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/sched/signal.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/sort.h>
24 #include <linux/rcupdate.h>
25 #include <linux/kthread.h>
26 #include <linux/slab.h>
27 #include <linux/ratelimit.h>
28 #include <linux/percpu_counter.h>
32 #include "print-tree.h"
36 #include "free-space-cache.h"
37 #include "free-space-tree.h"
42 #undef SCRAMBLE_DELAYED_REFS
45 * control flags for do_chunk_alloc's force field
46 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
47 * if we really need one.
49 * CHUNK_ALLOC_LIMITED means to only try and allocate one
50 * if we have very few chunks already allocated. This is
51 * used as part of the clustering code to help make sure
52 * we have a good pool of storage to cluster in, without
53 * filling the FS with empty chunks
55 * CHUNK_ALLOC_FORCE means it must try to allocate one
59 CHUNK_ALLOC_NO_FORCE = 0,
60 CHUNK_ALLOC_LIMITED = 1,
61 CHUNK_ALLOC_FORCE = 2,
64 static int update_block_group(struct btrfs_trans_handle *trans,
65 struct btrfs_fs_info *fs_info, u64 bytenr,
66 u64 num_bytes, int alloc);
67 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
68 struct btrfs_fs_info *fs_info,
69 struct btrfs_delayed_ref_node *node, u64 parent,
70 u64 root_objectid, u64 owner_objectid,
71 u64 owner_offset, int refs_to_drop,
72 struct btrfs_delayed_extent_op *extra_op);
73 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
74 struct extent_buffer *leaf,
75 struct btrfs_extent_item *ei);
76 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
77 struct btrfs_fs_info *fs_info,
78 u64 parent, u64 root_objectid,
79 u64 flags, u64 owner, u64 offset,
80 struct btrfs_key *ins, int ref_mod);
81 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
82 struct btrfs_fs_info *fs_info,
83 u64 parent, u64 root_objectid,
84 u64 flags, struct btrfs_disk_key *key,
85 int level, struct btrfs_key *ins);
86 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
87 struct btrfs_fs_info *fs_info, u64 flags,
89 static int find_next_key(struct btrfs_path *path, int level,
90 struct btrfs_key *key);
91 static void dump_space_info(struct btrfs_fs_info *fs_info,
92 struct btrfs_space_info *info, u64 bytes,
93 int dump_block_groups);
94 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
95 u64 ram_bytes, u64 num_bytes, int delalloc);
96 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
97 u64 num_bytes, int delalloc);
98 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
100 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
101 struct btrfs_space_info *space_info,
103 enum btrfs_reserve_flush_enum flush,
105 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
106 struct btrfs_space_info *space_info,
108 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
109 struct btrfs_space_info *space_info,
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
116 return cache->cached == BTRFS_CACHE_FINISHED ||
117 cache->cached == BTRFS_CACHE_ERROR;
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
122 return (cache->flags & bits) == bits;
125 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
127 atomic_inc(&cache->count);
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
132 if (atomic_dec_and_test(&cache->count)) {
133 WARN_ON(cache->pinned > 0);
134 WARN_ON(cache->reserved > 0);
137 * If not empty, someone is still holding mutex of
138 * full_stripe_lock, which can only be released by caller.
139 * And it will definitely cause use-after-free when caller
140 * tries to release full stripe lock.
142 * No better way to resolve, but only to warn.
144 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
145 kfree(cache->free_space_ctl);
151 * this adds the block group to the fs_info rb tree for the block group
154 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
155 struct btrfs_block_group_cache *block_group)
158 struct rb_node *parent = NULL;
159 struct btrfs_block_group_cache *cache;
161 spin_lock(&info->block_group_cache_lock);
162 p = &info->block_group_cache_tree.rb_node;
166 cache = rb_entry(parent, struct btrfs_block_group_cache,
168 if (block_group->key.objectid < cache->key.objectid) {
170 } else if (block_group->key.objectid > cache->key.objectid) {
173 spin_unlock(&info->block_group_cache_lock);
178 rb_link_node(&block_group->cache_node, parent, p);
179 rb_insert_color(&block_group->cache_node,
180 &info->block_group_cache_tree);
182 if (info->first_logical_byte > block_group->key.objectid)
183 info->first_logical_byte = block_group->key.objectid;
185 spin_unlock(&info->block_group_cache_lock);
191 * This will return the block group at or after bytenr if contains is 0, else
192 * it will return the block group that contains the bytenr
194 static struct btrfs_block_group_cache *
195 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
198 struct btrfs_block_group_cache *cache, *ret = NULL;
202 spin_lock(&info->block_group_cache_lock);
203 n = info->block_group_cache_tree.rb_node;
206 cache = rb_entry(n, struct btrfs_block_group_cache,
208 end = cache->key.objectid + cache->key.offset - 1;
209 start = cache->key.objectid;
211 if (bytenr < start) {
212 if (!contains && (!ret || start < ret->key.objectid))
215 } else if (bytenr > start) {
216 if (contains && bytenr <= end) {
227 btrfs_get_block_group(ret);
228 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
229 info->first_logical_byte = ret->key.objectid;
231 spin_unlock(&info->block_group_cache_lock);
236 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
237 u64 start, u64 num_bytes)
239 u64 end = start + num_bytes - 1;
240 set_extent_bits(&fs_info->freed_extents[0],
241 start, end, EXTENT_UPTODATE);
242 set_extent_bits(&fs_info->freed_extents[1],
243 start, end, EXTENT_UPTODATE);
247 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
248 struct btrfs_block_group_cache *cache)
252 start = cache->key.objectid;
253 end = start + cache->key.offset - 1;
255 clear_extent_bits(&fs_info->freed_extents[0],
256 start, end, EXTENT_UPTODATE);
257 clear_extent_bits(&fs_info->freed_extents[1],
258 start, end, EXTENT_UPTODATE);
261 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
262 struct btrfs_block_group_cache *cache)
269 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
270 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
271 cache->bytes_super += stripe_len;
272 ret = add_excluded_extent(fs_info, cache->key.objectid,
278 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
279 bytenr = btrfs_sb_offset(i);
280 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
281 bytenr, 0, &logical, &nr, &stripe_len);
288 if (logical[nr] > cache->key.objectid +
292 if (logical[nr] + stripe_len <= cache->key.objectid)
296 if (start < cache->key.objectid) {
297 start = cache->key.objectid;
298 len = (logical[nr] + stripe_len) - start;
300 len = min_t(u64, stripe_len,
301 cache->key.objectid +
302 cache->key.offset - start);
305 cache->bytes_super += len;
306 ret = add_excluded_extent(fs_info, start, len);
318 static struct btrfs_caching_control *
319 get_caching_control(struct btrfs_block_group_cache *cache)
321 struct btrfs_caching_control *ctl;
323 spin_lock(&cache->lock);
324 if (!cache->caching_ctl) {
325 spin_unlock(&cache->lock);
329 ctl = cache->caching_ctl;
330 refcount_inc(&ctl->count);
331 spin_unlock(&cache->lock);
335 static void put_caching_control(struct btrfs_caching_control *ctl)
337 if (refcount_dec_and_test(&ctl->count))
341 #ifdef CONFIG_BTRFS_DEBUG
342 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
344 struct btrfs_fs_info *fs_info = block_group->fs_info;
345 u64 start = block_group->key.objectid;
346 u64 len = block_group->key.offset;
347 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
348 fs_info->nodesize : fs_info->sectorsize;
349 u64 step = chunk << 1;
351 while (len > chunk) {
352 btrfs_remove_free_space(block_group, start, chunk);
363 * this is only called by cache_block_group, since we could have freed extents
364 * we need to check the pinned_extents for any extents that can't be used yet
365 * since their free space will be released as soon as the transaction commits.
367 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
368 struct btrfs_fs_info *info, u64 start, u64 end)
370 u64 extent_start, extent_end, size, total_added = 0;
373 while (start < end) {
374 ret = find_first_extent_bit(info->pinned_extents, start,
375 &extent_start, &extent_end,
376 EXTENT_DIRTY | EXTENT_UPTODATE,
381 if (extent_start <= start) {
382 start = extent_end + 1;
383 } else if (extent_start > start && extent_start < end) {
384 size = extent_start - start;
386 ret = btrfs_add_free_space(block_group, start,
388 BUG_ON(ret); /* -ENOMEM or logic error */
389 start = extent_end + 1;
398 ret = btrfs_add_free_space(block_group, start, size);
399 BUG_ON(ret); /* -ENOMEM or logic error */
405 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
407 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
408 struct btrfs_fs_info *fs_info = block_group->fs_info;
409 struct btrfs_root *extent_root = fs_info->extent_root;
410 struct btrfs_path *path;
411 struct extent_buffer *leaf;
412 struct btrfs_key key;
419 path = btrfs_alloc_path();
423 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
425 #ifdef CONFIG_BTRFS_DEBUG
427 * If we're fragmenting we don't want to make anybody think we can
428 * allocate from this block group until we've had a chance to fragment
431 if (btrfs_should_fragment_free_space(block_group))
435 * We don't want to deadlock with somebody trying to allocate a new
436 * extent for the extent root while also trying to search the extent
437 * root to add free space. So we skip locking and search the commit
438 * root, since its read-only
440 path->skip_locking = 1;
441 path->search_commit_root = 1;
442 path->reada = READA_FORWARD;
446 key.type = BTRFS_EXTENT_ITEM_KEY;
449 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
453 leaf = path->nodes[0];
454 nritems = btrfs_header_nritems(leaf);
457 if (btrfs_fs_closing(fs_info) > 1) {
462 if (path->slots[0] < nritems) {
463 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
465 ret = find_next_key(path, 0, &key);
469 if (need_resched() ||
470 rwsem_is_contended(&fs_info->commit_root_sem)) {
472 caching_ctl->progress = last;
473 btrfs_release_path(path);
474 up_read(&fs_info->commit_root_sem);
475 mutex_unlock(&caching_ctl->mutex);
477 mutex_lock(&caching_ctl->mutex);
478 down_read(&fs_info->commit_root_sem);
482 ret = btrfs_next_leaf(extent_root, path);
487 leaf = path->nodes[0];
488 nritems = btrfs_header_nritems(leaf);
492 if (key.objectid < last) {
495 key.type = BTRFS_EXTENT_ITEM_KEY;
498 caching_ctl->progress = last;
499 btrfs_release_path(path);
503 if (key.objectid < block_group->key.objectid) {
508 if (key.objectid >= block_group->key.objectid +
509 block_group->key.offset)
512 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
513 key.type == BTRFS_METADATA_ITEM_KEY) {
514 total_found += add_new_free_space(block_group,
517 if (key.type == BTRFS_METADATA_ITEM_KEY)
518 last = key.objectid +
521 last = key.objectid + key.offset;
523 if (total_found > CACHING_CTL_WAKE_UP) {
526 wake_up(&caching_ctl->wait);
533 total_found += add_new_free_space(block_group, fs_info, last,
534 block_group->key.objectid +
535 block_group->key.offset);
536 caching_ctl->progress = (u64)-1;
539 btrfs_free_path(path);
543 static noinline void caching_thread(struct btrfs_work *work)
545 struct btrfs_block_group_cache *block_group;
546 struct btrfs_fs_info *fs_info;
547 struct btrfs_caching_control *caching_ctl;
548 struct btrfs_root *extent_root;
551 caching_ctl = container_of(work, struct btrfs_caching_control, work);
552 block_group = caching_ctl->block_group;
553 fs_info = block_group->fs_info;
554 extent_root = fs_info->extent_root;
556 mutex_lock(&caching_ctl->mutex);
557 down_read(&fs_info->commit_root_sem);
559 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
560 ret = load_free_space_tree(caching_ctl);
562 ret = load_extent_tree_free(caching_ctl);
564 spin_lock(&block_group->lock);
565 block_group->caching_ctl = NULL;
566 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
567 spin_unlock(&block_group->lock);
569 #ifdef CONFIG_BTRFS_DEBUG
570 if (btrfs_should_fragment_free_space(block_group)) {
573 spin_lock(&block_group->space_info->lock);
574 spin_lock(&block_group->lock);
575 bytes_used = block_group->key.offset -
576 btrfs_block_group_used(&block_group->item);
577 block_group->space_info->bytes_used += bytes_used >> 1;
578 spin_unlock(&block_group->lock);
579 spin_unlock(&block_group->space_info->lock);
580 fragment_free_space(block_group);
584 caching_ctl->progress = (u64)-1;
586 up_read(&fs_info->commit_root_sem);
587 free_excluded_extents(fs_info, block_group);
588 mutex_unlock(&caching_ctl->mutex);
590 wake_up(&caching_ctl->wait);
592 put_caching_control(caching_ctl);
593 btrfs_put_block_group(block_group);
596 static int cache_block_group(struct btrfs_block_group_cache *cache,
600 struct btrfs_fs_info *fs_info = cache->fs_info;
601 struct btrfs_caching_control *caching_ctl;
604 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
608 INIT_LIST_HEAD(&caching_ctl->list);
609 mutex_init(&caching_ctl->mutex);
610 init_waitqueue_head(&caching_ctl->wait);
611 caching_ctl->block_group = cache;
612 caching_ctl->progress = cache->key.objectid;
613 refcount_set(&caching_ctl->count, 1);
614 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
615 caching_thread, NULL, NULL);
617 spin_lock(&cache->lock);
619 * This should be a rare occasion, but this could happen I think in the
620 * case where one thread starts to load the space cache info, and then
621 * some other thread starts a transaction commit which tries to do an
622 * allocation while the other thread is still loading the space cache
623 * info. The previous loop should have kept us from choosing this block
624 * group, but if we've moved to the state where we will wait on caching
625 * block groups we need to first check if we're doing a fast load here,
626 * so we can wait for it to finish, otherwise we could end up allocating
627 * from a block group who's cache gets evicted for one reason or
630 while (cache->cached == BTRFS_CACHE_FAST) {
631 struct btrfs_caching_control *ctl;
633 ctl = cache->caching_ctl;
634 refcount_inc(&ctl->count);
635 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
636 spin_unlock(&cache->lock);
640 finish_wait(&ctl->wait, &wait);
641 put_caching_control(ctl);
642 spin_lock(&cache->lock);
645 if (cache->cached != BTRFS_CACHE_NO) {
646 spin_unlock(&cache->lock);
650 WARN_ON(cache->caching_ctl);
651 cache->caching_ctl = caching_ctl;
652 cache->cached = BTRFS_CACHE_FAST;
653 spin_unlock(&cache->lock);
655 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
656 mutex_lock(&caching_ctl->mutex);
657 ret = load_free_space_cache(fs_info, cache);
659 spin_lock(&cache->lock);
661 cache->caching_ctl = NULL;
662 cache->cached = BTRFS_CACHE_FINISHED;
663 cache->last_byte_to_unpin = (u64)-1;
664 caching_ctl->progress = (u64)-1;
666 if (load_cache_only) {
667 cache->caching_ctl = NULL;
668 cache->cached = BTRFS_CACHE_NO;
670 cache->cached = BTRFS_CACHE_STARTED;
671 cache->has_caching_ctl = 1;
674 spin_unlock(&cache->lock);
675 #ifdef CONFIG_BTRFS_DEBUG
677 btrfs_should_fragment_free_space(cache)) {
680 spin_lock(&cache->space_info->lock);
681 spin_lock(&cache->lock);
682 bytes_used = cache->key.offset -
683 btrfs_block_group_used(&cache->item);
684 cache->space_info->bytes_used += bytes_used >> 1;
685 spin_unlock(&cache->lock);
686 spin_unlock(&cache->space_info->lock);
687 fragment_free_space(cache);
690 mutex_unlock(&caching_ctl->mutex);
692 wake_up(&caching_ctl->wait);
694 put_caching_control(caching_ctl);
695 free_excluded_extents(fs_info, cache);
700 * We're either using the free space tree or no caching at all.
701 * Set cached to the appropriate value and wakeup any waiters.
703 spin_lock(&cache->lock);
704 if (load_cache_only) {
705 cache->caching_ctl = NULL;
706 cache->cached = BTRFS_CACHE_NO;
708 cache->cached = BTRFS_CACHE_STARTED;
709 cache->has_caching_ctl = 1;
711 spin_unlock(&cache->lock);
712 wake_up(&caching_ctl->wait);
715 if (load_cache_only) {
716 put_caching_control(caching_ctl);
720 down_write(&fs_info->commit_root_sem);
721 refcount_inc(&caching_ctl->count);
722 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
723 up_write(&fs_info->commit_root_sem);
725 btrfs_get_block_group(cache);
727 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
733 * return the block group that starts at or after bytenr
735 static struct btrfs_block_group_cache *
736 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
738 return block_group_cache_tree_search(info, bytenr, 0);
742 * return the block group that contains the given bytenr
744 struct btrfs_block_group_cache *btrfs_lookup_block_group(
745 struct btrfs_fs_info *info,
748 return block_group_cache_tree_search(info, bytenr, 1);
751 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
754 struct list_head *head = &info->space_info;
755 struct btrfs_space_info *found;
757 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
760 list_for_each_entry_rcu(found, head, list) {
761 if (found->flags & flags) {
770 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
771 u64 owner, u64 root_objectid)
773 struct btrfs_space_info *space_info;
776 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
777 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
778 flags = BTRFS_BLOCK_GROUP_SYSTEM;
780 flags = BTRFS_BLOCK_GROUP_METADATA;
782 flags = BTRFS_BLOCK_GROUP_DATA;
785 space_info = __find_space_info(fs_info, flags);
787 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
791 * after adding space to the filesystem, we need to clear the full flags
792 * on all the space infos.
794 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
796 struct list_head *head = &info->space_info;
797 struct btrfs_space_info *found;
800 list_for_each_entry_rcu(found, head, list)
805 /* simple helper to search for an existing data extent at a given offset */
806 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
809 struct btrfs_key key;
810 struct btrfs_path *path;
812 path = btrfs_alloc_path();
816 key.objectid = start;
818 key.type = BTRFS_EXTENT_ITEM_KEY;
819 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
820 btrfs_free_path(path);
825 * helper function to lookup reference count and flags of a tree block.
827 * the head node for delayed ref is used to store the sum of all the
828 * reference count modifications queued up in the rbtree. the head
829 * node may also store the extent flags to set. This way you can check
830 * to see what the reference count and extent flags would be if all of
831 * the delayed refs are not processed.
833 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
834 struct btrfs_fs_info *fs_info, u64 bytenr,
835 u64 offset, int metadata, u64 *refs, u64 *flags)
837 struct btrfs_delayed_ref_head *head;
838 struct btrfs_delayed_ref_root *delayed_refs;
839 struct btrfs_path *path;
840 struct btrfs_extent_item *ei;
841 struct extent_buffer *leaf;
842 struct btrfs_key key;
849 * If we don't have skinny metadata, don't bother doing anything
852 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
853 offset = fs_info->nodesize;
857 path = btrfs_alloc_path();
862 path->skip_locking = 1;
863 path->search_commit_root = 1;
867 key.objectid = bytenr;
870 key.type = BTRFS_METADATA_ITEM_KEY;
872 key.type = BTRFS_EXTENT_ITEM_KEY;
874 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
878 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
879 if (path->slots[0]) {
881 btrfs_item_key_to_cpu(path->nodes[0], &key,
883 if (key.objectid == bytenr &&
884 key.type == BTRFS_EXTENT_ITEM_KEY &&
885 key.offset == fs_info->nodesize)
891 leaf = path->nodes[0];
892 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
893 if (item_size >= sizeof(*ei)) {
894 ei = btrfs_item_ptr(leaf, path->slots[0],
895 struct btrfs_extent_item);
896 num_refs = btrfs_extent_refs(leaf, ei);
897 extent_flags = btrfs_extent_flags(leaf, ei);
899 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
900 struct btrfs_extent_item_v0 *ei0;
901 BUG_ON(item_size != sizeof(*ei0));
902 ei0 = btrfs_item_ptr(leaf, path->slots[0],
903 struct btrfs_extent_item_v0);
904 num_refs = btrfs_extent_refs_v0(leaf, ei0);
905 /* FIXME: this isn't correct for data */
906 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
911 BUG_ON(num_refs == 0);
921 delayed_refs = &trans->transaction->delayed_refs;
922 spin_lock(&delayed_refs->lock);
923 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
925 if (!mutex_trylock(&head->mutex)) {
926 refcount_inc(&head->node.refs);
927 spin_unlock(&delayed_refs->lock);
929 btrfs_release_path(path);
932 * Mutex was contended, block until it's released and try
935 mutex_lock(&head->mutex);
936 mutex_unlock(&head->mutex);
937 btrfs_put_delayed_ref(&head->node);
940 spin_lock(&head->lock);
941 if (head->extent_op && head->extent_op->update_flags)
942 extent_flags |= head->extent_op->flags_to_set;
944 BUG_ON(num_refs == 0);
946 num_refs += head->node.ref_mod;
947 spin_unlock(&head->lock);
948 mutex_unlock(&head->mutex);
950 spin_unlock(&delayed_refs->lock);
952 WARN_ON(num_refs == 0);
956 *flags = extent_flags;
958 btrfs_free_path(path);
963 * Back reference rules. Back refs have three main goals:
965 * 1) differentiate between all holders of references to an extent so that
966 * when a reference is dropped we can make sure it was a valid reference
967 * before freeing the extent.
969 * 2) Provide enough information to quickly find the holders of an extent
970 * if we notice a given block is corrupted or bad.
972 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
973 * maintenance. This is actually the same as #2, but with a slightly
974 * different use case.
976 * There are two kinds of back refs. The implicit back refs is optimized
977 * for pointers in non-shared tree blocks. For a given pointer in a block,
978 * back refs of this kind provide information about the block's owner tree
979 * and the pointer's key. These information allow us to find the block by
980 * b-tree searching. The full back refs is for pointers in tree blocks not
981 * referenced by their owner trees. The location of tree block is recorded
982 * in the back refs. Actually the full back refs is generic, and can be
983 * used in all cases the implicit back refs is used. The major shortcoming
984 * of the full back refs is its overhead. Every time a tree block gets
985 * COWed, we have to update back refs entry for all pointers in it.
987 * For a newly allocated tree block, we use implicit back refs for
988 * pointers in it. This means most tree related operations only involve
989 * implicit back refs. For a tree block created in old transaction, the
990 * only way to drop a reference to it is COW it. So we can detect the
991 * event that tree block loses its owner tree's reference and do the
992 * back refs conversion.
994 * When a tree block is COWed through a tree, there are four cases:
996 * The reference count of the block is one and the tree is the block's
997 * owner tree. Nothing to do in this case.
999 * The reference count of the block is one and the tree is not the
1000 * block's owner tree. In this case, full back refs is used for pointers
1001 * in the block. Remove these full back refs, add implicit back refs for
1002 * every pointers in the new block.
1004 * The reference count of the block is greater than one and the tree is
1005 * the block's owner tree. In this case, implicit back refs is used for
1006 * pointers in the block. Add full back refs for every pointers in the
1007 * block, increase lower level extents' reference counts. The original
1008 * implicit back refs are entailed to the new block.
1010 * The reference count of the block is greater than one and the tree is
1011 * not the block's owner tree. Add implicit back refs for every pointer in
1012 * the new block, increase lower level extents' reference count.
1014 * Back Reference Key composing:
1016 * The key objectid corresponds to the first byte in the extent,
1017 * The key type is used to differentiate between types of back refs.
1018 * There are different meanings of the key offset for different types
1021 * File extents can be referenced by:
1023 * - multiple snapshots, subvolumes, or different generations in one subvol
1024 * - different files inside a single subvolume
1025 * - different offsets inside a file (bookend extents in file.c)
1027 * The extent ref structure for the implicit back refs has fields for:
1029 * - Objectid of the subvolume root
1030 * - objectid of the file holding the reference
1031 * - original offset in the file
1032 * - how many bookend extents
1034 * The key offset for the implicit back refs is hash of the first
1037 * The extent ref structure for the full back refs has field for:
1039 * - number of pointers in the tree leaf
1041 * The key offset for the implicit back refs is the first byte of
1044 * When a file extent is allocated, The implicit back refs is used.
1045 * the fields are filled in:
1047 * (root_key.objectid, inode objectid, offset in file, 1)
1049 * When a file extent is removed file truncation, we find the
1050 * corresponding implicit back refs and check the following fields:
1052 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1054 * Btree extents can be referenced by:
1056 * - Different subvolumes
1058 * Both the implicit back refs and the full back refs for tree blocks
1059 * only consist of key. The key offset for the implicit back refs is
1060 * objectid of block's owner tree. The key offset for the full back refs
1061 * is the first byte of parent block.
1063 * When implicit back refs is used, information about the lowest key and
1064 * level of the tree block are required. These information are stored in
1065 * tree block info structure.
1068 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1069 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1070 struct btrfs_fs_info *fs_info,
1071 struct btrfs_path *path,
1072 u64 owner, u32 extra_size)
1074 struct btrfs_root *root = fs_info->extent_root;
1075 struct btrfs_extent_item *item;
1076 struct btrfs_extent_item_v0 *ei0;
1077 struct btrfs_extent_ref_v0 *ref0;
1078 struct btrfs_tree_block_info *bi;
1079 struct extent_buffer *leaf;
1080 struct btrfs_key key;
1081 struct btrfs_key found_key;
1082 u32 new_size = sizeof(*item);
1086 leaf = path->nodes[0];
1087 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1089 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1090 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1091 struct btrfs_extent_item_v0);
1092 refs = btrfs_extent_refs_v0(leaf, ei0);
1094 if (owner == (u64)-1) {
1096 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1097 ret = btrfs_next_leaf(root, path);
1100 BUG_ON(ret > 0); /* Corruption */
1101 leaf = path->nodes[0];
1103 btrfs_item_key_to_cpu(leaf, &found_key,
1105 BUG_ON(key.objectid != found_key.objectid);
1106 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1110 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1111 struct btrfs_extent_ref_v0);
1112 owner = btrfs_ref_objectid_v0(leaf, ref0);
1116 btrfs_release_path(path);
1118 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1119 new_size += sizeof(*bi);
1121 new_size -= sizeof(*ei0);
1122 ret = btrfs_search_slot(trans, root, &key, path,
1123 new_size + extra_size, 1);
1126 BUG_ON(ret); /* Corruption */
1128 btrfs_extend_item(fs_info, path, new_size);
1130 leaf = path->nodes[0];
1131 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1132 btrfs_set_extent_refs(leaf, item, refs);
1133 /* FIXME: get real generation */
1134 btrfs_set_extent_generation(leaf, item, 0);
1135 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1136 btrfs_set_extent_flags(leaf, item,
1137 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1138 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1139 bi = (struct btrfs_tree_block_info *)(item + 1);
1140 /* FIXME: get first key of the block */
1141 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1142 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1144 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1146 btrfs_mark_buffer_dirty(leaf);
1152 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1153 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1154 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1156 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1157 struct btrfs_extent_inline_ref *iref,
1158 enum btrfs_inline_ref_type is_data)
1160 int type = btrfs_extent_inline_ref_type(eb, iref);
1161 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1163 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1164 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1165 type == BTRFS_SHARED_DATA_REF_KEY ||
1166 type == BTRFS_EXTENT_DATA_REF_KEY) {
1167 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1168 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1170 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1171 ASSERT(eb->fs_info);
1173 * Every shared one has parent tree block,
1174 * which must be aligned to sector size.
1177 IS_ALIGNED(offset, eb->fs_info->sectorsize))
1180 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1181 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1183 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1184 ASSERT(eb->fs_info);
1186 * Every shared one has parent tree block,
1187 * which must be aligned to sector size.
1190 IS_ALIGNED(offset, eb->fs_info->sectorsize))
1194 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1199 btrfs_print_leaf((struct extent_buffer *)eb);
1200 btrfs_err(eb->fs_info,
1201 "eb %llu iref 0x%lx invalid extent inline ref type %d",
1202 eb->start, (unsigned long)iref, type);
1205 return BTRFS_REF_TYPE_INVALID;
1208 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1210 u32 high_crc = ~(u32)0;
1211 u32 low_crc = ~(u32)0;
1214 lenum = cpu_to_le64(root_objectid);
1215 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1216 lenum = cpu_to_le64(owner);
1217 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1218 lenum = cpu_to_le64(offset);
1219 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1221 return ((u64)high_crc << 31) ^ (u64)low_crc;
1224 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1225 struct btrfs_extent_data_ref *ref)
1227 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1228 btrfs_extent_data_ref_objectid(leaf, ref),
1229 btrfs_extent_data_ref_offset(leaf, ref));
1232 static int match_extent_data_ref(struct extent_buffer *leaf,
1233 struct btrfs_extent_data_ref *ref,
1234 u64 root_objectid, u64 owner, u64 offset)
1236 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1237 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1238 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1243 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1244 struct btrfs_fs_info *fs_info,
1245 struct btrfs_path *path,
1246 u64 bytenr, u64 parent,
1248 u64 owner, u64 offset)
1250 struct btrfs_root *root = fs_info->extent_root;
1251 struct btrfs_key key;
1252 struct btrfs_extent_data_ref *ref;
1253 struct extent_buffer *leaf;
1259 key.objectid = bytenr;
1261 key.type = BTRFS_SHARED_DATA_REF_KEY;
1262 key.offset = parent;
1264 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1265 key.offset = hash_extent_data_ref(root_objectid,
1270 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1279 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1280 key.type = BTRFS_EXTENT_REF_V0_KEY;
1281 btrfs_release_path(path);
1282 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1293 leaf = path->nodes[0];
1294 nritems = btrfs_header_nritems(leaf);
1296 if (path->slots[0] >= nritems) {
1297 ret = btrfs_next_leaf(root, path);
1303 leaf = path->nodes[0];
1304 nritems = btrfs_header_nritems(leaf);
1308 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1309 if (key.objectid != bytenr ||
1310 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1313 ref = btrfs_item_ptr(leaf, path->slots[0],
1314 struct btrfs_extent_data_ref);
1316 if (match_extent_data_ref(leaf, ref, root_objectid,
1319 btrfs_release_path(path);
1331 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1332 struct btrfs_fs_info *fs_info,
1333 struct btrfs_path *path,
1334 u64 bytenr, u64 parent,
1335 u64 root_objectid, u64 owner,
1336 u64 offset, int refs_to_add)
1338 struct btrfs_root *root = fs_info->extent_root;
1339 struct btrfs_key key;
1340 struct extent_buffer *leaf;
1345 key.objectid = bytenr;
1347 key.type = BTRFS_SHARED_DATA_REF_KEY;
1348 key.offset = parent;
1349 size = sizeof(struct btrfs_shared_data_ref);
1351 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1352 key.offset = hash_extent_data_ref(root_objectid,
1354 size = sizeof(struct btrfs_extent_data_ref);
1357 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1358 if (ret && ret != -EEXIST)
1361 leaf = path->nodes[0];
1363 struct btrfs_shared_data_ref *ref;
1364 ref = btrfs_item_ptr(leaf, path->slots[0],
1365 struct btrfs_shared_data_ref);
1367 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1369 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1370 num_refs += refs_to_add;
1371 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1374 struct btrfs_extent_data_ref *ref;
1375 while (ret == -EEXIST) {
1376 ref = btrfs_item_ptr(leaf, path->slots[0],
1377 struct btrfs_extent_data_ref);
1378 if (match_extent_data_ref(leaf, ref, root_objectid,
1381 btrfs_release_path(path);
1383 ret = btrfs_insert_empty_item(trans, root, path, &key,
1385 if (ret && ret != -EEXIST)
1388 leaf = path->nodes[0];
1390 ref = btrfs_item_ptr(leaf, path->slots[0],
1391 struct btrfs_extent_data_ref);
1393 btrfs_set_extent_data_ref_root(leaf, ref,
1395 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1396 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1397 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1399 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1400 num_refs += refs_to_add;
1401 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1404 btrfs_mark_buffer_dirty(leaf);
1407 btrfs_release_path(path);
1411 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1412 struct btrfs_fs_info *fs_info,
1413 struct btrfs_path *path,
1414 int refs_to_drop, int *last_ref)
1416 struct btrfs_key key;
1417 struct btrfs_extent_data_ref *ref1 = NULL;
1418 struct btrfs_shared_data_ref *ref2 = NULL;
1419 struct extent_buffer *leaf;
1423 leaf = path->nodes[0];
1424 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1426 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1427 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1428 struct btrfs_extent_data_ref);
1429 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1430 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1431 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1432 struct btrfs_shared_data_ref);
1433 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1434 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1435 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1436 struct btrfs_extent_ref_v0 *ref0;
1437 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1438 struct btrfs_extent_ref_v0);
1439 num_refs = btrfs_ref_count_v0(leaf, ref0);
1445 BUG_ON(num_refs < refs_to_drop);
1446 num_refs -= refs_to_drop;
1448 if (num_refs == 0) {
1449 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1452 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1453 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1454 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1455 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1456 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1458 struct btrfs_extent_ref_v0 *ref0;
1459 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1460 struct btrfs_extent_ref_v0);
1461 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1464 btrfs_mark_buffer_dirty(leaf);
1469 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1470 struct btrfs_extent_inline_ref *iref)
1472 struct btrfs_key key;
1473 struct extent_buffer *leaf;
1474 struct btrfs_extent_data_ref *ref1;
1475 struct btrfs_shared_data_ref *ref2;
1479 leaf = path->nodes[0];
1480 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1483 * If type is invalid, we should have bailed out earlier than
1486 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1487 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1488 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1489 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1490 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1492 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1493 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1495 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1496 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1497 struct btrfs_extent_data_ref);
1498 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1499 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1500 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1501 struct btrfs_shared_data_ref);
1502 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1503 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1504 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1505 struct btrfs_extent_ref_v0 *ref0;
1506 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1507 struct btrfs_extent_ref_v0);
1508 num_refs = btrfs_ref_count_v0(leaf, ref0);
1516 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1517 struct btrfs_fs_info *fs_info,
1518 struct btrfs_path *path,
1519 u64 bytenr, u64 parent,
1522 struct btrfs_root *root = fs_info->extent_root;
1523 struct btrfs_key key;
1526 key.objectid = bytenr;
1528 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1529 key.offset = parent;
1531 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1532 key.offset = root_objectid;
1535 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1538 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1539 if (ret == -ENOENT && parent) {
1540 btrfs_release_path(path);
1541 key.type = BTRFS_EXTENT_REF_V0_KEY;
1542 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1550 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1551 struct btrfs_fs_info *fs_info,
1552 struct btrfs_path *path,
1553 u64 bytenr, u64 parent,
1556 struct btrfs_key key;
1559 key.objectid = bytenr;
1561 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1562 key.offset = parent;
1564 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1565 key.offset = root_objectid;
1568 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1570 btrfs_release_path(path);
1574 static inline int extent_ref_type(u64 parent, u64 owner)
1577 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1579 type = BTRFS_SHARED_BLOCK_REF_KEY;
1581 type = BTRFS_TREE_BLOCK_REF_KEY;
1584 type = BTRFS_SHARED_DATA_REF_KEY;
1586 type = BTRFS_EXTENT_DATA_REF_KEY;
1591 static int find_next_key(struct btrfs_path *path, int level,
1592 struct btrfs_key *key)
1595 for (; level < BTRFS_MAX_LEVEL; level++) {
1596 if (!path->nodes[level])
1598 if (path->slots[level] + 1 >=
1599 btrfs_header_nritems(path->nodes[level]))
1602 btrfs_item_key_to_cpu(path->nodes[level], key,
1603 path->slots[level] + 1);
1605 btrfs_node_key_to_cpu(path->nodes[level], key,
1606 path->slots[level] + 1);
1613 * look for inline back ref. if back ref is found, *ref_ret is set
1614 * to the address of inline back ref, and 0 is returned.
1616 * if back ref isn't found, *ref_ret is set to the address where it
1617 * should be inserted, and -ENOENT is returned.
1619 * if insert is true and there are too many inline back refs, the path
1620 * points to the extent item, and -EAGAIN is returned.
1622 * NOTE: inline back refs are ordered in the same way that back ref
1623 * items in the tree are ordered.
1625 static noinline_for_stack
1626 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1627 struct btrfs_fs_info *fs_info,
1628 struct btrfs_path *path,
1629 struct btrfs_extent_inline_ref **ref_ret,
1630 u64 bytenr, u64 num_bytes,
1631 u64 parent, u64 root_objectid,
1632 u64 owner, u64 offset, int insert)
1634 struct btrfs_root *root = fs_info->extent_root;
1635 struct btrfs_key key;
1636 struct extent_buffer *leaf;
1637 struct btrfs_extent_item *ei;
1638 struct btrfs_extent_inline_ref *iref;
1648 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1651 key.objectid = bytenr;
1652 key.type = BTRFS_EXTENT_ITEM_KEY;
1653 key.offset = num_bytes;
1655 want = extent_ref_type(parent, owner);
1657 extra_size = btrfs_extent_inline_ref_size(want);
1658 path->keep_locks = 1;
1663 * Owner is our parent level, so we can just add one to get the level
1664 * for the block we are interested in.
1666 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1667 key.type = BTRFS_METADATA_ITEM_KEY;
1672 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1679 * We may be a newly converted file system which still has the old fat
1680 * extent entries for metadata, so try and see if we have one of those.
1682 if (ret > 0 && skinny_metadata) {
1683 skinny_metadata = false;
1684 if (path->slots[0]) {
1686 btrfs_item_key_to_cpu(path->nodes[0], &key,
1688 if (key.objectid == bytenr &&
1689 key.type == BTRFS_EXTENT_ITEM_KEY &&
1690 key.offset == num_bytes)
1694 key.objectid = bytenr;
1695 key.type = BTRFS_EXTENT_ITEM_KEY;
1696 key.offset = num_bytes;
1697 btrfs_release_path(path);
1702 if (ret && !insert) {
1705 } else if (WARN_ON(ret)) {
1706 btrfs_print_leaf(path->nodes[0]);
1708 "extent item not found for insert, bytenr %llu num_bytes %llu parent %llu root_objectid %llu owner %llu offset %llu",
1709 bytenr, num_bytes, parent, root_objectid, owner,
1715 leaf = path->nodes[0];
1716 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1717 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1718 if (item_size < sizeof(*ei)) {
1723 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1729 leaf = path->nodes[0];
1730 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1733 BUG_ON(item_size < sizeof(*ei));
1735 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1736 flags = btrfs_extent_flags(leaf, ei);
1738 ptr = (unsigned long)(ei + 1);
1739 end = (unsigned long)ei + item_size;
1741 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1742 ptr += sizeof(struct btrfs_tree_block_info);
1746 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1747 needed = BTRFS_REF_TYPE_DATA;
1749 needed = BTRFS_REF_TYPE_BLOCK;
1757 iref = (struct btrfs_extent_inline_ref *)ptr;
1758 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1759 if (type == BTRFS_REF_TYPE_INVALID) {
1767 ptr += btrfs_extent_inline_ref_size(type);
1771 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1772 struct btrfs_extent_data_ref *dref;
1773 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1774 if (match_extent_data_ref(leaf, dref, root_objectid,
1779 if (hash_extent_data_ref_item(leaf, dref) <
1780 hash_extent_data_ref(root_objectid, owner, offset))
1784 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1786 if (parent == ref_offset) {
1790 if (ref_offset < parent)
1793 if (root_objectid == ref_offset) {
1797 if (ref_offset < root_objectid)
1801 ptr += btrfs_extent_inline_ref_size(type);
1803 if (err == -ENOENT && insert) {
1804 if (item_size + extra_size >=
1805 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1810 * To add new inline back ref, we have to make sure
1811 * there is no corresponding back ref item.
1812 * For simplicity, we just do not add new inline back
1813 * ref if there is any kind of item for this block
1815 if (find_next_key(path, 0, &key) == 0 &&
1816 key.objectid == bytenr &&
1817 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1822 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1825 path->keep_locks = 0;
1826 btrfs_unlock_up_safe(path, 1);
1832 * helper to add new inline back ref
1834 static noinline_for_stack
1835 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1836 struct btrfs_path *path,
1837 struct btrfs_extent_inline_ref *iref,
1838 u64 parent, u64 root_objectid,
1839 u64 owner, u64 offset, int refs_to_add,
1840 struct btrfs_delayed_extent_op *extent_op)
1842 struct extent_buffer *leaf;
1843 struct btrfs_extent_item *ei;
1846 unsigned long item_offset;
1851 leaf = path->nodes[0];
1852 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1853 item_offset = (unsigned long)iref - (unsigned long)ei;
1855 type = extent_ref_type(parent, owner);
1856 size = btrfs_extent_inline_ref_size(type);
1858 btrfs_extend_item(fs_info, path, size);
1860 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1861 refs = btrfs_extent_refs(leaf, ei);
1862 refs += refs_to_add;
1863 btrfs_set_extent_refs(leaf, ei, refs);
1865 __run_delayed_extent_op(extent_op, leaf, ei);
1867 ptr = (unsigned long)ei + item_offset;
1868 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1869 if (ptr < end - size)
1870 memmove_extent_buffer(leaf, ptr + size, ptr,
1873 iref = (struct btrfs_extent_inline_ref *)ptr;
1874 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1875 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1876 struct btrfs_extent_data_ref *dref;
1877 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1878 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1879 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1880 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1881 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1882 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1883 struct btrfs_shared_data_ref *sref;
1884 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1885 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1886 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1887 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1888 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1890 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1892 btrfs_mark_buffer_dirty(leaf);
1895 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1896 struct btrfs_fs_info *fs_info,
1897 struct btrfs_path *path,
1898 struct btrfs_extent_inline_ref **ref_ret,
1899 u64 bytenr, u64 num_bytes, u64 parent,
1900 u64 root_objectid, u64 owner, u64 offset)
1904 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1905 bytenr, num_bytes, parent,
1906 root_objectid, owner, offset, 0);
1910 btrfs_release_path(path);
1913 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1914 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1915 parent, root_objectid);
1917 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1918 parent, root_objectid, owner,
1925 * helper to update/remove inline back ref
1927 static noinline_for_stack
1928 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1929 struct btrfs_path *path,
1930 struct btrfs_extent_inline_ref *iref,
1932 struct btrfs_delayed_extent_op *extent_op,
1935 struct extent_buffer *leaf;
1936 struct btrfs_extent_item *ei;
1937 struct btrfs_extent_data_ref *dref = NULL;
1938 struct btrfs_shared_data_ref *sref = NULL;
1946 leaf = path->nodes[0];
1947 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1948 refs = btrfs_extent_refs(leaf, ei);
1949 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1950 refs += refs_to_mod;
1951 btrfs_set_extent_refs(leaf, ei, refs);
1953 __run_delayed_extent_op(extent_op, leaf, ei);
1956 * If type is invalid, we should have bailed out after
1957 * lookup_inline_extent_backref().
1959 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1960 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1962 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1963 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1964 refs = btrfs_extent_data_ref_count(leaf, dref);
1965 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1966 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1967 refs = btrfs_shared_data_ref_count(leaf, sref);
1970 BUG_ON(refs_to_mod != -1);
1973 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1974 refs += refs_to_mod;
1977 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1978 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1980 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1983 size = btrfs_extent_inline_ref_size(type);
1984 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1985 ptr = (unsigned long)iref;
1986 end = (unsigned long)ei + item_size;
1987 if (ptr + size < end)
1988 memmove_extent_buffer(leaf, ptr, ptr + size,
1991 btrfs_truncate_item(fs_info, path, item_size, 1);
1993 btrfs_mark_buffer_dirty(leaf);
1996 static noinline_for_stack
1997 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1998 struct btrfs_fs_info *fs_info,
1999 struct btrfs_path *path,
2000 u64 bytenr, u64 num_bytes, u64 parent,
2001 u64 root_objectid, u64 owner,
2002 u64 offset, int refs_to_add,
2003 struct btrfs_delayed_extent_op *extent_op)
2005 struct btrfs_extent_inline_ref *iref;
2008 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
2009 bytenr, num_bytes, parent,
2010 root_objectid, owner, offset, 1);
2012 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
2013 update_inline_extent_backref(fs_info, path, iref,
2014 refs_to_add, extent_op, NULL);
2015 } else if (ret == -ENOENT) {
2016 setup_inline_extent_backref(fs_info, path, iref, parent,
2017 root_objectid, owner, offset,
2018 refs_to_add, extent_op);
2024 static int insert_extent_backref(struct btrfs_trans_handle *trans,
2025 struct btrfs_fs_info *fs_info,
2026 struct btrfs_path *path,
2027 u64 bytenr, u64 parent, u64 root_objectid,
2028 u64 owner, u64 offset, int refs_to_add)
2031 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2032 BUG_ON(refs_to_add != 1);
2033 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
2034 parent, root_objectid);
2036 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
2037 parent, root_objectid,
2038 owner, offset, refs_to_add);
2043 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2044 struct btrfs_fs_info *fs_info,
2045 struct btrfs_path *path,
2046 struct btrfs_extent_inline_ref *iref,
2047 int refs_to_drop, int is_data, int *last_ref)
2051 BUG_ON(!is_data && refs_to_drop != 1);
2053 update_inline_extent_backref(fs_info, path, iref,
2054 -refs_to_drop, NULL, last_ref);
2055 } else if (is_data) {
2056 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
2060 ret = btrfs_del_item(trans, fs_info->extent_root, path);
2065 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2066 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2067 u64 *discarded_bytes)
2070 u64 bytes_left, end;
2071 u64 aligned_start = ALIGN(start, 1 << 9);
2073 if (WARN_ON(start != aligned_start)) {
2074 len -= aligned_start - start;
2075 len = round_down(len, 1 << 9);
2076 start = aligned_start;
2079 *discarded_bytes = 0;
2087 /* Skip any superblocks on this device. */
2088 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2089 u64 sb_start = btrfs_sb_offset(j);
2090 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2091 u64 size = sb_start - start;
2093 if (!in_range(sb_start, start, bytes_left) &&
2094 !in_range(sb_end, start, bytes_left) &&
2095 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2099 * Superblock spans beginning of range. Adjust start and
2102 if (sb_start <= start) {
2103 start += sb_end - start;
2108 bytes_left = end - start;
2113 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2116 *discarded_bytes += size;
2117 else if (ret != -EOPNOTSUPP)
2126 bytes_left = end - start;
2130 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2133 *discarded_bytes += bytes_left;
2138 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2139 u64 num_bytes, u64 *actual_bytes)
2142 u64 discarded_bytes = 0;
2143 struct btrfs_bio *bbio = NULL;
2147 * Avoid races with device replace and make sure our bbio has devices
2148 * associated to its stripes that don't go away while we are discarding.
2150 btrfs_bio_counter_inc_blocked(fs_info);
2151 /* Tell the block device(s) that the sectors can be discarded */
2152 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2154 /* Error condition is -ENOMEM */
2156 struct btrfs_bio_stripe *stripe = bbio->stripes;
2160 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2162 if (!stripe->dev->can_discard)
2165 ret = btrfs_issue_discard(stripe->dev->bdev,
2170 discarded_bytes += bytes;
2171 else if (ret != -EOPNOTSUPP)
2172 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2175 * Just in case we get back EOPNOTSUPP for some reason,
2176 * just ignore the return value so we don't screw up
2177 * people calling discard_extent.
2181 btrfs_put_bbio(bbio);
2183 btrfs_bio_counter_dec(fs_info);
2186 *actual_bytes = discarded_bytes;
2189 if (ret == -EOPNOTSUPP)
2194 /* Can return -ENOMEM */
2195 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2196 struct btrfs_fs_info *fs_info,
2197 u64 bytenr, u64 num_bytes, u64 parent,
2198 u64 root_objectid, u64 owner, u64 offset)
2200 int old_ref_mod, new_ref_mod;
2203 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2204 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2206 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2207 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2209 root_objectid, (int)owner,
2210 BTRFS_ADD_DELAYED_REF, NULL,
2211 &old_ref_mod, &new_ref_mod);
2213 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2215 root_objectid, owner, offset,
2216 0, BTRFS_ADD_DELAYED_REF,
2217 &old_ref_mod, &new_ref_mod);
2220 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2221 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2226 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2227 struct btrfs_fs_info *fs_info,
2228 struct btrfs_delayed_ref_node *node,
2229 u64 parent, u64 root_objectid,
2230 u64 owner, u64 offset, int refs_to_add,
2231 struct btrfs_delayed_extent_op *extent_op)
2233 struct btrfs_path *path;
2234 struct extent_buffer *leaf;
2235 struct btrfs_extent_item *item;
2236 struct btrfs_key key;
2237 u64 bytenr = node->bytenr;
2238 u64 num_bytes = node->num_bytes;
2242 path = btrfs_alloc_path();
2246 path->reada = READA_FORWARD;
2247 path->leave_spinning = 1;
2248 /* this will setup the path even if it fails to insert the back ref */
2249 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2250 num_bytes, parent, root_objectid,
2252 refs_to_add, extent_op);
2253 if ((ret < 0 && ret != -EAGAIN) || !ret)
2257 * Ok we had -EAGAIN which means we didn't have space to insert and
2258 * inline extent ref, so just update the reference count and add a
2261 leaf = path->nodes[0];
2262 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2263 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2264 refs = btrfs_extent_refs(leaf, item);
2265 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2267 __run_delayed_extent_op(extent_op, leaf, item);
2269 btrfs_mark_buffer_dirty(leaf);
2270 btrfs_release_path(path);
2272 path->reada = READA_FORWARD;
2273 path->leave_spinning = 1;
2274 /* now insert the actual backref */
2275 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2276 root_objectid, owner, offset, refs_to_add);
2278 btrfs_abort_transaction(trans, ret);
2280 btrfs_free_path(path);
2284 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2285 struct btrfs_fs_info *fs_info,
2286 struct btrfs_delayed_ref_node *node,
2287 struct btrfs_delayed_extent_op *extent_op,
2288 int insert_reserved)
2291 struct btrfs_delayed_data_ref *ref;
2292 struct btrfs_key ins;
2297 ins.objectid = node->bytenr;
2298 ins.offset = node->num_bytes;
2299 ins.type = BTRFS_EXTENT_ITEM_KEY;
2301 ref = btrfs_delayed_node_to_data_ref(node);
2302 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2304 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2305 parent = ref->parent;
2306 ref_root = ref->root;
2308 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2310 flags |= extent_op->flags_to_set;
2311 ret = alloc_reserved_file_extent(trans, fs_info,
2312 parent, ref_root, flags,
2313 ref->objectid, ref->offset,
2314 &ins, node->ref_mod);
2315 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2316 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2317 ref_root, ref->objectid,
2318 ref->offset, node->ref_mod,
2320 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2321 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2322 ref_root, ref->objectid,
2323 ref->offset, node->ref_mod,
2331 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2332 struct extent_buffer *leaf,
2333 struct btrfs_extent_item *ei)
2335 u64 flags = btrfs_extent_flags(leaf, ei);
2336 if (extent_op->update_flags) {
2337 flags |= extent_op->flags_to_set;
2338 btrfs_set_extent_flags(leaf, ei, flags);
2341 if (extent_op->update_key) {
2342 struct btrfs_tree_block_info *bi;
2343 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2344 bi = (struct btrfs_tree_block_info *)(ei + 1);
2345 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2349 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2350 struct btrfs_fs_info *fs_info,
2351 struct btrfs_delayed_ref_node *node,
2352 struct btrfs_delayed_extent_op *extent_op)
2354 struct btrfs_key key;
2355 struct btrfs_path *path;
2356 struct btrfs_extent_item *ei;
2357 struct extent_buffer *leaf;
2361 int metadata = !extent_op->is_data;
2366 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2369 path = btrfs_alloc_path();
2373 key.objectid = node->bytenr;
2376 key.type = BTRFS_METADATA_ITEM_KEY;
2377 key.offset = extent_op->level;
2379 key.type = BTRFS_EXTENT_ITEM_KEY;
2380 key.offset = node->num_bytes;
2384 path->reada = READA_FORWARD;
2385 path->leave_spinning = 1;
2386 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2393 if (path->slots[0] > 0) {
2395 btrfs_item_key_to_cpu(path->nodes[0], &key,
2397 if (key.objectid == node->bytenr &&
2398 key.type == BTRFS_EXTENT_ITEM_KEY &&
2399 key.offset == node->num_bytes)
2403 btrfs_release_path(path);
2406 key.objectid = node->bytenr;
2407 key.offset = node->num_bytes;
2408 key.type = BTRFS_EXTENT_ITEM_KEY;
2417 leaf = path->nodes[0];
2418 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2419 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2420 if (item_size < sizeof(*ei)) {
2421 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2426 leaf = path->nodes[0];
2427 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2430 BUG_ON(item_size < sizeof(*ei));
2431 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2432 __run_delayed_extent_op(extent_op, leaf, ei);
2434 btrfs_mark_buffer_dirty(leaf);
2436 btrfs_free_path(path);
2440 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2441 struct btrfs_fs_info *fs_info,
2442 struct btrfs_delayed_ref_node *node,
2443 struct btrfs_delayed_extent_op *extent_op,
2444 int insert_reserved)
2447 struct btrfs_delayed_tree_ref *ref;
2448 struct btrfs_key ins;
2451 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2453 ref = btrfs_delayed_node_to_tree_ref(node);
2454 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2456 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2457 parent = ref->parent;
2458 ref_root = ref->root;
2460 ins.objectid = node->bytenr;
2461 if (skinny_metadata) {
2462 ins.offset = ref->level;
2463 ins.type = BTRFS_METADATA_ITEM_KEY;
2465 ins.offset = node->num_bytes;
2466 ins.type = BTRFS_EXTENT_ITEM_KEY;
2469 if (node->ref_mod != 1) {
2471 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2472 node->bytenr, node->ref_mod, node->action, ref_root,
2476 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2477 BUG_ON(!extent_op || !extent_op->update_flags);
2478 ret = alloc_reserved_tree_block(trans, fs_info,
2480 extent_op->flags_to_set,
2483 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2484 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2488 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2489 ret = __btrfs_free_extent(trans, fs_info, node,
2491 ref->level, 0, 1, extent_op);
2498 /* helper function to actually process a single delayed ref entry */
2499 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2500 struct btrfs_fs_info *fs_info,
2501 struct btrfs_delayed_ref_node *node,
2502 struct btrfs_delayed_extent_op *extent_op,
2503 int insert_reserved)
2507 if (trans->aborted) {
2508 if (insert_reserved)
2509 btrfs_pin_extent(fs_info, node->bytenr,
2510 node->num_bytes, 1);
2514 if (btrfs_delayed_ref_is_head(node)) {
2515 struct btrfs_delayed_ref_head *head;
2517 * we've hit the end of the chain and we were supposed
2518 * to insert this extent into the tree. But, it got
2519 * deleted before we ever needed to insert it, so all
2520 * we have to do is clean up the accounting
2523 head = btrfs_delayed_node_to_head(node);
2524 trace_run_delayed_ref_head(fs_info, node, head, node->action);
2526 if (head->total_ref_mod < 0) {
2527 struct btrfs_block_group_cache *cache;
2529 cache = btrfs_lookup_block_group(fs_info, node->bytenr);
2531 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2533 btrfs_put_block_group(cache);
2536 if (insert_reserved) {
2537 btrfs_pin_extent(fs_info, node->bytenr,
2538 node->num_bytes, 1);
2539 if (head->is_data) {
2540 ret = btrfs_del_csums(trans, fs_info,
2546 /* Also free its reserved qgroup space */
2547 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2548 head->qgroup_reserved);
2552 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2553 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2554 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2556 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2557 node->type == BTRFS_SHARED_DATA_REF_KEY)
2558 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2565 static inline struct btrfs_delayed_ref_node *
2566 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2568 struct btrfs_delayed_ref_node *ref;
2570 if (list_empty(&head->ref_list))
2574 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2575 * This is to prevent a ref count from going down to zero, which deletes
2576 * the extent item from the extent tree, when there still are references
2577 * to add, which would fail because they would not find the extent item.
2579 if (!list_empty(&head->ref_add_list))
2580 return list_first_entry(&head->ref_add_list,
2581 struct btrfs_delayed_ref_node, add_list);
2583 ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
2585 ASSERT(list_empty(&ref->add_list));
2590 * Returns 0 on success or if called with an already aborted transaction.
2591 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2593 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2594 struct btrfs_fs_info *fs_info,
2597 struct btrfs_delayed_ref_root *delayed_refs;
2598 struct btrfs_delayed_ref_node *ref;
2599 struct btrfs_delayed_ref_head *locked_ref = NULL;
2600 struct btrfs_delayed_extent_op *extent_op;
2601 ktime_t start = ktime_get();
2603 unsigned long count = 0;
2604 unsigned long actual_count = 0;
2605 int must_insert_reserved = 0;
2607 delayed_refs = &trans->transaction->delayed_refs;
2613 spin_lock(&delayed_refs->lock);
2614 locked_ref = btrfs_select_ref_head(trans);
2616 spin_unlock(&delayed_refs->lock);
2620 /* grab the lock that says we are going to process
2621 * all the refs for this head */
2622 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2623 spin_unlock(&delayed_refs->lock);
2625 * we may have dropped the spin lock to get the head
2626 * mutex lock, and that might have given someone else
2627 * time to free the head. If that's true, it has been
2628 * removed from our list and we can move on.
2630 if (ret == -EAGAIN) {
2638 * We need to try and merge add/drops of the same ref since we
2639 * can run into issues with relocate dropping the implicit ref
2640 * and then it being added back again before the drop can
2641 * finish. If we merged anything we need to re-loop so we can
2643 * Or we can get node references of the same type that weren't
2644 * merged when created due to bumps in the tree mod seq, and
2645 * we need to merge them to prevent adding an inline extent
2646 * backref before dropping it (triggering a BUG_ON at
2647 * insert_inline_extent_backref()).
2649 spin_lock(&locked_ref->lock);
2650 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2654 * locked_ref is the head node, so we have to go one
2655 * node back for any delayed ref updates
2657 ref = select_delayed_ref(locked_ref);
2659 if (ref && ref->seq &&
2660 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2661 spin_unlock(&locked_ref->lock);
2662 spin_lock(&delayed_refs->lock);
2663 locked_ref->processing = 0;
2664 delayed_refs->num_heads_ready++;
2665 spin_unlock(&delayed_refs->lock);
2666 btrfs_delayed_ref_unlock(locked_ref);
2674 * record the must insert reserved flag before we
2675 * drop the spin lock.
2677 must_insert_reserved = locked_ref->must_insert_reserved;
2678 locked_ref->must_insert_reserved = 0;
2680 extent_op = locked_ref->extent_op;
2681 locked_ref->extent_op = NULL;
2686 /* All delayed refs have been processed, Go ahead
2687 * and send the head node to run_one_delayed_ref,
2688 * so that any accounting fixes can happen
2690 ref = &locked_ref->node;
2692 if (extent_op && must_insert_reserved) {
2693 btrfs_free_delayed_extent_op(extent_op);
2698 spin_unlock(&locked_ref->lock);
2699 ret = run_delayed_extent_op(trans, fs_info,
2701 btrfs_free_delayed_extent_op(extent_op);
2705 * Need to reset must_insert_reserved if
2706 * there was an error so the abort stuff
2707 * can cleanup the reserved space
2710 if (must_insert_reserved)
2711 locked_ref->must_insert_reserved = 1;
2712 spin_lock(&delayed_refs->lock);
2713 locked_ref->processing = 0;
2714 delayed_refs->num_heads_ready++;
2715 spin_unlock(&delayed_refs->lock);
2716 btrfs_debug(fs_info,
2717 "run_delayed_extent_op returned %d",
2719 btrfs_delayed_ref_unlock(locked_ref);
2726 * Need to drop our head ref lock and re-acquire the
2727 * delayed ref lock and then re-check to make sure
2730 spin_unlock(&locked_ref->lock);
2731 spin_lock(&delayed_refs->lock);
2732 spin_lock(&locked_ref->lock);
2733 if (!list_empty(&locked_ref->ref_list) ||
2734 locked_ref->extent_op) {
2735 spin_unlock(&locked_ref->lock);
2736 spin_unlock(&delayed_refs->lock);
2740 delayed_refs->num_heads--;
2741 rb_erase(&locked_ref->href_node,
2742 &delayed_refs->href_root);
2743 spin_unlock(&delayed_refs->lock);
2747 list_del(&ref->list);
2748 if (!list_empty(&ref->add_list))
2749 list_del(&ref->add_list);
2751 atomic_dec(&delayed_refs->num_entries);
2753 if (!btrfs_delayed_ref_is_head(ref)) {
2755 * when we play the delayed ref, also correct the
2758 switch (ref->action) {
2759 case BTRFS_ADD_DELAYED_REF:
2760 case BTRFS_ADD_DELAYED_EXTENT:
2761 locked_ref->node.ref_mod -= ref->ref_mod;
2763 case BTRFS_DROP_DELAYED_REF:
2764 locked_ref->node.ref_mod += ref->ref_mod;
2770 spin_unlock(&locked_ref->lock);
2772 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2773 must_insert_reserved);
2775 btrfs_free_delayed_extent_op(extent_op);
2777 spin_lock(&delayed_refs->lock);
2778 locked_ref->processing = 0;
2779 delayed_refs->num_heads_ready++;
2780 spin_unlock(&delayed_refs->lock);
2781 btrfs_delayed_ref_unlock(locked_ref);
2782 btrfs_put_delayed_ref(ref);
2783 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2789 * If this node is a head, that means all the refs in this head
2790 * have been dealt with, and we will pick the next head to deal
2791 * with, so we must unlock the head and drop it from the cluster
2792 * list before we release it.
2794 if (btrfs_delayed_ref_is_head(ref)) {
2795 if (locked_ref->is_data &&
2796 locked_ref->total_ref_mod < 0) {
2797 spin_lock(&delayed_refs->lock);
2798 delayed_refs->pending_csums -= ref->num_bytes;
2799 spin_unlock(&delayed_refs->lock);
2801 btrfs_delayed_ref_unlock(locked_ref);
2804 btrfs_put_delayed_ref(ref);
2810 * We don't want to include ref heads since we can have empty ref heads
2811 * and those will drastically skew our runtime down since we just do
2812 * accounting, no actual extent tree updates.
2814 if (actual_count > 0) {
2815 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2819 * We weigh the current average higher than our current runtime
2820 * to avoid large swings in the average.
2822 spin_lock(&delayed_refs->lock);
2823 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2824 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2825 spin_unlock(&delayed_refs->lock);
2830 #ifdef SCRAMBLE_DELAYED_REFS
2832 * Normally delayed refs get processed in ascending bytenr order. This
2833 * correlates in most cases to the order added. To expose dependencies on this
2834 * order, we start to process the tree in the middle instead of the beginning
2836 static u64 find_middle(struct rb_root *root)
2838 struct rb_node *n = root->rb_node;
2839 struct btrfs_delayed_ref_node *entry;
2842 u64 first = 0, last = 0;
2846 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2847 first = entry->bytenr;
2851 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2852 last = entry->bytenr;
2857 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2858 WARN_ON(!entry->in_tree);
2860 middle = entry->bytenr;
2873 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2877 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2878 sizeof(struct btrfs_extent_inline_ref));
2879 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2880 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2883 * We don't ever fill up leaves all the way so multiply by 2 just to be
2884 * closer to what we're really going to want to use.
2886 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2890 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2891 * would require to store the csums for that many bytes.
2893 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2896 u64 num_csums_per_leaf;
2899 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2900 num_csums_per_leaf = div64_u64(csum_size,
2901 (u64)btrfs_super_csum_size(fs_info->super_copy));
2902 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2903 num_csums += num_csums_per_leaf - 1;
2904 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2908 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2909 struct btrfs_fs_info *fs_info)
2911 struct btrfs_block_rsv *global_rsv;
2912 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2913 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2914 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2915 u64 num_bytes, num_dirty_bgs_bytes;
2918 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2919 num_heads = heads_to_leaves(fs_info, num_heads);
2921 num_bytes += (num_heads - 1) * fs_info->nodesize;
2923 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2925 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2927 global_rsv = &fs_info->global_block_rsv;
2930 * If we can't allocate any more chunks lets make sure we have _lots_ of
2931 * wiggle room since running delayed refs can create more delayed refs.
2933 if (global_rsv->space_info->full) {
2934 num_dirty_bgs_bytes <<= 1;
2938 spin_lock(&global_rsv->lock);
2939 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2941 spin_unlock(&global_rsv->lock);
2945 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2946 struct btrfs_fs_info *fs_info)
2949 atomic_read(&trans->transaction->delayed_refs.num_entries);
2954 avg_runtime = fs_info->avg_delayed_ref_runtime;
2955 val = num_entries * avg_runtime;
2956 if (val >= NSEC_PER_SEC)
2958 if (val >= NSEC_PER_SEC / 2)
2961 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2964 struct async_delayed_refs {
2965 struct btrfs_root *root;
2970 struct completion wait;
2971 struct btrfs_work work;
2974 static inline struct async_delayed_refs *
2975 to_async_delayed_refs(struct btrfs_work *work)
2977 return container_of(work, struct async_delayed_refs, work);
2980 static void delayed_ref_async_start(struct btrfs_work *work)
2982 struct async_delayed_refs *async = to_async_delayed_refs(work);
2983 struct btrfs_trans_handle *trans;
2984 struct btrfs_fs_info *fs_info = async->root->fs_info;
2987 /* if the commit is already started, we don't need to wait here */
2988 if (btrfs_transaction_blocked(fs_info))
2991 trans = btrfs_join_transaction(async->root);
2992 if (IS_ERR(trans)) {
2993 async->error = PTR_ERR(trans);
2998 * trans->sync means that when we call end_transaction, we won't
2999 * wait on delayed refs
3003 /* Don't bother flushing if we got into a different transaction */
3004 if (trans->transid > async->transid)
3007 ret = btrfs_run_delayed_refs(trans, fs_info, async->count);
3011 ret = btrfs_end_transaction(trans);
3012 if (ret && !async->error)
3016 complete(&async->wait);
3021 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
3022 unsigned long count, u64 transid, int wait)
3024 struct async_delayed_refs *async;
3027 async = kmalloc(sizeof(*async), GFP_NOFS);
3031 async->root = fs_info->tree_root;
3032 async->count = count;
3034 async->transid = transid;
3039 init_completion(&async->wait);
3041 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3042 delayed_ref_async_start, NULL, NULL);
3044 btrfs_queue_work(fs_info->extent_workers, &async->work);
3047 wait_for_completion(&async->wait);
3056 * this starts processing the delayed reference count updates and
3057 * extent insertions we have queued up so far. count can be
3058 * 0, which means to process everything in the tree at the start
3059 * of the run (but not newly added entries), or it can be some target
3060 * number you'd like to process.
3062 * Returns 0 on success or if called with an aborted transaction
3063 * Returns <0 on error and aborts the transaction
3065 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3066 struct btrfs_fs_info *fs_info, unsigned long count)
3068 struct rb_node *node;
3069 struct btrfs_delayed_ref_root *delayed_refs;
3070 struct btrfs_delayed_ref_head *head;
3072 int run_all = count == (unsigned long)-1;
3073 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3075 /* We'll clean this up in btrfs_cleanup_transaction */
3079 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3082 delayed_refs = &trans->transaction->delayed_refs;
3084 count = atomic_read(&delayed_refs->num_entries) * 2;
3087 #ifdef SCRAMBLE_DELAYED_REFS
3088 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3090 trans->can_flush_pending_bgs = false;
3091 ret = __btrfs_run_delayed_refs(trans, fs_info, count);
3093 btrfs_abort_transaction(trans, ret);
3098 if (!list_empty(&trans->new_bgs))
3099 btrfs_create_pending_block_groups(trans, fs_info);
3101 spin_lock(&delayed_refs->lock);
3102 node = rb_first(&delayed_refs->href_root);
3104 spin_unlock(&delayed_refs->lock);
3109 head = rb_entry(node, struct btrfs_delayed_ref_head,
3111 if (btrfs_delayed_ref_is_head(&head->node)) {
3112 struct btrfs_delayed_ref_node *ref;
3115 refcount_inc(&ref->refs);
3117 spin_unlock(&delayed_refs->lock);
3119 * Mutex was contended, block until it's
3120 * released and try again
3122 mutex_lock(&head->mutex);
3123 mutex_unlock(&head->mutex);
3125 btrfs_put_delayed_ref(ref);
3131 node = rb_next(node);
3133 spin_unlock(&delayed_refs->lock);
3138 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3142 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3143 struct btrfs_fs_info *fs_info,
3144 u64 bytenr, u64 num_bytes, u64 flags,
3145 int level, int is_data)
3147 struct btrfs_delayed_extent_op *extent_op;
3150 extent_op = btrfs_alloc_delayed_extent_op();
3154 extent_op->flags_to_set = flags;
3155 extent_op->update_flags = true;
3156 extent_op->update_key = false;
3157 extent_op->is_data = is_data ? true : false;
3158 extent_op->level = level;
3160 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3161 num_bytes, extent_op);
3163 btrfs_free_delayed_extent_op(extent_op);
3167 static noinline int check_delayed_ref(struct btrfs_root *root,
3168 struct btrfs_path *path,
3169 u64 objectid, u64 offset, u64 bytenr)
3171 struct btrfs_delayed_ref_head *head;
3172 struct btrfs_delayed_ref_node *ref;
3173 struct btrfs_delayed_data_ref *data_ref;
3174 struct btrfs_delayed_ref_root *delayed_refs;
3175 struct btrfs_transaction *cur_trans;
3178 spin_lock(&root->fs_info->trans_lock);
3179 cur_trans = root->fs_info->running_transaction;
3181 refcount_inc(&cur_trans->use_count);
3182 spin_unlock(&root->fs_info->trans_lock);
3186 delayed_refs = &cur_trans->delayed_refs;
3187 spin_lock(&delayed_refs->lock);
3188 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3190 spin_unlock(&delayed_refs->lock);
3191 btrfs_put_transaction(cur_trans);
3195 if (!mutex_trylock(&head->mutex)) {
3196 refcount_inc(&head->node.refs);
3197 spin_unlock(&delayed_refs->lock);
3199 btrfs_release_path(path);
3202 * Mutex was contended, block until it's released and let
3205 mutex_lock(&head->mutex);
3206 mutex_unlock(&head->mutex);
3207 btrfs_put_delayed_ref(&head->node);
3208 btrfs_put_transaction(cur_trans);
3211 spin_unlock(&delayed_refs->lock);
3213 spin_lock(&head->lock);
3214 list_for_each_entry(ref, &head->ref_list, list) {
3215 /* If it's a shared ref we know a cross reference exists */
3216 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3221 data_ref = btrfs_delayed_node_to_data_ref(ref);
3224 * If our ref doesn't match the one we're currently looking at
3225 * then we have a cross reference.
3227 if (data_ref->root != root->root_key.objectid ||
3228 data_ref->objectid != objectid ||
3229 data_ref->offset != offset) {
3234 spin_unlock(&head->lock);
3235 mutex_unlock(&head->mutex);
3236 btrfs_put_transaction(cur_trans);
3240 static noinline int check_committed_ref(struct btrfs_root *root,
3241 struct btrfs_path *path,
3242 u64 objectid, u64 offset, u64 bytenr)
3244 struct btrfs_fs_info *fs_info = root->fs_info;
3245 struct btrfs_root *extent_root = fs_info->extent_root;
3246 struct extent_buffer *leaf;
3247 struct btrfs_extent_data_ref *ref;
3248 struct btrfs_extent_inline_ref *iref;
3249 struct btrfs_extent_item *ei;
3250 struct btrfs_key key;
3255 key.objectid = bytenr;
3256 key.offset = (u64)-1;
3257 key.type = BTRFS_EXTENT_ITEM_KEY;
3259 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3262 BUG_ON(ret == 0); /* Corruption */
3265 if (path->slots[0] == 0)
3269 leaf = path->nodes[0];
3270 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3272 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3276 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3277 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3278 if (item_size < sizeof(*ei)) {
3279 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3283 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3285 if (item_size != sizeof(*ei) +
3286 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3289 if (btrfs_extent_generation(leaf, ei) <=
3290 btrfs_root_last_snapshot(&root->root_item))
3293 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3295 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3296 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3299 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3300 if (btrfs_extent_refs(leaf, ei) !=
3301 btrfs_extent_data_ref_count(leaf, ref) ||
3302 btrfs_extent_data_ref_root(leaf, ref) !=
3303 root->root_key.objectid ||
3304 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3305 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3313 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3316 struct btrfs_path *path;
3320 path = btrfs_alloc_path();
3325 ret = check_committed_ref(root, path, objectid,
3327 if (ret && ret != -ENOENT)
3330 ret2 = check_delayed_ref(root, path, objectid,
3332 } while (ret2 == -EAGAIN);
3334 if (ret2 && ret2 != -ENOENT) {
3339 if (ret != -ENOENT || ret2 != -ENOENT)
3342 btrfs_free_path(path);
3343 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3348 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3349 struct btrfs_root *root,
3350 struct extent_buffer *buf,
3351 int full_backref, int inc)
3353 struct btrfs_fs_info *fs_info = root->fs_info;
3359 struct btrfs_key key;
3360 struct btrfs_file_extent_item *fi;
3364 int (*process_func)(struct btrfs_trans_handle *,
3365 struct btrfs_fs_info *,
3366 u64, u64, u64, u64, u64, u64);
3369 if (btrfs_is_testing(fs_info))
3372 ref_root = btrfs_header_owner(buf);
3373 nritems = btrfs_header_nritems(buf);
3374 level = btrfs_header_level(buf);
3376 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3380 process_func = btrfs_inc_extent_ref;
3382 process_func = btrfs_free_extent;
3385 parent = buf->start;
3389 for (i = 0; i < nritems; i++) {
3391 btrfs_item_key_to_cpu(buf, &key, i);
3392 if (key.type != BTRFS_EXTENT_DATA_KEY)
3394 fi = btrfs_item_ptr(buf, i,
3395 struct btrfs_file_extent_item);
3396 if (btrfs_file_extent_type(buf, fi) ==
3397 BTRFS_FILE_EXTENT_INLINE)
3399 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3403 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3404 key.offset -= btrfs_file_extent_offset(buf, fi);
3405 ret = process_func(trans, fs_info, bytenr, num_bytes,
3406 parent, ref_root, key.objectid,
3411 bytenr = btrfs_node_blockptr(buf, i);
3412 num_bytes = fs_info->nodesize;
3413 ret = process_func(trans, fs_info, bytenr, num_bytes,
3414 parent, ref_root, level - 1, 0);
3424 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3425 struct extent_buffer *buf, int full_backref)
3427 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3430 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3431 struct extent_buffer *buf, int full_backref)
3433 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3436 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3437 struct btrfs_fs_info *fs_info,
3438 struct btrfs_path *path,
3439 struct btrfs_block_group_cache *cache)
3442 struct btrfs_root *extent_root = fs_info->extent_root;
3444 struct extent_buffer *leaf;
3446 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3453 leaf = path->nodes[0];
3454 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3455 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3456 btrfs_mark_buffer_dirty(leaf);
3458 btrfs_release_path(path);
3463 static struct btrfs_block_group_cache *
3464 next_block_group(struct btrfs_fs_info *fs_info,
3465 struct btrfs_block_group_cache *cache)
3467 struct rb_node *node;
3469 spin_lock(&fs_info->block_group_cache_lock);
3471 /* If our block group was removed, we need a full search. */
3472 if (RB_EMPTY_NODE(&cache->cache_node)) {
3473 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3475 spin_unlock(&fs_info->block_group_cache_lock);
3476 btrfs_put_block_group(cache);
3477 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3479 node = rb_next(&cache->cache_node);
3480 btrfs_put_block_group(cache);
3482 cache = rb_entry(node, struct btrfs_block_group_cache,
3484 btrfs_get_block_group(cache);
3487 spin_unlock(&fs_info->block_group_cache_lock);
3491 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3492 struct btrfs_trans_handle *trans,
3493 struct btrfs_path *path)
3495 struct btrfs_fs_info *fs_info = block_group->fs_info;
3496 struct btrfs_root *root = fs_info->tree_root;
3497 struct inode *inode = NULL;
3498 struct extent_changeset *data_reserved = NULL;
3500 int dcs = BTRFS_DC_ERROR;
3506 * If this block group is smaller than 100 megs don't bother caching the
3509 if (block_group->key.offset < (100 * SZ_1M)) {
3510 spin_lock(&block_group->lock);
3511 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3512 spin_unlock(&block_group->lock);
3519 inode = lookup_free_space_inode(fs_info, block_group, path);
3520 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3521 ret = PTR_ERR(inode);
3522 btrfs_release_path(path);
3526 if (IS_ERR(inode)) {
3530 if (block_group->ro)
3533 ret = create_free_space_inode(fs_info, trans, block_group,
3541 * We want to set the generation to 0, that way if anything goes wrong
3542 * from here on out we know not to trust this cache when we load up next
3545 BTRFS_I(inode)->generation = 0;
3546 ret = btrfs_update_inode(trans, root, inode);
3549 * So theoretically we could recover from this, simply set the
3550 * super cache generation to 0 so we know to invalidate the
3551 * cache, but then we'd have to keep track of the block groups
3552 * that fail this way so we know we _have_ to reset this cache
3553 * before the next commit or risk reading stale cache. So to
3554 * limit our exposure to horrible edge cases lets just abort the
3555 * transaction, this only happens in really bad situations
3558 btrfs_abort_transaction(trans, ret);
3563 /* We've already setup this transaction, go ahead and exit */
3564 if (block_group->cache_generation == trans->transid &&
3565 i_size_read(inode)) {
3566 dcs = BTRFS_DC_SETUP;
3570 if (i_size_read(inode) > 0) {
3571 ret = btrfs_check_trunc_cache_free_space(fs_info,
3572 &fs_info->global_block_rsv);
3576 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3581 spin_lock(&block_group->lock);
3582 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3583 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3585 * don't bother trying to write stuff out _if_
3586 * a) we're not cached,
3587 * b) we're with nospace_cache mount option,
3588 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3590 dcs = BTRFS_DC_WRITTEN;
3591 spin_unlock(&block_group->lock);
3594 spin_unlock(&block_group->lock);
3597 * We hit an ENOSPC when setting up the cache in this transaction, just
3598 * skip doing the setup, we've already cleared the cache so we're safe.
3600 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3606 * Try to preallocate enough space based on how big the block group is.
3607 * Keep in mind this has to include any pinned space which could end up
3608 * taking up quite a bit since it's not folded into the other space
3611 num_pages = div_u64(block_group->key.offset, SZ_256M);
3616 num_pages *= PAGE_SIZE;
3618 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3622 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3623 num_pages, num_pages,
3626 * Our cache requires contiguous chunks so that we don't modify a bunch
3627 * of metadata or split extents when writing the cache out, which means
3628 * we can enospc if we are heavily fragmented in addition to just normal
3629 * out of space conditions. So if we hit this just skip setting up any
3630 * other block groups for this transaction, maybe we'll unpin enough
3631 * space the next time around.
3634 dcs = BTRFS_DC_SETUP;
3635 else if (ret == -ENOSPC)
3636 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3641 btrfs_release_path(path);
3643 spin_lock(&block_group->lock);
3644 if (!ret && dcs == BTRFS_DC_SETUP)
3645 block_group->cache_generation = trans->transid;
3646 block_group->disk_cache_state = dcs;
3647 spin_unlock(&block_group->lock);
3649 extent_changeset_free(data_reserved);
3653 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3654 struct btrfs_fs_info *fs_info)
3656 struct btrfs_block_group_cache *cache, *tmp;
3657 struct btrfs_transaction *cur_trans = trans->transaction;
3658 struct btrfs_path *path;
3660 if (list_empty(&cur_trans->dirty_bgs) ||
3661 !btrfs_test_opt(fs_info, SPACE_CACHE))
3664 path = btrfs_alloc_path();
3668 /* Could add new block groups, use _safe just in case */
3669 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3671 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3672 cache_save_setup(cache, trans, path);
3675 btrfs_free_path(path);
3680 * transaction commit does final block group cache writeback during a
3681 * critical section where nothing is allowed to change the FS. This is
3682 * required in order for the cache to actually match the block group,
3683 * but can introduce a lot of latency into the commit.
3685 * So, btrfs_start_dirty_block_groups is here to kick off block group
3686 * cache IO. There's a chance we'll have to redo some of it if the
3687 * block group changes again during the commit, but it greatly reduces
3688 * the commit latency by getting rid of the easy block groups while
3689 * we're still allowing others to join the commit.
3691 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3692 struct btrfs_fs_info *fs_info)
3694 struct btrfs_block_group_cache *cache;
3695 struct btrfs_transaction *cur_trans = trans->transaction;
3698 struct btrfs_path *path = NULL;
3700 struct list_head *io = &cur_trans->io_bgs;
3701 int num_started = 0;
3704 spin_lock(&cur_trans->dirty_bgs_lock);
3705 if (list_empty(&cur_trans->dirty_bgs)) {
3706 spin_unlock(&cur_trans->dirty_bgs_lock);
3709 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3710 spin_unlock(&cur_trans->dirty_bgs_lock);
3714 * make sure all the block groups on our dirty list actually
3717 btrfs_create_pending_block_groups(trans, fs_info);
3720 path = btrfs_alloc_path();
3726 * cache_write_mutex is here only to save us from balance or automatic
3727 * removal of empty block groups deleting this block group while we are
3728 * writing out the cache
3730 mutex_lock(&trans->transaction->cache_write_mutex);
3731 while (!list_empty(&dirty)) {
3732 cache = list_first_entry(&dirty,
3733 struct btrfs_block_group_cache,
3736 * this can happen if something re-dirties a block
3737 * group that is already under IO. Just wait for it to
3738 * finish and then do it all again
3740 if (!list_empty(&cache->io_list)) {
3741 list_del_init(&cache->io_list);
3742 btrfs_wait_cache_io(trans, cache, path);
3743 btrfs_put_block_group(cache);
3748 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3749 * if it should update the cache_state. Don't delete
3750 * until after we wait.
3752 * Since we're not running in the commit critical section
3753 * we need the dirty_bgs_lock to protect from update_block_group
3755 spin_lock(&cur_trans->dirty_bgs_lock);
3756 list_del_init(&cache->dirty_list);
3757 spin_unlock(&cur_trans->dirty_bgs_lock);
3761 cache_save_setup(cache, trans, path);
3763 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3764 cache->io_ctl.inode = NULL;
3765 ret = btrfs_write_out_cache(fs_info, trans,
3767 if (ret == 0 && cache->io_ctl.inode) {
3772 * the cache_write_mutex is protecting
3775 list_add_tail(&cache->io_list, io);
3778 * if we failed to write the cache, the
3779 * generation will be bad and life goes on
3785 ret = write_one_cache_group(trans, fs_info,
3788 * Our block group might still be attached to the list
3789 * of new block groups in the transaction handle of some
3790 * other task (struct btrfs_trans_handle->new_bgs). This
3791 * means its block group item isn't yet in the extent
3792 * tree. If this happens ignore the error, as we will
3793 * try again later in the critical section of the
3794 * transaction commit.
3796 if (ret == -ENOENT) {
3798 spin_lock(&cur_trans->dirty_bgs_lock);
3799 if (list_empty(&cache->dirty_list)) {
3800 list_add_tail(&cache->dirty_list,
3801 &cur_trans->dirty_bgs);
3802 btrfs_get_block_group(cache);
3804 spin_unlock(&cur_trans->dirty_bgs_lock);
3806 btrfs_abort_transaction(trans, ret);
3810 /* if its not on the io list, we need to put the block group */
3812 btrfs_put_block_group(cache);
3818 * Avoid blocking other tasks for too long. It might even save
3819 * us from writing caches for block groups that are going to be
3822 mutex_unlock(&trans->transaction->cache_write_mutex);
3823 mutex_lock(&trans->transaction->cache_write_mutex);
3825 mutex_unlock(&trans->transaction->cache_write_mutex);
3828 * go through delayed refs for all the stuff we've just kicked off
3829 * and then loop back (just once)
3831 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
3832 if (!ret && loops == 0) {
3834 spin_lock(&cur_trans->dirty_bgs_lock);
3835 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3837 * dirty_bgs_lock protects us from concurrent block group
3838 * deletes too (not just cache_write_mutex).
3840 if (!list_empty(&dirty)) {
3841 spin_unlock(&cur_trans->dirty_bgs_lock);
3844 spin_unlock(&cur_trans->dirty_bgs_lock);
3845 } else if (ret < 0) {
3846 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3849 btrfs_free_path(path);
3853 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3854 struct btrfs_fs_info *fs_info)
3856 struct btrfs_block_group_cache *cache;
3857 struct btrfs_transaction *cur_trans = trans->transaction;
3860 struct btrfs_path *path;
3861 struct list_head *io = &cur_trans->io_bgs;
3862 int num_started = 0;
3864 path = btrfs_alloc_path();
3869 * Even though we are in the critical section of the transaction commit,
3870 * we can still have concurrent tasks adding elements to this
3871 * transaction's list of dirty block groups. These tasks correspond to
3872 * endio free space workers started when writeback finishes for a
3873 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3874 * allocate new block groups as a result of COWing nodes of the root
3875 * tree when updating the free space inode. The writeback for the space
3876 * caches is triggered by an earlier call to
3877 * btrfs_start_dirty_block_groups() and iterations of the following
3879 * Also we want to do the cache_save_setup first and then run the
3880 * delayed refs to make sure we have the best chance at doing this all
3883 spin_lock(&cur_trans->dirty_bgs_lock);
3884 while (!list_empty(&cur_trans->dirty_bgs)) {
3885 cache = list_first_entry(&cur_trans->dirty_bgs,
3886 struct btrfs_block_group_cache,
3890 * this can happen if cache_save_setup re-dirties a block
3891 * group that is already under IO. Just wait for it to
3892 * finish and then do it all again
3894 if (!list_empty(&cache->io_list)) {
3895 spin_unlock(&cur_trans->dirty_bgs_lock);
3896 list_del_init(&cache->io_list);
3897 btrfs_wait_cache_io(trans, cache, path);
3898 btrfs_put_block_group(cache);
3899 spin_lock(&cur_trans->dirty_bgs_lock);
3903 * don't remove from the dirty list until after we've waited
3906 list_del_init(&cache->dirty_list);
3907 spin_unlock(&cur_trans->dirty_bgs_lock);
3910 cache_save_setup(cache, trans, path);
3913 ret = btrfs_run_delayed_refs(trans, fs_info,
3914 (unsigned long) -1);
3916 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3917 cache->io_ctl.inode = NULL;
3918 ret = btrfs_write_out_cache(fs_info, trans,
3920 if (ret == 0 && cache->io_ctl.inode) {
3923 list_add_tail(&cache->io_list, io);
3926 * if we failed to write the cache, the
3927 * generation will be bad and life goes on
3933 ret = write_one_cache_group(trans, fs_info,
3936 * One of the free space endio workers might have
3937 * created a new block group while updating a free space
3938 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3939 * and hasn't released its transaction handle yet, in
3940 * which case the new block group is still attached to
3941 * its transaction handle and its creation has not
3942 * finished yet (no block group item in the extent tree
3943 * yet, etc). If this is the case, wait for all free
3944 * space endio workers to finish and retry. This is a
3945 * a very rare case so no need for a more efficient and
3948 if (ret == -ENOENT) {
3949 wait_event(cur_trans->writer_wait,
3950 atomic_read(&cur_trans->num_writers) == 1);
3951 ret = write_one_cache_group(trans, fs_info,
3955 btrfs_abort_transaction(trans, ret);
3958 /* if its not on the io list, we need to put the block group */
3960 btrfs_put_block_group(cache);
3961 spin_lock(&cur_trans->dirty_bgs_lock);
3963 spin_unlock(&cur_trans->dirty_bgs_lock);
3965 while (!list_empty(io)) {
3966 cache = list_first_entry(io, struct btrfs_block_group_cache,
3968 list_del_init(&cache->io_list);
3969 btrfs_wait_cache_io(trans, cache, path);
3970 btrfs_put_block_group(cache);
3973 btrfs_free_path(path);
3977 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3979 struct btrfs_block_group_cache *block_group;
3982 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3983 if (!block_group || block_group->ro)
3986 btrfs_put_block_group(block_group);
3990 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3992 struct btrfs_block_group_cache *bg;
3995 bg = btrfs_lookup_block_group(fs_info, bytenr);
3999 spin_lock(&bg->lock);
4003 atomic_inc(&bg->nocow_writers);
4004 spin_unlock(&bg->lock);
4006 /* no put on block group, done by btrfs_dec_nocow_writers */
4008 btrfs_put_block_group(bg);
4014 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
4016 struct btrfs_block_group_cache *bg;
4018 bg = btrfs_lookup_block_group(fs_info, bytenr);
4020 if (atomic_dec_and_test(&bg->nocow_writers))
4021 wake_up_atomic_t(&bg->nocow_writers);
4023 * Once for our lookup and once for the lookup done by a previous call
4024 * to btrfs_inc_nocow_writers()
4026 btrfs_put_block_group(bg);
4027 btrfs_put_block_group(bg);
4030 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
4036 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4038 wait_on_atomic_t(&bg->nocow_writers,
4039 btrfs_wait_nocow_writers_atomic_t,
4040 TASK_UNINTERRUPTIBLE);
4043 static const char *alloc_name(u64 flags)
4046 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4048 case BTRFS_BLOCK_GROUP_METADATA:
4050 case BTRFS_BLOCK_GROUP_DATA:
4052 case BTRFS_BLOCK_GROUP_SYSTEM:
4056 return "invalid-combination";
4060 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4061 struct btrfs_space_info **new)
4064 struct btrfs_space_info *space_info;
4068 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4072 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4079 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4080 INIT_LIST_HEAD(&space_info->block_groups[i]);
4081 init_rwsem(&space_info->groups_sem);
4082 spin_lock_init(&space_info->lock);
4083 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4084 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4085 init_waitqueue_head(&space_info->wait);
4086 INIT_LIST_HEAD(&space_info->ro_bgs);
4087 INIT_LIST_HEAD(&space_info->tickets);
4088 INIT_LIST_HEAD(&space_info->priority_tickets);
4090 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4091 info->space_info_kobj, "%s",
4092 alloc_name(space_info->flags));
4094 kobject_put(&space_info->kobj);
4099 list_add_rcu(&space_info->list, &info->space_info);
4100 if (flags & BTRFS_BLOCK_GROUP_DATA)
4101 info->data_sinfo = space_info;
4106 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4107 u64 total_bytes, u64 bytes_used,
4109 struct btrfs_space_info **space_info)
4111 struct btrfs_space_info *found;
4114 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4115 BTRFS_BLOCK_GROUP_RAID10))
4120 found = __find_space_info(info, flags);
4122 spin_lock(&found->lock);
4123 found->total_bytes += total_bytes;
4124 found->disk_total += total_bytes * factor;
4125 found->bytes_used += bytes_used;
4126 found->disk_used += bytes_used * factor;
4127 found->bytes_readonly += bytes_readonly;
4128 if (total_bytes > 0)
4130 space_info_add_new_bytes(info, found, total_bytes -
4131 bytes_used - bytes_readonly);
4132 spin_unlock(&found->lock);
4133 *space_info = found;
4136 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4138 u64 extra_flags = chunk_to_extended(flags) &
4139 BTRFS_EXTENDED_PROFILE_MASK;
4141 write_seqlock(&fs_info->profiles_lock);
4142 if (flags & BTRFS_BLOCK_GROUP_DATA)
4143 fs_info->avail_data_alloc_bits |= extra_flags;
4144 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4145 fs_info->avail_metadata_alloc_bits |= extra_flags;
4146 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4147 fs_info->avail_system_alloc_bits |= extra_flags;
4148 write_sequnlock(&fs_info->profiles_lock);
4152 * returns target flags in extended format or 0 if restripe for this
4153 * chunk_type is not in progress
4155 * should be called with either volume_mutex or balance_lock held
4157 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4159 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4165 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4166 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4167 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4168 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4169 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4170 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4171 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4172 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4173 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4180 * @flags: available profiles in extended format (see ctree.h)
4182 * Returns reduced profile in chunk format. If profile changing is in
4183 * progress (either running or paused) picks the target profile (if it's
4184 * already available), otherwise falls back to plain reducing.
4186 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4188 u64 num_devices = fs_info->fs_devices->rw_devices;
4194 * see if restripe for this chunk_type is in progress, if so
4195 * try to reduce to the target profile
4197 spin_lock(&fs_info->balance_lock);
4198 target = get_restripe_target(fs_info, flags);
4200 /* pick target profile only if it's already available */
4201 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4202 spin_unlock(&fs_info->balance_lock);
4203 return extended_to_chunk(target);
4206 spin_unlock(&fs_info->balance_lock);
4208 /* First, mask out the RAID levels which aren't possible */
4209 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4210 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4211 allowed |= btrfs_raid_group[raid_type];
4215 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4216 allowed = BTRFS_BLOCK_GROUP_RAID6;
4217 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4218 allowed = BTRFS_BLOCK_GROUP_RAID5;
4219 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4220 allowed = BTRFS_BLOCK_GROUP_RAID10;
4221 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4222 allowed = BTRFS_BLOCK_GROUP_RAID1;
4223 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4224 allowed = BTRFS_BLOCK_GROUP_RAID0;
4226 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4228 return extended_to_chunk(flags | allowed);
4231 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4238 seq = read_seqbegin(&fs_info->profiles_lock);
4240 if (flags & BTRFS_BLOCK_GROUP_DATA)
4241 flags |= fs_info->avail_data_alloc_bits;
4242 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4243 flags |= fs_info->avail_system_alloc_bits;
4244 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4245 flags |= fs_info->avail_metadata_alloc_bits;
4246 } while (read_seqretry(&fs_info->profiles_lock, seq));
4248 return btrfs_reduce_alloc_profile(fs_info, flags);
4251 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4253 struct btrfs_fs_info *fs_info = root->fs_info;
4258 flags = BTRFS_BLOCK_GROUP_DATA;
4259 else if (root == fs_info->chunk_root)
4260 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4262 flags = BTRFS_BLOCK_GROUP_METADATA;
4264 ret = get_alloc_profile(fs_info, flags);
4268 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4270 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4273 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4275 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4278 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4280 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4283 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4284 bool may_use_included)
4287 return s_info->bytes_used + s_info->bytes_reserved +
4288 s_info->bytes_pinned + s_info->bytes_readonly +
4289 (may_use_included ? s_info->bytes_may_use : 0);
4292 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4294 struct btrfs_root *root = inode->root;
4295 struct btrfs_fs_info *fs_info = root->fs_info;
4296 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4299 int need_commit = 2;
4300 int have_pinned_space;
4302 /* make sure bytes are sectorsize aligned */
4303 bytes = ALIGN(bytes, fs_info->sectorsize);
4305 if (btrfs_is_free_space_inode(inode)) {
4307 ASSERT(current->journal_info);
4311 /* make sure we have enough space to handle the data first */
4312 spin_lock(&data_sinfo->lock);
4313 used = btrfs_space_info_used(data_sinfo, true);
4315 if (used + bytes > data_sinfo->total_bytes) {
4316 struct btrfs_trans_handle *trans;
4319 * if we don't have enough free bytes in this space then we need
4320 * to alloc a new chunk.
4322 if (!data_sinfo->full) {
4325 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4326 spin_unlock(&data_sinfo->lock);
4328 alloc_target = btrfs_data_alloc_profile(fs_info);
4330 * It is ugly that we don't call nolock join
4331 * transaction for the free space inode case here.
4332 * But it is safe because we only do the data space
4333 * reservation for the free space cache in the
4334 * transaction context, the common join transaction
4335 * just increase the counter of the current transaction
4336 * handler, doesn't try to acquire the trans_lock of
4339 trans = btrfs_join_transaction(root);
4341 return PTR_ERR(trans);
4343 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4344 CHUNK_ALLOC_NO_FORCE);
4345 btrfs_end_transaction(trans);
4350 have_pinned_space = 1;
4359 * If we don't have enough pinned space to deal with this
4360 * allocation, and no removed chunk in current transaction,
4361 * don't bother committing the transaction.
4363 have_pinned_space = percpu_counter_compare(
4364 &data_sinfo->total_bytes_pinned,
4365 used + bytes - data_sinfo->total_bytes);
4366 spin_unlock(&data_sinfo->lock);
4368 /* commit the current transaction and try again */
4371 !atomic_read(&fs_info->open_ioctl_trans)) {
4374 if (need_commit > 0) {
4375 btrfs_start_delalloc_roots(fs_info, 0, -1);
4376 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4380 trans = btrfs_join_transaction(root);
4382 return PTR_ERR(trans);
4383 if (have_pinned_space >= 0 ||
4384 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4385 &trans->transaction->flags) ||
4387 ret = btrfs_commit_transaction(trans);
4391 * The cleaner kthread might still be doing iput
4392 * operations. Wait for it to finish so that
4393 * more space is released.
4395 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4396 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4399 btrfs_end_transaction(trans);
4403 trace_btrfs_space_reservation(fs_info,
4404 "space_info:enospc",
4405 data_sinfo->flags, bytes, 1);
4408 data_sinfo->bytes_may_use += bytes;
4409 trace_btrfs_space_reservation(fs_info, "space_info",
4410 data_sinfo->flags, bytes, 1);
4411 spin_unlock(&data_sinfo->lock);
4416 int btrfs_check_data_free_space(struct inode *inode,
4417 struct extent_changeset **reserved, u64 start, u64 len)
4419 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4422 /* align the range */
4423 len = round_up(start + len, fs_info->sectorsize) -
4424 round_down(start, fs_info->sectorsize);
4425 start = round_down(start, fs_info->sectorsize);
4427 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4431 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4432 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4434 btrfs_free_reserved_data_space_noquota(inode, start, len);
4441 * Called if we need to clear a data reservation for this inode
4442 * Normally in a error case.
4444 * This one will *NOT* use accurate qgroup reserved space API, just for case
4445 * which we can't sleep and is sure it won't affect qgroup reserved space.
4446 * Like clear_bit_hook().
4448 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4451 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4452 struct btrfs_space_info *data_sinfo;
4454 /* Make sure the range is aligned to sectorsize */
4455 len = round_up(start + len, fs_info->sectorsize) -
4456 round_down(start, fs_info->sectorsize);
4457 start = round_down(start, fs_info->sectorsize);
4459 data_sinfo = fs_info->data_sinfo;
4460 spin_lock(&data_sinfo->lock);
4461 if (WARN_ON(data_sinfo->bytes_may_use < len))
4462 data_sinfo->bytes_may_use = 0;
4464 data_sinfo->bytes_may_use -= len;
4465 trace_btrfs_space_reservation(fs_info, "space_info",
4466 data_sinfo->flags, len, 0);
4467 spin_unlock(&data_sinfo->lock);
4471 * Called if we need to clear a data reservation for this inode
4472 * Normally in a error case.
4474 * This one will handle the per-inode data rsv map for accurate reserved
4477 void btrfs_free_reserved_data_space(struct inode *inode,
4478 struct extent_changeset *reserved, u64 start, u64 len)
4480 struct btrfs_root *root = BTRFS_I(inode)->root;
4482 /* Make sure the range is aligned to sectorsize */
4483 len = round_up(start + len, root->fs_info->sectorsize) -
4484 round_down(start, root->fs_info->sectorsize);
4485 start = round_down(start, root->fs_info->sectorsize);
4487 btrfs_free_reserved_data_space_noquota(inode, start, len);
4488 btrfs_qgroup_free_data(inode, reserved, start, len);
4491 static void force_metadata_allocation(struct btrfs_fs_info *info)
4493 struct list_head *head = &info->space_info;
4494 struct btrfs_space_info *found;
4497 list_for_each_entry_rcu(found, head, list) {
4498 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4499 found->force_alloc = CHUNK_ALLOC_FORCE;
4504 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4506 return (global->size << 1);
4509 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4510 struct btrfs_space_info *sinfo, int force)
4512 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4513 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4516 if (force == CHUNK_ALLOC_FORCE)
4520 * We need to take into account the global rsv because for all intents
4521 * and purposes it's used space. Don't worry about locking the
4522 * global_rsv, it doesn't change except when the transaction commits.
4524 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4525 bytes_used += calc_global_rsv_need_space(global_rsv);
4528 * in limited mode, we want to have some free space up to
4529 * about 1% of the FS size.
4531 if (force == CHUNK_ALLOC_LIMITED) {
4532 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4533 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4535 if (sinfo->total_bytes - bytes_used < thresh)
4539 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4544 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4548 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4549 BTRFS_BLOCK_GROUP_RAID0 |
4550 BTRFS_BLOCK_GROUP_RAID5 |
4551 BTRFS_BLOCK_GROUP_RAID6))
4552 num_dev = fs_info->fs_devices->rw_devices;
4553 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4556 num_dev = 1; /* DUP or single */
4562 * If @is_allocation is true, reserve space in the system space info necessary
4563 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4566 void check_system_chunk(struct btrfs_trans_handle *trans,
4567 struct btrfs_fs_info *fs_info, u64 type)
4569 struct btrfs_space_info *info;
4576 * Needed because we can end up allocating a system chunk and for an
4577 * atomic and race free space reservation in the chunk block reserve.
4579 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4581 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4582 spin_lock(&info->lock);
4583 left = info->total_bytes - btrfs_space_info_used(info, true);
4584 spin_unlock(&info->lock);
4586 num_devs = get_profile_num_devs(fs_info, type);
4588 /* num_devs device items to update and 1 chunk item to add or remove */
4589 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4590 btrfs_calc_trans_metadata_size(fs_info, 1);
4592 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4593 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4594 left, thresh, type);
4595 dump_space_info(fs_info, info, 0, 0);
4598 if (left < thresh) {
4599 u64 flags = btrfs_system_alloc_profile(fs_info);
4602 * Ignore failure to create system chunk. We might end up not
4603 * needing it, as we might not need to COW all nodes/leafs from
4604 * the paths we visit in the chunk tree (they were already COWed
4605 * or created in the current transaction for example).
4607 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4611 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4612 &fs_info->chunk_block_rsv,
4613 thresh, BTRFS_RESERVE_NO_FLUSH);
4615 trans->chunk_bytes_reserved += thresh;
4620 * If force is CHUNK_ALLOC_FORCE:
4621 * - return 1 if it successfully allocates a chunk,
4622 * - return errors including -ENOSPC otherwise.
4623 * If force is NOT CHUNK_ALLOC_FORCE:
4624 * - return 0 if it doesn't need to allocate a new chunk,
4625 * - return 1 if it successfully allocates a chunk,
4626 * - return errors including -ENOSPC otherwise.
4628 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4629 struct btrfs_fs_info *fs_info, u64 flags, int force)
4631 struct btrfs_space_info *space_info;
4632 int wait_for_alloc = 0;
4635 /* Don't re-enter if we're already allocating a chunk */
4636 if (trans->allocating_chunk)
4639 space_info = __find_space_info(fs_info, flags);
4641 ret = create_space_info(fs_info, flags, &space_info);
4647 spin_lock(&space_info->lock);
4648 if (force < space_info->force_alloc)
4649 force = space_info->force_alloc;
4650 if (space_info->full) {
4651 if (should_alloc_chunk(fs_info, space_info, force))
4655 spin_unlock(&space_info->lock);
4659 if (!should_alloc_chunk(fs_info, space_info, force)) {
4660 spin_unlock(&space_info->lock);
4662 } else if (space_info->chunk_alloc) {
4665 space_info->chunk_alloc = 1;
4668 spin_unlock(&space_info->lock);
4670 mutex_lock(&fs_info->chunk_mutex);
4673 * The chunk_mutex is held throughout the entirety of a chunk
4674 * allocation, so once we've acquired the chunk_mutex we know that the
4675 * other guy is done and we need to recheck and see if we should
4678 if (wait_for_alloc) {
4679 mutex_unlock(&fs_info->chunk_mutex);
4685 trans->allocating_chunk = true;
4688 * If we have mixed data/metadata chunks we want to make sure we keep
4689 * allocating mixed chunks instead of individual chunks.
4691 if (btrfs_mixed_space_info(space_info))
4692 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4695 * if we're doing a data chunk, go ahead and make sure that
4696 * we keep a reasonable number of metadata chunks allocated in the
4699 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4700 fs_info->data_chunk_allocations++;
4701 if (!(fs_info->data_chunk_allocations %
4702 fs_info->metadata_ratio))
4703 force_metadata_allocation(fs_info);
4707 * Check if we have enough space in SYSTEM chunk because we may need
4708 * to update devices.
4710 check_system_chunk(trans, fs_info, flags);
4712 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4713 trans->allocating_chunk = false;
4715 spin_lock(&space_info->lock);
4716 if (ret < 0 && ret != -ENOSPC)
4719 space_info->full = 1;
4723 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4725 space_info->chunk_alloc = 0;
4726 spin_unlock(&space_info->lock);
4727 mutex_unlock(&fs_info->chunk_mutex);
4729 * When we allocate a new chunk we reserve space in the chunk block
4730 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4731 * add new nodes/leafs to it if we end up needing to do it when
4732 * inserting the chunk item and updating device items as part of the
4733 * second phase of chunk allocation, performed by
4734 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4735 * large number of new block groups to create in our transaction
4736 * handle's new_bgs list to avoid exhausting the chunk block reserve
4737 * in extreme cases - like having a single transaction create many new
4738 * block groups when starting to write out the free space caches of all
4739 * the block groups that were made dirty during the lifetime of the
4742 if (trans->can_flush_pending_bgs &&
4743 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4744 btrfs_create_pending_block_groups(trans, fs_info);
4745 btrfs_trans_release_chunk_metadata(trans);
4750 static int can_overcommit(struct btrfs_fs_info *fs_info,
4751 struct btrfs_space_info *space_info, u64 bytes,
4752 enum btrfs_reserve_flush_enum flush,
4755 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4761 /* Don't overcommit when in mixed mode. */
4762 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4766 profile = btrfs_system_alloc_profile(fs_info);
4768 profile = btrfs_metadata_alloc_profile(fs_info);
4770 used = btrfs_space_info_used(space_info, false);
4773 * We only want to allow over committing if we have lots of actual space
4774 * free, but if we don't have enough space to handle the global reserve
4775 * space then we could end up having a real enospc problem when trying
4776 * to allocate a chunk or some other such important allocation.
4778 spin_lock(&global_rsv->lock);
4779 space_size = calc_global_rsv_need_space(global_rsv);
4780 spin_unlock(&global_rsv->lock);
4781 if (used + space_size >= space_info->total_bytes)
4784 used += space_info->bytes_may_use;
4786 avail = atomic64_read(&fs_info->free_chunk_space);
4789 * If we have dup, raid1 or raid10 then only half of the free
4790 * space is actually useable. For raid56, the space info used
4791 * doesn't include the parity drive, so we don't have to
4794 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4795 BTRFS_BLOCK_GROUP_RAID1 |
4796 BTRFS_BLOCK_GROUP_RAID10))
4800 * If we aren't flushing all things, let us overcommit up to
4801 * 1/2th of the space. If we can flush, don't let us overcommit
4802 * too much, let it overcommit up to 1/8 of the space.
4804 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4809 if (used + bytes < space_info->total_bytes + avail)
4814 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4815 unsigned long nr_pages, int nr_items)
4817 struct super_block *sb = fs_info->sb;
4819 if (down_read_trylock(&sb->s_umount)) {
4820 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4821 up_read(&sb->s_umount);
4824 * We needn't worry the filesystem going from r/w to r/o though
4825 * we don't acquire ->s_umount mutex, because the filesystem
4826 * should guarantee the delalloc inodes list be empty after
4827 * the filesystem is readonly(all dirty pages are written to
4830 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4831 if (!current->journal_info)
4832 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4836 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4842 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4843 nr = div64_u64(to_reclaim, bytes);
4849 #define EXTENT_SIZE_PER_ITEM SZ_256K
4852 * shrink metadata reservation for delalloc
4854 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4855 u64 orig, bool wait_ordered)
4857 struct btrfs_block_rsv *block_rsv;
4858 struct btrfs_space_info *space_info;
4859 struct btrfs_trans_handle *trans;
4864 unsigned long nr_pages;
4866 enum btrfs_reserve_flush_enum flush;
4868 /* Calc the number of the pages we need flush for space reservation */
4869 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4870 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4872 trans = (struct btrfs_trans_handle *)current->journal_info;
4873 block_rsv = &fs_info->delalloc_block_rsv;
4874 space_info = block_rsv->space_info;
4876 delalloc_bytes = percpu_counter_sum_positive(
4877 &fs_info->delalloc_bytes);
4878 if (delalloc_bytes == 0) {
4882 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4887 while (delalloc_bytes && loops < 3) {
4888 max_reclaim = min(delalloc_bytes, to_reclaim);
4889 nr_pages = max_reclaim >> PAGE_SHIFT;
4890 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4892 * We need to wait for the async pages to actually start before
4895 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4899 if (max_reclaim <= nr_pages)
4902 max_reclaim -= nr_pages;
4904 wait_event(fs_info->async_submit_wait,
4905 atomic_read(&fs_info->async_delalloc_pages) <=
4909 flush = BTRFS_RESERVE_FLUSH_ALL;
4911 flush = BTRFS_RESERVE_NO_FLUSH;
4912 spin_lock(&space_info->lock);
4913 if (list_empty(&space_info->tickets) &&
4914 list_empty(&space_info->priority_tickets)) {
4915 spin_unlock(&space_info->lock);
4918 spin_unlock(&space_info->lock);
4921 if (wait_ordered && !trans) {
4922 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4924 time_left = schedule_timeout_killable(1);
4928 delalloc_bytes = percpu_counter_sum_positive(
4929 &fs_info->delalloc_bytes);
4933 struct reserve_ticket {
4936 struct list_head list;
4937 wait_queue_head_t wait;
4941 * maybe_commit_transaction - possibly commit the transaction if its ok to
4942 * @root - the root we're allocating for
4943 * @bytes - the number of bytes we want to reserve
4944 * @force - force the commit
4946 * This will check to make sure that committing the transaction will actually
4947 * get us somewhere and then commit the transaction if it does. Otherwise it
4948 * will return -ENOSPC.
4950 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4951 struct btrfs_space_info *space_info)
4953 struct reserve_ticket *ticket = NULL;
4954 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4955 struct btrfs_trans_handle *trans;
4958 trans = (struct btrfs_trans_handle *)current->journal_info;
4962 spin_lock(&space_info->lock);
4963 if (!list_empty(&space_info->priority_tickets))
4964 ticket = list_first_entry(&space_info->priority_tickets,
4965 struct reserve_ticket, list);
4966 else if (!list_empty(&space_info->tickets))
4967 ticket = list_first_entry(&space_info->tickets,
4968 struct reserve_ticket, list);
4969 bytes = (ticket) ? ticket->bytes : 0;
4970 spin_unlock(&space_info->lock);
4975 /* See if there is enough pinned space to make this reservation */
4976 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4981 * See if there is some space in the delayed insertion reservation for
4984 if (space_info != delayed_rsv->space_info)
4987 spin_lock(&delayed_rsv->lock);
4988 if (delayed_rsv->size > bytes)
4991 bytes -= delayed_rsv->size;
4992 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4994 spin_unlock(&delayed_rsv->lock);
4997 spin_unlock(&delayed_rsv->lock);
5000 trans = btrfs_join_transaction(fs_info->extent_root);
5004 return btrfs_commit_transaction(trans);
5008 * Try to flush some data based on policy set by @state. This is only advisory
5009 * and may fail for various reasons. The caller is supposed to examine the
5010 * state of @space_info to detect the outcome.
5012 static void flush_space(struct btrfs_fs_info *fs_info,
5013 struct btrfs_space_info *space_info, u64 num_bytes,
5016 struct btrfs_root *root = fs_info->extent_root;
5017 struct btrfs_trans_handle *trans;
5022 case FLUSH_DELAYED_ITEMS_NR:
5023 case FLUSH_DELAYED_ITEMS:
5024 if (state == FLUSH_DELAYED_ITEMS_NR)
5025 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
5029 trans = btrfs_join_transaction(root);
5030 if (IS_ERR(trans)) {
5031 ret = PTR_ERR(trans);
5034 ret = btrfs_run_delayed_items_nr(trans, fs_info, nr);
5035 btrfs_end_transaction(trans);
5037 case FLUSH_DELALLOC:
5038 case FLUSH_DELALLOC_WAIT:
5039 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
5040 state == FLUSH_DELALLOC_WAIT);
5043 trans = btrfs_join_transaction(root);
5044 if (IS_ERR(trans)) {
5045 ret = PTR_ERR(trans);
5048 ret = do_chunk_alloc(trans, fs_info,
5049 btrfs_metadata_alloc_profile(fs_info),
5050 CHUNK_ALLOC_NO_FORCE);
5051 btrfs_end_transaction(trans);
5052 if (ret > 0 || ret == -ENOSPC)
5056 ret = may_commit_transaction(fs_info, space_info);
5063 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5069 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5070 struct btrfs_space_info *space_info,
5073 struct reserve_ticket *ticket;
5078 list_for_each_entry(ticket, &space_info->tickets, list)
5079 to_reclaim += ticket->bytes;
5080 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5081 to_reclaim += ticket->bytes;
5085 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5086 if (can_overcommit(fs_info, space_info, to_reclaim,
5087 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5090 used = btrfs_space_info_used(space_info, true);
5092 if (can_overcommit(fs_info, space_info, SZ_1M,
5093 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5094 expected = div_factor_fine(space_info->total_bytes, 95);
5096 expected = div_factor_fine(space_info->total_bytes, 90);
5098 if (used > expected)
5099 to_reclaim = used - expected;
5102 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5103 space_info->bytes_reserved);
5107 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5108 struct btrfs_space_info *space_info,
5109 u64 used, bool system_chunk)
5111 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5113 /* If we're just plain full then async reclaim just slows us down. */
5114 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5117 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5121 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5122 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5125 static void wake_all_tickets(struct list_head *head)
5127 struct reserve_ticket *ticket;
5129 while (!list_empty(head)) {
5130 ticket = list_first_entry(head, struct reserve_ticket, list);
5131 list_del_init(&ticket->list);
5132 ticket->error = -ENOSPC;
5133 wake_up(&ticket->wait);
5138 * This is for normal flushers, we can wait all goddamned day if we want to. We
5139 * will loop and continuously try to flush as long as we are making progress.
5140 * We count progress as clearing off tickets each time we have to loop.
5142 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5144 struct btrfs_fs_info *fs_info;
5145 struct btrfs_space_info *space_info;
5148 int commit_cycles = 0;
5149 u64 last_tickets_id;
5151 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5152 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5154 spin_lock(&space_info->lock);
5155 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5158 space_info->flush = 0;
5159 spin_unlock(&space_info->lock);
5162 last_tickets_id = space_info->tickets_id;
5163 spin_unlock(&space_info->lock);
5165 flush_state = FLUSH_DELAYED_ITEMS_NR;
5167 flush_space(fs_info, space_info, to_reclaim, flush_state);
5168 spin_lock(&space_info->lock);
5169 if (list_empty(&space_info->tickets)) {
5170 space_info->flush = 0;
5171 spin_unlock(&space_info->lock);
5174 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5177 if (last_tickets_id == space_info->tickets_id) {
5180 last_tickets_id = space_info->tickets_id;
5181 flush_state = FLUSH_DELAYED_ITEMS_NR;
5186 if (flush_state > COMMIT_TRANS) {
5188 if (commit_cycles > 2) {
5189 wake_all_tickets(&space_info->tickets);
5190 space_info->flush = 0;
5192 flush_state = FLUSH_DELAYED_ITEMS_NR;
5195 spin_unlock(&space_info->lock);
5196 } while (flush_state <= COMMIT_TRANS);
5199 void btrfs_init_async_reclaim_work(struct work_struct *work)
5201 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5204 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5205 struct btrfs_space_info *space_info,
5206 struct reserve_ticket *ticket)
5209 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5211 spin_lock(&space_info->lock);
5212 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5215 spin_unlock(&space_info->lock);
5218 spin_unlock(&space_info->lock);
5221 flush_space(fs_info, space_info, to_reclaim, flush_state);
5223 spin_lock(&space_info->lock);
5224 if (ticket->bytes == 0) {
5225 spin_unlock(&space_info->lock);
5228 spin_unlock(&space_info->lock);
5231 * Priority flushers can't wait on delalloc without
5234 if (flush_state == FLUSH_DELALLOC ||
5235 flush_state == FLUSH_DELALLOC_WAIT)
5236 flush_state = ALLOC_CHUNK;
5237 } while (flush_state < COMMIT_TRANS);
5240 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5241 struct btrfs_space_info *space_info,
5242 struct reserve_ticket *ticket, u64 orig_bytes)
5248 spin_lock(&space_info->lock);
5249 while (ticket->bytes > 0 && ticket->error == 0) {
5250 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5255 spin_unlock(&space_info->lock);
5259 finish_wait(&ticket->wait, &wait);
5260 spin_lock(&space_info->lock);
5263 ret = ticket->error;
5264 if (!list_empty(&ticket->list))
5265 list_del_init(&ticket->list);
5266 if (ticket->bytes && ticket->bytes < orig_bytes) {
5267 u64 num_bytes = orig_bytes - ticket->bytes;
5268 space_info->bytes_may_use -= num_bytes;
5269 trace_btrfs_space_reservation(fs_info, "space_info",
5270 space_info->flags, num_bytes, 0);
5272 spin_unlock(&space_info->lock);
5278 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5279 * @root - the root we're allocating for
5280 * @space_info - the space info we want to allocate from
5281 * @orig_bytes - the number of bytes we want
5282 * @flush - whether or not we can flush to make our reservation
5284 * This will reserve orig_bytes number of bytes from the space info associated
5285 * with the block_rsv. If there is not enough space it will make an attempt to
5286 * flush out space to make room. It will do this by flushing delalloc if
5287 * possible or committing the transaction. If flush is 0 then no attempts to
5288 * regain reservations will be made and this will fail if there is not enough
5291 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5292 struct btrfs_space_info *space_info,
5294 enum btrfs_reserve_flush_enum flush,
5297 struct reserve_ticket ticket;
5302 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5304 spin_lock(&space_info->lock);
5306 used = btrfs_space_info_used(space_info, true);
5309 * If we have enough space then hooray, make our reservation and carry
5310 * on. If not see if we can overcommit, and if we can, hooray carry on.
5311 * If not things get more complicated.
5313 if (used + orig_bytes <= space_info->total_bytes) {
5314 space_info->bytes_may_use += orig_bytes;
5315 trace_btrfs_space_reservation(fs_info, "space_info",
5316 space_info->flags, orig_bytes, 1);
5318 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5320 space_info->bytes_may_use += orig_bytes;
5321 trace_btrfs_space_reservation(fs_info, "space_info",
5322 space_info->flags, orig_bytes, 1);
5327 * If we couldn't make a reservation then setup our reservation ticket
5328 * and kick the async worker if it's not already running.
5330 * If we are a priority flusher then we just need to add our ticket to
5331 * the list and we will do our own flushing further down.
5333 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5334 ticket.bytes = orig_bytes;
5336 init_waitqueue_head(&ticket.wait);
5337 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5338 list_add_tail(&ticket.list, &space_info->tickets);
5339 if (!space_info->flush) {
5340 space_info->flush = 1;
5341 trace_btrfs_trigger_flush(fs_info,
5345 queue_work(system_unbound_wq,
5346 &fs_info->async_reclaim_work);
5349 list_add_tail(&ticket.list,
5350 &space_info->priority_tickets);
5352 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5355 * We will do the space reservation dance during log replay,
5356 * which means we won't have fs_info->fs_root set, so don't do
5357 * the async reclaim as we will panic.
5359 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5360 need_do_async_reclaim(fs_info, space_info,
5361 used, system_chunk) &&
5362 !work_busy(&fs_info->async_reclaim_work)) {
5363 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5364 orig_bytes, flush, "preempt");
5365 queue_work(system_unbound_wq,
5366 &fs_info->async_reclaim_work);
5369 spin_unlock(&space_info->lock);
5370 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5373 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5374 return wait_reserve_ticket(fs_info, space_info, &ticket,
5378 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5379 spin_lock(&space_info->lock);
5381 if (ticket.bytes < orig_bytes) {
5382 u64 num_bytes = orig_bytes - ticket.bytes;
5383 space_info->bytes_may_use -= num_bytes;
5384 trace_btrfs_space_reservation(fs_info, "space_info",
5389 list_del_init(&ticket.list);
5392 spin_unlock(&space_info->lock);
5393 ASSERT(list_empty(&ticket.list));
5398 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5399 * @root - the root we're allocating for
5400 * @block_rsv - the block_rsv we're allocating for
5401 * @orig_bytes - the number of bytes we want
5402 * @flush - whether or not we can flush to make our reservation
5404 * This will reserve orgi_bytes number of bytes from the space info associated
5405 * with the block_rsv. If there is not enough space it will make an attempt to
5406 * flush out space to make room. It will do this by flushing delalloc if
5407 * possible or committing the transaction. If flush is 0 then no attempts to
5408 * regain reservations will be made and this will fail if there is not enough
5411 static int reserve_metadata_bytes(struct btrfs_root *root,
5412 struct btrfs_block_rsv *block_rsv,
5414 enum btrfs_reserve_flush_enum flush)
5416 struct btrfs_fs_info *fs_info = root->fs_info;
5417 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5419 bool system_chunk = (root == fs_info->chunk_root);
5421 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5422 orig_bytes, flush, system_chunk);
5423 if (ret == -ENOSPC &&
5424 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5425 if (block_rsv != global_rsv &&
5426 !block_rsv_use_bytes(global_rsv, orig_bytes))
5430 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5431 block_rsv->space_info->flags,
5436 static struct btrfs_block_rsv *get_block_rsv(
5437 const struct btrfs_trans_handle *trans,
5438 const struct btrfs_root *root)
5440 struct btrfs_fs_info *fs_info = root->fs_info;
5441 struct btrfs_block_rsv *block_rsv = NULL;
5443 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5444 (root == fs_info->csum_root && trans->adding_csums) ||
5445 (root == fs_info->uuid_root))
5446 block_rsv = trans->block_rsv;
5449 block_rsv = root->block_rsv;
5452 block_rsv = &fs_info->empty_block_rsv;
5457 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5461 spin_lock(&block_rsv->lock);
5462 if (block_rsv->reserved >= num_bytes) {
5463 block_rsv->reserved -= num_bytes;
5464 if (block_rsv->reserved < block_rsv->size)
5465 block_rsv->full = 0;
5468 spin_unlock(&block_rsv->lock);
5472 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5473 u64 num_bytes, int update_size)
5475 spin_lock(&block_rsv->lock);
5476 block_rsv->reserved += num_bytes;
5478 block_rsv->size += num_bytes;
5479 else if (block_rsv->reserved >= block_rsv->size)
5480 block_rsv->full = 1;
5481 spin_unlock(&block_rsv->lock);
5484 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5485 struct btrfs_block_rsv *dest, u64 num_bytes,
5488 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5491 if (global_rsv->space_info != dest->space_info)
5494 spin_lock(&global_rsv->lock);
5495 min_bytes = div_factor(global_rsv->size, min_factor);
5496 if (global_rsv->reserved < min_bytes + num_bytes) {
5497 spin_unlock(&global_rsv->lock);
5500 global_rsv->reserved -= num_bytes;
5501 if (global_rsv->reserved < global_rsv->size)
5502 global_rsv->full = 0;
5503 spin_unlock(&global_rsv->lock);
5505 block_rsv_add_bytes(dest, num_bytes, 1);
5510 * This is for space we already have accounted in space_info->bytes_may_use, so
5511 * basically when we're returning space from block_rsv's.
5513 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5514 struct btrfs_space_info *space_info,
5517 struct reserve_ticket *ticket;
5518 struct list_head *head;
5520 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5521 bool check_overcommit = false;
5523 spin_lock(&space_info->lock);
5524 head = &space_info->priority_tickets;
5527 * If we are over our limit then we need to check and see if we can
5528 * overcommit, and if we can't then we just need to free up our space
5529 * and not satisfy any requests.
5531 used = btrfs_space_info_used(space_info, true);
5532 if (used - num_bytes >= space_info->total_bytes)
5533 check_overcommit = true;
5535 while (!list_empty(head) && num_bytes) {
5536 ticket = list_first_entry(head, struct reserve_ticket,
5539 * We use 0 bytes because this space is already reserved, so
5540 * adding the ticket space would be a double count.
5542 if (check_overcommit &&
5543 !can_overcommit(fs_info, space_info, 0, flush, false))
5545 if (num_bytes >= ticket->bytes) {
5546 list_del_init(&ticket->list);
5547 num_bytes -= ticket->bytes;
5549 space_info->tickets_id++;
5550 wake_up(&ticket->wait);
5552 ticket->bytes -= num_bytes;
5557 if (num_bytes && head == &space_info->priority_tickets) {
5558 head = &space_info->tickets;
5559 flush = BTRFS_RESERVE_FLUSH_ALL;
5562 space_info->bytes_may_use -= num_bytes;
5563 trace_btrfs_space_reservation(fs_info, "space_info",
5564 space_info->flags, num_bytes, 0);
5565 spin_unlock(&space_info->lock);
5569 * This is for newly allocated space that isn't accounted in
5570 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5571 * we use this helper.
5573 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5574 struct btrfs_space_info *space_info,
5577 struct reserve_ticket *ticket;
5578 struct list_head *head = &space_info->priority_tickets;
5581 while (!list_empty(head) && num_bytes) {
5582 ticket = list_first_entry(head, struct reserve_ticket,
5584 if (num_bytes >= ticket->bytes) {
5585 trace_btrfs_space_reservation(fs_info, "space_info",
5588 list_del_init(&ticket->list);
5589 num_bytes -= ticket->bytes;
5590 space_info->bytes_may_use += ticket->bytes;
5592 space_info->tickets_id++;
5593 wake_up(&ticket->wait);
5595 trace_btrfs_space_reservation(fs_info, "space_info",
5598 space_info->bytes_may_use += num_bytes;
5599 ticket->bytes -= num_bytes;
5604 if (num_bytes && head == &space_info->priority_tickets) {
5605 head = &space_info->tickets;
5610 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5611 struct btrfs_block_rsv *block_rsv,
5612 struct btrfs_block_rsv *dest, u64 num_bytes)
5614 struct btrfs_space_info *space_info = block_rsv->space_info;
5616 spin_lock(&block_rsv->lock);
5617 if (num_bytes == (u64)-1)
5618 num_bytes = block_rsv->size;
5619 block_rsv->size -= num_bytes;
5620 if (block_rsv->reserved >= block_rsv->size) {
5621 num_bytes = block_rsv->reserved - block_rsv->size;
5622 block_rsv->reserved = block_rsv->size;
5623 block_rsv->full = 1;
5627 spin_unlock(&block_rsv->lock);
5629 if (num_bytes > 0) {
5631 spin_lock(&dest->lock);
5635 bytes_to_add = dest->size - dest->reserved;
5636 bytes_to_add = min(num_bytes, bytes_to_add);
5637 dest->reserved += bytes_to_add;
5638 if (dest->reserved >= dest->size)
5640 num_bytes -= bytes_to_add;
5642 spin_unlock(&dest->lock);
5645 space_info_add_old_bytes(fs_info, space_info,
5650 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5651 struct btrfs_block_rsv *dst, u64 num_bytes,
5656 ret = block_rsv_use_bytes(src, num_bytes);
5660 block_rsv_add_bytes(dst, num_bytes, update_size);
5664 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5666 memset(rsv, 0, sizeof(*rsv));
5667 spin_lock_init(&rsv->lock);
5671 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5672 unsigned short type)
5674 struct btrfs_block_rsv *block_rsv;
5676 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5680 btrfs_init_block_rsv(block_rsv, type);
5681 block_rsv->space_info = __find_space_info(fs_info,
5682 BTRFS_BLOCK_GROUP_METADATA);
5686 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5687 struct btrfs_block_rsv *rsv)
5691 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5695 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5700 int btrfs_block_rsv_add(struct btrfs_root *root,
5701 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5702 enum btrfs_reserve_flush_enum flush)
5709 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5711 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5718 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5726 spin_lock(&block_rsv->lock);
5727 num_bytes = div_factor(block_rsv->size, min_factor);
5728 if (block_rsv->reserved >= num_bytes)
5730 spin_unlock(&block_rsv->lock);
5735 int btrfs_block_rsv_refill(struct btrfs_root *root,
5736 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5737 enum btrfs_reserve_flush_enum flush)
5745 spin_lock(&block_rsv->lock);
5746 num_bytes = min_reserved;
5747 if (block_rsv->reserved >= num_bytes)
5750 num_bytes -= block_rsv->reserved;
5751 spin_unlock(&block_rsv->lock);
5756 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5758 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5765 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5766 struct btrfs_block_rsv *block_rsv,
5769 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5771 if (global_rsv == block_rsv ||
5772 block_rsv->space_info != global_rsv->space_info)
5774 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5777 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5779 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5780 struct btrfs_space_info *sinfo = block_rsv->space_info;
5784 * The global block rsv is based on the size of the extent tree, the
5785 * checksum tree and the root tree. If the fs is empty we want to set
5786 * it to a minimal amount for safety.
5788 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5789 btrfs_root_used(&fs_info->csum_root->root_item) +
5790 btrfs_root_used(&fs_info->tree_root->root_item);
5791 num_bytes = max_t(u64, num_bytes, SZ_16M);
5793 spin_lock(&sinfo->lock);
5794 spin_lock(&block_rsv->lock);
5796 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5798 if (block_rsv->reserved < block_rsv->size) {
5799 num_bytes = btrfs_space_info_used(sinfo, true);
5800 if (sinfo->total_bytes > num_bytes) {
5801 num_bytes = sinfo->total_bytes - num_bytes;
5802 num_bytes = min(num_bytes,
5803 block_rsv->size - block_rsv->reserved);
5804 block_rsv->reserved += num_bytes;
5805 sinfo->bytes_may_use += num_bytes;
5806 trace_btrfs_space_reservation(fs_info, "space_info",
5807 sinfo->flags, num_bytes,
5810 } else if (block_rsv->reserved > block_rsv->size) {
5811 num_bytes = block_rsv->reserved - block_rsv->size;
5812 sinfo->bytes_may_use -= num_bytes;
5813 trace_btrfs_space_reservation(fs_info, "space_info",
5814 sinfo->flags, num_bytes, 0);
5815 block_rsv->reserved = block_rsv->size;
5818 if (block_rsv->reserved == block_rsv->size)
5819 block_rsv->full = 1;
5821 block_rsv->full = 0;
5823 spin_unlock(&block_rsv->lock);
5824 spin_unlock(&sinfo->lock);
5827 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5829 struct btrfs_space_info *space_info;
5831 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5832 fs_info->chunk_block_rsv.space_info = space_info;
5834 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5835 fs_info->global_block_rsv.space_info = space_info;
5836 fs_info->delalloc_block_rsv.space_info = space_info;
5837 fs_info->trans_block_rsv.space_info = space_info;
5838 fs_info->empty_block_rsv.space_info = space_info;
5839 fs_info->delayed_block_rsv.space_info = space_info;
5841 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5842 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5843 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5844 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5845 if (fs_info->quota_root)
5846 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5847 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5849 update_global_block_rsv(fs_info);
5852 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5854 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5856 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5857 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5858 WARN_ON(fs_info->trans_block_rsv.size > 0);
5859 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5860 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5861 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5862 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5863 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5866 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5867 struct btrfs_fs_info *fs_info)
5869 if (!trans->block_rsv)
5872 if (!trans->bytes_reserved)
5875 trace_btrfs_space_reservation(fs_info, "transaction",
5876 trans->transid, trans->bytes_reserved, 0);
5877 btrfs_block_rsv_release(fs_info, trans->block_rsv,
5878 trans->bytes_reserved);
5879 trans->bytes_reserved = 0;
5883 * To be called after all the new block groups attached to the transaction
5884 * handle have been created (btrfs_create_pending_block_groups()).
5886 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5888 struct btrfs_fs_info *fs_info = trans->fs_info;
5890 if (!trans->chunk_bytes_reserved)
5893 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5895 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5896 trans->chunk_bytes_reserved);
5897 trans->chunk_bytes_reserved = 0;
5900 /* Can only return 0 or -ENOSPC */
5901 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5902 struct btrfs_inode *inode)
5904 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5905 struct btrfs_root *root = inode->root;
5907 * We always use trans->block_rsv here as we will have reserved space
5908 * for our orphan when starting the transaction, using get_block_rsv()
5909 * here will sometimes make us choose the wrong block rsv as we could be
5910 * doing a reloc inode for a non refcounted root.
5912 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5913 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5916 * We need to hold space in order to delete our orphan item once we've
5917 * added it, so this takes the reservation so we can release it later
5918 * when we are truly done with the orphan item.
5920 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5922 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5924 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5927 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5929 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5930 struct btrfs_root *root = inode->root;
5931 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5933 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5935 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5939 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5940 * root: the root of the parent directory
5941 * rsv: block reservation
5942 * items: the number of items that we need do reservation
5943 * qgroup_reserved: used to return the reserved size in qgroup
5945 * This function is used to reserve the space for snapshot/subvolume
5946 * creation and deletion. Those operations are different with the
5947 * common file/directory operations, they change two fs/file trees
5948 * and root tree, the number of items that the qgroup reserves is
5949 * different with the free space reservation. So we can not use
5950 * the space reservation mechanism in start_transaction().
5952 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5953 struct btrfs_block_rsv *rsv,
5955 u64 *qgroup_reserved,
5956 bool use_global_rsv)
5960 struct btrfs_fs_info *fs_info = root->fs_info;
5961 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5963 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5964 /* One for parent inode, two for dir entries */
5965 num_bytes = 3 * fs_info->nodesize;
5966 ret = btrfs_qgroup_reserve_meta(root, num_bytes, true);
5973 *qgroup_reserved = num_bytes;
5975 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5976 rsv->space_info = __find_space_info(fs_info,
5977 BTRFS_BLOCK_GROUP_METADATA);
5978 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5979 BTRFS_RESERVE_FLUSH_ALL);
5981 if (ret == -ENOSPC && use_global_rsv)
5982 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5984 if (ret && *qgroup_reserved)
5985 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5990 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5991 struct btrfs_block_rsv *rsv)
5993 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5997 * drop_outstanding_extent - drop an outstanding extent
5998 * @inode: the inode we're dropping the extent for
5999 * @num_bytes: the number of bytes we're releasing.
6001 * This is called when we are freeing up an outstanding extent, either called
6002 * after an error or after an extent is written. This will return the number of
6003 * reserved extents that need to be freed. This must be called with
6004 * BTRFS_I(inode)->lock held.
6006 static unsigned drop_outstanding_extent(struct btrfs_inode *inode,
6009 unsigned drop_inode_space = 0;
6010 unsigned dropped_extents = 0;
6011 unsigned num_extents;
6013 num_extents = count_max_extents(num_bytes);
6014 ASSERT(num_extents);
6015 ASSERT(inode->outstanding_extents >= num_extents);
6016 inode->outstanding_extents -= num_extents;
6018 if (inode->outstanding_extents == 0 &&
6019 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6020 &inode->runtime_flags))
6021 drop_inode_space = 1;
6024 * If we have more or the same amount of outstanding extents than we have
6025 * reserved then we need to leave the reserved extents count alone.
6027 if (inode->outstanding_extents >= inode->reserved_extents)
6028 return drop_inode_space;
6030 dropped_extents = inode->reserved_extents - inode->outstanding_extents;
6031 inode->reserved_extents -= dropped_extents;
6032 return dropped_extents + drop_inode_space;
6036 * calc_csum_metadata_size - return the amount of metadata space that must be
6037 * reserved/freed for the given bytes.
6038 * @inode: the inode we're manipulating
6039 * @num_bytes: the number of bytes in question
6040 * @reserve: 1 if we are reserving space, 0 if we are freeing space
6042 * This adjusts the number of csum_bytes in the inode and then returns the
6043 * correct amount of metadata that must either be reserved or freed. We
6044 * calculate how many checksums we can fit into one leaf and then divide the
6045 * number of bytes that will need to be checksumed by this value to figure out
6046 * how many checksums will be required. If we are adding bytes then the number
6047 * may go up and we will return the number of additional bytes that must be
6048 * reserved. If it is going down we will return the number of bytes that must
6051 * This must be called with BTRFS_I(inode)->lock held.
6053 static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes,
6056 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6057 u64 old_csums, num_csums;
6059 if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0)
6062 old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
6064 inode->csum_bytes += num_bytes;
6066 inode->csum_bytes -= num_bytes;
6067 num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
6069 /* No change, no need to reserve more */
6070 if (old_csums == num_csums)
6074 return btrfs_calc_trans_metadata_size(fs_info,
6075 num_csums - old_csums);
6077 return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums);
6080 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6082 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6083 struct btrfs_root *root = inode->root;
6084 struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv;
6087 unsigned nr_extents;
6088 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6090 bool delalloc_lock = true;
6093 bool release_extra = false;
6095 /* If we are a free space inode we need to not flush since we will be in
6096 * the middle of a transaction commit. We also don't need the delalloc
6097 * mutex since we won't race with anybody. We need this mostly to make
6098 * lockdep shut its filthy mouth.
6100 * If we have a transaction open (can happen if we call truncate_block
6101 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6103 if (btrfs_is_free_space_inode(inode)) {
6104 flush = BTRFS_RESERVE_NO_FLUSH;
6105 delalloc_lock = false;
6106 } else if (current->journal_info) {
6107 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6110 if (flush != BTRFS_RESERVE_NO_FLUSH &&
6111 btrfs_transaction_in_commit(fs_info))
6112 schedule_timeout(1);
6115 mutex_lock(&inode->delalloc_mutex);
6117 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6119 spin_lock(&inode->lock);
6120 nr_extents = count_max_extents(num_bytes);
6121 inode->outstanding_extents += nr_extents;
6124 if (inode->outstanding_extents > inode->reserved_extents)
6125 nr_extents += inode->outstanding_extents -
6126 inode->reserved_extents;
6128 /* We always want to reserve a slot for updating the inode. */
6129 to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1);
6130 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
6131 csum_bytes = inode->csum_bytes;
6132 spin_unlock(&inode->lock);
6134 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6135 ret = btrfs_qgroup_reserve_meta(root,
6136 nr_extents * fs_info->nodesize, true);
6141 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
6142 if (unlikely(ret)) {
6143 btrfs_qgroup_free_meta(root,
6144 nr_extents * fs_info->nodesize);
6148 spin_lock(&inode->lock);
6149 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6150 &inode->runtime_flags)) {
6151 to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1);
6152 release_extra = true;
6154 inode->reserved_extents += nr_extents;
6155 spin_unlock(&inode->lock);
6158 mutex_unlock(&inode->delalloc_mutex);
6161 trace_btrfs_space_reservation(fs_info, "delalloc",
6162 btrfs_ino(inode), to_reserve, 1);
6164 btrfs_block_rsv_release(fs_info, block_rsv,
6165 btrfs_calc_trans_metadata_size(fs_info, 1));
6169 spin_lock(&inode->lock);
6170 dropped = drop_outstanding_extent(inode, num_bytes);
6172 * If the inodes csum_bytes is the same as the original
6173 * csum_bytes then we know we haven't raced with any free()ers
6174 * so we can just reduce our inodes csum bytes and carry on.
6176 if (inode->csum_bytes == csum_bytes) {
6177 calc_csum_metadata_size(inode, num_bytes, 0);
6179 u64 orig_csum_bytes = inode->csum_bytes;
6183 * This is tricky, but first we need to figure out how much we
6184 * freed from any free-ers that occurred during this
6185 * reservation, so we reset ->csum_bytes to the csum_bytes
6186 * before we dropped our lock, and then call the free for the
6187 * number of bytes that were freed while we were trying our
6190 bytes = csum_bytes - inode->csum_bytes;
6191 inode->csum_bytes = csum_bytes;
6192 to_free = calc_csum_metadata_size(inode, bytes, 0);
6196 * Now we need to see how much we would have freed had we not
6197 * been making this reservation and our ->csum_bytes were not
6198 * artificially inflated.
6200 inode->csum_bytes = csum_bytes - num_bytes;
6201 bytes = csum_bytes - orig_csum_bytes;
6202 bytes = calc_csum_metadata_size(inode, bytes, 0);
6205 * Now reset ->csum_bytes to what it should be. If bytes is
6206 * more than to_free then we would have freed more space had we
6207 * not had an artificially high ->csum_bytes, so we need to free
6208 * the remainder. If bytes is the same or less then we don't
6209 * need to do anything, the other free-ers did the correct
6212 inode->csum_bytes = orig_csum_bytes - num_bytes;
6213 if (bytes > to_free)
6214 to_free = bytes - to_free;
6218 spin_unlock(&inode->lock);
6220 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6223 btrfs_block_rsv_release(fs_info, block_rsv, to_free);
6224 trace_btrfs_space_reservation(fs_info, "delalloc",
6225 btrfs_ino(inode), to_free, 0);
6228 mutex_unlock(&inode->delalloc_mutex);
6233 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6234 * @inode: the inode to release the reservation for
6235 * @num_bytes: the number of bytes we're releasing
6237 * This will release the metadata reservation for an inode. This can be called
6238 * once we complete IO for a given set of bytes to release their metadata
6241 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes)
6243 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6247 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6248 spin_lock(&inode->lock);
6249 dropped = drop_outstanding_extent(inode, num_bytes);
6252 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6253 spin_unlock(&inode->lock);
6255 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6257 if (btrfs_is_testing(fs_info))
6260 trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode),
6263 btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free);
6267 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6269 * @inode: inode we're writing to
6270 * @start: start range we are writing to
6271 * @len: how long the range we are writing to
6272 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6273 * current reservation.
6275 * This will do the following things
6277 * o reserve space in data space info for num bytes
6278 * and reserve precious corresponding qgroup space
6279 * (Done in check_data_free_space)
6281 * o reserve space for metadata space, based on the number of outstanding
6282 * extents and how much csums will be needed
6283 * also reserve metadata space in a per root over-reserve method.
6284 * o add to the inodes->delalloc_bytes
6285 * o add it to the fs_info's delalloc inodes list.
6286 * (Above 3 all done in delalloc_reserve_metadata)
6288 * Return 0 for success
6289 * Return <0 for error(-ENOSPC or -EQUOT)
6291 int btrfs_delalloc_reserve_space(struct inode *inode,
6292 struct extent_changeset **reserved, u64 start, u64 len)
6296 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6299 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6301 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6306 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6307 * @inode: inode we're releasing space for
6308 * @start: start position of the space already reserved
6309 * @len: the len of the space already reserved
6311 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6312 * called in the case that we don't need the metadata AND data reservations
6313 * anymore. So if there is an error or we insert an inline extent.
6315 * This function will release the metadata space that was not used and will
6316 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6317 * list if there are no delalloc bytes left.
6318 * Also it will handle the qgroup reserved space.
6320 void btrfs_delalloc_release_space(struct inode *inode,
6321 struct extent_changeset *reserved, u64 start, u64 len)
6323 btrfs_delalloc_release_metadata(BTRFS_I(inode), len);
6324 btrfs_free_reserved_data_space(inode, reserved, start, len);
6327 static int update_block_group(struct btrfs_trans_handle *trans,
6328 struct btrfs_fs_info *info, u64 bytenr,
6329 u64 num_bytes, int alloc)
6331 struct btrfs_block_group_cache *cache = NULL;
6332 u64 total = num_bytes;
6337 /* block accounting for super block */
6338 spin_lock(&info->delalloc_root_lock);
6339 old_val = btrfs_super_bytes_used(info->super_copy);
6341 old_val += num_bytes;
6343 old_val -= num_bytes;
6344 btrfs_set_super_bytes_used(info->super_copy, old_val);
6345 spin_unlock(&info->delalloc_root_lock);
6348 cache = btrfs_lookup_block_group(info, bytenr);
6351 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6352 BTRFS_BLOCK_GROUP_RAID1 |
6353 BTRFS_BLOCK_GROUP_RAID10))
6358 * If this block group has free space cache written out, we
6359 * need to make sure to load it if we are removing space. This
6360 * is because we need the unpinning stage to actually add the
6361 * space back to the block group, otherwise we will leak space.
6363 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6364 cache_block_group(cache, 1);
6366 byte_in_group = bytenr - cache->key.objectid;
6367 WARN_ON(byte_in_group > cache->key.offset);
6369 spin_lock(&cache->space_info->lock);
6370 spin_lock(&cache->lock);
6372 if (btrfs_test_opt(info, SPACE_CACHE) &&
6373 cache->disk_cache_state < BTRFS_DC_CLEAR)
6374 cache->disk_cache_state = BTRFS_DC_CLEAR;
6376 old_val = btrfs_block_group_used(&cache->item);
6377 num_bytes = min(total, cache->key.offset - byte_in_group);
6379 old_val += num_bytes;
6380 btrfs_set_block_group_used(&cache->item, old_val);
6381 cache->reserved -= num_bytes;
6382 cache->space_info->bytes_reserved -= num_bytes;
6383 cache->space_info->bytes_used += num_bytes;
6384 cache->space_info->disk_used += num_bytes * factor;
6385 spin_unlock(&cache->lock);
6386 spin_unlock(&cache->space_info->lock);
6388 old_val -= num_bytes;
6389 btrfs_set_block_group_used(&cache->item, old_val);
6390 cache->pinned += num_bytes;
6391 cache->space_info->bytes_pinned += num_bytes;
6392 cache->space_info->bytes_used -= num_bytes;
6393 cache->space_info->disk_used -= num_bytes * factor;
6394 spin_unlock(&cache->lock);
6395 spin_unlock(&cache->space_info->lock);
6397 trace_btrfs_space_reservation(info, "pinned",
6398 cache->space_info->flags,
6400 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6402 set_extent_dirty(info->pinned_extents,
6403 bytenr, bytenr + num_bytes - 1,
6404 GFP_NOFS | __GFP_NOFAIL);
6407 spin_lock(&trans->transaction->dirty_bgs_lock);
6408 if (list_empty(&cache->dirty_list)) {
6409 list_add_tail(&cache->dirty_list,
6410 &trans->transaction->dirty_bgs);
6411 trans->transaction->num_dirty_bgs++;
6412 btrfs_get_block_group(cache);
6414 spin_unlock(&trans->transaction->dirty_bgs_lock);
6417 * No longer have used bytes in this block group, queue it for
6418 * deletion. We do this after adding the block group to the
6419 * dirty list to avoid races between cleaner kthread and space
6422 if (!alloc && old_val == 0) {
6423 spin_lock(&info->unused_bgs_lock);
6424 if (list_empty(&cache->bg_list)) {
6425 btrfs_get_block_group(cache);
6426 list_add_tail(&cache->bg_list,
6429 spin_unlock(&info->unused_bgs_lock);
6432 btrfs_put_block_group(cache);
6434 bytenr += num_bytes;
6439 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6441 struct btrfs_block_group_cache *cache;
6444 spin_lock(&fs_info->block_group_cache_lock);
6445 bytenr = fs_info->first_logical_byte;
6446 spin_unlock(&fs_info->block_group_cache_lock);
6448 if (bytenr < (u64)-1)
6451 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6455 bytenr = cache->key.objectid;
6456 btrfs_put_block_group(cache);
6461 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6462 struct btrfs_block_group_cache *cache,
6463 u64 bytenr, u64 num_bytes, int reserved)
6465 spin_lock(&cache->space_info->lock);
6466 spin_lock(&cache->lock);
6467 cache->pinned += num_bytes;
6468 cache->space_info->bytes_pinned += num_bytes;
6470 cache->reserved -= num_bytes;
6471 cache->space_info->bytes_reserved -= num_bytes;
6473 spin_unlock(&cache->lock);
6474 spin_unlock(&cache->space_info->lock);
6476 trace_btrfs_space_reservation(fs_info, "pinned",
6477 cache->space_info->flags, num_bytes, 1);
6478 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6479 set_extent_dirty(fs_info->pinned_extents, bytenr,
6480 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6485 * this function must be called within transaction
6487 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6488 u64 bytenr, u64 num_bytes, int reserved)
6490 struct btrfs_block_group_cache *cache;
6492 cache = btrfs_lookup_block_group(fs_info, bytenr);
6493 BUG_ON(!cache); /* Logic error */
6495 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6497 btrfs_put_block_group(cache);
6502 * this function must be called within transaction
6504 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6505 u64 bytenr, u64 num_bytes)
6507 struct btrfs_block_group_cache *cache;
6510 cache = btrfs_lookup_block_group(fs_info, bytenr);
6515 * pull in the free space cache (if any) so that our pin
6516 * removes the free space from the cache. We have load_only set
6517 * to one because the slow code to read in the free extents does check
6518 * the pinned extents.
6520 cache_block_group(cache, 1);
6522 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6524 /* remove us from the free space cache (if we're there at all) */
6525 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6526 btrfs_put_block_group(cache);
6530 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6531 u64 start, u64 num_bytes)
6534 struct btrfs_block_group_cache *block_group;
6535 struct btrfs_caching_control *caching_ctl;
6537 block_group = btrfs_lookup_block_group(fs_info, start);
6541 cache_block_group(block_group, 0);
6542 caching_ctl = get_caching_control(block_group);
6546 BUG_ON(!block_group_cache_done(block_group));
6547 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6549 mutex_lock(&caching_ctl->mutex);
6551 if (start >= caching_ctl->progress) {
6552 ret = add_excluded_extent(fs_info, start, num_bytes);
6553 } else if (start + num_bytes <= caching_ctl->progress) {
6554 ret = btrfs_remove_free_space(block_group,
6557 num_bytes = caching_ctl->progress - start;
6558 ret = btrfs_remove_free_space(block_group,
6563 num_bytes = (start + num_bytes) -
6564 caching_ctl->progress;
6565 start = caching_ctl->progress;
6566 ret = add_excluded_extent(fs_info, start, num_bytes);
6569 mutex_unlock(&caching_ctl->mutex);
6570 put_caching_control(caching_ctl);
6572 btrfs_put_block_group(block_group);
6576 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6577 struct extent_buffer *eb)
6579 struct btrfs_file_extent_item *item;
6580 struct btrfs_key key;
6584 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6587 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6588 btrfs_item_key_to_cpu(eb, &key, i);
6589 if (key.type != BTRFS_EXTENT_DATA_KEY)
6591 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6592 found_type = btrfs_file_extent_type(eb, item);
6593 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6595 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6597 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6598 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6599 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6606 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6608 atomic_inc(&bg->reservations);
6611 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6614 struct btrfs_block_group_cache *bg;
6616 bg = btrfs_lookup_block_group(fs_info, start);
6618 if (atomic_dec_and_test(&bg->reservations))
6619 wake_up_atomic_t(&bg->reservations);
6620 btrfs_put_block_group(bg);
6623 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6629 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6631 struct btrfs_space_info *space_info = bg->space_info;
6635 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6639 * Our block group is read only but before we set it to read only,
6640 * some task might have had allocated an extent from it already, but it
6641 * has not yet created a respective ordered extent (and added it to a
6642 * root's list of ordered extents).
6643 * Therefore wait for any task currently allocating extents, since the
6644 * block group's reservations counter is incremented while a read lock
6645 * on the groups' semaphore is held and decremented after releasing
6646 * the read access on that semaphore and creating the ordered extent.
6648 down_write(&space_info->groups_sem);
6649 up_write(&space_info->groups_sem);
6651 wait_on_atomic_t(&bg->reservations,
6652 btrfs_wait_bg_reservations_atomic_t,
6653 TASK_UNINTERRUPTIBLE);
6657 * btrfs_add_reserved_bytes - update the block_group and space info counters
6658 * @cache: The cache we are manipulating
6659 * @ram_bytes: The number of bytes of file content, and will be same to
6660 * @num_bytes except for the compress path.
6661 * @num_bytes: The number of bytes in question
6662 * @delalloc: The blocks are allocated for the delalloc write
6664 * This is called by the allocator when it reserves space. If this is a
6665 * reservation and the block group has become read only we cannot make the
6666 * reservation and return -EAGAIN, otherwise this function always succeeds.
6668 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6669 u64 ram_bytes, u64 num_bytes, int delalloc)
6671 struct btrfs_space_info *space_info = cache->space_info;
6674 spin_lock(&space_info->lock);
6675 spin_lock(&cache->lock);
6679 cache->reserved += num_bytes;
6680 space_info->bytes_reserved += num_bytes;
6682 trace_btrfs_space_reservation(cache->fs_info,
6683 "space_info", space_info->flags,
6685 space_info->bytes_may_use -= ram_bytes;
6687 cache->delalloc_bytes += num_bytes;
6689 spin_unlock(&cache->lock);
6690 spin_unlock(&space_info->lock);
6695 * btrfs_free_reserved_bytes - update the block_group and space info counters
6696 * @cache: The cache we are manipulating
6697 * @num_bytes: The number of bytes in question
6698 * @delalloc: The blocks are allocated for the delalloc write
6700 * This is called by somebody who is freeing space that was never actually used
6701 * on disk. For example if you reserve some space for a new leaf in transaction
6702 * A and before transaction A commits you free that leaf, you call this with
6703 * reserve set to 0 in order to clear the reservation.
6706 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6707 u64 num_bytes, int delalloc)
6709 struct btrfs_space_info *space_info = cache->space_info;
6712 spin_lock(&space_info->lock);
6713 spin_lock(&cache->lock);
6715 space_info->bytes_readonly += num_bytes;
6716 cache->reserved -= num_bytes;
6717 space_info->bytes_reserved -= num_bytes;
6720 cache->delalloc_bytes -= num_bytes;
6721 spin_unlock(&cache->lock);
6722 spin_unlock(&space_info->lock);
6725 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6727 struct btrfs_caching_control *next;
6728 struct btrfs_caching_control *caching_ctl;
6729 struct btrfs_block_group_cache *cache;
6731 down_write(&fs_info->commit_root_sem);
6733 list_for_each_entry_safe(caching_ctl, next,
6734 &fs_info->caching_block_groups, list) {
6735 cache = caching_ctl->block_group;
6736 if (block_group_cache_done(cache)) {
6737 cache->last_byte_to_unpin = (u64)-1;
6738 list_del_init(&caching_ctl->list);
6739 put_caching_control(caching_ctl);
6741 cache->last_byte_to_unpin = caching_ctl->progress;
6745 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6746 fs_info->pinned_extents = &fs_info->freed_extents[1];
6748 fs_info->pinned_extents = &fs_info->freed_extents[0];
6750 up_write(&fs_info->commit_root_sem);
6752 update_global_block_rsv(fs_info);
6756 * Returns the free cluster for the given space info and sets empty_cluster to
6757 * what it should be based on the mount options.
6759 static struct btrfs_free_cluster *
6760 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6761 struct btrfs_space_info *space_info, u64 *empty_cluster)
6763 struct btrfs_free_cluster *ret = NULL;
6766 if (btrfs_mixed_space_info(space_info))
6769 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6770 ret = &fs_info->meta_alloc_cluster;
6771 if (btrfs_test_opt(fs_info, SSD))
6772 *empty_cluster = SZ_2M;
6774 *empty_cluster = SZ_64K;
6775 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6776 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6777 *empty_cluster = SZ_2M;
6778 ret = &fs_info->data_alloc_cluster;
6784 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6786 const bool return_free_space)
6788 struct btrfs_block_group_cache *cache = NULL;
6789 struct btrfs_space_info *space_info;
6790 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6791 struct btrfs_free_cluster *cluster = NULL;
6793 u64 total_unpinned = 0;
6794 u64 empty_cluster = 0;
6797 while (start <= end) {
6800 start >= cache->key.objectid + cache->key.offset) {
6802 btrfs_put_block_group(cache);
6804 cache = btrfs_lookup_block_group(fs_info, start);
6805 BUG_ON(!cache); /* Logic error */
6807 cluster = fetch_cluster_info(fs_info,
6810 empty_cluster <<= 1;
6813 len = cache->key.objectid + cache->key.offset - start;
6814 len = min(len, end + 1 - start);
6816 if (start < cache->last_byte_to_unpin) {
6817 len = min(len, cache->last_byte_to_unpin - start);
6818 if (return_free_space)
6819 btrfs_add_free_space(cache, start, len);
6823 total_unpinned += len;
6824 space_info = cache->space_info;
6827 * If this space cluster has been marked as fragmented and we've
6828 * unpinned enough in this block group to potentially allow a
6829 * cluster to be created inside of it go ahead and clear the
6832 if (cluster && cluster->fragmented &&
6833 total_unpinned > empty_cluster) {
6834 spin_lock(&cluster->lock);
6835 cluster->fragmented = 0;
6836 spin_unlock(&cluster->lock);
6839 spin_lock(&space_info->lock);
6840 spin_lock(&cache->lock);
6841 cache->pinned -= len;
6842 space_info->bytes_pinned -= len;
6844 trace_btrfs_space_reservation(fs_info, "pinned",
6845 space_info->flags, len, 0);
6846 space_info->max_extent_size = 0;
6847 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6849 space_info->bytes_readonly += len;
6852 spin_unlock(&cache->lock);
6853 if (!readonly && return_free_space &&
6854 global_rsv->space_info == space_info) {
6857 spin_lock(&global_rsv->lock);
6858 if (!global_rsv->full) {
6859 to_add = min(len, global_rsv->size -
6860 global_rsv->reserved);
6861 global_rsv->reserved += to_add;
6862 space_info->bytes_may_use += to_add;
6863 if (global_rsv->reserved >= global_rsv->size)
6864 global_rsv->full = 1;
6865 trace_btrfs_space_reservation(fs_info,
6871 spin_unlock(&global_rsv->lock);
6872 /* Add to any tickets we may have */
6874 space_info_add_new_bytes(fs_info, space_info,
6877 spin_unlock(&space_info->lock);
6881 btrfs_put_block_group(cache);
6885 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6886 struct btrfs_fs_info *fs_info)
6888 struct btrfs_block_group_cache *block_group, *tmp;
6889 struct list_head *deleted_bgs;
6890 struct extent_io_tree *unpin;
6895 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6896 unpin = &fs_info->freed_extents[1];
6898 unpin = &fs_info->freed_extents[0];
6900 while (!trans->aborted) {
6901 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6902 ret = find_first_extent_bit(unpin, 0, &start, &end,
6903 EXTENT_DIRTY, NULL);
6905 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6909 if (btrfs_test_opt(fs_info, DISCARD))
6910 ret = btrfs_discard_extent(fs_info, start,
6911 end + 1 - start, NULL);
6913 clear_extent_dirty(unpin, start, end);
6914 unpin_extent_range(fs_info, start, end, true);
6915 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6920 * Transaction is finished. We don't need the lock anymore. We
6921 * do need to clean up the block groups in case of a transaction
6924 deleted_bgs = &trans->transaction->deleted_bgs;
6925 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6929 if (!trans->aborted)
6930 ret = btrfs_discard_extent(fs_info,
6931 block_group->key.objectid,
6932 block_group->key.offset,
6935 list_del_init(&block_group->bg_list);
6936 btrfs_put_block_group_trimming(block_group);
6937 btrfs_put_block_group(block_group);
6940 const char *errstr = btrfs_decode_error(ret);
6942 "discard failed while removing blockgroup: errno=%d %s",
6950 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6951 struct btrfs_fs_info *info,
6952 struct btrfs_delayed_ref_node *node, u64 parent,
6953 u64 root_objectid, u64 owner_objectid,
6954 u64 owner_offset, int refs_to_drop,
6955 struct btrfs_delayed_extent_op *extent_op)
6957 struct btrfs_key key;
6958 struct btrfs_path *path;
6959 struct btrfs_root *extent_root = info->extent_root;
6960 struct extent_buffer *leaf;
6961 struct btrfs_extent_item *ei;
6962 struct btrfs_extent_inline_ref *iref;
6965 int extent_slot = 0;
6966 int found_extent = 0;
6970 u64 bytenr = node->bytenr;
6971 u64 num_bytes = node->num_bytes;
6973 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6975 path = btrfs_alloc_path();
6979 path->reada = READA_FORWARD;
6980 path->leave_spinning = 1;
6982 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6983 BUG_ON(!is_data && refs_to_drop != 1);
6986 skinny_metadata = 0;
6988 ret = lookup_extent_backref(trans, info, path, &iref,
6989 bytenr, num_bytes, parent,
6990 root_objectid, owner_objectid,
6993 extent_slot = path->slots[0];
6994 while (extent_slot >= 0) {
6995 btrfs_item_key_to_cpu(path->nodes[0], &key,
6997 if (key.objectid != bytenr)
6999 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
7000 key.offset == num_bytes) {
7004 if (key.type == BTRFS_METADATA_ITEM_KEY &&
7005 key.offset == owner_objectid) {
7009 if (path->slots[0] - extent_slot > 5)
7013 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7014 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
7015 if (found_extent && item_size < sizeof(*ei))
7018 if (!found_extent) {
7020 ret = remove_extent_backref(trans, info, path, NULL,
7022 is_data, &last_ref);
7024 btrfs_abort_transaction(trans, ret);
7027 btrfs_release_path(path);
7028 path->leave_spinning = 1;
7030 key.objectid = bytenr;
7031 key.type = BTRFS_EXTENT_ITEM_KEY;
7032 key.offset = num_bytes;
7034 if (!is_data && skinny_metadata) {
7035 key.type = BTRFS_METADATA_ITEM_KEY;
7036 key.offset = owner_objectid;
7039 ret = btrfs_search_slot(trans, extent_root,
7041 if (ret > 0 && skinny_metadata && path->slots[0]) {
7043 * Couldn't find our skinny metadata item,
7044 * see if we have ye olde extent item.
7047 btrfs_item_key_to_cpu(path->nodes[0], &key,
7049 if (key.objectid == bytenr &&
7050 key.type == BTRFS_EXTENT_ITEM_KEY &&
7051 key.offset == num_bytes)
7055 if (ret > 0 && skinny_metadata) {
7056 skinny_metadata = false;
7057 key.objectid = bytenr;
7058 key.type = BTRFS_EXTENT_ITEM_KEY;
7059 key.offset = num_bytes;
7060 btrfs_release_path(path);
7061 ret = btrfs_search_slot(trans, extent_root,
7067 "umm, got %d back from search, was looking for %llu",
7070 btrfs_print_leaf(path->nodes[0]);
7073 btrfs_abort_transaction(trans, ret);
7076 extent_slot = path->slots[0];
7078 } else if (WARN_ON(ret == -ENOENT)) {
7079 btrfs_print_leaf(path->nodes[0]);
7081 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7082 bytenr, parent, root_objectid, owner_objectid,
7084 btrfs_abort_transaction(trans, ret);
7087 btrfs_abort_transaction(trans, ret);
7091 leaf = path->nodes[0];
7092 item_size = btrfs_item_size_nr(leaf, extent_slot);
7093 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7094 if (item_size < sizeof(*ei)) {
7095 BUG_ON(found_extent || extent_slot != path->slots[0]);
7096 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
7099 btrfs_abort_transaction(trans, ret);
7103 btrfs_release_path(path);
7104 path->leave_spinning = 1;
7106 key.objectid = bytenr;
7107 key.type = BTRFS_EXTENT_ITEM_KEY;
7108 key.offset = num_bytes;
7110 ret = btrfs_search_slot(trans, extent_root, &key, path,
7114 "umm, got %d back from search, was looking for %llu",
7116 btrfs_print_leaf(path->nodes[0]);
7119 btrfs_abort_transaction(trans, ret);
7123 extent_slot = path->slots[0];
7124 leaf = path->nodes[0];
7125 item_size = btrfs_item_size_nr(leaf, extent_slot);
7128 BUG_ON(item_size < sizeof(*ei));
7129 ei = btrfs_item_ptr(leaf, extent_slot,
7130 struct btrfs_extent_item);
7131 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7132 key.type == BTRFS_EXTENT_ITEM_KEY) {
7133 struct btrfs_tree_block_info *bi;
7134 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7135 bi = (struct btrfs_tree_block_info *)(ei + 1);
7136 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7139 refs = btrfs_extent_refs(leaf, ei);
7140 if (refs < refs_to_drop) {
7142 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7143 refs_to_drop, refs, bytenr);
7145 btrfs_abort_transaction(trans, ret);
7148 refs -= refs_to_drop;
7152 __run_delayed_extent_op(extent_op, leaf, ei);
7154 * In the case of inline back ref, reference count will
7155 * be updated by remove_extent_backref
7158 BUG_ON(!found_extent);
7160 btrfs_set_extent_refs(leaf, ei, refs);
7161 btrfs_mark_buffer_dirty(leaf);
7164 ret = remove_extent_backref(trans, info, path,
7166 is_data, &last_ref);
7168 btrfs_abort_transaction(trans, ret);
7174 BUG_ON(is_data && refs_to_drop !=
7175 extent_data_ref_count(path, iref));
7177 BUG_ON(path->slots[0] != extent_slot);
7179 BUG_ON(path->slots[0] != extent_slot + 1);
7180 path->slots[0] = extent_slot;
7186 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7189 btrfs_abort_transaction(trans, ret);
7192 btrfs_release_path(path);
7195 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7197 btrfs_abort_transaction(trans, ret);
7202 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7204 btrfs_abort_transaction(trans, ret);
7208 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7210 btrfs_abort_transaction(trans, ret);
7214 btrfs_release_path(path);
7217 btrfs_free_path(path);
7222 * when we free an block, it is possible (and likely) that we free the last
7223 * delayed ref for that extent as well. This searches the delayed ref tree for
7224 * a given extent, and if there are no other delayed refs to be processed, it
7225 * removes it from the tree.
7227 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7230 struct btrfs_delayed_ref_head *head;
7231 struct btrfs_delayed_ref_root *delayed_refs;
7234 delayed_refs = &trans->transaction->delayed_refs;
7235 spin_lock(&delayed_refs->lock);
7236 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7238 goto out_delayed_unlock;
7240 spin_lock(&head->lock);
7241 if (!list_empty(&head->ref_list))
7244 if (head->extent_op) {
7245 if (!head->must_insert_reserved)
7247 btrfs_free_delayed_extent_op(head->extent_op);
7248 head->extent_op = NULL;
7252 * waiting for the lock here would deadlock. If someone else has it
7253 * locked they are already in the process of dropping it anyway
7255 if (!mutex_trylock(&head->mutex))
7259 * at this point we have a head with no other entries. Go
7260 * ahead and process it.
7262 head->node.in_tree = 0;
7263 rb_erase(&head->href_node, &delayed_refs->href_root);
7265 atomic_dec(&delayed_refs->num_entries);
7268 * we don't take a ref on the node because we're removing it from the
7269 * tree, so we just steal the ref the tree was holding.
7271 delayed_refs->num_heads--;
7272 if (head->processing == 0)
7273 delayed_refs->num_heads_ready--;
7274 head->processing = 0;
7275 spin_unlock(&head->lock);
7276 spin_unlock(&delayed_refs->lock);
7278 BUG_ON(head->extent_op);
7279 if (head->must_insert_reserved)
7282 mutex_unlock(&head->mutex);
7283 btrfs_put_delayed_ref(&head->node);
7286 spin_unlock(&head->lock);
7289 spin_unlock(&delayed_refs->lock);
7293 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7294 struct btrfs_root *root,
7295 struct extent_buffer *buf,
7296 u64 parent, int last_ref)
7298 struct btrfs_fs_info *fs_info = root->fs_info;
7302 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7303 int old_ref_mod, new_ref_mod;
7305 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7307 root->root_key.objectid,
7308 btrfs_header_level(buf),
7309 BTRFS_DROP_DELAYED_REF, NULL,
7310 &old_ref_mod, &new_ref_mod);
7311 BUG_ON(ret); /* -ENOMEM */
7312 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7315 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7316 struct btrfs_block_group_cache *cache;
7318 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7319 ret = check_ref_cleanup(trans, buf->start);
7325 cache = btrfs_lookup_block_group(fs_info, buf->start);
7327 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7328 pin_down_extent(fs_info, cache, buf->start,
7330 btrfs_put_block_group(cache);
7334 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7336 btrfs_add_free_space(cache, buf->start, buf->len);
7337 btrfs_free_reserved_bytes(cache, buf->len, 0);
7338 btrfs_put_block_group(cache);
7339 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7343 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7344 root->root_key.objectid);
7348 * Deleting the buffer, clear the corrupt flag since it doesn't
7351 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7355 /* Can return -ENOMEM */
7356 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7357 struct btrfs_fs_info *fs_info,
7358 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7359 u64 owner, u64 offset)
7361 int old_ref_mod, new_ref_mod;
7364 if (btrfs_is_testing(fs_info))
7369 * tree log blocks never actually go into the extent allocation
7370 * tree, just update pinning info and exit early.
7372 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7373 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7374 /* unlocks the pinned mutex */
7375 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7376 old_ref_mod = new_ref_mod = 0;
7378 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7379 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7381 root_objectid, (int)owner,
7382 BTRFS_DROP_DELAYED_REF, NULL,
7383 &old_ref_mod, &new_ref_mod);
7385 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7387 root_objectid, owner, offset,
7388 0, BTRFS_DROP_DELAYED_REF,
7389 &old_ref_mod, &new_ref_mod);
7392 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7393 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7399 * when we wait for progress in the block group caching, its because
7400 * our allocation attempt failed at least once. So, we must sleep
7401 * and let some progress happen before we try again.
7403 * This function will sleep at least once waiting for new free space to
7404 * show up, and then it will check the block group free space numbers
7405 * for our min num_bytes. Another option is to have it go ahead
7406 * and look in the rbtree for a free extent of a given size, but this
7409 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7410 * any of the information in this block group.
7412 static noinline void
7413 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7416 struct btrfs_caching_control *caching_ctl;
7418 caching_ctl = get_caching_control(cache);
7422 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7423 (cache->free_space_ctl->free_space >= num_bytes));
7425 put_caching_control(caching_ctl);
7429 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7431 struct btrfs_caching_control *caching_ctl;
7434 caching_ctl = get_caching_control(cache);
7436 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7438 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7439 if (cache->cached == BTRFS_CACHE_ERROR)
7441 put_caching_control(caching_ctl);
7445 int __get_raid_index(u64 flags)
7447 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7448 return BTRFS_RAID_RAID10;
7449 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7450 return BTRFS_RAID_RAID1;
7451 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7452 return BTRFS_RAID_DUP;
7453 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7454 return BTRFS_RAID_RAID0;
7455 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7456 return BTRFS_RAID_RAID5;
7457 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7458 return BTRFS_RAID_RAID6;
7460 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7463 int get_block_group_index(struct btrfs_block_group_cache *cache)
7465 return __get_raid_index(cache->flags);
7468 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7469 [BTRFS_RAID_RAID10] = "raid10",
7470 [BTRFS_RAID_RAID1] = "raid1",
7471 [BTRFS_RAID_DUP] = "dup",
7472 [BTRFS_RAID_RAID0] = "raid0",
7473 [BTRFS_RAID_SINGLE] = "single",
7474 [BTRFS_RAID_RAID5] = "raid5",
7475 [BTRFS_RAID_RAID6] = "raid6",
7478 static const char *get_raid_name(enum btrfs_raid_types type)
7480 if (type >= BTRFS_NR_RAID_TYPES)
7483 return btrfs_raid_type_names[type];
7486 enum btrfs_loop_type {
7487 LOOP_CACHING_NOWAIT = 0,
7488 LOOP_CACHING_WAIT = 1,
7489 LOOP_ALLOC_CHUNK = 2,
7490 LOOP_NO_EMPTY_SIZE = 3,
7494 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7498 down_read(&cache->data_rwsem);
7502 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7505 btrfs_get_block_group(cache);
7507 down_read(&cache->data_rwsem);
7510 static struct btrfs_block_group_cache *
7511 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7512 struct btrfs_free_cluster *cluster,
7515 struct btrfs_block_group_cache *used_bg = NULL;
7517 spin_lock(&cluster->refill_lock);
7519 used_bg = cluster->block_group;
7523 if (used_bg == block_group)
7526 btrfs_get_block_group(used_bg);
7531 if (down_read_trylock(&used_bg->data_rwsem))
7534 spin_unlock(&cluster->refill_lock);
7536 /* We should only have one-level nested. */
7537 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7539 spin_lock(&cluster->refill_lock);
7540 if (used_bg == cluster->block_group)
7543 up_read(&used_bg->data_rwsem);
7544 btrfs_put_block_group(used_bg);
7549 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7553 up_read(&cache->data_rwsem);
7554 btrfs_put_block_group(cache);
7558 * walks the btree of allocated extents and find a hole of a given size.
7559 * The key ins is changed to record the hole:
7560 * ins->objectid == start position
7561 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7562 * ins->offset == the size of the hole.
7563 * Any available blocks before search_start are skipped.
7565 * If there is no suitable free space, we will record the max size of
7566 * the free space extent currently.
7568 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7569 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7570 u64 hint_byte, struct btrfs_key *ins,
7571 u64 flags, int delalloc)
7574 struct btrfs_root *root = fs_info->extent_root;
7575 struct btrfs_free_cluster *last_ptr = NULL;
7576 struct btrfs_block_group_cache *block_group = NULL;
7577 u64 search_start = 0;
7578 u64 max_extent_size = 0;
7579 u64 max_free_space = 0;
7580 u64 empty_cluster = 0;
7581 struct btrfs_space_info *space_info;
7583 int index = __get_raid_index(flags);
7584 bool failed_cluster_refill = false;
7585 bool failed_alloc = false;
7586 bool use_cluster = true;
7587 bool have_caching_bg = false;
7588 bool orig_have_caching_bg = false;
7589 bool full_search = false;
7591 WARN_ON(num_bytes < fs_info->sectorsize);
7592 ins->type = BTRFS_EXTENT_ITEM_KEY;
7596 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7598 space_info = __find_space_info(fs_info, flags);
7600 btrfs_err(fs_info, "No space info for %llu", flags);
7605 * If our free space is heavily fragmented we may not be able to make
7606 * big contiguous allocations, so instead of doing the expensive search
7607 * for free space, simply return ENOSPC with our max_extent_size so we
7608 * can go ahead and search for a more manageable chunk.
7610 * If our max_extent_size is large enough for our allocation simply
7611 * disable clustering since we will likely not be able to find enough
7612 * space to create a cluster and induce latency trying.
7614 if (unlikely(space_info->max_extent_size)) {
7615 spin_lock(&space_info->lock);
7616 if (space_info->max_extent_size &&
7617 num_bytes > space_info->max_extent_size) {
7618 ins->offset = space_info->max_extent_size;
7619 spin_unlock(&space_info->lock);
7621 } else if (space_info->max_extent_size) {
7622 use_cluster = false;
7624 spin_unlock(&space_info->lock);
7627 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7629 spin_lock(&last_ptr->lock);
7630 if (last_ptr->block_group)
7631 hint_byte = last_ptr->window_start;
7632 if (last_ptr->fragmented) {
7634 * We still set window_start so we can keep track of the
7635 * last place we found an allocation to try and save
7638 hint_byte = last_ptr->window_start;
7639 use_cluster = false;
7641 spin_unlock(&last_ptr->lock);
7644 search_start = max(search_start, first_logical_byte(fs_info, 0));
7645 search_start = max(search_start, hint_byte);
7646 if (search_start == hint_byte) {
7647 block_group = btrfs_lookup_block_group(fs_info, search_start);
7649 * we don't want to use the block group if it doesn't match our
7650 * allocation bits, or if its not cached.
7652 * However if we are re-searching with an ideal block group
7653 * picked out then we don't care that the block group is cached.
7655 if (block_group && block_group_bits(block_group, flags) &&
7656 block_group->cached != BTRFS_CACHE_NO) {
7657 down_read(&space_info->groups_sem);
7658 if (list_empty(&block_group->list) ||
7661 * someone is removing this block group,
7662 * we can't jump into the have_block_group
7663 * target because our list pointers are not
7666 btrfs_put_block_group(block_group);
7667 up_read(&space_info->groups_sem);
7669 index = get_block_group_index(block_group);
7670 btrfs_lock_block_group(block_group, delalloc);
7671 goto have_block_group;
7673 } else if (block_group) {
7674 btrfs_put_block_group(block_group);
7678 have_caching_bg = false;
7679 if (index == 0 || index == __get_raid_index(flags))
7681 down_read(&space_info->groups_sem);
7682 list_for_each_entry(block_group, &space_info->block_groups[index],
7687 /* If the block group is read-only, we can skip it entirely. */
7688 if (unlikely(block_group->ro))
7691 btrfs_grab_block_group(block_group, delalloc);
7692 search_start = block_group->key.objectid;
7695 * this can happen if we end up cycling through all the
7696 * raid types, but we want to make sure we only allocate
7697 * for the proper type.
7699 if (!block_group_bits(block_group, flags)) {
7700 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7701 BTRFS_BLOCK_GROUP_RAID1 |
7702 BTRFS_BLOCK_GROUP_RAID5 |
7703 BTRFS_BLOCK_GROUP_RAID6 |
7704 BTRFS_BLOCK_GROUP_RAID10;
7707 * if they asked for extra copies and this block group
7708 * doesn't provide them, bail. This does allow us to
7709 * fill raid0 from raid1.
7711 if ((flags & extra) && !(block_group->flags & extra))
7715 * This block group has different flags than we want.
7716 * It's possible that we have MIXED_GROUP flag but no
7717 * block group is mixed. Just skip such block group.
7719 btrfs_release_block_group(block_group, delalloc);
7724 cached = block_group_cache_done(block_group);
7725 if (unlikely(!cached)) {
7726 have_caching_bg = true;
7727 ret = cache_block_group(block_group, 0);
7732 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7736 * Ok we want to try and use the cluster allocator, so
7739 if (last_ptr && use_cluster) {
7740 struct btrfs_block_group_cache *used_block_group;
7741 unsigned long aligned_cluster;
7743 * the refill lock keeps out other
7744 * people trying to start a new cluster
7746 used_block_group = btrfs_lock_cluster(block_group,
7749 if (!used_block_group)
7750 goto refill_cluster;
7752 if (used_block_group != block_group &&
7753 (used_block_group->ro ||
7754 !block_group_bits(used_block_group, flags)))
7755 goto release_cluster;
7757 offset = btrfs_alloc_from_cluster(used_block_group,
7760 used_block_group->key.objectid,
7763 /* we have a block, we're done */
7764 spin_unlock(&last_ptr->refill_lock);
7765 trace_btrfs_reserve_extent_cluster(fs_info,
7767 search_start, num_bytes);
7768 if (used_block_group != block_group) {
7769 btrfs_release_block_group(block_group,
7771 block_group = used_block_group;
7776 WARN_ON(last_ptr->block_group != used_block_group);
7778 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7779 * set up a new clusters, so lets just skip it
7780 * and let the allocator find whatever block
7781 * it can find. If we reach this point, we
7782 * will have tried the cluster allocator
7783 * plenty of times and not have found
7784 * anything, so we are likely way too
7785 * fragmented for the clustering stuff to find
7788 * However, if the cluster is taken from the
7789 * current block group, release the cluster
7790 * first, so that we stand a better chance of
7791 * succeeding in the unclustered
7793 if (loop >= LOOP_NO_EMPTY_SIZE &&
7794 used_block_group != block_group) {
7795 spin_unlock(&last_ptr->refill_lock);
7796 btrfs_release_block_group(used_block_group,
7798 goto unclustered_alloc;
7802 * this cluster didn't work out, free it and
7805 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7807 if (used_block_group != block_group)
7808 btrfs_release_block_group(used_block_group,
7811 if (loop >= LOOP_NO_EMPTY_SIZE) {
7812 spin_unlock(&last_ptr->refill_lock);
7813 goto unclustered_alloc;
7816 aligned_cluster = max_t(unsigned long,
7817 empty_cluster + empty_size,
7818 block_group->full_stripe_len);
7820 /* allocate a cluster in this block group */
7821 ret = btrfs_find_space_cluster(fs_info, block_group,
7822 last_ptr, search_start,
7827 * now pull our allocation out of this
7830 offset = btrfs_alloc_from_cluster(block_group,
7836 /* we found one, proceed */
7837 spin_unlock(&last_ptr->refill_lock);
7838 trace_btrfs_reserve_extent_cluster(fs_info,
7839 block_group, search_start,
7843 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7844 && !failed_cluster_refill) {
7845 spin_unlock(&last_ptr->refill_lock);
7847 failed_cluster_refill = true;
7848 wait_block_group_cache_progress(block_group,
7849 num_bytes + empty_cluster + empty_size);
7850 goto have_block_group;
7854 * at this point we either didn't find a cluster
7855 * or we weren't able to allocate a block from our
7856 * cluster. Free the cluster we've been trying
7857 * to use, and go to the next block group
7859 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7860 spin_unlock(&last_ptr->refill_lock);
7866 * We are doing an unclustered alloc, set the fragmented flag so
7867 * we don't bother trying to setup a cluster again until we get
7870 if (unlikely(last_ptr)) {
7871 spin_lock(&last_ptr->lock);
7872 last_ptr->fragmented = 1;
7873 spin_unlock(&last_ptr->lock);
7876 struct btrfs_free_space_ctl *ctl =
7877 block_group->free_space_ctl;
7879 spin_lock(&ctl->tree_lock);
7880 if (ctl->free_space <
7881 num_bytes + empty_cluster + empty_size) {
7882 max_free_space = max(max_free_space,
7884 spin_unlock(&ctl->tree_lock);
7887 spin_unlock(&ctl->tree_lock);
7890 offset = btrfs_find_space_for_alloc(block_group, search_start,
7891 num_bytes, empty_size,
7894 * If we didn't find a chunk, and we haven't failed on this
7895 * block group before, and this block group is in the middle of
7896 * caching and we are ok with waiting, then go ahead and wait
7897 * for progress to be made, and set failed_alloc to true.
7899 * If failed_alloc is true then we've already waited on this
7900 * block group once and should move on to the next block group.
7902 if (!offset && !failed_alloc && !cached &&
7903 loop > LOOP_CACHING_NOWAIT) {
7904 wait_block_group_cache_progress(block_group,
7905 num_bytes + empty_size);
7906 failed_alloc = true;
7907 goto have_block_group;
7908 } else if (!offset) {
7912 search_start = ALIGN(offset, fs_info->stripesize);
7914 /* move on to the next group */
7915 if (search_start + num_bytes >
7916 block_group->key.objectid + block_group->key.offset) {
7917 btrfs_add_free_space(block_group, offset, num_bytes);
7921 if (offset < search_start)
7922 btrfs_add_free_space(block_group, offset,
7923 search_start - offset);
7924 BUG_ON(offset > search_start);
7926 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7927 num_bytes, delalloc);
7928 if (ret == -EAGAIN) {
7929 btrfs_add_free_space(block_group, offset, num_bytes);
7932 btrfs_inc_block_group_reservations(block_group);
7934 /* we are all good, lets return */
7935 ins->objectid = search_start;
7936 ins->offset = num_bytes;
7938 trace_btrfs_reserve_extent(fs_info, block_group,
7939 search_start, num_bytes);
7940 btrfs_release_block_group(block_group, delalloc);
7943 failed_cluster_refill = false;
7944 failed_alloc = false;
7945 BUG_ON(index != get_block_group_index(block_group));
7946 btrfs_release_block_group(block_group, delalloc);
7949 up_read(&space_info->groups_sem);
7951 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7952 && !orig_have_caching_bg)
7953 orig_have_caching_bg = true;
7955 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7958 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7962 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7963 * caching kthreads as we move along
7964 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7965 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7966 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7969 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7971 if (loop == LOOP_CACHING_NOWAIT) {
7973 * We want to skip the LOOP_CACHING_WAIT step if we
7974 * don't have any uncached bgs and we've already done a
7975 * full search through.
7977 if (orig_have_caching_bg || !full_search)
7978 loop = LOOP_CACHING_WAIT;
7980 loop = LOOP_ALLOC_CHUNK;
7985 if (loop == LOOP_ALLOC_CHUNK) {
7986 struct btrfs_trans_handle *trans;
7989 trans = current->journal_info;
7993 trans = btrfs_join_transaction(root);
7995 if (IS_ERR(trans)) {
7996 ret = PTR_ERR(trans);
8000 ret = do_chunk_alloc(trans, fs_info, flags,
8004 * If we can't allocate a new chunk we've already looped
8005 * through at least once, move on to the NO_EMPTY_SIZE
8009 loop = LOOP_NO_EMPTY_SIZE;
8012 * Do not bail out on ENOSPC since we
8013 * can do more things.
8015 if (ret < 0 && ret != -ENOSPC)
8016 btrfs_abort_transaction(trans, ret);
8020 btrfs_end_transaction(trans);
8025 if (loop == LOOP_NO_EMPTY_SIZE) {
8027 * Don't loop again if we already have no empty_size and
8030 if (empty_size == 0 &&
8031 empty_cluster == 0) {
8040 } else if (!ins->objectid) {
8042 } else if (ins->objectid) {
8043 if (!use_cluster && last_ptr) {
8044 spin_lock(&last_ptr->lock);
8045 last_ptr->window_start = ins->objectid;
8046 spin_unlock(&last_ptr->lock);
8051 if (ret == -ENOSPC) {
8052 if (!max_extent_size)
8053 max_extent_size = max_free_space;
8054 spin_lock(&space_info->lock);
8055 space_info->max_extent_size = max_extent_size;
8056 spin_unlock(&space_info->lock);
8057 ins->offset = max_extent_size;
8062 static void dump_space_info(struct btrfs_fs_info *fs_info,
8063 struct btrfs_space_info *info, u64 bytes,
8064 int dump_block_groups)
8066 struct btrfs_block_group_cache *cache;
8069 spin_lock(&info->lock);
8070 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
8072 info->total_bytes - btrfs_space_info_used(info, true),
8073 info->full ? "" : "not ");
8075 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
8076 info->total_bytes, info->bytes_used, info->bytes_pinned,
8077 info->bytes_reserved, info->bytes_may_use,
8078 info->bytes_readonly);
8079 spin_unlock(&info->lock);
8081 if (!dump_block_groups)
8084 down_read(&info->groups_sem);
8086 list_for_each_entry(cache, &info->block_groups[index], list) {
8087 spin_lock(&cache->lock);
8089 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8090 cache->key.objectid, cache->key.offset,
8091 btrfs_block_group_used(&cache->item), cache->pinned,
8092 cache->reserved, cache->ro ? "[readonly]" : "");
8093 btrfs_dump_free_space(cache, bytes);
8094 spin_unlock(&cache->lock);
8096 if (++index < BTRFS_NR_RAID_TYPES)
8098 up_read(&info->groups_sem);
8101 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8102 u64 num_bytes, u64 min_alloc_size,
8103 u64 empty_size, u64 hint_byte,
8104 struct btrfs_key *ins, int is_data, int delalloc)
8106 struct btrfs_fs_info *fs_info = root->fs_info;
8107 bool final_tried = num_bytes == min_alloc_size;
8111 flags = get_alloc_profile_by_root(root, is_data);
8113 WARN_ON(num_bytes < fs_info->sectorsize);
8114 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8115 hint_byte, ins, flags, delalloc);
8116 if (!ret && !is_data) {
8117 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8118 } else if (ret == -ENOSPC) {
8119 if (!final_tried && ins->offset) {
8120 num_bytes = min(num_bytes >> 1, ins->offset);
8121 num_bytes = round_down(num_bytes,
8122 fs_info->sectorsize);
8123 num_bytes = max(num_bytes, min_alloc_size);
8124 ram_bytes = num_bytes;
8125 if (num_bytes == min_alloc_size)
8128 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8129 struct btrfs_space_info *sinfo;
8131 sinfo = __find_space_info(fs_info, flags);
8133 "allocation failed flags %llu, wanted %llu",
8136 dump_space_info(fs_info, sinfo, num_bytes, 1);
8143 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8145 int pin, int delalloc)
8147 struct btrfs_block_group_cache *cache;
8150 cache = btrfs_lookup_block_group(fs_info, start);
8152 btrfs_err(fs_info, "Unable to find block group for %llu",
8158 pin_down_extent(fs_info, cache, start, len, 1);
8160 if (btrfs_test_opt(fs_info, DISCARD))
8161 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8162 btrfs_add_free_space(cache, start, len);
8163 btrfs_free_reserved_bytes(cache, len, delalloc);
8164 trace_btrfs_reserved_extent_free(fs_info, start, len);
8167 btrfs_put_block_group(cache);
8171 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8172 u64 start, u64 len, int delalloc)
8174 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8177 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8180 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8183 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8184 struct btrfs_fs_info *fs_info,
8185 u64 parent, u64 root_objectid,
8186 u64 flags, u64 owner, u64 offset,
8187 struct btrfs_key *ins, int ref_mod)
8190 struct btrfs_extent_item *extent_item;
8191 struct btrfs_extent_inline_ref *iref;
8192 struct btrfs_path *path;
8193 struct extent_buffer *leaf;
8198 type = BTRFS_SHARED_DATA_REF_KEY;
8200 type = BTRFS_EXTENT_DATA_REF_KEY;
8202 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8204 path = btrfs_alloc_path();
8208 path->leave_spinning = 1;
8209 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8212 btrfs_free_path(path);
8216 leaf = path->nodes[0];
8217 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8218 struct btrfs_extent_item);
8219 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8220 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8221 btrfs_set_extent_flags(leaf, extent_item,
8222 flags | BTRFS_EXTENT_FLAG_DATA);
8224 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8225 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8227 struct btrfs_shared_data_ref *ref;
8228 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8229 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8230 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8232 struct btrfs_extent_data_ref *ref;
8233 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8234 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8235 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8236 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8237 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8240 btrfs_mark_buffer_dirty(path->nodes[0]);
8241 btrfs_free_path(path);
8243 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8248 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8249 if (ret) { /* -ENOENT, logic error */
8250 btrfs_err(fs_info, "update block group failed for %llu %llu",
8251 ins->objectid, ins->offset);
8254 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8258 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8259 struct btrfs_fs_info *fs_info,
8260 u64 parent, u64 root_objectid,
8261 u64 flags, struct btrfs_disk_key *key,
8262 int level, struct btrfs_key *ins)
8265 struct btrfs_extent_item *extent_item;
8266 struct btrfs_tree_block_info *block_info;
8267 struct btrfs_extent_inline_ref *iref;
8268 struct btrfs_path *path;
8269 struct extent_buffer *leaf;
8270 u32 size = sizeof(*extent_item) + sizeof(*iref);
8271 u64 num_bytes = ins->offset;
8272 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8274 if (!skinny_metadata)
8275 size += sizeof(*block_info);
8277 path = btrfs_alloc_path();
8279 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8284 path->leave_spinning = 1;
8285 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8288 btrfs_free_path(path);
8289 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8294 leaf = path->nodes[0];
8295 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8296 struct btrfs_extent_item);
8297 btrfs_set_extent_refs(leaf, extent_item, 1);
8298 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8299 btrfs_set_extent_flags(leaf, extent_item,
8300 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8302 if (skinny_metadata) {
8303 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8304 num_bytes = fs_info->nodesize;
8306 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8307 btrfs_set_tree_block_key(leaf, block_info, key);
8308 btrfs_set_tree_block_level(leaf, block_info, level);
8309 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8313 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8314 btrfs_set_extent_inline_ref_type(leaf, iref,
8315 BTRFS_SHARED_BLOCK_REF_KEY);
8316 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8318 btrfs_set_extent_inline_ref_type(leaf, iref,
8319 BTRFS_TREE_BLOCK_REF_KEY);
8320 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8323 btrfs_mark_buffer_dirty(leaf);
8324 btrfs_free_path(path);
8326 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8331 ret = update_block_group(trans, fs_info, ins->objectid,
8332 fs_info->nodesize, 1);
8333 if (ret) { /* -ENOENT, logic error */
8334 btrfs_err(fs_info, "update block group failed for %llu %llu",
8335 ins->objectid, ins->offset);
8339 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8344 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8345 u64 root_objectid, u64 owner,
8346 u64 offset, u64 ram_bytes,
8347 struct btrfs_key *ins)
8349 struct btrfs_fs_info *fs_info = trans->fs_info;
8352 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8354 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8355 ins->offset, 0, root_objectid, owner,
8357 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8362 * this is used by the tree logging recovery code. It records that
8363 * an extent has been allocated and makes sure to clear the free
8364 * space cache bits as well
8366 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8367 struct btrfs_fs_info *fs_info,
8368 u64 root_objectid, u64 owner, u64 offset,
8369 struct btrfs_key *ins)
8372 struct btrfs_block_group_cache *block_group;
8373 struct btrfs_space_info *space_info;
8376 * Mixed block groups will exclude before processing the log so we only
8377 * need to do the exclude dance if this fs isn't mixed.
8379 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8380 ret = __exclude_logged_extent(fs_info, ins->objectid,
8386 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8390 space_info = block_group->space_info;
8391 spin_lock(&space_info->lock);
8392 spin_lock(&block_group->lock);
8393 space_info->bytes_reserved += ins->offset;
8394 block_group->reserved += ins->offset;
8395 spin_unlock(&block_group->lock);
8396 spin_unlock(&space_info->lock);
8398 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8399 0, owner, offset, ins, 1);
8400 btrfs_put_block_group(block_group);
8404 static struct extent_buffer *
8405 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8406 u64 bytenr, int level)
8408 struct btrfs_fs_info *fs_info = root->fs_info;
8409 struct extent_buffer *buf;
8411 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8416 * Extra safety check in case the extent tree is corrupted and extent
8417 * allocator chooses to use a tree block which is already used and
8420 if (buf->lock_owner == current->pid) {
8421 btrfs_err_rl(fs_info,
8422 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8423 buf->start, btrfs_header_owner(buf), current->pid);
8424 free_extent_buffer(buf);
8425 return ERR_PTR(-EUCLEAN);
8428 btrfs_set_header_generation(buf, trans->transid);
8429 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8430 btrfs_tree_lock(buf);
8431 clean_tree_block(fs_info, buf);
8432 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8434 btrfs_set_lock_blocking(buf);
8435 set_extent_buffer_uptodate(buf);
8437 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8438 buf->log_index = root->log_transid % 2;
8440 * we allow two log transactions at a time, use different
8441 * EXENT bit to differentiate dirty pages.
8443 if (buf->log_index == 0)
8444 set_extent_dirty(&root->dirty_log_pages, buf->start,
8445 buf->start + buf->len - 1, GFP_NOFS);
8447 set_extent_new(&root->dirty_log_pages, buf->start,
8448 buf->start + buf->len - 1);
8450 buf->log_index = -1;
8451 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8452 buf->start + buf->len - 1, GFP_NOFS);
8454 trans->dirty = true;
8455 /* this returns a buffer locked for blocking */
8459 static struct btrfs_block_rsv *
8460 use_block_rsv(struct btrfs_trans_handle *trans,
8461 struct btrfs_root *root, u32 blocksize)
8463 struct btrfs_fs_info *fs_info = root->fs_info;
8464 struct btrfs_block_rsv *block_rsv;
8465 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8467 bool global_updated = false;
8469 block_rsv = get_block_rsv(trans, root);
8471 if (unlikely(block_rsv->size == 0))
8474 ret = block_rsv_use_bytes(block_rsv, blocksize);
8478 if (block_rsv->failfast)
8479 return ERR_PTR(ret);
8481 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8482 global_updated = true;
8483 update_global_block_rsv(fs_info);
8487 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8488 static DEFINE_RATELIMIT_STATE(_rs,
8489 DEFAULT_RATELIMIT_INTERVAL * 10,
8490 /*DEFAULT_RATELIMIT_BURST*/ 1);
8491 if (__ratelimit(&_rs))
8493 "BTRFS: block rsv returned %d\n", ret);
8496 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8497 BTRFS_RESERVE_NO_FLUSH);
8501 * If we couldn't reserve metadata bytes try and use some from
8502 * the global reserve if its space type is the same as the global
8505 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8506 block_rsv->space_info == global_rsv->space_info) {
8507 ret = block_rsv_use_bytes(global_rsv, blocksize);
8511 return ERR_PTR(ret);
8514 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8515 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8517 block_rsv_add_bytes(block_rsv, blocksize, 0);
8518 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8522 * finds a free extent and does all the dirty work required for allocation
8523 * returns the tree buffer or an ERR_PTR on error.
8525 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8526 struct btrfs_root *root,
8527 u64 parent, u64 root_objectid,
8528 const struct btrfs_disk_key *key,
8529 int level, u64 hint,
8532 struct btrfs_fs_info *fs_info = root->fs_info;
8533 struct btrfs_key ins;
8534 struct btrfs_block_rsv *block_rsv;
8535 struct extent_buffer *buf;
8536 struct btrfs_delayed_extent_op *extent_op;
8539 u32 blocksize = fs_info->nodesize;
8540 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8542 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8543 if (btrfs_is_testing(fs_info)) {
8544 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8547 root->alloc_bytenr += blocksize;
8552 block_rsv = use_block_rsv(trans, root, blocksize);
8553 if (IS_ERR(block_rsv))
8554 return ERR_CAST(block_rsv);
8556 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8557 empty_size, hint, &ins, 0, 0);
8561 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8564 goto out_free_reserved;
8567 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8569 parent = ins.objectid;
8570 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8574 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8575 extent_op = btrfs_alloc_delayed_extent_op();
8581 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8583 memset(&extent_op->key, 0, sizeof(extent_op->key));
8584 extent_op->flags_to_set = flags;
8585 extent_op->update_key = skinny_metadata ? false : true;
8586 extent_op->update_flags = true;
8587 extent_op->is_data = false;
8588 extent_op->level = level;
8590 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8592 root_objectid, level,
8593 BTRFS_ADD_DELAYED_EXTENT,
8594 extent_op, NULL, NULL);
8596 goto out_free_delayed;
8601 btrfs_free_delayed_extent_op(extent_op);
8603 btrfs_tree_unlock(buf);
8604 free_extent_buffer(buf);
8606 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8608 unuse_block_rsv(fs_info, block_rsv, blocksize);
8609 return ERR_PTR(ret);
8612 struct walk_control {
8613 u64 refs[BTRFS_MAX_LEVEL];
8614 u64 flags[BTRFS_MAX_LEVEL];
8615 struct btrfs_key update_progress;
8626 #define DROP_REFERENCE 1
8627 #define UPDATE_BACKREF 2
8629 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8630 struct btrfs_root *root,
8631 struct walk_control *wc,
8632 struct btrfs_path *path)
8634 struct btrfs_fs_info *fs_info = root->fs_info;
8640 struct btrfs_key key;
8641 struct extent_buffer *eb;
8646 if (path->slots[wc->level] < wc->reada_slot) {
8647 wc->reada_count = wc->reada_count * 2 / 3;
8648 wc->reada_count = max(wc->reada_count, 2);
8650 wc->reada_count = wc->reada_count * 3 / 2;
8651 wc->reada_count = min_t(int, wc->reada_count,
8652 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8655 eb = path->nodes[wc->level];
8656 nritems = btrfs_header_nritems(eb);
8658 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8659 if (nread >= wc->reada_count)
8663 bytenr = btrfs_node_blockptr(eb, slot);
8664 generation = btrfs_node_ptr_generation(eb, slot);
8666 if (slot == path->slots[wc->level])
8669 if (wc->stage == UPDATE_BACKREF &&
8670 generation <= root->root_key.offset)
8673 /* We don't lock the tree block, it's OK to be racy here */
8674 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8675 wc->level - 1, 1, &refs,
8677 /* We don't care about errors in readahead. */
8682 if (wc->stage == DROP_REFERENCE) {
8686 if (wc->level == 1 &&
8687 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8689 if (!wc->update_ref ||
8690 generation <= root->root_key.offset)
8692 btrfs_node_key_to_cpu(eb, &key, slot);
8693 ret = btrfs_comp_cpu_keys(&key,
8694 &wc->update_progress);
8698 if (wc->level == 1 &&
8699 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8703 readahead_tree_block(fs_info, bytenr);
8706 wc->reada_slot = slot;
8710 * helper to process tree block while walking down the tree.
8712 * when wc->stage == UPDATE_BACKREF, this function updates
8713 * back refs for pointers in the block.
8715 * NOTE: return value 1 means we should stop walking down.
8717 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8718 struct btrfs_root *root,
8719 struct btrfs_path *path,
8720 struct walk_control *wc, int lookup_info)
8722 struct btrfs_fs_info *fs_info = root->fs_info;
8723 int level = wc->level;
8724 struct extent_buffer *eb = path->nodes[level];
8725 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8728 if (wc->stage == UPDATE_BACKREF &&
8729 btrfs_header_owner(eb) != root->root_key.objectid)
8733 * when reference count of tree block is 1, it won't increase
8734 * again. once full backref flag is set, we never clear it.
8737 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8738 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8739 BUG_ON(!path->locks[level]);
8740 ret = btrfs_lookup_extent_info(trans, fs_info,
8741 eb->start, level, 1,
8744 BUG_ON(ret == -ENOMEM);
8747 BUG_ON(wc->refs[level] == 0);
8750 if (wc->stage == DROP_REFERENCE) {
8751 if (wc->refs[level] > 1)
8754 if (path->locks[level] && !wc->keep_locks) {
8755 btrfs_tree_unlock_rw(eb, path->locks[level]);
8756 path->locks[level] = 0;
8761 /* wc->stage == UPDATE_BACKREF */
8762 if (!(wc->flags[level] & flag)) {
8763 BUG_ON(!path->locks[level]);
8764 ret = btrfs_inc_ref(trans, root, eb, 1);
8765 BUG_ON(ret); /* -ENOMEM */
8766 ret = btrfs_dec_ref(trans, root, eb, 0);
8767 BUG_ON(ret); /* -ENOMEM */
8768 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8770 btrfs_header_level(eb), 0);
8771 BUG_ON(ret); /* -ENOMEM */
8772 wc->flags[level] |= flag;
8776 * the block is shared by multiple trees, so it's not good to
8777 * keep the tree lock
8779 if (path->locks[level] && level > 0) {
8780 btrfs_tree_unlock_rw(eb, path->locks[level]);
8781 path->locks[level] = 0;
8787 * helper to process tree block pointer.
8789 * when wc->stage == DROP_REFERENCE, this function checks
8790 * reference count of the block pointed to. if the block
8791 * is shared and we need update back refs for the subtree
8792 * rooted at the block, this function changes wc->stage to
8793 * UPDATE_BACKREF. if the block is shared and there is no
8794 * need to update back, this function drops the reference
8797 * NOTE: return value 1 means we should stop walking down.
8799 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8800 struct btrfs_root *root,
8801 struct btrfs_path *path,
8802 struct walk_control *wc, int *lookup_info)
8804 struct btrfs_fs_info *fs_info = root->fs_info;
8809 struct btrfs_key key;
8810 struct extent_buffer *next;
8811 int level = wc->level;
8814 bool need_account = false;
8816 generation = btrfs_node_ptr_generation(path->nodes[level],
8817 path->slots[level]);
8819 * if the lower level block was created before the snapshot
8820 * was created, we know there is no need to update back refs
8823 if (wc->stage == UPDATE_BACKREF &&
8824 generation <= root->root_key.offset) {
8829 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8830 blocksize = fs_info->nodesize;
8832 next = find_extent_buffer(fs_info, bytenr);
8834 next = btrfs_find_create_tree_block(fs_info, bytenr);
8836 return PTR_ERR(next);
8838 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8842 btrfs_tree_lock(next);
8843 btrfs_set_lock_blocking(next);
8845 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8846 &wc->refs[level - 1],
8847 &wc->flags[level - 1]);
8851 if (unlikely(wc->refs[level - 1] == 0)) {
8852 btrfs_err(fs_info, "Missing references.");
8858 if (wc->stage == DROP_REFERENCE) {
8859 if (wc->refs[level - 1] > 1) {
8860 need_account = true;
8862 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8865 if (!wc->update_ref ||
8866 generation <= root->root_key.offset)
8869 btrfs_node_key_to_cpu(path->nodes[level], &key,
8870 path->slots[level]);
8871 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8875 wc->stage = UPDATE_BACKREF;
8876 wc->shared_level = level - 1;
8880 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8884 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8885 btrfs_tree_unlock(next);
8886 free_extent_buffer(next);
8892 if (reada && level == 1)
8893 reada_walk_down(trans, root, wc, path);
8894 next = read_tree_block(fs_info, bytenr, generation);
8896 return PTR_ERR(next);
8897 } else if (!extent_buffer_uptodate(next)) {
8898 free_extent_buffer(next);
8901 btrfs_tree_lock(next);
8902 btrfs_set_lock_blocking(next);
8906 ASSERT(level == btrfs_header_level(next));
8907 if (level != btrfs_header_level(next)) {
8908 btrfs_err(root->fs_info, "mismatched level");
8912 path->nodes[level] = next;
8913 path->slots[level] = 0;
8914 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8920 wc->refs[level - 1] = 0;
8921 wc->flags[level - 1] = 0;
8922 if (wc->stage == DROP_REFERENCE) {
8923 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8924 parent = path->nodes[level]->start;
8926 ASSERT(root->root_key.objectid ==
8927 btrfs_header_owner(path->nodes[level]));
8928 if (root->root_key.objectid !=
8929 btrfs_header_owner(path->nodes[level])) {
8930 btrfs_err(root->fs_info,
8931 "mismatched block owner");
8939 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8940 generation, level - 1);
8942 btrfs_err_rl(fs_info,
8943 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8947 ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize,
8948 parent, root->root_key.objectid,
8958 btrfs_tree_unlock(next);
8959 free_extent_buffer(next);
8965 * helper to process tree block while walking up the tree.
8967 * when wc->stage == DROP_REFERENCE, this function drops
8968 * reference count on the block.
8970 * when wc->stage == UPDATE_BACKREF, this function changes
8971 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8972 * to UPDATE_BACKREF previously while processing the block.
8974 * NOTE: return value 1 means we should stop walking up.
8976 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8977 struct btrfs_root *root,
8978 struct btrfs_path *path,
8979 struct walk_control *wc)
8981 struct btrfs_fs_info *fs_info = root->fs_info;
8983 int level = wc->level;
8984 struct extent_buffer *eb = path->nodes[level];
8987 if (wc->stage == UPDATE_BACKREF) {
8988 BUG_ON(wc->shared_level < level);
8989 if (level < wc->shared_level)
8992 ret = find_next_key(path, level + 1, &wc->update_progress);
8996 wc->stage = DROP_REFERENCE;
8997 wc->shared_level = -1;
8998 path->slots[level] = 0;
9001 * check reference count again if the block isn't locked.
9002 * we should start walking down the tree again if reference
9005 if (!path->locks[level]) {
9007 btrfs_tree_lock(eb);
9008 btrfs_set_lock_blocking(eb);
9009 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9011 ret = btrfs_lookup_extent_info(trans, fs_info,
9012 eb->start, level, 1,
9016 btrfs_tree_unlock_rw(eb, path->locks[level]);
9017 path->locks[level] = 0;
9020 BUG_ON(wc->refs[level] == 0);
9021 if (wc->refs[level] == 1) {
9022 btrfs_tree_unlock_rw(eb, path->locks[level]);
9023 path->locks[level] = 0;
9029 /* wc->stage == DROP_REFERENCE */
9030 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9032 if (wc->refs[level] == 1) {
9034 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9035 ret = btrfs_dec_ref(trans, root, eb, 1);
9037 ret = btrfs_dec_ref(trans, root, eb, 0);
9038 BUG_ON(ret); /* -ENOMEM */
9039 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
9041 btrfs_err_rl(fs_info,
9042 "error %d accounting leaf items. Quota is out of sync, rescan required.",
9046 /* make block locked assertion in clean_tree_block happy */
9047 if (!path->locks[level] &&
9048 btrfs_header_generation(eb) == trans->transid) {
9049 btrfs_tree_lock(eb);
9050 btrfs_set_lock_blocking(eb);
9051 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9053 clean_tree_block(fs_info, eb);
9056 if (eb == root->node) {
9057 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9059 else if (root->root_key.objectid != btrfs_header_owner(eb))
9060 goto owner_mismatch;
9062 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9063 parent = path->nodes[level + 1]->start;
9064 else if (root->root_key.objectid !=
9065 btrfs_header_owner(path->nodes[level + 1]))
9066 goto owner_mismatch;
9069 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9071 wc->refs[level] = 0;
9072 wc->flags[level] = 0;
9076 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
9077 btrfs_header_owner(eb), root->root_key.objectid);
9081 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9082 struct btrfs_root *root,
9083 struct btrfs_path *path,
9084 struct walk_control *wc)
9086 int level = wc->level;
9087 int lookup_info = 1;
9090 while (level >= 0) {
9091 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9098 if (path->slots[level] >=
9099 btrfs_header_nritems(path->nodes[level]))
9102 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9104 path->slots[level]++;
9113 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9114 struct btrfs_root *root,
9115 struct btrfs_path *path,
9116 struct walk_control *wc, int max_level)
9118 int level = wc->level;
9121 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9122 while (level < max_level && path->nodes[level]) {
9124 if (path->slots[level] + 1 <
9125 btrfs_header_nritems(path->nodes[level])) {
9126 path->slots[level]++;
9129 ret = walk_up_proc(trans, root, path, wc);
9135 if (path->locks[level]) {
9136 btrfs_tree_unlock_rw(path->nodes[level],
9137 path->locks[level]);
9138 path->locks[level] = 0;
9140 free_extent_buffer(path->nodes[level]);
9141 path->nodes[level] = NULL;
9149 * drop a subvolume tree.
9151 * this function traverses the tree freeing any blocks that only
9152 * referenced by the tree.
9154 * when a shared tree block is found. this function decreases its
9155 * reference count by one. if update_ref is true, this function
9156 * also make sure backrefs for the shared block and all lower level
9157 * blocks are properly updated.
9159 * If called with for_reloc == 0, may exit early with -EAGAIN
9161 int btrfs_drop_snapshot(struct btrfs_root *root,
9162 struct btrfs_block_rsv *block_rsv, int update_ref,
9165 struct btrfs_fs_info *fs_info = root->fs_info;
9166 struct btrfs_path *path;
9167 struct btrfs_trans_handle *trans;
9168 struct btrfs_root *tree_root = fs_info->tree_root;
9169 struct btrfs_root_item *root_item = &root->root_item;
9170 struct walk_control *wc;
9171 struct btrfs_key key;
9175 bool root_dropped = false;
9177 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9179 path = btrfs_alloc_path();
9185 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9187 btrfs_free_path(path);
9192 trans = btrfs_start_transaction(tree_root, 0);
9193 if (IS_ERR(trans)) {
9194 err = PTR_ERR(trans);
9199 trans->block_rsv = block_rsv;
9201 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9202 level = btrfs_header_level(root->node);
9203 path->nodes[level] = btrfs_lock_root_node(root);
9204 btrfs_set_lock_blocking(path->nodes[level]);
9205 path->slots[level] = 0;
9206 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9207 memset(&wc->update_progress, 0,
9208 sizeof(wc->update_progress));
9210 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9211 memcpy(&wc->update_progress, &key,
9212 sizeof(wc->update_progress));
9214 level = root_item->drop_level;
9216 path->lowest_level = level;
9217 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9218 path->lowest_level = 0;
9226 * unlock our path, this is safe because only this
9227 * function is allowed to delete this snapshot
9229 btrfs_unlock_up_safe(path, 0);
9231 level = btrfs_header_level(root->node);
9233 btrfs_tree_lock(path->nodes[level]);
9234 btrfs_set_lock_blocking(path->nodes[level]);
9235 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9237 ret = btrfs_lookup_extent_info(trans, fs_info,
9238 path->nodes[level]->start,
9239 level, 1, &wc->refs[level],
9245 BUG_ON(wc->refs[level] == 0);
9247 if (level == root_item->drop_level)
9250 btrfs_tree_unlock(path->nodes[level]);
9251 path->locks[level] = 0;
9252 WARN_ON(wc->refs[level] != 1);
9258 wc->shared_level = -1;
9259 wc->stage = DROP_REFERENCE;
9260 wc->update_ref = update_ref;
9262 wc->for_reloc = for_reloc;
9263 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9267 ret = walk_down_tree(trans, root, path, wc);
9273 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9280 BUG_ON(wc->stage != DROP_REFERENCE);
9284 if (wc->stage == DROP_REFERENCE) {
9286 btrfs_node_key(path->nodes[level],
9287 &root_item->drop_progress,
9288 path->slots[level]);
9289 root_item->drop_level = level;
9292 BUG_ON(wc->level == 0);
9293 if (btrfs_should_end_transaction(trans) ||
9294 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9295 ret = btrfs_update_root(trans, tree_root,
9299 btrfs_abort_transaction(trans, ret);
9304 btrfs_end_transaction_throttle(trans);
9305 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9306 btrfs_debug(fs_info,
9307 "drop snapshot early exit");
9312 trans = btrfs_start_transaction(tree_root, 0);
9313 if (IS_ERR(trans)) {
9314 err = PTR_ERR(trans);
9318 trans->block_rsv = block_rsv;
9321 btrfs_release_path(path);
9325 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9327 btrfs_abort_transaction(trans, ret);
9332 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9333 ret = btrfs_find_root(tree_root, &root->root_key, path,
9336 btrfs_abort_transaction(trans, ret);
9339 } else if (ret > 0) {
9340 /* if we fail to delete the orphan item this time
9341 * around, it'll get picked up the next time.
9343 * The most common failure here is just -ENOENT.
9345 btrfs_del_orphan_item(trans, tree_root,
9346 root->root_key.objectid);
9350 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9351 btrfs_add_dropped_root(trans, root);
9353 free_extent_buffer(root->node);
9354 free_extent_buffer(root->commit_root);
9355 btrfs_put_fs_root(root);
9357 root_dropped = true;
9359 btrfs_end_transaction_throttle(trans);
9362 btrfs_free_path(path);
9365 * So if we need to stop dropping the snapshot for whatever reason we
9366 * need to make sure to add it back to the dead root list so that we
9367 * keep trying to do the work later. This also cleans up roots if we
9368 * don't have it in the radix (like when we recover after a power fail
9369 * or unmount) so we don't leak memory.
9371 if (!for_reloc && root_dropped == false)
9372 btrfs_add_dead_root(root);
9377 * drop subtree rooted at tree block 'node'.
9379 * NOTE: this function will unlock and release tree block 'node'
9380 * only used by relocation code
9382 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9383 struct btrfs_root *root,
9384 struct extent_buffer *node,
9385 struct extent_buffer *parent)
9387 struct btrfs_fs_info *fs_info = root->fs_info;
9388 struct btrfs_path *path;
9389 struct walk_control *wc;
9395 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9397 path = btrfs_alloc_path();
9401 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9403 btrfs_free_path(path);
9407 btrfs_assert_tree_locked(parent);
9408 parent_level = btrfs_header_level(parent);
9409 extent_buffer_get(parent);
9410 path->nodes[parent_level] = parent;
9411 path->slots[parent_level] = btrfs_header_nritems(parent);
9413 btrfs_assert_tree_locked(node);
9414 level = btrfs_header_level(node);
9415 path->nodes[level] = node;
9416 path->slots[level] = 0;
9417 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9419 wc->refs[parent_level] = 1;
9420 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9422 wc->shared_level = -1;
9423 wc->stage = DROP_REFERENCE;
9427 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9430 wret = walk_down_tree(trans, root, path, wc);
9436 wret = walk_up_tree(trans, root, path, wc, parent_level);
9444 btrfs_free_path(path);
9448 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9454 * if restripe for this chunk_type is on pick target profile and
9455 * return, otherwise do the usual balance
9457 stripped = get_restripe_target(fs_info, flags);
9459 return extended_to_chunk(stripped);
9461 num_devices = fs_info->fs_devices->rw_devices;
9463 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9464 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9465 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9467 if (num_devices == 1) {
9468 stripped |= BTRFS_BLOCK_GROUP_DUP;
9469 stripped = flags & ~stripped;
9471 /* turn raid0 into single device chunks */
9472 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9475 /* turn mirroring into duplication */
9476 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9477 BTRFS_BLOCK_GROUP_RAID10))
9478 return stripped | BTRFS_BLOCK_GROUP_DUP;
9480 /* they already had raid on here, just return */
9481 if (flags & stripped)
9484 stripped |= BTRFS_BLOCK_GROUP_DUP;
9485 stripped = flags & ~stripped;
9487 /* switch duplicated blocks with raid1 */
9488 if (flags & BTRFS_BLOCK_GROUP_DUP)
9489 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9491 /* this is drive concat, leave it alone */
9497 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9499 struct btrfs_space_info *sinfo = cache->space_info;
9501 u64 min_allocable_bytes;
9505 * We need some metadata space and system metadata space for
9506 * allocating chunks in some corner cases until we force to set
9507 * it to be readonly.
9510 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9512 min_allocable_bytes = SZ_1M;
9514 min_allocable_bytes = 0;
9516 spin_lock(&sinfo->lock);
9517 spin_lock(&cache->lock);
9525 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9526 cache->bytes_super - btrfs_block_group_used(&cache->item);
9528 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9529 min_allocable_bytes <= sinfo->total_bytes) {
9530 sinfo->bytes_readonly += num_bytes;
9532 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9536 spin_unlock(&cache->lock);
9537 spin_unlock(&sinfo->lock);
9541 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9542 struct btrfs_block_group_cache *cache)
9545 struct btrfs_trans_handle *trans;
9550 trans = btrfs_join_transaction(fs_info->extent_root);
9552 return PTR_ERR(trans);
9555 * we're not allowed to set block groups readonly after the dirty
9556 * block groups cache has started writing. If it already started,
9557 * back off and let this transaction commit
9559 mutex_lock(&fs_info->ro_block_group_mutex);
9560 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9561 u64 transid = trans->transid;
9563 mutex_unlock(&fs_info->ro_block_group_mutex);
9564 btrfs_end_transaction(trans);
9566 ret = btrfs_wait_for_commit(fs_info, transid);
9573 * if we are changing raid levels, try to allocate a corresponding
9574 * block group with the new raid level.
9576 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9577 if (alloc_flags != cache->flags) {
9578 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9581 * ENOSPC is allowed here, we may have enough space
9582 * already allocated at the new raid level to
9591 ret = inc_block_group_ro(cache, 0);
9594 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9595 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9599 ret = inc_block_group_ro(cache, 0);
9601 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9602 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9603 mutex_lock(&fs_info->chunk_mutex);
9604 check_system_chunk(trans, fs_info, alloc_flags);
9605 mutex_unlock(&fs_info->chunk_mutex);
9607 mutex_unlock(&fs_info->ro_block_group_mutex);
9609 btrfs_end_transaction(trans);
9613 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9614 struct btrfs_fs_info *fs_info, u64 type)
9616 u64 alloc_flags = get_alloc_profile(fs_info, type);
9618 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9622 * helper to account the unused space of all the readonly block group in the
9623 * space_info. takes mirrors into account.
9625 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9627 struct btrfs_block_group_cache *block_group;
9631 /* It's df, we don't care if it's racy */
9632 if (list_empty(&sinfo->ro_bgs))
9635 spin_lock(&sinfo->lock);
9636 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9637 spin_lock(&block_group->lock);
9639 if (!block_group->ro) {
9640 spin_unlock(&block_group->lock);
9644 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9645 BTRFS_BLOCK_GROUP_RAID10 |
9646 BTRFS_BLOCK_GROUP_DUP))
9651 free_bytes += (block_group->key.offset -
9652 btrfs_block_group_used(&block_group->item)) *
9655 spin_unlock(&block_group->lock);
9657 spin_unlock(&sinfo->lock);
9662 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9664 struct btrfs_space_info *sinfo = cache->space_info;
9669 spin_lock(&sinfo->lock);
9670 spin_lock(&cache->lock);
9672 num_bytes = cache->key.offset - cache->reserved -
9673 cache->pinned - cache->bytes_super -
9674 btrfs_block_group_used(&cache->item);
9675 sinfo->bytes_readonly -= num_bytes;
9676 list_del_init(&cache->ro_list);
9678 spin_unlock(&cache->lock);
9679 spin_unlock(&sinfo->lock);
9683 * checks to see if its even possible to relocate this block group.
9685 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9686 * ok to go ahead and try.
9688 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9690 struct btrfs_root *root = fs_info->extent_root;
9691 struct btrfs_block_group_cache *block_group;
9692 struct btrfs_space_info *space_info;
9693 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9694 struct btrfs_device *device;
9695 struct btrfs_trans_handle *trans;
9705 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9707 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9709 /* odd, couldn't find the block group, leave it alone */
9713 "can't find block group for bytenr %llu",
9718 min_free = btrfs_block_group_used(&block_group->item);
9720 /* no bytes used, we're good */
9724 space_info = block_group->space_info;
9725 spin_lock(&space_info->lock);
9727 full = space_info->full;
9730 * if this is the last block group we have in this space, we can't
9731 * relocate it unless we're able to allocate a new chunk below.
9733 * Otherwise, we need to make sure we have room in the space to handle
9734 * all of the extents from this block group. If we can, we're good
9736 if ((space_info->total_bytes != block_group->key.offset) &&
9737 (btrfs_space_info_used(space_info, false) + min_free <
9738 space_info->total_bytes)) {
9739 spin_unlock(&space_info->lock);
9742 spin_unlock(&space_info->lock);
9745 * ok we don't have enough space, but maybe we have free space on our
9746 * devices to allocate new chunks for relocation, so loop through our
9747 * alloc devices and guess if we have enough space. if this block
9748 * group is going to be restriped, run checks against the target
9749 * profile instead of the current one.
9761 target = get_restripe_target(fs_info, block_group->flags);
9763 index = __get_raid_index(extended_to_chunk(target));
9766 * this is just a balance, so if we were marked as full
9767 * we know there is no space for a new chunk
9772 "no space to alloc new chunk for block group %llu",
9773 block_group->key.objectid);
9777 index = get_block_group_index(block_group);
9780 if (index == BTRFS_RAID_RAID10) {
9784 } else if (index == BTRFS_RAID_RAID1) {
9786 } else if (index == BTRFS_RAID_DUP) {
9789 } else if (index == BTRFS_RAID_RAID0) {
9790 dev_min = fs_devices->rw_devices;
9791 min_free = div64_u64(min_free, dev_min);
9794 /* We need to do this so that we can look at pending chunks */
9795 trans = btrfs_join_transaction(root);
9796 if (IS_ERR(trans)) {
9797 ret = PTR_ERR(trans);
9801 mutex_lock(&fs_info->chunk_mutex);
9802 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9806 * check to make sure we can actually find a chunk with enough
9807 * space to fit our block group in.
9809 if (device->total_bytes > device->bytes_used + min_free &&
9810 !device->is_tgtdev_for_dev_replace) {
9811 ret = find_free_dev_extent(trans, device, min_free,
9816 if (dev_nr >= dev_min)
9822 if (debug && ret == -1)
9824 "no space to allocate a new chunk for block group %llu",
9825 block_group->key.objectid);
9826 mutex_unlock(&fs_info->chunk_mutex);
9827 btrfs_end_transaction(trans);
9829 btrfs_put_block_group(block_group);
9833 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9834 struct btrfs_path *path,
9835 struct btrfs_key *key)
9837 struct btrfs_root *root = fs_info->extent_root;
9839 struct btrfs_key found_key;
9840 struct extent_buffer *leaf;
9841 struct btrfs_block_group_item bg;
9845 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9850 slot = path->slots[0];
9851 leaf = path->nodes[0];
9852 if (slot >= btrfs_header_nritems(leaf)) {
9853 ret = btrfs_next_leaf(root, path);
9860 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9862 if (found_key.objectid >= key->objectid &&
9863 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9864 struct extent_map_tree *em_tree;
9865 struct extent_map *em;
9867 em_tree = &root->fs_info->mapping_tree.map_tree;
9868 read_lock(&em_tree->lock);
9869 em = lookup_extent_mapping(em_tree, found_key.objectid,
9871 read_unlock(&em_tree->lock);
9874 "logical %llu len %llu found bg but no related chunk",
9875 found_key.objectid, found_key.offset);
9877 } else if (em->start != found_key.objectid ||
9878 em->len != found_key.offset) {
9880 "block group %llu len %llu mismatch with chunk %llu len %llu",
9881 found_key.objectid, found_key.offset,
9882 em->start, em->len);
9885 read_extent_buffer(leaf, &bg,
9886 btrfs_item_ptr_offset(leaf, slot),
9888 flags = btrfs_block_group_flags(&bg) &
9889 BTRFS_BLOCK_GROUP_TYPE_MASK;
9891 if (flags != (em->map_lookup->type &
9892 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9894 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9896 found_key.offset, flags,
9897 (BTRFS_BLOCK_GROUP_TYPE_MASK &
9898 em->map_lookup->type));
9904 free_extent_map(em);
9913 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9915 struct btrfs_block_group_cache *block_group;
9919 struct inode *inode;
9921 block_group = btrfs_lookup_first_block_group(info, last);
9922 while (block_group) {
9923 wait_block_group_cache_done(block_group);
9924 spin_lock(&block_group->lock);
9925 if (block_group->iref)
9927 spin_unlock(&block_group->lock);
9928 block_group = next_block_group(info, block_group);
9937 inode = block_group->inode;
9938 block_group->iref = 0;
9939 block_group->inode = NULL;
9940 spin_unlock(&block_group->lock);
9941 ASSERT(block_group->io_ctl.inode == NULL);
9943 last = block_group->key.objectid + block_group->key.offset;
9944 btrfs_put_block_group(block_group);
9949 * Must be called only after stopping all workers, since we could have block
9950 * group caching kthreads running, and therefore they could race with us if we
9951 * freed the block groups before stopping them.
9953 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9955 struct btrfs_block_group_cache *block_group;
9956 struct btrfs_space_info *space_info;
9957 struct btrfs_caching_control *caching_ctl;
9960 down_write(&info->commit_root_sem);
9961 while (!list_empty(&info->caching_block_groups)) {
9962 caching_ctl = list_entry(info->caching_block_groups.next,
9963 struct btrfs_caching_control, list);
9964 list_del(&caching_ctl->list);
9965 put_caching_control(caching_ctl);
9967 up_write(&info->commit_root_sem);
9969 spin_lock(&info->unused_bgs_lock);
9970 while (!list_empty(&info->unused_bgs)) {
9971 block_group = list_first_entry(&info->unused_bgs,
9972 struct btrfs_block_group_cache,
9974 list_del_init(&block_group->bg_list);
9975 btrfs_put_block_group(block_group);
9977 spin_unlock(&info->unused_bgs_lock);
9979 spin_lock(&info->block_group_cache_lock);
9980 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9981 block_group = rb_entry(n, struct btrfs_block_group_cache,
9983 rb_erase(&block_group->cache_node,
9984 &info->block_group_cache_tree);
9985 RB_CLEAR_NODE(&block_group->cache_node);
9986 spin_unlock(&info->block_group_cache_lock);
9988 down_write(&block_group->space_info->groups_sem);
9989 list_del(&block_group->list);
9990 up_write(&block_group->space_info->groups_sem);
9993 * We haven't cached this block group, which means we could
9994 * possibly have excluded extents on this block group.
9996 if (block_group->cached == BTRFS_CACHE_NO ||
9997 block_group->cached == BTRFS_CACHE_ERROR)
9998 free_excluded_extents(info, block_group);
10000 btrfs_remove_free_space_cache(block_group);
10001 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
10002 ASSERT(list_empty(&block_group->dirty_list));
10003 ASSERT(list_empty(&block_group->io_list));
10004 ASSERT(list_empty(&block_group->bg_list));
10005 ASSERT(atomic_read(&block_group->count) == 1);
10006 btrfs_put_block_group(block_group);
10008 spin_lock(&info->block_group_cache_lock);
10010 spin_unlock(&info->block_group_cache_lock);
10012 /* now that all the block groups are freed, go through and
10013 * free all the space_info structs. This is only called during
10014 * the final stages of unmount, and so we know nobody is
10015 * using them. We call synchronize_rcu() once before we start,
10016 * just to be on the safe side.
10020 release_global_block_rsv(info);
10022 while (!list_empty(&info->space_info)) {
10025 space_info = list_entry(info->space_info.next,
10026 struct btrfs_space_info,
10030 * Do not hide this behind enospc_debug, this is actually
10031 * important and indicates a real bug if this happens.
10033 if (WARN_ON(space_info->bytes_pinned > 0 ||
10034 space_info->bytes_reserved > 0 ||
10035 space_info->bytes_may_use > 0))
10036 dump_space_info(info, space_info, 0, 0);
10037 list_del(&space_info->list);
10038 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10039 struct kobject *kobj;
10040 kobj = space_info->block_group_kobjs[i];
10041 space_info->block_group_kobjs[i] = NULL;
10047 kobject_del(&space_info->kobj);
10048 kobject_put(&space_info->kobj);
10053 static void __link_block_group(struct btrfs_space_info *space_info,
10054 struct btrfs_block_group_cache *cache)
10056 int index = get_block_group_index(cache);
10057 bool first = false;
10059 down_write(&space_info->groups_sem);
10060 if (list_empty(&space_info->block_groups[index]))
10062 list_add_tail(&cache->list, &space_info->block_groups[index]);
10063 up_write(&space_info->groups_sem);
10066 struct raid_kobject *rkobj;
10069 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10072 rkobj->raid_type = index;
10073 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10074 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10075 "%s", get_raid_name(index));
10077 kobject_put(&rkobj->kobj);
10080 space_info->block_group_kobjs[index] = &rkobj->kobj;
10085 btrfs_warn(cache->fs_info,
10086 "failed to add kobject for block cache, ignoring");
10089 static struct btrfs_block_group_cache *
10090 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10091 u64 start, u64 size)
10093 struct btrfs_block_group_cache *cache;
10095 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10099 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10101 if (!cache->free_space_ctl) {
10106 cache->key.objectid = start;
10107 cache->key.offset = size;
10108 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10110 cache->fs_info = fs_info;
10111 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10112 set_free_space_tree_thresholds(cache);
10114 atomic_set(&cache->count, 1);
10115 spin_lock_init(&cache->lock);
10116 init_rwsem(&cache->data_rwsem);
10117 INIT_LIST_HEAD(&cache->list);
10118 INIT_LIST_HEAD(&cache->cluster_list);
10119 INIT_LIST_HEAD(&cache->bg_list);
10120 INIT_LIST_HEAD(&cache->ro_list);
10121 INIT_LIST_HEAD(&cache->dirty_list);
10122 INIT_LIST_HEAD(&cache->io_list);
10123 btrfs_init_free_space_ctl(cache);
10124 atomic_set(&cache->trimming, 0);
10125 mutex_init(&cache->free_space_lock);
10126 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10133 * Iterate all chunks and verify that each of them has the corresponding block
10136 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10138 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
10139 struct extent_map *em;
10140 struct btrfs_block_group_cache *bg;
10145 read_lock(&map_tree->map_tree.lock);
10147 * lookup_extent_mapping will return the first extent map
10148 * intersecting the range, so setting @len to 1 is enough to
10149 * get the first chunk.
10151 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
10152 read_unlock(&map_tree->map_tree.lock);
10156 bg = btrfs_lookup_block_group(fs_info, em->start);
10159 "chunk start=%llu len=%llu doesn't have corresponding block group",
10160 em->start, em->len);
10162 free_extent_map(em);
10165 if (bg->key.objectid != em->start ||
10166 bg->key.offset != em->len ||
10167 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10168 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10170 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10171 em->start, em->len,
10172 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10173 bg->key.objectid, bg->key.offset,
10174 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10176 free_extent_map(em);
10177 btrfs_put_block_group(bg);
10180 start = em->start + em->len;
10181 free_extent_map(em);
10182 btrfs_put_block_group(bg);
10187 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10189 struct btrfs_path *path;
10191 struct btrfs_block_group_cache *cache;
10192 struct btrfs_space_info *space_info;
10193 struct btrfs_key key;
10194 struct btrfs_key found_key;
10195 struct extent_buffer *leaf;
10196 int need_clear = 0;
10201 feature = btrfs_super_incompat_flags(info->super_copy);
10202 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10206 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10207 path = btrfs_alloc_path();
10210 path->reada = READA_FORWARD;
10212 cache_gen = btrfs_super_cache_generation(info->super_copy);
10213 if (btrfs_test_opt(info, SPACE_CACHE) &&
10214 btrfs_super_generation(info->super_copy) != cache_gen)
10216 if (btrfs_test_opt(info, CLEAR_CACHE))
10220 ret = find_first_block_group(info, path, &key);
10226 leaf = path->nodes[0];
10227 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10229 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10238 * When we mount with old space cache, we need to
10239 * set BTRFS_DC_CLEAR and set dirty flag.
10241 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10242 * truncate the old free space cache inode and
10244 * b) Setting 'dirty flag' makes sure that we flush
10245 * the new space cache info onto disk.
10247 if (btrfs_test_opt(info, SPACE_CACHE))
10248 cache->disk_cache_state = BTRFS_DC_CLEAR;
10251 read_extent_buffer(leaf, &cache->item,
10252 btrfs_item_ptr_offset(leaf, path->slots[0]),
10253 sizeof(cache->item));
10254 cache->flags = btrfs_block_group_flags(&cache->item);
10256 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10257 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10259 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10260 cache->key.objectid);
10261 btrfs_put_block_group(cache);
10266 key.objectid = found_key.objectid + found_key.offset;
10267 btrfs_release_path(path);
10270 * We need to exclude the super stripes now so that the space
10271 * info has super bytes accounted for, otherwise we'll think
10272 * we have more space than we actually do.
10274 ret = exclude_super_stripes(info, cache);
10277 * We may have excluded something, so call this just in
10280 free_excluded_extents(info, cache);
10281 btrfs_put_block_group(cache);
10286 * check for two cases, either we are full, and therefore
10287 * don't need to bother with the caching work since we won't
10288 * find any space, or we are empty, and we can just add all
10289 * the space in and be done with it. This saves us _alot_ of
10290 * time, particularly in the full case.
10292 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10293 cache->last_byte_to_unpin = (u64)-1;
10294 cache->cached = BTRFS_CACHE_FINISHED;
10295 free_excluded_extents(info, cache);
10296 } else if (btrfs_block_group_used(&cache->item) == 0) {
10297 cache->last_byte_to_unpin = (u64)-1;
10298 cache->cached = BTRFS_CACHE_FINISHED;
10299 add_new_free_space(cache, info,
10300 found_key.objectid,
10301 found_key.objectid +
10303 free_excluded_extents(info, cache);
10306 ret = btrfs_add_block_group_cache(info, cache);
10308 btrfs_remove_free_space_cache(cache);
10309 btrfs_put_block_group(cache);
10313 trace_btrfs_add_block_group(info, cache, 0);
10314 update_space_info(info, cache->flags, found_key.offset,
10315 btrfs_block_group_used(&cache->item),
10316 cache->bytes_super, &space_info);
10318 cache->space_info = space_info;
10320 __link_block_group(space_info, cache);
10322 set_avail_alloc_bits(info, cache->flags);
10323 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10324 inc_block_group_ro(cache, 1);
10325 } else if (btrfs_block_group_used(&cache->item) == 0) {
10326 spin_lock(&info->unused_bgs_lock);
10327 /* Should always be true but just in case. */
10328 if (list_empty(&cache->bg_list)) {
10329 btrfs_get_block_group(cache);
10330 list_add_tail(&cache->bg_list,
10331 &info->unused_bgs);
10333 spin_unlock(&info->unused_bgs_lock);
10337 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10338 if (!(get_alloc_profile(info, space_info->flags) &
10339 (BTRFS_BLOCK_GROUP_RAID10 |
10340 BTRFS_BLOCK_GROUP_RAID1 |
10341 BTRFS_BLOCK_GROUP_RAID5 |
10342 BTRFS_BLOCK_GROUP_RAID6 |
10343 BTRFS_BLOCK_GROUP_DUP)))
10346 * avoid allocating from un-mirrored block group if there are
10347 * mirrored block groups.
10349 list_for_each_entry(cache,
10350 &space_info->block_groups[BTRFS_RAID_RAID0],
10352 inc_block_group_ro(cache, 1);
10353 list_for_each_entry(cache,
10354 &space_info->block_groups[BTRFS_RAID_SINGLE],
10356 inc_block_group_ro(cache, 1);
10359 init_global_block_rsv(info);
10360 ret = check_chunk_block_group_mappings(info);
10362 btrfs_free_path(path);
10366 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10367 struct btrfs_fs_info *fs_info)
10369 struct btrfs_block_group_cache *block_group;
10370 struct btrfs_root *extent_root = fs_info->extent_root;
10371 struct btrfs_block_group_item item;
10372 struct btrfs_key key;
10374 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10376 trans->can_flush_pending_bgs = false;
10377 while (!list_empty(&trans->new_bgs)) {
10378 block_group = list_first_entry(&trans->new_bgs,
10379 struct btrfs_block_group_cache,
10384 spin_lock(&block_group->lock);
10385 memcpy(&item, &block_group->item, sizeof(item));
10386 memcpy(&key, &block_group->key, sizeof(key));
10387 spin_unlock(&block_group->lock);
10389 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10392 btrfs_abort_transaction(trans, ret);
10393 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10396 btrfs_abort_transaction(trans, ret);
10397 add_block_group_free_space(trans, fs_info, block_group);
10398 /* already aborted the transaction if it failed. */
10400 list_del_init(&block_group->bg_list);
10402 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10405 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10406 struct btrfs_fs_info *fs_info, u64 bytes_used,
10407 u64 type, u64 chunk_offset, u64 size)
10409 struct btrfs_block_group_cache *cache;
10412 btrfs_set_log_full_commit(fs_info, trans);
10414 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10418 btrfs_set_block_group_used(&cache->item, bytes_used);
10419 btrfs_set_block_group_chunk_objectid(&cache->item,
10420 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10421 btrfs_set_block_group_flags(&cache->item, type);
10423 cache->flags = type;
10424 cache->last_byte_to_unpin = (u64)-1;
10425 cache->cached = BTRFS_CACHE_FINISHED;
10426 cache->needs_free_space = 1;
10427 ret = exclude_super_stripes(fs_info, cache);
10430 * We may have excluded something, so call this just in
10433 free_excluded_extents(fs_info, cache);
10434 btrfs_put_block_group(cache);
10438 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10440 free_excluded_extents(fs_info, cache);
10442 #ifdef CONFIG_BTRFS_DEBUG
10443 if (btrfs_should_fragment_free_space(cache)) {
10444 u64 new_bytes_used = size - bytes_used;
10446 bytes_used += new_bytes_used >> 1;
10447 fragment_free_space(cache);
10451 * Ensure the corresponding space_info object is created and
10452 * assigned to our block group. We want our bg to be added to the rbtree
10453 * with its ->space_info set.
10455 cache->space_info = __find_space_info(fs_info, cache->flags);
10456 if (!cache->space_info) {
10457 ret = create_space_info(fs_info, cache->flags,
10458 &cache->space_info);
10460 btrfs_remove_free_space_cache(cache);
10461 btrfs_put_block_group(cache);
10466 ret = btrfs_add_block_group_cache(fs_info, cache);
10468 btrfs_remove_free_space_cache(cache);
10469 btrfs_put_block_group(cache);
10474 * Now that our block group has its ->space_info set and is inserted in
10475 * the rbtree, update the space info's counters.
10477 trace_btrfs_add_block_group(fs_info, cache, 1);
10478 update_space_info(fs_info, cache->flags, size, bytes_used,
10479 cache->bytes_super, &cache->space_info);
10480 update_global_block_rsv(fs_info);
10482 __link_block_group(cache->space_info, cache);
10484 list_add_tail(&cache->bg_list, &trans->new_bgs);
10486 set_avail_alloc_bits(fs_info, type);
10490 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10492 u64 extra_flags = chunk_to_extended(flags) &
10493 BTRFS_EXTENDED_PROFILE_MASK;
10495 write_seqlock(&fs_info->profiles_lock);
10496 if (flags & BTRFS_BLOCK_GROUP_DATA)
10497 fs_info->avail_data_alloc_bits &= ~extra_flags;
10498 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10499 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10500 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10501 fs_info->avail_system_alloc_bits &= ~extra_flags;
10502 write_sequnlock(&fs_info->profiles_lock);
10505 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10506 struct btrfs_fs_info *fs_info, u64 group_start,
10507 struct extent_map *em)
10509 struct btrfs_root *root = fs_info->extent_root;
10510 struct btrfs_path *path;
10511 struct btrfs_block_group_cache *block_group;
10512 struct btrfs_free_cluster *cluster;
10513 struct btrfs_root *tree_root = fs_info->tree_root;
10514 struct btrfs_key key;
10515 struct inode *inode;
10516 struct kobject *kobj = NULL;
10520 struct btrfs_caching_control *caching_ctl = NULL;
10523 block_group = btrfs_lookup_block_group(fs_info, group_start);
10524 BUG_ON(!block_group);
10525 BUG_ON(!block_group->ro);
10528 * Free the reserved super bytes from this block group before
10531 free_excluded_extents(fs_info, block_group);
10533 memcpy(&key, &block_group->key, sizeof(key));
10534 index = get_block_group_index(block_group);
10535 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10536 BTRFS_BLOCK_GROUP_RAID1 |
10537 BTRFS_BLOCK_GROUP_RAID10))
10542 /* make sure this block group isn't part of an allocation cluster */
10543 cluster = &fs_info->data_alloc_cluster;
10544 spin_lock(&cluster->refill_lock);
10545 btrfs_return_cluster_to_free_space(block_group, cluster);
10546 spin_unlock(&cluster->refill_lock);
10549 * make sure this block group isn't part of a metadata
10550 * allocation cluster
10552 cluster = &fs_info->meta_alloc_cluster;
10553 spin_lock(&cluster->refill_lock);
10554 btrfs_return_cluster_to_free_space(block_group, cluster);
10555 spin_unlock(&cluster->refill_lock);
10557 path = btrfs_alloc_path();
10564 * get the inode first so any iput calls done for the io_list
10565 * aren't the final iput (no unlinks allowed now)
10567 inode = lookup_free_space_inode(fs_info, block_group, path);
10569 mutex_lock(&trans->transaction->cache_write_mutex);
10571 * make sure our free spache cache IO is done before remove the
10574 spin_lock(&trans->transaction->dirty_bgs_lock);
10575 if (!list_empty(&block_group->io_list)) {
10576 list_del_init(&block_group->io_list);
10578 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10580 spin_unlock(&trans->transaction->dirty_bgs_lock);
10581 btrfs_wait_cache_io(trans, block_group, path);
10582 btrfs_put_block_group(block_group);
10583 spin_lock(&trans->transaction->dirty_bgs_lock);
10586 if (!list_empty(&block_group->dirty_list)) {
10587 list_del_init(&block_group->dirty_list);
10588 btrfs_put_block_group(block_group);
10590 spin_unlock(&trans->transaction->dirty_bgs_lock);
10591 mutex_unlock(&trans->transaction->cache_write_mutex);
10593 if (!IS_ERR(inode)) {
10594 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10596 btrfs_add_delayed_iput(inode);
10599 clear_nlink(inode);
10600 /* One for the block groups ref */
10601 spin_lock(&block_group->lock);
10602 if (block_group->iref) {
10603 block_group->iref = 0;
10604 block_group->inode = NULL;
10605 spin_unlock(&block_group->lock);
10608 spin_unlock(&block_group->lock);
10610 /* One for our lookup ref */
10611 btrfs_add_delayed_iput(inode);
10614 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10615 key.offset = block_group->key.objectid;
10618 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10622 btrfs_release_path(path);
10624 ret = btrfs_del_item(trans, tree_root, path);
10627 btrfs_release_path(path);
10630 spin_lock(&fs_info->block_group_cache_lock);
10631 rb_erase(&block_group->cache_node,
10632 &fs_info->block_group_cache_tree);
10633 RB_CLEAR_NODE(&block_group->cache_node);
10635 /* Once for the block groups rbtree */
10636 btrfs_put_block_group(block_group);
10638 if (fs_info->first_logical_byte == block_group->key.objectid)
10639 fs_info->first_logical_byte = (u64)-1;
10640 spin_unlock(&fs_info->block_group_cache_lock);
10642 down_write(&block_group->space_info->groups_sem);
10644 * we must use list_del_init so people can check to see if they
10645 * are still on the list after taking the semaphore
10647 list_del_init(&block_group->list);
10648 if (list_empty(&block_group->space_info->block_groups[index])) {
10649 kobj = block_group->space_info->block_group_kobjs[index];
10650 block_group->space_info->block_group_kobjs[index] = NULL;
10651 clear_avail_alloc_bits(fs_info, block_group->flags);
10653 up_write(&block_group->space_info->groups_sem);
10659 if (block_group->has_caching_ctl)
10660 caching_ctl = get_caching_control(block_group);
10661 if (block_group->cached == BTRFS_CACHE_STARTED)
10662 wait_block_group_cache_done(block_group);
10663 if (block_group->has_caching_ctl) {
10664 down_write(&fs_info->commit_root_sem);
10665 if (!caching_ctl) {
10666 struct btrfs_caching_control *ctl;
10668 list_for_each_entry(ctl,
10669 &fs_info->caching_block_groups, list)
10670 if (ctl->block_group == block_group) {
10672 refcount_inc(&caching_ctl->count);
10677 list_del_init(&caching_ctl->list);
10678 up_write(&fs_info->commit_root_sem);
10680 /* Once for the caching bgs list and once for us. */
10681 put_caching_control(caching_ctl);
10682 put_caching_control(caching_ctl);
10686 spin_lock(&trans->transaction->dirty_bgs_lock);
10687 if (!list_empty(&block_group->dirty_list)) {
10690 if (!list_empty(&block_group->io_list)) {
10693 spin_unlock(&trans->transaction->dirty_bgs_lock);
10694 btrfs_remove_free_space_cache(block_group);
10696 spin_lock(&block_group->space_info->lock);
10697 list_del_init(&block_group->ro_list);
10699 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10700 WARN_ON(block_group->space_info->total_bytes
10701 < block_group->key.offset);
10702 WARN_ON(block_group->space_info->bytes_readonly
10703 < block_group->key.offset);
10704 WARN_ON(block_group->space_info->disk_total
10705 < block_group->key.offset * factor);
10707 block_group->space_info->total_bytes -= block_group->key.offset;
10708 block_group->space_info->bytes_readonly -= block_group->key.offset;
10709 block_group->space_info->disk_total -= block_group->key.offset * factor;
10711 spin_unlock(&block_group->space_info->lock);
10713 memcpy(&key, &block_group->key, sizeof(key));
10715 mutex_lock(&fs_info->chunk_mutex);
10716 if (!list_empty(&em->list)) {
10717 /* We're in the transaction->pending_chunks list. */
10718 free_extent_map(em);
10720 spin_lock(&block_group->lock);
10721 block_group->removed = 1;
10723 * At this point trimming can't start on this block group, because we
10724 * removed the block group from the tree fs_info->block_group_cache_tree
10725 * so no one can't find it anymore and even if someone already got this
10726 * block group before we removed it from the rbtree, they have already
10727 * incremented block_group->trimming - if they didn't, they won't find
10728 * any free space entries because we already removed them all when we
10729 * called btrfs_remove_free_space_cache().
10731 * And we must not remove the extent map from the fs_info->mapping_tree
10732 * to prevent the same logical address range and physical device space
10733 * ranges from being reused for a new block group. This is because our
10734 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10735 * completely transactionless, so while it is trimming a range the
10736 * currently running transaction might finish and a new one start,
10737 * allowing for new block groups to be created that can reuse the same
10738 * physical device locations unless we take this special care.
10740 * There may also be an implicit trim operation if the file system
10741 * is mounted with -odiscard. The same protections must remain
10742 * in place until the extents have been discarded completely when
10743 * the transaction commit has completed.
10745 remove_em = (atomic_read(&block_group->trimming) == 0);
10747 * Make sure a trimmer task always sees the em in the pinned_chunks list
10748 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10749 * before checking block_group->removed).
10753 * Our em might be in trans->transaction->pending_chunks which
10754 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10755 * and so is the fs_info->pinned_chunks list.
10757 * So at this point we must be holding the chunk_mutex to avoid
10758 * any races with chunk allocation (more specifically at
10759 * volumes.c:contains_pending_extent()), to ensure it always
10760 * sees the em, either in the pending_chunks list or in the
10761 * pinned_chunks list.
10763 list_move_tail(&em->list, &fs_info->pinned_chunks);
10765 spin_unlock(&block_group->lock);
10768 struct extent_map_tree *em_tree;
10770 em_tree = &fs_info->mapping_tree.map_tree;
10771 write_lock(&em_tree->lock);
10773 * The em might be in the pending_chunks list, so make sure the
10774 * chunk mutex is locked, since remove_extent_mapping() will
10775 * delete us from that list.
10777 remove_extent_mapping(em_tree, em);
10778 write_unlock(&em_tree->lock);
10779 /* once for the tree */
10780 free_extent_map(em);
10783 mutex_unlock(&fs_info->chunk_mutex);
10785 ret = remove_block_group_free_space(trans, fs_info, block_group);
10789 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10795 ret = btrfs_del_item(trans, root, path);
10798 /* Once for the lookup reference */
10799 btrfs_put_block_group(block_group);
10800 btrfs_free_path(path);
10804 struct btrfs_trans_handle *
10805 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10806 const u64 chunk_offset)
10808 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10809 struct extent_map *em;
10810 struct map_lookup *map;
10811 unsigned int num_items;
10813 read_lock(&em_tree->lock);
10814 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10815 read_unlock(&em_tree->lock);
10816 ASSERT(em && em->start == chunk_offset);
10819 * We need to reserve 3 + N units from the metadata space info in order
10820 * to remove a block group (done at btrfs_remove_chunk() and at
10821 * btrfs_remove_block_group()), which are used for:
10823 * 1 unit for adding the free space inode's orphan (located in the tree
10825 * 1 unit for deleting the block group item (located in the extent
10827 * 1 unit for deleting the free space item (located in tree of tree
10829 * N units for deleting N device extent items corresponding to each
10830 * stripe (located in the device tree).
10832 * In order to remove a block group we also need to reserve units in the
10833 * system space info in order to update the chunk tree (update one or
10834 * more device items and remove one chunk item), but this is done at
10835 * btrfs_remove_chunk() through a call to check_system_chunk().
10837 map = em->map_lookup;
10838 num_items = 3 + map->num_stripes;
10839 free_extent_map(em);
10841 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10846 * Process the unused_bgs list and remove any that don't have any allocated
10847 * space inside of them.
10849 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10851 struct btrfs_block_group_cache *block_group;
10852 struct btrfs_space_info *space_info;
10853 struct btrfs_trans_handle *trans;
10856 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10859 spin_lock(&fs_info->unused_bgs_lock);
10860 while (!list_empty(&fs_info->unused_bgs)) {
10864 block_group = list_first_entry(&fs_info->unused_bgs,
10865 struct btrfs_block_group_cache,
10867 list_del_init(&block_group->bg_list);
10869 space_info = block_group->space_info;
10871 if (ret || btrfs_mixed_space_info(space_info)) {
10872 btrfs_put_block_group(block_group);
10875 spin_unlock(&fs_info->unused_bgs_lock);
10877 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10879 /* Don't want to race with allocators so take the groups_sem */
10880 down_write(&space_info->groups_sem);
10881 spin_lock(&block_group->lock);
10882 if (block_group->reserved || block_group->pinned ||
10883 btrfs_block_group_used(&block_group->item) ||
10885 list_is_singular(&block_group->list)) {
10887 * We want to bail if we made new allocations or have
10888 * outstanding allocations in this block group. We do
10889 * the ro check in case balance is currently acting on
10890 * this block group.
10892 spin_unlock(&block_group->lock);
10893 up_write(&space_info->groups_sem);
10896 spin_unlock(&block_group->lock);
10898 /* We don't want to force the issue, only flip if it's ok. */
10899 ret = inc_block_group_ro(block_group, 0);
10900 up_write(&space_info->groups_sem);
10907 * Want to do this before we do anything else so we can recover
10908 * properly if we fail to join the transaction.
10910 trans = btrfs_start_trans_remove_block_group(fs_info,
10911 block_group->key.objectid);
10912 if (IS_ERR(trans)) {
10913 btrfs_dec_block_group_ro(block_group);
10914 ret = PTR_ERR(trans);
10919 * We could have pending pinned extents for this block group,
10920 * just delete them, we don't care about them anymore.
10922 start = block_group->key.objectid;
10923 end = start + block_group->key.offset - 1;
10925 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10926 * btrfs_finish_extent_commit(). If we are at transaction N,
10927 * another task might be running finish_extent_commit() for the
10928 * previous transaction N - 1, and have seen a range belonging
10929 * to the block group in freed_extents[] before we were able to
10930 * clear the whole block group range from freed_extents[]. This
10931 * means that task can lookup for the block group after we
10932 * unpinned it from freed_extents[] and removed it, leading to
10933 * a BUG_ON() at btrfs_unpin_extent_range().
10935 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10936 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10939 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10940 btrfs_dec_block_group_ro(block_group);
10943 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10946 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10947 btrfs_dec_block_group_ro(block_group);
10950 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10952 /* Reset pinned so btrfs_put_block_group doesn't complain */
10953 spin_lock(&space_info->lock);
10954 spin_lock(&block_group->lock);
10956 space_info->bytes_pinned -= block_group->pinned;
10957 space_info->bytes_readonly += block_group->pinned;
10958 percpu_counter_add(&space_info->total_bytes_pinned,
10959 -block_group->pinned);
10960 block_group->pinned = 0;
10962 spin_unlock(&block_group->lock);
10963 spin_unlock(&space_info->lock);
10965 /* DISCARD can flip during remount */
10966 trimming = btrfs_test_opt(fs_info, DISCARD);
10968 /* Implicit trim during transaction commit. */
10970 btrfs_get_block_group_trimming(block_group);
10973 * Btrfs_remove_chunk will abort the transaction if things go
10976 ret = btrfs_remove_chunk(trans, fs_info,
10977 block_group->key.objectid);
10981 btrfs_put_block_group_trimming(block_group);
10986 * If we're not mounted with -odiscard, we can just forget
10987 * about this block group. Otherwise we'll need to wait
10988 * until transaction commit to do the actual discard.
10991 spin_lock(&fs_info->unused_bgs_lock);
10993 * A concurrent scrub might have added us to the list
10994 * fs_info->unused_bgs, so use a list_move operation
10995 * to add the block group to the deleted_bgs list.
10997 list_move(&block_group->bg_list,
10998 &trans->transaction->deleted_bgs);
10999 spin_unlock(&fs_info->unused_bgs_lock);
11000 btrfs_get_block_group(block_group);
11003 btrfs_end_transaction(trans);
11005 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
11006 btrfs_put_block_group(block_group);
11007 spin_lock(&fs_info->unused_bgs_lock);
11009 spin_unlock(&fs_info->unused_bgs_lock);
11012 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
11014 struct btrfs_space_info *space_info;
11015 struct btrfs_super_block *disk_super;
11021 disk_super = fs_info->super_copy;
11022 if (!btrfs_super_root(disk_super))
11025 features = btrfs_super_incompat_flags(disk_super);
11026 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11029 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11030 ret = create_space_info(fs_info, flags, &space_info);
11035 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11036 ret = create_space_info(fs_info, flags, &space_info);
11038 flags = BTRFS_BLOCK_GROUP_METADATA;
11039 ret = create_space_info(fs_info, flags, &space_info);
11043 flags = BTRFS_BLOCK_GROUP_DATA;
11044 ret = create_space_info(fs_info, flags, &space_info);
11050 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
11051 u64 start, u64 end)
11053 return unpin_extent_range(fs_info, start, end, false);
11057 * It used to be that old block groups would be left around forever.
11058 * Iterating over them would be enough to trim unused space. Since we
11059 * now automatically remove them, we also need to iterate over unallocated
11062 * We don't want a transaction for this since the discard may take a
11063 * substantial amount of time. We don't require that a transaction be
11064 * running, but we do need to take a running transaction into account
11065 * to ensure that we're not discarding chunks that were released in
11066 * the current transaction.
11068 * Holding the chunks lock will prevent other threads from allocating
11069 * or releasing chunks, but it won't prevent a running transaction
11070 * from committing and releasing the memory that the pending chunks
11071 * list head uses. For that, we need to take a reference to the
11074 static int btrfs_trim_free_extents(struct btrfs_device *device,
11075 u64 minlen, u64 *trimmed)
11077 u64 start = 0, len = 0;
11082 /* Discard not supported = nothing to do. */
11083 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
11086 /* Not writeable = nothing to do. */
11087 if (!device->writeable)
11090 /* No free space = nothing to do. */
11091 if (device->total_bytes <= device->bytes_used)
11097 struct btrfs_fs_info *fs_info = device->fs_info;
11098 struct btrfs_transaction *trans;
11101 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11105 down_read(&fs_info->commit_root_sem);
11107 spin_lock(&fs_info->trans_lock);
11108 trans = fs_info->running_transaction;
11110 refcount_inc(&trans->use_count);
11111 spin_unlock(&fs_info->trans_lock);
11113 ret = find_free_dev_extent_start(trans, device, minlen, start,
11116 btrfs_put_transaction(trans);
11119 up_read(&fs_info->commit_root_sem);
11120 mutex_unlock(&fs_info->chunk_mutex);
11121 if (ret == -ENOSPC)
11126 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11127 up_read(&fs_info->commit_root_sem);
11128 mutex_unlock(&fs_info->chunk_mutex);
11136 if (fatal_signal_pending(current)) {
11137 ret = -ERESTARTSYS;
11148 * Trim the whole filesystem by:
11149 * 1) trimming the free space in each block group
11150 * 2) trimming the unallocated space on each device
11152 * This will also continue trimming even if a block group or device encounters
11153 * an error. The return value will be the last error, or 0 if nothing bad
11156 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11158 struct btrfs_block_group_cache *cache = NULL;
11159 struct btrfs_device *device;
11160 struct list_head *devices;
11166 u64 dev_failed = 0;
11171 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11172 for (; cache; cache = next_block_group(fs_info, cache)) {
11173 if (cache->key.objectid >= (range->start + range->len)) {
11174 btrfs_put_block_group(cache);
11178 start = max(range->start, cache->key.objectid);
11179 end = min(range->start + range->len,
11180 cache->key.objectid + cache->key.offset);
11182 if (end - start >= range->minlen) {
11183 if (!block_group_cache_done(cache)) {
11184 ret = cache_block_group(cache, 0);
11190 ret = wait_block_group_cache_done(cache);
11197 ret = btrfs_trim_block_group(cache,
11203 trimmed += group_trimmed;
11213 btrfs_warn(fs_info,
11214 "failed to trim %llu block group(s), last error %d",
11215 bg_failed, bg_ret);
11216 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11217 devices = &fs_info->fs_devices->devices;
11218 list_for_each_entry(device, devices, dev_list) {
11219 ret = btrfs_trim_free_extents(device, range->minlen,
11227 trimmed += group_trimmed;
11229 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11232 btrfs_warn(fs_info,
11233 "failed to trim %llu device(s), last error %d",
11234 dev_failed, dev_ret);
11235 range->len = trimmed;
11242 * btrfs_{start,end}_write_no_snapshotting() are similar to
11243 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11244 * data into the page cache through nocow before the subvolume is snapshoted,
11245 * but flush the data into disk after the snapshot creation, or to prevent
11246 * operations while snapshotting is ongoing and that cause the snapshot to be
11247 * inconsistent (writes followed by expanding truncates for example).
11249 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11251 percpu_counter_dec(&root->subv_writers->counter);
11253 * Make sure counter is updated before we wake up waiters.
11256 if (waitqueue_active(&root->subv_writers->wait))
11257 wake_up(&root->subv_writers->wait);
11260 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11262 if (atomic_read(&root->will_be_snapshotted))
11265 percpu_counter_inc(&root->subv_writers->counter);
11267 * Make sure counter is updated before we check for snapshot creation.
11270 if (atomic_read(&root->will_be_snapshotted)) {
11271 btrfs_end_write_no_snapshotting(root);
11277 static int wait_snapshotting_atomic_t(atomic_t *a)
11283 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11288 ret = btrfs_start_write_no_snapshotting(root);
11291 wait_on_atomic_t(&root->will_be_snapshotted,
11292 wait_snapshotting_atomic_t,
11293 TASK_UNINTERRUPTIBLE);