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
1174 * block, which must be aligned to
1178 IS_ALIGNED(offset, eb->fs_info->nodesize))
1181 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1182 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1184 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1185 ASSERT(eb->fs_info);
1187 * Every shared one has parent tree
1188 * block, which must be aligned to
1192 IS_ALIGNED(offset, eb->fs_info->nodesize))
1196 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1201 btrfs_print_leaf((struct extent_buffer *)eb);
1202 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1206 return BTRFS_REF_TYPE_INVALID;
1209 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1211 u32 high_crc = ~(u32)0;
1212 u32 low_crc = ~(u32)0;
1215 lenum = cpu_to_le64(root_objectid);
1216 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1217 lenum = cpu_to_le64(owner);
1218 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1219 lenum = cpu_to_le64(offset);
1220 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1222 return ((u64)high_crc << 31) ^ (u64)low_crc;
1225 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1226 struct btrfs_extent_data_ref *ref)
1228 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1229 btrfs_extent_data_ref_objectid(leaf, ref),
1230 btrfs_extent_data_ref_offset(leaf, ref));
1233 static int match_extent_data_ref(struct extent_buffer *leaf,
1234 struct btrfs_extent_data_ref *ref,
1235 u64 root_objectid, u64 owner, u64 offset)
1237 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1238 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1239 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1244 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1245 struct btrfs_fs_info *fs_info,
1246 struct btrfs_path *path,
1247 u64 bytenr, u64 parent,
1249 u64 owner, u64 offset)
1251 struct btrfs_root *root = fs_info->extent_root;
1252 struct btrfs_key key;
1253 struct btrfs_extent_data_ref *ref;
1254 struct extent_buffer *leaf;
1260 key.objectid = bytenr;
1262 key.type = BTRFS_SHARED_DATA_REF_KEY;
1263 key.offset = parent;
1265 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1266 key.offset = hash_extent_data_ref(root_objectid,
1271 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1281 key.type = BTRFS_EXTENT_REF_V0_KEY;
1282 btrfs_release_path(path);
1283 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1294 leaf = path->nodes[0];
1295 nritems = btrfs_header_nritems(leaf);
1297 if (path->slots[0] >= nritems) {
1298 ret = btrfs_next_leaf(root, path);
1304 leaf = path->nodes[0];
1305 nritems = btrfs_header_nritems(leaf);
1309 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1310 if (key.objectid != bytenr ||
1311 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1314 ref = btrfs_item_ptr(leaf, path->slots[0],
1315 struct btrfs_extent_data_ref);
1317 if (match_extent_data_ref(leaf, ref, root_objectid,
1320 btrfs_release_path(path);
1332 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1333 struct btrfs_fs_info *fs_info,
1334 struct btrfs_path *path,
1335 u64 bytenr, u64 parent,
1336 u64 root_objectid, u64 owner,
1337 u64 offset, int refs_to_add)
1339 struct btrfs_root *root = fs_info->extent_root;
1340 struct btrfs_key key;
1341 struct extent_buffer *leaf;
1346 key.objectid = bytenr;
1348 key.type = BTRFS_SHARED_DATA_REF_KEY;
1349 key.offset = parent;
1350 size = sizeof(struct btrfs_shared_data_ref);
1352 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1353 key.offset = hash_extent_data_ref(root_objectid,
1355 size = sizeof(struct btrfs_extent_data_ref);
1358 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1359 if (ret && ret != -EEXIST)
1362 leaf = path->nodes[0];
1364 struct btrfs_shared_data_ref *ref;
1365 ref = btrfs_item_ptr(leaf, path->slots[0],
1366 struct btrfs_shared_data_ref);
1368 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1370 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1371 num_refs += refs_to_add;
1372 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1375 struct btrfs_extent_data_ref *ref;
1376 while (ret == -EEXIST) {
1377 ref = btrfs_item_ptr(leaf, path->slots[0],
1378 struct btrfs_extent_data_ref);
1379 if (match_extent_data_ref(leaf, ref, root_objectid,
1382 btrfs_release_path(path);
1384 ret = btrfs_insert_empty_item(trans, root, path, &key,
1386 if (ret && ret != -EEXIST)
1389 leaf = path->nodes[0];
1391 ref = btrfs_item_ptr(leaf, path->slots[0],
1392 struct btrfs_extent_data_ref);
1394 btrfs_set_extent_data_ref_root(leaf, ref,
1396 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1397 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1398 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1400 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1401 num_refs += refs_to_add;
1402 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1405 btrfs_mark_buffer_dirty(leaf);
1408 btrfs_release_path(path);
1412 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1413 struct btrfs_fs_info *fs_info,
1414 struct btrfs_path *path,
1415 int refs_to_drop, int *last_ref)
1417 struct btrfs_key key;
1418 struct btrfs_extent_data_ref *ref1 = NULL;
1419 struct btrfs_shared_data_ref *ref2 = NULL;
1420 struct extent_buffer *leaf;
1424 leaf = path->nodes[0];
1425 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1427 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1428 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1429 struct btrfs_extent_data_ref);
1430 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1431 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1432 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1433 struct btrfs_shared_data_ref);
1434 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1435 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1436 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1437 struct btrfs_extent_ref_v0 *ref0;
1438 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1439 struct btrfs_extent_ref_v0);
1440 num_refs = btrfs_ref_count_v0(leaf, ref0);
1446 BUG_ON(num_refs < refs_to_drop);
1447 num_refs -= refs_to_drop;
1449 if (num_refs == 0) {
1450 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1453 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1454 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1455 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1456 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1457 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1459 struct btrfs_extent_ref_v0 *ref0;
1460 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1461 struct btrfs_extent_ref_v0);
1462 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1465 btrfs_mark_buffer_dirty(leaf);
1470 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1471 struct btrfs_extent_inline_ref *iref)
1473 struct btrfs_key key;
1474 struct extent_buffer *leaf;
1475 struct btrfs_extent_data_ref *ref1;
1476 struct btrfs_shared_data_ref *ref2;
1480 leaf = path->nodes[0];
1481 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1484 * If type is invalid, we should have bailed out earlier than
1487 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1488 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1489 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1490 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1491 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1493 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1494 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1496 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1497 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1498 struct btrfs_extent_data_ref);
1499 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1500 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1501 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1502 struct btrfs_shared_data_ref);
1503 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1504 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1505 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1506 struct btrfs_extent_ref_v0 *ref0;
1507 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1508 struct btrfs_extent_ref_v0);
1509 num_refs = btrfs_ref_count_v0(leaf, ref0);
1517 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1518 struct btrfs_fs_info *fs_info,
1519 struct btrfs_path *path,
1520 u64 bytenr, u64 parent,
1523 struct btrfs_root *root = fs_info->extent_root;
1524 struct btrfs_key key;
1527 key.objectid = bytenr;
1529 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1530 key.offset = parent;
1532 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1533 key.offset = root_objectid;
1536 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1539 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1540 if (ret == -ENOENT && parent) {
1541 btrfs_release_path(path);
1542 key.type = BTRFS_EXTENT_REF_V0_KEY;
1543 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1551 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1552 struct btrfs_fs_info *fs_info,
1553 struct btrfs_path *path,
1554 u64 bytenr, u64 parent,
1557 struct btrfs_key key;
1560 key.objectid = bytenr;
1562 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1563 key.offset = parent;
1565 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1566 key.offset = root_objectid;
1569 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1571 btrfs_release_path(path);
1575 static inline int extent_ref_type(u64 parent, u64 owner)
1578 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1580 type = BTRFS_SHARED_BLOCK_REF_KEY;
1582 type = BTRFS_TREE_BLOCK_REF_KEY;
1585 type = BTRFS_SHARED_DATA_REF_KEY;
1587 type = BTRFS_EXTENT_DATA_REF_KEY;
1592 static int find_next_key(struct btrfs_path *path, int level,
1593 struct btrfs_key *key)
1596 for (; level < BTRFS_MAX_LEVEL; level++) {
1597 if (!path->nodes[level])
1599 if (path->slots[level] + 1 >=
1600 btrfs_header_nritems(path->nodes[level]))
1603 btrfs_item_key_to_cpu(path->nodes[level], key,
1604 path->slots[level] + 1);
1606 btrfs_node_key_to_cpu(path->nodes[level], key,
1607 path->slots[level] + 1);
1614 * look for inline back ref. if back ref is found, *ref_ret is set
1615 * to the address of inline back ref, and 0 is returned.
1617 * if back ref isn't found, *ref_ret is set to the address where it
1618 * should be inserted, and -ENOENT is returned.
1620 * if insert is true and there are too many inline back refs, the path
1621 * points to the extent item, and -EAGAIN is returned.
1623 * NOTE: inline back refs are ordered in the same way that back ref
1624 * items in the tree are ordered.
1626 static noinline_for_stack
1627 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1628 struct btrfs_fs_info *fs_info,
1629 struct btrfs_path *path,
1630 struct btrfs_extent_inline_ref **ref_ret,
1631 u64 bytenr, u64 num_bytes,
1632 u64 parent, u64 root_objectid,
1633 u64 owner, u64 offset, int insert)
1635 struct btrfs_root *root = fs_info->extent_root;
1636 struct btrfs_key key;
1637 struct extent_buffer *leaf;
1638 struct btrfs_extent_item *ei;
1639 struct btrfs_extent_inline_ref *iref;
1649 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1652 key.objectid = bytenr;
1653 key.type = BTRFS_EXTENT_ITEM_KEY;
1654 key.offset = num_bytes;
1656 want = extent_ref_type(parent, owner);
1658 extra_size = btrfs_extent_inline_ref_size(want);
1659 path->keep_locks = 1;
1664 * Owner is our parent level, so we can just add one to get the level
1665 * for the block we are interested in.
1667 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1668 key.type = BTRFS_METADATA_ITEM_KEY;
1673 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1680 * We may be a newly converted file system which still has the old fat
1681 * extent entries for metadata, so try and see if we have one of those.
1683 if (ret > 0 && skinny_metadata) {
1684 skinny_metadata = false;
1685 if (path->slots[0]) {
1687 btrfs_item_key_to_cpu(path->nodes[0], &key,
1689 if (key.objectid == bytenr &&
1690 key.type == BTRFS_EXTENT_ITEM_KEY &&
1691 key.offset == num_bytes)
1695 key.objectid = bytenr;
1696 key.type = BTRFS_EXTENT_ITEM_KEY;
1697 key.offset = num_bytes;
1698 btrfs_release_path(path);
1703 if (ret && !insert) {
1706 } else if (WARN_ON(ret)) {
1711 leaf = path->nodes[0];
1712 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1713 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1714 if (item_size < sizeof(*ei)) {
1719 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1725 leaf = path->nodes[0];
1726 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1729 BUG_ON(item_size < sizeof(*ei));
1731 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1732 flags = btrfs_extent_flags(leaf, ei);
1734 ptr = (unsigned long)(ei + 1);
1735 end = (unsigned long)ei + item_size;
1737 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1738 ptr += sizeof(struct btrfs_tree_block_info);
1742 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1743 needed = BTRFS_REF_TYPE_DATA;
1745 needed = BTRFS_REF_TYPE_BLOCK;
1753 iref = (struct btrfs_extent_inline_ref *)ptr;
1754 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1755 if (type == BTRFS_REF_TYPE_INVALID) {
1763 ptr += btrfs_extent_inline_ref_size(type);
1767 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1768 struct btrfs_extent_data_ref *dref;
1769 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1770 if (match_extent_data_ref(leaf, dref, root_objectid,
1775 if (hash_extent_data_ref_item(leaf, dref) <
1776 hash_extent_data_ref(root_objectid, owner, offset))
1780 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1782 if (parent == ref_offset) {
1786 if (ref_offset < parent)
1789 if (root_objectid == ref_offset) {
1793 if (ref_offset < root_objectid)
1797 ptr += btrfs_extent_inline_ref_size(type);
1799 if (err == -ENOENT && insert) {
1800 if (item_size + extra_size >=
1801 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1806 * To add new inline back ref, we have to make sure
1807 * there is no corresponding back ref item.
1808 * For simplicity, we just do not add new inline back
1809 * ref if there is any kind of item for this block
1811 if (find_next_key(path, 0, &key) == 0 &&
1812 key.objectid == bytenr &&
1813 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1818 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1821 path->keep_locks = 0;
1822 btrfs_unlock_up_safe(path, 1);
1828 * helper to add new inline back ref
1830 static noinline_for_stack
1831 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1832 struct btrfs_path *path,
1833 struct btrfs_extent_inline_ref *iref,
1834 u64 parent, u64 root_objectid,
1835 u64 owner, u64 offset, int refs_to_add,
1836 struct btrfs_delayed_extent_op *extent_op)
1838 struct extent_buffer *leaf;
1839 struct btrfs_extent_item *ei;
1842 unsigned long item_offset;
1847 leaf = path->nodes[0];
1848 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1849 item_offset = (unsigned long)iref - (unsigned long)ei;
1851 type = extent_ref_type(parent, owner);
1852 size = btrfs_extent_inline_ref_size(type);
1854 btrfs_extend_item(fs_info, path, size);
1856 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1857 refs = btrfs_extent_refs(leaf, ei);
1858 refs += refs_to_add;
1859 btrfs_set_extent_refs(leaf, ei, refs);
1861 __run_delayed_extent_op(extent_op, leaf, ei);
1863 ptr = (unsigned long)ei + item_offset;
1864 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1865 if (ptr < end - size)
1866 memmove_extent_buffer(leaf, ptr + size, ptr,
1869 iref = (struct btrfs_extent_inline_ref *)ptr;
1870 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1871 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1872 struct btrfs_extent_data_ref *dref;
1873 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1874 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1875 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1876 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1877 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1878 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1879 struct btrfs_shared_data_ref *sref;
1880 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1881 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1882 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1883 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1884 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1886 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1888 btrfs_mark_buffer_dirty(leaf);
1891 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1892 struct btrfs_fs_info *fs_info,
1893 struct btrfs_path *path,
1894 struct btrfs_extent_inline_ref **ref_ret,
1895 u64 bytenr, u64 num_bytes, u64 parent,
1896 u64 root_objectid, u64 owner, u64 offset)
1900 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1901 bytenr, num_bytes, parent,
1902 root_objectid, owner, offset, 0);
1906 btrfs_release_path(path);
1909 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1910 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1911 parent, root_objectid);
1913 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1914 parent, root_objectid, owner,
1921 * helper to update/remove inline back ref
1923 static noinline_for_stack
1924 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1925 struct btrfs_path *path,
1926 struct btrfs_extent_inline_ref *iref,
1928 struct btrfs_delayed_extent_op *extent_op,
1931 struct extent_buffer *leaf;
1932 struct btrfs_extent_item *ei;
1933 struct btrfs_extent_data_ref *dref = NULL;
1934 struct btrfs_shared_data_ref *sref = NULL;
1942 leaf = path->nodes[0];
1943 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1944 refs = btrfs_extent_refs(leaf, ei);
1945 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1946 refs += refs_to_mod;
1947 btrfs_set_extent_refs(leaf, ei, refs);
1949 __run_delayed_extent_op(extent_op, leaf, ei);
1952 * If type is invalid, we should have bailed out after
1953 * lookup_inline_extent_backref().
1955 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1956 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1958 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1959 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1960 refs = btrfs_extent_data_ref_count(leaf, dref);
1961 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1962 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1963 refs = btrfs_shared_data_ref_count(leaf, sref);
1966 BUG_ON(refs_to_mod != -1);
1969 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1970 refs += refs_to_mod;
1973 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1974 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1976 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1979 size = btrfs_extent_inline_ref_size(type);
1980 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1981 ptr = (unsigned long)iref;
1982 end = (unsigned long)ei + item_size;
1983 if (ptr + size < end)
1984 memmove_extent_buffer(leaf, ptr, ptr + size,
1987 btrfs_truncate_item(fs_info, path, item_size, 1);
1989 btrfs_mark_buffer_dirty(leaf);
1992 static noinline_for_stack
1993 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1994 struct btrfs_fs_info *fs_info,
1995 struct btrfs_path *path,
1996 u64 bytenr, u64 num_bytes, u64 parent,
1997 u64 root_objectid, u64 owner,
1998 u64 offset, int refs_to_add,
1999 struct btrfs_delayed_extent_op *extent_op)
2001 struct btrfs_extent_inline_ref *iref;
2004 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
2005 bytenr, num_bytes, parent,
2006 root_objectid, owner, offset, 1);
2008 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
2009 update_inline_extent_backref(fs_info, path, iref,
2010 refs_to_add, extent_op, NULL);
2011 } else if (ret == -ENOENT) {
2012 setup_inline_extent_backref(fs_info, path, iref, parent,
2013 root_objectid, owner, offset,
2014 refs_to_add, extent_op);
2020 static int insert_extent_backref(struct btrfs_trans_handle *trans,
2021 struct btrfs_fs_info *fs_info,
2022 struct btrfs_path *path,
2023 u64 bytenr, u64 parent, u64 root_objectid,
2024 u64 owner, u64 offset, int refs_to_add)
2027 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2028 BUG_ON(refs_to_add != 1);
2029 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
2030 parent, root_objectid);
2032 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
2033 parent, root_objectid,
2034 owner, offset, refs_to_add);
2039 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2040 struct btrfs_fs_info *fs_info,
2041 struct btrfs_path *path,
2042 struct btrfs_extent_inline_ref *iref,
2043 int refs_to_drop, int is_data, int *last_ref)
2047 BUG_ON(!is_data && refs_to_drop != 1);
2049 update_inline_extent_backref(fs_info, path, iref,
2050 -refs_to_drop, NULL, last_ref);
2051 } else if (is_data) {
2052 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
2056 ret = btrfs_del_item(trans, fs_info->extent_root, path);
2061 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2062 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2063 u64 *discarded_bytes)
2066 u64 bytes_left, end;
2067 u64 aligned_start = ALIGN(start, 1 << 9);
2069 if (WARN_ON(start != aligned_start)) {
2070 len -= aligned_start - start;
2071 len = round_down(len, 1 << 9);
2072 start = aligned_start;
2075 *discarded_bytes = 0;
2083 /* Skip any superblocks on this device. */
2084 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2085 u64 sb_start = btrfs_sb_offset(j);
2086 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2087 u64 size = sb_start - start;
2089 if (!in_range(sb_start, start, bytes_left) &&
2090 !in_range(sb_end, start, bytes_left) &&
2091 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2095 * Superblock spans beginning of range. Adjust start and
2098 if (sb_start <= start) {
2099 start += sb_end - start;
2104 bytes_left = end - start;
2109 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2112 *discarded_bytes += size;
2113 else if (ret != -EOPNOTSUPP)
2122 bytes_left = end - start;
2126 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2129 *discarded_bytes += bytes_left;
2134 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2135 u64 num_bytes, u64 *actual_bytes)
2138 u64 discarded_bytes = 0;
2139 struct btrfs_bio *bbio = NULL;
2143 * Avoid races with device replace and make sure our bbio has devices
2144 * associated to its stripes that don't go away while we are discarding.
2146 btrfs_bio_counter_inc_blocked(fs_info);
2147 /* Tell the block device(s) that the sectors can be discarded */
2148 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2150 /* Error condition is -ENOMEM */
2152 struct btrfs_bio_stripe *stripe = bbio->stripes;
2156 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2158 if (!stripe->dev->can_discard)
2161 ret = btrfs_issue_discard(stripe->dev->bdev,
2166 discarded_bytes += bytes;
2167 else if (ret != -EOPNOTSUPP)
2168 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2171 * Just in case we get back EOPNOTSUPP for some reason,
2172 * just ignore the return value so we don't screw up
2173 * people calling discard_extent.
2177 btrfs_put_bbio(bbio);
2179 btrfs_bio_counter_dec(fs_info);
2182 *actual_bytes = discarded_bytes;
2185 if (ret == -EOPNOTSUPP)
2190 /* Can return -ENOMEM */
2191 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2192 struct btrfs_fs_info *fs_info,
2193 u64 bytenr, u64 num_bytes, u64 parent,
2194 u64 root_objectid, u64 owner, u64 offset)
2196 int old_ref_mod, new_ref_mod;
2199 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2200 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2202 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2203 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2205 root_objectid, (int)owner,
2206 BTRFS_ADD_DELAYED_REF, NULL,
2207 &old_ref_mod, &new_ref_mod);
2209 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2211 root_objectid, owner, offset,
2212 0, BTRFS_ADD_DELAYED_REF,
2213 &old_ref_mod, &new_ref_mod);
2216 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2217 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2222 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2223 struct btrfs_fs_info *fs_info,
2224 struct btrfs_delayed_ref_node *node,
2225 u64 parent, u64 root_objectid,
2226 u64 owner, u64 offset, int refs_to_add,
2227 struct btrfs_delayed_extent_op *extent_op)
2229 struct btrfs_path *path;
2230 struct extent_buffer *leaf;
2231 struct btrfs_extent_item *item;
2232 struct btrfs_key key;
2233 u64 bytenr = node->bytenr;
2234 u64 num_bytes = node->num_bytes;
2238 path = btrfs_alloc_path();
2242 path->reada = READA_FORWARD;
2243 path->leave_spinning = 1;
2244 /* this will setup the path even if it fails to insert the back ref */
2245 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2246 num_bytes, parent, root_objectid,
2248 refs_to_add, extent_op);
2249 if ((ret < 0 && ret != -EAGAIN) || !ret)
2253 * Ok we had -EAGAIN which means we didn't have space to insert and
2254 * inline extent ref, so just update the reference count and add a
2257 leaf = path->nodes[0];
2258 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2259 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2260 refs = btrfs_extent_refs(leaf, item);
2261 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2263 __run_delayed_extent_op(extent_op, leaf, item);
2265 btrfs_mark_buffer_dirty(leaf);
2266 btrfs_release_path(path);
2268 path->reada = READA_FORWARD;
2269 path->leave_spinning = 1;
2270 /* now insert the actual backref */
2271 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2272 root_objectid, owner, offset, refs_to_add);
2274 btrfs_abort_transaction(trans, ret);
2276 btrfs_free_path(path);
2280 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2281 struct btrfs_fs_info *fs_info,
2282 struct btrfs_delayed_ref_node *node,
2283 struct btrfs_delayed_extent_op *extent_op,
2284 int insert_reserved)
2287 struct btrfs_delayed_data_ref *ref;
2288 struct btrfs_key ins;
2293 ins.objectid = node->bytenr;
2294 ins.offset = node->num_bytes;
2295 ins.type = BTRFS_EXTENT_ITEM_KEY;
2297 ref = btrfs_delayed_node_to_data_ref(node);
2298 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2300 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2301 parent = ref->parent;
2302 ref_root = ref->root;
2304 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2306 flags |= extent_op->flags_to_set;
2307 ret = alloc_reserved_file_extent(trans, fs_info,
2308 parent, ref_root, flags,
2309 ref->objectid, ref->offset,
2310 &ins, node->ref_mod);
2311 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2312 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2313 ref_root, ref->objectid,
2314 ref->offset, node->ref_mod,
2316 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2317 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2318 ref_root, ref->objectid,
2319 ref->offset, node->ref_mod,
2327 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2328 struct extent_buffer *leaf,
2329 struct btrfs_extent_item *ei)
2331 u64 flags = btrfs_extent_flags(leaf, ei);
2332 if (extent_op->update_flags) {
2333 flags |= extent_op->flags_to_set;
2334 btrfs_set_extent_flags(leaf, ei, flags);
2337 if (extent_op->update_key) {
2338 struct btrfs_tree_block_info *bi;
2339 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2340 bi = (struct btrfs_tree_block_info *)(ei + 1);
2341 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2345 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2346 struct btrfs_fs_info *fs_info,
2347 struct btrfs_delayed_ref_node *node,
2348 struct btrfs_delayed_extent_op *extent_op)
2350 struct btrfs_key key;
2351 struct btrfs_path *path;
2352 struct btrfs_extent_item *ei;
2353 struct extent_buffer *leaf;
2357 int metadata = !extent_op->is_data;
2362 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2365 path = btrfs_alloc_path();
2369 key.objectid = node->bytenr;
2372 key.type = BTRFS_METADATA_ITEM_KEY;
2373 key.offset = extent_op->level;
2375 key.type = BTRFS_EXTENT_ITEM_KEY;
2376 key.offset = node->num_bytes;
2380 path->reada = READA_FORWARD;
2381 path->leave_spinning = 1;
2382 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2389 if (path->slots[0] > 0) {
2391 btrfs_item_key_to_cpu(path->nodes[0], &key,
2393 if (key.objectid == node->bytenr &&
2394 key.type == BTRFS_EXTENT_ITEM_KEY &&
2395 key.offset == node->num_bytes)
2399 btrfs_release_path(path);
2402 key.objectid = node->bytenr;
2403 key.offset = node->num_bytes;
2404 key.type = BTRFS_EXTENT_ITEM_KEY;
2413 leaf = path->nodes[0];
2414 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2415 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2416 if (item_size < sizeof(*ei)) {
2417 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2422 leaf = path->nodes[0];
2423 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2426 BUG_ON(item_size < sizeof(*ei));
2427 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2428 __run_delayed_extent_op(extent_op, leaf, ei);
2430 btrfs_mark_buffer_dirty(leaf);
2432 btrfs_free_path(path);
2436 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2437 struct btrfs_fs_info *fs_info,
2438 struct btrfs_delayed_ref_node *node,
2439 struct btrfs_delayed_extent_op *extent_op,
2440 int insert_reserved)
2443 struct btrfs_delayed_tree_ref *ref;
2444 struct btrfs_key ins;
2447 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2449 ref = btrfs_delayed_node_to_tree_ref(node);
2450 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2452 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2453 parent = ref->parent;
2454 ref_root = ref->root;
2456 ins.objectid = node->bytenr;
2457 if (skinny_metadata) {
2458 ins.offset = ref->level;
2459 ins.type = BTRFS_METADATA_ITEM_KEY;
2461 ins.offset = node->num_bytes;
2462 ins.type = BTRFS_EXTENT_ITEM_KEY;
2465 if (node->ref_mod != 1) {
2467 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2468 node->bytenr, node->ref_mod, node->action, ref_root,
2472 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2473 BUG_ON(!extent_op || !extent_op->update_flags);
2474 ret = alloc_reserved_tree_block(trans, fs_info,
2476 extent_op->flags_to_set,
2479 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2480 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2484 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2485 ret = __btrfs_free_extent(trans, fs_info, node,
2487 ref->level, 0, 1, extent_op);
2494 /* helper function to actually process a single delayed ref entry */
2495 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2496 struct btrfs_fs_info *fs_info,
2497 struct btrfs_delayed_ref_node *node,
2498 struct btrfs_delayed_extent_op *extent_op,
2499 int insert_reserved)
2503 if (trans->aborted) {
2504 if (insert_reserved)
2505 btrfs_pin_extent(fs_info, node->bytenr,
2506 node->num_bytes, 1);
2510 if (btrfs_delayed_ref_is_head(node)) {
2511 struct btrfs_delayed_ref_head *head;
2513 * we've hit the end of the chain and we were supposed
2514 * to insert this extent into the tree. But, it got
2515 * deleted before we ever needed to insert it, so all
2516 * we have to do is clean up the accounting
2519 head = btrfs_delayed_node_to_head(node);
2520 trace_run_delayed_ref_head(fs_info, node, head, node->action);
2522 if (head->total_ref_mod < 0) {
2523 struct btrfs_block_group_cache *cache;
2525 cache = btrfs_lookup_block_group(fs_info, node->bytenr);
2527 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2529 btrfs_put_block_group(cache);
2532 if (insert_reserved) {
2533 btrfs_pin_extent(fs_info, node->bytenr,
2534 node->num_bytes, 1);
2535 if (head->is_data) {
2536 ret = btrfs_del_csums(trans, fs_info,
2542 /* Also free its reserved qgroup space */
2543 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2544 head->qgroup_reserved);
2548 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2549 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2550 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2552 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2553 node->type == BTRFS_SHARED_DATA_REF_KEY)
2554 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2561 static inline struct btrfs_delayed_ref_node *
2562 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2564 struct btrfs_delayed_ref_node *ref;
2566 if (list_empty(&head->ref_list))
2570 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2571 * This is to prevent a ref count from going down to zero, which deletes
2572 * the extent item from the extent tree, when there still are references
2573 * to add, which would fail because they would not find the extent item.
2575 if (!list_empty(&head->ref_add_list))
2576 return list_first_entry(&head->ref_add_list,
2577 struct btrfs_delayed_ref_node, add_list);
2579 ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
2581 ASSERT(list_empty(&ref->add_list));
2586 * Returns 0 on success or if called with an already aborted transaction.
2587 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2589 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2590 struct btrfs_fs_info *fs_info,
2593 struct btrfs_delayed_ref_root *delayed_refs;
2594 struct btrfs_delayed_ref_node *ref;
2595 struct btrfs_delayed_ref_head *locked_ref = NULL;
2596 struct btrfs_delayed_extent_op *extent_op;
2597 ktime_t start = ktime_get();
2599 unsigned long count = 0;
2600 unsigned long actual_count = 0;
2601 int must_insert_reserved = 0;
2603 delayed_refs = &trans->transaction->delayed_refs;
2609 spin_lock(&delayed_refs->lock);
2610 locked_ref = btrfs_select_ref_head(trans);
2612 spin_unlock(&delayed_refs->lock);
2616 /* grab the lock that says we are going to process
2617 * all the refs for this head */
2618 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2619 spin_unlock(&delayed_refs->lock);
2621 * we may have dropped the spin lock to get the head
2622 * mutex lock, and that might have given someone else
2623 * time to free the head. If that's true, it has been
2624 * removed from our list and we can move on.
2626 if (ret == -EAGAIN) {
2634 * We need to try and merge add/drops of the same ref since we
2635 * can run into issues with relocate dropping the implicit ref
2636 * and then it being added back again before the drop can
2637 * finish. If we merged anything we need to re-loop so we can
2639 * Or we can get node references of the same type that weren't
2640 * merged when created due to bumps in the tree mod seq, and
2641 * we need to merge them to prevent adding an inline extent
2642 * backref before dropping it (triggering a BUG_ON at
2643 * insert_inline_extent_backref()).
2645 spin_lock(&locked_ref->lock);
2646 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2650 * locked_ref is the head node, so we have to go one
2651 * node back for any delayed ref updates
2653 ref = select_delayed_ref(locked_ref);
2655 if (ref && ref->seq &&
2656 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2657 spin_unlock(&locked_ref->lock);
2658 spin_lock(&delayed_refs->lock);
2659 locked_ref->processing = 0;
2660 delayed_refs->num_heads_ready++;
2661 spin_unlock(&delayed_refs->lock);
2662 btrfs_delayed_ref_unlock(locked_ref);
2670 * record the must insert reserved flag before we
2671 * drop the spin lock.
2673 must_insert_reserved = locked_ref->must_insert_reserved;
2674 locked_ref->must_insert_reserved = 0;
2676 extent_op = locked_ref->extent_op;
2677 locked_ref->extent_op = NULL;
2682 /* All delayed refs have been processed, Go ahead
2683 * and send the head node to run_one_delayed_ref,
2684 * so that any accounting fixes can happen
2686 ref = &locked_ref->node;
2688 if (extent_op && must_insert_reserved) {
2689 btrfs_free_delayed_extent_op(extent_op);
2694 spin_unlock(&locked_ref->lock);
2695 ret = run_delayed_extent_op(trans, fs_info,
2697 btrfs_free_delayed_extent_op(extent_op);
2701 * Need to reset must_insert_reserved if
2702 * there was an error so the abort stuff
2703 * can cleanup the reserved space
2706 if (must_insert_reserved)
2707 locked_ref->must_insert_reserved = 1;
2708 spin_lock(&delayed_refs->lock);
2709 locked_ref->processing = 0;
2710 delayed_refs->num_heads_ready++;
2711 spin_unlock(&delayed_refs->lock);
2712 btrfs_debug(fs_info,
2713 "run_delayed_extent_op returned %d",
2715 btrfs_delayed_ref_unlock(locked_ref);
2722 * Need to drop our head ref lock and re-acquire the
2723 * delayed ref lock and then re-check to make sure
2726 spin_unlock(&locked_ref->lock);
2727 spin_lock(&delayed_refs->lock);
2728 spin_lock(&locked_ref->lock);
2729 if (!list_empty(&locked_ref->ref_list) ||
2730 locked_ref->extent_op) {
2731 spin_unlock(&locked_ref->lock);
2732 spin_unlock(&delayed_refs->lock);
2736 delayed_refs->num_heads--;
2737 rb_erase(&locked_ref->href_node,
2738 &delayed_refs->href_root);
2739 spin_unlock(&delayed_refs->lock);
2743 list_del(&ref->list);
2744 if (!list_empty(&ref->add_list))
2745 list_del(&ref->add_list);
2747 atomic_dec(&delayed_refs->num_entries);
2749 if (!btrfs_delayed_ref_is_head(ref)) {
2751 * when we play the delayed ref, also correct the
2754 switch (ref->action) {
2755 case BTRFS_ADD_DELAYED_REF:
2756 case BTRFS_ADD_DELAYED_EXTENT:
2757 locked_ref->node.ref_mod -= ref->ref_mod;
2759 case BTRFS_DROP_DELAYED_REF:
2760 locked_ref->node.ref_mod += ref->ref_mod;
2766 spin_unlock(&locked_ref->lock);
2768 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2769 must_insert_reserved);
2771 btrfs_free_delayed_extent_op(extent_op);
2773 spin_lock(&delayed_refs->lock);
2774 locked_ref->processing = 0;
2775 delayed_refs->num_heads_ready++;
2776 spin_unlock(&delayed_refs->lock);
2777 btrfs_delayed_ref_unlock(locked_ref);
2778 btrfs_put_delayed_ref(ref);
2779 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2785 * If this node is a head, that means all the refs in this head
2786 * have been dealt with, and we will pick the next head to deal
2787 * with, so we must unlock the head and drop it from the cluster
2788 * list before we release it.
2790 if (btrfs_delayed_ref_is_head(ref)) {
2791 if (locked_ref->is_data &&
2792 locked_ref->total_ref_mod < 0) {
2793 spin_lock(&delayed_refs->lock);
2794 delayed_refs->pending_csums -= ref->num_bytes;
2795 spin_unlock(&delayed_refs->lock);
2797 btrfs_delayed_ref_unlock(locked_ref);
2800 btrfs_put_delayed_ref(ref);
2806 * We don't want to include ref heads since we can have empty ref heads
2807 * and those will drastically skew our runtime down since we just do
2808 * accounting, no actual extent tree updates.
2810 if (actual_count > 0) {
2811 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2815 * We weigh the current average higher than our current runtime
2816 * to avoid large swings in the average.
2818 spin_lock(&delayed_refs->lock);
2819 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2820 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2821 spin_unlock(&delayed_refs->lock);
2826 #ifdef SCRAMBLE_DELAYED_REFS
2828 * Normally delayed refs get processed in ascending bytenr order. This
2829 * correlates in most cases to the order added. To expose dependencies on this
2830 * order, we start to process the tree in the middle instead of the beginning
2832 static u64 find_middle(struct rb_root *root)
2834 struct rb_node *n = root->rb_node;
2835 struct btrfs_delayed_ref_node *entry;
2838 u64 first = 0, last = 0;
2842 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2843 first = entry->bytenr;
2847 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2848 last = entry->bytenr;
2853 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2854 WARN_ON(!entry->in_tree);
2856 middle = entry->bytenr;
2869 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2873 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2874 sizeof(struct btrfs_extent_inline_ref));
2875 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2876 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2879 * We don't ever fill up leaves all the way so multiply by 2 just to be
2880 * closer to what we're really going to want to use.
2882 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2886 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2887 * would require to store the csums for that many bytes.
2889 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2892 u64 num_csums_per_leaf;
2895 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2896 num_csums_per_leaf = div64_u64(csum_size,
2897 (u64)btrfs_super_csum_size(fs_info->super_copy));
2898 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2899 num_csums += num_csums_per_leaf - 1;
2900 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2904 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2905 struct btrfs_fs_info *fs_info)
2907 struct btrfs_block_rsv *global_rsv;
2908 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2909 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2910 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2911 u64 num_bytes, num_dirty_bgs_bytes;
2914 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2915 num_heads = heads_to_leaves(fs_info, num_heads);
2917 num_bytes += (num_heads - 1) * fs_info->nodesize;
2919 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2921 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2923 global_rsv = &fs_info->global_block_rsv;
2926 * If we can't allocate any more chunks lets make sure we have _lots_ of
2927 * wiggle room since running delayed refs can create more delayed refs.
2929 if (global_rsv->space_info->full) {
2930 num_dirty_bgs_bytes <<= 1;
2934 spin_lock(&global_rsv->lock);
2935 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2937 spin_unlock(&global_rsv->lock);
2941 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2942 struct btrfs_fs_info *fs_info)
2945 atomic_read(&trans->transaction->delayed_refs.num_entries);
2950 avg_runtime = fs_info->avg_delayed_ref_runtime;
2951 val = num_entries * avg_runtime;
2952 if (val >= NSEC_PER_SEC)
2954 if (val >= NSEC_PER_SEC / 2)
2957 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2960 struct async_delayed_refs {
2961 struct btrfs_root *root;
2966 struct completion wait;
2967 struct btrfs_work work;
2970 static inline struct async_delayed_refs *
2971 to_async_delayed_refs(struct btrfs_work *work)
2973 return container_of(work, struct async_delayed_refs, work);
2976 static void delayed_ref_async_start(struct btrfs_work *work)
2978 struct async_delayed_refs *async = to_async_delayed_refs(work);
2979 struct btrfs_trans_handle *trans;
2980 struct btrfs_fs_info *fs_info = async->root->fs_info;
2983 /* if the commit is already started, we don't need to wait here */
2984 if (btrfs_transaction_blocked(fs_info))
2987 trans = btrfs_join_transaction(async->root);
2988 if (IS_ERR(trans)) {
2989 async->error = PTR_ERR(trans);
2994 * trans->sync means that when we call end_transaction, we won't
2995 * wait on delayed refs
2999 /* Don't bother flushing if we got into a different transaction */
3000 if (trans->transid > async->transid)
3003 ret = btrfs_run_delayed_refs(trans, fs_info, async->count);
3007 ret = btrfs_end_transaction(trans);
3008 if (ret && !async->error)
3012 complete(&async->wait);
3017 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
3018 unsigned long count, u64 transid, int wait)
3020 struct async_delayed_refs *async;
3023 async = kmalloc(sizeof(*async), GFP_NOFS);
3027 async->root = fs_info->tree_root;
3028 async->count = count;
3030 async->transid = transid;
3035 init_completion(&async->wait);
3037 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3038 delayed_ref_async_start, NULL, NULL);
3040 btrfs_queue_work(fs_info->extent_workers, &async->work);
3043 wait_for_completion(&async->wait);
3052 * this starts processing the delayed reference count updates and
3053 * extent insertions we have queued up so far. count can be
3054 * 0, which means to process everything in the tree at the start
3055 * of the run (but not newly added entries), or it can be some target
3056 * number you'd like to process.
3058 * Returns 0 on success or if called with an aborted transaction
3059 * Returns <0 on error and aborts the transaction
3061 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3062 struct btrfs_fs_info *fs_info, unsigned long count)
3064 struct rb_node *node;
3065 struct btrfs_delayed_ref_root *delayed_refs;
3066 struct btrfs_delayed_ref_head *head;
3068 int run_all = count == (unsigned long)-1;
3069 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3071 /* We'll clean this up in btrfs_cleanup_transaction */
3075 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3078 delayed_refs = &trans->transaction->delayed_refs;
3080 count = atomic_read(&delayed_refs->num_entries) * 2;
3083 #ifdef SCRAMBLE_DELAYED_REFS
3084 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3086 trans->can_flush_pending_bgs = false;
3087 ret = __btrfs_run_delayed_refs(trans, fs_info, count);
3089 btrfs_abort_transaction(trans, ret);
3094 if (!list_empty(&trans->new_bgs))
3095 btrfs_create_pending_block_groups(trans, fs_info);
3097 spin_lock(&delayed_refs->lock);
3098 node = rb_first(&delayed_refs->href_root);
3100 spin_unlock(&delayed_refs->lock);
3105 head = rb_entry(node, struct btrfs_delayed_ref_head,
3107 if (btrfs_delayed_ref_is_head(&head->node)) {
3108 struct btrfs_delayed_ref_node *ref;
3111 refcount_inc(&ref->refs);
3113 spin_unlock(&delayed_refs->lock);
3115 * Mutex was contended, block until it's
3116 * released and try again
3118 mutex_lock(&head->mutex);
3119 mutex_unlock(&head->mutex);
3121 btrfs_put_delayed_ref(ref);
3127 node = rb_next(node);
3129 spin_unlock(&delayed_refs->lock);
3134 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3138 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3139 struct btrfs_fs_info *fs_info,
3140 u64 bytenr, u64 num_bytes, u64 flags,
3141 int level, int is_data)
3143 struct btrfs_delayed_extent_op *extent_op;
3146 extent_op = btrfs_alloc_delayed_extent_op();
3150 extent_op->flags_to_set = flags;
3151 extent_op->update_flags = true;
3152 extent_op->update_key = false;
3153 extent_op->is_data = is_data ? true : false;
3154 extent_op->level = level;
3156 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3157 num_bytes, extent_op);
3159 btrfs_free_delayed_extent_op(extent_op);
3163 static noinline int check_delayed_ref(struct btrfs_root *root,
3164 struct btrfs_path *path,
3165 u64 objectid, u64 offset, u64 bytenr)
3167 struct btrfs_delayed_ref_head *head;
3168 struct btrfs_delayed_ref_node *ref;
3169 struct btrfs_delayed_data_ref *data_ref;
3170 struct btrfs_delayed_ref_root *delayed_refs;
3171 struct btrfs_transaction *cur_trans;
3174 spin_lock(&root->fs_info->trans_lock);
3175 cur_trans = root->fs_info->running_transaction;
3177 refcount_inc(&cur_trans->use_count);
3178 spin_unlock(&root->fs_info->trans_lock);
3182 delayed_refs = &cur_trans->delayed_refs;
3183 spin_lock(&delayed_refs->lock);
3184 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3186 spin_unlock(&delayed_refs->lock);
3187 btrfs_put_transaction(cur_trans);
3191 if (!mutex_trylock(&head->mutex)) {
3192 refcount_inc(&head->node.refs);
3193 spin_unlock(&delayed_refs->lock);
3195 btrfs_release_path(path);
3198 * Mutex was contended, block until it's released and let
3201 mutex_lock(&head->mutex);
3202 mutex_unlock(&head->mutex);
3203 btrfs_put_delayed_ref(&head->node);
3204 btrfs_put_transaction(cur_trans);
3207 spin_unlock(&delayed_refs->lock);
3209 spin_lock(&head->lock);
3210 list_for_each_entry(ref, &head->ref_list, list) {
3211 /* If it's a shared ref we know a cross reference exists */
3212 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3217 data_ref = btrfs_delayed_node_to_data_ref(ref);
3220 * If our ref doesn't match the one we're currently looking at
3221 * then we have a cross reference.
3223 if (data_ref->root != root->root_key.objectid ||
3224 data_ref->objectid != objectid ||
3225 data_ref->offset != offset) {
3230 spin_unlock(&head->lock);
3231 mutex_unlock(&head->mutex);
3232 btrfs_put_transaction(cur_trans);
3236 static noinline int check_committed_ref(struct btrfs_root *root,
3237 struct btrfs_path *path,
3238 u64 objectid, u64 offset, u64 bytenr)
3240 struct btrfs_fs_info *fs_info = root->fs_info;
3241 struct btrfs_root *extent_root = fs_info->extent_root;
3242 struct extent_buffer *leaf;
3243 struct btrfs_extent_data_ref *ref;
3244 struct btrfs_extent_inline_ref *iref;
3245 struct btrfs_extent_item *ei;
3246 struct btrfs_key key;
3251 key.objectid = bytenr;
3252 key.offset = (u64)-1;
3253 key.type = BTRFS_EXTENT_ITEM_KEY;
3255 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3258 BUG_ON(ret == 0); /* Corruption */
3261 if (path->slots[0] == 0)
3265 leaf = path->nodes[0];
3266 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3268 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3272 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3273 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3274 if (item_size < sizeof(*ei)) {
3275 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3279 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3281 if (item_size != sizeof(*ei) +
3282 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3285 if (btrfs_extent_generation(leaf, ei) <=
3286 btrfs_root_last_snapshot(&root->root_item))
3289 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3291 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3292 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3295 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3296 if (btrfs_extent_refs(leaf, ei) !=
3297 btrfs_extent_data_ref_count(leaf, ref) ||
3298 btrfs_extent_data_ref_root(leaf, ref) !=
3299 root->root_key.objectid ||
3300 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3301 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3309 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3312 struct btrfs_path *path;
3316 path = btrfs_alloc_path();
3321 ret = check_committed_ref(root, path, objectid,
3323 if (ret && ret != -ENOENT)
3326 ret2 = check_delayed_ref(root, path, objectid,
3328 } while (ret2 == -EAGAIN);
3330 if (ret2 && ret2 != -ENOENT) {
3335 if (ret != -ENOENT || ret2 != -ENOENT)
3338 btrfs_free_path(path);
3339 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3344 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3345 struct btrfs_root *root,
3346 struct extent_buffer *buf,
3347 int full_backref, int inc)
3349 struct btrfs_fs_info *fs_info = root->fs_info;
3355 struct btrfs_key key;
3356 struct btrfs_file_extent_item *fi;
3360 int (*process_func)(struct btrfs_trans_handle *,
3361 struct btrfs_fs_info *,
3362 u64, u64, u64, u64, u64, u64);
3365 if (btrfs_is_testing(fs_info))
3368 ref_root = btrfs_header_owner(buf);
3369 nritems = btrfs_header_nritems(buf);
3370 level = btrfs_header_level(buf);
3372 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3376 process_func = btrfs_inc_extent_ref;
3378 process_func = btrfs_free_extent;
3381 parent = buf->start;
3385 for (i = 0; i < nritems; i++) {
3387 btrfs_item_key_to_cpu(buf, &key, i);
3388 if (key.type != BTRFS_EXTENT_DATA_KEY)
3390 fi = btrfs_item_ptr(buf, i,
3391 struct btrfs_file_extent_item);
3392 if (btrfs_file_extent_type(buf, fi) ==
3393 BTRFS_FILE_EXTENT_INLINE)
3395 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3399 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3400 key.offset -= btrfs_file_extent_offset(buf, fi);
3401 ret = process_func(trans, fs_info, bytenr, num_bytes,
3402 parent, ref_root, key.objectid,
3407 bytenr = btrfs_node_blockptr(buf, i);
3408 num_bytes = fs_info->nodesize;
3409 ret = process_func(trans, fs_info, bytenr, num_bytes,
3410 parent, ref_root, level - 1, 0);
3420 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3421 struct extent_buffer *buf, int full_backref)
3423 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3426 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3427 struct extent_buffer *buf, int full_backref)
3429 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3432 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3433 struct btrfs_fs_info *fs_info,
3434 struct btrfs_path *path,
3435 struct btrfs_block_group_cache *cache)
3438 struct btrfs_root *extent_root = fs_info->extent_root;
3440 struct extent_buffer *leaf;
3442 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3449 leaf = path->nodes[0];
3450 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3451 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3452 btrfs_mark_buffer_dirty(leaf);
3454 btrfs_release_path(path);
3459 static struct btrfs_block_group_cache *
3460 next_block_group(struct btrfs_fs_info *fs_info,
3461 struct btrfs_block_group_cache *cache)
3463 struct rb_node *node;
3465 spin_lock(&fs_info->block_group_cache_lock);
3467 /* If our block group was removed, we need a full search. */
3468 if (RB_EMPTY_NODE(&cache->cache_node)) {
3469 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3471 spin_unlock(&fs_info->block_group_cache_lock);
3472 btrfs_put_block_group(cache);
3473 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3475 node = rb_next(&cache->cache_node);
3476 btrfs_put_block_group(cache);
3478 cache = rb_entry(node, struct btrfs_block_group_cache,
3480 btrfs_get_block_group(cache);
3483 spin_unlock(&fs_info->block_group_cache_lock);
3487 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3488 struct btrfs_trans_handle *trans,
3489 struct btrfs_path *path)
3491 struct btrfs_fs_info *fs_info = block_group->fs_info;
3492 struct btrfs_root *root = fs_info->tree_root;
3493 struct inode *inode = NULL;
3494 struct extent_changeset *data_reserved = NULL;
3496 int dcs = BTRFS_DC_ERROR;
3502 * If this block group is smaller than 100 megs don't bother caching the
3505 if (block_group->key.offset < (100 * SZ_1M)) {
3506 spin_lock(&block_group->lock);
3507 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3508 spin_unlock(&block_group->lock);
3515 inode = lookup_free_space_inode(fs_info, block_group, path);
3516 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3517 ret = PTR_ERR(inode);
3518 btrfs_release_path(path);
3522 if (IS_ERR(inode)) {
3526 if (block_group->ro)
3529 ret = create_free_space_inode(fs_info, trans, block_group,
3537 * We want to set the generation to 0, that way if anything goes wrong
3538 * from here on out we know not to trust this cache when we load up next
3541 BTRFS_I(inode)->generation = 0;
3542 ret = btrfs_update_inode(trans, root, inode);
3545 * So theoretically we could recover from this, simply set the
3546 * super cache generation to 0 so we know to invalidate the
3547 * cache, but then we'd have to keep track of the block groups
3548 * that fail this way so we know we _have_ to reset this cache
3549 * before the next commit or risk reading stale cache. So to
3550 * limit our exposure to horrible edge cases lets just abort the
3551 * transaction, this only happens in really bad situations
3554 btrfs_abort_transaction(trans, ret);
3559 /* We've already setup this transaction, go ahead and exit */
3560 if (block_group->cache_generation == trans->transid &&
3561 i_size_read(inode)) {
3562 dcs = BTRFS_DC_SETUP;
3566 if (i_size_read(inode) > 0) {
3567 ret = btrfs_check_trunc_cache_free_space(fs_info,
3568 &fs_info->global_block_rsv);
3572 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3577 spin_lock(&block_group->lock);
3578 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3579 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3581 * don't bother trying to write stuff out _if_
3582 * a) we're not cached,
3583 * b) we're with nospace_cache mount option,
3584 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3586 dcs = BTRFS_DC_WRITTEN;
3587 spin_unlock(&block_group->lock);
3590 spin_unlock(&block_group->lock);
3593 * We hit an ENOSPC when setting up the cache in this transaction, just
3594 * skip doing the setup, we've already cleared the cache so we're safe.
3596 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3602 * Try to preallocate enough space based on how big the block group is.
3603 * Keep in mind this has to include any pinned space which could end up
3604 * taking up quite a bit since it's not folded into the other space
3607 num_pages = div_u64(block_group->key.offset, SZ_256M);
3612 num_pages *= PAGE_SIZE;
3614 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3618 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3619 num_pages, num_pages,
3622 * Our cache requires contiguous chunks so that we don't modify a bunch
3623 * of metadata or split extents when writing the cache out, which means
3624 * we can enospc if we are heavily fragmented in addition to just normal
3625 * out of space conditions. So if we hit this just skip setting up any
3626 * other block groups for this transaction, maybe we'll unpin enough
3627 * space the next time around.
3630 dcs = BTRFS_DC_SETUP;
3631 else if (ret == -ENOSPC)
3632 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3637 btrfs_release_path(path);
3639 spin_lock(&block_group->lock);
3640 if (!ret && dcs == BTRFS_DC_SETUP)
3641 block_group->cache_generation = trans->transid;
3642 block_group->disk_cache_state = dcs;
3643 spin_unlock(&block_group->lock);
3645 extent_changeset_free(data_reserved);
3649 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3650 struct btrfs_fs_info *fs_info)
3652 struct btrfs_block_group_cache *cache, *tmp;
3653 struct btrfs_transaction *cur_trans = trans->transaction;
3654 struct btrfs_path *path;
3656 if (list_empty(&cur_trans->dirty_bgs) ||
3657 !btrfs_test_opt(fs_info, SPACE_CACHE))
3660 path = btrfs_alloc_path();
3664 /* Could add new block groups, use _safe just in case */
3665 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3667 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3668 cache_save_setup(cache, trans, path);
3671 btrfs_free_path(path);
3676 * transaction commit does final block group cache writeback during a
3677 * critical section where nothing is allowed to change the FS. This is
3678 * required in order for the cache to actually match the block group,
3679 * but can introduce a lot of latency into the commit.
3681 * So, btrfs_start_dirty_block_groups is here to kick off block group
3682 * cache IO. There's a chance we'll have to redo some of it if the
3683 * block group changes again during the commit, but it greatly reduces
3684 * the commit latency by getting rid of the easy block groups while
3685 * we're still allowing others to join the commit.
3687 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3688 struct btrfs_fs_info *fs_info)
3690 struct btrfs_block_group_cache *cache;
3691 struct btrfs_transaction *cur_trans = trans->transaction;
3694 struct btrfs_path *path = NULL;
3696 struct list_head *io = &cur_trans->io_bgs;
3697 int num_started = 0;
3700 spin_lock(&cur_trans->dirty_bgs_lock);
3701 if (list_empty(&cur_trans->dirty_bgs)) {
3702 spin_unlock(&cur_trans->dirty_bgs_lock);
3705 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3706 spin_unlock(&cur_trans->dirty_bgs_lock);
3710 * make sure all the block groups on our dirty list actually
3713 btrfs_create_pending_block_groups(trans, fs_info);
3716 path = btrfs_alloc_path();
3722 * cache_write_mutex is here only to save us from balance or automatic
3723 * removal of empty block groups deleting this block group while we are
3724 * writing out the cache
3726 mutex_lock(&trans->transaction->cache_write_mutex);
3727 while (!list_empty(&dirty)) {
3728 cache = list_first_entry(&dirty,
3729 struct btrfs_block_group_cache,
3732 * this can happen if something re-dirties a block
3733 * group that is already under IO. Just wait for it to
3734 * finish and then do it all again
3736 if (!list_empty(&cache->io_list)) {
3737 list_del_init(&cache->io_list);
3738 btrfs_wait_cache_io(trans, cache, path);
3739 btrfs_put_block_group(cache);
3744 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3745 * if it should update the cache_state. Don't delete
3746 * until after we wait.
3748 * Since we're not running in the commit critical section
3749 * we need the dirty_bgs_lock to protect from update_block_group
3751 spin_lock(&cur_trans->dirty_bgs_lock);
3752 list_del_init(&cache->dirty_list);
3753 spin_unlock(&cur_trans->dirty_bgs_lock);
3757 cache_save_setup(cache, trans, path);
3759 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3760 cache->io_ctl.inode = NULL;
3761 ret = btrfs_write_out_cache(fs_info, trans,
3763 if (ret == 0 && cache->io_ctl.inode) {
3768 * the cache_write_mutex is protecting
3771 list_add_tail(&cache->io_list, io);
3774 * if we failed to write the cache, the
3775 * generation will be bad and life goes on
3781 ret = write_one_cache_group(trans, fs_info,
3784 * Our block group might still be attached to the list
3785 * of new block groups in the transaction handle of some
3786 * other task (struct btrfs_trans_handle->new_bgs). This
3787 * means its block group item isn't yet in the extent
3788 * tree. If this happens ignore the error, as we will
3789 * try again later in the critical section of the
3790 * transaction commit.
3792 if (ret == -ENOENT) {
3794 spin_lock(&cur_trans->dirty_bgs_lock);
3795 if (list_empty(&cache->dirty_list)) {
3796 list_add_tail(&cache->dirty_list,
3797 &cur_trans->dirty_bgs);
3798 btrfs_get_block_group(cache);
3800 spin_unlock(&cur_trans->dirty_bgs_lock);
3802 btrfs_abort_transaction(trans, ret);
3806 /* if its not on the io list, we need to put the block group */
3808 btrfs_put_block_group(cache);
3814 * Avoid blocking other tasks for too long. It might even save
3815 * us from writing caches for block groups that are going to be
3818 mutex_unlock(&trans->transaction->cache_write_mutex);
3819 mutex_lock(&trans->transaction->cache_write_mutex);
3821 mutex_unlock(&trans->transaction->cache_write_mutex);
3824 * go through delayed refs for all the stuff we've just kicked off
3825 * and then loop back (just once)
3827 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
3828 if (!ret && loops == 0) {
3830 spin_lock(&cur_trans->dirty_bgs_lock);
3831 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3833 * dirty_bgs_lock protects us from concurrent block group
3834 * deletes too (not just cache_write_mutex).
3836 if (!list_empty(&dirty)) {
3837 spin_unlock(&cur_trans->dirty_bgs_lock);
3840 spin_unlock(&cur_trans->dirty_bgs_lock);
3841 } else if (ret < 0) {
3842 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3845 btrfs_free_path(path);
3849 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3850 struct btrfs_fs_info *fs_info)
3852 struct btrfs_block_group_cache *cache;
3853 struct btrfs_transaction *cur_trans = trans->transaction;
3856 struct btrfs_path *path;
3857 struct list_head *io = &cur_trans->io_bgs;
3858 int num_started = 0;
3860 path = btrfs_alloc_path();
3865 * Even though we are in the critical section of the transaction commit,
3866 * we can still have concurrent tasks adding elements to this
3867 * transaction's list of dirty block groups. These tasks correspond to
3868 * endio free space workers started when writeback finishes for a
3869 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3870 * allocate new block groups as a result of COWing nodes of the root
3871 * tree when updating the free space inode. The writeback for the space
3872 * caches is triggered by an earlier call to
3873 * btrfs_start_dirty_block_groups() and iterations of the following
3875 * Also we want to do the cache_save_setup first and then run the
3876 * delayed refs to make sure we have the best chance at doing this all
3879 spin_lock(&cur_trans->dirty_bgs_lock);
3880 while (!list_empty(&cur_trans->dirty_bgs)) {
3881 cache = list_first_entry(&cur_trans->dirty_bgs,
3882 struct btrfs_block_group_cache,
3886 * this can happen if cache_save_setup re-dirties a block
3887 * group that is already under IO. Just wait for it to
3888 * finish and then do it all again
3890 if (!list_empty(&cache->io_list)) {
3891 spin_unlock(&cur_trans->dirty_bgs_lock);
3892 list_del_init(&cache->io_list);
3893 btrfs_wait_cache_io(trans, cache, path);
3894 btrfs_put_block_group(cache);
3895 spin_lock(&cur_trans->dirty_bgs_lock);
3899 * don't remove from the dirty list until after we've waited
3902 list_del_init(&cache->dirty_list);
3903 spin_unlock(&cur_trans->dirty_bgs_lock);
3906 cache_save_setup(cache, trans, path);
3909 ret = btrfs_run_delayed_refs(trans, fs_info,
3910 (unsigned long) -1);
3912 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3913 cache->io_ctl.inode = NULL;
3914 ret = btrfs_write_out_cache(fs_info, trans,
3916 if (ret == 0 && cache->io_ctl.inode) {
3919 list_add_tail(&cache->io_list, io);
3922 * if we failed to write the cache, the
3923 * generation will be bad and life goes on
3929 ret = write_one_cache_group(trans, fs_info,
3932 * One of the free space endio workers might have
3933 * created a new block group while updating a free space
3934 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3935 * and hasn't released its transaction handle yet, in
3936 * which case the new block group is still attached to
3937 * its transaction handle and its creation has not
3938 * finished yet (no block group item in the extent tree
3939 * yet, etc). If this is the case, wait for all free
3940 * space endio workers to finish and retry. This is a
3941 * a very rare case so no need for a more efficient and
3944 if (ret == -ENOENT) {
3945 wait_event(cur_trans->writer_wait,
3946 atomic_read(&cur_trans->num_writers) == 1);
3947 ret = write_one_cache_group(trans, fs_info,
3951 btrfs_abort_transaction(trans, ret);
3954 /* if its not on the io list, we need to put the block group */
3956 btrfs_put_block_group(cache);
3957 spin_lock(&cur_trans->dirty_bgs_lock);
3959 spin_unlock(&cur_trans->dirty_bgs_lock);
3961 while (!list_empty(io)) {
3962 cache = list_first_entry(io, struct btrfs_block_group_cache,
3964 list_del_init(&cache->io_list);
3965 btrfs_wait_cache_io(trans, cache, path);
3966 btrfs_put_block_group(cache);
3969 btrfs_free_path(path);
3973 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3975 struct btrfs_block_group_cache *block_group;
3978 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3979 if (!block_group || block_group->ro)
3982 btrfs_put_block_group(block_group);
3986 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3988 struct btrfs_block_group_cache *bg;
3991 bg = btrfs_lookup_block_group(fs_info, bytenr);
3995 spin_lock(&bg->lock);
3999 atomic_inc(&bg->nocow_writers);
4000 spin_unlock(&bg->lock);
4002 /* no put on block group, done by btrfs_dec_nocow_writers */
4004 btrfs_put_block_group(bg);
4010 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
4012 struct btrfs_block_group_cache *bg;
4014 bg = btrfs_lookup_block_group(fs_info, bytenr);
4016 if (atomic_dec_and_test(&bg->nocow_writers))
4017 wake_up_atomic_t(&bg->nocow_writers);
4019 * Once for our lookup and once for the lookup done by a previous call
4020 * to btrfs_inc_nocow_writers()
4022 btrfs_put_block_group(bg);
4023 btrfs_put_block_group(bg);
4026 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
4032 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4034 wait_on_atomic_t(&bg->nocow_writers,
4035 btrfs_wait_nocow_writers_atomic_t,
4036 TASK_UNINTERRUPTIBLE);
4039 static const char *alloc_name(u64 flags)
4042 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4044 case BTRFS_BLOCK_GROUP_METADATA:
4046 case BTRFS_BLOCK_GROUP_DATA:
4048 case BTRFS_BLOCK_GROUP_SYSTEM:
4052 return "invalid-combination";
4056 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4057 struct btrfs_space_info **new)
4060 struct btrfs_space_info *space_info;
4064 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4068 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4075 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4076 INIT_LIST_HEAD(&space_info->block_groups[i]);
4077 init_rwsem(&space_info->groups_sem);
4078 spin_lock_init(&space_info->lock);
4079 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4080 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4081 init_waitqueue_head(&space_info->wait);
4082 INIT_LIST_HEAD(&space_info->ro_bgs);
4083 INIT_LIST_HEAD(&space_info->tickets);
4084 INIT_LIST_HEAD(&space_info->priority_tickets);
4086 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4087 info->space_info_kobj, "%s",
4088 alloc_name(space_info->flags));
4090 percpu_counter_destroy(&space_info->total_bytes_pinned);
4096 list_add_rcu(&space_info->list, &info->space_info);
4097 if (flags & BTRFS_BLOCK_GROUP_DATA)
4098 info->data_sinfo = space_info;
4103 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4104 u64 total_bytes, u64 bytes_used,
4106 struct btrfs_space_info **space_info)
4108 struct btrfs_space_info *found;
4111 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4112 BTRFS_BLOCK_GROUP_RAID10))
4117 found = __find_space_info(info, flags);
4119 spin_lock(&found->lock);
4120 found->total_bytes += total_bytes;
4121 found->disk_total += total_bytes * factor;
4122 found->bytes_used += bytes_used;
4123 found->disk_used += bytes_used * factor;
4124 found->bytes_readonly += bytes_readonly;
4125 if (total_bytes > 0)
4127 space_info_add_new_bytes(info, found, total_bytes -
4128 bytes_used - bytes_readonly);
4129 spin_unlock(&found->lock);
4130 *space_info = found;
4133 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4135 u64 extra_flags = chunk_to_extended(flags) &
4136 BTRFS_EXTENDED_PROFILE_MASK;
4138 write_seqlock(&fs_info->profiles_lock);
4139 if (flags & BTRFS_BLOCK_GROUP_DATA)
4140 fs_info->avail_data_alloc_bits |= extra_flags;
4141 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4142 fs_info->avail_metadata_alloc_bits |= extra_flags;
4143 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4144 fs_info->avail_system_alloc_bits |= extra_flags;
4145 write_sequnlock(&fs_info->profiles_lock);
4149 * returns target flags in extended format or 0 if restripe for this
4150 * chunk_type is not in progress
4152 * should be called with either volume_mutex or balance_lock held
4154 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4156 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4162 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4163 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4164 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4165 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4166 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4167 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4168 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4169 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4170 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4177 * @flags: available profiles in extended format (see ctree.h)
4179 * Returns reduced profile in chunk format. If profile changing is in
4180 * progress (either running or paused) picks the target profile (if it's
4181 * already available), otherwise falls back to plain reducing.
4183 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4185 u64 num_devices = fs_info->fs_devices->rw_devices;
4191 * see if restripe for this chunk_type is in progress, if so
4192 * try to reduce to the target profile
4194 spin_lock(&fs_info->balance_lock);
4195 target = get_restripe_target(fs_info, flags);
4197 /* pick target profile only if it's already available */
4198 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4199 spin_unlock(&fs_info->balance_lock);
4200 return extended_to_chunk(target);
4203 spin_unlock(&fs_info->balance_lock);
4205 /* First, mask out the RAID levels which aren't possible */
4206 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4207 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4208 allowed |= btrfs_raid_group[raid_type];
4212 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4213 allowed = BTRFS_BLOCK_GROUP_RAID6;
4214 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4215 allowed = BTRFS_BLOCK_GROUP_RAID5;
4216 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4217 allowed = BTRFS_BLOCK_GROUP_RAID10;
4218 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4219 allowed = BTRFS_BLOCK_GROUP_RAID1;
4220 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4221 allowed = BTRFS_BLOCK_GROUP_RAID0;
4223 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4225 return extended_to_chunk(flags | allowed);
4228 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4235 seq = read_seqbegin(&fs_info->profiles_lock);
4237 if (flags & BTRFS_BLOCK_GROUP_DATA)
4238 flags |= fs_info->avail_data_alloc_bits;
4239 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4240 flags |= fs_info->avail_system_alloc_bits;
4241 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4242 flags |= fs_info->avail_metadata_alloc_bits;
4243 } while (read_seqretry(&fs_info->profiles_lock, seq));
4245 return btrfs_reduce_alloc_profile(fs_info, flags);
4248 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4250 struct btrfs_fs_info *fs_info = root->fs_info;
4255 flags = BTRFS_BLOCK_GROUP_DATA;
4256 else if (root == fs_info->chunk_root)
4257 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4259 flags = BTRFS_BLOCK_GROUP_METADATA;
4261 ret = get_alloc_profile(fs_info, flags);
4265 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4267 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4270 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4272 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4275 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4277 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4280 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4281 bool may_use_included)
4284 return s_info->bytes_used + s_info->bytes_reserved +
4285 s_info->bytes_pinned + s_info->bytes_readonly +
4286 (may_use_included ? s_info->bytes_may_use : 0);
4289 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4291 struct btrfs_root *root = inode->root;
4292 struct btrfs_fs_info *fs_info = root->fs_info;
4293 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4296 int need_commit = 2;
4297 int have_pinned_space;
4299 /* make sure bytes are sectorsize aligned */
4300 bytes = ALIGN(bytes, fs_info->sectorsize);
4302 if (btrfs_is_free_space_inode(inode)) {
4304 ASSERT(current->journal_info);
4308 /* make sure we have enough space to handle the data first */
4309 spin_lock(&data_sinfo->lock);
4310 used = btrfs_space_info_used(data_sinfo, true);
4312 if (used + bytes > data_sinfo->total_bytes) {
4313 struct btrfs_trans_handle *trans;
4316 * if we don't have enough free bytes in this space then we need
4317 * to alloc a new chunk.
4319 if (!data_sinfo->full) {
4322 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4323 spin_unlock(&data_sinfo->lock);
4325 alloc_target = btrfs_data_alloc_profile(fs_info);
4327 * It is ugly that we don't call nolock join
4328 * transaction for the free space inode case here.
4329 * But it is safe because we only do the data space
4330 * reservation for the free space cache in the
4331 * transaction context, the common join transaction
4332 * just increase the counter of the current transaction
4333 * handler, doesn't try to acquire the trans_lock of
4336 trans = btrfs_join_transaction(root);
4338 return PTR_ERR(trans);
4340 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4341 CHUNK_ALLOC_NO_FORCE);
4342 btrfs_end_transaction(trans);
4347 have_pinned_space = 1;
4356 * If we don't have enough pinned space to deal with this
4357 * allocation, and no removed chunk in current transaction,
4358 * don't bother committing the transaction.
4360 have_pinned_space = percpu_counter_compare(
4361 &data_sinfo->total_bytes_pinned,
4362 used + bytes - data_sinfo->total_bytes);
4363 spin_unlock(&data_sinfo->lock);
4365 /* commit the current transaction and try again */
4368 !atomic_read(&fs_info->open_ioctl_trans)) {
4371 if (need_commit > 0) {
4372 btrfs_start_delalloc_roots(fs_info, 0, -1);
4373 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4377 trans = btrfs_join_transaction(root);
4379 return PTR_ERR(trans);
4380 if (have_pinned_space >= 0 ||
4381 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4382 &trans->transaction->flags) ||
4384 ret = btrfs_commit_transaction(trans);
4388 * The cleaner kthread might still be doing iput
4389 * operations. Wait for it to finish so that
4390 * more space is released.
4392 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4393 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4396 btrfs_end_transaction(trans);
4400 trace_btrfs_space_reservation(fs_info,
4401 "space_info:enospc",
4402 data_sinfo->flags, bytes, 1);
4405 data_sinfo->bytes_may_use += bytes;
4406 trace_btrfs_space_reservation(fs_info, "space_info",
4407 data_sinfo->flags, bytes, 1);
4408 spin_unlock(&data_sinfo->lock);
4413 int btrfs_check_data_free_space(struct inode *inode,
4414 struct extent_changeset **reserved, u64 start, u64 len)
4416 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4419 /* align the range */
4420 len = round_up(start + len, fs_info->sectorsize) -
4421 round_down(start, fs_info->sectorsize);
4422 start = round_down(start, fs_info->sectorsize);
4424 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4428 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4429 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4431 btrfs_free_reserved_data_space_noquota(inode, start, len);
4438 * Called if we need to clear a data reservation for this inode
4439 * Normally in a error case.
4441 * This one will *NOT* use accurate qgroup reserved space API, just for case
4442 * which we can't sleep and is sure it won't affect qgroup reserved space.
4443 * Like clear_bit_hook().
4445 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4448 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4449 struct btrfs_space_info *data_sinfo;
4451 /* Make sure the range is aligned to sectorsize */
4452 len = round_up(start + len, fs_info->sectorsize) -
4453 round_down(start, fs_info->sectorsize);
4454 start = round_down(start, fs_info->sectorsize);
4456 data_sinfo = fs_info->data_sinfo;
4457 spin_lock(&data_sinfo->lock);
4458 if (WARN_ON(data_sinfo->bytes_may_use < len))
4459 data_sinfo->bytes_may_use = 0;
4461 data_sinfo->bytes_may_use -= len;
4462 trace_btrfs_space_reservation(fs_info, "space_info",
4463 data_sinfo->flags, len, 0);
4464 spin_unlock(&data_sinfo->lock);
4468 * Called if we need to clear a data reservation for this inode
4469 * Normally in a error case.
4471 * This one will handle the per-inode data rsv map for accurate reserved
4474 void btrfs_free_reserved_data_space(struct inode *inode,
4475 struct extent_changeset *reserved, u64 start, u64 len)
4477 struct btrfs_root *root = BTRFS_I(inode)->root;
4479 /* Make sure the range is aligned to sectorsize */
4480 len = round_up(start + len, root->fs_info->sectorsize) -
4481 round_down(start, root->fs_info->sectorsize);
4482 start = round_down(start, root->fs_info->sectorsize);
4484 btrfs_free_reserved_data_space_noquota(inode, start, len);
4485 btrfs_qgroup_free_data(inode, reserved, start, len);
4488 static void force_metadata_allocation(struct btrfs_fs_info *info)
4490 struct list_head *head = &info->space_info;
4491 struct btrfs_space_info *found;
4494 list_for_each_entry_rcu(found, head, list) {
4495 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4496 found->force_alloc = CHUNK_ALLOC_FORCE;
4501 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4503 return (global->size << 1);
4506 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4507 struct btrfs_space_info *sinfo, int force)
4509 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4510 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4513 if (force == CHUNK_ALLOC_FORCE)
4517 * We need to take into account the global rsv because for all intents
4518 * and purposes it's used space. Don't worry about locking the
4519 * global_rsv, it doesn't change except when the transaction commits.
4521 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4522 bytes_used += calc_global_rsv_need_space(global_rsv);
4525 * in limited mode, we want to have some free space up to
4526 * about 1% of the FS size.
4528 if (force == CHUNK_ALLOC_LIMITED) {
4529 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4530 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4532 if (sinfo->total_bytes - bytes_used < thresh)
4536 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4541 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4545 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4546 BTRFS_BLOCK_GROUP_RAID0 |
4547 BTRFS_BLOCK_GROUP_RAID5 |
4548 BTRFS_BLOCK_GROUP_RAID6))
4549 num_dev = fs_info->fs_devices->rw_devices;
4550 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4553 num_dev = 1; /* DUP or single */
4559 * If @is_allocation is true, reserve space in the system space info necessary
4560 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4563 void check_system_chunk(struct btrfs_trans_handle *trans,
4564 struct btrfs_fs_info *fs_info, u64 type)
4566 struct btrfs_space_info *info;
4573 * Needed because we can end up allocating a system chunk and for an
4574 * atomic and race free space reservation in the chunk block reserve.
4576 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4578 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4579 spin_lock(&info->lock);
4580 left = info->total_bytes - btrfs_space_info_used(info, true);
4581 spin_unlock(&info->lock);
4583 num_devs = get_profile_num_devs(fs_info, type);
4585 /* num_devs device items to update and 1 chunk item to add or remove */
4586 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4587 btrfs_calc_trans_metadata_size(fs_info, 1);
4589 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4590 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4591 left, thresh, type);
4592 dump_space_info(fs_info, info, 0, 0);
4595 if (left < thresh) {
4596 u64 flags = btrfs_system_alloc_profile(fs_info);
4599 * Ignore failure to create system chunk. We might end up not
4600 * needing it, as we might not need to COW all nodes/leafs from
4601 * the paths we visit in the chunk tree (they were already COWed
4602 * or created in the current transaction for example).
4604 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4608 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4609 &fs_info->chunk_block_rsv,
4610 thresh, BTRFS_RESERVE_NO_FLUSH);
4612 trans->chunk_bytes_reserved += thresh;
4617 * If force is CHUNK_ALLOC_FORCE:
4618 * - return 1 if it successfully allocates a chunk,
4619 * - return errors including -ENOSPC otherwise.
4620 * If force is NOT CHUNK_ALLOC_FORCE:
4621 * - return 0 if it doesn't need to allocate a new chunk,
4622 * - return 1 if it successfully allocates a chunk,
4623 * - return errors including -ENOSPC otherwise.
4625 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4626 struct btrfs_fs_info *fs_info, u64 flags, int force)
4628 struct btrfs_space_info *space_info;
4629 int wait_for_alloc = 0;
4632 /* Don't re-enter if we're already allocating a chunk */
4633 if (trans->allocating_chunk)
4636 space_info = __find_space_info(fs_info, flags);
4638 ret = create_space_info(fs_info, flags, &space_info);
4644 spin_lock(&space_info->lock);
4645 if (force < space_info->force_alloc)
4646 force = space_info->force_alloc;
4647 if (space_info->full) {
4648 if (should_alloc_chunk(fs_info, space_info, force))
4652 spin_unlock(&space_info->lock);
4656 if (!should_alloc_chunk(fs_info, space_info, force)) {
4657 spin_unlock(&space_info->lock);
4659 } else if (space_info->chunk_alloc) {
4662 space_info->chunk_alloc = 1;
4665 spin_unlock(&space_info->lock);
4667 mutex_lock(&fs_info->chunk_mutex);
4670 * The chunk_mutex is held throughout the entirety of a chunk
4671 * allocation, so once we've acquired the chunk_mutex we know that the
4672 * other guy is done and we need to recheck and see if we should
4675 if (wait_for_alloc) {
4676 mutex_unlock(&fs_info->chunk_mutex);
4682 trans->allocating_chunk = true;
4685 * If we have mixed data/metadata chunks we want to make sure we keep
4686 * allocating mixed chunks instead of individual chunks.
4688 if (btrfs_mixed_space_info(space_info))
4689 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4692 * if we're doing a data chunk, go ahead and make sure that
4693 * we keep a reasonable number of metadata chunks allocated in the
4696 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4697 fs_info->data_chunk_allocations++;
4698 if (!(fs_info->data_chunk_allocations %
4699 fs_info->metadata_ratio))
4700 force_metadata_allocation(fs_info);
4704 * Check if we have enough space in SYSTEM chunk because we may need
4705 * to update devices.
4707 check_system_chunk(trans, fs_info, flags);
4709 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4710 trans->allocating_chunk = false;
4712 spin_lock(&space_info->lock);
4713 if (ret < 0 && ret != -ENOSPC)
4716 space_info->full = 1;
4720 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4722 space_info->chunk_alloc = 0;
4723 spin_unlock(&space_info->lock);
4724 mutex_unlock(&fs_info->chunk_mutex);
4726 * When we allocate a new chunk we reserve space in the chunk block
4727 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4728 * add new nodes/leafs to it if we end up needing to do it when
4729 * inserting the chunk item and updating device items as part of the
4730 * second phase of chunk allocation, performed by
4731 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4732 * large number of new block groups to create in our transaction
4733 * handle's new_bgs list to avoid exhausting the chunk block reserve
4734 * in extreme cases - like having a single transaction create many new
4735 * block groups when starting to write out the free space caches of all
4736 * the block groups that were made dirty during the lifetime of the
4739 if (trans->can_flush_pending_bgs &&
4740 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4741 btrfs_create_pending_block_groups(trans, fs_info);
4742 btrfs_trans_release_chunk_metadata(trans);
4747 static int can_overcommit(struct btrfs_fs_info *fs_info,
4748 struct btrfs_space_info *space_info, u64 bytes,
4749 enum btrfs_reserve_flush_enum flush,
4752 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4758 /* Don't overcommit when in mixed mode. */
4759 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4763 profile = btrfs_system_alloc_profile(fs_info);
4765 profile = btrfs_metadata_alloc_profile(fs_info);
4767 used = btrfs_space_info_used(space_info, false);
4770 * We only want to allow over committing if we have lots of actual space
4771 * free, but if we don't have enough space to handle the global reserve
4772 * space then we could end up having a real enospc problem when trying
4773 * to allocate a chunk or some other such important allocation.
4775 spin_lock(&global_rsv->lock);
4776 space_size = calc_global_rsv_need_space(global_rsv);
4777 spin_unlock(&global_rsv->lock);
4778 if (used + space_size >= space_info->total_bytes)
4781 used += space_info->bytes_may_use;
4783 avail = atomic64_read(&fs_info->free_chunk_space);
4786 * If we have dup, raid1 or raid10 then only half of the free
4787 * space is actually useable. For raid56, the space info used
4788 * doesn't include the parity drive, so we don't have to
4791 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4792 BTRFS_BLOCK_GROUP_RAID1 |
4793 BTRFS_BLOCK_GROUP_RAID10))
4797 * If we aren't flushing all things, let us overcommit up to
4798 * 1/2th of the space. If we can flush, don't let us overcommit
4799 * too much, let it overcommit up to 1/8 of the space.
4801 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4806 if (used + bytes < space_info->total_bytes + avail)
4811 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4812 unsigned long nr_pages, int nr_items)
4814 struct super_block *sb = fs_info->sb;
4816 if (down_read_trylock(&sb->s_umount)) {
4817 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4818 up_read(&sb->s_umount);
4821 * We needn't worry the filesystem going from r/w to r/o though
4822 * we don't acquire ->s_umount mutex, because the filesystem
4823 * should guarantee the delalloc inodes list be empty after
4824 * the filesystem is readonly(all dirty pages are written to
4827 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4828 if (!current->journal_info)
4829 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4833 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4839 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4840 nr = div64_u64(to_reclaim, bytes);
4846 #define EXTENT_SIZE_PER_ITEM SZ_256K
4849 * shrink metadata reservation for delalloc
4851 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4852 u64 orig, bool wait_ordered)
4854 struct btrfs_block_rsv *block_rsv;
4855 struct btrfs_space_info *space_info;
4856 struct btrfs_trans_handle *trans;
4861 unsigned long nr_pages;
4863 enum btrfs_reserve_flush_enum flush;
4865 /* Calc the number of the pages we need flush for space reservation */
4866 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4867 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4869 trans = (struct btrfs_trans_handle *)current->journal_info;
4870 block_rsv = &fs_info->delalloc_block_rsv;
4871 space_info = block_rsv->space_info;
4873 delalloc_bytes = percpu_counter_sum_positive(
4874 &fs_info->delalloc_bytes);
4875 if (delalloc_bytes == 0) {
4879 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4884 while (delalloc_bytes && loops < 3) {
4885 max_reclaim = min(delalloc_bytes, to_reclaim);
4886 nr_pages = max_reclaim >> PAGE_SHIFT;
4887 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4889 * We need to wait for the async pages to actually start before
4892 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4896 if (max_reclaim <= nr_pages)
4899 max_reclaim -= nr_pages;
4901 wait_event(fs_info->async_submit_wait,
4902 atomic_read(&fs_info->async_delalloc_pages) <=
4906 flush = BTRFS_RESERVE_FLUSH_ALL;
4908 flush = BTRFS_RESERVE_NO_FLUSH;
4909 spin_lock(&space_info->lock);
4910 if (list_empty(&space_info->tickets) &&
4911 list_empty(&space_info->priority_tickets)) {
4912 spin_unlock(&space_info->lock);
4915 spin_unlock(&space_info->lock);
4918 if (wait_ordered && !trans) {
4919 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4921 time_left = schedule_timeout_killable(1);
4925 delalloc_bytes = percpu_counter_sum_positive(
4926 &fs_info->delalloc_bytes);
4930 struct reserve_ticket {
4933 struct list_head list;
4934 wait_queue_head_t wait;
4938 * maybe_commit_transaction - possibly commit the transaction if its ok to
4939 * @root - the root we're allocating for
4940 * @bytes - the number of bytes we want to reserve
4941 * @force - force the commit
4943 * This will check to make sure that committing the transaction will actually
4944 * get us somewhere and then commit the transaction if it does. Otherwise it
4945 * will return -ENOSPC.
4947 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4948 struct btrfs_space_info *space_info)
4950 struct reserve_ticket *ticket = NULL;
4951 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4952 struct btrfs_trans_handle *trans;
4955 trans = (struct btrfs_trans_handle *)current->journal_info;
4959 spin_lock(&space_info->lock);
4960 if (!list_empty(&space_info->priority_tickets))
4961 ticket = list_first_entry(&space_info->priority_tickets,
4962 struct reserve_ticket, list);
4963 else if (!list_empty(&space_info->tickets))
4964 ticket = list_first_entry(&space_info->tickets,
4965 struct reserve_ticket, list);
4966 bytes = (ticket) ? ticket->bytes : 0;
4967 spin_unlock(&space_info->lock);
4972 /* See if there is enough pinned space to make this reservation */
4973 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4978 * See if there is some space in the delayed insertion reservation for
4981 if (space_info != delayed_rsv->space_info)
4984 spin_lock(&delayed_rsv->lock);
4985 if (delayed_rsv->size > bytes)
4988 bytes -= delayed_rsv->size;
4989 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4991 spin_unlock(&delayed_rsv->lock);
4994 spin_unlock(&delayed_rsv->lock);
4997 trans = btrfs_join_transaction(fs_info->extent_root);
5001 return btrfs_commit_transaction(trans);
5005 * Try to flush some data based on policy set by @state. This is only advisory
5006 * and may fail for various reasons. The caller is supposed to examine the
5007 * state of @space_info to detect the outcome.
5009 static void flush_space(struct btrfs_fs_info *fs_info,
5010 struct btrfs_space_info *space_info, u64 num_bytes,
5013 struct btrfs_root *root = fs_info->extent_root;
5014 struct btrfs_trans_handle *trans;
5019 case FLUSH_DELAYED_ITEMS_NR:
5020 case FLUSH_DELAYED_ITEMS:
5021 if (state == FLUSH_DELAYED_ITEMS_NR)
5022 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
5026 trans = btrfs_join_transaction(root);
5027 if (IS_ERR(trans)) {
5028 ret = PTR_ERR(trans);
5031 ret = btrfs_run_delayed_items_nr(trans, fs_info, nr);
5032 btrfs_end_transaction(trans);
5034 case FLUSH_DELALLOC:
5035 case FLUSH_DELALLOC_WAIT:
5036 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
5037 state == FLUSH_DELALLOC_WAIT);
5040 trans = btrfs_join_transaction(root);
5041 if (IS_ERR(trans)) {
5042 ret = PTR_ERR(trans);
5045 ret = do_chunk_alloc(trans, fs_info,
5046 btrfs_metadata_alloc_profile(fs_info),
5047 CHUNK_ALLOC_NO_FORCE);
5048 btrfs_end_transaction(trans);
5049 if (ret > 0 || ret == -ENOSPC)
5053 ret = may_commit_transaction(fs_info, space_info);
5060 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5066 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5067 struct btrfs_space_info *space_info,
5070 struct reserve_ticket *ticket;
5075 list_for_each_entry(ticket, &space_info->tickets, list)
5076 to_reclaim += ticket->bytes;
5077 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5078 to_reclaim += ticket->bytes;
5082 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5083 if (can_overcommit(fs_info, space_info, to_reclaim,
5084 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5087 used = btrfs_space_info_used(space_info, true);
5089 if (can_overcommit(fs_info, space_info, SZ_1M,
5090 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5091 expected = div_factor_fine(space_info->total_bytes, 95);
5093 expected = div_factor_fine(space_info->total_bytes, 90);
5095 if (used > expected)
5096 to_reclaim = used - expected;
5099 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5100 space_info->bytes_reserved);
5104 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5105 struct btrfs_space_info *space_info,
5106 u64 used, bool system_chunk)
5108 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5110 /* If we're just plain full then async reclaim just slows us down. */
5111 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5114 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5118 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5119 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5122 static void wake_all_tickets(struct list_head *head)
5124 struct reserve_ticket *ticket;
5126 while (!list_empty(head)) {
5127 ticket = list_first_entry(head, struct reserve_ticket, list);
5128 list_del_init(&ticket->list);
5129 ticket->error = -ENOSPC;
5130 wake_up(&ticket->wait);
5135 * This is for normal flushers, we can wait all goddamned day if we want to. We
5136 * will loop and continuously try to flush as long as we are making progress.
5137 * We count progress as clearing off tickets each time we have to loop.
5139 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5141 struct btrfs_fs_info *fs_info;
5142 struct btrfs_space_info *space_info;
5145 int commit_cycles = 0;
5146 u64 last_tickets_id;
5148 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5149 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5151 spin_lock(&space_info->lock);
5152 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5155 space_info->flush = 0;
5156 spin_unlock(&space_info->lock);
5159 last_tickets_id = space_info->tickets_id;
5160 spin_unlock(&space_info->lock);
5162 flush_state = FLUSH_DELAYED_ITEMS_NR;
5164 flush_space(fs_info, space_info, to_reclaim, flush_state);
5165 spin_lock(&space_info->lock);
5166 if (list_empty(&space_info->tickets)) {
5167 space_info->flush = 0;
5168 spin_unlock(&space_info->lock);
5171 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5174 if (last_tickets_id == space_info->tickets_id) {
5177 last_tickets_id = space_info->tickets_id;
5178 flush_state = FLUSH_DELAYED_ITEMS_NR;
5183 if (flush_state > COMMIT_TRANS) {
5185 if (commit_cycles > 2) {
5186 wake_all_tickets(&space_info->tickets);
5187 space_info->flush = 0;
5189 flush_state = FLUSH_DELAYED_ITEMS_NR;
5192 spin_unlock(&space_info->lock);
5193 } while (flush_state <= COMMIT_TRANS);
5196 void btrfs_init_async_reclaim_work(struct work_struct *work)
5198 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5201 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5202 struct btrfs_space_info *space_info,
5203 struct reserve_ticket *ticket)
5206 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5208 spin_lock(&space_info->lock);
5209 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5212 spin_unlock(&space_info->lock);
5215 spin_unlock(&space_info->lock);
5218 flush_space(fs_info, space_info, to_reclaim, flush_state);
5220 spin_lock(&space_info->lock);
5221 if (ticket->bytes == 0) {
5222 spin_unlock(&space_info->lock);
5225 spin_unlock(&space_info->lock);
5228 * Priority flushers can't wait on delalloc without
5231 if (flush_state == FLUSH_DELALLOC ||
5232 flush_state == FLUSH_DELALLOC_WAIT)
5233 flush_state = ALLOC_CHUNK;
5234 } while (flush_state < COMMIT_TRANS);
5237 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5238 struct btrfs_space_info *space_info,
5239 struct reserve_ticket *ticket, u64 orig_bytes)
5245 spin_lock(&space_info->lock);
5246 while (ticket->bytes > 0 && ticket->error == 0) {
5247 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5252 spin_unlock(&space_info->lock);
5256 finish_wait(&ticket->wait, &wait);
5257 spin_lock(&space_info->lock);
5260 ret = ticket->error;
5261 if (!list_empty(&ticket->list))
5262 list_del_init(&ticket->list);
5263 if (ticket->bytes && ticket->bytes < orig_bytes) {
5264 u64 num_bytes = orig_bytes - ticket->bytes;
5265 space_info->bytes_may_use -= num_bytes;
5266 trace_btrfs_space_reservation(fs_info, "space_info",
5267 space_info->flags, num_bytes, 0);
5269 spin_unlock(&space_info->lock);
5275 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5276 * @root - the root we're allocating for
5277 * @space_info - the space info we want to allocate from
5278 * @orig_bytes - the number of bytes we want
5279 * @flush - whether or not we can flush to make our reservation
5281 * This will reserve orig_bytes number of bytes from the space info associated
5282 * with the block_rsv. If there is not enough space it will make an attempt to
5283 * flush out space to make room. It will do this by flushing delalloc if
5284 * possible or committing the transaction. If flush is 0 then no attempts to
5285 * regain reservations will be made and this will fail if there is not enough
5288 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5289 struct btrfs_space_info *space_info,
5291 enum btrfs_reserve_flush_enum flush,
5294 struct reserve_ticket ticket;
5299 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5301 spin_lock(&space_info->lock);
5303 used = btrfs_space_info_used(space_info, true);
5306 * If we have enough space then hooray, make our reservation and carry
5307 * on. If not see if we can overcommit, and if we can, hooray carry on.
5308 * If not things get more complicated.
5310 if (used + orig_bytes <= space_info->total_bytes) {
5311 space_info->bytes_may_use += orig_bytes;
5312 trace_btrfs_space_reservation(fs_info, "space_info",
5313 space_info->flags, orig_bytes, 1);
5315 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5317 space_info->bytes_may_use += orig_bytes;
5318 trace_btrfs_space_reservation(fs_info, "space_info",
5319 space_info->flags, orig_bytes, 1);
5324 * If we couldn't make a reservation then setup our reservation ticket
5325 * and kick the async worker if it's not already running.
5327 * If we are a priority flusher then we just need to add our ticket to
5328 * the list and we will do our own flushing further down.
5330 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5331 ticket.bytes = orig_bytes;
5333 init_waitqueue_head(&ticket.wait);
5334 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5335 list_add_tail(&ticket.list, &space_info->tickets);
5336 if (!space_info->flush) {
5337 space_info->flush = 1;
5338 trace_btrfs_trigger_flush(fs_info,
5342 queue_work(system_unbound_wq,
5343 &fs_info->async_reclaim_work);
5346 list_add_tail(&ticket.list,
5347 &space_info->priority_tickets);
5349 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5352 * We will do the space reservation dance during log replay,
5353 * which means we won't have fs_info->fs_root set, so don't do
5354 * the async reclaim as we will panic.
5356 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5357 need_do_async_reclaim(fs_info, space_info,
5358 used, system_chunk) &&
5359 !work_busy(&fs_info->async_reclaim_work)) {
5360 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5361 orig_bytes, flush, "preempt");
5362 queue_work(system_unbound_wq,
5363 &fs_info->async_reclaim_work);
5366 spin_unlock(&space_info->lock);
5367 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5370 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5371 return wait_reserve_ticket(fs_info, space_info, &ticket,
5375 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5376 spin_lock(&space_info->lock);
5378 if (ticket.bytes < orig_bytes) {
5379 u64 num_bytes = orig_bytes - ticket.bytes;
5380 space_info->bytes_may_use -= num_bytes;
5381 trace_btrfs_space_reservation(fs_info, "space_info",
5386 list_del_init(&ticket.list);
5389 spin_unlock(&space_info->lock);
5390 ASSERT(list_empty(&ticket.list));
5395 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5396 * @root - the root we're allocating for
5397 * @block_rsv - the block_rsv we're allocating for
5398 * @orig_bytes - the number of bytes we want
5399 * @flush - whether or not we can flush to make our reservation
5401 * This will reserve orgi_bytes number of bytes from the space info associated
5402 * with the block_rsv. If there is not enough space it will make an attempt to
5403 * flush out space to make room. It will do this by flushing delalloc if
5404 * possible or committing the transaction. If flush is 0 then no attempts to
5405 * regain reservations will be made and this will fail if there is not enough
5408 static int reserve_metadata_bytes(struct btrfs_root *root,
5409 struct btrfs_block_rsv *block_rsv,
5411 enum btrfs_reserve_flush_enum flush)
5413 struct btrfs_fs_info *fs_info = root->fs_info;
5414 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5416 bool system_chunk = (root == fs_info->chunk_root);
5418 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5419 orig_bytes, flush, system_chunk);
5420 if (ret == -ENOSPC &&
5421 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5422 if (block_rsv != global_rsv &&
5423 !block_rsv_use_bytes(global_rsv, orig_bytes))
5427 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5428 block_rsv->space_info->flags,
5433 static struct btrfs_block_rsv *get_block_rsv(
5434 const struct btrfs_trans_handle *trans,
5435 const struct btrfs_root *root)
5437 struct btrfs_fs_info *fs_info = root->fs_info;
5438 struct btrfs_block_rsv *block_rsv = NULL;
5440 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5441 (root == fs_info->csum_root && trans->adding_csums) ||
5442 (root == fs_info->uuid_root))
5443 block_rsv = trans->block_rsv;
5446 block_rsv = root->block_rsv;
5449 block_rsv = &fs_info->empty_block_rsv;
5454 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5458 spin_lock(&block_rsv->lock);
5459 if (block_rsv->reserved >= num_bytes) {
5460 block_rsv->reserved -= num_bytes;
5461 if (block_rsv->reserved < block_rsv->size)
5462 block_rsv->full = 0;
5465 spin_unlock(&block_rsv->lock);
5469 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5470 u64 num_bytes, int update_size)
5472 spin_lock(&block_rsv->lock);
5473 block_rsv->reserved += num_bytes;
5475 block_rsv->size += num_bytes;
5476 else if (block_rsv->reserved >= block_rsv->size)
5477 block_rsv->full = 1;
5478 spin_unlock(&block_rsv->lock);
5481 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5482 struct btrfs_block_rsv *dest, u64 num_bytes,
5485 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5488 if (global_rsv->space_info != dest->space_info)
5491 spin_lock(&global_rsv->lock);
5492 min_bytes = div_factor(global_rsv->size, min_factor);
5493 if (global_rsv->reserved < min_bytes + num_bytes) {
5494 spin_unlock(&global_rsv->lock);
5497 global_rsv->reserved -= num_bytes;
5498 if (global_rsv->reserved < global_rsv->size)
5499 global_rsv->full = 0;
5500 spin_unlock(&global_rsv->lock);
5502 block_rsv_add_bytes(dest, num_bytes, 1);
5507 * This is for space we already have accounted in space_info->bytes_may_use, so
5508 * basically when we're returning space from block_rsv's.
5510 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5511 struct btrfs_space_info *space_info,
5514 struct reserve_ticket *ticket;
5515 struct list_head *head;
5517 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5518 bool check_overcommit = false;
5520 spin_lock(&space_info->lock);
5521 head = &space_info->priority_tickets;
5524 * If we are over our limit then we need to check and see if we can
5525 * overcommit, and if we can't then we just need to free up our space
5526 * and not satisfy any requests.
5528 used = btrfs_space_info_used(space_info, true);
5529 if (used - num_bytes >= space_info->total_bytes)
5530 check_overcommit = true;
5532 while (!list_empty(head) && num_bytes) {
5533 ticket = list_first_entry(head, struct reserve_ticket,
5536 * We use 0 bytes because this space is already reserved, so
5537 * adding the ticket space would be a double count.
5539 if (check_overcommit &&
5540 !can_overcommit(fs_info, space_info, 0, flush, false))
5542 if (num_bytes >= ticket->bytes) {
5543 list_del_init(&ticket->list);
5544 num_bytes -= ticket->bytes;
5546 space_info->tickets_id++;
5547 wake_up(&ticket->wait);
5549 ticket->bytes -= num_bytes;
5554 if (num_bytes && head == &space_info->priority_tickets) {
5555 head = &space_info->tickets;
5556 flush = BTRFS_RESERVE_FLUSH_ALL;
5559 space_info->bytes_may_use -= num_bytes;
5560 trace_btrfs_space_reservation(fs_info, "space_info",
5561 space_info->flags, num_bytes, 0);
5562 spin_unlock(&space_info->lock);
5566 * This is for newly allocated space that isn't accounted in
5567 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5568 * we use this helper.
5570 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5571 struct btrfs_space_info *space_info,
5574 struct reserve_ticket *ticket;
5575 struct list_head *head = &space_info->priority_tickets;
5578 while (!list_empty(head) && num_bytes) {
5579 ticket = list_first_entry(head, struct reserve_ticket,
5581 if (num_bytes >= ticket->bytes) {
5582 trace_btrfs_space_reservation(fs_info, "space_info",
5585 list_del_init(&ticket->list);
5586 num_bytes -= ticket->bytes;
5587 space_info->bytes_may_use += ticket->bytes;
5589 space_info->tickets_id++;
5590 wake_up(&ticket->wait);
5592 trace_btrfs_space_reservation(fs_info, "space_info",
5595 space_info->bytes_may_use += num_bytes;
5596 ticket->bytes -= num_bytes;
5601 if (num_bytes && head == &space_info->priority_tickets) {
5602 head = &space_info->tickets;
5607 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5608 struct btrfs_block_rsv *block_rsv,
5609 struct btrfs_block_rsv *dest, u64 num_bytes)
5611 struct btrfs_space_info *space_info = block_rsv->space_info;
5613 spin_lock(&block_rsv->lock);
5614 if (num_bytes == (u64)-1)
5615 num_bytes = block_rsv->size;
5616 block_rsv->size -= num_bytes;
5617 if (block_rsv->reserved >= block_rsv->size) {
5618 num_bytes = block_rsv->reserved - block_rsv->size;
5619 block_rsv->reserved = block_rsv->size;
5620 block_rsv->full = 1;
5624 spin_unlock(&block_rsv->lock);
5626 if (num_bytes > 0) {
5628 spin_lock(&dest->lock);
5632 bytes_to_add = dest->size - dest->reserved;
5633 bytes_to_add = min(num_bytes, bytes_to_add);
5634 dest->reserved += bytes_to_add;
5635 if (dest->reserved >= dest->size)
5637 num_bytes -= bytes_to_add;
5639 spin_unlock(&dest->lock);
5642 space_info_add_old_bytes(fs_info, space_info,
5647 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5648 struct btrfs_block_rsv *dst, u64 num_bytes,
5653 ret = block_rsv_use_bytes(src, num_bytes);
5657 block_rsv_add_bytes(dst, num_bytes, update_size);
5661 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5663 memset(rsv, 0, sizeof(*rsv));
5664 spin_lock_init(&rsv->lock);
5668 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5669 unsigned short type)
5671 struct btrfs_block_rsv *block_rsv;
5673 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5677 btrfs_init_block_rsv(block_rsv, type);
5678 block_rsv->space_info = __find_space_info(fs_info,
5679 BTRFS_BLOCK_GROUP_METADATA);
5683 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5684 struct btrfs_block_rsv *rsv)
5688 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5692 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5697 int btrfs_block_rsv_add(struct btrfs_root *root,
5698 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5699 enum btrfs_reserve_flush_enum flush)
5706 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5708 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5715 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5723 spin_lock(&block_rsv->lock);
5724 num_bytes = div_factor(block_rsv->size, min_factor);
5725 if (block_rsv->reserved >= num_bytes)
5727 spin_unlock(&block_rsv->lock);
5732 int btrfs_block_rsv_refill(struct btrfs_root *root,
5733 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5734 enum btrfs_reserve_flush_enum flush)
5742 spin_lock(&block_rsv->lock);
5743 num_bytes = min_reserved;
5744 if (block_rsv->reserved >= num_bytes)
5747 num_bytes -= block_rsv->reserved;
5748 spin_unlock(&block_rsv->lock);
5753 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5755 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5762 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5763 struct btrfs_block_rsv *block_rsv,
5766 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5768 if (global_rsv == block_rsv ||
5769 block_rsv->space_info != global_rsv->space_info)
5771 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5774 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5776 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5777 struct btrfs_space_info *sinfo = block_rsv->space_info;
5781 * The global block rsv is based on the size of the extent tree, the
5782 * checksum tree and the root tree. If the fs is empty we want to set
5783 * it to a minimal amount for safety.
5785 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5786 btrfs_root_used(&fs_info->csum_root->root_item) +
5787 btrfs_root_used(&fs_info->tree_root->root_item);
5788 num_bytes = max_t(u64, num_bytes, SZ_16M);
5790 spin_lock(&sinfo->lock);
5791 spin_lock(&block_rsv->lock);
5793 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5795 if (block_rsv->reserved < block_rsv->size) {
5796 num_bytes = btrfs_space_info_used(sinfo, true);
5797 if (sinfo->total_bytes > num_bytes) {
5798 num_bytes = sinfo->total_bytes - num_bytes;
5799 num_bytes = min(num_bytes,
5800 block_rsv->size - block_rsv->reserved);
5801 block_rsv->reserved += num_bytes;
5802 sinfo->bytes_may_use += num_bytes;
5803 trace_btrfs_space_reservation(fs_info, "space_info",
5804 sinfo->flags, num_bytes,
5807 } else if (block_rsv->reserved > block_rsv->size) {
5808 num_bytes = block_rsv->reserved - block_rsv->size;
5809 sinfo->bytes_may_use -= num_bytes;
5810 trace_btrfs_space_reservation(fs_info, "space_info",
5811 sinfo->flags, num_bytes, 0);
5812 block_rsv->reserved = block_rsv->size;
5815 if (block_rsv->reserved == block_rsv->size)
5816 block_rsv->full = 1;
5818 block_rsv->full = 0;
5820 spin_unlock(&block_rsv->lock);
5821 spin_unlock(&sinfo->lock);
5824 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5826 struct btrfs_space_info *space_info;
5828 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5829 fs_info->chunk_block_rsv.space_info = space_info;
5831 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5832 fs_info->global_block_rsv.space_info = space_info;
5833 fs_info->delalloc_block_rsv.space_info = space_info;
5834 fs_info->trans_block_rsv.space_info = space_info;
5835 fs_info->empty_block_rsv.space_info = space_info;
5836 fs_info->delayed_block_rsv.space_info = space_info;
5838 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5839 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5840 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5841 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5842 if (fs_info->quota_root)
5843 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5844 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5846 update_global_block_rsv(fs_info);
5849 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5851 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5853 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5854 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5855 WARN_ON(fs_info->trans_block_rsv.size > 0);
5856 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5857 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5858 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5859 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5860 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5863 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5864 struct btrfs_fs_info *fs_info)
5866 if (!trans->block_rsv)
5869 if (!trans->bytes_reserved)
5872 trace_btrfs_space_reservation(fs_info, "transaction",
5873 trans->transid, trans->bytes_reserved, 0);
5874 btrfs_block_rsv_release(fs_info, trans->block_rsv,
5875 trans->bytes_reserved);
5876 trans->bytes_reserved = 0;
5880 * To be called after all the new block groups attached to the transaction
5881 * handle have been created (btrfs_create_pending_block_groups()).
5883 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5885 struct btrfs_fs_info *fs_info = trans->fs_info;
5887 if (!trans->chunk_bytes_reserved)
5890 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5892 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5893 trans->chunk_bytes_reserved);
5894 trans->chunk_bytes_reserved = 0;
5897 /* Can only return 0 or -ENOSPC */
5898 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5899 struct btrfs_inode *inode)
5901 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5902 struct btrfs_root *root = inode->root;
5904 * We always use trans->block_rsv here as we will have reserved space
5905 * for our orphan when starting the transaction, using get_block_rsv()
5906 * here will sometimes make us choose the wrong block rsv as we could be
5907 * doing a reloc inode for a non refcounted root.
5909 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5910 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5913 * We need to hold space in order to delete our orphan item once we've
5914 * added it, so this takes the reservation so we can release it later
5915 * when we are truly done with the orphan item.
5917 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5919 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5921 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5924 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5926 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5927 struct btrfs_root *root = inode->root;
5928 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5930 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5932 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5936 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5937 * root: the root of the parent directory
5938 * rsv: block reservation
5939 * items: the number of items that we need do reservation
5940 * qgroup_reserved: used to return the reserved size in qgroup
5942 * This function is used to reserve the space for snapshot/subvolume
5943 * creation and deletion. Those operations are different with the
5944 * common file/directory operations, they change two fs/file trees
5945 * and root tree, the number of items that the qgroup reserves is
5946 * different with the free space reservation. So we can not use
5947 * the space reservation mechanism in start_transaction().
5949 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5950 struct btrfs_block_rsv *rsv,
5952 u64 *qgroup_reserved,
5953 bool use_global_rsv)
5957 struct btrfs_fs_info *fs_info = root->fs_info;
5958 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5960 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5961 /* One for parent inode, two for dir entries */
5962 num_bytes = 3 * fs_info->nodesize;
5963 ret = btrfs_qgroup_reserve_meta(root, num_bytes, true);
5970 *qgroup_reserved = num_bytes;
5972 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5973 rsv->space_info = __find_space_info(fs_info,
5974 BTRFS_BLOCK_GROUP_METADATA);
5975 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5976 BTRFS_RESERVE_FLUSH_ALL);
5978 if (ret == -ENOSPC && use_global_rsv)
5979 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5981 if (ret && *qgroup_reserved)
5982 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5987 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5988 struct btrfs_block_rsv *rsv)
5990 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5994 * drop_outstanding_extent - drop an outstanding extent
5995 * @inode: the inode we're dropping the extent for
5996 * @num_bytes: the number of bytes we're releasing.
5998 * This is called when we are freeing up an outstanding extent, either called
5999 * after an error or after an extent is written. This will return the number of
6000 * reserved extents that need to be freed. This must be called with
6001 * BTRFS_I(inode)->lock held.
6003 static unsigned drop_outstanding_extent(struct btrfs_inode *inode,
6006 unsigned drop_inode_space = 0;
6007 unsigned dropped_extents = 0;
6008 unsigned num_extents;
6010 num_extents = count_max_extents(num_bytes);
6011 ASSERT(num_extents);
6012 ASSERT(inode->outstanding_extents >= num_extents);
6013 inode->outstanding_extents -= num_extents;
6015 if (inode->outstanding_extents == 0 &&
6016 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6017 &inode->runtime_flags))
6018 drop_inode_space = 1;
6021 * If we have more or the same amount of outstanding extents than we have
6022 * reserved then we need to leave the reserved extents count alone.
6024 if (inode->outstanding_extents >= inode->reserved_extents)
6025 return drop_inode_space;
6027 dropped_extents = inode->reserved_extents - inode->outstanding_extents;
6028 inode->reserved_extents -= dropped_extents;
6029 return dropped_extents + drop_inode_space;
6033 * calc_csum_metadata_size - return the amount of metadata space that must be
6034 * reserved/freed for the given bytes.
6035 * @inode: the inode we're manipulating
6036 * @num_bytes: the number of bytes in question
6037 * @reserve: 1 if we are reserving space, 0 if we are freeing space
6039 * This adjusts the number of csum_bytes in the inode and then returns the
6040 * correct amount of metadata that must either be reserved or freed. We
6041 * calculate how many checksums we can fit into one leaf and then divide the
6042 * number of bytes that will need to be checksumed by this value to figure out
6043 * how many checksums will be required. If we are adding bytes then the number
6044 * may go up and we will return the number of additional bytes that must be
6045 * reserved. If it is going down we will return the number of bytes that must
6048 * This must be called with BTRFS_I(inode)->lock held.
6050 static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes,
6053 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6054 u64 old_csums, num_csums;
6056 if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0)
6059 old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
6061 inode->csum_bytes += num_bytes;
6063 inode->csum_bytes -= num_bytes;
6064 num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
6066 /* No change, no need to reserve more */
6067 if (old_csums == num_csums)
6071 return btrfs_calc_trans_metadata_size(fs_info,
6072 num_csums - old_csums);
6074 return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums);
6077 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6079 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6080 struct btrfs_root *root = inode->root;
6081 struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv;
6084 unsigned nr_extents;
6085 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6087 bool delalloc_lock = true;
6090 bool release_extra = false;
6092 /* If we are a free space inode we need to not flush since we will be in
6093 * the middle of a transaction commit. We also don't need the delalloc
6094 * mutex since we won't race with anybody. We need this mostly to make
6095 * lockdep shut its filthy mouth.
6097 * If we have a transaction open (can happen if we call truncate_block
6098 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6100 if (btrfs_is_free_space_inode(inode)) {
6101 flush = BTRFS_RESERVE_NO_FLUSH;
6102 delalloc_lock = false;
6103 } else if (current->journal_info) {
6104 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6107 if (flush != BTRFS_RESERVE_NO_FLUSH &&
6108 btrfs_transaction_in_commit(fs_info))
6109 schedule_timeout(1);
6112 mutex_lock(&inode->delalloc_mutex);
6114 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6116 spin_lock(&inode->lock);
6117 nr_extents = count_max_extents(num_bytes);
6118 inode->outstanding_extents += nr_extents;
6121 if (inode->outstanding_extents > inode->reserved_extents)
6122 nr_extents += inode->outstanding_extents -
6123 inode->reserved_extents;
6125 /* We always want to reserve a slot for updating the inode. */
6126 to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1);
6127 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
6128 csum_bytes = inode->csum_bytes;
6129 spin_unlock(&inode->lock);
6131 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6132 ret = btrfs_qgroup_reserve_meta(root,
6133 nr_extents * fs_info->nodesize, true);
6138 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
6139 if (unlikely(ret)) {
6140 btrfs_qgroup_free_meta(root,
6141 nr_extents * fs_info->nodesize);
6145 spin_lock(&inode->lock);
6146 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6147 &inode->runtime_flags)) {
6148 to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1);
6149 release_extra = true;
6151 inode->reserved_extents += nr_extents;
6152 spin_unlock(&inode->lock);
6155 mutex_unlock(&inode->delalloc_mutex);
6158 trace_btrfs_space_reservation(fs_info, "delalloc",
6159 btrfs_ino(inode), to_reserve, 1);
6161 btrfs_block_rsv_release(fs_info, block_rsv,
6162 btrfs_calc_trans_metadata_size(fs_info, 1));
6166 spin_lock(&inode->lock);
6167 dropped = drop_outstanding_extent(inode, num_bytes);
6169 * If the inodes csum_bytes is the same as the original
6170 * csum_bytes then we know we haven't raced with any free()ers
6171 * so we can just reduce our inodes csum bytes and carry on.
6173 if (inode->csum_bytes == csum_bytes) {
6174 calc_csum_metadata_size(inode, num_bytes, 0);
6176 u64 orig_csum_bytes = inode->csum_bytes;
6180 * This is tricky, but first we need to figure out how much we
6181 * freed from any free-ers that occurred during this
6182 * reservation, so we reset ->csum_bytes to the csum_bytes
6183 * before we dropped our lock, and then call the free for the
6184 * number of bytes that were freed while we were trying our
6187 bytes = csum_bytes - inode->csum_bytes;
6188 inode->csum_bytes = csum_bytes;
6189 to_free = calc_csum_metadata_size(inode, bytes, 0);
6193 * Now we need to see how much we would have freed had we not
6194 * been making this reservation and our ->csum_bytes were not
6195 * artificially inflated.
6197 inode->csum_bytes = csum_bytes - num_bytes;
6198 bytes = csum_bytes - orig_csum_bytes;
6199 bytes = calc_csum_metadata_size(inode, bytes, 0);
6202 * Now reset ->csum_bytes to what it should be. If bytes is
6203 * more than to_free then we would have freed more space had we
6204 * not had an artificially high ->csum_bytes, so we need to free
6205 * the remainder. If bytes is the same or less then we don't
6206 * need to do anything, the other free-ers did the correct
6209 inode->csum_bytes = orig_csum_bytes - num_bytes;
6210 if (bytes > to_free)
6211 to_free = bytes - to_free;
6215 spin_unlock(&inode->lock);
6217 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6220 btrfs_block_rsv_release(fs_info, block_rsv, to_free);
6221 trace_btrfs_space_reservation(fs_info, "delalloc",
6222 btrfs_ino(inode), to_free, 0);
6225 mutex_unlock(&inode->delalloc_mutex);
6230 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6231 * @inode: the inode to release the reservation for
6232 * @num_bytes: the number of bytes we're releasing
6234 * This will release the metadata reservation for an inode. This can be called
6235 * once we complete IO for a given set of bytes to release their metadata
6238 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes)
6240 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6244 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6245 spin_lock(&inode->lock);
6246 dropped = drop_outstanding_extent(inode, num_bytes);
6249 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6250 spin_unlock(&inode->lock);
6252 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6254 if (btrfs_is_testing(fs_info))
6257 trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode),
6260 btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free);
6264 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6266 * @inode: inode we're writing to
6267 * @start: start range we are writing to
6268 * @len: how long the range we are writing to
6269 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6270 * current reservation.
6272 * This will do the following things
6274 * o reserve space in data space info for num bytes
6275 * and reserve precious corresponding qgroup space
6276 * (Done in check_data_free_space)
6278 * o reserve space for metadata space, based on the number of outstanding
6279 * extents and how much csums will be needed
6280 * also reserve metadata space in a per root over-reserve method.
6281 * o add to the inodes->delalloc_bytes
6282 * o add it to the fs_info's delalloc inodes list.
6283 * (Above 3 all done in delalloc_reserve_metadata)
6285 * Return 0 for success
6286 * Return <0 for error(-ENOSPC or -EQUOT)
6288 int btrfs_delalloc_reserve_space(struct inode *inode,
6289 struct extent_changeset **reserved, u64 start, u64 len)
6293 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6296 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6298 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6303 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6304 * @inode: inode we're releasing space for
6305 * @start: start position of the space already reserved
6306 * @len: the len of the space already reserved
6308 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6309 * called in the case that we don't need the metadata AND data reservations
6310 * anymore. So if there is an error or we insert an inline extent.
6312 * This function will release the metadata space that was not used and will
6313 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6314 * list if there are no delalloc bytes left.
6315 * Also it will handle the qgroup reserved space.
6317 void btrfs_delalloc_release_space(struct inode *inode,
6318 struct extent_changeset *reserved, u64 start, u64 len)
6320 btrfs_delalloc_release_metadata(BTRFS_I(inode), len);
6321 btrfs_free_reserved_data_space(inode, reserved, start, len);
6324 static int update_block_group(struct btrfs_trans_handle *trans,
6325 struct btrfs_fs_info *info, u64 bytenr,
6326 u64 num_bytes, int alloc)
6328 struct btrfs_block_group_cache *cache = NULL;
6329 u64 total = num_bytes;
6334 /* block accounting for super block */
6335 spin_lock(&info->delalloc_root_lock);
6336 old_val = btrfs_super_bytes_used(info->super_copy);
6338 old_val += num_bytes;
6340 old_val -= num_bytes;
6341 btrfs_set_super_bytes_used(info->super_copy, old_val);
6342 spin_unlock(&info->delalloc_root_lock);
6345 cache = btrfs_lookup_block_group(info, bytenr);
6348 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6349 BTRFS_BLOCK_GROUP_RAID1 |
6350 BTRFS_BLOCK_GROUP_RAID10))
6355 * If this block group has free space cache written out, we
6356 * need to make sure to load it if we are removing space. This
6357 * is because we need the unpinning stage to actually add the
6358 * space back to the block group, otherwise we will leak space.
6360 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6361 cache_block_group(cache, 1);
6363 byte_in_group = bytenr - cache->key.objectid;
6364 WARN_ON(byte_in_group > cache->key.offset);
6366 spin_lock(&cache->space_info->lock);
6367 spin_lock(&cache->lock);
6369 if (btrfs_test_opt(info, SPACE_CACHE) &&
6370 cache->disk_cache_state < BTRFS_DC_CLEAR)
6371 cache->disk_cache_state = BTRFS_DC_CLEAR;
6373 old_val = btrfs_block_group_used(&cache->item);
6374 num_bytes = min(total, cache->key.offset - byte_in_group);
6376 old_val += num_bytes;
6377 btrfs_set_block_group_used(&cache->item, old_val);
6378 cache->reserved -= num_bytes;
6379 cache->space_info->bytes_reserved -= num_bytes;
6380 cache->space_info->bytes_used += num_bytes;
6381 cache->space_info->disk_used += num_bytes * factor;
6382 spin_unlock(&cache->lock);
6383 spin_unlock(&cache->space_info->lock);
6385 old_val -= num_bytes;
6386 btrfs_set_block_group_used(&cache->item, old_val);
6387 cache->pinned += num_bytes;
6388 cache->space_info->bytes_pinned += num_bytes;
6389 cache->space_info->bytes_used -= num_bytes;
6390 cache->space_info->disk_used -= num_bytes * factor;
6391 spin_unlock(&cache->lock);
6392 spin_unlock(&cache->space_info->lock);
6394 trace_btrfs_space_reservation(info, "pinned",
6395 cache->space_info->flags,
6397 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6399 set_extent_dirty(info->pinned_extents,
6400 bytenr, bytenr + num_bytes - 1,
6401 GFP_NOFS | __GFP_NOFAIL);
6404 spin_lock(&trans->transaction->dirty_bgs_lock);
6405 if (list_empty(&cache->dirty_list)) {
6406 list_add_tail(&cache->dirty_list,
6407 &trans->transaction->dirty_bgs);
6408 trans->transaction->num_dirty_bgs++;
6409 btrfs_get_block_group(cache);
6411 spin_unlock(&trans->transaction->dirty_bgs_lock);
6414 * No longer have used bytes in this block group, queue it for
6415 * deletion. We do this after adding the block group to the
6416 * dirty list to avoid races between cleaner kthread and space
6419 if (!alloc && old_val == 0) {
6420 spin_lock(&info->unused_bgs_lock);
6421 if (list_empty(&cache->bg_list)) {
6422 btrfs_get_block_group(cache);
6423 list_add_tail(&cache->bg_list,
6426 spin_unlock(&info->unused_bgs_lock);
6429 btrfs_put_block_group(cache);
6431 bytenr += num_bytes;
6436 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6438 struct btrfs_block_group_cache *cache;
6441 spin_lock(&fs_info->block_group_cache_lock);
6442 bytenr = fs_info->first_logical_byte;
6443 spin_unlock(&fs_info->block_group_cache_lock);
6445 if (bytenr < (u64)-1)
6448 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6452 bytenr = cache->key.objectid;
6453 btrfs_put_block_group(cache);
6458 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6459 struct btrfs_block_group_cache *cache,
6460 u64 bytenr, u64 num_bytes, int reserved)
6462 spin_lock(&cache->space_info->lock);
6463 spin_lock(&cache->lock);
6464 cache->pinned += num_bytes;
6465 cache->space_info->bytes_pinned += num_bytes;
6467 cache->reserved -= num_bytes;
6468 cache->space_info->bytes_reserved -= num_bytes;
6470 spin_unlock(&cache->lock);
6471 spin_unlock(&cache->space_info->lock);
6473 trace_btrfs_space_reservation(fs_info, "pinned",
6474 cache->space_info->flags, num_bytes, 1);
6475 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6476 set_extent_dirty(fs_info->pinned_extents, bytenr,
6477 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6482 * this function must be called within transaction
6484 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6485 u64 bytenr, u64 num_bytes, int reserved)
6487 struct btrfs_block_group_cache *cache;
6489 cache = btrfs_lookup_block_group(fs_info, bytenr);
6490 BUG_ON(!cache); /* Logic error */
6492 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6494 btrfs_put_block_group(cache);
6499 * this function must be called within transaction
6501 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6502 u64 bytenr, u64 num_bytes)
6504 struct btrfs_block_group_cache *cache;
6507 cache = btrfs_lookup_block_group(fs_info, bytenr);
6512 * pull in the free space cache (if any) so that our pin
6513 * removes the free space from the cache. We have load_only set
6514 * to one because the slow code to read in the free extents does check
6515 * the pinned extents.
6517 cache_block_group(cache, 1);
6519 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6521 /* remove us from the free space cache (if we're there at all) */
6522 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6523 btrfs_put_block_group(cache);
6527 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6528 u64 start, u64 num_bytes)
6531 struct btrfs_block_group_cache *block_group;
6532 struct btrfs_caching_control *caching_ctl;
6534 block_group = btrfs_lookup_block_group(fs_info, start);
6538 cache_block_group(block_group, 0);
6539 caching_ctl = get_caching_control(block_group);
6543 BUG_ON(!block_group_cache_done(block_group));
6544 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6546 mutex_lock(&caching_ctl->mutex);
6548 if (start >= caching_ctl->progress) {
6549 ret = add_excluded_extent(fs_info, start, num_bytes);
6550 } else if (start + num_bytes <= caching_ctl->progress) {
6551 ret = btrfs_remove_free_space(block_group,
6554 num_bytes = caching_ctl->progress - start;
6555 ret = btrfs_remove_free_space(block_group,
6560 num_bytes = (start + num_bytes) -
6561 caching_ctl->progress;
6562 start = caching_ctl->progress;
6563 ret = add_excluded_extent(fs_info, start, num_bytes);
6566 mutex_unlock(&caching_ctl->mutex);
6567 put_caching_control(caching_ctl);
6569 btrfs_put_block_group(block_group);
6573 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6574 struct extent_buffer *eb)
6576 struct btrfs_file_extent_item *item;
6577 struct btrfs_key key;
6581 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6584 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6585 btrfs_item_key_to_cpu(eb, &key, i);
6586 if (key.type != BTRFS_EXTENT_DATA_KEY)
6588 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6589 found_type = btrfs_file_extent_type(eb, item);
6590 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6592 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6594 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6595 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6596 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6603 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6605 atomic_inc(&bg->reservations);
6608 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6611 struct btrfs_block_group_cache *bg;
6613 bg = btrfs_lookup_block_group(fs_info, start);
6615 if (atomic_dec_and_test(&bg->reservations))
6616 wake_up_atomic_t(&bg->reservations);
6617 btrfs_put_block_group(bg);
6620 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6626 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6628 struct btrfs_space_info *space_info = bg->space_info;
6632 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6636 * Our block group is read only but before we set it to read only,
6637 * some task might have had allocated an extent from it already, but it
6638 * has not yet created a respective ordered extent (and added it to a
6639 * root's list of ordered extents).
6640 * Therefore wait for any task currently allocating extents, since the
6641 * block group's reservations counter is incremented while a read lock
6642 * on the groups' semaphore is held and decremented after releasing
6643 * the read access on that semaphore and creating the ordered extent.
6645 down_write(&space_info->groups_sem);
6646 up_write(&space_info->groups_sem);
6648 wait_on_atomic_t(&bg->reservations,
6649 btrfs_wait_bg_reservations_atomic_t,
6650 TASK_UNINTERRUPTIBLE);
6654 * btrfs_add_reserved_bytes - update the block_group and space info counters
6655 * @cache: The cache we are manipulating
6656 * @ram_bytes: The number of bytes of file content, and will be same to
6657 * @num_bytes except for the compress path.
6658 * @num_bytes: The number of bytes in question
6659 * @delalloc: The blocks are allocated for the delalloc write
6661 * This is called by the allocator when it reserves space. If this is a
6662 * reservation and the block group has become read only we cannot make the
6663 * reservation and return -EAGAIN, otherwise this function always succeeds.
6665 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6666 u64 ram_bytes, u64 num_bytes, int delalloc)
6668 struct btrfs_space_info *space_info = cache->space_info;
6671 spin_lock(&space_info->lock);
6672 spin_lock(&cache->lock);
6676 cache->reserved += num_bytes;
6677 space_info->bytes_reserved += num_bytes;
6679 trace_btrfs_space_reservation(cache->fs_info,
6680 "space_info", space_info->flags,
6682 space_info->bytes_may_use -= ram_bytes;
6684 cache->delalloc_bytes += num_bytes;
6686 spin_unlock(&cache->lock);
6687 spin_unlock(&space_info->lock);
6692 * btrfs_free_reserved_bytes - update the block_group and space info counters
6693 * @cache: The cache we are manipulating
6694 * @num_bytes: The number of bytes in question
6695 * @delalloc: The blocks are allocated for the delalloc write
6697 * This is called by somebody who is freeing space that was never actually used
6698 * on disk. For example if you reserve some space for a new leaf in transaction
6699 * A and before transaction A commits you free that leaf, you call this with
6700 * reserve set to 0 in order to clear the reservation.
6703 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6704 u64 num_bytes, int delalloc)
6706 struct btrfs_space_info *space_info = cache->space_info;
6709 spin_lock(&space_info->lock);
6710 spin_lock(&cache->lock);
6712 space_info->bytes_readonly += num_bytes;
6713 cache->reserved -= num_bytes;
6714 space_info->bytes_reserved -= num_bytes;
6717 cache->delalloc_bytes -= num_bytes;
6718 spin_unlock(&cache->lock);
6719 spin_unlock(&space_info->lock);
6722 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6724 struct btrfs_caching_control *next;
6725 struct btrfs_caching_control *caching_ctl;
6726 struct btrfs_block_group_cache *cache;
6728 down_write(&fs_info->commit_root_sem);
6730 list_for_each_entry_safe(caching_ctl, next,
6731 &fs_info->caching_block_groups, list) {
6732 cache = caching_ctl->block_group;
6733 if (block_group_cache_done(cache)) {
6734 cache->last_byte_to_unpin = (u64)-1;
6735 list_del_init(&caching_ctl->list);
6736 put_caching_control(caching_ctl);
6738 cache->last_byte_to_unpin = caching_ctl->progress;
6742 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6743 fs_info->pinned_extents = &fs_info->freed_extents[1];
6745 fs_info->pinned_extents = &fs_info->freed_extents[0];
6747 up_write(&fs_info->commit_root_sem);
6749 update_global_block_rsv(fs_info);
6753 * Returns the free cluster for the given space info and sets empty_cluster to
6754 * what it should be based on the mount options.
6756 static struct btrfs_free_cluster *
6757 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6758 struct btrfs_space_info *space_info, u64 *empty_cluster)
6760 struct btrfs_free_cluster *ret = NULL;
6763 if (btrfs_mixed_space_info(space_info))
6766 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6767 ret = &fs_info->meta_alloc_cluster;
6768 if (btrfs_test_opt(fs_info, SSD))
6769 *empty_cluster = SZ_2M;
6771 *empty_cluster = SZ_64K;
6772 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6773 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6774 *empty_cluster = SZ_2M;
6775 ret = &fs_info->data_alloc_cluster;
6781 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6783 const bool return_free_space)
6785 struct btrfs_block_group_cache *cache = NULL;
6786 struct btrfs_space_info *space_info;
6787 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6788 struct btrfs_free_cluster *cluster = NULL;
6790 u64 total_unpinned = 0;
6791 u64 empty_cluster = 0;
6794 while (start <= end) {
6797 start >= cache->key.objectid + cache->key.offset) {
6799 btrfs_put_block_group(cache);
6801 cache = btrfs_lookup_block_group(fs_info, start);
6802 BUG_ON(!cache); /* Logic error */
6804 cluster = fetch_cluster_info(fs_info,
6807 empty_cluster <<= 1;
6810 len = cache->key.objectid + cache->key.offset - start;
6811 len = min(len, end + 1 - start);
6813 if (start < cache->last_byte_to_unpin) {
6814 len = min(len, cache->last_byte_to_unpin - start);
6815 if (return_free_space)
6816 btrfs_add_free_space(cache, start, len);
6820 total_unpinned += len;
6821 space_info = cache->space_info;
6824 * If this space cluster has been marked as fragmented and we've
6825 * unpinned enough in this block group to potentially allow a
6826 * cluster to be created inside of it go ahead and clear the
6829 if (cluster && cluster->fragmented &&
6830 total_unpinned > empty_cluster) {
6831 spin_lock(&cluster->lock);
6832 cluster->fragmented = 0;
6833 spin_unlock(&cluster->lock);
6836 spin_lock(&space_info->lock);
6837 spin_lock(&cache->lock);
6838 cache->pinned -= len;
6839 space_info->bytes_pinned -= len;
6841 trace_btrfs_space_reservation(fs_info, "pinned",
6842 space_info->flags, len, 0);
6843 space_info->max_extent_size = 0;
6844 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6846 space_info->bytes_readonly += len;
6849 spin_unlock(&cache->lock);
6850 if (!readonly && return_free_space &&
6851 global_rsv->space_info == space_info) {
6854 spin_lock(&global_rsv->lock);
6855 if (!global_rsv->full) {
6856 to_add = min(len, global_rsv->size -
6857 global_rsv->reserved);
6858 global_rsv->reserved += to_add;
6859 space_info->bytes_may_use += to_add;
6860 if (global_rsv->reserved >= global_rsv->size)
6861 global_rsv->full = 1;
6862 trace_btrfs_space_reservation(fs_info,
6868 spin_unlock(&global_rsv->lock);
6869 /* Add to any tickets we may have */
6871 space_info_add_new_bytes(fs_info, space_info,
6874 spin_unlock(&space_info->lock);
6878 btrfs_put_block_group(cache);
6882 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6883 struct btrfs_fs_info *fs_info)
6885 struct btrfs_block_group_cache *block_group, *tmp;
6886 struct list_head *deleted_bgs;
6887 struct extent_io_tree *unpin;
6892 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6893 unpin = &fs_info->freed_extents[1];
6895 unpin = &fs_info->freed_extents[0];
6897 while (!trans->aborted) {
6898 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6899 ret = find_first_extent_bit(unpin, 0, &start, &end,
6900 EXTENT_DIRTY, NULL);
6902 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6906 if (btrfs_test_opt(fs_info, DISCARD))
6907 ret = btrfs_discard_extent(fs_info, start,
6908 end + 1 - start, NULL);
6910 clear_extent_dirty(unpin, start, end);
6911 unpin_extent_range(fs_info, start, end, true);
6912 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6917 * Transaction is finished. We don't need the lock anymore. We
6918 * do need to clean up the block groups in case of a transaction
6921 deleted_bgs = &trans->transaction->deleted_bgs;
6922 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6926 if (!trans->aborted)
6927 ret = btrfs_discard_extent(fs_info,
6928 block_group->key.objectid,
6929 block_group->key.offset,
6932 list_del_init(&block_group->bg_list);
6933 btrfs_put_block_group_trimming(block_group);
6934 btrfs_put_block_group(block_group);
6937 const char *errstr = btrfs_decode_error(ret);
6939 "discard failed while removing blockgroup: errno=%d %s",
6947 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6948 struct btrfs_fs_info *info,
6949 struct btrfs_delayed_ref_node *node, u64 parent,
6950 u64 root_objectid, u64 owner_objectid,
6951 u64 owner_offset, int refs_to_drop,
6952 struct btrfs_delayed_extent_op *extent_op)
6954 struct btrfs_key key;
6955 struct btrfs_path *path;
6956 struct btrfs_root *extent_root = info->extent_root;
6957 struct extent_buffer *leaf;
6958 struct btrfs_extent_item *ei;
6959 struct btrfs_extent_inline_ref *iref;
6962 int extent_slot = 0;
6963 int found_extent = 0;
6967 u64 bytenr = node->bytenr;
6968 u64 num_bytes = node->num_bytes;
6970 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6972 path = btrfs_alloc_path();
6976 path->reada = READA_FORWARD;
6977 path->leave_spinning = 1;
6979 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6980 BUG_ON(!is_data && refs_to_drop != 1);
6983 skinny_metadata = 0;
6985 ret = lookup_extent_backref(trans, info, path, &iref,
6986 bytenr, num_bytes, parent,
6987 root_objectid, owner_objectid,
6990 extent_slot = path->slots[0];
6991 while (extent_slot >= 0) {
6992 btrfs_item_key_to_cpu(path->nodes[0], &key,
6994 if (key.objectid != bytenr)
6996 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6997 key.offset == num_bytes) {
7001 if (key.type == BTRFS_METADATA_ITEM_KEY &&
7002 key.offset == owner_objectid) {
7006 if (path->slots[0] - extent_slot > 5)
7010 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7011 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
7012 if (found_extent && item_size < sizeof(*ei))
7015 if (!found_extent) {
7017 ret = remove_extent_backref(trans, info, path, NULL,
7019 is_data, &last_ref);
7021 btrfs_abort_transaction(trans, ret);
7024 btrfs_release_path(path);
7025 path->leave_spinning = 1;
7027 key.objectid = bytenr;
7028 key.type = BTRFS_EXTENT_ITEM_KEY;
7029 key.offset = num_bytes;
7031 if (!is_data && skinny_metadata) {
7032 key.type = BTRFS_METADATA_ITEM_KEY;
7033 key.offset = owner_objectid;
7036 ret = btrfs_search_slot(trans, extent_root,
7038 if (ret > 0 && skinny_metadata && path->slots[0]) {
7040 * Couldn't find our skinny metadata item,
7041 * see if we have ye olde extent item.
7044 btrfs_item_key_to_cpu(path->nodes[0], &key,
7046 if (key.objectid == bytenr &&
7047 key.type == BTRFS_EXTENT_ITEM_KEY &&
7048 key.offset == num_bytes)
7052 if (ret > 0 && skinny_metadata) {
7053 skinny_metadata = false;
7054 key.objectid = bytenr;
7055 key.type = BTRFS_EXTENT_ITEM_KEY;
7056 key.offset = num_bytes;
7057 btrfs_release_path(path);
7058 ret = btrfs_search_slot(trans, extent_root,
7064 "umm, got %d back from search, was looking for %llu",
7067 btrfs_print_leaf(path->nodes[0]);
7070 btrfs_abort_transaction(trans, ret);
7073 extent_slot = path->slots[0];
7075 } else if (WARN_ON(ret == -ENOENT)) {
7076 btrfs_print_leaf(path->nodes[0]);
7078 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7079 bytenr, parent, root_objectid, owner_objectid,
7081 btrfs_abort_transaction(trans, ret);
7084 btrfs_abort_transaction(trans, ret);
7088 leaf = path->nodes[0];
7089 item_size = btrfs_item_size_nr(leaf, extent_slot);
7090 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7091 if (item_size < sizeof(*ei)) {
7092 BUG_ON(found_extent || extent_slot != path->slots[0]);
7093 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
7096 btrfs_abort_transaction(trans, ret);
7100 btrfs_release_path(path);
7101 path->leave_spinning = 1;
7103 key.objectid = bytenr;
7104 key.type = BTRFS_EXTENT_ITEM_KEY;
7105 key.offset = num_bytes;
7107 ret = btrfs_search_slot(trans, extent_root, &key, path,
7111 "umm, got %d back from search, was looking for %llu",
7113 btrfs_print_leaf(path->nodes[0]);
7116 btrfs_abort_transaction(trans, ret);
7120 extent_slot = path->slots[0];
7121 leaf = path->nodes[0];
7122 item_size = btrfs_item_size_nr(leaf, extent_slot);
7125 BUG_ON(item_size < sizeof(*ei));
7126 ei = btrfs_item_ptr(leaf, extent_slot,
7127 struct btrfs_extent_item);
7128 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7129 key.type == BTRFS_EXTENT_ITEM_KEY) {
7130 struct btrfs_tree_block_info *bi;
7131 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7132 bi = (struct btrfs_tree_block_info *)(ei + 1);
7133 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7136 refs = btrfs_extent_refs(leaf, ei);
7137 if (refs < refs_to_drop) {
7139 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7140 refs_to_drop, refs, bytenr);
7142 btrfs_abort_transaction(trans, ret);
7145 refs -= refs_to_drop;
7149 __run_delayed_extent_op(extent_op, leaf, ei);
7151 * In the case of inline back ref, reference count will
7152 * be updated by remove_extent_backref
7155 BUG_ON(!found_extent);
7157 btrfs_set_extent_refs(leaf, ei, refs);
7158 btrfs_mark_buffer_dirty(leaf);
7161 ret = remove_extent_backref(trans, info, path,
7163 is_data, &last_ref);
7165 btrfs_abort_transaction(trans, ret);
7171 BUG_ON(is_data && refs_to_drop !=
7172 extent_data_ref_count(path, iref));
7174 BUG_ON(path->slots[0] != extent_slot);
7176 BUG_ON(path->slots[0] != extent_slot + 1);
7177 path->slots[0] = extent_slot;
7183 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7186 btrfs_abort_transaction(trans, ret);
7189 btrfs_release_path(path);
7192 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7194 btrfs_abort_transaction(trans, ret);
7199 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7201 btrfs_abort_transaction(trans, ret);
7205 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7207 btrfs_abort_transaction(trans, ret);
7211 btrfs_release_path(path);
7214 btrfs_free_path(path);
7219 * when we free an block, it is possible (and likely) that we free the last
7220 * delayed ref for that extent as well. This searches the delayed ref tree for
7221 * a given extent, and if there are no other delayed refs to be processed, it
7222 * removes it from the tree.
7224 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7227 struct btrfs_delayed_ref_head *head;
7228 struct btrfs_delayed_ref_root *delayed_refs;
7231 delayed_refs = &trans->transaction->delayed_refs;
7232 spin_lock(&delayed_refs->lock);
7233 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7235 goto out_delayed_unlock;
7237 spin_lock(&head->lock);
7238 if (!list_empty(&head->ref_list))
7241 if (head->extent_op) {
7242 if (!head->must_insert_reserved)
7244 btrfs_free_delayed_extent_op(head->extent_op);
7245 head->extent_op = NULL;
7249 * waiting for the lock here would deadlock. If someone else has it
7250 * locked they are already in the process of dropping it anyway
7252 if (!mutex_trylock(&head->mutex))
7256 * at this point we have a head with no other entries. Go
7257 * ahead and process it.
7259 head->node.in_tree = 0;
7260 rb_erase(&head->href_node, &delayed_refs->href_root);
7262 atomic_dec(&delayed_refs->num_entries);
7265 * we don't take a ref on the node because we're removing it from the
7266 * tree, so we just steal the ref the tree was holding.
7268 delayed_refs->num_heads--;
7269 if (head->processing == 0)
7270 delayed_refs->num_heads_ready--;
7271 head->processing = 0;
7272 spin_unlock(&head->lock);
7273 spin_unlock(&delayed_refs->lock);
7275 BUG_ON(head->extent_op);
7276 if (head->must_insert_reserved)
7279 mutex_unlock(&head->mutex);
7280 btrfs_put_delayed_ref(&head->node);
7283 spin_unlock(&head->lock);
7286 spin_unlock(&delayed_refs->lock);
7290 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7291 struct btrfs_root *root,
7292 struct extent_buffer *buf,
7293 u64 parent, int last_ref)
7295 struct btrfs_fs_info *fs_info = root->fs_info;
7299 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7300 int old_ref_mod, new_ref_mod;
7302 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7304 root->root_key.objectid,
7305 btrfs_header_level(buf),
7306 BTRFS_DROP_DELAYED_REF, NULL,
7307 &old_ref_mod, &new_ref_mod);
7308 BUG_ON(ret); /* -ENOMEM */
7309 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7312 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7313 struct btrfs_block_group_cache *cache;
7315 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7316 ret = check_ref_cleanup(trans, buf->start);
7322 cache = btrfs_lookup_block_group(fs_info, buf->start);
7324 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7325 pin_down_extent(fs_info, cache, buf->start,
7327 btrfs_put_block_group(cache);
7331 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7333 btrfs_add_free_space(cache, buf->start, buf->len);
7334 btrfs_free_reserved_bytes(cache, buf->len, 0);
7335 btrfs_put_block_group(cache);
7336 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7340 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7341 root->root_key.objectid);
7345 * Deleting the buffer, clear the corrupt flag since it doesn't
7348 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7352 /* Can return -ENOMEM */
7353 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7354 struct btrfs_fs_info *fs_info,
7355 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7356 u64 owner, u64 offset)
7358 int old_ref_mod, new_ref_mod;
7361 if (btrfs_is_testing(fs_info))
7366 * tree log blocks never actually go into the extent allocation
7367 * tree, just update pinning info and exit early.
7369 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7370 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7371 /* unlocks the pinned mutex */
7372 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7373 old_ref_mod = new_ref_mod = 0;
7375 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7376 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7378 root_objectid, (int)owner,
7379 BTRFS_DROP_DELAYED_REF, NULL,
7380 &old_ref_mod, &new_ref_mod);
7382 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7384 root_objectid, owner, offset,
7385 0, BTRFS_DROP_DELAYED_REF,
7386 &old_ref_mod, &new_ref_mod);
7389 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7390 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7396 * when we wait for progress in the block group caching, its because
7397 * our allocation attempt failed at least once. So, we must sleep
7398 * and let some progress happen before we try again.
7400 * This function will sleep at least once waiting for new free space to
7401 * show up, and then it will check the block group free space numbers
7402 * for our min num_bytes. Another option is to have it go ahead
7403 * and look in the rbtree for a free extent of a given size, but this
7406 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7407 * any of the information in this block group.
7409 static noinline void
7410 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7413 struct btrfs_caching_control *caching_ctl;
7415 caching_ctl = get_caching_control(cache);
7419 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7420 (cache->free_space_ctl->free_space >= num_bytes));
7422 put_caching_control(caching_ctl);
7426 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7428 struct btrfs_caching_control *caching_ctl;
7431 caching_ctl = get_caching_control(cache);
7433 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7435 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7436 if (cache->cached == BTRFS_CACHE_ERROR)
7438 put_caching_control(caching_ctl);
7442 int __get_raid_index(u64 flags)
7444 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7445 return BTRFS_RAID_RAID10;
7446 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7447 return BTRFS_RAID_RAID1;
7448 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7449 return BTRFS_RAID_DUP;
7450 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7451 return BTRFS_RAID_RAID0;
7452 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7453 return BTRFS_RAID_RAID5;
7454 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7455 return BTRFS_RAID_RAID6;
7457 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7460 int get_block_group_index(struct btrfs_block_group_cache *cache)
7462 return __get_raid_index(cache->flags);
7465 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7466 [BTRFS_RAID_RAID10] = "raid10",
7467 [BTRFS_RAID_RAID1] = "raid1",
7468 [BTRFS_RAID_DUP] = "dup",
7469 [BTRFS_RAID_RAID0] = "raid0",
7470 [BTRFS_RAID_SINGLE] = "single",
7471 [BTRFS_RAID_RAID5] = "raid5",
7472 [BTRFS_RAID_RAID6] = "raid6",
7475 static const char *get_raid_name(enum btrfs_raid_types type)
7477 if (type >= BTRFS_NR_RAID_TYPES)
7480 return btrfs_raid_type_names[type];
7483 enum btrfs_loop_type {
7484 LOOP_CACHING_NOWAIT = 0,
7485 LOOP_CACHING_WAIT = 1,
7486 LOOP_ALLOC_CHUNK = 2,
7487 LOOP_NO_EMPTY_SIZE = 3,
7491 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7495 down_read(&cache->data_rwsem);
7499 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7502 btrfs_get_block_group(cache);
7504 down_read(&cache->data_rwsem);
7507 static struct btrfs_block_group_cache *
7508 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7509 struct btrfs_free_cluster *cluster,
7512 struct btrfs_block_group_cache *used_bg = NULL;
7514 spin_lock(&cluster->refill_lock);
7516 used_bg = cluster->block_group;
7520 if (used_bg == block_group)
7523 btrfs_get_block_group(used_bg);
7528 if (down_read_trylock(&used_bg->data_rwsem))
7531 spin_unlock(&cluster->refill_lock);
7533 /* We should only have one-level nested. */
7534 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7536 spin_lock(&cluster->refill_lock);
7537 if (used_bg == cluster->block_group)
7540 up_read(&used_bg->data_rwsem);
7541 btrfs_put_block_group(used_bg);
7546 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7550 up_read(&cache->data_rwsem);
7551 btrfs_put_block_group(cache);
7555 * walks the btree of allocated extents and find a hole of a given size.
7556 * The key ins is changed to record the hole:
7557 * ins->objectid == start position
7558 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7559 * ins->offset == the size of the hole.
7560 * Any available blocks before search_start are skipped.
7562 * If there is no suitable free space, we will record the max size of
7563 * the free space extent currently.
7565 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7566 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7567 u64 hint_byte, struct btrfs_key *ins,
7568 u64 flags, int delalloc)
7571 struct btrfs_root *root = fs_info->extent_root;
7572 struct btrfs_free_cluster *last_ptr = NULL;
7573 struct btrfs_block_group_cache *block_group = NULL;
7574 u64 search_start = 0;
7575 u64 max_extent_size = 0;
7576 u64 max_free_space = 0;
7577 u64 empty_cluster = 0;
7578 struct btrfs_space_info *space_info;
7580 int index = __get_raid_index(flags);
7581 bool failed_cluster_refill = false;
7582 bool failed_alloc = false;
7583 bool use_cluster = true;
7584 bool have_caching_bg = false;
7585 bool orig_have_caching_bg = false;
7586 bool full_search = false;
7588 WARN_ON(num_bytes < fs_info->sectorsize);
7589 ins->type = BTRFS_EXTENT_ITEM_KEY;
7593 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7595 space_info = __find_space_info(fs_info, flags);
7597 btrfs_err(fs_info, "No space info for %llu", flags);
7602 * If our free space is heavily fragmented we may not be able to make
7603 * big contiguous allocations, so instead of doing the expensive search
7604 * for free space, simply return ENOSPC with our max_extent_size so we
7605 * can go ahead and search for a more manageable chunk.
7607 * If our max_extent_size is large enough for our allocation simply
7608 * disable clustering since we will likely not be able to find enough
7609 * space to create a cluster and induce latency trying.
7611 if (unlikely(space_info->max_extent_size)) {
7612 spin_lock(&space_info->lock);
7613 if (space_info->max_extent_size &&
7614 num_bytes > space_info->max_extent_size) {
7615 ins->offset = space_info->max_extent_size;
7616 spin_unlock(&space_info->lock);
7618 } else if (space_info->max_extent_size) {
7619 use_cluster = false;
7621 spin_unlock(&space_info->lock);
7624 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7626 spin_lock(&last_ptr->lock);
7627 if (last_ptr->block_group)
7628 hint_byte = last_ptr->window_start;
7629 if (last_ptr->fragmented) {
7631 * We still set window_start so we can keep track of the
7632 * last place we found an allocation to try and save
7635 hint_byte = last_ptr->window_start;
7636 use_cluster = false;
7638 spin_unlock(&last_ptr->lock);
7641 search_start = max(search_start, first_logical_byte(fs_info, 0));
7642 search_start = max(search_start, hint_byte);
7643 if (search_start == hint_byte) {
7644 block_group = btrfs_lookup_block_group(fs_info, search_start);
7646 * we don't want to use the block group if it doesn't match our
7647 * allocation bits, or if its not cached.
7649 * However if we are re-searching with an ideal block group
7650 * picked out then we don't care that the block group is cached.
7652 if (block_group && block_group_bits(block_group, flags) &&
7653 block_group->cached != BTRFS_CACHE_NO) {
7654 down_read(&space_info->groups_sem);
7655 if (list_empty(&block_group->list) ||
7658 * someone is removing this block group,
7659 * we can't jump into the have_block_group
7660 * target because our list pointers are not
7663 btrfs_put_block_group(block_group);
7664 up_read(&space_info->groups_sem);
7666 index = get_block_group_index(block_group);
7667 btrfs_lock_block_group(block_group, delalloc);
7668 goto have_block_group;
7670 } else if (block_group) {
7671 btrfs_put_block_group(block_group);
7675 have_caching_bg = false;
7676 if (index == 0 || index == __get_raid_index(flags))
7678 down_read(&space_info->groups_sem);
7679 list_for_each_entry(block_group, &space_info->block_groups[index],
7684 /* If the block group is read-only, we can skip it entirely. */
7685 if (unlikely(block_group->ro))
7688 btrfs_grab_block_group(block_group, delalloc);
7689 search_start = block_group->key.objectid;
7692 * this can happen if we end up cycling through all the
7693 * raid types, but we want to make sure we only allocate
7694 * for the proper type.
7696 if (!block_group_bits(block_group, flags)) {
7697 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7698 BTRFS_BLOCK_GROUP_RAID1 |
7699 BTRFS_BLOCK_GROUP_RAID5 |
7700 BTRFS_BLOCK_GROUP_RAID6 |
7701 BTRFS_BLOCK_GROUP_RAID10;
7704 * if they asked for extra copies and this block group
7705 * doesn't provide them, bail. This does allow us to
7706 * fill raid0 from raid1.
7708 if ((flags & extra) && !(block_group->flags & extra))
7713 cached = block_group_cache_done(block_group);
7714 if (unlikely(!cached)) {
7715 have_caching_bg = true;
7716 ret = cache_block_group(block_group, 0);
7721 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7725 * Ok we want to try and use the cluster allocator, so
7728 if (last_ptr && use_cluster) {
7729 struct btrfs_block_group_cache *used_block_group;
7730 unsigned long aligned_cluster;
7732 * the refill lock keeps out other
7733 * people trying to start a new cluster
7735 used_block_group = btrfs_lock_cluster(block_group,
7738 if (!used_block_group)
7739 goto refill_cluster;
7741 if (used_block_group != block_group &&
7742 (used_block_group->ro ||
7743 !block_group_bits(used_block_group, flags)))
7744 goto release_cluster;
7746 offset = btrfs_alloc_from_cluster(used_block_group,
7749 used_block_group->key.objectid,
7752 /* we have a block, we're done */
7753 spin_unlock(&last_ptr->refill_lock);
7754 trace_btrfs_reserve_extent_cluster(fs_info,
7756 search_start, num_bytes);
7757 if (used_block_group != block_group) {
7758 btrfs_release_block_group(block_group,
7760 block_group = used_block_group;
7765 WARN_ON(last_ptr->block_group != used_block_group);
7767 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7768 * set up a new clusters, so lets just skip it
7769 * and let the allocator find whatever block
7770 * it can find. If we reach this point, we
7771 * will have tried the cluster allocator
7772 * plenty of times and not have found
7773 * anything, so we are likely way too
7774 * fragmented for the clustering stuff to find
7777 * However, if the cluster is taken from the
7778 * current block group, release the cluster
7779 * first, so that we stand a better chance of
7780 * succeeding in the unclustered
7782 if (loop >= LOOP_NO_EMPTY_SIZE &&
7783 used_block_group != block_group) {
7784 spin_unlock(&last_ptr->refill_lock);
7785 btrfs_release_block_group(used_block_group,
7787 goto unclustered_alloc;
7791 * this cluster didn't work out, free it and
7794 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7796 if (used_block_group != block_group)
7797 btrfs_release_block_group(used_block_group,
7800 if (loop >= LOOP_NO_EMPTY_SIZE) {
7801 spin_unlock(&last_ptr->refill_lock);
7802 goto unclustered_alloc;
7805 aligned_cluster = max_t(unsigned long,
7806 empty_cluster + empty_size,
7807 block_group->full_stripe_len);
7809 /* allocate a cluster in this block group */
7810 ret = btrfs_find_space_cluster(fs_info, block_group,
7811 last_ptr, search_start,
7816 * now pull our allocation out of this
7819 offset = btrfs_alloc_from_cluster(block_group,
7825 /* we found one, proceed */
7826 spin_unlock(&last_ptr->refill_lock);
7827 trace_btrfs_reserve_extent_cluster(fs_info,
7828 block_group, search_start,
7832 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7833 && !failed_cluster_refill) {
7834 spin_unlock(&last_ptr->refill_lock);
7836 failed_cluster_refill = true;
7837 wait_block_group_cache_progress(block_group,
7838 num_bytes + empty_cluster + empty_size);
7839 goto have_block_group;
7843 * at this point we either didn't find a cluster
7844 * or we weren't able to allocate a block from our
7845 * cluster. Free the cluster we've been trying
7846 * to use, and go to the next block group
7848 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7849 spin_unlock(&last_ptr->refill_lock);
7855 * We are doing an unclustered alloc, set the fragmented flag so
7856 * we don't bother trying to setup a cluster again until we get
7859 if (unlikely(last_ptr)) {
7860 spin_lock(&last_ptr->lock);
7861 last_ptr->fragmented = 1;
7862 spin_unlock(&last_ptr->lock);
7865 struct btrfs_free_space_ctl *ctl =
7866 block_group->free_space_ctl;
7868 spin_lock(&ctl->tree_lock);
7869 if (ctl->free_space <
7870 num_bytes + empty_cluster + empty_size) {
7871 max_free_space = max(max_free_space,
7873 spin_unlock(&ctl->tree_lock);
7876 spin_unlock(&ctl->tree_lock);
7879 offset = btrfs_find_space_for_alloc(block_group, search_start,
7880 num_bytes, empty_size,
7883 * If we didn't find a chunk, and we haven't failed on this
7884 * block group before, and this block group is in the middle of
7885 * caching and we are ok with waiting, then go ahead and wait
7886 * for progress to be made, and set failed_alloc to true.
7888 * If failed_alloc is true then we've already waited on this
7889 * block group once and should move on to the next block group.
7891 if (!offset && !failed_alloc && !cached &&
7892 loop > LOOP_CACHING_NOWAIT) {
7893 wait_block_group_cache_progress(block_group,
7894 num_bytes + empty_size);
7895 failed_alloc = true;
7896 goto have_block_group;
7897 } else if (!offset) {
7901 search_start = ALIGN(offset, fs_info->stripesize);
7903 /* move on to the next group */
7904 if (search_start + num_bytes >
7905 block_group->key.objectid + block_group->key.offset) {
7906 btrfs_add_free_space(block_group, offset, num_bytes);
7910 if (offset < search_start)
7911 btrfs_add_free_space(block_group, offset,
7912 search_start - offset);
7913 BUG_ON(offset > search_start);
7915 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7916 num_bytes, delalloc);
7917 if (ret == -EAGAIN) {
7918 btrfs_add_free_space(block_group, offset, num_bytes);
7921 btrfs_inc_block_group_reservations(block_group);
7923 /* we are all good, lets return */
7924 ins->objectid = search_start;
7925 ins->offset = num_bytes;
7927 trace_btrfs_reserve_extent(fs_info, block_group,
7928 search_start, num_bytes);
7929 btrfs_release_block_group(block_group, delalloc);
7932 failed_cluster_refill = false;
7933 failed_alloc = false;
7934 BUG_ON(index != get_block_group_index(block_group));
7935 btrfs_release_block_group(block_group, delalloc);
7938 up_read(&space_info->groups_sem);
7940 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7941 && !orig_have_caching_bg)
7942 orig_have_caching_bg = true;
7944 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7947 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7951 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7952 * caching kthreads as we move along
7953 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7954 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7955 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7958 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7960 if (loop == LOOP_CACHING_NOWAIT) {
7962 * We want to skip the LOOP_CACHING_WAIT step if we
7963 * don't have any uncached bgs and we've already done a
7964 * full search through.
7966 if (orig_have_caching_bg || !full_search)
7967 loop = LOOP_CACHING_WAIT;
7969 loop = LOOP_ALLOC_CHUNK;
7974 if (loop == LOOP_ALLOC_CHUNK) {
7975 struct btrfs_trans_handle *trans;
7978 trans = current->journal_info;
7982 trans = btrfs_join_transaction(root);
7984 if (IS_ERR(trans)) {
7985 ret = PTR_ERR(trans);
7989 ret = do_chunk_alloc(trans, fs_info, flags,
7993 * If we can't allocate a new chunk we've already looped
7994 * through at least once, move on to the NO_EMPTY_SIZE
7998 loop = LOOP_NO_EMPTY_SIZE;
8001 * Do not bail out on ENOSPC since we
8002 * can do more things.
8004 if (ret < 0 && ret != -ENOSPC)
8005 btrfs_abort_transaction(trans, ret);
8009 btrfs_end_transaction(trans);
8014 if (loop == LOOP_NO_EMPTY_SIZE) {
8016 * Don't loop again if we already have no empty_size and
8019 if (empty_size == 0 &&
8020 empty_cluster == 0) {
8029 } else if (!ins->objectid) {
8031 } else if (ins->objectid) {
8032 if (!use_cluster && last_ptr) {
8033 spin_lock(&last_ptr->lock);
8034 last_ptr->window_start = ins->objectid;
8035 spin_unlock(&last_ptr->lock);
8040 if (ret == -ENOSPC) {
8041 if (!max_extent_size)
8042 max_extent_size = max_free_space;
8043 spin_lock(&space_info->lock);
8044 space_info->max_extent_size = max_extent_size;
8045 spin_unlock(&space_info->lock);
8046 ins->offset = max_extent_size;
8051 static void dump_space_info(struct btrfs_fs_info *fs_info,
8052 struct btrfs_space_info *info, u64 bytes,
8053 int dump_block_groups)
8055 struct btrfs_block_group_cache *cache;
8058 spin_lock(&info->lock);
8059 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
8061 info->total_bytes - btrfs_space_info_used(info, true),
8062 info->full ? "" : "not ");
8064 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
8065 info->total_bytes, info->bytes_used, info->bytes_pinned,
8066 info->bytes_reserved, info->bytes_may_use,
8067 info->bytes_readonly);
8068 spin_unlock(&info->lock);
8070 if (!dump_block_groups)
8073 down_read(&info->groups_sem);
8075 list_for_each_entry(cache, &info->block_groups[index], list) {
8076 spin_lock(&cache->lock);
8078 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8079 cache->key.objectid, cache->key.offset,
8080 btrfs_block_group_used(&cache->item), cache->pinned,
8081 cache->reserved, cache->ro ? "[readonly]" : "");
8082 btrfs_dump_free_space(cache, bytes);
8083 spin_unlock(&cache->lock);
8085 if (++index < BTRFS_NR_RAID_TYPES)
8087 up_read(&info->groups_sem);
8090 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8091 u64 num_bytes, u64 min_alloc_size,
8092 u64 empty_size, u64 hint_byte,
8093 struct btrfs_key *ins, int is_data, int delalloc)
8095 struct btrfs_fs_info *fs_info = root->fs_info;
8096 bool final_tried = num_bytes == min_alloc_size;
8100 flags = get_alloc_profile_by_root(root, is_data);
8102 WARN_ON(num_bytes < fs_info->sectorsize);
8103 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8104 hint_byte, ins, flags, delalloc);
8105 if (!ret && !is_data) {
8106 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8107 } else if (ret == -ENOSPC) {
8108 if (!final_tried && ins->offset) {
8109 num_bytes = min(num_bytes >> 1, ins->offset);
8110 num_bytes = round_down(num_bytes,
8111 fs_info->sectorsize);
8112 num_bytes = max(num_bytes, min_alloc_size);
8113 ram_bytes = num_bytes;
8114 if (num_bytes == min_alloc_size)
8117 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8118 struct btrfs_space_info *sinfo;
8120 sinfo = __find_space_info(fs_info, flags);
8122 "allocation failed flags %llu, wanted %llu",
8125 dump_space_info(fs_info, sinfo, num_bytes, 1);
8132 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8134 int pin, int delalloc)
8136 struct btrfs_block_group_cache *cache;
8139 cache = btrfs_lookup_block_group(fs_info, start);
8141 btrfs_err(fs_info, "Unable to find block group for %llu",
8147 pin_down_extent(fs_info, cache, start, len, 1);
8149 if (btrfs_test_opt(fs_info, DISCARD))
8150 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8151 btrfs_add_free_space(cache, start, len);
8152 btrfs_free_reserved_bytes(cache, len, delalloc);
8153 trace_btrfs_reserved_extent_free(fs_info, start, len);
8156 btrfs_put_block_group(cache);
8160 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8161 u64 start, u64 len, int delalloc)
8163 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8166 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8169 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8172 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8173 struct btrfs_fs_info *fs_info,
8174 u64 parent, u64 root_objectid,
8175 u64 flags, u64 owner, u64 offset,
8176 struct btrfs_key *ins, int ref_mod)
8179 struct btrfs_extent_item *extent_item;
8180 struct btrfs_extent_inline_ref *iref;
8181 struct btrfs_path *path;
8182 struct extent_buffer *leaf;
8187 type = BTRFS_SHARED_DATA_REF_KEY;
8189 type = BTRFS_EXTENT_DATA_REF_KEY;
8191 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8193 path = btrfs_alloc_path();
8197 path->leave_spinning = 1;
8198 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8201 btrfs_free_path(path);
8205 leaf = path->nodes[0];
8206 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8207 struct btrfs_extent_item);
8208 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8209 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8210 btrfs_set_extent_flags(leaf, extent_item,
8211 flags | BTRFS_EXTENT_FLAG_DATA);
8213 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8214 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8216 struct btrfs_shared_data_ref *ref;
8217 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8218 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8219 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8221 struct btrfs_extent_data_ref *ref;
8222 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8223 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8224 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8225 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8226 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8229 btrfs_mark_buffer_dirty(path->nodes[0]);
8230 btrfs_free_path(path);
8232 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8237 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8238 if (ret) { /* -ENOENT, logic error */
8239 btrfs_err(fs_info, "update block group failed for %llu %llu",
8240 ins->objectid, ins->offset);
8243 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8247 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8248 struct btrfs_fs_info *fs_info,
8249 u64 parent, u64 root_objectid,
8250 u64 flags, struct btrfs_disk_key *key,
8251 int level, struct btrfs_key *ins)
8254 struct btrfs_extent_item *extent_item;
8255 struct btrfs_tree_block_info *block_info;
8256 struct btrfs_extent_inline_ref *iref;
8257 struct btrfs_path *path;
8258 struct extent_buffer *leaf;
8259 u32 size = sizeof(*extent_item) + sizeof(*iref);
8260 u64 num_bytes = ins->offset;
8261 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8263 if (!skinny_metadata)
8264 size += sizeof(*block_info);
8266 path = btrfs_alloc_path();
8268 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8273 path->leave_spinning = 1;
8274 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8277 btrfs_free_path(path);
8278 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8283 leaf = path->nodes[0];
8284 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8285 struct btrfs_extent_item);
8286 btrfs_set_extent_refs(leaf, extent_item, 1);
8287 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8288 btrfs_set_extent_flags(leaf, extent_item,
8289 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8291 if (skinny_metadata) {
8292 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8293 num_bytes = fs_info->nodesize;
8295 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8296 btrfs_set_tree_block_key(leaf, block_info, key);
8297 btrfs_set_tree_block_level(leaf, block_info, level);
8298 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8302 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8303 btrfs_set_extent_inline_ref_type(leaf, iref,
8304 BTRFS_SHARED_BLOCK_REF_KEY);
8305 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8307 btrfs_set_extent_inline_ref_type(leaf, iref,
8308 BTRFS_TREE_BLOCK_REF_KEY);
8309 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8312 btrfs_mark_buffer_dirty(leaf);
8313 btrfs_free_path(path);
8315 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8320 ret = update_block_group(trans, fs_info, ins->objectid,
8321 fs_info->nodesize, 1);
8322 if (ret) { /* -ENOENT, logic error */
8323 btrfs_err(fs_info, "update block group failed for %llu %llu",
8324 ins->objectid, ins->offset);
8328 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8333 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8334 u64 root_objectid, u64 owner,
8335 u64 offset, u64 ram_bytes,
8336 struct btrfs_key *ins)
8338 struct btrfs_fs_info *fs_info = trans->fs_info;
8341 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8343 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8344 ins->offset, 0, root_objectid, owner,
8346 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8351 * this is used by the tree logging recovery code. It records that
8352 * an extent has been allocated and makes sure to clear the free
8353 * space cache bits as well
8355 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8356 struct btrfs_fs_info *fs_info,
8357 u64 root_objectid, u64 owner, u64 offset,
8358 struct btrfs_key *ins)
8361 struct btrfs_block_group_cache *block_group;
8362 struct btrfs_space_info *space_info;
8365 * Mixed block groups will exclude before processing the log so we only
8366 * need to do the exclude dance if this fs isn't mixed.
8368 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8369 ret = __exclude_logged_extent(fs_info, ins->objectid,
8375 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8379 space_info = block_group->space_info;
8380 spin_lock(&space_info->lock);
8381 spin_lock(&block_group->lock);
8382 space_info->bytes_reserved += ins->offset;
8383 block_group->reserved += ins->offset;
8384 spin_unlock(&block_group->lock);
8385 spin_unlock(&space_info->lock);
8387 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8388 0, owner, offset, ins, 1);
8389 btrfs_put_block_group(block_group);
8393 static struct extent_buffer *
8394 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8395 u64 bytenr, int level)
8397 struct btrfs_fs_info *fs_info = root->fs_info;
8398 struct extent_buffer *buf;
8400 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8405 * Extra safety check in case the extent tree is corrupted and extent
8406 * allocator chooses to use a tree block which is already used and
8409 if (buf->lock_owner == current->pid) {
8410 btrfs_err_rl(fs_info,
8411 "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected",
8412 buf->start, btrfs_header_owner(buf), current->pid);
8413 free_extent_buffer(buf);
8414 return ERR_PTR(-EUCLEAN);
8417 btrfs_set_header_generation(buf, trans->transid);
8418 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8419 btrfs_tree_lock(buf);
8420 clean_tree_block(fs_info, buf);
8421 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8423 btrfs_set_lock_blocking(buf);
8424 set_extent_buffer_uptodate(buf);
8426 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8427 buf->log_index = root->log_transid % 2;
8429 * we allow two log transactions at a time, use different
8430 * EXENT bit to differentiate dirty pages.
8432 if (buf->log_index == 0)
8433 set_extent_dirty(&root->dirty_log_pages, buf->start,
8434 buf->start + buf->len - 1, GFP_NOFS);
8436 set_extent_new(&root->dirty_log_pages, buf->start,
8437 buf->start + buf->len - 1);
8439 buf->log_index = -1;
8440 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8441 buf->start + buf->len - 1, GFP_NOFS);
8443 trans->dirty = true;
8444 /* this returns a buffer locked for blocking */
8448 static struct btrfs_block_rsv *
8449 use_block_rsv(struct btrfs_trans_handle *trans,
8450 struct btrfs_root *root, u32 blocksize)
8452 struct btrfs_fs_info *fs_info = root->fs_info;
8453 struct btrfs_block_rsv *block_rsv;
8454 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8456 bool global_updated = false;
8458 block_rsv = get_block_rsv(trans, root);
8460 if (unlikely(block_rsv->size == 0))
8463 ret = block_rsv_use_bytes(block_rsv, blocksize);
8467 if (block_rsv->failfast)
8468 return ERR_PTR(ret);
8470 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8471 global_updated = true;
8472 update_global_block_rsv(fs_info);
8476 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8477 static DEFINE_RATELIMIT_STATE(_rs,
8478 DEFAULT_RATELIMIT_INTERVAL * 10,
8479 /*DEFAULT_RATELIMIT_BURST*/ 1);
8480 if (__ratelimit(&_rs))
8482 "BTRFS: block rsv returned %d\n", ret);
8485 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8486 BTRFS_RESERVE_NO_FLUSH);
8490 * If we couldn't reserve metadata bytes try and use some from
8491 * the global reserve if its space type is the same as the global
8494 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8495 block_rsv->space_info == global_rsv->space_info) {
8496 ret = block_rsv_use_bytes(global_rsv, blocksize);
8500 return ERR_PTR(ret);
8503 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8504 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8506 block_rsv_add_bytes(block_rsv, blocksize, 0);
8507 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8511 * finds a free extent and does all the dirty work required for allocation
8512 * returns the tree buffer or an ERR_PTR on error.
8514 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8515 struct btrfs_root *root,
8516 u64 parent, u64 root_objectid,
8517 const struct btrfs_disk_key *key,
8518 int level, u64 hint,
8521 struct btrfs_fs_info *fs_info = root->fs_info;
8522 struct btrfs_key ins;
8523 struct btrfs_block_rsv *block_rsv;
8524 struct extent_buffer *buf;
8525 struct btrfs_delayed_extent_op *extent_op;
8528 u32 blocksize = fs_info->nodesize;
8529 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8531 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8532 if (btrfs_is_testing(fs_info)) {
8533 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8536 root->alloc_bytenr += blocksize;
8541 block_rsv = use_block_rsv(trans, root, blocksize);
8542 if (IS_ERR(block_rsv))
8543 return ERR_CAST(block_rsv);
8545 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8546 empty_size, hint, &ins, 0, 0);
8550 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8553 goto out_free_reserved;
8556 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8558 parent = ins.objectid;
8559 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8563 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8564 extent_op = btrfs_alloc_delayed_extent_op();
8570 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8572 memset(&extent_op->key, 0, sizeof(extent_op->key));
8573 extent_op->flags_to_set = flags;
8574 extent_op->update_key = skinny_metadata ? false : true;
8575 extent_op->update_flags = true;
8576 extent_op->is_data = false;
8577 extent_op->level = level;
8579 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8581 root_objectid, level,
8582 BTRFS_ADD_DELAYED_EXTENT,
8583 extent_op, NULL, NULL);
8585 goto out_free_delayed;
8590 btrfs_free_delayed_extent_op(extent_op);
8592 free_extent_buffer(buf);
8594 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8596 unuse_block_rsv(fs_info, block_rsv, blocksize);
8597 return ERR_PTR(ret);
8600 struct walk_control {
8601 u64 refs[BTRFS_MAX_LEVEL];
8602 u64 flags[BTRFS_MAX_LEVEL];
8603 struct btrfs_key update_progress;
8614 #define DROP_REFERENCE 1
8615 #define UPDATE_BACKREF 2
8617 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8618 struct btrfs_root *root,
8619 struct walk_control *wc,
8620 struct btrfs_path *path)
8622 struct btrfs_fs_info *fs_info = root->fs_info;
8628 struct btrfs_key key;
8629 struct extent_buffer *eb;
8634 if (path->slots[wc->level] < wc->reada_slot) {
8635 wc->reada_count = wc->reada_count * 2 / 3;
8636 wc->reada_count = max(wc->reada_count, 2);
8638 wc->reada_count = wc->reada_count * 3 / 2;
8639 wc->reada_count = min_t(int, wc->reada_count,
8640 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8643 eb = path->nodes[wc->level];
8644 nritems = btrfs_header_nritems(eb);
8646 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8647 if (nread >= wc->reada_count)
8651 bytenr = btrfs_node_blockptr(eb, slot);
8652 generation = btrfs_node_ptr_generation(eb, slot);
8654 if (slot == path->slots[wc->level])
8657 if (wc->stage == UPDATE_BACKREF &&
8658 generation <= root->root_key.offset)
8661 /* We don't lock the tree block, it's OK to be racy here */
8662 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8663 wc->level - 1, 1, &refs,
8665 /* We don't care about errors in readahead. */
8670 if (wc->stage == DROP_REFERENCE) {
8674 if (wc->level == 1 &&
8675 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8677 if (!wc->update_ref ||
8678 generation <= root->root_key.offset)
8680 btrfs_node_key_to_cpu(eb, &key, slot);
8681 ret = btrfs_comp_cpu_keys(&key,
8682 &wc->update_progress);
8686 if (wc->level == 1 &&
8687 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8691 readahead_tree_block(fs_info, bytenr);
8694 wc->reada_slot = slot;
8698 * helper to process tree block while walking down the tree.
8700 * when wc->stage == UPDATE_BACKREF, this function updates
8701 * back refs for pointers in the block.
8703 * NOTE: return value 1 means we should stop walking down.
8705 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8706 struct btrfs_root *root,
8707 struct btrfs_path *path,
8708 struct walk_control *wc, int lookup_info)
8710 struct btrfs_fs_info *fs_info = root->fs_info;
8711 int level = wc->level;
8712 struct extent_buffer *eb = path->nodes[level];
8713 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8716 if (wc->stage == UPDATE_BACKREF &&
8717 btrfs_header_owner(eb) != root->root_key.objectid)
8721 * when reference count of tree block is 1, it won't increase
8722 * again. once full backref flag is set, we never clear it.
8725 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8726 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8727 BUG_ON(!path->locks[level]);
8728 ret = btrfs_lookup_extent_info(trans, fs_info,
8729 eb->start, level, 1,
8732 BUG_ON(ret == -ENOMEM);
8735 BUG_ON(wc->refs[level] == 0);
8738 if (wc->stage == DROP_REFERENCE) {
8739 if (wc->refs[level] > 1)
8742 if (path->locks[level] && !wc->keep_locks) {
8743 btrfs_tree_unlock_rw(eb, path->locks[level]);
8744 path->locks[level] = 0;
8749 /* wc->stage == UPDATE_BACKREF */
8750 if (!(wc->flags[level] & flag)) {
8751 BUG_ON(!path->locks[level]);
8752 ret = btrfs_inc_ref(trans, root, eb, 1);
8753 BUG_ON(ret); /* -ENOMEM */
8754 ret = btrfs_dec_ref(trans, root, eb, 0);
8755 BUG_ON(ret); /* -ENOMEM */
8756 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8758 btrfs_header_level(eb), 0);
8759 BUG_ON(ret); /* -ENOMEM */
8760 wc->flags[level] |= flag;
8764 * the block is shared by multiple trees, so it's not good to
8765 * keep the tree lock
8767 if (path->locks[level] && level > 0) {
8768 btrfs_tree_unlock_rw(eb, path->locks[level]);
8769 path->locks[level] = 0;
8775 * helper to process tree block pointer.
8777 * when wc->stage == DROP_REFERENCE, this function checks
8778 * reference count of the block pointed to. if the block
8779 * is shared and we need update back refs for the subtree
8780 * rooted at the block, this function changes wc->stage to
8781 * UPDATE_BACKREF. if the block is shared and there is no
8782 * need to update back, this function drops the reference
8785 * NOTE: return value 1 means we should stop walking down.
8787 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8788 struct btrfs_root *root,
8789 struct btrfs_path *path,
8790 struct walk_control *wc, int *lookup_info)
8792 struct btrfs_fs_info *fs_info = root->fs_info;
8797 struct btrfs_key key;
8798 struct extent_buffer *next;
8799 int level = wc->level;
8802 bool need_account = false;
8804 generation = btrfs_node_ptr_generation(path->nodes[level],
8805 path->slots[level]);
8807 * if the lower level block was created before the snapshot
8808 * was created, we know there is no need to update back refs
8811 if (wc->stage == UPDATE_BACKREF &&
8812 generation <= root->root_key.offset) {
8817 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8818 blocksize = fs_info->nodesize;
8820 next = find_extent_buffer(fs_info, bytenr);
8822 next = btrfs_find_create_tree_block(fs_info, bytenr);
8824 return PTR_ERR(next);
8826 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8830 btrfs_tree_lock(next);
8831 btrfs_set_lock_blocking(next);
8833 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8834 &wc->refs[level - 1],
8835 &wc->flags[level - 1]);
8839 if (unlikely(wc->refs[level - 1] == 0)) {
8840 btrfs_err(fs_info, "Missing references.");
8846 if (wc->stage == DROP_REFERENCE) {
8847 if (wc->refs[level - 1] > 1) {
8848 need_account = true;
8850 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8853 if (!wc->update_ref ||
8854 generation <= root->root_key.offset)
8857 btrfs_node_key_to_cpu(path->nodes[level], &key,
8858 path->slots[level]);
8859 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8863 wc->stage = UPDATE_BACKREF;
8864 wc->shared_level = level - 1;
8868 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8872 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8873 btrfs_tree_unlock(next);
8874 free_extent_buffer(next);
8880 if (reada && level == 1)
8881 reada_walk_down(trans, root, wc, path);
8882 next = read_tree_block(fs_info, bytenr, generation);
8884 return PTR_ERR(next);
8885 } else if (!extent_buffer_uptodate(next)) {
8886 free_extent_buffer(next);
8889 btrfs_tree_lock(next);
8890 btrfs_set_lock_blocking(next);
8894 ASSERT(level == btrfs_header_level(next));
8895 if (level != btrfs_header_level(next)) {
8896 btrfs_err(root->fs_info, "mismatched level");
8900 path->nodes[level] = next;
8901 path->slots[level] = 0;
8902 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8908 wc->refs[level - 1] = 0;
8909 wc->flags[level - 1] = 0;
8910 if (wc->stage == DROP_REFERENCE) {
8911 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8912 parent = path->nodes[level]->start;
8914 ASSERT(root->root_key.objectid ==
8915 btrfs_header_owner(path->nodes[level]));
8916 if (root->root_key.objectid !=
8917 btrfs_header_owner(path->nodes[level])) {
8918 btrfs_err(root->fs_info,
8919 "mismatched block owner");
8927 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8928 generation, level - 1);
8930 btrfs_err_rl(fs_info,
8931 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8935 ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize,
8936 parent, root->root_key.objectid,
8946 btrfs_tree_unlock(next);
8947 free_extent_buffer(next);
8953 * helper to process tree block while walking up the tree.
8955 * when wc->stage == DROP_REFERENCE, this function drops
8956 * reference count on the block.
8958 * when wc->stage == UPDATE_BACKREF, this function changes
8959 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8960 * to UPDATE_BACKREF previously while processing the block.
8962 * NOTE: return value 1 means we should stop walking up.
8964 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8965 struct btrfs_root *root,
8966 struct btrfs_path *path,
8967 struct walk_control *wc)
8969 struct btrfs_fs_info *fs_info = root->fs_info;
8971 int level = wc->level;
8972 struct extent_buffer *eb = path->nodes[level];
8975 if (wc->stage == UPDATE_BACKREF) {
8976 BUG_ON(wc->shared_level < level);
8977 if (level < wc->shared_level)
8980 ret = find_next_key(path, level + 1, &wc->update_progress);
8984 wc->stage = DROP_REFERENCE;
8985 wc->shared_level = -1;
8986 path->slots[level] = 0;
8989 * check reference count again if the block isn't locked.
8990 * we should start walking down the tree again if reference
8993 if (!path->locks[level]) {
8995 btrfs_tree_lock(eb);
8996 btrfs_set_lock_blocking(eb);
8997 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8999 ret = btrfs_lookup_extent_info(trans, fs_info,
9000 eb->start, level, 1,
9004 btrfs_tree_unlock_rw(eb, path->locks[level]);
9005 path->locks[level] = 0;
9008 BUG_ON(wc->refs[level] == 0);
9009 if (wc->refs[level] == 1) {
9010 btrfs_tree_unlock_rw(eb, path->locks[level]);
9011 path->locks[level] = 0;
9017 /* wc->stage == DROP_REFERENCE */
9018 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9020 if (wc->refs[level] == 1) {
9022 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9023 ret = btrfs_dec_ref(trans, root, eb, 1);
9025 ret = btrfs_dec_ref(trans, root, eb, 0);
9026 BUG_ON(ret); /* -ENOMEM */
9027 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
9029 btrfs_err_rl(fs_info,
9030 "error %d accounting leaf items. Quota is out of sync, rescan required.",
9034 /* make block locked assertion in clean_tree_block happy */
9035 if (!path->locks[level] &&
9036 btrfs_header_generation(eb) == trans->transid) {
9037 btrfs_tree_lock(eb);
9038 btrfs_set_lock_blocking(eb);
9039 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9041 clean_tree_block(fs_info, eb);
9044 if (eb == root->node) {
9045 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9047 else if (root->root_key.objectid != btrfs_header_owner(eb))
9048 goto owner_mismatch;
9050 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9051 parent = path->nodes[level + 1]->start;
9052 else if (root->root_key.objectid !=
9053 btrfs_header_owner(path->nodes[level + 1]))
9054 goto owner_mismatch;
9057 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9059 wc->refs[level] = 0;
9060 wc->flags[level] = 0;
9064 btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu",
9065 btrfs_header_owner(eb), root->root_key.objectid);
9069 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9070 struct btrfs_root *root,
9071 struct btrfs_path *path,
9072 struct walk_control *wc)
9074 int level = wc->level;
9075 int lookup_info = 1;
9078 while (level >= 0) {
9079 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9086 if (path->slots[level] >=
9087 btrfs_header_nritems(path->nodes[level]))
9090 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9092 path->slots[level]++;
9101 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9102 struct btrfs_root *root,
9103 struct btrfs_path *path,
9104 struct walk_control *wc, int max_level)
9106 int level = wc->level;
9109 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9110 while (level < max_level && path->nodes[level]) {
9112 if (path->slots[level] + 1 <
9113 btrfs_header_nritems(path->nodes[level])) {
9114 path->slots[level]++;
9117 ret = walk_up_proc(trans, root, path, wc);
9123 if (path->locks[level]) {
9124 btrfs_tree_unlock_rw(path->nodes[level],
9125 path->locks[level]);
9126 path->locks[level] = 0;
9128 free_extent_buffer(path->nodes[level]);
9129 path->nodes[level] = NULL;
9137 * drop a subvolume tree.
9139 * this function traverses the tree freeing any blocks that only
9140 * referenced by the tree.
9142 * when a shared tree block is found. this function decreases its
9143 * reference count by one. if update_ref is true, this function
9144 * also make sure backrefs for the shared block and all lower level
9145 * blocks are properly updated.
9147 * If called with for_reloc == 0, may exit early with -EAGAIN
9149 int btrfs_drop_snapshot(struct btrfs_root *root,
9150 struct btrfs_block_rsv *block_rsv, int update_ref,
9153 struct btrfs_fs_info *fs_info = root->fs_info;
9154 struct btrfs_path *path;
9155 struct btrfs_trans_handle *trans;
9156 struct btrfs_root *tree_root = fs_info->tree_root;
9157 struct btrfs_root_item *root_item = &root->root_item;
9158 struct walk_control *wc;
9159 struct btrfs_key key;
9163 bool root_dropped = false;
9165 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9167 path = btrfs_alloc_path();
9173 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9175 btrfs_free_path(path);
9180 trans = btrfs_start_transaction(tree_root, 0);
9181 if (IS_ERR(trans)) {
9182 err = PTR_ERR(trans);
9187 trans->block_rsv = block_rsv;
9189 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9190 level = btrfs_header_level(root->node);
9191 path->nodes[level] = btrfs_lock_root_node(root);
9192 btrfs_set_lock_blocking(path->nodes[level]);
9193 path->slots[level] = 0;
9194 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9195 memset(&wc->update_progress, 0,
9196 sizeof(wc->update_progress));
9198 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9199 memcpy(&wc->update_progress, &key,
9200 sizeof(wc->update_progress));
9202 level = root_item->drop_level;
9204 path->lowest_level = level;
9205 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9206 path->lowest_level = 0;
9214 * unlock our path, this is safe because only this
9215 * function is allowed to delete this snapshot
9217 btrfs_unlock_up_safe(path, 0);
9219 level = btrfs_header_level(root->node);
9221 btrfs_tree_lock(path->nodes[level]);
9222 btrfs_set_lock_blocking(path->nodes[level]);
9223 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9225 ret = btrfs_lookup_extent_info(trans, fs_info,
9226 path->nodes[level]->start,
9227 level, 1, &wc->refs[level],
9233 BUG_ON(wc->refs[level] == 0);
9235 if (level == root_item->drop_level)
9238 btrfs_tree_unlock(path->nodes[level]);
9239 path->locks[level] = 0;
9240 WARN_ON(wc->refs[level] != 1);
9246 wc->shared_level = -1;
9247 wc->stage = DROP_REFERENCE;
9248 wc->update_ref = update_ref;
9250 wc->for_reloc = for_reloc;
9251 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9255 ret = walk_down_tree(trans, root, path, wc);
9261 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9268 BUG_ON(wc->stage != DROP_REFERENCE);
9272 if (wc->stage == DROP_REFERENCE) {
9274 btrfs_node_key(path->nodes[level],
9275 &root_item->drop_progress,
9276 path->slots[level]);
9277 root_item->drop_level = level;
9280 BUG_ON(wc->level == 0);
9281 if (btrfs_should_end_transaction(trans) ||
9282 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9283 ret = btrfs_update_root(trans, tree_root,
9287 btrfs_abort_transaction(trans, ret);
9292 btrfs_end_transaction_throttle(trans);
9293 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9294 btrfs_debug(fs_info,
9295 "drop snapshot early exit");
9300 trans = btrfs_start_transaction(tree_root, 0);
9301 if (IS_ERR(trans)) {
9302 err = PTR_ERR(trans);
9306 trans->block_rsv = block_rsv;
9309 btrfs_release_path(path);
9313 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9315 btrfs_abort_transaction(trans, ret);
9320 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9321 ret = btrfs_find_root(tree_root, &root->root_key, path,
9324 btrfs_abort_transaction(trans, ret);
9327 } else if (ret > 0) {
9328 /* if we fail to delete the orphan item this time
9329 * around, it'll get picked up the next time.
9331 * The most common failure here is just -ENOENT.
9333 btrfs_del_orphan_item(trans, tree_root,
9334 root->root_key.objectid);
9338 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9339 btrfs_add_dropped_root(trans, root);
9341 free_extent_buffer(root->node);
9342 free_extent_buffer(root->commit_root);
9343 btrfs_put_fs_root(root);
9345 root_dropped = true;
9347 btrfs_end_transaction_throttle(trans);
9350 btrfs_free_path(path);
9353 * So if we need to stop dropping the snapshot for whatever reason we
9354 * need to make sure to add it back to the dead root list so that we
9355 * keep trying to do the work later. This also cleans up roots if we
9356 * don't have it in the radix (like when we recover after a power fail
9357 * or unmount) so we don't leak memory.
9359 if (!for_reloc && root_dropped == false)
9360 btrfs_add_dead_root(root);
9361 if (err && err != -EAGAIN)
9362 btrfs_handle_fs_error(fs_info, err, NULL);
9367 * drop subtree rooted at tree block 'node'.
9369 * NOTE: this function will unlock and release tree block 'node'
9370 * only used by relocation code
9372 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9373 struct btrfs_root *root,
9374 struct extent_buffer *node,
9375 struct extent_buffer *parent)
9377 struct btrfs_fs_info *fs_info = root->fs_info;
9378 struct btrfs_path *path;
9379 struct walk_control *wc;
9385 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9387 path = btrfs_alloc_path();
9391 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9393 btrfs_free_path(path);
9397 btrfs_assert_tree_locked(parent);
9398 parent_level = btrfs_header_level(parent);
9399 extent_buffer_get(parent);
9400 path->nodes[parent_level] = parent;
9401 path->slots[parent_level] = btrfs_header_nritems(parent);
9403 btrfs_assert_tree_locked(node);
9404 level = btrfs_header_level(node);
9405 path->nodes[level] = node;
9406 path->slots[level] = 0;
9407 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9409 wc->refs[parent_level] = 1;
9410 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9412 wc->shared_level = -1;
9413 wc->stage = DROP_REFERENCE;
9417 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9420 wret = walk_down_tree(trans, root, path, wc);
9426 wret = walk_up_tree(trans, root, path, wc, parent_level);
9434 btrfs_free_path(path);
9438 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9444 * if restripe for this chunk_type is on pick target profile and
9445 * return, otherwise do the usual balance
9447 stripped = get_restripe_target(fs_info, flags);
9449 return extended_to_chunk(stripped);
9451 num_devices = fs_info->fs_devices->rw_devices;
9453 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9454 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9455 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9457 if (num_devices == 1) {
9458 stripped |= BTRFS_BLOCK_GROUP_DUP;
9459 stripped = flags & ~stripped;
9461 /* turn raid0 into single device chunks */
9462 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9465 /* turn mirroring into duplication */
9466 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9467 BTRFS_BLOCK_GROUP_RAID10))
9468 return stripped | BTRFS_BLOCK_GROUP_DUP;
9470 /* they already had raid on here, just return */
9471 if (flags & stripped)
9474 stripped |= BTRFS_BLOCK_GROUP_DUP;
9475 stripped = flags & ~stripped;
9477 /* switch duplicated blocks with raid1 */
9478 if (flags & BTRFS_BLOCK_GROUP_DUP)
9479 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9481 /* this is drive concat, leave it alone */
9487 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9489 struct btrfs_space_info *sinfo = cache->space_info;
9491 u64 min_allocable_bytes;
9495 * We need some metadata space and system metadata space for
9496 * allocating chunks in some corner cases until we force to set
9497 * it to be readonly.
9500 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9502 min_allocable_bytes = SZ_1M;
9504 min_allocable_bytes = 0;
9506 spin_lock(&sinfo->lock);
9507 spin_lock(&cache->lock);
9515 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9516 cache->bytes_super - btrfs_block_group_used(&cache->item);
9518 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9519 min_allocable_bytes <= sinfo->total_bytes) {
9520 sinfo->bytes_readonly += num_bytes;
9522 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9526 spin_unlock(&cache->lock);
9527 spin_unlock(&sinfo->lock);
9531 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9532 struct btrfs_block_group_cache *cache)
9535 struct btrfs_trans_handle *trans;
9540 trans = btrfs_join_transaction(fs_info->extent_root);
9542 return PTR_ERR(trans);
9545 * we're not allowed to set block groups readonly after the dirty
9546 * block groups cache has started writing. If it already started,
9547 * back off and let this transaction commit
9549 mutex_lock(&fs_info->ro_block_group_mutex);
9550 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9551 u64 transid = trans->transid;
9553 mutex_unlock(&fs_info->ro_block_group_mutex);
9554 btrfs_end_transaction(trans);
9556 ret = btrfs_wait_for_commit(fs_info, transid);
9563 * if we are changing raid levels, try to allocate a corresponding
9564 * block group with the new raid level.
9566 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9567 if (alloc_flags != cache->flags) {
9568 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9571 * ENOSPC is allowed here, we may have enough space
9572 * already allocated at the new raid level to
9581 ret = inc_block_group_ro(cache, 0);
9584 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9585 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9589 ret = inc_block_group_ro(cache, 0);
9591 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9592 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9593 mutex_lock(&fs_info->chunk_mutex);
9594 check_system_chunk(trans, fs_info, alloc_flags);
9595 mutex_unlock(&fs_info->chunk_mutex);
9597 mutex_unlock(&fs_info->ro_block_group_mutex);
9599 btrfs_end_transaction(trans);
9603 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9604 struct btrfs_fs_info *fs_info, u64 type)
9606 u64 alloc_flags = get_alloc_profile(fs_info, type);
9608 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9612 * helper to account the unused space of all the readonly block group in the
9613 * space_info. takes mirrors into account.
9615 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9617 struct btrfs_block_group_cache *block_group;
9621 /* It's df, we don't care if it's racy */
9622 if (list_empty(&sinfo->ro_bgs))
9625 spin_lock(&sinfo->lock);
9626 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9627 spin_lock(&block_group->lock);
9629 if (!block_group->ro) {
9630 spin_unlock(&block_group->lock);
9634 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9635 BTRFS_BLOCK_GROUP_RAID10 |
9636 BTRFS_BLOCK_GROUP_DUP))
9641 free_bytes += (block_group->key.offset -
9642 btrfs_block_group_used(&block_group->item)) *
9645 spin_unlock(&block_group->lock);
9647 spin_unlock(&sinfo->lock);
9652 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9654 struct btrfs_space_info *sinfo = cache->space_info;
9659 spin_lock(&sinfo->lock);
9660 spin_lock(&cache->lock);
9662 num_bytes = cache->key.offset - cache->reserved -
9663 cache->pinned - cache->bytes_super -
9664 btrfs_block_group_used(&cache->item);
9665 sinfo->bytes_readonly -= num_bytes;
9666 list_del_init(&cache->ro_list);
9668 spin_unlock(&cache->lock);
9669 spin_unlock(&sinfo->lock);
9673 * checks to see if its even possible to relocate this block group.
9675 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9676 * ok to go ahead and try.
9678 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9680 struct btrfs_root *root = fs_info->extent_root;
9681 struct btrfs_block_group_cache *block_group;
9682 struct btrfs_space_info *space_info;
9683 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9684 struct btrfs_device *device;
9685 struct btrfs_trans_handle *trans;
9695 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9697 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9699 /* odd, couldn't find the block group, leave it alone */
9703 "can't find block group for bytenr %llu",
9708 min_free = btrfs_block_group_used(&block_group->item);
9710 /* no bytes used, we're good */
9714 space_info = block_group->space_info;
9715 spin_lock(&space_info->lock);
9717 full = space_info->full;
9720 * if this is the last block group we have in this space, we can't
9721 * relocate it unless we're able to allocate a new chunk below.
9723 * Otherwise, we need to make sure we have room in the space to handle
9724 * all of the extents from this block group. If we can, we're good
9726 if ((space_info->total_bytes != block_group->key.offset) &&
9727 (btrfs_space_info_used(space_info, false) + min_free <
9728 space_info->total_bytes)) {
9729 spin_unlock(&space_info->lock);
9732 spin_unlock(&space_info->lock);
9735 * ok we don't have enough space, but maybe we have free space on our
9736 * devices to allocate new chunks for relocation, so loop through our
9737 * alloc devices and guess if we have enough space. if this block
9738 * group is going to be restriped, run checks against the target
9739 * profile instead of the current one.
9751 target = get_restripe_target(fs_info, block_group->flags);
9753 index = __get_raid_index(extended_to_chunk(target));
9756 * this is just a balance, so if we were marked as full
9757 * we know there is no space for a new chunk
9762 "no space to alloc new chunk for block group %llu",
9763 block_group->key.objectid);
9767 index = get_block_group_index(block_group);
9770 if (index == BTRFS_RAID_RAID10) {
9774 } else if (index == BTRFS_RAID_RAID1) {
9776 } else if (index == BTRFS_RAID_DUP) {
9779 } else if (index == BTRFS_RAID_RAID0) {
9780 dev_min = fs_devices->rw_devices;
9781 min_free = div64_u64(min_free, dev_min);
9784 /* We need to do this so that we can look at pending chunks */
9785 trans = btrfs_join_transaction(root);
9786 if (IS_ERR(trans)) {
9787 ret = PTR_ERR(trans);
9791 mutex_lock(&fs_info->chunk_mutex);
9792 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9796 * check to make sure we can actually find a chunk with enough
9797 * space to fit our block group in.
9799 if (device->total_bytes > device->bytes_used + min_free &&
9800 !device->is_tgtdev_for_dev_replace) {
9801 ret = find_free_dev_extent(trans, device, min_free,
9806 if (dev_nr >= dev_min)
9812 if (debug && ret == -1)
9814 "no space to allocate a new chunk for block group %llu",
9815 block_group->key.objectid);
9816 mutex_unlock(&fs_info->chunk_mutex);
9817 btrfs_end_transaction(trans);
9819 btrfs_put_block_group(block_group);
9823 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9824 struct btrfs_path *path,
9825 struct btrfs_key *key)
9827 struct btrfs_root *root = fs_info->extent_root;
9829 struct btrfs_key found_key;
9830 struct extent_buffer *leaf;
9831 struct btrfs_block_group_item bg;
9835 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9840 slot = path->slots[0];
9841 leaf = path->nodes[0];
9842 if (slot >= btrfs_header_nritems(leaf)) {
9843 ret = btrfs_next_leaf(root, path);
9850 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9852 if (found_key.objectid >= key->objectid &&
9853 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9854 struct extent_map_tree *em_tree;
9855 struct extent_map *em;
9857 em_tree = &root->fs_info->mapping_tree.map_tree;
9858 read_lock(&em_tree->lock);
9859 em = lookup_extent_mapping(em_tree, found_key.objectid,
9861 read_unlock(&em_tree->lock);
9864 "logical %llu len %llu found bg but no related chunk",
9865 found_key.objectid, found_key.offset);
9867 } else if (em->start != found_key.objectid ||
9868 em->len != found_key.offset) {
9870 "block group %llu len %llu mismatch with chunk %llu len %llu",
9871 found_key.objectid, found_key.offset,
9872 em->start, em->len);
9875 read_extent_buffer(leaf, &bg,
9876 btrfs_item_ptr_offset(leaf, slot),
9878 flags = btrfs_block_group_flags(&bg) &
9879 BTRFS_BLOCK_GROUP_TYPE_MASK;
9881 if (flags != (em->map_lookup->type &
9882 BTRFS_BLOCK_GROUP_TYPE_MASK)) {
9884 "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx",
9886 found_key.offset, flags,
9887 (BTRFS_BLOCK_GROUP_TYPE_MASK &
9888 em->map_lookup->type));
9894 free_extent_map(em);
9903 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9905 struct btrfs_block_group_cache *block_group;
9909 struct inode *inode;
9911 block_group = btrfs_lookup_first_block_group(info, last);
9912 while (block_group) {
9913 wait_block_group_cache_done(block_group);
9914 spin_lock(&block_group->lock);
9915 if (block_group->iref)
9917 spin_unlock(&block_group->lock);
9918 block_group = next_block_group(info, block_group);
9927 inode = block_group->inode;
9928 block_group->iref = 0;
9929 block_group->inode = NULL;
9930 spin_unlock(&block_group->lock);
9931 ASSERT(block_group->io_ctl.inode == NULL);
9933 last = block_group->key.objectid + block_group->key.offset;
9934 btrfs_put_block_group(block_group);
9939 * Must be called only after stopping all workers, since we could have block
9940 * group caching kthreads running, and therefore they could race with us if we
9941 * freed the block groups before stopping them.
9943 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9945 struct btrfs_block_group_cache *block_group;
9946 struct btrfs_space_info *space_info;
9947 struct btrfs_caching_control *caching_ctl;
9950 down_write(&info->commit_root_sem);
9951 while (!list_empty(&info->caching_block_groups)) {
9952 caching_ctl = list_entry(info->caching_block_groups.next,
9953 struct btrfs_caching_control, list);
9954 list_del(&caching_ctl->list);
9955 put_caching_control(caching_ctl);
9957 up_write(&info->commit_root_sem);
9959 spin_lock(&info->unused_bgs_lock);
9960 while (!list_empty(&info->unused_bgs)) {
9961 block_group = list_first_entry(&info->unused_bgs,
9962 struct btrfs_block_group_cache,
9964 list_del_init(&block_group->bg_list);
9965 btrfs_put_block_group(block_group);
9967 spin_unlock(&info->unused_bgs_lock);
9969 spin_lock(&info->block_group_cache_lock);
9970 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9971 block_group = rb_entry(n, struct btrfs_block_group_cache,
9973 rb_erase(&block_group->cache_node,
9974 &info->block_group_cache_tree);
9975 RB_CLEAR_NODE(&block_group->cache_node);
9976 spin_unlock(&info->block_group_cache_lock);
9978 down_write(&block_group->space_info->groups_sem);
9979 list_del(&block_group->list);
9980 up_write(&block_group->space_info->groups_sem);
9983 * We haven't cached this block group, which means we could
9984 * possibly have excluded extents on this block group.
9986 if (block_group->cached == BTRFS_CACHE_NO ||
9987 block_group->cached == BTRFS_CACHE_ERROR)
9988 free_excluded_extents(info, block_group);
9990 btrfs_remove_free_space_cache(block_group);
9991 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9992 ASSERT(list_empty(&block_group->dirty_list));
9993 ASSERT(list_empty(&block_group->io_list));
9994 ASSERT(list_empty(&block_group->bg_list));
9995 ASSERT(atomic_read(&block_group->count) == 1);
9996 btrfs_put_block_group(block_group);
9998 spin_lock(&info->block_group_cache_lock);
10000 spin_unlock(&info->block_group_cache_lock);
10002 /* now that all the block groups are freed, go through and
10003 * free all the space_info structs. This is only called during
10004 * the final stages of unmount, and so we know nobody is
10005 * using them. We call synchronize_rcu() once before we start,
10006 * just to be on the safe side.
10010 release_global_block_rsv(info);
10012 while (!list_empty(&info->space_info)) {
10015 space_info = list_entry(info->space_info.next,
10016 struct btrfs_space_info,
10020 * Do not hide this behind enospc_debug, this is actually
10021 * important and indicates a real bug if this happens.
10023 if (WARN_ON(space_info->bytes_pinned > 0 ||
10024 space_info->bytes_reserved > 0 ||
10025 space_info->bytes_may_use > 0))
10026 dump_space_info(info, space_info, 0, 0);
10027 list_del(&space_info->list);
10028 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10029 struct kobject *kobj;
10030 kobj = space_info->block_group_kobjs[i];
10031 space_info->block_group_kobjs[i] = NULL;
10037 kobject_del(&space_info->kobj);
10038 kobject_put(&space_info->kobj);
10043 static void __link_block_group(struct btrfs_space_info *space_info,
10044 struct btrfs_block_group_cache *cache)
10046 int index = get_block_group_index(cache);
10047 bool first = false;
10049 down_write(&space_info->groups_sem);
10050 if (list_empty(&space_info->block_groups[index]))
10052 list_add_tail(&cache->list, &space_info->block_groups[index]);
10053 up_write(&space_info->groups_sem);
10056 struct raid_kobject *rkobj;
10059 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10062 rkobj->raid_type = index;
10063 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10064 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10065 "%s", get_raid_name(index));
10067 kobject_put(&rkobj->kobj);
10070 space_info->block_group_kobjs[index] = &rkobj->kobj;
10075 btrfs_warn(cache->fs_info,
10076 "failed to add kobject for block cache, ignoring");
10079 static struct btrfs_block_group_cache *
10080 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10081 u64 start, u64 size)
10083 struct btrfs_block_group_cache *cache;
10085 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10089 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10091 if (!cache->free_space_ctl) {
10096 cache->key.objectid = start;
10097 cache->key.offset = size;
10098 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10100 cache->fs_info = fs_info;
10101 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10102 set_free_space_tree_thresholds(cache);
10104 atomic_set(&cache->count, 1);
10105 spin_lock_init(&cache->lock);
10106 init_rwsem(&cache->data_rwsem);
10107 INIT_LIST_HEAD(&cache->list);
10108 INIT_LIST_HEAD(&cache->cluster_list);
10109 INIT_LIST_HEAD(&cache->bg_list);
10110 INIT_LIST_HEAD(&cache->ro_list);
10111 INIT_LIST_HEAD(&cache->dirty_list);
10112 INIT_LIST_HEAD(&cache->io_list);
10113 btrfs_init_free_space_ctl(cache);
10114 atomic_set(&cache->trimming, 0);
10115 mutex_init(&cache->free_space_lock);
10116 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10123 * Iterate all chunks and verify that each of them has the corresponding block
10126 static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info)
10128 struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
10129 struct extent_map *em;
10130 struct btrfs_block_group_cache *bg;
10135 read_lock(&map_tree->map_tree.lock);
10137 * lookup_extent_mapping will return the first extent map
10138 * intersecting the range, so setting @len to 1 is enough to
10139 * get the first chunk.
10141 em = lookup_extent_mapping(&map_tree->map_tree, start, 1);
10142 read_unlock(&map_tree->map_tree.lock);
10146 bg = btrfs_lookup_block_group(fs_info, em->start);
10149 "chunk start=%llu len=%llu doesn't have corresponding block group",
10150 em->start, em->len);
10152 free_extent_map(em);
10155 if (bg->key.objectid != em->start ||
10156 bg->key.offset != em->len ||
10157 (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) !=
10158 (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) {
10160 "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx",
10161 em->start, em->len,
10162 em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK,
10163 bg->key.objectid, bg->key.offset,
10164 bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK);
10166 free_extent_map(em);
10167 btrfs_put_block_group(bg);
10170 start = em->start + em->len;
10171 free_extent_map(em);
10172 btrfs_put_block_group(bg);
10177 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10179 struct btrfs_path *path;
10181 struct btrfs_block_group_cache *cache;
10182 struct btrfs_space_info *space_info;
10183 struct btrfs_key key;
10184 struct btrfs_key found_key;
10185 struct extent_buffer *leaf;
10186 int need_clear = 0;
10191 feature = btrfs_super_incompat_flags(info->super_copy);
10192 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10196 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10197 path = btrfs_alloc_path();
10200 path->reada = READA_FORWARD;
10202 cache_gen = btrfs_super_cache_generation(info->super_copy);
10203 if (btrfs_test_opt(info, SPACE_CACHE) &&
10204 btrfs_super_generation(info->super_copy) != cache_gen)
10206 if (btrfs_test_opt(info, CLEAR_CACHE))
10210 ret = find_first_block_group(info, path, &key);
10216 leaf = path->nodes[0];
10217 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10219 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10228 * When we mount with old space cache, we need to
10229 * set BTRFS_DC_CLEAR and set dirty flag.
10231 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10232 * truncate the old free space cache inode and
10234 * b) Setting 'dirty flag' makes sure that we flush
10235 * the new space cache info onto disk.
10237 if (btrfs_test_opt(info, SPACE_CACHE))
10238 cache->disk_cache_state = BTRFS_DC_CLEAR;
10241 read_extent_buffer(leaf, &cache->item,
10242 btrfs_item_ptr_offset(leaf, path->slots[0]),
10243 sizeof(cache->item));
10244 cache->flags = btrfs_block_group_flags(&cache->item);
10246 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10247 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10249 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10250 cache->key.objectid);
10255 key.objectid = found_key.objectid + found_key.offset;
10256 btrfs_release_path(path);
10259 * We need to exclude the super stripes now so that the space
10260 * info has super bytes accounted for, otherwise we'll think
10261 * we have more space than we actually do.
10263 ret = exclude_super_stripes(info, cache);
10266 * We may have excluded something, so call this just in
10269 free_excluded_extents(info, cache);
10270 btrfs_put_block_group(cache);
10275 * check for two cases, either we are full, and therefore
10276 * don't need to bother with the caching work since we won't
10277 * find any space, or we are empty, and we can just add all
10278 * the space in and be done with it. This saves us _alot_ of
10279 * time, particularly in the full case.
10281 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10282 cache->last_byte_to_unpin = (u64)-1;
10283 cache->cached = BTRFS_CACHE_FINISHED;
10284 free_excluded_extents(info, cache);
10285 } else if (btrfs_block_group_used(&cache->item) == 0) {
10286 cache->last_byte_to_unpin = (u64)-1;
10287 cache->cached = BTRFS_CACHE_FINISHED;
10288 add_new_free_space(cache, info,
10289 found_key.objectid,
10290 found_key.objectid +
10292 free_excluded_extents(info, cache);
10295 ret = btrfs_add_block_group_cache(info, cache);
10297 btrfs_remove_free_space_cache(cache);
10298 btrfs_put_block_group(cache);
10302 trace_btrfs_add_block_group(info, cache, 0);
10303 update_space_info(info, cache->flags, found_key.offset,
10304 btrfs_block_group_used(&cache->item),
10305 cache->bytes_super, &space_info);
10307 cache->space_info = space_info;
10309 __link_block_group(space_info, cache);
10311 set_avail_alloc_bits(info, cache->flags);
10312 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10313 inc_block_group_ro(cache, 1);
10314 } else if (btrfs_block_group_used(&cache->item) == 0) {
10315 spin_lock(&info->unused_bgs_lock);
10316 /* Should always be true but just in case. */
10317 if (list_empty(&cache->bg_list)) {
10318 btrfs_get_block_group(cache);
10319 list_add_tail(&cache->bg_list,
10320 &info->unused_bgs);
10322 spin_unlock(&info->unused_bgs_lock);
10326 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10327 if (!(get_alloc_profile(info, space_info->flags) &
10328 (BTRFS_BLOCK_GROUP_RAID10 |
10329 BTRFS_BLOCK_GROUP_RAID1 |
10330 BTRFS_BLOCK_GROUP_RAID5 |
10331 BTRFS_BLOCK_GROUP_RAID6 |
10332 BTRFS_BLOCK_GROUP_DUP)))
10335 * avoid allocating from un-mirrored block group if there are
10336 * mirrored block groups.
10338 list_for_each_entry(cache,
10339 &space_info->block_groups[BTRFS_RAID_RAID0],
10341 inc_block_group_ro(cache, 1);
10342 list_for_each_entry(cache,
10343 &space_info->block_groups[BTRFS_RAID_SINGLE],
10345 inc_block_group_ro(cache, 1);
10348 init_global_block_rsv(info);
10349 ret = check_chunk_block_group_mappings(info);
10351 btrfs_free_path(path);
10355 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10356 struct btrfs_fs_info *fs_info)
10358 struct btrfs_block_group_cache *block_group;
10359 struct btrfs_root *extent_root = fs_info->extent_root;
10360 struct btrfs_block_group_item item;
10361 struct btrfs_key key;
10363 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10365 trans->can_flush_pending_bgs = false;
10366 while (!list_empty(&trans->new_bgs)) {
10367 block_group = list_first_entry(&trans->new_bgs,
10368 struct btrfs_block_group_cache,
10373 spin_lock(&block_group->lock);
10374 memcpy(&item, &block_group->item, sizeof(item));
10375 memcpy(&key, &block_group->key, sizeof(key));
10376 spin_unlock(&block_group->lock);
10378 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10381 btrfs_abort_transaction(trans, ret);
10382 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10385 btrfs_abort_transaction(trans, ret);
10386 add_block_group_free_space(trans, fs_info, block_group);
10387 /* already aborted the transaction if it failed. */
10389 list_del_init(&block_group->bg_list);
10391 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10394 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10395 struct btrfs_fs_info *fs_info, u64 bytes_used,
10396 u64 type, u64 chunk_offset, u64 size)
10398 struct btrfs_block_group_cache *cache;
10401 btrfs_set_log_full_commit(fs_info, trans);
10403 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10407 btrfs_set_block_group_used(&cache->item, bytes_used);
10408 btrfs_set_block_group_chunk_objectid(&cache->item,
10409 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10410 btrfs_set_block_group_flags(&cache->item, type);
10412 cache->flags = type;
10413 cache->last_byte_to_unpin = (u64)-1;
10414 cache->cached = BTRFS_CACHE_FINISHED;
10415 cache->needs_free_space = 1;
10416 ret = exclude_super_stripes(fs_info, cache);
10419 * We may have excluded something, so call this just in
10422 free_excluded_extents(fs_info, cache);
10423 btrfs_put_block_group(cache);
10427 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10429 free_excluded_extents(fs_info, cache);
10431 #ifdef CONFIG_BTRFS_DEBUG
10432 if (btrfs_should_fragment_free_space(cache)) {
10433 u64 new_bytes_used = size - bytes_used;
10435 bytes_used += new_bytes_used >> 1;
10436 fragment_free_space(cache);
10440 * Ensure the corresponding space_info object is created and
10441 * assigned to our block group. We want our bg to be added to the rbtree
10442 * with its ->space_info set.
10444 cache->space_info = __find_space_info(fs_info, cache->flags);
10445 if (!cache->space_info) {
10446 ret = create_space_info(fs_info, cache->flags,
10447 &cache->space_info);
10449 btrfs_remove_free_space_cache(cache);
10450 btrfs_put_block_group(cache);
10455 ret = btrfs_add_block_group_cache(fs_info, cache);
10457 btrfs_remove_free_space_cache(cache);
10458 btrfs_put_block_group(cache);
10463 * Now that our block group has its ->space_info set and is inserted in
10464 * the rbtree, update the space info's counters.
10466 trace_btrfs_add_block_group(fs_info, cache, 1);
10467 update_space_info(fs_info, cache->flags, size, bytes_used,
10468 cache->bytes_super, &cache->space_info);
10469 update_global_block_rsv(fs_info);
10471 __link_block_group(cache->space_info, cache);
10473 list_add_tail(&cache->bg_list, &trans->new_bgs);
10475 set_avail_alloc_bits(fs_info, type);
10479 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10481 u64 extra_flags = chunk_to_extended(flags) &
10482 BTRFS_EXTENDED_PROFILE_MASK;
10484 write_seqlock(&fs_info->profiles_lock);
10485 if (flags & BTRFS_BLOCK_GROUP_DATA)
10486 fs_info->avail_data_alloc_bits &= ~extra_flags;
10487 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10488 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10489 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10490 fs_info->avail_system_alloc_bits &= ~extra_flags;
10491 write_sequnlock(&fs_info->profiles_lock);
10494 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10495 struct btrfs_fs_info *fs_info, u64 group_start,
10496 struct extent_map *em)
10498 struct btrfs_root *root = fs_info->extent_root;
10499 struct btrfs_path *path;
10500 struct btrfs_block_group_cache *block_group;
10501 struct btrfs_free_cluster *cluster;
10502 struct btrfs_root *tree_root = fs_info->tree_root;
10503 struct btrfs_key key;
10504 struct inode *inode;
10505 struct kobject *kobj = NULL;
10509 struct btrfs_caching_control *caching_ctl = NULL;
10512 block_group = btrfs_lookup_block_group(fs_info, group_start);
10513 BUG_ON(!block_group);
10514 BUG_ON(!block_group->ro);
10517 * Free the reserved super bytes from this block group before
10520 free_excluded_extents(fs_info, block_group);
10522 memcpy(&key, &block_group->key, sizeof(key));
10523 index = get_block_group_index(block_group);
10524 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10525 BTRFS_BLOCK_GROUP_RAID1 |
10526 BTRFS_BLOCK_GROUP_RAID10))
10531 /* make sure this block group isn't part of an allocation cluster */
10532 cluster = &fs_info->data_alloc_cluster;
10533 spin_lock(&cluster->refill_lock);
10534 btrfs_return_cluster_to_free_space(block_group, cluster);
10535 spin_unlock(&cluster->refill_lock);
10538 * make sure this block group isn't part of a metadata
10539 * allocation cluster
10541 cluster = &fs_info->meta_alloc_cluster;
10542 spin_lock(&cluster->refill_lock);
10543 btrfs_return_cluster_to_free_space(block_group, cluster);
10544 spin_unlock(&cluster->refill_lock);
10546 path = btrfs_alloc_path();
10553 * get the inode first so any iput calls done for the io_list
10554 * aren't the final iput (no unlinks allowed now)
10556 inode = lookup_free_space_inode(fs_info, block_group, path);
10558 mutex_lock(&trans->transaction->cache_write_mutex);
10560 * make sure our free spache cache IO is done before remove the
10563 spin_lock(&trans->transaction->dirty_bgs_lock);
10564 if (!list_empty(&block_group->io_list)) {
10565 list_del_init(&block_group->io_list);
10567 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10569 spin_unlock(&trans->transaction->dirty_bgs_lock);
10570 btrfs_wait_cache_io(trans, block_group, path);
10571 btrfs_put_block_group(block_group);
10572 spin_lock(&trans->transaction->dirty_bgs_lock);
10575 if (!list_empty(&block_group->dirty_list)) {
10576 list_del_init(&block_group->dirty_list);
10577 btrfs_put_block_group(block_group);
10579 spin_unlock(&trans->transaction->dirty_bgs_lock);
10580 mutex_unlock(&trans->transaction->cache_write_mutex);
10582 if (!IS_ERR(inode)) {
10583 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10585 btrfs_add_delayed_iput(inode);
10588 clear_nlink(inode);
10589 /* One for the block groups ref */
10590 spin_lock(&block_group->lock);
10591 if (block_group->iref) {
10592 block_group->iref = 0;
10593 block_group->inode = NULL;
10594 spin_unlock(&block_group->lock);
10597 spin_unlock(&block_group->lock);
10599 /* One for our lookup ref */
10600 btrfs_add_delayed_iput(inode);
10603 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10604 key.offset = block_group->key.objectid;
10607 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10611 btrfs_release_path(path);
10613 ret = btrfs_del_item(trans, tree_root, path);
10616 btrfs_release_path(path);
10619 spin_lock(&fs_info->block_group_cache_lock);
10620 rb_erase(&block_group->cache_node,
10621 &fs_info->block_group_cache_tree);
10622 RB_CLEAR_NODE(&block_group->cache_node);
10624 if (fs_info->first_logical_byte == block_group->key.objectid)
10625 fs_info->first_logical_byte = (u64)-1;
10626 spin_unlock(&fs_info->block_group_cache_lock);
10628 down_write(&block_group->space_info->groups_sem);
10630 * we must use list_del_init so people can check to see if they
10631 * are still on the list after taking the semaphore
10633 list_del_init(&block_group->list);
10634 if (list_empty(&block_group->space_info->block_groups[index])) {
10635 kobj = block_group->space_info->block_group_kobjs[index];
10636 block_group->space_info->block_group_kobjs[index] = NULL;
10637 clear_avail_alloc_bits(fs_info, block_group->flags);
10639 up_write(&block_group->space_info->groups_sem);
10645 if (block_group->has_caching_ctl)
10646 caching_ctl = get_caching_control(block_group);
10647 if (block_group->cached == BTRFS_CACHE_STARTED)
10648 wait_block_group_cache_done(block_group);
10649 if (block_group->has_caching_ctl) {
10650 down_write(&fs_info->commit_root_sem);
10651 if (!caching_ctl) {
10652 struct btrfs_caching_control *ctl;
10654 list_for_each_entry(ctl,
10655 &fs_info->caching_block_groups, list)
10656 if (ctl->block_group == block_group) {
10658 refcount_inc(&caching_ctl->count);
10663 list_del_init(&caching_ctl->list);
10664 up_write(&fs_info->commit_root_sem);
10666 /* Once for the caching bgs list and once for us. */
10667 put_caching_control(caching_ctl);
10668 put_caching_control(caching_ctl);
10672 spin_lock(&trans->transaction->dirty_bgs_lock);
10673 if (!list_empty(&block_group->dirty_list)) {
10676 if (!list_empty(&block_group->io_list)) {
10679 spin_unlock(&trans->transaction->dirty_bgs_lock);
10680 btrfs_remove_free_space_cache(block_group);
10682 spin_lock(&block_group->space_info->lock);
10683 list_del_init(&block_group->ro_list);
10685 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10686 WARN_ON(block_group->space_info->total_bytes
10687 < block_group->key.offset);
10688 WARN_ON(block_group->space_info->bytes_readonly
10689 < block_group->key.offset);
10690 WARN_ON(block_group->space_info->disk_total
10691 < block_group->key.offset * factor);
10693 block_group->space_info->total_bytes -= block_group->key.offset;
10694 block_group->space_info->bytes_readonly -= block_group->key.offset;
10695 block_group->space_info->disk_total -= block_group->key.offset * factor;
10697 spin_unlock(&block_group->space_info->lock);
10699 memcpy(&key, &block_group->key, sizeof(key));
10701 mutex_lock(&fs_info->chunk_mutex);
10702 if (!list_empty(&em->list)) {
10703 /* We're in the transaction->pending_chunks list. */
10704 free_extent_map(em);
10706 spin_lock(&block_group->lock);
10707 block_group->removed = 1;
10709 * At this point trimming can't start on this block group, because we
10710 * removed the block group from the tree fs_info->block_group_cache_tree
10711 * so no one can't find it anymore and even if someone already got this
10712 * block group before we removed it from the rbtree, they have already
10713 * incremented block_group->trimming - if they didn't, they won't find
10714 * any free space entries because we already removed them all when we
10715 * called btrfs_remove_free_space_cache().
10717 * And we must not remove the extent map from the fs_info->mapping_tree
10718 * to prevent the same logical address range and physical device space
10719 * ranges from being reused for a new block group. This is because our
10720 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10721 * completely transactionless, so while it is trimming a range the
10722 * currently running transaction might finish and a new one start,
10723 * allowing for new block groups to be created that can reuse the same
10724 * physical device locations unless we take this special care.
10726 * There may also be an implicit trim operation if the file system
10727 * is mounted with -odiscard. The same protections must remain
10728 * in place until the extents have been discarded completely when
10729 * the transaction commit has completed.
10731 remove_em = (atomic_read(&block_group->trimming) == 0);
10733 * Make sure a trimmer task always sees the em in the pinned_chunks list
10734 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10735 * before checking block_group->removed).
10739 * Our em might be in trans->transaction->pending_chunks which
10740 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10741 * and so is the fs_info->pinned_chunks list.
10743 * So at this point we must be holding the chunk_mutex to avoid
10744 * any races with chunk allocation (more specifically at
10745 * volumes.c:contains_pending_extent()), to ensure it always
10746 * sees the em, either in the pending_chunks list or in the
10747 * pinned_chunks list.
10749 list_move_tail(&em->list, &fs_info->pinned_chunks);
10751 spin_unlock(&block_group->lock);
10754 struct extent_map_tree *em_tree;
10756 em_tree = &fs_info->mapping_tree.map_tree;
10757 write_lock(&em_tree->lock);
10759 * The em might be in the pending_chunks list, so make sure the
10760 * chunk mutex is locked, since remove_extent_mapping() will
10761 * delete us from that list.
10763 remove_extent_mapping(em_tree, em);
10764 write_unlock(&em_tree->lock);
10765 /* once for the tree */
10766 free_extent_map(em);
10769 mutex_unlock(&fs_info->chunk_mutex);
10771 ret = remove_block_group_free_space(trans, fs_info, block_group);
10775 btrfs_put_block_group(block_group);
10776 btrfs_put_block_group(block_group);
10778 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10784 ret = btrfs_del_item(trans, root, path);
10786 btrfs_free_path(path);
10790 struct btrfs_trans_handle *
10791 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10792 const u64 chunk_offset)
10794 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10795 struct extent_map *em;
10796 struct map_lookup *map;
10797 unsigned int num_items;
10799 read_lock(&em_tree->lock);
10800 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10801 read_unlock(&em_tree->lock);
10802 ASSERT(em && em->start == chunk_offset);
10805 * We need to reserve 3 + N units from the metadata space info in order
10806 * to remove a block group (done at btrfs_remove_chunk() and at
10807 * btrfs_remove_block_group()), which are used for:
10809 * 1 unit for adding the free space inode's orphan (located in the tree
10811 * 1 unit for deleting the block group item (located in the extent
10813 * 1 unit for deleting the free space item (located in tree of tree
10815 * N units for deleting N device extent items corresponding to each
10816 * stripe (located in the device tree).
10818 * In order to remove a block group we also need to reserve units in the
10819 * system space info in order to update the chunk tree (update one or
10820 * more device items and remove one chunk item), but this is done at
10821 * btrfs_remove_chunk() through a call to check_system_chunk().
10823 map = em->map_lookup;
10824 num_items = 3 + map->num_stripes;
10825 free_extent_map(em);
10827 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10832 * Process the unused_bgs list and remove any that don't have any allocated
10833 * space inside of them.
10835 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10837 struct btrfs_block_group_cache *block_group;
10838 struct btrfs_space_info *space_info;
10839 struct btrfs_trans_handle *trans;
10842 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10845 spin_lock(&fs_info->unused_bgs_lock);
10846 while (!list_empty(&fs_info->unused_bgs)) {
10850 block_group = list_first_entry(&fs_info->unused_bgs,
10851 struct btrfs_block_group_cache,
10853 list_del_init(&block_group->bg_list);
10855 space_info = block_group->space_info;
10857 if (ret || btrfs_mixed_space_info(space_info)) {
10858 btrfs_put_block_group(block_group);
10861 spin_unlock(&fs_info->unused_bgs_lock);
10863 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10865 /* Don't want to race with allocators so take the groups_sem */
10866 down_write(&space_info->groups_sem);
10867 spin_lock(&block_group->lock);
10868 if (block_group->reserved || block_group->pinned ||
10869 btrfs_block_group_used(&block_group->item) ||
10871 list_is_singular(&block_group->list)) {
10873 * We want to bail if we made new allocations or have
10874 * outstanding allocations in this block group. We do
10875 * the ro check in case balance is currently acting on
10876 * this block group.
10878 spin_unlock(&block_group->lock);
10879 up_write(&space_info->groups_sem);
10882 spin_unlock(&block_group->lock);
10884 /* We don't want to force the issue, only flip if it's ok. */
10885 ret = inc_block_group_ro(block_group, 0);
10886 up_write(&space_info->groups_sem);
10893 * Want to do this before we do anything else so we can recover
10894 * properly if we fail to join the transaction.
10896 trans = btrfs_start_trans_remove_block_group(fs_info,
10897 block_group->key.objectid);
10898 if (IS_ERR(trans)) {
10899 btrfs_dec_block_group_ro(block_group);
10900 ret = PTR_ERR(trans);
10905 * We could have pending pinned extents for this block group,
10906 * just delete them, we don't care about them anymore.
10908 start = block_group->key.objectid;
10909 end = start + block_group->key.offset - 1;
10911 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10912 * btrfs_finish_extent_commit(). If we are at transaction N,
10913 * another task might be running finish_extent_commit() for the
10914 * previous transaction N - 1, and have seen a range belonging
10915 * to the block group in freed_extents[] before we were able to
10916 * clear the whole block group range from freed_extents[]. This
10917 * means that task can lookup for the block group after we
10918 * unpinned it from freed_extents[] and removed it, leading to
10919 * a BUG_ON() at btrfs_unpin_extent_range().
10921 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10922 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10925 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10926 btrfs_dec_block_group_ro(block_group);
10929 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10932 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10933 btrfs_dec_block_group_ro(block_group);
10936 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10938 /* Reset pinned so btrfs_put_block_group doesn't complain */
10939 spin_lock(&space_info->lock);
10940 spin_lock(&block_group->lock);
10942 space_info->bytes_pinned -= block_group->pinned;
10943 space_info->bytes_readonly += block_group->pinned;
10944 percpu_counter_add(&space_info->total_bytes_pinned,
10945 -block_group->pinned);
10946 block_group->pinned = 0;
10948 spin_unlock(&block_group->lock);
10949 spin_unlock(&space_info->lock);
10951 /* DISCARD can flip during remount */
10952 trimming = btrfs_test_opt(fs_info, DISCARD);
10954 /* Implicit trim during transaction commit. */
10956 btrfs_get_block_group_trimming(block_group);
10959 * Btrfs_remove_chunk will abort the transaction if things go
10962 ret = btrfs_remove_chunk(trans, fs_info,
10963 block_group->key.objectid);
10967 btrfs_put_block_group_trimming(block_group);
10972 * If we're not mounted with -odiscard, we can just forget
10973 * about this block group. Otherwise we'll need to wait
10974 * until transaction commit to do the actual discard.
10977 spin_lock(&fs_info->unused_bgs_lock);
10979 * A concurrent scrub might have added us to the list
10980 * fs_info->unused_bgs, so use a list_move operation
10981 * to add the block group to the deleted_bgs list.
10983 list_move(&block_group->bg_list,
10984 &trans->transaction->deleted_bgs);
10985 spin_unlock(&fs_info->unused_bgs_lock);
10986 btrfs_get_block_group(block_group);
10989 btrfs_end_transaction(trans);
10991 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10992 btrfs_put_block_group(block_group);
10993 spin_lock(&fs_info->unused_bgs_lock);
10995 spin_unlock(&fs_info->unused_bgs_lock);
10998 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
11000 struct btrfs_space_info *space_info;
11001 struct btrfs_super_block *disk_super;
11007 disk_super = fs_info->super_copy;
11008 if (!btrfs_super_root(disk_super))
11011 features = btrfs_super_incompat_flags(disk_super);
11012 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
11015 flags = BTRFS_BLOCK_GROUP_SYSTEM;
11016 ret = create_space_info(fs_info, flags, &space_info);
11021 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
11022 ret = create_space_info(fs_info, flags, &space_info);
11024 flags = BTRFS_BLOCK_GROUP_METADATA;
11025 ret = create_space_info(fs_info, flags, &space_info);
11029 flags = BTRFS_BLOCK_GROUP_DATA;
11030 ret = create_space_info(fs_info, flags, &space_info);
11036 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
11037 u64 start, u64 end)
11039 return unpin_extent_range(fs_info, start, end, false);
11043 * It used to be that old block groups would be left around forever.
11044 * Iterating over them would be enough to trim unused space. Since we
11045 * now automatically remove them, we also need to iterate over unallocated
11048 * We don't want a transaction for this since the discard may take a
11049 * substantial amount of time. We don't require that a transaction be
11050 * running, but we do need to take a running transaction into account
11051 * to ensure that we're not discarding chunks that were released in
11052 * the current transaction.
11054 * Holding the chunks lock will prevent other threads from allocating
11055 * or releasing chunks, but it won't prevent a running transaction
11056 * from committing and releasing the memory that the pending chunks
11057 * list head uses. For that, we need to take a reference to the
11060 static int btrfs_trim_free_extents(struct btrfs_device *device,
11061 u64 minlen, u64 *trimmed)
11063 u64 start = 0, len = 0;
11068 /* Discard not supported = nothing to do. */
11069 if (!blk_queue_discard(bdev_get_queue(device->bdev)))
11072 /* Not writeable = nothing to do. */
11073 if (!device->writeable)
11076 /* No free space = nothing to do. */
11077 if (device->total_bytes <= device->bytes_used)
11083 struct btrfs_fs_info *fs_info = device->fs_info;
11084 struct btrfs_transaction *trans;
11087 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11091 down_read(&fs_info->commit_root_sem);
11093 spin_lock(&fs_info->trans_lock);
11094 trans = fs_info->running_transaction;
11096 refcount_inc(&trans->use_count);
11097 spin_unlock(&fs_info->trans_lock);
11099 ret = find_free_dev_extent_start(trans, device, minlen, start,
11102 btrfs_put_transaction(trans);
11105 up_read(&fs_info->commit_root_sem);
11106 mutex_unlock(&fs_info->chunk_mutex);
11107 if (ret == -ENOSPC)
11112 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11113 up_read(&fs_info->commit_root_sem);
11114 mutex_unlock(&fs_info->chunk_mutex);
11122 if (fatal_signal_pending(current)) {
11123 ret = -ERESTARTSYS;
11134 * Trim the whole filesystem by:
11135 * 1) trimming the free space in each block group
11136 * 2) trimming the unallocated space on each device
11138 * This will also continue trimming even if a block group or device encounters
11139 * an error. The return value will be the last error, or 0 if nothing bad
11142 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11144 struct btrfs_block_group_cache *cache = NULL;
11145 struct btrfs_device *device;
11146 struct list_head *devices;
11152 u64 dev_failed = 0;
11157 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11158 for (; cache; cache = next_block_group(fs_info, cache)) {
11159 if (cache->key.objectid >= (range->start + range->len)) {
11160 btrfs_put_block_group(cache);
11164 start = max(range->start, cache->key.objectid);
11165 end = min(range->start + range->len,
11166 cache->key.objectid + cache->key.offset);
11168 if (end - start >= range->minlen) {
11169 if (!block_group_cache_done(cache)) {
11170 ret = cache_block_group(cache, 0);
11176 ret = wait_block_group_cache_done(cache);
11183 ret = btrfs_trim_block_group(cache,
11189 trimmed += group_trimmed;
11199 btrfs_warn(fs_info,
11200 "failed to trim %llu block group(s), last error %d",
11201 bg_failed, bg_ret);
11202 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11203 devices = &fs_info->fs_devices->devices;
11204 list_for_each_entry(device, devices, dev_list) {
11205 ret = btrfs_trim_free_extents(device, range->minlen,
11213 trimmed += group_trimmed;
11215 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11218 btrfs_warn(fs_info,
11219 "failed to trim %llu device(s), last error %d",
11220 dev_failed, dev_ret);
11221 range->len = trimmed;
11228 * btrfs_{start,end}_write_no_snapshotting() are similar to
11229 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11230 * data into the page cache through nocow before the subvolume is snapshoted,
11231 * but flush the data into disk after the snapshot creation, or to prevent
11232 * operations while snapshotting is ongoing and that cause the snapshot to be
11233 * inconsistent (writes followed by expanding truncates for example).
11235 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11237 percpu_counter_dec(&root->subv_writers->counter);
11239 * Make sure counter is updated before we wake up waiters.
11242 if (waitqueue_active(&root->subv_writers->wait))
11243 wake_up(&root->subv_writers->wait);
11246 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11248 if (atomic_read(&root->will_be_snapshotted))
11251 percpu_counter_inc(&root->subv_writers->counter);
11253 * Make sure counter is updated before we check for snapshot creation.
11256 if (atomic_read(&root->will_be_snapshotted)) {
11257 btrfs_end_write_no_snapshotting(root);
11263 static int wait_snapshotting_atomic_t(atomic_t *a)
11269 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11274 ret = btrfs_start_write_no_snapshotting(root);
11277 wait_on_atomic_t(&root->will_be_snapshotted,
11278 wait_snapshotting_atomic_t,
11279 TASK_UNINTERRUPTIBLE);