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 requiried,
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 void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info,
2460 struct btrfs_delayed_ref_root *delayed_refs,
2461 struct btrfs_delayed_ref_head *head)
2463 int nr_items = 1; /* Dropping this ref head update. */
2465 if (head->total_ref_mod < 0) {
2466 struct btrfs_space_info *space_info;
2470 flags = BTRFS_BLOCK_GROUP_DATA;
2471 else if (head->is_system)
2472 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2474 flags = BTRFS_BLOCK_GROUP_METADATA;
2475 space_info = __find_space_info(fs_info, flags);
2477 percpu_counter_add_batch(&space_info->total_bytes_pinned,
2479 BTRFS_TOTAL_BYTES_PINNED_BATCH);
2482 * We had csum deletions accounted for in our delayed refs rsv,
2483 * we need to drop the csum leaves for this update from our
2486 if (head->is_data) {
2487 spin_lock(&delayed_refs->lock);
2488 delayed_refs->pending_csums -= head->num_bytes;
2489 spin_unlock(&delayed_refs->lock);
2490 nr_items += btrfs_csum_bytes_to_leaves(fs_info,
2495 /* Also free its reserved qgroup space */
2496 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2497 head->qgroup_reserved);
2498 btrfs_delayed_refs_rsv_release(fs_info, nr_items);
2501 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2502 struct btrfs_delayed_ref_head *head)
2505 struct btrfs_fs_info *fs_info = trans->fs_info;
2506 struct btrfs_delayed_ref_root *delayed_refs;
2509 delayed_refs = &trans->transaction->delayed_refs;
2511 ret = run_and_cleanup_extent_op(trans, head);
2513 unselect_delayed_ref_head(delayed_refs, head);
2514 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2521 * Need to drop our head ref lock and re-acquire the delayed ref lock
2522 * and then re-check to make sure nobody got added.
2524 spin_unlock(&head->lock);
2525 spin_lock(&delayed_refs->lock);
2526 spin_lock(&head->lock);
2527 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) {
2528 spin_unlock(&head->lock);
2529 spin_unlock(&delayed_refs->lock);
2532 btrfs_delete_ref_head(delayed_refs, head);
2533 spin_unlock(&head->lock);
2534 spin_unlock(&delayed_refs->lock);
2536 if (head->must_insert_reserved) {
2537 btrfs_pin_extent(fs_info, head->bytenr,
2538 head->num_bytes, 1);
2539 if (head->is_data) {
2540 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2545 btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head);
2547 trace_run_delayed_ref_head(fs_info, head, 0);
2548 btrfs_delayed_ref_unlock(head);
2549 btrfs_put_delayed_ref_head(head);
2553 static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head(
2554 struct btrfs_trans_handle *trans)
2556 struct btrfs_delayed_ref_root *delayed_refs =
2557 &trans->transaction->delayed_refs;
2558 struct btrfs_delayed_ref_head *head = NULL;
2561 spin_lock(&delayed_refs->lock);
2562 head = btrfs_select_ref_head(delayed_refs);
2564 spin_unlock(&delayed_refs->lock);
2569 * Grab the lock that says we are going to process all the refs for
2572 ret = btrfs_delayed_ref_lock(delayed_refs, head);
2573 spin_unlock(&delayed_refs->lock);
2576 * We may have dropped the spin lock to get the head mutex lock, and
2577 * that might have given someone else time to free the head. If that's
2578 * true, it has been removed from our list and we can move on.
2581 head = ERR_PTR(-EAGAIN);
2586 static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans,
2587 struct btrfs_delayed_ref_head *locked_ref,
2588 unsigned long *run_refs)
2590 struct btrfs_fs_info *fs_info = trans->fs_info;
2591 struct btrfs_delayed_ref_root *delayed_refs;
2592 struct btrfs_delayed_extent_op *extent_op;
2593 struct btrfs_delayed_ref_node *ref;
2594 int must_insert_reserved = 0;
2597 delayed_refs = &trans->transaction->delayed_refs;
2599 lockdep_assert_held(&locked_ref->mutex);
2600 lockdep_assert_held(&locked_ref->lock);
2602 while ((ref = select_delayed_ref(locked_ref))) {
2604 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2605 spin_unlock(&locked_ref->lock);
2606 unselect_delayed_ref_head(delayed_refs, locked_ref);
2612 rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree);
2613 RB_CLEAR_NODE(&ref->ref_node);
2614 if (!list_empty(&ref->add_list))
2615 list_del(&ref->add_list);
2617 * When we play the delayed ref, also correct the ref_mod on
2620 switch (ref->action) {
2621 case BTRFS_ADD_DELAYED_REF:
2622 case BTRFS_ADD_DELAYED_EXTENT:
2623 locked_ref->ref_mod -= ref->ref_mod;
2625 case BTRFS_DROP_DELAYED_REF:
2626 locked_ref->ref_mod += ref->ref_mod;
2631 atomic_dec(&delayed_refs->num_entries);
2634 * Record the must_insert_reserved flag before we drop the
2637 must_insert_reserved = locked_ref->must_insert_reserved;
2638 locked_ref->must_insert_reserved = 0;
2640 extent_op = locked_ref->extent_op;
2641 locked_ref->extent_op = NULL;
2642 spin_unlock(&locked_ref->lock);
2644 ret = run_one_delayed_ref(trans, ref, extent_op,
2645 must_insert_reserved);
2647 btrfs_free_delayed_extent_op(extent_op);
2649 unselect_delayed_ref_head(delayed_refs, locked_ref);
2650 btrfs_put_delayed_ref(ref);
2651 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2656 btrfs_put_delayed_ref(ref);
2659 spin_lock(&locked_ref->lock);
2660 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2667 * Returns 0 on success or if called with an already aborted transaction.
2668 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2670 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2673 struct btrfs_fs_info *fs_info = trans->fs_info;
2674 struct btrfs_delayed_ref_root *delayed_refs;
2675 struct btrfs_delayed_ref_head *locked_ref = NULL;
2676 ktime_t start = ktime_get();
2678 unsigned long count = 0;
2679 unsigned long actual_count = 0;
2681 delayed_refs = &trans->transaction->delayed_refs;
2684 locked_ref = btrfs_obtain_ref_head(trans);
2685 if (IS_ERR_OR_NULL(locked_ref)) {
2686 if (PTR_ERR(locked_ref) == -EAGAIN) {
2695 * We need to try and merge add/drops of the same ref since we
2696 * can run into issues with relocate dropping the implicit ref
2697 * and then it being added back again before the drop can
2698 * finish. If we merged anything we need to re-loop so we can
2700 * Or we can get node references of the same type that weren't
2701 * merged when created due to bumps in the tree mod seq, and
2702 * we need to merge them to prevent adding an inline extent
2703 * backref before dropping it (triggering a BUG_ON at
2704 * insert_inline_extent_backref()).
2706 spin_lock(&locked_ref->lock);
2707 btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref);
2709 ret = btrfs_run_delayed_refs_for_head(trans, locked_ref,
2711 if (ret < 0 && ret != -EAGAIN) {
2713 * Error, btrfs_run_delayed_refs_for_head already
2714 * unlocked everything so just bail out
2719 * Success, perform the usual cleanup of a processed
2722 ret = cleanup_ref_head(trans, locked_ref);
2724 /* We dropped our lock, we need to loop. */
2733 * Either success case or btrfs_run_delayed_refs_for_head
2734 * returned -EAGAIN, meaning we need to select another head
2739 } while ((nr != -1 && count < nr) || locked_ref);
2742 * We don't want to include ref heads since we can have empty ref heads
2743 * and those will drastically skew our runtime down since we just do
2744 * accounting, no actual extent tree updates.
2746 if (actual_count > 0) {
2747 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2751 * We weigh the current average higher than our current runtime
2752 * to avoid large swings in the average.
2754 spin_lock(&delayed_refs->lock);
2755 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2756 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2757 spin_unlock(&delayed_refs->lock);
2762 #ifdef SCRAMBLE_DELAYED_REFS
2764 * Normally delayed refs get processed in ascending bytenr order. This
2765 * correlates in most cases to the order added. To expose dependencies on this
2766 * order, we start to process the tree in the middle instead of the beginning
2768 static u64 find_middle(struct rb_root *root)
2770 struct rb_node *n = root->rb_node;
2771 struct btrfs_delayed_ref_node *entry;
2774 u64 first = 0, last = 0;
2778 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2779 first = entry->bytenr;
2783 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2784 last = entry->bytenr;
2789 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2790 WARN_ON(!entry->in_tree);
2792 middle = entry->bytenr;
2805 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2809 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2810 sizeof(struct btrfs_extent_inline_ref));
2811 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2812 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2815 * We don't ever fill up leaves all the way so multiply by 2 just to be
2816 * closer to what we're really going to want to use.
2818 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2822 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2823 * would require to store the csums for that many bytes.
2825 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2828 u64 num_csums_per_leaf;
2831 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2832 num_csums_per_leaf = div64_u64(csum_size,
2833 (u64)btrfs_super_csum_size(fs_info->super_copy));
2834 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2835 num_csums += num_csums_per_leaf - 1;
2836 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2840 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info)
2842 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
2843 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
2847 spin_lock(&global_rsv->lock);
2848 reserved = global_rsv->reserved;
2849 spin_unlock(&global_rsv->lock);
2852 * Since the global reserve is just kind of magic we don't really want
2853 * to rely on it to save our bacon, so if our size is more than the
2854 * delayed_refs_rsv and the global rsv then it's time to think about
2857 spin_lock(&delayed_refs_rsv->lock);
2858 reserved += delayed_refs_rsv->reserved;
2859 if (delayed_refs_rsv->size >= reserved)
2861 spin_unlock(&delayed_refs_rsv->lock);
2865 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans)
2868 atomic_read(&trans->transaction->delayed_refs.num_entries);
2873 avg_runtime = trans->fs_info->avg_delayed_ref_runtime;
2874 val = num_entries * avg_runtime;
2875 if (val >= NSEC_PER_SEC)
2877 if (val >= NSEC_PER_SEC / 2)
2880 return btrfs_check_space_for_delayed_refs(trans->fs_info);
2883 struct async_delayed_refs {
2884 struct btrfs_root *root;
2889 struct completion wait;
2890 struct btrfs_work work;
2893 static inline struct async_delayed_refs *
2894 to_async_delayed_refs(struct btrfs_work *work)
2896 return container_of(work, struct async_delayed_refs, work);
2899 static void delayed_ref_async_start(struct btrfs_work *work)
2901 struct async_delayed_refs *async = to_async_delayed_refs(work);
2902 struct btrfs_trans_handle *trans;
2903 struct btrfs_fs_info *fs_info = async->root->fs_info;
2906 /* if the commit is already started, we don't need to wait here */
2907 if (btrfs_transaction_blocked(fs_info))
2910 trans = btrfs_join_transaction(async->root);
2911 if (IS_ERR(trans)) {
2912 async->error = PTR_ERR(trans);
2917 * trans->sync means that when we call end_transaction, we won't
2918 * wait on delayed refs
2922 /* Don't bother flushing if we got into a different transaction */
2923 if (trans->transid > async->transid)
2926 ret = btrfs_run_delayed_refs(trans, async->count);
2930 ret = btrfs_end_transaction(trans);
2931 if (ret && !async->error)
2935 complete(&async->wait);
2940 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2941 unsigned long count, u64 transid, int wait)
2943 struct async_delayed_refs *async;
2946 async = kmalloc(sizeof(*async), GFP_NOFS);
2950 async->root = fs_info->tree_root;
2951 async->count = count;
2953 async->transid = transid;
2958 init_completion(&async->wait);
2960 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2961 delayed_ref_async_start, NULL, NULL);
2963 btrfs_queue_work(fs_info->extent_workers, &async->work);
2966 wait_for_completion(&async->wait);
2975 * this starts processing the delayed reference count updates and
2976 * extent insertions we have queued up so far. count can be
2977 * 0, which means to process everything in the tree at the start
2978 * of the run (but not newly added entries), or it can be some target
2979 * number you'd like to process.
2981 * Returns 0 on success or if called with an aborted transaction
2982 * Returns <0 on error and aborts the transaction
2984 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2985 unsigned long count)
2987 struct btrfs_fs_info *fs_info = trans->fs_info;
2988 struct rb_node *node;
2989 struct btrfs_delayed_ref_root *delayed_refs;
2990 struct btrfs_delayed_ref_head *head;
2992 int run_all = count == (unsigned long)-1;
2994 /* We'll clean this up in btrfs_cleanup_transaction */
2998 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3001 delayed_refs = &trans->transaction->delayed_refs;
3003 count = atomic_read(&delayed_refs->num_entries) * 2;
3006 #ifdef SCRAMBLE_DELAYED_REFS
3007 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3009 ret = __btrfs_run_delayed_refs(trans, count);
3011 btrfs_abort_transaction(trans, ret);
3016 if (!list_empty(&trans->new_bgs))
3017 btrfs_create_pending_block_groups(trans);
3019 spin_lock(&delayed_refs->lock);
3020 node = rb_first_cached(&delayed_refs->href_root);
3022 spin_unlock(&delayed_refs->lock);
3025 head = rb_entry(node, struct btrfs_delayed_ref_head,
3027 refcount_inc(&head->refs);
3028 spin_unlock(&delayed_refs->lock);
3030 /* Mutex was contended, block until it's released and retry. */
3031 mutex_lock(&head->mutex);
3032 mutex_unlock(&head->mutex);
3034 btrfs_put_delayed_ref_head(head);
3042 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3043 struct btrfs_fs_info *fs_info,
3044 u64 bytenr, u64 num_bytes, u64 flags,
3045 int level, int is_data)
3047 struct btrfs_delayed_extent_op *extent_op;
3050 extent_op = btrfs_alloc_delayed_extent_op();
3054 extent_op->flags_to_set = flags;
3055 extent_op->update_flags = true;
3056 extent_op->update_key = false;
3057 extent_op->is_data = is_data ? true : false;
3058 extent_op->level = level;
3060 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3061 num_bytes, extent_op);
3063 btrfs_free_delayed_extent_op(extent_op);
3067 static noinline int check_delayed_ref(struct btrfs_root *root,
3068 struct btrfs_path *path,
3069 u64 objectid, u64 offset, u64 bytenr)
3071 struct btrfs_delayed_ref_head *head;
3072 struct btrfs_delayed_ref_node *ref;
3073 struct btrfs_delayed_data_ref *data_ref;
3074 struct btrfs_delayed_ref_root *delayed_refs;
3075 struct btrfs_transaction *cur_trans;
3076 struct rb_node *node;
3079 spin_lock(&root->fs_info->trans_lock);
3080 cur_trans = root->fs_info->running_transaction;
3082 refcount_inc(&cur_trans->use_count);
3083 spin_unlock(&root->fs_info->trans_lock);
3087 delayed_refs = &cur_trans->delayed_refs;
3088 spin_lock(&delayed_refs->lock);
3089 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3091 spin_unlock(&delayed_refs->lock);
3092 btrfs_put_transaction(cur_trans);
3096 if (!mutex_trylock(&head->mutex)) {
3097 refcount_inc(&head->refs);
3098 spin_unlock(&delayed_refs->lock);
3100 btrfs_release_path(path);
3103 * Mutex was contended, block until it's released and let
3106 mutex_lock(&head->mutex);
3107 mutex_unlock(&head->mutex);
3108 btrfs_put_delayed_ref_head(head);
3109 btrfs_put_transaction(cur_trans);
3112 spin_unlock(&delayed_refs->lock);
3114 spin_lock(&head->lock);
3116 * XXX: We should replace this with a proper search function in the
3119 for (node = rb_first_cached(&head->ref_tree); node;
3120 node = rb_next(node)) {
3121 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3122 /* If it's a shared ref we know a cross reference exists */
3123 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3128 data_ref = btrfs_delayed_node_to_data_ref(ref);
3131 * If our ref doesn't match the one we're currently looking at
3132 * then we have a cross reference.
3134 if (data_ref->root != root->root_key.objectid ||
3135 data_ref->objectid != objectid ||
3136 data_ref->offset != offset) {
3141 spin_unlock(&head->lock);
3142 mutex_unlock(&head->mutex);
3143 btrfs_put_transaction(cur_trans);
3147 static noinline int check_committed_ref(struct btrfs_root *root,
3148 struct btrfs_path *path,
3149 u64 objectid, u64 offset, u64 bytenr)
3151 struct btrfs_fs_info *fs_info = root->fs_info;
3152 struct btrfs_root *extent_root = fs_info->extent_root;
3153 struct extent_buffer *leaf;
3154 struct btrfs_extent_data_ref *ref;
3155 struct btrfs_extent_inline_ref *iref;
3156 struct btrfs_extent_item *ei;
3157 struct btrfs_key key;
3162 key.objectid = bytenr;
3163 key.offset = (u64)-1;
3164 key.type = BTRFS_EXTENT_ITEM_KEY;
3166 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3169 BUG_ON(ret == 0); /* Corruption */
3172 if (path->slots[0] == 0)
3176 leaf = path->nodes[0];
3177 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3179 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3183 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3184 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3186 if (item_size != sizeof(*ei) +
3187 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3190 if (btrfs_extent_generation(leaf, ei) <=
3191 btrfs_root_last_snapshot(&root->root_item))
3194 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3196 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3197 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3200 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3201 if (btrfs_extent_refs(leaf, ei) !=
3202 btrfs_extent_data_ref_count(leaf, ref) ||
3203 btrfs_extent_data_ref_root(leaf, ref) !=
3204 root->root_key.objectid ||
3205 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3206 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3214 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3217 struct btrfs_path *path;
3220 path = btrfs_alloc_path();
3225 ret = check_committed_ref(root, path, objectid,
3227 if (ret && ret != -ENOENT)
3230 ret = check_delayed_ref(root, path, objectid, offset, bytenr);
3231 } while (ret == -EAGAIN);
3234 btrfs_free_path(path);
3235 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3240 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3241 struct btrfs_root *root,
3242 struct extent_buffer *buf,
3243 int full_backref, int inc)
3245 struct btrfs_fs_info *fs_info = root->fs_info;
3251 struct btrfs_key key;
3252 struct btrfs_file_extent_item *fi;
3256 int (*process_func)(struct btrfs_trans_handle *,
3257 struct btrfs_root *,
3258 u64, u64, u64, u64, u64, u64);
3261 if (btrfs_is_testing(fs_info))
3264 ref_root = btrfs_header_owner(buf);
3265 nritems = btrfs_header_nritems(buf);
3266 level = btrfs_header_level(buf);
3268 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3272 process_func = btrfs_inc_extent_ref;
3274 process_func = btrfs_free_extent;
3277 parent = buf->start;
3281 for (i = 0; i < nritems; i++) {
3283 btrfs_item_key_to_cpu(buf, &key, i);
3284 if (key.type != BTRFS_EXTENT_DATA_KEY)
3286 fi = btrfs_item_ptr(buf, i,
3287 struct btrfs_file_extent_item);
3288 if (btrfs_file_extent_type(buf, fi) ==
3289 BTRFS_FILE_EXTENT_INLINE)
3291 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3295 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3296 key.offset -= btrfs_file_extent_offset(buf, fi);
3297 ret = process_func(trans, root, bytenr, num_bytes,
3298 parent, ref_root, key.objectid,
3303 bytenr = btrfs_node_blockptr(buf, i);
3304 num_bytes = fs_info->nodesize;
3305 ret = process_func(trans, root, bytenr, num_bytes,
3306 parent, ref_root, level - 1, 0);
3316 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3317 struct extent_buffer *buf, int full_backref)
3319 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3322 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3323 struct extent_buffer *buf, int full_backref)
3325 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3328 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3329 struct btrfs_fs_info *fs_info,
3330 struct btrfs_path *path,
3331 struct btrfs_block_group_cache *cache)
3334 struct btrfs_root *extent_root = fs_info->extent_root;
3336 struct extent_buffer *leaf;
3338 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3345 leaf = path->nodes[0];
3346 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3347 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3348 btrfs_mark_buffer_dirty(leaf);
3350 btrfs_release_path(path);
3355 static struct btrfs_block_group_cache *
3356 next_block_group(struct btrfs_fs_info *fs_info,
3357 struct btrfs_block_group_cache *cache)
3359 struct rb_node *node;
3361 spin_lock(&fs_info->block_group_cache_lock);
3363 /* If our block group was removed, we need a full search. */
3364 if (RB_EMPTY_NODE(&cache->cache_node)) {
3365 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3367 spin_unlock(&fs_info->block_group_cache_lock);
3368 btrfs_put_block_group(cache);
3369 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3371 node = rb_next(&cache->cache_node);
3372 btrfs_put_block_group(cache);
3374 cache = rb_entry(node, struct btrfs_block_group_cache,
3376 btrfs_get_block_group(cache);
3379 spin_unlock(&fs_info->block_group_cache_lock);
3383 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3384 struct btrfs_trans_handle *trans,
3385 struct btrfs_path *path)
3387 struct btrfs_fs_info *fs_info = block_group->fs_info;
3388 struct btrfs_root *root = fs_info->tree_root;
3389 struct inode *inode = NULL;
3390 struct extent_changeset *data_reserved = NULL;
3392 int dcs = BTRFS_DC_ERROR;
3398 * If this block group is smaller than 100 megs don't bother caching the
3401 if (block_group->key.offset < (100 * SZ_1M)) {
3402 spin_lock(&block_group->lock);
3403 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3404 spin_unlock(&block_group->lock);
3411 inode = lookup_free_space_inode(fs_info, block_group, path);
3412 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3413 ret = PTR_ERR(inode);
3414 btrfs_release_path(path);
3418 if (IS_ERR(inode)) {
3422 if (block_group->ro)
3425 ret = create_free_space_inode(fs_info, trans, block_group,
3433 * We want to set the generation to 0, that way if anything goes wrong
3434 * from here on out we know not to trust this cache when we load up next
3437 BTRFS_I(inode)->generation = 0;
3438 ret = btrfs_update_inode(trans, root, inode);
3441 * So theoretically we could recover from this, simply set the
3442 * super cache generation to 0 so we know to invalidate the
3443 * cache, but then we'd have to keep track of the block groups
3444 * that fail this way so we know we _have_ to reset this cache
3445 * before the next commit or risk reading stale cache. So to
3446 * limit our exposure to horrible edge cases lets just abort the
3447 * transaction, this only happens in really bad situations
3450 btrfs_abort_transaction(trans, ret);
3455 /* We've already setup this transaction, go ahead and exit */
3456 if (block_group->cache_generation == trans->transid &&
3457 i_size_read(inode)) {
3458 dcs = BTRFS_DC_SETUP;
3462 if (i_size_read(inode) > 0) {
3463 ret = btrfs_check_trunc_cache_free_space(fs_info,
3464 &fs_info->global_block_rsv);
3468 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3473 spin_lock(&block_group->lock);
3474 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3475 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3477 * don't bother trying to write stuff out _if_
3478 * a) we're not cached,
3479 * b) we're with nospace_cache mount option,
3480 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3482 dcs = BTRFS_DC_WRITTEN;
3483 spin_unlock(&block_group->lock);
3486 spin_unlock(&block_group->lock);
3489 * We hit an ENOSPC when setting up the cache in this transaction, just
3490 * skip doing the setup, we've already cleared the cache so we're safe.
3492 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3498 * Try to preallocate enough space based on how big the block group is.
3499 * Keep in mind this has to include any pinned space which could end up
3500 * taking up quite a bit since it's not folded into the other space
3503 num_pages = div_u64(block_group->key.offset, SZ_256M);
3508 num_pages *= PAGE_SIZE;
3510 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3514 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3515 num_pages, num_pages,
3518 * Our cache requires contiguous chunks so that we don't modify a bunch
3519 * of metadata or split extents when writing the cache out, which means
3520 * we can enospc if we are heavily fragmented in addition to just normal
3521 * out of space conditions. So if we hit this just skip setting up any
3522 * other block groups for this transaction, maybe we'll unpin enough
3523 * space the next time around.
3526 dcs = BTRFS_DC_SETUP;
3527 else if (ret == -ENOSPC)
3528 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3533 btrfs_release_path(path);
3535 spin_lock(&block_group->lock);
3536 if (!ret && dcs == BTRFS_DC_SETUP)
3537 block_group->cache_generation = trans->transid;
3538 block_group->disk_cache_state = dcs;
3539 spin_unlock(&block_group->lock);
3541 extent_changeset_free(data_reserved);
3545 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3546 struct btrfs_fs_info *fs_info)
3548 struct btrfs_block_group_cache *cache, *tmp;
3549 struct btrfs_transaction *cur_trans = trans->transaction;
3550 struct btrfs_path *path;
3552 if (list_empty(&cur_trans->dirty_bgs) ||
3553 !btrfs_test_opt(fs_info, SPACE_CACHE))
3556 path = btrfs_alloc_path();
3560 /* Could add new block groups, use _safe just in case */
3561 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3563 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3564 cache_save_setup(cache, trans, path);
3567 btrfs_free_path(path);
3572 * transaction commit does final block group cache writeback during a
3573 * critical section where nothing is allowed to change the FS. This is
3574 * required in order for the cache to actually match the block group,
3575 * but can introduce a lot of latency into the commit.
3577 * So, btrfs_start_dirty_block_groups is here to kick off block group
3578 * cache IO. There's a chance we'll have to redo some of it if the
3579 * block group changes again during the commit, but it greatly reduces
3580 * the commit latency by getting rid of the easy block groups while
3581 * we're still allowing others to join the commit.
3583 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3585 struct btrfs_fs_info *fs_info = trans->fs_info;
3586 struct btrfs_block_group_cache *cache;
3587 struct btrfs_transaction *cur_trans = trans->transaction;
3590 struct btrfs_path *path = NULL;
3592 struct list_head *io = &cur_trans->io_bgs;
3593 int num_started = 0;
3596 spin_lock(&cur_trans->dirty_bgs_lock);
3597 if (list_empty(&cur_trans->dirty_bgs)) {
3598 spin_unlock(&cur_trans->dirty_bgs_lock);
3601 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3602 spin_unlock(&cur_trans->dirty_bgs_lock);
3606 * make sure all the block groups on our dirty list actually
3609 btrfs_create_pending_block_groups(trans);
3612 path = btrfs_alloc_path();
3618 * cache_write_mutex is here only to save us from balance or automatic
3619 * removal of empty block groups deleting this block group while we are
3620 * writing out the cache
3622 mutex_lock(&trans->transaction->cache_write_mutex);
3623 while (!list_empty(&dirty)) {
3624 bool drop_reserve = true;
3626 cache = list_first_entry(&dirty,
3627 struct btrfs_block_group_cache,
3630 * this can happen if something re-dirties a block
3631 * group that is already under IO. Just wait for it to
3632 * finish and then do it all again
3634 if (!list_empty(&cache->io_list)) {
3635 list_del_init(&cache->io_list);
3636 btrfs_wait_cache_io(trans, cache, path);
3637 btrfs_put_block_group(cache);
3642 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3643 * if it should update the cache_state. Don't delete
3644 * until after we wait.
3646 * Since we're not running in the commit critical section
3647 * we need the dirty_bgs_lock to protect from update_block_group
3649 spin_lock(&cur_trans->dirty_bgs_lock);
3650 list_del_init(&cache->dirty_list);
3651 spin_unlock(&cur_trans->dirty_bgs_lock);
3655 cache_save_setup(cache, trans, path);
3657 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3658 cache->io_ctl.inode = NULL;
3659 ret = btrfs_write_out_cache(fs_info, trans,
3661 if (ret == 0 && cache->io_ctl.inode) {
3666 * The cache_write_mutex is protecting the
3667 * io_list, also refer to the definition of
3668 * btrfs_transaction::io_bgs for more details
3670 list_add_tail(&cache->io_list, io);
3673 * if we failed to write the cache, the
3674 * generation will be bad and life goes on
3680 ret = write_one_cache_group(trans, fs_info,
3683 * Our block group might still be attached to the list
3684 * of new block groups in the transaction handle of some
3685 * other task (struct btrfs_trans_handle->new_bgs). This
3686 * means its block group item isn't yet in the extent
3687 * tree. If this happens ignore the error, as we will
3688 * try again later in the critical section of the
3689 * transaction commit.
3691 if (ret == -ENOENT) {
3693 spin_lock(&cur_trans->dirty_bgs_lock);
3694 if (list_empty(&cache->dirty_list)) {
3695 list_add_tail(&cache->dirty_list,
3696 &cur_trans->dirty_bgs);
3697 btrfs_get_block_group(cache);
3698 drop_reserve = false;
3700 spin_unlock(&cur_trans->dirty_bgs_lock);
3702 btrfs_abort_transaction(trans, ret);
3706 /* if it's not on the io list, we need to put the block group */
3708 btrfs_put_block_group(cache);
3710 btrfs_delayed_refs_rsv_release(fs_info, 1);
3716 * Avoid blocking other tasks for too long. It might even save
3717 * us from writing caches for block groups that are going to be
3720 mutex_unlock(&trans->transaction->cache_write_mutex);
3721 mutex_lock(&trans->transaction->cache_write_mutex);
3723 mutex_unlock(&trans->transaction->cache_write_mutex);
3726 * go through delayed refs for all the stuff we've just kicked off
3727 * and then loop back (just once)
3729 ret = btrfs_run_delayed_refs(trans, 0);
3730 if (!ret && loops == 0) {
3732 spin_lock(&cur_trans->dirty_bgs_lock);
3733 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3735 * dirty_bgs_lock protects us from concurrent block group
3736 * deletes too (not just cache_write_mutex).
3738 if (!list_empty(&dirty)) {
3739 spin_unlock(&cur_trans->dirty_bgs_lock);
3742 spin_unlock(&cur_trans->dirty_bgs_lock);
3743 } else if (ret < 0) {
3744 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3747 btrfs_free_path(path);
3751 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3752 struct btrfs_fs_info *fs_info)
3754 struct btrfs_block_group_cache *cache;
3755 struct btrfs_transaction *cur_trans = trans->transaction;
3758 struct btrfs_path *path;
3759 struct list_head *io = &cur_trans->io_bgs;
3760 int num_started = 0;
3762 path = btrfs_alloc_path();
3767 * Even though we are in the critical section of the transaction commit,
3768 * we can still have concurrent tasks adding elements to this
3769 * transaction's list of dirty block groups. These tasks correspond to
3770 * endio free space workers started when writeback finishes for a
3771 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3772 * allocate new block groups as a result of COWing nodes of the root
3773 * tree when updating the free space inode. The writeback for the space
3774 * caches is triggered by an earlier call to
3775 * btrfs_start_dirty_block_groups() and iterations of the following
3777 * Also we want to do the cache_save_setup first and then run the
3778 * delayed refs to make sure we have the best chance at doing this all
3781 spin_lock(&cur_trans->dirty_bgs_lock);
3782 while (!list_empty(&cur_trans->dirty_bgs)) {
3783 cache = list_first_entry(&cur_trans->dirty_bgs,
3784 struct btrfs_block_group_cache,
3788 * this can happen if cache_save_setup re-dirties a block
3789 * group that is already under IO. Just wait for it to
3790 * finish and then do it all again
3792 if (!list_empty(&cache->io_list)) {
3793 spin_unlock(&cur_trans->dirty_bgs_lock);
3794 list_del_init(&cache->io_list);
3795 btrfs_wait_cache_io(trans, cache, path);
3796 btrfs_put_block_group(cache);
3797 spin_lock(&cur_trans->dirty_bgs_lock);
3801 * don't remove from the dirty list until after we've waited
3804 list_del_init(&cache->dirty_list);
3805 spin_unlock(&cur_trans->dirty_bgs_lock);
3808 cache_save_setup(cache, trans, path);
3811 ret = btrfs_run_delayed_refs(trans,
3812 (unsigned long) -1);
3814 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3815 cache->io_ctl.inode = NULL;
3816 ret = btrfs_write_out_cache(fs_info, trans,
3818 if (ret == 0 && cache->io_ctl.inode) {
3821 list_add_tail(&cache->io_list, io);
3824 * if we failed to write the cache, the
3825 * generation will be bad and life goes on
3831 ret = write_one_cache_group(trans, fs_info,
3834 * One of the free space endio workers might have
3835 * created a new block group while updating a free space
3836 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3837 * and hasn't released its transaction handle yet, in
3838 * which case the new block group is still attached to
3839 * its transaction handle and its creation has not
3840 * finished yet (no block group item in the extent tree
3841 * yet, etc). If this is the case, wait for all free
3842 * space endio workers to finish and retry. This is a
3843 * a very rare case so no need for a more efficient and
3846 if (ret == -ENOENT) {
3847 wait_event(cur_trans->writer_wait,
3848 atomic_read(&cur_trans->num_writers) == 1);
3849 ret = write_one_cache_group(trans, fs_info,
3853 btrfs_abort_transaction(trans, ret);
3856 /* if its not on the io list, we need to put the block group */
3858 btrfs_put_block_group(cache);
3859 btrfs_delayed_refs_rsv_release(fs_info, 1);
3860 spin_lock(&cur_trans->dirty_bgs_lock);
3862 spin_unlock(&cur_trans->dirty_bgs_lock);
3865 * Refer to the definition of io_bgs member for details why it's safe
3866 * to use it without any locking
3868 while (!list_empty(io)) {
3869 cache = list_first_entry(io, struct btrfs_block_group_cache,
3871 list_del_init(&cache->io_list);
3872 btrfs_wait_cache_io(trans, cache, path);
3873 btrfs_put_block_group(cache);
3876 btrfs_free_path(path);
3880 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3882 struct btrfs_block_group_cache *block_group;
3885 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3886 if (!block_group || block_group->ro)
3889 btrfs_put_block_group(block_group);
3893 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3895 struct btrfs_block_group_cache *bg;
3898 bg = btrfs_lookup_block_group(fs_info, bytenr);
3902 spin_lock(&bg->lock);
3906 atomic_inc(&bg->nocow_writers);
3907 spin_unlock(&bg->lock);
3909 /* no put on block group, done by btrfs_dec_nocow_writers */
3911 btrfs_put_block_group(bg);
3917 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3919 struct btrfs_block_group_cache *bg;
3921 bg = btrfs_lookup_block_group(fs_info, bytenr);
3923 if (atomic_dec_and_test(&bg->nocow_writers))
3924 wake_up_var(&bg->nocow_writers);
3926 * Once for our lookup and once for the lookup done by a previous call
3927 * to btrfs_inc_nocow_writers()
3929 btrfs_put_block_group(bg);
3930 btrfs_put_block_group(bg);
3933 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3935 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3938 static const char *alloc_name(u64 flags)
3941 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3943 case BTRFS_BLOCK_GROUP_METADATA:
3945 case BTRFS_BLOCK_GROUP_DATA:
3947 case BTRFS_BLOCK_GROUP_SYSTEM:
3951 return "invalid-combination";
3955 static int create_space_info(struct btrfs_fs_info *info, u64 flags)
3958 struct btrfs_space_info *space_info;
3962 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3966 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3973 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3974 INIT_LIST_HEAD(&space_info->block_groups[i]);
3975 init_rwsem(&space_info->groups_sem);
3976 spin_lock_init(&space_info->lock);
3977 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3978 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3979 init_waitqueue_head(&space_info->wait);
3980 INIT_LIST_HEAD(&space_info->ro_bgs);
3981 INIT_LIST_HEAD(&space_info->tickets);
3982 INIT_LIST_HEAD(&space_info->priority_tickets);
3984 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3985 info->space_info_kobj, "%s",
3986 alloc_name(space_info->flags));
3988 percpu_counter_destroy(&space_info->total_bytes_pinned);
3993 list_add_rcu(&space_info->list, &info->space_info);
3994 if (flags & BTRFS_BLOCK_GROUP_DATA)
3995 info->data_sinfo = space_info;
4000 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4001 u64 total_bytes, u64 bytes_used,
4003 struct btrfs_space_info **space_info)
4005 struct btrfs_space_info *found;
4008 factor = btrfs_bg_type_to_factor(flags);
4010 found = __find_space_info(info, flags);
4012 spin_lock(&found->lock);
4013 found->total_bytes += total_bytes;
4014 found->disk_total += total_bytes * factor;
4015 found->bytes_used += bytes_used;
4016 found->disk_used += bytes_used * factor;
4017 found->bytes_readonly += bytes_readonly;
4018 if (total_bytes > 0)
4020 space_info_add_new_bytes(info, found, total_bytes -
4021 bytes_used - bytes_readonly);
4022 spin_unlock(&found->lock);
4023 *space_info = found;
4026 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4028 u64 extra_flags = chunk_to_extended(flags) &
4029 BTRFS_EXTENDED_PROFILE_MASK;
4031 write_seqlock(&fs_info->profiles_lock);
4032 if (flags & BTRFS_BLOCK_GROUP_DATA)
4033 fs_info->avail_data_alloc_bits |= extra_flags;
4034 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4035 fs_info->avail_metadata_alloc_bits |= extra_flags;
4036 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4037 fs_info->avail_system_alloc_bits |= extra_flags;
4038 write_sequnlock(&fs_info->profiles_lock);
4042 * returns target flags in extended format or 0 if restripe for this
4043 * chunk_type is not in progress
4045 * should be called with balance_lock held
4047 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4049 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4055 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4056 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4057 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4058 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4059 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4060 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4061 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4062 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4063 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4070 * @flags: available profiles in extended format (see ctree.h)
4072 * Returns reduced profile in chunk format. If profile changing is in
4073 * progress (either running or paused) picks the target profile (if it's
4074 * already available), otherwise falls back to plain reducing.
4076 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4078 u64 num_devices = fs_info->fs_devices->rw_devices;
4084 * see if restripe for this chunk_type is in progress, if so
4085 * try to reduce to the target profile
4087 spin_lock(&fs_info->balance_lock);
4088 target = get_restripe_target(fs_info, flags);
4090 /* pick target profile only if it's already available */
4091 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4092 spin_unlock(&fs_info->balance_lock);
4093 return extended_to_chunk(target);
4096 spin_unlock(&fs_info->balance_lock);
4098 /* First, mask out the RAID levels which aren't possible */
4099 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4100 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4101 allowed |= btrfs_raid_array[raid_type].bg_flag;
4105 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4106 allowed = BTRFS_BLOCK_GROUP_RAID6;
4107 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4108 allowed = BTRFS_BLOCK_GROUP_RAID5;
4109 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4110 allowed = BTRFS_BLOCK_GROUP_RAID10;
4111 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4112 allowed = BTRFS_BLOCK_GROUP_RAID1;
4113 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4114 allowed = BTRFS_BLOCK_GROUP_RAID0;
4116 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4118 return extended_to_chunk(flags | allowed);
4121 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4128 seq = read_seqbegin(&fs_info->profiles_lock);
4130 if (flags & BTRFS_BLOCK_GROUP_DATA)
4131 flags |= fs_info->avail_data_alloc_bits;
4132 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4133 flags |= fs_info->avail_system_alloc_bits;
4134 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4135 flags |= fs_info->avail_metadata_alloc_bits;
4136 } while (read_seqretry(&fs_info->profiles_lock, seq));
4138 return btrfs_reduce_alloc_profile(fs_info, flags);
4141 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4143 struct btrfs_fs_info *fs_info = root->fs_info;
4148 flags = BTRFS_BLOCK_GROUP_DATA;
4149 else if (root == fs_info->chunk_root)
4150 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4152 flags = BTRFS_BLOCK_GROUP_METADATA;
4154 ret = get_alloc_profile(fs_info, flags);
4158 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4160 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4163 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4165 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4168 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4170 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4173 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4174 bool may_use_included)
4177 return s_info->bytes_used + s_info->bytes_reserved +
4178 s_info->bytes_pinned + s_info->bytes_readonly +
4179 (may_use_included ? s_info->bytes_may_use : 0);
4182 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4184 struct btrfs_root *root = inode->root;
4185 struct btrfs_fs_info *fs_info = root->fs_info;
4186 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4189 int need_commit = 2;
4190 int have_pinned_space;
4192 /* make sure bytes are sectorsize aligned */
4193 bytes = ALIGN(bytes, fs_info->sectorsize);
4195 if (btrfs_is_free_space_inode(inode)) {
4197 ASSERT(current->journal_info);
4201 /* make sure we have enough space to handle the data first */
4202 spin_lock(&data_sinfo->lock);
4203 used = btrfs_space_info_used(data_sinfo, true);
4205 if (used + bytes > data_sinfo->total_bytes) {
4206 struct btrfs_trans_handle *trans;
4209 * if we don't have enough free bytes in this space then we need
4210 * to alloc a new chunk.
4212 if (!data_sinfo->full) {
4215 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4216 spin_unlock(&data_sinfo->lock);
4218 alloc_target = btrfs_data_alloc_profile(fs_info);
4220 * It is ugly that we don't call nolock join
4221 * transaction for the free space inode case here.
4222 * But it is safe because we only do the data space
4223 * reservation for the free space cache in the
4224 * transaction context, the common join transaction
4225 * just increase the counter of the current transaction
4226 * handler, doesn't try to acquire the trans_lock of
4229 trans = btrfs_join_transaction(root);
4231 return PTR_ERR(trans);
4233 ret = do_chunk_alloc(trans, alloc_target,
4234 CHUNK_ALLOC_NO_FORCE);
4235 btrfs_end_transaction(trans);
4240 have_pinned_space = 1;
4249 * If we don't have enough pinned space to deal with this
4250 * allocation, and no removed chunk in current transaction,
4251 * don't bother committing the transaction.
4253 have_pinned_space = __percpu_counter_compare(
4254 &data_sinfo->total_bytes_pinned,
4255 used + bytes - data_sinfo->total_bytes,
4256 BTRFS_TOTAL_BYTES_PINNED_BATCH);
4257 spin_unlock(&data_sinfo->lock);
4259 /* commit the current transaction and try again */
4264 if (need_commit > 0) {
4265 btrfs_start_delalloc_roots(fs_info, -1);
4266 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4270 trans = btrfs_join_transaction(root);
4272 return PTR_ERR(trans);
4273 if (have_pinned_space >= 0 ||
4274 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4275 &trans->transaction->flags) ||
4277 ret = btrfs_commit_transaction(trans);
4281 * The cleaner kthread might still be doing iput
4282 * operations. Wait for it to finish so that
4283 * more space is released.
4285 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4286 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4289 btrfs_end_transaction(trans);
4293 trace_btrfs_space_reservation(fs_info,
4294 "space_info:enospc",
4295 data_sinfo->flags, bytes, 1);
4298 update_bytes_may_use(data_sinfo, bytes);
4299 trace_btrfs_space_reservation(fs_info, "space_info",
4300 data_sinfo->flags, bytes, 1);
4301 spin_unlock(&data_sinfo->lock);
4306 int btrfs_check_data_free_space(struct inode *inode,
4307 struct extent_changeset **reserved, u64 start, u64 len)
4309 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4312 /* align the range */
4313 len = round_up(start + len, fs_info->sectorsize) -
4314 round_down(start, fs_info->sectorsize);
4315 start = round_down(start, fs_info->sectorsize);
4317 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4321 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4322 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4324 btrfs_free_reserved_data_space_noquota(inode, start, len);
4331 * Called if we need to clear a data reservation for this inode
4332 * Normally in a error case.
4334 * This one will *NOT* use accurate qgroup reserved space API, just for case
4335 * which we can't sleep and is sure it won't affect qgroup reserved space.
4336 * Like clear_bit_hook().
4338 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4341 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4342 struct btrfs_space_info *data_sinfo;
4344 /* Make sure the range is aligned to sectorsize */
4345 len = round_up(start + len, fs_info->sectorsize) -
4346 round_down(start, fs_info->sectorsize);
4347 start = round_down(start, fs_info->sectorsize);
4349 data_sinfo = fs_info->data_sinfo;
4350 spin_lock(&data_sinfo->lock);
4351 update_bytes_may_use(data_sinfo, -len);
4352 trace_btrfs_space_reservation(fs_info, "space_info",
4353 data_sinfo->flags, len, 0);
4354 spin_unlock(&data_sinfo->lock);
4358 * Called if we need to clear a data reservation for this inode
4359 * Normally in a error case.
4361 * This one will handle the per-inode data rsv map for accurate reserved
4364 void btrfs_free_reserved_data_space(struct inode *inode,
4365 struct extent_changeset *reserved, u64 start, u64 len)
4367 struct btrfs_root *root = BTRFS_I(inode)->root;
4369 /* Make sure the range is aligned to sectorsize */
4370 len = round_up(start + len, root->fs_info->sectorsize) -
4371 round_down(start, root->fs_info->sectorsize);
4372 start = round_down(start, root->fs_info->sectorsize);
4374 btrfs_free_reserved_data_space_noquota(inode, start, len);
4375 btrfs_qgroup_free_data(inode, reserved, start, len);
4378 static void force_metadata_allocation(struct btrfs_fs_info *info)
4380 struct list_head *head = &info->space_info;
4381 struct btrfs_space_info *found;
4384 list_for_each_entry_rcu(found, head, list) {
4385 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4386 found->force_alloc = CHUNK_ALLOC_FORCE;
4391 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4393 return (global->size << 1);
4396 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4397 struct btrfs_space_info *sinfo, int force)
4399 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4400 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4403 if (force == CHUNK_ALLOC_FORCE)
4407 * We need to take into account the global rsv because for all intents
4408 * and purposes it's used space. Don't worry about locking the
4409 * global_rsv, it doesn't change except when the transaction commits.
4411 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4412 bytes_used += calc_global_rsv_need_space(global_rsv);
4415 * in limited mode, we want to have some free space up to
4416 * about 1% of the FS size.
4418 if (force == CHUNK_ALLOC_LIMITED) {
4419 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4420 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4422 if (sinfo->total_bytes - bytes_used < thresh)
4426 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4431 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4435 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4436 BTRFS_BLOCK_GROUP_RAID0 |
4437 BTRFS_BLOCK_GROUP_RAID5 |
4438 BTRFS_BLOCK_GROUP_RAID6))
4439 num_dev = fs_info->fs_devices->rw_devices;
4440 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4443 num_dev = 1; /* DUP or single */
4449 * If @is_allocation is true, reserve space in the system space info necessary
4450 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4453 void check_system_chunk(struct btrfs_trans_handle *trans, u64 type)
4455 struct btrfs_fs_info *fs_info = trans->fs_info;
4456 struct btrfs_space_info *info;
4463 * Needed because we can end up allocating a system chunk and for an
4464 * atomic and race free space reservation in the chunk block reserve.
4466 lockdep_assert_held(&fs_info->chunk_mutex);
4468 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4469 spin_lock(&info->lock);
4470 left = info->total_bytes - btrfs_space_info_used(info, true);
4471 spin_unlock(&info->lock);
4473 num_devs = get_profile_num_devs(fs_info, type);
4475 /* num_devs device items to update and 1 chunk item to add or remove */
4476 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4477 btrfs_calc_trans_metadata_size(fs_info, 1);
4479 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4480 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4481 left, thresh, type);
4482 dump_space_info(fs_info, info, 0, 0);
4485 if (left < thresh) {
4486 u64 flags = btrfs_system_alloc_profile(fs_info);
4489 * Ignore failure to create system chunk. We might end up not
4490 * needing it, as we might not need to COW all nodes/leafs from
4491 * the paths we visit in the chunk tree (they were already COWed
4492 * or created in the current transaction for example).
4494 ret = btrfs_alloc_chunk(trans, flags);
4498 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4499 &fs_info->chunk_block_rsv,
4500 thresh, BTRFS_RESERVE_NO_FLUSH);
4502 trans->chunk_bytes_reserved += thresh;
4507 * If force is CHUNK_ALLOC_FORCE:
4508 * - return 1 if it successfully allocates a chunk,
4509 * - return errors including -ENOSPC otherwise.
4510 * If force is NOT CHUNK_ALLOC_FORCE:
4511 * - return 0 if it doesn't need to allocate a new chunk,
4512 * - return 1 if it successfully allocates a chunk,
4513 * - return errors including -ENOSPC otherwise.
4515 static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags,
4518 struct btrfs_fs_info *fs_info = trans->fs_info;
4519 struct btrfs_space_info *space_info;
4520 bool wait_for_alloc = false;
4521 bool should_alloc = false;
4524 /* Don't re-enter if we're already allocating a chunk */
4525 if (trans->allocating_chunk)
4528 space_info = __find_space_info(fs_info, flags);
4532 spin_lock(&space_info->lock);
4533 if (force < space_info->force_alloc)
4534 force = space_info->force_alloc;
4535 should_alloc = should_alloc_chunk(fs_info, space_info, force);
4536 if (space_info->full) {
4537 /* No more free physical space */
4542 spin_unlock(&space_info->lock);
4544 } else if (!should_alloc) {
4545 spin_unlock(&space_info->lock);
4547 } else if (space_info->chunk_alloc) {
4549 * Someone is already allocating, so we need to block
4550 * until this someone is finished and then loop to
4551 * recheck if we should continue with our allocation
4554 wait_for_alloc = true;
4555 spin_unlock(&space_info->lock);
4556 mutex_lock(&fs_info->chunk_mutex);
4557 mutex_unlock(&fs_info->chunk_mutex);
4559 /* Proceed with allocation */
4560 space_info->chunk_alloc = 1;
4561 wait_for_alloc = false;
4562 spin_unlock(&space_info->lock);
4566 } while (wait_for_alloc);
4568 mutex_lock(&fs_info->chunk_mutex);
4569 trans->allocating_chunk = true;
4572 * If we have mixed data/metadata chunks we want to make sure we keep
4573 * allocating mixed chunks instead of individual chunks.
4575 if (btrfs_mixed_space_info(space_info))
4576 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4579 * if we're doing a data chunk, go ahead and make sure that
4580 * we keep a reasonable number of metadata chunks allocated in the
4583 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4584 fs_info->data_chunk_allocations++;
4585 if (!(fs_info->data_chunk_allocations %
4586 fs_info->metadata_ratio))
4587 force_metadata_allocation(fs_info);
4591 * Check if we have enough space in SYSTEM chunk because we may need
4592 * to update devices.
4594 check_system_chunk(trans, flags);
4596 ret = btrfs_alloc_chunk(trans, flags);
4597 trans->allocating_chunk = false;
4599 spin_lock(&space_info->lock);
4602 space_info->full = 1;
4607 space_info->max_extent_size = 0;
4610 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4612 space_info->chunk_alloc = 0;
4613 spin_unlock(&space_info->lock);
4614 mutex_unlock(&fs_info->chunk_mutex);
4616 * When we allocate a new chunk we reserve space in the chunk block
4617 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4618 * add new nodes/leafs to it if we end up needing to do it when
4619 * inserting the chunk item and updating device items as part of the
4620 * second phase of chunk allocation, performed by
4621 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4622 * large number of new block groups to create in our transaction
4623 * handle's new_bgs list to avoid exhausting the chunk block reserve
4624 * in extreme cases - like having a single transaction create many new
4625 * block groups when starting to write out the free space caches of all
4626 * the block groups that were made dirty during the lifetime of the
4629 if (trans->chunk_bytes_reserved >= (u64)SZ_2M)
4630 btrfs_create_pending_block_groups(trans);
4635 static int can_overcommit(struct btrfs_fs_info *fs_info,
4636 struct btrfs_space_info *space_info, u64 bytes,
4637 enum btrfs_reserve_flush_enum flush,
4640 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4647 /* Don't overcommit when in mixed mode. */
4648 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4652 profile = btrfs_system_alloc_profile(fs_info);
4654 profile = btrfs_metadata_alloc_profile(fs_info);
4656 used = btrfs_space_info_used(space_info, false);
4659 * We only want to allow over committing if we have lots of actual space
4660 * free, but if we don't have enough space to handle the global reserve
4661 * space then we could end up having a real enospc problem when trying
4662 * to allocate a chunk or some other such important allocation.
4664 spin_lock(&global_rsv->lock);
4665 space_size = calc_global_rsv_need_space(global_rsv);
4666 spin_unlock(&global_rsv->lock);
4667 if (used + space_size >= space_info->total_bytes)
4670 used += space_info->bytes_may_use;
4672 avail = atomic64_read(&fs_info->free_chunk_space);
4675 * If we have dup, raid1 or raid10 then only half of the free
4676 * space is actually usable. For raid56, the space info used
4677 * doesn't include the parity drive, so we don't have to
4680 factor = btrfs_bg_type_to_factor(profile);
4681 avail = div_u64(avail, factor);
4684 * If we aren't flushing all things, let us overcommit up to
4685 * 1/2th of the space. If we can flush, don't let us overcommit
4686 * too much, let it overcommit up to 1/8 of the space.
4688 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4693 if (used + bytes < space_info->total_bytes + avail)
4698 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4699 unsigned long nr_pages, int nr_items)
4701 struct super_block *sb = fs_info->sb;
4703 if (down_read_trylock(&sb->s_umount)) {
4704 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4705 up_read(&sb->s_umount);
4708 * We needn't worry the filesystem going from r/w to r/o though
4709 * we don't acquire ->s_umount mutex, because the filesystem
4710 * should guarantee the delalloc inodes list be empty after
4711 * the filesystem is readonly(all dirty pages are written to
4714 btrfs_start_delalloc_roots(fs_info, nr_items);
4715 if (!current->journal_info)
4716 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4720 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4726 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4727 nr = div64_u64(to_reclaim, bytes);
4733 #define EXTENT_SIZE_PER_ITEM SZ_256K
4736 * shrink metadata reservation for delalloc
4738 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4739 u64 orig, bool wait_ordered)
4741 struct btrfs_space_info *space_info;
4742 struct btrfs_trans_handle *trans;
4747 unsigned long nr_pages;
4750 /* Calc the number of the pages we need flush for space reservation */
4751 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4752 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4754 trans = (struct btrfs_trans_handle *)current->journal_info;
4755 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4757 delalloc_bytes = percpu_counter_sum_positive(
4758 &fs_info->delalloc_bytes);
4759 if (delalloc_bytes == 0) {
4763 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4768 while (delalloc_bytes && loops < 3) {
4769 max_reclaim = min(delalloc_bytes, to_reclaim);
4770 nr_pages = max_reclaim >> PAGE_SHIFT;
4771 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4773 * We need to wait for the async pages to actually start before
4776 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4780 if (max_reclaim <= nr_pages)
4783 max_reclaim -= nr_pages;
4785 wait_event(fs_info->async_submit_wait,
4786 atomic_read(&fs_info->async_delalloc_pages) <=
4789 spin_lock(&space_info->lock);
4790 if (list_empty(&space_info->tickets) &&
4791 list_empty(&space_info->priority_tickets)) {
4792 spin_unlock(&space_info->lock);
4795 spin_unlock(&space_info->lock);
4798 if (wait_ordered && !trans) {
4799 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4801 time_left = schedule_timeout_killable(1);
4805 delalloc_bytes = percpu_counter_sum_positive(
4806 &fs_info->delalloc_bytes);
4810 struct reserve_ticket {
4813 struct list_head list;
4814 wait_queue_head_t wait;
4818 * maybe_commit_transaction - possibly commit the transaction if its ok to
4819 * @root - the root we're allocating for
4820 * @bytes - the number of bytes we want to reserve
4821 * @force - force the commit
4823 * This will check to make sure that committing the transaction will actually
4824 * get us somewhere and then commit the transaction if it does. Otherwise it
4825 * will return -ENOSPC.
4827 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4828 struct btrfs_space_info *space_info)
4830 struct reserve_ticket *ticket = NULL;
4831 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4832 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
4833 struct btrfs_trans_handle *trans;
4835 u64 reclaim_bytes = 0;
4837 trans = (struct btrfs_trans_handle *)current->journal_info;
4841 spin_lock(&space_info->lock);
4842 if (!list_empty(&space_info->priority_tickets))
4843 ticket = list_first_entry(&space_info->priority_tickets,
4844 struct reserve_ticket, list);
4845 else if (!list_empty(&space_info->tickets))
4846 ticket = list_first_entry(&space_info->tickets,
4847 struct reserve_ticket, list);
4848 bytes_needed = (ticket) ? ticket->bytes : 0;
4849 spin_unlock(&space_info->lock);
4854 /* See if there is enough pinned space to make this reservation */
4855 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4857 BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0)
4861 * See if there is some space in the delayed insertion reservation for
4864 if (space_info != delayed_rsv->space_info)
4867 spin_lock(&delayed_rsv->lock);
4868 reclaim_bytes += delayed_rsv->reserved;
4869 spin_unlock(&delayed_rsv->lock);
4871 spin_lock(&delayed_refs_rsv->lock);
4872 reclaim_bytes += delayed_refs_rsv->reserved;
4873 spin_unlock(&delayed_refs_rsv->lock);
4874 if (reclaim_bytes >= bytes_needed)
4876 bytes_needed -= reclaim_bytes;
4878 if (__percpu_counter_compare(&space_info->total_bytes_pinned,
4880 BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) {
4885 trans = btrfs_join_transaction(fs_info->extent_root);
4889 return btrfs_commit_transaction(trans);
4893 * Try to flush some data based on policy set by @state. This is only advisory
4894 * and may fail for various reasons. The caller is supposed to examine the
4895 * state of @space_info to detect the outcome.
4897 static void flush_space(struct btrfs_fs_info *fs_info,
4898 struct btrfs_space_info *space_info, u64 num_bytes,
4901 struct btrfs_root *root = fs_info->extent_root;
4902 struct btrfs_trans_handle *trans;
4907 case FLUSH_DELAYED_ITEMS_NR:
4908 case FLUSH_DELAYED_ITEMS:
4909 if (state == FLUSH_DELAYED_ITEMS_NR)
4910 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4914 trans = btrfs_join_transaction(root);
4915 if (IS_ERR(trans)) {
4916 ret = PTR_ERR(trans);
4919 ret = btrfs_run_delayed_items_nr(trans, nr);
4920 btrfs_end_transaction(trans);
4922 case FLUSH_DELALLOC:
4923 case FLUSH_DELALLOC_WAIT:
4924 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4925 state == FLUSH_DELALLOC_WAIT);
4927 case FLUSH_DELAYED_REFS_NR:
4928 case FLUSH_DELAYED_REFS:
4929 trans = btrfs_join_transaction(root);
4930 if (IS_ERR(trans)) {
4931 ret = PTR_ERR(trans);
4934 if (state == FLUSH_DELAYED_REFS_NR)
4935 nr = calc_reclaim_items_nr(fs_info, num_bytes);
4938 btrfs_run_delayed_refs(trans, nr);
4939 btrfs_end_transaction(trans);
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)
4956 * If we have pending delayed iputs then we could free up a
4957 * bunch of pinned space, so make sure we run the iputs before
4958 * we do our pinned bytes check below.
4960 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4961 btrfs_run_delayed_iputs(fs_info);
4962 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4964 ret = may_commit_transaction(fs_info, space_info);
4971 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4977 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4978 struct btrfs_space_info *space_info,
4981 struct reserve_ticket *ticket;
4986 list_for_each_entry(ticket, &space_info->tickets, list)
4987 to_reclaim += ticket->bytes;
4988 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4989 to_reclaim += ticket->bytes;
4993 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4994 if (can_overcommit(fs_info, space_info, to_reclaim,
4995 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4998 used = btrfs_space_info_used(space_info, true);
5000 if (can_overcommit(fs_info, space_info, SZ_1M,
5001 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5002 expected = div_factor_fine(space_info->total_bytes, 95);
5004 expected = div_factor_fine(space_info->total_bytes, 90);
5006 if (used > expected)
5007 to_reclaim = used - expected;
5010 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5011 space_info->bytes_reserved);
5015 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5016 struct btrfs_space_info *space_info,
5017 u64 used, bool system_chunk)
5019 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5021 /* If we're just plain full then async reclaim just slows us down. */
5022 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5025 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5029 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5030 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5033 static void wake_all_tickets(struct list_head *head)
5035 struct reserve_ticket *ticket;
5037 while (!list_empty(head)) {
5038 ticket = list_first_entry(head, struct reserve_ticket, list);
5039 list_del_init(&ticket->list);
5040 ticket->error = -ENOSPC;
5041 wake_up(&ticket->wait);
5046 * This is for normal flushers, we can wait all goddamned day if we want to. We
5047 * will loop and continuously try to flush as long as we are making progress.
5048 * We count progress as clearing off tickets each time we have to loop.
5050 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5052 struct btrfs_fs_info *fs_info;
5053 struct btrfs_space_info *space_info;
5056 int commit_cycles = 0;
5057 u64 last_tickets_id;
5059 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5060 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5062 spin_lock(&space_info->lock);
5063 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5066 space_info->flush = 0;
5067 spin_unlock(&space_info->lock);
5070 last_tickets_id = space_info->tickets_id;
5071 spin_unlock(&space_info->lock);
5073 flush_state = FLUSH_DELAYED_ITEMS_NR;
5075 flush_space(fs_info, space_info, to_reclaim, flush_state);
5076 spin_lock(&space_info->lock);
5077 if (list_empty(&space_info->tickets)) {
5078 space_info->flush = 0;
5079 spin_unlock(&space_info->lock);
5082 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5085 if (last_tickets_id == space_info->tickets_id) {
5088 last_tickets_id = space_info->tickets_id;
5089 flush_state = FLUSH_DELAYED_ITEMS_NR;
5094 if (flush_state > COMMIT_TRANS) {
5096 if (commit_cycles > 2) {
5097 wake_all_tickets(&space_info->tickets);
5098 space_info->flush = 0;
5100 flush_state = FLUSH_DELAYED_ITEMS_NR;
5103 spin_unlock(&space_info->lock);
5104 } while (flush_state <= COMMIT_TRANS);
5107 void btrfs_init_async_reclaim_work(struct work_struct *work)
5109 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5112 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5113 struct btrfs_space_info *space_info,
5114 struct reserve_ticket *ticket)
5117 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5119 spin_lock(&space_info->lock);
5120 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5123 spin_unlock(&space_info->lock);
5126 spin_unlock(&space_info->lock);
5129 flush_space(fs_info, space_info, to_reclaim, flush_state);
5131 spin_lock(&space_info->lock);
5132 if (ticket->bytes == 0) {
5133 spin_unlock(&space_info->lock);
5136 spin_unlock(&space_info->lock);
5139 * Priority flushers can't wait on delalloc without
5142 if (flush_state == FLUSH_DELALLOC ||
5143 flush_state == FLUSH_DELALLOC_WAIT)
5144 flush_state = ALLOC_CHUNK;
5145 } while (flush_state < COMMIT_TRANS);
5148 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5149 struct btrfs_space_info *space_info,
5150 struct reserve_ticket *ticket, u64 orig_bytes)
5156 spin_lock(&space_info->lock);
5157 while (ticket->bytes > 0 && ticket->error == 0) {
5158 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5163 spin_unlock(&space_info->lock);
5167 finish_wait(&ticket->wait, &wait);
5168 spin_lock(&space_info->lock);
5171 ret = ticket->error;
5172 if (!list_empty(&ticket->list))
5173 list_del_init(&ticket->list);
5174 if (ticket->bytes && ticket->bytes < orig_bytes) {
5175 u64 num_bytes = orig_bytes - ticket->bytes;
5176 update_bytes_may_use(space_info, -num_bytes);
5177 trace_btrfs_space_reservation(fs_info, "space_info",
5178 space_info->flags, num_bytes, 0);
5180 spin_unlock(&space_info->lock);
5186 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5187 * @root - the root we're allocating for
5188 * @space_info - the space info we want to allocate from
5189 * @orig_bytes - the number of bytes we want
5190 * @flush - whether or not we can flush to make our reservation
5192 * This will reserve orig_bytes number of bytes from the space info associated
5193 * with the block_rsv. If there is not enough space it will make an attempt to
5194 * flush out space to make room. It will do this by flushing delalloc if
5195 * possible or committing the transaction. If flush is 0 then no attempts to
5196 * regain reservations will be made and this will fail if there is not enough
5199 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5200 struct btrfs_space_info *space_info,
5202 enum btrfs_reserve_flush_enum flush,
5205 struct reserve_ticket ticket;
5210 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5212 spin_lock(&space_info->lock);
5214 used = btrfs_space_info_used(space_info, true);
5217 * If we have enough space then hooray, make our reservation and carry
5218 * on. If not see if we can overcommit, and if we can, hooray carry on.
5219 * If not things get more complicated.
5221 if (used + orig_bytes <= space_info->total_bytes) {
5222 update_bytes_may_use(space_info, orig_bytes);
5223 trace_btrfs_space_reservation(fs_info, "space_info",
5224 space_info->flags, orig_bytes, 1);
5226 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5228 update_bytes_may_use(space_info, orig_bytes);
5229 trace_btrfs_space_reservation(fs_info, "space_info",
5230 space_info->flags, orig_bytes, 1);
5235 * If we couldn't make a reservation then setup our reservation ticket
5236 * and kick the async worker if it's not already running.
5238 * If we are a priority flusher then we just need to add our ticket to
5239 * the list and we will do our own flushing further down.
5241 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5242 ticket.bytes = orig_bytes;
5244 init_waitqueue_head(&ticket.wait);
5245 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5246 list_add_tail(&ticket.list, &space_info->tickets);
5247 if (!space_info->flush) {
5248 space_info->flush = 1;
5249 trace_btrfs_trigger_flush(fs_info,
5253 queue_work(system_unbound_wq,
5254 &fs_info->async_reclaim_work);
5257 list_add_tail(&ticket.list,
5258 &space_info->priority_tickets);
5260 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5263 * We will do the space reservation dance during log replay,
5264 * which means we won't have fs_info->fs_root set, so don't do
5265 * the async reclaim as we will panic.
5267 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5268 need_do_async_reclaim(fs_info, space_info,
5269 used, system_chunk) &&
5270 !work_busy(&fs_info->async_reclaim_work)) {
5271 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5272 orig_bytes, flush, "preempt");
5273 queue_work(system_unbound_wq,
5274 &fs_info->async_reclaim_work);
5277 spin_unlock(&space_info->lock);
5278 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5281 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5282 return wait_reserve_ticket(fs_info, space_info, &ticket,
5286 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5287 spin_lock(&space_info->lock);
5289 if (ticket.bytes < orig_bytes) {
5290 u64 num_bytes = orig_bytes - ticket.bytes;
5291 update_bytes_may_use(space_info, -num_bytes);
5292 trace_btrfs_space_reservation(fs_info, "space_info",
5297 list_del_init(&ticket.list);
5300 spin_unlock(&space_info->lock);
5301 ASSERT(list_empty(&ticket.list));
5306 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5307 * @root - the root we're allocating for
5308 * @block_rsv - the block_rsv we're allocating for
5309 * @orig_bytes - the number of bytes we want
5310 * @flush - whether or not we can flush to make our reservation
5312 * This will reserve orig_bytes number of bytes from the space info associated
5313 * with the block_rsv. If there is not enough space it will make an attempt to
5314 * flush out space to make room. It will do this by flushing delalloc if
5315 * possible or committing the transaction. If flush is 0 then no attempts to
5316 * regain reservations will be made and this will fail if there is not enough
5319 static int reserve_metadata_bytes(struct btrfs_root *root,
5320 struct btrfs_block_rsv *block_rsv,
5322 enum btrfs_reserve_flush_enum flush)
5324 struct btrfs_fs_info *fs_info = root->fs_info;
5325 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5327 bool system_chunk = (root == fs_info->chunk_root);
5329 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5330 orig_bytes, flush, system_chunk);
5331 if (ret == -ENOSPC &&
5332 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5333 if (block_rsv != global_rsv &&
5334 !block_rsv_use_bytes(global_rsv, orig_bytes))
5337 if (ret == -ENOSPC) {
5338 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5339 block_rsv->space_info->flags,
5342 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5343 dump_space_info(fs_info, block_rsv->space_info,
5349 static struct btrfs_block_rsv *get_block_rsv(
5350 const struct btrfs_trans_handle *trans,
5351 const struct btrfs_root *root)
5353 struct btrfs_fs_info *fs_info = root->fs_info;
5354 struct btrfs_block_rsv *block_rsv = NULL;
5356 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5357 (root == fs_info->csum_root && trans->adding_csums) ||
5358 (root == fs_info->uuid_root))
5359 block_rsv = trans->block_rsv;
5362 block_rsv = root->block_rsv;
5365 block_rsv = &fs_info->empty_block_rsv;
5370 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5374 spin_lock(&block_rsv->lock);
5375 if (block_rsv->reserved >= num_bytes) {
5376 block_rsv->reserved -= num_bytes;
5377 if (block_rsv->reserved < block_rsv->size)
5378 block_rsv->full = 0;
5381 spin_unlock(&block_rsv->lock);
5385 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5386 u64 num_bytes, bool update_size)
5388 spin_lock(&block_rsv->lock);
5389 block_rsv->reserved += num_bytes;
5391 block_rsv->size += num_bytes;
5392 else if (block_rsv->reserved >= block_rsv->size)
5393 block_rsv->full = 1;
5394 spin_unlock(&block_rsv->lock);
5397 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5398 struct btrfs_block_rsv *dest, u64 num_bytes,
5401 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5404 if (global_rsv->space_info != dest->space_info)
5407 spin_lock(&global_rsv->lock);
5408 min_bytes = div_factor(global_rsv->size, min_factor);
5409 if (global_rsv->reserved < min_bytes + num_bytes) {
5410 spin_unlock(&global_rsv->lock);
5413 global_rsv->reserved -= num_bytes;
5414 if (global_rsv->reserved < global_rsv->size)
5415 global_rsv->full = 0;
5416 spin_unlock(&global_rsv->lock);
5418 block_rsv_add_bytes(dest, num_bytes, true);
5423 * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv.
5424 * @fs_info - the fs info for our fs.
5425 * @src - the source block rsv to transfer from.
5426 * @num_bytes - the number of bytes to transfer.
5428 * This transfers up to the num_bytes amount from the src rsv to the
5429 * delayed_refs_rsv. Any extra bytes are returned to the space info.
5431 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
5432 struct btrfs_block_rsv *src,
5435 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
5438 spin_lock(&src->lock);
5439 src->reserved -= num_bytes;
5440 src->size -= num_bytes;
5441 spin_unlock(&src->lock);
5443 spin_lock(&delayed_refs_rsv->lock);
5444 if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) {
5445 u64 delta = delayed_refs_rsv->size -
5446 delayed_refs_rsv->reserved;
5447 if (num_bytes > delta) {
5448 to_free = num_bytes - delta;
5452 to_free = num_bytes;
5457 delayed_refs_rsv->reserved += num_bytes;
5458 if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size)
5459 delayed_refs_rsv->full = 1;
5460 spin_unlock(&delayed_refs_rsv->lock);
5463 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5466 space_info_add_old_bytes(fs_info, delayed_refs_rsv->space_info,
5471 * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage.
5472 * @fs_info - the fs_info for our fs.
5473 * @flush - control how we can flush for this reservation.
5475 * This will refill the delayed block_rsv up to 1 items size worth of space and
5476 * will return -ENOSPC if we can't make the reservation.
5478 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
5479 enum btrfs_reserve_flush_enum flush)
5481 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5482 u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1);
5486 spin_lock(&block_rsv->lock);
5487 if (block_rsv->reserved < block_rsv->size) {
5488 num_bytes = block_rsv->size - block_rsv->reserved;
5489 num_bytes = min(num_bytes, limit);
5491 spin_unlock(&block_rsv->lock);
5496 ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv,
5500 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5501 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5507 * This is for space we already have accounted in space_info->bytes_may_use, so
5508 * basically when we're returning space from block_rsv's.
5510 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5511 struct btrfs_space_info *space_info,
5514 struct reserve_ticket *ticket;
5515 struct list_head *head;
5517 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5518 bool check_overcommit = false;
5520 spin_lock(&space_info->lock);
5521 head = &space_info->priority_tickets;
5524 * If we are over our limit then we need to check and see if we can
5525 * overcommit, and if we can't then we just need to free up our space
5526 * and not satisfy any requests.
5528 used = btrfs_space_info_used(space_info, true);
5529 if (used - num_bytes >= space_info->total_bytes)
5530 check_overcommit = true;
5532 while (!list_empty(head) && num_bytes) {
5533 ticket = list_first_entry(head, struct reserve_ticket,
5536 * We use 0 bytes because this space is already reserved, so
5537 * adding the ticket space would be a double count.
5539 if (check_overcommit &&
5540 !can_overcommit(fs_info, space_info, 0, flush, false))
5542 if (num_bytes >= ticket->bytes) {
5543 list_del_init(&ticket->list);
5544 num_bytes -= ticket->bytes;
5546 space_info->tickets_id++;
5547 wake_up(&ticket->wait);
5549 ticket->bytes -= num_bytes;
5554 if (num_bytes && head == &space_info->priority_tickets) {
5555 head = &space_info->tickets;
5556 flush = BTRFS_RESERVE_FLUSH_ALL;
5559 update_bytes_may_use(space_info, -num_bytes);
5560 trace_btrfs_space_reservation(fs_info, "space_info",
5561 space_info->flags, num_bytes, 0);
5562 spin_unlock(&space_info->lock);
5566 * This is for newly allocated space that isn't accounted in
5567 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5568 * we use this helper.
5570 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5571 struct btrfs_space_info *space_info,
5574 struct reserve_ticket *ticket;
5575 struct list_head *head = &space_info->priority_tickets;
5578 while (!list_empty(head) && num_bytes) {
5579 ticket = list_first_entry(head, struct reserve_ticket,
5581 if (num_bytes >= ticket->bytes) {
5582 trace_btrfs_space_reservation(fs_info, "space_info",
5585 list_del_init(&ticket->list);
5586 num_bytes -= ticket->bytes;
5587 update_bytes_may_use(space_info, ticket->bytes);
5589 space_info->tickets_id++;
5590 wake_up(&ticket->wait);
5592 trace_btrfs_space_reservation(fs_info, "space_info",
5595 update_bytes_may_use(space_info, num_bytes);
5596 ticket->bytes -= num_bytes;
5601 if (num_bytes && head == &space_info->priority_tickets) {
5602 head = &space_info->tickets;
5607 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5608 struct btrfs_block_rsv *block_rsv,
5609 struct btrfs_block_rsv *dest, u64 num_bytes,
5610 u64 *qgroup_to_release_ret)
5612 struct btrfs_space_info *space_info = block_rsv->space_info;
5613 u64 qgroup_to_release = 0;
5616 spin_lock(&block_rsv->lock);
5617 if (num_bytes == (u64)-1) {
5618 num_bytes = block_rsv->size;
5619 qgroup_to_release = block_rsv->qgroup_rsv_size;
5621 block_rsv->size -= num_bytes;
5622 if (block_rsv->reserved >= block_rsv->size) {
5623 num_bytes = block_rsv->reserved - block_rsv->size;
5624 block_rsv->reserved = block_rsv->size;
5625 block_rsv->full = 1;
5629 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5630 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5631 block_rsv->qgroup_rsv_size;
5632 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5634 qgroup_to_release = 0;
5636 spin_unlock(&block_rsv->lock);
5639 if (num_bytes > 0) {
5641 spin_lock(&dest->lock);
5645 bytes_to_add = dest->size - dest->reserved;
5646 bytes_to_add = min(num_bytes, bytes_to_add);
5647 dest->reserved += bytes_to_add;
5648 if (dest->reserved >= dest->size)
5650 num_bytes -= bytes_to_add;
5652 spin_unlock(&dest->lock);
5655 space_info_add_old_bytes(fs_info, space_info,
5658 if (qgroup_to_release_ret)
5659 *qgroup_to_release_ret = qgroup_to_release;
5663 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5664 struct btrfs_block_rsv *dst, u64 num_bytes,
5669 ret = block_rsv_use_bytes(src, num_bytes);
5673 block_rsv_add_bytes(dst, num_bytes, update_size);
5677 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5679 memset(rsv, 0, sizeof(*rsv));
5680 spin_lock_init(&rsv->lock);
5684 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5685 struct btrfs_block_rsv *rsv,
5686 unsigned short type)
5688 btrfs_init_block_rsv(rsv, type);
5689 rsv->space_info = __find_space_info(fs_info,
5690 BTRFS_BLOCK_GROUP_METADATA);
5693 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5694 unsigned short type)
5696 struct btrfs_block_rsv *block_rsv;
5698 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5702 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5706 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5707 struct btrfs_block_rsv *rsv)
5711 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5715 int btrfs_block_rsv_add(struct btrfs_root *root,
5716 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5717 enum btrfs_reserve_flush_enum flush)
5724 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5726 block_rsv_add_bytes(block_rsv, num_bytes, true);
5731 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5739 spin_lock(&block_rsv->lock);
5740 num_bytes = div_factor(block_rsv->size, min_factor);
5741 if (block_rsv->reserved >= num_bytes)
5743 spin_unlock(&block_rsv->lock);
5748 int btrfs_block_rsv_refill(struct btrfs_root *root,
5749 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5750 enum btrfs_reserve_flush_enum flush)
5758 spin_lock(&block_rsv->lock);
5759 num_bytes = min_reserved;
5760 if (block_rsv->reserved >= num_bytes)
5763 num_bytes -= block_rsv->reserved;
5764 spin_unlock(&block_rsv->lock);
5769 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5771 block_rsv_add_bytes(block_rsv, num_bytes, false);
5779 * btrfs_inode_rsv_refill - refill the inode block rsv.
5780 * @inode - the inode we are refilling.
5781 * @flush - the flushing restriction.
5783 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5784 * block_rsv->size as the minimum size. We'll either refill the missing amount
5785 * or return if we already have enough space. This will also handle the reserve
5786 * tracepoint for the reserved amount.
5788 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5789 enum btrfs_reserve_flush_enum flush)
5791 struct btrfs_root *root = inode->root;
5792 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5794 u64 qgroup_num_bytes = 0;
5797 spin_lock(&block_rsv->lock);
5798 if (block_rsv->reserved < block_rsv->size)
5799 num_bytes = block_rsv->size - block_rsv->reserved;
5800 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5801 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5802 block_rsv->qgroup_rsv_reserved;
5803 spin_unlock(&block_rsv->lock);
5808 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5811 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5813 block_rsv_add_bytes(block_rsv, num_bytes, false);
5814 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5815 btrfs_ino(inode), num_bytes, 1);
5817 /* Don't forget to increase qgroup_rsv_reserved */
5818 spin_lock(&block_rsv->lock);
5819 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5820 spin_unlock(&block_rsv->lock);
5822 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5826 static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5827 struct btrfs_block_rsv *block_rsv,
5828 u64 num_bytes, u64 *qgroup_to_release)
5830 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5831 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
5832 struct btrfs_block_rsv *target = delayed_rsv;
5834 if (target->full || target == block_rsv)
5835 target = global_rsv;
5837 if (block_rsv->space_info != target->space_info)
5840 return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes,
5844 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5845 struct btrfs_block_rsv *block_rsv,
5848 __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL);
5852 * btrfs_inode_rsv_release - release any excessive reservation.
5853 * @inode - the inode we need to release from.
5854 * @qgroup_free - free or convert qgroup meta.
5855 * Unlike normal operation, qgroup meta reservation needs to know if we are
5856 * freeing qgroup reservation or just converting it into per-trans. Normally
5857 * @qgroup_free is true for error handling, and false for normal release.
5859 * This is the same as btrfs_block_rsv_release, except that it handles the
5860 * tracepoint for the reservation.
5862 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5864 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5865 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5867 u64 qgroup_to_release = 0;
5870 * Since we statically set the block_rsv->size we just want to say we
5871 * are releasing 0 bytes, and then we'll just get the reservation over
5874 released = __btrfs_block_rsv_release(fs_info, block_rsv, 0,
5875 &qgroup_to_release);
5877 trace_btrfs_space_reservation(fs_info, "delalloc",
5878 btrfs_ino(inode), released, 0);
5880 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5882 btrfs_qgroup_convert_reserved_meta(inode->root,
5887 * btrfs_delayed_refs_rsv_release - release a ref head's reservation.
5888 * @fs_info - the fs_info for our fs.
5889 * @nr - the number of items to drop.
5891 * This drops the delayed ref head's count from the delayed refs rsv and frees
5892 * any excess reservation we had.
5894 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr)
5896 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv;
5897 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5898 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr);
5901 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv,
5904 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv",
5908 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5910 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5911 struct btrfs_space_info *sinfo = block_rsv->space_info;
5915 * The global block rsv is based on the size of the extent tree, the
5916 * checksum tree and the root tree. If the fs is empty we want to set
5917 * it to a minimal amount for safety.
5919 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5920 btrfs_root_used(&fs_info->csum_root->root_item) +
5921 btrfs_root_used(&fs_info->tree_root->root_item);
5922 num_bytes = max_t(u64, num_bytes, SZ_16M);
5924 spin_lock(&sinfo->lock);
5925 spin_lock(&block_rsv->lock);
5927 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5929 if (block_rsv->reserved < block_rsv->size) {
5930 num_bytes = btrfs_space_info_used(sinfo, true);
5931 if (sinfo->total_bytes > num_bytes) {
5932 num_bytes = sinfo->total_bytes - num_bytes;
5933 num_bytes = min(num_bytes,
5934 block_rsv->size - block_rsv->reserved);
5935 block_rsv->reserved += num_bytes;
5936 update_bytes_may_use(sinfo, num_bytes);
5937 trace_btrfs_space_reservation(fs_info, "space_info",
5938 sinfo->flags, num_bytes,
5941 } else if (block_rsv->reserved > block_rsv->size) {
5942 num_bytes = block_rsv->reserved - block_rsv->size;
5943 update_bytes_may_use(sinfo, -num_bytes);
5944 trace_btrfs_space_reservation(fs_info, "space_info",
5945 sinfo->flags, num_bytes, 0);
5946 block_rsv->reserved = block_rsv->size;
5949 if (block_rsv->reserved == block_rsv->size)
5950 block_rsv->full = 1;
5952 block_rsv->full = 0;
5954 spin_unlock(&block_rsv->lock);
5955 spin_unlock(&sinfo->lock);
5958 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5960 struct btrfs_space_info *space_info;
5962 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5963 fs_info->chunk_block_rsv.space_info = space_info;
5965 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5966 fs_info->global_block_rsv.space_info = space_info;
5967 fs_info->trans_block_rsv.space_info = space_info;
5968 fs_info->empty_block_rsv.space_info = space_info;
5969 fs_info->delayed_block_rsv.space_info = space_info;
5970 fs_info->delayed_refs_rsv.space_info = space_info;
5972 fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv;
5973 fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv;
5974 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5975 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5976 if (fs_info->quota_root)
5977 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5978 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5980 update_global_block_rsv(fs_info);
5983 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5985 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5987 WARN_ON(fs_info->trans_block_rsv.size > 0);
5988 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5989 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5990 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5991 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5992 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5993 WARN_ON(fs_info->delayed_refs_rsv.reserved > 0);
5994 WARN_ON(fs_info->delayed_refs_rsv.size > 0);
5998 * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv
5999 * @trans - the trans that may have generated delayed refs
6001 * This is to be called anytime we may have adjusted trans->delayed_ref_updates,
6002 * it'll calculate the additional size and add it to the delayed_refs_rsv.
6004 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans)
6006 struct btrfs_fs_info *fs_info = trans->fs_info;
6007 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv;
6010 if (!trans->delayed_ref_updates)
6013 num_bytes = btrfs_calc_trans_metadata_size(fs_info,
6014 trans->delayed_ref_updates);
6015 spin_lock(&delayed_rsv->lock);
6016 delayed_rsv->size += num_bytes;
6017 delayed_rsv->full = 0;
6018 spin_unlock(&delayed_rsv->lock);
6019 trans->delayed_ref_updates = 0;
6023 * To be called after all the new block groups attached to the transaction
6024 * handle have been created (btrfs_create_pending_block_groups()).
6026 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
6028 struct btrfs_fs_info *fs_info = trans->fs_info;
6030 if (!trans->chunk_bytes_reserved)
6033 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
6035 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
6036 trans->chunk_bytes_reserved, NULL);
6037 trans->chunk_bytes_reserved = 0;
6041 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
6042 * root: the root of the parent directory
6043 * rsv: block reservation
6044 * items: the number of items that we need do reservation
6045 * use_global_rsv: allow fallback to the global block reservation
6047 * This function is used to reserve the space for snapshot/subvolume
6048 * creation and deletion. Those operations are different with the
6049 * common file/directory operations, they change two fs/file trees
6050 * and root tree, the number of items that the qgroup reserves is
6051 * different with the free space reservation. So we can not use
6052 * the space reservation mechanism in start_transaction().
6054 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
6055 struct btrfs_block_rsv *rsv, int items,
6056 bool use_global_rsv)
6058 u64 qgroup_num_bytes = 0;
6061 struct btrfs_fs_info *fs_info = root->fs_info;
6062 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6064 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6065 /* One for parent inode, two for dir entries */
6066 qgroup_num_bytes = 3 * fs_info->nodesize;
6067 ret = btrfs_qgroup_reserve_meta_prealloc(root,
6068 qgroup_num_bytes, true);
6073 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6074 rsv->space_info = __find_space_info(fs_info,
6075 BTRFS_BLOCK_GROUP_METADATA);
6076 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6077 BTRFS_RESERVE_FLUSH_ALL);
6079 if (ret == -ENOSPC && use_global_rsv)
6080 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true);
6082 if (ret && qgroup_num_bytes)
6083 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
6088 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6089 struct btrfs_block_rsv *rsv)
6091 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6094 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6095 struct btrfs_inode *inode)
6097 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6098 u64 reserve_size = 0;
6099 u64 qgroup_rsv_size = 0;
6101 unsigned outstanding_extents;
6103 lockdep_assert_held(&inode->lock);
6104 outstanding_extents = inode->outstanding_extents;
6105 if (outstanding_extents)
6106 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6107 outstanding_extents + 1);
6108 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6110 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6113 * For qgroup rsv, the calculation is very simple:
6114 * account one nodesize for each outstanding extent
6116 * This is overestimating in most cases.
6118 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
6120 spin_lock(&block_rsv->lock);
6121 block_rsv->size = reserve_size;
6122 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6123 spin_unlock(&block_rsv->lock);
6126 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6128 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6129 unsigned nr_extents;
6130 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6132 bool delalloc_lock = true;
6134 /* If we are a free space inode we need to not flush since we will be in
6135 * the middle of a transaction commit. We also don't need the delalloc
6136 * mutex since we won't race with anybody. We need this mostly to make
6137 * lockdep shut its filthy mouth.
6139 * If we have a transaction open (can happen if we call truncate_block
6140 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6142 if (btrfs_is_free_space_inode(inode)) {
6143 flush = BTRFS_RESERVE_NO_FLUSH;
6144 delalloc_lock = false;
6146 if (current->journal_info)
6147 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6149 if (btrfs_transaction_in_commit(fs_info))
6150 schedule_timeout(1);
6154 mutex_lock(&inode->delalloc_mutex);
6156 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6158 /* Add our new extents and calculate the new rsv size. */
6159 spin_lock(&inode->lock);
6160 nr_extents = count_max_extents(num_bytes);
6161 btrfs_mod_outstanding_extents(inode, nr_extents);
6162 inode->csum_bytes += num_bytes;
6163 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6164 spin_unlock(&inode->lock);
6166 ret = btrfs_inode_rsv_refill(inode, flush);
6171 mutex_unlock(&inode->delalloc_mutex);
6175 spin_lock(&inode->lock);
6176 nr_extents = count_max_extents(num_bytes);
6177 btrfs_mod_outstanding_extents(inode, -nr_extents);
6178 inode->csum_bytes -= num_bytes;
6179 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6180 spin_unlock(&inode->lock);
6182 btrfs_inode_rsv_release(inode, true);
6184 mutex_unlock(&inode->delalloc_mutex);
6189 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6190 * @inode: the inode to release the reservation for.
6191 * @num_bytes: the number of bytes we are releasing.
6192 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6194 * This will release the metadata reservation for an inode. This can be called
6195 * once we complete IO for a given set of bytes to release their metadata
6196 * reservations, or on error for the same reason.
6198 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6201 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6203 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6204 spin_lock(&inode->lock);
6205 inode->csum_bytes -= num_bytes;
6206 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6207 spin_unlock(&inode->lock);
6209 if (btrfs_is_testing(fs_info))
6212 btrfs_inode_rsv_release(inode, qgroup_free);
6216 * btrfs_delalloc_release_extents - release our outstanding_extents
6217 * @inode: the inode to balance the reservation for.
6218 * @num_bytes: the number of bytes we originally reserved with
6219 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6221 * When we reserve space we increase outstanding_extents for the extents we may
6222 * add. Once we've set the range as delalloc or created our ordered extents we
6223 * have outstanding_extents to track the real usage, so we use this to free our
6224 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6225 * with btrfs_delalloc_reserve_metadata.
6227 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6230 struct btrfs_fs_info *fs_info = inode->root->fs_info;
6231 unsigned num_extents;
6233 spin_lock(&inode->lock);
6234 num_extents = count_max_extents(num_bytes);
6235 btrfs_mod_outstanding_extents(inode, -num_extents);
6236 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6237 spin_unlock(&inode->lock);
6239 if (btrfs_is_testing(fs_info))
6242 btrfs_inode_rsv_release(inode, qgroup_free);
6246 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6248 * @inode: inode we're writing to
6249 * @start: start range we are writing to
6250 * @len: how long the range we are writing to
6251 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6252 * current reservation.
6254 * This will do the following things
6256 * o reserve space in data space info for num bytes
6257 * and reserve precious corresponding qgroup space
6258 * (Done in check_data_free_space)
6260 * o reserve space for metadata space, based on the number of outstanding
6261 * extents and how much csums will be needed
6262 * also reserve metadata space in a per root over-reserve method.
6263 * o add to the inodes->delalloc_bytes
6264 * o add it to the fs_info's delalloc inodes list.
6265 * (Above 3 all done in delalloc_reserve_metadata)
6267 * Return 0 for success
6268 * Return <0 for error(-ENOSPC or -EQUOT)
6270 int btrfs_delalloc_reserve_space(struct inode *inode,
6271 struct extent_changeset **reserved, u64 start, u64 len)
6275 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6278 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6280 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6285 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6286 * @inode: inode we're releasing space for
6287 * @start: start position of the space already reserved
6288 * @len: the len of the space already reserved
6289 * @release_bytes: the len of the space we consumed or didn't use
6291 * This function will release the metadata space that was not used and will
6292 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6293 * list if there are no delalloc bytes left.
6294 * Also it will handle the qgroup reserved space.
6296 void btrfs_delalloc_release_space(struct inode *inode,
6297 struct extent_changeset *reserved,
6298 u64 start, u64 len, bool qgroup_free)
6300 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6301 btrfs_free_reserved_data_space(inode, reserved, start, len);
6304 static int update_block_group(struct btrfs_trans_handle *trans,
6305 struct btrfs_fs_info *info, u64 bytenr,
6306 u64 num_bytes, int alloc)
6308 struct btrfs_block_group_cache *cache = NULL;
6309 u64 total = num_bytes;
6315 /* block accounting for super block */
6316 spin_lock(&info->delalloc_root_lock);
6317 old_val = btrfs_super_bytes_used(info->super_copy);
6319 old_val += num_bytes;
6321 old_val -= num_bytes;
6322 btrfs_set_super_bytes_used(info->super_copy, old_val);
6323 spin_unlock(&info->delalloc_root_lock);
6326 cache = btrfs_lookup_block_group(info, bytenr);
6331 factor = btrfs_bg_type_to_factor(cache->flags);
6334 * If this block group has free space cache written out, we
6335 * need to make sure to load it if we are removing space. This
6336 * is because we need the unpinning stage to actually add the
6337 * space back to the block group, otherwise we will leak space.
6339 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6340 cache_block_group(cache, 1);
6342 byte_in_group = bytenr - cache->key.objectid;
6343 WARN_ON(byte_in_group > cache->key.offset);
6345 spin_lock(&cache->space_info->lock);
6346 spin_lock(&cache->lock);
6348 if (btrfs_test_opt(info, SPACE_CACHE) &&
6349 cache->disk_cache_state < BTRFS_DC_CLEAR)
6350 cache->disk_cache_state = BTRFS_DC_CLEAR;
6352 old_val = btrfs_block_group_used(&cache->item);
6353 num_bytes = min(total, cache->key.offset - byte_in_group);
6355 old_val += num_bytes;
6356 btrfs_set_block_group_used(&cache->item, old_val);
6357 cache->reserved -= num_bytes;
6358 cache->space_info->bytes_reserved -= 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 old_val -= num_bytes;
6365 btrfs_set_block_group_used(&cache->item, old_val);
6366 cache->pinned += num_bytes;
6367 update_bytes_pinned(cache->space_info, num_bytes);
6368 cache->space_info->bytes_used -= num_bytes;
6369 cache->space_info->disk_used -= num_bytes * factor;
6370 spin_unlock(&cache->lock);
6371 spin_unlock(&cache->space_info->lock);
6373 trace_btrfs_space_reservation(info, "pinned",
6374 cache->space_info->flags,
6376 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6378 BTRFS_TOTAL_BYTES_PINNED_BATCH);
6379 set_extent_dirty(info->pinned_extents,
6380 bytenr, bytenr + num_bytes - 1,
6381 GFP_NOFS | __GFP_NOFAIL);
6384 spin_lock(&trans->transaction->dirty_bgs_lock);
6385 if (list_empty(&cache->dirty_list)) {
6386 list_add_tail(&cache->dirty_list,
6387 &trans->transaction->dirty_bgs);
6388 trans->transaction->num_dirty_bgs++;
6389 trans->delayed_ref_updates++;
6390 btrfs_get_block_group(cache);
6392 spin_unlock(&trans->transaction->dirty_bgs_lock);
6395 * No longer have used bytes in this block group, queue it for
6396 * deletion. We do this after adding the block group to the
6397 * dirty list to avoid races between cleaner kthread and space
6400 if (!alloc && old_val == 0)
6401 btrfs_mark_bg_unused(cache);
6403 btrfs_put_block_group(cache);
6405 bytenr += num_bytes;
6408 /* Modified block groups are accounted for in the delayed_refs_rsv. */
6409 btrfs_update_delayed_refs_rsv(trans);
6413 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6415 struct btrfs_block_group_cache *cache;
6418 spin_lock(&fs_info->block_group_cache_lock);
6419 bytenr = fs_info->first_logical_byte;
6420 spin_unlock(&fs_info->block_group_cache_lock);
6422 if (bytenr < (u64)-1)
6425 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6429 bytenr = cache->key.objectid;
6430 btrfs_put_block_group(cache);
6435 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6436 struct btrfs_block_group_cache *cache,
6437 u64 bytenr, u64 num_bytes, int reserved)
6439 spin_lock(&cache->space_info->lock);
6440 spin_lock(&cache->lock);
6441 cache->pinned += num_bytes;
6442 update_bytes_pinned(cache->space_info, num_bytes);
6444 cache->reserved -= num_bytes;
6445 cache->space_info->bytes_reserved -= num_bytes;
6447 spin_unlock(&cache->lock);
6448 spin_unlock(&cache->space_info->lock);
6450 trace_btrfs_space_reservation(fs_info, "pinned",
6451 cache->space_info->flags, num_bytes, 1);
6452 percpu_counter_add_batch(&cache->space_info->total_bytes_pinned,
6453 num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6454 set_extent_dirty(fs_info->pinned_extents, bytenr,
6455 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6460 * this function must be called within transaction
6462 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6463 u64 bytenr, u64 num_bytes, int reserved)
6465 struct btrfs_block_group_cache *cache;
6467 cache = btrfs_lookup_block_group(fs_info, bytenr);
6468 BUG_ON(!cache); /* Logic error */
6470 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6472 btrfs_put_block_group(cache);
6477 * this function must be called within transaction
6479 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6480 u64 bytenr, u64 num_bytes)
6482 struct btrfs_block_group_cache *cache;
6485 cache = btrfs_lookup_block_group(fs_info, bytenr);
6490 * pull in the free space cache (if any) so that our pin
6491 * removes the free space from the cache. We have load_only set
6492 * to one because the slow code to read in the free extents does check
6493 * the pinned extents.
6495 cache_block_group(cache, 1);
6497 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6499 /* remove us from the free space cache (if we're there at all) */
6500 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6501 btrfs_put_block_group(cache);
6505 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6506 u64 start, u64 num_bytes)
6509 struct btrfs_block_group_cache *block_group;
6510 struct btrfs_caching_control *caching_ctl;
6512 block_group = btrfs_lookup_block_group(fs_info, start);
6516 cache_block_group(block_group, 0);
6517 caching_ctl = get_caching_control(block_group);
6521 BUG_ON(!block_group_cache_done(block_group));
6522 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6524 mutex_lock(&caching_ctl->mutex);
6526 if (start >= caching_ctl->progress) {
6527 ret = add_excluded_extent(fs_info, start, num_bytes);
6528 } else if (start + num_bytes <= caching_ctl->progress) {
6529 ret = btrfs_remove_free_space(block_group,
6532 num_bytes = caching_ctl->progress - start;
6533 ret = btrfs_remove_free_space(block_group,
6538 num_bytes = (start + num_bytes) -
6539 caching_ctl->progress;
6540 start = caching_ctl->progress;
6541 ret = add_excluded_extent(fs_info, start, num_bytes);
6544 mutex_unlock(&caching_ctl->mutex);
6545 put_caching_control(caching_ctl);
6547 btrfs_put_block_group(block_group);
6551 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6552 struct extent_buffer *eb)
6554 struct btrfs_file_extent_item *item;
6555 struct btrfs_key key;
6560 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6563 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6564 btrfs_item_key_to_cpu(eb, &key, i);
6565 if (key.type != BTRFS_EXTENT_DATA_KEY)
6567 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6568 found_type = btrfs_file_extent_type(eb, item);
6569 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6571 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6573 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6574 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6575 ret = __exclude_logged_extent(fs_info, key.objectid, key.offset);
6584 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6586 atomic_inc(&bg->reservations);
6589 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6592 struct btrfs_block_group_cache *bg;
6594 bg = btrfs_lookup_block_group(fs_info, start);
6596 if (atomic_dec_and_test(&bg->reservations))
6597 wake_up_var(&bg->reservations);
6598 btrfs_put_block_group(bg);
6601 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6603 struct btrfs_space_info *space_info = bg->space_info;
6607 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6611 * Our block group is read only but before we set it to read only,
6612 * some task might have had allocated an extent from it already, but it
6613 * has not yet created a respective ordered extent (and added it to a
6614 * root's list of ordered extents).
6615 * Therefore wait for any task currently allocating extents, since the
6616 * block group's reservations counter is incremented while a read lock
6617 * on the groups' semaphore is held and decremented after releasing
6618 * the read access on that semaphore and creating the ordered extent.
6620 down_write(&space_info->groups_sem);
6621 up_write(&space_info->groups_sem);
6623 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6627 * btrfs_add_reserved_bytes - update the block_group and space info counters
6628 * @cache: The cache we are manipulating
6629 * @ram_bytes: The number of bytes of file content, and will be same to
6630 * @num_bytes except for the compress path.
6631 * @num_bytes: The number of bytes in question
6632 * @delalloc: The blocks are allocated for the delalloc write
6634 * This is called by the allocator when it reserves space. If this is a
6635 * reservation and the block group has become read only we cannot make the
6636 * reservation and return -EAGAIN, otherwise this function always succeeds.
6638 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6639 u64 ram_bytes, u64 num_bytes, int delalloc)
6641 struct btrfs_space_info *space_info = cache->space_info;
6644 spin_lock(&space_info->lock);
6645 spin_lock(&cache->lock);
6649 cache->reserved += num_bytes;
6650 space_info->bytes_reserved += num_bytes;
6651 update_bytes_may_use(space_info, -ram_bytes);
6653 cache->delalloc_bytes += num_bytes;
6655 spin_unlock(&cache->lock);
6656 spin_unlock(&space_info->lock);
6661 * btrfs_free_reserved_bytes - update the block_group and space info counters
6662 * @cache: The cache we are manipulating
6663 * @num_bytes: The number of bytes in question
6664 * @delalloc: The blocks are allocated for the delalloc write
6666 * This is called by somebody who is freeing space that was never actually used
6667 * on disk. For example if you reserve some space for a new leaf in transaction
6668 * A and before transaction A commits you free that leaf, you call this with
6669 * reserve set to 0 in order to clear the reservation.
6672 static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6673 u64 num_bytes, int delalloc)
6675 struct btrfs_space_info *space_info = cache->space_info;
6677 spin_lock(&space_info->lock);
6678 spin_lock(&cache->lock);
6680 space_info->bytes_readonly += num_bytes;
6681 cache->reserved -= num_bytes;
6682 space_info->bytes_reserved -= num_bytes;
6683 space_info->max_extent_size = 0;
6686 cache->delalloc_bytes -= num_bytes;
6687 spin_unlock(&cache->lock);
6688 spin_unlock(&space_info->lock);
6690 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6692 struct btrfs_caching_control *next;
6693 struct btrfs_caching_control *caching_ctl;
6694 struct btrfs_block_group_cache *cache;
6696 down_write(&fs_info->commit_root_sem);
6698 list_for_each_entry_safe(caching_ctl, next,
6699 &fs_info->caching_block_groups, list) {
6700 cache = caching_ctl->block_group;
6701 if (block_group_cache_done(cache)) {
6702 cache->last_byte_to_unpin = (u64)-1;
6703 list_del_init(&caching_ctl->list);
6704 put_caching_control(caching_ctl);
6706 cache->last_byte_to_unpin = caching_ctl->progress;
6710 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6711 fs_info->pinned_extents = &fs_info->freed_extents[1];
6713 fs_info->pinned_extents = &fs_info->freed_extents[0];
6715 up_write(&fs_info->commit_root_sem);
6717 update_global_block_rsv(fs_info);
6721 * Returns the free cluster for the given space info and sets empty_cluster to
6722 * what it should be based on the mount options.
6724 static struct btrfs_free_cluster *
6725 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6726 struct btrfs_space_info *space_info, u64 *empty_cluster)
6728 struct btrfs_free_cluster *ret = NULL;
6731 if (btrfs_mixed_space_info(space_info))
6734 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6735 ret = &fs_info->meta_alloc_cluster;
6736 if (btrfs_test_opt(fs_info, SSD))
6737 *empty_cluster = SZ_2M;
6739 *empty_cluster = SZ_64K;
6740 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6741 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6742 *empty_cluster = SZ_2M;
6743 ret = &fs_info->data_alloc_cluster;
6749 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6751 const bool return_free_space)
6753 struct btrfs_block_group_cache *cache = NULL;
6754 struct btrfs_space_info *space_info;
6755 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6756 struct btrfs_free_cluster *cluster = NULL;
6758 u64 total_unpinned = 0;
6759 u64 empty_cluster = 0;
6762 while (start <= end) {
6765 start >= cache->key.objectid + cache->key.offset) {
6767 btrfs_put_block_group(cache);
6769 cache = btrfs_lookup_block_group(fs_info, start);
6770 BUG_ON(!cache); /* Logic error */
6772 cluster = fetch_cluster_info(fs_info,
6775 empty_cluster <<= 1;
6778 len = cache->key.objectid + cache->key.offset - start;
6779 len = min(len, end + 1 - start);
6781 if (start < cache->last_byte_to_unpin) {
6782 len = min(len, cache->last_byte_to_unpin - start);
6783 if (return_free_space)
6784 btrfs_add_free_space(cache, start, len);
6788 total_unpinned += len;
6789 space_info = cache->space_info;
6792 * If this space cluster has been marked as fragmented and we've
6793 * unpinned enough in this block group to potentially allow a
6794 * cluster to be created inside of it go ahead and clear the
6797 if (cluster && cluster->fragmented &&
6798 total_unpinned > empty_cluster) {
6799 spin_lock(&cluster->lock);
6800 cluster->fragmented = 0;
6801 spin_unlock(&cluster->lock);
6804 spin_lock(&space_info->lock);
6805 spin_lock(&cache->lock);
6806 cache->pinned -= len;
6807 update_bytes_pinned(space_info, -len);
6809 trace_btrfs_space_reservation(fs_info, "pinned",
6810 space_info->flags, len, 0);
6811 space_info->max_extent_size = 0;
6812 percpu_counter_add_batch(&space_info->total_bytes_pinned,
6813 -len, BTRFS_TOTAL_BYTES_PINNED_BATCH);
6815 space_info->bytes_readonly += len;
6818 spin_unlock(&cache->lock);
6819 if (!readonly && return_free_space &&
6820 global_rsv->space_info == space_info) {
6823 spin_lock(&global_rsv->lock);
6824 if (!global_rsv->full) {
6825 to_add = min(len, global_rsv->size -
6826 global_rsv->reserved);
6827 global_rsv->reserved += to_add;
6828 update_bytes_may_use(space_info, to_add);
6829 if (global_rsv->reserved >= global_rsv->size)
6830 global_rsv->full = 1;
6831 trace_btrfs_space_reservation(fs_info,
6837 spin_unlock(&global_rsv->lock);
6838 /* Add to any tickets we may have */
6840 space_info_add_new_bytes(fs_info, space_info,
6843 spin_unlock(&space_info->lock);
6847 btrfs_put_block_group(cache);
6851 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6853 struct btrfs_fs_info *fs_info = trans->fs_info;
6854 struct btrfs_block_group_cache *block_group, *tmp;
6855 struct list_head *deleted_bgs;
6856 struct extent_io_tree *unpin;
6861 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6862 unpin = &fs_info->freed_extents[1];
6864 unpin = &fs_info->freed_extents[0];
6866 while (!trans->aborted) {
6867 struct extent_state *cached_state = NULL;
6869 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6870 ret = find_first_extent_bit(unpin, 0, &start, &end,
6871 EXTENT_DIRTY, &cached_state);
6873 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6877 if (btrfs_test_opt(fs_info, DISCARD))
6878 ret = btrfs_discard_extent(fs_info, start,
6879 end + 1 - start, NULL);
6881 clear_extent_dirty(unpin, start, end, &cached_state);
6882 unpin_extent_range(fs_info, start, end, true);
6883 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6884 free_extent_state(cached_state);
6889 * Transaction is finished. We don't need the lock anymore. We
6890 * do need to clean up the block groups in case of a transaction
6893 deleted_bgs = &trans->transaction->deleted_bgs;
6894 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6898 if (!trans->aborted)
6899 ret = btrfs_discard_extent(fs_info,
6900 block_group->key.objectid,
6901 block_group->key.offset,
6904 list_del_init(&block_group->bg_list);
6905 btrfs_put_block_group_trimming(block_group);
6906 btrfs_put_block_group(block_group);
6909 const char *errstr = btrfs_decode_error(ret);
6911 "discard failed while removing blockgroup: errno=%d %s",
6919 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6920 struct btrfs_delayed_ref_node *node, u64 parent,
6921 u64 root_objectid, u64 owner_objectid,
6922 u64 owner_offset, int refs_to_drop,
6923 struct btrfs_delayed_extent_op *extent_op)
6925 struct btrfs_fs_info *info = trans->fs_info;
6926 struct btrfs_key key;
6927 struct btrfs_path *path;
6928 struct btrfs_root *extent_root = info->extent_root;
6929 struct extent_buffer *leaf;
6930 struct btrfs_extent_item *ei;
6931 struct btrfs_extent_inline_ref *iref;
6934 int extent_slot = 0;
6935 int found_extent = 0;
6939 u64 bytenr = node->bytenr;
6940 u64 num_bytes = node->num_bytes;
6942 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6944 path = btrfs_alloc_path();
6948 path->reada = READA_FORWARD;
6949 path->leave_spinning = 1;
6951 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6952 BUG_ON(!is_data && refs_to_drop != 1);
6955 skinny_metadata = false;
6957 ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes,
6958 parent, root_objectid, owner_objectid,
6961 extent_slot = path->slots[0];
6962 while (extent_slot >= 0) {
6963 btrfs_item_key_to_cpu(path->nodes[0], &key,
6965 if (key.objectid != bytenr)
6967 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6968 key.offset == num_bytes) {
6972 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6973 key.offset == owner_objectid) {
6977 if (path->slots[0] - extent_slot > 5)
6982 if (!found_extent) {
6984 ret = remove_extent_backref(trans, path, NULL,
6986 is_data, &last_ref);
6988 btrfs_abort_transaction(trans, ret);
6991 btrfs_release_path(path);
6992 path->leave_spinning = 1;
6994 key.objectid = bytenr;
6995 key.type = BTRFS_EXTENT_ITEM_KEY;
6996 key.offset = num_bytes;
6998 if (!is_data && skinny_metadata) {
6999 key.type = BTRFS_METADATA_ITEM_KEY;
7000 key.offset = owner_objectid;
7003 ret = btrfs_search_slot(trans, extent_root,
7005 if (ret > 0 && skinny_metadata && path->slots[0]) {
7007 * Couldn't find our skinny metadata item,
7008 * see if we have ye olde extent item.
7011 btrfs_item_key_to_cpu(path->nodes[0], &key,
7013 if (key.objectid == bytenr &&
7014 key.type == BTRFS_EXTENT_ITEM_KEY &&
7015 key.offset == num_bytes)
7019 if (ret > 0 && skinny_metadata) {
7020 skinny_metadata = false;
7021 key.objectid = bytenr;
7022 key.type = BTRFS_EXTENT_ITEM_KEY;
7023 key.offset = num_bytes;
7024 btrfs_release_path(path);
7025 ret = btrfs_search_slot(trans, extent_root,
7031 "umm, got %d back from search, was looking for %llu",
7034 btrfs_print_leaf(path->nodes[0]);
7037 btrfs_abort_transaction(trans, ret);
7040 extent_slot = path->slots[0];
7042 } else if (WARN_ON(ret == -ENOENT)) {
7043 btrfs_print_leaf(path->nodes[0]);
7045 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7046 bytenr, parent, root_objectid, owner_objectid,
7048 btrfs_abort_transaction(trans, ret);
7051 btrfs_abort_transaction(trans, ret);
7055 leaf = path->nodes[0];
7056 item_size = btrfs_item_size_nr(leaf, extent_slot);
7057 if (unlikely(item_size < sizeof(*ei))) {
7059 btrfs_print_v0_err(info);
7060 btrfs_abort_transaction(trans, ret);
7063 ei = btrfs_item_ptr(leaf, extent_slot,
7064 struct btrfs_extent_item);
7065 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7066 key.type == BTRFS_EXTENT_ITEM_KEY) {
7067 struct btrfs_tree_block_info *bi;
7068 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7069 bi = (struct btrfs_tree_block_info *)(ei + 1);
7070 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7073 refs = btrfs_extent_refs(leaf, ei);
7074 if (refs < refs_to_drop) {
7076 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7077 refs_to_drop, refs, bytenr);
7079 btrfs_abort_transaction(trans, ret);
7082 refs -= refs_to_drop;
7086 __run_delayed_extent_op(extent_op, leaf, ei);
7088 * In the case of inline back ref, reference count will
7089 * be updated by remove_extent_backref
7092 BUG_ON(!found_extent);
7094 btrfs_set_extent_refs(leaf, ei, refs);
7095 btrfs_mark_buffer_dirty(leaf);
7098 ret = remove_extent_backref(trans, path, iref,
7099 refs_to_drop, is_data,
7102 btrfs_abort_transaction(trans, ret);
7108 BUG_ON(is_data && refs_to_drop !=
7109 extent_data_ref_count(path, iref));
7111 BUG_ON(path->slots[0] != extent_slot);
7113 BUG_ON(path->slots[0] != extent_slot + 1);
7114 path->slots[0] = extent_slot;
7120 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7123 btrfs_abort_transaction(trans, ret);
7126 btrfs_release_path(path);
7129 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7131 btrfs_abort_transaction(trans, ret);
7136 ret = add_to_free_space_tree(trans, bytenr, num_bytes);
7138 btrfs_abort_transaction(trans, ret);
7142 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7144 btrfs_abort_transaction(trans, ret);
7148 btrfs_release_path(path);
7151 btrfs_free_path(path);
7156 * when we free an block, it is possible (and likely) that we free the last
7157 * delayed ref for that extent as well. This searches the delayed ref tree for
7158 * a given extent, and if there are no other delayed refs to be processed, it
7159 * removes it from the tree.
7161 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7164 struct btrfs_delayed_ref_head *head;
7165 struct btrfs_delayed_ref_root *delayed_refs;
7168 delayed_refs = &trans->transaction->delayed_refs;
7169 spin_lock(&delayed_refs->lock);
7170 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7172 goto out_delayed_unlock;
7174 spin_lock(&head->lock);
7175 if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root))
7178 if (cleanup_extent_op(head) != NULL)
7182 * waiting for the lock here would deadlock. If someone else has it
7183 * locked they are already in the process of dropping it anyway
7185 if (!mutex_trylock(&head->mutex))
7188 btrfs_delete_ref_head(delayed_refs, head);
7189 head->processing = 0;
7191 spin_unlock(&head->lock);
7192 spin_unlock(&delayed_refs->lock);
7194 BUG_ON(head->extent_op);
7195 if (head->must_insert_reserved)
7198 btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head);
7199 mutex_unlock(&head->mutex);
7200 btrfs_put_delayed_ref_head(head);
7203 spin_unlock(&head->lock);
7206 spin_unlock(&delayed_refs->lock);
7210 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7211 struct btrfs_root *root,
7212 struct extent_buffer *buf,
7213 u64 parent, int last_ref)
7215 struct btrfs_fs_info *fs_info = root->fs_info;
7219 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7220 int old_ref_mod, new_ref_mod;
7222 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7223 root->root_key.objectid,
7224 btrfs_header_level(buf), 0,
7225 BTRFS_DROP_DELAYED_REF);
7226 ret = btrfs_add_delayed_tree_ref(trans, buf->start,
7228 root->root_key.objectid,
7229 btrfs_header_level(buf),
7230 BTRFS_DROP_DELAYED_REF, NULL,
7231 &old_ref_mod, &new_ref_mod);
7232 BUG_ON(ret); /* -ENOMEM */
7233 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7236 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7237 struct btrfs_block_group_cache *cache;
7239 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7240 ret = check_ref_cleanup(trans, buf->start);
7246 cache = btrfs_lookup_block_group(fs_info, buf->start);
7248 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7249 pin_down_extent(fs_info, cache, buf->start,
7251 btrfs_put_block_group(cache);
7255 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7257 btrfs_add_free_space(cache, buf->start, buf->len);
7258 btrfs_free_reserved_bytes(cache, buf->len, 0);
7259 btrfs_put_block_group(cache);
7260 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7264 add_pinned_bytes(fs_info, buf->len, true,
7265 root->root_key.objectid);
7269 * Deleting the buffer, clear the corrupt flag since it doesn't
7272 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7276 /* Can return -ENOMEM */
7277 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7278 struct btrfs_root *root,
7279 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7280 u64 owner, u64 offset)
7282 struct btrfs_fs_info *fs_info = root->fs_info;
7283 int old_ref_mod, new_ref_mod;
7286 if (btrfs_is_testing(fs_info))
7289 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7290 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7291 root_objectid, owner, offset,
7292 BTRFS_DROP_DELAYED_REF);
7295 * tree log blocks never actually go into the extent allocation
7296 * tree, just update pinning info and exit early.
7298 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7299 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7300 /* unlocks the pinned mutex */
7301 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7302 old_ref_mod = new_ref_mod = 0;
7304 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7305 ret = btrfs_add_delayed_tree_ref(trans, bytenr,
7307 root_objectid, (int)owner,
7308 BTRFS_DROP_DELAYED_REF, NULL,
7309 &old_ref_mod, &new_ref_mod);
7311 ret = btrfs_add_delayed_data_ref(trans, bytenr,
7313 root_objectid, owner, offset,
7314 0, BTRFS_DROP_DELAYED_REF,
7315 &old_ref_mod, &new_ref_mod);
7318 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7319 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7321 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7328 * when we wait for progress in the block group caching, its because
7329 * our allocation attempt failed at least once. So, we must sleep
7330 * and let some progress happen before we try again.
7332 * This function will sleep at least once waiting for new free space to
7333 * show up, and then it will check the block group free space numbers
7334 * for our min num_bytes. Another option is to have it go ahead
7335 * and look in the rbtree for a free extent of a given size, but this
7338 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7339 * any of the information in this block group.
7341 static noinline void
7342 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7345 struct btrfs_caching_control *caching_ctl;
7347 caching_ctl = get_caching_control(cache);
7351 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7352 (cache->free_space_ctl->free_space >= num_bytes));
7354 put_caching_control(caching_ctl);
7358 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7360 struct btrfs_caching_control *caching_ctl;
7363 caching_ctl = get_caching_control(cache);
7365 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7367 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7368 if (cache->cached == BTRFS_CACHE_ERROR)
7370 put_caching_control(caching_ctl);
7374 enum btrfs_loop_type {
7375 LOOP_CACHING_NOWAIT = 0,
7376 LOOP_CACHING_WAIT = 1,
7377 LOOP_ALLOC_CHUNK = 2,
7378 LOOP_NO_EMPTY_SIZE = 3,
7382 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7386 down_read(&cache->data_rwsem);
7390 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7393 btrfs_get_block_group(cache);
7395 down_read(&cache->data_rwsem);
7398 static struct btrfs_block_group_cache *
7399 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7400 struct btrfs_free_cluster *cluster,
7403 struct btrfs_block_group_cache *used_bg = NULL;
7405 spin_lock(&cluster->refill_lock);
7407 used_bg = cluster->block_group;
7411 if (used_bg == block_group)
7414 btrfs_get_block_group(used_bg);
7419 if (down_read_trylock(&used_bg->data_rwsem))
7422 spin_unlock(&cluster->refill_lock);
7424 /* We should only have one-level nested. */
7425 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7427 spin_lock(&cluster->refill_lock);
7428 if (used_bg == cluster->block_group)
7431 up_read(&used_bg->data_rwsem);
7432 btrfs_put_block_group(used_bg);
7437 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7441 up_read(&cache->data_rwsem);
7442 btrfs_put_block_group(cache);
7446 * Structure used internally for find_free_extent() function. Wraps needed
7449 struct find_free_extent_ctl {
7450 /* Basic allocation info */
7457 /* Where to start the search inside the bg */
7460 /* For clustered allocation */
7463 bool have_caching_bg;
7464 bool orig_have_caching_bg;
7466 /* RAID index, converted from flags */
7470 * Current loop number, check find_free_extent_update_loop() for details
7475 * Whether we're refilling a cluster, if true we need to re-search
7476 * current block group but don't try to refill the cluster again.
7478 bool retry_clustered;
7481 * Whether we're updating free space cache, if true we need to re-search
7482 * current block group but don't try updating free space cache again.
7484 bool retry_unclustered;
7486 /* If current block group is cached */
7489 /* Max contiguous hole found */
7490 u64 max_extent_size;
7492 /* Total free space from free space cache, not always contiguous */
7493 u64 total_free_space;
7501 * Helper function for find_free_extent().
7503 * Return -ENOENT to inform caller that we need fallback to unclustered mode.
7504 * Return -EAGAIN to inform caller that we need to re-search this block group
7505 * Return >0 to inform caller that we find nothing
7506 * Return 0 means we have found a location and set ffe_ctl->found_offset.
7508 static int find_free_extent_clustered(struct btrfs_block_group_cache *bg,
7509 struct btrfs_free_cluster *last_ptr,
7510 struct find_free_extent_ctl *ffe_ctl,
7511 struct btrfs_block_group_cache **cluster_bg_ret)
7513 struct btrfs_fs_info *fs_info = bg->fs_info;
7514 struct btrfs_block_group_cache *cluster_bg;
7515 u64 aligned_cluster;
7519 cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc);
7521 goto refill_cluster;
7522 if (cluster_bg != bg && (cluster_bg->ro ||
7523 !block_group_bits(cluster_bg, ffe_ctl->flags)))
7524 goto release_cluster;
7526 offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr,
7527 ffe_ctl->num_bytes, cluster_bg->key.objectid,
7528 &ffe_ctl->max_extent_size);
7530 /* We have a block, we're done */
7531 spin_unlock(&last_ptr->refill_lock);
7532 trace_btrfs_reserve_extent_cluster(cluster_bg,
7533 ffe_ctl->search_start, ffe_ctl->num_bytes);
7534 *cluster_bg_ret = cluster_bg;
7535 ffe_ctl->found_offset = offset;
7538 WARN_ON(last_ptr->block_group != cluster_bg);
7542 * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so
7543 * lets just skip it and let the allocator find whatever block it can
7544 * find. If we reach this point, we will have tried the cluster
7545 * allocator plenty of times and not have found anything, so we are
7546 * likely way too fragmented for the clustering stuff to find anything.
7548 * However, if the cluster is taken from the current block group,
7549 * release the cluster first, so that we stand a better chance of
7550 * succeeding in the unclustered allocation.
7552 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) {
7553 spin_unlock(&last_ptr->refill_lock);
7554 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7558 /* This cluster didn't work out, free it and start over */
7559 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7561 if (cluster_bg != bg)
7562 btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc);
7565 if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) {
7566 spin_unlock(&last_ptr->refill_lock);
7570 aligned_cluster = max_t(u64,
7571 ffe_ctl->empty_cluster + ffe_ctl->empty_size,
7572 bg->full_stripe_len);
7573 ret = btrfs_find_space_cluster(fs_info, bg, last_ptr,
7574 ffe_ctl->search_start, ffe_ctl->num_bytes,
7577 /* Now pull our allocation out of this cluster */
7578 offset = btrfs_alloc_from_cluster(bg, last_ptr,
7579 ffe_ctl->num_bytes, ffe_ctl->search_start,
7580 &ffe_ctl->max_extent_size);
7582 /* We found one, proceed */
7583 spin_unlock(&last_ptr->refill_lock);
7584 trace_btrfs_reserve_extent_cluster(bg,
7585 ffe_ctl->search_start,
7586 ffe_ctl->num_bytes);
7587 ffe_ctl->found_offset = offset;
7590 } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT &&
7591 !ffe_ctl->retry_clustered) {
7592 spin_unlock(&last_ptr->refill_lock);
7594 ffe_ctl->retry_clustered = true;
7595 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7596 ffe_ctl->empty_cluster + ffe_ctl->empty_size);
7600 * At this point we either didn't find a cluster or we weren't able to
7601 * allocate a block from our cluster. Free the cluster we've been
7602 * trying to use, and go to the next block group.
7604 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7605 spin_unlock(&last_ptr->refill_lock);
7610 * Return >0 to inform caller that we find nothing
7611 * Return 0 when we found an free extent and set ffe_ctrl->found_offset
7612 * Return -EAGAIN to inform caller that we need to re-search this block group
7614 static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg,
7615 struct btrfs_free_cluster *last_ptr,
7616 struct find_free_extent_ctl *ffe_ctl)
7621 * We are doing an unclustered allocation, set the fragmented flag so
7622 * we don't bother trying to setup a cluster again until we get more
7625 if (unlikely(last_ptr)) {
7626 spin_lock(&last_ptr->lock);
7627 last_ptr->fragmented = 1;
7628 spin_unlock(&last_ptr->lock);
7630 if (ffe_ctl->cached) {
7631 struct btrfs_free_space_ctl *free_space_ctl;
7633 free_space_ctl = bg->free_space_ctl;
7634 spin_lock(&free_space_ctl->tree_lock);
7635 if (free_space_ctl->free_space <
7636 ffe_ctl->num_bytes + ffe_ctl->empty_cluster +
7637 ffe_ctl->empty_size) {
7638 ffe_ctl->total_free_space = max_t(u64,
7639 ffe_ctl->total_free_space,
7640 free_space_ctl->free_space);
7641 spin_unlock(&free_space_ctl->tree_lock);
7644 spin_unlock(&free_space_ctl->tree_lock);
7647 offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start,
7648 ffe_ctl->num_bytes, ffe_ctl->empty_size,
7649 &ffe_ctl->max_extent_size);
7652 * If we didn't find a chunk, and we haven't failed on this block group
7653 * before, and this block group is in the middle of caching and we are
7654 * ok with waiting, then go ahead and wait for progress to be made, and
7655 * set @retry_unclustered to true.
7657 * If @retry_unclustered is true then we've already waited on this
7658 * block group once and should move on to the next block group.
7660 if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached &&
7661 ffe_ctl->loop > LOOP_CACHING_NOWAIT) {
7662 wait_block_group_cache_progress(bg, ffe_ctl->num_bytes +
7663 ffe_ctl->empty_size);
7664 ffe_ctl->retry_unclustered = true;
7666 } else if (!offset) {
7669 ffe_ctl->found_offset = offset;
7674 * Return >0 means caller needs to re-search for free extent
7675 * Return 0 means we have the needed free extent.
7676 * Return <0 means we failed to locate any free extent.
7678 static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info,
7679 struct btrfs_free_cluster *last_ptr,
7680 struct btrfs_key *ins,
7681 struct find_free_extent_ctl *ffe_ctl,
7682 int full_search, bool use_cluster)
7684 struct btrfs_root *root = fs_info->extent_root;
7687 if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) &&
7688 ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg)
7689 ffe_ctl->orig_have_caching_bg = true;
7691 if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT &&
7692 ffe_ctl->have_caching_bg)
7695 if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES)
7698 if (ins->objectid) {
7699 if (!use_cluster && last_ptr) {
7700 spin_lock(&last_ptr->lock);
7701 last_ptr->window_start = ins->objectid;
7702 spin_unlock(&last_ptr->lock);
7708 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7709 * caching kthreads as we move along
7710 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7711 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7712 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7715 if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) {
7717 if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) {
7719 * We want to skip the LOOP_CACHING_WAIT step if we
7720 * don't have any uncached bgs and we've already done a
7721 * full search through.
7723 if (ffe_ctl->orig_have_caching_bg || !full_search)
7724 ffe_ctl->loop = LOOP_CACHING_WAIT;
7726 ffe_ctl->loop = LOOP_ALLOC_CHUNK;
7731 if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) {
7732 struct btrfs_trans_handle *trans;
7735 trans = current->journal_info;
7739 trans = btrfs_join_transaction(root);
7741 if (IS_ERR(trans)) {
7742 ret = PTR_ERR(trans);
7746 ret = do_chunk_alloc(trans, ffe_ctl->flags,
7750 * If we can't allocate a new chunk we've already looped
7751 * through at least once, move on to the NO_EMPTY_SIZE
7755 ffe_ctl->loop = LOOP_NO_EMPTY_SIZE;
7757 /* Do not bail out on ENOSPC since we can do more. */
7758 if (ret < 0 && ret != -ENOSPC)
7759 btrfs_abort_transaction(trans, ret);
7763 btrfs_end_transaction(trans);
7768 if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) {
7770 * Don't loop again if we already have no empty_size and
7773 if (ffe_ctl->empty_size == 0 &&
7774 ffe_ctl->empty_cluster == 0)
7776 ffe_ctl->empty_size = 0;
7777 ffe_ctl->empty_cluster = 0;
7785 * walks the btree of allocated extents and find a hole of a given size.
7786 * The key ins is changed to record the hole:
7787 * ins->objectid == start position
7788 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7789 * ins->offset == the size of the hole.
7790 * Any available blocks before search_start are skipped.
7792 * If there is no suitable free space, we will record the max size of
7793 * the free space extent currently.
7795 * The overall logic and call chain:
7797 * find_free_extent()
7798 * |- Iterate through all block groups
7799 * | |- Get a valid block group
7800 * | |- Try to do clustered allocation in that block group
7801 * | |- Try to do unclustered allocation in that block group
7802 * | |- Check if the result is valid
7803 * | | |- If valid, then exit
7804 * | |- Jump to next block group
7806 * |- Push harder to find free extents
7807 * |- If not found, re-iterate all block groups
7809 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7810 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7811 u64 hint_byte, struct btrfs_key *ins,
7812 u64 flags, int delalloc)
7815 struct btrfs_free_cluster *last_ptr = NULL;
7816 struct btrfs_block_group_cache *block_group = NULL;
7817 struct find_free_extent_ctl ffe_ctl = {0};
7818 struct btrfs_space_info *space_info;
7819 bool use_cluster = true;
7820 bool full_search = false;
7822 WARN_ON(num_bytes < fs_info->sectorsize);
7824 ffe_ctl.ram_bytes = ram_bytes;
7825 ffe_ctl.num_bytes = num_bytes;
7826 ffe_ctl.empty_size = empty_size;
7827 ffe_ctl.flags = flags;
7828 ffe_ctl.search_start = 0;
7829 ffe_ctl.retry_clustered = false;
7830 ffe_ctl.retry_unclustered = false;
7831 ffe_ctl.delalloc = delalloc;
7832 ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags);
7833 ffe_ctl.have_caching_bg = false;
7834 ffe_ctl.orig_have_caching_bg = false;
7835 ffe_ctl.found_offset = 0;
7837 ins->type = BTRFS_EXTENT_ITEM_KEY;
7841 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7843 space_info = __find_space_info(fs_info, flags);
7845 btrfs_err(fs_info, "No space info for %llu", flags);
7850 * If our free space is heavily fragmented we may not be able to make
7851 * big contiguous allocations, so instead of doing the expensive search
7852 * for free space, simply return ENOSPC with our max_extent_size so we
7853 * can go ahead and search for a more manageable chunk.
7855 * If our max_extent_size is large enough for our allocation simply
7856 * disable clustering since we will likely not be able to find enough
7857 * space to create a cluster and induce latency trying.
7859 if (unlikely(space_info->max_extent_size)) {
7860 spin_lock(&space_info->lock);
7861 if (space_info->max_extent_size &&
7862 num_bytes > space_info->max_extent_size) {
7863 ins->offset = space_info->max_extent_size;
7864 spin_unlock(&space_info->lock);
7866 } else if (space_info->max_extent_size) {
7867 use_cluster = false;
7869 spin_unlock(&space_info->lock);
7872 last_ptr = fetch_cluster_info(fs_info, space_info,
7873 &ffe_ctl.empty_cluster);
7875 spin_lock(&last_ptr->lock);
7876 if (last_ptr->block_group)
7877 hint_byte = last_ptr->window_start;
7878 if (last_ptr->fragmented) {
7880 * We still set window_start so we can keep track of the
7881 * last place we found an allocation to try and save
7884 hint_byte = last_ptr->window_start;
7885 use_cluster = false;
7887 spin_unlock(&last_ptr->lock);
7890 ffe_ctl.search_start = max(ffe_ctl.search_start,
7891 first_logical_byte(fs_info, 0));
7892 ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte);
7893 if (ffe_ctl.search_start == hint_byte) {
7894 block_group = btrfs_lookup_block_group(fs_info,
7895 ffe_ctl.search_start);
7897 * we don't want to use the block group if it doesn't match our
7898 * allocation bits, or if its not cached.
7900 * However if we are re-searching with an ideal block group
7901 * picked out then we don't care that the block group is cached.
7903 if (block_group && block_group_bits(block_group, flags) &&
7904 block_group->cached != BTRFS_CACHE_NO) {
7905 down_read(&space_info->groups_sem);
7906 if (list_empty(&block_group->list) ||
7909 * someone is removing this block group,
7910 * we can't jump into the have_block_group
7911 * target because our list pointers are not
7914 btrfs_put_block_group(block_group);
7915 up_read(&space_info->groups_sem);
7917 ffe_ctl.index = btrfs_bg_flags_to_raid_index(
7918 block_group->flags);
7919 btrfs_lock_block_group(block_group, delalloc);
7920 goto have_block_group;
7922 } else if (block_group) {
7923 btrfs_put_block_group(block_group);
7927 ffe_ctl.have_caching_bg = false;
7928 if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) ||
7931 down_read(&space_info->groups_sem);
7932 list_for_each_entry(block_group,
7933 &space_info->block_groups[ffe_ctl.index], list) {
7934 /* If the block group is read-only, we can skip it entirely. */
7935 if (unlikely(block_group->ro))
7938 btrfs_grab_block_group(block_group, delalloc);
7939 ffe_ctl.search_start = block_group->key.objectid;
7942 * this can happen if we end up cycling through all the
7943 * raid types, but we want to make sure we only allocate
7944 * for the proper type.
7946 if (!block_group_bits(block_group, flags)) {
7947 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7948 BTRFS_BLOCK_GROUP_RAID1 |
7949 BTRFS_BLOCK_GROUP_RAID5 |
7950 BTRFS_BLOCK_GROUP_RAID6 |
7951 BTRFS_BLOCK_GROUP_RAID10;
7954 * if they asked for extra copies and this block group
7955 * doesn't provide them, bail. This does allow us to
7956 * fill raid0 from raid1.
7958 if ((flags & extra) && !(block_group->flags & extra))
7963 ffe_ctl.cached = block_group_cache_done(block_group);
7964 if (unlikely(!ffe_ctl.cached)) {
7965 ffe_ctl.have_caching_bg = true;
7966 ret = cache_block_group(block_group, 0);
7971 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7975 * Ok we want to try and use the cluster allocator, so
7978 if (last_ptr && use_cluster) {
7979 struct btrfs_block_group_cache *cluster_bg = NULL;
7981 ret = find_free_extent_clustered(block_group, last_ptr,
7982 &ffe_ctl, &cluster_bg);
7985 if (cluster_bg && cluster_bg != block_group) {
7986 btrfs_release_block_group(block_group,
7988 block_group = cluster_bg;
7991 } else if (ret == -EAGAIN) {
7992 goto have_block_group;
7993 } else if (ret > 0) {
7996 /* ret == -ENOENT case falls through */
7999 ret = find_free_extent_unclustered(block_group, last_ptr,
8002 goto have_block_group;
8005 /* ret == 0 case falls through */
8007 ffe_ctl.search_start = round_up(ffe_ctl.found_offset,
8008 fs_info->stripesize);
8010 /* move on to the next group */
8011 if (ffe_ctl.search_start + num_bytes >
8012 block_group->key.objectid + block_group->key.offset) {
8013 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8018 if (ffe_ctl.found_offset < ffe_ctl.search_start)
8019 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8020 ffe_ctl.search_start - ffe_ctl.found_offset);
8022 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
8023 num_bytes, delalloc);
8024 if (ret == -EAGAIN) {
8025 btrfs_add_free_space(block_group, ffe_ctl.found_offset,
8029 btrfs_inc_block_group_reservations(block_group);
8031 /* we are all good, lets return */
8032 ins->objectid = ffe_ctl.search_start;
8033 ins->offset = num_bytes;
8035 trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start,
8037 btrfs_release_block_group(block_group, delalloc);
8040 ffe_ctl.retry_clustered = false;
8041 ffe_ctl.retry_unclustered = false;
8042 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
8044 btrfs_release_block_group(block_group, delalloc);
8047 up_read(&space_info->groups_sem);
8049 ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl,
8050 full_search, use_cluster);
8054 if (ret == -ENOSPC) {
8056 * Use ffe_ctl->total_free_space as fallback if we can't find
8057 * any contiguous hole.
8059 if (!ffe_ctl.max_extent_size)
8060 ffe_ctl.max_extent_size = ffe_ctl.total_free_space;
8061 spin_lock(&space_info->lock);
8062 space_info->max_extent_size = ffe_ctl.max_extent_size;
8063 spin_unlock(&space_info->lock);
8064 ins->offset = ffe_ctl.max_extent_size;
8069 static void dump_space_info(struct btrfs_fs_info *fs_info,
8070 struct btrfs_space_info *info, u64 bytes,
8071 int dump_block_groups)
8073 struct btrfs_block_group_cache *cache;
8076 spin_lock(&info->lock);
8077 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
8079 info->total_bytes - btrfs_space_info_used(info, true),
8080 info->full ? "" : "not ");
8082 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
8083 info->total_bytes, info->bytes_used, info->bytes_pinned,
8084 info->bytes_reserved, info->bytes_may_use,
8085 info->bytes_readonly);
8086 spin_unlock(&info->lock);
8088 if (!dump_block_groups)
8091 down_read(&info->groups_sem);
8093 list_for_each_entry(cache, &info->block_groups[index], list) {
8094 spin_lock(&cache->lock);
8096 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8097 cache->key.objectid, cache->key.offset,
8098 btrfs_block_group_used(&cache->item), cache->pinned,
8099 cache->reserved, cache->ro ? "[readonly]" : "");
8100 btrfs_dump_free_space(cache, bytes);
8101 spin_unlock(&cache->lock);
8103 if (++index < BTRFS_NR_RAID_TYPES)
8105 up_read(&info->groups_sem);
8109 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
8110 * hole that is at least as big as @num_bytes.
8112 * @root - The root that will contain this extent
8114 * @ram_bytes - The amount of space in ram that @num_bytes take. This
8115 * is used for accounting purposes. This value differs
8116 * from @num_bytes only in the case of compressed extents.
8118 * @num_bytes - Number of bytes to allocate on-disk.
8120 * @min_alloc_size - Indicates the minimum amount of space that the
8121 * allocator should try to satisfy. In some cases
8122 * @num_bytes may be larger than what is required and if
8123 * the filesystem is fragmented then allocation fails.
8124 * However, the presence of @min_alloc_size gives a
8125 * chance to try and satisfy the smaller allocation.
8127 * @empty_size - A hint that you plan on doing more COW. This is the
8128 * size in bytes the allocator should try to find free
8129 * next to the block it returns. This is just a hint and
8130 * may be ignored by the allocator.
8132 * @hint_byte - Hint to the allocator to start searching above the byte
8133 * address passed. It might be ignored.
8135 * @ins - This key is modified to record the found hole. It will
8136 * have the following values:
8137 * ins->objectid == start position
8138 * ins->flags = BTRFS_EXTENT_ITEM_KEY
8139 * ins->offset == the size of the hole.
8141 * @is_data - Boolean flag indicating whether an extent is
8142 * allocated for data (true) or metadata (false)
8144 * @delalloc - Boolean flag indicating whether this allocation is for
8145 * delalloc or not. If 'true' data_rwsem of block groups
8146 * is going to be acquired.
8149 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8150 * case -ENOSPC is returned then @ins->offset will contain the size of the
8151 * largest available hole the allocator managed to find.
8153 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8154 u64 num_bytes, u64 min_alloc_size,
8155 u64 empty_size, u64 hint_byte,
8156 struct btrfs_key *ins, int is_data, int delalloc)
8158 struct btrfs_fs_info *fs_info = root->fs_info;
8159 bool final_tried = num_bytes == min_alloc_size;
8163 flags = get_alloc_profile_by_root(root, is_data);
8165 WARN_ON(num_bytes < fs_info->sectorsize);
8166 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8167 hint_byte, ins, flags, delalloc);
8168 if (!ret && !is_data) {
8169 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8170 } else if (ret == -ENOSPC) {
8171 if (!final_tried && ins->offset) {
8172 num_bytes = min(num_bytes >> 1, ins->offset);
8173 num_bytes = round_down(num_bytes,
8174 fs_info->sectorsize);
8175 num_bytes = max(num_bytes, min_alloc_size);
8176 ram_bytes = num_bytes;
8177 if (num_bytes == min_alloc_size)
8180 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8181 struct btrfs_space_info *sinfo;
8183 sinfo = __find_space_info(fs_info, flags);
8185 "allocation failed flags %llu, wanted %llu",
8188 dump_space_info(fs_info, sinfo, num_bytes, 1);
8195 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8197 int pin, int delalloc)
8199 struct btrfs_block_group_cache *cache;
8202 cache = btrfs_lookup_block_group(fs_info, start);
8204 btrfs_err(fs_info, "Unable to find block group for %llu",
8210 pin_down_extent(fs_info, cache, start, len, 1);
8212 if (btrfs_test_opt(fs_info, DISCARD))
8213 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8214 btrfs_add_free_space(cache, start, len);
8215 btrfs_free_reserved_bytes(cache, len, delalloc);
8216 trace_btrfs_reserved_extent_free(fs_info, start, len);
8219 btrfs_put_block_group(cache);
8223 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8224 u64 start, u64 len, int delalloc)
8226 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8229 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8232 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8235 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8236 u64 parent, u64 root_objectid,
8237 u64 flags, u64 owner, u64 offset,
8238 struct btrfs_key *ins, int ref_mod)
8240 struct btrfs_fs_info *fs_info = trans->fs_info;
8242 struct btrfs_extent_item *extent_item;
8243 struct btrfs_extent_inline_ref *iref;
8244 struct btrfs_path *path;
8245 struct extent_buffer *leaf;
8250 type = BTRFS_SHARED_DATA_REF_KEY;
8252 type = BTRFS_EXTENT_DATA_REF_KEY;
8254 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8256 path = btrfs_alloc_path();
8260 path->leave_spinning = 1;
8261 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8264 btrfs_free_path(path);
8268 leaf = path->nodes[0];
8269 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8270 struct btrfs_extent_item);
8271 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8272 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8273 btrfs_set_extent_flags(leaf, extent_item,
8274 flags | BTRFS_EXTENT_FLAG_DATA);
8276 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8277 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8279 struct btrfs_shared_data_ref *ref;
8280 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8281 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8282 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8284 struct btrfs_extent_data_ref *ref;
8285 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8286 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8287 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8288 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8289 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8292 btrfs_mark_buffer_dirty(path->nodes[0]);
8293 btrfs_free_path(path);
8295 ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset);
8299 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8300 if (ret) { /* -ENOENT, logic error */
8301 btrfs_err(fs_info, "update block group failed for %llu %llu",
8302 ins->objectid, ins->offset);
8305 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8309 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8310 struct btrfs_delayed_ref_node *node,
8311 struct btrfs_delayed_extent_op *extent_op)
8313 struct btrfs_fs_info *fs_info = trans->fs_info;
8315 struct btrfs_extent_item *extent_item;
8316 struct btrfs_key extent_key;
8317 struct btrfs_tree_block_info *block_info;
8318 struct btrfs_extent_inline_ref *iref;
8319 struct btrfs_path *path;
8320 struct extent_buffer *leaf;
8321 struct btrfs_delayed_tree_ref *ref;
8322 u32 size = sizeof(*extent_item) + sizeof(*iref);
8324 u64 flags = extent_op->flags_to_set;
8325 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8327 ref = btrfs_delayed_node_to_tree_ref(node);
8329 extent_key.objectid = node->bytenr;
8330 if (skinny_metadata) {
8331 extent_key.offset = ref->level;
8332 extent_key.type = BTRFS_METADATA_ITEM_KEY;
8333 num_bytes = fs_info->nodesize;
8335 extent_key.offset = node->num_bytes;
8336 extent_key.type = BTRFS_EXTENT_ITEM_KEY;
8337 size += sizeof(*block_info);
8338 num_bytes = node->num_bytes;
8341 path = btrfs_alloc_path();
8345 path->leave_spinning = 1;
8346 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8349 btrfs_free_path(path);
8353 leaf = path->nodes[0];
8354 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8355 struct btrfs_extent_item);
8356 btrfs_set_extent_refs(leaf, extent_item, 1);
8357 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8358 btrfs_set_extent_flags(leaf, extent_item,
8359 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8361 if (skinny_metadata) {
8362 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8364 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8365 btrfs_set_tree_block_key(leaf, block_info, &extent_op->key);
8366 btrfs_set_tree_block_level(leaf, block_info, ref->level);
8367 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8370 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) {
8371 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8372 btrfs_set_extent_inline_ref_type(leaf, iref,
8373 BTRFS_SHARED_BLOCK_REF_KEY);
8374 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent);
8376 btrfs_set_extent_inline_ref_type(leaf, iref,
8377 BTRFS_TREE_BLOCK_REF_KEY);
8378 btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root);
8381 btrfs_mark_buffer_dirty(leaf);
8382 btrfs_free_path(path);
8384 ret = remove_from_free_space_tree(trans, extent_key.objectid,
8389 ret = update_block_group(trans, fs_info, extent_key.objectid,
8390 fs_info->nodesize, 1);
8391 if (ret) { /* -ENOENT, logic error */
8392 btrfs_err(fs_info, "update block group failed for %llu %llu",
8393 extent_key.objectid, extent_key.offset);
8397 trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid,
8402 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8403 struct btrfs_root *root, u64 owner,
8404 u64 offset, u64 ram_bytes,
8405 struct btrfs_key *ins)
8409 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8411 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8412 root->root_key.objectid, owner, offset,
8413 BTRFS_ADD_DELAYED_EXTENT);
8415 ret = btrfs_add_delayed_data_ref(trans, ins->objectid,
8417 root->root_key.objectid, owner,
8419 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8424 * this is used by the tree logging recovery code. It records that
8425 * an extent has been allocated and makes sure to clear the free
8426 * space cache bits as well
8428 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8429 u64 root_objectid, u64 owner, u64 offset,
8430 struct btrfs_key *ins)
8432 struct btrfs_fs_info *fs_info = trans->fs_info;
8434 struct btrfs_block_group_cache *block_group;
8435 struct btrfs_space_info *space_info;
8438 * Mixed block groups will exclude before processing the log so we only
8439 * need to do the exclude dance if this fs isn't mixed.
8441 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8442 ret = __exclude_logged_extent(fs_info, ins->objectid,
8448 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8452 space_info = block_group->space_info;
8453 spin_lock(&space_info->lock);
8454 spin_lock(&block_group->lock);
8455 space_info->bytes_reserved += ins->offset;
8456 block_group->reserved += ins->offset;
8457 spin_unlock(&block_group->lock);
8458 spin_unlock(&space_info->lock);
8460 ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
8462 btrfs_put_block_group(block_group);
8466 static struct extent_buffer *
8467 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8468 u64 bytenr, int level, u64 owner)
8470 struct btrfs_fs_info *fs_info = root->fs_info;
8471 struct extent_buffer *buf;
8473 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8478 * Extra safety check in case the extent tree is corrupted and extent
8479 * allocator chooses to use a tree block which is already used and
8482 if (buf->lock_owner == current->pid) {
8483 btrfs_err_rl(fs_info,
8484 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8485 buf->start, btrfs_header_owner(buf), current->pid);
8486 free_extent_buffer(buf);
8487 return ERR_PTR(-EUCLEAN);
8490 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8491 btrfs_tree_lock(buf);
8492 clean_tree_block(fs_info, buf);
8493 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8495 btrfs_set_lock_blocking(buf);
8496 set_extent_buffer_uptodate(buf);
8498 memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header));
8499 btrfs_set_header_level(buf, level);
8500 btrfs_set_header_bytenr(buf, buf->start);
8501 btrfs_set_header_generation(buf, trans->transid);
8502 btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV);
8503 btrfs_set_header_owner(buf, owner);
8504 write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid);
8505 write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid);
8506 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8507 buf->log_index = root->log_transid % 2;
8509 * we allow two log transactions at a time, use different
8510 * EXTENT bit to differentiate dirty pages.
8512 if (buf->log_index == 0)
8513 set_extent_dirty(&root->dirty_log_pages, buf->start,
8514 buf->start + buf->len - 1, GFP_NOFS);
8516 set_extent_new(&root->dirty_log_pages, buf->start,
8517 buf->start + buf->len - 1);
8519 buf->log_index = -1;
8520 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8521 buf->start + buf->len - 1, GFP_NOFS);
8523 trans->dirty = true;
8524 /* this returns a buffer locked for blocking */
8528 static struct btrfs_block_rsv *
8529 use_block_rsv(struct btrfs_trans_handle *trans,
8530 struct btrfs_root *root, u32 blocksize)
8532 struct btrfs_fs_info *fs_info = root->fs_info;
8533 struct btrfs_block_rsv *block_rsv;
8534 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8536 bool global_updated = false;
8538 block_rsv = get_block_rsv(trans, root);
8540 if (unlikely(block_rsv->size == 0))
8543 ret = block_rsv_use_bytes(block_rsv, blocksize);
8547 if (block_rsv->failfast)
8548 return ERR_PTR(ret);
8550 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8551 global_updated = true;
8552 update_global_block_rsv(fs_info);
8557 * The global reserve still exists to save us from ourselves, so don't
8558 * warn_on if we are short on our delayed refs reserve.
8560 if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS &&
8561 btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8562 static DEFINE_RATELIMIT_STATE(_rs,
8563 DEFAULT_RATELIMIT_INTERVAL * 10,
8564 /*DEFAULT_RATELIMIT_BURST*/ 1);
8565 if (__ratelimit(&_rs))
8567 "BTRFS: block rsv returned %d\n", ret);
8570 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8571 BTRFS_RESERVE_NO_FLUSH);
8575 * If we couldn't reserve metadata bytes try and use some from
8576 * the global reserve if its space type is the same as the global
8579 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8580 block_rsv->space_info == global_rsv->space_info) {
8581 ret = block_rsv_use_bytes(global_rsv, blocksize);
8585 return ERR_PTR(ret);
8588 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8589 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8591 block_rsv_add_bytes(block_rsv, blocksize, false);
8592 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8596 * finds a free extent and does all the dirty work required for allocation
8597 * returns the tree buffer or an ERR_PTR on error.
8599 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8600 struct btrfs_root *root,
8601 u64 parent, u64 root_objectid,
8602 const struct btrfs_disk_key *key,
8603 int level, u64 hint,
8606 struct btrfs_fs_info *fs_info = root->fs_info;
8607 struct btrfs_key ins;
8608 struct btrfs_block_rsv *block_rsv;
8609 struct extent_buffer *buf;
8610 struct btrfs_delayed_extent_op *extent_op;
8613 u32 blocksize = fs_info->nodesize;
8614 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8616 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8617 if (btrfs_is_testing(fs_info)) {
8618 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8619 level, root_objectid);
8621 root->alloc_bytenr += blocksize;
8626 block_rsv = use_block_rsv(trans, root, blocksize);
8627 if (IS_ERR(block_rsv))
8628 return ERR_CAST(block_rsv);
8630 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8631 empty_size, hint, &ins, 0, 0);
8635 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level,
8639 goto out_free_reserved;
8642 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8644 parent = ins.objectid;
8645 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8649 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8650 extent_op = btrfs_alloc_delayed_extent_op();
8656 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8658 memset(&extent_op->key, 0, sizeof(extent_op->key));
8659 extent_op->flags_to_set = flags;
8660 extent_op->update_key = skinny_metadata ? false : true;
8661 extent_op->update_flags = true;
8662 extent_op->is_data = false;
8663 extent_op->level = level;
8665 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8666 root_objectid, level, 0,
8667 BTRFS_ADD_DELAYED_EXTENT);
8668 ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
8670 root_objectid, level,
8671 BTRFS_ADD_DELAYED_EXTENT,
8672 extent_op, NULL, NULL);
8674 goto out_free_delayed;
8679 btrfs_free_delayed_extent_op(extent_op);
8681 free_extent_buffer(buf);
8683 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8685 unuse_block_rsv(fs_info, block_rsv, blocksize);
8686 return ERR_PTR(ret);
8689 struct walk_control {
8690 u64 refs[BTRFS_MAX_LEVEL];
8691 u64 flags[BTRFS_MAX_LEVEL];
8692 struct btrfs_key update_progress;
8702 #define DROP_REFERENCE 1
8703 #define UPDATE_BACKREF 2
8705 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8706 struct btrfs_root *root,
8707 struct walk_control *wc,
8708 struct btrfs_path *path)
8710 struct btrfs_fs_info *fs_info = root->fs_info;
8716 struct btrfs_key key;
8717 struct extent_buffer *eb;
8722 if (path->slots[wc->level] < wc->reada_slot) {
8723 wc->reada_count = wc->reada_count * 2 / 3;
8724 wc->reada_count = max(wc->reada_count, 2);
8726 wc->reada_count = wc->reada_count * 3 / 2;
8727 wc->reada_count = min_t(int, wc->reada_count,
8728 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8731 eb = path->nodes[wc->level];
8732 nritems = btrfs_header_nritems(eb);
8734 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8735 if (nread >= wc->reada_count)
8739 bytenr = btrfs_node_blockptr(eb, slot);
8740 generation = btrfs_node_ptr_generation(eb, slot);
8742 if (slot == path->slots[wc->level])
8745 if (wc->stage == UPDATE_BACKREF &&
8746 generation <= root->root_key.offset)
8749 /* We don't lock the tree block, it's OK to be racy here */
8750 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8751 wc->level - 1, 1, &refs,
8753 /* We don't care about errors in readahead. */
8758 if (wc->stage == DROP_REFERENCE) {
8762 if (wc->level == 1 &&
8763 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8765 if (!wc->update_ref ||
8766 generation <= root->root_key.offset)
8768 btrfs_node_key_to_cpu(eb, &key, slot);
8769 ret = btrfs_comp_cpu_keys(&key,
8770 &wc->update_progress);
8774 if (wc->level == 1 &&
8775 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8779 readahead_tree_block(fs_info, bytenr);
8782 wc->reada_slot = slot;
8786 * helper to process tree block while walking down the tree.
8788 * when wc->stage == UPDATE_BACKREF, this function updates
8789 * back refs for pointers in the block.
8791 * NOTE: return value 1 means we should stop walking down.
8793 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8794 struct btrfs_root *root,
8795 struct btrfs_path *path,
8796 struct walk_control *wc, int lookup_info)
8798 struct btrfs_fs_info *fs_info = root->fs_info;
8799 int level = wc->level;
8800 struct extent_buffer *eb = path->nodes[level];
8801 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8804 if (wc->stage == UPDATE_BACKREF &&
8805 btrfs_header_owner(eb) != root->root_key.objectid)
8809 * when reference count of tree block is 1, it won't increase
8810 * again. once full backref flag is set, we never clear it.
8813 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8814 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8815 BUG_ON(!path->locks[level]);
8816 ret = btrfs_lookup_extent_info(trans, fs_info,
8817 eb->start, level, 1,
8820 BUG_ON(ret == -ENOMEM);
8823 BUG_ON(wc->refs[level] == 0);
8826 if (wc->stage == DROP_REFERENCE) {
8827 if (wc->refs[level] > 1)
8830 if (path->locks[level] && !wc->keep_locks) {
8831 btrfs_tree_unlock_rw(eb, path->locks[level]);
8832 path->locks[level] = 0;
8837 /* wc->stage == UPDATE_BACKREF */
8838 if (!(wc->flags[level] & flag)) {
8839 BUG_ON(!path->locks[level]);
8840 ret = btrfs_inc_ref(trans, root, eb, 1);
8841 BUG_ON(ret); /* -ENOMEM */
8842 ret = btrfs_dec_ref(trans, root, eb, 0);
8843 BUG_ON(ret); /* -ENOMEM */
8844 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8846 btrfs_header_level(eb), 0);
8847 BUG_ON(ret); /* -ENOMEM */
8848 wc->flags[level] |= flag;
8852 * the block is shared by multiple trees, so it's not good to
8853 * keep the tree lock
8855 if (path->locks[level] && level > 0) {
8856 btrfs_tree_unlock_rw(eb, path->locks[level]);
8857 path->locks[level] = 0;
8863 * helper to process tree block pointer.
8865 * when wc->stage == DROP_REFERENCE, this function checks
8866 * reference count of the block pointed to. if the block
8867 * is shared and we need update back refs for the subtree
8868 * rooted at the block, this function changes wc->stage to
8869 * UPDATE_BACKREF. if the block is shared and there is no
8870 * need to update back, this function drops the reference
8873 * NOTE: return value 1 means we should stop walking down.
8875 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8876 struct btrfs_root *root,
8877 struct btrfs_path *path,
8878 struct walk_control *wc, int *lookup_info)
8880 struct btrfs_fs_info *fs_info = root->fs_info;
8884 struct btrfs_key key;
8885 struct btrfs_key first_key;
8886 struct extent_buffer *next;
8887 int level = wc->level;
8890 bool need_account = false;
8892 generation = btrfs_node_ptr_generation(path->nodes[level],
8893 path->slots[level]);
8895 * if the lower level block was created before the snapshot
8896 * was created, we know there is no need to update back refs
8899 if (wc->stage == UPDATE_BACKREF &&
8900 generation <= root->root_key.offset) {
8905 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8906 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8907 path->slots[level]);
8909 next = find_extent_buffer(fs_info, bytenr);
8911 next = btrfs_find_create_tree_block(fs_info, bytenr);
8913 return PTR_ERR(next);
8915 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8919 btrfs_tree_lock(next);
8920 btrfs_set_lock_blocking(next);
8922 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8923 &wc->refs[level - 1],
8924 &wc->flags[level - 1]);
8928 if (unlikely(wc->refs[level - 1] == 0)) {
8929 btrfs_err(fs_info, "Missing references.");
8935 if (wc->stage == DROP_REFERENCE) {
8936 if (wc->refs[level - 1] > 1) {
8937 need_account = true;
8939 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8942 if (!wc->update_ref ||
8943 generation <= root->root_key.offset)
8946 btrfs_node_key_to_cpu(path->nodes[level], &key,
8947 path->slots[level]);
8948 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8952 wc->stage = UPDATE_BACKREF;
8953 wc->shared_level = level - 1;
8957 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8961 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8962 btrfs_tree_unlock(next);
8963 free_extent_buffer(next);
8969 if (reada && level == 1)
8970 reada_walk_down(trans, root, wc, path);
8971 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8974 return PTR_ERR(next);
8975 } else if (!extent_buffer_uptodate(next)) {
8976 free_extent_buffer(next);
8979 btrfs_tree_lock(next);
8980 btrfs_set_lock_blocking(next);
8984 ASSERT(level == btrfs_header_level(next));
8985 if (level != btrfs_header_level(next)) {
8986 btrfs_err(root->fs_info, "mismatched level");
8990 path->nodes[level] = next;
8991 path->slots[level] = 0;
8992 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8998 wc->refs[level - 1] = 0;
8999 wc->flags[level - 1] = 0;
9000 if (wc->stage == DROP_REFERENCE) {
9001 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
9002 parent = path->nodes[level]->start;
9004 ASSERT(root->root_key.objectid ==
9005 btrfs_header_owner(path->nodes[level]));
9006 if (root->root_key.objectid !=
9007 btrfs_header_owner(path->nodes[level])) {
9008 btrfs_err(root->fs_info,
9009 "mismatched block owner");
9017 * Reloc tree doesn't contribute to qgroup numbers, and we have
9018 * already accounted them at merge time (replace_path),
9019 * thus we could skip expensive subtree trace here.
9021 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
9023 ret = btrfs_qgroup_trace_subtree(trans, next,
9024 generation, level - 1);
9026 btrfs_err_rl(fs_info,
9027 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
9031 ret = btrfs_free_extent(trans, root, bytenr, fs_info->nodesize,
9032 parent, root->root_key.objectid,
9042 btrfs_tree_unlock(next);
9043 free_extent_buffer(next);
9049 * helper to process tree block while walking up the tree.
9051 * when wc->stage == DROP_REFERENCE, this function drops
9052 * reference count on the block.
9054 * when wc->stage == UPDATE_BACKREF, this function changes
9055 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9056 * to UPDATE_BACKREF previously while processing the block.
9058 * NOTE: return value 1 means we should stop walking up.
9060 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9061 struct btrfs_root *root,
9062 struct btrfs_path *path,
9063 struct walk_control *wc)
9065 struct btrfs_fs_info *fs_info = root->fs_info;
9067 int level = wc->level;
9068 struct extent_buffer *eb = path->nodes[level];
9071 if (wc->stage == UPDATE_BACKREF) {
9072 BUG_ON(wc->shared_level < level);
9073 if (level < wc->shared_level)
9076 ret = find_next_key(path, level + 1, &wc->update_progress);
9080 wc->stage = DROP_REFERENCE;
9081 wc->shared_level = -1;
9082 path->slots[level] = 0;
9085 * check reference count again if the block isn't locked.
9086 * we should start walking down the tree again if reference
9089 if (!path->locks[level]) {
9091 btrfs_tree_lock(eb);
9092 btrfs_set_lock_blocking(eb);
9093 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9095 ret = btrfs_lookup_extent_info(trans, fs_info,
9096 eb->start, level, 1,
9100 btrfs_tree_unlock_rw(eb, path->locks[level]);
9101 path->locks[level] = 0;
9104 BUG_ON(wc->refs[level] == 0);
9105 if (wc->refs[level] == 1) {
9106 btrfs_tree_unlock_rw(eb, path->locks[level]);
9107 path->locks[level] = 0;
9113 /* wc->stage == DROP_REFERENCE */
9114 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9116 if (wc->refs[level] == 1) {
9118 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9119 ret = btrfs_dec_ref(trans, root, eb, 1);
9121 ret = btrfs_dec_ref(trans, root, eb, 0);
9122 BUG_ON(ret); /* -ENOMEM */
9123 ret = btrfs_qgroup_trace_leaf_items(trans, eb);
9125 btrfs_err_rl(fs_info,
9126 "error %d accounting leaf items. Quota is out of sync, rescan required.",
9130 /* make block locked assertion in clean_tree_block happy */
9131 if (!path->locks[level] &&
9132 btrfs_header_generation(eb) == trans->transid) {
9133 btrfs_tree_lock(eb);
9134 btrfs_set_lock_blocking(eb);
9135 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9137 clean_tree_block(fs_info, eb);
9140 if (eb == root->node) {
9141 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9143 else if (root->root_key.objectid != btrfs_header_owner(eb))
9144 goto owner_mismatch;
9146 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9147 parent = path->nodes[level + 1]->start;
9148 else if (root->root_key.objectid !=
9149 btrfs_header_owner(path->nodes[level + 1]))
9150 goto owner_mismatch;
9153 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9155 wc->refs[level] = 0;
9156 wc->flags[level] = 0;
9160 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
9161 btrfs_header_owner(eb), root->root_key.objectid);
9165 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9166 struct btrfs_root *root,
9167 struct btrfs_path *path,
9168 struct walk_control *wc)
9170 int level = wc->level;
9171 int lookup_info = 1;
9174 while (level >= 0) {
9175 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9182 if (path->slots[level] >=
9183 btrfs_header_nritems(path->nodes[level]))
9186 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9188 path->slots[level]++;
9197 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9198 struct btrfs_root *root,
9199 struct btrfs_path *path,
9200 struct walk_control *wc, int max_level)
9202 int level = wc->level;
9205 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9206 while (level < max_level && path->nodes[level]) {
9208 if (path->slots[level] + 1 <
9209 btrfs_header_nritems(path->nodes[level])) {
9210 path->slots[level]++;
9213 ret = walk_up_proc(trans, root, path, wc);
9219 if (path->locks[level]) {
9220 btrfs_tree_unlock_rw(path->nodes[level],
9221 path->locks[level]);
9222 path->locks[level] = 0;
9224 free_extent_buffer(path->nodes[level]);
9225 path->nodes[level] = NULL;
9233 * drop a subvolume tree.
9235 * this function traverses the tree freeing any blocks that only
9236 * referenced by the tree.
9238 * when a shared tree block is found. this function decreases its
9239 * reference count by one. if update_ref is true, this function
9240 * also make sure backrefs for the shared block and all lower level
9241 * blocks are properly updated.
9243 * If called with for_reloc == 0, may exit early with -EAGAIN
9245 int btrfs_drop_snapshot(struct btrfs_root *root,
9246 struct btrfs_block_rsv *block_rsv, int update_ref,
9249 struct btrfs_fs_info *fs_info = root->fs_info;
9250 struct btrfs_path *path;
9251 struct btrfs_trans_handle *trans;
9252 struct btrfs_root *tree_root = fs_info->tree_root;
9253 struct btrfs_root_item *root_item = &root->root_item;
9254 struct walk_control *wc;
9255 struct btrfs_key key;
9259 bool root_dropped = false;
9261 btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid);
9263 path = btrfs_alloc_path();
9269 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9271 btrfs_free_path(path);
9276 trans = btrfs_start_transaction(tree_root, 0);
9277 if (IS_ERR(trans)) {
9278 err = PTR_ERR(trans);
9282 err = btrfs_run_delayed_items(trans);
9287 trans->block_rsv = block_rsv;
9290 * This will help us catch people modifying the fs tree while we're
9291 * dropping it. It is unsafe to mess with the fs tree while it's being
9292 * dropped as we unlock the root node and parent nodes as we walk down
9293 * the tree, assuming nothing will change. If something does change
9294 * then we'll have stale information and drop references to blocks we've
9297 set_bit(BTRFS_ROOT_DELETING, &root->state);
9298 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9299 level = btrfs_header_level(root->node);
9300 path->nodes[level] = btrfs_lock_root_node(root);
9301 btrfs_set_lock_blocking(path->nodes[level]);
9302 path->slots[level] = 0;
9303 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9304 memset(&wc->update_progress, 0,
9305 sizeof(wc->update_progress));
9307 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9308 memcpy(&wc->update_progress, &key,
9309 sizeof(wc->update_progress));
9311 level = root_item->drop_level;
9313 path->lowest_level = level;
9314 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9315 path->lowest_level = 0;
9323 * unlock our path, this is safe because only this
9324 * function is allowed to delete this snapshot
9326 btrfs_unlock_up_safe(path, 0);
9328 level = btrfs_header_level(root->node);
9330 btrfs_tree_lock(path->nodes[level]);
9331 btrfs_set_lock_blocking(path->nodes[level]);
9332 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9334 ret = btrfs_lookup_extent_info(trans, fs_info,
9335 path->nodes[level]->start,
9336 level, 1, &wc->refs[level],
9342 BUG_ON(wc->refs[level] == 0);
9344 if (level == root_item->drop_level)
9347 btrfs_tree_unlock(path->nodes[level]);
9348 path->locks[level] = 0;
9349 WARN_ON(wc->refs[level] != 1);
9355 wc->shared_level = -1;
9356 wc->stage = DROP_REFERENCE;
9357 wc->update_ref = update_ref;
9359 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9363 ret = walk_down_tree(trans, root, path, wc);
9369 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9376 BUG_ON(wc->stage != DROP_REFERENCE);
9380 if (wc->stage == DROP_REFERENCE) {
9382 btrfs_node_key(path->nodes[level],
9383 &root_item->drop_progress,
9384 path->slots[level]);
9385 root_item->drop_level = level;
9388 BUG_ON(wc->level == 0);
9389 if (btrfs_should_end_transaction(trans) ||
9390 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9391 ret = btrfs_update_root(trans, tree_root,
9395 btrfs_abort_transaction(trans, ret);
9400 btrfs_end_transaction_throttle(trans);
9401 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9402 btrfs_debug(fs_info,
9403 "drop snapshot early exit");
9408 trans = btrfs_start_transaction(tree_root, 0);
9409 if (IS_ERR(trans)) {
9410 err = PTR_ERR(trans);
9414 trans->block_rsv = block_rsv;
9417 btrfs_release_path(path);
9421 ret = btrfs_del_root(trans, &root->root_key);
9423 btrfs_abort_transaction(trans, ret);
9428 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9429 ret = btrfs_find_root(tree_root, &root->root_key, path,
9432 btrfs_abort_transaction(trans, ret);
9435 } else if (ret > 0) {
9436 /* if we fail to delete the orphan item this time
9437 * around, it'll get picked up the next time.
9439 * The most common failure here is just -ENOENT.
9441 btrfs_del_orphan_item(trans, tree_root,
9442 root->root_key.objectid);
9446 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9447 btrfs_add_dropped_root(trans, root);
9449 free_extent_buffer(root->node);
9450 free_extent_buffer(root->commit_root);
9451 btrfs_put_fs_root(root);
9453 root_dropped = true;
9455 btrfs_end_transaction_throttle(trans);
9458 btrfs_free_path(path);
9461 * So if we need to stop dropping the snapshot for whatever reason we
9462 * need to make sure to add it back to the dead root list so that we
9463 * keep trying to do the work later. This also cleans up roots if we
9464 * don't have it in the radix (like when we recover after a power fail
9465 * or unmount) so we don't leak memory.
9467 if (!for_reloc && !root_dropped)
9468 btrfs_add_dead_root(root);
9469 if (err && err != -EAGAIN)
9470 btrfs_handle_fs_error(fs_info, err, NULL);
9475 * drop subtree rooted at tree block 'node'.
9477 * NOTE: this function will unlock and release tree block 'node'
9478 * only used by relocation code
9480 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9481 struct btrfs_root *root,
9482 struct extent_buffer *node,
9483 struct extent_buffer *parent)
9485 struct btrfs_fs_info *fs_info = root->fs_info;
9486 struct btrfs_path *path;
9487 struct walk_control *wc;
9493 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9495 path = btrfs_alloc_path();
9499 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9501 btrfs_free_path(path);
9505 btrfs_assert_tree_locked(parent);
9506 parent_level = btrfs_header_level(parent);
9507 extent_buffer_get(parent);
9508 path->nodes[parent_level] = parent;
9509 path->slots[parent_level] = btrfs_header_nritems(parent);
9511 btrfs_assert_tree_locked(node);
9512 level = btrfs_header_level(node);
9513 path->nodes[level] = node;
9514 path->slots[level] = 0;
9515 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9517 wc->refs[parent_level] = 1;
9518 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9520 wc->shared_level = -1;
9521 wc->stage = DROP_REFERENCE;
9524 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9527 wret = walk_down_tree(trans, root, path, wc);
9533 wret = walk_up_tree(trans, root, path, wc, parent_level);
9541 btrfs_free_path(path);
9545 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9551 * if restripe for this chunk_type is on pick target profile and
9552 * return, otherwise do the usual balance
9554 stripped = get_restripe_target(fs_info, flags);
9556 return extended_to_chunk(stripped);
9558 num_devices = fs_info->fs_devices->rw_devices;
9560 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9561 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9562 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9564 if (num_devices == 1) {
9565 stripped |= BTRFS_BLOCK_GROUP_DUP;
9566 stripped = flags & ~stripped;
9568 /* turn raid0 into single device chunks */
9569 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9572 /* turn mirroring into duplication */
9573 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9574 BTRFS_BLOCK_GROUP_RAID10))
9575 return stripped | BTRFS_BLOCK_GROUP_DUP;
9577 /* they already had raid on here, just return */
9578 if (flags & stripped)
9581 stripped |= BTRFS_BLOCK_GROUP_DUP;
9582 stripped = flags & ~stripped;
9584 /* switch duplicated blocks with raid1 */
9585 if (flags & BTRFS_BLOCK_GROUP_DUP)
9586 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9588 /* this is drive concat, leave it alone */
9594 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9596 struct btrfs_space_info *sinfo = cache->space_info;
9598 u64 min_allocable_bytes;
9602 * We need some metadata space and system metadata space for
9603 * allocating chunks in some corner cases until we force to set
9604 * it to be readonly.
9607 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9609 min_allocable_bytes = SZ_1M;
9611 min_allocable_bytes = 0;
9613 spin_lock(&sinfo->lock);
9614 spin_lock(&cache->lock);
9622 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9623 cache->bytes_super - btrfs_block_group_used(&cache->item);
9625 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9626 min_allocable_bytes <= sinfo->total_bytes) {
9627 sinfo->bytes_readonly += num_bytes;
9629 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9633 spin_unlock(&cache->lock);
9634 spin_unlock(&sinfo->lock);
9638 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache)
9641 struct btrfs_fs_info *fs_info = cache->fs_info;
9642 struct btrfs_trans_handle *trans;
9647 trans = btrfs_join_transaction(fs_info->extent_root);
9649 return PTR_ERR(trans);
9652 * we're not allowed to set block groups readonly after the dirty
9653 * block groups cache has started writing. If it already started,
9654 * back off and let this transaction commit
9656 mutex_lock(&fs_info->ro_block_group_mutex);
9657 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9658 u64 transid = trans->transid;
9660 mutex_unlock(&fs_info->ro_block_group_mutex);
9661 btrfs_end_transaction(trans);
9663 ret = btrfs_wait_for_commit(fs_info, transid);
9670 * if we are changing raid levels, try to allocate a corresponding
9671 * block group with the new raid level.
9673 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9674 if (alloc_flags != cache->flags) {
9675 ret = do_chunk_alloc(trans, alloc_flags,
9678 * ENOSPC is allowed here, we may have enough space
9679 * already allocated at the new raid level to
9688 ret = inc_block_group_ro(cache, 0);
9691 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9692 ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9695 ret = inc_block_group_ro(cache, 0);
9697 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9698 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9699 mutex_lock(&fs_info->chunk_mutex);
9700 check_system_chunk(trans, alloc_flags);
9701 mutex_unlock(&fs_info->chunk_mutex);
9703 mutex_unlock(&fs_info->ro_block_group_mutex);
9705 btrfs_end_transaction(trans);
9709 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type)
9711 u64 alloc_flags = get_alloc_profile(trans->fs_info, type);
9713 return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE);
9717 * helper to account the unused space of all the readonly block group in the
9718 * space_info. takes mirrors into account.
9720 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9722 struct btrfs_block_group_cache *block_group;
9726 /* It's df, we don't care if it's racy */
9727 if (list_empty(&sinfo->ro_bgs))
9730 spin_lock(&sinfo->lock);
9731 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9732 spin_lock(&block_group->lock);
9734 if (!block_group->ro) {
9735 spin_unlock(&block_group->lock);
9739 factor = btrfs_bg_type_to_factor(block_group->flags);
9740 free_bytes += (block_group->key.offset -
9741 btrfs_block_group_used(&block_group->item)) *
9744 spin_unlock(&block_group->lock);
9746 spin_unlock(&sinfo->lock);
9751 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9753 struct btrfs_space_info *sinfo = cache->space_info;
9758 spin_lock(&sinfo->lock);
9759 spin_lock(&cache->lock);
9761 num_bytes = cache->key.offset - cache->reserved -
9762 cache->pinned - cache->bytes_super -
9763 btrfs_block_group_used(&cache->item);
9764 sinfo->bytes_readonly -= num_bytes;
9765 list_del_init(&cache->ro_list);
9767 spin_unlock(&cache->lock);
9768 spin_unlock(&sinfo->lock);
9772 * Checks to see if it's even possible to relocate this block group.
9774 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9775 * ok to go ahead and try.
9777 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9779 struct btrfs_root *root = fs_info->extent_root;
9780 struct btrfs_block_group_cache *block_group;
9781 struct btrfs_space_info *space_info;
9782 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9783 struct btrfs_device *device;
9784 struct btrfs_trans_handle *trans;
9794 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9796 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9798 /* odd, couldn't find the block group, leave it alone */
9802 "can't find block group for bytenr %llu",
9807 min_free = btrfs_block_group_used(&block_group->item);
9809 /* no bytes used, we're good */
9813 space_info = block_group->space_info;
9814 spin_lock(&space_info->lock);
9816 full = space_info->full;
9819 * if this is the last block group we have in this space, we can't
9820 * relocate it unless we're able to allocate a new chunk below.
9822 * Otherwise, we need to make sure we have room in the space to handle
9823 * all of the extents from this block group. If we can, we're good
9825 if ((space_info->total_bytes != block_group->key.offset) &&
9826 (btrfs_space_info_used(space_info, false) + min_free <
9827 space_info->total_bytes)) {
9828 spin_unlock(&space_info->lock);
9831 spin_unlock(&space_info->lock);
9834 * ok we don't have enough space, but maybe we have free space on our
9835 * devices to allocate new chunks for relocation, so loop through our
9836 * alloc devices and guess if we have enough space. if this block
9837 * group is going to be restriped, run checks against the target
9838 * profile instead of the current one.
9850 target = get_restripe_target(fs_info, block_group->flags);
9852 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9855 * this is just a balance, so if we were marked as full
9856 * we know there is no space for a new chunk
9861 "no space to alloc new chunk for block group %llu",
9862 block_group->key.objectid);
9866 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9869 if (index == BTRFS_RAID_RAID10) {
9873 } else if (index == BTRFS_RAID_RAID1) {
9875 } else if (index == BTRFS_RAID_DUP) {
9878 } else if (index == BTRFS_RAID_RAID0) {
9879 dev_min = fs_devices->rw_devices;
9880 min_free = div64_u64(min_free, dev_min);
9883 /* We need to do this so that we can look at pending chunks */
9884 trans = btrfs_join_transaction(root);
9885 if (IS_ERR(trans)) {
9886 ret = PTR_ERR(trans);
9890 mutex_lock(&fs_info->chunk_mutex);
9891 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9895 * check to make sure we can actually find a chunk with enough
9896 * space to fit our block group in.
9898 if (device->total_bytes > device->bytes_used + min_free &&
9899 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9900 ret = find_free_dev_extent(trans, device, min_free,
9905 if (dev_nr >= dev_min)
9911 if (debug && ret == -1)
9913 "no space to allocate a new chunk for block group %llu",
9914 block_group->key.objectid);
9915 mutex_unlock(&fs_info->chunk_mutex);
9916 btrfs_end_transaction(trans);
9918 btrfs_put_block_group(block_group);
9922 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9923 struct btrfs_path *path,
9924 struct btrfs_key *key)
9926 struct btrfs_root *root = fs_info->extent_root;
9928 struct btrfs_key found_key;
9929 struct extent_buffer *leaf;
9930 struct btrfs_block_group_item bg;
9934 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9939 slot = path->slots[0];
9940 leaf = path->nodes[0];
9941 if (slot >= btrfs_header_nritems(leaf)) {
9942 ret = btrfs_next_leaf(root, path);
9949 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9951 if (found_key.objectid >= key->objectid &&
9952 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9953 struct extent_map_tree *em_tree;
9954 struct extent_map *em;
9956 em_tree = &root->fs_info->mapping_tree.map_tree;
9957 read_lock(&em_tree->lock);
9958 em = lookup_extent_mapping(em_tree, found_key.objectid,
9960 read_unlock(&em_tree->lock);
9963 "logical %llu len %llu found bg but no related chunk",
9964 found_key.objectid, found_key.offset);
9966 } else if (em->start != found_key.objectid ||
9967 em->len != found_key.offset) {
9969 "block group %llu len %llu mismatch with chunk %llu len %llu",
9970 found_key.objectid, found_key.offset,
9971 em->start, em->len);
9974 read_extent_buffer(leaf, &bg,
9975 btrfs_item_ptr_offset(leaf, slot),
9977 flags = btrfs_block_group_flags(&bg) &
9978 BTRFS_BLOCK_GROUP_TYPE_MASK;
9980 if (flags != (em->map_lookup->type &
9981 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9983 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9985 found_key.offset, flags,
9986 (BTRFS_BLOCK_GROUP_TYPE_MASK &
9987 em->map_lookup->type));
9993 free_extent_map(em);
10002 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
10004 struct btrfs_block_group_cache *block_group;
10008 struct inode *inode;
10010 block_group = btrfs_lookup_first_block_group(info, last);
10011 while (block_group) {
10012 wait_block_group_cache_done(block_group);
10013 spin_lock(&block_group->lock);
10014 if (block_group->iref)
10016 spin_unlock(&block_group->lock);
10017 block_group = next_block_group(info, block_group);
10019 if (!block_group) {
10026 inode = block_group->inode;
10027 block_group->iref = 0;
10028 block_group->inode = NULL;
10029 spin_unlock(&block_group->lock);
10030 ASSERT(block_group->io_ctl.inode == NULL);
10032 last = block_group->key.objectid + block_group->key.offset;
10033 btrfs_put_block_group(block_group);
10038 * Must be called only after stopping all workers, since we could have block
10039 * group caching kthreads running, and therefore they could race with us if we
10040 * freed the block groups before stopping them.
10042 int btrfs_free_block_groups(struct btrfs_fs_info *info)
10044 struct btrfs_block_group_cache *block_group;
10045 struct btrfs_space_info *space_info;
10046 struct btrfs_caching_control *caching_ctl;
10049 down_write(&info->commit_root_sem);
10050 while (!list_empty(&info->caching_block_groups)) {
10051 caching_ctl = list_entry(info->caching_block_groups.next,
10052 struct btrfs_caching_control, list);
10053 list_del(&caching_ctl->list);
10054 put_caching_control(caching_ctl);
10056 up_write(&info->commit_root_sem);
10058 spin_lock(&info->unused_bgs_lock);
10059 while (!list_empty(&info->unused_bgs)) {
10060 block_group = list_first_entry(&info->unused_bgs,
10061 struct btrfs_block_group_cache,
10063 list_del_init(&block_group->bg_list);
10064 btrfs_put_block_group(block_group);
10066 spin_unlock(&info->unused_bgs_lock);
10068 spin_lock(&info->block_group_cache_lock);
10069 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
10070 block_group = rb_entry(n, struct btrfs_block_group_cache,
10072 rb_erase(&block_group->cache_node,
10073 &info->block_group_cache_tree);
10074 RB_CLEAR_NODE(&block_group->cache_node);
10075 spin_unlock(&info->block_group_cache_lock);
10077 down_write(&block_group->space_info->groups_sem);
10078 list_del(&block_group->list);
10079 up_write(&block_group->space_info->groups_sem);
10082 * We haven't cached this block group, which means we could
10083 * possibly have excluded extents on this block group.
10085 if (block_group->cached == BTRFS_CACHE_NO ||
10086 block_group->cached == BTRFS_CACHE_ERROR)
10087 free_excluded_extents(block_group);
10089 btrfs_remove_free_space_cache(block_group);
10090 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
10091 ASSERT(list_empty(&block_group->dirty_list));
10092 ASSERT(list_empty(&block_group->io_list));
10093 ASSERT(list_empty(&block_group->bg_list));
10094 ASSERT(atomic_read(&block_group->count) == 1);
10095 btrfs_put_block_group(block_group);
10097 spin_lock(&info->block_group_cache_lock);
10099 spin_unlock(&info->block_group_cache_lock);
10101 /* now that all the block groups are freed, go through and
10102 * free all the space_info structs. This is only called during
10103 * the final stages of unmount, and so we know nobody is
10104 * using them. We call synchronize_rcu() once before we start,
10105 * just to be on the safe side.
10109 release_global_block_rsv(info);
10111 while (!list_empty(&info->space_info)) {
10114 space_info = list_entry(info->space_info.next,
10115 struct btrfs_space_info,
10119 * Do not hide this behind enospc_debug, this is actually
10120 * important and indicates a real bug if this happens.
10122 if (WARN_ON(space_info->bytes_pinned > 0 ||
10123 space_info->bytes_reserved > 0 ||
10124 space_info->bytes_may_use > 0))
10125 dump_space_info(info, space_info, 0, 0);
10126 list_del(&space_info->list);
10127 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10128 struct kobject *kobj;
10129 kobj = space_info->block_group_kobjs[i];
10130 space_info->block_group_kobjs[i] = NULL;
10136 kobject_del(&space_info->kobj);
10137 kobject_put(&space_info->kobj);
10142 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
10143 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
10145 struct btrfs_space_info *space_info;
10146 struct raid_kobject *rkobj;
10151 spin_lock(&fs_info->pending_raid_kobjs_lock);
10152 list_splice_init(&fs_info->pending_raid_kobjs, &list);
10153 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10155 list_for_each_entry(rkobj, &list, list) {
10156 space_info = __find_space_info(fs_info, rkobj->flags);
10157 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
10159 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10160 "%s", get_raid_name(index));
10162 kobject_put(&rkobj->kobj);
10167 btrfs_warn(fs_info,
10168 "failed to add kobject for block cache, ignoring");
10171 static void link_block_group(struct btrfs_block_group_cache *cache)
10173 struct btrfs_space_info *space_info = cache->space_info;
10174 struct btrfs_fs_info *fs_info = cache->fs_info;
10175 int index = btrfs_bg_flags_to_raid_index(cache->flags);
10176 bool first = false;
10178 down_write(&space_info->groups_sem);
10179 if (list_empty(&space_info->block_groups[index]))
10181 list_add_tail(&cache->list, &space_info->block_groups[index]);
10182 up_write(&space_info->groups_sem);
10185 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10187 btrfs_warn(cache->fs_info,
10188 "couldn't alloc memory for raid level kobject");
10191 rkobj->flags = cache->flags;
10192 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10194 spin_lock(&fs_info->pending_raid_kobjs_lock);
10195 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
10196 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10197 space_info->block_group_kobjs[index] = &rkobj->kobj;
10201 static struct btrfs_block_group_cache *
10202 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10203 u64 start, u64 size)
10205 struct btrfs_block_group_cache *cache;
10207 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10211 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10213 if (!cache->free_space_ctl) {
10218 cache->key.objectid = start;
10219 cache->key.offset = size;
10220 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10222 cache->fs_info = fs_info;
10223 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10224 set_free_space_tree_thresholds(cache);
10226 atomic_set(&cache->count, 1);
10227 spin_lock_init(&cache->lock);
10228 init_rwsem(&cache->data_rwsem);
10229 INIT_LIST_HEAD(&cache->list);
10230 INIT_LIST_HEAD(&cache->cluster_list);
10231 INIT_LIST_HEAD(&cache->bg_list);
10232 INIT_LIST_HEAD(&cache->ro_list);
10233 INIT_LIST_HEAD(&cache->dirty_list);
10234 INIT_LIST_HEAD(&cache->io_list);
10235 btrfs_init_free_space_ctl(cache);
10236 atomic_set(&cache->trimming, 0);
10237 mutex_init(&cache->free_space_lock);
10238 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10245 * Iterate all chunks and verify that each of them has the corresponding block
10248 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10250 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
10251 struct extent_map *em;
10252 struct btrfs_block_group_cache *bg;
10257 read_lock(&map_tree->map_tree.lock);
10259 * lookup_extent_mapping will return the first extent map
10260 * intersecting the range, so setting @len to 1 is enough to
10261 * get the first chunk.
10263 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
10264 read_unlock(&map_tree->map_tree.lock);
10268 bg = btrfs_lookup_block_group(fs_info, em->start);
10271 "chunk start=%llu len=%llu doesn't have corresponding block group",
10272 em->start, em->len);
10274 free_extent_map(em);
10277 if (bg->key.objectid != em->start ||
10278 bg->key.offset != em->len ||
10279 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10280 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10282 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10283 em->start, em->len,
10284 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10285 bg->key.objectid, bg->key.offset,
10286 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10288 free_extent_map(em);
10289 btrfs_put_block_group(bg);
10292 start = em->start + em->len;
10293 free_extent_map(em);
10294 btrfs_put_block_group(bg);
10299 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10301 struct btrfs_path *path;
10303 struct btrfs_block_group_cache *cache;
10304 struct btrfs_space_info *space_info;
10305 struct btrfs_key key;
10306 struct btrfs_key found_key;
10307 struct extent_buffer *leaf;
10308 int need_clear = 0;
10313 feature = btrfs_super_incompat_flags(info->super_copy);
10314 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10318 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10319 path = btrfs_alloc_path();
10322 path->reada = READA_FORWARD;
10324 cache_gen = btrfs_super_cache_generation(info->super_copy);
10325 if (btrfs_test_opt(info, SPACE_CACHE) &&
10326 btrfs_super_generation(info->super_copy) != cache_gen)
10328 if (btrfs_test_opt(info, CLEAR_CACHE))
10332 ret = find_first_block_group(info, path, &key);
10338 leaf = path->nodes[0];
10339 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10341 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10350 * When we mount with old space cache, we need to
10351 * set BTRFS_DC_CLEAR and set dirty flag.
10353 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10354 * truncate the old free space cache inode and
10356 * b) Setting 'dirty flag' makes sure that we flush
10357 * the new space cache info onto disk.
10359 if (btrfs_test_opt(info, SPACE_CACHE))
10360 cache->disk_cache_state = BTRFS_DC_CLEAR;
10363 read_extent_buffer(leaf, &cache->item,
10364 btrfs_item_ptr_offset(leaf, path->slots[0]),
10365 sizeof(cache->item));
10366 cache->flags = btrfs_block_group_flags(&cache->item);
10368 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10369 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10371 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10372 cache->key.objectid);
10377 key.objectid = found_key.objectid + found_key.offset;
10378 btrfs_release_path(path);
10381 * We need to exclude the super stripes now so that the space
10382 * info has super bytes accounted for, otherwise we'll think
10383 * we have more space than we actually do.
10385 ret = exclude_super_stripes(cache);
10388 * We may have excluded something, so call this just in
10391 free_excluded_extents(cache);
10392 btrfs_put_block_group(cache);
10397 * check for two cases, either we are full, and therefore
10398 * don't need to bother with the caching work since we won't
10399 * find any space, or we are empty, and we can just add all
10400 * the space in and be done with it. This saves us _a_lot_ of
10401 * time, particularly in the full case.
10403 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10404 cache->last_byte_to_unpin = (u64)-1;
10405 cache->cached = BTRFS_CACHE_FINISHED;
10406 free_excluded_extents(cache);
10407 } else if (btrfs_block_group_used(&cache->item) == 0) {
10408 cache->last_byte_to_unpin = (u64)-1;
10409 cache->cached = BTRFS_CACHE_FINISHED;
10410 add_new_free_space(cache, found_key.objectid,
10411 found_key.objectid +
10413 free_excluded_extents(cache);
10416 ret = btrfs_add_block_group_cache(info, cache);
10418 btrfs_remove_free_space_cache(cache);
10419 btrfs_put_block_group(cache);
10423 trace_btrfs_add_block_group(info, cache, 0);
10424 update_space_info(info, cache->flags, found_key.offset,
10425 btrfs_block_group_used(&cache->item),
10426 cache->bytes_super, &space_info);
10428 cache->space_info = space_info;
10430 link_block_group(cache);
10432 set_avail_alloc_bits(info, cache->flags);
10433 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10434 inc_block_group_ro(cache, 1);
10435 } else if (btrfs_block_group_used(&cache->item) == 0) {
10436 ASSERT(list_empty(&cache->bg_list));
10437 btrfs_mark_bg_unused(cache);
10441 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10442 if (!(get_alloc_profile(info, space_info->flags) &
10443 (BTRFS_BLOCK_GROUP_RAID10 |
10444 BTRFS_BLOCK_GROUP_RAID1 |
10445 BTRFS_BLOCK_GROUP_RAID5 |
10446 BTRFS_BLOCK_GROUP_RAID6 |
10447 BTRFS_BLOCK_GROUP_DUP)))
10450 * avoid allocating from un-mirrored block group if there are
10451 * mirrored block groups.
10453 list_for_each_entry(cache,
10454 &space_info->block_groups[BTRFS_RAID_RAID0],
10456 inc_block_group_ro(cache, 1);
10457 list_for_each_entry(cache,
10458 &space_info->block_groups[BTRFS_RAID_SINGLE],
10460 inc_block_group_ro(cache, 1);
10463 btrfs_add_raid_kobjects(info);
10464 init_global_block_rsv(info);
10465 ret = check_chunk_block_group_mappings(info);
10467 btrfs_free_path(path);
10471 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10473 struct btrfs_fs_info *fs_info = trans->fs_info;
10474 struct btrfs_block_group_cache *block_group;
10475 struct btrfs_root *extent_root = fs_info->extent_root;
10476 struct btrfs_block_group_item item;
10477 struct btrfs_key key;
10480 if (!trans->can_flush_pending_bgs)
10483 while (!list_empty(&trans->new_bgs)) {
10484 block_group = list_first_entry(&trans->new_bgs,
10485 struct btrfs_block_group_cache,
10490 spin_lock(&block_group->lock);
10491 memcpy(&item, &block_group->item, sizeof(item));
10492 memcpy(&key, &block_group->key, sizeof(key));
10493 spin_unlock(&block_group->lock);
10495 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10498 btrfs_abort_transaction(trans, ret);
10499 ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset);
10501 btrfs_abort_transaction(trans, ret);
10502 add_block_group_free_space(trans, block_group);
10503 /* already aborted the transaction if it failed. */
10505 btrfs_delayed_refs_rsv_release(fs_info, 1);
10506 list_del_init(&block_group->bg_list);
10508 btrfs_trans_release_chunk_metadata(trans);
10511 int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used,
10512 u64 type, u64 chunk_offset, u64 size)
10514 struct btrfs_fs_info *fs_info = trans->fs_info;
10515 struct btrfs_block_group_cache *cache;
10518 btrfs_set_log_full_commit(fs_info, trans);
10520 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10524 btrfs_set_block_group_used(&cache->item, bytes_used);
10525 btrfs_set_block_group_chunk_objectid(&cache->item,
10526 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10527 btrfs_set_block_group_flags(&cache->item, type);
10529 cache->flags = type;
10530 cache->last_byte_to_unpin = (u64)-1;
10531 cache->cached = BTRFS_CACHE_FINISHED;
10532 cache->needs_free_space = 1;
10533 ret = exclude_super_stripes(cache);
10536 * We may have excluded something, so call this just in
10539 free_excluded_extents(cache);
10540 btrfs_put_block_group(cache);
10544 add_new_free_space(cache, chunk_offset, chunk_offset + size);
10546 free_excluded_extents(cache);
10548 #ifdef CONFIG_BTRFS_DEBUG
10549 if (btrfs_should_fragment_free_space(cache)) {
10550 u64 new_bytes_used = size - bytes_used;
10552 bytes_used += new_bytes_used >> 1;
10553 fragment_free_space(cache);
10557 * Ensure the corresponding space_info object is created and
10558 * assigned to our block group. We want our bg to be added to the rbtree
10559 * with its ->space_info set.
10561 cache->space_info = __find_space_info(fs_info, cache->flags);
10562 ASSERT(cache->space_info);
10564 ret = btrfs_add_block_group_cache(fs_info, cache);
10566 btrfs_remove_free_space_cache(cache);
10567 btrfs_put_block_group(cache);
10572 * Now that our block group has its ->space_info set and is inserted in
10573 * the rbtree, update the space info's counters.
10575 trace_btrfs_add_block_group(fs_info, cache, 1);
10576 update_space_info(fs_info, cache->flags, size, bytes_used,
10577 cache->bytes_super, &cache->space_info);
10578 update_global_block_rsv(fs_info);
10580 link_block_group(cache);
10582 list_add_tail(&cache->bg_list, &trans->new_bgs);
10583 trans->delayed_ref_updates++;
10584 btrfs_update_delayed_refs_rsv(trans);
10586 set_avail_alloc_bits(fs_info, type);
10590 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10592 u64 extra_flags = chunk_to_extended(flags) &
10593 BTRFS_EXTENDED_PROFILE_MASK;
10595 write_seqlock(&fs_info->profiles_lock);
10596 if (flags & BTRFS_BLOCK_GROUP_DATA)
10597 fs_info->avail_data_alloc_bits &= ~extra_flags;
10598 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10599 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10600 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10601 fs_info->avail_system_alloc_bits &= ~extra_flags;
10602 write_sequnlock(&fs_info->profiles_lock);
10605 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10606 u64 group_start, struct extent_map *em)
10608 struct btrfs_fs_info *fs_info = trans->fs_info;
10609 struct btrfs_root *root = fs_info->extent_root;
10610 struct btrfs_path *path;
10611 struct btrfs_block_group_cache *block_group;
10612 struct btrfs_free_cluster *cluster;
10613 struct btrfs_root *tree_root = fs_info->tree_root;
10614 struct btrfs_key key;
10615 struct inode *inode;
10616 struct kobject *kobj = NULL;
10620 struct btrfs_caching_control *caching_ctl = NULL;
10622 bool remove_rsv = false;
10624 block_group = btrfs_lookup_block_group(fs_info, group_start);
10625 BUG_ON(!block_group);
10626 BUG_ON(!block_group->ro);
10628 trace_btrfs_remove_block_group(block_group);
10630 * Free the reserved super bytes from this block group before
10633 free_excluded_extents(block_group);
10634 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10635 block_group->key.offset);
10637 memcpy(&key, &block_group->key, sizeof(key));
10638 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10639 factor = btrfs_bg_type_to_factor(block_group->flags);
10641 /* make sure this block group isn't part of an allocation cluster */
10642 cluster = &fs_info->data_alloc_cluster;
10643 spin_lock(&cluster->refill_lock);
10644 btrfs_return_cluster_to_free_space(block_group, cluster);
10645 spin_unlock(&cluster->refill_lock);
10648 * make sure this block group isn't part of a metadata
10649 * allocation cluster
10651 cluster = &fs_info->meta_alloc_cluster;
10652 spin_lock(&cluster->refill_lock);
10653 btrfs_return_cluster_to_free_space(block_group, cluster);
10654 spin_unlock(&cluster->refill_lock);
10656 path = btrfs_alloc_path();
10663 * get the inode first so any iput calls done for the io_list
10664 * aren't the final iput (no unlinks allowed now)
10666 inode = lookup_free_space_inode(fs_info, block_group, path);
10668 mutex_lock(&trans->transaction->cache_write_mutex);
10670 * Make sure our free space cache IO is done before removing the
10673 spin_lock(&trans->transaction->dirty_bgs_lock);
10674 if (!list_empty(&block_group->io_list)) {
10675 list_del_init(&block_group->io_list);
10677 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10679 spin_unlock(&trans->transaction->dirty_bgs_lock);
10680 btrfs_wait_cache_io(trans, block_group, path);
10681 btrfs_put_block_group(block_group);
10682 spin_lock(&trans->transaction->dirty_bgs_lock);
10685 if (!list_empty(&block_group->dirty_list)) {
10686 list_del_init(&block_group->dirty_list);
10688 btrfs_put_block_group(block_group);
10690 spin_unlock(&trans->transaction->dirty_bgs_lock);
10691 mutex_unlock(&trans->transaction->cache_write_mutex);
10693 if (!IS_ERR(inode)) {
10694 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10696 btrfs_add_delayed_iput(inode);
10699 clear_nlink(inode);
10700 /* One for the block groups ref */
10701 spin_lock(&block_group->lock);
10702 if (block_group->iref) {
10703 block_group->iref = 0;
10704 block_group->inode = NULL;
10705 spin_unlock(&block_group->lock);
10708 spin_unlock(&block_group->lock);
10710 /* One for our lookup ref */
10711 btrfs_add_delayed_iput(inode);
10714 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10715 key.offset = block_group->key.objectid;
10718 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10722 btrfs_release_path(path);
10724 ret = btrfs_del_item(trans, tree_root, path);
10727 btrfs_release_path(path);
10730 spin_lock(&fs_info->block_group_cache_lock);
10731 rb_erase(&block_group->cache_node,
10732 &fs_info->block_group_cache_tree);
10733 RB_CLEAR_NODE(&block_group->cache_node);
10735 if (fs_info->first_logical_byte == block_group->key.objectid)
10736 fs_info->first_logical_byte = (u64)-1;
10737 spin_unlock(&fs_info->block_group_cache_lock);
10739 down_write(&block_group->space_info->groups_sem);
10741 * we must use list_del_init so people can check to see if they
10742 * are still on the list after taking the semaphore
10744 list_del_init(&block_group->list);
10745 if (list_empty(&block_group->space_info->block_groups[index])) {
10746 kobj = block_group->space_info->block_group_kobjs[index];
10747 block_group->space_info->block_group_kobjs[index] = NULL;
10748 clear_avail_alloc_bits(fs_info, block_group->flags);
10750 up_write(&block_group->space_info->groups_sem);
10756 if (block_group->has_caching_ctl)
10757 caching_ctl = get_caching_control(block_group);
10758 if (block_group->cached == BTRFS_CACHE_STARTED)
10759 wait_block_group_cache_done(block_group);
10760 if (block_group->has_caching_ctl) {
10761 down_write(&fs_info->commit_root_sem);
10762 if (!caching_ctl) {
10763 struct btrfs_caching_control *ctl;
10765 list_for_each_entry(ctl,
10766 &fs_info->caching_block_groups, list)
10767 if (ctl->block_group == block_group) {
10769 refcount_inc(&caching_ctl->count);
10774 list_del_init(&caching_ctl->list);
10775 up_write(&fs_info->commit_root_sem);
10777 /* Once for the caching bgs list and once for us. */
10778 put_caching_control(caching_ctl);
10779 put_caching_control(caching_ctl);
10783 spin_lock(&trans->transaction->dirty_bgs_lock);
10784 if (!list_empty(&block_group->dirty_list)) {
10787 if (!list_empty(&block_group->io_list)) {
10790 spin_unlock(&trans->transaction->dirty_bgs_lock);
10791 btrfs_remove_free_space_cache(block_group);
10793 spin_lock(&block_group->space_info->lock);
10794 list_del_init(&block_group->ro_list);
10796 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10797 WARN_ON(block_group->space_info->total_bytes
10798 < block_group->key.offset);
10799 WARN_ON(block_group->space_info->bytes_readonly
10800 < block_group->key.offset);
10801 WARN_ON(block_group->space_info->disk_total
10802 < block_group->key.offset * factor);
10804 block_group->space_info->total_bytes -= block_group->key.offset;
10805 block_group->space_info->bytes_readonly -= block_group->key.offset;
10806 block_group->space_info->disk_total -= block_group->key.offset * factor;
10808 spin_unlock(&block_group->space_info->lock);
10810 memcpy(&key, &block_group->key, sizeof(key));
10812 mutex_lock(&fs_info->chunk_mutex);
10813 if (!list_empty(&em->list)) {
10814 /* We're in the transaction->pending_chunks list. */
10815 free_extent_map(em);
10817 spin_lock(&block_group->lock);
10818 block_group->removed = 1;
10820 * At this point trimming can't start on this block group, because we
10821 * removed the block group from the tree fs_info->block_group_cache_tree
10822 * so no one can't find it anymore and even if someone already got this
10823 * block group before we removed it from the rbtree, they have already
10824 * incremented block_group->trimming - if they didn't, they won't find
10825 * any free space entries because we already removed them all when we
10826 * called btrfs_remove_free_space_cache().
10828 * And we must not remove the extent map from the fs_info->mapping_tree
10829 * to prevent the same logical address range and physical device space
10830 * ranges from being reused for a new block group. This is because our
10831 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10832 * completely transactionless, so while it is trimming a range the
10833 * currently running transaction might finish and a new one start,
10834 * allowing for new block groups to be created that can reuse the same
10835 * physical device locations unless we take this special care.
10837 * There may also be an implicit trim operation if the file system
10838 * is mounted with -odiscard. The same protections must remain
10839 * in place until the extents have been discarded completely when
10840 * the transaction commit has completed.
10842 remove_em = (atomic_read(&block_group->trimming) == 0);
10844 * Make sure a trimmer task always sees the em in the pinned_chunks list
10845 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10846 * before checking block_group->removed).
10850 * Our em might be in trans->transaction->pending_chunks which
10851 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10852 * and so is the fs_info->pinned_chunks list.
10854 * So at this point we must be holding the chunk_mutex to avoid
10855 * any races with chunk allocation (more specifically at
10856 * volumes.c:contains_pending_extent()), to ensure it always
10857 * sees the em, either in the pending_chunks list or in the
10858 * pinned_chunks list.
10860 list_move_tail(&em->list, &fs_info->pinned_chunks);
10862 spin_unlock(&block_group->lock);
10865 struct extent_map_tree *em_tree;
10867 em_tree = &fs_info->mapping_tree.map_tree;
10868 write_lock(&em_tree->lock);
10870 * The em might be in the pending_chunks list, so make sure the
10871 * chunk mutex is locked, since remove_extent_mapping() will
10872 * delete us from that list.
10874 remove_extent_mapping(em_tree, em);
10875 write_unlock(&em_tree->lock);
10876 /* once for the tree */
10877 free_extent_map(em);
10880 mutex_unlock(&fs_info->chunk_mutex);
10882 ret = remove_block_group_free_space(trans, block_group);
10886 btrfs_put_block_group(block_group);
10887 btrfs_put_block_group(block_group);
10889 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10895 ret = btrfs_del_item(trans, root, path);
10898 btrfs_delayed_refs_rsv_release(fs_info, 1);
10899 btrfs_free_path(path);
10903 struct btrfs_trans_handle *
10904 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10905 const u64 chunk_offset)
10907 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10908 struct extent_map *em;
10909 struct map_lookup *map;
10910 unsigned int num_items;
10912 read_lock(&em_tree->lock);
10913 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10914 read_unlock(&em_tree->lock);
10915 ASSERT(em && em->start == chunk_offset);
10918 * We need to reserve 3 + N units from the metadata space info in order
10919 * to remove a block group (done at btrfs_remove_chunk() and at
10920 * btrfs_remove_block_group()), which are used for:
10922 * 1 unit for adding the free space inode's orphan (located in the tree
10924 * 1 unit for deleting the block group item (located in the extent
10926 * 1 unit for deleting the free space item (located in tree of tree
10928 * N units for deleting N device extent items corresponding to each
10929 * stripe (located in the device tree).
10931 * In order to remove a block group we also need to reserve units in the
10932 * system space info in order to update the chunk tree (update one or
10933 * more device items and remove one chunk item), but this is done at
10934 * btrfs_remove_chunk() through a call to check_system_chunk().
10936 map = em->map_lookup;
10937 num_items = 3 + map->num_stripes;
10938 free_extent_map(em);
10940 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10945 * Process the unused_bgs list and remove any that don't have any allocated
10946 * space inside of them.
10948 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10950 struct btrfs_block_group_cache *block_group;
10951 struct btrfs_space_info *space_info;
10952 struct btrfs_trans_handle *trans;
10955 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10958 spin_lock(&fs_info->unused_bgs_lock);
10959 while (!list_empty(&fs_info->unused_bgs)) {
10963 block_group = list_first_entry(&fs_info->unused_bgs,
10964 struct btrfs_block_group_cache,
10966 list_del_init(&block_group->bg_list);
10968 space_info = block_group->space_info;
10970 if (ret || btrfs_mixed_space_info(space_info)) {
10971 btrfs_put_block_group(block_group);
10974 spin_unlock(&fs_info->unused_bgs_lock);
10976 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10978 /* Don't want to race with allocators so take the groups_sem */
10979 down_write(&space_info->groups_sem);
10980 spin_lock(&block_group->lock);
10981 if (block_group->reserved || block_group->pinned ||
10982 btrfs_block_group_used(&block_group->item) ||
10984 list_is_singular(&block_group->list)) {
10986 * We want to bail if we made new allocations or have
10987 * outstanding allocations in this block group. We do
10988 * the ro check in case balance is currently acting on
10989 * this block group.
10991 trace_btrfs_skip_unused_block_group(block_group);
10992 spin_unlock(&block_group->lock);
10993 up_write(&space_info->groups_sem);
10996 spin_unlock(&block_group->lock);
10998 /* We don't want to force the issue, only flip if it's ok. */
10999 ret = inc_block_group_ro(block_group, 0);
11000 up_write(&space_info->groups_sem);
11007 * Want to do this before we do anything else so we can recover
11008 * properly if we fail to join the transaction.
11010 trans = btrfs_start_trans_remove_block_group(fs_info,
11011 block_group->key.objectid);
11012 if (IS_ERR(trans)) {
11013 btrfs_dec_block_group_ro(block_group);
11014 ret = PTR_ERR(trans);
11019 * We could have pending pinned extents for this block group,
11020 * just delete them, we don't care about them anymore.
11022 start = block_group->key.objectid;
11023 end = start + block_group->key.offset - 1;
11025 * Hold the unused_bg_unpin_mutex lock to avoid racing with
11026 * btrfs_finish_extent_commit(). If we are at transaction N,
11027 * another task might be running finish_extent_commit() for the
11028 * previous transaction N - 1, and have seen a range belonging
11029 * to the block group in freed_extents[] before we were able to
11030 * clear the whole block group range from freed_extents[]. This
11031 * means that task can lookup for the block group after we
11032 * unpinned it from freed_extents[] and removed it, leading to
11033 * a BUG_ON() at btrfs_unpin_extent_range().
11035 mutex_lock(&fs_info->unused_bg_unpin_mutex);
11036 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
11039 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11040 btrfs_dec_block_group_ro(block_group);
11043 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
11046 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11047 btrfs_dec_block_group_ro(block_group);
11050 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
11052 /* Reset pinned so btrfs_put_block_group doesn't complain */
11053 spin_lock(&space_info->lock);
11054 spin_lock(&block_group->lock);
11056 update_bytes_pinned(space_info, -block_group->pinned);
11057 space_info->bytes_readonly += block_group->pinned;
11058 percpu_counter_add_batch(&space_info->total_bytes_pinned,
11059 -block_group->pinned,
11060 BTRFS_TOTAL_BYTES_PINNED_BATCH);
11061 block_group->pinned = 0;
11063 spin_unlock(&block_group->lock);
11064 spin_unlock(&space_info->lock);
11066 /* DISCARD can flip during remount */
11067 trimming = btrfs_test_opt(fs_info, DISCARD);
11069 /* Implicit trim during transaction commit. */
11071 btrfs_get_block_group_trimming(block_group);
11074 * Btrfs_remove_chunk will abort the transaction if things go
11077 ret = btrfs_remove_chunk(trans, block_group->key.objectid);
11081 btrfs_put_block_group_trimming(block_group);
11086 * If we're not mounted with -odiscard, we can just forget
11087 * about this block group. Otherwise we'll need to wait
11088 * until transaction commit to do the actual discard.
11091 spin_lock(&fs_info->unused_bgs_lock);
11093 * A concurrent scrub might have added us to the list
11094 * fs_info->unused_bgs, so use a list_move operation
11095 * to add the block group to the deleted_bgs list.
11097 list_move(&block_group->bg_list,
11098 &trans->transaction->deleted_bgs);
11099 spin_unlock(&fs_info->unused_bgs_lock);
11100 btrfs_get_block_group(block_group);
11103 btrfs_end_transaction(trans);
11105 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
11106 btrfs_put_block_group(block_group);
11107 spin_lock(&fs_info->unused_bgs_lock);
11109 spin_unlock(&fs_info->unused_bgs_lock);
11112 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
11114 struct btrfs_super_block *disk_super;
11120 disk_super = fs_info->super_copy;
11121 if (!btrfs_super_root(disk_super))
11124 features = btrfs_super_incompat_flags(disk_super);
11125 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11128 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11129 ret = create_space_info(fs_info, flags);
11134 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11135 ret = create_space_info(fs_info, flags);
11137 flags = BTRFS_BLOCK_GROUP_METADATA;
11138 ret = create_space_info(fs_info, flags);
11142 flags = BTRFS_BLOCK_GROUP_DATA;
11143 ret = create_space_info(fs_info, flags);
11149 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
11150 u64 start, u64 end)
11152 return unpin_extent_range(fs_info, start, end, false);
11156 * It used to be that old block groups would be left around forever.
11157 * Iterating over them would be enough to trim unused space. Since we
11158 * now automatically remove them, we also need to iterate over unallocated
11161 * We don't want a transaction for this since the discard may take a
11162 * substantial amount of time. We don't require that a transaction be
11163 * running, but we do need to take a running transaction into account
11164 * to ensure that we're not discarding chunks that were released or
11165 * allocated in the current transaction.
11167 * Holding the chunks lock will prevent other threads from allocating
11168 * or releasing chunks, but it won't prevent a running transaction
11169 * from committing and releasing the memory that the pending chunks
11170 * list head uses. For that, we need to take a reference to the
11171 * transaction and hold the commit root sem. We only need to hold
11172 * it while performing the free space search since we have already
11173 * held back allocations.
11175 static int btrfs_trim_free_extents(struct btrfs_device *device,
11176 u64 minlen, u64 *trimmed)
11178 u64 start = 0, len = 0;
11183 /* Discard not supported = nothing to do. */
11184 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
11187 /* Not writable = nothing to do. */
11188 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
11191 /* No free space = nothing to do. */
11192 if (device->total_bytes <= device->bytes_used)
11198 struct btrfs_fs_info *fs_info = device->fs_info;
11199 struct btrfs_transaction *trans;
11202 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11206 ret = down_read_killable(&fs_info->commit_root_sem);
11208 mutex_unlock(&fs_info->chunk_mutex);
11212 spin_lock(&fs_info->trans_lock);
11213 trans = fs_info->running_transaction;
11215 refcount_inc(&trans->use_count);
11216 spin_unlock(&fs_info->trans_lock);
11219 up_read(&fs_info->commit_root_sem);
11221 ret = find_free_dev_extent_start(trans, device, minlen, start,
11224 up_read(&fs_info->commit_root_sem);
11225 btrfs_put_transaction(trans);
11229 mutex_unlock(&fs_info->chunk_mutex);
11230 if (ret == -ENOSPC)
11235 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11236 mutex_unlock(&fs_info->chunk_mutex);
11244 if (fatal_signal_pending(current)) {
11245 ret = -ERESTARTSYS;
11256 * Trim the whole filesystem by:
11257 * 1) trimming the free space in each block group
11258 * 2) trimming the unallocated space on each device
11260 * This will also continue trimming even if a block group or device encounters
11261 * an error. The return value will be the last error, or 0 if nothing bad
11264 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11266 struct btrfs_block_group_cache *cache = NULL;
11267 struct btrfs_device *device;
11268 struct list_head *devices;
11274 u64 dev_failed = 0;
11279 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11280 for (; cache; cache = next_block_group(fs_info, cache)) {
11281 if (cache->key.objectid >= (range->start + range->len)) {
11282 btrfs_put_block_group(cache);
11286 start = max(range->start, cache->key.objectid);
11287 end = min(range->start + range->len,
11288 cache->key.objectid + cache->key.offset);
11290 if (end - start >= range->minlen) {
11291 if (!block_group_cache_done(cache)) {
11292 ret = cache_block_group(cache, 0);
11298 ret = wait_block_group_cache_done(cache);
11305 ret = btrfs_trim_block_group(cache,
11311 trimmed += group_trimmed;
11321 btrfs_warn(fs_info,
11322 "failed to trim %llu block group(s), last error %d",
11323 bg_failed, bg_ret);
11324 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11325 devices = &fs_info->fs_devices->devices;
11326 list_for_each_entry(device, devices, dev_list) {
11327 ret = btrfs_trim_free_extents(device, range->minlen,
11335 trimmed += group_trimmed;
11337 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11340 btrfs_warn(fs_info,
11341 "failed to trim %llu device(s), last error %d",
11342 dev_failed, dev_ret);
11343 range->len = trimmed;
11350 * btrfs_{start,end}_write_no_snapshotting() are similar to
11351 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11352 * data into the page cache through nocow before the subvolume is snapshoted,
11353 * but flush the data into disk after the snapshot creation, or to prevent
11354 * operations while snapshotting is ongoing and that cause the snapshot to be
11355 * inconsistent (writes followed by expanding truncates for example).
11357 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11359 percpu_counter_dec(&root->subv_writers->counter);
11360 cond_wake_up(&root->subv_writers->wait);
11363 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11365 if (atomic_read(&root->will_be_snapshotted))
11368 percpu_counter_inc(&root->subv_writers->counter);
11370 * Make sure counter is updated before we check for snapshot creation.
11373 if (atomic_read(&root->will_be_snapshotted)) {
11374 btrfs_end_write_no_snapshotting(root);
11380 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11385 ret = btrfs_start_write_no_snapshotting(root);
11388 wait_var_event(&root->will_be_snapshotted,
11389 !atomic_read(&root->will_be_snapshotted));
11393 void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg)
11395 struct btrfs_fs_info *fs_info = bg->fs_info;
11397 spin_lock(&fs_info->unused_bgs_lock);
11398 if (list_empty(&bg->bg_list)) {
11399 btrfs_get_block_group(bg);
11400 trace_btrfs_add_unused_block_group(bg);
11401 list_add_tail(&bg->bg_list, &fs_info->unused_bgs);
11403 spin_unlock(&fs_info->unused_bgs_lock);