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>
16 #include "transaction.h"
21 #include "extent_io.h"
26 * - subvol delete -> delete when ref goes to 0? delete limits also?
30 * - copy also limits on subvol creation
32 * - caches fuer ulists
33 * - performance benchmarks
34 * - check all ioctl parameters
38 * Helpers to access qgroup reservation
40 * Callers should ensure the lock context and type are valid
43 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
48 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
49 ret += qgroup->rsv.values[i];
54 #ifdef CONFIG_BTRFS_DEBUG
55 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
57 if (type == BTRFS_QGROUP_RSV_DATA)
59 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
60 return "meta_pertrans";
61 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
62 return "meta_prealloc";
67 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
68 struct btrfs_qgroup *qgroup, u64 num_bytes,
69 enum btrfs_qgroup_rsv_type type)
71 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
72 qgroup->rsv.values[type] += num_bytes;
75 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
76 struct btrfs_qgroup *qgroup, u64 num_bytes,
77 enum btrfs_qgroup_rsv_type type)
79 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
80 if (qgroup->rsv.values[type] >= num_bytes) {
81 qgroup->rsv.values[type] -= num_bytes;
84 #ifdef CONFIG_BTRFS_DEBUG
86 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
87 qgroup->qgroupid, qgroup_rsv_type_str(type),
88 qgroup->rsv.values[type], num_bytes);
90 qgroup->rsv.values[type] = 0;
93 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
94 struct btrfs_qgroup *dest,
95 struct btrfs_qgroup *src)
99 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
100 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
103 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
104 struct btrfs_qgroup *dest,
105 struct btrfs_qgroup *src)
109 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
110 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
113 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
116 if (qg->old_refcnt < seq)
117 qg->old_refcnt = seq;
118 qg->old_refcnt += mod;
121 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
124 if (qg->new_refcnt < seq)
125 qg->new_refcnt = seq;
126 qg->new_refcnt += mod;
129 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
131 if (qg->old_refcnt < seq)
133 return qg->old_refcnt - seq;
136 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
138 if (qg->new_refcnt < seq)
140 return qg->new_refcnt - seq;
144 * glue structure to represent the relations between qgroups.
146 struct btrfs_qgroup_list {
147 struct list_head next_group;
148 struct list_head next_member;
149 struct btrfs_qgroup *group;
150 struct btrfs_qgroup *member;
153 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
155 return (u64)(uintptr_t)qg;
158 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
160 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
164 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
166 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
168 /* must be called with qgroup_ioctl_lock held */
169 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
172 struct rb_node *n = fs_info->qgroup_tree.rb_node;
173 struct btrfs_qgroup *qgroup;
176 qgroup = rb_entry(n, struct btrfs_qgroup, node);
177 if (qgroup->qgroupid < qgroupid)
179 else if (qgroup->qgroupid > qgroupid)
187 /* must be called with qgroup_lock held */
188 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
191 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
192 struct rb_node *parent = NULL;
193 struct btrfs_qgroup *qgroup;
197 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
199 if (qgroup->qgroupid < qgroupid)
201 else if (qgroup->qgroupid > qgroupid)
207 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
209 return ERR_PTR(-ENOMEM);
211 qgroup->qgroupid = qgroupid;
212 INIT_LIST_HEAD(&qgroup->groups);
213 INIT_LIST_HEAD(&qgroup->members);
214 INIT_LIST_HEAD(&qgroup->dirty);
216 rb_link_node(&qgroup->node, parent, p);
217 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
222 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
224 struct btrfs_qgroup_list *list;
226 list_del(&qgroup->dirty);
227 while (!list_empty(&qgroup->groups)) {
228 list = list_first_entry(&qgroup->groups,
229 struct btrfs_qgroup_list, next_group);
230 list_del(&list->next_group);
231 list_del(&list->next_member);
235 while (!list_empty(&qgroup->members)) {
236 list = list_first_entry(&qgroup->members,
237 struct btrfs_qgroup_list, next_member);
238 list_del(&list->next_group);
239 list_del(&list->next_member);
245 /* must be called with qgroup_lock held */
246 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
248 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
253 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
254 __del_qgroup_rb(qgroup);
258 /* must be called with qgroup_lock held */
259 static int add_relation_rb(struct btrfs_fs_info *fs_info,
260 u64 memberid, u64 parentid)
262 struct btrfs_qgroup *member;
263 struct btrfs_qgroup *parent;
264 struct btrfs_qgroup_list *list;
266 member = find_qgroup_rb(fs_info, memberid);
267 parent = find_qgroup_rb(fs_info, parentid);
268 if (!member || !parent)
271 list = kzalloc(sizeof(*list), GFP_ATOMIC);
275 list->group = parent;
276 list->member = member;
277 list_add_tail(&list->next_group, &member->groups);
278 list_add_tail(&list->next_member, &parent->members);
283 /* must be called with qgroup_lock held */
284 static int del_relation_rb(struct btrfs_fs_info *fs_info,
285 u64 memberid, u64 parentid)
287 struct btrfs_qgroup *member;
288 struct btrfs_qgroup *parent;
289 struct btrfs_qgroup_list *list;
291 member = find_qgroup_rb(fs_info, memberid);
292 parent = find_qgroup_rb(fs_info, parentid);
293 if (!member || !parent)
296 list_for_each_entry(list, &member->groups, next_group) {
297 if (list->group == parent) {
298 list_del(&list->next_group);
299 list_del(&list->next_member);
307 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
308 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
311 struct btrfs_qgroup *qgroup;
313 qgroup = find_qgroup_rb(fs_info, qgroupid);
316 if (qgroup->rfer != rfer || qgroup->excl != excl)
323 * The full config is read in one go, only called from open_ctree()
324 * It doesn't use any locking, as at this point we're still single-threaded
326 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
328 struct btrfs_key key;
329 struct btrfs_key found_key;
330 struct btrfs_root *quota_root = fs_info->quota_root;
331 struct btrfs_path *path = NULL;
332 struct extent_buffer *l;
336 u64 rescan_progress = 0;
338 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
341 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
342 if (!fs_info->qgroup_ulist) {
347 path = btrfs_alloc_path();
353 /* default this to quota off, in case no status key is found */
354 fs_info->qgroup_flags = 0;
357 * pass 1: read status, all qgroup infos and limits
362 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
367 struct btrfs_qgroup *qgroup;
369 slot = path->slots[0];
371 btrfs_item_key_to_cpu(l, &found_key, slot);
373 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
374 struct btrfs_qgroup_status_item *ptr;
376 ptr = btrfs_item_ptr(l, slot,
377 struct btrfs_qgroup_status_item);
379 if (btrfs_qgroup_status_version(l, ptr) !=
380 BTRFS_QGROUP_STATUS_VERSION) {
382 "old qgroup version, quota disabled");
385 if (btrfs_qgroup_status_generation(l, ptr) !=
386 fs_info->generation) {
387 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
389 "qgroup generation mismatch, marked as inconsistent");
391 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
393 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
397 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
398 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
401 qgroup = find_qgroup_rb(fs_info, found_key.offset);
402 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
403 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
404 btrfs_err(fs_info, "inconsistent qgroup config");
405 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
408 qgroup = add_qgroup_rb(fs_info, found_key.offset);
409 if (IS_ERR(qgroup)) {
410 ret = PTR_ERR(qgroup);
414 switch (found_key.type) {
415 case BTRFS_QGROUP_INFO_KEY: {
416 struct btrfs_qgroup_info_item *ptr;
418 ptr = btrfs_item_ptr(l, slot,
419 struct btrfs_qgroup_info_item);
420 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
421 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
422 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
423 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
424 /* generation currently unused */
427 case BTRFS_QGROUP_LIMIT_KEY: {
428 struct btrfs_qgroup_limit_item *ptr;
430 ptr = btrfs_item_ptr(l, slot,
431 struct btrfs_qgroup_limit_item);
432 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
433 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
434 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
435 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
436 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
441 ret = btrfs_next_item(quota_root, path);
447 btrfs_release_path(path);
450 * pass 2: read all qgroup relations
453 key.type = BTRFS_QGROUP_RELATION_KEY;
455 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
459 slot = path->slots[0];
461 btrfs_item_key_to_cpu(l, &found_key, slot);
463 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
466 if (found_key.objectid > found_key.offset) {
467 /* parent <- member, not needed to build config */
468 /* FIXME should we omit the key completely? */
472 ret = add_relation_rb(fs_info, found_key.objectid,
474 if (ret == -ENOENT) {
476 "orphan qgroup relation 0x%llx->0x%llx",
477 found_key.objectid, found_key.offset);
478 ret = 0; /* ignore the error */
483 ret = btrfs_next_item(quota_root, path);
490 fs_info->qgroup_flags |= flags;
491 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
492 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
493 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
495 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
496 btrfs_free_path(path);
499 ulist_free(fs_info->qgroup_ulist);
500 fs_info->qgroup_ulist = NULL;
501 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
504 return ret < 0 ? ret : 0;
508 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
509 * first two are in single-threaded paths.And for the third one, we have set
510 * quota_root to be null with qgroup_lock held before, so it is safe to clean
511 * up the in-memory structures without qgroup_lock held.
513 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
516 struct btrfs_qgroup *qgroup;
518 while ((n = rb_first(&fs_info->qgroup_tree))) {
519 qgroup = rb_entry(n, struct btrfs_qgroup, node);
520 rb_erase(n, &fs_info->qgroup_tree);
521 __del_qgroup_rb(qgroup);
524 * we call btrfs_free_qgroup_config() when umounting
525 * filesystem and disabling quota, so we set qgroup_ulist
526 * to be null here to avoid double free.
528 ulist_free(fs_info->qgroup_ulist);
529 fs_info->qgroup_ulist = NULL;
532 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans,
533 struct btrfs_root *quota_root,
537 struct btrfs_path *path;
538 struct btrfs_key key;
540 path = btrfs_alloc_path();
545 key.type = BTRFS_QGROUP_RELATION_KEY;
548 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
550 btrfs_mark_buffer_dirty(path->nodes[0]);
552 btrfs_free_path(path);
556 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans,
557 struct btrfs_root *quota_root,
561 struct btrfs_path *path;
562 struct btrfs_key key;
564 path = btrfs_alloc_path();
569 key.type = BTRFS_QGROUP_RELATION_KEY;
572 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
581 ret = btrfs_del_item(trans, quota_root, path);
583 btrfs_free_path(path);
587 static int add_qgroup_item(struct btrfs_trans_handle *trans,
588 struct btrfs_root *quota_root, u64 qgroupid)
591 struct btrfs_path *path;
592 struct btrfs_qgroup_info_item *qgroup_info;
593 struct btrfs_qgroup_limit_item *qgroup_limit;
594 struct extent_buffer *leaf;
595 struct btrfs_key key;
597 if (btrfs_is_testing(quota_root->fs_info))
600 path = btrfs_alloc_path();
605 key.type = BTRFS_QGROUP_INFO_KEY;
606 key.offset = qgroupid;
609 * Avoid a transaction abort by catching -EEXIST here. In that
610 * case, we proceed by re-initializing the existing structure
614 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
615 sizeof(*qgroup_info));
616 if (ret && ret != -EEXIST)
619 leaf = path->nodes[0];
620 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
621 struct btrfs_qgroup_info_item);
622 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
623 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
624 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
625 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
626 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
628 btrfs_mark_buffer_dirty(leaf);
630 btrfs_release_path(path);
632 key.type = BTRFS_QGROUP_LIMIT_KEY;
633 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
634 sizeof(*qgroup_limit));
635 if (ret && ret != -EEXIST)
638 leaf = path->nodes[0];
639 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
640 struct btrfs_qgroup_limit_item);
641 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
642 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
643 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
644 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
645 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
647 btrfs_mark_buffer_dirty(leaf);
651 btrfs_free_path(path);
655 static int del_qgroup_item(struct btrfs_trans_handle *trans,
656 struct btrfs_root *quota_root, u64 qgroupid)
659 struct btrfs_path *path;
660 struct btrfs_key key;
662 path = btrfs_alloc_path();
667 key.type = BTRFS_QGROUP_INFO_KEY;
668 key.offset = qgroupid;
669 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
678 ret = btrfs_del_item(trans, quota_root, path);
682 btrfs_release_path(path);
684 key.type = BTRFS_QGROUP_LIMIT_KEY;
685 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
694 ret = btrfs_del_item(trans, quota_root, path);
697 btrfs_free_path(path);
701 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
702 struct btrfs_root *root,
703 struct btrfs_qgroup *qgroup)
705 struct btrfs_path *path;
706 struct btrfs_key key;
707 struct extent_buffer *l;
708 struct btrfs_qgroup_limit_item *qgroup_limit;
713 key.type = BTRFS_QGROUP_LIMIT_KEY;
714 key.offset = qgroup->qgroupid;
716 path = btrfs_alloc_path();
720 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
728 slot = path->slots[0];
729 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
730 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
731 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
732 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
733 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
734 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
736 btrfs_mark_buffer_dirty(l);
739 btrfs_free_path(path);
743 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
744 struct btrfs_root *root,
745 struct btrfs_qgroup *qgroup)
747 struct btrfs_path *path;
748 struct btrfs_key key;
749 struct extent_buffer *l;
750 struct btrfs_qgroup_info_item *qgroup_info;
754 if (btrfs_is_testing(root->fs_info))
758 key.type = BTRFS_QGROUP_INFO_KEY;
759 key.offset = qgroup->qgroupid;
761 path = btrfs_alloc_path();
765 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
773 slot = path->slots[0];
774 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
775 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
776 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
777 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
778 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
779 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
781 btrfs_mark_buffer_dirty(l);
784 btrfs_free_path(path);
788 static int update_qgroup_status_item(struct btrfs_trans_handle *trans,
789 struct btrfs_fs_info *fs_info,
790 struct btrfs_root *root)
792 struct btrfs_path *path;
793 struct btrfs_key key;
794 struct extent_buffer *l;
795 struct btrfs_qgroup_status_item *ptr;
800 key.type = BTRFS_QGROUP_STATUS_KEY;
803 path = btrfs_alloc_path();
807 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
815 slot = path->slots[0];
816 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
817 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
818 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
819 btrfs_set_qgroup_status_rescan(l, ptr,
820 fs_info->qgroup_rescan_progress.objectid);
822 btrfs_mark_buffer_dirty(l);
825 btrfs_free_path(path);
830 * called with qgroup_lock held
832 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
833 struct btrfs_root *root)
835 struct btrfs_path *path;
836 struct btrfs_key key;
837 struct extent_buffer *leaf = NULL;
841 path = btrfs_alloc_path();
845 path->leave_spinning = 1;
852 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
855 leaf = path->nodes[0];
856 nr = btrfs_header_nritems(leaf);
860 * delete the leaf one by one
861 * since the whole tree is going
865 ret = btrfs_del_items(trans, root, path, 0, nr);
869 btrfs_release_path(path);
873 btrfs_free_path(path);
877 int btrfs_quota_enable(struct btrfs_trans_handle *trans,
878 struct btrfs_fs_info *fs_info)
880 struct btrfs_root *quota_root;
881 struct btrfs_root *tree_root = fs_info->tree_root;
882 struct btrfs_path *path = NULL;
883 struct btrfs_qgroup_status_item *ptr;
884 struct extent_buffer *leaf;
885 struct btrfs_key key;
886 struct btrfs_key found_key;
887 struct btrfs_qgroup *qgroup = NULL;
891 mutex_lock(&fs_info->qgroup_ioctl_lock);
892 if (fs_info->quota_root)
895 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
896 if (!fs_info->qgroup_ulist) {
902 * initially create the quota tree
904 quota_root = btrfs_create_tree(trans, fs_info,
905 BTRFS_QUOTA_TREE_OBJECTID);
906 if (IS_ERR(quota_root)) {
907 ret = PTR_ERR(quota_root);
911 path = btrfs_alloc_path();
918 key.type = BTRFS_QGROUP_STATUS_KEY;
921 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
926 leaf = path->nodes[0];
927 ptr = btrfs_item_ptr(leaf, path->slots[0],
928 struct btrfs_qgroup_status_item);
929 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
930 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
931 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
932 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
933 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
934 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
936 btrfs_mark_buffer_dirty(leaf);
939 key.type = BTRFS_ROOT_REF_KEY;
942 btrfs_release_path(path);
943 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
951 slot = path->slots[0];
952 leaf = path->nodes[0];
953 btrfs_item_key_to_cpu(leaf, &found_key, slot);
955 if (found_key.type == BTRFS_ROOT_REF_KEY) {
956 ret = add_qgroup_item(trans, quota_root,
961 qgroup = add_qgroup_rb(fs_info, found_key.offset);
962 if (IS_ERR(qgroup)) {
963 ret = PTR_ERR(qgroup);
967 ret = btrfs_next_item(tree_root, path);
975 btrfs_release_path(path);
976 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
980 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
981 if (IS_ERR(qgroup)) {
982 ret = PTR_ERR(qgroup);
985 spin_lock(&fs_info->qgroup_lock);
986 fs_info->quota_root = quota_root;
987 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
988 spin_unlock(&fs_info->qgroup_lock);
989 ret = qgroup_rescan_init(fs_info, 0, 1);
991 qgroup_rescan_zero_tracking(fs_info);
992 btrfs_queue_work(fs_info->qgroup_rescan_workers,
993 &fs_info->qgroup_rescan_work);
997 btrfs_free_path(path);
1000 free_extent_buffer(quota_root->node);
1001 free_extent_buffer(quota_root->commit_root);
1006 ulist_free(fs_info->qgroup_ulist);
1007 fs_info->qgroup_ulist = NULL;
1009 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1013 int btrfs_quota_disable(struct btrfs_trans_handle *trans,
1014 struct btrfs_fs_info *fs_info)
1016 struct btrfs_root *quota_root;
1019 mutex_lock(&fs_info->qgroup_ioctl_lock);
1020 if (!fs_info->quota_root)
1022 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1023 btrfs_qgroup_wait_for_completion(fs_info, false);
1024 spin_lock(&fs_info->qgroup_lock);
1025 quota_root = fs_info->quota_root;
1026 fs_info->quota_root = NULL;
1027 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1028 spin_unlock(&fs_info->qgroup_lock);
1030 btrfs_free_qgroup_config(fs_info);
1032 ret = btrfs_clean_quota_tree(trans, quota_root);
1036 ret = btrfs_del_root(trans, fs_info, "a_root->root_key);
1040 list_del("a_root->dirty_list);
1042 btrfs_tree_lock(quota_root->node);
1043 clean_tree_block(fs_info, quota_root->node);
1044 btrfs_tree_unlock(quota_root->node);
1045 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1047 free_extent_buffer(quota_root->node);
1048 free_extent_buffer(quota_root->commit_root);
1051 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1055 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1056 struct btrfs_qgroup *qgroup)
1058 if (list_empty(&qgroup->dirty))
1059 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1063 * The easy accounting, we're updating qgroup relationship whose child qgroup
1064 * only has exclusive extents.
1066 * In this case, all exclsuive extents will also be exlusive for parent, so
1067 * excl/rfer just get added/removed.
1069 * So is qgroup reservation space, which should also be added/removed to
1071 * Or when child tries to release reservation space, parent will underflow its
1072 * reservation (for relationship adding case).
1074 * Caller should hold fs_info->qgroup_lock.
1076 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1077 struct ulist *tmp, u64 ref_root,
1078 struct btrfs_qgroup *src, int sign)
1080 struct btrfs_qgroup *qgroup;
1081 struct btrfs_qgroup_list *glist;
1082 struct ulist_node *unode;
1083 struct ulist_iterator uiter;
1084 u64 num_bytes = src->excl;
1087 qgroup = find_qgroup_rb(fs_info, ref_root);
1091 qgroup->rfer += sign * num_bytes;
1092 qgroup->rfer_cmpr += sign * num_bytes;
1094 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1095 qgroup->excl += sign * num_bytes;
1096 qgroup->excl_cmpr += sign * num_bytes;
1099 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1101 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1103 qgroup_dirty(fs_info, qgroup);
1105 /* Get all of the parent groups that contain this qgroup */
1106 list_for_each_entry(glist, &qgroup->groups, next_group) {
1107 ret = ulist_add(tmp, glist->group->qgroupid,
1108 qgroup_to_aux(glist->group), GFP_ATOMIC);
1113 /* Iterate all of the parents and adjust their reference counts */
1114 ULIST_ITER_INIT(&uiter);
1115 while ((unode = ulist_next(tmp, &uiter))) {
1116 qgroup = unode_aux_to_qgroup(unode);
1117 qgroup->rfer += sign * num_bytes;
1118 qgroup->rfer_cmpr += sign * num_bytes;
1119 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1120 qgroup->excl += sign * num_bytes;
1122 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1124 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1125 qgroup->excl_cmpr += sign * num_bytes;
1126 qgroup_dirty(fs_info, qgroup);
1128 /* Add any parents of the parents */
1129 list_for_each_entry(glist, &qgroup->groups, next_group) {
1130 ret = ulist_add(tmp, glist->group->qgroupid,
1131 qgroup_to_aux(glist->group), GFP_ATOMIC);
1143 * Quick path for updating qgroup with only excl refs.
1145 * In that case, just update all parent will be enough.
1146 * Or we needs to do a full rescan.
1147 * Caller should also hold fs_info->qgroup_lock.
1149 * Return 0 for quick update, return >0 for need to full rescan
1150 * and mark INCONSISTENT flag.
1151 * Return < 0 for other error.
1153 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1154 struct ulist *tmp, u64 src, u64 dst,
1157 struct btrfs_qgroup *qgroup;
1161 qgroup = find_qgroup_rb(fs_info, src);
1164 if (qgroup->excl == qgroup->rfer) {
1166 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1175 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1179 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans,
1180 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1182 struct btrfs_root *quota_root;
1183 struct btrfs_qgroup *parent;
1184 struct btrfs_qgroup *member;
1185 struct btrfs_qgroup_list *list;
1189 /* Check the level of src and dst first */
1190 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1193 tmp = ulist_alloc(GFP_KERNEL);
1197 mutex_lock(&fs_info->qgroup_ioctl_lock);
1198 quota_root = fs_info->quota_root;
1203 member = find_qgroup_rb(fs_info, src);
1204 parent = find_qgroup_rb(fs_info, dst);
1205 if (!member || !parent) {
1210 /* check if such qgroup relation exist firstly */
1211 list_for_each_entry(list, &member->groups, next_group) {
1212 if (list->group == parent) {
1218 ret = add_qgroup_relation_item(trans, quota_root, src, dst);
1222 ret = add_qgroup_relation_item(trans, quota_root, dst, src);
1224 del_qgroup_relation_item(trans, quota_root, src, dst);
1228 spin_lock(&fs_info->qgroup_lock);
1229 ret = add_relation_rb(fs_info, src, dst);
1231 spin_unlock(&fs_info->qgroup_lock);
1234 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1235 spin_unlock(&fs_info->qgroup_lock);
1237 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1242 static int __del_qgroup_relation(struct btrfs_trans_handle *trans,
1243 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1245 struct btrfs_root *quota_root;
1246 struct btrfs_qgroup *parent;
1247 struct btrfs_qgroup *member;
1248 struct btrfs_qgroup_list *list;
1253 tmp = ulist_alloc(GFP_KERNEL);
1257 quota_root = fs_info->quota_root;
1263 member = find_qgroup_rb(fs_info, src);
1264 parent = find_qgroup_rb(fs_info, dst);
1265 if (!member || !parent) {
1270 /* check if such qgroup relation exist firstly */
1271 list_for_each_entry(list, &member->groups, next_group) {
1272 if (list->group == parent)
1278 ret = del_qgroup_relation_item(trans, quota_root, src, dst);
1279 err = del_qgroup_relation_item(trans, quota_root, dst, src);
1283 spin_lock(&fs_info->qgroup_lock);
1284 del_relation_rb(fs_info, src, dst);
1285 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1286 spin_unlock(&fs_info->qgroup_lock);
1292 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans,
1293 struct btrfs_fs_info *fs_info, u64 src, u64 dst)
1297 mutex_lock(&fs_info->qgroup_ioctl_lock);
1298 ret = __del_qgroup_relation(trans, fs_info, src, dst);
1299 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1304 int btrfs_create_qgroup(struct btrfs_trans_handle *trans,
1305 struct btrfs_fs_info *fs_info, u64 qgroupid)
1307 struct btrfs_root *quota_root;
1308 struct btrfs_qgroup *qgroup;
1311 mutex_lock(&fs_info->qgroup_ioctl_lock);
1312 quota_root = fs_info->quota_root;
1317 qgroup = find_qgroup_rb(fs_info, qgroupid);
1323 ret = add_qgroup_item(trans, quota_root, qgroupid);
1327 spin_lock(&fs_info->qgroup_lock);
1328 qgroup = add_qgroup_rb(fs_info, qgroupid);
1329 spin_unlock(&fs_info->qgroup_lock);
1332 ret = PTR_ERR(qgroup);
1334 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1338 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans,
1339 struct btrfs_fs_info *fs_info, u64 qgroupid)
1341 struct btrfs_root *quota_root;
1342 struct btrfs_qgroup *qgroup;
1343 struct btrfs_qgroup_list *list;
1346 mutex_lock(&fs_info->qgroup_ioctl_lock);
1347 quota_root = fs_info->quota_root;
1353 qgroup = find_qgroup_rb(fs_info, qgroupid);
1358 /* check if there are no children of this qgroup */
1359 if (!list_empty(&qgroup->members)) {
1364 ret = del_qgroup_item(trans, quota_root, qgroupid);
1365 if (ret && ret != -ENOENT)
1368 while (!list_empty(&qgroup->groups)) {
1369 list = list_first_entry(&qgroup->groups,
1370 struct btrfs_qgroup_list, next_group);
1371 ret = __del_qgroup_relation(trans, fs_info,
1373 list->group->qgroupid);
1378 spin_lock(&fs_info->qgroup_lock);
1379 del_qgroup_rb(fs_info, qgroupid);
1380 spin_unlock(&fs_info->qgroup_lock);
1382 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1386 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans,
1387 struct btrfs_fs_info *fs_info, u64 qgroupid,
1388 struct btrfs_qgroup_limit *limit)
1390 struct btrfs_root *quota_root;
1391 struct btrfs_qgroup *qgroup;
1393 /* Sometimes we would want to clear the limit on this qgroup.
1394 * To meet this requirement, we treat the -1 as a special value
1395 * which tell kernel to clear the limit on this qgroup.
1397 const u64 CLEAR_VALUE = -1;
1399 mutex_lock(&fs_info->qgroup_ioctl_lock);
1400 quota_root = fs_info->quota_root;
1406 qgroup = find_qgroup_rb(fs_info, qgroupid);
1412 spin_lock(&fs_info->qgroup_lock);
1413 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1414 if (limit->max_rfer == CLEAR_VALUE) {
1415 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1416 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1417 qgroup->max_rfer = 0;
1419 qgroup->max_rfer = limit->max_rfer;
1422 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1423 if (limit->max_excl == CLEAR_VALUE) {
1424 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1425 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1426 qgroup->max_excl = 0;
1428 qgroup->max_excl = limit->max_excl;
1431 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1432 if (limit->rsv_rfer == CLEAR_VALUE) {
1433 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1434 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1435 qgroup->rsv_rfer = 0;
1437 qgroup->rsv_rfer = limit->rsv_rfer;
1440 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1441 if (limit->rsv_excl == CLEAR_VALUE) {
1442 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1443 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1444 qgroup->rsv_excl = 0;
1446 qgroup->rsv_excl = limit->rsv_excl;
1449 qgroup->lim_flags |= limit->flags;
1451 spin_unlock(&fs_info->qgroup_lock);
1453 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
1455 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1456 btrfs_info(fs_info, "unable to update quota limit for %llu",
1461 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1465 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1466 struct btrfs_delayed_ref_root *delayed_refs,
1467 struct btrfs_qgroup_extent_record *record)
1469 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1470 struct rb_node *parent_node = NULL;
1471 struct btrfs_qgroup_extent_record *entry;
1472 u64 bytenr = record->bytenr;
1474 lockdep_assert_held(&delayed_refs->lock);
1475 trace_btrfs_qgroup_trace_extent(fs_info, record);
1479 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1481 if (bytenr < entry->bytenr)
1483 else if (bytenr > entry->bytenr)
1484 p = &(*p)->rb_right;
1489 rb_link_node(&record->node, parent_node, p);
1490 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1494 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1495 struct btrfs_qgroup_extent_record *qrecord)
1497 struct ulist *old_root;
1498 u64 bytenr = qrecord->bytenr;
1501 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1503 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1505 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1511 * Here we don't need to get the lock of
1512 * trans->transaction->delayed_refs, since inserted qrecord won't
1513 * be deleted, only qrecord->node may be modified (new qrecord insert)
1515 * So modifying qrecord->old_roots is safe here
1517 qrecord->old_roots = old_root;
1521 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
1522 struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
1525 struct btrfs_qgroup_extent_record *record;
1526 struct btrfs_delayed_ref_root *delayed_refs;
1529 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1530 || bytenr == 0 || num_bytes == 0)
1532 if (WARN_ON(trans == NULL))
1534 record = kmalloc(sizeof(*record), gfp_flag);
1538 delayed_refs = &trans->transaction->delayed_refs;
1539 record->bytenr = bytenr;
1540 record->num_bytes = num_bytes;
1541 record->old_roots = NULL;
1543 spin_lock(&delayed_refs->lock);
1544 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1545 spin_unlock(&delayed_refs->lock);
1550 return btrfs_qgroup_trace_extent_post(fs_info, record);
1553 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1554 struct btrfs_fs_info *fs_info,
1555 struct extent_buffer *eb)
1557 int nr = btrfs_header_nritems(eb);
1558 int i, extent_type, ret;
1559 struct btrfs_key key;
1560 struct btrfs_file_extent_item *fi;
1561 u64 bytenr, num_bytes;
1563 /* We can be called directly from walk_up_proc() */
1564 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1567 for (i = 0; i < nr; i++) {
1568 btrfs_item_key_to_cpu(eb, &key, i);
1570 if (key.type != BTRFS_EXTENT_DATA_KEY)
1573 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1574 /* filter out non qgroup-accountable extents */
1575 extent_type = btrfs_file_extent_type(eb, fi);
1577 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1580 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1584 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1586 ret = btrfs_qgroup_trace_extent(trans, fs_info, bytenr,
1587 num_bytes, GFP_NOFS);
1596 * Walk up the tree from the bottom, freeing leaves and any interior
1597 * nodes which have had all slots visited. If a node (leaf or
1598 * interior) is freed, the node above it will have it's slot
1599 * incremented. The root node will never be freed.
1601 * At the end of this function, we should have a path which has all
1602 * slots incremented to the next position for a search. If we need to
1603 * read a new node it will be NULL and the node above it will have the
1604 * correct slot selected for a later read.
1606 * If we increment the root nodes slot counter past the number of
1607 * elements, 1 is returned to signal completion of the search.
1609 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1613 struct extent_buffer *eb;
1615 if (root_level == 0)
1618 while (level <= root_level) {
1619 eb = path->nodes[level];
1620 nr = btrfs_header_nritems(eb);
1621 path->slots[level]++;
1622 slot = path->slots[level];
1623 if (slot >= nr || level == 0) {
1625 * Don't free the root - we will detect this
1626 * condition after our loop and return a
1627 * positive value for caller to stop walking the tree.
1629 if (level != root_level) {
1630 btrfs_tree_unlock_rw(eb, path->locks[level]);
1631 path->locks[level] = 0;
1633 free_extent_buffer(eb);
1634 path->nodes[level] = NULL;
1635 path->slots[level] = 0;
1639 * We have a valid slot to walk back down
1640 * from. Stop here so caller can process these
1649 eb = path->nodes[root_level];
1650 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1656 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
1657 struct btrfs_root *root,
1658 struct extent_buffer *root_eb,
1659 u64 root_gen, int root_level)
1661 struct btrfs_fs_info *fs_info = root->fs_info;
1664 struct extent_buffer *eb = root_eb;
1665 struct btrfs_path *path = NULL;
1667 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
1668 BUG_ON(root_eb == NULL);
1670 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1673 if (!extent_buffer_uptodate(root_eb)) {
1674 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
1679 if (root_level == 0) {
1680 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, root_eb);
1684 path = btrfs_alloc_path();
1689 * Walk down the tree. Missing extent blocks are filled in as
1690 * we go. Metadata is accounted every time we read a new
1693 * When we reach a leaf, we account for file extent items in it,
1694 * walk back up the tree (adjusting slot pointers as we go)
1695 * and restart the search process.
1697 extent_buffer_get(root_eb); /* For path */
1698 path->nodes[root_level] = root_eb;
1699 path->slots[root_level] = 0;
1700 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
1703 while (level >= 0) {
1704 if (path->nodes[level] == NULL) {
1705 struct btrfs_key first_key;
1711 * We need to get child blockptr/gen from parent before
1714 eb = path->nodes[level + 1];
1715 parent_slot = path->slots[level + 1];
1716 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1717 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1718 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1720 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1725 } else if (!extent_buffer_uptodate(eb)) {
1726 free_extent_buffer(eb);
1731 path->nodes[level] = eb;
1732 path->slots[level] = 0;
1734 btrfs_tree_read_lock(eb);
1735 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1736 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
1738 ret = btrfs_qgroup_trace_extent(trans, fs_info,
1747 ret = btrfs_qgroup_trace_leaf_items(trans,fs_info,
1748 path->nodes[level]);
1752 /* Nonzero return here means we completed our search */
1753 ret = adjust_slots_upwards(path, root_level);
1757 /* Restart search with new slots */
1766 btrfs_free_path(path);
1771 #define UPDATE_NEW 0
1772 #define UPDATE_OLD 1
1774 * Walk all of the roots that points to the bytenr and adjust their refcnts.
1776 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
1777 struct ulist *roots, struct ulist *tmp,
1778 struct ulist *qgroups, u64 seq, int update_old)
1780 struct ulist_node *unode;
1781 struct ulist_iterator uiter;
1782 struct ulist_node *tmp_unode;
1783 struct ulist_iterator tmp_uiter;
1784 struct btrfs_qgroup *qg;
1789 ULIST_ITER_INIT(&uiter);
1790 while ((unode = ulist_next(roots, &uiter))) {
1791 qg = find_qgroup_rb(fs_info, unode->val);
1796 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
1800 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
1803 ULIST_ITER_INIT(&tmp_uiter);
1804 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
1805 struct btrfs_qgroup_list *glist;
1807 qg = unode_aux_to_qgroup(tmp_unode);
1809 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
1811 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
1812 list_for_each_entry(glist, &qg->groups, next_group) {
1813 ret = ulist_add(qgroups, glist->group->qgroupid,
1814 qgroup_to_aux(glist->group),
1818 ret = ulist_add(tmp, glist->group->qgroupid,
1819 qgroup_to_aux(glist->group),
1830 * Update qgroup rfer/excl counters.
1831 * Rfer update is easy, codes can explain themselves.
1833 * Excl update is tricky, the update is split into 2 part.
1834 * Part 1: Possible exclusive <-> sharing detect:
1836 * -------------------------------------
1838 * -------------------------------------
1840 * -------------------------------------
1843 * A: cur_old_roots < nr_old_roots (not exclusive before)
1844 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
1845 * B: cur_new_roots < nr_new_roots (not exclusive now)
1846 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
1849 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
1850 * *: Definitely not changed. **: Possible unchanged.
1852 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
1854 * To make the logic clear, we first use condition A and B to split
1855 * combination into 4 results.
1857 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
1858 * only on variant maybe 0.
1860 * Lastly, check result **, since there are 2 variants maybe 0, split them
1862 * But this time we don't need to consider other things, the codes and logic
1863 * is easy to understand now.
1865 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
1866 struct ulist *qgroups,
1869 u64 num_bytes, u64 seq)
1871 struct ulist_node *unode;
1872 struct ulist_iterator uiter;
1873 struct btrfs_qgroup *qg;
1874 u64 cur_new_count, cur_old_count;
1876 ULIST_ITER_INIT(&uiter);
1877 while ((unode = ulist_next(qgroups, &uiter))) {
1880 qg = unode_aux_to_qgroup(unode);
1881 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
1882 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
1884 trace_qgroup_update_counters(fs_info, qg->qgroupid,
1885 cur_old_count, cur_new_count);
1887 /* Rfer update part */
1888 if (cur_old_count == 0 && cur_new_count > 0) {
1889 qg->rfer += num_bytes;
1890 qg->rfer_cmpr += num_bytes;
1893 if (cur_old_count > 0 && cur_new_count == 0) {
1894 qg->rfer -= num_bytes;
1895 qg->rfer_cmpr -= num_bytes;
1899 /* Excl update part */
1900 /* Exclusive/none -> shared case */
1901 if (cur_old_count == nr_old_roots &&
1902 cur_new_count < nr_new_roots) {
1903 /* Exclusive -> shared */
1904 if (cur_old_count != 0) {
1905 qg->excl -= num_bytes;
1906 qg->excl_cmpr -= num_bytes;
1911 /* Shared -> exclusive/none case */
1912 if (cur_old_count < nr_old_roots &&
1913 cur_new_count == nr_new_roots) {
1914 /* Shared->exclusive */
1915 if (cur_new_count != 0) {
1916 qg->excl += num_bytes;
1917 qg->excl_cmpr += num_bytes;
1922 /* Exclusive/none -> exclusive/none case */
1923 if (cur_old_count == nr_old_roots &&
1924 cur_new_count == nr_new_roots) {
1925 if (cur_old_count == 0) {
1926 /* None -> exclusive/none */
1928 if (cur_new_count != 0) {
1929 /* None -> exclusive */
1930 qg->excl += num_bytes;
1931 qg->excl_cmpr += num_bytes;
1934 /* None -> none, nothing changed */
1936 /* Exclusive -> exclusive/none */
1938 if (cur_new_count == 0) {
1939 /* Exclusive -> none */
1940 qg->excl -= num_bytes;
1941 qg->excl_cmpr -= num_bytes;
1944 /* Exclusive -> exclusive, nothing changed */
1949 qgroup_dirty(fs_info, qg);
1955 * Check if the @roots potentially is a list of fs tree roots
1957 * Return 0 for definitely not a fs/subvol tree roots ulist
1958 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
1961 static int maybe_fs_roots(struct ulist *roots)
1963 struct ulist_node *unode;
1964 struct ulist_iterator uiter;
1966 /* Empty one, still possible for fs roots */
1967 if (!roots || roots->nnodes == 0)
1970 ULIST_ITER_INIT(&uiter);
1971 unode = ulist_next(roots, &uiter);
1976 * If it contains fs tree roots, then it must belong to fs/subvol
1978 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
1980 return is_fstree(unode->val);
1984 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
1985 struct btrfs_fs_info *fs_info,
1986 u64 bytenr, u64 num_bytes,
1987 struct ulist *old_roots, struct ulist *new_roots)
1989 struct ulist *qgroups = NULL;
1990 struct ulist *tmp = NULL;
1992 u64 nr_new_roots = 0;
1993 u64 nr_old_roots = 0;
1996 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2000 if (!maybe_fs_roots(new_roots))
2002 nr_new_roots = new_roots->nnodes;
2005 if (!maybe_fs_roots(old_roots))
2007 nr_old_roots = old_roots->nnodes;
2010 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2011 if (nr_old_roots == 0 && nr_new_roots == 0)
2014 BUG_ON(!fs_info->quota_root);
2016 trace_btrfs_qgroup_account_extent(fs_info, bytenr, num_bytes,
2017 nr_old_roots, nr_new_roots);
2019 qgroups = ulist_alloc(GFP_NOFS);
2024 tmp = ulist_alloc(GFP_NOFS);
2030 mutex_lock(&fs_info->qgroup_rescan_lock);
2031 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2032 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2033 mutex_unlock(&fs_info->qgroup_rescan_lock);
2038 mutex_unlock(&fs_info->qgroup_rescan_lock);
2040 spin_lock(&fs_info->qgroup_lock);
2041 seq = fs_info->qgroup_seq;
2043 /* Update old refcnts using old_roots */
2044 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2049 /* Update new refcnts using new_roots */
2050 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2055 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2059 * Bump qgroup_seq to avoid seq overlap
2061 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2063 spin_unlock(&fs_info->qgroup_lock);
2066 ulist_free(qgroups);
2067 ulist_free(old_roots);
2068 ulist_free(new_roots);
2072 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2074 struct btrfs_fs_info *fs_info = trans->fs_info;
2075 struct btrfs_qgroup_extent_record *record;
2076 struct btrfs_delayed_ref_root *delayed_refs;
2077 struct ulist *new_roots = NULL;
2078 struct rb_node *node;
2082 delayed_refs = &trans->transaction->delayed_refs;
2083 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2084 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2085 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2088 trace_btrfs_qgroup_account_extents(fs_info, record);
2092 * Old roots should be searched when inserting qgroup
2095 if (WARN_ON(!record->old_roots)) {
2096 /* Search commit root to find old_roots */
2097 ret = btrfs_find_all_roots(NULL, fs_info,
2099 &record->old_roots, false);
2105 * Use SEQ_LAST as time_seq to do special search, which
2106 * doesn't lock tree or delayed_refs and search current
2107 * root. It's safe inside commit_transaction().
2109 ret = btrfs_find_all_roots(trans, fs_info,
2110 record->bytenr, SEQ_LAST, &new_roots, false);
2113 if (qgroup_to_skip) {
2114 ulist_del(new_roots, qgroup_to_skip, 0);
2115 ulist_del(record->old_roots, qgroup_to_skip,
2118 ret = btrfs_qgroup_account_extent(trans, fs_info,
2119 record->bytenr, record->num_bytes,
2120 record->old_roots, new_roots);
2121 record->old_roots = NULL;
2125 ulist_free(record->old_roots);
2126 ulist_free(new_roots);
2128 rb_erase(node, &delayed_refs->dirty_extent_root);
2136 * called from commit_transaction. Writes all changed qgroups to disk.
2138 int btrfs_run_qgroups(struct btrfs_trans_handle *trans,
2139 struct btrfs_fs_info *fs_info)
2141 struct btrfs_root *quota_root = fs_info->quota_root;
2147 spin_lock(&fs_info->qgroup_lock);
2148 while (!list_empty(&fs_info->dirty_qgroups)) {
2149 struct btrfs_qgroup *qgroup;
2150 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2151 struct btrfs_qgroup, dirty);
2152 list_del_init(&qgroup->dirty);
2153 spin_unlock(&fs_info->qgroup_lock);
2154 ret = update_qgroup_info_item(trans, quota_root, qgroup);
2156 fs_info->qgroup_flags |=
2157 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2158 ret = update_qgroup_limit_item(trans, quota_root, qgroup);
2160 fs_info->qgroup_flags |=
2161 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2162 spin_lock(&fs_info->qgroup_lock);
2164 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2165 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2167 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2168 spin_unlock(&fs_info->qgroup_lock);
2170 ret = update_qgroup_status_item(trans, fs_info, quota_root);
2172 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2178 * Copy the accounting information between qgroups. This is necessary
2179 * when a snapshot or a subvolume is created. Throwing an error will
2180 * cause a transaction abort so we take extra care here to only error
2181 * when a readonly fs is a reasonable outcome.
2183 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
2184 struct btrfs_fs_info *fs_info, u64 srcid, u64 objectid,
2185 struct btrfs_qgroup_inherit *inherit)
2190 struct btrfs_root *quota_root = fs_info->quota_root;
2191 struct btrfs_qgroup *srcgroup;
2192 struct btrfs_qgroup *dstgroup;
2196 mutex_lock(&fs_info->qgroup_ioctl_lock);
2197 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2206 i_qgroups = (u64 *)(inherit + 1);
2207 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2208 2 * inherit->num_excl_copies;
2209 for (i = 0; i < nums; ++i) {
2210 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2213 * Zero out invalid groups so we can ignore
2217 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2225 * create a tracking group for the subvol itself
2227 ret = add_qgroup_item(trans, quota_root, objectid);
2232 struct btrfs_root *srcroot;
2233 struct btrfs_key srckey;
2235 srckey.objectid = srcid;
2236 srckey.type = BTRFS_ROOT_ITEM_KEY;
2237 srckey.offset = (u64)-1;
2238 srcroot = btrfs_read_fs_root_no_name(fs_info, &srckey);
2239 if (IS_ERR(srcroot)) {
2240 ret = PTR_ERR(srcroot);
2244 level_size = fs_info->nodesize;
2248 * add qgroup to all inherited groups
2251 i_qgroups = (u64 *)(inherit + 1);
2252 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2253 if (*i_qgroups == 0)
2255 ret = add_qgroup_relation_item(trans, quota_root,
2256 objectid, *i_qgroups);
2257 if (ret && ret != -EEXIST)
2259 ret = add_qgroup_relation_item(trans, quota_root,
2260 *i_qgroups, objectid);
2261 if (ret && ret != -EEXIST)
2268 spin_lock(&fs_info->qgroup_lock);
2270 dstgroup = add_qgroup_rb(fs_info, objectid);
2271 if (IS_ERR(dstgroup)) {
2272 ret = PTR_ERR(dstgroup);
2276 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2277 dstgroup->lim_flags = inherit->lim.flags;
2278 dstgroup->max_rfer = inherit->lim.max_rfer;
2279 dstgroup->max_excl = inherit->lim.max_excl;
2280 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2281 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2283 ret = update_qgroup_limit_item(trans, quota_root, dstgroup);
2285 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2287 "unable to update quota limit for %llu",
2288 dstgroup->qgroupid);
2294 srcgroup = find_qgroup_rb(fs_info, srcid);
2299 * We call inherit after we clone the root in order to make sure
2300 * our counts don't go crazy, so at this point the only
2301 * difference between the two roots should be the root node.
2303 dstgroup->rfer = srcgroup->rfer;
2304 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2305 dstgroup->excl = level_size;
2306 dstgroup->excl_cmpr = level_size;
2307 srcgroup->excl = level_size;
2308 srcgroup->excl_cmpr = level_size;
2310 /* inherit the limit info */
2311 dstgroup->lim_flags = srcgroup->lim_flags;
2312 dstgroup->max_rfer = srcgroup->max_rfer;
2313 dstgroup->max_excl = srcgroup->max_excl;
2314 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2315 dstgroup->rsv_excl = srcgroup->rsv_excl;
2317 qgroup_dirty(fs_info, dstgroup);
2318 qgroup_dirty(fs_info, srcgroup);
2324 i_qgroups = (u64 *)(inherit + 1);
2325 for (i = 0; i < inherit->num_qgroups; ++i) {
2327 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2334 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2335 struct btrfs_qgroup *src;
2336 struct btrfs_qgroup *dst;
2338 if (!i_qgroups[0] || !i_qgroups[1])
2341 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2342 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2349 dst->rfer = src->rfer - level_size;
2350 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2352 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2353 struct btrfs_qgroup *src;
2354 struct btrfs_qgroup *dst;
2356 if (!i_qgroups[0] || !i_qgroups[1])
2359 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2360 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2367 dst->excl = src->excl + level_size;
2368 dst->excl_cmpr = src->excl_cmpr + level_size;
2372 spin_unlock(&fs_info->qgroup_lock);
2374 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2378 static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
2380 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2381 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2384 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2385 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2391 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2392 enum btrfs_qgroup_rsv_type type)
2394 struct btrfs_root *quota_root;
2395 struct btrfs_qgroup *qgroup;
2396 struct btrfs_fs_info *fs_info = root->fs_info;
2397 u64 ref_root = root->root_key.objectid;
2399 struct ulist_node *unode;
2400 struct ulist_iterator uiter;
2402 if (!is_fstree(ref_root))
2408 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2409 capable(CAP_SYS_RESOURCE))
2412 spin_lock(&fs_info->qgroup_lock);
2413 quota_root = fs_info->quota_root;
2417 qgroup = find_qgroup_rb(fs_info, ref_root);
2422 * in a first step, we check all affected qgroups if any limits would
2425 ulist_reinit(fs_info->qgroup_ulist);
2426 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2427 qgroup_to_aux(qgroup), GFP_ATOMIC);
2430 ULIST_ITER_INIT(&uiter);
2431 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2432 struct btrfs_qgroup *qg;
2433 struct btrfs_qgroup_list *glist;
2435 qg = unode_aux_to_qgroup(unode);
2437 if (enforce && !qgroup_check_limits(qg, num_bytes)) {
2442 list_for_each_entry(glist, &qg->groups, next_group) {
2443 ret = ulist_add(fs_info->qgroup_ulist,
2444 glist->group->qgroupid,
2445 qgroup_to_aux(glist->group), GFP_ATOMIC);
2452 * no limits exceeded, now record the reservation into all qgroups
2454 ULIST_ITER_INIT(&uiter);
2455 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2456 struct btrfs_qgroup *qg;
2458 qg = unode_aux_to_qgroup(unode);
2460 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2461 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2465 spin_unlock(&fs_info->qgroup_lock);
2470 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2473 * Will handle all higher level qgroup too.
2475 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2476 * This special case is only used for META_PERTRANS type.
2478 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2479 u64 ref_root, u64 num_bytes,
2480 enum btrfs_qgroup_rsv_type type)
2482 struct btrfs_root *quota_root;
2483 struct btrfs_qgroup *qgroup;
2484 struct ulist_node *unode;
2485 struct ulist_iterator uiter;
2488 if (!is_fstree(ref_root))
2494 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2495 WARN(1, "%s: Invalid type to free", __func__);
2498 spin_lock(&fs_info->qgroup_lock);
2500 quota_root = fs_info->quota_root;
2504 qgroup = find_qgroup_rb(fs_info, ref_root);
2508 if (num_bytes == (u64)-1)
2510 * We're freeing all pertrans rsv, get reserved value from
2511 * level 0 qgroup as real num_bytes to free.
2513 num_bytes = qgroup->rsv.values[type];
2515 ulist_reinit(fs_info->qgroup_ulist);
2516 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2517 qgroup_to_aux(qgroup), GFP_ATOMIC);
2520 ULIST_ITER_INIT(&uiter);
2521 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2522 struct btrfs_qgroup *qg;
2523 struct btrfs_qgroup_list *glist;
2525 qg = unode_aux_to_qgroup(unode);
2527 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2528 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2530 list_for_each_entry(glist, &qg->groups, next_group) {
2531 ret = ulist_add(fs_info->qgroup_ulist,
2532 glist->group->qgroupid,
2533 qgroup_to_aux(glist->group), GFP_ATOMIC);
2540 spin_unlock(&fs_info->qgroup_lock);
2544 * returns < 0 on error, 0 when more leafs are to be scanned.
2545 * returns 1 when done.
2548 qgroup_rescan_leaf(struct btrfs_fs_info *fs_info, struct btrfs_path *path,
2549 struct btrfs_trans_handle *trans)
2551 struct btrfs_key found;
2552 struct extent_buffer *scratch_leaf = NULL;
2553 struct ulist *roots = NULL;
2554 struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
2559 mutex_lock(&fs_info->qgroup_rescan_lock);
2560 ret = btrfs_search_slot_for_read(fs_info->extent_root,
2561 &fs_info->qgroup_rescan_progress,
2564 btrfs_debug(fs_info,
2565 "current progress key (%llu %u %llu), search_slot ret %d",
2566 fs_info->qgroup_rescan_progress.objectid,
2567 fs_info->qgroup_rescan_progress.type,
2568 fs_info->qgroup_rescan_progress.offset, ret);
2572 * The rescan is about to end, we will not be scanning any
2573 * further blocks. We cannot unset the RESCAN flag here, because
2574 * we want to commit the transaction if everything went well.
2575 * To make the live accounting work in this phase, we set our
2576 * scan progress pointer such that every real extent objectid
2579 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
2580 btrfs_release_path(path);
2581 mutex_unlock(&fs_info->qgroup_rescan_lock);
2585 btrfs_item_key_to_cpu(path->nodes[0], &found,
2586 btrfs_header_nritems(path->nodes[0]) - 1);
2587 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
2589 btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2590 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
2591 if (!scratch_leaf) {
2593 mutex_unlock(&fs_info->qgroup_rescan_lock);
2596 extent_buffer_get(scratch_leaf);
2597 btrfs_tree_read_lock(scratch_leaf);
2598 btrfs_set_lock_blocking_rw(scratch_leaf, BTRFS_READ_LOCK);
2599 slot = path->slots[0];
2600 btrfs_release_path(path);
2601 mutex_unlock(&fs_info->qgroup_rescan_lock);
2603 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
2604 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
2605 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
2606 found.type != BTRFS_METADATA_ITEM_KEY)
2608 if (found.type == BTRFS_METADATA_ITEM_KEY)
2609 num_bytes = fs_info->nodesize;
2611 num_bytes = found.offset;
2613 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
2617 /* For rescan, just pass old_roots as NULL */
2618 ret = btrfs_qgroup_account_extent(trans, fs_info,
2619 found.objectid, num_bytes, NULL, roots);
2625 btrfs_tree_read_unlock_blocking(scratch_leaf);
2626 free_extent_buffer(scratch_leaf);
2628 btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
2633 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
2635 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
2636 qgroup_rescan_work);
2637 struct btrfs_path *path;
2638 struct btrfs_trans_handle *trans = NULL;
2642 path = btrfs_alloc_path();
2647 while (!err && !btrfs_fs_closing(fs_info)) {
2648 trans = btrfs_start_transaction(fs_info->fs_root, 0);
2649 if (IS_ERR(trans)) {
2650 err = PTR_ERR(trans);
2653 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
2656 err = qgroup_rescan_leaf(fs_info, path, trans);
2659 btrfs_commit_transaction(trans);
2661 btrfs_end_transaction(trans);
2665 btrfs_free_path(path);
2667 mutex_lock(&fs_info->qgroup_rescan_lock);
2668 if (!btrfs_fs_closing(fs_info))
2669 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2672 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
2673 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2674 } else if (err < 0) {
2675 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2677 mutex_unlock(&fs_info->qgroup_rescan_lock);
2680 * only update status, since the previous part has already updated the
2683 trans = btrfs_start_transaction(fs_info->quota_root, 1);
2684 if (IS_ERR(trans)) {
2685 err = PTR_ERR(trans);
2687 "fail to start transaction for status update: %d",
2691 ret = update_qgroup_status_item(trans, fs_info, fs_info->quota_root);
2694 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
2696 btrfs_end_transaction(trans);
2698 if (btrfs_fs_closing(fs_info)) {
2699 btrfs_info(fs_info, "qgroup scan paused");
2700 } else if (err >= 0) {
2701 btrfs_info(fs_info, "qgroup scan completed%s",
2702 err > 0 ? " (inconsistency flag cleared)" : "");
2704 btrfs_err(fs_info, "qgroup scan failed with %d", err);
2708 mutex_lock(&fs_info->qgroup_rescan_lock);
2709 fs_info->qgroup_rescan_running = false;
2710 mutex_unlock(&fs_info->qgroup_rescan_lock);
2711 complete_all(&fs_info->qgroup_rescan_completion);
2715 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
2716 * memory required for the rescan context.
2719 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
2725 (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) ||
2726 !(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))) {
2731 mutex_lock(&fs_info->qgroup_rescan_lock);
2732 spin_lock(&fs_info->qgroup_lock);
2735 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2737 else if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
2741 spin_unlock(&fs_info->qgroup_lock);
2742 mutex_unlock(&fs_info->qgroup_rescan_lock);
2745 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2748 memset(&fs_info->qgroup_rescan_progress, 0,
2749 sizeof(fs_info->qgroup_rescan_progress));
2750 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
2751 init_completion(&fs_info->qgroup_rescan_completion);
2752 fs_info->qgroup_rescan_running = true;
2754 spin_unlock(&fs_info->qgroup_lock);
2755 mutex_unlock(&fs_info->qgroup_rescan_lock);
2757 memset(&fs_info->qgroup_rescan_work, 0,
2758 sizeof(fs_info->qgroup_rescan_work));
2759 btrfs_init_work(&fs_info->qgroup_rescan_work,
2760 btrfs_qgroup_rescan_helper,
2761 btrfs_qgroup_rescan_worker, NULL, NULL);
2765 btrfs_info(fs_info, "qgroup_rescan_init failed with %d", ret);
2773 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
2776 struct btrfs_qgroup *qgroup;
2778 spin_lock(&fs_info->qgroup_lock);
2779 /* clear all current qgroup tracking information */
2780 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
2781 qgroup = rb_entry(n, struct btrfs_qgroup, node);
2783 qgroup->rfer_cmpr = 0;
2785 qgroup->excl_cmpr = 0;
2787 spin_unlock(&fs_info->qgroup_lock);
2791 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
2794 struct btrfs_trans_handle *trans;
2796 ret = qgroup_rescan_init(fs_info, 0, 1);
2801 * We have set the rescan_progress to 0, which means no more
2802 * delayed refs will be accounted by btrfs_qgroup_account_ref.
2803 * However, btrfs_qgroup_account_ref may be right after its call
2804 * to btrfs_find_all_roots, in which case it would still do the
2806 * To solve this, we're committing the transaction, which will
2807 * ensure we run all delayed refs and only after that, we are
2808 * going to clear all tracking information for a clean start.
2811 trans = btrfs_join_transaction(fs_info->fs_root);
2812 if (IS_ERR(trans)) {
2813 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2814 return PTR_ERR(trans);
2816 ret = btrfs_commit_transaction(trans);
2818 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
2822 qgroup_rescan_zero_tracking(fs_info);
2824 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2825 &fs_info->qgroup_rescan_work);
2830 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
2836 mutex_lock(&fs_info->qgroup_rescan_lock);
2837 spin_lock(&fs_info->qgroup_lock);
2838 running = fs_info->qgroup_rescan_running;
2839 spin_unlock(&fs_info->qgroup_lock);
2840 mutex_unlock(&fs_info->qgroup_rescan_lock);
2846 ret = wait_for_completion_interruptible(
2847 &fs_info->qgroup_rescan_completion);
2849 wait_for_completion(&fs_info->qgroup_rescan_completion);
2855 * this is only called from open_ctree where we're still single threaded, thus
2856 * locking is omitted here.
2859 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
2861 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
2862 btrfs_queue_work(fs_info->qgroup_rescan_workers,
2863 &fs_info->qgroup_rescan_work);
2867 * Reserve qgroup space for range [start, start + len).
2869 * This function will either reserve space from related qgroups or doing
2870 * nothing if the range is already reserved.
2872 * Return 0 for successful reserve
2873 * Return <0 for error (including -EQUOT)
2875 * NOTE: this function may sleep for memory allocation.
2876 * if btrfs_qgroup_reserve_data() is called multiple times with
2877 * same @reserved, caller must ensure when error happens it's OK
2878 * to free *ALL* reserved space.
2880 int btrfs_qgroup_reserve_data(struct inode *inode,
2881 struct extent_changeset **reserved_ret, u64 start,
2884 struct btrfs_root *root = BTRFS_I(inode)->root;
2885 struct ulist_node *unode;
2886 struct ulist_iterator uiter;
2887 struct extent_changeset *reserved;
2892 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
2893 !is_fstree(root->objectid) || len == 0)
2896 /* @reserved parameter is mandatory for qgroup */
2897 if (WARN_ON(!reserved_ret))
2899 if (!*reserved_ret) {
2900 *reserved_ret = extent_changeset_alloc();
2904 reserved = *reserved_ret;
2905 /* Record already reserved space */
2906 orig_reserved = reserved->bytes_changed;
2907 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
2908 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
2910 /* Newly reserved space */
2911 to_reserve = reserved->bytes_changed - orig_reserved;
2912 trace_btrfs_qgroup_reserve_data(inode, start, len,
2913 to_reserve, QGROUP_RESERVE);
2916 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
2923 /* cleanup *ALL* already reserved ranges */
2924 ULIST_ITER_INIT(&uiter);
2925 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
2926 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
2927 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
2928 extent_changeset_release(reserved);
2932 /* Free ranges specified by @reserved, normally in error path */
2933 static int qgroup_free_reserved_data(struct inode *inode,
2934 struct extent_changeset *reserved, u64 start, u64 len)
2936 struct btrfs_root *root = BTRFS_I(inode)->root;
2937 struct ulist_node *unode;
2938 struct ulist_iterator uiter;
2939 struct extent_changeset changeset;
2943 extent_changeset_init(&changeset);
2944 len = round_up(start + len, root->fs_info->sectorsize);
2945 start = round_down(start, root->fs_info->sectorsize);
2947 ULIST_ITER_INIT(&uiter);
2948 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
2949 u64 range_start = unode->val;
2950 /* unode->aux is the inclusive end */
2951 u64 range_len = unode->aux - range_start + 1;
2955 extent_changeset_release(&changeset);
2957 /* Only free range in range [start, start + len) */
2958 if (range_start >= start + len ||
2959 range_start + range_len <= start)
2961 free_start = max(range_start, start);
2962 free_len = min(start + len, range_start + range_len) -
2965 * TODO: To also modify reserved->ranges_reserved to reflect
2968 * However as long as we free qgroup reserved according to
2969 * EXTENT_QGROUP_RESERVED, we won't double free.
2970 * So not need to rush.
2972 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
2973 free_start, free_start + free_len - 1,
2974 EXTENT_QGROUP_RESERVED, &changeset);
2977 freed += changeset.bytes_changed;
2979 btrfs_qgroup_free_refroot(root->fs_info, root->objectid, freed,
2980 BTRFS_QGROUP_RSV_DATA);
2983 extent_changeset_release(&changeset);
2987 static int __btrfs_qgroup_release_data(struct inode *inode,
2988 struct extent_changeset *reserved, u64 start, u64 len,
2991 struct extent_changeset changeset;
2992 int trace_op = QGROUP_RELEASE;
2995 /* In release case, we shouldn't have @reserved */
2996 WARN_ON(!free && reserved);
2997 if (free && reserved)
2998 return qgroup_free_reserved_data(inode, reserved, start, len);
2999 extent_changeset_init(&changeset);
3000 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3001 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3006 trace_op = QGROUP_FREE;
3007 trace_btrfs_qgroup_release_data(inode, start, len,
3008 changeset.bytes_changed, trace_op);
3010 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3011 BTRFS_I(inode)->root->objectid,
3012 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3013 ret = changeset.bytes_changed;
3015 extent_changeset_release(&changeset);
3020 * Free a reserved space range from io_tree and related qgroups
3022 * Should be called when a range of pages get invalidated before reaching disk.
3023 * Or for error cleanup case.
3024 * if @reserved is given, only reserved range in [@start, @start + @len) will
3027 * For data written to disk, use btrfs_qgroup_release_data().
3029 * NOTE: This function may sleep for memory allocation.
3031 int btrfs_qgroup_free_data(struct inode *inode,
3032 struct extent_changeset *reserved, u64 start, u64 len)
3034 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3038 * Release a reserved space range from io_tree only.
3040 * Should be called when a range of pages get written to disk and corresponding
3041 * FILE_EXTENT is inserted into corresponding root.
3043 * Since new qgroup accounting framework will only update qgroup numbers at
3044 * commit_transaction() time, its reserved space shouldn't be freed from
3047 * But we should release the range from io_tree, to allow further write to be
3050 * NOTE: This function may sleep for memory allocation.
3052 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3054 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3057 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3058 enum btrfs_qgroup_rsv_type type)
3060 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3061 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3066 spin_lock(&root->qgroup_meta_rsv_lock);
3067 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3068 root->qgroup_meta_rsv_prealloc += num_bytes;
3070 root->qgroup_meta_rsv_pertrans += num_bytes;
3071 spin_unlock(&root->qgroup_meta_rsv_lock);
3074 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3075 enum btrfs_qgroup_rsv_type type)
3077 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3078 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3083 spin_lock(&root->qgroup_meta_rsv_lock);
3084 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3085 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3087 root->qgroup_meta_rsv_prealloc -= num_bytes;
3089 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3091 root->qgroup_meta_rsv_pertrans -= num_bytes;
3093 spin_unlock(&root->qgroup_meta_rsv_lock);
3097 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3098 enum btrfs_qgroup_rsv_type type, bool enforce)
3100 struct btrfs_fs_info *fs_info = root->fs_info;
3103 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3104 !is_fstree(root->objectid) || num_bytes == 0)
3107 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3108 trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3109 ret = qgroup_reserve(root, num_bytes, enforce, type);
3113 * Record what we have reserved into root.
3115 * To avoid quota disabled->enabled underflow.
3116 * In that case, we may try to free space we haven't reserved
3117 * (since quota was disabled), so record what we reserved into root.
3118 * And ensure later release won't underflow this number.
3120 add_root_meta_rsv(root, num_bytes, type);
3124 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3126 struct btrfs_fs_info *fs_info = root->fs_info;
3128 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3129 !is_fstree(root->objectid))
3132 /* TODO: Update trace point to handle such free */
3133 trace_qgroup_meta_free_all_pertrans(root);
3134 /* Special value -1 means to free all reserved space */
3135 btrfs_qgroup_free_refroot(fs_info, root->objectid, (u64)-1,
3136 BTRFS_QGROUP_RSV_META_PERTRANS);
3139 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3140 enum btrfs_qgroup_rsv_type type)
3142 struct btrfs_fs_info *fs_info = root->fs_info;
3144 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3145 !is_fstree(root->objectid))
3149 * reservation for META_PREALLOC can happen before quota is enabled,
3150 * which can lead to underflow.
3151 * Here ensure we will only free what we really have reserved.
3153 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3154 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3155 trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3156 btrfs_qgroup_free_refroot(fs_info, root->objectid, num_bytes, type);
3159 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3162 struct btrfs_root *quota_root = fs_info->quota_root;
3163 struct btrfs_qgroup *qgroup;
3164 struct ulist_node *unode;
3165 struct ulist_iterator uiter;
3173 spin_lock(&fs_info->qgroup_lock);
3174 qgroup = find_qgroup_rb(fs_info, ref_root);
3177 ulist_reinit(fs_info->qgroup_ulist);
3178 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3179 qgroup_to_aux(qgroup), GFP_ATOMIC);
3182 ULIST_ITER_INIT(&uiter);
3183 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3184 struct btrfs_qgroup *qg;
3185 struct btrfs_qgroup_list *glist;
3187 qg = unode_aux_to_qgroup(unode);
3189 qgroup_rsv_release(fs_info, qg, num_bytes,
3190 BTRFS_QGROUP_RSV_META_PREALLOC);
3191 qgroup_rsv_add(fs_info, qg, num_bytes,
3192 BTRFS_QGROUP_RSV_META_PERTRANS);
3193 list_for_each_entry(glist, &qg->groups, next_group) {
3194 ret = ulist_add(fs_info->qgroup_ulist,
3195 glist->group->qgroupid,
3196 qgroup_to_aux(glist->group), GFP_ATOMIC);
3202 spin_unlock(&fs_info->qgroup_lock);
3205 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3207 struct btrfs_fs_info *fs_info = root->fs_info;
3209 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3210 !is_fstree(root->objectid))
3212 /* Same as btrfs_qgroup_free_meta_prealloc() */
3213 num_bytes = sub_root_meta_rsv(root, num_bytes,
3214 BTRFS_QGROUP_RSV_META_PREALLOC);
3215 trace_qgroup_meta_convert(root, num_bytes);
3216 qgroup_convert_meta(fs_info, root->objectid, num_bytes);
3220 * Check qgroup reserved space leaking, normally at destroy inode
3223 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3225 struct extent_changeset changeset;
3226 struct ulist_node *unode;
3227 struct ulist_iterator iter;
3230 extent_changeset_init(&changeset);
3231 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3232 EXTENT_QGROUP_RESERVED, &changeset);
3235 if (WARN_ON(changeset.bytes_changed)) {
3236 ULIST_ITER_INIT(&iter);
3237 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3238 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3239 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3240 inode->i_ino, unode->val, unode->aux);
3242 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3243 BTRFS_I(inode)->root->objectid,
3244 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3247 extent_changeset_release(&changeset);