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/sizes.h>
17 #include "transaction.h"
22 #include "extent_io.h"
27 * - subvol delete -> delete when ref goes to 0? delete limits also?
31 * - copy also limits on subvol creation
33 * - caches fuer ulists
34 * - performance benchmarks
35 * - check all ioctl parameters
39 * Helpers to access qgroup reservation
41 * Callers should ensure the lock context and type are valid
44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
49 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
50 ret += qgroup->rsv.values[i];
55 #ifdef CONFIG_BTRFS_DEBUG
56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
58 if (type == BTRFS_QGROUP_RSV_DATA)
60 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61 return "meta_pertrans";
62 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63 return "meta_prealloc";
68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69 struct btrfs_qgroup *qgroup, u64 num_bytes,
70 enum btrfs_qgroup_rsv_type type)
72 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
73 qgroup->rsv.values[type] += num_bytes;
76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77 struct btrfs_qgroup *qgroup, u64 num_bytes,
78 enum btrfs_qgroup_rsv_type type)
80 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
81 if (qgroup->rsv.values[type] >= num_bytes) {
82 qgroup->rsv.values[type] -= num_bytes;
85 #ifdef CONFIG_BTRFS_DEBUG
87 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
88 qgroup->qgroupid, qgroup_rsv_type_str(type),
89 qgroup->rsv.values[type], num_bytes);
91 qgroup->rsv.values[type] = 0;
94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
95 struct btrfs_qgroup *dest,
96 struct btrfs_qgroup *src)
100 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
101 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105 struct btrfs_qgroup *dest,
106 struct btrfs_qgroup *src)
110 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
111 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
117 if (qg->old_refcnt < seq)
118 qg->old_refcnt = seq;
119 qg->old_refcnt += mod;
122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
125 if (qg->new_refcnt < seq)
126 qg->new_refcnt = seq;
127 qg->new_refcnt += mod;
130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
132 if (qg->old_refcnt < seq)
134 return qg->old_refcnt - seq;
137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
139 if (qg->new_refcnt < seq)
141 return qg->new_refcnt - seq;
145 * glue structure to represent the relations between qgroups.
147 struct btrfs_qgroup_list {
148 struct list_head next_group;
149 struct list_head next_member;
150 struct btrfs_qgroup *group;
151 struct btrfs_qgroup *member;
154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
156 return (u64)(uintptr_t)qg;
159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
161 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
169 /* must be called with qgroup_ioctl_lock held */
170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
173 struct rb_node *n = fs_info->qgroup_tree.rb_node;
174 struct btrfs_qgroup *qgroup;
177 qgroup = rb_entry(n, struct btrfs_qgroup, node);
178 if (qgroup->qgroupid < qgroupid)
180 else if (qgroup->qgroupid > qgroupid)
188 /* must be called with qgroup_lock held */
189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
192 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
193 struct rb_node *parent = NULL;
194 struct btrfs_qgroup *qgroup;
198 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
200 if (qgroup->qgroupid < qgroupid)
202 else if (qgroup->qgroupid > qgroupid)
208 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
210 return ERR_PTR(-ENOMEM);
212 qgroup->qgroupid = qgroupid;
213 INIT_LIST_HEAD(&qgroup->groups);
214 INIT_LIST_HEAD(&qgroup->members);
215 INIT_LIST_HEAD(&qgroup->dirty);
217 rb_link_node(&qgroup->node, parent, p);
218 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
223 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
225 struct btrfs_qgroup_list *list;
227 list_del(&qgroup->dirty);
228 while (!list_empty(&qgroup->groups)) {
229 list = list_first_entry(&qgroup->groups,
230 struct btrfs_qgroup_list, next_group);
231 list_del(&list->next_group);
232 list_del(&list->next_member);
236 while (!list_empty(&qgroup->members)) {
237 list = list_first_entry(&qgroup->members,
238 struct btrfs_qgroup_list, next_member);
239 list_del(&list->next_group);
240 list_del(&list->next_member);
246 /* must be called with qgroup_lock held */
247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
249 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
254 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
255 __del_qgroup_rb(qgroup);
259 /* must be called with qgroup_lock held */
260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
261 u64 memberid, u64 parentid)
263 struct btrfs_qgroup *member;
264 struct btrfs_qgroup *parent;
265 struct btrfs_qgroup_list *list;
267 member = find_qgroup_rb(fs_info, memberid);
268 parent = find_qgroup_rb(fs_info, parentid);
269 if (!member || !parent)
272 list = kzalloc(sizeof(*list), GFP_ATOMIC);
276 list->group = parent;
277 list->member = member;
278 list_add_tail(&list->next_group, &member->groups);
279 list_add_tail(&list->next_member, &parent->members);
284 /* must be called with qgroup_lock held */
285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
286 u64 memberid, u64 parentid)
288 struct btrfs_qgroup *member;
289 struct btrfs_qgroup *parent;
290 struct btrfs_qgroup_list *list;
292 member = find_qgroup_rb(fs_info, memberid);
293 parent = find_qgroup_rb(fs_info, parentid);
294 if (!member || !parent)
297 list_for_each_entry(list, &member->groups, next_group) {
298 if (list->group == parent) {
299 list_del(&list->next_group);
300 list_del(&list->next_member);
308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
312 struct btrfs_qgroup *qgroup;
314 qgroup = find_qgroup_rb(fs_info, qgroupid);
317 if (qgroup->rfer != rfer || qgroup->excl != excl)
324 * The full config is read in one go, only called from open_ctree()
325 * It doesn't use any locking, as at this point we're still single-threaded
327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
329 struct btrfs_key key;
330 struct btrfs_key found_key;
331 struct btrfs_root *quota_root = fs_info->quota_root;
332 struct btrfs_path *path = NULL;
333 struct extent_buffer *l;
337 u64 rescan_progress = 0;
339 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
342 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
343 if (!fs_info->qgroup_ulist) {
348 path = btrfs_alloc_path();
354 /* default this to quota off, in case no status key is found */
355 fs_info->qgroup_flags = 0;
358 * pass 1: read status, all qgroup infos and limits
363 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
368 struct btrfs_qgroup *qgroup;
370 slot = path->slots[0];
372 btrfs_item_key_to_cpu(l, &found_key, slot);
374 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
375 struct btrfs_qgroup_status_item *ptr;
377 ptr = btrfs_item_ptr(l, slot,
378 struct btrfs_qgroup_status_item);
380 if (btrfs_qgroup_status_version(l, ptr) !=
381 BTRFS_QGROUP_STATUS_VERSION) {
383 "old qgroup version, quota disabled");
386 if (btrfs_qgroup_status_generation(l, ptr) !=
387 fs_info->generation) {
388 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
390 "qgroup generation mismatch, marked as inconsistent");
392 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
394 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
398 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
399 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
402 qgroup = find_qgroup_rb(fs_info, found_key.offset);
403 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
404 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
405 btrfs_err(fs_info, "inconsistent qgroup config");
406 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
409 qgroup = add_qgroup_rb(fs_info, found_key.offset);
410 if (IS_ERR(qgroup)) {
411 ret = PTR_ERR(qgroup);
415 switch (found_key.type) {
416 case BTRFS_QGROUP_INFO_KEY: {
417 struct btrfs_qgroup_info_item *ptr;
419 ptr = btrfs_item_ptr(l, slot,
420 struct btrfs_qgroup_info_item);
421 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
422 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
423 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
424 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
425 /* generation currently unused */
428 case BTRFS_QGROUP_LIMIT_KEY: {
429 struct btrfs_qgroup_limit_item *ptr;
431 ptr = btrfs_item_ptr(l, slot,
432 struct btrfs_qgroup_limit_item);
433 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
434 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
435 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
436 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
437 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
442 ret = btrfs_next_item(quota_root, path);
448 btrfs_release_path(path);
451 * pass 2: read all qgroup relations
454 key.type = BTRFS_QGROUP_RELATION_KEY;
456 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
460 slot = path->slots[0];
462 btrfs_item_key_to_cpu(l, &found_key, slot);
464 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
467 if (found_key.objectid > found_key.offset) {
468 /* parent <- member, not needed to build config */
469 /* FIXME should we omit the key completely? */
473 ret = add_relation_rb(fs_info, found_key.objectid,
475 if (ret == -ENOENT) {
477 "orphan qgroup relation 0x%llx->0x%llx",
478 found_key.objectid, found_key.offset);
479 ret = 0; /* ignore the error */
484 ret = btrfs_next_item(quota_root, path);
491 btrfs_free_path(path);
492 fs_info->qgroup_flags |= flags;
493 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
494 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
495 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
497 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
500 ulist_free(fs_info->qgroup_ulist);
501 fs_info->qgroup_ulist = NULL;
502 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
505 return ret < 0 ? ret : 0;
509 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
510 * first two are in single-threaded paths.And for the third one, we have set
511 * quota_root to be null with qgroup_lock held before, so it is safe to clean
512 * up the in-memory structures without qgroup_lock held.
514 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
517 struct btrfs_qgroup *qgroup;
519 while ((n = rb_first(&fs_info->qgroup_tree))) {
520 qgroup = rb_entry(n, struct btrfs_qgroup, node);
521 rb_erase(n, &fs_info->qgroup_tree);
522 __del_qgroup_rb(qgroup);
525 * we call btrfs_free_qgroup_config() when umounting
526 * filesystem and disabling quota, so we set qgroup_ulist
527 * to be null here to avoid double free.
529 ulist_free(fs_info->qgroup_ulist);
530 fs_info->qgroup_ulist = NULL;
533 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
537 struct btrfs_root *quota_root = trans->fs_info->quota_root;
538 struct btrfs_path *path;
539 struct btrfs_key key;
541 path = btrfs_alloc_path();
546 key.type = BTRFS_QGROUP_RELATION_KEY;
549 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
551 btrfs_mark_buffer_dirty(path->nodes[0]);
553 btrfs_free_path(path);
557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
561 struct btrfs_root *quota_root = trans->fs_info->quota_root;
562 struct btrfs_path *path;
563 struct btrfs_key key;
565 path = btrfs_alloc_path();
570 key.type = BTRFS_QGROUP_RELATION_KEY;
573 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
582 ret = btrfs_del_item(trans, quota_root, path);
584 btrfs_free_path(path);
588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
589 struct btrfs_root *quota_root, u64 qgroupid)
592 struct btrfs_path *path;
593 struct btrfs_qgroup_info_item *qgroup_info;
594 struct btrfs_qgroup_limit_item *qgroup_limit;
595 struct extent_buffer *leaf;
596 struct btrfs_key key;
598 if (btrfs_is_testing(quota_root->fs_info))
601 path = btrfs_alloc_path();
606 key.type = BTRFS_QGROUP_INFO_KEY;
607 key.offset = qgroupid;
610 * Avoid a transaction abort by catching -EEXIST here. In that
611 * case, we proceed by re-initializing the existing structure
615 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616 sizeof(*qgroup_info));
617 if (ret && ret != -EEXIST)
620 leaf = path->nodes[0];
621 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
622 struct btrfs_qgroup_info_item);
623 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
624 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
625 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
626 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
627 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
629 btrfs_mark_buffer_dirty(leaf);
631 btrfs_release_path(path);
633 key.type = BTRFS_QGROUP_LIMIT_KEY;
634 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
635 sizeof(*qgroup_limit));
636 if (ret && ret != -EEXIST)
639 leaf = path->nodes[0];
640 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
641 struct btrfs_qgroup_limit_item);
642 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
643 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
644 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
645 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
646 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
648 btrfs_mark_buffer_dirty(leaf);
652 btrfs_free_path(path);
656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
659 struct btrfs_root *quota_root = trans->fs_info->quota_root;
660 struct btrfs_path *path;
661 struct btrfs_key key;
663 path = btrfs_alloc_path();
668 key.type = BTRFS_QGROUP_INFO_KEY;
669 key.offset = qgroupid;
670 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
679 ret = btrfs_del_item(trans, quota_root, path);
683 btrfs_release_path(path);
685 key.type = BTRFS_QGROUP_LIMIT_KEY;
686 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
695 ret = btrfs_del_item(trans, quota_root, path);
698 btrfs_free_path(path);
702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
703 struct btrfs_qgroup *qgroup)
705 struct btrfs_root *quota_root = trans->fs_info->quota_root;
706 struct btrfs_path *path;
707 struct btrfs_key key;
708 struct extent_buffer *l;
709 struct btrfs_qgroup_limit_item *qgroup_limit;
714 key.type = BTRFS_QGROUP_LIMIT_KEY;
715 key.offset = qgroup->qgroupid;
717 path = btrfs_alloc_path();
721 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
729 slot = path->slots[0];
730 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
731 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
732 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
733 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
734 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
735 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
737 btrfs_mark_buffer_dirty(l);
740 btrfs_free_path(path);
744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
745 struct btrfs_qgroup *qgroup)
747 struct btrfs_fs_info *fs_info = trans->fs_info;
748 struct btrfs_root *quota_root = fs_info->quota_root;
749 struct btrfs_path *path;
750 struct btrfs_key key;
751 struct extent_buffer *l;
752 struct btrfs_qgroup_info_item *qgroup_info;
756 if (btrfs_is_testing(fs_info))
760 key.type = BTRFS_QGROUP_INFO_KEY;
761 key.offset = qgroup->qgroupid;
763 path = btrfs_alloc_path();
767 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
775 slot = path->slots[0];
776 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
777 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
778 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
779 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
780 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
781 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
783 btrfs_mark_buffer_dirty(l);
786 btrfs_free_path(path);
790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
792 struct btrfs_fs_info *fs_info = trans->fs_info;
793 struct btrfs_root *quota_root = fs_info->quota_root;
794 struct btrfs_path *path;
795 struct btrfs_key key;
796 struct extent_buffer *l;
797 struct btrfs_qgroup_status_item *ptr;
802 key.type = BTRFS_QGROUP_STATUS_KEY;
805 path = btrfs_alloc_path();
809 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
817 slot = path->slots[0];
818 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
819 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
820 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
821 btrfs_set_qgroup_status_rescan(l, ptr,
822 fs_info->qgroup_rescan_progress.objectid);
824 btrfs_mark_buffer_dirty(l);
827 btrfs_free_path(path);
832 * called with qgroup_lock held
834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
835 struct btrfs_root *root)
837 struct btrfs_path *path;
838 struct btrfs_key key;
839 struct extent_buffer *leaf = NULL;
843 path = btrfs_alloc_path();
847 path->leave_spinning = 1;
854 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
857 leaf = path->nodes[0];
858 nr = btrfs_header_nritems(leaf);
862 * delete the leaf one by one
863 * since the whole tree is going
867 ret = btrfs_del_items(trans, root, path, 0, nr);
871 btrfs_release_path(path);
875 btrfs_free_path(path);
879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
881 struct btrfs_root *quota_root;
882 struct btrfs_root *tree_root = fs_info->tree_root;
883 struct btrfs_path *path = NULL;
884 struct btrfs_qgroup_status_item *ptr;
885 struct extent_buffer *leaf;
886 struct btrfs_key key;
887 struct btrfs_key found_key;
888 struct btrfs_qgroup *qgroup = NULL;
889 struct btrfs_trans_handle *trans = NULL;
893 mutex_lock(&fs_info->qgroup_ioctl_lock);
894 if (fs_info->quota_root)
898 * 1 for quota root item
899 * 1 for BTRFS_QGROUP_STATUS item
901 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
902 * per subvolume. However those are not currently reserved since it
903 * would be a lot of overkill.
905 trans = btrfs_start_transaction(tree_root, 2);
907 ret = PTR_ERR(trans);
912 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
913 if (!fs_info->qgroup_ulist) {
915 btrfs_abort_transaction(trans, ret);
920 * initially create the quota tree
922 quota_root = btrfs_create_tree(trans, fs_info,
923 BTRFS_QUOTA_TREE_OBJECTID);
924 if (IS_ERR(quota_root)) {
925 ret = PTR_ERR(quota_root);
926 btrfs_abort_transaction(trans, ret);
930 path = btrfs_alloc_path();
933 btrfs_abort_transaction(trans, ret);
938 key.type = BTRFS_QGROUP_STATUS_KEY;
941 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
944 btrfs_abort_transaction(trans, ret);
948 leaf = path->nodes[0];
949 ptr = btrfs_item_ptr(leaf, path->slots[0],
950 struct btrfs_qgroup_status_item);
951 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
952 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
953 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
954 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
955 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
956 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
958 btrfs_mark_buffer_dirty(leaf);
961 key.type = BTRFS_ROOT_REF_KEY;
964 btrfs_release_path(path);
965 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
969 btrfs_abort_transaction(trans, ret);
974 slot = path->slots[0];
975 leaf = path->nodes[0];
976 btrfs_item_key_to_cpu(leaf, &found_key, slot);
978 if (found_key.type == BTRFS_ROOT_REF_KEY) {
979 ret = add_qgroup_item(trans, quota_root,
982 btrfs_abort_transaction(trans, ret);
986 qgroup = add_qgroup_rb(fs_info, found_key.offset);
987 if (IS_ERR(qgroup)) {
988 ret = PTR_ERR(qgroup);
989 btrfs_abort_transaction(trans, ret);
993 ret = btrfs_next_item(tree_root, path);
995 btrfs_abort_transaction(trans, ret);
1003 btrfs_release_path(path);
1004 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1006 btrfs_abort_transaction(trans, ret);
1010 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1011 if (IS_ERR(qgroup)) {
1012 ret = PTR_ERR(qgroup);
1013 btrfs_abort_transaction(trans, ret);
1017 ret = btrfs_commit_transaction(trans);
1023 * Set quota enabled flag after committing the transaction, to avoid
1024 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1027 spin_lock(&fs_info->qgroup_lock);
1028 fs_info->quota_root = quota_root;
1029 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1030 spin_unlock(&fs_info->qgroup_lock);
1032 ret = qgroup_rescan_init(fs_info, 0, 1);
1034 qgroup_rescan_zero_tracking(fs_info);
1035 fs_info->qgroup_rescan_running = true;
1036 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1037 &fs_info->qgroup_rescan_work);
1041 btrfs_free_path(path);
1044 free_extent_buffer(quota_root->node);
1045 free_extent_buffer(quota_root->commit_root);
1050 ulist_free(fs_info->qgroup_ulist);
1051 fs_info->qgroup_ulist = NULL;
1053 btrfs_end_transaction(trans);
1055 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1059 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1061 struct btrfs_root *quota_root;
1062 struct btrfs_trans_handle *trans = NULL;
1065 mutex_lock(&fs_info->qgroup_ioctl_lock);
1066 if (!fs_info->quota_root)
1070 * 1 For the root item
1072 * We should also reserve enough items for the quota tree deletion in
1073 * btrfs_clean_quota_tree but this is not done.
1075 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1076 if (IS_ERR(trans)) {
1077 ret = PTR_ERR(trans);
1081 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1082 btrfs_qgroup_wait_for_completion(fs_info, false);
1083 spin_lock(&fs_info->qgroup_lock);
1084 quota_root = fs_info->quota_root;
1085 fs_info->quota_root = NULL;
1086 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1087 spin_unlock(&fs_info->qgroup_lock);
1089 btrfs_free_qgroup_config(fs_info);
1091 ret = btrfs_clean_quota_tree(trans, quota_root);
1093 btrfs_abort_transaction(trans, ret);
1097 ret = btrfs_del_root(trans, "a_root->root_key);
1099 btrfs_abort_transaction(trans, ret);
1103 list_del("a_root->dirty_list);
1105 btrfs_tree_lock(quota_root->node);
1106 clean_tree_block(fs_info, quota_root->node);
1107 btrfs_tree_unlock(quota_root->node);
1108 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1110 free_extent_buffer(quota_root->node);
1111 free_extent_buffer(quota_root->commit_root);
1115 ret = btrfs_end_transaction(trans);
1117 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1121 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1122 struct btrfs_qgroup *qgroup)
1124 if (list_empty(&qgroup->dirty))
1125 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1129 * The easy accounting, we're updating qgroup relationship whose child qgroup
1130 * only has exclusive extents.
1132 * In this case, all exclsuive extents will also be exlusive for parent, so
1133 * excl/rfer just get added/removed.
1135 * So is qgroup reservation space, which should also be added/removed to
1137 * Or when child tries to release reservation space, parent will underflow its
1138 * reservation (for relationship adding case).
1140 * Caller should hold fs_info->qgroup_lock.
1142 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1143 struct ulist *tmp, u64 ref_root,
1144 struct btrfs_qgroup *src, int sign)
1146 struct btrfs_qgroup *qgroup;
1147 struct btrfs_qgroup_list *glist;
1148 struct ulist_node *unode;
1149 struct ulist_iterator uiter;
1150 u64 num_bytes = src->excl;
1153 qgroup = find_qgroup_rb(fs_info, ref_root);
1157 qgroup->rfer += sign * num_bytes;
1158 qgroup->rfer_cmpr += sign * num_bytes;
1160 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1161 qgroup->excl += sign * num_bytes;
1162 qgroup->excl_cmpr += sign * num_bytes;
1165 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1167 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1169 qgroup_dirty(fs_info, qgroup);
1171 /* Get all of the parent groups that contain this qgroup */
1172 list_for_each_entry(glist, &qgroup->groups, next_group) {
1173 ret = ulist_add(tmp, glist->group->qgroupid,
1174 qgroup_to_aux(glist->group), GFP_ATOMIC);
1179 /* Iterate all of the parents and adjust their reference counts */
1180 ULIST_ITER_INIT(&uiter);
1181 while ((unode = ulist_next(tmp, &uiter))) {
1182 qgroup = unode_aux_to_qgroup(unode);
1183 qgroup->rfer += sign * num_bytes;
1184 qgroup->rfer_cmpr += sign * num_bytes;
1185 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1186 qgroup->excl += sign * num_bytes;
1188 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1190 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1191 qgroup->excl_cmpr += sign * num_bytes;
1192 qgroup_dirty(fs_info, qgroup);
1194 /* Add any parents of the parents */
1195 list_for_each_entry(glist, &qgroup->groups, next_group) {
1196 ret = ulist_add(tmp, glist->group->qgroupid,
1197 qgroup_to_aux(glist->group), GFP_ATOMIC);
1209 * Quick path for updating qgroup with only excl refs.
1211 * In that case, just update all parent will be enough.
1212 * Or we needs to do a full rescan.
1213 * Caller should also hold fs_info->qgroup_lock.
1215 * Return 0 for quick update, return >0 for need to full rescan
1216 * and mark INCONSISTENT flag.
1217 * Return < 0 for other error.
1219 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1220 struct ulist *tmp, u64 src, u64 dst,
1223 struct btrfs_qgroup *qgroup;
1227 qgroup = find_qgroup_rb(fs_info, src);
1230 if (qgroup->excl == qgroup->rfer) {
1232 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1241 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1245 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1248 struct btrfs_fs_info *fs_info = trans->fs_info;
1249 struct btrfs_root *quota_root;
1250 struct btrfs_qgroup *parent;
1251 struct btrfs_qgroup *member;
1252 struct btrfs_qgroup_list *list;
1256 /* Check the level of src and dst first */
1257 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1260 tmp = ulist_alloc(GFP_KERNEL);
1264 mutex_lock(&fs_info->qgroup_ioctl_lock);
1265 quota_root = fs_info->quota_root;
1270 member = find_qgroup_rb(fs_info, src);
1271 parent = find_qgroup_rb(fs_info, dst);
1272 if (!member || !parent) {
1277 /* check if such qgroup relation exist firstly */
1278 list_for_each_entry(list, &member->groups, next_group) {
1279 if (list->group == parent) {
1285 ret = add_qgroup_relation_item(trans, src, dst);
1289 ret = add_qgroup_relation_item(trans, dst, src);
1291 del_qgroup_relation_item(trans, src, dst);
1295 spin_lock(&fs_info->qgroup_lock);
1296 ret = add_relation_rb(fs_info, src, dst);
1298 spin_unlock(&fs_info->qgroup_lock);
1301 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1302 spin_unlock(&fs_info->qgroup_lock);
1304 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1309 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1312 struct btrfs_fs_info *fs_info = trans->fs_info;
1313 struct btrfs_root *quota_root;
1314 struct btrfs_qgroup *parent;
1315 struct btrfs_qgroup *member;
1316 struct btrfs_qgroup_list *list;
1321 tmp = ulist_alloc(GFP_KERNEL);
1325 quota_root = fs_info->quota_root;
1331 member = find_qgroup_rb(fs_info, src);
1332 parent = find_qgroup_rb(fs_info, dst);
1333 if (!member || !parent) {
1338 /* check if such qgroup relation exist firstly */
1339 list_for_each_entry(list, &member->groups, next_group) {
1340 if (list->group == parent)
1346 ret = del_qgroup_relation_item(trans, src, dst);
1347 err = del_qgroup_relation_item(trans, dst, src);
1351 spin_lock(&fs_info->qgroup_lock);
1352 del_relation_rb(fs_info, src, dst);
1353 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1354 spin_unlock(&fs_info->qgroup_lock);
1360 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1363 struct btrfs_fs_info *fs_info = trans->fs_info;
1366 mutex_lock(&fs_info->qgroup_ioctl_lock);
1367 ret = __del_qgroup_relation(trans, src, dst);
1368 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1373 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1375 struct btrfs_fs_info *fs_info = trans->fs_info;
1376 struct btrfs_root *quota_root;
1377 struct btrfs_qgroup *qgroup;
1380 mutex_lock(&fs_info->qgroup_ioctl_lock);
1381 quota_root = fs_info->quota_root;
1386 qgroup = find_qgroup_rb(fs_info, qgroupid);
1392 ret = add_qgroup_item(trans, quota_root, qgroupid);
1396 spin_lock(&fs_info->qgroup_lock);
1397 qgroup = add_qgroup_rb(fs_info, qgroupid);
1398 spin_unlock(&fs_info->qgroup_lock);
1401 ret = PTR_ERR(qgroup);
1403 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1407 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1409 struct btrfs_fs_info *fs_info = trans->fs_info;
1410 struct btrfs_root *quota_root;
1411 struct btrfs_qgroup *qgroup;
1412 struct btrfs_qgroup_list *list;
1415 mutex_lock(&fs_info->qgroup_ioctl_lock);
1416 quota_root = fs_info->quota_root;
1422 qgroup = find_qgroup_rb(fs_info, qgroupid);
1427 /* check if there are no children of this qgroup */
1428 if (!list_empty(&qgroup->members)) {
1433 ret = del_qgroup_item(trans, qgroupid);
1434 if (ret && ret != -ENOENT)
1437 while (!list_empty(&qgroup->groups)) {
1438 list = list_first_entry(&qgroup->groups,
1439 struct btrfs_qgroup_list, next_group);
1440 ret = __del_qgroup_relation(trans, qgroupid,
1441 list->group->qgroupid);
1446 spin_lock(&fs_info->qgroup_lock);
1447 del_qgroup_rb(fs_info, qgroupid);
1448 spin_unlock(&fs_info->qgroup_lock);
1450 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1454 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1455 struct btrfs_qgroup_limit *limit)
1457 struct btrfs_fs_info *fs_info = trans->fs_info;
1458 struct btrfs_root *quota_root;
1459 struct btrfs_qgroup *qgroup;
1461 /* Sometimes we would want to clear the limit on this qgroup.
1462 * To meet this requirement, we treat the -1 as a special value
1463 * which tell kernel to clear the limit on this qgroup.
1465 const u64 CLEAR_VALUE = -1;
1467 mutex_lock(&fs_info->qgroup_ioctl_lock);
1468 quota_root = fs_info->quota_root;
1474 qgroup = find_qgroup_rb(fs_info, qgroupid);
1480 spin_lock(&fs_info->qgroup_lock);
1481 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1482 if (limit->max_rfer == CLEAR_VALUE) {
1483 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1484 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1485 qgroup->max_rfer = 0;
1487 qgroup->max_rfer = limit->max_rfer;
1490 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1491 if (limit->max_excl == CLEAR_VALUE) {
1492 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1493 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1494 qgroup->max_excl = 0;
1496 qgroup->max_excl = limit->max_excl;
1499 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1500 if (limit->rsv_rfer == CLEAR_VALUE) {
1501 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1502 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1503 qgroup->rsv_rfer = 0;
1505 qgroup->rsv_rfer = limit->rsv_rfer;
1508 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1509 if (limit->rsv_excl == CLEAR_VALUE) {
1510 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1511 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1512 qgroup->rsv_excl = 0;
1514 qgroup->rsv_excl = limit->rsv_excl;
1517 qgroup->lim_flags |= limit->flags;
1519 spin_unlock(&fs_info->qgroup_lock);
1521 ret = update_qgroup_limit_item(trans, qgroup);
1523 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1524 btrfs_info(fs_info, "unable to update quota limit for %llu",
1529 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1533 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1534 struct btrfs_delayed_ref_root *delayed_refs,
1535 struct btrfs_qgroup_extent_record *record)
1537 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1538 struct rb_node *parent_node = NULL;
1539 struct btrfs_qgroup_extent_record *entry;
1540 u64 bytenr = record->bytenr;
1542 lockdep_assert_held(&delayed_refs->lock);
1543 trace_btrfs_qgroup_trace_extent(fs_info, record);
1547 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1549 if (bytenr < entry->bytenr)
1551 else if (bytenr > entry->bytenr)
1552 p = &(*p)->rb_right;
1557 rb_link_node(&record->node, parent_node, p);
1558 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1562 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1563 struct btrfs_qgroup_extent_record *qrecord)
1565 struct ulist *old_root;
1566 u64 bytenr = qrecord->bytenr;
1569 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1571 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1573 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1579 * Here we don't need to get the lock of
1580 * trans->transaction->delayed_refs, since inserted qrecord won't
1581 * be deleted, only qrecord->node may be modified (new qrecord insert)
1583 * So modifying qrecord->old_roots is safe here
1585 qrecord->old_roots = old_root;
1589 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1590 u64 num_bytes, gfp_t gfp_flag)
1592 struct btrfs_fs_info *fs_info = trans->fs_info;
1593 struct btrfs_qgroup_extent_record *record;
1594 struct btrfs_delayed_ref_root *delayed_refs;
1597 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1598 || bytenr == 0 || num_bytes == 0)
1600 record = kmalloc(sizeof(*record), gfp_flag);
1604 delayed_refs = &trans->transaction->delayed_refs;
1605 record->bytenr = bytenr;
1606 record->num_bytes = num_bytes;
1607 record->old_roots = NULL;
1609 spin_lock(&delayed_refs->lock);
1610 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1611 spin_unlock(&delayed_refs->lock);
1616 return btrfs_qgroup_trace_extent_post(fs_info, record);
1619 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1620 struct extent_buffer *eb)
1622 struct btrfs_fs_info *fs_info = trans->fs_info;
1623 int nr = btrfs_header_nritems(eb);
1624 int i, extent_type, ret;
1625 struct btrfs_key key;
1626 struct btrfs_file_extent_item *fi;
1627 u64 bytenr, num_bytes;
1629 /* We can be called directly from walk_up_proc() */
1630 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1633 for (i = 0; i < nr; i++) {
1634 btrfs_item_key_to_cpu(eb, &key, i);
1636 if (key.type != BTRFS_EXTENT_DATA_KEY)
1639 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1640 /* filter out non qgroup-accountable extents */
1641 extent_type = btrfs_file_extent_type(eb, fi);
1643 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1646 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1650 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1652 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1662 * Walk up the tree from the bottom, freeing leaves and any interior
1663 * nodes which have had all slots visited. If a node (leaf or
1664 * interior) is freed, the node above it will have it's slot
1665 * incremented. The root node will never be freed.
1667 * At the end of this function, we should have a path which has all
1668 * slots incremented to the next position for a search. If we need to
1669 * read a new node it will be NULL and the node above it will have the
1670 * correct slot selected for a later read.
1672 * If we increment the root nodes slot counter past the number of
1673 * elements, 1 is returned to signal completion of the search.
1675 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1679 struct extent_buffer *eb;
1681 if (root_level == 0)
1684 while (level <= root_level) {
1685 eb = path->nodes[level];
1686 nr = btrfs_header_nritems(eb);
1687 path->slots[level]++;
1688 slot = path->slots[level];
1689 if (slot >= nr || level == 0) {
1691 * Don't free the root - we will detect this
1692 * condition after our loop and return a
1693 * positive value for caller to stop walking the tree.
1695 if (level != root_level) {
1696 btrfs_tree_unlock_rw(eb, path->locks[level]);
1697 path->locks[level] = 0;
1699 free_extent_buffer(eb);
1700 path->nodes[level] = NULL;
1701 path->slots[level] = 0;
1705 * We have a valid slot to walk back down
1706 * from. Stop here so caller can process these
1715 eb = path->nodes[root_level];
1716 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1722 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1723 struct extent_buffer *root_eb,
1724 u64 root_gen, int root_level)
1726 struct btrfs_fs_info *fs_info = trans->fs_info;
1729 struct extent_buffer *eb = root_eb;
1730 struct btrfs_path *path = NULL;
1732 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1733 BUG_ON(root_eb == NULL);
1735 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1738 if (!extent_buffer_uptodate(root_eb)) {
1739 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
1744 if (root_level == 0) {
1745 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
1749 path = btrfs_alloc_path();
1754 * Walk down the tree. Missing extent blocks are filled in as
1755 * we go. Metadata is accounted every time we read a new
1758 * When we reach a leaf, we account for file extent items in it,
1759 * walk back up the tree (adjusting slot pointers as we go)
1760 * and restart the search process.
1762 extent_buffer_get(root_eb); /* For path */
1763 path->nodes[root_level] = root_eb;
1764 path->slots[root_level] = 0;
1765 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1768 while (level >= 0) {
1769 if (path->nodes[level] == NULL) {
1770 struct btrfs_key first_key;
1776 * We need to get child blockptr/gen from parent before
1779 eb = path->nodes[level + 1];
1780 parent_slot = path->slots[level + 1];
1781 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1782 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1783 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1785 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1790 } else if (!extent_buffer_uptodate(eb)) {
1791 free_extent_buffer(eb);
1796 path->nodes[level] = eb;
1797 path->slots[level] = 0;
1799 btrfs_tree_read_lock(eb);
1800 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1801 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1803 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
1811 ret = btrfs_qgroup_trace_leaf_items(trans,
1812 path->nodes[level]);
1816 /* Nonzero return here means we completed our search */
1817 ret = adjust_slots_upwards(path, root_level);
1821 /* Restart search with new slots */
1830 btrfs_free_path(path);
1835 #define UPDATE_NEW 0
1836 #define UPDATE_OLD 1
1838 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1840 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1841 struct ulist *roots, struct ulist *tmp,
1842 struct ulist *qgroups, u64 seq, int update_old)
1844 struct ulist_node *unode;
1845 struct ulist_iterator uiter;
1846 struct ulist_node *tmp_unode;
1847 struct ulist_iterator tmp_uiter;
1848 struct btrfs_qgroup *qg;
1853 ULIST_ITER_INIT(&uiter);
1854 while ((unode = ulist_next(roots, &uiter))) {
1855 qg = find_qgroup_rb(fs_info, unode->val);
1860 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1864 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1867 ULIST_ITER_INIT(&tmp_uiter);
1868 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1869 struct btrfs_qgroup_list *glist;
1871 qg = unode_aux_to_qgroup(tmp_unode);
1873 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1875 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1876 list_for_each_entry(glist, &qg->groups, next_group) {
1877 ret = ulist_add(qgroups, glist->group->qgroupid,
1878 qgroup_to_aux(glist->group),
1882 ret = ulist_add(tmp, glist->group->qgroupid,
1883 qgroup_to_aux(glist->group),
1894 * Update qgroup rfer/excl counters.
1895 * Rfer update is easy, codes can explain themselves.
1897 * Excl update is tricky, the update is split into 2 part.
1898 * Part 1: Possible exclusive <-> sharing detect:
1900 * -------------------------------------
1902 * -------------------------------------
1904 * -------------------------------------
1907 * A: cur_old_roots < nr_old_roots (not exclusive before)
1908 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1909 * B: cur_new_roots < nr_new_roots (not exclusive now)
1910 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
1913 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1914 * *: Definitely not changed. **: Possible unchanged.
1916 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1918 * To make the logic clear, we first use condition A and B to split
1919 * combination into 4 results.
1921 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1922 * only on variant maybe 0.
1924 * Lastly, check result **, since there are 2 variants maybe 0, split them
1926 * But this time we don't need to consider other things, the codes and logic
1927 * is easy to understand now.
1929 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1930 struct ulist *qgroups,
1933 u64 num_bytes, u64 seq)
1935 struct ulist_node *unode;
1936 struct ulist_iterator uiter;
1937 struct btrfs_qgroup *qg;
1938 u64 cur_new_count, cur_old_count;
1940 ULIST_ITER_INIT(&uiter);
1941 while ((unode = ulist_next(qgroups, &uiter))) {
1944 qg = unode_aux_to_qgroup(unode);
1945 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1946 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1948 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
1951 /* Rfer update part */
1952 if (cur_old_count == 0 && cur_new_count > 0) {
1953 qg->rfer += num_bytes;
1954 qg->rfer_cmpr += num_bytes;
1957 if (cur_old_count > 0 && cur_new_count == 0) {
1958 qg->rfer -= num_bytes;
1959 qg->rfer_cmpr -= num_bytes;
1963 /* Excl update part */
1964 /* Exclusive/none -> shared case */
1965 if (cur_old_count == nr_old_roots &&
1966 cur_new_count < nr_new_roots) {
1967 /* Exclusive -> shared */
1968 if (cur_old_count != 0) {
1969 qg->excl -= num_bytes;
1970 qg->excl_cmpr -= num_bytes;
1975 /* Shared -> exclusive/none case */
1976 if (cur_old_count < nr_old_roots &&
1977 cur_new_count == nr_new_roots) {
1978 /* Shared->exclusive */
1979 if (cur_new_count != 0) {
1980 qg->excl += num_bytes;
1981 qg->excl_cmpr += num_bytes;
1986 /* Exclusive/none -> exclusive/none case */
1987 if (cur_old_count == nr_old_roots &&
1988 cur_new_count == nr_new_roots) {
1989 if (cur_old_count == 0) {
1990 /* None -> exclusive/none */
1992 if (cur_new_count != 0) {
1993 /* None -> exclusive */
1994 qg->excl += num_bytes;
1995 qg->excl_cmpr += num_bytes;
1998 /* None -> none, nothing changed */
2000 /* Exclusive -> exclusive/none */
2002 if (cur_new_count == 0) {
2003 /* Exclusive -> none */
2004 qg->excl -= num_bytes;
2005 qg->excl_cmpr -= num_bytes;
2008 /* Exclusive -> exclusive, nothing changed */
2013 qgroup_dirty(fs_info, qg);
2019 * Check if the @roots potentially is a list of fs tree roots
2021 * Return 0 for definitely not a fs/subvol tree roots ulist
2022 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2025 static int maybe_fs_roots(struct ulist *roots)
2027 struct ulist_node *unode;
2028 struct ulist_iterator uiter;
2030 /* Empty one, still possible for fs roots */
2031 if (!roots || roots->nnodes == 0)
2034 ULIST_ITER_INIT(&uiter);
2035 unode = ulist_next(roots, &uiter);
2040 * If it contains fs tree roots, then it must belong to fs/subvol
2042 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2044 return is_fstree(unode->val);
2047 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2048 u64 num_bytes, struct ulist *old_roots,
2049 struct ulist *new_roots)
2051 struct btrfs_fs_info *fs_info = trans->fs_info;
2052 struct ulist *qgroups = NULL;
2053 struct ulist *tmp = NULL;
2055 u64 nr_new_roots = 0;
2056 u64 nr_old_roots = 0;
2060 * If quotas get disabled meanwhile, the resouces need to be freed and
2061 * we can't just exit here.
2063 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2067 if (!maybe_fs_roots(new_roots))
2069 nr_new_roots = new_roots->nnodes;
2072 if (!maybe_fs_roots(old_roots))
2074 nr_old_roots = old_roots->nnodes;
2077 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2078 if (nr_old_roots == 0 && nr_new_roots == 0)
2081 BUG_ON(!fs_info->quota_root);
2083 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2084 num_bytes, nr_old_roots, nr_new_roots);
2086 qgroups = ulist_alloc(GFP_NOFS);
2091 tmp = ulist_alloc(GFP_NOFS);
2097 mutex_lock(&fs_info->qgroup_rescan_lock);
2098 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2099 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2100 mutex_unlock(&fs_info->qgroup_rescan_lock);
2105 mutex_unlock(&fs_info->qgroup_rescan_lock);
2107 spin_lock(&fs_info->qgroup_lock);
2108 seq = fs_info->qgroup_seq;
2110 /* Update old refcnts using old_roots */
2111 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2116 /* Update new refcnts using new_roots */
2117 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2122 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2126 * Bump qgroup_seq to avoid seq overlap
2128 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2130 spin_unlock(&fs_info->qgroup_lock);
2133 ulist_free(qgroups);
2134 ulist_free(old_roots);
2135 ulist_free(new_roots);
2139 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2141 struct btrfs_fs_info *fs_info = trans->fs_info;
2142 struct btrfs_qgroup_extent_record *record;
2143 struct btrfs_delayed_ref_root *delayed_refs;
2144 struct ulist *new_roots = NULL;
2145 struct rb_node *node;
2149 delayed_refs = &trans->transaction->delayed_refs;
2150 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2151 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2152 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2155 trace_btrfs_qgroup_account_extents(fs_info, record);
2159 * Old roots should be searched when inserting qgroup
2162 if (WARN_ON(!record->old_roots)) {
2163 /* Search commit root to find old_roots */
2164 ret = btrfs_find_all_roots(NULL, fs_info,
2166 &record->old_roots, false);
2172 * Use SEQ_LAST as time_seq to do special search, which
2173 * doesn't lock tree or delayed_refs and search current
2174 * root. It's safe inside commit_transaction().
2176 ret = btrfs_find_all_roots(trans, fs_info,
2177 record->bytenr, SEQ_LAST, &new_roots, false);
2180 if (qgroup_to_skip) {
2181 ulist_del(new_roots, qgroup_to_skip, 0);
2182 ulist_del(record->old_roots, qgroup_to_skip,
2185 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2189 record->old_roots = NULL;
2193 ulist_free(record->old_roots);
2194 ulist_free(new_roots);
2196 rb_erase(node, &delayed_refs->dirty_extent_root);
2204 * called from commit_transaction. Writes all changed qgroups to disk.
2206 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2208 struct btrfs_fs_info *fs_info = trans->fs_info;
2209 struct btrfs_root *quota_root = fs_info->quota_root;
2215 spin_lock(&fs_info->qgroup_lock);
2216 while (!list_empty(&fs_info->dirty_qgroups)) {
2217 struct btrfs_qgroup *qgroup;
2218 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2219 struct btrfs_qgroup, dirty);
2220 list_del_init(&qgroup->dirty);
2221 spin_unlock(&fs_info->qgroup_lock);
2222 ret = update_qgroup_info_item(trans, qgroup);
2224 fs_info->qgroup_flags |=
2225 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2226 ret = update_qgroup_limit_item(trans, qgroup);
2228 fs_info->qgroup_flags |=
2229 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2230 spin_lock(&fs_info->qgroup_lock);
2232 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2233 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2235 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2236 spin_unlock(&fs_info->qgroup_lock);
2238 ret = update_qgroup_status_item(trans);
2240 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2246 * Copy the accounting information between qgroups. This is necessary
2247 * when a snapshot or a subvolume is created. Throwing an error will
2248 * cause a transaction abort so we take extra care here to only error
2249 * when a readonly fs is a reasonable outcome.
2251 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2252 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2257 bool committing = false;
2258 struct btrfs_fs_info *fs_info = trans->fs_info;
2259 struct btrfs_root *quota_root;
2260 struct btrfs_qgroup *srcgroup;
2261 struct btrfs_qgroup *dstgroup;
2262 bool need_rescan = false;
2267 * There are only two callers of this function.
2269 * One in create_subvol() in the ioctl context, which needs to hold
2270 * the qgroup_ioctl_lock.
2272 * The other one in create_pending_snapshot() where no other qgroup
2273 * code can modify the fs as they all need to either start a new trans
2274 * or hold a trans handler, thus we don't need to hold
2275 * qgroup_ioctl_lock.
2276 * This would avoid long and complex lock chain and make lockdep happy.
2278 spin_lock(&fs_info->trans_lock);
2279 if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2281 spin_unlock(&fs_info->trans_lock);
2284 mutex_lock(&fs_info->qgroup_ioctl_lock);
2285 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2288 quota_root = fs_info->quota_root;
2295 i_qgroups = (u64 *)(inherit + 1);
2296 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2297 2 * inherit->num_excl_copies;
2298 for (i = 0; i < nums; ++i) {
2299 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2302 * Zero out invalid groups so we can ignore
2306 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2314 * create a tracking group for the subvol itself
2316 ret = add_qgroup_item(trans, quota_root, objectid);
2321 * add qgroup to all inherited groups
2324 i_qgroups = (u64 *)(inherit + 1);
2325 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2326 if (*i_qgroups == 0)
2328 ret = add_qgroup_relation_item(trans, objectid,
2330 if (ret && ret != -EEXIST)
2332 ret = add_qgroup_relation_item(trans, *i_qgroups,
2334 if (ret && ret != -EEXIST)
2341 spin_lock(&fs_info->qgroup_lock);
2343 dstgroup = add_qgroup_rb(fs_info, objectid);
2344 if (IS_ERR(dstgroup)) {
2345 ret = PTR_ERR(dstgroup);
2349 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2350 dstgroup->lim_flags = inherit->lim.flags;
2351 dstgroup->max_rfer = inherit->lim.max_rfer;
2352 dstgroup->max_excl = inherit->lim.max_excl;
2353 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2354 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2356 ret = update_qgroup_limit_item(trans, dstgroup);
2358 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2360 "unable to update quota limit for %llu",
2361 dstgroup->qgroupid);
2367 srcgroup = find_qgroup_rb(fs_info, srcid);
2372 * We call inherit after we clone the root in order to make sure
2373 * our counts don't go crazy, so at this point the only
2374 * difference between the two roots should be the root node.
2376 level_size = fs_info->nodesize;
2377 dstgroup->rfer = srcgroup->rfer;
2378 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2379 dstgroup->excl = level_size;
2380 dstgroup->excl_cmpr = level_size;
2381 srcgroup->excl = level_size;
2382 srcgroup->excl_cmpr = level_size;
2384 /* inherit the limit info */
2385 dstgroup->lim_flags = srcgroup->lim_flags;
2386 dstgroup->max_rfer = srcgroup->max_rfer;
2387 dstgroup->max_excl = srcgroup->max_excl;
2388 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2389 dstgroup->rsv_excl = srcgroup->rsv_excl;
2391 qgroup_dirty(fs_info, dstgroup);
2392 qgroup_dirty(fs_info, srcgroup);
2398 i_qgroups = (u64 *)(inherit + 1);
2399 for (i = 0; i < inherit->num_qgroups; ++i) {
2401 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2408 * If we're doing a snapshot, and adding the snapshot to a new
2409 * qgroup, the numbers are guaranteed to be incorrect.
2415 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2416 struct btrfs_qgroup *src;
2417 struct btrfs_qgroup *dst;
2419 if (!i_qgroups[0] || !i_qgroups[1])
2422 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2423 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2430 dst->rfer = src->rfer - level_size;
2431 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2433 /* Manually tweaking numbers certainly needs a rescan */
2436 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2437 struct btrfs_qgroup *src;
2438 struct btrfs_qgroup *dst;
2440 if (!i_qgroups[0] || !i_qgroups[1])
2443 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2444 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2451 dst->excl = src->excl + level_size;
2452 dst->excl_cmpr = src->excl_cmpr + level_size;
2457 spin_unlock(&fs_info->qgroup_lock);
2460 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2462 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2467 * Two limits to commit transaction in advance.
2469 * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
2470 * For SIZE, it will be in byte unit as threshold.
2472 #define QGROUP_FREE_RATIO 32
2473 #define QGROUP_FREE_SIZE SZ_32M
2474 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2475 const struct btrfs_qgroup *qg, u64 num_bytes)
2480 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2481 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2484 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2485 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2489 * Even if we passed the check, it's better to check if reservation
2490 * for meta_pertrans is pushing us near limit.
2491 * If there is too much pertrans reservation or it's near the limit,
2492 * let's try commit transaction to free some, using transaction_kthread
2494 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2495 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2496 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2497 free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2498 threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2501 free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2502 threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2507 * Use transaction_kthread to commit transaction, so we no
2508 * longer need to bother nested transaction nor lock context.
2510 if (free < threshold)
2511 btrfs_commit_transaction_locksafe(fs_info);
2517 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2518 enum btrfs_qgroup_rsv_type type)
2520 struct btrfs_root *quota_root;
2521 struct btrfs_qgroup *qgroup;
2522 struct btrfs_fs_info *fs_info = root->fs_info;
2523 u64 ref_root = root->root_key.objectid;
2525 struct ulist_node *unode;
2526 struct ulist_iterator uiter;
2528 if (!is_fstree(ref_root))
2534 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2535 capable(CAP_SYS_RESOURCE))
2538 spin_lock(&fs_info->qgroup_lock);
2539 quota_root = fs_info->quota_root;
2543 qgroup = find_qgroup_rb(fs_info, ref_root);
2548 * in a first step, we check all affected qgroups if any limits would
2551 ulist_reinit(fs_info->qgroup_ulist);
2552 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2553 qgroup_to_aux(qgroup), GFP_ATOMIC);
2556 ULIST_ITER_INIT(&uiter);
2557 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2558 struct btrfs_qgroup *qg;
2559 struct btrfs_qgroup_list *glist;
2561 qg = unode_aux_to_qgroup(unode);
2563 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2568 list_for_each_entry(glist, &qg->groups, next_group) {
2569 ret = ulist_add(fs_info->qgroup_ulist,
2570 glist->group->qgroupid,
2571 qgroup_to_aux(glist->group), GFP_ATOMIC);
2578 * no limits exceeded, now record the reservation into all qgroups
2580 ULIST_ITER_INIT(&uiter);
2581 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2582 struct btrfs_qgroup *qg;
2584 qg = unode_aux_to_qgroup(unode);
2586 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2587 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2591 spin_unlock(&fs_info->qgroup_lock);
2596 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2599 * Will handle all higher level qgroup too.
2601 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2602 * This special case is only used for META_PERTRANS type.
2604 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2605 u64 ref_root, u64 num_bytes,
2606 enum btrfs_qgroup_rsv_type type)
2608 struct btrfs_root *quota_root;
2609 struct btrfs_qgroup *qgroup;
2610 struct ulist_node *unode;
2611 struct ulist_iterator uiter;
2614 if (!is_fstree(ref_root))
2620 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2621 WARN(1, "%s: Invalid type to free", __func__);
2624 spin_lock(&fs_info->qgroup_lock);
2626 quota_root = fs_info->quota_root;
2630 qgroup = find_qgroup_rb(fs_info, ref_root);
2634 if (num_bytes == (u64)-1)
2636 * We're freeing all pertrans rsv, get reserved value from
2637 * level 0 qgroup as real num_bytes to free.
2639 num_bytes = qgroup->rsv.values[type];
2641 ulist_reinit(fs_info->qgroup_ulist);
2642 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2643 qgroup_to_aux(qgroup), GFP_ATOMIC);
2646 ULIST_ITER_INIT(&uiter);
2647 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2648 struct btrfs_qgroup *qg;
2649 struct btrfs_qgroup_list *glist;
2651 qg = unode_aux_to_qgroup(unode);
2653 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2654 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2656 list_for_each_entry(glist, &qg->groups, next_group) {
2657 ret = ulist_add(fs_info->qgroup_ulist,
2658 glist->group->qgroupid,
2659 qgroup_to_aux(glist->group), GFP_ATOMIC);
2666 spin_unlock(&fs_info->qgroup_lock);
2670 * Check if the leaf is the last leaf. Which means all node pointers
2671 * are at their last position.
2673 static bool is_last_leaf(struct btrfs_path *path)
2677 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2678 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
2685 * returns < 0 on error, 0 when more leafs are to be scanned.
2686 * returns 1 when done.
2688 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
2689 struct btrfs_path *path)
2691 struct btrfs_fs_info *fs_info = trans->fs_info;
2692 struct btrfs_key found;
2693 struct extent_buffer *scratch_leaf = NULL;
2694 struct ulist *roots = NULL;
2700 mutex_lock(&fs_info->qgroup_rescan_lock);
2701 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2702 &fs_info->qgroup_rescan_progress,
2705 btrfs_debug(fs_info,
2706 "current progress key (%llu %u %llu), search_slot ret %d",
2707 fs_info->qgroup_rescan_progress.objectid,
2708 fs_info->qgroup_rescan_progress.type,
2709 fs_info->qgroup_rescan_progress.offset, ret);
2713 * The rescan is about to end, we will not be scanning any
2714 * further blocks. We cannot unset the RESCAN flag here, because
2715 * we want to commit the transaction if everything went well.
2716 * To make the live accounting work in this phase, we set our
2717 * scan progress pointer such that every real extent objectid
2720 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2721 btrfs_release_path(path);
2722 mutex_unlock(&fs_info->qgroup_rescan_lock);
2725 done = is_last_leaf(path);
2727 btrfs_item_key_to_cpu(path->nodes[0], &found,
2728 btrfs_header_nritems(path->nodes[0]) - 1);
2729 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2731 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2732 if (!scratch_leaf) {
2734 mutex_unlock(&fs_info->qgroup_rescan_lock);
2737 extent_buffer_get(scratch_leaf);
2738 btrfs_tree_read_lock(scratch_leaf);
2739 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2740 slot = path->slots[0];
2741 btrfs_release_path(path);
2742 mutex_unlock(&fs_info->qgroup_rescan_lock);
2744 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2745 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2746 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2747 found.type != BTRFS_METADATA_ITEM_KEY)
2749 if (found.type == BTRFS_METADATA_ITEM_KEY)
2750 num_bytes = fs_info->nodesize;
2752 num_bytes = found.offset;
2754 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2758 /* For rescan, just pass old_roots as NULL */
2759 ret = btrfs_qgroup_account_extent(trans, found.objectid,
2760 num_bytes, NULL, roots);
2766 btrfs_tree_read_unlock_blocking(scratch_leaf);
2767 free_extent_buffer(scratch_leaf);
2772 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2777 static bool rescan_should_stop(struct btrfs_fs_info *fs_info)
2779 return btrfs_fs_closing(fs_info) ||
2780 test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state);
2783 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2785 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2786 qgroup_rescan_work);
2787 struct btrfs_path *path;
2788 struct btrfs_trans_handle *trans = NULL;
2791 bool stopped = false;
2793 path = btrfs_alloc_path();
2797 * Rescan should only search for commit root, and any later difference
2798 * should be recorded by qgroup
2800 path->search_commit_root = 1;
2801 path->skip_locking = 1;
2804 while (!err && !(stopped = rescan_should_stop(fs_info))) {
2805 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2806 if (IS_ERR(trans)) {
2807 err = PTR_ERR(trans);
2810 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2813 err = qgroup_rescan_leaf(trans, path);
2816 btrfs_commit_transaction(trans);
2818 btrfs_end_transaction(trans);
2822 btrfs_free_path(path);
2824 mutex_lock(&fs_info->qgroup_rescan_lock);
2826 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2827 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2828 } else if (err < 0) {
2829 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2831 mutex_unlock(&fs_info->qgroup_rescan_lock);
2834 * only update status, since the previous part has already updated the
2837 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2838 if (IS_ERR(trans)) {
2839 err = PTR_ERR(trans);
2842 "fail to start transaction for status update: %d",
2846 mutex_lock(&fs_info->qgroup_rescan_lock);
2848 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2850 ret = update_qgroup_status_item(trans);
2853 btrfs_err(fs_info, "fail to update qgroup status: %d",
2857 fs_info->qgroup_rescan_running = false;
2858 complete_all(&fs_info->qgroup_rescan_completion);
2859 mutex_unlock(&fs_info->qgroup_rescan_lock);
2864 btrfs_end_transaction(trans);
2867 btrfs_info(fs_info, "qgroup scan paused");
2868 } else if (err >= 0) {
2869 btrfs_info(fs_info, "qgroup scan completed%s",
2870 err > 0 ? " (inconsistency flag cleared)" : "");
2872 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2877 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2878 * memory required for the rescan context.
2881 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2887 /* we're resuming qgroup rescan at mount time */
2888 if (!(fs_info->qgroup_flags &
2889 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
2891 "qgroup rescan init failed, qgroup rescan is not queued");
2893 } else if (!(fs_info->qgroup_flags &
2894 BTRFS_QGROUP_STATUS_FLAG_ON)) {
2896 "qgroup rescan init failed, qgroup is not enabled");
2904 mutex_lock(&fs_info->qgroup_rescan_lock);
2905 spin_lock(&fs_info->qgroup_lock);
2908 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2910 "qgroup rescan is already in progress");
2912 } else if (!(fs_info->qgroup_flags &
2913 BTRFS_QGROUP_STATUS_FLAG_ON)) {
2915 "qgroup rescan init failed, qgroup is not enabled");
2920 spin_unlock(&fs_info->qgroup_lock);
2921 mutex_unlock(&fs_info->qgroup_rescan_lock);
2924 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2927 memset(&fs_info->qgroup_rescan_progress, 0,
2928 sizeof(fs_info->qgroup_rescan_progress));
2929 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2930 init_completion(&fs_info->qgroup_rescan_completion);
2932 spin_unlock(&fs_info->qgroup_lock);
2933 mutex_unlock(&fs_info->qgroup_rescan_lock);
2935 memset(&fs_info->qgroup_rescan_work, 0,
2936 sizeof(fs_info->qgroup_rescan_work));
2937 btrfs_init_work(&fs_info->qgroup_rescan_work,
2938 btrfs_qgroup_rescan_helper,
2939 btrfs_qgroup_rescan_worker, NULL, NULL);
2944 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2947 struct btrfs_qgroup *qgroup;
2949 spin_lock(&fs_info->qgroup_lock);
2950 /* clear all current qgroup tracking information */
2951 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2952 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2954 qgroup->rfer_cmpr = 0;
2956 qgroup->excl_cmpr = 0;
2957 qgroup_dirty(fs_info, qgroup);
2959 spin_unlock(&fs_info->qgroup_lock);
2963 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2966 struct btrfs_trans_handle *trans;
2968 ret = qgroup_rescan_init(fs_info, 0, 1);
2973 * We have set the rescan_progress to 0, which means no more
2974 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2975 * However, btrfs_qgroup_account_ref may be right after its call
2976 * to btrfs_find_all_roots, in which case it would still do the
2978 * To solve this, we're committing the transaction, which will
2979 * ensure we run all delayed refs and only after that, we are
2980 * going to clear all tracking information for a clean start.
2983 trans = btrfs_join_transaction(fs_info->fs_root);
2984 if (IS_ERR(trans)) {
2985 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2986 return PTR_ERR(trans);
2988 ret = btrfs_commit_transaction(trans);
2990 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2994 qgroup_rescan_zero_tracking(fs_info);
2996 mutex_lock(&fs_info->qgroup_rescan_lock);
2997 fs_info->qgroup_rescan_running = true;
2998 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2999 &fs_info->qgroup_rescan_work);
3000 mutex_unlock(&fs_info->qgroup_rescan_lock);
3005 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3011 mutex_lock(&fs_info->qgroup_rescan_lock);
3012 spin_lock(&fs_info->qgroup_lock);
3013 running = fs_info->qgroup_rescan_running;
3014 spin_unlock(&fs_info->qgroup_lock);
3015 mutex_unlock(&fs_info->qgroup_rescan_lock);
3021 ret = wait_for_completion_interruptible(
3022 &fs_info->qgroup_rescan_completion);
3024 wait_for_completion(&fs_info->qgroup_rescan_completion);
3030 * this is only called from open_ctree where we're still single threaded, thus
3031 * locking is omitted here.
3034 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3036 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3037 mutex_lock(&fs_info->qgroup_rescan_lock);
3038 fs_info->qgroup_rescan_running = true;
3039 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3040 &fs_info->qgroup_rescan_work);
3041 mutex_unlock(&fs_info->qgroup_rescan_lock);
3046 * Reserve qgroup space for range [start, start + len).
3048 * This function will either reserve space from related qgroups or doing
3049 * nothing if the range is already reserved.
3051 * Return 0 for successful reserve
3052 * Return <0 for error (including -EQUOT)
3054 * NOTE: this function may sleep for memory allocation.
3055 * if btrfs_qgroup_reserve_data() is called multiple times with
3056 * same @reserved, caller must ensure when error happens it's OK
3057 * to free *ALL* reserved space.
3059 int btrfs_qgroup_reserve_data(struct inode *inode,
3060 struct extent_changeset **reserved_ret, u64 start,
3063 struct btrfs_root *root = BTRFS_I(inode)->root;
3064 struct ulist_node *unode;
3065 struct ulist_iterator uiter;
3066 struct extent_changeset *reserved;
3071 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3072 !is_fstree(root->objectid) || len == 0)
3075 /* @reserved parameter is mandatory for qgroup */
3076 if (WARN_ON(!reserved_ret))
3078 if (!*reserved_ret) {
3079 *reserved_ret = extent_changeset_alloc();
3083 reserved = *reserved_ret;
3084 /* Record already reserved space */
3085 orig_reserved = reserved->bytes_changed;
3086 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3087 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3089 /* Newly reserved space */
3090 to_reserve = reserved->bytes_changed - orig_reserved;
3091 trace_btrfs_qgroup_reserve_data(inode, start, len,
3092 to_reserve, QGROUP_RESERVE);
3095 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3102 /* cleanup *ALL* already reserved ranges */
3103 ULIST_ITER_INIT(&uiter);
3104 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3105 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3106 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3107 /* Also free data bytes of already reserved one */
3108 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid,
3109 orig_reserved, BTRFS_QGROUP_RSV_DATA);
3110 extent_changeset_release(reserved);
3114 /* Free ranges specified by @reserved, normally in error path */
3115 static int qgroup_free_reserved_data(struct inode *inode,
3116 struct extent_changeset *reserved, u64 start, u64 len)
3118 struct btrfs_root *root = BTRFS_I(inode)->root;
3119 struct ulist_node *unode;
3120 struct ulist_iterator uiter;
3121 struct extent_changeset changeset;
3125 extent_changeset_init(&changeset);
3126 len = round_up(start + len, root->fs_info->sectorsize);
3127 start = round_down(start, root->fs_info->sectorsize);
3129 ULIST_ITER_INIT(&uiter);
3130 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3131 u64 range_start = unode->val;
3132 /* unode->aux is the inclusive end */
3133 u64 range_len = unode->aux - range_start + 1;
3137 extent_changeset_release(&changeset);
3139 /* Only free range in range [start, start + len) */
3140 if (range_start >= start + len ||
3141 range_start + range_len <= start)
3143 free_start = max(range_start, start);
3144 free_len = min(start + len, range_start + range_len) -
3147 * TODO: To also modify reserved->ranges_reserved to reflect
3150 * However as long as we free qgroup reserved according to
3151 * EXTENT_QGROUP_RESERVED, we won't double free.
3152 * So not need to rush.
3154 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree,
3155 free_start, free_start + free_len - 1,
3156 EXTENT_QGROUP_RESERVED, &changeset);
3159 freed += changeset.bytes_changed;
3161 btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed,
3162 BTRFS_QGROUP_RSV_DATA);
3165 extent_changeset_release(&changeset);
3169 static int __btrfs_qgroup_release_data(struct inode *inode,
3170 struct extent_changeset *reserved, u64 start, u64 len,
3173 struct extent_changeset changeset;
3174 int trace_op = QGROUP_RELEASE;
3177 if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3178 &BTRFS_I(inode)->root->fs_info->flags))
3181 /* In release case, we shouldn't have @reserved */
3182 WARN_ON(!free && reserved);
3183 if (free && reserved)
3184 return qgroup_free_reserved_data(inode, reserved, start, len);
3185 extent_changeset_init(&changeset);
3186 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3187 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3192 trace_op = QGROUP_FREE;
3193 trace_btrfs_qgroup_release_data(inode, start, len,
3194 changeset.bytes_changed, trace_op);
3196 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3197 BTRFS_I(inode)->root->objectid,
3198 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3199 ret = changeset.bytes_changed;
3201 extent_changeset_release(&changeset);
3206 * Free a reserved space range from io_tree and related qgroups
3208 * Should be called when a range of pages get invalidated before reaching disk.
3209 * Or for error cleanup case.
3210 * if @reserved is given, only reserved range in [@start, @start + @len) will
3213 * For data written to disk, use btrfs_qgroup_release_data().
3215 * NOTE: This function may sleep for memory allocation.
3217 int btrfs_qgroup_free_data(struct inode *inode,
3218 struct extent_changeset *reserved, u64 start, u64 len)
3220 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3224 * Release a reserved space range from io_tree only.
3226 * Should be called when a range of pages get written to disk and corresponding
3227 * FILE_EXTENT is inserted into corresponding root.
3229 * Since new qgroup accounting framework will only update qgroup numbers at
3230 * commit_transaction() time, its reserved space shouldn't be freed from
3233 * But we should release the range from io_tree, to allow further write to be
3236 * NOTE: This function may sleep for memory allocation.
3238 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3240 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3243 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3244 enum btrfs_qgroup_rsv_type type)
3246 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3247 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3252 spin_lock(&root->qgroup_meta_rsv_lock);
3253 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3254 root->qgroup_meta_rsv_prealloc += num_bytes;
3256 root->qgroup_meta_rsv_pertrans += num_bytes;
3257 spin_unlock(&root->qgroup_meta_rsv_lock);
3260 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3261 enum btrfs_qgroup_rsv_type type)
3263 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3264 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3269 spin_lock(&root->qgroup_meta_rsv_lock);
3270 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3271 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3273 root->qgroup_meta_rsv_prealloc -= num_bytes;
3275 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3277 root->qgroup_meta_rsv_pertrans -= num_bytes;
3279 spin_unlock(&root->qgroup_meta_rsv_lock);
3283 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3284 enum btrfs_qgroup_rsv_type type, bool enforce)
3286 struct btrfs_fs_info *fs_info = root->fs_info;
3289 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3290 !is_fstree(root->objectid) || num_bytes == 0)
3293 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3294 trace_qgroup_meta_reserve(root, (s64)num_bytes, type);
3295 ret = qgroup_reserve(root, num_bytes, enforce, type);
3299 * Record what we have reserved into root.
3301 * To avoid quota disabled->enabled underflow.
3302 * In that case, we may try to free space we haven't reserved
3303 * (since quota was disabled), so record what we reserved into root.
3304 * And ensure later release won't underflow this number.
3306 add_root_meta_rsv(root, num_bytes, type);
3310 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3312 struct btrfs_fs_info *fs_info = root->fs_info;
3314 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3315 !is_fstree(root->objectid))
3318 /* TODO: Update trace point to handle such free */
3319 trace_qgroup_meta_free_all_pertrans(root);
3320 /* Special value -1 means to free all reserved space */
3321 btrfs_qgroup_free_refroot(fs_info, root->objectid, (u64)-1,
3322 BTRFS_QGROUP_RSV_META_PERTRANS);
3325 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3326 enum btrfs_qgroup_rsv_type type)
3328 struct btrfs_fs_info *fs_info = root->fs_info;
3330 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3331 !is_fstree(root->objectid))
3335 * reservation for META_PREALLOC can happen before quota is enabled,
3336 * which can lead to underflow.
3337 * Here ensure we will only free what we really have reserved.
3339 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3340 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3341 trace_qgroup_meta_reserve(root, -(s64)num_bytes, type);
3342 btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes, type);
3345 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3348 struct btrfs_root *quota_root = fs_info->quota_root;
3349 struct btrfs_qgroup *qgroup;
3350 struct ulist_node *unode;
3351 struct ulist_iterator uiter;
3359 spin_lock(&fs_info->qgroup_lock);
3360 qgroup = find_qgroup_rb(fs_info, ref_root);
3363 ulist_reinit(fs_info->qgroup_ulist);
3364 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3365 qgroup_to_aux(qgroup), GFP_ATOMIC);
3368 ULIST_ITER_INIT(&uiter);
3369 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3370 struct btrfs_qgroup *qg;
3371 struct btrfs_qgroup_list *glist;
3373 qg = unode_aux_to_qgroup(unode);
3375 qgroup_rsv_release(fs_info, qg, num_bytes,
3376 BTRFS_QGROUP_RSV_META_PREALLOC);
3377 qgroup_rsv_add(fs_info, qg, num_bytes,
3378 BTRFS_QGROUP_RSV_META_PERTRANS);
3379 list_for_each_entry(glist, &qg->groups, next_group) {
3380 ret = ulist_add(fs_info->qgroup_ulist,
3381 glist->group->qgroupid,
3382 qgroup_to_aux(glist->group), GFP_ATOMIC);
3388 spin_unlock(&fs_info->qgroup_lock);
3391 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3393 struct btrfs_fs_info *fs_info = root->fs_info;
3395 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3396 !is_fstree(root->objectid))
3398 /* Same as btrfs_qgroup_free_meta_prealloc() */
3399 num_bytes = sub_root_meta_rsv(root, num_bytes,
3400 BTRFS_QGROUP_RSV_META_PREALLOC);
3401 trace_qgroup_meta_convert(root, num_bytes);
3402 qgroup_convert_meta(fs_info, root->objectid, num_bytes);
3406 * Check qgroup reserved space leaking, normally at destroy inode
3409 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3411 struct extent_changeset changeset;
3412 struct ulist_node *unode;
3413 struct ulist_iterator iter;
3416 extent_changeset_init(&changeset);
3417 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3418 EXTENT_QGROUP_RESERVED, &changeset);
3421 if (WARN_ON(changeset.bytes_changed)) {
3422 ULIST_ITER_INIT(&iter);
3423 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3424 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3425 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3426 inode->i_ino, unode->val, unode->aux);
3428 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3429 BTRFS_I(inode)->root->objectid,
3430 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3433 extent_changeset_release(&changeset);