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);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE);
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);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE);
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;
440 path->reada = READA_FORWARD;
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 COWed 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;
2046 * Avoid races with device replace and make sure our bbio has devices
2047 * associated to its stripes that don't go away while we are discarding.
2049 btrfs_bio_counter_inc_blocked(root->fs_info);
2050 /* Tell the block device(s) that the sectors can be discarded */
2051 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2052 bytenr, &num_bytes, &bbio, 0);
2053 /* Error condition is -ENOMEM */
2055 struct btrfs_bio_stripe *stripe = bbio->stripes;
2059 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2061 if (!stripe->dev->can_discard)
2064 ret = btrfs_issue_discard(stripe->dev->bdev,
2069 discarded_bytes += bytes;
2070 else if (ret != -EOPNOTSUPP)
2071 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2074 * Just in case we get back EOPNOTSUPP for some reason,
2075 * just ignore the return value so we don't screw up
2076 * people calling discard_extent.
2080 btrfs_put_bbio(bbio);
2082 btrfs_bio_counter_dec(root->fs_info);
2085 *actual_bytes = discarded_bytes;
2088 if (ret == -EOPNOTSUPP)
2093 /* Can return -ENOMEM */
2094 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2095 struct btrfs_root *root,
2096 u64 bytenr, u64 num_bytes, u64 parent,
2097 u64 root_objectid, u64 owner, u64 offset)
2100 struct btrfs_fs_info *fs_info = root->fs_info;
2102 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2103 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2105 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2106 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2108 parent, root_objectid, (int)owner,
2109 BTRFS_ADD_DELAYED_REF, NULL);
2111 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2112 num_bytes, parent, root_objectid,
2114 BTRFS_ADD_DELAYED_REF, NULL);
2119 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2120 struct btrfs_root *root,
2121 struct btrfs_delayed_ref_node *node,
2122 u64 parent, u64 root_objectid,
2123 u64 owner, u64 offset, int refs_to_add,
2124 struct btrfs_delayed_extent_op *extent_op)
2126 struct btrfs_fs_info *fs_info = root->fs_info;
2127 struct btrfs_path *path;
2128 struct extent_buffer *leaf;
2129 struct btrfs_extent_item *item;
2130 struct btrfs_key key;
2131 u64 bytenr = node->bytenr;
2132 u64 num_bytes = node->num_bytes;
2136 path = btrfs_alloc_path();
2140 path->reada = READA_FORWARD;
2141 path->leave_spinning = 1;
2142 /* this will setup the path even if it fails to insert the back ref */
2143 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2144 bytenr, num_bytes, parent,
2145 root_objectid, owner, offset,
2146 refs_to_add, extent_op);
2147 if ((ret < 0 && ret != -EAGAIN) || !ret)
2151 * Ok we had -EAGAIN which means we didn't have space to insert and
2152 * inline extent ref, so just update the reference count and add a
2155 leaf = path->nodes[0];
2156 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2157 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2158 refs = btrfs_extent_refs(leaf, item);
2159 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2161 __run_delayed_extent_op(extent_op, leaf, item);
2163 btrfs_mark_buffer_dirty(leaf);
2164 btrfs_release_path(path);
2166 path->reada = READA_FORWARD;
2167 path->leave_spinning = 1;
2168 /* now insert the actual backref */
2169 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2170 path, bytenr, parent, root_objectid,
2171 owner, offset, refs_to_add);
2173 btrfs_abort_transaction(trans, root, ret);
2175 btrfs_free_path(path);
2179 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2180 struct btrfs_root *root,
2181 struct btrfs_delayed_ref_node *node,
2182 struct btrfs_delayed_extent_op *extent_op,
2183 int insert_reserved)
2186 struct btrfs_delayed_data_ref *ref;
2187 struct btrfs_key ins;
2192 ins.objectid = node->bytenr;
2193 ins.offset = node->num_bytes;
2194 ins.type = BTRFS_EXTENT_ITEM_KEY;
2196 ref = btrfs_delayed_node_to_data_ref(node);
2197 trace_run_delayed_data_ref(node, ref, node->action);
2199 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2200 parent = ref->parent;
2201 ref_root = ref->root;
2203 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2205 flags |= extent_op->flags_to_set;
2206 ret = alloc_reserved_file_extent(trans, root,
2207 parent, ref_root, flags,
2208 ref->objectid, ref->offset,
2209 &ins, node->ref_mod);
2210 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2211 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2212 ref_root, ref->objectid,
2213 ref->offset, node->ref_mod,
2215 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2216 ret = __btrfs_free_extent(trans, root, node, parent,
2217 ref_root, ref->objectid,
2218 ref->offset, node->ref_mod,
2226 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2227 struct extent_buffer *leaf,
2228 struct btrfs_extent_item *ei)
2230 u64 flags = btrfs_extent_flags(leaf, ei);
2231 if (extent_op->update_flags) {
2232 flags |= extent_op->flags_to_set;
2233 btrfs_set_extent_flags(leaf, ei, flags);
2236 if (extent_op->update_key) {
2237 struct btrfs_tree_block_info *bi;
2238 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2239 bi = (struct btrfs_tree_block_info *)(ei + 1);
2240 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2244 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2245 struct btrfs_root *root,
2246 struct btrfs_delayed_ref_node *node,
2247 struct btrfs_delayed_extent_op *extent_op)
2249 struct btrfs_key key;
2250 struct btrfs_path *path;
2251 struct btrfs_extent_item *ei;
2252 struct extent_buffer *leaf;
2256 int metadata = !extent_op->is_data;
2261 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2264 path = btrfs_alloc_path();
2268 key.objectid = node->bytenr;
2271 key.type = BTRFS_METADATA_ITEM_KEY;
2272 key.offset = extent_op->level;
2274 key.type = BTRFS_EXTENT_ITEM_KEY;
2275 key.offset = node->num_bytes;
2279 path->reada = READA_FORWARD;
2280 path->leave_spinning = 1;
2281 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2289 if (path->slots[0] > 0) {
2291 btrfs_item_key_to_cpu(path->nodes[0], &key,
2293 if (key.objectid == node->bytenr &&
2294 key.type == BTRFS_EXTENT_ITEM_KEY &&
2295 key.offset == node->num_bytes)
2299 btrfs_release_path(path);
2302 key.objectid = node->bytenr;
2303 key.offset = node->num_bytes;
2304 key.type = BTRFS_EXTENT_ITEM_KEY;
2313 leaf = path->nodes[0];
2314 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2316 if (item_size < sizeof(*ei)) {
2317 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2323 leaf = path->nodes[0];
2324 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2327 BUG_ON(item_size < sizeof(*ei));
2328 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2329 __run_delayed_extent_op(extent_op, leaf, ei);
2331 btrfs_mark_buffer_dirty(leaf);
2333 btrfs_free_path(path);
2337 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2338 struct btrfs_root *root,
2339 struct btrfs_delayed_ref_node *node,
2340 struct btrfs_delayed_extent_op *extent_op,
2341 int insert_reserved)
2344 struct btrfs_delayed_tree_ref *ref;
2345 struct btrfs_key ins;
2348 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2351 ref = btrfs_delayed_node_to_tree_ref(node);
2352 trace_run_delayed_tree_ref(node, ref, node->action);
2354 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2355 parent = ref->parent;
2356 ref_root = ref->root;
2358 ins.objectid = node->bytenr;
2359 if (skinny_metadata) {
2360 ins.offset = ref->level;
2361 ins.type = BTRFS_METADATA_ITEM_KEY;
2363 ins.offset = node->num_bytes;
2364 ins.type = BTRFS_EXTENT_ITEM_KEY;
2367 BUG_ON(node->ref_mod != 1);
2368 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2369 BUG_ON(!extent_op || !extent_op->update_flags);
2370 ret = alloc_reserved_tree_block(trans, root,
2372 extent_op->flags_to_set,
2375 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2376 ret = __btrfs_inc_extent_ref(trans, root, node,
2380 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2381 ret = __btrfs_free_extent(trans, root, node,
2383 ref->level, 0, 1, extent_op);
2390 /* helper function to actually process a single delayed ref entry */
2391 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2392 struct btrfs_root *root,
2393 struct btrfs_delayed_ref_node *node,
2394 struct btrfs_delayed_extent_op *extent_op,
2395 int insert_reserved)
2399 if (trans->aborted) {
2400 if (insert_reserved)
2401 btrfs_pin_extent(root, node->bytenr,
2402 node->num_bytes, 1);
2406 if (btrfs_delayed_ref_is_head(node)) {
2407 struct btrfs_delayed_ref_head *head;
2409 * we've hit the end of the chain and we were supposed
2410 * to insert this extent into the tree. But, it got
2411 * deleted before we ever needed to insert it, so all
2412 * we have to do is clean up the accounting
2415 head = btrfs_delayed_node_to_head(node);
2416 trace_run_delayed_ref_head(node, head, node->action);
2418 if (insert_reserved) {
2419 btrfs_pin_extent(root, node->bytenr,
2420 node->num_bytes, 1);
2421 if (head->is_data) {
2422 ret = btrfs_del_csums(trans, root,
2428 /* Also free its reserved qgroup space */
2429 btrfs_qgroup_free_delayed_ref(root->fs_info,
2430 head->qgroup_ref_root,
2431 head->qgroup_reserved);
2435 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2436 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2437 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2439 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2440 node->type == BTRFS_SHARED_DATA_REF_KEY)
2441 ret = run_delayed_data_ref(trans, root, node, extent_op,
2448 static inline struct btrfs_delayed_ref_node *
2449 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2451 struct btrfs_delayed_ref_node *ref;
2453 if (list_empty(&head->ref_list))
2457 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2458 * This is to prevent a ref count from going down to zero, which deletes
2459 * the extent item from the extent tree, when there still are references
2460 * to add, which would fail because they would not find the extent item.
2462 list_for_each_entry(ref, &head->ref_list, list) {
2463 if (ref->action == BTRFS_ADD_DELAYED_REF)
2467 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2472 * Returns 0 on success or if called with an already aborted transaction.
2473 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2475 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2476 struct btrfs_root *root,
2479 struct btrfs_delayed_ref_root *delayed_refs;
2480 struct btrfs_delayed_ref_node *ref;
2481 struct btrfs_delayed_ref_head *locked_ref = NULL;
2482 struct btrfs_delayed_extent_op *extent_op;
2483 struct btrfs_fs_info *fs_info = root->fs_info;
2484 ktime_t start = ktime_get();
2486 unsigned long count = 0;
2487 unsigned long actual_count = 0;
2488 int must_insert_reserved = 0;
2490 delayed_refs = &trans->transaction->delayed_refs;
2496 spin_lock(&delayed_refs->lock);
2497 locked_ref = btrfs_select_ref_head(trans);
2499 spin_unlock(&delayed_refs->lock);
2503 /* grab the lock that says we are going to process
2504 * all the refs for this head */
2505 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2506 spin_unlock(&delayed_refs->lock);
2508 * we may have dropped the spin lock to get the head
2509 * mutex lock, and that might have given someone else
2510 * time to free the head. If that's true, it has been
2511 * removed from our list and we can move on.
2513 if (ret == -EAGAIN) {
2521 * We need to try and merge add/drops of the same ref since we
2522 * can run into issues with relocate dropping the implicit ref
2523 * and then it being added back again before the drop can
2524 * finish. If we merged anything we need to re-loop so we can
2526 * Or we can get node references of the same type that weren't
2527 * merged when created due to bumps in the tree mod seq, and
2528 * we need to merge them to prevent adding an inline extent
2529 * backref before dropping it (triggering a BUG_ON at
2530 * insert_inline_extent_backref()).
2532 spin_lock(&locked_ref->lock);
2533 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2537 * locked_ref is the head node, so we have to go one
2538 * node back for any delayed ref updates
2540 ref = select_delayed_ref(locked_ref);
2542 if (ref && ref->seq &&
2543 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2544 spin_unlock(&locked_ref->lock);
2545 btrfs_delayed_ref_unlock(locked_ref);
2546 spin_lock(&delayed_refs->lock);
2547 locked_ref->processing = 0;
2548 delayed_refs->num_heads_ready++;
2549 spin_unlock(&delayed_refs->lock);
2557 * record the must insert reserved flag before we
2558 * drop the spin lock.
2560 must_insert_reserved = locked_ref->must_insert_reserved;
2561 locked_ref->must_insert_reserved = 0;
2563 extent_op = locked_ref->extent_op;
2564 locked_ref->extent_op = NULL;
2569 /* All delayed refs have been processed, Go ahead
2570 * and send the head node to run_one_delayed_ref,
2571 * so that any accounting fixes can happen
2573 ref = &locked_ref->node;
2575 if (extent_op && must_insert_reserved) {
2576 btrfs_free_delayed_extent_op(extent_op);
2581 spin_unlock(&locked_ref->lock);
2582 ret = run_delayed_extent_op(trans, root,
2584 btrfs_free_delayed_extent_op(extent_op);
2588 * Need to reset must_insert_reserved if
2589 * there was an error so the abort stuff
2590 * can cleanup the reserved space
2593 if (must_insert_reserved)
2594 locked_ref->must_insert_reserved = 1;
2595 locked_ref->processing = 0;
2596 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2597 btrfs_delayed_ref_unlock(locked_ref);
2604 * Need to drop our head ref lock and re-acquire the
2605 * delayed ref lock and then re-check to make sure
2608 spin_unlock(&locked_ref->lock);
2609 spin_lock(&delayed_refs->lock);
2610 spin_lock(&locked_ref->lock);
2611 if (!list_empty(&locked_ref->ref_list) ||
2612 locked_ref->extent_op) {
2613 spin_unlock(&locked_ref->lock);
2614 spin_unlock(&delayed_refs->lock);
2618 delayed_refs->num_heads--;
2619 rb_erase(&locked_ref->href_node,
2620 &delayed_refs->href_root);
2621 spin_unlock(&delayed_refs->lock);
2625 list_del(&ref->list);
2627 atomic_dec(&delayed_refs->num_entries);
2629 if (!btrfs_delayed_ref_is_head(ref)) {
2631 * when we play the delayed ref, also correct the
2634 switch (ref->action) {
2635 case BTRFS_ADD_DELAYED_REF:
2636 case BTRFS_ADD_DELAYED_EXTENT:
2637 locked_ref->node.ref_mod -= ref->ref_mod;
2639 case BTRFS_DROP_DELAYED_REF:
2640 locked_ref->node.ref_mod += ref->ref_mod;
2646 spin_unlock(&locked_ref->lock);
2648 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2649 must_insert_reserved);
2651 btrfs_free_delayed_extent_op(extent_op);
2653 locked_ref->processing = 0;
2654 btrfs_delayed_ref_unlock(locked_ref);
2655 btrfs_put_delayed_ref(ref);
2656 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2661 * If this node is a head, that means all the refs in this head
2662 * have been dealt with, and we will pick the next head to deal
2663 * with, so we must unlock the head and drop it from the cluster
2664 * list before we release it.
2666 if (btrfs_delayed_ref_is_head(ref)) {
2667 if (locked_ref->is_data &&
2668 locked_ref->total_ref_mod < 0) {
2669 spin_lock(&delayed_refs->lock);
2670 delayed_refs->pending_csums -= ref->num_bytes;
2671 spin_unlock(&delayed_refs->lock);
2673 btrfs_delayed_ref_unlock(locked_ref);
2676 btrfs_put_delayed_ref(ref);
2682 * We don't want to include ref heads since we can have empty ref heads
2683 * and those will drastically skew our runtime down since we just do
2684 * accounting, no actual extent tree updates.
2686 if (actual_count > 0) {
2687 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2691 * We weigh the current average higher than our current runtime
2692 * to avoid large swings in the average.
2694 spin_lock(&delayed_refs->lock);
2695 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2696 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2697 spin_unlock(&delayed_refs->lock);
2702 #ifdef SCRAMBLE_DELAYED_REFS
2704 * Normally delayed refs get processed in ascending bytenr order. This
2705 * correlates in most cases to the order added. To expose dependencies on this
2706 * order, we start to process the tree in the middle instead of the beginning
2708 static u64 find_middle(struct rb_root *root)
2710 struct rb_node *n = root->rb_node;
2711 struct btrfs_delayed_ref_node *entry;
2714 u64 first = 0, last = 0;
2718 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2719 first = entry->bytenr;
2723 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2724 last = entry->bytenr;
2729 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2730 WARN_ON(!entry->in_tree);
2732 middle = entry->bytenr;
2745 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2749 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2750 sizeof(struct btrfs_extent_inline_ref));
2751 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2752 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2755 * We don't ever fill up leaves all the way so multiply by 2 just to be
2756 * closer to what we're really going to want to use.
2758 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2762 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2763 * would require to store the csums for that many bytes.
2765 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2768 u64 num_csums_per_leaf;
2771 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2772 num_csums_per_leaf = div64_u64(csum_size,
2773 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2774 num_csums = div64_u64(csum_bytes, root->sectorsize);
2775 num_csums += num_csums_per_leaf - 1;
2776 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2780 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2781 struct btrfs_root *root)
2783 struct btrfs_block_rsv *global_rsv;
2784 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2785 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2786 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2787 u64 num_bytes, num_dirty_bgs_bytes;
2790 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2791 num_heads = heads_to_leaves(root, num_heads);
2793 num_bytes += (num_heads - 1) * root->nodesize;
2795 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2796 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2798 global_rsv = &root->fs_info->global_block_rsv;
2801 * If we can't allocate any more chunks lets make sure we have _lots_ of
2802 * wiggle room since running delayed refs can create more delayed refs.
2804 if (global_rsv->space_info->full) {
2805 num_dirty_bgs_bytes <<= 1;
2809 spin_lock(&global_rsv->lock);
2810 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2812 spin_unlock(&global_rsv->lock);
2816 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2817 struct btrfs_root *root)
2819 struct btrfs_fs_info *fs_info = root->fs_info;
2821 atomic_read(&trans->transaction->delayed_refs.num_entries);
2826 avg_runtime = fs_info->avg_delayed_ref_runtime;
2827 val = num_entries * avg_runtime;
2828 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2830 if (val >= NSEC_PER_SEC / 2)
2833 return btrfs_check_space_for_delayed_refs(trans, root);
2836 struct async_delayed_refs {
2837 struct btrfs_root *root;
2842 struct completion wait;
2843 struct btrfs_work work;
2846 static void delayed_ref_async_start(struct btrfs_work *work)
2848 struct async_delayed_refs *async;
2849 struct btrfs_trans_handle *trans;
2852 async = container_of(work, struct async_delayed_refs, work);
2854 /* if the commit is already started, we don't need to wait here */
2855 if (btrfs_transaction_blocked(async->root->fs_info))
2858 trans = btrfs_join_transaction(async->root);
2859 if (IS_ERR(trans)) {
2860 async->error = PTR_ERR(trans);
2865 * trans->sync means that when we call end_transaction, we won't
2866 * wait on delayed refs
2870 /* Don't bother flushing if we got into a different transaction */
2871 if (trans->transid > async->transid)
2874 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2878 ret = btrfs_end_transaction(trans, async->root);
2879 if (ret && !async->error)
2883 complete(&async->wait);
2888 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2889 unsigned long count, u64 transid, int wait)
2891 struct async_delayed_refs *async;
2894 async = kmalloc(sizeof(*async), GFP_NOFS);
2898 async->root = root->fs_info->tree_root;
2899 async->count = count;
2901 async->transid = transid;
2906 init_completion(&async->wait);
2908 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2909 delayed_ref_async_start, NULL, NULL);
2911 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2914 wait_for_completion(&async->wait);
2923 * this starts processing the delayed reference count updates and
2924 * extent insertions we have queued up so far. count can be
2925 * 0, which means to process everything in the tree at the start
2926 * of the run (but not newly added entries), or it can be some target
2927 * number you'd like to process.
2929 * Returns 0 on success or if called with an aborted transaction
2930 * Returns <0 on error and aborts the transaction
2932 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2933 struct btrfs_root *root, unsigned long count)
2935 struct rb_node *node;
2936 struct btrfs_delayed_ref_root *delayed_refs;
2937 struct btrfs_delayed_ref_head *head;
2939 int run_all = count == (unsigned long)-1;
2940 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2942 /* We'll clean this up in btrfs_cleanup_transaction */
2946 if (root->fs_info->creating_free_space_tree)
2949 if (root == root->fs_info->extent_root)
2950 root = root->fs_info->tree_root;
2952 delayed_refs = &trans->transaction->delayed_refs;
2954 count = atomic_read(&delayed_refs->num_entries) * 2;
2957 #ifdef SCRAMBLE_DELAYED_REFS
2958 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2960 trans->can_flush_pending_bgs = false;
2961 ret = __btrfs_run_delayed_refs(trans, root, count);
2963 btrfs_abort_transaction(trans, root, ret);
2968 if (!list_empty(&trans->new_bgs))
2969 btrfs_create_pending_block_groups(trans, root);
2971 spin_lock(&delayed_refs->lock);
2972 node = rb_first(&delayed_refs->href_root);
2974 spin_unlock(&delayed_refs->lock);
2977 count = (unsigned long)-1;
2980 head = rb_entry(node, struct btrfs_delayed_ref_head,
2982 if (btrfs_delayed_ref_is_head(&head->node)) {
2983 struct btrfs_delayed_ref_node *ref;
2986 atomic_inc(&ref->refs);
2988 spin_unlock(&delayed_refs->lock);
2990 * Mutex was contended, block until it's
2991 * released and try again
2993 mutex_lock(&head->mutex);
2994 mutex_unlock(&head->mutex);
2996 btrfs_put_delayed_ref(ref);
3002 node = rb_next(node);
3004 spin_unlock(&delayed_refs->lock);
3009 assert_qgroups_uptodate(trans);
3010 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3014 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3015 struct btrfs_root *root,
3016 u64 bytenr, u64 num_bytes, u64 flags,
3017 int level, int is_data)
3019 struct btrfs_delayed_extent_op *extent_op;
3022 extent_op = btrfs_alloc_delayed_extent_op();
3026 extent_op->flags_to_set = flags;
3027 extent_op->update_flags = true;
3028 extent_op->update_key = false;
3029 extent_op->is_data = is_data ? true : false;
3030 extent_op->level = level;
3032 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3033 num_bytes, extent_op);
3035 btrfs_free_delayed_extent_op(extent_op);
3039 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3040 struct btrfs_root *root,
3041 struct btrfs_path *path,
3042 u64 objectid, u64 offset, u64 bytenr)
3044 struct btrfs_delayed_ref_head *head;
3045 struct btrfs_delayed_ref_node *ref;
3046 struct btrfs_delayed_data_ref *data_ref;
3047 struct btrfs_delayed_ref_root *delayed_refs;
3050 delayed_refs = &trans->transaction->delayed_refs;
3051 spin_lock(&delayed_refs->lock);
3052 head = btrfs_find_delayed_ref_head(trans, bytenr);
3054 spin_unlock(&delayed_refs->lock);
3058 if (!mutex_trylock(&head->mutex)) {
3059 atomic_inc(&head->node.refs);
3060 spin_unlock(&delayed_refs->lock);
3062 btrfs_release_path(path);
3065 * Mutex was contended, block until it's released and let
3068 mutex_lock(&head->mutex);
3069 mutex_unlock(&head->mutex);
3070 btrfs_put_delayed_ref(&head->node);
3073 spin_unlock(&delayed_refs->lock);
3075 spin_lock(&head->lock);
3076 list_for_each_entry(ref, &head->ref_list, list) {
3077 /* If it's a shared ref we know a cross reference exists */
3078 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3083 data_ref = btrfs_delayed_node_to_data_ref(ref);
3086 * If our ref doesn't match the one we're currently looking at
3087 * then we have a cross reference.
3089 if (data_ref->root != root->root_key.objectid ||
3090 data_ref->objectid != objectid ||
3091 data_ref->offset != offset) {
3096 spin_unlock(&head->lock);
3097 mutex_unlock(&head->mutex);
3101 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3102 struct btrfs_root *root,
3103 struct btrfs_path *path,
3104 u64 objectid, u64 offset, u64 bytenr)
3106 struct btrfs_root *extent_root = root->fs_info->extent_root;
3107 struct extent_buffer *leaf;
3108 struct btrfs_extent_data_ref *ref;
3109 struct btrfs_extent_inline_ref *iref;
3110 struct btrfs_extent_item *ei;
3111 struct btrfs_key key;
3115 key.objectid = bytenr;
3116 key.offset = (u64)-1;
3117 key.type = BTRFS_EXTENT_ITEM_KEY;
3119 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3122 BUG_ON(ret == 0); /* Corruption */
3125 if (path->slots[0] == 0)
3129 leaf = path->nodes[0];
3130 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3132 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3136 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3137 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3138 if (item_size < sizeof(*ei)) {
3139 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3143 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3145 if (item_size != sizeof(*ei) +
3146 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3149 if (btrfs_extent_generation(leaf, ei) <=
3150 btrfs_root_last_snapshot(&root->root_item))
3153 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3154 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3155 BTRFS_EXTENT_DATA_REF_KEY)
3158 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3159 if (btrfs_extent_refs(leaf, ei) !=
3160 btrfs_extent_data_ref_count(leaf, ref) ||
3161 btrfs_extent_data_ref_root(leaf, ref) !=
3162 root->root_key.objectid ||
3163 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3164 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3172 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3173 struct btrfs_root *root,
3174 u64 objectid, u64 offset, u64 bytenr)
3176 struct btrfs_path *path;
3180 path = btrfs_alloc_path();
3185 ret = check_committed_ref(trans, root, path, objectid,
3187 if (ret && ret != -ENOENT)
3190 ret2 = check_delayed_ref(trans, root, path, objectid,
3192 } while (ret2 == -EAGAIN);
3194 if (ret2 && ret2 != -ENOENT) {
3199 if (ret != -ENOENT || ret2 != -ENOENT)
3202 btrfs_free_path(path);
3203 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3208 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3209 struct btrfs_root *root,
3210 struct extent_buffer *buf,
3211 int full_backref, int inc)
3218 struct btrfs_key key;
3219 struct btrfs_file_extent_item *fi;
3223 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3224 u64, u64, u64, u64, u64, u64);
3227 if (btrfs_test_is_dummy_root(root))
3230 ref_root = btrfs_header_owner(buf);
3231 nritems = btrfs_header_nritems(buf);
3232 level = btrfs_header_level(buf);
3234 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3238 process_func = btrfs_inc_extent_ref;
3240 process_func = btrfs_free_extent;
3243 parent = buf->start;
3247 for (i = 0; i < nritems; i++) {
3249 btrfs_item_key_to_cpu(buf, &key, i);
3250 if (key.type != BTRFS_EXTENT_DATA_KEY)
3252 fi = btrfs_item_ptr(buf, i,
3253 struct btrfs_file_extent_item);
3254 if (btrfs_file_extent_type(buf, fi) ==
3255 BTRFS_FILE_EXTENT_INLINE)
3257 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3261 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3262 key.offset -= btrfs_file_extent_offset(buf, fi);
3263 ret = process_func(trans, root, bytenr, num_bytes,
3264 parent, ref_root, key.objectid,
3269 bytenr = btrfs_node_blockptr(buf, i);
3270 num_bytes = root->nodesize;
3271 ret = process_func(trans, root, bytenr, num_bytes,
3272 parent, ref_root, level - 1, 0);
3282 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3283 struct extent_buffer *buf, int full_backref)
3285 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3288 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3289 struct extent_buffer *buf, int full_backref)
3291 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3294 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3295 struct btrfs_root *root,
3296 struct btrfs_path *path,
3297 struct btrfs_block_group_cache *cache)
3300 struct btrfs_root *extent_root = root->fs_info->extent_root;
3302 struct extent_buffer *leaf;
3304 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3311 leaf = path->nodes[0];
3312 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3313 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3314 btrfs_mark_buffer_dirty(leaf);
3316 btrfs_release_path(path);
3321 static struct btrfs_block_group_cache *
3322 next_block_group(struct btrfs_root *root,
3323 struct btrfs_block_group_cache *cache)
3325 struct rb_node *node;
3327 spin_lock(&root->fs_info->block_group_cache_lock);
3329 /* If our block group was removed, we need a full search. */
3330 if (RB_EMPTY_NODE(&cache->cache_node)) {
3331 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3333 spin_unlock(&root->fs_info->block_group_cache_lock);
3334 btrfs_put_block_group(cache);
3335 cache = btrfs_lookup_first_block_group(root->fs_info,
3339 node = rb_next(&cache->cache_node);
3340 btrfs_put_block_group(cache);
3342 cache = rb_entry(node, struct btrfs_block_group_cache,
3344 btrfs_get_block_group(cache);
3347 spin_unlock(&root->fs_info->block_group_cache_lock);
3351 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3352 struct btrfs_trans_handle *trans,
3353 struct btrfs_path *path)
3355 struct btrfs_root *root = block_group->fs_info->tree_root;
3356 struct inode *inode = NULL;
3358 int dcs = BTRFS_DC_ERROR;
3364 * If this block group is smaller than 100 megs don't bother caching the
3367 if (block_group->key.offset < (100 * SZ_1M)) {
3368 spin_lock(&block_group->lock);
3369 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3370 spin_unlock(&block_group->lock);
3377 inode = lookup_free_space_inode(root, block_group, path);
3378 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3379 ret = PTR_ERR(inode);
3380 btrfs_release_path(path);
3384 if (IS_ERR(inode)) {
3388 if (block_group->ro)
3391 ret = create_free_space_inode(root, trans, block_group, path);
3397 /* We've already setup this transaction, go ahead and exit */
3398 if (block_group->cache_generation == trans->transid &&
3399 i_size_read(inode)) {
3400 dcs = BTRFS_DC_SETUP;
3405 * We want to set the generation to 0, that way if anything goes wrong
3406 * from here on out we know not to trust this cache when we load up next
3409 BTRFS_I(inode)->generation = 0;
3410 ret = btrfs_update_inode(trans, root, inode);
3413 * So theoretically we could recover from this, simply set the
3414 * super cache generation to 0 so we know to invalidate the
3415 * cache, but then we'd have to keep track of the block groups
3416 * that fail this way so we know we _have_ to reset this cache
3417 * before the next commit or risk reading stale cache. So to
3418 * limit our exposure to horrible edge cases lets just abort the
3419 * transaction, this only happens in really bad situations
3422 btrfs_abort_transaction(trans, root, ret);
3427 if (i_size_read(inode) > 0) {
3428 ret = btrfs_check_trunc_cache_free_space(root,
3429 &root->fs_info->global_block_rsv);
3433 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3438 spin_lock(&block_group->lock);
3439 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3440 !btrfs_test_opt(root, SPACE_CACHE)) {
3442 * don't bother trying to write stuff out _if_
3443 * a) we're not cached,
3444 * b) we're with nospace_cache mount option.
3446 dcs = BTRFS_DC_WRITTEN;
3447 spin_unlock(&block_group->lock);
3450 spin_unlock(&block_group->lock);
3453 * We hit an ENOSPC when setting up the cache in this transaction, just
3454 * skip doing the setup, we've already cleared the cache so we're safe.
3456 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3462 * Try to preallocate enough space based on how big the block group is.
3463 * Keep in mind this has to include any pinned space which could end up
3464 * taking up quite a bit since it's not folded into the other space
3467 num_pages = div_u64(block_group->key.offset, SZ_256M);
3472 num_pages *= PAGE_SIZE;
3474 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3478 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3479 num_pages, num_pages,
3482 * Our cache requires contiguous chunks so that we don't modify a bunch
3483 * of metadata or split extents when writing the cache out, which means
3484 * we can enospc if we are heavily fragmented in addition to just normal
3485 * out of space conditions. So if we hit this just skip setting up any
3486 * other block groups for this transaction, maybe we'll unpin enough
3487 * space the next time around.
3490 dcs = BTRFS_DC_SETUP;
3491 else if (ret == -ENOSPC)
3492 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3493 btrfs_free_reserved_data_space(inode, 0, num_pages);
3498 btrfs_release_path(path);
3500 spin_lock(&block_group->lock);
3501 if (!ret && dcs == BTRFS_DC_SETUP)
3502 block_group->cache_generation = trans->transid;
3503 block_group->disk_cache_state = dcs;
3504 spin_unlock(&block_group->lock);
3509 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3510 struct btrfs_root *root)
3512 struct btrfs_block_group_cache *cache, *tmp;
3513 struct btrfs_transaction *cur_trans = trans->transaction;
3514 struct btrfs_path *path;
3516 if (list_empty(&cur_trans->dirty_bgs) ||
3517 !btrfs_test_opt(root, SPACE_CACHE))
3520 path = btrfs_alloc_path();
3524 /* Could add new block groups, use _safe just in case */
3525 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3527 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3528 cache_save_setup(cache, trans, path);
3531 btrfs_free_path(path);
3536 * transaction commit does final block group cache writeback during a
3537 * critical section where nothing is allowed to change the FS. This is
3538 * required in order for the cache to actually match the block group,
3539 * but can introduce a lot of latency into the commit.
3541 * So, btrfs_start_dirty_block_groups is here to kick off block group
3542 * cache IO. There's a chance we'll have to redo some of it if the
3543 * block group changes again during the commit, but it greatly reduces
3544 * the commit latency by getting rid of the easy block groups while
3545 * we're still allowing others to join the commit.
3547 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3548 struct btrfs_root *root)
3550 struct btrfs_block_group_cache *cache;
3551 struct btrfs_transaction *cur_trans = trans->transaction;
3554 struct btrfs_path *path = NULL;
3556 struct list_head *io = &cur_trans->io_bgs;
3557 int num_started = 0;
3560 spin_lock(&cur_trans->dirty_bgs_lock);
3561 if (list_empty(&cur_trans->dirty_bgs)) {
3562 spin_unlock(&cur_trans->dirty_bgs_lock);
3565 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3566 spin_unlock(&cur_trans->dirty_bgs_lock);
3570 * make sure all the block groups on our dirty list actually
3573 btrfs_create_pending_block_groups(trans, root);
3576 path = btrfs_alloc_path();
3582 * cache_write_mutex is here only to save us from balance or automatic
3583 * removal of empty block groups deleting this block group while we are
3584 * writing out the cache
3586 mutex_lock(&trans->transaction->cache_write_mutex);
3587 while (!list_empty(&dirty)) {
3588 cache = list_first_entry(&dirty,
3589 struct btrfs_block_group_cache,
3592 * this can happen if something re-dirties a block
3593 * group that is already under IO. Just wait for it to
3594 * finish and then do it all again
3596 if (!list_empty(&cache->io_list)) {
3597 list_del_init(&cache->io_list);
3598 btrfs_wait_cache_io(root, trans, cache,
3599 &cache->io_ctl, path,
3600 cache->key.objectid);
3601 btrfs_put_block_group(cache);
3606 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3607 * if it should update the cache_state. Don't delete
3608 * until after we wait.
3610 * Since we're not running in the commit critical section
3611 * we need the dirty_bgs_lock to protect from update_block_group
3613 spin_lock(&cur_trans->dirty_bgs_lock);
3614 list_del_init(&cache->dirty_list);
3615 spin_unlock(&cur_trans->dirty_bgs_lock);
3619 cache_save_setup(cache, trans, path);
3621 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3622 cache->io_ctl.inode = NULL;
3623 ret = btrfs_write_out_cache(root, trans, cache, path);
3624 if (ret == 0 && cache->io_ctl.inode) {
3629 * the cache_write_mutex is protecting
3632 list_add_tail(&cache->io_list, io);
3635 * if we failed to write the cache, the
3636 * generation will be bad and life goes on
3642 ret = write_one_cache_group(trans, root, path, cache);
3644 * Our block group might still be attached to the list
3645 * of new block groups in the transaction handle of some
3646 * other task (struct btrfs_trans_handle->new_bgs). This
3647 * means its block group item isn't yet in the extent
3648 * tree. If this happens ignore the error, as we will
3649 * try again later in the critical section of the
3650 * transaction commit.
3652 if (ret == -ENOENT) {
3654 spin_lock(&cur_trans->dirty_bgs_lock);
3655 if (list_empty(&cache->dirty_list)) {
3656 list_add_tail(&cache->dirty_list,
3657 &cur_trans->dirty_bgs);
3658 btrfs_get_block_group(cache);
3660 spin_unlock(&cur_trans->dirty_bgs_lock);
3662 btrfs_abort_transaction(trans, root, ret);
3666 /* if its not on the io list, we need to put the block group */
3668 btrfs_put_block_group(cache);
3674 * Avoid blocking other tasks for too long. It might even save
3675 * us from writing caches for block groups that are going to be
3678 mutex_unlock(&trans->transaction->cache_write_mutex);
3679 mutex_lock(&trans->transaction->cache_write_mutex);
3681 mutex_unlock(&trans->transaction->cache_write_mutex);
3684 * go through delayed refs for all the stuff we've just kicked off
3685 * and then loop back (just once)
3687 ret = btrfs_run_delayed_refs(trans, root, 0);
3688 if (!ret && loops == 0) {
3690 spin_lock(&cur_trans->dirty_bgs_lock);
3691 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3693 * dirty_bgs_lock protects us from concurrent block group
3694 * deletes too (not just cache_write_mutex).
3696 if (!list_empty(&dirty)) {
3697 spin_unlock(&cur_trans->dirty_bgs_lock);
3700 spin_unlock(&cur_trans->dirty_bgs_lock);
3703 btrfs_free_path(path);
3707 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3708 struct btrfs_root *root)
3710 struct btrfs_block_group_cache *cache;
3711 struct btrfs_transaction *cur_trans = trans->transaction;
3714 struct btrfs_path *path;
3715 struct list_head *io = &cur_trans->io_bgs;
3716 int num_started = 0;
3718 path = btrfs_alloc_path();
3723 * Even though we are in the critical section of the transaction commit,
3724 * we can still have concurrent tasks adding elements to this
3725 * transaction's list of dirty block groups. These tasks correspond to
3726 * endio free space workers started when writeback finishes for a
3727 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3728 * allocate new block groups as a result of COWing nodes of the root
3729 * tree when updating the free space inode. The writeback for the space
3730 * caches is triggered by an earlier call to
3731 * btrfs_start_dirty_block_groups() and iterations of the following
3733 * Also we want to do the cache_save_setup first and then run the
3734 * delayed refs to make sure we have the best chance at doing this all
3737 spin_lock(&cur_trans->dirty_bgs_lock);
3738 while (!list_empty(&cur_trans->dirty_bgs)) {
3739 cache = list_first_entry(&cur_trans->dirty_bgs,
3740 struct btrfs_block_group_cache,
3744 * this can happen if cache_save_setup re-dirties a block
3745 * group that is already under IO. Just wait for it to
3746 * finish and then do it all again
3748 if (!list_empty(&cache->io_list)) {
3749 spin_unlock(&cur_trans->dirty_bgs_lock);
3750 list_del_init(&cache->io_list);
3751 btrfs_wait_cache_io(root, trans, cache,
3752 &cache->io_ctl, path,
3753 cache->key.objectid);
3754 btrfs_put_block_group(cache);
3755 spin_lock(&cur_trans->dirty_bgs_lock);
3759 * don't remove from the dirty list until after we've waited
3762 list_del_init(&cache->dirty_list);
3763 spin_unlock(&cur_trans->dirty_bgs_lock);
3766 cache_save_setup(cache, trans, path);
3769 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3771 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3772 cache->io_ctl.inode = NULL;
3773 ret = btrfs_write_out_cache(root, trans, cache, path);
3774 if (ret == 0 && cache->io_ctl.inode) {
3777 list_add_tail(&cache->io_list, io);
3780 * if we failed to write the cache, the
3781 * generation will be bad and life goes on
3787 ret = write_one_cache_group(trans, root, path, cache);
3789 * One of the free space endio workers might have
3790 * created a new block group while updating a free space
3791 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3792 * and hasn't released its transaction handle yet, in
3793 * which case the new block group is still attached to
3794 * its transaction handle and its creation has not
3795 * finished yet (no block group item in the extent tree
3796 * yet, etc). If this is the case, wait for all free
3797 * space endio workers to finish and retry. This is a
3798 * a very rare case so no need for a more efficient and
3801 if (ret == -ENOENT) {
3802 wait_event(cur_trans->writer_wait,
3803 atomic_read(&cur_trans->num_writers) == 1);
3804 ret = write_one_cache_group(trans, root, path,
3808 btrfs_abort_transaction(trans, root, ret);
3811 /* if its not on the io list, we need to put the block group */
3813 btrfs_put_block_group(cache);
3814 spin_lock(&cur_trans->dirty_bgs_lock);
3816 spin_unlock(&cur_trans->dirty_bgs_lock);
3818 while (!list_empty(io)) {
3819 cache = list_first_entry(io, struct btrfs_block_group_cache,
3821 list_del_init(&cache->io_list);
3822 btrfs_wait_cache_io(root, trans, cache,
3823 &cache->io_ctl, path, cache->key.objectid);
3824 btrfs_put_block_group(cache);
3827 btrfs_free_path(path);
3831 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3833 struct btrfs_block_group_cache *block_group;
3836 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3837 if (!block_group || block_group->ro)
3840 btrfs_put_block_group(block_group);
3844 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3846 struct btrfs_block_group_cache *bg;
3849 bg = btrfs_lookup_block_group(fs_info, bytenr);
3853 spin_lock(&bg->lock);
3857 atomic_inc(&bg->nocow_writers);
3858 spin_unlock(&bg->lock);
3860 /* no put on block group, done by btrfs_dec_nocow_writers */
3862 btrfs_put_block_group(bg);
3868 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3870 struct btrfs_block_group_cache *bg;
3872 bg = btrfs_lookup_block_group(fs_info, bytenr);
3874 if (atomic_dec_and_test(&bg->nocow_writers))
3875 wake_up_atomic_t(&bg->nocow_writers);
3877 * Once for our lookup and once for the lookup done by a previous call
3878 * to btrfs_inc_nocow_writers()
3880 btrfs_put_block_group(bg);
3881 btrfs_put_block_group(bg);
3884 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3890 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3892 wait_on_atomic_t(&bg->nocow_writers,
3893 btrfs_wait_nocow_writers_atomic_t,
3894 TASK_UNINTERRUPTIBLE);
3897 static const char *alloc_name(u64 flags)
3900 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3902 case BTRFS_BLOCK_GROUP_METADATA:
3904 case BTRFS_BLOCK_GROUP_DATA:
3906 case BTRFS_BLOCK_GROUP_SYSTEM:
3910 return "invalid-combination";
3914 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3915 u64 total_bytes, u64 bytes_used,
3917 struct btrfs_space_info **space_info)
3919 struct btrfs_space_info *found;
3924 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3925 BTRFS_BLOCK_GROUP_RAID10))
3930 found = __find_space_info(info, flags);
3932 spin_lock(&found->lock);
3933 found->total_bytes += total_bytes;
3934 found->disk_total += total_bytes * factor;
3935 found->bytes_used += bytes_used;
3936 found->disk_used += bytes_used * factor;
3937 found->bytes_readonly += bytes_readonly;
3938 if (total_bytes > 0)
3940 spin_unlock(&found->lock);
3941 *space_info = found;
3944 found = kzalloc(sizeof(*found), GFP_NOFS);
3948 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3954 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3955 INIT_LIST_HEAD(&found->block_groups[i]);
3956 init_rwsem(&found->groups_sem);
3957 spin_lock_init(&found->lock);
3958 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3959 found->total_bytes = total_bytes;
3960 found->disk_total = total_bytes * factor;
3961 found->bytes_used = bytes_used;
3962 found->disk_used = bytes_used * factor;
3963 found->bytes_pinned = 0;
3964 found->bytes_reserved = 0;
3965 found->bytes_readonly = bytes_readonly;
3966 found->bytes_may_use = 0;
3968 found->max_extent_size = 0;
3969 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3970 found->chunk_alloc = 0;
3972 init_waitqueue_head(&found->wait);
3973 INIT_LIST_HEAD(&found->ro_bgs);
3975 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3976 info->space_info_kobj, "%s",
3977 alloc_name(found->flags));
3983 *space_info = found;
3984 list_add_rcu(&found->list, &info->space_info);
3985 if (flags & BTRFS_BLOCK_GROUP_DATA)
3986 info->data_sinfo = found;
3991 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3993 u64 extra_flags = chunk_to_extended(flags) &
3994 BTRFS_EXTENDED_PROFILE_MASK;
3996 write_seqlock(&fs_info->profiles_lock);
3997 if (flags & BTRFS_BLOCK_GROUP_DATA)
3998 fs_info->avail_data_alloc_bits |= extra_flags;
3999 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4000 fs_info->avail_metadata_alloc_bits |= extra_flags;
4001 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4002 fs_info->avail_system_alloc_bits |= extra_flags;
4003 write_sequnlock(&fs_info->profiles_lock);
4007 * returns target flags in extended format or 0 if restripe for this
4008 * chunk_type is not in progress
4010 * should be called with either volume_mutex or balance_lock held
4012 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4014 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4020 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4021 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4022 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4023 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4024 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4025 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4026 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4027 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4028 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4035 * @flags: available profiles in extended format (see ctree.h)
4037 * Returns reduced profile in chunk format. If profile changing is in
4038 * progress (either running or paused) picks the target profile (if it's
4039 * already available), otherwise falls back to plain reducing.
4041 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
4043 u64 num_devices = root->fs_info->fs_devices->rw_devices;
4049 * see if restripe for this chunk_type is in progress, if so
4050 * try to reduce to the target profile
4052 spin_lock(&root->fs_info->balance_lock);
4053 target = get_restripe_target(root->fs_info, flags);
4055 /* pick target profile only if it's already available */
4056 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4057 spin_unlock(&root->fs_info->balance_lock);
4058 return extended_to_chunk(target);
4061 spin_unlock(&root->fs_info->balance_lock);
4063 /* First, mask out the RAID levels which aren't possible */
4064 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4065 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4066 allowed |= btrfs_raid_group[raid_type];
4070 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4071 allowed = BTRFS_BLOCK_GROUP_RAID6;
4072 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4073 allowed = BTRFS_BLOCK_GROUP_RAID5;
4074 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4075 allowed = BTRFS_BLOCK_GROUP_RAID10;
4076 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4077 allowed = BTRFS_BLOCK_GROUP_RAID1;
4078 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4079 allowed = BTRFS_BLOCK_GROUP_RAID0;
4081 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4083 return extended_to_chunk(flags | allowed);
4086 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
4093 seq = read_seqbegin(&root->fs_info->profiles_lock);
4095 if (flags & BTRFS_BLOCK_GROUP_DATA)
4096 flags |= root->fs_info->avail_data_alloc_bits;
4097 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4098 flags |= root->fs_info->avail_system_alloc_bits;
4099 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4100 flags |= root->fs_info->avail_metadata_alloc_bits;
4101 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4103 return btrfs_reduce_alloc_profile(root, flags);
4106 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4112 flags = BTRFS_BLOCK_GROUP_DATA;
4113 else if (root == root->fs_info->chunk_root)
4114 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4116 flags = BTRFS_BLOCK_GROUP_METADATA;
4118 ret = get_alloc_profile(root, flags);
4122 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4124 struct btrfs_space_info *data_sinfo;
4125 struct btrfs_root *root = BTRFS_I(inode)->root;
4126 struct btrfs_fs_info *fs_info = root->fs_info;
4129 int need_commit = 2;
4130 int have_pinned_space;
4132 /* make sure bytes are sectorsize aligned */
4133 bytes = ALIGN(bytes, root->sectorsize);
4135 if (btrfs_is_free_space_inode(inode)) {
4137 ASSERT(current->journal_info);
4140 data_sinfo = fs_info->data_sinfo;
4145 /* make sure we have enough space to handle the data first */
4146 spin_lock(&data_sinfo->lock);
4147 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4148 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4149 data_sinfo->bytes_may_use;
4151 if (used + bytes > data_sinfo->total_bytes) {
4152 struct btrfs_trans_handle *trans;
4155 * if we don't have enough free bytes in this space then we need
4156 * to alloc a new chunk.
4158 if (!data_sinfo->full) {
4161 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4162 spin_unlock(&data_sinfo->lock);
4164 alloc_target = btrfs_get_alloc_profile(root, 1);
4166 * It is ugly that we don't call nolock join
4167 * transaction for the free space inode case here.
4168 * But it is safe because we only do the data space
4169 * reservation for the free space cache in the
4170 * transaction context, the common join transaction
4171 * just increase the counter of the current transaction
4172 * handler, doesn't try to acquire the trans_lock of
4175 trans = btrfs_join_transaction(root);
4177 return PTR_ERR(trans);
4179 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4181 CHUNK_ALLOC_NO_FORCE);
4182 btrfs_end_transaction(trans, root);
4187 have_pinned_space = 1;
4193 data_sinfo = fs_info->data_sinfo;
4199 * If we don't have enough pinned space to deal with this
4200 * allocation, and no removed chunk in current transaction,
4201 * don't bother committing the transaction.
4203 have_pinned_space = percpu_counter_compare(
4204 &data_sinfo->total_bytes_pinned,
4205 used + bytes - data_sinfo->total_bytes);
4206 spin_unlock(&data_sinfo->lock);
4208 /* commit the current transaction and try again */
4211 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4214 if (need_commit > 0) {
4215 btrfs_start_delalloc_roots(fs_info, 0, -1);
4216 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
4219 trans = btrfs_join_transaction(root);
4221 return PTR_ERR(trans);
4222 if (have_pinned_space >= 0 ||
4223 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4224 &trans->transaction->flags) ||
4226 ret = btrfs_commit_transaction(trans, root);
4230 * The cleaner kthread might still be doing iput
4231 * operations. Wait for it to finish so that
4232 * more space is released.
4234 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4235 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4238 btrfs_end_transaction(trans, root);
4242 trace_btrfs_space_reservation(root->fs_info,
4243 "space_info:enospc",
4244 data_sinfo->flags, bytes, 1);
4247 data_sinfo->bytes_may_use += bytes;
4248 trace_btrfs_space_reservation(root->fs_info, "space_info",
4249 data_sinfo->flags, bytes, 1);
4250 spin_unlock(&data_sinfo->lock);
4256 * New check_data_free_space() with ability for precious data reservation
4257 * Will replace old btrfs_check_data_free_space(), but for patch split,
4258 * add a new function first and then replace it.
4260 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4262 struct btrfs_root *root = BTRFS_I(inode)->root;
4265 /* align the range */
4266 len = round_up(start + len, root->sectorsize) -
4267 round_down(start, root->sectorsize);
4268 start = round_down(start, root->sectorsize);
4270 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4275 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4277 * TODO: Find a good method to avoid reserve data space for NOCOW
4278 * range, but don't impact performance on quota disable case.
4280 ret = btrfs_qgroup_reserve_data(inode, start, len);
4285 * Called if we need to clear a data reservation for this inode
4286 * Normally in a error case.
4288 * This one will *NOT* use accurate qgroup reserved space API, just for case
4289 * which we can't sleep and is sure it won't affect qgroup reserved space.
4290 * Like clear_bit_hook().
4292 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4295 struct btrfs_root *root = BTRFS_I(inode)->root;
4296 struct btrfs_space_info *data_sinfo;
4298 /* Make sure the range is aligned to sectorsize */
4299 len = round_up(start + len, root->sectorsize) -
4300 round_down(start, root->sectorsize);
4301 start = round_down(start, root->sectorsize);
4303 data_sinfo = root->fs_info->data_sinfo;
4304 spin_lock(&data_sinfo->lock);
4305 if (WARN_ON(data_sinfo->bytes_may_use < len))
4306 data_sinfo->bytes_may_use = 0;
4308 data_sinfo->bytes_may_use -= len;
4309 trace_btrfs_space_reservation(root->fs_info, "space_info",
4310 data_sinfo->flags, len, 0);
4311 spin_unlock(&data_sinfo->lock);
4315 * Called if we need to clear a data reservation for this inode
4316 * Normally in a error case.
4318 * This one will handle the per-inode data rsv map for accurate reserved
4321 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4323 btrfs_free_reserved_data_space_noquota(inode, start, len);
4324 btrfs_qgroup_free_data(inode, start, len);
4327 static void force_metadata_allocation(struct btrfs_fs_info *info)
4329 struct list_head *head = &info->space_info;
4330 struct btrfs_space_info *found;
4333 list_for_each_entry_rcu(found, head, list) {
4334 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4335 found->force_alloc = CHUNK_ALLOC_FORCE;
4340 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4342 return (global->size << 1);
4345 static int should_alloc_chunk(struct btrfs_root *root,
4346 struct btrfs_space_info *sinfo, int force)
4348 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4349 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4350 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4353 if (force == CHUNK_ALLOC_FORCE)
4357 * We need to take into account the global rsv because for all intents
4358 * and purposes it's used space. Don't worry about locking the
4359 * global_rsv, it doesn't change except when the transaction commits.
4361 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4362 num_allocated += calc_global_rsv_need_space(global_rsv);
4365 * in limited mode, we want to have some free space up to
4366 * about 1% of the FS size.
4368 if (force == CHUNK_ALLOC_LIMITED) {
4369 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4370 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4372 if (num_bytes - num_allocated < thresh)
4376 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4381 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4385 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4386 BTRFS_BLOCK_GROUP_RAID0 |
4387 BTRFS_BLOCK_GROUP_RAID5 |
4388 BTRFS_BLOCK_GROUP_RAID6))
4389 num_dev = root->fs_info->fs_devices->rw_devices;
4390 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4393 num_dev = 1; /* DUP or single */
4399 * If @is_allocation is true, reserve space in the system space info necessary
4400 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4403 void check_system_chunk(struct btrfs_trans_handle *trans,
4404 struct btrfs_root *root,
4407 struct btrfs_space_info *info;
4414 * Needed because we can end up allocating a system chunk and for an
4415 * atomic and race free space reservation in the chunk block reserve.
4417 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4419 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4420 spin_lock(&info->lock);
4421 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4422 info->bytes_reserved - info->bytes_readonly -
4423 info->bytes_may_use;
4424 spin_unlock(&info->lock);
4426 num_devs = get_profile_num_devs(root, type);
4428 /* num_devs device items to update and 1 chunk item to add or remove */
4429 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4430 btrfs_calc_trans_metadata_size(root, 1);
4432 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4433 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4434 left, thresh, type);
4435 dump_space_info(info, 0, 0);
4438 if (left < thresh) {
4441 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4443 * Ignore failure to create system chunk. We might end up not
4444 * needing it, as we might not need to COW all nodes/leafs from
4445 * the paths we visit in the chunk tree (they were already COWed
4446 * or created in the current transaction for example).
4448 ret = btrfs_alloc_chunk(trans, root, flags);
4452 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4453 &root->fs_info->chunk_block_rsv,
4454 thresh, BTRFS_RESERVE_NO_FLUSH);
4456 trans->chunk_bytes_reserved += thresh;
4460 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4461 struct btrfs_root *extent_root, u64 flags, int force)
4463 struct btrfs_space_info *space_info;
4464 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4465 int wait_for_alloc = 0;
4468 /* Don't re-enter if we're already allocating a chunk */
4469 if (trans->allocating_chunk)
4472 space_info = __find_space_info(extent_root->fs_info, flags);
4474 ret = update_space_info(extent_root->fs_info, flags,
4475 0, 0, 0, &space_info);
4476 BUG_ON(ret); /* -ENOMEM */
4478 BUG_ON(!space_info); /* Logic error */
4481 spin_lock(&space_info->lock);
4482 if (force < space_info->force_alloc)
4483 force = space_info->force_alloc;
4484 if (space_info->full) {
4485 if (should_alloc_chunk(extent_root, space_info, force))
4489 spin_unlock(&space_info->lock);
4493 if (!should_alloc_chunk(extent_root, space_info, force)) {
4494 spin_unlock(&space_info->lock);
4496 } else if (space_info->chunk_alloc) {
4499 space_info->chunk_alloc = 1;
4502 spin_unlock(&space_info->lock);
4504 mutex_lock(&fs_info->chunk_mutex);
4507 * The chunk_mutex is held throughout the entirety of a chunk
4508 * allocation, so once we've acquired the chunk_mutex we know that the
4509 * other guy is done and we need to recheck and see if we should
4512 if (wait_for_alloc) {
4513 mutex_unlock(&fs_info->chunk_mutex);
4518 trans->allocating_chunk = true;
4521 * If we have mixed data/metadata chunks we want to make sure we keep
4522 * allocating mixed chunks instead of individual chunks.
4524 if (btrfs_mixed_space_info(space_info))
4525 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4528 * if we're doing a data chunk, go ahead and make sure that
4529 * we keep a reasonable number of metadata chunks allocated in the
4532 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4533 fs_info->data_chunk_allocations++;
4534 if (!(fs_info->data_chunk_allocations %
4535 fs_info->metadata_ratio))
4536 force_metadata_allocation(fs_info);
4540 * Check if we have enough space in SYSTEM chunk because we may need
4541 * to update devices.
4543 check_system_chunk(trans, extent_root, flags);
4545 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4546 trans->allocating_chunk = false;
4548 spin_lock(&space_info->lock);
4549 if (ret < 0 && ret != -ENOSPC)
4552 space_info->full = 1;
4556 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4558 space_info->chunk_alloc = 0;
4559 spin_unlock(&space_info->lock);
4560 mutex_unlock(&fs_info->chunk_mutex);
4562 * When we allocate a new chunk we reserve space in the chunk block
4563 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4564 * add new nodes/leafs to it if we end up needing to do it when
4565 * inserting the chunk item and updating device items as part of the
4566 * second phase of chunk allocation, performed by
4567 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4568 * large number of new block groups to create in our transaction
4569 * handle's new_bgs list to avoid exhausting the chunk block reserve
4570 * in extreme cases - like having a single transaction create many new
4571 * block groups when starting to write out the free space caches of all
4572 * the block groups that were made dirty during the lifetime of the
4575 if (trans->can_flush_pending_bgs &&
4576 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4577 btrfs_create_pending_block_groups(trans, trans->root);
4578 btrfs_trans_release_chunk_metadata(trans);
4583 static int can_overcommit(struct btrfs_root *root,
4584 struct btrfs_space_info *space_info, u64 bytes,
4585 enum btrfs_reserve_flush_enum flush)
4587 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4588 u64 profile = btrfs_get_alloc_profile(root, 0);
4593 used = space_info->bytes_used + space_info->bytes_reserved +
4594 space_info->bytes_pinned + space_info->bytes_readonly;
4597 * We only want to allow over committing if we have lots of actual space
4598 * free, but if we don't have enough space to handle the global reserve
4599 * space then we could end up having a real enospc problem when trying
4600 * to allocate a chunk or some other such important allocation.
4602 spin_lock(&global_rsv->lock);
4603 space_size = calc_global_rsv_need_space(global_rsv);
4604 spin_unlock(&global_rsv->lock);
4605 if (used + space_size >= space_info->total_bytes)
4608 used += space_info->bytes_may_use;
4610 spin_lock(&root->fs_info->free_chunk_lock);
4611 avail = root->fs_info->free_chunk_space;
4612 spin_unlock(&root->fs_info->free_chunk_lock);
4615 * If we have dup, raid1 or raid10 then only half of the free
4616 * space is actually useable. For raid56, the space info used
4617 * doesn't include the parity drive, so we don't have to
4620 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4621 BTRFS_BLOCK_GROUP_RAID1 |
4622 BTRFS_BLOCK_GROUP_RAID10))
4626 * If we aren't flushing all things, let us overcommit up to
4627 * 1/2th of the space. If we can flush, don't let us overcommit
4628 * too much, let it overcommit up to 1/8 of the space.
4630 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4635 if (used + bytes < space_info->total_bytes + avail)
4640 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4641 unsigned long nr_pages, int nr_items)
4643 struct super_block *sb = root->fs_info->sb;
4645 if (down_read_trylock(&sb->s_umount)) {
4646 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4647 up_read(&sb->s_umount);
4650 * We needn't worry the filesystem going from r/w to r/o though
4651 * we don't acquire ->s_umount mutex, because the filesystem
4652 * should guarantee the delalloc inodes list be empty after
4653 * the filesystem is readonly(all dirty pages are written to
4656 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4657 if (!current->journal_info)
4658 btrfs_wait_ordered_roots(root->fs_info, nr_items,
4663 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4668 bytes = btrfs_calc_trans_metadata_size(root, 1);
4669 nr = (int)div64_u64(to_reclaim, bytes);
4675 #define EXTENT_SIZE_PER_ITEM SZ_256K
4678 * shrink metadata reservation for delalloc
4680 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4683 struct btrfs_block_rsv *block_rsv;
4684 struct btrfs_space_info *space_info;
4685 struct btrfs_trans_handle *trans;
4689 unsigned long nr_pages;
4692 enum btrfs_reserve_flush_enum flush;
4694 /* Calc the number of the pages we need flush for space reservation */
4695 items = calc_reclaim_items_nr(root, to_reclaim);
4696 to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM;
4698 trans = (struct btrfs_trans_handle *)current->journal_info;
4699 block_rsv = &root->fs_info->delalloc_block_rsv;
4700 space_info = block_rsv->space_info;
4702 delalloc_bytes = percpu_counter_sum_positive(
4703 &root->fs_info->delalloc_bytes);
4704 if (delalloc_bytes == 0) {
4708 btrfs_wait_ordered_roots(root->fs_info, items,
4714 while (delalloc_bytes && loops < 3) {
4715 max_reclaim = min(delalloc_bytes, to_reclaim);
4716 nr_pages = max_reclaim >> PAGE_SHIFT;
4717 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4719 * We need to wait for the async pages to actually start before
4722 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4726 if (max_reclaim <= nr_pages)
4729 max_reclaim -= nr_pages;
4731 wait_event(root->fs_info->async_submit_wait,
4732 atomic_read(&root->fs_info->async_delalloc_pages) <=
4736 flush = BTRFS_RESERVE_FLUSH_ALL;
4738 flush = BTRFS_RESERVE_NO_FLUSH;
4739 spin_lock(&space_info->lock);
4740 if (can_overcommit(root, space_info, orig, flush)) {
4741 spin_unlock(&space_info->lock);
4744 spin_unlock(&space_info->lock);
4747 if (wait_ordered && !trans) {
4748 btrfs_wait_ordered_roots(root->fs_info, items,
4751 time_left = schedule_timeout_killable(1);
4755 delalloc_bytes = percpu_counter_sum_positive(
4756 &root->fs_info->delalloc_bytes);
4761 * maybe_commit_transaction - possibly commit the transaction if its ok to
4762 * @root - the root we're allocating for
4763 * @bytes - the number of bytes we want to reserve
4764 * @force - force the commit
4766 * This will check to make sure that committing the transaction will actually
4767 * get us somewhere and then commit the transaction if it does. Otherwise it
4768 * will return -ENOSPC.
4770 static int may_commit_transaction(struct btrfs_root *root,
4771 struct btrfs_space_info *space_info,
4772 u64 bytes, int force)
4774 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4775 struct btrfs_trans_handle *trans;
4777 trans = (struct btrfs_trans_handle *)current->journal_info;
4784 /* See if there is enough pinned space to make this reservation */
4785 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4790 * See if there is some space in the delayed insertion reservation for
4793 if (space_info != delayed_rsv->space_info)
4796 spin_lock(&delayed_rsv->lock);
4797 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4798 bytes - delayed_rsv->size) >= 0) {
4799 spin_unlock(&delayed_rsv->lock);
4802 spin_unlock(&delayed_rsv->lock);
4805 trans = btrfs_join_transaction(root);
4809 return btrfs_commit_transaction(trans, root);
4813 FLUSH_DELAYED_ITEMS_NR = 1,
4814 FLUSH_DELAYED_ITEMS = 2,
4816 FLUSH_DELALLOC_WAIT = 4,
4821 static int flush_space(struct btrfs_root *root,
4822 struct btrfs_space_info *space_info, u64 num_bytes,
4823 u64 orig_bytes, int state)
4825 struct btrfs_trans_handle *trans;
4830 case FLUSH_DELAYED_ITEMS_NR:
4831 case FLUSH_DELAYED_ITEMS:
4832 if (state == FLUSH_DELAYED_ITEMS_NR)
4833 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4837 trans = btrfs_join_transaction(root);
4838 if (IS_ERR(trans)) {
4839 ret = PTR_ERR(trans);
4842 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4843 btrfs_end_transaction(trans, root);
4845 case FLUSH_DELALLOC:
4846 case FLUSH_DELALLOC_WAIT:
4847 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4848 state == FLUSH_DELALLOC_WAIT);
4851 trans = btrfs_join_transaction(root);
4852 if (IS_ERR(trans)) {
4853 ret = PTR_ERR(trans);
4856 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4857 btrfs_get_alloc_profile(root, 0),
4858 CHUNK_ALLOC_NO_FORCE);
4859 btrfs_end_transaction(trans, root);
4864 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4875 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4876 struct btrfs_space_info *space_info)
4882 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4883 spin_lock(&space_info->lock);
4884 if (can_overcommit(root, space_info, to_reclaim,
4885 BTRFS_RESERVE_FLUSH_ALL)) {
4890 used = space_info->bytes_used + space_info->bytes_reserved +
4891 space_info->bytes_pinned + space_info->bytes_readonly +
4892 space_info->bytes_may_use;
4893 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4894 expected = div_factor_fine(space_info->total_bytes, 95);
4896 expected = div_factor_fine(space_info->total_bytes, 90);
4898 if (used > expected)
4899 to_reclaim = used - expected;
4902 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4903 space_info->bytes_reserved);
4905 spin_unlock(&space_info->lock);
4910 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4911 struct btrfs_fs_info *fs_info, u64 used)
4913 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4915 /* If we're just plain full then async reclaim just slows us down. */
4916 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4919 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4920 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4923 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4924 struct btrfs_fs_info *fs_info,
4929 spin_lock(&space_info->lock);
4931 * We run out of space and have not got any free space via flush_space,
4932 * so don't bother doing async reclaim.
4934 if (flush_state > COMMIT_TRANS && space_info->full) {
4935 spin_unlock(&space_info->lock);
4939 used = space_info->bytes_used + space_info->bytes_reserved +
4940 space_info->bytes_pinned + space_info->bytes_readonly +
4941 space_info->bytes_may_use;
4942 if (need_do_async_reclaim(space_info, fs_info, used)) {
4943 spin_unlock(&space_info->lock);
4946 spin_unlock(&space_info->lock);
4951 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4953 struct btrfs_fs_info *fs_info;
4954 struct btrfs_space_info *space_info;
4958 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4959 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4961 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4966 flush_state = FLUSH_DELAYED_ITEMS_NR;
4968 flush_space(fs_info->fs_root, space_info, to_reclaim,
4969 to_reclaim, flush_state);
4971 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4974 } while (flush_state < COMMIT_TRANS);
4977 void btrfs_init_async_reclaim_work(struct work_struct *work)
4979 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4983 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4984 * @root - the root we're allocating for
4985 * @block_rsv - the block_rsv we're allocating for
4986 * @orig_bytes - the number of bytes we want
4987 * @flush - whether or not we can flush to make our reservation
4989 * This will reserve orig_bytes number of bytes from the space info associated
4990 * with the block_rsv. If there is not enough space it will make an attempt to
4991 * flush out space to make room. It will do this by flushing delalloc if
4992 * possible or committing the transaction. If flush is 0 then no attempts to
4993 * regain reservations will be made and this will fail if there is not enough
4996 static int reserve_metadata_bytes(struct btrfs_root *root,
4997 struct btrfs_block_rsv *block_rsv,
4999 enum btrfs_reserve_flush_enum flush)
5001 struct btrfs_space_info *space_info = block_rsv->space_info;
5003 u64 num_bytes = orig_bytes;
5004 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5006 bool flushing = false;
5010 spin_lock(&space_info->lock);
5012 * We only want to wait if somebody other than us is flushing and we
5013 * are actually allowed to flush all things.
5015 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
5016 space_info->flush) {
5017 spin_unlock(&space_info->lock);
5019 * If we have a trans handle we can't wait because the flusher
5020 * may have to commit the transaction, which would mean we would
5021 * deadlock since we are waiting for the flusher to finish, but
5022 * hold the current transaction open.
5024 if (current->journal_info)
5026 ret = wait_event_killable(space_info->wait, !space_info->flush);
5027 /* Must have been killed, return */
5031 spin_lock(&space_info->lock);
5035 used = space_info->bytes_used + space_info->bytes_reserved +
5036 space_info->bytes_pinned + space_info->bytes_readonly +
5037 space_info->bytes_may_use;
5040 * The idea here is that we've not already over-reserved the block group
5041 * then we can go ahead and save our reservation first and then start
5042 * flushing if we need to. Otherwise if we've already overcommitted
5043 * lets start flushing stuff first and then come back and try to make
5046 if (used <= space_info->total_bytes) {
5047 if (used + orig_bytes <= space_info->total_bytes) {
5048 space_info->bytes_may_use += orig_bytes;
5049 trace_btrfs_space_reservation(root->fs_info,
5050 "space_info", space_info->flags, orig_bytes, 1);
5054 * Ok set num_bytes to orig_bytes since we aren't
5055 * overocmmitted, this way we only try and reclaim what
5058 num_bytes = orig_bytes;
5062 * Ok we're over committed, set num_bytes to the overcommitted
5063 * amount plus the amount of bytes that we need for this
5066 num_bytes = used - space_info->total_bytes +
5070 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
5071 space_info->bytes_may_use += orig_bytes;
5072 trace_btrfs_space_reservation(root->fs_info, "space_info",
5073 space_info->flags, orig_bytes,
5079 * Couldn't make our reservation, save our place so while we're trying
5080 * to reclaim space we can actually use it instead of somebody else
5081 * stealing it from us.
5083 * We make the other tasks wait for the flush only when we can flush
5086 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5088 space_info->flush = 1;
5089 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5092 * We will do the space reservation dance during log replay,
5093 * which means we won't have fs_info->fs_root set, so don't do
5094 * the async reclaim as we will panic.
5096 if (!root->fs_info->log_root_recovering &&
5097 need_do_async_reclaim(space_info, root->fs_info, used) &&
5098 !work_busy(&root->fs_info->async_reclaim_work))
5099 queue_work(system_unbound_wq,
5100 &root->fs_info->async_reclaim_work);
5102 spin_unlock(&space_info->lock);
5104 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5107 ret = flush_space(root, space_info, num_bytes, orig_bytes,
5112 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
5113 * would happen. So skip delalloc flush.
5115 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5116 (flush_state == FLUSH_DELALLOC ||
5117 flush_state == FLUSH_DELALLOC_WAIT))
5118 flush_state = ALLOC_CHUNK;
5122 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5123 flush_state < COMMIT_TRANS)
5125 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
5126 flush_state <= COMMIT_TRANS)
5130 if (ret == -ENOSPC &&
5131 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5132 struct btrfs_block_rsv *global_rsv =
5133 &root->fs_info->global_block_rsv;
5135 if (block_rsv != global_rsv &&
5136 !block_rsv_use_bytes(global_rsv, orig_bytes))
5140 trace_btrfs_space_reservation(root->fs_info,
5141 "space_info:enospc",
5142 space_info->flags, orig_bytes, 1);
5144 spin_lock(&space_info->lock);
5145 space_info->flush = 0;
5146 wake_up_all(&space_info->wait);
5147 spin_unlock(&space_info->lock);
5152 static struct btrfs_block_rsv *get_block_rsv(
5153 const struct btrfs_trans_handle *trans,
5154 const struct btrfs_root *root)
5156 struct btrfs_block_rsv *block_rsv = NULL;
5158 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5159 (root == root->fs_info->csum_root && trans->adding_csums) ||
5160 (root == root->fs_info->uuid_root))
5161 block_rsv = trans->block_rsv;
5164 block_rsv = root->block_rsv;
5167 block_rsv = &root->fs_info->empty_block_rsv;
5172 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5176 spin_lock(&block_rsv->lock);
5177 if (block_rsv->reserved >= num_bytes) {
5178 block_rsv->reserved -= num_bytes;
5179 if (block_rsv->reserved < block_rsv->size)
5180 block_rsv->full = 0;
5183 spin_unlock(&block_rsv->lock);
5187 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5188 u64 num_bytes, int update_size)
5190 spin_lock(&block_rsv->lock);
5191 block_rsv->reserved += num_bytes;
5193 block_rsv->size += num_bytes;
5194 else if (block_rsv->reserved >= block_rsv->size)
5195 block_rsv->full = 1;
5196 spin_unlock(&block_rsv->lock);
5199 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5200 struct btrfs_block_rsv *dest, u64 num_bytes,
5203 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5206 if (global_rsv->space_info != dest->space_info)
5209 spin_lock(&global_rsv->lock);
5210 min_bytes = div_factor(global_rsv->size, min_factor);
5211 if (global_rsv->reserved < min_bytes + num_bytes) {
5212 spin_unlock(&global_rsv->lock);
5215 global_rsv->reserved -= num_bytes;
5216 if (global_rsv->reserved < global_rsv->size)
5217 global_rsv->full = 0;
5218 spin_unlock(&global_rsv->lock);
5220 block_rsv_add_bytes(dest, num_bytes, 1);
5224 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5225 struct btrfs_block_rsv *block_rsv,
5226 struct btrfs_block_rsv *dest, u64 num_bytes)
5228 struct btrfs_space_info *space_info = block_rsv->space_info;
5230 spin_lock(&block_rsv->lock);
5231 if (num_bytes == (u64)-1)
5232 num_bytes = block_rsv->size;
5233 block_rsv->size -= num_bytes;
5234 if (block_rsv->reserved >= block_rsv->size) {
5235 num_bytes = block_rsv->reserved - block_rsv->size;
5236 block_rsv->reserved = block_rsv->size;
5237 block_rsv->full = 1;
5241 spin_unlock(&block_rsv->lock);
5243 if (num_bytes > 0) {
5245 spin_lock(&dest->lock);
5249 bytes_to_add = dest->size - dest->reserved;
5250 bytes_to_add = min(num_bytes, bytes_to_add);
5251 dest->reserved += bytes_to_add;
5252 if (dest->reserved >= dest->size)
5254 num_bytes -= bytes_to_add;
5256 spin_unlock(&dest->lock);
5259 spin_lock(&space_info->lock);
5260 space_info->bytes_may_use -= num_bytes;
5261 trace_btrfs_space_reservation(fs_info, "space_info",
5262 space_info->flags, num_bytes, 0);
5263 spin_unlock(&space_info->lock);
5268 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5269 struct btrfs_block_rsv *dst, u64 num_bytes)
5273 ret = block_rsv_use_bytes(src, num_bytes);
5277 block_rsv_add_bytes(dst, num_bytes, 1);
5281 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5283 memset(rsv, 0, sizeof(*rsv));
5284 spin_lock_init(&rsv->lock);
5288 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5289 unsigned short type)
5291 struct btrfs_block_rsv *block_rsv;
5292 struct btrfs_fs_info *fs_info = root->fs_info;
5294 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5298 btrfs_init_block_rsv(block_rsv, type);
5299 block_rsv->space_info = __find_space_info(fs_info,
5300 BTRFS_BLOCK_GROUP_METADATA);
5304 void btrfs_free_block_rsv(struct btrfs_root *root,
5305 struct btrfs_block_rsv *rsv)
5309 btrfs_block_rsv_release(root, rsv, (u64)-1);
5313 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5318 int btrfs_block_rsv_add(struct btrfs_root *root,
5319 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5320 enum btrfs_reserve_flush_enum flush)
5327 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5329 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5336 int btrfs_block_rsv_check(struct btrfs_root *root,
5337 struct btrfs_block_rsv *block_rsv, int min_factor)
5345 spin_lock(&block_rsv->lock);
5346 num_bytes = div_factor(block_rsv->size, min_factor);
5347 if (block_rsv->reserved >= num_bytes)
5349 spin_unlock(&block_rsv->lock);
5354 int btrfs_block_rsv_refill(struct btrfs_root *root,
5355 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5356 enum btrfs_reserve_flush_enum flush)
5364 spin_lock(&block_rsv->lock);
5365 num_bytes = min_reserved;
5366 if (block_rsv->reserved >= num_bytes)
5369 num_bytes -= block_rsv->reserved;
5370 spin_unlock(&block_rsv->lock);
5375 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5377 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5384 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5385 struct btrfs_block_rsv *dst_rsv,
5388 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5391 void btrfs_block_rsv_release(struct btrfs_root *root,
5392 struct btrfs_block_rsv *block_rsv,
5395 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5396 if (global_rsv == block_rsv ||
5397 block_rsv->space_info != global_rsv->space_info)
5399 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5404 * helper to calculate size of global block reservation.
5405 * the desired value is sum of space used by extent tree,
5406 * checksum tree and root tree
5408 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5410 struct btrfs_space_info *sinfo;
5414 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5416 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5417 spin_lock(&sinfo->lock);
5418 data_used = sinfo->bytes_used;
5419 spin_unlock(&sinfo->lock);
5421 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5422 spin_lock(&sinfo->lock);
5423 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5425 meta_used = sinfo->bytes_used;
5426 spin_unlock(&sinfo->lock);
5428 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5430 num_bytes += div_u64(data_used + meta_used, 50);
5432 if (num_bytes * 3 > meta_used)
5433 num_bytes = div_u64(meta_used, 3);
5435 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5438 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5440 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5441 struct btrfs_space_info *sinfo = block_rsv->space_info;
5444 num_bytes = calc_global_metadata_size(fs_info);
5446 spin_lock(&sinfo->lock);
5447 spin_lock(&block_rsv->lock);
5449 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5451 if (block_rsv->reserved < block_rsv->size) {
5452 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5453 sinfo->bytes_reserved + sinfo->bytes_readonly +
5454 sinfo->bytes_may_use;
5455 if (sinfo->total_bytes > num_bytes) {
5456 num_bytes = sinfo->total_bytes - num_bytes;
5457 num_bytes = min(num_bytes,
5458 block_rsv->size - block_rsv->reserved);
5459 block_rsv->reserved += num_bytes;
5460 sinfo->bytes_may_use += num_bytes;
5461 trace_btrfs_space_reservation(fs_info, "space_info",
5462 sinfo->flags, num_bytes,
5465 } else if (block_rsv->reserved > block_rsv->size) {
5466 num_bytes = block_rsv->reserved - block_rsv->size;
5467 sinfo->bytes_may_use -= num_bytes;
5468 trace_btrfs_space_reservation(fs_info, "space_info",
5469 sinfo->flags, num_bytes, 0);
5470 block_rsv->reserved = block_rsv->size;
5473 if (block_rsv->reserved == block_rsv->size)
5474 block_rsv->full = 1;
5476 block_rsv->full = 0;
5478 spin_unlock(&block_rsv->lock);
5479 spin_unlock(&sinfo->lock);
5482 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5484 struct btrfs_space_info *space_info;
5486 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5487 fs_info->chunk_block_rsv.space_info = space_info;
5489 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5490 fs_info->global_block_rsv.space_info = space_info;
5491 fs_info->delalloc_block_rsv.space_info = space_info;
5492 fs_info->trans_block_rsv.space_info = space_info;
5493 fs_info->empty_block_rsv.space_info = space_info;
5494 fs_info->delayed_block_rsv.space_info = space_info;
5496 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5497 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5498 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5499 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5500 if (fs_info->quota_root)
5501 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5502 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5504 update_global_block_rsv(fs_info);
5507 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5509 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5511 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5512 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5513 WARN_ON(fs_info->trans_block_rsv.size > 0);
5514 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5515 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5516 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5517 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5518 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5521 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5522 struct btrfs_root *root)
5524 if (!trans->block_rsv)
5527 if (!trans->bytes_reserved)
5530 trace_btrfs_space_reservation(root->fs_info, "transaction",
5531 trans->transid, trans->bytes_reserved, 0);
5532 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5533 trans->bytes_reserved = 0;
5537 * To be called after all the new block groups attached to the transaction
5538 * handle have been created (btrfs_create_pending_block_groups()).
5540 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5542 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5544 if (!trans->chunk_bytes_reserved)
5547 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5549 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5550 trans->chunk_bytes_reserved);
5551 trans->chunk_bytes_reserved = 0;
5554 /* Can only return 0 or -ENOSPC */
5555 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5556 struct inode *inode)
5558 struct btrfs_root *root = BTRFS_I(inode)->root;
5559 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5560 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5563 * We need to hold space in order to delete our orphan item once we've
5564 * added it, so this takes the reservation so we can release it later
5565 * when we are truly done with the orphan item.
5567 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5568 trace_btrfs_space_reservation(root->fs_info, "orphan",
5569 btrfs_ino(inode), num_bytes, 1);
5570 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5573 void btrfs_orphan_release_metadata(struct inode *inode)
5575 struct btrfs_root *root = BTRFS_I(inode)->root;
5576 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5577 trace_btrfs_space_reservation(root->fs_info, "orphan",
5578 btrfs_ino(inode), num_bytes, 0);
5579 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5583 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5584 * root: the root of the parent directory
5585 * rsv: block reservation
5586 * items: the number of items that we need do reservation
5587 * qgroup_reserved: used to return the reserved size in qgroup
5589 * This function is used to reserve the space for snapshot/subvolume
5590 * creation and deletion. Those operations are different with the
5591 * common file/directory operations, they change two fs/file trees
5592 * and root tree, the number of items that the qgroup reserves is
5593 * different with the free space reservation. So we can not use
5594 * the space reservation mechanism in start_transaction().
5596 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5597 struct btrfs_block_rsv *rsv,
5599 u64 *qgroup_reserved,
5600 bool use_global_rsv)
5604 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5606 if (root->fs_info->quota_enabled) {
5607 /* One for parent inode, two for dir entries */
5608 num_bytes = 3 * root->nodesize;
5609 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5616 *qgroup_reserved = num_bytes;
5618 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5619 rsv->space_info = __find_space_info(root->fs_info,
5620 BTRFS_BLOCK_GROUP_METADATA);
5621 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5622 BTRFS_RESERVE_FLUSH_ALL);
5624 if (ret == -ENOSPC && use_global_rsv)
5625 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5627 if (ret && *qgroup_reserved)
5628 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5633 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5634 struct btrfs_block_rsv *rsv,
5635 u64 qgroup_reserved)
5637 btrfs_block_rsv_release(root, rsv, (u64)-1);
5641 * drop_outstanding_extent - drop an outstanding extent
5642 * @inode: the inode we're dropping the extent for
5643 * @num_bytes: the number of bytes we're releasing.
5645 * This is called when we are freeing up an outstanding extent, either called
5646 * after an error or after an extent is written. This will return the number of
5647 * reserved extents that need to be freed. This must be called with
5648 * BTRFS_I(inode)->lock held.
5650 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5652 unsigned drop_inode_space = 0;
5653 unsigned dropped_extents = 0;
5654 unsigned num_extents = 0;
5656 num_extents = (unsigned)div64_u64(num_bytes +
5657 BTRFS_MAX_EXTENT_SIZE - 1,
5658 BTRFS_MAX_EXTENT_SIZE);
5659 ASSERT(num_extents);
5660 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5661 BTRFS_I(inode)->outstanding_extents -= num_extents;
5663 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5664 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5665 &BTRFS_I(inode)->runtime_flags))
5666 drop_inode_space = 1;
5669 * If we have more or the same amount of outstanding extents than we have
5670 * reserved then we need to leave the reserved extents count alone.
5672 if (BTRFS_I(inode)->outstanding_extents >=
5673 BTRFS_I(inode)->reserved_extents)
5674 return drop_inode_space;
5676 dropped_extents = BTRFS_I(inode)->reserved_extents -
5677 BTRFS_I(inode)->outstanding_extents;
5678 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5679 return dropped_extents + drop_inode_space;
5683 * calc_csum_metadata_size - return the amount of metadata space that must be
5684 * reserved/freed for the given bytes.
5685 * @inode: the inode we're manipulating
5686 * @num_bytes: the number of bytes in question
5687 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5689 * This adjusts the number of csum_bytes in the inode and then returns the
5690 * correct amount of metadata that must either be reserved or freed. We
5691 * calculate how many checksums we can fit into one leaf and then divide the
5692 * number of bytes that will need to be checksumed by this value to figure out
5693 * how many checksums will be required. If we are adding bytes then the number
5694 * may go up and we will return the number of additional bytes that must be
5695 * reserved. If it is going down we will return the number of bytes that must
5698 * This must be called with BTRFS_I(inode)->lock held.
5700 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5703 struct btrfs_root *root = BTRFS_I(inode)->root;
5704 u64 old_csums, num_csums;
5706 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5707 BTRFS_I(inode)->csum_bytes == 0)
5710 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5712 BTRFS_I(inode)->csum_bytes += num_bytes;
5714 BTRFS_I(inode)->csum_bytes -= num_bytes;
5715 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5717 /* No change, no need to reserve more */
5718 if (old_csums == num_csums)
5722 return btrfs_calc_trans_metadata_size(root,
5723 num_csums - old_csums);
5725 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5728 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5730 struct btrfs_root *root = BTRFS_I(inode)->root;
5731 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5734 unsigned nr_extents = 0;
5735 int extra_reserve = 0;
5736 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5738 bool delalloc_lock = true;
5742 /* If we are a free space inode we need to not flush since we will be in
5743 * the middle of a transaction commit. We also don't need the delalloc
5744 * mutex since we won't race with anybody. We need this mostly to make
5745 * lockdep shut its filthy mouth.
5747 if (btrfs_is_free_space_inode(inode)) {
5748 flush = BTRFS_RESERVE_NO_FLUSH;
5749 delalloc_lock = false;
5752 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5753 btrfs_transaction_in_commit(root->fs_info))
5754 schedule_timeout(1);
5757 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5759 num_bytes = ALIGN(num_bytes, root->sectorsize);
5761 spin_lock(&BTRFS_I(inode)->lock);
5762 nr_extents = (unsigned)div64_u64(num_bytes +
5763 BTRFS_MAX_EXTENT_SIZE - 1,
5764 BTRFS_MAX_EXTENT_SIZE);
5765 BTRFS_I(inode)->outstanding_extents += nr_extents;
5768 if (BTRFS_I(inode)->outstanding_extents >
5769 BTRFS_I(inode)->reserved_extents)
5770 nr_extents = BTRFS_I(inode)->outstanding_extents -
5771 BTRFS_I(inode)->reserved_extents;
5774 * Add an item to reserve for updating the inode when we complete the
5777 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5778 &BTRFS_I(inode)->runtime_flags)) {
5783 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5784 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5785 csum_bytes = BTRFS_I(inode)->csum_bytes;
5786 spin_unlock(&BTRFS_I(inode)->lock);
5788 if (root->fs_info->quota_enabled) {
5789 ret = btrfs_qgroup_reserve_meta(root,
5790 nr_extents * root->nodesize);
5795 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5796 if (unlikely(ret)) {
5797 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5801 spin_lock(&BTRFS_I(inode)->lock);
5802 if (extra_reserve) {
5803 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5804 &BTRFS_I(inode)->runtime_flags);
5807 BTRFS_I(inode)->reserved_extents += nr_extents;
5808 spin_unlock(&BTRFS_I(inode)->lock);
5811 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5814 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5815 btrfs_ino(inode), to_reserve, 1);
5816 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5821 spin_lock(&BTRFS_I(inode)->lock);
5822 dropped = drop_outstanding_extent(inode, num_bytes);
5824 * If the inodes csum_bytes is the same as the original
5825 * csum_bytes then we know we haven't raced with any free()ers
5826 * so we can just reduce our inodes csum bytes and carry on.
5828 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5829 calc_csum_metadata_size(inode, num_bytes, 0);
5831 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5835 * This is tricky, but first we need to figure out how much we
5836 * freed from any free-ers that occurred during this
5837 * reservation, so we reset ->csum_bytes to the csum_bytes
5838 * before we dropped our lock, and then call the free for the
5839 * number of bytes that were freed while we were trying our
5842 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5843 BTRFS_I(inode)->csum_bytes = csum_bytes;
5844 to_free = calc_csum_metadata_size(inode, bytes, 0);
5848 * Now we need to see how much we would have freed had we not
5849 * been making this reservation and our ->csum_bytes were not
5850 * artificially inflated.
5852 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5853 bytes = csum_bytes - orig_csum_bytes;
5854 bytes = calc_csum_metadata_size(inode, bytes, 0);
5857 * Now reset ->csum_bytes to what it should be. If bytes is
5858 * more than to_free then we would have freed more space had we
5859 * not had an artificially high ->csum_bytes, so we need to free
5860 * the remainder. If bytes is the same or less then we don't
5861 * need to do anything, the other free-ers did the correct
5864 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5865 if (bytes > to_free)
5866 to_free = bytes - to_free;
5870 spin_unlock(&BTRFS_I(inode)->lock);
5872 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5875 btrfs_block_rsv_release(root, block_rsv, to_free);
5876 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5877 btrfs_ino(inode), to_free, 0);
5880 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5885 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5886 * @inode: the inode to release the reservation for
5887 * @num_bytes: the number of bytes we're releasing
5889 * This will release the metadata reservation for an inode. This can be called
5890 * once we complete IO for a given set of bytes to release their metadata
5893 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5895 struct btrfs_root *root = BTRFS_I(inode)->root;
5899 num_bytes = ALIGN(num_bytes, root->sectorsize);
5900 spin_lock(&BTRFS_I(inode)->lock);
5901 dropped = drop_outstanding_extent(inode, num_bytes);
5904 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5905 spin_unlock(&BTRFS_I(inode)->lock);
5907 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5909 if (btrfs_test_is_dummy_root(root))
5912 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5913 btrfs_ino(inode), to_free, 0);
5915 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5920 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5922 * @inode: inode we're writing to
5923 * @start: start range we are writing to
5924 * @len: how long the range we are writing to
5926 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5928 * This will do the following things
5930 * o reserve space in data space info for num bytes
5931 * and reserve precious corresponding qgroup space
5932 * (Done in check_data_free_space)
5934 * o reserve space for metadata space, based on the number of outstanding
5935 * extents and how much csums will be needed
5936 * also reserve metadata space in a per root over-reserve method.
5937 * o add to the inodes->delalloc_bytes
5938 * o add it to the fs_info's delalloc inodes list.
5939 * (Above 3 all done in delalloc_reserve_metadata)
5941 * Return 0 for success
5942 * Return <0 for error(-ENOSPC or -EQUOT)
5944 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5948 ret = btrfs_check_data_free_space(inode, start, len);
5951 ret = btrfs_delalloc_reserve_metadata(inode, len);
5953 btrfs_free_reserved_data_space(inode, start, len);
5958 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5959 * @inode: inode we're releasing space for
5960 * @start: start position of the space already reserved
5961 * @len: the len of the space already reserved
5963 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5964 * called in the case that we don't need the metadata AND data reservations
5965 * anymore. So if there is an error or we insert an inline extent.
5967 * This function will release the metadata space that was not used and will
5968 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5969 * list if there are no delalloc bytes left.
5970 * Also it will handle the qgroup reserved space.
5972 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5974 btrfs_delalloc_release_metadata(inode, len);
5975 btrfs_free_reserved_data_space(inode, start, len);
5978 static int update_block_group(struct btrfs_trans_handle *trans,
5979 struct btrfs_root *root, u64 bytenr,
5980 u64 num_bytes, int alloc)
5982 struct btrfs_block_group_cache *cache = NULL;
5983 struct btrfs_fs_info *info = root->fs_info;
5984 u64 total = num_bytes;
5989 /* block accounting for super block */
5990 spin_lock(&info->delalloc_root_lock);
5991 old_val = btrfs_super_bytes_used(info->super_copy);
5993 old_val += num_bytes;
5995 old_val -= num_bytes;
5996 btrfs_set_super_bytes_used(info->super_copy, old_val);
5997 spin_unlock(&info->delalloc_root_lock);
6000 cache = btrfs_lookup_block_group(info, bytenr);
6003 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6004 BTRFS_BLOCK_GROUP_RAID1 |
6005 BTRFS_BLOCK_GROUP_RAID10))
6010 * If this block group has free space cache written out, we
6011 * need to make sure to load it if we are removing space. This
6012 * is because we need the unpinning stage to actually add the
6013 * space back to the block group, otherwise we will leak space.
6015 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6016 cache_block_group(cache, 1);
6018 byte_in_group = bytenr - cache->key.objectid;
6019 WARN_ON(byte_in_group > cache->key.offset);
6021 spin_lock(&cache->space_info->lock);
6022 spin_lock(&cache->lock);
6024 if (btrfs_test_opt(root, SPACE_CACHE) &&
6025 cache->disk_cache_state < BTRFS_DC_CLEAR)
6026 cache->disk_cache_state = BTRFS_DC_CLEAR;
6028 old_val = btrfs_block_group_used(&cache->item);
6029 num_bytes = min(total, cache->key.offset - byte_in_group);
6031 old_val += num_bytes;
6032 btrfs_set_block_group_used(&cache->item, old_val);
6033 cache->reserved -= num_bytes;
6034 cache->space_info->bytes_reserved -= num_bytes;
6035 cache->space_info->bytes_used += num_bytes;
6036 cache->space_info->disk_used += num_bytes * factor;
6037 spin_unlock(&cache->lock);
6038 spin_unlock(&cache->space_info->lock);
6040 old_val -= num_bytes;
6041 btrfs_set_block_group_used(&cache->item, old_val);
6042 cache->pinned += num_bytes;
6043 cache->space_info->bytes_pinned += num_bytes;
6044 cache->space_info->bytes_used -= num_bytes;
6045 cache->space_info->disk_used -= num_bytes * factor;
6046 spin_unlock(&cache->lock);
6047 spin_unlock(&cache->space_info->lock);
6049 set_extent_dirty(info->pinned_extents,
6050 bytenr, bytenr + num_bytes - 1,
6051 GFP_NOFS | __GFP_NOFAIL);
6054 spin_lock(&trans->transaction->dirty_bgs_lock);
6055 if (list_empty(&cache->dirty_list)) {
6056 list_add_tail(&cache->dirty_list,
6057 &trans->transaction->dirty_bgs);
6058 trans->transaction->num_dirty_bgs++;
6059 btrfs_get_block_group(cache);
6061 spin_unlock(&trans->transaction->dirty_bgs_lock);
6064 * No longer have used bytes in this block group, queue it for
6065 * deletion. We do this after adding the block group to the
6066 * dirty list to avoid races between cleaner kthread and space
6069 if (!alloc && old_val == 0) {
6070 spin_lock(&info->unused_bgs_lock);
6071 if (list_empty(&cache->bg_list)) {
6072 btrfs_get_block_group(cache);
6073 list_add_tail(&cache->bg_list,
6076 spin_unlock(&info->unused_bgs_lock);
6079 btrfs_put_block_group(cache);
6081 bytenr += num_bytes;
6086 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6088 struct btrfs_block_group_cache *cache;
6091 spin_lock(&root->fs_info->block_group_cache_lock);
6092 bytenr = root->fs_info->first_logical_byte;
6093 spin_unlock(&root->fs_info->block_group_cache_lock);
6095 if (bytenr < (u64)-1)
6098 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6102 bytenr = cache->key.objectid;
6103 btrfs_put_block_group(cache);
6108 static int pin_down_extent(struct btrfs_root *root,
6109 struct btrfs_block_group_cache *cache,
6110 u64 bytenr, u64 num_bytes, int reserved)
6112 spin_lock(&cache->space_info->lock);
6113 spin_lock(&cache->lock);
6114 cache->pinned += num_bytes;
6115 cache->space_info->bytes_pinned += num_bytes;
6117 cache->reserved -= num_bytes;
6118 cache->space_info->bytes_reserved -= num_bytes;
6120 spin_unlock(&cache->lock);
6121 spin_unlock(&cache->space_info->lock);
6123 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6124 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6126 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
6131 * this function must be called within transaction
6133 int btrfs_pin_extent(struct btrfs_root *root,
6134 u64 bytenr, u64 num_bytes, int reserved)
6136 struct btrfs_block_group_cache *cache;
6138 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6139 BUG_ON(!cache); /* Logic error */
6141 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6143 btrfs_put_block_group(cache);
6148 * this function must be called within transaction
6150 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6151 u64 bytenr, u64 num_bytes)
6153 struct btrfs_block_group_cache *cache;
6156 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6161 * pull in the free space cache (if any) so that our pin
6162 * removes the free space from the cache. We have load_only set
6163 * to one because the slow code to read in the free extents does check
6164 * the pinned extents.
6166 cache_block_group(cache, 1);
6168 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6170 /* remove us from the free space cache (if we're there at all) */
6171 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6172 btrfs_put_block_group(cache);
6176 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6179 struct btrfs_block_group_cache *block_group;
6180 struct btrfs_caching_control *caching_ctl;
6182 block_group = btrfs_lookup_block_group(root->fs_info, start);
6186 cache_block_group(block_group, 0);
6187 caching_ctl = get_caching_control(block_group);
6191 BUG_ON(!block_group_cache_done(block_group));
6192 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6194 mutex_lock(&caching_ctl->mutex);
6196 if (start >= caching_ctl->progress) {
6197 ret = add_excluded_extent(root, start, num_bytes);
6198 } else if (start + num_bytes <= caching_ctl->progress) {
6199 ret = btrfs_remove_free_space(block_group,
6202 num_bytes = caching_ctl->progress - start;
6203 ret = btrfs_remove_free_space(block_group,
6208 num_bytes = (start + num_bytes) -
6209 caching_ctl->progress;
6210 start = caching_ctl->progress;
6211 ret = add_excluded_extent(root, start, num_bytes);
6214 mutex_unlock(&caching_ctl->mutex);
6215 put_caching_control(caching_ctl);
6217 btrfs_put_block_group(block_group);
6221 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6222 struct extent_buffer *eb)
6224 struct btrfs_file_extent_item *item;
6225 struct btrfs_key key;
6229 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6232 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6233 btrfs_item_key_to_cpu(eb, &key, i);
6234 if (key.type != BTRFS_EXTENT_DATA_KEY)
6236 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6237 found_type = btrfs_file_extent_type(eb, item);
6238 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6240 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6242 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6243 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6244 __exclude_logged_extent(log, key.objectid, key.offset);
6251 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6253 atomic_inc(&bg->reservations);
6256 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6259 struct btrfs_block_group_cache *bg;
6261 bg = btrfs_lookup_block_group(fs_info, start);
6263 if (atomic_dec_and_test(&bg->reservations))
6264 wake_up_atomic_t(&bg->reservations);
6265 btrfs_put_block_group(bg);
6268 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6274 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6276 struct btrfs_space_info *space_info = bg->space_info;
6280 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6284 * Our block group is read only but before we set it to read only,
6285 * some task might have had allocated an extent from it already, but it
6286 * has not yet created a respective ordered extent (and added it to a
6287 * root's list of ordered extents).
6288 * Therefore wait for any task currently allocating extents, since the
6289 * block group's reservations counter is incremented while a read lock
6290 * on the groups' semaphore is held and decremented after releasing
6291 * the read access on that semaphore and creating the ordered extent.
6293 down_write(&space_info->groups_sem);
6294 up_write(&space_info->groups_sem);
6296 wait_on_atomic_t(&bg->reservations,
6297 btrfs_wait_bg_reservations_atomic_t,
6298 TASK_UNINTERRUPTIBLE);
6302 * btrfs_update_reserved_bytes - update the block_group and space info counters
6303 * @cache: The cache we are manipulating
6304 * @num_bytes: The number of bytes in question
6305 * @reserve: One of the reservation enums
6306 * @delalloc: The blocks are allocated for the delalloc write
6308 * This is called by the allocator when it reserves space, or by somebody who is
6309 * freeing space that was never actually used on disk. For example if you
6310 * reserve some space for a new leaf in transaction A and before transaction A
6311 * commits you free that leaf, you call this with reserve set to 0 in order to
6312 * clear the reservation.
6314 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6315 * ENOSPC accounting. For data we handle the reservation through clearing the
6316 * delalloc bits in the io_tree. We have to do this since we could end up
6317 * allocating less disk space for the amount of data we have reserved in the
6318 * case of compression.
6320 * If this is a reservation and the block group has become read only we cannot
6321 * make the reservation and return -EAGAIN, otherwise this function always
6324 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6325 u64 num_bytes, int reserve, int delalloc)
6327 struct btrfs_space_info *space_info = cache->space_info;
6330 spin_lock(&space_info->lock);
6331 spin_lock(&cache->lock);
6332 if (reserve != RESERVE_FREE) {
6336 cache->reserved += num_bytes;
6337 space_info->bytes_reserved += num_bytes;
6338 if (reserve == RESERVE_ALLOC) {
6339 trace_btrfs_space_reservation(cache->fs_info,
6340 "space_info", space_info->flags,
6342 space_info->bytes_may_use -= num_bytes;
6346 cache->delalloc_bytes += num_bytes;
6350 space_info->bytes_readonly += num_bytes;
6351 cache->reserved -= num_bytes;
6352 space_info->bytes_reserved -= num_bytes;
6355 cache->delalloc_bytes -= num_bytes;
6357 spin_unlock(&cache->lock);
6358 spin_unlock(&space_info->lock);
6362 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6363 struct btrfs_root *root)
6365 struct btrfs_fs_info *fs_info = root->fs_info;
6366 struct btrfs_caching_control *next;
6367 struct btrfs_caching_control *caching_ctl;
6368 struct btrfs_block_group_cache *cache;
6370 down_write(&fs_info->commit_root_sem);
6372 list_for_each_entry_safe(caching_ctl, next,
6373 &fs_info->caching_block_groups, list) {
6374 cache = caching_ctl->block_group;
6375 if (block_group_cache_done(cache)) {
6376 cache->last_byte_to_unpin = (u64)-1;
6377 list_del_init(&caching_ctl->list);
6378 put_caching_control(caching_ctl);
6380 cache->last_byte_to_unpin = caching_ctl->progress;
6384 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6385 fs_info->pinned_extents = &fs_info->freed_extents[1];
6387 fs_info->pinned_extents = &fs_info->freed_extents[0];
6389 up_write(&fs_info->commit_root_sem);
6391 update_global_block_rsv(fs_info);
6395 * Returns the free cluster for the given space info and sets empty_cluster to
6396 * what it should be based on the mount options.
6398 static struct btrfs_free_cluster *
6399 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6402 struct btrfs_free_cluster *ret = NULL;
6403 bool ssd = btrfs_test_opt(root, SSD);
6406 if (btrfs_mixed_space_info(space_info))
6410 *empty_cluster = SZ_2M;
6411 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6412 ret = &root->fs_info->meta_alloc_cluster;
6414 *empty_cluster = SZ_64K;
6415 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6416 ret = &root->fs_info->data_alloc_cluster;
6422 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6423 const bool return_free_space)
6425 struct btrfs_fs_info *fs_info = root->fs_info;
6426 struct btrfs_block_group_cache *cache = NULL;
6427 struct btrfs_space_info *space_info;
6428 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6429 struct btrfs_free_cluster *cluster = NULL;
6431 u64 total_unpinned = 0;
6432 u64 empty_cluster = 0;
6435 while (start <= end) {
6438 start >= cache->key.objectid + cache->key.offset) {
6440 btrfs_put_block_group(cache);
6442 cache = btrfs_lookup_block_group(fs_info, start);
6443 BUG_ON(!cache); /* Logic error */
6445 cluster = fetch_cluster_info(root,
6448 empty_cluster <<= 1;
6451 len = cache->key.objectid + cache->key.offset - start;
6452 len = min(len, end + 1 - start);
6454 if (start < cache->last_byte_to_unpin) {
6455 len = min(len, cache->last_byte_to_unpin - start);
6456 if (return_free_space)
6457 btrfs_add_free_space(cache, start, len);
6461 total_unpinned += len;
6462 space_info = cache->space_info;
6465 * If this space cluster has been marked as fragmented and we've
6466 * unpinned enough in this block group to potentially allow a
6467 * cluster to be created inside of it go ahead and clear the
6470 if (cluster && cluster->fragmented &&
6471 total_unpinned > empty_cluster) {
6472 spin_lock(&cluster->lock);
6473 cluster->fragmented = 0;
6474 spin_unlock(&cluster->lock);
6477 spin_lock(&space_info->lock);
6478 spin_lock(&cache->lock);
6479 cache->pinned -= len;
6480 space_info->bytes_pinned -= len;
6481 space_info->max_extent_size = 0;
6482 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6484 space_info->bytes_readonly += len;
6487 spin_unlock(&cache->lock);
6488 if (!readonly && global_rsv->space_info == space_info) {
6489 spin_lock(&global_rsv->lock);
6490 if (!global_rsv->full) {
6491 len = min(len, global_rsv->size -
6492 global_rsv->reserved);
6493 global_rsv->reserved += len;
6494 space_info->bytes_may_use += len;
6495 if (global_rsv->reserved >= global_rsv->size)
6496 global_rsv->full = 1;
6498 spin_unlock(&global_rsv->lock);
6500 spin_unlock(&space_info->lock);
6504 btrfs_put_block_group(cache);
6508 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6509 struct btrfs_root *root)
6511 struct btrfs_fs_info *fs_info = root->fs_info;
6512 struct btrfs_block_group_cache *block_group, *tmp;
6513 struct list_head *deleted_bgs;
6514 struct extent_io_tree *unpin;
6519 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6520 unpin = &fs_info->freed_extents[1];
6522 unpin = &fs_info->freed_extents[0];
6524 while (!trans->aborted) {
6525 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6526 ret = find_first_extent_bit(unpin, 0, &start, &end,
6527 EXTENT_DIRTY, NULL);
6529 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6533 if (btrfs_test_opt(root, DISCARD))
6534 ret = btrfs_discard_extent(root, start,
6535 end + 1 - start, NULL);
6537 clear_extent_dirty(unpin, start, end);
6538 unpin_extent_range(root, start, end, true);
6539 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6544 * Transaction is finished. We don't need the lock anymore. We
6545 * do need to clean up the block groups in case of a transaction
6548 deleted_bgs = &trans->transaction->deleted_bgs;
6549 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6553 if (!trans->aborted)
6554 ret = btrfs_discard_extent(root,
6555 block_group->key.objectid,
6556 block_group->key.offset,
6559 list_del_init(&block_group->bg_list);
6560 btrfs_put_block_group_trimming(block_group);
6561 btrfs_put_block_group(block_group);
6564 const char *errstr = btrfs_decode_error(ret);
6566 "Discard failed while removing blockgroup: errno=%d %s\n",
6574 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6575 u64 owner, u64 root_objectid)
6577 struct btrfs_space_info *space_info;
6580 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6581 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6582 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6584 flags = BTRFS_BLOCK_GROUP_METADATA;
6586 flags = BTRFS_BLOCK_GROUP_DATA;
6589 space_info = __find_space_info(fs_info, flags);
6590 BUG_ON(!space_info); /* Logic bug */
6591 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6595 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6596 struct btrfs_root *root,
6597 struct btrfs_delayed_ref_node *node, u64 parent,
6598 u64 root_objectid, u64 owner_objectid,
6599 u64 owner_offset, int refs_to_drop,
6600 struct btrfs_delayed_extent_op *extent_op)
6602 struct btrfs_key key;
6603 struct btrfs_path *path;
6604 struct btrfs_fs_info *info = root->fs_info;
6605 struct btrfs_root *extent_root = info->extent_root;
6606 struct extent_buffer *leaf;
6607 struct btrfs_extent_item *ei;
6608 struct btrfs_extent_inline_ref *iref;
6611 int extent_slot = 0;
6612 int found_extent = 0;
6616 u64 bytenr = node->bytenr;
6617 u64 num_bytes = node->num_bytes;
6619 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6622 path = btrfs_alloc_path();
6626 path->reada = READA_FORWARD;
6627 path->leave_spinning = 1;
6629 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6630 BUG_ON(!is_data && refs_to_drop != 1);
6633 skinny_metadata = 0;
6635 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6636 bytenr, num_bytes, parent,
6637 root_objectid, owner_objectid,
6640 extent_slot = path->slots[0];
6641 while (extent_slot >= 0) {
6642 btrfs_item_key_to_cpu(path->nodes[0], &key,
6644 if (key.objectid != bytenr)
6646 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6647 key.offset == num_bytes) {
6651 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6652 key.offset == owner_objectid) {
6656 if (path->slots[0] - extent_slot > 5)
6660 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6661 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6662 if (found_extent && item_size < sizeof(*ei))
6665 if (!found_extent) {
6667 ret = remove_extent_backref(trans, extent_root, path,
6669 is_data, &last_ref);
6671 btrfs_abort_transaction(trans, extent_root, ret);
6674 btrfs_release_path(path);
6675 path->leave_spinning = 1;
6677 key.objectid = bytenr;
6678 key.type = BTRFS_EXTENT_ITEM_KEY;
6679 key.offset = num_bytes;
6681 if (!is_data && skinny_metadata) {
6682 key.type = BTRFS_METADATA_ITEM_KEY;
6683 key.offset = owner_objectid;
6686 ret = btrfs_search_slot(trans, extent_root,
6688 if (ret > 0 && skinny_metadata && path->slots[0]) {
6690 * Couldn't find our skinny metadata item,
6691 * see if we have ye olde extent item.
6694 btrfs_item_key_to_cpu(path->nodes[0], &key,
6696 if (key.objectid == bytenr &&
6697 key.type == BTRFS_EXTENT_ITEM_KEY &&
6698 key.offset == num_bytes)
6702 if (ret > 0 && skinny_metadata) {
6703 skinny_metadata = false;
6704 key.objectid = bytenr;
6705 key.type = BTRFS_EXTENT_ITEM_KEY;
6706 key.offset = num_bytes;
6707 btrfs_release_path(path);
6708 ret = btrfs_search_slot(trans, extent_root,
6713 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6716 btrfs_print_leaf(extent_root,
6720 btrfs_abort_transaction(trans, extent_root, ret);
6723 extent_slot = path->slots[0];
6725 } else if (WARN_ON(ret == -ENOENT)) {
6726 btrfs_print_leaf(extent_root, path->nodes[0]);
6728 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6729 bytenr, parent, root_objectid, owner_objectid,
6731 btrfs_abort_transaction(trans, extent_root, ret);
6734 btrfs_abort_transaction(trans, extent_root, ret);
6738 leaf = path->nodes[0];
6739 item_size = btrfs_item_size_nr(leaf, extent_slot);
6740 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6741 if (item_size < sizeof(*ei)) {
6742 BUG_ON(found_extent || extent_slot != path->slots[0]);
6743 ret = convert_extent_item_v0(trans, extent_root, path,
6746 btrfs_abort_transaction(trans, extent_root, ret);
6750 btrfs_release_path(path);
6751 path->leave_spinning = 1;
6753 key.objectid = bytenr;
6754 key.type = BTRFS_EXTENT_ITEM_KEY;
6755 key.offset = num_bytes;
6757 ret = btrfs_search_slot(trans, extent_root, &key, path,
6760 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6762 btrfs_print_leaf(extent_root, path->nodes[0]);
6765 btrfs_abort_transaction(trans, extent_root, ret);
6769 extent_slot = path->slots[0];
6770 leaf = path->nodes[0];
6771 item_size = btrfs_item_size_nr(leaf, extent_slot);
6774 BUG_ON(item_size < sizeof(*ei));
6775 ei = btrfs_item_ptr(leaf, extent_slot,
6776 struct btrfs_extent_item);
6777 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6778 key.type == BTRFS_EXTENT_ITEM_KEY) {
6779 struct btrfs_tree_block_info *bi;
6780 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6781 bi = (struct btrfs_tree_block_info *)(ei + 1);
6782 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6785 refs = btrfs_extent_refs(leaf, ei);
6786 if (refs < refs_to_drop) {
6787 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6788 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6790 btrfs_abort_transaction(trans, extent_root, ret);
6793 refs -= refs_to_drop;
6797 __run_delayed_extent_op(extent_op, leaf, ei);
6799 * In the case of inline back ref, reference count will
6800 * be updated by remove_extent_backref
6803 BUG_ON(!found_extent);
6805 btrfs_set_extent_refs(leaf, ei, refs);
6806 btrfs_mark_buffer_dirty(leaf);
6809 ret = remove_extent_backref(trans, extent_root, path,
6811 is_data, &last_ref);
6813 btrfs_abort_transaction(trans, extent_root, ret);
6817 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6821 BUG_ON(is_data && refs_to_drop !=
6822 extent_data_ref_count(path, iref));
6824 BUG_ON(path->slots[0] != extent_slot);
6826 BUG_ON(path->slots[0] != extent_slot + 1);
6827 path->slots[0] = extent_slot;
6833 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6836 btrfs_abort_transaction(trans, extent_root, ret);
6839 btrfs_release_path(path);
6842 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6844 btrfs_abort_transaction(trans, extent_root, ret);
6849 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
6852 btrfs_abort_transaction(trans, extent_root, ret);
6856 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6858 btrfs_abort_transaction(trans, extent_root, ret);
6862 btrfs_release_path(path);
6865 btrfs_free_path(path);
6870 * when we free an block, it is possible (and likely) that we free the last
6871 * delayed ref for that extent as well. This searches the delayed ref tree for
6872 * a given extent, and if there are no other delayed refs to be processed, it
6873 * removes it from the tree.
6875 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6876 struct btrfs_root *root, u64 bytenr)
6878 struct btrfs_delayed_ref_head *head;
6879 struct btrfs_delayed_ref_root *delayed_refs;
6882 delayed_refs = &trans->transaction->delayed_refs;
6883 spin_lock(&delayed_refs->lock);
6884 head = btrfs_find_delayed_ref_head(trans, bytenr);
6886 goto out_delayed_unlock;
6888 spin_lock(&head->lock);
6889 if (!list_empty(&head->ref_list))
6892 if (head->extent_op) {
6893 if (!head->must_insert_reserved)
6895 btrfs_free_delayed_extent_op(head->extent_op);
6896 head->extent_op = NULL;
6900 * waiting for the lock here would deadlock. If someone else has it
6901 * locked they are already in the process of dropping it anyway
6903 if (!mutex_trylock(&head->mutex))
6907 * at this point we have a head with no other entries. Go
6908 * ahead and process it.
6910 head->node.in_tree = 0;
6911 rb_erase(&head->href_node, &delayed_refs->href_root);
6913 atomic_dec(&delayed_refs->num_entries);
6916 * we don't take a ref on the node because we're removing it from the
6917 * tree, so we just steal the ref the tree was holding.
6919 delayed_refs->num_heads--;
6920 if (head->processing == 0)
6921 delayed_refs->num_heads_ready--;
6922 head->processing = 0;
6923 spin_unlock(&head->lock);
6924 spin_unlock(&delayed_refs->lock);
6926 BUG_ON(head->extent_op);
6927 if (head->must_insert_reserved)
6930 mutex_unlock(&head->mutex);
6931 btrfs_put_delayed_ref(&head->node);
6934 spin_unlock(&head->lock);
6937 spin_unlock(&delayed_refs->lock);
6941 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6942 struct btrfs_root *root,
6943 struct extent_buffer *buf,
6944 u64 parent, int last_ref)
6949 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6950 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6951 buf->start, buf->len,
6952 parent, root->root_key.objectid,
6953 btrfs_header_level(buf),
6954 BTRFS_DROP_DELAYED_REF, NULL);
6955 BUG_ON(ret); /* -ENOMEM */
6961 if (btrfs_header_generation(buf) == trans->transid) {
6962 struct btrfs_block_group_cache *cache;
6964 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6965 ret = check_ref_cleanup(trans, root, buf->start);
6970 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6972 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6973 pin_down_extent(root, cache, buf->start, buf->len, 1);
6974 btrfs_put_block_group(cache);
6978 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6980 btrfs_add_free_space(cache, buf->start, buf->len);
6981 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6982 btrfs_put_block_group(cache);
6983 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6988 add_pinned_bytes(root->fs_info, buf->len,
6989 btrfs_header_level(buf),
6990 root->root_key.objectid);
6993 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6996 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6999 /* Can return -ENOMEM */
7000 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7001 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7002 u64 owner, u64 offset)
7005 struct btrfs_fs_info *fs_info = root->fs_info;
7007 if (btrfs_test_is_dummy_root(root))
7010 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
7013 * tree log blocks never actually go into the extent allocation
7014 * tree, just update pinning info and exit early.
7016 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7017 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7018 /* unlocks the pinned mutex */
7019 btrfs_pin_extent(root, bytenr, num_bytes, 1);
7021 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7022 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7024 parent, root_objectid, (int)owner,
7025 BTRFS_DROP_DELAYED_REF, NULL);
7027 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7029 parent, root_objectid, owner,
7031 BTRFS_DROP_DELAYED_REF, NULL);
7037 * when we wait for progress in the block group caching, its because
7038 * our allocation attempt failed at least once. So, we must sleep
7039 * and let some progress happen before we try again.
7041 * This function will sleep at least once waiting for new free space to
7042 * show up, and then it will check the block group free space numbers
7043 * for our min num_bytes. Another option is to have it go ahead
7044 * and look in the rbtree for a free extent of a given size, but this
7047 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7048 * any of the information in this block group.
7050 static noinline void
7051 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7054 struct btrfs_caching_control *caching_ctl;
7056 caching_ctl = get_caching_control(cache);
7060 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7061 (cache->free_space_ctl->free_space >= num_bytes));
7063 put_caching_control(caching_ctl);
7067 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7069 struct btrfs_caching_control *caching_ctl;
7072 caching_ctl = get_caching_control(cache);
7074 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7076 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7077 if (cache->cached == BTRFS_CACHE_ERROR)
7079 put_caching_control(caching_ctl);
7083 int __get_raid_index(u64 flags)
7085 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7086 return BTRFS_RAID_RAID10;
7087 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7088 return BTRFS_RAID_RAID1;
7089 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7090 return BTRFS_RAID_DUP;
7091 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7092 return BTRFS_RAID_RAID0;
7093 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7094 return BTRFS_RAID_RAID5;
7095 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7096 return BTRFS_RAID_RAID6;
7098 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7101 int get_block_group_index(struct btrfs_block_group_cache *cache)
7103 return __get_raid_index(cache->flags);
7106 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7107 [BTRFS_RAID_RAID10] = "raid10",
7108 [BTRFS_RAID_RAID1] = "raid1",
7109 [BTRFS_RAID_DUP] = "dup",
7110 [BTRFS_RAID_RAID0] = "raid0",
7111 [BTRFS_RAID_SINGLE] = "single",
7112 [BTRFS_RAID_RAID5] = "raid5",
7113 [BTRFS_RAID_RAID6] = "raid6",
7116 static const char *get_raid_name(enum btrfs_raid_types type)
7118 if (type >= BTRFS_NR_RAID_TYPES)
7121 return btrfs_raid_type_names[type];
7124 enum btrfs_loop_type {
7125 LOOP_CACHING_NOWAIT = 0,
7126 LOOP_CACHING_WAIT = 1,
7127 LOOP_ALLOC_CHUNK = 2,
7128 LOOP_NO_EMPTY_SIZE = 3,
7132 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7136 down_read(&cache->data_rwsem);
7140 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7143 btrfs_get_block_group(cache);
7145 down_read(&cache->data_rwsem);
7148 static struct btrfs_block_group_cache *
7149 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7150 struct btrfs_free_cluster *cluster,
7153 struct btrfs_block_group_cache *used_bg = NULL;
7155 spin_lock(&cluster->refill_lock);
7157 used_bg = cluster->block_group;
7161 if (used_bg == block_group)
7164 btrfs_get_block_group(used_bg);
7169 if (down_read_trylock(&used_bg->data_rwsem))
7172 spin_unlock(&cluster->refill_lock);
7174 down_read(&used_bg->data_rwsem);
7176 spin_lock(&cluster->refill_lock);
7177 if (used_bg == cluster->block_group)
7180 up_read(&used_bg->data_rwsem);
7181 btrfs_put_block_group(used_bg);
7186 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7190 up_read(&cache->data_rwsem);
7191 btrfs_put_block_group(cache);
7195 * walks the btree of allocated extents and find a hole of a given size.
7196 * The key ins is changed to record the hole:
7197 * ins->objectid == start position
7198 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7199 * ins->offset == the size of the hole.
7200 * Any available blocks before search_start are skipped.
7202 * If there is no suitable free space, we will record the max size of
7203 * the free space extent currently.
7205 static noinline int find_free_extent(struct btrfs_root *orig_root,
7206 u64 num_bytes, u64 empty_size,
7207 u64 hint_byte, struct btrfs_key *ins,
7208 u64 flags, int delalloc)
7211 struct btrfs_root *root = orig_root->fs_info->extent_root;
7212 struct btrfs_free_cluster *last_ptr = NULL;
7213 struct btrfs_block_group_cache *block_group = NULL;
7214 u64 search_start = 0;
7215 u64 max_extent_size = 0;
7216 u64 empty_cluster = 0;
7217 struct btrfs_space_info *space_info;
7219 int index = __get_raid_index(flags);
7220 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7221 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7222 bool failed_cluster_refill = false;
7223 bool failed_alloc = false;
7224 bool use_cluster = true;
7225 bool have_caching_bg = false;
7226 bool orig_have_caching_bg = false;
7227 bool full_search = false;
7229 WARN_ON(num_bytes < root->sectorsize);
7230 ins->type = BTRFS_EXTENT_ITEM_KEY;
7234 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7236 space_info = __find_space_info(root->fs_info, flags);
7238 btrfs_err(root->fs_info, "No space info for %llu", flags);
7243 * If our free space is heavily fragmented we may not be able to make
7244 * big contiguous allocations, so instead of doing the expensive search
7245 * for free space, simply return ENOSPC with our max_extent_size so we
7246 * can go ahead and search for a more manageable chunk.
7248 * If our max_extent_size is large enough for our allocation simply
7249 * disable clustering since we will likely not be able to find enough
7250 * space to create a cluster and induce latency trying.
7252 if (unlikely(space_info->max_extent_size)) {
7253 spin_lock(&space_info->lock);
7254 if (space_info->max_extent_size &&
7255 num_bytes > space_info->max_extent_size) {
7256 ins->offset = space_info->max_extent_size;
7257 spin_unlock(&space_info->lock);
7259 } else if (space_info->max_extent_size) {
7260 use_cluster = false;
7262 spin_unlock(&space_info->lock);
7265 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7267 spin_lock(&last_ptr->lock);
7268 if (last_ptr->block_group)
7269 hint_byte = last_ptr->window_start;
7270 if (last_ptr->fragmented) {
7272 * We still set window_start so we can keep track of the
7273 * last place we found an allocation to try and save
7276 hint_byte = last_ptr->window_start;
7277 use_cluster = false;
7279 spin_unlock(&last_ptr->lock);
7282 search_start = max(search_start, first_logical_byte(root, 0));
7283 search_start = max(search_start, hint_byte);
7284 if (search_start == hint_byte) {
7285 block_group = btrfs_lookup_block_group(root->fs_info,
7288 * we don't want to use the block group if it doesn't match our
7289 * allocation bits, or if its not cached.
7291 * However if we are re-searching with an ideal block group
7292 * picked out then we don't care that the block group is cached.
7294 if (block_group && block_group_bits(block_group, flags) &&
7295 block_group->cached != BTRFS_CACHE_NO) {
7296 down_read(&space_info->groups_sem);
7297 if (list_empty(&block_group->list) ||
7300 * someone is removing this block group,
7301 * we can't jump into the have_block_group
7302 * target because our list pointers are not
7305 btrfs_put_block_group(block_group);
7306 up_read(&space_info->groups_sem);
7308 index = get_block_group_index(block_group);
7309 btrfs_lock_block_group(block_group, delalloc);
7310 goto have_block_group;
7312 } else if (block_group) {
7313 btrfs_put_block_group(block_group);
7317 have_caching_bg = false;
7318 if (index == 0 || index == __get_raid_index(flags))
7320 down_read(&space_info->groups_sem);
7321 list_for_each_entry(block_group, &space_info->block_groups[index],
7326 btrfs_grab_block_group(block_group, delalloc);
7327 search_start = block_group->key.objectid;
7330 * this can happen if we end up cycling through all the
7331 * raid types, but we want to make sure we only allocate
7332 * for the proper type.
7334 if (!block_group_bits(block_group, flags)) {
7335 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7336 BTRFS_BLOCK_GROUP_RAID1 |
7337 BTRFS_BLOCK_GROUP_RAID5 |
7338 BTRFS_BLOCK_GROUP_RAID6 |
7339 BTRFS_BLOCK_GROUP_RAID10;
7342 * if they asked for extra copies and this block group
7343 * doesn't provide them, bail. This does allow us to
7344 * fill raid0 from raid1.
7346 if ((flags & extra) && !(block_group->flags & extra))
7351 cached = block_group_cache_done(block_group);
7352 if (unlikely(!cached)) {
7353 have_caching_bg = true;
7354 ret = cache_block_group(block_group, 0);
7359 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7361 if (unlikely(block_group->ro))
7365 * Ok we want to try and use the cluster allocator, so
7368 if (last_ptr && use_cluster) {
7369 struct btrfs_block_group_cache *used_block_group;
7370 unsigned long aligned_cluster;
7372 * the refill lock keeps out other
7373 * people trying to start a new cluster
7375 used_block_group = btrfs_lock_cluster(block_group,
7378 if (!used_block_group)
7379 goto refill_cluster;
7381 if (used_block_group != block_group &&
7382 (used_block_group->ro ||
7383 !block_group_bits(used_block_group, flags)))
7384 goto release_cluster;
7386 offset = btrfs_alloc_from_cluster(used_block_group,
7389 used_block_group->key.objectid,
7392 /* we have a block, we're done */
7393 spin_unlock(&last_ptr->refill_lock);
7394 trace_btrfs_reserve_extent_cluster(root,
7396 search_start, num_bytes);
7397 if (used_block_group != block_group) {
7398 btrfs_release_block_group(block_group,
7400 block_group = used_block_group;
7405 WARN_ON(last_ptr->block_group != used_block_group);
7407 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7408 * set up a new clusters, so lets just skip it
7409 * and let the allocator find whatever block
7410 * it can find. If we reach this point, we
7411 * will have tried the cluster allocator
7412 * plenty of times and not have found
7413 * anything, so we are likely way too
7414 * fragmented for the clustering stuff to find
7417 * However, if the cluster is taken from the
7418 * current block group, release the cluster
7419 * first, so that we stand a better chance of
7420 * succeeding in the unclustered
7422 if (loop >= LOOP_NO_EMPTY_SIZE &&
7423 used_block_group != block_group) {
7424 spin_unlock(&last_ptr->refill_lock);
7425 btrfs_release_block_group(used_block_group,
7427 goto unclustered_alloc;
7431 * this cluster didn't work out, free it and
7434 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7436 if (used_block_group != block_group)
7437 btrfs_release_block_group(used_block_group,
7440 if (loop >= LOOP_NO_EMPTY_SIZE) {
7441 spin_unlock(&last_ptr->refill_lock);
7442 goto unclustered_alloc;
7445 aligned_cluster = max_t(unsigned long,
7446 empty_cluster + empty_size,
7447 block_group->full_stripe_len);
7449 /* allocate a cluster in this block group */
7450 ret = btrfs_find_space_cluster(root, block_group,
7451 last_ptr, search_start,
7456 * now pull our allocation out of this
7459 offset = btrfs_alloc_from_cluster(block_group,
7465 /* we found one, proceed */
7466 spin_unlock(&last_ptr->refill_lock);
7467 trace_btrfs_reserve_extent_cluster(root,
7468 block_group, search_start,
7472 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7473 && !failed_cluster_refill) {
7474 spin_unlock(&last_ptr->refill_lock);
7476 failed_cluster_refill = true;
7477 wait_block_group_cache_progress(block_group,
7478 num_bytes + empty_cluster + empty_size);
7479 goto have_block_group;
7483 * at this point we either didn't find a cluster
7484 * or we weren't able to allocate a block from our
7485 * cluster. Free the cluster we've been trying
7486 * to use, and go to the next block group
7488 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7489 spin_unlock(&last_ptr->refill_lock);
7495 * We are doing an unclustered alloc, set the fragmented flag so
7496 * we don't bother trying to setup a cluster again until we get
7499 if (unlikely(last_ptr)) {
7500 spin_lock(&last_ptr->lock);
7501 last_ptr->fragmented = 1;
7502 spin_unlock(&last_ptr->lock);
7504 spin_lock(&block_group->free_space_ctl->tree_lock);
7506 block_group->free_space_ctl->free_space <
7507 num_bytes + empty_cluster + empty_size) {
7508 if (block_group->free_space_ctl->free_space >
7511 block_group->free_space_ctl->free_space;
7512 spin_unlock(&block_group->free_space_ctl->tree_lock);
7515 spin_unlock(&block_group->free_space_ctl->tree_lock);
7517 offset = btrfs_find_space_for_alloc(block_group, search_start,
7518 num_bytes, empty_size,
7521 * If we didn't find a chunk, and we haven't failed on this
7522 * block group before, and this block group is in the middle of
7523 * caching and we are ok with waiting, then go ahead and wait
7524 * for progress to be made, and set failed_alloc to true.
7526 * If failed_alloc is true then we've already waited on this
7527 * block group once and should move on to the next block group.
7529 if (!offset && !failed_alloc && !cached &&
7530 loop > LOOP_CACHING_NOWAIT) {
7531 wait_block_group_cache_progress(block_group,
7532 num_bytes + empty_size);
7533 failed_alloc = true;
7534 goto have_block_group;
7535 } else if (!offset) {
7539 search_start = ALIGN(offset, root->stripesize);
7541 /* move on to the next group */
7542 if (search_start + num_bytes >
7543 block_group->key.objectid + block_group->key.offset) {
7544 btrfs_add_free_space(block_group, offset, num_bytes);
7548 if (offset < search_start)
7549 btrfs_add_free_space(block_group, offset,
7550 search_start - offset);
7551 BUG_ON(offset > search_start);
7553 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7554 alloc_type, delalloc);
7555 if (ret == -EAGAIN) {
7556 btrfs_add_free_space(block_group, offset, num_bytes);
7559 btrfs_inc_block_group_reservations(block_group);
7561 /* we are all good, lets return */
7562 ins->objectid = search_start;
7563 ins->offset = num_bytes;
7565 trace_btrfs_reserve_extent(orig_root, block_group,
7566 search_start, num_bytes);
7567 btrfs_release_block_group(block_group, delalloc);
7570 failed_cluster_refill = false;
7571 failed_alloc = false;
7572 BUG_ON(index != get_block_group_index(block_group));
7573 btrfs_release_block_group(block_group, delalloc);
7575 up_read(&space_info->groups_sem);
7577 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7578 && !orig_have_caching_bg)
7579 orig_have_caching_bg = true;
7581 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7584 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7588 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7589 * caching kthreads as we move along
7590 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7591 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7592 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7595 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7597 if (loop == LOOP_CACHING_NOWAIT) {
7599 * We want to skip the LOOP_CACHING_WAIT step if we
7600 * don't have any uncached bgs and we've already done a
7601 * full search through.
7603 if (orig_have_caching_bg || !full_search)
7604 loop = LOOP_CACHING_WAIT;
7606 loop = LOOP_ALLOC_CHUNK;
7611 if (loop == LOOP_ALLOC_CHUNK) {
7612 struct btrfs_trans_handle *trans;
7615 trans = current->journal_info;
7619 trans = btrfs_join_transaction(root);
7621 if (IS_ERR(trans)) {
7622 ret = PTR_ERR(trans);
7626 ret = do_chunk_alloc(trans, root, flags,
7630 * If we can't allocate a new chunk we've already looped
7631 * through at least once, move on to the NO_EMPTY_SIZE
7635 loop = LOOP_NO_EMPTY_SIZE;
7638 * Do not bail out on ENOSPC since we
7639 * can do more things.
7641 if (ret < 0 && ret != -ENOSPC)
7642 btrfs_abort_transaction(trans,
7647 btrfs_end_transaction(trans, root);
7652 if (loop == LOOP_NO_EMPTY_SIZE) {
7654 * Don't loop again if we already have no empty_size and
7657 if (empty_size == 0 &&
7658 empty_cluster == 0) {
7667 } else if (!ins->objectid) {
7669 } else if (ins->objectid) {
7670 if (!use_cluster && last_ptr) {
7671 spin_lock(&last_ptr->lock);
7672 last_ptr->window_start = ins->objectid;
7673 spin_unlock(&last_ptr->lock);
7678 if (ret == -ENOSPC) {
7679 spin_lock(&space_info->lock);
7680 space_info->max_extent_size = max_extent_size;
7681 spin_unlock(&space_info->lock);
7682 ins->offset = max_extent_size;
7687 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7688 int dump_block_groups)
7690 struct btrfs_block_group_cache *cache;
7693 spin_lock(&info->lock);
7694 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7696 info->total_bytes - info->bytes_used - info->bytes_pinned -
7697 info->bytes_reserved - info->bytes_readonly,
7698 (info->full) ? "" : "not ");
7699 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7700 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7701 info->total_bytes, info->bytes_used, info->bytes_pinned,
7702 info->bytes_reserved, info->bytes_may_use,
7703 info->bytes_readonly);
7704 spin_unlock(&info->lock);
7706 if (!dump_block_groups)
7709 down_read(&info->groups_sem);
7711 list_for_each_entry(cache, &info->block_groups[index], list) {
7712 spin_lock(&cache->lock);
7713 printk(KERN_INFO "BTRFS: "
7714 "block group %llu has %llu bytes, "
7715 "%llu used %llu pinned %llu reserved %s\n",
7716 cache->key.objectid, cache->key.offset,
7717 btrfs_block_group_used(&cache->item), cache->pinned,
7718 cache->reserved, cache->ro ? "[readonly]" : "");
7719 btrfs_dump_free_space(cache, bytes);
7720 spin_unlock(&cache->lock);
7722 if (++index < BTRFS_NR_RAID_TYPES)
7724 up_read(&info->groups_sem);
7727 int btrfs_reserve_extent(struct btrfs_root *root,
7728 u64 num_bytes, u64 min_alloc_size,
7729 u64 empty_size, u64 hint_byte,
7730 struct btrfs_key *ins, int is_data, int delalloc)
7732 bool final_tried = num_bytes == min_alloc_size;
7736 flags = btrfs_get_alloc_profile(root, is_data);
7738 WARN_ON(num_bytes < root->sectorsize);
7739 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7741 if (!ret && !is_data) {
7742 btrfs_dec_block_group_reservations(root->fs_info,
7744 } else if (ret == -ENOSPC) {
7745 if (!final_tried && ins->offset) {
7746 num_bytes = min(num_bytes >> 1, ins->offset);
7747 num_bytes = round_down(num_bytes, root->sectorsize);
7748 num_bytes = max(num_bytes, min_alloc_size);
7749 if (num_bytes == min_alloc_size)
7752 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7753 struct btrfs_space_info *sinfo;
7755 sinfo = __find_space_info(root->fs_info, flags);
7756 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7759 dump_space_info(sinfo, num_bytes, 1);
7766 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7768 int pin, int delalloc)
7770 struct btrfs_block_group_cache *cache;
7773 cache = btrfs_lookup_block_group(root->fs_info, start);
7775 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7781 pin_down_extent(root, cache, start, len, 1);
7783 if (btrfs_test_opt(root, DISCARD))
7784 ret = btrfs_discard_extent(root, start, len, NULL);
7785 btrfs_add_free_space(cache, start, len);
7786 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7789 btrfs_put_block_group(cache);
7791 trace_btrfs_reserved_extent_free(root, start, len);
7796 int btrfs_free_reserved_extent(struct btrfs_root *root,
7797 u64 start, u64 len, int delalloc)
7799 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7802 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7805 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7808 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7809 struct btrfs_root *root,
7810 u64 parent, u64 root_objectid,
7811 u64 flags, u64 owner, u64 offset,
7812 struct btrfs_key *ins, int ref_mod)
7815 struct btrfs_fs_info *fs_info = root->fs_info;
7816 struct btrfs_extent_item *extent_item;
7817 struct btrfs_extent_inline_ref *iref;
7818 struct btrfs_path *path;
7819 struct extent_buffer *leaf;
7824 type = BTRFS_SHARED_DATA_REF_KEY;
7826 type = BTRFS_EXTENT_DATA_REF_KEY;
7828 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7830 path = btrfs_alloc_path();
7834 path->leave_spinning = 1;
7835 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7838 btrfs_free_path(path);
7842 leaf = path->nodes[0];
7843 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7844 struct btrfs_extent_item);
7845 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7846 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7847 btrfs_set_extent_flags(leaf, extent_item,
7848 flags | BTRFS_EXTENT_FLAG_DATA);
7850 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7851 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7853 struct btrfs_shared_data_ref *ref;
7854 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7855 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7856 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7858 struct btrfs_extent_data_ref *ref;
7859 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7860 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7861 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7862 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7863 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7866 btrfs_mark_buffer_dirty(path->nodes[0]);
7867 btrfs_free_path(path);
7869 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7874 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7875 if (ret) { /* -ENOENT, logic error */
7876 btrfs_err(fs_info, "update block group failed for %llu %llu",
7877 ins->objectid, ins->offset);
7880 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7884 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7885 struct btrfs_root *root,
7886 u64 parent, u64 root_objectid,
7887 u64 flags, struct btrfs_disk_key *key,
7888 int level, struct btrfs_key *ins)
7891 struct btrfs_fs_info *fs_info = root->fs_info;
7892 struct btrfs_extent_item *extent_item;
7893 struct btrfs_tree_block_info *block_info;
7894 struct btrfs_extent_inline_ref *iref;
7895 struct btrfs_path *path;
7896 struct extent_buffer *leaf;
7897 u32 size = sizeof(*extent_item) + sizeof(*iref);
7898 u64 num_bytes = ins->offset;
7899 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7902 if (!skinny_metadata)
7903 size += sizeof(*block_info);
7905 path = btrfs_alloc_path();
7907 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7912 path->leave_spinning = 1;
7913 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7916 btrfs_free_path(path);
7917 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7922 leaf = path->nodes[0];
7923 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7924 struct btrfs_extent_item);
7925 btrfs_set_extent_refs(leaf, extent_item, 1);
7926 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7927 btrfs_set_extent_flags(leaf, extent_item,
7928 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7930 if (skinny_metadata) {
7931 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7932 num_bytes = root->nodesize;
7934 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7935 btrfs_set_tree_block_key(leaf, block_info, key);
7936 btrfs_set_tree_block_level(leaf, block_info, level);
7937 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7941 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7942 btrfs_set_extent_inline_ref_type(leaf, iref,
7943 BTRFS_SHARED_BLOCK_REF_KEY);
7944 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7946 btrfs_set_extent_inline_ref_type(leaf, iref,
7947 BTRFS_TREE_BLOCK_REF_KEY);
7948 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7951 btrfs_mark_buffer_dirty(leaf);
7952 btrfs_free_path(path);
7954 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
7959 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7961 if (ret) { /* -ENOENT, logic error */
7962 btrfs_err(fs_info, "update block group failed for %llu %llu",
7963 ins->objectid, ins->offset);
7967 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7971 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7972 struct btrfs_root *root,
7973 u64 root_objectid, u64 owner,
7974 u64 offset, u64 ram_bytes,
7975 struct btrfs_key *ins)
7979 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7981 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7983 root_objectid, owner, offset,
7984 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
7990 * this is used by the tree logging recovery code. It records that
7991 * an extent has been allocated and makes sure to clear the free
7992 * space cache bits as well
7994 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7995 struct btrfs_root *root,
7996 u64 root_objectid, u64 owner, u64 offset,
7997 struct btrfs_key *ins)
8000 struct btrfs_block_group_cache *block_group;
8003 * Mixed block groups will exclude before processing the log so we only
8004 * need to do the exclude dance if this fs isn't mixed.
8006 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
8007 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
8012 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
8016 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
8017 RESERVE_ALLOC_NO_ACCOUNT, 0);
8018 BUG_ON(ret); /* logic error */
8019 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
8020 0, owner, offset, ins, 1);
8021 btrfs_put_block_group(block_group);
8025 static struct extent_buffer *
8026 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8027 u64 bytenr, int level)
8029 struct extent_buffer *buf;
8031 buf = btrfs_find_create_tree_block(root, bytenr);
8035 btrfs_set_header_generation(buf, trans->transid);
8036 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8037 btrfs_tree_lock(buf);
8038 clean_tree_block(trans, root->fs_info, buf);
8039 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8041 btrfs_set_lock_blocking(buf);
8042 set_extent_buffer_uptodate(buf);
8044 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8045 buf->log_index = root->log_transid % 2;
8047 * we allow two log transactions at a time, use different
8048 * EXENT bit to differentiate dirty pages.
8050 if (buf->log_index == 0)
8051 set_extent_dirty(&root->dirty_log_pages, buf->start,
8052 buf->start + buf->len - 1, GFP_NOFS);
8054 set_extent_new(&root->dirty_log_pages, buf->start,
8055 buf->start + buf->len - 1);
8057 buf->log_index = -1;
8058 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8059 buf->start + buf->len - 1, GFP_NOFS);
8061 trans->dirty = true;
8062 /* this returns a buffer locked for blocking */
8066 static struct btrfs_block_rsv *
8067 use_block_rsv(struct btrfs_trans_handle *trans,
8068 struct btrfs_root *root, u32 blocksize)
8070 struct btrfs_block_rsv *block_rsv;
8071 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
8073 bool global_updated = false;
8075 block_rsv = get_block_rsv(trans, root);
8077 if (unlikely(block_rsv->size == 0))
8080 ret = block_rsv_use_bytes(block_rsv, blocksize);
8084 if (block_rsv->failfast)
8085 return ERR_PTR(ret);
8087 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8088 global_updated = true;
8089 update_global_block_rsv(root->fs_info);
8093 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
8094 static DEFINE_RATELIMIT_STATE(_rs,
8095 DEFAULT_RATELIMIT_INTERVAL * 10,
8096 /*DEFAULT_RATELIMIT_BURST*/ 1);
8097 if (__ratelimit(&_rs))
8099 "BTRFS: block rsv returned %d\n", ret);
8102 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8103 BTRFS_RESERVE_NO_FLUSH);
8107 * If we couldn't reserve metadata bytes try and use some from
8108 * the global reserve if its space type is the same as the global
8111 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8112 block_rsv->space_info == global_rsv->space_info) {
8113 ret = block_rsv_use_bytes(global_rsv, blocksize);
8117 return ERR_PTR(ret);
8120 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8121 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8123 block_rsv_add_bytes(block_rsv, blocksize, 0);
8124 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8128 * finds a free extent and does all the dirty work required for allocation
8129 * returns the tree buffer or an ERR_PTR on error.
8131 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8132 struct btrfs_root *root,
8133 u64 parent, u64 root_objectid,
8134 struct btrfs_disk_key *key, int level,
8135 u64 hint, u64 empty_size)
8137 struct btrfs_key ins;
8138 struct btrfs_block_rsv *block_rsv;
8139 struct extent_buffer *buf;
8140 struct btrfs_delayed_extent_op *extent_op;
8143 u32 blocksize = root->nodesize;
8144 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8147 if (btrfs_test_is_dummy_root(root)) {
8148 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8151 root->alloc_bytenr += blocksize;
8155 block_rsv = use_block_rsv(trans, root, blocksize);
8156 if (IS_ERR(block_rsv))
8157 return ERR_CAST(block_rsv);
8159 ret = btrfs_reserve_extent(root, blocksize, blocksize,
8160 empty_size, hint, &ins, 0, 0);
8164 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8167 goto out_free_reserved;
8170 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8172 parent = ins.objectid;
8173 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8177 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8178 extent_op = btrfs_alloc_delayed_extent_op();
8184 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8186 memset(&extent_op->key, 0, sizeof(extent_op->key));
8187 extent_op->flags_to_set = flags;
8188 extent_op->update_key = skinny_metadata ? false : true;
8189 extent_op->update_flags = true;
8190 extent_op->is_data = false;
8191 extent_op->level = level;
8193 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8194 ins.objectid, ins.offset,
8195 parent, root_objectid, level,
8196 BTRFS_ADD_DELAYED_EXTENT,
8199 goto out_free_delayed;
8204 btrfs_free_delayed_extent_op(extent_op);
8206 free_extent_buffer(buf);
8208 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8210 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8211 return ERR_PTR(ret);
8214 struct walk_control {
8215 u64 refs[BTRFS_MAX_LEVEL];
8216 u64 flags[BTRFS_MAX_LEVEL];
8217 struct btrfs_key update_progress;
8228 #define DROP_REFERENCE 1
8229 #define UPDATE_BACKREF 2
8231 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8232 struct btrfs_root *root,
8233 struct walk_control *wc,
8234 struct btrfs_path *path)
8242 struct btrfs_key key;
8243 struct extent_buffer *eb;
8248 if (path->slots[wc->level] < wc->reada_slot) {
8249 wc->reada_count = wc->reada_count * 2 / 3;
8250 wc->reada_count = max(wc->reada_count, 2);
8252 wc->reada_count = wc->reada_count * 3 / 2;
8253 wc->reada_count = min_t(int, wc->reada_count,
8254 BTRFS_NODEPTRS_PER_BLOCK(root));
8257 eb = path->nodes[wc->level];
8258 nritems = btrfs_header_nritems(eb);
8259 blocksize = root->nodesize;
8261 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8262 if (nread >= wc->reada_count)
8266 bytenr = btrfs_node_blockptr(eb, slot);
8267 generation = btrfs_node_ptr_generation(eb, slot);
8269 if (slot == path->slots[wc->level])
8272 if (wc->stage == UPDATE_BACKREF &&
8273 generation <= root->root_key.offset)
8276 /* We don't lock the tree block, it's OK to be racy here */
8277 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8278 wc->level - 1, 1, &refs,
8280 /* We don't care about errors in readahead. */
8285 if (wc->stage == DROP_REFERENCE) {
8289 if (wc->level == 1 &&
8290 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8292 if (!wc->update_ref ||
8293 generation <= root->root_key.offset)
8295 btrfs_node_key_to_cpu(eb, &key, slot);
8296 ret = btrfs_comp_cpu_keys(&key,
8297 &wc->update_progress);
8301 if (wc->level == 1 &&
8302 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8306 readahead_tree_block(root, bytenr);
8309 wc->reada_slot = slot;
8313 * These may not be seen by the usual inc/dec ref code so we have to
8316 static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8317 struct btrfs_root *root, u64 bytenr,
8320 struct btrfs_qgroup_extent_record *qrecord;
8321 struct btrfs_delayed_ref_root *delayed_refs;
8323 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8327 qrecord->bytenr = bytenr;
8328 qrecord->num_bytes = num_bytes;
8329 qrecord->old_roots = NULL;
8331 delayed_refs = &trans->transaction->delayed_refs;
8332 spin_lock(&delayed_refs->lock);
8333 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8335 spin_unlock(&delayed_refs->lock);
8340 static int account_leaf_items(struct btrfs_trans_handle *trans,
8341 struct btrfs_root *root,
8342 struct extent_buffer *eb)
8344 int nr = btrfs_header_nritems(eb);
8345 int i, extent_type, ret;
8346 struct btrfs_key key;
8347 struct btrfs_file_extent_item *fi;
8348 u64 bytenr, num_bytes;
8350 /* We can be called directly from walk_up_proc() */
8351 if (!root->fs_info->quota_enabled)
8354 for (i = 0; i < nr; i++) {
8355 btrfs_item_key_to_cpu(eb, &key, i);
8357 if (key.type != BTRFS_EXTENT_DATA_KEY)
8360 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8361 /* filter out non qgroup-accountable extents */
8362 extent_type = btrfs_file_extent_type(eb, fi);
8364 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8367 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8371 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8373 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8381 * Walk up the tree from the bottom, freeing leaves and any interior
8382 * nodes which have had all slots visited. If a node (leaf or
8383 * interior) is freed, the node above it will have it's slot
8384 * incremented. The root node will never be freed.
8386 * At the end of this function, we should have a path which has all
8387 * slots incremented to the next position for a search. If we need to
8388 * read a new node it will be NULL and the node above it will have the
8389 * correct slot selected for a later read.
8391 * If we increment the root nodes slot counter past the number of
8392 * elements, 1 is returned to signal completion of the search.
8394 static int adjust_slots_upwards(struct btrfs_root *root,
8395 struct btrfs_path *path, int root_level)
8399 struct extent_buffer *eb;
8401 if (root_level == 0)
8404 while (level <= root_level) {
8405 eb = path->nodes[level];
8406 nr = btrfs_header_nritems(eb);
8407 path->slots[level]++;
8408 slot = path->slots[level];
8409 if (slot >= nr || level == 0) {
8411 * Don't free the root - we will detect this
8412 * condition after our loop and return a
8413 * positive value for caller to stop walking the tree.
8415 if (level != root_level) {
8416 btrfs_tree_unlock_rw(eb, path->locks[level]);
8417 path->locks[level] = 0;
8419 free_extent_buffer(eb);
8420 path->nodes[level] = NULL;
8421 path->slots[level] = 0;
8425 * We have a valid slot to walk back down
8426 * from. Stop here so caller can process these
8435 eb = path->nodes[root_level];
8436 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8443 * root_eb is the subtree root and is locked before this function is called.
8445 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8446 struct btrfs_root *root,
8447 struct extent_buffer *root_eb,
8453 struct extent_buffer *eb = root_eb;
8454 struct btrfs_path *path = NULL;
8456 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8457 BUG_ON(root_eb == NULL);
8459 if (!root->fs_info->quota_enabled)
8462 if (!extent_buffer_uptodate(root_eb)) {
8463 ret = btrfs_read_buffer(root_eb, root_gen);
8468 if (root_level == 0) {
8469 ret = account_leaf_items(trans, root, root_eb);
8473 path = btrfs_alloc_path();
8478 * Walk down the tree. Missing extent blocks are filled in as
8479 * we go. Metadata is accounted every time we read a new
8482 * When we reach a leaf, we account for file extent items in it,
8483 * walk back up the tree (adjusting slot pointers as we go)
8484 * and restart the search process.
8486 extent_buffer_get(root_eb); /* For path */
8487 path->nodes[root_level] = root_eb;
8488 path->slots[root_level] = 0;
8489 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8492 while (level >= 0) {
8493 if (path->nodes[level] == NULL) {
8498 /* We need to get child blockptr/gen from
8499 * parent before we can read it. */
8500 eb = path->nodes[level + 1];
8501 parent_slot = path->slots[level + 1];
8502 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8503 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8505 eb = read_tree_block(root, child_bytenr, child_gen);
8509 } else if (!extent_buffer_uptodate(eb)) {
8510 free_extent_buffer(eb);
8515 path->nodes[level] = eb;
8516 path->slots[level] = 0;
8518 btrfs_tree_read_lock(eb);
8519 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8520 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8522 ret = record_one_subtree_extent(trans, root, child_bytenr,
8529 ret = account_leaf_items(trans, root, path->nodes[level]);
8533 /* Nonzero return here means we completed our search */
8534 ret = adjust_slots_upwards(root, path, root_level);
8538 /* Restart search with new slots */
8547 btrfs_free_path(path);
8553 * helper to process tree block while walking down the tree.
8555 * when wc->stage == UPDATE_BACKREF, this function updates
8556 * back refs for pointers in the block.
8558 * NOTE: return value 1 means we should stop walking down.
8560 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8561 struct btrfs_root *root,
8562 struct btrfs_path *path,
8563 struct walk_control *wc, int lookup_info)
8565 int level = wc->level;
8566 struct extent_buffer *eb = path->nodes[level];
8567 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8570 if (wc->stage == UPDATE_BACKREF &&
8571 btrfs_header_owner(eb) != root->root_key.objectid)
8575 * when reference count of tree block is 1, it won't increase
8576 * again. once full backref flag is set, we never clear it.
8579 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8580 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8581 BUG_ON(!path->locks[level]);
8582 ret = btrfs_lookup_extent_info(trans, root,
8583 eb->start, level, 1,
8586 BUG_ON(ret == -ENOMEM);
8589 BUG_ON(wc->refs[level] == 0);
8592 if (wc->stage == DROP_REFERENCE) {
8593 if (wc->refs[level] > 1)
8596 if (path->locks[level] && !wc->keep_locks) {
8597 btrfs_tree_unlock_rw(eb, path->locks[level]);
8598 path->locks[level] = 0;
8603 /* wc->stage == UPDATE_BACKREF */
8604 if (!(wc->flags[level] & flag)) {
8605 BUG_ON(!path->locks[level]);
8606 ret = btrfs_inc_ref(trans, root, eb, 1);
8607 BUG_ON(ret); /* -ENOMEM */
8608 ret = btrfs_dec_ref(trans, root, eb, 0);
8609 BUG_ON(ret); /* -ENOMEM */
8610 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8612 btrfs_header_level(eb), 0);
8613 BUG_ON(ret); /* -ENOMEM */
8614 wc->flags[level] |= flag;
8618 * the block is shared by multiple trees, so it's not good to
8619 * keep the tree lock
8621 if (path->locks[level] && level > 0) {
8622 btrfs_tree_unlock_rw(eb, path->locks[level]);
8623 path->locks[level] = 0;
8629 * helper to process tree block pointer.
8631 * when wc->stage == DROP_REFERENCE, this function checks
8632 * reference count of the block pointed to. if the block
8633 * is shared and we need update back refs for the subtree
8634 * rooted at the block, this function changes wc->stage to
8635 * UPDATE_BACKREF. if the block is shared and there is no
8636 * need to update back, this function drops the reference
8639 * NOTE: return value 1 means we should stop walking down.
8641 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8642 struct btrfs_root *root,
8643 struct btrfs_path *path,
8644 struct walk_control *wc, int *lookup_info)
8650 struct btrfs_key key;
8651 struct extent_buffer *next;
8652 int level = wc->level;
8655 bool need_account = false;
8657 generation = btrfs_node_ptr_generation(path->nodes[level],
8658 path->slots[level]);
8660 * if the lower level block was created before the snapshot
8661 * was created, we know there is no need to update back refs
8664 if (wc->stage == UPDATE_BACKREF &&
8665 generation <= root->root_key.offset) {
8670 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8671 blocksize = root->nodesize;
8673 next = btrfs_find_tree_block(root->fs_info, bytenr);
8675 next = btrfs_find_create_tree_block(root, bytenr);
8677 return PTR_ERR(next);
8679 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8683 btrfs_tree_lock(next);
8684 btrfs_set_lock_blocking(next);
8686 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8687 &wc->refs[level - 1],
8688 &wc->flags[level - 1]);
8690 btrfs_tree_unlock(next);
8694 if (unlikely(wc->refs[level - 1] == 0)) {
8695 btrfs_err(root->fs_info, "Missing references.");
8700 if (wc->stage == DROP_REFERENCE) {
8701 if (wc->refs[level - 1] > 1) {
8702 need_account = true;
8704 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8707 if (!wc->update_ref ||
8708 generation <= root->root_key.offset)
8711 btrfs_node_key_to_cpu(path->nodes[level], &key,
8712 path->slots[level]);
8713 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8717 wc->stage = UPDATE_BACKREF;
8718 wc->shared_level = level - 1;
8722 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8726 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8727 btrfs_tree_unlock(next);
8728 free_extent_buffer(next);
8734 if (reada && level == 1)
8735 reada_walk_down(trans, root, wc, path);
8736 next = read_tree_block(root, bytenr, generation);
8738 return PTR_ERR(next);
8739 } else if (!extent_buffer_uptodate(next)) {
8740 free_extent_buffer(next);
8743 btrfs_tree_lock(next);
8744 btrfs_set_lock_blocking(next);
8748 BUG_ON(level != btrfs_header_level(next));
8749 path->nodes[level] = next;
8750 path->slots[level] = 0;
8751 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8757 wc->refs[level - 1] = 0;
8758 wc->flags[level - 1] = 0;
8759 if (wc->stage == DROP_REFERENCE) {
8760 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8761 parent = path->nodes[level]->start;
8763 BUG_ON(root->root_key.objectid !=
8764 btrfs_header_owner(path->nodes[level]));
8769 ret = account_shared_subtree(trans, root, next,
8770 generation, level - 1);
8772 btrfs_err_rl(root->fs_info,
8774 "%d accounting shared subtree. Quota "
8775 "is out of sync, rescan required.",
8779 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8780 root->root_key.objectid, level - 1, 0);
8781 BUG_ON(ret); /* -ENOMEM */
8783 btrfs_tree_unlock(next);
8784 free_extent_buffer(next);
8790 * helper to process tree block while walking up the tree.
8792 * when wc->stage == DROP_REFERENCE, this function drops
8793 * reference count on the block.
8795 * when wc->stage == UPDATE_BACKREF, this function changes
8796 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8797 * to UPDATE_BACKREF previously while processing the block.
8799 * NOTE: return value 1 means we should stop walking up.
8801 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8802 struct btrfs_root *root,
8803 struct btrfs_path *path,
8804 struct walk_control *wc)
8807 int level = wc->level;
8808 struct extent_buffer *eb = path->nodes[level];
8811 if (wc->stage == UPDATE_BACKREF) {
8812 BUG_ON(wc->shared_level < level);
8813 if (level < wc->shared_level)
8816 ret = find_next_key(path, level + 1, &wc->update_progress);
8820 wc->stage = DROP_REFERENCE;
8821 wc->shared_level = -1;
8822 path->slots[level] = 0;
8825 * check reference count again if the block isn't locked.
8826 * we should start walking down the tree again if reference
8829 if (!path->locks[level]) {
8831 btrfs_tree_lock(eb);
8832 btrfs_set_lock_blocking(eb);
8833 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8835 ret = btrfs_lookup_extent_info(trans, root,
8836 eb->start, level, 1,
8840 btrfs_tree_unlock_rw(eb, path->locks[level]);
8841 path->locks[level] = 0;
8844 BUG_ON(wc->refs[level] == 0);
8845 if (wc->refs[level] == 1) {
8846 btrfs_tree_unlock_rw(eb, path->locks[level]);
8847 path->locks[level] = 0;
8853 /* wc->stage == DROP_REFERENCE */
8854 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8856 if (wc->refs[level] == 1) {
8858 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8859 ret = btrfs_dec_ref(trans, root, eb, 1);
8861 ret = btrfs_dec_ref(trans, root, eb, 0);
8862 BUG_ON(ret); /* -ENOMEM */
8863 ret = account_leaf_items(trans, root, eb);
8865 btrfs_err_rl(root->fs_info,
8867 "%d accounting leaf items. Quota "
8868 "is out of sync, rescan required.",
8872 /* make block locked assertion in clean_tree_block happy */
8873 if (!path->locks[level] &&
8874 btrfs_header_generation(eb) == trans->transid) {
8875 btrfs_tree_lock(eb);
8876 btrfs_set_lock_blocking(eb);
8877 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8879 clean_tree_block(trans, root->fs_info, eb);
8882 if (eb == root->node) {
8883 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8886 BUG_ON(root->root_key.objectid !=
8887 btrfs_header_owner(eb));
8889 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8890 parent = path->nodes[level + 1]->start;
8892 BUG_ON(root->root_key.objectid !=
8893 btrfs_header_owner(path->nodes[level + 1]));
8896 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8898 wc->refs[level] = 0;
8899 wc->flags[level] = 0;
8903 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8904 struct btrfs_root *root,
8905 struct btrfs_path *path,
8906 struct walk_control *wc)
8908 int level = wc->level;
8909 int lookup_info = 1;
8912 while (level >= 0) {
8913 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8920 if (path->slots[level] >=
8921 btrfs_header_nritems(path->nodes[level]))
8924 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8926 path->slots[level]++;
8935 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8936 struct btrfs_root *root,
8937 struct btrfs_path *path,
8938 struct walk_control *wc, int max_level)
8940 int level = wc->level;
8943 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8944 while (level < max_level && path->nodes[level]) {
8946 if (path->slots[level] + 1 <
8947 btrfs_header_nritems(path->nodes[level])) {
8948 path->slots[level]++;
8951 ret = walk_up_proc(trans, root, path, wc);
8955 if (path->locks[level]) {
8956 btrfs_tree_unlock_rw(path->nodes[level],
8957 path->locks[level]);
8958 path->locks[level] = 0;
8960 free_extent_buffer(path->nodes[level]);
8961 path->nodes[level] = NULL;
8969 * drop a subvolume tree.
8971 * this function traverses the tree freeing any blocks that only
8972 * referenced by the tree.
8974 * when a shared tree block is found. this function decreases its
8975 * reference count by one. if update_ref is true, this function
8976 * also make sure backrefs for the shared block and all lower level
8977 * blocks are properly updated.
8979 * If called with for_reloc == 0, may exit early with -EAGAIN
8981 int btrfs_drop_snapshot(struct btrfs_root *root,
8982 struct btrfs_block_rsv *block_rsv, int update_ref,
8985 struct btrfs_path *path;
8986 struct btrfs_trans_handle *trans;
8987 struct btrfs_root *tree_root = root->fs_info->tree_root;
8988 struct btrfs_root_item *root_item = &root->root_item;
8989 struct walk_control *wc;
8990 struct btrfs_key key;
8994 bool root_dropped = false;
8996 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8998 path = btrfs_alloc_path();
9004 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9006 btrfs_free_path(path);
9011 trans = btrfs_start_transaction(tree_root, 0);
9012 if (IS_ERR(trans)) {
9013 err = PTR_ERR(trans);
9018 trans->block_rsv = block_rsv;
9020 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9021 level = btrfs_header_level(root->node);
9022 path->nodes[level] = btrfs_lock_root_node(root);
9023 btrfs_set_lock_blocking(path->nodes[level]);
9024 path->slots[level] = 0;
9025 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9026 memset(&wc->update_progress, 0,
9027 sizeof(wc->update_progress));
9029 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9030 memcpy(&wc->update_progress, &key,
9031 sizeof(wc->update_progress));
9033 level = root_item->drop_level;
9035 path->lowest_level = level;
9036 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9037 path->lowest_level = 0;
9045 * unlock our path, this is safe because only this
9046 * function is allowed to delete this snapshot
9048 btrfs_unlock_up_safe(path, 0);
9050 level = btrfs_header_level(root->node);
9052 btrfs_tree_lock(path->nodes[level]);
9053 btrfs_set_lock_blocking(path->nodes[level]);
9054 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9056 ret = btrfs_lookup_extent_info(trans, root,
9057 path->nodes[level]->start,
9058 level, 1, &wc->refs[level],
9064 BUG_ON(wc->refs[level] == 0);
9066 if (level == root_item->drop_level)
9069 btrfs_tree_unlock(path->nodes[level]);
9070 path->locks[level] = 0;
9071 WARN_ON(wc->refs[level] != 1);
9077 wc->shared_level = -1;
9078 wc->stage = DROP_REFERENCE;
9079 wc->update_ref = update_ref;
9081 wc->for_reloc = for_reloc;
9082 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9086 ret = walk_down_tree(trans, root, path, wc);
9092 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9099 BUG_ON(wc->stage != DROP_REFERENCE);
9103 if (wc->stage == DROP_REFERENCE) {
9105 btrfs_node_key(path->nodes[level],
9106 &root_item->drop_progress,
9107 path->slots[level]);
9108 root_item->drop_level = level;
9111 BUG_ON(wc->level == 0);
9112 if (btrfs_should_end_transaction(trans, tree_root) ||
9113 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
9114 ret = btrfs_update_root(trans, tree_root,
9118 btrfs_abort_transaction(trans, tree_root, ret);
9123 btrfs_end_transaction_throttle(trans, tree_root);
9124 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
9125 pr_debug("BTRFS: drop snapshot early exit\n");
9130 trans = btrfs_start_transaction(tree_root, 0);
9131 if (IS_ERR(trans)) {
9132 err = PTR_ERR(trans);
9136 trans->block_rsv = block_rsv;
9139 btrfs_release_path(path);
9143 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9145 btrfs_abort_transaction(trans, tree_root, ret);
9149 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9150 ret = btrfs_find_root(tree_root, &root->root_key, path,
9153 btrfs_abort_transaction(trans, tree_root, ret);
9156 } else if (ret > 0) {
9157 /* if we fail to delete the orphan item this time
9158 * around, it'll get picked up the next time.
9160 * The most common failure here is just -ENOENT.
9162 btrfs_del_orphan_item(trans, tree_root,
9163 root->root_key.objectid);
9167 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9168 btrfs_add_dropped_root(trans, root);
9170 free_extent_buffer(root->node);
9171 free_extent_buffer(root->commit_root);
9172 btrfs_put_fs_root(root);
9174 root_dropped = true;
9176 btrfs_end_transaction_throttle(trans, tree_root);
9179 btrfs_free_path(path);
9182 * So if we need to stop dropping the snapshot for whatever reason we
9183 * need to make sure to add it back to the dead root list so that we
9184 * keep trying to do the work later. This also cleans up roots if we
9185 * don't have it in the radix (like when we recover after a power fail
9186 * or unmount) so we don't leak memory.
9188 if (!for_reloc && root_dropped == false)
9189 btrfs_add_dead_root(root);
9190 if (err && err != -EAGAIN)
9191 btrfs_handle_fs_error(root->fs_info, err, NULL);
9196 * drop subtree rooted at tree block 'node'.
9198 * NOTE: this function will unlock and release tree block 'node'
9199 * only used by relocation code
9201 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9202 struct btrfs_root *root,
9203 struct extent_buffer *node,
9204 struct extent_buffer *parent)
9206 struct btrfs_path *path;
9207 struct walk_control *wc;
9213 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9215 path = btrfs_alloc_path();
9219 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9221 btrfs_free_path(path);
9225 btrfs_assert_tree_locked(parent);
9226 parent_level = btrfs_header_level(parent);
9227 extent_buffer_get(parent);
9228 path->nodes[parent_level] = parent;
9229 path->slots[parent_level] = btrfs_header_nritems(parent);
9231 btrfs_assert_tree_locked(node);
9232 level = btrfs_header_level(node);
9233 path->nodes[level] = node;
9234 path->slots[level] = 0;
9235 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9237 wc->refs[parent_level] = 1;
9238 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9240 wc->shared_level = -1;
9241 wc->stage = DROP_REFERENCE;
9245 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9248 wret = walk_down_tree(trans, root, path, wc);
9254 wret = walk_up_tree(trans, root, path, wc, parent_level);
9262 btrfs_free_path(path);
9266 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9272 * if restripe for this chunk_type is on pick target profile and
9273 * return, otherwise do the usual balance
9275 stripped = get_restripe_target(root->fs_info, flags);
9277 return extended_to_chunk(stripped);
9279 num_devices = root->fs_info->fs_devices->rw_devices;
9281 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9282 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9283 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9285 if (num_devices == 1) {
9286 stripped |= BTRFS_BLOCK_GROUP_DUP;
9287 stripped = flags & ~stripped;
9289 /* turn raid0 into single device chunks */
9290 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9293 /* turn mirroring into duplication */
9294 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9295 BTRFS_BLOCK_GROUP_RAID10))
9296 return stripped | BTRFS_BLOCK_GROUP_DUP;
9298 /* they already had raid on here, just return */
9299 if (flags & stripped)
9302 stripped |= BTRFS_BLOCK_GROUP_DUP;
9303 stripped = flags & ~stripped;
9305 /* switch duplicated blocks with raid1 */
9306 if (flags & BTRFS_BLOCK_GROUP_DUP)
9307 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9309 /* this is drive concat, leave it alone */
9315 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9317 struct btrfs_space_info *sinfo = cache->space_info;
9319 u64 min_allocable_bytes;
9323 * We need some metadata space and system metadata space for
9324 * allocating chunks in some corner cases until we force to set
9325 * it to be readonly.
9328 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9330 min_allocable_bytes = SZ_1M;
9332 min_allocable_bytes = 0;
9334 spin_lock(&sinfo->lock);
9335 spin_lock(&cache->lock);
9343 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9344 cache->bytes_super - btrfs_block_group_used(&cache->item);
9346 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9347 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9348 min_allocable_bytes <= sinfo->total_bytes) {
9349 sinfo->bytes_readonly += num_bytes;
9351 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9355 spin_unlock(&cache->lock);
9356 spin_unlock(&sinfo->lock);
9360 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9361 struct btrfs_block_group_cache *cache)
9364 struct btrfs_trans_handle *trans;
9369 trans = btrfs_join_transaction(root);
9371 return PTR_ERR(trans);
9374 * we're not allowed to set block groups readonly after the dirty
9375 * block groups cache has started writing. If it already started,
9376 * back off and let this transaction commit
9378 mutex_lock(&root->fs_info->ro_block_group_mutex);
9379 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9380 u64 transid = trans->transid;
9382 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9383 btrfs_end_transaction(trans, root);
9385 ret = btrfs_wait_for_commit(root, transid);
9392 * if we are changing raid levels, try to allocate a corresponding
9393 * block group with the new raid level.
9395 alloc_flags = update_block_group_flags(root, cache->flags);
9396 if (alloc_flags != cache->flags) {
9397 ret = do_chunk_alloc(trans, root, alloc_flags,
9400 * ENOSPC is allowed here, we may have enough space
9401 * already allocated at the new raid level to
9410 ret = inc_block_group_ro(cache, 0);
9413 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9414 ret = do_chunk_alloc(trans, root, alloc_flags,
9418 ret = inc_block_group_ro(cache, 0);
9420 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9421 alloc_flags = update_block_group_flags(root, cache->flags);
9422 lock_chunks(root->fs_info->chunk_root);
9423 check_system_chunk(trans, root, alloc_flags);
9424 unlock_chunks(root->fs_info->chunk_root);
9426 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9428 btrfs_end_transaction(trans, root);
9432 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9433 struct btrfs_root *root, u64 type)
9435 u64 alloc_flags = get_alloc_profile(root, type);
9436 return do_chunk_alloc(trans, root, alloc_flags,
9441 * helper to account the unused space of all the readonly block group in the
9442 * space_info. takes mirrors into account.
9444 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9446 struct btrfs_block_group_cache *block_group;
9450 /* It's df, we don't care if it's racy */
9451 if (list_empty(&sinfo->ro_bgs))
9454 spin_lock(&sinfo->lock);
9455 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9456 spin_lock(&block_group->lock);
9458 if (!block_group->ro) {
9459 spin_unlock(&block_group->lock);
9463 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9464 BTRFS_BLOCK_GROUP_RAID10 |
9465 BTRFS_BLOCK_GROUP_DUP))
9470 free_bytes += (block_group->key.offset -
9471 btrfs_block_group_used(&block_group->item)) *
9474 spin_unlock(&block_group->lock);
9476 spin_unlock(&sinfo->lock);
9481 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9482 struct btrfs_block_group_cache *cache)
9484 struct btrfs_space_info *sinfo = cache->space_info;
9489 spin_lock(&sinfo->lock);
9490 spin_lock(&cache->lock);
9492 num_bytes = cache->key.offset - cache->reserved -
9493 cache->pinned - cache->bytes_super -
9494 btrfs_block_group_used(&cache->item);
9495 sinfo->bytes_readonly -= num_bytes;
9496 list_del_init(&cache->ro_list);
9498 spin_unlock(&cache->lock);
9499 spin_unlock(&sinfo->lock);
9503 * checks to see if its even possible to relocate this block group.
9505 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9506 * ok to go ahead and try.
9508 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9510 struct btrfs_block_group_cache *block_group;
9511 struct btrfs_space_info *space_info;
9512 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9513 struct btrfs_device *device;
9514 struct btrfs_trans_handle *trans;
9524 debug = btrfs_test_opt(root, ENOSPC_DEBUG);
9526 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9528 /* odd, couldn't find the block group, leave it alone */
9531 btrfs_warn(root->fs_info,
9532 "can't find block group for bytenr %llu",
9537 min_free = btrfs_block_group_used(&block_group->item);
9539 /* no bytes used, we're good */
9543 space_info = block_group->space_info;
9544 spin_lock(&space_info->lock);
9546 full = space_info->full;
9549 * if this is the last block group we have in this space, we can't
9550 * relocate it unless we're able to allocate a new chunk below.
9552 * Otherwise, we need to make sure we have room in the space to handle
9553 * all of the extents from this block group. If we can, we're good
9555 if ((space_info->total_bytes != block_group->key.offset) &&
9556 (space_info->bytes_used + space_info->bytes_reserved +
9557 space_info->bytes_pinned + space_info->bytes_readonly +
9558 min_free < space_info->total_bytes)) {
9559 spin_unlock(&space_info->lock);
9562 spin_unlock(&space_info->lock);
9565 * ok we don't have enough space, but maybe we have free space on our
9566 * devices to allocate new chunks for relocation, so loop through our
9567 * alloc devices and guess if we have enough space. if this block
9568 * group is going to be restriped, run checks against the target
9569 * profile instead of the current one.
9581 target = get_restripe_target(root->fs_info, block_group->flags);
9583 index = __get_raid_index(extended_to_chunk(target));
9586 * this is just a balance, so if we were marked as full
9587 * we know there is no space for a new chunk
9591 btrfs_warn(root->fs_info,
9592 "no space to alloc new chunk for block group %llu",
9593 block_group->key.objectid);
9597 index = get_block_group_index(block_group);
9600 if (index == BTRFS_RAID_RAID10) {
9604 } else if (index == BTRFS_RAID_RAID1) {
9606 } else if (index == BTRFS_RAID_DUP) {
9609 } else if (index == BTRFS_RAID_RAID0) {
9610 dev_min = fs_devices->rw_devices;
9611 min_free = div64_u64(min_free, dev_min);
9614 /* We need to do this so that we can look at pending chunks */
9615 trans = btrfs_join_transaction(root);
9616 if (IS_ERR(trans)) {
9617 ret = PTR_ERR(trans);
9621 mutex_lock(&root->fs_info->chunk_mutex);
9622 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9626 * check to make sure we can actually find a chunk with enough
9627 * space to fit our block group in.
9629 if (device->total_bytes > device->bytes_used + min_free &&
9630 !device->is_tgtdev_for_dev_replace) {
9631 ret = find_free_dev_extent(trans, device, min_free,
9636 if (dev_nr >= dev_min)
9642 if (debug && ret == -1)
9643 btrfs_warn(root->fs_info,
9644 "no space to allocate a new chunk for block group %llu",
9645 block_group->key.objectid);
9646 mutex_unlock(&root->fs_info->chunk_mutex);
9647 btrfs_end_transaction(trans, root);
9649 btrfs_put_block_group(block_group);
9653 static int find_first_block_group(struct btrfs_root *root,
9654 struct btrfs_path *path, struct btrfs_key *key)
9657 struct btrfs_key found_key;
9658 struct extent_buffer *leaf;
9661 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9666 slot = path->slots[0];
9667 leaf = path->nodes[0];
9668 if (slot >= btrfs_header_nritems(leaf)) {
9669 ret = btrfs_next_leaf(root, path);
9676 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9678 if (found_key.objectid >= key->objectid &&
9679 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9689 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9691 struct btrfs_block_group_cache *block_group;
9695 struct inode *inode;
9697 block_group = btrfs_lookup_first_block_group(info, last);
9698 while (block_group) {
9699 spin_lock(&block_group->lock);
9700 if (block_group->iref)
9702 spin_unlock(&block_group->lock);
9703 block_group = next_block_group(info->tree_root,
9713 inode = block_group->inode;
9714 block_group->iref = 0;
9715 block_group->inode = NULL;
9716 spin_unlock(&block_group->lock);
9718 last = block_group->key.objectid + block_group->key.offset;
9719 btrfs_put_block_group(block_group);
9723 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9725 struct btrfs_block_group_cache *block_group;
9726 struct btrfs_space_info *space_info;
9727 struct btrfs_caching_control *caching_ctl;
9730 down_write(&info->commit_root_sem);
9731 while (!list_empty(&info->caching_block_groups)) {
9732 caching_ctl = list_entry(info->caching_block_groups.next,
9733 struct btrfs_caching_control, list);
9734 list_del(&caching_ctl->list);
9735 put_caching_control(caching_ctl);
9737 up_write(&info->commit_root_sem);
9739 spin_lock(&info->unused_bgs_lock);
9740 while (!list_empty(&info->unused_bgs)) {
9741 block_group = list_first_entry(&info->unused_bgs,
9742 struct btrfs_block_group_cache,
9744 list_del_init(&block_group->bg_list);
9745 btrfs_put_block_group(block_group);
9747 spin_unlock(&info->unused_bgs_lock);
9749 spin_lock(&info->block_group_cache_lock);
9750 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9751 block_group = rb_entry(n, struct btrfs_block_group_cache,
9753 rb_erase(&block_group->cache_node,
9754 &info->block_group_cache_tree);
9755 RB_CLEAR_NODE(&block_group->cache_node);
9756 spin_unlock(&info->block_group_cache_lock);
9758 down_write(&block_group->space_info->groups_sem);
9759 list_del(&block_group->list);
9760 up_write(&block_group->space_info->groups_sem);
9762 if (block_group->cached == BTRFS_CACHE_STARTED)
9763 wait_block_group_cache_done(block_group);
9766 * We haven't cached this block group, which means we could
9767 * possibly have excluded extents on this block group.
9769 if (block_group->cached == BTRFS_CACHE_NO ||
9770 block_group->cached == BTRFS_CACHE_ERROR)
9771 free_excluded_extents(info->extent_root, block_group);
9773 btrfs_remove_free_space_cache(block_group);
9774 btrfs_put_block_group(block_group);
9776 spin_lock(&info->block_group_cache_lock);
9778 spin_unlock(&info->block_group_cache_lock);
9780 /* now that all the block groups are freed, go through and
9781 * free all the space_info structs. This is only called during
9782 * the final stages of unmount, and so we know nobody is
9783 * using them. We call synchronize_rcu() once before we start,
9784 * just to be on the safe side.
9788 release_global_block_rsv(info);
9790 while (!list_empty(&info->space_info)) {
9793 space_info = list_entry(info->space_info.next,
9794 struct btrfs_space_info,
9796 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9797 if (WARN_ON(space_info->bytes_pinned > 0 ||
9798 space_info->bytes_reserved > 0 ||
9799 space_info->bytes_may_use > 0)) {
9800 dump_space_info(space_info, 0, 0);
9803 list_del(&space_info->list);
9804 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9805 struct kobject *kobj;
9806 kobj = space_info->block_group_kobjs[i];
9807 space_info->block_group_kobjs[i] = NULL;
9813 kobject_del(&space_info->kobj);
9814 kobject_put(&space_info->kobj);
9819 static void __link_block_group(struct btrfs_space_info *space_info,
9820 struct btrfs_block_group_cache *cache)
9822 int index = get_block_group_index(cache);
9825 down_write(&space_info->groups_sem);
9826 if (list_empty(&space_info->block_groups[index]))
9828 list_add_tail(&cache->list, &space_info->block_groups[index]);
9829 up_write(&space_info->groups_sem);
9832 struct raid_kobject *rkobj;
9835 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9838 rkobj->raid_type = index;
9839 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9840 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9841 "%s", get_raid_name(index));
9843 kobject_put(&rkobj->kobj);
9846 space_info->block_group_kobjs[index] = &rkobj->kobj;
9851 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9854 static struct btrfs_block_group_cache *
9855 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9857 struct btrfs_block_group_cache *cache;
9859 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9863 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9865 if (!cache->free_space_ctl) {
9870 cache->key.objectid = start;
9871 cache->key.offset = size;
9872 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9874 cache->sectorsize = root->sectorsize;
9875 cache->fs_info = root->fs_info;
9876 cache->full_stripe_len = btrfs_full_stripe_len(root,
9877 &root->fs_info->mapping_tree,
9879 set_free_space_tree_thresholds(cache);
9881 atomic_set(&cache->count, 1);
9882 spin_lock_init(&cache->lock);
9883 init_rwsem(&cache->data_rwsem);
9884 INIT_LIST_HEAD(&cache->list);
9885 INIT_LIST_HEAD(&cache->cluster_list);
9886 INIT_LIST_HEAD(&cache->bg_list);
9887 INIT_LIST_HEAD(&cache->ro_list);
9888 INIT_LIST_HEAD(&cache->dirty_list);
9889 INIT_LIST_HEAD(&cache->io_list);
9890 btrfs_init_free_space_ctl(cache);
9891 atomic_set(&cache->trimming, 0);
9892 mutex_init(&cache->free_space_lock);
9897 int btrfs_read_block_groups(struct btrfs_root *root)
9899 struct btrfs_path *path;
9901 struct btrfs_block_group_cache *cache;
9902 struct btrfs_fs_info *info = root->fs_info;
9903 struct btrfs_space_info *space_info;
9904 struct btrfs_key key;
9905 struct btrfs_key found_key;
9906 struct extent_buffer *leaf;
9910 root = info->extent_root;
9913 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9914 path = btrfs_alloc_path();
9917 path->reada = READA_FORWARD;
9919 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9920 if (btrfs_test_opt(root, SPACE_CACHE) &&
9921 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9923 if (btrfs_test_opt(root, CLEAR_CACHE))
9927 ret = find_first_block_group(root, path, &key);
9933 leaf = path->nodes[0];
9934 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9936 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9945 * When we mount with old space cache, we need to
9946 * set BTRFS_DC_CLEAR and set dirty flag.
9948 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9949 * truncate the old free space cache inode and
9951 * b) Setting 'dirty flag' makes sure that we flush
9952 * the new space cache info onto disk.
9954 if (btrfs_test_opt(root, SPACE_CACHE))
9955 cache->disk_cache_state = BTRFS_DC_CLEAR;
9958 read_extent_buffer(leaf, &cache->item,
9959 btrfs_item_ptr_offset(leaf, path->slots[0]),
9960 sizeof(cache->item));
9961 cache->flags = btrfs_block_group_flags(&cache->item);
9963 key.objectid = found_key.objectid + found_key.offset;
9964 btrfs_release_path(path);
9967 * We need to exclude the super stripes now so that the space
9968 * info has super bytes accounted for, otherwise we'll think
9969 * we have more space than we actually do.
9971 ret = exclude_super_stripes(root, cache);
9974 * We may have excluded something, so call this just in
9977 free_excluded_extents(root, cache);
9978 btrfs_put_block_group(cache);
9983 * check for two cases, either we are full, and therefore
9984 * don't need to bother with the caching work since we won't
9985 * find any space, or we are empty, and we can just add all
9986 * the space in and be done with it. This saves us _alot_ of
9987 * time, particularly in the full case.
9989 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9990 cache->last_byte_to_unpin = (u64)-1;
9991 cache->cached = BTRFS_CACHE_FINISHED;
9992 free_excluded_extents(root, cache);
9993 } else if (btrfs_block_group_used(&cache->item) == 0) {
9994 cache->last_byte_to_unpin = (u64)-1;
9995 cache->cached = BTRFS_CACHE_FINISHED;
9996 add_new_free_space(cache, root->fs_info,
9998 found_key.objectid +
10000 free_excluded_extents(root, cache);
10003 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10005 btrfs_remove_free_space_cache(cache);
10006 btrfs_put_block_group(cache);
10010 ret = update_space_info(info, cache->flags, found_key.offset,
10011 btrfs_block_group_used(&cache->item),
10012 cache->bytes_super, &space_info);
10014 btrfs_remove_free_space_cache(cache);
10015 spin_lock(&info->block_group_cache_lock);
10016 rb_erase(&cache->cache_node,
10017 &info->block_group_cache_tree);
10018 RB_CLEAR_NODE(&cache->cache_node);
10019 spin_unlock(&info->block_group_cache_lock);
10020 btrfs_put_block_group(cache);
10024 cache->space_info = space_info;
10026 __link_block_group(space_info, cache);
10028 set_avail_alloc_bits(root->fs_info, cache->flags);
10029 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
10030 inc_block_group_ro(cache, 1);
10031 } else if (btrfs_block_group_used(&cache->item) == 0) {
10032 spin_lock(&info->unused_bgs_lock);
10033 /* Should always be true but just in case. */
10034 if (list_empty(&cache->bg_list)) {
10035 btrfs_get_block_group(cache);
10036 list_add_tail(&cache->bg_list,
10037 &info->unused_bgs);
10039 spin_unlock(&info->unused_bgs_lock);
10043 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
10044 if (!(get_alloc_profile(root, space_info->flags) &
10045 (BTRFS_BLOCK_GROUP_RAID10 |
10046 BTRFS_BLOCK_GROUP_RAID1 |
10047 BTRFS_BLOCK_GROUP_RAID5 |
10048 BTRFS_BLOCK_GROUP_RAID6 |
10049 BTRFS_BLOCK_GROUP_DUP)))
10052 * avoid allocating from un-mirrored block group if there are
10053 * mirrored block groups.
10055 list_for_each_entry(cache,
10056 &space_info->block_groups[BTRFS_RAID_RAID0],
10058 inc_block_group_ro(cache, 1);
10059 list_for_each_entry(cache,
10060 &space_info->block_groups[BTRFS_RAID_SINGLE],
10062 inc_block_group_ro(cache, 1);
10065 init_global_block_rsv(info);
10068 btrfs_free_path(path);
10072 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10073 struct btrfs_root *root)
10075 struct btrfs_block_group_cache *block_group, *tmp;
10076 struct btrfs_root *extent_root = root->fs_info->extent_root;
10077 struct btrfs_block_group_item item;
10078 struct btrfs_key key;
10080 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10082 trans->can_flush_pending_bgs = false;
10083 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10087 spin_lock(&block_group->lock);
10088 memcpy(&item, &block_group->item, sizeof(item));
10089 memcpy(&key, &block_group->key, sizeof(key));
10090 spin_unlock(&block_group->lock);
10092 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10095 btrfs_abort_transaction(trans, extent_root, ret);
10096 ret = btrfs_finish_chunk_alloc(trans, extent_root,
10097 key.objectid, key.offset);
10099 btrfs_abort_transaction(trans, extent_root, ret);
10100 add_block_group_free_space(trans, root->fs_info, block_group);
10101 /* already aborted the transaction if it failed. */
10103 list_del_init(&block_group->bg_list);
10105 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10108 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10109 struct btrfs_root *root, u64 bytes_used,
10110 u64 type, u64 chunk_objectid, u64 chunk_offset,
10114 struct btrfs_root *extent_root;
10115 struct btrfs_block_group_cache *cache;
10116 extent_root = root->fs_info->extent_root;
10118 btrfs_set_log_full_commit(root->fs_info, trans);
10120 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
10124 btrfs_set_block_group_used(&cache->item, bytes_used);
10125 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10126 btrfs_set_block_group_flags(&cache->item, type);
10128 cache->flags = type;
10129 cache->last_byte_to_unpin = (u64)-1;
10130 cache->cached = BTRFS_CACHE_FINISHED;
10131 cache->needs_free_space = 1;
10132 ret = exclude_super_stripes(root, cache);
10135 * We may have excluded something, so call this just in
10138 free_excluded_extents(root, cache);
10139 btrfs_put_block_group(cache);
10143 add_new_free_space(cache, root->fs_info, chunk_offset,
10144 chunk_offset + size);
10146 free_excluded_extents(root, cache);
10148 #ifdef CONFIG_BTRFS_DEBUG
10149 if (btrfs_should_fragment_free_space(root, cache)) {
10150 u64 new_bytes_used = size - bytes_used;
10152 bytes_used += new_bytes_used >> 1;
10153 fragment_free_space(root, cache);
10157 * Call to ensure the corresponding space_info object is created and
10158 * assigned to our block group, but don't update its counters just yet.
10159 * We want our bg to be added to the rbtree with its ->space_info set.
10161 ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0,
10162 &cache->space_info);
10164 btrfs_remove_free_space_cache(cache);
10165 btrfs_put_block_group(cache);
10169 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10171 btrfs_remove_free_space_cache(cache);
10172 btrfs_put_block_group(cache);
10177 * Now that our block group has its ->space_info set and is inserted in
10178 * the rbtree, update the space info's counters.
10180 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10181 cache->bytes_super, &cache->space_info);
10183 btrfs_remove_free_space_cache(cache);
10184 spin_lock(&root->fs_info->block_group_cache_lock);
10185 rb_erase(&cache->cache_node,
10186 &root->fs_info->block_group_cache_tree);
10187 RB_CLEAR_NODE(&cache->cache_node);
10188 spin_unlock(&root->fs_info->block_group_cache_lock);
10189 btrfs_put_block_group(cache);
10192 update_global_block_rsv(root->fs_info);
10194 __link_block_group(cache->space_info, cache);
10196 list_add_tail(&cache->bg_list, &trans->new_bgs);
10198 set_avail_alloc_bits(extent_root->fs_info, type);
10202 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10204 u64 extra_flags = chunk_to_extended(flags) &
10205 BTRFS_EXTENDED_PROFILE_MASK;
10207 write_seqlock(&fs_info->profiles_lock);
10208 if (flags & BTRFS_BLOCK_GROUP_DATA)
10209 fs_info->avail_data_alloc_bits &= ~extra_flags;
10210 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10211 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10212 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10213 fs_info->avail_system_alloc_bits &= ~extra_flags;
10214 write_sequnlock(&fs_info->profiles_lock);
10217 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10218 struct btrfs_root *root, u64 group_start,
10219 struct extent_map *em)
10221 struct btrfs_path *path;
10222 struct btrfs_block_group_cache *block_group;
10223 struct btrfs_free_cluster *cluster;
10224 struct btrfs_root *tree_root = root->fs_info->tree_root;
10225 struct btrfs_key key;
10226 struct inode *inode;
10227 struct kobject *kobj = NULL;
10231 struct btrfs_caching_control *caching_ctl = NULL;
10234 root = root->fs_info->extent_root;
10236 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10237 BUG_ON(!block_group);
10238 BUG_ON(!block_group->ro);
10241 * Free the reserved super bytes from this block group before
10244 free_excluded_extents(root, block_group);
10246 memcpy(&key, &block_group->key, sizeof(key));
10247 index = get_block_group_index(block_group);
10248 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10249 BTRFS_BLOCK_GROUP_RAID1 |
10250 BTRFS_BLOCK_GROUP_RAID10))
10255 /* make sure this block group isn't part of an allocation cluster */
10256 cluster = &root->fs_info->data_alloc_cluster;
10257 spin_lock(&cluster->refill_lock);
10258 btrfs_return_cluster_to_free_space(block_group, cluster);
10259 spin_unlock(&cluster->refill_lock);
10262 * make sure this block group isn't part of a metadata
10263 * allocation cluster
10265 cluster = &root->fs_info->meta_alloc_cluster;
10266 spin_lock(&cluster->refill_lock);
10267 btrfs_return_cluster_to_free_space(block_group, cluster);
10268 spin_unlock(&cluster->refill_lock);
10270 path = btrfs_alloc_path();
10277 * get the inode first so any iput calls done for the io_list
10278 * aren't the final iput (no unlinks allowed now)
10280 inode = lookup_free_space_inode(tree_root, block_group, path);
10282 mutex_lock(&trans->transaction->cache_write_mutex);
10284 * make sure our free spache cache IO is done before remove the
10287 spin_lock(&trans->transaction->dirty_bgs_lock);
10288 if (!list_empty(&block_group->io_list)) {
10289 list_del_init(&block_group->io_list);
10291 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10293 spin_unlock(&trans->transaction->dirty_bgs_lock);
10294 btrfs_wait_cache_io(root, trans, block_group,
10295 &block_group->io_ctl, path,
10296 block_group->key.objectid);
10297 btrfs_put_block_group(block_group);
10298 spin_lock(&trans->transaction->dirty_bgs_lock);
10301 if (!list_empty(&block_group->dirty_list)) {
10302 list_del_init(&block_group->dirty_list);
10303 btrfs_put_block_group(block_group);
10305 spin_unlock(&trans->transaction->dirty_bgs_lock);
10306 mutex_unlock(&trans->transaction->cache_write_mutex);
10308 if (!IS_ERR(inode)) {
10309 ret = btrfs_orphan_add(trans, inode);
10311 btrfs_add_delayed_iput(inode);
10314 clear_nlink(inode);
10315 /* One for the block groups ref */
10316 spin_lock(&block_group->lock);
10317 if (block_group->iref) {
10318 block_group->iref = 0;
10319 block_group->inode = NULL;
10320 spin_unlock(&block_group->lock);
10323 spin_unlock(&block_group->lock);
10325 /* One for our lookup ref */
10326 btrfs_add_delayed_iput(inode);
10329 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10330 key.offset = block_group->key.objectid;
10333 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10337 btrfs_release_path(path);
10339 ret = btrfs_del_item(trans, tree_root, path);
10342 btrfs_release_path(path);
10345 spin_lock(&root->fs_info->block_group_cache_lock);
10346 rb_erase(&block_group->cache_node,
10347 &root->fs_info->block_group_cache_tree);
10348 RB_CLEAR_NODE(&block_group->cache_node);
10350 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10351 root->fs_info->first_logical_byte = (u64)-1;
10352 spin_unlock(&root->fs_info->block_group_cache_lock);
10354 down_write(&block_group->space_info->groups_sem);
10356 * we must use list_del_init so people can check to see if they
10357 * are still on the list after taking the semaphore
10359 list_del_init(&block_group->list);
10360 if (list_empty(&block_group->space_info->block_groups[index])) {
10361 kobj = block_group->space_info->block_group_kobjs[index];
10362 block_group->space_info->block_group_kobjs[index] = NULL;
10363 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10365 up_write(&block_group->space_info->groups_sem);
10371 if (block_group->has_caching_ctl)
10372 caching_ctl = get_caching_control(block_group);
10373 if (block_group->cached == BTRFS_CACHE_STARTED)
10374 wait_block_group_cache_done(block_group);
10375 if (block_group->has_caching_ctl) {
10376 down_write(&root->fs_info->commit_root_sem);
10377 if (!caching_ctl) {
10378 struct btrfs_caching_control *ctl;
10380 list_for_each_entry(ctl,
10381 &root->fs_info->caching_block_groups, list)
10382 if (ctl->block_group == block_group) {
10384 atomic_inc(&caching_ctl->count);
10389 list_del_init(&caching_ctl->list);
10390 up_write(&root->fs_info->commit_root_sem);
10392 /* Once for the caching bgs list and once for us. */
10393 put_caching_control(caching_ctl);
10394 put_caching_control(caching_ctl);
10398 spin_lock(&trans->transaction->dirty_bgs_lock);
10399 if (!list_empty(&block_group->dirty_list)) {
10402 if (!list_empty(&block_group->io_list)) {
10405 spin_unlock(&trans->transaction->dirty_bgs_lock);
10406 btrfs_remove_free_space_cache(block_group);
10408 spin_lock(&block_group->space_info->lock);
10409 list_del_init(&block_group->ro_list);
10411 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10412 WARN_ON(block_group->space_info->total_bytes
10413 < block_group->key.offset);
10414 WARN_ON(block_group->space_info->bytes_readonly
10415 < block_group->key.offset);
10416 WARN_ON(block_group->space_info->disk_total
10417 < block_group->key.offset * factor);
10419 block_group->space_info->total_bytes -= block_group->key.offset;
10420 block_group->space_info->bytes_readonly -= block_group->key.offset;
10421 block_group->space_info->disk_total -= block_group->key.offset * factor;
10423 spin_unlock(&block_group->space_info->lock);
10425 memcpy(&key, &block_group->key, sizeof(key));
10428 if (!list_empty(&em->list)) {
10429 /* We're in the transaction->pending_chunks list. */
10430 free_extent_map(em);
10432 spin_lock(&block_group->lock);
10433 block_group->removed = 1;
10435 * At this point trimming can't start on this block group, because we
10436 * removed the block group from the tree fs_info->block_group_cache_tree
10437 * so no one can't find it anymore and even if someone already got this
10438 * block group before we removed it from the rbtree, they have already
10439 * incremented block_group->trimming - if they didn't, they won't find
10440 * any free space entries because we already removed them all when we
10441 * called btrfs_remove_free_space_cache().
10443 * And we must not remove the extent map from the fs_info->mapping_tree
10444 * to prevent the same logical address range and physical device space
10445 * ranges from being reused for a new block group. This is because our
10446 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10447 * completely transactionless, so while it is trimming a range the
10448 * currently running transaction might finish and a new one start,
10449 * allowing for new block groups to be created that can reuse the same
10450 * physical device locations unless we take this special care.
10452 * There may also be an implicit trim operation if the file system
10453 * is mounted with -odiscard. The same protections must remain
10454 * in place until the extents have been discarded completely when
10455 * the transaction commit has completed.
10457 remove_em = (atomic_read(&block_group->trimming) == 0);
10459 * Make sure a trimmer task always sees the em in the pinned_chunks list
10460 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10461 * before checking block_group->removed).
10465 * Our em might be in trans->transaction->pending_chunks which
10466 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10467 * and so is the fs_info->pinned_chunks list.
10469 * So at this point we must be holding the chunk_mutex to avoid
10470 * any races with chunk allocation (more specifically at
10471 * volumes.c:contains_pending_extent()), to ensure it always
10472 * sees the em, either in the pending_chunks list or in the
10473 * pinned_chunks list.
10475 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10477 spin_unlock(&block_group->lock);
10480 struct extent_map_tree *em_tree;
10482 em_tree = &root->fs_info->mapping_tree.map_tree;
10483 write_lock(&em_tree->lock);
10485 * The em might be in the pending_chunks list, so make sure the
10486 * chunk mutex is locked, since remove_extent_mapping() will
10487 * delete us from that list.
10489 remove_extent_mapping(em_tree, em);
10490 write_unlock(&em_tree->lock);
10491 /* once for the tree */
10492 free_extent_map(em);
10495 unlock_chunks(root);
10497 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10501 btrfs_put_block_group(block_group);
10502 btrfs_put_block_group(block_group);
10504 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10510 ret = btrfs_del_item(trans, root, path);
10512 btrfs_free_path(path);
10516 struct btrfs_trans_handle *
10517 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10518 const u64 chunk_offset)
10520 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10521 struct extent_map *em;
10522 struct map_lookup *map;
10523 unsigned int num_items;
10525 read_lock(&em_tree->lock);
10526 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10527 read_unlock(&em_tree->lock);
10528 ASSERT(em && em->start == chunk_offset);
10531 * We need to reserve 3 + N units from the metadata space info in order
10532 * to remove a block group (done at btrfs_remove_chunk() and at
10533 * btrfs_remove_block_group()), which are used for:
10535 * 1 unit for adding the free space inode's orphan (located in the tree
10537 * 1 unit for deleting the block group item (located in the extent
10539 * 1 unit for deleting the free space item (located in tree of tree
10541 * N units for deleting N device extent items corresponding to each
10542 * stripe (located in the device tree).
10544 * In order to remove a block group we also need to reserve units in the
10545 * system space info in order to update the chunk tree (update one or
10546 * more device items and remove one chunk item), but this is done at
10547 * btrfs_remove_chunk() through a call to check_system_chunk().
10549 map = em->map_lookup;
10550 num_items = 3 + map->num_stripes;
10551 free_extent_map(em);
10553 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10558 * Process the unused_bgs list and remove any that don't have any allocated
10559 * space inside of them.
10561 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10563 struct btrfs_block_group_cache *block_group;
10564 struct btrfs_space_info *space_info;
10565 struct btrfs_root *root = fs_info->extent_root;
10566 struct btrfs_trans_handle *trans;
10569 if (!fs_info->open)
10572 spin_lock(&fs_info->unused_bgs_lock);
10573 while (!list_empty(&fs_info->unused_bgs)) {
10577 block_group = list_first_entry(&fs_info->unused_bgs,
10578 struct btrfs_block_group_cache,
10580 list_del_init(&block_group->bg_list);
10582 space_info = block_group->space_info;
10584 if (ret || btrfs_mixed_space_info(space_info)) {
10585 btrfs_put_block_group(block_group);
10588 spin_unlock(&fs_info->unused_bgs_lock);
10590 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10592 /* Don't want to race with allocators so take the groups_sem */
10593 down_write(&space_info->groups_sem);
10594 spin_lock(&block_group->lock);
10595 if (block_group->reserved ||
10596 btrfs_block_group_used(&block_group->item) ||
10598 list_is_singular(&block_group->list)) {
10600 * We want to bail if we made new allocations or have
10601 * outstanding allocations in this block group. We do
10602 * the ro check in case balance is currently acting on
10603 * this block group.
10605 spin_unlock(&block_group->lock);
10606 up_write(&space_info->groups_sem);
10609 spin_unlock(&block_group->lock);
10611 /* We don't want to force the issue, only flip if it's ok. */
10612 ret = inc_block_group_ro(block_group, 0);
10613 up_write(&space_info->groups_sem);
10620 * Want to do this before we do anything else so we can recover
10621 * properly if we fail to join the transaction.
10623 trans = btrfs_start_trans_remove_block_group(fs_info,
10624 block_group->key.objectid);
10625 if (IS_ERR(trans)) {
10626 btrfs_dec_block_group_ro(root, block_group);
10627 ret = PTR_ERR(trans);
10632 * We could have pending pinned extents for this block group,
10633 * just delete them, we don't care about them anymore.
10635 start = block_group->key.objectid;
10636 end = start + block_group->key.offset - 1;
10638 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10639 * btrfs_finish_extent_commit(). If we are at transaction N,
10640 * another task might be running finish_extent_commit() for the
10641 * previous transaction N - 1, and have seen a range belonging
10642 * to the block group in freed_extents[] before we were able to
10643 * clear the whole block group range from freed_extents[]. This
10644 * means that task can lookup for the block group after we
10645 * unpinned it from freed_extents[] and removed it, leading to
10646 * a BUG_ON() at btrfs_unpin_extent_range().
10648 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10649 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10652 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10653 btrfs_dec_block_group_ro(root, block_group);
10656 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10659 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10660 btrfs_dec_block_group_ro(root, block_group);
10663 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10665 /* Reset pinned so btrfs_put_block_group doesn't complain */
10666 spin_lock(&space_info->lock);
10667 spin_lock(&block_group->lock);
10669 space_info->bytes_pinned -= block_group->pinned;
10670 space_info->bytes_readonly += block_group->pinned;
10671 percpu_counter_add(&space_info->total_bytes_pinned,
10672 -block_group->pinned);
10673 block_group->pinned = 0;
10675 spin_unlock(&block_group->lock);
10676 spin_unlock(&space_info->lock);
10678 /* DISCARD can flip during remount */
10679 trimming = btrfs_test_opt(root, DISCARD);
10681 /* Implicit trim during transaction commit. */
10683 btrfs_get_block_group_trimming(block_group);
10686 * Btrfs_remove_chunk will abort the transaction if things go
10689 ret = btrfs_remove_chunk(trans, root,
10690 block_group->key.objectid);
10694 btrfs_put_block_group_trimming(block_group);
10699 * If we're not mounted with -odiscard, we can just forget
10700 * about this block group. Otherwise we'll need to wait
10701 * until transaction commit to do the actual discard.
10704 spin_lock(&fs_info->unused_bgs_lock);
10706 * A concurrent scrub might have added us to the list
10707 * fs_info->unused_bgs, so use a list_move operation
10708 * to add the block group to the deleted_bgs list.
10710 list_move(&block_group->bg_list,
10711 &trans->transaction->deleted_bgs);
10712 spin_unlock(&fs_info->unused_bgs_lock);
10713 btrfs_get_block_group(block_group);
10716 btrfs_end_transaction(trans, root);
10718 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10719 btrfs_put_block_group(block_group);
10720 spin_lock(&fs_info->unused_bgs_lock);
10722 spin_unlock(&fs_info->unused_bgs_lock);
10725 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10727 struct btrfs_space_info *space_info;
10728 struct btrfs_super_block *disk_super;
10734 disk_super = fs_info->super_copy;
10735 if (!btrfs_super_root(disk_super))
10738 features = btrfs_super_incompat_flags(disk_super);
10739 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10742 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10743 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10748 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10749 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10751 flags = BTRFS_BLOCK_GROUP_METADATA;
10752 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10756 flags = BTRFS_BLOCK_GROUP_DATA;
10757 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10763 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10765 return unpin_extent_range(root, start, end, false);
10769 * It used to be that old block groups would be left around forever.
10770 * Iterating over them would be enough to trim unused space. Since we
10771 * now automatically remove them, we also need to iterate over unallocated
10774 * We don't want a transaction for this since the discard may take a
10775 * substantial amount of time. We don't require that a transaction be
10776 * running, but we do need to take a running transaction into account
10777 * to ensure that we're not discarding chunks that were released in
10778 * the current transaction.
10780 * Holding the chunks lock will prevent other threads from allocating
10781 * or releasing chunks, but it won't prevent a running transaction
10782 * from committing and releasing the memory that the pending chunks
10783 * list head uses. For that, we need to take a reference to the
10786 static int btrfs_trim_free_extents(struct btrfs_device *device,
10787 u64 minlen, u64 *trimmed)
10789 u64 start = 0, len = 0;
10794 /* Not writeable = nothing to do. */
10795 if (!device->writeable)
10798 /* No free space = nothing to do. */
10799 if (device->total_bytes <= device->bytes_used)
10805 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10806 struct btrfs_transaction *trans;
10809 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10813 down_read(&fs_info->commit_root_sem);
10815 spin_lock(&fs_info->trans_lock);
10816 trans = fs_info->running_transaction;
10818 atomic_inc(&trans->use_count);
10819 spin_unlock(&fs_info->trans_lock);
10821 ret = find_free_dev_extent_start(trans, device, minlen, start,
10824 btrfs_put_transaction(trans);
10827 up_read(&fs_info->commit_root_sem);
10828 mutex_unlock(&fs_info->chunk_mutex);
10829 if (ret == -ENOSPC)
10834 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10835 up_read(&fs_info->commit_root_sem);
10836 mutex_unlock(&fs_info->chunk_mutex);
10844 if (fatal_signal_pending(current)) {
10845 ret = -ERESTARTSYS;
10855 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10857 struct btrfs_fs_info *fs_info = root->fs_info;
10858 struct btrfs_block_group_cache *cache = NULL;
10859 struct btrfs_device *device;
10860 struct list_head *devices;
10865 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10869 * try to trim all FS space, our block group may start from non-zero.
10871 if (range->len == total_bytes)
10872 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10874 cache = btrfs_lookup_block_group(fs_info, range->start);
10877 if (cache->key.objectid >= (range->start + range->len)) {
10878 btrfs_put_block_group(cache);
10882 start = max(range->start, cache->key.objectid);
10883 end = min(range->start + range->len,
10884 cache->key.objectid + cache->key.offset);
10886 if (end - start >= range->minlen) {
10887 if (!block_group_cache_done(cache)) {
10888 ret = cache_block_group(cache, 0);
10890 btrfs_put_block_group(cache);
10893 ret = wait_block_group_cache_done(cache);
10895 btrfs_put_block_group(cache);
10899 ret = btrfs_trim_block_group(cache,
10905 trimmed += group_trimmed;
10907 btrfs_put_block_group(cache);
10912 cache = next_block_group(fs_info->tree_root, cache);
10915 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10916 devices = &root->fs_info->fs_devices->alloc_list;
10917 list_for_each_entry(device, devices, dev_alloc_list) {
10918 ret = btrfs_trim_free_extents(device, range->minlen,
10923 trimmed += group_trimmed;
10925 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10927 range->len = trimmed;
10932 * btrfs_{start,end}_write_no_snapshoting() are similar to
10933 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10934 * data into the page cache through nocow before the subvolume is snapshoted,
10935 * but flush the data into disk after the snapshot creation, or to prevent
10936 * operations while snapshoting is ongoing and that cause the snapshot to be
10937 * inconsistent (writes followed by expanding truncates for example).
10939 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10941 percpu_counter_dec(&root->subv_writers->counter);
10943 * Make sure counter is updated before we wake up waiters.
10946 if (waitqueue_active(&root->subv_writers->wait))
10947 wake_up(&root->subv_writers->wait);
10950 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10952 if (atomic_read(&root->will_be_snapshoted))
10955 percpu_counter_inc(&root->subv_writers->counter);
10957 * Make sure counter is updated before we check for snapshot creation.
10960 if (atomic_read(&root->will_be_snapshoted)) {
10961 btrfs_end_write_no_snapshoting(root);
10967 static int wait_snapshoting_atomic_t(atomic_t *a)
10973 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
10978 ret = btrfs_start_write_no_snapshoting(root);
10981 wait_on_atomic_t(&root->will_be_snapshoted,
10982 wait_snapshoting_atomic_t,
10983 TASK_UNINTERRUPTIBLE);