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
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 fs_info->qgroup_flags |= flags;
492 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
493 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
494 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
496 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
497 btrfs_free_path(path);
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 unmounting
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)
897 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
898 if (!fs_info->qgroup_ulist) {
904 * 1 for quota root item
905 * 1 for BTRFS_QGROUP_STATUS item
907 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
908 * per subvolume. However those are not currently reserved since it
909 * would be a lot of overkill.
911 trans = btrfs_start_transaction(tree_root, 2);
913 ret = PTR_ERR(trans);
919 * initially create the quota tree
921 quota_root = btrfs_create_tree(trans, BTRFS_QUOTA_TREE_OBJECTID);
922 if (IS_ERR(quota_root)) {
923 ret = PTR_ERR(quota_root);
924 btrfs_abort_transaction(trans, ret);
928 path = btrfs_alloc_path();
931 btrfs_abort_transaction(trans, ret);
936 key.type = BTRFS_QGROUP_STATUS_KEY;
939 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
942 btrfs_abort_transaction(trans, ret);
946 leaf = path->nodes[0];
947 ptr = btrfs_item_ptr(leaf, path->slots[0],
948 struct btrfs_qgroup_status_item);
949 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
950 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
951 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
952 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
953 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
954 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
956 btrfs_mark_buffer_dirty(leaf);
959 key.type = BTRFS_ROOT_REF_KEY;
962 btrfs_release_path(path);
963 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
967 btrfs_abort_transaction(trans, ret);
972 slot = path->slots[0];
973 leaf = path->nodes[0];
974 btrfs_item_key_to_cpu(leaf, &found_key, slot);
976 if (found_key.type == BTRFS_ROOT_REF_KEY) {
977 ret = add_qgroup_item(trans, quota_root,
980 btrfs_abort_transaction(trans, ret);
984 qgroup = add_qgroup_rb(fs_info, found_key.offset);
985 if (IS_ERR(qgroup)) {
986 ret = PTR_ERR(qgroup);
987 btrfs_abort_transaction(trans, ret);
991 ret = btrfs_next_item(tree_root, path);
993 btrfs_abort_transaction(trans, ret);
1001 btrfs_release_path(path);
1002 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1004 btrfs_abort_transaction(trans, ret);
1008 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1009 if (IS_ERR(qgroup)) {
1010 ret = PTR_ERR(qgroup);
1011 btrfs_abort_transaction(trans, ret);
1015 ret = btrfs_commit_transaction(trans);
1021 * Set quota enabled flag after committing the transaction, to avoid
1022 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1025 spin_lock(&fs_info->qgroup_lock);
1026 fs_info->quota_root = quota_root;
1027 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1028 spin_unlock(&fs_info->qgroup_lock);
1030 ret = qgroup_rescan_init(fs_info, 0, 1);
1032 qgroup_rescan_zero_tracking(fs_info);
1033 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1034 &fs_info->qgroup_rescan_work);
1038 btrfs_free_path(path);
1041 free_extent_buffer(quota_root->node);
1042 free_extent_buffer(quota_root->commit_root);
1047 ulist_free(fs_info->qgroup_ulist);
1048 fs_info->qgroup_ulist = NULL;
1050 btrfs_end_transaction(trans);
1052 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1056 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1058 struct btrfs_root *quota_root;
1059 struct btrfs_trans_handle *trans = NULL;
1062 mutex_lock(&fs_info->qgroup_ioctl_lock);
1063 if (!fs_info->quota_root)
1067 * 1 For the root item
1069 * We should also reserve enough items for the quota tree deletion in
1070 * btrfs_clean_quota_tree but this is not done.
1072 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1073 if (IS_ERR(trans)) {
1074 ret = PTR_ERR(trans);
1078 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1079 btrfs_qgroup_wait_for_completion(fs_info, false);
1080 spin_lock(&fs_info->qgroup_lock);
1081 quota_root = fs_info->quota_root;
1082 fs_info->quota_root = NULL;
1083 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1084 spin_unlock(&fs_info->qgroup_lock);
1086 btrfs_free_qgroup_config(fs_info);
1088 ret = btrfs_clean_quota_tree(trans, quota_root);
1090 btrfs_abort_transaction(trans, ret);
1094 ret = btrfs_del_root(trans, "a_root->root_key);
1096 btrfs_abort_transaction(trans, ret);
1100 list_del("a_root->dirty_list);
1102 btrfs_tree_lock(quota_root->node);
1103 btrfs_clean_tree_block(quota_root->node);
1104 btrfs_tree_unlock(quota_root->node);
1105 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1107 free_extent_buffer(quota_root->node);
1108 free_extent_buffer(quota_root->commit_root);
1112 ret = btrfs_end_transaction(trans);
1114 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1118 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1119 struct btrfs_qgroup *qgroup)
1121 if (list_empty(&qgroup->dirty))
1122 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1126 * The easy accounting, we're updating qgroup relationship whose child qgroup
1127 * only has exclusive extents.
1129 * In this case, all exclusive extents will also be exclusive for parent, so
1130 * excl/rfer just get added/removed.
1132 * So is qgroup reservation space, which should also be added/removed to
1134 * Or when child tries to release reservation space, parent will underflow its
1135 * reservation (for relationship adding case).
1137 * Caller should hold fs_info->qgroup_lock.
1139 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1140 struct ulist *tmp, u64 ref_root,
1141 struct btrfs_qgroup *src, int sign)
1143 struct btrfs_qgroup *qgroup;
1144 struct btrfs_qgroup_list *glist;
1145 struct ulist_node *unode;
1146 struct ulist_iterator uiter;
1147 u64 num_bytes = src->excl;
1150 qgroup = find_qgroup_rb(fs_info, ref_root);
1154 qgroup->rfer += sign * num_bytes;
1155 qgroup->rfer_cmpr += sign * num_bytes;
1157 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1158 qgroup->excl += sign * num_bytes;
1159 qgroup->excl_cmpr += sign * num_bytes;
1162 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1164 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1166 qgroup_dirty(fs_info, qgroup);
1168 /* Get all of the parent groups that contain this qgroup */
1169 list_for_each_entry(glist, &qgroup->groups, next_group) {
1170 ret = ulist_add(tmp, glist->group->qgroupid,
1171 qgroup_to_aux(glist->group), GFP_ATOMIC);
1176 /* Iterate all of the parents and adjust their reference counts */
1177 ULIST_ITER_INIT(&uiter);
1178 while ((unode = ulist_next(tmp, &uiter))) {
1179 qgroup = unode_aux_to_qgroup(unode);
1180 qgroup->rfer += sign * num_bytes;
1181 qgroup->rfer_cmpr += sign * num_bytes;
1182 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1183 qgroup->excl += sign * num_bytes;
1185 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1187 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1188 qgroup->excl_cmpr += sign * num_bytes;
1189 qgroup_dirty(fs_info, qgroup);
1191 /* Add any parents of the parents */
1192 list_for_each_entry(glist, &qgroup->groups, next_group) {
1193 ret = ulist_add(tmp, glist->group->qgroupid,
1194 qgroup_to_aux(glist->group), GFP_ATOMIC);
1206 * Quick path for updating qgroup with only excl refs.
1208 * In that case, just update all parent will be enough.
1209 * Or we needs to do a full rescan.
1210 * Caller should also hold fs_info->qgroup_lock.
1212 * Return 0 for quick update, return >0 for need to full rescan
1213 * and mark INCONSISTENT flag.
1214 * Return < 0 for other error.
1216 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1217 struct ulist *tmp, u64 src, u64 dst,
1220 struct btrfs_qgroup *qgroup;
1224 qgroup = find_qgroup_rb(fs_info, src);
1227 if (qgroup->excl == qgroup->rfer) {
1229 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1238 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1242 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1245 struct btrfs_fs_info *fs_info = trans->fs_info;
1246 struct btrfs_root *quota_root;
1247 struct btrfs_qgroup *parent;
1248 struct btrfs_qgroup *member;
1249 struct btrfs_qgroup_list *list;
1253 /* Check the level of src and dst first */
1254 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1257 tmp = ulist_alloc(GFP_KERNEL);
1261 mutex_lock(&fs_info->qgroup_ioctl_lock);
1262 quota_root = fs_info->quota_root;
1267 member = find_qgroup_rb(fs_info, src);
1268 parent = find_qgroup_rb(fs_info, dst);
1269 if (!member || !parent) {
1274 /* check if such qgroup relation exist firstly */
1275 list_for_each_entry(list, &member->groups, next_group) {
1276 if (list->group == parent) {
1282 ret = add_qgroup_relation_item(trans, src, dst);
1286 ret = add_qgroup_relation_item(trans, dst, src);
1288 del_qgroup_relation_item(trans, src, dst);
1292 spin_lock(&fs_info->qgroup_lock);
1293 ret = add_relation_rb(fs_info, src, dst);
1295 spin_unlock(&fs_info->qgroup_lock);
1298 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1299 spin_unlock(&fs_info->qgroup_lock);
1301 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1306 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1309 struct btrfs_fs_info *fs_info = trans->fs_info;
1310 struct btrfs_root *quota_root;
1311 struct btrfs_qgroup *parent;
1312 struct btrfs_qgroup *member;
1313 struct btrfs_qgroup_list *list;
1318 tmp = ulist_alloc(GFP_KERNEL);
1322 quota_root = fs_info->quota_root;
1328 member = find_qgroup_rb(fs_info, src);
1329 parent = find_qgroup_rb(fs_info, dst);
1330 if (!member || !parent) {
1335 /* check if such qgroup relation exist firstly */
1336 list_for_each_entry(list, &member->groups, next_group) {
1337 if (list->group == parent)
1343 ret = del_qgroup_relation_item(trans, src, dst);
1344 err = del_qgroup_relation_item(trans, dst, src);
1348 spin_lock(&fs_info->qgroup_lock);
1349 del_relation_rb(fs_info, src, dst);
1350 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1351 spin_unlock(&fs_info->qgroup_lock);
1357 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1360 struct btrfs_fs_info *fs_info = trans->fs_info;
1363 mutex_lock(&fs_info->qgroup_ioctl_lock);
1364 ret = __del_qgroup_relation(trans, src, dst);
1365 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1370 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1372 struct btrfs_fs_info *fs_info = trans->fs_info;
1373 struct btrfs_root *quota_root;
1374 struct btrfs_qgroup *qgroup;
1377 mutex_lock(&fs_info->qgroup_ioctl_lock);
1378 quota_root = fs_info->quota_root;
1383 qgroup = find_qgroup_rb(fs_info, qgroupid);
1389 ret = add_qgroup_item(trans, quota_root, qgroupid);
1393 spin_lock(&fs_info->qgroup_lock);
1394 qgroup = add_qgroup_rb(fs_info, qgroupid);
1395 spin_unlock(&fs_info->qgroup_lock);
1398 ret = PTR_ERR(qgroup);
1400 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1404 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1406 struct btrfs_fs_info *fs_info = trans->fs_info;
1407 struct btrfs_root *quota_root;
1408 struct btrfs_qgroup *qgroup;
1409 struct btrfs_qgroup_list *list;
1412 mutex_lock(&fs_info->qgroup_ioctl_lock);
1413 quota_root = fs_info->quota_root;
1419 qgroup = find_qgroup_rb(fs_info, qgroupid);
1425 /* Check if there are no children of this qgroup */
1426 if (!list_empty(&qgroup->members)) {
1431 ret = del_qgroup_item(trans, qgroupid);
1432 if (ret && ret != -ENOENT)
1435 while (!list_empty(&qgroup->groups)) {
1436 list = list_first_entry(&qgroup->groups,
1437 struct btrfs_qgroup_list, next_group);
1438 ret = __del_qgroup_relation(trans, qgroupid,
1439 list->group->qgroupid);
1444 spin_lock(&fs_info->qgroup_lock);
1445 del_qgroup_rb(fs_info, qgroupid);
1446 spin_unlock(&fs_info->qgroup_lock);
1448 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1452 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1453 struct btrfs_qgroup_limit *limit)
1455 struct btrfs_fs_info *fs_info = trans->fs_info;
1456 struct btrfs_root *quota_root;
1457 struct btrfs_qgroup *qgroup;
1459 /* Sometimes we would want to clear the limit on this qgroup.
1460 * To meet this requirement, we treat the -1 as a special value
1461 * which tell kernel to clear the limit on this qgroup.
1463 const u64 CLEAR_VALUE = -1;
1465 mutex_lock(&fs_info->qgroup_ioctl_lock);
1466 quota_root = fs_info->quota_root;
1472 qgroup = find_qgroup_rb(fs_info, qgroupid);
1478 spin_lock(&fs_info->qgroup_lock);
1479 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1480 if (limit->max_rfer == CLEAR_VALUE) {
1481 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1482 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1483 qgroup->max_rfer = 0;
1485 qgroup->max_rfer = limit->max_rfer;
1488 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1489 if (limit->max_excl == CLEAR_VALUE) {
1490 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1491 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1492 qgroup->max_excl = 0;
1494 qgroup->max_excl = limit->max_excl;
1497 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1498 if (limit->rsv_rfer == CLEAR_VALUE) {
1499 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1500 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1501 qgroup->rsv_rfer = 0;
1503 qgroup->rsv_rfer = limit->rsv_rfer;
1506 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1507 if (limit->rsv_excl == CLEAR_VALUE) {
1508 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1509 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1510 qgroup->rsv_excl = 0;
1512 qgroup->rsv_excl = limit->rsv_excl;
1515 qgroup->lim_flags |= limit->flags;
1517 spin_unlock(&fs_info->qgroup_lock);
1519 ret = update_qgroup_limit_item(trans, qgroup);
1521 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1522 btrfs_info(fs_info, "unable to update quota limit for %llu",
1527 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1531 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1532 struct btrfs_delayed_ref_root *delayed_refs,
1533 struct btrfs_qgroup_extent_record *record)
1535 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1536 struct rb_node *parent_node = NULL;
1537 struct btrfs_qgroup_extent_record *entry;
1538 u64 bytenr = record->bytenr;
1540 lockdep_assert_held(&delayed_refs->lock);
1541 trace_btrfs_qgroup_trace_extent(fs_info, record);
1545 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1547 if (bytenr < entry->bytenr) {
1549 } else if (bytenr > entry->bytenr) {
1550 p = &(*p)->rb_right;
1552 if (record->data_rsv && !entry->data_rsv) {
1553 entry->data_rsv = record->data_rsv;
1554 entry->data_rsv_refroot =
1555 record->data_rsv_refroot;
1561 rb_link_node(&record->node, parent_node, p);
1562 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1566 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1567 struct btrfs_qgroup_extent_record *qrecord)
1569 struct ulist *old_root;
1570 u64 bytenr = qrecord->bytenr;
1573 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1575 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1577 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1583 * Here we don't need to get the lock of
1584 * trans->transaction->delayed_refs, since inserted qrecord won't
1585 * be deleted, only qrecord->node may be modified (new qrecord insert)
1587 * So modifying qrecord->old_roots is safe here
1589 qrecord->old_roots = old_root;
1593 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1594 u64 num_bytes, gfp_t gfp_flag)
1596 struct btrfs_fs_info *fs_info = trans->fs_info;
1597 struct btrfs_qgroup_extent_record *record;
1598 struct btrfs_delayed_ref_root *delayed_refs;
1601 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1602 || bytenr == 0 || num_bytes == 0)
1604 record = kzalloc(sizeof(*record), gfp_flag);
1608 delayed_refs = &trans->transaction->delayed_refs;
1609 record->bytenr = bytenr;
1610 record->num_bytes = num_bytes;
1611 record->old_roots = NULL;
1613 spin_lock(&delayed_refs->lock);
1614 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1615 spin_unlock(&delayed_refs->lock);
1620 return btrfs_qgroup_trace_extent_post(fs_info, record);
1623 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1624 struct extent_buffer *eb)
1626 struct btrfs_fs_info *fs_info = trans->fs_info;
1627 int nr = btrfs_header_nritems(eb);
1628 int i, extent_type, ret;
1629 struct btrfs_key key;
1630 struct btrfs_file_extent_item *fi;
1631 u64 bytenr, num_bytes;
1633 /* We can be called directly from walk_up_proc() */
1634 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1637 for (i = 0; i < nr; i++) {
1638 btrfs_item_key_to_cpu(eb, &key, i);
1640 if (key.type != BTRFS_EXTENT_DATA_KEY)
1643 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1644 /* filter out non qgroup-accountable extents */
1645 extent_type = btrfs_file_extent_type(eb, fi);
1647 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1650 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1654 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1656 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1666 * Walk up the tree from the bottom, freeing leaves and any interior
1667 * nodes which have had all slots visited. If a node (leaf or
1668 * interior) is freed, the node above it will have it's slot
1669 * incremented. The root node will never be freed.
1671 * At the end of this function, we should have a path which has all
1672 * slots incremented to the next position for a search. If we need to
1673 * read a new node it will be NULL and the node above it will have the
1674 * correct slot selected for a later read.
1676 * If we increment the root nodes slot counter past the number of
1677 * elements, 1 is returned to signal completion of the search.
1679 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1683 struct extent_buffer *eb;
1685 if (root_level == 0)
1688 while (level <= root_level) {
1689 eb = path->nodes[level];
1690 nr = btrfs_header_nritems(eb);
1691 path->slots[level]++;
1692 slot = path->slots[level];
1693 if (slot >= nr || level == 0) {
1695 * Don't free the root - we will detect this
1696 * condition after our loop and return a
1697 * positive value for caller to stop walking the tree.
1699 if (level != root_level) {
1700 btrfs_tree_unlock_rw(eb, path->locks[level]);
1701 path->locks[level] = 0;
1703 free_extent_buffer(eb);
1704 path->nodes[level] = NULL;
1705 path->slots[level] = 0;
1709 * We have a valid slot to walk back down
1710 * from. Stop here so caller can process these
1719 eb = path->nodes[root_level];
1720 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1727 * Helper function to trace a subtree tree block swap.
1729 * The swap will happen in highest tree block, but there may be a lot of
1730 * tree blocks involved.
1733 * OO = Old tree blocks
1734 * NN = New tree blocks allocated during balance
1736 * File tree (257) Reloc tree for 257
1739 * L1 OO OO (a) OO NN (a)
1741 * L0 OO OO OO OO OO OO NN NN
1744 * When calling qgroup_trace_extent_swap(), we will pass:
1746 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1750 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1751 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1753 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1755 * 1) Tree search from @src_eb
1756 * It should acts as a simplified btrfs_search_slot().
1757 * The key for search can be extracted from @dst_path->nodes[dst_level]
1760 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1761 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1762 * They should be marked during previous (@dst_level = 1) iteration.
1764 * 3) Mark file extents in leaves dirty
1765 * We don't have good way to pick out new file extents only.
1766 * So we still follow the old method by scanning all file extents in
1769 * This function can free us from keeping two paths, thus later we only need
1770 * to care about how to iterate all new tree blocks in reloc tree.
1772 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1773 struct extent_buffer *src_eb,
1774 struct btrfs_path *dst_path,
1775 int dst_level, int root_level,
1778 struct btrfs_key key;
1779 struct btrfs_path *src_path;
1780 struct btrfs_fs_info *fs_info = trans->fs_info;
1781 u32 nodesize = fs_info->nodesize;
1782 int cur_level = root_level;
1785 BUG_ON(dst_level > root_level);
1786 /* Level mismatch */
1787 if (btrfs_header_level(src_eb) != root_level)
1790 src_path = btrfs_alloc_path();
1797 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1799 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1802 extent_buffer_get(src_eb);
1803 src_path->nodes[root_level] = src_eb;
1804 src_path->slots[root_level] = dst_path->slots[root_level];
1805 src_path->locks[root_level] = 0;
1807 /* A simplified version of btrfs_search_slot() */
1808 while (cur_level >= dst_level) {
1809 struct btrfs_key src_key;
1810 struct btrfs_key dst_key;
1812 if (src_path->nodes[cur_level] == NULL) {
1813 struct btrfs_key first_key;
1814 struct extent_buffer *eb;
1819 eb = src_path->nodes[cur_level + 1];
1820 parent_slot = src_path->slots[cur_level + 1];
1821 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1822 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1823 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1825 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1826 cur_level, &first_key);
1830 } else if (!extent_buffer_uptodate(eb)) {
1831 free_extent_buffer(eb);
1836 src_path->nodes[cur_level] = eb;
1838 btrfs_tree_read_lock(eb);
1839 btrfs_set_lock_blocking_read(eb);
1840 src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1843 src_path->slots[cur_level] = dst_path->slots[cur_level];
1845 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1846 &dst_key, dst_path->slots[cur_level]);
1847 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1848 &src_key, src_path->slots[cur_level]);
1850 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1851 &dst_key, dst_path->slots[cur_level]);
1852 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1853 &src_key, src_path->slots[cur_level]);
1855 /* Content mismatch, something went wrong */
1856 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1864 * Now both @dst_path and @src_path have been populated, record the tree
1865 * blocks for qgroup accounting.
1867 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1868 nodesize, GFP_NOFS);
1871 ret = btrfs_qgroup_trace_extent(trans,
1872 dst_path->nodes[dst_level]->start,
1873 nodesize, GFP_NOFS);
1877 /* Record leaf file extents */
1878 if (dst_level == 0 && trace_leaf) {
1879 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1882 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1885 btrfs_free_path(src_path);
1890 * Helper function to do recursive generation-aware depth-first search, to
1891 * locate all new tree blocks in a subtree of reloc tree.
1893 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1902 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1906 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1907 * above tree blocks along with their counter parts in file tree.
1908 * While during search, old tree blocks OO(c) will be skipped as tree block swap
1909 * won't affect OO(c).
1911 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1912 struct extent_buffer *src_eb,
1913 struct btrfs_path *dst_path,
1914 int cur_level, int root_level,
1915 u64 last_snapshot, bool trace_leaf)
1917 struct btrfs_fs_info *fs_info = trans->fs_info;
1918 struct extent_buffer *eb;
1919 bool need_cleanup = false;
1923 /* Level sanity check */
1924 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1925 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
1926 root_level < cur_level) {
1927 btrfs_err_rl(fs_info,
1928 "%s: bad levels, cur_level=%d root_level=%d",
1929 __func__, cur_level, root_level);
1933 /* Read the tree block if needed */
1934 if (dst_path->nodes[cur_level] == NULL) {
1935 struct btrfs_key first_key;
1941 * dst_path->nodes[root_level] must be initialized before
1942 * calling this function.
1944 if (cur_level == root_level) {
1945 btrfs_err_rl(fs_info,
1946 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1947 __func__, root_level, root_level, cur_level);
1952 * We need to get child blockptr/gen from parent before we can
1955 eb = dst_path->nodes[cur_level + 1];
1956 parent_slot = dst_path->slots[cur_level + 1];
1957 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1958 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1959 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1961 /* This node is old, no need to trace */
1962 if (child_gen < last_snapshot)
1965 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1966 cur_level, &first_key);
1970 } else if (!extent_buffer_uptodate(eb)) {
1971 free_extent_buffer(eb);
1976 dst_path->nodes[cur_level] = eb;
1977 dst_path->slots[cur_level] = 0;
1979 btrfs_tree_read_lock(eb);
1980 btrfs_set_lock_blocking_read(eb);
1981 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1982 need_cleanup = true;
1985 /* Now record this tree block and its counter part for qgroups */
1986 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1987 root_level, trace_leaf);
1991 eb = dst_path->nodes[cur_level];
1993 if (cur_level > 0) {
1994 /* Iterate all child tree blocks */
1995 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1996 /* Skip old tree blocks as they won't be swapped */
1997 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
1999 dst_path->slots[cur_level] = i;
2001 /* Recursive call (at most 7 times) */
2002 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2003 dst_path, cur_level - 1, root_level,
2004 last_snapshot, trace_leaf);
2013 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2014 dst_path->locks[cur_level]);
2015 free_extent_buffer(dst_path->nodes[cur_level]);
2016 dst_path->nodes[cur_level] = NULL;
2017 dst_path->slots[cur_level] = 0;
2018 dst_path->locks[cur_level] = 0;
2024 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2025 struct extent_buffer *src_eb,
2026 struct extent_buffer *dst_eb,
2027 u64 last_snapshot, bool trace_leaf)
2029 struct btrfs_fs_info *fs_info = trans->fs_info;
2030 struct btrfs_path *dst_path = NULL;
2034 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2037 /* Wrong parameter order */
2038 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2039 btrfs_err_rl(fs_info,
2040 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2041 btrfs_header_generation(src_eb),
2042 btrfs_header_generation(dst_eb));
2046 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2051 level = btrfs_header_level(dst_eb);
2052 dst_path = btrfs_alloc_path();
2058 extent_buffer_get(dst_eb);
2059 dst_path->nodes[level] = dst_eb;
2060 dst_path->slots[level] = 0;
2061 dst_path->locks[level] = 0;
2063 /* Do the generation aware breadth-first search */
2064 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2065 level, last_snapshot, trace_leaf);
2071 btrfs_free_path(dst_path);
2073 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2077 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2078 struct extent_buffer *root_eb,
2079 u64 root_gen, int root_level)
2081 struct btrfs_fs_info *fs_info = trans->fs_info;
2084 struct extent_buffer *eb = root_eb;
2085 struct btrfs_path *path = NULL;
2087 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2088 BUG_ON(root_eb == NULL);
2090 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2093 if (!extent_buffer_uptodate(root_eb)) {
2094 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2099 if (root_level == 0) {
2100 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2104 path = btrfs_alloc_path();
2109 * Walk down the tree. Missing extent blocks are filled in as
2110 * we go. Metadata is accounted every time we read a new
2113 * When we reach a leaf, we account for file extent items in it,
2114 * walk back up the tree (adjusting slot pointers as we go)
2115 * and restart the search process.
2117 extent_buffer_get(root_eb); /* For path */
2118 path->nodes[root_level] = root_eb;
2119 path->slots[root_level] = 0;
2120 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2123 while (level >= 0) {
2124 if (path->nodes[level] == NULL) {
2125 struct btrfs_key first_key;
2131 * We need to get child blockptr/gen from parent before
2134 eb = path->nodes[level + 1];
2135 parent_slot = path->slots[level + 1];
2136 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2137 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2138 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2140 eb = read_tree_block(fs_info, child_bytenr, child_gen,
2145 } else if (!extent_buffer_uptodate(eb)) {
2146 free_extent_buffer(eb);
2151 path->nodes[level] = eb;
2152 path->slots[level] = 0;
2154 btrfs_tree_read_lock(eb);
2155 btrfs_set_lock_blocking_read(eb);
2156 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2158 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2166 ret = btrfs_qgroup_trace_leaf_items(trans,
2167 path->nodes[level]);
2171 /* Nonzero return here means we completed our search */
2172 ret = adjust_slots_upwards(path, root_level);
2176 /* Restart search with new slots */
2185 btrfs_free_path(path);
2190 #define UPDATE_NEW 0
2191 #define UPDATE_OLD 1
2193 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2195 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2196 struct ulist *roots, struct ulist *tmp,
2197 struct ulist *qgroups, u64 seq, int update_old)
2199 struct ulist_node *unode;
2200 struct ulist_iterator uiter;
2201 struct ulist_node *tmp_unode;
2202 struct ulist_iterator tmp_uiter;
2203 struct btrfs_qgroup *qg;
2208 ULIST_ITER_INIT(&uiter);
2209 while ((unode = ulist_next(roots, &uiter))) {
2210 qg = find_qgroup_rb(fs_info, unode->val);
2215 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2219 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2222 ULIST_ITER_INIT(&tmp_uiter);
2223 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2224 struct btrfs_qgroup_list *glist;
2226 qg = unode_aux_to_qgroup(tmp_unode);
2228 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2230 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2231 list_for_each_entry(glist, &qg->groups, next_group) {
2232 ret = ulist_add(qgroups, glist->group->qgroupid,
2233 qgroup_to_aux(glist->group),
2237 ret = ulist_add(tmp, glist->group->qgroupid,
2238 qgroup_to_aux(glist->group),
2249 * Update qgroup rfer/excl counters.
2250 * Rfer update is easy, codes can explain themselves.
2252 * Excl update is tricky, the update is split into 2 part.
2253 * Part 1: Possible exclusive <-> sharing detect:
2255 * -------------------------------------
2257 * -------------------------------------
2259 * -------------------------------------
2262 * A: cur_old_roots < nr_old_roots (not exclusive before)
2263 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2264 * B: cur_new_roots < nr_new_roots (not exclusive now)
2265 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2268 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2269 * *: Definitely not changed. **: Possible unchanged.
2271 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2273 * To make the logic clear, we first use condition A and B to split
2274 * combination into 4 results.
2276 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2277 * only on variant maybe 0.
2279 * Lastly, check result **, since there are 2 variants maybe 0, split them
2281 * But this time we don't need to consider other things, the codes and logic
2282 * is easy to understand now.
2284 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2285 struct ulist *qgroups,
2288 u64 num_bytes, u64 seq)
2290 struct ulist_node *unode;
2291 struct ulist_iterator uiter;
2292 struct btrfs_qgroup *qg;
2293 u64 cur_new_count, cur_old_count;
2295 ULIST_ITER_INIT(&uiter);
2296 while ((unode = ulist_next(qgroups, &uiter))) {
2299 qg = unode_aux_to_qgroup(unode);
2300 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2301 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2303 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2306 /* Rfer update part */
2307 if (cur_old_count == 0 && cur_new_count > 0) {
2308 qg->rfer += num_bytes;
2309 qg->rfer_cmpr += num_bytes;
2312 if (cur_old_count > 0 && cur_new_count == 0) {
2313 qg->rfer -= num_bytes;
2314 qg->rfer_cmpr -= num_bytes;
2318 /* Excl update part */
2319 /* Exclusive/none -> shared case */
2320 if (cur_old_count == nr_old_roots &&
2321 cur_new_count < nr_new_roots) {
2322 /* Exclusive -> shared */
2323 if (cur_old_count != 0) {
2324 qg->excl -= num_bytes;
2325 qg->excl_cmpr -= num_bytes;
2330 /* Shared -> exclusive/none case */
2331 if (cur_old_count < nr_old_roots &&
2332 cur_new_count == nr_new_roots) {
2333 /* Shared->exclusive */
2334 if (cur_new_count != 0) {
2335 qg->excl += num_bytes;
2336 qg->excl_cmpr += num_bytes;
2341 /* Exclusive/none -> exclusive/none case */
2342 if (cur_old_count == nr_old_roots &&
2343 cur_new_count == nr_new_roots) {
2344 if (cur_old_count == 0) {
2345 /* None -> exclusive/none */
2347 if (cur_new_count != 0) {
2348 /* None -> exclusive */
2349 qg->excl += num_bytes;
2350 qg->excl_cmpr += num_bytes;
2353 /* None -> none, nothing changed */
2355 /* Exclusive -> exclusive/none */
2357 if (cur_new_count == 0) {
2358 /* Exclusive -> none */
2359 qg->excl -= num_bytes;
2360 qg->excl_cmpr -= num_bytes;
2363 /* Exclusive -> exclusive, nothing changed */
2368 qgroup_dirty(fs_info, qg);
2374 * Check if the @roots potentially is a list of fs tree roots
2376 * Return 0 for definitely not a fs/subvol tree roots ulist
2377 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2380 static int maybe_fs_roots(struct ulist *roots)
2382 struct ulist_node *unode;
2383 struct ulist_iterator uiter;
2385 /* Empty one, still possible for fs roots */
2386 if (!roots || roots->nnodes == 0)
2389 ULIST_ITER_INIT(&uiter);
2390 unode = ulist_next(roots, &uiter);
2395 * If it contains fs tree roots, then it must belong to fs/subvol
2397 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2399 return is_fstree(unode->val);
2402 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2403 u64 num_bytes, struct ulist *old_roots,
2404 struct ulist *new_roots)
2406 struct btrfs_fs_info *fs_info = trans->fs_info;
2407 struct ulist *qgroups = NULL;
2408 struct ulist *tmp = NULL;
2410 u64 nr_new_roots = 0;
2411 u64 nr_old_roots = 0;
2414 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2418 if (!maybe_fs_roots(new_roots))
2420 nr_new_roots = new_roots->nnodes;
2423 if (!maybe_fs_roots(old_roots))
2425 nr_old_roots = old_roots->nnodes;
2428 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2429 if (nr_old_roots == 0 && nr_new_roots == 0)
2432 BUG_ON(!fs_info->quota_root);
2434 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2435 num_bytes, nr_old_roots, nr_new_roots);
2437 qgroups = ulist_alloc(GFP_NOFS);
2442 tmp = ulist_alloc(GFP_NOFS);
2448 mutex_lock(&fs_info->qgroup_rescan_lock);
2449 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2450 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2451 mutex_unlock(&fs_info->qgroup_rescan_lock);
2456 mutex_unlock(&fs_info->qgroup_rescan_lock);
2458 spin_lock(&fs_info->qgroup_lock);
2459 seq = fs_info->qgroup_seq;
2461 /* Update old refcnts using old_roots */
2462 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2467 /* Update new refcnts using new_roots */
2468 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2473 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2477 * Bump qgroup_seq to avoid seq overlap
2479 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2481 spin_unlock(&fs_info->qgroup_lock);
2484 ulist_free(qgroups);
2485 ulist_free(old_roots);
2486 ulist_free(new_roots);
2490 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2492 struct btrfs_fs_info *fs_info = trans->fs_info;
2493 struct btrfs_qgroup_extent_record *record;
2494 struct btrfs_delayed_ref_root *delayed_refs;
2495 struct ulist *new_roots = NULL;
2496 struct rb_node *node;
2497 u64 num_dirty_extents = 0;
2501 delayed_refs = &trans->transaction->delayed_refs;
2502 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2503 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2504 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2507 num_dirty_extents++;
2508 trace_btrfs_qgroup_account_extents(fs_info, record);
2512 * Old roots should be searched when inserting qgroup
2515 if (WARN_ON(!record->old_roots)) {
2516 /* Search commit root to find old_roots */
2517 ret = btrfs_find_all_roots(NULL, fs_info,
2519 &record->old_roots, false);
2524 /* Free the reserved data space */
2525 btrfs_qgroup_free_refroot(fs_info,
2526 record->data_rsv_refroot,
2528 BTRFS_QGROUP_RSV_DATA);
2530 * Use SEQ_LAST as time_seq to do special search, which
2531 * doesn't lock tree or delayed_refs and search current
2532 * root. It's safe inside commit_transaction().
2534 ret = btrfs_find_all_roots(trans, fs_info,
2535 record->bytenr, SEQ_LAST, &new_roots, false);
2538 if (qgroup_to_skip) {
2539 ulist_del(new_roots, qgroup_to_skip, 0);
2540 ulist_del(record->old_roots, qgroup_to_skip,
2543 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2547 record->old_roots = NULL;
2551 ulist_free(record->old_roots);
2552 ulist_free(new_roots);
2554 rb_erase(node, &delayed_refs->dirty_extent_root);
2558 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2564 * called from commit_transaction. Writes all changed qgroups to disk.
2566 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2568 struct btrfs_fs_info *fs_info = trans->fs_info;
2569 struct btrfs_root *quota_root = fs_info->quota_root;
2575 spin_lock(&fs_info->qgroup_lock);
2576 while (!list_empty(&fs_info->dirty_qgroups)) {
2577 struct btrfs_qgroup *qgroup;
2578 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2579 struct btrfs_qgroup, dirty);
2580 list_del_init(&qgroup->dirty);
2581 spin_unlock(&fs_info->qgroup_lock);
2582 ret = update_qgroup_info_item(trans, qgroup);
2584 fs_info->qgroup_flags |=
2585 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2586 ret = update_qgroup_limit_item(trans, qgroup);
2588 fs_info->qgroup_flags |=
2589 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2590 spin_lock(&fs_info->qgroup_lock);
2592 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2593 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2595 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2596 spin_unlock(&fs_info->qgroup_lock);
2598 ret = update_qgroup_status_item(trans);
2600 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2606 * Copy the accounting information between qgroups. This is necessary
2607 * when a snapshot or a subvolume is created. Throwing an error will
2608 * cause a transaction abort so we take extra care here to only error
2609 * when a readonly fs is a reasonable outcome.
2611 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2612 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2617 struct btrfs_fs_info *fs_info = trans->fs_info;
2618 struct btrfs_root *quota_root;
2619 struct btrfs_qgroup *srcgroup;
2620 struct btrfs_qgroup *dstgroup;
2624 mutex_lock(&fs_info->qgroup_ioctl_lock);
2625 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2628 quota_root = fs_info->quota_root;
2635 i_qgroups = (u64 *)(inherit + 1);
2636 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2637 2 * inherit->num_excl_copies;
2638 for (i = 0; i < nums; ++i) {
2639 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2642 * Zero out invalid groups so we can ignore
2646 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2654 * create a tracking group for the subvol itself
2656 ret = add_qgroup_item(trans, quota_root, objectid);
2661 * add qgroup to all inherited groups
2664 i_qgroups = (u64 *)(inherit + 1);
2665 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2666 if (*i_qgroups == 0)
2668 ret = add_qgroup_relation_item(trans, objectid,
2670 if (ret && ret != -EEXIST)
2672 ret = add_qgroup_relation_item(trans, *i_qgroups,
2674 if (ret && ret != -EEXIST)
2681 spin_lock(&fs_info->qgroup_lock);
2683 dstgroup = add_qgroup_rb(fs_info, objectid);
2684 if (IS_ERR(dstgroup)) {
2685 ret = PTR_ERR(dstgroup);
2689 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2690 dstgroup->lim_flags = inherit->lim.flags;
2691 dstgroup->max_rfer = inherit->lim.max_rfer;
2692 dstgroup->max_excl = inherit->lim.max_excl;
2693 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2694 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2696 ret = update_qgroup_limit_item(trans, dstgroup);
2698 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2700 "unable to update quota limit for %llu",
2701 dstgroup->qgroupid);
2707 srcgroup = find_qgroup_rb(fs_info, srcid);
2712 * We call inherit after we clone the root in order to make sure
2713 * our counts don't go crazy, so at this point the only
2714 * difference between the two roots should be the root node.
2716 level_size = fs_info->nodesize;
2717 dstgroup->rfer = srcgroup->rfer;
2718 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2719 dstgroup->excl = level_size;
2720 dstgroup->excl_cmpr = level_size;
2721 srcgroup->excl = level_size;
2722 srcgroup->excl_cmpr = level_size;
2724 /* inherit the limit info */
2725 dstgroup->lim_flags = srcgroup->lim_flags;
2726 dstgroup->max_rfer = srcgroup->max_rfer;
2727 dstgroup->max_excl = srcgroup->max_excl;
2728 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2729 dstgroup->rsv_excl = srcgroup->rsv_excl;
2731 qgroup_dirty(fs_info, dstgroup);
2732 qgroup_dirty(fs_info, srcgroup);
2738 i_qgroups = (u64 *)(inherit + 1);
2739 for (i = 0; i < inherit->num_qgroups; ++i) {
2741 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2748 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2749 struct btrfs_qgroup *src;
2750 struct btrfs_qgroup *dst;
2752 if (!i_qgroups[0] || !i_qgroups[1])
2755 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2756 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2763 dst->rfer = src->rfer - level_size;
2764 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2766 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2767 struct btrfs_qgroup *src;
2768 struct btrfs_qgroup *dst;
2770 if (!i_qgroups[0] || !i_qgroups[1])
2773 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2774 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2781 dst->excl = src->excl + level_size;
2782 dst->excl_cmpr = src->excl_cmpr + level_size;
2786 spin_unlock(&fs_info->qgroup_lock);
2788 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2793 * Two limits to commit transaction in advance.
2795 * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
2796 * For SIZE, it will be in byte unit as threshold.
2798 #define QGROUP_FREE_RATIO 32
2799 #define QGROUP_FREE_SIZE SZ_32M
2800 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2801 const struct btrfs_qgroup *qg, u64 num_bytes)
2806 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2807 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2810 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2811 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2815 * Even if we passed the check, it's better to check if reservation
2816 * for meta_pertrans is pushing us near limit.
2817 * If there is too much pertrans reservation or it's near the limit,
2818 * let's try commit transaction to free some, using transaction_kthread
2820 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2821 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2822 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2823 free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2824 threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2827 free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2828 threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2833 * Use transaction_kthread to commit transaction, so we no
2834 * longer need to bother nested transaction nor lock context.
2836 if (free < threshold)
2837 btrfs_commit_transaction_locksafe(fs_info);
2843 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2844 enum btrfs_qgroup_rsv_type type)
2846 struct btrfs_root *quota_root;
2847 struct btrfs_qgroup *qgroup;
2848 struct btrfs_fs_info *fs_info = root->fs_info;
2849 u64 ref_root = root->root_key.objectid;
2851 struct ulist_node *unode;
2852 struct ulist_iterator uiter;
2854 if (!is_fstree(ref_root))
2860 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2861 capable(CAP_SYS_RESOURCE))
2864 spin_lock(&fs_info->qgroup_lock);
2865 quota_root = fs_info->quota_root;
2869 qgroup = find_qgroup_rb(fs_info, ref_root);
2874 * in a first step, we check all affected qgroups if any limits would
2877 ulist_reinit(fs_info->qgroup_ulist);
2878 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2879 qgroup_to_aux(qgroup), GFP_ATOMIC);
2882 ULIST_ITER_INIT(&uiter);
2883 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2884 struct btrfs_qgroup *qg;
2885 struct btrfs_qgroup_list *glist;
2887 qg = unode_aux_to_qgroup(unode);
2889 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2894 list_for_each_entry(glist, &qg->groups, next_group) {
2895 ret = ulist_add(fs_info->qgroup_ulist,
2896 glist->group->qgroupid,
2897 qgroup_to_aux(glist->group), GFP_ATOMIC);
2904 * no limits exceeded, now record the reservation into all qgroups
2906 ULIST_ITER_INIT(&uiter);
2907 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2908 struct btrfs_qgroup *qg;
2910 qg = unode_aux_to_qgroup(unode);
2912 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2916 spin_unlock(&fs_info->qgroup_lock);
2921 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2924 * Will handle all higher level qgroup too.
2926 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2927 * This special case is only used for META_PERTRANS type.
2929 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2930 u64 ref_root, u64 num_bytes,
2931 enum btrfs_qgroup_rsv_type type)
2933 struct btrfs_root *quota_root;
2934 struct btrfs_qgroup *qgroup;
2935 struct ulist_node *unode;
2936 struct ulist_iterator uiter;
2939 if (!is_fstree(ref_root))
2945 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2946 WARN(1, "%s: Invalid type to free", __func__);
2949 spin_lock(&fs_info->qgroup_lock);
2951 quota_root = fs_info->quota_root;
2955 qgroup = find_qgroup_rb(fs_info, ref_root);
2959 if (num_bytes == (u64)-1)
2961 * We're freeing all pertrans rsv, get reserved value from
2962 * level 0 qgroup as real num_bytes to free.
2964 num_bytes = qgroup->rsv.values[type];
2966 ulist_reinit(fs_info->qgroup_ulist);
2967 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2968 qgroup_to_aux(qgroup), GFP_ATOMIC);
2971 ULIST_ITER_INIT(&uiter);
2972 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2973 struct btrfs_qgroup *qg;
2974 struct btrfs_qgroup_list *glist;
2976 qg = unode_aux_to_qgroup(unode);
2978 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2980 list_for_each_entry(glist, &qg->groups, next_group) {
2981 ret = ulist_add(fs_info->qgroup_ulist,
2982 glist->group->qgroupid,
2983 qgroup_to_aux(glist->group), GFP_ATOMIC);
2990 spin_unlock(&fs_info->qgroup_lock);
2994 * Check if the leaf is the last leaf. Which means all node pointers
2995 * are at their last position.
2997 static bool is_last_leaf(struct btrfs_path *path)
3001 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3002 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3009 * returns < 0 on error, 0 when more leafs are to be scanned.
3010 * returns 1 when done.
3012 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3013 struct btrfs_path *path)
3015 struct btrfs_fs_info *fs_info = trans->fs_info;
3016 struct btrfs_key found;
3017 struct extent_buffer *scratch_leaf = NULL;
3018 struct ulist *roots = NULL;
3024 mutex_lock(&fs_info->qgroup_rescan_lock);
3025 ret = btrfs_search_slot_for_read(fs_info->extent_root,
3026 &fs_info->qgroup_rescan_progress,
3029 btrfs_debug(fs_info,
3030 "current progress key (%llu %u %llu), search_slot ret %d",
3031 fs_info->qgroup_rescan_progress.objectid,
3032 fs_info->qgroup_rescan_progress.type,
3033 fs_info->qgroup_rescan_progress.offset, ret);
3037 * The rescan is about to end, we will not be scanning any
3038 * further blocks. We cannot unset the RESCAN flag here, because
3039 * we want to commit the transaction if everything went well.
3040 * To make the live accounting work in this phase, we set our
3041 * scan progress pointer such that every real extent objectid
3044 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3045 btrfs_release_path(path);
3046 mutex_unlock(&fs_info->qgroup_rescan_lock);
3049 done = is_last_leaf(path);
3051 btrfs_item_key_to_cpu(path->nodes[0], &found,
3052 btrfs_header_nritems(path->nodes[0]) - 1);
3053 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3055 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3056 if (!scratch_leaf) {
3058 mutex_unlock(&fs_info->qgroup_rescan_lock);
3061 slot = path->slots[0];
3062 btrfs_release_path(path);
3063 mutex_unlock(&fs_info->qgroup_rescan_lock);
3065 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3066 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3067 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3068 found.type != BTRFS_METADATA_ITEM_KEY)
3070 if (found.type == BTRFS_METADATA_ITEM_KEY)
3071 num_bytes = fs_info->nodesize;
3073 num_bytes = found.offset;
3075 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3079 /* For rescan, just pass old_roots as NULL */
3080 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3081 num_bytes, NULL, roots);
3087 free_extent_buffer(scratch_leaf);
3091 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3096 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3098 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3099 qgroup_rescan_work);
3100 struct btrfs_path *path;
3101 struct btrfs_trans_handle *trans = NULL;
3105 path = btrfs_alloc_path();
3109 * Rescan should only search for commit root, and any later difference
3110 * should be recorded by qgroup
3112 path->search_commit_root = 1;
3113 path->skip_locking = 1;
3116 while (!err && !btrfs_fs_closing(fs_info)) {
3117 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3118 if (IS_ERR(trans)) {
3119 err = PTR_ERR(trans);
3122 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3125 err = qgroup_rescan_leaf(trans, path);
3128 btrfs_commit_transaction(trans);
3130 btrfs_end_transaction(trans);
3134 btrfs_free_path(path);
3136 mutex_lock(&fs_info->qgroup_rescan_lock);
3137 if (!btrfs_fs_closing(fs_info))
3138 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3141 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3142 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3143 } else if (err < 0) {
3144 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3146 mutex_unlock(&fs_info->qgroup_rescan_lock);
3149 * only update status, since the previous part has already updated the
3152 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3153 if (IS_ERR(trans)) {
3154 err = PTR_ERR(trans);
3156 "fail to start transaction for status update: %d",
3160 ret = update_qgroup_status_item(trans);
3163 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
3165 btrfs_end_transaction(trans);
3167 if (btrfs_fs_closing(fs_info)) {
3168 btrfs_info(fs_info, "qgroup scan paused");
3169 } else if (err >= 0) {
3170 btrfs_info(fs_info, "qgroup scan completed%s",
3171 err > 0 ? " (inconsistency flag cleared)" : "");
3173 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3177 mutex_lock(&fs_info->qgroup_rescan_lock);
3178 fs_info->qgroup_rescan_running = false;
3179 mutex_unlock(&fs_info->qgroup_rescan_lock);
3180 complete_all(&fs_info->qgroup_rescan_completion);
3184 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3185 * memory required for the rescan context.
3188 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3194 /* we're resuming qgroup rescan at mount time */
3195 if (!(fs_info->qgroup_flags &
3196 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3198 "qgroup rescan init failed, qgroup is not enabled");
3200 } else if (!(fs_info->qgroup_flags &
3201 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3203 "qgroup rescan init failed, qgroup rescan is not queued");
3211 mutex_lock(&fs_info->qgroup_rescan_lock);
3212 spin_lock(&fs_info->qgroup_lock);
3215 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3217 "qgroup rescan is already in progress");
3219 } else if (!(fs_info->qgroup_flags &
3220 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3222 "qgroup rescan init failed, qgroup is not enabled");
3227 spin_unlock(&fs_info->qgroup_lock);
3228 mutex_unlock(&fs_info->qgroup_rescan_lock);
3231 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3234 memset(&fs_info->qgroup_rescan_progress, 0,
3235 sizeof(fs_info->qgroup_rescan_progress));
3236 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3237 init_completion(&fs_info->qgroup_rescan_completion);
3238 fs_info->qgroup_rescan_running = true;
3240 spin_unlock(&fs_info->qgroup_lock);
3241 mutex_unlock(&fs_info->qgroup_rescan_lock);
3243 memset(&fs_info->qgroup_rescan_work, 0,
3244 sizeof(fs_info->qgroup_rescan_work));
3245 btrfs_init_work(&fs_info->qgroup_rescan_work,
3246 btrfs_qgroup_rescan_helper,
3247 btrfs_qgroup_rescan_worker, NULL, NULL);
3252 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3255 struct btrfs_qgroup *qgroup;
3257 spin_lock(&fs_info->qgroup_lock);
3258 /* clear all current qgroup tracking information */
3259 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3260 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3262 qgroup->rfer_cmpr = 0;
3264 qgroup->excl_cmpr = 0;
3265 qgroup_dirty(fs_info, qgroup);
3267 spin_unlock(&fs_info->qgroup_lock);
3271 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3274 struct btrfs_trans_handle *trans;
3276 ret = qgroup_rescan_init(fs_info, 0, 1);
3281 * We have set the rescan_progress to 0, which means no more
3282 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3283 * However, btrfs_qgroup_account_ref may be right after its call
3284 * to btrfs_find_all_roots, in which case it would still do the
3286 * To solve this, we're committing the transaction, which will
3287 * ensure we run all delayed refs and only after that, we are
3288 * going to clear all tracking information for a clean start.
3291 trans = btrfs_join_transaction(fs_info->fs_root);
3292 if (IS_ERR(trans)) {
3293 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3294 return PTR_ERR(trans);
3296 ret = btrfs_commit_transaction(trans);
3298 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3302 qgroup_rescan_zero_tracking(fs_info);
3304 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3305 &fs_info->qgroup_rescan_work);
3310 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3316 mutex_lock(&fs_info->qgroup_rescan_lock);
3317 spin_lock(&fs_info->qgroup_lock);
3318 running = fs_info->qgroup_rescan_running;
3319 spin_unlock(&fs_info->qgroup_lock);
3320 mutex_unlock(&fs_info->qgroup_rescan_lock);
3326 ret = wait_for_completion_interruptible(
3327 &fs_info->qgroup_rescan_completion);
3329 wait_for_completion(&fs_info->qgroup_rescan_completion);
3335 * this is only called from open_ctree where we're still single threaded, thus
3336 * locking is omitted here.
3339 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3341 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3342 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3343 &fs_info->qgroup_rescan_work);
3347 * Reserve qgroup space for range [start, start + len).
3349 * This function will either reserve space from related qgroups or doing
3350 * nothing if the range is already reserved.
3352 * Return 0 for successful reserve
3353 * Return <0 for error (including -EQUOT)
3355 * NOTE: this function may sleep for memory allocation.
3356 * if btrfs_qgroup_reserve_data() is called multiple times with
3357 * same @reserved, caller must ensure when error happens it's OK
3358 * to free *ALL* reserved space.
3360 int btrfs_qgroup_reserve_data(struct inode *inode,
3361 struct extent_changeset **reserved_ret, u64 start,
3364 struct btrfs_root *root = BTRFS_I(inode)->root;
3365 struct ulist_node *unode;
3366 struct ulist_iterator uiter;
3367 struct extent_changeset *reserved;
3372 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3373 !is_fstree(root->root_key.objectid) || len == 0)
3376 /* @reserved parameter is mandatory for qgroup */
3377 if (WARN_ON(!reserved_ret))
3379 if (!*reserved_ret) {
3380 *reserved_ret = extent_changeset_alloc();
3384 reserved = *reserved_ret;
3385 /* Record already reserved space */
3386 orig_reserved = reserved->bytes_changed;
3387 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3388 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3390 /* Newly reserved space */
3391 to_reserve = reserved->bytes_changed - orig_reserved;
3392 trace_btrfs_qgroup_reserve_data(inode, start, len,
3393 to_reserve, QGROUP_RESERVE);
3396 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3403 /* cleanup *ALL* already reserved ranges */
3404 ULIST_ITER_INIT(&uiter);
3405 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3406 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3407 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3408 extent_changeset_release(reserved);
3412 /* Free ranges specified by @reserved, normally in error path */
3413 static int qgroup_free_reserved_data(struct inode *inode,
3414 struct extent_changeset *reserved, u64 start, u64 len)
3416 struct btrfs_root *root = BTRFS_I(inode)->root;
3417 struct ulist_node *unode;
3418 struct ulist_iterator uiter;
3419 struct extent_changeset changeset;
3423 extent_changeset_init(&changeset);
3424 len = round_up(start + len, root->fs_info->sectorsize);
3425 start = round_down(start, root->fs_info->sectorsize);
3427 ULIST_ITER_INIT(&uiter);
3428 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3429 u64 range_start = unode->val;
3430 /* unode->aux is the inclusive end */
3431 u64 range_len = unode->aux - range_start + 1;
3435 extent_changeset_release(&changeset);
3437 /* Only free range in range [start, start + len) */
3438 if (range_start >= start + len ||
3439 range_start + range_len <= start)
3441 free_start = max(range_start, start);
3442 free_len = min(start + len, range_start + range_len) -
3445 * TODO: To also modify reserved->ranges_reserved to reflect
3448 * However as long as we free qgroup reserved according to
3449 * EXTENT_QGROUP_RESERVED, we won't double free.
3450 * So not need to rush.
3452 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3453 free_start, free_start + free_len - 1,
3454 EXTENT_QGROUP_RESERVED, &changeset);
3457 freed += changeset.bytes_changed;
3459 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3460 BTRFS_QGROUP_RSV_DATA);
3463 extent_changeset_release(&changeset);
3467 static int __btrfs_qgroup_release_data(struct inode *inode,
3468 struct extent_changeset *reserved, u64 start, u64 len,
3471 struct extent_changeset changeset;
3472 int trace_op = QGROUP_RELEASE;
3475 if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3476 &BTRFS_I(inode)->root->fs_info->flags))
3479 /* In release case, we shouldn't have @reserved */
3480 WARN_ON(!free && reserved);
3481 if (free && reserved)
3482 return qgroup_free_reserved_data(inode, reserved, start, len);
3483 extent_changeset_init(&changeset);
3484 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3485 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3490 trace_op = QGROUP_FREE;
3491 trace_btrfs_qgroup_release_data(inode, start, len,
3492 changeset.bytes_changed, trace_op);
3494 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3495 BTRFS_I(inode)->root->root_key.objectid,
3496 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3497 ret = changeset.bytes_changed;
3499 extent_changeset_release(&changeset);
3504 * Free a reserved space range from io_tree and related qgroups
3506 * Should be called when a range of pages get invalidated before reaching disk.
3507 * Or for error cleanup case.
3508 * if @reserved is given, only reserved range in [@start, @start + @len) will
3511 * For data written to disk, use btrfs_qgroup_release_data().
3513 * NOTE: This function may sleep for memory allocation.
3515 int btrfs_qgroup_free_data(struct inode *inode,
3516 struct extent_changeset *reserved, u64 start, u64 len)
3518 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3522 * Release a reserved space range from io_tree only.
3524 * Should be called when a range of pages get written to disk and corresponding
3525 * FILE_EXTENT is inserted into corresponding root.
3527 * Since new qgroup accounting framework will only update qgroup numbers at
3528 * commit_transaction() time, its reserved space shouldn't be freed from
3531 * But we should release the range from io_tree, to allow further write to be
3534 * NOTE: This function may sleep for memory allocation.
3536 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3538 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3541 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3542 enum btrfs_qgroup_rsv_type type)
3544 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3545 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3550 spin_lock(&root->qgroup_meta_rsv_lock);
3551 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3552 root->qgroup_meta_rsv_prealloc += num_bytes;
3554 root->qgroup_meta_rsv_pertrans += num_bytes;
3555 spin_unlock(&root->qgroup_meta_rsv_lock);
3558 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3559 enum btrfs_qgroup_rsv_type type)
3561 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3562 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3567 spin_lock(&root->qgroup_meta_rsv_lock);
3568 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3569 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3571 root->qgroup_meta_rsv_prealloc -= num_bytes;
3573 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3575 root->qgroup_meta_rsv_pertrans -= num_bytes;
3577 spin_unlock(&root->qgroup_meta_rsv_lock);
3581 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3582 enum btrfs_qgroup_rsv_type type, bool enforce)
3584 struct btrfs_fs_info *fs_info = root->fs_info;
3587 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3588 !is_fstree(root->root_key.objectid) || num_bytes == 0)
3591 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3592 trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3593 ret = qgroup_reserve(root, num_bytes, enforce, type);
3597 * Record what we have reserved into root.
3599 * To avoid quota disabled->enabled underflow.
3600 * In that case, we may try to free space we haven't reserved
3601 * (since quota was disabled), so record what we reserved into root.
3602 * And ensure later release won't underflow this number.
3604 add_root_meta_rsv(root, num_bytes, type);
3608 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3610 struct btrfs_fs_info *fs_info = root->fs_info;
3612 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3613 !is_fstree(root->root_key.objectid))
3616 /* TODO: Update trace point to handle such free */
3617 trace_qgroup_meta_free_all_pertrans(root);
3618 /* Special value -1 means to free all reserved space */
3619 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3620 BTRFS_QGROUP_RSV_META_PERTRANS);
3623 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3624 enum btrfs_qgroup_rsv_type type)
3626 struct btrfs_fs_info *fs_info = root->fs_info;
3628 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3629 !is_fstree(root->root_key.objectid))
3633 * reservation for META_PREALLOC can happen before quota is enabled,
3634 * which can lead to underflow.
3635 * Here ensure we will only free what we really have reserved.
3637 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3638 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3639 trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3640 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3644 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3647 struct btrfs_root *quota_root = fs_info->quota_root;
3648 struct btrfs_qgroup *qgroup;
3649 struct ulist_node *unode;
3650 struct ulist_iterator uiter;
3658 spin_lock(&fs_info->qgroup_lock);
3659 qgroup = find_qgroup_rb(fs_info, ref_root);
3662 ulist_reinit(fs_info->qgroup_ulist);
3663 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3664 qgroup_to_aux(qgroup), GFP_ATOMIC);
3667 ULIST_ITER_INIT(&uiter);
3668 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3669 struct btrfs_qgroup *qg;
3670 struct btrfs_qgroup_list *glist;
3672 qg = unode_aux_to_qgroup(unode);
3674 qgroup_rsv_release(fs_info, qg, num_bytes,
3675 BTRFS_QGROUP_RSV_META_PREALLOC);
3676 qgroup_rsv_add(fs_info, qg, num_bytes,
3677 BTRFS_QGROUP_RSV_META_PERTRANS);
3678 list_for_each_entry(glist, &qg->groups, next_group) {
3679 ret = ulist_add(fs_info->qgroup_ulist,
3680 glist->group->qgroupid,
3681 qgroup_to_aux(glist->group), GFP_ATOMIC);
3687 spin_unlock(&fs_info->qgroup_lock);
3690 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3692 struct btrfs_fs_info *fs_info = root->fs_info;
3694 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3695 !is_fstree(root->root_key.objectid))
3697 /* Same as btrfs_qgroup_free_meta_prealloc() */
3698 num_bytes = sub_root_meta_rsv(root, num_bytes,
3699 BTRFS_QGROUP_RSV_META_PREALLOC);
3700 trace_qgroup_meta_convert(root, num_bytes);
3701 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3705 * Check qgroup reserved space leaking, normally at destroy inode
3708 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3710 struct extent_changeset changeset;
3711 struct ulist_node *unode;
3712 struct ulist_iterator iter;
3715 extent_changeset_init(&changeset);
3716 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3717 EXTENT_QGROUP_RESERVED, &changeset);
3720 if (WARN_ON(changeset.bytes_changed)) {
3721 ULIST_ITER_INIT(&iter);
3722 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3723 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3724 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3725 inode->i_ino, unode->val, unode->aux);
3727 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3728 BTRFS_I(inode)->root->root_key.objectid,
3729 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3732 extent_changeset_release(&changeset);
3735 void btrfs_qgroup_init_swapped_blocks(
3736 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3740 spin_lock_init(&swapped_blocks->lock);
3741 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3742 swapped_blocks->blocks[i] = RB_ROOT;
3743 swapped_blocks->swapped = false;
3747 * Delete all swapped blocks record of @root.
3748 * Every record here means we skipped a full subtree scan for qgroup.
3750 * Gets called when committing one transaction.
3752 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3754 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3757 swapped_blocks = &root->swapped_blocks;
3759 spin_lock(&swapped_blocks->lock);
3760 if (!swapped_blocks->swapped)
3762 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3763 struct rb_root *cur_root = &swapped_blocks->blocks[i];
3764 struct btrfs_qgroup_swapped_block *entry;
3765 struct btrfs_qgroup_swapped_block *next;
3767 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3770 swapped_blocks->blocks[i] = RB_ROOT;
3772 swapped_blocks->swapped = false;
3774 spin_unlock(&swapped_blocks->lock);
3778 * Add subtree roots record into @subvol_root.
3780 * @subvol_root: tree root of the subvolume tree get swapped
3781 * @bg: block group under balance
3782 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
3783 * @reloc_parent/slot: pointer to the subtree root in reloc tree
3784 * BOTH POINTERS ARE BEFORE TREE SWAP
3785 * @last_snapshot: last snapshot generation of the subvolume tree
3787 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3788 struct btrfs_root *subvol_root,
3789 struct btrfs_block_group_cache *bg,
3790 struct extent_buffer *subvol_parent, int subvol_slot,
3791 struct extent_buffer *reloc_parent, int reloc_slot,
3794 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3795 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3796 struct btrfs_qgroup_swapped_block *block;
3797 struct rb_node **cur;
3798 struct rb_node *parent = NULL;
3799 int level = btrfs_header_level(subvol_parent) - 1;
3802 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3805 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3806 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3807 btrfs_err_rl(fs_info,
3808 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3810 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3811 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3815 block = kmalloc(sizeof(*block), GFP_NOFS);
3822 * @reloc_parent/slot is still before swap, while @block is going to
3823 * record the bytenr after swap, so we do the swap here.
3825 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3826 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3828 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3829 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3831 block->last_snapshot = last_snapshot;
3832 block->level = level;
3833 if (bg->flags & BTRFS_BLOCK_GROUP_DATA)
3834 block->trace_leaf = true;
3836 block->trace_leaf = false;
3837 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3839 /* Insert @block into @blocks */
3840 spin_lock(&blocks->lock);
3841 cur = &blocks->blocks[level].rb_node;
3843 struct btrfs_qgroup_swapped_block *entry;
3846 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3849 if (entry->subvol_bytenr < block->subvol_bytenr) {
3850 cur = &(*cur)->rb_left;
3851 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
3852 cur = &(*cur)->rb_right;
3854 if (entry->subvol_generation !=
3855 block->subvol_generation ||
3856 entry->reloc_bytenr != block->reloc_bytenr ||
3857 entry->reloc_generation !=
3858 block->reloc_generation) {
3860 * Duplicated but mismatch entry found.
3863 * Marking qgroup inconsistent should be enough
3866 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3873 rb_link_node(&block->node, parent, cur);
3874 rb_insert_color(&block->node, &blocks->blocks[level]);
3875 blocks->swapped = true;
3877 spin_unlock(&blocks->lock);
3880 fs_info->qgroup_flags |=
3881 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3886 * Check if the tree block is a subtree root, and if so do the needed
3887 * delayed subtree trace for qgroup.
3889 * This is called during btrfs_cow_block().
3891 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3892 struct btrfs_root *root,
3893 struct extent_buffer *subvol_eb)
3895 struct btrfs_fs_info *fs_info = root->fs_info;
3896 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3897 struct btrfs_qgroup_swapped_block *block;
3898 struct extent_buffer *reloc_eb = NULL;
3899 struct rb_node *node;
3901 bool swapped = false;
3902 int level = btrfs_header_level(subvol_eb);
3906 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3908 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3911 spin_lock(&blocks->lock);
3912 if (!blocks->swapped) {
3913 spin_unlock(&blocks->lock);
3916 node = blocks->blocks[level].rb_node;
3919 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3920 if (block->subvol_bytenr < subvol_eb->start) {
3921 node = node->rb_left;
3922 } else if (block->subvol_bytenr > subvol_eb->start) {
3923 node = node->rb_right;
3930 spin_unlock(&blocks->lock);
3933 /* Found one, remove it from @blocks first and update blocks->swapped */
3934 rb_erase(&block->node, &blocks->blocks[level]);
3935 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3936 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3941 blocks->swapped = swapped;
3942 spin_unlock(&blocks->lock);
3944 /* Read out reloc subtree root */
3945 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3946 block->reloc_generation, block->level,
3948 if (IS_ERR(reloc_eb)) {
3949 ret = PTR_ERR(reloc_eb);
3953 if (!extent_buffer_uptodate(reloc_eb)) {
3958 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
3959 block->last_snapshot, block->trace_leaf);
3962 free_extent_buffer(reloc_eb);
3965 btrfs_err_rl(fs_info,
3966 "failed to account subtree at bytenr %llu: %d",
3967 subvol_eb->start, ret);
3968 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;