2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/sched/signal.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/sort.h>
24 #include <linux/rcupdate.h>
25 #include <linux/kthread.h>
26 #include <linux/slab.h>
27 #include <linux/ratelimit.h>
28 #include <linux/percpu_counter.h>
32 #include "print-tree.h"
36 #include "free-space-cache.h"
37 #include "free-space-tree.h"
42 #undef SCRAMBLE_DELAYED_REFS
45 * control flags for do_chunk_alloc's force field
46 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
47 * if we really need one.
49 * CHUNK_ALLOC_LIMITED means to only try and allocate one
50 * if we have very few chunks already allocated. This is
51 * used as part of the clustering code to help make sure
52 * we have a good pool of storage to cluster in, without
53 * filling the FS with empty chunks
55 * CHUNK_ALLOC_FORCE means it must try to allocate one
59 CHUNK_ALLOC_NO_FORCE = 0,
60 CHUNK_ALLOC_LIMITED = 1,
61 CHUNK_ALLOC_FORCE = 2,
64 static int update_block_group(struct btrfs_trans_handle *trans,
65 struct btrfs_fs_info *fs_info, u64 bytenr,
66 u64 num_bytes, int alloc);
67 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
68 struct btrfs_fs_info *fs_info,
69 struct btrfs_delayed_ref_node *node, u64 parent,
70 u64 root_objectid, u64 owner_objectid,
71 u64 owner_offset, int refs_to_drop,
72 struct btrfs_delayed_extent_op *extra_op);
73 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
74 struct extent_buffer *leaf,
75 struct btrfs_extent_item *ei);
76 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
77 struct btrfs_fs_info *fs_info,
78 u64 parent, u64 root_objectid,
79 u64 flags, u64 owner, u64 offset,
80 struct btrfs_key *ins, int ref_mod);
81 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
82 struct btrfs_fs_info *fs_info,
83 u64 parent, u64 root_objectid,
84 u64 flags, struct btrfs_disk_key *key,
85 int level, struct btrfs_key *ins);
86 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
87 struct btrfs_fs_info *fs_info, u64 flags,
89 static int find_next_key(struct btrfs_path *path, int level,
90 struct btrfs_key *key);
91 static void dump_space_info(struct btrfs_fs_info *fs_info,
92 struct btrfs_space_info *info, u64 bytes,
93 int dump_block_groups);
94 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
95 u64 ram_bytes, u64 num_bytes, int delalloc);
96 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
97 u64 num_bytes, int delalloc);
98 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
100 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
101 struct btrfs_space_info *space_info,
103 enum btrfs_reserve_flush_enum flush,
105 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
106 struct btrfs_space_info *space_info,
108 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
109 struct btrfs_space_info *space_info,
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
116 return cache->cached == BTRFS_CACHE_FINISHED ||
117 cache->cached == BTRFS_CACHE_ERROR;
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
122 return (cache->flags & bits) == bits;
125 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
127 atomic_inc(&cache->count);
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
132 if (atomic_dec_and_test(&cache->count)) {
133 WARN_ON(cache->pinned > 0);
134 WARN_ON(cache->reserved > 0);
137 * If not empty, someone is still holding mutex of
138 * full_stripe_lock, which can only be released by caller.
139 * And it will definitely cause use-after-free when caller
140 * tries to release full stripe lock.
142 * No better way to resolve, but only to warn.
144 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
145 kfree(cache->free_space_ctl);
151 * this adds the block group to the fs_info rb tree for the block group
154 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
155 struct btrfs_block_group_cache *block_group)
158 struct rb_node *parent = NULL;
159 struct btrfs_block_group_cache *cache;
161 spin_lock(&info->block_group_cache_lock);
162 p = &info->block_group_cache_tree.rb_node;
166 cache = rb_entry(parent, struct btrfs_block_group_cache,
168 if (block_group->key.objectid < cache->key.objectid) {
170 } else if (block_group->key.objectid > cache->key.objectid) {
173 spin_unlock(&info->block_group_cache_lock);
178 rb_link_node(&block_group->cache_node, parent, p);
179 rb_insert_color(&block_group->cache_node,
180 &info->block_group_cache_tree);
182 if (info->first_logical_byte > block_group->key.objectid)
183 info->first_logical_byte = block_group->key.objectid;
185 spin_unlock(&info->block_group_cache_lock);
191 * This will return the block group at or after bytenr if contains is 0, else
192 * it will return the block group that contains the bytenr
194 static struct btrfs_block_group_cache *
195 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
198 struct btrfs_block_group_cache *cache, *ret = NULL;
202 spin_lock(&info->block_group_cache_lock);
203 n = info->block_group_cache_tree.rb_node;
206 cache = rb_entry(n, struct btrfs_block_group_cache,
208 end = cache->key.objectid + cache->key.offset - 1;
209 start = cache->key.objectid;
211 if (bytenr < start) {
212 if (!contains && (!ret || start < ret->key.objectid))
215 } else if (bytenr > start) {
216 if (contains && bytenr <= end) {
227 btrfs_get_block_group(ret);
228 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
229 info->first_logical_byte = ret->key.objectid;
231 spin_unlock(&info->block_group_cache_lock);
236 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
237 u64 start, u64 num_bytes)
239 u64 end = start + num_bytes - 1;
240 set_extent_bits(&fs_info->freed_extents[0],
241 start, end, EXTENT_UPTODATE);
242 set_extent_bits(&fs_info->freed_extents[1],
243 start, end, EXTENT_UPTODATE);
247 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
248 struct btrfs_block_group_cache *cache)
252 start = cache->key.objectid;
253 end = start + cache->key.offset - 1;
255 clear_extent_bits(&fs_info->freed_extents[0],
256 start, end, EXTENT_UPTODATE);
257 clear_extent_bits(&fs_info->freed_extents[1],
258 start, end, EXTENT_UPTODATE);
261 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
262 struct btrfs_block_group_cache *cache)
269 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
270 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
271 cache->bytes_super += stripe_len;
272 ret = add_excluded_extent(fs_info, cache->key.objectid,
278 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
279 bytenr = btrfs_sb_offset(i);
280 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
281 bytenr, 0, &logical, &nr, &stripe_len);
288 if (logical[nr] > cache->key.objectid +
292 if (logical[nr] + stripe_len <= cache->key.objectid)
296 if (start < cache->key.objectid) {
297 start = cache->key.objectid;
298 len = (logical[nr] + stripe_len) - start;
300 len = min_t(u64, stripe_len,
301 cache->key.objectid +
302 cache->key.offset - start);
305 cache->bytes_super += len;
306 ret = add_excluded_extent(fs_info, start, len);
318 static struct btrfs_caching_control *
319 get_caching_control(struct btrfs_block_group_cache *cache)
321 struct btrfs_caching_control *ctl;
323 spin_lock(&cache->lock);
324 if (!cache->caching_ctl) {
325 spin_unlock(&cache->lock);
329 ctl = cache->caching_ctl;
330 refcount_inc(&ctl->count);
331 spin_unlock(&cache->lock);
335 static void put_caching_control(struct btrfs_caching_control *ctl)
337 if (refcount_dec_and_test(&ctl->count))
341 #ifdef CONFIG_BTRFS_DEBUG
342 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
344 struct btrfs_fs_info *fs_info = block_group->fs_info;
345 u64 start = block_group->key.objectid;
346 u64 len = block_group->key.offset;
347 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
348 fs_info->nodesize : fs_info->sectorsize;
349 u64 step = chunk << 1;
351 while (len > chunk) {
352 btrfs_remove_free_space(block_group, start, chunk);
363 * this is only called by cache_block_group, since we could have freed extents
364 * we need to check the pinned_extents for any extents that can't be used yet
365 * since their free space will be released as soon as the transaction commits.
367 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
368 struct btrfs_fs_info *info, u64 start, u64 end)
370 u64 extent_start, extent_end, size, total_added = 0;
373 while (start < end) {
374 ret = find_first_extent_bit(info->pinned_extents, start,
375 &extent_start, &extent_end,
376 EXTENT_DIRTY | EXTENT_UPTODATE,
381 if (extent_start <= start) {
382 start = extent_end + 1;
383 } else if (extent_start > start && extent_start < end) {
384 size = extent_start - start;
386 ret = btrfs_add_free_space(block_group, start,
388 BUG_ON(ret); /* -ENOMEM or logic error */
389 start = extent_end + 1;
398 ret = btrfs_add_free_space(block_group, start, size);
399 BUG_ON(ret); /* -ENOMEM or logic error */
405 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
407 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
408 struct btrfs_fs_info *fs_info = block_group->fs_info;
409 struct btrfs_root *extent_root = fs_info->extent_root;
410 struct btrfs_path *path;
411 struct extent_buffer *leaf;
412 struct btrfs_key key;
419 path = btrfs_alloc_path();
423 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
425 #ifdef CONFIG_BTRFS_DEBUG
427 * If we're fragmenting we don't want to make anybody think we can
428 * allocate from this block group until we've had a chance to fragment
431 if (btrfs_should_fragment_free_space(block_group))
435 * We don't want to deadlock with somebody trying to allocate a new
436 * extent for the extent root while also trying to search the extent
437 * root to add free space. So we skip locking and search the commit
438 * root, since its read-only
440 path->skip_locking = 1;
441 path->search_commit_root = 1;
442 path->reada = READA_FORWARD;
446 key.type = BTRFS_EXTENT_ITEM_KEY;
449 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
453 leaf = path->nodes[0];
454 nritems = btrfs_header_nritems(leaf);
457 if (btrfs_fs_closing(fs_info) > 1) {
462 if (path->slots[0] < nritems) {
463 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
465 ret = find_next_key(path, 0, &key);
469 if (need_resched() ||
470 rwsem_is_contended(&fs_info->commit_root_sem)) {
472 caching_ctl->progress = last;
473 btrfs_release_path(path);
474 up_read(&fs_info->commit_root_sem);
475 mutex_unlock(&caching_ctl->mutex);
477 mutex_lock(&caching_ctl->mutex);
478 down_read(&fs_info->commit_root_sem);
482 ret = btrfs_next_leaf(extent_root, path);
487 leaf = path->nodes[0];
488 nritems = btrfs_header_nritems(leaf);
492 if (key.objectid < last) {
495 key.type = BTRFS_EXTENT_ITEM_KEY;
498 caching_ctl->progress = last;
499 btrfs_release_path(path);
503 if (key.objectid < block_group->key.objectid) {
508 if (key.objectid >= block_group->key.objectid +
509 block_group->key.offset)
512 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
513 key.type == BTRFS_METADATA_ITEM_KEY) {
514 total_found += add_new_free_space(block_group,
517 if (key.type == BTRFS_METADATA_ITEM_KEY)
518 last = key.objectid +
521 last = key.objectid + key.offset;
523 if (total_found > CACHING_CTL_WAKE_UP) {
526 wake_up(&caching_ctl->wait);
533 total_found += add_new_free_space(block_group, fs_info, last,
534 block_group->key.objectid +
535 block_group->key.offset);
536 caching_ctl->progress = (u64)-1;
539 btrfs_free_path(path);
543 static noinline void caching_thread(struct btrfs_work *work)
545 struct btrfs_block_group_cache *block_group;
546 struct btrfs_fs_info *fs_info;
547 struct btrfs_caching_control *caching_ctl;
548 struct btrfs_root *extent_root;
551 caching_ctl = container_of(work, struct btrfs_caching_control, work);
552 block_group = caching_ctl->block_group;
553 fs_info = block_group->fs_info;
554 extent_root = fs_info->extent_root;
556 mutex_lock(&caching_ctl->mutex);
557 down_read(&fs_info->commit_root_sem);
559 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
560 ret = load_free_space_tree(caching_ctl);
562 ret = load_extent_tree_free(caching_ctl);
564 spin_lock(&block_group->lock);
565 block_group->caching_ctl = NULL;
566 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
567 spin_unlock(&block_group->lock);
569 #ifdef CONFIG_BTRFS_DEBUG
570 if (btrfs_should_fragment_free_space(block_group)) {
573 spin_lock(&block_group->space_info->lock);
574 spin_lock(&block_group->lock);
575 bytes_used = block_group->key.offset -
576 btrfs_block_group_used(&block_group->item);
577 block_group->space_info->bytes_used += bytes_used >> 1;
578 spin_unlock(&block_group->lock);
579 spin_unlock(&block_group->space_info->lock);
580 fragment_free_space(block_group);
584 caching_ctl->progress = (u64)-1;
586 up_read(&fs_info->commit_root_sem);
587 free_excluded_extents(fs_info, block_group);
588 mutex_unlock(&caching_ctl->mutex);
590 wake_up(&caching_ctl->wait);
592 put_caching_control(caching_ctl);
593 btrfs_put_block_group(block_group);
596 static int cache_block_group(struct btrfs_block_group_cache *cache,
600 struct btrfs_fs_info *fs_info = cache->fs_info;
601 struct btrfs_caching_control *caching_ctl;
604 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
608 INIT_LIST_HEAD(&caching_ctl->list);
609 mutex_init(&caching_ctl->mutex);
610 init_waitqueue_head(&caching_ctl->wait);
611 caching_ctl->block_group = cache;
612 caching_ctl->progress = cache->key.objectid;
613 refcount_set(&caching_ctl->count, 1);
614 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
615 caching_thread, NULL, NULL);
617 spin_lock(&cache->lock);
619 * This should be a rare occasion, but this could happen I think in the
620 * case where one thread starts to load the space cache info, and then
621 * some other thread starts a transaction commit which tries to do an
622 * allocation while the other thread is still loading the space cache
623 * info. The previous loop should have kept us from choosing this block
624 * group, but if we've moved to the state where we will wait on caching
625 * block groups we need to first check if we're doing a fast load here,
626 * so we can wait for it to finish, otherwise we could end up allocating
627 * from a block group who's cache gets evicted for one reason or
630 while (cache->cached == BTRFS_CACHE_FAST) {
631 struct btrfs_caching_control *ctl;
633 ctl = cache->caching_ctl;
634 refcount_inc(&ctl->count);
635 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
636 spin_unlock(&cache->lock);
640 finish_wait(&ctl->wait, &wait);
641 put_caching_control(ctl);
642 spin_lock(&cache->lock);
645 if (cache->cached != BTRFS_CACHE_NO) {
646 spin_unlock(&cache->lock);
650 WARN_ON(cache->caching_ctl);
651 cache->caching_ctl = caching_ctl;
652 cache->cached = BTRFS_CACHE_FAST;
653 spin_unlock(&cache->lock);
655 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
656 mutex_lock(&caching_ctl->mutex);
657 ret = load_free_space_cache(fs_info, cache);
659 spin_lock(&cache->lock);
661 cache->caching_ctl = NULL;
662 cache->cached = BTRFS_CACHE_FINISHED;
663 cache->last_byte_to_unpin = (u64)-1;
664 caching_ctl->progress = (u64)-1;
666 if (load_cache_only) {
667 cache->caching_ctl = NULL;
668 cache->cached = BTRFS_CACHE_NO;
670 cache->cached = BTRFS_CACHE_STARTED;
671 cache->has_caching_ctl = 1;
674 spin_unlock(&cache->lock);
675 #ifdef CONFIG_BTRFS_DEBUG
677 btrfs_should_fragment_free_space(cache)) {
680 spin_lock(&cache->space_info->lock);
681 spin_lock(&cache->lock);
682 bytes_used = cache->key.offset -
683 btrfs_block_group_used(&cache->item);
684 cache->space_info->bytes_used += bytes_used >> 1;
685 spin_unlock(&cache->lock);
686 spin_unlock(&cache->space_info->lock);
687 fragment_free_space(cache);
690 mutex_unlock(&caching_ctl->mutex);
692 wake_up(&caching_ctl->wait);
694 put_caching_control(caching_ctl);
695 free_excluded_extents(fs_info, cache);
700 * We're either using the free space tree or no caching at all.
701 * Set cached to the appropriate value and wakeup any waiters.
703 spin_lock(&cache->lock);
704 if (load_cache_only) {
705 cache->caching_ctl = NULL;
706 cache->cached = BTRFS_CACHE_NO;
708 cache->cached = BTRFS_CACHE_STARTED;
709 cache->has_caching_ctl = 1;
711 spin_unlock(&cache->lock);
712 wake_up(&caching_ctl->wait);
715 if (load_cache_only) {
716 put_caching_control(caching_ctl);
720 down_write(&fs_info->commit_root_sem);
721 refcount_inc(&caching_ctl->count);
722 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
723 up_write(&fs_info->commit_root_sem);
725 btrfs_get_block_group(cache);
727 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
733 * return the block group that starts at or after bytenr
735 static struct btrfs_block_group_cache *
736 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
738 return block_group_cache_tree_search(info, bytenr, 0);
742 * return the block group that contains the given bytenr
744 struct btrfs_block_group_cache *btrfs_lookup_block_group(
745 struct btrfs_fs_info *info,
748 return block_group_cache_tree_search(info, bytenr, 1);
751 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
754 struct list_head *head = &info->space_info;
755 struct btrfs_space_info *found;
757 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
760 list_for_each_entry_rcu(found, head, list) {
761 if (found->flags & flags) {
770 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
771 u64 owner, u64 root_objectid)
773 struct btrfs_space_info *space_info;
776 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
777 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
778 flags = BTRFS_BLOCK_GROUP_SYSTEM;
780 flags = BTRFS_BLOCK_GROUP_METADATA;
782 flags = BTRFS_BLOCK_GROUP_DATA;
785 space_info = __find_space_info(fs_info, flags);
787 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
791 * after adding space to the filesystem, we need to clear the full flags
792 * on all the space infos.
794 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
796 struct list_head *head = &info->space_info;
797 struct btrfs_space_info *found;
800 list_for_each_entry_rcu(found, head, list)
805 /* simple helper to search for an existing data extent at a given offset */
806 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
809 struct btrfs_key key;
810 struct btrfs_path *path;
812 path = btrfs_alloc_path();
816 key.objectid = start;
818 key.type = BTRFS_EXTENT_ITEM_KEY;
819 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
820 btrfs_free_path(path);
825 * helper function to lookup reference count and flags of a tree block.
827 * the head node for delayed ref is used to store the sum of all the
828 * reference count modifications queued up in the rbtree. the head
829 * node may also store the extent flags to set. This way you can check
830 * to see what the reference count and extent flags would be if all of
831 * the delayed refs are not processed.
833 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
834 struct btrfs_fs_info *fs_info, u64 bytenr,
835 u64 offset, int metadata, u64 *refs, u64 *flags)
837 struct btrfs_delayed_ref_head *head;
838 struct btrfs_delayed_ref_root *delayed_refs;
839 struct btrfs_path *path;
840 struct btrfs_extent_item *ei;
841 struct extent_buffer *leaf;
842 struct btrfs_key key;
849 * If we don't have skinny metadata, don't bother doing anything
852 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
853 offset = fs_info->nodesize;
857 path = btrfs_alloc_path();
862 path->skip_locking = 1;
863 path->search_commit_root = 1;
867 key.objectid = bytenr;
870 key.type = BTRFS_METADATA_ITEM_KEY;
872 key.type = BTRFS_EXTENT_ITEM_KEY;
874 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
878 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
879 if (path->slots[0]) {
881 btrfs_item_key_to_cpu(path->nodes[0], &key,
883 if (key.objectid == bytenr &&
884 key.type == BTRFS_EXTENT_ITEM_KEY &&
885 key.offset == fs_info->nodesize)
891 leaf = path->nodes[0];
892 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
893 if (item_size >= sizeof(*ei)) {
894 ei = btrfs_item_ptr(leaf, path->slots[0],
895 struct btrfs_extent_item);
896 num_refs = btrfs_extent_refs(leaf, ei);
897 extent_flags = btrfs_extent_flags(leaf, ei);
899 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
900 struct btrfs_extent_item_v0 *ei0;
901 BUG_ON(item_size != sizeof(*ei0));
902 ei0 = btrfs_item_ptr(leaf, path->slots[0],
903 struct btrfs_extent_item_v0);
904 num_refs = btrfs_extent_refs_v0(leaf, ei0);
905 /* FIXME: this isn't correct for data */
906 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
911 BUG_ON(num_refs == 0);
921 delayed_refs = &trans->transaction->delayed_refs;
922 spin_lock(&delayed_refs->lock);
923 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
925 if (!mutex_trylock(&head->mutex)) {
926 refcount_inc(&head->node.refs);
927 spin_unlock(&delayed_refs->lock);
929 btrfs_release_path(path);
932 * Mutex was contended, block until it's released and try
935 mutex_lock(&head->mutex);
936 mutex_unlock(&head->mutex);
937 btrfs_put_delayed_ref(&head->node);
940 spin_lock(&head->lock);
941 if (head->extent_op && head->extent_op->update_flags)
942 extent_flags |= head->extent_op->flags_to_set;
944 BUG_ON(num_refs == 0);
946 num_refs += head->node.ref_mod;
947 spin_unlock(&head->lock);
948 mutex_unlock(&head->mutex);
950 spin_unlock(&delayed_refs->lock);
952 WARN_ON(num_refs == 0);
956 *flags = extent_flags;
958 btrfs_free_path(path);
963 * Back reference rules. Back refs have three main goals:
965 * 1) differentiate between all holders of references to an extent so that
966 * when a reference is dropped we can make sure it was a valid reference
967 * before freeing the extent.
969 * 2) Provide enough information to quickly find the holders of an extent
970 * if we notice a given block is corrupted or bad.
972 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
973 * maintenance. This is actually the same as #2, but with a slightly
974 * different use case.
976 * There are two kinds of back refs. The implicit back refs is optimized
977 * for pointers in non-shared tree blocks. For a given pointer in a block,
978 * back refs of this kind provide information about the block's owner tree
979 * and the pointer's key. These information allow us to find the block by
980 * b-tree searching. The full back refs is for pointers in tree blocks not
981 * referenced by their owner trees. The location of tree block is recorded
982 * in the back refs. Actually the full back refs is generic, and can be
983 * used in all cases the implicit back refs is used. The major shortcoming
984 * of the full back refs is its overhead. Every time a tree block gets
985 * COWed, we have to update back refs entry for all pointers in it.
987 * For a newly allocated tree block, we use implicit back refs for
988 * pointers in it. This means most tree related operations only involve
989 * implicit back refs. For a tree block created in old transaction, the
990 * only way to drop a reference to it is COW it. So we can detect the
991 * event that tree block loses its owner tree's reference and do the
992 * back refs conversion.
994 * When a tree block is COWed through a tree, there are four cases:
996 * The reference count of the block is one and the tree is the block's
997 * owner tree. Nothing to do in this case.
999 * The reference count of the block is one and the tree is not the
1000 * block's owner tree. In this case, full back refs is used for pointers
1001 * in the block. Remove these full back refs, add implicit back refs for
1002 * every pointers in the new block.
1004 * The reference count of the block is greater than one and the tree is
1005 * the block's owner tree. In this case, implicit back refs is used for
1006 * pointers in the block. Add full back refs for every pointers in the
1007 * block, increase lower level extents' reference counts. The original
1008 * implicit back refs are entailed to the new block.
1010 * The reference count of the block is greater than one and the tree is
1011 * not the block's owner tree. Add implicit back refs for every pointer in
1012 * the new block, increase lower level extents' reference count.
1014 * Back Reference Key composing:
1016 * The key objectid corresponds to the first byte in the extent,
1017 * The key type is used to differentiate between types of back refs.
1018 * There are different meanings of the key offset for different types
1021 * File extents can be referenced by:
1023 * - multiple snapshots, subvolumes, or different generations in one subvol
1024 * - different files inside a single subvolume
1025 * - different offsets inside a file (bookend extents in file.c)
1027 * The extent ref structure for the implicit back refs has fields for:
1029 * - Objectid of the subvolume root
1030 * - objectid of the file holding the reference
1031 * - original offset in the file
1032 * - how many bookend extents
1034 * The key offset for the implicit back refs is hash of the first
1037 * The extent ref structure for the full back refs has field for:
1039 * - number of pointers in the tree leaf
1041 * The key offset for the implicit back refs is the first byte of
1044 * When a file extent is allocated, The implicit back refs is used.
1045 * the fields are filled in:
1047 * (root_key.objectid, inode objectid, offset in file, 1)
1049 * When a file extent is removed file truncation, we find the
1050 * corresponding implicit back refs and check the following fields:
1052 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1054 * Btree extents can be referenced by:
1056 * - Different subvolumes
1058 * Both the implicit back refs and the full back refs for tree blocks
1059 * only consist of key. The key offset for the implicit back refs is
1060 * objectid of block's owner tree. The key offset for the full back refs
1061 * is the first byte of parent block.
1063 * When implicit back refs is used, information about the lowest key and
1064 * level of the tree block are required. These information are stored in
1065 * tree block info structure.
1068 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1069 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1070 struct btrfs_fs_info *fs_info,
1071 struct btrfs_path *path,
1072 u64 owner, u32 extra_size)
1074 struct btrfs_root *root = fs_info->extent_root;
1075 struct btrfs_extent_item *item;
1076 struct btrfs_extent_item_v0 *ei0;
1077 struct btrfs_extent_ref_v0 *ref0;
1078 struct btrfs_tree_block_info *bi;
1079 struct extent_buffer *leaf;
1080 struct btrfs_key key;
1081 struct btrfs_key found_key;
1082 u32 new_size = sizeof(*item);
1086 leaf = path->nodes[0];
1087 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1089 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1090 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1091 struct btrfs_extent_item_v0);
1092 refs = btrfs_extent_refs_v0(leaf, ei0);
1094 if (owner == (u64)-1) {
1096 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1097 ret = btrfs_next_leaf(root, path);
1100 BUG_ON(ret > 0); /* Corruption */
1101 leaf = path->nodes[0];
1103 btrfs_item_key_to_cpu(leaf, &found_key,
1105 BUG_ON(key.objectid != found_key.objectid);
1106 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1110 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1111 struct btrfs_extent_ref_v0);
1112 owner = btrfs_ref_objectid_v0(leaf, ref0);
1116 btrfs_release_path(path);
1118 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1119 new_size += sizeof(*bi);
1121 new_size -= sizeof(*ei0);
1122 ret = btrfs_search_slot(trans, root, &key, path,
1123 new_size + extra_size, 1);
1126 BUG_ON(ret); /* Corruption */
1128 btrfs_extend_item(fs_info, path, new_size);
1130 leaf = path->nodes[0];
1131 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1132 btrfs_set_extent_refs(leaf, item, refs);
1133 /* FIXME: get real generation */
1134 btrfs_set_extent_generation(leaf, item, 0);
1135 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1136 btrfs_set_extent_flags(leaf, item,
1137 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1138 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1139 bi = (struct btrfs_tree_block_info *)(item + 1);
1140 /* FIXME: get first key of the block */
1141 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1142 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1144 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1146 btrfs_mark_buffer_dirty(leaf);
1152 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1153 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1154 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1156 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1157 struct btrfs_extent_inline_ref *iref,
1158 enum btrfs_inline_ref_type is_data)
1160 int type = btrfs_extent_inline_ref_type(eb, iref);
1161 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1163 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1164 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1165 type == BTRFS_SHARED_DATA_REF_KEY ||
1166 type == BTRFS_EXTENT_DATA_REF_KEY) {
1167 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1168 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1170 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1171 ASSERT(eb->fs_info);
1173 * Every shared one has parent tree
1174 * block, which must be aligned to
1178 IS_ALIGNED(offset, eb->fs_info->nodesize))
1181 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1182 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1184 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1185 ASSERT(eb->fs_info);
1187 * Every shared one has parent tree
1188 * block, which must be aligned to
1192 IS_ALIGNED(offset, eb->fs_info->nodesize))
1196 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1201 btrfs_print_leaf((struct extent_buffer *)eb);
1202 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1206 return BTRFS_REF_TYPE_INVALID;
1209 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1211 u32 high_crc = ~(u32)0;
1212 u32 low_crc = ~(u32)0;
1215 lenum = cpu_to_le64(root_objectid);
1216 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1217 lenum = cpu_to_le64(owner);
1218 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1219 lenum = cpu_to_le64(offset);
1220 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1222 return ((u64)high_crc << 31) ^ (u64)low_crc;
1225 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1226 struct btrfs_extent_data_ref *ref)
1228 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1229 btrfs_extent_data_ref_objectid(leaf, ref),
1230 btrfs_extent_data_ref_offset(leaf, ref));
1233 static int match_extent_data_ref(struct extent_buffer *leaf,
1234 struct btrfs_extent_data_ref *ref,
1235 u64 root_objectid, u64 owner, u64 offset)
1237 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1238 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1239 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1244 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1245 struct btrfs_fs_info *fs_info,
1246 struct btrfs_path *path,
1247 u64 bytenr, u64 parent,
1249 u64 owner, u64 offset)
1251 struct btrfs_root *root = fs_info->extent_root;
1252 struct btrfs_key key;
1253 struct btrfs_extent_data_ref *ref;
1254 struct extent_buffer *leaf;
1260 key.objectid = bytenr;
1262 key.type = BTRFS_SHARED_DATA_REF_KEY;
1263 key.offset = parent;
1265 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1266 key.offset = hash_extent_data_ref(root_objectid,
1271 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1280 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1281 key.type = BTRFS_EXTENT_REF_V0_KEY;
1282 btrfs_release_path(path);
1283 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1294 leaf = path->nodes[0];
1295 nritems = btrfs_header_nritems(leaf);
1297 if (path->slots[0] >= nritems) {
1298 ret = btrfs_next_leaf(root, path);
1304 leaf = path->nodes[0];
1305 nritems = btrfs_header_nritems(leaf);
1309 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1310 if (key.objectid != bytenr ||
1311 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1314 ref = btrfs_item_ptr(leaf, path->slots[0],
1315 struct btrfs_extent_data_ref);
1317 if (match_extent_data_ref(leaf, ref, root_objectid,
1320 btrfs_release_path(path);
1332 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1333 struct btrfs_fs_info *fs_info,
1334 struct btrfs_path *path,
1335 u64 bytenr, u64 parent,
1336 u64 root_objectid, u64 owner,
1337 u64 offset, int refs_to_add)
1339 struct btrfs_root *root = fs_info->extent_root;
1340 struct btrfs_key key;
1341 struct extent_buffer *leaf;
1346 key.objectid = bytenr;
1348 key.type = BTRFS_SHARED_DATA_REF_KEY;
1349 key.offset = parent;
1350 size = sizeof(struct btrfs_shared_data_ref);
1352 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1353 key.offset = hash_extent_data_ref(root_objectid,
1355 size = sizeof(struct btrfs_extent_data_ref);
1358 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1359 if (ret && ret != -EEXIST)
1362 leaf = path->nodes[0];
1364 struct btrfs_shared_data_ref *ref;
1365 ref = btrfs_item_ptr(leaf, path->slots[0],
1366 struct btrfs_shared_data_ref);
1368 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1370 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1371 num_refs += refs_to_add;
1372 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1375 struct btrfs_extent_data_ref *ref;
1376 while (ret == -EEXIST) {
1377 ref = btrfs_item_ptr(leaf, path->slots[0],
1378 struct btrfs_extent_data_ref);
1379 if (match_extent_data_ref(leaf, ref, root_objectid,
1382 btrfs_release_path(path);
1384 ret = btrfs_insert_empty_item(trans, root, path, &key,
1386 if (ret && ret != -EEXIST)
1389 leaf = path->nodes[0];
1391 ref = btrfs_item_ptr(leaf, path->slots[0],
1392 struct btrfs_extent_data_ref);
1394 btrfs_set_extent_data_ref_root(leaf, ref,
1396 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1397 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1398 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1400 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1401 num_refs += refs_to_add;
1402 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1405 btrfs_mark_buffer_dirty(leaf);
1408 btrfs_release_path(path);
1412 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1413 struct btrfs_fs_info *fs_info,
1414 struct btrfs_path *path,
1415 int refs_to_drop, int *last_ref)
1417 struct btrfs_key key;
1418 struct btrfs_extent_data_ref *ref1 = NULL;
1419 struct btrfs_shared_data_ref *ref2 = NULL;
1420 struct extent_buffer *leaf;
1424 leaf = path->nodes[0];
1425 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1427 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1428 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1429 struct btrfs_extent_data_ref);
1430 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1431 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1432 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1433 struct btrfs_shared_data_ref);
1434 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1435 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1436 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1437 struct btrfs_extent_ref_v0 *ref0;
1438 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1439 struct btrfs_extent_ref_v0);
1440 num_refs = btrfs_ref_count_v0(leaf, ref0);
1446 BUG_ON(num_refs < refs_to_drop);
1447 num_refs -= refs_to_drop;
1449 if (num_refs == 0) {
1450 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1453 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1454 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1455 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1456 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1457 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1459 struct btrfs_extent_ref_v0 *ref0;
1460 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1461 struct btrfs_extent_ref_v0);
1462 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1465 btrfs_mark_buffer_dirty(leaf);
1470 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1471 struct btrfs_extent_inline_ref *iref)
1473 struct btrfs_key key;
1474 struct extent_buffer *leaf;
1475 struct btrfs_extent_data_ref *ref1;
1476 struct btrfs_shared_data_ref *ref2;
1480 leaf = path->nodes[0];
1481 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1484 * If type is invalid, we should have bailed out earlier than
1487 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1488 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1489 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1490 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1491 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1493 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1494 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1496 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1497 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1498 struct btrfs_extent_data_ref);
1499 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1500 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1501 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1502 struct btrfs_shared_data_ref);
1503 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1504 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1505 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1506 struct btrfs_extent_ref_v0 *ref0;
1507 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1508 struct btrfs_extent_ref_v0);
1509 num_refs = btrfs_ref_count_v0(leaf, ref0);
1517 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1518 struct btrfs_fs_info *fs_info,
1519 struct btrfs_path *path,
1520 u64 bytenr, u64 parent,
1523 struct btrfs_root *root = fs_info->extent_root;
1524 struct btrfs_key key;
1527 key.objectid = bytenr;
1529 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1530 key.offset = parent;
1532 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1533 key.offset = root_objectid;
1536 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1539 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1540 if (ret == -ENOENT && parent) {
1541 btrfs_release_path(path);
1542 key.type = BTRFS_EXTENT_REF_V0_KEY;
1543 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1551 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1552 struct btrfs_fs_info *fs_info,
1553 struct btrfs_path *path,
1554 u64 bytenr, u64 parent,
1557 struct btrfs_key key;
1560 key.objectid = bytenr;
1562 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1563 key.offset = parent;
1565 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1566 key.offset = root_objectid;
1569 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1571 btrfs_release_path(path);
1575 static inline int extent_ref_type(u64 parent, u64 owner)
1578 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1580 type = BTRFS_SHARED_BLOCK_REF_KEY;
1582 type = BTRFS_TREE_BLOCK_REF_KEY;
1585 type = BTRFS_SHARED_DATA_REF_KEY;
1587 type = BTRFS_EXTENT_DATA_REF_KEY;
1592 static int find_next_key(struct btrfs_path *path, int level,
1593 struct btrfs_key *key)
1596 for (; level < BTRFS_MAX_LEVEL; level++) {
1597 if (!path->nodes[level])
1599 if (path->slots[level] + 1 >=
1600 btrfs_header_nritems(path->nodes[level]))
1603 btrfs_item_key_to_cpu(path->nodes[level], key,
1604 path->slots[level] + 1);
1606 btrfs_node_key_to_cpu(path->nodes[level], key,
1607 path->slots[level] + 1);
1614 * look for inline back ref. if back ref is found, *ref_ret is set
1615 * to the address of inline back ref, and 0 is returned.
1617 * if back ref isn't found, *ref_ret is set to the address where it
1618 * should be inserted, and -ENOENT is returned.
1620 * if insert is true and there are too many inline back refs, the path
1621 * points to the extent item, and -EAGAIN is returned.
1623 * NOTE: inline back refs are ordered in the same way that back ref
1624 * items in the tree are ordered.
1626 static noinline_for_stack
1627 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1628 struct btrfs_fs_info *fs_info,
1629 struct btrfs_path *path,
1630 struct btrfs_extent_inline_ref **ref_ret,
1631 u64 bytenr, u64 num_bytes,
1632 u64 parent, u64 root_objectid,
1633 u64 owner, u64 offset, int insert)
1635 struct btrfs_root *root = fs_info->extent_root;
1636 struct btrfs_key key;
1637 struct extent_buffer *leaf;
1638 struct btrfs_extent_item *ei;
1639 struct btrfs_extent_inline_ref *iref;
1649 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1652 key.objectid = bytenr;
1653 key.type = BTRFS_EXTENT_ITEM_KEY;
1654 key.offset = num_bytes;
1656 want = extent_ref_type(parent, owner);
1658 extra_size = btrfs_extent_inline_ref_size(want);
1659 path->keep_locks = 1;
1664 * Owner is our parent level, so we can just add one to get the level
1665 * for the block we are interested in.
1667 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1668 key.type = BTRFS_METADATA_ITEM_KEY;
1673 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1680 * We may be a newly converted file system which still has the old fat
1681 * extent entries for metadata, so try and see if we have one of those.
1683 if (ret > 0 && skinny_metadata) {
1684 skinny_metadata = false;
1685 if (path->slots[0]) {
1687 btrfs_item_key_to_cpu(path->nodes[0], &key,
1689 if (key.objectid == bytenr &&
1690 key.type == BTRFS_EXTENT_ITEM_KEY &&
1691 key.offset == num_bytes)
1695 key.objectid = bytenr;
1696 key.type = BTRFS_EXTENT_ITEM_KEY;
1697 key.offset = num_bytes;
1698 btrfs_release_path(path);
1703 if (ret && !insert) {
1706 } else if (WARN_ON(ret)) {
1711 leaf = path->nodes[0];
1712 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1713 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1714 if (item_size < sizeof(*ei)) {
1719 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1725 leaf = path->nodes[0];
1726 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1729 BUG_ON(item_size < sizeof(*ei));
1731 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1732 flags = btrfs_extent_flags(leaf, ei);
1734 ptr = (unsigned long)(ei + 1);
1735 end = (unsigned long)ei + item_size;
1737 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1738 ptr += sizeof(struct btrfs_tree_block_info);
1742 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1743 needed = BTRFS_REF_TYPE_DATA;
1745 needed = BTRFS_REF_TYPE_BLOCK;
1753 iref = (struct btrfs_extent_inline_ref *)ptr;
1754 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1755 if (type == BTRFS_REF_TYPE_INVALID) {
1763 ptr += btrfs_extent_inline_ref_size(type);
1767 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1768 struct btrfs_extent_data_ref *dref;
1769 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1770 if (match_extent_data_ref(leaf, dref, root_objectid,
1775 if (hash_extent_data_ref_item(leaf, dref) <
1776 hash_extent_data_ref(root_objectid, owner, offset))
1780 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1782 if (parent == ref_offset) {
1786 if (ref_offset < parent)
1789 if (root_objectid == ref_offset) {
1793 if (ref_offset < root_objectid)
1797 ptr += btrfs_extent_inline_ref_size(type);
1799 if (err == -ENOENT && insert) {
1800 if (item_size + extra_size >=
1801 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1806 * To add new inline back ref, we have to make sure
1807 * there is no corresponding back ref item.
1808 * For simplicity, we just do not add new inline back
1809 * ref if there is any kind of item for this block
1811 if (find_next_key(path, 0, &key) == 0 &&
1812 key.objectid == bytenr &&
1813 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1818 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1821 path->keep_locks = 0;
1822 btrfs_unlock_up_safe(path, 1);
1828 * helper to add new inline back ref
1830 static noinline_for_stack
1831 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1832 struct btrfs_path *path,
1833 struct btrfs_extent_inline_ref *iref,
1834 u64 parent, u64 root_objectid,
1835 u64 owner, u64 offset, int refs_to_add,
1836 struct btrfs_delayed_extent_op *extent_op)
1838 struct extent_buffer *leaf;
1839 struct btrfs_extent_item *ei;
1842 unsigned long item_offset;
1847 leaf = path->nodes[0];
1848 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1849 item_offset = (unsigned long)iref - (unsigned long)ei;
1851 type = extent_ref_type(parent, owner);
1852 size = btrfs_extent_inline_ref_size(type);
1854 btrfs_extend_item(fs_info, path, size);
1856 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1857 refs = btrfs_extent_refs(leaf, ei);
1858 refs += refs_to_add;
1859 btrfs_set_extent_refs(leaf, ei, refs);
1861 __run_delayed_extent_op(extent_op, leaf, ei);
1863 ptr = (unsigned long)ei + item_offset;
1864 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1865 if (ptr < end - size)
1866 memmove_extent_buffer(leaf, ptr + size, ptr,
1869 iref = (struct btrfs_extent_inline_ref *)ptr;
1870 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1871 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1872 struct btrfs_extent_data_ref *dref;
1873 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1874 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1875 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1876 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1877 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1878 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1879 struct btrfs_shared_data_ref *sref;
1880 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1881 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1882 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1883 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1884 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1886 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1888 btrfs_mark_buffer_dirty(leaf);
1891 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1892 struct btrfs_fs_info *fs_info,
1893 struct btrfs_path *path,
1894 struct btrfs_extent_inline_ref **ref_ret,
1895 u64 bytenr, u64 num_bytes, u64 parent,
1896 u64 root_objectid, u64 owner, u64 offset)
1900 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1901 bytenr, num_bytes, parent,
1902 root_objectid, owner, offset, 0);
1906 btrfs_release_path(path);
1909 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1910 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1911 parent, root_objectid);
1913 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1914 parent, root_objectid, owner,
1921 * helper to update/remove inline back ref
1923 static noinline_for_stack
1924 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1925 struct btrfs_path *path,
1926 struct btrfs_extent_inline_ref *iref,
1928 struct btrfs_delayed_extent_op *extent_op,
1931 struct extent_buffer *leaf;
1932 struct btrfs_extent_item *ei;
1933 struct btrfs_extent_data_ref *dref = NULL;
1934 struct btrfs_shared_data_ref *sref = NULL;
1942 leaf = path->nodes[0];
1943 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1944 refs = btrfs_extent_refs(leaf, ei);
1945 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1946 refs += refs_to_mod;
1947 btrfs_set_extent_refs(leaf, ei, refs);
1949 __run_delayed_extent_op(extent_op, leaf, ei);
1952 * If type is invalid, we should have bailed out after
1953 * lookup_inline_extent_backref().
1955 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1956 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1958 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1959 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1960 refs = btrfs_extent_data_ref_count(leaf, dref);
1961 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1962 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1963 refs = btrfs_shared_data_ref_count(leaf, sref);
1966 BUG_ON(refs_to_mod != -1);
1969 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1970 refs += refs_to_mod;
1973 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1974 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1976 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1979 size = btrfs_extent_inline_ref_size(type);
1980 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1981 ptr = (unsigned long)iref;
1982 end = (unsigned long)ei + item_size;
1983 if (ptr + size < end)
1984 memmove_extent_buffer(leaf, ptr, ptr + size,
1987 btrfs_truncate_item(fs_info, path, item_size, 1);
1989 btrfs_mark_buffer_dirty(leaf);
1992 static noinline_for_stack
1993 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1994 struct btrfs_fs_info *fs_info,
1995 struct btrfs_path *path,
1996 u64 bytenr, u64 num_bytes, u64 parent,
1997 u64 root_objectid, u64 owner,
1998 u64 offset, int refs_to_add,
1999 struct btrfs_delayed_extent_op *extent_op)
2001 struct btrfs_extent_inline_ref *iref;
2004 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
2005 bytenr, num_bytes, parent,
2006 root_objectid, owner, offset, 1);
2008 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
2009 update_inline_extent_backref(fs_info, path, iref,
2010 refs_to_add, extent_op, NULL);
2011 } else if (ret == -ENOENT) {
2012 setup_inline_extent_backref(fs_info, path, iref, parent,
2013 root_objectid, owner, offset,
2014 refs_to_add, extent_op);
2020 static int insert_extent_backref(struct btrfs_trans_handle *trans,
2021 struct btrfs_fs_info *fs_info,
2022 struct btrfs_path *path,
2023 u64 bytenr, u64 parent, u64 root_objectid,
2024 u64 owner, u64 offset, int refs_to_add)
2027 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2028 BUG_ON(refs_to_add != 1);
2029 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
2030 parent, root_objectid);
2032 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
2033 parent, root_objectid,
2034 owner, offset, refs_to_add);
2039 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2040 struct btrfs_fs_info *fs_info,
2041 struct btrfs_path *path,
2042 struct btrfs_extent_inline_ref *iref,
2043 int refs_to_drop, int is_data, int *last_ref)
2047 BUG_ON(!is_data && refs_to_drop != 1);
2049 update_inline_extent_backref(fs_info, path, iref,
2050 -refs_to_drop, NULL, last_ref);
2051 } else if (is_data) {
2052 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
2056 ret = btrfs_del_item(trans, fs_info->extent_root, path);
2061 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2062 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2063 u64 *discarded_bytes)
2066 u64 bytes_left, end;
2067 u64 aligned_start = ALIGN(start, 1 << 9);
2069 if (WARN_ON(start != aligned_start)) {
2070 len -= aligned_start - start;
2071 len = round_down(len, 1 << 9);
2072 start = aligned_start;
2075 *discarded_bytes = 0;
2083 /* Skip any superblocks on this device. */
2084 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2085 u64 sb_start = btrfs_sb_offset(j);
2086 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2087 u64 size = sb_start - start;
2089 if (!in_range(sb_start, start, bytes_left) &&
2090 !in_range(sb_end, start, bytes_left) &&
2091 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2095 * Superblock spans beginning of range. Adjust start and
2098 if (sb_start <= start) {
2099 start += sb_end - start;
2104 bytes_left = end - start;
2109 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2112 *discarded_bytes += size;
2113 else if (ret != -EOPNOTSUPP)
2122 bytes_left = end - start;
2126 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2129 *discarded_bytes += bytes_left;
2134 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2135 u64 num_bytes, u64 *actual_bytes)
2138 u64 discarded_bytes = 0;
2139 struct btrfs_bio *bbio = NULL;
2143 * Avoid races with device replace and make sure our bbio has devices
2144 * associated to its stripes that don't go away while we are discarding.
2146 btrfs_bio_counter_inc_blocked(fs_info);
2147 /* Tell the block device(s) that the sectors can be discarded */
2148 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2150 /* Error condition is -ENOMEM */
2152 struct btrfs_bio_stripe *stripe = bbio->stripes;
2156 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2158 if (!stripe->dev->can_discard)
2161 ret = btrfs_issue_discard(stripe->dev->bdev,
2166 discarded_bytes += bytes;
2167 else if (ret != -EOPNOTSUPP)
2168 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2171 * Just in case we get back EOPNOTSUPP for some reason,
2172 * just ignore the return value so we don't screw up
2173 * people calling discard_extent.
2177 btrfs_put_bbio(bbio);
2179 btrfs_bio_counter_dec(fs_info);
2182 *actual_bytes = discarded_bytes;
2185 if (ret == -EOPNOTSUPP)
2190 /* Can return -ENOMEM */
2191 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2192 struct btrfs_fs_info *fs_info,
2193 u64 bytenr, u64 num_bytes, u64 parent,
2194 u64 root_objectid, u64 owner, u64 offset)
2196 int old_ref_mod, new_ref_mod;
2199 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2200 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2202 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2203 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2205 root_objectid, (int)owner,
2206 BTRFS_ADD_DELAYED_REF, NULL,
2207 &old_ref_mod, &new_ref_mod);
2209 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2211 root_objectid, owner, offset,
2212 0, BTRFS_ADD_DELAYED_REF,
2213 &old_ref_mod, &new_ref_mod);
2216 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2217 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2222 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2223 struct btrfs_fs_info *fs_info,
2224 struct btrfs_delayed_ref_node *node,
2225 u64 parent, u64 root_objectid,
2226 u64 owner, u64 offset, int refs_to_add,
2227 struct btrfs_delayed_extent_op *extent_op)
2229 struct btrfs_path *path;
2230 struct extent_buffer *leaf;
2231 struct btrfs_extent_item *item;
2232 struct btrfs_key key;
2233 u64 bytenr = node->bytenr;
2234 u64 num_bytes = node->num_bytes;
2238 path = btrfs_alloc_path();
2242 path->reada = READA_FORWARD;
2243 path->leave_spinning = 1;
2244 /* this will setup the path even if it fails to insert the back ref */
2245 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2246 num_bytes, parent, root_objectid,
2248 refs_to_add, extent_op);
2249 if ((ret < 0 && ret != -EAGAIN) || !ret)
2253 * Ok we had -EAGAIN which means we didn't have space to insert and
2254 * inline extent ref, so just update the reference count and add a
2257 leaf = path->nodes[0];
2258 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2259 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2260 refs = btrfs_extent_refs(leaf, item);
2261 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2263 __run_delayed_extent_op(extent_op, leaf, item);
2265 btrfs_mark_buffer_dirty(leaf);
2266 btrfs_release_path(path);
2268 path->reada = READA_FORWARD;
2269 path->leave_spinning = 1;
2270 /* now insert the actual backref */
2271 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2272 root_objectid, owner, offset, refs_to_add);
2274 btrfs_abort_transaction(trans, ret);
2276 btrfs_free_path(path);
2280 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2281 struct btrfs_fs_info *fs_info,
2282 struct btrfs_delayed_ref_node *node,
2283 struct btrfs_delayed_extent_op *extent_op,
2284 int insert_reserved)
2287 struct btrfs_delayed_data_ref *ref;
2288 struct btrfs_key ins;
2293 ins.objectid = node->bytenr;
2294 ins.offset = node->num_bytes;
2295 ins.type = BTRFS_EXTENT_ITEM_KEY;
2297 ref = btrfs_delayed_node_to_data_ref(node);
2298 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2300 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2301 parent = ref->parent;
2302 ref_root = ref->root;
2304 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2306 flags |= extent_op->flags_to_set;
2307 ret = alloc_reserved_file_extent(trans, fs_info,
2308 parent, ref_root, flags,
2309 ref->objectid, ref->offset,
2310 &ins, node->ref_mod);
2311 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2312 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2313 ref_root, ref->objectid,
2314 ref->offset, node->ref_mod,
2316 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2317 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2318 ref_root, ref->objectid,
2319 ref->offset, node->ref_mod,
2327 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2328 struct extent_buffer *leaf,
2329 struct btrfs_extent_item *ei)
2331 u64 flags = btrfs_extent_flags(leaf, ei);
2332 if (extent_op->update_flags) {
2333 flags |= extent_op->flags_to_set;
2334 btrfs_set_extent_flags(leaf, ei, flags);
2337 if (extent_op->update_key) {
2338 struct btrfs_tree_block_info *bi;
2339 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2340 bi = (struct btrfs_tree_block_info *)(ei + 1);
2341 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2345 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2346 struct btrfs_fs_info *fs_info,
2347 struct btrfs_delayed_ref_node *node,
2348 struct btrfs_delayed_extent_op *extent_op)
2350 struct btrfs_key key;
2351 struct btrfs_path *path;
2352 struct btrfs_extent_item *ei;
2353 struct extent_buffer *leaf;
2357 int metadata = !extent_op->is_data;
2362 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2365 path = btrfs_alloc_path();
2369 key.objectid = node->bytenr;
2372 key.type = BTRFS_METADATA_ITEM_KEY;
2373 key.offset = extent_op->level;
2375 key.type = BTRFS_EXTENT_ITEM_KEY;
2376 key.offset = node->num_bytes;
2380 path->reada = READA_FORWARD;
2381 path->leave_spinning = 1;
2382 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2389 if (path->slots[0] > 0) {
2391 btrfs_item_key_to_cpu(path->nodes[0], &key,
2393 if (key.objectid == node->bytenr &&
2394 key.type == BTRFS_EXTENT_ITEM_KEY &&
2395 key.offset == node->num_bytes)
2399 btrfs_release_path(path);
2402 key.objectid = node->bytenr;
2403 key.offset = node->num_bytes;
2404 key.type = BTRFS_EXTENT_ITEM_KEY;
2413 leaf = path->nodes[0];
2414 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2415 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2416 if (item_size < sizeof(*ei)) {
2417 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2422 leaf = path->nodes[0];
2423 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2426 BUG_ON(item_size < sizeof(*ei));
2427 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2428 __run_delayed_extent_op(extent_op, leaf, ei);
2430 btrfs_mark_buffer_dirty(leaf);
2432 btrfs_free_path(path);
2436 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2437 struct btrfs_fs_info *fs_info,
2438 struct btrfs_delayed_ref_node *node,
2439 struct btrfs_delayed_extent_op *extent_op,
2440 int insert_reserved)
2443 struct btrfs_delayed_tree_ref *ref;
2444 struct btrfs_key ins;
2447 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2449 ref = btrfs_delayed_node_to_tree_ref(node);
2450 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2452 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2453 parent = ref->parent;
2454 ref_root = ref->root;
2456 ins.objectid = node->bytenr;
2457 if (skinny_metadata) {
2458 ins.offset = ref->level;
2459 ins.type = BTRFS_METADATA_ITEM_KEY;
2461 ins.offset = node->num_bytes;
2462 ins.type = BTRFS_EXTENT_ITEM_KEY;
2465 if (node->ref_mod != 1) {
2467 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2468 node->bytenr, node->ref_mod, node->action, ref_root,
2472 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2473 BUG_ON(!extent_op || !extent_op->update_flags);
2474 ret = alloc_reserved_tree_block(trans, fs_info,
2476 extent_op->flags_to_set,
2479 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2480 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2484 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2485 ret = __btrfs_free_extent(trans, fs_info, node,
2487 ref->level, 0, 1, extent_op);
2494 /* helper function to actually process a single delayed ref entry */
2495 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2496 struct btrfs_fs_info *fs_info,
2497 struct btrfs_delayed_ref_node *node,
2498 struct btrfs_delayed_extent_op *extent_op,
2499 int insert_reserved)
2503 if (trans->aborted) {
2504 if (insert_reserved)
2505 btrfs_pin_extent(fs_info, node->bytenr,
2506 node->num_bytes, 1);
2510 if (btrfs_delayed_ref_is_head(node)) {
2511 struct btrfs_delayed_ref_head *head;
2513 * we've hit the end of the chain and we were supposed
2514 * to insert this extent into the tree. But, it got
2515 * deleted before we ever needed to insert it, so all
2516 * we have to do is clean up the accounting
2519 head = btrfs_delayed_node_to_head(node);
2520 trace_run_delayed_ref_head(fs_info, node, head, node->action);
2522 if (head->total_ref_mod < 0) {
2523 struct btrfs_block_group_cache *cache;
2525 cache = btrfs_lookup_block_group(fs_info, node->bytenr);
2527 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2529 btrfs_put_block_group(cache);
2532 if (insert_reserved) {
2533 btrfs_pin_extent(fs_info, node->bytenr,
2534 node->num_bytes, 1);
2535 if (head->is_data) {
2536 ret = btrfs_del_csums(trans, fs_info,
2542 /* Also free its reserved qgroup space */
2543 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2544 head->qgroup_reserved);
2548 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2549 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2550 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2552 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2553 node->type == BTRFS_SHARED_DATA_REF_KEY)
2554 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2561 static inline struct btrfs_delayed_ref_node *
2562 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2564 struct btrfs_delayed_ref_node *ref;
2566 if (list_empty(&head->ref_list))
2570 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2571 * This is to prevent a ref count from going down to zero, which deletes
2572 * the extent item from the extent tree, when there still are references
2573 * to add, which would fail because they would not find the extent item.
2575 if (!list_empty(&head->ref_add_list))
2576 return list_first_entry(&head->ref_add_list,
2577 struct btrfs_delayed_ref_node, add_list);
2579 ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
2581 ASSERT(list_empty(&ref->add_list));
2586 * Returns 0 on success or if called with an already aborted transaction.
2587 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2589 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2590 struct btrfs_fs_info *fs_info,
2593 struct btrfs_delayed_ref_root *delayed_refs;
2594 struct btrfs_delayed_ref_node *ref;
2595 struct btrfs_delayed_ref_head *locked_ref = NULL;
2596 struct btrfs_delayed_extent_op *extent_op;
2597 ktime_t start = ktime_get();
2599 unsigned long count = 0;
2600 unsigned long actual_count = 0;
2601 int must_insert_reserved = 0;
2603 delayed_refs = &trans->transaction->delayed_refs;
2609 spin_lock(&delayed_refs->lock);
2610 locked_ref = btrfs_select_ref_head(trans);
2612 spin_unlock(&delayed_refs->lock);
2616 /* grab the lock that says we are going to process
2617 * all the refs for this head */
2618 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2619 spin_unlock(&delayed_refs->lock);
2621 * we may have dropped the spin lock to get the head
2622 * mutex lock, and that might have given someone else
2623 * time to free the head. If that's true, it has been
2624 * removed from our list and we can move on.
2626 if (ret == -EAGAIN) {
2634 * We need to try and merge add/drops of the same ref since we
2635 * can run into issues with relocate dropping the implicit ref
2636 * and then it being added back again before the drop can
2637 * finish. If we merged anything we need to re-loop so we can
2639 * Or we can get node references of the same type that weren't
2640 * merged when created due to bumps in the tree mod seq, and
2641 * we need to merge them to prevent adding an inline extent
2642 * backref before dropping it (triggering a BUG_ON at
2643 * insert_inline_extent_backref()).
2645 spin_lock(&locked_ref->lock);
2646 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2650 * locked_ref is the head node, so we have to go one
2651 * node back for any delayed ref updates
2653 ref = select_delayed_ref(locked_ref);
2655 if (ref && ref->seq &&
2656 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2657 spin_unlock(&locked_ref->lock);
2658 spin_lock(&delayed_refs->lock);
2659 locked_ref->processing = 0;
2660 delayed_refs->num_heads_ready++;
2661 spin_unlock(&delayed_refs->lock);
2662 btrfs_delayed_ref_unlock(locked_ref);
2670 * record the must insert reserved flag before we
2671 * drop the spin lock.
2673 must_insert_reserved = locked_ref->must_insert_reserved;
2674 locked_ref->must_insert_reserved = 0;
2676 extent_op = locked_ref->extent_op;
2677 locked_ref->extent_op = NULL;
2682 /* All delayed refs have been processed, Go ahead
2683 * and send the head node to run_one_delayed_ref,
2684 * so that any accounting fixes can happen
2686 ref = &locked_ref->node;
2688 if (extent_op && must_insert_reserved) {
2689 btrfs_free_delayed_extent_op(extent_op);
2694 spin_unlock(&locked_ref->lock);
2695 ret = run_delayed_extent_op(trans, fs_info,
2697 btrfs_free_delayed_extent_op(extent_op);
2701 * Need to reset must_insert_reserved if
2702 * there was an error so the abort stuff
2703 * can cleanup the reserved space
2706 if (must_insert_reserved)
2707 locked_ref->must_insert_reserved = 1;
2708 spin_lock(&delayed_refs->lock);
2709 locked_ref->processing = 0;
2710 delayed_refs->num_heads_ready++;
2711 spin_unlock(&delayed_refs->lock);
2712 btrfs_debug(fs_info,
2713 "run_delayed_extent_op returned %d",
2715 btrfs_delayed_ref_unlock(locked_ref);
2722 * Need to drop our head ref lock and re-acquire the
2723 * delayed ref lock and then re-check to make sure
2726 spin_unlock(&locked_ref->lock);
2727 spin_lock(&delayed_refs->lock);
2728 spin_lock(&locked_ref->lock);
2729 if (!list_empty(&locked_ref->ref_list) ||
2730 locked_ref->extent_op) {
2731 spin_unlock(&locked_ref->lock);
2732 spin_unlock(&delayed_refs->lock);
2736 delayed_refs->num_heads--;
2737 rb_erase(&locked_ref->href_node,
2738 &delayed_refs->href_root);
2739 spin_unlock(&delayed_refs->lock);
2743 list_del(&ref->list);
2744 if (!list_empty(&ref->add_list))
2745 list_del(&ref->add_list);
2747 atomic_dec(&delayed_refs->num_entries);
2749 if (!btrfs_delayed_ref_is_head(ref)) {
2751 * when we play the delayed ref, also correct the
2754 switch (ref->action) {
2755 case BTRFS_ADD_DELAYED_REF:
2756 case BTRFS_ADD_DELAYED_EXTENT:
2757 locked_ref->node.ref_mod -= ref->ref_mod;
2759 case BTRFS_DROP_DELAYED_REF:
2760 locked_ref->node.ref_mod += ref->ref_mod;
2766 spin_unlock(&locked_ref->lock);
2768 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2769 must_insert_reserved);
2771 btrfs_free_delayed_extent_op(extent_op);
2773 spin_lock(&delayed_refs->lock);
2774 locked_ref->processing = 0;
2775 delayed_refs->num_heads_ready++;
2776 spin_unlock(&delayed_refs->lock);
2777 btrfs_delayed_ref_unlock(locked_ref);
2778 btrfs_put_delayed_ref(ref);
2779 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2785 * If this node is a head, that means all the refs in this head
2786 * have been dealt with, and we will pick the next head to deal
2787 * with, so we must unlock the head and drop it from the cluster
2788 * list before we release it.
2790 if (btrfs_delayed_ref_is_head(ref)) {
2791 if (locked_ref->is_data &&
2792 locked_ref->total_ref_mod < 0) {
2793 spin_lock(&delayed_refs->lock);
2794 delayed_refs->pending_csums -= ref->num_bytes;
2795 spin_unlock(&delayed_refs->lock);
2797 btrfs_delayed_ref_unlock(locked_ref);
2800 btrfs_put_delayed_ref(ref);
2806 * We don't want to include ref heads since we can have empty ref heads
2807 * and those will drastically skew our runtime down since we just do
2808 * accounting, no actual extent tree updates.
2810 if (actual_count > 0) {
2811 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2815 * We weigh the current average higher than our current runtime
2816 * to avoid large swings in the average.
2818 spin_lock(&delayed_refs->lock);
2819 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2820 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2821 spin_unlock(&delayed_refs->lock);
2826 #ifdef SCRAMBLE_DELAYED_REFS
2828 * Normally delayed refs get processed in ascending bytenr order. This
2829 * correlates in most cases to the order added. To expose dependencies on this
2830 * order, we start to process the tree in the middle instead of the beginning
2832 static u64 find_middle(struct rb_root *root)
2834 struct rb_node *n = root->rb_node;
2835 struct btrfs_delayed_ref_node *entry;
2838 u64 first = 0, last = 0;
2842 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2843 first = entry->bytenr;
2847 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2848 last = entry->bytenr;
2853 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2854 WARN_ON(!entry->in_tree);
2856 middle = entry->bytenr;
2869 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2873 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2874 sizeof(struct btrfs_extent_inline_ref));
2875 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2876 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2879 * We don't ever fill up leaves all the way so multiply by 2 just to be
2880 * closer to what we're really going to want to use.
2882 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2886 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2887 * would require to store the csums for that many bytes.
2889 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2892 u64 num_csums_per_leaf;
2895 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2896 num_csums_per_leaf = div64_u64(csum_size,
2897 (u64)btrfs_super_csum_size(fs_info->super_copy));
2898 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2899 num_csums += num_csums_per_leaf - 1;
2900 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2904 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2905 struct btrfs_fs_info *fs_info)
2907 struct btrfs_block_rsv *global_rsv;
2908 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2909 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2910 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2911 u64 num_bytes, num_dirty_bgs_bytes;
2914 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2915 num_heads = heads_to_leaves(fs_info, num_heads);
2917 num_bytes += (num_heads - 1) * fs_info->nodesize;
2919 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2921 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2923 global_rsv = &fs_info->global_block_rsv;
2926 * If we can't allocate any more chunks lets make sure we have _lots_ of
2927 * wiggle room since running delayed refs can create more delayed refs.
2929 if (global_rsv->space_info->full) {
2930 num_dirty_bgs_bytes <<= 1;
2934 spin_lock(&global_rsv->lock);
2935 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2937 spin_unlock(&global_rsv->lock);
2941 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2942 struct btrfs_fs_info *fs_info)
2945 atomic_read(&trans->transaction->delayed_refs.num_entries);
2950 avg_runtime = fs_info->avg_delayed_ref_runtime;
2951 val = num_entries * avg_runtime;
2952 if (val >= NSEC_PER_SEC)
2954 if (val >= NSEC_PER_SEC / 2)
2957 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2960 struct async_delayed_refs {
2961 struct btrfs_root *root;
2966 struct completion wait;
2967 struct btrfs_work work;
2970 static inline struct async_delayed_refs *
2971 to_async_delayed_refs(struct btrfs_work *work)
2973 return container_of(work, struct async_delayed_refs, work);
2976 static void delayed_ref_async_start(struct btrfs_work *work)
2978 struct async_delayed_refs *async = to_async_delayed_refs(work);
2979 struct btrfs_trans_handle *trans;
2980 struct btrfs_fs_info *fs_info = async->root->fs_info;
2983 /* if the commit is already started, we don't need to wait here */
2984 if (btrfs_transaction_blocked(fs_info))
2987 trans = btrfs_join_transaction(async->root);
2988 if (IS_ERR(trans)) {
2989 async->error = PTR_ERR(trans);
2994 * trans->sync means that when we call end_transaction, we won't
2995 * wait on delayed refs
2999 /* Don't bother flushing if we got into a different transaction */
3000 if (trans->transid > async->transid)
3003 ret = btrfs_run_delayed_refs(trans, fs_info, async->count);
3007 ret = btrfs_end_transaction(trans);
3008 if (ret && !async->error)
3012 complete(&async->wait);
3017 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
3018 unsigned long count, u64 transid, int wait)
3020 struct async_delayed_refs *async;
3023 async = kmalloc(sizeof(*async), GFP_NOFS);
3027 async->root = fs_info->tree_root;
3028 async->count = count;
3030 async->transid = transid;
3035 init_completion(&async->wait);
3037 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3038 delayed_ref_async_start, NULL, NULL);
3040 btrfs_queue_work(fs_info->extent_workers, &async->work);
3043 wait_for_completion(&async->wait);
3052 * this starts processing the delayed reference count updates and
3053 * extent insertions we have queued up so far. count can be
3054 * 0, which means to process everything in the tree at the start
3055 * of the run (but not newly added entries), or it can be some target
3056 * number you'd like to process.
3058 * Returns 0 on success or if called with an aborted transaction
3059 * Returns <0 on error and aborts the transaction
3061 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3062 struct btrfs_fs_info *fs_info, unsigned long count)
3064 struct rb_node *node;
3065 struct btrfs_delayed_ref_root *delayed_refs;
3066 struct btrfs_delayed_ref_head *head;
3068 int run_all = count == (unsigned long)-1;
3069 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3071 /* We'll clean this up in btrfs_cleanup_transaction */
3075 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3078 delayed_refs = &trans->transaction->delayed_refs;
3080 count = atomic_read(&delayed_refs->num_entries) * 2;
3083 #ifdef SCRAMBLE_DELAYED_REFS
3084 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3086 trans->can_flush_pending_bgs = false;
3087 ret = __btrfs_run_delayed_refs(trans, fs_info, count);
3089 btrfs_abort_transaction(trans, ret);
3094 if (!list_empty(&trans->new_bgs))
3095 btrfs_create_pending_block_groups(trans, fs_info);
3097 spin_lock(&delayed_refs->lock);
3098 node = rb_first(&delayed_refs->href_root);
3100 spin_unlock(&delayed_refs->lock);
3105 head = rb_entry(node, struct btrfs_delayed_ref_head,
3107 if (btrfs_delayed_ref_is_head(&head->node)) {
3108 struct btrfs_delayed_ref_node *ref;
3111 refcount_inc(&ref->refs);
3113 spin_unlock(&delayed_refs->lock);
3115 * Mutex was contended, block until it's
3116 * released and try again
3118 mutex_lock(&head->mutex);
3119 mutex_unlock(&head->mutex);
3121 btrfs_put_delayed_ref(ref);
3127 node = rb_next(node);
3129 spin_unlock(&delayed_refs->lock);
3134 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3138 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3139 struct btrfs_fs_info *fs_info,
3140 u64 bytenr, u64 num_bytes, u64 flags,
3141 int level, int is_data)
3143 struct btrfs_delayed_extent_op *extent_op;
3146 extent_op = btrfs_alloc_delayed_extent_op();
3150 extent_op->flags_to_set = flags;
3151 extent_op->update_flags = true;
3152 extent_op->update_key = false;
3153 extent_op->is_data = is_data ? true : false;
3154 extent_op->level = level;
3156 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3157 num_bytes, extent_op);
3159 btrfs_free_delayed_extent_op(extent_op);
3163 static noinline int check_delayed_ref(struct btrfs_root *root,
3164 struct btrfs_path *path,
3165 u64 objectid, u64 offset, u64 bytenr)
3167 struct btrfs_delayed_ref_head *head;
3168 struct btrfs_delayed_ref_node *ref;
3169 struct btrfs_delayed_data_ref *data_ref;
3170 struct btrfs_delayed_ref_root *delayed_refs;
3171 struct btrfs_transaction *cur_trans;
3174 cur_trans = root->fs_info->running_transaction;
3178 delayed_refs = &cur_trans->delayed_refs;
3179 spin_lock(&delayed_refs->lock);
3180 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3182 spin_unlock(&delayed_refs->lock);
3186 if (!mutex_trylock(&head->mutex)) {
3187 refcount_inc(&head->node.refs);
3188 spin_unlock(&delayed_refs->lock);
3190 btrfs_release_path(path);
3193 * Mutex was contended, block until it's released and let
3196 mutex_lock(&head->mutex);
3197 mutex_unlock(&head->mutex);
3198 btrfs_put_delayed_ref(&head->node);
3201 spin_unlock(&delayed_refs->lock);
3203 spin_lock(&head->lock);
3204 list_for_each_entry(ref, &head->ref_list, list) {
3205 /* If it's a shared ref we know a cross reference exists */
3206 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3211 data_ref = btrfs_delayed_node_to_data_ref(ref);
3214 * If our ref doesn't match the one we're currently looking at
3215 * then we have a cross reference.
3217 if (data_ref->root != root->root_key.objectid ||
3218 data_ref->objectid != objectid ||
3219 data_ref->offset != offset) {
3224 spin_unlock(&head->lock);
3225 mutex_unlock(&head->mutex);
3229 static noinline int check_committed_ref(struct btrfs_root *root,
3230 struct btrfs_path *path,
3231 u64 objectid, u64 offset, u64 bytenr)
3233 struct btrfs_fs_info *fs_info = root->fs_info;
3234 struct btrfs_root *extent_root = fs_info->extent_root;
3235 struct extent_buffer *leaf;
3236 struct btrfs_extent_data_ref *ref;
3237 struct btrfs_extent_inline_ref *iref;
3238 struct btrfs_extent_item *ei;
3239 struct btrfs_key key;
3244 key.objectid = bytenr;
3245 key.offset = (u64)-1;
3246 key.type = BTRFS_EXTENT_ITEM_KEY;
3248 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3251 BUG_ON(ret == 0); /* Corruption */
3254 if (path->slots[0] == 0)
3258 leaf = path->nodes[0];
3259 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3261 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3265 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3266 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3267 if (item_size < sizeof(*ei)) {
3268 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3272 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3274 if (item_size != sizeof(*ei) +
3275 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3278 if (btrfs_extent_generation(leaf, ei) <=
3279 btrfs_root_last_snapshot(&root->root_item))
3282 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3284 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3285 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3288 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3289 if (btrfs_extent_refs(leaf, ei) !=
3290 btrfs_extent_data_ref_count(leaf, ref) ||
3291 btrfs_extent_data_ref_root(leaf, ref) !=
3292 root->root_key.objectid ||
3293 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3294 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3302 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3305 struct btrfs_path *path;
3309 path = btrfs_alloc_path();
3314 ret = check_committed_ref(root, path, objectid,
3316 if (ret && ret != -ENOENT)
3319 ret2 = check_delayed_ref(root, path, objectid,
3321 } while (ret2 == -EAGAIN);
3323 if (ret2 && ret2 != -ENOENT) {
3328 if (ret != -ENOENT || ret2 != -ENOENT)
3331 btrfs_free_path(path);
3332 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3337 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3338 struct btrfs_root *root,
3339 struct extent_buffer *buf,
3340 int full_backref, int inc)
3342 struct btrfs_fs_info *fs_info = root->fs_info;
3348 struct btrfs_key key;
3349 struct btrfs_file_extent_item *fi;
3353 int (*process_func)(struct btrfs_trans_handle *,
3354 struct btrfs_fs_info *,
3355 u64, u64, u64, u64, u64, u64);
3358 if (btrfs_is_testing(fs_info))
3361 ref_root = btrfs_header_owner(buf);
3362 nritems = btrfs_header_nritems(buf);
3363 level = btrfs_header_level(buf);
3365 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3369 process_func = btrfs_inc_extent_ref;
3371 process_func = btrfs_free_extent;
3374 parent = buf->start;
3378 for (i = 0; i < nritems; i++) {
3380 btrfs_item_key_to_cpu(buf, &key, i);
3381 if (key.type != BTRFS_EXTENT_DATA_KEY)
3383 fi = btrfs_item_ptr(buf, i,
3384 struct btrfs_file_extent_item);
3385 if (btrfs_file_extent_type(buf, fi) ==
3386 BTRFS_FILE_EXTENT_INLINE)
3388 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3392 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3393 key.offset -= btrfs_file_extent_offset(buf, fi);
3394 ret = process_func(trans, fs_info, bytenr, num_bytes,
3395 parent, ref_root, key.objectid,
3400 bytenr = btrfs_node_blockptr(buf, i);
3401 num_bytes = fs_info->nodesize;
3402 ret = process_func(trans, fs_info, bytenr, num_bytes,
3403 parent, ref_root, level - 1, 0);
3413 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3414 struct extent_buffer *buf, int full_backref)
3416 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3419 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3420 struct extent_buffer *buf, int full_backref)
3422 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3425 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3426 struct btrfs_fs_info *fs_info,
3427 struct btrfs_path *path,
3428 struct btrfs_block_group_cache *cache)
3431 struct btrfs_root *extent_root = fs_info->extent_root;
3433 struct extent_buffer *leaf;
3435 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3442 leaf = path->nodes[0];
3443 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3444 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3445 btrfs_mark_buffer_dirty(leaf);
3447 btrfs_release_path(path);
3452 static struct btrfs_block_group_cache *
3453 next_block_group(struct btrfs_fs_info *fs_info,
3454 struct btrfs_block_group_cache *cache)
3456 struct rb_node *node;
3458 spin_lock(&fs_info->block_group_cache_lock);
3460 /* If our block group was removed, we need a full search. */
3461 if (RB_EMPTY_NODE(&cache->cache_node)) {
3462 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3464 spin_unlock(&fs_info->block_group_cache_lock);
3465 btrfs_put_block_group(cache);
3466 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3468 node = rb_next(&cache->cache_node);
3469 btrfs_put_block_group(cache);
3471 cache = rb_entry(node, struct btrfs_block_group_cache,
3473 btrfs_get_block_group(cache);
3476 spin_unlock(&fs_info->block_group_cache_lock);
3480 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3481 struct btrfs_trans_handle *trans,
3482 struct btrfs_path *path)
3484 struct btrfs_fs_info *fs_info = block_group->fs_info;
3485 struct btrfs_root *root = fs_info->tree_root;
3486 struct inode *inode = NULL;
3487 struct extent_changeset *data_reserved = NULL;
3489 int dcs = BTRFS_DC_ERROR;
3495 * If this block group is smaller than 100 megs don't bother caching the
3498 if (block_group->key.offset < (100 * SZ_1M)) {
3499 spin_lock(&block_group->lock);
3500 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3501 spin_unlock(&block_group->lock);
3508 inode = lookup_free_space_inode(fs_info, block_group, path);
3509 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3510 ret = PTR_ERR(inode);
3511 btrfs_release_path(path);
3515 if (IS_ERR(inode)) {
3519 if (block_group->ro)
3522 ret = create_free_space_inode(fs_info, trans, block_group,
3529 /* We've already setup this transaction, go ahead and exit */
3530 if (block_group->cache_generation == trans->transid &&
3531 i_size_read(inode)) {
3532 dcs = BTRFS_DC_SETUP;
3537 * We want to set the generation to 0, that way if anything goes wrong
3538 * from here on out we know not to trust this cache when we load up next
3541 BTRFS_I(inode)->generation = 0;
3542 ret = btrfs_update_inode(trans, root, inode);
3545 * So theoretically we could recover from this, simply set the
3546 * super cache generation to 0 so we know to invalidate the
3547 * cache, but then we'd have to keep track of the block groups
3548 * that fail this way so we know we _have_ to reset this cache
3549 * before the next commit or risk reading stale cache. So to
3550 * limit our exposure to horrible edge cases lets just abort the
3551 * transaction, this only happens in really bad situations
3554 btrfs_abort_transaction(trans, ret);
3559 if (i_size_read(inode) > 0) {
3560 ret = btrfs_check_trunc_cache_free_space(fs_info,
3561 &fs_info->global_block_rsv);
3565 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3570 spin_lock(&block_group->lock);
3571 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3572 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3574 * don't bother trying to write stuff out _if_
3575 * a) we're not cached,
3576 * b) we're with nospace_cache mount option,
3577 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3579 dcs = BTRFS_DC_WRITTEN;
3580 spin_unlock(&block_group->lock);
3583 spin_unlock(&block_group->lock);
3586 * We hit an ENOSPC when setting up the cache in this transaction, just
3587 * skip doing the setup, we've already cleared the cache so we're safe.
3589 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3595 * Try to preallocate enough space based on how big the block group is.
3596 * Keep in mind this has to include any pinned space which could end up
3597 * taking up quite a bit since it's not folded into the other space
3600 num_pages = div_u64(block_group->key.offset, SZ_256M);
3605 num_pages *= PAGE_SIZE;
3607 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3611 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3612 num_pages, num_pages,
3615 * Our cache requires contiguous chunks so that we don't modify a bunch
3616 * of metadata or split extents when writing the cache out, which means
3617 * we can enospc if we are heavily fragmented in addition to just normal
3618 * out of space conditions. So if we hit this just skip setting up any
3619 * other block groups for this transaction, maybe we'll unpin enough
3620 * space the next time around.
3623 dcs = BTRFS_DC_SETUP;
3624 else if (ret == -ENOSPC)
3625 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3630 btrfs_release_path(path);
3632 spin_lock(&block_group->lock);
3633 if (!ret && dcs == BTRFS_DC_SETUP)
3634 block_group->cache_generation = trans->transid;
3635 block_group->disk_cache_state = dcs;
3636 spin_unlock(&block_group->lock);
3638 extent_changeset_free(data_reserved);
3642 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3643 struct btrfs_fs_info *fs_info)
3645 struct btrfs_block_group_cache *cache, *tmp;
3646 struct btrfs_transaction *cur_trans = trans->transaction;
3647 struct btrfs_path *path;
3649 if (list_empty(&cur_trans->dirty_bgs) ||
3650 !btrfs_test_opt(fs_info, SPACE_CACHE))
3653 path = btrfs_alloc_path();
3657 /* Could add new block groups, use _safe just in case */
3658 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3660 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3661 cache_save_setup(cache, trans, path);
3664 btrfs_free_path(path);
3669 * transaction commit does final block group cache writeback during a
3670 * critical section where nothing is allowed to change the FS. This is
3671 * required in order for the cache to actually match the block group,
3672 * but can introduce a lot of latency into the commit.
3674 * So, btrfs_start_dirty_block_groups is here to kick off block group
3675 * cache IO. There's a chance we'll have to redo some of it if the
3676 * block group changes again during the commit, but it greatly reduces
3677 * the commit latency by getting rid of the easy block groups while
3678 * we're still allowing others to join the commit.
3680 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3681 struct btrfs_fs_info *fs_info)
3683 struct btrfs_block_group_cache *cache;
3684 struct btrfs_transaction *cur_trans = trans->transaction;
3687 struct btrfs_path *path = NULL;
3689 struct list_head *io = &cur_trans->io_bgs;
3690 int num_started = 0;
3693 spin_lock(&cur_trans->dirty_bgs_lock);
3694 if (list_empty(&cur_trans->dirty_bgs)) {
3695 spin_unlock(&cur_trans->dirty_bgs_lock);
3698 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3699 spin_unlock(&cur_trans->dirty_bgs_lock);
3703 * make sure all the block groups on our dirty list actually
3706 btrfs_create_pending_block_groups(trans, fs_info);
3709 path = btrfs_alloc_path();
3715 * cache_write_mutex is here only to save us from balance or automatic
3716 * removal of empty block groups deleting this block group while we are
3717 * writing out the cache
3719 mutex_lock(&trans->transaction->cache_write_mutex);
3720 while (!list_empty(&dirty)) {
3721 cache = list_first_entry(&dirty,
3722 struct btrfs_block_group_cache,
3725 * this can happen if something re-dirties a block
3726 * group that is already under IO. Just wait for it to
3727 * finish and then do it all again
3729 if (!list_empty(&cache->io_list)) {
3730 list_del_init(&cache->io_list);
3731 btrfs_wait_cache_io(trans, cache, path);
3732 btrfs_put_block_group(cache);
3737 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3738 * if it should update the cache_state. Don't delete
3739 * until after we wait.
3741 * Since we're not running in the commit critical section
3742 * we need the dirty_bgs_lock to protect from update_block_group
3744 spin_lock(&cur_trans->dirty_bgs_lock);
3745 list_del_init(&cache->dirty_list);
3746 spin_unlock(&cur_trans->dirty_bgs_lock);
3750 cache_save_setup(cache, trans, path);
3752 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3753 cache->io_ctl.inode = NULL;
3754 ret = btrfs_write_out_cache(fs_info, trans,
3756 if (ret == 0 && cache->io_ctl.inode) {
3761 * the cache_write_mutex is protecting
3764 list_add_tail(&cache->io_list, io);
3767 * if we failed to write the cache, the
3768 * generation will be bad and life goes on
3774 ret = write_one_cache_group(trans, fs_info,
3777 * Our block group might still be attached to the list
3778 * of new block groups in the transaction handle of some
3779 * other task (struct btrfs_trans_handle->new_bgs). This
3780 * means its block group item isn't yet in the extent
3781 * tree. If this happens ignore the error, as we will
3782 * try again later in the critical section of the
3783 * transaction commit.
3785 if (ret == -ENOENT) {
3787 spin_lock(&cur_trans->dirty_bgs_lock);
3788 if (list_empty(&cache->dirty_list)) {
3789 list_add_tail(&cache->dirty_list,
3790 &cur_trans->dirty_bgs);
3791 btrfs_get_block_group(cache);
3793 spin_unlock(&cur_trans->dirty_bgs_lock);
3795 btrfs_abort_transaction(trans, ret);
3799 /* if its not on the io list, we need to put the block group */
3801 btrfs_put_block_group(cache);
3807 * Avoid blocking other tasks for too long. It might even save
3808 * us from writing caches for block groups that are going to be
3811 mutex_unlock(&trans->transaction->cache_write_mutex);
3812 mutex_lock(&trans->transaction->cache_write_mutex);
3814 mutex_unlock(&trans->transaction->cache_write_mutex);
3817 * go through delayed refs for all the stuff we've just kicked off
3818 * and then loop back (just once)
3820 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
3821 if (!ret && loops == 0) {
3823 spin_lock(&cur_trans->dirty_bgs_lock);
3824 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3826 * dirty_bgs_lock protects us from concurrent block group
3827 * deletes too (not just cache_write_mutex).
3829 if (!list_empty(&dirty)) {
3830 spin_unlock(&cur_trans->dirty_bgs_lock);
3833 spin_unlock(&cur_trans->dirty_bgs_lock);
3834 } else if (ret < 0) {
3835 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3838 btrfs_free_path(path);
3842 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3843 struct btrfs_fs_info *fs_info)
3845 struct btrfs_block_group_cache *cache;
3846 struct btrfs_transaction *cur_trans = trans->transaction;
3849 struct btrfs_path *path;
3850 struct list_head *io = &cur_trans->io_bgs;
3851 int num_started = 0;
3853 path = btrfs_alloc_path();
3858 * Even though we are in the critical section of the transaction commit,
3859 * we can still have concurrent tasks adding elements to this
3860 * transaction's list of dirty block groups. These tasks correspond to
3861 * endio free space workers started when writeback finishes for a
3862 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3863 * allocate new block groups as a result of COWing nodes of the root
3864 * tree when updating the free space inode. The writeback for the space
3865 * caches is triggered by an earlier call to
3866 * btrfs_start_dirty_block_groups() and iterations of the following
3868 * Also we want to do the cache_save_setup first and then run the
3869 * delayed refs to make sure we have the best chance at doing this all
3872 spin_lock(&cur_trans->dirty_bgs_lock);
3873 while (!list_empty(&cur_trans->dirty_bgs)) {
3874 cache = list_first_entry(&cur_trans->dirty_bgs,
3875 struct btrfs_block_group_cache,
3879 * this can happen if cache_save_setup re-dirties a block
3880 * group that is already under IO. Just wait for it to
3881 * finish and then do it all again
3883 if (!list_empty(&cache->io_list)) {
3884 spin_unlock(&cur_trans->dirty_bgs_lock);
3885 list_del_init(&cache->io_list);
3886 btrfs_wait_cache_io(trans, cache, path);
3887 btrfs_put_block_group(cache);
3888 spin_lock(&cur_trans->dirty_bgs_lock);
3892 * don't remove from the dirty list until after we've waited
3895 list_del_init(&cache->dirty_list);
3896 spin_unlock(&cur_trans->dirty_bgs_lock);
3899 cache_save_setup(cache, trans, path);
3902 ret = btrfs_run_delayed_refs(trans, fs_info,
3903 (unsigned long) -1);
3905 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3906 cache->io_ctl.inode = NULL;
3907 ret = btrfs_write_out_cache(fs_info, trans,
3909 if (ret == 0 && cache->io_ctl.inode) {
3912 list_add_tail(&cache->io_list, io);
3915 * if we failed to write the cache, the
3916 * generation will be bad and life goes on
3922 ret = write_one_cache_group(trans, fs_info,
3925 * One of the free space endio workers might have
3926 * created a new block group while updating a free space
3927 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3928 * and hasn't released its transaction handle yet, in
3929 * which case the new block group is still attached to
3930 * its transaction handle and its creation has not
3931 * finished yet (no block group item in the extent tree
3932 * yet, etc). If this is the case, wait for all free
3933 * space endio workers to finish and retry. This is a
3934 * a very rare case so no need for a more efficient and
3937 if (ret == -ENOENT) {
3938 wait_event(cur_trans->writer_wait,
3939 atomic_read(&cur_trans->num_writers) == 1);
3940 ret = write_one_cache_group(trans, fs_info,
3944 btrfs_abort_transaction(trans, ret);
3947 /* if its not on the io list, we need to put the block group */
3949 btrfs_put_block_group(cache);
3950 spin_lock(&cur_trans->dirty_bgs_lock);
3952 spin_unlock(&cur_trans->dirty_bgs_lock);
3954 while (!list_empty(io)) {
3955 cache = list_first_entry(io, struct btrfs_block_group_cache,
3957 list_del_init(&cache->io_list);
3958 btrfs_wait_cache_io(trans, cache, path);
3959 btrfs_put_block_group(cache);
3962 btrfs_free_path(path);
3966 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3968 struct btrfs_block_group_cache *block_group;
3971 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3972 if (!block_group || block_group->ro)
3975 btrfs_put_block_group(block_group);
3979 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3981 struct btrfs_block_group_cache *bg;
3984 bg = btrfs_lookup_block_group(fs_info, bytenr);
3988 spin_lock(&bg->lock);
3992 atomic_inc(&bg->nocow_writers);
3993 spin_unlock(&bg->lock);
3995 /* no put on block group, done by btrfs_dec_nocow_writers */
3997 btrfs_put_block_group(bg);
4003 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
4005 struct btrfs_block_group_cache *bg;
4007 bg = btrfs_lookup_block_group(fs_info, bytenr);
4009 if (atomic_dec_and_test(&bg->nocow_writers))
4010 wake_up_atomic_t(&bg->nocow_writers);
4012 * Once for our lookup and once for the lookup done by a previous call
4013 * to btrfs_inc_nocow_writers()
4015 btrfs_put_block_group(bg);
4016 btrfs_put_block_group(bg);
4019 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
4025 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4027 wait_on_atomic_t(&bg->nocow_writers,
4028 btrfs_wait_nocow_writers_atomic_t,
4029 TASK_UNINTERRUPTIBLE);
4032 static const char *alloc_name(u64 flags)
4035 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4037 case BTRFS_BLOCK_GROUP_METADATA:
4039 case BTRFS_BLOCK_GROUP_DATA:
4041 case BTRFS_BLOCK_GROUP_SYSTEM:
4045 return "invalid-combination";
4049 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4050 struct btrfs_space_info **new)
4053 struct btrfs_space_info *space_info;
4057 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4061 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4068 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4069 INIT_LIST_HEAD(&space_info->block_groups[i]);
4070 init_rwsem(&space_info->groups_sem);
4071 spin_lock_init(&space_info->lock);
4072 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4073 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4074 init_waitqueue_head(&space_info->wait);
4075 INIT_LIST_HEAD(&space_info->ro_bgs);
4076 INIT_LIST_HEAD(&space_info->tickets);
4077 INIT_LIST_HEAD(&space_info->priority_tickets);
4079 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4080 info->space_info_kobj, "%s",
4081 alloc_name(space_info->flags));
4083 percpu_counter_destroy(&space_info->total_bytes_pinned);
4089 list_add_rcu(&space_info->list, &info->space_info);
4090 if (flags & BTRFS_BLOCK_GROUP_DATA)
4091 info->data_sinfo = space_info;
4096 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4097 u64 total_bytes, u64 bytes_used,
4099 struct btrfs_space_info **space_info)
4101 struct btrfs_space_info *found;
4104 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4105 BTRFS_BLOCK_GROUP_RAID10))
4110 found = __find_space_info(info, flags);
4112 spin_lock(&found->lock);
4113 found->total_bytes += total_bytes;
4114 found->disk_total += total_bytes * factor;
4115 found->bytes_used += bytes_used;
4116 found->disk_used += bytes_used * factor;
4117 found->bytes_readonly += bytes_readonly;
4118 if (total_bytes > 0)
4120 space_info_add_new_bytes(info, found, total_bytes -
4121 bytes_used - bytes_readonly);
4122 spin_unlock(&found->lock);
4123 *space_info = found;
4126 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4128 u64 extra_flags = chunk_to_extended(flags) &
4129 BTRFS_EXTENDED_PROFILE_MASK;
4131 write_seqlock(&fs_info->profiles_lock);
4132 if (flags & BTRFS_BLOCK_GROUP_DATA)
4133 fs_info->avail_data_alloc_bits |= extra_flags;
4134 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4135 fs_info->avail_metadata_alloc_bits |= extra_flags;
4136 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4137 fs_info->avail_system_alloc_bits |= extra_flags;
4138 write_sequnlock(&fs_info->profiles_lock);
4142 * returns target flags in extended format or 0 if restripe for this
4143 * chunk_type is not in progress
4145 * should be called with either volume_mutex or balance_lock held
4147 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4149 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4155 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4156 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4157 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4158 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4159 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4160 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4161 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4162 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4163 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4170 * @flags: available profiles in extended format (see ctree.h)
4172 * Returns reduced profile in chunk format. If profile changing is in
4173 * progress (either running or paused) picks the target profile (if it's
4174 * already available), otherwise falls back to plain reducing.
4176 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4178 u64 num_devices = fs_info->fs_devices->rw_devices;
4184 * see if restripe for this chunk_type is in progress, if so
4185 * try to reduce to the target profile
4187 spin_lock(&fs_info->balance_lock);
4188 target = get_restripe_target(fs_info, flags);
4190 /* pick target profile only if it's already available */
4191 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4192 spin_unlock(&fs_info->balance_lock);
4193 return extended_to_chunk(target);
4196 spin_unlock(&fs_info->balance_lock);
4198 /* First, mask out the RAID levels which aren't possible */
4199 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4200 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4201 allowed |= btrfs_raid_group[raid_type];
4205 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4206 allowed = BTRFS_BLOCK_GROUP_RAID6;
4207 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4208 allowed = BTRFS_BLOCK_GROUP_RAID5;
4209 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4210 allowed = BTRFS_BLOCK_GROUP_RAID10;
4211 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4212 allowed = BTRFS_BLOCK_GROUP_RAID1;
4213 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4214 allowed = BTRFS_BLOCK_GROUP_RAID0;
4216 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4218 return extended_to_chunk(flags | allowed);
4221 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4228 seq = read_seqbegin(&fs_info->profiles_lock);
4230 if (flags & BTRFS_BLOCK_GROUP_DATA)
4231 flags |= fs_info->avail_data_alloc_bits;
4232 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4233 flags |= fs_info->avail_system_alloc_bits;
4234 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4235 flags |= fs_info->avail_metadata_alloc_bits;
4236 } while (read_seqretry(&fs_info->profiles_lock, seq));
4238 return btrfs_reduce_alloc_profile(fs_info, flags);
4241 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4243 struct btrfs_fs_info *fs_info = root->fs_info;
4248 flags = BTRFS_BLOCK_GROUP_DATA;
4249 else if (root == fs_info->chunk_root)
4250 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4252 flags = BTRFS_BLOCK_GROUP_METADATA;
4254 ret = get_alloc_profile(fs_info, flags);
4258 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4260 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4263 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4265 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4268 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4270 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4273 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4274 bool may_use_included)
4277 return s_info->bytes_used + s_info->bytes_reserved +
4278 s_info->bytes_pinned + s_info->bytes_readonly +
4279 (may_use_included ? s_info->bytes_may_use : 0);
4282 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4284 struct btrfs_root *root = inode->root;
4285 struct btrfs_fs_info *fs_info = root->fs_info;
4286 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4289 int need_commit = 2;
4290 int have_pinned_space;
4292 /* make sure bytes are sectorsize aligned */
4293 bytes = ALIGN(bytes, fs_info->sectorsize);
4295 if (btrfs_is_free_space_inode(inode)) {
4297 ASSERT(current->journal_info);
4301 /* make sure we have enough space to handle the data first */
4302 spin_lock(&data_sinfo->lock);
4303 used = btrfs_space_info_used(data_sinfo, true);
4305 if (used + bytes > data_sinfo->total_bytes) {
4306 struct btrfs_trans_handle *trans;
4309 * if we don't have enough free bytes in this space then we need
4310 * to alloc a new chunk.
4312 if (!data_sinfo->full) {
4315 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4316 spin_unlock(&data_sinfo->lock);
4318 alloc_target = btrfs_data_alloc_profile(fs_info);
4320 * It is ugly that we don't call nolock join
4321 * transaction for the free space inode case here.
4322 * But it is safe because we only do the data space
4323 * reservation for the free space cache in the
4324 * transaction context, the common join transaction
4325 * just increase the counter of the current transaction
4326 * handler, doesn't try to acquire the trans_lock of
4329 trans = btrfs_join_transaction(root);
4331 return PTR_ERR(trans);
4333 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4334 CHUNK_ALLOC_NO_FORCE);
4335 btrfs_end_transaction(trans);
4340 have_pinned_space = 1;
4349 * If we don't have enough pinned space to deal with this
4350 * allocation, and no removed chunk in current transaction,
4351 * don't bother committing the transaction.
4353 have_pinned_space = percpu_counter_compare(
4354 &data_sinfo->total_bytes_pinned,
4355 used + bytes - data_sinfo->total_bytes);
4356 spin_unlock(&data_sinfo->lock);
4358 /* commit the current transaction and try again */
4361 !atomic_read(&fs_info->open_ioctl_trans)) {
4364 if (need_commit > 0) {
4365 btrfs_start_delalloc_roots(fs_info, 0, -1);
4366 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4370 trans = btrfs_join_transaction(root);
4372 return PTR_ERR(trans);
4373 if (have_pinned_space >= 0 ||
4374 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4375 &trans->transaction->flags) ||
4377 ret = btrfs_commit_transaction(trans);
4381 * The cleaner kthread might still be doing iput
4382 * operations. Wait for it to finish so that
4383 * more space is released.
4385 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4386 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4389 btrfs_end_transaction(trans);
4393 trace_btrfs_space_reservation(fs_info,
4394 "space_info:enospc",
4395 data_sinfo->flags, bytes, 1);
4398 data_sinfo->bytes_may_use += bytes;
4399 trace_btrfs_space_reservation(fs_info, "space_info",
4400 data_sinfo->flags, bytes, 1);
4401 spin_unlock(&data_sinfo->lock);
4406 int btrfs_check_data_free_space(struct inode *inode,
4407 struct extent_changeset **reserved, u64 start, u64 len)
4409 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4412 /* align the range */
4413 len = round_up(start + len, fs_info->sectorsize) -
4414 round_down(start, fs_info->sectorsize);
4415 start = round_down(start, fs_info->sectorsize);
4417 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4421 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4422 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4424 btrfs_free_reserved_data_space_noquota(inode, start, len);
4431 * Called if we need to clear a data reservation for this inode
4432 * Normally in a error case.
4434 * This one will *NOT* use accurate qgroup reserved space API, just for case
4435 * which we can't sleep and is sure it won't affect qgroup reserved space.
4436 * Like clear_bit_hook().
4438 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4441 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4442 struct btrfs_space_info *data_sinfo;
4444 /* Make sure the range is aligned to sectorsize */
4445 len = round_up(start + len, fs_info->sectorsize) -
4446 round_down(start, fs_info->sectorsize);
4447 start = round_down(start, fs_info->sectorsize);
4449 data_sinfo = fs_info->data_sinfo;
4450 spin_lock(&data_sinfo->lock);
4451 if (WARN_ON(data_sinfo->bytes_may_use < len))
4452 data_sinfo->bytes_may_use = 0;
4454 data_sinfo->bytes_may_use -= len;
4455 trace_btrfs_space_reservation(fs_info, "space_info",
4456 data_sinfo->flags, len, 0);
4457 spin_unlock(&data_sinfo->lock);
4461 * Called if we need to clear a data reservation for this inode
4462 * Normally in a error case.
4464 * This one will handle the per-inode data rsv map for accurate reserved
4467 void btrfs_free_reserved_data_space(struct inode *inode,
4468 struct extent_changeset *reserved, u64 start, u64 len)
4470 struct btrfs_root *root = BTRFS_I(inode)->root;
4472 /* Make sure the range is aligned to sectorsize */
4473 len = round_up(start + len, root->fs_info->sectorsize) -
4474 round_down(start, root->fs_info->sectorsize);
4475 start = round_down(start, root->fs_info->sectorsize);
4477 btrfs_free_reserved_data_space_noquota(inode, start, len);
4478 btrfs_qgroup_free_data(inode, reserved, start, len);
4481 static void force_metadata_allocation(struct btrfs_fs_info *info)
4483 struct list_head *head = &info->space_info;
4484 struct btrfs_space_info *found;
4487 list_for_each_entry_rcu(found, head, list) {
4488 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4489 found->force_alloc = CHUNK_ALLOC_FORCE;
4494 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4496 return (global->size << 1);
4499 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4500 struct btrfs_space_info *sinfo, int force)
4502 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4503 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4506 if (force == CHUNK_ALLOC_FORCE)
4510 * We need to take into account the global rsv because for all intents
4511 * and purposes it's used space. Don't worry about locking the
4512 * global_rsv, it doesn't change except when the transaction commits.
4514 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4515 bytes_used += calc_global_rsv_need_space(global_rsv);
4518 * in limited mode, we want to have some free space up to
4519 * about 1% of the FS size.
4521 if (force == CHUNK_ALLOC_LIMITED) {
4522 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4523 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4525 if (sinfo->total_bytes - bytes_used < thresh)
4529 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4534 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4538 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4539 BTRFS_BLOCK_GROUP_RAID0 |
4540 BTRFS_BLOCK_GROUP_RAID5 |
4541 BTRFS_BLOCK_GROUP_RAID6))
4542 num_dev = fs_info->fs_devices->rw_devices;
4543 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4546 num_dev = 1; /* DUP or single */
4552 * If @is_allocation is true, reserve space in the system space info necessary
4553 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4556 void check_system_chunk(struct btrfs_trans_handle *trans,
4557 struct btrfs_fs_info *fs_info, u64 type)
4559 struct btrfs_space_info *info;
4566 * Needed because we can end up allocating a system chunk and for an
4567 * atomic and race free space reservation in the chunk block reserve.
4569 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4571 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4572 spin_lock(&info->lock);
4573 left = info->total_bytes - btrfs_space_info_used(info, true);
4574 spin_unlock(&info->lock);
4576 num_devs = get_profile_num_devs(fs_info, type);
4578 /* num_devs device items to update and 1 chunk item to add or remove */
4579 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4580 btrfs_calc_trans_metadata_size(fs_info, 1);
4582 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4583 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4584 left, thresh, type);
4585 dump_space_info(fs_info, info, 0, 0);
4588 if (left < thresh) {
4589 u64 flags = btrfs_system_alloc_profile(fs_info);
4592 * Ignore failure to create system chunk. We might end up not
4593 * needing it, as we might not need to COW all nodes/leafs from
4594 * the paths we visit in the chunk tree (they were already COWed
4595 * or created in the current transaction for example).
4597 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4601 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4602 &fs_info->chunk_block_rsv,
4603 thresh, BTRFS_RESERVE_NO_FLUSH);
4605 trans->chunk_bytes_reserved += thresh;
4610 * If force is CHUNK_ALLOC_FORCE:
4611 * - return 1 if it successfully allocates a chunk,
4612 * - return errors including -ENOSPC otherwise.
4613 * If force is NOT CHUNK_ALLOC_FORCE:
4614 * - return 0 if it doesn't need to allocate a new chunk,
4615 * - return 1 if it successfully allocates a chunk,
4616 * - return errors including -ENOSPC otherwise.
4618 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4619 struct btrfs_fs_info *fs_info, u64 flags, int force)
4621 struct btrfs_space_info *space_info;
4622 int wait_for_alloc = 0;
4625 /* Don't re-enter if we're already allocating a chunk */
4626 if (trans->allocating_chunk)
4629 space_info = __find_space_info(fs_info, flags);
4631 ret = create_space_info(fs_info, flags, &space_info);
4637 spin_lock(&space_info->lock);
4638 if (force < space_info->force_alloc)
4639 force = space_info->force_alloc;
4640 if (space_info->full) {
4641 if (should_alloc_chunk(fs_info, space_info, force))
4645 spin_unlock(&space_info->lock);
4649 if (!should_alloc_chunk(fs_info, space_info, force)) {
4650 spin_unlock(&space_info->lock);
4652 } else if (space_info->chunk_alloc) {
4655 space_info->chunk_alloc = 1;
4658 spin_unlock(&space_info->lock);
4660 mutex_lock(&fs_info->chunk_mutex);
4663 * The chunk_mutex is held throughout the entirety of a chunk
4664 * allocation, so once we've acquired the chunk_mutex we know that the
4665 * other guy is done and we need to recheck and see if we should
4668 if (wait_for_alloc) {
4669 mutex_unlock(&fs_info->chunk_mutex);
4674 trans->allocating_chunk = true;
4677 * If we have mixed data/metadata chunks we want to make sure we keep
4678 * allocating mixed chunks instead of individual chunks.
4680 if (btrfs_mixed_space_info(space_info))
4681 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4684 * if we're doing a data chunk, go ahead and make sure that
4685 * we keep a reasonable number of metadata chunks allocated in the
4688 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4689 fs_info->data_chunk_allocations++;
4690 if (!(fs_info->data_chunk_allocations %
4691 fs_info->metadata_ratio))
4692 force_metadata_allocation(fs_info);
4696 * Check if we have enough space in SYSTEM chunk because we may need
4697 * to update devices.
4699 check_system_chunk(trans, fs_info, flags);
4701 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4702 trans->allocating_chunk = false;
4704 spin_lock(&space_info->lock);
4705 if (ret < 0 && ret != -ENOSPC)
4708 space_info->full = 1;
4712 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4714 space_info->chunk_alloc = 0;
4715 spin_unlock(&space_info->lock);
4716 mutex_unlock(&fs_info->chunk_mutex);
4718 * When we allocate a new chunk we reserve space in the chunk block
4719 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4720 * add new nodes/leafs to it if we end up needing to do it when
4721 * inserting the chunk item and updating device items as part of the
4722 * second phase of chunk allocation, performed by
4723 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4724 * large number of new block groups to create in our transaction
4725 * handle's new_bgs list to avoid exhausting the chunk block reserve
4726 * in extreme cases - like having a single transaction create many new
4727 * block groups when starting to write out the free space caches of all
4728 * the block groups that were made dirty during the lifetime of the
4731 if (trans->can_flush_pending_bgs &&
4732 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4733 btrfs_create_pending_block_groups(trans, fs_info);
4734 btrfs_trans_release_chunk_metadata(trans);
4739 static int can_overcommit(struct btrfs_fs_info *fs_info,
4740 struct btrfs_space_info *space_info, u64 bytes,
4741 enum btrfs_reserve_flush_enum flush,
4744 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4750 /* Don't overcommit when in mixed mode. */
4751 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4755 profile = btrfs_system_alloc_profile(fs_info);
4757 profile = btrfs_metadata_alloc_profile(fs_info);
4759 used = btrfs_space_info_used(space_info, false);
4762 * We only want to allow over committing if we have lots of actual space
4763 * free, but if we don't have enough space to handle the global reserve
4764 * space then we could end up having a real enospc problem when trying
4765 * to allocate a chunk or some other such important allocation.
4767 spin_lock(&global_rsv->lock);
4768 space_size = calc_global_rsv_need_space(global_rsv);
4769 spin_unlock(&global_rsv->lock);
4770 if (used + space_size >= space_info->total_bytes)
4773 used += space_info->bytes_may_use;
4775 avail = atomic64_read(&fs_info->free_chunk_space);
4778 * If we have dup, raid1 or raid10 then only half of the free
4779 * space is actually useable. For raid56, the space info used
4780 * doesn't include the parity drive, so we don't have to
4783 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4784 BTRFS_BLOCK_GROUP_RAID1 |
4785 BTRFS_BLOCK_GROUP_RAID10))
4789 * If we aren't flushing all things, let us overcommit up to
4790 * 1/2th of the space. If we can flush, don't let us overcommit
4791 * too much, let it overcommit up to 1/8 of the space.
4793 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4798 if (used + bytes < space_info->total_bytes + avail)
4803 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4804 unsigned long nr_pages, int nr_items)
4806 struct super_block *sb = fs_info->sb;
4808 if (down_read_trylock(&sb->s_umount)) {
4809 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4810 up_read(&sb->s_umount);
4813 * We needn't worry the filesystem going from r/w to r/o though
4814 * we don't acquire ->s_umount mutex, because the filesystem
4815 * should guarantee the delalloc inodes list be empty after
4816 * the filesystem is readonly(all dirty pages are written to
4819 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4820 if (!current->journal_info)
4821 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4825 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4831 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4832 nr = div64_u64(to_reclaim, bytes);
4838 #define EXTENT_SIZE_PER_ITEM SZ_256K
4841 * shrink metadata reservation for delalloc
4843 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4844 u64 orig, bool wait_ordered)
4846 struct btrfs_block_rsv *block_rsv;
4847 struct btrfs_space_info *space_info;
4848 struct btrfs_trans_handle *trans;
4853 unsigned long nr_pages;
4855 enum btrfs_reserve_flush_enum flush;
4857 /* Calc the number of the pages we need flush for space reservation */
4858 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4859 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4861 trans = (struct btrfs_trans_handle *)current->journal_info;
4862 block_rsv = &fs_info->delalloc_block_rsv;
4863 space_info = block_rsv->space_info;
4865 delalloc_bytes = percpu_counter_sum_positive(
4866 &fs_info->delalloc_bytes);
4867 if (delalloc_bytes == 0) {
4871 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4876 while (delalloc_bytes && loops < 3) {
4877 max_reclaim = min(delalloc_bytes, to_reclaim);
4878 nr_pages = max_reclaim >> PAGE_SHIFT;
4879 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4881 * We need to wait for the async pages to actually start before
4884 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4888 if (max_reclaim <= nr_pages)
4891 max_reclaim -= nr_pages;
4893 wait_event(fs_info->async_submit_wait,
4894 atomic_read(&fs_info->async_delalloc_pages) <=
4898 flush = BTRFS_RESERVE_FLUSH_ALL;
4900 flush = BTRFS_RESERVE_NO_FLUSH;
4901 spin_lock(&space_info->lock);
4902 if (list_empty(&space_info->tickets) &&
4903 list_empty(&space_info->priority_tickets)) {
4904 spin_unlock(&space_info->lock);
4907 spin_unlock(&space_info->lock);
4910 if (wait_ordered && !trans) {
4911 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4913 time_left = schedule_timeout_killable(1);
4917 delalloc_bytes = percpu_counter_sum_positive(
4918 &fs_info->delalloc_bytes);
4923 * maybe_commit_transaction - possibly commit the transaction if its ok to
4924 * @root - the root we're allocating for
4925 * @bytes - the number of bytes we want to reserve
4926 * @force - force the commit
4928 * This will check to make sure that committing the transaction will actually
4929 * get us somewhere and then commit the transaction if it does. Otherwise it
4930 * will return -ENOSPC.
4932 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4933 struct btrfs_space_info *space_info,
4934 u64 bytes, int force)
4936 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4937 struct btrfs_trans_handle *trans;
4939 trans = (struct btrfs_trans_handle *)current->journal_info;
4946 /* See if there is enough pinned space to make this reservation */
4947 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4952 * See if there is some space in the delayed insertion reservation for
4955 if (space_info != delayed_rsv->space_info)
4958 spin_lock(&delayed_rsv->lock);
4959 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4960 bytes - delayed_rsv->size) < 0) {
4961 spin_unlock(&delayed_rsv->lock);
4964 spin_unlock(&delayed_rsv->lock);
4967 trans = btrfs_join_transaction(fs_info->extent_root);
4971 return btrfs_commit_transaction(trans);
4974 struct reserve_ticket {
4977 struct list_head list;
4978 wait_queue_head_t wait;
4982 * Try to flush some data based on policy set by @state. This is only advisory
4983 * and may fail for various reasons. The caller is supposed to examine the
4984 * state of @space_info to detect the outcome.
4986 static void flush_space(struct btrfs_fs_info *fs_info,
4987 struct btrfs_space_info *space_info, u64 num_bytes,
4990 struct btrfs_root *root = fs_info->extent_root;
4991 struct btrfs_trans_handle *trans;
4996 case FLUSH_DELAYED_ITEMS_NR:
4997 case FLUSH_DELAYED_ITEMS:
4998 if (state == FLUSH_DELAYED_ITEMS_NR)
4999 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
5003 trans = btrfs_join_transaction(root);
5004 if (IS_ERR(trans)) {
5005 ret = PTR_ERR(trans);
5008 ret = btrfs_run_delayed_items_nr(trans, fs_info, nr);
5009 btrfs_end_transaction(trans);
5011 case FLUSH_DELALLOC:
5012 case FLUSH_DELALLOC_WAIT:
5013 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
5014 state == FLUSH_DELALLOC_WAIT);
5017 trans = btrfs_join_transaction(root);
5018 if (IS_ERR(trans)) {
5019 ret = PTR_ERR(trans);
5022 ret = do_chunk_alloc(trans, fs_info,
5023 btrfs_metadata_alloc_profile(fs_info),
5024 CHUNK_ALLOC_NO_FORCE);
5025 btrfs_end_transaction(trans);
5026 if (ret > 0 || ret == -ENOSPC)
5030 ret = may_commit_transaction(fs_info, space_info,
5038 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5044 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5045 struct btrfs_space_info *space_info,
5048 struct reserve_ticket *ticket;
5053 list_for_each_entry(ticket, &space_info->tickets, list)
5054 to_reclaim += ticket->bytes;
5055 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5056 to_reclaim += ticket->bytes;
5060 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5061 if (can_overcommit(fs_info, space_info, to_reclaim,
5062 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5065 used = btrfs_space_info_used(space_info, true);
5067 if (can_overcommit(fs_info, space_info, SZ_1M,
5068 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5069 expected = div_factor_fine(space_info->total_bytes, 95);
5071 expected = div_factor_fine(space_info->total_bytes, 90);
5073 if (used > expected)
5074 to_reclaim = used - expected;
5077 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5078 space_info->bytes_reserved);
5082 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5083 struct btrfs_space_info *space_info,
5084 u64 used, bool system_chunk)
5086 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5088 /* If we're just plain full then async reclaim just slows us down. */
5089 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5092 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5096 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5097 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5100 static void wake_all_tickets(struct list_head *head)
5102 struct reserve_ticket *ticket;
5104 while (!list_empty(head)) {
5105 ticket = list_first_entry(head, struct reserve_ticket, list);
5106 list_del_init(&ticket->list);
5107 ticket->error = -ENOSPC;
5108 wake_up(&ticket->wait);
5113 * This is for normal flushers, we can wait all goddamned day if we want to. We
5114 * will loop and continuously try to flush as long as we are making progress.
5115 * We count progress as clearing off tickets each time we have to loop.
5117 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5119 struct btrfs_fs_info *fs_info;
5120 struct btrfs_space_info *space_info;
5123 int commit_cycles = 0;
5124 u64 last_tickets_id;
5126 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5127 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5129 spin_lock(&space_info->lock);
5130 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5133 space_info->flush = 0;
5134 spin_unlock(&space_info->lock);
5137 last_tickets_id = space_info->tickets_id;
5138 spin_unlock(&space_info->lock);
5140 flush_state = FLUSH_DELAYED_ITEMS_NR;
5142 flush_space(fs_info, space_info, to_reclaim, flush_state);
5143 spin_lock(&space_info->lock);
5144 if (list_empty(&space_info->tickets)) {
5145 space_info->flush = 0;
5146 spin_unlock(&space_info->lock);
5149 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5152 if (last_tickets_id == space_info->tickets_id) {
5155 last_tickets_id = space_info->tickets_id;
5156 flush_state = FLUSH_DELAYED_ITEMS_NR;
5161 if (flush_state > COMMIT_TRANS) {
5163 if (commit_cycles > 2) {
5164 wake_all_tickets(&space_info->tickets);
5165 space_info->flush = 0;
5167 flush_state = FLUSH_DELAYED_ITEMS_NR;
5170 spin_unlock(&space_info->lock);
5171 } while (flush_state <= COMMIT_TRANS);
5174 void btrfs_init_async_reclaim_work(struct work_struct *work)
5176 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5179 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5180 struct btrfs_space_info *space_info,
5181 struct reserve_ticket *ticket)
5184 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5186 spin_lock(&space_info->lock);
5187 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5190 spin_unlock(&space_info->lock);
5193 spin_unlock(&space_info->lock);
5196 flush_space(fs_info, space_info, to_reclaim, flush_state);
5198 spin_lock(&space_info->lock);
5199 if (ticket->bytes == 0) {
5200 spin_unlock(&space_info->lock);
5203 spin_unlock(&space_info->lock);
5206 * Priority flushers can't wait on delalloc without
5209 if (flush_state == FLUSH_DELALLOC ||
5210 flush_state == FLUSH_DELALLOC_WAIT)
5211 flush_state = ALLOC_CHUNK;
5212 } while (flush_state < COMMIT_TRANS);
5215 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5216 struct btrfs_space_info *space_info,
5217 struct reserve_ticket *ticket, u64 orig_bytes)
5223 spin_lock(&space_info->lock);
5224 while (ticket->bytes > 0 && ticket->error == 0) {
5225 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5230 spin_unlock(&space_info->lock);
5234 finish_wait(&ticket->wait, &wait);
5235 spin_lock(&space_info->lock);
5238 ret = ticket->error;
5239 if (!list_empty(&ticket->list))
5240 list_del_init(&ticket->list);
5241 if (ticket->bytes && ticket->bytes < orig_bytes) {
5242 u64 num_bytes = orig_bytes - ticket->bytes;
5243 space_info->bytes_may_use -= num_bytes;
5244 trace_btrfs_space_reservation(fs_info, "space_info",
5245 space_info->flags, num_bytes, 0);
5247 spin_unlock(&space_info->lock);
5253 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5254 * @root - the root we're allocating for
5255 * @space_info - the space info we want to allocate from
5256 * @orig_bytes - the number of bytes we want
5257 * @flush - whether or not we can flush to make our reservation
5259 * This will reserve orig_bytes number of bytes from the space info associated
5260 * with the block_rsv. If there is not enough space it will make an attempt to
5261 * flush out space to make room. It will do this by flushing delalloc if
5262 * possible or committing the transaction. If flush is 0 then no attempts to
5263 * regain reservations will be made and this will fail if there is not enough
5266 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5267 struct btrfs_space_info *space_info,
5269 enum btrfs_reserve_flush_enum flush,
5272 struct reserve_ticket ticket;
5277 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5279 spin_lock(&space_info->lock);
5281 used = btrfs_space_info_used(space_info, true);
5284 * If we have enough space then hooray, make our reservation and carry
5285 * on. If not see if we can overcommit, and if we can, hooray carry on.
5286 * If not things get more complicated.
5288 if (used + orig_bytes <= space_info->total_bytes) {
5289 space_info->bytes_may_use += orig_bytes;
5290 trace_btrfs_space_reservation(fs_info, "space_info",
5291 space_info->flags, orig_bytes, 1);
5293 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5295 space_info->bytes_may_use += orig_bytes;
5296 trace_btrfs_space_reservation(fs_info, "space_info",
5297 space_info->flags, orig_bytes, 1);
5302 * If we couldn't make a reservation then setup our reservation ticket
5303 * and kick the async worker if it's not already running.
5305 * If we are a priority flusher then we just need to add our ticket to
5306 * the list and we will do our own flushing further down.
5308 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5309 ticket.bytes = orig_bytes;
5311 init_waitqueue_head(&ticket.wait);
5312 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5313 list_add_tail(&ticket.list, &space_info->tickets);
5314 if (!space_info->flush) {
5315 space_info->flush = 1;
5316 trace_btrfs_trigger_flush(fs_info,
5320 queue_work(system_unbound_wq,
5321 &fs_info->async_reclaim_work);
5324 list_add_tail(&ticket.list,
5325 &space_info->priority_tickets);
5327 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5330 * We will do the space reservation dance during log replay,
5331 * which means we won't have fs_info->fs_root set, so don't do
5332 * the async reclaim as we will panic.
5334 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5335 need_do_async_reclaim(fs_info, space_info,
5336 used, system_chunk) &&
5337 !work_busy(&fs_info->async_reclaim_work)) {
5338 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5339 orig_bytes, flush, "preempt");
5340 queue_work(system_unbound_wq,
5341 &fs_info->async_reclaim_work);
5344 spin_unlock(&space_info->lock);
5345 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5348 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5349 return wait_reserve_ticket(fs_info, space_info, &ticket,
5353 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5354 spin_lock(&space_info->lock);
5356 if (ticket.bytes < orig_bytes) {
5357 u64 num_bytes = orig_bytes - ticket.bytes;
5358 space_info->bytes_may_use -= num_bytes;
5359 trace_btrfs_space_reservation(fs_info, "space_info",
5364 list_del_init(&ticket.list);
5367 spin_unlock(&space_info->lock);
5368 ASSERT(list_empty(&ticket.list));
5373 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5374 * @root - the root we're allocating for
5375 * @block_rsv - the block_rsv we're allocating for
5376 * @orig_bytes - the number of bytes we want
5377 * @flush - whether or not we can flush to make our reservation
5379 * This will reserve orgi_bytes number of bytes from the space info associated
5380 * with the block_rsv. If there is not enough space it will make an attempt to
5381 * flush out space to make room. It will do this by flushing delalloc if
5382 * possible or committing the transaction. If flush is 0 then no attempts to
5383 * regain reservations will be made and this will fail if there is not enough
5386 static int reserve_metadata_bytes(struct btrfs_root *root,
5387 struct btrfs_block_rsv *block_rsv,
5389 enum btrfs_reserve_flush_enum flush)
5391 struct btrfs_fs_info *fs_info = root->fs_info;
5392 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5394 bool system_chunk = (root == fs_info->chunk_root);
5396 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5397 orig_bytes, flush, system_chunk);
5398 if (ret == -ENOSPC &&
5399 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5400 if (block_rsv != global_rsv &&
5401 !block_rsv_use_bytes(global_rsv, orig_bytes))
5405 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5406 block_rsv->space_info->flags,
5411 static struct btrfs_block_rsv *get_block_rsv(
5412 const struct btrfs_trans_handle *trans,
5413 const struct btrfs_root *root)
5415 struct btrfs_fs_info *fs_info = root->fs_info;
5416 struct btrfs_block_rsv *block_rsv = NULL;
5418 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5419 (root == fs_info->csum_root && trans->adding_csums) ||
5420 (root == fs_info->uuid_root))
5421 block_rsv = trans->block_rsv;
5424 block_rsv = root->block_rsv;
5427 block_rsv = &fs_info->empty_block_rsv;
5432 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5436 spin_lock(&block_rsv->lock);
5437 if (block_rsv->reserved >= num_bytes) {
5438 block_rsv->reserved -= num_bytes;
5439 if (block_rsv->reserved < block_rsv->size)
5440 block_rsv->full = 0;
5443 spin_unlock(&block_rsv->lock);
5447 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5448 u64 num_bytes, int update_size)
5450 spin_lock(&block_rsv->lock);
5451 block_rsv->reserved += num_bytes;
5453 block_rsv->size += num_bytes;
5454 else if (block_rsv->reserved >= block_rsv->size)
5455 block_rsv->full = 1;
5456 spin_unlock(&block_rsv->lock);
5459 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5460 struct btrfs_block_rsv *dest, u64 num_bytes,
5463 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5466 if (global_rsv->space_info != dest->space_info)
5469 spin_lock(&global_rsv->lock);
5470 min_bytes = div_factor(global_rsv->size, min_factor);
5471 if (global_rsv->reserved < min_bytes + num_bytes) {
5472 spin_unlock(&global_rsv->lock);
5475 global_rsv->reserved -= num_bytes;
5476 if (global_rsv->reserved < global_rsv->size)
5477 global_rsv->full = 0;
5478 spin_unlock(&global_rsv->lock);
5480 block_rsv_add_bytes(dest, num_bytes, 1);
5485 * This is for space we already have accounted in space_info->bytes_may_use, so
5486 * basically when we're returning space from block_rsv's.
5488 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5489 struct btrfs_space_info *space_info,
5492 struct reserve_ticket *ticket;
5493 struct list_head *head;
5495 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5496 bool check_overcommit = false;
5498 spin_lock(&space_info->lock);
5499 head = &space_info->priority_tickets;
5502 * If we are over our limit then we need to check and see if we can
5503 * overcommit, and if we can't then we just need to free up our space
5504 * and not satisfy any requests.
5506 used = btrfs_space_info_used(space_info, true);
5507 if (used - num_bytes >= space_info->total_bytes)
5508 check_overcommit = true;
5510 while (!list_empty(head) && num_bytes) {
5511 ticket = list_first_entry(head, struct reserve_ticket,
5514 * We use 0 bytes because this space is already reserved, so
5515 * adding the ticket space would be a double count.
5517 if (check_overcommit &&
5518 !can_overcommit(fs_info, space_info, 0, flush, false))
5520 if (num_bytes >= ticket->bytes) {
5521 list_del_init(&ticket->list);
5522 num_bytes -= ticket->bytes;
5524 space_info->tickets_id++;
5525 wake_up(&ticket->wait);
5527 ticket->bytes -= num_bytes;
5532 if (num_bytes && head == &space_info->priority_tickets) {
5533 head = &space_info->tickets;
5534 flush = BTRFS_RESERVE_FLUSH_ALL;
5537 space_info->bytes_may_use -= num_bytes;
5538 trace_btrfs_space_reservation(fs_info, "space_info",
5539 space_info->flags, num_bytes, 0);
5540 spin_unlock(&space_info->lock);
5544 * This is for newly allocated space that isn't accounted in
5545 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5546 * we use this helper.
5548 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5549 struct btrfs_space_info *space_info,
5552 struct reserve_ticket *ticket;
5553 struct list_head *head = &space_info->priority_tickets;
5556 while (!list_empty(head) && num_bytes) {
5557 ticket = list_first_entry(head, struct reserve_ticket,
5559 if (num_bytes >= ticket->bytes) {
5560 trace_btrfs_space_reservation(fs_info, "space_info",
5563 list_del_init(&ticket->list);
5564 num_bytes -= ticket->bytes;
5565 space_info->bytes_may_use += ticket->bytes;
5567 space_info->tickets_id++;
5568 wake_up(&ticket->wait);
5570 trace_btrfs_space_reservation(fs_info, "space_info",
5573 space_info->bytes_may_use += num_bytes;
5574 ticket->bytes -= num_bytes;
5579 if (num_bytes && head == &space_info->priority_tickets) {
5580 head = &space_info->tickets;
5585 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5586 struct btrfs_block_rsv *block_rsv,
5587 struct btrfs_block_rsv *dest, u64 num_bytes)
5589 struct btrfs_space_info *space_info = block_rsv->space_info;
5591 spin_lock(&block_rsv->lock);
5592 if (num_bytes == (u64)-1)
5593 num_bytes = block_rsv->size;
5594 block_rsv->size -= num_bytes;
5595 if (block_rsv->reserved >= block_rsv->size) {
5596 num_bytes = block_rsv->reserved - block_rsv->size;
5597 block_rsv->reserved = block_rsv->size;
5598 block_rsv->full = 1;
5602 spin_unlock(&block_rsv->lock);
5604 if (num_bytes > 0) {
5606 spin_lock(&dest->lock);
5610 bytes_to_add = dest->size - dest->reserved;
5611 bytes_to_add = min(num_bytes, bytes_to_add);
5612 dest->reserved += bytes_to_add;
5613 if (dest->reserved >= dest->size)
5615 num_bytes -= bytes_to_add;
5617 spin_unlock(&dest->lock);
5620 space_info_add_old_bytes(fs_info, space_info,
5625 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5626 struct btrfs_block_rsv *dst, u64 num_bytes,
5631 ret = block_rsv_use_bytes(src, num_bytes);
5635 block_rsv_add_bytes(dst, num_bytes, update_size);
5639 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5641 memset(rsv, 0, sizeof(*rsv));
5642 spin_lock_init(&rsv->lock);
5646 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5647 unsigned short type)
5649 struct btrfs_block_rsv *block_rsv;
5651 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5655 btrfs_init_block_rsv(block_rsv, type);
5656 block_rsv->space_info = __find_space_info(fs_info,
5657 BTRFS_BLOCK_GROUP_METADATA);
5661 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5662 struct btrfs_block_rsv *rsv)
5666 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5670 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5675 int btrfs_block_rsv_add(struct btrfs_root *root,
5676 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5677 enum btrfs_reserve_flush_enum flush)
5684 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5686 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5693 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5701 spin_lock(&block_rsv->lock);
5702 num_bytes = div_factor(block_rsv->size, min_factor);
5703 if (block_rsv->reserved >= num_bytes)
5705 spin_unlock(&block_rsv->lock);
5710 int btrfs_block_rsv_refill(struct btrfs_root *root,
5711 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5712 enum btrfs_reserve_flush_enum flush)
5720 spin_lock(&block_rsv->lock);
5721 num_bytes = min_reserved;
5722 if (block_rsv->reserved >= num_bytes)
5725 num_bytes -= block_rsv->reserved;
5726 spin_unlock(&block_rsv->lock);
5731 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5733 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5740 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5741 struct btrfs_block_rsv *block_rsv,
5744 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5746 if (global_rsv == block_rsv ||
5747 block_rsv->space_info != global_rsv->space_info)
5749 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5752 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5754 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5755 struct btrfs_space_info *sinfo = block_rsv->space_info;
5759 * The global block rsv is based on the size of the extent tree, the
5760 * checksum tree and the root tree. If the fs is empty we want to set
5761 * it to a minimal amount for safety.
5763 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5764 btrfs_root_used(&fs_info->csum_root->root_item) +
5765 btrfs_root_used(&fs_info->tree_root->root_item);
5766 num_bytes = max_t(u64, num_bytes, SZ_16M);
5768 spin_lock(&sinfo->lock);
5769 spin_lock(&block_rsv->lock);
5771 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5773 if (block_rsv->reserved < block_rsv->size) {
5774 num_bytes = btrfs_space_info_used(sinfo, true);
5775 if (sinfo->total_bytes > num_bytes) {
5776 num_bytes = sinfo->total_bytes - num_bytes;
5777 num_bytes = min(num_bytes,
5778 block_rsv->size - block_rsv->reserved);
5779 block_rsv->reserved += num_bytes;
5780 sinfo->bytes_may_use += num_bytes;
5781 trace_btrfs_space_reservation(fs_info, "space_info",
5782 sinfo->flags, num_bytes,
5785 } else if (block_rsv->reserved > block_rsv->size) {
5786 num_bytes = block_rsv->reserved - block_rsv->size;
5787 sinfo->bytes_may_use -= num_bytes;
5788 trace_btrfs_space_reservation(fs_info, "space_info",
5789 sinfo->flags, num_bytes, 0);
5790 block_rsv->reserved = block_rsv->size;
5793 if (block_rsv->reserved == block_rsv->size)
5794 block_rsv->full = 1;
5796 block_rsv->full = 0;
5798 spin_unlock(&block_rsv->lock);
5799 spin_unlock(&sinfo->lock);
5802 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5804 struct btrfs_space_info *space_info;
5806 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5807 fs_info->chunk_block_rsv.space_info = space_info;
5809 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5810 fs_info->global_block_rsv.space_info = space_info;
5811 fs_info->delalloc_block_rsv.space_info = space_info;
5812 fs_info->trans_block_rsv.space_info = space_info;
5813 fs_info->empty_block_rsv.space_info = space_info;
5814 fs_info->delayed_block_rsv.space_info = space_info;
5816 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5817 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5818 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5819 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5820 if (fs_info->quota_root)
5821 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5822 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5824 update_global_block_rsv(fs_info);
5827 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5829 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5831 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5832 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5833 WARN_ON(fs_info->trans_block_rsv.size > 0);
5834 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5835 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5836 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5837 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5838 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5841 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5842 struct btrfs_fs_info *fs_info)
5844 if (!trans->block_rsv)
5847 if (!trans->bytes_reserved)
5850 trace_btrfs_space_reservation(fs_info, "transaction",
5851 trans->transid, trans->bytes_reserved, 0);
5852 btrfs_block_rsv_release(fs_info, trans->block_rsv,
5853 trans->bytes_reserved);
5854 trans->bytes_reserved = 0;
5858 * To be called after all the new block groups attached to the transaction
5859 * handle have been created (btrfs_create_pending_block_groups()).
5861 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5863 struct btrfs_fs_info *fs_info = trans->fs_info;
5865 if (!trans->chunk_bytes_reserved)
5868 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5870 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5871 trans->chunk_bytes_reserved);
5872 trans->chunk_bytes_reserved = 0;
5875 /* Can only return 0 or -ENOSPC */
5876 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5877 struct btrfs_inode *inode)
5879 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5880 struct btrfs_root *root = inode->root;
5882 * We always use trans->block_rsv here as we will have reserved space
5883 * for our orphan when starting the transaction, using get_block_rsv()
5884 * here will sometimes make us choose the wrong block rsv as we could be
5885 * doing a reloc inode for a non refcounted root.
5887 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5888 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5891 * We need to hold space in order to delete our orphan item once we've
5892 * added it, so this takes the reservation so we can release it later
5893 * when we are truly done with the orphan item.
5895 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5897 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5899 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5902 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5904 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5905 struct btrfs_root *root = inode->root;
5906 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5908 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5910 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5914 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5915 * root: the root of the parent directory
5916 * rsv: block reservation
5917 * items: the number of items that we need do reservation
5918 * qgroup_reserved: used to return the reserved size in qgroup
5920 * This function is used to reserve the space for snapshot/subvolume
5921 * creation and deletion. Those operations are different with the
5922 * common file/directory operations, they change two fs/file trees
5923 * and root tree, the number of items that the qgroup reserves is
5924 * different with the free space reservation. So we can not use
5925 * the space reservation mechanism in start_transaction().
5927 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5928 struct btrfs_block_rsv *rsv,
5930 u64 *qgroup_reserved,
5931 bool use_global_rsv)
5935 struct btrfs_fs_info *fs_info = root->fs_info;
5936 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5938 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5939 /* One for parent inode, two for dir entries */
5940 num_bytes = 3 * fs_info->nodesize;
5941 ret = btrfs_qgroup_reserve_meta(root, num_bytes, true);
5948 *qgroup_reserved = num_bytes;
5950 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5951 rsv->space_info = __find_space_info(fs_info,
5952 BTRFS_BLOCK_GROUP_METADATA);
5953 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5954 BTRFS_RESERVE_FLUSH_ALL);
5956 if (ret == -ENOSPC && use_global_rsv)
5957 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5959 if (ret && *qgroup_reserved)
5960 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5965 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5966 struct btrfs_block_rsv *rsv)
5968 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5972 * drop_outstanding_extent - drop an outstanding extent
5973 * @inode: the inode we're dropping the extent for
5974 * @num_bytes: the number of bytes we're releasing.
5976 * This is called when we are freeing up an outstanding extent, either called
5977 * after an error or after an extent is written. This will return the number of
5978 * reserved extents that need to be freed. This must be called with
5979 * BTRFS_I(inode)->lock held.
5981 static unsigned drop_outstanding_extent(struct btrfs_inode *inode,
5984 unsigned drop_inode_space = 0;
5985 unsigned dropped_extents = 0;
5986 unsigned num_extents;
5988 num_extents = count_max_extents(num_bytes);
5989 ASSERT(num_extents);
5990 ASSERT(inode->outstanding_extents >= num_extents);
5991 inode->outstanding_extents -= num_extents;
5993 if (inode->outstanding_extents == 0 &&
5994 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5995 &inode->runtime_flags))
5996 drop_inode_space = 1;
5999 * If we have more or the same amount of outstanding extents than we have
6000 * reserved then we need to leave the reserved extents count alone.
6002 if (inode->outstanding_extents >= inode->reserved_extents)
6003 return drop_inode_space;
6005 dropped_extents = inode->reserved_extents - inode->outstanding_extents;
6006 inode->reserved_extents -= dropped_extents;
6007 return dropped_extents + drop_inode_space;
6011 * calc_csum_metadata_size - return the amount of metadata space that must be
6012 * reserved/freed for the given bytes.
6013 * @inode: the inode we're manipulating
6014 * @num_bytes: the number of bytes in question
6015 * @reserve: 1 if we are reserving space, 0 if we are freeing space
6017 * This adjusts the number of csum_bytes in the inode and then returns the
6018 * correct amount of metadata that must either be reserved or freed. We
6019 * calculate how many checksums we can fit into one leaf and then divide the
6020 * number of bytes that will need to be checksumed by this value to figure out
6021 * how many checksums will be required. If we are adding bytes then the number
6022 * may go up and we will return the number of additional bytes that must be
6023 * reserved. If it is going down we will return the number of bytes that must
6026 * This must be called with BTRFS_I(inode)->lock held.
6028 static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes,
6031 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6032 u64 old_csums, num_csums;
6034 if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0)
6037 old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
6039 inode->csum_bytes += num_bytes;
6041 inode->csum_bytes -= num_bytes;
6042 num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
6044 /* No change, no need to reserve more */
6045 if (old_csums == num_csums)
6049 return btrfs_calc_trans_metadata_size(fs_info,
6050 num_csums - old_csums);
6052 return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums);
6055 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6057 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6058 struct btrfs_root *root = inode->root;
6059 struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv;
6062 unsigned nr_extents;
6063 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6065 bool delalloc_lock = true;
6068 bool release_extra = false;
6070 /* If we are a free space inode we need to not flush since we will be in
6071 * the middle of a transaction commit. We also don't need the delalloc
6072 * mutex since we won't race with anybody. We need this mostly to make
6073 * lockdep shut its filthy mouth.
6075 * If we have a transaction open (can happen if we call truncate_block
6076 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6078 if (btrfs_is_free_space_inode(inode)) {
6079 flush = BTRFS_RESERVE_NO_FLUSH;
6080 delalloc_lock = false;
6081 } else if (current->journal_info) {
6082 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6085 if (flush != BTRFS_RESERVE_NO_FLUSH &&
6086 btrfs_transaction_in_commit(fs_info))
6087 schedule_timeout(1);
6090 mutex_lock(&inode->delalloc_mutex);
6092 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6094 spin_lock(&inode->lock);
6095 nr_extents = count_max_extents(num_bytes);
6096 inode->outstanding_extents += nr_extents;
6099 if (inode->outstanding_extents > inode->reserved_extents)
6100 nr_extents += inode->outstanding_extents -
6101 inode->reserved_extents;
6103 /* We always want to reserve a slot for updating the inode. */
6104 to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1);
6105 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
6106 csum_bytes = inode->csum_bytes;
6107 spin_unlock(&inode->lock);
6109 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6110 ret = btrfs_qgroup_reserve_meta(root,
6111 nr_extents * fs_info->nodesize, true);
6116 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
6117 if (unlikely(ret)) {
6118 btrfs_qgroup_free_meta(root,
6119 nr_extents * fs_info->nodesize);
6123 spin_lock(&inode->lock);
6124 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6125 &inode->runtime_flags)) {
6126 to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1);
6127 release_extra = true;
6129 inode->reserved_extents += nr_extents;
6130 spin_unlock(&inode->lock);
6133 mutex_unlock(&inode->delalloc_mutex);
6136 trace_btrfs_space_reservation(fs_info, "delalloc",
6137 btrfs_ino(inode), to_reserve, 1);
6139 btrfs_block_rsv_release(fs_info, block_rsv,
6140 btrfs_calc_trans_metadata_size(fs_info, 1));
6144 spin_lock(&inode->lock);
6145 dropped = drop_outstanding_extent(inode, num_bytes);
6147 * If the inodes csum_bytes is the same as the original
6148 * csum_bytes then we know we haven't raced with any free()ers
6149 * so we can just reduce our inodes csum bytes and carry on.
6151 if (inode->csum_bytes == csum_bytes) {
6152 calc_csum_metadata_size(inode, num_bytes, 0);
6154 u64 orig_csum_bytes = inode->csum_bytes;
6158 * This is tricky, but first we need to figure out how much we
6159 * freed from any free-ers that occurred during this
6160 * reservation, so we reset ->csum_bytes to the csum_bytes
6161 * before we dropped our lock, and then call the free for the
6162 * number of bytes that were freed while we were trying our
6165 bytes = csum_bytes - inode->csum_bytes;
6166 inode->csum_bytes = csum_bytes;
6167 to_free = calc_csum_metadata_size(inode, bytes, 0);
6171 * Now we need to see how much we would have freed had we not
6172 * been making this reservation and our ->csum_bytes were not
6173 * artificially inflated.
6175 inode->csum_bytes = csum_bytes - num_bytes;
6176 bytes = csum_bytes - orig_csum_bytes;
6177 bytes = calc_csum_metadata_size(inode, bytes, 0);
6180 * Now reset ->csum_bytes to what it should be. If bytes is
6181 * more than to_free then we would have freed more space had we
6182 * not had an artificially high ->csum_bytes, so we need to free
6183 * the remainder. If bytes is the same or less then we don't
6184 * need to do anything, the other free-ers did the correct
6187 inode->csum_bytes = orig_csum_bytes - num_bytes;
6188 if (bytes > to_free)
6189 to_free = bytes - to_free;
6193 spin_unlock(&inode->lock);
6195 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6198 btrfs_block_rsv_release(fs_info, block_rsv, to_free);
6199 trace_btrfs_space_reservation(fs_info, "delalloc",
6200 btrfs_ino(inode), to_free, 0);
6203 mutex_unlock(&inode->delalloc_mutex);
6208 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6209 * @inode: the inode to release the reservation for
6210 * @num_bytes: the number of bytes we're releasing
6212 * This will release the metadata reservation for an inode. This can be called
6213 * once we complete IO for a given set of bytes to release their metadata
6216 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes)
6218 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6222 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6223 spin_lock(&inode->lock);
6224 dropped = drop_outstanding_extent(inode, num_bytes);
6227 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6228 spin_unlock(&inode->lock);
6230 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6232 if (btrfs_is_testing(fs_info))
6235 trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode),
6238 btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free);
6242 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6244 * @inode: inode we're writing to
6245 * @start: start range we are writing to
6246 * @len: how long the range we are writing to
6247 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6248 * current reservation.
6250 * This will do the following things
6252 * o reserve space in data space info for num bytes
6253 * and reserve precious corresponding qgroup space
6254 * (Done in check_data_free_space)
6256 * o reserve space for metadata space, based on the number of outstanding
6257 * extents and how much csums will be needed
6258 * also reserve metadata space in a per root over-reserve method.
6259 * o add to the inodes->delalloc_bytes
6260 * o add it to the fs_info's delalloc inodes list.
6261 * (Above 3 all done in delalloc_reserve_metadata)
6263 * Return 0 for success
6264 * Return <0 for error(-ENOSPC or -EQUOT)
6266 int btrfs_delalloc_reserve_space(struct inode *inode,
6267 struct extent_changeset **reserved, u64 start, u64 len)
6271 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6274 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6276 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6281 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6282 * @inode: inode we're releasing space for
6283 * @start: start position of the space already reserved
6284 * @len: the len of the space already reserved
6286 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6287 * called in the case that we don't need the metadata AND data reservations
6288 * anymore. So if there is an error or we insert an inline extent.
6290 * This function will release the metadata space that was not used and will
6291 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6292 * list if there are no delalloc bytes left.
6293 * Also it will handle the qgroup reserved space.
6295 void btrfs_delalloc_release_space(struct inode *inode,
6296 struct extent_changeset *reserved, u64 start, u64 len)
6298 btrfs_delalloc_release_metadata(BTRFS_I(inode), len);
6299 btrfs_free_reserved_data_space(inode, reserved, start, len);
6302 static int update_block_group(struct btrfs_trans_handle *trans,
6303 struct btrfs_fs_info *info, u64 bytenr,
6304 u64 num_bytes, int alloc)
6306 struct btrfs_block_group_cache *cache = NULL;
6307 u64 total = num_bytes;
6312 /* block accounting for super block */
6313 spin_lock(&info->delalloc_root_lock);
6314 old_val = btrfs_super_bytes_used(info->super_copy);
6316 old_val += num_bytes;
6318 old_val -= num_bytes;
6319 btrfs_set_super_bytes_used(info->super_copy, old_val);
6320 spin_unlock(&info->delalloc_root_lock);
6323 cache = btrfs_lookup_block_group(info, bytenr);
6326 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6327 BTRFS_BLOCK_GROUP_RAID1 |
6328 BTRFS_BLOCK_GROUP_RAID10))
6333 * If this block group has free space cache written out, we
6334 * need to make sure to load it if we are removing space. This
6335 * is because we need the unpinning stage to actually add the
6336 * space back to the block group, otherwise we will leak space.
6338 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6339 cache_block_group(cache, 1);
6341 byte_in_group = bytenr - cache->key.objectid;
6342 WARN_ON(byte_in_group > cache->key.offset);
6344 spin_lock(&cache->space_info->lock);
6345 spin_lock(&cache->lock);
6347 if (btrfs_test_opt(info, SPACE_CACHE) &&
6348 cache->disk_cache_state < BTRFS_DC_CLEAR)
6349 cache->disk_cache_state = BTRFS_DC_CLEAR;
6351 old_val = btrfs_block_group_used(&cache->item);
6352 num_bytes = min(total, cache->key.offset - byte_in_group);
6354 old_val += num_bytes;
6355 btrfs_set_block_group_used(&cache->item, old_val);
6356 cache->reserved -= num_bytes;
6357 cache->space_info->bytes_reserved -= num_bytes;
6358 cache->space_info->bytes_used += num_bytes;
6359 cache->space_info->disk_used += num_bytes * factor;
6360 spin_unlock(&cache->lock);
6361 spin_unlock(&cache->space_info->lock);
6363 old_val -= num_bytes;
6364 btrfs_set_block_group_used(&cache->item, old_val);
6365 cache->pinned += num_bytes;
6366 cache->space_info->bytes_pinned += num_bytes;
6367 cache->space_info->bytes_used -= num_bytes;
6368 cache->space_info->disk_used -= num_bytes * factor;
6369 spin_unlock(&cache->lock);
6370 spin_unlock(&cache->space_info->lock);
6372 trace_btrfs_space_reservation(info, "pinned",
6373 cache->space_info->flags,
6375 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6377 set_extent_dirty(info->pinned_extents,
6378 bytenr, bytenr + num_bytes - 1,
6379 GFP_NOFS | __GFP_NOFAIL);
6382 spin_lock(&trans->transaction->dirty_bgs_lock);
6383 if (list_empty(&cache->dirty_list)) {
6384 list_add_tail(&cache->dirty_list,
6385 &trans->transaction->dirty_bgs);
6386 trans->transaction->num_dirty_bgs++;
6387 btrfs_get_block_group(cache);
6389 spin_unlock(&trans->transaction->dirty_bgs_lock);
6392 * No longer have used bytes in this block group, queue it for
6393 * deletion. We do this after adding the block group to the
6394 * dirty list to avoid races between cleaner kthread and space
6397 if (!alloc && old_val == 0) {
6398 spin_lock(&info->unused_bgs_lock);
6399 if (list_empty(&cache->bg_list)) {
6400 btrfs_get_block_group(cache);
6401 list_add_tail(&cache->bg_list,
6404 spin_unlock(&info->unused_bgs_lock);
6407 btrfs_put_block_group(cache);
6409 bytenr += num_bytes;
6414 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6416 struct btrfs_block_group_cache *cache;
6419 spin_lock(&fs_info->block_group_cache_lock);
6420 bytenr = fs_info->first_logical_byte;
6421 spin_unlock(&fs_info->block_group_cache_lock);
6423 if (bytenr < (u64)-1)
6426 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6430 bytenr = cache->key.objectid;
6431 btrfs_put_block_group(cache);
6436 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6437 struct btrfs_block_group_cache *cache,
6438 u64 bytenr, u64 num_bytes, int reserved)
6440 spin_lock(&cache->space_info->lock);
6441 spin_lock(&cache->lock);
6442 cache->pinned += num_bytes;
6443 cache->space_info->bytes_pinned += num_bytes;
6445 cache->reserved -= num_bytes;
6446 cache->space_info->bytes_reserved -= num_bytes;
6448 spin_unlock(&cache->lock);
6449 spin_unlock(&cache->space_info->lock);
6451 trace_btrfs_space_reservation(fs_info, "pinned",
6452 cache->space_info->flags, num_bytes, 1);
6453 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6454 set_extent_dirty(fs_info->pinned_extents, bytenr,
6455 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6460 * this function must be called within transaction
6462 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6463 u64 bytenr, u64 num_bytes, int reserved)
6465 struct btrfs_block_group_cache *cache;
6467 cache = btrfs_lookup_block_group(fs_info, bytenr);
6468 BUG_ON(!cache); /* Logic error */
6470 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6472 btrfs_put_block_group(cache);
6477 * this function must be called within transaction
6479 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6480 u64 bytenr, u64 num_bytes)
6482 struct btrfs_block_group_cache *cache;
6485 cache = btrfs_lookup_block_group(fs_info, bytenr);
6490 * pull in the free space cache (if any) so that our pin
6491 * removes the free space from the cache. We have load_only set
6492 * to one because the slow code to read in the free extents does check
6493 * the pinned extents.
6495 cache_block_group(cache, 1);
6497 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6499 /* remove us from the free space cache (if we're there at all) */
6500 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6501 btrfs_put_block_group(cache);
6505 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6506 u64 start, u64 num_bytes)
6509 struct btrfs_block_group_cache *block_group;
6510 struct btrfs_caching_control *caching_ctl;
6512 block_group = btrfs_lookup_block_group(fs_info, start);
6516 cache_block_group(block_group, 0);
6517 caching_ctl = get_caching_control(block_group);
6521 BUG_ON(!block_group_cache_done(block_group));
6522 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6524 mutex_lock(&caching_ctl->mutex);
6526 if (start >= caching_ctl->progress) {
6527 ret = add_excluded_extent(fs_info, start, num_bytes);
6528 } else if (start + num_bytes <= caching_ctl->progress) {
6529 ret = btrfs_remove_free_space(block_group,
6532 num_bytes = caching_ctl->progress - start;
6533 ret = btrfs_remove_free_space(block_group,
6538 num_bytes = (start + num_bytes) -
6539 caching_ctl->progress;
6540 start = caching_ctl->progress;
6541 ret = add_excluded_extent(fs_info, start, num_bytes);
6544 mutex_unlock(&caching_ctl->mutex);
6545 put_caching_control(caching_ctl);
6547 btrfs_put_block_group(block_group);
6551 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6552 struct extent_buffer *eb)
6554 struct btrfs_file_extent_item *item;
6555 struct btrfs_key key;
6559 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6562 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6563 btrfs_item_key_to_cpu(eb, &key, i);
6564 if (key.type != BTRFS_EXTENT_DATA_KEY)
6566 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6567 found_type = btrfs_file_extent_type(eb, item);
6568 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6570 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6572 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6573 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6574 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6581 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6583 atomic_inc(&bg->reservations);
6586 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6589 struct btrfs_block_group_cache *bg;
6591 bg = btrfs_lookup_block_group(fs_info, start);
6593 if (atomic_dec_and_test(&bg->reservations))
6594 wake_up_atomic_t(&bg->reservations);
6595 btrfs_put_block_group(bg);
6598 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6604 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6606 struct btrfs_space_info *space_info = bg->space_info;
6610 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6614 * Our block group is read only but before we set it to read only,
6615 * some task might have had allocated an extent from it already, but it
6616 * has not yet created a respective ordered extent (and added it to a
6617 * root's list of ordered extents).
6618 * Therefore wait for any task currently allocating extents, since the
6619 * block group's reservations counter is incremented while a read lock
6620 * on the groups' semaphore is held and decremented after releasing
6621 * the read access on that semaphore and creating the ordered extent.
6623 down_write(&space_info->groups_sem);
6624 up_write(&space_info->groups_sem);
6626 wait_on_atomic_t(&bg->reservations,
6627 btrfs_wait_bg_reservations_atomic_t,
6628 TASK_UNINTERRUPTIBLE);
6632 * btrfs_add_reserved_bytes - update the block_group and space info counters
6633 * @cache: The cache we are manipulating
6634 * @ram_bytes: The number of bytes of file content, and will be same to
6635 * @num_bytes except for the compress path.
6636 * @num_bytes: The number of bytes in question
6637 * @delalloc: The blocks are allocated for the delalloc write
6639 * This is called by the allocator when it reserves space. If this is a
6640 * reservation and the block group has become read only we cannot make the
6641 * reservation and return -EAGAIN, otherwise this function always succeeds.
6643 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6644 u64 ram_bytes, u64 num_bytes, int delalloc)
6646 struct btrfs_space_info *space_info = cache->space_info;
6649 spin_lock(&space_info->lock);
6650 spin_lock(&cache->lock);
6654 cache->reserved += num_bytes;
6655 space_info->bytes_reserved += num_bytes;
6657 trace_btrfs_space_reservation(cache->fs_info,
6658 "space_info", space_info->flags,
6660 space_info->bytes_may_use -= ram_bytes;
6662 cache->delalloc_bytes += num_bytes;
6664 spin_unlock(&cache->lock);
6665 spin_unlock(&space_info->lock);
6670 * btrfs_free_reserved_bytes - update the block_group and space info counters
6671 * @cache: The cache we are manipulating
6672 * @num_bytes: The number of bytes in question
6673 * @delalloc: The blocks are allocated for the delalloc write
6675 * This is called by somebody who is freeing space that was never actually used
6676 * on disk. For example if you reserve some space for a new leaf in transaction
6677 * A and before transaction A commits you free that leaf, you call this with
6678 * reserve set to 0 in order to clear the reservation.
6681 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6682 u64 num_bytes, int delalloc)
6684 struct btrfs_space_info *space_info = cache->space_info;
6687 spin_lock(&space_info->lock);
6688 spin_lock(&cache->lock);
6690 space_info->bytes_readonly += num_bytes;
6691 cache->reserved -= num_bytes;
6692 space_info->bytes_reserved -= num_bytes;
6695 cache->delalloc_bytes -= num_bytes;
6696 spin_unlock(&cache->lock);
6697 spin_unlock(&space_info->lock);
6700 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6702 struct btrfs_caching_control *next;
6703 struct btrfs_caching_control *caching_ctl;
6704 struct btrfs_block_group_cache *cache;
6706 down_write(&fs_info->commit_root_sem);
6708 list_for_each_entry_safe(caching_ctl, next,
6709 &fs_info->caching_block_groups, list) {
6710 cache = caching_ctl->block_group;
6711 if (block_group_cache_done(cache)) {
6712 cache->last_byte_to_unpin = (u64)-1;
6713 list_del_init(&caching_ctl->list);
6714 put_caching_control(caching_ctl);
6716 cache->last_byte_to_unpin = caching_ctl->progress;
6720 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6721 fs_info->pinned_extents = &fs_info->freed_extents[1];
6723 fs_info->pinned_extents = &fs_info->freed_extents[0];
6725 up_write(&fs_info->commit_root_sem);
6727 update_global_block_rsv(fs_info);
6731 * Returns the free cluster for the given space info and sets empty_cluster to
6732 * what it should be based on the mount options.
6734 static struct btrfs_free_cluster *
6735 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6736 struct btrfs_space_info *space_info, u64 *empty_cluster)
6738 struct btrfs_free_cluster *ret = NULL;
6741 if (btrfs_mixed_space_info(space_info))
6744 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6745 ret = &fs_info->meta_alloc_cluster;
6746 if (btrfs_test_opt(fs_info, SSD))
6747 *empty_cluster = SZ_2M;
6749 *empty_cluster = SZ_64K;
6750 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6751 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6752 *empty_cluster = SZ_2M;
6753 ret = &fs_info->data_alloc_cluster;
6759 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6761 const bool return_free_space)
6763 struct btrfs_block_group_cache *cache = NULL;
6764 struct btrfs_space_info *space_info;
6765 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6766 struct btrfs_free_cluster *cluster = NULL;
6768 u64 total_unpinned = 0;
6769 u64 empty_cluster = 0;
6772 while (start <= end) {
6775 start >= cache->key.objectid + cache->key.offset) {
6777 btrfs_put_block_group(cache);
6779 cache = btrfs_lookup_block_group(fs_info, start);
6780 BUG_ON(!cache); /* Logic error */
6782 cluster = fetch_cluster_info(fs_info,
6785 empty_cluster <<= 1;
6788 len = cache->key.objectid + cache->key.offset - start;
6789 len = min(len, end + 1 - start);
6791 if (start < cache->last_byte_to_unpin) {
6792 len = min(len, cache->last_byte_to_unpin - start);
6793 if (return_free_space)
6794 btrfs_add_free_space(cache, start, len);
6798 total_unpinned += len;
6799 space_info = cache->space_info;
6802 * If this space cluster has been marked as fragmented and we've
6803 * unpinned enough in this block group to potentially allow a
6804 * cluster to be created inside of it go ahead and clear the
6807 if (cluster && cluster->fragmented &&
6808 total_unpinned > empty_cluster) {
6809 spin_lock(&cluster->lock);
6810 cluster->fragmented = 0;
6811 spin_unlock(&cluster->lock);
6814 spin_lock(&space_info->lock);
6815 spin_lock(&cache->lock);
6816 cache->pinned -= len;
6817 space_info->bytes_pinned -= len;
6819 trace_btrfs_space_reservation(fs_info, "pinned",
6820 space_info->flags, len, 0);
6821 space_info->max_extent_size = 0;
6822 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6824 space_info->bytes_readonly += len;
6827 spin_unlock(&cache->lock);
6828 if (!readonly && return_free_space &&
6829 global_rsv->space_info == space_info) {
6832 spin_lock(&global_rsv->lock);
6833 if (!global_rsv->full) {
6834 to_add = min(len, global_rsv->size -
6835 global_rsv->reserved);
6836 global_rsv->reserved += to_add;
6837 space_info->bytes_may_use += to_add;
6838 if (global_rsv->reserved >= global_rsv->size)
6839 global_rsv->full = 1;
6840 trace_btrfs_space_reservation(fs_info,
6846 spin_unlock(&global_rsv->lock);
6847 /* Add to any tickets we may have */
6849 space_info_add_new_bytes(fs_info, space_info,
6852 spin_unlock(&space_info->lock);
6856 btrfs_put_block_group(cache);
6860 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6861 struct btrfs_fs_info *fs_info)
6863 struct btrfs_block_group_cache *block_group, *tmp;
6864 struct list_head *deleted_bgs;
6865 struct extent_io_tree *unpin;
6870 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6871 unpin = &fs_info->freed_extents[1];
6873 unpin = &fs_info->freed_extents[0];
6875 while (!trans->aborted) {
6876 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6877 ret = find_first_extent_bit(unpin, 0, &start, &end,
6878 EXTENT_DIRTY, NULL);
6880 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6884 if (btrfs_test_opt(fs_info, DISCARD))
6885 ret = btrfs_discard_extent(fs_info, start,
6886 end + 1 - start, NULL);
6888 clear_extent_dirty(unpin, start, end);
6889 unpin_extent_range(fs_info, start, end, true);
6890 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6895 * Transaction is finished. We don't need the lock anymore. We
6896 * do need to clean up the block groups in case of a transaction
6899 deleted_bgs = &trans->transaction->deleted_bgs;
6900 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6904 if (!trans->aborted)
6905 ret = btrfs_discard_extent(fs_info,
6906 block_group->key.objectid,
6907 block_group->key.offset,
6910 list_del_init(&block_group->bg_list);
6911 btrfs_put_block_group_trimming(block_group);
6912 btrfs_put_block_group(block_group);
6915 const char *errstr = btrfs_decode_error(ret);
6917 "discard failed while removing blockgroup: errno=%d %s",
6925 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6926 struct btrfs_fs_info *info,
6927 struct btrfs_delayed_ref_node *node, u64 parent,
6928 u64 root_objectid, u64 owner_objectid,
6929 u64 owner_offset, int refs_to_drop,
6930 struct btrfs_delayed_extent_op *extent_op)
6932 struct btrfs_key key;
6933 struct btrfs_path *path;
6934 struct btrfs_root *extent_root = info->extent_root;
6935 struct extent_buffer *leaf;
6936 struct btrfs_extent_item *ei;
6937 struct btrfs_extent_inline_ref *iref;
6940 int extent_slot = 0;
6941 int found_extent = 0;
6945 u64 bytenr = node->bytenr;
6946 u64 num_bytes = node->num_bytes;
6948 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6950 path = btrfs_alloc_path();
6954 path->reada = READA_FORWARD;
6955 path->leave_spinning = 1;
6957 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6958 BUG_ON(!is_data && refs_to_drop != 1);
6961 skinny_metadata = 0;
6963 ret = lookup_extent_backref(trans, info, path, &iref,
6964 bytenr, num_bytes, parent,
6965 root_objectid, owner_objectid,
6968 extent_slot = path->slots[0];
6969 while (extent_slot >= 0) {
6970 btrfs_item_key_to_cpu(path->nodes[0], &key,
6972 if (key.objectid != bytenr)
6974 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6975 key.offset == num_bytes) {
6979 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6980 key.offset == owner_objectid) {
6984 if (path->slots[0] - extent_slot > 5)
6988 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6989 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6990 if (found_extent && item_size < sizeof(*ei))
6993 if (!found_extent) {
6995 ret = remove_extent_backref(trans, info, path, NULL,
6997 is_data, &last_ref);
6999 btrfs_abort_transaction(trans, ret);
7002 btrfs_release_path(path);
7003 path->leave_spinning = 1;
7005 key.objectid = bytenr;
7006 key.type = BTRFS_EXTENT_ITEM_KEY;
7007 key.offset = num_bytes;
7009 if (!is_data && skinny_metadata) {
7010 key.type = BTRFS_METADATA_ITEM_KEY;
7011 key.offset = owner_objectid;
7014 ret = btrfs_search_slot(trans, extent_root,
7016 if (ret > 0 && skinny_metadata && path->slots[0]) {
7018 * Couldn't find our skinny metadata item,
7019 * see if we have ye olde extent item.
7022 btrfs_item_key_to_cpu(path->nodes[0], &key,
7024 if (key.objectid == bytenr &&
7025 key.type == BTRFS_EXTENT_ITEM_KEY &&
7026 key.offset == num_bytes)
7030 if (ret > 0 && skinny_metadata) {
7031 skinny_metadata = false;
7032 key.objectid = bytenr;
7033 key.type = BTRFS_EXTENT_ITEM_KEY;
7034 key.offset = num_bytes;
7035 btrfs_release_path(path);
7036 ret = btrfs_search_slot(trans, extent_root,
7042 "umm, got %d back from search, was looking for %llu",
7045 btrfs_print_leaf(path->nodes[0]);
7048 btrfs_abort_transaction(trans, ret);
7051 extent_slot = path->slots[0];
7053 } else if (WARN_ON(ret == -ENOENT)) {
7054 btrfs_print_leaf(path->nodes[0]);
7056 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
7057 bytenr, parent, root_objectid, owner_objectid,
7059 btrfs_abort_transaction(trans, ret);
7062 btrfs_abort_transaction(trans, ret);
7066 leaf = path->nodes[0];
7067 item_size = btrfs_item_size_nr(leaf, extent_slot);
7068 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
7069 if (item_size < sizeof(*ei)) {
7070 BUG_ON(found_extent || extent_slot != path->slots[0]);
7071 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
7074 btrfs_abort_transaction(trans, ret);
7078 btrfs_release_path(path);
7079 path->leave_spinning = 1;
7081 key.objectid = bytenr;
7082 key.type = BTRFS_EXTENT_ITEM_KEY;
7083 key.offset = num_bytes;
7085 ret = btrfs_search_slot(trans, extent_root, &key, path,
7089 "umm, got %d back from search, was looking for %llu",
7091 btrfs_print_leaf(path->nodes[0]);
7094 btrfs_abort_transaction(trans, ret);
7098 extent_slot = path->slots[0];
7099 leaf = path->nodes[0];
7100 item_size = btrfs_item_size_nr(leaf, extent_slot);
7103 BUG_ON(item_size < sizeof(*ei));
7104 ei = btrfs_item_ptr(leaf, extent_slot,
7105 struct btrfs_extent_item);
7106 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7107 key.type == BTRFS_EXTENT_ITEM_KEY) {
7108 struct btrfs_tree_block_info *bi;
7109 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7110 bi = (struct btrfs_tree_block_info *)(ei + 1);
7111 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7114 refs = btrfs_extent_refs(leaf, ei);
7115 if (refs < refs_to_drop) {
7117 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7118 refs_to_drop, refs, bytenr);
7120 btrfs_abort_transaction(trans, ret);
7123 refs -= refs_to_drop;
7127 __run_delayed_extent_op(extent_op, leaf, ei);
7129 * In the case of inline back ref, reference count will
7130 * be updated by remove_extent_backref
7133 BUG_ON(!found_extent);
7135 btrfs_set_extent_refs(leaf, ei, refs);
7136 btrfs_mark_buffer_dirty(leaf);
7139 ret = remove_extent_backref(trans, info, path,
7141 is_data, &last_ref);
7143 btrfs_abort_transaction(trans, ret);
7149 BUG_ON(is_data && refs_to_drop !=
7150 extent_data_ref_count(path, iref));
7152 BUG_ON(path->slots[0] != extent_slot);
7154 BUG_ON(path->slots[0] != extent_slot + 1);
7155 path->slots[0] = extent_slot;
7161 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7164 btrfs_abort_transaction(trans, ret);
7167 btrfs_release_path(path);
7170 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7172 btrfs_abort_transaction(trans, ret);
7177 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7179 btrfs_abort_transaction(trans, ret);
7183 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7185 btrfs_abort_transaction(trans, ret);
7189 btrfs_release_path(path);
7192 btrfs_free_path(path);
7197 * when we free an block, it is possible (and likely) that we free the last
7198 * delayed ref for that extent as well. This searches the delayed ref tree for
7199 * a given extent, and if there are no other delayed refs to be processed, it
7200 * removes it from the tree.
7202 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7205 struct btrfs_delayed_ref_head *head;
7206 struct btrfs_delayed_ref_root *delayed_refs;
7209 delayed_refs = &trans->transaction->delayed_refs;
7210 spin_lock(&delayed_refs->lock);
7211 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7213 goto out_delayed_unlock;
7215 spin_lock(&head->lock);
7216 if (!list_empty(&head->ref_list))
7219 if (head->extent_op) {
7220 if (!head->must_insert_reserved)
7222 btrfs_free_delayed_extent_op(head->extent_op);
7223 head->extent_op = NULL;
7227 * waiting for the lock here would deadlock. If someone else has it
7228 * locked they are already in the process of dropping it anyway
7230 if (!mutex_trylock(&head->mutex))
7234 * at this point we have a head with no other entries. Go
7235 * ahead and process it.
7237 head->node.in_tree = 0;
7238 rb_erase(&head->href_node, &delayed_refs->href_root);
7240 atomic_dec(&delayed_refs->num_entries);
7243 * we don't take a ref on the node because we're removing it from the
7244 * tree, so we just steal the ref the tree was holding.
7246 delayed_refs->num_heads--;
7247 if (head->processing == 0)
7248 delayed_refs->num_heads_ready--;
7249 head->processing = 0;
7250 spin_unlock(&head->lock);
7251 spin_unlock(&delayed_refs->lock);
7253 BUG_ON(head->extent_op);
7254 if (head->must_insert_reserved)
7257 mutex_unlock(&head->mutex);
7258 btrfs_put_delayed_ref(&head->node);
7261 spin_unlock(&head->lock);
7264 spin_unlock(&delayed_refs->lock);
7268 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7269 struct btrfs_root *root,
7270 struct extent_buffer *buf,
7271 u64 parent, int last_ref)
7273 struct btrfs_fs_info *fs_info = root->fs_info;
7277 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7278 int old_ref_mod, new_ref_mod;
7280 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7282 root->root_key.objectid,
7283 btrfs_header_level(buf),
7284 BTRFS_DROP_DELAYED_REF, NULL,
7285 &old_ref_mod, &new_ref_mod);
7286 BUG_ON(ret); /* -ENOMEM */
7287 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7290 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7291 struct btrfs_block_group_cache *cache;
7293 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7294 ret = check_ref_cleanup(trans, buf->start);
7300 cache = btrfs_lookup_block_group(fs_info, buf->start);
7302 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7303 pin_down_extent(fs_info, cache, buf->start,
7305 btrfs_put_block_group(cache);
7309 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7311 btrfs_add_free_space(cache, buf->start, buf->len);
7312 btrfs_free_reserved_bytes(cache, buf->len, 0);
7313 btrfs_put_block_group(cache);
7314 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7318 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7319 root->root_key.objectid);
7323 * Deleting the buffer, clear the corrupt flag since it doesn't
7326 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7330 /* Can return -ENOMEM */
7331 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7332 struct btrfs_fs_info *fs_info,
7333 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7334 u64 owner, u64 offset)
7336 int old_ref_mod, new_ref_mod;
7339 if (btrfs_is_testing(fs_info))
7344 * tree log blocks never actually go into the extent allocation
7345 * tree, just update pinning info and exit early.
7347 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7348 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7349 /* unlocks the pinned mutex */
7350 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7351 old_ref_mod = new_ref_mod = 0;
7353 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7354 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7356 root_objectid, (int)owner,
7357 BTRFS_DROP_DELAYED_REF, NULL,
7358 &old_ref_mod, &new_ref_mod);
7360 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7362 root_objectid, owner, offset,
7363 0, BTRFS_DROP_DELAYED_REF,
7364 &old_ref_mod, &new_ref_mod);
7367 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7368 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7374 * when we wait for progress in the block group caching, its because
7375 * our allocation attempt failed at least once. So, we must sleep
7376 * and let some progress happen before we try again.
7378 * This function will sleep at least once waiting for new free space to
7379 * show up, and then it will check the block group free space numbers
7380 * for our min num_bytes. Another option is to have it go ahead
7381 * and look in the rbtree for a free extent of a given size, but this
7384 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7385 * any of the information in this block group.
7387 static noinline void
7388 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7391 struct btrfs_caching_control *caching_ctl;
7393 caching_ctl = get_caching_control(cache);
7397 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7398 (cache->free_space_ctl->free_space >= num_bytes));
7400 put_caching_control(caching_ctl);
7404 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7406 struct btrfs_caching_control *caching_ctl;
7409 caching_ctl = get_caching_control(cache);
7411 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7413 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7414 if (cache->cached == BTRFS_CACHE_ERROR)
7416 put_caching_control(caching_ctl);
7420 int __get_raid_index(u64 flags)
7422 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7423 return BTRFS_RAID_RAID10;
7424 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7425 return BTRFS_RAID_RAID1;
7426 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7427 return BTRFS_RAID_DUP;
7428 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7429 return BTRFS_RAID_RAID0;
7430 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7431 return BTRFS_RAID_RAID5;
7432 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7433 return BTRFS_RAID_RAID6;
7435 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7438 int get_block_group_index(struct btrfs_block_group_cache *cache)
7440 return __get_raid_index(cache->flags);
7443 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7444 [BTRFS_RAID_RAID10] = "raid10",
7445 [BTRFS_RAID_RAID1] = "raid1",
7446 [BTRFS_RAID_DUP] = "dup",
7447 [BTRFS_RAID_RAID0] = "raid0",
7448 [BTRFS_RAID_SINGLE] = "single",
7449 [BTRFS_RAID_RAID5] = "raid5",
7450 [BTRFS_RAID_RAID6] = "raid6",
7453 static const char *get_raid_name(enum btrfs_raid_types type)
7455 if (type >= BTRFS_NR_RAID_TYPES)
7458 return btrfs_raid_type_names[type];
7461 enum btrfs_loop_type {
7462 LOOP_CACHING_NOWAIT = 0,
7463 LOOP_CACHING_WAIT = 1,
7464 LOOP_ALLOC_CHUNK = 2,
7465 LOOP_NO_EMPTY_SIZE = 3,
7469 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7473 down_read(&cache->data_rwsem);
7477 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7480 btrfs_get_block_group(cache);
7482 down_read(&cache->data_rwsem);
7485 static struct btrfs_block_group_cache *
7486 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7487 struct btrfs_free_cluster *cluster,
7490 struct btrfs_block_group_cache *used_bg = NULL;
7492 spin_lock(&cluster->refill_lock);
7494 used_bg = cluster->block_group;
7498 if (used_bg == block_group)
7501 btrfs_get_block_group(used_bg);
7506 if (down_read_trylock(&used_bg->data_rwsem))
7509 spin_unlock(&cluster->refill_lock);
7511 /* We should only have one-level nested. */
7512 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7514 spin_lock(&cluster->refill_lock);
7515 if (used_bg == cluster->block_group)
7518 up_read(&used_bg->data_rwsem);
7519 btrfs_put_block_group(used_bg);
7524 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7528 up_read(&cache->data_rwsem);
7529 btrfs_put_block_group(cache);
7533 * walks the btree of allocated extents and find a hole of a given size.
7534 * The key ins is changed to record the hole:
7535 * ins->objectid == start position
7536 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7537 * ins->offset == the size of the hole.
7538 * Any available blocks before search_start are skipped.
7540 * If there is no suitable free space, we will record the max size of
7541 * the free space extent currently.
7543 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7544 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7545 u64 hint_byte, struct btrfs_key *ins,
7546 u64 flags, int delalloc)
7549 struct btrfs_root *root = fs_info->extent_root;
7550 struct btrfs_free_cluster *last_ptr = NULL;
7551 struct btrfs_block_group_cache *block_group = NULL;
7552 u64 search_start = 0;
7553 u64 max_extent_size = 0;
7554 u64 empty_cluster = 0;
7555 struct btrfs_space_info *space_info;
7557 int index = __get_raid_index(flags);
7558 bool failed_cluster_refill = false;
7559 bool failed_alloc = false;
7560 bool use_cluster = true;
7561 bool have_caching_bg = false;
7562 bool orig_have_caching_bg = false;
7563 bool full_search = false;
7565 WARN_ON(num_bytes < fs_info->sectorsize);
7566 ins->type = BTRFS_EXTENT_ITEM_KEY;
7570 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7572 space_info = __find_space_info(fs_info, flags);
7574 btrfs_err(fs_info, "No space info for %llu", flags);
7579 * If our free space is heavily fragmented we may not be able to make
7580 * big contiguous allocations, so instead of doing the expensive search
7581 * for free space, simply return ENOSPC with our max_extent_size so we
7582 * can go ahead and search for a more manageable chunk.
7584 * If our max_extent_size is large enough for our allocation simply
7585 * disable clustering since we will likely not be able to find enough
7586 * space to create a cluster and induce latency trying.
7588 if (unlikely(space_info->max_extent_size)) {
7589 spin_lock(&space_info->lock);
7590 if (space_info->max_extent_size &&
7591 num_bytes > space_info->max_extent_size) {
7592 ins->offset = space_info->max_extent_size;
7593 spin_unlock(&space_info->lock);
7595 } else if (space_info->max_extent_size) {
7596 use_cluster = false;
7598 spin_unlock(&space_info->lock);
7601 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7603 spin_lock(&last_ptr->lock);
7604 if (last_ptr->block_group)
7605 hint_byte = last_ptr->window_start;
7606 if (last_ptr->fragmented) {
7608 * We still set window_start so we can keep track of the
7609 * last place we found an allocation to try and save
7612 hint_byte = last_ptr->window_start;
7613 use_cluster = false;
7615 spin_unlock(&last_ptr->lock);
7618 search_start = max(search_start, first_logical_byte(fs_info, 0));
7619 search_start = max(search_start, hint_byte);
7620 if (search_start == hint_byte) {
7621 block_group = btrfs_lookup_block_group(fs_info, search_start);
7623 * we don't want to use the block group if it doesn't match our
7624 * allocation bits, or if its not cached.
7626 * However if we are re-searching with an ideal block group
7627 * picked out then we don't care that the block group is cached.
7629 if (block_group && block_group_bits(block_group, flags) &&
7630 block_group->cached != BTRFS_CACHE_NO) {
7631 down_read(&space_info->groups_sem);
7632 if (list_empty(&block_group->list) ||
7635 * someone is removing this block group,
7636 * we can't jump into the have_block_group
7637 * target because our list pointers are not
7640 btrfs_put_block_group(block_group);
7641 up_read(&space_info->groups_sem);
7643 index = get_block_group_index(block_group);
7644 btrfs_lock_block_group(block_group, delalloc);
7645 goto have_block_group;
7647 } else if (block_group) {
7648 btrfs_put_block_group(block_group);
7652 have_caching_bg = false;
7653 if (index == 0 || index == __get_raid_index(flags))
7655 down_read(&space_info->groups_sem);
7656 list_for_each_entry(block_group, &space_info->block_groups[index],
7661 /* If the block group is read-only, we can skip it entirely. */
7662 if (unlikely(block_group->ro))
7665 btrfs_grab_block_group(block_group, delalloc);
7666 search_start = block_group->key.objectid;
7669 * this can happen if we end up cycling through all the
7670 * raid types, but we want to make sure we only allocate
7671 * for the proper type.
7673 if (!block_group_bits(block_group, flags)) {
7674 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7675 BTRFS_BLOCK_GROUP_RAID1 |
7676 BTRFS_BLOCK_GROUP_RAID5 |
7677 BTRFS_BLOCK_GROUP_RAID6 |
7678 BTRFS_BLOCK_GROUP_RAID10;
7681 * if they asked for extra copies and this block group
7682 * doesn't provide them, bail. This does allow us to
7683 * fill raid0 from raid1.
7685 if ((flags & extra) && !(block_group->flags & extra))
7690 cached = block_group_cache_done(block_group);
7691 if (unlikely(!cached)) {
7692 have_caching_bg = true;
7693 ret = cache_block_group(block_group, 0);
7698 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7702 * Ok we want to try and use the cluster allocator, so
7705 if (last_ptr && use_cluster) {
7706 struct btrfs_block_group_cache *used_block_group;
7707 unsigned long aligned_cluster;
7709 * the refill lock keeps out other
7710 * people trying to start a new cluster
7712 used_block_group = btrfs_lock_cluster(block_group,
7715 if (!used_block_group)
7716 goto refill_cluster;
7718 if (used_block_group != block_group &&
7719 (used_block_group->ro ||
7720 !block_group_bits(used_block_group, flags)))
7721 goto release_cluster;
7723 offset = btrfs_alloc_from_cluster(used_block_group,
7726 used_block_group->key.objectid,
7729 /* we have a block, we're done */
7730 spin_unlock(&last_ptr->refill_lock);
7731 trace_btrfs_reserve_extent_cluster(fs_info,
7733 search_start, num_bytes);
7734 if (used_block_group != block_group) {
7735 btrfs_release_block_group(block_group,
7737 block_group = used_block_group;
7742 WARN_ON(last_ptr->block_group != used_block_group);
7744 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7745 * set up a new clusters, so lets just skip it
7746 * and let the allocator find whatever block
7747 * it can find. If we reach this point, we
7748 * will have tried the cluster allocator
7749 * plenty of times and not have found
7750 * anything, so we are likely way too
7751 * fragmented for the clustering stuff to find
7754 * However, if the cluster is taken from the
7755 * current block group, release the cluster
7756 * first, so that we stand a better chance of
7757 * succeeding in the unclustered
7759 if (loop >= LOOP_NO_EMPTY_SIZE &&
7760 used_block_group != block_group) {
7761 spin_unlock(&last_ptr->refill_lock);
7762 btrfs_release_block_group(used_block_group,
7764 goto unclustered_alloc;
7768 * this cluster didn't work out, free it and
7771 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7773 if (used_block_group != block_group)
7774 btrfs_release_block_group(used_block_group,
7777 if (loop >= LOOP_NO_EMPTY_SIZE) {
7778 spin_unlock(&last_ptr->refill_lock);
7779 goto unclustered_alloc;
7782 aligned_cluster = max_t(unsigned long,
7783 empty_cluster + empty_size,
7784 block_group->full_stripe_len);
7786 /* allocate a cluster in this block group */
7787 ret = btrfs_find_space_cluster(fs_info, block_group,
7788 last_ptr, search_start,
7793 * now pull our allocation out of this
7796 offset = btrfs_alloc_from_cluster(block_group,
7802 /* we found one, proceed */
7803 spin_unlock(&last_ptr->refill_lock);
7804 trace_btrfs_reserve_extent_cluster(fs_info,
7805 block_group, search_start,
7809 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7810 && !failed_cluster_refill) {
7811 spin_unlock(&last_ptr->refill_lock);
7813 failed_cluster_refill = true;
7814 wait_block_group_cache_progress(block_group,
7815 num_bytes + empty_cluster + empty_size);
7816 goto have_block_group;
7820 * at this point we either didn't find a cluster
7821 * or we weren't able to allocate a block from our
7822 * cluster. Free the cluster we've been trying
7823 * to use, and go to the next block group
7825 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7826 spin_unlock(&last_ptr->refill_lock);
7832 * We are doing an unclustered alloc, set the fragmented flag so
7833 * we don't bother trying to setup a cluster again until we get
7836 if (unlikely(last_ptr)) {
7837 spin_lock(&last_ptr->lock);
7838 last_ptr->fragmented = 1;
7839 spin_unlock(&last_ptr->lock);
7842 struct btrfs_free_space_ctl *ctl =
7843 block_group->free_space_ctl;
7845 spin_lock(&ctl->tree_lock);
7846 if (ctl->free_space <
7847 num_bytes + empty_cluster + empty_size) {
7848 if (ctl->free_space > max_extent_size)
7849 max_extent_size = ctl->free_space;
7850 spin_unlock(&ctl->tree_lock);
7853 spin_unlock(&ctl->tree_lock);
7856 offset = btrfs_find_space_for_alloc(block_group, search_start,
7857 num_bytes, empty_size,
7860 * If we didn't find a chunk, and we haven't failed on this
7861 * block group before, and this block group is in the middle of
7862 * caching and we are ok with waiting, then go ahead and wait
7863 * for progress to be made, and set failed_alloc to true.
7865 * If failed_alloc is true then we've already waited on this
7866 * block group once and should move on to the next block group.
7868 if (!offset && !failed_alloc && !cached &&
7869 loop > LOOP_CACHING_NOWAIT) {
7870 wait_block_group_cache_progress(block_group,
7871 num_bytes + empty_size);
7872 failed_alloc = true;
7873 goto have_block_group;
7874 } else if (!offset) {
7878 search_start = ALIGN(offset, fs_info->stripesize);
7880 /* move on to the next group */
7881 if (search_start + num_bytes >
7882 block_group->key.objectid + block_group->key.offset) {
7883 btrfs_add_free_space(block_group, offset, num_bytes);
7887 if (offset < search_start)
7888 btrfs_add_free_space(block_group, offset,
7889 search_start - offset);
7890 BUG_ON(offset > search_start);
7892 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7893 num_bytes, delalloc);
7894 if (ret == -EAGAIN) {
7895 btrfs_add_free_space(block_group, offset, num_bytes);
7898 btrfs_inc_block_group_reservations(block_group);
7900 /* we are all good, lets return */
7901 ins->objectid = search_start;
7902 ins->offset = num_bytes;
7904 trace_btrfs_reserve_extent(fs_info, block_group,
7905 search_start, num_bytes);
7906 btrfs_release_block_group(block_group, delalloc);
7909 failed_cluster_refill = false;
7910 failed_alloc = false;
7911 BUG_ON(index != get_block_group_index(block_group));
7912 btrfs_release_block_group(block_group, delalloc);
7915 up_read(&space_info->groups_sem);
7917 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7918 && !orig_have_caching_bg)
7919 orig_have_caching_bg = true;
7921 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7924 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7928 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7929 * caching kthreads as we move along
7930 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7931 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7932 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7935 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7937 if (loop == LOOP_CACHING_NOWAIT) {
7939 * We want to skip the LOOP_CACHING_WAIT step if we
7940 * don't have any uncached bgs and we've already done a
7941 * full search through.
7943 if (orig_have_caching_bg || !full_search)
7944 loop = LOOP_CACHING_WAIT;
7946 loop = LOOP_ALLOC_CHUNK;
7951 if (loop == LOOP_ALLOC_CHUNK) {
7952 struct btrfs_trans_handle *trans;
7955 trans = current->journal_info;
7959 trans = btrfs_join_transaction(root);
7961 if (IS_ERR(trans)) {
7962 ret = PTR_ERR(trans);
7966 ret = do_chunk_alloc(trans, fs_info, flags,
7970 * If we can't allocate a new chunk we've already looped
7971 * through at least once, move on to the NO_EMPTY_SIZE
7975 loop = LOOP_NO_EMPTY_SIZE;
7978 * Do not bail out on ENOSPC since we
7979 * can do more things.
7981 if (ret < 0 && ret != -ENOSPC)
7982 btrfs_abort_transaction(trans, ret);
7986 btrfs_end_transaction(trans);
7991 if (loop == LOOP_NO_EMPTY_SIZE) {
7993 * Don't loop again if we already have no empty_size and
7996 if (empty_size == 0 &&
7997 empty_cluster == 0) {
8006 } else if (!ins->objectid) {
8008 } else if (ins->objectid) {
8009 if (!use_cluster && last_ptr) {
8010 spin_lock(&last_ptr->lock);
8011 last_ptr->window_start = ins->objectid;
8012 spin_unlock(&last_ptr->lock);
8017 if (ret == -ENOSPC) {
8018 spin_lock(&space_info->lock);
8019 space_info->max_extent_size = max_extent_size;
8020 spin_unlock(&space_info->lock);
8021 ins->offset = max_extent_size;
8026 static void dump_space_info(struct btrfs_fs_info *fs_info,
8027 struct btrfs_space_info *info, u64 bytes,
8028 int dump_block_groups)
8030 struct btrfs_block_group_cache *cache;
8033 spin_lock(&info->lock);
8034 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
8036 info->total_bytes - btrfs_space_info_used(info, true),
8037 info->full ? "" : "not ");
8039 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
8040 info->total_bytes, info->bytes_used, info->bytes_pinned,
8041 info->bytes_reserved, info->bytes_may_use,
8042 info->bytes_readonly);
8043 spin_unlock(&info->lock);
8045 if (!dump_block_groups)
8048 down_read(&info->groups_sem);
8050 list_for_each_entry(cache, &info->block_groups[index], list) {
8051 spin_lock(&cache->lock);
8053 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
8054 cache->key.objectid, cache->key.offset,
8055 btrfs_block_group_used(&cache->item), cache->pinned,
8056 cache->reserved, cache->ro ? "[readonly]" : "");
8057 btrfs_dump_free_space(cache, bytes);
8058 spin_unlock(&cache->lock);
8060 if (++index < BTRFS_NR_RAID_TYPES)
8062 up_read(&info->groups_sem);
8065 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8066 u64 num_bytes, u64 min_alloc_size,
8067 u64 empty_size, u64 hint_byte,
8068 struct btrfs_key *ins, int is_data, int delalloc)
8070 struct btrfs_fs_info *fs_info = root->fs_info;
8071 bool final_tried = num_bytes == min_alloc_size;
8075 flags = get_alloc_profile_by_root(root, is_data);
8077 WARN_ON(num_bytes < fs_info->sectorsize);
8078 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8079 hint_byte, ins, flags, delalloc);
8080 if (!ret && !is_data) {
8081 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8082 } else if (ret == -ENOSPC) {
8083 if (!final_tried && ins->offset) {
8084 num_bytes = min(num_bytes >> 1, ins->offset);
8085 num_bytes = round_down(num_bytes,
8086 fs_info->sectorsize);
8087 num_bytes = max(num_bytes, min_alloc_size);
8088 ram_bytes = num_bytes;
8089 if (num_bytes == min_alloc_size)
8092 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8093 struct btrfs_space_info *sinfo;
8095 sinfo = __find_space_info(fs_info, flags);
8097 "allocation failed flags %llu, wanted %llu",
8100 dump_space_info(fs_info, sinfo, num_bytes, 1);
8107 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8109 int pin, int delalloc)
8111 struct btrfs_block_group_cache *cache;
8114 cache = btrfs_lookup_block_group(fs_info, start);
8116 btrfs_err(fs_info, "Unable to find block group for %llu",
8122 pin_down_extent(fs_info, cache, start, len, 1);
8124 if (btrfs_test_opt(fs_info, DISCARD))
8125 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8126 btrfs_add_free_space(cache, start, len);
8127 btrfs_free_reserved_bytes(cache, len, delalloc);
8128 trace_btrfs_reserved_extent_free(fs_info, start, len);
8131 btrfs_put_block_group(cache);
8135 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8136 u64 start, u64 len, int delalloc)
8138 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8141 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8144 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8147 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8148 struct btrfs_fs_info *fs_info,
8149 u64 parent, u64 root_objectid,
8150 u64 flags, u64 owner, u64 offset,
8151 struct btrfs_key *ins, int ref_mod)
8154 struct btrfs_extent_item *extent_item;
8155 struct btrfs_extent_inline_ref *iref;
8156 struct btrfs_path *path;
8157 struct extent_buffer *leaf;
8162 type = BTRFS_SHARED_DATA_REF_KEY;
8164 type = BTRFS_EXTENT_DATA_REF_KEY;
8166 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8168 path = btrfs_alloc_path();
8172 path->leave_spinning = 1;
8173 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8176 btrfs_free_path(path);
8180 leaf = path->nodes[0];
8181 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8182 struct btrfs_extent_item);
8183 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8184 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8185 btrfs_set_extent_flags(leaf, extent_item,
8186 flags | BTRFS_EXTENT_FLAG_DATA);
8188 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8189 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8191 struct btrfs_shared_data_ref *ref;
8192 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8193 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8194 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8196 struct btrfs_extent_data_ref *ref;
8197 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8198 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8199 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8200 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8201 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8204 btrfs_mark_buffer_dirty(path->nodes[0]);
8205 btrfs_free_path(path);
8207 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8212 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8213 if (ret) { /* -ENOENT, logic error */
8214 btrfs_err(fs_info, "update block group failed for %llu %llu",
8215 ins->objectid, ins->offset);
8218 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8222 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8223 struct btrfs_fs_info *fs_info,
8224 u64 parent, u64 root_objectid,
8225 u64 flags, struct btrfs_disk_key *key,
8226 int level, struct btrfs_key *ins)
8229 struct btrfs_extent_item *extent_item;
8230 struct btrfs_tree_block_info *block_info;
8231 struct btrfs_extent_inline_ref *iref;
8232 struct btrfs_path *path;
8233 struct extent_buffer *leaf;
8234 u32 size = sizeof(*extent_item) + sizeof(*iref);
8235 u64 num_bytes = ins->offset;
8236 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8238 if (!skinny_metadata)
8239 size += sizeof(*block_info);
8241 path = btrfs_alloc_path();
8243 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8248 path->leave_spinning = 1;
8249 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8252 btrfs_free_path(path);
8253 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8258 leaf = path->nodes[0];
8259 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8260 struct btrfs_extent_item);
8261 btrfs_set_extent_refs(leaf, extent_item, 1);
8262 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8263 btrfs_set_extent_flags(leaf, extent_item,
8264 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8266 if (skinny_metadata) {
8267 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8268 num_bytes = fs_info->nodesize;
8270 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8271 btrfs_set_tree_block_key(leaf, block_info, key);
8272 btrfs_set_tree_block_level(leaf, block_info, level);
8273 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8277 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8278 btrfs_set_extent_inline_ref_type(leaf, iref,
8279 BTRFS_SHARED_BLOCK_REF_KEY);
8280 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8282 btrfs_set_extent_inline_ref_type(leaf, iref,
8283 BTRFS_TREE_BLOCK_REF_KEY);
8284 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8287 btrfs_mark_buffer_dirty(leaf);
8288 btrfs_free_path(path);
8290 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8295 ret = update_block_group(trans, fs_info, ins->objectid,
8296 fs_info->nodesize, 1);
8297 if (ret) { /* -ENOENT, logic error */
8298 btrfs_err(fs_info, "update block group failed for %llu %llu",
8299 ins->objectid, ins->offset);
8303 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8308 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8309 u64 root_objectid, u64 owner,
8310 u64 offset, u64 ram_bytes,
8311 struct btrfs_key *ins)
8313 struct btrfs_fs_info *fs_info = trans->fs_info;
8316 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8318 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8319 ins->offset, 0, root_objectid, owner,
8321 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8326 * this is used by the tree logging recovery code. It records that
8327 * an extent has been allocated and makes sure to clear the free
8328 * space cache bits as well
8330 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8331 struct btrfs_fs_info *fs_info,
8332 u64 root_objectid, u64 owner, u64 offset,
8333 struct btrfs_key *ins)
8336 struct btrfs_block_group_cache *block_group;
8337 struct btrfs_space_info *space_info;
8340 * Mixed block groups will exclude before processing the log so we only
8341 * need to do the exclude dance if this fs isn't mixed.
8343 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8344 ret = __exclude_logged_extent(fs_info, ins->objectid,
8350 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8354 space_info = block_group->space_info;
8355 spin_lock(&space_info->lock);
8356 spin_lock(&block_group->lock);
8357 space_info->bytes_reserved += ins->offset;
8358 block_group->reserved += ins->offset;
8359 spin_unlock(&block_group->lock);
8360 spin_unlock(&space_info->lock);
8362 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8363 0, owner, offset, ins, 1);
8364 btrfs_put_block_group(block_group);
8368 static struct extent_buffer *
8369 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8370 u64 bytenr, int level)
8372 struct btrfs_fs_info *fs_info = root->fs_info;
8373 struct extent_buffer *buf;
8375 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8379 btrfs_set_header_generation(buf, trans->transid);
8380 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8381 btrfs_tree_lock(buf);
8382 clean_tree_block(fs_info, buf);
8383 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8385 btrfs_set_lock_blocking(buf);
8386 set_extent_buffer_uptodate(buf);
8388 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8389 buf->log_index = root->log_transid % 2;
8391 * we allow two log transactions at a time, use different
8392 * EXENT bit to differentiate dirty pages.
8394 if (buf->log_index == 0)
8395 set_extent_dirty(&root->dirty_log_pages, buf->start,
8396 buf->start + buf->len - 1, GFP_NOFS);
8398 set_extent_new(&root->dirty_log_pages, buf->start,
8399 buf->start + buf->len - 1);
8401 buf->log_index = -1;
8402 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8403 buf->start + buf->len - 1, GFP_NOFS);
8405 trans->dirty = true;
8406 /* this returns a buffer locked for blocking */
8410 static struct btrfs_block_rsv *
8411 use_block_rsv(struct btrfs_trans_handle *trans,
8412 struct btrfs_root *root, u32 blocksize)
8414 struct btrfs_fs_info *fs_info = root->fs_info;
8415 struct btrfs_block_rsv *block_rsv;
8416 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8418 bool global_updated = false;
8420 block_rsv = get_block_rsv(trans, root);
8422 if (unlikely(block_rsv->size == 0))
8425 ret = block_rsv_use_bytes(block_rsv, blocksize);
8429 if (block_rsv->failfast)
8430 return ERR_PTR(ret);
8432 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8433 global_updated = true;
8434 update_global_block_rsv(fs_info);
8438 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8439 static DEFINE_RATELIMIT_STATE(_rs,
8440 DEFAULT_RATELIMIT_INTERVAL * 10,
8441 /*DEFAULT_RATELIMIT_BURST*/ 1);
8442 if (__ratelimit(&_rs))
8444 "BTRFS: block rsv returned %d\n", ret);
8447 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8448 BTRFS_RESERVE_NO_FLUSH);
8452 * If we couldn't reserve metadata bytes try and use some from
8453 * the global reserve if its space type is the same as the global
8456 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8457 block_rsv->space_info == global_rsv->space_info) {
8458 ret = block_rsv_use_bytes(global_rsv, blocksize);
8462 return ERR_PTR(ret);
8465 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8466 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8468 block_rsv_add_bytes(block_rsv, blocksize, 0);
8469 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8473 * finds a free extent and does all the dirty work required for allocation
8474 * returns the tree buffer or an ERR_PTR on error.
8476 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8477 struct btrfs_root *root,
8478 u64 parent, u64 root_objectid,
8479 const struct btrfs_disk_key *key,
8480 int level, u64 hint,
8483 struct btrfs_fs_info *fs_info = root->fs_info;
8484 struct btrfs_key ins;
8485 struct btrfs_block_rsv *block_rsv;
8486 struct extent_buffer *buf;
8487 struct btrfs_delayed_extent_op *extent_op;
8490 u32 blocksize = fs_info->nodesize;
8491 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8493 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8494 if (btrfs_is_testing(fs_info)) {
8495 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8498 root->alloc_bytenr += blocksize;
8503 block_rsv = use_block_rsv(trans, root, blocksize);
8504 if (IS_ERR(block_rsv))
8505 return ERR_CAST(block_rsv);
8507 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8508 empty_size, hint, &ins, 0, 0);
8512 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8515 goto out_free_reserved;
8518 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8520 parent = ins.objectid;
8521 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8525 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8526 extent_op = btrfs_alloc_delayed_extent_op();
8532 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8534 memset(&extent_op->key, 0, sizeof(extent_op->key));
8535 extent_op->flags_to_set = flags;
8536 extent_op->update_key = skinny_metadata ? false : true;
8537 extent_op->update_flags = true;
8538 extent_op->is_data = false;
8539 extent_op->level = level;
8541 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8543 root_objectid, level,
8544 BTRFS_ADD_DELAYED_EXTENT,
8545 extent_op, NULL, NULL);
8547 goto out_free_delayed;
8552 btrfs_free_delayed_extent_op(extent_op);
8554 free_extent_buffer(buf);
8556 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8558 unuse_block_rsv(fs_info, block_rsv, blocksize);
8559 return ERR_PTR(ret);
8562 struct walk_control {
8563 u64 refs[BTRFS_MAX_LEVEL];
8564 u64 flags[BTRFS_MAX_LEVEL];
8565 struct btrfs_key update_progress;
8576 #define DROP_REFERENCE 1
8577 #define UPDATE_BACKREF 2
8579 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8580 struct btrfs_root *root,
8581 struct walk_control *wc,
8582 struct btrfs_path *path)
8584 struct btrfs_fs_info *fs_info = root->fs_info;
8590 struct btrfs_key key;
8591 struct extent_buffer *eb;
8596 if (path->slots[wc->level] < wc->reada_slot) {
8597 wc->reada_count = wc->reada_count * 2 / 3;
8598 wc->reada_count = max(wc->reada_count, 2);
8600 wc->reada_count = wc->reada_count * 3 / 2;
8601 wc->reada_count = min_t(int, wc->reada_count,
8602 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8605 eb = path->nodes[wc->level];
8606 nritems = btrfs_header_nritems(eb);
8608 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8609 if (nread >= wc->reada_count)
8613 bytenr = btrfs_node_blockptr(eb, slot);
8614 generation = btrfs_node_ptr_generation(eb, slot);
8616 if (slot == path->slots[wc->level])
8619 if (wc->stage == UPDATE_BACKREF &&
8620 generation <= root->root_key.offset)
8623 /* We don't lock the tree block, it's OK to be racy here */
8624 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8625 wc->level - 1, 1, &refs,
8627 /* We don't care about errors in readahead. */
8632 if (wc->stage == DROP_REFERENCE) {
8636 if (wc->level == 1 &&
8637 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8639 if (!wc->update_ref ||
8640 generation <= root->root_key.offset)
8642 btrfs_node_key_to_cpu(eb, &key, slot);
8643 ret = btrfs_comp_cpu_keys(&key,
8644 &wc->update_progress);
8648 if (wc->level == 1 &&
8649 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8653 readahead_tree_block(fs_info, bytenr);
8656 wc->reada_slot = slot;
8660 * helper to process tree block while walking down the tree.
8662 * when wc->stage == UPDATE_BACKREF, this function updates
8663 * back refs for pointers in the block.
8665 * NOTE: return value 1 means we should stop walking down.
8667 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8668 struct btrfs_root *root,
8669 struct btrfs_path *path,
8670 struct walk_control *wc, int lookup_info)
8672 struct btrfs_fs_info *fs_info = root->fs_info;
8673 int level = wc->level;
8674 struct extent_buffer *eb = path->nodes[level];
8675 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8678 if (wc->stage == UPDATE_BACKREF &&
8679 btrfs_header_owner(eb) != root->root_key.objectid)
8683 * when reference count of tree block is 1, it won't increase
8684 * again. once full backref flag is set, we never clear it.
8687 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8688 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8689 BUG_ON(!path->locks[level]);
8690 ret = btrfs_lookup_extent_info(trans, fs_info,
8691 eb->start, level, 1,
8694 BUG_ON(ret == -ENOMEM);
8697 BUG_ON(wc->refs[level] == 0);
8700 if (wc->stage == DROP_REFERENCE) {
8701 if (wc->refs[level] > 1)
8704 if (path->locks[level] && !wc->keep_locks) {
8705 btrfs_tree_unlock_rw(eb, path->locks[level]);
8706 path->locks[level] = 0;
8711 /* wc->stage == UPDATE_BACKREF */
8712 if (!(wc->flags[level] & flag)) {
8713 BUG_ON(!path->locks[level]);
8714 ret = btrfs_inc_ref(trans, root, eb, 1);
8715 BUG_ON(ret); /* -ENOMEM */
8716 ret = btrfs_dec_ref(trans, root, eb, 0);
8717 BUG_ON(ret); /* -ENOMEM */
8718 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8720 btrfs_header_level(eb), 0);
8721 BUG_ON(ret); /* -ENOMEM */
8722 wc->flags[level] |= flag;
8726 * the block is shared by multiple trees, so it's not good to
8727 * keep the tree lock
8729 if (path->locks[level] && level > 0) {
8730 btrfs_tree_unlock_rw(eb, path->locks[level]);
8731 path->locks[level] = 0;
8737 * helper to process tree block pointer.
8739 * when wc->stage == DROP_REFERENCE, this function checks
8740 * reference count of the block pointed to. if the block
8741 * is shared and we need update back refs for the subtree
8742 * rooted at the block, this function changes wc->stage to
8743 * UPDATE_BACKREF. if the block is shared and there is no
8744 * need to update back, this function drops the reference
8747 * NOTE: return value 1 means we should stop walking down.
8749 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8750 struct btrfs_root *root,
8751 struct btrfs_path *path,
8752 struct walk_control *wc, int *lookup_info)
8754 struct btrfs_fs_info *fs_info = root->fs_info;
8759 struct btrfs_key key;
8760 struct extent_buffer *next;
8761 int level = wc->level;
8764 bool need_account = false;
8766 generation = btrfs_node_ptr_generation(path->nodes[level],
8767 path->slots[level]);
8769 * if the lower level block was created before the snapshot
8770 * was created, we know there is no need to update back refs
8773 if (wc->stage == UPDATE_BACKREF &&
8774 generation <= root->root_key.offset) {
8779 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8780 blocksize = fs_info->nodesize;
8782 next = find_extent_buffer(fs_info, bytenr);
8784 next = btrfs_find_create_tree_block(fs_info, bytenr);
8786 return PTR_ERR(next);
8788 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8792 btrfs_tree_lock(next);
8793 btrfs_set_lock_blocking(next);
8795 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8796 &wc->refs[level - 1],
8797 &wc->flags[level - 1]);
8801 if (unlikely(wc->refs[level - 1] == 0)) {
8802 btrfs_err(fs_info, "Missing references.");
8808 if (wc->stage == DROP_REFERENCE) {
8809 if (wc->refs[level - 1] > 1) {
8810 need_account = true;
8812 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8815 if (!wc->update_ref ||
8816 generation <= root->root_key.offset)
8819 btrfs_node_key_to_cpu(path->nodes[level], &key,
8820 path->slots[level]);
8821 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8825 wc->stage = UPDATE_BACKREF;
8826 wc->shared_level = level - 1;
8830 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8834 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8835 btrfs_tree_unlock(next);
8836 free_extent_buffer(next);
8842 if (reada && level == 1)
8843 reada_walk_down(trans, root, wc, path);
8844 next = read_tree_block(fs_info, bytenr, generation);
8846 return PTR_ERR(next);
8847 } else if (!extent_buffer_uptodate(next)) {
8848 free_extent_buffer(next);
8851 btrfs_tree_lock(next);
8852 btrfs_set_lock_blocking(next);
8856 ASSERT(level == btrfs_header_level(next));
8857 if (level != btrfs_header_level(next)) {
8858 btrfs_err(root->fs_info, "mismatched level");
8862 path->nodes[level] = next;
8863 path->slots[level] = 0;
8864 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8870 wc->refs[level - 1] = 0;
8871 wc->flags[level - 1] = 0;
8872 if (wc->stage == DROP_REFERENCE) {
8873 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8874 parent = path->nodes[level]->start;
8876 ASSERT(root->root_key.objectid ==
8877 btrfs_header_owner(path->nodes[level]));
8878 if (root->root_key.objectid !=
8879 btrfs_header_owner(path->nodes[level])) {
8880 btrfs_err(root->fs_info,
8881 "mismatched block owner");
8889 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8890 generation, level - 1);
8892 btrfs_err_rl(fs_info,
8893 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8897 ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize,
8898 parent, root->root_key.objectid,
8908 btrfs_tree_unlock(next);
8909 free_extent_buffer(next);
8915 * helper to process tree block while walking up the tree.
8917 * when wc->stage == DROP_REFERENCE, this function drops
8918 * reference count on the block.
8920 * when wc->stage == UPDATE_BACKREF, this function changes
8921 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8922 * to UPDATE_BACKREF previously while processing the block.
8924 * NOTE: return value 1 means we should stop walking up.
8926 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8927 struct btrfs_root *root,
8928 struct btrfs_path *path,
8929 struct walk_control *wc)
8931 struct btrfs_fs_info *fs_info = root->fs_info;
8933 int level = wc->level;
8934 struct extent_buffer *eb = path->nodes[level];
8937 if (wc->stage == UPDATE_BACKREF) {
8938 BUG_ON(wc->shared_level < level);
8939 if (level < wc->shared_level)
8942 ret = find_next_key(path, level + 1, &wc->update_progress);
8946 wc->stage = DROP_REFERENCE;
8947 wc->shared_level = -1;
8948 path->slots[level] = 0;
8951 * check reference count again if the block isn't locked.
8952 * we should start walking down the tree again if reference
8955 if (!path->locks[level]) {
8957 btrfs_tree_lock(eb);
8958 btrfs_set_lock_blocking(eb);
8959 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8961 ret = btrfs_lookup_extent_info(trans, fs_info,
8962 eb->start, level, 1,
8966 btrfs_tree_unlock_rw(eb, path->locks[level]);
8967 path->locks[level] = 0;
8970 BUG_ON(wc->refs[level] == 0);
8971 if (wc->refs[level] == 1) {
8972 btrfs_tree_unlock_rw(eb, path->locks[level]);
8973 path->locks[level] = 0;
8979 /* wc->stage == DROP_REFERENCE */
8980 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8982 if (wc->refs[level] == 1) {
8984 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8985 ret = btrfs_dec_ref(trans, root, eb, 1);
8987 ret = btrfs_dec_ref(trans, root, eb, 0);
8988 BUG_ON(ret); /* -ENOMEM */
8989 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8991 btrfs_err_rl(fs_info,
8992 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8996 /* make block locked assertion in clean_tree_block happy */
8997 if (!path->locks[level] &&
8998 btrfs_header_generation(eb) == trans->transid) {
8999 btrfs_tree_lock(eb);
9000 btrfs_set_lock_blocking(eb);
9001 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9003 clean_tree_block(fs_info, eb);
9006 if (eb == root->node) {
9007 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9010 BUG_ON(root->root_key.objectid !=
9011 btrfs_header_owner(eb));
9013 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9014 parent = path->nodes[level + 1]->start;
9016 BUG_ON(root->root_key.objectid !=
9017 btrfs_header_owner(path->nodes[level + 1]));
9020 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9022 wc->refs[level] = 0;
9023 wc->flags[level] = 0;
9027 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9028 struct btrfs_root *root,
9029 struct btrfs_path *path,
9030 struct walk_control *wc)
9032 int level = wc->level;
9033 int lookup_info = 1;
9036 while (level >= 0) {
9037 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9044 if (path->slots[level] >=
9045 btrfs_header_nritems(path->nodes[level]))
9048 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9050 path->slots[level]++;
9059 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9060 struct btrfs_root *root,
9061 struct btrfs_path *path,
9062 struct walk_control *wc, int max_level)
9064 int level = wc->level;
9067 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9068 while (level < max_level && path->nodes[level]) {
9070 if (path->slots[level] + 1 <
9071 btrfs_header_nritems(path->nodes[level])) {
9072 path->slots[level]++;
9075 ret = walk_up_proc(trans, root, path, wc);
9079 if (path->locks[level]) {
9080 btrfs_tree_unlock_rw(path->nodes[level],
9081 path->locks[level]);
9082 path->locks[level] = 0;
9084 free_extent_buffer(path->nodes[level]);
9085 path->nodes[level] = NULL;
9093 * drop a subvolume tree.
9095 * this function traverses the tree freeing any blocks that only
9096 * referenced by the tree.
9098 * when a shared tree block is found. this function decreases its
9099 * reference count by one. if update_ref is true, this function
9100 * also make sure backrefs for the shared block and all lower level
9101 * blocks are properly updated.
9103 * If called with for_reloc == 0, may exit early with -EAGAIN
9105 int btrfs_drop_snapshot(struct btrfs_root *root,
9106 struct btrfs_block_rsv *block_rsv, int update_ref,
9109 struct btrfs_fs_info *fs_info = root->fs_info;
9110 struct btrfs_path *path;
9111 struct btrfs_trans_handle *trans;
9112 struct btrfs_root *tree_root = fs_info->tree_root;
9113 struct btrfs_root_item *root_item = &root->root_item;
9114 struct walk_control *wc;
9115 struct btrfs_key key;
9119 bool root_dropped = false;
9121 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9123 path = btrfs_alloc_path();
9129 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9131 btrfs_free_path(path);
9136 trans = btrfs_start_transaction(tree_root, 0);
9137 if (IS_ERR(trans)) {
9138 err = PTR_ERR(trans);
9143 trans->block_rsv = block_rsv;
9145 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9146 level = btrfs_header_level(root->node);
9147 path->nodes[level] = btrfs_lock_root_node(root);
9148 btrfs_set_lock_blocking(path->nodes[level]);
9149 path->slots[level] = 0;
9150 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9151 memset(&wc->update_progress, 0,
9152 sizeof(wc->update_progress));
9154 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9155 memcpy(&wc->update_progress, &key,
9156 sizeof(wc->update_progress));
9158 level = root_item->drop_level;
9160 path->lowest_level = level;
9161 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9162 path->lowest_level = 0;
9170 * unlock our path, this is safe because only this
9171 * function is allowed to delete this snapshot
9173 btrfs_unlock_up_safe(path, 0);
9175 level = btrfs_header_level(root->node);
9177 btrfs_tree_lock(path->nodes[level]);
9178 btrfs_set_lock_blocking(path->nodes[level]);
9179 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9181 ret = btrfs_lookup_extent_info(trans, fs_info,
9182 path->nodes[level]->start,
9183 level, 1, &wc->refs[level],
9189 BUG_ON(wc->refs[level] == 0);
9191 if (level == root_item->drop_level)
9194 btrfs_tree_unlock(path->nodes[level]);
9195 path->locks[level] = 0;
9196 WARN_ON(wc->refs[level] != 1);
9202 wc->shared_level = -1;
9203 wc->stage = DROP_REFERENCE;
9204 wc->update_ref = update_ref;
9206 wc->for_reloc = for_reloc;
9207 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9211 ret = walk_down_tree(trans, root, path, wc);
9217 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9224 BUG_ON(wc->stage != DROP_REFERENCE);
9228 if (wc->stage == DROP_REFERENCE) {
9230 btrfs_node_key(path->nodes[level],
9231 &root_item->drop_progress,
9232 path->slots[level]);
9233 root_item->drop_level = level;
9236 BUG_ON(wc->level == 0);
9237 if (btrfs_should_end_transaction(trans) ||
9238 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9239 ret = btrfs_update_root(trans, tree_root,
9243 btrfs_abort_transaction(trans, ret);
9248 btrfs_end_transaction_throttle(trans);
9249 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9250 btrfs_debug(fs_info,
9251 "drop snapshot early exit");
9256 trans = btrfs_start_transaction(tree_root, 0);
9257 if (IS_ERR(trans)) {
9258 err = PTR_ERR(trans);
9262 trans->block_rsv = block_rsv;
9265 btrfs_release_path(path);
9269 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9271 btrfs_abort_transaction(trans, ret);
9275 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9276 ret = btrfs_find_root(tree_root, &root->root_key, path,
9279 btrfs_abort_transaction(trans, ret);
9282 } else if (ret > 0) {
9283 /* if we fail to delete the orphan item this time
9284 * around, it'll get picked up the next time.
9286 * The most common failure here is just -ENOENT.
9288 btrfs_del_orphan_item(trans, tree_root,
9289 root->root_key.objectid);
9293 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9294 btrfs_add_dropped_root(trans, root);
9296 free_extent_buffer(root->node);
9297 free_extent_buffer(root->commit_root);
9298 btrfs_put_fs_root(root);
9300 root_dropped = true;
9302 btrfs_end_transaction_throttle(trans);
9305 btrfs_free_path(path);
9308 * So if we need to stop dropping the snapshot for whatever reason we
9309 * need to make sure to add it back to the dead root list so that we
9310 * keep trying to do the work later. This also cleans up roots if we
9311 * don't have it in the radix (like when we recover after a power fail
9312 * or unmount) so we don't leak memory.
9314 if (!for_reloc && root_dropped == false)
9315 btrfs_add_dead_root(root);
9316 if (err && err != -EAGAIN)
9317 btrfs_handle_fs_error(fs_info, err, NULL);
9322 * drop subtree rooted at tree block 'node'.
9324 * NOTE: this function will unlock and release tree block 'node'
9325 * only used by relocation code
9327 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9328 struct btrfs_root *root,
9329 struct extent_buffer *node,
9330 struct extent_buffer *parent)
9332 struct btrfs_fs_info *fs_info = root->fs_info;
9333 struct btrfs_path *path;
9334 struct walk_control *wc;
9340 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9342 path = btrfs_alloc_path();
9346 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9348 btrfs_free_path(path);
9352 btrfs_assert_tree_locked(parent);
9353 parent_level = btrfs_header_level(parent);
9354 extent_buffer_get(parent);
9355 path->nodes[parent_level] = parent;
9356 path->slots[parent_level] = btrfs_header_nritems(parent);
9358 btrfs_assert_tree_locked(node);
9359 level = btrfs_header_level(node);
9360 path->nodes[level] = node;
9361 path->slots[level] = 0;
9362 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9364 wc->refs[parent_level] = 1;
9365 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9367 wc->shared_level = -1;
9368 wc->stage = DROP_REFERENCE;
9372 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9375 wret = walk_down_tree(trans, root, path, wc);
9381 wret = walk_up_tree(trans, root, path, wc, parent_level);
9389 btrfs_free_path(path);
9393 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9399 * if restripe for this chunk_type is on pick target profile and
9400 * return, otherwise do the usual balance
9402 stripped = get_restripe_target(fs_info, flags);
9404 return extended_to_chunk(stripped);
9406 num_devices = fs_info->fs_devices->rw_devices;
9408 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9409 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9410 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9412 if (num_devices == 1) {
9413 stripped |= BTRFS_BLOCK_GROUP_DUP;
9414 stripped = flags & ~stripped;
9416 /* turn raid0 into single device chunks */
9417 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9420 /* turn mirroring into duplication */
9421 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9422 BTRFS_BLOCK_GROUP_RAID10))
9423 return stripped | BTRFS_BLOCK_GROUP_DUP;
9425 /* they already had raid on here, just return */
9426 if (flags & stripped)
9429 stripped |= BTRFS_BLOCK_GROUP_DUP;
9430 stripped = flags & ~stripped;
9432 /* switch duplicated blocks with raid1 */
9433 if (flags & BTRFS_BLOCK_GROUP_DUP)
9434 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9436 /* this is drive concat, leave it alone */
9442 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9444 struct btrfs_space_info *sinfo = cache->space_info;
9446 u64 min_allocable_bytes;
9450 * We need some metadata space and system metadata space for
9451 * allocating chunks in some corner cases until we force to set
9452 * it to be readonly.
9455 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9457 min_allocable_bytes = SZ_1M;
9459 min_allocable_bytes = 0;
9461 spin_lock(&sinfo->lock);
9462 spin_lock(&cache->lock);
9470 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9471 cache->bytes_super - btrfs_block_group_used(&cache->item);
9473 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9474 min_allocable_bytes <= sinfo->total_bytes) {
9475 sinfo->bytes_readonly += num_bytes;
9477 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9481 spin_unlock(&cache->lock);
9482 spin_unlock(&sinfo->lock);
9486 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9487 struct btrfs_block_group_cache *cache)
9490 struct btrfs_trans_handle *trans;
9495 trans = btrfs_join_transaction(fs_info->extent_root);
9497 return PTR_ERR(trans);
9500 * we're not allowed to set block groups readonly after the dirty
9501 * block groups cache has started writing. If it already started,
9502 * back off and let this transaction commit
9504 mutex_lock(&fs_info->ro_block_group_mutex);
9505 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9506 u64 transid = trans->transid;
9508 mutex_unlock(&fs_info->ro_block_group_mutex);
9509 btrfs_end_transaction(trans);
9511 ret = btrfs_wait_for_commit(fs_info, transid);
9518 * if we are changing raid levels, try to allocate a corresponding
9519 * block group with the new raid level.
9521 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9522 if (alloc_flags != cache->flags) {
9523 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9526 * ENOSPC is allowed here, we may have enough space
9527 * already allocated at the new raid level to
9536 ret = inc_block_group_ro(cache, 0);
9539 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9540 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9544 ret = inc_block_group_ro(cache, 0);
9546 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9547 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9548 mutex_lock(&fs_info->chunk_mutex);
9549 check_system_chunk(trans, fs_info, alloc_flags);
9550 mutex_unlock(&fs_info->chunk_mutex);
9552 mutex_unlock(&fs_info->ro_block_group_mutex);
9554 btrfs_end_transaction(trans);
9558 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9559 struct btrfs_fs_info *fs_info, u64 type)
9561 u64 alloc_flags = get_alloc_profile(fs_info, type);
9563 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9567 * helper to account the unused space of all the readonly block group in the
9568 * space_info. takes mirrors into account.
9570 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9572 struct btrfs_block_group_cache *block_group;
9576 /* It's df, we don't care if it's racy */
9577 if (list_empty(&sinfo->ro_bgs))
9580 spin_lock(&sinfo->lock);
9581 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9582 spin_lock(&block_group->lock);
9584 if (!block_group->ro) {
9585 spin_unlock(&block_group->lock);
9589 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9590 BTRFS_BLOCK_GROUP_RAID10 |
9591 BTRFS_BLOCK_GROUP_DUP))
9596 free_bytes += (block_group->key.offset -
9597 btrfs_block_group_used(&block_group->item)) *
9600 spin_unlock(&block_group->lock);
9602 spin_unlock(&sinfo->lock);
9607 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9609 struct btrfs_space_info *sinfo = cache->space_info;
9614 spin_lock(&sinfo->lock);
9615 spin_lock(&cache->lock);
9617 num_bytes = cache->key.offset - cache->reserved -
9618 cache->pinned - cache->bytes_super -
9619 btrfs_block_group_used(&cache->item);
9620 sinfo->bytes_readonly -= num_bytes;
9621 list_del_init(&cache->ro_list);
9623 spin_unlock(&cache->lock);
9624 spin_unlock(&sinfo->lock);
9628 * checks to see if its even possible to relocate this block group.
9630 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9631 * ok to go ahead and try.
9633 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9635 struct btrfs_root *root = fs_info->extent_root;
9636 struct btrfs_block_group_cache *block_group;
9637 struct btrfs_space_info *space_info;
9638 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9639 struct btrfs_device *device;
9640 struct btrfs_trans_handle *trans;
9650 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9652 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9654 /* odd, couldn't find the block group, leave it alone */
9658 "can't find block group for bytenr %llu",
9663 min_free = btrfs_block_group_used(&block_group->item);
9665 /* no bytes used, we're good */
9669 space_info = block_group->space_info;
9670 spin_lock(&space_info->lock);
9672 full = space_info->full;
9675 * if this is the last block group we have in this space, we can't
9676 * relocate it unless we're able to allocate a new chunk below.
9678 * Otherwise, we need to make sure we have room in the space to handle
9679 * all of the extents from this block group. If we can, we're good
9681 if ((space_info->total_bytes != block_group->key.offset) &&
9682 (btrfs_space_info_used(space_info, false) + min_free <
9683 space_info->total_bytes)) {
9684 spin_unlock(&space_info->lock);
9687 spin_unlock(&space_info->lock);
9690 * ok we don't have enough space, but maybe we have free space on our
9691 * devices to allocate new chunks for relocation, so loop through our
9692 * alloc devices and guess if we have enough space. if this block
9693 * group is going to be restriped, run checks against the target
9694 * profile instead of the current one.
9706 target = get_restripe_target(fs_info, block_group->flags);
9708 index = __get_raid_index(extended_to_chunk(target));
9711 * this is just a balance, so if we were marked as full
9712 * we know there is no space for a new chunk
9717 "no space to alloc new chunk for block group %llu",
9718 block_group->key.objectid);
9722 index = get_block_group_index(block_group);
9725 if (index == BTRFS_RAID_RAID10) {
9729 } else if (index == BTRFS_RAID_RAID1) {
9731 } else if (index == BTRFS_RAID_DUP) {
9734 } else if (index == BTRFS_RAID_RAID0) {
9735 dev_min = fs_devices->rw_devices;
9736 min_free = div64_u64(min_free, dev_min);
9739 /* We need to do this so that we can look at pending chunks */
9740 trans = btrfs_join_transaction(root);
9741 if (IS_ERR(trans)) {
9742 ret = PTR_ERR(trans);
9746 mutex_lock(&fs_info->chunk_mutex);
9747 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9751 * check to make sure we can actually find a chunk with enough
9752 * space to fit our block group in.
9754 if (device->total_bytes > device->bytes_used + min_free &&
9755 !device->is_tgtdev_for_dev_replace) {
9756 ret = find_free_dev_extent(trans, device, min_free,
9761 if (dev_nr >= dev_min)
9767 if (debug && ret == -1)
9769 "no space to allocate a new chunk for block group %llu",
9770 block_group->key.objectid);
9771 mutex_unlock(&fs_info->chunk_mutex);
9772 btrfs_end_transaction(trans);
9774 btrfs_put_block_group(block_group);
9778 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9779 struct btrfs_path *path,
9780 struct btrfs_key *key)
9782 struct btrfs_root *root = fs_info->extent_root;
9784 struct btrfs_key found_key;
9785 struct extent_buffer *leaf;
9788 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9793 slot = path->slots[0];
9794 leaf = path->nodes[0];
9795 if (slot >= btrfs_header_nritems(leaf)) {
9796 ret = btrfs_next_leaf(root, path);
9803 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9805 if (found_key.objectid >= key->objectid &&
9806 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9807 struct extent_map_tree *em_tree;
9808 struct extent_map *em;
9810 em_tree = &root->fs_info->mapping_tree.map_tree;
9811 read_lock(&em_tree->lock);
9812 em = lookup_extent_mapping(em_tree, found_key.objectid,
9814 read_unlock(&em_tree->lock);
9817 "logical %llu len %llu found bg but no related chunk",
9818 found_key.objectid, found_key.offset);
9823 free_extent_map(em);
9832 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9834 struct btrfs_block_group_cache *block_group;
9838 struct inode *inode;
9840 block_group = btrfs_lookup_first_block_group(info, last);
9841 while (block_group) {
9842 spin_lock(&block_group->lock);
9843 if (block_group->iref)
9845 spin_unlock(&block_group->lock);
9846 block_group = next_block_group(info, block_group);
9855 inode = block_group->inode;
9856 block_group->iref = 0;
9857 block_group->inode = NULL;
9858 spin_unlock(&block_group->lock);
9859 ASSERT(block_group->io_ctl.inode == NULL);
9861 last = block_group->key.objectid + block_group->key.offset;
9862 btrfs_put_block_group(block_group);
9867 * Must be called only after stopping all workers, since we could have block
9868 * group caching kthreads running, and therefore they could race with us if we
9869 * freed the block groups before stopping them.
9871 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9873 struct btrfs_block_group_cache *block_group;
9874 struct btrfs_space_info *space_info;
9875 struct btrfs_caching_control *caching_ctl;
9878 down_write(&info->commit_root_sem);
9879 while (!list_empty(&info->caching_block_groups)) {
9880 caching_ctl = list_entry(info->caching_block_groups.next,
9881 struct btrfs_caching_control, list);
9882 list_del(&caching_ctl->list);
9883 put_caching_control(caching_ctl);
9885 up_write(&info->commit_root_sem);
9887 spin_lock(&info->unused_bgs_lock);
9888 while (!list_empty(&info->unused_bgs)) {
9889 block_group = list_first_entry(&info->unused_bgs,
9890 struct btrfs_block_group_cache,
9892 list_del_init(&block_group->bg_list);
9893 btrfs_put_block_group(block_group);
9895 spin_unlock(&info->unused_bgs_lock);
9897 spin_lock(&info->block_group_cache_lock);
9898 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9899 block_group = rb_entry(n, struct btrfs_block_group_cache,
9901 rb_erase(&block_group->cache_node,
9902 &info->block_group_cache_tree);
9903 RB_CLEAR_NODE(&block_group->cache_node);
9904 spin_unlock(&info->block_group_cache_lock);
9906 down_write(&block_group->space_info->groups_sem);
9907 list_del(&block_group->list);
9908 up_write(&block_group->space_info->groups_sem);
9911 * We haven't cached this block group, which means we could
9912 * possibly have excluded extents on this block group.
9914 if (block_group->cached == BTRFS_CACHE_NO ||
9915 block_group->cached == BTRFS_CACHE_ERROR)
9916 free_excluded_extents(info, block_group);
9918 btrfs_remove_free_space_cache(block_group);
9919 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9920 ASSERT(list_empty(&block_group->dirty_list));
9921 ASSERT(list_empty(&block_group->io_list));
9922 ASSERT(list_empty(&block_group->bg_list));
9923 ASSERT(atomic_read(&block_group->count) == 1);
9924 btrfs_put_block_group(block_group);
9926 spin_lock(&info->block_group_cache_lock);
9928 spin_unlock(&info->block_group_cache_lock);
9930 /* now that all the block groups are freed, go through and
9931 * free all the space_info structs. This is only called during
9932 * the final stages of unmount, and so we know nobody is
9933 * using them. We call synchronize_rcu() once before we start,
9934 * just to be on the safe side.
9938 release_global_block_rsv(info);
9940 while (!list_empty(&info->space_info)) {
9943 space_info = list_entry(info->space_info.next,
9944 struct btrfs_space_info,
9948 * Do not hide this behind enospc_debug, this is actually
9949 * important and indicates a real bug if this happens.
9951 if (WARN_ON(space_info->bytes_pinned > 0 ||
9952 space_info->bytes_reserved > 0 ||
9953 space_info->bytes_may_use > 0))
9954 dump_space_info(info, space_info, 0, 0);
9955 list_del(&space_info->list);
9956 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9957 struct kobject *kobj;
9958 kobj = space_info->block_group_kobjs[i];
9959 space_info->block_group_kobjs[i] = NULL;
9965 kobject_del(&space_info->kobj);
9966 kobject_put(&space_info->kobj);
9971 static void __link_block_group(struct btrfs_space_info *space_info,
9972 struct btrfs_block_group_cache *cache)
9974 int index = get_block_group_index(cache);
9977 down_write(&space_info->groups_sem);
9978 if (list_empty(&space_info->block_groups[index]))
9980 list_add_tail(&cache->list, &space_info->block_groups[index]);
9981 up_write(&space_info->groups_sem);
9984 struct raid_kobject *rkobj;
9987 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9990 rkobj->raid_type = index;
9991 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9992 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9993 "%s", get_raid_name(index));
9995 kobject_put(&rkobj->kobj);
9998 space_info->block_group_kobjs[index] = &rkobj->kobj;
10003 btrfs_warn(cache->fs_info,
10004 "failed to add kobject for block cache, ignoring");
10007 static struct btrfs_block_group_cache *
10008 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
10009 u64 start, u64 size)
10011 struct btrfs_block_group_cache *cache;
10013 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10017 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10019 if (!cache->free_space_ctl) {
10024 cache->key.objectid = start;
10025 cache->key.offset = size;
10026 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10028 cache->fs_info = fs_info;
10029 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10030 set_free_space_tree_thresholds(cache);
10032 atomic_set(&cache->count, 1);
10033 spin_lock_init(&cache->lock);
10034 init_rwsem(&cache->data_rwsem);
10035 INIT_LIST_HEAD(&cache->list);
10036 INIT_LIST_HEAD(&cache->cluster_list);
10037 INIT_LIST_HEAD(&cache->bg_list);
10038 INIT_LIST_HEAD(&cache->ro_list);
10039 INIT_LIST_HEAD(&cache->dirty_list);
10040 INIT_LIST_HEAD(&cache->io_list);
10041 btrfs_init_free_space_ctl(cache);
10042 atomic_set(&cache->trimming, 0);
10043 mutex_init(&cache->free_space_lock);
10044 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10049 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10051 struct btrfs_path *path;
10053 struct btrfs_block_group_cache *cache;
10054 struct btrfs_space_info *space_info;
10055 struct btrfs_key key;
10056 struct btrfs_key found_key;
10057 struct extent_buffer *leaf;
10058 int need_clear = 0;
10063 feature = btrfs_super_incompat_flags(info->super_copy);
10064 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10068 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10069 path = btrfs_alloc_path();
10072 path->reada = READA_FORWARD;
10074 cache_gen = btrfs_super_cache_generation(info->super_copy);
10075 if (btrfs_test_opt(info, SPACE_CACHE) &&
10076 btrfs_super_generation(info->super_copy) != cache_gen)
10078 if (btrfs_test_opt(info, CLEAR_CACHE))
10082 ret = find_first_block_group(info, path, &key);
10088 leaf = path->nodes[0];
10089 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10091 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10100 * When we mount with old space cache, we need to
10101 * set BTRFS_DC_CLEAR and set dirty flag.
10103 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10104 * truncate the old free space cache inode and
10106 * b) Setting 'dirty flag' makes sure that we flush
10107 * the new space cache info onto disk.
10109 if (btrfs_test_opt(info, SPACE_CACHE))
10110 cache->disk_cache_state = BTRFS_DC_CLEAR;
10113 read_extent_buffer(leaf, &cache->item,
10114 btrfs_item_ptr_offset(leaf, path->slots[0]),
10115 sizeof(cache->item));
10116 cache->flags = btrfs_block_group_flags(&cache->item);
10118 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10119 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10121 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10122 cache->key.objectid);
10127 key.objectid = found_key.objectid + found_key.offset;
10128 btrfs_release_path(path);
10131 * We need to exclude the super stripes now so that the space
10132 * info has super bytes accounted for, otherwise we'll think
10133 * we have more space than we actually do.
10135 ret = exclude_super_stripes(info, cache);
10138 * We may have excluded something, so call this just in
10141 free_excluded_extents(info, cache);
10142 btrfs_put_block_group(cache);
10147 * check for two cases, either we are full, and therefore
10148 * don't need to bother with the caching work since we won't
10149 * find any space, or we are empty, and we can just add all
10150 * the space in and be done with it. This saves us _alot_ of
10151 * time, particularly in the full case.
10153 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10154 cache->last_byte_to_unpin = (u64)-1;
10155 cache->cached = BTRFS_CACHE_FINISHED;
10156 free_excluded_extents(info, cache);
10157 } else if (btrfs_block_group_used(&cache->item) == 0) {
10158 cache->last_byte_to_unpin = (u64)-1;
10159 cache->cached = BTRFS_CACHE_FINISHED;
10160 add_new_free_space(cache, info,
10161 found_key.objectid,
10162 found_key.objectid +
10164 free_excluded_extents(info, cache);
10167 ret = btrfs_add_block_group_cache(info, cache);
10169 btrfs_remove_free_space_cache(cache);
10170 btrfs_put_block_group(cache);
10174 trace_btrfs_add_block_group(info, cache, 0);
10175 update_space_info(info, cache->flags, found_key.offset,
10176 btrfs_block_group_used(&cache->item),
10177 cache->bytes_super, &space_info);
10179 cache->space_info = space_info;
10181 __link_block_group(space_info, cache);
10183 set_avail_alloc_bits(info, cache->flags);
10184 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10185 inc_block_group_ro(cache, 1);
10186 } else if (btrfs_block_group_used(&cache->item) == 0) {
10187 spin_lock(&info->unused_bgs_lock);
10188 /* Should always be true but just in case. */
10189 if (list_empty(&cache->bg_list)) {
10190 btrfs_get_block_group(cache);
10191 list_add_tail(&cache->bg_list,
10192 &info->unused_bgs);
10194 spin_unlock(&info->unused_bgs_lock);
10198 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10199 if (!(get_alloc_profile(info, space_info->flags) &
10200 (BTRFS_BLOCK_GROUP_RAID10 |
10201 BTRFS_BLOCK_GROUP_RAID1 |
10202 BTRFS_BLOCK_GROUP_RAID5 |
10203 BTRFS_BLOCK_GROUP_RAID6 |
10204 BTRFS_BLOCK_GROUP_DUP)))
10207 * avoid allocating from un-mirrored block group if there are
10208 * mirrored block groups.
10210 list_for_each_entry(cache,
10211 &space_info->block_groups[BTRFS_RAID_RAID0],
10213 inc_block_group_ro(cache, 1);
10214 list_for_each_entry(cache,
10215 &space_info->block_groups[BTRFS_RAID_SINGLE],
10217 inc_block_group_ro(cache, 1);
10220 init_global_block_rsv(info);
10223 btrfs_free_path(path);
10227 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10228 struct btrfs_fs_info *fs_info)
10230 struct btrfs_block_group_cache *block_group, *tmp;
10231 struct btrfs_root *extent_root = fs_info->extent_root;
10232 struct btrfs_block_group_item item;
10233 struct btrfs_key key;
10235 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10237 trans->can_flush_pending_bgs = false;
10238 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10242 spin_lock(&block_group->lock);
10243 memcpy(&item, &block_group->item, sizeof(item));
10244 memcpy(&key, &block_group->key, sizeof(key));
10245 spin_unlock(&block_group->lock);
10247 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10250 btrfs_abort_transaction(trans, ret);
10251 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10254 btrfs_abort_transaction(trans, ret);
10255 add_block_group_free_space(trans, fs_info, block_group);
10256 /* already aborted the transaction if it failed. */
10258 list_del_init(&block_group->bg_list);
10260 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10263 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10264 struct btrfs_fs_info *fs_info, u64 bytes_used,
10265 u64 type, u64 chunk_offset, u64 size)
10267 struct btrfs_block_group_cache *cache;
10270 btrfs_set_log_full_commit(fs_info, trans);
10272 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10276 btrfs_set_block_group_used(&cache->item, bytes_used);
10277 btrfs_set_block_group_chunk_objectid(&cache->item,
10278 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10279 btrfs_set_block_group_flags(&cache->item, type);
10281 cache->flags = type;
10282 cache->last_byte_to_unpin = (u64)-1;
10283 cache->cached = BTRFS_CACHE_FINISHED;
10284 cache->needs_free_space = 1;
10285 ret = exclude_super_stripes(fs_info, cache);
10288 * We may have excluded something, so call this just in
10291 free_excluded_extents(fs_info, cache);
10292 btrfs_put_block_group(cache);
10296 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10298 free_excluded_extents(fs_info, cache);
10300 #ifdef CONFIG_BTRFS_DEBUG
10301 if (btrfs_should_fragment_free_space(cache)) {
10302 u64 new_bytes_used = size - bytes_used;
10304 bytes_used += new_bytes_used >> 1;
10305 fragment_free_space(cache);
10309 * Ensure the corresponding space_info object is created and
10310 * assigned to our block group. We want our bg to be added to the rbtree
10311 * with its ->space_info set.
10313 cache->space_info = __find_space_info(fs_info, cache->flags);
10314 if (!cache->space_info) {
10315 ret = create_space_info(fs_info, cache->flags,
10316 &cache->space_info);
10318 btrfs_remove_free_space_cache(cache);
10319 btrfs_put_block_group(cache);
10324 ret = btrfs_add_block_group_cache(fs_info, cache);
10326 btrfs_remove_free_space_cache(cache);
10327 btrfs_put_block_group(cache);
10332 * Now that our block group has its ->space_info set and is inserted in
10333 * the rbtree, update the space info's counters.
10335 trace_btrfs_add_block_group(fs_info, cache, 1);
10336 update_space_info(fs_info, cache->flags, size, bytes_used,
10337 cache->bytes_super, &cache->space_info);
10338 update_global_block_rsv(fs_info);
10340 __link_block_group(cache->space_info, cache);
10342 list_add_tail(&cache->bg_list, &trans->new_bgs);
10344 set_avail_alloc_bits(fs_info, type);
10348 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10350 u64 extra_flags = chunk_to_extended(flags) &
10351 BTRFS_EXTENDED_PROFILE_MASK;
10353 write_seqlock(&fs_info->profiles_lock);
10354 if (flags & BTRFS_BLOCK_GROUP_DATA)
10355 fs_info->avail_data_alloc_bits &= ~extra_flags;
10356 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10357 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10358 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10359 fs_info->avail_system_alloc_bits &= ~extra_flags;
10360 write_sequnlock(&fs_info->profiles_lock);
10363 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10364 struct btrfs_fs_info *fs_info, u64 group_start,
10365 struct extent_map *em)
10367 struct btrfs_root *root = fs_info->extent_root;
10368 struct btrfs_path *path;
10369 struct btrfs_block_group_cache *block_group;
10370 struct btrfs_free_cluster *cluster;
10371 struct btrfs_root *tree_root = fs_info->tree_root;
10372 struct btrfs_key key;
10373 struct inode *inode;
10374 struct kobject *kobj = NULL;
10378 struct btrfs_caching_control *caching_ctl = NULL;
10381 block_group = btrfs_lookup_block_group(fs_info, group_start);
10382 BUG_ON(!block_group);
10383 BUG_ON(!block_group->ro);
10386 * Free the reserved super bytes from this block group before
10389 free_excluded_extents(fs_info, block_group);
10391 memcpy(&key, &block_group->key, sizeof(key));
10392 index = get_block_group_index(block_group);
10393 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10394 BTRFS_BLOCK_GROUP_RAID1 |
10395 BTRFS_BLOCK_GROUP_RAID10))
10400 /* make sure this block group isn't part of an allocation cluster */
10401 cluster = &fs_info->data_alloc_cluster;
10402 spin_lock(&cluster->refill_lock);
10403 btrfs_return_cluster_to_free_space(block_group, cluster);
10404 spin_unlock(&cluster->refill_lock);
10407 * make sure this block group isn't part of a metadata
10408 * allocation cluster
10410 cluster = &fs_info->meta_alloc_cluster;
10411 spin_lock(&cluster->refill_lock);
10412 btrfs_return_cluster_to_free_space(block_group, cluster);
10413 spin_unlock(&cluster->refill_lock);
10415 path = btrfs_alloc_path();
10422 * get the inode first so any iput calls done for the io_list
10423 * aren't the final iput (no unlinks allowed now)
10425 inode = lookup_free_space_inode(fs_info, block_group, path);
10427 mutex_lock(&trans->transaction->cache_write_mutex);
10429 * make sure our free spache cache IO is done before remove the
10432 spin_lock(&trans->transaction->dirty_bgs_lock);
10433 if (!list_empty(&block_group->io_list)) {
10434 list_del_init(&block_group->io_list);
10436 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10438 spin_unlock(&trans->transaction->dirty_bgs_lock);
10439 btrfs_wait_cache_io(trans, block_group, path);
10440 btrfs_put_block_group(block_group);
10441 spin_lock(&trans->transaction->dirty_bgs_lock);
10444 if (!list_empty(&block_group->dirty_list)) {
10445 list_del_init(&block_group->dirty_list);
10446 btrfs_put_block_group(block_group);
10448 spin_unlock(&trans->transaction->dirty_bgs_lock);
10449 mutex_unlock(&trans->transaction->cache_write_mutex);
10451 if (!IS_ERR(inode)) {
10452 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10454 btrfs_add_delayed_iput(inode);
10457 clear_nlink(inode);
10458 /* One for the block groups ref */
10459 spin_lock(&block_group->lock);
10460 if (block_group->iref) {
10461 block_group->iref = 0;
10462 block_group->inode = NULL;
10463 spin_unlock(&block_group->lock);
10466 spin_unlock(&block_group->lock);
10468 /* One for our lookup ref */
10469 btrfs_add_delayed_iput(inode);
10472 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10473 key.offset = block_group->key.objectid;
10476 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10480 btrfs_release_path(path);
10482 ret = btrfs_del_item(trans, tree_root, path);
10485 btrfs_release_path(path);
10488 spin_lock(&fs_info->block_group_cache_lock);
10489 rb_erase(&block_group->cache_node,
10490 &fs_info->block_group_cache_tree);
10491 RB_CLEAR_NODE(&block_group->cache_node);
10493 if (fs_info->first_logical_byte == block_group->key.objectid)
10494 fs_info->first_logical_byte = (u64)-1;
10495 spin_unlock(&fs_info->block_group_cache_lock);
10497 down_write(&block_group->space_info->groups_sem);
10499 * we must use list_del_init so people can check to see if they
10500 * are still on the list after taking the semaphore
10502 list_del_init(&block_group->list);
10503 if (list_empty(&block_group->space_info->block_groups[index])) {
10504 kobj = block_group->space_info->block_group_kobjs[index];
10505 block_group->space_info->block_group_kobjs[index] = NULL;
10506 clear_avail_alloc_bits(fs_info, block_group->flags);
10508 up_write(&block_group->space_info->groups_sem);
10514 if (block_group->has_caching_ctl)
10515 caching_ctl = get_caching_control(block_group);
10516 if (block_group->cached == BTRFS_CACHE_STARTED)
10517 wait_block_group_cache_done(block_group);
10518 if (block_group->has_caching_ctl) {
10519 down_write(&fs_info->commit_root_sem);
10520 if (!caching_ctl) {
10521 struct btrfs_caching_control *ctl;
10523 list_for_each_entry(ctl,
10524 &fs_info->caching_block_groups, list)
10525 if (ctl->block_group == block_group) {
10527 refcount_inc(&caching_ctl->count);
10532 list_del_init(&caching_ctl->list);
10533 up_write(&fs_info->commit_root_sem);
10535 /* Once for the caching bgs list and once for us. */
10536 put_caching_control(caching_ctl);
10537 put_caching_control(caching_ctl);
10541 spin_lock(&trans->transaction->dirty_bgs_lock);
10542 if (!list_empty(&block_group->dirty_list)) {
10545 if (!list_empty(&block_group->io_list)) {
10548 spin_unlock(&trans->transaction->dirty_bgs_lock);
10549 btrfs_remove_free_space_cache(block_group);
10551 spin_lock(&block_group->space_info->lock);
10552 list_del_init(&block_group->ro_list);
10554 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10555 WARN_ON(block_group->space_info->total_bytes
10556 < block_group->key.offset);
10557 WARN_ON(block_group->space_info->bytes_readonly
10558 < block_group->key.offset);
10559 WARN_ON(block_group->space_info->disk_total
10560 < block_group->key.offset * factor);
10562 block_group->space_info->total_bytes -= block_group->key.offset;
10563 block_group->space_info->bytes_readonly -= block_group->key.offset;
10564 block_group->space_info->disk_total -= block_group->key.offset * factor;
10566 spin_unlock(&block_group->space_info->lock);
10568 memcpy(&key, &block_group->key, sizeof(key));
10570 mutex_lock(&fs_info->chunk_mutex);
10571 if (!list_empty(&em->list)) {
10572 /* We're in the transaction->pending_chunks list. */
10573 free_extent_map(em);
10575 spin_lock(&block_group->lock);
10576 block_group->removed = 1;
10578 * At this point trimming can't start on this block group, because we
10579 * removed the block group from the tree fs_info->block_group_cache_tree
10580 * so no one can't find it anymore and even if someone already got this
10581 * block group before we removed it from the rbtree, they have already
10582 * incremented block_group->trimming - if they didn't, they won't find
10583 * any free space entries because we already removed them all when we
10584 * called btrfs_remove_free_space_cache().
10586 * And we must not remove the extent map from the fs_info->mapping_tree
10587 * to prevent the same logical address range and physical device space
10588 * ranges from being reused for a new block group. This is because our
10589 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10590 * completely transactionless, so while it is trimming a range the
10591 * currently running transaction might finish and a new one start,
10592 * allowing for new block groups to be created that can reuse the same
10593 * physical device locations unless we take this special care.
10595 * There may also be an implicit trim operation if the file system
10596 * is mounted with -odiscard. The same protections must remain
10597 * in place until the extents have been discarded completely when
10598 * the transaction commit has completed.
10600 remove_em = (atomic_read(&block_group->trimming) == 0);
10602 * Make sure a trimmer task always sees the em in the pinned_chunks list
10603 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10604 * before checking block_group->removed).
10608 * Our em might be in trans->transaction->pending_chunks which
10609 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10610 * and so is the fs_info->pinned_chunks list.
10612 * So at this point we must be holding the chunk_mutex to avoid
10613 * any races with chunk allocation (more specifically at
10614 * volumes.c:contains_pending_extent()), to ensure it always
10615 * sees the em, either in the pending_chunks list or in the
10616 * pinned_chunks list.
10618 list_move_tail(&em->list, &fs_info->pinned_chunks);
10620 spin_unlock(&block_group->lock);
10623 struct extent_map_tree *em_tree;
10625 em_tree = &fs_info->mapping_tree.map_tree;
10626 write_lock(&em_tree->lock);
10628 * The em might be in the pending_chunks list, so make sure the
10629 * chunk mutex is locked, since remove_extent_mapping() will
10630 * delete us from that list.
10632 remove_extent_mapping(em_tree, em);
10633 write_unlock(&em_tree->lock);
10634 /* once for the tree */
10635 free_extent_map(em);
10638 mutex_unlock(&fs_info->chunk_mutex);
10640 ret = remove_block_group_free_space(trans, fs_info, block_group);
10644 btrfs_put_block_group(block_group);
10645 btrfs_put_block_group(block_group);
10647 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10653 ret = btrfs_del_item(trans, root, path);
10655 btrfs_free_path(path);
10659 struct btrfs_trans_handle *
10660 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10661 const u64 chunk_offset)
10663 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10664 struct extent_map *em;
10665 struct map_lookup *map;
10666 unsigned int num_items;
10668 read_lock(&em_tree->lock);
10669 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10670 read_unlock(&em_tree->lock);
10671 ASSERT(em && em->start == chunk_offset);
10674 * We need to reserve 3 + N units from the metadata space info in order
10675 * to remove a block group (done at btrfs_remove_chunk() and at
10676 * btrfs_remove_block_group()), which are used for:
10678 * 1 unit for adding the free space inode's orphan (located in the tree
10680 * 1 unit for deleting the block group item (located in the extent
10682 * 1 unit for deleting the free space item (located in tree of tree
10684 * N units for deleting N device extent items corresponding to each
10685 * stripe (located in the device tree).
10687 * In order to remove a block group we also need to reserve units in the
10688 * system space info in order to update the chunk tree (update one or
10689 * more device items and remove one chunk item), but this is done at
10690 * btrfs_remove_chunk() through a call to check_system_chunk().
10692 map = em->map_lookup;
10693 num_items = 3 + map->num_stripes;
10694 free_extent_map(em);
10696 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10701 * Process the unused_bgs list and remove any that don't have any allocated
10702 * space inside of them.
10704 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10706 struct btrfs_block_group_cache *block_group;
10707 struct btrfs_space_info *space_info;
10708 struct btrfs_trans_handle *trans;
10711 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10714 spin_lock(&fs_info->unused_bgs_lock);
10715 while (!list_empty(&fs_info->unused_bgs)) {
10719 block_group = list_first_entry(&fs_info->unused_bgs,
10720 struct btrfs_block_group_cache,
10722 list_del_init(&block_group->bg_list);
10724 space_info = block_group->space_info;
10726 if (ret || btrfs_mixed_space_info(space_info)) {
10727 btrfs_put_block_group(block_group);
10730 spin_unlock(&fs_info->unused_bgs_lock);
10732 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10734 /* Don't want to race with allocators so take the groups_sem */
10735 down_write(&space_info->groups_sem);
10736 spin_lock(&block_group->lock);
10737 if (block_group->reserved ||
10738 btrfs_block_group_used(&block_group->item) ||
10740 list_is_singular(&block_group->list)) {
10742 * We want to bail if we made new allocations or have
10743 * outstanding allocations in this block group. We do
10744 * the ro check in case balance is currently acting on
10745 * this block group.
10747 spin_unlock(&block_group->lock);
10748 up_write(&space_info->groups_sem);
10751 spin_unlock(&block_group->lock);
10753 /* We don't want to force the issue, only flip if it's ok. */
10754 ret = inc_block_group_ro(block_group, 0);
10755 up_write(&space_info->groups_sem);
10762 * Want to do this before we do anything else so we can recover
10763 * properly if we fail to join the transaction.
10765 trans = btrfs_start_trans_remove_block_group(fs_info,
10766 block_group->key.objectid);
10767 if (IS_ERR(trans)) {
10768 btrfs_dec_block_group_ro(block_group);
10769 ret = PTR_ERR(trans);
10774 * We could have pending pinned extents for this block group,
10775 * just delete them, we don't care about them anymore.
10777 start = block_group->key.objectid;
10778 end = start + block_group->key.offset - 1;
10780 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10781 * btrfs_finish_extent_commit(). If we are at transaction N,
10782 * another task might be running finish_extent_commit() for the
10783 * previous transaction N - 1, and have seen a range belonging
10784 * to the block group in freed_extents[] before we were able to
10785 * clear the whole block group range from freed_extents[]. This
10786 * means that task can lookup for the block group after we
10787 * unpinned it from freed_extents[] and removed it, leading to
10788 * a BUG_ON() at btrfs_unpin_extent_range().
10790 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10791 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10794 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10795 btrfs_dec_block_group_ro(block_group);
10798 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10801 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10802 btrfs_dec_block_group_ro(block_group);
10805 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10807 /* Reset pinned so btrfs_put_block_group doesn't complain */
10808 spin_lock(&space_info->lock);
10809 spin_lock(&block_group->lock);
10811 space_info->bytes_pinned -= block_group->pinned;
10812 space_info->bytes_readonly += block_group->pinned;
10813 percpu_counter_add(&space_info->total_bytes_pinned,
10814 -block_group->pinned);
10815 block_group->pinned = 0;
10817 spin_unlock(&block_group->lock);
10818 spin_unlock(&space_info->lock);
10820 /* DISCARD can flip during remount */
10821 trimming = btrfs_test_opt(fs_info, DISCARD);
10823 /* Implicit trim during transaction commit. */
10825 btrfs_get_block_group_trimming(block_group);
10828 * Btrfs_remove_chunk will abort the transaction if things go
10831 ret = btrfs_remove_chunk(trans, fs_info,
10832 block_group->key.objectid);
10836 btrfs_put_block_group_trimming(block_group);
10841 * If we're not mounted with -odiscard, we can just forget
10842 * about this block group. Otherwise we'll need to wait
10843 * until transaction commit to do the actual discard.
10846 spin_lock(&fs_info->unused_bgs_lock);
10848 * A concurrent scrub might have added us to the list
10849 * fs_info->unused_bgs, so use a list_move operation
10850 * to add the block group to the deleted_bgs list.
10852 list_move(&block_group->bg_list,
10853 &trans->transaction->deleted_bgs);
10854 spin_unlock(&fs_info->unused_bgs_lock);
10855 btrfs_get_block_group(block_group);
10858 btrfs_end_transaction(trans);
10860 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10861 btrfs_put_block_group(block_group);
10862 spin_lock(&fs_info->unused_bgs_lock);
10864 spin_unlock(&fs_info->unused_bgs_lock);
10867 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10869 struct btrfs_space_info *space_info;
10870 struct btrfs_super_block *disk_super;
10876 disk_super = fs_info->super_copy;
10877 if (!btrfs_super_root(disk_super))
10880 features = btrfs_super_incompat_flags(disk_super);
10881 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10884 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10885 ret = create_space_info(fs_info, flags, &space_info);
10890 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10891 ret = create_space_info(fs_info, flags, &space_info);
10893 flags = BTRFS_BLOCK_GROUP_METADATA;
10894 ret = create_space_info(fs_info, flags, &space_info);
10898 flags = BTRFS_BLOCK_GROUP_DATA;
10899 ret = create_space_info(fs_info, flags, &space_info);
10905 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10906 u64 start, u64 end)
10908 return unpin_extent_range(fs_info, start, end, false);
10912 * It used to be that old block groups would be left around forever.
10913 * Iterating over them would be enough to trim unused space. Since we
10914 * now automatically remove them, we also need to iterate over unallocated
10917 * We don't want a transaction for this since the discard may take a
10918 * substantial amount of time. We don't require that a transaction be
10919 * running, but we do need to take a running transaction into account
10920 * to ensure that we're not discarding chunks that were released in
10921 * the current transaction.
10923 * Holding the chunks lock will prevent other threads from allocating
10924 * or releasing chunks, but it won't prevent a running transaction
10925 * from committing and releasing the memory that the pending chunks
10926 * list head uses. For that, we need to take a reference to the
10929 static int btrfs_trim_free_extents(struct btrfs_device *device,
10930 u64 minlen, u64 *trimmed)
10932 u64 start = 0, len = 0;
10937 /* Not writeable = nothing to do. */
10938 if (!device->writeable)
10941 /* No free space = nothing to do. */
10942 if (device->total_bytes <= device->bytes_used)
10948 struct btrfs_fs_info *fs_info = device->fs_info;
10949 struct btrfs_transaction *trans;
10952 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10956 down_read(&fs_info->commit_root_sem);
10958 spin_lock(&fs_info->trans_lock);
10959 trans = fs_info->running_transaction;
10961 refcount_inc(&trans->use_count);
10962 spin_unlock(&fs_info->trans_lock);
10964 ret = find_free_dev_extent_start(trans, device, minlen, start,
10967 btrfs_put_transaction(trans);
10970 up_read(&fs_info->commit_root_sem);
10971 mutex_unlock(&fs_info->chunk_mutex);
10972 if (ret == -ENOSPC)
10977 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10978 up_read(&fs_info->commit_root_sem);
10979 mutex_unlock(&fs_info->chunk_mutex);
10987 if (fatal_signal_pending(current)) {
10988 ret = -ERESTARTSYS;
10998 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
11000 struct btrfs_block_group_cache *cache = NULL;
11001 struct btrfs_device *device;
11002 struct list_head *devices;
11007 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11011 * try to trim all FS space, our block group may start from non-zero.
11013 if (range->len == total_bytes)
11014 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11016 cache = btrfs_lookup_block_group(fs_info, range->start);
11019 if (cache->key.objectid >= (range->start + range->len)) {
11020 btrfs_put_block_group(cache);
11024 start = max(range->start, cache->key.objectid);
11025 end = min(range->start + range->len,
11026 cache->key.objectid + cache->key.offset);
11028 if (end - start >= range->minlen) {
11029 if (!block_group_cache_done(cache)) {
11030 ret = cache_block_group(cache, 0);
11032 btrfs_put_block_group(cache);
11035 ret = wait_block_group_cache_done(cache);
11037 btrfs_put_block_group(cache);
11041 ret = btrfs_trim_block_group(cache,
11047 trimmed += group_trimmed;
11049 btrfs_put_block_group(cache);
11054 cache = next_block_group(fs_info, cache);
11057 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11058 devices = &fs_info->fs_devices->alloc_list;
11059 list_for_each_entry(device, devices, dev_alloc_list) {
11060 ret = btrfs_trim_free_extents(device, range->minlen,
11065 trimmed += group_trimmed;
11067 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11069 range->len = trimmed;
11074 * btrfs_{start,end}_write_no_snapshotting() are similar to
11075 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11076 * data into the page cache through nocow before the subvolume is snapshoted,
11077 * but flush the data into disk after the snapshot creation, or to prevent
11078 * operations while snapshotting is ongoing and that cause the snapshot to be
11079 * inconsistent (writes followed by expanding truncates for example).
11081 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11083 percpu_counter_dec(&root->subv_writers->counter);
11085 * Make sure counter is updated before we wake up waiters.
11088 if (waitqueue_active(&root->subv_writers->wait))
11089 wake_up(&root->subv_writers->wait);
11092 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11094 if (atomic_read(&root->will_be_snapshotted))
11097 percpu_counter_inc(&root->subv_writers->counter);
11099 * Make sure counter is updated before we check for snapshot creation.
11102 if (atomic_read(&root->will_be_snapshotted)) {
11103 btrfs_end_write_no_snapshotting(root);
11109 static int wait_snapshotting_atomic_t(atomic_t *a)
11115 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11120 ret = btrfs_start_write_no_snapshotting(root);
11123 wait_on_atomic_t(&root->will_be_snapshotted,
11124 wait_snapshotting_atomic_t,
11125 TASK_UNINTERRUPTIBLE);