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_block_group_cache *cache;
2606 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
2608 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2610 btrfs_put_block_group(cache);
2612 if (head->is_data) {
2613 spin_lock(&delayed_refs->lock);
2614 delayed_refs->pending_csums -= head->num_bytes;
2615 spin_unlock(&delayed_refs->lock);
2619 if (head->must_insert_reserved) {
2620 btrfs_pin_extent(fs_info, head->bytenr,
2621 head->num_bytes, 1);
2622 if (head->is_data) {
2623 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2628 /* Also free its reserved qgroup space */
2629 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2630 head->qgroup_reserved);
2631 btrfs_delayed_ref_unlock(head);
2632 btrfs_put_delayed_ref_head(head);
2637 * Returns 0 on success or if called with an already aborted transaction.
2638 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2640 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2643 struct btrfs_fs_info *fs_info = trans->fs_info;
2644 struct btrfs_delayed_ref_root *delayed_refs;
2645 struct btrfs_delayed_ref_node *ref;
2646 struct btrfs_delayed_ref_head *locked_ref = NULL;
2647 struct btrfs_delayed_extent_op *extent_op;
2648 ktime_t start = ktime_get();
2650 unsigned long count = 0;
2651 unsigned long actual_count = 0;
2652 int must_insert_reserved = 0;
2654 delayed_refs = &trans->transaction->delayed_refs;
2660 spin_lock(&delayed_refs->lock);
2661 locked_ref = btrfs_select_ref_head(trans);
2663 spin_unlock(&delayed_refs->lock);
2667 /* grab the lock that says we are going to process
2668 * all the refs for this head */
2669 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2670 spin_unlock(&delayed_refs->lock);
2672 * we may have dropped the spin lock to get the head
2673 * mutex lock, and that might have given someone else
2674 * time to free the head. If that's true, it has been
2675 * removed from our list and we can move on.
2677 if (ret == -EAGAIN) {
2685 * We need to try and merge add/drops of the same ref since we
2686 * can run into issues with relocate dropping the implicit ref
2687 * and then it being added back again before the drop can
2688 * finish. If we merged anything we need to re-loop so we can
2690 * Or we can get node references of the same type that weren't
2691 * merged when created due to bumps in the tree mod seq, and
2692 * we need to merge them to prevent adding an inline extent
2693 * backref before dropping it (triggering a BUG_ON at
2694 * insert_inline_extent_backref()).
2696 spin_lock(&locked_ref->lock);
2697 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2701 * locked_ref is the head node, so we have to go one
2702 * node back for any delayed ref updates
2704 ref = select_delayed_ref(locked_ref);
2706 if (ref && ref->seq &&
2707 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2708 spin_unlock(&locked_ref->lock);
2709 unselect_delayed_ref_head(delayed_refs, locked_ref);
2717 * We're done processing refs in this ref_head, clean everything
2718 * up and move on to the next ref_head.
2721 ret = cleanup_ref_head(trans, fs_info, locked_ref);
2723 /* We dropped our lock, we need to loop. */
2736 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2737 RB_CLEAR_NODE(&ref->ref_node);
2738 if (!list_empty(&ref->add_list))
2739 list_del(&ref->add_list);
2741 * When we play the delayed ref, also correct the ref_mod on
2744 switch (ref->action) {
2745 case BTRFS_ADD_DELAYED_REF:
2746 case BTRFS_ADD_DELAYED_EXTENT:
2747 locked_ref->ref_mod -= ref->ref_mod;
2749 case BTRFS_DROP_DELAYED_REF:
2750 locked_ref->ref_mod += ref->ref_mod;
2755 atomic_dec(&delayed_refs->num_entries);
2758 * Record the must-insert_reserved flag before we drop the spin
2761 must_insert_reserved = locked_ref->must_insert_reserved;
2762 locked_ref->must_insert_reserved = 0;
2764 extent_op = locked_ref->extent_op;
2765 locked_ref->extent_op = NULL;
2766 spin_unlock(&locked_ref->lock);
2768 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2769 must_insert_reserved);
2771 btrfs_free_delayed_extent_op(extent_op);
2773 unselect_delayed_ref_head(delayed_refs, locked_ref);
2774 btrfs_put_delayed_ref(ref);
2775 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2780 btrfs_put_delayed_ref(ref);
2786 * We don't want to include ref heads since we can have empty ref heads
2787 * and those will drastically skew our runtime down since we just do
2788 * accounting, no actual extent tree updates.
2790 if (actual_count > 0) {
2791 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2795 * We weigh the current average higher than our current runtime
2796 * to avoid large swings in the average.
2798 spin_lock(&delayed_refs->lock);
2799 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2800 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2801 spin_unlock(&delayed_refs->lock);
2806 #ifdef SCRAMBLE_DELAYED_REFS
2808 * Normally delayed refs get processed in ascending bytenr order. This
2809 * correlates in most cases to the order added. To expose dependencies on this
2810 * order, we start to process the tree in the middle instead of the beginning
2812 static u64 find_middle(struct rb_root *root)
2814 struct rb_node *n = root->rb_node;
2815 struct btrfs_delayed_ref_node *entry;
2818 u64 first = 0, last = 0;
2822 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2823 first = entry->bytenr;
2827 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2828 last = entry->bytenr;
2833 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2834 WARN_ON(!entry->in_tree);
2836 middle = entry->bytenr;
2849 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2853 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2854 sizeof(struct btrfs_extent_inline_ref));
2855 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2856 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2859 * We don't ever fill up leaves all the way so multiply by 2 just to be
2860 * closer to what we're really going to want to use.
2862 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2866 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2867 * would require to store the csums for that many bytes.
2869 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2872 u64 num_csums_per_leaf;
2875 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2876 num_csums_per_leaf = div64_u64(csum_size,
2877 (u64)btrfs_super_csum_size(fs_info->super_copy));
2878 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2879 num_csums += num_csums_per_leaf - 1;
2880 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2884 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2885 struct btrfs_fs_info *fs_info)
2887 struct btrfs_block_rsv *global_rsv;
2888 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2889 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2890 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2891 u64 num_bytes, num_dirty_bgs_bytes;
2894 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2895 num_heads = heads_to_leaves(fs_info, num_heads);
2897 num_bytes += (num_heads - 1) * fs_info->nodesize;
2899 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2901 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2903 global_rsv = &fs_info->global_block_rsv;
2906 * If we can't allocate any more chunks lets make sure we have _lots_ of
2907 * wiggle room since running delayed refs can create more delayed refs.
2909 if (global_rsv->space_info->full) {
2910 num_dirty_bgs_bytes <<= 1;
2914 spin_lock(&global_rsv->lock);
2915 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2917 spin_unlock(&global_rsv->lock);
2921 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2922 struct btrfs_fs_info *fs_info)
2925 atomic_read(&trans->transaction->delayed_refs.num_entries);
2930 avg_runtime = fs_info->avg_delayed_ref_runtime;
2931 val = num_entries * avg_runtime;
2932 if (val >= NSEC_PER_SEC)
2934 if (val >= NSEC_PER_SEC / 2)
2937 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2940 struct async_delayed_refs {
2941 struct btrfs_root *root;
2946 struct completion wait;
2947 struct btrfs_work work;
2950 static inline struct async_delayed_refs *
2951 to_async_delayed_refs(struct btrfs_work *work)
2953 return container_of(work, struct async_delayed_refs, work);
2956 static void delayed_ref_async_start(struct btrfs_work *work)
2958 struct async_delayed_refs *async = to_async_delayed_refs(work);
2959 struct btrfs_trans_handle *trans;
2960 struct btrfs_fs_info *fs_info = async->root->fs_info;
2963 /* if the commit is already started, we don't need to wait here */
2964 if (btrfs_transaction_blocked(fs_info))
2967 trans = btrfs_join_transaction(async->root);
2968 if (IS_ERR(trans)) {
2969 async->error = PTR_ERR(trans);
2974 * trans->sync means that when we call end_transaction, we won't
2975 * wait on delayed refs
2979 /* Don't bother flushing if we got into a different transaction */
2980 if (trans->transid > async->transid)
2983 ret = btrfs_run_delayed_refs(trans, async->count);
2987 ret = btrfs_end_transaction(trans);
2988 if (ret && !async->error)
2992 complete(&async->wait);
2997 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2998 unsigned long count, u64 transid, int wait)
3000 struct async_delayed_refs *async;
3003 async = kmalloc(sizeof(*async), GFP_NOFS);
3007 async->root = fs_info->tree_root;
3008 async->count = count;
3010 async->transid = transid;
3015 init_completion(&async->wait);
3017 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3018 delayed_ref_async_start, NULL, NULL);
3020 btrfs_queue_work(fs_info->extent_workers, &async->work);
3023 wait_for_completion(&async->wait);
3032 * this starts processing the delayed reference count updates and
3033 * extent insertions we have queued up so far. count can be
3034 * 0, which means to process everything in the tree at the start
3035 * of the run (but not newly added entries), or it can be some target
3036 * number you'd like to process.
3038 * Returns 0 on success or if called with an aborted transaction
3039 * Returns <0 on error and aborts the transaction
3041 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3042 unsigned long count)
3044 struct btrfs_fs_info *fs_info = trans->fs_info;
3045 struct rb_node *node;
3046 struct btrfs_delayed_ref_root *delayed_refs;
3047 struct btrfs_delayed_ref_head *head;
3049 int run_all = count == (unsigned long)-1;
3050 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3052 /* We'll clean this up in btrfs_cleanup_transaction */
3056 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3059 delayed_refs = &trans->transaction->delayed_refs;
3061 count = atomic_read(&delayed_refs->num_entries) * 2;
3064 #ifdef SCRAMBLE_DELAYED_REFS
3065 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3067 trans->can_flush_pending_bgs = false;
3068 ret = __btrfs_run_delayed_refs(trans, count);
3070 btrfs_abort_transaction(trans, ret);
3075 if (!list_empty(&trans->new_bgs))
3076 btrfs_create_pending_block_groups(trans);
3078 spin_lock(&delayed_refs->lock);
3079 node = rb_first(&delayed_refs->href_root);
3081 spin_unlock(&delayed_refs->lock);
3084 head = rb_entry(node, struct btrfs_delayed_ref_head,
3086 refcount_inc(&head->refs);
3087 spin_unlock(&delayed_refs->lock);
3089 /* Mutex was contended, block until it's released and retry. */
3090 mutex_lock(&head->mutex);
3091 mutex_unlock(&head->mutex);
3093 btrfs_put_delayed_ref_head(head);
3098 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3102 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3103 struct btrfs_fs_info *fs_info,
3104 u64 bytenr, u64 num_bytes, u64 flags,
3105 int level, int is_data)
3107 struct btrfs_delayed_extent_op *extent_op;
3110 extent_op = btrfs_alloc_delayed_extent_op();
3114 extent_op->flags_to_set = flags;
3115 extent_op->update_flags = true;
3116 extent_op->update_key = false;
3117 extent_op->is_data = is_data ? true : false;
3118 extent_op->level = level;
3120 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3121 num_bytes, extent_op);
3123 btrfs_free_delayed_extent_op(extent_op);
3127 static noinline int check_delayed_ref(struct btrfs_root *root,
3128 struct btrfs_path *path,
3129 u64 objectid, u64 offset, u64 bytenr)
3131 struct btrfs_delayed_ref_head *head;
3132 struct btrfs_delayed_ref_node *ref;
3133 struct btrfs_delayed_data_ref *data_ref;
3134 struct btrfs_delayed_ref_root *delayed_refs;
3135 struct btrfs_transaction *cur_trans;
3136 struct rb_node *node;
3139 cur_trans = root->fs_info->running_transaction;
3143 delayed_refs = &cur_trans->delayed_refs;
3144 spin_lock(&delayed_refs->lock);
3145 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3147 spin_unlock(&delayed_refs->lock);
3151 if (!mutex_trylock(&head->mutex)) {
3152 refcount_inc(&head->refs);
3153 spin_unlock(&delayed_refs->lock);
3155 btrfs_release_path(path);
3158 * Mutex was contended, block until it's released and let
3161 mutex_lock(&head->mutex);
3162 mutex_unlock(&head->mutex);
3163 btrfs_put_delayed_ref_head(head);
3166 spin_unlock(&delayed_refs->lock);
3168 spin_lock(&head->lock);
3170 * XXX: We should replace this with a proper search function in the
3173 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3174 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3175 /* If it's a shared ref we know a cross reference exists */
3176 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3181 data_ref = btrfs_delayed_node_to_data_ref(ref);
3184 * If our ref doesn't match the one we're currently looking at
3185 * then we have a cross reference.
3187 if (data_ref->root != root->root_key.objectid ||
3188 data_ref->objectid != objectid ||
3189 data_ref->offset != offset) {
3194 spin_unlock(&head->lock);
3195 mutex_unlock(&head->mutex);
3199 static noinline int check_committed_ref(struct btrfs_root *root,
3200 struct btrfs_path *path,
3201 u64 objectid, u64 offset, u64 bytenr)
3203 struct btrfs_fs_info *fs_info = root->fs_info;
3204 struct btrfs_root *extent_root = fs_info->extent_root;
3205 struct extent_buffer *leaf;
3206 struct btrfs_extent_data_ref *ref;
3207 struct btrfs_extent_inline_ref *iref;
3208 struct btrfs_extent_item *ei;
3209 struct btrfs_key key;
3214 key.objectid = bytenr;
3215 key.offset = (u64)-1;
3216 key.type = BTRFS_EXTENT_ITEM_KEY;
3218 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3221 BUG_ON(ret == 0); /* Corruption */
3224 if (path->slots[0] == 0)
3228 leaf = path->nodes[0];
3229 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3231 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3235 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3236 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3237 if (item_size < sizeof(*ei)) {
3238 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3242 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3244 if (item_size != sizeof(*ei) +
3245 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3248 if (btrfs_extent_generation(leaf, ei) <=
3249 btrfs_root_last_snapshot(&root->root_item))
3252 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3254 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3255 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3258 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3259 if (btrfs_extent_refs(leaf, ei) !=
3260 btrfs_extent_data_ref_count(leaf, ref) ||
3261 btrfs_extent_data_ref_root(leaf, ref) !=
3262 root->root_key.objectid ||
3263 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3264 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3272 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3275 struct btrfs_path *path;
3279 path = btrfs_alloc_path();
3284 ret = check_committed_ref(root, path, objectid,
3286 if (ret && ret != -ENOENT)
3289 ret2 = check_delayed_ref(root, path, objectid,
3291 } while (ret2 == -EAGAIN);
3293 if (ret2 && ret2 != -ENOENT) {
3298 if (ret != -ENOENT || ret2 != -ENOENT)
3301 btrfs_free_path(path);
3302 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3307 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3308 struct btrfs_root *root,
3309 struct extent_buffer *buf,
3310 int full_backref, int inc)
3312 struct btrfs_fs_info *fs_info = root->fs_info;
3318 struct btrfs_key key;
3319 struct btrfs_file_extent_item *fi;
3323 int (*process_func)(struct btrfs_trans_handle *,
3324 struct btrfs_root *,
3325 u64, u64, u64, u64, u64, u64);
3328 if (btrfs_is_testing(fs_info))
3331 ref_root = btrfs_header_owner(buf);
3332 nritems = btrfs_header_nritems(buf);
3333 level = btrfs_header_level(buf);
3335 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3339 process_func = btrfs_inc_extent_ref;
3341 process_func = btrfs_free_extent;
3344 parent = buf->start;
3348 for (i = 0; i < nritems; i++) {
3350 btrfs_item_key_to_cpu(buf, &key, i);
3351 if (key.type != BTRFS_EXTENT_DATA_KEY)
3353 fi = btrfs_item_ptr(buf, i,
3354 struct btrfs_file_extent_item);
3355 if (btrfs_file_extent_type(buf, fi) ==
3356 BTRFS_FILE_EXTENT_INLINE)
3358 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3362 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3363 key.offset -= btrfs_file_extent_offset(buf, fi);
3364 ret = process_func(trans, root, bytenr, num_bytes,
3365 parent, ref_root, key.objectid,
3370 bytenr = btrfs_node_blockptr(buf, i);
3371 num_bytes = fs_info->nodesize;
3372 ret = process_func(trans, root, bytenr, num_bytes,
3373 parent, ref_root, level - 1, 0);
3383 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3384 struct extent_buffer *buf, int full_backref)
3386 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3389 int btrfs_dec_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, 0);
3395 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3396 struct btrfs_fs_info *fs_info,
3397 struct btrfs_path *path,
3398 struct btrfs_block_group_cache *cache)
3401 struct btrfs_root *extent_root = fs_info->extent_root;
3403 struct extent_buffer *leaf;
3405 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3412 leaf = path->nodes[0];
3413 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3414 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3415 btrfs_mark_buffer_dirty(leaf);
3417 btrfs_release_path(path);
3422 static struct btrfs_block_group_cache *
3423 next_block_group(struct btrfs_fs_info *fs_info,
3424 struct btrfs_block_group_cache *cache)
3426 struct rb_node *node;
3428 spin_lock(&fs_info->block_group_cache_lock);
3430 /* If our block group was removed, we need a full search. */
3431 if (RB_EMPTY_NODE(&cache->cache_node)) {
3432 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3434 spin_unlock(&fs_info->block_group_cache_lock);
3435 btrfs_put_block_group(cache);
3436 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3438 node = rb_next(&cache->cache_node);
3439 btrfs_put_block_group(cache);
3441 cache = rb_entry(node, struct btrfs_block_group_cache,
3443 btrfs_get_block_group(cache);
3446 spin_unlock(&fs_info->block_group_cache_lock);
3450 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3451 struct btrfs_trans_handle *trans,
3452 struct btrfs_path *path)
3454 struct btrfs_fs_info *fs_info = block_group->fs_info;
3455 struct btrfs_root *root = fs_info->tree_root;
3456 struct inode *inode = NULL;
3457 struct extent_changeset *data_reserved = NULL;
3459 int dcs = BTRFS_DC_ERROR;
3465 * If this block group is smaller than 100 megs don't bother caching the
3468 if (block_group->key.offset < (100 * SZ_1M)) {
3469 spin_lock(&block_group->lock);
3470 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3471 spin_unlock(&block_group->lock);
3478 inode = lookup_free_space_inode(fs_info, block_group, path);
3479 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3480 ret = PTR_ERR(inode);
3481 btrfs_release_path(path);
3485 if (IS_ERR(inode)) {
3489 if (block_group->ro)
3492 ret = create_free_space_inode(fs_info, trans, block_group,
3500 * We want to set the generation to 0, that way if anything goes wrong
3501 * from here on out we know not to trust this cache when we load up next
3504 BTRFS_I(inode)->generation = 0;
3505 ret = btrfs_update_inode(trans, root, inode);
3508 * So theoretically we could recover from this, simply set the
3509 * super cache generation to 0 so we know to invalidate the
3510 * cache, but then we'd have to keep track of the block groups
3511 * that fail this way so we know we _have_ to reset this cache
3512 * before the next commit or risk reading stale cache. So to
3513 * limit our exposure to horrible edge cases lets just abort the
3514 * transaction, this only happens in really bad situations
3517 btrfs_abort_transaction(trans, ret);
3522 /* We've already setup this transaction, go ahead and exit */
3523 if (block_group->cache_generation == trans->transid &&
3524 i_size_read(inode)) {
3525 dcs = BTRFS_DC_SETUP;
3529 if (i_size_read(inode) > 0) {
3530 ret = btrfs_check_trunc_cache_free_space(fs_info,
3531 &fs_info->global_block_rsv);
3535 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3540 spin_lock(&block_group->lock);
3541 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3542 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3544 * don't bother trying to write stuff out _if_
3545 * a) we're not cached,
3546 * b) we're with nospace_cache mount option,
3547 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3549 dcs = BTRFS_DC_WRITTEN;
3550 spin_unlock(&block_group->lock);
3553 spin_unlock(&block_group->lock);
3556 * We hit an ENOSPC when setting up the cache in this transaction, just
3557 * skip doing the setup, we've already cleared the cache so we're safe.
3559 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3565 * Try to preallocate enough space based on how big the block group is.
3566 * Keep in mind this has to include any pinned space which could end up
3567 * taking up quite a bit since it's not folded into the other space
3570 num_pages = div_u64(block_group->key.offset, SZ_256M);
3575 num_pages *= PAGE_SIZE;
3577 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3581 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3582 num_pages, num_pages,
3585 * Our cache requires contiguous chunks so that we don't modify a bunch
3586 * of metadata or split extents when writing the cache out, which means
3587 * we can enospc if we are heavily fragmented in addition to just normal
3588 * out of space conditions. So if we hit this just skip setting up any
3589 * other block groups for this transaction, maybe we'll unpin enough
3590 * space the next time around.
3593 dcs = BTRFS_DC_SETUP;
3594 else if (ret == -ENOSPC)
3595 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3600 btrfs_release_path(path);
3602 spin_lock(&block_group->lock);
3603 if (!ret && dcs == BTRFS_DC_SETUP)
3604 block_group->cache_generation = trans->transid;
3605 block_group->disk_cache_state = dcs;
3606 spin_unlock(&block_group->lock);
3608 extent_changeset_free(data_reserved);
3612 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3613 struct btrfs_fs_info *fs_info)
3615 struct btrfs_block_group_cache *cache, *tmp;
3616 struct btrfs_transaction *cur_trans = trans->transaction;
3617 struct btrfs_path *path;
3619 if (list_empty(&cur_trans->dirty_bgs) ||
3620 !btrfs_test_opt(fs_info, SPACE_CACHE))
3623 path = btrfs_alloc_path();
3627 /* Could add new block groups, use _safe just in case */
3628 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3630 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3631 cache_save_setup(cache, trans, path);
3634 btrfs_free_path(path);
3639 * transaction commit does final block group cache writeback during a
3640 * critical section where nothing is allowed to change the FS. This is
3641 * required in order for the cache to actually match the block group,
3642 * but can introduce a lot of latency into the commit.
3644 * So, btrfs_start_dirty_block_groups is here to kick off block group
3645 * cache IO. There's a chance we'll have to redo some of it if the
3646 * block group changes again during the commit, but it greatly reduces
3647 * the commit latency by getting rid of the easy block groups while
3648 * we're still allowing others to join the commit.
3650 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3652 struct btrfs_fs_info *fs_info = trans->fs_info;
3653 struct btrfs_block_group_cache *cache;
3654 struct btrfs_transaction *cur_trans = trans->transaction;
3657 struct btrfs_path *path = NULL;
3659 struct list_head *io = &cur_trans->io_bgs;
3660 int num_started = 0;
3663 spin_lock(&cur_trans->dirty_bgs_lock);
3664 if (list_empty(&cur_trans->dirty_bgs)) {
3665 spin_unlock(&cur_trans->dirty_bgs_lock);
3668 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3669 spin_unlock(&cur_trans->dirty_bgs_lock);
3673 * make sure all the block groups on our dirty list actually
3676 btrfs_create_pending_block_groups(trans);
3679 path = btrfs_alloc_path();
3685 * cache_write_mutex is here only to save us from balance or automatic
3686 * removal of empty block groups deleting this block group while we are
3687 * writing out the cache
3689 mutex_lock(&trans->transaction->cache_write_mutex);
3690 while (!list_empty(&dirty)) {
3691 cache = list_first_entry(&dirty,
3692 struct btrfs_block_group_cache,
3695 * this can happen if something re-dirties a block
3696 * group that is already under IO. Just wait for it to
3697 * finish and then do it all again
3699 if (!list_empty(&cache->io_list)) {
3700 list_del_init(&cache->io_list);
3701 btrfs_wait_cache_io(trans, cache, path);
3702 btrfs_put_block_group(cache);
3707 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3708 * if it should update the cache_state. Don't delete
3709 * until after we wait.
3711 * Since we're not running in the commit critical section
3712 * we need the dirty_bgs_lock to protect from update_block_group
3714 spin_lock(&cur_trans->dirty_bgs_lock);
3715 list_del_init(&cache->dirty_list);
3716 spin_unlock(&cur_trans->dirty_bgs_lock);
3720 cache_save_setup(cache, trans, path);
3722 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3723 cache->io_ctl.inode = NULL;
3724 ret = btrfs_write_out_cache(fs_info, trans,
3726 if (ret == 0 && cache->io_ctl.inode) {
3731 * The cache_write_mutex is protecting the
3732 * io_list, also refer to the definition of
3733 * btrfs_transaction::io_bgs for more details
3735 list_add_tail(&cache->io_list, io);
3738 * if we failed to write the cache, the
3739 * generation will be bad and life goes on
3745 ret = write_one_cache_group(trans, fs_info,
3748 * Our block group might still be attached to the list
3749 * of new block groups in the transaction handle of some
3750 * other task (struct btrfs_trans_handle->new_bgs). This
3751 * means its block group item isn't yet in the extent
3752 * tree. If this happens ignore the error, as we will
3753 * try again later in the critical section of the
3754 * transaction commit.
3756 if (ret == -ENOENT) {
3758 spin_lock(&cur_trans->dirty_bgs_lock);
3759 if (list_empty(&cache->dirty_list)) {
3760 list_add_tail(&cache->dirty_list,
3761 &cur_trans->dirty_bgs);
3762 btrfs_get_block_group(cache);
3764 spin_unlock(&cur_trans->dirty_bgs_lock);
3766 btrfs_abort_transaction(trans, ret);
3770 /* if its not on the io list, we need to put the block group */
3772 btrfs_put_block_group(cache);
3778 * Avoid blocking other tasks for too long. It might even save
3779 * us from writing caches for block groups that are going to be
3782 mutex_unlock(&trans->transaction->cache_write_mutex);
3783 mutex_lock(&trans->transaction->cache_write_mutex);
3785 mutex_unlock(&trans->transaction->cache_write_mutex);
3788 * go through delayed refs for all the stuff we've just kicked off
3789 * and then loop back (just once)
3791 ret = btrfs_run_delayed_refs(trans, 0);
3792 if (!ret && loops == 0) {
3794 spin_lock(&cur_trans->dirty_bgs_lock);
3795 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3797 * dirty_bgs_lock protects us from concurrent block group
3798 * deletes too (not just cache_write_mutex).
3800 if (!list_empty(&dirty)) {
3801 spin_unlock(&cur_trans->dirty_bgs_lock);
3804 spin_unlock(&cur_trans->dirty_bgs_lock);
3805 } else if (ret < 0) {
3806 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3809 btrfs_free_path(path);
3813 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3814 struct btrfs_fs_info *fs_info)
3816 struct btrfs_block_group_cache *cache;
3817 struct btrfs_transaction *cur_trans = trans->transaction;
3820 struct btrfs_path *path;
3821 struct list_head *io = &cur_trans->io_bgs;
3822 int num_started = 0;
3824 path = btrfs_alloc_path();
3829 * Even though we are in the critical section of the transaction commit,
3830 * we can still have concurrent tasks adding elements to this
3831 * transaction's list of dirty block groups. These tasks correspond to
3832 * endio free space workers started when writeback finishes for a
3833 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3834 * allocate new block groups as a result of COWing nodes of the root
3835 * tree when updating the free space inode. The writeback for the space
3836 * caches is triggered by an earlier call to
3837 * btrfs_start_dirty_block_groups() and iterations of the following
3839 * Also we want to do the cache_save_setup first and then run the
3840 * delayed refs to make sure we have the best chance at doing this all
3843 spin_lock(&cur_trans->dirty_bgs_lock);
3844 while (!list_empty(&cur_trans->dirty_bgs)) {
3845 cache = list_first_entry(&cur_trans->dirty_bgs,
3846 struct btrfs_block_group_cache,
3850 * this can happen if cache_save_setup re-dirties a block
3851 * group that is already under IO. Just wait for it to
3852 * finish and then do it all again
3854 if (!list_empty(&cache->io_list)) {
3855 spin_unlock(&cur_trans->dirty_bgs_lock);
3856 list_del_init(&cache->io_list);
3857 btrfs_wait_cache_io(trans, cache, path);
3858 btrfs_put_block_group(cache);
3859 spin_lock(&cur_trans->dirty_bgs_lock);
3863 * don't remove from the dirty list until after we've waited
3866 list_del_init(&cache->dirty_list);
3867 spin_unlock(&cur_trans->dirty_bgs_lock);
3870 cache_save_setup(cache, trans, path);
3873 ret = btrfs_run_delayed_refs(trans,
3874 (unsigned long) -1);
3876 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3877 cache->io_ctl.inode = NULL;
3878 ret = btrfs_write_out_cache(fs_info, trans,
3880 if (ret == 0 && cache->io_ctl.inode) {
3883 list_add_tail(&cache->io_list, io);
3886 * if we failed to write the cache, the
3887 * generation will be bad and life goes on
3893 ret = write_one_cache_group(trans, fs_info,
3896 * One of the free space endio workers might have
3897 * created a new block group while updating a free space
3898 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3899 * and hasn't released its transaction handle yet, in
3900 * which case the new block group is still attached to
3901 * its transaction handle and its creation has not
3902 * finished yet (no block group item in the extent tree
3903 * yet, etc). If this is the case, wait for all free
3904 * space endio workers to finish and retry. This is a
3905 * a very rare case so no need for a more efficient and
3908 if (ret == -ENOENT) {
3909 wait_event(cur_trans->writer_wait,
3910 atomic_read(&cur_trans->num_writers) == 1);
3911 ret = write_one_cache_group(trans, fs_info,
3915 btrfs_abort_transaction(trans, ret);
3918 /* if its not on the io list, we need to put the block group */
3920 btrfs_put_block_group(cache);
3921 spin_lock(&cur_trans->dirty_bgs_lock);
3923 spin_unlock(&cur_trans->dirty_bgs_lock);
3926 * Refer to the definition of io_bgs member for details why it's safe
3927 * to use it without any locking
3929 while (!list_empty(io)) {
3930 cache = list_first_entry(io, struct btrfs_block_group_cache,
3932 list_del_init(&cache->io_list);
3933 btrfs_wait_cache_io(trans, cache, path);
3934 btrfs_put_block_group(cache);
3937 btrfs_free_path(path);
3941 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3943 struct btrfs_block_group_cache *block_group;
3946 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3947 if (!block_group || block_group->ro)
3950 btrfs_put_block_group(block_group);
3954 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3956 struct btrfs_block_group_cache *bg;
3959 bg = btrfs_lookup_block_group(fs_info, bytenr);
3963 spin_lock(&bg->lock);
3967 atomic_inc(&bg->nocow_writers);
3968 spin_unlock(&bg->lock);
3970 /* no put on block group, done by btrfs_dec_nocow_writers */
3972 btrfs_put_block_group(bg);
3978 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3980 struct btrfs_block_group_cache *bg;
3982 bg = btrfs_lookup_block_group(fs_info, bytenr);
3984 if (atomic_dec_and_test(&bg->nocow_writers))
3985 wake_up_var(&bg->nocow_writers);
3987 * Once for our lookup and once for the lookup done by a previous call
3988 * to btrfs_inc_nocow_writers()
3990 btrfs_put_block_group(bg);
3991 btrfs_put_block_group(bg);
3994 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3996 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
3999 static const char *alloc_name(u64 flags)
4002 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4004 case BTRFS_BLOCK_GROUP_METADATA:
4006 case BTRFS_BLOCK_GROUP_DATA:
4008 case BTRFS_BLOCK_GROUP_SYSTEM:
4012 return "invalid-combination";
4016 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4017 struct btrfs_space_info **new)
4020 struct btrfs_space_info *space_info;
4024 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4028 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4035 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4036 INIT_LIST_HEAD(&space_info->block_groups[i]);
4037 init_rwsem(&space_info->groups_sem);
4038 spin_lock_init(&space_info->lock);
4039 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4040 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4041 init_waitqueue_head(&space_info->wait);
4042 INIT_LIST_HEAD(&space_info->ro_bgs);
4043 INIT_LIST_HEAD(&space_info->tickets);
4044 INIT_LIST_HEAD(&space_info->priority_tickets);
4046 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4047 info->space_info_kobj, "%s",
4048 alloc_name(space_info->flags));
4050 percpu_counter_destroy(&space_info->total_bytes_pinned);
4056 list_add_rcu(&space_info->list, &info->space_info);
4057 if (flags & BTRFS_BLOCK_GROUP_DATA)
4058 info->data_sinfo = space_info;
4063 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4064 u64 total_bytes, u64 bytes_used,
4066 struct btrfs_space_info **space_info)
4068 struct btrfs_space_info *found;
4071 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4072 BTRFS_BLOCK_GROUP_RAID10))
4077 found = __find_space_info(info, flags);
4079 spin_lock(&found->lock);
4080 found->total_bytes += total_bytes;
4081 found->disk_total += total_bytes * factor;
4082 found->bytes_used += bytes_used;
4083 found->disk_used += bytes_used * factor;
4084 found->bytes_readonly += bytes_readonly;
4085 if (total_bytes > 0)
4087 space_info_add_new_bytes(info, found, total_bytes -
4088 bytes_used - bytes_readonly);
4089 spin_unlock(&found->lock);
4090 *space_info = found;
4093 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4095 u64 extra_flags = chunk_to_extended(flags) &
4096 BTRFS_EXTENDED_PROFILE_MASK;
4098 write_seqlock(&fs_info->profiles_lock);
4099 if (flags & BTRFS_BLOCK_GROUP_DATA)
4100 fs_info->avail_data_alloc_bits |= extra_flags;
4101 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4102 fs_info->avail_metadata_alloc_bits |= extra_flags;
4103 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4104 fs_info->avail_system_alloc_bits |= extra_flags;
4105 write_sequnlock(&fs_info->profiles_lock);
4109 * returns target flags in extended format or 0 if restripe for this
4110 * chunk_type is not in progress
4112 * should be called with either volume_mutex or balance_lock held
4114 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4116 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4122 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4123 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4124 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4125 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4126 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4127 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4128 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4129 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4130 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4137 * @flags: available profiles in extended format (see ctree.h)
4139 * Returns reduced profile in chunk format. If profile changing is in
4140 * progress (either running or paused) picks the target profile (if it's
4141 * already available), otherwise falls back to plain reducing.
4143 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4145 u64 num_devices = fs_info->fs_devices->rw_devices;
4151 * see if restripe for this chunk_type is in progress, if so
4152 * try to reduce to the target profile
4154 spin_lock(&fs_info->balance_lock);
4155 target = get_restripe_target(fs_info, flags);
4157 /* pick target profile only if it's already available */
4158 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4159 spin_unlock(&fs_info->balance_lock);
4160 return extended_to_chunk(target);
4163 spin_unlock(&fs_info->balance_lock);
4165 /* First, mask out the RAID levels which aren't possible */
4166 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4167 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4168 allowed |= btrfs_raid_group[raid_type];
4172 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4173 allowed = BTRFS_BLOCK_GROUP_RAID6;
4174 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4175 allowed = BTRFS_BLOCK_GROUP_RAID5;
4176 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4177 allowed = BTRFS_BLOCK_GROUP_RAID10;
4178 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4179 allowed = BTRFS_BLOCK_GROUP_RAID1;
4180 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4181 allowed = BTRFS_BLOCK_GROUP_RAID0;
4183 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4185 return extended_to_chunk(flags | allowed);
4188 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4195 seq = read_seqbegin(&fs_info->profiles_lock);
4197 if (flags & BTRFS_BLOCK_GROUP_DATA)
4198 flags |= fs_info->avail_data_alloc_bits;
4199 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4200 flags |= fs_info->avail_system_alloc_bits;
4201 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4202 flags |= fs_info->avail_metadata_alloc_bits;
4203 } while (read_seqretry(&fs_info->profiles_lock, seq));
4205 return btrfs_reduce_alloc_profile(fs_info, flags);
4208 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4210 struct btrfs_fs_info *fs_info = root->fs_info;
4215 flags = BTRFS_BLOCK_GROUP_DATA;
4216 else if (root == fs_info->chunk_root)
4217 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4219 flags = BTRFS_BLOCK_GROUP_METADATA;
4221 ret = get_alloc_profile(fs_info, flags);
4225 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4227 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4230 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4232 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4235 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4237 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4240 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4241 bool may_use_included)
4244 return s_info->bytes_used + s_info->bytes_reserved +
4245 s_info->bytes_pinned + s_info->bytes_readonly +
4246 (may_use_included ? s_info->bytes_may_use : 0);
4249 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4251 struct btrfs_root *root = inode->root;
4252 struct btrfs_fs_info *fs_info = root->fs_info;
4253 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4256 int need_commit = 2;
4257 int have_pinned_space;
4259 /* make sure bytes are sectorsize aligned */
4260 bytes = ALIGN(bytes, fs_info->sectorsize);
4262 if (btrfs_is_free_space_inode(inode)) {
4264 ASSERT(current->journal_info);
4268 /* make sure we have enough space to handle the data first */
4269 spin_lock(&data_sinfo->lock);
4270 used = btrfs_space_info_used(data_sinfo, true);
4272 if (used + bytes > data_sinfo->total_bytes) {
4273 struct btrfs_trans_handle *trans;
4276 * if we don't have enough free bytes in this space then we need
4277 * to alloc a new chunk.
4279 if (!data_sinfo->full) {
4282 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4283 spin_unlock(&data_sinfo->lock);
4285 alloc_target = btrfs_data_alloc_profile(fs_info);
4287 * It is ugly that we don't call nolock join
4288 * transaction for the free space inode case here.
4289 * But it is safe because we only do the data space
4290 * reservation for the free space cache in the
4291 * transaction context, the common join transaction
4292 * just increase the counter of the current transaction
4293 * handler, doesn't try to acquire the trans_lock of
4296 trans = btrfs_join_transaction(root);
4298 return PTR_ERR(trans);
4300 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4301 CHUNK_ALLOC_NO_FORCE);
4302 btrfs_end_transaction(trans);
4307 have_pinned_space = 1;
4316 * If we don't have enough pinned space to deal with this
4317 * allocation, and no removed chunk in current transaction,
4318 * don't bother committing the transaction.
4320 have_pinned_space = percpu_counter_compare(
4321 &data_sinfo->total_bytes_pinned,
4322 used + bytes - data_sinfo->total_bytes);
4323 spin_unlock(&data_sinfo->lock);
4325 /* commit the current transaction and try again */
4330 if (need_commit > 0) {
4331 btrfs_start_delalloc_roots(fs_info, 0, -1);
4332 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4336 trans = btrfs_join_transaction(root);
4338 return PTR_ERR(trans);
4339 if (have_pinned_space >= 0 ||
4340 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4341 &trans->transaction->flags) ||
4343 ret = btrfs_commit_transaction(trans);
4347 * The cleaner kthread might still be doing iput
4348 * operations. Wait for it to finish so that
4349 * more space is released.
4351 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4352 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4355 btrfs_end_transaction(trans);
4359 trace_btrfs_space_reservation(fs_info,
4360 "space_info:enospc",
4361 data_sinfo->flags, bytes, 1);
4364 data_sinfo->bytes_may_use += bytes;
4365 trace_btrfs_space_reservation(fs_info, "space_info",
4366 data_sinfo->flags, bytes, 1);
4367 spin_unlock(&data_sinfo->lock);
4372 int btrfs_check_data_free_space(struct inode *inode,
4373 struct extent_changeset **reserved, u64 start, u64 len)
4375 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4378 /* align the range */
4379 len = round_up(start + len, fs_info->sectorsize) -
4380 round_down(start, fs_info->sectorsize);
4381 start = round_down(start, fs_info->sectorsize);
4383 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4387 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4388 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4390 btrfs_free_reserved_data_space_noquota(inode, start, len);
4397 * Called if we need to clear a data reservation for this inode
4398 * Normally in a error case.
4400 * This one will *NOT* use accurate qgroup reserved space API, just for case
4401 * which we can't sleep and is sure it won't affect qgroup reserved space.
4402 * Like clear_bit_hook().
4404 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4407 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4408 struct btrfs_space_info *data_sinfo;
4410 /* Make sure the range is aligned to sectorsize */
4411 len = round_up(start + len, fs_info->sectorsize) -
4412 round_down(start, fs_info->sectorsize);
4413 start = round_down(start, fs_info->sectorsize);
4415 data_sinfo = fs_info->data_sinfo;
4416 spin_lock(&data_sinfo->lock);
4417 if (WARN_ON(data_sinfo->bytes_may_use < len))
4418 data_sinfo->bytes_may_use = 0;
4420 data_sinfo->bytes_may_use -= len;
4421 trace_btrfs_space_reservation(fs_info, "space_info",
4422 data_sinfo->flags, len, 0);
4423 spin_unlock(&data_sinfo->lock);
4427 * Called if we need to clear a data reservation for this inode
4428 * Normally in a error case.
4430 * This one will handle the per-inode data rsv map for accurate reserved
4433 void btrfs_free_reserved_data_space(struct inode *inode,
4434 struct extent_changeset *reserved, u64 start, u64 len)
4436 struct btrfs_root *root = BTRFS_I(inode)->root;
4438 /* Make sure the range is aligned to sectorsize */
4439 len = round_up(start + len, root->fs_info->sectorsize) -
4440 round_down(start, root->fs_info->sectorsize);
4441 start = round_down(start, root->fs_info->sectorsize);
4443 btrfs_free_reserved_data_space_noquota(inode, start, len);
4444 btrfs_qgroup_free_data(inode, reserved, start, len);
4447 static void force_metadata_allocation(struct btrfs_fs_info *info)
4449 struct list_head *head = &info->space_info;
4450 struct btrfs_space_info *found;
4453 list_for_each_entry_rcu(found, head, list) {
4454 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4455 found->force_alloc = CHUNK_ALLOC_FORCE;
4460 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4462 return (global->size << 1);
4465 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4466 struct btrfs_space_info *sinfo, int force)
4468 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4469 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4472 if (force == CHUNK_ALLOC_FORCE)
4476 * We need to take into account the global rsv because for all intents
4477 * and purposes it's used space. Don't worry about locking the
4478 * global_rsv, it doesn't change except when the transaction commits.
4480 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4481 bytes_used += calc_global_rsv_need_space(global_rsv);
4484 * in limited mode, we want to have some free space up to
4485 * about 1% of the FS size.
4487 if (force == CHUNK_ALLOC_LIMITED) {
4488 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4489 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4491 if (sinfo->total_bytes - bytes_used < thresh)
4495 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4500 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4504 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4505 BTRFS_BLOCK_GROUP_RAID0 |
4506 BTRFS_BLOCK_GROUP_RAID5 |
4507 BTRFS_BLOCK_GROUP_RAID6))
4508 num_dev = fs_info->fs_devices->rw_devices;
4509 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4512 num_dev = 1; /* DUP or single */
4518 * If @is_allocation is true, reserve space in the system space info necessary
4519 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4522 void check_system_chunk(struct btrfs_trans_handle *trans,
4523 struct btrfs_fs_info *fs_info, u64 type)
4525 struct btrfs_space_info *info;
4532 * Needed because we can end up allocating a system chunk and for an
4533 * atomic and race free space reservation in the chunk block reserve.
4535 lockdep_assert_held(&fs_info->chunk_mutex);
4537 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4538 spin_lock(&info->lock);
4539 left = info->total_bytes - btrfs_space_info_used(info, true);
4540 spin_unlock(&info->lock);
4542 num_devs = get_profile_num_devs(fs_info, type);
4544 /* num_devs device items to update and 1 chunk item to add or remove */
4545 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4546 btrfs_calc_trans_metadata_size(fs_info, 1);
4548 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4549 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4550 left, thresh, type);
4551 dump_space_info(fs_info, info, 0, 0);
4554 if (left < thresh) {
4555 u64 flags = btrfs_system_alloc_profile(fs_info);
4558 * Ignore failure to create system chunk. We might end up not
4559 * needing it, as we might not need to COW all nodes/leafs from
4560 * the paths we visit in the chunk tree (they were already COWed
4561 * or created in the current transaction for example).
4563 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4567 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4568 &fs_info->chunk_block_rsv,
4569 thresh, BTRFS_RESERVE_NO_FLUSH);
4571 trans->chunk_bytes_reserved += thresh;
4576 * If force is CHUNK_ALLOC_FORCE:
4577 * - return 1 if it successfully allocates a chunk,
4578 * - return errors including -ENOSPC otherwise.
4579 * If force is NOT CHUNK_ALLOC_FORCE:
4580 * - return 0 if it doesn't need to allocate a new chunk,
4581 * - return 1 if it successfully allocates a chunk,
4582 * - return errors including -ENOSPC otherwise.
4584 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4585 struct btrfs_fs_info *fs_info, u64 flags, int force)
4587 struct btrfs_space_info *space_info;
4588 int wait_for_alloc = 0;
4591 /* Don't re-enter if we're already allocating a chunk */
4592 if (trans->allocating_chunk)
4595 space_info = __find_space_info(fs_info, flags);
4599 spin_lock(&space_info->lock);
4600 if (force < space_info->force_alloc)
4601 force = space_info->force_alloc;
4602 if (space_info->full) {
4603 if (should_alloc_chunk(fs_info, space_info, force))
4607 spin_unlock(&space_info->lock);
4611 if (!should_alloc_chunk(fs_info, space_info, force)) {
4612 spin_unlock(&space_info->lock);
4614 } else if (space_info->chunk_alloc) {
4617 space_info->chunk_alloc = 1;
4620 spin_unlock(&space_info->lock);
4622 mutex_lock(&fs_info->chunk_mutex);
4625 * The chunk_mutex is held throughout the entirety of a chunk
4626 * allocation, so once we've acquired the chunk_mutex we know that the
4627 * other guy is done and we need to recheck and see if we should
4630 if (wait_for_alloc) {
4631 mutex_unlock(&fs_info->chunk_mutex);
4637 trans->allocating_chunk = true;
4640 * If we have mixed data/metadata chunks we want to make sure we keep
4641 * allocating mixed chunks instead of individual chunks.
4643 if (btrfs_mixed_space_info(space_info))
4644 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4647 * if we're doing a data chunk, go ahead and make sure that
4648 * we keep a reasonable number of metadata chunks allocated in the
4651 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4652 fs_info->data_chunk_allocations++;
4653 if (!(fs_info->data_chunk_allocations %
4654 fs_info->metadata_ratio))
4655 force_metadata_allocation(fs_info);
4659 * Check if we have enough space in SYSTEM chunk because we may need
4660 * to update devices.
4662 check_system_chunk(trans, fs_info, flags);
4664 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4665 trans->allocating_chunk = false;
4667 spin_lock(&space_info->lock);
4668 if (ret < 0 && ret != -ENOSPC)
4671 space_info->full = 1;
4675 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4677 space_info->chunk_alloc = 0;
4678 spin_unlock(&space_info->lock);
4679 mutex_unlock(&fs_info->chunk_mutex);
4681 * When we allocate a new chunk we reserve space in the chunk block
4682 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4683 * add new nodes/leafs to it if we end up needing to do it when
4684 * inserting the chunk item and updating device items as part of the
4685 * second phase of chunk allocation, performed by
4686 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4687 * large number of new block groups to create in our transaction
4688 * handle's new_bgs list to avoid exhausting the chunk block reserve
4689 * in extreme cases - like having a single transaction create many new
4690 * block groups when starting to write out the free space caches of all
4691 * the block groups that were made dirty during the lifetime of the
4694 if (trans->can_flush_pending_bgs &&
4695 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4696 btrfs_create_pending_block_groups(trans);
4697 btrfs_trans_release_chunk_metadata(trans);
4702 static int can_overcommit(struct btrfs_fs_info *fs_info,
4703 struct btrfs_space_info *space_info, u64 bytes,
4704 enum btrfs_reserve_flush_enum flush,
4707 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4713 /* Don't overcommit when in mixed mode. */
4714 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4718 profile = btrfs_system_alloc_profile(fs_info);
4720 profile = btrfs_metadata_alloc_profile(fs_info);
4722 used = btrfs_space_info_used(space_info, false);
4725 * We only want to allow over committing if we have lots of actual space
4726 * free, but if we don't have enough space to handle the global reserve
4727 * space then we could end up having a real enospc problem when trying
4728 * to allocate a chunk or some other such important allocation.
4730 spin_lock(&global_rsv->lock);
4731 space_size = calc_global_rsv_need_space(global_rsv);
4732 spin_unlock(&global_rsv->lock);
4733 if (used + space_size >= space_info->total_bytes)
4736 used += space_info->bytes_may_use;
4738 avail = atomic64_read(&fs_info->free_chunk_space);
4741 * If we have dup, raid1 or raid10 then only half of the free
4742 * space is actually useable. For raid56, the space info used
4743 * doesn't include the parity drive, so we don't have to
4746 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4747 BTRFS_BLOCK_GROUP_RAID1 |
4748 BTRFS_BLOCK_GROUP_RAID10))
4752 * If we aren't flushing all things, let us overcommit up to
4753 * 1/2th of the space. If we can flush, don't let us overcommit
4754 * too much, let it overcommit up to 1/8 of the space.
4756 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4761 if (used + bytes < space_info->total_bytes + avail)
4766 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4767 unsigned long nr_pages, int nr_items)
4769 struct super_block *sb = fs_info->sb;
4771 if (down_read_trylock(&sb->s_umount)) {
4772 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4773 up_read(&sb->s_umount);
4776 * We needn't worry the filesystem going from r/w to r/o though
4777 * we don't acquire ->s_umount mutex, because the filesystem
4778 * should guarantee the delalloc inodes list be empty after
4779 * the filesystem is readonly(all dirty pages are written to
4782 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4783 if (!current->journal_info)
4784 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4788 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4794 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4795 nr = div64_u64(to_reclaim, bytes);
4801 #define EXTENT_SIZE_PER_ITEM SZ_256K
4804 * shrink metadata reservation for delalloc
4806 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4807 u64 orig, bool wait_ordered)
4809 struct btrfs_space_info *space_info;
4810 struct btrfs_trans_handle *trans;
4815 unsigned long nr_pages;
4818 /* Calc the number of the pages we need flush for space reservation */
4819 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4820 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4822 trans = (struct btrfs_trans_handle *)current->journal_info;
4823 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4825 delalloc_bytes = percpu_counter_sum_positive(
4826 &fs_info->delalloc_bytes);
4827 if (delalloc_bytes == 0) {
4831 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4836 while (delalloc_bytes && loops < 3) {
4837 max_reclaim = min(delalloc_bytes, to_reclaim);
4838 nr_pages = max_reclaim >> PAGE_SHIFT;
4839 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4841 * We need to wait for the async pages to actually start before
4844 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4848 if (max_reclaim <= nr_pages)
4851 max_reclaim -= nr_pages;
4853 wait_event(fs_info->async_submit_wait,
4854 atomic_read(&fs_info->async_delalloc_pages) <=
4857 spin_lock(&space_info->lock);
4858 if (list_empty(&space_info->tickets) &&
4859 list_empty(&space_info->priority_tickets)) {
4860 spin_unlock(&space_info->lock);
4863 spin_unlock(&space_info->lock);
4866 if (wait_ordered && !trans) {
4867 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4869 time_left = schedule_timeout_killable(1);
4873 delalloc_bytes = percpu_counter_sum_positive(
4874 &fs_info->delalloc_bytes);
4878 struct reserve_ticket {
4881 struct list_head list;
4882 wait_queue_head_t wait;
4886 * maybe_commit_transaction - possibly commit the transaction if its ok to
4887 * @root - the root we're allocating for
4888 * @bytes - the number of bytes we want to reserve
4889 * @force - force the commit
4891 * This will check to make sure that committing the transaction will actually
4892 * get us somewhere and then commit the transaction if it does. Otherwise it
4893 * will return -ENOSPC.
4895 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4896 struct btrfs_space_info *space_info)
4898 struct reserve_ticket *ticket = NULL;
4899 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4900 struct btrfs_trans_handle *trans;
4903 trans = (struct btrfs_trans_handle *)current->journal_info;
4907 spin_lock(&space_info->lock);
4908 if (!list_empty(&space_info->priority_tickets))
4909 ticket = list_first_entry(&space_info->priority_tickets,
4910 struct reserve_ticket, list);
4911 else if (!list_empty(&space_info->tickets))
4912 ticket = list_first_entry(&space_info->tickets,
4913 struct reserve_ticket, list);
4914 bytes = (ticket) ? ticket->bytes : 0;
4915 spin_unlock(&space_info->lock);
4920 /* See if there is enough pinned space to make this reservation */
4921 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4926 * See if there is some space in the delayed insertion reservation for
4929 if (space_info != delayed_rsv->space_info)
4932 spin_lock(&delayed_rsv->lock);
4933 if (delayed_rsv->size > bytes)
4936 bytes -= delayed_rsv->size;
4937 spin_unlock(&delayed_rsv->lock);
4939 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4945 trans = btrfs_join_transaction(fs_info->extent_root);
4949 return btrfs_commit_transaction(trans);
4953 * Try to flush some data based on policy set by @state. This is only advisory
4954 * and may fail for various reasons. The caller is supposed to examine the
4955 * state of @space_info to detect the outcome.
4957 static void flush_space(struct btrfs_fs_info *fs_info,
4958 struct btrfs_space_info *space_info, u64 num_bytes,
4961 struct btrfs_root *root = fs_info->extent_root;
4962 struct btrfs_trans_handle *trans;
4967 case FLUSH_DELAYED_ITEMS_NR:
4968 case FLUSH_DELAYED_ITEMS:
4969 if (state == FLUSH_DELAYED_ITEMS_NR)
4970 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4974 trans = btrfs_join_transaction(root);
4975 if (IS_ERR(trans)) {
4976 ret = PTR_ERR(trans);
4979 ret = btrfs_run_delayed_items_nr(trans, nr);
4980 btrfs_end_transaction(trans);
4982 case FLUSH_DELALLOC:
4983 case FLUSH_DELALLOC_WAIT:
4984 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4985 state == FLUSH_DELALLOC_WAIT);
4988 trans = btrfs_join_transaction(root);
4989 if (IS_ERR(trans)) {
4990 ret = PTR_ERR(trans);
4993 ret = do_chunk_alloc(trans, fs_info,
4994 btrfs_metadata_alloc_profile(fs_info),
4995 CHUNK_ALLOC_NO_FORCE);
4996 btrfs_end_transaction(trans);
4997 if (ret > 0 || ret == -ENOSPC)
5001 ret = may_commit_transaction(fs_info, space_info);
5008 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5014 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5015 struct btrfs_space_info *space_info,
5018 struct reserve_ticket *ticket;
5023 list_for_each_entry(ticket, &space_info->tickets, list)
5024 to_reclaim += ticket->bytes;
5025 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5026 to_reclaim += ticket->bytes;
5030 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5031 if (can_overcommit(fs_info, space_info, to_reclaim,
5032 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5035 used = btrfs_space_info_used(space_info, true);
5037 if (can_overcommit(fs_info, space_info, SZ_1M,
5038 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5039 expected = div_factor_fine(space_info->total_bytes, 95);
5041 expected = div_factor_fine(space_info->total_bytes, 90);
5043 if (used > expected)
5044 to_reclaim = used - expected;
5047 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5048 space_info->bytes_reserved);
5052 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5053 struct btrfs_space_info *space_info,
5054 u64 used, bool system_chunk)
5056 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5058 /* If we're just plain full then async reclaim just slows us down. */
5059 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5062 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5066 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5067 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5070 static void wake_all_tickets(struct list_head *head)
5072 struct reserve_ticket *ticket;
5074 while (!list_empty(head)) {
5075 ticket = list_first_entry(head, struct reserve_ticket, list);
5076 list_del_init(&ticket->list);
5077 ticket->error = -ENOSPC;
5078 wake_up(&ticket->wait);
5083 * This is for normal flushers, we can wait all goddamned day if we want to. We
5084 * will loop and continuously try to flush as long as we are making progress.
5085 * We count progress as clearing off tickets each time we have to loop.
5087 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5089 struct btrfs_fs_info *fs_info;
5090 struct btrfs_space_info *space_info;
5093 int commit_cycles = 0;
5094 u64 last_tickets_id;
5096 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5097 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5099 spin_lock(&space_info->lock);
5100 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5103 space_info->flush = 0;
5104 spin_unlock(&space_info->lock);
5107 last_tickets_id = space_info->tickets_id;
5108 spin_unlock(&space_info->lock);
5110 flush_state = FLUSH_DELAYED_ITEMS_NR;
5112 flush_space(fs_info, space_info, to_reclaim, flush_state);
5113 spin_lock(&space_info->lock);
5114 if (list_empty(&space_info->tickets)) {
5115 space_info->flush = 0;
5116 spin_unlock(&space_info->lock);
5119 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5122 if (last_tickets_id == space_info->tickets_id) {
5125 last_tickets_id = space_info->tickets_id;
5126 flush_state = FLUSH_DELAYED_ITEMS_NR;
5131 if (flush_state > COMMIT_TRANS) {
5133 if (commit_cycles > 2) {
5134 wake_all_tickets(&space_info->tickets);
5135 space_info->flush = 0;
5137 flush_state = FLUSH_DELAYED_ITEMS_NR;
5140 spin_unlock(&space_info->lock);
5141 } while (flush_state <= COMMIT_TRANS);
5144 void btrfs_init_async_reclaim_work(struct work_struct *work)
5146 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5149 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5150 struct btrfs_space_info *space_info,
5151 struct reserve_ticket *ticket)
5154 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5156 spin_lock(&space_info->lock);
5157 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5160 spin_unlock(&space_info->lock);
5163 spin_unlock(&space_info->lock);
5166 flush_space(fs_info, space_info, to_reclaim, flush_state);
5168 spin_lock(&space_info->lock);
5169 if (ticket->bytes == 0) {
5170 spin_unlock(&space_info->lock);
5173 spin_unlock(&space_info->lock);
5176 * Priority flushers can't wait on delalloc without
5179 if (flush_state == FLUSH_DELALLOC ||
5180 flush_state == FLUSH_DELALLOC_WAIT)
5181 flush_state = ALLOC_CHUNK;
5182 } while (flush_state < COMMIT_TRANS);
5185 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5186 struct btrfs_space_info *space_info,
5187 struct reserve_ticket *ticket, u64 orig_bytes)
5193 spin_lock(&space_info->lock);
5194 while (ticket->bytes > 0 && ticket->error == 0) {
5195 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5200 spin_unlock(&space_info->lock);
5204 finish_wait(&ticket->wait, &wait);
5205 spin_lock(&space_info->lock);
5208 ret = ticket->error;
5209 if (!list_empty(&ticket->list))
5210 list_del_init(&ticket->list);
5211 if (ticket->bytes && ticket->bytes < orig_bytes) {
5212 u64 num_bytes = orig_bytes - ticket->bytes;
5213 space_info->bytes_may_use -= num_bytes;
5214 trace_btrfs_space_reservation(fs_info, "space_info",
5215 space_info->flags, num_bytes, 0);
5217 spin_unlock(&space_info->lock);
5223 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5224 * @root - the root we're allocating for
5225 * @space_info - the space info we want to allocate from
5226 * @orig_bytes - the number of bytes we want
5227 * @flush - whether or not we can flush to make our reservation
5229 * This will reserve orig_bytes number of bytes from the space info associated
5230 * with the block_rsv. If there is not enough space it will make an attempt to
5231 * flush out space to make room. It will do this by flushing delalloc if
5232 * possible or committing the transaction. If flush is 0 then no attempts to
5233 * regain reservations will be made and this will fail if there is not enough
5236 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5237 struct btrfs_space_info *space_info,
5239 enum btrfs_reserve_flush_enum flush,
5242 struct reserve_ticket ticket;
5247 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5249 spin_lock(&space_info->lock);
5251 used = btrfs_space_info_used(space_info, true);
5254 * If we have enough space then hooray, make our reservation and carry
5255 * on. If not see if we can overcommit, and if we can, hooray carry on.
5256 * If not things get more complicated.
5258 if (used + orig_bytes <= space_info->total_bytes) {
5259 space_info->bytes_may_use += orig_bytes;
5260 trace_btrfs_space_reservation(fs_info, "space_info",
5261 space_info->flags, orig_bytes, 1);
5263 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5265 space_info->bytes_may_use += orig_bytes;
5266 trace_btrfs_space_reservation(fs_info, "space_info",
5267 space_info->flags, orig_bytes, 1);
5272 * If we couldn't make a reservation then setup our reservation ticket
5273 * and kick the async worker if it's not already running.
5275 * If we are a priority flusher then we just need to add our ticket to
5276 * the list and we will do our own flushing further down.
5278 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5279 ticket.bytes = orig_bytes;
5281 init_waitqueue_head(&ticket.wait);
5282 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5283 list_add_tail(&ticket.list, &space_info->tickets);
5284 if (!space_info->flush) {
5285 space_info->flush = 1;
5286 trace_btrfs_trigger_flush(fs_info,
5290 queue_work(system_unbound_wq,
5291 &fs_info->async_reclaim_work);
5294 list_add_tail(&ticket.list,
5295 &space_info->priority_tickets);
5297 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5300 * We will do the space reservation dance during log replay,
5301 * which means we won't have fs_info->fs_root set, so don't do
5302 * the async reclaim as we will panic.
5304 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5305 need_do_async_reclaim(fs_info, space_info,
5306 used, system_chunk) &&
5307 !work_busy(&fs_info->async_reclaim_work)) {
5308 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5309 orig_bytes, flush, "preempt");
5310 queue_work(system_unbound_wq,
5311 &fs_info->async_reclaim_work);
5314 spin_unlock(&space_info->lock);
5315 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5318 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5319 return wait_reserve_ticket(fs_info, space_info, &ticket,
5323 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5324 spin_lock(&space_info->lock);
5326 if (ticket.bytes < orig_bytes) {
5327 u64 num_bytes = orig_bytes - ticket.bytes;
5328 space_info->bytes_may_use -= num_bytes;
5329 trace_btrfs_space_reservation(fs_info, "space_info",
5334 list_del_init(&ticket.list);
5337 spin_unlock(&space_info->lock);
5338 ASSERT(list_empty(&ticket.list));
5343 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5344 * @root - the root we're allocating for
5345 * @block_rsv - the block_rsv we're allocating for
5346 * @orig_bytes - the number of bytes we want
5347 * @flush - whether or not we can flush to make our reservation
5349 * This will reserve orgi_bytes number of bytes from the space info associated
5350 * with the block_rsv. If there is not enough space it will make an attempt to
5351 * flush out space to make room. It will do this by flushing delalloc if
5352 * possible or committing the transaction. If flush is 0 then no attempts to
5353 * regain reservations will be made and this will fail if there is not enough
5356 static int reserve_metadata_bytes(struct btrfs_root *root,
5357 struct btrfs_block_rsv *block_rsv,
5359 enum btrfs_reserve_flush_enum flush)
5361 struct btrfs_fs_info *fs_info = root->fs_info;
5362 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5364 bool system_chunk = (root == fs_info->chunk_root);
5366 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5367 orig_bytes, flush, system_chunk);
5368 if (ret == -ENOSPC &&
5369 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5370 if (block_rsv != global_rsv &&
5371 !block_rsv_use_bytes(global_rsv, orig_bytes))
5374 if (ret == -ENOSPC) {
5375 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5376 block_rsv->space_info->flags,
5379 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5380 dump_space_info(fs_info, block_rsv->space_info,
5386 static struct btrfs_block_rsv *get_block_rsv(
5387 const struct btrfs_trans_handle *trans,
5388 const struct btrfs_root *root)
5390 struct btrfs_fs_info *fs_info = root->fs_info;
5391 struct btrfs_block_rsv *block_rsv = NULL;
5393 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5394 (root == fs_info->csum_root && trans->adding_csums) ||
5395 (root == fs_info->uuid_root))
5396 block_rsv = trans->block_rsv;
5399 block_rsv = root->block_rsv;
5402 block_rsv = &fs_info->empty_block_rsv;
5407 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5411 spin_lock(&block_rsv->lock);
5412 if (block_rsv->reserved >= num_bytes) {
5413 block_rsv->reserved -= num_bytes;
5414 if (block_rsv->reserved < block_rsv->size)
5415 block_rsv->full = 0;
5418 spin_unlock(&block_rsv->lock);
5422 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5423 u64 num_bytes, int update_size)
5425 spin_lock(&block_rsv->lock);
5426 block_rsv->reserved += num_bytes;
5428 block_rsv->size += num_bytes;
5429 else if (block_rsv->reserved >= block_rsv->size)
5430 block_rsv->full = 1;
5431 spin_unlock(&block_rsv->lock);
5434 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5435 struct btrfs_block_rsv *dest, u64 num_bytes,
5438 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5441 if (global_rsv->space_info != dest->space_info)
5444 spin_lock(&global_rsv->lock);
5445 min_bytes = div_factor(global_rsv->size, min_factor);
5446 if (global_rsv->reserved < min_bytes + num_bytes) {
5447 spin_unlock(&global_rsv->lock);
5450 global_rsv->reserved -= num_bytes;
5451 if (global_rsv->reserved < global_rsv->size)
5452 global_rsv->full = 0;
5453 spin_unlock(&global_rsv->lock);
5455 block_rsv_add_bytes(dest, num_bytes, 1);
5460 * This is for space we already have accounted in space_info->bytes_may_use, so
5461 * basically when we're returning space from block_rsv's.
5463 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5464 struct btrfs_space_info *space_info,
5467 struct reserve_ticket *ticket;
5468 struct list_head *head;
5470 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5471 bool check_overcommit = false;
5473 spin_lock(&space_info->lock);
5474 head = &space_info->priority_tickets;
5477 * If we are over our limit then we need to check and see if we can
5478 * overcommit, and if we can't then we just need to free up our space
5479 * and not satisfy any requests.
5481 used = btrfs_space_info_used(space_info, true);
5482 if (used - num_bytes >= space_info->total_bytes)
5483 check_overcommit = true;
5485 while (!list_empty(head) && num_bytes) {
5486 ticket = list_first_entry(head, struct reserve_ticket,
5489 * We use 0 bytes because this space is already reserved, so
5490 * adding the ticket space would be a double count.
5492 if (check_overcommit &&
5493 !can_overcommit(fs_info, space_info, 0, flush, false))
5495 if (num_bytes >= ticket->bytes) {
5496 list_del_init(&ticket->list);
5497 num_bytes -= ticket->bytes;
5499 space_info->tickets_id++;
5500 wake_up(&ticket->wait);
5502 ticket->bytes -= num_bytes;
5507 if (num_bytes && head == &space_info->priority_tickets) {
5508 head = &space_info->tickets;
5509 flush = BTRFS_RESERVE_FLUSH_ALL;
5512 space_info->bytes_may_use -= num_bytes;
5513 trace_btrfs_space_reservation(fs_info, "space_info",
5514 space_info->flags, num_bytes, 0);
5515 spin_unlock(&space_info->lock);
5519 * This is for newly allocated space that isn't accounted in
5520 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5521 * we use this helper.
5523 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5524 struct btrfs_space_info *space_info,
5527 struct reserve_ticket *ticket;
5528 struct list_head *head = &space_info->priority_tickets;
5531 while (!list_empty(head) && num_bytes) {
5532 ticket = list_first_entry(head, struct reserve_ticket,
5534 if (num_bytes >= ticket->bytes) {
5535 trace_btrfs_space_reservation(fs_info, "space_info",
5538 list_del_init(&ticket->list);
5539 num_bytes -= ticket->bytes;
5540 space_info->bytes_may_use += ticket->bytes;
5542 space_info->tickets_id++;
5543 wake_up(&ticket->wait);
5545 trace_btrfs_space_reservation(fs_info, "space_info",
5548 space_info->bytes_may_use += num_bytes;
5549 ticket->bytes -= num_bytes;
5554 if (num_bytes && head == &space_info->priority_tickets) {
5555 head = &space_info->tickets;
5560 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5561 struct btrfs_block_rsv *block_rsv,
5562 struct btrfs_block_rsv *dest, u64 num_bytes)
5564 struct btrfs_space_info *space_info = block_rsv->space_info;
5567 spin_lock(&block_rsv->lock);
5568 if (num_bytes == (u64)-1)
5569 num_bytes = block_rsv->size;
5570 block_rsv->size -= num_bytes;
5571 if (block_rsv->reserved >= block_rsv->size) {
5572 num_bytes = block_rsv->reserved - block_rsv->size;
5573 block_rsv->reserved = block_rsv->size;
5574 block_rsv->full = 1;
5578 spin_unlock(&block_rsv->lock);
5581 if (num_bytes > 0) {
5583 spin_lock(&dest->lock);
5587 bytes_to_add = dest->size - dest->reserved;
5588 bytes_to_add = min(num_bytes, bytes_to_add);
5589 dest->reserved += bytes_to_add;
5590 if (dest->reserved >= dest->size)
5592 num_bytes -= bytes_to_add;
5594 spin_unlock(&dest->lock);
5597 space_info_add_old_bytes(fs_info, space_info,
5603 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5604 struct btrfs_block_rsv *dst, u64 num_bytes,
5609 ret = block_rsv_use_bytes(src, num_bytes);
5613 block_rsv_add_bytes(dst, num_bytes, update_size);
5617 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5619 memset(rsv, 0, sizeof(*rsv));
5620 spin_lock_init(&rsv->lock);
5624 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5625 struct btrfs_block_rsv *rsv,
5626 unsigned short type)
5628 btrfs_init_block_rsv(rsv, type);
5629 rsv->space_info = __find_space_info(fs_info,
5630 BTRFS_BLOCK_GROUP_METADATA);
5633 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5634 unsigned short type)
5636 struct btrfs_block_rsv *block_rsv;
5638 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5642 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5646 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5647 struct btrfs_block_rsv *rsv)
5651 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5655 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5660 int btrfs_block_rsv_add(struct btrfs_root *root,
5661 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5662 enum btrfs_reserve_flush_enum flush)
5669 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5671 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5678 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5686 spin_lock(&block_rsv->lock);
5687 num_bytes = div_factor(block_rsv->size, min_factor);
5688 if (block_rsv->reserved >= num_bytes)
5690 spin_unlock(&block_rsv->lock);
5695 int btrfs_block_rsv_refill(struct btrfs_root *root,
5696 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5697 enum btrfs_reserve_flush_enum flush)
5705 spin_lock(&block_rsv->lock);
5706 num_bytes = min_reserved;
5707 if (block_rsv->reserved >= num_bytes)
5710 num_bytes -= block_rsv->reserved;
5711 spin_unlock(&block_rsv->lock);
5716 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5718 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5726 * btrfs_inode_rsv_refill - refill the inode block rsv.
5727 * @inode - the inode we are refilling.
5728 * @flush - the flusing restriction.
5730 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5731 * block_rsv->size as the minimum size. We'll either refill the missing amount
5732 * or return if we already have enough space. This will also handle the resreve
5733 * tracepoint for the reserved amount.
5735 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5736 enum btrfs_reserve_flush_enum flush)
5738 struct btrfs_root *root = inode->root;
5739 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5743 spin_lock(&block_rsv->lock);
5744 if (block_rsv->reserved < block_rsv->size)
5745 num_bytes = block_rsv->size - block_rsv->reserved;
5746 spin_unlock(&block_rsv->lock);
5751 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
5754 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5756 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5757 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5758 btrfs_ino(inode), num_bytes, 1);
5764 * btrfs_inode_rsv_release - release any excessive reservation.
5765 * @inode - the inode we need to release from.
5766 * @qgroup_free - free or convert qgroup meta.
5767 * Unlike normal operation, qgroup meta reservation needs to know if we are
5768 * freeing qgroup reservation or just converting it into per-trans. Normally
5769 * @qgroup_free is true for error handling, and false for normal release.
5771 * This is the same as btrfs_block_rsv_release, except that it handles the
5772 * tracepoint for the reservation.
5774 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5776 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5777 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5778 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5782 * Since we statically set the block_rsv->size we just want to say we
5783 * are releasing 0 bytes, and then we'll just get the reservation over
5786 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0);
5788 trace_btrfs_space_reservation(fs_info, "delalloc",
5789 btrfs_ino(inode), released, 0);
5791 btrfs_qgroup_free_meta_prealloc(inode->root, released);
5793 btrfs_qgroup_convert_reserved_meta(inode->root, released);
5796 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5797 struct btrfs_block_rsv *block_rsv,
5800 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5802 if (global_rsv == block_rsv ||
5803 block_rsv->space_info != global_rsv->space_info)
5805 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5808 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5810 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5811 struct btrfs_space_info *sinfo = block_rsv->space_info;
5815 * The global block rsv is based on the size of the extent tree, the
5816 * checksum tree and the root tree. If the fs is empty we want to set
5817 * it to a minimal amount for safety.
5819 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5820 btrfs_root_used(&fs_info->csum_root->root_item) +
5821 btrfs_root_used(&fs_info->tree_root->root_item);
5822 num_bytes = max_t(u64, num_bytes, SZ_16M);
5824 spin_lock(&sinfo->lock);
5825 spin_lock(&block_rsv->lock);
5827 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5829 if (block_rsv->reserved < block_rsv->size) {
5830 num_bytes = btrfs_space_info_used(sinfo, true);
5831 if (sinfo->total_bytes > num_bytes) {
5832 num_bytes = sinfo->total_bytes - num_bytes;
5833 num_bytes = min(num_bytes,
5834 block_rsv->size - block_rsv->reserved);
5835 block_rsv->reserved += num_bytes;
5836 sinfo->bytes_may_use += num_bytes;
5837 trace_btrfs_space_reservation(fs_info, "space_info",
5838 sinfo->flags, num_bytes,
5841 } else if (block_rsv->reserved > block_rsv->size) {
5842 num_bytes = block_rsv->reserved - block_rsv->size;
5843 sinfo->bytes_may_use -= num_bytes;
5844 trace_btrfs_space_reservation(fs_info, "space_info",
5845 sinfo->flags, num_bytes, 0);
5846 block_rsv->reserved = block_rsv->size;
5849 if (block_rsv->reserved == block_rsv->size)
5850 block_rsv->full = 1;
5852 block_rsv->full = 0;
5854 spin_unlock(&block_rsv->lock);
5855 spin_unlock(&sinfo->lock);
5858 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5860 struct btrfs_space_info *space_info;
5862 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5863 fs_info->chunk_block_rsv.space_info = space_info;
5865 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5866 fs_info->global_block_rsv.space_info = space_info;
5867 fs_info->trans_block_rsv.space_info = space_info;
5868 fs_info->empty_block_rsv.space_info = space_info;
5869 fs_info->delayed_block_rsv.space_info = space_info;
5871 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5872 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5873 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5874 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5875 if (fs_info->quota_root)
5876 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5877 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5879 update_global_block_rsv(fs_info);
5882 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5884 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5886 WARN_ON(fs_info->trans_block_rsv.size > 0);
5887 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5888 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5889 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5890 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5891 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5896 * To be called after all the new block groups attached to the transaction
5897 * handle have been created (btrfs_create_pending_block_groups()).
5899 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5901 struct btrfs_fs_info *fs_info = trans->fs_info;
5903 if (!trans->chunk_bytes_reserved)
5906 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5908 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5909 trans->chunk_bytes_reserved);
5910 trans->chunk_bytes_reserved = 0;
5913 /* Can only return 0 or -ENOSPC */
5914 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5915 struct btrfs_inode *inode)
5917 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5918 struct btrfs_root *root = inode->root;
5920 * We always use trans->block_rsv here as we will have reserved space
5921 * for our orphan when starting the transaction, using get_block_rsv()
5922 * here will sometimes make us choose the wrong block rsv as we could be
5923 * doing a reloc inode for a non refcounted root.
5925 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5926 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5929 * We need to hold space in order to delete our orphan item once we've
5930 * added it, so this takes the reservation so we can release it later
5931 * when we are truly done with the orphan item.
5933 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5935 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5937 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5940 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5942 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5943 struct btrfs_root *root = inode->root;
5944 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5946 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5948 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5952 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5953 * root: the root of the parent directory
5954 * rsv: block reservation
5955 * items: the number of items that we need do reservation
5956 * qgroup_reserved: used to return the reserved size in qgroup
5958 * This function is used to reserve the space for snapshot/subvolume
5959 * creation and deletion. Those operations are different with the
5960 * common file/directory operations, they change two fs/file trees
5961 * and root tree, the number of items that the qgroup reserves is
5962 * different with the free space reservation. So we can not use
5963 * the space reservation mechanism in start_transaction().
5965 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5966 struct btrfs_block_rsv *rsv,
5968 u64 *qgroup_reserved,
5969 bool use_global_rsv)
5973 struct btrfs_fs_info *fs_info = root->fs_info;
5974 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5976 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5977 /* One for parent inode, two for dir entries */
5978 num_bytes = 3 * fs_info->nodesize;
5979 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
5986 *qgroup_reserved = num_bytes;
5988 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5989 rsv->space_info = __find_space_info(fs_info,
5990 BTRFS_BLOCK_GROUP_METADATA);
5991 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5992 BTRFS_RESERVE_FLUSH_ALL);
5994 if (ret == -ENOSPC && use_global_rsv)
5995 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5997 if (ret && *qgroup_reserved)
5998 btrfs_qgroup_free_meta_prealloc(root, *qgroup_reserved);
6003 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6004 struct btrfs_block_rsv *rsv)
6006 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6009 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6010 struct btrfs_inode *inode)
6012 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6013 u64 reserve_size = 0;
6015 unsigned outstanding_extents;
6017 lockdep_assert_held(&inode->lock);
6018 outstanding_extents = inode->outstanding_extents;
6019 if (outstanding_extents)
6020 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6021 outstanding_extents + 1);
6022 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6024 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6027 spin_lock(&block_rsv->lock);
6028 block_rsv->size = reserve_size;
6029 spin_unlock(&block_rsv->lock);
6032 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6034 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6035 unsigned nr_extents;
6036 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6038 bool delalloc_lock = true;
6040 /* If we are a free space inode we need to not flush since we will be in
6041 * the middle of a transaction commit. We also don't need the delalloc
6042 * mutex since we won't race with anybody. We need this mostly to make
6043 * lockdep shut its filthy mouth.
6045 * If we have a transaction open (can happen if we call truncate_block
6046 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6048 if (btrfs_is_free_space_inode(inode)) {
6049 flush = BTRFS_RESERVE_NO_FLUSH;
6050 delalloc_lock = false;
6052 if (current->journal_info)
6053 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6055 if (btrfs_transaction_in_commit(fs_info))
6056 schedule_timeout(1);
6060 mutex_lock(&inode->delalloc_mutex);
6062 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6064 /* Add our new extents and calculate the new rsv size. */
6065 spin_lock(&inode->lock);
6066 nr_extents = count_max_extents(num_bytes);
6067 btrfs_mod_outstanding_extents(inode, nr_extents);
6068 inode->csum_bytes += num_bytes;
6069 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6070 spin_unlock(&inode->lock);
6072 ret = btrfs_inode_rsv_refill(inode, flush);
6077 mutex_unlock(&inode->delalloc_mutex);
6081 spin_lock(&inode->lock);
6082 nr_extents = count_max_extents(num_bytes);
6083 btrfs_mod_outstanding_extents(inode, -nr_extents);
6084 inode->csum_bytes -= num_bytes;
6085 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6086 spin_unlock(&inode->lock);
6088 btrfs_inode_rsv_release(inode, true);
6090 mutex_unlock(&inode->delalloc_mutex);
6095 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6096 * @inode: the inode to release the reservation for.
6097 * @num_bytes: the number of bytes we are releasing.
6098 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6100 * This will release the metadata reservation for an inode. This can be called
6101 * once we complete IO for a given set of bytes to release their metadata
6102 * reservations, or on error for the same reason.
6104 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6107 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6109 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6110 spin_lock(&inode->lock);
6111 inode->csum_bytes -= num_bytes;
6112 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6113 spin_unlock(&inode->lock);
6115 if (btrfs_is_testing(fs_info))
6118 btrfs_inode_rsv_release(inode, qgroup_free);
6122 * btrfs_delalloc_release_extents - release our outstanding_extents
6123 * @inode: the inode to balance the reservation for.
6124 * @num_bytes: the number of bytes we originally reserved with
6125 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6127 * When we reserve space we increase outstanding_extents for the extents we may
6128 * add. Once we've set the range as delalloc or created our ordered extents we
6129 * have outstanding_extents to track the real usage, so we use this to free our
6130 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6131 * with btrfs_delalloc_reserve_metadata.
6133 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6136 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6137 unsigned num_extents;
6139 spin_lock(&inode->lock);
6140 num_extents = count_max_extents(num_bytes);
6141 btrfs_mod_outstanding_extents(inode, -num_extents);
6142 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6143 spin_unlock(&inode->lock);
6145 if (btrfs_is_testing(fs_info))
6148 btrfs_inode_rsv_release(inode, qgroup_free);
6152 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6154 * @inode: inode we're writing to
6155 * @start: start range we are writing to
6156 * @len: how long the range we are writing to
6157 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6158 * current reservation.
6160 * This will do the following things
6162 * o reserve space in data space info for num bytes
6163 * and reserve precious corresponding qgroup space
6164 * (Done in check_data_free_space)
6166 * o reserve space for metadata space, based on the number of outstanding
6167 * extents and how much csums will be needed
6168 * also reserve metadata space in a per root over-reserve method.
6169 * o add to the inodes->delalloc_bytes
6170 * o add it to the fs_info's delalloc inodes list.
6171 * (Above 3 all done in delalloc_reserve_metadata)
6173 * Return 0 for success
6174 * Return <0 for error(-ENOSPC or -EQUOT)
6176 int btrfs_delalloc_reserve_space(struct inode *inode,
6177 struct extent_changeset **reserved, u64 start, u64 len)
6181 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6184 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6186 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6191 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6192 * @inode: inode we're releasing space for
6193 * @start: start position of the space already reserved
6194 * @len: the len of the space already reserved
6195 * @release_bytes: the len of the space we consumed or didn't use
6197 * This function will release the metadata space that was not used and will
6198 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6199 * list if there are no delalloc bytes left.
6200 * Also it will handle the qgroup reserved space.
6202 void btrfs_delalloc_release_space(struct inode *inode,
6203 struct extent_changeset *reserved,
6204 u64 start, u64 len, bool qgroup_free)
6206 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6207 btrfs_free_reserved_data_space(inode, reserved, start, len);
6210 static int update_block_group(struct btrfs_trans_handle *trans,
6211 struct btrfs_fs_info *info, u64 bytenr,
6212 u64 num_bytes, int alloc)
6214 struct btrfs_block_group_cache *cache = NULL;
6215 u64 total = num_bytes;
6220 /* block accounting for super block */
6221 spin_lock(&info->delalloc_root_lock);
6222 old_val = btrfs_super_bytes_used(info->super_copy);
6224 old_val += num_bytes;
6226 old_val -= num_bytes;
6227 btrfs_set_super_bytes_used(info->super_copy, old_val);
6228 spin_unlock(&info->delalloc_root_lock);
6231 cache = btrfs_lookup_block_group(info, bytenr);
6234 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6235 BTRFS_BLOCK_GROUP_RAID1 |
6236 BTRFS_BLOCK_GROUP_RAID10))
6241 * If this block group has free space cache written out, we
6242 * need to make sure to load it if we are removing space. This
6243 * is because we need the unpinning stage to actually add the
6244 * space back to the block group, otherwise we will leak space.
6246 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6247 cache_block_group(cache, 1);
6249 byte_in_group = bytenr - cache->key.objectid;
6250 WARN_ON(byte_in_group > cache->key.offset);
6252 spin_lock(&cache->space_info->lock);
6253 spin_lock(&cache->lock);
6255 if (btrfs_test_opt(info, SPACE_CACHE) &&
6256 cache->disk_cache_state < BTRFS_DC_CLEAR)
6257 cache->disk_cache_state = BTRFS_DC_CLEAR;
6259 old_val = btrfs_block_group_used(&cache->item);
6260 num_bytes = min(total, cache->key.offset - byte_in_group);
6262 old_val += num_bytes;
6263 btrfs_set_block_group_used(&cache->item, old_val);
6264 cache->reserved -= num_bytes;
6265 cache->space_info->bytes_reserved -= num_bytes;
6266 cache->space_info->bytes_used += num_bytes;
6267 cache->space_info->disk_used += num_bytes * factor;
6268 spin_unlock(&cache->lock);
6269 spin_unlock(&cache->space_info->lock);
6271 old_val -= num_bytes;
6272 btrfs_set_block_group_used(&cache->item, old_val);
6273 cache->pinned += num_bytes;
6274 cache->space_info->bytes_pinned += num_bytes;
6275 cache->space_info->bytes_used -= num_bytes;
6276 cache->space_info->disk_used -= num_bytes * factor;
6277 spin_unlock(&cache->lock);
6278 spin_unlock(&cache->space_info->lock);
6280 trace_btrfs_space_reservation(info, "pinned",
6281 cache->space_info->flags,
6283 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6285 set_extent_dirty(info->pinned_extents,
6286 bytenr, bytenr + num_bytes - 1,
6287 GFP_NOFS | __GFP_NOFAIL);
6290 spin_lock(&trans->transaction->dirty_bgs_lock);
6291 if (list_empty(&cache->dirty_list)) {
6292 list_add_tail(&cache->dirty_list,
6293 &trans->transaction->dirty_bgs);
6294 trans->transaction->num_dirty_bgs++;
6295 btrfs_get_block_group(cache);
6297 spin_unlock(&trans->transaction->dirty_bgs_lock);
6300 * No longer have used bytes in this block group, queue it for
6301 * deletion. We do this after adding the block group to the
6302 * dirty list to avoid races between cleaner kthread and space
6305 if (!alloc && old_val == 0) {
6306 spin_lock(&info->unused_bgs_lock);
6307 if (list_empty(&cache->bg_list)) {
6308 btrfs_get_block_group(cache);
6309 list_add_tail(&cache->bg_list,
6312 spin_unlock(&info->unused_bgs_lock);
6315 btrfs_put_block_group(cache);
6317 bytenr += num_bytes;
6322 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6324 struct btrfs_block_group_cache *cache;
6327 spin_lock(&fs_info->block_group_cache_lock);
6328 bytenr = fs_info->first_logical_byte;
6329 spin_unlock(&fs_info->block_group_cache_lock);
6331 if (bytenr < (u64)-1)
6334 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6338 bytenr = cache->key.objectid;
6339 btrfs_put_block_group(cache);
6344 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6345 struct btrfs_block_group_cache *cache,
6346 u64 bytenr, u64 num_bytes, int reserved)
6348 spin_lock(&cache->space_info->lock);
6349 spin_lock(&cache->lock);
6350 cache->pinned += num_bytes;
6351 cache->space_info->bytes_pinned += num_bytes;
6353 cache->reserved -= num_bytes;
6354 cache->space_info->bytes_reserved -= num_bytes;
6356 spin_unlock(&cache->lock);
6357 spin_unlock(&cache->space_info->lock);
6359 trace_btrfs_space_reservation(fs_info, "pinned",
6360 cache->space_info->flags, num_bytes, 1);
6361 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6362 set_extent_dirty(fs_info->pinned_extents, bytenr,
6363 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6368 * this function must be called within transaction
6370 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6371 u64 bytenr, u64 num_bytes, int reserved)
6373 struct btrfs_block_group_cache *cache;
6375 cache = btrfs_lookup_block_group(fs_info, bytenr);
6376 BUG_ON(!cache); /* Logic error */
6378 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6380 btrfs_put_block_group(cache);
6385 * this function must be called within transaction
6387 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6388 u64 bytenr, u64 num_bytes)
6390 struct btrfs_block_group_cache *cache;
6393 cache = btrfs_lookup_block_group(fs_info, bytenr);
6398 * pull in the free space cache (if any) so that our pin
6399 * removes the free space from the cache. We have load_only set
6400 * to one because the slow code to read in the free extents does check
6401 * the pinned extents.
6403 cache_block_group(cache, 1);
6405 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6407 /* remove us from the free space cache (if we're there at all) */
6408 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6409 btrfs_put_block_group(cache);
6413 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6414 u64 start, u64 num_bytes)
6417 struct btrfs_block_group_cache *block_group;
6418 struct btrfs_caching_control *caching_ctl;
6420 block_group = btrfs_lookup_block_group(fs_info, start);
6424 cache_block_group(block_group, 0);
6425 caching_ctl = get_caching_control(block_group);
6429 BUG_ON(!block_group_cache_done(block_group));
6430 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6432 mutex_lock(&caching_ctl->mutex);
6434 if (start >= caching_ctl->progress) {
6435 ret = add_excluded_extent(fs_info, start, num_bytes);
6436 } else if (start + num_bytes <= caching_ctl->progress) {
6437 ret = btrfs_remove_free_space(block_group,
6440 num_bytes = caching_ctl->progress - start;
6441 ret = btrfs_remove_free_space(block_group,
6446 num_bytes = (start + num_bytes) -
6447 caching_ctl->progress;
6448 start = caching_ctl->progress;
6449 ret = add_excluded_extent(fs_info, start, num_bytes);
6452 mutex_unlock(&caching_ctl->mutex);
6453 put_caching_control(caching_ctl);
6455 btrfs_put_block_group(block_group);
6459 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6460 struct extent_buffer *eb)
6462 struct btrfs_file_extent_item *item;
6463 struct btrfs_key key;
6467 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6470 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6471 btrfs_item_key_to_cpu(eb, &key, i);
6472 if (key.type != BTRFS_EXTENT_DATA_KEY)
6474 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6475 found_type = btrfs_file_extent_type(eb, item);
6476 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6478 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6480 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6481 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6482 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6489 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6491 atomic_inc(&bg->reservations);
6494 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6497 struct btrfs_block_group_cache *bg;
6499 bg = btrfs_lookup_block_group(fs_info, start);
6501 if (atomic_dec_and_test(&bg->reservations))
6502 wake_up_var(&bg->reservations);
6503 btrfs_put_block_group(bg);
6506 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6508 struct btrfs_space_info *space_info = bg->space_info;
6512 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6516 * Our block group is read only but before we set it to read only,
6517 * some task might have had allocated an extent from it already, but it
6518 * has not yet created a respective ordered extent (and added it to a
6519 * root's list of ordered extents).
6520 * Therefore wait for any task currently allocating extents, since the
6521 * block group's reservations counter is incremented while a read lock
6522 * on the groups' semaphore is held and decremented after releasing
6523 * the read access on that semaphore and creating the ordered extent.
6525 down_write(&space_info->groups_sem);
6526 up_write(&space_info->groups_sem);
6528 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6532 * btrfs_add_reserved_bytes - update the block_group and space info counters
6533 * @cache: The cache we are manipulating
6534 * @ram_bytes: The number of bytes of file content, and will be same to
6535 * @num_bytes except for the compress path.
6536 * @num_bytes: The number of bytes in question
6537 * @delalloc: The blocks are allocated for the delalloc write
6539 * This is called by the allocator when it reserves space. If this is a
6540 * reservation and the block group has become read only we cannot make the
6541 * reservation and return -EAGAIN, otherwise this function always succeeds.
6543 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6544 u64 ram_bytes, u64 num_bytes, int delalloc)
6546 struct btrfs_space_info *space_info = cache->space_info;
6549 spin_lock(&space_info->lock);
6550 spin_lock(&cache->lock);
6554 cache->reserved += num_bytes;
6555 space_info->bytes_reserved += num_bytes;
6557 trace_btrfs_space_reservation(cache->fs_info,
6558 "space_info", space_info->flags,
6560 space_info->bytes_may_use -= ram_bytes;
6562 cache->delalloc_bytes += num_bytes;
6564 spin_unlock(&cache->lock);
6565 spin_unlock(&space_info->lock);
6570 * btrfs_free_reserved_bytes - update the block_group and space info counters
6571 * @cache: The cache we are manipulating
6572 * @num_bytes: The number of bytes in question
6573 * @delalloc: The blocks are allocated for the delalloc write
6575 * This is called by somebody who is freeing space that was never actually used
6576 * on disk. For example if you reserve some space for a new leaf in transaction
6577 * A and before transaction A commits you free that leaf, you call this with
6578 * reserve set to 0 in order to clear the reservation.
6581 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6582 u64 num_bytes, int delalloc)
6584 struct btrfs_space_info *space_info = cache->space_info;
6587 spin_lock(&space_info->lock);
6588 spin_lock(&cache->lock);
6590 space_info->bytes_readonly += num_bytes;
6591 cache->reserved -= num_bytes;
6592 space_info->bytes_reserved -= num_bytes;
6595 cache->delalloc_bytes -= num_bytes;
6596 spin_unlock(&cache->lock);
6597 spin_unlock(&space_info->lock);
6600 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6602 struct btrfs_caching_control *next;
6603 struct btrfs_caching_control *caching_ctl;
6604 struct btrfs_block_group_cache *cache;
6606 down_write(&fs_info->commit_root_sem);
6608 list_for_each_entry_safe(caching_ctl, next,
6609 &fs_info->caching_block_groups, list) {
6610 cache = caching_ctl->block_group;
6611 if (block_group_cache_done(cache)) {
6612 cache->last_byte_to_unpin = (u64)-1;
6613 list_del_init(&caching_ctl->list);
6614 put_caching_control(caching_ctl);
6616 cache->last_byte_to_unpin = caching_ctl->progress;
6620 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6621 fs_info->pinned_extents = &fs_info->freed_extents[1];
6623 fs_info->pinned_extents = &fs_info->freed_extents[0];
6625 up_write(&fs_info->commit_root_sem);
6627 update_global_block_rsv(fs_info);
6631 * Returns the free cluster for the given space info and sets empty_cluster to
6632 * what it should be based on the mount options.
6634 static struct btrfs_free_cluster *
6635 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6636 struct btrfs_space_info *space_info, u64 *empty_cluster)
6638 struct btrfs_free_cluster *ret = NULL;
6641 if (btrfs_mixed_space_info(space_info))
6644 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6645 ret = &fs_info->meta_alloc_cluster;
6646 if (btrfs_test_opt(fs_info, SSD))
6647 *empty_cluster = SZ_2M;
6649 *empty_cluster = SZ_64K;
6650 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6651 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6652 *empty_cluster = SZ_2M;
6653 ret = &fs_info->data_alloc_cluster;
6659 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6661 const bool return_free_space)
6663 struct btrfs_block_group_cache *cache = NULL;
6664 struct btrfs_space_info *space_info;
6665 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6666 struct btrfs_free_cluster *cluster = NULL;
6668 u64 total_unpinned = 0;
6669 u64 empty_cluster = 0;
6672 while (start <= end) {
6675 start >= cache->key.objectid + cache->key.offset) {
6677 btrfs_put_block_group(cache);
6679 cache = btrfs_lookup_block_group(fs_info, start);
6680 BUG_ON(!cache); /* Logic error */
6682 cluster = fetch_cluster_info(fs_info,
6685 empty_cluster <<= 1;
6688 len = cache->key.objectid + cache->key.offset - start;
6689 len = min(len, end + 1 - start);
6691 if (start < cache->last_byte_to_unpin) {
6692 len = min(len, cache->last_byte_to_unpin - start);
6693 if (return_free_space)
6694 btrfs_add_free_space(cache, start, len);
6698 total_unpinned += len;
6699 space_info = cache->space_info;
6702 * If this space cluster has been marked as fragmented and we've
6703 * unpinned enough in this block group to potentially allow a
6704 * cluster to be created inside of it go ahead and clear the
6707 if (cluster && cluster->fragmented &&
6708 total_unpinned > empty_cluster) {
6709 spin_lock(&cluster->lock);
6710 cluster->fragmented = 0;
6711 spin_unlock(&cluster->lock);
6714 spin_lock(&space_info->lock);
6715 spin_lock(&cache->lock);
6716 cache->pinned -= len;
6717 space_info->bytes_pinned -= len;
6719 trace_btrfs_space_reservation(fs_info, "pinned",
6720 space_info->flags, len, 0);
6721 space_info->max_extent_size = 0;
6722 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6724 space_info->bytes_readonly += len;
6727 spin_unlock(&cache->lock);
6728 if (!readonly && return_free_space &&
6729 global_rsv->space_info == space_info) {
6732 spin_lock(&global_rsv->lock);
6733 if (!global_rsv->full) {
6734 to_add = min(len, global_rsv->size -
6735 global_rsv->reserved);
6736 global_rsv->reserved += to_add;
6737 space_info->bytes_may_use += to_add;
6738 if (global_rsv->reserved >= global_rsv->size)
6739 global_rsv->full = 1;
6740 trace_btrfs_space_reservation(fs_info,
6746 spin_unlock(&global_rsv->lock);
6747 /* Add to any tickets we may have */
6749 space_info_add_new_bytes(fs_info, space_info,
6752 spin_unlock(&space_info->lock);
6756 btrfs_put_block_group(cache);
6760 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6762 struct btrfs_fs_info *fs_info = trans->fs_info;
6763 struct btrfs_block_group_cache *block_group, *tmp;
6764 struct list_head *deleted_bgs;
6765 struct extent_io_tree *unpin;
6770 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6771 unpin = &fs_info->freed_extents[1];
6773 unpin = &fs_info->freed_extents[0];
6775 while (!trans->aborted) {
6776 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6777 ret = find_first_extent_bit(unpin, 0, &start, &end,
6778 EXTENT_DIRTY, NULL);
6780 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6784 if (btrfs_test_opt(fs_info, DISCARD))
6785 ret = btrfs_discard_extent(fs_info, start,
6786 end + 1 - start, NULL);
6788 clear_extent_dirty(unpin, start, end);
6789 unpin_extent_range(fs_info, start, end, true);
6790 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6795 * Transaction is finished. We don't need the lock anymore. We
6796 * do need to clean up the block groups in case of a transaction
6799 deleted_bgs = &trans->transaction->deleted_bgs;
6800 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6804 if (!trans->aborted)
6805 ret = btrfs_discard_extent(fs_info,
6806 block_group->key.objectid,
6807 block_group->key.offset,
6810 list_del_init(&block_group->bg_list);
6811 btrfs_put_block_group_trimming(block_group);
6812 btrfs_put_block_group(block_group);
6815 const char *errstr = btrfs_decode_error(ret);
6817 "discard failed while removing blockgroup: errno=%d %s",
6825 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6826 struct btrfs_fs_info *info,
6827 struct btrfs_delayed_ref_node *node, u64 parent,
6828 u64 root_objectid, u64 owner_objectid,
6829 u64 owner_offset, int refs_to_drop,
6830 struct btrfs_delayed_extent_op *extent_op)
6832 struct btrfs_key key;
6833 struct btrfs_path *path;
6834 struct btrfs_root *extent_root = info->extent_root;
6835 struct extent_buffer *leaf;
6836 struct btrfs_extent_item *ei;
6837 struct btrfs_extent_inline_ref *iref;
6840 int extent_slot = 0;
6841 int found_extent = 0;
6845 u64 bytenr = node->bytenr;
6846 u64 num_bytes = node->num_bytes;
6848 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6850 path = btrfs_alloc_path();
6854 path->reada = READA_FORWARD;
6855 path->leave_spinning = 1;
6857 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6858 BUG_ON(!is_data && refs_to_drop != 1);
6861 skinny_metadata = false;
6863 ret = lookup_extent_backref(trans, info, path, &iref,
6864 bytenr, num_bytes, parent,
6865 root_objectid, owner_objectid,
6868 extent_slot = path->slots[0];
6869 while (extent_slot >= 0) {
6870 btrfs_item_key_to_cpu(path->nodes[0], &key,
6872 if (key.objectid != bytenr)
6874 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6875 key.offset == num_bytes) {
6879 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6880 key.offset == owner_objectid) {
6884 if (path->slots[0] - extent_slot > 5)
6888 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6889 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6890 if (found_extent && item_size < sizeof(*ei))
6893 if (!found_extent) {
6895 ret = remove_extent_backref(trans, info, path, NULL,
6897 is_data, &last_ref);
6899 btrfs_abort_transaction(trans, ret);
6902 btrfs_release_path(path);
6903 path->leave_spinning = 1;
6905 key.objectid = bytenr;
6906 key.type = BTRFS_EXTENT_ITEM_KEY;
6907 key.offset = num_bytes;
6909 if (!is_data && skinny_metadata) {
6910 key.type = BTRFS_METADATA_ITEM_KEY;
6911 key.offset = owner_objectid;
6914 ret = btrfs_search_slot(trans, extent_root,
6916 if (ret > 0 && skinny_metadata && path->slots[0]) {
6918 * Couldn't find our skinny metadata item,
6919 * see if we have ye olde extent item.
6922 btrfs_item_key_to_cpu(path->nodes[0], &key,
6924 if (key.objectid == bytenr &&
6925 key.type == BTRFS_EXTENT_ITEM_KEY &&
6926 key.offset == num_bytes)
6930 if (ret > 0 && skinny_metadata) {
6931 skinny_metadata = false;
6932 key.objectid = bytenr;
6933 key.type = BTRFS_EXTENT_ITEM_KEY;
6934 key.offset = num_bytes;
6935 btrfs_release_path(path);
6936 ret = btrfs_search_slot(trans, extent_root,
6942 "umm, got %d back from search, was looking for %llu",
6945 btrfs_print_leaf(path->nodes[0]);
6948 btrfs_abort_transaction(trans, ret);
6951 extent_slot = path->slots[0];
6953 } else if (WARN_ON(ret == -ENOENT)) {
6954 btrfs_print_leaf(path->nodes[0]);
6956 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6957 bytenr, parent, root_objectid, owner_objectid,
6959 btrfs_abort_transaction(trans, ret);
6962 btrfs_abort_transaction(trans, ret);
6966 leaf = path->nodes[0];
6967 item_size = btrfs_item_size_nr(leaf, extent_slot);
6968 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6969 if (item_size < sizeof(*ei)) {
6970 BUG_ON(found_extent || extent_slot != path->slots[0]);
6971 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
6974 btrfs_abort_transaction(trans, ret);
6978 btrfs_release_path(path);
6979 path->leave_spinning = 1;
6981 key.objectid = bytenr;
6982 key.type = BTRFS_EXTENT_ITEM_KEY;
6983 key.offset = num_bytes;
6985 ret = btrfs_search_slot(trans, extent_root, &key, path,
6989 "umm, got %d back from search, was looking for %llu",
6991 btrfs_print_leaf(path->nodes[0]);
6994 btrfs_abort_transaction(trans, ret);
6998 extent_slot = path->slots[0];
6999 leaf = path->nodes[0];
7000 item_size = btrfs_item_size_nr(leaf, extent_slot);
7003 BUG_ON(item_size < sizeof(*ei));
7004 ei = btrfs_item_ptr(leaf, extent_slot,
7005 struct btrfs_extent_item);
7006 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7007 key.type == BTRFS_EXTENT_ITEM_KEY) {
7008 struct btrfs_tree_block_info *bi;
7009 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7010 bi = (struct btrfs_tree_block_info *)(ei + 1);
7011 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7014 refs = btrfs_extent_refs(leaf, ei);
7015 if (refs < refs_to_drop) {
7017 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7018 refs_to_drop, refs, bytenr);
7020 btrfs_abort_transaction(trans, ret);
7023 refs -= refs_to_drop;
7027 __run_delayed_extent_op(extent_op, leaf, ei);
7029 * In the case of inline back ref, reference count will
7030 * be updated by remove_extent_backref
7033 BUG_ON(!found_extent);
7035 btrfs_set_extent_refs(leaf, ei, refs);
7036 btrfs_mark_buffer_dirty(leaf);
7039 ret = remove_extent_backref(trans, info, path,
7041 is_data, &last_ref);
7043 btrfs_abort_transaction(trans, ret);
7049 BUG_ON(is_data && refs_to_drop !=
7050 extent_data_ref_count(path, iref));
7052 BUG_ON(path->slots[0] != extent_slot);
7054 BUG_ON(path->slots[0] != extent_slot + 1);
7055 path->slots[0] = extent_slot;
7061 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7064 btrfs_abort_transaction(trans, ret);
7067 btrfs_release_path(path);
7070 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7072 btrfs_abort_transaction(trans, ret);
7077 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7079 btrfs_abort_transaction(trans, ret);
7083 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7085 btrfs_abort_transaction(trans, ret);
7089 btrfs_release_path(path);
7092 btrfs_free_path(path);
7097 * when we free an block, it is possible (and likely) that we free the last
7098 * delayed ref for that extent as well. This searches the delayed ref tree for
7099 * a given extent, and if there are no other delayed refs to be processed, it
7100 * removes it from the tree.
7102 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7105 struct btrfs_delayed_ref_head *head;
7106 struct btrfs_delayed_ref_root *delayed_refs;
7109 delayed_refs = &trans->transaction->delayed_refs;
7110 spin_lock(&delayed_refs->lock);
7111 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7113 goto out_delayed_unlock;
7115 spin_lock(&head->lock);
7116 if (!RB_EMPTY_ROOT(&head->ref_tree))
7119 if (head->extent_op) {
7120 if (!head->must_insert_reserved)
7122 btrfs_free_delayed_extent_op(head->extent_op);
7123 head->extent_op = NULL;
7127 * waiting for the lock here would deadlock. If someone else has it
7128 * locked they are already in the process of dropping it anyway
7130 if (!mutex_trylock(&head->mutex))
7134 * at this point we have a head with no other entries. Go
7135 * ahead and process it.
7137 rb_erase(&head->href_node, &delayed_refs->href_root);
7138 RB_CLEAR_NODE(&head->href_node);
7139 atomic_dec(&delayed_refs->num_entries);
7142 * we don't take a ref on the node because we're removing it from the
7143 * tree, so we just steal the ref the tree was holding.
7145 delayed_refs->num_heads--;
7146 if (head->processing == 0)
7147 delayed_refs->num_heads_ready--;
7148 head->processing = 0;
7149 spin_unlock(&head->lock);
7150 spin_unlock(&delayed_refs->lock);
7152 BUG_ON(head->extent_op);
7153 if (head->must_insert_reserved)
7156 mutex_unlock(&head->mutex);
7157 btrfs_put_delayed_ref_head(head);
7160 spin_unlock(&head->lock);
7163 spin_unlock(&delayed_refs->lock);
7167 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7168 struct btrfs_root *root,
7169 struct extent_buffer *buf,
7170 u64 parent, int last_ref)
7172 struct btrfs_fs_info *fs_info = root->fs_info;
7176 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7177 int old_ref_mod, new_ref_mod;
7179 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7180 root->root_key.objectid,
7181 btrfs_header_level(buf), 0,
7182 BTRFS_DROP_DELAYED_REF);
7183 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7185 root->root_key.objectid,
7186 btrfs_header_level(buf),
7187 BTRFS_DROP_DELAYED_REF, NULL,
7188 &old_ref_mod, &new_ref_mod);
7189 BUG_ON(ret); /* -ENOMEM */
7190 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7193 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7194 struct btrfs_block_group_cache *cache;
7196 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7197 ret = check_ref_cleanup(trans, buf->start);
7203 cache = btrfs_lookup_block_group(fs_info, buf->start);
7205 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7206 pin_down_extent(fs_info, cache, buf->start,
7208 btrfs_put_block_group(cache);
7212 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7214 btrfs_add_free_space(cache, buf->start, buf->len);
7215 btrfs_free_reserved_bytes(cache, buf->len, 0);
7216 btrfs_put_block_group(cache);
7217 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7221 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7222 root->root_key.objectid);
7226 * Deleting the buffer, clear the corrupt flag since it doesn't
7229 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7233 /* Can return -ENOMEM */
7234 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7235 struct btrfs_root *root,
7236 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7237 u64 owner, u64 offset)
7239 struct btrfs_fs_info *fs_info = root->fs_info;
7240 int old_ref_mod, new_ref_mod;
7243 if (btrfs_is_testing(fs_info))
7246 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7247 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7248 root_objectid, owner, offset,
7249 BTRFS_DROP_DELAYED_REF);
7252 * tree log blocks never actually go into the extent allocation
7253 * tree, just update pinning info and exit early.
7255 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7256 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7257 /* unlocks the pinned mutex */
7258 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7259 old_ref_mod = new_ref_mod = 0;
7261 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7262 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7264 root_objectid, (int)owner,
7265 BTRFS_DROP_DELAYED_REF, NULL,
7266 &old_ref_mod, &new_ref_mod);
7268 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7270 root_objectid, owner, offset,
7271 0, BTRFS_DROP_DELAYED_REF,
7272 &old_ref_mod, &new_ref_mod);
7275 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7276 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7282 * when we wait for progress in the block group caching, its because
7283 * our allocation attempt failed at least once. So, we must sleep
7284 * and let some progress happen before we try again.
7286 * This function will sleep at least once waiting for new free space to
7287 * show up, and then it will check the block group free space numbers
7288 * for our min num_bytes. Another option is to have it go ahead
7289 * and look in the rbtree for a free extent of a given size, but this
7292 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7293 * any of the information in this block group.
7295 static noinline void
7296 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7299 struct btrfs_caching_control *caching_ctl;
7301 caching_ctl = get_caching_control(cache);
7305 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7306 (cache->free_space_ctl->free_space >= num_bytes));
7308 put_caching_control(caching_ctl);
7312 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7314 struct btrfs_caching_control *caching_ctl;
7317 caching_ctl = get_caching_control(cache);
7319 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7321 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7322 if (cache->cached == BTRFS_CACHE_ERROR)
7324 put_caching_control(caching_ctl);
7328 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7329 [BTRFS_RAID_RAID10] = "raid10",
7330 [BTRFS_RAID_RAID1] = "raid1",
7331 [BTRFS_RAID_DUP] = "dup",
7332 [BTRFS_RAID_RAID0] = "raid0",
7333 [BTRFS_RAID_SINGLE] = "single",
7334 [BTRFS_RAID_RAID5] = "raid5",
7335 [BTRFS_RAID_RAID6] = "raid6",
7338 static const char *get_raid_name(enum btrfs_raid_types type)
7340 if (type >= BTRFS_NR_RAID_TYPES)
7343 return btrfs_raid_type_names[type];
7346 enum btrfs_loop_type {
7347 LOOP_CACHING_NOWAIT = 0,
7348 LOOP_CACHING_WAIT = 1,
7349 LOOP_ALLOC_CHUNK = 2,
7350 LOOP_NO_EMPTY_SIZE = 3,
7354 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7358 down_read(&cache->data_rwsem);
7362 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7365 btrfs_get_block_group(cache);
7367 down_read(&cache->data_rwsem);
7370 static struct btrfs_block_group_cache *
7371 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7372 struct btrfs_free_cluster *cluster,
7375 struct btrfs_block_group_cache *used_bg = NULL;
7377 spin_lock(&cluster->refill_lock);
7379 used_bg = cluster->block_group;
7383 if (used_bg == block_group)
7386 btrfs_get_block_group(used_bg);
7391 if (down_read_trylock(&used_bg->data_rwsem))
7394 spin_unlock(&cluster->refill_lock);
7396 /* We should only have one-level nested. */
7397 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7399 spin_lock(&cluster->refill_lock);
7400 if (used_bg == cluster->block_group)
7403 up_read(&used_bg->data_rwsem);
7404 btrfs_put_block_group(used_bg);
7409 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7413 up_read(&cache->data_rwsem);
7414 btrfs_put_block_group(cache);
7418 * walks the btree of allocated extents and find a hole of a given size.
7419 * The key ins is changed to record the hole:
7420 * ins->objectid == start position
7421 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7422 * ins->offset == the size of the hole.
7423 * Any available blocks before search_start are skipped.
7425 * If there is no suitable free space, we will record the max size of
7426 * the free space extent currently.
7428 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7429 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7430 u64 hint_byte, struct btrfs_key *ins,
7431 u64 flags, int delalloc)
7434 struct btrfs_root *root = fs_info->extent_root;
7435 struct btrfs_free_cluster *last_ptr = NULL;
7436 struct btrfs_block_group_cache *block_group = NULL;
7437 u64 search_start = 0;
7438 u64 max_extent_size = 0;
7439 u64 empty_cluster = 0;
7440 struct btrfs_space_info *space_info;
7442 int index = btrfs_bg_flags_to_raid_index(flags);
7443 bool failed_cluster_refill = false;
7444 bool failed_alloc = false;
7445 bool use_cluster = true;
7446 bool have_caching_bg = false;
7447 bool orig_have_caching_bg = false;
7448 bool full_search = false;
7450 WARN_ON(num_bytes < fs_info->sectorsize);
7451 ins->type = BTRFS_EXTENT_ITEM_KEY;
7455 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7457 space_info = __find_space_info(fs_info, flags);
7459 btrfs_err(fs_info, "No space info for %llu", flags);
7464 * If our free space is heavily fragmented we may not be able to make
7465 * big contiguous allocations, so instead of doing the expensive search
7466 * for free space, simply return ENOSPC with our max_extent_size so we
7467 * can go ahead and search for a more manageable chunk.
7469 * If our max_extent_size is large enough for our allocation simply
7470 * disable clustering since we will likely not be able to find enough
7471 * space to create a cluster and induce latency trying.
7473 if (unlikely(space_info->max_extent_size)) {
7474 spin_lock(&space_info->lock);
7475 if (space_info->max_extent_size &&
7476 num_bytes > space_info->max_extent_size) {
7477 ins->offset = space_info->max_extent_size;
7478 spin_unlock(&space_info->lock);
7480 } else if (space_info->max_extent_size) {
7481 use_cluster = false;
7483 spin_unlock(&space_info->lock);
7486 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7488 spin_lock(&last_ptr->lock);
7489 if (last_ptr->block_group)
7490 hint_byte = last_ptr->window_start;
7491 if (last_ptr->fragmented) {
7493 * We still set window_start so we can keep track of the
7494 * last place we found an allocation to try and save
7497 hint_byte = last_ptr->window_start;
7498 use_cluster = false;
7500 spin_unlock(&last_ptr->lock);
7503 search_start = max(search_start, first_logical_byte(fs_info, 0));
7504 search_start = max(search_start, hint_byte);
7505 if (search_start == hint_byte) {
7506 block_group = btrfs_lookup_block_group(fs_info, search_start);
7508 * we don't want to use the block group if it doesn't match our
7509 * allocation bits, or if its not cached.
7511 * However if we are re-searching with an ideal block group
7512 * picked out then we don't care that the block group is cached.
7514 if (block_group && block_group_bits(block_group, flags) &&
7515 block_group->cached != BTRFS_CACHE_NO) {
7516 down_read(&space_info->groups_sem);
7517 if (list_empty(&block_group->list) ||
7520 * someone is removing this block group,
7521 * we can't jump into the have_block_group
7522 * target because our list pointers are not
7525 btrfs_put_block_group(block_group);
7526 up_read(&space_info->groups_sem);
7528 index = btrfs_bg_flags_to_raid_index(
7529 block_group->flags);
7530 btrfs_lock_block_group(block_group, delalloc);
7531 goto have_block_group;
7533 } else if (block_group) {
7534 btrfs_put_block_group(block_group);
7538 have_caching_bg = false;
7539 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7541 down_read(&space_info->groups_sem);
7542 list_for_each_entry(block_group, &space_info->block_groups[index],
7547 /* If the block group is read-only, we can skip it entirely. */
7548 if (unlikely(block_group->ro))
7551 btrfs_grab_block_group(block_group, delalloc);
7552 search_start = block_group->key.objectid;
7555 * this can happen if we end up cycling through all the
7556 * raid types, but we want to make sure we only allocate
7557 * for the proper type.
7559 if (!block_group_bits(block_group, flags)) {
7560 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7561 BTRFS_BLOCK_GROUP_RAID1 |
7562 BTRFS_BLOCK_GROUP_RAID5 |
7563 BTRFS_BLOCK_GROUP_RAID6 |
7564 BTRFS_BLOCK_GROUP_RAID10;
7567 * if they asked for extra copies and this block group
7568 * doesn't provide them, bail. This does allow us to
7569 * fill raid0 from raid1.
7571 if ((flags & extra) && !(block_group->flags & extra))
7576 cached = block_group_cache_done(block_group);
7577 if (unlikely(!cached)) {
7578 have_caching_bg = true;
7579 ret = cache_block_group(block_group, 0);
7584 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7588 * Ok we want to try and use the cluster allocator, so
7591 if (last_ptr && use_cluster) {
7592 struct btrfs_block_group_cache *used_block_group;
7593 unsigned long aligned_cluster;
7595 * the refill lock keeps out other
7596 * people trying to start a new cluster
7598 used_block_group = btrfs_lock_cluster(block_group,
7601 if (!used_block_group)
7602 goto refill_cluster;
7604 if (used_block_group != block_group &&
7605 (used_block_group->ro ||
7606 !block_group_bits(used_block_group, flags)))
7607 goto release_cluster;
7609 offset = btrfs_alloc_from_cluster(used_block_group,
7612 used_block_group->key.objectid,
7615 /* we have a block, we're done */
7616 spin_unlock(&last_ptr->refill_lock);
7617 trace_btrfs_reserve_extent_cluster(fs_info,
7619 search_start, num_bytes);
7620 if (used_block_group != block_group) {
7621 btrfs_release_block_group(block_group,
7623 block_group = used_block_group;
7628 WARN_ON(last_ptr->block_group != used_block_group);
7630 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7631 * set up a new clusters, so lets just skip it
7632 * and let the allocator find whatever block
7633 * it can find. If we reach this point, we
7634 * will have tried the cluster allocator
7635 * plenty of times and not have found
7636 * anything, so we are likely way too
7637 * fragmented for the clustering stuff to find
7640 * However, if the cluster is taken from the
7641 * current block group, release the cluster
7642 * first, so that we stand a better chance of
7643 * succeeding in the unclustered
7645 if (loop >= LOOP_NO_EMPTY_SIZE &&
7646 used_block_group != block_group) {
7647 spin_unlock(&last_ptr->refill_lock);
7648 btrfs_release_block_group(used_block_group,
7650 goto unclustered_alloc;
7654 * this cluster didn't work out, free it and
7657 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7659 if (used_block_group != block_group)
7660 btrfs_release_block_group(used_block_group,
7663 if (loop >= LOOP_NO_EMPTY_SIZE) {
7664 spin_unlock(&last_ptr->refill_lock);
7665 goto unclustered_alloc;
7668 aligned_cluster = max_t(unsigned long,
7669 empty_cluster + empty_size,
7670 block_group->full_stripe_len);
7672 /* allocate a cluster in this block group */
7673 ret = btrfs_find_space_cluster(fs_info, block_group,
7674 last_ptr, search_start,
7679 * now pull our allocation out of this
7682 offset = btrfs_alloc_from_cluster(block_group,
7688 /* we found one, proceed */
7689 spin_unlock(&last_ptr->refill_lock);
7690 trace_btrfs_reserve_extent_cluster(fs_info,
7691 block_group, search_start,
7695 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7696 && !failed_cluster_refill) {
7697 spin_unlock(&last_ptr->refill_lock);
7699 failed_cluster_refill = true;
7700 wait_block_group_cache_progress(block_group,
7701 num_bytes + empty_cluster + empty_size);
7702 goto have_block_group;
7706 * at this point we either didn't find a cluster
7707 * or we weren't able to allocate a block from our
7708 * cluster. Free the cluster we've been trying
7709 * to use, and go to the next block group
7711 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7712 spin_unlock(&last_ptr->refill_lock);
7718 * We are doing an unclustered alloc, set the fragmented flag so
7719 * we don't bother trying to setup a cluster again until we get
7722 if (unlikely(last_ptr)) {
7723 spin_lock(&last_ptr->lock);
7724 last_ptr->fragmented = 1;
7725 spin_unlock(&last_ptr->lock);
7728 struct btrfs_free_space_ctl *ctl =
7729 block_group->free_space_ctl;
7731 spin_lock(&ctl->tree_lock);
7732 if (ctl->free_space <
7733 num_bytes + empty_cluster + empty_size) {
7734 if (ctl->free_space > max_extent_size)
7735 max_extent_size = ctl->free_space;
7736 spin_unlock(&ctl->tree_lock);
7739 spin_unlock(&ctl->tree_lock);
7742 offset = btrfs_find_space_for_alloc(block_group, search_start,
7743 num_bytes, empty_size,
7746 * If we didn't find a chunk, and we haven't failed on this
7747 * block group before, and this block group is in the middle of
7748 * caching and we are ok with waiting, then go ahead and wait
7749 * for progress to be made, and set failed_alloc to true.
7751 * If failed_alloc is true then we've already waited on this
7752 * block group once and should move on to the next block group.
7754 if (!offset && !failed_alloc && !cached &&
7755 loop > LOOP_CACHING_NOWAIT) {
7756 wait_block_group_cache_progress(block_group,
7757 num_bytes + empty_size);
7758 failed_alloc = true;
7759 goto have_block_group;
7760 } else if (!offset) {
7764 search_start = ALIGN(offset, fs_info->stripesize);
7766 /* move on to the next group */
7767 if (search_start + num_bytes >
7768 block_group->key.objectid + block_group->key.offset) {
7769 btrfs_add_free_space(block_group, offset, num_bytes);
7773 if (offset < search_start)
7774 btrfs_add_free_space(block_group, offset,
7775 search_start - offset);
7776 BUG_ON(offset > search_start);
7778 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7779 num_bytes, delalloc);
7780 if (ret == -EAGAIN) {
7781 btrfs_add_free_space(block_group, offset, num_bytes);
7784 btrfs_inc_block_group_reservations(block_group);
7786 /* we are all good, lets return */
7787 ins->objectid = search_start;
7788 ins->offset = num_bytes;
7790 trace_btrfs_reserve_extent(fs_info, block_group,
7791 search_start, num_bytes);
7792 btrfs_release_block_group(block_group, delalloc);
7795 failed_cluster_refill = false;
7796 failed_alloc = false;
7797 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7799 btrfs_release_block_group(block_group, delalloc);
7802 up_read(&space_info->groups_sem);
7804 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7805 && !orig_have_caching_bg)
7806 orig_have_caching_bg = true;
7808 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7811 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7815 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7816 * caching kthreads as we move along
7817 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7818 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7819 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7822 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7824 if (loop == LOOP_CACHING_NOWAIT) {
7826 * We want to skip the LOOP_CACHING_WAIT step if we
7827 * don't have any uncached bgs and we've already done a
7828 * full search through.
7830 if (orig_have_caching_bg || !full_search)
7831 loop = LOOP_CACHING_WAIT;
7833 loop = LOOP_ALLOC_CHUNK;
7838 if (loop == LOOP_ALLOC_CHUNK) {
7839 struct btrfs_trans_handle *trans;
7842 trans = current->journal_info;
7846 trans = btrfs_join_transaction(root);
7848 if (IS_ERR(trans)) {
7849 ret = PTR_ERR(trans);
7853 ret = do_chunk_alloc(trans, fs_info, flags,
7857 * If we can't allocate a new chunk we've already looped
7858 * through at least once, move on to the NO_EMPTY_SIZE
7862 loop = LOOP_NO_EMPTY_SIZE;
7865 * Do not bail out on ENOSPC since we
7866 * can do more things.
7868 if (ret < 0 && ret != -ENOSPC)
7869 btrfs_abort_transaction(trans, ret);
7873 btrfs_end_transaction(trans);
7878 if (loop == LOOP_NO_EMPTY_SIZE) {
7880 * Don't loop again if we already have no empty_size and
7883 if (empty_size == 0 &&
7884 empty_cluster == 0) {
7893 } else if (!ins->objectid) {
7895 } else if (ins->objectid) {
7896 if (!use_cluster && last_ptr) {
7897 spin_lock(&last_ptr->lock);
7898 last_ptr->window_start = ins->objectid;
7899 spin_unlock(&last_ptr->lock);
7904 if (ret == -ENOSPC) {
7905 spin_lock(&space_info->lock);
7906 space_info->max_extent_size = max_extent_size;
7907 spin_unlock(&space_info->lock);
7908 ins->offset = max_extent_size;
7913 static void dump_space_info(struct btrfs_fs_info *fs_info,
7914 struct btrfs_space_info *info, u64 bytes,
7915 int dump_block_groups)
7917 struct btrfs_block_group_cache *cache;
7920 spin_lock(&info->lock);
7921 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7923 info->total_bytes - btrfs_space_info_used(info, true),
7924 info->full ? "" : "not ");
7926 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7927 info->total_bytes, info->bytes_used, info->bytes_pinned,
7928 info->bytes_reserved, info->bytes_may_use,
7929 info->bytes_readonly);
7930 spin_unlock(&info->lock);
7932 if (!dump_block_groups)
7935 down_read(&info->groups_sem);
7937 list_for_each_entry(cache, &info->block_groups[index], list) {
7938 spin_lock(&cache->lock);
7940 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7941 cache->key.objectid, cache->key.offset,
7942 btrfs_block_group_used(&cache->item), cache->pinned,
7943 cache->reserved, cache->ro ? "[readonly]" : "");
7944 btrfs_dump_free_space(cache, bytes);
7945 spin_unlock(&cache->lock);
7947 if (++index < BTRFS_NR_RAID_TYPES)
7949 up_read(&info->groups_sem);
7953 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7954 * hole that is at least as big as @num_bytes.
7956 * @root - The root that will contain this extent
7958 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7959 * is used for accounting purposes. This value differs
7960 * from @num_bytes only in the case of compressed extents.
7962 * @num_bytes - Number of bytes to allocate on-disk.
7964 * @min_alloc_size - Indicates the minimum amount of space that the
7965 * allocator should try to satisfy. In some cases
7966 * @num_bytes may be larger than what is required and if
7967 * the filesystem is fragmented then allocation fails.
7968 * However, the presence of @min_alloc_size gives a
7969 * chance to try and satisfy the smaller allocation.
7971 * @empty_size - A hint that you plan on doing more COW. This is the
7972 * size in bytes the allocator should try to find free
7973 * next to the block it returns. This is just a hint and
7974 * may be ignored by the allocator.
7976 * @hint_byte - Hint to the allocator to start searching above the byte
7977 * address passed. It might be ignored.
7979 * @ins - This key is modified to record the found hole. It will
7980 * have the following values:
7981 * ins->objectid == start position
7982 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7983 * ins->offset == the size of the hole.
7985 * @is_data - Boolean flag indicating whether an extent is
7986 * allocated for data (true) or metadata (false)
7988 * @delalloc - Boolean flag indicating whether this allocation is for
7989 * delalloc or not. If 'true' data_rwsem of block groups
7990 * is going to be acquired.
7993 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
7994 * case -ENOSPC is returned then @ins->offset will contain the size of the
7995 * largest available hole the allocator managed to find.
7997 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7998 u64 num_bytes, u64 min_alloc_size,
7999 u64 empty_size, u64 hint_byte,
8000 struct btrfs_key *ins, int is_data, int delalloc)
8002 struct btrfs_fs_info *fs_info = root->fs_info;
8003 bool final_tried = num_bytes == min_alloc_size;
8007 flags = get_alloc_profile_by_root(root, is_data);
8009 WARN_ON(num_bytes < fs_info->sectorsize);
8010 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8011 hint_byte, ins, flags, delalloc);
8012 if (!ret && !is_data) {
8013 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8014 } else if (ret == -ENOSPC) {
8015 if (!final_tried && ins->offset) {
8016 num_bytes = min(num_bytes >> 1, ins->offset);
8017 num_bytes = round_down(num_bytes,
8018 fs_info->sectorsize);
8019 num_bytes = max(num_bytes, min_alloc_size);
8020 ram_bytes = num_bytes;
8021 if (num_bytes == min_alloc_size)
8024 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8025 struct btrfs_space_info *sinfo;
8027 sinfo = __find_space_info(fs_info, flags);
8029 "allocation failed flags %llu, wanted %llu",
8032 dump_space_info(fs_info, sinfo, num_bytes, 1);
8039 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8041 int pin, int delalloc)
8043 struct btrfs_block_group_cache *cache;
8046 cache = btrfs_lookup_block_group(fs_info, start);
8048 btrfs_err(fs_info, "Unable to find block group for %llu",
8054 pin_down_extent(fs_info, cache, start, len, 1);
8056 if (btrfs_test_opt(fs_info, DISCARD))
8057 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8058 btrfs_add_free_space(cache, start, len);
8059 btrfs_free_reserved_bytes(cache, len, delalloc);
8060 trace_btrfs_reserved_extent_free(fs_info, start, len);
8063 btrfs_put_block_group(cache);
8067 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8068 u64 start, u64 len, int delalloc)
8070 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8073 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8076 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8079 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8080 struct btrfs_fs_info *fs_info,
8081 u64 parent, u64 root_objectid,
8082 u64 flags, u64 owner, u64 offset,
8083 struct btrfs_key *ins, int ref_mod)
8086 struct btrfs_extent_item *extent_item;
8087 struct btrfs_extent_inline_ref *iref;
8088 struct btrfs_path *path;
8089 struct extent_buffer *leaf;
8094 type = BTRFS_SHARED_DATA_REF_KEY;
8096 type = BTRFS_EXTENT_DATA_REF_KEY;
8098 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8100 path = btrfs_alloc_path();
8104 path->leave_spinning = 1;
8105 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8108 btrfs_free_path(path);
8112 leaf = path->nodes[0];
8113 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8114 struct btrfs_extent_item);
8115 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8116 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8117 btrfs_set_extent_flags(leaf, extent_item,
8118 flags | BTRFS_EXTENT_FLAG_DATA);
8120 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8121 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8123 struct btrfs_shared_data_ref *ref;
8124 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8125 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8126 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8128 struct btrfs_extent_data_ref *ref;
8129 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8130 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8131 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8132 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8133 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8136 btrfs_mark_buffer_dirty(path->nodes[0]);
8137 btrfs_free_path(path);
8139 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8144 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8145 if (ret) { /* -ENOENT, logic error */
8146 btrfs_err(fs_info, "update block group failed for %llu %llu",
8147 ins->objectid, ins->offset);
8150 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8154 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8155 struct btrfs_fs_info *fs_info,
8156 u64 parent, u64 root_objectid,
8157 u64 flags, struct btrfs_disk_key *key,
8158 int level, struct btrfs_key *ins)
8161 struct btrfs_extent_item *extent_item;
8162 struct btrfs_tree_block_info *block_info;
8163 struct btrfs_extent_inline_ref *iref;
8164 struct btrfs_path *path;
8165 struct extent_buffer *leaf;
8166 u32 size = sizeof(*extent_item) + sizeof(*iref);
8167 u64 num_bytes = ins->offset;
8168 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8170 if (!skinny_metadata)
8171 size += sizeof(*block_info);
8173 path = btrfs_alloc_path();
8175 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8180 path->leave_spinning = 1;
8181 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8184 btrfs_free_path(path);
8185 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8190 leaf = path->nodes[0];
8191 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8192 struct btrfs_extent_item);
8193 btrfs_set_extent_refs(leaf, extent_item, 1);
8194 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8195 btrfs_set_extent_flags(leaf, extent_item,
8196 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8198 if (skinny_metadata) {
8199 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8200 num_bytes = fs_info->nodesize;
8202 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8203 btrfs_set_tree_block_key(leaf, block_info, key);
8204 btrfs_set_tree_block_level(leaf, block_info, level);
8205 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8209 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8210 btrfs_set_extent_inline_ref_type(leaf, iref,
8211 BTRFS_SHARED_BLOCK_REF_KEY);
8212 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8214 btrfs_set_extent_inline_ref_type(leaf, iref,
8215 BTRFS_TREE_BLOCK_REF_KEY);
8216 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8219 btrfs_mark_buffer_dirty(leaf);
8220 btrfs_free_path(path);
8222 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8227 ret = update_block_group(trans, fs_info, ins->objectid,
8228 fs_info->nodesize, 1);
8229 if (ret) { /* -ENOENT, logic error */
8230 btrfs_err(fs_info, "update block group failed for %llu %llu",
8231 ins->objectid, ins->offset);
8235 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8240 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8241 struct btrfs_root *root, u64 owner,
8242 u64 offset, u64 ram_bytes,
8243 struct btrfs_key *ins)
8245 struct btrfs_fs_info *fs_info = root->fs_info;
8248 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8250 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8251 root->root_key.objectid, owner, offset,
8252 BTRFS_ADD_DELAYED_EXTENT);
8254 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8256 root->root_key.objectid, owner,
8258 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8263 * this is used by the tree logging recovery code. It records that
8264 * an extent has been allocated and makes sure to clear the free
8265 * space cache bits as well
8267 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8268 struct btrfs_fs_info *fs_info,
8269 u64 root_objectid, u64 owner, u64 offset,
8270 struct btrfs_key *ins)
8273 struct btrfs_block_group_cache *block_group;
8274 struct btrfs_space_info *space_info;
8277 * Mixed block groups will exclude before processing the log so we only
8278 * need to do the exclude dance if this fs isn't mixed.
8280 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8281 ret = __exclude_logged_extent(fs_info, ins->objectid,
8287 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8291 space_info = block_group->space_info;
8292 spin_lock(&space_info->lock);
8293 spin_lock(&block_group->lock);
8294 space_info->bytes_reserved += ins->offset;
8295 block_group->reserved += ins->offset;
8296 spin_unlock(&block_group->lock);
8297 spin_unlock(&space_info->lock);
8299 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8300 0, owner, offset, ins, 1);
8301 btrfs_put_block_group(block_group);
8305 static struct extent_buffer *
8306 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8307 u64 bytenr, int level)
8309 struct btrfs_fs_info *fs_info = root->fs_info;
8310 struct extent_buffer *buf;
8312 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8316 btrfs_set_header_generation(buf, trans->transid);
8317 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8318 btrfs_tree_lock(buf);
8319 clean_tree_block(fs_info, buf);
8320 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8322 btrfs_set_lock_blocking(buf);
8323 set_extent_buffer_uptodate(buf);
8325 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8326 buf->log_index = root->log_transid % 2;
8328 * we allow two log transactions at a time, use different
8329 * EXENT bit to differentiate dirty pages.
8331 if (buf->log_index == 0)
8332 set_extent_dirty(&root->dirty_log_pages, buf->start,
8333 buf->start + buf->len - 1, GFP_NOFS);
8335 set_extent_new(&root->dirty_log_pages, buf->start,
8336 buf->start + buf->len - 1);
8338 buf->log_index = -1;
8339 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8340 buf->start + buf->len - 1, GFP_NOFS);
8342 trans->dirty = true;
8343 /* this returns a buffer locked for blocking */
8347 static struct btrfs_block_rsv *
8348 use_block_rsv(struct btrfs_trans_handle *trans,
8349 struct btrfs_root *root, u32 blocksize)
8351 struct btrfs_fs_info *fs_info = root->fs_info;
8352 struct btrfs_block_rsv *block_rsv;
8353 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8355 bool global_updated = false;
8357 block_rsv = get_block_rsv(trans, root);
8359 if (unlikely(block_rsv->size == 0))
8362 ret = block_rsv_use_bytes(block_rsv, blocksize);
8366 if (block_rsv->failfast)
8367 return ERR_PTR(ret);
8369 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8370 global_updated = true;
8371 update_global_block_rsv(fs_info);
8375 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8376 static DEFINE_RATELIMIT_STATE(_rs,
8377 DEFAULT_RATELIMIT_INTERVAL * 10,
8378 /*DEFAULT_RATELIMIT_BURST*/ 1);
8379 if (__ratelimit(&_rs))
8381 "BTRFS: block rsv returned %d\n", ret);
8384 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8385 BTRFS_RESERVE_NO_FLUSH);
8389 * If we couldn't reserve metadata bytes try and use some from
8390 * the global reserve if its space type is the same as the global
8393 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8394 block_rsv->space_info == global_rsv->space_info) {
8395 ret = block_rsv_use_bytes(global_rsv, blocksize);
8399 return ERR_PTR(ret);
8402 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8403 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8405 block_rsv_add_bytes(block_rsv, blocksize, 0);
8406 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8410 * finds a free extent and does all the dirty work required for allocation
8411 * returns the tree buffer or an ERR_PTR on error.
8413 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8414 struct btrfs_root *root,
8415 u64 parent, u64 root_objectid,
8416 const struct btrfs_disk_key *key,
8417 int level, u64 hint,
8420 struct btrfs_fs_info *fs_info = root->fs_info;
8421 struct btrfs_key ins;
8422 struct btrfs_block_rsv *block_rsv;
8423 struct extent_buffer *buf;
8424 struct btrfs_delayed_extent_op *extent_op;
8427 u32 blocksize = fs_info->nodesize;
8428 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8430 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8431 if (btrfs_is_testing(fs_info)) {
8432 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8435 root->alloc_bytenr += blocksize;
8440 block_rsv = use_block_rsv(trans, root, blocksize);
8441 if (IS_ERR(block_rsv))
8442 return ERR_CAST(block_rsv);
8444 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8445 empty_size, hint, &ins, 0, 0);
8449 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8452 goto out_free_reserved;
8455 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8457 parent = ins.objectid;
8458 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8462 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8463 extent_op = btrfs_alloc_delayed_extent_op();
8469 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8471 memset(&extent_op->key, 0, sizeof(extent_op->key));
8472 extent_op->flags_to_set = flags;
8473 extent_op->update_key = skinny_metadata ? false : true;
8474 extent_op->update_flags = true;
8475 extent_op->is_data = false;
8476 extent_op->level = level;
8478 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8479 root_objectid, level, 0,
8480 BTRFS_ADD_DELAYED_EXTENT);
8481 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8483 root_objectid, level,
8484 BTRFS_ADD_DELAYED_EXTENT,
8485 extent_op, NULL, NULL);
8487 goto out_free_delayed;
8492 btrfs_free_delayed_extent_op(extent_op);
8494 free_extent_buffer(buf);
8496 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8498 unuse_block_rsv(fs_info, block_rsv, blocksize);
8499 return ERR_PTR(ret);
8502 struct walk_control {
8503 u64 refs[BTRFS_MAX_LEVEL];
8504 u64 flags[BTRFS_MAX_LEVEL];
8505 struct btrfs_key update_progress;
8516 #define DROP_REFERENCE 1
8517 #define UPDATE_BACKREF 2
8519 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8520 struct btrfs_root *root,
8521 struct walk_control *wc,
8522 struct btrfs_path *path)
8524 struct btrfs_fs_info *fs_info = root->fs_info;
8530 struct btrfs_key key;
8531 struct extent_buffer *eb;
8536 if (path->slots[wc->level] < wc->reada_slot) {
8537 wc->reada_count = wc->reada_count * 2 / 3;
8538 wc->reada_count = max(wc->reada_count, 2);
8540 wc->reada_count = wc->reada_count * 3 / 2;
8541 wc->reada_count = min_t(int, wc->reada_count,
8542 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8545 eb = path->nodes[wc->level];
8546 nritems = btrfs_header_nritems(eb);
8548 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8549 if (nread >= wc->reada_count)
8553 bytenr = btrfs_node_blockptr(eb, slot);
8554 generation = btrfs_node_ptr_generation(eb, slot);
8556 if (slot == path->slots[wc->level])
8559 if (wc->stage == UPDATE_BACKREF &&
8560 generation <= root->root_key.offset)
8563 /* We don't lock the tree block, it's OK to be racy here */
8564 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8565 wc->level - 1, 1, &refs,
8567 /* We don't care about errors in readahead. */
8572 if (wc->stage == DROP_REFERENCE) {
8576 if (wc->level == 1 &&
8577 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8579 if (!wc->update_ref ||
8580 generation <= root->root_key.offset)
8582 btrfs_node_key_to_cpu(eb, &key, slot);
8583 ret = btrfs_comp_cpu_keys(&key,
8584 &wc->update_progress);
8588 if (wc->level == 1 &&
8589 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8593 readahead_tree_block(fs_info, bytenr);
8596 wc->reada_slot = slot;
8600 * helper to process tree block while walking down the tree.
8602 * when wc->stage == UPDATE_BACKREF, this function updates
8603 * back refs for pointers in the block.
8605 * NOTE: return value 1 means we should stop walking down.
8607 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8608 struct btrfs_root *root,
8609 struct btrfs_path *path,
8610 struct walk_control *wc, int lookup_info)
8612 struct btrfs_fs_info *fs_info = root->fs_info;
8613 int level = wc->level;
8614 struct extent_buffer *eb = path->nodes[level];
8615 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8618 if (wc->stage == UPDATE_BACKREF &&
8619 btrfs_header_owner(eb) != root->root_key.objectid)
8623 * when reference count of tree block is 1, it won't increase
8624 * again. once full backref flag is set, we never clear it.
8627 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8628 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8629 BUG_ON(!path->locks[level]);
8630 ret = btrfs_lookup_extent_info(trans, fs_info,
8631 eb->start, level, 1,
8634 BUG_ON(ret == -ENOMEM);
8637 BUG_ON(wc->refs[level] == 0);
8640 if (wc->stage == DROP_REFERENCE) {
8641 if (wc->refs[level] > 1)
8644 if (path->locks[level] && !wc->keep_locks) {
8645 btrfs_tree_unlock_rw(eb, path->locks[level]);
8646 path->locks[level] = 0;
8651 /* wc->stage == UPDATE_BACKREF */
8652 if (!(wc->flags[level] & flag)) {
8653 BUG_ON(!path->locks[level]);
8654 ret = btrfs_inc_ref(trans, root, eb, 1);
8655 BUG_ON(ret); /* -ENOMEM */
8656 ret = btrfs_dec_ref(trans, root, eb, 0);
8657 BUG_ON(ret); /* -ENOMEM */
8658 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8660 btrfs_header_level(eb), 0);
8661 BUG_ON(ret); /* -ENOMEM */
8662 wc->flags[level] |= flag;
8666 * the block is shared by multiple trees, so it's not good to
8667 * keep the tree lock
8669 if (path->locks[level] && level > 0) {
8670 btrfs_tree_unlock_rw(eb, path->locks[level]);
8671 path->locks[level] = 0;
8677 * helper to process tree block pointer.
8679 * when wc->stage == DROP_REFERENCE, this function checks
8680 * reference count of the block pointed to. if the block
8681 * is shared and we need update back refs for the subtree
8682 * rooted at the block, this function changes wc->stage to
8683 * UPDATE_BACKREF. if the block is shared and there is no
8684 * need to update back, this function drops the reference
8687 * NOTE: return value 1 means we should stop walking down.
8689 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8690 struct btrfs_root *root,
8691 struct btrfs_path *path,
8692 struct walk_control *wc, int *lookup_info)
8694 struct btrfs_fs_info *fs_info = root->fs_info;
8699 struct btrfs_key key;
8700 struct btrfs_key first_key;
8701 struct extent_buffer *next;
8702 int level = wc->level;
8705 bool need_account = false;
8707 generation = btrfs_node_ptr_generation(path->nodes[level],
8708 path->slots[level]);
8710 * if the lower level block was created before the snapshot
8711 * was created, we know there is no need to update back refs
8714 if (wc->stage == UPDATE_BACKREF &&
8715 generation <= root->root_key.offset) {
8720 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8721 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8722 path->slots[level]);
8723 blocksize = fs_info->nodesize;
8725 next = find_extent_buffer(fs_info, bytenr);
8727 next = btrfs_find_create_tree_block(fs_info, bytenr);
8729 return PTR_ERR(next);
8731 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8735 btrfs_tree_lock(next);
8736 btrfs_set_lock_blocking(next);
8738 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8739 &wc->refs[level - 1],
8740 &wc->flags[level - 1]);
8744 if (unlikely(wc->refs[level - 1] == 0)) {
8745 btrfs_err(fs_info, "Missing references.");
8751 if (wc->stage == DROP_REFERENCE) {
8752 if (wc->refs[level - 1] > 1) {
8753 need_account = true;
8755 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8758 if (!wc->update_ref ||
8759 generation <= root->root_key.offset)
8762 btrfs_node_key_to_cpu(path->nodes[level], &key,
8763 path->slots[level]);
8764 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8768 wc->stage = UPDATE_BACKREF;
8769 wc->shared_level = level - 1;
8773 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8777 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8778 btrfs_tree_unlock(next);
8779 free_extent_buffer(next);
8785 if (reada && level == 1)
8786 reada_walk_down(trans, root, wc, path);
8787 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8790 return PTR_ERR(next);
8791 } else if (!extent_buffer_uptodate(next)) {
8792 free_extent_buffer(next);
8795 btrfs_tree_lock(next);
8796 btrfs_set_lock_blocking(next);
8800 ASSERT(level == btrfs_header_level(next));
8801 if (level != btrfs_header_level(next)) {
8802 btrfs_err(root->fs_info, "mismatched level");
8806 path->nodes[level] = next;
8807 path->slots[level] = 0;
8808 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8814 wc->refs[level - 1] = 0;
8815 wc->flags[level - 1] = 0;
8816 if (wc->stage == DROP_REFERENCE) {
8817 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8818 parent = path->nodes[level]->start;
8820 ASSERT(root->root_key.objectid ==
8821 btrfs_header_owner(path->nodes[level]));
8822 if (root->root_key.objectid !=
8823 btrfs_header_owner(path->nodes[level])) {
8824 btrfs_err(root->fs_info,
8825 "mismatched block owner");
8833 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8834 generation, level - 1);
8836 btrfs_err_rl(fs_info,
8837 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8841 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8842 parent, root->root_key.objectid,
8852 btrfs_tree_unlock(next);
8853 free_extent_buffer(next);
8859 * helper to process tree block while walking up the tree.
8861 * when wc->stage == DROP_REFERENCE, this function drops
8862 * reference count on the block.
8864 * when wc->stage == UPDATE_BACKREF, this function changes
8865 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8866 * to UPDATE_BACKREF previously while processing the block.
8868 * NOTE: return value 1 means we should stop walking up.
8870 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8871 struct btrfs_root *root,
8872 struct btrfs_path *path,
8873 struct walk_control *wc)
8875 struct btrfs_fs_info *fs_info = root->fs_info;
8877 int level = wc->level;
8878 struct extent_buffer *eb = path->nodes[level];
8881 if (wc->stage == UPDATE_BACKREF) {
8882 BUG_ON(wc->shared_level < level);
8883 if (level < wc->shared_level)
8886 ret = find_next_key(path, level + 1, &wc->update_progress);
8890 wc->stage = DROP_REFERENCE;
8891 wc->shared_level = -1;
8892 path->slots[level] = 0;
8895 * check reference count again if the block isn't locked.
8896 * we should start walking down the tree again if reference
8899 if (!path->locks[level]) {
8901 btrfs_tree_lock(eb);
8902 btrfs_set_lock_blocking(eb);
8903 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8905 ret = btrfs_lookup_extent_info(trans, fs_info,
8906 eb->start, level, 1,
8910 btrfs_tree_unlock_rw(eb, path->locks[level]);
8911 path->locks[level] = 0;
8914 BUG_ON(wc->refs[level] == 0);
8915 if (wc->refs[level] == 1) {
8916 btrfs_tree_unlock_rw(eb, path->locks[level]);
8917 path->locks[level] = 0;
8923 /* wc->stage == DROP_REFERENCE */
8924 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8926 if (wc->refs[level] == 1) {
8928 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8929 ret = btrfs_dec_ref(trans, root, eb, 1);
8931 ret = btrfs_dec_ref(trans, root, eb, 0);
8932 BUG_ON(ret); /* -ENOMEM */
8933 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8935 btrfs_err_rl(fs_info,
8936 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8940 /* make block locked assertion in clean_tree_block happy */
8941 if (!path->locks[level] &&
8942 btrfs_header_generation(eb) == trans->transid) {
8943 btrfs_tree_lock(eb);
8944 btrfs_set_lock_blocking(eb);
8945 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8947 clean_tree_block(fs_info, eb);
8950 if (eb == root->node) {
8951 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8954 BUG_ON(root->root_key.objectid !=
8955 btrfs_header_owner(eb));
8957 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8958 parent = path->nodes[level + 1]->start;
8960 BUG_ON(root->root_key.objectid !=
8961 btrfs_header_owner(path->nodes[level + 1]));
8964 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8966 wc->refs[level] = 0;
8967 wc->flags[level] = 0;
8971 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8972 struct btrfs_root *root,
8973 struct btrfs_path *path,
8974 struct walk_control *wc)
8976 int level = wc->level;
8977 int lookup_info = 1;
8980 while (level >= 0) {
8981 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8988 if (path->slots[level] >=
8989 btrfs_header_nritems(path->nodes[level]))
8992 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8994 path->slots[level]++;
9003 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9004 struct btrfs_root *root,
9005 struct btrfs_path *path,
9006 struct walk_control *wc, int max_level)
9008 int level = wc->level;
9011 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9012 while (level < max_level && path->nodes[level]) {
9014 if (path->slots[level] + 1 <
9015 btrfs_header_nritems(path->nodes[level])) {
9016 path->slots[level]++;
9019 ret = walk_up_proc(trans, root, path, wc);
9023 if (path->locks[level]) {
9024 btrfs_tree_unlock_rw(path->nodes[level],
9025 path->locks[level]);
9026 path->locks[level] = 0;
9028 free_extent_buffer(path->nodes[level]);
9029 path->nodes[level] = NULL;
9037 * drop a subvolume tree.
9039 * this function traverses the tree freeing any blocks that only
9040 * referenced by the tree.
9042 * when a shared tree block is found. this function decreases its
9043 * reference count by one. if update_ref is true, this function
9044 * also make sure backrefs for the shared block and all lower level
9045 * blocks are properly updated.
9047 * If called with for_reloc == 0, may exit early with -EAGAIN
9049 int btrfs_drop_snapshot(struct btrfs_root *root,
9050 struct btrfs_block_rsv *block_rsv, int update_ref,
9053 struct btrfs_fs_info *fs_info = root->fs_info;
9054 struct btrfs_path *path;
9055 struct btrfs_trans_handle *trans;
9056 struct btrfs_root *tree_root = fs_info->tree_root;
9057 struct btrfs_root_item *root_item = &root->root_item;
9058 struct walk_control *wc;
9059 struct btrfs_key key;
9063 bool root_dropped = false;
9065 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9067 path = btrfs_alloc_path();
9073 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9075 btrfs_free_path(path);
9080 trans = btrfs_start_transaction(tree_root, 0);
9081 if (IS_ERR(trans)) {
9082 err = PTR_ERR(trans);
9087 trans->block_rsv = block_rsv;
9089 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9090 level = btrfs_header_level(root->node);
9091 path->nodes[level] = btrfs_lock_root_node(root);
9092 btrfs_set_lock_blocking(path->nodes[level]);
9093 path->slots[level] = 0;
9094 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9095 memset(&wc->update_progress, 0,
9096 sizeof(wc->update_progress));
9098 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9099 memcpy(&wc->update_progress, &key,
9100 sizeof(wc->update_progress));
9102 level = root_item->drop_level;
9104 path->lowest_level = level;
9105 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9106 path->lowest_level = 0;
9114 * unlock our path, this is safe because only this
9115 * function is allowed to delete this snapshot
9117 btrfs_unlock_up_safe(path, 0);
9119 level = btrfs_header_level(root->node);
9121 btrfs_tree_lock(path->nodes[level]);
9122 btrfs_set_lock_blocking(path->nodes[level]);
9123 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9125 ret = btrfs_lookup_extent_info(trans, fs_info,
9126 path->nodes[level]->start,
9127 level, 1, &wc->refs[level],
9133 BUG_ON(wc->refs[level] == 0);
9135 if (level == root_item->drop_level)
9138 btrfs_tree_unlock(path->nodes[level]);
9139 path->locks[level] = 0;
9140 WARN_ON(wc->refs[level] != 1);
9146 wc->shared_level = -1;
9147 wc->stage = DROP_REFERENCE;
9148 wc->update_ref = update_ref;
9150 wc->for_reloc = for_reloc;
9151 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9155 ret = walk_down_tree(trans, root, path, wc);
9161 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9168 BUG_ON(wc->stage != DROP_REFERENCE);
9172 if (wc->stage == DROP_REFERENCE) {
9174 btrfs_node_key(path->nodes[level],
9175 &root_item->drop_progress,
9176 path->slots[level]);
9177 root_item->drop_level = level;
9180 BUG_ON(wc->level == 0);
9181 if (btrfs_should_end_transaction(trans) ||
9182 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9183 ret = btrfs_update_root(trans, tree_root,
9187 btrfs_abort_transaction(trans, ret);
9192 btrfs_end_transaction_throttle(trans);
9193 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9194 btrfs_debug(fs_info,
9195 "drop snapshot early exit");
9200 trans = btrfs_start_transaction(tree_root, 0);
9201 if (IS_ERR(trans)) {
9202 err = PTR_ERR(trans);
9206 trans->block_rsv = block_rsv;
9209 btrfs_release_path(path);
9213 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9215 btrfs_abort_transaction(trans, ret);
9220 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9221 ret = btrfs_find_root(tree_root, &root->root_key, path,
9224 btrfs_abort_transaction(trans, ret);
9227 } else if (ret > 0) {
9228 /* if we fail to delete the orphan item this time
9229 * around, it'll get picked up the next time.
9231 * The most common failure here is just -ENOENT.
9233 btrfs_del_orphan_item(trans, tree_root,
9234 root->root_key.objectid);
9238 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9239 btrfs_add_dropped_root(trans, root);
9241 free_extent_buffer(root->node);
9242 free_extent_buffer(root->commit_root);
9243 btrfs_put_fs_root(root);
9245 root_dropped = true;
9247 btrfs_end_transaction_throttle(trans);
9250 btrfs_free_path(path);
9253 * So if we need to stop dropping the snapshot for whatever reason we
9254 * need to make sure to add it back to the dead root list so that we
9255 * keep trying to do the work later. This also cleans up roots if we
9256 * don't have it in the radix (like when we recover after a power fail
9257 * or unmount) so we don't leak memory.
9259 if (!for_reloc && !root_dropped)
9260 btrfs_add_dead_root(root);
9261 if (err && err != -EAGAIN)
9262 btrfs_handle_fs_error(fs_info, err, NULL);
9267 * drop subtree rooted at tree block 'node'.
9269 * NOTE: this function will unlock and release tree block 'node'
9270 * only used by relocation code
9272 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9273 struct btrfs_root *root,
9274 struct extent_buffer *node,
9275 struct extent_buffer *parent)
9277 struct btrfs_fs_info *fs_info = root->fs_info;
9278 struct btrfs_path *path;
9279 struct walk_control *wc;
9285 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9287 path = btrfs_alloc_path();
9291 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9293 btrfs_free_path(path);
9297 btrfs_assert_tree_locked(parent);
9298 parent_level = btrfs_header_level(parent);
9299 extent_buffer_get(parent);
9300 path->nodes[parent_level] = parent;
9301 path->slots[parent_level] = btrfs_header_nritems(parent);
9303 btrfs_assert_tree_locked(node);
9304 level = btrfs_header_level(node);
9305 path->nodes[level] = node;
9306 path->slots[level] = 0;
9307 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9309 wc->refs[parent_level] = 1;
9310 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9312 wc->shared_level = -1;
9313 wc->stage = DROP_REFERENCE;
9317 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9320 wret = walk_down_tree(trans, root, path, wc);
9326 wret = walk_up_tree(trans, root, path, wc, parent_level);
9334 btrfs_free_path(path);
9338 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9344 * if restripe for this chunk_type is on pick target profile and
9345 * return, otherwise do the usual balance
9347 stripped = get_restripe_target(fs_info, flags);
9349 return extended_to_chunk(stripped);
9351 num_devices = fs_info->fs_devices->rw_devices;
9353 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9354 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9355 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9357 if (num_devices == 1) {
9358 stripped |= BTRFS_BLOCK_GROUP_DUP;
9359 stripped = flags & ~stripped;
9361 /* turn raid0 into single device chunks */
9362 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9365 /* turn mirroring into duplication */
9366 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9367 BTRFS_BLOCK_GROUP_RAID10))
9368 return stripped | BTRFS_BLOCK_GROUP_DUP;
9370 /* they already had raid on here, just return */
9371 if (flags & stripped)
9374 stripped |= BTRFS_BLOCK_GROUP_DUP;
9375 stripped = flags & ~stripped;
9377 /* switch duplicated blocks with raid1 */
9378 if (flags & BTRFS_BLOCK_GROUP_DUP)
9379 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9381 /* this is drive concat, leave it alone */
9387 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9389 struct btrfs_space_info *sinfo = cache->space_info;
9391 u64 min_allocable_bytes;
9395 * We need some metadata space and system metadata space for
9396 * allocating chunks in some corner cases until we force to set
9397 * it to be readonly.
9400 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9402 min_allocable_bytes = SZ_1M;
9404 min_allocable_bytes = 0;
9406 spin_lock(&sinfo->lock);
9407 spin_lock(&cache->lock);
9415 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9416 cache->bytes_super - btrfs_block_group_used(&cache->item);
9418 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9419 min_allocable_bytes <= sinfo->total_bytes) {
9420 sinfo->bytes_readonly += num_bytes;
9422 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9426 spin_unlock(&cache->lock);
9427 spin_unlock(&sinfo->lock);
9431 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9432 struct btrfs_block_group_cache *cache)
9435 struct btrfs_trans_handle *trans;
9440 trans = btrfs_join_transaction(fs_info->extent_root);
9442 return PTR_ERR(trans);
9445 * we're not allowed to set block groups readonly after the dirty
9446 * block groups cache has started writing. If it already started,
9447 * back off and let this transaction commit
9449 mutex_lock(&fs_info->ro_block_group_mutex);
9450 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9451 u64 transid = trans->transid;
9453 mutex_unlock(&fs_info->ro_block_group_mutex);
9454 btrfs_end_transaction(trans);
9456 ret = btrfs_wait_for_commit(fs_info, transid);
9463 * if we are changing raid levels, try to allocate a corresponding
9464 * block group with the new raid level.
9466 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9467 if (alloc_flags != cache->flags) {
9468 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9471 * ENOSPC is allowed here, we may have enough space
9472 * already allocated at the new raid level to
9481 ret = inc_block_group_ro(cache, 0);
9484 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9485 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9489 ret = inc_block_group_ro(cache, 0);
9491 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9492 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9493 mutex_lock(&fs_info->chunk_mutex);
9494 check_system_chunk(trans, fs_info, alloc_flags);
9495 mutex_unlock(&fs_info->chunk_mutex);
9497 mutex_unlock(&fs_info->ro_block_group_mutex);
9499 btrfs_end_transaction(trans);
9503 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9504 struct btrfs_fs_info *fs_info, u64 type)
9506 u64 alloc_flags = get_alloc_profile(fs_info, type);
9508 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9512 * helper to account the unused space of all the readonly block group in the
9513 * space_info. takes mirrors into account.
9515 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9517 struct btrfs_block_group_cache *block_group;
9521 /* It's df, we don't care if it's racy */
9522 if (list_empty(&sinfo->ro_bgs))
9525 spin_lock(&sinfo->lock);
9526 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9527 spin_lock(&block_group->lock);
9529 if (!block_group->ro) {
9530 spin_unlock(&block_group->lock);
9534 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9535 BTRFS_BLOCK_GROUP_RAID10 |
9536 BTRFS_BLOCK_GROUP_DUP))
9541 free_bytes += (block_group->key.offset -
9542 btrfs_block_group_used(&block_group->item)) *
9545 spin_unlock(&block_group->lock);
9547 spin_unlock(&sinfo->lock);
9552 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9554 struct btrfs_space_info *sinfo = cache->space_info;
9559 spin_lock(&sinfo->lock);
9560 spin_lock(&cache->lock);
9562 num_bytes = cache->key.offset - cache->reserved -
9563 cache->pinned - cache->bytes_super -
9564 btrfs_block_group_used(&cache->item);
9565 sinfo->bytes_readonly -= num_bytes;
9566 list_del_init(&cache->ro_list);
9568 spin_unlock(&cache->lock);
9569 spin_unlock(&sinfo->lock);
9573 * checks to see if its even possible to relocate this block group.
9575 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9576 * ok to go ahead and try.
9578 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9580 struct btrfs_root *root = fs_info->extent_root;
9581 struct btrfs_block_group_cache *block_group;
9582 struct btrfs_space_info *space_info;
9583 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9584 struct btrfs_device *device;
9585 struct btrfs_trans_handle *trans;
9595 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9597 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9599 /* odd, couldn't find the block group, leave it alone */
9603 "can't find block group for bytenr %llu",
9608 min_free = btrfs_block_group_used(&block_group->item);
9610 /* no bytes used, we're good */
9614 space_info = block_group->space_info;
9615 spin_lock(&space_info->lock);
9617 full = space_info->full;
9620 * if this is the last block group we have in this space, we can't
9621 * relocate it unless we're able to allocate a new chunk below.
9623 * Otherwise, we need to make sure we have room in the space to handle
9624 * all of the extents from this block group. If we can, we're good
9626 if ((space_info->total_bytes != block_group->key.offset) &&
9627 (btrfs_space_info_used(space_info, false) + min_free <
9628 space_info->total_bytes)) {
9629 spin_unlock(&space_info->lock);
9632 spin_unlock(&space_info->lock);
9635 * ok we don't have enough space, but maybe we have free space on our
9636 * devices to allocate new chunks for relocation, so loop through our
9637 * alloc devices and guess if we have enough space. if this block
9638 * group is going to be restriped, run checks against the target
9639 * profile instead of the current one.
9651 target = get_restripe_target(fs_info, block_group->flags);
9653 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9656 * this is just a balance, so if we were marked as full
9657 * we know there is no space for a new chunk
9662 "no space to alloc new chunk for block group %llu",
9663 block_group->key.objectid);
9667 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9670 if (index == BTRFS_RAID_RAID10) {
9674 } else if (index == BTRFS_RAID_RAID1) {
9676 } else if (index == BTRFS_RAID_DUP) {
9679 } else if (index == BTRFS_RAID_RAID0) {
9680 dev_min = fs_devices->rw_devices;
9681 min_free = div64_u64(min_free, dev_min);
9684 /* We need to do this so that we can look at pending chunks */
9685 trans = btrfs_join_transaction(root);
9686 if (IS_ERR(trans)) {
9687 ret = PTR_ERR(trans);
9691 mutex_lock(&fs_info->chunk_mutex);
9692 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9696 * check to make sure we can actually find a chunk with enough
9697 * space to fit our block group in.
9699 if (device->total_bytes > device->bytes_used + min_free &&
9700 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9701 ret = find_free_dev_extent(trans, device, min_free,
9706 if (dev_nr >= dev_min)
9712 if (debug && ret == -1)
9714 "no space to allocate a new chunk for block group %llu",
9715 block_group->key.objectid);
9716 mutex_unlock(&fs_info->chunk_mutex);
9717 btrfs_end_transaction(trans);
9719 btrfs_put_block_group(block_group);
9723 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9724 struct btrfs_path *path,
9725 struct btrfs_key *key)
9727 struct btrfs_root *root = fs_info->extent_root;
9729 struct btrfs_key found_key;
9730 struct extent_buffer *leaf;
9733 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9738 slot = path->slots[0];
9739 leaf = path->nodes[0];
9740 if (slot >= btrfs_header_nritems(leaf)) {
9741 ret = btrfs_next_leaf(root, path);
9748 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9750 if (found_key.objectid >= key->objectid &&
9751 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9752 struct extent_map_tree *em_tree;
9753 struct extent_map *em;
9755 em_tree = &root->fs_info->mapping_tree.map_tree;
9756 read_lock(&em_tree->lock);
9757 em = lookup_extent_mapping(em_tree, found_key.objectid,
9759 read_unlock(&em_tree->lock);
9762 "logical %llu len %llu found bg but no related chunk",
9763 found_key.objectid, found_key.offset);
9768 free_extent_map(em);
9777 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9779 struct btrfs_block_group_cache *block_group;
9783 struct inode *inode;
9785 block_group = btrfs_lookup_first_block_group(info, last);
9786 while (block_group) {
9787 spin_lock(&block_group->lock);
9788 if (block_group->iref)
9790 spin_unlock(&block_group->lock);
9791 block_group = next_block_group(info, block_group);
9800 inode = block_group->inode;
9801 block_group->iref = 0;
9802 block_group->inode = NULL;
9803 spin_unlock(&block_group->lock);
9804 ASSERT(block_group->io_ctl.inode == NULL);
9806 last = block_group->key.objectid + block_group->key.offset;
9807 btrfs_put_block_group(block_group);
9812 * Must be called only after stopping all workers, since we could have block
9813 * group caching kthreads running, and therefore they could race with us if we
9814 * freed the block groups before stopping them.
9816 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9818 struct btrfs_block_group_cache *block_group;
9819 struct btrfs_space_info *space_info;
9820 struct btrfs_caching_control *caching_ctl;
9823 down_write(&info->commit_root_sem);
9824 while (!list_empty(&info->caching_block_groups)) {
9825 caching_ctl = list_entry(info->caching_block_groups.next,
9826 struct btrfs_caching_control, list);
9827 list_del(&caching_ctl->list);
9828 put_caching_control(caching_ctl);
9830 up_write(&info->commit_root_sem);
9832 spin_lock(&info->unused_bgs_lock);
9833 while (!list_empty(&info->unused_bgs)) {
9834 block_group = list_first_entry(&info->unused_bgs,
9835 struct btrfs_block_group_cache,
9837 list_del_init(&block_group->bg_list);
9838 btrfs_put_block_group(block_group);
9840 spin_unlock(&info->unused_bgs_lock);
9842 spin_lock(&info->block_group_cache_lock);
9843 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9844 block_group = rb_entry(n, struct btrfs_block_group_cache,
9846 rb_erase(&block_group->cache_node,
9847 &info->block_group_cache_tree);
9848 RB_CLEAR_NODE(&block_group->cache_node);
9849 spin_unlock(&info->block_group_cache_lock);
9851 down_write(&block_group->space_info->groups_sem);
9852 list_del(&block_group->list);
9853 up_write(&block_group->space_info->groups_sem);
9856 * We haven't cached this block group, which means we could
9857 * possibly have excluded extents on this block group.
9859 if (block_group->cached == BTRFS_CACHE_NO ||
9860 block_group->cached == BTRFS_CACHE_ERROR)
9861 free_excluded_extents(info, block_group);
9863 btrfs_remove_free_space_cache(block_group);
9864 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9865 ASSERT(list_empty(&block_group->dirty_list));
9866 ASSERT(list_empty(&block_group->io_list));
9867 ASSERT(list_empty(&block_group->bg_list));
9868 ASSERT(atomic_read(&block_group->count) == 1);
9869 btrfs_put_block_group(block_group);
9871 spin_lock(&info->block_group_cache_lock);
9873 spin_unlock(&info->block_group_cache_lock);
9875 /* now that all the block groups are freed, go through and
9876 * free all the space_info structs. This is only called during
9877 * the final stages of unmount, and so we know nobody is
9878 * using them. We call synchronize_rcu() once before we start,
9879 * just to be on the safe side.
9883 release_global_block_rsv(info);
9885 while (!list_empty(&info->space_info)) {
9888 space_info = list_entry(info->space_info.next,
9889 struct btrfs_space_info,
9893 * Do not hide this behind enospc_debug, this is actually
9894 * important and indicates a real bug if this happens.
9896 if (WARN_ON(space_info->bytes_pinned > 0 ||
9897 space_info->bytes_reserved > 0 ||
9898 space_info->bytes_may_use > 0))
9899 dump_space_info(info, space_info, 0, 0);
9900 list_del(&space_info->list);
9901 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9902 struct kobject *kobj;
9903 kobj = space_info->block_group_kobjs[i];
9904 space_info->block_group_kobjs[i] = NULL;
9910 kobject_del(&space_info->kobj);
9911 kobject_put(&space_info->kobj);
9916 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9917 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9919 struct btrfs_space_info *space_info;
9920 struct raid_kobject *rkobj;
9925 spin_lock(&fs_info->pending_raid_kobjs_lock);
9926 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9927 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9929 list_for_each_entry(rkobj, &list, list) {
9930 space_info = __find_space_info(fs_info, rkobj->flags);
9931 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9933 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9934 "%s", get_raid_name(index));
9936 kobject_put(&rkobj->kobj);
9942 "failed to add kobject for block cache, ignoring");
9945 static void link_block_group(struct btrfs_block_group_cache *cache)
9947 struct btrfs_space_info *space_info = cache->space_info;
9948 struct btrfs_fs_info *fs_info = cache->fs_info;
9949 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9952 down_write(&space_info->groups_sem);
9953 if (list_empty(&space_info->block_groups[index]))
9955 list_add_tail(&cache->list, &space_info->block_groups[index]);
9956 up_write(&space_info->groups_sem);
9959 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9961 btrfs_warn(cache->fs_info,
9962 "couldn't alloc memory for raid level kobject");
9965 rkobj->flags = cache->flags;
9966 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9968 spin_lock(&fs_info->pending_raid_kobjs_lock);
9969 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
9970 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9971 space_info->block_group_kobjs[index] = &rkobj->kobj;
9975 static struct btrfs_block_group_cache *
9976 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9977 u64 start, u64 size)
9979 struct btrfs_block_group_cache *cache;
9981 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9985 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9987 if (!cache->free_space_ctl) {
9992 cache->key.objectid = start;
9993 cache->key.offset = size;
9994 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9996 cache->fs_info = fs_info;
9997 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
9998 set_free_space_tree_thresholds(cache);
10000 atomic_set(&cache->count, 1);
10001 spin_lock_init(&cache->lock);
10002 init_rwsem(&cache->data_rwsem);
10003 INIT_LIST_HEAD(&cache->list);
10004 INIT_LIST_HEAD(&cache->cluster_list);
10005 INIT_LIST_HEAD(&cache->bg_list);
10006 INIT_LIST_HEAD(&cache->ro_list);
10007 INIT_LIST_HEAD(&cache->dirty_list);
10008 INIT_LIST_HEAD(&cache->io_list);
10009 btrfs_init_free_space_ctl(cache);
10010 atomic_set(&cache->trimming, 0);
10011 mutex_init(&cache->free_space_lock);
10012 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10017 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10019 struct btrfs_path *path;
10021 struct btrfs_block_group_cache *cache;
10022 struct btrfs_space_info *space_info;
10023 struct btrfs_key key;
10024 struct btrfs_key found_key;
10025 struct extent_buffer *leaf;
10026 int need_clear = 0;
10031 feature = btrfs_super_incompat_flags(info->super_copy);
10032 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10036 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10037 path = btrfs_alloc_path();
10040 path->reada = READA_FORWARD;
10042 cache_gen = btrfs_super_cache_generation(info->super_copy);
10043 if (btrfs_test_opt(info, SPACE_CACHE) &&
10044 btrfs_super_generation(info->super_copy) != cache_gen)
10046 if (btrfs_test_opt(info, CLEAR_CACHE))
10050 ret = find_first_block_group(info, path, &key);
10056 leaf = path->nodes[0];
10057 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10059 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10068 * When we mount with old space cache, we need to
10069 * set BTRFS_DC_CLEAR and set dirty flag.
10071 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10072 * truncate the old free space cache inode and
10074 * b) Setting 'dirty flag' makes sure that we flush
10075 * the new space cache info onto disk.
10077 if (btrfs_test_opt(info, SPACE_CACHE))
10078 cache->disk_cache_state = BTRFS_DC_CLEAR;
10081 read_extent_buffer(leaf, &cache->item,
10082 btrfs_item_ptr_offset(leaf, path->slots[0]),
10083 sizeof(cache->item));
10084 cache->flags = btrfs_block_group_flags(&cache->item);
10086 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10087 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10089 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10090 cache->key.objectid);
10095 key.objectid = found_key.objectid + found_key.offset;
10096 btrfs_release_path(path);
10099 * We need to exclude the super stripes now so that the space
10100 * info has super bytes accounted for, otherwise we'll think
10101 * we have more space than we actually do.
10103 ret = exclude_super_stripes(info, cache);
10106 * We may have excluded something, so call this just in
10109 free_excluded_extents(info, cache);
10110 btrfs_put_block_group(cache);
10115 * check for two cases, either we are full, and therefore
10116 * don't need to bother with the caching work since we won't
10117 * find any space, or we are empty, and we can just add all
10118 * the space in and be done with it. This saves us _alot_ of
10119 * time, particularly in the full case.
10121 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10122 cache->last_byte_to_unpin = (u64)-1;
10123 cache->cached = BTRFS_CACHE_FINISHED;
10124 free_excluded_extents(info, cache);
10125 } else if (btrfs_block_group_used(&cache->item) == 0) {
10126 cache->last_byte_to_unpin = (u64)-1;
10127 cache->cached = BTRFS_CACHE_FINISHED;
10128 add_new_free_space(cache, info,
10129 found_key.objectid,
10130 found_key.objectid +
10132 free_excluded_extents(info, cache);
10135 ret = btrfs_add_block_group_cache(info, cache);
10137 btrfs_remove_free_space_cache(cache);
10138 btrfs_put_block_group(cache);
10142 trace_btrfs_add_block_group(info, cache, 0);
10143 update_space_info(info, cache->flags, found_key.offset,
10144 btrfs_block_group_used(&cache->item),
10145 cache->bytes_super, &space_info);
10147 cache->space_info = space_info;
10149 link_block_group(cache);
10151 set_avail_alloc_bits(info, cache->flags);
10152 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10153 inc_block_group_ro(cache, 1);
10154 } else if (btrfs_block_group_used(&cache->item) == 0) {
10155 spin_lock(&info->unused_bgs_lock);
10156 /* Should always be true but just in case. */
10157 if (list_empty(&cache->bg_list)) {
10158 btrfs_get_block_group(cache);
10159 list_add_tail(&cache->bg_list,
10160 &info->unused_bgs);
10162 spin_unlock(&info->unused_bgs_lock);
10166 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10167 if (!(get_alloc_profile(info, space_info->flags) &
10168 (BTRFS_BLOCK_GROUP_RAID10 |
10169 BTRFS_BLOCK_GROUP_RAID1 |
10170 BTRFS_BLOCK_GROUP_RAID5 |
10171 BTRFS_BLOCK_GROUP_RAID6 |
10172 BTRFS_BLOCK_GROUP_DUP)))
10175 * avoid allocating from un-mirrored block group if there are
10176 * mirrored block groups.
10178 list_for_each_entry(cache,
10179 &space_info->block_groups[BTRFS_RAID_RAID0],
10181 inc_block_group_ro(cache, 1);
10182 list_for_each_entry(cache,
10183 &space_info->block_groups[BTRFS_RAID_SINGLE],
10185 inc_block_group_ro(cache, 1);
10188 btrfs_add_raid_kobjects(info);
10189 init_global_block_rsv(info);
10192 btrfs_free_path(path);
10196 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10198 struct btrfs_fs_info *fs_info = trans->fs_info;
10199 struct btrfs_block_group_cache *block_group, *tmp;
10200 struct btrfs_root *extent_root = fs_info->extent_root;
10201 struct btrfs_block_group_item item;
10202 struct btrfs_key key;
10204 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10206 trans->can_flush_pending_bgs = false;
10207 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10211 spin_lock(&block_group->lock);
10212 memcpy(&item, &block_group->item, sizeof(item));
10213 memcpy(&key, &block_group->key, sizeof(key));
10214 spin_unlock(&block_group->lock);
10216 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10219 btrfs_abort_transaction(trans, ret);
10220 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10223 btrfs_abort_transaction(trans, ret);
10224 add_block_group_free_space(trans, fs_info, block_group);
10225 /* already aborted the transaction if it failed. */
10227 list_del_init(&block_group->bg_list);
10229 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10232 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10233 struct btrfs_fs_info *fs_info, u64 bytes_used,
10234 u64 type, u64 chunk_offset, u64 size)
10236 struct btrfs_block_group_cache *cache;
10239 btrfs_set_log_full_commit(fs_info, trans);
10241 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10245 btrfs_set_block_group_used(&cache->item, bytes_used);
10246 btrfs_set_block_group_chunk_objectid(&cache->item,
10247 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10248 btrfs_set_block_group_flags(&cache->item, type);
10250 cache->flags = type;
10251 cache->last_byte_to_unpin = (u64)-1;
10252 cache->cached = BTRFS_CACHE_FINISHED;
10253 cache->needs_free_space = 1;
10254 ret = exclude_super_stripes(fs_info, cache);
10257 * We may have excluded something, so call this just in
10260 free_excluded_extents(fs_info, cache);
10261 btrfs_put_block_group(cache);
10265 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10267 free_excluded_extents(fs_info, cache);
10269 #ifdef CONFIG_BTRFS_DEBUG
10270 if (btrfs_should_fragment_free_space(cache)) {
10271 u64 new_bytes_used = size - bytes_used;
10273 bytes_used += new_bytes_used >> 1;
10274 fragment_free_space(cache);
10278 * Ensure the corresponding space_info object is created and
10279 * assigned to our block group. We want our bg to be added to the rbtree
10280 * with its ->space_info set.
10282 cache->space_info = __find_space_info(fs_info, cache->flags);
10283 ASSERT(cache->space_info);
10285 ret = btrfs_add_block_group_cache(fs_info, cache);
10287 btrfs_remove_free_space_cache(cache);
10288 btrfs_put_block_group(cache);
10293 * Now that our block group has its ->space_info set and is inserted in
10294 * the rbtree, update the space info's counters.
10296 trace_btrfs_add_block_group(fs_info, cache, 1);
10297 update_space_info(fs_info, cache->flags, size, bytes_used,
10298 cache->bytes_super, &cache->space_info);
10299 update_global_block_rsv(fs_info);
10301 link_block_group(cache);
10303 list_add_tail(&cache->bg_list, &trans->new_bgs);
10305 set_avail_alloc_bits(fs_info, type);
10309 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10311 u64 extra_flags = chunk_to_extended(flags) &
10312 BTRFS_EXTENDED_PROFILE_MASK;
10314 write_seqlock(&fs_info->profiles_lock);
10315 if (flags & BTRFS_BLOCK_GROUP_DATA)
10316 fs_info->avail_data_alloc_bits &= ~extra_flags;
10317 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10318 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10319 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10320 fs_info->avail_system_alloc_bits &= ~extra_flags;
10321 write_sequnlock(&fs_info->profiles_lock);
10324 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10325 struct btrfs_fs_info *fs_info, u64 group_start,
10326 struct extent_map *em)
10328 struct btrfs_root *root = fs_info->extent_root;
10329 struct btrfs_path *path;
10330 struct btrfs_block_group_cache *block_group;
10331 struct btrfs_free_cluster *cluster;
10332 struct btrfs_root *tree_root = fs_info->tree_root;
10333 struct btrfs_key key;
10334 struct inode *inode;
10335 struct kobject *kobj = NULL;
10339 struct btrfs_caching_control *caching_ctl = NULL;
10342 block_group = btrfs_lookup_block_group(fs_info, group_start);
10343 BUG_ON(!block_group);
10344 BUG_ON(!block_group->ro);
10347 * Free the reserved super bytes from this block group before
10350 free_excluded_extents(fs_info, block_group);
10351 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10352 block_group->key.offset);
10354 memcpy(&key, &block_group->key, sizeof(key));
10355 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10356 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10357 BTRFS_BLOCK_GROUP_RAID1 |
10358 BTRFS_BLOCK_GROUP_RAID10))
10363 /* make sure this block group isn't part of an allocation cluster */
10364 cluster = &fs_info->data_alloc_cluster;
10365 spin_lock(&cluster->refill_lock);
10366 btrfs_return_cluster_to_free_space(block_group, cluster);
10367 spin_unlock(&cluster->refill_lock);
10370 * make sure this block group isn't part of a metadata
10371 * allocation cluster
10373 cluster = &fs_info->meta_alloc_cluster;
10374 spin_lock(&cluster->refill_lock);
10375 btrfs_return_cluster_to_free_space(block_group, cluster);
10376 spin_unlock(&cluster->refill_lock);
10378 path = btrfs_alloc_path();
10385 * get the inode first so any iput calls done for the io_list
10386 * aren't the final iput (no unlinks allowed now)
10388 inode = lookup_free_space_inode(fs_info, block_group, path);
10390 mutex_lock(&trans->transaction->cache_write_mutex);
10392 * make sure our free spache cache IO is done before remove the
10395 spin_lock(&trans->transaction->dirty_bgs_lock);
10396 if (!list_empty(&block_group->io_list)) {
10397 list_del_init(&block_group->io_list);
10399 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10401 spin_unlock(&trans->transaction->dirty_bgs_lock);
10402 btrfs_wait_cache_io(trans, block_group, path);
10403 btrfs_put_block_group(block_group);
10404 spin_lock(&trans->transaction->dirty_bgs_lock);
10407 if (!list_empty(&block_group->dirty_list)) {
10408 list_del_init(&block_group->dirty_list);
10409 btrfs_put_block_group(block_group);
10411 spin_unlock(&trans->transaction->dirty_bgs_lock);
10412 mutex_unlock(&trans->transaction->cache_write_mutex);
10414 if (!IS_ERR(inode)) {
10415 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10417 btrfs_add_delayed_iput(inode);
10420 clear_nlink(inode);
10421 /* One for the block groups ref */
10422 spin_lock(&block_group->lock);
10423 if (block_group->iref) {
10424 block_group->iref = 0;
10425 block_group->inode = NULL;
10426 spin_unlock(&block_group->lock);
10429 spin_unlock(&block_group->lock);
10431 /* One for our lookup ref */
10432 btrfs_add_delayed_iput(inode);
10435 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10436 key.offset = block_group->key.objectid;
10439 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10443 btrfs_release_path(path);
10445 ret = btrfs_del_item(trans, tree_root, path);
10448 btrfs_release_path(path);
10451 spin_lock(&fs_info->block_group_cache_lock);
10452 rb_erase(&block_group->cache_node,
10453 &fs_info->block_group_cache_tree);
10454 RB_CLEAR_NODE(&block_group->cache_node);
10456 if (fs_info->first_logical_byte == block_group->key.objectid)
10457 fs_info->first_logical_byte = (u64)-1;
10458 spin_unlock(&fs_info->block_group_cache_lock);
10460 down_write(&block_group->space_info->groups_sem);
10462 * we must use list_del_init so people can check to see if they
10463 * are still on the list after taking the semaphore
10465 list_del_init(&block_group->list);
10466 if (list_empty(&block_group->space_info->block_groups[index])) {
10467 kobj = block_group->space_info->block_group_kobjs[index];
10468 block_group->space_info->block_group_kobjs[index] = NULL;
10469 clear_avail_alloc_bits(fs_info, block_group->flags);
10471 up_write(&block_group->space_info->groups_sem);
10477 if (block_group->has_caching_ctl)
10478 caching_ctl = get_caching_control(block_group);
10479 if (block_group->cached == BTRFS_CACHE_STARTED)
10480 wait_block_group_cache_done(block_group);
10481 if (block_group->has_caching_ctl) {
10482 down_write(&fs_info->commit_root_sem);
10483 if (!caching_ctl) {
10484 struct btrfs_caching_control *ctl;
10486 list_for_each_entry(ctl,
10487 &fs_info->caching_block_groups, list)
10488 if (ctl->block_group == block_group) {
10490 refcount_inc(&caching_ctl->count);
10495 list_del_init(&caching_ctl->list);
10496 up_write(&fs_info->commit_root_sem);
10498 /* Once for the caching bgs list and once for us. */
10499 put_caching_control(caching_ctl);
10500 put_caching_control(caching_ctl);
10504 spin_lock(&trans->transaction->dirty_bgs_lock);
10505 if (!list_empty(&block_group->dirty_list)) {
10508 if (!list_empty(&block_group->io_list)) {
10511 spin_unlock(&trans->transaction->dirty_bgs_lock);
10512 btrfs_remove_free_space_cache(block_group);
10514 spin_lock(&block_group->space_info->lock);
10515 list_del_init(&block_group->ro_list);
10517 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10518 WARN_ON(block_group->space_info->total_bytes
10519 < block_group->key.offset);
10520 WARN_ON(block_group->space_info->bytes_readonly
10521 < block_group->key.offset);
10522 WARN_ON(block_group->space_info->disk_total
10523 < block_group->key.offset * factor);
10525 block_group->space_info->total_bytes -= block_group->key.offset;
10526 block_group->space_info->bytes_readonly -= block_group->key.offset;
10527 block_group->space_info->disk_total -= block_group->key.offset * factor;
10529 spin_unlock(&block_group->space_info->lock);
10531 memcpy(&key, &block_group->key, sizeof(key));
10533 mutex_lock(&fs_info->chunk_mutex);
10534 if (!list_empty(&em->list)) {
10535 /* We're in the transaction->pending_chunks list. */
10536 free_extent_map(em);
10538 spin_lock(&block_group->lock);
10539 block_group->removed = 1;
10541 * At this point trimming can't start on this block group, because we
10542 * removed the block group from the tree fs_info->block_group_cache_tree
10543 * so no one can't find it anymore and even if someone already got this
10544 * block group before we removed it from the rbtree, they have already
10545 * incremented block_group->trimming - if they didn't, they won't find
10546 * any free space entries because we already removed them all when we
10547 * called btrfs_remove_free_space_cache().
10549 * And we must not remove the extent map from the fs_info->mapping_tree
10550 * to prevent the same logical address range and physical device space
10551 * ranges from being reused for a new block group. This is because our
10552 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10553 * completely transactionless, so while it is trimming a range the
10554 * currently running transaction might finish and a new one start,
10555 * allowing for new block groups to be created that can reuse the same
10556 * physical device locations unless we take this special care.
10558 * There may also be an implicit trim operation if the file system
10559 * is mounted with -odiscard. The same protections must remain
10560 * in place until the extents have been discarded completely when
10561 * the transaction commit has completed.
10563 remove_em = (atomic_read(&block_group->trimming) == 0);
10565 * Make sure a trimmer task always sees the em in the pinned_chunks list
10566 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10567 * before checking block_group->removed).
10571 * Our em might be in trans->transaction->pending_chunks which
10572 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10573 * and so is the fs_info->pinned_chunks list.
10575 * So at this point we must be holding the chunk_mutex to avoid
10576 * any races with chunk allocation (more specifically at
10577 * volumes.c:contains_pending_extent()), to ensure it always
10578 * sees the em, either in the pending_chunks list or in the
10579 * pinned_chunks list.
10581 list_move_tail(&em->list, &fs_info->pinned_chunks);
10583 spin_unlock(&block_group->lock);
10586 struct extent_map_tree *em_tree;
10588 em_tree = &fs_info->mapping_tree.map_tree;
10589 write_lock(&em_tree->lock);
10591 * The em might be in the pending_chunks list, so make sure the
10592 * chunk mutex is locked, since remove_extent_mapping() will
10593 * delete us from that list.
10595 remove_extent_mapping(em_tree, em);
10596 write_unlock(&em_tree->lock);
10597 /* once for the tree */
10598 free_extent_map(em);
10601 mutex_unlock(&fs_info->chunk_mutex);
10603 ret = remove_block_group_free_space(trans, fs_info, block_group);
10607 btrfs_put_block_group(block_group);
10608 btrfs_put_block_group(block_group);
10610 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10616 ret = btrfs_del_item(trans, root, path);
10618 btrfs_free_path(path);
10622 struct btrfs_trans_handle *
10623 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10624 const u64 chunk_offset)
10626 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10627 struct extent_map *em;
10628 struct map_lookup *map;
10629 unsigned int num_items;
10631 read_lock(&em_tree->lock);
10632 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10633 read_unlock(&em_tree->lock);
10634 ASSERT(em && em->start == chunk_offset);
10637 * We need to reserve 3 + N units from the metadata space info in order
10638 * to remove a block group (done at btrfs_remove_chunk() and at
10639 * btrfs_remove_block_group()), which are used for:
10641 * 1 unit for adding the free space inode's orphan (located in the tree
10643 * 1 unit for deleting the block group item (located in the extent
10645 * 1 unit for deleting the free space item (located in tree of tree
10647 * N units for deleting N device extent items corresponding to each
10648 * stripe (located in the device tree).
10650 * In order to remove a block group we also need to reserve units in the
10651 * system space info in order to update the chunk tree (update one or
10652 * more device items and remove one chunk item), but this is done at
10653 * btrfs_remove_chunk() through a call to check_system_chunk().
10655 map = em->map_lookup;
10656 num_items = 3 + map->num_stripes;
10657 free_extent_map(em);
10659 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10664 * Process the unused_bgs list and remove any that don't have any allocated
10665 * space inside of them.
10667 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10669 struct btrfs_block_group_cache *block_group;
10670 struct btrfs_space_info *space_info;
10671 struct btrfs_trans_handle *trans;
10674 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10677 spin_lock(&fs_info->unused_bgs_lock);
10678 while (!list_empty(&fs_info->unused_bgs)) {
10682 block_group = list_first_entry(&fs_info->unused_bgs,
10683 struct btrfs_block_group_cache,
10685 list_del_init(&block_group->bg_list);
10687 space_info = block_group->space_info;
10689 if (ret || btrfs_mixed_space_info(space_info)) {
10690 btrfs_put_block_group(block_group);
10693 spin_unlock(&fs_info->unused_bgs_lock);
10695 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10697 /* Don't want to race with allocators so take the groups_sem */
10698 down_write(&space_info->groups_sem);
10699 spin_lock(&block_group->lock);
10700 if (block_group->reserved ||
10701 btrfs_block_group_used(&block_group->item) ||
10703 list_is_singular(&block_group->list)) {
10705 * We want to bail if we made new allocations or have
10706 * outstanding allocations in this block group. We do
10707 * the ro check in case balance is currently acting on
10708 * this block group.
10710 spin_unlock(&block_group->lock);
10711 up_write(&space_info->groups_sem);
10714 spin_unlock(&block_group->lock);
10716 /* We don't want to force the issue, only flip if it's ok. */
10717 ret = inc_block_group_ro(block_group, 0);
10718 up_write(&space_info->groups_sem);
10725 * Want to do this before we do anything else so we can recover
10726 * properly if we fail to join the transaction.
10728 trans = btrfs_start_trans_remove_block_group(fs_info,
10729 block_group->key.objectid);
10730 if (IS_ERR(trans)) {
10731 btrfs_dec_block_group_ro(block_group);
10732 ret = PTR_ERR(trans);
10737 * We could have pending pinned extents for this block group,
10738 * just delete them, we don't care about them anymore.
10740 start = block_group->key.objectid;
10741 end = start + block_group->key.offset - 1;
10743 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10744 * btrfs_finish_extent_commit(). If we are at transaction N,
10745 * another task might be running finish_extent_commit() for the
10746 * previous transaction N - 1, and have seen a range belonging
10747 * to the block group in freed_extents[] before we were able to
10748 * clear the whole block group range from freed_extents[]. This
10749 * means that task can lookup for the block group after we
10750 * unpinned it from freed_extents[] and removed it, leading to
10751 * a BUG_ON() at btrfs_unpin_extent_range().
10753 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10754 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10757 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10758 btrfs_dec_block_group_ro(block_group);
10761 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10764 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10765 btrfs_dec_block_group_ro(block_group);
10768 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10770 /* Reset pinned so btrfs_put_block_group doesn't complain */
10771 spin_lock(&space_info->lock);
10772 spin_lock(&block_group->lock);
10774 space_info->bytes_pinned -= block_group->pinned;
10775 space_info->bytes_readonly += block_group->pinned;
10776 percpu_counter_add(&space_info->total_bytes_pinned,
10777 -block_group->pinned);
10778 block_group->pinned = 0;
10780 spin_unlock(&block_group->lock);
10781 spin_unlock(&space_info->lock);
10783 /* DISCARD can flip during remount */
10784 trimming = btrfs_test_opt(fs_info, DISCARD);
10786 /* Implicit trim during transaction commit. */
10788 btrfs_get_block_group_trimming(block_group);
10791 * Btrfs_remove_chunk will abort the transaction if things go
10794 ret = btrfs_remove_chunk(trans, fs_info,
10795 block_group->key.objectid);
10799 btrfs_put_block_group_trimming(block_group);
10804 * If we're not mounted with -odiscard, we can just forget
10805 * about this block group. Otherwise we'll need to wait
10806 * until transaction commit to do the actual discard.
10809 spin_lock(&fs_info->unused_bgs_lock);
10811 * A concurrent scrub might have added us to the list
10812 * fs_info->unused_bgs, so use a list_move operation
10813 * to add the block group to the deleted_bgs list.
10815 list_move(&block_group->bg_list,
10816 &trans->transaction->deleted_bgs);
10817 spin_unlock(&fs_info->unused_bgs_lock);
10818 btrfs_get_block_group(block_group);
10821 btrfs_end_transaction(trans);
10823 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10824 btrfs_put_block_group(block_group);
10825 spin_lock(&fs_info->unused_bgs_lock);
10827 spin_unlock(&fs_info->unused_bgs_lock);
10830 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10832 struct btrfs_space_info *space_info;
10833 struct btrfs_super_block *disk_super;
10839 disk_super = fs_info->super_copy;
10840 if (!btrfs_super_root(disk_super))
10843 features = btrfs_super_incompat_flags(disk_super);
10844 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10847 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10848 ret = create_space_info(fs_info, flags, &space_info);
10853 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10854 ret = create_space_info(fs_info, flags, &space_info);
10856 flags = BTRFS_BLOCK_GROUP_METADATA;
10857 ret = create_space_info(fs_info, flags, &space_info);
10861 flags = BTRFS_BLOCK_GROUP_DATA;
10862 ret = create_space_info(fs_info, flags, &space_info);
10868 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10869 u64 start, u64 end)
10871 return unpin_extent_range(fs_info, start, end, false);
10875 * It used to be that old block groups would be left around forever.
10876 * Iterating over them would be enough to trim unused space. Since we
10877 * now automatically remove them, we also need to iterate over unallocated
10880 * We don't want a transaction for this since the discard may take a
10881 * substantial amount of time. We don't require that a transaction be
10882 * running, but we do need to take a running transaction into account
10883 * to ensure that we're not discarding chunks that were released in
10884 * the current transaction.
10886 * Holding the chunks lock will prevent other threads from allocating
10887 * or releasing chunks, but it won't prevent a running transaction
10888 * from committing and releasing the memory that the pending chunks
10889 * list head uses. For that, we need to take a reference to the
10892 static int btrfs_trim_free_extents(struct btrfs_device *device,
10893 u64 minlen, u64 *trimmed)
10895 u64 start = 0, len = 0;
10900 /* Not writeable = nothing to do. */
10901 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10904 /* No free space = nothing to do. */
10905 if (device->total_bytes <= device->bytes_used)
10911 struct btrfs_fs_info *fs_info = device->fs_info;
10912 struct btrfs_transaction *trans;
10915 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10919 down_read(&fs_info->commit_root_sem);
10921 spin_lock(&fs_info->trans_lock);
10922 trans = fs_info->running_transaction;
10924 refcount_inc(&trans->use_count);
10925 spin_unlock(&fs_info->trans_lock);
10927 ret = find_free_dev_extent_start(trans, device, minlen, start,
10930 btrfs_put_transaction(trans);
10933 up_read(&fs_info->commit_root_sem);
10934 mutex_unlock(&fs_info->chunk_mutex);
10935 if (ret == -ENOSPC)
10940 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10941 up_read(&fs_info->commit_root_sem);
10942 mutex_unlock(&fs_info->chunk_mutex);
10950 if (fatal_signal_pending(current)) {
10951 ret = -ERESTARTSYS;
10961 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10963 struct btrfs_block_group_cache *cache = NULL;
10964 struct btrfs_device *device;
10965 struct list_head *devices;
10970 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10974 * try to trim all FS space, our block group may start from non-zero.
10976 if (range->len == total_bytes)
10977 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10979 cache = btrfs_lookup_block_group(fs_info, range->start);
10982 if (cache->key.objectid >= (range->start + range->len)) {
10983 btrfs_put_block_group(cache);
10987 start = max(range->start, cache->key.objectid);
10988 end = min(range->start + range->len,
10989 cache->key.objectid + cache->key.offset);
10991 if (end - start >= range->minlen) {
10992 if (!block_group_cache_done(cache)) {
10993 ret = cache_block_group(cache, 0);
10995 btrfs_put_block_group(cache);
10998 ret = wait_block_group_cache_done(cache);
11000 btrfs_put_block_group(cache);
11004 ret = btrfs_trim_block_group(cache,
11010 trimmed += group_trimmed;
11012 btrfs_put_block_group(cache);
11017 cache = next_block_group(fs_info, cache);
11020 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11021 devices = &fs_info->fs_devices->alloc_list;
11022 list_for_each_entry(device, devices, dev_alloc_list) {
11023 ret = btrfs_trim_free_extents(device, range->minlen,
11028 trimmed += group_trimmed;
11030 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11032 range->len = trimmed;
11037 * btrfs_{start,end}_write_no_snapshotting() are similar to
11038 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11039 * data into the page cache through nocow before the subvolume is snapshoted,
11040 * but flush the data into disk after the snapshot creation, or to prevent
11041 * operations while snapshotting is ongoing and that cause the snapshot to be
11042 * inconsistent (writes followed by expanding truncates for example).
11044 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11046 percpu_counter_dec(&root->subv_writers->counter);
11048 * Make sure counter is updated before we wake up waiters.
11051 if (waitqueue_active(&root->subv_writers->wait))
11052 wake_up(&root->subv_writers->wait);
11055 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11057 if (atomic_read(&root->will_be_snapshotted))
11060 percpu_counter_inc(&root->subv_writers->counter);
11062 * Make sure counter is updated before we check for snapshot creation.
11065 if (atomic_read(&root->will_be_snapshotted)) {
11066 btrfs_end_write_no_snapshotting(root);
11072 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11077 ret = btrfs_start_write_no_snapshotting(root);
11080 wait_var_event(&root->will_be_snapshotted,
11081 !atomic_read(&root->will_be_snapshotted));