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 bool metadata, u64 root_objectid)
749 struct btrfs_space_info *space_info;
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 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
2206 add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid);
2212 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2213 struct btrfs_fs_info *fs_info,
2214 struct btrfs_delayed_ref_node *node,
2215 u64 parent, u64 root_objectid,
2216 u64 owner, u64 offset, int refs_to_add,
2217 struct btrfs_delayed_extent_op *extent_op)
2219 struct btrfs_path *path;
2220 struct extent_buffer *leaf;
2221 struct btrfs_extent_item *item;
2222 struct btrfs_key key;
2223 u64 bytenr = node->bytenr;
2224 u64 num_bytes = node->num_bytes;
2228 path = btrfs_alloc_path();
2232 path->reada = READA_FORWARD;
2233 path->leave_spinning = 1;
2234 /* this will setup the path even if it fails to insert the back ref */
2235 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2236 num_bytes, parent, root_objectid,
2238 refs_to_add, extent_op);
2239 if ((ret < 0 && ret != -EAGAIN) || !ret)
2243 * Ok we had -EAGAIN which means we didn't have space to insert and
2244 * inline extent ref, so just update the reference count and add a
2247 leaf = path->nodes[0];
2248 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2249 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2250 refs = btrfs_extent_refs(leaf, item);
2251 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2253 __run_delayed_extent_op(extent_op, leaf, item);
2255 btrfs_mark_buffer_dirty(leaf);
2256 btrfs_release_path(path);
2258 path->reada = READA_FORWARD;
2259 path->leave_spinning = 1;
2260 /* now insert the actual backref */
2261 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2262 root_objectid, owner, offset, refs_to_add);
2264 btrfs_abort_transaction(trans, ret);
2266 btrfs_free_path(path);
2270 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2271 struct btrfs_fs_info *fs_info,
2272 struct btrfs_delayed_ref_node *node,
2273 struct btrfs_delayed_extent_op *extent_op,
2274 int insert_reserved)
2277 struct btrfs_delayed_data_ref *ref;
2278 struct btrfs_key ins;
2283 ins.objectid = node->bytenr;
2284 ins.offset = node->num_bytes;
2285 ins.type = BTRFS_EXTENT_ITEM_KEY;
2287 ref = btrfs_delayed_node_to_data_ref(node);
2288 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2290 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2291 parent = ref->parent;
2292 ref_root = ref->root;
2294 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2296 flags |= extent_op->flags_to_set;
2297 ret = alloc_reserved_file_extent(trans, fs_info,
2298 parent, ref_root, flags,
2299 ref->objectid, ref->offset,
2300 &ins, node->ref_mod);
2301 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2302 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2303 ref_root, ref->objectid,
2304 ref->offset, node->ref_mod,
2306 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2307 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2308 ref_root, ref->objectid,
2309 ref->offset, node->ref_mod,
2317 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2318 struct extent_buffer *leaf,
2319 struct btrfs_extent_item *ei)
2321 u64 flags = btrfs_extent_flags(leaf, ei);
2322 if (extent_op->update_flags) {
2323 flags |= extent_op->flags_to_set;
2324 btrfs_set_extent_flags(leaf, ei, flags);
2327 if (extent_op->update_key) {
2328 struct btrfs_tree_block_info *bi;
2329 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2330 bi = (struct btrfs_tree_block_info *)(ei + 1);
2331 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2335 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2336 struct btrfs_fs_info *fs_info,
2337 struct btrfs_delayed_ref_head *head,
2338 struct btrfs_delayed_extent_op *extent_op)
2340 struct btrfs_key key;
2341 struct btrfs_path *path;
2342 struct btrfs_extent_item *ei;
2343 struct extent_buffer *leaf;
2347 int metadata = !extent_op->is_data;
2352 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2355 path = btrfs_alloc_path();
2359 key.objectid = head->bytenr;
2362 key.type = BTRFS_METADATA_ITEM_KEY;
2363 key.offset = extent_op->level;
2365 key.type = BTRFS_EXTENT_ITEM_KEY;
2366 key.offset = head->num_bytes;
2370 path->reada = READA_FORWARD;
2371 path->leave_spinning = 1;
2372 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2379 if (path->slots[0] > 0) {
2381 btrfs_item_key_to_cpu(path->nodes[0], &key,
2383 if (key.objectid == head->bytenr &&
2384 key.type == BTRFS_EXTENT_ITEM_KEY &&
2385 key.offset == head->num_bytes)
2389 btrfs_release_path(path);
2392 key.objectid = head->bytenr;
2393 key.offset = head->num_bytes;
2394 key.type = BTRFS_EXTENT_ITEM_KEY;
2403 leaf = path->nodes[0];
2404 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2405 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2406 if (item_size < sizeof(*ei)) {
2407 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2412 leaf = path->nodes[0];
2413 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2416 BUG_ON(item_size < sizeof(*ei));
2417 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2418 __run_delayed_extent_op(extent_op, leaf, ei);
2420 btrfs_mark_buffer_dirty(leaf);
2422 btrfs_free_path(path);
2426 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2427 struct btrfs_fs_info *fs_info,
2428 struct btrfs_delayed_ref_node *node,
2429 struct btrfs_delayed_extent_op *extent_op,
2430 int insert_reserved)
2433 struct btrfs_delayed_tree_ref *ref;
2434 struct btrfs_key ins;
2437 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2439 ref = btrfs_delayed_node_to_tree_ref(node);
2440 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2442 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2443 parent = ref->parent;
2444 ref_root = ref->root;
2446 ins.objectid = node->bytenr;
2447 if (skinny_metadata) {
2448 ins.offset = ref->level;
2449 ins.type = BTRFS_METADATA_ITEM_KEY;
2451 ins.offset = node->num_bytes;
2452 ins.type = BTRFS_EXTENT_ITEM_KEY;
2455 if (node->ref_mod != 1) {
2457 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2458 node->bytenr, node->ref_mod, node->action, ref_root,
2462 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2463 BUG_ON(!extent_op || !extent_op->update_flags);
2464 ret = alloc_reserved_tree_block(trans, fs_info,
2466 extent_op->flags_to_set,
2469 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2470 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2474 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2475 ret = __btrfs_free_extent(trans, fs_info, node,
2477 ref->level, 0, 1, extent_op);
2484 /* helper function to actually process a single delayed ref entry */
2485 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2486 struct btrfs_fs_info *fs_info,
2487 struct btrfs_delayed_ref_node *node,
2488 struct btrfs_delayed_extent_op *extent_op,
2489 int insert_reserved)
2493 if (trans->aborted) {
2494 if (insert_reserved)
2495 btrfs_pin_extent(fs_info, node->bytenr,
2496 node->num_bytes, 1);
2500 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2501 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2502 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2504 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2505 node->type == BTRFS_SHARED_DATA_REF_KEY)
2506 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2513 static inline struct btrfs_delayed_ref_node *
2514 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2516 struct btrfs_delayed_ref_node *ref;
2518 if (RB_EMPTY_ROOT(&head->ref_tree))
2522 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2523 * This is to prevent a ref count from going down to zero, which deletes
2524 * the extent item from the extent tree, when there still are references
2525 * to add, which would fail because they would not find the extent item.
2527 if (!list_empty(&head->ref_add_list))
2528 return list_first_entry(&head->ref_add_list,
2529 struct btrfs_delayed_ref_node, add_list);
2531 ref = rb_entry(rb_first(&head->ref_tree),
2532 struct btrfs_delayed_ref_node, ref_node);
2533 ASSERT(list_empty(&ref->add_list));
2537 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2538 struct btrfs_delayed_ref_head *head)
2540 spin_lock(&delayed_refs->lock);
2541 head->processing = 0;
2542 delayed_refs->num_heads_ready++;
2543 spin_unlock(&delayed_refs->lock);
2544 btrfs_delayed_ref_unlock(head);
2547 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2548 struct btrfs_fs_info *fs_info,
2549 struct btrfs_delayed_ref_head *head)
2551 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2556 head->extent_op = NULL;
2557 if (head->must_insert_reserved) {
2558 btrfs_free_delayed_extent_op(extent_op);
2561 spin_unlock(&head->lock);
2562 ret = run_delayed_extent_op(trans, fs_info, head, extent_op);
2563 btrfs_free_delayed_extent_op(extent_op);
2564 return ret ? ret : 1;
2567 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2568 struct btrfs_fs_info *fs_info,
2569 struct btrfs_delayed_ref_head *head)
2571 struct btrfs_delayed_ref_root *delayed_refs;
2574 delayed_refs = &trans->transaction->delayed_refs;
2576 ret = cleanup_extent_op(trans, fs_info, head);
2578 unselect_delayed_ref_head(delayed_refs, head);
2579 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2586 * Need to drop our head ref lock and re-acquire the delayed ref lock
2587 * and then re-check to make sure nobody got added.
2589 spin_unlock(&head->lock);
2590 spin_lock(&delayed_refs->lock);
2591 spin_lock(&head->lock);
2592 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2593 spin_unlock(&head->lock);
2594 spin_unlock(&delayed_refs->lock);
2597 delayed_refs->num_heads--;
2598 rb_erase(&head->href_node, &delayed_refs->href_root);
2599 RB_CLEAR_NODE(&head->href_node);
2600 spin_unlock(&head->lock);
2601 spin_unlock(&delayed_refs->lock);
2602 atomic_dec(&delayed_refs->num_entries);
2604 trace_run_delayed_ref_head(fs_info, head, 0);
2606 if (head->total_ref_mod < 0) {
2607 struct btrfs_space_info *space_info;
2611 flags = BTRFS_BLOCK_GROUP_DATA;
2612 else if (head->is_system)
2613 flags = BTRFS_BLOCK_GROUP_SYSTEM;
2615 flags = BTRFS_BLOCK_GROUP_METADATA;
2616 space_info = __find_space_info(fs_info, flags);
2618 percpu_counter_add(&space_info->total_bytes_pinned,
2621 if (head->is_data) {
2622 spin_lock(&delayed_refs->lock);
2623 delayed_refs->pending_csums -= head->num_bytes;
2624 spin_unlock(&delayed_refs->lock);
2628 if (head->must_insert_reserved) {
2629 btrfs_pin_extent(fs_info, head->bytenr,
2630 head->num_bytes, 1);
2631 if (head->is_data) {
2632 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2637 /* Also free its reserved qgroup space */
2638 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2639 head->qgroup_reserved);
2640 btrfs_delayed_ref_unlock(head);
2641 btrfs_put_delayed_ref_head(head);
2646 * Returns 0 on success or if called with an already aborted transaction.
2647 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2649 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2652 struct btrfs_fs_info *fs_info = trans->fs_info;
2653 struct btrfs_delayed_ref_root *delayed_refs;
2654 struct btrfs_delayed_ref_node *ref;
2655 struct btrfs_delayed_ref_head *locked_ref = NULL;
2656 struct btrfs_delayed_extent_op *extent_op;
2657 ktime_t start = ktime_get();
2659 unsigned long count = 0;
2660 unsigned long actual_count = 0;
2661 int must_insert_reserved = 0;
2663 delayed_refs = &trans->transaction->delayed_refs;
2669 spin_lock(&delayed_refs->lock);
2670 locked_ref = btrfs_select_ref_head(trans);
2672 spin_unlock(&delayed_refs->lock);
2676 /* grab the lock that says we are going to process
2677 * all the refs for this head */
2678 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2679 spin_unlock(&delayed_refs->lock);
2681 * we may have dropped the spin lock to get the head
2682 * mutex lock, and that might have given someone else
2683 * time to free the head. If that's true, it has been
2684 * removed from our list and we can move on.
2686 if (ret == -EAGAIN) {
2694 * We need to try and merge add/drops of the same ref since we
2695 * can run into issues with relocate dropping the implicit ref
2696 * and then it being added back again before the drop can
2697 * finish. If we merged anything we need to re-loop so we can
2699 * Or we can get node references of the same type that weren't
2700 * merged when created due to bumps in the tree mod seq, and
2701 * we need to merge them to prevent adding an inline extent
2702 * backref before dropping it (triggering a BUG_ON at
2703 * insert_inline_extent_backref()).
2705 spin_lock(&locked_ref->lock);
2706 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2710 * locked_ref is the head node, so we have to go one
2711 * node back for any delayed ref updates
2713 ref = select_delayed_ref(locked_ref);
2715 if (ref && ref->seq &&
2716 btrfs_check_delayed_seq(fs_info, ref->seq)) {
2717 spin_unlock(&locked_ref->lock);
2718 unselect_delayed_ref_head(delayed_refs, locked_ref);
2726 * We're done processing refs in this ref_head, clean everything
2727 * up and move on to the next ref_head.
2730 ret = cleanup_ref_head(trans, fs_info, locked_ref);
2732 /* We dropped our lock, we need to loop. */
2745 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2746 RB_CLEAR_NODE(&ref->ref_node);
2747 if (!list_empty(&ref->add_list))
2748 list_del(&ref->add_list);
2750 * When we play the delayed ref, also correct the ref_mod on
2753 switch (ref->action) {
2754 case BTRFS_ADD_DELAYED_REF:
2755 case BTRFS_ADD_DELAYED_EXTENT:
2756 locked_ref->ref_mod -= ref->ref_mod;
2758 case BTRFS_DROP_DELAYED_REF:
2759 locked_ref->ref_mod += ref->ref_mod;
2764 atomic_dec(&delayed_refs->num_entries);
2767 * Record the must-insert_reserved flag before we drop the spin
2770 must_insert_reserved = locked_ref->must_insert_reserved;
2771 locked_ref->must_insert_reserved = 0;
2773 extent_op = locked_ref->extent_op;
2774 locked_ref->extent_op = NULL;
2775 spin_unlock(&locked_ref->lock);
2777 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2778 must_insert_reserved);
2780 btrfs_free_delayed_extent_op(extent_op);
2782 unselect_delayed_ref_head(delayed_refs, locked_ref);
2783 btrfs_put_delayed_ref(ref);
2784 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2789 btrfs_put_delayed_ref(ref);
2795 * We don't want to include ref heads since we can have empty ref heads
2796 * and those will drastically skew our runtime down since we just do
2797 * accounting, no actual extent tree updates.
2799 if (actual_count > 0) {
2800 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2804 * We weigh the current average higher than our current runtime
2805 * to avoid large swings in the average.
2807 spin_lock(&delayed_refs->lock);
2808 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2809 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2810 spin_unlock(&delayed_refs->lock);
2815 #ifdef SCRAMBLE_DELAYED_REFS
2817 * Normally delayed refs get processed in ascending bytenr order. This
2818 * correlates in most cases to the order added. To expose dependencies on this
2819 * order, we start to process the tree in the middle instead of the beginning
2821 static u64 find_middle(struct rb_root *root)
2823 struct rb_node *n = root->rb_node;
2824 struct btrfs_delayed_ref_node *entry;
2827 u64 first = 0, last = 0;
2831 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2832 first = entry->bytenr;
2836 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2837 last = entry->bytenr;
2842 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2843 WARN_ON(!entry->in_tree);
2845 middle = entry->bytenr;
2858 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2862 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2863 sizeof(struct btrfs_extent_inline_ref));
2864 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2865 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2868 * We don't ever fill up leaves all the way so multiply by 2 just to be
2869 * closer to what we're really going to want to use.
2871 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2875 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2876 * would require to store the csums for that many bytes.
2878 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2881 u64 num_csums_per_leaf;
2884 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2885 num_csums_per_leaf = div64_u64(csum_size,
2886 (u64)btrfs_super_csum_size(fs_info->super_copy));
2887 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2888 num_csums += num_csums_per_leaf - 1;
2889 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2893 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2894 struct btrfs_fs_info *fs_info)
2896 struct btrfs_block_rsv *global_rsv;
2897 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2898 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2899 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2900 u64 num_bytes, num_dirty_bgs_bytes;
2903 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2904 num_heads = heads_to_leaves(fs_info, num_heads);
2906 num_bytes += (num_heads - 1) * fs_info->nodesize;
2908 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2910 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2912 global_rsv = &fs_info->global_block_rsv;
2915 * If we can't allocate any more chunks lets make sure we have _lots_ of
2916 * wiggle room since running delayed refs can create more delayed refs.
2918 if (global_rsv->space_info->full) {
2919 num_dirty_bgs_bytes <<= 1;
2923 spin_lock(&global_rsv->lock);
2924 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2926 spin_unlock(&global_rsv->lock);
2930 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2931 struct btrfs_fs_info *fs_info)
2934 atomic_read(&trans->transaction->delayed_refs.num_entries);
2939 avg_runtime = fs_info->avg_delayed_ref_runtime;
2940 val = num_entries * avg_runtime;
2941 if (val >= NSEC_PER_SEC)
2943 if (val >= NSEC_PER_SEC / 2)
2946 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2949 struct async_delayed_refs {
2950 struct btrfs_root *root;
2955 struct completion wait;
2956 struct btrfs_work work;
2959 static inline struct async_delayed_refs *
2960 to_async_delayed_refs(struct btrfs_work *work)
2962 return container_of(work, struct async_delayed_refs, work);
2965 static void delayed_ref_async_start(struct btrfs_work *work)
2967 struct async_delayed_refs *async = to_async_delayed_refs(work);
2968 struct btrfs_trans_handle *trans;
2969 struct btrfs_fs_info *fs_info = async->root->fs_info;
2972 /* if the commit is already started, we don't need to wait here */
2973 if (btrfs_transaction_blocked(fs_info))
2976 trans = btrfs_join_transaction(async->root);
2977 if (IS_ERR(trans)) {
2978 async->error = PTR_ERR(trans);
2983 * trans->sync means that when we call end_transaction, we won't
2984 * wait on delayed refs
2988 /* Don't bother flushing if we got into a different transaction */
2989 if (trans->transid > async->transid)
2992 ret = btrfs_run_delayed_refs(trans, async->count);
2996 ret = btrfs_end_transaction(trans);
2997 if (ret && !async->error)
3001 complete(&async->wait);
3006 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
3007 unsigned long count, u64 transid, int wait)
3009 struct async_delayed_refs *async;
3012 async = kmalloc(sizeof(*async), GFP_NOFS);
3016 async->root = fs_info->tree_root;
3017 async->count = count;
3019 async->transid = transid;
3024 init_completion(&async->wait);
3026 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3027 delayed_ref_async_start, NULL, NULL);
3029 btrfs_queue_work(fs_info->extent_workers, &async->work);
3032 wait_for_completion(&async->wait);
3041 * this starts processing the delayed reference count updates and
3042 * extent insertions we have queued up so far. count can be
3043 * 0, which means to process everything in the tree at the start
3044 * of the run (but not newly added entries), or it can be some target
3045 * number you'd like to process.
3047 * Returns 0 on success or if called with an aborted transaction
3048 * Returns <0 on error and aborts the transaction
3050 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3051 unsigned long count)
3053 struct btrfs_fs_info *fs_info = trans->fs_info;
3054 struct rb_node *node;
3055 struct btrfs_delayed_ref_root *delayed_refs;
3056 struct btrfs_delayed_ref_head *head;
3058 int run_all = count == (unsigned long)-1;
3059 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3061 /* We'll clean this up in btrfs_cleanup_transaction */
3065 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3068 delayed_refs = &trans->transaction->delayed_refs;
3070 count = atomic_read(&delayed_refs->num_entries) * 2;
3073 #ifdef SCRAMBLE_DELAYED_REFS
3074 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3076 trans->can_flush_pending_bgs = false;
3077 ret = __btrfs_run_delayed_refs(trans, count);
3079 btrfs_abort_transaction(trans, ret);
3084 if (!list_empty(&trans->new_bgs))
3085 btrfs_create_pending_block_groups(trans);
3087 spin_lock(&delayed_refs->lock);
3088 node = rb_first(&delayed_refs->href_root);
3090 spin_unlock(&delayed_refs->lock);
3093 head = rb_entry(node, struct btrfs_delayed_ref_head,
3095 refcount_inc(&head->refs);
3096 spin_unlock(&delayed_refs->lock);
3098 /* Mutex was contended, block until it's released and retry. */
3099 mutex_lock(&head->mutex);
3100 mutex_unlock(&head->mutex);
3102 btrfs_put_delayed_ref_head(head);
3107 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3111 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3112 struct btrfs_fs_info *fs_info,
3113 u64 bytenr, u64 num_bytes, u64 flags,
3114 int level, int is_data)
3116 struct btrfs_delayed_extent_op *extent_op;
3119 extent_op = btrfs_alloc_delayed_extent_op();
3123 extent_op->flags_to_set = flags;
3124 extent_op->update_flags = true;
3125 extent_op->update_key = false;
3126 extent_op->is_data = is_data ? true : false;
3127 extent_op->level = level;
3129 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3130 num_bytes, extent_op);
3132 btrfs_free_delayed_extent_op(extent_op);
3136 static noinline int check_delayed_ref(struct btrfs_root *root,
3137 struct btrfs_path *path,
3138 u64 objectid, u64 offset, u64 bytenr)
3140 struct btrfs_delayed_ref_head *head;
3141 struct btrfs_delayed_ref_node *ref;
3142 struct btrfs_delayed_data_ref *data_ref;
3143 struct btrfs_delayed_ref_root *delayed_refs;
3144 struct btrfs_transaction *cur_trans;
3145 struct rb_node *node;
3148 spin_lock(&root->fs_info->trans_lock);
3149 cur_trans = root->fs_info->running_transaction;
3151 refcount_inc(&cur_trans->use_count);
3152 spin_unlock(&root->fs_info->trans_lock);
3156 delayed_refs = &cur_trans->delayed_refs;
3157 spin_lock(&delayed_refs->lock);
3158 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3160 spin_unlock(&delayed_refs->lock);
3161 btrfs_put_transaction(cur_trans);
3165 if (!mutex_trylock(&head->mutex)) {
3166 refcount_inc(&head->refs);
3167 spin_unlock(&delayed_refs->lock);
3169 btrfs_release_path(path);
3172 * Mutex was contended, block until it's released and let
3175 mutex_lock(&head->mutex);
3176 mutex_unlock(&head->mutex);
3177 btrfs_put_delayed_ref_head(head);
3178 btrfs_put_transaction(cur_trans);
3181 spin_unlock(&delayed_refs->lock);
3183 spin_lock(&head->lock);
3185 * XXX: We should replace this with a proper search function in the
3188 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3189 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3190 /* If it's a shared ref we know a cross reference exists */
3191 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3196 data_ref = btrfs_delayed_node_to_data_ref(ref);
3199 * If our ref doesn't match the one we're currently looking at
3200 * then we have a cross reference.
3202 if (data_ref->root != root->root_key.objectid ||
3203 data_ref->objectid != objectid ||
3204 data_ref->offset != offset) {
3209 spin_unlock(&head->lock);
3210 mutex_unlock(&head->mutex);
3211 btrfs_put_transaction(cur_trans);
3215 static noinline int check_committed_ref(struct btrfs_root *root,
3216 struct btrfs_path *path,
3217 u64 objectid, u64 offset, u64 bytenr)
3219 struct btrfs_fs_info *fs_info = root->fs_info;
3220 struct btrfs_root *extent_root = fs_info->extent_root;
3221 struct extent_buffer *leaf;
3222 struct btrfs_extent_data_ref *ref;
3223 struct btrfs_extent_inline_ref *iref;
3224 struct btrfs_extent_item *ei;
3225 struct btrfs_key key;
3230 key.objectid = bytenr;
3231 key.offset = (u64)-1;
3232 key.type = BTRFS_EXTENT_ITEM_KEY;
3234 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3237 BUG_ON(ret == 0); /* Corruption */
3240 if (path->slots[0] == 0)
3244 leaf = path->nodes[0];
3245 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3247 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3251 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3252 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3253 if (item_size < sizeof(*ei)) {
3254 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3258 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3260 if (item_size != sizeof(*ei) +
3261 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3264 if (btrfs_extent_generation(leaf, ei) <=
3265 btrfs_root_last_snapshot(&root->root_item))
3268 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3270 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3271 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3274 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3275 if (btrfs_extent_refs(leaf, ei) !=
3276 btrfs_extent_data_ref_count(leaf, ref) ||
3277 btrfs_extent_data_ref_root(leaf, ref) !=
3278 root->root_key.objectid ||
3279 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3280 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3288 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3291 struct btrfs_path *path;
3295 path = btrfs_alloc_path();
3300 ret = check_committed_ref(root, path, objectid,
3302 if (ret && ret != -ENOENT)
3305 ret2 = check_delayed_ref(root, path, objectid,
3307 } while (ret2 == -EAGAIN);
3309 if (ret2 && ret2 != -ENOENT) {
3314 if (ret != -ENOENT || ret2 != -ENOENT)
3317 btrfs_free_path(path);
3318 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3323 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3324 struct btrfs_root *root,
3325 struct extent_buffer *buf,
3326 int full_backref, int inc)
3328 struct btrfs_fs_info *fs_info = root->fs_info;
3334 struct btrfs_key key;
3335 struct btrfs_file_extent_item *fi;
3339 int (*process_func)(struct btrfs_trans_handle *,
3340 struct btrfs_root *,
3341 u64, u64, u64, u64, u64, u64);
3344 if (btrfs_is_testing(fs_info))
3347 ref_root = btrfs_header_owner(buf);
3348 nritems = btrfs_header_nritems(buf);
3349 level = btrfs_header_level(buf);
3351 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3355 process_func = btrfs_inc_extent_ref;
3357 process_func = btrfs_free_extent;
3360 parent = buf->start;
3364 for (i = 0; i < nritems; i++) {
3366 btrfs_item_key_to_cpu(buf, &key, i);
3367 if (key.type != BTRFS_EXTENT_DATA_KEY)
3369 fi = btrfs_item_ptr(buf, i,
3370 struct btrfs_file_extent_item);
3371 if (btrfs_file_extent_type(buf, fi) ==
3372 BTRFS_FILE_EXTENT_INLINE)
3374 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3378 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3379 key.offset -= btrfs_file_extent_offset(buf, fi);
3380 ret = process_func(trans, root, bytenr, num_bytes,
3381 parent, ref_root, key.objectid,
3386 bytenr = btrfs_node_blockptr(buf, i);
3387 num_bytes = fs_info->nodesize;
3388 ret = process_func(trans, root, bytenr, num_bytes,
3389 parent, ref_root, level - 1, 0);
3399 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3400 struct extent_buffer *buf, int full_backref)
3402 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3405 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3406 struct extent_buffer *buf, int full_backref)
3408 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3411 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3412 struct btrfs_fs_info *fs_info,
3413 struct btrfs_path *path,
3414 struct btrfs_block_group_cache *cache)
3417 struct btrfs_root *extent_root = fs_info->extent_root;
3419 struct extent_buffer *leaf;
3421 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3428 leaf = path->nodes[0];
3429 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3430 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3431 btrfs_mark_buffer_dirty(leaf);
3433 btrfs_release_path(path);
3438 static struct btrfs_block_group_cache *
3439 next_block_group(struct btrfs_fs_info *fs_info,
3440 struct btrfs_block_group_cache *cache)
3442 struct rb_node *node;
3444 spin_lock(&fs_info->block_group_cache_lock);
3446 /* If our block group was removed, we need a full search. */
3447 if (RB_EMPTY_NODE(&cache->cache_node)) {
3448 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3450 spin_unlock(&fs_info->block_group_cache_lock);
3451 btrfs_put_block_group(cache);
3452 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3454 node = rb_next(&cache->cache_node);
3455 btrfs_put_block_group(cache);
3457 cache = rb_entry(node, struct btrfs_block_group_cache,
3459 btrfs_get_block_group(cache);
3462 spin_unlock(&fs_info->block_group_cache_lock);
3466 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3467 struct btrfs_trans_handle *trans,
3468 struct btrfs_path *path)
3470 struct btrfs_fs_info *fs_info = block_group->fs_info;
3471 struct btrfs_root *root = fs_info->tree_root;
3472 struct inode *inode = NULL;
3473 struct extent_changeset *data_reserved = NULL;
3475 int dcs = BTRFS_DC_ERROR;
3481 * If this block group is smaller than 100 megs don't bother caching the
3484 if (block_group->key.offset < (100 * SZ_1M)) {
3485 spin_lock(&block_group->lock);
3486 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3487 spin_unlock(&block_group->lock);
3494 inode = lookup_free_space_inode(fs_info, block_group, path);
3495 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3496 ret = PTR_ERR(inode);
3497 btrfs_release_path(path);
3501 if (IS_ERR(inode)) {
3505 if (block_group->ro)
3508 ret = create_free_space_inode(fs_info, trans, block_group,
3516 * We want to set the generation to 0, that way if anything goes wrong
3517 * from here on out we know not to trust this cache when we load up next
3520 BTRFS_I(inode)->generation = 0;
3521 ret = btrfs_update_inode(trans, root, inode);
3524 * So theoretically we could recover from this, simply set the
3525 * super cache generation to 0 so we know to invalidate the
3526 * cache, but then we'd have to keep track of the block groups
3527 * that fail this way so we know we _have_ to reset this cache
3528 * before the next commit or risk reading stale cache. So to
3529 * limit our exposure to horrible edge cases lets just abort the
3530 * transaction, this only happens in really bad situations
3533 btrfs_abort_transaction(trans, ret);
3538 /* We've already setup this transaction, go ahead and exit */
3539 if (block_group->cache_generation == trans->transid &&
3540 i_size_read(inode)) {
3541 dcs = BTRFS_DC_SETUP;
3545 if (i_size_read(inode) > 0) {
3546 ret = btrfs_check_trunc_cache_free_space(fs_info,
3547 &fs_info->global_block_rsv);
3551 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3556 spin_lock(&block_group->lock);
3557 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3558 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3560 * don't bother trying to write stuff out _if_
3561 * a) we're not cached,
3562 * b) we're with nospace_cache mount option,
3563 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3565 dcs = BTRFS_DC_WRITTEN;
3566 spin_unlock(&block_group->lock);
3569 spin_unlock(&block_group->lock);
3572 * We hit an ENOSPC when setting up the cache in this transaction, just
3573 * skip doing the setup, we've already cleared the cache so we're safe.
3575 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3581 * Try to preallocate enough space based on how big the block group is.
3582 * Keep in mind this has to include any pinned space which could end up
3583 * taking up quite a bit since it's not folded into the other space
3586 num_pages = div_u64(block_group->key.offset, SZ_256M);
3591 num_pages *= PAGE_SIZE;
3593 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3597 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3598 num_pages, num_pages,
3601 * Our cache requires contiguous chunks so that we don't modify a bunch
3602 * of metadata or split extents when writing the cache out, which means
3603 * we can enospc if we are heavily fragmented in addition to just normal
3604 * out of space conditions. So if we hit this just skip setting up any
3605 * other block groups for this transaction, maybe we'll unpin enough
3606 * space the next time around.
3609 dcs = BTRFS_DC_SETUP;
3610 else if (ret == -ENOSPC)
3611 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3616 btrfs_release_path(path);
3618 spin_lock(&block_group->lock);
3619 if (!ret && dcs == BTRFS_DC_SETUP)
3620 block_group->cache_generation = trans->transid;
3621 block_group->disk_cache_state = dcs;
3622 spin_unlock(&block_group->lock);
3624 extent_changeset_free(data_reserved);
3628 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3629 struct btrfs_fs_info *fs_info)
3631 struct btrfs_block_group_cache *cache, *tmp;
3632 struct btrfs_transaction *cur_trans = trans->transaction;
3633 struct btrfs_path *path;
3635 if (list_empty(&cur_trans->dirty_bgs) ||
3636 !btrfs_test_opt(fs_info, SPACE_CACHE))
3639 path = btrfs_alloc_path();
3643 /* Could add new block groups, use _safe just in case */
3644 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3646 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3647 cache_save_setup(cache, trans, path);
3650 btrfs_free_path(path);
3655 * transaction commit does final block group cache writeback during a
3656 * critical section where nothing is allowed to change the FS. This is
3657 * required in order for the cache to actually match the block group,
3658 * but can introduce a lot of latency into the commit.
3660 * So, btrfs_start_dirty_block_groups is here to kick off block group
3661 * cache IO. There's a chance we'll have to redo some of it if the
3662 * block group changes again during the commit, but it greatly reduces
3663 * the commit latency by getting rid of the easy block groups while
3664 * we're still allowing others to join the commit.
3666 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3668 struct btrfs_fs_info *fs_info = trans->fs_info;
3669 struct btrfs_block_group_cache *cache;
3670 struct btrfs_transaction *cur_trans = trans->transaction;
3673 struct btrfs_path *path = NULL;
3675 struct list_head *io = &cur_trans->io_bgs;
3676 int num_started = 0;
3679 spin_lock(&cur_trans->dirty_bgs_lock);
3680 if (list_empty(&cur_trans->dirty_bgs)) {
3681 spin_unlock(&cur_trans->dirty_bgs_lock);
3684 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3685 spin_unlock(&cur_trans->dirty_bgs_lock);
3689 * make sure all the block groups on our dirty list actually
3692 btrfs_create_pending_block_groups(trans);
3695 path = btrfs_alloc_path();
3701 * cache_write_mutex is here only to save us from balance or automatic
3702 * removal of empty block groups deleting this block group while we are
3703 * writing out the cache
3705 mutex_lock(&trans->transaction->cache_write_mutex);
3706 while (!list_empty(&dirty)) {
3707 cache = list_first_entry(&dirty,
3708 struct btrfs_block_group_cache,
3711 * this can happen if something re-dirties a block
3712 * group that is already under IO. Just wait for it to
3713 * finish and then do it all again
3715 if (!list_empty(&cache->io_list)) {
3716 list_del_init(&cache->io_list);
3717 btrfs_wait_cache_io(trans, cache, path);
3718 btrfs_put_block_group(cache);
3723 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3724 * if it should update the cache_state. Don't delete
3725 * until after we wait.
3727 * Since we're not running in the commit critical section
3728 * we need the dirty_bgs_lock to protect from update_block_group
3730 spin_lock(&cur_trans->dirty_bgs_lock);
3731 list_del_init(&cache->dirty_list);
3732 spin_unlock(&cur_trans->dirty_bgs_lock);
3736 cache_save_setup(cache, trans, path);
3738 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3739 cache->io_ctl.inode = NULL;
3740 ret = btrfs_write_out_cache(fs_info, trans,
3742 if (ret == 0 && cache->io_ctl.inode) {
3747 * The cache_write_mutex is protecting the
3748 * io_list, also refer to the definition of
3749 * btrfs_transaction::io_bgs for more details
3751 list_add_tail(&cache->io_list, io);
3754 * if we failed to write the cache, the
3755 * generation will be bad and life goes on
3761 ret = write_one_cache_group(trans, fs_info,
3764 * Our block group might still be attached to the list
3765 * of new block groups in the transaction handle of some
3766 * other task (struct btrfs_trans_handle->new_bgs). This
3767 * means its block group item isn't yet in the extent
3768 * tree. If this happens ignore the error, as we will
3769 * try again later in the critical section of the
3770 * transaction commit.
3772 if (ret == -ENOENT) {
3774 spin_lock(&cur_trans->dirty_bgs_lock);
3775 if (list_empty(&cache->dirty_list)) {
3776 list_add_tail(&cache->dirty_list,
3777 &cur_trans->dirty_bgs);
3778 btrfs_get_block_group(cache);
3780 spin_unlock(&cur_trans->dirty_bgs_lock);
3782 btrfs_abort_transaction(trans, ret);
3786 /* if its not on the io list, we need to put the block group */
3788 btrfs_put_block_group(cache);
3794 * Avoid blocking other tasks for too long. It might even save
3795 * us from writing caches for block groups that are going to be
3798 mutex_unlock(&trans->transaction->cache_write_mutex);
3799 mutex_lock(&trans->transaction->cache_write_mutex);
3801 mutex_unlock(&trans->transaction->cache_write_mutex);
3804 * go through delayed refs for all the stuff we've just kicked off
3805 * and then loop back (just once)
3807 ret = btrfs_run_delayed_refs(trans, 0);
3808 if (!ret && loops == 0) {
3810 spin_lock(&cur_trans->dirty_bgs_lock);
3811 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3813 * dirty_bgs_lock protects us from concurrent block group
3814 * deletes too (not just cache_write_mutex).
3816 if (!list_empty(&dirty)) {
3817 spin_unlock(&cur_trans->dirty_bgs_lock);
3820 spin_unlock(&cur_trans->dirty_bgs_lock);
3821 } else if (ret < 0) {
3822 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3825 btrfs_free_path(path);
3829 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3830 struct btrfs_fs_info *fs_info)
3832 struct btrfs_block_group_cache *cache;
3833 struct btrfs_transaction *cur_trans = trans->transaction;
3836 struct btrfs_path *path;
3837 struct list_head *io = &cur_trans->io_bgs;
3838 int num_started = 0;
3840 path = btrfs_alloc_path();
3845 * Even though we are in the critical section of the transaction commit,
3846 * we can still have concurrent tasks adding elements to this
3847 * transaction's list of dirty block groups. These tasks correspond to
3848 * endio free space workers started when writeback finishes for a
3849 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3850 * allocate new block groups as a result of COWing nodes of the root
3851 * tree when updating the free space inode. The writeback for the space
3852 * caches is triggered by an earlier call to
3853 * btrfs_start_dirty_block_groups() and iterations of the following
3855 * Also we want to do the cache_save_setup first and then run the
3856 * delayed refs to make sure we have the best chance at doing this all
3859 spin_lock(&cur_trans->dirty_bgs_lock);
3860 while (!list_empty(&cur_trans->dirty_bgs)) {
3861 cache = list_first_entry(&cur_trans->dirty_bgs,
3862 struct btrfs_block_group_cache,
3866 * this can happen if cache_save_setup re-dirties a block
3867 * group that is already under IO. Just wait for it to
3868 * finish and then do it all again
3870 if (!list_empty(&cache->io_list)) {
3871 spin_unlock(&cur_trans->dirty_bgs_lock);
3872 list_del_init(&cache->io_list);
3873 btrfs_wait_cache_io(trans, cache, path);
3874 btrfs_put_block_group(cache);
3875 spin_lock(&cur_trans->dirty_bgs_lock);
3879 * don't remove from the dirty list until after we've waited
3882 list_del_init(&cache->dirty_list);
3883 spin_unlock(&cur_trans->dirty_bgs_lock);
3886 cache_save_setup(cache, trans, path);
3889 ret = btrfs_run_delayed_refs(trans,
3890 (unsigned long) -1);
3892 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3893 cache->io_ctl.inode = NULL;
3894 ret = btrfs_write_out_cache(fs_info, trans,
3896 if (ret == 0 && cache->io_ctl.inode) {
3899 list_add_tail(&cache->io_list, io);
3902 * if we failed to write the cache, the
3903 * generation will be bad and life goes on
3909 ret = write_one_cache_group(trans, fs_info,
3912 * One of the free space endio workers might have
3913 * created a new block group while updating a free space
3914 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3915 * and hasn't released its transaction handle yet, in
3916 * which case the new block group is still attached to
3917 * its transaction handle and its creation has not
3918 * finished yet (no block group item in the extent tree
3919 * yet, etc). If this is the case, wait for all free
3920 * space endio workers to finish and retry. This is a
3921 * a very rare case so no need for a more efficient and
3924 if (ret == -ENOENT) {
3925 wait_event(cur_trans->writer_wait,
3926 atomic_read(&cur_trans->num_writers) == 1);
3927 ret = write_one_cache_group(trans, fs_info,
3931 btrfs_abort_transaction(trans, ret);
3934 /* if its not on the io list, we need to put the block group */
3936 btrfs_put_block_group(cache);
3937 spin_lock(&cur_trans->dirty_bgs_lock);
3939 spin_unlock(&cur_trans->dirty_bgs_lock);
3942 * Refer to the definition of io_bgs member for details why it's safe
3943 * to use it without any locking
3945 while (!list_empty(io)) {
3946 cache = list_first_entry(io, struct btrfs_block_group_cache,
3948 list_del_init(&cache->io_list);
3949 btrfs_wait_cache_io(trans, cache, path);
3950 btrfs_put_block_group(cache);
3953 btrfs_free_path(path);
3957 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3959 struct btrfs_block_group_cache *block_group;
3962 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3963 if (!block_group || block_group->ro)
3966 btrfs_put_block_group(block_group);
3970 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3972 struct btrfs_block_group_cache *bg;
3975 bg = btrfs_lookup_block_group(fs_info, bytenr);
3979 spin_lock(&bg->lock);
3983 atomic_inc(&bg->nocow_writers);
3984 spin_unlock(&bg->lock);
3986 /* no put on block group, done by btrfs_dec_nocow_writers */
3988 btrfs_put_block_group(bg);
3994 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3996 struct btrfs_block_group_cache *bg;
3998 bg = btrfs_lookup_block_group(fs_info, bytenr);
4000 if (atomic_dec_and_test(&bg->nocow_writers))
4001 wake_up_var(&bg->nocow_writers);
4003 * Once for our lookup and once for the lookup done by a previous call
4004 * to btrfs_inc_nocow_writers()
4006 btrfs_put_block_group(bg);
4007 btrfs_put_block_group(bg);
4010 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4012 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
4015 static const char *alloc_name(u64 flags)
4018 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4020 case BTRFS_BLOCK_GROUP_METADATA:
4022 case BTRFS_BLOCK_GROUP_DATA:
4024 case BTRFS_BLOCK_GROUP_SYSTEM:
4028 return "invalid-combination";
4032 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4033 struct btrfs_space_info **new)
4036 struct btrfs_space_info *space_info;
4040 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4044 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4051 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4052 INIT_LIST_HEAD(&space_info->block_groups[i]);
4053 init_rwsem(&space_info->groups_sem);
4054 spin_lock_init(&space_info->lock);
4055 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4056 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4057 init_waitqueue_head(&space_info->wait);
4058 INIT_LIST_HEAD(&space_info->ro_bgs);
4059 INIT_LIST_HEAD(&space_info->tickets);
4060 INIT_LIST_HEAD(&space_info->priority_tickets);
4062 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4063 info->space_info_kobj, "%s",
4064 alloc_name(space_info->flags));
4066 percpu_counter_destroy(&space_info->total_bytes_pinned);
4072 list_add_rcu(&space_info->list, &info->space_info);
4073 if (flags & BTRFS_BLOCK_GROUP_DATA)
4074 info->data_sinfo = space_info;
4079 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4080 u64 total_bytes, u64 bytes_used,
4082 struct btrfs_space_info **space_info)
4084 struct btrfs_space_info *found;
4087 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4088 BTRFS_BLOCK_GROUP_RAID10))
4093 found = __find_space_info(info, flags);
4095 spin_lock(&found->lock);
4096 found->total_bytes += total_bytes;
4097 found->disk_total += total_bytes * factor;
4098 found->bytes_used += bytes_used;
4099 found->disk_used += bytes_used * factor;
4100 found->bytes_readonly += bytes_readonly;
4101 if (total_bytes > 0)
4103 space_info_add_new_bytes(info, found, total_bytes -
4104 bytes_used - bytes_readonly);
4105 spin_unlock(&found->lock);
4106 *space_info = found;
4109 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4111 u64 extra_flags = chunk_to_extended(flags) &
4112 BTRFS_EXTENDED_PROFILE_MASK;
4114 write_seqlock(&fs_info->profiles_lock);
4115 if (flags & BTRFS_BLOCK_GROUP_DATA)
4116 fs_info->avail_data_alloc_bits |= extra_flags;
4117 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4118 fs_info->avail_metadata_alloc_bits |= extra_flags;
4119 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4120 fs_info->avail_system_alloc_bits |= extra_flags;
4121 write_sequnlock(&fs_info->profiles_lock);
4125 * returns target flags in extended format or 0 if restripe for this
4126 * chunk_type is not in progress
4128 * should be called with either volume_mutex or balance_lock held
4130 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4132 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4138 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4139 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4140 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4141 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4142 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4143 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4144 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4145 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4146 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4153 * @flags: available profiles in extended format (see ctree.h)
4155 * Returns reduced profile in chunk format. If profile changing is in
4156 * progress (either running or paused) picks the target profile (if it's
4157 * already available), otherwise falls back to plain reducing.
4159 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4161 u64 num_devices = fs_info->fs_devices->rw_devices;
4167 * see if restripe for this chunk_type is in progress, if so
4168 * try to reduce to the target profile
4170 spin_lock(&fs_info->balance_lock);
4171 target = get_restripe_target(fs_info, flags);
4173 /* pick target profile only if it's already available */
4174 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4175 spin_unlock(&fs_info->balance_lock);
4176 return extended_to_chunk(target);
4179 spin_unlock(&fs_info->balance_lock);
4181 /* First, mask out the RAID levels which aren't possible */
4182 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4183 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4184 allowed |= btrfs_raid_group[raid_type];
4188 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4189 allowed = BTRFS_BLOCK_GROUP_RAID6;
4190 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4191 allowed = BTRFS_BLOCK_GROUP_RAID5;
4192 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4193 allowed = BTRFS_BLOCK_GROUP_RAID10;
4194 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4195 allowed = BTRFS_BLOCK_GROUP_RAID1;
4196 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4197 allowed = BTRFS_BLOCK_GROUP_RAID0;
4199 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4201 return extended_to_chunk(flags | allowed);
4204 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4211 seq = read_seqbegin(&fs_info->profiles_lock);
4213 if (flags & BTRFS_BLOCK_GROUP_DATA)
4214 flags |= fs_info->avail_data_alloc_bits;
4215 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4216 flags |= fs_info->avail_system_alloc_bits;
4217 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4218 flags |= fs_info->avail_metadata_alloc_bits;
4219 } while (read_seqretry(&fs_info->profiles_lock, seq));
4221 return btrfs_reduce_alloc_profile(fs_info, flags);
4224 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4226 struct btrfs_fs_info *fs_info = root->fs_info;
4231 flags = BTRFS_BLOCK_GROUP_DATA;
4232 else if (root == fs_info->chunk_root)
4233 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4235 flags = BTRFS_BLOCK_GROUP_METADATA;
4237 ret = get_alloc_profile(fs_info, flags);
4241 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4243 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4246 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4248 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4251 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4253 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4256 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4257 bool may_use_included)
4260 return s_info->bytes_used + s_info->bytes_reserved +
4261 s_info->bytes_pinned + s_info->bytes_readonly +
4262 (may_use_included ? s_info->bytes_may_use : 0);
4265 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4267 struct btrfs_root *root = inode->root;
4268 struct btrfs_fs_info *fs_info = root->fs_info;
4269 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4272 int need_commit = 2;
4273 int have_pinned_space;
4275 /* make sure bytes are sectorsize aligned */
4276 bytes = ALIGN(bytes, fs_info->sectorsize);
4278 if (btrfs_is_free_space_inode(inode)) {
4280 ASSERT(current->journal_info);
4284 /* make sure we have enough space to handle the data first */
4285 spin_lock(&data_sinfo->lock);
4286 used = btrfs_space_info_used(data_sinfo, true);
4288 if (used + bytes > data_sinfo->total_bytes) {
4289 struct btrfs_trans_handle *trans;
4292 * if we don't have enough free bytes in this space then we need
4293 * to alloc a new chunk.
4295 if (!data_sinfo->full) {
4298 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4299 spin_unlock(&data_sinfo->lock);
4301 alloc_target = btrfs_data_alloc_profile(fs_info);
4303 * It is ugly that we don't call nolock join
4304 * transaction for the free space inode case here.
4305 * But it is safe because we only do the data space
4306 * reservation for the free space cache in the
4307 * transaction context, the common join transaction
4308 * just increase the counter of the current transaction
4309 * handler, doesn't try to acquire the trans_lock of
4312 trans = btrfs_join_transaction(root);
4314 return PTR_ERR(trans);
4316 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4317 CHUNK_ALLOC_NO_FORCE);
4318 btrfs_end_transaction(trans);
4323 have_pinned_space = 1;
4332 * If we don't have enough pinned space to deal with this
4333 * allocation, and no removed chunk in current transaction,
4334 * don't bother committing the transaction.
4336 have_pinned_space = percpu_counter_compare(
4337 &data_sinfo->total_bytes_pinned,
4338 used + bytes - data_sinfo->total_bytes);
4339 spin_unlock(&data_sinfo->lock);
4341 /* commit the current transaction and try again */
4346 if (need_commit > 0) {
4347 btrfs_start_delalloc_roots(fs_info, 0, -1);
4348 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4352 trans = btrfs_join_transaction(root);
4354 return PTR_ERR(trans);
4355 if (have_pinned_space >= 0 ||
4356 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4357 &trans->transaction->flags) ||
4359 ret = btrfs_commit_transaction(trans);
4363 * The cleaner kthread might still be doing iput
4364 * operations. Wait for it to finish so that
4365 * more space is released.
4367 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4368 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4371 btrfs_end_transaction(trans);
4375 trace_btrfs_space_reservation(fs_info,
4376 "space_info:enospc",
4377 data_sinfo->flags, bytes, 1);
4380 data_sinfo->bytes_may_use += bytes;
4381 trace_btrfs_space_reservation(fs_info, "space_info",
4382 data_sinfo->flags, bytes, 1);
4383 spin_unlock(&data_sinfo->lock);
4388 int btrfs_check_data_free_space(struct inode *inode,
4389 struct extent_changeset **reserved, u64 start, u64 len)
4391 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4394 /* align the range */
4395 len = round_up(start + len, fs_info->sectorsize) -
4396 round_down(start, fs_info->sectorsize);
4397 start = round_down(start, fs_info->sectorsize);
4399 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4403 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4404 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4406 btrfs_free_reserved_data_space_noquota(inode, start, len);
4413 * Called if we need to clear a data reservation for this inode
4414 * Normally in a error case.
4416 * This one will *NOT* use accurate qgroup reserved space API, just for case
4417 * which we can't sleep and is sure it won't affect qgroup reserved space.
4418 * Like clear_bit_hook().
4420 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4423 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4424 struct btrfs_space_info *data_sinfo;
4426 /* Make sure the range is aligned to sectorsize */
4427 len = round_up(start + len, fs_info->sectorsize) -
4428 round_down(start, fs_info->sectorsize);
4429 start = round_down(start, fs_info->sectorsize);
4431 data_sinfo = fs_info->data_sinfo;
4432 spin_lock(&data_sinfo->lock);
4433 if (WARN_ON(data_sinfo->bytes_may_use < len))
4434 data_sinfo->bytes_may_use = 0;
4436 data_sinfo->bytes_may_use -= len;
4437 trace_btrfs_space_reservation(fs_info, "space_info",
4438 data_sinfo->flags, len, 0);
4439 spin_unlock(&data_sinfo->lock);
4443 * Called if we need to clear a data reservation for this inode
4444 * Normally in a error case.
4446 * This one will handle the per-inode data rsv map for accurate reserved
4449 void btrfs_free_reserved_data_space(struct inode *inode,
4450 struct extent_changeset *reserved, u64 start, u64 len)
4452 struct btrfs_root *root = BTRFS_I(inode)->root;
4454 /* Make sure the range is aligned to sectorsize */
4455 len = round_up(start + len, root->fs_info->sectorsize) -
4456 round_down(start, root->fs_info->sectorsize);
4457 start = round_down(start, root->fs_info->sectorsize);
4459 btrfs_free_reserved_data_space_noquota(inode, start, len);
4460 btrfs_qgroup_free_data(inode, reserved, start, len);
4463 static void force_metadata_allocation(struct btrfs_fs_info *info)
4465 struct list_head *head = &info->space_info;
4466 struct btrfs_space_info *found;
4469 list_for_each_entry_rcu(found, head, list) {
4470 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4471 found->force_alloc = CHUNK_ALLOC_FORCE;
4476 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4478 return (global->size << 1);
4481 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4482 struct btrfs_space_info *sinfo, int force)
4484 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4485 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4488 if (force == CHUNK_ALLOC_FORCE)
4492 * We need to take into account the global rsv because for all intents
4493 * and purposes it's used space. Don't worry about locking the
4494 * global_rsv, it doesn't change except when the transaction commits.
4496 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4497 bytes_used += calc_global_rsv_need_space(global_rsv);
4500 * in limited mode, we want to have some free space up to
4501 * about 1% of the FS size.
4503 if (force == CHUNK_ALLOC_LIMITED) {
4504 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4505 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4507 if (sinfo->total_bytes - bytes_used < thresh)
4511 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4516 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4520 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4521 BTRFS_BLOCK_GROUP_RAID0 |
4522 BTRFS_BLOCK_GROUP_RAID5 |
4523 BTRFS_BLOCK_GROUP_RAID6))
4524 num_dev = fs_info->fs_devices->rw_devices;
4525 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4528 num_dev = 1; /* DUP or single */
4534 * If @is_allocation is true, reserve space in the system space info necessary
4535 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4538 void check_system_chunk(struct btrfs_trans_handle *trans,
4539 struct btrfs_fs_info *fs_info, u64 type)
4541 struct btrfs_space_info *info;
4548 * Needed because we can end up allocating a system chunk and for an
4549 * atomic and race free space reservation in the chunk block reserve.
4551 lockdep_assert_held(&fs_info->chunk_mutex);
4553 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4554 spin_lock(&info->lock);
4555 left = info->total_bytes - btrfs_space_info_used(info, true);
4556 spin_unlock(&info->lock);
4558 num_devs = get_profile_num_devs(fs_info, type);
4560 /* num_devs device items to update and 1 chunk item to add or remove */
4561 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4562 btrfs_calc_trans_metadata_size(fs_info, 1);
4564 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4565 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4566 left, thresh, type);
4567 dump_space_info(fs_info, info, 0, 0);
4570 if (left < thresh) {
4571 u64 flags = btrfs_system_alloc_profile(fs_info);
4574 * Ignore failure to create system chunk. We might end up not
4575 * needing it, as we might not need to COW all nodes/leafs from
4576 * the paths we visit in the chunk tree (they were already COWed
4577 * or created in the current transaction for example).
4579 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4583 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4584 &fs_info->chunk_block_rsv,
4585 thresh, BTRFS_RESERVE_NO_FLUSH);
4587 trans->chunk_bytes_reserved += thresh;
4592 * If force is CHUNK_ALLOC_FORCE:
4593 * - return 1 if it successfully allocates a chunk,
4594 * - return errors including -ENOSPC otherwise.
4595 * If force is NOT CHUNK_ALLOC_FORCE:
4596 * - return 0 if it doesn't need to allocate a new chunk,
4597 * - return 1 if it successfully allocates a chunk,
4598 * - return errors including -ENOSPC otherwise.
4600 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4601 struct btrfs_fs_info *fs_info, u64 flags, int force)
4603 struct btrfs_space_info *space_info;
4604 int wait_for_alloc = 0;
4607 /* Don't re-enter if we're already allocating a chunk */
4608 if (trans->allocating_chunk)
4611 space_info = __find_space_info(fs_info, flags);
4615 spin_lock(&space_info->lock);
4616 if (force < space_info->force_alloc)
4617 force = space_info->force_alloc;
4618 if (space_info->full) {
4619 if (should_alloc_chunk(fs_info, space_info, force))
4623 spin_unlock(&space_info->lock);
4627 if (!should_alloc_chunk(fs_info, space_info, force)) {
4628 spin_unlock(&space_info->lock);
4630 } else if (space_info->chunk_alloc) {
4633 space_info->chunk_alloc = 1;
4636 spin_unlock(&space_info->lock);
4638 mutex_lock(&fs_info->chunk_mutex);
4641 * The chunk_mutex is held throughout the entirety of a chunk
4642 * allocation, so once we've acquired the chunk_mutex we know that the
4643 * other guy is done and we need to recheck and see if we should
4646 if (wait_for_alloc) {
4647 mutex_unlock(&fs_info->chunk_mutex);
4653 trans->allocating_chunk = true;
4656 * If we have mixed data/metadata chunks we want to make sure we keep
4657 * allocating mixed chunks instead of individual chunks.
4659 if (btrfs_mixed_space_info(space_info))
4660 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4663 * if we're doing a data chunk, go ahead and make sure that
4664 * we keep a reasonable number of metadata chunks allocated in the
4667 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4668 fs_info->data_chunk_allocations++;
4669 if (!(fs_info->data_chunk_allocations %
4670 fs_info->metadata_ratio))
4671 force_metadata_allocation(fs_info);
4675 * Check if we have enough space in SYSTEM chunk because we may need
4676 * to update devices.
4678 check_system_chunk(trans, fs_info, flags);
4680 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4681 trans->allocating_chunk = false;
4683 spin_lock(&space_info->lock);
4684 if (ret < 0 && ret != -ENOSPC)
4687 space_info->full = 1;
4691 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4693 space_info->chunk_alloc = 0;
4694 spin_unlock(&space_info->lock);
4695 mutex_unlock(&fs_info->chunk_mutex);
4697 * When we allocate a new chunk we reserve space in the chunk block
4698 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4699 * add new nodes/leafs to it if we end up needing to do it when
4700 * inserting the chunk item and updating device items as part of the
4701 * second phase of chunk allocation, performed by
4702 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4703 * large number of new block groups to create in our transaction
4704 * handle's new_bgs list to avoid exhausting the chunk block reserve
4705 * in extreme cases - like having a single transaction create many new
4706 * block groups when starting to write out the free space caches of all
4707 * the block groups that were made dirty during the lifetime of the
4710 if (trans->can_flush_pending_bgs &&
4711 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4712 btrfs_create_pending_block_groups(trans);
4713 btrfs_trans_release_chunk_metadata(trans);
4718 static int can_overcommit(struct btrfs_fs_info *fs_info,
4719 struct btrfs_space_info *space_info, u64 bytes,
4720 enum btrfs_reserve_flush_enum flush,
4723 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4729 /* Don't overcommit when in mixed mode. */
4730 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4734 profile = btrfs_system_alloc_profile(fs_info);
4736 profile = btrfs_metadata_alloc_profile(fs_info);
4738 used = btrfs_space_info_used(space_info, false);
4741 * We only want to allow over committing if we have lots of actual space
4742 * free, but if we don't have enough space to handle the global reserve
4743 * space then we could end up having a real enospc problem when trying
4744 * to allocate a chunk or some other such important allocation.
4746 spin_lock(&global_rsv->lock);
4747 space_size = calc_global_rsv_need_space(global_rsv);
4748 spin_unlock(&global_rsv->lock);
4749 if (used + space_size >= space_info->total_bytes)
4752 used += space_info->bytes_may_use;
4754 avail = atomic64_read(&fs_info->free_chunk_space);
4757 * If we have dup, raid1 or raid10 then only half of the free
4758 * space is actually useable. For raid56, the space info used
4759 * doesn't include the parity drive, so we don't have to
4762 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4763 BTRFS_BLOCK_GROUP_RAID1 |
4764 BTRFS_BLOCK_GROUP_RAID10))
4768 * If we aren't flushing all things, let us overcommit up to
4769 * 1/2th of the space. If we can flush, don't let us overcommit
4770 * too much, let it overcommit up to 1/8 of the space.
4772 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4777 if (used + bytes < space_info->total_bytes + avail)
4782 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4783 unsigned long nr_pages, int nr_items)
4785 struct super_block *sb = fs_info->sb;
4787 if (down_read_trylock(&sb->s_umount)) {
4788 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4789 up_read(&sb->s_umount);
4792 * We needn't worry the filesystem going from r/w to r/o though
4793 * we don't acquire ->s_umount mutex, because the filesystem
4794 * should guarantee the delalloc inodes list be empty after
4795 * the filesystem is readonly(all dirty pages are written to
4798 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4799 if (!current->journal_info)
4800 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4804 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4810 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4811 nr = div64_u64(to_reclaim, bytes);
4817 #define EXTENT_SIZE_PER_ITEM SZ_256K
4820 * shrink metadata reservation for delalloc
4822 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4823 u64 orig, bool wait_ordered)
4825 struct btrfs_space_info *space_info;
4826 struct btrfs_trans_handle *trans;
4831 unsigned long nr_pages;
4834 /* Calc the number of the pages we need flush for space reservation */
4835 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4836 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4838 trans = (struct btrfs_trans_handle *)current->journal_info;
4839 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4841 delalloc_bytes = percpu_counter_sum_positive(
4842 &fs_info->delalloc_bytes);
4843 if (delalloc_bytes == 0) {
4847 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4852 while (delalloc_bytes && loops < 3) {
4853 max_reclaim = min(delalloc_bytes, to_reclaim);
4854 nr_pages = max_reclaim >> PAGE_SHIFT;
4855 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4857 * We need to wait for the async pages to actually start before
4860 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4864 if (max_reclaim <= nr_pages)
4867 max_reclaim -= nr_pages;
4869 wait_event(fs_info->async_submit_wait,
4870 atomic_read(&fs_info->async_delalloc_pages) <=
4873 spin_lock(&space_info->lock);
4874 if (list_empty(&space_info->tickets) &&
4875 list_empty(&space_info->priority_tickets)) {
4876 spin_unlock(&space_info->lock);
4879 spin_unlock(&space_info->lock);
4882 if (wait_ordered && !trans) {
4883 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4885 time_left = schedule_timeout_killable(1);
4889 delalloc_bytes = percpu_counter_sum_positive(
4890 &fs_info->delalloc_bytes);
4894 struct reserve_ticket {
4897 struct list_head list;
4898 wait_queue_head_t wait;
4902 * maybe_commit_transaction - possibly commit the transaction if its ok to
4903 * @root - the root we're allocating for
4904 * @bytes - the number of bytes we want to reserve
4905 * @force - force the commit
4907 * This will check to make sure that committing the transaction will actually
4908 * get us somewhere and then commit the transaction if it does. Otherwise it
4909 * will return -ENOSPC.
4911 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4912 struct btrfs_space_info *space_info)
4914 struct reserve_ticket *ticket = NULL;
4915 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4916 struct btrfs_trans_handle *trans;
4919 trans = (struct btrfs_trans_handle *)current->journal_info;
4923 spin_lock(&space_info->lock);
4924 if (!list_empty(&space_info->priority_tickets))
4925 ticket = list_first_entry(&space_info->priority_tickets,
4926 struct reserve_ticket, list);
4927 else if (!list_empty(&space_info->tickets))
4928 ticket = list_first_entry(&space_info->tickets,
4929 struct reserve_ticket, list);
4930 bytes = (ticket) ? ticket->bytes : 0;
4931 spin_unlock(&space_info->lock);
4936 /* See if there is enough pinned space to make this reservation */
4937 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4942 * See if there is some space in the delayed insertion reservation for
4945 if (space_info != delayed_rsv->space_info)
4948 spin_lock(&delayed_rsv->lock);
4949 if (delayed_rsv->size > bytes)
4952 bytes -= delayed_rsv->size;
4953 spin_unlock(&delayed_rsv->lock);
4955 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4961 trans = btrfs_join_transaction(fs_info->extent_root);
4965 return btrfs_commit_transaction(trans);
4969 * Try to flush some data based on policy set by @state. This is only advisory
4970 * and may fail for various reasons. The caller is supposed to examine the
4971 * state of @space_info to detect the outcome.
4973 static void flush_space(struct btrfs_fs_info *fs_info,
4974 struct btrfs_space_info *space_info, u64 num_bytes,
4977 struct btrfs_root *root = fs_info->extent_root;
4978 struct btrfs_trans_handle *trans;
4983 case FLUSH_DELAYED_ITEMS_NR:
4984 case FLUSH_DELAYED_ITEMS:
4985 if (state == FLUSH_DELAYED_ITEMS_NR)
4986 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4990 trans = btrfs_join_transaction(root);
4991 if (IS_ERR(trans)) {
4992 ret = PTR_ERR(trans);
4995 ret = btrfs_run_delayed_items_nr(trans, nr);
4996 btrfs_end_transaction(trans);
4998 case FLUSH_DELALLOC:
4999 case FLUSH_DELALLOC_WAIT:
5000 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
5001 state == FLUSH_DELALLOC_WAIT);
5004 trans = btrfs_join_transaction(root);
5005 if (IS_ERR(trans)) {
5006 ret = PTR_ERR(trans);
5009 ret = do_chunk_alloc(trans, fs_info,
5010 btrfs_metadata_alloc_profile(fs_info),
5011 CHUNK_ALLOC_NO_FORCE);
5012 btrfs_end_transaction(trans);
5013 if (ret > 0 || ret == -ENOSPC)
5017 ret = may_commit_transaction(fs_info, space_info);
5024 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5030 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5031 struct btrfs_space_info *space_info,
5034 struct reserve_ticket *ticket;
5039 list_for_each_entry(ticket, &space_info->tickets, list)
5040 to_reclaim += ticket->bytes;
5041 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5042 to_reclaim += ticket->bytes;
5046 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5047 if (can_overcommit(fs_info, space_info, to_reclaim,
5048 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5051 used = btrfs_space_info_used(space_info, true);
5053 if (can_overcommit(fs_info, space_info, SZ_1M,
5054 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5055 expected = div_factor_fine(space_info->total_bytes, 95);
5057 expected = div_factor_fine(space_info->total_bytes, 90);
5059 if (used > expected)
5060 to_reclaim = used - expected;
5063 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5064 space_info->bytes_reserved);
5068 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5069 struct btrfs_space_info *space_info,
5070 u64 used, bool system_chunk)
5072 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5074 /* If we're just plain full then async reclaim just slows us down. */
5075 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5078 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5082 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5083 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5086 static void wake_all_tickets(struct list_head *head)
5088 struct reserve_ticket *ticket;
5090 while (!list_empty(head)) {
5091 ticket = list_first_entry(head, struct reserve_ticket, list);
5092 list_del_init(&ticket->list);
5093 ticket->error = -ENOSPC;
5094 wake_up(&ticket->wait);
5099 * This is for normal flushers, we can wait all goddamned day if we want to. We
5100 * will loop and continuously try to flush as long as we are making progress.
5101 * We count progress as clearing off tickets each time we have to loop.
5103 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5105 struct btrfs_fs_info *fs_info;
5106 struct btrfs_space_info *space_info;
5109 int commit_cycles = 0;
5110 u64 last_tickets_id;
5112 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5113 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5115 spin_lock(&space_info->lock);
5116 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5119 space_info->flush = 0;
5120 spin_unlock(&space_info->lock);
5123 last_tickets_id = space_info->tickets_id;
5124 spin_unlock(&space_info->lock);
5126 flush_state = FLUSH_DELAYED_ITEMS_NR;
5128 flush_space(fs_info, space_info, to_reclaim, flush_state);
5129 spin_lock(&space_info->lock);
5130 if (list_empty(&space_info->tickets)) {
5131 space_info->flush = 0;
5132 spin_unlock(&space_info->lock);
5135 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5138 if (last_tickets_id == space_info->tickets_id) {
5141 last_tickets_id = space_info->tickets_id;
5142 flush_state = FLUSH_DELAYED_ITEMS_NR;
5147 if (flush_state > COMMIT_TRANS) {
5149 if (commit_cycles > 2) {
5150 wake_all_tickets(&space_info->tickets);
5151 space_info->flush = 0;
5153 flush_state = FLUSH_DELAYED_ITEMS_NR;
5156 spin_unlock(&space_info->lock);
5157 } while (flush_state <= COMMIT_TRANS);
5160 void btrfs_init_async_reclaim_work(struct work_struct *work)
5162 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5165 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5166 struct btrfs_space_info *space_info,
5167 struct reserve_ticket *ticket)
5170 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5172 spin_lock(&space_info->lock);
5173 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5176 spin_unlock(&space_info->lock);
5179 spin_unlock(&space_info->lock);
5182 flush_space(fs_info, space_info, to_reclaim, flush_state);
5184 spin_lock(&space_info->lock);
5185 if (ticket->bytes == 0) {
5186 spin_unlock(&space_info->lock);
5189 spin_unlock(&space_info->lock);
5192 * Priority flushers can't wait on delalloc without
5195 if (flush_state == FLUSH_DELALLOC ||
5196 flush_state == FLUSH_DELALLOC_WAIT)
5197 flush_state = ALLOC_CHUNK;
5198 } while (flush_state < COMMIT_TRANS);
5201 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5202 struct btrfs_space_info *space_info,
5203 struct reserve_ticket *ticket, u64 orig_bytes)
5209 spin_lock(&space_info->lock);
5210 while (ticket->bytes > 0 && ticket->error == 0) {
5211 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5216 spin_unlock(&space_info->lock);
5220 finish_wait(&ticket->wait, &wait);
5221 spin_lock(&space_info->lock);
5224 ret = ticket->error;
5225 if (!list_empty(&ticket->list))
5226 list_del_init(&ticket->list);
5227 if (ticket->bytes && ticket->bytes < orig_bytes) {
5228 u64 num_bytes = orig_bytes - ticket->bytes;
5229 space_info->bytes_may_use -= num_bytes;
5230 trace_btrfs_space_reservation(fs_info, "space_info",
5231 space_info->flags, num_bytes, 0);
5233 spin_unlock(&space_info->lock);
5239 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5240 * @root - the root we're allocating for
5241 * @space_info - the space info we want to allocate from
5242 * @orig_bytes - the number of bytes we want
5243 * @flush - whether or not we can flush to make our reservation
5245 * This will reserve orig_bytes number of bytes from the space info associated
5246 * with the block_rsv. If there is not enough space it will make an attempt to
5247 * flush out space to make room. It will do this by flushing delalloc if
5248 * possible or committing the transaction. If flush is 0 then no attempts to
5249 * regain reservations will be made and this will fail if there is not enough
5252 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5253 struct btrfs_space_info *space_info,
5255 enum btrfs_reserve_flush_enum flush,
5258 struct reserve_ticket ticket;
5263 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5265 spin_lock(&space_info->lock);
5267 used = btrfs_space_info_used(space_info, true);
5270 * If we have enough space then hooray, make our reservation and carry
5271 * on. If not see if we can overcommit, and if we can, hooray carry on.
5272 * If not things get more complicated.
5274 if (used + orig_bytes <= space_info->total_bytes) {
5275 space_info->bytes_may_use += orig_bytes;
5276 trace_btrfs_space_reservation(fs_info, "space_info",
5277 space_info->flags, orig_bytes, 1);
5279 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5281 space_info->bytes_may_use += orig_bytes;
5282 trace_btrfs_space_reservation(fs_info, "space_info",
5283 space_info->flags, orig_bytes, 1);
5288 * If we couldn't make a reservation then setup our reservation ticket
5289 * and kick the async worker if it's not already running.
5291 * If we are a priority flusher then we just need to add our ticket to
5292 * the list and we will do our own flushing further down.
5294 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5295 ticket.bytes = orig_bytes;
5297 init_waitqueue_head(&ticket.wait);
5298 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5299 list_add_tail(&ticket.list, &space_info->tickets);
5300 if (!space_info->flush) {
5301 space_info->flush = 1;
5302 trace_btrfs_trigger_flush(fs_info,
5306 queue_work(system_unbound_wq,
5307 &fs_info->async_reclaim_work);
5310 list_add_tail(&ticket.list,
5311 &space_info->priority_tickets);
5313 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5316 * We will do the space reservation dance during log replay,
5317 * which means we won't have fs_info->fs_root set, so don't do
5318 * the async reclaim as we will panic.
5320 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5321 need_do_async_reclaim(fs_info, space_info,
5322 used, system_chunk) &&
5323 !work_busy(&fs_info->async_reclaim_work)) {
5324 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5325 orig_bytes, flush, "preempt");
5326 queue_work(system_unbound_wq,
5327 &fs_info->async_reclaim_work);
5330 spin_unlock(&space_info->lock);
5331 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5334 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5335 return wait_reserve_ticket(fs_info, space_info, &ticket,
5339 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5340 spin_lock(&space_info->lock);
5342 if (ticket.bytes < orig_bytes) {
5343 u64 num_bytes = orig_bytes - ticket.bytes;
5344 space_info->bytes_may_use -= num_bytes;
5345 trace_btrfs_space_reservation(fs_info, "space_info",
5350 list_del_init(&ticket.list);
5353 spin_unlock(&space_info->lock);
5354 ASSERT(list_empty(&ticket.list));
5359 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5360 * @root - the root we're allocating for
5361 * @block_rsv - the block_rsv we're allocating for
5362 * @orig_bytes - the number of bytes we want
5363 * @flush - whether or not we can flush to make our reservation
5365 * This will reserve orgi_bytes number of bytes from the space info associated
5366 * with the block_rsv. If there is not enough space it will make an attempt to
5367 * flush out space to make room. It will do this by flushing delalloc if
5368 * possible or committing the transaction. If flush is 0 then no attempts to
5369 * regain reservations will be made and this will fail if there is not enough
5372 static int reserve_metadata_bytes(struct btrfs_root *root,
5373 struct btrfs_block_rsv *block_rsv,
5375 enum btrfs_reserve_flush_enum flush)
5377 struct btrfs_fs_info *fs_info = root->fs_info;
5378 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5380 bool system_chunk = (root == fs_info->chunk_root);
5382 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5383 orig_bytes, flush, system_chunk);
5384 if (ret == -ENOSPC &&
5385 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5386 if (block_rsv != global_rsv &&
5387 !block_rsv_use_bytes(global_rsv, orig_bytes))
5390 if (ret == -ENOSPC) {
5391 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5392 block_rsv->space_info->flags,
5395 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5396 dump_space_info(fs_info, block_rsv->space_info,
5402 static struct btrfs_block_rsv *get_block_rsv(
5403 const struct btrfs_trans_handle *trans,
5404 const struct btrfs_root *root)
5406 struct btrfs_fs_info *fs_info = root->fs_info;
5407 struct btrfs_block_rsv *block_rsv = NULL;
5409 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5410 (root == fs_info->csum_root && trans->adding_csums) ||
5411 (root == fs_info->uuid_root))
5412 block_rsv = trans->block_rsv;
5415 block_rsv = root->block_rsv;
5418 block_rsv = &fs_info->empty_block_rsv;
5423 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5427 spin_lock(&block_rsv->lock);
5428 if (block_rsv->reserved >= num_bytes) {
5429 block_rsv->reserved -= num_bytes;
5430 if (block_rsv->reserved < block_rsv->size)
5431 block_rsv->full = 0;
5434 spin_unlock(&block_rsv->lock);
5438 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5439 u64 num_bytes, int update_size)
5441 spin_lock(&block_rsv->lock);
5442 block_rsv->reserved += num_bytes;
5444 block_rsv->size += num_bytes;
5445 else if (block_rsv->reserved >= block_rsv->size)
5446 block_rsv->full = 1;
5447 spin_unlock(&block_rsv->lock);
5450 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5451 struct btrfs_block_rsv *dest, u64 num_bytes,
5454 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5457 if (global_rsv->space_info != dest->space_info)
5460 spin_lock(&global_rsv->lock);
5461 min_bytes = div_factor(global_rsv->size, min_factor);
5462 if (global_rsv->reserved < min_bytes + num_bytes) {
5463 spin_unlock(&global_rsv->lock);
5466 global_rsv->reserved -= num_bytes;
5467 if (global_rsv->reserved < global_rsv->size)
5468 global_rsv->full = 0;
5469 spin_unlock(&global_rsv->lock);
5471 block_rsv_add_bytes(dest, num_bytes, 1);
5476 * This is for space we already have accounted in space_info->bytes_may_use, so
5477 * basically when we're returning space from block_rsv's.
5479 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5480 struct btrfs_space_info *space_info,
5483 struct reserve_ticket *ticket;
5484 struct list_head *head;
5486 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5487 bool check_overcommit = false;
5489 spin_lock(&space_info->lock);
5490 head = &space_info->priority_tickets;
5493 * If we are over our limit then we need to check and see if we can
5494 * overcommit, and if we can't then we just need to free up our space
5495 * and not satisfy any requests.
5497 used = btrfs_space_info_used(space_info, true);
5498 if (used - num_bytes >= space_info->total_bytes)
5499 check_overcommit = true;
5501 while (!list_empty(head) && num_bytes) {
5502 ticket = list_first_entry(head, struct reserve_ticket,
5505 * We use 0 bytes because this space is already reserved, so
5506 * adding the ticket space would be a double count.
5508 if (check_overcommit &&
5509 !can_overcommit(fs_info, space_info, 0, flush, false))
5511 if (num_bytes >= ticket->bytes) {
5512 list_del_init(&ticket->list);
5513 num_bytes -= ticket->bytes;
5515 space_info->tickets_id++;
5516 wake_up(&ticket->wait);
5518 ticket->bytes -= num_bytes;
5523 if (num_bytes && head == &space_info->priority_tickets) {
5524 head = &space_info->tickets;
5525 flush = BTRFS_RESERVE_FLUSH_ALL;
5528 space_info->bytes_may_use -= num_bytes;
5529 trace_btrfs_space_reservation(fs_info, "space_info",
5530 space_info->flags, num_bytes, 0);
5531 spin_unlock(&space_info->lock);
5535 * This is for newly allocated space that isn't accounted in
5536 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5537 * we use this helper.
5539 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5540 struct btrfs_space_info *space_info,
5543 struct reserve_ticket *ticket;
5544 struct list_head *head = &space_info->priority_tickets;
5547 while (!list_empty(head) && num_bytes) {
5548 ticket = list_first_entry(head, struct reserve_ticket,
5550 if (num_bytes >= ticket->bytes) {
5551 trace_btrfs_space_reservation(fs_info, "space_info",
5554 list_del_init(&ticket->list);
5555 num_bytes -= ticket->bytes;
5556 space_info->bytes_may_use += ticket->bytes;
5558 space_info->tickets_id++;
5559 wake_up(&ticket->wait);
5561 trace_btrfs_space_reservation(fs_info, "space_info",
5564 space_info->bytes_may_use += num_bytes;
5565 ticket->bytes -= num_bytes;
5570 if (num_bytes && head == &space_info->priority_tickets) {
5571 head = &space_info->tickets;
5576 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5577 struct btrfs_block_rsv *block_rsv,
5578 struct btrfs_block_rsv *dest, u64 num_bytes,
5579 u64 *qgroup_to_release_ret)
5581 struct btrfs_space_info *space_info = block_rsv->space_info;
5582 u64 qgroup_to_release = 0;
5585 spin_lock(&block_rsv->lock);
5586 if (num_bytes == (u64)-1) {
5587 num_bytes = block_rsv->size;
5588 qgroup_to_release = block_rsv->qgroup_rsv_size;
5590 block_rsv->size -= num_bytes;
5591 if (block_rsv->reserved >= block_rsv->size) {
5592 num_bytes = block_rsv->reserved - block_rsv->size;
5593 block_rsv->reserved = block_rsv->size;
5594 block_rsv->full = 1;
5598 if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) {
5599 qgroup_to_release = block_rsv->qgroup_rsv_reserved -
5600 block_rsv->qgroup_rsv_size;
5601 block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size;
5603 qgroup_to_release = 0;
5605 spin_unlock(&block_rsv->lock);
5608 if (num_bytes > 0) {
5610 spin_lock(&dest->lock);
5614 bytes_to_add = dest->size - dest->reserved;
5615 bytes_to_add = min(num_bytes, bytes_to_add);
5616 dest->reserved += bytes_to_add;
5617 if (dest->reserved >= dest->size)
5619 num_bytes -= bytes_to_add;
5621 spin_unlock(&dest->lock);
5624 space_info_add_old_bytes(fs_info, space_info,
5627 if (qgroup_to_release_ret)
5628 *qgroup_to_release_ret = qgroup_to_release;
5632 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5633 struct btrfs_block_rsv *dst, u64 num_bytes,
5638 ret = block_rsv_use_bytes(src, num_bytes);
5642 block_rsv_add_bytes(dst, num_bytes, update_size);
5646 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5648 memset(rsv, 0, sizeof(*rsv));
5649 spin_lock_init(&rsv->lock);
5653 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5654 struct btrfs_block_rsv *rsv,
5655 unsigned short type)
5657 btrfs_init_block_rsv(rsv, type);
5658 rsv->space_info = __find_space_info(fs_info,
5659 BTRFS_BLOCK_GROUP_METADATA);
5662 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5663 unsigned short type)
5665 struct btrfs_block_rsv *block_rsv;
5667 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5671 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5675 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5676 struct btrfs_block_rsv *rsv)
5680 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5684 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5689 int btrfs_block_rsv_add(struct btrfs_root *root,
5690 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5691 enum btrfs_reserve_flush_enum flush)
5698 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5700 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5707 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5715 spin_lock(&block_rsv->lock);
5716 num_bytes = div_factor(block_rsv->size, min_factor);
5717 if (block_rsv->reserved >= num_bytes)
5719 spin_unlock(&block_rsv->lock);
5724 int btrfs_block_rsv_refill(struct btrfs_root *root,
5725 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5726 enum btrfs_reserve_flush_enum flush)
5734 spin_lock(&block_rsv->lock);
5735 num_bytes = min_reserved;
5736 if (block_rsv->reserved >= num_bytes)
5739 num_bytes -= block_rsv->reserved;
5740 spin_unlock(&block_rsv->lock);
5745 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5747 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5755 * btrfs_inode_rsv_refill - refill the inode block rsv.
5756 * @inode - the inode we are refilling.
5757 * @flush - the flusing restriction.
5759 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5760 * block_rsv->size as the minimum size. We'll either refill the missing amount
5761 * or return if we already have enough space. This will also handle the resreve
5762 * tracepoint for the reserved amount.
5764 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5765 enum btrfs_reserve_flush_enum flush)
5767 struct btrfs_root *root = inode->root;
5768 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5770 u64 qgroup_num_bytes = 0;
5773 spin_lock(&block_rsv->lock);
5774 if (block_rsv->reserved < block_rsv->size)
5775 num_bytes = block_rsv->size - block_rsv->reserved;
5776 if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size)
5777 qgroup_num_bytes = block_rsv->qgroup_rsv_size -
5778 block_rsv->qgroup_rsv_reserved;
5779 spin_unlock(&block_rsv->lock);
5784 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true);
5787 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5789 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5790 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5791 btrfs_ino(inode), num_bytes, 1);
5793 /* Don't forget to increase qgroup_rsv_reserved */
5794 spin_lock(&block_rsv->lock);
5795 block_rsv->qgroup_rsv_reserved += qgroup_num_bytes;
5796 spin_unlock(&block_rsv->lock);
5798 btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes);
5803 * btrfs_inode_rsv_release - release any excessive reservation.
5804 * @inode - the inode we need to release from.
5805 * @qgroup_free - free or convert qgroup meta.
5806 * Unlike normal operation, qgroup meta reservation needs to know if we are
5807 * freeing qgroup reservation or just converting it into per-trans. Normally
5808 * @qgroup_free is true for error handling, and false for normal release.
5810 * This is the same as btrfs_block_rsv_release, except that it handles the
5811 * tracepoint for the reservation.
5813 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5815 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5816 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5817 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5819 u64 qgroup_to_release = 0;
5822 * Since we statically set the block_rsv->size we just want to say we
5823 * are releasing 0 bytes, and then we'll just get the reservation over
5826 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0,
5827 &qgroup_to_release);
5829 trace_btrfs_space_reservation(fs_info, "delalloc",
5830 btrfs_ino(inode), released, 0);
5832 btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release);
5834 btrfs_qgroup_convert_reserved_meta(inode->root,
5838 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5839 struct btrfs_block_rsv *block_rsv,
5842 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5844 if (global_rsv == block_rsv ||
5845 block_rsv->space_info != global_rsv->space_info)
5847 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL);
5850 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5852 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5853 struct btrfs_space_info *sinfo = block_rsv->space_info;
5857 * The global block rsv is based on the size of the extent tree, the
5858 * checksum tree and the root tree. If the fs is empty we want to set
5859 * it to a minimal amount for safety.
5861 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5862 btrfs_root_used(&fs_info->csum_root->root_item) +
5863 btrfs_root_used(&fs_info->tree_root->root_item);
5864 num_bytes = max_t(u64, num_bytes, SZ_16M);
5866 spin_lock(&sinfo->lock);
5867 spin_lock(&block_rsv->lock);
5869 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5871 if (block_rsv->reserved < block_rsv->size) {
5872 num_bytes = btrfs_space_info_used(sinfo, true);
5873 if (sinfo->total_bytes > num_bytes) {
5874 num_bytes = sinfo->total_bytes - num_bytes;
5875 num_bytes = min(num_bytes,
5876 block_rsv->size - block_rsv->reserved);
5877 block_rsv->reserved += num_bytes;
5878 sinfo->bytes_may_use += num_bytes;
5879 trace_btrfs_space_reservation(fs_info, "space_info",
5880 sinfo->flags, num_bytes,
5883 } else if (block_rsv->reserved > block_rsv->size) {
5884 num_bytes = block_rsv->reserved - block_rsv->size;
5885 sinfo->bytes_may_use -= num_bytes;
5886 trace_btrfs_space_reservation(fs_info, "space_info",
5887 sinfo->flags, num_bytes, 0);
5888 block_rsv->reserved = block_rsv->size;
5891 if (block_rsv->reserved == block_rsv->size)
5892 block_rsv->full = 1;
5894 block_rsv->full = 0;
5896 spin_unlock(&block_rsv->lock);
5897 spin_unlock(&sinfo->lock);
5900 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5902 struct btrfs_space_info *space_info;
5904 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5905 fs_info->chunk_block_rsv.space_info = space_info;
5907 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5908 fs_info->global_block_rsv.space_info = space_info;
5909 fs_info->trans_block_rsv.space_info = space_info;
5910 fs_info->empty_block_rsv.space_info = space_info;
5911 fs_info->delayed_block_rsv.space_info = space_info;
5913 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5914 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5915 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5916 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5917 if (fs_info->quota_root)
5918 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5919 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5921 update_global_block_rsv(fs_info);
5924 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5926 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5928 WARN_ON(fs_info->trans_block_rsv.size > 0);
5929 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5930 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5931 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5932 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5933 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5938 * To be called after all the new block groups attached to the transaction
5939 * handle have been created (btrfs_create_pending_block_groups()).
5941 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5943 struct btrfs_fs_info *fs_info = trans->fs_info;
5945 if (!trans->chunk_bytes_reserved)
5948 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5950 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5951 trans->chunk_bytes_reserved, NULL);
5952 trans->chunk_bytes_reserved = 0;
5955 /* Can only return 0 or -ENOSPC */
5956 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5957 struct btrfs_inode *inode)
5959 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5960 struct btrfs_root *root = inode->root;
5962 * We always use trans->block_rsv here as we will have reserved space
5963 * for our orphan when starting the transaction, using get_block_rsv()
5964 * here will sometimes make us choose the wrong block rsv as we could be
5965 * doing a reloc inode for a non refcounted root.
5967 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5968 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5971 * We need to hold space in order to delete our orphan item once we've
5972 * added it, so this takes the reservation so we can release it later
5973 * when we are truly done with the orphan item.
5975 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5977 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5979 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5982 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5984 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5985 struct btrfs_root *root = inode->root;
5986 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5988 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5990 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5994 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5995 * root: the root of the parent directory
5996 * rsv: block reservation
5997 * items: the number of items that we need do reservation
5998 * qgroup_reserved: used to return the reserved size in qgroup
6000 * This function is used to reserve the space for snapshot/subvolume
6001 * creation and deletion. Those operations are different with the
6002 * common file/directory operations, they change two fs/file trees
6003 * and root tree, the number of items that the qgroup reserves is
6004 * different with the free space reservation. So we can not use
6005 * the space reservation mechanism in start_transaction().
6007 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
6008 struct btrfs_block_rsv *rsv,
6010 u64 *qgroup_reserved,
6011 bool use_global_rsv)
6015 struct btrfs_fs_info *fs_info = root->fs_info;
6016 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6018 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6019 /* One for parent inode, two for dir entries */
6020 num_bytes = 3 * fs_info->nodesize;
6021 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
6028 *qgroup_reserved = num_bytes;
6030 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6031 rsv->space_info = __find_space_info(fs_info,
6032 BTRFS_BLOCK_GROUP_METADATA);
6033 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6034 BTRFS_RESERVE_FLUSH_ALL);
6036 if (ret == -ENOSPC && use_global_rsv)
6037 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
6039 if (ret && *qgroup_reserved)
6040 btrfs_qgroup_free_meta_prealloc(root, *qgroup_reserved);
6045 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6046 struct btrfs_block_rsv *rsv)
6048 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6051 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6052 struct btrfs_inode *inode)
6054 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6055 u64 reserve_size = 0;
6056 u64 qgroup_rsv_size = 0;
6058 unsigned outstanding_extents;
6060 lockdep_assert_held(&inode->lock);
6061 outstanding_extents = inode->outstanding_extents;
6062 if (outstanding_extents)
6063 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6064 outstanding_extents + 1);
6065 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6067 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6070 * For qgroup rsv, the calculation is very simple:
6071 * account one nodesize for each outstanding extent
6073 * This is overestimating in most cases.
6075 qgroup_rsv_size = outstanding_extents * fs_info->nodesize;
6077 spin_lock(&block_rsv->lock);
6078 block_rsv->size = reserve_size;
6079 block_rsv->qgroup_rsv_size = qgroup_rsv_size;
6080 spin_unlock(&block_rsv->lock);
6083 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6085 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6086 unsigned nr_extents;
6087 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6089 bool delalloc_lock = true;
6091 /* If we are a free space inode we need to not flush since we will be in
6092 * the middle of a transaction commit. We also don't need the delalloc
6093 * mutex since we won't race with anybody. We need this mostly to make
6094 * lockdep shut its filthy mouth.
6096 * If we have a transaction open (can happen if we call truncate_block
6097 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6099 if (btrfs_is_free_space_inode(inode)) {
6100 flush = BTRFS_RESERVE_NO_FLUSH;
6101 delalloc_lock = false;
6103 if (current->journal_info)
6104 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6106 if (btrfs_transaction_in_commit(fs_info))
6107 schedule_timeout(1);
6111 mutex_lock(&inode->delalloc_mutex);
6113 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6115 /* Add our new extents and calculate the new rsv size. */
6116 spin_lock(&inode->lock);
6117 nr_extents = count_max_extents(num_bytes);
6118 btrfs_mod_outstanding_extents(inode, nr_extents);
6119 inode->csum_bytes += num_bytes;
6120 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6121 spin_unlock(&inode->lock);
6123 ret = btrfs_inode_rsv_refill(inode, flush);
6128 mutex_unlock(&inode->delalloc_mutex);
6132 spin_lock(&inode->lock);
6133 nr_extents = count_max_extents(num_bytes);
6134 btrfs_mod_outstanding_extents(inode, -nr_extents);
6135 inode->csum_bytes -= num_bytes;
6136 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6137 spin_unlock(&inode->lock);
6139 btrfs_inode_rsv_release(inode, true);
6141 mutex_unlock(&inode->delalloc_mutex);
6146 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6147 * @inode: the inode to release the reservation for.
6148 * @num_bytes: the number of bytes we are releasing.
6149 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6151 * This will release the metadata reservation for an inode. This can be called
6152 * once we complete IO for a given set of bytes to release their metadata
6153 * reservations, or on error for the same reason.
6155 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6158 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6160 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6161 spin_lock(&inode->lock);
6162 inode->csum_bytes -= num_bytes;
6163 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6164 spin_unlock(&inode->lock);
6166 if (btrfs_is_testing(fs_info))
6169 btrfs_inode_rsv_release(inode, qgroup_free);
6173 * btrfs_delalloc_release_extents - release our outstanding_extents
6174 * @inode: the inode to balance the reservation for.
6175 * @num_bytes: the number of bytes we originally reserved with
6176 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6178 * When we reserve space we increase outstanding_extents for the extents we may
6179 * add. Once we've set the range as delalloc or created our ordered extents we
6180 * have outstanding_extents to track the real usage, so we use this to free our
6181 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6182 * with btrfs_delalloc_reserve_metadata.
6184 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6187 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6188 unsigned num_extents;
6190 spin_lock(&inode->lock);
6191 num_extents = count_max_extents(num_bytes);
6192 btrfs_mod_outstanding_extents(inode, -num_extents);
6193 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6194 spin_unlock(&inode->lock);
6196 if (btrfs_is_testing(fs_info))
6199 btrfs_inode_rsv_release(inode, qgroup_free);
6203 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6205 * @inode: inode we're writing to
6206 * @start: start range we are writing to
6207 * @len: how long the range we are writing to
6208 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6209 * current reservation.
6211 * This will do the following things
6213 * o reserve space in data space info for num bytes
6214 * and reserve precious corresponding qgroup space
6215 * (Done in check_data_free_space)
6217 * o reserve space for metadata space, based on the number of outstanding
6218 * extents and how much csums will be needed
6219 * also reserve metadata space in a per root over-reserve method.
6220 * o add to the inodes->delalloc_bytes
6221 * o add it to the fs_info's delalloc inodes list.
6222 * (Above 3 all done in delalloc_reserve_metadata)
6224 * Return 0 for success
6225 * Return <0 for error(-ENOSPC or -EQUOT)
6227 int btrfs_delalloc_reserve_space(struct inode *inode,
6228 struct extent_changeset **reserved, u64 start, u64 len)
6232 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6235 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6237 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6242 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6243 * @inode: inode we're releasing space for
6244 * @start: start position of the space already reserved
6245 * @len: the len of the space already reserved
6246 * @release_bytes: the len of the space we consumed or didn't use
6248 * This function will release the metadata space that was not used and will
6249 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6250 * list if there are no delalloc bytes left.
6251 * Also it will handle the qgroup reserved space.
6253 void btrfs_delalloc_release_space(struct inode *inode,
6254 struct extent_changeset *reserved,
6255 u64 start, u64 len, bool qgroup_free)
6257 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6258 btrfs_free_reserved_data_space(inode, reserved, start, len);
6261 static int update_block_group(struct btrfs_trans_handle *trans,
6262 struct btrfs_fs_info *info, u64 bytenr,
6263 u64 num_bytes, int alloc)
6265 struct btrfs_block_group_cache *cache = NULL;
6266 u64 total = num_bytes;
6271 /* block accounting for super block */
6272 spin_lock(&info->delalloc_root_lock);
6273 old_val = btrfs_super_bytes_used(info->super_copy);
6275 old_val += num_bytes;
6277 old_val -= num_bytes;
6278 btrfs_set_super_bytes_used(info->super_copy, old_val);
6279 spin_unlock(&info->delalloc_root_lock);
6282 cache = btrfs_lookup_block_group(info, bytenr);
6285 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6286 BTRFS_BLOCK_GROUP_RAID1 |
6287 BTRFS_BLOCK_GROUP_RAID10))
6292 * If this block group has free space cache written out, we
6293 * need to make sure to load it if we are removing space. This
6294 * is because we need the unpinning stage to actually add the
6295 * space back to the block group, otherwise we will leak space.
6297 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6298 cache_block_group(cache, 1);
6300 byte_in_group = bytenr - cache->key.objectid;
6301 WARN_ON(byte_in_group > cache->key.offset);
6303 spin_lock(&cache->space_info->lock);
6304 spin_lock(&cache->lock);
6306 if (btrfs_test_opt(info, SPACE_CACHE) &&
6307 cache->disk_cache_state < BTRFS_DC_CLEAR)
6308 cache->disk_cache_state = BTRFS_DC_CLEAR;
6310 old_val = btrfs_block_group_used(&cache->item);
6311 num_bytes = min(total, cache->key.offset - byte_in_group);
6313 old_val += num_bytes;
6314 btrfs_set_block_group_used(&cache->item, old_val);
6315 cache->reserved -= num_bytes;
6316 cache->space_info->bytes_reserved -= num_bytes;
6317 cache->space_info->bytes_used += num_bytes;
6318 cache->space_info->disk_used += num_bytes * factor;
6319 spin_unlock(&cache->lock);
6320 spin_unlock(&cache->space_info->lock);
6322 old_val -= num_bytes;
6323 btrfs_set_block_group_used(&cache->item, old_val);
6324 cache->pinned += num_bytes;
6325 cache->space_info->bytes_pinned += num_bytes;
6326 cache->space_info->bytes_used -= num_bytes;
6327 cache->space_info->disk_used -= num_bytes * factor;
6328 spin_unlock(&cache->lock);
6329 spin_unlock(&cache->space_info->lock);
6331 trace_btrfs_space_reservation(info, "pinned",
6332 cache->space_info->flags,
6334 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6336 set_extent_dirty(info->pinned_extents,
6337 bytenr, bytenr + num_bytes - 1,
6338 GFP_NOFS | __GFP_NOFAIL);
6341 spin_lock(&trans->transaction->dirty_bgs_lock);
6342 if (list_empty(&cache->dirty_list)) {
6343 list_add_tail(&cache->dirty_list,
6344 &trans->transaction->dirty_bgs);
6345 trans->transaction->num_dirty_bgs++;
6346 btrfs_get_block_group(cache);
6348 spin_unlock(&trans->transaction->dirty_bgs_lock);
6351 * No longer have used bytes in this block group, queue it for
6352 * deletion. We do this after adding the block group to the
6353 * dirty list to avoid races between cleaner kthread and space
6356 if (!alloc && old_val == 0) {
6357 spin_lock(&info->unused_bgs_lock);
6358 if (list_empty(&cache->bg_list)) {
6359 btrfs_get_block_group(cache);
6360 list_add_tail(&cache->bg_list,
6363 spin_unlock(&info->unused_bgs_lock);
6366 btrfs_put_block_group(cache);
6368 bytenr += num_bytes;
6373 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6375 struct btrfs_block_group_cache *cache;
6378 spin_lock(&fs_info->block_group_cache_lock);
6379 bytenr = fs_info->first_logical_byte;
6380 spin_unlock(&fs_info->block_group_cache_lock);
6382 if (bytenr < (u64)-1)
6385 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6389 bytenr = cache->key.objectid;
6390 btrfs_put_block_group(cache);
6395 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6396 struct btrfs_block_group_cache *cache,
6397 u64 bytenr, u64 num_bytes, int reserved)
6399 spin_lock(&cache->space_info->lock);
6400 spin_lock(&cache->lock);
6401 cache->pinned += num_bytes;
6402 cache->space_info->bytes_pinned += num_bytes;
6404 cache->reserved -= num_bytes;
6405 cache->space_info->bytes_reserved -= num_bytes;
6407 spin_unlock(&cache->lock);
6408 spin_unlock(&cache->space_info->lock);
6410 trace_btrfs_space_reservation(fs_info, "pinned",
6411 cache->space_info->flags, num_bytes, 1);
6412 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6413 set_extent_dirty(fs_info->pinned_extents, bytenr,
6414 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6419 * this function must be called within transaction
6421 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6422 u64 bytenr, u64 num_bytes, int reserved)
6424 struct btrfs_block_group_cache *cache;
6426 cache = btrfs_lookup_block_group(fs_info, bytenr);
6427 BUG_ON(!cache); /* Logic error */
6429 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6431 btrfs_put_block_group(cache);
6436 * this function must be called within transaction
6438 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6439 u64 bytenr, u64 num_bytes)
6441 struct btrfs_block_group_cache *cache;
6444 cache = btrfs_lookup_block_group(fs_info, bytenr);
6449 * pull in the free space cache (if any) so that our pin
6450 * removes the free space from the cache. We have load_only set
6451 * to one because the slow code to read in the free extents does check
6452 * the pinned extents.
6454 cache_block_group(cache, 1);
6456 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6458 /* remove us from the free space cache (if we're there at all) */
6459 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6460 btrfs_put_block_group(cache);
6464 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6465 u64 start, u64 num_bytes)
6468 struct btrfs_block_group_cache *block_group;
6469 struct btrfs_caching_control *caching_ctl;
6471 block_group = btrfs_lookup_block_group(fs_info, start);
6475 cache_block_group(block_group, 0);
6476 caching_ctl = get_caching_control(block_group);
6480 BUG_ON(!block_group_cache_done(block_group));
6481 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6483 mutex_lock(&caching_ctl->mutex);
6485 if (start >= caching_ctl->progress) {
6486 ret = add_excluded_extent(fs_info, start, num_bytes);
6487 } else if (start + num_bytes <= caching_ctl->progress) {
6488 ret = btrfs_remove_free_space(block_group,
6491 num_bytes = caching_ctl->progress - start;
6492 ret = btrfs_remove_free_space(block_group,
6497 num_bytes = (start + num_bytes) -
6498 caching_ctl->progress;
6499 start = caching_ctl->progress;
6500 ret = add_excluded_extent(fs_info, start, num_bytes);
6503 mutex_unlock(&caching_ctl->mutex);
6504 put_caching_control(caching_ctl);
6506 btrfs_put_block_group(block_group);
6510 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6511 struct extent_buffer *eb)
6513 struct btrfs_file_extent_item *item;
6514 struct btrfs_key key;
6518 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6521 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6522 btrfs_item_key_to_cpu(eb, &key, i);
6523 if (key.type != BTRFS_EXTENT_DATA_KEY)
6525 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6526 found_type = btrfs_file_extent_type(eb, item);
6527 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6529 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6531 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6532 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6533 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6540 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6542 atomic_inc(&bg->reservations);
6545 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6548 struct btrfs_block_group_cache *bg;
6550 bg = btrfs_lookup_block_group(fs_info, start);
6552 if (atomic_dec_and_test(&bg->reservations))
6553 wake_up_var(&bg->reservations);
6554 btrfs_put_block_group(bg);
6557 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6559 struct btrfs_space_info *space_info = bg->space_info;
6563 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6567 * Our block group is read only but before we set it to read only,
6568 * some task might have had allocated an extent from it already, but it
6569 * has not yet created a respective ordered extent (and added it to a
6570 * root's list of ordered extents).
6571 * Therefore wait for any task currently allocating extents, since the
6572 * block group's reservations counter is incremented while a read lock
6573 * on the groups' semaphore is held and decremented after releasing
6574 * the read access on that semaphore and creating the ordered extent.
6576 down_write(&space_info->groups_sem);
6577 up_write(&space_info->groups_sem);
6579 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6583 * btrfs_add_reserved_bytes - update the block_group and space info counters
6584 * @cache: The cache we are manipulating
6585 * @ram_bytes: The number of bytes of file content, and will be same to
6586 * @num_bytes except for the compress path.
6587 * @num_bytes: The number of bytes in question
6588 * @delalloc: The blocks are allocated for the delalloc write
6590 * This is called by the allocator when it reserves space. If this is a
6591 * reservation and the block group has become read only we cannot make the
6592 * reservation and return -EAGAIN, otherwise this function always succeeds.
6594 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6595 u64 ram_bytes, u64 num_bytes, int delalloc)
6597 struct btrfs_space_info *space_info = cache->space_info;
6600 spin_lock(&space_info->lock);
6601 spin_lock(&cache->lock);
6605 cache->reserved += num_bytes;
6606 space_info->bytes_reserved += num_bytes;
6608 trace_btrfs_space_reservation(cache->fs_info,
6609 "space_info", space_info->flags,
6611 space_info->bytes_may_use -= ram_bytes;
6613 cache->delalloc_bytes += num_bytes;
6615 spin_unlock(&cache->lock);
6616 spin_unlock(&space_info->lock);
6621 * btrfs_free_reserved_bytes - update the block_group and space info counters
6622 * @cache: The cache we are manipulating
6623 * @num_bytes: The number of bytes in question
6624 * @delalloc: The blocks are allocated for the delalloc write
6626 * This is called by somebody who is freeing space that was never actually used
6627 * on disk. For example if you reserve some space for a new leaf in transaction
6628 * A and before transaction A commits you free that leaf, you call this with
6629 * reserve set to 0 in order to clear the reservation.
6632 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6633 u64 num_bytes, int delalloc)
6635 struct btrfs_space_info *space_info = cache->space_info;
6638 spin_lock(&space_info->lock);
6639 spin_lock(&cache->lock);
6641 space_info->bytes_readonly += num_bytes;
6642 cache->reserved -= num_bytes;
6643 space_info->bytes_reserved -= num_bytes;
6646 cache->delalloc_bytes -= num_bytes;
6647 spin_unlock(&cache->lock);
6648 spin_unlock(&space_info->lock);
6651 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6653 struct btrfs_caching_control *next;
6654 struct btrfs_caching_control *caching_ctl;
6655 struct btrfs_block_group_cache *cache;
6657 down_write(&fs_info->commit_root_sem);
6659 list_for_each_entry_safe(caching_ctl, next,
6660 &fs_info->caching_block_groups, list) {
6661 cache = caching_ctl->block_group;
6662 if (block_group_cache_done(cache)) {
6663 cache->last_byte_to_unpin = (u64)-1;
6664 list_del_init(&caching_ctl->list);
6665 put_caching_control(caching_ctl);
6667 cache->last_byte_to_unpin = caching_ctl->progress;
6671 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6672 fs_info->pinned_extents = &fs_info->freed_extents[1];
6674 fs_info->pinned_extents = &fs_info->freed_extents[0];
6676 up_write(&fs_info->commit_root_sem);
6678 update_global_block_rsv(fs_info);
6682 * Returns the free cluster for the given space info and sets empty_cluster to
6683 * what it should be based on the mount options.
6685 static struct btrfs_free_cluster *
6686 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6687 struct btrfs_space_info *space_info, u64 *empty_cluster)
6689 struct btrfs_free_cluster *ret = NULL;
6692 if (btrfs_mixed_space_info(space_info))
6695 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6696 ret = &fs_info->meta_alloc_cluster;
6697 if (btrfs_test_opt(fs_info, SSD))
6698 *empty_cluster = SZ_2M;
6700 *empty_cluster = SZ_64K;
6701 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6702 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6703 *empty_cluster = SZ_2M;
6704 ret = &fs_info->data_alloc_cluster;
6710 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6712 const bool return_free_space)
6714 struct btrfs_block_group_cache *cache = NULL;
6715 struct btrfs_space_info *space_info;
6716 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6717 struct btrfs_free_cluster *cluster = NULL;
6719 u64 total_unpinned = 0;
6720 u64 empty_cluster = 0;
6723 while (start <= end) {
6726 start >= cache->key.objectid + cache->key.offset) {
6728 btrfs_put_block_group(cache);
6730 cache = btrfs_lookup_block_group(fs_info, start);
6731 BUG_ON(!cache); /* Logic error */
6733 cluster = fetch_cluster_info(fs_info,
6736 empty_cluster <<= 1;
6739 len = cache->key.objectid + cache->key.offset - start;
6740 len = min(len, end + 1 - start);
6742 if (start < cache->last_byte_to_unpin) {
6743 len = min(len, cache->last_byte_to_unpin - start);
6744 if (return_free_space)
6745 btrfs_add_free_space(cache, start, len);
6749 total_unpinned += len;
6750 space_info = cache->space_info;
6753 * If this space cluster has been marked as fragmented and we've
6754 * unpinned enough in this block group to potentially allow a
6755 * cluster to be created inside of it go ahead and clear the
6758 if (cluster && cluster->fragmented &&
6759 total_unpinned > empty_cluster) {
6760 spin_lock(&cluster->lock);
6761 cluster->fragmented = 0;
6762 spin_unlock(&cluster->lock);
6765 spin_lock(&space_info->lock);
6766 spin_lock(&cache->lock);
6767 cache->pinned -= len;
6768 space_info->bytes_pinned -= len;
6770 trace_btrfs_space_reservation(fs_info, "pinned",
6771 space_info->flags, len, 0);
6772 space_info->max_extent_size = 0;
6773 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6775 space_info->bytes_readonly += len;
6778 spin_unlock(&cache->lock);
6779 if (!readonly && return_free_space &&
6780 global_rsv->space_info == space_info) {
6783 spin_lock(&global_rsv->lock);
6784 if (!global_rsv->full) {
6785 to_add = min(len, global_rsv->size -
6786 global_rsv->reserved);
6787 global_rsv->reserved += to_add;
6788 space_info->bytes_may_use += to_add;
6789 if (global_rsv->reserved >= global_rsv->size)
6790 global_rsv->full = 1;
6791 trace_btrfs_space_reservation(fs_info,
6797 spin_unlock(&global_rsv->lock);
6798 /* Add to any tickets we may have */
6800 space_info_add_new_bytes(fs_info, space_info,
6803 spin_unlock(&space_info->lock);
6807 btrfs_put_block_group(cache);
6811 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6813 struct btrfs_fs_info *fs_info = trans->fs_info;
6814 struct btrfs_block_group_cache *block_group, *tmp;
6815 struct list_head *deleted_bgs;
6816 struct extent_io_tree *unpin;
6821 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6822 unpin = &fs_info->freed_extents[1];
6824 unpin = &fs_info->freed_extents[0];
6826 while (!trans->aborted) {
6827 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6828 ret = find_first_extent_bit(unpin, 0, &start, &end,
6829 EXTENT_DIRTY, NULL);
6831 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6835 if (btrfs_test_opt(fs_info, DISCARD))
6836 ret = btrfs_discard_extent(fs_info, start,
6837 end + 1 - start, NULL);
6839 clear_extent_dirty(unpin, start, end);
6840 unpin_extent_range(fs_info, start, end, true);
6841 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6846 * Transaction is finished. We don't need the lock anymore. We
6847 * do need to clean up the block groups in case of a transaction
6850 deleted_bgs = &trans->transaction->deleted_bgs;
6851 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6855 if (!trans->aborted)
6856 ret = btrfs_discard_extent(fs_info,
6857 block_group->key.objectid,
6858 block_group->key.offset,
6861 list_del_init(&block_group->bg_list);
6862 btrfs_put_block_group_trimming(block_group);
6863 btrfs_put_block_group(block_group);
6866 const char *errstr = btrfs_decode_error(ret);
6868 "discard failed while removing blockgroup: errno=%d %s",
6876 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6877 struct btrfs_fs_info *info,
6878 struct btrfs_delayed_ref_node *node, u64 parent,
6879 u64 root_objectid, u64 owner_objectid,
6880 u64 owner_offset, int refs_to_drop,
6881 struct btrfs_delayed_extent_op *extent_op)
6883 struct btrfs_key key;
6884 struct btrfs_path *path;
6885 struct btrfs_root *extent_root = info->extent_root;
6886 struct extent_buffer *leaf;
6887 struct btrfs_extent_item *ei;
6888 struct btrfs_extent_inline_ref *iref;
6891 int extent_slot = 0;
6892 int found_extent = 0;
6896 u64 bytenr = node->bytenr;
6897 u64 num_bytes = node->num_bytes;
6899 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6901 path = btrfs_alloc_path();
6905 path->reada = READA_FORWARD;
6906 path->leave_spinning = 1;
6908 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6909 BUG_ON(!is_data && refs_to_drop != 1);
6912 skinny_metadata = false;
6914 ret = lookup_extent_backref(trans, info, path, &iref,
6915 bytenr, num_bytes, parent,
6916 root_objectid, owner_objectid,
6919 extent_slot = path->slots[0];
6920 while (extent_slot >= 0) {
6921 btrfs_item_key_to_cpu(path->nodes[0], &key,
6923 if (key.objectid != bytenr)
6925 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6926 key.offset == num_bytes) {
6930 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6931 key.offset == owner_objectid) {
6935 if (path->slots[0] - extent_slot > 5)
6939 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6940 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6941 if (found_extent && item_size < sizeof(*ei))
6944 if (!found_extent) {
6946 ret = remove_extent_backref(trans, info, path, NULL,
6948 is_data, &last_ref);
6950 btrfs_abort_transaction(trans, ret);
6953 btrfs_release_path(path);
6954 path->leave_spinning = 1;
6956 key.objectid = bytenr;
6957 key.type = BTRFS_EXTENT_ITEM_KEY;
6958 key.offset = num_bytes;
6960 if (!is_data && skinny_metadata) {
6961 key.type = BTRFS_METADATA_ITEM_KEY;
6962 key.offset = owner_objectid;
6965 ret = btrfs_search_slot(trans, extent_root,
6967 if (ret > 0 && skinny_metadata && path->slots[0]) {
6969 * Couldn't find our skinny metadata item,
6970 * see if we have ye olde extent item.
6973 btrfs_item_key_to_cpu(path->nodes[0], &key,
6975 if (key.objectid == bytenr &&
6976 key.type == BTRFS_EXTENT_ITEM_KEY &&
6977 key.offset == num_bytes)
6981 if (ret > 0 && skinny_metadata) {
6982 skinny_metadata = false;
6983 key.objectid = bytenr;
6984 key.type = BTRFS_EXTENT_ITEM_KEY;
6985 key.offset = num_bytes;
6986 btrfs_release_path(path);
6987 ret = btrfs_search_slot(trans, extent_root,
6993 "umm, got %d back from search, was looking for %llu",
6996 btrfs_print_leaf(path->nodes[0]);
6999 btrfs_abort_transaction(trans, ret);
7002 extent_slot = path->slots[0];
7004 } else if (WARN_ON(ret == -ENOENT)) {
7005 btrfs_print_leaf(path->nodes[0]);
7007 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7008 bytenr, parent, root_objectid, owner_objectid,
7010 btrfs_abort_transaction(trans, ret);
7013 btrfs_abort_transaction(trans, ret);
7017 leaf = path->nodes[0];
7018 item_size = btrfs_item_size_nr(leaf, extent_slot);
7019 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7020 if (item_size < sizeof(*ei)) {
7021 BUG_ON(found_extent || extent_slot != path->slots[0]);
7022 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
7025 btrfs_abort_transaction(trans, ret);
7029 btrfs_release_path(path);
7030 path->leave_spinning = 1;
7032 key.objectid = bytenr;
7033 key.type = BTRFS_EXTENT_ITEM_KEY;
7034 key.offset = num_bytes;
7036 ret = btrfs_search_slot(trans, extent_root, &key, path,
7040 "umm, got %d back from search, was looking for %llu",
7042 btrfs_print_leaf(path->nodes[0]);
7045 btrfs_abort_transaction(trans, ret);
7049 extent_slot = path->slots[0];
7050 leaf = path->nodes[0];
7051 item_size = btrfs_item_size_nr(leaf, extent_slot);
7054 BUG_ON(item_size < sizeof(*ei));
7055 ei = btrfs_item_ptr(leaf, extent_slot,
7056 struct btrfs_extent_item);
7057 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7058 key.type == BTRFS_EXTENT_ITEM_KEY) {
7059 struct btrfs_tree_block_info *bi;
7060 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7061 bi = (struct btrfs_tree_block_info *)(ei + 1);
7062 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7065 refs = btrfs_extent_refs(leaf, ei);
7066 if (refs < refs_to_drop) {
7068 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7069 refs_to_drop, refs, bytenr);
7071 btrfs_abort_transaction(trans, ret);
7074 refs -= refs_to_drop;
7078 __run_delayed_extent_op(extent_op, leaf, ei);
7080 * In the case of inline back ref, reference count will
7081 * be updated by remove_extent_backref
7084 BUG_ON(!found_extent);
7086 btrfs_set_extent_refs(leaf, ei, refs);
7087 btrfs_mark_buffer_dirty(leaf);
7090 ret = remove_extent_backref(trans, info, path,
7092 is_data, &last_ref);
7094 btrfs_abort_transaction(trans, ret);
7100 BUG_ON(is_data && refs_to_drop !=
7101 extent_data_ref_count(path, iref));
7103 BUG_ON(path->slots[0] != extent_slot);
7105 BUG_ON(path->slots[0] != extent_slot + 1);
7106 path->slots[0] = extent_slot;
7112 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7115 btrfs_abort_transaction(trans, ret);
7118 btrfs_release_path(path);
7121 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7123 btrfs_abort_transaction(trans, ret);
7128 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7130 btrfs_abort_transaction(trans, ret);
7134 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7136 btrfs_abort_transaction(trans, ret);
7140 btrfs_release_path(path);
7143 btrfs_free_path(path);
7148 * when we free an block, it is possible (and likely) that we free the last
7149 * delayed ref for that extent as well. This searches the delayed ref tree for
7150 * a given extent, and if there are no other delayed refs to be processed, it
7151 * removes it from the tree.
7153 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7156 struct btrfs_delayed_ref_head *head;
7157 struct btrfs_delayed_ref_root *delayed_refs;
7160 delayed_refs = &trans->transaction->delayed_refs;
7161 spin_lock(&delayed_refs->lock);
7162 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7164 goto out_delayed_unlock;
7166 spin_lock(&head->lock);
7167 if (!RB_EMPTY_ROOT(&head->ref_tree))
7170 if (head->extent_op) {
7171 if (!head->must_insert_reserved)
7173 btrfs_free_delayed_extent_op(head->extent_op);
7174 head->extent_op = NULL;
7178 * waiting for the lock here would deadlock. If someone else has it
7179 * locked they are already in the process of dropping it anyway
7181 if (!mutex_trylock(&head->mutex))
7185 * at this point we have a head with no other entries. Go
7186 * ahead and process it.
7188 rb_erase(&head->href_node, &delayed_refs->href_root);
7189 RB_CLEAR_NODE(&head->href_node);
7190 atomic_dec(&delayed_refs->num_entries);
7193 * we don't take a ref on the node because we're removing it from the
7194 * tree, so we just steal the ref the tree was holding.
7196 delayed_refs->num_heads--;
7197 if (head->processing == 0)
7198 delayed_refs->num_heads_ready--;
7199 head->processing = 0;
7200 spin_unlock(&head->lock);
7201 spin_unlock(&delayed_refs->lock);
7203 BUG_ON(head->extent_op);
7204 if (head->must_insert_reserved)
7207 mutex_unlock(&head->mutex);
7208 btrfs_put_delayed_ref_head(head);
7211 spin_unlock(&head->lock);
7214 spin_unlock(&delayed_refs->lock);
7218 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7219 struct btrfs_root *root,
7220 struct extent_buffer *buf,
7221 u64 parent, int last_ref)
7223 struct btrfs_fs_info *fs_info = root->fs_info;
7227 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7228 int old_ref_mod, new_ref_mod;
7230 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7231 root->root_key.objectid,
7232 btrfs_header_level(buf), 0,
7233 BTRFS_DROP_DELAYED_REF);
7234 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7236 root->root_key.objectid,
7237 btrfs_header_level(buf),
7238 BTRFS_DROP_DELAYED_REF, NULL,
7239 &old_ref_mod, &new_ref_mod);
7240 BUG_ON(ret); /* -ENOMEM */
7241 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7244 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7245 struct btrfs_block_group_cache *cache;
7247 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7248 ret = check_ref_cleanup(trans, buf->start);
7254 cache = btrfs_lookup_block_group(fs_info, buf->start);
7256 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7257 pin_down_extent(fs_info, cache, buf->start,
7259 btrfs_put_block_group(cache);
7263 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7265 btrfs_add_free_space(cache, buf->start, buf->len);
7266 btrfs_free_reserved_bytes(cache, buf->len, 0);
7267 btrfs_put_block_group(cache);
7268 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7272 add_pinned_bytes(fs_info, buf->len, true,
7273 root->root_key.objectid);
7277 * Deleting the buffer, clear the corrupt flag since it doesn't
7280 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7284 /* Can return -ENOMEM */
7285 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7286 struct btrfs_root *root,
7287 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7288 u64 owner, u64 offset)
7290 struct btrfs_fs_info *fs_info = root->fs_info;
7291 int old_ref_mod, new_ref_mod;
7294 if (btrfs_is_testing(fs_info))
7297 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7298 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7299 root_objectid, owner, offset,
7300 BTRFS_DROP_DELAYED_REF);
7303 * tree log blocks never actually go into the extent allocation
7304 * tree, just update pinning info and exit early.
7306 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7307 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7308 /* unlocks the pinned mutex */
7309 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7310 old_ref_mod = new_ref_mod = 0;
7312 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7313 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7315 root_objectid, (int)owner,
7316 BTRFS_DROP_DELAYED_REF, NULL,
7317 &old_ref_mod, &new_ref_mod);
7319 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7321 root_objectid, owner, offset,
7322 0, BTRFS_DROP_DELAYED_REF,
7323 &old_ref_mod, &new_ref_mod);
7326 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) {
7327 bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID;
7329 add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid);
7336 * when we wait for progress in the block group caching, its because
7337 * our allocation attempt failed at least once. So, we must sleep
7338 * and let some progress happen before we try again.
7340 * This function will sleep at least once waiting for new free space to
7341 * show up, and then it will check the block group free space numbers
7342 * for our min num_bytes. Another option is to have it go ahead
7343 * and look in the rbtree for a free extent of a given size, but this
7346 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7347 * any of the information in this block group.
7349 static noinline void
7350 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7353 struct btrfs_caching_control *caching_ctl;
7355 caching_ctl = get_caching_control(cache);
7359 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7360 (cache->free_space_ctl->free_space >= num_bytes));
7362 put_caching_control(caching_ctl);
7366 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7368 struct btrfs_caching_control *caching_ctl;
7371 caching_ctl = get_caching_control(cache);
7373 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7375 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7376 if (cache->cached == BTRFS_CACHE_ERROR)
7378 put_caching_control(caching_ctl);
7382 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7383 [BTRFS_RAID_RAID10] = "raid10",
7384 [BTRFS_RAID_RAID1] = "raid1",
7385 [BTRFS_RAID_DUP] = "dup",
7386 [BTRFS_RAID_RAID0] = "raid0",
7387 [BTRFS_RAID_SINGLE] = "single",
7388 [BTRFS_RAID_RAID5] = "raid5",
7389 [BTRFS_RAID_RAID6] = "raid6",
7392 static const char *get_raid_name(enum btrfs_raid_types type)
7394 if (type >= BTRFS_NR_RAID_TYPES)
7397 return btrfs_raid_type_names[type];
7400 enum btrfs_loop_type {
7401 LOOP_CACHING_NOWAIT = 0,
7402 LOOP_CACHING_WAIT = 1,
7403 LOOP_ALLOC_CHUNK = 2,
7404 LOOP_NO_EMPTY_SIZE = 3,
7408 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7412 down_read(&cache->data_rwsem);
7416 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7419 btrfs_get_block_group(cache);
7421 down_read(&cache->data_rwsem);
7424 static struct btrfs_block_group_cache *
7425 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7426 struct btrfs_free_cluster *cluster,
7429 struct btrfs_block_group_cache *used_bg = NULL;
7431 spin_lock(&cluster->refill_lock);
7433 used_bg = cluster->block_group;
7437 if (used_bg == block_group)
7440 btrfs_get_block_group(used_bg);
7445 if (down_read_trylock(&used_bg->data_rwsem))
7448 spin_unlock(&cluster->refill_lock);
7450 /* We should only have one-level nested. */
7451 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7453 spin_lock(&cluster->refill_lock);
7454 if (used_bg == cluster->block_group)
7457 up_read(&used_bg->data_rwsem);
7458 btrfs_put_block_group(used_bg);
7463 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7467 up_read(&cache->data_rwsem);
7468 btrfs_put_block_group(cache);
7472 * walks the btree of allocated extents and find a hole of a given size.
7473 * The key ins is changed to record the hole:
7474 * ins->objectid == start position
7475 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7476 * ins->offset == the size of the hole.
7477 * Any available blocks before search_start are skipped.
7479 * If there is no suitable free space, we will record the max size of
7480 * the free space extent currently.
7482 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7483 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7484 u64 hint_byte, struct btrfs_key *ins,
7485 u64 flags, int delalloc)
7488 struct btrfs_root *root = fs_info->extent_root;
7489 struct btrfs_free_cluster *last_ptr = NULL;
7490 struct btrfs_block_group_cache *block_group = NULL;
7491 u64 search_start = 0;
7492 u64 max_extent_size = 0;
7493 u64 empty_cluster = 0;
7494 struct btrfs_space_info *space_info;
7496 int index = btrfs_bg_flags_to_raid_index(flags);
7497 bool failed_cluster_refill = false;
7498 bool failed_alloc = false;
7499 bool use_cluster = true;
7500 bool have_caching_bg = false;
7501 bool orig_have_caching_bg = false;
7502 bool full_search = false;
7504 WARN_ON(num_bytes < fs_info->sectorsize);
7505 ins->type = BTRFS_EXTENT_ITEM_KEY;
7509 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7511 space_info = __find_space_info(fs_info, flags);
7513 btrfs_err(fs_info, "No space info for %llu", flags);
7518 * If our free space is heavily fragmented we may not be able to make
7519 * big contiguous allocations, so instead of doing the expensive search
7520 * for free space, simply return ENOSPC with our max_extent_size so we
7521 * can go ahead and search for a more manageable chunk.
7523 * If our max_extent_size is large enough for our allocation simply
7524 * disable clustering since we will likely not be able to find enough
7525 * space to create a cluster and induce latency trying.
7527 if (unlikely(space_info->max_extent_size)) {
7528 spin_lock(&space_info->lock);
7529 if (space_info->max_extent_size &&
7530 num_bytes > space_info->max_extent_size) {
7531 ins->offset = space_info->max_extent_size;
7532 spin_unlock(&space_info->lock);
7534 } else if (space_info->max_extent_size) {
7535 use_cluster = false;
7537 spin_unlock(&space_info->lock);
7540 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7542 spin_lock(&last_ptr->lock);
7543 if (last_ptr->block_group)
7544 hint_byte = last_ptr->window_start;
7545 if (last_ptr->fragmented) {
7547 * We still set window_start so we can keep track of the
7548 * last place we found an allocation to try and save
7551 hint_byte = last_ptr->window_start;
7552 use_cluster = false;
7554 spin_unlock(&last_ptr->lock);
7557 search_start = max(search_start, first_logical_byte(fs_info, 0));
7558 search_start = max(search_start, hint_byte);
7559 if (search_start == hint_byte) {
7560 block_group = btrfs_lookup_block_group(fs_info, search_start);
7562 * we don't want to use the block group if it doesn't match our
7563 * allocation bits, or if its not cached.
7565 * However if we are re-searching with an ideal block group
7566 * picked out then we don't care that the block group is cached.
7568 if (block_group && block_group_bits(block_group, flags) &&
7569 block_group->cached != BTRFS_CACHE_NO) {
7570 down_read(&space_info->groups_sem);
7571 if (list_empty(&block_group->list) ||
7574 * someone is removing this block group,
7575 * we can't jump into the have_block_group
7576 * target because our list pointers are not
7579 btrfs_put_block_group(block_group);
7580 up_read(&space_info->groups_sem);
7582 index = btrfs_bg_flags_to_raid_index(
7583 block_group->flags);
7584 btrfs_lock_block_group(block_group, delalloc);
7585 goto have_block_group;
7587 } else if (block_group) {
7588 btrfs_put_block_group(block_group);
7592 have_caching_bg = false;
7593 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7595 down_read(&space_info->groups_sem);
7596 list_for_each_entry(block_group, &space_info->block_groups[index],
7601 /* If the block group is read-only, we can skip it entirely. */
7602 if (unlikely(block_group->ro))
7605 btrfs_grab_block_group(block_group, delalloc);
7606 search_start = block_group->key.objectid;
7609 * this can happen if we end up cycling through all the
7610 * raid types, but we want to make sure we only allocate
7611 * for the proper type.
7613 if (!block_group_bits(block_group, flags)) {
7614 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7615 BTRFS_BLOCK_GROUP_RAID1 |
7616 BTRFS_BLOCK_GROUP_RAID5 |
7617 BTRFS_BLOCK_GROUP_RAID6 |
7618 BTRFS_BLOCK_GROUP_RAID10;
7621 * if they asked for extra copies and this block group
7622 * doesn't provide them, bail. This does allow us to
7623 * fill raid0 from raid1.
7625 if ((flags & extra) && !(block_group->flags & extra))
7630 cached = block_group_cache_done(block_group);
7631 if (unlikely(!cached)) {
7632 have_caching_bg = true;
7633 ret = cache_block_group(block_group, 0);
7638 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7642 * Ok we want to try and use the cluster allocator, so
7645 if (last_ptr && use_cluster) {
7646 struct btrfs_block_group_cache *used_block_group;
7647 unsigned long aligned_cluster;
7649 * the refill lock keeps out other
7650 * people trying to start a new cluster
7652 used_block_group = btrfs_lock_cluster(block_group,
7655 if (!used_block_group)
7656 goto refill_cluster;
7658 if (used_block_group != block_group &&
7659 (used_block_group->ro ||
7660 !block_group_bits(used_block_group, flags)))
7661 goto release_cluster;
7663 offset = btrfs_alloc_from_cluster(used_block_group,
7666 used_block_group->key.objectid,
7669 /* we have a block, we're done */
7670 spin_unlock(&last_ptr->refill_lock);
7671 trace_btrfs_reserve_extent_cluster(fs_info,
7673 search_start, num_bytes);
7674 if (used_block_group != block_group) {
7675 btrfs_release_block_group(block_group,
7677 block_group = used_block_group;
7682 WARN_ON(last_ptr->block_group != used_block_group);
7684 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7685 * set up a new clusters, so lets just skip it
7686 * and let the allocator find whatever block
7687 * it can find. If we reach this point, we
7688 * will have tried the cluster allocator
7689 * plenty of times and not have found
7690 * anything, so we are likely way too
7691 * fragmented for the clustering stuff to find
7694 * However, if the cluster is taken from the
7695 * current block group, release the cluster
7696 * first, so that we stand a better chance of
7697 * succeeding in the unclustered
7699 if (loop >= LOOP_NO_EMPTY_SIZE &&
7700 used_block_group != block_group) {
7701 spin_unlock(&last_ptr->refill_lock);
7702 btrfs_release_block_group(used_block_group,
7704 goto unclustered_alloc;
7708 * this cluster didn't work out, free it and
7711 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7713 if (used_block_group != block_group)
7714 btrfs_release_block_group(used_block_group,
7717 if (loop >= LOOP_NO_EMPTY_SIZE) {
7718 spin_unlock(&last_ptr->refill_lock);
7719 goto unclustered_alloc;
7722 aligned_cluster = max_t(unsigned long,
7723 empty_cluster + empty_size,
7724 block_group->full_stripe_len);
7726 /* allocate a cluster in this block group */
7727 ret = btrfs_find_space_cluster(fs_info, block_group,
7728 last_ptr, search_start,
7733 * now pull our allocation out of this
7736 offset = btrfs_alloc_from_cluster(block_group,
7742 /* we found one, proceed */
7743 spin_unlock(&last_ptr->refill_lock);
7744 trace_btrfs_reserve_extent_cluster(fs_info,
7745 block_group, search_start,
7749 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7750 && !failed_cluster_refill) {
7751 spin_unlock(&last_ptr->refill_lock);
7753 failed_cluster_refill = true;
7754 wait_block_group_cache_progress(block_group,
7755 num_bytes + empty_cluster + empty_size);
7756 goto have_block_group;
7760 * at this point we either didn't find a cluster
7761 * or we weren't able to allocate a block from our
7762 * cluster. Free the cluster we've been trying
7763 * to use, and go to the next block group
7765 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7766 spin_unlock(&last_ptr->refill_lock);
7772 * We are doing an unclustered alloc, set the fragmented flag so
7773 * we don't bother trying to setup a cluster again until we get
7776 if (unlikely(last_ptr)) {
7777 spin_lock(&last_ptr->lock);
7778 last_ptr->fragmented = 1;
7779 spin_unlock(&last_ptr->lock);
7782 struct btrfs_free_space_ctl *ctl =
7783 block_group->free_space_ctl;
7785 spin_lock(&ctl->tree_lock);
7786 if (ctl->free_space <
7787 num_bytes + empty_cluster + empty_size) {
7788 if (ctl->free_space > max_extent_size)
7789 max_extent_size = ctl->free_space;
7790 spin_unlock(&ctl->tree_lock);
7793 spin_unlock(&ctl->tree_lock);
7796 offset = btrfs_find_space_for_alloc(block_group, search_start,
7797 num_bytes, empty_size,
7800 * If we didn't find a chunk, and we haven't failed on this
7801 * block group before, and this block group is in the middle of
7802 * caching and we are ok with waiting, then go ahead and wait
7803 * for progress to be made, and set failed_alloc to true.
7805 * If failed_alloc is true then we've already waited on this
7806 * block group once and should move on to the next block group.
7808 if (!offset && !failed_alloc && !cached &&
7809 loop > LOOP_CACHING_NOWAIT) {
7810 wait_block_group_cache_progress(block_group,
7811 num_bytes + empty_size);
7812 failed_alloc = true;
7813 goto have_block_group;
7814 } else if (!offset) {
7818 search_start = ALIGN(offset, fs_info->stripesize);
7820 /* move on to the next group */
7821 if (search_start + num_bytes >
7822 block_group->key.objectid + block_group->key.offset) {
7823 btrfs_add_free_space(block_group, offset, num_bytes);
7827 if (offset < search_start)
7828 btrfs_add_free_space(block_group, offset,
7829 search_start - offset);
7830 BUG_ON(offset > search_start);
7832 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7833 num_bytes, delalloc);
7834 if (ret == -EAGAIN) {
7835 btrfs_add_free_space(block_group, offset, num_bytes);
7838 btrfs_inc_block_group_reservations(block_group);
7840 /* we are all good, lets return */
7841 ins->objectid = search_start;
7842 ins->offset = num_bytes;
7844 trace_btrfs_reserve_extent(fs_info, block_group,
7845 search_start, num_bytes);
7846 btrfs_release_block_group(block_group, delalloc);
7849 failed_cluster_refill = false;
7850 failed_alloc = false;
7851 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7853 btrfs_release_block_group(block_group, delalloc);
7856 up_read(&space_info->groups_sem);
7858 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7859 && !orig_have_caching_bg)
7860 orig_have_caching_bg = true;
7862 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7865 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7869 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7870 * caching kthreads as we move along
7871 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7872 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7873 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7876 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7878 if (loop == LOOP_CACHING_NOWAIT) {
7880 * We want to skip the LOOP_CACHING_WAIT step if we
7881 * don't have any uncached bgs and we've already done a
7882 * full search through.
7884 if (orig_have_caching_bg || !full_search)
7885 loop = LOOP_CACHING_WAIT;
7887 loop = LOOP_ALLOC_CHUNK;
7892 if (loop == LOOP_ALLOC_CHUNK) {
7893 struct btrfs_trans_handle *trans;
7896 trans = current->journal_info;
7900 trans = btrfs_join_transaction(root);
7902 if (IS_ERR(trans)) {
7903 ret = PTR_ERR(trans);
7907 ret = do_chunk_alloc(trans, fs_info, flags,
7911 * If we can't allocate a new chunk we've already looped
7912 * through at least once, move on to the NO_EMPTY_SIZE
7916 loop = LOOP_NO_EMPTY_SIZE;
7919 * Do not bail out on ENOSPC since we
7920 * can do more things.
7922 if (ret < 0 && ret != -ENOSPC)
7923 btrfs_abort_transaction(trans, ret);
7927 btrfs_end_transaction(trans);
7932 if (loop == LOOP_NO_EMPTY_SIZE) {
7934 * Don't loop again if we already have no empty_size and
7937 if (empty_size == 0 &&
7938 empty_cluster == 0) {
7947 } else if (!ins->objectid) {
7949 } else if (ins->objectid) {
7950 if (!use_cluster && last_ptr) {
7951 spin_lock(&last_ptr->lock);
7952 last_ptr->window_start = ins->objectid;
7953 spin_unlock(&last_ptr->lock);
7958 if (ret == -ENOSPC) {
7959 spin_lock(&space_info->lock);
7960 space_info->max_extent_size = max_extent_size;
7961 spin_unlock(&space_info->lock);
7962 ins->offset = max_extent_size;
7967 static void dump_space_info(struct btrfs_fs_info *fs_info,
7968 struct btrfs_space_info *info, u64 bytes,
7969 int dump_block_groups)
7971 struct btrfs_block_group_cache *cache;
7974 spin_lock(&info->lock);
7975 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7977 info->total_bytes - btrfs_space_info_used(info, true),
7978 info->full ? "" : "not ");
7980 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7981 info->total_bytes, info->bytes_used, info->bytes_pinned,
7982 info->bytes_reserved, info->bytes_may_use,
7983 info->bytes_readonly);
7984 spin_unlock(&info->lock);
7986 if (!dump_block_groups)
7989 down_read(&info->groups_sem);
7991 list_for_each_entry(cache, &info->block_groups[index], list) {
7992 spin_lock(&cache->lock);
7994 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7995 cache->key.objectid, cache->key.offset,
7996 btrfs_block_group_used(&cache->item), cache->pinned,
7997 cache->reserved, cache->ro ? "[readonly]" : "");
7998 btrfs_dump_free_space(cache, bytes);
7999 spin_unlock(&cache->lock);
8001 if (++index < BTRFS_NR_RAID_TYPES)
8003 up_read(&info->groups_sem);
8007 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
8008 * hole that is at least as big as @num_bytes.
8010 * @root - The root that will contain this extent
8012 * @ram_bytes - The amount of space in ram that @num_bytes take. This
8013 * is used for accounting purposes. This value differs
8014 * from @num_bytes only in the case of compressed extents.
8016 * @num_bytes - Number of bytes to allocate on-disk.
8018 * @min_alloc_size - Indicates the minimum amount of space that the
8019 * allocator should try to satisfy. In some cases
8020 * @num_bytes may be larger than what is required and if
8021 * the filesystem is fragmented then allocation fails.
8022 * However, the presence of @min_alloc_size gives a
8023 * chance to try and satisfy the smaller allocation.
8025 * @empty_size - A hint that you plan on doing more COW. This is the
8026 * size in bytes the allocator should try to find free
8027 * next to the block it returns. This is just a hint and
8028 * may be ignored by the allocator.
8030 * @hint_byte - Hint to the allocator to start searching above the byte
8031 * address passed. It might be ignored.
8033 * @ins - This key is modified to record the found hole. It will
8034 * have the following values:
8035 * ins->objectid == start position
8036 * ins->flags = BTRFS_EXTENT_ITEM_KEY
8037 * ins->offset == the size of the hole.
8039 * @is_data - Boolean flag indicating whether an extent is
8040 * allocated for data (true) or metadata (false)
8042 * @delalloc - Boolean flag indicating whether this allocation is for
8043 * delalloc or not. If 'true' data_rwsem of block groups
8044 * is going to be acquired.
8047 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8048 * case -ENOSPC is returned then @ins->offset will contain the size of the
8049 * largest available hole the allocator managed to find.
8051 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8052 u64 num_bytes, u64 min_alloc_size,
8053 u64 empty_size, u64 hint_byte,
8054 struct btrfs_key *ins, int is_data, int delalloc)
8056 struct btrfs_fs_info *fs_info = root->fs_info;
8057 bool final_tried = num_bytes == min_alloc_size;
8061 flags = get_alloc_profile_by_root(root, is_data);
8063 WARN_ON(num_bytes < fs_info->sectorsize);
8064 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8065 hint_byte, ins, flags, delalloc);
8066 if (!ret && !is_data) {
8067 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8068 } else if (ret == -ENOSPC) {
8069 if (!final_tried && ins->offset) {
8070 num_bytes = min(num_bytes >> 1, ins->offset);
8071 num_bytes = round_down(num_bytes,
8072 fs_info->sectorsize);
8073 num_bytes = max(num_bytes, min_alloc_size);
8074 ram_bytes = num_bytes;
8075 if (num_bytes == min_alloc_size)
8078 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8079 struct btrfs_space_info *sinfo;
8081 sinfo = __find_space_info(fs_info, flags);
8083 "allocation failed flags %llu, wanted %llu",
8086 dump_space_info(fs_info, sinfo, num_bytes, 1);
8093 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8095 int pin, int delalloc)
8097 struct btrfs_block_group_cache *cache;
8100 cache = btrfs_lookup_block_group(fs_info, start);
8102 btrfs_err(fs_info, "Unable to find block group for %llu",
8108 pin_down_extent(fs_info, cache, start, len, 1);
8110 if (btrfs_test_opt(fs_info, DISCARD))
8111 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8112 btrfs_add_free_space(cache, start, len);
8113 btrfs_free_reserved_bytes(cache, len, delalloc);
8114 trace_btrfs_reserved_extent_free(fs_info, start, len);
8117 btrfs_put_block_group(cache);
8121 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8122 u64 start, u64 len, int delalloc)
8124 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8127 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8130 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8133 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8134 struct btrfs_fs_info *fs_info,
8135 u64 parent, u64 root_objectid,
8136 u64 flags, u64 owner, u64 offset,
8137 struct btrfs_key *ins, int ref_mod)
8140 struct btrfs_extent_item *extent_item;
8141 struct btrfs_extent_inline_ref *iref;
8142 struct btrfs_path *path;
8143 struct extent_buffer *leaf;
8148 type = BTRFS_SHARED_DATA_REF_KEY;
8150 type = BTRFS_EXTENT_DATA_REF_KEY;
8152 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8154 path = btrfs_alloc_path();
8158 path->leave_spinning = 1;
8159 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8162 btrfs_free_path(path);
8166 leaf = path->nodes[0];
8167 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8168 struct btrfs_extent_item);
8169 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8170 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8171 btrfs_set_extent_flags(leaf, extent_item,
8172 flags | BTRFS_EXTENT_FLAG_DATA);
8174 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8175 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8177 struct btrfs_shared_data_ref *ref;
8178 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8179 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8180 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8182 struct btrfs_extent_data_ref *ref;
8183 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8184 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8185 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8186 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8187 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8190 btrfs_mark_buffer_dirty(path->nodes[0]);
8191 btrfs_free_path(path);
8193 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8198 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8199 if (ret) { /* -ENOENT, logic error */
8200 btrfs_err(fs_info, "update block group failed for %llu %llu",
8201 ins->objectid, ins->offset);
8204 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8208 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8209 struct btrfs_fs_info *fs_info,
8210 u64 parent, u64 root_objectid,
8211 u64 flags, struct btrfs_disk_key *key,
8212 int level, struct btrfs_key *ins)
8215 struct btrfs_extent_item *extent_item;
8216 struct btrfs_tree_block_info *block_info;
8217 struct btrfs_extent_inline_ref *iref;
8218 struct btrfs_path *path;
8219 struct extent_buffer *leaf;
8220 u32 size = sizeof(*extent_item) + sizeof(*iref);
8221 u64 num_bytes = ins->offset;
8222 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8224 if (!skinny_metadata)
8225 size += sizeof(*block_info);
8227 path = btrfs_alloc_path();
8229 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8234 path->leave_spinning = 1;
8235 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8238 btrfs_free_path(path);
8239 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8244 leaf = path->nodes[0];
8245 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8246 struct btrfs_extent_item);
8247 btrfs_set_extent_refs(leaf, extent_item, 1);
8248 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8249 btrfs_set_extent_flags(leaf, extent_item,
8250 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8252 if (skinny_metadata) {
8253 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8254 num_bytes = fs_info->nodesize;
8256 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8257 btrfs_set_tree_block_key(leaf, block_info, key);
8258 btrfs_set_tree_block_level(leaf, block_info, level);
8259 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8263 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8264 btrfs_set_extent_inline_ref_type(leaf, iref,
8265 BTRFS_SHARED_BLOCK_REF_KEY);
8266 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8268 btrfs_set_extent_inline_ref_type(leaf, iref,
8269 BTRFS_TREE_BLOCK_REF_KEY);
8270 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8273 btrfs_mark_buffer_dirty(leaf);
8274 btrfs_free_path(path);
8276 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8281 ret = update_block_group(trans, fs_info, ins->objectid,
8282 fs_info->nodesize, 1);
8283 if (ret) { /* -ENOENT, logic error */
8284 btrfs_err(fs_info, "update block group failed for %llu %llu",
8285 ins->objectid, ins->offset);
8289 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8294 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8295 struct btrfs_root *root, u64 owner,
8296 u64 offset, u64 ram_bytes,
8297 struct btrfs_key *ins)
8299 struct btrfs_fs_info *fs_info = root->fs_info;
8302 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8304 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8305 root->root_key.objectid, owner, offset,
8306 BTRFS_ADD_DELAYED_EXTENT);
8308 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8310 root->root_key.objectid, owner,
8312 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8317 * this is used by the tree logging recovery code. It records that
8318 * an extent has been allocated and makes sure to clear the free
8319 * space cache bits as well
8321 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8322 struct btrfs_fs_info *fs_info,
8323 u64 root_objectid, u64 owner, u64 offset,
8324 struct btrfs_key *ins)
8327 struct btrfs_block_group_cache *block_group;
8328 struct btrfs_space_info *space_info;
8331 * Mixed block groups will exclude before processing the log so we only
8332 * need to do the exclude dance if this fs isn't mixed.
8334 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8335 ret = __exclude_logged_extent(fs_info, ins->objectid,
8341 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8345 space_info = block_group->space_info;
8346 spin_lock(&space_info->lock);
8347 spin_lock(&block_group->lock);
8348 space_info->bytes_reserved += ins->offset;
8349 block_group->reserved += ins->offset;
8350 spin_unlock(&block_group->lock);
8351 spin_unlock(&space_info->lock);
8353 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8354 0, owner, offset, ins, 1);
8355 btrfs_put_block_group(block_group);
8359 static struct extent_buffer *
8360 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8361 u64 bytenr, int level)
8363 struct btrfs_fs_info *fs_info = root->fs_info;
8364 struct extent_buffer *buf;
8366 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8370 btrfs_set_header_generation(buf, trans->transid);
8371 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8372 btrfs_tree_lock(buf);
8373 clean_tree_block(fs_info, buf);
8374 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8376 btrfs_set_lock_blocking(buf);
8377 set_extent_buffer_uptodate(buf);
8379 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8380 buf->log_index = root->log_transid % 2;
8382 * we allow two log transactions at a time, use different
8383 * EXENT bit to differentiate dirty pages.
8385 if (buf->log_index == 0)
8386 set_extent_dirty(&root->dirty_log_pages, buf->start,
8387 buf->start + buf->len - 1, GFP_NOFS);
8389 set_extent_new(&root->dirty_log_pages, buf->start,
8390 buf->start + buf->len - 1);
8392 buf->log_index = -1;
8393 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8394 buf->start + buf->len - 1, GFP_NOFS);
8396 trans->dirty = true;
8397 /* this returns a buffer locked for blocking */
8401 static struct btrfs_block_rsv *
8402 use_block_rsv(struct btrfs_trans_handle *trans,
8403 struct btrfs_root *root, u32 blocksize)
8405 struct btrfs_fs_info *fs_info = root->fs_info;
8406 struct btrfs_block_rsv *block_rsv;
8407 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8409 bool global_updated = false;
8411 block_rsv = get_block_rsv(trans, root);
8413 if (unlikely(block_rsv->size == 0))
8416 ret = block_rsv_use_bytes(block_rsv, blocksize);
8420 if (block_rsv->failfast)
8421 return ERR_PTR(ret);
8423 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8424 global_updated = true;
8425 update_global_block_rsv(fs_info);
8429 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8430 static DEFINE_RATELIMIT_STATE(_rs,
8431 DEFAULT_RATELIMIT_INTERVAL * 10,
8432 /*DEFAULT_RATELIMIT_BURST*/ 1);
8433 if (__ratelimit(&_rs))
8435 "BTRFS: block rsv returned %d\n", ret);
8438 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8439 BTRFS_RESERVE_NO_FLUSH);
8443 * If we couldn't reserve metadata bytes try and use some from
8444 * the global reserve if its space type is the same as the global
8447 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8448 block_rsv->space_info == global_rsv->space_info) {
8449 ret = block_rsv_use_bytes(global_rsv, blocksize);
8453 return ERR_PTR(ret);
8456 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8457 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8459 block_rsv_add_bytes(block_rsv, blocksize, 0);
8460 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL);
8464 * finds a free extent and does all the dirty work required for allocation
8465 * returns the tree buffer or an ERR_PTR on error.
8467 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8468 struct btrfs_root *root,
8469 u64 parent, u64 root_objectid,
8470 const struct btrfs_disk_key *key,
8471 int level, u64 hint,
8474 struct btrfs_fs_info *fs_info = root->fs_info;
8475 struct btrfs_key ins;
8476 struct btrfs_block_rsv *block_rsv;
8477 struct extent_buffer *buf;
8478 struct btrfs_delayed_extent_op *extent_op;
8481 u32 blocksize = fs_info->nodesize;
8482 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8484 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8485 if (btrfs_is_testing(fs_info)) {
8486 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8489 root->alloc_bytenr += blocksize;
8494 block_rsv = use_block_rsv(trans, root, blocksize);
8495 if (IS_ERR(block_rsv))
8496 return ERR_CAST(block_rsv);
8498 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8499 empty_size, hint, &ins, 0, 0);
8503 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8506 goto out_free_reserved;
8509 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8511 parent = ins.objectid;
8512 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8516 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8517 extent_op = btrfs_alloc_delayed_extent_op();
8523 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8525 memset(&extent_op->key, 0, sizeof(extent_op->key));
8526 extent_op->flags_to_set = flags;
8527 extent_op->update_key = skinny_metadata ? false : true;
8528 extent_op->update_flags = true;
8529 extent_op->is_data = false;
8530 extent_op->level = level;
8532 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8533 root_objectid, level, 0,
8534 BTRFS_ADD_DELAYED_EXTENT);
8535 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8537 root_objectid, level,
8538 BTRFS_ADD_DELAYED_EXTENT,
8539 extent_op, NULL, NULL);
8541 goto out_free_delayed;
8546 btrfs_free_delayed_extent_op(extent_op);
8548 free_extent_buffer(buf);
8550 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8552 unuse_block_rsv(fs_info, block_rsv, blocksize);
8553 return ERR_PTR(ret);
8556 struct walk_control {
8557 u64 refs[BTRFS_MAX_LEVEL];
8558 u64 flags[BTRFS_MAX_LEVEL];
8559 struct btrfs_key update_progress;
8570 #define DROP_REFERENCE 1
8571 #define UPDATE_BACKREF 2
8573 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8574 struct btrfs_root *root,
8575 struct walk_control *wc,
8576 struct btrfs_path *path)
8578 struct btrfs_fs_info *fs_info = root->fs_info;
8584 struct btrfs_key key;
8585 struct extent_buffer *eb;
8590 if (path->slots[wc->level] < wc->reada_slot) {
8591 wc->reada_count = wc->reada_count * 2 / 3;
8592 wc->reada_count = max(wc->reada_count, 2);
8594 wc->reada_count = wc->reada_count * 3 / 2;
8595 wc->reada_count = min_t(int, wc->reada_count,
8596 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8599 eb = path->nodes[wc->level];
8600 nritems = btrfs_header_nritems(eb);
8602 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8603 if (nread >= wc->reada_count)
8607 bytenr = btrfs_node_blockptr(eb, slot);
8608 generation = btrfs_node_ptr_generation(eb, slot);
8610 if (slot == path->slots[wc->level])
8613 if (wc->stage == UPDATE_BACKREF &&
8614 generation <= root->root_key.offset)
8617 /* We don't lock the tree block, it's OK to be racy here */
8618 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8619 wc->level - 1, 1, &refs,
8621 /* We don't care about errors in readahead. */
8626 if (wc->stage == DROP_REFERENCE) {
8630 if (wc->level == 1 &&
8631 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8633 if (!wc->update_ref ||
8634 generation <= root->root_key.offset)
8636 btrfs_node_key_to_cpu(eb, &key, slot);
8637 ret = btrfs_comp_cpu_keys(&key,
8638 &wc->update_progress);
8642 if (wc->level == 1 &&
8643 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8647 readahead_tree_block(fs_info, bytenr);
8650 wc->reada_slot = slot;
8654 * helper to process tree block while walking down the tree.
8656 * when wc->stage == UPDATE_BACKREF, this function updates
8657 * back refs for pointers in the block.
8659 * NOTE: return value 1 means we should stop walking down.
8661 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8662 struct btrfs_root *root,
8663 struct btrfs_path *path,
8664 struct walk_control *wc, int lookup_info)
8666 struct btrfs_fs_info *fs_info = root->fs_info;
8667 int level = wc->level;
8668 struct extent_buffer *eb = path->nodes[level];
8669 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8672 if (wc->stage == UPDATE_BACKREF &&
8673 btrfs_header_owner(eb) != root->root_key.objectid)
8677 * when reference count of tree block is 1, it won't increase
8678 * again. once full backref flag is set, we never clear it.
8681 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8682 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8683 BUG_ON(!path->locks[level]);
8684 ret = btrfs_lookup_extent_info(trans, fs_info,
8685 eb->start, level, 1,
8688 BUG_ON(ret == -ENOMEM);
8691 BUG_ON(wc->refs[level] == 0);
8694 if (wc->stage == DROP_REFERENCE) {
8695 if (wc->refs[level] > 1)
8698 if (path->locks[level] && !wc->keep_locks) {
8699 btrfs_tree_unlock_rw(eb, path->locks[level]);
8700 path->locks[level] = 0;
8705 /* wc->stage == UPDATE_BACKREF */
8706 if (!(wc->flags[level] & flag)) {
8707 BUG_ON(!path->locks[level]);
8708 ret = btrfs_inc_ref(trans, root, eb, 1);
8709 BUG_ON(ret); /* -ENOMEM */
8710 ret = btrfs_dec_ref(trans, root, eb, 0);
8711 BUG_ON(ret); /* -ENOMEM */
8712 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8714 btrfs_header_level(eb), 0);
8715 BUG_ON(ret); /* -ENOMEM */
8716 wc->flags[level] |= flag;
8720 * the block is shared by multiple trees, so it's not good to
8721 * keep the tree lock
8723 if (path->locks[level] && level > 0) {
8724 btrfs_tree_unlock_rw(eb, path->locks[level]);
8725 path->locks[level] = 0;
8731 * helper to process tree block pointer.
8733 * when wc->stage == DROP_REFERENCE, this function checks
8734 * reference count of the block pointed to. if the block
8735 * is shared and we need update back refs for the subtree
8736 * rooted at the block, this function changes wc->stage to
8737 * UPDATE_BACKREF. if the block is shared and there is no
8738 * need to update back, this function drops the reference
8741 * NOTE: return value 1 means we should stop walking down.
8743 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8744 struct btrfs_root *root,
8745 struct btrfs_path *path,
8746 struct walk_control *wc, int *lookup_info)
8748 struct btrfs_fs_info *fs_info = root->fs_info;
8753 struct btrfs_key key;
8754 struct btrfs_key first_key;
8755 struct extent_buffer *next;
8756 int level = wc->level;
8759 bool need_account = false;
8761 generation = btrfs_node_ptr_generation(path->nodes[level],
8762 path->slots[level]);
8764 * if the lower level block was created before the snapshot
8765 * was created, we know there is no need to update back refs
8768 if (wc->stage == UPDATE_BACKREF &&
8769 generation <= root->root_key.offset) {
8774 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8775 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8776 path->slots[level]);
8777 blocksize = fs_info->nodesize;
8779 next = find_extent_buffer(fs_info, bytenr);
8781 next = btrfs_find_create_tree_block(fs_info, bytenr);
8783 return PTR_ERR(next);
8785 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8789 btrfs_tree_lock(next);
8790 btrfs_set_lock_blocking(next);
8792 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8793 &wc->refs[level - 1],
8794 &wc->flags[level - 1]);
8798 if (unlikely(wc->refs[level - 1] == 0)) {
8799 btrfs_err(fs_info, "Missing references.");
8805 if (wc->stage == DROP_REFERENCE) {
8806 if (wc->refs[level - 1] > 1) {
8807 need_account = true;
8809 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8812 if (!wc->update_ref ||
8813 generation <= root->root_key.offset)
8816 btrfs_node_key_to_cpu(path->nodes[level], &key,
8817 path->slots[level]);
8818 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8822 wc->stage = UPDATE_BACKREF;
8823 wc->shared_level = level - 1;
8827 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8831 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8832 btrfs_tree_unlock(next);
8833 free_extent_buffer(next);
8839 if (reada && level == 1)
8840 reada_walk_down(trans, root, wc, path);
8841 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8844 return PTR_ERR(next);
8845 } else if (!extent_buffer_uptodate(next)) {
8846 free_extent_buffer(next);
8849 btrfs_tree_lock(next);
8850 btrfs_set_lock_blocking(next);
8854 ASSERT(level == btrfs_header_level(next));
8855 if (level != btrfs_header_level(next)) {
8856 btrfs_err(root->fs_info, "mismatched level");
8860 path->nodes[level] = next;
8861 path->slots[level] = 0;
8862 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8868 wc->refs[level - 1] = 0;
8869 wc->flags[level - 1] = 0;
8870 if (wc->stage == DROP_REFERENCE) {
8871 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8872 parent = path->nodes[level]->start;
8874 ASSERT(root->root_key.objectid ==
8875 btrfs_header_owner(path->nodes[level]));
8876 if (root->root_key.objectid !=
8877 btrfs_header_owner(path->nodes[level])) {
8878 btrfs_err(root->fs_info,
8879 "mismatched block owner");
8887 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8888 generation, level - 1);
8890 btrfs_err_rl(fs_info,
8891 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8895 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8896 parent, root->root_key.objectid,
8906 btrfs_tree_unlock(next);
8907 free_extent_buffer(next);
8913 * helper to process tree block while walking up the tree.
8915 * when wc->stage == DROP_REFERENCE, this function drops
8916 * reference count on the block.
8918 * when wc->stage == UPDATE_BACKREF, this function changes
8919 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8920 * to UPDATE_BACKREF previously while processing the block.
8922 * NOTE: return value 1 means we should stop walking up.
8924 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8925 struct btrfs_root *root,
8926 struct btrfs_path *path,
8927 struct walk_control *wc)
8929 struct btrfs_fs_info *fs_info = root->fs_info;
8931 int level = wc->level;
8932 struct extent_buffer *eb = path->nodes[level];
8935 if (wc->stage == UPDATE_BACKREF) {
8936 BUG_ON(wc->shared_level < level);
8937 if (level < wc->shared_level)
8940 ret = find_next_key(path, level + 1, &wc->update_progress);
8944 wc->stage = DROP_REFERENCE;
8945 wc->shared_level = -1;
8946 path->slots[level] = 0;
8949 * check reference count again if the block isn't locked.
8950 * we should start walking down the tree again if reference
8953 if (!path->locks[level]) {
8955 btrfs_tree_lock(eb);
8956 btrfs_set_lock_blocking(eb);
8957 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8959 ret = btrfs_lookup_extent_info(trans, fs_info,
8960 eb->start, level, 1,
8964 btrfs_tree_unlock_rw(eb, path->locks[level]);
8965 path->locks[level] = 0;
8968 BUG_ON(wc->refs[level] == 0);
8969 if (wc->refs[level] == 1) {
8970 btrfs_tree_unlock_rw(eb, path->locks[level]);
8971 path->locks[level] = 0;
8977 /* wc->stage == DROP_REFERENCE */
8978 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8980 if (wc->refs[level] == 1) {
8982 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8983 ret = btrfs_dec_ref(trans, root, eb, 1);
8985 ret = btrfs_dec_ref(trans, root, eb, 0);
8986 BUG_ON(ret); /* -ENOMEM */
8987 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8989 btrfs_err_rl(fs_info,
8990 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8994 /* make block locked assertion in clean_tree_block happy */
8995 if (!path->locks[level] &&
8996 btrfs_header_generation(eb) == trans->transid) {
8997 btrfs_tree_lock(eb);
8998 btrfs_set_lock_blocking(eb);
8999 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9001 clean_tree_block(fs_info, eb);
9004 if (eb == root->node) {
9005 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9008 BUG_ON(root->root_key.objectid !=
9009 btrfs_header_owner(eb));
9011 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9012 parent = path->nodes[level + 1]->start;
9014 BUG_ON(root->root_key.objectid !=
9015 btrfs_header_owner(path->nodes[level + 1]));
9018 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9020 wc->refs[level] = 0;
9021 wc->flags[level] = 0;
9025 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9026 struct btrfs_root *root,
9027 struct btrfs_path *path,
9028 struct walk_control *wc)
9030 int level = wc->level;
9031 int lookup_info = 1;
9034 while (level >= 0) {
9035 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9042 if (path->slots[level] >=
9043 btrfs_header_nritems(path->nodes[level]))
9046 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9048 path->slots[level]++;
9057 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9058 struct btrfs_root *root,
9059 struct btrfs_path *path,
9060 struct walk_control *wc, int max_level)
9062 int level = wc->level;
9065 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9066 while (level < max_level && path->nodes[level]) {
9068 if (path->slots[level] + 1 <
9069 btrfs_header_nritems(path->nodes[level])) {
9070 path->slots[level]++;
9073 ret = walk_up_proc(trans, root, path, wc);
9077 if (path->locks[level]) {
9078 btrfs_tree_unlock_rw(path->nodes[level],
9079 path->locks[level]);
9080 path->locks[level] = 0;
9082 free_extent_buffer(path->nodes[level]);
9083 path->nodes[level] = NULL;
9091 * drop a subvolume tree.
9093 * this function traverses the tree freeing any blocks that only
9094 * referenced by the tree.
9096 * when a shared tree block is found. this function decreases its
9097 * reference count by one. if update_ref is true, this function
9098 * also make sure backrefs for the shared block and all lower level
9099 * blocks are properly updated.
9101 * If called with for_reloc == 0, may exit early with -EAGAIN
9103 int btrfs_drop_snapshot(struct btrfs_root *root,
9104 struct btrfs_block_rsv *block_rsv, int update_ref,
9107 struct btrfs_fs_info *fs_info = root->fs_info;
9108 struct btrfs_path *path;
9109 struct btrfs_trans_handle *trans;
9110 struct btrfs_root *tree_root = fs_info->tree_root;
9111 struct btrfs_root_item *root_item = &root->root_item;
9112 struct walk_control *wc;
9113 struct btrfs_key key;
9117 bool root_dropped = false;
9119 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9121 path = btrfs_alloc_path();
9127 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9129 btrfs_free_path(path);
9134 trans = btrfs_start_transaction(tree_root, 0);
9135 if (IS_ERR(trans)) {
9136 err = PTR_ERR(trans);
9141 trans->block_rsv = block_rsv;
9143 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9144 level = btrfs_header_level(root->node);
9145 path->nodes[level] = btrfs_lock_root_node(root);
9146 btrfs_set_lock_blocking(path->nodes[level]);
9147 path->slots[level] = 0;
9148 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9149 memset(&wc->update_progress, 0,
9150 sizeof(wc->update_progress));
9152 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9153 memcpy(&wc->update_progress, &key,
9154 sizeof(wc->update_progress));
9156 level = root_item->drop_level;
9158 path->lowest_level = level;
9159 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9160 path->lowest_level = 0;
9168 * unlock our path, this is safe because only this
9169 * function is allowed to delete this snapshot
9171 btrfs_unlock_up_safe(path, 0);
9173 level = btrfs_header_level(root->node);
9175 btrfs_tree_lock(path->nodes[level]);
9176 btrfs_set_lock_blocking(path->nodes[level]);
9177 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9179 ret = btrfs_lookup_extent_info(trans, fs_info,
9180 path->nodes[level]->start,
9181 level, 1, &wc->refs[level],
9187 BUG_ON(wc->refs[level] == 0);
9189 if (level == root_item->drop_level)
9192 btrfs_tree_unlock(path->nodes[level]);
9193 path->locks[level] = 0;
9194 WARN_ON(wc->refs[level] != 1);
9200 wc->shared_level = -1;
9201 wc->stage = DROP_REFERENCE;
9202 wc->update_ref = update_ref;
9204 wc->for_reloc = for_reloc;
9205 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9209 ret = walk_down_tree(trans, root, path, wc);
9215 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9222 BUG_ON(wc->stage != DROP_REFERENCE);
9226 if (wc->stage == DROP_REFERENCE) {
9228 btrfs_node_key(path->nodes[level],
9229 &root_item->drop_progress,
9230 path->slots[level]);
9231 root_item->drop_level = level;
9234 BUG_ON(wc->level == 0);
9235 if (btrfs_should_end_transaction(trans) ||
9236 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9237 ret = btrfs_update_root(trans, tree_root,
9241 btrfs_abort_transaction(trans, ret);
9246 btrfs_end_transaction_throttle(trans);
9247 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9248 btrfs_debug(fs_info,
9249 "drop snapshot early exit");
9254 trans = btrfs_start_transaction(tree_root, 0);
9255 if (IS_ERR(trans)) {
9256 err = PTR_ERR(trans);
9260 trans->block_rsv = block_rsv;
9263 btrfs_release_path(path);
9267 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9269 btrfs_abort_transaction(trans, ret);
9274 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9275 ret = btrfs_find_root(tree_root, &root->root_key, path,
9278 btrfs_abort_transaction(trans, ret);
9281 } else if (ret > 0) {
9282 /* if we fail to delete the orphan item this time
9283 * around, it'll get picked up the next time.
9285 * The most common failure here is just -ENOENT.
9287 btrfs_del_orphan_item(trans, tree_root,
9288 root->root_key.objectid);
9292 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9293 btrfs_add_dropped_root(trans, root);
9295 free_extent_buffer(root->node);
9296 free_extent_buffer(root->commit_root);
9297 btrfs_put_fs_root(root);
9299 root_dropped = true;
9301 btrfs_end_transaction_throttle(trans);
9304 btrfs_free_path(path);
9307 * So if we need to stop dropping the snapshot for whatever reason we
9308 * need to make sure to add it back to the dead root list so that we
9309 * keep trying to do the work later. This also cleans up roots if we
9310 * don't have it in the radix (like when we recover after a power fail
9311 * or unmount) so we don't leak memory.
9313 if (!for_reloc && !root_dropped)
9314 btrfs_add_dead_root(root);
9315 if (err && err != -EAGAIN)
9316 btrfs_handle_fs_error(fs_info, err, NULL);
9321 * drop subtree rooted at tree block 'node'.
9323 * NOTE: this function will unlock and release tree block 'node'
9324 * only used by relocation code
9326 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9327 struct btrfs_root *root,
9328 struct extent_buffer *node,
9329 struct extent_buffer *parent)
9331 struct btrfs_fs_info *fs_info = root->fs_info;
9332 struct btrfs_path *path;
9333 struct walk_control *wc;
9339 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9341 path = btrfs_alloc_path();
9345 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9347 btrfs_free_path(path);
9351 btrfs_assert_tree_locked(parent);
9352 parent_level = btrfs_header_level(parent);
9353 extent_buffer_get(parent);
9354 path->nodes[parent_level] = parent;
9355 path->slots[parent_level] = btrfs_header_nritems(parent);
9357 btrfs_assert_tree_locked(node);
9358 level = btrfs_header_level(node);
9359 path->nodes[level] = node;
9360 path->slots[level] = 0;
9361 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9363 wc->refs[parent_level] = 1;
9364 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9366 wc->shared_level = -1;
9367 wc->stage = DROP_REFERENCE;
9371 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9374 wret = walk_down_tree(trans, root, path, wc);
9380 wret = walk_up_tree(trans, root, path, wc, parent_level);
9388 btrfs_free_path(path);
9392 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9398 * if restripe for this chunk_type is on pick target profile and
9399 * return, otherwise do the usual balance
9401 stripped = get_restripe_target(fs_info, flags);
9403 return extended_to_chunk(stripped);
9405 num_devices = fs_info->fs_devices->rw_devices;
9407 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9408 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9409 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9411 if (num_devices == 1) {
9412 stripped |= BTRFS_BLOCK_GROUP_DUP;
9413 stripped = flags & ~stripped;
9415 /* turn raid0 into single device chunks */
9416 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9419 /* turn mirroring into duplication */
9420 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9421 BTRFS_BLOCK_GROUP_RAID10))
9422 return stripped | BTRFS_BLOCK_GROUP_DUP;
9424 /* they already had raid on here, just return */
9425 if (flags & stripped)
9428 stripped |= BTRFS_BLOCK_GROUP_DUP;
9429 stripped = flags & ~stripped;
9431 /* switch duplicated blocks with raid1 */
9432 if (flags & BTRFS_BLOCK_GROUP_DUP)
9433 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9435 /* this is drive concat, leave it alone */
9441 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9443 struct btrfs_space_info *sinfo = cache->space_info;
9445 u64 min_allocable_bytes;
9449 * We need some metadata space and system metadata space for
9450 * allocating chunks in some corner cases until we force to set
9451 * it to be readonly.
9454 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9456 min_allocable_bytes = SZ_1M;
9458 min_allocable_bytes = 0;
9460 spin_lock(&sinfo->lock);
9461 spin_lock(&cache->lock);
9469 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9470 cache->bytes_super - btrfs_block_group_used(&cache->item);
9472 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9473 min_allocable_bytes <= sinfo->total_bytes) {
9474 sinfo->bytes_readonly += num_bytes;
9476 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9480 spin_unlock(&cache->lock);
9481 spin_unlock(&sinfo->lock);
9485 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9486 struct btrfs_block_group_cache *cache)
9489 struct btrfs_trans_handle *trans;
9494 trans = btrfs_join_transaction(fs_info->extent_root);
9496 return PTR_ERR(trans);
9499 * we're not allowed to set block groups readonly after the dirty
9500 * block groups cache has started writing. If it already started,
9501 * back off and let this transaction commit
9503 mutex_lock(&fs_info->ro_block_group_mutex);
9504 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9505 u64 transid = trans->transid;
9507 mutex_unlock(&fs_info->ro_block_group_mutex);
9508 btrfs_end_transaction(trans);
9510 ret = btrfs_wait_for_commit(fs_info, transid);
9517 * if we are changing raid levels, try to allocate a corresponding
9518 * block group with the new raid level.
9520 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9521 if (alloc_flags != cache->flags) {
9522 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9525 * ENOSPC is allowed here, we may have enough space
9526 * already allocated at the new raid level to
9535 ret = inc_block_group_ro(cache, 0);
9538 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9539 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9543 ret = inc_block_group_ro(cache, 0);
9545 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9546 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9547 mutex_lock(&fs_info->chunk_mutex);
9548 check_system_chunk(trans, fs_info, alloc_flags);
9549 mutex_unlock(&fs_info->chunk_mutex);
9551 mutex_unlock(&fs_info->ro_block_group_mutex);
9553 btrfs_end_transaction(trans);
9557 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9558 struct btrfs_fs_info *fs_info, u64 type)
9560 u64 alloc_flags = get_alloc_profile(fs_info, type);
9562 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9566 * helper to account the unused space of all the readonly block group in the
9567 * space_info. takes mirrors into account.
9569 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9571 struct btrfs_block_group_cache *block_group;
9575 /* It's df, we don't care if it's racy */
9576 if (list_empty(&sinfo->ro_bgs))
9579 spin_lock(&sinfo->lock);
9580 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9581 spin_lock(&block_group->lock);
9583 if (!block_group->ro) {
9584 spin_unlock(&block_group->lock);
9588 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9589 BTRFS_BLOCK_GROUP_RAID10 |
9590 BTRFS_BLOCK_GROUP_DUP))
9595 free_bytes += (block_group->key.offset -
9596 btrfs_block_group_used(&block_group->item)) *
9599 spin_unlock(&block_group->lock);
9601 spin_unlock(&sinfo->lock);
9606 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9608 struct btrfs_space_info *sinfo = cache->space_info;
9613 spin_lock(&sinfo->lock);
9614 spin_lock(&cache->lock);
9616 num_bytes = cache->key.offset - cache->reserved -
9617 cache->pinned - cache->bytes_super -
9618 btrfs_block_group_used(&cache->item);
9619 sinfo->bytes_readonly -= num_bytes;
9620 list_del_init(&cache->ro_list);
9622 spin_unlock(&cache->lock);
9623 spin_unlock(&sinfo->lock);
9627 * checks to see if its even possible to relocate this block group.
9629 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9630 * ok to go ahead and try.
9632 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9634 struct btrfs_root *root = fs_info->extent_root;
9635 struct btrfs_block_group_cache *block_group;
9636 struct btrfs_space_info *space_info;
9637 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9638 struct btrfs_device *device;
9639 struct btrfs_trans_handle *trans;
9649 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9651 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9653 /* odd, couldn't find the block group, leave it alone */
9657 "can't find block group for bytenr %llu",
9662 min_free = btrfs_block_group_used(&block_group->item);
9664 /* no bytes used, we're good */
9668 space_info = block_group->space_info;
9669 spin_lock(&space_info->lock);
9671 full = space_info->full;
9674 * if this is the last block group we have in this space, we can't
9675 * relocate it unless we're able to allocate a new chunk below.
9677 * Otherwise, we need to make sure we have room in the space to handle
9678 * all of the extents from this block group. If we can, we're good
9680 if ((space_info->total_bytes != block_group->key.offset) &&
9681 (btrfs_space_info_used(space_info, false) + min_free <
9682 space_info->total_bytes)) {
9683 spin_unlock(&space_info->lock);
9686 spin_unlock(&space_info->lock);
9689 * ok we don't have enough space, but maybe we have free space on our
9690 * devices to allocate new chunks for relocation, so loop through our
9691 * alloc devices and guess if we have enough space. if this block
9692 * group is going to be restriped, run checks against the target
9693 * profile instead of the current one.
9705 target = get_restripe_target(fs_info, block_group->flags);
9707 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9710 * this is just a balance, so if we were marked as full
9711 * we know there is no space for a new chunk
9716 "no space to alloc new chunk for block group %llu",
9717 block_group->key.objectid);
9721 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9724 if (index == BTRFS_RAID_RAID10) {
9728 } else if (index == BTRFS_RAID_RAID1) {
9730 } else if (index == BTRFS_RAID_DUP) {
9733 } else if (index == BTRFS_RAID_RAID0) {
9734 dev_min = fs_devices->rw_devices;
9735 min_free = div64_u64(min_free, dev_min);
9738 /* We need to do this so that we can look at pending chunks */
9739 trans = btrfs_join_transaction(root);
9740 if (IS_ERR(trans)) {
9741 ret = PTR_ERR(trans);
9745 mutex_lock(&fs_info->chunk_mutex);
9746 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9750 * check to make sure we can actually find a chunk with enough
9751 * space to fit our block group in.
9753 if (device->total_bytes > device->bytes_used + min_free &&
9754 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9755 ret = find_free_dev_extent(trans, device, min_free,
9760 if (dev_nr >= dev_min)
9766 if (debug && ret == -1)
9768 "no space to allocate a new chunk for block group %llu",
9769 block_group->key.objectid);
9770 mutex_unlock(&fs_info->chunk_mutex);
9771 btrfs_end_transaction(trans);
9773 btrfs_put_block_group(block_group);
9777 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9778 struct btrfs_path *path,
9779 struct btrfs_key *key)
9781 struct btrfs_root *root = fs_info->extent_root;
9783 struct btrfs_key found_key;
9784 struct extent_buffer *leaf;
9787 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9792 slot = path->slots[0];
9793 leaf = path->nodes[0];
9794 if (slot >= btrfs_header_nritems(leaf)) {
9795 ret = btrfs_next_leaf(root, path);
9802 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9804 if (found_key.objectid >= key->objectid &&
9805 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9806 struct extent_map_tree *em_tree;
9807 struct extent_map *em;
9809 em_tree = &root->fs_info->mapping_tree.map_tree;
9810 read_lock(&em_tree->lock);
9811 em = lookup_extent_mapping(em_tree, found_key.objectid,
9813 read_unlock(&em_tree->lock);
9816 "logical %llu len %llu found bg but no related chunk",
9817 found_key.objectid, found_key.offset);
9822 free_extent_map(em);
9831 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9833 struct btrfs_block_group_cache *block_group;
9837 struct inode *inode;
9839 block_group = btrfs_lookup_first_block_group(info, last);
9840 while (block_group) {
9841 spin_lock(&block_group->lock);
9842 if (block_group->iref)
9844 spin_unlock(&block_group->lock);
9845 block_group = next_block_group(info, block_group);
9854 inode = block_group->inode;
9855 block_group->iref = 0;
9856 block_group->inode = NULL;
9857 spin_unlock(&block_group->lock);
9858 ASSERT(block_group->io_ctl.inode == NULL);
9860 last = block_group->key.objectid + block_group->key.offset;
9861 btrfs_put_block_group(block_group);
9866 * Must be called only after stopping all workers, since we could have block
9867 * group caching kthreads running, and therefore they could race with us if we
9868 * freed the block groups before stopping them.
9870 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9872 struct btrfs_block_group_cache *block_group;
9873 struct btrfs_space_info *space_info;
9874 struct btrfs_caching_control *caching_ctl;
9877 down_write(&info->commit_root_sem);
9878 while (!list_empty(&info->caching_block_groups)) {
9879 caching_ctl = list_entry(info->caching_block_groups.next,
9880 struct btrfs_caching_control, list);
9881 list_del(&caching_ctl->list);
9882 put_caching_control(caching_ctl);
9884 up_write(&info->commit_root_sem);
9886 spin_lock(&info->unused_bgs_lock);
9887 while (!list_empty(&info->unused_bgs)) {
9888 block_group = list_first_entry(&info->unused_bgs,
9889 struct btrfs_block_group_cache,
9891 list_del_init(&block_group->bg_list);
9892 btrfs_put_block_group(block_group);
9894 spin_unlock(&info->unused_bgs_lock);
9896 spin_lock(&info->block_group_cache_lock);
9897 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9898 block_group = rb_entry(n, struct btrfs_block_group_cache,
9900 rb_erase(&block_group->cache_node,
9901 &info->block_group_cache_tree);
9902 RB_CLEAR_NODE(&block_group->cache_node);
9903 spin_unlock(&info->block_group_cache_lock);
9905 down_write(&block_group->space_info->groups_sem);
9906 list_del(&block_group->list);
9907 up_write(&block_group->space_info->groups_sem);
9910 * We haven't cached this block group, which means we could
9911 * possibly have excluded extents on this block group.
9913 if (block_group->cached == BTRFS_CACHE_NO ||
9914 block_group->cached == BTRFS_CACHE_ERROR)
9915 free_excluded_extents(info, block_group);
9917 btrfs_remove_free_space_cache(block_group);
9918 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9919 ASSERT(list_empty(&block_group->dirty_list));
9920 ASSERT(list_empty(&block_group->io_list));
9921 ASSERT(list_empty(&block_group->bg_list));
9922 ASSERT(atomic_read(&block_group->count) == 1);
9923 btrfs_put_block_group(block_group);
9925 spin_lock(&info->block_group_cache_lock);
9927 spin_unlock(&info->block_group_cache_lock);
9929 /* now that all the block groups are freed, go through and
9930 * free all the space_info structs. This is only called during
9931 * the final stages of unmount, and so we know nobody is
9932 * using them. We call synchronize_rcu() once before we start,
9933 * just to be on the safe side.
9937 release_global_block_rsv(info);
9939 while (!list_empty(&info->space_info)) {
9942 space_info = list_entry(info->space_info.next,
9943 struct btrfs_space_info,
9947 * Do not hide this behind enospc_debug, this is actually
9948 * important and indicates a real bug if this happens.
9950 if (WARN_ON(space_info->bytes_pinned > 0 ||
9951 space_info->bytes_reserved > 0 ||
9952 space_info->bytes_may_use > 0))
9953 dump_space_info(info, space_info, 0, 0);
9954 list_del(&space_info->list);
9955 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9956 struct kobject *kobj;
9957 kobj = space_info->block_group_kobjs[i];
9958 space_info->block_group_kobjs[i] = NULL;
9964 kobject_del(&space_info->kobj);
9965 kobject_put(&space_info->kobj);
9970 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9971 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9973 struct btrfs_space_info *space_info;
9974 struct raid_kobject *rkobj;
9979 spin_lock(&fs_info->pending_raid_kobjs_lock);
9980 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9981 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9983 list_for_each_entry(rkobj, &list, list) {
9984 space_info = __find_space_info(fs_info, rkobj->flags);
9985 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9987 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9988 "%s", get_raid_name(index));
9990 kobject_put(&rkobj->kobj);
9996 "failed to add kobject for block cache, ignoring");
9999 static void link_block_group(struct btrfs_block_group_cache *cache)
10001 struct btrfs_space_info *space_info = cache->space_info;
10002 struct btrfs_fs_info *fs_info = cache->fs_info;
10003 int index = btrfs_bg_flags_to_raid_index(cache->flags);
10004 bool first = false;
10006 down_write(&space_info->groups_sem);
10007 if (list_empty(&space_info->block_groups[index]))
10009 list_add_tail(&cache->list, &space_info->block_groups[index]);
10010 up_write(&space_info->groups_sem);
10013 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10015 btrfs_warn(cache->fs_info,
10016 "couldn't alloc memory for raid level kobject");
10019 rkobj->flags = cache->flags;
10020 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10022 spin_lock(&fs_info->pending_raid_kobjs_lock);
10023 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
10024 spin_unlock(&fs_info->pending_raid_kobjs_lock);
10025 space_info->block_group_kobjs[index] = &rkobj->kobj;
10029 static struct btrfs_block_group_cache *
10030 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10031 u64 start, u64 size)
10033 struct btrfs_block_group_cache *cache;
10035 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10039 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10041 if (!cache->free_space_ctl) {
10046 cache->key.objectid = start;
10047 cache->key.offset = size;
10048 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10050 cache->fs_info = fs_info;
10051 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10052 set_free_space_tree_thresholds(cache);
10054 atomic_set(&cache->count, 1);
10055 spin_lock_init(&cache->lock);
10056 init_rwsem(&cache->data_rwsem);
10057 INIT_LIST_HEAD(&cache->list);
10058 INIT_LIST_HEAD(&cache->cluster_list);
10059 INIT_LIST_HEAD(&cache->bg_list);
10060 INIT_LIST_HEAD(&cache->ro_list);
10061 INIT_LIST_HEAD(&cache->dirty_list);
10062 INIT_LIST_HEAD(&cache->io_list);
10063 btrfs_init_free_space_ctl(cache);
10064 atomic_set(&cache->trimming, 0);
10065 mutex_init(&cache->free_space_lock);
10066 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10071 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10073 struct btrfs_path *path;
10075 struct btrfs_block_group_cache *cache;
10076 struct btrfs_space_info *space_info;
10077 struct btrfs_key key;
10078 struct btrfs_key found_key;
10079 struct extent_buffer *leaf;
10080 int need_clear = 0;
10085 feature = btrfs_super_incompat_flags(info->super_copy);
10086 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10090 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10091 path = btrfs_alloc_path();
10094 path->reada = READA_FORWARD;
10096 cache_gen = btrfs_super_cache_generation(info->super_copy);
10097 if (btrfs_test_opt(info, SPACE_CACHE) &&
10098 btrfs_super_generation(info->super_copy) != cache_gen)
10100 if (btrfs_test_opt(info, CLEAR_CACHE))
10104 ret = find_first_block_group(info, path, &key);
10110 leaf = path->nodes[0];
10111 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10113 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10122 * When we mount with old space cache, we need to
10123 * set BTRFS_DC_CLEAR and set dirty flag.
10125 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10126 * truncate the old free space cache inode and
10128 * b) Setting 'dirty flag' makes sure that we flush
10129 * the new space cache info onto disk.
10131 if (btrfs_test_opt(info, SPACE_CACHE))
10132 cache->disk_cache_state = BTRFS_DC_CLEAR;
10135 read_extent_buffer(leaf, &cache->item,
10136 btrfs_item_ptr_offset(leaf, path->slots[0]),
10137 sizeof(cache->item));
10138 cache->flags = btrfs_block_group_flags(&cache->item);
10140 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10141 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10143 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10144 cache->key.objectid);
10149 key.objectid = found_key.objectid + found_key.offset;
10150 btrfs_release_path(path);
10153 * We need to exclude the super stripes now so that the space
10154 * info has super bytes accounted for, otherwise we'll think
10155 * we have more space than we actually do.
10157 ret = exclude_super_stripes(info, cache);
10160 * We may have excluded something, so call this just in
10163 free_excluded_extents(info, cache);
10164 btrfs_put_block_group(cache);
10169 * check for two cases, either we are full, and therefore
10170 * don't need to bother with the caching work since we won't
10171 * find any space, or we are empty, and we can just add all
10172 * the space in and be done with it. This saves us _alot_ of
10173 * time, particularly in the full case.
10175 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10176 cache->last_byte_to_unpin = (u64)-1;
10177 cache->cached = BTRFS_CACHE_FINISHED;
10178 free_excluded_extents(info, cache);
10179 } else if (btrfs_block_group_used(&cache->item) == 0) {
10180 cache->last_byte_to_unpin = (u64)-1;
10181 cache->cached = BTRFS_CACHE_FINISHED;
10182 add_new_free_space(cache, info,
10183 found_key.objectid,
10184 found_key.objectid +
10186 free_excluded_extents(info, cache);
10189 ret = btrfs_add_block_group_cache(info, cache);
10191 btrfs_remove_free_space_cache(cache);
10192 btrfs_put_block_group(cache);
10196 trace_btrfs_add_block_group(info, cache, 0);
10197 update_space_info(info, cache->flags, found_key.offset,
10198 btrfs_block_group_used(&cache->item),
10199 cache->bytes_super, &space_info);
10201 cache->space_info = space_info;
10203 link_block_group(cache);
10205 set_avail_alloc_bits(info, cache->flags);
10206 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10207 inc_block_group_ro(cache, 1);
10208 } else if (btrfs_block_group_used(&cache->item) == 0) {
10209 spin_lock(&info->unused_bgs_lock);
10210 /* Should always be true but just in case. */
10211 if (list_empty(&cache->bg_list)) {
10212 btrfs_get_block_group(cache);
10213 list_add_tail(&cache->bg_list,
10214 &info->unused_bgs);
10216 spin_unlock(&info->unused_bgs_lock);
10220 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10221 if (!(get_alloc_profile(info, space_info->flags) &
10222 (BTRFS_BLOCK_GROUP_RAID10 |
10223 BTRFS_BLOCK_GROUP_RAID1 |
10224 BTRFS_BLOCK_GROUP_RAID5 |
10225 BTRFS_BLOCK_GROUP_RAID6 |
10226 BTRFS_BLOCK_GROUP_DUP)))
10229 * avoid allocating from un-mirrored block group if there are
10230 * mirrored block groups.
10232 list_for_each_entry(cache,
10233 &space_info->block_groups[BTRFS_RAID_RAID0],
10235 inc_block_group_ro(cache, 1);
10236 list_for_each_entry(cache,
10237 &space_info->block_groups[BTRFS_RAID_SINGLE],
10239 inc_block_group_ro(cache, 1);
10242 btrfs_add_raid_kobjects(info);
10243 init_global_block_rsv(info);
10246 btrfs_free_path(path);
10250 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10252 struct btrfs_fs_info *fs_info = trans->fs_info;
10253 struct btrfs_block_group_cache *block_group, *tmp;
10254 struct btrfs_root *extent_root = fs_info->extent_root;
10255 struct btrfs_block_group_item item;
10256 struct btrfs_key key;
10258 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10260 trans->can_flush_pending_bgs = false;
10261 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10265 spin_lock(&block_group->lock);
10266 memcpy(&item, &block_group->item, sizeof(item));
10267 memcpy(&key, &block_group->key, sizeof(key));
10268 spin_unlock(&block_group->lock);
10270 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10273 btrfs_abort_transaction(trans, ret);
10274 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10277 btrfs_abort_transaction(trans, ret);
10278 add_block_group_free_space(trans, fs_info, block_group);
10279 /* already aborted the transaction if it failed. */
10281 list_del_init(&block_group->bg_list);
10283 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10286 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10287 struct btrfs_fs_info *fs_info, u64 bytes_used,
10288 u64 type, u64 chunk_offset, u64 size)
10290 struct btrfs_block_group_cache *cache;
10293 btrfs_set_log_full_commit(fs_info, trans);
10295 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10299 btrfs_set_block_group_used(&cache->item, bytes_used);
10300 btrfs_set_block_group_chunk_objectid(&cache->item,
10301 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10302 btrfs_set_block_group_flags(&cache->item, type);
10304 cache->flags = type;
10305 cache->last_byte_to_unpin = (u64)-1;
10306 cache->cached = BTRFS_CACHE_FINISHED;
10307 cache->needs_free_space = 1;
10308 ret = exclude_super_stripes(fs_info, cache);
10311 * We may have excluded something, so call this just in
10314 free_excluded_extents(fs_info, cache);
10315 btrfs_put_block_group(cache);
10319 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10321 free_excluded_extents(fs_info, cache);
10323 #ifdef CONFIG_BTRFS_DEBUG
10324 if (btrfs_should_fragment_free_space(cache)) {
10325 u64 new_bytes_used = size - bytes_used;
10327 bytes_used += new_bytes_used >> 1;
10328 fragment_free_space(cache);
10332 * Ensure the corresponding space_info object is created and
10333 * assigned to our block group. We want our bg to be added to the rbtree
10334 * with its ->space_info set.
10336 cache->space_info = __find_space_info(fs_info, cache->flags);
10337 ASSERT(cache->space_info);
10339 ret = btrfs_add_block_group_cache(fs_info, cache);
10341 btrfs_remove_free_space_cache(cache);
10342 btrfs_put_block_group(cache);
10347 * Now that our block group has its ->space_info set and is inserted in
10348 * the rbtree, update the space info's counters.
10350 trace_btrfs_add_block_group(fs_info, cache, 1);
10351 update_space_info(fs_info, cache->flags, size, bytes_used,
10352 cache->bytes_super, &cache->space_info);
10353 update_global_block_rsv(fs_info);
10355 link_block_group(cache);
10357 list_add_tail(&cache->bg_list, &trans->new_bgs);
10359 set_avail_alloc_bits(fs_info, type);
10363 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10365 u64 extra_flags = chunk_to_extended(flags) &
10366 BTRFS_EXTENDED_PROFILE_MASK;
10368 write_seqlock(&fs_info->profiles_lock);
10369 if (flags & BTRFS_BLOCK_GROUP_DATA)
10370 fs_info->avail_data_alloc_bits &= ~extra_flags;
10371 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10372 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10373 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10374 fs_info->avail_system_alloc_bits &= ~extra_flags;
10375 write_sequnlock(&fs_info->profiles_lock);
10378 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10379 struct btrfs_fs_info *fs_info, u64 group_start,
10380 struct extent_map *em)
10382 struct btrfs_root *root = fs_info->extent_root;
10383 struct btrfs_path *path;
10384 struct btrfs_block_group_cache *block_group;
10385 struct btrfs_free_cluster *cluster;
10386 struct btrfs_root *tree_root = fs_info->tree_root;
10387 struct btrfs_key key;
10388 struct inode *inode;
10389 struct kobject *kobj = NULL;
10393 struct btrfs_caching_control *caching_ctl = NULL;
10396 block_group = btrfs_lookup_block_group(fs_info, group_start);
10397 BUG_ON(!block_group);
10398 BUG_ON(!block_group->ro);
10401 * Free the reserved super bytes from this block group before
10404 free_excluded_extents(fs_info, block_group);
10405 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10406 block_group->key.offset);
10408 memcpy(&key, &block_group->key, sizeof(key));
10409 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10410 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10411 BTRFS_BLOCK_GROUP_RAID1 |
10412 BTRFS_BLOCK_GROUP_RAID10))
10417 /* make sure this block group isn't part of an allocation cluster */
10418 cluster = &fs_info->data_alloc_cluster;
10419 spin_lock(&cluster->refill_lock);
10420 btrfs_return_cluster_to_free_space(block_group, cluster);
10421 spin_unlock(&cluster->refill_lock);
10424 * make sure this block group isn't part of a metadata
10425 * allocation cluster
10427 cluster = &fs_info->meta_alloc_cluster;
10428 spin_lock(&cluster->refill_lock);
10429 btrfs_return_cluster_to_free_space(block_group, cluster);
10430 spin_unlock(&cluster->refill_lock);
10432 path = btrfs_alloc_path();
10439 * get the inode first so any iput calls done for the io_list
10440 * aren't the final iput (no unlinks allowed now)
10442 inode = lookup_free_space_inode(fs_info, block_group, path);
10444 mutex_lock(&trans->transaction->cache_write_mutex);
10446 * make sure our free spache cache IO is done before remove the
10449 spin_lock(&trans->transaction->dirty_bgs_lock);
10450 if (!list_empty(&block_group->io_list)) {
10451 list_del_init(&block_group->io_list);
10453 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10455 spin_unlock(&trans->transaction->dirty_bgs_lock);
10456 btrfs_wait_cache_io(trans, block_group, path);
10457 btrfs_put_block_group(block_group);
10458 spin_lock(&trans->transaction->dirty_bgs_lock);
10461 if (!list_empty(&block_group->dirty_list)) {
10462 list_del_init(&block_group->dirty_list);
10463 btrfs_put_block_group(block_group);
10465 spin_unlock(&trans->transaction->dirty_bgs_lock);
10466 mutex_unlock(&trans->transaction->cache_write_mutex);
10468 if (!IS_ERR(inode)) {
10469 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10471 btrfs_add_delayed_iput(inode);
10474 clear_nlink(inode);
10475 /* One for the block groups ref */
10476 spin_lock(&block_group->lock);
10477 if (block_group->iref) {
10478 block_group->iref = 0;
10479 block_group->inode = NULL;
10480 spin_unlock(&block_group->lock);
10483 spin_unlock(&block_group->lock);
10485 /* One for our lookup ref */
10486 btrfs_add_delayed_iput(inode);
10489 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10490 key.offset = block_group->key.objectid;
10493 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10497 btrfs_release_path(path);
10499 ret = btrfs_del_item(trans, tree_root, path);
10502 btrfs_release_path(path);
10505 spin_lock(&fs_info->block_group_cache_lock);
10506 rb_erase(&block_group->cache_node,
10507 &fs_info->block_group_cache_tree);
10508 RB_CLEAR_NODE(&block_group->cache_node);
10510 if (fs_info->first_logical_byte == block_group->key.objectid)
10511 fs_info->first_logical_byte = (u64)-1;
10512 spin_unlock(&fs_info->block_group_cache_lock);
10514 down_write(&block_group->space_info->groups_sem);
10516 * we must use list_del_init so people can check to see if they
10517 * are still on the list after taking the semaphore
10519 list_del_init(&block_group->list);
10520 if (list_empty(&block_group->space_info->block_groups[index])) {
10521 kobj = block_group->space_info->block_group_kobjs[index];
10522 block_group->space_info->block_group_kobjs[index] = NULL;
10523 clear_avail_alloc_bits(fs_info, block_group->flags);
10525 up_write(&block_group->space_info->groups_sem);
10531 if (block_group->has_caching_ctl)
10532 caching_ctl = get_caching_control(block_group);
10533 if (block_group->cached == BTRFS_CACHE_STARTED)
10534 wait_block_group_cache_done(block_group);
10535 if (block_group->has_caching_ctl) {
10536 down_write(&fs_info->commit_root_sem);
10537 if (!caching_ctl) {
10538 struct btrfs_caching_control *ctl;
10540 list_for_each_entry(ctl,
10541 &fs_info->caching_block_groups, list)
10542 if (ctl->block_group == block_group) {
10544 refcount_inc(&caching_ctl->count);
10549 list_del_init(&caching_ctl->list);
10550 up_write(&fs_info->commit_root_sem);
10552 /* Once for the caching bgs list and once for us. */
10553 put_caching_control(caching_ctl);
10554 put_caching_control(caching_ctl);
10558 spin_lock(&trans->transaction->dirty_bgs_lock);
10559 if (!list_empty(&block_group->dirty_list)) {
10562 if (!list_empty(&block_group->io_list)) {
10565 spin_unlock(&trans->transaction->dirty_bgs_lock);
10566 btrfs_remove_free_space_cache(block_group);
10568 spin_lock(&block_group->space_info->lock);
10569 list_del_init(&block_group->ro_list);
10571 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10572 WARN_ON(block_group->space_info->total_bytes
10573 < block_group->key.offset);
10574 WARN_ON(block_group->space_info->bytes_readonly
10575 < block_group->key.offset);
10576 WARN_ON(block_group->space_info->disk_total
10577 < block_group->key.offset * factor);
10579 block_group->space_info->total_bytes -= block_group->key.offset;
10580 block_group->space_info->bytes_readonly -= block_group->key.offset;
10581 block_group->space_info->disk_total -= block_group->key.offset * factor;
10583 spin_unlock(&block_group->space_info->lock);
10585 memcpy(&key, &block_group->key, sizeof(key));
10587 mutex_lock(&fs_info->chunk_mutex);
10588 if (!list_empty(&em->list)) {
10589 /* We're in the transaction->pending_chunks list. */
10590 free_extent_map(em);
10592 spin_lock(&block_group->lock);
10593 block_group->removed = 1;
10595 * At this point trimming can't start on this block group, because we
10596 * removed the block group from the tree fs_info->block_group_cache_tree
10597 * so no one can't find it anymore and even if someone already got this
10598 * block group before we removed it from the rbtree, they have already
10599 * incremented block_group->trimming - if they didn't, they won't find
10600 * any free space entries because we already removed them all when we
10601 * called btrfs_remove_free_space_cache().
10603 * And we must not remove the extent map from the fs_info->mapping_tree
10604 * to prevent the same logical address range and physical device space
10605 * ranges from being reused for a new block group. This is because our
10606 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10607 * completely transactionless, so while it is trimming a range the
10608 * currently running transaction might finish and a new one start,
10609 * allowing for new block groups to be created that can reuse the same
10610 * physical device locations unless we take this special care.
10612 * There may also be an implicit trim operation if the file system
10613 * is mounted with -odiscard. The same protections must remain
10614 * in place until the extents have been discarded completely when
10615 * the transaction commit has completed.
10617 remove_em = (atomic_read(&block_group->trimming) == 0);
10619 * Make sure a trimmer task always sees the em in the pinned_chunks list
10620 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10621 * before checking block_group->removed).
10625 * Our em might be in trans->transaction->pending_chunks which
10626 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10627 * and so is the fs_info->pinned_chunks list.
10629 * So at this point we must be holding the chunk_mutex to avoid
10630 * any races with chunk allocation (more specifically at
10631 * volumes.c:contains_pending_extent()), to ensure it always
10632 * sees the em, either in the pending_chunks list or in the
10633 * pinned_chunks list.
10635 list_move_tail(&em->list, &fs_info->pinned_chunks);
10637 spin_unlock(&block_group->lock);
10640 struct extent_map_tree *em_tree;
10642 em_tree = &fs_info->mapping_tree.map_tree;
10643 write_lock(&em_tree->lock);
10645 * The em might be in the pending_chunks list, so make sure the
10646 * chunk mutex is locked, since remove_extent_mapping() will
10647 * delete us from that list.
10649 remove_extent_mapping(em_tree, em);
10650 write_unlock(&em_tree->lock);
10651 /* once for the tree */
10652 free_extent_map(em);
10655 mutex_unlock(&fs_info->chunk_mutex);
10657 ret = remove_block_group_free_space(trans, fs_info, block_group);
10661 btrfs_put_block_group(block_group);
10662 btrfs_put_block_group(block_group);
10664 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10670 ret = btrfs_del_item(trans, root, path);
10672 btrfs_free_path(path);
10676 struct btrfs_trans_handle *
10677 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10678 const u64 chunk_offset)
10680 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10681 struct extent_map *em;
10682 struct map_lookup *map;
10683 unsigned int num_items;
10685 read_lock(&em_tree->lock);
10686 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10687 read_unlock(&em_tree->lock);
10688 ASSERT(em && em->start == chunk_offset);
10691 * We need to reserve 3 + N units from the metadata space info in order
10692 * to remove a block group (done at btrfs_remove_chunk() and at
10693 * btrfs_remove_block_group()), which are used for:
10695 * 1 unit for adding the free space inode's orphan (located in the tree
10697 * 1 unit for deleting the block group item (located in the extent
10699 * 1 unit for deleting the free space item (located in tree of tree
10701 * N units for deleting N device extent items corresponding to each
10702 * stripe (located in the device tree).
10704 * In order to remove a block group we also need to reserve units in the
10705 * system space info in order to update the chunk tree (update one or
10706 * more device items and remove one chunk item), but this is done at
10707 * btrfs_remove_chunk() through a call to check_system_chunk().
10709 map = em->map_lookup;
10710 num_items = 3 + map->num_stripes;
10711 free_extent_map(em);
10713 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10718 * Process the unused_bgs list and remove any that don't have any allocated
10719 * space inside of them.
10721 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10723 struct btrfs_block_group_cache *block_group;
10724 struct btrfs_space_info *space_info;
10725 struct btrfs_trans_handle *trans;
10728 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10731 spin_lock(&fs_info->unused_bgs_lock);
10732 while (!list_empty(&fs_info->unused_bgs)) {
10736 block_group = list_first_entry(&fs_info->unused_bgs,
10737 struct btrfs_block_group_cache,
10739 list_del_init(&block_group->bg_list);
10741 space_info = block_group->space_info;
10743 if (ret || btrfs_mixed_space_info(space_info)) {
10744 btrfs_put_block_group(block_group);
10747 spin_unlock(&fs_info->unused_bgs_lock);
10749 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10751 /* Don't want to race with allocators so take the groups_sem */
10752 down_write(&space_info->groups_sem);
10753 spin_lock(&block_group->lock);
10754 if (block_group->reserved ||
10755 btrfs_block_group_used(&block_group->item) ||
10757 list_is_singular(&block_group->list)) {
10759 * We want to bail if we made new allocations or have
10760 * outstanding allocations in this block group. We do
10761 * the ro check in case balance is currently acting on
10762 * this block group.
10764 spin_unlock(&block_group->lock);
10765 up_write(&space_info->groups_sem);
10768 spin_unlock(&block_group->lock);
10770 /* We don't want to force the issue, only flip if it's ok. */
10771 ret = inc_block_group_ro(block_group, 0);
10772 up_write(&space_info->groups_sem);
10779 * Want to do this before we do anything else so we can recover
10780 * properly if we fail to join the transaction.
10782 trans = btrfs_start_trans_remove_block_group(fs_info,
10783 block_group->key.objectid);
10784 if (IS_ERR(trans)) {
10785 btrfs_dec_block_group_ro(block_group);
10786 ret = PTR_ERR(trans);
10791 * We could have pending pinned extents for this block group,
10792 * just delete them, we don't care about them anymore.
10794 start = block_group->key.objectid;
10795 end = start + block_group->key.offset - 1;
10797 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10798 * btrfs_finish_extent_commit(). If we are at transaction N,
10799 * another task might be running finish_extent_commit() for the
10800 * previous transaction N - 1, and have seen a range belonging
10801 * to the block group in freed_extents[] before we were able to
10802 * clear the whole block group range from freed_extents[]. This
10803 * means that task can lookup for the block group after we
10804 * unpinned it from freed_extents[] and removed it, leading to
10805 * a BUG_ON() at btrfs_unpin_extent_range().
10807 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10808 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10811 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10812 btrfs_dec_block_group_ro(block_group);
10815 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10818 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10819 btrfs_dec_block_group_ro(block_group);
10822 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10824 /* Reset pinned so btrfs_put_block_group doesn't complain */
10825 spin_lock(&space_info->lock);
10826 spin_lock(&block_group->lock);
10828 space_info->bytes_pinned -= block_group->pinned;
10829 space_info->bytes_readonly += block_group->pinned;
10830 percpu_counter_add(&space_info->total_bytes_pinned,
10831 -block_group->pinned);
10832 block_group->pinned = 0;
10834 spin_unlock(&block_group->lock);
10835 spin_unlock(&space_info->lock);
10837 /* DISCARD can flip during remount */
10838 trimming = btrfs_test_opt(fs_info, DISCARD);
10840 /* Implicit trim during transaction commit. */
10842 btrfs_get_block_group_trimming(block_group);
10845 * Btrfs_remove_chunk will abort the transaction if things go
10848 ret = btrfs_remove_chunk(trans, fs_info,
10849 block_group->key.objectid);
10853 btrfs_put_block_group_trimming(block_group);
10858 * If we're not mounted with -odiscard, we can just forget
10859 * about this block group. Otherwise we'll need to wait
10860 * until transaction commit to do the actual discard.
10863 spin_lock(&fs_info->unused_bgs_lock);
10865 * A concurrent scrub might have added us to the list
10866 * fs_info->unused_bgs, so use a list_move operation
10867 * to add the block group to the deleted_bgs list.
10869 list_move(&block_group->bg_list,
10870 &trans->transaction->deleted_bgs);
10871 spin_unlock(&fs_info->unused_bgs_lock);
10872 btrfs_get_block_group(block_group);
10875 btrfs_end_transaction(trans);
10877 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10878 btrfs_put_block_group(block_group);
10879 spin_lock(&fs_info->unused_bgs_lock);
10881 spin_unlock(&fs_info->unused_bgs_lock);
10884 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10886 struct btrfs_space_info *space_info;
10887 struct btrfs_super_block *disk_super;
10893 disk_super = fs_info->super_copy;
10894 if (!btrfs_super_root(disk_super))
10897 features = btrfs_super_incompat_flags(disk_super);
10898 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10901 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10902 ret = create_space_info(fs_info, flags, &space_info);
10907 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10908 ret = create_space_info(fs_info, flags, &space_info);
10910 flags = BTRFS_BLOCK_GROUP_METADATA;
10911 ret = create_space_info(fs_info, flags, &space_info);
10915 flags = BTRFS_BLOCK_GROUP_DATA;
10916 ret = create_space_info(fs_info, flags, &space_info);
10922 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10923 u64 start, u64 end)
10925 return unpin_extent_range(fs_info, start, end, false);
10929 * It used to be that old block groups would be left around forever.
10930 * Iterating over them would be enough to trim unused space. Since we
10931 * now automatically remove them, we also need to iterate over unallocated
10934 * We don't want a transaction for this since the discard may take a
10935 * substantial amount of time. We don't require that a transaction be
10936 * running, but we do need to take a running transaction into account
10937 * to ensure that we're not discarding chunks that were released in
10938 * the current transaction.
10940 * Holding the chunks lock will prevent other threads from allocating
10941 * or releasing chunks, but it won't prevent a running transaction
10942 * from committing and releasing the memory that the pending chunks
10943 * list head uses. For that, we need to take a reference to the
10946 static int btrfs_trim_free_extents(struct btrfs_device *device,
10947 u64 minlen, u64 *trimmed)
10949 u64 start = 0, len = 0;
10954 /* Not writeable = nothing to do. */
10955 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10958 /* No free space = nothing to do. */
10959 if (device->total_bytes <= device->bytes_used)
10965 struct btrfs_fs_info *fs_info = device->fs_info;
10966 struct btrfs_transaction *trans;
10969 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10973 down_read(&fs_info->commit_root_sem);
10975 spin_lock(&fs_info->trans_lock);
10976 trans = fs_info->running_transaction;
10978 refcount_inc(&trans->use_count);
10979 spin_unlock(&fs_info->trans_lock);
10981 ret = find_free_dev_extent_start(trans, device, minlen, start,
10984 btrfs_put_transaction(trans);
10987 up_read(&fs_info->commit_root_sem);
10988 mutex_unlock(&fs_info->chunk_mutex);
10989 if (ret == -ENOSPC)
10994 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10995 up_read(&fs_info->commit_root_sem);
10996 mutex_unlock(&fs_info->chunk_mutex);
11004 if (fatal_signal_pending(current)) {
11005 ret = -ERESTARTSYS;
11015 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11017 struct btrfs_block_group_cache *cache = NULL;
11018 struct btrfs_device *device;
11019 struct list_head *devices;
11024 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11028 * try to trim all FS space, our block group may start from non-zero.
11030 if (range->len == total_bytes)
11031 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11033 cache = btrfs_lookup_block_group(fs_info, range->start);
11036 if (cache->key.objectid >= (range->start + range->len)) {
11037 btrfs_put_block_group(cache);
11041 start = max(range->start, cache->key.objectid);
11042 end = min(range->start + range->len,
11043 cache->key.objectid + cache->key.offset);
11045 if (end - start >= range->minlen) {
11046 if (!block_group_cache_done(cache)) {
11047 ret = cache_block_group(cache, 0);
11049 btrfs_put_block_group(cache);
11052 ret = wait_block_group_cache_done(cache);
11054 btrfs_put_block_group(cache);
11058 ret = btrfs_trim_block_group(cache,
11064 trimmed += group_trimmed;
11066 btrfs_put_block_group(cache);
11071 cache = next_block_group(fs_info, cache);
11074 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11075 devices = &fs_info->fs_devices->alloc_list;
11076 list_for_each_entry(device, devices, dev_alloc_list) {
11077 ret = btrfs_trim_free_extents(device, range->minlen,
11082 trimmed += group_trimmed;
11084 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11086 range->len = trimmed;
11091 * btrfs_{start,end}_write_no_snapshotting() are similar to
11092 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11093 * data into the page cache through nocow before the subvolume is snapshoted,
11094 * but flush the data into disk after the snapshot creation, or to prevent
11095 * operations while snapshotting is ongoing and that cause the snapshot to be
11096 * inconsistent (writes followed by expanding truncates for example).
11098 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11100 percpu_counter_dec(&root->subv_writers->counter);
11102 * Make sure counter is updated before we wake up waiters.
11105 if (waitqueue_active(&root->subv_writers->wait))
11106 wake_up(&root->subv_writers->wait);
11109 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11111 if (atomic_read(&root->will_be_snapshotted))
11114 percpu_counter_inc(&root->subv_writers->counter);
11116 * Make sure counter is updated before we check for snapshot creation.
11119 if (atomic_read(&root->will_be_snapshotted)) {
11120 btrfs_end_write_no_snapshotting(root);
11126 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11131 ret = btrfs_start_write_no_snapshotting(root);
11134 wait_var_event(&root->will_be_snapshotted,
11135 !atomic_read(&root->will_be_snapshotted));