1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2011 STRATO. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h>
9 #include <linux/blkdev.h>
10 #include <linux/rbtree.h>
11 #include <linux/slab.h>
12 #include <linux/workqueue.h>
13 #include <linux/btrfs.h>
14 #include <linux/sizes.h>
17 #include "transaction.h"
22 #include "extent_io.h"
27 * - subvol delete -> delete when ref goes to 0? delete limits also?
31 * - copy also limits on subvol creation
34 * - performance benchmarks
35 * - check all ioctl parameters
39 * Helpers to access qgroup reservation
41 * Callers should ensure the lock context and type are valid
44 static u64 qgroup_rsv_total(const struct btrfs_qgroup *qgroup)
49 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
50 ret += qgroup->rsv.values[i];
55 #ifdef CONFIG_BTRFS_DEBUG
56 static const char *qgroup_rsv_type_str(enum btrfs_qgroup_rsv_type type)
58 if (type == BTRFS_QGROUP_RSV_DATA)
60 if (type == BTRFS_QGROUP_RSV_META_PERTRANS)
61 return "meta_pertrans";
62 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
63 return "meta_prealloc";
68 static void qgroup_rsv_add(struct btrfs_fs_info *fs_info,
69 struct btrfs_qgroup *qgroup, u64 num_bytes,
70 enum btrfs_qgroup_rsv_type type)
72 trace_qgroup_update_reserve(fs_info, qgroup, num_bytes, type);
73 qgroup->rsv.values[type] += num_bytes;
76 static void qgroup_rsv_release(struct btrfs_fs_info *fs_info,
77 struct btrfs_qgroup *qgroup, u64 num_bytes,
78 enum btrfs_qgroup_rsv_type type)
80 trace_qgroup_update_reserve(fs_info, qgroup, -(s64)num_bytes, type);
81 if (qgroup->rsv.values[type] >= num_bytes) {
82 qgroup->rsv.values[type] -= num_bytes;
85 #ifdef CONFIG_BTRFS_DEBUG
87 "qgroup %llu %s reserved space underflow, have %llu to free %llu",
88 qgroup->qgroupid, qgroup_rsv_type_str(type),
89 qgroup->rsv.values[type], num_bytes);
91 qgroup->rsv.values[type] = 0;
94 static void qgroup_rsv_add_by_qgroup(struct btrfs_fs_info *fs_info,
95 struct btrfs_qgroup *dest,
96 struct btrfs_qgroup *src)
100 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
101 qgroup_rsv_add(fs_info, dest, src->rsv.values[i], i);
104 static void qgroup_rsv_release_by_qgroup(struct btrfs_fs_info *fs_info,
105 struct btrfs_qgroup *dest,
106 struct btrfs_qgroup *src)
110 for (i = 0; i < BTRFS_QGROUP_RSV_LAST; i++)
111 qgroup_rsv_release(fs_info, dest, src->rsv.values[i], i);
114 static void btrfs_qgroup_update_old_refcnt(struct btrfs_qgroup *qg, u64 seq,
117 if (qg->old_refcnt < seq)
118 qg->old_refcnt = seq;
119 qg->old_refcnt += mod;
122 static void btrfs_qgroup_update_new_refcnt(struct btrfs_qgroup *qg, u64 seq,
125 if (qg->new_refcnt < seq)
126 qg->new_refcnt = seq;
127 qg->new_refcnt += mod;
130 static inline u64 btrfs_qgroup_get_old_refcnt(struct btrfs_qgroup *qg, u64 seq)
132 if (qg->old_refcnt < seq)
134 return qg->old_refcnt - seq;
137 static inline u64 btrfs_qgroup_get_new_refcnt(struct btrfs_qgroup *qg, u64 seq)
139 if (qg->new_refcnt < seq)
141 return qg->new_refcnt - seq;
145 * glue structure to represent the relations between qgroups.
147 struct btrfs_qgroup_list {
148 struct list_head next_group;
149 struct list_head next_member;
150 struct btrfs_qgroup *group;
151 struct btrfs_qgroup *member;
154 static inline u64 qgroup_to_aux(struct btrfs_qgroup *qg)
156 return (u64)(uintptr_t)qg;
159 static inline struct btrfs_qgroup* unode_aux_to_qgroup(struct ulist_node *n)
161 return (struct btrfs_qgroup *)(uintptr_t)n->aux;
165 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
167 static void qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info);
169 /* must be called with qgroup_ioctl_lock held */
170 static struct btrfs_qgroup *find_qgroup_rb(struct btrfs_fs_info *fs_info,
173 struct rb_node *n = fs_info->qgroup_tree.rb_node;
174 struct btrfs_qgroup *qgroup;
177 qgroup = rb_entry(n, struct btrfs_qgroup, node);
178 if (qgroup->qgroupid < qgroupid)
180 else if (qgroup->qgroupid > qgroupid)
188 /* must be called with qgroup_lock held */
189 static struct btrfs_qgroup *add_qgroup_rb(struct btrfs_fs_info *fs_info,
192 struct rb_node **p = &fs_info->qgroup_tree.rb_node;
193 struct rb_node *parent = NULL;
194 struct btrfs_qgroup *qgroup;
198 qgroup = rb_entry(parent, struct btrfs_qgroup, node);
200 if (qgroup->qgroupid < qgroupid)
202 else if (qgroup->qgroupid > qgroupid)
208 qgroup = kzalloc(sizeof(*qgroup), GFP_ATOMIC);
210 return ERR_PTR(-ENOMEM);
212 qgroup->qgroupid = qgroupid;
213 INIT_LIST_HEAD(&qgroup->groups);
214 INIT_LIST_HEAD(&qgroup->members);
215 INIT_LIST_HEAD(&qgroup->dirty);
217 rb_link_node(&qgroup->node, parent, p);
218 rb_insert_color(&qgroup->node, &fs_info->qgroup_tree);
223 static void __del_qgroup_rb(struct btrfs_qgroup *qgroup)
225 struct btrfs_qgroup_list *list;
227 list_del(&qgroup->dirty);
228 while (!list_empty(&qgroup->groups)) {
229 list = list_first_entry(&qgroup->groups,
230 struct btrfs_qgroup_list, next_group);
231 list_del(&list->next_group);
232 list_del(&list->next_member);
236 while (!list_empty(&qgroup->members)) {
237 list = list_first_entry(&qgroup->members,
238 struct btrfs_qgroup_list, next_member);
239 list_del(&list->next_group);
240 list_del(&list->next_member);
246 /* must be called with qgroup_lock held */
247 static int del_qgroup_rb(struct btrfs_fs_info *fs_info, u64 qgroupid)
249 struct btrfs_qgroup *qgroup = find_qgroup_rb(fs_info, qgroupid);
254 rb_erase(&qgroup->node, &fs_info->qgroup_tree);
255 __del_qgroup_rb(qgroup);
259 /* must be called with qgroup_lock held */
260 static int add_relation_rb(struct btrfs_fs_info *fs_info,
261 u64 memberid, u64 parentid)
263 struct btrfs_qgroup *member;
264 struct btrfs_qgroup *parent;
265 struct btrfs_qgroup_list *list;
267 member = find_qgroup_rb(fs_info, memberid);
268 parent = find_qgroup_rb(fs_info, parentid);
269 if (!member || !parent)
272 list = kzalloc(sizeof(*list), GFP_ATOMIC);
276 list->group = parent;
277 list->member = member;
278 list_add_tail(&list->next_group, &member->groups);
279 list_add_tail(&list->next_member, &parent->members);
284 /* must be called with qgroup_lock held */
285 static int del_relation_rb(struct btrfs_fs_info *fs_info,
286 u64 memberid, u64 parentid)
288 struct btrfs_qgroup *member;
289 struct btrfs_qgroup *parent;
290 struct btrfs_qgroup_list *list;
292 member = find_qgroup_rb(fs_info, memberid);
293 parent = find_qgroup_rb(fs_info, parentid);
294 if (!member || !parent)
297 list_for_each_entry(list, &member->groups, next_group) {
298 if (list->group == parent) {
299 list_del(&list->next_group);
300 list_del(&list->next_member);
308 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
309 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
312 struct btrfs_qgroup *qgroup;
314 qgroup = find_qgroup_rb(fs_info, qgroupid);
317 if (qgroup->rfer != rfer || qgroup->excl != excl)
324 * The full config is read in one go, only called from open_ctree()
325 * It doesn't use any locking, as at this point we're still single-threaded
327 int btrfs_read_qgroup_config(struct btrfs_fs_info *fs_info)
329 struct btrfs_key key;
330 struct btrfs_key found_key;
331 struct btrfs_root *quota_root = fs_info->quota_root;
332 struct btrfs_path *path = NULL;
333 struct extent_buffer *l;
337 u64 rescan_progress = 0;
339 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
342 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
343 if (!fs_info->qgroup_ulist) {
348 path = btrfs_alloc_path();
354 /* default this to quota off, in case no status key is found */
355 fs_info->qgroup_flags = 0;
358 * pass 1: read status, all qgroup infos and limits
363 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 1);
368 struct btrfs_qgroup *qgroup;
370 slot = path->slots[0];
372 btrfs_item_key_to_cpu(l, &found_key, slot);
374 if (found_key.type == BTRFS_QGROUP_STATUS_KEY) {
375 struct btrfs_qgroup_status_item *ptr;
377 ptr = btrfs_item_ptr(l, slot,
378 struct btrfs_qgroup_status_item);
380 if (btrfs_qgroup_status_version(l, ptr) !=
381 BTRFS_QGROUP_STATUS_VERSION) {
383 "old qgroup version, quota disabled");
386 if (btrfs_qgroup_status_generation(l, ptr) !=
387 fs_info->generation) {
388 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
390 "qgroup generation mismatch, marked as inconsistent");
392 fs_info->qgroup_flags = btrfs_qgroup_status_flags(l,
394 rescan_progress = btrfs_qgroup_status_rescan(l, ptr);
398 if (found_key.type != BTRFS_QGROUP_INFO_KEY &&
399 found_key.type != BTRFS_QGROUP_LIMIT_KEY)
402 qgroup = find_qgroup_rb(fs_info, found_key.offset);
403 if ((qgroup && found_key.type == BTRFS_QGROUP_INFO_KEY) ||
404 (!qgroup && found_key.type == BTRFS_QGROUP_LIMIT_KEY)) {
405 btrfs_err(fs_info, "inconsistent qgroup config");
406 flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
409 qgroup = add_qgroup_rb(fs_info, found_key.offset);
410 if (IS_ERR(qgroup)) {
411 ret = PTR_ERR(qgroup);
415 switch (found_key.type) {
416 case BTRFS_QGROUP_INFO_KEY: {
417 struct btrfs_qgroup_info_item *ptr;
419 ptr = btrfs_item_ptr(l, slot,
420 struct btrfs_qgroup_info_item);
421 qgroup->rfer = btrfs_qgroup_info_rfer(l, ptr);
422 qgroup->rfer_cmpr = btrfs_qgroup_info_rfer_cmpr(l, ptr);
423 qgroup->excl = btrfs_qgroup_info_excl(l, ptr);
424 qgroup->excl_cmpr = btrfs_qgroup_info_excl_cmpr(l, ptr);
425 /* generation currently unused */
428 case BTRFS_QGROUP_LIMIT_KEY: {
429 struct btrfs_qgroup_limit_item *ptr;
431 ptr = btrfs_item_ptr(l, slot,
432 struct btrfs_qgroup_limit_item);
433 qgroup->lim_flags = btrfs_qgroup_limit_flags(l, ptr);
434 qgroup->max_rfer = btrfs_qgroup_limit_max_rfer(l, ptr);
435 qgroup->max_excl = btrfs_qgroup_limit_max_excl(l, ptr);
436 qgroup->rsv_rfer = btrfs_qgroup_limit_rsv_rfer(l, ptr);
437 qgroup->rsv_excl = btrfs_qgroup_limit_rsv_excl(l, ptr);
442 ret = btrfs_next_item(quota_root, path);
448 btrfs_release_path(path);
451 * pass 2: read all qgroup relations
454 key.type = BTRFS_QGROUP_RELATION_KEY;
456 ret = btrfs_search_slot_for_read(quota_root, &key, path, 1, 0);
460 slot = path->slots[0];
462 btrfs_item_key_to_cpu(l, &found_key, slot);
464 if (found_key.type != BTRFS_QGROUP_RELATION_KEY)
467 if (found_key.objectid > found_key.offset) {
468 /* parent <- member, not needed to build config */
469 /* FIXME should we omit the key completely? */
473 ret = add_relation_rb(fs_info, found_key.objectid,
475 if (ret == -ENOENT) {
477 "orphan qgroup relation 0x%llx->0x%llx",
478 found_key.objectid, found_key.offset);
479 ret = 0; /* ignore the error */
484 ret = btrfs_next_item(quota_root, path);
491 fs_info->qgroup_flags |= flags;
492 if (!(fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_ON))
493 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
494 else if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN &&
496 ret = qgroup_rescan_init(fs_info, rescan_progress, 0);
497 btrfs_free_path(path);
500 ulist_free(fs_info->qgroup_ulist);
501 fs_info->qgroup_ulist = NULL;
502 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
505 return ret < 0 ? ret : 0;
509 * This is called from close_ctree() or open_ctree() or btrfs_quota_disable(),
510 * first two are in single-threaded paths.And for the third one, we have set
511 * quota_root to be null with qgroup_lock held before, so it is safe to clean
512 * up the in-memory structures without qgroup_lock held.
514 void btrfs_free_qgroup_config(struct btrfs_fs_info *fs_info)
517 struct btrfs_qgroup *qgroup;
519 while ((n = rb_first(&fs_info->qgroup_tree))) {
520 qgroup = rb_entry(n, struct btrfs_qgroup, node);
521 rb_erase(n, &fs_info->qgroup_tree);
522 __del_qgroup_rb(qgroup);
525 * We call btrfs_free_qgroup_config() when unmounting
526 * filesystem and disabling quota, so we set qgroup_ulist
527 * to be null here to avoid double free.
529 ulist_free(fs_info->qgroup_ulist);
530 fs_info->qgroup_ulist = NULL;
533 static int add_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
537 struct btrfs_root *quota_root = trans->fs_info->quota_root;
538 struct btrfs_path *path;
539 struct btrfs_key key;
541 path = btrfs_alloc_path();
546 key.type = BTRFS_QGROUP_RELATION_KEY;
549 ret = btrfs_insert_empty_item(trans, quota_root, path, &key, 0);
551 btrfs_mark_buffer_dirty(path->nodes[0]);
553 btrfs_free_path(path);
557 static int del_qgroup_relation_item(struct btrfs_trans_handle *trans, u64 src,
561 struct btrfs_root *quota_root = trans->fs_info->quota_root;
562 struct btrfs_path *path;
563 struct btrfs_key key;
565 path = btrfs_alloc_path();
570 key.type = BTRFS_QGROUP_RELATION_KEY;
573 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
582 ret = btrfs_del_item(trans, quota_root, path);
584 btrfs_free_path(path);
588 static int add_qgroup_item(struct btrfs_trans_handle *trans,
589 struct btrfs_root *quota_root, u64 qgroupid)
592 struct btrfs_path *path;
593 struct btrfs_qgroup_info_item *qgroup_info;
594 struct btrfs_qgroup_limit_item *qgroup_limit;
595 struct extent_buffer *leaf;
596 struct btrfs_key key;
598 if (btrfs_is_testing(quota_root->fs_info))
601 path = btrfs_alloc_path();
606 key.type = BTRFS_QGROUP_INFO_KEY;
607 key.offset = qgroupid;
610 * Avoid a transaction abort by catching -EEXIST here. In that
611 * case, we proceed by re-initializing the existing structure
615 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
616 sizeof(*qgroup_info));
617 if (ret && ret != -EEXIST)
620 leaf = path->nodes[0];
621 qgroup_info = btrfs_item_ptr(leaf, path->slots[0],
622 struct btrfs_qgroup_info_item);
623 btrfs_set_qgroup_info_generation(leaf, qgroup_info, trans->transid);
624 btrfs_set_qgroup_info_rfer(leaf, qgroup_info, 0);
625 btrfs_set_qgroup_info_rfer_cmpr(leaf, qgroup_info, 0);
626 btrfs_set_qgroup_info_excl(leaf, qgroup_info, 0);
627 btrfs_set_qgroup_info_excl_cmpr(leaf, qgroup_info, 0);
629 btrfs_mark_buffer_dirty(leaf);
631 btrfs_release_path(path);
633 key.type = BTRFS_QGROUP_LIMIT_KEY;
634 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
635 sizeof(*qgroup_limit));
636 if (ret && ret != -EEXIST)
639 leaf = path->nodes[0];
640 qgroup_limit = btrfs_item_ptr(leaf, path->slots[0],
641 struct btrfs_qgroup_limit_item);
642 btrfs_set_qgroup_limit_flags(leaf, qgroup_limit, 0);
643 btrfs_set_qgroup_limit_max_rfer(leaf, qgroup_limit, 0);
644 btrfs_set_qgroup_limit_max_excl(leaf, qgroup_limit, 0);
645 btrfs_set_qgroup_limit_rsv_rfer(leaf, qgroup_limit, 0);
646 btrfs_set_qgroup_limit_rsv_excl(leaf, qgroup_limit, 0);
648 btrfs_mark_buffer_dirty(leaf);
652 btrfs_free_path(path);
656 static int del_qgroup_item(struct btrfs_trans_handle *trans, u64 qgroupid)
659 struct btrfs_root *quota_root = trans->fs_info->quota_root;
660 struct btrfs_path *path;
661 struct btrfs_key key;
663 path = btrfs_alloc_path();
668 key.type = BTRFS_QGROUP_INFO_KEY;
669 key.offset = qgroupid;
670 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
679 ret = btrfs_del_item(trans, quota_root, path);
683 btrfs_release_path(path);
685 key.type = BTRFS_QGROUP_LIMIT_KEY;
686 ret = btrfs_search_slot(trans, quota_root, &key, path, -1, 1);
695 ret = btrfs_del_item(trans, quota_root, path);
698 btrfs_free_path(path);
702 static int update_qgroup_limit_item(struct btrfs_trans_handle *trans,
703 struct btrfs_qgroup *qgroup)
705 struct btrfs_root *quota_root = trans->fs_info->quota_root;
706 struct btrfs_path *path;
707 struct btrfs_key key;
708 struct extent_buffer *l;
709 struct btrfs_qgroup_limit_item *qgroup_limit;
714 key.type = BTRFS_QGROUP_LIMIT_KEY;
715 key.offset = qgroup->qgroupid;
717 path = btrfs_alloc_path();
721 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
729 slot = path->slots[0];
730 qgroup_limit = btrfs_item_ptr(l, slot, struct btrfs_qgroup_limit_item);
731 btrfs_set_qgroup_limit_flags(l, qgroup_limit, qgroup->lim_flags);
732 btrfs_set_qgroup_limit_max_rfer(l, qgroup_limit, qgroup->max_rfer);
733 btrfs_set_qgroup_limit_max_excl(l, qgroup_limit, qgroup->max_excl);
734 btrfs_set_qgroup_limit_rsv_rfer(l, qgroup_limit, qgroup->rsv_rfer);
735 btrfs_set_qgroup_limit_rsv_excl(l, qgroup_limit, qgroup->rsv_excl);
737 btrfs_mark_buffer_dirty(l);
740 btrfs_free_path(path);
744 static int update_qgroup_info_item(struct btrfs_trans_handle *trans,
745 struct btrfs_qgroup *qgroup)
747 struct btrfs_fs_info *fs_info = trans->fs_info;
748 struct btrfs_root *quota_root = fs_info->quota_root;
749 struct btrfs_path *path;
750 struct btrfs_key key;
751 struct extent_buffer *l;
752 struct btrfs_qgroup_info_item *qgroup_info;
756 if (btrfs_is_testing(fs_info))
760 key.type = BTRFS_QGROUP_INFO_KEY;
761 key.offset = qgroup->qgroupid;
763 path = btrfs_alloc_path();
767 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
775 slot = path->slots[0];
776 qgroup_info = btrfs_item_ptr(l, slot, struct btrfs_qgroup_info_item);
777 btrfs_set_qgroup_info_generation(l, qgroup_info, trans->transid);
778 btrfs_set_qgroup_info_rfer(l, qgroup_info, qgroup->rfer);
779 btrfs_set_qgroup_info_rfer_cmpr(l, qgroup_info, qgroup->rfer_cmpr);
780 btrfs_set_qgroup_info_excl(l, qgroup_info, qgroup->excl);
781 btrfs_set_qgroup_info_excl_cmpr(l, qgroup_info, qgroup->excl_cmpr);
783 btrfs_mark_buffer_dirty(l);
786 btrfs_free_path(path);
790 static int update_qgroup_status_item(struct btrfs_trans_handle *trans)
792 struct btrfs_fs_info *fs_info = trans->fs_info;
793 struct btrfs_root *quota_root = fs_info->quota_root;
794 struct btrfs_path *path;
795 struct btrfs_key key;
796 struct extent_buffer *l;
797 struct btrfs_qgroup_status_item *ptr;
802 key.type = BTRFS_QGROUP_STATUS_KEY;
805 path = btrfs_alloc_path();
809 ret = btrfs_search_slot(trans, quota_root, &key, path, 0, 1);
817 slot = path->slots[0];
818 ptr = btrfs_item_ptr(l, slot, struct btrfs_qgroup_status_item);
819 btrfs_set_qgroup_status_flags(l, ptr, fs_info->qgroup_flags);
820 btrfs_set_qgroup_status_generation(l, ptr, trans->transid);
821 btrfs_set_qgroup_status_rescan(l, ptr,
822 fs_info->qgroup_rescan_progress.objectid);
824 btrfs_mark_buffer_dirty(l);
827 btrfs_free_path(path);
832 * called with qgroup_lock held
834 static int btrfs_clean_quota_tree(struct btrfs_trans_handle *trans,
835 struct btrfs_root *root)
837 struct btrfs_path *path;
838 struct btrfs_key key;
839 struct extent_buffer *leaf = NULL;
843 path = btrfs_alloc_path();
847 path->leave_spinning = 1;
854 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
857 leaf = path->nodes[0];
858 nr = btrfs_header_nritems(leaf);
862 * delete the leaf one by one
863 * since the whole tree is going
867 ret = btrfs_del_items(trans, root, path, 0, nr);
871 btrfs_release_path(path);
875 btrfs_free_path(path);
879 int btrfs_quota_enable(struct btrfs_fs_info *fs_info)
881 struct btrfs_root *quota_root;
882 struct btrfs_root *tree_root = fs_info->tree_root;
883 struct btrfs_path *path = NULL;
884 struct btrfs_qgroup_status_item *ptr;
885 struct extent_buffer *leaf;
886 struct btrfs_key key;
887 struct btrfs_key found_key;
888 struct btrfs_qgroup *qgroup = NULL;
889 struct btrfs_trans_handle *trans = NULL;
893 mutex_lock(&fs_info->qgroup_ioctl_lock);
894 if (fs_info->quota_root)
898 * 1 for quota root item
899 * 1 for BTRFS_QGROUP_STATUS item
901 * Yet we also need 2*n items for a QGROUP_INFO/QGROUP_LIMIT items
902 * per subvolume. However those are not currently reserved since it
903 * would be a lot of overkill.
905 trans = btrfs_start_transaction(tree_root, 2);
907 ret = PTR_ERR(trans);
912 fs_info->qgroup_ulist = ulist_alloc(GFP_KERNEL);
913 if (!fs_info->qgroup_ulist) {
915 btrfs_abort_transaction(trans, ret);
920 * initially create the quota tree
922 quota_root = btrfs_create_tree(trans, fs_info,
923 BTRFS_QUOTA_TREE_OBJECTID);
924 if (IS_ERR(quota_root)) {
925 ret = PTR_ERR(quota_root);
926 btrfs_abort_transaction(trans, ret);
930 path = btrfs_alloc_path();
933 btrfs_abort_transaction(trans, ret);
938 key.type = BTRFS_QGROUP_STATUS_KEY;
941 ret = btrfs_insert_empty_item(trans, quota_root, path, &key,
944 btrfs_abort_transaction(trans, ret);
948 leaf = path->nodes[0];
949 ptr = btrfs_item_ptr(leaf, path->slots[0],
950 struct btrfs_qgroup_status_item);
951 btrfs_set_qgroup_status_generation(leaf, ptr, trans->transid);
952 btrfs_set_qgroup_status_version(leaf, ptr, BTRFS_QGROUP_STATUS_VERSION);
953 fs_info->qgroup_flags = BTRFS_QGROUP_STATUS_FLAG_ON |
954 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
955 btrfs_set_qgroup_status_flags(leaf, ptr, fs_info->qgroup_flags);
956 btrfs_set_qgroup_status_rescan(leaf, ptr, 0);
958 btrfs_mark_buffer_dirty(leaf);
961 key.type = BTRFS_ROOT_REF_KEY;
964 btrfs_release_path(path);
965 ret = btrfs_search_slot_for_read(tree_root, &key, path, 1, 0);
969 btrfs_abort_transaction(trans, ret);
974 slot = path->slots[0];
975 leaf = path->nodes[0];
976 btrfs_item_key_to_cpu(leaf, &found_key, slot);
978 if (found_key.type == BTRFS_ROOT_REF_KEY) {
979 ret = add_qgroup_item(trans, quota_root,
982 btrfs_abort_transaction(trans, ret);
986 qgroup = add_qgroup_rb(fs_info, found_key.offset);
987 if (IS_ERR(qgroup)) {
988 ret = PTR_ERR(qgroup);
989 btrfs_abort_transaction(trans, ret);
993 ret = btrfs_next_item(tree_root, path);
995 btrfs_abort_transaction(trans, ret);
1003 btrfs_release_path(path);
1004 ret = add_qgroup_item(trans, quota_root, BTRFS_FS_TREE_OBJECTID);
1006 btrfs_abort_transaction(trans, ret);
1010 qgroup = add_qgroup_rb(fs_info, BTRFS_FS_TREE_OBJECTID);
1011 if (IS_ERR(qgroup)) {
1012 ret = PTR_ERR(qgroup);
1013 btrfs_abort_transaction(trans, ret);
1017 ret = btrfs_commit_transaction(trans);
1023 * Set quota enabled flag after committing the transaction, to avoid
1024 * deadlocks on fs_info->qgroup_ioctl_lock with concurrent snapshot
1027 spin_lock(&fs_info->qgroup_lock);
1028 fs_info->quota_root = quota_root;
1029 set_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1030 spin_unlock(&fs_info->qgroup_lock);
1032 ret = qgroup_rescan_init(fs_info, 0, 1);
1034 qgroup_rescan_zero_tracking(fs_info);
1035 btrfs_queue_work(fs_info->qgroup_rescan_workers,
1036 &fs_info->qgroup_rescan_work);
1040 btrfs_free_path(path);
1043 free_extent_buffer(quota_root->node);
1044 free_extent_buffer(quota_root->commit_root);
1049 ulist_free(fs_info->qgroup_ulist);
1050 fs_info->qgroup_ulist = NULL;
1052 btrfs_end_transaction(trans);
1054 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1058 int btrfs_quota_disable(struct btrfs_fs_info *fs_info)
1060 struct btrfs_root *quota_root;
1061 struct btrfs_trans_handle *trans = NULL;
1064 mutex_lock(&fs_info->qgroup_ioctl_lock);
1065 if (!fs_info->quota_root)
1069 * 1 For the root item
1071 * We should also reserve enough items for the quota tree deletion in
1072 * btrfs_clean_quota_tree but this is not done.
1074 trans = btrfs_start_transaction(fs_info->tree_root, 1);
1075 if (IS_ERR(trans)) {
1076 ret = PTR_ERR(trans);
1080 clear_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags);
1081 btrfs_qgroup_wait_for_completion(fs_info, false);
1082 spin_lock(&fs_info->qgroup_lock);
1083 quota_root = fs_info->quota_root;
1084 fs_info->quota_root = NULL;
1085 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
1086 spin_unlock(&fs_info->qgroup_lock);
1088 btrfs_free_qgroup_config(fs_info);
1090 ret = btrfs_clean_quota_tree(trans, quota_root);
1092 btrfs_abort_transaction(trans, ret);
1096 ret = btrfs_del_root(trans, "a_root->root_key);
1098 btrfs_abort_transaction(trans, ret);
1102 list_del("a_root->dirty_list);
1104 btrfs_tree_lock(quota_root->node);
1105 clean_tree_block(fs_info, quota_root->node);
1106 btrfs_tree_unlock(quota_root->node);
1107 btrfs_free_tree_block(trans, quota_root, quota_root->node, 0, 1);
1109 free_extent_buffer(quota_root->node);
1110 free_extent_buffer(quota_root->commit_root);
1114 ret = btrfs_end_transaction(trans);
1116 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1120 static void qgroup_dirty(struct btrfs_fs_info *fs_info,
1121 struct btrfs_qgroup *qgroup)
1123 if (list_empty(&qgroup->dirty))
1124 list_add(&qgroup->dirty, &fs_info->dirty_qgroups);
1128 * The easy accounting, we're updating qgroup relationship whose child qgroup
1129 * only has exclusive extents.
1131 * In this case, all exclusive extents will also be exclusive for parent, so
1132 * excl/rfer just get added/removed.
1134 * So is qgroup reservation space, which should also be added/removed to
1136 * Or when child tries to release reservation space, parent will underflow its
1137 * reservation (for relationship adding case).
1139 * Caller should hold fs_info->qgroup_lock.
1141 static int __qgroup_excl_accounting(struct btrfs_fs_info *fs_info,
1142 struct ulist *tmp, u64 ref_root,
1143 struct btrfs_qgroup *src, int sign)
1145 struct btrfs_qgroup *qgroup;
1146 struct btrfs_qgroup_list *glist;
1147 struct ulist_node *unode;
1148 struct ulist_iterator uiter;
1149 u64 num_bytes = src->excl;
1152 qgroup = find_qgroup_rb(fs_info, ref_root);
1156 qgroup->rfer += sign * num_bytes;
1157 qgroup->rfer_cmpr += sign * num_bytes;
1159 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1160 qgroup->excl += sign * num_bytes;
1161 qgroup->excl_cmpr += sign * num_bytes;
1164 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1166 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1168 qgroup_dirty(fs_info, qgroup);
1170 /* Get all of the parent groups that contain this qgroup */
1171 list_for_each_entry(glist, &qgroup->groups, next_group) {
1172 ret = ulist_add(tmp, glist->group->qgroupid,
1173 qgroup_to_aux(glist->group), GFP_ATOMIC);
1178 /* Iterate all of the parents and adjust their reference counts */
1179 ULIST_ITER_INIT(&uiter);
1180 while ((unode = ulist_next(tmp, &uiter))) {
1181 qgroup = unode_aux_to_qgroup(unode);
1182 qgroup->rfer += sign * num_bytes;
1183 qgroup->rfer_cmpr += sign * num_bytes;
1184 WARN_ON(sign < 0 && qgroup->excl < num_bytes);
1185 qgroup->excl += sign * num_bytes;
1187 qgroup_rsv_add_by_qgroup(fs_info, qgroup, src);
1189 qgroup_rsv_release_by_qgroup(fs_info, qgroup, src);
1190 qgroup->excl_cmpr += sign * num_bytes;
1191 qgroup_dirty(fs_info, qgroup);
1193 /* Add any parents of the parents */
1194 list_for_each_entry(glist, &qgroup->groups, next_group) {
1195 ret = ulist_add(tmp, glist->group->qgroupid,
1196 qgroup_to_aux(glist->group), GFP_ATOMIC);
1208 * Quick path for updating qgroup with only excl refs.
1210 * In that case, just update all parent will be enough.
1211 * Or we needs to do a full rescan.
1212 * Caller should also hold fs_info->qgroup_lock.
1214 * Return 0 for quick update, return >0 for need to full rescan
1215 * and mark INCONSISTENT flag.
1216 * Return < 0 for other error.
1218 static int quick_update_accounting(struct btrfs_fs_info *fs_info,
1219 struct ulist *tmp, u64 src, u64 dst,
1222 struct btrfs_qgroup *qgroup;
1226 qgroup = find_qgroup_rb(fs_info, src);
1229 if (qgroup->excl == qgroup->rfer) {
1231 err = __qgroup_excl_accounting(fs_info, tmp, dst,
1240 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1244 int btrfs_add_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1247 struct btrfs_fs_info *fs_info = trans->fs_info;
1248 struct btrfs_root *quota_root;
1249 struct btrfs_qgroup *parent;
1250 struct btrfs_qgroup *member;
1251 struct btrfs_qgroup_list *list;
1255 /* Check the level of src and dst first */
1256 if (btrfs_qgroup_level(src) >= btrfs_qgroup_level(dst))
1259 tmp = ulist_alloc(GFP_KERNEL);
1263 mutex_lock(&fs_info->qgroup_ioctl_lock);
1264 quota_root = fs_info->quota_root;
1269 member = find_qgroup_rb(fs_info, src);
1270 parent = find_qgroup_rb(fs_info, dst);
1271 if (!member || !parent) {
1276 /* check if such qgroup relation exist firstly */
1277 list_for_each_entry(list, &member->groups, next_group) {
1278 if (list->group == parent) {
1284 ret = add_qgroup_relation_item(trans, src, dst);
1288 ret = add_qgroup_relation_item(trans, dst, src);
1290 del_qgroup_relation_item(trans, src, dst);
1294 spin_lock(&fs_info->qgroup_lock);
1295 ret = add_relation_rb(fs_info, src, dst);
1297 spin_unlock(&fs_info->qgroup_lock);
1300 ret = quick_update_accounting(fs_info, tmp, src, dst, 1);
1301 spin_unlock(&fs_info->qgroup_lock);
1303 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1308 static int __del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1311 struct btrfs_fs_info *fs_info = trans->fs_info;
1312 struct btrfs_root *quota_root;
1313 struct btrfs_qgroup *parent;
1314 struct btrfs_qgroup *member;
1315 struct btrfs_qgroup_list *list;
1320 tmp = ulist_alloc(GFP_KERNEL);
1324 quota_root = fs_info->quota_root;
1330 member = find_qgroup_rb(fs_info, src);
1331 parent = find_qgroup_rb(fs_info, dst);
1332 if (!member || !parent) {
1337 /* check if such qgroup relation exist firstly */
1338 list_for_each_entry(list, &member->groups, next_group) {
1339 if (list->group == parent)
1345 ret = del_qgroup_relation_item(trans, src, dst);
1346 err = del_qgroup_relation_item(trans, dst, src);
1350 spin_lock(&fs_info->qgroup_lock);
1351 del_relation_rb(fs_info, src, dst);
1352 ret = quick_update_accounting(fs_info, tmp, src, dst, -1);
1353 spin_unlock(&fs_info->qgroup_lock);
1359 int btrfs_del_qgroup_relation(struct btrfs_trans_handle *trans, u64 src,
1362 struct btrfs_fs_info *fs_info = trans->fs_info;
1365 mutex_lock(&fs_info->qgroup_ioctl_lock);
1366 ret = __del_qgroup_relation(trans, src, dst);
1367 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1372 int btrfs_create_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1374 struct btrfs_fs_info *fs_info = trans->fs_info;
1375 struct btrfs_root *quota_root;
1376 struct btrfs_qgroup *qgroup;
1379 mutex_lock(&fs_info->qgroup_ioctl_lock);
1380 quota_root = fs_info->quota_root;
1385 qgroup = find_qgroup_rb(fs_info, qgroupid);
1391 ret = add_qgroup_item(trans, quota_root, qgroupid);
1395 spin_lock(&fs_info->qgroup_lock);
1396 qgroup = add_qgroup_rb(fs_info, qgroupid);
1397 spin_unlock(&fs_info->qgroup_lock);
1400 ret = PTR_ERR(qgroup);
1402 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1406 int btrfs_remove_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid)
1408 struct btrfs_fs_info *fs_info = trans->fs_info;
1409 struct btrfs_root *quota_root;
1410 struct btrfs_qgroup *qgroup;
1411 struct btrfs_qgroup_list *list;
1414 mutex_lock(&fs_info->qgroup_ioctl_lock);
1415 quota_root = fs_info->quota_root;
1421 qgroup = find_qgroup_rb(fs_info, qgroupid);
1427 /* Check if there are no children of this qgroup */
1428 if (!list_empty(&qgroup->members)) {
1433 ret = del_qgroup_item(trans, qgroupid);
1434 if (ret && ret != -ENOENT)
1437 while (!list_empty(&qgroup->groups)) {
1438 list = list_first_entry(&qgroup->groups,
1439 struct btrfs_qgroup_list, next_group);
1440 ret = __del_qgroup_relation(trans, qgroupid,
1441 list->group->qgroupid);
1446 spin_lock(&fs_info->qgroup_lock);
1447 del_qgroup_rb(fs_info, qgroupid);
1448 spin_unlock(&fs_info->qgroup_lock);
1450 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1454 int btrfs_limit_qgroup(struct btrfs_trans_handle *trans, u64 qgroupid,
1455 struct btrfs_qgroup_limit *limit)
1457 struct btrfs_fs_info *fs_info = trans->fs_info;
1458 struct btrfs_root *quota_root;
1459 struct btrfs_qgroup *qgroup;
1461 /* Sometimes we would want to clear the limit on this qgroup.
1462 * To meet this requirement, we treat the -1 as a special value
1463 * which tell kernel to clear the limit on this qgroup.
1465 const u64 CLEAR_VALUE = -1;
1467 mutex_lock(&fs_info->qgroup_ioctl_lock);
1468 quota_root = fs_info->quota_root;
1474 qgroup = find_qgroup_rb(fs_info, qgroupid);
1480 spin_lock(&fs_info->qgroup_lock);
1481 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_RFER) {
1482 if (limit->max_rfer == CLEAR_VALUE) {
1483 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1484 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_RFER;
1485 qgroup->max_rfer = 0;
1487 qgroup->max_rfer = limit->max_rfer;
1490 if (limit->flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) {
1491 if (limit->max_excl == CLEAR_VALUE) {
1492 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1493 limit->flags &= ~BTRFS_QGROUP_LIMIT_MAX_EXCL;
1494 qgroup->max_excl = 0;
1496 qgroup->max_excl = limit->max_excl;
1499 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_RFER) {
1500 if (limit->rsv_rfer == CLEAR_VALUE) {
1501 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1502 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_RFER;
1503 qgroup->rsv_rfer = 0;
1505 qgroup->rsv_rfer = limit->rsv_rfer;
1508 if (limit->flags & BTRFS_QGROUP_LIMIT_RSV_EXCL) {
1509 if (limit->rsv_excl == CLEAR_VALUE) {
1510 qgroup->lim_flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1511 limit->flags &= ~BTRFS_QGROUP_LIMIT_RSV_EXCL;
1512 qgroup->rsv_excl = 0;
1514 qgroup->rsv_excl = limit->rsv_excl;
1517 qgroup->lim_flags |= limit->flags;
1519 spin_unlock(&fs_info->qgroup_lock);
1521 ret = update_qgroup_limit_item(trans, qgroup);
1523 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1524 btrfs_info(fs_info, "unable to update quota limit for %llu",
1529 mutex_unlock(&fs_info->qgroup_ioctl_lock);
1533 int btrfs_qgroup_trace_extent_nolock(struct btrfs_fs_info *fs_info,
1534 struct btrfs_delayed_ref_root *delayed_refs,
1535 struct btrfs_qgroup_extent_record *record)
1537 struct rb_node **p = &delayed_refs->dirty_extent_root.rb_node;
1538 struct rb_node *parent_node = NULL;
1539 struct btrfs_qgroup_extent_record *entry;
1540 u64 bytenr = record->bytenr;
1542 lockdep_assert_held(&delayed_refs->lock);
1543 trace_btrfs_qgroup_trace_extent(fs_info, record);
1547 entry = rb_entry(parent_node, struct btrfs_qgroup_extent_record,
1549 if (bytenr < entry->bytenr)
1551 else if (bytenr > entry->bytenr)
1552 p = &(*p)->rb_right;
1557 rb_link_node(&record->node, parent_node, p);
1558 rb_insert_color(&record->node, &delayed_refs->dirty_extent_root);
1562 int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
1563 struct btrfs_qgroup_extent_record *qrecord)
1565 struct ulist *old_root;
1566 u64 bytenr = qrecord->bytenr;
1569 ret = btrfs_find_all_roots(NULL, fs_info, bytenr, 0, &old_root, false);
1571 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
1573 "error accounting new delayed refs extent (err code: %d), quota inconsistent",
1579 * Here we don't need to get the lock of
1580 * trans->transaction->delayed_refs, since inserted qrecord won't
1581 * be deleted, only qrecord->node may be modified (new qrecord insert)
1583 * So modifying qrecord->old_roots is safe here
1585 qrecord->old_roots = old_root;
1589 int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans, u64 bytenr,
1590 u64 num_bytes, gfp_t gfp_flag)
1592 struct btrfs_fs_info *fs_info = trans->fs_info;
1593 struct btrfs_qgroup_extent_record *record;
1594 struct btrfs_delayed_ref_root *delayed_refs;
1597 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)
1598 || bytenr == 0 || num_bytes == 0)
1600 record = kmalloc(sizeof(*record), gfp_flag);
1604 delayed_refs = &trans->transaction->delayed_refs;
1605 record->bytenr = bytenr;
1606 record->num_bytes = num_bytes;
1607 record->old_roots = NULL;
1609 spin_lock(&delayed_refs->lock);
1610 ret = btrfs_qgroup_trace_extent_nolock(fs_info, delayed_refs, record);
1611 spin_unlock(&delayed_refs->lock);
1616 return btrfs_qgroup_trace_extent_post(fs_info, record);
1619 int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
1620 struct extent_buffer *eb)
1622 struct btrfs_fs_info *fs_info = trans->fs_info;
1623 int nr = btrfs_header_nritems(eb);
1624 int i, extent_type, ret;
1625 struct btrfs_key key;
1626 struct btrfs_file_extent_item *fi;
1627 u64 bytenr, num_bytes;
1629 /* We can be called directly from walk_up_proc() */
1630 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
1633 for (i = 0; i < nr; i++) {
1634 btrfs_item_key_to_cpu(eb, &key, i);
1636 if (key.type != BTRFS_EXTENT_DATA_KEY)
1639 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
1640 /* filter out non qgroup-accountable extents */
1641 extent_type = btrfs_file_extent_type(eb, fi);
1643 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
1646 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
1650 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
1652 ret = btrfs_qgroup_trace_extent(trans, bytenr, num_bytes,
1662 * Walk up the tree from the bottom, freeing leaves and any interior
1663 * nodes which have had all slots visited. If a node (leaf or
1664 * interior) is freed, the node above it will have it's slot
1665 * incremented. The root node will never be freed.
1667 * At the end of this function, we should have a path which has all
1668 * slots incremented to the next position for a search. If we need to
1669 * read a new node it will be NULL and the node above it will have the
1670 * correct slot selected for a later read.
1672 * If we increment the root nodes slot counter past the number of
1673 * elements, 1 is returned to signal completion of the search.
1675 static int adjust_slots_upwards(struct btrfs_path *path, int root_level)
1679 struct extent_buffer *eb;
1681 if (root_level == 0)
1684 while (level <= root_level) {
1685 eb = path->nodes[level];
1686 nr = btrfs_header_nritems(eb);
1687 path->slots[level]++;
1688 slot = path->slots[level];
1689 if (slot >= nr || level == 0) {
1691 * Don't free the root - we will detect this
1692 * condition after our loop and return a
1693 * positive value for caller to stop walking the tree.
1695 if (level != root_level) {
1696 btrfs_tree_unlock_rw(eb, path->locks[level]);
1697 path->locks[level] = 0;
1699 free_extent_buffer(eb);
1700 path->nodes[level] = NULL;
1701 path->slots[level] = 0;
1705 * We have a valid slot to walk back down
1706 * from. Stop here so caller can process these
1715 eb = path->nodes[root_level];
1716 if (path->slots[root_level] >= btrfs_header_nritems(eb))
1723 * Helper function to trace a subtree tree block swap.
1725 * The swap will happen in highest tree block, but there may be a lot of
1726 * tree blocks involved.
1729 * OO = Old tree blocks
1730 * NN = New tree blocks allocated during balance
1732 * File tree (257) Reloc tree for 257
1735 * L1 OO OO (a) OO NN (a)
1737 * L0 OO OO OO OO OO OO NN NN
1740 * When calling qgroup_trace_extent_swap(), we will pass:
1742 * @dst_path = [ nodes[1] = NN(a), nodes[0] = NN(c) ]
1746 * In that case, qgroup_trace_extent_swap() will search from OO(a) to
1747 * reach OO(c), then mark both OO(c) and NN(c) as qgroup dirty.
1749 * The main work of qgroup_trace_extent_swap() can be split into 3 parts:
1751 * 1) Tree search from @src_eb
1752 * It should acts as a simplified btrfs_search_slot().
1753 * The key for search can be extracted from @dst_path->nodes[dst_level]
1756 * 2) Mark the final tree blocks in @src_path and @dst_path qgroup dirty
1757 * NOTE: In above case, OO(a) and NN(a) won't be marked qgroup dirty.
1758 * They should be marked during previous (@dst_level = 1) iteration.
1760 * 3) Mark file extents in leaves dirty
1761 * We don't have good way to pick out new file extents only.
1762 * So we still follow the old method by scanning all file extents in
1765 * This function can free us from keeping two paths, thus later we only need
1766 * to care about how to iterate all new tree blocks in reloc tree.
1768 static int qgroup_trace_extent_swap(struct btrfs_trans_handle* trans,
1769 struct extent_buffer *src_eb,
1770 struct btrfs_path *dst_path,
1771 int dst_level, int root_level,
1774 struct btrfs_key key;
1775 struct btrfs_path *src_path;
1776 struct btrfs_fs_info *fs_info = trans->fs_info;
1777 u32 nodesize = fs_info->nodesize;
1778 int cur_level = root_level;
1781 BUG_ON(dst_level > root_level);
1782 /* Level mismatch */
1783 if (btrfs_header_level(src_eb) != root_level)
1786 src_path = btrfs_alloc_path();
1793 btrfs_node_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1795 btrfs_item_key_to_cpu(dst_path->nodes[dst_level], &key, 0);
1798 extent_buffer_get(src_eb);
1799 src_path->nodes[root_level] = src_eb;
1800 src_path->slots[root_level] = dst_path->slots[root_level];
1801 src_path->locks[root_level] = 0;
1803 /* A simplified version of btrfs_search_slot() */
1804 while (cur_level >= dst_level) {
1805 struct btrfs_key src_key;
1806 struct btrfs_key dst_key;
1808 if (src_path->nodes[cur_level] == NULL) {
1809 struct btrfs_key first_key;
1810 struct extent_buffer *eb;
1815 eb = src_path->nodes[cur_level + 1];
1816 parent_slot = src_path->slots[cur_level + 1];
1817 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1818 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1819 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1821 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1822 cur_level, &first_key);
1826 } else if (!extent_buffer_uptodate(eb)) {
1827 free_extent_buffer(eb);
1832 src_path->nodes[cur_level] = eb;
1834 btrfs_tree_read_lock(eb);
1835 btrfs_set_lock_blocking_read(eb);
1836 src_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1839 src_path->slots[cur_level] = dst_path->slots[cur_level];
1841 btrfs_node_key_to_cpu(dst_path->nodes[cur_level],
1842 &dst_key, dst_path->slots[cur_level]);
1843 btrfs_node_key_to_cpu(src_path->nodes[cur_level],
1844 &src_key, src_path->slots[cur_level]);
1846 btrfs_item_key_to_cpu(dst_path->nodes[cur_level],
1847 &dst_key, dst_path->slots[cur_level]);
1848 btrfs_item_key_to_cpu(src_path->nodes[cur_level],
1849 &src_key, src_path->slots[cur_level]);
1851 /* Content mismatch, something went wrong */
1852 if (btrfs_comp_cpu_keys(&dst_key, &src_key)) {
1860 * Now both @dst_path and @src_path have been populated, record the tree
1861 * blocks for qgroup accounting.
1863 ret = btrfs_qgroup_trace_extent(trans, src_path->nodes[dst_level]->start,
1864 nodesize, GFP_NOFS);
1867 ret = btrfs_qgroup_trace_extent(trans,
1868 dst_path->nodes[dst_level]->start,
1869 nodesize, GFP_NOFS);
1873 /* Record leaf file extents */
1874 if (dst_level == 0 && trace_leaf) {
1875 ret = btrfs_qgroup_trace_leaf_items(trans, src_path->nodes[0]);
1878 ret = btrfs_qgroup_trace_leaf_items(trans, dst_path->nodes[0]);
1881 btrfs_free_path(src_path);
1886 * Helper function to do recursive generation-aware depth-first search, to
1887 * locate all new tree blocks in a subtree of reloc tree.
1889 * E.g. (OO = Old tree blocks, NN = New tree blocks, whose gen == last_snapshot)
1898 * @dst_path = [ nodes[1] = NN(b), nodes[0] = NULL ],
1902 * We will iterate through tree blocks NN(b), NN(d) and info qgroup to trace
1903 * above tree blocks along with their counter parts in file tree.
1904 * While during search, old tree blocks OO(c) will be skipped as tree block swap
1905 * won't affect OO(c).
1907 static int qgroup_trace_new_subtree_blocks(struct btrfs_trans_handle* trans,
1908 struct extent_buffer *src_eb,
1909 struct btrfs_path *dst_path,
1910 int cur_level, int root_level,
1911 u64 last_snapshot, bool trace_leaf)
1913 struct btrfs_fs_info *fs_info = trans->fs_info;
1914 struct extent_buffer *eb;
1915 bool need_cleanup = false;
1919 /* Level sanity check */
1920 if (cur_level < 0 || cur_level >= BTRFS_MAX_LEVEL ||
1921 root_level < 0 || root_level >= BTRFS_MAX_LEVEL ||
1922 root_level < cur_level) {
1923 btrfs_err_rl(fs_info,
1924 "%s: bad levels, cur_level=%d root_level=%d",
1925 __func__, cur_level, root_level);
1929 /* Read the tree block if needed */
1930 if (dst_path->nodes[cur_level] == NULL) {
1931 struct btrfs_key first_key;
1937 * dst_path->nodes[root_level] must be initialized before
1938 * calling this function.
1940 if (cur_level == root_level) {
1941 btrfs_err_rl(fs_info,
1942 "%s: dst_path->nodes[%d] not initialized, root_level=%d cur_level=%d",
1943 __func__, root_level, root_level, cur_level);
1948 * We need to get child blockptr/gen from parent before we can
1951 eb = dst_path->nodes[cur_level + 1];
1952 parent_slot = dst_path->slots[cur_level + 1];
1953 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
1954 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
1955 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
1957 /* This node is old, no need to trace */
1958 if (child_gen < last_snapshot)
1961 eb = read_tree_block(fs_info, child_bytenr, child_gen,
1962 cur_level, &first_key);
1966 } else if (!extent_buffer_uptodate(eb)) {
1967 free_extent_buffer(eb);
1972 dst_path->nodes[cur_level] = eb;
1973 dst_path->slots[cur_level] = 0;
1975 btrfs_tree_read_lock(eb);
1976 btrfs_set_lock_blocking_read(eb);
1977 dst_path->locks[cur_level] = BTRFS_READ_LOCK_BLOCKING;
1978 need_cleanup = true;
1981 /* Now record this tree block and its counter part for qgroups */
1982 ret = qgroup_trace_extent_swap(trans, src_eb, dst_path, cur_level,
1983 root_level, trace_leaf);
1987 eb = dst_path->nodes[cur_level];
1989 if (cur_level > 0) {
1990 /* Iterate all child tree blocks */
1991 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1992 /* Skip old tree blocks as they won't be swapped */
1993 if (btrfs_node_ptr_generation(eb, i) < last_snapshot)
1995 dst_path->slots[cur_level] = i;
1997 /* Recursive call (at most 7 times) */
1998 ret = qgroup_trace_new_subtree_blocks(trans, src_eb,
1999 dst_path, cur_level - 1, root_level,
2000 last_snapshot, trace_leaf);
2009 btrfs_tree_unlock_rw(dst_path->nodes[cur_level],
2010 dst_path->locks[cur_level]);
2011 free_extent_buffer(dst_path->nodes[cur_level]);
2012 dst_path->nodes[cur_level] = NULL;
2013 dst_path->slots[cur_level] = 0;
2014 dst_path->locks[cur_level] = 0;
2020 static int qgroup_trace_subtree_swap(struct btrfs_trans_handle *trans,
2021 struct extent_buffer *src_eb,
2022 struct extent_buffer *dst_eb,
2023 u64 last_snapshot, bool trace_leaf)
2025 struct btrfs_fs_info *fs_info = trans->fs_info;
2026 struct btrfs_path *dst_path = NULL;
2030 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2033 /* Wrong parameter order */
2034 if (btrfs_header_generation(src_eb) > btrfs_header_generation(dst_eb)) {
2035 btrfs_err_rl(fs_info,
2036 "%s: bad parameter order, src_gen=%llu dst_gen=%llu", __func__,
2037 btrfs_header_generation(src_eb),
2038 btrfs_header_generation(dst_eb));
2042 if (!extent_buffer_uptodate(src_eb) || !extent_buffer_uptodate(dst_eb)) {
2047 level = btrfs_header_level(dst_eb);
2048 dst_path = btrfs_alloc_path();
2054 extent_buffer_get(dst_eb);
2055 dst_path->nodes[level] = dst_eb;
2056 dst_path->slots[level] = 0;
2057 dst_path->locks[level] = 0;
2059 /* Do the generation aware breadth-first search */
2060 ret = qgroup_trace_new_subtree_blocks(trans, src_eb, dst_path, level,
2061 level, last_snapshot, trace_leaf);
2067 btrfs_free_path(dst_path);
2069 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2073 int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
2074 struct extent_buffer *root_eb,
2075 u64 root_gen, int root_level)
2077 struct btrfs_fs_info *fs_info = trans->fs_info;
2080 struct extent_buffer *eb = root_eb;
2081 struct btrfs_path *path = NULL;
2083 BUG_ON(root_level < 0 || root_level >= BTRFS_MAX_LEVEL);
2084 BUG_ON(root_eb == NULL);
2086 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2089 if (!extent_buffer_uptodate(root_eb)) {
2090 ret = btrfs_read_buffer(root_eb, root_gen, root_level, NULL);
2095 if (root_level == 0) {
2096 ret = btrfs_qgroup_trace_leaf_items(trans, root_eb);
2100 path = btrfs_alloc_path();
2105 * Walk down the tree. Missing extent blocks are filled in as
2106 * we go. Metadata is accounted every time we read a new
2109 * When we reach a leaf, we account for file extent items in it,
2110 * walk back up the tree (adjusting slot pointers as we go)
2111 * and restart the search process.
2113 extent_buffer_get(root_eb); /* For path */
2114 path->nodes[root_level] = root_eb;
2115 path->slots[root_level] = 0;
2116 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
2119 while (level >= 0) {
2120 if (path->nodes[level] == NULL) {
2121 struct btrfs_key first_key;
2127 * We need to get child blockptr/gen from parent before
2130 eb = path->nodes[level + 1];
2131 parent_slot = path->slots[level + 1];
2132 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
2133 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
2134 btrfs_node_key_to_cpu(eb, &first_key, parent_slot);
2136 eb = read_tree_block(fs_info, child_bytenr, child_gen,
2141 } else if (!extent_buffer_uptodate(eb)) {
2142 free_extent_buffer(eb);
2147 path->nodes[level] = eb;
2148 path->slots[level] = 0;
2150 btrfs_tree_read_lock(eb);
2151 btrfs_set_lock_blocking_read(eb);
2152 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
2154 ret = btrfs_qgroup_trace_extent(trans, child_bytenr,
2162 ret = btrfs_qgroup_trace_leaf_items(trans,
2163 path->nodes[level]);
2167 /* Nonzero return here means we completed our search */
2168 ret = adjust_slots_upwards(path, root_level);
2172 /* Restart search with new slots */
2181 btrfs_free_path(path);
2186 #define UPDATE_NEW 0
2187 #define UPDATE_OLD 1
2189 * Walk all of the roots that points to the bytenr and adjust their refcnts.
2191 static int qgroup_update_refcnt(struct btrfs_fs_info *fs_info,
2192 struct ulist *roots, struct ulist *tmp,
2193 struct ulist *qgroups, u64 seq, int update_old)
2195 struct ulist_node *unode;
2196 struct ulist_iterator uiter;
2197 struct ulist_node *tmp_unode;
2198 struct ulist_iterator tmp_uiter;
2199 struct btrfs_qgroup *qg;
2204 ULIST_ITER_INIT(&uiter);
2205 while ((unode = ulist_next(roots, &uiter))) {
2206 qg = find_qgroup_rb(fs_info, unode->val);
2211 ret = ulist_add(qgroups, qg->qgroupid, qgroup_to_aux(qg),
2215 ret = ulist_add(tmp, qg->qgroupid, qgroup_to_aux(qg), GFP_ATOMIC);
2218 ULIST_ITER_INIT(&tmp_uiter);
2219 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
2220 struct btrfs_qgroup_list *glist;
2222 qg = unode_aux_to_qgroup(tmp_unode);
2224 btrfs_qgroup_update_old_refcnt(qg, seq, 1);
2226 btrfs_qgroup_update_new_refcnt(qg, seq, 1);
2227 list_for_each_entry(glist, &qg->groups, next_group) {
2228 ret = ulist_add(qgroups, glist->group->qgroupid,
2229 qgroup_to_aux(glist->group),
2233 ret = ulist_add(tmp, glist->group->qgroupid,
2234 qgroup_to_aux(glist->group),
2245 * Update qgroup rfer/excl counters.
2246 * Rfer update is easy, codes can explain themselves.
2248 * Excl update is tricky, the update is split into 2 part.
2249 * Part 1: Possible exclusive <-> sharing detect:
2251 * -------------------------------------
2253 * -------------------------------------
2255 * -------------------------------------
2258 * A: cur_old_roots < nr_old_roots (not exclusive before)
2259 * !A: cur_old_roots == nr_old_roots (possible exclusive before)
2260 * B: cur_new_roots < nr_new_roots (not exclusive now)
2261 * !B: cur_new_roots == nr_new_roots (possible exclusive now)
2264 * +: Possible sharing -> exclusive -: Possible exclusive -> sharing
2265 * *: Definitely not changed. **: Possible unchanged.
2267 * For !A and !B condition, the exception is cur_old/new_roots == 0 case.
2269 * To make the logic clear, we first use condition A and B to split
2270 * combination into 4 results.
2272 * Then, for result "+" and "-", check old/new_roots == 0 case, as in them
2273 * only on variant maybe 0.
2275 * Lastly, check result **, since there are 2 variants maybe 0, split them
2277 * But this time we don't need to consider other things, the codes and logic
2278 * is easy to understand now.
2280 static int qgroup_update_counters(struct btrfs_fs_info *fs_info,
2281 struct ulist *qgroups,
2284 u64 num_bytes, u64 seq)
2286 struct ulist_node *unode;
2287 struct ulist_iterator uiter;
2288 struct btrfs_qgroup *qg;
2289 u64 cur_new_count, cur_old_count;
2291 ULIST_ITER_INIT(&uiter);
2292 while ((unode = ulist_next(qgroups, &uiter))) {
2295 qg = unode_aux_to_qgroup(unode);
2296 cur_old_count = btrfs_qgroup_get_old_refcnt(qg, seq);
2297 cur_new_count = btrfs_qgroup_get_new_refcnt(qg, seq);
2299 trace_qgroup_update_counters(fs_info, qg, cur_old_count,
2302 /* Rfer update part */
2303 if (cur_old_count == 0 && cur_new_count > 0) {
2304 qg->rfer += num_bytes;
2305 qg->rfer_cmpr += num_bytes;
2308 if (cur_old_count > 0 && cur_new_count == 0) {
2309 qg->rfer -= num_bytes;
2310 qg->rfer_cmpr -= num_bytes;
2314 /* Excl update part */
2315 /* Exclusive/none -> shared case */
2316 if (cur_old_count == nr_old_roots &&
2317 cur_new_count < nr_new_roots) {
2318 /* Exclusive -> shared */
2319 if (cur_old_count != 0) {
2320 qg->excl -= num_bytes;
2321 qg->excl_cmpr -= num_bytes;
2326 /* Shared -> exclusive/none case */
2327 if (cur_old_count < nr_old_roots &&
2328 cur_new_count == nr_new_roots) {
2329 /* Shared->exclusive */
2330 if (cur_new_count != 0) {
2331 qg->excl += num_bytes;
2332 qg->excl_cmpr += num_bytes;
2337 /* Exclusive/none -> exclusive/none case */
2338 if (cur_old_count == nr_old_roots &&
2339 cur_new_count == nr_new_roots) {
2340 if (cur_old_count == 0) {
2341 /* None -> exclusive/none */
2343 if (cur_new_count != 0) {
2344 /* None -> exclusive */
2345 qg->excl += num_bytes;
2346 qg->excl_cmpr += num_bytes;
2349 /* None -> none, nothing changed */
2351 /* Exclusive -> exclusive/none */
2353 if (cur_new_count == 0) {
2354 /* Exclusive -> none */
2355 qg->excl -= num_bytes;
2356 qg->excl_cmpr -= num_bytes;
2359 /* Exclusive -> exclusive, nothing changed */
2364 qgroup_dirty(fs_info, qg);
2370 * Check if the @roots potentially is a list of fs tree roots
2372 * Return 0 for definitely not a fs/subvol tree roots ulist
2373 * Return 1 for possible fs/subvol tree roots in the list (considering an empty
2376 static int maybe_fs_roots(struct ulist *roots)
2378 struct ulist_node *unode;
2379 struct ulist_iterator uiter;
2381 /* Empty one, still possible for fs roots */
2382 if (!roots || roots->nnodes == 0)
2385 ULIST_ITER_INIT(&uiter);
2386 unode = ulist_next(roots, &uiter);
2391 * If it contains fs tree roots, then it must belong to fs/subvol
2393 * If it contains a non-fs tree, it won't be shared with fs/subvol trees.
2395 return is_fstree(unode->val);
2398 int btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans, u64 bytenr,
2399 u64 num_bytes, struct ulist *old_roots,
2400 struct ulist *new_roots)
2402 struct btrfs_fs_info *fs_info = trans->fs_info;
2403 struct ulist *qgroups = NULL;
2404 struct ulist *tmp = NULL;
2406 u64 nr_new_roots = 0;
2407 u64 nr_old_roots = 0;
2410 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2414 if (!maybe_fs_roots(new_roots))
2416 nr_new_roots = new_roots->nnodes;
2419 if (!maybe_fs_roots(old_roots))
2421 nr_old_roots = old_roots->nnodes;
2424 /* Quick exit, either not fs tree roots, or won't affect any qgroup */
2425 if (nr_old_roots == 0 && nr_new_roots == 0)
2428 BUG_ON(!fs_info->quota_root);
2430 trace_btrfs_qgroup_account_extent(fs_info, trans->transid, bytenr,
2431 num_bytes, nr_old_roots, nr_new_roots);
2433 qgroups = ulist_alloc(GFP_NOFS);
2438 tmp = ulist_alloc(GFP_NOFS);
2444 mutex_lock(&fs_info->qgroup_rescan_lock);
2445 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
2446 if (fs_info->qgroup_rescan_progress.objectid <= bytenr) {
2447 mutex_unlock(&fs_info->qgroup_rescan_lock);
2452 mutex_unlock(&fs_info->qgroup_rescan_lock);
2454 spin_lock(&fs_info->qgroup_lock);
2455 seq = fs_info->qgroup_seq;
2457 /* Update old refcnts using old_roots */
2458 ret = qgroup_update_refcnt(fs_info, old_roots, tmp, qgroups, seq,
2463 /* Update new refcnts using new_roots */
2464 ret = qgroup_update_refcnt(fs_info, new_roots, tmp, qgroups, seq,
2469 qgroup_update_counters(fs_info, qgroups, nr_old_roots, nr_new_roots,
2473 * Bump qgroup_seq to avoid seq overlap
2475 fs_info->qgroup_seq += max(nr_old_roots, nr_new_roots) + 1;
2477 spin_unlock(&fs_info->qgroup_lock);
2480 ulist_free(qgroups);
2481 ulist_free(old_roots);
2482 ulist_free(new_roots);
2486 int btrfs_qgroup_account_extents(struct btrfs_trans_handle *trans)
2488 struct btrfs_fs_info *fs_info = trans->fs_info;
2489 struct btrfs_qgroup_extent_record *record;
2490 struct btrfs_delayed_ref_root *delayed_refs;
2491 struct ulist *new_roots = NULL;
2492 struct rb_node *node;
2493 u64 num_dirty_extents = 0;
2497 delayed_refs = &trans->transaction->delayed_refs;
2498 qgroup_to_skip = delayed_refs->qgroup_to_skip;
2499 while ((node = rb_first(&delayed_refs->dirty_extent_root))) {
2500 record = rb_entry(node, struct btrfs_qgroup_extent_record,
2503 num_dirty_extents++;
2504 trace_btrfs_qgroup_account_extents(fs_info, record);
2508 * Old roots should be searched when inserting qgroup
2511 if (WARN_ON(!record->old_roots)) {
2512 /* Search commit root to find old_roots */
2513 ret = btrfs_find_all_roots(NULL, fs_info,
2515 &record->old_roots, false);
2521 * Use SEQ_LAST as time_seq to do special search, which
2522 * doesn't lock tree or delayed_refs and search current
2523 * root. It's safe inside commit_transaction().
2525 ret = btrfs_find_all_roots(trans, fs_info,
2526 record->bytenr, SEQ_LAST, &new_roots, false);
2529 if (qgroup_to_skip) {
2530 ulist_del(new_roots, qgroup_to_skip, 0);
2531 ulist_del(record->old_roots, qgroup_to_skip,
2534 ret = btrfs_qgroup_account_extent(trans, record->bytenr,
2538 record->old_roots = NULL;
2542 ulist_free(record->old_roots);
2543 ulist_free(new_roots);
2545 rb_erase(node, &delayed_refs->dirty_extent_root);
2549 trace_qgroup_num_dirty_extents(fs_info, trans->transid,
2555 * called from commit_transaction. Writes all changed qgroups to disk.
2557 int btrfs_run_qgroups(struct btrfs_trans_handle *trans)
2559 struct btrfs_fs_info *fs_info = trans->fs_info;
2560 struct btrfs_root *quota_root = fs_info->quota_root;
2566 spin_lock(&fs_info->qgroup_lock);
2567 while (!list_empty(&fs_info->dirty_qgroups)) {
2568 struct btrfs_qgroup *qgroup;
2569 qgroup = list_first_entry(&fs_info->dirty_qgroups,
2570 struct btrfs_qgroup, dirty);
2571 list_del_init(&qgroup->dirty);
2572 spin_unlock(&fs_info->qgroup_lock);
2573 ret = update_qgroup_info_item(trans, qgroup);
2575 fs_info->qgroup_flags |=
2576 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2577 ret = update_qgroup_limit_item(trans, qgroup);
2579 fs_info->qgroup_flags |=
2580 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2581 spin_lock(&fs_info->qgroup_lock);
2583 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2584 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_ON;
2586 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_ON;
2587 spin_unlock(&fs_info->qgroup_lock);
2589 ret = update_qgroup_status_item(trans);
2591 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2597 * Copy the accounting information between qgroups. This is necessary
2598 * when a snapshot or a subvolume is created. Throwing an error will
2599 * cause a transaction abort so we take extra care here to only error
2600 * when a readonly fs is a reasonable outcome.
2602 int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans, u64 srcid,
2603 u64 objectid, struct btrfs_qgroup_inherit *inherit)
2608 struct btrfs_fs_info *fs_info = trans->fs_info;
2609 struct btrfs_root *quota_root;
2610 struct btrfs_qgroup *srcgroup;
2611 struct btrfs_qgroup *dstgroup;
2615 mutex_lock(&fs_info->qgroup_ioctl_lock);
2616 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
2619 quota_root = fs_info->quota_root;
2626 i_qgroups = (u64 *)(inherit + 1);
2627 nums = inherit->num_qgroups + 2 * inherit->num_ref_copies +
2628 2 * inherit->num_excl_copies;
2629 for (i = 0; i < nums; ++i) {
2630 srcgroup = find_qgroup_rb(fs_info, *i_qgroups);
2633 * Zero out invalid groups so we can ignore
2637 ((srcgroup->qgroupid >> 48) <= (objectid >> 48)))
2645 * create a tracking group for the subvol itself
2647 ret = add_qgroup_item(trans, quota_root, objectid);
2652 * add qgroup to all inherited groups
2655 i_qgroups = (u64 *)(inherit + 1);
2656 for (i = 0; i < inherit->num_qgroups; ++i, ++i_qgroups) {
2657 if (*i_qgroups == 0)
2659 ret = add_qgroup_relation_item(trans, objectid,
2661 if (ret && ret != -EEXIST)
2663 ret = add_qgroup_relation_item(trans, *i_qgroups,
2665 if (ret && ret != -EEXIST)
2672 spin_lock(&fs_info->qgroup_lock);
2674 dstgroup = add_qgroup_rb(fs_info, objectid);
2675 if (IS_ERR(dstgroup)) {
2676 ret = PTR_ERR(dstgroup);
2680 if (inherit && inherit->flags & BTRFS_QGROUP_INHERIT_SET_LIMITS) {
2681 dstgroup->lim_flags = inherit->lim.flags;
2682 dstgroup->max_rfer = inherit->lim.max_rfer;
2683 dstgroup->max_excl = inherit->lim.max_excl;
2684 dstgroup->rsv_rfer = inherit->lim.rsv_rfer;
2685 dstgroup->rsv_excl = inherit->lim.rsv_excl;
2687 ret = update_qgroup_limit_item(trans, dstgroup);
2689 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
2691 "unable to update quota limit for %llu",
2692 dstgroup->qgroupid);
2698 srcgroup = find_qgroup_rb(fs_info, srcid);
2703 * We call inherit after we clone the root in order to make sure
2704 * our counts don't go crazy, so at this point the only
2705 * difference between the two roots should be the root node.
2707 level_size = fs_info->nodesize;
2708 dstgroup->rfer = srcgroup->rfer;
2709 dstgroup->rfer_cmpr = srcgroup->rfer_cmpr;
2710 dstgroup->excl = level_size;
2711 dstgroup->excl_cmpr = level_size;
2712 srcgroup->excl = level_size;
2713 srcgroup->excl_cmpr = level_size;
2715 /* inherit the limit info */
2716 dstgroup->lim_flags = srcgroup->lim_flags;
2717 dstgroup->max_rfer = srcgroup->max_rfer;
2718 dstgroup->max_excl = srcgroup->max_excl;
2719 dstgroup->rsv_rfer = srcgroup->rsv_rfer;
2720 dstgroup->rsv_excl = srcgroup->rsv_excl;
2722 qgroup_dirty(fs_info, dstgroup);
2723 qgroup_dirty(fs_info, srcgroup);
2729 i_qgroups = (u64 *)(inherit + 1);
2730 for (i = 0; i < inherit->num_qgroups; ++i) {
2732 ret = add_relation_rb(fs_info, objectid, *i_qgroups);
2739 for (i = 0; i < inherit->num_ref_copies; ++i, i_qgroups += 2) {
2740 struct btrfs_qgroup *src;
2741 struct btrfs_qgroup *dst;
2743 if (!i_qgroups[0] || !i_qgroups[1])
2746 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2747 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2754 dst->rfer = src->rfer - level_size;
2755 dst->rfer_cmpr = src->rfer_cmpr - level_size;
2757 for (i = 0; i < inherit->num_excl_copies; ++i, i_qgroups += 2) {
2758 struct btrfs_qgroup *src;
2759 struct btrfs_qgroup *dst;
2761 if (!i_qgroups[0] || !i_qgroups[1])
2764 src = find_qgroup_rb(fs_info, i_qgroups[0]);
2765 dst = find_qgroup_rb(fs_info, i_qgroups[1]);
2772 dst->excl = src->excl + level_size;
2773 dst->excl_cmpr = src->excl_cmpr + level_size;
2777 spin_unlock(&fs_info->qgroup_lock);
2779 mutex_unlock(&fs_info->qgroup_ioctl_lock);
2784 * Two limits to commit transaction in advance.
2786 * For RATIO, it will be 1/RATIO of the remaining limit
2787 * (excluding data and prealloc meta) as threshold.
2788 * For SIZE, it will be in byte unit as threshold.
2790 #define QGROUP_PERTRANS_RATIO 32
2791 #define QGROUP_PERTRANS_SIZE SZ_32M
2792 static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
2793 const struct btrfs_qgroup *qg, u64 num_bytes)
2798 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
2799 qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
2802 if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL) &&
2803 qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
2807 * Even if we passed the check, it's better to check if reservation
2808 * for meta_pertrans is pushing us near limit.
2809 * If there is too much pertrans reservation or it's near the limit,
2810 * let's try commit transaction to free some, using transaction_kthread
2812 if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
2813 BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
2814 if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
2815 limit = qg->max_excl;
2817 limit = qg->max_rfer;
2818 threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
2819 qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
2820 QGROUP_PERTRANS_RATIO;
2821 threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
2824 * Use transaction_kthread to commit transaction, so we no
2825 * longer need to bother nested transaction nor lock context.
2827 if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
2828 btrfs_commit_transaction_locksafe(fs_info);
2834 static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
2835 enum btrfs_qgroup_rsv_type type)
2837 struct btrfs_root *quota_root;
2838 struct btrfs_qgroup *qgroup;
2839 struct btrfs_fs_info *fs_info = root->fs_info;
2840 u64 ref_root = root->root_key.objectid;
2842 struct ulist_node *unode;
2843 struct ulist_iterator uiter;
2845 if (!is_fstree(ref_root))
2851 if (test_bit(BTRFS_FS_QUOTA_OVERRIDE, &fs_info->flags) &&
2852 capable(CAP_SYS_RESOURCE))
2855 spin_lock(&fs_info->qgroup_lock);
2856 quota_root = fs_info->quota_root;
2860 qgroup = find_qgroup_rb(fs_info, ref_root);
2865 * in a first step, we check all affected qgroups if any limits would
2868 ulist_reinit(fs_info->qgroup_ulist);
2869 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2870 qgroup_to_aux(qgroup), GFP_ATOMIC);
2873 ULIST_ITER_INIT(&uiter);
2874 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2875 struct btrfs_qgroup *qg;
2876 struct btrfs_qgroup_list *glist;
2878 qg = unode_aux_to_qgroup(unode);
2880 if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
2885 list_for_each_entry(glist, &qg->groups, next_group) {
2886 ret = ulist_add(fs_info->qgroup_ulist,
2887 glist->group->qgroupid,
2888 qgroup_to_aux(glist->group), GFP_ATOMIC);
2895 * no limits exceeded, now record the reservation into all qgroups
2897 ULIST_ITER_INIT(&uiter);
2898 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2899 struct btrfs_qgroup *qg;
2901 qg = unode_aux_to_qgroup(unode);
2903 trace_qgroup_update_reserve(fs_info, qg, num_bytes, type);
2904 qgroup_rsv_add(fs_info, qg, num_bytes, type);
2908 spin_unlock(&fs_info->qgroup_lock);
2913 * Free @num_bytes of reserved space with @type for qgroup. (Normally level 0
2916 * Will handle all higher level qgroup too.
2918 * NOTE: If @num_bytes is (u64)-1, this means to free all bytes of this qgroup.
2919 * This special case is only used for META_PERTRANS type.
2921 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
2922 u64 ref_root, u64 num_bytes,
2923 enum btrfs_qgroup_rsv_type type)
2925 struct btrfs_root *quota_root;
2926 struct btrfs_qgroup *qgroup;
2927 struct ulist_node *unode;
2928 struct ulist_iterator uiter;
2931 if (!is_fstree(ref_root))
2937 if (num_bytes == (u64)-1 && type != BTRFS_QGROUP_RSV_META_PERTRANS) {
2938 WARN(1, "%s: Invalid type to free", __func__);
2941 spin_lock(&fs_info->qgroup_lock);
2943 quota_root = fs_info->quota_root;
2947 qgroup = find_qgroup_rb(fs_info, ref_root);
2951 if (num_bytes == (u64)-1)
2953 * We're freeing all pertrans rsv, get reserved value from
2954 * level 0 qgroup as real num_bytes to free.
2956 num_bytes = qgroup->rsv.values[type];
2958 ulist_reinit(fs_info->qgroup_ulist);
2959 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
2960 qgroup_to_aux(qgroup), GFP_ATOMIC);
2963 ULIST_ITER_INIT(&uiter);
2964 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
2965 struct btrfs_qgroup *qg;
2966 struct btrfs_qgroup_list *glist;
2968 qg = unode_aux_to_qgroup(unode);
2970 trace_qgroup_update_reserve(fs_info, qg, -(s64)num_bytes, type);
2971 qgroup_rsv_release(fs_info, qg, num_bytes, type);
2973 list_for_each_entry(glist, &qg->groups, next_group) {
2974 ret = ulist_add(fs_info->qgroup_ulist,
2975 glist->group->qgroupid,
2976 qgroup_to_aux(glist->group), GFP_ATOMIC);
2983 spin_unlock(&fs_info->qgroup_lock);
2987 * Check if the leaf is the last leaf. Which means all node pointers
2988 * are at their last position.
2990 static bool is_last_leaf(struct btrfs_path *path)
2994 for (i = 1; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
2995 if (path->slots[i] != btrfs_header_nritems(path->nodes[i]) - 1)
3002 * returns < 0 on error, 0 when more leafs are to be scanned.
3003 * returns 1 when done.
3005 static int qgroup_rescan_leaf(struct btrfs_trans_handle *trans,
3006 struct btrfs_path *path)
3008 struct btrfs_fs_info *fs_info = trans->fs_info;
3009 struct btrfs_key found;
3010 struct extent_buffer *scratch_leaf = NULL;
3011 struct ulist *roots = NULL;
3017 mutex_lock(&fs_info->qgroup_rescan_lock);
3018 ret = btrfs_search_slot_for_read(fs_info->extent_root,
3019 &fs_info->qgroup_rescan_progress,
3022 btrfs_debug(fs_info,
3023 "current progress key (%llu %u %llu), search_slot ret %d",
3024 fs_info->qgroup_rescan_progress.objectid,
3025 fs_info->qgroup_rescan_progress.type,
3026 fs_info->qgroup_rescan_progress.offset, ret);
3030 * The rescan is about to end, we will not be scanning any
3031 * further blocks. We cannot unset the RESCAN flag here, because
3032 * we want to commit the transaction if everything went well.
3033 * To make the live accounting work in this phase, we set our
3034 * scan progress pointer such that every real extent objectid
3037 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3038 btrfs_release_path(path);
3039 mutex_unlock(&fs_info->qgroup_rescan_lock);
3042 done = is_last_leaf(path);
3044 btrfs_item_key_to_cpu(path->nodes[0], &found,
3045 btrfs_header_nritems(path->nodes[0]) - 1);
3046 fs_info->qgroup_rescan_progress.objectid = found.objectid + 1;
3048 scratch_leaf = btrfs_clone_extent_buffer(path->nodes[0]);
3049 if (!scratch_leaf) {
3051 mutex_unlock(&fs_info->qgroup_rescan_lock);
3054 slot = path->slots[0];
3055 btrfs_release_path(path);
3056 mutex_unlock(&fs_info->qgroup_rescan_lock);
3058 for (; slot < btrfs_header_nritems(scratch_leaf); ++slot) {
3059 btrfs_item_key_to_cpu(scratch_leaf, &found, slot);
3060 if (found.type != BTRFS_EXTENT_ITEM_KEY &&
3061 found.type != BTRFS_METADATA_ITEM_KEY)
3063 if (found.type == BTRFS_METADATA_ITEM_KEY)
3064 num_bytes = fs_info->nodesize;
3066 num_bytes = found.offset;
3068 ret = btrfs_find_all_roots(NULL, fs_info, found.objectid, 0,
3072 /* For rescan, just pass old_roots as NULL */
3073 ret = btrfs_qgroup_account_extent(trans, found.objectid,
3074 num_bytes, NULL, roots);
3080 free_extent_buffer(scratch_leaf);
3084 fs_info->qgroup_rescan_progress.objectid = (u64)-1;
3089 static void btrfs_qgroup_rescan_worker(struct btrfs_work *work)
3091 struct btrfs_fs_info *fs_info = container_of(work, struct btrfs_fs_info,
3092 qgroup_rescan_work);
3093 struct btrfs_path *path;
3094 struct btrfs_trans_handle *trans = NULL;
3098 path = btrfs_alloc_path();
3102 * Rescan should only search for commit root, and any later difference
3103 * should be recorded by qgroup
3105 path->search_commit_root = 1;
3106 path->skip_locking = 1;
3109 while (!err && !btrfs_fs_closing(fs_info)) {
3110 trans = btrfs_start_transaction(fs_info->fs_root, 0);
3111 if (IS_ERR(trans)) {
3112 err = PTR_ERR(trans);
3115 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
3118 err = qgroup_rescan_leaf(trans, path);
3121 btrfs_commit_transaction(trans);
3123 btrfs_end_transaction(trans);
3127 btrfs_free_path(path);
3129 mutex_lock(&fs_info->qgroup_rescan_lock);
3130 if (!btrfs_fs_closing(fs_info))
3131 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3134 fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT) {
3135 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3136 } else if (err < 0) {
3137 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3139 mutex_unlock(&fs_info->qgroup_rescan_lock);
3142 * only update status, since the previous part has already updated the
3145 trans = btrfs_start_transaction(fs_info->quota_root, 1);
3146 if (IS_ERR(trans)) {
3147 err = PTR_ERR(trans);
3149 "fail to start transaction for status update: %d",
3153 ret = update_qgroup_status_item(trans);
3156 btrfs_err(fs_info, "fail to update qgroup status: %d", err);
3158 btrfs_end_transaction(trans);
3160 if (btrfs_fs_closing(fs_info)) {
3161 btrfs_info(fs_info, "qgroup scan paused");
3162 } else if (err >= 0) {
3163 btrfs_info(fs_info, "qgroup scan completed%s",
3164 err > 0 ? " (inconsistency flag cleared)" : "");
3166 btrfs_err(fs_info, "qgroup scan failed with %d", err);
3170 mutex_lock(&fs_info->qgroup_rescan_lock);
3171 fs_info->qgroup_rescan_running = false;
3172 mutex_unlock(&fs_info->qgroup_rescan_lock);
3173 complete_all(&fs_info->qgroup_rescan_completion);
3177 * Checks that (a) no rescan is running and (b) quota is enabled. Allocates all
3178 * memory required for the rescan context.
3181 qgroup_rescan_init(struct btrfs_fs_info *fs_info, u64 progress_objectid,
3187 /* we're resuming qgroup rescan at mount time */
3188 if (!(fs_info->qgroup_flags &
3189 BTRFS_QGROUP_STATUS_FLAG_RESCAN)) {
3191 "qgroup rescan init failed, qgroup is not enabled");
3193 } else if (!(fs_info->qgroup_flags &
3194 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3196 "qgroup rescan init failed, qgroup rescan is not queued");
3204 mutex_lock(&fs_info->qgroup_rescan_lock);
3205 spin_lock(&fs_info->qgroup_lock);
3208 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN) {
3210 "qgroup rescan is already in progress");
3212 } else if (!(fs_info->qgroup_flags &
3213 BTRFS_QGROUP_STATUS_FLAG_ON)) {
3215 "qgroup rescan init failed, qgroup is not enabled");
3220 spin_unlock(&fs_info->qgroup_lock);
3221 mutex_unlock(&fs_info->qgroup_rescan_lock);
3224 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3227 memset(&fs_info->qgroup_rescan_progress, 0,
3228 sizeof(fs_info->qgroup_rescan_progress));
3229 fs_info->qgroup_rescan_progress.objectid = progress_objectid;
3230 init_completion(&fs_info->qgroup_rescan_completion);
3231 fs_info->qgroup_rescan_running = true;
3233 spin_unlock(&fs_info->qgroup_lock);
3234 mutex_unlock(&fs_info->qgroup_rescan_lock);
3236 memset(&fs_info->qgroup_rescan_work, 0,
3237 sizeof(fs_info->qgroup_rescan_work));
3238 btrfs_init_work(&fs_info->qgroup_rescan_work,
3239 btrfs_qgroup_rescan_helper,
3240 btrfs_qgroup_rescan_worker, NULL, NULL);
3245 qgroup_rescan_zero_tracking(struct btrfs_fs_info *fs_info)
3248 struct btrfs_qgroup *qgroup;
3250 spin_lock(&fs_info->qgroup_lock);
3251 /* clear all current qgroup tracking information */
3252 for (n = rb_first(&fs_info->qgroup_tree); n; n = rb_next(n)) {
3253 qgroup = rb_entry(n, struct btrfs_qgroup, node);
3255 qgroup->rfer_cmpr = 0;
3257 qgroup->excl_cmpr = 0;
3258 qgroup_dirty(fs_info, qgroup);
3260 spin_unlock(&fs_info->qgroup_lock);
3264 btrfs_qgroup_rescan(struct btrfs_fs_info *fs_info)
3267 struct btrfs_trans_handle *trans;
3269 ret = qgroup_rescan_init(fs_info, 0, 1);
3274 * We have set the rescan_progress to 0, which means no more
3275 * delayed refs will be accounted by btrfs_qgroup_account_ref.
3276 * However, btrfs_qgroup_account_ref may be right after its call
3277 * to btrfs_find_all_roots, in which case it would still do the
3279 * To solve this, we're committing the transaction, which will
3280 * ensure we run all delayed refs and only after that, we are
3281 * going to clear all tracking information for a clean start.
3284 trans = btrfs_join_transaction(fs_info->fs_root);
3285 if (IS_ERR(trans)) {
3286 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3287 return PTR_ERR(trans);
3289 ret = btrfs_commit_transaction(trans);
3291 fs_info->qgroup_flags &= ~BTRFS_QGROUP_STATUS_FLAG_RESCAN;
3295 qgroup_rescan_zero_tracking(fs_info);
3297 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3298 &fs_info->qgroup_rescan_work);
3303 int btrfs_qgroup_wait_for_completion(struct btrfs_fs_info *fs_info,
3309 mutex_lock(&fs_info->qgroup_rescan_lock);
3310 spin_lock(&fs_info->qgroup_lock);
3311 running = fs_info->qgroup_rescan_running;
3312 spin_unlock(&fs_info->qgroup_lock);
3313 mutex_unlock(&fs_info->qgroup_rescan_lock);
3319 ret = wait_for_completion_interruptible(
3320 &fs_info->qgroup_rescan_completion);
3322 wait_for_completion(&fs_info->qgroup_rescan_completion);
3328 * this is only called from open_ctree where we're still single threaded, thus
3329 * locking is omitted here.
3332 btrfs_qgroup_rescan_resume(struct btrfs_fs_info *fs_info)
3334 if (fs_info->qgroup_flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN)
3335 btrfs_queue_work(fs_info->qgroup_rescan_workers,
3336 &fs_info->qgroup_rescan_work);
3340 * Reserve qgroup space for range [start, start + len).
3342 * This function will either reserve space from related qgroups or doing
3343 * nothing if the range is already reserved.
3345 * Return 0 for successful reserve
3346 * Return <0 for error (including -EQUOT)
3348 * NOTE: this function may sleep for memory allocation.
3349 * if btrfs_qgroup_reserve_data() is called multiple times with
3350 * same @reserved, caller must ensure when error happens it's OK
3351 * to free *ALL* reserved space.
3353 int btrfs_qgroup_reserve_data(struct inode *inode,
3354 struct extent_changeset **reserved_ret, u64 start,
3357 struct btrfs_root *root = BTRFS_I(inode)->root;
3358 struct ulist_node *unode;
3359 struct ulist_iterator uiter;
3360 struct extent_changeset *reserved;
3365 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &root->fs_info->flags) ||
3366 !is_fstree(root->root_key.objectid) || len == 0)
3369 /* @reserved parameter is mandatory for qgroup */
3370 if (WARN_ON(!reserved_ret))
3372 if (!*reserved_ret) {
3373 *reserved_ret = extent_changeset_alloc();
3377 reserved = *reserved_ret;
3378 /* Record already reserved space */
3379 orig_reserved = reserved->bytes_changed;
3380 ret = set_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3381 start + len -1, EXTENT_QGROUP_RESERVED, reserved);
3383 /* Newly reserved space */
3384 to_reserve = reserved->bytes_changed - orig_reserved;
3385 trace_btrfs_qgroup_reserve_data(inode, start, len,
3386 to_reserve, QGROUP_RESERVE);
3389 ret = qgroup_reserve(root, to_reserve, true, BTRFS_QGROUP_RSV_DATA);
3396 /* cleanup *ALL* already reserved ranges */
3397 ULIST_ITER_INIT(&uiter);
3398 while ((unode = ulist_next(&reserved->range_changed, &uiter)))
3399 clear_extent_bit(&BTRFS_I(inode)->io_tree, unode->val,
3400 unode->aux, EXTENT_QGROUP_RESERVED, 0, 0, NULL);
3401 extent_changeset_release(reserved);
3405 /* Free ranges specified by @reserved, normally in error path */
3406 static int qgroup_free_reserved_data(struct inode *inode,
3407 struct extent_changeset *reserved, u64 start, u64 len)
3409 struct btrfs_root *root = BTRFS_I(inode)->root;
3410 struct ulist_node *unode;
3411 struct ulist_iterator uiter;
3412 struct extent_changeset changeset;
3416 extent_changeset_init(&changeset);
3417 len = round_up(start + len, root->fs_info->sectorsize);
3418 start = round_down(start, root->fs_info->sectorsize);
3420 ULIST_ITER_INIT(&uiter);
3421 while ((unode = ulist_next(&reserved->range_changed, &uiter))) {
3422 u64 range_start = unode->val;
3423 /* unode->aux is the inclusive end */
3424 u64 range_len = unode->aux - range_start + 1;
3428 extent_changeset_release(&changeset);
3430 /* Only free range in range [start, start + len) */
3431 if (range_start >= start + len ||
3432 range_start + range_len <= start)
3434 free_start = max(range_start, start);
3435 free_len = min(start + len, range_start + range_len) -
3438 * TODO: To also modify reserved->ranges_reserved to reflect
3441 * However as long as we free qgroup reserved according to
3442 * EXTENT_QGROUP_RESERVED, we won't double free.
3443 * So not need to rush.
3445 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_failure_tree,
3446 free_start, free_start + free_len - 1,
3447 EXTENT_QGROUP_RESERVED, &changeset);
3450 freed += changeset.bytes_changed;
3452 btrfs_qgroup_free_refroot(root->fs_info, root->root_key.objectid, freed,
3453 BTRFS_QGROUP_RSV_DATA);
3456 extent_changeset_release(&changeset);
3460 static int __btrfs_qgroup_release_data(struct inode *inode,
3461 struct extent_changeset *reserved, u64 start, u64 len,
3464 struct extent_changeset changeset;
3465 int trace_op = QGROUP_RELEASE;
3468 if (!test_bit(BTRFS_FS_QUOTA_ENABLED,
3469 &BTRFS_I(inode)->root->fs_info->flags))
3472 /* In release case, we shouldn't have @reserved */
3473 WARN_ON(!free && reserved);
3474 if (free && reserved)
3475 return qgroup_free_reserved_data(inode, reserved, start, len);
3476 extent_changeset_init(&changeset);
3477 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, start,
3478 start + len -1, EXTENT_QGROUP_RESERVED, &changeset);
3483 trace_op = QGROUP_FREE;
3484 trace_btrfs_qgroup_release_data(inode, start, len,
3485 changeset.bytes_changed, trace_op);
3487 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3488 BTRFS_I(inode)->root->root_key.objectid,
3489 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3490 ret = changeset.bytes_changed;
3492 extent_changeset_release(&changeset);
3497 * Free a reserved space range from io_tree and related qgroups
3499 * Should be called when a range of pages get invalidated before reaching disk.
3500 * Or for error cleanup case.
3501 * if @reserved is given, only reserved range in [@start, @start + @len) will
3504 * For data written to disk, use btrfs_qgroup_release_data().
3506 * NOTE: This function may sleep for memory allocation.
3508 int btrfs_qgroup_free_data(struct inode *inode,
3509 struct extent_changeset *reserved, u64 start, u64 len)
3511 return __btrfs_qgroup_release_data(inode, reserved, start, len, 1);
3515 * Release a reserved space range from io_tree only.
3517 * Should be called when a range of pages get written to disk and corresponding
3518 * FILE_EXTENT is inserted into corresponding root.
3520 * Since new qgroup accounting framework will only update qgroup numbers at
3521 * commit_transaction() time, its reserved space shouldn't be freed from
3524 * But we should release the range from io_tree, to allow further write to be
3527 * NOTE: This function may sleep for memory allocation.
3529 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len)
3531 return __btrfs_qgroup_release_data(inode, NULL, start, len, 0);
3534 static void add_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3535 enum btrfs_qgroup_rsv_type type)
3537 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3538 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3543 spin_lock(&root->qgroup_meta_rsv_lock);
3544 if (type == BTRFS_QGROUP_RSV_META_PREALLOC)
3545 root->qgroup_meta_rsv_prealloc += num_bytes;
3547 root->qgroup_meta_rsv_pertrans += num_bytes;
3548 spin_unlock(&root->qgroup_meta_rsv_lock);
3551 static int sub_root_meta_rsv(struct btrfs_root *root, int num_bytes,
3552 enum btrfs_qgroup_rsv_type type)
3554 if (type != BTRFS_QGROUP_RSV_META_PREALLOC &&
3555 type != BTRFS_QGROUP_RSV_META_PERTRANS)
3560 spin_lock(&root->qgroup_meta_rsv_lock);
3561 if (type == BTRFS_QGROUP_RSV_META_PREALLOC) {
3562 num_bytes = min_t(u64, root->qgroup_meta_rsv_prealloc,
3564 root->qgroup_meta_rsv_prealloc -= num_bytes;
3566 num_bytes = min_t(u64, root->qgroup_meta_rsv_pertrans,
3568 root->qgroup_meta_rsv_pertrans -= num_bytes;
3570 spin_unlock(&root->qgroup_meta_rsv_lock);
3574 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
3575 enum btrfs_qgroup_rsv_type type, bool enforce)
3577 struct btrfs_fs_info *fs_info = root->fs_info;
3580 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3581 !is_fstree(root->root_key.objectid) || num_bytes == 0)
3584 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3585 trace_qgroup_meta_reserve(root, type, (s64)num_bytes);
3586 ret = qgroup_reserve(root, num_bytes, enforce, type);
3590 * Record what we have reserved into root.
3592 * To avoid quota disabled->enabled underflow.
3593 * In that case, we may try to free space we haven't reserved
3594 * (since quota was disabled), so record what we reserved into root.
3595 * And ensure later release won't underflow this number.
3597 add_root_meta_rsv(root, num_bytes, type);
3601 void btrfs_qgroup_free_meta_all_pertrans(struct btrfs_root *root)
3603 struct btrfs_fs_info *fs_info = root->fs_info;
3605 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3606 !is_fstree(root->root_key.objectid))
3609 /* TODO: Update trace point to handle such free */
3610 trace_qgroup_meta_free_all_pertrans(root);
3611 /* Special value -1 means to free all reserved space */
3612 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid, (u64)-1,
3613 BTRFS_QGROUP_RSV_META_PERTRANS);
3616 void __btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes,
3617 enum btrfs_qgroup_rsv_type type)
3619 struct btrfs_fs_info *fs_info = root->fs_info;
3621 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3622 !is_fstree(root->root_key.objectid))
3626 * reservation for META_PREALLOC can happen before quota is enabled,
3627 * which can lead to underflow.
3628 * Here ensure we will only free what we really have reserved.
3630 num_bytes = sub_root_meta_rsv(root, num_bytes, type);
3631 BUG_ON(num_bytes != round_down(num_bytes, fs_info->nodesize));
3632 trace_qgroup_meta_reserve(root, type, -(s64)num_bytes);
3633 btrfs_qgroup_free_refroot(fs_info, root->root_key.objectid,
3637 static void qgroup_convert_meta(struct btrfs_fs_info *fs_info, u64 ref_root,
3640 struct btrfs_root *quota_root = fs_info->quota_root;
3641 struct btrfs_qgroup *qgroup;
3642 struct ulist_node *unode;
3643 struct ulist_iterator uiter;
3651 spin_lock(&fs_info->qgroup_lock);
3652 qgroup = find_qgroup_rb(fs_info, ref_root);
3655 ulist_reinit(fs_info->qgroup_ulist);
3656 ret = ulist_add(fs_info->qgroup_ulist, qgroup->qgroupid,
3657 qgroup_to_aux(qgroup), GFP_ATOMIC);
3660 ULIST_ITER_INIT(&uiter);
3661 while ((unode = ulist_next(fs_info->qgroup_ulist, &uiter))) {
3662 struct btrfs_qgroup *qg;
3663 struct btrfs_qgroup_list *glist;
3665 qg = unode_aux_to_qgroup(unode);
3667 qgroup_rsv_release(fs_info, qg, num_bytes,
3668 BTRFS_QGROUP_RSV_META_PREALLOC);
3669 qgroup_rsv_add(fs_info, qg, num_bytes,
3670 BTRFS_QGROUP_RSV_META_PERTRANS);
3671 list_for_each_entry(glist, &qg->groups, next_group) {
3672 ret = ulist_add(fs_info->qgroup_ulist,
3673 glist->group->qgroupid,
3674 qgroup_to_aux(glist->group), GFP_ATOMIC);
3680 spin_unlock(&fs_info->qgroup_lock);
3683 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes)
3685 struct btrfs_fs_info *fs_info = root->fs_info;
3687 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) ||
3688 !is_fstree(root->root_key.objectid))
3690 /* Same as btrfs_qgroup_free_meta_prealloc() */
3691 num_bytes = sub_root_meta_rsv(root, num_bytes,
3692 BTRFS_QGROUP_RSV_META_PREALLOC);
3693 trace_qgroup_meta_convert(root, num_bytes);
3694 qgroup_convert_meta(fs_info, root->root_key.objectid, num_bytes);
3698 * Check qgroup reserved space leaking, normally at destroy inode
3701 void btrfs_qgroup_check_reserved_leak(struct inode *inode)
3703 struct extent_changeset changeset;
3704 struct ulist_node *unode;
3705 struct ulist_iterator iter;
3708 extent_changeset_init(&changeset);
3709 ret = clear_record_extent_bits(&BTRFS_I(inode)->io_tree, 0, (u64)-1,
3710 EXTENT_QGROUP_RESERVED, &changeset);
3713 if (WARN_ON(changeset.bytes_changed)) {
3714 ULIST_ITER_INIT(&iter);
3715 while ((unode = ulist_next(&changeset.range_changed, &iter))) {
3716 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3717 "leaking qgroup reserved space, ino: %lu, start: %llu, end: %llu",
3718 inode->i_ino, unode->val, unode->aux);
3720 btrfs_qgroup_free_refroot(BTRFS_I(inode)->root->fs_info,
3721 BTRFS_I(inode)->root->root_key.objectid,
3722 changeset.bytes_changed, BTRFS_QGROUP_RSV_DATA);
3725 extent_changeset_release(&changeset);
3728 void btrfs_qgroup_init_swapped_blocks(
3729 struct btrfs_qgroup_swapped_blocks *swapped_blocks)
3733 spin_lock_init(&swapped_blocks->lock);
3734 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3735 swapped_blocks->blocks[i] = RB_ROOT;
3736 swapped_blocks->swapped = false;
3740 * Delete all swapped blocks record of @root.
3741 * Every record here means we skipped a full subtree scan for qgroup.
3743 * Gets called when committing one transaction.
3745 void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root)
3747 struct btrfs_qgroup_swapped_blocks *swapped_blocks;
3750 swapped_blocks = &root->swapped_blocks;
3752 spin_lock(&swapped_blocks->lock);
3753 if (!swapped_blocks->swapped)
3755 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3756 struct rb_root *cur_root = &swapped_blocks->blocks[i];
3757 struct btrfs_qgroup_swapped_block *entry;
3758 struct btrfs_qgroup_swapped_block *next;
3760 rbtree_postorder_for_each_entry_safe(entry, next, cur_root,
3763 swapped_blocks->blocks[i] = RB_ROOT;
3765 swapped_blocks->swapped = false;
3767 spin_unlock(&swapped_blocks->lock);
3771 * Add subtree roots record into @subvol_root.
3773 * @subvol_root: tree root of the subvolume tree get swapped
3774 * @bg: block group under balance
3775 * @subvol_parent/slot: pointer to the subtree root in subvolume tree
3776 * @reloc_parent/slot: pointer to the subtree root in reloc tree
3777 * BOTH POINTERS ARE BEFORE TREE SWAP
3778 * @last_snapshot: last snapshot generation of the subvolume tree
3780 int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
3781 struct btrfs_root *subvol_root,
3782 struct btrfs_block_group_cache *bg,
3783 struct extent_buffer *subvol_parent, int subvol_slot,
3784 struct extent_buffer *reloc_parent, int reloc_slot,
3787 struct btrfs_fs_info *fs_info = subvol_root->fs_info;
3788 struct btrfs_qgroup_swapped_blocks *blocks = &subvol_root->swapped_blocks;
3789 struct btrfs_qgroup_swapped_block *block;
3790 struct rb_node **cur;
3791 struct rb_node *parent = NULL;
3792 int level = btrfs_header_level(subvol_parent) - 1;
3795 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3798 if (btrfs_node_ptr_generation(subvol_parent, subvol_slot) >
3799 btrfs_node_ptr_generation(reloc_parent, reloc_slot)) {
3800 btrfs_err_rl(fs_info,
3801 "%s: bad parameter order, subvol_gen=%llu reloc_gen=%llu",
3803 btrfs_node_ptr_generation(subvol_parent, subvol_slot),
3804 btrfs_node_ptr_generation(reloc_parent, reloc_slot));
3808 block = kmalloc(sizeof(*block), GFP_NOFS);
3815 * @reloc_parent/slot is still before swap, while @block is going to
3816 * record the bytenr after swap, so we do the swap here.
3818 block->subvol_bytenr = btrfs_node_blockptr(reloc_parent, reloc_slot);
3819 block->subvol_generation = btrfs_node_ptr_generation(reloc_parent,
3821 block->reloc_bytenr = btrfs_node_blockptr(subvol_parent, subvol_slot);
3822 block->reloc_generation = btrfs_node_ptr_generation(subvol_parent,
3824 block->last_snapshot = last_snapshot;
3825 block->level = level;
3826 if (bg->flags & BTRFS_BLOCK_GROUP_DATA)
3827 block->trace_leaf = true;
3829 block->trace_leaf = false;
3830 btrfs_node_key_to_cpu(reloc_parent, &block->first_key, reloc_slot);
3832 /* Insert @block into @blocks */
3833 spin_lock(&blocks->lock);
3834 cur = &blocks->blocks[level].rb_node;
3836 struct btrfs_qgroup_swapped_block *entry;
3839 entry = rb_entry(parent, struct btrfs_qgroup_swapped_block,
3842 if (entry->subvol_bytenr < block->subvol_bytenr) {
3843 cur = &(*cur)->rb_left;
3844 } else if (entry->subvol_bytenr > block->subvol_bytenr) {
3845 cur = &(*cur)->rb_right;
3847 if (entry->subvol_generation !=
3848 block->subvol_generation ||
3849 entry->reloc_bytenr != block->reloc_bytenr ||
3850 entry->reloc_generation !=
3851 block->reloc_generation) {
3853 * Duplicated but mismatch entry found.
3856 * Marking qgroup inconsistent should be enough
3859 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
3866 rb_link_node(&block->node, parent, cur);
3867 rb_insert_color(&block->node, &blocks->blocks[level]);
3868 blocks->swapped = true;
3870 spin_unlock(&blocks->lock);
3873 fs_info->qgroup_flags |=
3874 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;
3879 * Check if the tree block is a subtree root, and if so do the needed
3880 * delayed subtree trace for qgroup.
3882 * This is called during btrfs_cow_block().
3884 int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
3885 struct btrfs_root *root,
3886 struct extent_buffer *subvol_eb)
3888 struct btrfs_fs_info *fs_info = root->fs_info;
3889 struct btrfs_qgroup_swapped_blocks *blocks = &root->swapped_blocks;
3890 struct btrfs_qgroup_swapped_block *block;
3891 struct extent_buffer *reloc_eb = NULL;
3892 struct rb_node *node;
3894 bool swapped = false;
3895 int level = btrfs_header_level(subvol_eb);
3899 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
3901 if (!is_fstree(root->root_key.objectid) || !root->reloc_root)
3904 spin_lock(&blocks->lock);
3905 if (!blocks->swapped) {
3906 spin_unlock(&blocks->lock);
3909 node = blocks->blocks[level].rb_node;
3912 block = rb_entry(node, struct btrfs_qgroup_swapped_block, node);
3913 if (block->subvol_bytenr < subvol_eb->start) {
3914 node = node->rb_left;
3915 } else if (block->subvol_bytenr > subvol_eb->start) {
3916 node = node->rb_right;
3923 spin_unlock(&blocks->lock);
3926 /* Found one, remove it from @blocks first and update blocks->swapped */
3927 rb_erase(&block->node, &blocks->blocks[level]);
3928 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3929 if (RB_EMPTY_ROOT(&blocks->blocks[i])) {
3934 blocks->swapped = swapped;
3935 spin_unlock(&blocks->lock);
3937 /* Read out reloc subtree root */
3938 reloc_eb = read_tree_block(fs_info, block->reloc_bytenr,
3939 block->reloc_generation, block->level,
3941 if (IS_ERR(reloc_eb)) {
3942 ret = PTR_ERR(reloc_eb);
3946 if (!extent_buffer_uptodate(reloc_eb)) {
3951 ret = qgroup_trace_subtree_swap(trans, reloc_eb, subvol_eb,
3952 block->last_snapshot, block->trace_leaf);
3955 free_extent_buffer(reloc_eb);
3958 btrfs_err_rl(fs_info,
3959 "failed to account subtree at bytenr %llu: %d",
3960 subvol_eb->start, ret);
3961 fs_info->qgroup_flags |= BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT;