2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
36 #include "free-space-tree.h"
41 #undef SCRAMBLE_DELAYED_REFS
44 * control flags for do_chunk_alloc's force field
45 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
46 * if we really need one.
48 * CHUNK_ALLOC_LIMITED means to only try and allocate one
49 * if we have very few chunks already allocated. This is
50 * used as part of the clustering code to help make sure
51 * we have a good pool of storage to cluster in, without
52 * filling the FS with empty chunks
54 * CHUNK_ALLOC_FORCE means it must try to allocate one
58 CHUNK_ALLOC_NO_FORCE = 0,
59 CHUNK_ALLOC_LIMITED = 1,
60 CHUNK_ALLOC_FORCE = 2,
64 * Control how reservations are dealt with.
66 * RESERVE_FREE - freeing a reservation.
67 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
69 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
70 * bytes_may_use as the ENOSPC accounting is done elsewhere
75 RESERVE_ALLOC_NO_ACCOUNT = 2,
78 static int update_block_group(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root, u64 bytenr,
80 u64 num_bytes, int alloc);
81 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
82 struct btrfs_root *root,
83 struct btrfs_delayed_ref_node *node, u64 parent,
84 u64 root_objectid, u64 owner_objectid,
85 u64 owner_offset, int refs_to_drop,
86 struct btrfs_delayed_extent_op *extra_op);
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
88 struct extent_buffer *leaf,
89 struct btrfs_extent_item *ei);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
91 struct btrfs_root *root,
92 u64 parent, u64 root_objectid,
93 u64 flags, u64 owner, u64 offset,
94 struct btrfs_key *ins, int ref_mod);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 u64 parent, u64 root_objectid,
98 u64 flags, struct btrfs_disk_key *key,
99 int level, struct btrfs_key *ins);
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 return (cache->flags & bits) == bits;
128 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 atomic_inc(&cache->count);
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
161 if (block_group->key.objectid < cache->key.objectid) {
163 } else if (block_group->key.objectid > cache->key.objectid) {
166 spin_unlock(&info->block_group_cache_lock);
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
178 spin_unlock(&info->block_group_cache_lock);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
191 struct btrfs_block_group_cache *cache, *ret = NULL;
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
199 cache = rb_entry(n, struct btrfs_block_group_cache,
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
224 spin_unlock(&info->block_group_cache_lock);
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
282 if (logical[nr] > cache->key.objectid +
286 if (logical[nr] + stripe_len <= cache->key.objectid)
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
315 struct btrfs_caching_control *ctl;
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
329 static void put_caching_control(struct btrfs_caching_control *ctl)
331 if (atomic_dec_and_test(&ctl->count))
335 #ifdef CONFIG_BTRFS_DEBUG
336 static void fragment_free_space(struct btrfs_root *root,
337 struct btrfs_block_group_cache *block_group)
339 u64 start = block_group->key.objectid;
340 u64 len = block_group->key.offset;
341 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
342 root->nodesize : root->sectorsize;
343 u64 step = chunk << 1;
345 while (len > chunk) {
346 btrfs_remove_free_space(block_group, start, chunk);
357 * this is only called by cache_block_group, since we could have freed extents
358 * we need to check the pinned_extents for any extents that can't be used yet
359 * since their free space will be released as soon as the transaction commits.
361 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
362 struct btrfs_fs_info *info, u64 start, u64 end)
364 u64 extent_start, extent_end, size, total_added = 0;
367 while (start < end) {
368 ret = find_first_extent_bit(info->pinned_extents, start,
369 &extent_start, &extent_end,
370 EXTENT_DIRTY | EXTENT_UPTODATE,
375 if (extent_start <= start) {
376 start = extent_end + 1;
377 } else if (extent_start > start && extent_start < end) {
378 size = extent_start - start;
380 ret = btrfs_add_free_space(block_group, start,
382 BUG_ON(ret); /* -ENOMEM or logic error */
383 start = extent_end + 1;
392 ret = btrfs_add_free_space(block_group, start, size);
393 BUG_ON(ret); /* -ENOMEM or logic error */
399 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
401 struct btrfs_block_group_cache *block_group;
402 struct btrfs_fs_info *fs_info;
403 struct btrfs_root *extent_root;
404 struct btrfs_path *path;
405 struct extent_buffer *leaf;
406 struct btrfs_key key;
413 block_group = caching_ctl->block_group;
414 fs_info = block_group->fs_info;
415 extent_root = fs_info->extent_root;
417 path = btrfs_alloc_path();
421 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
423 #ifdef CONFIG_BTRFS_DEBUG
425 * If we're fragmenting we don't want to make anybody think we can
426 * allocate from this block group until we've had a chance to fragment
429 if (btrfs_should_fragment_free_space(extent_root, block_group))
433 * We don't want to deadlock with somebody trying to allocate a new
434 * extent for the extent root while also trying to search the extent
435 * root to add free space. So we skip locking and search the commit
436 * root, since its read-only
438 path->skip_locking = 1;
439 path->search_commit_root = 1;
444 key.type = BTRFS_EXTENT_ITEM_KEY;
447 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
451 leaf = path->nodes[0];
452 nritems = btrfs_header_nritems(leaf);
455 if (btrfs_fs_closing(fs_info) > 1) {
460 if (path->slots[0] < nritems) {
461 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
463 ret = find_next_key(path, 0, &key);
467 if (need_resched() ||
468 rwsem_is_contended(&fs_info->commit_root_sem)) {
470 caching_ctl->progress = last;
471 btrfs_release_path(path);
472 up_read(&fs_info->commit_root_sem);
473 mutex_unlock(&caching_ctl->mutex);
475 mutex_lock(&caching_ctl->mutex);
476 down_read(&fs_info->commit_root_sem);
480 ret = btrfs_next_leaf(extent_root, path);
485 leaf = path->nodes[0];
486 nritems = btrfs_header_nritems(leaf);
490 if (key.objectid < last) {
493 key.type = BTRFS_EXTENT_ITEM_KEY;
496 caching_ctl->progress = last;
497 btrfs_release_path(path);
501 if (key.objectid < block_group->key.objectid) {
506 if (key.objectid >= block_group->key.objectid +
507 block_group->key.offset)
510 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
511 key.type == BTRFS_METADATA_ITEM_KEY) {
512 total_found += add_new_free_space(block_group,
515 if (key.type == BTRFS_METADATA_ITEM_KEY)
516 last = key.objectid +
517 fs_info->tree_root->nodesize;
519 last = key.objectid + key.offset;
521 if (total_found > CACHING_CTL_WAKE_UP) {
524 wake_up(&caching_ctl->wait);
531 total_found += add_new_free_space(block_group, fs_info, last,
532 block_group->key.objectid +
533 block_group->key.offset);
534 caching_ctl->progress = (u64)-1;
537 btrfs_free_path(path);
541 static noinline void caching_thread(struct btrfs_work *work)
543 struct btrfs_block_group_cache *block_group;
544 struct btrfs_fs_info *fs_info;
545 struct btrfs_caching_control *caching_ctl;
546 struct btrfs_root *extent_root;
549 caching_ctl = container_of(work, struct btrfs_caching_control, work);
550 block_group = caching_ctl->block_group;
551 fs_info = block_group->fs_info;
552 extent_root = fs_info->extent_root;
554 mutex_lock(&caching_ctl->mutex);
555 down_read(&fs_info->commit_root_sem);
557 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
558 ret = load_free_space_tree(caching_ctl);
560 ret = load_extent_tree_free(caching_ctl);
562 spin_lock(&block_group->lock);
563 block_group->caching_ctl = NULL;
564 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
565 spin_unlock(&block_group->lock);
567 #ifdef CONFIG_BTRFS_DEBUG
568 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
571 spin_lock(&block_group->space_info->lock);
572 spin_lock(&block_group->lock);
573 bytes_used = block_group->key.offset -
574 btrfs_block_group_used(&block_group->item);
575 block_group->space_info->bytes_used += bytes_used >> 1;
576 spin_unlock(&block_group->lock);
577 spin_unlock(&block_group->space_info->lock);
578 fragment_free_space(extent_root, block_group);
582 caching_ctl->progress = (u64)-1;
584 up_read(&fs_info->commit_root_sem);
585 free_excluded_extents(fs_info->extent_root, block_group);
586 mutex_unlock(&caching_ctl->mutex);
588 wake_up(&caching_ctl->wait);
590 put_caching_control(caching_ctl);
591 btrfs_put_block_group(block_group);
594 static int cache_block_group(struct btrfs_block_group_cache *cache,
598 struct btrfs_fs_info *fs_info = cache->fs_info;
599 struct btrfs_caching_control *caching_ctl;
602 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
606 INIT_LIST_HEAD(&caching_ctl->list);
607 mutex_init(&caching_ctl->mutex);
608 init_waitqueue_head(&caching_ctl->wait);
609 caching_ctl->block_group = cache;
610 caching_ctl->progress = cache->key.objectid;
611 atomic_set(&caching_ctl->count, 1);
612 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
613 caching_thread, NULL, NULL);
615 spin_lock(&cache->lock);
617 * This should be a rare occasion, but this could happen I think in the
618 * case where one thread starts to load the space cache info, and then
619 * some other thread starts a transaction commit which tries to do an
620 * allocation while the other thread is still loading the space cache
621 * info. The previous loop should have kept us from choosing this block
622 * group, but if we've moved to the state where we will wait on caching
623 * block groups we need to first check if we're doing a fast load here,
624 * so we can wait for it to finish, otherwise we could end up allocating
625 * from a block group who's cache gets evicted for one reason or
628 while (cache->cached == BTRFS_CACHE_FAST) {
629 struct btrfs_caching_control *ctl;
631 ctl = cache->caching_ctl;
632 atomic_inc(&ctl->count);
633 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
634 spin_unlock(&cache->lock);
638 finish_wait(&ctl->wait, &wait);
639 put_caching_control(ctl);
640 spin_lock(&cache->lock);
643 if (cache->cached != BTRFS_CACHE_NO) {
644 spin_unlock(&cache->lock);
648 WARN_ON(cache->caching_ctl);
649 cache->caching_ctl = caching_ctl;
650 cache->cached = BTRFS_CACHE_FAST;
651 spin_unlock(&cache->lock);
653 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
654 mutex_lock(&caching_ctl->mutex);
655 ret = load_free_space_cache(fs_info, cache);
657 spin_lock(&cache->lock);
659 cache->caching_ctl = NULL;
660 cache->cached = BTRFS_CACHE_FINISHED;
661 cache->last_byte_to_unpin = (u64)-1;
662 caching_ctl->progress = (u64)-1;
664 if (load_cache_only) {
665 cache->caching_ctl = NULL;
666 cache->cached = BTRFS_CACHE_NO;
668 cache->cached = BTRFS_CACHE_STARTED;
669 cache->has_caching_ctl = 1;
672 spin_unlock(&cache->lock);
673 #ifdef CONFIG_BTRFS_DEBUG
675 btrfs_should_fragment_free_space(fs_info->extent_root,
679 spin_lock(&cache->space_info->lock);
680 spin_lock(&cache->lock);
681 bytes_used = cache->key.offset -
682 btrfs_block_group_used(&cache->item);
683 cache->space_info->bytes_used += bytes_used >> 1;
684 spin_unlock(&cache->lock);
685 spin_unlock(&cache->space_info->lock);
686 fragment_free_space(fs_info->extent_root, cache);
689 mutex_unlock(&caching_ctl->mutex);
691 wake_up(&caching_ctl->wait);
693 put_caching_control(caching_ctl);
694 free_excluded_extents(fs_info->extent_root, cache);
699 * We're either using the free space tree or no caching at all.
700 * Set cached to the appropriate value and wakeup any waiters.
702 spin_lock(&cache->lock);
703 if (load_cache_only) {
704 cache->caching_ctl = NULL;
705 cache->cached = BTRFS_CACHE_NO;
707 cache->cached = BTRFS_CACHE_STARTED;
708 cache->has_caching_ctl = 1;
710 spin_unlock(&cache->lock);
711 wake_up(&caching_ctl->wait);
714 if (load_cache_only) {
715 put_caching_control(caching_ctl);
719 down_write(&fs_info->commit_root_sem);
720 atomic_inc(&caching_ctl->count);
721 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
722 up_write(&fs_info->commit_root_sem);
724 btrfs_get_block_group(cache);
726 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
732 * return the block group that starts at or after bytenr
734 static struct btrfs_block_group_cache *
735 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
737 struct btrfs_block_group_cache *cache;
739 cache = block_group_cache_tree_search(info, bytenr, 0);
745 * return the block group that contains the given bytenr
747 struct btrfs_block_group_cache *btrfs_lookup_block_group(
748 struct btrfs_fs_info *info,
751 struct btrfs_block_group_cache *cache;
753 cache = block_group_cache_tree_search(info, bytenr, 1);
758 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
761 struct list_head *head = &info->space_info;
762 struct btrfs_space_info *found;
764 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
767 list_for_each_entry_rcu(found, head, list) {
768 if (found->flags & flags) {
778 * after adding space to the filesystem, we need to clear the full flags
779 * on all the space infos.
781 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
783 struct list_head *head = &info->space_info;
784 struct btrfs_space_info *found;
787 list_for_each_entry_rcu(found, head, list)
792 /* simple helper to search for an existing data extent at a given offset */
793 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
796 struct btrfs_key key;
797 struct btrfs_path *path;
799 path = btrfs_alloc_path();
803 key.objectid = start;
805 key.type = BTRFS_EXTENT_ITEM_KEY;
806 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
808 btrfs_free_path(path);
813 * helper function to lookup reference count and flags of a tree block.
815 * the head node for delayed ref is used to store the sum of all the
816 * reference count modifications queued up in the rbtree. the head
817 * node may also store the extent flags to set. This way you can check
818 * to see what the reference count and extent flags would be if all of
819 * the delayed refs are not processed.
821 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
822 struct btrfs_root *root, u64 bytenr,
823 u64 offset, int metadata, u64 *refs, u64 *flags)
825 struct btrfs_delayed_ref_head *head;
826 struct btrfs_delayed_ref_root *delayed_refs;
827 struct btrfs_path *path;
828 struct btrfs_extent_item *ei;
829 struct extent_buffer *leaf;
830 struct btrfs_key key;
837 * If we don't have skinny metadata, don't bother doing anything
840 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
841 offset = root->nodesize;
845 path = btrfs_alloc_path();
850 path->skip_locking = 1;
851 path->search_commit_root = 1;
855 key.objectid = bytenr;
858 key.type = BTRFS_METADATA_ITEM_KEY;
860 key.type = BTRFS_EXTENT_ITEM_KEY;
862 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
867 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
868 if (path->slots[0]) {
870 btrfs_item_key_to_cpu(path->nodes[0], &key,
872 if (key.objectid == bytenr &&
873 key.type == BTRFS_EXTENT_ITEM_KEY &&
874 key.offset == root->nodesize)
880 leaf = path->nodes[0];
881 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
882 if (item_size >= sizeof(*ei)) {
883 ei = btrfs_item_ptr(leaf, path->slots[0],
884 struct btrfs_extent_item);
885 num_refs = btrfs_extent_refs(leaf, ei);
886 extent_flags = btrfs_extent_flags(leaf, ei);
888 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
889 struct btrfs_extent_item_v0 *ei0;
890 BUG_ON(item_size != sizeof(*ei0));
891 ei0 = btrfs_item_ptr(leaf, path->slots[0],
892 struct btrfs_extent_item_v0);
893 num_refs = btrfs_extent_refs_v0(leaf, ei0);
894 /* FIXME: this isn't correct for data */
895 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
900 BUG_ON(num_refs == 0);
910 delayed_refs = &trans->transaction->delayed_refs;
911 spin_lock(&delayed_refs->lock);
912 head = btrfs_find_delayed_ref_head(trans, bytenr);
914 if (!mutex_trylock(&head->mutex)) {
915 atomic_inc(&head->node.refs);
916 spin_unlock(&delayed_refs->lock);
918 btrfs_release_path(path);
921 * Mutex was contended, block until it's released and try
924 mutex_lock(&head->mutex);
925 mutex_unlock(&head->mutex);
926 btrfs_put_delayed_ref(&head->node);
929 spin_lock(&head->lock);
930 if (head->extent_op && head->extent_op->update_flags)
931 extent_flags |= head->extent_op->flags_to_set;
933 BUG_ON(num_refs == 0);
935 num_refs += head->node.ref_mod;
936 spin_unlock(&head->lock);
937 mutex_unlock(&head->mutex);
939 spin_unlock(&delayed_refs->lock);
941 WARN_ON(num_refs == 0);
945 *flags = extent_flags;
947 btrfs_free_path(path);
952 * Back reference rules. Back refs have three main goals:
954 * 1) differentiate between all holders of references to an extent so that
955 * when a reference is dropped we can make sure it was a valid reference
956 * before freeing the extent.
958 * 2) Provide enough information to quickly find the holders of an extent
959 * if we notice a given block is corrupted or bad.
961 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
962 * maintenance. This is actually the same as #2, but with a slightly
963 * different use case.
965 * There are two kinds of back refs. The implicit back refs is optimized
966 * for pointers in non-shared tree blocks. For a given pointer in a block,
967 * back refs of this kind provide information about the block's owner tree
968 * and the pointer's key. These information allow us to find the block by
969 * b-tree searching. The full back refs is for pointers in tree blocks not
970 * referenced by their owner trees. The location of tree block is recorded
971 * in the back refs. Actually the full back refs is generic, and can be
972 * used in all cases the implicit back refs is used. The major shortcoming
973 * of the full back refs is its overhead. Every time a tree block gets
974 * COWed, we have to update back refs entry for all pointers in it.
976 * For a newly allocated tree block, we use implicit back refs for
977 * pointers in it. This means most tree related operations only involve
978 * implicit back refs. For a tree block created in old transaction, the
979 * only way to drop a reference to it is COW it. So we can detect the
980 * event that tree block loses its owner tree's reference and do the
981 * back refs conversion.
983 * When a tree block is COW'd through a tree, there are four cases:
985 * The reference count of the block is one and the tree is the block's
986 * owner tree. Nothing to do in this case.
988 * The reference count of the block is one and the tree is not the
989 * block's owner tree. In this case, full back refs is used for pointers
990 * in the block. Remove these full back refs, add implicit back refs for
991 * every pointers in the new block.
993 * The reference count of the block is greater than one and the tree is
994 * the block's owner tree. In this case, implicit back refs is used for
995 * pointers in the block. Add full back refs for every pointers in the
996 * block, increase lower level extents' reference counts. The original
997 * implicit back refs are entailed to the new block.
999 * The reference count of the block is greater than one and the tree is
1000 * not the block's owner tree. Add implicit back refs for every pointer in
1001 * the new block, increase lower level extents' reference count.
1003 * Back Reference Key composing:
1005 * The key objectid corresponds to the first byte in the extent,
1006 * The key type is used to differentiate between types of back refs.
1007 * There are different meanings of the key offset for different types
1010 * File extents can be referenced by:
1012 * - multiple snapshots, subvolumes, or different generations in one subvol
1013 * - different files inside a single subvolume
1014 * - different offsets inside a file (bookend extents in file.c)
1016 * The extent ref structure for the implicit back refs has fields for:
1018 * - Objectid of the subvolume root
1019 * - objectid of the file holding the reference
1020 * - original offset in the file
1021 * - how many bookend extents
1023 * The key offset for the implicit back refs is hash of the first
1026 * The extent ref structure for the full back refs has field for:
1028 * - number of pointers in the tree leaf
1030 * The key offset for the implicit back refs is the first byte of
1033 * When a file extent is allocated, The implicit back refs is used.
1034 * the fields are filled in:
1036 * (root_key.objectid, inode objectid, offset in file, 1)
1038 * When a file extent is removed file truncation, we find the
1039 * corresponding implicit back refs and check the following fields:
1041 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1043 * Btree extents can be referenced by:
1045 * - Different subvolumes
1047 * Both the implicit back refs and the full back refs for tree blocks
1048 * only consist of key. The key offset for the implicit back refs is
1049 * objectid of block's owner tree. The key offset for the full back refs
1050 * is the first byte of parent block.
1052 * When implicit back refs is used, information about the lowest key and
1053 * level of the tree block are required. These information are stored in
1054 * tree block info structure.
1057 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1058 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1059 struct btrfs_root *root,
1060 struct btrfs_path *path,
1061 u64 owner, u32 extra_size)
1063 struct btrfs_extent_item *item;
1064 struct btrfs_extent_item_v0 *ei0;
1065 struct btrfs_extent_ref_v0 *ref0;
1066 struct btrfs_tree_block_info *bi;
1067 struct extent_buffer *leaf;
1068 struct btrfs_key key;
1069 struct btrfs_key found_key;
1070 u32 new_size = sizeof(*item);
1074 leaf = path->nodes[0];
1075 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1077 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1078 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1079 struct btrfs_extent_item_v0);
1080 refs = btrfs_extent_refs_v0(leaf, ei0);
1082 if (owner == (u64)-1) {
1084 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1085 ret = btrfs_next_leaf(root, path);
1088 BUG_ON(ret > 0); /* Corruption */
1089 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key,
1093 BUG_ON(key.objectid != found_key.objectid);
1094 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1098 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1099 struct btrfs_extent_ref_v0);
1100 owner = btrfs_ref_objectid_v0(leaf, ref0);
1104 btrfs_release_path(path);
1106 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1107 new_size += sizeof(*bi);
1109 new_size -= sizeof(*ei0);
1110 ret = btrfs_search_slot(trans, root, &key, path,
1111 new_size + extra_size, 1);
1114 BUG_ON(ret); /* Corruption */
1116 btrfs_extend_item(root, path, new_size);
1118 leaf = path->nodes[0];
1119 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1120 btrfs_set_extent_refs(leaf, item, refs);
1121 /* FIXME: get real generation */
1122 btrfs_set_extent_generation(leaf, item, 0);
1123 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1124 btrfs_set_extent_flags(leaf, item,
1125 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1126 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1127 bi = (struct btrfs_tree_block_info *)(item + 1);
1128 /* FIXME: get first key of the block */
1129 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1130 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1132 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1134 btrfs_mark_buffer_dirty(leaf);
1139 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1141 u32 high_crc = ~(u32)0;
1142 u32 low_crc = ~(u32)0;
1145 lenum = cpu_to_le64(root_objectid);
1146 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1147 lenum = cpu_to_le64(owner);
1148 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1149 lenum = cpu_to_le64(offset);
1150 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1152 return ((u64)high_crc << 31) ^ (u64)low_crc;
1155 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1156 struct btrfs_extent_data_ref *ref)
1158 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1159 btrfs_extent_data_ref_objectid(leaf, ref),
1160 btrfs_extent_data_ref_offset(leaf, ref));
1163 static int match_extent_data_ref(struct extent_buffer *leaf,
1164 struct btrfs_extent_data_ref *ref,
1165 u64 root_objectid, u64 owner, u64 offset)
1167 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1168 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1169 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1174 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1175 struct btrfs_root *root,
1176 struct btrfs_path *path,
1177 u64 bytenr, u64 parent,
1179 u64 owner, u64 offset)
1181 struct btrfs_key key;
1182 struct btrfs_extent_data_ref *ref;
1183 struct extent_buffer *leaf;
1189 key.objectid = bytenr;
1191 key.type = BTRFS_SHARED_DATA_REF_KEY;
1192 key.offset = parent;
1194 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1195 key.offset = hash_extent_data_ref(root_objectid,
1200 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1209 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1210 key.type = BTRFS_EXTENT_REF_V0_KEY;
1211 btrfs_release_path(path);
1212 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1223 leaf = path->nodes[0];
1224 nritems = btrfs_header_nritems(leaf);
1226 if (path->slots[0] >= nritems) {
1227 ret = btrfs_next_leaf(root, path);
1233 leaf = path->nodes[0];
1234 nritems = btrfs_header_nritems(leaf);
1238 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1239 if (key.objectid != bytenr ||
1240 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1243 ref = btrfs_item_ptr(leaf, path->slots[0],
1244 struct btrfs_extent_data_ref);
1246 if (match_extent_data_ref(leaf, ref, root_objectid,
1249 btrfs_release_path(path);
1261 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root,
1263 struct btrfs_path *path,
1264 u64 bytenr, u64 parent,
1265 u64 root_objectid, u64 owner,
1266 u64 offset, int refs_to_add)
1268 struct btrfs_key key;
1269 struct extent_buffer *leaf;
1274 key.objectid = bytenr;
1276 key.type = BTRFS_SHARED_DATA_REF_KEY;
1277 key.offset = parent;
1278 size = sizeof(struct btrfs_shared_data_ref);
1280 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1281 key.offset = hash_extent_data_ref(root_objectid,
1283 size = sizeof(struct btrfs_extent_data_ref);
1286 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1287 if (ret && ret != -EEXIST)
1290 leaf = path->nodes[0];
1292 struct btrfs_shared_data_ref *ref;
1293 ref = btrfs_item_ptr(leaf, path->slots[0],
1294 struct btrfs_shared_data_ref);
1296 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1298 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1299 num_refs += refs_to_add;
1300 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1303 struct btrfs_extent_data_ref *ref;
1304 while (ret == -EEXIST) {
1305 ref = btrfs_item_ptr(leaf, path->slots[0],
1306 struct btrfs_extent_data_ref);
1307 if (match_extent_data_ref(leaf, ref, root_objectid,
1310 btrfs_release_path(path);
1312 ret = btrfs_insert_empty_item(trans, root, path, &key,
1314 if (ret && ret != -EEXIST)
1317 leaf = path->nodes[0];
1319 ref = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_extent_data_ref);
1322 btrfs_set_extent_data_ref_root(leaf, ref,
1324 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1325 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1326 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1328 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1329 num_refs += refs_to_add;
1330 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1333 btrfs_mark_buffer_dirty(leaf);
1336 btrfs_release_path(path);
1340 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1341 struct btrfs_root *root,
1342 struct btrfs_path *path,
1343 int refs_to_drop, int *last_ref)
1345 struct btrfs_key key;
1346 struct btrfs_extent_data_ref *ref1 = NULL;
1347 struct btrfs_shared_data_ref *ref2 = NULL;
1348 struct extent_buffer *leaf;
1352 leaf = path->nodes[0];
1353 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1355 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1356 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1357 struct btrfs_extent_data_ref);
1358 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1359 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1360 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1361 struct btrfs_shared_data_ref);
1362 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1363 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1364 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1365 struct btrfs_extent_ref_v0 *ref0;
1366 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1367 struct btrfs_extent_ref_v0);
1368 num_refs = btrfs_ref_count_v0(leaf, ref0);
1374 BUG_ON(num_refs < refs_to_drop);
1375 num_refs -= refs_to_drop;
1377 if (num_refs == 0) {
1378 ret = btrfs_del_item(trans, root, path);
1381 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1382 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1383 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1384 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1385 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1387 struct btrfs_extent_ref_v0 *ref0;
1388 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1389 struct btrfs_extent_ref_v0);
1390 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1393 btrfs_mark_buffer_dirty(leaf);
1398 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1399 struct btrfs_extent_inline_ref *iref)
1401 struct btrfs_key key;
1402 struct extent_buffer *leaf;
1403 struct btrfs_extent_data_ref *ref1;
1404 struct btrfs_shared_data_ref *ref2;
1407 leaf = path->nodes[0];
1408 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1410 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1411 BTRFS_EXTENT_DATA_REF_KEY) {
1412 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1413 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1415 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1416 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1418 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1419 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1420 struct btrfs_extent_data_ref);
1421 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1422 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1423 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1424 struct btrfs_shared_data_ref);
1425 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1426 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1427 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1428 struct btrfs_extent_ref_v0 *ref0;
1429 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1430 struct btrfs_extent_ref_v0);
1431 num_refs = btrfs_ref_count_v0(leaf, ref0);
1439 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1440 struct btrfs_root *root,
1441 struct btrfs_path *path,
1442 u64 bytenr, u64 parent,
1445 struct btrfs_key key;
1448 key.objectid = bytenr;
1450 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1451 key.offset = parent;
1453 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1454 key.offset = root_objectid;
1457 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1460 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1461 if (ret == -ENOENT && parent) {
1462 btrfs_release_path(path);
1463 key.type = BTRFS_EXTENT_REF_V0_KEY;
1464 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1472 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1473 struct btrfs_root *root,
1474 struct btrfs_path *path,
1475 u64 bytenr, u64 parent,
1478 struct btrfs_key key;
1481 key.objectid = bytenr;
1483 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1484 key.offset = parent;
1486 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1487 key.offset = root_objectid;
1490 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1491 btrfs_release_path(path);
1495 static inline int extent_ref_type(u64 parent, u64 owner)
1498 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1500 type = BTRFS_SHARED_BLOCK_REF_KEY;
1502 type = BTRFS_TREE_BLOCK_REF_KEY;
1505 type = BTRFS_SHARED_DATA_REF_KEY;
1507 type = BTRFS_EXTENT_DATA_REF_KEY;
1512 static int find_next_key(struct btrfs_path *path, int level,
1513 struct btrfs_key *key)
1516 for (; level < BTRFS_MAX_LEVEL; level++) {
1517 if (!path->nodes[level])
1519 if (path->slots[level] + 1 >=
1520 btrfs_header_nritems(path->nodes[level]))
1523 btrfs_item_key_to_cpu(path->nodes[level], key,
1524 path->slots[level] + 1);
1526 btrfs_node_key_to_cpu(path->nodes[level], key,
1527 path->slots[level] + 1);
1534 * look for inline back ref. if back ref is found, *ref_ret is set
1535 * to the address of inline back ref, and 0 is returned.
1537 * if back ref isn't found, *ref_ret is set to the address where it
1538 * should be inserted, and -ENOENT is returned.
1540 * if insert is true and there are too many inline back refs, the path
1541 * points to the extent item, and -EAGAIN is returned.
1543 * NOTE: inline back refs are ordered in the same way that back ref
1544 * items in the tree are ordered.
1546 static noinline_for_stack
1547 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1548 struct btrfs_root *root,
1549 struct btrfs_path *path,
1550 struct btrfs_extent_inline_ref **ref_ret,
1551 u64 bytenr, u64 num_bytes,
1552 u64 parent, u64 root_objectid,
1553 u64 owner, u64 offset, int insert)
1555 struct btrfs_key key;
1556 struct extent_buffer *leaf;
1557 struct btrfs_extent_item *ei;
1558 struct btrfs_extent_inline_ref *iref;
1568 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1571 key.objectid = bytenr;
1572 key.type = BTRFS_EXTENT_ITEM_KEY;
1573 key.offset = num_bytes;
1575 want = extent_ref_type(parent, owner);
1577 extra_size = btrfs_extent_inline_ref_size(want);
1578 path->keep_locks = 1;
1583 * Owner is our parent level, so we can just add one to get the level
1584 * for the block we are interested in.
1586 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1587 key.type = BTRFS_METADATA_ITEM_KEY;
1592 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1599 * We may be a newly converted file system which still has the old fat
1600 * extent entries for metadata, so try and see if we have one of those.
1602 if (ret > 0 && skinny_metadata) {
1603 skinny_metadata = false;
1604 if (path->slots[0]) {
1606 btrfs_item_key_to_cpu(path->nodes[0], &key,
1608 if (key.objectid == bytenr &&
1609 key.type == BTRFS_EXTENT_ITEM_KEY &&
1610 key.offset == num_bytes)
1614 key.objectid = bytenr;
1615 key.type = BTRFS_EXTENT_ITEM_KEY;
1616 key.offset = num_bytes;
1617 btrfs_release_path(path);
1622 if (ret && !insert) {
1625 } else if (WARN_ON(ret)) {
1630 leaf = path->nodes[0];
1631 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1632 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1633 if (item_size < sizeof(*ei)) {
1638 ret = convert_extent_item_v0(trans, root, path, owner,
1644 leaf = path->nodes[0];
1645 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1648 BUG_ON(item_size < sizeof(*ei));
1650 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1651 flags = btrfs_extent_flags(leaf, ei);
1653 ptr = (unsigned long)(ei + 1);
1654 end = (unsigned long)ei + item_size;
1656 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1657 ptr += sizeof(struct btrfs_tree_block_info);
1667 iref = (struct btrfs_extent_inline_ref *)ptr;
1668 type = btrfs_extent_inline_ref_type(leaf, iref);
1672 ptr += btrfs_extent_inline_ref_size(type);
1676 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1677 struct btrfs_extent_data_ref *dref;
1678 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1679 if (match_extent_data_ref(leaf, dref, root_objectid,
1684 if (hash_extent_data_ref_item(leaf, dref) <
1685 hash_extent_data_ref(root_objectid, owner, offset))
1689 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1691 if (parent == ref_offset) {
1695 if (ref_offset < parent)
1698 if (root_objectid == ref_offset) {
1702 if (ref_offset < root_objectid)
1706 ptr += btrfs_extent_inline_ref_size(type);
1708 if (err == -ENOENT && insert) {
1709 if (item_size + extra_size >=
1710 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1715 * To add new inline back ref, we have to make sure
1716 * there is no corresponding back ref item.
1717 * For simplicity, we just do not add new inline back
1718 * ref if there is any kind of item for this block
1720 if (find_next_key(path, 0, &key) == 0 &&
1721 key.objectid == bytenr &&
1722 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1727 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1730 path->keep_locks = 0;
1731 btrfs_unlock_up_safe(path, 1);
1737 * helper to add new inline back ref
1739 static noinline_for_stack
1740 void setup_inline_extent_backref(struct btrfs_root *root,
1741 struct btrfs_path *path,
1742 struct btrfs_extent_inline_ref *iref,
1743 u64 parent, u64 root_objectid,
1744 u64 owner, u64 offset, int refs_to_add,
1745 struct btrfs_delayed_extent_op *extent_op)
1747 struct extent_buffer *leaf;
1748 struct btrfs_extent_item *ei;
1751 unsigned long item_offset;
1756 leaf = path->nodes[0];
1757 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1758 item_offset = (unsigned long)iref - (unsigned long)ei;
1760 type = extent_ref_type(parent, owner);
1761 size = btrfs_extent_inline_ref_size(type);
1763 btrfs_extend_item(root, path, size);
1765 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1766 refs = btrfs_extent_refs(leaf, ei);
1767 refs += refs_to_add;
1768 btrfs_set_extent_refs(leaf, ei, refs);
1770 __run_delayed_extent_op(extent_op, leaf, ei);
1772 ptr = (unsigned long)ei + item_offset;
1773 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1774 if (ptr < end - size)
1775 memmove_extent_buffer(leaf, ptr + size, ptr,
1778 iref = (struct btrfs_extent_inline_ref *)ptr;
1779 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1780 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1781 struct btrfs_extent_data_ref *dref;
1782 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1783 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1784 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1785 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1786 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1787 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1788 struct btrfs_shared_data_ref *sref;
1789 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1790 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1791 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1792 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1793 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1795 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1797 btrfs_mark_buffer_dirty(leaf);
1800 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1801 struct btrfs_root *root,
1802 struct btrfs_path *path,
1803 struct btrfs_extent_inline_ref **ref_ret,
1804 u64 bytenr, u64 num_bytes, u64 parent,
1805 u64 root_objectid, u64 owner, u64 offset)
1809 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1810 bytenr, num_bytes, parent,
1811 root_objectid, owner, offset, 0);
1815 btrfs_release_path(path);
1818 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1819 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1822 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1823 root_objectid, owner, offset);
1829 * helper to update/remove inline back ref
1831 static noinline_for_stack
1832 void update_inline_extent_backref(struct btrfs_root *root,
1833 struct btrfs_path *path,
1834 struct btrfs_extent_inline_ref *iref,
1836 struct btrfs_delayed_extent_op *extent_op,
1839 struct extent_buffer *leaf;
1840 struct btrfs_extent_item *ei;
1841 struct btrfs_extent_data_ref *dref = NULL;
1842 struct btrfs_shared_data_ref *sref = NULL;
1850 leaf = path->nodes[0];
1851 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1852 refs = btrfs_extent_refs(leaf, ei);
1853 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1854 refs += refs_to_mod;
1855 btrfs_set_extent_refs(leaf, ei, refs);
1857 __run_delayed_extent_op(extent_op, leaf, ei);
1859 type = btrfs_extent_inline_ref_type(leaf, iref);
1861 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1862 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1863 refs = btrfs_extent_data_ref_count(leaf, dref);
1864 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1865 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1866 refs = btrfs_shared_data_ref_count(leaf, sref);
1869 BUG_ON(refs_to_mod != -1);
1872 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1873 refs += refs_to_mod;
1876 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1877 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1879 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1882 size = btrfs_extent_inline_ref_size(type);
1883 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1884 ptr = (unsigned long)iref;
1885 end = (unsigned long)ei + item_size;
1886 if (ptr + size < end)
1887 memmove_extent_buffer(leaf, ptr, ptr + size,
1890 btrfs_truncate_item(root, path, item_size, 1);
1892 btrfs_mark_buffer_dirty(leaf);
1895 static noinline_for_stack
1896 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1897 struct btrfs_root *root,
1898 struct btrfs_path *path,
1899 u64 bytenr, u64 num_bytes, u64 parent,
1900 u64 root_objectid, u64 owner,
1901 u64 offset, int refs_to_add,
1902 struct btrfs_delayed_extent_op *extent_op)
1904 struct btrfs_extent_inline_ref *iref;
1907 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1908 bytenr, num_bytes, parent,
1909 root_objectid, owner, offset, 1);
1911 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1912 update_inline_extent_backref(root, path, iref,
1913 refs_to_add, extent_op, NULL);
1914 } else if (ret == -ENOENT) {
1915 setup_inline_extent_backref(root, path, iref, parent,
1916 root_objectid, owner, offset,
1917 refs_to_add, extent_op);
1923 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1924 struct btrfs_root *root,
1925 struct btrfs_path *path,
1926 u64 bytenr, u64 parent, u64 root_objectid,
1927 u64 owner, u64 offset, int refs_to_add)
1930 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1931 BUG_ON(refs_to_add != 1);
1932 ret = insert_tree_block_ref(trans, root, path, bytenr,
1933 parent, root_objectid);
1935 ret = insert_extent_data_ref(trans, root, path, bytenr,
1936 parent, root_objectid,
1937 owner, offset, refs_to_add);
1942 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1943 struct btrfs_root *root,
1944 struct btrfs_path *path,
1945 struct btrfs_extent_inline_ref *iref,
1946 int refs_to_drop, int is_data, int *last_ref)
1950 BUG_ON(!is_data && refs_to_drop != 1);
1952 update_inline_extent_backref(root, path, iref,
1953 -refs_to_drop, NULL, last_ref);
1954 } else if (is_data) {
1955 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1959 ret = btrfs_del_item(trans, root, path);
1964 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1965 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1966 u64 *discarded_bytes)
1969 u64 bytes_left, end;
1970 u64 aligned_start = ALIGN(start, 1 << 9);
1972 if (WARN_ON(start != aligned_start)) {
1973 len -= aligned_start - start;
1974 len = round_down(len, 1 << 9);
1975 start = aligned_start;
1978 *discarded_bytes = 0;
1986 /* Skip any superblocks on this device. */
1987 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1988 u64 sb_start = btrfs_sb_offset(j);
1989 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1990 u64 size = sb_start - start;
1992 if (!in_range(sb_start, start, bytes_left) &&
1993 !in_range(sb_end, start, bytes_left) &&
1994 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1998 * Superblock spans beginning of range. Adjust start and
2001 if (sb_start <= start) {
2002 start += sb_end - start;
2007 bytes_left = end - start;
2012 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2015 *discarded_bytes += size;
2016 else if (ret != -EOPNOTSUPP)
2025 bytes_left = end - start;
2029 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2032 *discarded_bytes += bytes_left;
2037 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2038 u64 num_bytes, u64 *actual_bytes)
2041 u64 discarded_bytes = 0;
2042 struct btrfs_bio *bbio = NULL;
2045 /* Tell the block device(s) that the sectors can be discarded */
2046 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2047 bytenr, &num_bytes, &bbio, 0);
2048 /* Error condition is -ENOMEM */
2050 struct btrfs_bio_stripe *stripe = bbio->stripes;
2054 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2056 if (!stripe->dev->can_discard)
2059 ret = btrfs_issue_discard(stripe->dev->bdev,
2064 discarded_bytes += bytes;
2065 else if (ret != -EOPNOTSUPP)
2066 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2069 * Just in case we get back EOPNOTSUPP for some reason,
2070 * just ignore the return value so we don't screw up
2071 * people calling discard_extent.
2075 btrfs_put_bbio(bbio);
2079 *actual_bytes = discarded_bytes;
2082 if (ret == -EOPNOTSUPP)
2087 /* Can return -ENOMEM */
2088 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2089 struct btrfs_root *root,
2090 u64 bytenr, u64 num_bytes, u64 parent,
2091 u64 root_objectid, u64 owner, u64 offset)
2094 struct btrfs_fs_info *fs_info = root->fs_info;
2096 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2097 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2099 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2100 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2102 parent, root_objectid, (int)owner,
2103 BTRFS_ADD_DELAYED_REF, NULL);
2105 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2106 num_bytes, parent, root_objectid,
2108 BTRFS_ADD_DELAYED_REF, NULL);
2113 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2114 struct btrfs_root *root,
2115 struct btrfs_delayed_ref_node *node,
2116 u64 parent, u64 root_objectid,
2117 u64 owner, u64 offset, int refs_to_add,
2118 struct btrfs_delayed_extent_op *extent_op)
2120 struct btrfs_fs_info *fs_info = root->fs_info;
2121 struct btrfs_path *path;
2122 struct extent_buffer *leaf;
2123 struct btrfs_extent_item *item;
2124 struct btrfs_key key;
2125 u64 bytenr = node->bytenr;
2126 u64 num_bytes = node->num_bytes;
2130 path = btrfs_alloc_path();
2135 path->leave_spinning = 1;
2136 /* this will setup the path even if it fails to insert the back ref */
2137 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2138 bytenr, num_bytes, parent,
2139 root_objectid, owner, offset,
2140 refs_to_add, extent_op);
2141 if ((ret < 0 && ret != -EAGAIN) || !ret)
2145 * Ok we had -EAGAIN which means we didn't have space to insert and
2146 * inline extent ref, so just update the reference count and add a
2149 leaf = path->nodes[0];
2150 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2151 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2152 refs = btrfs_extent_refs(leaf, item);
2153 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2155 __run_delayed_extent_op(extent_op, leaf, item);
2157 btrfs_mark_buffer_dirty(leaf);
2158 btrfs_release_path(path);
2161 path->leave_spinning = 1;
2162 /* now insert the actual backref */
2163 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2164 path, bytenr, parent, root_objectid,
2165 owner, offset, refs_to_add);
2167 btrfs_abort_transaction(trans, root, ret);
2169 btrfs_free_path(path);
2173 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2174 struct btrfs_root *root,
2175 struct btrfs_delayed_ref_node *node,
2176 struct btrfs_delayed_extent_op *extent_op,
2177 int insert_reserved)
2180 struct btrfs_delayed_data_ref *ref;
2181 struct btrfs_key ins;
2186 ins.objectid = node->bytenr;
2187 ins.offset = node->num_bytes;
2188 ins.type = BTRFS_EXTENT_ITEM_KEY;
2190 ref = btrfs_delayed_node_to_data_ref(node);
2191 trace_run_delayed_data_ref(node, ref, node->action);
2193 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2194 parent = ref->parent;
2195 ref_root = ref->root;
2197 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2199 flags |= extent_op->flags_to_set;
2200 ret = alloc_reserved_file_extent(trans, root,
2201 parent, ref_root, flags,
2202 ref->objectid, ref->offset,
2203 &ins, node->ref_mod);
2204 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2205 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2206 ref_root, ref->objectid,
2207 ref->offset, node->ref_mod,
2209 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2210 ret = __btrfs_free_extent(trans, root, node, parent,
2211 ref_root, ref->objectid,
2212 ref->offset, node->ref_mod,
2220 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2221 struct extent_buffer *leaf,
2222 struct btrfs_extent_item *ei)
2224 u64 flags = btrfs_extent_flags(leaf, ei);
2225 if (extent_op->update_flags) {
2226 flags |= extent_op->flags_to_set;
2227 btrfs_set_extent_flags(leaf, ei, flags);
2230 if (extent_op->update_key) {
2231 struct btrfs_tree_block_info *bi;
2232 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2233 bi = (struct btrfs_tree_block_info *)(ei + 1);
2234 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2238 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2239 struct btrfs_root *root,
2240 struct btrfs_delayed_ref_node *node,
2241 struct btrfs_delayed_extent_op *extent_op)
2243 struct btrfs_key key;
2244 struct btrfs_path *path;
2245 struct btrfs_extent_item *ei;
2246 struct extent_buffer *leaf;
2250 int metadata = !extent_op->is_data;
2255 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2258 path = btrfs_alloc_path();
2262 key.objectid = node->bytenr;
2265 key.type = BTRFS_METADATA_ITEM_KEY;
2266 key.offset = extent_op->level;
2268 key.type = BTRFS_EXTENT_ITEM_KEY;
2269 key.offset = node->num_bytes;
2274 path->leave_spinning = 1;
2275 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2283 if (path->slots[0] > 0) {
2285 btrfs_item_key_to_cpu(path->nodes[0], &key,
2287 if (key.objectid == node->bytenr &&
2288 key.type == BTRFS_EXTENT_ITEM_KEY &&
2289 key.offset == node->num_bytes)
2293 btrfs_release_path(path);
2296 key.objectid = node->bytenr;
2297 key.offset = node->num_bytes;
2298 key.type = BTRFS_EXTENT_ITEM_KEY;
2307 leaf = path->nodes[0];
2308 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2309 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2310 if (item_size < sizeof(*ei)) {
2311 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2317 leaf = path->nodes[0];
2318 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2321 BUG_ON(item_size < sizeof(*ei));
2322 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2323 __run_delayed_extent_op(extent_op, leaf, ei);
2325 btrfs_mark_buffer_dirty(leaf);
2327 btrfs_free_path(path);
2331 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2332 struct btrfs_root *root,
2333 struct btrfs_delayed_ref_node *node,
2334 struct btrfs_delayed_extent_op *extent_op,
2335 int insert_reserved)
2338 struct btrfs_delayed_tree_ref *ref;
2339 struct btrfs_key ins;
2342 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2345 ref = btrfs_delayed_node_to_tree_ref(node);
2346 trace_run_delayed_tree_ref(node, ref, node->action);
2348 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2349 parent = ref->parent;
2350 ref_root = ref->root;
2352 ins.objectid = node->bytenr;
2353 if (skinny_metadata) {
2354 ins.offset = ref->level;
2355 ins.type = BTRFS_METADATA_ITEM_KEY;
2357 ins.offset = node->num_bytes;
2358 ins.type = BTRFS_EXTENT_ITEM_KEY;
2361 BUG_ON(node->ref_mod != 1);
2362 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2363 BUG_ON(!extent_op || !extent_op->update_flags);
2364 ret = alloc_reserved_tree_block(trans, root,
2366 extent_op->flags_to_set,
2369 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2370 ret = __btrfs_inc_extent_ref(trans, root, node,
2374 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2375 ret = __btrfs_free_extent(trans, root, node,
2377 ref->level, 0, 1, extent_op);
2384 /* helper function to actually process a single delayed ref entry */
2385 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2386 struct btrfs_root *root,
2387 struct btrfs_delayed_ref_node *node,
2388 struct btrfs_delayed_extent_op *extent_op,
2389 int insert_reserved)
2393 if (trans->aborted) {
2394 if (insert_reserved)
2395 btrfs_pin_extent(root, node->bytenr,
2396 node->num_bytes, 1);
2400 if (btrfs_delayed_ref_is_head(node)) {
2401 struct btrfs_delayed_ref_head *head;
2403 * we've hit the end of the chain and we were supposed
2404 * to insert this extent into the tree. But, it got
2405 * deleted before we ever needed to insert it, so all
2406 * we have to do is clean up the accounting
2409 head = btrfs_delayed_node_to_head(node);
2410 trace_run_delayed_ref_head(node, head, node->action);
2412 if (insert_reserved) {
2413 btrfs_pin_extent(root, node->bytenr,
2414 node->num_bytes, 1);
2415 if (head->is_data) {
2416 ret = btrfs_del_csums(trans, root,
2422 /* Also free its reserved qgroup space */
2423 btrfs_qgroup_free_delayed_ref(root->fs_info,
2424 head->qgroup_ref_root,
2425 head->qgroup_reserved);
2429 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2430 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2431 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2433 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2434 node->type == BTRFS_SHARED_DATA_REF_KEY)
2435 ret = run_delayed_data_ref(trans, root, node, extent_op,
2442 static inline struct btrfs_delayed_ref_node *
2443 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2445 struct btrfs_delayed_ref_node *ref;
2447 if (list_empty(&head->ref_list))
2451 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2452 * This is to prevent a ref count from going down to zero, which deletes
2453 * the extent item from the extent tree, when there still are references
2454 * to add, which would fail because they would not find the extent item.
2456 list_for_each_entry(ref, &head->ref_list, list) {
2457 if (ref->action == BTRFS_ADD_DELAYED_REF)
2461 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2466 * Returns 0 on success or if called with an already aborted transaction.
2467 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2469 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2470 struct btrfs_root *root,
2473 struct btrfs_delayed_ref_root *delayed_refs;
2474 struct btrfs_delayed_ref_node *ref;
2475 struct btrfs_delayed_ref_head *locked_ref = NULL;
2476 struct btrfs_delayed_extent_op *extent_op;
2477 struct btrfs_fs_info *fs_info = root->fs_info;
2478 ktime_t start = ktime_get();
2480 unsigned long count = 0;
2481 unsigned long actual_count = 0;
2482 int must_insert_reserved = 0;
2484 delayed_refs = &trans->transaction->delayed_refs;
2490 spin_lock(&delayed_refs->lock);
2491 locked_ref = btrfs_select_ref_head(trans);
2493 spin_unlock(&delayed_refs->lock);
2497 /* grab the lock that says we are going to process
2498 * all the refs for this head */
2499 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2500 spin_unlock(&delayed_refs->lock);
2502 * we may have dropped the spin lock to get the head
2503 * mutex lock, and that might have given someone else
2504 * time to free the head. If that's true, it has been
2505 * removed from our list and we can move on.
2507 if (ret == -EAGAIN) {
2515 * We need to try and merge add/drops of the same ref since we
2516 * can run into issues with relocate dropping the implicit ref
2517 * and then it being added back again before the drop can
2518 * finish. If we merged anything we need to re-loop so we can
2520 * Or we can get node references of the same type that weren't
2521 * merged when created due to bumps in the tree mod seq, and
2522 * we need to merge them to prevent adding an inline extent
2523 * backref before dropping it (triggering a BUG_ON at
2524 * insert_inline_extent_backref()).
2526 spin_lock(&locked_ref->lock);
2527 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2531 * locked_ref is the head node, so we have to go one
2532 * node back for any delayed ref updates
2534 ref = select_delayed_ref(locked_ref);
2536 if (ref && ref->seq &&
2537 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2538 spin_unlock(&locked_ref->lock);
2539 btrfs_delayed_ref_unlock(locked_ref);
2540 spin_lock(&delayed_refs->lock);
2541 locked_ref->processing = 0;
2542 delayed_refs->num_heads_ready++;
2543 spin_unlock(&delayed_refs->lock);
2551 * record the must insert reserved flag before we
2552 * drop the spin lock.
2554 must_insert_reserved = locked_ref->must_insert_reserved;
2555 locked_ref->must_insert_reserved = 0;
2557 extent_op = locked_ref->extent_op;
2558 locked_ref->extent_op = NULL;
2563 /* All delayed refs have been processed, Go ahead
2564 * and send the head node to run_one_delayed_ref,
2565 * so that any accounting fixes can happen
2567 ref = &locked_ref->node;
2569 if (extent_op && must_insert_reserved) {
2570 btrfs_free_delayed_extent_op(extent_op);
2575 spin_unlock(&locked_ref->lock);
2576 ret = run_delayed_extent_op(trans, root,
2578 btrfs_free_delayed_extent_op(extent_op);
2582 * Need to reset must_insert_reserved if
2583 * there was an error so the abort stuff
2584 * can cleanup the reserved space
2587 if (must_insert_reserved)
2588 locked_ref->must_insert_reserved = 1;
2589 locked_ref->processing = 0;
2590 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2591 btrfs_delayed_ref_unlock(locked_ref);
2598 * Need to drop our head ref lock and re-aqcuire the
2599 * delayed ref lock and then re-check to make sure
2602 spin_unlock(&locked_ref->lock);
2603 spin_lock(&delayed_refs->lock);
2604 spin_lock(&locked_ref->lock);
2605 if (!list_empty(&locked_ref->ref_list) ||
2606 locked_ref->extent_op) {
2607 spin_unlock(&locked_ref->lock);
2608 spin_unlock(&delayed_refs->lock);
2612 delayed_refs->num_heads--;
2613 rb_erase(&locked_ref->href_node,
2614 &delayed_refs->href_root);
2615 spin_unlock(&delayed_refs->lock);
2619 list_del(&ref->list);
2621 atomic_dec(&delayed_refs->num_entries);
2623 if (!btrfs_delayed_ref_is_head(ref)) {
2625 * when we play the delayed ref, also correct the
2628 switch (ref->action) {
2629 case BTRFS_ADD_DELAYED_REF:
2630 case BTRFS_ADD_DELAYED_EXTENT:
2631 locked_ref->node.ref_mod -= ref->ref_mod;
2633 case BTRFS_DROP_DELAYED_REF:
2634 locked_ref->node.ref_mod += ref->ref_mod;
2640 spin_unlock(&locked_ref->lock);
2642 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2643 must_insert_reserved);
2645 btrfs_free_delayed_extent_op(extent_op);
2647 locked_ref->processing = 0;
2648 btrfs_delayed_ref_unlock(locked_ref);
2649 btrfs_put_delayed_ref(ref);
2650 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2655 * If this node is a head, that means all the refs in this head
2656 * have been dealt with, and we will pick the next head to deal
2657 * with, so we must unlock the head and drop it from the cluster
2658 * list before we release it.
2660 if (btrfs_delayed_ref_is_head(ref)) {
2661 if (locked_ref->is_data &&
2662 locked_ref->total_ref_mod < 0) {
2663 spin_lock(&delayed_refs->lock);
2664 delayed_refs->pending_csums -= ref->num_bytes;
2665 spin_unlock(&delayed_refs->lock);
2667 btrfs_delayed_ref_unlock(locked_ref);
2670 btrfs_put_delayed_ref(ref);
2676 * We don't want to include ref heads since we can have empty ref heads
2677 * and those will drastically skew our runtime down since we just do
2678 * accounting, no actual extent tree updates.
2680 if (actual_count > 0) {
2681 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2685 * We weigh the current average higher than our current runtime
2686 * to avoid large swings in the average.
2688 spin_lock(&delayed_refs->lock);
2689 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2690 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2691 spin_unlock(&delayed_refs->lock);
2696 #ifdef SCRAMBLE_DELAYED_REFS
2698 * Normally delayed refs get processed in ascending bytenr order. This
2699 * correlates in most cases to the order added. To expose dependencies on this
2700 * order, we start to process the tree in the middle instead of the beginning
2702 static u64 find_middle(struct rb_root *root)
2704 struct rb_node *n = root->rb_node;
2705 struct btrfs_delayed_ref_node *entry;
2708 u64 first = 0, last = 0;
2712 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2713 first = entry->bytenr;
2717 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2718 last = entry->bytenr;
2723 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2724 WARN_ON(!entry->in_tree);
2726 middle = entry->bytenr;
2739 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2743 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2744 sizeof(struct btrfs_extent_inline_ref));
2745 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2746 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2749 * We don't ever fill up leaves all the way so multiply by 2 just to be
2750 * closer to what we're really going to want to ouse.
2752 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2756 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2757 * would require to store the csums for that many bytes.
2759 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2762 u64 num_csums_per_leaf;
2765 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2766 num_csums_per_leaf = div64_u64(csum_size,
2767 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2768 num_csums = div64_u64(csum_bytes, root->sectorsize);
2769 num_csums += num_csums_per_leaf - 1;
2770 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2774 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2775 struct btrfs_root *root)
2777 struct btrfs_block_rsv *global_rsv;
2778 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2779 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2780 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2781 u64 num_bytes, num_dirty_bgs_bytes;
2784 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2785 num_heads = heads_to_leaves(root, num_heads);
2787 num_bytes += (num_heads - 1) * root->nodesize;
2789 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2790 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2792 global_rsv = &root->fs_info->global_block_rsv;
2795 * If we can't allocate any more chunks lets make sure we have _lots_ of
2796 * wiggle room since running delayed refs can create more delayed refs.
2798 if (global_rsv->space_info->full) {
2799 num_dirty_bgs_bytes <<= 1;
2803 spin_lock(&global_rsv->lock);
2804 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2806 spin_unlock(&global_rsv->lock);
2810 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2811 struct btrfs_root *root)
2813 struct btrfs_fs_info *fs_info = root->fs_info;
2815 atomic_read(&trans->transaction->delayed_refs.num_entries);
2820 avg_runtime = fs_info->avg_delayed_ref_runtime;
2821 val = num_entries * avg_runtime;
2822 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2824 if (val >= NSEC_PER_SEC / 2)
2827 return btrfs_check_space_for_delayed_refs(trans, root);
2830 struct async_delayed_refs {
2831 struct btrfs_root *root;
2835 struct completion wait;
2836 struct btrfs_work work;
2839 static void delayed_ref_async_start(struct btrfs_work *work)
2841 struct async_delayed_refs *async;
2842 struct btrfs_trans_handle *trans;
2845 async = container_of(work, struct async_delayed_refs, work);
2847 trans = btrfs_join_transaction(async->root);
2848 if (IS_ERR(trans)) {
2849 async->error = PTR_ERR(trans);
2854 * trans->sync means that when we call end_transaciton, we won't
2855 * wait on delayed refs
2858 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2862 ret = btrfs_end_transaction(trans, async->root);
2863 if (ret && !async->error)
2867 complete(&async->wait);
2872 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2873 unsigned long count, int wait)
2875 struct async_delayed_refs *async;
2878 async = kmalloc(sizeof(*async), GFP_NOFS);
2882 async->root = root->fs_info->tree_root;
2883 async->count = count;
2889 init_completion(&async->wait);
2891 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2892 delayed_ref_async_start, NULL, NULL);
2894 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2897 wait_for_completion(&async->wait);
2906 * this starts processing the delayed reference count updates and
2907 * extent insertions we have queued up so far. count can be
2908 * 0, which means to process everything in the tree at the start
2909 * of the run (but not newly added entries), or it can be some target
2910 * number you'd like to process.
2912 * Returns 0 on success or if called with an aborted transaction
2913 * Returns <0 on error and aborts the transaction
2915 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2916 struct btrfs_root *root, unsigned long count)
2918 struct rb_node *node;
2919 struct btrfs_delayed_ref_root *delayed_refs;
2920 struct btrfs_delayed_ref_head *head;
2922 int run_all = count == (unsigned long)-1;
2923 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2925 /* We'll clean this up in btrfs_cleanup_transaction */
2929 if (root == root->fs_info->extent_root)
2930 root = root->fs_info->tree_root;
2932 delayed_refs = &trans->transaction->delayed_refs;
2934 count = atomic_read(&delayed_refs->num_entries) * 2;
2937 #ifdef SCRAMBLE_DELAYED_REFS
2938 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2940 trans->can_flush_pending_bgs = false;
2941 ret = __btrfs_run_delayed_refs(trans, root, count);
2943 btrfs_abort_transaction(trans, root, ret);
2948 if (!list_empty(&trans->new_bgs))
2949 btrfs_create_pending_block_groups(trans, root);
2951 spin_lock(&delayed_refs->lock);
2952 node = rb_first(&delayed_refs->href_root);
2954 spin_unlock(&delayed_refs->lock);
2957 count = (unsigned long)-1;
2960 head = rb_entry(node, struct btrfs_delayed_ref_head,
2962 if (btrfs_delayed_ref_is_head(&head->node)) {
2963 struct btrfs_delayed_ref_node *ref;
2966 atomic_inc(&ref->refs);
2968 spin_unlock(&delayed_refs->lock);
2970 * Mutex was contended, block until it's
2971 * released and try again
2973 mutex_lock(&head->mutex);
2974 mutex_unlock(&head->mutex);
2976 btrfs_put_delayed_ref(ref);
2982 node = rb_next(node);
2984 spin_unlock(&delayed_refs->lock);
2989 assert_qgroups_uptodate(trans);
2990 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2994 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2995 struct btrfs_root *root,
2996 u64 bytenr, u64 num_bytes, u64 flags,
2997 int level, int is_data)
2999 struct btrfs_delayed_extent_op *extent_op;
3002 extent_op = btrfs_alloc_delayed_extent_op();
3006 extent_op->flags_to_set = flags;
3007 extent_op->update_flags = 1;
3008 extent_op->update_key = 0;
3009 extent_op->is_data = is_data ? 1 : 0;
3010 extent_op->level = level;
3012 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3013 num_bytes, extent_op);
3015 btrfs_free_delayed_extent_op(extent_op);
3019 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3020 struct btrfs_root *root,
3021 struct btrfs_path *path,
3022 u64 objectid, u64 offset, u64 bytenr)
3024 struct btrfs_delayed_ref_head *head;
3025 struct btrfs_delayed_ref_node *ref;
3026 struct btrfs_delayed_data_ref *data_ref;
3027 struct btrfs_delayed_ref_root *delayed_refs;
3030 delayed_refs = &trans->transaction->delayed_refs;
3031 spin_lock(&delayed_refs->lock);
3032 head = btrfs_find_delayed_ref_head(trans, bytenr);
3034 spin_unlock(&delayed_refs->lock);
3038 if (!mutex_trylock(&head->mutex)) {
3039 atomic_inc(&head->node.refs);
3040 spin_unlock(&delayed_refs->lock);
3042 btrfs_release_path(path);
3045 * Mutex was contended, block until it's released and let
3048 mutex_lock(&head->mutex);
3049 mutex_unlock(&head->mutex);
3050 btrfs_put_delayed_ref(&head->node);
3053 spin_unlock(&delayed_refs->lock);
3055 spin_lock(&head->lock);
3056 list_for_each_entry(ref, &head->ref_list, list) {
3057 /* If it's a shared ref we know a cross reference exists */
3058 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3063 data_ref = btrfs_delayed_node_to_data_ref(ref);
3066 * If our ref doesn't match the one we're currently looking at
3067 * then we have a cross reference.
3069 if (data_ref->root != root->root_key.objectid ||
3070 data_ref->objectid != objectid ||
3071 data_ref->offset != offset) {
3076 spin_unlock(&head->lock);
3077 mutex_unlock(&head->mutex);
3081 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3082 struct btrfs_root *root,
3083 struct btrfs_path *path,
3084 u64 objectid, u64 offset, u64 bytenr)
3086 struct btrfs_root *extent_root = root->fs_info->extent_root;
3087 struct extent_buffer *leaf;
3088 struct btrfs_extent_data_ref *ref;
3089 struct btrfs_extent_inline_ref *iref;
3090 struct btrfs_extent_item *ei;
3091 struct btrfs_key key;
3095 key.objectid = bytenr;
3096 key.offset = (u64)-1;
3097 key.type = BTRFS_EXTENT_ITEM_KEY;
3099 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3102 BUG_ON(ret == 0); /* Corruption */
3105 if (path->slots[0] == 0)
3109 leaf = path->nodes[0];
3110 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3112 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3116 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3117 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3118 if (item_size < sizeof(*ei)) {
3119 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3123 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3125 if (item_size != sizeof(*ei) +
3126 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3129 if (btrfs_extent_generation(leaf, ei) <=
3130 btrfs_root_last_snapshot(&root->root_item))
3133 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3134 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3135 BTRFS_EXTENT_DATA_REF_KEY)
3138 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3139 if (btrfs_extent_refs(leaf, ei) !=
3140 btrfs_extent_data_ref_count(leaf, ref) ||
3141 btrfs_extent_data_ref_root(leaf, ref) !=
3142 root->root_key.objectid ||
3143 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3144 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3152 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3153 struct btrfs_root *root,
3154 u64 objectid, u64 offset, u64 bytenr)
3156 struct btrfs_path *path;
3160 path = btrfs_alloc_path();
3165 ret = check_committed_ref(trans, root, path, objectid,
3167 if (ret && ret != -ENOENT)
3170 ret2 = check_delayed_ref(trans, root, path, objectid,
3172 } while (ret2 == -EAGAIN);
3174 if (ret2 && ret2 != -ENOENT) {
3179 if (ret != -ENOENT || ret2 != -ENOENT)
3182 btrfs_free_path(path);
3183 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3188 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3189 struct btrfs_root *root,
3190 struct extent_buffer *buf,
3191 int full_backref, int inc)
3198 struct btrfs_key key;
3199 struct btrfs_file_extent_item *fi;
3203 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3204 u64, u64, u64, u64, u64, u64);
3207 if (btrfs_test_is_dummy_root(root))
3210 ref_root = btrfs_header_owner(buf);
3211 nritems = btrfs_header_nritems(buf);
3212 level = btrfs_header_level(buf);
3214 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3218 process_func = btrfs_inc_extent_ref;
3220 process_func = btrfs_free_extent;
3223 parent = buf->start;
3227 for (i = 0; i < nritems; i++) {
3229 btrfs_item_key_to_cpu(buf, &key, i);
3230 if (key.type != BTRFS_EXTENT_DATA_KEY)
3232 fi = btrfs_item_ptr(buf, i,
3233 struct btrfs_file_extent_item);
3234 if (btrfs_file_extent_type(buf, fi) ==
3235 BTRFS_FILE_EXTENT_INLINE)
3237 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3241 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3242 key.offset -= btrfs_file_extent_offset(buf, fi);
3243 ret = process_func(trans, root, bytenr, num_bytes,
3244 parent, ref_root, key.objectid,
3249 bytenr = btrfs_node_blockptr(buf, i);
3250 num_bytes = root->nodesize;
3251 ret = process_func(trans, root, bytenr, num_bytes,
3252 parent, ref_root, level - 1, 0);
3262 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3263 struct extent_buffer *buf, int full_backref)
3265 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3268 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3269 struct extent_buffer *buf, int full_backref)
3271 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3274 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3275 struct btrfs_root *root,
3276 struct btrfs_path *path,
3277 struct btrfs_block_group_cache *cache)
3280 struct btrfs_root *extent_root = root->fs_info->extent_root;
3282 struct extent_buffer *leaf;
3284 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3291 leaf = path->nodes[0];
3292 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3293 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3294 btrfs_mark_buffer_dirty(leaf);
3296 btrfs_release_path(path);
3301 static struct btrfs_block_group_cache *
3302 next_block_group(struct btrfs_root *root,
3303 struct btrfs_block_group_cache *cache)
3305 struct rb_node *node;
3307 spin_lock(&root->fs_info->block_group_cache_lock);
3309 /* If our block group was removed, we need a full search. */
3310 if (RB_EMPTY_NODE(&cache->cache_node)) {
3311 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3313 spin_unlock(&root->fs_info->block_group_cache_lock);
3314 btrfs_put_block_group(cache);
3315 cache = btrfs_lookup_first_block_group(root->fs_info,
3319 node = rb_next(&cache->cache_node);
3320 btrfs_put_block_group(cache);
3322 cache = rb_entry(node, struct btrfs_block_group_cache,
3324 btrfs_get_block_group(cache);
3327 spin_unlock(&root->fs_info->block_group_cache_lock);
3331 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3332 struct btrfs_trans_handle *trans,
3333 struct btrfs_path *path)
3335 struct btrfs_root *root = block_group->fs_info->tree_root;
3336 struct inode *inode = NULL;
3338 int dcs = BTRFS_DC_ERROR;
3344 * If this block group is smaller than 100 megs don't bother caching the
3347 if (block_group->key.offset < (100 * 1024 * 1024)) {
3348 spin_lock(&block_group->lock);
3349 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3350 spin_unlock(&block_group->lock);
3357 inode = lookup_free_space_inode(root, block_group, path);
3358 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3359 ret = PTR_ERR(inode);
3360 btrfs_release_path(path);
3364 if (IS_ERR(inode)) {
3368 if (block_group->ro)
3371 ret = create_free_space_inode(root, trans, block_group, path);
3377 /* We've already setup this transaction, go ahead and exit */
3378 if (block_group->cache_generation == trans->transid &&
3379 i_size_read(inode)) {
3380 dcs = BTRFS_DC_SETUP;
3385 * We want to set the generation to 0, that way if anything goes wrong
3386 * from here on out we know not to trust this cache when we load up next
3389 BTRFS_I(inode)->generation = 0;
3390 ret = btrfs_update_inode(trans, root, inode);
3393 * So theoretically we could recover from this, simply set the
3394 * super cache generation to 0 so we know to invalidate the
3395 * cache, but then we'd have to keep track of the block groups
3396 * that fail this way so we know we _have_ to reset this cache
3397 * before the next commit or risk reading stale cache. So to
3398 * limit our exposure to horrible edge cases lets just abort the
3399 * transaction, this only happens in really bad situations
3402 btrfs_abort_transaction(trans, root, ret);
3407 if (i_size_read(inode) > 0) {
3408 ret = btrfs_check_trunc_cache_free_space(root,
3409 &root->fs_info->global_block_rsv);
3413 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3418 spin_lock(&block_group->lock);
3419 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3420 !btrfs_test_opt(root, SPACE_CACHE)) {
3422 * don't bother trying to write stuff out _if_
3423 * a) we're not cached,
3424 * b) we're with nospace_cache mount option.
3426 dcs = BTRFS_DC_WRITTEN;
3427 spin_unlock(&block_group->lock);
3430 spin_unlock(&block_group->lock);
3433 * We hit an ENOSPC when setting up the cache in this transaction, just
3434 * skip doing the setup, we've already cleared the cache so we're safe.
3436 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3442 * Try to preallocate enough space based on how big the block group is.
3443 * Keep in mind this has to include any pinned space which could end up
3444 * taking up quite a bit since it's not folded into the other space
3447 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3452 num_pages *= PAGE_CACHE_SIZE;
3454 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3458 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3459 num_pages, num_pages,
3462 * Our cache requires contiguous chunks so that we don't modify a bunch
3463 * of metadata or split extents when writing the cache out, which means
3464 * we can enospc if we are heavily fragmented in addition to just normal
3465 * out of space conditions. So if we hit this just skip setting up any
3466 * other block groups for this transaction, maybe we'll unpin enough
3467 * space the next time around.
3470 dcs = BTRFS_DC_SETUP;
3471 else if (ret == -ENOSPC)
3472 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3473 btrfs_free_reserved_data_space(inode, 0, num_pages);
3478 btrfs_release_path(path);
3480 spin_lock(&block_group->lock);
3481 if (!ret && dcs == BTRFS_DC_SETUP)
3482 block_group->cache_generation = trans->transid;
3483 block_group->disk_cache_state = dcs;
3484 spin_unlock(&block_group->lock);
3489 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3490 struct btrfs_root *root)
3492 struct btrfs_block_group_cache *cache, *tmp;
3493 struct btrfs_transaction *cur_trans = trans->transaction;
3494 struct btrfs_path *path;
3496 if (list_empty(&cur_trans->dirty_bgs) ||
3497 !btrfs_test_opt(root, SPACE_CACHE))
3500 path = btrfs_alloc_path();
3504 /* Could add new block groups, use _safe just in case */
3505 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3507 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3508 cache_save_setup(cache, trans, path);
3511 btrfs_free_path(path);
3516 * transaction commit does final block group cache writeback during a
3517 * critical section where nothing is allowed to change the FS. This is
3518 * required in order for the cache to actually match the block group,
3519 * but can introduce a lot of latency into the commit.
3521 * So, btrfs_start_dirty_block_groups is here to kick off block group
3522 * cache IO. There's a chance we'll have to redo some of it if the
3523 * block group changes again during the commit, but it greatly reduces
3524 * the commit latency by getting rid of the easy block groups while
3525 * we're still allowing others to join the commit.
3527 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3528 struct btrfs_root *root)
3530 struct btrfs_block_group_cache *cache;
3531 struct btrfs_transaction *cur_trans = trans->transaction;
3534 struct btrfs_path *path = NULL;
3536 struct list_head *io = &cur_trans->io_bgs;
3537 int num_started = 0;
3540 spin_lock(&cur_trans->dirty_bgs_lock);
3541 if (list_empty(&cur_trans->dirty_bgs)) {
3542 spin_unlock(&cur_trans->dirty_bgs_lock);
3545 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3546 spin_unlock(&cur_trans->dirty_bgs_lock);
3550 * make sure all the block groups on our dirty list actually
3553 btrfs_create_pending_block_groups(trans, root);
3556 path = btrfs_alloc_path();
3562 * cache_write_mutex is here only to save us from balance or automatic
3563 * removal of empty block groups deleting this block group while we are
3564 * writing out the cache
3566 mutex_lock(&trans->transaction->cache_write_mutex);
3567 while (!list_empty(&dirty)) {
3568 cache = list_first_entry(&dirty,
3569 struct btrfs_block_group_cache,
3572 * this can happen if something re-dirties a block
3573 * group that is already under IO. Just wait for it to
3574 * finish and then do it all again
3576 if (!list_empty(&cache->io_list)) {
3577 list_del_init(&cache->io_list);
3578 btrfs_wait_cache_io(root, trans, cache,
3579 &cache->io_ctl, path,
3580 cache->key.objectid);
3581 btrfs_put_block_group(cache);
3586 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3587 * if it should update the cache_state. Don't delete
3588 * until after we wait.
3590 * Since we're not running in the commit critical section
3591 * we need the dirty_bgs_lock to protect from update_block_group
3593 spin_lock(&cur_trans->dirty_bgs_lock);
3594 list_del_init(&cache->dirty_list);
3595 spin_unlock(&cur_trans->dirty_bgs_lock);
3599 cache_save_setup(cache, trans, path);
3601 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3602 cache->io_ctl.inode = NULL;
3603 ret = btrfs_write_out_cache(root, trans, cache, path);
3604 if (ret == 0 && cache->io_ctl.inode) {
3609 * the cache_write_mutex is protecting
3612 list_add_tail(&cache->io_list, io);
3615 * if we failed to write the cache, the
3616 * generation will be bad and life goes on
3622 ret = write_one_cache_group(trans, root, path, cache);
3624 * Our block group might still be attached to the list
3625 * of new block groups in the transaction handle of some
3626 * other task (struct btrfs_trans_handle->new_bgs). This
3627 * means its block group item isn't yet in the extent
3628 * tree. If this happens ignore the error, as we will
3629 * try again later in the critical section of the
3630 * transaction commit.
3632 if (ret == -ENOENT) {
3634 spin_lock(&cur_trans->dirty_bgs_lock);
3635 if (list_empty(&cache->dirty_list)) {
3636 list_add_tail(&cache->dirty_list,
3637 &cur_trans->dirty_bgs);
3638 btrfs_get_block_group(cache);
3640 spin_unlock(&cur_trans->dirty_bgs_lock);
3642 btrfs_abort_transaction(trans, root, ret);
3646 /* if its not on the io list, we need to put the block group */
3648 btrfs_put_block_group(cache);
3654 * Avoid blocking other tasks for too long. It might even save
3655 * us from writing caches for block groups that are going to be
3658 mutex_unlock(&trans->transaction->cache_write_mutex);
3659 mutex_lock(&trans->transaction->cache_write_mutex);
3661 mutex_unlock(&trans->transaction->cache_write_mutex);
3664 * go through delayed refs for all the stuff we've just kicked off
3665 * and then loop back (just once)
3667 ret = btrfs_run_delayed_refs(trans, root, 0);
3668 if (!ret && loops == 0) {
3670 spin_lock(&cur_trans->dirty_bgs_lock);
3671 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3673 * dirty_bgs_lock protects us from concurrent block group
3674 * deletes too (not just cache_write_mutex).
3676 if (!list_empty(&dirty)) {
3677 spin_unlock(&cur_trans->dirty_bgs_lock);
3680 spin_unlock(&cur_trans->dirty_bgs_lock);
3683 btrfs_free_path(path);
3687 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3688 struct btrfs_root *root)
3690 struct btrfs_block_group_cache *cache;
3691 struct btrfs_transaction *cur_trans = trans->transaction;
3694 struct btrfs_path *path;
3695 struct list_head *io = &cur_trans->io_bgs;
3696 int num_started = 0;
3698 path = btrfs_alloc_path();
3703 * Even though we are in the critical section of the transaction commit,
3704 * we can still have concurrent tasks adding elements to this
3705 * transaction's list of dirty block groups. These tasks correspond to
3706 * endio free space workers started when writeback finishes for a
3707 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3708 * allocate new block groups as a result of COWing nodes of the root
3709 * tree when updating the free space inode. The writeback for the space
3710 * caches is triggered by an earlier call to
3711 * btrfs_start_dirty_block_groups() and iterations of the following
3713 * Also we want to do the cache_save_setup first and then run the
3714 * delayed refs to make sure we have the best chance at doing this all
3717 spin_lock(&cur_trans->dirty_bgs_lock);
3718 while (!list_empty(&cur_trans->dirty_bgs)) {
3719 cache = list_first_entry(&cur_trans->dirty_bgs,
3720 struct btrfs_block_group_cache,
3724 * this can happen if cache_save_setup re-dirties a block
3725 * group that is already under IO. Just wait for it to
3726 * finish and then do it all again
3728 if (!list_empty(&cache->io_list)) {
3729 spin_unlock(&cur_trans->dirty_bgs_lock);
3730 list_del_init(&cache->io_list);
3731 btrfs_wait_cache_io(root, trans, cache,
3732 &cache->io_ctl, path,
3733 cache->key.objectid);
3734 btrfs_put_block_group(cache);
3735 spin_lock(&cur_trans->dirty_bgs_lock);
3739 * don't remove from the dirty list until after we've waited
3742 list_del_init(&cache->dirty_list);
3743 spin_unlock(&cur_trans->dirty_bgs_lock);
3746 cache_save_setup(cache, trans, path);
3749 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3751 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3752 cache->io_ctl.inode = NULL;
3753 ret = btrfs_write_out_cache(root, trans, cache, path);
3754 if (ret == 0 && cache->io_ctl.inode) {
3757 list_add_tail(&cache->io_list, io);
3760 * if we failed to write the cache, the
3761 * generation will be bad and life goes on
3767 ret = write_one_cache_group(trans, root, path, cache);
3769 btrfs_abort_transaction(trans, root, ret);
3772 /* if its not on the io list, we need to put the block group */
3774 btrfs_put_block_group(cache);
3775 spin_lock(&cur_trans->dirty_bgs_lock);
3777 spin_unlock(&cur_trans->dirty_bgs_lock);
3779 while (!list_empty(io)) {
3780 cache = list_first_entry(io, struct btrfs_block_group_cache,
3782 list_del_init(&cache->io_list);
3783 btrfs_wait_cache_io(root, trans, cache,
3784 &cache->io_ctl, path, cache->key.objectid);
3785 btrfs_put_block_group(cache);
3788 btrfs_free_path(path);
3792 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3794 struct btrfs_block_group_cache *block_group;
3797 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3798 if (!block_group || block_group->ro)
3801 btrfs_put_block_group(block_group);
3805 static const char *alloc_name(u64 flags)
3808 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3810 case BTRFS_BLOCK_GROUP_METADATA:
3812 case BTRFS_BLOCK_GROUP_DATA:
3814 case BTRFS_BLOCK_GROUP_SYSTEM:
3818 return "invalid-combination";
3822 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3823 u64 total_bytes, u64 bytes_used,
3824 struct btrfs_space_info **space_info)
3826 struct btrfs_space_info *found;
3831 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3832 BTRFS_BLOCK_GROUP_RAID10))
3837 found = __find_space_info(info, flags);
3839 spin_lock(&found->lock);
3840 found->total_bytes += total_bytes;
3841 found->disk_total += total_bytes * factor;
3842 found->bytes_used += bytes_used;
3843 found->disk_used += bytes_used * factor;
3844 if (total_bytes > 0)
3846 spin_unlock(&found->lock);
3847 *space_info = found;
3850 found = kzalloc(sizeof(*found), GFP_NOFS);
3854 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3860 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3861 INIT_LIST_HEAD(&found->block_groups[i]);
3862 init_rwsem(&found->groups_sem);
3863 spin_lock_init(&found->lock);
3864 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3865 found->total_bytes = total_bytes;
3866 found->disk_total = total_bytes * factor;
3867 found->bytes_used = bytes_used;
3868 found->disk_used = bytes_used * factor;
3869 found->bytes_pinned = 0;
3870 found->bytes_reserved = 0;
3871 found->bytes_readonly = 0;
3872 found->bytes_may_use = 0;
3874 found->max_extent_size = 0;
3875 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3876 found->chunk_alloc = 0;
3878 init_waitqueue_head(&found->wait);
3879 INIT_LIST_HEAD(&found->ro_bgs);
3881 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3882 info->space_info_kobj, "%s",
3883 alloc_name(found->flags));
3889 *space_info = found;
3890 list_add_rcu(&found->list, &info->space_info);
3891 if (flags & BTRFS_BLOCK_GROUP_DATA)
3892 info->data_sinfo = found;
3897 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3899 u64 extra_flags = chunk_to_extended(flags) &
3900 BTRFS_EXTENDED_PROFILE_MASK;
3902 write_seqlock(&fs_info->profiles_lock);
3903 if (flags & BTRFS_BLOCK_GROUP_DATA)
3904 fs_info->avail_data_alloc_bits |= extra_flags;
3905 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3906 fs_info->avail_metadata_alloc_bits |= extra_flags;
3907 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3908 fs_info->avail_system_alloc_bits |= extra_flags;
3909 write_sequnlock(&fs_info->profiles_lock);
3913 * returns target flags in extended format or 0 if restripe for this
3914 * chunk_type is not in progress
3916 * should be called with either volume_mutex or balance_lock held
3918 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3920 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3926 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3927 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3928 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3929 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3930 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3931 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3932 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3933 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3934 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3941 * @flags: available profiles in extended format (see ctree.h)
3943 * Returns reduced profile in chunk format. If profile changing is in
3944 * progress (either running or paused) picks the target profile (if it's
3945 * already available), otherwise falls back to plain reducing.
3947 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3949 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3955 * see if restripe for this chunk_type is in progress, if so
3956 * try to reduce to the target profile
3958 spin_lock(&root->fs_info->balance_lock);
3959 target = get_restripe_target(root->fs_info, flags);
3961 /* pick target profile only if it's already available */
3962 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3963 spin_unlock(&root->fs_info->balance_lock);
3964 return extended_to_chunk(target);
3967 spin_unlock(&root->fs_info->balance_lock);
3969 /* First, mask out the RAID levels which aren't possible */
3970 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3971 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
3972 allowed |= btrfs_raid_group[raid_type];
3976 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
3977 allowed = BTRFS_BLOCK_GROUP_RAID6;
3978 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
3979 allowed = BTRFS_BLOCK_GROUP_RAID5;
3980 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
3981 allowed = BTRFS_BLOCK_GROUP_RAID10;
3982 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
3983 allowed = BTRFS_BLOCK_GROUP_RAID1;
3984 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
3985 allowed = BTRFS_BLOCK_GROUP_RAID0;
3987 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
3989 return extended_to_chunk(flags | allowed);
3992 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3999 seq = read_seqbegin(&root->fs_info->profiles_lock);
4001 if (flags & BTRFS_BLOCK_GROUP_DATA)
4002 flags |= root->fs_info->avail_data_alloc_bits;
4003 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4004 flags |= root->fs_info->avail_system_alloc_bits;
4005 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4006 flags |= root->fs_info->avail_metadata_alloc_bits;
4007 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4009 return btrfs_reduce_alloc_profile(root, flags);
4012 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4018 flags = BTRFS_BLOCK_GROUP_DATA;
4019 else if (root == root->fs_info->chunk_root)
4020 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4022 flags = BTRFS_BLOCK_GROUP_METADATA;
4024 ret = get_alloc_profile(root, flags);
4028 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4030 struct btrfs_space_info *data_sinfo;
4031 struct btrfs_root *root = BTRFS_I(inode)->root;
4032 struct btrfs_fs_info *fs_info = root->fs_info;
4035 int need_commit = 2;
4036 int have_pinned_space;
4038 /* make sure bytes are sectorsize aligned */
4039 bytes = ALIGN(bytes, root->sectorsize);
4041 if (btrfs_is_free_space_inode(inode)) {
4043 ASSERT(current->journal_info);
4046 data_sinfo = fs_info->data_sinfo;
4051 /* make sure we have enough space to handle the data first */
4052 spin_lock(&data_sinfo->lock);
4053 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4054 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4055 data_sinfo->bytes_may_use;
4057 if (used + bytes > data_sinfo->total_bytes) {
4058 struct btrfs_trans_handle *trans;
4061 * if we don't have enough free bytes in this space then we need
4062 * to alloc a new chunk.
4064 if (!data_sinfo->full) {
4067 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4068 spin_unlock(&data_sinfo->lock);
4070 alloc_target = btrfs_get_alloc_profile(root, 1);
4072 * It is ugly that we don't call nolock join
4073 * transaction for the free space inode case here.
4074 * But it is safe because we only do the data space
4075 * reservation for the free space cache in the
4076 * transaction context, the common join transaction
4077 * just increase the counter of the current transaction
4078 * handler, doesn't try to acquire the trans_lock of
4081 trans = btrfs_join_transaction(root);
4083 return PTR_ERR(trans);
4085 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4087 CHUNK_ALLOC_NO_FORCE);
4088 btrfs_end_transaction(trans, root);
4093 have_pinned_space = 1;
4099 data_sinfo = fs_info->data_sinfo;
4105 * If we don't have enough pinned space to deal with this
4106 * allocation, and no removed chunk in current transaction,
4107 * don't bother committing the transaction.
4109 have_pinned_space = percpu_counter_compare(
4110 &data_sinfo->total_bytes_pinned,
4111 used + bytes - data_sinfo->total_bytes);
4112 spin_unlock(&data_sinfo->lock);
4114 /* commit the current transaction and try again */
4117 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4120 if (need_commit > 0)
4121 btrfs_wait_ordered_roots(fs_info, -1);
4123 trans = btrfs_join_transaction(root);
4125 return PTR_ERR(trans);
4126 if (have_pinned_space >= 0 ||
4127 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4128 &trans->transaction->flags) ||
4130 ret = btrfs_commit_transaction(trans, root);
4134 * make sure that all running delayed iput are
4137 down_write(&root->fs_info->delayed_iput_sem);
4138 up_write(&root->fs_info->delayed_iput_sem);
4141 btrfs_end_transaction(trans, root);
4145 trace_btrfs_space_reservation(root->fs_info,
4146 "space_info:enospc",
4147 data_sinfo->flags, bytes, 1);
4150 data_sinfo->bytes_may_use += bytes;
4151 trace_btrfs_space_reservation(root->fs_info, "space_info",
4152 data_sinfo->flags, bytes, 1);
4153 spin_unlock(&data_sinfo->lock);
4159 * New check_data_free_space() with ability for precious data reservation
4160 * Will replace old btrfs_check_data_free_space(), but for patch split,
4161 * add a new function first and then replace it.
4163 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4165 struct btrfs_root *root = BTRFS_I(inode)->root;
4168 /* align the range */
4169 len = round_up(start + len, root->sectorsize) -
4170 round_down(start, root->sectorsize);
4171 start = round_down(start, root->sectorsize);
4173 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4178 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4180 * TODO: Find a good method to avoid reserve data space for NOCOW
4181 * range, but don't impact performance on quota disable case.
4183 ret = btrfs_qgroup_reserve_data(inode, start, len);
4188 * Called if we need to clear a data reservation for this inode
4189 * Normally in a error case.
4191 * This one will *NOT* use accurate qgroup reserved space API, just for case
4192 * which we can't sleep and is sure it won't affect qgroup reserved space.
4193 * Like clear_bit_hook().
4195 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4198 struct btrfs_root *root = BTRFS_I(inode)->root;
4199 struct btrfs_space_info *data_sinfo;
4201 /* Make sure the range is aligned to sectorsize */
4202 len = round_up(start + len, root->sectorsize) -
4203 round_down(start, root->sectorsize);
4204 start = round_down(start, root->sectorsize);
4206 data_sinfo = root->fs_info->data_sinfo;
4207 spin_lock(&data_sinfo->lock);
4208 if (WARN_ON(data_sinfo->bytes_may_use < len))
4209 data_sinfo->bytes_may_use = 0;
4211 data_sinfo->bytes_may_use -= len;
4212 trace_btrfs_space_reservation(root->fs_info, "space_info",
4213 data_sinfo->flags, len, 0);
4214 spin_unlock(&data_sinfo->lock);
4218 * Called if we need to clear a data reservation for this inode
4219 * Normally in a error case.
4221 * This one will handle the per-indoe data rsv map for accurate reserved
4224 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4226 btrfs_free_reserved_data_space_noquota(inode, start, len);
4227 btrfs_qgroup_free_data(inode, start, len);
4230 static void force_metadata_allocation(struct btrfs_fs_info *info)
4232 struct list_head *head = &info->space_info;
4233 struct btrfs_space_info *found;
4236 list_for_each_entry_rcu(found, head, list) {
4237 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4238 found->force_alloc = CHUNK_ALLOC_FORCE;
4243 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4245 return (global->size << 1);
4248 static int should_alloc_chunk(struct btrfs_root *root,
4249 struct btrfs_space_info *sinfo, int force)
4251 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4252 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4253 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4256 if (force == CHUNK_ALLOC_FORCE)
4260 * We need to take into account the global rsv because for all intents
4261 * and purposes it's used space. Don't worry about locking the
4262 * global_rsv, it doesn't change except when the transaction commits.
4264 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4265 num_allocated += calc_global_rsv_need_space(global_rsv);
4268 * in limited mode, we want to have some free space up to
4269 * about 1% of the FS size.
4271 if (force == CHUNK_ALLOC_LIMITED) {
4272 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4273 thresh = max_t(u64, 64 * 1024 * 1024,
4274 div_factor_fine(thresh, 1));
4276 if (num_bytes - num_allocated < thresh)
4280 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4285 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4289 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4290 BTRFS_BLOCK_GROUP_RAID0 |
4291 BTRFS_BLOCK_GROUP_RAID5 |
4292 BTRFS_BLOCK_GROUP_RAID6))
4293 num_dev = root->fs_info->fs_devices->rw_devices;
4294 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4297 num_dev = 1; /* DUP or single */
4303 * If @is_allocation is true, reserve space in the system space info necessary
4304 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4307 void check_system_chunk(struct btrfs_trans_handle *trans,
4308 struct btrfs_root *root,
4311 struct btrfs_space_info *info;
4318 * Needed because we can end up allocating a system chunk and for an
4319 * atomic and race free space reservation in the chunk block reserve.
4321 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4323 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4324 spin_lock(&info->lock);
4325 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4326 info->bytes_reserved - info->bytes_readonly -
4327 info->bytes_may_use;
4328 spin_unlock(&info->lock);
4330 num_devs = get_profile_num_devs(root, type);
4332 /* num_devs device items to update and 1 chunk item to add or remove */
4333 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4334 btrfs_calc_trans_metadata_size(root, 1);
4336 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4337 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4338 left, thresh, type);
4339 dump_space_info(info, 0, 0);
4342 if (left < thresh) {
4345 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4347 * Ignore failure to create system chunk. We might end up not
4348 * needing it, as we might not need to COW all nodes/leafs from
4349 * the paths we visit in the chunk tree (they were already COWed
4350 * or created in the current transaction for example).
4352 ret = btrfs_alloc_chunk(trans, root, flags);
4356 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4357 &root->fs_info->chunk_block_rsv,
4358 thresh, BTRFS_RESERVE_NO_FLUSH);
4360 trans->chunk_bytes_reserved += thresh;
4364 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4365 struct btrfs_root *extent_root, u64 flags, int force)
4367 struct btrfs_space_info *space_info;
4368 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4369 int wait_for_alloc = 0;
4372 /* Don't re-enter if we're already allocating a chunk */
4373 if (trans->allocating_chunk)
4376 space_info = __find_space_info(extent_root->fs_info, flags);
4378 ret = update_space_info(extent_root->fs_info, flags,
4380 BUG_ON(ret); /* -ENOMEM */
4382 BUG_ON(!space_info); /* Logic error */
4385 spin_lock(&space_info->lock);
4386 if (force < space_info->force_alloc)
4387 force = space_info->force_alloc;
4388 if (space_info->full) {
4389 if (should_alloc_chunk(extent_root, space_info, force))
4393 spin_unlock(&space_info->lock);
4397 if (!should_alloc_chunk(extent_root, space_info, force)) {
4398 spin_unlock(&space_info->lock);
4400 } else if (space_info->chunk_alloc) {
4403 space_info->chunk_alloc = 1;
4406 spin_unlock(&space_info->lock);
4408 mutex_lock(&fs_info->chunk_mutex);
4411 * The chunk_mutex is held throughout the entirety of a chunk
4412 * allocation, so once we've acquired the chunk_mutex we know that the
4413 * other guy is done and we need to recheck and see if we should
4416 if (wait_for_alloc) {
4417 mutex_unlock(&fs_info->chunk_mutex);
4422 trans->allocating_chunk = true;
4425 * If we have mixed data/metadata chunks we want to make sure we keep
4426 * allocating mixed chunks instead of individual chunks.
4428 if (btrfs_mixed_space_info(space_info))
4429 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4432 * if we're doing a data chunk, go ahead and make sure that
4433 * we keep a reasonable number of metadata chunks allocated in the
4436 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4437 fs_info->data_chunk_allocations++;
4438 if (!(fs_info->data_chunk_allocations %
4439 fs_info->metadata_ratio))
4440 force_metadata_allocation(fs_info);
4444 * Check if we have enough space in SYSTEM chunk because we may need
4445 * to update devices.
4447 check_system_chunk(trans, extent_root, flags);
4449 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4450 trans->allocating_chunk = false;
4452 spin_lock(&space_info->lock);
4453 if (ret < 0 && ret != -ENOSPC)
4456 space_info->full = 1;
4460 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4462 space_info->chunk_alloc = 0;
4463 spin_unlock(&space_info->lock);
4464 mutex_unlock(&fs_info->chunk_mutex);
4466 * When we allocate a new chunk we reserve space in the chunk block
4467 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4468 * add new nodes/leafs to it if we end up needing to do it when
4469 * inserting the chunk item and updating device items as part of the
4470 * second phase of chunk allocation, performed by
4471 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4472 * large number of new block groups to create in our transaction
4473 * handle's new_bgs list to avoid exhausting the chunk block reserve
4474 * in extreme cases - like having a single transaction create many new
4475 * block groups when starting to write out the free space caches of all
4476 * the block groups that were made dirty during the lifetime of the
4479 if (trans->can_flush_pending_bgs &&
4480 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4481 btrfs_create_pending_block_groups(trans, trans->root);
4482 btrfs_trans_release_chunk_metadata(trans);
4487 static int can_overcommit(struct btrfs_root *root,
4488 struct btrfs_space_info *space_info, u64 bytes,
4489 enum btrfs_reserve_flush_enum flush)
4491 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4492 u64 profile = btrfs_get_alloc_profile(root, 0);
4497 used = space_info->bytes_used + space_info->bytes_reserved +
4498 space_info->bytes_pinned + space_info->bytes_readonly;
4501 * We only want to allow over committing if we have lots of actual space
4502 * free, but if we don't have enough space to handle the global reserve
4503 * space then we could end up having a real enospc problem when trying
4504 * to allocate a chunk or some other such important allocation.
4506 spin_lock(&global_rsv->lock);
4507 space_size = calc_global_rsv_need_space(global_rsv);
4508 spin_unlock(&global_rsv->lock);
4509 if (used + space_size >= space_info->total_bytes)
4512 used += space_info->bytes_may_use;
4514 spin_lock(&root->fs_info->free_chunk_lock);
4515 avail = root->fs_info->free_chunk_space;
4516 spin_unlock(&root->fs_info->free_chunk_lock);
4519 * If we have dup, raid1 or raid10 then only half of the free
4520 * space is actually useable. For raid56, the space info used
4521 * doesn't include the parity drive, so we don't have to
4524 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4525 BTRFS_BLOCK_GROUP_RAID1 |
4526 BTRFS_BLOCK_GROUP_RAID10))
4530 * If we aren't flushing all things, let us overcommit up to
4531 * 1/2th of the space. If we can flush, don't let us overcommit
4532 * too much, let it overcommit up to 1/8 of the space.
4534 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4539 if (used + bytes < space_info->total_bytes + avail)
4544 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4545 unsigned long nr_pages, int nr_items)
4547 struct super_block *sb = root->fs_info->sb;
4549 if (down_read_trylock(&sb->s_umount)) {
4550 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4551 up_read(&sb->s_umount);
4554 * We needn't worry the filesystem going from r/w to r/o though
4555 * we don't acquire ->s_umount mutex, because the filesystem
4556 * should guarantee the delalloc inodes list be empty after
4557 * the filesystem is readonly(all dirty pages are written to
4560 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4561 if (!current->journal_info)
4562 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4566 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4571 bytes = btrfs_calc_trans_metadata_size(root, 1);
4572 nr = (int)div64_u64(to_reclaim, bytes);
4578 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4581 * shrink metadata reservation for delalloc
4583 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4586 struct btrfs_block_rsv *block_rsv;
4587 struct btrfs_space_info *space_info;
4588 struct btrfs_trans_handle *trans;
4592 unsigned long nr_pages;
4595 enum btrfs_reserve_flush_enum flush;
4597 /* Calc the number of the pages we need flush for space reservation */
4598 items = calc_reclaim_items_nr(root, to_reclaim);
4599 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4601 trans = (struct btrfs_trans_handle *)current->journal_info;
4602 block_rsv = &root->fs_info->delalloc_block_rsv;
4603 space_info = block_rsv->space_info;
4605 delalloc_bytes = percpu_counter_sum_positive(
4606 &root->fs_info->delalloc_bytes);
4607 if (delalloc_bytes == 0) {
4611 btrfs_wait_ordered_roots(root->fs_info, items);
4616 while (delalloc_bytes && loops < 3) {
4617 max_reclaim = min(delalloc_bytes, to_reclaim);
4618 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4619 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4621 * We need to wait for the async pages to actually start before
4624 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4628 if (max_reclaim <= nr_pages)
4631 max_reclaim -= nr_pages;
4633 wait_event(root->fs_info->async_submit_wait,
4634 atomic_read(&root->fs_info->async_delalloc_pages) <=
4638 flush = BTRFS_RESERVE_FLUSH_ALL;
4640 flush = BTRFS_RESERVE_NO_FLUSH;
4641 spin_lock(&space_info->lock);
4642 if (can_overcommit(root, space_info, orig, flush)) {
4643 spin_unlock(&space_info->lock);
4646 spin_unlock(&space_info->lock);
4649 if (wait_ordered && !trans) {
4650 btrfs_wait_ordered_roots(root->fs_info, items);
4652 time_left = schedule_timeout_killable(1);
4656 delalloc_bytes = percpu_counter_sum_positive(
4657 &root->fs_info->delalloc_bytes);
4662 * maybe_commit_transaction - possibly commit the transaction if its ok to
4663 * @root - the root we're allocating for
4664 * @bytes - the number of bytes we want to reserve
4665 * @force - force the commit
4667 * This will check to make sure that committing the transaction will actually
4668 * get us somewhere and then commit the transaction if it does. Otherwise it
4669 * will return -ENOSPC.
4671 static int may_commit_transaction(struct btrfs_root *root,
4672 struct btrfs_space_info *space_info,
4673 u64 bytes, int force)
4675 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4676 struct btrfs_trans_handle *trans;
4678 trans = (struct btrfs_trans_handle *)current->journal_info;
4685 /* See if there is enough pinned space to make this reservation */
4686 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4691 * See if there is some space in the delayed insertion reservation for
4694 if (space_info != delayed_rsv->space_info)
4697 spin_lock(&delayed_rsv->lock);
4698 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4699 bytes - delayed_rsv->size) >= 0) {
4700 spin_unlock(&delayed_rsv->lock);
4703 spin_unlock(&delayed_rsv->lock);
4706 trans = btrfs_join_transaction(root);
4710 return btrfs_commit_transaction(trans, root);
4714 FLUSH_DELAYED_ITEMS_NR = 1,
4715 FLUSH_DELAYED_ITEMS = 2,
4717 FLUSH_DELALLOC_WAIT = 4,
4722 static int flush_space(struct btrfs_root *root,
4723 struct btrfs_space_info *space_info, u64 num_bytes,
4724 u64 orig_bytes, int state)
4726 struct btrfs_trans_handle *trans;
4731 case FLUSH_DELAYED_ITEMS_NR:
4732 case FLUSH_DELAYED_ITEMS:
4733 if (state == FLUSH_DELAYED_ITEMS_NR)
4734 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4738 trans = btrfs_join_transaction(root);
4739 if (IS_ERR(trans)) {
4740 ret = PTR_ERR(trans);
4743 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4744 btrfs_end_transaction(trans, root);
4746 case FLUSH_DELALLOC:
4747 case FLUSH_DELALLOC_WAIT:
4748 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4749 state == FLUSH_DELALLOC_WAIT);
4752 trans = btrfs_join_transaction(root);
4753 if (IS_ERR(trans)) {
4754 ret = PTR_ERR(trans);
4757 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4758 btrfs_get_alloc_profile(root, 0),
4759 CHUNK_ALLOC_NO_FORCE);
4760 btrfs_end_transaction(trans, root);
4765 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4776 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4777 struct btrfs_space_info *space_info)
4783 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4785 spin_lock(&space_info->lock);
4786 if (can_overcommit(root, space_info, to_reclaim,
4787 BTRFS_RESERVE_FLUSH_ALL)) {
4792 used = space_info->bytes_used + space_info->bytes_reserved +
4793 space_info->bytes_pinned + space_info->bytes_readonly +
4794 space_info->bytes_may_use;
4795 if (can_overcommit(root, space_info, 1024 * 1024,
4796 BTRFS_RESERVE_FLUSH_ALL))
4797 expected = div_factor_fine(space_info->total_bytes, 95);
4799 expected = div_factor_fine(space_info->total_bytes, 90);
4801 if (used > expected)
4802 to_reclaim = used - expected;
4805 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4806 space_info->bytes_reserved);
4808 spin_unlock(&space_info->lock);
4813 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4814 struct btrfs_fs_info *fs_info, u64 used)
4816 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4818 /* If we're just plain full then async reclaim just slows us down. */
4819 if (space_info->bytes_used >= thresh)
4822 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4823 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4826 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4827 struct btrfs_fs_info *fs_info,
4832 spin_lock(&space_info->lock);
4834 * We run out of space and have not got any free space via flush_space,
4835 * so don't bother doing async reclaim.
4837 if (flush_state > COMMIT_TRANS && space_info->full) {
4838 spin_unlock(&space_info->lock);
4842 used = space_info->bytes_used + space_info->bytes_reserved +
4843 space_info->bytes_pinned + space_info->bytes_readonly +
4844 space_info->bytes_may_use;
4845 if (need_do_async_reclaim(space_info, fs_info, used)) {
4846 spin_unlock(&space_info->lock);
4849 spin_unlock(&space_info->lock);
4854 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4856 struct btrfs_fs_info *fs_info;
4857 struct btrfs_space_info *space_info;
4861 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4862 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4864 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4869 flush_state = FLUSH_DELAYED_ITEMS_NR;
4871 flush_space(fs_info->fs_root, space_info, to_reclaim,
4872 to_reclaim, flush_state);
4874 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4877 } while (flush_state < COMMIT_TRANS);
4880 void btrfs_init_async_reclaim_work(struct work_struct *work)
4882 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4886 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4887 * @root - the root we're allocating for
4888 * @block_rsv - the block_rsv we're allocating for
4889 * @orig_bytes - the number of bytes we want
4890 * @flush - whether or not we can flush to make our reservation
4892 * This will reserve orgi_bytes number of bytes from the space info associated
4893 * with the block_rsv. If there is not enough space it will make an attempt to
4894 * flush out space to make room. It will do this by flushing delalloc if
4895 * possible or committing the transaction. If flush is 0 then no attempts to
4896 * regain reservations will be made and this will fail if there is not enough
4899 static int reserve_metadata_bytes(struct btrfs_root *root,
4900 struct btrfs_block_rsv *block_rsv,
4902 enum btrfs_reserve_flush_enum flush)
4904 struct btrfs_space_info *space_info = block_rsv->space_info;
4906 u64 num_bytes = orig_bytes;
4907 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4909 bool flushing = false;
4913 spin_lock(&space_info->lock);
4915 * We only want to wait if somebody other than us is flushing and we
4916 * are actually allowed to flush all things.
4918 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4919 space_info->flush) {
4920 spin_unlock(&space_info->lock);
4922 * If we have a trans handle we can't wait because the flusher
4923 * may have to commit the transaction, which would mean we would
4924 * deadlock since we are waiting for the flusher to finish, but
4925 * hold the current transaction open.
4927 if (current->journal_info)
4929 ret = wait_event_killable(space_info->wait, !space_info->flush);
4930 /* Must have been killed, return */
4934 spin_lock(&space_info->lock);
4938 used = space_info->bytes_used + space_info->bytes_reserved +
4939 space_info->bytes_pinned + space_info->bytes_readonly +
4940 space_info->bytes_may_use;
4943 * The idea here is that we've not already over-reserved the block group
4944 * then we can go ahead and save our reservation first and then start
4945 * flushing if we need to. Otherwise if we've already overcommitted
4946 * lets start flushing stuff first and then come back and try to make
4949 if (used <= space_info->total_bytes) {
4950 if (used + orig_bytes <= space_info->total_bytes) {
4951 space_info->bytes_may_use += orig_bytes;
4952 trace_btrfs_space_reservation(root->fs_info,
4953 "space_info", space_info->flags, orig_bytes, 1);
4957 * Ok set num_bytes to orig_bytes since we aren't
4958 * overocmmitted, this way we only try and reclaim what
4961 num_bytes = orig_bytes;
4965 * Ok we're over committed, set num_bytes to the overcommitted
4966 * amount plus the amount of bytes that we need for this
4969 num_bytes = used - space_info->total_bytes +
4973 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4974 space_info->bytes_may_use += orig_bytes;
4975 trace_btrfs_space_reservation(root->fs_info, "space_info",
4976 space_info->flags, orig_bytes,
4982 * Couldn't make our reservation, save our place so while we're trying
4983 * to reclaim space we can actually use it instead of somebody else
4984 * stealing it from us.
4986 * We make the other tasks wait for the flush only when we can flush
4989 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4991 space_info->flush = 1;
4992 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4995 * We will do the space reservation dance during log replay,
4996 * which means we won't have fs_info->fs_root set, so don't do
4997 * the async reclaim as we will panic.
4999 if (!root->fs_info->log_root_recovering &&
5000 need_do_async_reclaim(space_info, root->fs_info, used) &&
5001 !work_busy(&root->fs_info->async_reclaim_work))
5002 queue_work(system_unbound_wq,
5003 &root->fs_info->async_reclaim_work);
5005 spin_unlock(&space_info->lock);
5007 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5010 ret = flush_space(root, space_info, num_bytes, orig_bytes,
5015 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
5016 * would happen. So skip delalloc flush.
5018 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5019 (flush_state == FLUSH_DELALLOC ||
5020 flush_state == FLUSH_DELALLOC_WAIT))
5021 flush_state = ALLOC_CHUNK;
5025 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5026 flush_state < COMMIT_TRANS)
5028 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
5029 flush_state <= COMMIT_TRANS)
5033 if (ret == -ENOSPC &&
5034 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5035 struct btrfs_block_rsv *global_rsv =
5036 &root->fs_info->global_block_rsv;
5038 if (block_rsv != global_rsv &&
5039 !block_rsv_use_bytes(global_rsv, orig_bytes))
5043 trace_btrfs_space_reservation(root->fs_info,
5044 "space_info:enospc",
5045 space_info->flags, orig_bytes, 1);
5047 spin_lock(&space_info->lock);
5048 space_info->flush = 0;
5049 wake_up_all(&space_info->wait);
5050 spin_unlock(&space_info->lock);
5055 static struct btrfs_block_rsv *get_block_rsv(
5056 const struct btrfs_trans_handle *trans,
5057 const struct btrfs_root *root)
5059 struct btrfs_block_rsv *block_rsv = NULL;
5061 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5062 (root == root->fs_info->csum_root && trans->adding_csums) ||
5063 (root == root->fs_info->uuid_root))
5064 block_rsv = trans->block_rsv;
5067 block_rsv = root->block_rsv;
5070 block_rsv = &root->fs_info->empty_block_rsv;
5075 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5079 spin_lock(&block_rsv->lock);
5080 if (block_rsv->reserved >= num_bytes) {
5081 block_rsv->reserved -= num_bytes;
5082 if (block_rsv->reserved < block_rsv->size)
5083 block_rsv->full = 0;
5086 spin_unlock(&block_rsv->lock);
5090 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5091 u64 num_bytes, int update_size)
5093 spin_lock(&block_rsv->lock);
5094 block_rsv->reserved += num_bytes;
5096 block_rsv->size += num_bytes;
5097 else if (block_rsv->reserved >= block_rsv->size)
5098 block_rsv->full = 1;
5099 spin_unlock(&block_rsv->lock);
5102 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5103 struct btrfs_block_rsv *dest, u64 num_bytes,
5106 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5109 if (global_rsv->space_info != dest->space_info)
5112 spin_lock(&global_rsv->lock);
5113 min_bytes = div_factor(global_rsv->size, min_factor);
5114 if (global_rsv->reserved < min_bytes + num_bytes) {
5115 spin_unlock(&global_rsv->lock);
5118 global_rsv->reserved -= num_bytes;
5119 if (global_rsv->reserved < global_rsv->size)
5120 global_rsv->full = 0;
5121 spin_unlock(&global_rsv->lock);
5123 block_rsv_add_bytes(dest, num_bytes, 1);
5127 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5128 struct btrfs_block_rsv *block_rsv,
5129 struct btrfs_block_rsv *dest, u64 num_bytes)
5131 struct btrfs_space_info *space_info = block_rsv->space_info;
5133 spin_lock(&block_rsv->lock);
5134 if (num_bytes == (u64)-1)
5135 num_bytes = block_rsv->size;
5136 block_rsv->size -= num_bytes;
5137 if (block_rsv->reserved >= block_rsv->size) {
5138 num_bytes = block_rsv->reserved - block_rsv->size;
5139 block_rsv->reserved = block_rsv->size;
5140 block_rsv->full = 1;
5144 spin_unlock(&block_rsv->lock);
5146 if (num_bytes > 0) {
5148 spin_lock(&dest->lock);
5152 bytes_to_add = dest->size - dest->reserved;
5153 bytes_to_add = min(num_bytes, bytes_to_add);
5154 dest->reserved += bytes_to_add;
5155 if (dest->reserved >= dest->size)
5157 num_bytes -= bytes_to_add;
5159 spin_unlock(&dest->lock);
5162 spin_lock(&space_info->lock);
5163 space_info->bytes_may_use -= num_bytes;
5164 trace_btrfs_space_reservation(fs_info, "space_info",
5165 space_info->flags, num_bytes, 0);
5166 spin_unlock(&space_info->lock);
5171 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5172 struct btrfs_block_rsv *dst, u64 num_bytes)
5176 ret = block_rsv_use_bytes(src, num_bytes);
5180 block_rsv_add_bytes(dst, num_bytes, 1);
5184 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5186 memset(rsv, 0, sizeof(*rsv));
5187 spin_lock_init(&rsv->lock);
5191 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5192 unsigned short type)
5194 struct btrfs_block_rsv *block_rsv;
5195 struct btrfs_fs_info *fs_info = root->fs_info;
5197 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5201 btrfs_init_block_rsv(block_rsv, type);
5202 block_rsv->space_info = __find_space_info(fs_info,
5203 BTRFS_BLOCK_GROUP_METADATA);
5207 void btrfs_free_block_rsv(struct btrfs_root *root,
5208 struct btrfs_block_rsv *rsv)
5212 btrfs_block_rsv_release(root, rsv, (u64)-1);
5216 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5221 int btrfs_block_rsv_add(struct btrfs_root *root,
5222 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5223 enum btrfs_reserve_flush_enum flush)
5230 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5232 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5239 int btrfs_block_rsv_check(struct btrfs_root *root,
5240 struct btrfs_block_rsv *block_rsv, int min_factor)
5248 spin_lock(&block_rsv->lock);
5249 num_bytes = div_factor(block_rsv->size, min_factor);
5250 if (block_rsv->reserved >= num_bytes)
5252 spin_unlock(&block_rsv->lock);
5257 int btrfs_block_rsv_refill(struct btrfs_root *root,
5258 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5259 enum btrfs_reserve_flush_enum flush)
5267 spin_lock(&block_rsv->lock);
5268 num_bytes = min_reserved;
5269 if (block_rsv->reserved >= num_bytes)
5272 num_bytes -= block_rsv->reserved;
5273 spin_unlock(&block_rsv->lock);
5278 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5280 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5287 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5288 struct btrfs_block_rsv *dst_rsv,
5291 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5294 void btrfs_block_rsv_release(struct btrfs_root *root,
5295 struct btrfs_block_rsv *block_rsv,
5298 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5299 if (global_rsv == block_rsv ||
5300 block_rsv->space_info != global_rsv->space_info)
5302 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5307 * helper to calculate size of global block reservation.
5308 * the desired value is sum of space used by extent tree,
5309 * checksum tree and root tree
5311 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5313 struct btrfs_space_info *sinfo;
5317 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5319 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5320 spin_lock(&sinfo->lock);
5321 data_used = sinfo->bytes_used;
5322 spin_unlock(&sinfo->lock);
5324 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5325 spin_lock(&sinfo->lock);
5326 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5328 meta_used = sinfo->bytes_used;
5329 spin_unlock(&sinfo->lock);
5331 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5333 num_bytes += div_u64(data_used + meta_used, 50);
5335 if (num_bytes * 3 > meta_used)
5336 num_bytes = div_u64(meta_used, 3);
5338 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5341 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5343 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5344 struct btrfs_space_info *sinfo = block_rsv->space_info;
5347 num_bytes = calc_global_metadata_size(fs_info);
5349 spin_lock(&sinfo->lock);
5350 spin_lock(&block_rsv->lock);
5352 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5354 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5355 sinfo->bytes_reserved + sinfo->bytes_readonly +
5356 sinfo->bytes_may_use;
5358 if (sinfo->total_bytes > num_bytes) {
5359 num_bytes = sinfo->total_bytes - num_bytes;
5360 block_rsv->reserved += num_bytes;
5361 sinfo->bytes_may_use += num_bytes;
5362 trace_btrfs_space_reservation(fs_info, "space_info",
5363 sinfo->flags, num_bytes, 1);
5366 if (block_rsv->reserved >= block_rsv->size) {
5367 num_bytes = block_rsv->reserved - block_rsv->size;
5368 sinfo->bytes_may_use -= num_bytes;
5369 trace_btrfs_space_reservation(fs_info, "space_info",
5370 sinfo->flags, num_bytes, 0);
5371 block_rsv->reserved = block_rsv->size;
5372 block_rsv->full = 1;
5375 spin_unlock(&block_rsv->lock);
5376 spin_unlock(&sinfo->lock);
5379 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5381 struct btrfs_space_info *space_info;
5383 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5384 fs_info->chunk_block_rsv.space_info = space_info;
5386 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5387 fs_info->global_block_rsv.space_info = space_info;
5388 fs_info->delalloc_block_rsv.space_info = space_info;
5389 fs_info->trans_block_rsv.space_info = space_info;
5390 fs_info->empty_block_rsv.space_info = space_info;
5391 fs_info->delayed_block_rsv.space_info = space_info;
5393 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5394 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5395 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5396 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5397 if (fs_info->quota_root)
5398 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5399 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5401 update_global_block_rsv(fs_info);
5404 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5406 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5408 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5409 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5410 WARN_ON(fs_info->trans_block_rsv.size > 0);
5411 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5412 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5413 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5414 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5415 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5418 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5419 struct btrfs_root *root)
5421 if (!trans->block_rsv)
5424 if (!trans->bytes_reserved)
5427 trace_btrfs_space_reservation(root->fs_info, "transaction",
5428 trans->transid, trans->bytes_reserved, 0);
5429 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5430 trans->bytes_reserved = 0;
5434 * To be called after all the new block groups attached to the transaction
5435 * handle have been created (btrfs_create_pending_block_groups()).
5437 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5439 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5441 if (!trans->chunk_bytes_reserved)
5444 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5446 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5447 trans->chunk_bytes_reserved);
5448 trans->chunk_bytes_reserved = 0;
5451 /* Can only return 0 or -ENOSPC */
5452 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5453 struct inode *inode)
5455 struct btrfs_root *root = BTRFS_I(inode)->root;
5456 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5457 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5460 * We need to hold space in order to delete our orphan item once we've
5461 * added it, so this takes the reservation so we can release it later
5462 * when we are truly done with the orphan item.
5464 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5465 trace_btrfs_space_reservation(root->fs_info, "orphan",
5466 btrfs_ino(inode), num_bytes, 1);
5467 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5470 void btrfs_orphan_release_metadata(struct inode *inode)
5472 struct btrfs_root *root = BTRFS_I(inode)->root;
5473 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5474 trace_btrfs_space_reservation(root->fs_info, "orphan",
5475 btrfs_ino(inode), num_bytes, 0);
5476 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5480 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5481 * root: the root of the parent directory
5482 * rsv: block reservation
5483 * items: the number of items that we need do reservation
5484 * qgroup_reserved: used to return the reserved size in qgroup
5486 * This function is used to reserve the space for snapshot/subvolume
5487 * creation and deletion. Those operations are different with the
5488 * common file/directory operations, they change two fs/file trees
5489 * and root tree, the number of items that the qgroup reserves is
5490 * different with the free space reservation. So we can not use
5491 * the space reseravtion mechanism in start_transaction().
5493 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5494 struct btrfs_block_rsv *rsv,
5496 u64 *qgroup_reserved,
5497 bool use_global_rsv)
5501 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5503 if (root->fs_info->quota_enabled) {
5504 /* One for parent inode, two for dir entries */
5505 num_bytes = 3 * root->nodesize;
5506 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5513 *qgroup_reserved = num_bytes;
5515 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5516 rsv->space_info = __find_space_info(root->fs_info,
5517 BTRFS_BLOCK_GROUP_METADATA);
5518 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5519 BTRFS_RESERVE_FLUSH_ALL);
5521 if (ret == -ENOSPC && use_global_rsv)
5522 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5524 if (ret && *qgroup_reserved)
5525 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5530 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5531 struct btrfs_block_rsv *rsv,
5532 u64 qgroup_reserved)
5534 btrfs_block_rsv_release(root, rsv, (u64)-1);
5538 * drop_outstanding_extent - drop an outstanding extent
5539 * @inode: the inode we're dropping the extent for
5540 * @num_bytes: the number of bytes we're relaseing.
5542 * This is called when we are freeing up an outstanding extent, either called
5543 * after an error or after an extent is written. This will return the number of
5544 * reserved extents that need to be freed. This must be called with
5545 * BTRFS_I(inode)->lock held.
5547 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5549 unsigned drop_inode_space = 0;
5550 unsigned dropped_extents = 0;
5551 unsigned num_extents = 0;
5553 num_extents = (unsigned)div64_u64(num_bytes +
5554 BTRFS_MAX_EXTENT_SIZE - 1,
5555 BTRFS_MAX_EXTENT_SIZE);
5556 ASSERT(num_extents);
5557 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5558 BTRFS_I(inode)->outstanding_extents -= num_extents;
5560 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5561 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5562 &BTRFS_I(inode)->runtime_flags))
5563 drop_inode_space = 1;
5566 * If we have more or the same amount of outsanding extents than we have
5567 * reserved then we need to leave the reserved extents count alone.
5569 if (BTRFS_I(inode)->outstanding_extents >=
5570 BTRFS_I(inode)->reserved_extents)
5571 return drop_inode_space;
5573 dropped_extents = BTRFS_I(inode)->reserved_extents -
5574 BTRFS_I(inode)->outstanding_extents;
5575 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5576 return dropped_extents + drop_inode_space;
5580 * calc_csum_metadata_size - return the amount of metada space that must be
5581 * reserved/free'd for the given bytes.
5582 * @inode: the inode we're manipulating
5583 * @num_bytes: the number of bytes in question
5584 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5586 * This adjusts the number of csum_bytes in the inode and then returns the
5587 * correct amount of metadata that must either be reserved or freed. We
5588 * calculate how many checksums we can fit into one leaf and then divide the
5589 * number of bytes that will need to be checksumed by this value to figure out
5590 * how many checksums will be required. If we are adding bytes then the number
5591 * may go up and we will return the number of additional bytes that must be
5592 * reserved. If it is going down we will return the number of bytes that must
5595 * This must be called with BTRFS_I(inode)->lock held.
5597 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5600 struct btrfs_root *root = BTRFS_I(inode)->root;
5601 u64 old_csums, num_csums;
5603 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5604 BTRFS_I(inode)->csum_bytes == 0)
5607 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5609 BTRFS_I(inode)->csum_bytes += num_bytes;
5611 BTRFS_I(inode)->csum_bytes -= num_bytes;
5612 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5614 /* No change, no need to reserve more */
5615 if (old_csums == num_csums)
5619 return btrfs_calc_trans_metadata_size(root,
5620 num_csums - old_csums);
5622 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5625 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5627 struct btrfs_root *root = BTRFS_I(inode)->root;
5628 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5631 unsigned nr_extents = 0;
5632 int extra_reserve = 0;
5633 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5635 bool delalloc_lock = true;
5639 /* If we are a free space inode we need to not flush since we will be in
5640 * the middle of a transaction commit. We also don't need the delalloc
5641 * mutex since we won't race with anybody. We need this mostly to make
5642 * lockdep shut its filthy mouth.
5644 if (btrfs_is_free_space_inode(inode)) {
5645 flush = BTRFS_RESERVE_NO_FLUSH;
5646 delalloc_lock = false;
5649 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5650 btrfs_transaction_in_commit(root->fs_info))
5651 schedule_timeout(1);
5654 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5656 num_bytes = ALIGN(num_bytes, root->sectorsize);
5658 spin_lock(&BTRFS_I(inode)->lock);
5659 nr_extents = (unsigned)div64_u64(num_bytes +
5660 BTRFS_MAX_EXTENT_SIZE - 1,
5661 BTRFS_MAX_EXTENT_SIZE);
5662 BTRFS_I(inode)->outstanding_extents += nr_extents;
5665 if (BTRFS_I(inode)->outstanding_extents >
5666 BTRFS_I(inode)->reserved_extents)
5667 nr_extents = BTRFS_I(inode)->outstanding_extents -
5668 BTRFS_I(inode)->reserved_extents;
5671 * Add an item to reserve for updating the inode when we complete the
5674 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5675 &BTRFS_I(inode)->runtime_flags)) {
5680 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5681 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5682 csum_bytes = BTRFS_I(inode)->csum_bytes;
5683 spin_unlock(&BTRFS_I(inode)->lock);
5685 if (root->fs_info->quota_enabled) {
5686 ret = btrfs_qgroup_reserve_meta(root,
5687 nr_extents * root->nodesize);
5692 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5693 if (unlikely(ret)) {
5694 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5698 spin_lock(&BTRFS_I(inode)->lock);
5699 if (extra_reserve) {
5700 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5701 &BTRFS_I(inode)->runtime_flags);
5704 BTRFS_I(inode)->reserved_extents += nr_extents;
5705 spin_unlock(&BTRFS_I(inode)->lock);
5708 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5711 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5712 btrfs_ino(inode), to_reserve, 1);
5713 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5718 spin_lock(&BTRFS_I(inode)->lock);
5719 dropped = drop_outstanding_extent(inode, num_bytes);
5721 * If the inodes csum_bytes is the same as the original
5722 * csum_bytes then we know we haven't raced with any free()ers
5723 * so we can just reduce our inodes csum bytes and carry on.
5725 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5726 calc_csum_metadata_size(inode, num_bytes, 0);
5728 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5732 * This is tricky, but first we need to figure out how much we
5733 * free'd from any free-ers that occured during this
5734 * reservation, so we reset ->csum_bytes to the csum_bytes
5735 * before we dropped our lock, and then call the free for the
5736 * number of bytes that were freed while we were trying our
5739 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5740 BTRFS_I(inode)->csum_bytes = csum_bytes;
5741 to_free = calc_csum_metadata_size(inode, bytes, 0);
5745 * Now we need to see how much we would have freed had we not
5746 * been making this reservation and our ->csum_bytes were not
5747 * artificially inflated.
5749 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5750 bytes = csum_bytes - orig_csum_bytes;
5751 bytes = calc_csum_metadata_size(inode, bytes, 0);
5754 * Now reset ->csum_bytes to what it should be. If bytes is
5755 * more than to_free then we would have free'd more space had we
5756 * not had an artificially high ->csum_bytes, so we need to free
5757 * the remainder. If bytes is the same or less then we don't
5758 * need to do anything, the other free-ers did the correct
5761 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5762 if (bytes > to_free)
5763 to_free = bytes - to_free;
5767 spin_unlock(&BTRFS_I(inode)->lock);
5769 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5772 btrfs_block_rsv_release(root, block_rsv, to_free);
5773 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5774 btrfs_ino(inode), to_free, 0);
5777 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5782 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5783 * @inode: the inode to release the reservation for
5784 * @num_bytes: the number of bytes we're releasing
5786 * This will release the metadata reservation for an inode. This can be called
5787 * once we complete IO for a given set of bytes to release their metadata
5790 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5792 struct btrfs_root *root = BTRFS_I(inode)->root;
5796 num_bytes = ALIGN(num_bytes, root->sectorsize);
5797 spin_lock(&BTRFS_I(inode)->lock);
5798 dropped = drop_outstanding_extent(inode, num_bytes);
5801 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5802 spin_unlock(&BTRFS_I(inode)->lock);
5804 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5806 if (btrfs_test_is_dummy_root(root))
5809 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5810 btrfs_ino(inode), to_free, 0);
5812 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5817 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5819 * @inode: inode we're writing to
5820 * @start: start range we are writing to
5821 * @len: how long the range we are writing to
5823 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5825 * This will do the following things
5827 * o reserve space in data space info for num bytes
5828 * and reserve precious corresponding qgroup space
5829 * (Done in check_data_free_space)
5831 * o reserve space for metadata space, based on the number of outstanding
5832 * extents and how much csums will be needed
5833 * also reserve metadata space in a per root over-reserve method.
5834 * o add to the inodes->delalloc_bytes
5835 * o add it to the fs_info's delalloc inodes list.
5836 * (Above 3 all done in delalloc_reserve_metadata)
5838 * Return 0 for success
5839 * Return <0 for error(-ENOSPC or -EQUOT)
5841 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5845 ret = btrfs_check_data_free_space(inode, start, len);
5848 ret = btrfs_delalloc_reserve_metadata(inode, len);
5850 btrfs_free_reserved_data_space(inode, start, len);
5855 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5856 * @inode: inode we're releasing space for
5857 * @start: start position of the space already reserved
5858 * @len: the len of the space already reserved
5860 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5861 * called in the case that we don't need the metadata AND data reservations
5862 * anymore. So if there is an error or we insert an inline extent.
5864 * This function will release the metadata space that was not used and will
5865 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5866 * list if there are no delalloc bytes left.
5867 * Also it will handle the qgroup reserved space.
5869 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5871 btrfs_delalloc_release_metadata(inode, len);
5872 btrfs_free_reserved_data_space(inode, start, len);
5875 static int update_block_group(struct btrfs_trans_handle *trans,
5876 struct btrfs_root *root, u64 bytenr,
5877 u64 num_bytes, int alloc)
5879 struct btrfs_block_group_cache *cache = NULL;
5880 struct btrfs_fs_info *info = root->fs_info;
5881 u64 total = num_bytes;
5886 /* block accounting for super block */
5887 spin_lock(&info->delalloc_root_lock);
5888 old_val = btrfs_super_bytes_used(info->super_copy);
5890 old_val += num_bytes;
5892 old_val -= num_bytes;
5893 btrfs_set_super_bytes_used(info->super_copy, old_val);
5894 spin_unlock(&info->delalloc_root_lock);
5897 cache = btrfs_lookup_block_group(info, bytenr);
5900 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5901 BTRFS_BLOCK_GROUP_RAID1 |
5902 BTRFS_BLOCK_GROUP_RAID10))
5907 * If this block group has free space cache written out, we
5908 * need to make sure to load it if we are removing space. This
5909 * is because we need the unpinning stage to actually add the
5910 * space back to the block group, otherwise we will leak space.
5912 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5913 cache_block_group(cache, 1);
5915 byte_in_group = bytenr - cache->key.objectid;
5916 WARN_ON(byte_in_group > cache->key.offset);
5918 spin_lock(&cache->space_info->lock);
5919 spin_lock(&cache->lock);
5921 if (btrfs_test_opt(root, SPACE_CACHE) &&
5922 cache->disk_cache_state < BTRFS_DC_CLEAR)
5923 cache->disk_cache_state = BTRFS_DC_CLEAR;
5925 old_val = btrfs_block_group_used(&cache->item);
5926 num_bytes = min(total, cache->key.offset - byte_in_group);
5928 old_val += num_bytes;
5929 btrfs_set_block_group_used(&cache->item, old_val);
5930 cache->reserved -= num_bytes;
5931 cache->space_info->bytes_reserved -= num_bytes;
5932 cache->space_info->bytes_used += num_bytes;
5933 cache->space_info->disk_used += num_bytes * factor;
5934 spin_unlock(&cache->lock);
5935 spin_unlock(&cache->space_info->lock);
5937 old_val -= num_bytes;
5938 btrfs_set_block_group_used(&cache->item, old_val);
5939 cache->pinned += num_bytes;
5940 cache->space_info->bytes_pinned += num_bytes;
5941 cache->space_info->bytes_used -= num_bytes;
5942 cache->space_info->disk_used -= num_bytes * factor;
5943 spin_unlock(&cache->lock);
5944 spin_unlock(&cache->space_info->lock);
5946 set_extent_dirty(info->pinned_extents,
5947 bytenr, bytenr + num_bytes - 1,
5948 GFP_NOFS | __GFP_NOFAIL);
5951 spin_lock(&trans->transaction->dirty_bgs_lock);
5952 if (list_empty(&cache->dirty_list)) {
5953 list_add_tail(&cache->dirty_list,
5954 &trans->transaction->dirty_bgs);
5955 trans->transaction->num_dirty_bgs++;
5956 btrfs_get_block_group(cache);
5958 spin_unlock(&trans->transaction->dirty_bgs_lock);
5961 * No longer have used bytes in this block group, queue it for
5962 * deletion. We do this after adding the block group to the
5963 * dirty list to avoid races between cleaner kthread and space
5966 if (!alloc && old_val == 0) {
5967 spin_lock(&info->unused_bgs_lock);
5968 if (list_empty(&cache->bg_list)) {
5969 btrfs_get_block_group(cache);
5970 list_add_tail(&cache->bg_list,
5973 spin_unlock(&info->unused_bgs_lock);
5976 btrfs_put_block_group(cache);
5978 bytenr += num_bytes;
5983 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5985 struct btrfs_block_group_cache *cache;
5988 spin_lock(&root->fs_info->block_group_cache_lock);
5989 bytenr = root->fs_info->first_logical_byte;
5990 spin_unlock(&root->fs_info->block_group_cache_lock);
5992 if (bytenr < (u64)-1)
5995 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5999 bytenr = cache->key.objectid;
6000 btrfs_put_block_group(cache);
6005 static int pin_down_extent(struct btrfs_root *root,
6006 struct btrfs_block_group_cache *cache,
6007 u64 bytenr, u64 num_bytes, int reserved)
6009 spin_lock(&cache->space_info->lock);
6010 spin_lock(&cache->lock);
6011 cache->pinned += num_bytes;
6012 cache->space_info->bytes_pinned += num_bytes;
6014 cache->reserved -= num_bytes;
6015 cache->space_info->bytes_reserved -= num_bytes;
6017 spin_unlock(&cache->lock);
6018 spin_unlock(&cache->space_info->lock);
6020 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6021 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6023 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
6028 * this function must be called within transaction
6030 int btrfs_pin_extent(struct btrfs_root *root,
6031 u64 bytenr, u64 num_bytes, int reserved)
6033 struct btrfs_block_group_cache *cache;
6035 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6036 BUG_ON(!cache); /* Logic error */
6038 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6040 btrfs_put_block_group(cache);
6045 * this function must be called within transaction
6047 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6048 u64 bytenr, u64 num_bytes)
6050 struct btrfs_block_group_cache *cache;
6053 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6058 * pull in the free space cache (if any) so that our pin
6059 * removes the free space from the cache. We have load_only set
6060 * to one because the slow code to read in the free extents does check
6061 * the pinned extents.
6063 cache_block_group(cache, 1);
6065 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6067 /* remove us from the free space cache (if we're there at all) */
6068 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6069 btrfs_put_block_group(cache);
6073 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6076 struct btrfs_block_group_cache *block_group;
6077 struct btrfs_caching_control *caching_ctl;
6079 block_group = btrfs_lookup_block_group(root->fs_info, start);
6083 cache_block_group(block_group, 0);
6084 caching_ctl = get_caching_control(block_group);
6088 BUG_ON(!block_group_cache_done(block_group));
6089 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6091 mutex_lock(&caching_ctl->mutex);
6093 if (start >= caching_ctl->progress) {
6094 ret = add_excluded_extent(root, start, num_bytes);
6095 } else if (start + num_bytes <= caching_ctl->progress) {
6096 ret = btrfs_remove_free_space(block_group,
6099 num_bytes = caching_ctl->progress - start;
6100 ret = btrfs_remove_free_space(block_group,
6105 num_bytes = (start + num_bytes) -
6106 caching_ctl->progress;
6107 start = caching_ctl->progress;
6108 ret = add_excluded_extent(root, start, num_bytes);
6111 mutex_unlock(&caching_ctl->mutex);
6112 put_caching_control(caching_ctl);
6114 btrfs_put_block_group(block_group);
6118 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6119 struct extent_buffer *eb)
6121 struct btrfs_file_extent_item *item;
6122 struct btrfs_key key;
6126 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6129 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6130 btrfs_item_key_to_cpu(eb, &key, i);
6131 if (key.type != BTRFS_EXTENT_DATA_KEY)
6133 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6134 found_type = btrfs_file_extent_type(eb, item);
6135 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6137 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6139 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6140 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6141 __exclude_logged_extent(log, key.objectid, key.offset);
6148 * btrfs_update_reserved_bytes - update the block_group and space info counters
6149 * @cache: The cache we are manipulating
6150 * @num_bytes: The number of bytes in question
6151 * @reserve: One of the reservation enums
6152 * @delalloc: The blocks are allocated for the delalloc write
6154 * This is called by the allocator when it reserves space, or by somebody who is
6155 * freeing space that was never actually used on disk. For example if you
6156 * reserve some space for a new leaf in transaction A and before transaction A
6157 * commits you free that leaf, you call this with reserve set to 0 in order to
6158 * clear the reservation.
6160 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6161 * ENOSPC accounting. For data we handle the reservation through clearing the
6162 * delalloc bits in the io_tree. We have to do this since we could end up
6163 * allocating less disk space for the amount of data we have reserved in the
6164 * case of compression.
6166 * If this is a reservation and the block group has become read only we cannot
6167 * make the reservation and return -EAGAIN, otherwise this function always
6170 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6171 u64 num_bytes, int reserve, int delalloc)
6173 struct btrfs_space_info *space_info = cache->space_info;
6176 spin_lock(&space_info->lock);
6177 spin_lock(&cache->lock);
6178 if (reserve != RESERVE_FREE) {
6182 cache->reserved += num_bytes;
6183 space_info->bytes_reserved += num_bytes;
6184 if (reserve == RESERVE_ALLOC) {
6185 trace_btrfs_space_reservation(cache->fs_info,
6186 "space_info", space_info->flags,
6188 space_info->bytes_may_use -= num_bytes;
6192 cache->delalloc_bytes += num_bytes;
6196 space_info->bytes_readonly += num_bytes;
6197 cache->reserved -= num_bytes;
6198 space_info->bytes_reserved -= num_bytes;
6201 cache->delalloc_bytes -= num_bytes;
6203 spin_unlock(&cache->lock);
6204 spin_unlock(&space_info->lock);
6208 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6209 struct btrfs_root *root)
6211 struct btrfs_fs_info *fs_info = root->fs_info;
6212 struct btrfs_caching_control *next;
6213 struct btrfs_caching_control *caching_ctl;
6214 struct btrfs_block_group_cache *cache;
6216 down_write(&fs_info->commit_root_sem);
6218 list_for_each_entry_safe(caching_ctl, next,
6219 &fs_info->caching_block_groups, list) {
6220 cache = caching_ctl->block_group;
6221 if (block_group_cache_done(cache)) {
6222 cache->last_byte_to_unpin = (u64)-1;
6223 list_del_init(&caching_ctl->list);
6224 put_caching_control(caching_ctl);
6226 cache->last_byte_to_unpin = caching_ctl->progress;
6230 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6231 fs_info->pinned_extents = &fs_info->freed_extents[1];
6233 fs_info->pinned_extents = &fs_info->freed_extents[0];
6235 up_write(&fs_info->commit_root_sem);
6237 update_global_block_rsv(fs_info);
6241 * Returns the free cluster for the given space info and sets empty_cluster to
6242 * what it should be based on the mount options.
6244 static struct btrfs_free_cluster *
6245 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6248 struct btrfs_free_cluster *ret = NULL;
6249 bool ssd = btrfs_test_opt(root, SSD);
6252 if (btrfs_mixed_space_info(space_info))
6256 *empty_cluster = 2 * 1024 * 1024;
6257 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6258 ret = &root->fs_info->meta_alloc_cluster;
6260 *empty_cluster = 64 * 1024;
6261 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6262 ret = &root->fs_info->data_alloc_cluster;
6268 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6269 const bool return_free_space)
6271 struct btrfs_fs_info *fs_info = root->fs_info;
6272 struct btrfs_block_group_cache *cache = NULL;
6273 struct btrfs_space_info *space_info;
6274 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6275 struct btrfs_free_cluster *cluster = NULL;
6277 u64 total_unpinned = 0;
6278 u64 empty_cluster = 0;
6281 while (start <= end) {
6284 start >= cache->key.objectid + cache->key.offset) {
6286 btrfs_put_block_group(cache);
6288 cache = btrfs_lookup_block_group(fs_info, start);
6289 BUG_ON(!cache); /* Logic error */
6291 cluster = fetch_cluster_info(root,
6294 empty_cluster <<= 1;
6297 len = cache->key.objectid + cache->key.offset - start;
6298 len = min(len, end + 1 - start);
6300 if (start < cache->last_byte_to_unpin) {
6301 len = min(len, cache->last_byte_to_unpin - start);
6302 if (return_free_space)
6303 btrfs_add_free_space(cache, start, len);
6307 total_unpinned += len;
6308 space_info = cache->space_info;
6311 * If this space cluster has been marked as fragmented and we've
6312 * unpinned enough in this block group to potentially allow a
6313 * cluster to be created inside of it go ahead and clear the
6316 if (cluster && cluster->fragmented &&
6317 total_unpinned > empty_cluster) {
6318 spin_lock(&cluster->lock);
6319 cluster->fragmented = 0;
6320 spin_unlock(&cluster->lock);
6323 spin_lock(&space_info->lock);
6324 spin_lock(&cache->lock);
6325 cache->pinned -= len;
6326 space_info->bytes_pinned -= len;
6327 space_info->max_extent_size = 0;
6328 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6330 space_info->bytes_readonly += len;
6333 spin_unlock(&cache->lock);
6334 if (!readonly && global_rsv->space_info == space_info) {
6335 spin_lock(&global_rsv->lock);
6336 if (!global_rsv->full) {
6337 len = min(len, global_rsv->size -
6338 global_rsv->reserved);
6339 global_rsv->reserved += len;
6340 space_info->bytes_may_use += len;
6341 if (global_rsv->reserved >= global_rsv->size)
6342 global_rsv->full = 1;
6344 spin_unlock(&global_rsv->lock);
6346 spin_unlock(&space_info->lock);
6350 btrfs_put_block_group(cache);
6354 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6355 struct btrfs_root *root)
6357 struct btrfs_fs_info *fs_info = root->fs_info;
6358 struct btrfs_block_group_cache *block_group, *tmp;
6359 struct list_head *deleted_bgs;
6360 struct extent_io_tree *unpin;
6365 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6366 unpin = &fs_info->freed_extents[1];
6368 unpin = &fs_info->freed_extents[0];
6370 while (!trans->aborted) {
6371 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6372 ret = find_first_extent_bit(unpin, 0, &start, &end,
6373 EXTENT_DIRTY, NULL);
6375 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6379 if (btrfs_test_opt(root, DISCARD))
6380 ret = btrfs_discard_extent(root, start,
6381 end + 1 - start, NULL);
6383 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6384 unpin_extent_range(root, start, end, true);
6385 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6390 * Transaction is finished. We don't need the lock anymore. We
6391 * do need to clean up the block groups in case of a transaction
6394 deleted_bgs = &trans->transaction->deleted_bgs;
6395 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6399 if (!trans->aborted)
6400 ret = btrfs_discard_extent(root,
6401 block_group->key.objectid,
6402 block_group->key.offset,
6405 list_del_init(&block_group->bg_list);
6406 btrfs_put_block_group_trimming(block_group);
6407 btrfs_put_block_group(block_group);
6410 const char *errstr = btrfs_decode_error(ret);
6412 "Discard failed while removing blockgroup: errno=%d %s\n",
6420 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6421 u64 owner, u64 root_objectid)
6423 struct btrfs_space_info *space_info;
6426 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6427 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6428 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6430 flags = BTRFS_BLOCK_GROUP_METADATA;
6432 flags = BTRFS_BLOCK_GROUP_DATA;
6435 space_info = __find_space_info(fs_info, flags);
6436 BUG_ON(!space_info); /* Logic bug */
6437 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6441 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6442 struct btrfs_root *root,
6443 struct btrfs_delayed_ref_node *node, u64 parent,
6444 u64 root_objectid, u64 owner_objectid,
6445 u64 owner_offset, int refs_to_drop,
6446 struct btrfs_delayed_extent_op *extent_op)
6448 struct btrfs_key key;
6449 struct btrfs_path *path;
6450 struct btrfs_fs_info *info = root->fs_info;
6451 struct btrfs_root *extent_root = info->extent_root;
6452 struct extent_buffer *leaf;
6453 struct btrfs_extent_item *ei;
6454 struct btrfs_extent_inline_ref *iref;
6457 int extent_slot = 0;
6458 int found_extent = 0;
6462 u64 bytenr = node->bytenr;
6463 u64 num_bytes = node->num_bytes;
6465 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6468 path = btrfs_alloc_path();
6473 path->leave_spinning = 1;
6475 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6476 BUG_ON(!is_data && refs_to_drop != 1);
6479 skinny_metadata = 0;
6481 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6482 bytenr, num_bytes, parent,
6483 root_objectid, owner_objectid,
6486 extent_slot = path->slots[0];
6487 while (extent_slot >= 0) {
6488 btrfs_item_key_to_cpu(path->nodes[0], &key,
6490 if (key.objectid != bytenr)
6492 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6493 key.offset == num_bytes) {
6497 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6498 key.offset == owner_objectid) {
6502 if (path->slots[0] - extent_slot > 5)
6506 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6507 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6508 if (found_extent && item_size < sizeof(*ei))
6511 if (!found_extent) {
6513 ret = remove_extent_backref(trans, extent_root, path,
6515 is_data, &last_ref);
6517 btrfs_abort_transaction(trans, extent_root, ret);
6520 btrfs_release_path(path);
6521 path->leave_spinning = 1;
6523 key.objectid = bytenr;
6524 key.type = BTRFS_EXTENT_ITEM_KEY;
6525 key.offset = num_bytes;
6527 if (!is_data && skinny_metadata) {
6528 key.type = BTRFS_METADATA_ITEM_KEY;
6529 key.offset = owner_objectid;
6532 ret = btrfs_search_slot(trans, extent_root,
6534 if (ret > 0 && skinny_metadata && path->slots[0]) {
6536 * Couldn't find our skinny metadata item,
6537 * see if we have ye olde extent item.
6540 btrfs_item_key_to_cpu(path->nodes[0], &key,
6542 if (key.objectid == bytenr &&
6543 key.type == BTRFS_EXTENT_ITEM_KEY &&
6544 key.offset == num_bytes)
6548 if (ret > 0 && skinny_metadata) {
6549 skinny_metadata = false;
6550 key.objectid = bytenr;
6551 key.type = BTRFS_EXTENT_ITEM_KEY;
6552 key.offset = num_bytes;
6553 btrfs_release_path(path);
6554 ret = btrfs_search_slot(trans, extent_root,
6559 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6562 btrfs_print_leaf(extent_root,
6566 btrfs_abort_transaction(trans, extent_root, ret);
6569 extent_slot = path->slots[0];
6571 } else if (WARN_ON(ret == -ENOENT)) {
6572 btrfs_print_leaf(extent_root, path->nodes[0]);
6574 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6575 bytenr, parent, root_objectid, owner_objectid,
6577 btrfs_abort_transaction(trans, extent_root, ret);
6580 btrfs_abort_transaction(trans, extent_root, ret);
6584 leaf = path->nodes[0];
6585 item_size = btrfs_item_size_nr(leaf, extent_slot);
6586 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6587 if (item_size < sizeof(*ei)) {
6588 BUG_ON(found_extent || extent_slot != path->slots[0]);
6589 ret = convert_extent_item_v0(trans, extent_root, path,
6592 btrfs_abort_transaction(trans, extent_root, ret);
6596 btrfs_release_path(path);
6597 path->leave_spinning = 1;
6599 key.objectid = bytenr;
6600 key.type = BTRFS_EXTENT_ITEM_KEY;
6601 key.offset = num_bytes;
6603 ret = btrfs_search_slot(trans, extent_root, &key, path,
6606 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6608 btrfs_print_leaf(extent_root, path->nodes[0]);
6611 btrfs_abort_transaction(trans, extent_root, ret);
6615 extent_slot = path->slots[0];
6616 leaf = path->nodes[0];
6617 item_size = btrfs_item_size_nr(leaf, extent_slot);
6620 BUG_ON(item_size < sizeof(*ei));
6621 ei = btrfs_item_ptr(leaf, extent_slot,
6622 struct btrfs_extent_item);
6623 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6624 key.type == BTRFS_EXTENT_ITEM_KEY) {
6625 struct btrfs_tree_block_info *bi;
6626 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6627 bi = (struct btrfs_tree_block_info *)(ei + 1);
6628 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6631 refs = btrfs_extent_refs(leaf, ei);
6632 if (refs < refs_to_drop) {
6633 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6634 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6636 btrfs_abort_transaction(trans, extent_root, ret);
6639 refs -= refs_to_drop;
6643 __run_delayed_extent_op(extent_op, leaf, ei);
6645 * In the case of inline back ref, reference count will
6646 * be updated by remove_extent_backref
6649 BUG_ON(!found_extent);
6651 btrfs_set_extent_refs(leaf, ei, refs);
6652 btrfs_mark_buffer_dirty(leaf);
6655 ret = remove_extent_backref(trans, extent_root, path,
6657 is_data, &last_ref);
6659 btrfs_abort_transaction(trans, extent_root, ret);
6663 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6667 BUG_ON(is_data && refs_to_drop !=
6668 extent_data_ref_count(path, iref));
6670 BUG_ON(path->slots[0] != extent_slot);
6672 BUG_ON(path->slots[0] != extent_slot + 1);
6673 path->slots[0] = extent_slot;
6679 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6682 btrfs_abort_transaction(trans, extent_root, ret);
6685 btrfs_release_path(path);
6688 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6690 btrfs_abort_transaction(trans, extent_root, ret);
6695 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
6698 btrfs_abort_transaction(trans, extent_root, ret);
6702 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6704 btrfs_abort_transaction(trans, extent_root, ret);
6708 btrfs_release_path(path);
6711 btrfs_free_path(path);
6716 * when we free an block, it is possible (and likely) that we free the last
6717 * delayed ref for that extent as well. This searches the delayed ref tree for
6718 * a given extent, and if there are no other delayed refs to be processed, it
6719 * removes it from the tree.
6721 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6722 struct btrfs_root *root, u64 bytenr)
6724 struct btrfs_delayed_ref_head *head;
6725 struct btrfs_delayed_ref_root *delayed_refs;
6728 delayed_refs = &trans->transaction->delayed_refs;
6729 spin_lock(&delayed_refs->lock);
6730 head = btrfs_find_delayed_ref_head(trans, bytenr);
6732 goto out_delayed_unlock;
6734 spin_lock(&head->lock);
6735 if (!list_empty(&head->ref_list))
6738 if (head->extent_op) {
6739 if (!head->must_insert_reserved)
6741 btrfs_free_delayed_extent_op(head->extent_op);
6742 head->extent_op = NULL;
6746 * waiting for the lock here would deadlock. If someone else has it
6747 * locked they are already in the process of dropping it anyway
6749 if (!mutex_trylock(&head->mutex))
6753 * at this point we have a head with no other entries. Go
6754 * ahead and process it.
6756 head->node.in_tree = 0;
6757 rb_erase(&head->href_node, &delayed_refs->href_root);
6759 atomic_dec(&delayed_refs->num_entries);
6762 * we don't take a ref on the node because we're removing it from the
6763 * tree, so we just steal the ref the tree was holding.
6765 delayed_refs->num_heads--;
6766 if (head->processing == 0)
6767 delayed_refs->num_heads_ready--;
6768 head->processing = 0;
6769 spin_unlock(&head->lock);
6770 spin_unlock(&delayed_refs->lock);
6772 BUG_ON(head->extent_op);
6773 if (head->must_insert_reserved)
6776 mutex_unlock(&head->mutex);
6777 btrfs_put_delayed_ref(&head->node);
6780 spin_unlock(&head->lock);
6783 spin_unlock(&delayed_refs->lock);
6787 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6788 struct btrfs_root *root,
6789 struct extent_buffer *buf,
6790 u64 parent, int last_ref)
6795 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6796 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6797 buf->start, buf->len,
6798 parent, root->root_key.objectid,
6799 btrfs_header_level(buf),
6800 BTRFS_DROP_DELAYED_REF, NULL);
6801 BUG_ON(ret); /* -ENOMEM */
6807 if (btrfs_header_generation(buf) == trans->transid) {
6808 struct btrfs_block_group_cache *cache;
6810 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6811 ret = check_ref_cleanup(trans, root, buf->start);
6816 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6818 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6819 pin_down_extent(root, cache, buf->start, buf->len, 1);
6820 btrfs_put_block_group(cache);
6824 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6826 btrfs_add_free_space(cache, buf->start, buf->len);
6827 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6828 btrfs_put_block_group(cache);
6829 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6834 add_pinned_bytes(root->fs_info, buf->len,
6835 btrfs_header_level(buf),
6836 root->root_key.objectid);
6839 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6842 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6845 /* Can return -ENOMEM */
6846 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6847 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6848 u64 owner, u64 offset)
6851 struct btrfs_fs_info *fs_info = root->fs_info;
6853 if (btrfs_test_is_dummy_root(root))
6856 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6859 * tree log blocks never actually go into the extent allocation
6860 * tree, just update pinning info and exit early.
6862 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6863 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6864 /* unlocks the pinned mutex */
6865 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6867 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6868 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6870 parent, root_objectid, (int)owner,
6871 BTRFS_DROP_DELAYED_REF, NULL);
6873 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6875 parent, root_objectid, owner,
6877 BTRFS_DROP_DELAYED_REF, NULL);
6883 * when we wait for progress in the block group caching, its because
6884 * our allocation attempt failed at least once. So, we must sleep
6885 * and let some progress happen before we try again.
6887 * This function will sleep at least once waiting for new free space to
6888 * show up, and then it will check the block group free space numbers
6889 * for our min num_bytes. Another option is to have it go ahead
6890 * and look in the rbtree for a free extent of a given size, but this
6893 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6894 * any of the information in this block group.
6896 static noinline void
6897 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6900 struct btrfs_caching_control *caching_ctl;
6902 caching_ctl = get_caching_control(cache);
6906 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6907 (cache->free_space_ctl->free_space >= num_bytes));
6909 put_caching_control(caching_ctl);
6913 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6915 struct btrfs_caching_control *caching_ctl;
6918 caching_ctl = get_caching_control(cache);
6920 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6922 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6923 if (cache->cached == BTRFS_CACHE_ERROR)
6925 put_caching_control(caching_ctl);
6929 int __get_raid_index(u64 flags)
6931 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6932 return BTRFS_RAID_RAID10;
6933 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6934 return BTRFS_RAID_RAID1;
6935 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6936 return BTRFS_RAID_DUP;
6937 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6938 return BTRFS_RAID_RAID0;
6939 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6940 return BTRFS_RAID_RAID5;
6941 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6942 return BTRFS_RAID_RAID6;
6944 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6947 int get_block_group_index(struct btrfs_block_group_cache *cache)
6949 return __get_raid_index(cache->flags);
6952 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6953 [BTRFS_RAID_RAID10] = "raid10",
6954 [BTRFS_RAID_RAID1] = "raid1",
6955 [BTRFS_RAID_DUP] = "dup",
6956 [BTRFS_RAID_RAID0] = "raid0",
6957 [BTRFS_RAID_SINGLE] = "single",
6958 [BTRFS_RAID_RAID5] = "raid5",
6959 [BTRFS_RAID_RAID6] = "raid6",
6962 static const char *get_raid_name(enum btrfs_raid_types type)
6964 if (type >= BTRFS_NR_RAID_TYPES)
6967 return btrfs_raid_type_names[type];
6970 enum btrfs_loop_type {
6971 LOOP_CACHING_NOWAIT = 0,
6972 LOOP_CACHING_WAIT = 1,
6973 LOOP_ALLOC_CHUNK = 2,
6974 LOOP_NO_EMPTY_SIZE = 3,
6978 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6982 down_read(&cache->data_rwsem);
6986 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6989 btrfs_get_block_group(cache);
6991 down_read(&cache->data_rwsem);
6994 static struct btrfs_block_group_cache *
6995 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6996 struct btrfs_free_cluster *cluster,
6999 struct btrfs_block_group_cache *used_bg;
7000 bool locked = false;
7002 spin_lock(&cluster->refill_lock);
7004 if (used_bg == cluster->block_group)
7007 up_read(&used_bg->data_rwsem);
7008 btrfs_put_block_group(used_bg);
7011 used_bg = cluster->block_group;
7015 if (used_bg == block_group)
7018 btrfs_get_block_group(used_bg);
7023 if (down_read_trylock(&used_bg->data_rwsem))
7026 spin_unlock(&cluster->refill_lock);
7027 down_read(&used_bg->data_rwsem);
7033 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7037 up_read(&cache->data_rwsem);
7038 btrfs_put_block_group(cache);
7042 * walks the btree of allocated extents and find a hole of a given size.
7043 * The key ins is changed to record the hole:
7044 * ins->objectid == start position
7045 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7046 * ins->offset == the size of the hole.
7047 * Any available blocks before search_start are skipped.
7049 * If there is no suitable free space, we will record the max size of
7050 * the free space extent currently.
7052 static noinline int find_free_extent(struct btrfs_root *orig_root,
7053 u64 num_bytes, u64 empty_size,
7054 u64 hint_byte, struct btrfs_key *ins,
7055 u64 flags, int delalloc)
7058 struct btrfs_root *root = orig_root->fs_info->extent_root;
7059 struct btrfs_free_cluster *last_ptr = NULL;
7060 struct btrfs_block_group_cache *block_group = NULL;
7061 u64 search_start = 0;
7062 u64 max_extent_size = 0;
7063 u64 empty_cluster = 0;
7064 struct btrfs_space_info *space_info;
7066 int index = __get_raid_index(flags);
7067 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7068 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7069 bool failed_cluster_refill = false;
7070 bool failed_alloc = false;
7071 bool use_cluster = true;
7072 bool have_caching_bg = false;
7073 bool orig_have_caching_bg = false;
7074 bool full_search = false;
7076 WARN_ON(num_bytes < root->sectorsize);
7077 ins->type = BTRFS_EXTENT_ITEM_KEY;
7081 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7083 space_info = __find_space_info(root->fs_info, flags);
7085 btrfs_err(root->fs_info, "No space info for %llu", flags);
7090 * If our free space is heavily fragmented we may not be able to make
7091 * big contiguous allocations, so instead of doing the expensive search
7092 * for free space, simply return ENOSPC with our max_extent_size so we
7093 * can go ahead and search for a more manageable chunk.
7095 * If our max_extent_size is large enough for our allocation simply
7096 * disable clustering since we will likely not be able to find enough
7097 * space to create a cluster and induce latency trying.
7099 if (unlikely(space_info->max_extent_size)) {
7100 spin_lock(&space_info->lock);
7101 if (space_info->max_extent_size &&
7102 num_bytes > space_info->max_extent_size) {
7103 ins->offset = space_info->max_extent_size;
7104 spin_unlock(&space_info->lock);
7106 } else if (space_info->max_extent_size) {
7107 use_cluster = false;
7109 spin_unlock(&space_info->lock);
7112 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7114 spin_lock(&last_ptr->lock);
7115 if (last_ptr->block_group)
7116 hint_byte = last_ptr->window_start;
7117 if (last_ptr->fragmented) {
7119 * We still set window_start so we can keep track of the
7120 * last place we found an allocation to try and save
7123 hint_byte = last_ptr->window_start;
7124 use_cluster = false;
7126 spin_unlock(&last_ptr->lock);
7129 search_start = max(search_start, first_logical_byte(root, 0));
7130 search_start = max(search_start, hint_byte);
7131 if (search_start == hint_byte) {
7132 block_group = btrfs_lookup_block_group(root->fs_info,
7135 * we don't want to use the block group if it doesn't match our
7136 * allocation bits, or if its not cached.
7138 * However if we are re-searching with an ideal block group
7139 * picked out then we don't care that the block group is cached.
7141 if (block_group && block_group_bits(block_group, flags) &&
7142 block_group->cached != BTRFS_CACHE_NO) {
7143 down_read(&space_info->groups_sem);
7144 if (list_empty(&block_group->list) ||
7147 * someone is removing this block group,
7148 * we can't jump into the have_block_group
7149 * target because our list pointers are not
7152 btrfs_put_block_group(block_group);
7153 up_read(&space_info->groups_sem);
7155 index = get_block_group_index(block_group);
7156 btrfs_lock_block_group(block_group, delalloc);
7157 goto have_block_group;
7159 } else if (block_group) {
7160 btrfs_put_block_group(block_group);
7164 have_caching_bg = false;
7165 if (index == 0 || index == __get_raid_index(flags))
7167 down_read(&space_info->groups_sem);
7168 list_for_each_entry(block_group, &space_info->block_groups[index],
7173 btrfs_grab_block_group(block_group, delalloc);
7174 search_start = block_group->key.objectid;
7177 * this can happen if we end up cycling through all the
7178 * raid types, but we want to make sure we only allocate
7179 * for the proper type.
7181 if (!block_group_bits(block_group, flags)) {
7182 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7183 BTRFS_BLOCK_GROUP_RAID1 |
7184 BTRFS_BLOCK_GROUP_RAID5 |
7185 BTRFS_BLOCK_GROUP_RAID6 |
7186 BTRFS_BLOCK_GROUP_RAID10;
7189 * if they asked for extra copies and this block group
7190 * doesn't provide them, bail. This does allow us to
7191 * fill raid0 from raid1.
7193 if ((flags & extra) && !(block_group->flags & extra))
7198 cached = block_group_cache_done(block_group);
7199 if (unlikely(!cached)) {
7200 have_caching_bg = true;
7201 ret = cache_block_group(block_group, 0);
7206 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7208 if (unlikely(block_group->ro))
7212 * Ok we want to try and use the cluster allocator, so
7215 if (last_ptr && use_cluster) {
7216 struct btrfs_block_group_cache *used_block_group;
7217 unsigned long aligned_cluster;
7219 * the refill lock keeps out other
7220 * people trying to start a new cluster
7222 used_block_group = btrfs_lock_cluster(block_group,
7225 if (!used_block_group)
7226 goto refill_cluster;
7228 if (used_block_group != block_group &&
7229 (used_block_group->ro ||
7230 !block_group_bits(used_block_group, flags)))
7231 goto release_cluster;
7233 offset = btrfs_alloc_from_cluster(used_block_group,
7236 used_block_group->key.objectid,
7239 /* we have a block, we're done */
7240 spin_unlock(&last_ptr->refill_lock);
7241 trace_btrfs_reserve_extent_cluster(root,
7243 search_start, num_bytes);
7244 if (used_block_group != block_group) {
7245 btrfs_release_block_group(block_group,
7247 block_group = used_block_group;
7252 WARN_ON(last_ptr->block_group != used_block_group);
7254 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7255 * set up a new clusters, so lets just skip it
7256 * and let the allocator find whatever block
7257 * it can find. If we reach this point, we
7258 * will have tried the cluster allocator
7259 * plenty of times and not have found
7260 * anything, so we are likely way too
7261 * fragmented for the clustering stuff to find
7264 * However, if the cluster is taken from the
7265 * current block group, release the cluster
7266 * first, so that we stand a better chance of
7267 * succeeding in the unclustered
7269 if (loop >= LOOP_NO_EMPTY_SIZE &&
7270 used_block_group != block_group) {
7271 spin_unlock(&last_ptr->refill_lock);
7272 btrfs_release_block_group(used_block_group,
7274 goto unclustered_alloc;
7278 * this cluster didn't work out, free it and
7281 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7283 if (used_block_group != block_group)
7284 btrfs_release_block_group(used_block_group,
7287 if (loop >= LOOP_NO_EMPTY_SIZE) {
7288 spin_unlock(&last_ptr->refill_lock);
7289 goto unclustered_alloc;
7292 aligned_cluster = max_t(unsigned long,
7293 empty_cluster + empty_size,
7294 block_group->full_stripe_len);
7296 /* allocate a cluster in this block group */
7297 ret = btrfs_find_space_cluster(root, block_group,
7298 last_ptr, search_start,
7303 * now pull our allocation out of this
7306 offset = btrfs_alloc_from_cluster(block_group,
7312 /* we found one, proceed */
7313 spin_unlock(&last_ptr->refill_lock);
7314 trace_btrfs_reserve_extent_cluster(root,
7315 block_group, search_start,
7319 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7320 && !failed_cluster_refill) {
7321 spin_unlock(&last_ptr->refill_lock);
7323 failed_cluster_refill = true;
7324 wait_block_group_cache_progress(block_group,
7325 num_bytes + empty_cluster + empty_size);
7326 goto have_block_group;
7330 * at this point we either didn't find a cluster
7331 * or we weren't able to allocate a block from our
7332 * cluster. Free the cluster we've been trying
7333 * to use, and go to the next block group
7335 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7336 spin_unlock(&last_ptr->refill_lock);
7342 * We are doing an unclustered alloc, set the fragmented flag so
7343 * we don't bother trying to setup a cluster again until we get
7346 if (unlikely(last_ptr)) {
7347 spin_lock(&last_ptr->lock);
7348 last_ptr->fragmented = 1;
7349 spin_unlock(&last_ptr->lock);
7351 spin_lock(&block_group->free_space_ctl->tree_lock);
7353 block_group->free_space_ctl->free_space <
7354 num_bytes + empty_cluster + empty_size) {
7355 if (block_group->free_space_ctl->free_space >
7358 block_group->free_space_ctl->free_space;
7359 spin_unlock(&block_group->free_space_ctl->tree_lock);
7362 spin_unlock(&block_group->free_space_ctl->tree_lock);
7364 offset = btrfs_find_space_for_alloc(block_group, search_start,
7365 num_bytes, empty_size,
7368 * If we didn't find a chunk, and we haven't failed on this
7369 * block group before, and this block group is in the middle of
7370 * caching and we are ok with waiting, then go ahead and wait
7371 * for progress to be made, and set failed_alloc to true.
7373 * If failed_alloc is true then we've already waited on this
7374 * block group once and should move on to the next block group.
7376 if (!offset && !failed_alloc && !cached &&
7377 loop > LOOP_CACHING_NOWAIT) {
7378 wait_block_group_cache_progress(block_group,
7379 num_bytes + empty_size);
7380 failed_alloc = true;
7381 goto have_block_group;
7382 } else if (!offset) {
7386 search_start = ALIGN(offset, root->stripesize);
7388 /* move on to the next group */
7389 if (search_start + num_bytes >
7390 block_group->key.objectid + block_group->key.offset) {
7391 btrfs_add_free_space(block_group, offset, num_bytes);
7395 if (offset < search_start)
7396 btrfs_add_free_space(block_group, offset,
7397 search_start - offset);
7398 BUG_ON(offset > search_start);
7400 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7401 alloc_type, delalloc);
7402 if (ret == -EAGAIN) {
7403 btrfs_add_free_space(block_group, offset, num_bytes);
7407 /* we are all good, lets return */
7408 ins->objectid = search_start;
7409 ins->offset = num_bytes;
7411 trace_btrfs_reserve_extent(orig_root, block_group,
7412 search_start, num_bytes);
7413 btrfs_release_block_group(block_group, delalloc);
7416 failed_cluster_refill = false;
7417 failed_alloc = false;
7418 BUG_ON(index != get_block_group_index(block_group));
7419 btrfs_release_block_group(block_group, delalloc);
7421 up_read(&space_info->groups_sem);
7423 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7424 && !orig_have_caching_bg)
7425 orig_have_caching_bg = true;
7427 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7430 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7434 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7435 * caching kthreads as we move along
7436 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7437 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7438 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7441 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7443 if (loop == LOOP_CACHING_NOWAIT) {
7445 * We want to skip the LOOP_CACHING_WAIT step if we
7446 * don't have any unached bgs and we've alrelady done a
7447 * full search through.
7449 if (orig_have_caching_bg || !full_search)
7450 loop = LOOP_CACHING_WAIT;
7452 loop = LOOP_ALLOC_CHUNK;
7457 if (loop == LOOP_ALLOC_CHUNK) {
7458 struct btrfs_trans_handle *trans;
7461 trans = current->journal_info;
7465 trans = btrfs_join_transaction(root);
7467 if (IS_ERR(trans)) {
7468 ret = PTR_ERR(trans);
7472 ret = do_chunk_alloc(trans, root, flags,
7476 * If we can't allocate a new chunk we've already looped
7477 * through at least once, move on to the NO_EMPTY_SIZE
7481 loop = LOOP_NO_EMPTY_SIZE;
7484 * Do not bail out on ENOSPC since we
7485 * can do more things.
7487 if (ret < 0 && ret != -ENOSPC)
7488 btrfs_abort_transaction(trans,
7493 btrfs_end_transaction(trans, root);
7498 if (loop == LOOP_NO_EMPTY_SIZE) {
7500 * Don't loop again if we already have no empty_size and
7503 if (empty_size == 0 &&
7504 empty_cluster == 0) {
7513 } else if (!ins->objectid) {
7515 } else if (ins->objectid) {
7516 if (!use_cluster && last_ptr) {
7517 spin_lock(&last_ptr->lock);
7518 last_ptr->window_start = ins->objectid;
7519 spin_unlock(&last_ptr->lock);
7524 if (ret == -ENOSPC) {
7525 spin_lock(&space_info->lock);
7526 space_info->max_extent_size = max_extent_size;
7527 spin_unlock(&space_info->lock);
7528 ins->offset = max_extent_size;
7533 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7534 int dump_block_groups)
7536 struct btrfs_block_group_cache *cache;
7539 spin_lock(&info->lock);
7540 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7542 info->total_bytes - info->bytes_used - info->bytes_pinned -
7543 info->bytes_reserved - info->bytes_readonly,
7544 (info->full) ? "" : "not ");
7545 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7546 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7547 info->total_bytes, info->bytes_used, info->bytes_pinned,
7548 info->bytes_reserved, info->bytes_may_use,
7549 info->bytes_readonly);
7550 spin_unlock(&info->lock);
7552 if (!dump_block_groups)
7555 down_read(&info->groups_sem);
7557 list_for_each_entry(cache, &info->block_groups[index], list) {
7558 spin_lock(&cache->lock);
7559 printk(KERN_INFO "BTRFS: "
7560 "block group %llu has %llu bytes, "
7561 "%llu used %llu pinned %llu reserved %s\n",
7562 cache->key.objectid, cache->key.offset,
7563 btrfs_block_group_used(&cache->item), cache->pinned,
7564 cache->reserved, cache->ro ? "[readonly]" : "");
7565 btrfs_dump_free_space(cache, bytes);
7566 spin_unlock(&cache->lock);
7568 if (++index < BTRFS_NR_RAID_TYPES)
7570 up_read(&info->groups_sem);
7573 int btrfs_reserve_extent(struct btrfs_root *root,
7574 u64 num_bytes, u64 min_alloc_size,
7575 u64 empty_size, u64 hint_byte,
7576 struct btrfs_key *ins, int is_data, int delalloc)
7578 bool final_tried = num_bytes == min_alloc_size;
7582 flags = btrfs_get_alloc_profile(root, is_data);
7584 WARN_ON(num_bytes < root->sectorsize);
7585 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7588 if (ret == -ENOSPC) {
7589 if (!final_tried && ins->offset) {
7590 num_bytes = min(num_bytes >> 1, ins->offset);
7591 num_bytes = round_down(num_bytes, root->sectorsize);
7592 num_bytes = max(num_bytes, min_alloc_size);
7593 if (num_bytes == min_alloc_size)
7596 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7597 struct btrfs_space_info *sinfo;
7599 sinfo = __find_space_info(root->fs_info, flags);
7600 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7603 dump_space_info(sinfo, num_bytes, 1);
7610 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7612 int pin, int delalloc)
7614 struct btrfs_block_group_cache *cache;
7617 cache = btrfs_lookup_block_group(root->fs_info, start);
7619 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7625 pin_down_extent(root, cache, start, len, 1);
7627 if (btrfs_test_opt(root, DISCARD))
7628 ret = btrfs_discard_extent(root, start, len, NULL);
7629 btrfs_add_free_space(cache, start, len);
7630 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7633 btrfs_put_block_group(cache);
7635 trace_btrfs_reserved_extent_free(root, start, len);
7640 int btrfs_free_reserved_extent(struct btrfs_root *root,
7641 u64 start, u64 len, int delalloc)
7643 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7646 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7649 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7652 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7653 struct btrfs_root *root,
7654 u64 parent, u64 root_objectid,
7655 u64 flags, u64 owner, u64 offset,
7656 struct btrfs_key *ins, int ref_mod)
7659 struct btrfs_fs_info *fs_info = root->fs_info;
7660 struct btrfs_extent_item *extent_item;
7661 struct btrfs_extent_inline_ref *iref;
7662 struct btrfs_path *path;
7663 struct extent_buffer *leaf;
7668 type = BTRFS_SHARED_DATA_REF_KEY;
7670 type = BTRFS_EXTENT_DATA_REF_KEY;
7672 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7674 path = btrfs_alloc_path();
7678 path->leave_spinning = 1;
7679 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7682 btrfs_free_path(path);
7686 leaf = path->nodes[0];
7687 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7688 struct btrfs_extent_item);
7689 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7690 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7691 btrfs_set_extent_flags(leaf, extent_item,
7692 flags | BTRFS_EXTENT_FLAG_DATA);
7694 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7695 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7697 struct btrfs_shared_data_ref *ref;
7698 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7699 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7700 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7702 struct btrfs_extent_data_ref *ref;
7703 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7704 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7705 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7706 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7707 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7710 btrfs_mark_buffer_dirty(path->nodes[0]);
7711 btrfs_free_path(path);
7713 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7718 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7719 if (ret) { /* -ENOENT, logic error */
7720 btrfs_err(fs_info, "update block group failed for %llu %llu",
7721 ins->objectid, ins->offset);
7724 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7728 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7729 struct btrfs_root *root,
7730 u64 parent, u64 root_objectid,
7731 u64 flags, struct btrfs_disk_key *key,
7732 int level, struct btrfs_key *ins)
7735 struct btrfs_fs_info *fs_info = root->fs_info;
7736 struct btrfs_extent_item *extent_item;
7737 struct btrfs_tree_block_info *block_info;
7738 struct btrfs_extent_inline_ref *iref;
7739 struct btrfs_path *path;
7740 struct extent_buffer *leaf;
7741 u32 size = sizeof(*extent_item) + sizeof(*iref);
7742 u64 num_bytes = ins->offset;
7743 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7746 if (!skinny_metadata)
7747 size += sizeof(*block_info);
7749 path = btrfs_alloc_path();
7751 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7756 path->leave_spinning = 1;
7757 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7760 btrfs_free_path(path);
7761 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7766 leaf = path->nodes[0];
7767 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7768 struct btrfs_extent_item);
7769 btrfs_set_extent_refs(leaf, extent_item, 1);
7770 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7771 btrfs_set_extent_flags(leaf, extent_item,
7772 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7774 if (skinny_metadata) {
7775 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7776 num_bytes = root->nodesize;
7778 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7779 btrfs_set_tree_block_key(leaf, block_info, key);
7780 btrfs_set_tree_block_level(leaf, block_info, level);
7781 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7785 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7786 btrfs_set_extent_inline_ref_type(leaf, iref,
7787 BTRFS_SHARED_BLOCK_REF_KEY);
7788 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7790 btrfs_set_extent_inline_ref_type(leaf, iref,
7791 BTRFS_TREE_BLOCK_REF_KEY);
7792 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7795 btrfs_mark_buffer_dirty(leaf);
7796 btrfs_free_path(path);
7798 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7803 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7805 if (ret) { /* -ENOENT, logic error */
7806 btrfs_err(fs_info, "update block group failed for %llu %llu",
7807 ins->objectid, ins->offset);
7811 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7815 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7816 struct btrfs_root *root,
7817 u64 root_objectid, u64 owner,
7818 u64 offset, u64 ram_bytes,
7819 struct btrfs_key *ins)
7823 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7825 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7827 root_objectid, owner, offset,
7828 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
7834 * this is used by the tree logging recovery code. It records that
7835 * an extent has been allocated and makes sure to clear the free
7836 * space cache bits as well
7838 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7839 struct btrfs_root *root,
7840 u64 root_objectid, u64 owner, u64 offset,
7841 struct btrfs_key *ins)
7844 struct btrfs_block_group_cache *block_group;
7847 * Mixed block groups will exclude before processing the log so we only
7848 * need to do the exlude dance if this fs isn't mixed.
7850 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7851 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7856 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7860 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7861 RESERVE_ALLOC_NO_ACCOUNT, 0);
7862 BUG_ON(ret); /* logic error */
7863 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7864 0, owner, offset, ins, 1);
7865 btrfs_put_block_group(block_group);
7869 static struct extent_buffer *
7870 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7871 u64 bytenr, int level)
7873 struct extent_buffer *buf;
7875 buf = btrfs_find_create_tree_block(root, bytenr);
7877 return ERR_PTR(-ENOMEM);
7878 btrfs_set_header_generation(buf, trans->transid);
7879 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7880 btrfs_tree_lock(buf);
7881 clean_tree_block(trans, root->fs_info, buf);
7882 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7884 btrfs_set_lock_blocking(buf);
7885 set_extent_buffer_uptodate(buf);
7887 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7888 buf->log_index = root->log_transid % 2;
7890 * we allow two log transactions at a time, use different
7891 * EXENT bit to differentiate dirty pages.
7893 if (buf->log_index == 0)
7894 set_extent_dirty(&root->dirty_log_pages, buf->start,
7895 buf->start + buf->len - 1, GFP_NOFS);
7897 set_extent_new(&root->dirty_log_pages, buf->start,
7898 buf->start + buf->len - 1, GFP_NOFS);
7900 buf->log_index = -1;
7901 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7902 buf->start + buf->len - 1, GFP_NOFS);
7904 trans->blocks_used++;
7905 /* this returns a buffer locked for blocking */
7909 static struct btrfs_block_rsv *
7910 use_block_rsv(struct btrfs_trans_handle *trans,
7911 struct btrfs_root *root, u32 blocksize)
7913 struct btrfs_block_rsv *block_rsv;
7914 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7916 bool global_updated = false;
7918 block_rsv = get_block_rsv(trans, root);
7920 if (unlikely(block_rsv->size == 0))
7923 ret = block_rsv_use_bytes(block_rsv, blocksize);
7927 if (block_rsv->failfast)
7928 return ERR_PTR(ret);
7930 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7931 global_updated = true;
7932 update_global_block_rsv(root->fs_info);
7936 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7937 static DEFINE_RATELIMIT_STATE(_rs,
7938 DEFAULT_RATELIMIT_INTERVAL * 10,
7939 /*DEFAULT_RATELIMIT_BURST*/ 1);
7940 if (__ratelimit(&_rs))
7942 "BTRFS: block rsv returned %d\n", ret);
7945 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7946 BTRFS_RESERVE_NO_FLUSH);
7950 * If we couldn't reserve metadata bytes try and use some from
7951 * the global reserve if its space type is the same as the global
7954 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7955 block_rsv->space_info == global_rsv->space_info) {
7956 ret = block_rsv_use_bytes(global_rsv, blocksize);
7960 return ERR_PTR(ret);
7963 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7964 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7966 block_rsv_add_bytes(block_rsv, blocksize, 0);
7967 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7971 * finds a free extent and does all the dirty work required for allocation
7972 * returns the tree buffer or an ERR_PTR on error.
7974 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7975 struct btrfs_root *root,
7976 u64 parent, u64 root_objectid,
7977 struct btrfs_disk_key *key, int level,
7978 u64 hint, u64 empty_size)
7980 struct btrfs_key ins;
7981 struct btrfs_block_rsv *block_rsv;
7982 struct extent_buffer *buf;
7983 struct btrfs_delayed_extent_op *extent_op;
7986 u32 blocksize = root->nodesize;
7987 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7990 if (btrfs_test_is_dummy_root(root)) {
7991 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7994 root->alloc_bytenr += blocksize;
7998 block_rsv = use_block_rsv(trans, root, blocksize);
7999 if (IS_ERR(block_rsv))
8000 return ERR_CAST(block_rsv);
8002 ret = btrfs_reserve_extent(root, blocksize, blocksize,
8003 empty_size, hint, &ins, 0, 0);
8007 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8010 goto out_free_reserved;
8013 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8015 parent = ins.objectid;
8016 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8020 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8021 extent_op = btrfs_alloc_delayed_extent_op();
8027 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8029 memset(&extent_op->key, 0, sizeof(extent_op->key));
8030 extent_op->flags_to_set = flags;
8031 if (skinny_metadata)
8032 extent_op->update_key = 0;
8034 extent_op->update_key = 1;
8035 extent_op->update_flags = 1;
8036 extent_op->is_data = 0;
8037 extent_op->level = level;
8039 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8040 ins.objectid, ins.offset,
8041 parent, root_objectid, level,
8042 BTRFS_ADD_DELAYED_EXTENT,
8045 goto out_free_delayed;
8050 btrfs_free_delayed_extent_op(extent_op);
8052 free_extent_buffer(buf);
8054 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8056 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8057 return ERR_PTR(ret);
8060 struct walk_control {
8061 u64 refs[BTRFS_MAX_LEVEL];
8062 u64 flags[BTRFS_MAX_LEVEL];
8063 struct btrfs_key update_progress;
8074 #define DROP_REFERENCE 1
8075 #define UPDATE_BACKREF 2
8077 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8078 struct btrfs_root *root,
8079 struct walk_control *wc,
8080 struct btrfs_path *path)
8088 struct btrfs_key key;
8089 struct extent_buffer *eb;
8094 if (path->slots[wc->level] < wc->reada_slot) {
8095 wc->reada_count = wc->reada_count * 2 / 3;
8096 wc->reada_count = max(wc->reada_count, 2);
8098 wc->reada_count = wc->reada_count * 3 / 2;
8099 wc->reada_count = min_t(int, wc->reada_count,
8100 BTRFS_NODEPTRS_PER_BLOCK(root));
8103 eb = path->nodes[wc->level];
8104 nritems = btrfs_header_nritems(eb);
8105 blocksize = root->nodesize;
8107 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8108 if (nread >= wc->reada_count)
8112 bytenr = btrfs_node_blockptr(eb, slot);
8113 generation = btrfs_node_ptr_generation(eb, slot);
8115 if (slot == path->slots[wc->level])
8118 if (wc->stage == UPDATE_BACKREF &&
8119 generation <= root->root_key.offset)
8122 /* We don't lock the tree block, it's OK to be racy here */
8123 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8124 wc->level - 1, 1, &refs,
8126 /* We don't care about errors in readahead. */
8131 if (wc->stage == DROP_REFERENCE) {
8135 if (wc->level == 1 &&
8136 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8138 if (!wc->update_ref ||
8139 generation <= root->root_key.offset)
8141 btrfs_node_key_to_cpu(eb, &key, slot);
8142 ret = btrfs_comp_cpu_keys(&key,
8143 &wc->update_progress);
8147 if (wc->level == 1 &&
8148 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8152 readahead_tree_block(root, bytenr);
8155 wc->reada_slot = slot;
8159 * These may not be seen by the usual inc/dec ref code so we have to
8162 static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8163 struct btrfs_root *root, u64 bytenr,
8166 struct btrfs_qgroup_extent_record *qrecord;
8167 struct btrfs_delayed_ref_root *delayed_refs;
8169 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8173 qrecord->bytenr = bytenr;
8174 qrecord->num_bytes = num_bytes;
8175 qrecord->old_roots = NULL;
8177 delayed_refs = &trans->transaction->delayed_refs;
8178 spin_lock(&delayed_refs->lock);
8179 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8181 spin_unlock(&delayed_refs->lock);
8186 static int account_leaf_items(struct btrfs_trans_handle *trans,
8187 struct btrfs_root *root,
8188 struct extent_buffer *eb)
8190 int nr = btrfs_header_nritems(eb);
8191 int i, extent_type, ret;
8192 struct btrfs_key key;
8193 struct btrfs_file_extent_item *fi;
8194 u64 bytenr, num_bytes;
8196 /* We can be called directly from walk_up_proc() */
8197 if (!root->fs_info->quota_enabled)
8200 for (i = 0; i < nr; i++) {
8201 btrfs_item_key_to_cpu(eb, &key, i);
8203 if (key.type != BTRFS_EXTENT_DATA_KEY)
8206 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8207 /* filter out non qgroup-accountable extents */
8208 extent_type = btrfs_file_extent_type(eb, fi);
8210 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8213 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8217 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8219 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8227 * Walk up the tree from the bottom, freeing leaves and any interior
8228 * nodes which have had all slots visited. If a node (leaf or
8229 * interior) is freed, the node above it will have it's slot
8230 * incremented. The root node will never be freed.
8232 * At the end of this function, we should have a path which has all
8233 * slots incremented to the next position for a search. If we need to
8234 * read a new node it will be NULL and the node above it will have the
8235 * correct slot selected for a later read.
8237 * If we increment the root nodes slot counter past the number of
8238 * elements, 1 is returned to signal completion of the search.
8240 static int adjust_slots_upwards(struct btrfs_root *root,
8241 struct btrfs_path *path, int root_level)
8245 struct extent_buffer *eb;
8247 if (root_level == 0)
8250 while (level <= root_level) {
8251 eb = path->nodes[level];
8252 nr = btrfs_header_nritems(eb);
8253 path->slots[level]++;
8254 slot = path->slots[level];
8255 if (slot >= nr || level == 0) {
8257 * Don't free the root - we will detect this
8258 * condition after our loop and return a
8259 * positive value for caller to stop walking the tree.
8261 if (level != root_level) {
8262 btrfs_tree_unlock_rw(eb, path->locks[level]);
8263 path->locks[level] = 0;
8265 free_extent_buffer(eb);
8266 path->nodes[level] = NULL;
8267 path->slots[level] = 0;
8271 * We have a valid slot to walk back down
8272 * from. Stop here so caller can process these
8281 eb = path->nodes[root_level];
8282 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8289 * root_eb is the subtree root and is locked before this function is called.
8291 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8292 struct btrfs_root *root,
8293 struct extent_buffer *root_eb,
8299 struct extent_buffer *eb = root_eb;
8300 struct btrfs_path *path = NULL;
8302 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8303 BUG_ON(root_eb == NULL);
8305 if (!root->fs_info->quota_enabled)
8308 if (!extent_buffer_uptodate(root_eb)) {
8309 ret = btrfs_read_buffer(root_eb, root_gen);
8314 if (root_level == 0) {
8315 ret = account_leaf_items(trans, root, root_eb);
8319 path = btrfs_alloc_path();
8324 * Walk down the tree. Missing extent blocks are filled in as
8325 * we go. Metadata is accounted every time we read a new
8328 * When we reach a leaf, we account for file extent items in it,
8329 * walk back up the tree (adjusting slot pointers as we go)
8330 * and restart the search process.
8332 extent_buffer_get(root_eb); /* For path */
8333 path->nodes[root_level] = root_eb;
8334 path->slots[root_level] = 0;
8335 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8338 while (level >= 0) {
8339 if (path->nodes[level] == NULL) {
8344 /* We need to get child blockptr/gen from
8345 * parent before we can read it. */
8346 eb = path->nodes[level + 1];
8347 parent_slot = path->slots[level + 1];
8348 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8349 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8351 eb = read_tree_block(root, child_bytenr, child_gen);
8355 } else if (!extent_buffer_uptodate(eb)) {
8356 free_extent_buffer(eb);
8361 path->nodes[level] = eb;
8362 path->slots[level] = 0;
8364 btrfs_tree_read_lock(eb);
8365 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8366 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8368 ret = record_one_subtree_extent(trans, root, child_bytenr,
8375 ret = account_leaf_items(trans, root, path->nodes[level]);
8379 /* Nonzero return here means we completed our search */
8380 ret = adjust_slots_upwards(root, path, root_level);
8384 /* Restart search with new slots */
8393 btrfs_free_path(path);
8399 * helper to process tree block while walking down the tree.
8401 * when wc->stage == UPDATE_BACKREF, this function updates
8402 * back refs for pointers in the block.
8404 * NOTE: return value 1 means we should stop walking down.
8406 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8407 struct btrfs_root *root,
8408 struct btrfs_path *path,
8409 struct walk_control *wc, int lookup_info)
8411 int level = wc->level;
8412 struct extent_buffer *eb = path->nodes[level];
8413 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8416 if (wc->stage == UPDATE_BACKREF &&
8417 btrfs_header_owner(eb) != root->root_key.objectid)
8421 * when reference count of tree block is 1, it won't increase
8422 * again. once full backref flag is set, we never clear it.
8425 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8426 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8427 BUG_ON(!path->locks[level]);
8428 ret = btrfs_lookup_extent_info(trans, root,
8429 eb->start, level, 1,
8432 BUG_ON(ret == -ENOMEM);
8435 BUG_ON(wc->refs[level] == 0);
8438 if (wc->stage == DROP_REFERENCE) {
8439 if (wc->refs[level] > 1)
8442 if (path->locks[level] && !wc->keep_locks) {
8443 btrfs_tree_unlock_rw(eb, path->locks[level]);
8444 path->locks[level] = 0;
8449 /* wc->stage == UPDATE_BACKREF */
8450 if (!(wc->flags[level] & flag)) {
8451 BUG_ON(!path->locks[level]);
8452 ret = btrfs_inc_ref(trans, root, eb, 1);
8453 BUG_ON(ret); /* -ENOMEM */
8454 ret = btrfs_dec_ref(trans, root, eb, 0);
8455 BUG_ON(ret); /* -ENOMEM */
8456 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8458 btrfs_header_level(eb), 0);
8459 BUG_ON(ret); /* -ENOMEM */
8460 wc->flags[level] |= flag;
8464 * the block is shared by multiple trees, so it's not good to
8465 * keep the tree lock
8467 if (path->locks[level] && level > 0) {
8468 btrfs_tree_unlock_rw(eb, path->locks[level]);
8469 path->locks[level] = 0;
8475 * helper to process tree block pointer.
8477 * when wc->stage == DROP_REFERENCE, this function checks
8478 * reference count of the block pointed to. if the block
8479 * is shared and we need update back refs for the subtree
8480 * rooted at the block, this function changes wc->stage to
8481 * UPDATE_BACKREF. if the block is shared and there is no
8482 * need to update back, this function drops the reference
8485 * NOTE: return value 1 means we should stop walking down.
8487 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8488 struct btrfs_root *root,
8489 struct btrfs_path *path,
8490 struct walk_control *wc, int *lookup_info)
8496 struct btrfs_key key;
8497 struct extent_buffer *next;
8498 int level = wc->level;
8501 bool need_account = false;
8503 generation = btrfs_node_ptr_generation(path->nodes[level],
8504 path->slots[level]);
8506 * if the lower level block was created before the snapshot
8507 * was created, we know there is no need to update back refs
8510 if (wc->stage == UPDATE_BACKREF &&
8511 generation <= root->root_key.offset) {
8516 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8517 blocksize = root->nodesize;
8519 next = btrfs_find_tree_block(root->fs_info, bytenr);
8521 next = btrfs_find_create_tree_block(root, bytenr);
8524 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8528 btrfs_tree_lock(next);
8529 btrfs_set_lock_blocking(next);
8531 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8532 &wc->refs[level - 1],
8533 &wc->flags[level - 1]);
8535 btrfs_tree_unlock(next);
8539 if (unlikely(wc->refs[level - 1] == 0)) {
8540 btrfs_err(root->fs_info, "Missing references.");
8545 if (wc->stage == DROP_REFERENCE) {
8546 if (wc->refs[level - 1] > 1) {
8547 need_account = true;
8549 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8552 if (!wc->update_ref ||
8553 generation <= root->root_key.offset)
8556 btrfs_node_key_to_cpu(path->nodes[level], &key,
8557 path->slots[level]);
8558 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8562 wc->stage = UPDATE_BACKREF;
8563 wc->shared_level = level - 1;
8567 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8571 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8572 btrfs_tree_unlock(next);
8573 free_extent_buffer(next);
8579 if (reada && level == 1)
8580 reada_walk_down(trans, root, wc, path);
8581 next = read_tree_block(root, bytenr, generation);
8583 return PTR_ERR(next);
8584 } else if (!extent_buffer_uptodate(next)) {
8585 free_extent_buffer(next);
8588 btrfs_tree_lock(next);
8589 btrfs_set_lock_blocking(next);
8593 BUG_ON(level != btrfs_header_level(next));
8594 path->nodes[level] = next;
8595 path->slots[level] = 0;
8596 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8602 wc->refs[level - 1] = 0;
8603 wc->flags[level - 1] = 0;
8604 if (wc->stage == DROP_REFERENCE) {
8605 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8606 parent = path->nodes[level]->start;
8608 BUG_ON(root->root_key.objectid !=
8609 btrfs_header_owner(path->nodes[level]));
8614 ret = account_shared_subtree(trans, root, next,
8615 generation, level - 1);
8617 btrfs_err_rl(root->fs_info,
8619 "%d accounting shared subtree. Quota "
8620 "is out of sync, rescan required.",
8624 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8625 root->root_key.objectid, level - 1, 0);
8626 BUG_ON(ret); /* -ENOMEM */
8628 btrfs_tree_unlock(next);
8629 free_extent_buffer(next);
8635 * helper to process tree block while walking up the tree.
8637 * when wc->stage == DROP_REFERENCE, this function drops
8638 * reference count on the block.
8640 * when wc->stage == UPDATE_BACKREF, this function changes
8641 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8642 * to UPDATE_BACKREF previously while processing the block.
8644 * NOTE: return value 1 means we should stop walking up.
8646 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8647 struct btrfs_root *root,
8648 struct btrfs_path *path,
8649 struct walk_control *wc)
8652 int level = wc->level;
8653 struct extent_buffer *eb = path->nodes[level];
8656 if (wc->stage == UPDATE_BACKREF) {
8657 BUG_ON(wc->shared_level < level);
8658 if (level < wc->shared_level)
8661 ret = find_next_key(path, level + 1, &wc->update_progress);
8665 wc->stage = DROP_REFERENCE;
8666 wc->shared_level = -1;
8667 path->slots[level] = 0;
8670 * check reference count again if the block isn't locked.
8671 * we should start walking down the tree again if reference
8674 if (!path->locks[level]) {
8676 btrfs_tree_lock(eb);
8677 btrfs_set_lock_blocking(eb);
8678 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8680 ret = btrfs_lookup_extent_info(trans, root,
8681 eb->start, level, 1,
8685 btrfs_tree_unlock_rw(eb, path->locks[level]);
8686 path->locks[level] = 0;
8689 BUG_ON(wc->refs[level] == 0);
8690 if (wc->refs[level] == 1) {
8691 btrfs_tree_unlock_rw(eb, path->locks[level]);
8692 path->locks[level] = 0;
8698 /* wc->stage == DROP_REFERENCE */
8699 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8701 if (wc->refs[level] == 1) {
8703 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8704 ret = btrfs_dec_ref(trans, root, eb, 1);
8706 ret = btrfs_dec_ref(trans, root, eb, 0);
8707 BUG_ON(ret); /* -ENOMEM */
8708 ret = account_leaf_items(trans, root, eb);
8710 btrfs_err_rl(root->fs_info,
8712 "%d accounting leaf items. Quota "
8713 "is out of sync, rescan required.",
8717 /* make block locked assertion in clean_tree_block happy */
8718 if (!path->locks[level] &&
8719 btrfs_header_generation(eb) == trans->transid) {
8720 btrfs_tree_lock(eb);
8721 btrfs_set_lock_blocking(eb);
8722 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8724 clean_tree_block(trans, root->fs_info, eb);
8727 if (eb == root->node) {
8728 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8731 BUG_ON(root->root_key.objectid !=
8732 btrfs_header_owner(eb));
8734 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8735 parent = path->nodes[level + 1]->start;
8737 BUG_ON(root->root_key.objectid !=
8738 btrfs_header_owner(path->nodes[level + 1]));
8741 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8743 wc->refs[level] = 0;
8744 wc->flags[level] = 0;
8748 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8749 struct btrfs_root *root,
8750 struct btrfs_path *path,
8751 struct walk_control *wc)
8753 int level = wc->level;
8754 int lookup_info = 1;
8757 while (level >= 0) {
8758 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8765 if (path->slots[level] >=
8766 btrfs_header_nritems(path->nodes[level]))
8769 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8771 path->slots[level]++;
8780 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8781 struct btrfs_root *root,
8782 struct btrfs_path *path,
8783 struct walk_control *wc, int max_level)
8785 int level = wc->level;
8788 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8789 while (level < max_level && path->nodes[level]) {
8791 if (path->slots[level] + 1 <
8792 btrfs_header_nritems(path->nodes[level])) {
8793 path->slots[level]++;
8796 ret = walk_up_proc(trans, root, path, wc);
8800 if (path->locks[level]) {
8801 btrfs_tree_unlock_rw(path->nodes[level],
8802 path->locks[level]);
8803 path->locks[level] = 0;
8805 free_extent_buffer(path->nodes[level]);
8806 path->nodes[level] = NULL;
8814 * drop a subvolume tree.
8816 * this function traverses the tree freeing any blocks that only
8817 * referenced by the tree.
8819 * when a shared tree block is found. this function decreases its
8820 * reference count by one. if update_ref is true, this function
8821 * also make sure backrefs for the shared block and all lower level
8822 * blocks are properly updated.
8824 * If called with for_reloc == 0, may exit early with -EAGAIN
8826 int btrfs_drop_snapshot(struct btrfs_root *root,
8827 struct btrfs_block_rsv *block_rsv, int update_ref,
8830 struct btrfs_path *path;
8831 struct btrfs_trans_handle *trans;
8832 struct btrfs_root *tree_root = root->fs_info->tree_root;
8833 struct btrfs_root_item *root_item = &root->root_item;
8834 struct walk_control *wc;
8835 struct btrfs_key key;
8839 bool root_dropped = false;
8841 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8843 path = btrfs_alloc_path();
8849 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8851 btrfs_free_path(path);
8856 trans = btrfs_start_transaction(tree_root, 0);
8857 if (IS_ERR(trans)) {
8858 err = PTR_ERR(trans);
8863 trans->block_rsv = block_rsv;
8865 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8866 level = btrfs_header_level(root->node);
8867 path->nodes[level] = btrfs_lock_root_node(root);
8868 btrfs_set_lock_blocking(path->nodes[level]);
8869 path->slots[level] = 0;
8870 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8871 memset(&wc->update_progress, 0,
8872 sizeof(wc->update_progress));
8874 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8875 memcpy(&wc->update_progress, &key,
8876 sizeof(wc->update_progress));
8878 level = root_item->drop_level;
8880 path->lowest_level = level;
8881 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8882 path->lowest_level = 0;
8890 * unlock our path, this is safe because only this
8891 * function is allowed to delete this snapshot
8893 btrfs_unlock_up_safe(path, 0);
8895 level = btrfs_header_level(root->node);
8897 btrfs_tree_lock(path->nodes[level]);
8898 btrfs_set_lock_blocking(path->nodes[level]);
8899 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8901 ret = btrfs_lookup_extent_info(trans, root,
8902 path->nodes[level]->start,
8903 level, 1, &wc->refs[level],
8909 BUG_ON(wc->refs[level] == 0);
8911 if (level == root_item->drop_level)
8914 btrfs_tree_unlock(path->nodes[level]);
8915 path->locks[level] = 0;
8916 WARN_ON(wc->refs[level] != 1);
8922 wc->shared_level = -1;
8923 wc->stage = DROP_REFERENCE;
8924 wc->update_ref = update_ref;
8926 wc->for_reloc = for_reloc;
8927 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8931 ret = walk_down_tree(trans, root, path, wc);
8937 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8944 BUG_ON(wc->stage != DROP_REFERENCE);
8948 if (wc->stage == DROP_REFERENCE) {
8950 btrfs_node_key(path->nodes[level],
8951 &root_item->drop_progress,
8952 path->slots[level]);
8953 root_item->drop_level = level;
8956 BUG_ON(wc->level == 0);
8957 if (btrfs_should_end_transaction(trans, tree_root) ||
8958 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8959 ret = btrfs_update_root(trans, tree_root,
8963 btrfs_abort_transaction(trans, tree_root, ret);
8968 btrfs_end_transaction_throttle(trans, tree_root);
8969 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8970 pr_debug("BTRFS: drop snapshot early exit\n");
8975 trans = btrfs_start_transaction(tree_root, 0);
8976 if (IS_ERR(trans)) {
8977 err = PTR_ERR(trans);
8981 trans->block_rsv = block_rsv;
8984 btrfs_release_path(path);
8988 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8990 btrfs_abort_transaction(trans, tree_root, ret);
8994 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8995 ret = btrfs_find_root(tree_root, &root->root_key, path,
8998 btrfs_abort_transaction(trans, tree_root, ret);
9001 } else if (ret > 0) {
9002 /* if we fail to delete the orphan item this time
9003 * around, it'll get picked up the next time.
9005 * The most common failure here is just -ENOENT.
9007 btrfs_del_orphan_item(trans, tree_root,
9008 root->root_key.objectid);
9012 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9013 btrfs_add_dropped_root(trans, root);
9015 free_extent_buffer(root->node);
9016 free_extent_buffer(root->commit_root);
9017 btrfs_put_fs_root(root);
9019 root_dropped = true;
9021 btrfs_end_transaction_throttle(trans, tree_root);
9024 btrfs_free_path(path);
9027 * So if we need to stop dropping the snapshot for whatever reason we
9028 * need to make sure to add it back to the dead root list so that we
9029 * keep trying to do the work later. This also cleans up roots if we
9030 * don't have it in the radix (like when we recover after a power fail
9031 * or unmount) so we don't leak memory.
9033 if (!for_reloc && root_dropped == false)
9034 btrfs_add_dead_root(root);
9035 if (err && err != -EAGAIN)
9036 btrfs_std_error(root->fs_info, err, NULL);
9041 * drop subtree rooted at tree block 'node'.
9043 * NOTE: this function will unlock and release tree block 'node'
9044 * only used by relocation code
9046 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9047 struct btrfs_root *root,
9048 struct extent_buffer *node,
9049 struct extent_buffer *parent)
9051 struct btrfs_path *path;
9052 struct walk_control *wc;
9058 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9060 path = btrfs_alloc_path();
9064 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9066 btrfs_free_path(path);
9070 btrfs_assert_tree_locked(parent);
9071 parent_level = btrfs_header_level(parent);
9072 extent_buffer_get(parent);
9073 path->nodes[parent_level] = parent;
9074 path->slots[parent_level] = btrfs_header_nritems(parent);
9076 btrfs_assert_tree_locked(node);
9077 level = btrfs_header_level(node);
9078 path->nodes[level] = node;
9079 path->slots[level] = 0;
9080 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9082 wc->refs[parent_level] = 1;
9083 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9085 wc->shared_level = -1;
9086 wc->stage = DROP_REFERENCE;
9090 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9093 wret = walk_down_tree(trans, root, path, wc);
9099 wret = walk_up_tree(trans, root, path, wc, parent_level);
9107 btrfs_free_path(path);
9111 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9117 * if restripe for this chunk_type is on pick target profile and
9118 * return, otherwise do the usual balance
9120 stripped = get_restripe_target(root->fs_info, flags);
9122 return extended_to_chunk(stripped);
9124 num_devices = root->fs_info->fs_devices->rw_devices;
9126 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9127 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9128 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9130 if (num_devices == 1) {
9131 stripped |= BTRFS_BLOCK_GROUP_DUP;
9132 stripped = flags & ~stripped;
9134 /* turn raid0 into single device chunks */
9135 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9138 /* turn mirroring into duplication */
9139 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9140 BTRFS_BLOCK_GROUP_RAID10))
9141 return stripped | BTRFS_BLOCK_GROUP_DUP;
9143 /* they already had raid on here, just return */
9144 if (flags & stripped)
9147 stripped |= BTRFS_BLOCK_GROUP_DUP;
9148 stripped = flags & ~stripped;
9150 /* switch duplicated blocks with raid1 */
9151 if (flags & BTRFS_BLOCK_GROUP_DUP)
9152 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9154 /* this is drive concat, leave it alone */
9160 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9162 struct btrfs_space_info *sinfo = cache->space_info;
9164 u64 min_allocable_bytes;
9168 * We need some metadata space and system metadata space for
9169 * allocating chunks in some corner cases until we force to set
9170 * it to be readonly.
9173 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9175 min_allocable_bytes = 1 * 1024 * 1024;
9177 min_allocable_bytes = 0;
9179 spin_lock(&sinfo->lock);
9180 spin_lock(&cache->lock);
9188 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9189 cache->bytes_super - btrfs_block_group_used(&cache->item);
9191 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9192 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9193 min_allocable_bytes <= sinfo->total_bytes) {
9194 sinfo->bytes_readonly += num_bytes;
9196 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9200 spin_unlock(&cache->lock);
9201 spin_unlock(&sinfo->lock);
9205 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9206 struct btrfs_block_group_cache *cache)
9209 struct btrfs_trans_handle *trans;
9214 trans = btrfs_join_transaction(root);
9216 return PTR_ERR(trans);
9219 * we're not allowed to set block groups readonly after the dirty
9220 * block groups cache has started writing. If it already started,
9221 * back off and let this transaction commit
9223 mutex_lock(&root->fs_info->ro_block_group_mutex);
9224 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9225 u64 transid = trans->transid;
9227 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9228 btrfs_end_transaction(trans, root);
9230 ret = btrfs_wait_for_commit(root, transid);
9237 * if we are changing raid levels, try to allocate a corresponding
9238 * block group with the new raid level.
9240 alloc_flags = update_block_group_flags(root, cache->flags);
9241 if (alloc_flags != cache->flags) {
9242 ret = do_chunk_alloc(trans, root, alloc_flags,
9245 * ENOSPC is allowed here, we may have enough space
9246 * already allocated at the new raid level to
9255 ret = inc_block_group_ro(cache, 0);
9258 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9259 ret = do_chunk_alloc(trans, root, alloc_flags,
9263 ret = inc_block_group_ro(cache, 0);
9265 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9266 alloc_flags = update_block_group_flags(root, cache->flags);
9267 lock_chunks(root->fs_info->chunk_root);
9268 check_system_chunk(trans, root, alloc_flags);
9269 unlock_chunks(root->fs_info->chunk_root);
9271 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9273 btrfs_end_transaction(trans, root);
9277 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9278 struct btrfs_root *root, u64 type)
9280 u64 alloc_flags = get_alloc_profile(root, type);
9281 return do_chunk_alloc(trans, root, alloc_flags,
9286 * helper to account the unused space of all the readonly block group in the
9287 * space_info. takes mirrors into account.
9289 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9291 struct btrfs_block_group_cache *block_group;
9295 /* It's df, we don't care if it's racey */
9296 if (list_empty(&sinfo->ro_bgs))
9299 spin_lock(&sinfo->lock);
9300 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9301 spin_lock(&block_group->lock);
9303 if (!block_group->ro) {
9304 spin_unlock(&block_group->lock);
9308 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9309 BTRFS_BLOCK_GROUP_RAID10 |
9310 BTRFS_BLOCK_GROUP_DUP))
9315 free_bytes += (block_group->key.offset -
9316 btrfs_block_group_used(&block_group->item)) *
9319 spin_unlock(&block_group->lock);
9321 spin_unlock(&sinfo->lock);
9326 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9327 struct btrfs_block_group_cache *cache)
9329 struct btrfs_space_info *sinfo = cache->space_info;
9334 spin_lock(&sinfo->lock);
9335 spin_lock(&cache->lock);
9337 num_bytes = cache->key.offset - cache->reserved -
9338 cache->pinned - cache->bytes_super -
9339 btrfs_block_group_used(&cache->item);
9340 sinfo->bytes_readonly -= num_bytes;
9341 list_del_init(&cache->ro_list);
9343 spin_unlock(&cache->lock);
9344 spin_unlock(&sinfo->lock);
9348 * checks to see if its even possible to relocate this block group.
9350 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9351 * ok to go ahead and try.
9353 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9355 struct btrfs_block_group_cache *block_group;
9356 struct btrfs_space_info *space_info;
9357 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9358 struct btrfs_device *device;
9359 struct btrfs_trans_handle *trans;
9368 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9370 /* odd, couldn't find the block group, leave it alone */
9374 min_free = btrfs_block_group_used(&block_group->item);
9376 /* no bytes used, we're good */
9380 space_info = block_group->space_info;
9381 spin_lock(&space_info->lock);
9383 full = space_info->full;
9386 * if this is the last block group we have in this space, we can't
9387 * relocate it unless we're able to allocate a new chunk below.
9389 * Otherwise, we need to make sure we have room in the space to handle
9390 * all of the extents from this block group. If we can, we're good
9392 if ((space_info->total_bytes != block_group->key.offset) &&
9393 (space_info->bytes_used + space_info->bytes_reserved +
9394 space_info->bytes_pinned + space_info->bytes_readonly +
9395 min_free < space_info->total_bytes)) {
9396 spin_unlock(&space_info->lock);
9399 spin_unlock(&space_info->lock);
9402 * ok we don't have enough space, but maybe we have free space on our
9403 * devices to allocate new chunks for relocation, so loop through our
9404 * alloc devices and guess if we have enough space. if this block
9405 * group is going to be restriped, run checks against the target
9406 * profile instead of the current one.
9418 target = get_restripe_target(root->fs_info, block_group->flags);
9420 index = __get_raid_index(extended_to_chunk(target));
9423 * this is just a balance, so if we were marked as full
9424 * we know there is no space for a new chunk
9429 index = get_block_group_index(block_group);
9432 if (index == BTRFS_RAID_RAID10) {
9436 } else if (index == BTRFS_RAID_RAID1) {
9438 } else if (index == BTRFS_RAID_DUP) {
9441 } else if (index == BTRFS_RAID_RAID0) {
9442 dev_min = fs_devices->rw_devices;
9443 min_free = div64_u64(min_free, dev_min);
9446 /* We need to do this so that we can look at pending chunks */
9447 trans = btrfs_join_transaction(root);
9448 if (IS_ERR(trans)) {
9449 ret = PTR_ERR(trans);
9453 mutex_lock(&root->fs_info->chunk_mutex);
9454 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9458 * check to make sure we can actually find a chunk with enough
9459 * space to fit our block group in.
9461 if (device->total_bytes > device->bytes_used + min_free &&
9462 !device->is_tgtdev_for_dev_replace) {
9463 ret = find_free_dev_extent(trans, device, min_free,
9468 if (dev_nr >= dev_min)
9474 mutex_unlock(&root->fs_info->chunk_mutex);
9475 btrfs_end_transaction(trans, root);
9477 btrfs_put_block_group(block_group);
9481 static int find_first_block_group(struct btrfs_root *root,
9482 struct btrfs_path *path, struct btrfs_key *key)
9485 struct btrfs_key found_key;
9486 struct extent_buffer *leaf;
9489 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9494 slot = path->slots[0];
9495 leaf = path->nodes[0];
9496 if (slot >= btrfs_header_nritems(leaf)) {
9497 ret = btrfs_next_leaf(root, path);
9504 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9506 if (found_key.objectid >= key->objectid &&
9507 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9517 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9519 struct btrfs_block_group_cache *block_group;
9523 struct inode *inode;
9525 block_group = btrfs_lookup_first_block_group(info, last);
9526 while (block_group) {
9527 spin_lock(&block_group->lock);
9528 if (block_group->iref)
9530 spin_unlock(&block_group->lock);
9531 block_group = next_block_group(info->tree_root,
9541 inode = block_group->inode;
9542 block_group->iref = 0;
9543 block_group->inode = NULL;
9544 spin_unlock(&block_group->lock);
9546 last = block_group->key.objectid + block_group->key.offset;
9547 btrfs_put_block_group(block_group);
9551 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9553 struct btrfs_block_group_cache *block_group;
9554 struct btrfs_space_info *space_info;
9555 struct btrfs_caching_control *caching_ctl;
9558 down_write(&info->commit_root_sem);
9559 while (!list_empty(&info->caching_block_groups)) {
9560 caching_ctl = list_entry(info->caching_block_groups.next,
9561 struct btrfs_caching_control, list);
9562 list_del(&caching_ctl->list);
9563 put_caching_control(caching_ctl);
9565 up_write(&info->commit_root_sem);
9567 spin_lock(&info->unused_bgs_lock);
9568 while (!list_empty(&info->unused_bgs)) {
9569 block_group = list_first_entry(&info->unused_bgs,
9570 struct btrfs_block_group_cache,
9572 list_del_init(&block_group->bg_list);
9573 btrfs_put_block_group(block_group);
9575 spin_unlock(&info->unused_bgs_lock);
9577 spin_lock(&info->block_group_cache_lock);
9578 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9579 block_group = rb_entry(n, struct btrfs_block_group_cache,
9581 rb_erase(&block_group->cache_node,
9582 &info->block_group_cache_tree);
9583 RB_CLEAR_NODE(&block_group->cache_node);
9584 spin_unlock(&info->block_group_cache_lock);
9586 down_write(&block_group->space_info->groups_sem);
9587 list_del(&block_group->list);
9588 up_write(&block_group->space_info->groups_sem);
9590 if (block_group->cached == BTRFS_CACHE_STARTED)
9591 wait_block_group_cache_done(block_group);
9594 * We haven't cached this block group, which means we could
9595 * possibly have excluded extents on this block group.
9597 if (block_group->cached == BTRFS_CACHE_NO ||
9598 block_group->cached == BTRFS_CACHE_ERROR)
9599 free_excluded_extents(info->extent_root, block_group);
9601 btrfs_remove_free_space_cache(block_group);
9602 btrfs_put_block_group(block_group);
9604 spin_lock(&info->block_group_cache_lock);
9606 spin_unlock(&info->block_group_cache_lock);
9608 /* now that all the block groups are freed, go through and
9609 * free all the space_info structs. This is only called during
9610 * the final stages of unmount, and so we know nobody is
9611 * using them. We call synchronize_rcu() once before we start,
9612 * just to be on the safe side.
9616 release_global_block_rsv(info);
9618 while (!list_empty(&info->space_info)) {
9621 space_info = list_entry(info->space_info.next,
9622 struct btrfs_space_info,
9624 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9625 if (WARN_ON(space_info->bytes_pinned > 0 ||
9626 space_info->bytes_reserved > 0 ||
9627 space_info->bytes_may_use > 0)) {
9628 dump_space_info(space_info, 0, 0);
9631 list_del(&space_info->list);
9632 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9633 struct kobject *kobj;
9634 kobj = space_info->block_group_kobjs[i];
9635 space_info->block_group_kobjs[i] = NULL;
9641 kobject_del(&space_info->kobj);
9642 kobject_put(&space_info->kobj);
9647 static void __link_block_group(struct btrfs_space_info *space_info,
9648 struct btrfs_block_group_cache *cache)
9650 int index = get_block_group_index(cache);
9653 down_write(&space_info->groups_sem);
9654 if (list_empty(&space_info->block_groups[index]))
9656 list_add_tail(&cache->list, &space_info->block_groups[index]);
9657 up_write(&space_info->groups_sem);
9660 struct raid_kobject *rkobj;
9663 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9666 rkobj->raid_type = index;
9667 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9668 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9669 "%s", get_raid_name(index));
9671 kobject_put(&rkobj->kobj);
9674 space_info->block_group_kobjs[index] = &rkobj->kobj;
9679 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9682 static struct btrfs_block_group_cache *
9683 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9685 struct btrfs_block_group_cache *cache;
9687 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9691 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9693 if (!cache->free_space_ctl) {
9698 cache->key.objectid = start;
9699 cache->key.offset = size;
9700 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9702 cache->sectorsize = root->sectorsize;
9703 cache->fs_info = root->fs_info;
9704 cache->full_stripe_len = btrfs_full_stripe_len(root,
9705 &root->fs_info->mapping_tree,
9707 set_free_space_tree_thresholds(cache);
9709 atomic_set(&cache->count, 1);
9710 spin_lock_init(&cache->lock);
9711 init_rwsem(&cache->data_rwsem);
9712 INIT_LIST_HEAD(&cache->list);
9713 INIT_LIST_HEAD(&cache->cluster_list);
9714 INIT_LIST_HEAD(&cache->bg_list);
9715 INIT_LIST_HEAD(&cache->ro_list);
9716 INIT_LIST_HEAD(&cache->dirty_list);
9717 INIT_LIST_HEAD(&cache->io_list);
9718 btrfs_init_free_space_ctl(cache);
9719 atomic_set(&cache->trimming, 0);
9720 mutex_init(&cache->free_space_lock);
9725 int btrfs_read_block_groups(struct btrfs_root *root)
9727 struct btrfs_path *path;
9729 struct btrfs_block_group_cache *cache;
9730 struct btrfs_fs_info *info = root->fs_info;
9731 struct btrfs_space_info *space_info;
9732 struct btrfs_key key;
9733 struct btrfs_key found_key;
9734 struct extent_buffer *leaf;
9738 root = info->extent_root;
9741 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9742 path = btrfs_alloc_path();
9747 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9748 if (btrfs_test_opt(root, SPACE_CACHE) &&
9749 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9751 if (btrfs_test_opt(root, CLEAR_CACHE))
9755 ret = find_first_block_group(root, path, &key);
9761 leaf = path->nodes[0];
9762 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9764 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9773 * When we mount with old space cache, we need to
9774 * set BTRFS_DC_CLEAR and set dirty flag.
9776 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9777 * truncate the old free space cache inode and
9779 * b) Setting 'dirty flag' makes sure that we flush
9780 * the new space cache info onto disk.
9782 if (btrfs_test_opt(root, SPACE_CACHE))
9783 cache->disk_cache_state = BTRFS_DC_CLEAR;
9786 read_extent_buffer(leaf, &cache->item,
9787 btrfs_item_ptr_offset(leaf, path->slots[0]),
9788 sizeof(cache->item));
9789 cache->flags = btrfs_block_group_flags(&cache->item);
9791 key.objectid = found_key.objectid + found_key.offset;
9792 btrfs_release_path(path);
9795 * We need to exclude the super stripes now so that the space
9796 * info has super bytes accounted for, otherwise we'll think
9797 * we have more space than we actually do.
9799 ret = exclude_super_stripes(root, cache);
9802 * We may have excluded something, so call this just in
9805 free_excluded_extents(root, cache);
9806 btrfs_put_block_group(cache);
9811 * check for two cases, either we are full, and therefore
9812 * don't need to bother with the caching work since we won't
9813 * find any space, or we are empty, and we can just add all
9814 * the space in and be done with it. This saves us _alot_ of
9815 * time, particularly in the full case.
9817 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9818 cache->last_byte_to_unpin = (u64)-1;
9819 cache->cached = BTRFS_CACHE_FINISHED;
9820 free_excluded_extents(root, cache);
9821 } else if (btrfs_block_group_used(&cache->item) == 0) {
9822 cache->last_byte_to_unpin = (u64)-1;
9823 cache->cached = BTRFS_CACHE_FINISHED;
9824 add_new_free_space(cache, root->fs_info,
9826 found_key.objectid +
9828 free_excluded_extents(root, cache);
9831 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9833 btrfs_remove_free_space_cache(cache);
9834 btrfs_put_block_group(cache);
9838 ret = update_space_info(info, cache->flags, found_key.offset,
9839 btrfs_block_group_used(&cache->item),
9842 btrfs_remove_free_space_cache(cache);
9843 spin_lock(&info->block_group_cache_lock);
9844 rb_erase(&cache->cache_node,
9845 &info->block_group_cache_tree);
9846 RB_CLEAR_NODE(&cache->cache_node);
9847 spin_unlock(&info->block_group_cache_lock);
9848 btrfs_put_block_group(cache);
9852 cache->space_info = space_info;
9853 spin_lock(&cache->space_info->lock);
9854 cache->space_info->bytes_readonly += cache->bytes_super;
9855 spin_unlock(&cache->space_info->lock);
9857 __link_block_group(space_info, cache);
9859 set_avail_alloc_bits(root->fs_info, cache->flags);
9860 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9861 inc_block_group_ro(cache, 1);
9862 } else if (btrfs_block_group_used(&cache->item) == 0) {
9863 spin_lock(&info->unused_bgs_lock);
9864 /* Should always be true but just in case. */
9865 if (list_empty(&cache->bg_list)) {
9866 btrfs_get_block_group(cache);
9867 list_add_tail(&cache->bg_list,
9870 spin_unlock(&info->unused_bgs_lock);
9874 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9875 if (!(get_alloc_profile(root, space_info->flags) &
9876 (BTRFS_BLOCK_GROUP_RAID10 |
9877 BTRFS_BLOCK_GROUP_RAID1 |
9878 BTRFS_BLOCK_GROUP_RAID5 |
9879 BTRFS_BLOCK_GROUP_RAID6 |
9880 BTRFS_BLOCK_GROUP_DUP)))
9883 * avoid allocating from un-mirrored block group if there are
9884 * mirrored block groups.
9886 list_for_each_entry(cache,
9887 &space_info->block_groups[BTRFS_RAID_RAID0],
9889 inc_block_group_ro(cache, 1);
9890 list_for_each_entry(cache,
9891 &space_info->block_groups[BTRFS_RAID_SINGLE],
9893 inc_block_group_ro(cache, 1);
9896 init_global_block_rsv(info);
9899 btrfs_free_path(path);
9903 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9904 struct btrfs_root *root)
9906 struct btrfs_block_group_cache *block_group, *tmp;
9907 struct btrfs_root *extent_root = root->fs_info->extent_root;
9908 struct btrfs_block_group_item item;
9909 struct btrfs_key key;
9911 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9913 trans->can_flush_pending_bgs = false;
9914 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9918 spin_lock(&block_group->lock);
9919 memcpy(&item, &block_group->item, sizeof(item));
9920 memcpy(&key, &block_group->key, sizeof(key));
9921 spin_unlock(&block_group->lock);
9923 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9926 btrfs_abort_transaction(trans, extent_root, ret);
9927 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9928 key.objectid, key.offset);
9930 btrfs_abort_transaction(trans, extent_root, ret);
9931 add_block_group_free_space(trans, root->fs_info, block_group);
9932 /* already aborted the transaction if it failed. */
9934 list_del_init(&block_group->bg_list);
9936 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9939 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9940 struct btrfs_root *root, u64 bytes_used,
9941 u64 type, u64 chunk_objectid, u64 chunk_offset,
9945 struct btrfs_root *extent_root;
9946 struct btrfs_block_group_cache *cache;
9948 extent_root = root->fs_info->extent_root;
9950 btrfs_set_log_full_commit(root->fs_info, trans);
9952 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9956 btrfs_set_block_group_used(&cache->item, bytes_used);
9957 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9958 btrfs_set_block_group_flags(&cache->item, type);
9960 cache->flags = type;
9961 cache->last_byte_to_unpin = (u64)-1;
9962 cache->cached = BTRFS_CACHE_FINISHED;
9963 cache->needs_free_space = 1;
9964 ret = exclude_super_stripes(root, cache);
9967 * We may have excluded something, so call this just in
9970 free_excluded_extents(root, cache);
9971 btrfs_put_block_group(cache);
9975 add_new_free_space(cache, root->fs_info, chunk_offset,
9976 chunk_offset + size);
9978 free_excluded_extents(root, cache);
9980 #ifdef CONFIG_BTRFS_DEBUG
9981 if (btrfs_should_fragment_free_space(root, cache)) {
9982 u64 new_bytes_used = size - bytes_used;
9984 bytes_used += new_bytes_used >> 1;
9985 fragment_free_space(root, cache);
9989 * Call to ensure the corresponding space_info object is created and
9990 * assigned to our block group, but don't update its counters just yet.
9991 * We want our bg to be added to the rbtree with its ->space_info set.
9993 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9994 &cache->space_info);
9996 btrfs_remove_free_space_cache(cache);
9997 btrfs_put_block_group(cache);
10001 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10003 btrfs_remove_free_space_cache(cache);
10004 btrfs_put_block_group(cache);
10009 * Now that our block group has its ->space_info set and is inserted in
10010 * the rbtree, update the space info's counters.
10012 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10013 &cache->space_info);
10015 btrfs_remove_free_space_cache(cache);
10016 spin_lock(&root->fs_info->block_group_cache_lock);
10017 rb_erase(&cache->cache_node,
10018 &root->fs_info->block_group_cache_tree);
10019 RB_CLEAR_NODE(&cache->cache_node);
10020 spin_unlock(&root->fs_info->block_group_cache_lock);
10021 btrfs_put_block_group(cache);
10024 update_global_block_rsv(root->fs_info);
10026 spin_lock(&cache->space_info->lock);
10027 cache->space_info->bytes_readonly += cache->bytes_super;
10028 spin_unlock(&cache->space_info->lock);
10030 __link_block_group(cache->space_info, cache);
10032 list_add_tail(&cache->bg_list, &trans->new_bgs);
10034 set_avail_alloc_bits(extent_root->fs_info, type);
10039 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10041 u64 extra_flags = chunk_to_extended(flags) &
10042 BTRFS_EXTENDED_PROFILE_MASK;
10044 write_seqlock(&fs_info->profiles_lock);
10045 if (flags & BTRFS_BLOCK_GROUP_DATA)
10046 fs_info->avail_data_alloc_bits &= ~extra_flags;
10047 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10048 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10049 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10050 fs_info->avail_system_alloc_bits &= ~extra_flags;
10051 write_sequnlock(&fs_info->profiles_lock);
10054 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10055 struct btrfs_root *root, u64 group_start,
10056 struct extent_map *em)
10058 struct btrfs_path *path;
10059 struct btrfs_block_group_cache *block_group;
10060 struct btrfs_free_cluster *cluster;
10061 struct btrfs_root *tree_root = root->fs_info->tree_root;
10062 struct btrfs_key key;
10063 struct inode *inode;
10064 struct kobject *kobj = NULL;
10068 struct btrfs_caching_control *caching_ctl = NULL;
10071 root = root->fs_info->extent_root;
10073 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10074 BUG_ON(!block_group);
10075 BUG_ON(!block_group->ro);
10078 * Free the reserved super bytes from this block group before
10081 free_excluded_extents(root, block_group);
10083 memcpy(&key, &block_group->key, sizeof(key));
10084 index = get_block_group_index(block_group);
10085 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10086 BTRFS_BLOCK_GROUP_RAID1 |
10087 BTRFS_BLOCK_GROUP_RAID10))
10092 /* make sure this block group isn't part of an allocation cluster */
10093 cluster = &root->fs_info->data_alloc_cluster;
10094 spin_lock(&cluster->refill_lock);
10095 btrfs_return_cluster_to_free_space(block_group, cluster);
10096 spin_unlock(&cluster->refill_lock);
10099 * make sure this block group isn't part of a metadata
10100 * allocation cluster
10102 cluster = &root->fs_info->meta_alloc_cluster;
10103 spin_lock(&cluster->refill_lock);
10104 btrfs_return_cluster_to_free_space(block_group, cluster);
10105 spin_unlock(&cluster->refill_lock);
10107 path = btrfs_alloc_path();
10114 * get the inode first so any iput calls done for the io_list
10115 * aren't the final iput (no unlinks allowed now)
10117 inode = lookup_free_space_inode(tree_root, block_group, path);
10119 mutex_lock(&trans->transaction->cache_write_mutex);
10121 * make sure our free spache cache IO is done before remove the
10124 spin_lock(&trans->transaction->dirty_bgs_lock);
10125 if (!list_empty(&block_group->io_list)) {
10126 list_del_init(&block_group->io_list);
10128 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10130 spin_unlock(&trans->transaction->dirty_bgs_lock);
10131 btrfs_wait_cache_io(root, trans, block_group,
10132 &block_group->io_ctl, path,
10133 block_group->key.objectid);
10134 btrfs_put_block_group(block_group);
10135 spin_lock(&trans->transaction->dirty_bgs_lock);
10138 if (!list_empty(&block_group->dirty_list)) {
10139 list_del_init(&block_group->dirty_list);
10140 btrfs_put_block_group(block_group);
10142 spin_unlock(&trans->transaction->dirty_bgs_lock);
10143 mutex_unlock(&trans->transaction->cache_write_mutex);
10145 if (!IS_ERR(inode)) {
10146 ret = btrfs_orphan_add(trans, inode);
10148 btrfs_add_delayed_iput(inode);
10151 clear_nlink(inode);
10152 /* One for the block groups ref */
10153 spin_lock(&block_group->lock);
10154 if (block_group->iref) {
10155 block_group->iref = 0;
10156 block_group->inode = NULL;
10157 spin_unlock(&block_group->lock);
10160 spin_unlock(&block_group->lock);
10162 /* One for our lookup ref */
10163 btrfs_add_delayed_iput(inode);
10166 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10167 key.offset = block_group->key.objectid;
10170 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10174 btrfs_release_path(path);
10176 ret = btrfs_del_item(trans, tree_root, path);
10179 btrfs_release_path(path);
10182 spin_lock(&root->fs_info->block_group_cache_lock);
10183 rb_erase(&block_group->cache_node,
10184 &root->fs_info->block_group_cache_tree);
10185 RB_CLEAR_NODE(&block_group->cache_node);
10187 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10188 root->fs_info->first_logical_byte = (u64)-1;
10189 spin_unlock(&root->fs_info->block_group_cache_lock);
10191 down_write(&block_group->space_info->groups_sem);
10193 * we must use list_del_init so people can check to see if they
10194 * are still on the list after taking the semaphore
10196 list_del_init(&block_group->list);
10197 if (list_empty(&block_group->space_info->block_groups[index])) {
10198 kobj = block_group->space_info->block_group_kobjs[index];
10199 block_group->space_info->block_group_kobjs[index] = NULL;
10200 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10202 up_write(&block_group->space_info->groups_sem);
10208 if (block_group->has_caching_ctl)
10209 caching_ctl = get_caching_control(block_group);
10210 if (block_group->cached == BTRFS_CACHE_STARTED)
10211 wait_block_group_cache_done(block_group);
10212 if (block_group->has_caching_ctl) {
10213 down_write(&root->fs_info->commit_root_sem);
10214 if (!caching_ctl) {
10215 struct btrfs_caching_control *ctl;
10217 list_for_each_entry(ctl,
10218 &root->fs_info->caching_block_groups, list)
10219 if (ctl->block_group == block_group) {
10221 atomic_inc(&caching_ctl->count);
10226 list_del_init(&caching_ctl->list);
10227 up_write(&root->fs_info->commit_root_sem);
10229 /* Once for the caching bgs list and once for us. */
10230 put_caching_control(caching_ctl);
10231 put_caching_control(caching_ctl);
10235 spin_lock(&trans->transaction->dirty_bgs_lock);
10236 if (!list_empty(&block_group->dirty_list)) {
10239 if (!list_empty(&block_group->io_list)) {
10242 spin_unlock(&trans->transaction->dirty_bgs_lock);
10243 btrfs_remove_free_space_cache(block_group);
10245 spin_lock(&block_group->space_info->lock);
10246 list_del_init(&block_group->ro_list);
10248 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10249 WARN_ON(block_group->space_info->total_bytes
10250 < block_group->key.offset);
10251 WARN_ON(block_group->space_info->bytes_readonly
10252 < block_group->key.offset);
10253 WARN_ON(block_group->space_info->disk_total
10254 < block_group->key.offset * factor);
10256 block_group->space_info->total_bytes -= block_group->key.offset;
10257 block_group->space_info->bytes_readonly -= block_group->key.offset;
10258 block_group->space_info->disk_total -= block_group->key.offset * factor;
10260 spin_unlock(&block_group->space_info->lock);
10262 memcpy(&key, &block_group->key, sizeof(key));
10265 if (!list_empty(&em->list)) {
10266 /* We're in the transaction->pending_chunks list. */
10267 free_extent_map(em);
10269 spin_lock(&block_group->lock);
10270 block_group->removed = 1;
10272 * At this point trimming can't start on this block group, because we
10273 * removed the block group from the tree fs_info->block_group_cache_tree
10274 * so no one can't find it anymore and even if someone already got this
10275 * block group before we removed it from the rbtree, they have already
10276 * incremented block_group->trimming - if they didn't, they won't find
10277 * any free space entries because we already removed them all when we
10278 * called btrfs_remove_free_space_cache().
10280 * And we must not remove the extent map from the fs_info->mapping_tree
10281 * to prevent the same logical address range and physical device space
10282 * ranges from being reused for a new block group. This is because our
10283 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10284 * completely transactionless, so while it is trimming a range the
10285 * currently running transaction might finish and a new one start,
10286 * allowing for new block groups to be created that can reuse the same
10287 * physical device locations unless we take this special care.
10289 * There may also be an implicit trim operation if the file system
10290 * is mounted with -odiscard. The same protections must remain
10291 * in place until the extents have been discarded completely when
10292 * the transaction commit has completed.
10294 remove_em = (atomic_read(&block_group->trimming) == 0);
10296 * Make sure a trimmer task always sees the em in the pinned_chunks list
10297 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10298 * before checking block_group->removed).
10302 * Our em might be in trans->transaction->pending_chunks which
10303 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10304 * and so is the fs_info->pinned_chunks list.
10306 * So at this point we must be holding the chunk_mutex to avoid
10307 * any races with chunk allocation (more specifically at
10308 * volumes.c:contains_pending_extent()), to ensure it always
10309 * sees the em, either in the pending_chunks list or in the
10310 * pinned_chunks list.
10312 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10314 spin_unlock(&block_group->lock);
10317 struct extent_map_tree *em_tree;
10319 em_tree = &root->fs_info->mapping_tree.map_tree;
10320 write_lock(&em_tree->lock);
10322 * The em might be in the pending_chunks list, so make sure the
10323 * chunk mutex is locked, since remove_extent_mapping() will
10324 * delete us from that list.
10326 remove_extent_mapping(em_tree, em);
10327 write_unlock(&em_tree->lock);
10328 /* once for the tree */
10329 free_extent_map(em);
10332 unlock_chunks(root);
10334 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10338 btrfs_put_block_group(block_group);
10339 btrfs_put_block_group(block_group);
10341 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10347 ret = btrfs_del_item(trans, root, path);
10349 btrfs_free_path(path);
10353 struct btrfs_trans_handle *
10354 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10355 const u64 chunk_offset)
10357 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10358 struct extent_map *em;
10359 struct map_lookup *map;
10360 unsigned int num_items;
10362 read_lock(&em_tree->lock);
10363 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10364 read_unlock(&em_tree->lock);
10365 ASSERT(em && em->start == chunk_offset);
10368 * We need to reserve 3 + N units from the metadata space info in order
10369 * to remove a block group (done at btrfs_remove_chunk() and at
10370 * btrfs_remove_block_group()), which are used for:
10372 * 1 unit for adding the free space inode's orphan (located in the tree
10374 * 1 unit for deleting the block group item (located in the extent
10376 * 1 unit for deleting the free space item (located in tree of tree
10378 * N units for deleting N device extent items corresponding to each
10379 * stripe (located in the device tree).
10381 * In order to remove a block group we also need to reserve units in the
10382 * system space info in order to update the chunk tree (update one or
10383 * more device items and remove one chunk item), but this is done at
10384 * btrfs_remove_chunk() through a call to check_system_chunk().
10386 map = (struct map_lookup *)em->bdev;
10387 num_items = 3 + map->num_stripes;
10388 free_extent_map(em);
10390 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10395 * Process the unused_bgs list and remove any that don't have any allocated
10396 * space inside of them.
10398 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10400 struct btrfs_block_group_cache *block_group;
10401 struct btrfs_space_info *space_info;
10402 struct btrfs_root *root = fs_info->extent_root;
10403 struct btrfs_trans_handle *trans;
10406 if (!fs_info->open)
10409 spin_lock(&fs_info->unused_bgs_lock);
10410 while (!list_empty(&fs_info->unused_bgs)) {
10414 block_group = list_first_entry(&fs_info->unused_bgs,
10415 struct btrfs_block_group_cache,
10417 list_del_init(&block_group->bg_list);
10419 space_info = block_group->space_info;
10421 if (ret || btrfs_mixed_space_info(space_info)) {
10422 btrfs_put_block_group(block_group);
10425 spin_unlock(&fs_info->unused_bgs_lock);
10427 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10429 /* Don't want to race with allocators so take the groups_sem */
10430 down_write(&space_info->groups_sem);
10431 spin_lock(&block_group->lock);
10432 if (block_group->reserved ||
10433 btrfs_block_group_used(&block_group->item) ||
10435 list_is_singular(&block_group->list)) {
10437 * We want to bail if we made new allocations or have
10438 * outstanding allocations in this block group. We do
10439 * the ro check in case balance is currently acting on
10440 * this block group.
10442 spin_unlock(&block_group->lock);
10443 up_write(&space_info->groups_sem);
10446 spin_unlock(&block_group->lock);
10448 /* We don't want to force the issue, only flip if it's ok. */
10449 ret = inc_block_group_ro(block_group, 0);
10450 up_write(&space_info->groups_sem);
10457 * Want to do this before we do anything else so we can recover
10458 * properly if we fail to join the transaction.
10460 trans = btrfs_start_trans_remove_block_group(fs_info,
10461 block_group->key.objectid);
10462 if (IS_ERR(trans)) {
10463 btrfs_dec_block_group_ro(root, block_group);
10464 ret = PTR_ERR(trans);
10469 * We could have pending pinned extents for this block group,
10470 * just delete them, we don't care about them anymore.
10472 start = block_group->key.objectid;
10473 end = start + block_group->key.offset - 1;
10475 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10476 * btrfs_finish_extent_commit(). If we are at transaction N,
10477 * another task might be running finish_extent_commit() for the
10478 * previous transaction N - 1, and have seen a range belonging
10479 * to the block group in freed_extents[] before we were able to
10480 * clear the whole block group range from freed_extents[]. This
10481 * means that task can lookup for the block group after we
10482 * unpinned it from freed_extents[] and removed it, leading to
10483 * a BUG_ON() at btrfs_unpin_extent_range().
10485 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10486 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10487 EXTENT_DIRTY, GFP_NOFS);
10489 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10490 btrfs_dec_block_group_ro(root, block_group);
10493 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10494 EXTENT_DIRTY, GFP_NOFS);
10496 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10497 btrfs_dec_block_group_ro(root, block_group);
10500 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10502 /* Reset pinned so btrfs_put_block_group doesn't complain */
10503 spin_lock(&space_info->lock);
10504 spin_lock(&block_group->lock);
10506 space_info->bytes_pinned -= block_group->pinned;
10507 space_info->bytes_readonly += block_group->pinned;
10508 percpu_counter_add(&space_info->total_bytes_pinned,
10509 -block_group->pinned);
10510 block_group->pinned = 0;
10512 spin_unlock(&block_group->lock);
10513 spin_unlock(&space_info->lock);
10515 /* DISCARD can flip during remount */
10516 trimming = btrfs_test_opt(root, DISCARD);
10518 /* Implicit trim during transaction commit. */
10520 btrfs_get_block_group_trimming(block_group);
10523 * Btrfs_remove_chunk will abort the transaction if things go
10526 ret = btrfs_remove_chunk(trans, root,
10527 block_group->key.objectid);
10531 btrfs_put_block_group_trimming(block_group);
10536 * If we're not mounted with -odiscard, we can just forget
10537 * about this block group. Otherwise we'll need to wait
10538 * until transaction commit to do the actual discard.
10541 spin_lock(&fs_info->unused_bgs_lock);
10543 * A concurrent scrub might have added us to the list
10544 * fs_info->unused_bgs, so use a list_move operation
10545 * to add the block group to the deleted_bgs list.
10547 list_move(&block_group->bg_list,
10548 &trans->transaction->deleted_bgs);
10549 spin_unlock(&fs_info->unused_bgs_lock);
10550 btrfs_get_block_group(block_group);
10553 btrfs_end_transaction(trans, root);
10555 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10556 btrfs_put_block_group(block_group);
10557 spin_lock(&fs_info->unused_bgs_lock);
10559 spin_unlock(&fs_info->unused_bgs_lock);
10562 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10564 struct btrfs_space_info *space_info;
10565 struct btrfs_super_block *disk_super;
10571 disk_super = fs_info->super_copy;
10572 if (!btrfs_super_root(disk_super))
10575 features = btrfs_super_incompat_flags(disk_super);
10576 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10579 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10580 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10585 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10586 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10588 flags = BTRFS_BLOCK_GROUP_METADATA;
10589 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10593 flags = BTRFS_BLOCK_GROUP_DATA;
10594 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10600 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10602 return unpin_extent_range(root, start, end, false);
10606 * It used to be that old block groups would be left around forever.
10607 * Iterating over them would be enough to trim unused space. Since we
10608 * now automatically remove them, we also need to iterate over unallocated
10611 * We don't want a transaction for this since the discard may take a
10612 * substantial amount of time. We don't require that a transaction be
10613 * running, but we do need to take a running transaction into account
10614 * to ensure that we're not discarding chunks that were released in
10615 * the current transaction.
10617 * Holding the chunks lock will prevent other threads from allocating
10618 * or releasing chunks, but it won't prevent a running transaction
10619 * from committing and releasing the memory that the pending chunks
10620 * list head uses. For that, we need to take a reference to the
10623 static int btrfs_trim_free_extents(struct btrfs_device *device,
10624 u64 minlen, u64 *trimmed)
10626 u64 start = 0, len = 0;
10631 /* Not writeable = nothing to do. */
10632 if (!device->writeable)
10635 /* No free space = nothing to do. */
10636 if (device->total_bytes <= device->bytes_used)
10642 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10643 struct btrfs_transaction *trans;
10646 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10650 down_read(&fs_info->commit_root_sem);
10652 spin_lock(&fs_info->trans_lock);
10653 trans = fs_info->running_transaction;
10655 atomic_inc(&trans->use_count);
10656 spin_unlock(&fs_info->trans_lock);
10658 ret = find_free_dev_extent_start(trans, device, minlen, start,
10661 btrfs_put_transaction(trans);
10664 up_read(&fs_info->commit_root_sem);
10665 mutex_unlock(&fs_info->chunk_mutex);
10666 if (ret == -ENOSPC)
10671 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10672 up_read(&fs_info->commit_root_sem);
10673 mutex_unlock(&fs_info->chunk_mutex);
10681 if (fatal_signal_pending(current)) {
10682 ret = -ERESTARTSYS;
10692 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10694 struct btrfs_fs_info *fs_info = root->fs_info;
10695 struct btrfs_block_group_cache *cache = NULL;
10696 struct btrfs_device *device;
10697 struct list_head *devices;
10702 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10706 * try to trim all FS space, our block group may start from non-zero.
10708 if (range->len == total_bytes)
10709 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10711 cache = btrfs_lookup_block_group(fs_info, range->start);
10714 if (cache->key.objectid >= (range->start + range->len)) {
10715 btrfs_put_block_group(cache);
10719 start = max(range->start, cache->key.objectid);
10720 end = min(range->start + range->len,
10721 cache->key.objectid + cache->key.offset);
10723 if (end - start >= range->minlen) {
10724 if (!block_group_cache_done(cache)) {
10725 ret = cache_block_group(cache, 0);
10727 btrfs_put_block_group(cache);
10730 ret = wait_block_group_cache_done(cache);
10732 btrfs_put_block_group(cache);
10736 ret = btrfs_trim_block_group(cache,
10742 trimmed += group_trimmed;
10744 btrfs_put_block_group(cache);
10749 cache = next_block_group(fs_info->tree_root, cache);
10752 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10753 devices = &root->fs_info->fs_devices->alloc_list;
10754 list_for_each_entry(device, devices, dev_alloc_list) {
10755 ret = btrfs_trim_free_extents(device, range->minlen,
10760 trimmed += group_trimmed;
10762 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10764 range->len = trimmed;
10769 * btrfs_{start,end}_write_no_snapshoting() are similar to
10770 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10771 * data into the page cache through nocow before the subvolume is snapshoted,
10772 * but flush the data into disk after the snapshot creation, or to prevent
10773 * operations while snapshoting is ongoing and that cause the snapshot to be
10774 * inconsistent (writes followed by expanding truncates for example).
10776 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10778 percpu_counter_dec(&root->subv_writers->counter);
10780 * Make sure counter is updated before we wake up waiters.
10783 if (waitqueue_active(&root->subv_writers->wait))
10784 wake_up(&root->subv_writers->wait);
10787 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10789 if (atomic_read(&root->will_be_snapshoted))
10792 percpu_counter_inc(&root->subv_writers->counter);
10794 * Make sure counter is updated before we check for snapshot creation.
10797 if (atomic_read(&root->will_be_snapshoted)) {
10798 btrfs_end_write_no_snapshoting(root);