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,
54 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
55 struct btrfs_fs_info *fs_info,
56 struct btrfs_delayed_ref_node *node, u64 parent,
57 u64 root_objectid, u64 owner_objectid,
58 u64 owner_offset, int refs_to_drop,
59 struct btrfs_delayed_extent_op *extra_op);
60 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
61 struct extent_buffer *leaf,
62 struct btrfs_extent_item *ei);
63 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
64 struct btrfs_fs_info *fs_info,
65 u64 parent, u64 root_objectid,
66 u64 flags, u64 owner, u64 offset,
67 struct btrfs_key *ins, int ref_mod);
68 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
69 struct btrfs_fs_info *fs_info,
70 u64 parent, u64 root_objectid,
71 u64 flags, struct btrfs_disk_key *key,
72 int level, struct btrfs_key *ins);
73 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
74 struct btrfs_fs_info *fs_info, u64 flags,
76 static int find_next_key(struct btrfs_path *path, int level,
77 struct btrfs_key *key);
78 static void dump_space_info(struct btrfs_fs_info *fs_info,
79 struct btrfs_space_info *info, u64 bytes,
80 int dump_block_groups);
81 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
83 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
84 struct btrfs_space_info *space_info,
86 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
87 struct btrfs_space_info *space_info,
91 block_group_cache_done(struct btrfs_block_group_cache *cache)
94 return cache->cached == BTRFS_CACHE_FINISHED ||
95 cache->cached == BTRFS_CACHE_ERROR;
98 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
100 return (cache->flags & bits) == bits;
103 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
105 atomic_inc(&cache->count);
108 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
110 if (atomic_dec_and_test(&cache->count)) {
111 WARN_ON(cache->pinned > 0);
112 WARN_ON(cache->reserved > 0);
115 * If not empty, someone is still holding mutex of
116 * full_stripe_lock, which can only be released by caller.
117 * And it will definitely cause use-after-free when caller
118 * tries to release full stripe lock.
120 * No better way to resolve, but only to warn.
122 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
123 kfree(cache->free_space_ctl);
129 * this adds the block group to the fs_info rb tree for the block group
132 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
133 struct btrfs_block_group_cache *block_group)
136 struct rb_node *parent = NULL;
137 struct btrfs_block_group_cache *cache;
139 spin_lock(&info->block_group_cache_lock);
140 p = &info->block_group_cache_tree.rb_node;
144 cache = rb_entry(parent, struct btrfs_block_group_cache,
146 if (block_group->key.objectid < cache->key.objectid) {
148 } else if (block_group->key.objectid > cache->key.objectid) {
151 spin_unlock(&info->block_group_cache_lock);
156 rb_link_node(&block_group->cache_node, parent, p);
157 rb_insert_color(&block_group->cache_node,
158 &info->block_group_cache_tree);
160 if (info->first_logical_byte > block_group->key.objectid)
161 info->first_logical_byte = block_group->key.objectid;
163 spin_unlock(&info->block_group_cache_lock);
169 * This will return the block group at or after bytenr if contains is 0, else
170 * it will return the block group that contains the bytenr
172 static struct btrfs_block_group_cache *
173 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
176 struct btrfs_block_group_cache *cache, *ret = NULL;
180 spin_lock(&info->block_group_cache_lock);
181 n = info->block_group_cache_tree.rb_node;
184 cache = rb_entry(n, struct btrfs_block_group_cache,
186 end = cache->key.objectid + cache->key.offset - 1;
187 start = cache->key.objectid;
189 if (bytenr < start) {
190 if (!contains && (!ret || start < ret->key.objectid))
193 } else if (bytenr > start) {
194 if (contains && bytenr <= end) {
205 btrfs_get_block_group(ret);
206 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
207 info->first_logical_byte = ret->key.objectid;
209 spin_unlock(&info->block_group_cache_lock);
214 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
215 u64 start, u64 num_bytes)
217 u64 end = start + num_bytes - 1;
218 set_extent_bits(&fs_info->freed_extents[0],
219 start, end, EXTENT_UPTODATE);
220 set_extent_bits(&fs_info->freed_extents[1],
221 start, end, EXTENT_UPTODATE);
225 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
226 struct btrfs_block_group_cache *cache)
230 start = cache->key.objectid;
231 end = start + cache->key.offset - 1;
233 clear_extent_bits(&fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE);
235 clear_extent_bits(&fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE);
239 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
240 struct btrfs_block_group_cache *cache)
247 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
248 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
249 cache->bytes_super += stripe_len;
250 ret = add_excluded_extent(fs_info, cache->key.objectid,
256 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
257 bytenr = btrfs_sb_offset(i);
258 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
259 bytenr, 0, &logical, &nr, &stripe_len);
266 if (logical[nr] > cache->key.objectid +
270 if (logical[nr] + stripe_len <= cache->key.objectid)
274 if (start < cache->key.objectid) {
275 start = cache->key.objectid;
276 len = (logical[nr] + stripe_len) - start;
278 len = min_t(u64, stripe_len,
279 cache->key.objectid +
280 cache->key.offset - start);
283 cache->bytes_super += len;
284 ret = add_excluded_extent(fs_info, start, len);
296 static struct btrfs_caching_control *
297 get_caching_control(struct btrfs_block_group_cache *cache)
299 struct btrfs_caching_control *ctl;
301 spin_lock(&cache->lock);
302 if (!cache->caching_ctl) {
303 spin_unlock(&cache->lock);
307 ctl = cache->caching_ctl;
308 refcount_inc(&ctl->count);
309 spin_unlock(&cache->lock);
313 static void put_caching_control(struct btrfs_caching_control *ctl)
315 if (refcount_dec_and_test(&ctl->count))
319 #ifdef CONFIG_BTRFS_DEBUG
320 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
322 struct btrfs_fs_info *fs_info = block_group->fs_info;
323 u64 start = block_group->key.objectid;
324 u64 len = block_group->key.offset;
325 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
326 fs_info->nodesize : fs_info->sectorsize;
327 u64 step = chunk << 1;
329 while (len > chunk) {
330 btrfs_remove_free_space(block_group, start, chunk);
341 * this is only called by cache_block_group, since we could have freed extents
342 * we need to check the pinned_extents for any extents that can't be used yet
343 * since their free space will be released as soon as the transaction commits.
345 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
346 struct btrfs_fs_info *info, u64 start, u64 end)
348 u64 extent_start, extent_end, size, total_added = 0;
351 while (start < end) {
352 ret = find_first_extent_bit(info->pinned_extents, start,
353 &extent_start, &extent_end,
354 EXTENT_DIRTY | EXTENT_UPTODATE,
359 if (extent_start <= start) {
360 start = extent_end + 1;
361 } else if (extent_start > start && extent_start < end) {
362 size = extent_start - start;
364 ret = btrfs_add_free_space(block_group, start,
366 BUG_ON(ret); /* -ENOMEM or logic error */
367 start = extent_end + 1;
376 ret = btrfs_add_free_space(block_group, start, size);
377 BUG_ON(ret); /* -ENOMEM or logic error */
383 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
385 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
386 struct btrfs_fs_info *fs_info = block_group->fs_info;
387 struct btrfs_root *extent_root = fs_info->extent_root;
388 struct btrfs_path *path;
389 struct extent_buffer *leaf;
390 struct btrfs_key key;
397 path = btrfs_alloc_path();
401 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
403 #ifdef CONFIG_BTRFS_DEBUG
405 * If we're fragmenting we don't want to make anybody think we can
406 * allocate from this block group until we've had a chance to fragment
409 if (btrfs_should_fragment_free_space(block_group))
413 * We don't want to deadlock with somebody trying to allocate a new
414 * extent for the extent root while also trying to search the extent
415 * root to add free space. So we skip locking and search the commit
416 * root, since its read-only
418 path->skip_locking = 1;
419 path->search_commit_root = 1;
420 path->reada = READA_FORWARD;
424 key.type = BTRFS_EXTENT_ITEM_KEY;
427 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
431 leaf = path->nodes[0];
432 nritems = btrfs_header_nritems(leaf);
435 if (btrfs_fs_closing(fs_info) > 1) {
440 if (path->slots[0] < nritems) {
441 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
443 ret = find_next_key(path, 0, &key);
447 if (need_resched() ||
448 rwsem_is_contended(&fs_info->commit_root_sem)) {
450 caching_ctl->progress = last;
451 btrfs_release_path(path);
452 up_read(&fs_info->commit_root_sem);
453 mutex_unlock(&caching_ctl->mutex);
455 mutex_lock(&caching_ctl->mutex);
456 down_read(&fs_info->commit_root_sem);
460 ret = btrfs_next_leaf(extent_root, path);
465 leaf = path->nodes[0];
466 nritems = btrfs_header_nritems(leaf);
470 if (key.objectid < last) {
473 key.type = BTRFS_EXTENT_ITEM_KEY;
476 caching_ctl->progress = last;
477 btrfs_release_path(path);
481 if (key.objectid < block_group->key.objectid) {
486 if (key.objectid >= block_group->key.objectid +
487 block_group->key.offset)
490 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
491 key.type == BTRFS_METADATA_ITEM_KEY) {
492 total_found += add_new_free_space(block_group,
495 if (key.type == BTRFS_METADATA_ITEM_KEY)
496 last = key.objectid +
499 last = key.objectid + key.offset;
501 if (total_found > CACHING_CTL_WAKE_UP) {
504 wake_up(&caching_ctl->wait);
511 total_found += add_new_free_space(block_group, fs_info, last,
512 block_group->key.objectid +
513 block_group->key.offset);
514 caching_ctl->progress = (u64)-1;
517 btrfs_free_path(path);
521 static noinline void caching_thread(struct btrfs_work *work)
523 struct btrfs_block_group_cache *block_group;
524 struct btrfs_fs_info *fs_info;
525 struct btrfs_caching_control *caching_ctl;
528 caching_ctl = container_of(work, struct btrfs_caching_control, work);
529 block_group = caching_ctl->block_group;
530 fs_info = block_group->fs_info;
532 mutex_lock(&caching_ctl->mutex);
533 down_read(&fs_info->commit_root_sem);
535 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
536 ret = load_free_space_tree(caching_ctl);
538 ret = load_extent_tree_free(caching_ctl);
540 spin_lock(&block_group->lock);
541 block_group->caching_ctl = NULL;
542 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
543 spin_unlock(&block_group->lock);
545 #ifdef CONFIG_BTRFS_DEBUG
546 if (btrfs_should_fragment_free_space(block_group)) {
549 spin_lock(&block_group->space_info->lock);
550 spin_lock(&block_group->lock);
551 bytes_used = block_group->key.offset -
552 btrfs_block_group_used(&block_group->item);
553 block_group->space_info->bytes_used += bytes_used >> 1;
554 spin_unlock(&block_group->lock);
555 spin_unlock(&block_group->space_info->lock);
556 fragment_free_space(block_group);
560 caching_ctl->progress = (u64)-1;
562 up_read(&fs_info->commit_root_sem);
563 free_excluded_extents(fs_info, block_group);
564 mutex_unlock(&caching_ctl->mutex);
566 wake_up(&caching_ctl->wait);
568 put_caching_control(caching_ctl);
569 btrfs_put_block_group(block_group);
572 static int cache_block_group(struct btrfs_block_group_cache *cache,
576 struct btrfs_fs_info *fs_info = cache->fs_info;
577 struct btrfs_caching_control *caching_ctl;
580 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
584 INIT_LIST_HEAD(&caching_ctl->list);
585 mutex_init(&caching_ctl->mutex);
586 init_waitqueue_head(&caching_ctl->wait);
587 caching_ctl->block_group = cache;
588 caching_ctl->progress = cache->key.objectid;
589 refcount_set(&caching_ctl->count, 1);
590 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
591 caching_thread, NULL, NULL);
593 spin_lock(&cache->lock);
595 * This should be a rare occasion, but this could happen I think in the
596 * case where one thread starts to load the space cache info, and then
597 * some other thread starts a transaction commit which tries to do an
598 * allocation while the other thread is still loading the space cache
599 * info. The previous loop should have kept us from choosing this block
600 * group, but if we've moved to the state where we will wait on caching
601 * block groups we need to first check if we're doing a fast load here,
602 * so we can wait for it to finish, otherwise we could end up allocating
603 * from a block group who's cache gets evicted for one reason or
606 while (cache->cached == BTRFS_CACHE_FAST) {
607 struct btrfs_caching_control *ctl;
609 ctl = cache->caching_ctl;
610 refcount_inc(&ctl->count);
611 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
612 spin_unlock(&cache->lock);
616 finish_wait(&ctl->wait, &wait);
617 put_caching_control(ctl);
618 spin_lock(&cache->lock);
621 if (cache->cached != BTRFS_CACHE_NO) {
622 spin_unlock(&cache->lock);
626 WARN_ON(cache->caching_ctl);
627 cache->caching_ctl = caching_ctl;
628 cache->cached = BTRFS_CACHE_FAST;
629 spin_unlock(&cache->lock);
631 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
632 mutex_lock(&caching_ctl->mutex);
633 ret = load_free_space_cache(fs_info, cache);
635 spin_lock(&cache->lock);
637 cache->caching_ctl = NULL;
638 cache->cached = BTRFS_CACHE_FINISHED;
639 cache->last_byte_to_unpin = (u64)-1;
640 caching_ctl->progress = (u64)-1;
642 if (load_cache_only) {
643 cache->caching_ctl = NULL;
644 cache->cached = BTRFS_CACHE_NO;
646 cache->cached = BTRFS_CACHE_STARTED;
647 cache->has_caching_ctl = 1;
650 spin_unlock(&cache->lock);
651 #ifdef CONFIG_BTRFS_DEBUG
653 btrfs_should_fragment_free_space(cache)) {
656 spin_lock(&cache->space_info->lock);
657 spin_lock(&cache->lock);
658 bytes_used = cache->key.offset -
659 btrfs_block_group_used(&cache->item);
660 cache->space_info->bytes_used += bytes_used >> 1;
661 spin_unlock(&cache->lock);
662 spin_unlock(&cache->space_info->lock);
663 fragment_free_space(cache);
666 mutex_unlock(&caching_ctl->mutex);
668 wake_up(&caching_ctl->wait);
670 put_caching_control(caching_ctl);
671 free_excluded_extents(fs_info, cache);
676 * We're either using the free space tree or no caching at all.
677 * Set cached to the appropriate value and wakeup any waiters.
679 spin_lock(&cache->lock);
680 if (load_cache_only) {
681 cache->caching_ctl = NULL;
682 cache->cached = BTRFS_CACHE_NO;
684 cache->cached = BTRFS_CACHE_STARTED;
685 cache->has_caching_ctl = 1;
687 spin_unlock(&cache->lock);
688 wake_up(&caching_ctl->wait);
691 if (load_cache_only) {
692 put_caching_control(caching_ctl);
696 down_write(&fs_info->commit_root_sem);
697 refcount_inc(&caching_ctl->count);
698 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
699 up_write(&fs_info->commit_root_sem);
701 btrfs_get_block_group(cache);
703 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
709 * return the block group that starts at or after bytenr
711 static struct btrfs_block_group_cache *
712 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
714 return block_group_cache_tree_search(info, bytenr, 0);
718 * return the block group that contains the given bytenr
720 struct btrfs_block_group_cache *btrfs_lookup_block_group(
721 struct btrfs_fs_info *info,
724 return block_group_cache_tree_search(info, bytenr, 1);
727 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
730 struct list_head *head = &info->space_info;
731 struct btrfs_space_info *found;
733 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
736 list_for_each_entry_rcu(found, head, list) {
737 if (found->flags & flags) {
746 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
747 u64 owner, u64 root_objectid)
749 struct btrfs_space_info *space_info;
752 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
753 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
754 flags = BTRFS_BLOCK_GROUP_SYSTEM;
756 flags = BTRFS_BLOCK_GROUP_METADATA;
758 flags = BTRFS_BLOCK_GROUP_DATA;
761 space_info = __find_space_info(fs_info, flags);
763 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
767 * after adding space to the filesystem, we need to clear the full flags
768 * on all the space infos.
770 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
772 struct list_head *head = &info->space_info;
773 struct btrfs_space_info *found;
776 list_for_each_entry_rcu(found, head, list)
781 /* simple helper to search for an existing data extent at a given offset */
782 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
785 struct btrfs_key key;
786 struct btrfs_path *path;
788 path = btrfs_alloc_path();
792 key.objectid = start;
794 key.type = BTRFS_EXTENT_ITEM_KEY;
795 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
796 btrfs_free_path(path);
801 * helper function to lookup reference count and flags of a tree block.
803 * the head node for delayed ref is used to store the sum of all the
804 * reference count modifications queued up in the rbtree. the head
805 * node may also store the extent flags to set. This way you can check
806 * to see what the reference count and extent flags would be if all of
807 * the delayed refs are not processed.
809 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
810 struct btrfs_fs_info *fs_info, u64 bytenr,
811 u64 offset, int metadata, u64 *refs, u64 *flags)
813 struct btrfs_delayed_ref_head *head;
814 struct btrfs_delayed_ref_root *delayed_refs;
815 struct btrfs_path *path;
816 struct btrfs_extent_item *ei;
817 struct extent_buffer *leaf;
818 struct btrfs_key key;
825 * If we don't have skinny metadata, don't bother doing anything
828 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
829 offset = fs_info->nodesize;
833 path = btrfs_alloc_path();
838 path->skip_locking = 1;
839 path->search_commit_root = 1;
843 key.objectid = bytenr;
846 key.type = BTRFS_METADATA_ITEM_KEY;
848 key.type = BTRFS_EXTENT_ITEM_KEY;
850 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
854 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
855 if (path->slots[0]) {
857 btrfs_item_key_to_cpu(path->nodes[0], &key,
859 if (key.objectid == bytenr &&
860 key.type == BTRFS_EXTENT_ITEM_KEY &&
861 key.offset == fs_info->nodesize)
867 leaf = path->nodes[0];
868 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
869 if (item_size >= sizeof(*ei)) {
870 ei = btrfs_item_ptr(leaf, path->slots[0],
871 struct btrfs_extent_item);
872 num_refs = btrfs_extent_refs(leaf, ei);
873 extent_flags = btrfs_extent_flags(leaf, ei);
875 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
876 struct btrfs_extent_item_v0 *ei0;
877 BUG_ON(item_size != sizeof(*ei0));
878 ei0 = btrfs_item_ptr(leaf, path->slots[0],
879 struct btrfs_extent_item_v0);
880 num_refs = btrfs_extent_refs_v0(leaf, ei0);
881 /* FIXME: this isn't correct for data */
882 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
887 BUG_ON(num_refs == 0);
897 delayed_refs = &trans->transaction->delayed_refs;
898 spin_lock(&delayed_refs->lock);
899 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
901 if (!mutex_trylock(&head->mutex)) {
902 refcount_inc(&head->refs);
903 spin_unlock(&delayed_refs->lock);
905 btrfs_release_path(path);
908 * Mutex was contended, block until it's released and try
911 mutex_lock(&head->mutex);
912 mutex_unlock(&head->mutex);
913 btrfs_put_delayed_ref_head(head);
916 spin_lock(&head->lock);
917 if (head->extent_op && head->extent_op->update_flags)
918 extent_flags |= head->extent_op->flags_to_set;
920 BUG_ON(num_refs == 0);
922 num_refs += head->ref_mod;
923 spin_unlock(&head->lock);
924 mutex_unlock(&head->mutex);
926 spin_unlock(&delayed_refs->lock);
928 WARN_ON(num_refs == 0);
932 *flags = extent_flags;
934 btrfs_free_path(path);
939 * Back reference rules. Back refs have three main goals:
941 * 1) differentiate between all holders of references to an extent so that
942 * when a reference is dropped we can make sure it was a valid reference
943 * before freeing the extent.
945 * 2) Provide enough information to quickly find the holders of an extent
946 * if we notice a given block is corrupted or bad.
948 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
949 * maintenance. This is actually the same as #2, but with a slightly
950 * different use case.
952 * There are two kinds of back refs. The implicit back refs is optimized
953 * for pointers in non-shared tree blocks. For a given pointer in a block,
954 * back refs of this kind provide information about the block's owner tree
955 * and the pointer's key. These information allow us to find the block by
956 * b-tree searching. The full back refs is for pointers in tree blocks not
957 * referenced by their owner trees. The location of tree block is recorded
958 * in the back refs. Actually the full back refs is generic, and can be
959 * used in all cases the implicit back refs is used. The major shortcoming
960 * of the full back refs is its overhead. Every time a tree block gets
961 * COWed, we have to update back refs entry for all pointers in it.
963 * For a newly allocated tree block, we use implicit back refs for
964 * pointers in it. This means most tree related operations only involve
965 * implicit back refs. For a tree block created in old transaction, the
966 * only way to drop a reference to it is COW it. So we can detect the
967 * event that tree block loses its owner tree's reference and do the
968 * back refs conversion.
970 * When a tree block is COWed through a tree, there are four cases:
972 * The reference count of the block is one and the tree is the block's
973 * owner tree. Nothing to do in this case.
975 * The reference count of the block is one and the tree is not the
976 * block's owner tree. In this case, full back refs is used for pointers
977 * in the block. Remove these full back refs, add implicit back refs for
978 * every pointers in the new block.
980 * The reference count of the block is greater than one and the tree is
981 * the block's owner tree. In this case, implicit back refs is used for
982 * pointers in the block. Add full back refs for every pointers in the
983 * block, increase lower level extents' reference counts. The original
984 * implicit back refs are entailed to the new block.
986 * The reference count of the block is greater than one and the tree is
987 * not the block's owner tree. Add implicit back refs for every pointer in
988 * the new block, increase lower level extents' reference count.
990 * Back Reference Key composing:
992 * The key objectid corresponds to the first byte in the extent,
993 * The key type is used to differentiate between types of back refs.
994 * There are different meanings of the key offset for different types
997 * File extents can be referenced by:
999 * - multiple snapshots, subvolumes, or different generations in one subvol
1000 * - different files inside a single subvolume
1001 * - different offsets inside a file (bookend extents in file.c)
1003 * The extent ref structure for the implicit back refs has fields for:
1005 * - Objectid of the subvolume root
1006 * - objectid of the file holding the reference
1007 * - original offset in the file
1008 * - how many bookend extents
1010 * The key offset for the implicit back refs is hash of the first
1013 * The extent ref structure for the full back refs has field for:
1015 * - number of pointers in the tree leaf
1017 * The key offset for the implicit back refs is the first byte of
1020 * When a file extent is allocated, The implicit back refs is used.
1021 * the fields are filled in:
1023 * (root_key.objectid, inode objectid, offset in file, 1)
1025 * When a file extent is removed file truncation, we find the
1026 * corresponding implicit back refs and check the following fields:
1028 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1030 * Btree extents can be referenced by:
1032 * - Different subvolumes
1034 * Both the implicit back refs and the full back refs for tree blocks
1035 * only consist of key. The key offset for the implicit back refs is
1036 * objectid of block's owner tree. The key offset for the full back refs
1037 * is the first byte of parent block.
1039 * When implicit back refs is used, information about the lowest key and
1040 * level of the tree block are required. These information are stored in
1041 * tree block info structure.
1044 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1045 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1046 struct btrfs_fs_info *fs_info,
1047 struct btrfs_path *path,
1048 u64 owner, u32 extra_size)
1050 struct btrfs_root *root = fs_info->extent_root;
1051 struct btrfs_extent_item *item;
1052 struct btrfs_extent_item_v0 *ei0;
1053 struct btrfs_extent_ref_v0 *ref0;
1054 struct btrfs_tree_block_info *bi;
1055 struct extent_buffer *leaf;
1056 struct btrfs_key key;
1057 struct btrfs_key found_key;
1058 u32 new_size = sizeof(*item);
1062 leaf = path->nodes[0];
1063 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1065 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1066 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1067 struct btrfs_extent_item_v0);
1068 refs = btrfs_extent_refs_v0(leaf, ei0);
1070 if (owner == (u64)-1) {
1072 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1073 ret = btrfs_next_leaf(root, path);
1076 BUG_ON(ret > 0); /* Corruption */
1077 leaf = path->nodes[0];
1079 btrfs_item_key_to_cpu(leaf, &found_key,
1081 BUG_ON(key.objectid != found_key.objectid);
1082 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1086 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1087 struct btrfs_extent_ref_v0);
1088 owner = btrfs_ref_objectid_v0(leaf, ref0);
1092 btrfs_release_path(path);
1094 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1095 new_size += sizeof(*bi);
1097 new_size -= sizeof(*ei0);
1098 ret = btrfs_search_slot(trans, root, &key, path,
1099 new_size + extra_size, 1);
1102 BUG_ON(ret); /* Corruption */
1104 btrfs_extend_item(fs_info, path, new_size);
1106 leaf = path->nodes[0];
1107 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1108 btrfs_set_extent_refs(leaf, item, refs);
1109 /* FIXME: get real generation */
1110 btrfs_set_extent_generation(leaf, item, 0);
1111 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1112 btrfs_set_extent_flags(leaf, item,
1113 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1114 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1115 bi = (struct btrfs_tree_block_info *)(item + 1);
1116 /* FIXME: get first key of the block */
1117 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1118 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1120 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1122 btrfs_mark_buffer_dirty(leaf);
1128 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1129 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1130 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1132 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1133 struct btrfs_extent_inline_ref *iref,
1134 enum btrfs_inline_ref_type is_data)
1136 int type = btrfs_extent_inline_ref_type(eb, iref);
1137 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1139 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1140 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1141 type == BTRFS_SHARED_DATA_REF_KEY ||
1142 type == BTRFS_EXTENT_DATA_REF_KEY) {
1143 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1144 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1146 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1147 ASSERT(eb->fs_info);
1149 * Every shared one has parent tree
1150 * block, which must be aligned to
1154 IS_ALIGNED(offset, eb->fs_info->nodesize))
1157 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1158 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1160 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1161 ASSERT(eb->fs_info);
1163 * Every shared one has parent tree
1164 * block, which must be aligned to
1168 IS_ALIGNED(offset, eb->fs_info->nodesize))
1172 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1177 btrfs_print_leaf((struct extent_buffer *)eb);
1178 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1182 return BTRFS_REF_TYPE_INVALID;
1185 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1187 u32 high_crc = ~(u32)0;
1188 u32 low_crc = ~(u32)0;
1191 lenum = cpu_to_le64(root_objectid);
1192 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1193 lenum = cpu_to_le64(owner);
1194 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1195 lenum = cpu_to_le64(offset);
1196 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1198 return ((u64)high_crc << 31) ^ (u64)low_crc;
1201 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1202 struct btrfs_extent_data_ref *ref)
1204 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1205 btrfs_extent_data_ref_objectid(leaf, ref),
1206 btrfs_extent_data_ref_offset(leaf, ref));
1209 static int match_extent_data_ref(struct extent_buffer *leaf,
1210 struct btrfs_extent_data_ref *ref,
1211 u64 root_objectid, u64 owner, u64 offset)
1213 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1214 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1215 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1220 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1221 struct btrfs_fs_info *fs_info,
1222 struct btrfs_path *path,
1223 u64 bytenr, u64 parent,
1225 u64 owner, u64 offset)
1227 struct btrfs_root *root = fs_info->extent_root;
1228 struct btrfs_key key;
1229 struct btrfs_extent_data_ref *ref;
1230 struct extent_buffer *leaf;
1236 key.objectid = bytenr;
1238 key.type = BTRFS_SHARED_DATA_REF_KEY;
1239 key.offset = parent;
1241 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1242 key.offset = hash_extent_data_ref(root_objectid,
1247 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1256 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1257 key.type = BTRFS_EXTENT_REF_V0_KEY;
1258 btrfs_release_path(path);
1259 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1270 leaf = path->nodes[0];
1271 nritems = btrfs_header_nritems(leaf);
1273 if (path->slots[0] >= nritems) {
1274 ret = btrfs_next_leaf(root, path);
1280 leaf = path->nodes[0];
1281 nritems = btrfs_header_nritems(leaf);
1285 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1286 if (key.objectid != bytenr ||
1287 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1290 ref = btrfs_item_ptr(leaf, path->slots[0],
1291 struct btrfs_extent_data_ref);
1293 if (match_extent_data_ref(leaf, ref, root_objectid,
1296 btrfs_release_path(path);
1308 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1309 struct btrfs_fs_info *fs_info,
1310 struct btrfs_path *path,
1311 u64 bytenr, u64 parent,
1312 u64 root_objectid, u64 owner,
1313 u64 offset, int refs_to_add)
1315 struct btrfs_root *root = fs_info->extent_root;
1316 struct btrfs_key key;
1317 struct extent_buffer *leaf;
1322 key.objectid = bytenr;
1324 key.type = BTRFS_SHARED_DATA_REF_KEY;
1325 key.offset = parent;
1326 size = sizeof(struct btrfs_shared_data_ref);
1328 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1329 key.offset = hash_extent_data_ref(root_objectid,
1331 size = sizeof(struct btrfs_extent_data_ref);
1334 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1335 if (ret && ret != -EEXIST)
1338 leaf = path->nodes[0];
1340 struct btrfs_shared_data_ref *ref;
1341 ref = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_shared_data_ref);
1344 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1347 num_refs += refs_to_add;
1348 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1351 struct btrfs_extent_data_ref *ref;
1352 while (ret == -EEXIST) {
1353 ref = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_extent_data_ref);
1355 if (match_extent_data_ref(leaf, ref, root_objectid,
1358 btrfs_release_path(path);
1360 ret = btrfs_insert_empty_item(trans, root, path, &key,
1362 if (ret && ret != -EEXIST)
1365 leaf = path->nodes[0];
1367 ref = btrfs_item_ptr(leaf, path->slots[0],
1368 struct btrfs_extent_data_ref);
1370 btrfs_set_extent_data_ref_root(leaf, ref,
1372 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1373 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1374 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1376 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1377 num_refs += refs_to_add;
1378 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1381 btrfs_mark_buffer_dirty(leaf);
1384 btrfs_release_path(path);
1388 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1389 struct btrfs_fs_info *fs_info,
1390 struct btrfs_path *path,
1391 int refs_to_drop, int *last_ref)
1393 struct btrfs_key key;
1394 struct btrfs_extent_data_ref *ref1 = NULL;
1395 struct btrfs_shared_data_ref *ref2 = NULL;
1396 struct extent_buffer *leaf;
1400 leaf = path->nodes[0];
1401 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1403 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1404 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1405 struct btrfs_extent_data_ref);
1406 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1407 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1408 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1409 struct btrfs_shared_data_ref);
1410 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1411 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1412 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1413 struct btrfs_extent_ref_v0 *ref0;
1414 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1415 struct btrfs_extent_ref_v0);
1416 num_refs = btrfs_ref_count_v0(leaf, ref0);
1422 BUG_ON(num_refs < refs_to_drop);
1423 num_refs -= refs_to_drop;
1425 if (num_refs == 0) {
1426 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1429 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1430 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1431 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1432 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1433 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1435 struct btrfs_extent_ref_v0 *ref0;
1436 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1437 struct btrfs_extent_ref_v0);
1438 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1441 btrfs_mark_buffer_dirty(leaf);
1446 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1447 struct btrfs_extent_inline_ref *iref)
1449 struct btrfs_key key;
1450 struct extent_buffer *leaf;
1451 struct btrfs_extent_data_ref *ref1;
1452 struct btrfs_shared_data_ref *ref2;
1456 leaf = path->nodes[0];
1457 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1460 * If type is invalid, we should have bailed out earlier than
1463 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1464 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1465 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1466 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1467 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1469 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1470 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1472 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1473 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1474 struct btrfs_extent_data_ref);
1475 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1476 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1477 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1478 struct btrfs_shared_data_ref);
1479 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1480 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1481 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1482 struct btrfs_extent_ref_v0 *ref0;
1483 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1484 struct btrfs_extent_ref_v0);
1485 num_refs = btrfs_ref_count_v0(leaf, ref0);
1493 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1494 struct btrfs_fs_info *fs_info,
1495 struct btrfs_path *path,
1496 u64 bytenr, u64 parent,
1499 struct btrfs_root *root = fs_info->extent_root;
1500 struct btrfs_key key;
1503 key.objectid = bytenr;
1505 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1506 key.offset = parent;
1508 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1509 key.offset = root_objectid;
1512 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1515 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1516 if (ret == -ENOENT && parent) {
1517 btrfs_release_path(path);
1518 key.type = BTRFS_EXTENT_REF_V0_KEY;
1519 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1527 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1528 struct btrfs_fs_info *fs_info,
1529 struct btrfs_path *path,
1530 u64 bytenr, u64 parent,
1533 struct btrfs_key key;
1536 key.objectid = bytenr;
1538 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1539 key.offset = parent;
1541 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1542 key.offset = root_objectid;
1545 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1547 btrfs_release_path(path);
1551 static inline int extent_ref_type(u64 parent, u64 owner)
1554 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1556 type = BTRFS_SHARED_BLOCK_REF_KEY;
1558 type = BTRFS_TREE_BLOCK_REF_KEY;
1561 type = BTRFS_SHARED_DATA_REF_KEY;
1563 type = BTRFS_EXTENT_DATA_REF_KEY;
1568 static int find_next_key(struct btrfs_path *path, int level,
1569 struct btrfs_key *key)
1572 for (; level < BTRFS_MAX_LEVEL; level++) {
1573 if (!path->nodes[level])
1575 if (path->slots[level] + 1 >=
1576 btrfs_header_nritems(path->nodes[level]))
1579 btrfs_item_key_to_cpu(path->nodes[level], key,
1580 path->slots[level] + 1);
1582 btrfs_node_key_to_cpu(path->nodes[level], key,
1583 path->slots[level] + 1);
1590 * look for inline back ref. if back ref is found, *ref_ret is set
1591 * to the address of inline back ref, and 0 is returned.
1593 * if back ref isn't found, *ref_ret is set to the address where it
1594 * should be inserted, and -ENOENT is returned.
1596 * if insert is true and there are too many inline back refs, the path
1597 * points to the extent item, and -EAGAIN is returned.
1599 * NOTE: inline back refs are ordered in the same way that back ref
1600 * items in the tree are ordered.
1602 static noinline_for_stack
1603 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1604 struct btrfs_fs_info *fs_info,
1605 struct btrfs_path *path,
1606 struct btrfs_extent_inline_ref **ref_ret,
1607 u64 bytenr, u64 num_bytes,
1608 u64 parent, u64 root_objectid,
1609 u64 owner, u64 offset, int insert)
1611 struct btrfs_root *root = fs_info->extent_root;
1612 struct btrfs_key key;
1613 struct extent_buffer *leaf;
1614 struct btrfs_extent_item *ei;
1615 struct btrfs_extent_inline_ref *iref;
1625 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1628 key.objectid = bytenr;
1629 key.type = BTRFS_EXTENT_ITEM_KEY;
1630 key.offset = num_bytes;
1632 want = extent_ref_type(parent, owner);
1634 extra_size = btrfs_extent_inline_ref_size(want);
1635 path->keep_locks = 1;
1640 * Owner is our parent level, so we can just add one to get the level
1641 * for the block we are interested in.
1643 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1644 key.type = BTRFS_METADATA_ITEM_KEY;
1649 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1656 * We may be a newly converted file system which still has the old fat
1657 * extent entries for metadata, so try and see if we have one of those.
1659 if (ret > 0 && skinny_metadata) {
1660 skinny_metadata = false;
1661 if (path->slots[0]) {
1663 btrfs_item_key_to_cpu(path->nodes[0], &key,
1665 if (key.objectid == bytenr &&
1666 key.type == BTRFS_EXTENT_ITEM_KEY &&
1667 key.offset == num_bytes)
1671 key.objectid = bytenr;
1672 key.type = BTRFS_EXTENT_ITEM_KEY;
1673 key.offset = num_bytes;
1674 btrfs_release_path(path);
1679 if (ret && !insert) {
1682 } else if (WARN_ON(ret)) {
1687 leaf = path->nodes[0];
1688 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1689 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1690 if (item_size < sizeof(*ei)) {
1695 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1701 leaf = path->nodes[0];
1702 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1705 BUG_ON(item_size < sizeof(*ei));
1707 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1708 flags = btrfs_extent_flags(leaf, ei);
1710 ptr = (unsigned long)(ei + 1);
1711 end = (unsigned long)ei + item_size;
1713 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1714 ptr += sizeof(struct btrfs_tree_block_info);
1718 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1719 needed = BTRFS_REF_TYPE_DATA;
1721 needed = BTRFS_REF_TYPE_BLOCK;
1729 iref = (struct btrfs_extent_inline_ref *)ptr;
1730 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1731 if (type == BTRFS_REF_TYPE_INVALID) {
1739 ptr += btrfs_extent_inline_ref_size(type);
1743 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1744 struct btrfs_extent_data_ref *dref;
1745 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1746 if (match_extent_data_ref(leaf, dref, root_objectid,
1751 if (hash_extent_data_ref_item(leaf, dref) <
1752 hash_extent_data_ref(root_objectid, owner, offset))
1756 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1758 if (parent == ref_offset) {
1762 if (ref_offset < parent)
1765 if (root_objectid == ref_offset) {
1769 if (ref_offset < root_objectid)
1773 ptr += btrfs_extent_inline_ref_size(type);
1775 if (err == -ENOENT && insert) {
1776 if (item_size + extra_size >=
1777 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1782 * To add new inline back ref, we have to make sure
1783 * there is no corresponding back ref item.
1784 * For simplicity, we just do not add new inline back
1785 * ref if there is any kind of item for this block
1787 if (find_next_key(path, 0, &key) == 0 &&
1788 key.objectid == bytenr &&
1789 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1794 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1797 path->keep_locks = 0;
1798 btrfs_unlock_up_safe(path, 1);
1804 * helper to add new inline back ref
1806 static noinline_for_stack
1807 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1808 struct btrfs_path *path,
1809 struct btrfs_extent_inline_ref *iref,
1810 u64 parent, u64 root_objectid,
1811 u64 owner, u64 offset, int refs_to_add,
1812 struct btrfs_delayed_extent_op *extent_op)
1814 struct extent_buffer *leaf;
1815 struct btrfs_extent_item *ei;
1818 unsigned long item_offset;
1823 leaf = path->nodes[0];
1824 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1825 item_offset = (unsigned long)iref - (unsigned long)ei;
1827 type = extent_ref_type(parent, owner);
1828 size = btrfs_extent_inline_ref_size(type);
1830 btrfs_extend_item(fs_info, path, size);
1832 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1833 refs = btrfs_extent_refs(leaf, ei);
1834 refs += refs_to_add;
1835 btrfs_set_extent_refs(leaf, ei, refs);
1837 __run_delayed_extent_op(extent_op, leaf, ei);
1839 ptr = (unsigned long)ei + item_offset;
1840 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1841 if (ptr < end - size)
1842 memmove_extent_buffer(leaf, ptr + size, ptr,
1845 iref = (struct btrfs_extent_inline_ref *)ptr;
1846 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1847 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1848 struct btrfs_extent_data_ref *dref;
1849 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1850 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1851 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1852 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1853 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1854 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1855 struct btrfs_shared_data_ref *sref;
1856 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1857 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1858 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1859 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1860 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1862 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1864 btrfs_mark_buffer_dirty(leaf);
1867 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1868 struct btrfs_fs_info *fs_info,
1869 struct btrfs_path *path,
1870 struct btrfs_extent_inline_ref **ref_ret,
1871 u64 bytenr, u64 num_bytes, u64 parent,
1872 u64 root_objectid, u64 owner, u64 offset)
1876 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1877 bytenr, num_bytes, parent,
1878 root_objectid, owner, offset, 0);
1882 btrfs_release_path(path);
1885 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1886 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1887 parent, root_objectid);
1889 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1890 parent, root_objectid, owner,
1897 * helper to update/remove inline back ref
1899 static noinline_for_stack
1900 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1901 struct btrfs_path *path,
1902 struct btrfs_extent_inline_ref *iref,
1904 struct btrfs_delayed_extent_op *extent_op,
1907 struct extent_buffer *leaf;
1908 struct btrfs_extent_item *ei;
1909 struct btrfs_extent_data_ref *dref = NULL;
1910 struct btrfs_shared_data_ref *sref = NULL;
1918 leaf = path->nodes[0];
1919 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1920 refs = btrfs_extent_refs(leaf, ei);
1921 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1922 refs += refs_to_mod;
1923 btrfs_set_extent_refs(leaf, ei, refs);
1925 __run_delayed_extent_op(extent_op, leaf, ei);
1928 * If type is invalid, we should have bailed out after
1929 * lookup_inline_extent_backref().
1931 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1932 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1934 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1935 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1936 refs = btrfs_extent_data_ref_count(leaf, dref);
1937 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1938 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1939 refs = btrfs_shared_data_ref_count(leaf, sref);
1942 BUG_ON(refs_to_mod != -1);
1945 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1946 refs += refs_to_mod;
1949 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1950 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1952 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1955 size = btrfs_extent_inline_ref_size(type);
1956 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1957 ptr = (unsigned long)iref;
1958 end = (unsigned long)ei + item_size;
1959 if (ptr + size < end)
1960 memmove_extent_buffer(leaf, ptr, ptr + size,
1963 btrfs_truncate_item(fs_info, path, item_size, 1);
1965 btrfs_mark_buffer_dirty(leaf);
1968 static noinline_for_stack
1969 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1970 struct btrfs_fs_info *fs_info,
1971 struct btrfs_path *path,
1972 u64 bytenr, u64 num_bytes, u64 parent,
1973 u64 root_objectid, u64 owner,
1974 u64 offset, int refs_to_add,
1975 struct btrfs_delayed_extent_op *extent_op)
1977 struct btrfs_extent_inline_ref *iref;
1980 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1981 bytenr, num_bytes, parent,
1982 root_objectid, owner, offset, 1);
1984 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1985 update_inline_extent_backref(fs_info, path, iref,
1986 refs_to_add, extent_op, NULL);
1987 } else if (ret == -ENOENT) {
1988 setup_inline_extent_backref(fs_info, path, iref, parent,
1989 root_objectid, owner, offset,
1990 refs_to_add, extent_op);
1996 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1997 struct btrfs_fs_info *fs_info,
1998 struct btrfs_path *path,
1999 u64 bytenr, u64 parent, u64 root_objectid,
2000 u64 owner, u64 offset, int refs_to_add)
2003 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2004 BUG_ON(refs_to_add != 1);
2005 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
2006 parent, root_objectid);
2008 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
2009 parent, root_objectid,
2010 owner, offset, refs_to_add);
2015 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2016 struct btrfs_fs_info *fs_info,
2017 struct btrfs_path *path,
2018 struct btrfs_extent_inline_ref *iref,
2019 int refs_to_drop, int is_data, int *last_ref)
2023 BUG_ON(!is_data && refs_to_drop != 1);
2025 update_inline_extent_backref(fs_info, path, iref,
2026 -refs_to_drop, NULL, last_ref);
2027 } else if (is_data) {
2028 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
2032 ret = btrfs_del_item(trans, fs_info->extent_root, path);
2037 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2038 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2039 u64 *discarded_bytes)
2042 u64 bytes_left, end;
2043 u64 aligned_start = ALIGN(start, 1 << 9);
2045 if (WARN_ON(start != aligned_start)) {
2046 len -= aligned_start - start;
2047 len = round_down(len, 1 << 9);
2048 start = aligned_start;
2051 *discarded_bytes = 0;
2059 /* Skip any superblocks on this device. */
2060 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2061 u64 sb_start = btrfs_sb_offset(j);
2062 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2063 u64 size = sb_start - start;
2065 if (!in_range(sb_start, start, bytes_left) &&
2066 !in_range(sb_end, start, bytes_left) &&
2067 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2071 * Superblock spans beginning of range. Adjust start and
2074 if (sb_start <= start) {
2075 start += sb_end - start;
2080 bytes_left = end - start;
2085 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2088 *discarded_bytes += size;
2089 else if (ret != -EOPNOTSUPP)
2098 bytes_left = end - start;
2102 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2105 *discarded_bytes += bytes_left;
2110 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2111 u64 num_bytes, u64 *actual_bytes)
2114 u64 discarded_bytes = 0;
2115 struct btrfs_bio *bbio = NULL;
2119 * Avoid races with device replace and make sure our bbio has devices
2120 * associated to its stripes that don't go away while we are discarding.
2122 btrfs_bio_counter_inc_blocked(fs_info);
2123 /* Tell the block device(s) that the sectors can be discarded */
2124 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2126 /* Error condition is -ENOMEM */
2128 struct btrfs_bio_stripe *stripe = bbio->stripes;
2132 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2134 struct request_queue *req_q;
2136 if (!stripe->dev->bdev) {
2137 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2140 req_q = bdev_get_queue(stripe->dev->bdev);
2141 if (!blk_queue_discard(req_q))
2144 ret = btrfs_issue_discard(stripe->dev->bdev,
2149 discarded_bytes += bytes;
2150 else if (ret != -EOPNOTSUPP)
2151 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2154 * Just in case we get back EOPNOTSUPP for some reason,
2155 * just ignore the return value so we don't screw up
2156 * people calling discard_extent.
2160 btrfs_put_bbio(bbio);
2162 btrfs_bio_counter_dec(fs_info);
2165 *actual_bytes = discarded_bytes;
2168 if (ret == -EOPNOTSUPP)
2173 /* Can return -ENOMEM */
2174 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2175 struct btrfs_root *root,
2176 u64 bytenr, u64 num_bytes, u64 parent,
2177 u64 root_objectid, u64 owner, u64 offset)
2179 struct btrfs_fs_info *fs_info = root->fs_info;
2180 int old_ref_mod, new_ref_mod;
2183 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2184 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2186 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2187 owner, offset, BTRFS_ADD_DELAYED_REF);
2189 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2190 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2192 root_objectid, (int)owner,
2193 BTRFS_ADD_DELAYED_REF, NULL,
2194 &old_ref_mod, &new_ref_mod);
2196 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2198 root_objectid, owner, offset,
2199 0, BTRFS_ADD_DELAYED_REF,
2200 &old_ref_mod, &new_ref_mod);
2203 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2204 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2209 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2210 struct btrfs_fs_info *fs_info,
2211 struct btrfs_delayed_ref_node *node,
2212 u64 parent, u64 root_objectid,
2213 u64 owner, u64 offset, int refs_to_add,
2214 struct btrfs_delayed_extent_op *extent_op)
2216 struct btrfs_path *path;
2217 struct extent_buffer *leaf;
2218 struct btrfs_extent_item *item;
2219 struct btrfs_key key;
2220 u64 bytenr = node->bytenr;
2221 u64 num_bytes = node->num_bytes;
2225 path = btrfs_alloc_path();
2229 path->reada = READA_FORWARD;
2230 path->leave_spinning = 1;
2231 /* this will setup the path even if it fails to insert the back ref */
2232 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2233 num_bytes, parent, root_objectid,
2235 refs_to_add, extent_op);
2236 if ((ret < 0 && ret != -EAGAIN) || !ret)
2240 * Ok we had -EAGAIN which means we didn't have space to insert and
2241 * inline extent ref, so just update the reference count and add a
2244 leaf = path->nodes[0];
2245 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2246 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2247 refs = btrfs_extent_refs(leaf, item);
2248 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2250 __run_delayed_extent_op(extent_op, leaf, item);
2252 btrfs_mark_buffer_dirty(leaf);
2253 btrfs_release_path(path);
2255 path->reada = READA_FORWARD;
2256 path->leave_spinning = 1;
2257 /* now insert the actual backref */
2258 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2259 root_objectid, owner, offset, refs_to_add);
2261 btrfs_abort_transaction(trans, ret);
2263 btrfs_free_path(path);
2267 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2268 struct btrfs_fs_info *fs_info,
2269 struct btrfs_delayed_ref_node *node,
2270 struct btrfs_delayed_extent_op *extent_op,
2271 int insert_reserved)
2274 struct btrfs_delayed_data_ref *ref;
2275 struct btrfs_key ins;
2280 ins.objectid = node->bytenr;
2281 ins.offset = node->num_bytes;
2282 ins.type = BTRFS_EXTENT_ITEM_KEY;
2284 ref = btrfs_delayed_node_to_data_ref(node);
2285 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2287 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2288 parent = ref->parent;
2289 ref_root = ref->root;
2291 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2293 flags |= extent_op->flags_to_set;
2294 ret = alloc_reserved_file_extent(trans, fs_info,
2295 parent, ref_root, flags,
2296 ref->objectid, ref->offset,
2297 &ins, node->ref_mod);
2298 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2299 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2300 ref_root, ref->objectid,
2301 ref->offset, node->ref_mod,
2303 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2304 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2305 ref_root, ref->objectid,
2306 ref->offset, node->ref_mod,
2314 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2315 struct extent_buffer *leaf,
2316 struct btrfs_extent_item *ei)
2318 u64 flags = btrfs_extent_flags(leaf, ei);
2319 if (extent_op->update_flags) {
2320 flags |= extent_op->flags_to_set;
2321 btrfs_set_extent_flags(leaf, ei, flags);
2324 if (extent_op->update_key) {
2325 struct btrfs_tree_block_info *bi;
2326 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2327 bi = (struct btrfs_tree_block_info *)(ei + 1);
2328 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2332 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2333 struct btrfs_fs_info *fs_info,
2334 struct btrfs_delayed_ref_head *head,
2335 struct btrfs_delayed_extent_op *extent_op)
2337 struct btrfs_key key;
2338 struct btrfs_path *path;
2339 struct btrfs_extent_item *ei;
2340 struct extent_buffer *leaf;
2344 int metadata = !extent_op->is_data;
2349 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2352 path = btrfs_alloc_path();
2356 key.objectid = head->bytenr;
2359 key.type = BTRFS_METADATA_ITEM_KEY;
2360 key.offset = extent_op->level;
2362 key.type = BTRFS_EXTENT_ITEM_KEY;
2363 key.offset = head->num_bytes;
2367 path->reada = READA_FORWARD;
2368 path->leave_spinning = 1;
2369 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2376 if (path->slots[0] > 0) {
2378 btrfs_item_key_to_cpu(path->nodes[0], &key,
2380 if (key.objectid == head->bytenr &&
2381 key.type == BTRFS_EXTENT_ITEM_KEY &&
2382 key.offset == head->num_bytes)
2386 btrfs_release_path(path);
2389 key.objectid = head->bytenr;
2390 key.offset = head->num_bytes;
2391 key.type = BTRFS_EXTENT_ITEM_KEY;
2400 leaf = path->nodes[0];
2401 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2402 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2403 if (item_size < sizeof(*ei)) {
2404 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2409 leaf = path->nodes[0];
2410 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2413 BUG_ON(item_size < sizeof(*ei));
2414 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2415 __run_delayed_extent_op(extent_op, leaf, ei);
2417 btrfs_mark_buffer_dirty(leaf);
2419 btrfs_free_path(path);
2423 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2424 struct btrfs_fs_info *fs_info,
2425 struct btrfs_delayed_ref_node *node,
2426 struct btrfs_delayed_extent_op *extent_op,
2427 int insert_reserved)
2430 struct btrfs_delayed_tree_ref *ref;
2431 struct btrfs_key ins;
2434 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2436 ref = btrfs_delayed_node_to_tree_ref(node);
2437 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2439 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2440 parent = ref->parent;
2441 ref_root = ref->root;
2443 ins.objectid = node->bytenr;
2444 if (skinny_metadata) {
2445 ins.offset = ref->level;
2446 ins.type = BTRFS_METADATA_ITEM_KEY;
2448 ins.offset = node->num_bytes;
2449 ins.type = BTRFS_EXTENT_ITEM_KEY;
2452 if (node->ref_mod != 1) {
2454 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2455 node->bytenr, node->ref_mod, node->action, ref_root,
2459 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2460 BUG_ON(!extent_op || !extent_op->update_flags);
2461 ret = alloc_reserved_tree_block(trans, fs_info,
2463 extent_op->flags_to_set,
2466 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2467 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2471 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2472 ret = __btrfs_free_extent(trans, fs_info, node,
2474 ref->level, 0, 1, extent_op);
2481 /* helper function to actually process a single delayed ref entry */
2482 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2483 struct btrfs_fs_info *fs_info,
2484 struct btrfs_delayed_ref_node *node,
2485 struct btrfs_delayed_extent_op *extent_op,
2486 int insert_reserved)
2490 if (trans->aborted) {
2491 if (insert_reserved)
2492 btrfs_pin_extent(fs_info, node->bytenr,
2493 node->num_bytes, 1);
2497 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2498 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2499 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2501 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2502 node->type == BTRFS_SHARED_DATA_REF_KEY)
2503 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2510 static inline struct btrfs_delayed_ref_node *
2511 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2513 struct btrfs_delayed_ref_node *ref;
2515 if (RB_EMPTY_ROOT(&head->ref_tree))
2519 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2520 * This is to prevent a ref count from going down to zero, which deletes
2521 * the extent item from the extent tree, when there still are references
2522 * to add, which would fail because they would not find the extent item.
2524 if (!list_empty(&head->ref_add_list))
2525 return list_first_entry(&head->ref_add_list,
2526 struct btrfs_delayed_ref_node, add_list);
2528 ref = rb_entry(rb_first(&head->ref_tree),
2529 struct btrfs_delayed_ref_node, ref_node);
2530 ASSERT(list_empty(&ref->add_list));
2534 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2535 struct btrfs_delayed_ref_head *head)
2537 spin_lock(&delayed_refs->lock);
2538 head->processing = 0;
2539 delayed_refs->num_heads_ready++;
2540 spin_unlock(&delayed_refs->lock);
2541 btrfs_delayed_ref_unlock(head);
2544 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2545 struct btrfs_fs_info *fs_info,
2546 struct btrfs_delayed_ref_head *head)
2548 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2553 head->extent_op = NULL;
2554 if (head->must_insert_reserved) {
2555 btrfs_free_delayed_extent_op(extent_op);
2558 spin_unlock(&head->lock);
2559 ret = run_delayed_extent_op(trans, fs_info, head, extent_op);
2560 btrfs_free_delayed_extent_op(extent_op);
2561 return ret ? ret : 1;
2564 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2565 struct btrfs_fs_info *fs_info,
2566 struct btrfs_delayed_ref_head *head)
2568 struct btrfs_delayed_ref_root *delayed_refs;
2571 delayed_refs = &trans->transaction->delayed_refs;
2573 ret = cleanup_extent_op(trans, fs_info, head);
2575 unselect_delayed_ref_head(delayed_refs, head);
2576 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2583 * Need to drop our head ref lock and re-acquire the delayed ref lock
2584 * and then re-check to make sure nobody got added.
2586 spin_unlock(&head->lock);
2587 spin_lock(&delayed_refs->lock);
2588 spin_lock(&head->lock);
2589 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2590 spin_unlock(&head->lock);
2591 spin_unlock(&delayed_refs->lock);
2594 delayed_refs->num_heads--;
2595 rb_erase(&head->href_node, &delayed_refs->href_root);
2596 RB_CLEAR_NODE(&head->href_node);
2597 spin_unlock(&delayed_refs->lock);
2598 spin_unlock(&head->lock);
2599 atomic_dec(&delayed_refs->num_entries);
2601 trace_run_delayed_ref_head(fs_info, head, 0);
2603 if (head->total_ref_mod < 0) {
2604 struct btrfs_space_info *space_info;
2608 flags = BTRFS_BLOCK_GROUP_DATA;
2609 else if (head->is_system)
2610 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2612 flags = BTRFS_BLOCK_GROUP_METADATA;
2613 space_info = __find_space_info(fs_info, flags);
2615 percpu_counter_add(&space_info->total_bytes_pinned,
2618 if (head->is_data) {
2619 spin_lock(&delayed_refs->lock);
2620 delayed_refs->pending_csums -= head->num_bytes;
2621 spin_unlock(&delayed_refs->lock);
2625 if (head->must_insert_reserved) {
2626 btrfs_pin_extent(fs_info, head->bytenr,
2627 head->num_bytes, 1);
2628 if (head->is_data) {
2629 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2634 /* Also free its reserved qgroup space */
2635 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2636 head->qgroup_reserved);
2637 btrfs_delayed_ref_unlock(head);
2638 btrfs_put_delayed_ref_head(head);
2643 * Returns 0 on success or if called with an already aborted transaction.
2644 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2646 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2649 struct btrfs_fs_info *fs_info = trans->fs_info;
2650 struct btrfs_delayed_ref_root *delayed_refs;
2651 struct btrfs_delayed_ref_node *ref;
2652 struct btrfs_delayed_ref_head *locked_ref = NULL;
2653 struct btrfs_delayed_extent_op *extent_op;
2654 ktime_t start = ktime_get();
2656 unsigned long count = 0;
2657 unsigned long actual_count = 0;
2658 int must_insert_reserved = 0;
2660 delayed_refs = &trans->transaction->delayed_refs;
2666 spin_lock(&delayed_refs->lock);
2667 locked_ref = btrfs_select_ref_head(trans);
2669 spin_unlock(&delayed_refs->lock);
2673 /* grab the lock that says we are going to process
2674 * all the refs for this head */
2675 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2676 spin_unlock(&delayed_refs->lock);
2678 * we may have dropped the spin lock to get the head
2679 * mutex lock, and that might have given someone else
2680 * time to free the head. If that's true, it has been
2681 * removed from our list and we can move on.
2683 if (ret == -EAGAIN) {
2691 * We need to try and merge add/drops of the same ref since we
2692 * can run into issues with relocate dropping the implicit ref
2693 * and then it being added back again before the drop can
2694 * finish. If we merged anything we need to re-loop so we can
2696 * Or we can get node references of the same type that weren't
2697 * merged when created due to bumps in the tree mod seq, and
2698 * we need to merge them to prevent adding an inline extent
2699 * backref before dropping it (triggering a BUG_ON at
2700 * insert_inline_extent_backref()).
2702 spin_lock(&locked_ref->lock);
2703 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2707 * locked_ref is the head node, so we have to go one
2708 * node back for any delayed ref updates
2710 ref = select_delayed_ref(locked_ref);
2712 if (ref && ref->seq &&
2713 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2714 spin_unlock(&locked_ref->lock);
2715 unselect_delayed_ref_head(delayed_refs, locked_ref);
2723 * We're done processing refs in this ref_head, clean everything
2724 * up and move on to the next ref_head.
2727 ret = cleanup_ref_head(trans, fs_info, locked_ref);
2729 /* We dropped our lock, we need to loop. */
2742 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2743 RB_CLEAR_NODE(&ref->ref_node);
2744 if (!list_empty(&ref->add_list))
2745 list_del(&ref->add_list);
2747 * When we play the delayed ref, also correct the ref_mod on
2750 switch (ref->action) {
2751 case BTRFS_ADD_DELAYED_REF:
2752 case BTRFS_ADD_DELAYED_EXTENT:
2753 locked_ref->ref_mod -= ref->ref_mod;
2755 case BTRFS_DROP_DELAYED_REF:
2756 locked_ref->ref_mod += ref->ref_mod;
2761 atomic_dec(&delayed_refs->num_entries);
2764 * Record the must-insert_reserved flag before we drop the spin
2767 must_insert_reserved = locked_ref->must_insert_reserved;
2768 locked_ref->must_insert_reserved = 0;
2770 extent_op = locked_ref->extent_op;
2771 locked_ref->extent_op = NULL;
2772 spin_unlock(&locked_ref->lock);
2774 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2775 must_insert_reserved);
2777 btrfs_free_delayed_extent_op(extent_op);
2779 unselect_delayed_ref_head(delayed_refs, locked_ref);
2780 btrfs_put_delayed_ref(ref);
2781 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2786 btrfs_put_delayed_ref(ref);
2792 * We don't want to include ref heads since we can have empty ref heads
2793 * and those will drastically skew our runtime down since we just do
2794 * accounting, no actual extent tree updates.
2796 if (actual_count > 0) {
2797 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2801 * We weigh the current average higher than our current runtime
2802 * to avoid large swings in the average.
2804 spin_lock(&delayed_refs->lock);
2805 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2806 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2807 spin_unlock(&delayed_refs->lock);
2812 #ifdef SCRAMBLE_DELAYED_REFS
2814 * Normally delayed refs get processed in ascending bytenr order. This
2815 * correlates in most cases to the order added. To expose dependencies on this
2816 * order, we start to process the tree in the middle instead of the beginning
2818 static u64 find_middle(struct rb_root *root)
2820 struct rb_node *n = root->rb_node;
2821 struct btrfs_delayed_ref_node *entry;
2824 u64 first = 0, last = 0;
2828 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2829 first = entry->bytenr;
2833 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2834 last = entry->bytenr;
2839 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2840 WARN_ON(!entry->in_tree);
2842 middle = entry->bytenr;
2855 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2859 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2860 sizeof(struct btrfs_extent_inline_ref));
2861 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2862 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2865 * We don't ever fill up leaves all the way so multiply by 2 just to be
2866 * closer to what we're really going to want to use.
2868 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2872 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2873 * would require to store the csums for that many bytes.
2875 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2878 u64 num_csums_per_leaf;
2881 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2882 num_csums_per_leaf = div64_u64(csum_size,
2883 (u64)btrfs_super_csum_size(fs_info->super_copy));
2884 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2885 num_csums += num_csums_per_leaf - 1;
2886 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2890 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2891 struct btrfs_fs_info *fs_info)
2893 struct btrfs_block_rsv *global_rsv;
2894 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2895 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2896 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2897 u64 num_bytes, num_dirty_bgs_bytes;
2900 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2901 num_heads = heads_to_leaves(fs_info, num_heads);
2903 num_bytes += (num_heads - 1) * fs_info->nodesize;
2905 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2907 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2909 global_rsv = &fs_info->global_block_rsv;
2912 * If we can't allocate any more chunks lets make sure we have _lots_ of
2913 * wiggle room since running delayed refs can create more delayed refs.
2915 if (global_rsv->space_info->full) {
2916 num_dirty_bgs_bytes <<= 1;
2920 spin_lock(&global_rsv->lock);
2921 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2923 spin_unlock(&global_rsv->lock);
2927 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2928 struct btrfs_fs_info *fs_info)
2931 atomic_read(&trans->transaction->delayed_refs.num_entries);
2936 avg_runtime = fs_info->avg_delayed_ref_runtime;
2937 val = num_entries * avg_runtime;
2938 if (val >= NSEC_PER_SEC)
2940 if (val >= NSEC_PER_SEC / 2)
2943 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2946 struct async_delayed_refs {
2947 struct btrfs_root *root;
2952 struct completion wait;
2953 struct btrfs_work work;
2956 static inline struct async_delayed_refs *
2957 to_async_delayed_refs(struct btrfs_work *work)
2959 return container_of(work, struct async_delayed_refs, work);
2962 static void delayed_ref_async_start(struct btrfs_work *work)
2964 struct async_delayed_refs *async = to_async_delayed_refs(work);
2965 struct btrfs_trans_handle *trans;
2966 struct btrfs_fs_info *fs_info = async->root->fs_info;
2969 /* if the commit is already started, we don't need to wait here */
2970 if (btrfs_transaction_blocked(fs_info))
2973 trans = btrfs_join_transaction(async->root);
2974 if (IS_ERR(trans)) {
2975 async->error = PTR_ERR(trans);
2980 * trans->sync means that when we call end_transaction, we won't
2981 * wait on delayed refs
2985 /* Don't bother flushing if we got into a different transaction */
2986 if (trans->transid > async->transid)
2989 ret = btrfs_run_delayed_refs(trans, async->count);
2993 ret = btrfs_end_transaction(trans);
2994 if (ret && !async->error)
2998 complete(&async->wait);
3003 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
3004 unsigned long count, u64 transid, int wait)
3006 struct async_delayed_refs *async;
3009 async = kmalloc(sizeof(*async), GFP_NOFS);
3013 async->root = fs_info->tree_root;
3014 async->count = count;
3016 async->transid = transid;
3021 init_completion(&async->wait);
3023 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3024 delayed_ref_async_start, NULL, NULL);
3026 btrfs_queue_work(fs_info->extent_workers, &async->work);
3029 wait_for_completion(&async->wait);
3038 * this starts processing the delayed reference count updates and
3039 * extent insertions we have queued up so far. count can be
3040 * 0, which means to process everything in the tree at the start
3041 * of the run (but not newly added entries), or it can be some target
3042 * number you'd like to process.
3044 * Returns 0 on success or if called with an aborted transaction
3045 * Returns <0 on error and aborts the transaction
3047 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3048 unsigned long count)
3050 struct btrfs_fs_info *fs_info = trans->fs_info;
3051 struct rb_node *node;
3052 struct btrfs_delayed_ref_root *delayed_refs;
3053 struct btrfs_delayed_ref_head *head;
3055 int run_all = count == (unsigned long)-1;
3056 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3058 /* We'll clean this up in btrfs_cleanup_transaction */
3062 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3065 delayed_refs = &trans->transaction->delayed_refs;
3067 count = atomic_read(&delayed_refs->num_entries) * 2;
3070 #ifdef SCRAMBLE_DELAYED_REFS
3071 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3073 trans->can_flush_pending_bgs = false;
3074 ret = __btrfs_run_delayed_refs(trans, count);
3076 btrfs_abort_transaction(trans, ret);
3081 if (!list_empty(&trans->new_bgs))
3082 btrfs_create_pending_block_groups(trans);
3084 spin_lock(&delayed_refs->lock);
3085 node = rb_first(&delayed_refs->href_root);
3087 spin_unlock(&delayed_refs->lock);
3090 head = rb_entry(node, struct btrfs_delayed_ref_head,
3092 refcount_inc(&head->refs);
3093 spin_unlock(&delayed_refs->lock);
3095 /* Mutex was contended, block until it's released and retry. */
3096 mutex_lock(&head->mutex);
3097 mutex_unlock(&head->mutex);
3099 btrfs_put_delayed_ref_head(head);
3104 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3108 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3109 struct btrfs_fs_info *fs_info,
3110 u64 bytenr, u64 num_bytes, u64 flags,
3111 int level, int is_data)
3113 struct btrfs_delayed_extent_op *extent_op;
3116 extent_op = btrfs_alloc_delayed_extent_op();
3120 extent_op->flags_to_set = flags;
3121 extent_op->update_flags = true;
3122 extent_op->update_key = false;
3123 extent_op->is_data = is_data ? true : false;
3124 extent_op->level = level;
3126 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3127 num_bytes, extent_op);
3129 btrfs_free_delayed_extent_op(extent_op);
3133 static noinline int check_delayed_ref(struct btrfs_root *root,
3134 struct btrfs_path *path,
3135 u64 objectid, u64 offset, u64 bytenr)
3137 struct btrfs_delayed_ref_head *head;
3138 struct btrfs_delayed_ref_node *ref;
3139 struct btrfs_delayed_data_ref *data_ref;
3140 struct btrfs_delayed_ref_root *delayed_refs;
3141 struct btrfs_transaction *cur_trans;
3142 struct rb_node *node;
3145 cur_trans = root->fs_info->running_transaction;
3149 delayed_refs = &cur_trans->delayed_refs;
3150 spin_lock(&delayed_refs->lock);
3151 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3153 spin_unlock(&delayed_refs->lock);
3157 if (!mutex_trylock(&head->mutex)) {
3158 refcount_inc(&head->refs);
3159 spin_unlock(&delayed_refs->lock);
3161 btrfs_release_path(path);
3164 * Mutex was contended, block until it's released and let
3167 mutex_lock(&head->mutex);
3168 mutex_unlock(&head->mutex);
3169 btrfs_put_delayed_ref_head(head);
3172 spin_unlock(&delayed_refs->lock);
3174 spin_lock(&head->lock);
3176 * XXX: We should replace this with a proper search function in the
3179 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3180 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3181 /* If it's a shared ref we know a cross reference exists */
3182 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3187 data_ref = btrfs_delayed_node_to_data_ref(ref);
3190 * If our ref doesn't match the one we're currently looking at
3191 * then we have a cross reference.
3193 if (data_ref->root != root->root_key.objectid ||
3194 data_ref->objectid != objectid ||
3195 data_ref->offset != offset) {
3200 spin_unlock(&head->lock);
3201 mutex_unlock(&head->mutex);
3205 static noinline int check_committed_ref(struct btrfs_root *root,
3206 struct btrfs_path *path,
3207 u64 objectid, u64 offset, u64 bytenr)
3209 struct btrfs_fs_info *fs_info = root->fs_info;
3210 struct btrfs_root *extent_root = fs_info->extent_root;
3211 struct extent_buffer *leaf;
3212 struct btrfs_extent_data_ref *ref;
3213 struct btrfs_extent_inline_ref *iref;
3214 struct btrfs_extent_item *ei;
3215 struct btrfs_key key;
3220 key.objectid = bytenr;
3221 key.offset = (u64)-1;
3222 key.type = BTRFS_EXTENT_ITEM_KEY;
3224 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3227 BUG_ON(ret == 0); /* Corruption */
3230 if (path->slots[0] == 0)
3234 leaf = path->nodes[0];
3235 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3237 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3241 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3242 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3243 if (item_size < sizeof(*ei)) {
3244 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3248 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3250 if (item_size != sizeof(*ei) +
3251 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3254 if (btrfs_extent_generation(leaf, ei) <=
3255 btrfs_root_last_snapshot(&root->root_item))
3258 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3260 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3261 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3264 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3265 if (btrfs_extent_refs(leaf, ei) !=
3266 btrfs_extent_data_ref_count(leaf, ref) ||
3267 btrfs_extent_data_ref_root(leaf, ref) !=
3268 root->root_key.objectid ||
3269 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3270 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3278 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3281 struct btrfs_path *path;
3285 path = btrfs_alloc_path();
3290 ret = check_committed_ref(root, path, objectid,
3292 if (ret && ret != -ENOENT)
3295 ret2 = check_delayed_ref(root, path, objectid,
3297 } while (ret2 == -EAGAIN);
3299 if (ret2 && ret2 != -ENOENT) {
3304 if (ret != -ENOENT || ret2 != -ENOENT)
3307 btrfs_free_path(path);
3308 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3313 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3314 struct btrfs_root *root,
3315 struct extent_buffer *buf,
3316 int full_backref, int inc)
3318 struct btrfs_fs_info *fs_info = root->fs_info;
3324 struct btrfs_key key;
3325 struct btrfs_file_extent_item *fi;
3329 int (*process_func)(struct btrfs_trans_handle *,
3330 struct btrfs_root *,
3331 u64, u64, u64, u64, u64, u64);
3334 if (btrfs_is_testing(fs_info))
3337 ref_root = btrfs_header_owner(buf);
3338 nritems = btrfs_header_nritems(buf);
3339 level = btrfs_header_level(buf);
3341 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3345 process_func = btrfs_inc_extent_ref;
3347 process_func = btrfs_free_extent;
3350 parent = buf->start;
3354 for (i = 0; i < nritems; i++) {
3356 btrfs_item_key_to_cpu(buf, &key, i);
3357 if (key.type != BTRFS_EXTENT_DATA_KEY)
3359 fi = btrfs_item_ptr(buf, i,
3360 struct btrfs_file_extent_item);
3361 if (btrfs_file_extent_type(buf, fi) ==
3362 BTRFS_FILE_EXTENT_INLINE)
3364 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3368 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3369 key.offset -= btrfs_file_extent_offset(buf, fi);
3370 ret = process_func(trans, root, bytenr, num_bytes,
3371 parent, ref_root, key.objectid,
3376 bytenr = btrfs_node_blockptr(buf, i);
3377 num_bytes = fs_info->nodesize;
3378 ret = process_func(trans, root, bytenr, num_bytes,
3379 parent, ref_root, level - 1, 0);
3389 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3390 struct extent_buffer *buf, int full_backref)
3392 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3395 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3396 struct extent_buffer *buf, int full_backref)
3398 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3401 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3402 struct btrfs_fs_info *fs_info,
3403 struct btrfs_path *path,
3404 struct btrfs_block_group_cache *cache)
3407 struct btrfs_root *extent_root = fs_info->extent_root;
3409 struct extent_buffer *leaf;
3411 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3418 leaf = path->nodes[0];
3419 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3420 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3421 btrfs_mark_buffer_dirty(leaf);
3423 btrfs_release_path(path);
3428 static struct btrfs_block_group_cache *
3429 next_block_group(struct btrfs_fs_info *fs_info,
3430 struct btrfs_block_group_cache *cache)
3432 struct rb_node *node;
3434 spin_lock(&fs_info->block_group_cache_lock);
3436 /* If our block group was removed, we need a full search. */
3437 if (RB_EMPTY_NODE(&cache->cache_node)) {
3438 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3440 spin_unlock(&fs_info->block_group_cache_lock);
3441 btrfs_put_block_group(cache);
3442 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3444 node = rb_next(&cache->cache_node);
3445 btrfs_put_block_group(cache);
3447 cache = rb_entry(node, struct btrfs_block_group_cache,
3449 btrfs_get_block_group(cache);
3452 spin_unlock(&fs_info->block_group_cache_lock);
3456 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3457 struct btrfs_trans_handle *trans,
3458 struct btrfs_path *path)
3460 struct btrfs_fs_info *fs_info = block_group->fs_info;
3461 struct btrfs_root *root = fs_info->tree_root;
3462 struct inode *inode = NULL;
3463 struct extent_changeset *data_reserved = NULL;
3465 int dcs = BTRFS_DC_ERROR;
3471 * If this block group is smaller than 100 megs don't bother caching the
3474 if (block_group->key.offset < (100 * SZ_1M)) {
3475 spin_lock(&block_group->lock);
3476 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3477 spin_unlock(&block_group->lock);
3484 inode = lookup_free_space_inode(fs_info, block_group, path);
3485 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3486 ret = PTR_ERR(inode);
3487 btrfs_release_path(path);
3491 if (IS_ERR(inode)) {
3495 if (block_group->ro)
3498 ret = create_free_space_inode(fs_info, trans, block_group,
3506 * We want to set the generation to 0, that way if anything goes wrong
3507 * from here on out we know not to trust this cache when we load up next
3510 BTRFS_I(inode)->generation = 0;
3511 ret = btrfs_update_inode(trans, root, inode);
3514 * So theoretically we could recover from this, simply set the
3515 * super cache generation to 0 so we know to invalidate the
3516 * cache, but then we'd have to keep track of the block groups
3517 * that fail this way so we know we _have_ to reset this cache
3518 * before the next commit or risk reading stale cache. So to
3519 * limit our exposure to horrible edge cases lets just abort the
3520 * transaction, this only happens in really bad situations
3523 btrfs_abort_transaction(trans, ret);
3528 /* We've already setup this transaction, go ahead and exit */
3529 if (block_group->cache_generation == trans->transid &&
3530 i_size_read(inode)) {
3531 dcs = BTRFS_DC_SETUP;
3535 if (i_size_read(inode) > 0) {
3536 ret = btrfs_check_trunc_cache_free_space(fs_info,
3537 &fs_info->global_block_rsv);
3541 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3546 spin_lock(&block_group->lock);
3547 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3548 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3550 * don't bother trying to write stuff out _if_
3551 * a) we're not cached,
3552 * b) we're with nospace_cache mount option,
3553 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3555 dcs = BTRFS_DC_WRITTEN;
3556 spin_unlock(&block_group->lock);
3559 spin_unlock(&block_group->lock);
3562 * We hit an ENOSPC when setting up the cache in this transaction, just
3563 * skip doing the setup, we've already cleared the cache so we're safe.
3565 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3571 * Try to preallocate enough space based on how big the block group is.
3572 * Keep in mind this has to include any pinned space which could end up
3573 * taking up quite a bit since it's not folded into the other space
3576 num_pages = div_u64(block_group->key.offset, SZ_256M);
3581 num_pages *= PAGE_SIZE;
3583 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3587 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3588 num_pages, num_pages,
3591 * Our cache requires contiguous chunks so that we don't modify a bunch
3592 * of metadata or split extents when writing the cache out, which means
3593 * we can enospc if we are heavily fragmented in addition to just normal
3594 * out of space conditions. So if we hit this just skip setting up any
3595 * other block groups for this transaction, maybe we'll unpin enough
3596 * space the next time around.
3599 dcs = BTRFS_DC_SETUP;
3600 else if (ret == -ENOSPC)
3601 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3606 btrfs_release_path(path);
3608 spin_lock(&block_group->lock);
3609 if (!ret && dcs == BTRFS_DC_SETUP)
3610 block_group->cache_generation = trans->transid;
3611 block_group->disk_cache_state = dcs;
3612 spin_unlock(&block_group->lock);
3614 extent_changeset_free(data_reserved);
3618 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3619 struct btrfs_fs_info *fs_info)
3621 struct btrfs_block_group_cache *cache, *tmp;
3622 struct btrfs_transaction *cur_trans = trans->transaction;
3623 struct btrfs_path *path;
3625 if (list_empty(&cur_trans->dirty_bgs) ||
3626 !btrfs_test_opt(fs_info, SPACE_CACHE))
3629 path = btrfs_alloc_path();
3633 /* Could add new block groups, use _safe just in case */
3634 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3636 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3637 cache_save_setup(cache, trans, path);
3640 btrfs_free_path(path);
3645 * transaction commit does final block group cache writeback during a
3646 * critical section where nothing is allowed to change the FS. This is
3647 * required in order for the cache to actually match the block group,
3648 * but can introduce a lot of latency into the commit.
3650 * So, btrfs_start_dirty_block_groups is here to kick off block group
3651 * cache IO. There's a chance we'll have to redo some of it if the
3652 * block group changes again during the commit, but it greatly reduces
3653 * the commit latency by getting rid of the easy block groups while
3654 * we're still allowing others to join the commit.
3656 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3658 struct btrfs_fs_info *fs_info = trans->fs_info;
3659 struct btrfs_block_group_cache *cache;
3660 struct btrfs_transaction *cur_trans = trans->transaction;
3663 struct btrfs_path *path = NULL;
3665 struct list_head *io = &cur_trans->io_bgs;
3666 int num_started = 0;
3669 spin_lock(&cur_trans->dirty_bgs_lock);
3670 if (list_empty(&cur_trans->dirty_bgs)) {
3671 spin_unlock(&cur_trans->dirty_bgs_lock);
3674 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3675 spin_unlock(&cur_trans->dirty_bgs_lock);
3679 * make sure all the block groups on our dirty list actually
3682 btrfs_create_pending_block_groups(trans);
3685 path = btrfs_alloc_path();
3691 * cache_write_mutex is here only to save us from balance or automatic
3692 * removal of empty block groups deleting this block group while we are
3693 * writing out the cache
3695 mutex_lock(&trans->transaction->cache_write_mutex);
3696 while (!list_empty(&dirty)) {
3697 cache = list_first_entry(&dirty,
3698 struct btrfs_block_group_cache,
3701 * this can happen if something re-dirties a block
3702 * group that is already under IO. Just wait for it to
3703 * finish and then do it all again
3705 if (!list_empty(&cache->io_list)) {
3706 list_del_init(&cache->io_list);
3707 btrfs_wait_cache_io(trans, cache, path);
3708 btrfs_put_block_group(cache);
3713 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3714 * if it should update the cache_state. Don't delete
3715 * until after we wait.
3717 * Since we're not running in the commit critical section
3718 * we need the dirty_bgs_lock to protect from update_block_group
3720 spin_lock(&cur_trans->dirty_bgs_lock);
3721 list_del_init(&cache->dirty_list);
3722 spin_unlock(&cur_trans->dirty_bgs_lock);
3726 cache_save_setup(cache, trans, path);
3728 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3729 cache->io_ctl.inode = NULL;
3730 ret = btrfs_write_out_cache(fs_info, trans,
3732 if (ret == 0 && cache->io_ctl.inode) {
3737 * The cache_write_mutex is protecting the
3738 * io_list, also refer to the definition of
3739 * btrfs_transaction::io_bgs for more details
3741 list_add_tail(&cache->io_list, io);
3744 * if we failed to write the cache, the
3745 * generation will be bad and life goes on
3751 ret = write_one_cache_group(trans, fs_info,
3754 * Our block group might still be attached to the list
3755 * of new block groups in the transaction handle of some
3756 * other task (struct btrfs_trans_handle->new_bgs). This
3757 * means its block group item isn't yet in the extent
3758 * tree. If this happens ignore the error, as we will
3759 * try again later in the critical section of the
3760 * transaction commit.
3762 if (ret == -ENOENT) {
3764 spin_lock(&cur_trans->dirty_bgs_lock);
3765 if (list_empty(&cache->dirty_list)) {
3766 list_add_tail(&cache->dirty_list,
3767 &cur_trans->dirty_bgs);
3768 btrfs_get_block_group(cache);
3770 spin_unlock(&cur_trans->dirty_bgs_lock);
3772 btrfs_abort_transaction(trans, ret);
3776 /* if its not on the io list, we need to put the block group */
3778 btrfs_put_block_group(cache);
3784 * Avoid blocking other tasks for too long. It might even save
3785 * us from writing caches for block groups that are going to be
3788 mutex_unlock(&trans->transaction->cache_write_mutex);
3789 mutex_lock(&trans->transaction->cache_write_mutex);
3791 mutex_unlock(&trans->transaction->cache_write_mutex);
3794 * go through delayed refs for all the stuff we've just kicked off
3795 * and then loop back (just once)
3797 ret = btrfs_run_delayed_refs(trans, 0);
3798 if (!ret && loops == 0) {
3800 spin_lock(&cur_trans->dirty_bgs_lock);
3801 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3803 * dirty_bgs_lock protects us from concurrent block group
3804 * deletes too (not just cache_write_mutex).
3806 if (!list_empty(&dirty)) {
3807 spin_unlock(&cur_trans->dirty_bgs_lock);
3810 spin_unlock(&cur_trans->dirty_bgs_lock);
3811 } else if (ret < 0) {
3812 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3815 btrfs_free_path(path);
3819 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3820 struct btrfs_fs_info *fs_info)
3822 struct btrfs_block_group_cache *cache;
3823 struct btrfs_transaction *cur_trans = trans->transaction;
3826 struct btrfs_path *path;
3827 struct list_head *io = &cur_trans->io_bgs;
3828 int num_started = 0;
3830 path = btrfs_alloc_path();
3835 * Even though we are in the critical section of the transaction commit,
3836 * we can still have concurrent tasks adding elements to this
3837 * transaction's list of dirty block groups. These tasks correspond to
3838 * endio free space workers started when writeback finishes for a
3839 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3840 * allocate new block groups as a result of COWing nodes of the root
3841 * tree when updating the free space inode. The writeback for the space
3842 * caches is triggered by an earlier call to
3843 * btrfs_start_dirty_block_groups() and iterations of the following
3845 * Also we want to do the cache_save_setup first and then run the
3846 * delayed refs to make sure we have the best chance at doing this all
3849 spin_lock(&cur_trans->dirty_bgs_lock);
3850 while (!list_empty(&cur_trans->dirty_bgs)) {
3851 cache = list_first_entry(&cur_trans->dirty_bgs,
3852 struct btrfs_block_group_cache,
3856 * this can happen if cache_save_setup re-dirties a block
3857 * group that is already under IO. Just wait for it to
3858 * finish and then do it all again
3860 if (!list_empty(&cache->io_list)) {
3861 spin_unlock(&cur_trans->dirty_bgs_lock);
3862 list_del_init(&cache->io_list);
3863 btrfs_wait_cache_io(trans, cache, path);
3864 btrfs_put_block_group(cache);
3865 spin_lock(&cur_trans->dirty_bgs_lock);
3869 * don't remove from the dirty list until after we've waited
3872 list_del_init(&cache->dirty_list);
3873 spin_unlock(&cur_trans->dirty_bgs_lock);
3876 cache_save_setup(cache, trans, path);
3879 ret = btrfs_run_delayed_refs(trans,
3880 (unsigned long) -1);
3882 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3883 cache->io_ctl.inode = NULL;
3884 ret = btrfs_write_out_cache(fs_info, trans,
3886 if (ret == 0 && cache->io_ctl.inode) {
3889 list_add_tail(&cache->io_list, io);
3892 * if we failed to write the cache, the
3893 * generation will be bad and life goes on
3899 ret = write_one_cache_group(trans, fs_info,
3902 * One of the free space endio workers might have
3903 * created a new block group while updating a free space
3904 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3905 * and hasn't released its transaction handle yet, in
3906 * which case the new block group is still attached to
3907 * its transaction handle and its creation has not
3908 * finished yet (no block group item in the extent tree
3909 * yet, etc). If this is the case, wait for all free
3910 * space endio workers to finish and retry. This is a
3911 * a very rare case so no need for a more efficient and
3914 if (ret == -ENOENT) {
3915 wait_event(cur_trans->writer_wait,
3916 atomic_read(&cur_trans->num_writers) == 1);
3917 ret = write_one_cache_group(trans, fs_info,
3921 btrfs_abort_transaction(trans, ret);
3924 /* if its not on the io list, we need to put the block group */
3926 btrfs_put_block_group(cache);
3927 spin_lock(&cur_trans->dirty_bgs_lock);
3929 spin_unlock(&cur_trans->dirty_bgs_lock);
3932 * Refer to the definition of io_bgs member for details why it's safe
3933 * to use it without any locking
3935 while (!list_empty(io)) {
3936 cache = list_first_entry(io, struct btrfs_block_group_cache,
3938 list_del_init(&cache->io_list);
3939 btrfs_wait_cache_io(trans, cache, path);
3940 btrfs_put_block_group(cache);
3943 btrfs_free_path(path);
3947 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3949 struct btrfs_block_group_cache *block_group;
3952 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3953 if (!block_group || block_group->ro)
3956 btrfs_put_block_group(block_group);
3960 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3962 struct btrfs_block_group_cache *bg;
3965 bg = btrfs_lookup_block_group(fs_info, bytenr);
3969 spin_lock(&bg->lock);
3973 atomic_inc(&bg->nocow_writers);
3974 spin_unlock(&bg->lock);
3976 /* no put on block group, done by btrfs_dec_nocow_writers */
3978 btrfs_put_block_group(bg);
3984 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3986 struct btrfs_block_group_cache *bg;
3988 bg = btrfs_lookup_block_group(fs_info, bytenr);
3990 if (atomic_dec_and_test(&bg->nocow_writers))
3991 wake_up_var(&bg->nocow_writers);
3993 * Once for our lookup and once for the lookup done by a previous call
3994 * to btrfs_inc_nocow_writers()
3996 btrfs_put_block_group(bg);
3997 btrfs_put_block_group(bg);
4000 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4002 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
4005 static const char *alloc_name(u64 flags)
4008 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4010 case BTRFS_BLOCK_GROUP_METADATA:
4012 case BTRFS_BLOCK_GROUP_DATA:
4014 case BTRFS_BLOCK_GROUP_SYSTEM:
4018 return "invalid-combination";
4022 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4023 struct btrfs_space_info **new)
4026 struct btrfs_space_info *space_info;
4030 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4034 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4041 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4042 INIT_LIST_HEAD(&space_info->block_groups[i]);
4043 init_rwsem(&space_info->groups_sem);
4044 spin_lock_init(&space_info->lock);
4045 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4046 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4047 init_waitqueue_head(&space_info->wait);
4048 INIT_LIST_HEAD(&space_info->ro_bgs);
4049 INIT_LIST_HEAD(&space_info->tickets);
4050 INIT_LIST_HEAD(&space_info->priority_tickets);
4052 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4053 info->space_info_kobj, "%s",
4054 alloc_name(space_info->flags));
4056 percpu_counter_destroy(&space_info->total_bytes_pinned);
4062 list_add_rcu(&space_info->list, &info->space_info);
4063 if (flags & BTRFS_BLOCK_GROUP_DATA)
4064 info->data_sinfo = space_info;
4069 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4070 u64 total_bytes, u64 bytes_used,
4072 struct btrfs_space_info **space_info)
4074 struct btrfs_space_info *found;
4077 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4078 BTRFS_BLOCK_GROUP_RAID10))
4083 found = __find_space_info(info, flags);
4085 spin_lock(&found->lock);
4086 found->total_bytes += total_bytes;
4087 found->disk_total += total_bytes * factor;
4088 found->bytes_used += bytes_used;
4089 found->disk_used += bytes_used * factor;
4090 found->bytes_readonly += bytes_readonly;
4091 if (total_bytes > 0)
4093 space_info_add_new_bytes(info, found, total_bytes -
4094 bytes_used - bytes_readonly);
4095 spin_unlock(&found->lock);
4096 *space_info = found;
4099 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4101 u64 extra_flags = chunk_to_extended(flags) &
4102 BTRFS_EXTENDED_PROFILE_MASK;
4104 write_seqlock(&fs_info->profiles_lock);
4105 if (flags & BTRFS_BLOCK_GROUP_DATA)
4106 fs_info->avail_data_alloc_bits |= extra_flags;
4107 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4108 fs_info->avail_metadata_alloc_bits |= extra_flags;
4109 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4110 fs_info->avail_system_alloc_bits |= extra_flags;
4111 write_sequnlock(&fs_info->profiles_lock);
4115 * returns target flags in extended format or 0 if restripe for this
4116 * chunk_type is not in progress
4118 * should be called with either volume_mutex or balance_lock held
4120 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4122 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4128 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4129 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4130 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4131 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4132 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4133 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4134 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4135 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4136 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4143 * @flags: available profiles in extended format (see ctree.h)
4145 * Returns reduced profile in chunk format. If profile changing is in
4146 * progress (either running or paused) picks the target profile (if it's
4147 * already available), otherwise falls back to plain reducing.
4149 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4151 u64 num_devices = fs_info->fs_devices->rw_devices;
4157 * see if restripe for this chunk_type is in progress, if so
4158 * try to reduce to the target profile
4160 spin_lock(&fs_info->balance_lock);
4161 target = get_restripe_target(fs_info, flags);
4163 /* pick target profile only if it's already available */
4164 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4165 spin_unlock(&fs_info->balance_lock);
4166 return extended_to_chunk(target);
4169 spin_unlock(&fs_info->balance_lock);
4171 /* First, mask out the RAID levels which aren't possible */
4172 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4173 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4174 allowed |= btrfs_raid_group[raid_type];
4178 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4179 allowed = BTRFS_BLOCK_GROUP_RAID6;
4180 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4181 allowed = BTRFS_BLOCK_GROUP_RAID5;
4182 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4183 allowed = BTRFS_BLOCK_GROUP_RAID10;
4184 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4185 allowed = BTRFS_BLOCK_GROUP_RAID1;
4186 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4187 allowed = BTRFS_BLOCK_GROUP_RAID0;
4189 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4191 return extended_to_chunk(flags | allowed);
4194 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4201 seq = read_seqbegin(&fs_info->profiles_lock);
4203 if (flags & BTRFS_BLOCK_GROUP_DATA)
4204 flags |= fs_info->avail_data_alloc_bits;
4205 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4206 flags |= fs_info->avail_system_alloc_bits;
4207 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4208 flags |= fs_info->avail_metadata_alloc_bits;
4209 } while (read_seqretry(&fs_info->profiles_lock, seq));
4211 return btrfs_reduce_alloc_profile(fs_info, flags);
4214 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4216 struct btrfs_fs_info *fs_info = root->fs_info;
4221 flags = BTRFS_BLOCK_GROUP_DATA;
4222 else if (root == fs_info->chunk_root)
4223 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4225 flags = BTRFS_BLOCK_GROUP_METADATA;
4227 ret = get_alloc_profile(fs_info, flags);
4231 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4233 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4236 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4238 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4241 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4243 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4246 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4247 bool may_use_included)
4250 return s_info->bytes_used + s_info->bytes_reserved +
4251 s_info->bytes_pinned + s_info->bytes_readonly +
4252 (may_use_included ? s_info->bytes_may_use : 0);
4255 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4257 struct btrfs_root *root = inode->root;
4258 struct btrfs_fs_info *fs_info = root->fs_info;
4259 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4262 int need_commit = 2;
4263 int have_pinned_space;
4265 /* make sure bytes are sectorsize aligned */
4266 bytes = ALIGN(bytes, fs_info->sectorsize);
4268 if (btrfs_is_free_space_inode(inode)) {
4270 ASSERT(current->journal_info);
4274 /* make sure we have enough space to handle the data first */
4275 spin_lock(&data_sinfo->lock);
4276 used = btrfs_space_info_used(data_sinfo, true);
4278 if (used + bytes > data_sinfo->total_bytes) {
4279 struct btrfs_trans_handle *trans;
4282 * if we don't have enough free bytes in this space then we need
4283 * to alloc a new chunk.
4285 if (!data_sinfo->full) {
4288 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4289 spin_unlock(&data_sinfo->lock);
4291 alloc_target = btrfs_data_alloc_profile(fs_info);
4293 * It is ugly that we don't call nolock join
4294 * transaction for the free space inode case here.
4295 * But it is safe because we only do the data space
4296 * reservation for the free space cache in the
4297 * transaction context, the common join transaction
4298 * just increase the counter of the current transaction
4299 * handler, doesn't try to acquire the trans_lock of
4302 trans = btrfs_join_transaction(root);
4304 return PTR_ERR(trans);
4306 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4307 CHUNK_ALLOC_NO_FORCE);
4308 btrfs_end_transaction(trans);
4313 have_pinned_space = 1;
4322 * If we don't have enough pinned space to deal with this
4323 * allocation, and no removed chunk in current transaction,
4324 * don't bother committing the transaction.
4326 have_pinned_space = percpu_counter_compare(
4327 &data_sinfo->total_bytes_pinned,
4328 used + bytes - data_sinfo->total_bytes);
4329 spin_unlock(&data_sinfo->lock);
4331 /* commit the current transaction and try again */
4336 if (need_commit > 0) {
4337 btrfs_start_delalloc_roots(fs_info, 0, -1);
4338 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4342 trans = btrfs_join_transaction(root);
4344 return PTR_ERR(trans);
4345 if (have_pinned_space >= 0 ||
4346 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4347 &trans->transaction->flags) ||
4349 ret = btrfs_commit_transaction(trans);
4353 * The cleaner kthread might still be doing iput
4354 * operations. Wait for it to finish so that
4355 * more space is released.
4357 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4358 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4361 btrfs_end_transaction(trans);
4365 trace_btrfs_space_reservation(fs_info,
4366 "space_info:enospc",
4367 data_sinfo->flags, bytes, 1);
4370 data_sinfo->bytes_may_use += bytes;
4371 trace_btrfs_space_reservation(fs_info, "space_info",
4372 data_sinfo->flags, bytes, 1);
4373 spin_unlock(&data_sinfo->lock);
4378 int btrfs_check_data_free_space(struct inode *inode,
4379 struct extent_changeset **reserved, u64 start, u64 len)
4381 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4384 /* align the range */
4385 len = round_up(start + len, fs_info->sectorsize) -
4386 round_down(start, fs_info->sectorsize);
4387 start = round_down(start, fs_info->sectorsize);
4389 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4393 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4394 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4396 btrfs_free_reserved_data_space_noquota(inode, start, len);
4403 * Called if we need to clear a data reservation for this inode
4404 * Normally in a error case.
4406 * This one will *NOT* use accurate qgroup reserved space API, just for case
4407 * which we can't sleep and is sure it won't affect qgroup reserved space.
4408 * Like clear_bit_hook().
4410 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4413 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4414 struct btrfs_space_info *data_sinfo;
4416 /* Make sure the range is aligned to sectorsize */
4417 len = round_up(start + len, fs_info->sectorsize) -
4418 round_down(start, fs_info->sectorsize);
4419 start = round_down(start, fs_info->sectorsize);
4421 data_sinfo = fs_info->data_sinfo;
4422 spin_lock(&data_sinfo->lock);
4423 if (WARN_ON(data_sinfo->bytes_may_use < len))
4424 data_sinfo->bytes_may_use = 0;
4426 data_sinfo->bytes_may_use -= len;
4427 trace_btrfs_space_reservation(fs_info, "space_info",
4428 data_sinfo->flags, len, 0);
4429 spin_unlock(&data_sinfo->lock);
4433 * Called if we need to clear a data reservation for this inode
4434 * Normally in a error case.
4436 * This one will handle the per-inode data rsv map for accurate reserved
4439 void btrfs_free_reserved_data_space(struct inode *inode,
4440 struct extent_changeset *reserved, u64 start, u64 len)
4442 struct btrfs_root *root = BTRFS_I(inode)->root;
4444 /* Make sure the range is aligned to sectorsize */
4445 len = round_up(start + len, root->fs_info->sectorsize) -
4446 round_down(start, root->fs_info->sectorsize);
4447 start = round_down(start, root->fs_info->sectorsize);
4449 btrfs_free_reserved_data_space_noquota(inode, start, len);
4450 btrfs_qgroup_free_data(inode, reserved, start, len);
4453 static void force_metadata_allocation(struct btrfs_fs_info *info)
4455 struct list_head *head = &info->space_info;
4456 struct btrfs_space_info *found;
4459 list_for_each_entry_rcu(found, head, list) {
4460 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4461 found->force_alloc = CHUNK_ALLOC_FORCE;
4466 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4468 return (global->size << 1);
4471 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4472 struct btrfs_space_info *sinfo, int force)
4474 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4475 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4478 if (force == CHUNK_ALLOC_FORCE)
4482 * We need to take into account the global rsv because for all intents
4483 * and purposes it's used space. Don't worry about locking the
4484 * global_rsv, it doesn't change except when the transaction commits.
4486 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4487 bytes_used += calc_global_rsv_need_space(global_rsv);
4490 * in limited mode, we want to have some free space up to
4491 * about 1% of the FS size.
4493 if (force == CHUNK_ALLOC_LIMITED) {
4494 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4495 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4497 if (sinfo->total_bytes - bytes_used < thresh)
4501 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4506 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4510 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4511 BTRFS_BLOCK_GROUP_RAID0 |
4512 BTRFS_BLOCK_GROUP_RAID5 |
4513 BTRFS_BLOCK_GROUP_RAID6))
4514 num_dev = fs_info->fs_devices->rw_devices;
4515 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4518 num_dev = 1; /* DUP or single */
4524 * If @is_allocation is true, reserve space in the system space info necessary
4525 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4528 void check_system_chunk(struct btrfs_trans_handle *trans,
4529 struct btrfs_fs_info *fs_info, u64 type)
4531 struct btrfs_space_info *info;
4538 * Needed because we can end up allocating a system chunk and for an
4539 * atomic and race free space reservation in the chunk block reserve.
4541 lockdep_assert_held(&fs_info->chunk_mutex);
4543 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4544 spin_lock(&info->lock);
4545 left = info->total_bytes - btrfs_space_info_used(info, true);
4546 spin_unlock(&info->lock);
4548 num_devs = get_profile_num_devs(fs_info, type);
4550 /* num_devs device items to update and 1 chunk item to add or remove */
4551 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4552 btrfs_calc_trans_metadata_size(fs_info, 1);
4554 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4555 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4556 left, thresh, type);
4557 dump_space_info(fs_info, info, 0, 0);
4560 if (left < thresh) {
4561 u64 flags = btrfs_system_alloc_profile(fs_info);
4564 * Ignore failure to create system chunk. We might end up not
4565 * needing it, as we might not need to COW all nodes/leafs from
4566 * the paths we visit in the chunk tree (they were already COWed
4567 * or created in the current transaction for example).
4569 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4573 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4574 &fs_info->chunk_block_rsv,
4575 thresh, BTRFS_RESERVE_NO_FLUSH);
4577 trans->chunk_bytes_reserved += thresh;
4582 * If force is CHUNK_ALLOC_FORCE:
4583 * - return 1 if it successfully allocates a chunk,
4584 * - return errors including -ENOSPC otherwise.
4585 * If force is NOT CHUNK_ALLOC_FORCE:
4586 * - return 0 if it doesn't need to allocate a new chunk,
4587 * - return 1 if it successfully allocates a chunk,
4588 * - return errors including -ENOSPC otherwise.
4590 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4591 struct btrfs_fs_info *fs_info, u64 flags, int force)
4593 struct btrfs_space_info *space_info;
4594 int wait_for_alloc = 0;
4597 /* Don't re-enter if we're already allocating a chunk */
4598 if (trans->allocating_chunk)
4601 space_info = __find_space_info(fs_info, flags);
4605 spin_lock(&space_info->lock);
4606 if (force < space_info->force_alloc)
4607 force = space_info->force_alloc;
4608 if (space_info->full) {
4609 if (should_alloc_chunk(fs_info, space_info, force))
4613 spin_unlock(&space_info->lock);
4617 if (!should_alloc_chunk(fs_info, space_info, force)) {
4618 spin_unlock(&space_info->lock);
4620 } else if (space_info->chunk_alloc) {
4623 space_info->chunk_alloc = 1;
4626 spin_unlock(&space_info->lock);
4628 mutex_lock(&fs_info->chunk_mutex);
4631 * The chunk_mutex is held throughout the entirety of a chunk
4632 * allocation, so once we've acquired the chunk_mutex we know that the
4633 * other guy is done and we need to recheck and see if we should
4636 if (wait_for_alloc) {
4637 mutex_unlock(&fs_info->chunk_mutex);
4643 trans->allocating_chunk = true;
4646 * If we have mixed data/metadata chunks we want to make sure we keep
4647 * allocating mixed chunks instead of individual chunks.
4649 if (btrfs_mixed_space_info(space_info))
4650 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4653 * if we're doing a data chunk, go ahead and make sure that
4654 * we keep a reasonable number of metadata chunks allocated in the
4657 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4658 fs_info->data_chunk_allocations++;
4659 if (!(fs_info->data_chunk_allocations %
4660 fs_info->metadata_ratio))
4661 force_metadata_allocation(fs_info);
4665 * Check if we have enough space in SYSTEM chunk because we may need
4666 * to update devices.
4668 check_system_chunk(trans, fs_info, flags);
4670 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4671 trans->allocating_chunk = false;
4673 spin_lock(&space_info->lock);
4674 if (ret < 0 && ret != -ENOSPC)
4677 space_info->full = 1;
4681 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4683 space_info->chunk_alloc = 0;
4684 spin_unlock(&space_info->lock);
4685 mutex_unlock(&fs_info->chunk_mutex);
4687 * When we allocate a new chunk we reserve space in the chunk block
4688 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4689 * add new nodes/leafs to it if we end up needing to do it when
4690 * inserting the chunk item and updating device items as part of the
4691 * second phase of chunk allocation, performed by
4692 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4693 * large number of new block groups to create in our transaction
4694 * handle's new_bgs list to avoid exhausting the chunk block reserve
4695 * in extreme cases - like having a single transaction create many new
4696 * block groups when starting to write out the free space caches of all
4697 * the block groups that were made dirty during the lifetime of the
4700 if (trans->can_flush_pending_bgs &&
4701 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4702 btrfs_create_pending_block_groups(trans);
4703 btrfs_trans_release_chunk_metadata(trans);
4708 static int can_overcommit(struct btrfs_fs_info *fs_info,
4709 struct btrfs_space_info *space_info, u64 bytes,
4710 enum btrfs_reserve_flush_enum flush,
4713 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4719 /* Don't overcommit when in mixed mode. */
4720 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4724 profile = btrfs_system_alloc_profile(fs_info);
4726 profile = btrfs_metadata_alloc_profile(fs_info);
4728 used = btrfs_space_info_used(space_info, false);
4731 * We only want to allow over committing if we have lots of actual space
4732 * free, but if we don't have enough space to handle the global reserve
4733 * space then we could end up having a real enospc problem when trying
4734 * to allocate a chunk or some other such important allocation.
4736 spin_lock(&global_rsv->lock);
4737 space_size = calc_global_rsv_need_space(global_rsv);
4738 spin_unlock(&global_rsv->lock);
4739 if (used + space_size >= space_info->total_bytes)
4742 used += space_info->bytes_may_use;
4744 avail = atomic64_read(&fs_info->free_chunk_space);
4747 * If we have dup, raid1 or raid10 then only half of the free
4748 * space is actually useable. For raid56, the space info used
4749 * doesn't include the parity drive, so we don't have to
4752 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4753 BTRFS_BLOCK_GROUP_RAID1 |
4754 BTRFS_BLOCK_GROUP_RAID10))
4758 * If we aren't flushing all things, let us overcommit up to
4759 * 1/2th of the space. If we can flush, don't let us overcommit
4760 * too much, let it overcommit up to 1/8 of the space.
4762 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4767 if (used + bytes < space_info->total_bytes + avail)
4772 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4773 unsigned long nr_pages, int nr_items)
4775 struct super_block *sb = fs_info->sb;
4777 if (down_read_trylock(&sb->s_umount)) {
4778 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4779 up_read(&sb->s_umount);
4782 * We needn't worry the filesystem going from r/w to r/o though
4783 * we don't acquire ->s_umount mutex, because the filesystem
4784 * should guarantee the delalloc inodes list be empty after
4785 * the filesystem is readonly(all dirty pages are written to
4788 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4789 if (!current->journal_info)
4790 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4794 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4800 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4801 nr = div64_u64(to_reclaim, bytes);
4807 #define EXTENT_SIZE_PER_ITEM SZ_256K
4810 * shrink metadata reservation for delalloc
4812 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4813 u64 orig, bool wait_ordered)
4815 struct btrfs_space_info *space_info;
4816 struct btrfs_trans_handle *trans;
4821 unsigned long nr_pages;
4824 /* Calc the number of the pages we need flush for space reservation */
4825 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4826 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4828 trans = (struct btrfs_trans_handle *)current->journal_info;
4829 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4831 delalloc_bytes = percpu_counter_sum_positive(
4832 &fs_info->delalloc_bytes);
4833 if (delalloc_bytes == 0) {
4837 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4842 while (delalloc_bytes && loops < 3) {
4843 max_reclaim = min(delalloc_bytes, to_reclaim);
4844 nr_pages = max_reclaim >> PAGE_SHIFT;
4845 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4847 * We need to wait for the async pages to actually start before
4850 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4854 if (max_reclaim <= nr_pages)
4857 max_reclaim -= nr_pages;
4859 wait_event(fs_info->async_submit_wait,
4860 atomic_read(&fs_info->async_delalloc_pages) <=
4863 spin_lock(&space_info->lock);
4864 if (list_empty(&space_info->tickets) &&
4865 list_empty(&space_info->priority_tickets)) {
4866 spin_unlock(&space_info->lock);
4869 spin_unlock(&space_info->lock);
4872 if (wait_ordered && !trans) {
4873 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4875 time_left = schedule_timeout_killable(1);
4879 delalloc_bytes = percpu_counter_sum_positive(
4880 &fs_info->delalloc_bytes);
4884 struct reserve_ticket {
4887 struct list_head list;
4888 wait_queue_head_t wait;
4892 * maybe_commit_transaction - possibly commit the transaction if its ok to
4893 * @root - the root we're allocating for
4894 * @bytes - the number of bytes we want to reserve
4895 * @force - force the commit
4897 * This will check to make sure that committing the transaction will actually
4898 * get us somewhere and then commit the transaction if it does. Otherwise it
4899 * will return -ENOSPC.
4901 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4902 struct btrfs_space_info *space_info)
4904 struct reserve_ticket *ticket = NULL;
4905 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4906 struct btrfs_trans_handle *trans;
4909 trans = (struct btrfs_trans_handle *)current->journal_info;
4913 spin_lock(&space_info->lock);
4914 if (!list_empty(&space_info->priority_tickets))
4915 ticket = list_first_entry(&space_info->priority_tickets,
4916 struct reserve_ticket, list);
4917 else if (!list_empty(&space_info->tickets))
4918 ticket = list_first_entry(&space_info->tickets,
4919 struct reserve_ticket, list);
4920 bytes = (ticket) ? ticket->bytes : 0;
4921 spin_unlock(&space_info->lock);
4926 /* See if there is enough pinned space to make this reservation */
4927 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4932 * See if there is some space in the delayed insertion reservation for
4935 if (space_info != delayed_rsv->space_info)
4938 spin_lock(&delayed_rsv->lock);
4939 if (delayed_rsv->size > bytes)
4942 bytes -= delayed_rsv->size;
4943 spin_unlock(&delayed_rsv->lock);
4945 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4951 trans = btrfs_join_transaction(fs_info->extent_root);
4955 return btrfs_commit_transaction(trans);
4959 * Try to flush some data based on policy set by @state. This is only advisory
4960 * and may fail for various reasons. The caller is supposed to examine the
4961 * state of @space_info to detect the outcome.
4963 static void flush_space(struct btrfs_fs_info *fs_info,
4964 struct btrfs_space_info *space_info, u64 num_bytes,
4967 struct btrfs_root *root = fs_info->extent_root;
4968 struct btrfs_trans_handle *trans;
4973 case FLUSH_DELAYED_ITEMS_NR:
4974 case FLUSH_DELAYED_ITEMS:
4975 if (state == FLUSH_DELAYED_ITEMS_NR)
4976 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4980 trans = btrfs_join_transaction(root);
4981 if (IS_ERR(trans)) {
4982 ret = PTR_ERR(trans);
4985 ret = btrfs_run_delayed_items_nr(trans, nr);
4986 btrfs_end_transaction(trans);
4988 case FLUSH_DELALLOC:
4989 case FLUSH_DELALLOC_WAIT:
4990 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4991 state == FLUSH_DELALLOC_WAIT);
4994 trans = btrfs_join_transaction(root);
4995 if (IS_ERR(trans)) {
4996 ret = PTR_ERR(trans);
4999 ret = do_chunk_alloc(trans, fs_info,
5000 btrfs_metadata_alloc_profile(fs_info),
5001 CHUNK_ALLOC_NO_FORCE);
5002 btrfs_end_transaction(trans);
5003 if (ret > 0 || ret == -ENOSPC)
5007 ret = may_commit_transaction(fs_info, space_info);
5014 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5020 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5021 struct btrfs_space_info *space_info,
5024 struct reserve_ticket *ticket;
5029 list_for_each_entry(ticket, &space_info->tickets, list)
5030 to_reclaim += ticket->bytes;
5031 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5032 to_reclaim += ticket->bytes;
5036 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5037 if (can_overcommit(fs_info, space_info, to_reclaim,
5038 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5041 used = btrfs_space_info_used(space_info, true);
5043 if (can_overcommit(fs_info, space_info, SZ_1M,
5044 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5045 expected = div_factor_fine(space_info->total_bytes, 95);
5047 expected = div_factor_fine(space_info->total_bytes, 90);
5049 if (used > expected)
5050 to_reclaim = used - expected;
5053 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5054 space_info->bytes_reserved);
5058 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5059 struct btrfs_space_info *space_info,
5060 u64 used, bool system_chunk)
5062 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5064 /* If we're just plain full then async reclaim just slows us down. */
5065 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5068 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5072 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5073 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5076 static void wake_all_tickets(struct list_head *head)
5078 struct reserve_ticket *ticket;
5080 while (!list_empty(head)) {
5081 ticket = list_first_entry(head, struct reserve_ticket, list);
5082 list_del_init(&ticket->list);
5083 ticket->error = -ENOSPC;
5084 wake_up(&ticket->wait);
5089 * This is for normal flushers, we can wait all goddamned day if we want to. We
5090 * will loop and continuously try to flush as long as we are making progress.
5091 * We count progress as clearing off tickets each time we have to loop.
5093 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5095 struct btrfs_fs_info *fs_info;
5096 struct btrfs_space_info *space_info;
5099 int commit_cycles = 0;
5100 u64 last_tickets_id;
5102 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5103 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5105 spin_lock(&space_info->lock);
5106 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5109 space_info->flush = 0;
5110 spin_unlock(&space_info->lock);
5113 last_tickets_id = space_info->tickets_id;
5114 spin_unlock(&space_info->lock);
5116 flush_state = FLUSH_DELAYED_ITEMS_NR;
5118 flush_space(fs_info, space_info, to_reclaim, flush_state);
5119 spin_lock(&space_info->lock);
5120 if (list_empty(&space_info->tickets)) {
5121 space_info->flush = 0;
5122 spin_unlock(&space_info->lock);
5125 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5128 if (last_tickets_id == space_info->tickets_id) {
5131 last_tickets_id = space_info->tickets_id;
5132 flush_state = FLUSH_DELAYED_ITEMS_NR;
5137 if (flush_state > COMMIT_TRANS) {
5139 if (commit_cycles > 2) {
5140 wake_all_tickets(&space_info->tickets);
5141 space_info->flush = 0;
5143 flush_state = FLUSH_DELAYED_ITEMS_NR;
5146 spin_unlock(&space_info->lock);
5147 } while (flush_state <= COMMIT_TRANS);
5150 void btrfs_init_async_reclaim_work(struct work_struct *work)
5152 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5155 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5156 struct btrfs_space_info *space_info,
5157 struct reserve_ticket *ticket)
5160 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5162 spin_lock(&space_info->lock);
5163 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5166 spin_unlock(&space_info->lock);
5169 spin_unlock(&space_info->lock);
5172 flush_space(fs_info, space_info, to_reclaim, flush_state);
5174 spin_lock(&space_info->lock);
5175 if (ticket->bytes == 0) {
5176 spin_unlock(&space_info->lock);
5179 spin_unlock(&space_info->lock);
5182 * Priority flushers can't wait on delalloc without
5185 if (flush_state == FLUSH_DELALLOC ||
5186 flush_state == FLUSH_DELALLOC_WAIT)
5187 flush_state = ALLOC_CHUNK;
5188 } while (flush_state < COMMIT_TRANS);
5191 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5192 struct btrfs_space_info *space_info,
5193 struct reserve_ticket *ticket, u64 orig_bytes)
5199 spin_lock(&space_info->lock);
5200 while (ticket->bytes > 0 && ticket->error == 0) {
5201 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5206 spin_unlock(&space_info->lock);
5210 finish_wait(&ticket->wait, &wait);
5211 spin_lock(&space_info->lock);
5214 ret = ticket->error;
5215 if (!list_empty(&ticket->list))
5216 list_del_init(&ticket->list);
5217 if (ticket->bytes && ticket->bytes < orig_bytes) {
5218 u64 num_bytes = orig_bytes - ticket->bytes;
5219 space_info->bytes_may_use -= num_bytes;
5220 trace_btrfs_space_reservation(fs_info, "space_info",
5221 space_info->flags, num_bytes, 0);
5223 spin_unlock(&space_info->lock);
5229 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5230 * @root - the root we're allocating for
5231 * @space_info - the space info we want to allocate from
5232 * @orig_bytes - the number of bytes we want
5233 * @flush - whether or not we can flush to make our reservation
5235 * This will reserve orig_bytes number of bytes from the space info associated
5236 * with the block_rsv. If there is not enough space it will make an attempt to
5237 * flush out space to make room. It will do this by flushing delalloc if
5238 * possible or committing the transaction. If flush is 0 then no attempts to
5239 * regain reservations will be made and this will fail if there is not enough
5242 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5243 struct btrfs_space_info *space_info,
5245 enum btrfs_reserve_flush_enum flush,
5248 struct reserve_ticket ticket;
5253 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5255 spin_lock(&space_info->lock);
5257 used = btrfs_space_info_used(space_info, true);
5260 * If we have enough space then hooray, make our reservation and carry
5261 * on. If not see if we can overcommit, and if we can, hooray carry on.
5262 * If not things get more complicated.
5264 if (used + orig_bytes <= space_info->total_bytes) {
5265 space_info->bytes_may_use += orig_bytes;
5266 trace_btrfs_space_reservation(fs_info, "space_info",
5267 space_info->flags, orig_bytes, 1);
5269 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5271 space_info->bytes_may_use += orig_bytes;
5272 trace_btrfs_space_reservation(fs_info, "space_info",
5273 space_info->flags, orig_bytes, 1);
5278 * If we couldn't make a reservation then setup our reservation ticket
5279 * and kick the async worker if it's not already running.
5281 * If we are a priority flusher then we just need to add our ticket to
5282 * the list and we will do our own flushing further down.
5284 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5285 ticket.bytes = orig_bytes;
5287 init_waitqueue_head(&ticket.wait);
5288 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5289 list_add_tail(&ticket.list, &space_info->tickets);
5290 if (!space_info->flush) {
5291 space_info->flush = 1;
5292 trace_btrfs_trigger_flush(fs_info,
5296 queue_work(system_unbound_wq,
5297 &fs_info->async_reclaim_work);
5300 list_add_tail(&ticket.list,
5301 &space_info->priority_tickets);
5303 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5306 * We will do the space reservation dance during log replay,
5307 * which means we won't have fs_info->fs_root set, so don't do
5308 * the async reclaim as we will panic.
5310 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5311 need_do_async_reclaim(fs_info, space_info,
5312 used, system_chunk) &&
5313 !work_busy(&fs_info->async_reclaim_work)) {
5314 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5315 orig_bytes, flush, "preempt");
5316 queue_work(system_unbound_wq,
5317 &fs_info->async_reclaim_work);
5320 spin_unlock(&space_info->lock);
5321 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5324 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5325 return wait_reserve_ticket(fs_info, space_info, &ticket,
5329 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5330 spin_lock(&space_info->lock);
5332 if (ticket.bytes < orig_bytes) {
5333 u64 num_bytes = orig_bytes - ticket.bytes;
5334 space_info->bytes_may_use -= num_bytes;
5335 trace_btrfs_space_reservation(fs_info, "space_info",
5340 list_del_init(&ticket.list);
5343 spin_unlock(&space_info->lock);
5344 ASSERT(list_empty(&ticket.list));
5349 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5350 * @root - the root we're allocating for
5351 * @block_rsv - the block_rsv we're allocating for
5352 * @orig_bytes - the number of bytes we want
5353 * @flush - whether or not we can flush to make our reservation
5355 * This will reserve orgi_bytes number of bytes from the space info associated
5356 * with the block_rsv. If there is not enough space it will make an attempt to
5357 * flush out space to make room. It will do this by flushing delalloc if
5358 * possible or committing the transaction. If flush is 0 then no attempts to
5359 * regain reservations will be made and this will fail if there is not enough
5362 static int reserve_metadata_bytes(struct btrfs_root *root,
5363 struct btrfs_block_rsv *block_rsv,
5365 enum btrfs_reserve_flush_enum flush)
5367 struct btrfs_fs_info *fs_info = root->fs_info;
5368 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5370 bool system_chunk = (root == fs_info->chunk_root);
5372 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5373 orig_bytes, flush, system_chunk);
5374 if (ret == -ENOSPC &&
5375 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5376 if (block_rsv != global_rsv &&
5377 !block_rsv_use_bytes(global_rsv, orig_bytes))
5380 if (ret == -ENOSPC) {
5381 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5382 block_rsv->space_info->flags,
5385 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5386 dump_space_info(fs_info, block_rsv->space_info,
5392 static struct btrfs_block_rsv *get_block_rsv(
5393 const struct btrfs_trans_handle *trans,
5394 const struct btrfs_root *root)
5396 struct btrfs_fs_info *fs_info = root->fs_info;
5397 struct btrfs_block_rsv *block_rsv = NULL;
5399 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5400 (root == fs_info->csum_root && trans->adding_csums) ||
5401 (root == fs_info->uuid_root))
5402 block_rsv = trans->block_rsv;
5405 block_rsv = root->block_rsv;
5408 block_rsv = &fs_info->empty_block_rsv;
5413 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5417 spin_lock(&block_rsv->lock);
5418 if (block_rsv->reserved >= num_bytes) {
5419 block_rsv->reserved -= num_bytes;
5420 if (block_rsv->reserved < block_rsv->size)
5421 block_rsv->full = 0;
5424 spin_unlock(&block_rsv->lock);
5428 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5429 u64 num_bytes, int update_size)
5431 spin_lock(&block_rsv->lock);
5432 block_rsv->reserved += num_bytes;
5434 block_rsv->size += num_bytes;
5435 else if (block_rsv->reserved >= block_rsv->size)
5436 block_rsv->full = 1;
5437 spin_unlock(&block_rsv->lock);
5440 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5441 struct btrfs_block_rsv *dest, u64 num_bytes,
5444 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5447 if (global_rsv->space_info != dest->space_info)
5450 spin_lock(&global_rsv->lock);
5451 min_bytes = div_factor(global_rsv->size, min_factor);
5452 if (global_rsv->reserved < min_bytes + num_bytes) {
5453 spin_unlock(&global_rsv->lock);
5456 global_rsv->reserved -= num_bytes;
5457 if (global_rsv->reserved < global_rsv->size)
5458 global_rsv->full = 0;
5459 spin_unlock(&global_rsv->lock);
5461 block_rsv_add_bytes(dest, num_bytes, 1);
5466 * This is for space we already have accounted in space_info->bytes_may_use, so
5467 * basically when we're returning space from block_rsv's.
5469 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5470 struct btrfs_space_info *space_info,
5473 struct reserve_ticket *ticket;
5474 struct list_head *head;
5476 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5477 bool check_overcommit = false;
5479 spin_lock(&space_info->lock);
5480 head = &space_info->priority_tickets;
5483 * If we are over our limit then we need to check and see if we can
5484 * overcommit, and if we can't then we just need to free up our space
5485 * and not satisfy any requests.
5487 used = btrfs_space_info_used(space_info, true);
5488 if (used - num_bytes >= space_info->total_bytes)
5489 check_overcommit = true;
5491 while (!list_empty(head) && num_bytes) {
5492 ticket = list_first_entry(head, struct reserve_ticket,
5495 * We use 0 bytes because this space is already reserved, so
5496 * adding the ticket space would be a double count.
5498 if (check_overcommit &&
5499 !can_overcommit(fs_info, space_info, 0, flush, false))
5501 if (num_bytes >= ticket->bytes) {
5502 list_del_init(&ticket->list);
5503 num_bytes -= ticket->bytes;
5505 space_info->tickets_id++;
5506 wake_up(&ticket->wait);
5508 ticket->bytes -= num_bytes;
5513 if (num_bytes && head == &space_info->priority_tickets) {
5514 head = &space_info->tickets;
5515 flush = BTRFS_RESERVE_FLUSH_ALL;
5518 space_info->bytes_may_use -= num_bytes;
5519 trace_btrfs_space_reservation(fs_info, "space_info",
5520 space_info->flags, num_bytes, 0);
5521 spin_unlock(&space_info->lock);
5525 * This is for newly allocated space that isn't accounted in
5526 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5527 * we use this helper.
5529 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5530 struct btrfs_space_info *space_info,
5533 struct reserve_ticket *ticket;
5534 struct list_head *head = &space_info->priority_tickets;
5537 while (!list_empty(head) && num_bytes) {
5538 ticket = list_first_entry(head, struct reserve_ticket,
5540 if (num_bytes >= ticket->bytes) {
5541 trace_btrfs_space_reservation(fs_info, "space_info",
5544 list_del_init(&ticket->list);
5545 num_bytes -= ticket->bytes;
5546 space_info->bytes_may_use += ticket->bytes;
5548 space_info->tickets_id++;
5549 wake_up(&ticket->wait);
5551 trace_btrfs_space_reservation(fs_info, "space_info",
5554 space_info->bytes_may_use += num_bytes;
5555 ticket->bytes -= num_bytes;
5560 if (num_bytes && head == &space_info->priority_tickets) {
5561 head = &space_info->tickets;
5566 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5567 struct btrfs_block_rsv *block_rsv,
5568 struct btrfs_block_rsv *dest, u64 num_bytes,
5569 u64 *qgroup_to_release_ret)
5571 struct btrfs_space_info *space_info = block_rsv->space_info;
5572 u64 qgroup_to_release = 0;
5575 spin_lock(&block_rsv->lock);
5576 if (num_bytes == (u64)-1) {
5577 num_bytes = block_rsv->size;
5578 qgroup_to_release = block_rsv->qgroup_rsv_size;
5580 block_rsv->size -= num_bytes;
5581 if (block_rsv->reserved >= block_rsv->size) {
5582 num_bytes = block_rsv->reserved - block_rsv->size;
5583 block_rsv->reserved = block_rsv->size;
5584 block_rsv->full = 1;
5588 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5589 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5590 block_rsv->qgroup_rsv_size;
5591 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5593 qgroup_to_release = 0;
5595 spin_unlock(&block_rsv->lock);
5598 if (num_bytes > 0) {
5600 spin_lock(&dest->lock);
5604 bytes_to_add = dest->size - dest->reserved;
5605 bytes_to_add = min(num_bytes, bytes_to_add);
5606 dest->reserved += bytes_to_add;
5607 if (dest->reserved >= dest->size)
5609 num_bytes -= bytes_to_add;
5611 spin_unlock(&dest->lock);
5614 space_info_add_old_bytes(fs_info, space_info,
5617 if (qgroup_to_release_ret)
5618 *qgroup_to_release_ret = qgroup_to_release;
5622 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5623 struct btrfs_block_rsv *dst, u64 num_bytes,
5628 ret = block_rsv_use_bytes(src, num_bytes);
5632 block_rsv_add_bytes(dst, num_bytes, update_size);
5636 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5638 memset(rsv, 0, sizeof(*rsv));
5639 spin_lock_init(&rsv->lock);
5643 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5644 struct btrfs_block_rsv *rsv,
5645 unsigned short type)
5647 btrfs_init_block_rsv(rsv, type);
5648 rsv->space_info = __find_space_info(fs_info,
5649 BTRFS_BLOCK_GROUP_METADATA);
5652 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5653 unsigned short type)
5655 struct btrfs_block_rsv *block_rsv;
5657 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5661 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5665 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5666 struct btrfs_block_rsv *rsv)
5670 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5674 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5679 int btrfs_block_rsv_add(struct btrfs_root *root,
5680 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5681 enum btrfs_reserve_flush_enum flush)
5688 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5690 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5697 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5705 spin_lock(&block_rsv->lock);
5706 num_bytes = div_factor(block_rsv->size, min_factor);
5707 if (block_rsv->reserved >= num_bytes)
5709 spin_unlock(&block_rsv->lock);
5714 int btrfs_block_rsv_refill(struct btrfs_root *root,
5715 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5716 enum btrfs_reserve_flush_enum flush)
5724 spin_lock(&block_rsv->lock);
5725 num_bytes = min_reserved;
5726 if (block_rsv->reserved >= num_bytes)
5729 num_bytes -= block_rsv->reserved;
5730 spin_unlock(&block_rsv->lock);
5735 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5737 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5745 * btrfs_inode_rsv_refill - refill the inode block rsv.
5746 * @inode - the inode we are refilling.
5747 * @flush - the flusing restriction.
5749 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5750 * block_rsv->size as the minimum size. We'll either refill the missing amount
5751 * or return if we already have enough space. This will also handle the resreve
5752 * tracepoint for the reserved amount.
5754 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5755 enum btrfs_reserve_flush_enum flush)
5757 struct btrfs_root *root = inode->root;
5758 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5760 u64 qgroup_num_bytes = 0;
5763 spin_lock(&block_rsv->lock);
5764 if (block_rsv->reserved < block_rsv->size)
5765 num_bytes = block_rsv->size - block_rsv->reserved;
5766 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5767 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5768 block_rsv->qgroup_rsv_reserved;
5769 spin_unlock(&block_rsv->lock);
5774 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5777 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5779 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5780 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5781 btrfs_ino(inode), num_bytes, 1);
5783 /* Don't forget to increase qgroup_rsv_reserved */
5784 spin_lock(&block_rsv->lock);
5785 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5786 spin_unlock(&block_rsv->lock);
5788 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5793 * btrfs_inode_rsv_release - release any excessive reservation.
5794 * @inode - the inode we need to release from.
5795 * @qgroup_free - free or convert qgroup meta.
5796 * Unlike normal operation, qgroup meta reservation needs to know if we are
5797 * freeing qgroup reservation or just converting it into per-trans. Normally
5798 * @qgroup_free is true for error handling, and false for normal release.
5800 * This is the same as btrfs_block_rsv_release, except that it handles the
5801 * tracepoint for the reservation.
5803 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5805 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5806 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5807 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5809 u64 qgroup_to_release = 0;
5812 * Since we statically set the block_rsv->size we just want to say we
5813 * are releasing 0 bytes, and then we'll just get the reservation over
5816 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
5817 &qgroup_to_release);
5819 trace_btrfs_space_reservation(fs_info, "delalloc",
5820 btrfs_ino(inode), released, 0);
5822 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5824 btrfs_qgroup_convert_reserved_meta(inode->root,
5828 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5829 struct btrfs_block_rsv *block_rsv,
5832 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5834 if (global_rsv == block_rsv ||
5835 block_rsv->space_info != global_rsv->space_info)
5837 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
5840 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5842 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5843 struct btrfs_space_info *sinfo = block_rsv->space_info;
5847 * The global block rsv is based on the size of the extent tree, the
5848 * checksum tree and the root tree. If the fs is empty we want to set
5849 * it to a minimal amount for safety.
5851 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5852 btrfs_root_used(&fs_info->csum_root->root_item) +
5853 btrfs_root_used(&fs_info->tree_root->root_item);
5854 num_bytes = max_t(u64, num_bytes, SZ_16M);
5856 spin_lock(&sinfo->lock);
5857 spin_lock(&block_rsv->lock);
5859 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5861 if (block_rsv->reserved < block_rsv->size) {
5862 num_bytes = btrfs_space_info_used(sinfo, true);
5863 if (sinfo->total_bytes > num_bytes) {
5864 num_bytes = sinfo->total_bytes - num_bytes;
5865 num_bytes = min(num_bytes,
5866 block_rsv->size - block_rsv->reserved);
5867 block_rsv->reserved += num_bytes;
5868 sinfo->bytes_may_use += num_bytes;
5869 trace_btrfs_space_reservation(fs_info, "space_info",
5870 sinfo->flags, num_bytes,
5873 } else if (block_rsv->reserved > block_rsv->size) {
5874 num_bytes = block_rsv->reserved - block_rsv->size;
5875 sinfo->bytes_may_use -= num_bytes;
5876 trace_btrfs_space_reservation(fs_info, "space_info",
5877 sinfo->flags, num_bytes, 0);
5878 block_rsv->reserved = block_rsv->size;
5881 if (block_rsv->reserved == block_rsv->size)
5882 block_rsv->full = 1;
5884 block_rsv->full = 0;
5886 spin_unlock(&block_rsv->lock);
5887 spin_unlock(&sinfo->lock);
5890 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5892 struct btrfs_space_info *space_info;
5894 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5895 fs_info->chunk_block_rsv.space_info = space_info;
5897 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5898 fs_info->global_block_rsv.space_info = space_info;
5899 fs_info->trans_block_rsv.space_info = space_info;
5900 fs_info->empty_block_rsv.space_info = space_info;
5901 fs_info->delayed_block_rsv.space_info = space_info;
5903 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5904 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5905 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5906 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5907 if (fs_info->quota_root)
5908 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5909 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5911 update_global_block_rsv(fs_info);
5914 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5916 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5918 WARN_ON(fs_info->trans_block_rsv.size > 0);
5919 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5920 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5921 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5922 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5923 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5928 * To be called after all the new block groups attached to the transaction
5929 * handle have been created (btrfs_create_pending_block_groups()).
5931 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5933 struct btrfs_fs_info *fs_info = trans->fs_info;
5935 if (!trans->chunk_bytes_reserved)
5938 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5940 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5941 trans->chunk_bytes_reserved, NULL);
5942 trans->chunk_bytes_reserved = 0;
5945 /* Can only return 0 or -ENOSPC */
5946 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5947 struct btrfs_inode *inode)
5949 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5950 struct btrfs_root *root = inode->root;
5952 * We always use trans->block_rsv here as we will have reserved space
5953 * for our orphan when starting the transaction, using get_block_rsv()
5954 * here will sometimes make us choose the wrong block rsv as we could be
5955 * doing a reloc inode for a non refcounted root.
5957 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5958 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5961 * We need to hold space in order to delete our orphan item once we've
5962 * added it, so this takes the reservation so we can release it later
5963 * when we are truly done with the orphan item.
5965 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5967 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5969 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5972 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5974 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5975 struct btrfs_root *root = inode->root;
5976 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5978 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5980 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5984 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5985 * root: the root of the parent directory
5986 * rsv: block reservation
5987 * items: the number of items that we need do reservation
5988 * qgroup_reserved: used to return the reserved size in qgroup
5990 * This function is used to reserve the space for snapshot/subvolume
5991 * creation and deletion. Those operations are different with the
5992 * common file/directory operations, they change two fs/file trees
5993 * and root tree, the number of items that the qgroup reserves is
5994 * different with the free space reservation. So we can not use
5995 * the space reservation mechanism in start_transaction().
5997 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5998 struct btrfs_block_rsv *rsv,
6000 u64 *qgroup_reserved,
6001 bool use_global_rsv)
6005 struct btrfs_fs_info *fs_info = root->fs_info;
6006 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6008 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6009 /* One for parent inode, two for dir entries */
6010 num_bytes = 3 * fs_info->nodesize;
6011 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
6018 *qgroup_reserved = num_bytes;
6020 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6021 rsv->space_info = __find_space_info(fs_info,
6022 BTRFS_BLOCK_GROUP_METADATA);
6023 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6024 BTRFS_RESERVE_FLUSH_ALL);
6026 if (ret == -ENOSPC && use_global_rsv)
6027 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
6029 if (ret && *qgroup_reserved)
6030 btrfs_qgroup_free_meta_prealloc(root, *qgroup_reserved);
6035 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6036 struct btrfs_block_rsv *rsv)
6038 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6041 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6042 struct btrfs_inode *inode)
6044 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6045 u64 reserve_size = 0;
6046 u64 qgroup_rsv_size = 0;
6048 unsigned outstanding_extents;
6050 lockdep_assert_held(&inode->lock);
6051 outstanding_extents = inode->outstanding_extents;
6052 if (outstanding_extents)
6053 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6054 outstanding_extents + 1);
6055 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6057 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6060 * For qgroup rsv, the calculation is very simple:
6061 * account one nodesize for each outstanding extent
6063 * This is overestimating in most cases.
6065 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
6067 spin_lock(&block_rsv->lock);
6068 block_rsv->size = reserve_size;
6069 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6070 spin_unlock(&block_rsv->lock);
6073 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6075 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6076 unsigned nr_extents;
6077 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6079 bool delalloc_lock = true;
6081 /* If we are a free space inode we need to not flush since we will be in
6082 * the middle of a transaction commit. We also don't need the delalloc
6083 * mutex since we won't race with anybody. We need this mostly to make
6084 * lockdep shut its filthy mouth.
6086 * If we have a transaction open (can happen if we call truncate_block
6087 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6089 if (btrfs_is_free_space_inode(inode)) {
6090 flush = BTRFS_RESERVE_NO_FLUSH;
6091 delalloc_lock = false;
6093 if (current->journal_info)
6094 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6096 if (btrfs_transaction_in_commit(fs_info))
6097 schedule_timeout(1);
6101 mutex_lock(&inode->delalloc_mutex);
6103 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6105 /* Add our new extents and calculate the new rsv size. */
6106 spin_lock(&inode->lock);
6107 nr_extents = count_max_extents(num_bytes);
6108 btrfs_mod_outstanding_extents(inode, nr_extents);
6109 inode->csum_bytes += num_bytes;
6110 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6111 spin_unlock(&inode->lock);
6113 ret = btrfs_inode_rsv_refill(inode, flush);
6118 mutex_unlock(&inode->delalloc_mutex);
6122 spin_lock(&inode->lock);
6123 nr_extents = count_max_extents(num_bytes);
6124 btrfs_mod_outstanding_extents(inode, -nr_extents);
6125 inode->csum_bytes -= num_bytes;
6126 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6127 spin_unlock(&inode->lock);
6129 btrfs_inode_rsv_release(inode, true);
6131 mutex_unlock(&inode->delalloc_mutex);
6136 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6137 * @inode: the inode to release the reservation for.
6138 * @num_bytes: the number of bytes we are releasing.
6139 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6141 * This will release the metadata reservation for an inode. This can be called
6142 * once we complete IO for a given set of bytes to release their metadata
6143 * reservations, or on error for the same reason.
6145 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6148 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6150 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6151 spin_lock(&inode->lock);
6152 inode->csum_bytes -= num_bytes;
6153 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6154 spin_unlock(&inode->lock);
6156 if (btrfs_is_testing(fs_info))
6159 btrfs_inode_rsv_release(inode, qgroup_free);
6163 * btrfs_delalloc_release_extents - release our outstanding_extents
6164 * @inode: the inode to balance the reservation for.
6165 * @num_bytes: the number of bytes we originally reserved with
6166 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6168 * When we reserve space we increase outstanding_extents for the extents we may
6169 * add. Once we've set the range as delalloc or created our ordered extents we
6170 * have outstanding_extents to track the real usage, so we use this to free our
6171 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6172 * with btrfs_delalloc_reserve_metadata.
6174 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6177 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6178 unsigned num_extents;
6180 spin_lock(&inode->lock);
6181 num_extents = count_max_extents(num_bytes);
6182 btrfs_mod_outstanding_extents(inode, -num_extents);
6183 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6184 spin_unlock(&inode->lock);
6186 if (btrfs_is_testing(fs_info))
6189 btrfs_inode_rsv_release(inode, qgroup_free);
6193 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6195 * @inode: inode we're writing to
6196 * @start: start range we are writing to
6197 * @len: how long the range we are writing to
6198 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6199 * current reservation.
6201 * This will do the following things
6203 * o reserve space in data space info for num bytes
6204 * and reserve precious corresponding qgroup space
6205 * (Done in check_data_free_space)
6207 * o reserve space for metadata space, based on the number of outstanding
6208 * extents and how much csums will be needed
6209 * also reserve metadata space in a per root over-reserve method.
6210 * o add to the inodes->delalloc_bytes
6211 * o add it to the fs_info's delalloc inodes list.
6212 * (Above 3 all done in delalloc_reserve_metadata)
6214 * Return 0 for success
6215 * Return <0 for error(-ENOSPC or -EQUOT)
6217 int btrfs_delalloc_reserve_space(struct inode *inode,
6218 struct extent_changeset **reserved, u64 start, u64 len)
6222 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6225 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6227 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6232 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6233 * @inode: inode we're releasing space for
6234 * @start: start position of the space already reserved
6235 * @len: the len of the space already reserved
6236 * @release_bytes: the len of the space we consumed or didn't use
6238 * This function will release the metadata space that was not used and will
6239 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6240 * list if there are no delalloc bytes left.
6241 * Also it will handle the qgroup reserved space.
6243 void btrfs_delalloc_release_space(struct inode *inode,
6244 struct extent_changeset *reserved,
6245 u64 start, u64 len, bool qgroup_free)
6247 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6248 btrfs_free_reserved_data_space(inode, reserved, start, len);
6251 static int update_block_group(struct btrfs_trans_handle *trans,
6252 struct btrfs_fs_info *info, u64 bytenr,
6253 u64 num_bytes, int alloc)
6255 struct btrfs_block_group_cache *cache = NULL;
6256 u64 total = num_bytes;
6261 /* block accounting for super block */
6262 spin_lock(&info->delalloc_root_lock);
6263 old_val = btrfs_super_bytes_used(info->super_copy);
6265 old_val += num_bytes;
6267 old_val -= num_bytes;
6268 btrfs_set_super_bytes_used(info->super_copy, old_val);
6269 spin_unlock(&info->delalloc_root_lock);
6272 cache = btrfs_lookup_block_group(info, bytenr);
6275 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6276 BTRFS_BLOCK_GROUP_RAID1 |
6277 BTRFS_BLOCK_GROUP_RAID10))
6282 * If this block group has free space cache written out, we
6283 * need to make sure to load it if we are removing space. This
6284 * is because we need the unpinning stage to actually add the
6285 * space back to the block group, otherwise we will leak space.
6287 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6288 cache_block_group(cache, 1);
6290 byte_in_group = bytenr - cache->key.objectid;
6291 WARN_ON(byte_in_group > cache->key.offset);
6293 spin_lock(&cache->space_info->lock);
6294 spin_lock(&cache->lock);
6296 if (btrfs_test_opt(info, SPACE_CACHE) &&
6297 cache->disk_cache_state < BTRFS_DC_CLEAR)
6298 cache->disk_cache_state = BTRFS_DC_CLEAR;
6300 old_val = btrfs_block_group_used(&cache->item);
6301 num_bytes = min(total, cache->key.offset - byte_in_group);
6303 old_val += num_bytes;
6304 btrfs_set_block_group_used(&cache->item, old_val);
6305 cache->reserved -= num_bytes;
6306 cache->space_info->bytes_reserved -= num_bytes;
6307 cache->space_info->bytes_used += num_bytes;
6308 cache->space_info->disk_used += num_bytes * factor;
6309 spin_unlock(&cache->lock);
6310 spin_unlock(&cache->space_info->lock);
6312 old_val -= num_bytes;
6313 btrfs_set_block_group_used(&cache->item, old_val);
6314 cache->pinned += num_bytes;
6315 cache->space_info->bytes_pinned += num_bytes;
6316 cache->space_info->bytes_used -= num_bytes;
6317 cache->space_info->disk_used -= num_bytes * factor;
6318 spin_unlock(&cache->lock);
6319 spin_unlock(&cache->space_info->lock);
6321 trace_btrfs_space_reservation(info, "pinned",
6322 cache->space_info->flags,
6324 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6326 set_extent_dirty(info->pinned_extents,
6327 bytenr, bytenr + num_bytes - 1,
6328 GFP_NOFS | __GFP_NOFAIL);
6331 spin_lock(&trans->transaction->dirty_bgs_lock);
6332 if (list_empty(&cache->dirty_list)) {
6333 list_add_tail(&cache->dirty_list,
6334 &trans->transaction->dirty_bgs);
6335 trans->transaction->num_dirty_bgs++;
6336 btrfs_get_block_group(cache);
6338 spin_unlock(&trans->transaction->dirty_bgs_lock);
6341 * No longer have used bytes in this block group, queue it for
6342 * deletion. We do this after adding the block group to the
6343 * dirty list to avoid races between cleaner kthread and space
6346 if (!alloc && old_val == 0) {
6347 spin_lock(&info->unused_bgs_lock);
6348 if (list_empty(&cache->bg_list)) {
6349 btrfs_get_block_group(cache);
6350 list_add_tail(&cache->bg_list,
6353 spin_unlock(&info->unused_bgs_lock);
6356 btrfs_put_block_group(cache);
6358 bytenr += num_bytes;
6363 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6365 struct btrfs_block_group_cache *cache;
6368 spin_lock(&fs_info->block_group_cache_lock);
6369 bytenr = fs_info->first_logical_byte;
6370 spin_unlock(&fs_info->block_group_cache_lock);
6372 if (bytenr < (u64)-1)
6375 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6379 bytenr = cache->key.objectid;
6380 btrfs_put_block_group(cache);
6385 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6386 struct btrfs_block_group_cache *cache,
6387 u64 bytenr, u64 num_bytes, int reserved)
6389 spin_lock(&cache->space_info->lock);
6390 spin_lock(&cache->lock);
6391 cache->pinned += num_bytes;
6392 cache->space_info->bytes_pinned += num_bytes;
6394 cache->reserved -= num_bytes;
6395 cache->space_info->bytes_reserved -= num_bytes;
6397 spin_unlock(&cache->lock);
6398 spin_unlock(&cache->space_info->lock);
6400 trace_btrfs_space_reservation(fs_info, "pinned",
6401 cache->space_info->flags, num_bytes, 1);
6402 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6403 set_extent_dirty(fs_info->pinned_extents, bytenr,
6404 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6409 * this function must be called within transaction
6411 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6412 u64 bytenr, u64 num_bytes, int reserved)
6414 struct btrfs_block_group_cache *cache;
6416 cache = btrfs_lookup_block_group(fs_info, bytenr);
6417 BUG_ON(!cache); /* Logic error */
6419 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6421 btrfs_put_block_group(cache);
6426 * this function must be called within transaction
6428 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6429 u64 bytenr, u64 num_bytes)
6431 struct btrfs_block_group_cache *cache;
6434 cache = btrfs_lookup_block_group(fs_info, bytenr);
6439 * pull in the free space cache (if any) so that our pin
6440 * removes the free space from the cache. We have load_only set
6441 * to one because the slow code to read in the free extents does check
6442 * the pinned extents.
6444 cache_block_group(cache, 1);
6446 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6448 /* remove us from the free space cache (if we're there at all) */
6449 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6450 btrfs_put_block_group(cache);
6454 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6455 u64 start, u64 num_bytes)
6458 struct btrfs_block_group_cache *block_group;
6459 struct btrfs_caching_control *caching_ctl;
6461 block_group = btrfs_lookup_block_group(fs_info, start);
6465 cache_block_group(block_group, 0);
6466 caching_ctl = get_caching_control(block_group);
6470 BUG_ON(!block_group_cache_done(block_group));
6471 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6473 mutex_lock(&caching_ctl->mutex);
6475 if (start >= caching_ctl->progress) {
6476 ret = add_excluded_extent(fs_info, start, num_bytes);
6477 } else if (start + num_bytes <= caching_ctl->progress) {
6478 ret = btrfs_remove_free_space(block_group,
6481 num_bytes = caching_ctl->progress - start;
6482 ret = btrfs_remove_free_space(block_group,
6487 num_bytes = (start + num_bytes) -
6488 caching_ctl->progress;
6489 start = caching_ctl->progress;
6490 ret = add_excluded_extent(fs_info, start, num_bytes);
6493 mutex_unlock(&caching_ctl->mutex);
6494 put_caching_control(caching_ctl);
6496 btrfs_put_block_group(block_group);
6500 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6501 struct extent_buffer *eb)
6503 struct btrfs_file_extent_item *item;
6504 struct btrfs_key key;
6508 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6511 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6512 btrfs_item_key_to_cpu(eb, &key, i);
6513 if (key.type != BTRFS_EXTENT_DATA_KEY)
6515 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6516 found_type = btrfs_file_extent_type(eb, item);
6517 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6519 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6521 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6522 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6523 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6530 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6532 atomic_inc(&bg->reservations);
6535 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6538 struct btrfs_block_group_cache *bg;
6540 bg = btrfs_lookup_block_group(fs_info, start);
6542 if (atomic_dec_and_test(&bg->reservations))
6543 wake_up_var(&bg->reservations);
6544 btrfs_put_block_group(bg);
6547 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6549 struct btrfs_space_info *space_info = bg->space_info;
6553 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6557 * Our block group is read only but before we set it to read only,
6558 * some task might have had allocated an extent from it already, but it
6559 * has not yet created a respective ordered extent (and added it to a
6560 * root's list of ordered extents).
6561 * Therefore wait for any task currently allocating extents, since the
6562 * block group's reservations counter is incremented while a read lock
6563 * on the groups' semaphore is held and decremented after releasing
6564 * the read access on that semaphore and creating the ordered extent.
6566 down_write(&space_info->groups_sem);
6567 up_write(&space_info->groups_sem);
6569 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6573 * btrfs_add_reserved_bytes - update the block_group and space info counters
6574 * @cache: The cache we are manipulating
6575 * @ram_bytes: The number of bytes of file content, and will be same to
6576 * @num_bytes except for the compress path.
6577 * @num_bytes: The number of bytes in question
6578 * @delalloc: The blocks are allocated for the delalloc write
6580 * This is called by the allocator when it reserves space. If this is a
6581 * reservation and the block group has become read only we cannot make the
6582 * reservation and return -EAGAIN, otherwise this function always succeeds.
6584 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6585 u64 ram_bytes, u64 num_bytes, int delalloc)
6587 struct btrfs_space_info *space_info = cache->space_info;
6590 spin_lock(&space_info->lock);
6591 spin_lock(&cache->lock);
6595 cache->reserved += num_bytes;
6596 space_info->bytes_reserved += num_bytes;
6598 trace_btrfs_space_reservation(cache->fs_info,
6599 "space_info", space_info->flags,
6601 space_info->bytes_may_use -= ram_bytes;
6603 cache->delalloc_bytes += num_bytes;
6605 spin_unlock(&cache->lock);
6606 spin_unlock(&space_info->lock);
6611 * btrfs_free_reserved_bytes - update the block_group and space info counters
6612 * @cache: The cache we are manipulating
6613 * @num_bytes: The number of bytes in question
6614 * @delalloc: The blocks are allocated for the delalloc write
6616 * This is called by somebody who is freeing space that was never actually used
6617 * on disk. For example if you reserve some space for a new leaf in transaction
6618 * A and before transaction A commits you free that leaf, you call this with
6619 * reserve set to 0 in order to clear the reservation.
6622 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6623 u64 num_bytes, int delalloc)
6625 struct btrfs_space_info *space_info = cache->space_info;
6628 spin_lock(&space_info->lock);
6629 spin_lock(&cache->lock);
6631 space_info->bytes_readonly += num_bytes;
6632 cache->reserved -= num_bytes;
6633 space_info->bytes_reserved -= num_bytes;
6636 cache->delalloc_bytes -= num_bytes;
6637 spin_unlock(&cache->lock);
6638 spin_unlock(&space_info->lock);
6641 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6643 struct btrfs_caching_control *next;
6644 struct btrfs_caching_control *caching_ctl;
6645 struct btrfs_block_group_cache *cache;
6647 down_write(&fs_info->commit_root_sem);
6649 list_for_each_entry_safe(caching_ctl, next,
6650 &fs_info->caching_block_groups, list) {
6651 cache = caching_ctl->block_group;
6652 if (block_group_cache_done(cache)) {
6653 cache->last_byte_to_unpin = (u64)-1;
6654 list_del_init(&caching_ctl->list);
6655 put_caching_control(caching_ctl);
6657 cache->last_byte_to_unpin = caching_ctl->progress;
6661 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6662 fs_info->pinned_extents = &fs_info->freed_extents[1];
6664 fs_info->pinned_extents = &fs_info->freed_extents[0];
6666 up_write(&fs_info->commit_root_sem);
6668 update_global_block_rsv(fs_info);
6672 * Returns the free cluster for the given space info and sets empty_cluster to
6673 * what it should be based on the mount options.
6675 static struct btrfs_free_cluster *
6676 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6677 struct btrfs_space_info *space_info, u64 *empty_cluster)
6679 struct btrfs_free_cluster *ret = NULL;
6682 if (btrfs_mixed_space_info(space_info))
6685 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6686 ret = &fs_info->meta_alloc_cluster;
6687 if (btrfs_test_opt(fs_info, SSD))
6688 *empty_cluster = SZ_2M;
6690 *empty_cluster = SZ_64K;
6691 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6692 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6693 *empty_cluster = SZ_2M;
6694 ret = &fs_info->data_alloc_cluster;
6700 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6702 const bool return_free_space)
6704 struct btrfs_block_group_cache *cache = NULL;
6705 struct btrfs_space_info *space_info;
6706 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6707 struct btrfs_free_cluster *cluster = NULL;
6709 u64 total_unpinned = 0;
6710 u64 empty_cluster = 0;
6713 while (start <= end) {
6716 start >= cache->key.objectid + cache->key.offset) {
6718 btrfs_put_block_group(cache);
6720 cache = btrfs_lookup_block_group(fs_info, start);
6721 BUG_ON(!cache); /* Logic error */
6723 cluster = fetch_cluster_info(fs_info,
6726 empty_cluster <<= 1;
6729 len = cache->key.objectid + cache->key.offset - start;
6730 len = min(len, end + 1 - start);
6732 if (start < cache->last_byte_to_unpin) {
6733 len = min(len, cache->last_byte_to_unpin - start);
6734 if (return_free_space)
6735 btrfs_add_free_space(cache, start, len);
6739 total_unpinned += len;
6740 space_info = cache->space_info;
6743 * If this space cluster has been marked as fragmented and we've
6744 * unpinned enough in this block group to potentially allow a
6745 * cluster to be created inside of it go ahead and clear the
6748 if (cluster && cluster->fragmented &&
6749 total_unpinned > empty_cluster) {
6750 spin_lock(&cluster->lock);
6751 cluster->fragmented = 0;
6752 spin_unlock(&cluster->lock);
6755 spin_lock(&space_info->lock);
6756 spin_lock(&cache->lock);
6757 cache->pinned -= len;
6758 space_info->bytes_pinned -= len;
6760 trace_btrfs_space_reservation(fs_info, "pinned",
6761 space_info->flags, len, 0);
6762 space_info->max_extent_size = 0;
6763 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6765 space_info->bytes_readonly += len;
6768 spin_unlock(&cache->lock);
6769 if (!readonly && return_free_space &&
6770 global_rsv->space_info == space_info) {
6773 spin_lock(&global_rsv->lock);
6774 if (!global_rsv->full) {
6775 to_add = min(len, global_rsv->size -
6776 global_rsv->reserved);
6777 global_rsv->reserved += to_add;
6778 space_info->bytes_may_use += to_add;
6779 if (global_rsv->reserved >= global_rsv->size)
6780 global_rsv->full = 1;
6781 trace_btrfs_space_reservation(fs_info,
6787 spin_unlock(&global_rsv->lock);
6788 /* Add to any tickets we may have */
6790 space_info_add_new_bytes(fs_info, space_info,
6793 spin_unlock(&space_info->lock);
6797 btrfs_put_block_group(cache);
6801 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6803 struct btrfs_fs_info *fs_info = trans->fs_info;
6804 struct btrfs_block_group_cache *block_group, *tmp;
6805 struct list_head *deleted_bgs;
6806 struct extent_io_tree *unpin;
6811 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6812 unpin = &fs_info->freed_extents[1];
6814 unpin = &fs_info->freed_extents[0];
6816 while (!trans->aborted) {
6817 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6818 ret = find_first_extent_bit(unpin, 0, &start, &end,
6819 EXTENT_DIRTY, NULL);
6821 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6825 if (btrfs_test_opt(fs_info, DISCARD))
6826 ret = btrfs_discard_extent(fs_info, start,
6827 end + 1 - start, NULL);
6829 clear_extent_dirty(unpin, start, end);
6830 unpin_extent_range(fs_info, start, end, true);
6831 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6836 * Transaction is finished. We don't need the lock anymore. We
6837 * do need to clean up the block groups in case of a transaction
6840 deleted_bgs = &trans->transaction->deleted_bgs;
6841 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6845 if (!trans->aborted)
6846 ret = btrfs_discard_extent(fs_info,
6847 block_group->key.objectid,
6848 block_group->key.offset,
6851 list_del_init(&block_group->bg_list);
6852 btrfs_put_block_group_trimming(block_group);
6853 btrfs_put_block_group(block_group);
6856 const char *errstr = btrfs_decode_error(ret);
6858 "discard failed while removing blockgroup: errno=%d %s",
6866 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6867 struct btrfs_fs_info *info,
6868 struct btrfs_delayed_ref_node *node, u64 parent,
6869 u64 root_objectid, u64 owner_objectid,
6870 u64 owner_offset, int refs_to_drop,
6871 struct btrfs_delayed_extent_op *extent_op)
6873 struct btrfs_key key;
6874 struct btrfs_path *path;
6875 struct btrfs_root *extent_root = info->extent_root;
6876 struct extent_buffer *leaf;
6877 struct btrfs_extent_item *ei;
6878 struct btrfs_extent_inline_ref *iref;
6881 int extent_slot = 0;
6882 int found_extent = 0;
6886 u64 bytenr = node->bytenr;
6887 u64 num_bytes = node->num_bytes;
6889 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6891 path = btrfs_alloc_path();
6895 path->reada = READA_FORWARD;
6896 path->leave_spinning = 1;
6898 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6899 BUG_ON(!is_data && refs_to_drop != 1);
6902 skinny_metadata = false;
6904 ret = lookup_extent_backref(trans, info, path, &iref,
6905 bytenr, num_bytes, parent,
6906 root_objectid, owner_objectid,
6909 extent_slot = path->slots[0];
6910 while (extent_slot >= 0) {
6911 btrfs_item_key_to_cpu(path->nodes[0], &key,
6913 if (key.objectid != bytenr)
6915 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6916 key.offset == num_bytes) {
6920 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6921 key.offset == owner_objectid) {
6925 if (path->slots[0] - extent_slot > 5)
6929 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6930 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6931 if (found_extent && item_size < sizeof(*ei))
6934 if (!found_extent) {
6936 ret = remove_extent_backref(trans, info, path, NULL,
6938 is_data, &last_ref);
6940 btrfs_abort_transaction(trans, ret);
6943 btrfs_release_path(path);
6944 path->leave_spinning = 1;
6946 key.objectid = bytenr;
6947 key.type = BTRFS_EXTENT_ITEM_KEY;
6948 key.offset = num_bytes;
6950 if (!is_data && skinny_metadata) {
6951 key.type = BTRFS_METADATA_ITEM_KEY;
6952 key.offset = owner_objectid;
6955 ret = btrfs_search_slot(trans, extent_root,
6957 if (ret > 0 && skinny_metadata && path->slots[0]) {
6959 * Couldn't find our skinny metadata item,
6960 * see if we have ye olde extent item.
6963 btrfs_item_key_to_cpu(path->nodes[0], &key,
6965 if (key.objectid == bytenr &&
6966 key.type == BTRFS_EXTENT_ITEM_KEY &&
6967 key.offset == num_bytes)
6971 if (ret > 0 && skinny_metadata) {
6972 skinny_metadata = false;
6973 key.objectid = bytenr;
6974 key.type = BTRFS_EXTENT_ITEM_KEY;
6975 key.offset = num_bytes;
6976 btrfs_release_path(path);
6977 ret = btrfs_search_slot(trans, extent_root,
6983 "umm, got %d back from search, was looking for %llu",
6986 btrfs_print_leaf(path->nodes[0]);
6989 btrfs_abort_transaction(trans, ret);
6992 extent_slot = path->slots[0];
6994 } else if (WARN_ON(ret == -ENOENT)) {
6995 btrfs_print_leaf(path->nodes[0]);
6997 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6998 bytenr, parent, root_objectid, owner_objectid,
7000 btrfs_abort_transaction(trans, ret);
7003 btrfs_abort_transaction(trans, ret);
7007 leaf = path->nodes[0];
7008 item_size = btrfs_item_size_nr(leaf, extent_slot);
7009 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7010 if (item_size < sizeof(*ei)) {
7011 BUG_ON(found_extent || extent_slot != path->slots[0]);
7012 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
7015 btrfs_abort_transaction(trans, ret);
7019 btrfs_release_path(path);
7020 path->leave_spinning = 1;
7022 key.objectid = bytenr;
7023 key.type = BTRFS_EXTENT_ITEM_KEY;
7024 key.offset = num_bytes;
7026 ret = btrfs_search_slot(trans, extent_root, &key, path,
7030 "umm, got %d back from search, was looking for %llu",
7032 btrfs_print_leaf(path->nodes[0]);
7035 btrfs_abort_transaction(trans, ret);
7039 extent_slot = path->slots[0];
7040 leaf = path->nodes[0];
7041 item_size = btrfs_item_size_nr(leaf, extent_slot);
7044 BUG_ON(item_size < sizeof(*ei));
7045 ei = btrfs_item_ptr(leaf, extent_slot,
7046 struct btrfs_extent_item);
7047 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7048 key.type == BTRFS_EXTENT_ITEM_KEY) {
7049 struct btrfs_tree_block_info *bi;
7050 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7051 bi = (struct btrfs_tree_block_info *)(ei + 1);
7052 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7055 refs = btrfs_extent_refs(leaf, ei);
7056 if (refs < refs_to_drop) {
7058 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7059 refs_to_drop, refs, bytenr);
7061 btrfs_abort_transaction(trans, ret);
7064 refs -= refs_to_drop;
7068 __run_delayed_extent_op(extent_op, leaf, ei);
7070 * In the case of inline back ref, reference count will
7071 * be updated by remove_extent_backref
7074 BUG_ON(!found_extent);
7076 btrfs_set_extent_refs(leaf, ei, refs);
7077 btrfs_mark_buffer_dirty(leaf);
7080 ret = remove_extent_backref(trans, info, path,
7082 is_data, &last_ref);
7084 btrfs_abort_transaction(trans, ret);
7090 BUG_ON(is_data && refs_to_drop !=
7091 extent_data_ref_count(path, iref));
7093 BUG_ON(path->slots[0] != extent_slot);
7095 BUG_ON(path->slots[0] != extent_slot + 1);
7096 path->slots[0] = extent_slot;
7102 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7105 btrfs_abort_transaction(trans, ret);
7108 btrfs_release_path(path);
7111 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7113 btrfs_abort_transaction(trans, ret);
7118 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7120 btrfs_abort_transaction(trans, ret);
7124 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7126 btrfs_abort_transaction(trans, ret);
7130 btrfs_release_path(path);
7133 btrfs_free_path(path);
7138 * when we free an block, it is possible (and likely) that we free the last
7139 * delayed ref for that extent as well. This searches the delayed ref tree for
7140 * a given extent, and if there are no other delayed refs to be processed, it
7141 * removes it from the tree.
7143 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7146 struct btrfs_delayed_ref_head *head;
7147 struct btrfs_delayed_ref_root *delayed_refs;
7150 delayed_refs = &trans->transaction->delayed_refs;
7151 spin_lock(&delayed_refs->lock);
7152 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7154 goto out_delayed_unlock;
7156 spin_lock(&head->lock);
7157 if (!RB_EMPTY_ROOT(&head->ref_tree))
7160 if (head->extent_op) {
7161 if (!head->must_insert_reserved)
7163 btrfs_free_delayed_extent_op(head->extent_op);
7164 head->extent_op = NULL;
7168 * waiting for the lock here would deadlock. If someone else has it
7169 * locked they are already in the process of dropping it anyway
7171 if (!mutex_trylock(&head->mutex))
7175 * at this point we have a head with no other entries. Go
7176 * ahead and process it.
7178 rb_erase(&head->href_node, &delayed_refs->href_root);
7179 RB_CLEAR_NODE(&head->href_node);
7180 atomic_dec(&delayed_refs->num_entries);
7183 * we don't take a ref on the node because we're removing it from the
7184 * tree, so we just steal the ref the tree was holding.
7186 delayed_refs->num_heads--;
7187 if (head->processing == 0)
7188 delayed_refs->num_heads_ready--;
7189 head->processing = 0;
7190 spin_unlock(&head->lock);
7191 spin_unlock(&delayed_refs->lock);
7193 BUG_ON(head->extent_op);
7194 if (head->must_insert_reserved)
7197 mutex_unlock(&head->mutex);
7198 btrfs_put_delayed_ref_head(head);
7201 spin_unlock(&head->lock);
7204 spin_unlock(&delayed_refs->lock);
7208 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7209 struct btrfs_root *root,
7210 struct extent_buffer *buf,
7211 u64 parent, int last_ref)
7213 struct btrfs_fs_info *fs_info = root->fs_info;
7217 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7218 int old_ref_mod, new_ref_mod;
7220 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7221 root->root_key.objectid,
7222 btrfs_header_level(buf), 0,
7223 BTRFS_DROP_DELAYED_REF);
7224 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7226 root->root_key.objectid,
7227 btrfs_header_level(buf),
7228 BTRFS_DROP_DELAYED_REF, NULL,
7229 &old_ref_mod, &new_ref_mod);
7230 BUG_ON(ret); /* -ENOMEM */
7231 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7234 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7235 struct btrfs_block_group_cache *cache;
7237 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7238 ret = check_ref_cleanup(trans, buf->start);
7244 cache = btrfs_lookup_block_group(fs_info, buf->start);
7246 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7247 pin_down_extent(fs_info, cache, buf->start,
7249 btrfs_put_block_group(cache);
7253 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7255 btrfs_add_free_space(cache, buf->start, buf->len);
7256 btrfs_free_reserved_bytes(cache, buf->len, 0);
7257 btrfs_put_block_group(cache);
7258 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7262 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7263 root->root_key.objectid);
7267 * Deleting the buffer, clear the corrupt flag since it doesn't
7270 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7274 /* Can return -ENOMEM */
7275 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7276 struct btrfs_root *root,
7277 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7278 u64 owner, u64 offset)
7280 struct btrfs_fs_info *fs_info = root->fs_info;
7281 int old_ref_mod, new_ref_mod;
7284 if (btrfs_is_testing(fs_info))
7287 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7288 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7289 root_objectid, owner, offset,
7290 BTRFS_DROP_DELAYED_REF);
7293 * tree log blocks never actually go into the extent allocation
7294 * tree, just update pinning info and exit early.
7296 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7297 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7298 /* unlocks the pinned mutex */
7299 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7300 old_ref_mod = new_ref_mod = 0;
7302 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7303 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7305 root_objectid, (int)owner,
7306 BTRFS_DROP_DELAYED_REF, NULL,
7307 &old_ref_mod, &new_ref_mod);
7309 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7311 root_objectid, owner, offset,
7312 0, BTRFS_DROP_DELAYED_REF,
7313 &old_ref_mod, &new_ref_mod);
7316 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7317 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7323 * when we wait for progress in the block group caching, its because
7324 * our allocation attempt failed at least once. So, we must sleep
7325 * and let some progress happen before we try again.
7327 * This function will sleep at least once waiting for new free space to
7328 * show up, and then it will check the block group free space numbers
7329 * for our min num_bytes. Another option is to have it go ahead
7330 * and look in the rbtree for a free extent of a given size, but this
7333 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7334 * any of the information in this block group.
7336 static noinline void
7337 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7340 struct btrfs_caching_control *caching_ctl;
7342 caching_ctl = get_caching_control(cache);
7346 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7347 (cache->free_space_ctl->free_space >= num_bytes));
7349 put_caching_control(caching_ctl);
7353 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7355 struct btrfs_caching_control *caching_ctl;
7358 caching_ctl = get_caching_control(cache);
7360 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7362 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7363 if (cache->cached == BTRFS_CACHE_ERROR)
7365 put_caching_control(caching_ctl);
7369 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7370 [BTRFS_RAID_RAID10] = "raid10",
7371 [BTRFS_RAID_RAID1] = "raid1",
7372 [BTRFS_RAID_DUP] = "dup",
7373 [BTRFS_RAID_RAID0] = "raid0",
7374 [BTRFS_RAID_SINGLE] = "single",
7375 [BTRFS_RAID_RAID5] = "raid5",
7376 [BTRFS_RAID_RAID6] = "raid6",
7379 static const char *get_raid_name(enum btrfs_raid_types type)
7381 if (type >= BTRFS_NR_RAID_TYPES)
7384 return btrfs_raid_type_names[type];
7387 enum btrfs_loop_type {
7388 LOOP_CACHING_NOWAIT = 0,
7389 LOOP_CACHING_WAIT = 1,
7390 LOOP_ALLOC_CHUNK = 2,
7391 LOOP_NO_EMPTY_SIZE = 3,
7395 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7399 down_read(&cache->data_rwsem);
7403 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7406 btrfs_get_block_group(cache);
7408 down_read(&cache->data_rwsem);
7411 static struct btrfs_block_group_cache *
7412 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7413 struct btrfs_free_cluster *cluster,
7416 struct btrfs_block_group_cache *used_bg = NULL;
7418 spin_lock(&cluster->refill_lock);
7420 used_bg = cluster->block_group;
7424 if (used_bg == block_group)
7427 btrfs_get_block_group(used_bg);
7432 if (down_read_trylock(&used_bg->data_rwsem))
7435 spin_unlock(&cluster->refill_lock);
7437 /* We should only have one-level nested. */
7438 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7440 spin_lock(&cluster->refill_lock);
7441 if (used_bg == cluster->block_group)
7444 up_read(&used_bg->data_rwsem);
7445 btrfs_put_block_group(used_bg);
7450 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7454 up_read(&cache->data_rwsem);
7455 btrfs_put_block_group(cache);
7459 * walks the btree of allocated extents and find a hole of a given size.
7460 * The key ins is changed to record the hole:
7461 * ins->objectid == start position
7462 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7463 * ins->offset == the size of the hole.
7464 * Any available blocks before search_start are skipped.
7466 * If there is no suitable free space, we will record the max size of
7467 * the free space extent currently.
7469 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7470 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7471 u64 hint_byte, struct btrfs_key *ins,
7472 u64 flags, int delalloc)
7475 struct btrfs_root *root = fs_info->extent_root;
7476 struct btrfs_free_cluster *last_ptr = NULL;
7477 struct btrfs_block_group_cache *block_group = NULL;
7478 u64 search_start = 0;
7479 u64 max_extent_size = 0;
7480 u64 empty_cluster = 0;
7481 struct btrfs_space_info *space_info;
7483 int index = btrfs_bg_flags_to_raid_index(flags);
7484 bool failed_cluster_refill = false;
7485 bool failed_alloc = false;
7486 bool use_cluster = true;
7487 bool have_caching_bg = false;
7488 bool orig_have_caching_bg = false;
7489 bool full_search = false;
7491 WARN_ON(num_bytes < fs_info->sectorsize);
7492 ins->type = BTRFS_EXTENT_ITEM_KEY;
7496 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7498 space_info = __find_space_info(fs_info, flags);
7500 btrfs_err(fs_info, "No space info for %llu", flags);
7505 * If our free space is heavily fragmented we may not be able to make
7506 * big contiguous allocations, so instead of doing the expensive search
7507 * for free space, simply return ENOSPC with our max_extent_size so we
7508 * can go ahead and search for a more manageable chunk.
7510 * If our max_extent_size is large enough for our allocation simply
7511 * disable clustering since we will likely not be able to find enough
7512 * space to create a cluster and induce latency trying.
7514 if (unlikely(space_info->max_extent_size)) {
7515 spin_lock(&space_info->lock);
7516 if (space_info->max_extent_size &&
7517 num_bytes > space_info->max_extent_size) {
7518 ins->offset = space_info->max_extent_size;
7519 spin_unlock(&space_info->lock);
7521 } else if (space_info->max_extent_size) {
7522 use_cluster = false;
7524 spin_unlock(&space_info->lock);
7527 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7529 spin_lock(&last_ptr->lock);
7530 if (last_ptr->block_group)
7531 hint_byte = last_ptr->window_start;
7532 if (last_ptr->fragmented) {
7534 * We still set window_start so we can keep track of the
7535 * last place we found an allocation to try and save
7538 hint_byte = last_ptr->window_start;
7539 use_cluster = false;
7541 spin_unlock(&last_ptr->lock);
7544 search_start = max(search_start, first_logical_byte(fs_info, 0));
7545 search_start = max(search_start, hint_byte);
7546 if (search_start == hint_byte) {
7547 block_group = btrfs_lookup_block_group(fs_info, search_start);
7549 * we don't want to use the block group if it doesn't match our
7550 * allocation bits, or if its not cached.
7552 * However if we are re-searching with an ideal block group
7553 * picked out then we don't care that the block group is cached.
7555 if (block_group && block_group_bits(block_group, flags) &&
7556 block_group->cached != BTRFS_CACHE_NO) {
7557 down_read(&space_info->groups_sem);
7558 if (list_empty(&block_group->list) ||
7561 * someone is removing this block group,
7562 * we can't jump into the have_block_group
7563 * target because our list pointers are not
7566 btrfs_put_block_group(block_group);
7567 up_read(&space_info->groups_sem);
7569 index = btrfs_bg_flags_to_raid_index(
7570 block_group->flags);
7571 btrfs_lock_block_group(block_group, delalloc);
7572 goto have_block_group;
7574 } else if (block_group) {
7575 btrfs_put_block_group(block_group);
7579 have_caching_bg = false;
7580 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7582 down_read(&space_info->groups_sem);
7583 list_for_each_entry(block_group, &space_info->block_groups[index],
7588 /* If the block group is read-only, we can skip it entirely. */
7589 if (unlikely(block_group->ro))
7592 btrfs_grab_block_group(block_group, delalloc);
7593 search_start = block_group->key.objectid;
7596 * this can happen if we end up cycling through all the
7597 * raid types, but we want to make sure we only allocate
7598 * for the proper type.
7600 if (!block_group_bits(block_group, flags)) {
7601 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7602 BTRFS_BLOCK_GROUP_RAID1 |
7603 BTRFS_BLOCK_GROUP_RAID5 |
7604 BTRFS_BLOCK_GROUP_RAID6 |
7605 BTRFS_BLOCK_GROUP_RAID10;
7608 * if they asked for extra copies and this block group
7609 * doesn't provide them, bail. This does allow us to
7610 * fill raid0 from raid1.
7612 if ((flags & extra) && !(block_group->flags & extra))
7617 cached = block_group_cache_done(block_group);
7618 if (unlikely(!cached)) {
7619 have_caching_bg = true;
7620 ret = cache_block_group(block_group, 0);
7625 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7629 * Ok we want to try and use the cluster allocator, so
7632 if (last_ptr && use_cluster) {
7633 struct btrfs_block_group_cache *used_block_group;
7634 unsigned long aligned_cluster;
7636 * the refill lock keeps out other
7637 * people trying to start a new cluster
7639 used_block_group = btrfs_lock_cluster(block_group,
7642 if (!used_block_group)
7643 goto refill_cluster;
7645 if (used_block_group != block_group &&
7646 (used_block_group->ro ||
7647 !block_group_bits(used_block_group, flags)))
7648 goto release_cluster;
7650 offset = btrfs_alloc_from_cluster(used_block_group,
7653 used_block_group->key.objectid,
7656 /* we have a block, we're done */
7657 spin_unlock(&last_ptr->refill_lock);
7658 trace_btrfs_reserve_extent_cluster(fs_info,
7660 search_start, num_bytes);
7661 if (used_block_group != block_group) {
7662 btrfs_release_block_group(block_group,
7664 block_group = used_block_group;
7669 WARN_ON(last_ptr->block_group != used_block_group);
7671 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7672 * set up a new clusters, so lets just skip it
7673 * and let the allocator find whatever block
7674 * it can find. If we reach this point, we
7675 * will have tried the cluster allocator
7676 * plenty of times and not have found
7677 * anything, so we are likely way too
7678 * fragmented for the clustering stuff to find
7681 * However, if the cluster is taken from the
7682 * current block group, release the cluster
7683 * first, so that we stand a better chance of
7684 * succeeding in the unclustered
7686 if (loop >= LOOP_NO_EMPTY_SIZE &&
7687 used_block_group != block_group) {
7688 spin_unlock(&last_ptr->refill_lock);
7689 btrfs_release_block_group(used_block_group,
7691 goto unclustered_alloc;
7695 * this cluster didn't work out, free it and
7698 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7700 if (used_block_group != block_group)
7701 btrfs_release_block_group(used_block_group,
7704 if (loop >= LOOP_NO_EMPTY_SIZE) {
7705 spin_unlock(&last_ptr->refill_lock);
7706 goto unclustered_alloc;
7709 aligned_cluster = max_t(unsigned long,
7710 empty_cluster + empty_size,
7711 block_group->full_stripe_len);
7713 /* allocate a cluster in this block group */
7714 ret = btrfs_find_space_cluster(fs_info, block_group,
7715 last_ptr, search_start,
7720 * now pull our allocation out of this
7723 offset = btrfs_alloc_from_cluster(block_group,
7729 /* we found one, proceed */
7730 spin_unlock(&last_ptr->refill_lock);
7731 trace_btrfs_reserve_extent_cluster(fs_info,
7732 block_group, search_start,
7736 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7737 && !failed_cluster_refill) {
7738 spin_unlock(&last_ptr->refill_lock);
7740 failed_cluster_refill = true;
7741 wait_block_group_cache_progress(block_group,
7742 num_bytes + empty_cluster + empty_size);
7743 goto have_block_group;
7747 * at this point we either didn't find a cluster
7748 * or we weren't able to allocate a block from our
7749 * cluster. Free the cluster we've been trying
7750 * to use, and go to the next block group
7752 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7753 spin_unlock(&last_ptr->refill_lock);
7759 * We are doing an unclustered alloc, set the fragmented flag so
7760 * we don't bother trying to setup a cluster again until we get
7763 if (unlikely(last_ptr)) {
7764 spin_lock(&last_ptr->lock);
7765 last_ptr->fragmented = 1;
7766 spin_unlock(&last_ptr->lock);
7769 struct btrfs_free_space_ctl *ctl =
7770 block_group->free_space_ctl;
7772 spin_lock(&ctl->tree_lock);
7773 if (ctl->free_space <
7774 num_bytes + empty_cluster + empty_size) {
7775 if (ctl->free_space > max_extent_size)
7776 max_extent_size = ctl->free_space;
7777 spin_unlock(&ctl->tree_lock);
7780 spin_unlock(&ctl->tree_lock);
7783 offset = btrfs_find_space_for_alloc(block_group, search_start,
7784 num_bytes, empty_size,
7787 * If we didn't find a chunk, and we haven't failed on this
7788 * block group before, and this block group is in the middle of
7789 * caching and we are ok with waiting, then go ahead and wait
7790 * for progress to be made, and set failed_alloc to true.
7792 * If failed_alloc is true then we've already waited on this
7793 * block group once and should move on to the next block group.
7795 if (!offset && !failed_alloc && !cached &&
7796 loop > LOOP_CACHING_NOWAIT) {
7797 wait_block_group_cache_progress(block_group,
7798 num_bytes + empty_size);
7799 failed_alloc = true;
7800 goto have_block_group;
7801 } else if (!offset) {
7805 search_start = ALIGN(offset, fs_info->stripesize);
7807 /* move on to the next group */
7808 if (search_start + num_bytes >
7809 block_group->key.objectid + block_group->key.offset) {
7810 btrfs_add_free_space(block_group, offset, num_bytes);
7814 if (offset < search_start)
7815 btrfs_add_free_space(block_group, offset,
7816 search_start - offset);
7817 BUG_ON(offset > search_start);
7819 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7820 num_bytes, delalloc);
7821 if (ret == -EAGAIN) {
7822 btrfs_add_free_space(block_group, offset, num_bytes);
7825 btrfs_inc_block_group_reservations(block_group);
7827 /* we are all good, lets return */
7828 ins->objectid = search_start;
7829 ins->offset = num_bytes;
7831 trace_btrfs_reserve_extent(fs_info, block_group,
7832 search_start, num_bytes);
7833 btrfs_release_block_group(block_group, delalloc);
7836 failed_cluster_refill = false;
7837 failed_alloc = false;
7838 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7840 btrfs_release_block_group(block_group, delalloc);
7843 up_read(&space_info->groups_sem);
7845 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7846 && !orig_have_caching_bg)
7847 orig_have_caching_bg = true;
7849 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7852 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7856 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7857 * caching kthreads as we move along
7858 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7859 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7860 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7863 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7865 if (loop == LOOP_CACHING_NOWAIT) {
7867 * We want to skip the LOOP_CACHING_WAIT step if we
7868 * don't have any uncached bgs and we've already done a
7869 * full search through.
7871 if (orig_have_caching_bg || !full_search)
7872 loop = LOOP_CACHING_WAIT;
7874 loop = LOOP_ALLOC_CHUNK;
7879 if (loop == LOOP_ALLOC_CHUNK) {
7880 struct btrfs_trans_handle *trans;
7883 trans = current->journal_info;
7887 trans = btrfs_join_transaction(root);
7889 if (IS_ERR(trans)) {
7890 ret = PTR_ERR(trans);
7894 ret = do_chunk_alloc(trans, fs_info, flags,
7898 * If we can't allocate a new chunk we've already looped
7899 * through at least once, move on to the NO_EMPTY_SIZE
7903 loop = LOOP_NO_EMPTY_SIZE;
7906 * Do not bail out on ENOSPC since we
7907 * can do more things.
7909 if (ret < 0 && ret != -ENOSPC)
7910 btrfs_abort_transaction(trans, ret);
7914 btrfs_end_transaction(trans);
7919 if (loop == LOOP_NO_EMPTY_SIZE) {
7921 * Don't loop again if we already have no empty_size and
7924 if (empty_size == 0 &&
7925 empty_cluster == 0) {
7934 } else if (!ins->objectid) {
7936 } else if (ins->objectid) {
7937 if (!use_cluster && last_ptr) {
7938 spin_lock(&last_ptr->lock);
7939 last_ptr->window_start = ins->objectid;
7940 spin_unlock(&last_ptr->lock);
7945 if (ret == -ENOSPC) {
7946 spin_lock(&space_info->lock);
7947 space_info->max_extent_size = max_extent_size;
7948 spin_unlock(&space_info->lock);
7949 ins->offset = max_extent_size;
7954 static void dump_space_info(struct btrfs_fs_info *fs_info,
7955 struct btrfs_space_info *info, u64 bytes,
7956 int dump_block_groups)
7958 struct btrfs_block_group_cache *cache;
7961 spin_lock(&info->lock);
7962 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7964 info->total_bytes - btrfs_space_info_used(info, true),
7965 info->full ? "" : "not ");
7967 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7968 info->total_bytes, info->bytes_used, info->bytes_pinned,
7969 info->bytes_reserved, info->bytes_may_use,
7970 info->bytes_readonly);
7971 spin_unlock(&info->lock);
7973 if (!dump_block_groups)
7976 down_read(&info->groups_sem);
7978 list_for_each_entry(cache, &info->block_groups[index], list) {
7979 spin_lock(&cache->lock);
7981 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7982 cache->key.objectid, cache->key.offset,
7983 btrfs_block_group_used(&cache->item), cache->pinned,
7984 cache->reserved, cache->ro ? "[readonly]" : "");
7985 btrfs_dump_free_space(cache, bytes);
7986 spin_unlock(&cache->lock);
7988 if (++index < BTRFS_NR_RAID_TYPES)
7990 up_read(&info->groups_sem);
7994 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7995 * hole that is at least as big as @num_bytes.
7997 * @root - The root that will contain this extent
7999 * @ram_bytes - The amount of space in ram that @num_bytes take. This
8000 * is used for accounting purposes. This value differs
8001 * from @num_bytes only in the case of compressed extents.
8003 * @num_bytes - Number of bytes to allocate on-disk.
8005 * @min_alloc_size - Indicates the minimum amount of space that the
8006 * allocator should try to satisfy. In some cases
8007 * @num_bytes may be larger than what is required and if
8008 * the filesystem is fragmented then allocation fails.
8009 * However, the presence of @min_alloc_size gives a
8010 * chance to try and satisfy the smaller allocation.
8012 * @empty_size - A hint that you plan on doing more COW. This is the
8013 * size in bytes the allocator should try to find free
8014 * next to the block it returns. This is just a hint and
8015 * may be ignored by the allocator.
8017 * @hint_byte - Hint to the allocator to start searching above the byte
8018 * address passed. It might be ignored.
8020 * @ins - This key is modified to record the found hole. It will
8021 * have the following values:
8022 * ins->objectid == start position
8023 * ins->flags = BTRFS_EXTENT_ITEM_KEY
8024 * ins->offset == the size of the hole.
8026 * @is_data - Boolean flag indicating whether an extent is
8027 * allocated for data (true) or metadata (false)
8029 * @delalloc - Boolean flag indicating whether this allocation is for
8030 * delalloc or not. If 'true' data_rwsem of block groups
8031 * is going to be acquired.
8034 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8035 * case -ENOSPC is returned then @ins->offset will contain the size of the
8036 * largest available hole the allocator managed to find.
8038 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8039 u64 num_bytes, u64 min_alloc_size,
8040 u64 empty_size, u64 hint_byte,
8041 struct btrfs_key *ins, int is_data, int delalloc)
8043 struct btrfs_fs_info *fs_info = root->fs_info;
8044 bool final_tried = num_bytes == min_alloc_size;
8048 flags = get_alloc_profile_by_root(root, is_data);
8050 WARN_ON(num_bytes < fs_info->sectorsize);
8051 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8052 hint_byte, ins, flags, delalloc);
8053 if (!ret && !is_data) {
8054 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8055 } else if (ret == -ENOSPC) {
8056 if (!final_tried && ins->offset) {
8057 num_bytes = min(num_bytes >> 1, ins->offset);
8058 num_bytes = round_down(num_bytes,
8059 fs_info->sectorsize);
8060 num_bytes = max(num_bytes, min_alloc_size);
8061 ram_bytes = num_bytes;
8062 if (num_bytes == min_alloc_size)
8065 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8066 struct btrfs_space_info *sinfo;
8068 sinfo = __find_space_info(fs_info, flags);
8070 "allocation failed flags %llu, wanted %llu",
8073 dump_space_info(fs_info, sinfo, num_bytes, 1);
8080 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8082 int pin, int delalloc)
8084 struct btrfs_block_group_cache *cache;
8087 cache = btrfs_lookup_block_group(fs_info, start);
8089 btrfs_err(fs_info, "Unable to find block group for %llu",
8095 pin_down_extent(fs_info, cache, start, len, 1);
8097 if (btrfs_test_opt(fs_info, DISCARD))
8098 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8099 btrfs_add_free_space(cache, start, len);
8100 btrfs_free_reserved_bytes(cache, len, delalloc);
8101 trace_btrfs_reserved_extent_free(fs_info, start, len);
8104 btrfs_put_block_group(cache);
8108 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8109 u64 start, u64 len, int delalloc)
8111 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8114 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8117 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8120 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8121 struct btrfs_fs_info *fs_info,
8122 u64 parent, u64 root_objectid,
8123 u64 flags, u64 owner, u64 offset,
8124 struct btrfs_key *ins, int ref_mod)
8127 struct btrfs_extent_item *extent_item;
8128 struct btrfs_extent_inline_ref *iref;
8129 struct btrfs_path *path;
8130 struct extent_buffer *leaf;
8135 type = BTRFS_SHARED_DATA_REF_KEY;
8137 type = BTRFS_EXTENT_DATA_REF_KEY;
8139 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8141 path = btrfs_alloc_path();
8145 path->leave_spinning = 1;
8146 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8149 btrfs_free_path(path);
8153 leaf = path->nodes[0];
8154 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8155 struct btrfs_extent_item);
8156 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8157 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8158 btrfs_set_extent_flags(leaf, extent_item,
8159 flags | BTRFS_EXTENT_FLAG_DATA);
8161 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8162 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8164 struct btrfs_shared_data_ref *ref;
8165 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8166 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8167 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8169 struct btrfs_extent_data_ref *ref;
8170 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8171 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8172 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8173 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8174 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8177 btrfs_mark_buffer_dirty(path->nodes[0]);
8178 btrfs_free_path(path);
8180 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8185 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8186 if (ret) { /* -ENOENT, logic error */
8187 btrfs_err(fs_info, "update block group failed for %llu %llu",
8188 ins->objectid, ins->offset);
8191 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8195 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8196 struct btrfs_fs_info *fs_info,
8197 u64 parent, u64 root_objectid,
8198 u64 flags, struct btrfs_disk_key *key,
8199 int level, struct btrfs_key *ins)
8202 struct btrfs_extent_item *extent_item;
8203 struct btrfs_tree_block_info *block_info;
8204 struct btrfs_extent_inline_ref *iref;
8205 struct btrfs_path *path;
8206 struct extent_buffer *leaf;
8207 u32 size = sizeof(*extent_item) + sizeof(*iref);
8208 u64 num_bytes = ins->offset;
8209 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8211 if (!skinny_metadata)
8212 size += sizeof(*block_info);
8214 path = btrfs_alloc_path();
8216 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8221 path->leave_spinning = 1;
8222 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8225 btrfs_free_path(path);
8226 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8231 leaf = path->nodes[0];
8232 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8233 struct btrfs_extent_item);
8234 btrfs_set_extent_refs(leaf, extent_item, 1);
8235 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8236 btrfs_set_extent_flags(leaf, extent_item,
8237 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8239 if (skinny_metadata) {
8240 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8241 num_bytes = fs_info->nodesize;
8243 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8244 btrfs_set_tree_block_key(leaf, block_info, key);
8245 btrfs_set_tree_block_level(leaf, block_info, level);
8246 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8250 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8251 btrfs_set_extent_inline_ref_type(leaf, iref,
8252 BTRFS_SHARED_BLOCK_REF_KEY);
8253 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8255 btrfs_set_extent_inline_ref_type(leaf, iref,
8256 BTRFS_TREE_BLOCK_REF_KEY);
8257 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8260 btrfs_mark_buffer_dirty(leaf);
8261 btrfs_free_path(path);
8263 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8268 ret = update_block_group(trans, fs_info, ins->objectid,
8269 fs_info->nodesize, 1);
8270 if (ret) { /* -ENOENT, logic error */
8271 btrfs_err(fs_info, "update block group failed for %llu %llu",
8272 ins->objectid, ins->offset);
8276 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8281 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8282 struct btrfs_root *root, u64 owner,
8283 u64 offset, u64 ram_bytes,
8284 struct btrfs_key *ins)
8286 struct btrfs_fs_info *fs_info = root->fs_info;
8289 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8291 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8292 root->root_key.objectid, owner, offset,
8293 BTRFS_ADD_DELAYED_EXTENT);
8295 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8297 root->root_key.objectid, owner,
8299 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8304 * this is used by the tree logging recovery code. It records that
8305 * an extent has been allocated and makes sure to clear the free
8306 * space cache bits as well
8308 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8309 struct btrfs_fs_info *fs_info,
8310 u64 root_objectid, u64 owner, u64 offset,
8311 struct btrfs_key *ins)
8314 struct btrfs_block_group_cache *block_group;
8315 struct btrfs_space_info *space_info;
8318 * Mixed block groups will exclude before processing the log so we only
8319 * need to do the exclude dance if this fs isn't mixed.
8321 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8322 ret = __exclude_logged_extent(fs_info, ins->objectid,
8328 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8332 space_info = block_group->space_info;
8333 spin_lock(&space_info->lock);
8334 spin_lock(&block_group->lock);
8335 space_info->bytes_reserved += ins->offset;
8336 block_group->reserved += ins->offset;
8337 spin_unlock(&block_group->lock);
8338 spin_unlock(&space_info->lock);
8340 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8341 0, owner, offset, ins, 1);
8342 btrfs_put_block_group(block_group);
8346 static struct extent_buffer *
8347 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8348 u64 bytenr, int level)
8350 struct btrfs_fs_info *fs_info = root->fs_info;
8351 struct extent_buffer *buf;
8353 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8357 btrfs_set_header_generation(buf, trans->transid);
8358 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8359 btrfs_tree_lock(buf);
8360 clean_tree_block(fs_info, buf);
8361 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8363 btrfs_set_lock_blocking(buf);
8364 set_extent_buffer_uptodate(buf);
8366 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8367 buf->log_index = root->log_transid % 2;
8369 * we allow two log transactions at a time, use different
8370 * EXENT bit to differentiate dirty pages.
8372 if (buf->log_index == 0)
8373 set_extent_dirty(&root->dirty_log_pages, buf->start,
8374 buf->start + buf->len - 1, GFP_NOFS);
8376 set_extent_new(&root->dirty_log_pages, buf->start,
8377 buf->start + buf->len - 1);
8379 buf->log_index = -1;
8380 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8381 buf->start + buf->len - 1, GFP_NOFS);
8383 trans->dirty = true;
8384 /* this returns a buffer locked for blocking */
8388 static struct btrfs_block_rsv *
8389 use_block_rsv(struct btrfs_trans_handle *trans,
8390 struct btrfs_root *root, u32 blocksize)
8392 struct btrfs_fs_info *fs_info = root->fs_info;
8393 struct btrfs_block_rsv *block_rsv;
8394 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8396 bool global_updated = false;
8398 block_rsv = get_block_rsv(trans, root);
8400 if (unlikely(block_rsv->size == 0))
8403 ret = block_rsv_use_bytes(block_rsv, blocksize);
8407 if (block_rsv->failfast)
8408 return ERR_PTR(ret);
8410 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8411 global_updated = true;
8412 update_global_block_rsv(fs_info);
8416 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8417 static DEFINE_RATELIMIT_STATE(_rs,
8418 DEFAULT_RATELIMIT_INTERVAL * 10,
8419 /*DEFAULT_RATELIMIT_BURST*/ 1);
8420 if (__ratelimit(&_rs))
8422 "BTRFS: block rsv returned %d\n", ret);
8425 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8426 BTRFS_RESERVE_NO_FLUSH);
8430 * If we couldn't reserve metadata bytes try and use some from
8431 * the global reserve if its space type is the same as the global
8434 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8435 block_rsv->space_info == global_rsv->space_info) {
8436 ret = block_rsv_use_bytes(global_rsv, blocksize);
8440 return ERR_PTR(ret);
8443 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8444 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8446 block_rsv_add_bytes(block_rsv, blocksize, 0);
8447 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8451 * finds a free extent and does all the dirty work required for allocation
8452 * returns the tree buffer or an ERR_PTR on error.
8454 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8455 struct btrfs_root *root,
8456 u64 parent, u64 root_objectid,
8457 const struct btrfs_disk_key *key,
8458 int level, u64 hint,
8461 struct btrfs_fs_info *fs_info = root->fs_info;
8462 struct btrfs_key ins;
8463 struct btrfs_block_rsv *block_rsv;
8464 struct extent_buffer *buf;
8465 struct btrfs_delayed_extent_op *extent_op;
8468 u32 blocksize = fs_info->nodesize;
8469 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8471 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8472 if (btrfs_is_testing(fs_info)) {
8473 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8476 root->alloc_bytenr += blocksize;
8481 block_rsv = use_block_rsv(trans, root, blocksize);
8482 if (IS_ERR(block_rsv))
8483 return ERR_CAST(block_rsv);
8485 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8486 empty_size, hint, &ins, 0, 0);
8490 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8493 goto out_free_reserved;
8496 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8498 parent = ins.objectid;
8499 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8503 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8504 extent_op = btrfs_alloc_delayed_extent_op();
8510 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8512 memset(&extent_op->key, 0, sizeof(extent_op->key));
8513 extent_op->flags_to_set = flags;
8514 extent_op->update_key = skinny_metadata ? false : true;
8515 extent_op->update_flags = true;
8516 extent_op->is_data = false;
8517 extent_op->level = level;
8519 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8520 root_objectid, level, 0,
8521 BTRFS_ADD_DELAYED_EXTENT);
8522 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8524 root_objectid, level,
8525 BTRFS_ADD_DELAYED_EXTENT,
8526 extent_op, NULL, NULL);
8528 goto out_free_delayed;
8533 btrfs_free_delayed_extent_op(extent_op);
8535 free_extent_buffer(buf);
8537 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8539 unuse_block_rsv(fs_info, block_rsv, blocksize);
8540 return ERR_PTR(ret);
8543 struct walk_control {
8544 u64 refs[BTRFS_MAX_LEVEL];
8545 u64 flags[BTRFS_MAX_LEVEL];
8546 struct btrfs_key update_progress;
8557 #define DROP_REFERENCE 1
8558 #define UPDATE_BACKREF 2
8560 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8561 struct btrfs_root *root,
8562 struct walk_control *wc,
8563 struct btrfs_path *path)
8565 struct btrfs_fs_info *fs_info = root->fs_info;
8571 struct btrfs_key key;
8572 struct extent_buffer *eb;
8577 if (path->slots[wc->level] < wc->reada_slot) {
8578 wc->reada_count = wc->reada_count * 2 / 3;
8579 wc->reada_count = max(wc->reada_count, 2);
8581 wc->reada_count = wc->reada_count * 3 / 2;
8582 wc->reada_count = min_t(int, wc->reada_count,
8583 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8586 eb = path->nodes[wc->level];
8587 nritems = btrfs_header_nritems(eb);
8589 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8590 if (nread >= wc->reada_count)
8594 bytenr = btrfs_node_blockptr(eb, slot);
8595 generation = btrfs_node_ptr_generation(eb, slot);
8597 if (slot == path->slots[wc->level])
8600 if (wc->stage == UPDATE_BACKREF &&
8601 generation <= root->root_key.offset)
8604 /* We don't lock the tree block, it's OK to be racy here */
8605 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8606 wc->level - 1, 1, &refs,
8608 /* We don't care about errors in readahead. */
8613 if (wc->stage == DROP_REFERENCE) {
8617 if (wc->level == 1 &&
8618 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8620 if (!wc->update_ref ||
8621 generation <= root->root_key.offset)
8623 btrfs_node_key_to_cpu(eb, &key, slot);
8624 ret = btrfs_comp_cpu_keys(&key,
8625 &wc->update_progress);
8629 if (wc->level == 1 &&
8630 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8634 readahead_tree_block(fs_info, bytenr);
8637 wc->reada_slot = slot;
8641 * helper to process tree block while walking down the tree.
8643 * when wc->stage == UPDATE_BACKREF, this function updates
8644 * back refs for pointers in the block.
8646 * NOTE: return value 1 means we should stop walking down.
8648 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8649 struct btrfs_root *root,
8650 struct btrfs_path *path,
8651 struct walk_control *wc, int lookup_info)
8653 struct btrfs_fs_info *fs_info = root->fs_info;
8654 int level = wc->level;
8655 struct extent_buffer *eb = path->nodes[level];
8656 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8659 if (wc->stage == UPDATE_BACKREF &&
8660 btrfs_header_owner(eb) != root->root_key.objectid)
8664 * when reference count of tree block is 1, it won't increase
8665 * again. once full backref flag is set, we never clear it.
8668 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8669 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8670 BUG_ON(!path->locks[level]);
8671 ret = btrfs_lookup_extent_info(trans, fs_info,
8672 eb->start, level, 1,
8675 BUG_ON(ret == -ENOMEM);
8678 BUG_ON(wc->refs[level] == 0);
8681 if (wc->stage == DROP_REFERENCE) {
8682 if (wc->refs[level] > 1)
8685 if (path->locks[level] && !wc->keep_locks) {
8686 btrfs_tree_unlock_rw(eb, path->locks[level]);
8687 path->locks[level] = 0;
8692 /* wc->stage == UPDATE_BACKREF */
8693 if (!(wc->flags[level] & flag)) {
8694 BUG_ON(!path->locks[level]);
8695 ret = btrfs_inc_ref(trans, root, eb, 1);
8696 BUG_ON(ret); /* -ENOMEM */
8697 ret = btrfs_dec_ref(trans, root, eb, 0);
8698 BUG_ON(ret); /* -ENOMEM */
8699 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8701 btrfs_header_level(eb), 0);
8702 BUG_ON(ret); /* -ENOMEM */
8703 wc->flags[level] |= flag;
8707 * the block is shared by multiple trees, so it's not good to
8708 * keep the tree lock
8710 if (path->locks[level] && level > 0) {
8711 btrfs_tree_unlock_rw(eb, path->locks[level]);
8712 path->locks[level] = 0;
8718 * helper to process tree block pointer.
8720 * when wc->stage == DROP_REFERENCE, this function checks
8721 * reference count of the block pointed to. if the block
8722 * is shared and we need update back refs for the subtree
8723 * rooted at the block, this function changes wc->stage to
8724 * UPDATE_BACKREF. if the block is shared and there is no
8725 * need to update back, this function drops the reference
8728 * NOTE: return value 1 means we should stop walking down.
8730 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8731 struct btrfs_root *root,
8732 struct btrfs_path *path,
8733 struct walk_control *wc, int *lookup_info)
8735 struct btrfs_fs_info *fs_info = root->fs_info;
8740 struct btrfs_key key;
8741 struct btrfs_key first_key;
8742 struct extent_buffer *next;
8743 int level = wc->level;
8746 bool need_account = false;
8748 generation = btrfs_node_ptr_generation(path->nodes[level],
8749 path->slots[level]);
8751 * if the lower level block was created before the snapshot
8752 * was created, we know there is no need to update back refs
8755 if (wc->stage == UPDATE_BACKREF &&
8756 generation <= root->root_key.offset) {
8761 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8762 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8763 path->slots[level]);
8764 blocksize = fs_info->nodesize;
8766 next = find_extent_buffer(fs_info, bytenr);
8768 next = btrfs_find_create_tree_block(fs_info, bytenr);
8770 return PTR_ERR(next);
8772 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8776 btrfs_tree_lock(next);
8777 btrfs_set_lock_blocking(next);
8779 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8780 &wc->refs[level - 1],
8781 &wc->flags[level - 1]);
8785 if (unlikely(wc->refs[level - 1] == 0)) {
8786 btrfs_err(fs_info, "Missing references.");
8792 if (wc->stage == DROP_REFERENCE) {
8793 if (wc->refs[level - 1] > 1) {
8794 need_account = true;
8796 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8799 if (!wc->update_ref ||
8800 generation <= root->root_key.offset)
8803 btrfs_node_key_to_cpu(path->nodes[level], &key,
8804 path->slots[level]);
8805 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8809 wc->stage = UPDATE_BACKREF;
8810 wc->shared_level = level - 1;
8814 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8818 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8819 btrfs_tree_unlock(next);
8820 free_extent_buffer(next);
8826 if (reada && level == 1)
8827 reada_walk_down(trans, root, wc, path);
8828 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8831 return PTR_ERR(next);
8832 } else if (!extent_buffer_uptodate(next)) {
8833 free_extent_buffer(next);
8836 btrfs_tree_lock(next);
8837 btrfs_set_lock_blocking(next);
8841 ASSERT(level == btrfs_header_level(next));
8842 if (level != btrfs_header_level(next)) {
8843 btrfs_err(root->fs_info, "mismatched level");
8847 path->nodes[level] = next;
8848 path->slots[level] = 0;
8849 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8855 wc->refs[level - 1] = 0;
8856 wc->flags[level - 1] = 0;
8857 if (wc->stage == DROP_REFERENCE) {
8858 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8859 parent = path->nodes[level]->start;
8861 ASSERT(root->root_key.objectid ==
8862 btrfs_header_owner(path->nodes[level]));
8863 if (root->root_key.objectid !=
8864 btrfs_header_owner(path->nodes[level])) {
8865 btrfs_err(root->fs_info,
8866 "mismatched block owner");
8874 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8875 generation, level - 1);
8877 btrfs_err_rl(fs_info,
8878 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8882 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8883 parent, root->root_key.objectid,
8893 btrfs_tree_unlock(next);
8894 free_extent_buffer(next);
8900 * helper to process tree block while walking up the tree.
8902 * when wc->stage == DROP_REFERENCE, this function drops
8903 * reference count on the block.
8905 * when wc->stage == UPDATE_BACKREF, this function changes
8906 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8907 * to UPDATE_BACKREF previously while processing the block.
8909 * NOTE: return value 1 means we should stop walking up.
8911 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8912 struct btrfs_root *root,
8913 struct btrfs_path *path,
8914 struct walk_control *wc)
8916 struct btrfs_fs_info *fs_info = root->fs_info;
8918 int level = wc->level;
8919 struct extent_buffer *eb = path->nodes[level];
8922 if (wc->stage == UPDATE_BACKREF) {
8923 BUG_ON(wc->shared_level < level);
8924 if (level < wc->shared_level)
8927 ret = find_next_key(path, level + 1, &wc->update_progress);
8931 wc->stage = DROP_REFERENCE;
8932 wc->shared_level = -1;
8933 path->slots[level] = 0;
8936 * check reference count again if the block isn't locked.
8937 * we should start walking down the tree again if reference
8940 if (!path->locks[level]) {
8942 btrfs_tree_lock(eb);
8943 btrfs_set_lock_blocking(eb);
8944 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8946 ret = btrfs_lookup_extent_info(trans, fs_info,
8947 eb->start, level, 1,
8951 btrfs_tree_unlock_rw(eb, path->locks[level]);
8952 path->locks[level] = 0;
8955 BUG_ON(wc->refs[level] == 0);
8956 if (wc->refs[level] == 1) {
8957 btrfs_tree_unlock_rw(eb, path->locks[level]);
8958 path->locks[level] = 0;
8964 /* wc->stage == DROP_REFERENCE */
8965 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8967 if (wc->refs[level] == 1) {
8969 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8970 ret = btrfs_dec_ref(trans, root, eb, 1);
8972 ret = btrfs_dec_ref(trans, root, eb, 0);
8973 BUG_ON(ret); /* -ENOMEM */
8974 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8976 btrfs_err_rl(fs_info,
8977 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8981 /* make block locked assertion in clean_tree_block happy */
8982 if (!path->locks[level] &&
8983 btrfs_header_generation(eb) == trans->transid) {
8984 btrfs_tree_lock(eb);
8985 btrfs_set_lock_blocking(eb);
8986 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8988 clean_tree_block(fs_info, eb);
8991 if (eb == root->node) {
8992 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8995 BUG_ON(root->root_key.objectid !=
8996 btrfs_header_owner(eb));
8998 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8999 parent = path->nodes[level + 1]->start;
9001 BUG_ON(root->root_key.objectid !=
9002 btrfs_header_owner(path->nodes[level + 1]));
9005 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9007 wc->refs[level] = 0;
9008 wc->flags[level] = 0;
9012 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9013 struct btrfs_root *root,
9014 struct btrfs_path *path,
9015 struct walk_control *wc)
9017 int level = wc->level;
9018 int lookup_info = 1;
9021 while (level >= 0) {
9022 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9029 if (path->slots[level] >=
9030 btrfs_header_nritems(path->nodes[level]))
9033 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9035 path->slots[level]++;
9044 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9045 struct btrfs_root *root,
9046 struct btrfs_path *path,
9047 struct walk_control *wc, int max_level)
9049 int level = wc->level;
9052 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9053 while (level < max_level && path->nodes[level]) {
9055 if (path->slots[level] + 1 <
9056 btrfs_header_nritems(path->nodes[level])) {
9057 path->slots[level]++;
9060 ret = walk_up_proc(trans, root, path, wc);
9064 if (path->locks[level]) {
9065 btrfs_tree_unlock_rw(path->nodes[level],
9066 path->locks[level]);
9067 path->locks[level] = 0;
9069 free_extent_buffer(path->nodes[level]);
9070 path->nodes[level] = NULL;
9078 * drop a subvolume tree.
9080 * this function traverses the tree freeing any blocks that only
9081 * referenced by the tree.
9083 * when a shared tree block is found. this function decreases its
9084 * reference count by one. if update_ref is true, this function
9085 * also make sure backrefs for the shared block and all lower level
9086 * blocks are properly updated.
9088 * If called with for_reloc == 0, may exit early with -EAGAIN
9090 int btrfs_drop_snapshot(struct btrfs_root *root,
9091 struct btrfs_block_rsv *block_rsv, int update_ref,
9094 struct btrfs_fs_info *fs_info = root->fs_info;
9095 struct btrfs_path *path;
9096 struct btrfs_trans_handle *trans;
9097 struct btrfs_root *tree_root = fs_info->tree_root;
9098 struct btrfs_root_item *root_item = &root->root_item;
9099 struct walk_control *wc;
9100 struct btrfs_key key;
9104 bool root_dropped = false;
9106 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9108 path = btrfs_alloc_path();
9114 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9116 btrfs_free_path(path);
9121 trans = btrfs_start_transaction(tree_root, 0);
9122 if (IS_ERR(trans)) {
9123 err = PTR_ERR(trans);
9128 trans->block_rsv = block_rsv;
9130 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9131 level = btrfs_header_level(root->node);
9132 path->nodes[level] = btrfs_lock_root_node(root);
9133 btrfs_set_lock_blocking(path->nodes[level]);
9134 path->slots[level] = 0;
9135 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9136 memset(&wc->update_progress, 0,
9137 sizeof(wc->update_progress));
9139 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9140 memcpy(&wc->update_progress, &key,
9141 sizeof(wc->update_progress));
9143 level = root_item->drop_level;
9145 path->lowest_level = level;
9146 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9147 path->lowest_level = 0;
9155 * unlock our path, this is safe because only this
9156 * function is allowed to delete this snapshot
9158 btrfs_unlock_up_safe(path, 0);
9160 level = btrfs_header_level(root->node);
9162 btrfs_tree_lock(path->nodes[level]);
9163 btrfs_set_lock_blocking(path->nodes[level]);
9164 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9166 ret = btrfs_lookup_extent_info(trans, fs_info,
9167 path->nodes[level]->start,
9168 level, 1, &wc->refs[level],
9174 BUG_ON(wc->refs[level] == 0);
9176 if (level == root_item->drop_level)
9179 btrfs_tree_unlock(path->nodes[level]);
9180 path->locks[level] = 0;
9181 WARN_ON(wc->refs[level] != 1);
9187 wc->shared_level = -1;
9188 wc->stage = DROP_REFERENCE;
9189 wc->update_ref = update_ref;
9191 wc->for_reloc = for_reloc;
9192 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9196 ret = walk_down_tree(trans, root, path, wc);
9202 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9209 BUG_ON(wc->stage != DROP_REFERENCE);
9213 if (wc->stage == DROP_REFERENCE) {
9215 btrfs_node_key(path->nodes[level],
9216 &root_item->drop_progress,
9217 path->slots[level]);
9218 root_item->drop_level = level;
9221 BUG_ON(wc->level == 0);
9222 if (btrfs_should_end_transaction(trans) ||
9223 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9224 ret = btrfs_update_root(trans, tree_root,
9228 btrfs_abort_transaction(trans, ret);
9233 btrfs_end_transaction_throttle(trans);
9234 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9235 btrfs_debug(fs_info,
9236 "drop snapshot early exit");
9241 trans = btrfs_start_transaction(tree_root, 0);
9242 if (IS_ERR(trans)) {
9243 err = PTR_ERR(trans);
9247 trans->block_rsv = block_rsv;
9250 btrfs_release_path(path);
9254 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9256 btrfs_abort_transaction(trans, ret);
9261 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9262 ret = btrfs_find_root(tree_root, &root->root_key, path,
9265 btrfs_abort_transaction(trans, ret);
9268 } else if (ret > 0) {
9269 /* if we fail to delete the orphan item this time
9270 * around, it'll get picked up the next time.
9272 * The most common failure here is just -ENOENT.
9274 btrfs_del_orphan_item(trans, tree_root,
9275 root->root_key.objectid);
9279 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9280 btrfs_add_dropped_root(trans, root);
9282 free_extent_buffer(root->node);
9283 free_extent_buffer(root->commit_root);
9284 btrfs_put_fs_root(root);
9286 root_dropped = true;
9288 btrfs_end_transaction_throttle(trans);
9291 btrfs_free_path(path);
9294 * So if we need to stop dropping the snapshot for whatever reason we
9295 * need to make sure to add it back to the dead root list so that we
9296 * keep trying to do the work later. This also cleans up roots if we
9297 * don't have it in the radix (like when we recover after a power fail
9298 * or unmount) so we don't leak memory.
9300 if (!for_reloc && !root_dropped)
9301 btrfs_add_dead_root(root);
9302 if (err && err != -EAGAIN)
9303 btrfs_handle_fs_error(fs_info, err, NULL);
9308 * drop subtree rooted at tree block 'node'.
9310 * NOTE: this function will unlock and release tree block 'node'
9311 * only used by relocation code
9313 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9314 struct btrfs_root *root,
9315 struct extent_buffer *node,
9316 struct extent_buffer *parent)
9318 struct btrfs_fs_info *fs_info = root->fs_info;
9319 struct btrfs_path *path;
9320 struct walk_control *wc;
9326 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9328 path = btrfs_alloc_path();
9332 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9334 btrfs_free_path(path);
9338 btrfs_assert_tree_locked(parent);
9339 parent_level = btrfs_header_level(parent);
9340 extent_buffer_get(parent);
9341 path->nodes[parent_level] = parent;
9342 path->slots[parent_level] = btrfs_header_nritems(parent);
9344 btrfs_assert_tree_locked(node);
9345 level = btrfs_header_level(node);
9346 path->nodes[level] = node;
9347 path->slots[level] = 0;
9348 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9350 wc->refs[parent_level] = 1;
9351 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9353 wc->shared_level = -1;
9354 wc->stage = DROP_REFERENCE;
9358 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9361 wret = walk_down_tree(trans, root, path, wc);
9367 wret = walk_up_tree(trans, root, path, wc, parent_level);
9375 btrfs_free_path(path);
9379 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9385 * if restripe for this chunk_type is on pick target profile and
9386 * return, otherwise do the usual balance
9388 stripped = get_restripe_target(fs_info, flags);
9390 return extended_to_chunk(stripped);
9392 num_devices = fs_info->fs_devices->rw_devices;
9394 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9395 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9396 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9398 if (num_devices == 1) {
9399 stripped |= BTRFS_BLOCK_GROUP_DUP;
9400 stripped = flags & ~stripped;
9402 /* turn raid0 into single device chunks */
9403 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9406 /* turn mirroring into duplication */
9407 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9408 BTRFS_BLOCK_GROUP_RAID10))
9409 return stripped | BTRFS_BLOCK_GROUP_DUP;
9411 /* they already had raid on here, just return */
9412 if (flags & stripped)
9415 stripped |= BTRFS_BLOCK_GROUP_DUP;
9416 stripped = flags & ~stripped;
9418 /* switch duplicated blocks with raid1 */
9419 if (flags & BTRFS_BLOCK_GROUP_DUP)
9420 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9422 /* this is drive concat, leave it alone */
9428 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9430 struct btrfs_space_info *sinfo = cache->space_info;
9432 u64 min_allocable_bytes;
9436 * We need some metadata space and system metadata space for
9437 * allocating chunks in some corner cases until we force to set
9438 * it to be readonly.
9441 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9443 min_allocable_bytes = SZ_1M;
9445 min_allocable_bytes = 0;
9447 spin_lock(&sinfo->lock);
9448 spin_lock(&cache->lock);
9456 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9457 cache->bytes_super - btrfs_block_group_used(&cache->item);
9459 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9460 min_allocable_bytes <= sinfo->total_bytes) {
9461 sinfo->bytes_readonly += num_bytes;
9463 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9467 spin_unlock(&cache->lock);
9468 spin_unlock(&sinfo->lock);
9472 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9473 struct btrfs_block_group_cache *cache)
9476 struct btrfs_trans_handle *trans;
9481 trans = btrfs_join_transaction(fs_info->extent_root);
9483 return PTR_ERR(trans);
9486 * we're not allowed to set block groups readonly after the dirty
9487 * block groups cache has started writing. If it already started,
9488 * back off and let this transaction commit
9490 mutex_lock(&fs_info->ro_block_group_mutex);
9491 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9492 u64 transid = trans->transid;
9494 mutex_unlock(&fs_info->ro_block_group_mutex);
9495 btrfs_end_transaction(trans);
9497 ret = btrfs_wait_for_commit(fs_info, transid);
9504 * if we are changing raid levels, try to allocate a corresponding
9505 * block group with the new raid level.
9507 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9508 if (alloc_flags != cache->flags) {
9509 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9512 * ENOSPC is allowed here, we may have enough space
9513 * already allocated at the new raid level to
9522 ret = inc_block_group_ro(cache, 0);
9525 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9526 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9530 ret = inc_block_group_ro(cache, 0);
9532 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9533 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9534 mutex_lock(&fs_info->chunk_mutex);
9535 check_system_chunk(trans, fs_info, alloc_flags);
9536 mutex_unlock(&fs_info->chunk_mutex);
9538 mutex_unlock(&fs_info->ro_block_group_mutex);
9540 btrfs_end_transaction(trans);
9544 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9545 struct btrfs_fs_info *fs_info, u64 type)
9547 u64 alloc_flags = get_alloc_profile(fs_info, type);
9549 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9553 * helper to account the unused space of all the readonly block group in the
9554 * space_info. takes mirrors into account.
9556 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9558 struct btrfs_block_group_cache *block_group;
9562 /* It's df, we don't care if it's racy */
9563 if (list_empty(&sinfo->ro_bgs))
9566 spin_lock(&sinfo->lock);
9567 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9568 spin_lock(&block_group->lock);
9570 if (!block_group->ro) {
9571 spin_unlock(&block_group->lock);
9575 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9576 BTRFS_BLOCK_GROUP_RAID10 |
9577 BTRFS_BLOCK_GROUP_DUP))
9582 free_bytes += (block_group->key.offset -
9583 btrfs_block_group_used(&block_group->item)) *
9586 spin_unlock(&block_group->lock);
9588 spin_unlock(&sinfo->lock);
9593 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9595 struct btrfs_space_info *sinfo = cache->space_info;
9600 spin_lock(&sinfo->lock);
9601 spin_lock(&cache->lock);
9603 num_bytes = cache->key.offset - cache->reserved -
9604 cache->pinned - cache->bytes_super -
9605 btrfs_block_group_used(&cache->item);
9606 sinfo->bytes_readonly -= num_bytes;
9607 list_del_init(&cache->ro_list);
9609 spin_unlock(&cache->lock);
9610 spin_unlock(&sinfo->lock);
9614 * checks to see if its even possible to relocate this block group.
9616 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9617 * ok to go ahead and try.
9619 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9621 struct btrfs_root *root = fs_info->extent_root;
9622 struct btrfs_block_group_cache *block_group;
9623 struct btrfs_space_info *space_info;
9624 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9625 struct btrfs_device *device;
9626 struct btrfs_trans_handle *trans;
9636 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9638 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9640 /* odd, couldn't find the block group, leave it alone */
9644 "can't find block group for bytenr %llu",
9649 min_free = btrfs_block_group_used(&block_group->item);
9651 /* no bytes used, we're good */
9655 space_info = block_group->space_info;
9656 spin_lock(&space_info->lock);
9658 full = space_info->full;
9661 * if this is the last block group we have in this space, we can't
9662 * relocate it unless we're able to allocate a new chunk below.
9664 * Otherwise, we need to make sure we have room in the space to handle
9665 * all of the extents from this block group. If we can, we're good
9667 if ((space_info->total_bytes != block_group->key.offset) &&
9668 (btrfs_space_info_used(space_info, false) + min_free <
9669 space_info->total_bytes)) {
9670 spin_unlock(&space_info->lock);
9673 spin_unlock(&space_info->lock);
9676 * ok we don't have enough space, but maybe we have free space on our
9677 * devices to allocate new chunks for relocation, so loop through our
9678 * alloc devices and guess if we have enough space. if this block
9679 * group is going to be restriped, run checks against the target
9680 * profile instead of the current one.
9692 target = get_restripe_target(fs_info, block_group->flags);
9694 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9697 * this is just a balance, so if we were marked as full
9698 * we know there is no space for a new chunk
9703 "no space to alloc new chunk for block group %llu",
9704 block_group->key.objectid);
9708 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9711 if (index == BTRFS_RAID_RAID10) {
9715 } else if (index == BTRFS_RAID_RAID1) {
9717 } else if (index == BTRFS_RAID_DUP) {
9720 } else if (index == BTRFS_RAID_RAID0) {
9721 dev_min = fs_devices->rw_devices;
9722 min_free = div64_u64(min_free, dev_min);
9725 /* We need to do this so that we can look at pending chunks */
9726 trans = btrfs_join_transaction(root);
9727 if (IS_ERR(trans)) {
9728 ret = PTR_ERR(trans);
9732 mutex_lock(&fs_info->chunk_mutex);
9733 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9737 * check to make sure we can actually find a chunk with enough
9738 * space to fit our block group in.
9740 if (device->total_bytes > device->bytes_used + min_free &&
9741 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9742 ret = find_free_dev_extent(trans, device, min_free,
9747 if (dev_nr >= dev_min)
9753 if (debug && ret == -1)
9755 "no space to allocate a new chunk for block group %llu",
9756 block_group->key.objectid);
9757 mutex_unlock(&fs_info->chunk_mutex);
9758 btrfs_end_transaction(trans);
9760 btrfs_put_block_group(block_group);
9764 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9765 struct btrfs_path *path,
9766 struct btrfs_key *key)
9768 struct btrfs_root *root = fs_info->extent_root;
9770 struct btrfs_key found_key;
9771 struct extent_buffer *leaf;
9774 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9779 slot = path->slots[0];
9780 leaf = path->nodes[0];
9781 if (slot >= btrfs_header_nritems(leaf)) {
9782 ret = btrfs_next_leaf(root, path);
9789 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9791 if (found_key.objectid >= key->objectid &&
9792 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9793 struct extent_map_tree *em_tree;
9794 struct extent_map *em;
9796 em_tree = &root->fs_info->mapping_tree.map_tree;
9797 read_lock(&em_tree->lock);
9798 em = lookup_extent_mapping(em_tree, found_key.objectid,
9800 read_unlock(&em_tree->lock);
9803 "logical %llu len %llu found bg but no related chunk",
9804 found_key.objectid, found_key.offset);
9809 free_extent_map(em);
9818 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9820 struct btrfs_block_group_cache *block_group;
9824 struct inode *inode;
9826 block_group = btrfs_lookup_first_block_group(info, last);
9827 while (block_group) {
9828 spin_lock(&block_group->lock);
9829 if (block_group->iref)
9831 spin_unlock(&block_group->lock);
9832 block_group = next_block_group(info, block_group);
9841 inode = block_group->inode;
9842 block_group->iref = 0;
9843 block_group->inode = NULL;
9844 spin_unlock(&block_group->lock);
9845 ASSERT(block_group->io_ctl.inode == NULL);
9847 last = block_group->key.objectid + block_group->key.offset;
9848 btrfs_put_block_group(block_group);
9853 * Must be called only after stopping all workers, since we could have block
9854 * group caching kthreads running, and therefore they could race with us if we
9855 * freed the block groups before stopping them.
9857 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9859 struct btrfs_block_group_cache *block_group;
9860 struct btrfs_space_info *space_info;
9861 struct btrfs_caching_control *caching_ctl;
9864 down_write(&info->commit_root_sem);
9865 while (!list_empty(&info->caching_block_groups)) {
9866 caching_ctl = list_entry(info->caching_block_groups.next,
9867 struct btrfs_caching_control, list);
9868 list_del(&caching_ctl->list);
9869 put_caching_control(caching_ctl);
9871 up_write(&info->commit_root_sem);
9873 spin_lock(&info->unused_bgs_lock);
9874 while (!list_empty(&info->unused_bgs)) {
9875 block_group = list_first_entry(&info->unused_bgs,
9876 struct btrfs_block_group_cache,
9878 list_del_init(&block_group->bg_list);
9879 btrfs_put_block_group(block_group);
9881 spin_unlock(&info->unused_bgs_lock);
9883 spin_lock(&info->block_group_cache_lock);
9884 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9885 block_group = rb_entry(n, struct btrfs_block_group_cache,
9887 rb_erase(&block_group->cache_node,
9888 &info->block_group_cache_tree);
9889 RB_CLEAR_NODE(&block_group->cache_node);
9890 spin_unlock(&info->block_group_cache_lock);
9892 down_write(&block_group->space_info->groups_sem);
9893 list_del(&block_group->list);
9894 up_write(&block_group->space_info->groups_sem);
9897 * We haven't cached this block group, which means we could
9898 * possibly have excluded extents on this block group.
9900 if (block_group->cached == BTRFS_CACHE_NO ||
9901 block_group->cached == BTRFS_CACHE_ERROR)
9902 free_excluded_extents(info, block_group);
9904 btrfs_remove_free_space_cache(block_group);
9905 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9906 ASSERT(list_empty(&block_group->dirty_list));
9907 ASSERT(list_empty(&block_group->io_list));
9908 ASSERT(list_empty(&block_group->bg_list));
9909 ASSERT(atomic_read(&block_group->count) == 1);
9910 btrfs_put_block_group(block_group);
9912 spin_lock(&info->block_group_cache_lock);
9914 spin_unlock(&info->block_group_cache_lock);
9916 /* now that all the block groups are freed, go through and
9917 * free all the space_info structs. This is only called during
9918 * the final stages of unmount, and so we know nobody is
9919 * using them. We call synchronize_rcu() once before we start,
9920 * just to be on the safe side.
9924 release_global_block_rsv(info);
9926 while (!list_empty(&info->space_info)) {
9929 space_info = list_entry(info->space_info.next,
9930 struct btrfs_space_info,
9934 * Do not hide this behind enospc_debug, this is actually
9935 * important and indicates a real bug if this happens.
9937 if (WARN_ON(space_info->bytes_pinned > 0 ||
9938 space_info->bytes_reserved > 0 ||
9939 space_info->bytes_may_use > 0))
9940 dump_space_info(info, space_info, 0, 0);
9941 list_del(&space_info->list);
9942 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9943 struct kobject *kobj;
9944 kobj = space_info->block_group_kobjs[i];
9945 space_info->block_group_kobjs[i] = NULL;
9951 kobject_del(&space_info->kobj);
9952 kobject_put(&space_info->kobj);
9957 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9958 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9960 struct btrfs_space_info *space_info;
9961 struct raid_kobject *rkobj;
9966 spin_lock(&fs_info->pending_raid_kobjs_lock);
9967 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9968 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9970 list_for_each_entry(rkobj, &list, list) {
9971 space_info = __find_space_info(fs_info, rkobj->flags);
9972 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9974 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9975 "%s", get_raid_name(index));
9977 kobject_put(&rkobj->kobj);
9983 "failed to add kobject for block cache, ignoring");
9986 static void link_block_group(struct btrfs_block_group_cache *cache)
9988 struct btrfs_space_info *space_info = cache->space_info;
9989 struct btrfs_fs_info *fs_info = cache->fs_info;
9990 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9993 down_write(&space_info->groups_sem);
9994 if (list_empty(&space_info->block_groups[index]))
9996 list_add_tail(&cache->list, &space_info->block_groups[index]);
9997 up_write(&space_info->groups_sem);
10000 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10002 btrfs_warn(cache->fs_info,
10003 "couldn't alloc memory for raid level kobject");
10006 rkobj->flags = cache->flags;
10007 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10009 spin_lock(&fs_info->pending_raid_kobjs_lock);
10010 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
10011 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10012 space_info->block_group_kobjs[index] = &rkobj->kobj;
10016 static struct btrfs_block_group_cache *
10017 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10018 u64 start, u64 size)
10020 struct btrfs_block_group_cache *cache;
10022 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10026 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10028 if (!cache->free_space_ctl) {
10033 cache->key.objectid = start;
10034 cache->key.offset = size;
10035 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10037 cache->fs_info = fs_info;
10038 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10039 set_free_space_tree_thresholds(cache);
10041 atomic_set(&cache->count, 1);
10042 spin_lock_init(&cache->lock);
10043 init_rwsem(&cache->data_rwsem);
10044 INIT_LIST_HEAD(&cache->list);
10045 INIT_LIST_HEAD(&cache->cluster_list);
10046 INIT_LIST_HEAD(&cache->bg_list);
10047 INIT_LIST_HEAD(&cache->ro_list);
10048 INIT_LIST_HEAD(&cache->dirty_list);
10049 INIT_LIST_HEAD(&cache->io_list);
10050 btrfs_init_free_space_ctl(cache);
10051 atomic_set(&cache->trimming, 0);
10052 mutex_init(&cache->free_space_lock);
10053 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10058 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10060 struct btrfs_path *path;
10062 struct btrfs_block_group_cache *cache;
10063 struct btrfs_space_info *space_info;
10064 struct btrfs_key key;
10065 struct btrfs_key found_key;
10066 struct extent_buffer *leaf;
10067 int need_clear = 0;
10072 feature = btrfs_super_incompat_flags(info->super_copy);
10073 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10077 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10078 path = btrfs_alloc_path();
10081 path->reada = READA_FORWARD;
10083 cache_gen = btrfs_super_cache_generation(info->super_copy);
10084 if (btrfs_test_opt(info, SPACE_CACHE) &&
10085 btrfs_super_generation(info->super_copy) != cache_gen)
10087 if (btrfs_test_opt(info, CLEAR_CACHE))
10091 ret = find_first_block_group(info, path, &key);
10097 leaf = path->nodes[0];
10098 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10100 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10109 * When we mount with old space cache, we need to
10110 * set BTRFS_DC_CLEAR and set dirty flag.
10112 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10113 * truncate the old free space cache inode and
10115 * b) Setting 'dirty flag' makes sure that we flush
10116 * the new space cache info onto disk.
10118 if (btrfs_test_opt(info, SPACE_CACHE))
10119 cache->disk_cache_state = BTRFS_DC_CLEAR;
10122 read_extent_buffer(leaf, &cache->item,
10123 btrfs_item_ptr_offset(leaf, path->slots[0]),
10124 sizeof(cache->item));
10125 cache->flags = btrfs_block_group_flags(&cache->item);
10127 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10128 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10130 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10131 cache->key.objectid);
10136 key.objectid = found_key.objectid + found_key.offset;
10137 btrfs_release_path(path);
10140 * We need to exclude the super stripes now so that the space
10141 * info has super bytes accounted for, otherwise we'll think
10142 * we have more space than we actually do.
10144 ret = exclude_super_stripes(info, cache);
10147 * We may have excluded something, so call this just in
10150 free_excluded_extents(info, cache);
10151 btrfs_put_block_group(cache);
10156 * check for two cases, either we are full, and therefore
10157 * don't need to bother with the caching work since we won't
10158 * find any space, or we are empty, and we can just add all
10159 * the space in and be done with it. This saves us _alot_ of
10160 * time, particularly in the full case.
10162 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10163 cache->last_byte_to_unpin = (u64)-1;
10164 cache->cached = BTRFS_CACHE_FINISHED;
10165 free_excluded_extents(info, cache);
10166 } else if (btrfs_block_group_used(&cache->item) == 0) {
10167 cache->last_byte_to_unpin = (u64)-1;
10168 cache->cached = BTRFS_CACHE_FINISHED;
10169 add_new_free_space(cache, info,
10170 found_key.objectid,
10171 found_key.objectid +
10173 free_excluded_extents(info, cache);
10176 ret = btrfs_add_block_group_cache(info, cache);
10178 btrfs_remove_free_space_cache(cache);
10179 btrfs_put_block_group(cache);
10183 trace_btrfs_add_block_group(info, cache, 0);
10184 update_space_info(info, cache->flags, found_key.offset,
10185 btrfs_block_group_used(&cache->item),
10186 cache->bytes_super, &space_info);
10188 cache->space_info = space_info;
10190 link_block_group(cache);
10192 set_avail_alloc_bits(info, cache->flags);
10193 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10194 inc_block_group_ro(cache, 1);
10195 } else if (btrfs_block_group_used(&cache->item) == 0) {
10196 spin_lock(&info->unused_bgs_lock);
10197 /* Should always be true but just in case. */
10198 if (list_empty(&cache->bg_list)) {
10199 btrfs_get_block_group(cache);
10200 list_add_tail(&cache->bg_list,
10201 &info->unused_bgs);
10203 spin_unlock(&info->unused_bgs_lock);
10207 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10208 if (!(get_alloc_profile(info, space_info->flags) &
10209 (BTRFS_BLOCK_GROUP_RAID10 |
10210 BTRFS_BLOCK_GROUP_RAID1 |
10211 BTRFS_BLOCK_GROUP_RAID5 |
10212 BTRFS_BLOCK_GROUP_RAID6 |
10213 BTRFS_BLOCK_GROUP_DUP)))
10216 * avoid allocating from un-mirrored block group if there are
10217 * mirrored block groups.
10219 list_for_each_entry(cache,
10220 &space_info->block_groups[BTRFS_RAID_RAID0],
10222 inc_block_group_ro(cache, 1);
10223 list_for_each_entry(cache,
10224 &space_info->block_groups[BTRFS_RAID_SINGLE],
10226 inc_block_group_ro(cache, 1);
10229 btrfs_add_raid_kobjects(info);
10230 init_global_block_rsv(info);
10233 btrfs_free_path(path);
10237 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10239 struct btrfs_fs_info *fs_info = trans->fs_info;
10240 struct btrfs_block_group_cache *block_group, *tmp;
10241 struct btrfs_root *extent_root = fs_info->extent_root;
10242 struct btrfs_block_group_item item;
10243 struct btrfs_key key;
10245 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10247 trans->can_flush_pending_bgs = false;
10248 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10252 spin_lock(&block_group->lock);
10253 memcpy(&item, &block_group->item, sizeof(item));
10254 memcpy(&key, &block_group->key, sizeof(key));
10255 spin_unlock(&block_group->lock);
10257 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10260 btrfs_abort_transaction(trans, ret);
10261 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10264 btrfs_abort_transaction(trans, ret);
10265 add_block_group_free_space(trans, fs_info, block_group);
10266 /* already aborted the transaction if it failed. */
10268 list_del_init(&block_group->bg_list);
10270 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10273 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10274 struct btrfs_fs_info *fs_info, u64 bytes_used,
10275 u64 type, u64 chunk_offset, u64 size)
10277 struct btrfs_block_group_cache *cache;
10280 btrfs_set_log_full_commit(fs_info, trans);
10282 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10286 btrfs_set_block_group_used(&cache->item, bytes_used);
10287 btrfs_set_block_group_chunk_objectid(&cache->item,
10288 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10289 btrfs_set_block_group_flags(&cache->item, type);
10291 cache->flags = type;
10292 cache->last_byte_to_unpin = (u64)-1;
10293 cache->cached = BTRFS_CACHE_FINISHED;
10294 cache->needs_free_space = 1;
10295 ret = exclude_super_stripes(fs_info, cache);
10298 * We may have excluded something, so call this just in
10301 free_excluded_extents(fs_info, cache);
10302 btrfs_put_block_group(cache);
10306 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10308 free_excluded_extents(fs_info, cache);
10310 #ifdef CONFIG_BTRFS_DEBUG
10311 if (btrfs_should_fragment_free_space(cache)) {
10312 u64 new_bytes_used = size - bytes_used;
10314 bytes_used += new_bytes_used >> 1;
10315 fragment_free_space(cache);
10319 * Ensure the corresponding space_info object is created and
10320 * assigned to our block group. We want our bg to be added to the rbtree
10321 * with its ->space_info set.
10323 cache->space_info = __find_space_info(fs_info, cache->flags);
10324 ASSERT(cache->space_info);
10326 ret = btrfs_add_block_group_cache(fs_info, cache);
10328 btrfs_remove_free_space_cache(cache);
10329 btrfs_put_block_group(cache);
10334 * Now that our block group has its ->space_info set and is inserted in
10335 * the rbtree, update the space info's counters.
10337 trace_btrfs_add_block_group(fs_info, cache, 1);
10338 update_space_info(fs_info, cache->flags, size, bytes_used,
10339 cache->bytes_super, &cache->space_info);
10340 update_global_block_rsv(fs_info);
10342 link_block_group(cache);
10344 list_add_tail(&cache->bg_list, &trans->new_bgs);
10346 set_avail_alloc_bits(fs_info, type);
10350 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10352 u64 extra_flags = chunk_to_extended(flags) &
10353 BTRFS_EXTENDED_PROFILE_MASK;
10355 write_seqlock(&fs_info->profiles_lock);
10356 if (flags & BTRFS_BLOCK_GROUP_DATA)
10357 fs_info->avail_data_alloc_bits &= ~extra_flags;
10358 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10359 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10360 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10361 fs_info->avail_system_alloc_bits &= ~extra_flags;
10362 write_sequnlock(&fs_info->profiles_lock);
10365 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10366 struct btrfs_fs_info *fs_info, u64 group_start,
10367 struct extent_map *em)
10369 struct btrfs_root *root = fs_info->extent_root;
10370 struct btrfs_path *path;
10371 struct btrfs_block_group_cache *block_group;
10372 struct btrfs_free_cluster *cluster;
10373 struct btrfs_root *tree_root = fs_info->tree_root;
10374 struct btrfs_key key;
10375 struct inode *inode;
10376 struct kobject *kobj = NULL;
10380 struct btrfs_caching_control *caching_ctl = NULL;
10383 block_group = btrfs_lookup_block_group(fs_info, group_start);
10384 BUG_ON(!block_group);
10385 BUG_ON(!block_group->ro);
10388 * Free the reserved super bytes from this block group before
10391 free_excluded_extents(fs_info, block_group);
10392 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10393 block_group->key.offset);
10395 memcpy(&key, &block_group->key, sizeof(key));
10396 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10397 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10398 BTRFS_BLOCK_GROUP_RAID1 |
10399 BTRFS_BLOCK_GROUP_RAID10))
10404 /* make sure this block group isn't part of an allocation cluster */
10405 cluster = &fs_info->data_alloc_cluster;
10406 spin_lock(&cluster->refill_lock);
10407 btrfs_return_cluster_to_free_space(block_group, cluster);
10408 spin_unlock(&cluster->refill_lock);
10411 * make sure this block group isn't part of a metadata
10412 * allocation cluster
10414 cluster = &fs_info->meta_alloc_cluster;
10415 spin_lock(&cluster->refill_lock);
10416 btrfs_return_cluster_to_free_space(block_group, cluster);
10417 spin_unlock(&cluster->refill_lock);
10419 path = btrfs_alloc_path();
10426 * get the inode first so any iput calls done for the io_list
10427 * aren't the final iput (no unlinks allowed now)
10429 inode = lookup_free_space_inode(fs_info, block_group, path);
10431 mutex_lock(&trans->transaction->cache_write_mutex);
10433 * make sure our free spache cache IO is done before remove the
10436 spin_lock(&trans->transaction->dirty_bgs_lock);
10437 if (!list_empty(&block_group->io_list)) {
10438 list_del_init(&block_group->io_list);
10440 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10442 spin_unlock(&trans->transaction->dirty_bgs_lock);
10443 btrfs_wait_cache_io(trans, block_group, path);
10444 btrfs_put_block_group(block_group);
10445 spin_lock(&trans->transaction->dirty_bgs_lock);
10448 if (!list_empty(&block_group->dirty_list)) {
10449 list_del_init(&block_group->dirty_list);
10450 btrfs_put_block_group(block_group);
10452 spin_unlock(&trans->transaction->dirty_bgs_lock);
10453 mutex_unlock(&trans->transaction->cache_write_mutex);
10455 if (!IS_ERR(inode)) {
10456 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10458 btrfs_add_delayed_iput(inode);
10461 clear_nlink(inode);
10462 /* One for the block groups ref */
10463 spin_lock(&block_group->lock);
10464 if (block_group->iref) {
10465 block_group->iref = 0;
10466 block_group->inode = NULL;
10467 spin_unlock(&block_group->lock);
10470 spin_unlock(&block_group->lock);
10472 /* One for our lookup ref */
10473 btrfs_add_delayed_iput(inode);
10476 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10477 key.offset = block_group->key.objectid;
10480 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10484 btrfs_release_path(path);
10486 ret = btrfs_del_item(trans, tree_root, path);
10489 btrfs_release_path(path);
10492 spin_lock(&fs_info->block_group_cache_lock);
10493 rb_erase(&block_group->cache_node,
10494 &fs_info->block_group_cache_tree);
10495 RB_CLEAR_NODE(&block_group->cache_node);
10497 if (fs_info->first_logical_byte == block_group->key.objectid)
10498 fs_info->first_logical_byte = (u64)-1;
10499 spin_unlock(&fs_info->block_group_cache_lock);
10501 down_write(&block_group->space_info->groups_sem);
10503 * we must use list_del_init so people can check to see if they
10504 * are still on the list after taking the semaphore
10506 list_del_init(&block_group->list);
10507 if (list_empty(&block_group->space_info->block_groups[index])) {
10508 kobj = block_group->space_info->block_group_kobjs[index];
10509 block_group->space_info->block_group_kobjs[index] = NULL;
10510 clear_avail_alloc_bits(fs_info, block_group->flags);
10512 up_write(&block_group->space_info->groups_sem);
10518 if (block_group->has_caching_ctl)
10519 caching_ctl = get_caching_control(block_group);
10520 if (block_group->cached == BTRFS_CACHE_STARTED)
10521 wait_block_group_cache_done(block_group);
10522 if (block_group->has_caching_ctl) {
10523 down_write(&fs_info->commit_root_sem);
10524 if (!caching_ctl) {
10525 struct btrfs_caching_control *ctl;
10527 list_for_each_entry(ctl,
10528 &fs_info->caching_block_groups, list)
10529 if (ctl->block_group == block_group) {
10531 refcount_inc(&caching_ctl->count);
10536 list_del_init(&caching_ctl->list);
10537 up_write(&fs_info->commit_root_sem);
10539 /* Once for the caching bgs list and once for us. */
10540 put_caching_control(caching_ctl);
10541 put_caching_control(caching_ctl);
10545 spin_lock(&trans->transaction->dirty_bgs_lock);
10546 if (!list_empty(&block_group->dirty_list)) {
10549 if (!list_empty(&block_group->io_list)) {
10552 spin_unlock(&trans->transaction->dirty_bgs_lock);
10553 btrfs_remove_free_space_cache(block_group);
10555 spin_lock(&block_group->space_info->lock);
10556 list_del_init(&block_group->ro_list);
10558 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10559 WARN_ON(block_group->space_info->total_bytes
10560 < block_group->key.offset);
10561 WARN_ON(block_group->space_info->bytes_readonly
10562 < block_group->key.offset);
10563 WARN_ON(block_group->space_info->disk_total
10564 < block_group->key.offset * factor);
10566 block_group->space_info->total_bytes -= block_group->key.offset;
10567 block_group->space_info->bytes_readonly -= block_group->key.offset;
10568 block_group->space_info->disk_total -= block_group->key.offset * factor;
10570 spin_unlock(&block_group->space_info->lock);
10572 memcpy(&key, &block_group->key, sizeof(key));
10574 mutex_lock(&fs_info->chunk_mutex);
10575 if (!list_empty(&em->list)) {
10576 /* We're in the transaction->pending_chunks list. */
10577 free_extent_map(em);
10579 spin_lock(&block_group->lock);
10580 block_group->removed = 1;
10582 * At this point trimming can't start on this block group, because we
10583 * removed the block group from the tree fs_info->block_group_cache_tree
10584 * so no one can't find it anymore and even if someone already got this
10585 * block group before we removed it from the rbtree, they have already
10586 * incremented block_group->trimming - if they didn't, they won't find
10587 * any free space entries because we already removed them all when we
10588 * called btrfs_remove_free_space_cache().
10590 * And we must not remove the extent map from the fs_info->mapping_tree
10591 * to prevent the same logical address range and physical device space
10592 * ranges from being reused for a new block group. This is because our
10593 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10594 * completely transactionless, so while it is trimming a range the
10595 * currently running transaction might finish and a new one start,
10596 * allowing for new block groups to be created that can reuse the same
10597 * physical device locations unless we take this special care.
10599 * There may also be an implicit trim operation if the file system
10600 * is mounted with -odiscard. The same protections must remain
10601 * in place until the extents have been discarded completely when
10602 * the transaction commit has completed.
10604 remove_em = (atomic_read(&block_group->trimming) == 0);
10606 * Make sure a trimmer task always sees the em in the pinned_chunks list
10607 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10608 * before checking block_group->removed).
10612 * Our em might be in trans->transaction->pending_chunks which
10613 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10614 * and so is the fs_info->pinned_chunks list.
10616 * So at this point we must be holding the chunk_mutex to avoid
10617 * any races with chunk allocation (more specifically at
10618 * volumes.c:contains_pending_extent()), to ensure it always
10619 * sees the em, either in the pending_chunks list or in the
10620 * pinned_chunks list.
10622 list_move_tail(&em->list, &fs_info->pinned_chunks);
10624 spin_unlock(&block_group->lock);
10627 struct extent_map_tree *em_tree;
10629 em_tree = &fs_info->mapping_tree.map_tree;
10630 write_lock(&em_tree->lock);
10632 * The em might be in the pending_chunks list, so make sure the
10633 * chunk mutex is locked, since remove_extent_mapping() will
10634 * delete us from that list.
10636 remove_extent_mapping(em_tree, em);
10637 write_unlock(&em_tree->lock);
10638 /* once for the tree */
10639 free_extent_map(em);
10642 mutex_unlock(&fs_info->chunk_mutex);
10644 ret = remove_block_group_free_space(trans, fs_info, block_group);
10648 btrfs_put_block_group(block_group);
10649 btrfs_put_block_group(block_group);
10651 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10657 ret = btrfs_del_item(trans, root, path);
10659 btrfs_free_path(path);
10663 struct btrfs_trans_handle *
10664 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10665 const u64 chunk_offset)
10667 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10668 struct extent_map *em;
10669 struct map_lookup *map;
10670 unsigned int num_items;
10672 read_lock(&em_tree->lock);
10673 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10674 read_unlock(&em_tree->lock);
10675 ASSERT(em && em->start == chunk_offset);
10678 * We need to reserve 3 + N units from the metadata space info in order
10679 * to remove a block group (done at btrfs_remove_chunk() and at
10680 * btrfs_remove_block_group()), which are used for:
10682 * 1 unit for adding the free space inode's orphan (located in the tree
10684 * 1 unit for deleting the block group item (located in the extent
10686 * 1 unit for deleting the free space item (located in tree of tree
10688 * N units for deleting N device extent items corresponding to each
10689 * stripe (located in the device tree).
10691 * In order to remove a block group we also need to reserve units in the
10692 * system space info in order to update the chunk tree (update one or
10693 * more device items and remove one chunk item), but this is done at
10694 * btrfs_remove_chunk() through a call to check_system_chunk().
10696 map = em->map_lookup;
10697 num_items = 3 + map->num_stripes;
10698 free_extent_map(em);
10700 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10705 * Process the unused_bgs list and remove any that don't have any allocated
10706 * space inside of them.
10708 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10710 struct btrfs_block_group_cache *block_group;
10711 struct btrfs_space_info *space_info;
10712 struct btrfs_trans_handle *trans;
10715 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10718 spin_lock(&fs_info->unused_bgs_lock);
10719 while (!list_empty(&fs_info->unused_bgs)) {
10723 block_group = list_first_entry(&fs_info->unused_bgs,
10724 struct btrfs_block_group_cache,
10726 list_del_init(&block_group->bg_list);
10728 space_info = block_group->space_info;
10730 if (ret || btrfs_mixed_space_info(space_info)) {
10731 btrfs_put_block_group(block_group);
10734 spin_unlock(&fs_info->unused_bgs_lock);
10736 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10738 /* Don't want to race with allocators so take the groups_sem */
10739 down_write(&space_info->groups_sem);
10740 spin_lock(&block_group->lock);
10741 if (block_group->reserved ||
10742 btrfs_block_group_used(&block_group->item) ||
10744 list_is_singular(&block_group->list)) {
10746 * We want to bail if we made new allocations or have
10747 * outstanding allocations in this block group. We do
10748 * the ro check in case balance is currently acting on
10749 * this block group.
10751 spin_unlock(&block_group->lock);
10752 up_write(&space_info->groups_sem);
10755 spin_unlock(&block_group->lock);
10757 /* We don't want to force the issue, only flip if it's ok. */
10758 ret = inc_block_group_ro(block_group, 0);
10759 up_write(&space_info->groups_sem);
10766 * Want to do this before we do anything else so we can recover
10767 * properly if we fail to join the transaction.
10769 trans = btrfs_start_trans_remove_block_group(fs_info,
10770 block_group->key.objectid);
10771 if (IS_ERR(trans)) {
10772 btrfs_dec_block_group_ro(block_group);
10773 ret = PTR_ERR(trans);
10778 * We could have pending pinned extents for this block group,
10779 * just delete them, we don't care about them anymore.
10781 start = block_group->key.objectid;
10782 end = start + block_group->key.offset - 1;
10784 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10785 * btrfs_finish_extent_commit(). If we are at transaction N,
10786 * another task might be running finish_extent_commit() for the
10787 * previous transaction N - 1, and have seen a range belonging
10788 * to the block group in freed_extents[] before we were able to
10789 * clear the whole block group range from freed_extents[]. This
10790 * means that task can lookup for the block group after we
10791 * unpinned it from freed_extents[] and removed it, leading to
10792 * a BUG_ON() at btrfs_unpin_extent_range().
10794 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10795 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10798 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10799 btrfs_dec_block_group_ro(block_group);
10802 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10805 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10806 btrfs_dec_block_group_ro(block_group);
10809 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10811 /* Reset pinned so btrfs_put_block_group doesn't complain */
10812 spin_lock(&space_info->lock);
10813 spin_lock(&block_group->lock);
10815 space_info->bytes_pinned -= block_group->pinned;
10816 space_info->bytes_readonly += block_group->pinned;
10817 percpu_counter_add(&space_info->total_bytes_pinned,
10818 -block_group->pinned);
10819 block_group->pinned = 0;
10821 spin_unlock(&block_group->lock);
10822 spin_unlock(&space_info->lock);
10824 /* DISCARD can flip during remount */
10825 trimming = btrfs_test_opt(fs_info, DISCARD);
10827 /* Implicit trim during transaction commit. */
10829 btrfs_get_block_group_trimming(block_group);
10832 * Btrfs_remove_chunk will abort the transaction if things go
10835 ret = btrfs_remove_chunk(trans, fs_info,
10836 block_group->key.objectid);
10840 btrfs_put_block_group_trimming(block_group);
10845 * If we're not mounted with -odiscard, we can just forget
10846 * about this block group. Otherwise we'll need to wait
10847 * until transaction commit to do the actual discard.
10850 spin_lock(&fs_info->unused_bgs_lock);
10852 * A concurrent scrub might have added us to the list
10853 * fs_info->unused_bgs, so use a list_move operation
10854 * to add the block group to the deleted_bgs list.
10856 list_move(&block_group->bg_list,
10857 &trans->transaction->deleted_bgs);
10858 spin_unlock(&fs_info->unused_bgs_lock);
10859 btrfs_get_block_group(block_group);
10862 btrfs_end_transaction(trans);
10864 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10865 btrfs_put_block_group(block_group);
10866 spin_lock(&fs_info->unused_bgs_lock);
10868 spin_unlock(&fs_info->unused_bgs_lock);
10871 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10873 struct btrfs_space_info *space_info;
10874 struct btrfs_super_block *disk_super;
10880 disk_super = fs_info->super_copy;
10881 if (!btrfs_super_root(disk_super))
10884 features = btrfs_super_incompat_flags(disk_super);
10885 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10888 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10889 ret = create_space_info(fs_info, flags, &space_info);
10894 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10895 ret = create_space_info(fs_info, flags, &space_info);
10897 flags = BTRFS_BLOCK_GROUP_METADATA;
10898 ret = create_space_info(fs_info, flags, &space_info);
10902 flags = BTRFS_BLOCK_GROUP_DATA;
10903 ret = create_space_info(fs_info, flags, &space_info);
10909 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10910 u64 start, u64 end)
10912 return unpin_extent_range(fs_info, start, end, false);
10916 * It used to be that old block groups would be left around forever.
10917 * Iterating over them would be enough to trim unused space. Since we
10918 * now automatically remove them, we also need to iterate over unallocated
10921 * We don't want a transaction for this since the discard may take a
10922 * substantial amount of time. We don't require that a transaction be
10923 * running, but we do need to take a running transaction into account
10924 * to ensure that we're not discarding chunks that were released in
10925 * the current transaction.
10927 * Holding the chunks lock will prevent other threads from allocating
10928 * or releasing chunks, but it won't prevent a running transaction
10929 * from committing and releasing the memory that the pending chunks
10930 * list head uses. For that, we need to take a reference to the
10933 static int btrfs_trim_free_extents(struct btrfs_device *device,
10934 u64 minlen, u64 *trimmed)
10936 u64 start = 0, len = 0;
10941 /* Not writeable = nothing to do. */
10942 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10945 /* No free space = nothing to do. */
10946 if (device->total_bytes <= device->bytes_used)
10952 struct btrfs_fs_info *fs_info = device->fs_info;
10953 struct btrfs_transaction *trans;
10956 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10960 down_read(&fs_info->commit_root_sem);
10962 spin_lock(&fs_info->trans_lock);
10963 trans = fs_info->running_transaction;
10965 refcount_inc(&trans->use_count);
10966 spin_unlock(&fs_info->trans_lock);
10968 ret = find_free_dev_extent_start(trans, device, minlen, start,
10971 btrfs_put_transaction(trans);
10974 up_read(&fs_info->commit_root_sem);
10975 mutex_unlock(&fs_info->chunk_mutex);
10976 if (ret == -ENOSPC)
10981 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10982 up_read(&fs_info->commit_root_sem);
10983 mutex_unlock(&fs_info->chunk_mutex);
10991 if (fatal_signal_pending(current)) {
10992 ret = -ERESTARTSYS;
11002 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11004 struct btrfs_block_group_cache *cache = NULL;
11005 struct btrfs_device *device;
11006 struct list_head *devices;
11011 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11015 * try to trim all FS space, our block group may start from non-zero.
11017 if (range->len == total_bytes)
11018 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11020 cache = btrfs_lookup_block_group(fs_info, range->start);
11023 if (cache->key.objectid >= (range->start + range->len)) {
11024 btrfs_put_block_group(cache);
11028 start = max(range->start, cache->key.objectid);
11029 end = min(range->start + range->len,
11030 cache->key.objectid + cache->key.offset);
11032 if (end - start >= range->minlen) {
11033 if (!block_group_cache_done(cache)) {
11034 ret = cache_block_group(cache, 0);
11036 btrfs_put_block_group(cache);
11039 ret = wait_block_group_cache_done(cache);
11041 btrfs_put_block_group(cache);
11045 ret = btrfs_trim_block_group(cache,
11051 trimmed += group_trimmed;
11053 btrfs_put_block_group(cache);
11058 cache = next_block_group(fs_info, cache);
11061 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11062 devices = &fs_info->fs_devices->alloc_list;
11063 list_for_each_entry(device, devices, dev_alloc_list) {
11064 ret = btrfs_trim_free_extents(device, range->minlen,
11069 trimmed += group_trimmed;
11071 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11073 range->len = trimmed;
11078 * btrfs_{start,end}_write_no_snapshotting() are similar to
11079 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11080 * data into the page cache through nocow before the subvolume is snapshoted,
11081 * but flush the data into disk after the snapshot creation, or to prevent
11082 * operations while snapshotting is ongoing and that cause the snapshot to be
11083 * inconsistent (writes followed by expanding truncates for example).
11085 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11087 percpu_counter_dec(&root->subv_writers->counter);
11089 * Make sure counter is updated before we wake up waiters.
11092 if (waitqueue_active(&root->subv_writers->wait))
11093 wake_up(&root->subv_writers->wait);
11096 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11098 if (atomic_read(&root->will_be_snapshotted))
11101 percpu_counter_inc(&root->subv_writers->counter);
11103 * Make sure counter is updated before we check for snapshot creation.
11106 if (atomic_read(&root->will_be_snapshotted)) {
11107 btrfs_end_write_no_snapshotting(root);
11113 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11118 ret = btrfs_start_write_no_snapshotting(root);
11121 wait_var_event(&root->will_be_snapshotted,
11122 !atomic_read(&root->will_be_snapshotted));