1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/sched/signal.h>
8 #include <linux/pagemap.h>
9 #include <linux/writeback.h>
10 #include <linux/blkdev.h>
11 #include <linux/sort.h>
12 #include <linux/rcupdate.h>
13 #include <linux/kthread.h>
14 #include <linux/slab.h>
15 #include <linux/ratelimit.h>
16 #include <linux/percpu_counter.h>
17 #include <linux/lockdep.h>
18 #include <linux/crc32c.h>
21 #include "print-tree.h"
25 #include "free-space-cache.h"
26 #include "free-space-tree.h"
30 #include "ref-verify.h"
32 #undef SCRAMBLE_DELAYED_REFS
35 * control flags for do_chunk_alloc's force field
36 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
37 * if we really need one.
39 * CHUNK_ALLOC_LIMITED means to only try and allocate one
40 * if we have very few chunks already allocated. This is
41 * used as part of the clustering code to help make sure
42 * we have a good pool of storage to cluster in, without
43 * filling the FS with empty chunks
45 * CHUNK_ALLOC_FORCE means it must try to allocate one
49 CHUNK_ALLOC_NO_FORCE = 0,
50 CHUNK_ALLOC_LIMITED = 1,
51 CHUNK_ALLOC_FORCE = 2,
55 * Declare a helper function to detect underflow of various space info members
57 #define DECLARE_SPACE_INFO_UPDATE(name) \
58 static inline void update_##name(struct btrfs_space_info *sinfo, \
61 if (bytes < 0 && sinfo->name < -bytes) { \
66 sinfo->name += bytes; \
69 DECLARE_SPACE_INFO_UPDATE(bytes_may_use);
70 DECLARE_SPACE_INFO_UPDATE(bytes_pinned);
72 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
73 struct btrfs_delayed_ref_node *node, u64 parent,
74 u64 root_objectid, u64 owner_objectid,
75 u64 owner_offset, int refs_to_drop,
76 struct btrfs_delayed_extent_op *extra_op);
77 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
78 struct extent_buffer *leaf,
79 struct btrfs_extent_item *ei);
80 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
81 u64 parent, u64 root_objectid,
82 u64 flags, u64 owner, u64 offset,
83 struct btrfs_key *ins, int ref_mod);
84 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
85 struct btrfs_delayed_ref_node *node,
86 struct btrfs_delayed_extent_op *extent_op);
87 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
89 static int find_next_key(struct btrfs_path *path, int level,
90 struct btrfs_key *key);
91 static void dump_space_info(struct btrfs_fs_info *fs_info,
92 struct btrfs_space_info *info, u64 bytes,
93 int dump_block_groups);
94 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
96 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
97 struct btrfs_space_info *space_info,
99 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
100 struct btrfs_space_info *space_info,
104 block_group_cache_done(struct btrfs_block_group_cache *cache)
107 return cache->cached == BTRFS_CACHE_FINISHED ||
108 cache->cached == BTRFS_CACHE_ERROR;
111 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
113 return (cache->flags & bits) == bits;
116 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
118 atomic_inc(&cache->count);
121 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
123 if (atomic_dec_and_test(&cache->count)) {
124 WARN_ON(cache->pinned > 0);
125 WARN_ON(cache->reserved > 0);
128 * If not empty, someone is still holding mutex of
129 * full_stripe_lock, which can only be released by caller.
130 * And it will definitely cause use-after-free when caller
131 * tries to release full stripe lock.
133 * No better way to resolve, but only to warn.
135 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
136 kfree(cache->free_space_ctl);
142 * this adds the block group to the fs_info rb tree for the block group
145 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
146 struct btrfs_block_group_cache *block_group)
149 struct rb_node *parent = NULL;
150 struct btrfs_block_group_cache *cache;
152 spin_lock(&info->block_group_cache_lock);
153 p = &info->block_group_cache_tree.rb_node;
157 cache = rb_entry(parent, struct btrfs_block_group_cache,
159 if (block_group->key.objectid < cache->key.objectid) {
161 } else if (block_group->key.objectid > cache->key.objectid) {
164 spin_unlock(&info->block_group_cache_lock);
169 rb_link_node(&block_group->cache_node, parent, p);
170 rb_insert_color(&block_group->cache_node,
171 &info->block_group_cache_tree);
173 if (info->first_logical_byte > block_group->key.objectid)
174 info->first_logical_byte = block_group->key.objectid;
176 spin_unlock(&info->block_group_cache_lock);
182 * This will return the block group at or after bytenr if contains is 0, else
183 * it will return the block group that contains the bytenr
185 static struct btrfs_block_group_cache *
186 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
189 struct btrfs_block_group_cache *cache, *ret = NULL;
193 spin_lock(&info->block_group_cache_lock);
194 n = info->block_group_cache_tree.rb_node;
197 cache = rb_entry(n, struct btrfs_block_group_cache,
199 end = cache->key.objectid + cache->key.offset - 1;
200 start = cache->key.objectid;
202 if (bytenr < start) {
203 if (!contains && (!ret || start < ret->key.objectid))
206 } else if (bytenr > start) {
207 if (contains && bytenr <= end) {
218 btrfs_get_block_group(ret);
219 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
220 info->first_logical_byte = ret->key.objectid;
222 spin_unlock(&info->block_group_cache_lock);
227 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
228 u64 start, u64 num_bytes)
230 u64 end = start + num_bytes - 1;
231 set_extent_bits(&fs_info->freed_extents[0],
232 start, end, EXTENT_UPTODATE);
233 set_extent_bits(&fs_info->freed_extents[1],
234 start, end, EXTENT_UPTODATE);
238 static void free_excluded_extents(struct btrfs_block_group_cache *cache)
240 struct btrfs_fs_info *fs_info = cache->fs_info;
243 start = cache->key.objectid;
244 end = start + cache->key.offset - 1;
246 clear_extent_bits(&fs_info->freed_extents[0],
247 start, end, EXTENT_UPTODATE);
248 clear_extent_bits(&fs_info->freed_extents[1],
249 start, end, EXTENT_UPTODATE);
252 static int exclude_super_stripes(struct btrfs_block_group_cache *cache)
254 struct btrfs_fs_info *fs_info = cache->fs_info;
260 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
261 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
262 cache->bytes_super += stripe_len;
263 ret = add_excluded_extent(fs_info, cache->key.objectid,
269 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
270 bytenr = btrfs_sb_offset(i);
271 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
272 bytenr, &logical, &nr, &stripe_len);
279 if (logical[nr] > cache->key.objectid +
283 if (logical[nr] + stripe_len <= cache->key.objectid)
287 if (start < cache->key.objectid) {
288 start = cache->key.objectid;
289 len = (logical[nr] + stripe_len) - start;
291 len = min_t(u64, stripe_len,
292 cache->key.objectid +
293 cache->key.offset - start);
296 cache->bytes_super += len;
297 ret = add_excluded_extent(fs_info, start, len);
309 static struct btrfs_caching_control *
310 get_caching_control(struct btrfs_block_group_cache *cache)
312 struct btrfs_caching_control *ctl;
314 spin_lock(&cache->lock);
315 if (!cache->caching_ctl) {
316 spin_unlock(&cache->lock);
320 ctl = cache->caching_ctl;
321 refcount_inc(&ctl->count);
322 spin_unlock(&cache->lock);
326 static void put_caching_control(struct btrfs_caching_control *ctl)
328 if (refcount_dec_and_test(&ctl->count))
332 #ifdef CONFIG_BTRFS_DEBUG
333 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
335 struct btrfs_fs_info *fs_info = block_group->fs_info;
336 u64 start = block_group->key.objectid;
337 u64 len = block_group->key.offset;
338 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
339 fs_info->nodesize : fs_info->sectorsize;
340 u64 step = chunk << 1;
342 while (len > chunk) {
343 btrfs_remove_free_space(block_group, start, chunk);
354 * this is only called by cache_block_group, since we could have freed extents
355 * we need to check the pinned_extents for any extents that can't be used yet
356 * since their free space will be released as soon as the transaction commits.
358 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
361 struct btrfs_fs_info *info = block_group->fs_info;
362 u64 extent_start, extent_end, size, total_added = 0;
365 while (start < end) {
366 ret = find_first_extent_bit(info->pinned_extents, start,
367 &extent_start, &extent_end,
368 EXTENT_DIRTY | EXTENT_UPTODATE,
373 if (extent_start <= start) {
374 start = extent_end + 1;
375 } else if (extent_start > start && extent_start < end) {
376 size = extent_start - start;
378 ret = btrfs_add_free_space(block_group, start,
380 BUG_ON(ret); /* -ENOMEM or logic error */
381 start = extent_end + 1;
390 ret = btrfs_add_free_space(block_group, start, size);
391 BUG_ON(ret); /* -ENOMEM or logic error */
397 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
399 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
400 struct btrfs_fs_info *fs_info = block_group->fs_info;
401 struct btrfs_root *extent_root = fs_info->extent_root;
402 struct btrfs_path *path;
403 struct extent_buffer *leaf;
404 struct btrfs_key key;
411 path = btrfs_alloc_path();
415 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
417 #ifdef CONFIG_BTRFS_DEBUG
419 * If we're fragmenting we don't want to make anybody think we can
420 * allocate from this block group until we've had a chance to fragment
423 if (btrfs_should_fragment_free_space(block_group))
427 * We don't want to deadlock with somebody trying to allocate a new
428 * extent for the extent root while also trying to search the extent
429 * root to add free space. So we skip locking and search the commit
430 * root, since its read-only
432 path->skip_locking = 1;
433 path->search_commit_root = 1;
434 path->reada = READA_FORWARD;
438 key.type = BTRFS_EXTENT_ITEM_KEY;
441 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
445 leaf = path->nodes[0];
446 nritems = btrfs_header_nritems(leaf);
449 if (btrfs_fs_closing(fs_info) > 1) {
454 if (path->slots[0] < nritems) {
455 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
457 ret = find_next_key(path, 0, &key);
461 if (need_resched() ||
462 rwsem_is_contended(&fs_info->commit_root_sem)) {
464 caching_ctl->progress = last;
465 btrfs_release_path(path);
466 up_read(&fs_info->commit_root_sem);
467 mutex_unlock(&caching_ctl->mutex);
469 mutex_lock(&caching_ctl->mutex);
470 down_read(&fs_info->commit_root_sem);
474 ret = btrfs_next_leaf(extent_root, path);
479 leaf = path->nodes[0];
480 nritems = btrfs_header_nritems(leaf);
484 if (key.objectid < last) {
487 key.type = BTRFS_EXTENT_ITEM_KEY;
490 caching_ctl->progress = last;
491 btrfs_release_path(path);
495 if (key.objectid < block_group->key.objectid) {
500 if (key.objectid >= block_group->key.objectid +
501 block_group->key.offset)
504 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
505 key.type == BTRFS_METADATA_ITEM_KEY) {
506 total_found += add_new_free_space(block_group, last,
508 if (key.type == BTRFS_METADATA_ITEM_KEY)
509 last = key.objectid +
512 last = key.objectid + key.offset;
514 if (total_found > CACHING_CTL_WAKE_UP) {
517 wake_up(&caching_ctl->wait);
524 total_found += add_new_free_space(block_group, last,
525 block_group->key.objectid +
526 block_group->key.offset);
527 caching_ctl->progress = (u64)-1;
530 btrfs_free_path(path);
534 static noinline void caching_thread(struct btrfs_work *work)
536 struct btrfs_block_group_cache *block_group;
537 struct btrfs_fs_info *fs_info;
538 struct btrfs_caching_control *caching_ctl;
541 caching_ctl = container_of(work, struct btrfs_caching_control, work);
542 block_group = caching_ctl->block_group;
543 fs_info = block_group->fs_info;
545 mutex_lock(&caching_ctl->mutex);
546 down_read(&fs_info->commit_root_sem);
548 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
549 ret = load_free_space_tree(caching_ctl);
551 ret = load_extent_tree_free(caching_ctl);
553 spin_lock(&block_group->lock);
554 block_group->caching_ctl = NULL;
555 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
556 spin_unlock(&block_group->lock);
558 #ifdef CONFIG_BTRFS_DEBUG
559 if (btrfs_should_fragment_free_space(block_group)) {
562 spin_lock(&block_group->space_info->lock);
563 spin_lock(&block_group->lock);
564 bytes_used = block_group->key.offset -
565 btrfs_block_group_used(&block_group->item);
566 block_group->space_info->bytes_used += bytes_used >> 1;
567 spin_unlock(&block_group->lock);
568 spin_unlock(&block_group->space_info->lock);
569 fragment_free_space(block_group);
573 caching_ctl->progress = (u64)-1;
575 up_read(&fs_info->commit_root_sem);
576 free_excluded_extents(block_group);
577 mutex_unlock(&caching_ctl->mutex);
579 wake_up(&caching_ctl->wait);
581 put_caching_control(caching_ctl);
582 btrfs_put_block_group(block_group);
585 static int cache_block_group(struct btrfs_block_group_cache *cache,
589 struct btrfs_fs_info *fs_info = cache->fs_info;
590 struct btrfs_caching_control *caching_ctl;
593 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
597 INIT_LIST_HEAD(&caching_ctl->list);
598 mutex_init(&caching_ctl->mutex);
599 init_waitqueue_head(&caching_ctl->wait);
600 caching_ctl->block_group = cache;
601 caching_ctl->progress = cache->key.objectid;
602 refcount_set(&caching_ctl->count, 1);
603 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
604 caching_thread, NULL, NULL);
606 spin_lock(&cache->lock);
608 * This should be a rare occasion, but this could happen I think in the
609 * case where one thread starts to load the space cache info, and then
610 * some other thread starts a transaction commit which tries to do an
611 * allocation while the other thread is still loading the space cache
612 * info. The previous loop should have kept us from choosing this block
613 * group, but if we've moved to the state where we will wait on caching
614 * block groups we need to first check if we're doing a fast load here,
615 * so we can wait for it to finish, otherwise we could end up allocating
616 * from a block group who's cache gets evicted for one reason or
619 while (cache->cached == BTRFS_CACHE_FAST) {
620 struct btrfs_caching_control *ctl;
622 ctl = cache->caching_ctl;
623 refcount_inc(&ctl->count);
624 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
625 spin_unlock(&cache->lock);
629 finish_wait(&ctl->wait, &wait);
630 put_caching_control(ctl);
631 spin_lock(&cache->lock);
634 if (cache->cached != BTRFS_CACHE_NO) {
635 spin_unlock(&cache->lock);
639 WARN_ON(cache->caching_ctl);
640 cache->caching_ctl = caching_ctl;
641 cache->cached = BTRFS_CACHE_FAST;
642 spin_unlock(&cache->lock);
644 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
645 mutex_lock(&caching_ctl->mutex);
646 ret = load_free_space_cache(fs_info, cache);
648 spin_lock(&cache->lock);
650 cache->caching_ctl = NULL;
651 cache->cached = BTRFS_CACHE_FINISHED;
652 cache->last_byte_to_unpin = (u64)-1;
653 caching_ctl->progress = (u64)-1;
655 if (load_cache_only) {
656 cache->caching_ctl = NULL;
657 cache->cached = BTRFS_CACHE_NO;
659 cache->cached = BTRFS_CACHE_STARTED;
660 cache->has_caching_ctl = 1;
663 spin_unlock(&cache->lock);
664 #ifdef CONFIG_BTRFS_DEBUG
666 btrfs_should_fragment_free_space(cache)) {
669 spin_lock(&cache->space_info->lock);
670 spin_lock(&cache->lock);
671 bytes_used = cache->key.offset -
672 btrfs_block_group_used(&cache->item);
673 cache->space_info->bytes_used += bytes_used >> 1;
674 spin_unlock(&cache->lock);
675 spin_unlock(&cache->space_info->lock);
676 fragment_free_space(cache);
679 mutex_unlock(&caching_ctl->mutex);
681 wake_up(&caching_ctl->wait);
683 put_caching_control(caching_ctl);
684 free_excluded_extents(cache);
689 * We're either using the free space tree or no caching at all.
690 * Set cached to the appropriate value and wakeup any waiters.
692 spin_lock(&cache->lock);
693 if (load_cache_only) {
694 cache->caching_ctl = NULL;
695 cache->cached = BTRFS_CACHE_NO;
697 cache->cached = BTRFS_CACHE_STARTED;
698 cache->has_caching_ctl = 1;
700 spin_unlock(&cache->lock);
701 wake_up(&caching_ctl->wait);
704 if (load_cache_only) {
705 put_caching_control(caching_ctl);
709 down_write(&fs_info->commit_root_sem);
710 refcount_inc(&caching_ctl->count);
711 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
712 up_write(&fs_info->commit_root_sem);
714 btrfs_get_block_group(cache);
716 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
722 * return the block group that starts at or after bytenr
724 static struct btrfs_block_group_cache *
725 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
727 return block_group_cache_tree_search(info, bytenr, 0);
731 * return the block group that contains the given bytenr
733 struct btrfs_block_group_cache *btrfs_lookup_block_group(
734 struct btrfs_fs_info *info,
737 return block_group_cache_tree_search(info, bytenr, 1);
740 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
743 struct list_head *head = &info->space_info;
744 struct btrfs_space_info *found;
746 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
749 list_for_each_entry_rcu(found, head, list) {
750 if (found->flags & flags) {
759 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
760 bool metadata, u64 root_objectid)
762 struct btrfs_space_info *space_info;
766 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
767 flags = BTRFS_BLOCK_GROUP_SYSTEM;
769 flags = BTRFS_BLOCK_GROUP_METADATA;
771 flags = BTRFS_BLOCK_GROUP_DATA;
774 space_info = __find_space_info(fs_info, flags);
776 percpu_counter_add_batch(&space_info->total_bytes_pinned, num_bytes,
777 BTRFS_TOTAL_BYTES_PINNED_BATCH);
781 * after adding space to the filesystem, we need to clear the full flags
782 * on all the space infos.
784 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
786 struct list_head *head = &info->space_info;
787 struct btrfs_space_info *found;
790 list_for_each_entry_rcu(found, head, list)
795 /* simple helper to search for an existing data extent at a given offset */
796 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
799 struct btrfs_key key;
800 struct btrfs_path *path;
802 path = btrfs_alloc_path();
806 key.objectid = start;
808 key.type = BTRFS_EXTENT_ITEM_KEY;
809 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
810 btrfs_free_path(path);
815 * helper function to lookup reference count and flags of a tree block.
817 * the head node for delayed ref is used to store the sum of all the
818 * reference count modifications queued up in the rbtree. the head
819 * node may also store the extent flags to set. This way you can check
820 * to see what the reference count and extent flags would be if all of
821 * the delayed refs are not processed.
823 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
824 struct btrfs_fs_info *fs_info, u64 bytenr,
825 u64 offset, int metadata, u64 *refs, u64 *flags)
827 struct btrfs_delayed_ref_head *head;
828 struct btrfs_delayed_ref_root *delayed_refs;
829 struct btrfs_path *path;
830 struct btrfs_extent_item *ei;
831 struct extent_buffer *leaf;
832 struct btrfs_key key;
839 * If we don't have skinny metadata, don't bother doing anything
842 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
843 offset = fs_info->nodesize;
847 path = btrfs_alloc_path();
852 path->skip_locking = 1;
853 path->search_commit_root = 1;
857 key.objectid = bytenr;
860 key.type = BTRFS_METADATA_ITEM_KEY;
862 key.type = BTRFS_EXTENT_ITEM_KEY;
864 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
868 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
869 if (path->slots[0]) {
871 btrfs_item_key_to_cpu(path->nodes[0], &key,
873 if (key.objectid == bytenr &&
874 key.type == BTRFS_EXTENT_ITEM_KEY &&
875 key.offset == fs_info->nodesize)
881 leaf = path->nodes[0];
882 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
883 if (item_size >= sizeof(*ei)) {
884 ei = btrfs_item_ptr(leaf, path->slots[0],
885 struct btrfs_extent_item);
886 num_refs = btrfs_extent_refs(leaf, ei);
887 extent_flags = btrfs_extent_flags(leaf, ei);
890 btrfs_print_v0_err(fs_info);
892 btrfs_abort_transaction(trans, ret);
894 btrfs_handle_fs_error(fs_info, ret, NULL);
899 BUG_ON(num_refs == 0);
909 delayed_refs = &trans->transaction->delayed_refs;
910 spin_lock(&delayed_refs->lock);
911 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
913 if (!mutex_trylock(&head->mutex)) {
914 refcount_inc(&head->refs);
915 spin_unlock(&delayed_refs->lock);
917 btrfs_release_path(path);
920 * Mutex was contended, block until it's released and try
923 mutex_lock(&head->mutex);
924 mutex_unlock(&head->mutex);
925 btrfs_put_delayed_ref_head(head);
928 spin_lock(&head->lock);
929 if (head->extent_op && head->extent_op->update_flags)
930 extent_flags |= head->extent_op->flags_to_set;
932 BUG_ON(num_refs == 0);
934 num_refs += head->ref_mod;
935 spin_unlock(&head->lock);
936 mutex_unlock(&head->mutex);
938 spin_unlock(&delayed_refs->lock);
940 WARN_ON(num_refs == 0);
944 *flags = extent_flags;
946 btrfs_free_path(path);
951 * Back reference rules. Back refs have three main goals:
953 * 1) differentiate between all holders of references to an extent so that
954 * when a reference is dropped we can make sure it was a valid reference
955 * before freeing the extent.
957 * 2) Provide enough information to quickly find the holders of an extent
958 * if we notice a given block is corrupted or bad.
960 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
961 * maintenance. This is actually the same as #2, but with a slightly
962 * different use case.
964 * There are two kinds of back refs. The implicit back refs is optimized
965 * for pointers in non-shared tree blocks. For a given pointer in a block,
966 * back refs of this kind provide information about the block's owner tree
967 * and the pointer's key. These information allow us to find the block by
968 * b-tree searching. The full back refs is for pointers in tree blocks not
969 * referenced by their owner trees. The location of tree block is recorded
970 * in the back refs. Actually the full back refs is generic, and can be
971 * used in all cases the implicit back refs is used. The major shortcoming
972 * of the full back refs is its overhead. Every time a tree block gets
973 * COWed, we have to update back refs entry for all pointers in it.
975 * For a newly allocated tree block, we use implicit back refs for
976 * pointers in it. This means most tree related operations only involve
977 * implicit back refs. For a tree block created in old transaction, the
978 * only way to drop a reference to it is COW it. So we can detect the
979 * event that tree block loses its owner tree's reference and do the
980 * back refs conversion.
982 * When a tree block is COWed through a tree, there are four cases:
984 * The reference count of the block is one and the tree is the block's
985 * owner tree. Nothing to do in this case.
987 * The reference count of the block is one and the tree is not the
988 * block's owner tree. In this case, full back refs is used for pointers
989 * in the block. Remove these full back refs, add implicit back refs for
990 * every pointers in the new block.
992 * The reference count of the block is greater than one and the tree is
993 * the block's owner tree. In this case, implicit back refs is used for
994 * pointers in the block. Add full back refs for every pointers in the
995 * block, increase lower level extents' reference counts. The original
996 * implicit back refs are entailed to the new block.
998 * The reference count of the block is greater than one and the tree is
999 * not the block's owner tree. Add implicit back refs for every pointer in
1000 * the new block, increase lower level extents' reference count.
1002 * Back Reference Key composing:
1004 * The key objectid corresponds to the first byte in the extent,
1005 * The key type is used to differentiate between types of back refs.
1006 * There are different meanings of the key offset for different types
1009 * File extents can be referenced by:
1011 * - multiple snapshots, subvolumes, or different generations in one subvol
1012 * - different files inside a single subvolume
1013 * - different offsets inside a file (bookend extents in file.c)
1015 * The extent ref structure for the implicit back refs has fields for:
1017 * - Objectid of the subvolume root
1018 * - objectid of the file holding the reference
1019 * - original offset in the file
1020 * - how many bookend extents
1022 * The key offset for the implicit back refs is hash of the first
1025 * The extent ref structure for the full back refs has field for:
1027 * - number of pointers in the tree leaf
1029 * The key offset for the implicit back refs is the first byte of
1032 * When a file extent is allocated, The implicit back refs is used.
1033 * the fields are filled in:
1035 * (root_key.objectid, inode objectid, offset in file, 1)
1037 * When a file extent is removed file truncation, we find the
1038 * corresponding implicit back refs and check the following fields:
1040 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1042 * Btree extents can be referenced by:
1044 * - Different subvolumes
1046 * Both the implicit back refs and the full back refs for tree blocks
1047 * only consist of key. The key offset for the implicit back refs is
1048 * objectid of block's owner tree. The key offset for the full back refs
1049 * is the first byte of parent block.
1051 * When implicit back refs is used, information about the lowest key and
1052 * level of the tree block are required. These information are stored in
1053 * tree block info structure.
1057 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1058 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1059 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1061 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1062 struct btrfs_extent_inline_ref *iref,
1063 enum btrfs_inline_ref_type is_data)
1065 int type = btrfs_extent_inline_ref_type(eb, iref);
1066 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1068 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1069 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1070 type == BTRFS_SHARED_DATA_REF_KEY ||
1071 type == BTRFS_EXTENT_DATA_REF_KEY) {
1072 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1073 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1075 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1076 ASSERT(eb->fs_info);
1078 * Every shared one has parent tree
1079 * block, which must be aligned to
1083 IS_ALIGNED(offset, eb->fs_info->nodesize))
1086 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1087 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1089 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1090 ASSERT(eb->fs_info);
1092 * Every shared one has parent tree
1093 * block, which must be aligned to
1097 IS_ALIGNED(offset, eb->fs_info->nodesize))
1101 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1106 btrfs_print_leaf((struct extent_buffer *)eb);
1107 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1111 return BTRFS_REF_TYPE_INVALID;
1114 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1116 u32 high_crc = ~(u32)0;
1117 u32 low_crc = ~(u32)0;
1120 lenum = cpu_to_le64(root_objectid);
1121 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1122 lenum = cpu_to_le64(owner);
1123 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1124 lenum = cpu_to_le64(offset);
1125 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1127 return ((u64)high_crc << 31) ^ (u64)low_crc;
1130 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1131 struct btrfs_extent_data_ref *ref)
1133 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1134 btrfs_extent_data_ref_objectid(leaf, ref),
1135 btrfs_extent_data_ref_offset(leaf, ref));
1138 static int match_extent_data_ref(struct extent_buffer *leaf,
1139 struct btrfs_extent_data_ref *ref,
1140 u64 root_objectid, u64 owner, u64 offset)
1142 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1143 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1144 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1149 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1150 struct btrfs_path *path,
1151 u64 bytenr, u64 parent,
1153 u64 owner, u64 offset)
1155 struct btrfs_root *root = trans->fs_info->extent_root;
1156 struct btrfs_key key;
1157 struct btrfs_extent_data_ref *ref;
1158 struct extent_buffer *leaf;
1164 key.objectid = bytenr;
1166 key.type = BTRFS_SHARED_DATA_REF_KEY;
1167 key.offset = parent;
1169 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1170 key.offset = hash_extent_data_ref(root_objectid,
1175 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1187 leaf = path->nodes[0];
1188 nritems = btrfs_header_nritems(leaf);
1190 if (path->slots[0] >= nritems) {
1191 ret = btrfs_next_leaf(root, path);
1197 leaf = path->nodes[0];
1198 nritems = btrfs_header_nritems(leaf);
1202 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1203 if (key.objectid != bytenr ||
1204 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1207 ref = btrfs_item_ptr(leaf, path->slots[0],
1208 struct btrfs_extent_data_ref);
1210 if (match_extent_data_ref(leaf, ref, root_objectid,
1213 btrfs_release_path(path);
1225 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1226 struct btrfs_path *path,
1227 u64 bytenr, u64 parent,
1228 u64 root_objectid, u64 owner,
1229 u64 offset, int refs_to_add)
1231 struct btrfs_root *root = trans->fs_info->extent_root;
1232 struct btrfs_key key;
1233 struct extent_buffer *leaf;
1238 key.objectid = bytenr;
1240 key.type = BTRFS_SHARED_DATA_REF_KEY;
1241 key.offset = parent;
1242 size = sizeof(struct btrfs_shared_data_ref);
1244 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1245 key.offset = hash_extent_data_ref(root_objectid,
1247 size = sizeof(struct btrfs_extent_data_ref);
1250 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1251 if (ret && ret != -EEXIST)
1254 leaf = path->nodes[0];
1256 struct btrfs_shared_data_ref *ref;
1257 ref = btrfs_item_ptr(leaf, path->slots[0],
1258 struct btrfs_shared_data_ref);
1260 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1262 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1263 num_refs += refs_to_add;
1264 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1267 struct btrfs_extent_data_ref *ref;
1268 while (ret == -EEXIST) {
1269 ref = btrfs_item_ptr(leaf, path->slots[0],
1270 struct btrfs_extent_data_ref);
1271 if (match_extent_data_ref(leaf, ref, root_objectid,
1274 btrfs_release_path(path);
1276 ret = btrfs_insert_empty_item(trans, root, path, &key,
1278 if (ret && ret != -EEXIST)
1281 leaf = path->nodes[0];
1283 ref = btrfs_item_ptr(leaf, path->slots[0],
1284 struct btrfs_extent_data_ref);
1286 btrfs_set_extent_data_ref_root(leaf, ref,
1288 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1289 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1290 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1292 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1293 num_refs += refs_to_add;
1294 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1297 btrfs_mark_buffer_dirty(leaf);
1300 btrfs_release_path(path);
1304 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1305 struct btrfs_path *path,
1306 int refs_to_drop, int *last_ref)
1308 struct btrfs_key key;
1309 struct btrfs_extent_data_ref *ref1 = NULL;
1310 struct btrfs_shared_data_ref *ref2 = NULL;
1311 struct extent_buffer *leaf;
1315 leaf = path->nodes[0];
1316 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1318 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1319 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_extent_data_ref);
1321 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1322 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1323 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1324 struct btrfs_shared_data_ref);
1325 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1326 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
1327 btrfs_print_v0_err(trans->fs_info);
1328 btrfs_abort_transaction(trans, -EINVAL);
1334 BUG_ON(num_refs < refs_to_drop);
1335 num_refs -= refs_to_drop;
1337 if (num_refs == 0) {
1338 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1341 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1342 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1343 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1344 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1345 btrfs_mark_buffer_dirty(leaf);
1350 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1351 struct btrfs_extent_inline_ref *iref)
1353 struct btrfs_key key;
1354 struct extent_buffer *leaf;
1355 struct btrfs_extent_data_ref *ref1;
1356 struct btrfs_shared_data_ref *ref2;
1360 leaf = path->nodes[0];
1361 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1363 BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
1366 * If type is invalid, we should have bailed out earlier than
1369 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1370 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1371 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1372 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1373 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1375 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1376 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1378 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1379 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1380 struct btrfs_extent_data_ref);
1381 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1382 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1383 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1384 struct btrfs_shared_data_ref);
1385 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1392 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1393 struct btrfs_path *path,
1394 u64 bytenr, u64 parent,
1397 struct btrfs_root *root = trans->fs_info->extent_root;
1398 struct btrfs_key key;
1401 key.objectid = bytenr;
1403 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1404 key.offset = parent;
1406 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1407 key.offset = root_objectid;
1410 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1416 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1417 struct btrfs_path *path,
1418 u64 bytenr, u64 parent,
1421 struct btrfs_key key;
1424 key.objectid = bytenr;
1426 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1427 key.offset = parent;
1429 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1430 key.offset = root_objectid;
1433 ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root,
1435 btrfs_release_path(path);
1439 static inline int extent_ref_type(u64 parent, u64 owner)
1442 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1444 type = BTRFS_SHARED_BLOCK_REF_KEY;
1446 type = BTRFS_TREE_BLOCK_REF_KEY;
1449 type = BTRFS_SHARED_DATA_REF_KEY;
1451 type = BTRFS_EXTENT_DATA_REF_KEY;
1456 static int find_next_key(struct btrfs_path *path, int level,
1457 struct btrfs_key *key)
1460 for (; level < BTRFS_MAX_LEVEL; level++) {
1461 if (!path->nodes[level])
1463 if (path->slots[level] + 1 >=
1464 btrfs_header_nritems(path->nodes[level]))
1467 btrfs_item_key_to_cpu(path->nodes[level], key,
1468 path->slots[level] + 1);
1470 btrfs_node_key_to_cpu(path->nodes[level], key,
1471 path->slots[level] + 1);
1478 * look for inline back ref. if back ref is found, *ref_ret is set
1479 * to the address of inline back ref, and 0 is returned.
1481 * if back ref isn't found, *ref_ret is set to the address where it
1482 * should be inserted, and -ENOENT is returned.
1484 * if insert is true and there are too many inline back refs, the path
1485 * points to the extent item, and -EAGAIN is returned.
1487 * NOTE: inline back refs are ordered in the same way that back ref
1488 * items in the tree are ordered.
1490 static noinline_for_stack
1491 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1492 struct btrfs_path *path,
1493 struct btrfs_extent_inline_ref **ref_ret,
1494 u64 bytenr, u64 num_bytes,
1495 u64 parent, u64 root_objectid,
1496 u64 owner, u64 offset, int insert)
1498 struct btrfs_fs_info *fs_info = trans->fs_info;
1499 struct btrfs_root *root = fs_info->extent_root;
1500 struct btrfs_key key;
1501 struct extent_buffer *leaf;
1502 struct btrfs_extent_item *ei;
1503 struct btrfs_extent_inline_ref *iref;
1513 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1516 key.objectid = bytenr;
1517 key.type = BTRFS_EXTENT_ITEM_KEY;
1518 key.offset = num_bytes;
1520 want = extent_ref_type(parent, owner);
1522 extra_size = btrfs_extent_inline_ref_size(want);
1523 path->keep_locks = 1;
1528 * Owner is our level, so we can just add one to get the level for the
1529 * block we are interested in.
1531 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1532 key.type = BTRFS_METADATA_ITEM_KEY;
1537 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1544 * We may be a newly converted file system which still has the old fat
1545 * extent entries for metadata, so try and see if we have one of those.
1547 if (ret > 0 && skinny_metadata) {
1548 skinny_metadata = false;
1549 if (path->slots[0]) {
1551 btrfs_item_key_to_cpu(path->nodes[0], &key,
1553 if (key.objectid == bytenr &&
1554 key.type == BTRFS_EXTENT_ITEM_KEY &&
1555 key.offset == num_bytes)
1559 key.objectid = bytenr;
1560 key.type = BTRFS_EXTENT_ITEM_KEY;
1561 key.offset = num_bytes;
1562 btrfs_release_path(path);
1567 if (ret && !insert) {
1570 } else if (WARN_ON(ret)) {
1575 leaf = path->nodes[0];
1576 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1577 if (unlikely(item_size < sizeof(*ei))) {
1579 btrfs_print_v0_err(fs_info);
1580 btrfs_abort_transaction(trans, err);
1584 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1585 flags = btrfs_extent_flags(leaf, ei);
1587 ptr = (unsigned long)(ei + 1);
1588 end = (unsigned long)ei + item_size;
1590 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1591 ptr += sizeof(struct btrfs_tree_block_info);
1595 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1596 needed = BTRFS_REF_TYPE_DATA;
1598 needed = BTRFS_REF_TYPE_BLOCK;
1606 iref = (struct btrfs_extent_inline_ref *)ptr;
1607 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1608 if (type == BTRFS_REF_TYPE_INVALID) {
1616 ptr += btrfs_extent_inline_ref_size(type);
1620 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1621 struct btrfs_extent_data_ref *dref;
1622 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1623 if (match_extent_data_ref(leaf, dref, root_objectid,
1628 if (hash_extent_data_ref_item(leaf, dref) <
1629 hash_extent_data_ref(root_objectid, owner, offset))
1633 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1635 if (parent == ref_offset) {
1639 if (ref_offset < parent)
1642 if (root_objectid == ref_offset) {
1646 if (ref_offset < root_objectid)
1650 ptr += btrfs_extent_inline_ref_size(type);
1652 if (err == -ENOENT && insert) {
1653 if (item_size + extra_size >=
1654 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1659 * To add new inline back ref, we have to make sure
1660 * there is no corresponding back ref item.
1661 * For simplicity, we just do not add new inline back
1662 * ref if there is any kind of item for this block
1664 if (find_next_key(path, 0, &key) == 0 &&
1665 key.objectid == bytenr &&
1666 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1671 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1674 path->keep_locks = 0;
1675 btrfs_unlock_up_safe(path, 1);
1681 * helper to add new inline back ref
1683 static noinline_for_stack
1684 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1685 struct btrfs_path *path,
1686 struct btrfs_extent_inline_ref *iref,
1687 u64 parent, u64 root_objectid,
1688 u64 owner, u64 offset, int refs_to_add,
1689 struct btrfs_delayed_extent_op *extent_op)
1691 struct extent_buffer *leaf;
1692 struct btrfs_extent_item *ei;
1695 unsigned long item_offset;
1700 leaf = path->nodes[0];
1701 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1702 item_offset = (unsigned long)iref - (unsigned long)ei;
1704 type = extent_ref_type(parent, owner);
1705 size = btrfs_extent_inline_ref_size(type);
1707 btrfs_extend_item(fs_info, path, size);
1709 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1710 refs = btrfs_extent_refs(leaf, ei);
1711 refs += refs_to_add;
1712 btrfs_set_extent_refs(leaf, ei, refs);
1714 __run_delayed_extent_op(extent_op, leaf, ei);
1716 ptr = (unsigned long)ei + item_offset;
1717 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1718 if (ptr < end - size)
1719 memmove_extent_buffer(leaf, ptr + size, ptr,
1722 iref = (struct btrfs_extent_inline_ref *)ptr;
1723 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1724 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1725 struct btrfs_extent_data_ref *dref;
1726 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1727 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1728 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1729 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1730 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1731 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1732 struct btrfs_shared_data_ref *sref;
1733 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1734 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1735 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1736 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1737 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1739 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1741 btrfs_mark_buffer_dirty(leaf);
1744 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1745 struct btrfs_path *path,
1746 struct btrfs_extent_inline_ref **ref_ret,
1747 u64 bytenr, u64 num_bytes, u64 parent,
1748 u64 root_objectid, u64 owner, u64 offset)
1752 ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr,
1753 num_bytes, parent, root_objectid,
1758 btrfs_release_path(path);
1761 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1762 ret = lookup_tree_block_ref(trans, path, bytenr, parent,
1765 ret = lookup_extent_data_ref(trans, path, bytenr, parent,
1766 root_objectid, owner, offset);
1772 * helper to update/remove inline back ref
1774 static noinline_for_stack
1775 void update_inline_extent_backref(struct btrfs_path *path,
1776 struct btrfs_extent_inline_ref *iref,
1778 struct btrfs_delayed_extent_op *extent_op,
1781 struct extent_buffer *leaf = path->nodes[0];
1782 struct btrfs_fs_info *fs_info = leaf->fs_info;
1783 struct btrfs_extent_item *ei;
1784 struct btrfs_extent_data_ref *dref = NULL;
1785 struct btrfs_shared_data_ref *sref = NULL;
1793 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1794 refs = btrfs_extent_refs(leaf, ei);
1795 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1796 refs += refs_to_mod;
1797 btrfs_set_extent_refs(leaf, ei, refs);
1799 __run_delayed_extent_op(extent_op, leaf, ei);
1802 * If type is invalid, we should have bailed out after
1803 * lookup_inline_extent_backref().
1805 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1806 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1808 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1809 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1810 refs = btrfs_extent_data_ref_count(leaf, dref);
1811 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1812 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1813 refs = btrfs_shared_data_ref_count(leaf, sref);
1816 BUG_ON(refs_to_mod != -1);
1819 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1820 refs += refs_to_mod;
1823 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1824 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1826 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1829 size = btrfs_extent_inline_ref_size(type);
1830 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1831 ptr = (unsigned long)iref;
1832 end = (unsigned long)ei + item_size;
1833 if (ptr + size < end)
1834 memmove_extent_buffer(leaf, ptr, ptr + size,
1837 btrfs_truncate_item(fs_info, path, item_size, 1);
1839 btrfs_mark_buffer_dirty(leaf);
1842 static noinline_for_stack
1843 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1844 struct btrfs_path *path,
1845 u64 bytenr, u64 num_bytes, u64 parent,
1846 u64 root_objectid, u64 owner,
1847 u64 offset, int refs_to_add,
1848 struct btrfs_delayed_extent_op *extent_op)
1850 struct btrfs_extent_inline_ref *iref;
1853 ret = lookup_inline_extent_backref(trans, path, &iref, bytenr,
1854 num_bytes, parent, root_objectid,
1857 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1858 update_inline_extent_backref(path, iref, refs_to_add,
1860 } else if (ret == -ENOENT) {
1861 setup_inline_extent_backref(trans->fs_info, path, iref, parent,
1862 root_objectid, owner, offset,
1863 refs_to_add, extent_op);
1869 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1870 struct btrfs_path *path,
1871 u64 bytenr, u64 parent, u64 root_objectid,
1872 u64 owner, u64 offset, int refs_to_add)
1875 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1876 BUG_ON(refs_to_add != 1);
1877 ret = insert_tree_block_ref(trans, path, bytenr, parent,
1880 ret = insert_extent_data_ref(trans, path, bytenr, parent,
1881 root_objectid, owner, offset,
1887 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1888 struct btrfs_path *path,
1889 struct btrfs_extent_inline_ref *iref,
1890 int refs_to_drop, int is_data, int *last_ref)
1894 BUG_ON(!is_data && refs_to_drop != 1);
1896 update_inline_extent_backref(path, iref, -refs_to_drop, NULL,
1898 } else if (is_data) {
1899 ret = remove_extent_data_ref(trans, path, refs_to_drop,
1903 ret = btrfs_del_item(trans, trans->fs_info->extent_root, path);
1908 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1909 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1910 u64 *discarded_bytes)
1913 u64 bytes_left, end;
1914 u64 aligned_start = ALIGN(start, 1 << 9);
1916 if (WARN_ON(start != aligned_start)) {
1917 len -= aligned_start - start;
1918 len = round_down(len, 1 << 9);
1919 start = aligned_start;
1922 *discarded_bytes = 0;
1930 /* Skip any superblocks on this device. */
1931 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1932 u64 sb_start = btrfs_sb_offset(j);
1933 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1934 u64 size = sb_start - start;
1936 if (!in_range(sb_start, start, bytes_left) &&
1937 !in_range(sb_end, start, bytes_left) &&
1938 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1942 * Superblock spans beginning of range. Adjust start and
1945 if (sb_start <= start) {
1946 start += sb_end - start;
1951 bytes_left = end - start;
1956 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1959 *discarded_bytes += size;
1960 else if (ret != -EOPNOTSUPP)
1969 bytes_left = end - start;
1973 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
1976 *discarded_bytes += bytes_left;
1981 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
1982 u64 num_bytes, u64 *actual_bytes)
1985 u64 discarded_bytes = 0;
1986 struct btrfs_bio *bbio = NULL;
1990 * Avoid races with device replace and make sure our bbio has devices
1991 * associated to its stripes that don't go away while we are discarding.
1993 btrfs_bio_counter_inc_blocked(fs_info);
1994 /* Tell the block device(s) that the sectors can be discarded */
1995 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
1997 /* Error condition is -ENOMEM */
1999 struct btrfs_bio_stripe *stripe = bbio->stripes;
2003 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2005 struct request_queue *req_q;
2007 if (!stripe->dev->bdev) {
2008 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2011 req_q = bdev_get_queue(stripe->dev->bdev);
2012 if (!blk_queue_discard(req_q))
2015 ret = btrfs_issue_discard(stripe->dev->bdev,
2020 discarded_bytes += bytes;
2021 else if (ret != -EOPNOTSUPP)
2022 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2025 * Just in case we get back EOPNOTSUPP for some reason,
2026 * just ignore the return value so we don't screw up
2027 * people calling discard_extent.
2031 btrfs_put_bbio(bbio);
2033 btrfs_bio_counter_dec(fs_info);
2036 *actual_bytes = discarded_bytes;
2039 if (ret == -EOPNOTSUPP)
2044 /* Can return -ENOMEM */
2045 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2046 struct btrfs_root *root,
2047 u64 bytenr, u64 num_bytes, u64 parent,
2048 u64 root_objectid, u64 owner, u64 offset)
2050 struct btrfs_fs_info *fs_info = root->fs_info;
2051 int old_ref_mod, new_ref_mod;
2054 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2055 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2057 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2058 owner, offset, BTRFS_ADD_DELAYED_REF);
2060 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2061 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
2063 root_objectid, (int)owner,
2064 BTRFS_ADD_DELAYED_REF, NULL,
2065 &old_ref_mod, &new_ref_mod);
2067 ret = btrfs_add_delayed_data_ref(trans, bytenr,
2069 root_objectid, owner, offset,
2070 0, BTRFS_ADD_DELAYED_REF,
2071 &old_ref_mod, &new_ref_mod);
2074 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) {
2075 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
2077 add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid);
2084 * __btrfs_inc_extent_ref - insert backreference for a given extent
2086 * @trans: Handle of transaction
2088 * @node: The delayed ref node used to get the bytenr/length for
2089 * extent whose references are incremented.
2091 * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/
2092 * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical
2093 * bytenr of the parent block. Since new extents are always
2094 * created with indirect references, this will only be the case
2095 * when relocating a shared extent. In that case, root_objectid
2096 * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must
2099 * @root_objectid: The id of the root where this modification has originated,
2100 * this can be either one of the well-known metadata trees or
2101 * the subvolume id which references this extent.
2103 * @owner: For data extents it is the inode number of the owning file.
2104 * For metadata extents this parameter holds the level in the
2105 * tree of the extent.
2107 * @offset: For metadata extents the offset is ignored and is currently
2108 * always passed as 0. For data extents it is the fileoffset
2109 * this extent belongs to.
2111 * @refs_to_add Number of references to add
2113 * @extent_op Pointer to a structure, holding information necessary when
2114 * updating a tree block's flags
2117 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2118 struct btrfs_delayed_ref_node *node,
2119 u64 parent, u64 root_objectid,
2120 u64 owner, u64 offset, int refs_to_add,
2121 struct btrfs_delayed_extent_op *extent_op)
2123 struct btrfs_path *path;
2124 struct extent_buffer *leaf;
2125 struct btrfs_extent_item *item;
2126 struct btrfs_key key;
2127 u64 bytenr = node->bytenr;
2128 u64 num_bytes = node->num_bytes;
2132 path = btrfs_alloc_path();
2136 path->reada = READA_FORWARD;
2137 path->leave_spinning = 1;
2138 /* this will setup the path even if it fails to insert the back ref */
2139 ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes,
2140 parent, root_objectid, owner,
2141 offset, refs_to_add, extent_op);
2142 if ((ret < 0 && ret != -EAGAIN) || !ret)
2146 * Ok we had -EAGAIN which means we didn't have space to insert and
2147 * inline extent ref, so just update the reference count and add a
2150 leaf = path->nodes[0];
2151 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2152 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2153 refs = btrfs_extent_refs(leaf, item);
2154 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2156 __run_delayed_extent_op(extent_op, leaf, item);
2158 btrfs_mark_buffer_dirty(leaf);
2159 btrfs_release_path(path);
2161 path->reada = READA_FORWARD;
2162 path->leave_spinning = 1;
2163 /* now insert the actual backref */
2164 ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid,
2165 owner, offset, refs_to_add);
2167 btrfs_abort_transaction(trans, ret);
2169 btrfs_free_path(path);
2173 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2174 struct btrfs_delayed_ref_node *node,
2175 struct btrfs_delayed_extent_op *extent_op,
2176 int insert_reserved)
2179 struct btrfs_delayed_data_ref *ref;
2180 struct btrfs_key ins;
2185 ins.objectid = node->bytenr;
2186 ins.offset = node->num_bytes;
2187 ins.type = BTRFS_EXTENT_ITEM_KEY;
2189 ref = btrfs_delayed_node_to_data_ref(node);
2190 trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action);
2192 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2193 parent = ref->parent;
2194 ref_root = ref->root;
2196 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2198 flags |= extent_op->flags_to_set;
2199 ret = alloc_reserved_file_extent(trans, parent, ref_root,
2200 flags, ref->objectid,
2203 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2204 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2205 ref->objectid, ref->offset,
2206 node->ref_mod, extent_op);
2207 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2208 ret = __btrfs_free_extent(trans, node, parent,
2209 ref_root, ref->objectid,
2210 ref->offset, node->ref_mod,
2218 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2219 struct extent_buffer *leaf,
2220 struct btrfs_extent_item *ei)
2222 u64 flags = btrfs_extent_flags(leaf, ei);
2223 if (extent_op->update_flags) {
2224 flags |= extent_op->flags_to_set;
2225 btrfs_set_extent_flags(leaf, ei, flags);
2228 if (extent_op->update_key) {
2229 struct btrfs_tree_block_info *bi;
2230 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2231 bi = (struct btrfs_tree_block_info *)(ei + 1);
2232 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2236 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2237 struct btrfs_delayed_ref_head *head,
2238 struct btrfs_delayed_extent_op *extent_op)
2240 struct btrfs_fs_info *fs_info = trans->fs_info;
2241 struct btrfs_key key;
2242 struct btrfs_path *path;
2243 struct btrfs_extent_item *ei;
2244 struct extent_buffer *leaf;
2248 int metadata = !extent_op->is_data;
2253 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2256 path = btrfs_alloc_path();
2260 key.objectid = head->bytenr;
2263 key.type = BTRFS_METADATA_ITEM_KEY;
2264 key.offset = extent_op->level;
2266 key.type = BTRFS_EXTENT_ITEM_KEY;
2267 key.offset = head->num_bytes;
2271 path->reada = READA_FORWARD;
2272 path->leave_spinning = 1;
2273 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2280 if (path->slots[0] > 0) {
2282 btrfs_item_key_to_cpu(path->nodes[0], &key,
2284 if (key.objectid == head->bytenr &&
2285 key.type == BTRFS_EXTENT_ITEM_KEY &&
2286 key.offset == head->num_bytes)
2290 btrfs_release_path(path);
2293 key.objectid = head->bytenr;
2294 key.offset = head->num_bytes;
2295 key.type = BTRFS_EXTENT_ITEM_KEY;
2304 leaf = path->nodes[0];
2305 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2307 if (unlikely(item_size < sizeof(*ei))) {
2309 btrfs_print_v0_err(fs_info);
2310 btrfs_abort_transaction(trans, err);
2314 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2315 __run_delayed_extent_op(extent_op, leaf, ei);
2317 btrfs_mark_buffer_dirty(leaf);
2319 btrfs_free_path(path);
2323 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2324 struct btrfs_delayed_ref_node *node,
2325 struct btrfs_delayed_extent_op *extent_op,
2326 int insert_reserved)
2329 struct btrfs_delayed_tree_ref *ref;
2333 ref = btrfs_delayed_node_to_tree_ref(node);
2334 trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action);
2336 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2337 parent = ref->parent;
2338 ref_root = ref->root;
2340 if (node->ref_mod != 1) {
2341 btrfs_err(trans->fs_info,
2342 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2343 node->bytenr, node->ref_mod, node->action, ref_root,
2347 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2348 BUG_ON(!extent_op || !extent_op->update_flags);
2349 ret = alloc_reserved_tree_block(trans, node, extent_op);
2350 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2351 ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root,
2352 ref->level, 0, 1, extent_op);
2353 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2354 ret = __btrfs_free_extent(trans, node, parent, ref_root,
2355 ref->level, 0, 1, extent_op);
2362 /* helper function to actually process a single delayed ref entry */
2363 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2364 struct btrfs_delayed_ref_node *node,
2365 struct btrfs_delayed_extent_op *extent_op,
2366 int insert_reserved)
2370 if (trans->aborted) {
2371 if (insert_reserved)
2372 btrfs_pin_extent(trans->fs_info, node->bytenr,
2373 node->num_bytes, 1);
2377 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2378 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2379 ret = run_delayed_tree_ref(trans, node, extent_op,
2381 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2382 node->type == BTRFS_SHARED_DATA_REF_KEY)
2383 ret = run_delayed_data_ref(trans, node, extent_op,
2387 if (ret && insert_reserved)
2388 btrfs_pin_extent(trans->fs_info, node->bytenr,
2389 node->num_bytes, 1);
2393 static inline struct btrfs_delayed_ref_node *
2394 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2396 struct btrfs_delayed_ref_node *ref;
2398 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root))
2402 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2403 * This is to prevent a ref count from going down to zero, which deletes
2404 * the extent item from the extent tree, when there still are references
2405 * to add, which would fail because they would not find the extent item.
2407 if (!list_empty(&head->ref_add_list))
2408 return list_first_entry(&head->ref_add_list,
2409 struct btrfs_delayed_ref_node, add_list);
2411 ref = rb_entry(rb_first_cached(&head->ref_tree),
2412 struct btrfs_delayed_ref_node, ref_node);
2413 ASSERT(list_empty(&ref->add_list));
2417 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2418 struct btrfs_delayed_ref_head *head)
2420 spin_lock(&delayed_refs->lock);
2421 head->processing = 0;
2422 delayed_refs->num_heads_ready++;
2423 spin_unlock(&delayed_refs->lock);
2424 btrfs_delayed_ref_unlock(head);
2427 static struct btrfs_delayed_extent_op *cleanup_extent_op(
2428 struct btrfs_delayed_ref_head *head)
2430 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2435 if (head->must_insert_reserved) {
2436 head->extent_op = NULL;
2437 btrfs_free_delayed_extent_op(extent_op);
2443 static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans,
2444 struct btrfs_delayed_ref_head *head)
2446 struct btrfs_delayed_extent_op *extent_op;
2449 extent_op = cleanup_extent_op(head);
2452 head->extent_op = NULL;
2453 spin_unlock(&head->lock);
2454 ret = run_delayed_extent_op(trans, head, extent_op);
2455 btrfs_free_delayed_extent_op(extent_op);
2456 return ret ? ret : 1;
2459 static void cleanup_ref_head_accounting(struct btrfs_trans_handle *trans,
2460 struct btrfs_delayed_ref_head *head)
2462 struct btrfs_fs_info *fs_info = trans->fs_info;
2463 struct btrfs_delayed_ref_root *delayed_refs =
2464 &trans->transaction->delayed_refs;
2465 int nr_items = 1; /* Dropping this ref head update. */
2467 if (head->total_ref_mod < 0) {
2468 struct btrfs_space_info *space_info;
2472 flags = BTRFS_BLOCK_GROUP_DATA;
2473 else if (head->is_system)
2474 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2476 flags = BTRFS_BLOCK_GROUP_METADATA;
2477 space_info = __find_space_info(fs_info, flags);
2479 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2481 BTRFS_TOTAL_BYTES_PINNED_BATCH);
2484 * We had csum deletions accounted for in our delayed refs rsv,
2485 * we need to drop the csum leaves for this update from our
2488 if (head->is_data) {
2489 spin_lock(&delayed_refs->lock);
2490 delayed_refs->pending_csums -= head->num_bytes;
2491 spin_unlock(&delayed_refs->lock);
2492 nr_items += btrfs_csum_bytes_to_leaves(fs_info,
2497 /* Also free its reserved qgroup space */
2498 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2499 head->qgroup_reserved);
2500 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
2503 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2504 struct btrfs_delayed_ref_head *head)
2507 struct btrfs_fs_info *fs_info = trans->fs_info;
2508 struct btrfs_delayed_ref_root *delayed_refs;
2511 delayed_refs = &trans->transaction->delayed_refs;
2513 ret = run_and_cleanup_extent_op(trans, head);
2515 unselect_delayed_ref_head(delayed_refs, head);
2516 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2523 * Need to drop our head ref lock and re-acquire the delayed ref lock
2524 * and then re-check to make sure nobody got added.
2526 spin_unlock(&head->lock);
2527 spin_lock(&delayed_refs->lock);
2528 spin_lock(&head->lock);
2529 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
2530 spin_unlock(&head->lock);
2531 spin_unlock(&delayed_refs->lock);
2534 btrfs_delete_ref_head(delayed_refs, head);
2535 spin_unlock(&head->lock);
2536 spin_unlock(&delayed_refs->lock);
2538 if (head->must_insert_reserved) {
2539 btrfs_pin_extent(fs_info, head->bytenr,
2540 head->num_bytes, 1);
2541 if (head->is_data) {
2542 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2547 cleanup_ref_head_accounting(trans, head);
2549 trace_run_delayed_ref_head(fs_info, head, 0);
2550 btrfs_delayed_ref_unlock(head);
2551 btrfs_put_delayed_ref_head(head);
2555 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
2556 struct btrfs_trans_handle *trans)
2558 struct btrfs_delayed_ref_root *delayed_refs =
2559 &trans->transaction->delayed_refs;
2560 struct btrfs_delayed_ref_head *head = NULL;
2563 spin_lock(&delayed_refs->lock);
2564 head = btrfs_select_ref_head(delayed_refs);
2566 spin_unlock(&delayed_refs->lock);
2571 * Grab the lock that says we are going to process all the refs for
2574 ret = btrfs_delayed_ref_lock(delayed_refs, head);
2575 spin_unlock(&delayed_refs->lock);
2578 * We may have dropped the spin lock to get the head mutex lock, and
2579 * that might have given someone else time to free the head. If that's
2580 * true, it has been removed from our list and we can move on.
2583 head = ERR_PTR(-EAGAIN);
2588 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2589 struct btrfs_delayed_ref_head *locked_ref,
2590 unsigned long *run_refs)
2592 struct btrfs_fs_info *fs_info = trans->fs_info;
2593 struct btrfs_delayed_ref_root *delayed_refs;
2594 struct btrfs_delayed_extent_op *extent_op;
2595 struct btrfs_delayed_ref_node *ref;
2596 int must_insert_reserved = 0;
2599 delayed_refs = &trans->transaction->delayed_refs;
2601 lockdep_assert_held(&locked_ref->mutex);
2602 lockdep_assert_held(&locked_ref->lock);
2604 while ((ref = select_delayed_ref(locked_ref))) {
2606 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2607 spin_unlock(&locked_ref->lock);
2608 unselect_delayed_ref_head(delayed_refs, locked_ref);
2614 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2615 RB_CLEAR_NODE(&ref->ref_node);
2616 if (!list_empty(&ref->add_list))
2617 list_del(&ref->add_list);
2619 * When we play the delayed ref, also correct the ref_mod on
2622 switch (ref->action) {
2623 case BTRFS_ADD_DELAYED_REF:
2624 case BTRFS_ADD_DELAYED_EXTENT:
2625 locked_ref->ref_mod -= ref->ref_mod;
2627 case BTRFS_DROP_DELAYED_REF:
2628 locked_ref->ref_mod += ref->ref_mod;
2633 atomic_dec(&delayed_refs->num_entries);
2636 * Record the must_insert_reserved flag before we drop the
2639 must_insert_reserved = locked_ref->must_insert_reserved;
2640 locked_ref->must_insert_reserved = 0;
2642 extent_op = locked_ref->extent_op;
2643 locked_ref->extent_op = NULL;
2644 spin_unlock(&locked_ref->lock);
2646 ret = run_one_delayed_ref(trans, ref, extent_op,
2647 must_insert_reserved);
2649 btrfs_free_delayed_extent_op(extent_op);
2651 unselect_delayed_ref_head(delayed_refs, locked_ref);
2652 btrfs_put_delayed_ref(ref);
2653 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2658 btrfs_put_delayed_ref(ref);
2661 spin_lock(&locked_ref->lock);
2662 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2669 * Returns 0 on success or if called with an already aborted transaction.
2670 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2672 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2675 struct btrfs_fs_info *fs_info = trans->fs_info;
2676 struct btrfs_delayed_ref_root *delayed_refs;
2677 struct btrfs_delayed_ref_head *locked_ref = NULL;
2678 ktime_t start = ktime_get();
2680 unsigned long count = 0;
2681 unsigned long actual_count = 0;
2683 delayed_refs = &trans->transaction->delayed_refs;
2686 locked_ref = btrfs_obtain_ref_head(trans);
2687 if (IS_ERR_OR_NULL(locked_ref)) {
2688 if (PTR_ERR(locked_ref) == -EAGAIN) {
2697 * We need to try and merge add/drops of the same ref since we
2698 * can run into issues with relocate dropping the implicit ref
2699 * and then it being added back again before the drop can
2700 * finish. If we merged anything we need to re-loop so we can
2702 * Or we can get node references of the same type that weren't
2703 * merged when created due to bumps in the tree mod seq, and
2704 * we need to merge them to prevent adding an inline extent
2705 * backref before dropping it (triggering a BUG_ON at
2706 * insert_inline_extent_backref()).
2708 spin_lock(&locked_ref->lock);
2709 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2711 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2713 if (ret < 0 && ret != -EAGAIN) {
2715 * Error, btrfs_run_delayed_refs_for_head already
2716 * unlocked everything so just bail out
2721 * Success, perform the usual cleanup of a processed
2724 ret = cleanup_ref_head(trans, locked_ref);
2726 /* We dropped our lock, we need to loop. */
2735 * Either success case or btrfs_run_delayed_refs_for_head
2736 * returned -EAGAIN, meaning we need to select another head
2741 } while ((nr != -1 && count < nr) || locked_ref);
2744 * We don't want to include ref heads since we can have empty ref heads
2745 * and those will drastically skew our runtime down since we just do
2746 * accounting, no actual extent tree updates.
2748 if (actual_count > 0) {
2749 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2753 * We weigh the current average higher than our current runtime
2754 * to avoid large swings in the average.
2756 spin_lock(&delayed_refs->lock);
2757 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2758 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2759 spin_unlock(&delayed_refs->lock);
2764 #ifdef SCRAMBLE_DELAYED_REFS
2766 * Normally delayed refs get processed in ascending bytenr order. This
2767 * correlates in most cases to the order added. To expose dependencies on this
2768 * order, we start to process the tree in the middle instead of the beginning
2770 static u64 find_middle(struct rb_root *root)
2772 struct rb_node *n = root->rb_node;
2773 struct btrfs_delayed_ref_node *entry;
2776 u64 first = 0, last = 0;
2780 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2781 first = entry->bytenr;
2785 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2786 last = entry->bytenr;
2791 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2792 WARN_ON(!entry->in_tree);
2794 middle = entry->bytenr;
2807 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2811 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2812 sizeof(struct btrfs_extent_inline_ref));
2813 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2814 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2817 * We don't ever fill up leaves all the way so multiply by 2 just to be
2818 * closer to what we're really going to want to use.
2820 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2824 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2825 * would require to store the csums for that many bytes.
2827 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2830 u64 num_csums_per_leaf;
2833 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2834 num_csums_per_leaf = div64_u64(csum_size,
2835 (u64)btrfs_super_csum_size(fs_info->super_copy));
2836 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2837 num_csums += num_csums_per_leaf - 1;
2838 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2842 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans)
2844 struct btrfs_fs_info *fs_info = trans->fs_info;
2845 struct btrfs_block_rsv *global_rsv;
2846 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2847 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2848 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2849 u64 num_bytes, num_dirty_bgs_bytes;
2852 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2853 num_heads = heads_to_leaves(fs_info, num_heads);
2855 num_bytes += (num_heads - 1) * fs_info->nodesize;
2857 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2859 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2861 global_rsv = &fs_info->global_block_rsv;
2864 * If we can't allocate any more chunks lets make sure we have _lots_ of
2865 * wiggle room since running delayed refs can create more delayed refs.
2867 if (global_rsv->space_info->full) {
2868 num_dirty_bgs_bytes <<= 1;
2872 spin_lock(&global_rsv->lock);
2873 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2875 spin_unlock(&global_rsv->lock);
2879 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans)
2882 atomic_read(&trans->transaction->delayed_refs.num_entries);
2887 avg_runtime = trans->fs_info->avg_delayed_ref_runtime;
2888 val = num_entries * avg_runtime;
2889 if (val >= NSEC_PER_SEC)
2891 if (val >= NSEC_PER_SEC / 2)
2894 return btrfs_check_space_for_delayed_refs(trans);
2897 struct async_delayed_refs {
2898 struct btrfs_root *root;
2903 struct completion wait;
2904 struct btrfs_work work;
2907 static inline struct async_delayed_refs *
2908 to_async_delayed_refs(struct btrfs_work *work)
2910 return container_of(work, struct async_delayed_refs, work);
2913 static void delayed_ref_async_start(struct btrfs_work *work)
2915 struct async_delayed_refs *async = to_async_delayed_refs(work);
2916 struct btrfs_trans_handle *trans;
2917 struct btrfs_fs_info *fs_info = async->root->fs_info;
2920 /* if the commit is already started, we don't need to wait here */
2921 if (btrfs_transaction_blocked(fs_info))
2924 trans = btrfs_join_transaction(async->root);
2925 if (IS_ERR(trans)) {
2926 async->error = PTR_ERR(trans);
2931 * trans->sync means that when we call end_transaction, we won't
2932 * wait on delayed refs
2936 /* Don't bother flushing if we got into a different transaction */
2937 if (trans->transid > async->transid)
2940 ret = btrfs_run_delayed_refs(trans, async->count);
2944 ret = btrfs_end_transaction(trans);
2945 if (ret && !async->error)
2949 complete(&async->wait);
2954 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2955 unsigned long count, u64 transid, int wait)
2957 struct async_delayed_refs *async;
2960 async = kmalloc(sizeof(*async), GFP_NOFS);
2964 async->root = fs_info->tree_root;
2965 async->count = count;
2967 async->transid = transid;
2972 init_completion(&async->wait);
2974 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2975 delayed_ref_async_start, NULL, NULL);
2977 btrfs_queue_work(fs_info->extent_workers, &async->work);
2980 wait_for_completion(&async->wait);
2989 * this starts processing the delayed reference count updates and
2990 * extent insertions we have queued up so far. count can be
2991 * 0, which means to process everything in the tree at the start
2992 * of the run (but not newly added entries), or it can be some target
2993 * number you'd like to process.
2995 * Returns 0 on success or if called with an aborted transaction
2996 * Returns <0 on error and aborts the transaction
2998 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2999 unsigned long count)
3001 struct btrfs_fs_info *fs_info = trans->fs_info;
3002 struct rb_node *node;
3003 struct btrfs_delayed_ref_root *delayed_refs;
3004 struct btrfs_delayed_ref_head *head;
3006 int run_all = count == (unsigned long)-1;
3008 /* We'll clean this up in btrfs_cleanup_transaction */
3012 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3015 delayed_refs = &trans->transaction->delayed_refs;
3017 count = atomic_read(&delayed_refs->num_entries) * 2;
3020 #ifdef SCRAMBLE_DELAYED_REFS
3021 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3023 ret = __btrfs_run_delayed_refs(trans, count);
3025 btrfs_abort_transaction(trans, ret);
3030 if (!list_empty(&trans->new_bgs))
3031 btrfs_create_pending_block_groups(trans);
3033 spin_lock(&delayed_refs->lock);
3034 node = rb_first_cached(&delayed_refs->href_root);
3036 spin_unlock(&delayed_refs->lock);
3039 head = rb_entry(node, struct btrfs_delayed_ref_head,
3041 refcount_inc(&head->refs);
3042 spin_unlock(&delayed_refs->lock);
3044 /* Mutex was contended, block until it's released and retry. */
3045 mutex_lock(&head->mutex);
3046 mutex_unlock(&head->mutex);
3048 btrfs_put_delayed_ref_head(head);
3056 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3057 struct btrfs_fs_info *fs_info,
3058 u64 bytenr, u64 num_bytes, u64 flags,
3059 int level, int is_data)
3061 struct btrfs_delayed_extent_op *extent_op;
3064 extent_op = btrfs_alloc_delayed_extent_op();
3068 extent_op->flags_to_set = flags;
3069 extent_op->update_flags = true;
3070 extent_op->update_key = false;
3071 extent_op->is_data = is_data ? true : false;
3072 extent_op->level = level;
3074 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3075 num_bytes, extent_op);
3077 btrfs_free_delayed_extent_op(extent_op);
3081 static noinline int check_delayed_ref(struct btrfs_root *root,
3082 struct btrfs_path *path,
3083 u64 objectid, u64 offset, u64 bytenr)
3085 struct btrfs_delayed_ref_head *head;
3086 struct btrfs_delayed_ref_node *ref;
3087 struct btrfs_delayed_data_ref *data_ref;
3088 struct btrfs_delayed_ref_root *delayed_refs;
3089 struct btrfs_transaction *cur_trans;
3090 struct rb_node *node;
3093 spin_lock(&root->fs_info->trans_lock);
3094 cur_trans = root->fs_info->running_transaction;
3096 refcount_inc(&cur_trans->use_count);
3097 spin_unlock(&root->fs_info->trans_lock);
3101 delayed_refs = &cur_trans->delayed_refs;
3102 spin_lock(&delayed_refs->lock);
3103 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3105 spin_unlock(&delayed_refs->lock);
3106 btrfs_put_transaction(cur_trans);
3110 if (!mutex_trylock(&head->mutex)) {
3111 refcount_inc(&head->refs);
3112 spin_unlock(&delayed_refs->lock);
3114 btrfs_release_path(path);
3117 * Mutex was contended, block until it's released and let
3120 mutex_lock(&head->mutex);
3121 mutex_unlock(&head->mutex);
3122 btrfs_put_delayed_ref_head(head);
3123 btrfs_put_transaction(cur_trans);
3126 spin_unlock(&delayed_refs->lock);
3128 spin_lock(&head->lock);
3130 * XXX: We should replace this with a proper search function in the
3133 for (node = rb_first_cached(&head->ref_tree); node;
3134 node = rb_next(node)) {
3135 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3136 /* If it's a shared ref we know a cross reference exists */
3137 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3142 data_ref = btrfs_delayed_node_to_data_ref(ref);
3145 * If our ref doesn't match the one we're currently looking at
3146 * then we have a cross reference.
3148 if (data_ref->root != root->root_key.objectid ||
3149 data_ref->objectid != objectid ||
3150 data_ref->offset != offset) {
3155 spin_unlock(&head->lock);
3156 mutex_unlock(&head->mutex);
3157 btrfs_put_transaction(cur_trans);
3161 static noinline int check_committed_ref(struct btrfs_root *root,
3162 struct btrfs_path *path,
3163 u64 objectid, u64 offset, u64 bytenr)
3165 struct btrfs_fs_info *fs_info = root->fs_info;
3166 struct btrfs_root *extent_root = fs_info->extent_root;
3167 struct extent_buffer *leaf;
3168 struct btrfs_extent_data_ref *ref;
3169 struct btrfs_extent_inline_ref *iref;
3170 struct btrfs_extent_item *ei;
3171 struct btrfs_key key;
3176 key.objectid = bytenr;
3177 key.offset = (u64)-1;
3178 key.type = BTRFS_EXTENT_ITEM_KEY;
3180 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3183 BUG_ON(ret == 0); /* Corruption */
3186 if (path->slots[0] == 0)
3190 leaf = path->nodes[0];
3191 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3193 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3197 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3198 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3200 if (item_size != sizeof(*ei) +
3201 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3204 if (btrfs_extent_generation(leaf, ei) <=
3205 btrfs_root_last_snapshot(&root->root_item))
3208 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3210 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3211 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3214 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3215 if (btrfs_extent_refs(leaf, ei) !=
3216 btrfs_extent_data_ref_count(leaf, ref) ||
3217 btrfs_extent_data_ref_root(leaf, ref) !=
3218 root->root_key.objectid ||
3219 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3220 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3228 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3231 struct btrfs_path *path;
3234 path = btrfs_alloc_path();
3239 ret = check_committed_ref(root, path, objectid,
3241 if (ret && ret != -ENOENT)
3244 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
3245 } while (ret == -EAGAIN);
3248 btrfs_free_path(path);
3249 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3254 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3255 struct btrfs_root *root,
3256 struct extent_buffer *buf,
3257 int full_backref, int inc)
3259 struct btrfs_fs_info *fs_info = root->fs_info;
3265 struct btrfs_key key;
3266 struct btrfs_file_extent_item *fi;
3270 int (*process_func)(struct btrfs_trans_handle *,
3271 struct btrfs_root *,
3272 u64, u64, u64, u64, u64, u64);
3275 if (btrfs_is_testing(fs_info))
3278 ref_root = btrfs_header_owner(buf);
3279 nritems = btrfs_header_nritems(buf);
3280 level = btrfs_header_level(buf);
3282 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3286 process_func = btrfs_inc_extent_ref;
3288 process_func = btrfs_free_extent;
3291 parent = buf->start;
3295 for (i = 0; i < nritems; i++) {
3297 btrfs_item_key_to_cpu(buf, &key, i);
3298 if (key.type != BTRFS_EXTENT_DATA_KEY)
3300 fi = btrfs_item_ptr(buf, i,
3301 struct btrfs_file_extent_item);
3302 if (btrfs_file_extent_type(buf, fi) ==
3303 BTRFS_FILE_EXTENT_INLINE)
3305 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3309 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3310 key.offset -= btrfs_file_extent_offset(buf, fi);
3311 ret = process_func(trans, root, bytenr, num_bytes,
3312 parent, ref_root, key.objectid,
3317 bytenr = btrfs_node_blockptr(buf, i);
3318 num_bytes = fs_info->nodesize;
3319 ret = process_func(trans, root, bytenr, num_bytes,
3320 parent, ref_root, level - 1, 0);
3330 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3331 struct extent_buffer *buf, int full_backref)
3333 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3336 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3337 struct extent_buffer *buf, int full_backref)
3339 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3342 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3343 struct btrfs_fs_info *fs_info,
3344 struct btrfs_path *path,
3345 struct btrfs_block_group_cache *cache)
3348 struct btrfs_root *extent_root = fs_info->extent_root;
3350 struct extent_buffer *leaf;
3352 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3359 leaf = path->nodes[0];
3360 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3361 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3362 btrfs_mark_buffer_dirty(leaf);
3364 btrfs_release_path(path);
3369 static struct btrfs_block_group_cache *
3370 next_block_group(struct btrfs_fs_info *fs_info,
3371 struct btrfs_block_group_cache *cache)
3373 struct rb_node *node;
3375 spin_lock(&fs_info->block_group_cache_lock);
3377 /* If our block group was removed, we need a full search. */
3378 if (RB_EMPTY_NODE(&cache->cache_node)) {
3379 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3381 spin_unlock(&fs_info->block_group_cache_lock);
3382 btrfs_put_block_group(cache);
3383 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3385 node = rb_next(&cache->cache_node);
3386 btrfs_put_block_group(cache);
3388 cache = rb_entry(node, struct btrfs_block_group_cache,
3390 btrfs_get_block_group(cache);
3393 spin_unlock(&fs_info->block_group_cache_lock);
3397 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3398 struct btrfs_trans_handle *trans,
3399 struct btrfs_path *path)
3401 struct btrfs_fs_info *fs_info = block_group->fs_info;
3402 struct btrfs_root *root = fs_info->tree_root;
3403 struct inode *inode = NULL;
3404 struct extent_changeset *data_reserved = NULL;
3406 int dcs = BTRFS_DC_ERROR;
3412 * If this block group is smaller than 100 megs don't bother caching the
3415 if (block_group->key.offset < (100 * SZ_1M)) {
3416 spin_lock(&block_group->lock);
3417 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3418 spin_unlock(&block_group->lock);
3425 inode = lookup_free_space_inode(fs_info, block_group, path);
3426 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3427 ret = PTR_ERR(inode);
3428 btrfs_release_path(path);
3432 if (IS_ERR(inode)) {
3436 if (block_group->ro)
3439 ret = create_free_space_inode(fs_info, trans, block_group,
3447 * We want to set the generation to 0, that way if anything goes wrong
3448 * from here on out we know not to trust this cache when we load up next
3451 BTRFS_I(inode)->generation = 0;
3452 ret = btrfs_update_inode(trans, root, inode);
3455 * So theoretically we could recover from this, simply set the
3456 * super cache generation to 0 so we know to invalidate the
3457 * cache, but then we'd have to keep track of the block groups
3458 * that fail this way so we know we _have_ to reset this cache
3459 * before the next commit or risk reading stale cache. So to
3460 * limit our exposure to horrible edge cases lets just abort the
3461 * transaction, this only happens in really bad situations
3464 btrfs_abort_transaction(trans, ret);
3469 /* We've already setup this transaction, go ahead and exit */
3470 if (block_group->cache_generation == trans->transid &&
3471 i_size_read(inode)) {
3472 dcs = BTRFS_DC_SETUP;
3476 if (i_size_read(inode) > 0) {
3477 ret = btrfs_check_trunc_cache_free_space(fs_info,
3478 &fs_info->global_block_rsv);
3482 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3487 spin_lock(&block_group->lock);
3488 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3489 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3491 * don't bother trying to write stuff out _if_
3492 * a) we're not cached,
3493 * b) we're with nospace_cache mount option,
3494 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3496 dcs = BTRFS_DC_WRITTEN;
3497 spin_unlock(&block_group->lock);
3500 spin_unlock(&block_group->lock);
3503 * We hit an ENOSPC when setting up the cache in this transaction, just
3504 * skip doing the setup, we've already cleared the cache so we're safe.
3506 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3512 * Try to preallocate enough space based on how big the block group is.
3513 * Keep in mind this has to include any pinned space which could end up
3514 * taking up quite a bit since it's not folded into the other space
3517 num_pages = div_u64(block_group->key.offset, SZ_256M);
3522 num_pages *= PAGE_SIZE;
3524 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3528 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3529 num_pages, num_pages,
3532 * Our cache requires contiguous chunks so that we don't modify a bunch
3533 * of metadata or split extents when writing the cache out, which means
3534 * we can enospc if we are heavily fragmented in addition to just normal
3535 * out of space conditions. So if we hit this just skip setting up any
3536 * other block groups for this transaction, maybe we'll unpin enough
3537 * space the next time around.
3540 dcs = BTRFS_DC_SETUP;
3541 else if (ret == -ENOSPC)
3542 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3547 btrfs_release_path(path);
3549 spin_lock(&block_group->lock);
3550 if (!ret && dcs == BTRFS_DC_SETUP)
3551 block_group->cache_generation = trans->transid;
3552 block_group->disk_cache_state = dcs;
3553 spin_unlock(&block_group->lock);
3555 extent_changeset_free(data_reserved);
3559 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3560 struct btrfs_fs_info *fs_info)
3562 struct btrfs_block_group_cache *cache, *tmp;
3563 struct btrfs_transaction *cur_trans = trans->transaction;
3564 struct btrfs_path *path;
3566 if (list_empty(&cur_trans->dirty_bgs) ||
3567 !btrfs_test_opt(fs_info, SPACE_CACHE))
3570 path = btrfs_alloc_path();
3574 /* Could add new block groups, use _safe just in case */
3575 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3577 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3578 cache_save_setup(cache, trans, path);
3581 btrfs_free_path(path);
3586 * transaction commit does final block group cache writeback during a
3587 * critical section where nothing is allowed to change the FS. This is
3588 * required in order for the cache to actually match the block group,
3589 * but can introduce a lot of latency into the commit.
3591 * So, btrfs_start_dirty_block_groups is here to kick off block group
3592 * cache IO. There's a chance we'll have to redo some of it if the
3593 * block group changes again during the commit, but it greatly reduces
3594 * the commit latency by getting rid of the easy block groups while
3595 * we're still allowing others to join the commit.
3597 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3599 struct btrfs_fs_info *fs_info = trans->fs_info;
3600 struct btrfs_block_group_cache *cache;
3601 struct btrfs_transaction *cur_trans = trans->transaction;
3604 struct btrfs_path *path = NULL;
3606 struct list_head *io = &cur_trans->io_bgs;
3607 int num_started = 0;
3610 spin_lock(&cur_trans->dirty_bgs_lock);
3611 if (list_empty(&cur_trans->dirty_bgs)) {
3612 spin_unlock(&cur_trans->dirty_bgs_lock);
3615 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3616 spin_unlock(&cur_trans->dirty_bgs_lock);
3620 * make sure all the block groups on our dirty list actually
3623 btrfs_create_pending_block_groups(trans);
3626 path = btrfs_alloc_path();
3632 * cache_write_mutex is here only to save us from balance or automatic
3633 * removal of empty block groups deleting this block group while we are
3634 * writing out the cache
3636 mutex_lock(&trans->transaction->cache_write_mutex);
3637 while (!list_empty(&dirty)) {
3638 bool drop_reserve = true;
3640 cache = list_first_entry(&dirty,
3641 struct btrfs_block_group_cache,
3644 * this can happen if something re-dirties a block
3645 * group that is already under IO. Just wait for it to
3646 * finish and then do it all again
3648 if (!list_empty(&cache->io_list)) {
3649 list_del_init(&cache->io_list);
3650 btrfs_wait_cache_io(trans, cache, path);
3651 btrfs_put_block_group(cache);
3656 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3657 * if it should update the cache_state. Don't delete
3658 * until after we wait.
3660 * Since we're not running in the commit critical section
3661 * we need the dirty_bgs_lock to protect from update_block_group
3663 spin_lock(&cur_trans->dirty_bgs_lock);
3664 list_del_init(&cache->dirty_list);
3665 spin_unlock(&cur_trans->dirty_bgs_lock);
3669 cache_save_setup(cache, trans, path);
3671 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3672 cache->io_ctl.inode = NULL;
3673 ret = btrfs_write_out_cache(fs_info, trans,
3675 if (ret == 0 && cache->io_ctl.inode) {
3680 * The cache_write_mutex is protecting the
3681 * io_list, also refer to the definition of
3682 * btrfs_transaction::io_bgs for more details
3684 list_add_tail(&cache->io_list, io);
3687 * if we failed to write the cache, the
3688 * generation will be bad and life goes on
3694 ret = write_one_cache_group(trans, fs_info,
3697 * Our block group might still be attached to the list
3698 * of new block groups in the transaction handle of some
3699 * other task (struct btrfs_trans_handle->new_bgs). This
3700 * means its block group item isn't yet in the extent
3701 * tree. If this happens ignore the error, as we will
3702 * try again later in the critical section of the
3703 * transaction commit.
3705 if (ret == -ENOENT) {
3707 spin_lock(&cur_trans->dirty_bgs_lock);
3708 if (list_empty(&cache->dirty_list)) {
3709 list_add_tail(&cache->dirty_list,
3710 &cur_trans->dirty_bgs);
3711 btrfs_get_block_group(cache);
3712 drop_reserve = false;
3714 spin_unlock(&cur_trans->dirty_bgs_lock);
3716 btrfs_abort_transaction(trans, ret);
3720 /* if its not on the io list, we need to put the block group */
3722 btrfs_put_block_group(cache);
3724 btrfs_delayed_refs_rsv_release(fs_info, 1);
3730 * Avoid blocking other tasks for too long. It might even save
3731 * us from writing caches for block groups that are going to be
3734 mutex_unlock(&trans->transaction->cache_write_mutex);
3735 mutex_lock(&trans->transaction->cache_write_mutex);
3737 mutex_unlock(&trans->transaction->cache_write_mutex);
3740 * go through delayed refs for all the stuff we've just kicked off
3741 * and then loop back (just once)
3743 ret = btrfs_run_delayed_refs(trans, 0);
3744 if (!ret && loops == 0) {
3746 spin_lock(&cur_trans->dirty_bgs_lock);
3747 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3749 * dirty_bgs_lock protects us from concurrent block group
3750 * deletes too (not just cache_write_mutex).
3752 if (!list_empty(&dirty)) {
3753 spin_unlock(&cur_trans->dirty_bgs_lock);
3756 spin_unlock(&cur_trans->dirty_bgs_lock);
3757 } else if (ret < 0) {
3758 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3761 btrfs_free_path(path);
3765 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3766 struct btrfs_fs_info *fs_info)
3768 struct btrfs_block_group_cache *cache;
3769 struct btrfs_transaction *cur_trans = trans->transaction;
3772 struct btrfs_path *path;
3773 struct list_head *io = &cur_trans->io_bgs;
3774 int num_started = 0;
3776 path = btrfs_alloc_path();
3781 * Even though we are in the critical section of the transaction commit,
3782 * we can still have concurrent tasks adding elements to this
3783 * transaction's list of dirty block groups. These tasks correspond to
3784 * endio free space workers started when writeback finishes for a
3785 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3786 * allocate new block groups as a result of COWing nodes of the root
3787 * tree when updating the free space inode. The writeback for the space
3788 * caches is triggered by an earlier call to
3789 * btrfs_start_dirty_block_groups() and iterations of the following
3791 * Also we want to do the cache_save_setup first and then run the
3792 * delayed refs to make sure we have the best chance at doing this all
3795 spin_lock(&cur_trans->dirty_bgs_lock);
3796 while (!list_empty(&cur_trans->dirty_bgs)) {
3797 cache = list_first_entry(&cur_trans->dirty_bgs,
3798 struct btrfs_block_group_cache,
3802 * this can happen if cache_save_setup re-dirties a block
3803 * group that is already under IO. Just wait for it to
3804 * finish and then do it all again
3806 if (!list_empty(&cache->io_list)) {
3807 spin_unlock(&cur_trans->dirty_bgs_lock);
3808 list_del_init(&cache->io_list);
3809 btrfs_wait_cache_io(trans, cache, path);
3810 btrfs_put_block_group(cache);
3811 spin_lock(&cur_trans->dirty_bgs_lock);
3815 * don't remove from the dirty list until after we've waited
3818 list_del_init(&cache->dirty_list);
3819 spin_unlock(&cur_trans->dirty_bgs_lock);
3822 cache_save_setup(cache, trans, path);
3825 ret = btrfs_run_delayed_refs(trans,
3826 (unsigned long) -1);
3828 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3829 cache->io_ctl.inode = NULL;
3830 ret = btrfs_write_out_cache(fs_info, trans,
3832 if (ret == 0 && cache->io_ctl.inode) {
3835 list_add_tail(&cache->io_list, io);
3838 * if we failed to write the cache, the
3839 * generation will be bad and life goes on
3845 ret = write_one_cache_group(trans, fs_info,
3848 * One of the free space endio workers might have
3849 * created a new block group while updating a free space
3850 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3851 * and hasn't released its transaction handle yet, in
3852 * which case the new block group is still attached to
3853 * its transaction handle and its creation has not
3854 * finished yet (no block group item in the extent tree
3855 * yet, etc). If this is the case, wait for all free
3856 * space endio workers to finish and retry. This is a
3857 * a very rare case so no need for a more efficient and
3860 if (ret == -ENOENT) {
3861 wait_event(cur_trans->writer_wait,
3862 atomic_read(&cur_trans->num_writers) == 1);
3863 ret = write_one_cache_group(trans, fs_info,
3867 btrfs_abort_transaction(trans, ret);
3870 /* if its not on the io list, we need to put the block group */
3872 btrfs_put_block_group(cache);
3873 btrfs_delayed_refs_rsv_release(fs_info, 1);
3874 spin_lock(&cur_trans->dirty_bgs_lock);
3876 spin_unlock(&cur_trans->dirty_bgs_lock);
3879 * Refer to the definition of io_bgs member for details why it's safe
3880 * to use it without any locking
3882 while (!list_empty(io)) {
3883 cache = list_first_entry(io, struct btrfs_block_group_cache,
3885 list_del_init(&cache->io_list);
3886 btrfs_wait_cache_io(trans, cache, path);
3887 btrfs_put_block_group(cache);
3890 btrfs_free_path(path);
3894 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3896 struct btrfs_block_group_cache *block_group;
3899 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3900 if (!block_group || block_group->ro)
3903 btrfs_put_block_group(block_group);
3907 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3909 struct btrfs_block_group_cache *bg;
3912 bg = btrfs_lookup_block_group(fs_info, bytenr);
3916 spin_lock(&bg->lock);
3920 atomic_inc(&bg->nocow_writers);
3921 spin_unlock(&bg->lock);
3923 /* no put on block group, done by btrfs_dec_nocow_writers */
3925 btrfs_put_block_group(bg);
3931 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3933 struct btrfs_block_group_cache *bg;
3935 bg = btrfs_lookup_block_group(fs_info, bytenr);
3937 if (atomic_dec_and_test(&bg->nocow_writers))
3938 wake_up_var(&bg->nocow_writers);
3940 * Once for our lookup and once for the lookup done by a previous call
3941 * to btrfs_inc_nocow_writers()
3943 btrfs_put_block_group(bg);
3944 btrfs_put_block_group(bg);
3947 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3949 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3952 static const char *alloc_name(u64 flags)
3955 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3957 case BTRFS_BLOCK_GROUP_METADATA:
3959 case BTRFS_BLOCK_GROUP_DATA:
3961 case BTRFS_BLOCK_GROUP_SYSTEM:
3965 return "invalid-combination";
3969 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3972 struct btrfs_space_info *space_info;
3976 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3980 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3987 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3988 INIT_LIST_HEAD(&space_info->block_groups[i]);
3989 init_rwsem(&space_info->groups_sem);
3990 spin_lock_init(&space_info->lock);
3991 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3992 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3993 init_waitqueue_head(&space_info->wait);
3994 INIT_LIST_HEAD(&space_info->ro_bgs);
3995 INIT_LIST_HEAD(&space_info->tickets);
3996 INIT_LIST_HEAD(&space_info->priority_tickets);
3998 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3999 info->space_info_kobj, "%s",
4000 alloc_name(space_info->flags));
4002 percpu_counter_destroy(&space_info->total_bytes_pinned);
4007 list_add_rcu(&space_info->list, &info->space_info);
4008 if (flags & BTRFS_BLOCK_GROUP_DATA)
4009 info->data_sinfo = space_info;
4014 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4015 u64 total_bytes, u64 bytes_used,
4017 struct btrfs_space_info **space_info)
4019 struct btrfs_space_info *found;
4022 factor = btrfs_bg_type_to_factor(flags);
4024 found = __find_space_info(info, flags);
4026 spin_lock(&found->lock);
4027 found->total_bytes += total_bytes;
4028 found->disk_total += total_bytes * factor;
4029 found->bytes_used += bytes_used;
4030 found->disk_used += bytes_used * factor;
4031 found->bytes_readonly += bytes_readonly;
4032 if (total_bytes > 0)
4034 space_info_add_new_bytes(info, found, total_bytes -
4035 bytes_used - bytes_readonly);
4036 spin_unlock(&found->lock);
4037 *space_info = found;
4040 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4042 u64 extra_flags = chunk_to_extended(flags) &
4043 BTRFS_EXTENDED_PROFILE_MASK;
4045 write_seqlock(&fs_info->profiles_lock);
4046 if (flags & BTRFS_BLOCK_GROUP_DATA)
4047 fs_info->avail_data_alloc_bits |= extra_flags;
4048 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4049 fs_info->avail_metadata_alloc_bits |= extra_flags;
4050 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4051 fs_info->avail_system_alloc_bits |= extra_flags;
4052 write_sequnlock(&fs_info->profiles_lock);
4056 * returns target flags in extended format or 0 if restripe for this
4057 * chunk_type is not in progress
4059 * should be called with balance_lock held
4061 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4063 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4069 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4070 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4071 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4072 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4073 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4074 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4075 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4076 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4077 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4084 * @flags: available profiles in extended format (see ctree.h)
4086 * Returns reduced profile in chunk format. If profile changing is in
4087 * progress (either running or paused) picks the target profile (if it's
4088 * already available), otherwise falls back to plain reducing.
4090 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4092 u64 num_devices = fs_info->fs_devices->rw_devices;
4098 * see if restripe for this chunk_type is in progress, if so
4099 * try to reduce to the target profile
4101 spin_lock(&fs_info->balance_lock);
4102 target = get_restripe_target(fs_info, flags);
4104 /* pick target profile only if it's already available */
4105 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4106 spin_unlock(&fs_info->balance_lock);
4107 return extended_to_chunk(target);
4110 spin_unlock(&fs_info->balance_lock);
4112 /* First, mask out the RAID levels which aren't possible */
4113 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4114 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4115 allowed |= btrfs_raid_array[raid_type].bg_flag;
4119 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4120 allowed = BTRFS_BLOCK_GROUP_RAID6;
4121 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4122 allowed = BTRFS_BLOCK_GROUP_RAID5;
4123 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4124 allowed = BTRFS_BLOCK_GROUP_RAID10;
4125 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4126 allowed = BTRFS_BLOCK_GROUP_RAID1;
4127 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4128 allowed = BTRFS_BLOCK_GROUP_RAID0;
4130 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4132 return extended_to_chunk(flags | allowed);
4135 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4142 seq = read_seqbegin(&fs_info->profiles_lock);
4144 if (flags & BTRFS_BLOCK_GROUP_DATA)
4145 flags |= fs_info->avail_data_alloc_bits;
4146 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4147 flags |= fs_info->avail_system_alloc_bits;
4148 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4149 flags |= fs_info->avail_metadata_alloc_bits;
4150 } while (read_seqretry(&fs_info->profiles_lock, seq));
4152 return btrfs_reduce_alloc_profile(fs_info, flags);
4155 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4157 struct btrfs_fs_info *fs_info = root->fs_info;
4162 flags = BTRFS_BLOCK_GROUP_DATA;
4163 else if (root == fs_info->chunk_root)
4164 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4166 flags = BTRFS_BLOCK_GROUP_METADATA;
4168 ret = get_alloc_profile(fs_info, flags);
4172 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4174 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4177 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4179 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4182 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4184 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4187 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4188 bool may_use_included)
4191 return s_info->bytes_used + s_info->bytes_reserved +
4192 s_info->bytes_pinned + s_info->bytes_readonly +
4193 (may_use_included ? s_info->bytes_may_use : 0);
4196 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4198 struct btrfs_root *root = inode->root;
4199 struct btrfs_fs_info *fs_info = root->fs_info;
4200 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4203 int need_commit = 2;
4204 int have_pinned_space;
4206 /* make sure bytes are sectorsize aligned */
4207 bytes = ALIGN(bytes, fs_info->sectorsize);
4209 if (btrfs_is_free_space_inode(inode)) {
4211 ASSERT(current->journal_info);
4215 /* make sure we have enough space to handle the data first */
4216 spin_lock(&data_sinfo->lock);
4217 used = btrfs_space_info_used(data_sinfo, true);
4219 if (used + bytes > data_sinfo->total_bytes) {
4220 struct btrfs_trans_handle *trans;
4223 * if we don't have enough free bytes in this space then we need
4224 * to alloc a new chunk.
4226 if (!data_sinfo->full) {
4229 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4230 spin_unlock(&data_sinfo->lock);
4232 alloc_target = btrfs_data_alloc_profile(fs_info);
4234 * It is ugly that we don't call nolock join
4235 * transaction for the free space inode case here.
4236 * But it is safe because we only do the data space
4237 * reservation for the free space cache in the
4238 * transaction context, the common join transaction
4239 * just increase the counter of the current transaction
4240 * handler, doesn't try to acquire the trans_lock of
4243 trans = btrfs_join_transaction(root);
4245 return PTR_ERR(trans);
4247 ret = do_chunk_alloc(trans, alloc_target,
4248 CHUNK_ALLOC_NO_FORCE);
4249 btrfs_end_transaction(trans);
4254 have_pinned_space = 1;
4263 * If we don't have enough pinned space to deal with this
4264 * allocation, and no removed chunk in current transaction,
4265 * don't bother committing the transaction.
4267 have_pinned_space = __percpu_counter_compare(
4268 &data_sinfo->total_bytes_pinned,
4269 used + bytes - data_sinfo->total_bytes,
4270 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4271 spin_unlock(&data_sinfo->lock);
4273 /* commit the current transaction and try again */
4278 if (need_commit > 0) {
4279 btrfs_start_delalloc_roots(fs_info, -1);
4280 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4284 trans = btrfs_join_transaction(root);
4286 return PTR_ERR(trans);
4287 if (have_pinned_space >= 0 ||
4288 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4289 &trans->transaction->flags) ||
4291 ret = btrfs_commit_transaction(trans);
4295 * The cleaner kthread might still be doing iput
4296 * operations. Wait for it to finish so that
4297 * more space is released.
4299 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4300 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4303 btrfs_end_transaction(trans);
4307 trace_btrfs_space_reservation(fs_info,
4308 "space_info:enospc",
4309 data_sinfo->flags, bytes, 1);
4312 update_bytes_may_use(data_sinfo, bytes);
4313 trace_btrfs_space_reservation(fs_info, "space_info",
4314 data_sinfo->flags, bytes, 1);
4315 spin_unlock(&data_sinfo->lock);
4320 int btrfs_check_data_free_space(struct inode *inode,
4321 struct extent_changeset **reserved, u64 start, u64 len)
4323 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4326 /* align the range */
4327 len = round_up(start + len, fs_info->sectorsize) -
4328 round_down(start, fs_info->sectorsize);
4329 start = round_down(start, fs_info->sectorsize);
4331 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4335 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4336 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4338 btrfs_free_reserved_data_space_noquota(inode, start, len);
4345 * Called if we need to clear a data reservation for this inode
4346 * Normally in a error case.
4348 * This one will *NOT* use accurate qgroup reserved space API, just for case
4349 * which we can't sleep and is sure it won't affect qgroup reserved space.
4350 * Like clear_bit_hook().
4352 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4355 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4356 struct btrfs_space_info *data_sinfo;
4358 /* Make sure the range is aligned to sectorsize */
4359 len = round_up(start + len, fs_info->sectorsize) -
4360 round_down(start, fs_info->sectorsize);
4361 start = round_down(start, fs_info->sectorsize);
4363 data_sinfo = fs_info->data_sinfo;
4364 spin_lock(&data_sinfo->lock);
4365 update_bytes_may_use(data_sinfo, -len);
4366 trace_btrfs_space_reservation(fs_info, "space_info",
4367 data_sinfo->flags, len, 0);
4368 spin_unlock(&data_sinfo->lock);
4372 * Called if we need to clear a data reservation for this inode
4373 * Normally in a error case.
4375 * This one will handle the per-inode data rsv map for accurate reserved
4378 void btrfs_free_reserved_data_space(struct inode *inode,
4379 struct extent_changeset *reserved, u64 start, u64 len)
4381 struct btrfs_root *root = BTRFS_I(inode)->root;
4383 /* Make sure the range is aligned to sectorsize */
4384 len = round_up(start + len, root->fs_info->sectorsize) -
4385 round_down(start, root->fs_info->sectorsize);
4386 start = round_down(start, root->fs_info->sectorsize);
4388 btrfs_free_reserved_data_space_noquota(inode, start, len);
4389 btrfs_qgroup_free_data(inode, reserved, start, len);
4392 static void force_metadata_allocation(struct btrfs_fs_info *info)
4394 struct list_head *head = &info->space_info;
4395 struct btrfs_space_info *found;
4398 list_for_each_entry_rcu(found, head, list) {
4399 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4400 found->force_alloc = CHUNK_ALLOC_FORCE;
4405 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4407 return (global->size << 1);
4410 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4411 struct btrfs_space_info *sinfo, int force)
4413 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4414 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4417 if (force == CHUNK_ALLOC_FORCE)
4421 * We need to take into account the global rsv because for all intents
4422 * and purposes it's used space. Don't worry about locking the
4423 * global_rsv, it doesn't change except when the transaction commits.
4425 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4426 bytes_used += calc_global_rsv_need_space(global_rsv);
4429 * in limited mode, we want to have some free space up to
4430 * about 1% of the FS size.
4432 if (force == CHUNK_ALLOC_LIMITED) {
4433 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4434 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4436 if (sinfo->total_bytes - bytes_used < thresh)
4440 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4445 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4449 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4450 BTRFS_BLOCK_GROUP_RAID0 |
4451 BTRFS_BLOCK_GROUP_RAID5 |
4452 BTRFS_BLOCK_GROUP_RAID6))
4453 num_dev = fs_info->fs_devices->rw_devices;
4454 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4457 num_dev = 1; /* DUP or single */
4463 * If @is_allocation is true, reserve space in the system space info necessary
4464 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4467 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4469 struct btrfs_fs_info *fs_info = trans->fs_info;
4470 struct btrfs_space_info *info;
4477 * Needed because we can end up allocating a system chunk and for an
4478 * atomic and race free space reservation in the chunk block reserve.
4480 lockdep_assert_held(&fs_info->chunk_mutex);
4482 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4483 spin_lock(&info->lock);
4484 left = info->total_bytes - btrfs_space_info_used(info, true);
4485 spin_unlock(&info->lock);
4487 num_devs = get_profile_num_devs(fs_info, type);
4489 /* num_devs device items to update and 1 chunk item to add or remove */
4490 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4491 btrfs_calc_trans_metadata_size(fs_info, 1);
4493 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4494 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4495 left, thresh, type);
4496 dump_space_info(fs_info, info, 0, 0);
4499 if (left < thresh) {
4500 u64 flags = btrfs_system_alloc_profile(fs_info);
4503 * Ignore failure to create system chunk. We might end up not
4504 * needing it, as we might not need to COW all nodes/leafs from
4505 * the paths we visit in the chunk tree (they were already COWed
4506 * or created in the current transaction for example).
4508 ret = btrfs_alloc_chunk(trans, flags);
4512 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4513 &fs_info->chunk_block_rsv,
4514 thresh, BTRFS_RESERVE_NO_FLUSH);
4516 trans->chunk_bytes_reserved += thresh;
4521 * If force is CHUNK_ALLOC_FORCE:
4522 * - return 1 if it successfully allocates a chunk,
4523 * - return errors including -ENOSPC otherwise.
4524 * If force is NOT CHUNK_ALLOC_FORCE:
4525 * - return 0 if it doesn't need to allocate a new chunk,
4526 * - return 1 if it successfully allocates a chunk,
4527 * - return errors including -ENOSPC otherwise.
4529 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4532 struct btrfs_fs_info *fs_info = trans->fs_info;
4533 struct btrfs_space_info *space_info;
4534 bool wait_for_alloc = false;
4535 bool should_alloc = false;
4538 /* Don't re-enter if we're already allocating a chunk */
4539 if (trans->allocating_chunk)
4542 space_info = __find_space_info(fs_info, flags);
4546 spin_lock(&space_info->lock);
4547 if (force < space_info->force_alloc)
4548 force = space_info->force_alloc;
4549 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4550 if (space_info->full) {
4551 /* No more free physical space */
4556 spin_unlock(&space_info->lock);
4558 } else if (!should_alloc) {
4559 spin_unlock(&space_info->lock);
4561 } else if (space_info->chunk_alloc) {
4563 * Someone is already allocating, so we need to block
4564 * until this someone is finished and then loop to
4565 * recheck if we should continue with our allocation
4568 wait_for_alloc = true;
4569 spin_unlock(&space_info->lock);
4570 mutex_lock(&fs_info->chunk_mutex);
4571 mutex_unlock(&fs_info->chunk_mutex);
4573 /* Proceed with allocation */
4574 space_info->chunk_alloc = 1;
4575 wait_for_alloc = false;
4576 spin_unlock(&space_info->lock);
4580 } while (wait_for_alloc);
4582 mutex_lock(&fs_info->chunk_mutex);
4583 trans->allocating_chunk = true;
4586 * If we have mixed data/metadata chunks we want to make sure we keep
4587 * allocating mixed chunks instead of individual chunks.
4589 if (btrfs_mixed_space_info(space_info))
4590 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4593 * if we're doing a data chunk, go ahead and make sure that
4594 * we keep a reasonable number of metadata chunks allocated in the
4597 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4598 fs_info->data_chunk_allocations++;
4599 if (!(fs_info->data_chunk_allocations %
4600 fs_info->metadata_ratio))
4601 force_metadata_allocation(fs_info);
4605 * Check if we have enough space in SYSTEM chunk because we may need
4606 * to update devices.
4608 check_system_chunk(trans, flags);
4610 ret = btrfs_alloc_chunk(trans, flags);
4611 trans->allocating_chunk = false;
4613 spin_lock(&space_info->lock);
4616 space_info->full = 1;
4621 space_info->max_extent_size = 0;
4624 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4626 space_info->chunk_alloc = 0;
4627 spin_unlock(&space_info->lock);
4628 mutex_unlock(&fs_info->chunk_mutex);
4630 * When we allocate a new chunk we reserve space in the chunk block
4631 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4632 * add new nodes/leafs to it if we end up needing to do it when
4633 * inserting the chunk item and updating device items as part of the
4634 * second phase of chunk allocation, performed by
4635 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4636 * large number of new block groups to create in our transaction
4637 * handle's new_bgs list to avoid exhausting the chunk block reserve
4638 * in extreme cases - like having a single transaction create many new
4639 * block groups when starting to write out the free space caches of all
4640 * the block groups that were made dirty during the lifetime of the
4643 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
4644 btrfs_create_pending_block_groups(trans);
4649 static int can_overcommit(struct btrfs_fs_info *fs_info,
4650 struct btrfs_space_info *space_info, u64 bytes,
4651 enum btrfs_reserve_flush_enum flush,
4654 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4661 /* Don't overcommit when in mixed mode. */
4662 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4666 profile = btrfs_system_alloc_profile(fs_info);
4668 profile = btrfs_metadata_alloc_profile(fs_info);
4670 used = btrfs_space_info_used(space_info, false);
4673 * We only want to allow over committing if we have lots of actual space
4674 * free, but if we don't have enough space to handle the global reserve
4675 * space then we could end up having a real enospc problem when trying
4676 * to allocate a chunk or some other such important allocation.
4678 spin_lock(&global_rsv->lock);
4679 space_size = calc_global_rsv_need_space(global_rsv);
4680 spin_unlock(&global_rsv->lock);
4681 if (used + space_size >= space_info->total_bytes)
4684 used += space_info->bytes_may_use;
4686 avail = atomic64_read(&fs_info->free_chunk_space);
4689 * If we have dup, raid1 or raid10 then only half of the free
4690 * space is actually useable. For raid56, the space info used
4691 * doesn't include the parity drive, so we don't have to
4694 factor = btrfs_bg_type_to_factor(profile);
4695 avail = div_u64(avail, factor);
4698 * If we aren't flushing all things, let us overcommit up to
4699 * 1/2th of the space. If we can flush, don't let us overcommit
4700 * too much, let it overcommit up to 1/8 of the space.
4702 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4707 if (used + bytes < space_info->total_bytes + avail)
4712 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4713 unsigned long nr_pages, int nr_items)
4715 struct super_block *sb = fs_info->sb;
4717 if (down_read_trylock(&sb->s_umount)) {
4718 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4719 up_read(&sb->s_umount);
4722 * We needn't worry the filesystem going from r/w to r/o though
4723 * we don't acquire ->s_umount mutex, because the filesystem
4724 * should guarantee the delalloc inodes list be empty after
4725 * the filesystem is readonly(all dirty pages are written to
4728 btrfs_start_delalloc_roots(fs_info, nr_items);
4729 if (!current->journal_info)
4730 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4734 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4740 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4741 nr = div64_u64(to_reclaim, bytes);
4747 #define EXTENT_SIZE_PER_ITEM SZ_256K
4750 * shrink metadata reservation for delalloc
4752 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4753 u64 orig, bool wait_ordered)
4755 struct btrfs_space_info *space_info;
4756 struct btrfs_trans_handle *trans;
4761 unsigned long nr_pages;
4764 /* Calc the number of the pages we need flush for space reservation */
4765 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4766 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4768 trans = (struct btrfs_trans_handle *)current->journal_info;
4769 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4771 delalloc_bytes = percpu_counter_sum_positive(
4772 &fs_info->delalloc_bytes);
4773 if (delalloc_bytes == 0) {
4777 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4782 while (delalloc_bytes && loops < 3) {
4783 max_reclaim = min(delalloc_bytes, to_reclaim);
4784 nr_pages = max_reclaim >> PAGE_SHIFT;
4785 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4787 * We need to wait for the async pages to actually start before
4790 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4794 if (max_reclaim <= nr_pages)
4797 max_reclaim -= nr_pages;
4799 wait_event(fs_info->async_submit_wait,
4800 atomic_read(&fs_info->async_delalloc_pages) <=
4803 spin_lock(&space_info->lock);
4804 if (list_empty(&space_info->tickets) &&
4805 list_empty(&space_info->priority_tickets)) {
4806 spin_unlock(&space_info->lock);
4809 spin_unlock(&space_info->lock);
4812 if (wait_ordered && !trans) {
4813 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4815 time_left = schedule_timeout_killable(1);
4819 delalloc_bytes = percpu_counter_sum_positive(
4820 &fs_info->delalloc_bytes);
4824 struct reserve_ticket {
4827 struct list_head list;
4828 wait_queue_head_t wait;
4832 * maybe_commit_transaction - possibly commit the transaction if its ok to
4833 * @root - the root we're allocating for
4834 * @bytes - the number of bytes we want to reserve
4835 * @force - force the commit
4837 * This will check to make sure that committing the transaction will actually
4838 * get us somewhere and then commit the transaction if it does. Otherwise it
4839 * will return -ENOSPC.
4841 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4842 struct btrfs_space_info *space_info)
4844 struct reserve_ticket *ticket = NULL;
4845 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4846 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
4847 struct btrfs_trans_handle *trans;
4849 u64 reclaim_bytes = 0;
4851 trans = (struct btrfs_trans_handle *)current->journal_info;
4855 spin_lock(&space_info->lock);
4856 if (!list_empty(&space_info->priority_tickets))
4857 ticket = list_first_entry(&space_info->priority_tickets,
4858 struct reserve_ticket, list);
4859 else if (!list_empty(&space_info->tickets))
4860 ticket = list_first_entry(&space_info->tickets,
4861 struct reserve_ticket, list);
4862 bytes_needed = (ticket) ? ticket->bytes : 0;
4863 spin_unlock(&space_info->lock);
4868 /* See if there is enough pinned space to make this reservation */
4869 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4871 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4875 * See if there is some space in the delayed insertion reservation for
4878 if (space_info != delayed_rsv->space_info)
4881 spin_lock(&delayed_rsv->lock);
4882 reclaim_bytes += delayed_rsv->reserved;
4883 spin_unlock(&delayed_rsv->lock);
4885 spin_lock(&delayed_refs_rsv->lock);
4886 reclaim_bytes += delayed_refs_rsv->reserved;
4887 spin_unlock(&delayed_refs_rsv->lock);
4888 if (reclaim_bytes >= bytes_needed)
4890 bytes_needed -= reclaim_bytes;
4892 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4894 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) {
4899 trans = btrfs_join_transaction(fs_info->extent_root);
4903 return btrfs_commit_transaction(trans);
4907 * Try to flush some data based on policy set by @state. This is only advisory
4908 * and may fail for various reasons. The caller is supposed to examine the
4909 * state of @space_info to detect the outcome.
4911 static void flush_space(struct btrfs_fs_info *fs_info,
4912 struct btrfs_space_info *space_info, u64 num_bytes,
4915 struct btrfs_root *root = fs_info->extent_root;
4916 struct btrfs_trans_handle *trans;
4921 case FLUSH_DELAYED_ITEMS_NR:
4922 case FLUSH_DELAYED_ITEMS:
4923 if (state == FLUSH_DELAYED_ITEMS_NR)
4924 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4928 trans = btrfs_join_transaction(root);
4929 if (IS_ERR(trans)) {
4930 ret = PTR_ERR(trans);
4933 ret = btrfs_run_delayed_items_nr(trans, nr);
4934 btrfs_end_transaction(trans);
4936 case FLUSH_DELALLOC:
4937 case FLUSH_DELALLOC_WAIT:
4938 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4939 state == FLUSH_DELALLOC_WAIT);
4942 trans = btrfs_join_transaction(root);
4943 if (IS_ERR(trans)) {
4944 ret = PTR_ERR(trans);
4947 ret = do_chunk_alloc(trans,
4948 btrfs_metadata_alloc_profile(fs_info),
4949 CHUNK_ALLOC_NO_FORCE);
4950 btrfs_end_transaction(trans);
4951 if (ret > 0 || ret == -ENOSPC)
4955 ret = may_commit_transaction(fs_info, space_info);
4962 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4968 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4969 struct btrfs_space_info *space_info,
4972 struct reserve_ticket *ticket;
4977 list_for_each_entry(ticket, &space_info->tickets, list)
4978 to_reclaim += ticket->bytes;
4979 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4980 to_reclaim += ticket->bytes;
4984 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4985 if (can_overcommit(fs_info, space_info, to_reclaim,
4986 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4989 used = btrfs_space_info_used(space_info, true);
4991 if (can_overcommit(fs_info, space_info, SZ_1M,
4992 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4993 expected = div_factor_fine(space_info->total_bytes, 95);
4995 expected = div_factor_fine(space_info->total_bytes, 90);
4997 if (used > expected)
4998 to_reclaim = used - expected;
5001 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5002 space_info->bytes_reserved);
5006 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5007 struct btrfs_space_info *space_info,
5008 u64 used, bool system_chunk)
5010 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5012 /* If we're just plain full then async reclaim just slows us down. */
5013 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5016 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5020 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5021 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5024 static void wake_all_tickets(struct list_head *head)
5026 struct reserve_ticket *ticket;
5028 while (!list_empty(head)) {
5029 ticket = list_first_entry(head, struct reserve_ticket, list);
5030 list_del_init(&ticket->list);
5031 ticket->error = -ENOSPC;
5032 wake_up(&ticket->wait);
5037 * This is for normal flushers, we can wait all goddamned day if we want to. We
5038 * will loop and continuously try to flush as long as we are making progress.
5039 * We count progress as clearing off tickets each time we have to loop.
5041 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5043 struct btrfs_fs_info *fs_info;
5044 struct btrfs_space_info *space_info;
5047 int commit_cycles = 0;
5048 u64 last_tickets_id;
5050 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5051 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5053 spin_lock(&space_info->lock);
5054 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5057 space_info->flush = 0;
5058 spin_unlock(&space_info->lock);
5061 last_tickets_id = space_info->tickets_id;
5062 spin_unlock(&space_info->lock);
5064 flush_state = FLUSH_DELAYED_ITEMS_NR;
5066 flush_space(fs_info, space_info, to_reclaim, flush_state);
5067 spin_lock(&space_info->lock);
5068 if (list_empty(&space_info->tickets)) {
5069 space_info->flush = 0;
5070 spin_unlock(&space_info->lock);
5073 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5076 if (last_tickets_id == space_info->tickets_id) {
5079 last_tickets_id = space_info->tickets_id;
5080 flush_state = FLUSH_DELAYED_ITEMS_NR;
5085 if (flush_state > COMMIT_TRANS) {
5087 if (commit_cycles > 2) {
5088 wake_all_tickets(&space_info->tickets);
5089 space_info->flush = 0;
5091 flush_state = FLUSH_DELAYED_ITEMS_NR;
5094 spin_unlock(&space_info->lock);
5095 } while (flush_state <= COMMIT_TRANS);
5098 void btrfs_init_async_reclaim_work(struct work_struct *work)
5100 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5103 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5104 struct btrfs_space_info *space_info,
5105 struct reserve_ticket *ticket)
5108 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5110 spin_lock(&space_info->lock);
5111 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5114 spin_unlock(&space_info->lock);
5117 spin_unlock(&space_info->lock);
5120 flush_space(fs_info, space_info, to_reclaim, flush_state);
5122 spin_lock(&space_info->lock);
5123 if (ticket->bytes == 0) {
5124 spin_unlock(&space_info->lock);
5127 spin_unlock(&space_info->lock);
5130 * Priority flushers can't wait on delalloc without
5133 if (flush_state == FLUSH_DELALLOC ||
5134 flush_state == FLUSH_DELALLOC_WAIT)
5135 flush_state = ALLOC_CHUNK;
5136 } while (flush_state < COMMIT_TRANS);
5139 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5140 struct btrfs_space_info *space_info,
5141 struct reserve_ticket *ticket, u64 orig_bytes)
5147 spin_lock(&space_info->lock);
5148 while (ticket->bytes > 0 && ticket->error == 0) {
5149 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5154 spin_unlock(&space_info->lock);
5158 finish_wait(&ticket->wait, &wait);
5159 spin_lock(&space_info->lock);
5162 ret = ticket->error;
5163 if (!list_empty(&ticket->list))
5164 list_del_init(&ticket->list);
5165 if (ticket->bytes && ticket->bytes < orig_bytes) {
5166 u64 num_bytes = orig_bytes - ticket->bytes;
5167 update_bytes_may_use(space_info, -num_bytes);
5168 trace_btrfs_space_reservation(fs_info, "space_info",
5169 space_info->flags, num_bytes, 0);
5171 spin_unlock(&space_info->lock);
5177 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5178 * @root - the root we're allocating for
5179 * @space_info - the space info we want to allocate from
5180 * @orig_bytes - the number of bytes we want
5181 * @flush - whether or not we can flush to make our reservation
5183 * This will reserve orig_bytes number of bytes from the space info associated
5184 * with the block_rsv. If there is not enough space it will make an attempt to
5185 * flush out space to make room. It will do this by flushing delalloc if
5186 * possible or committing the transaction. If flush is 0 then no attempts to
5187 * regain reservations will be made and this will fail if there is not enough
5190 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5191 struct btrfs_space_info *space_info,
5193 enum btrfs_reserve_flush_enum flush,
5196 struct reserve_ticket ticket;
5201 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5203 spin_lock(&space_info->lock);
5205 used = btrfs_space_info_used(space_info, true);
5208 * If we have enough space then hooray, make our reservation and carry
5209 * on. If not see if we can overcommit, and if we can, hooray carry on.
5210 * If not things get more complicated.
5212 if (used + orig_bytes <= space_info->total_bytes) {
5213 update_bytes_may_use(space_info, orig_bytes);
5214 trace_btrfs_space_reservation(fs_info, "space_info",
5215 space_info->flags, orig_bytes, 1);
5217 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5219 update_bytes_may_use(space_info, orig_bytes);
5220 trace_btrfs_space_reservation(fs_info, "space_info",
5221 space_info->flags, orig_bytes, 1);
5226 * If we couldn't make a reservation then setup our reservation ticket
5227 * and kick the async worker if it's not already running.
5229 * If we are a priority flusher then we just need to add our ticket to
5230 * the list and we will do our own flushing further down.
5232 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5233 ticket.bytes = orig_bytes;
5235 init_waitqueue_head(&ticket.wait);
5236 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5237 list_add_tail(&ticket.list, &space_info->tickets);
5238 if (!space_info->flush) {
5239 space_info->flush = 1;
5240 trace_btrfs_trigger_flush(fs_info,
5244 queue_work(system_unbound_wq,
5245 &fs_info->async_reclaim_work);
5248 list_add_tail(&ticket.list,
5249 &space_info->priority_tickets);
5251 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5254 * We will do the space reservation dance during log replay,
5255 * which means we won't have fs_info->fs_root set, so don't do
5256 * the async reclaim as we will panic.
5258 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5259 need_do_async_reclaim(fs_info, space_info,
5260 used, system_chunk) &&
5261 !work_busy(&fs_info->async_reclaim_work)) {
5262 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5263 orig_bytes, flush, "preempt");
5264 queue_work(system_unbound_wq,
5265 &fs_info->async_reclaim_work);
5268 spin_unlock(&space_info->lock);
5269 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5272 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5273 return wait_reserve_ticket(fs_info, space_info, &ticket,
5277 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5278 spin_lock(&space_info->lock);
5280 if (ticket.bytes < orig_bytes) {
5281 u64 num_bytes = orig_bytes - ticket.bytes;
5282 update_bytes_may_use(space_info, -num_bytes);
5283 trace_btrfs_space_reservation(fs_info, "space_info",
5288 list_del_init(&ticket.list);
5291 spin_unlock(&space_info->lock);
5292 ASSERT(list_empty(&ticket.list));
5297 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5298 * @root - the root we're allocating for
5299 * @block_rsv - the block_rsv we're allocating for
5300 * @orig_bytes - the number of bytes we want
5301 * @flush - whether or not we can flush to make our reservation
5303 * This will reserve orgi_bytes number of bytes from the space info associated
5304 * with the block_rsv. If there is not enough space it will make an attempt to
5305 * flush out space to make room. It will do this by flushing delalloc if
5306 * possible or committing the transaction. If flush is 0 then no attempts to
5307 * regain reservations will be made and this will fail if there is not enough
5310 static int reserve_metadata_bytes(struct btrfs_root *root,
5311 struct btrfs_block_rsv *block_rsv,
5313 enum btrfs_reserve_flush_enum flush)
5315 struct btrfs_fs_info *fs_info = root->fs_info;
5316 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5318 bool system_chunk = (root == fs_info->chunk_root);
5320 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5321 orig_bytes, flush, system_chunk);
5322 if (ret == -ENOSPC &&
5323 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5324 if (block_rsv != global_rsv &&
5325 !block_rsv_use_bytes(global_rsv, orig_bytes))
5328 if (ret == -ENOSPC) {
5329 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5330 block_rsv->space_info->flags,
5333 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5334 dump_space_info(fs_info, block_rsv->space_info,
5340 static struct btrfs_block_rsv *get_block_rsv(
5341 const struct btrfs_trans_handle *trans,
5342 const struct btrfs_root *root)
5344 struct btrfs_fs_info *fs_info = root->fs_info;
5345 struct btrfs_block_rsv *block_rsv = NULL;
5347 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5348 (root == fs_info->csum_root && trans->adding_csums) ||
5349 (root == fs_info->uuid_root))
5350 block_rsv = trans->block_rsv;
5353 block_rsv = root->block_rsv;
5356 block_rsv = &fs_info->empty_block_rsv;
5361 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5365 spin_lock(&block_rsv->lock);
5366 if (block_rsv->reserved >= num_bytes) {
5367 block_rsv->reserved -= num_bytes;
5368 if (block_rsv->reserved < block_rsv->size)
5369 block_rsv->full = 0;
5372 spin_unlock(&block_rsv->lock);
5376 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5377 u64 num_bytes, bool update_size)
5379 spin_lock(&block_rsv->lock);
5380 block_rsv->reserved += num_bytes;
5382 block_rsv->size += num_bytes;
5383 else if (block_rsv->reserved >= block_rsv->size)
5384 block_rsv->full = 1;
5385 spin_unlock(&block_rsv->lock);
5388 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5389 struct btrfs_block_rsv *dest, u64 num_bytes,
5392 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5395 if (global_rsv->space_info != dest->space_info)
5398 spin_lock(&global_rsv->lock);
5399 min_bytes = div_factor(global_rsv->size, min_factor);
5400 if (global_rsv->reserved < min_bytes + num_bytes) {
5401 spin_unlock(&global_rsv->lock);
5404 global_rsv->reserved -= num_bytes;
5405 if (global_rsv->reserved < global_rsv->size)
5406 global_rsv->full = 0;
5407 spin_unlock(&global_rsv->lock);
5409 block_rsv_add_bytes(dest, num_bytes, true);
5414 * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv.
5415 * @fs_info - the fs info for our fs.
5416 * @src - the source block rsv to transfer from.
5417 * @num_bytes - the number of bytes to transfer.
5419 * This transfers up to the num_bytes amount from the src rsv to the
5420 * delayed_refs_rsv. Any extra bytes are returned to the space info.
5422 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
5423 struct btrfs_block_rsv *src,
5426 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
5429 spin_lock(&src->lock);
5430 src->reserved -= num_bytes;
5431 src->size -= num_bytes;
5432 spin_unlock(&src->lock);
5434 spin_lock(&delayed_refs_rsv->lock);
5435 if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) {
5436 u64 delta = delayed_refs_rsv->size -
5437 delayed_refs_rsv->reserved;
5438 if (num_bytes > delta) {
5439 to_free = num_bytes - delta;
5443 to_free = num_bytes;
5448 delayed_refs_rsv->reserved += num_bytes;
5449 if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size)
5450 delayed_refs_rsv->full = 1;
5451 spin_unlock(&delayed_refs_rsv->lock);
5454 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5457 space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info,
5462 * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage.
5463 * @fs_info - the fs_info for our fs.
5464 * @flush - control how we can flush for this reservation.
5466 * This will refill the delayed block_rsv up to 1 items size worth of space and
5467 * will return -ENOSPC if we can't make the reservation.
5469 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
5470 enum btrfs_reserve_flush_enum flush)
5472 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5473 u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1);
5477 spin_lock(&block_rsv->lock);
5478 if (block_rsv->reserved < block_rsv->size) {
5479 num_bytes = block_rsv->size - block_rsv->reserved;
5480 num_bytes = min(num_bytes, limit);
5482 spin_unlock(&block_rsv->lock);
5487 ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv,
5491 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5492 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5498 * This is for space we already have accounted in space_info->bytes_may_use, so
5499 * basically when we're returning space from block_rsv's.
5501 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5502 struct btrfs_space_info *space_info,
5505 struct reserve_ticket *ticket;
5506 struct list_head *head;
5508 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5509 bool check_overcommit = false;
5511 spin_lock(&space_info->lock);
5512 head = &space_info->priority_tickets;
5515 * If we are over our limit then we need to check and see if we can
5516 * overcommit, and if we can't then we just need to free up our space
5517 * and not satisfy any requests.
5519 used = btrfs_space_info_used(space_info, true);
5520 if (used - num_bytes >= space_info->total_bytes)
5521 check_overcommit = true;
5523 while (!list_empty(head) && num_bytes) {
5524 ticket = list_first_entry(head, struct reserve_ticket,
5527 * We use 0 bytes because this space is already reserved, so
5528 * adding the ticket space would be a double count.
5530 if (check_overcommit &&
5531 !can_overcommit(fs_info, space_info, 0, flush, false))
5533 if (num_bytes >= ticket->bytes) {
5534 list_del_init(&ticket->list);
5535 num_bytes -= ticket->bytes;
5537 space_info->tickets_id++;
5538 wake_up(&ticket->wait);
5540 ticket->bytes -= num_bytes;
5545 if (num_bytes && head == &space_info->priority_tickets) {
5546 head = &space_info->tickets;
5547 flush = BTRFS_RESERVE_FLUSH_ALL;
5550 update_bytes_may_use(space_info, -num_bytes);
5551 trace_btrfs_space_reservation(fs_info, "space_info",
5552 space_info->flags, num_bytes, 0);
5553 spin_unlock(&space_info->lock);
5557 * This is for newly allocated space that isn't accounted in
5558 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5559 * we use this helper.
5561 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5562 struct btrfs_space_info *space_info,
5565 struct reserve_ticket *ticket;
5566 struct list_head *head = &space_info->priority_tickets;
5569 while (!list_empty(head) && num_bytes) {
5570 ticket = list_first_entry(head, struct reserve_ticket,
5572 if (num_bytes >= ticket->bytes) {
5573 trace_btrfs_space_reservation(fs_info, "space_info",
5576 list_del_init(&ticket->list);
5577 num_bytes -= ticket->bytes;
5578 update_bytes_may_use(space_info, ticket->bytes);
5580 space_info->tickets_id++;
5581 wake_up(&ticket->wait);
5583 trace_btrfs_space_reservation(fs_info, "space_info",
5586 update_bytes_may_use(space_info, num_bytes);
5587 ticket->bytes -= num_bytes;
5592 if (num_bytes && head == &space_info->priority_tickets) {
5593 head = &space_info->tickets;
5598 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5599 struct btrfs_block_rsv *block_rsv,
5600 struct btrfs_block_rsv *dest, u64 num_bytes,
5601 u64 *qgroup_to_release_ret)
5603 struct btrfs_space_info *space_info = block_rsv->space_info;
5604 u64 qgroup_to_release = 0;
5607 spin_lock(&block_rsv->lock);
5608 if (num_bytes == (u64)-1) {
5609 num_bytes = block_rsv->size;
5610 qgroup_to_release = block_rsv->qgroup_rsv_size;
5612 block_rsv->size -= num_bytes;
5613 if (block_rsv->reserved >= block_rsv->size) {
5614 num_bytes = block_rsv->reserved - block_rsv->size;
5615 block_rsv->reserved = block_rsv->size;
5616 block_rsv->full = 1;
5620 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5621 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5622 block_rsv->qgroup_rsv_size;
5623 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5625 qgroup_to_release = 0;
5627 spin_unlock(&block_rsv->lock);
5630 if (num_bytes > 0) {
5632 spin_lock(&dest->lock);
5636 bytes_to_add = dest->size - dest->reserved;
5637 bytes_to_add = min(num_bytes, bytes_to_add);
5638 dest->reserved += bytes_to_add;
5639 if (dest->reserved >= dest->size)
5641 num_bytes -= bytes_to_add;
5643 spin_unlock(&dest->lock);
5646 space_info_add_old_bytes(fs_info, space_info,
5649 if (qgroup_to_release_ret)
5650 *qgroup_to_release_ret = qgroup_to_release;
5654 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5655 struct btrfs_block_rsv *dst, u64 num_bytes,
5660 ret = block_rsv_use_bytes(src, num_bytes);
5664 block_rsv_add_bytes(dst, num_bytes, update_size);
5668 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5670 memset(rsv, 0, sizeof(*rsv));
5671 spin_lock_init(&rsv->lock);
5675 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5676 struct btrfs_block_rsv *rsv,
5677 unsigned short type)
5679 btrfs_init_block_rsv(rsv, type);
5680 rsv->space_info = __find_space_info(fs_info,
5681 BTRFS_BLOCK_GROUP_METADATA);
5684 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5685 unsigned short type)
5687 struct btrfs_block_rsv *block_rsv;
5689 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5693 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5697 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5698 struct btrfs_block_rsv *rsv)
5702 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5706 int btrfs_block_rsv_add(struct btrfs_root *root,
5707 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5708 enum btrfs_reserve_flush_enum flush)
5715 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5717 block_rsv_add_bytes(block_rsv, num_bytes, true);
5722 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5730 spin_lock(&block_rsv->lock);
5731 num_bytes = div_factor(block_rsv->size, min_factor);
5732 if (block_rsv->reserved >= num_bytes)
5734 spin_unlock(&block_rsv->lock);
5739 int btrfs_block_rsv_refill(struct btrfs_root *root,
5740 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5741 enum btrfs_reserve_flush_enum flush)
5749 spin_lock(&block_rsv->lock);
5750 num_bytes = min_reserved;
5751 if (block_rsv->reserved >= num_bytes)
5754 num_bytes -= block_rsv->reserved;
5755 spin_unlock(&block_rsv->lock);
5760 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5762 block_rsv_add_bytes(block_rsv, num_bytes, false);
5770 * btrfs_inode_rsv_refill - refill the inode block rsv.
5771 * @inode - the inode we are refilling.
5772 * @flush - the flusing restriction.
5774 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5775 * block_rsv->size as the minimum size. We'll either refill the missing amount
5776 * or return if we already have enough space. This will also handle the resreve
5777 * tracepoint for the reserved amount.
5779 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5780 enum btrfs_reserve_flush_enum flush)
5782 struct btrfs_root *root = inode->root;
5783 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5785 u64 qgroup_num_bytes = 0;
5788 spin_lock(&block_rsv->lock);
5789 if (block_rsv->reserved < block_rsv->size)
5790 num_bytes = block_rsv->size - block_rsv->reserved;
5791 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5792 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5793 block_rsv->qgroup_rsv_reserved;
5794 spin_unlock(&block_rsv->lock);
5799 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5802 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5804 block_rsv_add_bytes(block_rsv, num_bytes, false);
5805 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5806 btrfs_ino(inode), num_bytes, 1);
5808 /* Don't forget to increase qgroup_rsv_reserved */
5809 spin_lock(&block_rsv->lock);
5810 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5811 spin_unlock(&block_rsv->lock);
5813 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5817 static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5818 struct btrfs_block_rsv *block_rsv,
5819 u64 num_bytes, u64 *qgroup_to_release)
5821 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5822 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
5823 struct btrfs_block_rsv *target = delayed_rsv;
5825 if (target->full || target == block_rsv)
5826 target = global_rsv;
5828 if (block_rsv->space_info != target->space_info)
5831 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
5835 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5836 struct btrfs_block_rsv *block_rsv,
5839 __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL);
5843 * btrfs_inode_rsv_release - release any excessive reservation.
5844 * @inode - the inode we need to release from.
5845 * @qgroup_free - free or convert qgroup meta.
5846 * Unlike normal operation, qgroup meta reservation needs to know if we are
5847 * freeing qgroup reservation or just converting it into per-trans. Normally
5848 * @qgroup_free is true for error handling, and false for normal release.
5850 * This is the same as btrfs_block_rsv_release, except that it handles the
5851 * tracepoint for the reservation.
5853 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5855 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5856 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5858 u64 qgroup_to_release = 0;
5861 * Since we statically set the block_rsv->size we just want to say we
5862 * are releasing 0 bytes, and then we'll just get the reservation over
5865 released = __btrfs_block_rsv_release(fs_info, block_rsv, 0,
5866 &qgroup_to_release);
5868 trace_btrfs_space_reservation(fs_info, "delalloc",
5869 btrfs_ino(inode), released, 0);
5871 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5873 btrfs_qgroup_convert_reserved_meta(inode->root,
5878 * btrfs_delayed_refs_rsv_release - release a ref head's reservation.
5879 * @fs_info - the fs_info for our fs.
5880 * @nr - the number of items to drop.
5882 * This drops the delayed ref head's count from the delayed refs rsv and frees
5883 * any excess reservation we had.
5885 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr)
5887 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5888 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5889 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr);
5892 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv,
5895 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5899 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5901 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5902 struct btrfs_space_info *sinfo = block_rsv->space_info;
5906 * The global block rsv is based on the size of the extent tree, the
5907 * checksum tree and the root tree. If the fs is empty we want to set
5908 * it to a minimal amount for safety.
5910 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5911 btrfs_root_used(&fs_info->csum_root->root_item) +
5912 btrfs_root_used(&fs_info->tree_root->root_item);
5913 num_bytes = max_t(u64, num_bytes, SZ_16M);
5915 spin_lock(&sinfo->lock);
5916 spin_lock(&block_rsv->lock);
5918 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5920 if (block_rsv->reserved < block_rsv->size) {
5921 num_bytes = btrfs_space_info_used(sinfo, true);
5922 if (sinfo->total_bytes > num_bytes) {
5923 num_bytes = sinfo->total_bytes - num_bytes;
5924 num_bytes = min(num_bytes,
5925 block_rsv->size - block_rsv->reserved);
5926 block_rsv->reserved += num_bytes;
5927 update_bytes_may_use(sinfo, num_bytes);
5928 trace_btrfs_space_reservation(fs_info, "space_info",
5929 sinfo->flags, num_bytes,
5932 } else if (block_rsv->reserved > block_rsv->size) {
5933 num_bytes = block_rsv->reserved - block_rsv->size;
5934 update_bytes_may_use(sinfo, -num_bytes);
5935 trace_btrfs_space_reservation(fs_info, "space_info",
5936 sinfo->flags, num_bytes, 0);
5937 block_rsv->reserved = block_rsv->size;
5940 if (block_rsv->reserved == block_rsv->size)
5941 block_rsv->full = 1;
5943 block_rsv->full = 0;
5945 spin_unlock(&block_rsv->lock);
5946 spin_unlock(&sinfo->lock);
5949 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5951 struct btrfs_space_info *space_info;
5953 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5954 fs_info->chunk_block_rsv.space_info = space_info;
5956 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5957 fs_info->global_block_rsv.space_info = space_info;
5958 fs_info->trans_block_rsv.space_info = space_info;
5959 fs_info->empty_block_rsv.space_info = space_info;
5960 fs_info->delayed_block_rsv.space_info = space_info;
5961 fs_info->delayed_refs_rsv.space_info = space_info;
5963 fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
5964 fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
5965 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5966 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5967 if (fs_info->quota_root)
5968 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5969 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5971 update_global_block_rsv(fs_info);
5974 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5976 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5978 WARN_ON(fs_info->trans_block_rsv.size > 0);
5979 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5980 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5981 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5982 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5983 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5984 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
5985 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
5989 * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv
5990 * @trans - the trans that may have generated delayed refs
5992 * This is to be called anytime we may have adjusted trans->delayed_ref_updates,
5993 * it'll calculate the additional size and add it to the delayed_refs_rsv.
5995 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans)
5997 struct btrfs_fs_info *fs_info = trans->fs_info;
5998 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
6001 if (!trans->delayed_ref_updates)
6004 num_bytes = btrfs_calc_trans_metadata_size(fs_info,
6005 trans->delayed_ref_updates);
6006 spin_lock(&delayed_rsv->lock);
6007 delayed_rsv->size += num_bytes;
6008 delayed_rsv->full = 0;
6009 spin_unlock(&delayed_rsv->lock);
6010 trans->delayed_ref_updates = 0;
6014 * To be called after all the new block groups attached to the transaction
6015 * handle have been created (btrfs_create_pending_block_groups()).
6017 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
6019 struct btrfs_fs_info *fs_info = trans->fs_info;
6021 if (!trans->chunk_bytes_reserved)
6024 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
6026 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
6027 trans->chunk_bytes_reserved, NULL);
6028 trans->chunk_bytes_reserved = 0;
6032 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
6033 * root: the root of the parent directory
6034 * rsv: block reservation
6035 * items: the number of items that we need do reservation
6036 * use_global_rsv: allow fallback to the global block reservation
6038 * This function is used to reserve the space for snapshot/subvolume
6039 * creation and deletion. Those operations are different with the
6040 * common file/directory operations, they change two fs/file trees
6041 * and root tree, the number of items that the qgroup reserves is
6042 * different with the free space reservation. So we can not use
6043 * the space reservation mechanism in start_transaction().
6045 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
6046 struct btrfs_block_rsv *rsv, int items,
6047 bool use_global_rsv)
6049 u64 qgroup_num_bytes = 0;
6052 struct btrfs_fs_info *fs_info = root->fs_info;
6053 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6055 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6056 /* One for parent inode, two for dir entries */
6057 qgroup_num_bytes = 3 * fs_info->nodesize;
6058 ret = btrfs_qgroup_reserve_meta_prealloc(root,
6059 qgroup_num_bytes, true);
6064 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6065 rsv->space_info = __find_space_info(fs_info,
6066 BTRFS_BLOCK_GROUP_METADATA);
6067 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6068 BTRFS_RESERVE_FLUSH_ALL);
6070 if (ret == -ENOSPC && use_global_rsv)
6071 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
6073 if (ret && qgroup_num_bytes)
6074 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
6079 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6080 struct btrfs_block_rsv *rsv)
6082 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6085 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6086 struct btrfs_inode *inode)
6088 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6089 u64 reserve_size = 0;
6090 u64 qgroup_rsv_size = 0;
6092 unsigned outstanding_extents;
6094 lockdep_assert_held(&inode->lock);
6095 outstanding_extents = inode->outstanding_extents;
6096 if (outstanding_extents)
6097 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6098 outstanding_extents + 1);
6099 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6101 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6104 * For qgroup rsv, the calculation is very simple:
6105 * account one nodesize for each outstanding extent
6107 * This is overestimating in most cases.
6109 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
6111 spin_lock(&block_rsv->lock);
6112 block_rsv->size = reserve_size;
6113 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6114 spin_unlock(&block_rsv->lock);
6117 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6119 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6120 unsigned nr_extents;
6121 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6123 bool delalloc_lock = true;
6125 /* If we are a free space inode we need to not flush since we will be in
6126 * the middle of a transaction commit. We also don't need the delalloc
6127 * mutex since we won't race with anybody. We need this mostly to make
6128 * lockdep shut its filthy mouth.
6130 * If we have a transaction open (can happen if we call truncate_block
6131 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6133 if (btrfs_is_free_space_inode(inode)) {
6134 flush = BTRFS_RESERVE_NO_FLUSH;
6135 delalloc_lock = false;
6137 if (current->journal_info)
6138 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6140 if (btrfs_transaction_in_commit(fs_info))
6141 schedule_timeout(1);
6145 mutex_lock(&inode->delalloc_mutex);
6147 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6149 /* Add our new extents and calculate the new rsv size. */
6150 spin_lock(&inode->lock);
6151 nr_extents = count_max_extents(num_bytes);
6152 btrfs_mod_outstanding_extents(inode, nr_extents);
6153 inode->csum_bytes += num_bytes;
6154 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6155 spin_unlock(&inode->lock);
6157 ret = btrfs_inode_rsv_refill(inode, flush);
6162 mutex_unlock(&inode->delalloc_mutex);
6166 spin_lock(&inode->lock);
6167 nr_extents = count_max_extents(num_bytes);
6168 btrfs_mod_outstanding_extents(inode, -nr_extents);
6169 inode->csum_bytes -= num_bytes;
6170 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6171 spin_unlock(&inode->lock);
6173 btrfs_inode_rsv_release(inode, true);
6175 mutex_unlock(&inode->delalloc_mutex);
6180 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6181 * @inode: the inode to release the reservation for.
6182 * @num_bytes: the number of bytes we are releasing.
6183 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6185 * This will release the metadata reservation for an inode. This can be called
6186 * once we complete IO for a given set of bytes to release their metadata
6187 * reservations, or on error for the same reason.
6189 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6192 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6194 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6195 spin_lock(&inode->lock);
6196 inode->csum_bytes -= num_bytes;
6197 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6198 spin_unlock(&inode->lock);
6200 if (btrfs_is_testing(fs_info))
6203 btrfs_inode_rsv_release(inode, qgroup_free);
6207 * btrfs_delalloc_release_extents - release our outstanding_extents
6208 * @inode: the inode to balance the reservation for.
6209 * @num_bytes: the number of bytes we originally reserved with
6210 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6212 * When we reserve space we increase outstanding_extents for the extents we may
6213 * add. Once we've set the range as delalloc or created our ordered extents we
6214 * have outstanding_extents to track the real usage, so we use this to free our
6215 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6216 * with btrfs_delalloc_reserve_metadata.
6218 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6221 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6222 unsigned num_extents;
6224 spin_lock(&inode->lock);
6225 num_extents = count_max_extents(num_bytes);
6226 btrfs_mod_outstanding_extents(inode, -num_extents);
6227 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6228 spin_unlock(&inode->lock);
6230 if (btrfs_is_testing(fs_info))
6233 btrfs_inode_rsv_release(inode, qgroup_free);
6237 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6239 * @inode: inode we're writing to
6240 * @start: start range we are writing to
6241 * @len: how long the range we are writing to
6242 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6243 * current reservation.
6245 * This will do the following things
6247 * o reserve space in data space info for num bytes
6248 * and reserve precious corresponding qgroup space
6249 * (Done in check_data_free_space)
6251 * o reserve space for metadata space, based on the number of outstanding
6252 * extents and how much csums will be needed
6253 * also reserve metadata space in a per root over-reserve method.
6254 * o add to the inodes->delalloc_bytes
6255 * o add it to the fs_info's delalloc inodes list.
6256 * (Above 3 all done in delalloc_reserve_metadata)
6258 * Return 0 for success
6259 * Return <0 for error(-ENOSPC or -EQUOT)
6261 int btrfs_delalloc_reserve_space(struct inode *inode,
6262 struct extent_changeset **reserved, u64 start, u64 len)
6266 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6269 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6271 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6276 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6277 * @inode: inode we're releasing space for
6278 * @start: start position of the space already reserved
6279 * @len: the len of the space already reserved
6280 * @release_bytes: the len of the space we consumed or didn't use
6282 * This function will release the metadata space that was not used and will
6283 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6284 * list if there are no delalloc bytes left.
6285 * Also it will handle the qgroup reserved space.
6287 void btrfs_delalloc_release_space(struct inode *inode,
6288 struct extent_changeset *reserved,
6289 u64 start, u64 len, bool qgroup_free)
6291 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6292 btrfs_free_reserved_data_space(inode, reserved, start, len);
6295 static int update_block_group(struct btrfs_trans_handle *trans,
6296 struct btrfs_fs_info *info, u64 bytenr,
6297 u64 num_bytes, int alloc)
6299 struct btrfs_block_group_cache *cache = NULL;
6300 u64 total = num_bytes;
6306 /* block accounting for super block */
6307 spin_lock(&info->delalloc_root_lock);
6308 old_val = btrfs_super_bytes_used(info->super_copy);
6310 old_val += num_bytes;
6312 old_val -= num_bytes;
6313 btrfs_set_super_bytes_used(info->super_copy, old_val);
6314 spin_unlock(&info->delalloc_root_lock);
6317 cache = btrfs_lookup_block_group(info, bytenr);
6322 factor = btrfs_bg_type_to_factor(cache->flags);
6325 * If this block group has free space cache written out, we
6326 * need to make sure to load it if we are removing space. This
6327 * is because we need the unpinning stage to actually add the
6328 * space back to the block group, otherwise we will leak space.
6330 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6331 cache_block_group(cache, 1);
6333 byte_in_group = bytenr - cache->key.objectid;
6334 WARN_ON(byte_in_group > cache->key.offset);
6336 spin_lock(&cache->space_info->lock);
6337 spin_lock(&cache->lock);
6339 if (btrfs_test_opt(info, SPACE_CACHE) &&
6340 cache->disk_cache_state < BTRFS_DC_CLEAR)
6341 cache->disk_cache_state = BTRFS_DC_CLEAR;
6343 old_val = btrfs_block_group_used(&cache->item);
6344 num_bytes = min(total, cache->key.offset - byte_in_group);
6346 old_val += num_bytes;
6347 btrfs_set_block_group_used(&cache->item, old_val);
6348 cache->reserved -= num_bytes;
6349 cache->space_info->bytes_reserved -= num_bytes;
6350 cache->space_info->bytes_used += num_bytes;
6351 cache->space_info->disk_used += num_bytes * factor;
6352 spin_unlock(&cache->lock);
6353 spin_unlock(&cache->space_info->lock);
6355 old_val -= num_bytes;
6356 btrfs_set_block_group_used(&cache->item, old_val);
6357 cache->pinned += num_bytes;
6358 update_bytes_pinned(cache->space_info, num_bytes);
6359 cache->space_info->bytes_used -= num_bytes;
6360 cache->space_info->disk_used -= num_bytes * factor;
6361 spin_unlock(&cache->lock);
6362 spin_unlock(&cache->space_info->lock);
6364 trace_btrfs_space_reservation(info, "pinned",
6365 cache->space_info->flags,
6367 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6369 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6370 set_extent_dirty(info->pinned_extents,
6371 bytenr, bytenr + num_bytes - 1,
6372 GFP_NOFS | __GFP_NOFAIL);
6375 spin_lock(&trans->transaction->dirty_bgs_lock);
6376 if (list_empty(&cache->dirty_list)) {
6377 list_add_tail(&cache->dirty_list,
6378 &trans->transaction->dirty_bgs);
6379 trans->transaction->num_dirty_bgs++;
6380 trans->delayed_ref_updates++;
6381 btrfs_get_block_group(cache);
6383 spin_unlock(&trans->transaction->dirty_bgs_lock);
6386 * No longer have used bytes in this block group, queue it for
6387 * deletion. We do this after adding the block group to the
6388 * dirty list to avoid races between cleaner kthread and space
6391 if (!alloc && old_val == 0)
6392 btrfs_mark_bg_unused(cache);
6394 btrfs_put_block_group(cache);
6396 bytenr += num_bytes;
6399 /* Modified block groups are accounted for in the delayed_refs_rsv. */
6400 btrfs_update_delayed_refs_rsv(trans);
6404 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6406 struct btrfs_block_group_cache *cache;
6409 spin_lock(&fs_info->block_group_cache_lock);
6410 bytenr = fs_info->first_logical_byte;
6411 spin_unlock(&fs_info->block_group_cache_lock);
6413 if (bytenr < (u64)-1)
6416 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6420 bytenr = cache->key.objectid;
6421 btrfs_put_block_group(cache);
6426 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6427 struct btrfs_block_group_cache *cache,
6428 u64 bytenr, u64 num_bytes, int reserved)
6430 spin_lock(&cache->space_info->lock);
6431 spin_lock(&cache->lock);
6432 cache->pinned += num_bytes;
6433 update_bytes_pinned(cache->space_info, num_bytes);
6435 cache->reserved -= num_bytes;
6436 cache->space_info->bytes_reserved -= num_bytes;
6438 spin_unlock(&cache->lock);
6439 spin_unlock(&cache->space_info->lock);
6441 trace_btrfs_space_reservation(fs_info, "pinned",
6442 cache->space_info->flags, num_bytes, 1);
6443 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6444 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6445 set_extent_dirty(fs_info->pinned_extents, bytenr,
6446 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6451 * this function must be called within transaction
6453 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6454 u64 bytenr, u64 num_bytes, int reserved)
6456 struct btrfs_block_group_cache *cache;
6458 cache = btrfs_lookup_block_group(fs_info, bytenr);
6459 BUG_ON(!cache); /* Logic error */
6461 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6463 btrfs_put_block_group(cache);
6468 * this function must be called within transaction
6470 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6471 u64 bytenr, u64 num_bytes)
6473 struct btrfs_block_group_cache *cache;
6476 cache = btrfs_lookup_block_group(fs_info, bytenr);
6481 * pull in the free space cache (if any) so that our pin
6482 * removes the free space from the cache. We have load_only set
6483 * to one because the slow code to read in the free extents does check
6484 * the pinned extents.
6486 cache_block_group(cache, 1);
6488 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6490 /* remove us from the free space cache (if we're there at all) */
6491 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6492 btrfs_put_block_group(cache);
6496 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6497 u64 start, u64 num_bytes)
6500 struct btrfs_block_group_cache *block_group;
6501 struct btrfs_caching_control *caching_ctl;
6503 block_group = btrfs_lookup_block_group(fs_info, start);
6507 cache_block_group(block_group, 0);
6508 caching_ctl = get_caching_control(block_group);
6512 BUG_ON(!block_group_cache_done(block_group));
6513 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6515 mutex_lock(&caching_ctl->mutex);
6517 if (start >= caching_ctl->progress) {
6518 ret = add_excluded_extent(fs_info, start, num_bytes);
6519 } else if (start + num_bytes <= caching_ctl->progress) {
6520 ret = btrfs_remove_free_space(block_group,
6523 num_bytes = caching_ctl->progress - start;
6524 ret = btrfs_remove_free_space(block_group,
6529 num_bytes = (start + num_bytes) -
6530 caching_ctl->progress;
6531 start = caching_ctl->progress;
6532 ret = add_excluded_extent(fs_info, start, num_bytes);
6535 mutex_unlock(&caching_ctl->mutex);
6536 put_caching_control(caching_ctl);
6538 btrfs_put_block_group(block_group);
6542 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6543 struct extent_buffer *eb)
6545 struct btrfs_file_extent_item *item;
6546 struct btrfs_key key;
6551 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6554 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6555 btrfs_item_key_to_cpu(eb, &key, i);
6556 if (key.type != BTRFS_EXTENT_DATA_KEY)
6558 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6559 found_type = btrfs_file_extent_type(eb, item);
6560 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6562 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6564 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6565 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6566 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6575 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6577 atomic_inc(&bg->reservations);
6580 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6583 struct btrfs_block_group_cache *bg;
6585 bg = btrfs_lookup_block_group(fs_info, start);
6587 if (atomic_dec_and_test(&bg->reservations))
6588 wake_up_var(&bg->reservations);
6589 btrfs_put_block_group(bg);
6592 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6594 struct btrfs_space_info *space_info = bg->space_info;
6598 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6602 * Our block group is read only but before we set it to read only,
6603 * some task might have had allocated an extent from it already, but it
6604 * has not yet created a respective ordered extent (and added it to a
6605 * root's list of ordered extents).
6606 * Therefore wait for any task currently allocating extents, since the
6607 * block group's reservations counter is incremented while a read lock
6608 * on the groups' semaphore is held and decremented after releasing
6609 * the read access on that semaphore and creating the ordered extent.
6611 down_write(&space_info->groups_sem);
6612 up_write(&space_info->groups_sem);
6614 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6618 * btrfs_add_reserved_bytes - update the block_group and space info counters
6619 * @cache: The cache we are manipulating
6620 * @ram_bytes: The number of bytes of file content, and will be same to
6621 * @num_bytes except for the compress path.
6622 * @num_bytes: The number of bytes in question
6623 * @delalloc: The blocks are allocated for the delalloc write
6625 * This is called by the allocator when it reserves space. If this is a
6626 * reservation and the block group has become read only we cannot make the
6627 * reservation and return -EAGAIN, otherwise this function always succeeds.
6629 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6630 u64 ram_bytes, u64 num_bytes, int delalloc)
6632 struct btrfs_space_info *space_info = cache->space_info;
6635 spin_lock(&space_info->lock);
6636 spin_lock(&cache->lock);
6640 cache->reserved += num_bytes;
6641 space_info->bytes_reserved += num_bytes;
6642 update_bytes_may_use(space_info, -ram_bytes);
6644 cache->delalloc_bytes += num_bytes;
6646 spin_unlock(&cache->lock);
6647 spin_unlock(&space_info->lock);
6652 * btrfs_free_reserved_bytes - update the block_group and space info counters
6653 * @cache: The cache we are manipulating
6654 * @num_bytes: The number of bytes in question
6655 * @delalloc: The blocks are allocated for the delalloc write
6657 * This is called by somebody who is freeing space that was never actually used
6658 * on disk. For example if you reserve some space for a new leaf in transaction
6659 * A and before transaction A commits you free that leaf, you call this with
6660 * reserve set to 0 in order to clear the reservation.
6663 static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6664 u64 num_bytes, int delalloc)
6666 struct btrfs_space_info *space_info = cache->space_info;
6668 spin_lock(&space_info->lock);
6669 spin_lock(&cache->lock);
6671 space_info->bytes_readonly += num_bytes;
6672 cache->reserved -= num_bytes;
6673 space_info->bytes_reserved -= num_bytes;
6674 space_info->max_extent_size = 0;
6677 cache->delalloc_bytes -= num_bytes;
6678 spin_unlock(&cache->lock);
6679 spin_unlock(&space_info->lock);
6681 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6683 struct btrfs_caching_control *next;
6684 struct btrfs_caching_control *caching_ctl;
6685 struct btrfs_block_group_cache *cache;
6687 down_write(&fs_info->commit_root_sem);
6689 list_for_each_entry_safe(caching_ctl, next,
6690 &fs_info->caching_block_groups, list) {
6691 cache = caching_ctl->block_group;
6692 if (block_group_cache_done(cache)) {
6693 cache->last_byte_to_unpin = (u64)-1;
6694 list_del_init(&caching_ctl->list);
6695 put_caching_control(caching_ctl);
6697 cache->last_byte_to_unpin = caching_ctl->progress;
6701 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6702 fs_info->pinned_extents = &fs_info->freed_extents[1];
6704 fs_info->pinned_extents = &fs_info->freed_extents[0];
6706 up_write(&fs_info->commit_root_sem);
6708 update_global_block_rsv(fs_info);
6712 * Returns the free cluster for the given space info and sets empty_cluster to
6713 * what it should be based on the mount options.
6715 static struct btrfs_free_cluster *
6716 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6717 struct btrfs_space_info *space_info, u64 *empty_cluster)
6719 struct btrfs_free_cluster *ret = NULL;
6722 if (btrfs_mixed_space_info(space_info))
6725 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6726 ret = &fs_info->meta_alloc_cluster;
6727 if (btrfs_test_opt(fs_info, SSD))
6728 *empty_cluster = SZ_2M;
6730 *empty_cluster = SZ_64K;
6731 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6732 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6733 *empty_cluster = SZ_2M;
6734 ret = &fs_info->data_alloc_cluster;
6740 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6742 const bool return_free_space)
6744 struct btrfs_block_group_cache *cache = NULL;
6745 struct btrfs_space_info *space_info;
6746 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6747 struct btrfs_free_cluster *cluster = NULL;
6749 u64 total_unpinned = 0;
6750 u64 empty_cluster = 0;
6753 while (start <= end) {
6756 start >= cache->key.objectid + cache->key.offset) {
6758 btrfs_put_block_group(cache);
6760 cache = btrfs_lookup_block_group(fs_info, start);
6761 BUG_ON(!cache); /* Logic error */
6763 cluster = fetch_cluster_info(fs_info,
6766 empty_cluster <<= 1;
6769 len = cache->key.objectid + cache->key.offset - start;
6770 len = min(len, end + 1 - start);
6772 if (start < cache->last_byte_to_unpin) {
6773 len = min(len, cache->last_byte_to_unpin - start);
6774 if (return_free_space)
6775 btrfs_add_free_space(cache, start, len);
6779 total_unpinned += len;
6780 space_info = cache->space_info;
6783 * If this space cluster has been marked as fragmented and we've
6784 * unpinned enough in this block group to potentially allow a
6785 * cluster to be created inside of it go ahead and clear the
6788 if (cluster && cluster->fragmented &&
6789 total_unpinned > empty_cluster) {
6790 spin_lock(&cluster->lock);
6791 cluster->fragmented = 0;
6792 spin_unlock(&cluster->lock);
6795 spin_lock(&space_info->lock);
6796 spin_lock(&cache->lock);
6797 cache->pinned -= len;
6798 update_bytes_pinned(space_info, -len);
6800 trace_btrfs_space_reservation(fs_info, "pinned",
6801 space_info->flags, len, 0);
6802 space_info->max_extent_size = 0;
6803 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6804 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6806 space_info->bytes_readonly += len;
6809 spin_unlock(&cache->lock);
6810 if (!readonly && return_free_space &&
6811 global_rsv->space_info == space_info) {
6814 spin_lock(&global_rsv->lock);
6815 if (!global_rsv->full) {
6816 to_add = min(len, global_rsv->size -
6817 global_rsv->reserved);
6818 global_rsv->reserved += to_add;
6819 update_bytes_may_use(space_info, to_add);
6820 if (global_rsv->reserved >= global_rsv->size)
6821 global_rsv->full = 1;
6822 trace_btrfs_space_reservation(fs_info,
6828 spin_unlock(&global_rsv->lock);
6829 /* Add to any tickets we may have */
6831 space_info_add_new_bytes(fs_info, space_info,
6834 spin_unlock(&space_info->lock);
6838 btrfs_put_block_group(cache);
6842 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6844 struct btrfs_fs_info *fs_info = trans->fs_info;
6845 struct btrfs_block_group_cache *block_group, *tmp;
6846 struct list_head *deleted_bgs;
6847 struct extent_io_tree *unpin;
6852 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6853 unpin = &fs_info->freed_extents[1];
6855 unpin = &fs_info->freed_extents[0];
6857 while (!trans->aborted) {
6858 struct extent_state *cached_state = NULL;
6860 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6861 ret = find_first_extent_bit(unpin, 0, &start, &end,
6862 EXTENT_DIRTY, &cached_state);
6864 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6868 if (btrfs_test_opt(fs_info, DISCARD))
6869 ret = btrfs_discard_extent(fs_info, start,
6870 end + 1 - start, NULL);
6872 clear_extent_dirty(unpin, start, end, &cached_state);
6873 unpin_extent_range(fs_info, start, end, true);
6874 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6875 free_extent_state(cached_state);
6880 * Transaction is finished. We don't need the lock anymore. We
6881 * do need to clean up the block groups in case of a transaction
6884 deleted_bgs = &trans->transaction->deleted_bgs;
6885 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6889 if (!trans->aborted)
6890 ret = btrfs_discard_extent(fs_info,
6891 block_group->key.objectid,
6892 block_group->key.offset,
6895 list_del_init(&block_group->bg_list);
6896 btrfs_put_block_group_trimming(block_group);
6897 btrfs_put_block_group(block_group);
6900 const char *errstr = btrfs_decode_error(ret);
6902 "discard failed while removing blockgroup: errno=%d %s",
6910 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6911 struct btrfs_delayed_ref_node *node, u64 parent,
6912 u64 root_objectid, u64 owner_objectid,
6913 u64 owner_offset, int refs_to_drop,
6914 struct btrfs_delayed_extent_op *extent_op)
6916 struct btrfs_fs_info *info = trans->fs_info;
6917 struct btrfs_key key;
6918 struct btrfs_path *path;
6919 struct btrfs_root *extent_root = info->extent_root;
6920 struct extent_buffer *leaf;
6921 struct btrfs_extent_item *ei;
6922 struct btrfs_extent_inline_ref *iref;
6925 int extent_slot = 0;
6926 int found_extent = 0;
6930 u64 bytenr = node->bytenr;
6931 u64 num_bytes = node->num_bytes;
6933 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6935 path = btrfs_alloc_path();
6939 path->reada = READA_FORWARD;
6940 path->leave_spinning = 1;
6942 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6943 BUG_ON(!is_data && refs_to_drop != 1);
6946 skinny_metadata = false;
6948 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
6949 parent, root_objectid, owner_objectid,
6952 extent_slot = path->slots[0];
6953 while (extent_slot >= 0) {
6954 btrfs_item_key_to_cpu(path->nodes[0], &key,
6956 if (key.objectid != bytenr)
6958 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6959 key.offset == num_bytes) {
6963 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6964 key.offset == owner_objectid) {
6968 if (path->slots[0] - extent_slot > 5)
6973 if (!found_extent) {
6975 ret = remove_extent_backref(trans, path, NULL,
6977 is_data, &last_ref);
6979 btrfs_abort_transaction(trans, ret);
6982 btrfs_release_path(path);
6983 path->leave_spinning = 1;
6985 key.objectid = bytenr;
6986 key.type = BTRFS_EXTENT_ITEM_KEY;
6987 key.offset = num_bytes;
6989 if (!is_data && skinny_metadata) {
6990 key.type = BTRFS_METADATA_ITEM_KEY;
6991 key.offset = owner_objectid;
6994 ret = btrfs_search_slot(trans, extent_root,
6996 if (ret > 0 && skinny_metadata && path->slots[0]) {
6998 * Couldn't find our skinny metadata item,
6999 * see if we have ye olde extent item.
7002 btrfs_item_key_to_cpu(path->nodes[0], &key,
7004 if (key.objectid == bytenr &&
7005 key.type == BTRFS_EXTENT_ITEM_KEY &&
7006 key.offset == num_bytes)
7010 if (ret > 0 && skinny_metadata) {
7011 skinny_metadata = false;
7012 key.objectid = bytenr;
7013 key.type = BTRFS_EXTENT_ITEM_KEY;
7014 key.offset = num_bytes;
7015 btrfs_release_path(path);
7016 ret = btrfs_search_slot(trans, extent_root,
7022 "umm, got %d back from search, was looking for %llu",
7025 btrfs_print_leaf(path->nodes[0]);
7028 btrfs_abort_transaction(trans, ret);
7031 extent_slot = path->slots[0];
7033 } else if (WARN_ON(ret == -ENOENT)) {
7034 btrfs_print_leaf(path->nodes[0]);
7036 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7037 bytenr, parent, root_objectid, owner_objectid,
7039 btrfs_abort_transaction(trans, ret);
7042 btrfs_abort_transaction(trans, ret);
7046 leaf = path->nodes[0];
7047 item_size = btrfs_item_size_nr(leaf, extent_slot);
7048 if (unlikely(item_size < sizeof(*ei))) {
7050 btrfs_print_v0_err(info);
7051 btrfs_abort_transaction(trans, ret);
7054 ei = btrfs_item_ptr(leaf, extent_slot,
7055 struct btrfs_extent_item);
7056 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7057 key.type == BTRFS_EXTENT_ITEM_KEY) {
7058 struct btrfs_tree_block_info *bi;
7059 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7060 bi = (struct btrfs_tree_block_info *)(ei + 1);
7061 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7064 refs = btrfs_extent_refs(leaf, ei);
7065 if (refs < refs_to_drop) {
7067 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7068 refs_to_drop, refs, bytenr);
7070 btrfs_abort_transaction(trans, ret);
7073 refs -= refs_to_drop;
7077 __run_delayed_extent_op(extent_op, leaf, ei);
7079 * In the case of inline back ref, reference count will
7080 * be updated by remove_extent_backref
7083 BUG_ON(!found_extent);
7085 btrfs_set_extent_refs(leaf, ei, refs);
7086 btrfs_mark_buffer_dirty(leaf);
7089 ret = remove_extent_backref(trans, path, iref,
7090 refs_to_drop, is_data,
7093 btrfs_abort_transaction(trans, ret);
7099 BUG_ON(is_data && refs_to_drop !=
7100 extent_data_ref_count(path, iref));
7102 BUG_ON(path->slots[0] != extent_slot);
7104 BUG_ON(path->slots[0] != extent_slot + 1);
7105 path->slots[0] = extent_slot;
7111 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7114 btrfs_abort_transaction(trans, ret);
7117 btrfs_release_path(path);
7120 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7122 btrfs_abort_transaction(trans, ret);
7127 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
7129 btrfs_abort_transaction(trans, ret);
7133 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7135 btrfs_abort_transaction(trans, ret);
7139 btrfs_release_path(path);
7142 btrfs_free_path(path);
7147 * when we free an block, it is possible (and likely) that we free the last
7148 * delayed ref for that extent as well. This searches the delayed ref tree for
7149 * a given extent, and if there are no other delayed refs to be processed, it
7150 * removes it from the tree.
7152 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7155 struct btrfs_delayed_ref_head *head;
7156 struct btrfs_delayed_ref_root *delayed_refs;
7159 delayed_refs = &trans->transaction->delayed_refs;
7160 spin_lock(&delayed_refs->lock);
7161 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7163 goto out_delayed_unlock;
7165 spin_lock(&head->lock);
7166 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
7169 if (cleanup_extent_op(head) != NULL)
7173 * waiting for the lock here would deadlock. If someone else has it
7174 * locked they are already in the process of dropping it anyway
7176 if (!mutex_trylock(&head->mutex))
7179 btrfs_delete_ref_head(delayed_refs, head);
7180 head->processing = 0;
7182 spin_unlock(&head->lock);
7183 spin_unlock(&delayed_refs->lock);
7185 BUG_ON(head->extent_op);
7186 if (head->must_insert_reserved)
7189 cleanup_ref_head_accounting(trans, head);
7190 mutex_unlock(&head->mutex);
7191 btrfs_put_delayed_ref_head(head);
7194 spin_unlock(&head->lock);
7197 spin_unlock(&delayed_refs->lock);
7201 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7202 struct btrfs_root *root,
7203 struct extent_buffer *buf,
7204 u64 parent, int last_ref)
7206 struct btrfs_fs_info *fs_info = root->fs_info;
7210 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7211 int old_ref_mod, new_ref_mod;
7213 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7214 root->root_key.objectid,
7215 btrfs_header_level(buf), 0,
7216 BTRFS_DROP_DELAYED_REF);
7217 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
7219 root->root_key.objectid,
7220 btrfs_header_level(buf),
7221 BTRFS_DROP_DELAYED_REF, NULL,
7222 &old_ref_mod, &new_ref_mod);
7223 BUG_ON(ret); /* -ENOMEM */
7224 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7227 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7228 struct btrfs_block_group_cache *cache;
7230 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7231 ret = check_ref_cleanup(trans, buf->start);
7237 cache = btrfs_lookup_block_group(fs_info, buf->start);
7239 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7240 pin_down_extent(fs_info, cache, buf->start,
7242 btrfs_put_block_group(cache);
7246 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7248 btrfs_add_free_space(cache, buf->start, buf->len);
7249 btrfs_free_reserved_bytes(cache, buf->len, 0);
7250 btrfs_put_block_group(cache);
7251 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7255 add_pinned_bytes(fs_info, buf->len, true,
7256 root->root_key.objectid);
7260 * Deleting the buffer, clear the corrupt flag since it doesn't
7263 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7267 /* Can return -ENOMEM */
7268 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7269 struct btrfs_root *root,
7270 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7271 u64 owner, u64 offset)
7273 struct btrfs_fs_info *fs_info = root->fs_info;
7274 int old_ref_mod, new_ref_mod;
7277 if (btrfs_is_testing(fs_info))
7280 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7281 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7282 root_objectid, owner, offset,
7283 BTRFS_DROP_DELAYED_REF);
7286 * tree log blocks never actually go into the extent allocation
7287 * tree, just update pinning info and exit early.
7289 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7290 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7291 /* unlocks the pinned mutex */
7292 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7293 old_ref_mod = new_ref_mod = 0;
7295 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7296 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7298 root_objectid, (int)owner,
7299 BTRFS_DROP_DELAYED_REF, NULL,
7300 &old_ref_mod, &new_ref_mod);
7302 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7304 root_objectid, owner, offset,
7305 0, BTRFS_DROP_DELAYED_REF,
7306 &old_ref_mod, &new_ref_mod);
7309 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7310 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7312 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7319 * when we wait for progress in the block group caching, its because
7320 * our allocation attempt failed at least once. So, we must sleep
7321 * and let some progress happen before we try again.
7323 * This function will sleep at least once waiting for new free space to
7324 * show up, and then it will check the block group free space numbers
7325 * for our min num_bytes. Another option is to have it go ahead
7326 * and look in the rbtree for a free extent of a given size, but this
7329 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7330 * any of the information in this block group.
7332 static noinline void
7333 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7336 struct btrfs_caching_control *caching_ctl;
7338 caching_ctl = get_caching_control(cache);
7342 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7343 (cache->free_space_ctl->free_space >= num_bytes));
7345 put_caching_control(caching_ctl);
7349 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7351 struct btrfs_caching_control *caching_ctl;
7354 caching_ctl = get_caching_control(cache);
7356 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7358 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7359 if (cache->cached == BTRFS_CACHE_ERROR)
7361 put_caching_control(caching_ctl);
7365 enum btrfs_loop_type {
7366 LOOP_CACHING_NOWAIT = 0,
7367 LOOP_CACHING_WAIT = 1,
7368 LOOP_ALLOC_CHUNK = 2,
7369 LOOP_NO_EMPTY_SIZE = 3,
7373 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7377 down_read(&cache->data_rwsem);
7381 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7384 btrfs_get_block_group(cache);
7386 down_read(&cache->data_rwsem);
7389 static struct btrfs_block_group_cache *
7390 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7391 struct btrfs_free_cluster *cluster,
7394 struct btrfs_block_group_cache *used_bg = NULL;
7396 spin_lock(&cluster->refill_lock);
7398 used_bg = cluster->block_group;
7402 if (used_bg == block_group)
7405 btrfs_get_block_group(used_bg);
7410 if (down_read_trylock(&used_bg->data_rwsem))
7413 spin_unlock(&cluster->refill_lock);
7415 /* We should only have one-level nested. */
7416 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7418 spin_lock(&cluster->refill_lock);
7419 if (used_bg == cluster->block_group)
7422 up_read(&used_bg->data_rwsem);
7423 btrfs_put_block_group(used_bg);
7428 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7432 up_read(&cache->data_rwsem);
7433 btrfs_put_block_group(cache);
7437 * Structure used internally for find_free_extent() function. Wraps needed
7440 struct find_free_extent_ctl {
7441 /* Basic allocation info */
7448 /* Where to start the search inside the bg */
7451 /* For clustered allocation */
7454 bool have_caching_bg;
7455 bool orig_have_caching_bg;
7457 /* RAID index, converted from flags */
7461 * Current loop number, check find_free_extent_update_loop() for details
7466 * Whether we're refilling a cluster, if true we need to re-search
7467 * current block group but don't try to refill the cluster again.
7469 bool retry_clustered;
7472 * Whether we're updating free space cache, if true we need to re-search
7473 * current block group but don't try updating free space cache again.
7475 bool retry_unclustered;
7477 /* If current block group is cached */
7480 /* Max contiguous hole found */
7481 u64 max_extent_size;
7483 /* Total free space from free space cache, not always contiguous */
7484 u64 total_free_space;
7492 * Helper function for find_free_extent().
7494 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
7495 * Return -EAGAIN to inform caller that we need to re-search this block group
7496 * Return >0 to inform caller that we find nothing
7497 * Return 0 means we have found a location and set ffe_ctl->found_offset.
7499 static int find_free_extent_clustered(struct btrfs_block_group_cache *bg,
7500 struct btrfs_free_cluster *last_ptr,
7501 struct find_free_extent_ctl *ffe_ctl,
7502 struct btrfs_block_group_cache **cluster_bg_ret)
7504 struct btrfs_fs_info *fs_info = bg->fs_info;
7505 struct btrfs_block_group_cache *cluster_bg;
7506 u64 aligned_cluster;
7510 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
7512 goto refill_cluster;
7513 if (cluster_bg != bg && (cluster_bg->ro ||
7514 !block_group_bits(cluster_bg, ffe_ctl->flags)))
7515 goto release_cluster;
7517 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
7518 ffe_ctl->num_bytes, cluster_bg->key.objectid,
7519 &ffe_ctl->max_extent_size);
7521 /* We have a block, we're done */
7522 spin_unlock(&last_ptr->refill_lock);
7523 trace_btrfs_reserve_extent_cluster(cluster_bg,
7524 ffe_ctl->search_start, ffe_ctl->num_bytes);
7525 *cluster_bg_ret = cluster_bg;
7526 ffe_ctl->found_offset = offset;
7529 WARN_ON(last_ptr->block_group != cluster_bg);
7533 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
7534 * lets just skip it and let the allocator find whatever block it can
7535 * find. If we reach this point, we will have tried the cluster
7536 * allocator plenty of times and not have found anything, so we are
7537 * likely way too fragmented for the clustering stuff to find anything.
7539 * However, if the cluster is taken from the current block group,
7540 * release the cluster first, so that we stand a better chance of
7541 * succeeding in the unclustered allocation.
7543 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
7544 spin_unlock(&last_ptr->refill_lock);
7545 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7549 /* This cluster didn't work out, free it and start over */
7550 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7552 if (cluster_bg != bg)
7553 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7556 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
7557 spin_unlock(&last_ptr->refill_lock);
7561 aligned_cluster = max_t(u64,
7562 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
7563 bg->full_stripe_len);
7564 ret = btrfs_find_space_cluster(fs_info, bg, last_ptr,
7565 ffe_ctl->search_start, ffe_ctl->num_bytes,
7568 /* Now pull our allocation out of this cluster */
7569 offset = btrfs_alloc_from_cluster(bg, last_ptr,
7570 ffe_ctl->num_bytes, ffe_ctl->search_start,
7571 &ffe_ctl->max_extent_size);
7573 /* We found one, proceed */
7574 spin_unlock(&last_ptr->refill_lock);
7575 trace_btrfs_reserve_extent_cluster(bg,
7576 ffe_ctl->search_start,
7577 ffe_ctl->num_bytes);
7578 ffe_ctl->found_offset = offset;
7581 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
7582 !ffe_ctl->retry_clustered) {
7583 spin_unlock(&last_ptr->refill_lock);
7585 ffe_ctl->retry_clustered = true;
7586 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7587 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
7591 * At this point we either didn't find a cluster or we weren't able to
7592 * allocate a block from our cluster. Free the cluster we've been
7593 * trying to use, and go to the next block group.
7595 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7596 spin_unlock(&last_ptr->refill_lock);
7601 * Return >0 to inform caller that we find nothing
7602 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
7603 * Return -EAGAIN to inform caller that we need to re-search this block group
7605 static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg,
7606 struct btrfs_free_cluster *last_ptr,
7607 struct find_free_extent_ctl *ffe_ctl)
7612 * We are doing an unclustered allocation, set the fragmented flag so
7613 * we don't bother trying to setup a cluster again until we get more
7616 if (unlikely(last_ptr)) {
7617 spin_lock(&last_ptr->lock);
7618 last_ptr->fragmented = 1;
7619 spin_unlock(&last_ptr->lock);
7621 if (ffe_ctl->cached) {
7622 struct btrfs_free_space_ctl *free_space_ctl;
7624 free_space_ctl = bg->free_space_ctl;
7625 spin_lock(&free_space_ctl->tree_lock);
7626 if (free_space_ctl->free_space <
7627 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
7628 ffe_ctl->empty_size) {
7629 ffe_ctl->total_free_space = max_t(u64,
7630 ffe_ctl->total_free_space,
7631 free_space_ctl->free_space);
7632 spin_unlock(&free_space_ctl->tree_lock);
7635 spin_unlock(&free_space_ctl->tree_lock);
7638 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
7639 ffe_ctl->num_bytes, ffe_ctl->empty_size,
7640 &ffe_ctl->max_extent_size);
7643 * If we didn't find a chunk, and we haven't failed on this block group
7644 * before, and this block group is in the middle of caching and we are
7645 * ok with waiting, then go ahead and wait for progress to be made, and
7646 * set @retry_unclustered to true.
7648 * If @retry_unclustered is true then we've already waited on this
7649 * block group once and should move on to the next block group.
7651 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
7652 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
7653 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7654 ffe_ctl->empty_size);
7655 ffe_ctl->retry_unclustered = true;
7657 } else if (!offset) {
7660 ffe_ctl->found_offset = offset;
7665 * Return >0 means caller needs to re-search for free extent
7666 * Return 0 means we have the needed free extent.
7667 * Return <0 means we failed to locate any free extent.
7669 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
7670 struct btrfs_free_cluster *last_ptr,
7671 struct btrfs_key *ins,
7672 struct find_free_extent_ctl *ffe_ctl,
7673 int full_search, bool use_cluster)
7675 struct btrfs_root *root = fs_info->extent_root;
7678 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
7679 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
7680 ffe_ctl->orig_have_caching_bg = true;
7682 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
7683 ffe_ctl->have_caching_bg)
7686 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
7689 if (ins->objectid) {
7690 if (!use_cluster && last_ptr) {
7691 spin_lock(&last_ptr->lock);
7692 last_ptr->window_start = ins->objectid;
7693 spin_unlock(&last_ptr->lock);
7699 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7700 * caching kthreads as we move along
7701 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7702 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7703 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7706 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
7708 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
7710 * We want to skip the LOOP_CACHING_WAIT step if we
7711 * don't have any uncached bgs and we've already done a
7712 * full search through.
7714 if (ffe_ctl->orig_have_caching_bg || !full_search)
7715 ffe_ctl->loop = LOOP_CACHING_WAIT;
7717 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
7722 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
7723 struct btrfs_trans_handle *trans;
7726 trans = current->journal_info;
7730 trans = btrfs_join_transaction(root);
7732 if (IS_ERR(trans)) {
7733 ret = PTR_ERR(trans);
7737 ret = do_chunk_alloc(trans, ffe_ctl->flags,
7741 * If we can't allocate a new chunk we've already looped
7742 * through at least once, move on to the NO_EMPTY_SIZE
7746 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
7748 /* Do not bail out on ENOSPC since we can do more. */
7749 if (ret < 0 && ret != -ENOSPC)
7750 btrfs_abort_transaction(trans, ret);
7754 btrfs_end_transaction(trans);
7759 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
7761 * Don't loop again if we already have no empty_size and
7764 if (ffe_ctl->empty_size == 0 &&
7765 ffe_ctl->empty_cluster == 0)
7767 ffe_ctl->empty_size = 0;
7768 ffe_ctl->empty_cluster = 0;
7776 * walks the btree of allocated extents and find a hole of a given size.
7777 * The key ins is changed to record the hole:
7778 * ins->objectid == start position
7779 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7780 * ins->offset == the size of the hole.
7781 * Any available blocks before search_start are skipped.
7783 * If there is no suitable free space, we will record the max size of
7784 * the free space extent currently.
7786 * The overall logic and call chain:
7788 * find_free_extent()
7789 * |- Iterate through all block groups
7790 * | |- Get a valid block group
7791 * | |- Try to do clustered allocation in that block group
7792 * | |- Try to do unclustered allocation in that block group
7793 * | |- Check if the result is valid
7794 * | | |- If valid, then exit
7795 * | |- Jump to next block group
7797 * |- Push harder to find free extents
7798 * |- If not found, re-iterate all block groups
7800 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7801 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7802 u64 hint_byte, struct btrfs_key *ins,
7803 u64 flags, int delalloc)
7806 struct btrfs_free_cluster *last_ptr = NULL;
7807 struct btrfs_block_group_cache *block_group = NULL;
7808 struct find_free_extent_ctl ffe_ctl = {0};
7809 struct btrfs_space_info *space_info;
7810 bool use_cluster = true;
7811 bool full_search = false;
7813 WARN_ON(num_bytes < fs_info->sectorsize);
7815 ffe_ctl.ram_bytes = ram_bytes;
7816 ffe_ctl.num_bytes = num_bytes;
7817 ffe_ctl.empty_size = empty_size;
7818 ffe_ctl.flags = flags;
7819 ffe_ctl.search_start = 0;
7820 ffe_ctl.retry_clustered = false;
7821 ffe_ctl.retry_unclustered = false;
7822 ffe_ctl.delalloc = delalloc;
7823 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
7824 ffe_ctl.have_caching_bg = false;
7825 ffe_ctl.orig_have_caching_bg = false;
7826 ffe_ctl.found_offset = 0;
7828 ins->type = BTRFS_EXTENT_ITEM_KEY;
7832 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7834 space_info = __find_space_info(fs_info, flags);
7836 btrfs_err(fs_info, "No space info for %llu", flags);
7841 * If our free space is heavily fragmented we may not be able to make
7842 * big contiguous allocations, so instead of doing the expensive search
7843 * for free space, simply return ENOSPC with our max_extent_size so we
7844 * can go ahead and search for a more manageable chunk.
7846 * If our max_extent_size is large enough for our allocation simply
7847 * disable clustering since we will likely not be able to find enough
7848 * space to create a cluster and induce latency trying.
7850 if (unlikely(space_info->max_extent_size)) {
7851 spin_lock(&space_info->lock);
7852 if (space_info->max_extent_size &&
7853 num_bytes > space_info->max_extent_size) {
7854 ins->offset = space_info->max_extent_size;
7855 spin_unlock(&space_info->lock);
7857 } else if (space_info->max_extent_size) {
7858 use_cluster = false;
7860 spin_unlock(&space_info->lock);
7863 last_ptr = fetch_cluster_info(fs_info, space_info,
7864 &ffe_ctl.empty_cluster);
7866 spin_lock(&last_ptr->lock);
7867 if (last_ptr->block_group)
7868 hint_byte = last_ptr->window_start;
7869 if (last_ptr->fragmented) {
7871 * We still set window_start so we can keep track of the
7872 * last place we found an allocation to try and save
7875 hint_byte = last_ptr->window_start;
7876 use_cluster = false;
7878 spin_unlock(&last_ptr->lock);
7881 ffe_ctl.search_start = max(ffe_ctl.search_start,
7882 first_logical_byte(fs_info, 0));
7883 ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
7884 if (ffe_ctl.search_start == hint_byte) {
7885 block_group = btrfs_lookup_block_group(fs_info,
7886 ffe_ctl.search_start);
7888 * we don't want to use the block group if it doesn't match our
7889 * allocation bits, or if its not cached.
7891 * However if we are re-searching with an ideal block group
7892 * picked out then we don't care that the block group is cached.
7894 if (block_group && block_group_bits(block_group, flags) &&
7895 block_group->cached != BTRFS_CACHE_NO) {
7896 down_read(&space_info->groups_sem);
7897 if (list_empty(&block_group->list) ||
7900 * someone is removing this block group,
7901 * we can't jump into the have_block_group
7902 * target because our list pointers are not
7905 btrfs_put_block_group(block_group);
7906 up_read(&space_info->groups_sem);
7908 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
7909 block_group->flags);
7910 btrfs_lock_block_group(block_group, delalloc);
7911 goto have_block_group;
7913 } else if (block_group) {
7914 btrfs_put_block_group(block_group);
7918 ffe_ctl.have_caching_bg = false;
7919 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
7922 down_read(&space_info->groups_sem);
7923 list_for_each_entry(block_group,
7924 &space_info->block_groups[ffe_ctl.index], list) {
7925 /* If the block group is read-only, we can skip it entirely. */
7926 if (unlikely(block_group->ro))
7929 btrfs_grab_block_group(block_group, delalloc);
7930 ffe_ctl.search_start = block_group->key.objectid;
7933 * this can happen if we end up cycling through all the
7934 * raid types, but we want to make sure we only allocate
7935 * for the proper type.
7937 if (!block_group_bits(block_group, flags)) {
7938 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7939 BTRFS_BLOCK_GROUP_RAID1 |
7940 BTRFS_BLOCK_GROUP_RAID5 |
7941 BTRFS_BLOCK_GROUP_RAID6 |
7942 BTRFS_BLOCK_GROUP_RAID10;
7945 * if they asked for extra copies and this block group
7946 * doesn't provide them, bail. This does allow us to
7947 * fill raid0 from raid1.
7949 if ((flags & extra) && !(block_group->flags & extra))
7954 ffe_ctl.cached = block_group_cache_done(block_group);
7955 if (unlikely(!ffe_ctl.cached)) {
7956 ffe_ctl.have_caching_bg = true;
7957 ret = cache_block_group(block_group, 0);
7962 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7966 * Ok we want to try and use the cluster allocator, so
7969 if (last_ptr && use_cluster) {
7970 struct btrfs_block_group_cache *cluster_bg = NULL;
7972 ret = find_free_extent_clustered(block_group, last_ptr,
7973 &ffe_ctl, &cluster_bg);
7976 if (cluster_bg && cluster_bg != block_group) {
7977 btrfs_release_block_group(block_group,
7979 block_group = cluster_bg;
7982 } else if (ret == -EAGAIN) {
7983 goto have_block_group;
7984 } else if (ret > 0) {
7987 /* ret == -ENOENT case falls through */
7990 ret = find_free_extent_unclustered(block_group, last_ptr,
7993 goto have_block_group;
7996 /* ret == 0 case falls through */
7998 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
7999 fs_info->stripesize);
8001 /* move on to the next group */
8002 if (ffe_ctl.search_start + num_bytes >
8003 block_group->key.objectid + block_group->key.offset) {
8004 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8009 if (ffe_ctl.found_offset < ffe_ctl.search_start)
8010 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8011 ffe_ctl.search_start - ffe_ctl.found_offset);
8013 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
8014 num_bytes, delalloc);
8015 if (ret == -EAGAIN) {
8016 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8020 btrfs_inc_block_group_reservations(block_group);
8022 /* we are all good, lets return */
8023 ins->objectid = ffe_ctl.search_start;
8024 ins->offset = num_bytes;
8026 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
8028 btrfs_release_block_group(block_group, delalloc);
8031 ffe_ctl.retry_clustered = false;
8032 ffe_ctl.retry_unclustered = false;
8033 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
8035 btrfs_release_block_group(block_group, delalloc);
8038 up_read(&space_info->groups_sem);
8040 ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
8041 full_search, use_cluster);
8045 if (ret == -ENOSPC) {
8047 * Use ffe_ctl->total_free_space as fallback if we can't find
8048 * any contiguous hole.
8050 if (!ffe_ctl.max_extent_size)
8051 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
8052 spin_lock(&space_info->lock);
8053 space_info->max_extent_size = ffe_ctl.max_extent_size;
8054 spin_unlock(&space_info->lock);
8055 ins->offset = ffe_ctl.max_extent_size;
8060 static void dump_space_info(struct btrfs_fs_info *fs_info,
8061 struct btrfs_space_info *info, u64 bytes,
8062 int dump_block_groups)
8064 struct btrfs_block_group_cache *cache;
8067 spin_lock(&info->lock);
8068 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
8070 info->total_bytes - btrfs_space_info_used(info, true),
8071 info->full ? "" : "not ");
8073 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
8074 info->total_bytes, info->bytes_used, info->bytes_pinned,
8075 info->bytes_reserved, info->bytes_may_use,
8076 info->bytes_readonly);
8077 spin_unlock(&info->lock);
8079 if (!dump_block_groups)
8082 down_read(&info->groups_sem);
8084 list_for_each_entry(cache, &info->block_groups[index], list) {
8085 spin_lock(&cache->lock);
8087 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8088 cache->key.objectid, cache->key.offset,
8089 btrfs_block_group_used(&cache->item), cache->pinned,
8090 cache->reserved, cache->ro ? "[readonly]" : "");
8091 btrfs_dump_free_space(cache, bytes);
8092 spin_unlock(&cache->lock);
8094 if (++index < BTRFS_NR_RAID_TYPES)
8096 up_read(&info->groups_sem);
8100 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
8101 * hole that is at least as big as @num_bytes.
8103 * @root - The root that will contain this extent
8105 * @ram_bytes - The amount of space in ram that @num_bytes take. This
8106 * is used for accounting purposes. This value differs
8107 * from @num_bytes only in the case of compressed extents.
8109 * @num_bytes - Number of bytes to allocate on-disk.
8111 * @min_alloc_size - Indicates the minimum amount of space that the
8112 * allocator should try to satisfy. In some cases
8113 * @num_bytes may be larger than what is required and if
8114 * the filesystem is fragmented then allocation fails.
8115 * However, the presence of @min_alloc_size gives a
8116 * chance to try and satisfy the smaller allocation.
8118 * @empty_size - A hint that you plan on doing more COW. This is the
8119 * size in bytes the allocator should try to find free
8120 * next to the block it returns. This is just a hint and
8121 * may be ignored by the allocator.
8123 * @hint_byte - Hint to the allocator to start searching above the byte
8124 * address passed. It might be ignored.
8126 * @ins - This key is modified to record the found hole. It will
8127 * have the following values:
8128 * ins->objectid == start position
8129 * ins->flags = BTRFS_EXTENT_ITEM_KEY
8130 * ins->offset == the size of the hole.
8132 * @is_data - Boolean flag indicating whether an extent is
8133 * allocated for data (true) or metadata (false)
8135 * @delalloc - Boolean flag indicating whether this allocation is for
8136 * delalloc or not. If 'true' data_rwsem of block groups
8137 * is going to be acquired.
8140 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8141 * case -ENOSPC is returned then @ins->offset will contain the size of the
8142 * largest available hole the allocator managed to find.
8144 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8145 u64 num_bytes, u64 min_alloc_size,
8146 u64 empty_size, u64 hint_byte,
8147 struct btrfs_key *ins, int is_data, int delalloc)
8149 struct btrfs_fs_info *fs_info = root->fs_info;
8150 bool final_tried = num_bytes == min_alloc_size;
8154 flags = get_alloc_profile_by_root(root, is_data);
8156 WARN_ON(num_bytes < fs_info->sectorsize);
8157 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8158 hint_byte, ins, flags, delalloc);
8159 if (!ret && !is_data) {
8160 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8161 } else if (ret == -ENOSPC) {
8162 if (!final_tried && ins->offset) {
8163 num_bytes = min(num_bytes >> 1, ins->offset);
8164 num_bytes = round_down(num_bytes,
8165 fs_info->sectorsize);
8166 num_bytes = max(num_bytes, min_alloc_size);
8167 ram_bytes = num_bytes;
8168 if (num_bytes == min_alloc_size)
8171 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8172 struct btrfs_space_info *sinfo;
8174 sinfo = __find_space_info(fs_info, flags);
8176 "allocation failed flags %llu, wanted %llu",
8179 dump_space_info(fs_info, sinfo, num_bytes, 1);
8186 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8188 int pin, int delalloc)
8190 struct btrfs_block_group_cache *cache;
8193 cache = btrfs_lookup_block_group(fs_info, start);
8195 btrfs_err(fs_info, "Unable to find block group for %llu",
8201 pin_down_extent(fs_info, cache, start, len, 1);
8203 if (btrfs_test_opt(fs_info, DISCARD))
8204 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8205 btrfs_add_free_space(cache, start, len);
8206 btrfs_free_reserved_bytes(cache, len, delalloc);
8207 trace_btrfs_reserved_extent_free(fs_info, start, len);
8210 btrfs_put_block_group(cache);
8214 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8215 u64 start, u64 len, int delalloc)
8217 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8220 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8223 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8226 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8227 u64 parent, u64 root_objectid,
8228 u64 flags, u64 owner, u64 offset,
8229 struct btrfs_key *ins, int ref_mod)
8231 struct btrfs_fs_info *fs_info = trans->fs_info;
8233 struct btrfs_extent_item *extent_item;
8234 struct btrfs_extent_inline_ref *iref;
8235 struct btrfs_path *path;
8236 struct extent_buffer *leaf;
8241 type = BTRFS_SHARED_DATA_REF_KEY;
8243 type = BTRFS_EXTENT_DATA_REF_KEY;
8245 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8247 path = btrfs_alloc_path();
8251 path->leave_spinning = 1;
8252 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8255 btrfs_free_path(path);
8259 leaf = path->nodes[0];
8260 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8261 struct btrfs_extent_item);
8262 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8263 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8264 btrfs_set_extent_flags(leaf, extent_item,
8265 flags | BTRFS_EXTENT_FLAG_DATA);
8267 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8268 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8270 struct btrfs_shared_data_ref *ref;
8271 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8272 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8273 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8275 struct btrfs_extent_data_ref *ref;
8276 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8277 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8278 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8279 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8280 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8283 btrfs_mark_buffer_dirty(path->nodes[0]);
8284 btrfs_free_path(path);
8286 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8290 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8291 if (ret) { /* -ENOENT, logic error */
8292 btrfs_err(fs_info, "update block group failed for %llu %llu",
8293 ins->objectid, ins->offset);
8296 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8300 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8301 struct btrfs_delayed_ref_node *node,
8302 struct btrfs_delayed_extent_op *extent_op)
8304 struct btrfs_fs_info *fs_info = trans->fs_info;
8306 struct btrfs_extent_item *extent_item;
8307 struct btrfs_key extent_key;
8308 struct btrfs_tree_block_info *block_info;
8309 struct btrfs_extent_inline_ref *iref;
8310 struct btrfs_path *path;
8311 struct extent_buffer *leaf;
8312 struct btrfs_delayed_tree_ref *ref;
8313 u32 size = sizeof(*extent_item) + sizeof(*iref);
8315 u64 flags = extent_op->flags_to_set;
8316 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8318 ref = btrfs_delayed_node_to_tree_ref(node);
8320 extent_key.objectid = node->bytenr;
8321 if (skinny_metadata) {
8322 extent_key.offset = ref->level;
8323 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8324 num_bytes = fs_info->nodesize;
8326 extent_key.offset = node->num_bytes;
8327 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8328 size += sizeof(*block_info);
8329 num_bytes = node->num_bytes;
8332 path = btrfs_alloc_path();
8336 path->leave_spinning = 1;
8337 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8340 btrfs_free_path(path);
8344 leaf = path->nodes[0];
8345 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8346 struct btrfs_extent_item);
8347 btrfs_set_extent_refs(leaf, extent_item, 1);
8348 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8349 btrfs_set_extent_flags(leaf, extent_item,
8350 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8352 if (skinny_metadata) {
8353 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8355 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8356 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8357 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8358 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8361 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8362 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8363 btrfs_set_extent_inline_ref_type(leaf, iref,
8364 BTRFS_SHARED_BLOCK_REF_KEY);
8365 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8367 btrfs_set_extent_inline_ref_type(leaf, iref,
8368 BTRFS_TREE_BLOCK_REF_KEY);
8369 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8372 btrfs_mark_buffer_dirty(leaf);
8373 btrfs_free_path(path);
8375 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8380 ret = update_block_group(trans, fs_info, extent_key.objectid,
8381 fs_info->nodesize, 1);
8382 if (ret) { /* -ENOENT, logic error */
8383 btrfs_err(fs_info, "update block group failed for %llu %llu",
8384 extent_key.objectid, extent_key.offset);
8388 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8393 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8394 struct btrfs_root *root, u64 owner,
8395 u64 offset, u64 ram_bytes,
8396 struct btrfs_key *ins)
8400 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8402 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8403 root->root_key.objectid, owner, offset,
8404 BTRFS_ADD_DELAYED_EXTENT);
8406 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8408 root->root_key.objectid, owner,
8410 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8415 * this is used by the tree logging recovery code. It records that
8416 * an extent has been allocated and makes sure to clear the free
8417 * space cache bits as well
8419 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8420 u64 root_objectid, u64 owner, u64 offset,
8421 struct btrfs_key *ins)
8423 struct btrfs_fs_info *fs_info = trans->fs_info;
8425 struct btrfs_block_group_cache *block_group;
8426 struct btrfs_space_info *space_info;
8429 * Mixed block groups will exclude before processing the log so we only
8430 * need to do the exclude dance if this fs isn't mixed.
8432 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8433 ret = __exclude_logged_extent(fs_info, ins->objectid,
8439 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8443 space_info = block_group->space_info;
8444 spin_lock(&space_info->lock);
8445 spin_lock(&block_group->lock);
8446 space_info->bytes_reserved += ins->offset;
8447 block_group->reserved += ins->offset;
8448 spin_unlock(&block_group->lock);
8449 spin_unlock(&space_info->lock);
8451 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8453 btrfs_put_block_group(block_group);
8457 static struct extent_buffer *
8458 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8459 u64 bytenr, int level, u64 owner)
8461 struct btrfs_fs_info *fs_info = root->fs_info;
8462 struct extent_buffer *buf;
8464 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8469 * Extra safety check in case the extent tree is corrupted and extent
8470 * allocator chooses to use a tree block which is already used and
8473 if (buf->lock_owner == current->pid) {
8474 btrfs_err_rl(fs_info,
8475 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8476 buf->start, btrfs_header_owner(buf), current->pid);
8477 free_extent_buffer(buf);
8478 return ERR_PTR(-EUCLEAN);
8481 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8482 btrfs_tree_lock(buf);
8483 clean_tree_block(fs_info, buf);
8484 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8486 btrfs_set_lock_blocking(buf);
8487 set_extent_buffer_uptodate(buf);
8489 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8490 btrfs_set_header_level(buf, level);
8491 btrfs_set_header_bytenr(buf, buf->start);
8492 btrfs_set_header_generation(buf, trans->transid);
8493 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8494 btrfs_set_header_owner(buf, owner);
8495 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
8496 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8497 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8498 buf->log_index = root->log_transid % 2;
8500 * we allow two log transactions at a time, use different
8501 * EXENT bit to differentiate dirty pages.
8503 if (buf->log_index == 0)
8504 set_extent_dirty(&root->dirty_log_pages, buf->start,
8505 buf->start + buf->len - 1, GFP_NOFS);
8507 set_extent_new(&root->dirty_log_pages, buf->start,
8508 buf->start + buf->len - 1);
8510 buf->log_index = -1;
8511 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8512 buf->start + buf->len - 1, GFP_NOFS);
8514 trans->dirty = true;
8515 /* this returns a buffer locked for blocking */
8519 static struct btrfs_block_rsv *
8520 use_block_rsv(struct btrfs_trans_handle *trans,
8521 struct btrfs_root *root, u32 blocksize)
8523 struct btrfs_fs_info *fs_info = root->fs_info;
8524 struct btrfs_block_rsv *block_rsv;
8525 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8527 bool global_updated = false;
8529 block_rsv = get_block_rsv(trans, root);
8531 if (unlikely(block_rsv->size == 0))
8534 ret = block_rsv_use_bytes(block_rsv, blocksize);
8538 if (block_rsv->failfast)
8539 return ERR_PTR(ret);
8541 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8542 global_updated = true;
8543 update_global_block_rsv(fs_info);
8548 * The global reserve still exists to save us from ourselves, so don't
8549 * warn_on if we are short on our delayed refs reserve.
8551 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
8552 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8553 static DEFINE_RATELIMIT_STATE(_rs,
8554 DEFAULT_RATELIMIT_INTERVAL * 10,
8555 /*DEFAULT_RATELIMIT_BURST*/ 1);
8556 if (__ratelimit(&_rs))
8558 "BTRFS: block rsv returned %d\n", ret);
8561 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8562 BTRFS_RESERVE_NO_FLUSH);
8566 * If we couldn't reserve metadata bytes try and use some from
8567 * the global reserve if its space type is the same as the global
8570 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8571 block_rsv->space_info == global_rsv->space_info) {
8572 ret = block_rsv_use_bytes(global_rsv, blocksize);
8576 return ERR_PTR(ret);
8579 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8580 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8582 block_rsv_add_bytes(block_rsv, blocksize, false);
8583 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8587 * finds a free extent and does all the dirty work required for allocation
8588 * returns the tree buffer or an ERR_PTR on error.
8590 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8591 struct btrfs_root *root,
8592 u64 parent, u64 root_objectid,
8593 const struct btrfs_disk_key *key,
8594 int level, u64 hint,
8597 struct btrfs_fs_info *fs_info = root->fs_info;
8598 struct btrfs_key ins;
8599 struct btrfs_block_rsv *block_rsv;
8600 struct extent_buffer *buf;
8601 struct btrfs_delayed_extent_op *extent_op;
8604 u32 blocksize = fs_info->nodesize;
8605 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8607 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8608 if (btrfs_is_testing(fs_info)) {
8609 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8610 level, root_objectid);
8612 root->alloc_bytenr += blocksize;
8617 block_rsv = use_block_rsv(trans, root, blocksize);
8618 if (IS_ERR(block_rsv))
8619 return ERR_CAST(block_rsv);
8621 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8622 empty_size, hint, &ins, 0, 0);
8626 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8630 goto out_free_reserved;
8633 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8635 parent = ins.objectid;
8636 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8640 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8641 extent_op = btrfs_alloc_delayed_extent_op();
8647 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8649 memset(&extent_op->key, 0, sizeof(extent_op->key));
8650 extent_op->flags_to_set = flags;
8651 extent_op->update_key = skinny_metadata ? false : true;
8652 extent_op->update_flags = true;
8653 extent_op->is_data = false;
8654 extent_op->level = level;
8656 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8657 root_objectid, level, 0,
8658 BTRFS_ADD_DELAYED_EXTENT);
8659 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8661 root_objectid, level,
8662 BTRFS_ADD_DELAYED_EXTENT,
8663 extent_op, NULL, NULL);
8665 goto out_free_delayed;
8670 btrfs_free_delayed_extent_op(extent_op);
8672 free_extent_buffer(buf);
8674 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8676 unuse_block_rsv(fs_info, block_rsv, blocksize);
8677 return ERR_PTR(ret);
8680 struct walk_control {
8681 u64 refs[BTRFS_MAX_LEVEL];
8682 u64 flags[BTRFS_MAX_LEVEL];
8683 struct btrfs_key update_progress;
8693 #define DROP_REFERENCE 1
8694 #define UPDATE_BACKREF 2
8696 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8697 struct btrfs_root *root,
8698 struct walk_control *wc,
8699 struct btrfs_path *path)
8701 struct btrfs_fs_info *fs_info = root->fs_info;
8707 struct btrfs_key key;
8708 struct extent_buffer *eb;
8713 if (path->slots[wc->level] < wc->reada_slot) {
8714 wc->reada_count = wc->reada_count * 2 / 3;
8715 wc->reada_count = max(wc->reada_count, 2);
8717 wc->reada_count = wc->reada_count * 3 / 2;
8718 wc->reada_count = min_t(int, wc->reada_count,
8719 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8722 eb = path->nodes[wc->level];
8723 nritems = btrfs_header_nritems(eb);
8725 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8726 if (nread >= wc->reada_count)
8730 bytenr = btrfs_node_blockptr(eb, slot);
8731 generation = btrfs_node_ptr_generation(eb, slot);
8733 if (slot == path->slots[wc->level])
8736 if (wc->stage == UPDATE_BACKREF &&
8737 generation <= root->root_key.offset)
8740 /* We don't lock the tree block, it's OK to be racy here */
8741 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8742 wc->level - 1, 1, &refs,
8744 /* We don't care about errors in readahead. */
8749 if (wc->stage == DROP_REFERENCE) {
8753 if (wc->level == 1 &&
8754 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8756 if (!wc->update_ref ||
8757 generation <= root->root_key.offset)
8759 btrfs_node_key_to_cpu(eb, &key, slot);
8760 ret = btrfs_comp_cpu_keys(&key,
8761 &wc->update_progress);
8765 if (wc->level == 1 &&
8766 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8770 readahead_tree_block(fs_info, bytenr);
8773 wc->reada_slot = slot;
8777 * helper to process tree block while walking down the tree.
8779 * when wc->stage == UPDATE_BACKREF, this function updates
8780 * back refs for pointers in the block.
8782 * NOTE: return value 1 means we should stop walking down.
8784 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8785 struct btrfs_root *root,
8786 struct btrfs_path *path,
8787 struct walk_control *wc, int lookup_info)
8789 struct btrfs_fs_info *fs_info = root->fs_info;
8790 int level = wc->level;
8791 struct extent_buffer *eb = path->nodes[level];
8792 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8795 if (wc->stage == UPDATE_BACKREF &&
8796 btrfs_header_owner(eb) != root->root_key.objectid)
8800 * when reference count of tree block is 1, it won't increase
8801 * again. once full backref flag is set, we never clear it.
8804 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8805 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8806 BUG_ON(!path->locks[level]);
8807 ret = btrfs_lookup_extent_info(trans, fs_info,
8808 eb->start, level, 1,
8811 BUG_ON(ret == -ENOMEM);
8814 BUG_ON(wc->refs[level] == 0);
8817 if (wc->stage == DROP_REFERENCE) {
8818 if (wc->refs[level] > 1)
8821 if (path->locks[level] && !wc->keep_locks) {
8822 btrfs_tree_unlock_rw(eb, path->locks[level]);
8823 path->locks[level] = 0;
8828 /* wc->stage == UPDATE_BACKREF */
8829 if (!(wc->flags[level] & flag)) {
8830 BUG_ON(!path->locks[level]);
8831 ret = btrfs_inc_ref(trans, root, eb, 1);
8832 BUG_ON(ret); /* -ENOMEM */
8833 ret = btrfs_dec_ref(trans, root, eb, 0);
8834 BUG_ON(ret); /* -ENOMEM */
8835 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8837 btrfs_header_level(eb), 0);
8838 BUG_ON(ret); /* -ENOMEM */
8839 wc->flags[level] |= flag;
8843 * the block is shared by multiple trees, so it's not good to
8844 * keep the tree lock
8846 if (path->locks[level] && level > 0) {
8847 btrfs_tree_unlock_rw(eb, path->locks[level]);
8848 path->locks[level] = 0;
8854 * helper to process tree block pointer.
8856 * when wc->stage == DROP_REFERENCE, this function checks
8857 * reference count of the block pointed to. if the block
8858 * is shared and we need update back refs for the subtree
8859 * rooted at the block, this function changes wc->stage to
8860 * UPDATE_BACKREF. if the block is shared and there is no
8861 * need to update back, this function drops the reference
8864 * NOTE: return value 1 means we should stop walking down.
8866 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8867 struct btrfs_root *root,
8868 struct btrfs_path *path,
8869 struct walk_control *wc, int *lookup_info)
8871 struct btrfs_fs_info *fs_info = root->fs_info;
8876 struct btrfs_key key;
8877 struct btrfs_key first_key;
8878 struct extent_buffer *next;
8879 int level = wc->level;
8882 bool need_account = false;
8884 generation = btrfs_node_ptr_generation(path->nodes[level],
8885 path->slots[level]);
8887 * if the lower level block was created before the snapshot
8888 * was created, we know there is no need to update back refs
8891 if (wc->stage == UPDATE_BACKREF &&
8892 generation <= root->root_key.offset) {
8897 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8898 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8899 path->slots[level]);
8900 blocksize = fs_info->nodesize;
8902 next = find_extent_buffer(fs_info, bytenr);
8904 next = btrfs_find_create_tree_block(fs_info, bytenr);
8906 return PTR_ERR(next);
8908 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8912 btrfs_tree_lock(next);
8913 btrfs_set_lock_blocking(next);
8915 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8916 &wc->refs[level - 1],
8917 &wc->flags[level - 1]);
8921 if (unlikely(wc->refs[level - 1] == 0)) {
8922 btrfs_err(fs_info, "Missing references.");
8928 if (wc->stage == DROP_REFERENCE) {
8929 if (wc->refs[level - 1] > 1) {
8930 need_account = true;
8932 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8935 if (!wc->update_ref ||
8936 generation <= root->root_key.offset)
8939 btrfs_node_key_to_cpu(path->nodes[level], &key,
8940 path->slots[level]);
8941 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8945 wc->stage = UPDATE_BACKREF;
8946 wc->shared_level = level - 1;
8950 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8954 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8955 btrfs_tree_unlock(next);
8956 free_extent_buffer(next);
8962 if (reada && level == 1)
8963 reada_walk_down(trans, root, wc, path);
8964 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8967 return PTR_ERR(next);
8968 } else if (!extent_buffer_uptodate(next)) {
8969 free_extent_buffer(next);
8972 btrfs_tree_lock(next);
8973 btrfs_set_lock_blocking(next);
8977 ASSERT(level == btrfs_header_level(next));
8978 if (level != btrfs_header_level(next)) {
8979 btrfs_err(root->fs_info, "mismatched level");
8983 path->nodes[level] = next;
8984 path->slots[level] = 0;
8985 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8991 wc->refs[level - 1] = 0;
8992 wc->flags[level - 1] = 0;
8993 if (wc->stage == DROP_REFERENCE) {
8994 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8995 parent = path->nodes[level]->start;
8997 ASSERT(root->root_key.objectid ==
8998 btrfs_header_owner(path->nodes[level]));
8999 if (root->root_key.objectid !=
9000 btrfs_header_owner(path->nodes[level])) {
9001 btrfs_err(root->fs_info,
9002 "mismatched block owner");
9010 * Reloc tree doesn't contribute to qgroup numbers, and we have
9011 * already accounted them at merge time (replace_path),
9012 * thus we could skip expensive subtree trace here.
9014 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
9016 ret = btrfs_qgroup_trace_subtree(trans, next,
9017 generation, level - 1);
9019 btrfs_err_rl(fs_info,
9020 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
9024 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
9025 parent, root->root_key.objectid,
9035 btrfs_tree_unlock(next);
9036 free_extent_buffer(next);
9042 * helper to process tree block while walking up the tree.
9044 * when wc->stage == DROP_REFERENCE, this function drops
9045 * reference count on the block.
9047 * when wc->stage == UPDATE_BACKREF, this function changes
9048 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9049 * to UPDATE_BACKREF previously while processing the block.
9051 * NOTE: return value 1 means we should stop walking up.
9053 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9054 struct btrfs_root *root,
9055 struct btrfs_path *path,
9056 struct walk_control *wc)
9058 struct btrfs_fs_info *fs_info = root->fs_info;
9060 int level = wc->level;
9061 struct extent_buffer *eb = path->nodes[level];
9064 if (wc->stage == UPDATE_BACKREF) {
9065 BUG_ON(wc->shared_level < level);
9066 if (level < wc->shared_level)
9069 ret = find_next_key(path, level + 1, &wc->update_progress);
9073 wc->stage = DROP_REFERENCE;
9074 wc->shared_level = -1;
9075 path->slots[level] = 0;
9078 * check reference count again if the block isn't locked.
9079 * we should start walking down the tree again if reference
9082 if (!path->locks[level]) {
9084 btrfs_tree_lock(eb);
9085 btrfs_set_lock_blocking(eb);
9086 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9088 ret = btrfs_lookup_extent_info(trans, fs_info,
9089 eb->start, level, 1,
9093 btrfs_tree_unlock_rw(eb, path->locks[level]);
9094 path->locks[level] = 0;
9097 BUG_ON(wc->refs[level] == 0);
9098 if (wc->refs[level] == 1) {
9099 btrfs_tree_unlock_rw(eb, path->locks[level]);
9100 path->locks[level] = 0;
9106 /* wc->stage == DROP_REFERENCE */
9107 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9109 if (wc->refs[level] == 1) {
9111 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9112 ret = btrfs_dec_ref(trans, root, eb, 1);
9114 ret = btrfs_dec_ref(trans, root, eb, 0);
9115 BUG_ON(ret); /* -ENOMEM */
9116 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
9118 btrfs_err_rl(fs_info,
9119 "error %d accounting leaf items. Quota is out of sync, rescan required.",
9123 /* make block locked assertion in clean_tree_block happy */
9124 if (!path->locks[level] &&
9125 btrfs_header_generation(eb) == trans->transid) {
9126 btrfs_tree_lock(eb);
9127 btrfs_set_lock_blocking(eb);
9128 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9130 clean_tree_block(fs_info, eb);
9133 if (eb == root->node) {
9134 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9136 else if (root->root_key.objectid != btrfs_header_owner(eb))
9137 goto owner_mismatch;
9139 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9140 parent = path->nodes[level + 1]->start;
9141 else if (root->root_key.objectid !=
9142 btrfs_header_owner(path->nodes[level + 1]))
9143 goto owner_mismatch;
9146 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9148 wc->refs[level] = 0;
9149 wc->flags[level] = 0;
9153 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
9154 btrfs_header_owner(eb), root->root_key.objectid);
9158 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9159 struct btrfs_root *root,
9160 struct btrfs_path *path,
9161 struct walk_control *wc)
9163 int level = wc->level;
9164 int lookup_info = 1;
9167 while (level >= 0) {
9168 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9175 if (path->slots[level] >=
9176 btrfs_header_nritems(path->nodes[level]))
9179 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9181 path->slots[level]++;
9190 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9191 struct btrfs_root *root,
9192 struct btrfs_path *path,
9193 struct walk_control *wc, int max_level)
9195 int level = wc->level;
9198 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9199 while (level < max_level && path->nodes[level]) {
9201 if (path->slots[level] + 1 <
9202 btrfs_header_nritems(path->nodes[level])) {
9203 path->slots[level]++;
9206 ret = walk_up_proc(trans, root, path, wc);
9212 if (path->locks[level]) {
9213 btrfs_tree_unlock_rw(path->nodes[level],
9214 path->locks[level]);
9215 path->locks[level] = 0;
9217 free_extent_buffer(path->nodes[level]);
9218 path->nodes[level] = NULL;
9226 * drop a subvolume tree.
9228 * this function traverses the tree freeing any blocks that only
9229 * referenced by the tree.
9231 * when a shared tree block is found. this function decreases its
9232 * reference count by one. if update_ref is true, this function
9233 * also make sure backrefs for the shared block and all lower level
9234 * blocks are properly updated.
9236 * If called with for_reloc == 0, may exit early with -EAGAIN
9238 int btrfs_drop_snapshot(struct btrfs_root *root,
9239 struct btrfs_block_rsv *block_rsv, int update_ref,
9242 struct btrfs_fs_info *fs_info = root->fs_info;
9243 struct btrfs_path *path;
9244 struct btrfs_trans_handle *trans;
9245 struct btrfs_root *tree_root = fs_info->tree_root;
9246 struct btrfs_root_item *root_item = &root->root_item;
9247 struct walk_control *wc;
9248 struct btrfs_key key;
9252 bool root_dropped = false;
9254 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
9256 path = btrfs_alloc_path();
9262 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9264 btrfs_free_path(path);
9269 trans = btrfs_start_transaction(tree_root, 0);
9270 if (IS_ERR(trans)) {
9271 err = PTR_ERR(trans);
9276 trans->block_rsv = block_rsv;
9278 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9279 level = btrfs_header_level(root->node);
9280 path->nodes[level] = btrfs_lock_root_node(root);
9281 btrfs_set_lock_blocking(path->nodes[level]);
9282 path->slots[level] = 0;
9283 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9284 memset(&wc->update_progress, 0,
9285 sizeof(wc->update_progress));
9287 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9288 memcpy(&wc->update_progress, &key,
9289 sizeof(wc->update_progress));
9291 level = root_item->drop_level;
9293 path->lowest_level = level;
9294 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9295 path->lowest_level = 0;
9303 * unlock our path, this is safe because only this
9304 * function is allowed to delete this snapshot
9306 btrfs_unlock_up_safe(path, 0);
9308 level = btrfs_header_level(root->node);
9310 btrfs_tree_lock(path->nodes[level]);
9311 btrfs_set_lock_blocking(path->nodes[level]);
9312 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9314 ret = btrfs_lookup_extent_info(trans, fs_info,
9315 path->nodes[level]->start,
9316 level, 1, &wc->refs[level],
9322 BUG_ON(wc->refs[level] == 0);
9324 if (level == root_item->drop_level)
9327 btrfs_tree_unlock(path->nodes[level]);
9328 path->locks[level] = 0;
9329 WARN_ON(wc->refs[level] != 1);
9335 wc->shared_level = -1;
9336 wc->stage = DROP_REFERENCE;
9337 wc->update_ref = update_ref;
9339 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9343 ret = walk_down_tree(trans, root, path, wc);
9349 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9356 BUG_ON(wc->stage != DROP_REFERENCE);
9360 if (wc->stage == DROP_REFERENCE) {
9362 btrfs_node_key(path->nodes[level],
9363 &root_item->drop_progress,
9364 path->slots[level]);
9365 root_item->drop_level = level;
9368 BUG_ON(wc->level == 0);
9369 if (btrfs_should_end_transaction(trans) ||
9370 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9371 ret = btrfs_update_root(trans, tree_root,
9375 btrfs_abort_transaction(trans, ret);
9380 btrfs_end_transaction_throttle(trans);
9381 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9382 btrfs_debug(fs_info,
9383 "drop snapshot early exit");
9388 trans = btrfs_start_transaction(tree_root, 0);
9389 if (IS_ERR(trans)) {
9390 err = PTR_ERR(trans);
9394 trans->block_rsv = block_rsv;
9397 btrfs_release_path(path);
9401 ret = btrfs_del_root(trans, &root->root_key);
9403 btrfs_abort_transaction(trans, ret);
9408 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9409 ret = btrfs_find_root(tree_root, &root->root_key, path,
9412 btrfs_abort_transaction(trans, ret);
9415 } else if (ret > 0) {
9416 /* if we fail to delete the orphan item this time
9417 * around, it'll get picked up the next time.
9419 * The most common failure here is just -ENOENT.
9421 btrfs_del_orphan_item(trans, tree_root,
9422 root->root_key.objectid);
9426 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9427 btrfs_add_dropped_root(trans, root);
9429 free_extent_buffer(root->node);
9430 free_extent_buffer(root->commit_root);
9431 btrfs_put_fs_root(root);
9433 root_dropped = true;
9435 btrfs_end_transaction_throttle(trans);
9438 btrfs_free_path(path);
9441 * So if we need to stop dropping the snapshot for whatever reason we
9442 * need to make sure to add it back to the dead root list so that we
9443 * keep trying to do the work later. This also cleans up roots if we
9444 * don't have it in the radix (like when we recover after a power fail
9445 * or unmount) so we don't leak memory.
9447 if (!for_reloc && !root_dropped)
9448 btrfs_add_dead_root(root);
9449 if (err && err != -EAGAIN)
9450 btrfs_handle_fs_error(fs_info, err, NULL);
9455 * drop subtree rooted at tree block 'node'.
9457 * NOTE: this function will unlock and release tree block 'node'
9458 * only used by relocation code
9460 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9461 struct btrfs_root *root,
9462 struct extent_buffer *node,
9463 struct extent_buffer *parent)
9465 struct btrfs_fs_info *fs_info = root->fs_info;
9466 struct btrfs_path *path;
9467 struct walk_control *wc;
9473 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9475 path = btrfs_alloc_path();
9479 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9481 btrfs_free_path(path);
9485 btrfs_assert_tree_locked(parent);
9486 parent_level = btrfs_header_level(parent);
9487 extent_buffer_get(parent);
9488 path->nodes[parent_level] = parent;
9489 path->slots[parent_level] = btrfs_header_nritems(parent);
9491 btrfs_assert_tree_locked(node);
9492 level = btrfs_header_level(node);
9493 path->nodes[level] = node;
9494 path->slots[level] = 0;
9495 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9497 wc->refs[parent_level] = 1;
9498 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9500 wc->shared_level = -1;
9501 wc->stage = DROP_REFERENCE;
9504 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9507 wret = walk_down_tree(trans, root, path, wc);
9513 wret = walk_up_tree(trans, root, path, wc, parent_level);
9521 btrfs_free_path(path);
9525 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9531 * if restripe for this chunk_type is on pick target profile and
9532 * return, otherwise do the usual balance
9534 stripped = get_restripe_target(fs_info, flags);
9536 return extended_to_chunk(stripped);
9538 num_devices = fs_info->fs_devices->rw_devices;
9540 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9541 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9542 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9544 if (num_devices == 1) {
9545 stripped |= BTRFS_BLOCK_GROUP_DUP;
9546 stripped = flags & ~stripped;
9548 /* turn raid0 into single device chunks */
9549 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9552 /* turn mirroring into duplication */
9553 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9554 BTRFS_BLOCK_GROUP_RAID10))
9555 return stripped | BTRFS_BLOCK_GROUP_DUP;
9557 /* they already had raid on here, just return */
9558 if (flags & stripped)
9561 stripped |= BTRFS_BLOCK_GROUP_DUP;
9562 stripped = flags & ~stripped;
9564 /* switch duplicated blocks with raid1 */
9565 if (flags & BTRFS_BLOCK_GROUP_DUP)
9566 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9568 /* this is drive concat, leave it alone */
9574 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9576 struct btrfs_space_info *sinfo = cache->space_info;
9578 u64 min_allocable_bytes;
9582 * We need some metadata space and system metadata space for
9583 * allocating chunks in some corner cases until we force to set
9584 * it to be readonly.
9587 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9589 min_allocable_bytes = SZ_1M;
9591 min_allocable_bytes = 0;
9593 spin_lock(&sinfo->lock);
9594 spin_lock(&cache->lock);
9602 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9603 cache->bytes_super - btrfs_block_group_used(&cache->item);
9605 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9606 min_allocable_bytes <= sinfo->total_bytes) {
9607 sinfo->bytes_readonly += num_bytes;
9609 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9613 spin_unlock(&cache->lock);
9614 spin_unlock(&sinfo->lock);
9618 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9621 struct btrfs_fs_info *fs_info = cache->fs_info;
9622 struct btrfs_trans_handle *trans;
9627 trans = btrfs_join_transaction(fs_info->extent_root);
9629 return PTR_ERR(trans);
9632 * we're not allowed to set block groups readonly after the dirty
9633 * block groups cache has started writing. If it already started,
9634 * back off and let this transaction commit
9636 mutex_lock(&fs_info->ro_block_group_mutex);
9637 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9638 u64 transid = trans->transid;
9640 mutex_unlock(&fs_info->ro_block_group_mutex);
9641 btrfs_end_transaction(trans);
9643 ret = btrfs_wait_for_commit(fs_info, transid);
9650 * if we are changing raid levels, try to allocate a corresponding
9651 * block group with the new raid level.
9653 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9654 if (alloc_flags != cache->flags) {
9655 ret = do_chunk_alloc(trans, alloc_flags,
9658 * ENOSPC is allowed here, we may have enough space
9659 * already allocated at the new raid level to
9668 ret = inc_block_group_ro(cache, 0);
9671 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9672 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9675 ret = inc_block_group_ro(cache, 0);
9677 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9678 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9679 mutex_lock(&fs_info->chunk_mutex);
9680 check_system_chunk(trans, alloc_flags);
9681 mutex_unlock(&fs_info->chunk_mutex);
9683 mutex_unlock(&fs_info->ro_block_group_mutex);
9685 btrfs_end_transaction(trans);
9689 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9691 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9693 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9697 * helper to account the unused space of all the readonly block group in the
9698 * space_info. takes mirrors into account.
9700 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9702 struct btrfs_block_group_cache *block_group;
9706 /* It's df, we don't care if it's racy */
9707 if (list_empty(&sinfo->ro_bgs))
9710 spin_lock(&sinfo->lock);
9711 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9712 spin_lock(&block_group->lock);
9714 if (!block_group->ro) {
9715 spin_unlock(&block_group->lock);
9719 factor = btrfs_bg_type_to_factor(block_group->flags);
9720 free_bytes += (block_group->key.offset -
9721 btrfs_block_group_used(&block_group->item)) *
9724 spin_unlock(&block_group->lock);
9726 spin_unlock(&sinfo->lock);
9731 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9733 struct btrfs_space_info *sinfo = cache->space_info;
9738 spin_lock(&sinfo->lock);
9739 spin_lock(&cache->lock);
9741 num_bytes = cache->key.offset - cache->reserved -
9742 cache->pinned - cache->bytes_super -
9743 btrfs_block_group_used(&cache->item);
9744 sinfo->bytes_readonly -= num_bytes;
9745 list_del_init(&cache->ro_list);
9747 spin_unlock(&cache->lock);
9748 spin_unlock(&sinfo->lock);
9752 * checks to see if its even possible to relocate this block group.
9754 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9755 * ok to go ahead and try.
9757 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9759 struct btrfs_root *root = fs_info->extent_root;
9760 struct btrfs_block_group_cache *block_group;
9761 struct btrfs_space_info *space_info;
9762 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9763 struct btrfs_device *device;
9764 struct btrfs_trans_handle *trans;
9774 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9776 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9778 /* odd, couldn't find the block group, leave it alone */
9782 "can't find block group for bytenr %llu",
9787 min_free = btrfs_block_group_used(&block_group->item);
9789 /* no bytes used, we're good */
9793 space_info = block_group->space_info;
9794 spin_lock(&space_info->lock);
9796 full = space_info->full;
9799 * if this is the last block group we have in this space, we can't
9800 * relocate it unless we're able to allocate a new chunk below.
9802 * Otherwise, we need to make sure we have room in the space to handle
9803 * all of the extents from this block group. If we can, we're good
9805 if ((space_info->total_bytes != block_group->key.offset) &&
9806 (btrfs_space_info_used(space_info, false) + min_free <
9807 space_info->total_bytes)) {
9808 spin_unlock(&space_info->lock);
9811 spin_unlock(&space_info->lock);
9814 * ok we don't have enough space, but maybe we have free space on our
9815 * devices to allocate new chunks for relocation, so loop through our
9816 * alloc devices and guess if we have enough space. if this block
9817 * group is going to be restriped, run checks against the target
9818 * profile instead of the current one.
9830 target = get_restripe_target(fs_info, block_group->flags);
9832 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9835 * this is just a balance, so if we were marked as full
9836 * we know there is no space for a new chunk
9841 "no space to alloc new chunk for block group %llu",
9842 block_group->key.objectid);
9846 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9849 if (index == BTRFS_RAID_RAID10) {
9853 } else if (index == BTRFS_RAID_RAID1) {
9855 } else if (index == BTRFS_RAID_DUP) {
9858 } else if (index == BTRFS_RAID_RAID0) {
9859 dev_min = fs_devices->rw_devices;
9860 min_free = div64_u64(min_free, dev_min);
9863 /* We need to do this so that we can look at pending chunks */
9864 trans = btrfs_join_transaction(root);
9865 if (IS_ERR(trans)) {
9866 ret = PTR_ERR(trans);
9870 mutex_lock(&fs_info->chunk_mutex);
9871 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9875 * check to make sure we can actually find a chunk with enough
9876 * space to fit our block group in.
9878 if (device->total_bytes > device->bytes_used + min_free &&
9879 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9880 ret = find_free_dev_extent(trans, device, min_free,
9885 if (dev_nr >= dev_min)
9891 if (debug && ret == -1)
9893 "no space to allocate a new chunk for block group %llu",
9894 block_group->key.objectid);
9895 mutex_unlock(&fs_info->chunk_mutex);
9896 btrfs_end_transaction(trans);
9898 btrfs_put_block_group(block_group);
9902 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9903 struct btrfs_path *path,
9904 struct btrfs_key *key)
9906 struct btrfs_root *root = fs_info->extent_root;
9908 struct btrfs_key found_key;
9909 struct extent_buffer *leaf;
9910 struct btrfs_block_group_item bg;
9914 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9919 slot = path->slots[0];
9920 leaf = path->nodes[0];
9921 if (slot >= btrfs_header_nritems(leaf)) {
9922 ret = btrfs_next_leaf(root, path);
9929 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9931 if (found_key.objectid >= key->objectid &&
9932 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9933 struct extent_map_tree *em_tree;
9934 struct extent_map *em;
9936 em_tree = &root->fs_info->mapping_tree.map_tree;
9937 read_lock(&em_tree->lock);
9938 em = lookup_extent_mapping(em_tree, found_key.objectid,
9940 read_unlock(&em_tree->lock);
9943 "logical %llu len %llu found bg but no related chunk",
9944 found_key.objectid, found_key.offset);
9946 } else if (em->start != found_key.objectid ||
9947 em->len != found_key.offset) {
9949 "block group %llu len %llu mismatch with chunk %llu len %llu",
9950 found_key.objectid, found_key.offset,
9951 em->start, em->len);
9954 read_extent_buffer(leaf, &bg,
9955 btrfs_item_ptr_offset(leaf, slot),
9957 flags = btrfs_block_group_flags(&bg) &
9958 BTRFS_BLOCK_GROUP_TYPE_MASK;
9960 if (flags != (em->map_lookup->type &
9961 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9963 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9965 found_key.offset, flags,
9966 (BTRFS_BLOCK_GROUP_TYPE_MASK &
9967 em->map_lookup->type));
9973 free_extent_map(em);
9982 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9984 struct btrfs_block_group_cache *block_group;
9988 struct inode *inode;
9990 block_group = btrfs_lookup_first_block_group(info, last);
9991 while (block_group) {
9992 wait_block_group_cache_done(block_group);
9993 spin_lock(&block_group->lock);
9994 if (block_group->iref)
9996 spin_unlock(&block_group->lock);
9997 block_group = next_block_group(info, block_group);
10006 inode = block_group->inode;
10007 block_group->iref = 0;
10008 block_group->inode = NULL;
10009 spin_unlock(&block_group->lock);
10010 ASSERT(block_group->io_ctl.inode == NULL);
10012 last = block_group->key.objectid + block_group->key.offset;
10013 btrfs_put_block_group(block_group);
10018 * Must be called only after stopping all workers, since we could have block
10019 * group caching kthreads running, and therefore they could race with us if we
10020 * freed the block groups before stopping them.
10022 int btrfs_free_block_groups(struct btrfs_fs_info *info)
10024 struct btrfs_block_group_cache *block_group;
10025 struct btrfs_space_info *space_info;
10026 struct btrfs_caching_control *caching_ctl;
10029 down_write(&info->commit_root_sem);
10030 while (!list_empty(&info->caching_block_groups)) {
10031 caching_ctl = list_entry(info->caching_block_groups.next,
10032 struct btrfs_caching_control, list);
10033 list_del(&caching_ctl->list);
10034 put_caching_control(caching_ctl);
10036 up_write(&info->commit_root_sem);
10038 spin_lock(&info->unused_bgs_lock);
10039 while (!list_empty(&info->unused_bgs)) {
10040 block_group = list_first_entry(&info->unused_bgs,
10041 struct btrfs_block_group_cache,
10043 list_del_init(&block_group->bg_list);
10044 btrfs_put_block_group(block_group);
10046 spin_unlock(&info->unused_bgs_lock);
10048 spin_lock(&info->block_group_cache_lock);
10049 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
10050 block_group = rb_entry(n, struct btrfs_block_group_cache,
10052 rb_erase(&block_group->cache_node,
10053 &info->block_group_cache_tree);
10054 RB_CLEAR_NODE(&block_group->cache_node);
10055 spin_unlock(&info->block_group_cache_lock);
10057 down_write(&block_group->space_info->groups_sem);
10058 list_del(&block_group->list);
10059 up_write(&block_group->space_info->groups_sem);
10062 * We haven't cached this block group, which means we could
10063 * possibly have excluded extents on this block group.
10065 if (block_group->cached == BTRFS_CACHE_NO ||
10066 block_group->cached == BTRFS_CACHE_ERROR)
10067 free_excluded_extents(block_group);
10069 btrfs_remove_free_space_cache(block_group);
10070 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
10071 ASSERT(list_empty(&block_group->dirty_list));
10072 ASSERT(list_empty(&block_group->io_list));
10073 ASSERT(list_empty(&block_group->bg_list));
10074 ASSERT(atomic_read(&block_group->count) == 1);
10075 btrfs_put_block_group(block_group);
10077 spin_lock(&info->block_group_cache_lock);
10079 spin_unlock(&info->block_group_cache_lock);
10081 /* now that all the block groups are freed, go through and
10082 * free all the space_info structs. This is only called during
10083 * the final stages of unmount, and so we know nobody is
10084 * using them. We call synchronize_rcu() once before we start,
10085 * just to be on the safe side.
10089 release_global_block_rsv(info);
10091 while (!list_empty(&info->space_info)) {
10094 space_info = list_entry(info->space_info.next,
10095 struct btrfs_space_info,
10099 * Do not hide this behind enospc_debug, this is actually
10100 * important and indicates a real bug if this happens.
10102 if (WARN_ON(space_info->bytes_pinned > 0 ||
10103 space_info->bytes_reserved > 0 ||
10104 space_info->bytes_may_use > 0))
10105 dump_space_info(info, space_info, 0, 0);
10106 list_del(&space_info->list);
10107 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10108 struct kobject *kobj;
10109 kobj = space_info->block_group_kobjs[i];
10110 space_info->block_group_kobjs[i] = NULL;
10116 kobject_del(&space_info->kobj);
10117 kobject_put(&space_info->kobj);
10122 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
10123 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
10125 struct btrfs_space_info *space_info;
10126 struct raid_kobject *rkobj;
10131 spin_lock(&fs_info->pending_raid_kobjs_lock);
10132 list_splice_init(&fs_info->pending_raid_kobjs, &list);
10133 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10135 list_for_each_entry(rkobj, &list, list) {
10136 space_info = __find_space_info(fs_info, rkobj->flags);
10137 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
10139 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10140 "%s", get_raid_name(index));
10142 kobject_put(&rkobj->kobj);
10147 btrfs_warn(fs_info,
10148 "failed to add kobject for block cache, ignoring");
10151 static void link_block_group(struct btrfs_block_group_cache *cache)
10153 struct btrfs_space_info *space_info = cache->space_info;
10154 struct btrfs_fs_info *fs_info = cache->fs_info;
10155 int index = btrfs_bg_flags_to_raid_index(cache->flags);
10156 bool first = false;
10158 down_write(&space_info->groups_sem);
10159 if (list_empty(&space_info->block_groups[index]))
10161 list_add_tail(&cache->list, &space_info->block_groups[index]);
10162 up_write(&space_info->groups_sem);
10165 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10167 btrfs_warn(cache->fs_info,
10168 "couldn't alloc memory for raid level kobject");
10171 rkobj->flags = cache->flags;
10172 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10174 spin_lock(&fs_info->pending_raid_kobjs_lock);
10175 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
10176 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10177 space_info->block_group_kobjs[index] = &rkobj->kobj;
10181 static struct btrfs_block_group_cache *
10182 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10183 u64 start, u64 size)
10185 struct btrfs_block_group_cache *cache;
10187 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10191 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10193 if (!cache->free_space_ctl) {
10198 cache->key.objectid = start;
10199 cache->key.offset = size;
10200 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10202 cache->fs_info = fs_info;
10203 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10204 set_free_space_tree_thresholds(cache);
10206 atomic_set(&cache->count, 1);
10207 spin_lock_init(&cache->lock);
10208 init_rwsem(&cache->data_rwsem);
10209 INIT_LIST_HEAD(&cache->list);
10210 INIT_LIST_HEAD(&cache->cluster_list);
10211 INIT_LIST_HEAD(&cache->bg_list);
10212 INIT_LIST_HEAD(&cache->ro_list);
10213 INIT_LIST_HEAD(&cache->dirty_list);
10214 INIT_LIST_HEAD(&cache->io_list);
10215 btrfs_init_free_space_ctl(cache);
10216 atomic_set(&cache->trimming, 0);
10217 mutex_init(&cache->free_space_lock);
10218 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10225 * Iterate all chunks and verify that each of them has the corresponding block
10228 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10230 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
10231 struct extent_map *em;
10232 struct btrfs_block_group_cache *bg;
10237 read_lock(&map_tree->map_tree.lock);
10239 * lookup_extent_mapping will return the first extent map
10240 * intersecting the range, so setting @len to 1 is enough to
10241 * get the first chunk.
10243 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
10244 read_unlock(&map_tree->map_tree.lock);
10248 bg = btrfs_lookup_block_group(fs_info, em->start);
10251 "chunk start=%llu len=%llu doesn't have corresponding block group",
10252 em->start, em->len);
10254 free_extent_map(em);
10257 if (bg->key.objectid != em->start ||
10258 bg->key.offset != em->len ||
10259 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10260 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10262 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10263 em->start, em->len,
10264 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10265 bg->key.objectid, bg->key.offset,
10266 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10268 free_extent_map(em);
10269 btrfs_put_block_group(bg);
10272 start = em->start + em->len;
10273 free_extent_map(em);
10274 btrfs_put_block_group(bg);
10279 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10281 struct btrfs_path *path;
10283 struct btrfs_block_group_cache *cache;
10284 struct btrfs_space_info *space_info;
10285 struct btrfs_key key;
10286 struct btrfs_key found_key;
10287 struct extent_buffer *leaf;
10288 int need_clear = 0;
10293 feature = btrfs_super_incompat_flags(info->super_copy);
10294 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10298 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10299 path = btrfs_alloc_path();
10302 path->reada = READA_FORWARD;
10304 cache_gen = btrfs_super_cache_generation(info->super_copy);
10305 if (btrfs_test_opt(info, SPACE_CACHE) &&
10306 btrfs_super_generation(info->super_copy) != cache_gen)
10308 if (btrfs_test_opt(info, CLEAR_CACHE))
10312 ret = find_first_block_group(info, path, &key);
10318 leaf = path->nodes[0];
10319 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10321 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10330 * When we mount with old space cache, we need to
10331 * set BTRFS_DC_CLEAR and set dirty flag.
10333 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10334 * truncate the old free space cache inode and
10336 * b) Setting 'dirty flag' makes sure that we flush
10337 * the new space cache info onto disk.
10339 if (btrfs_test_opt(info, SPACE_CACHE))
10340 cache->disk_cache_state = BTRFS_DC_CLEAR;
10343 read_extent_buffer(leaf, &cache->item,
10344 btrfs_item_ptr_offset(leaf, path->slots[0]),
10345 sizeof(cache->item));
10346 cache->flags = btrfs_block_group_flags(&cache->item);
10348 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10349 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10351 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10352 cache->key.objectid);
10357 key.objectid = found_key.objectid + found_key.offset;
10358 btrfs_release_path(path);
10361 * We need to exclude the super stripes now so that the space
10362 * info has super bytes accounted for, otherwise we'll think
10363 * we have more space than we actually do.
10365 ret = exclude_super_stripes(cache);
10368 * We may have excluded something, so call this just in
10371 free_excluded_extents(cache);
10372 btrfs_put_block_group(cache);
10377 * check for two cases, either we are full, and therefore
10378 * don't need to bother with the caching work since we won't
10379 * find any space, or we are empty, and we can just add all
10380 * the space in and be done with it. This saves us _alot_ of
10381 * time, particularly in the full case.
10383 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10384 cache->last_byte_to_unpin = (u64)-1;
10385 cache->cached = BTRFS_CACHE_FINISHED;
10386 free_excluded_extents(cache);
10387 } else if (btrfs_block_group_used(&cache->item) == 0) {
10388 cache->last_byte_to_unpin = (u64)-1;
10389 cache->cached = BTRFS_CACHE_FINISHED;
10390 add_new_free_space(cache, found_key.objectid,
10391 found_key.objectid +
10393 free_excluded_extents(cache);
10396 ret = btrfs_add_block_group_cache(info, cache);
10398 btrfs_remove_free_space_cache(cache);
10399 btrfs_put_block_group(cache);
10403 trace_btrfs_add_block_group(info, cache, 0);
10404 update_space_info(info, cache->flags, found_key.offset,
10405 btrfs_block_group_used(&cache->item),
10406 cache->bytes_super, &space_info);
10408 cache->space_info = space_info;
10410 link_block_group(cache);
10412 set_avail_alloc_bits(info, cache->flags);
10413 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10414 inc_block_group_ro(cache, 1);
10415 } else if (btrfs_block_group_used(&cache->item) == 0) {
10416 ASSERT(list_empty(&cache->bg_list));
10417 btrfs_mark_bg_unused(cache);
10421 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10422 if (!(get_alloc_profile(info, space_info->flags) &
10423 (BTRFS_BLOCK_GROUP_RAID10 |
10424 BTRFS_BLOCK_GROUP_RAID1 |
10425 BTRFS_BLOCK_GROUP_RAID5 |
10426 BTRFS_BLOCK_GROUP_RAID6 |
10427 BTRFS_BLOCK_GROUP_DUP)))
10430 * avoid allocating from un-mirrored block group if there are
10431 * mirrored block groups.
10433 list_for_each_entry(cache,
10434 &space_info->block_groups[BTRFS_RAID_RAID0],
10436 inc_block_group_ro(cache, 1);
10437 list_for_each_entry(cache,
10438 &space_info->block_groups[BTRFS_RAID_SINGLE],
10440 inc_block_group_ro(cache, 1);
10443 btrfs_add_raid_kobjects(info);
10444 init_global_block_rsv(info);
10445 ret = check_chunk_block_group_mappings(info);
10447 btrfs_free_path(path);
10451 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10453 struct btrfs_fs_info *fs_info = trans->fs_info;
10454 struct btrfs_block_group_cache *block_group;
10455 struct btrfs_root *extent_root = fs_info->extent_root;
10456 struct btrfs_block_group_item item;
10457 struct btrfs_key key;
10460 if (!trans->can_flush_pending_bgs)
10463 while (!list_empty(&trans->new_bgs)) {
10464 block_group = list_first_entry(&trans->new_bgs,
10465 struct btrfs_block_group_cache,
10470 spin_lock(&block_group->lock);
10471 memcpy(&item, &block_group->item, sizeof(item));
10472 memcpy(&key, &block_group->key, sizeof(key));
10473 spin_unlock(&block_group->lock);
10475 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10478 btrfs_abort_transaction(trans, ret);
10479 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10481 btrfs_abort_transaction(trans, ret);
10482 add_block_group_free_space(trans, block_group);
10483 /* already aborted the transaction if it failed. */
10485 btrfs_delayed_refs_rsv_release(fs_info, 1);
10486 list_del_init(&block_group->bg_list);
10488 btrfs_trans_release_chunk_metadata(trans);
10491 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10492 u64 type, u64 chunk_offset, u64 size)
10494 struct btrfs_fs_info *fs_info = trans->fs_info;
10495 struct btrfs_block_group_cache *cache;
10498 btrfs_set_log_full_commit(fs_info, trans);
10500 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10504 btrfs_set_block_group_used(&cache->item, bytes_used);
10505 btrfs_set_block_group_chunk_objectid(&cache->item,
10506 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10507 btrfs_set_block_group_flags(&cache->item, type);
10509 cache->flags = type;
10510 cache->last_byte_to_unpin = (u64)-1;
10511 cache->cached = BTRFS_CACHE_FINISHED;
10512 cache->needs_free_space = 1;
10513 ret = exclude_super_stripes(cache);
10516 * We may have excluded something, so call this just in
10519 free_excluded_extents(cache);
10520 btrfs_put_block_group(cache);
10524 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10526 free_excluded_extents(cache);
10528 #ifdef CONFIG_BTRFS_DEBUG
10529 if (btrfs_should_fragment_free_space(cache)) {
10530 u64 new_bytes_used = size - bytes_used;
10532 bytes_used += new_bytes_used >> 1;
10533 fragment_free_space(cache);
10537 * Ensure the corresponding space_info object is created and
10538 * assigned to our block group. We want our bg to be added to the rbtree
10539 * with its ->space_info set.
10541 cache->space_info = __find_space_info(fs_info, cache->flags);
10542 ASSERT(cache->space_info);
10544 ret = btrfs_add_block_group_cache(fs_info, cache);
10546 btrfs_remove_free_space_cache(cache);
10547 btrfs_put_block_group(cache);
10552 * Now that our block group has its ->space_info set and is inserted in
10553 * the rbtree, update the space info's counters.
10555 trace_btrfs_add_block_group(fs_info, cache, 1);
10556 update_space_info(fs_info, cache->flags, size, bytes_used,
10557 cache->bytes_super, &cache->space_info);
10558 update_global_block_rsv(fs_info);
10560 link_block_group(cache);
10562 list_add_tail(&cache->bg_list, &trans->new_bgs);
10563 trans->delayed_ref_updates++;
10564 btrfs_update_delayed_refs_rsv(trans);
10566 set_avail_alloc_bits(fs_info, type);
10570 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10572 u64 extra_flags = chunk_to_extended(flags) &
10573 BTRFS_EXTENDED_PROFILE_MASK;
10575 write_seqlock(&fs_info->profiles_lock);
10576 if (flags & BTRFS_BLOCK_GROUP_DATA)
10577 fs_info->avail_data_alloc_bits &= ~extra_flags;
10578 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10579 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10580 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10581 fs_info->avail_system_alloc_bits &= ~extra_flags;
10582 write_sequnlock(&fs_info->profiles_lock);
10585 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10586 u64 group_start, struct extent_map *em)
10588 struct btrfs_fs_info *fs_info = trans->fs_info;
10589 struct btrfs_root *root = fs_info->extent_root;
10590 struct btrfs_path *path;
10591 struct btrfs_block_group_cache *block_group;
10592 struct btrfs_free_cluster *cluster;
10593 struct btrfs_root *tree_root = fs_info->tree_root;
10594 struct btrfs_key key;
10595 struct inode *inode;
10596 struct kobject *kobj = NULL;
10600 struct btrfs_caching_control *caching_ctl = NULL;
10602 bool remove_rsv = false;
10604 block_group = btrfs_lookup_block_group(fs_info, group_start);
10605 BUG_ON(!block_group);
10606 BUG_ON(!block_group->ro);
10608 trace_btrfs_remove_block_group(block_group);
10610 * Free the reserved super bytes from this block group before
10613 free_excluded_extents(block_group);
10614 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10615 block_group->key.offset);
10617 memcpy(&key, &block_group->key, sizeof(key));
10618 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10619 factor = btrfs_bg_type_to_factor(block_group->flags);
10621 /* make sure this block group isn't part of an allocation cluster */
10622 cluster = &fs_info->data_alloc_cluster;
10623 spin_lock(&cluster->refill_lock);
10624 btrfs_return_cluster_to_free_space(block_group, cluster);
10625 spin_unlock(&cluster->refill_lock);
10628 * make sure this block group isn't part of a metadata
10629 * allocation cluster
10631 cluster = &fs_info->meta_alloc_cluster;
10632 spin_lock(&cluster->refill_lock);
10633 btrfs_return_cluster_to_free_space(block_group, cluster);
10634 spin_unlock(&cluster->refill_lock);
10636 path = btrfs_alloc_path();
10643 * get the inode first so any iput calls done for the io_list
10644 * aren't the final iput (no unlinks allowed now)
10646 inode = lookup_free_space_inode(fs_info, block_group, path);
10648 mutex_lock(&trans->transaction->cache_write_mutex);
10650 * make sure our free spache cache IO is done before remove the
10653 spin_lock(&trans->transaction->dirty_bgs_lock);
10654 if (!list_empty(&block_group->io_list)) {
10655 list_del_init(&block_group->io_list);
10657 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10659 spin_unlock(&trans->transaction->dirty_bgs_lock);
10660 btrfs_wait_cache_io(trans, block_group, path);
10661 btrfs_put_block_group(block_group);
10662 spin_lock(&trans->transaction->dirty_bgs_lock);
10665 if (!list_empty(&block_group->dirty_list)) {
10666 list_del_init(&block_group->dirty_list);
10668 btrfs_put_block_group(block_group);
10670 spin_unlock(&trans->transaction->dirty_bgs_lock);
10671 mutex_unlock(&trans->transaction->cache_write_mutex);
10673 if (!IS_ERR(inode)) {
10674 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10676 btrfs_add_delayed_iput(inode);
10679 clear_nlink(inode);
10680 /* One for the block groups ref */
10681 spin_lock(&block_group->lock);
10682 if (block_group->iref) {
10683 block_group->iref = 0;
10684 block_group->inode = NULL;
10685 spin_unlock(&block_group->lock);
10688 spin_unlock(&block_group->lock);
10690 /* One for our lookup ref */
10691 btrfs_add_delayed_iput(inode);
10694 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10695 key.offset = block_group->key.objectid;
10698 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10702 btrfs_release_path(path);
10704 ret = btrfs_del_item(trans, tree_root, path);
10707 btrfs_release_path(path);
10710 spin_lock(&fs_info->block_group_cache_lock);
10711 rb_erase(&block_group->cache_node,
10712 &fs_info->block_group_cache_tree);
10713 RB_CLEAR_NODE(&block_group->cache_node);
10715 if (fs_info->first_logical_byte == block_group->key.objectid)
10716 fs_info->first_logical_byte = (u64)-1;
10717 spin_unlock(&fs_info->block_group_cache_lock);
10719 down_write(&block_group->space_info->groups_sem);
10721 * we must use list_del_init so people can check to see if they
10722 * are still on the list after taking the semaphore
10724 list_del_init(&block_group->list);
10725 if (list_empty(&block_group->space_info->block_groups[index])) {
10726 kobj = block_group->space_info->block_group_kobjs[index];
10727 block_group->space_info->block_group_kobjs[index] = NULL;
10728 clear_avail_alloc_bits(fs_info, block_group->flags);
10730 up_write(&block_group->space_info->groups_sem);
10736 if (block_group->has_caching_ctl)
10737 caching_ctl = get_caching_control(block_group);
10738 if (block_group->cached == BTRFS_CACHE_STARTED)
10739 wait_block_group_cache_done(block_group);
10740 if (block_group->has_caching_ctl) {
10741 down_write(&fs_info->commit_root_sem);
10742 if (!caching_ctl) {
10743 struct btrfs_caching_control *ctl;
10745 list_for_each_entry(ctl,
10746 &fs_info->caching_block_groups, list)
10747 if (ctl->block_group == block_group) {
10749 refcount_inc(&caching_ctl->count);
10754 list_del_init(&caching_ctl->list);
10755 up_write(&fs_info->commit_root_sem);
10757 /* Once for the caching bgs list and once for us. */
10758 put_caching_control(caching_ctl);
10759 put_caching_control(caching_ctl);
10763 spin_lock(&trans->transaction->dirty_bgs_lock);
10764 if (!list_empty(&block_group->dirty_list)) {
10767 if (!list_empty(&block_group->io_list)) {
10770 spin_unlock(&trans->transaction->dirty_bgs_lock);
10771 btrfs_remove_free_space_cache(block_group);
10773 spin_lock(&block_group->space_info->lock);
10774 list_del_init(&block_group->ro_list);
10776 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10777 WARN_ON(block_group->space_info->total_bytes
10778 < block_group->key.offset);
10779 WARN_ON(block_group->space_info->bytes_readonly
10780 < block_group->key.offset);
10781 WARN_ON(block_group->space_info->disk_total
10782 < block_group->key.offset * factor);
10784 block_group->space_info->total_bytes -= block_group->key.offset;
10785 block_group->space_info->bytes_readonly -= block_group->key.offset;
10786 block_group->space_info->disk_total -= block_group->key.offset * factor;
10788 spin_unlock(&block_group->space_info->lock);
10790 memcpy(&key, &block_group->key, sizeof(key));
10792 mutex_lock(&fs_info->chunk_mutex);
10793 if (!list_empty(&em->list)) {
10794 /* We're in the transaction->pending_chunks list. */
10795 free_extent_map(em);
10797 spin_lock(&block_group->lock);
10798 block_group->removed = 1;
10800 * At this point trimming can't start on this block group, because we
10801 * removed the block group from the tree fs_info->block_group_cache_tree
10802 * so no one can't find it anymore and even if someone already got this
10803 * block group before we removed it from the rbtree, they have already
10804 * incremented block_group->trimming - if they didn't, they won't find
10805 * any free space entries because we already removed them all when we
10806 * called btrfs_remove_free_space_cache().
10808 * And we must not remove the extent map from the fs_info->mapping_tree
10809 * to prevent the same logical address range and physical device space
10810 * ranges from being reused for a new block group. This is because our
10811 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10812 * completely transactionless, so while it is trimming a range the
10813 * currently running transaction might finish and a new one start,
10814 * allowing for new block groups to be created that can reuse the same
10815 * physical device locations unless we take this special care.
10817 * There may also be an implicit trim operation if the file system
10818 * is mounted with -odiscard. The same protections must remain
10819 * in place until the extents have been discarded completely when
10820 * the transaction commit has completed.
10822 remove_em = (atomic_read(&block_group->trimming) == 0);
10824 * Make sure a trimmer task always sees the em in the pinned_chunks list
10825 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10826 * before checking block_group->removed).
10830 * Our em might be in trans->transaction->pending_chunks which
10831 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10832 * and so is the fs_info->pinned_chunks list.
10834 * So at this point we must be holding the chunk_mutex to avoid
10835 * any races with chunk allocation (more specifically at
10836 * volumes.c:contains_pending_extent()), to ensure it always
10837 * sees the em, either in the pending_chunks list or in the
10838 * pinned_chunks list.
10840 list_move_tail(&em->list, &fs_info->pinned_chunks);
10842 spin_unlock(&block_group->lock);
10845 struct extent_map_tree *em_tree;
10847 em_tree = &fs_info->mapping_tree.map_tree;
10848 write_lock(&em_tree->lock);
10850 * The em might be in the pending_chunks list, so make sure the
10851 * chunk mutex is locked, since remove_extent_mapping() will
10852 * delete us from that list.
10854 remove_extent_mapping(em_tree, em);
10855 write_unlock(&em_tree->lock);
10856 /* once for the tree */
10857 free_extent_map(em);
10860 mutex_unlock(&fs_info->chunk_mutex);
10862 ret = remove_block_group_free_space(trans, block_group);
10866 btrfs_put_block_group(block_group);
10867 btrfs_put_block_group(block_group);
10869 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10875 ret = btrfs_del_item(trans, root, path);
10878 btrfs_delayed_refs_rsv_release(fs_info, 1);
10879 btrfs_free_path(path);
10883 struct btrfs_trans_handle *
10884 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10885 const u64 chunk_offset)
10887 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10888 struct extent_map *em;
10889 struct map_lookup *map;
10890 unsigned int num_items;
10892 read_lock(&em_tree->lock);
10893 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10894 read_unlock(&em_tree->lock);
10895 ASSERT(em && em->start == chunk_offset);
10898 * We need to reserve 3 + N units from the metadata space info in order
10899 * to remove a block group (done at btrfs_remove_chunk() and at
10900 * btrfs_remove_block_group()), which are used for:
10902 * 1 unit for adding the free space inode's orphan (located in the tree
10904 * 1 unit for deleting the block group item (located in the extent
10906 * 1 unit for deleting the free space item (located in tree of tree
10908 * N units for deleting N device extent items corresponding to each
10909 * stripe (located in the device tree).
10911 * In order to remove a block group we also need to reserve units in the
10912 * system space info in order to update the chunk tree (update one or
10913 * more device items and remove one chunk item), but this is done at
10914 * btrfs_remove_chunk() through a call to check_system_chunk().
10916 map = em->map_lookup;
10917 num_items = 3 + map->num_stripes;
10918 free_extent_map(em);
10920 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10925 * Process the unused_bgs list and remove any that don't have any allocated
10926 * space inside of them.
10928 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10930 struct btrfs_block_group_cache *block_group;
10931 struct btrfs_space_info *space_info;
10932 struct btrfs_trans_handle *trans;
10935 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10938 spin_lock(&fs_info->unused_bgs_lock);
10939 while (!list_empty(&fs_info->unused_bgs)) {
10943 block_group = list_first_entry(&fs_info->unused_bgs,
10944 struct btrfs_block_group_cache,
10946 list_del_init(&block_group->bg_list);
10948 space_info = block_group->space_info;
10950 if (ret || btrfs_mixed_space_info(space_info)) {
10951 btrfs_put_block_group(block_group);
10954 spin_unlock(&fs_info->unused_bgs_lock);
10956 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10958 /* Don't want to race with allocators so take the groups_sem */
10959 down_write(&space_info->groups_sem);
10960 spin_lock(&block_group->lock);
10961 if (block_group->reserved || block_group->pinned ||
10962 btrfs_block_group_used(&block_group->item) ||
10964 list_is_singular(&block_group->list)) {
10966 * We want to bail if we made new allocations or have
10967 * outstanding allocations in this block group. We do
10968 * the ro check in case balance is currently acting on
10969 * this block group.
10971 trace_btrfs_skip_unused_block_group(block_group);
10972 spin_unlock(&block_group->lock);
10973 up_write(&space_info->groups_sem);
10976 spin_unlock(&block_group->lock);
10978 /* We don't want to force the issue, only flip if it's ok. */
10979 ret = inc_block_group_ro(block_group, 0);
10980 up_write(&space_info->groups_sem);
10987 * Want to do this before we do anything else so we can recover
10988 * properly if we fail to join the transaction.
10990 trans = btrfs_start_trans_remove_block_group(fs_info,
10991 block_group->key.objectid);
10992 if (IS_ERR(trans)) {
10993 btrfs_dec_block_group_ro(block_group);
10994 ret = PTR_ERR(trans);
10999 * We could have pending pinned extents for this block group,
11000 * just delete them, we don't care about them anymore.
11002 start = block_group->key.objectid;
11003 end = start + block_group->key.offset - 1;
11005 * Hold the unused_bg_unpin_mutex lock to avoid racing with
11006 * btrfs_finish_extent_commit(). If we are at transaction N,
11007 * another task might be running finish_extent_commit() for the
11008 * previous transaction N - 1, and have seen a range belonging
11009 * to the block group in freed_extents[] before we were able to
11010 * clear the whole block group range from freed_extents[]. This
11011 * means that task can lookup for the block group after we
11012 * unpinned it from freed_extents[] and removed it, leading to
11013 * a BUG_ON() at btrfs_unpin_extent_range().
11015 mutex_lock(&fs_info->unused_bg_unpin_mutex);
11016 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
11019 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11020 btrfs_dec_block_group_ro(block_group);
11023 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
11026 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11027 btrfs_dec_block_group_ro(block_group);
11030 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11032 /* Reset pinned so btrfs_put_block_group doesn't complain */
11033 spin_lock(&space_info->lock);
11034 spin_lock(&block_group->lock);
11036 update_bytes_pinned(space_info, -block_group->pinned);
11037 space_info->bytes_readonly += block_group->pinned;
11038 percpu_counter_add_batch(&space_info->total_bytes_pinned,
11039 -block_group->pinned,
11040 BTRFS_TOTAL_BYTES_PINNED_BATCH);
11041 block_group->pinned = 0;
11043 spin_unlock(&block_group->lock);
11044 spin_unlock(&space_info->lock);
11046 /* DISCARD can flip during remount */
11047 trimming = btrfs_test_opt(fs_info, DISCARD);
11049 /* Implicit trim during transaction commit. */
11051 btrfs_get_block_group_trimming(block_group);
11054 * Btrfs_remove_chunk will abort the transaction if things go
11057 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
11061 btrfs_put_block_group_trimming(block_group);
11066 * If we're not mounted with -odiscard, we can just forget
11067 * about this block group. Otherwise we'll need to wait
11068 * until transaction commit to do the actual discard.
11071 spin_lock(&fs_info->unused_bgs_lock);
11073 * A concurrent scrub might have added us to the list
11074 * fs_info->unused_bgs, so use a list_move operation
11075 * to add the block group to the deleted_bgs list.
11077 list_move(&block_group->bg_list,
11078 &trans->transaction->deleted_bgs);
11079 spin_unlock(&fs_info->unused_bgs_lock);
11080 btrfs_get_block_group(block_group);
11083 btrfs_end_transaction(trans);
11085 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
11086 btrfs_put_block_group(block_group);
11087 spin_lock(&fs_info->unused_bgs_lock);
11089 spin_unlock(&fs_info->unused_bgs_lock);
11092 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
11094 struct btrfs_super_block *disk_super;
11100 disk_super = fs_info->super_copy;
11101 if (!btrfs_super_root(disk_super))
11104 features = btrfs_super_incompat_flags(disk_super);
11105 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11108 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11109 ret = create_space_info(fs_info, flags);
11114 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11115 ret = create_space_info(fs_info, flags);
11117 flags = BTRFS_BLOCK_GROUP_METADATA;
11118 ret = create_space_info(fs_info, flags);
11122 flags = BTRFS_BLOCK_GROUP_DATA;
11123 ret = create_space_info(fs_info, flags);
11129 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
11130 u64 start, u64 end)
11132 return unpin_extent_range(fs_info, start, end, false);
11136 * It used to be that old block groups would be left around forever.
11137 * Iterating over them would be enough to trim unused space. Since we
11138 * now automatically remove them, we also need to iterate over unallocated
11141 * We don't want a transaction for this since the discard may take a
11142 * substantial amount of time. We don't require that a transaction be
11143 * running, but we do need to take a running transaction into account
11144 * to ensure that we're not discarding chunks that were released or
11145 * allocated in the current transaction.
11147 * Holding the chunks lock will prevent other threads from allocating
11148 * or releasing chunks, but it won't prevent a running transaction
11149 * from committing and releasing the memory that the pending chunks
11150 * list head uses. For that, we need to take a reference to the
11151 * transaction and hold the commit root sem. We only need to hold
11152 * it while performing the free space search since we have already
11153 * held back allocations.
11155 static int btrfs_trim_free_extents(struct btrfs_device *device,
11156 u64 minlen, u64 *trimmed)
11158 u64 start = 0, len = 0;
11163 /* Discard not supported = nothing to do. */
11164 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
11167 /* Not writeable = nothing to do. */
11168 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
11171 /* No free space = nothing to do. */
11172 if (device->total_bytes <= device->bytes_used)
11178 struct btrfs_fs_info *fs_info = device->fs_info;
11179 struct btrfs_transaction *trans;
11182 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11186 ret = down_read_killable(&fs_info->commit_root_sem);
11188 mutex_unlock(&fs_info->chunk_mutex);
11192 spin_lock(&fs_info->trans_lock);
11193 trans = fs_info->running_transaction;
11195 refcount_inc(&trans->use_count);
11196 spin_unlock(&fs_info->trans_lock);
11199 up_read(&fs_info->commit_root_sem);
11201 ret = find_free_dev_extent_start(trans, device, minlen, start,
11204 up_read(&fs_info->commit_root_sem);
11205 btrfs_put_transaction(trans);
11209 mutex_unlock(&fs_info->chunk_mutex);
11210 if (ret == -ENOSPC)
11215 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11216 mutex_unlock(&fs_info->chunk_mutex);
11224 if (fatal_signal_pending(current)) {
11225 ret = -ERESTARTSYS;
11236 * Trim the whole filesystem by:
11237 * 1) trimming the free space in each block group
11238 * 2) trimming the unallocated space on each device
11240 * This will also continue trimming even if a block group or device encounters
11241 * an error. The return value will be the last error, or 0 if nothing bad
11244 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11246 struct btrfs_block_group_cache *cache = NULL;
11247 struct btrfs_device *device;
11248 struct list_head *devices;
11254 u64 dev_failed = 0;
11259 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11260 for (; cache; cache = next_block_group(fs_info, cache)) {
11261 if (cache->key.objectid >= (range->start + range->len)) {
11262 btrfs_put_block_group(cache);
11266 start = max(range->start, cache->key.objectid);
11267 end = min(range->start + range->len,
11268 cache->key.objectid + cache->key.offset);
11270 if (end - start >= range->minlen) {
11271 if (!block_group_cache_done(cache)) {
11272 ret = cache_block_group(cache, 0);
11278 ret = wait_block_group_cache_done(cache);
11285 ret = btrfs_trim_block_group(cache,
11291 trimmed += group_trimmed;
11301 btrfs_warn(fs_info,
11302 "failed to trim %llu block group(s), last error %d",
11303 bg_failed, bg_ret);
11304 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11305 devices = &fs_info->fs_devices->devices;
11306 list_for_each_entry(device, devices, dev_list) {
11307 ret = btrfs_trim_free_extents(device, range->minlen,
11315 trimmed += group_trimmed;
11317 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11320 btrfs_warn(fs_info,
11321 "failed to trim %llu device(s), last error %d",
11322 dev_failed, dev_ret);
11323 range->len = trimmed;
11330 * btrfs_{start,end}_write_no_snapshotting() are similar to
11331 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11332 * data into the page cache through nocow before the subvolume is snapshoted,
11333 * but flush the data into disk after the snapshot creation, or to prevent
11334 * operations while snapshotting is ongoing and that cause the snapshot to be
11335 * inconsistent (writes followed by expanding truncates for example).
11337 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11339 percpu_counter_dec(&root->subv_writers->counter);
11340 cond_wake_up(&root->subv_writers->wait);
11343 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11345 if (atomic_read(&root->will_be_snapshotted))
11348 percpu_counter_inc(&root->subv_writers->counter);
11350 * Make sure counter is updated before we check for snapshot creation.
11353 if (atomic_read(&root->will_be_snapshotted)) {
11354 btrfs_end_write_no_snapshotting(root);
11360 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11365 ret = btrfs_start_write_no_snapshotting(root);
11368 wait_var_event(&root->will_be_snapshotted,
11369 !atomic_read(&root->will_be_snapshotted));
11373 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
11375 struct btrfs_fs_info *fs_info = bg->fs_info;
11377 spin_lock(&fs_info->unused_bgs_lock);
11378 if (list_empty(&bg->bg_list)) {
11379 btrfs_get_block_group(bg);
11380 trace_btrfs_add_unused_block_group(bg);
11381 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
11383 spin_unlock(&fs_info->unused_bgs_lock);