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"
24 #include "block-group.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;
1319 tmp = ulist_alloc(GFP_KERNEL);
1323 quota_root = fs_info->quota_root;
1329 member = find_qgroup_rb(fs_info, src);
1330 parent = find_qgroup_rb(fs_info, dst);
1332 * The parent/member pair doesn't exist, then try to delete the dead
1333 * relation items only.
1335 if (!member || !parent)
1338 /* check if such qgroup relation exist firstly */
1339 list_for_each_entry(list, &member->groups, next_group) {
1340 if (list->group == parent) {
1347 ret = del_qgroup_relation_item(trans, src, dst);
1348 if (ret < 0 && ret != -ENOENT)
1350 ret2 = del_qgroup_relation_item(trans, dst, src);
1351 if (ret2 < 0 && ret2 != -ENOENT)
1354 /* At least one deletion succeeded, return 0 */
1359 spin_lock(&fs_info->qgroup_lock);
1360 del_relation_rb(fs_info, src, dst);
1361 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1362 spin_unlock(&fs_info->qgroup_lock);
1369 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1372 struct btrfs_fs_info *fs_info = trans->fs_info;
1375 mutex_lock(&fs_info->qgroup_ioctl_lock);
1376 ret = __del_qgroup_relation(trans, src, dst);
1377 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1382 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1384 struct btrfs_fs_info *fs_info = trans->fs_info;
1385 struct btrfs_root *quota_root;
1386 struct btrfs_qgroup *qgroup;
1389 mutex_lock(&fs_info->qgroup_ioctl_lock);
1390 quota_root = fs_info->quota_root;
1395 qgroup = find_qgroup_rb(fs_info, qgroupid);
1401 ret = add_qgroup_item(trans, quota_root, qgroupid);
1405 spin_lock(&fs_info->qgroup_lock);
1406 qgroup = add_qgroup_rb(fs_info, qgroupid);
1407 spin_unlock(&fs_info->qgroup_lock);
1410 ret = PTR_ERR(qgroup);
1412 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1416 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1418 struct btrfs_fs_info *fs_info = trans->fs_info;
1419 struct btrfs_root *quota_root;
1420 struct btrfs_qgroup *qgroup;
1421 struct btrfs_qgroup_list *list;
1424 mutex_lock(&fs_info->qgroup_ioctl_lock);
1425 quota_root = fs_info->quota_root;
1431 qgroup = find_qgroup_rb(fs_info, qgroupid);
1437 /* Check if there are no children of this qgroup */
1438 if (!list_empty(&qgroup->members)) {
1443 ret = del_qgroup_item(trans, qgroupid);
1444 if (ret && ret != -ENOENT)
1447 while (!list_empty(&qgroup->groups)) {
1448 list = list_first_entry(&qgroup->groups,
1449 struct btrfs_qgroup_list, next_group);
1450 ret = __del_qgroup_relation(trans, qgroupid,
1451 list->group->qgroupid);
1456 spin_lock(&fs_info->qgroup_lock);
1457 del_qgroup_rb(fs_info, qgroupid);
1458 spin_unlock(&fs_info->qgroup_lock);
1460 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1464 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1465 struct btrfs_qgroup_limit *limit)
1467 struct btrfs_fs_info *fs_info = trans->fs_info;
1468 struct btrfs_root *quota_root;
1469 struct btrfs_qgroup *qgroup;
1471 /* Sometimes we would want to clear the limit on this qgroup.
1472 * To meet this requirement, we treat the -1 as a special value
1473 * which tell kernel to clear the limit on this qgroup.
1475 const u64 CLEAR_VALUE = -1;
1477 mutex_lock(&fs_info->qgroup_ioctl_lock);
1478 quota_root = fs_info->quota_root;
1484 qgroup = find_qgroup_rb(fs_info, qgroupid);
1490 spin_lock(&fs_info->qgroup_lock);
1491 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1492 if (limit->max_rfer == CLEAR_VALUE) {
1493 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1494 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1495 qgroup->max_rfer = 0;
1497 qgroup->max_rfer = limit->max_rfer;
1500 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1501 if (limit->max_excl == CLEAR_VALUE) {
1502 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1503 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1504 qgroup->max_excl = 0;
1506 qgroup->max_excl = limit->max_excl;
1509 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1510 if (limit->rsv_rfer == CLEAR_VALUE) {
1511 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1512 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1513 qgroup->rsv_rfer = 0;
1515 qgroup->rsv_rfer = limit->rsv_rfer;
1518 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1519 if (limit->rsv_excl == CLEAR_VALUE) {
1520 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1521 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1522 qgroup->rsv_excl = 0;
1524 qgroup->rsv_excl = limit->rsv_excl;
1527 qgroup->lim_flags |= limit->flags;
1529 spin_unlock(&fs_info->qgroup_lock);
1531 ret = update_qgroup_limit_item(trans, qgroup);
1533 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1534 btrfs_info(fs_info, "unable to update quota limit for %llu",
1539 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1543 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1544 struct btrfs_delayed_ref_root *delayed_refs,
1545 struct btrfs_qgroup_extent_record *record)
1547 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1548 struct rb_node *parent_node = NULL;
1549 struct btrfs_qgroup_extent_record *entry;
1550 u64 bytenr = record->bytenr;
1552 lockdep_assert_held(&delayed_refs->lock);
1553 trace_btrfs_qgroup_trace_extent(fs_info, record);
1557 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1559 if (bytenr < entry->bytenr) {
1561 } else if (bytenr > entry->bytenr) {
1562 p = &(*p)->rb_right;
1564 if (record->data_rsv && !entry->data_rsv) {
1565 entry->data_rsv = record->data_rsv;
1566 entry->data_rsv_refroot =
1567 record->data_rsv_refroot;
1573 rb_link_node(&record->node, parent_node, p);
1574 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1578 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1579 struct btrfs_qgroup_extent_record *qrecord)
1581 struct ulist *old_root;
1582 u64 bytenr = qrecord->bytenr;
1585 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1587 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1589 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1595 * Here we don't need to get the lock of
1596 * trans->transaction->delayed_refs, since inserted qrecord won't
1597 * be deleted, only qrecord->node may be modified (new qrecord insert)
1599 * So modifying qrecord->old_roots is safe here
1601 qrecord->old_roots = old_root;
1605 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1606 u64 num_bytes, gfp_t gfp_flag)
1608 struct btrfs_fs_info *fs_info = trans->fs_info;
1609 struct btrfs_qgroup_extent_record *record;
1610 struct btrfs_delayed_ref_root *delayed_refs;
1613 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1614 || bytenr == 0 || num_bytes == 0)
1616 record = kzalloc(sizeof(*record), gfp_flag);
1620 delayed_refs = &trans->transaction->delayed_refs;
1621 record->bytenr = bytenr;
1622 record->num_bytes = num_bytes;
1623 record->old_roots = NULL;
1625 spin_lock(&delayed_refs->lock);
1626 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1627 spin_unlock(&delayed_refs->lock);
1632 return btrfs_qgroup_trace_extent_post(fs_info, record);
1635 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1636 struct extent_buffer *eb)
1638 struct btrfs_fs_info *fs_info = trans->fs_info;
1639 int nr = btrfs_header_nritems(eb);
1640 int i, extent_type, ret;
1641 struct btrfs_key key;
1642 struct btrfs_file_extent_item *fi;
1643 u64 bytenr, num_bytes;
1645 /* We can be called directly from walk_up_proc() */
1646 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1649 for (i = 0; i < nr; i++) {
1650 btrfs_item_key_to_cpu(eb, &key, i);
1652 if (key.type != BTRFS_EXTENT_DATA_KEY)
1655 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1656 /* filter out non qgroup-accountable extents */
1657 extent_type = btrfs_file_extent_type(eb, fi);
1659 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1662 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1666 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1668 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1678 * Walk up the tree from the bottom, freeing leaves and any interior
1679 * nodes which have had all slots visited. If a node (leaf or
1680 * interior) is freed, the node above it will have it's slot
1681 * incremented. The root node will never be freed.
1683 * At the end of this function, we should have a path which has all
1684 * slots incremented to the next position for a search. If we need to
1685 * read a new node it will be NULL and the node above it will have the
1686 * correct slot selected for a later read.
1688 * If we increment the root nodes slot counter past the number of
1689 * elements, 1 is returned to signal completion of the search.
1691 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1695 struct extent_buffer *eb;
1697 if (root_level == 0)
1700 while (level <= root_level) {
1701 eb = path->nodes[level];
1702 nr = btrfs_header_nritems(eb);
1703 path->slots[level]++;
1704 slot = path->slots[level];
1705 if (slot >= nr || level == 0) {
1707 * Don't free the root - we will detect this
1708 * condition after our loop and return a
1709 * positive value for caller to stop walking the tree.
1711 if (level != root_level) {
1712 btrfs_tree_unlock_rw(eb, path->locks[level]);
1713 path->locks[level] = 0;
1715 free_extent_buffer(eb);
1716 path->nodes[level] = NULL;
1717 path->slots[level] = 0;
1721 * We have a valid slot to walk back down
1722 * from. Stop here so caller can process these
1731 eb = path->nodes[root_level];
1732 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1739 * Helper function to trace a subtree tree block swap.
1741 * The swap will happen in highest tree block, but there may be a lot of
1742 * tree blocks involved.
1745 * OO = Old tree blocks
1746 * NN = New tree blocks allocated during balance
1748 * File tree (257) Reloc tree for 257
1751 * L1 OO OO (a) OO NN (a)
1753 * L0 OO OO OO OO OO OO NN NN
1756 * When calling qgroup_trace_extent_swap(), we will pass:
1758 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1762 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1763 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1765 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1767 * 1) Tree search from @src_eb
1768 * It should acts as a simplified btrfs_search_slot().
1769 * The key for search can be extracted from @dst_path->nodes[dst_level]
1772 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1773 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1774 * They should be marked during previous (@dst_level = 1) iteration.
1776 * 3) Mark file extents in leaves dirty
1777 * We don't have good way to pick out new file extents only.
1778 * So we still follow the old method by scanning all file extents in
1781 * This function can free us from keeping two paths, thus later we only need
1782 * to care about how to iterate all new tree blocks in reloc tree.
1784 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1785 struct extent_buffer *src_eb,
1786 struct btrfs_path *dst_path,
1787 int dst_level, int root_level,
1790 struct btrfs_key key;
1791 struct btrfs_path *src_path;
1792 struct btrfs_fs_info *fs_info = trans->fs_info;
1793 u32 nodesize = fs_info->nodesize;
1794 int cur_level = root_level;
1797 BUG_ON(dst_level > root_level);
1798 /* Level mismatch */
1799 if (btrfs_header_level(src_eb) != root_level)
1802 src_path = btrfs_alloc_path();
1809 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1811 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1814 extent_buffer_get(src_eb);
1815 src_path->nodes[root_level] = src_eb;
1816 src_path->slots[root_level] = dst_path->slots[root_level];
1817 src_path->locks[root_level] = 0;
1819 /* A simplified version of btrfs_search_slot() */
1820 while (cur_level >= dst_level) {
1821 struct btrfs_key src_key;
1822 struct btrfs_key dst_key;
1824 if (src_path->nodes[cur_level] == NULL) {
1825 struct btrfs_key first_key;
1826 struct extent_buffer *eb;
1831 eb = src_path->nodes[cur_level + 1];
1832 parent_slot = src_path->slots[cur_level + 1];
1833 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1834 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1835 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1837 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1838 cur_level, &first_key);
1842 } else if (!extent_buffer_uptodate(eb)) {
1843 free_extent_buffer(eb);
1848 src_path->nodes[cur_level] = eb;
1850 btrfs_tree_read_lock(eb);
1851 btrfs_set_lock_blocking_read(eb);
1852 src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1855 src_path->slots[cur_level] = dst_path->slots[cur_level];
1857 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1858 &dst_key, dst_path->slots[cur_level]);
1859 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1860 &src_key, src_path->slots[cur_level]);
1862 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1863 &dst_key, dst_path->slots[cur_level]);
1864 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1865 &src_key, src_path->slots[cur_level]);
1867 /* Content mismatch, something went wrong */
1868 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1876 * Now both @dst_path and @src_path have been populated, record the tree
1877 * blocks for qgroup accounting.
1879 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1880 nodesize, GFP_NOFS);
1883 ret = btrfs_qgroup_trace_extent(trans,
1884 dst_path->nodes[dst_level]->start,
1885 nodesize, GFP_NOFS);
1889 /* Record leaf file extents */
1890 if (dst_level == 0 && trace_leaf) {
1891 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1894 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1897 btrfs_free_path(src_path);
1902 * Helper function to do recursive generation-aware depth-first search, to
1903 * locate all new tree blocks in a subtree of reloc tree.
1905 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1914 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1918 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1919 * above tree blocks along with their counter parts in file tree.
1920 * While during search, old tree blocks OO(c) will be skipped as tree block swap
1921 * won't affect OO(c).
1923 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1924 struct extent_buffer *src_eb,
1925 struct btrfs_path *dst_path,
1926 int cur_level, int root_level,
1927 u64 last_snapshot, bool trace_leaf)
1929 struct btrfs_fs_info *fs_info = trans->fs_info;
1930 struct extent_buffer *eb;
1931 bool need_cleanup = false;
1935 /* Level sanity check */
1936 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL - 1 ||
1937 root_level < 0 || root_level >= BTRFS_MAX_LEVEL - 1 ||
1938 root_level < cur_level) {
1939 btrfs_err_rl(fs_info,
1940 "%s: bad levels, cur_level=%d root_level=%d",
1941 __func__, cur_level, root_level);
1945 /* Read the tree block if needed */
1946 if (dst_path->nodes[cur_level] == NULL) {
1947 struct btrfs_key first_key;
1953 * dst_path->nodes[root_level] must be initialized before
1954 * calling this function.
1956 if (cur_level == root_level) {
1957 btrfs_err_rl(fs_info,
1958 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1959 __func__, root_level, root_level, cur_level);
1964 * We need to get child blockptr/gen from parent before we can
1967 eb = dst_path->nodes[cur_level + 1];
1968 parent_slot = dst_path->slots[cur_level + 1];
1969 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1970 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1971 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1973 /* This node is old, no need to trace */
1974 if (child_gen < last_snapshot)
1977 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1978 cur_level, &first_key);
1982 } else if (!extent_buffer_uptodate(eb)) {
1983 free_extent_buffer(eb);
1988 dst_path->nodes[cur_level] = eb;
1989 dst_path->slots[cur_level] = 0;
1991 btrfs_tree_read_lock(eb);
1992 btrfs_set_lock_blocking_read(eb);
1993 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1994 need_cleanup = true;
1997 /* Now record this tree block and its counter part for qgroups */
1998 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1999 root_level, trace_leaf);
2003 eb = dst_path->nodes[cur_level];
2005 if (cur_level > 0) {
2006 /* Iterate all child tree blocks */
2007 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2008 /* Skip old tree blocks as they won't be swapped */
2009 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
2011 dst_path->slots[cur_level] = i;
2013 /* Recursive call (at most 7 times) */
2014 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
2015 dst_path, cur_level - 1, root_level,
2016 last_snapshot, trace_leaf);
2025 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2026 dst_path->locks[cur_level]);
2027 free_extent_buffer(dst_path->nodes[cur_level]);
2028 dst_path->nodes[cur_level] = NULL;
2029 dst_path->slots[cur_level] = 0;
2030 dst_path->locks[cur_level] = 0;
2036 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2037 struct extent_buffer *src_eb,
2038 struct extent_buffer *dst_eb,
2039 u64 last_snapshot, bool trace_leaf)
2041 struct btrfs_fs_info *fs_info = trans->fs_info;
2042 struct btrfs_path *dst_path = NULL;
2046 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2049 /* Wrong parameter order */
2050 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2051 btrfs_err_rl(fs_info,
2052 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2053 btrfs_header_generation(src_eb),
2054 btrfs_header_generation(dst_eb));
2058 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2063 level = btrfs_header_level(dst_eb);
2064 dst_path = btrfs_alloc_path();
2070 extent_buffer_get(dst_eb);
2071 dst_path->nodes[level] = dst_eb;
2072 dst_path->slots[level] = 0;
2073 dst_path->locks[level] = 0;
2075 /* Do the generation aware breadth-first search */
2076 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2077 level, last_snapshot, trace_leaf);
2083 btrfs_free_path(dst_path);
2085 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2089 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2090 struct extent_buffer *root_eb,
2091 u64 root_gen, int root_level)
2093 struct btrfs_fs_info *fs_info = trans->fs_info;
2096 struct extent_buffer *eb = root_eb;
2097 struct btrfs_path *path = NULL;
2099 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2100 BUG_ON(root_eb == NULL);
2102 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2105 if (!extent_buffer_uptodate(root_eb)) {
2106 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2111 if (root_level == 0) {
2112 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2116 path = btrfs_alloc_path();
2121 * Walk down the tree. Missing extent blocks are filled in as
2122 * we go. Metadata is accounted every time we read a new
2125 * When we reach a leaf, we account for file extent items in it,
2126 * walk back up the tree (adjusting slot pointers as we go)
2127 * and restart the search process.
2129 extent_buffer_get(root_eb); /* For path */
2130 path->nodes[root_level] = root_eb;
2131 path->slots[root_level] = 0;
2132 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2135 while (level >= 0) {
2136 if (path->nodes[level] == NULL) {
2137 struct btrfs_key first_key;
2143 * We need to get child blockptr/gen from parent before
2146 eb = path->nodes[level + 1];
2147 parent_slot = path->slots[level + 1];
2148 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2149 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2150 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2152 eb = read_tree_block(fs_info, child_bytenr, child_gen,
2157 } else if (!extent_buffer_uptodate(eb)) {
2158 free_extent_buffer(eb);
2163 path->nodes[level] = eb;
2164 path->slots[level] = 0;
2166 btrfs_tree_read_lock(eb);
2167 btrfs_set_lock_blocking_read(eb);
2168 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2170 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2178 ret = btrfs_qgroup_trace_leaf_items(trans,
2179 path->nodes[level]);
2183 /* Nonzero return here means we completed our search */
2184 ret = adjust_slots_upwards(path, root_level);
2188 /* Restart search with new slots */
2197 btrfs_free_path(path);
2202 #define UPDATE_NEW 0
2203 #define UPDATE_OLD 1
2205 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2207 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2208 struct ulist *roots, struct ulist *tmp,
2209 struct ulist *qgroups, u64 seq, int update_old)
2211 struct ulist_node *unode;
2212 struct ulist_iterator uiter;
2213 struct ulist_node *tmp_unode;
2214 struct ulist_iterator tmp_uiter;
2215 struct btrfs_qgroup *qg;
2220 ULIST_ITER_INIT(&uiter);
2221 while ((unode = ulist_next(roots, &uiter))) {
2222 qg = find_qgroup_rb(fs_info, unode->val);
2227 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2231 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2234 ULIST_ITER_INIT(&tmp_uiter);
2235 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2236 struct btrfs_qgroup_list *glist;
2238 qg = unode_aux_to_qgroup(tmp_unode);
2240 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2242 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2243 list_for_each_entry(glist, &qg->groups, next_group) {
2244 ret = ulist_add(qgroups, glist->group->qgroupid,
2245 qgroup_to_aux(glist->group),
2249 ret = ulist_add(tmp, glist->group->qgroupid,
2250 qgroup_to_aux(glist->group),
2261 * Update qgroup rfer/excl counters.
2262 * Rfer update is easy, codes can explain themselves.
2264 * Excl update is tricky, the update is split into 2 part.
2265 * Part 1: Possible exclusive <-> sharing detect:
2267 * -------------------------------------
2269 * -------------------------------------
2271 * -------------------------------------
2274 * A: cur_old_roots < nr_old_roots (not exclusive before)
2275 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2276 * B: cur_new_roots < nr_new_roots (not exclusive now)
2277 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2280 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2281 * *: Definitely not changed. **: Possible unchanged.
2283 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2285 * To make the logic clear, we first use condition A and B to split
2286 * combination into 4 results.
2288 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2289 * only on variant maybe 0.
2291 * Lastly, check result **, since there are 2 variants maybe 0, split them
2293 * But this time we don't need to consider other things, the codes and logic
2294 * is easy to understand now.
2296 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2297 struct ulist *qgroups,
2300 u64 num_bytes, u64 seq)
2302 struct ulist_node *unode;
2303 struct ulist_iterator uiter;
2304 struct btrfs_qgroup *qg;
2305 u64 cur_new_count, cur_old_count;
2307 ULIST_ITER_INIT(&uiter);
2308 while ((unode = ulist_next(qgroups, &uiter))) {
2311 qg = unode_aux_to_qgroup(unode);
2312 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2313 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2315 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2318 /* Rfer update part */
2319 if (cur_old_count == 0 && cur_new_count > 0) {
2320 qg->rfer += num_bytes;
2321 qg->rfer_cmpr += num_bytes;
2324 if (cur_old_count > 0 && cur_new_count == 0) {
2325 qg->rfer -= num_bytes;
2326 qg->rfer_cmpr -= num_bytes;
2330 /* Excl update part */
2331 /* Exclusive/none -> shared case */
2332 if (cur_old_count == nr_old_roots &&
2333 cur_new_count < nr_new_roots) {
2334 /* Exclusive -> shared */
2335 if (cur_old_count != 0) {
2336 qg->excl -= num_bytes;
2337 qg->excl_cmpr -= num_bytes;
2342 /* Shared -> exclusive/none case */
2343 if (cur_old_count < nr_old_roots &&
2344 cur_new_count == nr_new_roots) {
2345 /* Shared->exclusive */
2346 if (cur_new_count != 0) {
2347 qg->excl += num_bytes;
2348 qg->excl_cmpr += num_bytes;
2353 /* Exclusive/none -> exclusive/none case */
2354 if (cur_old_count == nr_old_roots &&
2355 cur_new_count == nr_new_roots) {
2356 if (cur_old_count == 0) {
2357 /* None -> exclusive/none */
2359 if (cur_new_count != 0) {
2360 /* None -> exclusive */
2361 qg->excl += num_bytes;
2362 qg->excl_cmpr += num_bytes;
2365 /* None -> none, nothing changed */
2367 /* Exclusive -> exclusive/none */
2369 if (cur_new_count == 0) {
2370 /* Exclusive -> none */
2371 qg->excl -= num_bytes;
2372 qg->excl_cmpr -= num_bytes;
2375 /* Exclusive -> exclusive, nothing changed */
2380 qgroup_dirty(fs_info, qg);
2386 * Check if the @roots potentially is a list of fs tree roots
2388 * Return 0 for definitely not a fs/subvol tree roots ulist
2389 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2392 static int maybe_fs_roots(struct ulist *roots)
2394 struct ulist_node *unode;
2395 struct ulist_iterator uiter;
2397 /* Empty one, still possible for fs roots */
2398 if (!roots || roots->nnodes == 0)
2401 ULIST_ITER_INIT(&uiter);
2402 unode = ulist_next(roots, &uiter);
2407 * If it contains fs tree roots, then it must belong to fs/subvol
2409 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2411 return is_fstree(unode->val);
2414 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2415 u64 num_bytes, struct ulist *old_roots,
2416 struct ulist *new_roots)
2418 struct btrfs_fs_info *fs_info = trans->fs_info;
2419 struct ulist *qgroups = NULL;
2420 struct ulist *tmp = NULL;
2422 u64 nr_new_roots = 0;
2423 u64 nr_old_roots = 0;
2426 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2430 if (!maybe_fs_roots(new_roots))
2432 nr_new_roots = new_roots->nnodes;
2435 if (!maybe_fs_roots(old_roots))
2437 nr_old_roots = old_roots->nnodes;
2440 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2441 if (nr_old_roots == 0 && nr_new_roots == 0)
2444 BUG_ON(!fs_info->quota_root);
2446 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2447 num_bytes, nr_old_roots, nr_new_roots);
2449 qgroups = ulist_alloc(GFP_NOFS);
2454 tmp = ulist_alloc(GFP_NOFS);
2460 mutex_lock(&fs_info->qgroup_rescan_lock);
2461 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2462 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2463 mutex_unlock(&fs_info->qgroup_rescan_lock);
2468 mutex_unlock(&fs_info->qgroup_rescan_lock);
2470 spin_lock(&fs_info->qgroup_lock);
2471 seq = fs_info->qgroup_seq;
2473 /* Update old refcnts using old_roots */
2474 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2479 /* Update new refcnts using new_roots */
2480 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2485 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2489 * Bump qgroup_seq to avoid seq overlap
2491 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2493 spin_unlock(&fs_info->qgroup_lock);
2496 ulist_free(qgroups);
2497 ulist_free(old_roots);
2498 ulist_free(new_roots);
2502 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2504 struct btrfs_fs_info *fs_info = trans->fs_info;
2505 struct btrfs_qgroup_extent_record *record;
2506 struct btrfs_delayed_ref_root *delayed_refs;
2507 struct ulist *new_roots = NULL;
2508 struct rb_node *node;
2509 u64 num_dirty_extents = 0;
2513 delayed_refs = &trans->transaction->delayed_refs;
2514 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2515 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2516 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2519 num_dirty_extents++;
2520 trace_btrfs_qgroup_account_extents(fs_info, record);
2524 * Old roots should be searched when inserting qgroup
2527 if (WARN_ON(!record->old_roots)) {
2528 /* Search commit root to find old_roots */
2529 ret = btrfs_find_all_roots(NULL, fs_info,
2531 &record->old_roots, false);
2536 /* Free the reserved data space */
2537 btrfs_qgroup_free_refroot(fs_info,
2538 record->data_rsv_refroot,
2540 BTRFS_QGROUP_RSV_DATA);
2542 * Use SEQ_LAST as time_seq to do special search, which
2543 * doesn't lock tree or delayed_refs and search current
2544 * root. It's safe inside commit_transaction().
2546 ret = btrfs_find_all_roots(trans, fs_info,
2547 record->bytenr, SEQ_LAST, &new_roots, false);
2550 if (qgroup_to_skip) {
2551 ulist_del(new_roots, qgroup_to_skip, 0);
2552 ulist_del(record->old_roots, qgroup_to_skip,
2555 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2559 record->old_roots = NULL;
2563 ulist_free(record->old_roots);
2564 ulist_free(new_roots);
2566 rb_erase(node, &delayed_refs->dirty_extent_root);
2570 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2576 * called from commit_transaction. Writes all changed qgroups to disk.
2578 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2580 struct btrfs_fs_info *fs_info = trans->fs_info;
2581 struct btrfs_root *quota_root = fs_info->quota_root;
2587 spin_lock(&fs_info->qgroup_lock);
2588 while (!list_empty(&fs_info->dirty_qgroups)) {
2589 struct btrfs_qgroup *qgroup;
2590 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2591 struct btrfs_qgroup, dirty);
2592 list_del_init(&qgroup->dirty);
2593 spin_unlock(&fs_info->qgroup_lock);
2594 ret = update_qgroup_info_item(trans, qgroup);
2596 fs_info->qgroup_flags |=
2597 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2598 ret = update_qgroup_limit_item(trans, qgroup);
2600 fs_info->qgroup_flags |=
2601 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2602 spin_lock(&fs_info->qgroup_lock);
2604 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2605 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2607 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2608 spin_unlock(&fs_info->qgroup_lock);
2610 ret = update_qgroup_status_item(trans);
2612 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2618 * Copy the accounting information between qgroups. This is necessary
2619 * when a snapshot or a subvolume is created. Throwing an error will
2620 * cause a transaction abort so we take extra care here to only error
2621 * when a readonly fs is a reasonable outcome.
2623 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2624 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2629 bool committing = false;
2630 struct btrfs_fs_info *fs_info = trans->fs_info;
2631 struct btrfs_root *quota_root;
2632 struct btrfs_qgroup *srcgroup;
2633 struct btrfs_qgroup *dstgroup;
2638 * There are only two callers of this function.
2640 * One in create_subvol() in the ioctl context, which needs to hold
2641 * the qgroup_ioctl_lock.
2643 * The other one in create_pending_snapshot() where no other qgroup
2644 * code can modify the fs as they all need to either start a new trans
2645 * or hold a trans handler, thus we don't need to hold
2646 * qgroup_ioctl_lock.
2647 * This would avoid long and complex lock chain and make lockdep happy.
2649 spin_lock(&fs_info->trans_lock);
2650 if (trans->transaction->state == TRANS_STATE_COMMIT_DOING)
2652 spin_unlock(&fs_info->trans_lock);
2655 mutex_lock(&fs_info->qgroup_ioctl_lock);
2656 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2659 quota_root = fs_info->quota_root;
2666 i_qgroups = (u64 *)(inherit + 1);
2667 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2668 2 * inherit->num_excl_copies;
2669 for (i = 0; i < nums; ++i) {
2670 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2673 * Zero out invalid groups so we can ignore
2677 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2685 * create a tracking group for the subvol itself
2687 ret = add_qgroup_item(trans, quota_root, objectid);
2692 * add qgroup to all inherited groups
2695 i_qgroups = (u64 *)(inherit + 1);
2696 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2697 if (*i_qgroups == 0)
2699 ret = add_qgroup_relation_item(trans, objectid,
2701 if (ret && ret != -EEXIST)
2703 ret = add_qgroup_relation_item(trans, *i_qgroups,
2705 if (ret && ret != -EEXIST)
2712 spin_lock(&fs_info->qgroup_lock);
2714 dstgroup = add_qgroup_rb(fs_info, objectid);
2715 if (IS_ERR(dstgroup)) {
2716 ret = PTR_ERR(dstgroup);
2720 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2721 dstgroup->lim_flags = inherit->lim.flags;
2722 dstgroup->max_rfer = inherit->lim.max_rfer;
2723 dstgroup->max_excl = inherit->lim.max_excl;
2724 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2725 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2727 ret = update_qgroup_limit_item(trans, dstgroup);
2729 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2731 "unable to update quota limit for %llu",
2732 dstgroup->qgroupid);
2738 srcgroup = find_qgroup_rb(fs_info, srcid);
2743 * We call inherit after we clone the root in order to make sure
2744 * our counts don't go crazy, so at this point the only
2745 * difference between the two roots should be the root node.
2747 level_size = fs_info->nodesize;
2748 dstgroup->rfer = srcgroup->rfer;
2749 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2750 dstgroup->excl = level_size;
2751 dstgroup->excl_cmpr = level_size;
2752 srcgroup->excl = level_size;
2753 srcgroup->excl_cmpr = level_size;
2755 /* inherit the limit info */
2756 dstgroup->lim_flags = srcgroup->lim_flags;
2757 dstgroup->max_rfer = srcgroup->max_rfer;
2758 dstgroup->max_excl = srcgroup->max_excl;
2759 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2760 dstgroup->rsv_excl = srcgroup->rsv_excl;
2762 qgroup_dirty(fs_info, dstgroup);
2763 qgroup_dirty(fs_info, srcgroup);
2769 i_qgroups = (u64 *)(inherit + 1);
2770 for (i = 0; i < inherit->num_qgroups; ++i) {
2772 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2779 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2780 struct btrfs_qgroup *src;
2781 struct btrfs_qgroup *dst;
2783 if (!i_qgroups[0] || !i_qgroups[1])
2786 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2787 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2794 dst->rfer = src->rfer - level_size;
2795 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2797 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2798 struct btrfs_qgroup *src;
2799 struct btrfs_qgroup *dst;
2801 if (!i_qgroups[0] || !i_qgroups[1])
2804 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2805 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2812 dst->excl = src->excl + level_size;
2813 dst->excl_cmpr = src->excl_cmpr + level_size;
2817 spin_unlock(&fs_info->qgroup_lock);
2820 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2825 * Two limits to commit transaction in advance.
2827 * For RATIO, it will be 1/RATIO of the remaining limit as threshold.
2828 * For SIZE, it will be in byte unit as threshold.
2830 #define QGROUP_FREE_RATIO 32
2831 #define QGROUP_FREE_SIZE SZ_32M
2832 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2833 const struct btrfs_qgroup *qg, u64 num_bytes)
2838 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2839 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2842 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2843 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2847 * Even if we passed the check, it's better to check if reservation
2848 * for meta_pertrans is pushing us near limit.
2849 * If there is too much pertrans reservation or it's near the limit,
2850 * let's try commit transaction to free some, using transaction_kthread
2852 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2853 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2854 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
2855 free = qg->max_excl - qgroup_rsv_total(qg) - qg->excl;
2856 threshold = min_t(u64, qg->max_excl / QGROUP_FREE_RATIO,
2859 free = qg->max_rfer - qgroup_rsv_total(qg) - qg->rfer;
2860 threshold = min_t(u64, qg->max_rfer / QGROUP_FREE_RATIO,
2865 * Use transaction_kthread to commit transaction, so we no
2866 * longer need to bother nested transaction nor lock context.
2868 if (free < threshold)
2869 btrfs_commit_transaction_locksafe(fs_info);
2875 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2876 enum btrfs_qgroup_rsv_type type)
2878 struct btrfs_root *quota_root;
2879 struct btrfs_qgroup *qgroup;
2880 struct btrfs_fs_info *fs_info = root->fs_info;
2881 u64 ref_root = root->root_key.objectid;
2883 struct ulist_node *unode;
2884 struct ulist_iterator uiter;
2886 if (!is_fstree(ref_root))
2892 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2893 capable(CAP_SYS_RESOURCE))
2896 spin_lock(&fs_info->qgroup_lock);
2897 quota_root = fs_info->quota_root;
2901 qgroup = find_qgroup_rb(fs_info, ref_root);
2906 * in a first step, we check all affected qgroups if any limits would
2909 ulist_reinit(fs_info->qgroup_ulist);
2910 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2911 qgroup_to_aux(qgroup), GFP_ATOMIC);
2914 ULIST_ITER_INIT(&uiter);
2915 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2916 struct btrfs_qgroup *qg;
2917 struct btrfs_qgroup_list *glist;
2919 qg = unode_aux_to_qgroup(unode);
2921 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2926 list_for_each_entry(glist, &qg->groups, next_group) {
2927 ret = ulist_add(fs_info->qgroup_ulist,
2928 glist->group->qgroupid,
2929 qgroup_to_aux(glist->group), GFP_ATOMIC);
2936 * no limits exceeded, now record the reservation into all qgroups
2938 ULIST_ITER_INIT(&uiter);
2939 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2940 struct btrfs_qgroup *qg;
2942 qg = unode_aux_to_qgroup(unode);
2944 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2948 spin_unlock(&fs_info->qgroup_lock);
2953 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2956 * Will handle all higher level qgroup too.
2958 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2959 * This special case is only used for META_PERTRANS type.
2961 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2962 u64 ref_root, u64 num_bytes,
2963 enum btrfs_qgroup_rsv_type type)
2965 struct btrfs_root *quota_root;
2966 struct btrfs_qgroup *qgroup;
2967 struct ulist_node *unode;
2968 struct ulist_iterator uiter;
2971 if (!is_fstree(ref_root))
2977 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2978 WARN(1, "%s: Invalid type to free", __func__);
2981 spin_lock(&fs_info->qgroup_lock);
2983 quota_root = fs_info->quota_root;
2987 qgroup = find_qgroup_rb(fs_info, ref_root);
2991 if (num_bytes == (u64)-1)
2993 * We're freeing all pertrans rsv, get reserved value from
2994 * level 0 qgroup as real num_bytes to free.
2996 num_bytes = qgroup->rsv.values[type];
2998 ulist_reinit(fs_info->qgroup_ulist);
2999 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3000 qgroup_to_aux(qgroup), GFP_ATOMIC);
3003 ULIST_ITER_INIT(&uiter);
3004 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3005 struct btrfs_qgroup *qg;
3006 struct btrfs_qgroup_list *glist;
3008 qg = unode_aux_to_qgroup(unode);
3010 qgroup_rsv_release(fs_info, qg, num_bytes, type);
3012 list_for_each_entry(glist, &qg->groups, next_group) {
3013 ret = ulist_add(fs_info->qgroup_ulist,
3014 glist->group->qgroupid,
3015 qgroup_to_aux(glist->group), GFP_ATOMIC);
3022 spin_unlock(&fs_info->qgroup_lock);
3026 * Check if the leaf is the last leaf. Which means all node pointers
3027 * are at their last position.
3029 static bool is_last_leaf(struct btrfs_path *path)
3033 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
3034 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3041 * returns < 0 on error, 0 when more leafs are to be scanned.
3042 * returns 1 when done.
3044 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3045 struct btrfs_path *path)
3047 struct btrfs_fs_info *fs_info = trans->fs_info;
3048 struct btrfs_key found;
3049 struct extent_buffer *scratch_leaf = NULL;
3050 struct ulist *roots = NULL;
3056 mutex_lock(&fs_info->qgroup_rescan_lock);
3057 ret = btrfs_search_slot_for_read(fs_info->extent_root,
3058 &fs_info->qgroup_rescan_progress,
3061 btrfs_debug(fs_info,
3062 "current progress key (%llu %u %llu), search_slot ret %d",
3063 fs_info->qgroup_rescan_progress.objectid,
3064 fs_info->qgroup_rescan_progress.type,
3065 fs_info->qgroup_rescan_progress.offset, ret);
3069 * The rescan is about to end, we will not be scanning any
3070 * further blocks. We cannot unset the RESCAN flag here, because
3071 * we want to commit the transaction if everything went well.
3072 * To make the live accounting work in this phase, we set our
3073 * scan progress pointer such that every real extent objectid
3076 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3077 btrfs_release_path(path);
3078 mutex_unlock(&fs_info->qgroup_rescan_lock);
3081 done = is_last_leaf(path);
3083 btrfs_item_key_to_cpu(path->nodes[0], &found,
3084 btrfs_header_nritems(path->nodes[0]) - 1);
3085 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3087 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3088 if (!scratch_leaf) {
3090 mutex_unlock(&fs_info->qgroup_rescan_lock);
3093 slot = path->slots[0];
3094 btrfs_release_path(path);
3095 mutex_unlock(&fs_info->qgroup_rescan_lock);
3097 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3098 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3099 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3100 found.type != BTRFS_METADATA_ITEM_KEY)
3102 if (found.type == BTRFS_METADATA_ITEM_KEY)
3103 num_bytes = fs_info->nodesize;
3105 num_bytes = found.offset;
3107 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3111 /* For rescan, just pass old_roots as NULL */
3112 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3113 num_bytes, NULL, roots);
3119 free_extent_buffer(scratch_leaf);
3123 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3128 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3130 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3131 qgroup_rescan_work);
3132 struct btrfs_path *path;
3133 struct btrfs_trans_handle *trans = NULL;
3137 path = btrfs_alloc_path();
3141 * Rescan should only search for commit root, and any later difference
3142 * should be recorded by qgroup
3144 path->search_commit_root = 1;
3145 path->skip_locking = 1;
3148 while (!err && !btrfs_fs_closing(fs_info)) {
3149 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3150 if (IS_ERR(trans)) {
3151 err = PTR_ERR(trans);
3154 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3157 err = qgroup_rescan_leaf(trans, path);
3160 btrfs_commit_transaction(trans);
3162 btrfs_end_transaction(trans);
3166 btrfs_free_path(path);
3168 mutex_lock(&fs_info->qgroup_rescan_lock);
3169 if (!btrfs_fs_closing(fs_info))
3170 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3173 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3174 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3175 } else if (err < 0) {
3176 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3178 mutex_unlock(&fs_info->qgroup_rescan_lock);
3181 * only update status, since the previous part has already updated the
3184 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3185 if (IS_ERR(trans)) {
3186 err = PTR_ERR(trans);
3188 "fail to start transaction for status update: %d",
3192 ret = update_qgroup_status_item(trans);
3195 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
3197 btrfs_end_transaction(trans);
3199 if (btrfs_fs_closing(fs_info)) {
3200 btrfs_info(fs_info, "qgroup scan paused");
3201 } else if (err >= 0) {
3202 btrfs_info(fs_info, "qgroup scan completed%s",
3203 err > 0 ? " (inconsistency flag cleared)" : "");
3205 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3209 mutex_lock(&fs_info->qgroup_rescan_lock);
3210 fs_info->qgroup_rescan_running = false;
3211 mutex_unlock(&fs_info->qgroup_rescan_lock);
3212 complete_all(&fs_info->qgroup_rescan_completion);
3216 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3217 * memory required for the rescan context.
3220 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3226 /* we're resuming qgroup rescan at mount time */
3227 if (!(fs_info->qgroup_flags &
3228 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3230 "qgroup rescan init failed, qgroup is not enabled");
3232 } else if (!(fs_info->qgroup_flags &
3233 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3235 "qgroup rescan init failed, qgroup rescan is not queued");
3243 mutex_lock(&fs_info->qgroup_rescan_lock);
3244 spin_lock(&fs_info->qgroup_lock);
3247 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3249 "qgroup rescan is already in progress");
3251 } else if (!(fs_info->qgroup_flags &
3252 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3254 "qgroup rescan init failed, qgroup is not enabled");
3259 spin_unlock(&fs_info->qgroup_lock);
3260 mutex_unlock(&fs_info->qgroup_rescan_lock);
3263 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3266 memset(&fs_info->qgroup_rescan_progress, 0,
3267 sizeof(fs_info->qgroup_rescan_progress));
3268 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3269 init_completion(&fs_info->qgroup_rescan_completion);
3270 fs_info->qgroup_rescan_running = true;
3272 spin_unlock(&fs_info->qgroup_lock);
3273 mutex_unlock(&fs_info->qgroup_rescan_lock);
3275 memset(&fs_info->qgroup_rescan_work, 0,
3276 sizeof(fs_info->qgroup_rescan_work));
3277 btrfs_init_work(&fs_info->qgroup_rescan_work,
3278 btrfs_qgroup_rescan_helper,
3279 btrfs_qgroup_rescan_worker, NULL, NULL);
3284 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3287 struct btrfs_qgroup *qgroup;
3289 spin_lock(&fs_info->qgroup_lock);
3290 /* clear all current qgroup tracking information */
3291 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3292 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3294 qgroup->rfer_cmpr = 0;
3296 qgroup->excl_cmpr = 0;
3297 qgroup_dirty(fs_info, qgroup);
3299 spin_unlock(&fs_info->qgroup_lock);
3303 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3306 struct btrfs_trans_handle *trans;
3308 ret = qgroup_rescan_init(fs_info, 0, 1);
3313 * We have set the rescan_progress to 0, which means no more
3314 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3315 * However, btrfs_qgroup_account_ref may be right after its call
3316 * to btrfs_find_all_roots, in which case it would still do the
3318 * To solve this, we're committing the transaction, which will
3319 * ensure we run all delayed refs and only after that, we are
3320 * going to clear all tracking information for a clean start.
3323 trans = btrfs_join_transaction(fs_info->fs_root);
3324 if (IS_ERR(trans)) {
3325 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3326 return PTR_ERR(trans);
3328 ret = btrfs_commit_transaction(trans);
3330 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3334 qgroup_rescan_zero_tracking(fs_info);
3336 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3337 &fs_info->qgroup_rescan_work);
3342 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3348 mutex_lock(&fs_info->qgroup_rescan_lock);
3349 spin_lock(&fs_info->qgroup_lock);
3350 running = fs_info->qgroup_rescan_running;
3351 spin_unlock(&fs_info->qgroup_lock);
3352 mutex_unlock(&fs_info->qgroup_rescan_lock);
3358 ret = wait_for_completion_interruptible(
3359 &fs_info->qgroup_rescan_completion);
3361 wait_for_completion(&fs_info->qgroup_rescan_completion);
3367 * this is only called from open_ctree where we're still single threaded, thus
3368 * locking is omitted here.
3371 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3373 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3374 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3375 &fs_info->qgroup_rescan_work);
3379 * Reserve qgroup space for range [start, start + len).
3381 * This function will either reserve space from related qgroups or doing
3382 * nothing if the range is already reserved.
3384 * Return 0 for successful reserve
3385 * Return <0 for error (including -EQUOT)
3387 * NOTE: this function may sleep for memory allocation.
3388 * if btrfs_qgroup_reserve_data() is called multiple times with
3389 * same @reserved, caller must ensure when error happens it's OK
3390 * to free *ALL* reserved space.
3392 int btrfs_qgroup_reserve_data(struct inode *inode,
3393 struct extent_changeset **reserved_ret, u64 start,
3396 struct btrfs_root *root = BTRFS_I(inode)->root;
3397 struct ulist_node *unode;
3398 struct ulist_iterator uiter;
3399 struct extent_changeset *reserved;
3404 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3405 !is_fstree(root->root_key.objectid) || len == 0)
3408 /* @reserved parameter is mandatory for qgroup */
3409 if (WARN_ON(!reserved_ret))
3411 if (!*reserved_ret) {
3412 *reserved_ret = extent_changeset_alloc();
3416 reserved = *reserved_ret;
3417 /* Record already reserved space */
3418 orig_reserved = reserved->bytes_changed;
3419 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3420 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3422 /* Newly reserved space */
3423 to_reserve = reserved->bytes_changed - orig_reserved;
3424 trace_btrfs_qgroup_reserve_data(inode, start, len,
3425 to_reserve, QGROUP_RESERVE);
3428 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3435 /* cleanup *ALL* already reserved ranges */
3436 ULIST_ITER_INIT(&uiter);
3437 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3438 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3439 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3440 extent_changeset_release(reserved);
3444 /* Free ranges specified by @reserved, normally in error path */
3445 static int qgroup_free_reserved_data(struct inode *inode,
3446 struct extent_changeset *reserved, u64 start, u64 len)
3448 struct btrfs_root *root = BTRFS_I(inode)->root;
3449 struct ulist_node *unode;
3450 struct ulist_iterator uiter;
3451 struct extent_changeset changeset;
3455 extent_changeset_init(&changeset);
3456 len = round_up(start + len, root->fs_info->sectorsize);
3457 start = round_down(start, root->fs_info->sectorsize);
3459 ULIST_ITER_INIT(&uiter);
3460 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3461 u64 range_start = unode->val;
3462 /* unode->aux is the inclusive end */
3463 u64 range_len = unode->aux - range_start + 1;
3467 extent_changeset_release(&changeset);
3469 /* Only free range in range [start, start + len) */
3470 if (range_start >= start + len ||
3471 range_start + range_len <= start)
3473 free_start = max(range_start, start);
3474 free_len = min(start + len, range_start + range_len) -
3477 * TODO: To also modify reserved->ranges_reserved to reflect
3480 * However as long as we free qgroup reserved according to
3481 * EXTENT_QGROUP_RESERVED, we won't double free.
3482 * So not need to rush.
3484 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3485 free_start, free_start + free_len - 1,
3486 EXTENT_QGROUP_RESERVED, &changeset);
3489 freed += changeset.bytes_changed;
3491 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3492 BTRFS_QGROUP_RSV_DATA);
3495 extent_changeset_release(&changeset);
3499 static int __btrfs_qgroup_release_data(struct inode *inode,
3500 struct extent_changeset *reserved, u64 start, u64 len,
3503 struct extent_changeset changeset;
3504 int trace_op = QGROUP_RELEASE;
3507 if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3508 &BTRFS_I(inode)->root->fs_info->flags))
3511 /* In release case, we shouldn't have @reserved */
3512 WARN_ON(!free && reserved);
3513 if (free && reserved)
3514 return qgroup_free_reserved_data(inode, reserved, start, len);
3515 extent_changeset_init(&changeset);
3516 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3517 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3522 trace_op = QGROUP_FREE;
3523 trace_btrfs_qgroup_release_data(inode, start, len,
3524 changeset.bytes_changed, trace_op);
3526 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3527 BTRFS_I(inode)->root->root_key.objectid,
3528 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3529 ret = changeset.bytes_changed;
3531 extent_changeset_release(&changeset);
3536 * Free a reserved space range from io_tree and related qgroups
3538 * Should be called when a range of pages get invalidated before reaching disk.
3539 * Or for error cleanup case.
3540 * if @reserved is given, only reserved range in [@start, @start + @len) will
3543 * For data written to disk, use btrfs_qgroup_release_data().
3545 * NOTE: This function may sleep for memory allocation.
3547 int btrfs_qgroup_free_data(struct inode *inode,
3548 struct extent_changeset *reserved, u64 start, u64 len)
3550 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3554 * Release a reserved space range from io_tree only.
3556 * Should be called when a range of pages get written to disk and corresponding
3557 * FILE_EXTENT is inserted into corresponding root.
3559 * Since new qgroup accounting framework will only update qgroup numbers at
3560 * commit_transaction() time, its reserved space shouldn't be freed from
3563 * But we should release the range from io_tree, to allow further write to be
3566 * NOTE: This function may sleep for memory allocation.
3568 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3570 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3573 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3574 enum btrfs_qgroup_rsv_type type)
3576 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3577 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3582 spin_lock(&root->qgroup_meta_rsv_lock);
3583 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3584 root->qgroup_meta_rsv_prealloc += num_bytes;
3586 root->qgroup_meta_rsv_pertrans += num_bytes;
3587 spin_unlock(&root->qgroup_meta_rsv_lock);
3590 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3591 enum btrfs_qgroup_rsv_type type)
3593 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3594 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3599 spin_lock(&root->qgroup_meta_rsv_lock);
3600 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3601 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3603 root->qgroup_meta_rsv_prealloc -= num_bytes;
3605 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3607 root->qgroup_meta_rsv_pertrans -= num_bytes;
3609 spin_unlock(&root->qgroup_meta_rsv_lock);
3613 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3614 enum btrfs_qgroup_rsv_type type, bool enforce)
3616 struct btrfs_fs_info *fs_info = root->fs_info;
3619 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3620 !is_fstree(root->root_key.objectid) || num_bytes == 0)
3623 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3624 trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3625 ret = qgroup_reserve(root, num_bytes, enforce, type);
3629 * Record what we have reserved into root.
3631 * To avoid quota disabled->enabled underflow.
3632 * In that case, we may try to free space we haven't reserved
3633 * (since quota was disabled), so record what we reserved into root.
3634 * And ensure later release won't underflow this number.
3636 add_root_meta_rsv(root, num_bytes, type);
3640 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3642 struct btrfs_fs_info *fs_info = root->fs_info;
3644 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3645 !is_fstree(root->root_key.objectid))
3648 /* TODO: Update trace point to handle such free */
3649 trace_qgroup_meta_free_all_pertrans(root);
3650 /* Special value -1 means to free all reserved space */
3651 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3652 BTRFS_QGROUP_RSV_META_PERTRANS);
3655 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3656 enum btrfs_qgroup_rsv_type type)
3658 struct btrfs_fs_info *fs_info = root->fs_info;
3660 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3661 !is_fstree(root->root_key.objectid))
3665 * reservation for META_PREALLOC can happen before quota is enabled,
3666 * which can lead to underflow.
3667 * Here ensure we will only free what we really have reserved.
3669 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3670 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3671 trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3672 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3676 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3679 struct btrfs_root *quota_root = fs_info->quota_root;
3680 struct btrfs_qgroup *qgroup;
3681 struct ulist_node *unode;
3682 struct ulist_iterator uiter;
3690 spin_lock(&fs_info->qgroup_lock);
3691 qgroup = find_qgroup_rb(fs_info, ref_root);
3694 ulist_reinit(fs_info->qgroup_ulist);
3695 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3696 qgroup_to_aux(qgroup), GFP_ATOMIC);
3699 ULIST_ITER_INIT(&uiter);
3700 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3701 struct btrfs_qgroup *qg;
3702 struct btrfs_qgroup_list *glist;
3704 qg = unode_aux_to_qgroup(unode);
3706 qgroup_rsv_release(fs_info, qg, num_bytes,
3707 BTRFS_QGROUP_RSV_META_PREALLOC);
3708 qgroup_rsv_add(fs_info, qg, num_bytes,
3709 BTRFS_QGROUP_RSV_META_PERTRANS);
3710 list_for_each_entry(glist, &qg->groups, next_group) {
3711 ret = ulist_add(fs_info->qgroup_ulist,
3712 glist->group->qgroupid,
3713 qgroup_to_aux(glist->group), GFP_ATOMIC);
3719 spin_unlock(&fs_info->qgroup_lock);
3722 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3724 struct btrfs_fs_info *fs_info = root->fs_info;
3726 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3727 !is_fstree(root->root_key.objectid))
3729 /* Same as btrfs_qgroup_free_meta_prealloc() */
3730 num_bytes = sub_root_meta_rsv(root, num_bytes,
3731 BTRFS_QGROUP_RSV_META_PREALLOC);
3732 trace_qgroup_meta_convert(root, num_bytes);
3733 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3737 * Check qgroup reserved space leaking, normally at destroy inode
3740 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3742 struct extent_changeset changeset;
3743 struct ulist_node *unode;
3744 struct ulist_iterator iter;
3747 extent_changeset_init(&changeset);
3748 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3749 EXTENT_QGROUP_RESERVED, &changeset);
3752 if (WARN_ON(changeset.bytes_changed)) {
3753 ULIST_ITER_INIT(&iter);
3754 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3755 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3756 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3757 inode->i_ino, unode->val, unode->aux);
3759 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3760 BTRFS_I(inode)->root->root_key.objectid,
3761 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3764 extent_changeset_release(&changeset);
3767 void btrfs_qgroup_init_swapped_blocks(
3768 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3772 spin_lock_init(&swapped_blocks->lock);
3773 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3774 swapped_blocks->blocks[i] = RB_ROOT;
3775 swapped_blocks->swapped = false;
3779 * Delete all swapped blocks record of @root.
3780 * Every record here means we skipped a full subtree scan for qgroup.
3782 * Gets called when committing one transaction.
3784 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3786 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3789 swapped_blocks = &root->swapped_blocks;
3791 spin_lock(&swapped_blocks->lock);
3792 if (!swapped_blocks->swapped)
3794 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3795 struct rb_root *cur_root = &swapped_blocks->blocks[i];
3796 struct btrfs_qgroup_swapped_block *entry;
3797 struct btrfs_qgroup_swapped_block *next;
3799 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3802 swapped_blocks->blocks[i] = RB_ROOT;
3804 swapped_blocks->swapped = false;
3806 spin_unlock(&swapped_blocks->lock);
3810 * Add subtree roots record into @subvol_root.
3812 * @subvol_root: tree root of the subvolume tree get swapped
3813 * @bg: block group under balance
3814 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
3815 * @reloc_parent/slot: pointer to the subtree root in reloc tree
3816 * BOTH POINTERS ARE BEFORE TREE SWAP
3817 * @last_snapshot: last snapshot generation of the subvolume tree
3819 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3820 struct btrfs_root *subvol_root,
3821 struct btrfs_block_group_cache *bg,
3822 struct extent_buffer *subvol_parent, int subvol_slot,
3823 struct extent_buffer *reloc_parent, int reloc_slot,
3826 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3827 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3828 struct btrfs_qgroup_swapped_block *block;
3829 struct rb_node **cur;
3830 struct rb_node *parent = NULL;
3831 int level = btrfs_header_level(subvol_parent) - 1;
3834 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3837 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3838 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3839 btrfs_err_rl(fs_info,
3840 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3842 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3843 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3847 block = kmalloc(sizeof(*block), GFP_NOFS);
3854 * @reloc_parent/slot is still before swap, while @block is going to
3855 * record the bytenr after swap, so we do the swap here.
3857 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3858 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3860 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3861 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3863 block->last_snapshot = last_snapshot;
3864 block->level = level;
3867 * If we have bg == NULL, we're called from btrfs_recover_relocation(),
3868 * no one else can modify tree blocks thus we qgroup will not change
3869 * no matter the value of trace_leaf.
3871 if (bg && bg->flags & BTRFS_BLOCK_GROUP_DATA)
3872 block->trace_leaf = true;
3874 block->trace_leaf = false;
3875 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3877 /* Insert @block into @blocks */
3878 spin_lock(&blocks->lock);
3879 cur = &blocks->blocks[level].rb_node;
3881 struct btrfs_qgroup_swapped_block *entry;
3884 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3887 if (entry->subvol_bytenr < block->subvol_bytenr) {
3888 cur = &(*cur)->rb_left;
3889 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
3890 cur = &(*cur)->rb_right;
3892 if (entry->subvol_generation !=
3893 block->subvol_generation ||
3894 entry->reloc_bytenr != block->reloc_bytenr ||
3895 entry->reloc_generation !=
3896 block->reloc_generation) {
3898 * Duplicated but mismatch entry found.
3901 * Marking qgroup inconsistent should be enough
3904 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3911 rb_link_node(&block->node, parent, cur);
3912 rb_insert_color(&block->node, &blocks->blocks[level]);
3913 blocks->swapped = true;
3915 spin_unlock(&blocks->lock);
3918 fs_info->qgroup_flags |=
3919 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3924 * Check if the tree block is a subtree root, and if so do the needed
3925 * delayed subtree trace for qgroup.
3927 * This is called during btrfs_cow_block().
3929 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3930 struct btrfs_root *root,
3931 struct extent_buffer *subvol_eb)
3933 struct btrfs_fs_info *fs_info = root->fs_info;
3934 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3935 struct btrfs_qgroup_swapped_block *block;
3936 struct extent_buffer *reloc_eb = NULL;
3937 struct rb_node *node;
3939 bool swapped = false;
3940 int level = btrfs_header_level(subvol_eb);
3944 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3946 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3949 spin_lock(&blocks->lock);
3950 if (!blocks->swapped) {
3951 spin_unlock(&blocks->lock);
3954 node = blocks->blocks[level].rb_node;
3957 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3958 if (block->subvol_bytenr < subvol_eb->start) {
3959 node = node->rb_left;
3960 } else if (block->subvol_bytenr > subvol_eb->start) {
3961 node = node->rb_right;
3968 spin_unlock(&blocks->lock);
3971 /* Found one, remove it from @blocks first and update blocks->swapped */
3972 rb_erase(&block->node, &blocks->blocks[level]);
3973 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3974 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3979 blocks->swapped = swapped;
3980 spin_unlock(&blocks->lock);
3982 /* Read out reloc subtree root */
3983 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3984 block->reloc_generation, block->level,
3986 if (IS_ERR(reloc_eb)) {
3987 ret = PTR_ERR(reloc_eb);
3991 if (!extent_buffer_uptodate(reloc_eb)) {
3996 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
3997 block->last_snapshot, block->trace_leaf);
4000 free_extent_buffer(reloc_eb);
4003 btrfs_err_rl(fs_info,
4004 "failed to account subtree at bytenr %llu: %d",
4005 subvol_eb->start, ret);
4006 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;