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);
1151 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1153 u32 high_crc = ~(u32)0;
1154 u32 low_crc = ~(u32)0;
1157 lenum = cpu_to_le64(root_objectid);
1158 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1159 lenum = cpu_to_le64(owner);
1160 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1161 lenum = cpu_to_le64(offset);
1162 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1164 return ((u64)high_crc << 31) ^ (u64)low_crc;
1167 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1168 struct btrfs_extent_data_ref *ref)
1170 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1171 btrfs_extent_data_ref_objectid(leaf, ref),
1172 btrfs_extent_data_ref_offset(leaf, ref));
1175 static int match_extent_data_ref(struct extent_buffer *leaf,
1176 struct btrfs_extent_data_ref *ref,
1177 u64 root_objectid, u64 owner, u64 offset)
1179 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1180 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1181 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1186 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1187 struct btrfs_fs_info *fs_info,
1188 struct btrfs_path *path,
1189 u64 bytenr, u64 parent,
1191 u64 owner, u64 offset)
1193 struct btrfs_root *root = fs_info->extent_root;
1194 struct btrfs_key key;
1195 struct btrfs_extent_data_ref *ref;
1196 struct extent_buffer *leaf;
1202 key.objectid = bytenr;
1204 key.type = BTRFS_SHARED_DATA_REF_KEY;
1205 key.offset = parent;
1207 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1208 key.offset = hash_extent_data_ref(root_objectid,
1213 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1222 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1223 key.type = BTRFS_EXTENT_REF_V0_KEY;
1224 btrfs_release_path(path);
1225 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1236 leaf = path->nodes[0];
1237 nritems = btrfs_header_nritems(leaf);
1239 if (path->slots[0] >= nritems) {
1240 ret = btrfs_next_leaf(root, path);
1246 leaf = path->nodes[0];
1247 nritems = btrfs_header_nritems(leaf);
1251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1252 if (key.objectid != bytenr ||
1253 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1256 ref = btrfs_item_ptr(leaf, path->slots[0],
1257 struct btrfs_extent_data_ref);
1259 if (match_extent_data_ref(leaf, ref, root_objectid,
1262 btrfs_release_path(path);
1274 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1275 struct btrfs_fs_info *fs_info,
1276 struct btrfs_path *path,
1277 u64 bytenr, u64 parent,
1278 u64 root_objectid, u64 owner,
1279 u64 offset, int refs_to_add)
1281 struct btrfs_root *root = fs_info->extent_root;
1282 struct btrfs_key key;
1283 struct extent_buffer *leaf;
1288 key.objectid = bytenr;
1290 key.type = BTRFS_SHARED_DATA_REF_KEY;
1291 key.offset = parent;
1292 size = sizeof(struct btrfs_shared_data_ref);
1294 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1295 key.offset = hash_extent_data_ref(root_objectid,
1297 size = sizeof(struct btrfs_extent_data_ref);
1300 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1301 if (ret && ret != -EEXIST)
1304 leaf = path->nodes[0];
1306 struct btrfs_shared_data_ref *ref;
1307 ref = btrfs_item_ptr(leaf, path->slots[0],
1308 struct btrfs_shared_data_ref);
1310 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1312 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1313 num_refs += refs_to_add;
1314 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1317 struct btrfs_extent_data_ref *ref;
1318 while (ret == -EEXIST) {
1319 ref = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_extent_data_ref);
1321 if (match_extent_data_ref(leaf, ref, root_objectid,
1324 btrfs_release_path(path);
1326 ret = btrfs_insert_empty_item(trans, root, path, &key,
1328 if (ret && ret != -EEXIST)
1331 leaf = path->nodes[0];
1333 ref = btrfs_item_ptr(leaf, path->slots[0],
1334 struct btrfs_extent_data_ref);
1336 btrfs_set_extent_data_ref_root(leaf, ref,
1338 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1339 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1340 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1343 num_refs += refs_to_add;
1344 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1347 btrfs_mark_buffer_dirty(leaf);
1350 btrfs_release_path(path);
1354 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1355 struct btrfs_fs_info *fs_info,
1356 struct btrfs_path *path,
1357 int refs_to_drop, int *last_ref)
1359 struct btrfs_key key;
1360 struct btrfs_extent_data_ref *ref1 = NULL;
1361 struct btrfs_shared_data_ref *ref2 = NULL;
1362 struct extent_buffer *leaf;
1366 leaf = path->nodes[0];
1367 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1369 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1370 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1371 struct btrfs_extent_data_ref);
1372 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1373 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1374 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1375 struct btrfs_shared_data_ref);
1376 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1377 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1378 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1379 struct btrfs_extent_ref_v0 *ref0;
1380 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1381 struct btrfs_extent_ref_v0);
1382 num_refs = btrfs_ref_count_v0(leaf, ref0);
1388 BUG_ON(num_refs < refs_to_drop);
1389 num_refs -= refs_to_drop;
1391 if (num_refs == 0) {
1392 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1395 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1396 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1397 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1398 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1399 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1401 struct btrfs_extent_ref_v0 *ref0;
1402 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1403 struct btrfs_extent_ref_v0);
1404 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1407 btrfs_mark_buffer_dirty(leaf);
1412 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1413 struct btrfs_extent_inline_ref *iref)
1415 struct btrfs_key key;
1416 struct extent_buffer *leaf;
1417 struct btrfs_extent_data_ref *ref1;
1418 struct btrfs_shared_data_ref *ref2;
1421 leaf = path->nodes[0];
1422 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1424 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1425 BTRFS_EXTENT_DATA_REF_KEY) {
1426 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1427 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1429 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1430 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1432 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1433 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1434 struct btrfs_extent_data_ref);
1435 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1436 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1437 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1438 struct btrfs_shared_data_ref);
1439 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1440 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1441 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1442 struct btrfs_extent_ref_v0 *ref0;
1443 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1444 struct btrfs_extent_ref_v0);
1445 num_refs = btrfs_ref_count_v0(leaf, ref0);
1453 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1454 struct btrfs_fs_info *fs_info,
1455 struct btrfs_path *path,
1456 u64 bytenr, u64 parent,
1459 struct btrfs_root *root = fs_info->extent_root;
1460 struct btrfs_key key;
1463 key.objectid = bytenr;
1465 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1466 key.offset = parent;
1468 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1469 key.offset = root_objectid;
1472 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1475 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1476 if (ret == -ENOENT && parent) {
1477 btrfs_release_path(path);
1478 key.type = BTRFS_EXTENT_REF_V0_KEY;
1479 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1487 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1488 struct btrfs_fs_info *fs_info,
1489 struct btrfs_path *path,
1490 u64 bytenr, u64 parent,
1493 struct btrfs_key key;
1496 key.objectid = bytenr;
1498 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1499 key.offset = parent;
1501 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1502 key.offset = root_objectid;
1505 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1507 btrfs_release_path(path);
1511 static inline int extent_ref_type(u64 parent, u64 owner)
1514 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1516 type = BTRFS_SHARED_BLOCK_REF_KEY;
1518 type = BTRFS_TREE_BLOCK_REF_KEY;
1521 type = BTRFS_SHARED_DATA_REF_KEY;
1523 type = BTRFS_EXTENT_DATA_REF_KEY;
1528 static int find_next_key(struct btrfs_path *path, int level,
1529 struct btrfs_key *key)
1532 for (; level < BTRFS_MAX_LEVEL; level++) {
1533 if (!path->nodes[level])
1535 if (path->slots[level] + 1 >=
1536 btrfs_header_nritems(path->nodes[level]))
1539 btrfs_item_key_to_cpu(path->nodes[level], key,
1540 path->slots[level] + 1);
1542 btrfs_node_key_to_cpu(path->nodes[level], key,
1543 path->slots[level] + 1);
1550 * look for inline back ref. if back ref is found, *ref_ret is set
1551 * to the address of inline back ref, and 0 is returned.
1553 * if back ref isn't found, *ref_ret is set to the address where it
1554 * should be inserted, and -ENOENT is returned.
1556 * if insert is true and there are too many inline back refs, the path
1557 * points to the extent item, and -EAGAIN is returned.
1559 * NOTE: inline back refs are ordered in the same way that back ref
1560 * items in the tree are ordered.
1562 static noinline_for_stack
1563 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1564 struct btrfs_fs_info *fs_info,
1565 struct btrfs_path *path,
1566 struct btrfs_extent_inline_ref **ref_ret,
1567 u64 bytenr, u64 num_bytes,
1568 u64 parent, u64 root_objectid,
1569 u64 owner, u64 offset, int insert)
1571 struct btrfs_root *root = fs_info->extent_root;
1572 struct btrfs_key key;
1573 struct extent_buffer *leaf;
1574 struct btrfs_extent_item *ei;
1575 struct btrfs_extent_inline_ref *iref;
1585 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1587 key.objectid = bytenr;
1588 key.type = BTRFS_EXTENT_ITEM_KEY;
1589 key.offset = num_bytes;
1591 want = extent_ref_type(parent, owner);
1593 extra_size = btrfs_extent_inline_ref_size(want);
1594 path->keep_locks = 1;
1599 * Owner is our parent level, so we can just add one to get the level
1600 * for the block we are interested in.
1602 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1603 key.type = BTRFS_METADATA_ITEM_KEY;
1608 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1615 * We may be a newly converted file system which still has the old fat
1616 * extent entries for metadata, so try and see if we have one of those.
1618 if (ret > 0 && skinny_metadata) {
1619 skinny_metadata = false;
1620 if (path->slots[0]) {
1622 btrfs_item_key_to_cpu(path->nodes[0], &key,
1624 if (key.objectid == bytenr &&
1625 key.type == BTRFS_EXTENT_ITEM_KEY &&
1626 key.offset == num_bytes)
1630 key.objectid = bytenr;
1631 key.type = BTRFS_EXTENT_ITEM_KEY;
1632 key.offset = num_bytes;
1633 btrfs_release_path(path);
1638 if (ret && !insert) {
1641 } else if (WARN_ON(ret)) {
1646 leaf = path->nodes[0];
1647 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1648 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1649 if (item_size < sizeof(*ei)) {
1654 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1660 leaf = path->nodes[0];
1661 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1664 BUG_ON(item_size < sizeof(*ei));
1666 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1667 flags = btrfs_extent_flags(leaf, ei);
1669 ptr = (unsigned long)(ei + 1);
1670 end = (unsigned long)ei + item_size;
1672 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1673 ptr += sizeof(struct btrfs_tree_block_info);
1683 iref = (struct btrfs_extent_inline_ref *)ptr;
1684 type = btrfs_extent_inline_ref_type(leaf, iref);
1688 ptr += btrfs_extent_inline_ref_size(type);
1692 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1693 struct btrfs_extent_data_ref *dref;
1694 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1695 if (match_extent_data_ref(leaf, dref, root_objectid,
1700 if (hash_extent_data_ref_item(leaf, dref) <
1701 hash_extent_data_ref(root_objectid, owner, offset))
1705 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1707 if (parent == ref_offset) {
1711 if (ref_offset < parent)
1714 if (root_objectid == ref_offset) {
1718 if (ref_offset < root_objectid)
1722 ptr += btrfs_extent_inline_ref_size(type);
1724 if (err == -ENOENT && insert) {
1725 if (item_size + extra_size >=
1726 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1731 * To add new inline back ref, we have to make sure
1732 * there is no corresponding back ref item.
1733 * For simplicity, we just do not add new inline back
1734 * ref if there is any kind of item for this block
1736 if (find_next_key(path, 0, &key) == 0 &&
1737 key.objectid == bytenr &&
1738 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1743 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1746 path->keep_locks = 0;
1747 btrfs_unlock_up_safe(path, 1);
1753 * helper to add new inline back ref
1755 static noinline_for_stack
1756 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1757 struct btrfs_path *path,
1758 struct btrfs_extent_inline_ref *iref,
1759 u64 parent, u64 root_objectid,
1760 u64 owner, u64 offset, int refs_to_add,
1761 struct btrfs_delayed_extent_op *extent_op)
1763 struct extent_buffer *leaf;
1764 struct btrfs_extent_item *ei;
1767 unsigned long item_offset;
1772 leaf = path->nodes[0];
1773 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1774 item_offset = (unsigned long)iref - (unsigned long)ei;
1776 type = extent_ref_type(parent, owner);
1777 size = btrfs_extent_inline_ref_size(type);
1779 btrfs_extend_item(fs_info, path, size);
1781 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1782 refs = btrfs_extent_refs(leaf, ei);
1783 refs += refs_to_add;
1784 btrfs_set_extent_refs(leaf, ei, refs);
1786 __run_delayed_extent_op(extent_op, leaf, ei);
1788 ptr = (unsigned long)ei + item_offset;
1789 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1790 if (ptr < end - size)
1791 memmove_extent_buffer(leaf, ptr + size, ptr,
1794 iref = (struct btrfs_extent_inline_ref *)ptr;
1795 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1796 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1797 struct btrfs_extent_data_ref *dref;
1798 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1799 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1800 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1801 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1802 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1803 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1804 struct btrfs_shared_data_ref *sref;
1805 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1806 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1807 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1808 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1809 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1811 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1813 btrfs_mark_buffer_dirty(leaf);
1816 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1817 struct btrfs_fs_info *fs_info,
1818 struct btrfs_path *path,
1819 struct btrfs_extent_inline_ref **ref_ret,
1820 u64 bytenr, u64 num_bytes, u64 parent,
1821 u64 root_objectid, u64 owner, u64 offset)
1825 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1826 bytenr, num_bytes, parent,
1827 root_objectid, owner, offset, 0);
1831 btrfs_release_path(path);
1834 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1835 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1836 parent, root_objectid);
1838 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1839 parent, root_objectid, owner,
1846 * helper to update/remove inline back ref
1848 static noinline_for_stack
1849 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1850 struct btrfs_path *path,
1851 struct btrfs_extent_inline_ref *iref,
1853 struct btrfs_delayed_extent_op *extent_op,
1856 struct extent_buffer *leaf;
1857 struct btrfs_extent_item *ei;
1858 struct btrfs_extent_data_ref *dref = NULL;
1859 struct btrfs_shared_data_ref *sref = NULL;
1867 leaf = path->nodes[0];
1868 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1869 refs = btrfs_extent_refs(leaf, ei);
1870 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1871 refs += refs_to_mod;
1872 btrfs_set_extent_refs(leaf, ei, refs);
1874 __run_delayed_extent_op(extent_op, leaf, ei);
1876 type = btrfs_extent_inline_ref_type(leaf, iref);
1878 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1879 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1880 refs = btrfs_extent_data_ref_count(leaf, dref);
1881 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1882 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1883 refs = btrfs_shared_data_ref_count(leaf, sref);
1886 BUG_ON(refs_to_mod != -1);
1889 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1890 refs += refs_to_mod;
1893 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1894 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1896 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1899 size = btrfs_extent_inline_ref_size(type);
1900 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1901 ptr = (unsigned long)iref;
1902 end = (unsigned long)ei + item_size;
1903 if (ptr + size < end)
1904 memmove_extent_buffer(leaf, ptr, ptr + size,
1907 btrfs_truncate_item(fs_info, path, item_size, 1);
1909 btrfs_mark_buffer_dirty(leaf);
1912 static noinline_for_stack
1913 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1914 struct btrfs_fs_info *fs_info,
1915 struct btrfs_path *path,
1916 u64 bytenr, u64 num_bytes, u64 parent,
1917 u64 root_objectid, u64 owner,
1918 u64 offset, int refs_to_add,
1919 struct btrfs_delayed_extent_op *extent_op)
1921 struct btrfs_extent_inline_ref *iref;
1924 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1925 bytenr, num_bytes, parent,
1926 root_objectid, owner, offset, 1);
1928 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1929 update_inline_extent_backref(fs_info, path, iref,
1930 refs_to_add, extent_op, NULL);
1931 } else if (ret == -ENOENT) {
1932 setup_inline_extent_backref(fs_info, path, iref, parent,
1933 root_objectid, owner, offset,
1934 refs_to_add, extent_op);
1940 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1941 struct btrfs_fs_info *fs_info,
1942 struct btrfs_path *path,
1943 u64 bytenr, u64 parent, u64 root_objectid,
1944 u64 owner, u64 offset, int refs_to_add)
1947 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1948 BUG_ON(refs_to_add != 1);
1949 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
1950 parent, root_objectid);
1952 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
1953 parent, root_objectid,
1954 owner, offset, refs_to_add);
1959 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1960 struct btrfs_fs_info *fs_info,
1961 struct btrfs_path *path,
1962 struct btrfs_extent_inline_ref *iref,
1963 int refs_to_drop, int is_data, int *last_ref)
1967 BUG_ON(!is_data && refs_to_drop != 1);
1969 update_inline_extent_backref(fs_info, path, iref,
1970 -refs_to_drop, NULL, last_ref);
1971 } else if (is_data) {
1972 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
1976 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1981 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1982 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1983 u64 *discarded_bytes)
1986 u64 bytes_left, end;
1987 u64 aligned_start = ALIGN(start, 1 << 9);
1989 if (WARN_ON(start != aligned_start)) {
1990 len -= aligned_start - start;
1991 len = round_down(len, 1 << 9);
1992 start = aligned_start;
1995 *discarded_bytes = 0;
2003 /* Skip any superblocks on this device. */
2004 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2005 u64 sb_start = btrfs_sb_offset(j);
2006 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2007 u64 size = sb_start - start;
2009 if (!in_range(sb_start, start, bytes_left) &&
2010 !in_range(sb_end, start, bytes_left) &&
2011 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2015 * Superblock spans beginning of range. Adjust start and
2018 if (sb_start <= start) {
2019 start += sb_end - start;
2024 bytes_left = end - start;
2029 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2032 *discarded_bytes += size;
2033 else if (ret != -EOPNOTSUPP)
2042 bytes_left = end - start;
2046 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2049 *discarded_bytes += bytes_left;
2054 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2055 u64 num_bytes, u64 *actual_bytes)
2058 u64 discarded_bytes = 0;
2059 struct btrfs_bio *bbio = NULL;
2063 * Avoid races with device replace and make sure our bbio has devices
2064 * associated to its stripes that don't go away while we are discarding.
2066 btrfs_bio_counter_inc_blocked(fs_info);
2067 /* Tell the block device(s) that the sectors can be discarded */
2068 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2070 /* Error condition is -ENOMEM */
2072 struct btrfs_bio_stripe *stripe = bbio->stripes;
2076 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2078 if (!stripe->dev->can_discard)
2081 ret = btrfs_issue_discard(stripe->dev->bdev,
2086 discarded_bytes += bytes;
2087 else if (ret != -EOPNOTSUPP)
2088 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2091 * Just in case we get back EOPNOTSUPP for some reason,
2092 * just ignore the return value so we don't screw up
2093 * people calling discard_extent.
2097 btrfs_put_bbio(bbio);
2099 btrfs_bio_counter_dec(fs_info);
2102 *actual_bytes = discarded_bytes;
2105 if (ret == -EOPNOTSUPP)
2110 /* Can return -ENOMEM */
2111 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2112 struct btrfs_fs_info *fs_info,
2113 u64 bytenr, u64 num_bytes, u64 parent,
2114 u64 root_objectid, u64 owner, u64 offset)
2116 int old_ref_mod, new_ref_mod;
2119 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2120 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2122 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2123 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2125 root_objectid, (int)owner,
2126 BTRFS_ADD_DELAYED_REF, NULL,
2127 &old_ref_mod, &new_ref_mod);
2129 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2131 root_objectid, owner, offset,
2132 0, BTRFS_ADD_DELAYED_REF,
2133 &old_ref_mod, &new_ref_mod);
2136 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2137 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2142 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2143 struct btrfs_fs_info *fs_info,
2144 struct btrfs_delayed_ref_node *node,
2145 u64 parent, u64 root_objectid,
2146 u64 owner, u64 offset, int refs_to_add,
2147 struct btrfs_delayed_extent_op *extent_op)
2149 struct btrfs_path *path;
2150 struct extent_buffer *leaf;
2151 struct btrfs_extent_item *item;
2152 struct btrfs_key key;
2153 u64 bytenr = node->bytenr;
2154 u64 num_bytes = node->num_bytes;
2158 path = btrfs_alloc_path();
2162 path->reada = READA_FORWARD;
2163 path->leave_spinning = 1;
2164 /* this will setup the path even if it fails to insert the back ref */
2165 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2166 num_bytes, parent, root_objectid,
2168 refs_to_add, extent_op);
2169 if ((ret < 0 && ret != -EAGAIN) || !ret)
2173 * Ok we had -EAGAIN which means we didn't have space to insert and
2174 * inline extent ref, so just update the reference count and add a
2177 leaf = path->nodes[0];
2178 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2179 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2180 refs = btrfs_extent_refs(leaf, item);
2181 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2183 __run_delayed_extent_op(extent_op, leaf, item);
2185 btrfs_mark_buffer_dirty(leaf);
2186 btrfs_release_path(path);
2188 path->reada = READA_FORWARD;
2189 path->leave_spinning = 1;
2190 /* now insert the actual backref */
2191 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2192 root_objectid, owner, offset, refs_to_add);
2194 btrfs_abort_transaction(trans, ret);
2196 btrfs_free_path(path);
2200 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2201 struct btrfs_fs_info *fs_info,
2202 struct btrfs_delayed_ref_node *node,
2203 struct btrfs_delayed_extent_op *extent_op,
2204 int insert_reserved)
2207 struct btrfs_delayed_data_ref *ref;
2208 struct btrfs_key ins;
2213 ins.objectid = node->bytenr;
2214 ins.offset = node->num_bytes;
2215 ins.type = BTRFS_EXTENT_ITEM_KEY;
2217 ref = btrfs_delayed_node_to_data_ref(node);
2218 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2220 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2221 parent = ref->parent;
2222 ref_root = ref->root;
2224 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2226 flags |= extent_op->flags_to_set;
2227 ret = alloc_reserved_file_extent(trans, fs_info,
2228 parent, ref_root, flags,
2229 ref->objectid, ref->offset,
2230 &ins, node->ref_mod);
2231 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2232 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2233 ref_root, ref->objectid,
2234 ref->offset, node->ref_mod,
2236 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2237 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2238 ref_root, ref->objectid,
2239 ref->offset, node->ref_mod,
2247 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2248 struct extent_buffer *leaf,
2249 struct btrfs_extent_item *ei)
2251 u64 flags = btrfs_extent_flags(leaf, ei);
2252 if (extent_op->update_flags) {
2253 flags |= extent_op->flags_to_set;
2254 btrfs_set_extent_flags(leaf, ei, flags);
2257 if (extent_op->update_key) {
2258 struct btrfs_tree_block_info *bi;
2259 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2260 bi = (struct btrfs_tree_block_info *)(ei + 1);
2261 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2265 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2266 struct btrfs_fs_info *fs_info,
2267 struct btrfs_delayed_ref_node *node,
2268 struct btrfs_delayed_extent_op *extent_op)
2270 struct btrfs_key key;
2271 struct btrfs_path *path;
2272 struct btrfs_extent_item *ei;
2273 struct extent_buffer *leaf;
2277 int metadata = !extent_op->is_data;
2282 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2285 path = btrfs_alloc_path();
2289 key.objectid = node->bytenr;
2292 key.type = BTRFS_METADATA_ITEM_KEY;
2293 key.offset = extent_op->level;
2295 key.type = BTRFS_EXTENT_ITEM_KEY;
2296 key.offset = node->num_bytes;
2300 path->reada = READA_FORWARD;
2301 path->leave_spinning = 1;
2302 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2309 if (path->slots[0] > 0) {
2311 btrfs_item_key_to_cpu(path->nodes[0], &key,
2313 if (key.objectid == node->bytenr &&
2314 key.type == BTRFS_EXTENT_ITEM_KEY &&
2315 key.offset == node->num_bytes)
2319 btrfs_release_path(path);
2322 key.objectid = node->bytenr;
2323 key.offset = node->num_bytes;
2324 key.type = BTRFS_EXTENT_ITEM_KEY;
2333 leaf = path->nodes[0];
2334 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2335 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2336 if (item_size < sizeof(*ei)) {
2337 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2342 leaf = path->nodes[0];
2343 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2346 BUG_ON(item_size < sizeof(*ei));
2347 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2348 __run_delayed_extent_op(extent_op, leaf, ei);
2350 btrfs_mark_buffer_dirty(leaf);
2352 btrfs_free_path(path);
2356 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2357 struct btrfs_fs_info *fs_info,
2358 struct btrfs_delayed_ref_node *node,
2359 struct btrfs_delayed_extent_op *extent_op,
2360 int insert_reserved)
2363 struct btrfs_delayed_tree_ref *ref;
2364 struct btrfs_key ins;
2367 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2369 ref = btrfs_delayed_node_to_tree_ref(node);
2370 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2372 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2373 parent = ref->parent;
2374 ref_root = ref->root;
2376 ins.objectid = node->bytenr;
2377 if (skinny_metadata) {
2378 ins.offset = ref->level;
2379 ins.type = BTRFS_METADATA_ITEM_KEY;
2381 ins.offset = node->num_bytes;
2382 ins.type = BTRFS_EXTENT_ITEM_KEY;
2385 if (node->ref_mod != 1) {
2387 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2388 node->bytenr, node->ref_mod, node->action, ref_root,
2392 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2393 BUG_ON(!extent_op || !extent_op->update_flags);
2394 ret = alloc_reserved_tree_block(trans, fs_info,
2396 extent_op->flags_to_set,
2399 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2400 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2404 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2405 ret = __btrfs_free_extent(trans, fs_info, node,
2407 ref->level, 0, 1, extent_op);
2414 /* helper function to actually process a single delayed ref entry */
2415 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2416 struct btrfs_fs_info *fs_info,
2417 struct btrfs_delayed_ref_node *node,
2418 struct btrfs_delayed_extent_op *extent_op,
2419 int insert_reserved)
2423 if (trans->aborted) {
2424 if (insert_reserved)
2425 btrfs_pin_extent(fs_info, node->bytenr,
2426 node->num_bytes, 1);
2430 if (btrfs_delayed_ref_is_head(node)) {
2431 struct btrfs_delayed_ref_head *head;
2433 * we've hit the end of the chain and we were supposed
2434 * to insert this extent into the tree. But, it got
2435 * deleted before we ever needed to insert it, so all
2436 * we have to do is clean up the accounting
2439 head = btrfs_delayed_node_to_head(node);
2440 trace_run_delayed_ref_head(fs_info, node, head, node->action);
2442 if (head->total_ref_mod < 0) {
2443 struct btrfs_block_group_cache *cache;
2445 cache = btrfs_lookup_block_group(fs_info, node->bytenr);
2447 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2449 btrfs_put_block_group(cache);
2452 if (insert_reserved) {
2453 btrfs_pin_extent(fs_info, node->bytenr,
2454 node->num_bytes, 1);
2455 if (head->is_data) {
2456 ret = btrfs_del_csums(trans, fs_info,
2462 /* Also free its reserved qgroup space */
2463 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2464 head->qgroup_reserved);
2468 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2469 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2470 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2472 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2473 node->type == BTRFS_SHARED_DATA_REF_KEY)
2474 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2481 static inline struct btrfs_delayed_ref_node *
2482 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2484 struct btrfs_delayed_ref_node *ref;
2486 if (list_empty(&head->ref_list))
2490 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2491 * This is to prevent a ref count from going down to zero, which deletes
2492 * the extent item from the extent tree, when there still are references
2493 * to add, which would fail because they would not find the extent item.
2495 if (!list_empty(&head->ref_add_list))
2496 return list_first_entry(&head->ref_add_list,
2497 struct btrfs_delayed_ref_node, add_list);
2499 ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
2501 ASSERT(list_empty(&ref->add_list));
2506 * Returns 0 on success or if called with an already aborted transaction.
2507 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2509 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2510 struct btrfs_fs_info *fs_info,
2513 struct btrfs_delayed_ref_root *delayed_refs;
2514 struct btrfs_delayed_ref_node *ref;
2515 struct btrfs_delayed_ref_head *locked_ref = NULL;
2516 struct btrfs_delayed_extent_op *extent_op;
2517 ktime_t start = ktime_get();
2519 unsigned long count = 0;
2520 unsigned long actual_count = 0;
2521 int must_insert_reserved = 0;
2523 delayed_refs = &trans->transaction->delayed_refs;
2529 spin_lock(&delayed_refs->lock);
2530 locked_ref = btrfs_select_ref_head(trans);
2532 spin_unlock(&delayed_refs->lock);
2536 /* grab the lock that says we are going to process
2537 * all the refs for this head */
2538 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2539 spin_unlock(&delayed_refs->lock);
2541 * we may have dropped the spin lock to get the head
2542 * mutex lock, and that might have given someone else
2543 * time to free the head. If that's true, it has been
2544 * removed from our list and we can move on.
2546 if (ret == -EAGAIN) {
2554 * We need to try and merge add/drops of the same ref since we
2555 * can run into issues with relocate dropping the implicit ref
2556 * and then it being added back again before the drop can
2557 * finish. If we merged anything we need to re-loop so we can
2559 * Or we can get node references of the same type that weren't
2560 * merged when created due to bumps in the tree mod seq, and
2561 * we need to merge them to prevent adding an inline extent
2562 * backref before dropping it (triggering a BUG_ON at
2563 * insert_inline_extent_backref()).
2565 spin_lock(&locked_ref->lock);
2566 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2570 * locked_ref is the head node, so we have to go one
2571 * node back for any delayed ref updates
2573 ref = select_delayed_ref(locked_ref);
2575 if (ref && ref->seq &&
2576 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2577 spin_unlock(&locked_ref->lock);
2578 spin_lock(&delayed_refs->lock);
2579 locked_ref->processing = 0;
2580 delayed_refs->num_heads_ready++;
2581 spin_unlock(&delayed_refs->lock);
2582 btrfs_delayed_ref_unlock(locked_ref);
2590 * record the must insert reserved flag before we
2591 * drop the spin lock.
2593 must_insert_reserved = locked_ref->must_insert_reserved;
2594 locked_ref->must_insert_reserved = 0;
2596 extent_op = locked_ref->extent_op;
2597 locked_ref->extent_op = NULL;
2602 /* All delayed refs have been processed, Go ahead
2603 * and send the head node to run_one_delayed_ref,
2604 * so that any accounting fixes can happen
2606 ref = &locked_ref->node;
2608 if (extent_op && must_insert_reserved) {
2609 btrfs_free_delayed_extent_op(extent_op);
2614 spin_unlock(&locked_ref->lock);
2615 ret = run_delayed_extent_op(trans, fs_info,
2617 btrfs_free_delayed_extent_op(extent_op);
2621 * Need to reset must_insert_reserved if
2622 * there was an error so the abort stuff
2623 * can cleanup the reserved space
2626 if (must_insert_reserved)
2627 locked_ref->must_insert_reserved = 1;
2628 spin_lock(&delayed_refs->lock);
2629 locked_ref->processing = 0;
2630 delayed_refs->num_heads_ready++;
2631 spin_unlock(&delayed_refs->lock);
2632 btrfs_debug(fs_info,
2633 "run_delayed_extent_op returned %d",
2635 btrfs_delayed_ref_unlock(locked_ref);
2642 * Need to drop our head ref lock and re-acquire the
2643 * delayed ref lock and then re-check to make sure
2646 spin_unlock(&locked_ref->lock);
2647 spin_lock(&delayed_refs->lock);
2648 spin_lock(&locked_ref->lock);
2649 if (!list_empty(&locked_ref->ref_list) ||
2650 locked_ref->extent_op) {
2651 spin_unlock(&locked_ref->lock);
2652 spin_unlock(&delayed_refs->lock);
2656 delayed_refs->num_heads--;
2657 rb_erase(&locked_ref->href_node,
2658 &delayed_refs->href_root);
2659 spin_unlock(&delayed_refs->lock);
2663 list_del(&ref->list);
2664 if (!list_empty(&ref->add_list))
2665 list_del(&ref->add_list);
2667 atomic_dec(&delayed_refs->num_entries);
2669 if (!btrfs_delayed_ref_is_head(ref)) {
2671 * when we play the delayed ref, also correct the
2674 switch (ref->action) {
2675 case BTRFS_ADD_DELAYED_REF:
2676 case BTRFS_ADD_DELAYED_EXTENT:
2677 locked_ref->node.ref_mod -= ref->ref_mod;
2679 case BTRFS_DROP_DELAYED_REF:
2680 locked_ref->node.ref_mod += ref->ref_mod;
2686 spin_unlock(&locked_ref->lock);
2688 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2689 must_insert_reserved);
2691 btrfs_free_delayed_extent_op(extent_op);
2693 spin_lock(&delayed_refs->lock);
2694 locked_ref->processing = 0;
2695 delayed_refs->num_heads_ready++;
2696 spin_unlock(&delayed_refs->lock);
2697 btrfs_delayed_ref_unlock(locked_ref);
2698 btrfs_put_delayed_ref(ref);
2699 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2705 * If this node is a head, that means all the refs in this head
2706 * have been dealt with, and we will pick the next head to deal
2707 * with, so we must unlock the head and drop it from the cluster
2708 * list before we release it.
2710 if (btrfs_delayed_ref_is_head(ref)) {
2711 if (locked_ref->is_data &&
2712 locked_ref->total_ref_mod < 0) {
2713 spin_lock(&delayed_refs->lock);
2714 delayed_refs->pending_csums -= ref->num_bytes;
2715 spin_unlock(&delayed_refs->lock);
2717 btrfs_delayed_ref_unlock(locked_ref);
2720 btrfs_put_delayed_ref(ref);
2726 * We don't want to include ref heads since we can have empty ref heads
2727 * and those will drastically skew our runtime down since we just do
2728 * accounting, no actual extent tree updates.
2730 if (actual_count > 0) {
2731 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2735 * We weigh the current average higher than our current runtime
2736 * to avoid large swings in the average.
2738 spin_lock(&delayed_refs->lock);
2739 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2740 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2741 spin_unlock(&delayed_refs->lock);
2746 #ifdef SCRAMBLE_DELAYED_REFS
2748 * Normally delayed refs get processed in ascending bytenr order. This
2749 * correlates in most cases to the order added. To expose dependencies on this
2750 * order, we start to process the tree in the middle instead of the beginning
2752 static u64 find_middle(struct rb_root *root)
2754 struct rb_node *n = root->rb_node;
2755 struct btrfs_delayed_ref_node *entry;
2758 u64 first = 0, last = 0;
2762 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2763 first = entry->bytenr;
2767 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2768 last = entry->bytenr;
2773 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2774 WARN_ON(!entry->in_tree);
2776 middle = entry->bytenr;
2789 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2793 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2794 sizeof(struct btrfs_extent_inline_ref));
2795 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2796 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2799 * We don't ever fill up leaves all the way so multiply by 2 just to be
2800 * closer to what we're really going to want to use.
2802 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2806 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2807 * would require to store the csums for that many bytes.
2809 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2812 u64 num_csums_per_leaf;
2815 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2816 num_csums_per_leaf = div64_u64(csum_size,
2817 (u64)btrfs_super_csum_size(fs_info->super_copy));
2818 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2819 num_csums += num_csums_per_leaf - 1;
2820 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2824 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2825 struct btrfs_fs_info *fs_info)
2827 struct btrfs_block_rsv *global_rsv;
2828 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2829 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2830 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2831 u64 num_bytes, num_dirty_bgs_bytes;
2834 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2835 num_heads = heads_to_leaves(fs_info, num_heads);
2837 num_bytes += (num_heads - 1) * fs_info->nodesize;
2839 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2841 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2843 global_rsv = &fs_info->global_block_rsv;
2846 * If we can't allocate any more chunks lets make sure we have _lots_ of
2847 * wiggle room since running delayed refs can create more delayed refs.
2849 if (global_rsv->space_info->full) {
2850 num_dirty_bgs_bytes <<= 1;
2854 spin_lock(&global_rsv->lock);
2855 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2857 spin_unlock(&global_rsv->lock);
2861 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2862 struct btrfs_fs_info *fs_info)
2865 atomic_read(&trans->transaction->delayed_refs.num_entries);
2870 avg_runtime = fs_info->avg_delayed_ref_runtime;
2871 val = num_entries * avg_runtime;
2872 if (val >= NSEC_PER_SEC)
2874 if (val >= NSEC_PER_SEC / 2)
2877 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2880 struct async_delayed_refs {
2881 struct btrfs_root *root;
2886 struct completion wait;
2887 struct btrfs_work work;
2890 static inline struct async_delayed_refs *
2891 to_async_delayed_refs(struct btrfs_work *work)
2893 return container_of(work, struct async_delayed_refs, work);
2896 static void delayed_ref_async_start(struct btrfs_work *work)
2898 struct async_delayed_refs *async = to_async_delayed_refs(work);
2899 struct btrfs_trans_handle *trans;
2900 struct btrfs_fs_info *fs_info = async->root->fs_info;
2903 /* if the commit is already started, we don't need to wait here */
2904 if (btrfs_transaction_blocked(fs_info))
2907 trans = btrfs_join_transaction(async->root);
2908 if (IS_ERR(trans)) {
2909 async->error = PTR_ERR(trans);
2914 * trans->sync means that when we call end_transaction, we won't
2915 * wait on delayed refs
2919 /* Don't bother flushing if we got into a different transaction */
2920 if (trans->transid > async->transid)
2923 ret = btrfs_run_delayed_refs(trans, fs_info, async->count);
2927 ret = btrfs_end_transaction(trans);
2928 if (ret && !async->error)
2932 complete(&async->wait);
2937 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2938 unsigned long count, u64 transid, int wait)
2940 struct async_delayed_refs *async;
2943 async = kmalloc(sizeof(*async), GFP_NOFS);
2947 async->root = fs_info->tree_root;
2948 async->count = count;
2950 async->transid = transid;
2955 init_completion(&async->wait);
2957 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2958 delayed_ref_async_start, NULL, NULL);
2960 btrfs_queue_work(fs_info->extent_workers, &async->work);
2963 wait_for_completion(&async->wait);
2972 * this starts processing the delayed reference count updates and
2973 * extent insertions we have queued up so far. count can be
2974 * 0, which means to process everything in the tree at the start
2975 * of the run (but not newly added entries), or it can be some target
2976 * number you'd like to process.
2978 * Returns 0 on success or if called with an aborted transaction
2979 * Returns <0 on error and aborts the transaction
2981 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2982 struct btrfs_fs_info *fs_info, unsigned long count)
2984 struct rb_node *node;
2985 struct btrfs_delayed_ref_root *delayed_refs;
2986 struct btrfs_delayed_ref_head *head;
2988 int run_all = count == (unsigned long)-1;
2989 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2991 /* We'll clean this up in btrfs_cleanup_transaction */
2995 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2998 delayed_refs = &trans->transaction->delayed_refs;
3000 count = atomic_read(&delayed_refs->num_entries) * 2;
3003 #ifdef SCRAMBLE_DELAYED_REFS
3004 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3006 trans->can_flush_pending_bgs = false;
3007 ret = __btrfs_run_delayed_refs(trans, fs_info, count);
3009 btrfs_abort_transaction(trans, ret);
3014 if (!list_empty(&trans->new_bgs))
3015 btrfs_create_pending_block_groups(trans, fs_info);
3017 spin_lock(&delayed_refs->lock);
3018 node = rb_first(&delayed_refs->href_root);
3020 spin_unlock(&delayed_refs->lock);
3025 head = rb_entry(node, struct btrfs_delayed_ref_head,
3027 if (btrfs_delayed_ref_is_head(&head->node)) {
3028 struct btrfs_delayed_ref_node *ref;
3031 refcount_inc(&ref->refs);
3033 spin_unlock(&delayed_refs->lock);
3035 * Mutex was contended, block until it's
3036 * released and try again
3038 mutex_lock(&head->mutex);
3039 mutex_unlock(&head->mutex);
3041 btrfs_put_delayed_ref(ref);
3047 node = rb_next(node);
3049 spin_unlock(&delayed_refs->lock);
3054 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3058 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3059 struct btrfs_fs_info *fs_info,
3060 u64 bytenr, u64 num_bytes, u64 flags,
3061 int level, int is_data)
3063 struct btrfs_delayed_extent_op *extent_op;
3066 extent_op = btrfs_alloc_delayed_extent_op();
3070 extent_op->flags_to_set = flags;
3071 extent_op->update_flags = true;
3072 extent_op->update_key = false;
3073 extent_op->is_data = is_data ? true : false;
3074 extent_op->level = level;
3076 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3077 num_bytes, extent_op);
3079 btrfs_free_delayed_extent_op(extent_op);
3083 static noinline int check_delayed_ref(struct btrfs_root *root,
3084 struct btrfs_path *path,
3085 u64 objectid, u64 offset, u64 bytenr)
3087 struct btrfs_delayed_ref_head *head;
3088 struct btrfs_delayed_ref_node *ref;
3089 struct btrfs_delayed_data_ref *data_ref;
3090 struct btrfs_delayed_ref_root *delayed_refs;
3091 struct btrfs_transaction *cur_trans;
3094 cur_trans = root->fs_info->running_transaction;
3098 delayed_refs = &cur_trans->delayed_refs;
3099 spin_lock(&delayed_refs->lock);
3100 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3102 spin_unlock(&delayed_refs->lock);
3106 if (!mutex_trylock(&head->mutex)) {
3107 refcount_inc(&head->node.refs);
3108 spin_unlock(&delayed_refs->lock);
3110 btrfs_release_path(path);
3113 * Mutex was contended, block until it's released and let
3116 mutex_lock(&head->mutex);
3117 mutex_unlock(&head->mutex);
3118 btrfs_put_delayed_ref(&head->node);
3121 spin_unlock(&delayed_refs->lock);
3123 spin_lock(&head->lock);
3124 list_for_each_entry(ref, &head->ref_list, list) {
3125 /* If it's a shared ref we know a cross reference exists */
3126 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3131 data_ref = btrfs_delayed_node_to_data_ref(ref);
3134 * If our ref doesn't match the one we're currently looking at
3135 * then we have a cross reference.
3137 if (data_ref->root != root->root_key.objectid ||
3138 data_ref->objectid != objectid ||
3139 data_ref->offset != offset) {
3144 spin_unlock(&head->lock);
3145 mutex_unlock(&head->mutex);
3149 static noinline int check_committed_ref(struct btrfs_root *root,
3150 struct btrfs_path *path,
3151 u64 objectid, u64 offset, u64 bytenr)
3153 struct btrfs_fs_info *fs_info = root->fs_info;
3154 struct btrfs_root *extent_root = fs_info->extent_root;
3155 struct extent_buffer *leaf;
3156 struct btrfs_extent_data_ref *ref;
3157 struct btrfs_extent_inline_ref *iref;
3158 struct btrfs_extent_item *ei;
3159 struct btrfs_key key;
3163 key.objectid = bytenr;
3164 key.offset = (u64)-1;
3165 key.type = BTRFS_EXTENT_ITEM_KEY;
3167 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3170 BUG_ON(ret == 0); /* Corruption */
3173 if (path->slots[0] == 0)
3177 leaf = path->nodes[0];
3178 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3180 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3184 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3185 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3186 if (item_size < sizeof(*ei)) {
3187 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3191 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3193 if (item_size != sizeof(*ei) +
3194 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3197 if (btrfs_extent_generation(leaf, ei) <=
3198 btrfs_root_last_snapshot(&root->root_item))
3201 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3202 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3203 BTRFS_EXTENT_DATA_REF_KEY)
3206 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3207 if (btrfs_extent_refs(leaf, ei) !=
3208 btrfs_extent_data_ref_count(leaf, ref) ||
3209 btrfs_extent_data_ref_root(leaf, ref) !=
3210 root->root_key.objectid ||
3211 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3212 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3220 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3223 struct btrfs_path *path;
3227 path = btrfs_alloc_path();
3232 ret = check_committed_ref(root, path, objectid,
3234 if (ret && ret != -ENOENT)
3237 ret2 = check_delayed_ref(root, path, objectid,
3239 } while (ret2 == -EAGAIN);
3241 if (ret2 && ret2 != -ENOENT) {
3246 if (ret != -ENOENT || ret2 != -ENOENT)
3249 btrfs_free_path(path);
3250 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3255 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3256 struct btrfs_root *root,
3257 struct extent_buffer *buf,
3258 int full_backref, int inc)
3260 struct btrfs_fs_info *fs_info = root->fs_info;
3266 struct btrfs_key key;
3267 struct btrfs_file_extent_item *fi;
3271 int (*process_func)(struct btrfs_trans_handle *,
3272 struct btrfs_fs_info *,
3273 u64, u64, u64, u64, u64, u64);
3276 if (btrfs_is_testing(fs_info))
3279 ref_root = btrfs_header_owner(buf);
3280 nritems = btrfs_header_nritems(buf);
3281 level = btrfs_header_level(buf);
3283 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3287 process_func = btrfs_inc_extent_ref;
3289 process_func = btrfs_free_extent;
3292 parent = buf->start;
3296 for (i = 0; i < nritems; i++) {
3298 btrfs_item_key_to_cpu(buf, &key, i);
3299 if (key.type != BTRFS_EXTENT_DATA_KEY)
3301 fi = btrfs_item_ptr(buf, i,
3302 struct btrfs_file_extent_item);
3303 if (btrfs_file_extent_type(buf, fi) ==
3304 BTRFS_FILE_EXTENT_INLINE)
3306 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3310 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3311 key.offset -= btrfs_file_extent_offset(buf, fi);
3312 ret = process_func(trans, fs_info, bytenr, num_bytes,
3313 parent, ref_root, key.objectid,
3318 bytenr = btrfs_node_blockptr(buf, i);
3319 num_bytes = fs_info->nodesize;
3320 ret = process_func(trans, fs_info, bytenr, num_bytes,
3321 parent, ref_root, level - 1, 0);
3331 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3332 struct extent_buffer *buf, int full_backref)
3334 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3337 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3338 struct extent_buffer *buf, int full_backref)
3340 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3343 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3344 struct btrfs_fs_info *fs_info,
3345 struct btrfs_path *path,
3346 struct btrfs_block_group_cache *cache)
3349 struct btrfs_root *extent_root = fs_info->extent_root;
3351 struct extent_buffer *leaf;
3353 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3360 leaf = path->nodes[0];
3361 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3362 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3363 btrfs_mark_buffer_dirty(leaf);
3365 btrfs_release_path(path);
3370 static struct btrfs_block_group_cache *
3371 next_block_group(struct btrfs_fs_info *fs_info,
3372 struct btrfs_block_group_cache *cache)
3374 struct rb_node *node;
3376 spin_lock(&fs_info->block_group_cache_lock);
3378 /* If our block group was removed, we need a full search. */
3379 if (RB_EMPTY_NODE(&cache->cache_node)) {
3380 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3382 spin_unlock(&fs_info->block_group_cache_lock);
3383 btrfs_put_block_group(cache);
3384 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3386 node = rb_next(&cache->cache_node);
3387 btrfs_put_block_group(cache);
3389 cache = rb_entry(node, struct btrfs_block_group_cache,
3391 btrfs_get_block_group(cache);
3394 spin_unlock(&fs_info->block_group_cache_lock);
3398 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3399 struct btrfs_trans_handle *trans,
3400 struct btrfs_path *path)
3402 struct btrfs_fs_info *fs_info = block_group->fs_info;
3403 struct btrfs_root *root = fs_info->tree_root;
3404 struct inode *inode = NULL;
3405 struct extent_changeset *data_reserved = NULL;
3407 int dcs = BTRFS_DC_ERROR;
3413 * If this block group is smaller than 100 megs don't bother caching the
3416 if (block_group->key.offset < (100 * SZ_1M)) {
3417 spin_lock(&block_group->lock);
3418 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3419 spin_unlock(&block_group->lock);
3426 inode = lookup_free_space_inode(fs_info, block_group, path);
3427 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3428 ret = PTR_ERR(inode);
3429 btrfs_release_path(path);
3433 if (IS_ERR(inode)) {
3437 if (block_group->ro)
3440 ret = create_free_space_inode(fs_info, trans, block_group,
3447 /* We've already setup this transaction, go ahead and exit */
3448 if (block_group->cache_generation == trans->transid &&
3449 i_size_read(inode)) {
3450 dcs = BTRFS_DC_SETUP;
3455 * We want to set the generation to 0, that way if anything goes wrong
3456 * from here on out we know not to trust this cache when we load up next
3459 BTRFS_I(inode)->generation = 0;
3460 ret = btrfs_update_inode(trans, root, inode);
3463 * So theoretically we could recover from this, simply set the
3464 * super cache generation to 0 so we know to invalidate the
3465 * cache, but then we'd have to keep track of the block groups
3466 * that fail this way so we know we _have_ to reset this cache
3467 * before the next commit or risk reading stale cache. So to
3468 * limit our exposure to horrible edge cases lets just abort the
3469 * transaction, this only happens in really bad situations
3472 btrfs_abort_transaction(trans, ret);
3477 if (i_size_read(inode) > 0) {
3478 ret = btrfs_check_trunc_cache_free_space(fs_info,
3479 &fs_info->global_block_rsv);
3483 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3488 spin_lock(&block_group->lock);
3489 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3490 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3492 * don't bother trying to write stuff out _if_
3493 * a) we're not cached,
3494 * b) we're with nospace_cache mount option,
3495 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3497 dcs = BTRFS_DC_WRITTEN;
3498 spin_unlock(&block_group->lock);
3501 spin_unlock(&block_group->lock);
3504 * We hit an ENOSPC when setting up the cache in this transaction, just
3505 * skip doing the setup, we've already cleared the cache so we're safe.
3507 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3513 * Try to preallocate enough space based on how big the block group is.
3514 * Keep in mind this has to include any pinned space which could end up
3515 * taking up quite a bit since it's not folded into the other space
3518 num_pages = div_u64(block_group->key.offset, SZ_256M);
3523 num_pages *= PAGE_SIZE;
3525 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3529 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3530 num_pages, num_pages,
3533 * Our cache requires contiguous chunks so that we don't modify a bunch
3534 * of metadata or split extents when writing the cache out, which means
3535 * we can enospc if we are heavily fragmented in addition to just normal
3536 * out of space conditions. So if we hit this just skip setting up any
3537 * other block groups for this transaction, maybe we'll unpin enough
3538 * space the next time around.
3541 dcs = BTRFS_DC_SETUP;
3542 else if (ret == -ENOSPC)
3543 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3548 btrfs_release_path(path);
3550 spin_lock(&block_group->lock);
3551 if (!ret && dcs == BTRFS_DC_SETUP)
3552 block_group->cache_generation = trans->transid;
3553 block_group->disk_cache_state = dcs;
3554 spin_unlock(&block_group->lock);
3556 extent_changeset_free(data_reserved);
3560 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3561 struct btrfs_fs_info *fs_info)
3563 struct btrfs_block_group_cache *cache, *tmp;
3564 struct btrfs_transaction *cur_trans = trans->transaction;
3565 struct btrfs_path *path;
3567 if (list_empty(&cur_trans->dirty_bgs) ||
3568 !btrfs_test_opt(fs_info, SPACE_CACHE))
3571 path = btrfs_alloc_path();
3575 /* Could add new block groups, use _safe just in case */
3576 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3578 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3579 cache_save_setup(cache, trans, path);
3582 btrfs_free_path(path);
3587 * transaction commit does final block group cache writeback during a
3588 * critical section where nothing is allowed to change the FS. This is
3589 * required in order for the cache to actually match the block group,
3590 * but can introduce a lot of latency into the commit.
3592 * So, btrfs_start_dirty_block_groups is here to kick off block group
3593 * cache IO. There's a chance we'll have to redo some of it if the
3594 * block group changes again during the commit, but it greatly reduces
3595 * the commit latency by getting rid of the easy block groups while
3596 * we're still allowing others to join the commit.
3598 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3599 struct btrfs_fs_info *fs_info)
3601 struct btrfs_block_group_cache *cache;
3602 struct btrfs_transaction *cur_trans = trans->transaction;
3605 struct btrfs_path *path = NULL;
3607 struct list_head *io = &cur_trans->io_bgs;
3608 int num_started = 0;
3611 spin_lock(&cur_trans->dirty_bgs_lock);
3612 if (list_empty(&cur_trans->dirty_bgs)) {
3613 spin_unlock(&cur_trans->dirty_bgs_lock);
3616 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3617 spin_unlock(&cur_trans->dirty_bgs_lock);
3621 * make sure all the block groups on our dirty list actually
3624 btrfs_create_pending_block_groups(trans, fs_info);
3627 path = btrfs_alloc_path();
3633 * cache_write_mutex is here only to save us from balance or automatic
3634 * removal of empty block groups deleting this block group while we are
3635 * writing out the cache
3637 mutex_lock(&trans->transaction->cache_write_mutex);
3638 while (!list_empty(&dirty)) {
3639 cache = list_first_entry(&dirty,
3640 struct btrfs_block_group_cache,
3643 * this can happen if something re-dirties a block
3644 * group that is already under IO. Just wait for it to
3645 * finish and then do it all again
3647 if (!list_empty(&cache->io_list)) {
3648 list_del_init(&cache->io_list);
3649 btrfs_wait_cache_io(trans, cache, path);
3650 btrfs_put_block_group(cache);
3655 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3656 * if it should update the cache_state. Don't delete
3657 * until after we wait.
3659 * Since we're not running in the commit critical section
3660 * we need the dirty_bgs_lock to protect from update_block_group
3662 spin_lock(&cur_trans->dirty_bgs_lock);
3663 list_del_init(&cache->dirty_list);
3664 spin_unlock(&cur_trans->dirty_bgs_lock);
3668 cache_save_setup(cache, trans, path);
3670 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3671 cache->io_ctl.inode = NULL;
3672 ret = btrfs_write_out_cache(fs_info, trans,
3674 if (ret == 0 && cache->io_ctl.inode) {
3679 * the cache_write_mutex is protecting
3682 list_add_tail(&cache->io_list, io);
3685 * if we failed to write the cache, the
3686 * generation will be bad and life goes on
3692 ret = write_one_cache_group(trans, fs_info,
3695 * Our block group might still be attached to the list
3696 * of new block groups in the transaction handle of some
3697 * other task (struct btrfs_trans_handle->new_bgs). This
3698 * means its block group item isn't yet in the extent
3699 * tree. If this happens ignore the error, as we will
3700 * try again later in the critical section of the
3701 * transaction commit.
3703 if (ret == -ENOENT) {
3705 spin_lock(&cur_trans->dirty_bgs_lock);
3706 if (list_empty(&cache->dirty_list)) {
3707 list_add_tail(&cache->dirty_list,
3708 &cur_trans->dirty_bgs);
3709 btrfs_get_block_group(cache);
3711 spin_unlock(&cur_trans->dirty_bgs_lock);
3713 btrfs_abort_transaction(trans, ret);
3717 /* if its not on the io list, we need to put the block group */
3719 btrfs_put_block_group(cache);
3725 * Avoid blocking other tasks for too long. It might even save
3726 * us from writing caches for block groups that are going to be
3729 mutex_unlock(&trans->transaction->cache_write_mutex);
3730 mutex_lock(&trans->transaction->cache_write_mutex);
3732 mutex_unlock(&trans->transaction->cache_write_mutex);
3735 * go through delayed refs for all the stuff we've just kicked off
3736 * and then loop back (just once)
3738 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
3739 if (!ret && loops == 0) {
3741 spin_lock(&cur_trans->dirty_bgs_lock);
3742 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3744 * dirty_bgs_lock protects us from concurrent block group
3745 * deletes too (not just cache_write_mutex).
3747 if (!list_empty(&dirty)) {
3748 spin_unlock(&cur_trans->dirty_bgs_lock);
3751 spin_unlock(&cur_trans->dirty_bgs_lock);
3752 } else if (ret < 0) {
3753 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3756 btrfs_free_path(path);
3760 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3761 struct btrfs_fs_info *fs_info)
3763 struct btrfs_block_group_cache *cache;
3764 struct btrfs_transaction *cur_trans = trans->transaction;
3767 struct btrfs_path *path;
3768 struct list_head *io = &cur_trans->io_bgs;
3769 int num_started = 0;
3771 path = btrfs_alloc_path();
3776 * Even though we are in the critical section of the transaction commit,
3777 * we can still have concurrent tasks adding elements to this
3778 * transaction's list of dirty block groups. These tasks correspond to
3779 * endio free space workers started when writeback finishes for a
3780 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3781 * allocate new block groups as a result of COWing nodes of the root
3782 * tree when updating the free space inode. The writeback for the space
3783 * caches is triggered by an earlier call to
3784 * btrfs_start_dirty_block_groups() and iterations of the following
3786 * Also we want to do the cache_save_setup first and then run the
3787 * delayed refs to make sure we have the best chance at doing this all
3790 spin_lock(&cur_trans->dirty_bgs_lock);
3791 while (!list_empty(&cur_trans->dirty_bgs)) {
3792 cache = list_first_entry(&cur_trans->dirty_bgs,
3793 struct btrfs_block_group_cache,
3797 * this can happen if cache_save_setup re-dirties a block
3798 * group that is already under IO. Just wait for it to
3799 * finish and then do it all again
3801 if (!list_empty(&cache->io_list)) {
3802 spin_unlock(&cur_trans->dirty_bgs_lock);
3803 list_del_init(&cache->io_list);
3804 btrfs_wait_cache_io(trans, cache, path);
3805 btrfs_put_block_group(cache);
3806 spin_lock(&cur_trans->dirty_bgs_lock);
3810 * don't remove from the dirty list until after we've waited
3813 list_del_init(&cache->dirty_list);
3814 spin_unlock(&cur_trans->dirty_bgs_lock);
3817 cache_save_setup(cache, trans, path);
3820 ret = btrfs_run_delayed_refs(trans, fs_info,
3821 (unsigned long) -1);
3823 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3824 cache->io_ctl.inode = NULL;
3825 ret = btrfs_write_out_cache(fs_info, trans,
3827 if (ret == 0 && cache->io_ctl.inode) {
3830 list_add_tail(&cache->io_list, io);
3833 * if we failed to write the cache, the
3834 * generation will be bad and life goes on
3840 ret = write_one_cache_group(trans, fs_info,
3843 * One of the free space endio workers might have
3844 * created a new block group while updating a free space
3845 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3846 * and hasn't released its transaction handle yet, in
3847 * which case the new block group is still attached to
3848 * its transaction handle and its creation has not
3849 * finished yet (no block group item in the extent tree
3850 * yet, etc). If this is the case, wait for all free
3851 * space endio workers to finish and retry. This is a
3852 * a very rare case so no need for a more efficient and
3855 if (ret == -ENOENT) {
3856 wait_event(cur_trans->writer_wait,
3857 atomic_read(&cur_trans->num_writers) == 1);
3858 ret = write_one_cache_group(trans, fs_info,
3862 btrfs_abort_transaction(trans, ret);
3865 /* if its not on the io list, we need to put the block group */
3867 btrfs_put_block_group(cache);
3868 spin_lock(&cur_trans->dirty_bgs_lock);
3870 spin_unlock(&cur_trans->dirty_bgs_lock);
3872 while (!list_empty(io)) {
3873 cache = list_first_entry(io, struct btrfs_block_group_cache,
3875 list_del_init(&cache->io_list);
3876 btrfs_wait_cache_io(trans, cache, path);
3877 btrfs_put_block_group(cache);
3880 btrfs_free_path(path);
3884 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3886 struct btrfs_block_group_cache *block_group;
3889 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3890 if (!block_group || block_group->ro)
3893 btrfs_put_block_group(block_group);
3897 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3899 struct btrfs_block_group_cache *bg;
3902 bg = btrfs_lookup_block_group(fs_info, bytenr);
3906 spin_lock(&bg->lock);
3910 atomic_inc(&bg->nocow_writers);
3911 spin_unlock(&bg->lock);
3913 /* no put on block group, done by btrfs_dec_nocow_writers */
3915 btrfs_put_block_group(bg);
3921 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3923 struct btrfs_block_group_cache *bg;
3925 bg = btrfs_lookup_block_group(fs_info, bytenr);
3927 if (atomic_dec_and_test(&bg->nocow_writers))
3928 wake_up_atomic_t(&bg->nocow_writers);
3930 * Once for our lookup and once for the lookup done by a previous call
3931 * to btrfs_inc_nocow_writers()
3933 btrfs_put_block_group(bg);
3934 btrfs_put_block_group(bg);
3937 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3943 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3945 wait_on_atomic_t(&bg->nocow_writers,
3946 btrfs_wait_nocow_writers_atomic_t,
3947 TASK_UNINTERRUPTIBLE);
3950 static const char *alloc_name(u64 flags)
3953 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3955 case BTRFS_BLOCK_GROUP_METADATA:
3957 case BTRFS_BLOCK_GROUP_DATA:
3959 case BTRFS_BLOCK_GROUP_SYSTEM:
3963 return "invalid-combination";
3967 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
3968 struct btrfs_space_info **new)
3971 struct btrfs_space_info *space_info;
3975 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3979 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3986 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3987 INIT_LIST_HEAD(&space_info->block_groups[i]);
3988 init_rwsem(&space_info->groups_sem);
3989 spin_lock_init(&space_info->lock);
3990 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3991 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3992 init_waitqueue_head(&space_info->wait);
3993 INIT_LIST_HEAD(&space_info->ro_bgs);
3994 INIT_LIST_HEAD(&space_info->tickets);
3995 INIT_LIST_HEAD(&space_info->priority_tickets);
3997 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3998 info->space_info_kobj, "%s",
3999 alloc_name(space_info->flags));
4001 percpu_counter_destroy(&space_info->total_bytes_pinned);
4007 list_add_rcu(&space_info->list, &info->space_info);
4008 if (flags & BTRFS_BLOCK_GROUP_DATA)
4009 info->data_sinfo = space_info;
4014 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4015 u64 total_bytes, u64 bytes_used,
4017 struct btrfs_space_info **space_info)
4019 struct btrfs_space_info *found;
4022 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4023 BTRFS_BLOCK_GROUP_RAID10))
4028 found = __find_space_info(info, flags);
4030 spin_lock(&found->lock);
4031 found->total_bytes += total_bytes;
4032 found->disk_total += total_bytes * factor;
4033 found->bytes_used += bytes_used;
4034 found->disk_used += bytes_used * factor;
4035 found->bytes_readonly += bytes_readonly;
4036 if (total_bytes > 0)
4038 space_info_add_new_bytes(info, found, total_bytes -
4039 bytes_used - bytes_readonly);
4040 spin_unlock(&found->lock);
4041 *space_info = found;
4044 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4046 u64 extra_flags = chunk_to_extended(flags) &
4047 BTRFS_EXTENDED_PROFILE_MASK;
4049 write_seqlock(&fs_info->profiles_lock);
4050 if (flags & BTRFS_BLOCK_GROUP_DATA)
4051 fs_info->avail_data_alloc_bits |= extra_flags;
4052 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4053 fs_info->avail_metadata_alloc_bits |= extra_flags;
4054 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4055 fs_info->avail_system_alloc_bits |= extra_flags;
4056 write_sequnlock(&fs_info->profiles_lock);
4060 * returns target flags in extended format or 0 if restripe for this
4061 * chunk_type is not in progress
4063 * should be called with either volume_mutex or balance_lock held
4065 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4067 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4073 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4074 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4075 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4076 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4077 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4078 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4079 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4080 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4081 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4088 * @flags: available profiles in extended format (see ctree.h)
4090 * Returns reduced profile in chunk format. If profile changing is in
4091 * progress (either running or paused) picks the target profile (if it's
4092 * already available), otherwise falls back to plain reducing.
4094 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4096 u64 num_devices = fs_info->fs_devices->rw_devices;
4102 * see if restripe for this chunk_type is in progress, if so
4103 * try to reduce to the target profile
4105 spin_lock(&fs_info->balance_lock);
4106 target = get_restripe_target(fs_info, flags);
4108 /* pick target profile only if it's already available */
4109 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4110 spin_unlock(&fs_info->balance_lock);
4111 return extended_to_chunk(target);
4114 spin_unlock(&fs_info->balance_lock);
4116 /* First, mask out the RAID levels which aren't possible */
4117 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4118 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4119 allowed |= btrfs_raid_group[raid_type];
4123 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4124 allowed = BTRFS_BLOCK_GROUP_RAID6;
4125 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4126 allowed = BTRFS_BLOCK_GROUP_RAID5;
4127 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4128 allowed = BTRFS_BLOCK_GROUP_RAID10;
4129 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4130 allowed = BTRFS_BLOCK_GROUP_RAID1;
4131 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4132 allowed = BTRFS_BLOCK_GROUP_RAID0;
4134 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4136 return extended_to_chunk(flags | allowed);
4139 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4146 seq = read_seqbegin(&fs_info->profiles_lock);
4148 if (flags & BTRFS_BLOCK_GROUP_DATA)
4149 flags |= fs_info->avail_data_alloc_bits;
4150 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4151 flags |= fs_info->avail_system_alloc_bits;
4152 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4153 flags |= fs_info->avail_metadata_alloc_bits;
4154 } while (read_seqretry(&fs_info->profiles_lock, seq));
4156 return btrfs_reduce_alloc_profile(fs_info, flags);
4159 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4161 struct btrfs_fs_info *fs_info = root->fs_info;
4166 flags = BTRFS_BLOCK_GROUP_DATA;
4167 else if (root == fs_info->chunk_root)
4168 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4170 flags = BTRFS_BLOCK_GROUP_METADATA;
4172 ret = get_alloc_profile(fs_info, flags);
4176 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4178 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4181 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4183 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4186 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4188 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4191 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4192 bool may_use_included)
4195 return s_info->bytes_used + s_info->bytes_reserved +
4196 s_info->bytes_pinned + s_info->bytes_readonly +
4197 (may_use_included ? s_info->bytes_may_use : 0);
4200 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4202 struct btrfs_space_info *data_sinfo;
4203 struct btrfs_root *root = inode->root;
4204 struct btrfs_fs_info *fs_info = root->fs_info;
4207 int need_commit = 2;
4208 int have_pinned_space;
4210 /* make sure bytes are sectorsize aligned */
4211 bytes = ALIGN(bytes, fs_info->sectorsize);
4213 if (btrfs_is_free_space_inode(inode)) {
4215 ASSERT(current->journal_info);
4218 data_sinfo = fs_info->data_sinfo;
4223 /* make sure we have enough space to handle the data first */
4224 spin_lock(&data_sinfo->lock);
4225 used = btrfs_space_info_used(data_sinfo, true);
4227 if (used + bytes > data_sinfo->total_bytes) {
4228 struct btrfs_trans_handle *trans;
4231 * if we don't have enough free bytes in this space then we need
4232 * to alloc a new chunk.
4234 if (!data_sinfo->full) {
4237 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4238 spin_unlock(&data_sinfo->lock);
4240 alloc_target = btrfs_data_alloc_profile(fs_info);
4242 * It is ugly that we don't call nolock join
4243 * transaction for the free space inode case here.
4244 * But it is safe because we only do the data space
4245 * reservation for the free space cache in the
4246 * transaction context, the common join transaction
4247 * just increase the counter of the current transaction
4248 * handler, doesn't try to acquire the trans_lock of
4251 trans = btrfs_join_transaction(root);
4253 return PTR_ERR(trans);
4255 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4256 CHUNK_ALLOC_NO_FORCE);
4257 btrfs_end_transaction(trans);
4262 have_pinned_space = 1;
4268 data_sinfo = fs_info->data_sinfo;
4274 * If we don't have enough pinned space to deal with this
4275 * allocation, and no removed chunk in current transaction,
4276 * don't bother committing the transaction.
4278 have_pinned_space = percpu_counter_compare(
4279 &data_sinfo->total_bytes_pinned,
4280 used + bytes - data_sinfo->total_bytes);
4281 spin_unlock(&data_sinfo->lock);
4283 /* commit the current transaction and try again */
4286 !atomic_read(&fs_info->open_ioctl_trans)) {
4289 if (need_commit > 0) {
4290 btrfs_start_delalloc_roots(fs_info, 0, -1);
4291 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4295 trans = btrfs_join_transaction(root);
4297 return PTR_ERR(trans);
4298 if (have_pinned_space >= 0 ||
4299 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4300 &trans->transaction->flags) ||
4302 ret = btrfs_commit_transaction(trans);
4306 * The cleaner kthread might still be doing iput
4307 * operations. Wait for it to finish so that
4308 * more space is released.
4310 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4311 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4314 btrfs_end_transaction(trans);
4318 trace_btrfs_space_reservation(fs_info,
4319 "space_info:enospc",
4320 data_sinfo->flags, bytes, 1);
4323 data_sinfo->bytes_may_use += bytes;
4324 trace_btrfs_space_reservation(fs_info, "space_info",
4325 data_sinfo->flags, bytes, 1);
4326 spin_unlock(&data_sinfo->lock);
4331 int btrfs_check_data_free_space(struct inode *inode,
4332 struct extent_changeset **reserved, u64 start, u64 len)
4334 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4337 /* align the range */
4338 len = round_up(start + len, fs_info->sectorsize) -
4339 round_down(start, fs_info->sectorsize);
4340 start = round_down(start, fs_info->sectorsize);
4342 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4346 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4347 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4349 btrfs_free_reserved_data_space_noquota(inode, start, len);
4356 * Called if we need to clear a data reservation for this inode
4357 * Normally in a error case.
4359 * This one will *NOT* use accurate qgroup reserved space API, just for case
4360 * which we can't sleep and is sure it won't affect qgroup reserved space.
4361 * Like clear_bit_hook().
4363 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4366 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4367 struct btrfs_space_info *data_sinfo;
4369 /* Make sure the range is aligned to sectorsize */
4370 len = round_up(start + len, fs_info->sectorsize) -
4371 round_down(start, fs_info->sectorsize);
4372 start = round_down(start, fs_info->sectorsize);
4374 data_sinfo = fs_info->data_sinfo;
4375 spin_lock(&data_sinfo->lock);
4376 if (WARN_ON(data_sinfo->bytes_may_use < len))
4377 data_sinfo->bytes_may_use = 0;
4379 data_sinfo->bytes_may_use -= len;
4380 trace_btrfs_space_reservation(fs_info, "space_info",
4381 data_sinfo->flags, len, 0);
4382 spin_unlock(&data_sinfo->lock);
4386 * Called if we need to clear a data reservation for this inode
4387 * Normally in a error case.
4389 * This one will handle the per-inode data rsv map for accurate reserved
4392 void btrfs_free_reserved_data_space(struct inode *inode,
4393 struct extent_changeset *reserved, u64 start, u64 len)
4395 struct btrfs_root *root = BTRFS_I(inode)->root;
4397 /* Make sure the range is aligned to sectorsize */
4398 len = round_up(start + len, root->fs_info->sectorsize) -
4399 round_down(start, root->fs_info->sectorsize);
4400 start = round_down(start, root->fs_info->sectorsize);
4402 btrfs_free_reserved_data_space_noquota(inode, start, len);
4403 btrfs_qgroup_free_data(inode, reserved, start, len);
4406 static void force_metadata_allocation(struct btrfs_fs_info *info)
4408 struct list_head *head = &info->space_info;
4409 struct btrfs_space_info *found;
4412 list_for_each_entry_rcu(found, head, list) {
4413 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4414 found->force_alloc = CHUNK_ALLOC_FORCE;
4419 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4421 return (global->size << 1);
4424 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4425 struct btrfs_space_info *sinfo, int force)
4427 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4428 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4429 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4432 if (force == CHUNK_ALLOC_FORCE)
4436 * We need to take into account the global rsv because for all intents
4437 * and purposes it's used space. Don't worry about locking the
4438 * global_rsv, it doesn't change except when the transaction commits.
4440 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4441 num_allocated += calc_global_rsv_need_space(global_rsv);
4444 * in limited mode, we want to have some free space up to
4445 * about 1% of the FS size.
4447 if (force == CHUNK_ALLOC_LIMITED) {
4448 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4449 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4451 if (num_bytes - num_allocated < thresh)
4455 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4460 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4464 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4465 BTRFS_BLOCK_GROUP_RAID0 |
4466 BTRFS_BLOCK_GROUP_RAID5 |
4467 BTRFS_BLOCK_GROUP_RAID6))
4468 num_dev = fs_info->fs_devices->rw_devices;
4469 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4472 num_dev = 1; /* DUP or single */
4478 * If @is_allocation is true, reserve space in the system space info necessary
4479 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4482 void check_system_chunk(struct btrfs_trans_handle *trans,
4483 struct btrfs_fs_info *fs_info, u64 type)
4485 struct btrfs_space_info *info;
4492 * Needed because we can end up allocating a system chunk and for an
4493 * atomic and race free space reservation in the chunk block reserve.
4495 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4497 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4498 spin_lock(&info->lock);
4499 left = info->total_bytes - btrfs_space_info_used(info, true);
4500 spin_unlock(&info->lock);
4502 num_devs = get_profile_num_devs(fs_info, type);
4504 /* num_devs device items to update and 1 chunk item to add or remove */
4505 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4506 btrfs_calc_trans_metadata_size(fs_info, 1);
4508 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4509 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4510 left, thresh, type);
4511 dump_space_info(fs_info, info, 0, 0);
4514 if (left < thresh) {
4515 u64 flags = btrfs_system_alloc_profile(fs_info);
4518 * Ignore failure to create system chunk. We might end up not
4519 * needing it, as we might not need to COW all nodes/leafs from
4520 * the paths we visit in the chunk tree (they were already COWed
4521 * or created in the current transaction for example).
4523 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4527 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4528 &fs_info->chunk_block_rsv,
4529 thresh, BTRFS_RESERVE_NO_FLUSH);
4531 trans->chunk_bytes_reserved += thresh;
4536 * If force is CHUNK_ALLOC_FORCE:
4537 * - return 1 if it successfully allocates a chunk,
4538 * - return errors including -ENOSPC otherwise.
4539 * If force is NOT CHUNK_ALLOC_FORCE:
4540 * - return 0 if it doesn't need to allocate a new chunk,
4541 * - return 1 if it successfully allocates a chunk,
4542 * - return errors including -ENOSPC otherwise.
4544 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4545 struct btrfs_fs_info *fs_info, u64 flags, int force)
4547 struct btrfs_space_info *space_info;
4548 int wait_for_alloc = 0;
4551 /* Don't re-enter if we're already allocating a chunk */
4552 if (trans->allocating_chunk)
4555 space_info = __find_space_info(fs_info, flags);
4557 ret = create_space_info(fs_info, flags, &space_info);
4563 spin_lock(&space_info->lock);
4564 if (force < space_info->force_alloc)
4565 force = space_info->force_alloc;
4566 if (space_info->full) {
4567 if (should_alloc_chunk(fs_info, space_info, force))
4571 spin_unlock(&space_info->lock);
4575 if (!should_alloc_chunk(fs_info, space_info, force)) {
4576 spin_unlock(&space_info->lock);
4578 } else if (space_info->chunk_alloc) {
4581 space_info->chunk_alloc = 1;
4584 spin_unlock(&space_info->lock);
4586 mutex_lock(&fs_info->chunk_mutex);
4589 * The chunk_mutex is held throughout the entirety of a chunk
4590 * allocation, so once we've acquired the chunk_mutex we know that the
4591 * other guy is done and we need to recheck and see if we should
4594 if (wait_for_alloc) {
4595 mutex_unlock(&fs_info->chunk_mutex);
4600 trans->allocating_chunk = true;
4603 * If we have mixed data/metadata chunks we want to make sure we keep
4604 * allocating mixed chunks instead of individual chunks.
4606 if (btrfs_mixed_space_info(space_info))
4607 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4610 * if we're doing a data chunk, go ahead and make sure that
4611 * we keep a reasonable number of metadata chunks allocated in the
4614 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4615 fs_info->data_chunk_allocations++;
4616 if (!(fs_info->data_chunk_allocations %
4617 fs_info->metadata_ratio))
4618 force_metadata_allocation(fs_info);
4622 * Check if we have enough space in SYSTEM chunk because we may need
4623 * to update devices.
4625 check_system_chunk(trans, fs_info, flags);
4627 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4628 trans->allocating_chunk = false;
4630 spin_lock(&space_info->lock);
4631 if (ret < 0 && ret != -ENOSPC)
4634 space_info->full = 1;
4638 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4640 space_info->chunk_alloc = 0;
4641 spin_unlock(&space_info->lock);
4642 mutex_unlock(&fs_info->chunk_mutex);
4644 * When we allocate a new chunk we reserve space in the chunk block
4645 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4646 * add new nodes/leafs to it if we end up needing to do it when
4647 * inserting the chunk item and updating device items as part of the
4648 * second phase of chunk allocation, performed by
4649 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4650 * large number of new block groups to create in our transaction
4651 * handle's new_bgs list to avoid exhausting the chunk block reserve
4652 * in extreme cases - like having a single transaction create many new
4653 * block groups when starting to write out the free space caches of all
4654 * the block groups that were made dirty during the lifetime of the
4657 if (trans->can_flush_pending_bgs &&
4658 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4659 btrfs_create_pending_block_groups(trans, fs_info);
4660 btrfs_trans_release_chunk_metadata(trans);
4665 static int can_overcommit(struct btrfs_fs_info *fs_info,
4666 struct btrfs_space_info *space_info, u64 bytes,
4667 enum btrfs_reserve_flush_enum flush,
4670 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4676 /* Don't overcommit when in mixed mode. */
4677 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4681 profile = btrfs_system_alloc_profile(fs_info);
4683 profile = btrfs_metadata_alloc_profile(fs_info);
4685 used = btrfs_space_info_used(space_info, false);
4688 * We only want to allow over committing if we have lots of actual space
4689 * free, but if we don't have enough space to handle the global reserve
4690 * space then we could end up having a real enospc problem when trying
4691 * to allocate a chunk or some other such important allocation.
4693 spin_lock(&global_rsv->lock);
4694 space_size = calc_global_rsv_need_space(global_rsv);
4695 spin_unlock(&global_rsv->lock);
4696 if (used + space_size >= space_info->total_bytes)
4699 used += space_info->bytes_may_use;
4701 avail = atomic64_read(&fs_info->free_chunk_space);
4704 * If we have dup, raid1 or raid10 then only half of the free
4705 * space is actually useable. For raid56, the space info used
4706 * doesn't include the parity drive, so we don't have to
4709 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4710 BTRFS_BLOCK_GROUP_RAID1 |
4711 BTRFS_BLOCK_GROUP_RAID10))
4715 * If we aren't flushing all things, let us overcommit up to
4716 * 1/2th of the space. If we can flush, don't let us overcommit
4717 * too much, let it overcommit up to 1/8 of the space.
4719 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4724 if (used + bytes < space_info->total_bytes + avail)
4729 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4730 unsigned long nr_pages, int nr_items)
4732 struct super_block *sb = fs_info->sb;
4734 if (down_read_trylock(&sb->s_umount)) {
4735 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4736 up_read(&sb->s_umount);
4739 * We needn't worry the filesystem going from r/w to r/o though
4740 * we don't acquire ->s_umount mutex, because the filesystem
4741 * should guarantee the delalloc inodes list be empty after
4742 * the filesystem is readonly(all dirty pages are written to
4745 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4746 if (!current->journal_info)
4747 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4751 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4757 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4758 nr = div64_u64(to_reclaim, bytes);
4764 #define EXTENT_SIZE_PER_ITEM SZ_256K
4767 * shrink metadata reservation for delalloc
4769 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4770 u64 orig, bool wait_ordered)
4772 struct btrfs_block_rsv *block_rsv;
4773 struct btrfs_space_info *space_info;
4774 struct btrfs_trans_handle *trans;
4779 unsigned long nr_pages;
4781 enum btrfs_reserve_flush_enum flush;
4783 /* Calc the number of the pages we need flush for space reservation */
4784 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4785 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4787 trans = (struct btrfs_trans_handle *)current->journal_info;
4788 block_rsv = &fs_info->delalloc_block_rsv;
4789 space_info = block_rsv->space_info;
4791 delalloc_bytes = percpu_counter_sum_positive(
4792 &fs_info->delalloc_bytes);
4793 if (delalloc_bytes == 0) {
4797 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4802 while (delalloc_bytes && loops < 3) {
4803 max_reclaim = min(delalloc_bytes, to_reclaim);
4804 nr_pages = max_reclaim >> PAGE_SHIFT;
4805 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4807 * We need to wait for the async pages to actually start before
4810 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4814 if (max_reclaim <= nr_pages)
4817 max_reclaim -= nr_pages;
4819 wait_event(fs_info->async_submit_wait,
4820 atomic_read(&fs_info->async_delalloc_pages) <=
4824 flush = BTRFS_RESERVE_FLUSH_ALL;
4826 flush = BTRFS_RESERVE_NO_FLUSH;
4827 spin_lock(&space_info->lock);
4828 if (list_empty(&space_info->tickets) &&
4829 list_empty(&space_info->priority_tickets)) {
4830 spin_unlock(&space_info->lock);
4833 spin_unlock(&space_info->lock);
4836 if (wait_ordered && !trans) {
4837 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4839 time_left = schedule_timeout_killable(1);
4843 delalloc_bytes = percpu_counter_sum_positive(
4844 &fs_info->delalloc_bytes);
4849 * maybe_commit_transaction - possibly commit the transaction if its ok to
4850 * @root - the root we're allocating for
4851 * @bytes - the number of bytes we want to reserve
4852 * @force - force the commit
4854 * This will check to make sure that committing the transaction will actually
4855 * get us somewhere and then commit the transaction if it does. Otherwise it
4856 * will return -ENOSPC.
4858 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4859 struct btrfs_space_info *space_info,
4860 u64 bytes, int force)
4862 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4863 struct btrfs_trans_handle *trans;
4865 trans = (struct btrfs_trans_handle *)current->journal_info;
4872 /* See if there is enough pinned space to make this reservation */
4873 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4878 * See if there is some space in the delayed insertion reservation for
4881 if (space_info != delayed_rsv->space_info)
4884 spin_lock(&delayed_rsv->lock);
4885 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4886 bytes - delayed_rsv->size) < 0) {
4887 spin_unlock(&delayed_rsv->lock);
4890 spin_unlock(&delayed_rsv->lock);
4893 trans = btrfs_join_transaction(fs_info->extent_root);
4897 return btrfs_commit_transaction(trans);
4900 struct reserve_ticket {
4903 struct list_head list;
4904 wait_queue_head_t wait;
4907 static int flush_space(struct btrfs_fs_info *fs_info,
4908 struct btrfs_space_info *space_info, u64 num_bytes,
4909 u64 orig_bytes, int state)
4911 struct btrfs_root *root = fs_info->extent_root;
4912 struct btrfs_trans_handle *trans;
4917 case FLUSH_DELAYED_ITEMS_NR:
4918 case FLUSH_DELAYED_ITEMS:
4919 if (state == FLUSH_DELAYED_ITEMS_NR)
4920 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4924 trans = btrfs_join_transaction(root);
4925 if (IS_ERR(trans)) {
4926 ret = PTR_ERR(trans);
4929 ret = btrfs_run_delayed_items_nr(trans, fs_info, nr);
4930 btrfs_end_transaction(trans);
4932 case FLUSH_DELALLOC:
4933 case FLUSH_DELALLOC_WAIT:
4934 shrink_delalloc(fs_info, num_bytes * 2, orig_bytes,
4935 state == FLUSH_DELALLOC_WAIT);
4938 trans = btrfs_join_transaction(root);
4939 if (IS_ERR(trans)) {
4940 ret = PTR_ERR(trans);
4943 ret = do_chunk_alloc(trans, fs_info,
4944 btrfs_metadata_alloc_profile(fs_info),
4945 CHUNK_ALLOC_NO_FORCE);
4946 btrfs_end_transaction(trans);
4947 if (ret > 0 || ret == -ENOSPC)
4951 ret = may_commit_transaction(fs_info, space_info,
4959 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes,
4960 orig_bytes, state, ret);
4965 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4966 struct btrfs_space_info *space_info,
4969 struct reserve_ticket *ticket;
4974 list_for_each_entry(ticket, &space_info->tickets, list)
4975 to_reclaim += ticket->bytes;
4976 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4977 to_reclaim += ticket->bytes;
4981 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4982 if (can_overcommit(fs_info, space_info, to_reclaim,
4983 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4986 used = btrfs_space_info_used(space_info, true);
4988 if (can_overcommit(fs_info, space_info, SZ_1M,
4989 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4990 expected = div_factor_fine(space_info->total_bytes, 95);
4992 expected = div_factor_fine(space_info->total_bytes, 90);
4994 if (used > expected)
4995 to_reclaim = used - expected;
4998 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4999 space_info->bytes_reserved);
5003 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5004 struct btrfs_space_info *space_info,
5005 u64 used, bool system_chunk)
5007 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5009 /* If we're just plain full then async reclaim just slows us down. */
5010 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5013 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5017 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5018 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5021 static void wake_all_tickets(struct list_head *head)
5023 struct reserve_ticket *ticket;
5025 while (!list_empty(head)) {
5026 ticket = list_first_entry(head, struct reserve_ticket, list);
5027 list_del_init(&ticket->list);
5028 ticket->error = -ENOSPC;
5029 wake_up(&ticket->wait);
5034 * This is for normal flushers, we can wait all goddamned day if we want to. We
5035 * will loop and continuously try to flush as long as we are making progress.
5036 * We count progress as clearing off tickets each time we have to loop.
5038 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5040 struct btrfs_fs_info *fs_info;
5041 struct btrfs_space_info *space_info;
5044 int commit_cycles = 0;
5045 u64 last_tickets_id;
5047 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5048 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5050 spin_lock(&space_info->lock);
5051 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5054 space_info->flush = 0;
5055 spin_unlock(&space_info->lock);
5058 last_tickets_id = space_info->tickets_id;
5059 spin_unlock(&space_info->lock);
5061 flush_state = FLUSH_DELAYED_ITEMS_NR;
5063 struct reserve_ticket *ticket;
5066 ret = flush_space(fs_info, space_info, to_reclaim, to_reclaim,
5068 spin_lock(&space_info->lock);
5069 if (list_empty(&space_info->tickets)) {
5070 space_info->flush = 0;
5071 spin_unlock(&space_info->lock);
5074 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5077 ticket = list_first_entry(&space_info->tickets,
5078 struct reserve_ticket, list);
5079 if (last_tickets_id == space_info->tickets_id) {
5082 last_tickets_id = space_info->tickets_id;
5083 flush_state = FLUSH_DELAYED_ITEMS_NR;
5088 if (flush_state > COMMIT_TRANS) {
5090 if (commit_cycles > 2) {
5091 wake_all_tickets(&space_info->tickets);
5092 space_info->flush = 0;
5094 flush_state = FLUSH_DELAYED_ITEMS_NR;
5097 spin_unlock(&space_info->lock);
5098 } while (flush_state <= COMMIT_TRANS);
5101 void btrfs_init_async_reclaim_work(struct work_struct *work)
5103 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5106 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5107 struct btrfs_space_info *space_info,
5108 struct reserve_ticket *ticket)
5111 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5113 spin_lock(&space_info->lock);
5114 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5117 spin_unlock(&space_info->lock);
5120 spin_unlock(&space_info->lock);
5123 flush_space(fs_info, space_info, to_reclaim, to_reclaim,
5126 spin_lock(&space_info->lock);
5127 if (ticket->bytes == 0) {
5128 spin_unlock(&space_info->lock);
5131 spin_unlock(&space_info->lock);
5134 * Priority flushers can't wait on delalloc without
5137 if (flush_state == FLUSH_DELALLOC ||
5138 flush_state == FLUSH_DELALLOC_WAIT)
5139 flush_state = ALLOC_CHUNK;
5140 } while (flush_state < COMMIT_TRANS);
5143 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5144 struct btrfs_space_info *space_info,
5145 struct reserve_ticket *ticket, u64 orig_bytes)
5151 spin_lock(&space_info->lock);
5152 while (ticket->bytes > 0 && ticket->error == 0) {
5153 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5158 spin_unlock(&space_info->lock);
5162 finish_wait(&ticket->wait, &wait);
5163 spin_lock(&space_info->lock);
5166 ret = ticket->error;
5167 if (!list_empty(&ticket->list))
5168 list_del_init(&ticket->list);
5169 if (ticket->bytes && ticket->bytes < orig_bytes) {
5170 u64 num_bytes = orig_bytes - ticket->bytes;
5171 space_info->bytes_may_use -= num_bytes;
5172 trace_btrfs_space_reservation(fs_info, "space_info",
5173 space_info->flags, num_bytes, 0);
5175 spin_unlock(&space_info->lock);
5181 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5182 * @root - the root we're allocating for
5183 * @space_info - the space info we want to allocate from
5184 * @orig_bytes - the number of bytes we want
5185 * @flush - whether or not we can flush to make our reservation
5187 * This will reserve orig_bytes number of bytes from the space info associated
5188 * with the block_rsv. If there is not enough space it will make an attempt to
5189 * flush out space to make room. It will do this by flushing delalloc if
5190 * possible or committing the transaction. If flush is 0 then no attempts to
5191 * regain reservations will be made and this will fail if there is not enough
5194 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5195 struct btrfs_space_info *space_info,
5197 enum btrfs_reserve_flush_enum flush,
5200 struct reserve_ticket ticket;
5205 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5207 spin_lock(&space_info->lock);
5209 used = btrfs_space_info_used(space_info, true);
5212 * If we have enough space then hooray, make our reservation and carry
5213 * on. If not see if we can overcommit, and if we can, hooray carry on.
5214 * If not things get more complicated.
5216 if (used + orig_bytes <= space_info->total_bytes) {
5217 space_info->bytes_may_use += orig_bytes;
5218 trace_btrfs_space_reservation(fs_info, "space_info",
5219 space_info->flags, orig_bytes, 1);
5221 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5223 space_info->bytes_may_use += orig_bytes;
5224 trace_btrfs_space_reservation(fs_info, "space_info",
5225 space_info->flags, orig_bytes, 1);
5230 * If we couldn't make a reservation then setup our reservation ticket
5231 * and kick the async worker if it's not already running.
5233 * If we are a priority flusher then we just need to add our ticket to
5234 * the list and we will do our own flushing further down.
5236 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5237 ticket.bytes = orig_bytes;
5239 init_waitqueue_head(&ticket.wait);
5240 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5241 list_add_tail(&ticket.list, &space_info->tickets);
5242 if (!space_info->flush) {
5243 space_info->flush = 1;
5244 trace_btrfs_trigger_flush(fs_info,
5248 queue_work(system_unbound_wq,
5249 &fs_info->async_reclaim_work);
5252 list_add_tail(&ticket.list,
5253 &space_info->priority_tickets);
5255 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5258 * We will do the space reservation dance during log replay,
5259 * which means we won't have fs_info->fs_root set, so don't do
5260 * the async reclaim as we will panic.
5262 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5263 need_do_async_reclaim(fs_info, space_info,
5264 used, system_chunk) &&
5265 !work_busy(&fs_info->async_reclaim_work)) {
5266 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5267 orig_bytes, flush, "preempt");
5268 queue_work(system_unbound_wq,
5269 &fs_info->async_reclaim_work);
5272 spin_unlock(&space_info->lock);
5273 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5276 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5277 return wait_reserve_ticket(fs_info, space_info, &ticket,
5281 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5282 spin_lock(&space_info->lock);
5284 if (ticket.bytes < orig_bytes) {
5285 u64 num_bytes = orig_bytes - ticket.bytes;
5286 space_info->bytes_may_use -= num_bytes;
5287 trace_btrfs_space_reservation(fs_info, "space_info",
5292 list_del_init(&ticket.list);
5295 spin_unlock(&space_info->lock);
5296 ASSERT(list_empty(&ticket.list));
5301 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5302 * @root - the root we're allocating for
5303 * @block_rsv - the block_rsv we're allocating for
5304 * @orig_bytes - the number of bytes we want
5305 * @flush - whether or not we can flush to make our reservation
5307 * This will reserve orgi_bytes number of bytes from the space info associated
5308 * with the block_rsv. If there is not enough space it will make an attempt to
5309 * flush out space to make room. It will do this by flushing delalloc if
5310 * possible or committing the transaction. If flush is 0 then no attempts to
5311 * regain reservations will be made and this will fail if there is not enough
5314 static int reserve_metadata_bytes(struct btrfs_root *root,
5315 struct btrfs_block_rsv *block_rsv,
5317 enum btrfs_reserve_flush_enum flush)
5319 struct btrfs_fs_info *fs_info = root->fs_info;
5320 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5322 bool system_chunk = (root == fs_info->chunk_root);
5324 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5325 orig_bytes, flush, system_chunk);
5326 if (ret == -ENOSPC &&
5327 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5328 if (block_rsv != global_rsv &&
5329 !block_rsv_use_bytes(global_rsv, orig_bytes))
5333 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5334 block_rsv->space_info->flags,
5339 static struct btrfs_block_rsv *get_block_rsv(
5340 const struct btrfs_trans_handle *trans,
5341 const struct btrfs_root *root)
5343 struct btrfs_fs_info *fs_info = root->fs_info;
5344 struct btrfs_block_rsv *block_rsv = NULL;
5346 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5347 (root == fs_info->csum_root && trans->adding_csums) ||
5348 (root == fs_info->uuid_root))
5349 block_rsv = trans->block_rsv;
5352 block_rsv = root->block_rsv;
5355 block_rsv = &fs_info->empty_block_rsv;
5360 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5364 spin_lock(&block_rsv->lock);
5365 if (block_rsv->reserved >= num_bytes) {
5366 block_rsv->reserved -= num_bytes;
5367 if (block_rsv->reserved < block_rsv->size)
5368 block_rsv->full = 0;
5371 spin_unlock(&block_rsv->lock);
5375 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5376 u64 num_bytes, int update_size)
5378 spin_lock(&block_rsv->lock);
5379 block_rsv->reserved += num_bytes;
5381 block_rsv->size += num_bytes;
5382 else if (block_rsv->reserved >= block_rsv->size)
5383 block_rsv->full = 1;
5384 spin_unlock(&block_rsv->lock);
5387 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5388 struct btrfs_block_rsv *dest, u64 num_bytes,
5391 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5394 if (global_rsv->space_info != dest->space_info)
5397 spin_lock(&global_rsv->lock);
5398 min_bytes = div_factor(global_rsv->size, min_factor);
5399 if (global_rsv->reserved < min_bytes + num_bytes) {
5400 spin_unlock(&global_rsv->lock);
5403 global_rsv->reserved -= num_bytes;
5404 if (global_rsv->reserved < global_rsv->size)
5405 global_rsv->full = 0;
5406 spin_unlock(&global_rsv->lock);
5408 block_rsv_add_bytes(dest, num_bytes, 1);
5413 * This is for space we already have accounted in space_info->bytes_may_use, so
5414 * basically when we're returning space from block_rsv's.
5416 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5417 struct btrfs_space_info *space_info,
5420 struct reserve_ticket *ticket;
5421 struct list_head *head;
5423 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5424 bool check_overcommit = false;
5426 spin_lock(&space_info->lock);
5427 head = &space_info->priority_tickets;
5430 * If we are over our limit then we need to check and see if we can
5431 * overcommit, and if we can't then we just need to free up our space
5432 * and not satisfy any requests.
5434 used = btrfs_space_info_used(space_info, true);
5435 if (used - num_bytes >= space_info->total_bytes)
5436 check_overcommit = true;
5438 while (!list_empty(head) && num_bytes) {
5439 ticket = list_first_entry(head, struct reserve_ticket,
5442 * We use 0 bytes because this space is already reserved, so
5443 * adding the ticket space would be a double count.
5445 if (check_overcommit &&
5446 !can_overcommit(fs_info, space_info, 0, flush, false))
5448 if (num_bytes >= ticket->bytes) {
5449 list_del_init(&ticket->list);
5450 num_bytes -= ticket->bytes;
5452 space_info->tickets_id++;
5453 wake_up(&ticket->wait);
5455 ticket->bytes -= num_bytes;
5460 if (num_bytes && head == &space_info->priority_tickets) {
5461 head = &space_info->tickets;
5462 flush = BTRFS_RESERVE_FLUSH_ALL;
5465 space_info->bytes_may_use -= num_bytes;
5466 trace_btrfs_space_reservation(fs_info, "space_info",
5467 space_info->flags, num_bytes, 0);
5468 spin_unlock(&space_info->lock);
5472 * This is for newly allocated space that isn't accounted in
5473 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5474 * we use this helper.
5476 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5477 struct btrfs_space_info *space_info,
5480 struct reserve_ticket *ticket;
5481 struct list_head *head = &space_info->priority_tickets;
5484 while (!list_empty(head) && num_bytes) {
5485 ticket = list_first_entry(head, struct reserve_ticket,
5487 if (num_bytes >= ticket->bytes) {
5488 trace_btrfs_space_reservation(fs_info, "space_info",
5491 list_del_init(&ticket->list);
5492 num_bytes -= ticket->bytes;
5493 space_info->bytes_may_use += ticket->bytes;
5495 space_info->tickets_id++;
5496 wake_up(&ticket->wait);
5498 trace_btrfs_space_reservation(fs_info, "space_info",
5501 space_info->bytes_may_use += num_bytes;
5502 ticket->bytes -= num_bytes;
5507 if (num_bytes && head == &space_info->priority_tickets) {
5508 head = &space_info->tickets;
5513 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5514 struct btrfs_block_rsv *block_rsv,
5515 struct btrfs_block_rsv *dest, u64 num_bytes)
5517 struct btrfs_space_info *space_info = block_rsv->space_info;
5519 spin_lock(&block_rsv->lock);
5520 if (num_bytes == (u64)-1)
5521 num_bytes = block_rsv->size;
5522 block_rsv->size -= num_bytes;
5523 if (block_rsv->reserved >= block_rsv->size) {
5524 num_bytes = block_rsv->reserved - block_rsv->size;
5525 block_rsv->reserved = block_rsv->size;
5526 block_rsv->full = 1;
5530 spin_unlock(&block_rsv->lock);
5532 if (num_bytes > 0) {
5534 spin_lock(&dest->lock);
5538 bytes_to_add = dest->size - dest->reserved;
5539 bytes_to_add = min(num_bytes, bytes_to_add);
5540 dest->reserved += bytes_to_add;
5541 if (dest->reserved >= dest->size)
5543 num_bytes -= bytes_to_add;
5545 spin_unlock(&dest->lock);
5548 space_info_add_old_bytes(fs_info, space_info,
5553 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5554 struct btrfs_block_rsv *dst, u64 num_bytes,
5559 ret = block_rsv_use_bytes(src, num_bytes);
5563 block_rsv_add_bytes(dst, num_bytes, update_size);
5567 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5569 memset(rsv, 0, sizeof(*rsv));
5570 spin_lock_init(&rsv->lock);
5574 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5575 unsigned short type)
5577 struct btrfs_block_rsv *block_rsv;
5579 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5583 btrfs_init_block_rsv(block_rsv, type);
5584 block_rsv->space_info = __find_space_info(fs_info,
5585 BTRFS_BLOCK_GROUP_METADATA);
5589 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5590 struct btrfs_block_rsv *rsv)
5594 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5598 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5603 int btrfs_block_rsv_add(struct btrfs_root *root,
5604 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5605 enum btrfs_reserve_flush_enum flush)
5612 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5614 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5621 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5629 spin_lock(&block_rsv->lock);
5630 num_bytes = div_factor(block_rsv->size, min_factor);
5631 if (block_rsv->reserved >= num_bytes)
5633 spin_unlock(&block_rsv->lock);
5638 int btrfs_block_rsv_refill(struct btrfs_root *root,
5639 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5640 enum btrfs_reserve_flush_enum flush)
5648 spin_lock(&block_rsv->lock);
5649 num_bytes = min_reserved;
5650 if (block_rsv->reserved >= num_bytes)
5653 num_bytes -= block_rsv->reserved;
5654 spin_unlock(&block_rsv->lock);
5659 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5661 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5668 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5669 struct btrfs_block_rsv *block_rsv,
5672 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5674 if (global_rsv == block_rsv ||
5675 block_rsv->space_info != global_rsv->space_info)
5677 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5680 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5682 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5683 struct btrfs_space_info *sinfo = block_rsv->space_info;
5687 * The global block rsv is based on the size of the extent tree, the
5688 * checksum tree and the root tree. If the fs is empty we want to set
5689 * it to a minimal amount for safety.
5691 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5692 btrfs_root_used(&fs_info->csum_root->root_item) +
5693 btrfs_root_used(&fs_info->tree_root->root_item);
5694 num_bytes = max_t(u64, num_bytes, SZ_16M);
5696 spin_lock(&sinfo->lock);
5697 spin_lock(&block_rsv->lock);
5699 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5701 if (block_rsv->reserved < block_rsv->size) {
5702 num_bytes = btrfs_space_info_used(sinfo, true);
5703 if (sinfo->total_bytes > num_bytes) {
5704 num_bytes = sinfo->total_bytes - num_bytes;
5705 num_bytes = min(num_bytes,
5706 block_rsv->size - block_rsv->reserved);
5707 block_rsv->reserved += num_bytes;
5708 sinfo->bytes_may_use += num_bytes;
5709 trace_btrfs_space_reservation(fs_info, "space_info",
5710 sinfo->flags, num_bytes,
5713 } else if (block_rsv->reserved > block_rsv->size) {
5714 num_bytes = block_rsv->reserved - block_rsv->size;
5715 sinfo->bytes_may_use -= num_bytes;
5716 trace_btrfs_space_reservation(fs_info, "space_info",
5717 sinfo->flags, num_bytes, 0);
5718 block_rsv->reserved = block_rsv->size;
5721 if (block_rsv->reserved == block_rsv->size)
5722 block_rsv->full = 1;
5724 block_rsv->full = 0;
5726 spin_unlock(&block_rsv->lock);
5727 spin_unlock(&sinfo->lock);
5730 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5732 struct btrfs_space_info *space_info;
5734 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5735 fs_info->chunk_block_rsv.space_info = space_info;
5737 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5738 fs_info->global_block_rsv.space_info = space_info;
5739 fs_info->delalloc_block_rsv.space_info = space_info;
5740 fs_info->trans_block_rsv.space_info = space_info;
5741 fs_info->empty_block_rsv.space_info = space_info;
5742 fs_info->delayed_block_rsv.space_info = space_info;
5744 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5745 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5746 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5747 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5748 if (fs_info->quota_root)
5749 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5750 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5752 update_global_block_rsv(fs_info);
5755 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5757 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5759 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5760 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5761 WARN_ON(fs_info->trans_block_rsv.size > 0);
5762 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5763 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5764 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5765 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5766 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5769 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5770 struct btrfs_fs_info *fs_info)
5772 if (!trans->block_rsv)
5775 if (!trans->bytes_reserved)
5778 trace_btrfs_space_reservation(fs_info, "transaction",
5779 trans->transid, trans->bytes_reserved, 0);
5780 btrfs_block_rsv_release(fs_info, trans->block_rsv,
5781 trans->bytes_reserved);
5782 trans->bytes_reserved = 0;
5786 * To be called after all the new block groups attached to the transaction
5787 * handle have been created (btrfs_create_pending_block_groups()).
5789 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5791 struct btrfs_fs_info *fs_info = trans->fs_info;
5793 if (!trans->chunk_bytes_reserved)
5796 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5798 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5799 trans->chunk_bytes_reserved);
5800 trans->chunk_bytes_reserved = 0;
5803 /* Can only return 0 or -ENOSPC */
5804 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5805 struct btrfs_inode *inode)
5807 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5808 struct btrfs_root *root = inode->root;
5810 * We always use trans->block_rsv here as we will have reserved space
5811 * for our orphan when starting the transaction, using get_block_rsv()
5812 * here will sometimes make us choose the wrong block rsv as we could be
5813 * doing a reloc inode for a non refcounted root.
5815 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5816 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5819 * We need to hold space in order to delete our orphan item once we've
5820 * added it, so this takes the reservation so we can release it later
5821 * when we are truly done with the orphan item.
5823 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5825 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5827 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5830 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5832 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5833 struct btrfs_root *root = inode->root;
5834 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5836 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5838 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5842 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5843 * root: the root of the parent directory
5844 * rsv: block reservation
5845 * items: the number of items that we need do reservation
5846 * qgroup_reserved: used to return the reserved size in qgroup
5848 * This function is used to reserve the space for snapshot/subvolume
5849 * creation and deletion. Those operations are different with the
5850 * common file/directory operations, they change two fs/file trees
5851 * and root tree, the number of items that the qgroup reserves is
5852 * different with the free space reservation. So we can not use
5853 * the space reservation mechanism in start_transaction().
5855 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5856 struct btrfs_block_rsv *rsv,
5858 u64 *qgroup_reserved,
5859 bool use_global_rsv)
5863 struct btrfs_fs_info *fs_info = root->fs_info;
5864 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5866 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5867 /* One for parent inode, two for dir entries */
5868 num_bytes = 3 * fs_info->nodesize;
5869 ret = btrfs_qgroup_reserve_meta(root, num_bytes, true);
5876 *qgroup_reserved = num_bytes;
5878 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5879 rsv->space_info = __find_space_info(fs_info,
5880 BTRFS_BLOCK_GROUP_METADATA);
5881 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5882 BTRFS_RESERVE_FLUSH_ALL);
5884 if (ret == -ENOSPC && use_global_rsv)
5885 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5887 if (ret && *qgroup_reserved)
5888 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5893 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5894 struct btrfs_block_rsv *rsv)
5896 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5900 * drop_outstanding_extent - drop an outstanding extent
5901 * @inode: the inode we're dropping the extent for
5902 * @num_bytes: the number of bytes we're releasing.
5904 * This is called when we are freeing up an outstanding extent, either called
5905 * after an error or after an extent is written. This will return the number of
5906 * reserved extents that need to be freed. This must be called with
5907 * BTRFS_I(inode)->lock held.
5909 static unsigned drop_outstanding_extent(struct btrfs_inode *inode,
5912 unsigned drop_inode_space = 0;
5913 unsigned dropped_extents = 0;
5914 unsigned num_extents;
5916 num_extents = count_max_extents(num_bytes);
5917 ASSERT(num_extents);
5918 ASSERT(inode->outstanding_extents >= num_extents);
5919 inode->outstanding_extents -= num_extents;
5921 if (inode->outstanding_extents == 0 &&
5922 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5923 &inode->runtime_flags))
5924 drop_inode_space = 1;
5927 * If we have more or the same amount of outstanding extents than we have
5928 * reserved then we need to leave the reserved extents count alone.
5930 if (inode->outstanding_extents >= inode->reserved_extents)
5931 return drop_inode_space;
5933 dropped_extents = inode->reserved_extents - inode->outstanding_extents;
5934 inode->reserved_extents -= dropped_extents;
5935 return dropped_extents + drop_inode_space;
5939 * calc_csum_metadata_size - return the amount of metadata space that must be
5940 * reserved/freed for the given bytes.
5941 * @inode: the inode we're manipulating
5942 * @num_bytes: the number of bytes in question
5943 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5945 * This adjusts the number of csum_bytes in the inode and then returns the
5946 * correct amount of metadata that must either be reserved or freed. We
5947 * calculate how many checksums we can fit into one leaf and then divide the
5948 * number of bytes that will need to be checksumed by this value to figure out
5949 * how many checksums will be required. If we are adding bytes then the number
5950 * may go up and we will return the number of additional bytes that must be
5951 * reserved. If it is going down we will return the number of bytes that must
5954 * This must be called with BTRFS_I(inode)->lock held.
5956 static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes,
5959 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5960 u64 old_csums, num_csums;
5962 if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0)
5965 old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
5967 inode->csum_bytes += num_bytes;
5969 inode->csum_bytes -= num_bytes;
5970 num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
5972 /* No change, no need to reserve more */
5973 if (old_csums == num_csums)
5977 return btrfs_calc_trans_metadata_size(fs_info,
5978 num_csums - old_csums);
5980 return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums);
5983 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
5985 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5986 struct btrfs_root *root = inode->root;
5987 struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv;
5990 unsigned nr_extents;
5991 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5993 bool delalloc_lock = true;
5996 bool release_extra = false;
5998 /* If we are a free space inode we need to not flush since we will be in
5999 * the middle of a transaction commit. We also don't need the delalloc
6000 * mutex since we won't race with anybody. We need this mostly to make
6001 * lockdep shut its filthy mouth.
6003 * If we have a transaction open (can happen if we call truncate_block
6004 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6006 if (btrfs_is_free_space_inode(inode)) {
6007 flush = BTRFS_RESERVE_NO_FLUSH;
6008 delalloc_lock = false;
6009 } else if (current->journal_info) {
6010 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6013 if (flush != BTRFS_RESERVE_NO_FLUSH &&
6014 btrfs_transaction_in_commit(fs_info))
6015 schedule_timeout(1);
6018 mutex_lock(&inode->delalloc_mutex);
6020 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6022 spin_lock(&inode->lock);
6023 nr_extents = count_max_extents(num_bytes);
6024 inode->outstanding_extents += nr_extents;
6027 if (inode->outstanding_extents > inode->reserved_extents)
6028 nr_extents += inode->outstanding_extents -
6029 inode->reserved_extents;
6031 /* We always want to reserve a slot for updating the inode. */
6032 to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1);
6033 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
6034 csum_bytes = inode->csum_bytes;
6035 spin_unlock(&inode->lock);
6037 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6038 ret = btrfs_qgroup_reserve_meta(root,
6039 nr_extents * fs_info->nodesize, true);
6044 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
6045 if (unlikely(ret)) {
6046 btrfs_qgroup_free_meta(root,
6047 nr_extents * fs_info->nodesize);
6051 spin_lock(&inode->lock);
6052 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6053 &inode->runtime_flags)) {
6054 to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1);
6055 release_extra = true;
6057 inode->reserved_extents += nr_extents;
6058 spin_unlock(&inode->lock);
6061 mutex_unlock(&inode->delalloc_mutex);
6064 trace_btrfs_space_reservation(fs_info, "delalloc",
6065 btrfs_ino(inode), to_reserve, 1);
6067 btrfs_block_rsv_release(fs_info, block_rsv,
6068 btrfs_calc_trans_metadata_size(fs_info, 1));
6072 spin_lock(&inode->lock);
6073 dropped = drop_outstanding_extent(inode, num_bytes);
6075 * If the inodes csum_bytes is the same as the original
6076 * csum_bytes then we know we haven't raced with any free()ers
6077 * so we can just reduce our inodes csum bytes and carry on.
6079 if (inode->csum_bytes == csum_bytes) {
6080 calc_csum_metadata_size(inode, num_bytes, 0);
6082 u64 orig_csum_bytes = inode->csum_bytes;
6086 * This is tricky, but first we need to figure out how much we
6087 * freed from any free-ers that occurred during this
6088 * reservation, so we reset ->csum_bytes to the csum_bytes
6089 * before we dropped our lock, and then call the free for the
6090 * number of bytes that were freed while we were trying our
6093 bytes = csum_bytes - inode->csum_bytes;
6094 inode->csum_bytes = csum_bytes;
6095 to_free = calc_csum_metadata_size(inode, bytes, 0);
6099 * Now we need to see how much we would have freed had we not
6100 * been making this reservation and our ->csum_bytes were not
6101 * artificially inflated.
6103 inode->csum_bytes = csum_bytes - num_bytes;
6104 bytes = csum_bytes - orig_csum_bytes;
6105 bytes = calc_csum_metadata_size(inode, bytes, 0);
6108 * Now reset ->csum_bytes to what it should be. If bytes is
6109 * more than to_free then we would have freed more space had we
6110 * not had an artificially high ->csum_bytes, so we need to free
6111 * the remainder. If bytes is the same or less then we don't
6112 * need to do anything, the other free-ers did the correct
6115 inode->csum_bytes = orig_csum_bytes - num_bytes;
6116 if (bytes > to_free)
6117 to_free = bytes - to_free;
6121 spin_unlock(&inode->lock);
6123 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6126 btrfs_block_rsv_release(fs_info, block_rsv, to_free);
6127 trace_btrfs_space_reservation(fs_info, "delalloc",
6128 btrfs_ino(inode), to_free, 0);
6131 mutex_unlock(&inode->delalloc_mutex);
6136 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6137 * @inode: the inode to release the reservation for
6138 * @num_bytes: the number of bytes we're releasing
6140 * This will release the metadata reservation for an inode. This can be called
6141 * once we complete IO for a given set of bytes to release their metadata
6144 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes)
6146 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6150 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6151 spin_lock(&inode->lock);
6152 dropped = drop_outstanding_extent(inode, num_bytes);
6155 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6156 spin_unlock(&inode->lock);
6158 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6160 if (btrfs_is_testing(fs_info))
6163 trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode),
6166 btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free);
6170 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6172 * @inode: inode we're writing to
6173 * @start: start range we are writing to
6174 * @len: how long the range we are writing to
6175 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6176 * current reservation.
6178 * This will do the following things
6180 * o reserve space in data space info for num bytes
6181 * and reserve precious corresponding qgroup space
6182 * (Done in check_data_free_space)
6184 * o reserve space for metadata space, based on the number of outstanding
6185 * extents and how much csums will be needed
6186 * also reserve metadata space in a per root over-reserve method.
6187 * o add to the inodes->delalloc_bytes
6188 * o add it to the fs_info's delalloc inodes list.
6189 * (Above 3 all done in delalloc_reserve_metadata)
6191 * Return 0 for success
6192 * Return <0 for error(-ENOSPC or -EQUOT)
6194 int btrfs_delalloc_reserve_space(struct inode *inode,
6195 struct extent_changeset **reserved, u64 start, u64 len)
6199 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6202 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6204 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6209 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6210 * @inode: inode we're releasing space for
6211 * @start: start position of the space already reserved
6212 * @len: the len of the space already reserved
6214 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6215 * called in the case that we don't need the metadata AND data reservations
6216 * anymore. So if there is an error or we insert an inline extent.
6218 * This function will release the metadata space that was not used and will
6219 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6220 * list if there are no delalloc bytes left.
6221 * Also it will handle the qgroup reserved space.
6223 void btrfs_delalloc_release_space(struct inode *inode,
6224 struct extent_changeset *reserved, u64 start, u64 len)
6226 btrfs_delalloc_release_metadata(BTRFS_I(inode), len);
6227 btrfs_free_reserved_data_space(inode, reserved, start, len);
6230 static int update_block_group(struct btrfs_trans_handle *trans,
6231 struct btrfs_fs_info *info, u64 bytenr,
6232 u64 num_bytes, int alloc)
6234 struct btrfs_block_group_cache *cache = NULL;
6235 u64 total = num_bytes;
6240 /* block accounting for super block */
6241 spin_lock(&info->delalloc_root_lock);
6242 old_val = btrfs_super_bytes_used(info->super_copy);
6244 old_val += num_bytes;
6246 old_val -= num_bytes;
6247 btrfs_set_super_bytes_used(info->super_copy, old_val);
6248 spin_unlock(&info->delalloc_root_lock);
6251 cache = btrfs_lookup_block_group(info, bytenr);
6254 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6255 BTRFS_BLOCK_GROUP_RAID1 |
6256 BTRFS_BLOCK_GROUP_RAID10))
6261 * If this block group has free space cache written out, we
6262 * need to make sure to load it if we are removing space. This
6263 * is because we need the unpinning stage to actually add the
6264 * space back to the block group, otherwise we will leak space.
6266 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6267 cache_block_group(cache, 1);
6269 byte_in_group = bytenr - cache->key.objectid;
6270 WARN_ON(byte_in_group > cache->key.offset);
6272 spin_lock(&cache->space_info->lock);
6273 spin_lock(&cache->lock);
6275 if (btrfs_test_opt(info, SPACE_CACHE) &&
6276 cache->disk_cache_state < BTRFS_DC_CLEAR)
6277 cache->disk_cache_state = BTRFS_DC_CLEAR;
6279 old_val = btrfs_block_group_used(&cache->item);
6280 num_bytes = min(total, cache->key.offset - byte_in_group);
6282 old_val += num_bytes;
6283 btrfs_set_block_group_used(&cache->item, old_val);
6284 cache->reserved -= num_bytes;
6285 cache->space_info->bytes_reserved -= num_bytes;
6286 cache->space_info->bytes_used += num_bytes;
6287 cache->space_info->disk_used += num_bytes * factor;
6288 spin_unlock(&cache->lock);
6289 spin_unlock(&cache->space_info->lock);
6291 old_val -= num_bytes;
6292 btrfs_set_block_group_used(&cache->item, old_val);
6293 cache->pinned += num_bytes;
6294 cache->space_info->bytes_pinned += num_bytes;
6295 cache->space_info->bytes_used -= num_bytes;
6296 cache->space_info->disk_used -= num_bytes * factor;
6297 spin_unlock(&cache->lock);
6298 spin_unlock(&cache->space_info->lock);
6300 trace_btrfs_space_reservation(info, "pinned",
6301 cache->space_info->flags,
6303 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6305 set_extent_dirty(info->pinned_extents,
6306 bytenr, bytenr + num_bytes - 1,
6307 GFP_NOFS | __GFP_NOFAIL);
6310 spin_lock(&trans->transaction->dirty_bgs_lock);
6311 if (list_empty(&cache->dirty_list)) {
6312 list_add_tail(&cache->dirty_list,
6313 &trans->transaction->dirty_bgs);
6314 trans->transaction->num_dirty_bgs++;
6315 btrfs_get_block_group(cache);
6317 spin_unlock(&trans->transaction->dirty_bgs_lock);
6320 * No longer have used bytes in this block group, queue it for
6321 * deletion. We do this after adding the block group to the
6322 * dirty list to avoid races between cleaner kthread and space
6325 if (!alloc && old_val == 0) {
6326 spin_lock(&info->unused_bgs_lock);
6327 if (list_empty(&cache->bg_list)) {
6328 btrfs_get_block_group(cache);
6329 list_add_tail(&cache->bg_list,
6332 spin_unlock(&info->unused_bgs_lock);
6335 btrfs_put_block_group(cache);
6337 bytenr += num_bytes;
6342 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6344 struct btrfs_block_group_cache *cache;
6347 spin_lock(&fs_info->block_group_cache_lock);
6348 bytenr = fs_info->first_logical_byte;
6349 spin_unlock(&fs_info->block_group_cache_lock);
6351 if (bytenr < (u64)-1)
6354 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6358 bytenr = cache->key.objectid;
6359 btrfs_put_block_group(cache);
6364 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6365 struct btrfs_block_group_cache *cache,
6366 u64 bytenr, u64 num_bytes, int reserved)
6368 spin_lock(&cache->space_info->lock);
6369 spin_lock(&cache->lock);
6370 cache->pinned += num_bytes;
6371 cache->space_info->bytes_pinned += num_bytes;
6373 cache->reserved -= num_bytes;
6374 cache->space_info->bytes_reserved -= num_bytes;
6376 spin_unlock(&cache->lock);
6377 spin_unlock(&cache->space_info->lock);
6379 trace_btrfs_space_reservation(fs_info, "pinned",
6380 cache->space_info->flags, num_bytes, 1);
6381 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6382 set_extent_dirty(fs_info->pinned_extents, bytenr,
6383 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6388 * this function must be called within transaction
6390 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6391 u64 bytenr, u64 num_bytes, int reserved)
6393 struct btrfs_block_group_cache *cache;
6395 cache = btrfs_lookup_block_group(fs_info, bytenr);
6396 BUG_ON(!cache); /* Logic error */
6398 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6400 btrfs_put_block_group(cache);
6405 * this function must be called within transaction
6407 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6408 u64 bytenr, u64 num_bytes)
6410 struct btrfs_block_group_cache *cache;
6413 cache = btrfs_lookup_block_group(fs_info, bytenr);
6418 * pull in the free space cache (if any) so that our pin
6419 * removes the free space from the cache. We have load_only set
6420 * to one because the slow code to read in the free extents does check
6421 * the pinned extents.
6423 cache_block_group(cache, 1);
6425 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6427 /* remove us from the free space cache (if we're there at all) */
6428 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6429 btrfs_put_block_group(cache);
6433 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6434 u64 start, u64 num_bytes)
6437 struct btrfs_block_group_cache *block_group;
6438 struct btrfs_caching_control *caching_ctl;
6440 block_group = btrfs_lookup_block_group(fs_info, start);
6444 cache_block_group(block_group, 0);
6445 caching_ctl = get_caching_control(block_group);
6449 BUG_ON(!block_group_cache_done(block_group));
6450 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6452 mutex_lock(&caching_ctl->mutex);
6454 if (start >= caching_ctl->progress) {
6455 ret = add_excluded_extent(fs_info, start, num_bytes);
6456 } else if (start + num_bytes <= caching_ctl->progress) {
6457 ret = btrfs_remove_free_space(block_group,
6460 num_bytes = caching_ctl->progress - start;
6461 ret = btrfs_remove_free_space(block_group,
6466 num_bytes = (start + num_bytes) -
6467 caching_ctl->progress;
6468 start = caching_ctl->progress;
6469 ret = add_excluded_extent(fs_info, start, num_bytes);
6472 mutex_unlock(&caching_ctl->mutex);
6473 put_caching_control(caching_ctl);
6475 btrfs_put_block_group(block_group);
6479 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6480 struct extent_buffer *eb)
6482 struct btrfs_file_extent_item *item;
6483 struct btrfs_key key;
6487 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6490 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6491 btrfs_item_key_to_cpu(eb, &key, i);
6492 if (key.type != BTRFS_EXTENT_DATA_KEY)
6494 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6495 found_type = btrfs_file_extent_type(eb, item);
6496 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6498 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6500 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6501 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6502 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6509 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6511 atomic_inc(&bg->reservations);
6514 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6517 struct btrfs_block_group_cache *bg;
6519 bg = btrfs_lookup_block_group(fs_info, start);
6521 if (atomic_dec_and_test(&bg->reservations))
6522 wake_up_atomic_t(&bg->reservations);
6523 btrfs_put_block_group(bg);
6526 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6532 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6534 struct btrfs_space_info *space_info = bg->space_info;
6538 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6542 * Our block group is read only but before we set it to read only,
6543 * some task might have had allocated an extent from it already, but it
6544 * has not yet created a respective ordered extent (and added it to a
6545 * root's list of ordered extents).
6546 * Therefore wait for any task currently allocating extents, since the
6547 * block group's reservations counter is incremented while a read lock
6548 * on the groups' semaphore is held and decremented after releasing
6549 * the read access on that semaphore and creating the ordered extent.
6551 down_write(&space_info->groups_sem);
6552 up_write(&space_info->groups_sem);
6554 wait_on_atomic_t(&bg->reservations,
6555 btrfs_wait_bg_reservations_atomic_t,
6556 TASK_UNINTERRUPTIBLE);
6560 * btrfs_add_reserved_bytes - update the block_group and space info counters
6561 * @cache: The cache we are manipulating
6562 * @ram_bytes: The number of bytes of file content, and will be same to
6563 * @num_bytes except for the compress path.
6564 * @num_bytes: The number of bytes in question
6565 * @delalloc: The blocks are allocated for the delalloc write
6567 * This is called by the allocator when it reserves space. If this is a
6568 * reservation and the block group has become read only we cannot make the
6569 * reservation and return -EAGAIN, otherwise this function always succeeds.
6571 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6572 u64 ram_bytes, u64 num_bytes, int delalloc)
6574 struct btrfs_space_info *space_info = cache->space_info;
6577 spin_lock(&space_info->lock);
6578 spin_lock(&cache->lock);
6582 cache->reserved += num_bytes;
6583 space_info->bytes_reserved += num_bytes;
6585 trace_btrfs_space_reservation(cache->fs_info,
6586 "space_info", space_info->flags,
6588 space_info->bytes_may_use -= ram_bytes;
6590 cache->delalloc_bytes += num_bytes;
6592 spin_unlock(&cache->lock);
6593 spin_unlock(&space_info->lock);
6598 * btrfs_free_reserved_bytes - update the block_group and space info counters
6599 * @cache: The cache we are manipulating
6600 * @num_bytes: The number of bytes in question
6601 * @delalloc: The blocks are allocated for the delalloc write
6603 * This is called by somebody who is freeing space that was never actually used
6604 * on disk. For example if you reserve some space for a new leaf in transaction
6605 * A and before transaction A commits you free that leaf, you call this with
6606 * reserve set to 0 in order to clear the reservation.
6609 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6610 u64 num_bytes, int delalloc)
6612 struct btrfs_space_info *space_info = cache->space_info;
6615 spin_lock(&space_info->lock);
6616 spin_lock(&cache->lock);
6618 space_info->bytes_readonly += num_bytes;
6619 cache->reserved -= num_bytes;
6620 space_info->bytes_reserved -= num_bytes;
6623 cache->delalloc_bytes -= num_bytes;
6624 spin_unlock(&cache->lock);
6625 spin_unlock(&space_info->lock);
6628 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6630 struct btrfs_caching_control *next;
6631 struct btrfs_caching_control *caching_ctl;
6632 struct btrfs_block_group_cache *cache;
6634 down_write(&fs_info->commit_root_sem);
6636 list_for_each_entry_safe(caching_ctl, next,
6637 &fs_info->caching_block_groups, list) {
6638 cache = caching_ctl->block_group;
6639 if (block_group_cache_done(cache)) {
6640 cache->last_byte_to_unpin = (u64)-1;
6641 list_del_init(&caching_ctl->list);
6642 put_caching_control(caching_ctl);
6644 cache->last_byte_to_unpin = caching_ctl->progress;
6648 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6649 fs_info->pinned_extents = &fs_info->freed_extents[1];
6651 fs_info->pinned_extents = &fs_info->freed_extents[0];
6653 up_write(&fs_info->commit_root_sem);
6655 update_global_block_rsv(fs_info);
6659 * Returns the free cluster for the given space info and sets empty_cluster to
6660 * what it should be based on the mount options.
6662 static struct btrfs_free_cluster *
6663 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6664 struct btrfs_space_info *space_info, u64 *empty_cluster)
6666 struct btrfs_free_cluster *ret = NULL;
6667 bool ssd = btrfs_test_opt(fs_info, SSD);
6670 if (btrfs_mixed_space_info(space_info))
6674 *empty_cluster = SZ_2M;
6675 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6676 ret = &fs_info->meta_alloc_cluster;
6678 *empty_cluster = SZ_64K;
6679 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6680 ret = &fs_info->data_alloc_cluster;
6686 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6688 const bool return_free_space)
6690 struct btrfs_block_group_cache *cache = NULL;
6691 struct btrfs_space_info *space_info;
6692 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6693 struct btrfs_free_cluster *cluster = NULL;
6695 u64 total_unpinned = 0;
6696 u64 empty_cluster = 0;
6699 while (start <= end) {
6702 start >= cache->key.objectid + cache->key.offset) {
6704 btrfs_put_block_group(cache);
6706 cache = btrfs_lookup_block_group(fs_info, start);
6707 BUG_ON(!cache); /* Logic error */
6709 cluster = fetch_cluster_info(fs_info,
6712 empty_cluster <<= 1;
6715 len = cache->key.objectid + cache->key.offset - start;
6716 len = min(len, end + 1 - start);
6718 if (start < cache->last_byte_to_unpin) {
6719 len = min(len, cache->last_byte_to_unpin - start);
6720 if (return_free_space)
6721 btrfs_add_free_space(cache, start, len);
6725 total_unpinned += len;
6726 space_info = cache->space_info;
6729 * If this space cluster has been marked as fragmented and we've
6730 * unpinned enough in this block group to potentially allow a
6731 * cluster to be created inside of it go ahead and clear the
6734 if (cluster && cluster->fragmented &&
6735 total_unpinned > empty_cluster) {
6736 spin_lock(&cluster->lock);
6737 cluster->fragmented = 0;
6738 spin_unlock(&cluster->lock);
6741 spin_lock(&space_info->lock);
6742 spin_lock(&cache->lock);
6743 cache->pinned -= len;
6744 space_info->bytes_pinned -= len;
6746 trace_btrfs_space_reservation(fs_info, "pinned",
6747 space_info->flags, len, 0);
6748 space_info->max_extent_size = 0;
6749 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6751 space_info->bytes_readonly += len;
6754 spin_unlock(&cache->lock);
6755 if (!readonly && return_free_space &&
6756 global_rsv->space_info == space_info) {
6758 WARN_ON(!return_free_space);
6759 spin_lock(&global_rsv->lock);
6760 if (!global_rsv->full) {
6761 to_add = min(len, global_rsv->size -
6762 global_rsv->reserved);
6763 global_rsv->reserved += to_add;
6764 space_info->bytes_may_use += to_add;
6765 if (global_rsv->reserved >= global_rsv->size)
6766 global_rsv->full = 1;
6767 trace_btrfs_space_reservation(fs_info,
6773 spin_unlock(&global_rsv->lock);
6774 /* Add to any tickets we may have */
6776 space_info_add_new_bytes(fs_info, space_info,
6779 spin_unlock(&space_info->lock);
6783 btrfs_put_block_group(cache);
6787 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6788 struct btrfs_fs_info *fs_info)
6790 struct btrfs_block_group_cache *block_group, *tmp;
6791 struct list_head *deleted_bgs;
6792 struct extent_io_tree *unpin;
6797 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6798 unpin = &fs_info->freed_extents[1];
6800 unpin = &fs_info->freed_extents[0];
6802 while (!trans->aborted) {
6803 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6804 ret = find_first_extent_bit(unpin, 0, &start, &end,
6805 EXTENT_DIRTY, NULL);
6807 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6811 if (btrfs_test_opt(fs_info, DISCARD))
6812 ret = btrfs_discard_extent(fs_info, start,
6813 end + 1 - start, NULL);
6815 clear_extent_dirty(unpin, start, end);
6816 unpin_extent_range(fs_info, start, end, true);
6817 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6822 * Transaction is finished. We don't need the lock anymore. We
6823 * do need to clean up the block groups in case of a transaction
6826 deleted_bgs = &trans->transaction->deleted_bgs;
6827 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6831 if (!trans->aborted)
6832 ret = btrfs_discard_extent(fs_info,
6833 block_group->key.objectid,
6834 block_group->key.offset,
6837 list_del_init(&block_group->bg_list);
6838 btrfs_put_block_group_trimming(block_group);
6839 btrfs_put_block_group(block_group);
6842 const char *errstr = btrfs_decode_error(ret);
6844 "Discard failed while removing blockgroup: errno=%d %s\n",
6852 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6853 struct btrfs_fs_info *info,
6854 struct btrfs_delayed_ref_node *node, u64 parent,
6855 u64 root_objectid, u64 owner_objectid,
6856 u64 owner_offset, int refs_to_drop,
6857 struct btrfs_delayed_extent_op *extent_op)
6859 struct btrfs_key key;
6860 struct btrfs_path *path;
6861 struct btrfs_root *extent_root = info->extent_root;
6862 struct extent_buffer *leaf;
6863 struct btrfs_extent_item *ei;
6864 struct btrfs_extent_inline_ref *iref;
6867 int extent_slot = 0;
6868 int found_extent = 0;
6872 u64 bytenr = node->bytenr;
6873 u64 num_bytes = node->num_bytes;
6875 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6877 path = btrfs_alloc_path();
6881 path->reada = READA_FORWARD;
6882 path->leave_spinning = 1;
6884 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6885 BUG_ON(!is_data && refs_to_drop != 1);
6888 skinny_metadata = 0;
6890 ret = lookup_extent_backref(trans, info, path, &iref,
6891 bytenr, num_bytes, parent,
6892 root_objectid, owner_objectid,
6895 extent_slot = path->slots[0];
6896 while (extent_slot >= 0) {
6897 btrfs_item_key_to_cpu(path->nodes[0], &key,
6899 if (key.objectid != bytenr)
6901 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6902 key.offset == num_bytes) {
6906 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6907 key.offset == owner_objectid) {
6911 if (path->slots[0] - extent_slot > 5)
6915 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6916 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6917 if (found_extent && item_size < sizeof(*ei))
6920 if (!found_extent) {
6922 ret = remove_extent_backref(trans, info, path, NULL,
6924 is_data, &last_ref);
6926 btrfs_abort_transaction(trans, ret);
6929 btrfs_release_path(path);
6930 path->leave_spinning = 1;
6932 key.objectid = bytenr;
6933 key.type = BTRFS_EXTENT_ITEM_KEY;
6934 key.offset = num_bytes;
6936 if (!is_data && skinny_metadata) {
6937 key.type = BTRFS_METADATA_ITEM_KEY;
6938 key.offset = owner_objectid;
6941 ret = btrfs_search_slot(trans, extent_root,
6943 if (ret > 0 && skinny_metadata && path->slots[0]) {
6945 * Couldn't find our skinny metadata item,
6946 * see if we have ye olde extent item.
6949 btrfs_item_key_to_cpu(path->nodes[0], &key,
6951 if (key.objectid == bytenr &&
6952 key.type == BTRFS_EXTENT_ITEM_KEY &&
6953 key.offset == num_bytes)
6957 if (ret > 0 && skinny_metadata) {
6958 skinny_metadata = false;
6959 key.objectid = bytenr;
6960 key.type = BTRFS_EXTENT_ITEM_KEY;
6961 key.offset = num_bytes;
6962 btrfs_release_path(path);
6963 ret = btrfs_search_slot(trans, extent_root,
6969 "umm, got %d back from search, was looking for %llu",
6972 btrfs_print_leaf(info, path->nodes[0]);
6975 btrfs_abort_transaction(trans, ret);
6978 extent_slot = path->slots[0];
6980 } else if (WARN_ON(ret == -ENOENT)) {
6981 btrfs_print_leaf(info, path->nodes[0]);
6983 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6984 bytenr, parent, root_objectid, owner_objectid,
6986 btrfs_abort_transaction(trans, ret);
6989 btrfs_abort_transaction(trans, ret);
6993 leaf = path->nodes[0];
6994 item_size = btrfs_item_size_nr(leaf, extent_slot);
6995 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6996 if (item_size < sizeof(*ei)) {
6997 BUG_ON(found_extent || extent_slot != path->slots[0]);
6998 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
7001 btrfs_abort_transaction(trans, ret);
7005 btrfs_release_path(path);
7006 path->leave_spinning = 1;
7008 key.objectid = bytenr;
7009 key.type = BTRFS_EXTENT_ITEM_KEY;
7010 key.offset = num_bytes;
7012 ret = btrfs_search_slot(trans, extent_root, &key, path,
7016 "umm, got %d back from search, was looking for %llu",
7018 btrfs_print_leaf(info, path->nodes[0]);
7021 btrfs_abort_transaction(trans, ret);
7025 extent_slot = path->slots[0];
7026 leaf = path->nodes[0];
7027 item_size = btrfs_item_size_nr(leaf, extent_slot);
7030 BUG_ON(item_size < sizeof(*ei));
7031 ei = btrfs_item_ptr(leaf, extent_slot,
7032 struct btrfs_extent_item);
7033 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7034 key.type == BTRFS_EXTENT_ITEM_KEY) {
7035 struct btrfs_tree_block_info *bi;
7036 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7037 bi = (struct btrfs_tree_block_info *)(ei + 1);
7038 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7041 refs = btrfs_extent_refs(leaf, ei);
7042 if (refs < refs_to_drop) {
7044 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7045 refs_to_drop, refs, bytenr);
7047 btrfs_abort_transaction(trans, ret);
7050 refs -= refs_to_drop;
7054 __run_delayed_extent_op(extent_op, leaf, ei);
7056 * In the case of inline back ref, reference count will
7057 * be updated by remove_extent_backref
7060 BUG_ON(!found_extent);
7062 btrfs_set_extent_refs(leaf, ei, refs);
7063 btrfs_mark_buffer_dirty(leaf);
7066 ret = remove_extent_backref(trans, info, path,
7068 is_data, &last_ref);
7070 btrfs_abort_transaction(trans, ret);
7076 BUG_ON(is_data && refs_to_drop !=
7077 extent_data_ref_count(path, iref));
7079 BUG_ON(path->slots[0] != extent_slot);
7081 BUG_ON(path->slots[0] != extent_slot + 1);
7082 path->slots[0] = extent_slot;
7088 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7091 btrfs_abort_transaction(trans, ret);
7094 btrfs_release_path(path);
7097 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7099 btrfs_abort_transaction(trans, ret);
7104 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7106 btrfs_abort_transaction(trans, ret);
7110 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7112 btrfs_abort_transaction(trans, ret);
7116 btrfs_release_path(path);
7119 btrfs_free_path(path);
7124 * when we free an block, it is possible (and likely) that we free the last
7125 * delayed ref for that extent as well. This searches the delayed ref tree for
7126 * a given extent, and if there are no other delayed refs to be processed, it
7127 * removes it from the tree.
7129 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7132 struct btrfs_delayed_ref_head *head;
7133 struct btrfs_delayed_ref_root *delayed_refs;
7136 delayed_refs = &trans->transaction->delayed_refs;
7137 spin_lock(&delayed_refs->lock);
7138 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7140 goto out_delayed_unlock;
7142 spin_lock(&head->lock);
7143 if (!list_empty(&head->ref_list))
7146 if (head->extent_op) {
7147 if (!head->must_insert_reserved)
7149 btrfs_free_delayed_extent_op(head->extent_op);
7150 head->extent_op = NULL;
7154 * waiting for the lock here would deadlock. If someone else has it
7155 * locked they are already in the process of dropping it anyway
7157 if (!mutex_trylock(&head->mutex))
7161 * at this point we have a head with no other entries. Go
7162 * ahead and process it.
7164 head->node.in_tree = 0;
7165 rb_erase(&head->href_node, &delayed_refs->href_root);
7167 atomic_dec(&delayed_refs->num_entries);
7170 * we don't take a ref on the node because we're removing it from the
7171 * tree, so we just steal the ref the tree was holding.
7173 delayed_refs->num_heads--;
7174 if (head->processing == 0)
7175 delayed_refs->num_heads_ready--;
7176 head->processing = 0;
7177 spin_unlock(&head->lock);
7178 spin_unlock(&delayed_refs->lock);
7180 BUG_ON(head->extent_op);
7181 if (head->must_insert_reserved)
7184 mutex_unlock(&head->mutex);
7185 btrfs_put_delayed_ref(&head->node);
7188 spin_unlock(&head->lock);
7191 spin_unlock(&delayed_refs->lock);
7195 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7196 struct btrfs_root *root,
7197 struct extent_buffer *buf,
7198 u64 parent, int last_ref)
7200 struct btrfs_fs_info *fs_info = root->fs_info;
7204 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7205 int old_ref_mod, new_ref_mod;
7207 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7209 root->root_key.objectid,
7210 btrfs_header_level(buf),
7211 BTRFS_DROP_DELAYED_REF, NULL,
7212 &old_ref_mod, &new_ref_mod);
7213 BUG_ON(ret); /* -ENOMEM */
7214 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7217 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7218 struct btrfs_block_group_cache *cache;
7220 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7221 ret = check_ref_cleanup(trans, buf->start);
7227 cache = btrfs_lookup_block_group(fs_info, buf->start);
7229 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7230 pin_down_extent(fs_info, cache, buf->start,
7232 btrfs_put_block_group(cache);
7236 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7238 btrfs_add_free_space(cache, buf->start, buf->len);
7239 btrfs_free_reserved_bytes(cache, buf->len, 0);
7240 btrfs_put_block_group(cache);
7241 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7245 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7246 root->root_key.objectid);
7250 * Deleting the buffer, clear the corrupt flag since it doesn't
7253 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7257 /* Can return -ENOMEM */
7258 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7259 struct btrfs_fs_info *fs_info,
7260 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7261 u64 owner, u64 offset)
7263 int old_ref_mod, new_ref_mod;
7266 if (btrfs_is_testing(fs_info))
7271 * tree log blocks never actually go into the extent allocation
7272 * tree, just update pinning info and exit early.
7274 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7275 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7276 /* unlocks the pinned mutex */
7277 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7278 old_ref_mod = new_ref_mod = 0;
7280 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7281 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7283 root_objectid, (int)owner,
7284 BTRFS_DROP_DELAYED_REF, NULL,
7285 &old_ref_mod, &new_ref_mod);
7287 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7289 root_objectid, owner, offset,
7290 0, BTRFS_DROP_DELAYED_REF,
7291 &old_ref_mod, &new_ref_mod);
7294 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7295 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7301 * when we wait for progress in the block group caching, its because
7302 * our allocation attempt failed at least once. So, we must sleep
7303 * and let some progress happen before we try again.
7305 * This function will sleep at least once waiting for new free space to
7306 * show up, and then it will check the block group free space numbers
7307 * for our min num_bytes. Another option is to have it go ahead
7308 * and look in the rbtree for a free extent of a given size, but this
7311 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7312 * any of the information in this block group.
7314 static noinline void
7315 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7318 struct btrfs_caching_control *caching_ctl;
7320 caching_ctl = get_caching_control(cache);
7324 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7325 (cache->free_space_ctl->free_space >= num_bytes));
7327 put_caching_control(caching_ctl);
7331 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7333 struct btrfs_caching_control *caching_ctl;
7336 caching_ctl = get_caching_control(cache);
7338 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7340 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7341 if (cache->cached == BTRFS_CACHE_ERROR)
7343 put_caching_control(caching_ctl);
7347 int __get_raid_index(u64 flags)
7349 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7350 return BTRFS_RAID_RAID10;
7351 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7352 return BTRFS_RAID_RAID1;
7353 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7354 return BTRFS_RAID_DUP;
7355 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7356 return BTRFS_RAID_RAID0;
7357 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7358 return BTRFS_RAID_RAID5;
7359 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7360 return BTRFS_RAID_RAID6;
7362 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7365 int get_block_group_index(struct btrfs_block_group_cache *cache)
7367 return __get_raid_index(cache->flags);
7370 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7371 [BTRFS_RAID_RAID10] = "raid10",
7372 [BTRFS_RAID_RAID1] = "raid1",
7373 [BTRFS_RAID_DUP] = "dup",
7374 [BTRFS_RAID_RAID0] = "raid0",
7375 [BTRFS_RAID_SINGLE] = "single",
7376 [BTRFS_RAID_RAID5] = "raid5",
7377 [BTRFS_RAID_RAID6] = "raid6",
7380 static const char *get_raid_name(enum btrfs_raid_types type)
7382 if (type >= BTRFS_NR_RAID_TYPES)
7385 return btrfs_raid_type_names[type];
7388 enum btrfs_loop_type {
7389 LOOP_CACHING_NOWAIT = 0,
7390 LOOP_CACHING_WAIT = 1,
7391 LOOP_ALLOC_CHUNK = 2,
7392 LOOP_NO_EMPTY_SIZE = 3,
7396 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7400 down_read(&cache->data_rwsem);
7404 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7407 btrfs_get_block_group(cache);
7409 down_read(&cache->data_rwsem);
7412 static struct btrfs_block_group_cache *
7413 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7414 struct btrfs_free_cluster *cluster,
7417 struct btrfs_block_group_cache *used_bg = NULL;
7419 spin_lock(&cluster->refill_lock);
7421 used_bg = cluster->block_group;
7425 if (used_bg == block_group)
7428 btrfs_get_block_group(used_bg);
7433 if (down_read_trylock(&used_bg->data_rwsem))
7436 spin_unlock(&cluster->refill_lock);
7438 /* We should only have one-level nested. */
7439 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7441 spin_lock(&cluster->refill_lock);
7442 if (used_bg == cluster->block_group)
7445 up_read(&used_bg->data_rwsem);
7446 btrfs_put_block_group(used_bg);
7451 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7455 up_read(&cache->data_rwsem);
7456 btrfs_put_block_group(cache);
7460 * walks the btree of allocated extents and find a hole of a given size.
7461 * The key ins is changed to record the hole:
7462 * ins->objectid == start position
7463 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7464 * ins->offset == the size of the hole.
7465 * Any available blocks before search_start are skipped.
7467 * If there is no suitable free space, we will record the max size of
7468 * the free space extent currently.
7470 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7471 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7472 u64 hint_byte, struct btrfs_key *ins,
7473 u64 flags, int delalloc)
7476 struct btrfs_root *root = fs_info->extent_root;
7477 struct btrfs_free_cluster *last_ptr = NULL;
7478 struct btrfs_block_group_cache *block_group = NULL;
7479 u64 search_start = 0;
7480 u64 max_extent_size = 0;
7481 u64 empty_cluster = 0;
7482 struct btrfs_space_info *space_info;
7484 int index = __get_raid_index(flags);
7485 bool failed_cluster_refill = false;
7486 bool failed_alloc = false;
7487 bool use_cluster = true;
7488 bool have_caching_bg = false;
7489 bool orig_have_caching_bg = false;
7490 bool full_search = false;
7492 WARN_ON(num_bytes < fs_info->sectorsize);
7493 ins->type = BTRFS_EXTENT_ITEM_KEY;
7497 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7499 space_info = __find_space_info(fs_info, flags);
7501 btrfs_err(fs_info, "No space info for %llu", flags);
7506 * If our free space is heavily fragmented we may not be able to make
7507 * big contiguous allocations, so instead of doing the expensive search
7508 * for free space, simply return ENOSPC with our max_extent_size so we
7509 * can go ahead and search for a more manageable chunk.
7511 * If our max_extent_size is large enough for our allocation simply
7512 * disable clustering since we will likely not be able to find enough
7513 * space to create a cluster and induce latency trying.
7515 if (unlikely(space_info->max_extent_size)) {
7516 spin_lock(&space_info->lock);
7517 if (space_info->max_extent_size &&
7518 num_bytes > space_info->max_extent_size) {
7519 ins->offset = space_info->max_extent_size;
7520 spin_unlock(&space_info->lock);
7522 } else if (space_info->max_extent_size) {
7523 use_cluster = false;
7525 spin_unlock(&space_info->lock);
7528 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7530 spin_lock(&last_ptr->lock);
7531 if (last_ptr->block_group)
7532 hint_byte = last_ptr->window_start;
7533 if (last_ptr->fragmented) {
7535 * We still set window_start so we can keep track of the
7536 * last place we found an allocation to try and save
7539 hint_byte = last_ptr->window_start;
7540 use_cluster = false;
7542 spin_unlock(&last_ptr->lock);
7545 search_start = max(search_start, first_logical_byte(fs_info, 0));
7546 search_start = max(search_start, hint_byte);
7547 if (search_start == hint_byte) {
7548 block_group = btrfs_lookup_block_group(fs_info, search_start);
7550 * we don't want to use the block group if it doesn't match our
7551 * allocation bits, or if its not cached.
7553 * However if we are re-searching with an ideal block group
7554 * picked out then we don't care that the block group is cached.
7556 if (block_group && block_group_bits(block_group, flags) &&
7557 block_group->cached != BTRFS_CACHE_NO) {
7558 down_read(&space_info->groups_sem);
7559 if (list_empty(&block_group->list) ||
7562 * someone is removing this block group,
7563 * we can't jump into the have_block_group
7564 * target because our list pointers are not
7567 btrfs_put_block_group(block_group);
7568 up_read(&space_info->groups_sem);
7570 index = get_block_group_index(block_group);
7571 btrfs_lock_block_group(block_group, delalloc);
7572 goto have_block_group;
7574 } else if (block_group) {
7575 btrfs_put_block_group(block_group);
7579 have_caching_bg = false;
7580 if (index == 0 || index == __get_raid_index(flags))
7582 down_read(&space_info->groups_sem);
7583 list_for_each_entry(block_group, &space_info->block_groups[index],
7588 /* If the block group is read-only, we can skip it entirely. */
7589 if (unlikely(block_group->ro))
7592 btrfs_grab_block_group(block_group, delalloc);
7593 search_start = block_group->key.objectid;
7596 * this can happen if we end up cycling through all the
7597 * raid types, but we want to make sure we only allocate
7598 * for the proper type.
7600 if (!block_group_bits(block_group, flags)) {
7601 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7602 BTRFS_BLOCK_GROUP_RAID1 |
7603 BTRFS_BLOCK_GROUP_RAID5 |
7604 BTRFS_BLOCK_GROUP_RAID6 |
7605 BTRFS_BLOCK_GROUP_RAID10;
7608 * if they asked for extra copies and this block group
7609 * doesn't provide them, bail. This does allow us to
7610 * fill raid0 from raid1.
7612 if ((flags & extra) && !(block_group->flags & extra))
7617 cached = block_group_cache_done(block_group);
7618 if (unlikely(!cached)) {
7619 have_caching_bg = true;
7620 ret = cache_block_group(block_group, 0);
7625 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7629 * Ok we want to try and use the cluster allocator, so
7632 if (last_ptr && use_cluster) {
7633 struct btrfs_block_group_cache *used_block_group;
7634 unsigned long aligned_cluster;
7636 * the refill lock keeps out other
7637 * people trying to start a new cluster
7639 used_block_group = btrfs_lock_cluster(block_group,
7642 if (!used_block_group)
7643 goto refill_cluster;
7645 if (used_block_group != block_group &&
7646 (used_block_group->ro ||
7647 !block_group_bits(used_block_group, flags)))
7648 goto release_cluster;
7650 offset = btrfs_alloc_from_cluster(used_block_group,
7653 used_block_group->key.objectid,
7656 /* we have a block, we're done */
7657 spin_unlock(&last_ptr->refill_lock);
7658 trace_btrfs_reserve_extent_cluster(fs_info,
7660 search_start, num_bytes);
7661 if (used_block_group != block_group) {
7662 btrfs_release_block_group(block_group,
7664 block_group = used_block_group;
7669 WARN_ON(last_ptr->block_group != used_block_group);
7671 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7672 * set up a new clusters, so lets just skip it
7673 * and let the allocator find whatever block
7674 * it can find. If we reach this point, we
7675 * will have tried the cluster allocator
7676 * plenty of times and not have found
7677 * anything, so we are likely way too
7678 * fragmented for the clustering stuff to find
7681 * However, if the cluster is taken from the
7682 * current block group, release the cluster
7683 * first, so that we stand a better chance of
7684 * succeeding in the unclustered
7686 if (loop >= LOOP_NO_EMPTY_SIZE &&
7687 used_block_group != block_group) {
7688 spin_unlock(&last_ptr->refill_lock);
7689 btrfs_release_block_group(used_block_group,
7691 goto unclustered_alloc;
7695 * this cluster didn't work out, free it and
7698 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7700 if (used_block_group != block_group)
7701 btrfs_release_block_group(used_block_group,
7704 if (loop >= LOOP_NO_EMPTY_SIZE) {
7705 spin_unlock(&last_ptr->refill_lock);
7706 goto unclustered_alloc;
7709 aligned_cluster = max_t(unsigned long,
7710 empty_cluster + empty_size,
7711 block_group->full_stripe_len);
7713 /* allocate a cluster in this block group */
7714 ret = btrfs_find_space_cluster(fs_info, block_group,
7715 last_ptr, search_start,
7720 * now pull our allocation out of this
7723 offset = btrfs_alloc_from_cluster(block_group,
7729 /* we found one, proceed */
7730 spin_unlock(&last_ptr->refill_lock);
7731 trace_btrfs_reserve_extent_cluster(fs_info,
7732 block_group, search_start,
7736 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7737 && !failed_cluster_refill) {
7738 spin_unlock(&last_ptr->refill_lock);
7740 failed_cluster_refill = true;
7741 wait_block_group_cache_progress(block_group,
7742 num_bytes + empty_cluster + empty_size);
7743 goto have_block_group;
7747 * at this point we either didn't find a cluster
7748 * or we weren't able to allocate a block from our
7749 * cluster. Free the cluster we've been trying
7750 * to use, and go to the next block group
7752 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7753 spin_unlock(&last_ptr->refill_lock);
7759 * We are doing an unclustered alloc, set the fragmented flag so
7760 * we don't bother trying to setup a cluster again until we get
7763 if (unlikely(last_ptr)) {
7764 spin_lock(&last_ptr->lock);
7765 last_ptr->fragmented = 1;
7766 spin_unlock(&last_ptr->lock);
7769 struct btrfs_free_space_ctl *ctl =
7770 block_group->free_space_ctl;
7772 spin_lock(&ctl->tree_lock);
7773 if (ctl->free_space <
7774 num_bytes + empty_cluster + empty_size) {
7775 if (ctl->free_space > max_extent_size)
7776 max_extent_size = ctl->free_space;
7777 spin_unlock(&ctl->tree_lock);
7780 spin_unlock(&ctl->tree_lock);
7783 offset = btrfs_find_space_for_alloc(block_group, search_start,
7784 num_bytes, empty_size,
7787 * If we didn't find a chunk, and we haven't failed on this
7788 * block group before, and this block group is in the middle of
7789 * caching and we are ok with waiting, then go ahead and wait
7790 * for progress to be made, and set failed_alloc to true.
7792 * If failed_alloc is true then we've already waited on this
7793 * block group once and should move on to the next block group.
7795 if (!offset && !failed_alloc && !cached &&
7796 loop > LOOP_CACHING_NOWAIT) {
7797 wait_block_group_cache_progress(block_group,
7798 num_bytes + empty_size);
7799 failed_alloc = true;
7800 goto have_block_group;
7801 } else if (!offset) {
7805 search_start = ALIGN(offset, fs_info->stripesize);
7807 /* move on to the next group */
7808 if (search_start + num_bytes >
7809 block_group->key.objectid + block_group->key.offset) {
7810 btrfs_add_free_space(block_group, offset, num_bytes);
7814 if (offset < search_start)
7815 btrfs_add_free_space(block_group, offset,
7816 search_start - offset);
7817 BUG_ON(offset > search_start);
7819 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7820 num_bytes, delalloc);
7821 if (ret == -EAGAIN) {
7822 btrfs_add_free_space(block_group, offset, num_bytes);
7825 btrfs_inc_block_group_reservations(block_group);
7827 /* we are all good, lets return */
7828 ins->objectid = search_start;
7829 ins->offset = num_bytes;
7831 trace_btrfs_reserve_extent(fs_info, block_group,
7832 search_start, num_bytes);
7833 btrfs_release_block_group(block_group, delalloc);
7836 failed_cluster_refill = false;
7837 failed_alloc = false;
7838 BUG_ON(index != get_block_group_index(block_group));
7839 btrfs_release_block_group(block_group, delalloc);
7842 up_read(&space_info->groups_sem);
7844 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7845 && !orig_have_caching_bg)
7846 orig_have_caching_bg = true;
7848 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7851 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7855 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7856 * caching kthreads as we move along
7857 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7858 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7859 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7862 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7864 if (loop == LOOP_CACHING_NOWAIT) {
7866 * We want to skip the LOOP_CACHING_WAIT step if we
7867 * don't have any uncached bgs and we've already done a
7868 * full search through.
7870 if (orig_have_caching_bg || !full_search)
7871 loop = LOOP_CACHING_WAIT;
7873 loop = LOOP_ALLOC_CHUNK;
7878 if (loop == LOOP_ALLOC_CHUNK) {
7879 struct btrfs_trans_handle *trans;
7882 trans = current->journal_info;
7886 trans = btrfs_join_transaction(root);
7888 if (IS_ERR(trans)) {
7889 ret = PTR_ERR(trans);
7893 ret = do_chunk_alloc(trans, fs_info, flags,
7897 * If we can't allocate a new chunk we've already looped
7898 * through at least once, move on to the NO_EMPTY_SIZE
7902 loop = LOOP_NO_EMPTY_SIZE;
7905 * Do not bail out on ENOSPC since we
7906 * can do more things.
7908 if (ret < 0 && ret != -ENOSPC)
7909 btrfs_abort_transaction(trans, ret);
7913 btrfs_end_transaction(trans);
7918 if (loop == LOOP_NO_EMPTY_SIZE) {
7920 * Don't loop again if we already have no empty_size and
7923 if (empty_size == 0 &&
7924 empty_cluster == 0) {
7933 } else if (!ins->objectid) {
7935 } else if (ins->objectid) {
7936 if (!use_cluster && last_ptr) {
7937 spin_lock(&last_ptr->lock);
7938 last_ptr->window_start = ins->objectid;
7939 spin_unlock(&last_ptr->lock);
7944 if (ret == -ENOSPC) {
7945 spin_lock(&space_info->lock);
7946 space_info->max_extent_size = max_extent_size;
7947 spin_unlock(&space_info->lock);
7948 ins->offset = max_extent_size;
7953 static void dump_space_info(struct btrfs_fs_info *fs_info,
7954 struct btrfs_space_info *info, u64 bytes,
7955 int dump_block_groups)
7957 struct btrfs_block_group_cache *cache;
7960 spin_lock(&info->lock);
7961 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7963 info->total_bytes - btrfs_space_info_used(info, true),
7964 info->full ? "" : "not ");
7966 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7967 info->total_bytes, info->bytes_used, info->bytes_pinned,
7968 info->bytes_reserved, info->bytes_may_use,
7969 info->bytes_readonly);
7970 spin_unlock(&info->lock);
7972 if (!dump_block_groups)
7975 down_read(&info->groups_sem);
7977 list_for_each_entry(cache, &info->block_groups[index], list) {
7978 spin_lock(&cache->lock);
7980 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7981 cache->key.objectid, cache->key.offset,
7982 btrfs_block_group_used(&cache->item), cache->pinned,
7983 cache->reserved, cache->ro ? "[readonly]" : "");
7984 btrfs_dump_free_space(cache, bytes);
7985 spin_unlock(&cache->lock);
7987 if (++index < BTRFS_NR_RAID_TYPES)
7989 up_read(&info->groups_sem);
7992 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7993 u64 num_bytes, u64 min_alloc_size,
7994 u64 empty_size, u64 hint_byte,
7995 struct btrfs_key *ins, int is_data, int delalloc)
7997 struct btrfs_fs_info *fs_info = root->fs_info;
7998 bool final_tried = num_bytes == min_alloc_size;
8002 flags = get_alloc_profile_by_root(root, is_data);
8004 WARN_ON(num_bytes < fs_info->sectorsize);
8005 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8006 hint_byte, ins, flags, delalloc);
8007 if (!ret && !is_data) {
8008 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8009 } else if (ret == -ENOSPC) {
8010 if (!final_tried && ins->offset) {
8011 num_bytes = min(num_bytes >> 1, ins->offset);
8012 num_bytes = round_down(num_bytes,
8013 fs_info->sectorsize);
8014 num_bytes = max(num_bytes, min_alloc_size);
8015 ram_bytes = num_bytes;
8016 if (num_bytes == min_alloc_size)
8019 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8020 struct btrfs_space_info *sinfo;
8022 sinfo = __find_space_info(fs_info, flags);
8024 "allocation failed flags %llu, wanted %llu",
8027 dump_space_info(fs_info, sinfo, num_bytes, 1);
8034 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8036 int pin, int delalloc)
8038 struct btrfs_block_group_cache *cache;
8041 cache = btrfs_lookup_block_group(fs_info, start);
8043 btrfs_err(fs_info, "Unable to find block group for %llu",
8049 pin_down_extent(fs_info, cache, start, len, 1);
8051 if (btrfs_test_opt(fs_info, DISCARD))
8052 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8053 btrfs_add_free_space(cache, start, len);
8054 btrfs_free_reserved_bytes(cache, len, delalloc);
8055 trace_btrfs_reserved_extent_free(fs_info, start, len);
8058 btrfs_put_block_group(cache);
8062 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8063 u64 start, u64 len, int delalloc)
8065 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8068 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8071 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8074 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8075 struct btrfs_fs_info *fs_info,
8076 u64 parent, u64 root_objectid,
8077 u64 flags, u64 owner, u64 offset,
8078 struct btrfs_key *ins, int ref_mod)
8081 struct btrfs_extent_item *extent_item;
8082 struct btrfs_extent_inline_ref *iref;
8083 struct btrfs_path *path;
8084 struct extent_buffer *leaf;
8089 type = BTRFS_SHARED_DATA_REF_KEY;
8091 type = BTRFS_EXTENT_DATA_REF_KEY;
8093 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8095 path = btrfs_alloc_path();
8099 path->leave_spinning = 1;
8100 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8103 btrfs_free_path(path);
8107 leaf = path->nodes[0];
8108 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8109 struct btrfs_extent_item);
8110 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8111 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8112 btrfs_set_extent_flags(leaf, extent_item,
8113 flags | BTRFS_EXTENT_FLAG_DATA);
8115 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8116 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8118 struct btrfs_shared_data_ref *ref;
8119 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8120 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8121 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8123 struct btrfs_extent_data_ref *ref;
8124 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8125 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8126 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8127 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8128 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8131 btrfs_mark_buffer_dirty(path->nodes[0]);
8132 btrfs_free_path(path);
8134 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8139 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8140 if (ret) { /* -ENOENT, logic error */
8141 btrfs_err(fs_info, "update block group failed for %llu %llu",
8142 ins->objectid, ins->offset);
8145 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8149 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8150 struct btrfs_fs_info *fs_info,
8151 u64 parent, u64 root_objectid,
8152 u64 flags, struct btrfs_disk_key *key,
8153 int level, struct btrfs_key *ins)
8156 struct btrfs_extent_item *extent_item;
8157 struct btrfs_tree_block_info *block_info;
8158 struct btrfs_extent_inline_ref *iref;
8159 struct btrfs_path *path;
8160 struct extent_buffer *leaf;
8161 u32 size = sizeof(*extent_item) + sizeof(*iref);
8162 u64 num_bytes = ins->offset;
8163 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8165 if (!skinny_metadata)
8166 size += sizeof(*block_info);
8168 path = btrfs_alloc_path();
8170 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8175 path->leave_spinning = 1;
8176 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8179 btrfs_free_path(path);
8180 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8185 leaf = path->nodes[0];
8186 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8187 struct btrfs_extent_item);
8188 btrfs_set_extent_refs(leaf, extent_item, 1);
8189 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8190 btrfs_set_extent_flags(leaf, extent_item,
8191 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8193 if (skinny_metadata) {
8194 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8195 num_bytes = fs_info->nodesize;
8197 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8198 btrfs_set_tree_block_key(leaf, block_info, key);
8199 btrfs_set_tree_block_level(leaf, block_info, level);
8200 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8204 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8205 btrfs_set_extent_inline_ref_type(leaf, iref,
8206 BTRFS_SHARED_BLOCK_REF_KEY);
8207 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8209 btrfs_set_extent_inline_ref_type(leaf, iref,
8210 BTRFS_TREE_BLOCK_REF_KEY);
8211 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8214 btrfs_mark_buffer_dirty(leaf);
8215 btrfs_free_path(path);
8217 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8222 ret = update_block_group(trans, fs_info, ins->objectid,
8223 fs_info->nodesize, 1);
8224 if (ret) { /* -ENOENT, logic error */
8225 btrfs_err(fs_info, "update block group failed for %llu %llu",
8226 ins->objectid, ins->offset);
8230 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8235 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8236 u64 root_objectid, u64 owner,
8237 u64 offset, u64 ram_bytes,
8238 struct btrfs_key *ins)
8240 struct btrfs_fs_info *fs_info = trans->fs_info;
8243 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8245 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8246 ins->offset, 0, root_objectid, owner,
8248 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8253 * this is used by the tree logging recovery code. It records that
8254 * an extent has been allocated and makes sure to clear the free
8255 * space cache bits as well
8257 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8258 struct btrfs_fs_info *fs_info,
8259 u64 root_objectid, u64 owner, u64 offset,
8260 struct btrfs_key *ins)
8263 struct btrfs_block_group_cache *block_group;
8264 struct btrfs_space_info *space_info;
8267 * Mixed block groups will exclude before processing the log so we only
8268 * need to do the exclude dance if this fs isn't mixed.
8270 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8271 ret = __exclude_logged_extent(fs_info, ins->objectid,
8277 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8281 space_info = block_group->space_info;
8282 spin_lock(&space_info->lock);
8283 spin_lock(&block_group->lock);
8284 space_info->bytes_reserved += ins->offset;
8285 block_group->reserved += ins->offset;
8286 spin_unlock(&block_group->lock);
8287 spin_unlock(&space_info->lock);
8289 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8290 0, owner, offset, ins, 1);
8291 btrfs_put_block_group(block_group);
8295 static struct extent_buffer *
8296 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8297 u64 bytenr, int level)
8299 struct btrfs_fs_info *fs_info = root->fs_info;
8300 struct extent_buffer *buf;
8302 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8306 btrfs_set_header_generation(buf, trans->transid);
8307 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8308 btrfs_tree_lock(buf);
8309 clean_tree_block(fs_info, buf);
8310 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8312 btrfs_set_lock_blocking(buf);
8313 set_extent_buffer_uptodate(buf);
8315 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8316 buf->log_index = root->log_transid % 2;
8318 * we allow two log transactions at a time, use different
8319 * EXENT bit to differentiate dirty pages.
8321 if (buf->log_index == 0)
8322 set_extent_dirty(&root->dirty_log_pages, buf->start,
8323 buf->start + buf->len - 1, GFP_NOFS);
8325 set_extent_new(&root->dirty_log_pages, buf->start,
8326 buf->start + buf->len - 1);
8328 buf->log_index = -1;
8329 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8330 buf->start + buf->len - 1, GFP_NOFS);
8332 trans->dirty = true;
8333 /* this returns a buffer locked for blocking */
8337 static struct btrfs_block_rsv *
8338 use_block_rsv(struct btrfs_trans_handle *trans,
8339 struct btrfs_root *root, u32 blocksize)
8341 struct btrfs_fs_info *fs_info = root->fs_info;
8342 struct btrfs_block_rsv *block_rsv;
8343 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8345 bool global_updated = false;
8347 block_rsv = get_block_rsv(trans, root);
8349 if (unlikely(block_rsv->size == 0))
8352 ret = block_rsv_use_bytes(block_rsv, blocksize);
8356 if (block_rsv->failfast)
8357 return ERR_PTR(ret);
8359 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8360 global_updated = true;
8361 update_global_block_rsv(fs_info);
8365 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8366 static DEFINE_RATELIMIT_STATE(_rs,
8367 DEFAULT_RATELIMIT_INTERVAL * 10,
8368 /*DEFAULT_RATELIMIT_BURST*/ 1);
8369 if (__ratelimit(&_rs))
8371 "BTRFS: block rsv returned %d\n", ret);
8374 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8375 BTRFS_RESERVE_NO_FLUSH);
8379 * If we couldn't reserve metadata bytes try and use some from
8380 * the global reserve if its space type is the same as the global
8383 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8384 block_rsv->space_info == global_rsv->space_info) {
8385 ret = block_rsv_use_bytes(global_rsv, blocksize);
8389 return ERR_PTR(ret);
8392 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8393 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8395 block_rsv_add_bytes(block_rsv, blocksize, 0);
8396 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8400 * finds a free extent and does all the dirty work required for allocation
8401 * returns the tree buffer or an ERR_PTR on error.
8403 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8404 struct btrfs_root *root,
8405 u64 parent, u64 root_objectid,
8406 const struct btrfs_disk_key *key,
8407 int level, u64 hint,
8410 struct btrfs_fs_info *fs_info = root->fs_info;
8411 struct btrfs_key ins;
8412 struct btrfs_block_rsv *block_rsv;
8413 struct extent_buffer *buf;
8414 struct btrfs_delayed_extent_op *extent_op;
8417 u32 blocksize = fs_info->nodesize;
8418 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8420 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8421 if (btrfs_is_testing(fs_info)) {
8422 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8425 root->alloc_bytenr += blocksize;
8430 block_rsv = use_block_rsv(trans, root, blocksize);
8431 if (IS_ERR(block_rsv))
8432 return ERR_CAST(block_rsv);
8434 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8435 empty_size, hint, &ins, 0, 0);
8439 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8442 goto out_free_reserved;
8445 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8447 parent = ins.objectid;
8448 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8452 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8453 extent_op = btrfs_alloc_delayed_extent_op();
8459 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8461 memset(&extent_op->key, 0, sizeof(extent_op->key));
8462 extent_op->flags_to_set = flags;
8463 extent_op->update_key = skinny_metadata ? false : true;
8464 extent_op->update_flags = true;
8465 extent_op->is_data = false;
8466 extent_op->level = level;
8468 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8470 root_objectid, level,
8471 BTRFS_ADD_DELAYED_EXTENT,
8472 extent_op, NULL, NULL);
8474 goto out_free_delayed;
8479 btrfs_free_delayed_extent_op(extent_op);
8481 free_extent_buffer(buf);
8483 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8485 unuse_block_rsv(fs_info, block_rsv, blocksize);
8486 return ERR_PTR(ret);
8489 struct walk_control {
8490 u64 refs[BTRFS_MAX_LEVEL];
8491 u64 flags[BTRFS_MAX_LEVEL];
8492 struct btrfs_key update_progress;
8503 #define DROP_REFERENCE 1
8504 #define UPDATE_BACKREF 2
8506 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8507 struct btrfs_root *root,
8508 struct walk_control *wc,
8509 struct btrfs_path *path)
8511 struct btrfs_fs_info *fs_info = root->fs_info;
8517 struct btrfs_key key;
8518 struct extent_buffer *eb;
8523 if (path->slots[wc->level] < wc->reada_slot) {
8524 wc->reada_count = wc->reada_count * 2 / 3;
8525 wc->reada_count = max(wc->reada_count, 2);
8527 wc->reada_count = wc->reada_count * 3 / 2;
8528 wc->reada_count = min_t(int, wc->reada_count,
8529 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8532 eb = path->nodes[wc->level];
8533 nritems = btrfs_header_nritems(eb);
8535 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8536 if (nread >= wc->reada_count)
8540 bytenr = btrfs_node_blockptr(eb, slot);
8541 generation = btrfs_node_ptr_generation(eb, slot);
8543 if (slot == path->slots[wc->level])
8546 if (wc->stage == UPDATE_BACKREF &&
8547 generation <= root->root_key.offset)
8550 /* We don't lock the tree block, it's OK to be racy here */
8551 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8552 wc->level - 1, 1, &refs,
8554 /* We don't care about errors in readahead. */
8559 if (wc->stage == DROP_REFERENCE) {
8563 if (wc->level == 1 &&
8564 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8566 if (!wc->update_ref ||
8567 generation <= root->root_key.offset)
8569 btrfs_node_key_to_cpu(eb, &key, slot);
8570 ret = btrfs_comp_cpu_keys(&key,
8571 &wc->update_progress);
8575 if (wc->level == 1 &&
8576 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8580 readahead_tree_block(fs_info, bytenr);
8583 wc->reada_slot = slot;
8587 * helper to process tree block while walking down the tree.
8589 * when wc->stage == UPDATE_BACKREF, this function updates
8590 * back refs for pointers in the block.
8592 * NOTE: return value 1 means we should stop walking down.
8594 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8595 struct btrfs_root *root,
8596 struct btrfs_path *path,
8597 struct walk_control *wc, int lookup_info)
8599 struct btrfs_fs_info *fs_info = root->fs_info;
8600 int level = wc->level;
8601 struct extent_buffer *eb = path->nodes[level];
8602 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8605 if (wc->stage == UPDATE_BACKREF &&
8606 btrfs_header_owner(eb) != root->root_key.objectid)
8610 * when reference count of tree block is 1, it won't increase
8611 * again. once full backref flag is set, we never clear it.
8614 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8615 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8616 BUG_ON(!path->locks[level]);
8617 ret = btrfs_lookup_extent_info(trans, fs_info,
8618 eb->start, level, 1,
8621 BUG_ON(ret == -ENOMEM);
8624 BUG_ON(wc->refs[level] == 0);
8627 if (wc->stage == DROP_REFERENCE) {
8628 if (wc->refs[level] > 1)
8631 if (path->locks[level] && !wc->keep_locks) {
8632 btrfs_tree_unlock_rw(eb, path->locks[level]);
8633 path->locks[level] = 0;
8638 /* wc->stage == UPDATE_BACKREF */
8639 if (!(wc->flags[level] & flag)) {
8640 BUG_ON(!path->locks[level]);
8641 ret = btrfs_inc_ref(trans, root, eb, 1);
8642 BUG_ON(ret); /* -ENOMEM */
8643 ret = btrfs_dec_ref(trans, root, eb, 0);
8644 BUG_ON(ret); /* -ENOMEM */
8645 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8647 btrfs_header_level(eb), 0);
8648 BUG_ON(ret); /* -ENOMEM */
8649 wc->flags[level] |= flag;
8653 * the block is shared by multiple trees, so it's not good to
8654 * keep the tree lock
8656 if (path->locks[level] && level > 0) {
8657 btrfs_tree_unlock_rw(eb, path->locks[level]);
8658 path->locks[level] = 0;
8664 * helper to process tree block pointer.
8666 * when wc->stage == DROP_REFERENCE, this function checks
8667 * reference count of the block pointed to. if the block
8668 * is shared and we need update back refs for the subtree
8669 * rooted at the block, this function changes wc->stage to
8670 * UPDATE_BACKREF. if the block is shared and there is no
8671 * need to update back, this function drops the reference
8674 * NOTE: return value 1 means we should stop walking down.
8676 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8677 struct btrfs_root *root,
8678 struct btrfs_path *path,
8679 struct walk_control *wc, int *lookup_info)
8681 struct btrfs_fs_info *fs_info = root->fs_info;
8686 struct btrfs_key key;
8687 struct extent_buffer *next;
8688 int level = wc->level;
8691 bool need_account = false;
8693 generation = btrfs_node_ptr_generation(path->nodes[level],
8694 path->slots[level]);
8696 * if the lower level block was created before the snapshot
8697 * was created, we know there is no need to update back refs
8700 if (wc->stage == UPDATE_BACKREF &&
8701 generation <= root->root_key.offset) {
8706 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8707 blocksize = fs_info->nodesize;
8709 next = find_extent_buffer(fs_info, bytenr);
8711 next = btrfs_find_create_tree_block(fs_info, bytenr);
8713 return PTR_ERR(next);
8715 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8719 btrfs_tree_lock(next);
8720 btrfs_set_lock_blocking(next);
8722 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8723 &wc->refs[level - 1],
8724 &wc->flags[level - 1]);
8728 if (unlikely(wc->refs[level - 1] == 0)) {
8729 btrfs_err(fs_info, "Missing references.");
8735 if (wc->stage == DROP_REFERENCE) {
8736 if (wc->refs[level - 1] > 1) {
8737 need_account = true;
8739 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8742 if (!wc->update_ref ||
8743 generation <= root->root_key.offset)
8746 btrfs_node_key_to_cpu(path->nodes[level], &key,
8747 path->slots[level]);
8748 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8752 wc->stage = UPDATE_BACKREF;
8753 wc->shared_level = level - 1;
8757 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8761 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8762 btrfs_tree_unlock(next);
8763 free_extent_buffer(next);
8769 if (reada && level == 1)
8770 reada_walk_down(trans, root, wc, path);
8771 next = read_tree_block(fs_info, bytenr, generation);
8773 return PTR_ERR(next);
8774 } else if (!extent_buffer_uptodate(next)) {
8775 free_extent_buffer(next);
8778 btrfs_tree_lock(next);
8779 btrfs_set_lock_blocking(next);
8783 ASSERT(level == btrfs_header_level(next));
8784 if (level != btrfs_header_level(next)) {
8785 btrfs_err(root->fs_info, "mismatched level");
8789 path->nodes[level] = next;
8790 path->slots[level] = 0;
8791 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8797 wc->refs[level - 1] = 0;
8798 wc->flags[level - 1] = 0;
8799 if (wc->stage == DROP_REFERENCE) {
8800 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8801 parent = path->nodes[level]->start;
8803 ASSERT(root->root_key.objectid ==
8804 btrfs_header_owner(path->nodes[level]));
8805 if (root->root_key.objectid !=
8806 btrfs_header_owner(path->nodes[level])) {
8807 btrfs_err(root->fs_info,
8808 "mismatched block owner");
8816 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8817 generation, level - 1);
8819 btrfs_err_rl(fs_info,
8820 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8824 ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize,
8825 parent, root->root_key.objectid,
8835 btrfs_tree_unlock(next);
8836 free_extent_buffer(next);
8842 * helper to process tree block while walking up the tree.
8844 * when wc->stage == DROP_REFERENCE, this function drops
8845 * reference count on the block.
8847 * when wc->stage == UPDATE_BACKREF, this function changes
8848 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8849 * to UPDATE_BACKREF previously while processing the block.
8851 * NOTE: return value 1 means we should stop walking up.
8853 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8854 struct btrfs_root *root,
8855 struct btrfs_path *path,
8856 struct walk_control *wc)
8858 struct btrfs_fs_info *fs_info = root->fs_info;
8860 int level = wc->level;
8861 struct extent_buffer *eb = path->nodes[level];
8864 if (wc->stage == UPDATE_BACKREF) {
8865 BUG_ON(wc->shared_level < level);
8866 if (level < wc->shared_level)
8869 ret = find_next_key(path, level + 1, &wc->update_progress);
8873 wc->stage = DROP_REFERENCE;
8874 wc->shared_level = -1;
8875 path->slots[level] = 0;
8878 * check reference count again if the block isn't locked.
8879 * we should start walking down the tree again if reference
8882 if (!path->locks[level]) {
8884 btrfs_tree_lock(eb);
8885 btrfs_set_lock_blocking(eb);
8886 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8888 ret = btrfs_lookup_extent_info(trans, fs_info,
8889 eb->start, level, 1,
8893 btrfs_tree_unlock_rw(eb, path->locks[level]);
8894 path->locks[level] = 0;
8897 BUG_ON(wc->refs[level] == 0);
8898 if (wc->refs[level] == 1) {
8899 btrfs_tree_unlock_rw(eb, path->locks[level]);
8900 path->locks[level] = 0;
8906 /* wc->stage == DROP_REFERENCE */
8907 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8909 if (wc->refs[level] == 1) {
8911 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8912 ret = btrfs_dec_ref(trans, root, eb, 1);
8914 ret = btrfs_dec_ref(trans, root, eb, 0);
8915 BUG_ON(ret); /* -ENOMEM */
8916 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8918 btrfs_err_rl(fs_info,
8919 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8923 /* make block locked assertion in clean_tree_block happy */
8924 if (!path->locks[level] &&
8925 btrfs_header_generation(eb) == trans->transid) {
8926 btrfs_tree_lock(eb);
8927 btrfs_set_lock_blocking(eb);
8928 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8930 clean_tree_block(fs_info, eb);
8933 if (eb == root->node) {
8934 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8937 BUG_ON(root->root_key.objectid !=
8938 btrfs_header_owner(eb));
8940 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8941 parent = path->nodes[level + 1]->start;
8943 BUG_ON(root->root_key.objectid !=
8944 btrfs_header_owner(path->nodes[level + 1]));
8947 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8949 wc->refs[level] = 0;
8950 wc->flags[level] = 0;
8954 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8955 struct btrfs_root *root,
8956 struct btrfs_path *path,
8957 struct walk_control *wc)
8959 int level = wc->level;
8960 int lookup_info = 1;
8963 while (level >= 0) {
8964 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8971 if (path->slots[level] >=
8972 btrfs_header_nritems(path->nodes[level]))
8975 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8977 path->slots[level]++;
8986 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8987 struct btrfs_root *root,
8988 struct btrfs_path *path,
8989 struct walk_control *wc, int max_level)
8991 int level = wc->level;
8994 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8995 while (level < max_level && path->nodes[level]) {
8997 if (path->slots[level] + 1 <
8998 btrfs_header_nritems(path->nodes[level])) {
8999 path->slots[level]++;
9002 ret = walk_up_proc(trans, root, path, wc);
9006 if (path->locks[level]) {
9007 btrfs_tree_unlock_rw(path->nodes[level],
9008 path->locks[level]);
9009 path->locks[level] = 0;
9011 free_extent_buffer(path->nodes[level]);
9012 path->nodes[level] = NULL;
9020 * drop a subvolume tree.
9022 * this function traverses the tree freeing any blocks that only
9023 * referenced by the tree.
9025 * when a shared tree block is found. this function decreases its
9026 * reference count by one. if update_ref is true, this function
9027 * also make sure backrefs for the shared block and all lower level
9028 * blocks are properly updated.
9030 * If called with for_reloc == 0, may exit early with -EAGAIN
9032 int btrfs_drop_snapshot(struct btrfs_root *root,
9033 struct btrfs_block_rsv *block_rsv, int update_ref,
9036 struct btrfs_fs_info *fs_info = root->fs_info;
9037 struct btrfs_path *path;
9038 struct btrfs_trans_handle *trans;
9039 struct btrfs_root *tree_root = fs_info->tree_root;
9040 struct btrfs_root_item *root_item = &root->root_item;
9041 struct walk_control *wc;
9042 struct btrfs_key key;
9046 bool root_dropped = false;
9048 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9050 path = btrfs_alloc_path();
9056 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9058 btrfs_free_path(path);
9063 trans = btrfs_start_transaction(tree_root, 0);
9064 if (IS_ERR(trans)) {
9065 err = PTR_ERR(trans);
9070 trans->block_rsv = block_rsv;
9072 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9073 level = btrfs_header_level(root->node);
9074 path->nodes[level] = btrfs_lock_root_node(root);
9075 btrfs_set_lock_blocking(path->nodes[level]);
9076 path->slots[level] = 0;
9077 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9078 memset(&wc->update_progress, 0,
9079 sizeof(wc->update_progress));
9081 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9082 memcpy(&wc->update_progress, &key,
9083 sizeof(wc->update_progress));
9085 level = root_item->drop_level;
9087 path->lowest_level = level;
9088 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9089 path->lowest_level = 0;
9097 * unlock our path, this is safe because only this
9098 * function is allowed to delete this snapshot
9100 btrfs_unlock_up_safe(path, 0);
9102 level = btrfs_header_level(root->node);
9104 btrfs_tree_lock(path->nodes[level]);
9105 btrfs_set_lock_blocking(path->nodes[level]);
9106 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9108 ret = btrfs_lookup_extent_info(trans, fs_info,
9109 path->nodes[level]->start,
9110 level, 1, &wc->refs[level],
9116 BUG_ON(wc->refs[level] == 0);
9118 if (level == root_item->drop_level)
9121 btrfs_tree_unlock(path->nodes[level]);
9122 path->locks[level] = 0;
9123 WARN_ON(wc->refs[level] != 1);
9129 wc->shared_level = -1;
9130 wc->stage = DROP_REFERENCE;
9131 wc->update_ref = update_ref;
9133 wc->for_reloc = for_reloc;
9134 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9138 ret = walk_down_tree(trans, root, path, wc);
9144 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9151 BUG_ON(wc->stage != DROP_REFERENCE);
9155 if (wc->stage == DROP_REFERENCE) {
9157 btrfs_node_key(path->nodes[level],
9158 &root_item->drop_progress,
9159 path->slots[level]);
9160 root_item->drop_level = level;
9163 BUG_ON(wc->level == 0);
9164 if (btrfs_should_end_transaction(trans) ||
9165 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9166 ret = btrfs_update_root(trans, tree_root,
9170 btrfs_abort_transaction(trans, ret);
9175 btrfs_end_transaction_throttle(trans);
9176 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9177 btrfs_debug(fs_info,
9178 "drop snapshot early exit");
9183 trans = btrfs_start_transaction(tree_root, 0);
9184 if (IS_ERR(trans)) {
9185 err = PTR_ERR(trans);
9189 trans->block_rsv = block_rsv;
9192 btrfs_release_path(path);
9196 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9198 btrfs_abort_transaction(trans, ret);
9202 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9203 ret = btrfs_find_root(tree_root, &root->root_key, path,
9206 btrfs_abort_transaction(trans, ret);
9209 } else if (ret > 0) {
9210 /* if we fail to delete the orphan item this time
9211 * around, it'll get picked up the next time.
9213 * The most common failure here is just -ENOENT.
9215 btrfs_del_orphan_item(trans, tree_root,
9216 root->root_key.objectid);
9220 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9221 btrfs_add_dropped_root(trans, root);
9223 free_extent_buffer(root->node);
9224 free_extent_buffer(root->commit_root);
9225 btrfs_put_fs_root(root);
9227 root_dropped = true;
9229 btrfs_end_transaction_throttle(trans);
9232 btrfs_free_path(path);
9235 * So if we need to stop dropping the snapshot for whatever reason we
9236 * need to make sure to add it back to the dead root list so that we
9237 * keep trying to do the work later. This also cleans up roots if we
9238 * don't have it in the radix (like when we recover after a power fail
9239 * or unmount) so we don't leak memory.
9241 if (!for_reloc && root_dropped == false)
9242 btrfs_add_dead_root(root);
9243 if (err && err != -EAGAIN)
9244 btrfs_handle_fs_error(fs_info, err, NULL);
9249 * drop subtree rooted at tree block 'node'.
9251 * NOTE: this function will unlock and release tree block 'node'
9252 * only used by relocation code
9254 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9255 struct btrfs_root *root,
9256 struct extent_buffer *node,
9257 struct extent_buffer *parent)
9259 struct btrfs_fs_info *fs_info = root->fs_info;
9260 struct btrfs_path *path;
9261 struct walk_control *wc;
9267 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9269 path = btrfs_alloc_path();
9273 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9275 btrfs_free_path(path);
9279 btrfs_assert_tree_locked(parent);
9280 parent_level = btrfs_header_level(parent);
9281 extent_buffer_get(parent);
9282 path->nodes[parent_level] = parent;
9283 path->slots[parent_level] = btrfs_header_nritems(parent);
9285 btrfs_assert_tree_locked(node);
9286 level = btrfs_header_level(node);
9287 path->nodes[level] = node;
9288 path->slots[level] = 0;
9289 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9291 wc->refs[parent_level] = 1;
9292 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9294 wc->shared_level = -1;
9295 wc->stage = DROP_REFERENCE;
9299 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9302 wret = walk_down_tree(trans, root, path, wc);
9308 wret = walk_up_tree(trans, root, path, wc, parent_level);
9316 btrfs_free_path(path);
9320 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9326 * if restripe for this chunk_type is on pick target profile and
9327 * return, otherwise do the usual balance
9329 stripped = get_restripe_target(fs_info, flags);
9331 return extended_to_chunk(stripped);
9333 num_devices = fs_info->fs_devices->rw_devices;
9335 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9336 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9337 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9339 if (num_devices == 1) {
9340 stripped |= BTRFS_BLOCK_GROUP_DUP;
9341 stripped = flags & ~stripped;
9343 /* turn raid0 into single device chunks */
9344 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9347 /* turn mirroring into duplication */
9348 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9349 BTRFS_BLOCK_GROUP_RAID10))
9350 return stripped | BTRFS_BLOCK_GROUP_DUP;
9352 /* they already had raid on here, just return */
9353 if (flags & stripped)
9356 stripped |= BTRFS_BLOCK_GROUP_DUP;
9357 stripped = flags & ~stripped;
9359 /* switch duplicated blocks with raid1 */
9360 if (flags & BTRFS_BLOCK_GROUP_DUP)
9361 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9363 /* this is drive concat, leave it alone */
9369 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9371 struct btrfs_space_info *sinfo = cache->space_info;
9373 u64 min_allocable_bytes;
9377 * We need some metadata space and system metadata space for
9378 * allocating chunks in some corner cases until we force to set
9379 * it to be readonly.
9382 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9384 min_allocable_bytes = SZ_1M;
9386 min_allocable_bytes = 0;
9388 spin_lock(&sinfo->lock);
9389 spin_lock(&cache->lock);
9397 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9398 cache->bytes_super - btrfs_block_group_used(&cache->item);
9400 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9401 min_allocable_bytes <= sinfo->total_bytes) {
9402 sinfo->bytes_readonly += num_bytes;
9404 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9408 spin_unlock(&cache->lock);
9409 spin_unlock(&sinfo->lock);
9413 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9414 struct btrfs_block_group_cache *cache)
9417 struct btrfs_trans_handle *trans;
9422 trans = btrfs_join_transaction(fs_info->extent_root);
9424 return PTR_ERR(trans);
9427 * we're not allowed to set block groups readonly after the dirty
9428 * block groups cache has started writing. If it already started,
9429 * back off and let this transaction commit
9431 mutex_lock(&fs_info->ro_block_group_mutex);
9432 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9433 u64 transid = trans->transid;
9435 mutex_unlock(&fs_info->ro_block_group_mutex);
9436 btrfs_end_transaction(trans);
9438 ret = btrfs_wait_for_commit(fs_info, transid);
9445 * if we are changing raid levels, try to allocate a corresponding
9446 * block group with the new raid level.
9448 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9449 if (alloc_flags != cache->flags) {
9450 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9453 * ENOSPC is allowed here, we may have enough space
9454 * already allocated at the new raid level to
9463 ret = inc_block_group_ro(cache, 0);
9466 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9467 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9471 ret = inc_block_group_ro(cache, 0);
9473 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9474 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9475 mutex_lock(&fs_info->chunk_mutex);
9476 check_system_chunk(trans, fs_info, alloc_flags);
9477 mutex_unlock(&fs_info->chunk_mutex);
9479 mutex_unlock(&fs_info->ro_block_group_mutex);
9481 btrfs_end_transaction(trans);
9485 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9486 struct btrfs_fs_info *fs_info, u64 type)
9488 u64 alloc_flags = get_alloc_profile(fs_info, type);
9490 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9494 * helper to account the unused space of all the readonly block group in the
9495 * space_info. takes mirrors into account.
9497 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9499 struct btrfs_block_group_cache *block_group;
9503 /* It's df, we don't care if it's racy */
9504 if (list_empty(&sinfo->ro_bgs))
9507 spin_lock(&sinfo->lock);
9508 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9509 spin_lock(&block_group->lock);
9511 if (!block_group->ro) {
9512 spin_unlock(&block_group->lock);
9516 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9517 BTRFS_BLOCK_GROUP_RAID10 |
9518 BTRFS_BLOCK_GROUP_DUP))
9523 free_bytes += (block_group->key.offset -
9524 btrfs_block_group_used(&block_group->item)) *
9527 spin_unlock(&block_group->lock);
9529 spin_unlock(&sinfo->lock);
9534 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9536 struct btrfs_space_info *sinfo = cache->space_info;
9541 spin_lock(&sinfo->lock);
9542 spin_lock(&cache->lock);
9544 num_bytes = cache->key.offset - cache->reserved -
9545 cache->pinned - cache->bytes_super -
9546 btrfs_block_group_used(&cache->item);
9547 sinfo->bytes_readonly -= num_bytes;
9548 list_del_init(&cache->ro_list);
9550 spin_unlock(&cache->lock);
9551 spin_unlock(&sinfo->lock);
9555 * checks to see if its even possible to relocate this block group.
9557 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9558 * ok to go ahead and try.
9560 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9562 struct btrfs_root *root = fs_info->extent_root;
9563 struct btrfs_block_group_cache *block_group;
9564 struct btrfs_space_info *space_info;
9565 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9566 struct btrfs_device *device;
9567 struct btrfs_trans_handle *trans;
9577 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9579 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9581 /* odd, couldn't find the block group, leave it alone */
9585 "can't find block group for bytenr %llu",
9590 min_free = btrfs_block_group_used(&block_group->item);
9592 /* no bytes used, we're good */
9596 space_info = block_group->space_info;
9597 spin_lock(&space_info->lock);
9599 full = space_info->full;
9602 * if this is the last block group we have in this space, we can't
9603 * relocate it unless we're able to allocate a new chunk below.
9605 * Otherwise, we need to make sure we have room in the space to handle
9606 * all of the extents from this block group. If we can, we're good
9608 if ((space_info->total_bytes != block_group->key.offset) &&
9609 (btrfs_space_info_used(space_info, false) + min_free <
9610 space_info->total_bytes)) {
9611 spin_unlock(&space_info->lock);
9614 spin_unlock(&space_info->lock);
9617 * ok we don't have enough space, but maybe we have free space on our
9618 * devices to allocate new chunks for relocation, so loop through our
9619 * alloc devices and guess if we have enough space. if this block
9620 * group is going to be restriped, run checks against the target
9621 * profile instead of the current one.
9633 target = get_restripe_target(fs_info, block_group->flags);
9635 index = __get_raid_index(extended_to_chunk(target));
9638 * this is just a balance, so if we were marked as full
9639 * we know there is no space for a new chunk
9644 "no space to alloc new chunk for block group %llu",
9645 block_group->key.objectid);
9649 index = get_block_group_index(block_group);
9652 if (index == BTRFS_RAID_RAID10) {
9656 } else if (index == BTRFS_RAID_RAID1) {
9658 } else if (index == BTRFS_RAID_DUP) {
9661 } else if (index == BTRFS_RAID_RAID0) {
9662 dev_min = fs_devices->rw_devices;
9663 min_free = div64_u64(min_free, dev_min);
9666 /* We need to do this so that we can look at pending chunks */
9667 trans = btrfs_join_transaction(root);
9668 if (IS_ERR(trans)) {
9669 ret = PTR_ERR(trans);
9673 mutex_lock(&fs_info->chunk_mutex);
9674 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9678 * check to make sure we can actually find a chunk with enough
9679 * space to fit our block group in.
9681 if (device->total_bytes > device->bytes_used + min_free &&
9682 !device->is_tgtdev_for_dev_replace) {
9683 ret = find_free_dev_extent(trans, device, min_free,
9688 if (dev_nr >= dev_min)
9694 if (debug && ret == -1)
9696 "no space to allocate a new chunk for block group %llu",
9697 block_group->key.objectid);
9698 mutex_unlock(&fs_info->chunk_mutex);
9699 btrfs_end_transaction(trans);
9701 btrfs_put_block_group(block_group);
9705 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9706 struct btrfs_path *path,
9707 struct btrfs_key *key)
9709 struct btrfs_root *root = fs_info->extent_root;
9711 struct btrfs_key found_key;
9712 struct extent_buffer *leaf;
9715 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9720 slot = path->slots[0];
9721 leaf = path->nodes[0];
9722 if (slot >= btrfs_header_nritems(leaf)) {
9723 ret = btrfs_next_leaf(root, path);
9730 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9732 if (found_key.objectid >= key->objectid &&
9733 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9734 struct extent_map_tree *em_tree;
9735 struct extent_map *em;
9737 em_tree = &root->fs_info->mapping_tree.map_tree;
9738 read_lock(&em_tree->lock);
9739 em = lookup_extent_mapping(em_tree, found_key.objectid,
9741 read_unlock(&em_tree->lock);
9744 "logical %llu len %llu found bg but no related chunk",
9745 found_key.objectid, found_key.offset);
9750 free_extent_map(em);
9759 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9761 struct btrfs_block_group_cache *block_group;
9765 struct inode *inode;
9767 block_group = btrfs_lookup_first_block_group(info, last);
9768 while (block_group) {
9769 spin_lock(&block_group->lock);
9770 if (block_group->iref)
9772 spin_unlock(&block_group->lock);
9773 block_group = next_block_group(info, block_group);
9782 inode = block_group->inode;
9783 block_group->iref = 0;
9784 block_group->inode = NULL;
9785 spin_unlock(&block_group->lock);
9786 ASSERT(block_group->io_ctl.inode == NULL);
9788 last = block_group->key.objectid + block_group->key.offset;
9789 btrfs_put_block_group(block_group);
9794 * Must be called only after stopping all workers, since we could have block
9795 * group caching kthreads running, and therefore they could race with us if we
9796 * freed the block groups before stopping them.
9798 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9800 struct btrfs_block_group_cache *block_group;
9801 struct btrfs_space_info *space_info;
9802 struct btrfs_caching_control *caching_ctl;
9805 down_write(&info->commit_root_sem);
9806 while (!list_empty(&info->caching_block_groups)) {
9807 caching_ctl = list_entry(info->caching_block_groups.next,
9808 struct btrfs_caching_control, list);
9809 list_del(&caching_ctl->list);
9810 put_caching_control(caching_ctl);
9812 up_write(&info->commit_root_sem);
9814 spin_lock(&info->unused_bgs_lock);
9815 while (!list_empty(&info->unused_bgs)) {
9816 block_group = list_first_entry(&info->unused_bgs,
9817 struct btrfs_block_group_cache,
9819 list_del_init(&block_group->bg_list);
9820 btrfs_put_block_group(block_group);
9822 spin_unlock(&info->unused_bgs_lock);
9824 spin_lock(&info->block_group_cache_lock);
9825 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9826 block_group = rb_entry(n, struct btrfs_block_group_cache,
9828 rb_erase(&block_group->cache_node,
9829 &info->block_group_cache_tree);
9830 RB_CLEAR_NODE(&block_group->cache_node);
9831 spin_unlock(&info->block_group_cache_lock);
9833 down_write(&block_group->space_info->groups_sem);
9834 list_del(&block_group->list);
9835 up_write(&block_group->space_info->groups_sem);
9838 * We haven't cached this block group, which means we could
9839 * possibly have excluded extents on this block group.
9841 if (block_group->cached == BTRFS_CACHE_NO ||
9842 block_group->cached == BTRFS_CACHE_ERROR)
9843 free_excluded_extents(info, block_group);
9845 btrfs_remove_free_space_cache(block_group);
9846 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9847 ASSERT(list_empty(&block_group->dirty_list));
9848 ASSERT(list_empty(&block_group->io_list));
9849 ASSERT(list_empty(&block_group->bg_list));
9850 ASSERT(atomic_read(&block_group->count) == 1);
9851 btrfs_put_block_group(block_group);
9853 spin_lock(&info->block_group_cache_lock);
9855 spin_unlock(&info->block_group_cache_lock);
9857 /* now that all the block groups are freed, go through and
9858 * free all the space_info structs. This is only called during
9859 * the final stages of unmount, and so we know nobody is
9860 * using them. We call synchronize_rcu() once before we start,
9861 * just to be on the safe side.
9865 release_global_block_rsv(info);
9867 while (!list_empty(&info->space_info)) {
9870 space_info = list_entry(info->space_info.next,
9871 struct btrfs_space_info,
9875 * Do not hide this behind enospc_debug, this is actually
9876 * important and indicates a real bug if this happens.
9878 if (WARN_ON(space_info->bytes_pinned > 0 ||
9879 space_info->bytes_reserved > 0 ||
9880 space_info->bytes_may_use > 0))
9881 dump_space_info(info, space_info, 0, 0);
9882 list_del(&space_info->list);
9883 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9884 struct kobject *kobj;
9885 kobj = space_info->block_group_kobjs[i];
9886 space_info->block_group_kobjs[i] = NULL;
9892 kobject_del(&space_info->kobj);
9893 kobject_put(&space_info->kobj);
9898 static void __link_block_group(struct btrfs_space_info *space_info,
9899 struct btrfs_block_group_cache *cache)
9901 int index = get_block_group_index(cache);
9904 down_write(&space_info->groups_sem);
9905 if (list_empty(&space_info->block_groups[index]))
9907 list_add_tail(&cache->list, &space_info->block_groups[index]);
9908 up_write(&space_info->groups_sem);
9911 struct raid_kobject *rkobj;
9914 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9917 rkobj->raid_type = index;
9918 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9919 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9920 "%s", get_raid_name(index));
9922 kobject_put(&rkobj->kobj);
9925 space_info->block_group_kobjs[index] = &rkobj->kobj;
9930 btrfs_warn(cache->fs_info,
9931 "failed to add kobject for block cache, ignoring");
9934 static struct btrfs_block_group_cache *
9935 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9936 u64 start, u64 size)
9938 struct btrfs_block_group_cache *cache;
9940 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9944 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9946 if (!cache->free_space_ctl) {
9951 cache->key.objectid = start;
9952 cache->key.offset = size;
9953 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9955 cache->sectorsize = fs_info->sectorsize;
9956 cache->fs_info = fs_info;
9957 cache->full_stripe_len = btrfs_full_stripe_len(fs_info,
9958 &fs_info->mapping_tree,
9960 set_free_space_tree_thresholds(cache);
9962 atomic_set(&cache->count, 1);
9963 spin_lock_init(&cache->lock);
9964 init_rwsem(&cache->data_rwsem);
9965 INIT_LIST_HEAD(&cache->list);
9966 INIT_LIST_HEAD(&cache->cluster_list);
9967 INIT_LIST_HEAD(&cache->bg_list);
9968 INIT_LIST_HEAD(&cache->ro_list);
9969 INIT_LIST_HEAD(&cache->dirty_list);
9970 INIT_LIST_HEAD(&cache->io_list);
9971 btrfs_init_free_space_ctl(cache);
9972 atomic_set(&cache->trimming, 0);
9973 mutex_init(&cache->free_space_lock);
9974 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
9979 int btrfs_read_block_groups(struct btrfs_fs_info *info)
9981 struct btrfs_path *path;
9983 struct btrfs_block_group_cache *cache;
9984 struct btrfs_space_info *space_info;
9985 struct btrfs_key key;
9986 struct btrfs_key found_key;
9987 struct extent_buffer *leaf;
9993 feature = btrfs_super_incompat_flags(info->super_copy);
9994 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
9998 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9999 path = btrfs_alloc_path();
10002 path->reada = READA_FORWARD;
10004 cache_gen = btrfs_super_cache_generation(info->super_copy);
10005 if (btrfs_test_opt(info, SPACE_CACHE) &&
10006 btrfs_super_generation(info->super_copy) != cache_gen)
10008 if (btrfs_test_opt(info, CLEAR_CACHE))
10012 ret = find_first_block_group(info, path, &key);
10018 leaf = path->nodes[0];
10019 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10021 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10030 * When we mount with old space cache, we need to
10031 * set BTRFS_DC_CLEAR and set dirty flag.
10033 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10034 * truncate the old free space cache inode and
10036 * b) Setting 'dirty flag' makes sure that we flush
10037 * the new space cache info onto disk.
10039 if (btrfs_test_opt(info, SPACE_CACHE))
10040 cache->disk_cache_state = BTRFS_DC_CLEAR;
10043 read_extent_buffer(leaf, &cache->item,
10044 btrfs_item_ptr_offset(leaf, path->slots[0]),
10045 sizeof(cache->item));
10046 cache->flags = btrfs_block_group_flags(&cache->item);
10048 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10049 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10051 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10052 cache->key.objectid);
10057 key.objectid = found_key.objectid + found_key.offset;
10058 btrfs_release_path(path);
10061 * We need to exclude the super stripes now so that the space
10062 * info has super bytes accounted for, otherwise we'll think
10063 * we have more space than we actually do.
10065 ret = exclude_super_stripes(info, cache);
10068 * We may have excluded something, so call this just in
10071 free_excluded_extents(info, cache);
10072 btrfs_put_block_group(cache);
10077 * check for two cases, either we are full, and therefore
10078 * don't need to bother with the caching work since we won't
10079 * find any space, or we are empty, and we can just add all
10080 * the space in and be done with it. This saves us _alot_ of
10081 * time, particularly in the full case.
10083 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10084 cache->last_byte_to_unpin = (u64)-1;
10085 cache->cached = BTRFS_CACHE_FINISHED;
10086 free_excluded_extents(info, cache);
10087 } else if (btrfs_block_group_used(&cache->item) == 0) {
10088 cache->last_byte_to_unpin = (u64)-1;
10089 cache->cached = BTRFS_CACHE_FINISHED;
10090 add_new_free_space(cache, info,
10091 found_key.objectid,
10092 found_key.objectid +
10094 free_excluded_extents(info, cache);
10097 ret = btrfs_add_block_group_cache(info, cache);
10099 btrfs_remove_free_space_cache(cache);
10100 btrfs_put_block_group(cache);
10104 trace_btrfs_add_block_group(info, cache, 0);
10105 update_space_info(info, cache->flags, found_key.offset,
10106 btrfs_block_group_used(&cache->item),
10107 cache->bytes_super, &space_info);
10109 cache->space_info = space_info;
10111 __link_block_group(space_info, cache);
10113 set_avail_alloc_bits(info, cache->flags);
10114 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10115 inc_block_group_ro(cache, 1);
10116 } else if (btrfs_block_group_used(&cache->item) == 0) {
10117 spin_lock(&info->unused_bgs_lock);
10118 /* Should always be true but just in case. */
10119 if (list_empty(&cache->bg_list)) {
10120 btrfs_get_block_group(cache);
10121 list_add_tail(&cache->bg_list,
10122 &info->unused_bgs);
10124 spin_unlock(&info->unused_bgs_lock);
10128 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10129 if (!(get_alloc_profile(info, space_info->flags) &
10130 (BTRFS_BLOCK_GROUP_RAID10 |
10131 BTRFS_BLOCK_GROUP_RAID1 |
10132 BTRFS_BLOCK_GROUP_RAID5 |
10133 BTRFS_BLOCK_GROUP_RAID6 |
10134 BTRFS_BLOCK_GROUP_DUP)))
10137 * avoid allocating from un-mirrored block group if there are
10138 * mirrored block groups.
10140 list_for_each_entry(cache,
10141 &space_info->block_groups[BTRFS_RAID_RAID0],
10143 inc_block_group_ro(cache, 1);
10144 list_for_each_entry(cache,
10145 &space_info->block_groups[BTRFS_RAID_SINGLE],
10147 inc_block_group_ro(cache, 1);
10150 init_global_block_rsv(info);
10153 btrfs_free_path(path);
10157 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10158 struct btrfs_fs_info *fs_info)
10160 struct btrfs_block_group_cache *block_group, *tmp;
10161 struct btrfs_root *extent_root = fs_info->extent_root;
10162 struct btrfs_block_group_item item;
10163 struct btrfs_key key;
10165 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10167 trans->can_flush_pending_bgs = false;
10168 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10172 spin_lock(&block_group->lock);
10173 memcpy(&item, &block_group->item, sizeof(item));
10174 memcpy(&key, &block_group->key, sizeof(key));
10175 spin_unlock(&block_group->lock);
10177 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10180 btrfs_abort_transaction(trans, ret);
10181 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10184 btrfs_abort_transaction(trans, ret);
10185 add_block_group_free_space(trans, fs_info, block_group);
10186 /* already aborted the transaction if it failed. */
10188 list_del_init(&block_group->bg_list);
10190 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10193 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10194 struct btrfs_fs_info *fs_info, u64 bytes_used,
10195 u64 type, u64 chunk_objectid, u64 chunk_offset,
10198 struct btrfs_block_group_cache *cache;
10201 btrfs_set_log_full_commit(fs_info, trans);
10203 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10207 btrfs_set_block_group_used(&cache->item, bytes_used);
10208 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10209 btrfs_set_block_group_flags(&cache->item, type);
10211 cache->flags = type;
10212 cache->last_byte_to_unpin = (u64)-1;
10213 cache->cached = BTRFS_CACHE_FINISHED;
10214 cache->needs_free_space = 1;
10215 ret = exclude_super_stripes(fs_info, cache);
10218 * We may have excluded something, so call this just in
10221 free_excluded_extents(fs_info, cache);
10222 btrfs_put_block_group(cache);
10226 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10228 free_excluded_extents(fs_info, cache);
10230 #ifdef CONFIG_BTRFS_DEBUG
10231 if (btrfs_should_fragment_free_space(cache)) {
10232 u64 new_bytes_used = size - bytes_used;
10234 bytes_used += new_bytes_used >> 1;
10235 fragment_free_space(cache);
10239 * Ensure the corresponding space_info object is created and
10240 * assigned to our block group. We want our bg to be added to the rbtree
10241 * with its ->space_info set.
10243 cache->space_info = __find_space_info(fs_info, cache->flags);
10244 if (!cache->space_info) {
10245 ret = create_space_info(fs_info, cache->flags,
10246 &cache->space_info);
10248 btrfs_remove_free_space_cache(cache);
10249 btrfs_put_block_group(cache);
10254 ret = btrfs_add_block_group_cache(fs_info, cache);
10256 btrfs_remove_free_space_cache(cache);
10257 btrfs_put_block_group(cache);
10262 * Now that our block group has its ->space_info set and is inserted in
10263 * the rbtree, update the space info's counters.
10265 trace_btrfs_add_block_group(fs_info, cache, 1);
10266 update_space_info(fs_info, cache->flags, size, bytes_used,
10267 cache->bytes_super, &cache->space_info);
10268 update_global_block_rsv(fs_info);
10270 __link_block_group(cache->space_info, cache);
10272 list_add_tail(&cache->bg_list, &trans->new_bgs);
10274 set_avail_alloc_bits(fs_info, type);
10278 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10280 u64 extra_flags = chunk_to_extended(flags) &
10281 BTRFS_EXTENDED_PROFILE_MASK;
10283 write_seqlock(&fs_info->profiles_lock);
10284 if (flags & BTRFS_BLOCK_GROUP_DATA)
10285 fs_info->avail_data_alloc_bits &= ~extra_flags;
10286 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10287 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10288 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10289 fs_info->avail_system_alloc_bits &= ~extra_flags;
10290 write_sequnlock(&fs_info->profiles_lock);
10293 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10294 struct btrfs_fs_info *fs_info, u64 group_start,
10295 struct extent_map *em)
10297 struct btrfs_root *root = fs_info->extent_root;
10298 struct btrfs_path *path;
10299 struct btrfs_block_group_cache *block_group;
10300 struct btrfs_free_cluster *cluster;
10301 struct btrfs_root *tree_root = fs_info->tree_root;
10302 struct btrfs_key key;
10303 struct inode *inode;
10304 struct kobject *kobj = NULL;
10308 struct btrfs_caching_control *caching_ctl = NULL;
10311 block_group = btrfs_lookup_block_group(fs_info, group_start);
10312 BUG_ON(!block_group);
10313 BUG_ON(!block_group->ro);
10316 * Free the reserved super bytes from this block group before
10319 free_excluded_extents(fs_info, block_group);
10321 memcpy(&key, &block_group->key, sizeof(key));
10322 index = get_block_group_index(block_group);
10323 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10324 BTRFS_BLOCK_GROUP_RAID1 |
10325 BTRFS_BLOCK_GROUP_RAID10))
10330 /* make sure this block group isn't part of an allocation cluster */
10331 cluster = &fs_info->data_alloc_cluster;
10332 spin_lock(&cluster->refill_lock);
10333 btrfs_return_cluster_to_free_space(block_group, cluster);
10334 spin_unlock(&cluster->refill_lock);
10337 * make sure this block group isn't part of a metadata
10338 * allocation cluster
10340 cluster = &fs_info->meta_alloc_cluster;
10341 spin_lock(&cluster->refill_lock);
10342 btrfs_return_cluster_to_free_space(block_group, cluster);
10343 spin_unlock(&cluster->refill_lock);
10345 path = btrfs_alloc_path();
10352 * get the inode first so any iput calls done for the io_list
10353 * aren't the final iput (no unlinks allowed now)
10355 inode = lookup_free_space_inode(fs_info, block_group, path);
10357 mutex_lock(&trans->transaction->cache_write_mutex);
10359 * make sure our free spache cache IO is done before remove the
10362 spin_lock(&trans->transaction->dirty_bgs_lock);
10363 if (!list_empty(&block_group->io_list)) {
10364 list_del_init(&block_group->io_list);
10366 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10368 spin_unlock(&trans->transaction->dirty_bgs_lock);
10369 btrfs_wait_cache_io(trans, block_group, path);
10370 btrfs_put_block_group(block_group);
10371 spin_lock(&trans->transaction->dirty_bgs_lock);
10374 if (!list_empty(&block_group->dirty_list)) {
10375 list_del_init(&block_group->dirty_list);
10376 btrfs_put_block_group(block_group);
10378 spin_unlock(&trans->transaction->dirty_bgs_lock);
10379 mutex_unlock(&trans->transaction->cache_write_mutex);
10381 if (!IS_ERR(inode)) {
10382 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10384 btrfs_add_delayed_iput(inode);
10387 clear_nlink(inode);
10388 /* One for the block groups ref */
10389 spin_lock(&block_group->lock);
10390 if (block_group->iref) {
10391 block_group->iref = 0;
10392 block_group->inode = NULL;
10393 spin_unlock(&block_group->lock);
10396 spin_unlock(&block_group->lock);
10398 /* One for our lookup ref */
10399 btrfs_add_delayed_iput(inode);
10402 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10403 key.offset = block_group->key.objectid;
10406 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10410 btrfs_release_path(path);
10412 ret = btrfs_del_item(trans, tree_root, path);
10415 btrfs_release_path(path);
10418 spin_lock(&fs_info->block_group_cache_lock);
10419 rb_erase(&block_group->cache_node,
10420 &fs_info->block_group_cache_tree);
10421 RB_CLEAR_NODE(&block_group->cache_node);
10423 if (fs_info->first_logical_byte == block_group->key.objectid)
10424 fs_info->first_logical_byte = (u64)-1;
10425 spin_unlock(&fs_info->block_group_cache_lock);
10427 down_write(&block_group->space_info->groups_sem);
10429 * we must use list_del_init so people can check to see if they
10430 * are still on the list after taking the semaphore
10432 list_del_init(&block_group->list);
10433 if (list_empty(&block_group->space_info->block_groups[index])) {
10434 kobj = block_group->space_info->block_group_kobjs[index];
10435 block_group->space_info->block_group_kobjs[index] = NULL;
10436 clear_avail_alloc_bits(fs_info, block_group->flags);
10438 up_write(&block_group->space_info->groups_sem);
10444 if (block_group->has_caching_ctl)
10445 caching_ctl = get_caching_control(block_group);
10446 if (block_group->cached == BTRFS_CACHE_STARTED)
10447 wait_block_group_cache_done(block_group);
10448 if (block_group->has_caching_ctl) {
10449 down_write(&fs_info->commit_root_sem);
10450 if (!caching_ctl) {
10451 struct btrfs_caching_control *ctl;
10453 list_for_each_entry(ctl,
10454 &fs_info->caching_block_groups, list)
10455 if (ctl->block_group == block_group) {
10457 refcount_inc(&caching_ctl->count);
10462 list_del_init(&caching_ctl->list);
10463 up_write(&fs_info->commit_root_sem);
10465 /* Once for the caching bgs list and once for us. */
10466 put_caching_control(caching_ctl);
10467 put_caching_control(caching_ctl);
10471 spin_lock(&trans->transaction->dirty_bgs_lock);
10472 if (!list_empty(&block_group->dirty_list)) {
10475 if (!list_empty(&block_group->io_list)) {
10478 spin_unlock(&trans->transaction->dirty_bgs_lock);
10479 btrfs_remove_free_space_cache(block_group);
10481 spin_lock(&block_group->space_info->lock);
10482 list_del_init(&block_group->ro_list);
10484 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10485 WARN_ON(block_group->space_info->total_bytes
10486 < block_group->key.offset);
10487 WARN_ON(block_group->space_info->bytes_readonly
10488 < block_group->key.offset);
10489 WARN_ON(block_group->space_info->disk_total
10490 < block_group->key.offset * factor);
10492 block_group->space_info->total_bytes -= block_group->key.offset;
10493 block_group->space_info->bytes_readonly -= block_group->key.offset;
10494 block_group->space_info->disk_total -= block_group->key.offset * factor;
10496 spin_unlock(&block_group->space_info->lock);
10498 memcpy(&key, &block_group->key, sizeof(key));
10500 mutex_lock(&fs_info->chunk_mutex);
10501 if (!list_empty(&em->list)) {
10502 /* We're in the transaction->pending_chunks list. */
10503 free_extent_map(em);
10505 spin_lock(&block_group->lock);
10506 block_group->removed = 1;
10508 * At this point trimming can't start on this block group, because we
10509 * removed the block group from the tree fs_info->block_group_cache_tree
10510 * so no one can't find it anymore and even if someone already got this
10511 * block group before we removed it from the rbtree, they have already
10512 * incremented block_group->trimming - if they didn't, they won't find
10513 * any free space entries because we already removed them all when we
10514 * called btrfs_remove_free_space_cache().
10516 * And we must not remove the extent map from the fs_info->mapping_tree
10517 * to prevent the same logical address range and physical device space
10518 * ranges from being reused for a new block group. This is because our
10519 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10520 * completely transactionless, so while it is trimming a range the
10521 * currently running transaction might finish and a new one start,
10522 * allowing for new block groups to be created that can reuse the same
10523 * physical device locations unless we take this special care.
10525 * There may also be an implicit trim operation if the file system
10526 * is mounted with -odiscard. The same protections must remain
10527 * in place until the extents have been discarded completely when
10528 * the transaction commit has completed.
10530 remove_em = (atomic_read(&block_group->trimming) == 0);
10532 * Make sure a trimmer task always sees the em in the pinned_chunks list
10533 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10534 * before checking block_group->removed).
10538 * Our em might be in trans->transaction->pending_chunks which
10539 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10540 * and so is the fs_info->pinned_chunks list.
10542 * So at this point we must be holding the chunk_mutex to avoid
10543 * any races with chunk allocation (more specifically at
10544 * volumes.c:contains_pending_extent()), to ensure it always
10545 * sees the em, either in the pending_chunks list or in the
10546 * pinned_chunks list.
10548 list_move_tail(&em->list, &fs_info->pinned_chunks);
10550 spin_unlock(&block_group->lock);
10553 struct extent_map_tree *em_tree;
10555 em_tree = &fs_info->mapping_tree.map_tree;
10556 write_lock(&em_tree->lock);
10558 * The em might be in the pending_chunks list, so make sure the
10559 * chunk mutex is locked, since remove_extent_mapping() will
10560 * delete us from that list.
10562 remove_extent_mapping(em_tree, em);
10563 write_unlock(&em_tree->lock);
10564 /* once for the tree */
10565 free_extent_map(em);
10568 mutex_unlock(&fs_info->chunk_mutex);
10570 ret = remove_block_group_free_space(trans, fs_info, block_group);
10574 btrfs_put_block_group(block_group);
10575 btrfs_put_block_group(block_group);
10577 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10583 ret = btrfs_del_item(trans, root, path);
10585 btrfs_free_path(path);
10589 struct btrfs_trans_handle *
10590 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10591 const u64 chunk_offset)
10593 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10594 struct extent_map *em;
10595 struct map_lookup *map;
10596 unsigned int num_items;
10598 read_lock(&em_tree->lock);
10599 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10600 read_unlock(&em_tree->lock);
10601 ASSERT(em && em->start == chunk_offset);
10604 * We need to reserve 3 + N units from the metadata space info in order
10605 * to remove a block group (done at btrfs_remove_chunk() and at
10606 * btrfs_remove_block_group()), which are used for:
10608 * 1 unit for adding the free space inode's orphan (located in the tree
10610 * 1 unit for deleting the block group item (located in the extent
10612 * 1 unit for deleting the free space item (located in tree of tree
10614 * N units for deleting N device extent items corresponding to each
10615 * stripe (located in the device tree).
10617 * In order to remove a block group we also need to reserve units in the
10618 * system space info in order to update the chunk tree (update one or
10619 * more device items and remove one chunk item), but this is done at
10620 * btrfs_remove_chunk() through a call to check_system_chunk().
10622 map = em->map_lookup;
10623 num_items = 3 + map->num_stripes;
10624 free_extent_map(em);
10626 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10631 * Process the unused_bgs list and remove any that don't have any allocated
10632 * space inside of them.
10634 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10636 struct btrfs_block_group_cache *block_group;
10637 struct btrfs_space_info *space_info;
10638 struct btrfs_trans_handle *trans;
10641 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10644 spin_lock(&fs_info->unused_bgs_lock);
10645 while (!list_empty(&fs_info->unused_bgs)) {
10649 block_group = list_first_entry(&fs_info->unused_bgs,
10650 struct btrfs_block_group_cache,
10652 list_del_init(&block_group->bg_list);
10654 space_info = block_group->space_info;
10656 if (ret || btrfs_mixed_space_info(space_info)) {
10657 btrfs_put_block_group(block_group);
10660 spin_unlock(&fs_info->unused_bgs_lock);
10662 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10664 /* Don't want to race with allocators so take the groups_sem */
10665 down_write(&space_info->groups_sem);
10666 spin_lock(&block_group->lock);
10667 if (block_group->reserved ||
10668 btrfs_block_group_used(&block_group->item) ||
10670 list_is_singular(&block_group->list)) {
10672 * We want to bail if we made new allocations or have
10673 * outstanding allocations in this block group. We do
10674 * the ro check in case balance is currently acting on
10675 * this block group.
10677 spin_unlock(&block_group->lock);
10678 up_write(&space_info->groups_sem);
10681 spin_unlock(&block_group->lock);
10683 /* We don't want to force the issue, only flip if it's ok. */
10684 ret = inc_block_group_ro(block_group, 0);
10685 up_write(&space_info->groups_sem);
10692 * Want to do this before we do anything else so we can recover
10693 * properly if we fail to join the transaction.
10695 trans = btrfs_start_trans_remove_block_group(fs_info,
10696 block_group->key.objectid);
10697 if (IS_ERR(trans)) {
10698 btrfs_dec_block_group_ro(block_group);
10699 ret = PTR_ERR(trans);
10704 * We could have pending pinned extents for this block group,
10705 * just delete them, we don't care about them anymore.
10707 start = block_group->key.objectid;
10708 end = start + block_group->key.offset - 1;
10710 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10711 * btrfs_finish_extent_commit(). If we are at transaction N,
10712 * another task might be running finish_extent_commit() for the
10713 * previous transaction N - 1, and have seen a range belonging
10714 * to the block group in freed_extents[] before we were able to
10715 * clear the whole block group range from freed_extents[]. This
10716 * means that task can lookup for the block group after we
10717 * unpinned it from freed_extents[] and removed it, leading to
10718 * a BUG_ON() at btrfs_unpin_extent_range().
10720 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10721 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10724 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10725 btrfs_dec_block_group_ro(block_group);
10728 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10731 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10732 btrfs_dec_block_group_ro(block_group);
10735 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10737 /* Reset pinned so btrfs_put_block_group doesn't complain */
10738 spin_lock(&space_info->lock);
10739 spin_lock(&block_group->lock);
10741 space_info->bytes_pinned -= block_group->pinned;
10742 space_info->bytes_readonly += block_group->pinned;
10743 percpu_counter_add(&space_info->total_bytes_pinned,
10744 -block_group->pinned);
10745 block_group->pinned = 0;
10747 spin_unlock(&block_group->lock);
10748 spin_unlock(&space_info->lock);
10750 /* DISCARD can flip during remount */
10751 trimming = btrfs_test_opt(fs_info, DISCARD);
10753 /* Implicit trim during transaction commit. */
10755 btrfs_get_block_group_trimming(block_group);
10758 * Btrfs_remove_chunk will abort the transaction if things go
10761 ret = btrfs_remove_chunk(trans, fs_info,
10762 block_group->key.objectid);
10766 btrfs_put_block_group_trimming(block_group);
10771 * If we're not mounted with -odiscard, we can just forget
10772 * about this block group. Otherwise we'll need to wait
10773 * until transaction commit to do the actual discard.
10776 spin_lock(&fs_info->unused_bgs_lock);
10778 * A concurrent scrub might have added us to the list
10779 * fs_info->unused_bgs, so use a list_move operation
10780 * to add the block group to the deleted_bgs list.
10782 list_move(&block_group->bg_list,
10783 &trans->transaction->deleted_bgs);
10784 spin_unlock(&fs_info->unused_bgs_lock);
10785 btrfs_get_block_group(block_group);
10788 btrfs_end_transaction(trans);
10790 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10791 btrfs_put_block_group(block_group);
10792 spin_lock(&fs_info->unused_bgs_lock);
10794 spin_unlock(&fs_info->unused_bgs_lock);
10797 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10799 struct btrfs_space_info *space_info;
10800 struct btrfs_super_block *disk_super;
10806 disk_super = fs_info->super_copy;
10807 if (!btrfs_super_root(disk_super))
10810 features = btrfs_super_incompat_flags(disk_super);
10811 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10814 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10815 ret = create_space_info(fs_info, flags, &space_info);
10820 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10821 ret = create_space_info(fs_info, flags, &space_info);
10823 flags = BTRFS_BLOCK_GROUP_METADATA;
10824 ret = create_space_info(fs_info, flags, &space_info);
10828 flags = BTRFS_BLOCK_GROUP_DATA;
10829 ret = create_space_info(fs_info, flags, &space_info);
10835 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10836 u64 start, u64 end)
10838 return unpin_extent_range(fs_info, start, end, false);
10842 * It used to be that old block groups would be left around forever.
10843 * Iterating over them would be enough to trim unused space. Since we
10844 * now automatically remove them, we also need to iterate over unallocated
10847 * We don't want a transaction for this since the discard may take a
10848 * substantial amount of time. We don't require that a transaction be
10849 * running, but we do need to take a running transaction into account
10850 * to ensure that we're not discarding chunks that were released in
10851 * the current transaction.
10853 * Holding the chunks lock will prevent other threads from allocating
10854 * or releasing chunks, but it won't prevent a running transaction
10855 * from committing and releasing the memory that the pending chunks
10856 * list head uses. For that, we need to take a reference to the
10859 static int btrfs_trim_free_extents(struct btrfs_device *device,
10860 u64 minlen, u64 *trimmed)
10862 u64 start = 0, len = 0;
10867 /* Not writeable = nothing to do. */
10868 if (!device->writeable)
10871 /* No free space = nothing to do. */
10872 if (device->total_bytes <= device->bytes_used)
10878 struct btrfs_fs_info *fs_info = device->fs_info;
10879 struct btrfs_transaction *trans;
10882 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10886 down_read(&fs_info->commit_root_sem);
10888 spin_lock(&fs_info->trans_lock);
10889 trans = fs_info->running_transaction;
10891 refcount_inc(&trans->use_count);
10892 spin_unlock(&fs_info->trans_lock);
10894 ret = find_free_dev_extent_start(trans, device, minlen, start,
10897 btrfs_put_transaction(trans);
10900 up_read(&fs_info->commit_root_sem);
10901 mutex_unlock(&fs_info->chunk_mutex);
10902 if (ret == -ENOSPC)
10907 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10908 up_read(&fs_info->commit_root_sem);
10909 mutex_unlock(&fs_info->chunk_mutex);
10917 if (fatal_signal_pending(current)) {
10918 ret = -ERESTARTSYS;
10928 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10930 struct btrfs_block_group_cache *cache = NULL;
10931 struct btrfs_device *device;
10932 struct list_head *devices;
10937 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10941 * try to trim all FS space, our block group may start from non-zero.
10943 if (range->len == total_bytes)
10944 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10946 cache = btrfs_lookup_block_group(fs_info, range->start);
10949 if (cache->key.objectid >= (range->start + range->len)) {
10950 btrfs_put_block_group(cache);
10954 start = max(range->start, cache->key.objectid);
10955 end = min(range->start + range->len,
10956 cache->key.objectid + cache->key.offset);
10958 if (end - start >= range->minlen) {
10959 if (!block_group_cache_done(cache)) {
10960 ret = cache_block_group(cache, 0);
10962 btrfs_put_block_group(cache);
10965 ret = wait_block_group_cache_done(cache);
10967 btrfs_put_block_group(cache);
10971 ret = btrfs_trim_block_group(cache,
10977 trimmed += group_trimmed;
10979 btrfs_put_block_group(cache);
10984 cache = next_block_group(fs_info, cache);
10987 mutex_lock(&fs_info->fs_devices->device_list_mutex);
10988 devices = &fs_info->fs_devices->alloc_list;
10989 list_for_each_entry(device, devices, dev_alloc_list) {
10990 ret = btrfs_trim_free_extents(device, range->minlen,
10995 trimmed += group_trimmed;
10997 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
10999 range->len = trimmed;
11004 * btrfs_{start,end}_write_no_snapshoting() are similar to
11005 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11006 * data into the page cache through nocow before the subvolume is snapshoted,
11007 * but flush the data into disk after the snapshot creation, or to prevent
11008 * operations while snapshoting is ongoing and that cause the snapshot to be
11009 * inconsistent (writes followed by expanding truncates for example).
11011 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
11013 percpu_counter_dec(&root->subv_writers->counter);
11015 * Make sure counter is updated before we wake up waiters.
11018 if (waitqueue_active(&root->subv_writers->wait))
11019 wake_up(&root->subv_writers->wait);
11022 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
11024 if (atomic_read(&root->will_be_snapshoted))
11027 percpu_counter_inc(&root->subv_writers->counter);
11029 * Make sure counter is updated before we check for snapshot creation.
11032 if (atomic_read(&root->will_be_snapshoted)) {
11033 btrfs_end_write_no_snapshoting(root);
11039 static int wait_snapshoting_atomic_t(atomic_t *a)
11045 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11050 ret = btrfs_start_write_no_snapshoting(root);
11053 wait_on_atomic_t(&root->will_be_snapshoted,
11054 wait_snapshoting_atomic_t,
11055 TASK_UNINTERRUPTIBLE);