info = trans->fs_info;
1692 struct btrfs_root *quota_root;
1693 struct btrfs_qgroup *qgroup;
1694 struct btrfs_qgroup *prealloc = NULL;
1697 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED)
1700 mutex_lock(&fs_info->qgroup_ioctl_lock);
1701 if (!fs_info->quota_root) {
1705 quota_root = fs_info->quota_root;
1706 qgroup = find_qgroup_rb(fs_info, qgroupid);
1712 prealloc = kzalloc(sizeof(*prealloc), GFP_NOFS);
1718 ret = add_qgroup_item(trans, quota_root, qgroupid);
1722 spin_lock(&fs_info->qgroup_lock);
1723 qgroup = add_qgroup_rb(fs_info, prealloc, qgroupid);
1724 spin_unlock(&fs_info->qgroup_lock);
1727 ret = btrfs_sysfs_add_one_qgroup(fs_info, qgroup);
1729 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1734 static bool qgroup_has_usage(struct btrfs_qgroup *qgroup)
1736 return (qgroup->rfer > 0 || qgroup->rfer_cmpr > 0 ||
1737 qgroup->excl > 0 || qgroup->excl_cmpr > 0 ||
1738 qgroup->rsv.values[BTRFS_QGROUP_RSV_DATA] > 0 ||
1739 qgroup->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC] > 0 ||
1740 qgroup->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > 0);
1743 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1745 struct btrfs_fs_info *fs_info = trans->fs_info;
1746 struct btrfs_qgroup *qgroup;
1747 struct btrfs_qgroup_list *list;
1750 mutex_lock(&fs_info->qgroup_ioctl_lock);
1751 if (!fs_info->quota_root) {
1756 qgroup = find_qgroup_rb(fs_info, qgroupid);
1762 if (is_fstree(qgroupid) && qgroup_has_usage(qgroup)) {
1767 /* Check if there are no children of this qgroup */
1768 if (!list_empty(&qgroup->members)) {
1773 ret = del_qgroup_item(trans, qgroupid);
1774 if (ret && ret != -ENOENT)
1777 while (!list_empty(&qgroup->groups)) {
1778 list = list_first_entry(&qgroup->groups,
1779 struct btrfs_qgroup_list, next_group);
1780 ret = __del_qgroup_relation(trans, qgroupid,
1781 list->group->qgroupid);
1786 spin_lock(&fs_info->qgroup_lock);
1787 del_qgroup_rb(fs_info, qgroupid);
1788 spin_unlock(&fs_info->qgroup_lock);
1791 * Remove the qgroup from sysfs now without holding the qgroup_lock
1792 * spinlock, since the sysfs_remove_group() function needs to take
1793 * the mutex kernfs_mutex through kernfs_remove_by_name_ns().
1795 btrfs_sysfs_del_one_qgroup(fs_info, qgroup);
1798 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1802 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1803 struct btrfs_qgroup_limit *limit)
1805 struct btrfs_fs_info *fs_info = trans->fs_info;
1806 struct btrfs_qgroup *qgroup;
1808 /* Sometimes we would want to clear the limit on this qgroup.
1809 * To meet this requirement, we treat the -1 as a special value
1810 * which tell kernel to clear the limit on this qgroup.
1812 const u64 CLEAR_VALUE = -1;
1814 mutex_lock(&fs_info->qgroup_ioctl_lock);
1815 if (!fs_info->quota_root) {
1820 qgroup = find_qgroup_rb(fs_info, qgroupid);
1826 spin_lock(&fs_info->qgroup_lock);
1827 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1828 if (limit->max_rfer == CLEAR_VALUE) {
1829 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1830 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1831 qgroup->max_rfer = 0;
1833 qgroup->max_rfer = limit->max_rfer;
1836 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1837 if (limit->max_excl == CLEAR_VALUE) {
1838 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1839 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1840 qgroup->max_excl = 0;
1842 qgroup->max_excl = limit->max_excl;
1845 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1846 if (limit->rsv_rfer == CLEAR_VALUE) {
1847 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1848 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1849 qgroup->rsv_rfer = 0;
1851 qgroup->rsv_rfer = limit->rsv_rfer;
1854 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1855 if (limit->rsv_excl == CLEAR_VALUE) {
1856 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1857 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1858 qgroup->rsv_excl = 0;
1860 qgroup->rsv_excl = limit->rsv_excl;
1863 qgroup->lim_flags |= limit->flags;
1865 spin_unlock(&fs_info->qgroup_lock);
1867 ret = update_qgroup_limit_item(trans, qgroup);
1869 qgroup_mark_inconsistent(fs_info);
1870 btrfs_info(fs_info, "unable to update quota limit for %llu",
1875 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1880 * Inform qgroup to trace one dirty extent, its info is recorded in @record.
1881 * So qgroup can account it at transaction committing time.
1883 * No lock version, caller must acquire delayed ref lock and allocated memory,
1884 * then call btrfs_qgroup_trace_extent_post() after exiting lock context.
1886 * Return 0 for success insert
1887 * Return >0 for existing record, caller can free @record safely.
1888 * Error is not possible
1890 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1891 struct btrfs_delayed_ref_root *delayed_refs,
1892 struct btrfs_qgroup_extent_record *record)
1894 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1895 struct rb_node *parent_node = NULL;
1896 struct btrfs_qgroup_extent_record *entry;
1897 u64 bytenr = record->bytenr;
1899 if (!btrfs_qgroup_full_accounting(fs_info))
1902 lockdep_assert_held(&delayed_refs->lock);
1903 trace_btrfs_qgroup_trace_extent(fs_info, record);
1907 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1909 if (bytenr < entry->bytenr) {
1911 } else if (bytenr > entry->bytenr) {
1912 p = &(*p)->rb_right;
1914 if (record->data_rsv && !entry->data_rsv) {
1915 entry->data_rsv = record->data_rsv;
1916 entry->data_rsv_refroot =
1917 record->data_rsv_refroot;
1923 rb_link_node(&record->node, parent_node, p);
1924 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1929 * Post handler after qgroup_trace_extent_nolock().
1931 * NOTE: Current qgroup does the expensive backref walk at transaction
1932 * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
1934 * This is designed to allow btrfs_find_all_roots() to get correct new_roots
1937 * However for old_roots there is no need to do backref walk at that time,
1938 * since we search commit roots to walk backref and result will always be
1941 * Due to the nature of no lock version, we can't do backref there.
1942 * So we must call btrfs_qgroup_trace_extent_post() after exiting
1945 * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result
1946 * using current root, then we can move all expensive backref walk out of
1947 * transaction committing, but not now as qgroup accounting will be wrong again.
1949 int btrfs_qgroup_trace_extent_post(struct btrfs_trans_handle *trans,
1950 struct btrfs_qgroup_extent_record *qrecord)
1952 struct btrfs_backref_walk_ctx ctx = { 0 };
1955 if (!btrfs_qgroup_full_accounting(trans->fs_info))
1958 * We are always called in a context where we are already holding a
1959 * transaction handle. Often we are called when adding a data delayed
1960 * reference from btrfs_truncate_inode_items() (truncating or unlinking),
1961 * in which case we will be holding a write lock on extent buffer from a
1962 * subvolume tree. In this case we can't allow btrfs_find_all_roots() to
1963 * acquire fs_info->commit_root_sem, because that is a higher level lock
1964 * that must be acquired before locking any extent buffers.
1966 * So we want btrfs_find_all_roots() to not acquire the commit_root_sem
1967 * but we can't pass it a non-NULL transaction handle, because otherwise
1968 * it would not use commit roots and would lock extent buffers, causing
1969 * a deadlock if it ends up trying to read lock the same extent buffer
1970 * that was previously write locked at btrfs_truncate_inode_items().
1972 * So pass a NULL transaction handle to btrfs_find_all_roots() and
1973 * explicitly tell it to not acquire the commit_root_sem - if we are
1974 * holding a transaction handle we don't need its protection.
1976 ASSERT(trans != NULL);
1978 if (trans->fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
1981 ctx.bytenr = qrecord->bytenr;
1982 ctx.fs_info = trans->fs_info;
1984 ret = btrfs_find_all_roots(&ctx, true);
1986 qgroup_mark_inconsistent(trans->fs_info);
1987 btrfs_warn(trans->fs_info,
1988 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1994 * Here we don't need to get the lock of
1995 * trans->transaction->delayed_refs, since inserted qrecord won't
1996 * be deleted, only qrecord->node may be modified (new qrecord insert)
1998 * So modifying qrecord->old_roots is safe here
2000 qrecord->old_roots = ctx.roots;
2005 * Inform qgroup to trace one dirty extent, specified by @bytenr and
2007 * So qgroup can account it at commit trans time.
2009 * Better encapsulated version, with memory allocation and backref walk for
2011 * So this can sleep.
2013 * Return 0 if the operation is done.
2014 * Return <0 for error, like memory allocation failure or invalid parameter
2017 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2020 struct btrfs_fs_info *fs_info = trans->fs_info;
2021 struct btrfs_qgroup_extent_record *record;
2022 struct btrfs_delayed_ref_root *delayed_refs;
2025 if (!btrfs_qgroup_full_accounting(fs_info) || bytenr == 0 || num_bytes == 0)
2027 record = kzalloc(sizeof(*record), GFP_NOFS);
2031 delayed_refs = &trans->transaction->delayed_refs;
2032 record->bytenr = bytenr;
2033 record->num_bytes = num_bytes;
2034 record->old_roots = NULL;
2036 spin_lock(&delayed_refs->lock);
2037 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
2038 spin_unlock(&delayed_refs->lock);
2043 return btrfs_qgroup_trace_extent_post(trans, record);
2047 * Inform qgroup to trace all leaf items of data
2049 * Return 0 for success
2050 * Return <0 for error(ENOMEM)
2052 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
2053 struct extent_buffer *eb)
2055 struct btrfs_fs_info *fs_info = trans->fs_info;
2056 int nr = btrfs_header_nritems(eb);
2057 int i, extent_type, ret;
2058 struct btrfs_key key;
2059 struct btrfs_file_extent_item *fi;
2060 u64 bytenr, num_bytes;
2062 /* We can be called directly from walk_up_proc() */
2063 if (!btrfs_qgroup_full_accounting(fs_info))
2066 for (i = 0; i < nr; i++) {
2067 btrfs_item_key_to_cpu(eb, &key, i);
2069 if (key.type != BTRFS_EXTENT_DATA_KEY)
2072 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
2073 /* filter out non qgroup-accountable extents */
2074 extent_type = btrfs_file_extent_type(eb, fi);
2076 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
2079 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
2083 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
2085 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes);
2094 * Walk up the tree from the bottom, freeing leaves and any interior
2095 * nodes which have had all slots visited. If a node (leaf or
2096 * interior) is freed, the node above it will have it's slot
2097 * incremented. The root node will never be freed.
2099 * At the end of this function, we should have a path which has all
2100 * slots incremented to the next position for a search. If we need to
2101 * read a new node it will be NULL and the node above it will have the
2102 * correct slot selected for a later read.
2104 * If we increment the root nodes slot counter past the number of
2105 * elements, 1 is returned to signal completion of the search.
2107 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
2111 struct extent_buffer *eb;
2113 if (root_level == 0)
2116 while (level <= root_level) {
2117 eb = path->nodes[level];
2118 nr = btrfs_header_nritems(eb);
2119 path->slots[level]++;
2120 slot = path->slots[level];
2121 if (slot >= nr || level == 0) {
2123 * Don't free the root - we will detect this
2124 * condition after our loop and return a
2125 * positive value for caller to stop walking the tree.
2127 if (level != root_level) {
2128 btrfs_tree_unlock_rw(eb, path->locks[level]);
2129 path->locks[level] = 0;
2131 free_extent_buffer(eb);
2132 path->nodes[level] = NULL;
2133 path->slots[level] = 0;
2137 * We have a valid slot to walk back down
2138 * from. Stop here so caller can process these
2147 eb = path->nodes[root_level];
2148 if (path->slots[root_level] >= btrfs_header_nritems(eb))
2155 * Helper function to trace a subtree tree block swap.
2157 * The swap will happen in highest tree block, but there may be a lot of
2158 * tree blocks involved.
2161 * OO = Old tree blocks
2162 * NN = New tree blocks allocated during balance
2164 * File tree (257) Reloc tree for 257
2167 * L1 OO OO (a) OO NN (a)
2169 * L0 OO OO OO OO OO OO NN NN
2172 * When calling qgroup_trace_extent_swap(), we will pass:
2174 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
2178 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
2179 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
2181 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
2183 * 1) Tree search from @src_eb
2184 * It should acts as a simplified btrfs_search_slot().
2185 * The key for search can be extracted from @dst_path->nodes[dst_level]
2188 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
2189 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
2190 * They should be marked during previous (@dst_level = 1) iteration.
2192 * 3) Mark file extents in leaves dirty
2193 * We don't have good way to pick out new file extents only.
2194 * So we still follow the old method by scanning all file extents in
2197 * This function can free us from keeping two paths, thus later we only need
2198 * to care about how to iterate all new tree blocks in reloc tree.
2200 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
2201 struct extent_buffer *src_eb,
2202 struct btrfs_path *dst_path,
2203 int dst_level, int root_level,
2206 struct btrfs_key key;
2207 struct btrfs_path *src_path;
2208 struct btrfs_fs_info *fs_info = trans->fs_info;
2209 u32 nodesize = fs_info->nodesize;
2210 int cur_level = root_level;
2213 BUG_ON(dst_level > root_level);
2214 /* Level mismatch */
2215 if (btrfs_header_level(src_eb) != root_level)
2218 src_path = btrfs_alloc_path();
2225 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2227 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
2230 atomic_inc(&src_eb->refs);
2231 src_path->nodes[root_level] = src_eb;
2232 src_path->slots[root_level] = dst_path->slots[root_level];
2233 src_path->locks[root_level] = 0;
2235 /* A simplified version of btrfs_search_slot() */
2236 while (cur_level >= dst_level) {
2237 struct btrfs_key src_key;
2238 struct btrfs_key dst_key;
2240 if (src_path->nodes[cur_level] == NULL) {
2241 struct extent_buffer *eb;
2244 eb = src_path->nodes[cur_level + 1];
2245 parent_slot = src_path->slots[cur_level + 1];
2247 eb = btrfs_read_node_slot(eb, parent_slot);
2253 src_path->nodes[cur_level] = eb;
2255 btrfs_tree_read_lock(eb);
2256 src_path->locks[cur_level] = BTRFS_READ_LOCK;
2259 src_path->slots[cur_level] = dst_path->slots[cur_level];
2261 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
2262 &dst_key, dst_path->slots[cur_level]);
2263 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
2264 &src_key, src_path->slots[cur_level]);
2266 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
2267 &dst_key, dst_path->slots[cur_level]);
2268 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
2269 &src_key, src_path->slots[cur_level]);
2271 /* Content mismatch, something went wrong */
2272 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
2280 * Now both @dst_path and @src_path have been populated, record the tree
2281 * blocks for qgroup accounting.
2283 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
2287 ret = btrfs_qgroup_trace_extent(trans, dst_path->nodes[dst_level]->start,
2292 /* Record leaf file extents */
2293 if (dst_level == 0 && trace_leaf) {
2294 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
2297 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
2300 btrfs_free_path(src_path);
2305 * Helper function to do recursive generation-aware depth-first search, to
2306 * locate all new tree blocks in a subtree of reloc tree.
2308 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
2317 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
2321 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
2322 * above tree blocks along with their counter parts in file tree.
2323 * While during search, old tree blocks OO(c) will be skipped as tree block swap
2324 * won't affect OO(c).
2326 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
2327 struct extent_buffer *src_eb,
2328 struct btrfs_path *dst_path,
2329 int cur_level, int root_level,
2330 u64 last_snapshot, bool trace_leaf)
2332 struct btrfs_fs_info *fs_info = trans->fs_info;
2333 struct extent_buffer *eb;
2334 bool need_cleanup = false;
2338 /* Level sanity check */
2339 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
2340 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
2341 root_level < cur_level) {
2342 btrfs_err_rl(fs_info,
2343 "%s: bad levels, cur_level=%d root_level=%d",
2344 __func__, cur_level, root_level);
2348 /* Read the tree block if needed */
2349 if (dst_path->nodes[cur_level] == NULL) {
2354 * dst_path->nodes[root_level] must be initialized before
2355 * calling this function.
2357 if (cur_level == root_level) {
2358 btrfs_err_rl(fs_info,
2359 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
2360 __func__, root_level, root_level, cur_level);
2365 * We need to get child blockptr/gen from parent before we can
2368 eb = dst_path->nodes[cur_level + 1];
2369 parent_slot = dst_path->slots[cur_level + 1];
2370 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2372 /* This node is old, no need to trace */
2373 if (child_gen < last_snapshot)
2376 eb = btrfs_read_node_slot(eb, parent_slot);
2382 dst_path->nodes[cur_level] = eb;
2383 dst_path->slots[cur_level] = 0;
2385 btrfs_tree_read_lock(eb);
2386 dst_path->locks[cur_level] = BTRFS_READ_LOCK;
2387 need_cleanup = true;
2390 /* Now record this tree block and its counter part for qgroups */
2391 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
2392 root_level, trace_leaf);
2396 eb = dst_path->nodes[cur_level];
2398 if (cur_level > 0) {
2399 /* Iterate all child tree blocks */
2400 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2401 /* Skip old tree blocks as they won't be swapped */
2402 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2404 dst_path->slots[cur_level] = i;
2406 /* Recursive call (at most 7 times) */
2407 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2408 dst_path, cur_level - 1, root_level,
2409 last_snapshot, trace_leaf);
2418 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2419 dst_path->locks[cur_level]);
2420 free_extent_buffer(dst_path->nodes[cur_level]);
2421 dst_path->nodes[cur_level] = NULL;
2422 dst_path->slots[cur_level] = 0;
2423 dst_path->locks[cur_level] = 0;
2429 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2430 struct extent_buffer *src_eb,
2431 struct extent_buffer *dst_eb,
2432 u64 last_snapshot, bool trace_leaf)
2434 struct btrfs_fs_info *fs_info = trans->fs_info;
2435 struct btrfs_path *dst_path = NULL;
2439 if (!btrfs_qgroup_full_accounting(fs_info))
2442 /* Wrong parameter order */
2443 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2444 btrfs_err_rl(fs_info,
2445 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2446 btrfs_header_generation(src_eb),
2447 btrfs_header_generation(dst_eb));
2451 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2456 level = btrfs_header_level(dst_eb);
2457 dst_path = btrfs_alloc_path();
2463 atomic_inc(&dst_eb->refs);
2464 dst_path->nodes[level] = dst_eb;
2465 dst_path->slots[level] = 0;
2466 dst_path->locks[level] = 0;
2468 /* Do the generation aware breadth-first search */
2469 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2470 level, last_snapshot, trace_leaf);
2476 btrfs_free_path(dst_path);
2478 qgroup_mark_inconsistent(fs_info);
2483 * Inform qgroup to trace a whole subtree, including all its child tree
2485 * The root tree block is specified by @root_eb.
2487 * Normally used by relocation(tree block swap) and subvolume deletion.
2489 * Return 0 for success
2490 * Return <0 for error(ENOMEM or tree search error)
2492 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2493 struct extent_buffer *root_eb,
2494 u64 root_gen, int root_level)
2496 struct btrfs_fs_info *fs_info = trans->fs_info;
2499 u8 drop_subptree_thres;
2500 struct extent_buffer *eb = root_eb;
2501 struct btrfs_path *path = NULL;
2503 ASSERT(0 <= root_level && root_level < BTRFS_MAX_LEVEL);
2504 ASSERT(root_eb != NULL);
2506 if (!btrfs_qgroup_full_accounting(fs_info))
2509 spin_lock(&fs_info->qgroup_lock);
2510 drop_subptree_thres = fs_info->qgroup_drop_subtree_thres;
2511 spin_unlock(&fs_info->qgroup_lock);
2514 * This function only gets called for snapshot drop, if we hit a high
2515 * node here, it means we are going to change ownership for quite a lot
2516 * of extents, which will greatly slow down btrfs_commit_transaction().
2518 * So here if we find a high tree here, we just skip the accounting and
2519 * mark qgroup inconsistent.
2521 if (root_level >= drop_subptree_thres) {
2522 qgroup_mark_inconsistent(fs_info);
2526 if (!extent_buffer_uptodate(root_eb)) {
2527 struct btrfs_tree_parent_check check = {
2528 .has_first_key = false,
2529 .transid = root_gen,
2533 ret = btrfs_read_extent_buffer(root_eb, &check);
2538 if (root_level == 0) {
2539 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2543 path = btrfs_alloc_path();
2548 * Walk down the tree. Missing extent blocks are filled in as
2549 * we go. Metadata is accounted every time we read a new
2552 * When we reach a leaf, we account for file extent items in it,
2553 * walk back up the tree (adjusting slot pointers as we go)
2554 * and restart the search process.
2556 atomic_inc(&root_eb->refs); /* For path */
2557 path->nodes[root_level] = root_eb;
2558 path->slots[root_level] = 0;
2559 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2562 while (level >= 0) {
2563 if (path->nodes[level] == NULL) {
2568 * We need to get child blockptr from parent before we
2571 eb = path->nodes[level + 1];
2572 parent_slot = path->slots[level + 1];
2573 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2575 eb = btrfs_read_node_slot(eb, parent_slot);
2581 path->nodes[level] = eb;
2582 path->slots[level] = 0;
2584 btrfs_tree_read_lock(eb);
2585 path->locks[level] = BTRFS_READ_LOCK;
2587 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2594 ret = btrfs_qgroup_trace_leaf_items(trans,
2595 path->nodes[level]);
2599 /* Nonzero return here means we completed our search */
2600 ret = adjust_slots_upwards(path, root_level);
2604 /* Restart search with new slots */
2613 btrfs_free_path(path);
2618 static void qgroup_iterator_nested_add(struct list_head *head, struct btrfs_qgroup *qgroup)
2620 if (!list_empty(&qgroup->nested_iterator))
2623 list_add_tail(&qgroup->nested_iterator, head);
2626 static void qgroup_iterator_nested_clean(struct list_head *head)
2628 while (!list_empty(head)) {
2629 struct btrfs_qgroup *qgroup;
2631 qgroup = list_first_entry(head, struct btrfs_qgroup, nested_iterator);
2632 list_del_init(&qgroup->nested_iterator);
2636 #define UPDATE_NEW 0
2637 #define UPDATE_OLD 1
2639 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2641 static void qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2642 struct ulist *roots, struct list_head *qgroups,
2643 u64 seq, int update_old)
2645 struct ulist_node *unode;
2646 struct ulist_iterator uiter;
2647 struct btrfs_qgroup *qg;
2651 ULIST_ITER_INIT(&uiter);
2652 while ((unode = ulist_next(roots, &uiter))) {
2655 qg = find_qgroup_rb(fs_info, unode->val);
2659 qgroup_iterator_nested_add(qgroups, qg);
2660 qgroup_iterator_add(&tmp, qg);
2661 list_for_each_entry(qg, &tmp, iterator) {
2662 struct btrfs_qgroup_list *glist;
2665 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2667 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2669 list_for_each_entry(glist, &qg->groups, next_group) {
2670 qgroup_iterator_nested_add(qgroups, glist->group);
2671 qgroup_iterator_add(&tmp, glist->group);
2674 qgroup_iterator_clean(&tmp);
2679 * Update qgroup rfer/excl counters.
2680 * Rfer update is easy, codes can explain themselves.
2682 * Excl update is tricky, the update is split into 2 parts.
2683 * Part 1: Possible exclusive <-> sharing detect:
2685 * -------------------------------------
2687 * -------------------------------------
2689 * -------------------------------------
2692 * A: cur_old_roots < nr_old_roots (not exclusive before)
2693 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2694 * B: cur_new_roots < nr_new_roots (not exclusive now)
2695 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2698 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2699 * *: Definitely not changed. **: Possible unchanged.
2701 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2703 * To make the logic clear, we first use condition A and B to split
2704 * combination into 4 results.
2706 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2707 * only on variant maybe 0.
2709 * Lastly, check result **, since there are 2 variants maybe 0, split them
2711 * But this time we don't need to consider other things, the codes and logic
2712 * is easy to understand now.
2714 static void qgroup_update_counters(struct btrfs_fs_info *fs_info,
2715 struct list_head *qgroups, u64 nr_old_roots,
2716 u64 nr_new_roots, u64 num_bytes, u64 seq)
2718 struct btrfs_qgroup *qg;
2720 list_for_each_entry(qg, qgroups, nested_iterator) {
2721 u64 cur_new_count, cur_old_count;
2724 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2725 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2727 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2730 /* Rfer update part */
2731 if (cur_old_count == 0 && cur_new_count > 0) {
2732 qg->rfer += num_bytes;
2733 qg->rfer_cmpr += num_bytes;
2736 if (cur_old_count > 0 && cur_new_count == 0) {
2737 qg->rfer -= num_bytes;
2738 qg->rfer_cmpr -= num_bytes;
2742 /* Excl update part */
2743 /* Exclusive/none -> shared case */
2744 if (cur_old_count == nr_old_roots &&
2745 cur_new_count < nr_new_roots) {
2746 /* Exclusive -> shared */
2747 if (cur_old_count != 0) {
2748 qg->excl -= num_bytes;
2749 qg->excl_cmpr -= num_bytes;
2754 /* Shared -> exclusive/none case */
2755 if (cur_old_count < nr_old_roots &&
2756 cur_new_count == nr_new_roots) {
2757 /* Shared->exclusive */
2758 if (cur_new_count != 0) {
2759 qg->excl += num_bytes;
2760 qg->excl_cmpr += num_bytes;
2765 /* Exclusive/none -> exclusive/none case */
2766 if (cur_old_count == nr_old_roots &&
2767 cur_new_count == nr_new_roots) {
2768 if (cur_old_count == 0) {
2769 /* None -> exclusive/none */
2771 if (cur_new_count != 0) {
2772 /* None -> exclusive */
2773 qg->excl += num_bytes;
2774 qg->excl_cmpr += num_bytes;
2777 /* None -> none, nothing changed */
2779 /* Exclusive -> exclusive/none */
2781 if (cur_new_count == 0) {
2782 /* Exclusive -> none */
2783 qg->excl -= num_bytes;
2784 qg->excl_cmpr -= num_bytes;
2787 /* Exclusive -> exclusive, nothing changed */
2792 qgroup_dirty(fs_info, qg);
2797 * Check if the @roots potentially is a list of fs tree roots
2799 * Return 0 for definitely not a fs/subvol tree roots ulist
2800 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2803 static int maybe_fs_roots(struct ulist *roots)
2805 struct ulist_node *unode;
2806 struct ulist_iterator uiter;
2808 /* Empty one, still possible for fs roots */
2809 if (!roots || roots->nnodes == 0)
2812 ULIST_ITER_INIT(&uiter);
2813 unode = ulist_next(roots, &uiter);
2818 * If it contains fs tree roots, then it must belong to fs/subvol
2820 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2822 return is_fstree(unode->val);
2825 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2826 u64 num_bytes, struct ulist *old_roots,
2827 struct ulist *new_roots)
2829 struct btrfs_fs_info *fs_info = trans->fs_info;
2832 u64 nr_new_roots = 0;
2833 u64 nr_old_roots = 0;
2837 * If quotas get disabled meanwhile, the resources need to be freed and
2838 * we can't just exit here.
2840 if (!btrfs_qgroup_full_accounting(fs_info) ||
2841 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)
2845 if (!maybe_fs_roots(new_roots))
2847 nr_new_roots = new_roots->nnodes;
2850 if (!maybe_fs_roots(old_roots))
2852 nr_old_roots = old_roots->nnodes;
2855 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2856 if (nr_old_roots == 0 && nr_new_roots == 0)
2859 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2860 num_bytes, nr_old_roots, nr_new_roots);
2862 mutex_lock(&fs_info->qgroup_rescan_lock);
2863 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2864 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2865 mutex_unlock(&fs_info->qgroup_rescan_lock);
2870 mutex_unlock(&fs_info->qgroup_rescan_lock);
2872 spin_lock(&fs_info->qgroup_lock);
2873 seq = fs_info->qgroup_seq;
2875 /* Update old refcnts using old_roots */
2876 qgroup_update_refcnt(fs_info, old_roots, &qgroups, seq, UPDATE_OLD);
2878 /* Update new refcnts using new_roots */
2879 qgroup_update_refcnt(fs_info, new_roots, &qgroups, seq, UPDATE_NEW);
2881 qgroup_update_counters(fs_info, &qgroups, nr_old_roots, nr_new_roots,
2885 * We're done using the iterator, release all its qgroups while holding
2886 * fs_info->qgroup_lock so that we don't race with btrfs_remove_qgroup()
2887 * and trigger use-after-free accesses to qgroups.
2889 qgroup_iterator_nested_clean(&qgroups);
2892 * Bump qgroup_seq to avoid seq overlap
2894 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2895 spin_unlock(&fs_info->qgroup_lock);
2897 ulist_free(old_roots);
2898 ulist_free(new_roots);
2902 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2904 struct btrfs_fs_info *fs_info = trans->fs_info;
2905 struct btrfs_qgroup_extent_record *record;
2906 struct btrfs_delayed_ref_root *delayed_refs;
2907 struct ulist *new_roots = NULL;
2908 struct rb_node *node;
2909 u64 num_dirty_extents = 0;
2913 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
2916 delayed_refs = &trans->transaction->delayed_refs;
2917 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2918 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2919 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2922 num_dirty_extents++;
2923 trace_btrfs_qgroup_account_extents(fs_info, record);
2925 if (!ret && !(fs_info->qgroup_flags &
2926 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING)) {
2927 struct btrfs_backref_walk_ctx ctx = { 0 };
2929 ctx.bytenr = record->bytenr;
2930 ctx.fs_info = fs_info;
2933 * Old roots should be searched when inserting qgroup
2936 * But for INCONSISTENT (NO_ACCOUNTING) -> rescan case,
2937 * we may have some record inserted during
2938 * NO_ACCOUNTING (thus no old_roots populated), but
2939 * later we start rescan, which clears NO_ACCOUNTING,
2940 * leaving some inserted records without old_roots
2943 * Those cases are rare and should not cause too much
2944 * time spent during commit_transaction().
2946 if (!record->old_roots) {
2947 /* Search commit root to find old_roots */
2948 ret = btrfs_find_all_roots(&ctx, false);
2951 record->old_roots = ctx.roots;
2956 * Use BTRFS_SEQ_LAST as time_seq to do special search,
2957 * which doesn't lock tree or delayed_refs and search
2958 * current root. It's safe inside commit_transaction().
2961 ctx.time_seq = BTRFS_SEQ_LAST;
2962 ret = btrfs_find_all_roots(&ctx, false);
2965 new_roots = ctx.roots;
2966 if (qgroup_to_skip) {
2967 ulist_del(new_roots, qgroup_to_skip, 0);
2968 ulist_del(record->old_roots, qgroup_to_skip,
2971 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2975 record->old_roots = NULL;
2978 /* Free the reserved data space */
2979 btrfs_qgroup_free_refroot(fs_info,
2980 record->data_rsv_refroot,
2982 BTRFS_QGROUP_RSV_DATA);
2984 ulist_free(record->old_roots);
2985 ulist_free(new_roots);
2987 rb_erase(node, &delayed_refs->dirty_extent_root);
2991 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2997 * Writes all changed qgroups to disk.
2998 * Called by the transaction commit path and the qgroup assign ioctl.
3000 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
3002 struct btrfs_fs_info *fs_info = trans->fs_info;
3006 * In case we are called from the qgroup assign ioctl, assert that we
3007 * are holding the qgroup_ioctl_lock, otherwise we can race with a quota
3008 * disable operation (ioctl) and access a freed quota root.
3010 if (trans->transaction->state != TRANS_STATE_COMMIT_DOING)
3011 lockdep_assert_held(&fs_info->qgroup_ioctl_lock);
3013 if (!fs_info->quota_root)
3016 spin_lock(&fs_info->qgroup_lock);
3017 while (!list_empty(&fs_info->dirty_qgroups)) {
3018 struct btrfs_qgroup *qgroup;
3019 qgroup = list_first_entry(&fs_info->dirty_qgroups,
3020 struct btrfs_qgroup, dirty);
3021 list_del_init(&qgroup->dirty);
3022 spin_unlock(&fs_info->qgroup_lock);
3023 ret = update_qgroup_info_item(trans, qgroup);
3025 qgroup_mark_inconsistent(fs_info);
3026 ret = update_qgroup_limit_item(trans, qgroup);
3028 qgroup_mark_inconsistent(fs_info);
3029 spin_lock(&fs_info->qgroup_lock);
3031 if (btrfs_qgroup_enabled(fs_info))
3032 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
3034 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
3035 spin_unlock(&fs_info->qgroup_lock);
3037 ret = update_qgroup_status_item(trans);
3039 qgroup_mark_inconsistent(fs_info);
3044 int btrfs_qgroup_check_inherit(struct btrfs_fs_info *fs_info,
3045 struct btrfs_qgroup_inherit *inherit,
3048 if (inherit->flags & ~BTRFS_QGROUP_INHERIT_FLAGS_SUPP)
3050 if (size < sizeof(*inherit) || size > PAGE_SIZE)
3054 * In the past we allowed btrfs_qgroup_inherit to specify to copy
3055 * rfer/excl numbers directly from other qgroups. This behavior has
3056 * been disabled in userspace for a very long time, but here we should
3057 * also disable it in kernel, as this behavior is known to mark qgroup
3058 * inconsistent, and a rescan would wipe out the changes anyway.
3060 * Reject any btrfs_qgroup_inherit with num_ref_copies or num_excl_copies.
3062 if (inherit->num_ref_copies > 0 || inherit->num_excl_copies > 0)
3065 if (inherit->num_qgroups > PAGE_SIZE)
3068 if (size != struct_size(inherit, qgroups, inherit->num_qgroups))
3072 * Now check all the remaining qgroups, they should all:
3075 * - Be higher level qgroups.
3077 for (int i = 0; i < inherit->num_qgroups; i++) {
3078 struct btrfs_qgroup *qgroup;
3079 u64 qgroupid = inherit->qgroups[i];
3081 if (btrfs_qgroup_level(qgroupid) == 0)
3084 spin_lock(&fs_info->qgroup_lock);
3085 qgroup = find_qgroup_rb(fs_info, qgroupid);
3087 spin_unlock(&fs_info->qgroup_lock);
3090 spin_unlock(&fs_info->qgroup_lock);
3095 static int qgroup_auto_inherit(struct btrfs_fs_info *fs_info,
3097 struct btrfs_qgroup_inherit **inherit)
3100 u64 num_qgroups = 0;
3101 struct btrfs_qgroup *inode_qg;
3102 struct btrfs_qgroup_list *qg_list;
3103 struct btrfs_qgroup_inherit *res;
3110 inode_qg = find_qgroup_rb(fs_info, inode_rootid);
3114 num_qgroups = list_count_nodes(&inode_qg->groups);
3119 struct_sz = struct_size(res, qgroups, num_qgroups);
3120 if (struct_sz == SIZE_MAX)
3123 res = kzalloc(struct_sz, GFP_NOFS);
3126 res->num_qgroups = num_qgroups;
3127 qgids = res->qgroups;
3129 list_for_each_entry(qg_list, &inode_qg->groups, next_group)
3130 qgids[i] = qg_list->group->qgroupid;
3137 * Check if we can skip rescan when inheriting qgroups. If @src has a single
3138 * @parent, and that @parent is owning all its bytes exclusively, we can skip
3139 * the full rescan, by just adding nodesize to the @parent's excl/rfer.
3141 * Return <0 for fatal errors (like srcid/parentid has no qgroup).
3142 * Return 0 if a quick inherit is done.
3143 * Return >0 if a quick inherit is not possible, and a full rescan is needed.
3145 static int qgroup_snapshot_quick_inherit(struct btrfs_fs_info *fs_info,
3146 u64 srcid, u64 parentid)
3148 struct btrfs_qgroup *src;
3149 struct btrfs_qgroup *parent;
3150 struct btrfs_qgroup_list *list;
3153 src = find_qgroup_rb(fs_info, srcid);
3156 parent = find_qgroup_rb(fs_info, parentid);
3161 * Source has no parent qgroup, but our new qgroup would have one.
3162 * Qgroup numbers would become inconsistent.
3164 if (list_empty(&src->groups))
3167 list_for_each_entry(list, &src->groups, next_group) {
3168 /* The parent is not the same, quick update is not possible. */
3169 if (list->group->qgroupid != parentid)
3173 * More than one parent qgroup, we can't be sure about accounting
3181 * The parent is not exclusively owning all its bytes. We're not sure
3182 * if the source has any bytes not fully owned by the parent.
3184 if (parent->excl != parent->rfer)
3187 parent->excl += fs_info->nodesize;
3188 parent->rfer += fs_info->nodesize;
3193 * Copy the accounting information between qgroups. This is necessary
3194 * when a snapshot or a subvolume is created. Throwing an error will
3195 * cause a transaction abort so we take extra care here to only error
3196 * when a readonly fs is a reasonable outcome.
3198 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
3199 u64 objectid, u64 inode_rootid,
3200 struct btrfs_qgroup_inherit *inherit)
3205 bool committing = false;
3206 struct btrfs_fs_info *fs_info = trans->fs_info;
3207 struct btrfs_root *quota_root;
3208 struct btrfs_qgroup *srcgroup;
3209 struct btrfs_qgroup *dstgroup;
3210 struct btrfs_qgroup *prealloc;
3211 struct btrfs_qgroup_list **qlist_prealloc = NULL;
3212 bool free_inherit = false;
3213 bool need_rescan = false;
3217 prealloc = kzalloc(sizeof(*prealloc), GFP_NOFS);
3222 * There are only two callers of this function.
3224 * One in create_subvol() in the ioctl context, which needs to hold
3225 * the qgroup_ioctl_lock.
3227 * The other one in create_pending_snapshot() where no other qgroup
3228 * code can modify the fs as they all need to either start a new trans
3229 * or hold a trans handler, thus we don't need to hold
3230 * qgroup_ioctl_lock.
3231 * This would avoid long and complex lock chain and make lockdep happy.
3233 spin_lock(&fs_info->trans_lock);
3234 if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
3236 spin_unlock(&fs_info->trans_lock);
3239 mutex_lock(&fs_info->qgroup_ioctl_lock);
3240 if (!btrfs_qgroup_enabled(fs_info))
3243 quota_root = fs_info->quota_root;
3249 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE && !inherit) {
3250 ret = qgroup_auto_inherit(fs_info, inode_rootid, &inherit);
3253 free_inherit = true;
3257 i_qgroups = (u64 *)(inherit + 1);
3258 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
3259 2 * inherit->num_excl_copies;
3260 for (i = 0; i < nums; ++i) {
3261 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
3264 * Zero out invalid groups so we can ignore
3268 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
3276 * create a tracking group for the subvol itself
3278 ret = add_qgroup_item(trans, quota_root, objectid);
3283 * add qgroup to all inherited groups
3286 i_qgroups = (u64 *)(inherit + 1);
3287 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
3288 if (*i_qgroups == 0)
3290 ret = add_qgroup_relation_item(trans, objectid,
3292 if (ret && ret != -EEXIST)
3294 ret = add_qgroup_relation_item(trans, *i_qgroups,
3296 if (ret && ret != -EEXIST)
3301 qlist_prealloc = kcalloc(inherit->num_qgroups,
3302 sizeof(struct btrfs_qgroup_list *),
3304 if (!qlist_prealloc) {
3308 for (int i = 0; i < inherit->num_qgroups; i++) {
3309 qlist_prealloc[i] = kzalloc(sizeof(struct btrfs_qgroup_list),
3311 if (!qlist_prealloc[i]) {
3318 spin_lock(&fs_info->qgroup_lock);
3320 dstgroup = add_qgroup_rb(fs_info, prealloc, objectid);
3323 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
3324 dstgroup->lim_flags = inherit->lim.flags;
3325 dstgroup->max_rfer = inherit->lim.max_rfer;
3326 dstgroup->max_excl = inherit->lim.max_excl;
3327 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
3328 dstgroup->rsv_excl = inherit->lim.rsv_excl;
3330 qgroup_dirty(fs_info, dstgroup);
3333 if (srcid && btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_FULL) {
3334 srcgroup = find_qgroup_rb(fs_info, srcid);
3339 * We call inherit after we clone the root in order to make sure
3340 * our counts don't go crazy, so at this point the only
3341 * difference between the two roots should be the root node.
3343 level_size = fs_info->nodesize;
3344 dstgroup->rfer = srcgroup->rfer;
3345 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
3346 dstgroup->excl = level_size;
3347 dstgroup->excl_cmpr = level_size;
3348 srcgroup->excl = level_size;
3349 srcgroup->excl_cmpr = level_size;
3351 /* inherit the limit info */
3352 dstgroup->lim_flags = srcgroup->lim_flags;
3353 dstgroup->max_rfer = srcgroup->max_rfer;
3354 dstgroup->max_excl = srcgroup->max_excl;
3355 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
3356 dstgroup->rsv_excl = srcgroup->rsv_excl;
3358 qgroup_dirty(fs_info, dstgroup);
3359 qgroup_dirty(fs_info, srcgroup);
3362 * If the source qgroup has parent but the new one doesn't,
3363 * we need a full rescan.
3365 if (!inherit && !list_empty(&srcgroup->groups))
3372 i_qgroups = (u64 *)(inherit + 1);
3373 for (i = 0; i < inherit->num_qgroups; ++i) {
3375 ret = add_relation_rb(fs_info, qlist_prealloc[i], objectid,
3377 qlist_prealloc[i] = NULL;
3382 /* Check if we can do a quick inherit. */
3383 ret = qgroup_snapshot_quick_inherit(fs_info, srcid, *i_qgroups);
3393 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
3394 struct btrfs_qgroup *src;
3395 struct btrfs_qgroup *dst;
3397 if (!i_qgroups[0] || !i_qgroups[1])
3400 src = find_qgroup_rb(fs_info, i_qgroups[0]);
3401 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3408 dst->rfer = src->rfer - level_size;
3409 dst->rfer_cmpr = src->rfer_cmpr - level_size;
3411 /* Manually tweaking numbers certainly needs a rescan */
3414 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
3415 struct btrfs_qgroup *src;
3416 struct btrfs_qgroup *dst;
3418 if (!i_qgroups[0] || !i_qgroups[1])
3421 src = find_qgroup_rb(fs_info, i_qgroups[0]);
3422 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
3429 dst->excl = src->excl + level_size;
3430 dst->excl_cmpr = src->excl_cmpr + level_size;
3435 spin_unlock(&fs_info->qgroup_lock);
3437 ret = btrfs_sysfs_add_one_qgroup(fs_info, dstgroup);
3440 mutex_unlock(&fs_info->qgroup_ioctl_lock);
3442 qgroup_mark_inconsistent(fs_info);
3443 if (qlist_prealloc) {
3444 for (int i = 0; i < inherit->num_qgroups; i++)
3445 kfree(qlist_prealloc[i]);
3446 kfree(qlist_prealloc);
3454 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
3456 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
3457 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
3460 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
3461 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
3467 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
3468 enum btrfs_qgroup_rsv_type type)
3470 struct btrfs_qgroup *qgroup;
3471 struct btrfs_fs_info *fs_info = root->fs_info;
3472 u64 ref_root = root->root_key.objectid;
3474 LIST_HEAD(qgroup_list);
3476 if (!is_fstree(ref_root))
3482 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
3483 capable(CAP_SYS_RESOURCE))
3486 spin_lock(&fs_info->qgroup_lock);
3487 if (!fs_info->quota_root)
3490 qgroup = find_qgroup_rb(fs_info, ref_root);
3494 qgroup_iterator_add(&qgroup_list, qgroup);
3495 list_for_each_entry(qgroup, &qgroup_list, iterator) {
3496 struct btrfs_qgroup_list *glist;
3498 if (enforce && !qgroup_check_limits(qgroup, num_bytes)) {
3503 list_for_each_entry(glist, &qgroup->groups, next_group)
3504 qgroup_iterator_add(&qgroup_list, glist->group);
3509 * no limits exceeded, now record the reservation into all qgroups
3511 list_for_each_entry(qgroup, &qgroup_list, iterator)
3512 qgroup_rsv_add(fs_info, qgroup, num_bytes, type);
3515 qgroup_iterator_clean(&qgroup_list);
3516 spin_unlock(&fs_info->qgroup_lock);
3521 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
3524 * Will handle all higher level qgroup too.
3526 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
3527 * This special case is only used for META_PERTRANS type.
3529 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
3530 u64 ref_root, u64 num_bytes,
3531 enum btrfs_qgroup_rsv_type type)
3533 struct btrfs_qgroup *qgroup;
3534 LIST_HEAD(qgroup_list);
3536 if (!is_fstree(ref_root))
3542 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
3543 WARN(1, "%s: Invalid type to free", __func__);
3546 spin_lock(&fs_info->qgroup_lock);
3548 if (!fs_info->quota_root)
3551 qgroup = find_qgroup_rb(fs_info, ref_root);
3555 if (num_bytes == (u64)-1)
3557 * We're freeing all pertrans rsv, get reserved value from
3558 * level 0 qgroup as real num_bytes to free.
3560 num_bytes = qgroup->rsv.values[type];
3562 qgroup_iterator_add(&qgroup_list, qgroup);
3563 list_for_each_entry(qgroup, &qgroup_list, iterator) {
3564 struct btrfs_qgroup_list *glist;
3566 qgroup_rsv_release(fs_info, qgroup, num_bytes, type);
3567 list_for_each_entry(glist, &qgroup->groups, next_group) {
3568 qgroup_iterator_add(&qgroup_list, glist->group);
3572 qgroup_iterator_clean(&qgroup_list);
3573 spin_unlock(&fs_info->qgroup_lock);
3577 * Check if the leaf is the last leaf. Which means all node pointers
3578 * are at their last position.
3580 static bool is_last_leaf(struct btrfs_path *path)
3584 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3585 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3592 * returns < 0 on error, 0 when more leafs are to be scanned.
3593 * returns 1 when done.
3595 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3596 struct btrfs_path *path)
3598 struct btrfs_fs_info *fs_info = trans->fs_info;
3599 struct btrfs_root *extent_root;
3600 struct btrfs_key found;
3601 struct extent_buffer *scratch_leaf = NULL;
3607 if (!btrfs_qgroup_full_accounting(fs_info))
3610 mutex_lock(&fs_info->qgroup_rescan_lock);
3611 extent_root = btrfs_extent_root(fs_info,
3612 fs_info->qgroup_rescan_progress.objectid);
3613 ret = btrfs_search_slot_for_read(extent_root,
3614 &fs_info->qgroup_rescan_progress,
3617 btrfs_debug(fs_info,
3618 "current progress key (%llu %u %llu), search_slot ret %d",
3619 fs_info->qgroup_rescan_progress.objectid,
3620 fs_info->qgroup_rescan_progress.type,
3621 fs_info->qgroup_rescan_progress.offset, ret);
3625 * The rescan is about to end, we will not be scanning any
3626 * further blocks. We cannot unset the RESCAN flag here, because
3627 * we want to commit the transaction if everything went well.
3628 * To make the live accounting work in this phase, we set our
3629 * scan progress pointer such that every real extent objectid
3632 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3633 btrfs_release_path(path);
3634 mutex_unlock(&fs_info->qgroup_rescan_lock);
3637 done = is_last_leaf(path);
3639 btrfs_item_key_to_cpu(path->nodes[0], &found,
3640 btrfs_header_nritems(path->nodes[0]) - 1);
3641 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3643 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3644 if (!scratch_leaf) {
3646 mutex_unlock(&fs_info->qgroup_rescan_lock);
3649 slot = path->slots[0];
3650 btrfs_release_path(path);
3651 mutex_unlock(&fs_info->qgroup_rescan_lock);
3653 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3654 struct btrfs_backref_walk_ctx ctx = { 0 };
3656 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3657 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3658 found.type != BTRFS_METADATA_ITEM_KEY)
3660 if (found.type == BTRFS_METADATA_ITEM_KEY)
3661 num_bytes = fs_info->nodesize;
3663 num_bytes = found.offset;
3665 ctx.bytenr = found.objectid;
3666 ctx.fs_info = fs_info;
3668 ret = btrfs_find_all_roots(&ctx, false);
3671 /* For rescan, just pass old_roots as NULL */
3672 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3673 num_bytes, NULL, ctx.roots);
3679 free_extent_buffer(scratch_leaf);
3683 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3688 static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
3690 if (btrfs_fs_closing(fs_info))
3692 if (test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state))
3694 if (!btrfs_qgroup_enabled(fs_info))
3696 if (fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN)
3701 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3703 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3704 qgroup_rescan_work);
3705 struct btrfs_path *path;
3706 struct btrfs_trans_handle *trans = NULL;
3709 bool stopped = false;
3710 bool did_leaf_rescans = false;
3712 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE)
3715 path = btrfs_alloc_path();
3719 * Rescan should only search for commit root, and any later difference
3720 * should be recorded by qgroup
3722 path->search_commit_root = 1;
3723 path->skip_locking = 1;
3726 while (!err && !(stopped = rescan_should_stop(fs_info))) {
3727 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3728 if (IS_ERR(trans)) {
3729 err = PTR_ERR(trans);
3733 err = qgroup_rescan_leaf(trans, path);
3734 did_leaf_rescans = true;
3737 btrfs_commit_transaction(trans);
3739 btrfs_end_transaction(trans);
3743 btrfs_free_path(path);
3745 mutex_lock(&fs_info->qgroup_rescan_lock);
3747 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3748 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3749 } else if (err < 0 || stopped) {
3750 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3752 mutex_unlock(&fs_info->qgroup_rescan_lock);
3755 * Only update status, since the previous part has already updated the
3756 * qgroup info, and only if we did any actual work. This also prevents
3757 * race with a concurrent quota disable, which has already set
3758 * fs_info->quota_root to NULL and cleared BTRFS_FS_QUOTA_ENABLED at
3759 * btrfs_quota_disable().
3761 if (did_leaf_rescans) {
3762 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3763 if (IS_ERR(trans)) {
3764 err = PTR_ERR(trans);
3767 "fail to start transaction for status update: %d",
3774 mutex_lock(&fs_info->qgroup_rescan_lock);
3776 fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN)
3777 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3779 ret = update_qgroup_status_item(trans);
3782 btrfs_err(fs_info, "fail to update qgroup status: %d",
3786 fs_info->qgroup_rescan_running = false;
3787 fs_info->qgroup_flags &= ~BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN;
3788 complete_all(&fs_info->qgroup_rescan_completion);
3789 mutex_unlock(&fs_info->qgroup_rescan_lock);
3794 btrfs_end_transaction(trans);
3797 btrfs_info(fs_info, "qgroup scan paused");
3798 } else if (fs_info->qgroup_flags & BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN) {
3799 btrfs_info(fs_info, "qgroup scan cancelled");
3800 } else if (err >= 0) {
3801 btrfs_info(fs_info, "qgroup scan completed%s",
3802 err > 0 ? " (inconsistency flag cleared)" : "");
3804 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3809 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3810 * memory required for the rescan context.
3813 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3818 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE) {
3819 btrfs_warn(fs_info, "qgroup rescan init failed, running in simple mode");
3824 /* we're resuming qgroup rescan at mount time */
3825 if (!(fs_info->qgroup_flags &
3826 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3828 "qgroup rescan init failed, qgroup rescan is not queued");
3830 } else if (!(fs_info->qgroup_flags &
3831 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3833 "qgroup rescan init failed, qgroup is not enabled");
3841 mutex_lock(&fs_info->qgroup_rescan_lock);
3844 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3846 "qgroup rescan is already in progress");
3848 } else if (!(fs_info->qgroup_flags &
3849 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3851 "qgroup rescan init failed, qgroup is not enabled");
3853 } else if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED) {
3854 /* Quota disable is in progress */
3859 mutex_unlock(&fs_info->qgroup_rescan_lock);
3862 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3865 memset(&fs_info->qgroup_rescan_progress, 0,
3866 sizeof(fs_info->qgroup_rescan_progress));
3867 fs_info->qgroup_flags &= ~(BTRFS_QGROUP_RUNTIME_FLAG_CANCEL_RESCAN |
3868 BTRFS_QGROUP_RUNTIME_FLAG_NO_ACCOUNTING);
3869 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3870 init_completion(&fs_info->qgroup_rescan_completion);
3871 mutex_unlock(&fs_info->qgroup_rescan_lock);
3873 btrfs_init_work(&fs_info->qgroup_rescan_work,
3874 btrfs_qgroup_rescan_worker, NULL);
3879 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3882 struct btrfs_qgroup *qgroup;
3884 spin_lock(&fs_info->qgroup_lock);
3885 /* clear all current qgroup tracking information */
3886 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3887 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3889 qgroup->rfer_cmpr = 0;
3891 qgroup->excl_cmpr = 0;
3892 qgroup_dirty(fs_info, qgroup);
3894 spin_unlock(&fs_info->qgroup_lock);
3898 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3901 struct btrfs_trans_handle *trans;
3903 ret = qgroup_rescan_init(fs_info, 0, 1);
3908 * We have set the rescan_progress to 0, which means no more
3909 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3910 * However, btrfs_qgroup_account_ref may be right after its call
3911 * to btrfs_find_all_roots, in which case it would still do the
3913 * To solve this, we're committing the transaction, which will
3914 * ensure we run all delayed refs and only after that, we are
3915 * going to clear all tracking information for a clean start.
3918 trans = btrfs_attach_transaction_barrier(fs_info->fs_root);
3919 if (IS_ERR(trans) && trans != ERR_PTR(-ENOENT)) {
3920 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3921 return PTR_ERR(trans);
3922 } else if (trans != ERR_PTR(-ENOENT)) {
3923 ret = btrfs_commit_transaction(trans);
3925 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3930 qgroup_rescan_zero_tracking(fs_info);
3932 mutex_lock(&fs_info->qgroup_rescan_lock);
3933 fs_info->qgroup_rescan_running = true;
3934 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3935 &fs_info->qgroup_rescan_work);
3936 mutex_unlock(&fs_info->qgroup_rescan_lock);
3941 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3947 mutex_lock(&fs_info->qgroup_rescan_lock);
3948 running = fs_info->qgroup_rescan_running;
3949 mutex_unlock(&fs_info->qgroup_rescan_lock);
3955 ret = wait_for_completion_interruptible(
3956 &fs_info->qgroup_rescan_completion);
3958 wait_for_completion(&fs_info->qgroup_rescan_completion);
3964 * this is only called from open_ctree where we're still single threaded, thus
3965 * locking is omitted here.
3968 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3970 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3971 mutex_lock(&fs_info->qgroup_rescan_lock);
3972 fs_info->qgroup_rescan_running = true;
3973 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3974 &fs_info->qgroup_rescan_work);
3975 mutex_unlock(&fs_info->qgroup_rescan_lock);
3979 #define rbtree_iterate_from_safe(node, next, start) \
3980 for (node = start; node && ({ next = rb_next(node); 1;}); node = next)
3982 static int qgroup_unreserve_range(struct btrfs_inode *inode,
3983 struct extent_changeset *reserved, u64 start,
3986 struct rb_node *node;
3987 struct rb_node *next;
3988 struct ulist_node *entry;
3991 node = reserved->range_changed.root.rb_node;
3995 entry = rb_entry(node, struct ulist_node, rb_node);
3996 if (entry->val < start)
3997 node = node->rb_right;
3999 node = node->rb_left;
4002 if (entry->val > start && rb_prev(&entry->rb_node))
4003 entry = rb_entry(rb_prev(&entry->rb_node), struct ulist_node,
4006 rbtree_iterate_from_safe(node, next, &entry->rb_node) {
4012 entry = rb_entry(node, struct ulist_node, rb_node);
4013 entry_start = entry->val;
4014 entry_end = entry->aux;
4015 entry_len = entry_end - entry_start + 1;
4017 if (entry_start >= start + len)
4019 if (entry_start + entry_len <= start)
4022 * Now the entry is in [start, start + len), revert the
4023 * EXTENT_QGROUP_RESERVED bit.
4025 clear_ret = clear_extent_bits(&inode->io_tree, entry_start,
4026 entry_end, EXTENT_QGROUP_RESERVED);
4027 if (!ret && clear_ret < 0)
4030 ulist_del(&reserved->range_changed, entry->val, entry->aux);
4031 if (likely(reserved->bytes_changed >= entry_len)) {
4032 reserved->bytes_changed -= entry_len;
4035 reserved->bytes_changed = 0;
4043 * Try to free some space for qgroup.
4045 * For qgroup, there are only 3 ways to free qgroup space:
4046 * - Flush nodatacow write
4047 * Any nodatacow write will free its reserved data space at run_delalloc_range().
4048 * In theory, we should only flush nodatacow inodes, but it's not yet
4049 * possible, so we need to flush the whole root.
4051 * - Wait for ordered extents
4052 * When ordered extents are finished, their reserved metadata is finally
4053 * converted to per_trans status, which can be freed by later commit
4056 * - Commit transaction
4057 * This would free the meta_per_trans space.
4058 * In theory this shouldn't provide much space, but any more qgroup space
4061 static int try_flush_qgroup(struct btrfs_root *root)
4063 struct btrfs_trans_handle *trans;
4066 /* Can't hold an open transaction or we run the risk of deadlocking. */
4067 ASSERT(current->journal_info == NULL);
4068 if (WARN_ON(current->journal_info))
4072 * We don't want to run flush again and again, so if there is a running
4073 * one, we won't try to start a new flush, but exit directly.
4075 if (test_and_set_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state)) {
4076 wait_event(root->qgroup_flush_wait,
4077 !test_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state));
4081 ret = btrfs_start_delalloc_snapshot(root, true);
4084 btrfs_wait_ordered_extents(root, U64_MAX, 0, (u64)-1);
4086 trans = btrfs_attach_transaction_barrier(root);
4087 if (IS_ERR(trans)) {
4088 ret = PTR_ERR(trans);
4094 ret = btrfs_commit_transaction(trans);
4096 clear_bit(BTRFS_ROOT_QGROUP_FLUSHING, &root->state);
4097 wake_up(&root->qgroup_flush_wait);
4101 static int qgroup_reserve_data(struct btrfs_inode *inode,
4102 struct extent_changeset **reserved_ret, u64 start,
4105 struct btrfs_root *root = inode->root;
4106 struct extent_changeset *reserved;
4107 bool new_reserved = false;
4112 if (btrfs_qgroup_mode(root->fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
4113 !is_fstree(root->root_key.objectid) || len == 0)
4116 /* @reserved parameter is mandatory for qgroup */
4117 if (WARN_ON(!reserved_ret))
4119 if (!*reserved_ret) {
4120 new_reserved = true;
4121 *reserved_ret = extent_changeset_alloc();
4125 reserved = *reserved_ret;
4126 /* Record already reserved space */
4127 orig_reserved = reserved->bytes_changed;
4128 ret = set_record_extent_bits(&inode->io_tree, start,
4129 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
4131 /* Newly reserved space */
4132 to_reserve = reserved->bytes_changed - orig_reserved;
4133 trace_btrfs_qgroup_reserve_data(&inode->vfs_inode, start, len,
4134 to_reserve, QGROUP_RESERVE);
4137 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
4144 qgroup_unreserve_range(inode, reserved, start, len);
4147 extent_changeset_free(reserved);
4148 *reserved_ret = NULL;
4154 * Reserve qgroup space for range [start, start + len).
4156 * This function will either reserve space from related qgroups or do nothing
4157 * if the range is already reserved.
4159 * Return 0 for successful reservation
4160 * Return <0 for error (including -EQUOT)
4162 * NOTE: This function may sleep for memory allocation, dirty page flushing and
4163 * commit transaction. So caller should not hold any dirty page locked.
4165 int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
4166 struct extent_changeset **reserved_ret, u64 start,
4171 ret = qgroup_reserve_data(inode, reserved_ret, start, len);
4172 if (ret <= 0 && ret != -EDQUOT)
4175 ret = try_flush_qgroup(inode->root);
4178 return qgroup_reserve_data(inode, reserved_ret, start, len);
4181 /* Free ranges specified by @reserved, normally in error path */
4182 static int qgroup_free_reserved_data(struct btrfs_inode *inode,
4183 struct extent_changeset *reserved,
4184 u64 start, u64 len, u64 *freed_ret)
4186 struct btrfs_root *root = inode->root;
4187 struct ulist_node *unode;
4188 struct ulist_iterator uiter;
4189 struct extent_changeset changeset;
4193 extent_changeset_init(&changeset);
4194 len = round_up(start + len, root->fs_info->sectorsize);
4195 start = round_down(start, root->fs_info->sectorsize);
4197 ULIST_ITER_INIT(&uiter);
4198 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
4199 u64 range_start = unode->val;
4200 /* unode->aux is the inclusive end */
4201 u64 range_len = unode->aux - range_start + 1;
4205 extent_changeset_release(&changeset);
4207 /* Only free range in range [start, start + len) */
4208 if (range_start >= start + len ||
4209 range_start + range_len <= start)
4211 free_start = max(range_start, start);
4212 free_len = min(start + len, range_start + range_len) -
4215 * TODO: To also modify reserved->ranges_reserved to reflect
4218 * However as long as we free qgroup reserved according to
4219 * EXTENT_QGROUP_RESERVED, we won't double free.
4220 * So not need to rush.
4222 ret = clear_record_extent_bits(&inode->io_tree, free_start,
4223 free_start + free_len - 1,
4224 EXTENT_QGROUP_RESERVED, &changeset);
4227 freed += changeset.bytes_changed;
4229 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
4230 BTRFS_QGROUP_RSV_DATA);
4235 extent_changeset_release(&changeset);
4239 static int __btrfs_qgroup_release_data(struct btrfs_inode *inode,
4240 struct extent_changeset *reserved, u64 start, u64 len,
4241 u64 *released, int free)
4243 struct extent_changeset changeset;
4244 int trace_op = QGROUP_RELEASE;
4247 if (btrfs_qgroup_mode(inode->root->fs_info) == BTRFS_QGROUP_MODE_DISABLED) {
4248 extent_changeset_init(&changeset);
4249 return clear_record_extent_bits(&inode->io_tree, start,
4251 EXTENT_QGROUP_RESERVED, &changeset);
4254 /* In release case, we shouldn't have @reserved */
4255 WARN_ON(!free && reserved);
4256 if (free && reserved)
4257 return qgroup_free_reserved_data(inode, reserved, start, len, released);
4258 extent_changeset_init(&changeset);
4259 ret = clear_record_extent_bits(&inode->io_tree, start, start + len -1,
4260 EXTENT_QGROUP_RESERVED, &changeset);
4265 trace_op = QGROUP_FREE;
4266 trace_btrfs_qgroup_release_data(&inode->vfs_inode, start, len,
4267 changeset.bytes_changed, trace_op);
4269 btrfs_qgroup_free_refroot(inode->root->fs_info,
4270 inode->root->root_key.objectid,
4271 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4273 *released = changeset.bytes_changed;
4275 extent_changeset_release(&changeset);
4280 * Free a reserved space range from io_tree and related qgroups
4282 * Should be called when a range of pages get invalidated before reaching disk.
4283 * Or for error cleanup case.
4284 * if @reserved is given, only reserved range in [@start, @start + @len) will
4287 * For data written to disk, use btrfs_qgroup_release_data().
4289 * NOTE: This function may sleep for memory allocation.
4291 int btrfs_qgroup_free_data(struct btrfs_inode *inode,
4292 struct extent_changeset *reserved,
4293 u64 start, u64 len, u64 *freed)
4295 return __btrfs_qgroup_release_data(inode, reserved, start, len, freed, 1);
4299 * Release a reserved space range from io_tree only.
4301 * Should be called when a range of pages get written to disk and corresponding
4302 * FILE_EXTENT is inserted into corresponding root.
4304 * Since new qgroup accounting framework will only update qgroup numbers at
4305 * commit_transaction() time, its reserved space shouldn't be freed from
4308 * But we should release the range from io_tree, to allow further write to be
4311 * NOTE: This function may sleep for memory allocation.
4313 int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len, u64 *released)
4315 return __btrfs_qgroup_release_data(inode, NULL, start, len, released, 0);
4318 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
4319 enum btrfs_qgroup_rsv_type type)
4321 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
4322 type != BTRFS_QGROUP_RSV_META_PERTRANS)
4327 spin_lock(&root->qgroup_meta_rsv_lock);
4328 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
4329 root->qgroup_meta_rsv_prealloc += num_bytes;
4331 root->qgroup_meta_rsv_pertrans += num_bytes;
4332 spin_unlock(&root->qgroup_meta_rsv_lock);
4335 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
4336 enum btrfs_qgroup_rsv_type type)
4338 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
4339 type != BTRFS_QGROUP_RSV_META_PERTRANS)
4344 spin_lock(&root->qgroup_meta_rsv_lock);
4345 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
4346 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
4348 root->qgroup_meta_rsv_prealloc -= num_bytes;
4350 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
4352 root->qgroup_meta_rsv_pertrans -= num_bytes;
4354 spin_unlock(&root->qgroup_meta_rsv_lock);
4358 int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
4359 enum btrfs_qgroup_rsv_type type, bool enforce)
4361 struct btrfs_fs_info *fs_info = root->fs_info;
4364 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
4365 !is_fstree(root->root_key.objectid) || num_bytes == 0)
4368 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4369 trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
4370 ret = qgroup_reserve(root, num_bytes, enforce, type);
4374 * Record what we have reserved into root.
4376 * To avoid quota disabled->enabled underflow.
4377 * In that case, we may try to free space we haven't reserved
4378 * (since quota was disabled), so record what we reserved into root.
4379 * And ensure later release won't underflow this number.
4381 add_root_meta_rsv(root, num_bytes, type);
4385 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
4386 enum btrfs_qgroup_rsv_type type, bool enforce,
4391 ret = btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4392 if ((ret <= 0 && ret != -EDQUOT) || noflush)
4395 ret = try_flush_qgroup(root);
4398 return btrfs_qgroup_reserve_meta(root, num_bytes, type, enforce);
4402 * Per-transaction meta reservation should be all freed at transaction commit
4405 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
4407 struct btrfs_fs_info *fs_info = root->fs_info;
4409 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
4410 !is_fstree(root->root_key.objectid))
4413 /* TODO: Update trace point to handle such free */
4414 trace_qgroup_meta_free_all_pertrans(root);
4415 /* Special value -1 means to free all reserved space */
4416 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
4417 BTRFS_QGROUP_RSV_META_PERTRANS);
4420 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
4421 enum btrfs_qgroup_rsv_type type)
4423 struct btrfs_fs_info *fs_info = root->fs_info;
4425 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
4426 !is_fstree(root->root_key.objectid))
4430 * reservation for META_PREALLOC can happen before quota is enabled,
4431 * which can lead to underflow.
4432 * Here ensure we will only free what we really have reserved.
4434 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
4435 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
4436 trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
4437 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
4441 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
4444 struct btrfs_qgroup *qgroup;
4445 LIST_HEAD(qgroup_list);
4449 if (!fs_info->quota_root)
4452 spin_lock(&fs_info->qgroup_lock);
4453 qgroup = find_qgroup_rb(fs_info, ref_root);
4457 qgroup_iterator_add(&qgroup_list, qgroup);
4458 list_for_each_entry(qgroup, &qgroup_list, iterator) {
4459 struct btrfs_qgroup_list *glist;
4461 qgroup_rsv_release(fs_info, qgroup, num_bytes,
4462 BTRFS_QGROUP_RSV_META_PREALLOC);
4463 if (!sb_rdonly(fs_info->sb))
4464 qgroup_rsv_add(fs_info, qgroup, num_bytes,
4465 BTRFS_QGROUP_RSV_META_PERTRANS);
4467 list_for_each_entry(glist, &qgroup->groups, next_group)
4468 qgroup_iterator_add(&qgroup_list, glist->group);
4471 qgroup_iterator_clean(&qgroup_list);
4472 spin_unlock(&fs_info->qgroup_lock);
4476 * Convert @num_bytes of META_PREALLOCATED reservation to META_PERTRANS.
4478 * This is called when preallocated meta reservation needs to be used.
4479 * Normally after btrfs_join_transaction() call.
4481 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
4483 struct btrfs_fs_info *fs_info = root->fs_info;
4485 if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_DISABLED ||
4486 !is_fstree(root->root_key.objectid))
4488 /* Same as btrfs_qgroup_free_meta_prealloc() */
4489 num_bytes = sub_root_meta_rsv(root, num_bytes,
4490 BTRFS_QGROUP_RSV_META_PREALLOC);
4491 trace_qgroup_meta_convert(root, num_bytes);
4492 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
4493 if (!sb_rdonly(fs_info->sb))
4494 add_root_meta_rsv(root, num_bytes, BTRFS_QGROUP_RSV_META_PERTRANS);
4498 * Check qgroup reserved space leaking, normally at destroy inode
4501 void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode)
4503 struct extent_changeset changeset;
4504 struct ulist_node *unode;
4505 struct ulist_iterator iter;
4508 extent_changeset_init(&changeset);
4509 ret = clear_record_extent_bits(&inode->io_tree, 0, (u64)-1,
4510 EXTENT_QGROUP_RESERVED, &changeset);
4513 if (WARN_ON(changeset.bytes_changed)) {
4514 ULIST_ITER_INIT(&iter);
4515 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
4516 btrfs_warn(inode->root->fs_info,
4517 "leaking qgroup reserved space, ino: %llu, start: %llu, end: %llu",
4518 btrfs_ino(inode), unode->val, unode->aux);
4520 btrfs_qgroup_free_refroot(inode->root->fs_info,
4521 inode->root->root_key.objectid,
4522 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
4525 extent_changeset_release(&changeset);
4528 void btrfs_qgroup_init_swapped_blocks(
4529 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
4533 spin_lock_init(&swapped_blocks->lock);
4534 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
4535 swapped_blocks->blocks[i] = RB_ROOT;
4536 swapped_blocks->swapped = false;
4540 * Delete all swapped blocks record of @root.
4541 * Every record here means we skipped a full subtree scan for qgroup.
4543 * Gets called when committing one transaction.
4545 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
4547 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
4550 swapped_blocks = &root->swapped_blocks;
4552 spin_lock(&swapped_blocks->lock);
4553 if (!swapped_blocks->swapped)
4555 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4556 struct rb_root *cur_root = &swapped_blocks->blocks[i];
4557 struct btrfs_qgroup_swapped_block *entry;
4558 struct btrfs_qgroup_swapped_block *next;
4560 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
4563 swapped_blocks->blocks[i] = RB_ROOT;
4565 swapped_blocks->swapped = false;
4567 spin_unlock(&swapped_blocks->lock);
4571 * Add subtree roots record into @subvol_root.
4573 * @subvol_root: tree root of the subvolume tree get swapped
4574 * @bg: block group under balance
4575 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
4576 * @reloc_parent/slot: pointer to the subtree root in reloc tree
4577 * BOTH POINTERS ARE BEFORE TREE SWAP
4578 * @last_snapshot: last snapshot generation of the subvolume tree
4580 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
4581 struct btrfs_root *subvol_root,
4582 struct btrfs_block_group *bg,
4583 struct extent_buffer *subvol_parent, int subvol_slot,
4584 struct extent_buffer *reloc_parent, int reloc_slot,
4587 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
4588 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
4589 struct btrfs_qgroup_swapped_block *block;
4590 struct rb_node **cur;
4591 struct rb_node *parent = NULL;
4592 int level = btrfs_header_level(subvol_parent) - 1;
4595 if (!btrfs_qgroup_full_accounting(fs_info))
4598 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
4599 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
4600 btrfs_err_rl(fs_info,
4601 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
4603 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
4604 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
4608 block = kmalloc(sizeof(*block), GFP_NOFS);
4615 * @reloc_parent/slot is still before swap, while @block is going to
4616 * record the bytenr after swap, so we do the swap here.
4618 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
4619 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
4621 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
4622 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
4624 block->last_snapshot = last_snapshot;
4625 block->level = level;
4628 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
4629 * no one else can modify tree blocks thus we qgroup will not change
4630 * no matter the value of trace_leaf.
4632 if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
4633 block->trace_leaf = true;
4635 block->trace_leaf = false;
4636 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
4638 /* Insert @block into @blocks */
4639 spin_lock(&blocks->lock);
4640 cur = &blocks->blocks[level].rb_node;
4642 struct btrfs_qgroup_swapped_block *entry;
4645 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
4648 if (entry->subvol_bytenr < block->subvol_bytenr) {
4649 cur = &(*cur)->rb_left;
4650 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
4651 cur = &(*cur)->rb_right;
4653 if (entry->subvol_generation !=
4654 block->subvol_generation ||
4655 entry->reloc_bytenr != block->reloc_bytenr ||
4656 entry->reloc_generation !=
4657 block->reloc_generation) {
4659 * Duplicated but mismatch entry found.
4662 * Marking qgroup inconsistent should be enough
4665 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
4672 rb_link_node(&block->node, parent, cur);
4673 rb_insert_color(&block->node, &blocks->blocks[level]);
4674 blocks->swapped = true;
4676 spin_unlock(&blocks->lock);
4679 qgroup_mark_inconsistent(fs_info);
4684 * Check if the tree block is a subtree root, and if so do the needed
4685 * delayed subtree trace for qgroup.
4687 * This is called during btrfs_cow_block().
4689 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
4690 struct btrfs_root *root,
4691 struct extent_buffer *subvol_eb)
4693 struct btrfs_fs_info *fs_info = root->fs_info;
4694 struct btrfs_tree_parent_check check = { 0 };
4695 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
4696 struct btrfs_qgroup_swapped_block *block;
4697 struct extent_buffer *reloc_eb = NULL;
4698 struct rb_node *node;
4700 bool swapped = false;
4701 int level = btrfs_header_level(subvol_eb);
4705 if (!btrfs_qgroup_full_accounting(fs_info))
4707 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
4710 spin_lock(&blocks->lock);
4711 if (!blocks->swapped) {
4712 spin_unlock(&blocks->lock);
4715 node = blocks->blocks[level].rb_node;
4718 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
4719 if (block->subvol_bytenr < subvol_eb->start) {
4720 node = node->rb_left;
4721 } else if (block->subvol_bytenr > subvol_eb->start) {
4722 node = node->rb_right;
4729 spin_unlock(&blocks->lock);
4732 /* Found one, remove it from @blocks first and update blocks->swapped */
4733 rb_erase(&block->node, &blocks->blocks[level]);
4734 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
4735 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
4740 blocks->swapped = swapped;
4741 spin_unlock(&blocks->lock);
4743 check.level = block->level;
4744 check.transid = block->reloc_generation;
4745 check.has_first_key = true;
4746 memcpy(&check.first_key, &block->first_key, sizeof(check.first_key));
4748 /* Read out reloc subtree root */
4749 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr, &check);
4750 if (IS_ERR(reloc_eb)) {
4751 ret = PTR_ERR(reloc_eb);
4755 if (!extent_buffer_uptodate(reloc_eb)) {
4760 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
4761 block->last_snapshot, block->trace_leaf);
4764 free_extent_buffer(reloc_eb);
4767 btrfs_err_rl(fs_info,
4768 "failed to account subtree at bytenr %llu: %d",
4769 subvol_eb->start, ret);
4770 qgroup_mark_inconsistent(fs_info);
4775 void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans)
4777 struct btrfs_qgroup_extent_record *entry;
4778 struct btrfs_qgroup_extent_record *next;
4779 struct rb_root *root;
4781 root = &trans->delayed_refs.dirty_extent_root;
4782 rbtree_postorder_for_each_entry_safe(entry, next, root, node) {
4783 ulist_free(entry->old_roots);
4789 void btrfs_free_squota_rsv(struct btrfs_fs_info *fs_info, u64 root, u64 rsv_bytes)
4791 if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE)
4794 if (!is_fstree(root))
4797 btrfs_qgroup_free_refroot(fs_info, root, rsv_bytes, BTRFS_QGROUP_RSV_DATA);
4800 int btrfs_record_squota_delta(struct btrfs_fs_info *fs_info,
4801 struct btrfs_squota_delta *delta)
4804 struct btrfs_qgroup *qgroup;
4805 struct btrfs_qgroup *qg;
4806 LIST_HEAD(qgroup_list);
4807 u64 root = delta->root;
4808 u64 num_bytes = delta->num_bytes;
4809 const int sign = (delta->is_inc ? 1 : -1);
4811 if (btrfs_qgroup_mode(fs_info) != BTRFS_QGROUP_MODE_SIMPLE)
4814 if (!is_fstree(root))
4817 /* If the extent predates enabling quotas, don't count it. */
4818 if (delta->generation < fs_info->qgroup_enable_gen)
4821 spin_lock(&fs_info->qgroup_lock);
4822 qgroup = find_qgroup_rb(fs_info, root);
4829 qgroup_iterator_add(&qgroup_list, qgroup);
4830 list_for_each_entry(qg, &qgroup_list, iterator) {
4831 struct btrfs_qgroup_list *glist;
4833 qg->excl += num_bytes * sign;
4834 qg->rfer += num_bytes * sign;
4835 qgroup_dirty(fs_info, qg);
4837 list_for_each_entry(glist, &qg->groups, next_group)
4838 qgroup_iterator_add(&qgroup_list, glist->group);
4840 qgroup_iterator_clean(&qgroup_list);
4843 spin_unlock(&fs_info->qgroup_lock);