2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/sched/signal.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/sort.h>
24 #include <linux/rcupdate.h>
25 #include <linux/kthread.h>
26 #include <linux/slab.h>
27 #include <linux/ratelimit.h>
28 #include <linux/percpu_counter.h>
32 #include "print-tree.h"
36 #include "free-space-cache.h"
37 #include "free-space-tree.h"
42 #undef SCRAMBLE_DELAYED_REFS
45 * control flags for do_chunk_alloc's force field
46 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
47 * if we really need one.
49 * CHUNK_ALLOC_LIMITED means to only try and allocate one
50 * if we have very few chunks already allocated. This is
51 * used as part of the clustering code to help make sure
52 * we have a good pool of storage to cluster in, without
53 * filling the FS with empty chunks
55 * CHUNK_ALLOC_FORCE means it must try to allocate one
59 CHUNK_ALLOC_NO_FORCE = 0,
60 CHUNK_ALLOC_LIMITED = 1,
61 CHUNK_ALLOC_FORCE = 2,
64 static int update_block_group(struct btrfs_trans_handle *trans,
65 struct btrfs_fs_info *fs_info, u64 bytenr,
66 u64 num_bytes, int alloc);
67 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
68 struct btrfs_fs_info *fs_info,
69 struct btrfs_delayed_ref_node *node, u64 parent,
70 u64 root_objectid, u64 owner_objectid,
71 u64 owner_offset, int refs_to_drop,
72 struct btrfs_delayed_extent_op *extra_op);
73 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
74 struct extent_buffer *leaf,
75 struct btrfs_extent_item *ei);
76 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
77 struct btrfs_fs_info *fs_info,
78 u64 parent, u64 root_objectid,
79 u64 flags, u64 owner, u64 offset,
80 struct btrfs_key *ins, int ref_mod);
81 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
82 struct btrfs_fs_info *fs_info,
83 u64 parent, u64 root_objectid,
84 u64 flags, struct btrfs_disk_key *key,
85 int level, struct btrfs_key *ins);
86 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
87 struct btrfs_fs_info *fs_info, u64 flags,
89 static int find_next_key(struct btrfs_path *path, int level,
90 struct btrfs_key *key);
91 static void dump_space_info(struct btrfs_fs_info *fs_info,
92 struct btrfs_space_info *info, u64 bytes,
93 int dump_block_groups);
94 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
95 u64 ram_bytes, u64 num_bytes, int delalloc);
96 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
97 u64 num_bytes, int delalloc);
98 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
100 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
101 struct btrfs_space_info *space_info,
103 enum btrfs_reserve_flush_enum flush,
105 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
106 struct btrfs_space_info *space_info,
108 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
109 struct btrfs_space_info *space_info,
113 block_group_cache_done(struct btrfs_block_group_cache *cache)
116 return cache->cached == BTRFS_CACHE_FINISHED ||
117 cache->cached == BTRFS_CACHE_ERROR;
120 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
122 return (cache->flags & bits) == bits;
125 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
127 atomic_inc(&cache->count);
130 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
132 if (atomic_dec_and_test(&cache->count)) {
133 WARN_ON(cache->pinned > 0);
134 WARN_ON(cache->reserved > 0);
137 * If not empty, someone is still holding mutex of
138 * full_stripe_lock, which can only be released by caller.
139 * And it will definitely cause use-after-free when caller
140 * tries to release full stripe lock.
142 * No better way to resolve, but only to warn.
144 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
145 kfree(cache->free_space_ctl);
151 * this adds the block group to the fs_info rb tree for the block group
154 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
155 struct btrfs_block_group_cache *block_group)
158 struct rb_node *parent = NULL;
159 struct btrfs_block_group_cache *cache;
161 spin_lock(&info->block_group_cache_lock);
162 p = &info->block_group_cache_tree.rb_node;
166 cache = rb_entry(parent, struct btrfs_block_group_cache,
168 if (block_group->key.objectid < cache->key.objectid) {
170 } else if (block_group->key.objectid > cache->key.objectid) {
173 spin_unlock(&info->block_group_cache_lock);
178 rb_link_node(&block_group->cache_node, parent, p);
179 rb_insert_color(&block_group->cache_node,
180 &info->block_group_cache_tree);
182 if (info->first_logical_byte > block_group->key.objectid)
183 info->first_logical_byte = block_group->key.objectid;
185 spin_unlock(&info->block_group_cache_lock);
191 * This will return the block group at or after bytenr if contains is 0, else
192 * it will return the block group that contains the bytenr
194 static struct btrfs_block_group_cache *
195 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
198 struct btrfs_block_group_cache *cache, *ret = NULL;
202 spin_lock(&info->block_group_cache_lock);
203 n = info->block_group_cache_tree.rb_node;
206 cache = rb_entry(n, struct btrfs_block_group_cache,
208 end = cache->key.objectid + cache->key.offset - 1;
209 start = cache->key.objectid;
211 if (bytenr < start) {
212 if (!contains && (!ret || start < ret->key.objectid))
215 } else if (bytenr > start) {
216 if (contains && bytenr <= end) {
227 btrfs_get_block_group(ret);
228 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
229 info->first_logical_byte = ret->key.objectid;
231 spin_unlock(&info->block_group_cache_lock);
236 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
237 u64 start, u64 num_bytes)
239 u64 end = start + num_bytes - 1;
240 set_extent_bits(&fs_info->freed_extents[0],
241 start, end, EXTENT_UPTODATE);
242 set_extent_bits(&fs_info->freed_extents[1],
243 start, end, EXTENT_UPTODATE);
247 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
248 struct btrfs_block_group_cache *cache)
252 start = cache->key.objectid;
253 end = start + cache->key.offset - 1;
255 clear_extent_bits(&fs_info->freed_extents[0],
256 start, end, EXTENT_UPTODATE);
257 clear_extent_bits(&fs_info->freed_extents[1],
258 start, end, EXTENT_UPTODATE);
261 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
262 struct btrfs_block_group_cache *cache)
269 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
270 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
271 cache->bytes_super += stripe_len;
272 ret = add_excluded_extent(fs_info, cache->key.objectid,
278 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
279 bytenr = btrfs_sb_offset(i);
280 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
281 bytenr, 0, &logical, &nr, &stripe_len);
288 if (logical[nr] > cache->key.objectid +
292 if (logical[nr] + stripe_len <= cache->key.objectid)
296 if (start < cache->key.objectid) {
297 start = cache->key.objectid;
298 len = (logical[nr] + stripe_len) - start;
300 len = min_t(u64, stripe_len,
301 cache->key.objectid +
302 cache->key.offset - start);
305 cache->bytes_super += len;
306 ret = add_excluded_extent(fs_info, start, len);
318 static struct btrfs_caching_control *
319 get_caching_control(struct btrfs_block_group_cache *cache)
321 struct btrfs_caching_control *ctl;
323 spin_lock(&cache->lock);
324 if (!cache->caching_ctl) {
325 spin_unlock(&cache->lock);
329 ctl = cache->caching_ctl;
330 refcount_inc(&ctl->count);
331 spin_unlock(&cache->lock);
335 static void put_caching_control(struct btrfs_caching_control *ctl)
337 if (refcount_dec_and_test(&ctl->count))
341 #ifdef CONFIG_BTRFS_DEBUG
342 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
344 struct btrfs_fs_info *fs_info = block_group->fs_info;
345 u64 start = block_group->key.objectid;
346 u64 len = block_group->key.offset;
347 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
348 fs_info->nodesize : fs_info->sectorsize;
349 u64 step = chunk << 1;
351 while (len > chunk) {
352 btrfs_remove_free_space(block_group, start, chunk);
363 * this is only called by cache_block_group, since we could have freed extents
364 * we need to check the pinned_extents for any extents that can't be used yet
365 * since their free space will be released as soon as the transaction commits.
367 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
368 struct btrfs_fs_info *info, u64 start, u64 end)
370 u64 extent_start, extent_end, size, total_added = 0;
373 while (start < end) {
374 ret = find_first_extent_bit(info->pinned_extents, start,
375 &extent_start, &extent_end,
376 EXTENT_DIRTY | EXTENT_UPTODATE,
381 if (extent_start <= start) {
382 start = extent_end + 1;
383 } else if (extent_start > start && extent_start < end) {
384 size = extent_start - start;
386 ret = btrfs_add_free_space(block_group, start,
388 BUG_ON(ret); /* -ENOMEM or logic error */
389 start = extent_end + 1;
398 ret = btrfs_add_free_space(block_group, start, size);
399 BUG_ON(ret); /* -ENOMEM or logic error */
405 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
407 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
408 struct btrfs_fs_info *fs_info = block_group->fs_info;
409 struct btrfs_root *extent_root = fs_info->extent_root;
410 struct btrfs_path *path;
411 struct extent_buffer *leaf;
412 struct btrfs_key key;
419 path = btrfs_alloc_path();
423 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
425 #ifdef CONFIG_BTRFS_DEBUG
427 * If we're fragmenting we don't want to make anybody think we can
428 * allocate from this block group until we've had a chance to fragment
431 if (btrfs_should_fragment_free_space(block_group))
435 * We don't want to deadlock with somebody trying to allocate a new
436 * extent for the extent root while also trying to search the extent
437 * root to add free space. So we skip locking and search the commit
438 * root, since its read-only
440 path->skip_locking = 1;
441 path->search_commit_root = 1;
442 path->reada = READA_FORWARD;
446 key.type = BTRFS_EXTENT_ITEM_KEY;
449 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
453 leaf = path->nodes[0];
454 nritems = btrfs_header_nritems(leaf);
457 if (btrfs_fs_closing(fs_info) > 1) {
462 if (path->slots[0] < nritems) {
463 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
465 ret = find_next_key(path, 0, &key);
469 if (need_resched() ||
470 rwsem_is_contended(&fs_info->commit_root_sem)) {
472 caching_ctl->progress = last;
473 btrfs_release_path(path);
474 up_read(&fs_info->commit_root_sem);
475 mutex_unlock(&caching_ctl->mutex);
477 mutex_lock(&caching_ctl->mutex);
478 down_read(&fs_info->commit_root_sem);
482 ret = btrfs_next_leaf(extent_root, path);
487 leaf = path->nodes[0];
488 nritems = btrfs_header_nritems(leaf);
492 if (key.objectid < last) {
495 key.type = BTRFS_EXTENT_ITEM_KEY;
498 caching_ctl->progress = last;
499 btrfs_release_path(path);
503 if (key.objectid < block_group->key.objectid) {
508 if (key.objectid >= block_group->key.objectid +
509 block_group->key.offset)
512 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
513 key.type == BTRFS_METADATA_ITEM_KEY) {
514 total_found += add_new_free_space(block_group,
517 if (key.type == BTRFS_METADATA_ITEM_KEY)
518 last = key.objectid +
521 last = key.objectid + key.offset;
523 if (total_found > CACHING_CTL_WAKE_UP) {
526 wake_up(&caching_ctl->wait);
533 total_found += add_new_free_space(block_group, fs_info, last,
534 block_group->key.objectid +
535 block_group->key.offset);
536 caching_ctl->progress = (u64)-1;
539 btrfs_free_path(path);
543 static noinline void caching_thread(struct btrfs_work *work)
545 struct btrfs_block_group_cache *block_group;
546 struct btrfs_fs_info *fs_info;
547 struct btrfs_caching_control *caching_ctl;
548 struct btrfs_root *extent_root;
551 caching_ctl = container_of(work, struct btrfs_caching_control, work);
552 block_group = caching_ctl->block_group;
553 fs_info = block_group->fs_info;
554 extent_root = fs_info->extent_root;
556 mutex_lock(&caching_ctl->mutex);
557 down_read(&fs_info->commit_root_sem);
559 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
560 ret = load_free_space_tree(caching_ctl);
562 ret = load_extent_tree_free(caching_ctl);
564 spin_lock(&block_group->lock);
565 block_group->caching_ctl = NULL;
566 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
567 spin_unlock(&block_group->lock);
569 #ifdef CONFIG_BTRFS_DEBUG
570 if (btrfs_should_fragment_free_space(block_group)) {
573 spin_lock(&block_group->space_info->lock);
574 spin_lock(&block_group->lock);
575 bytes_used = block_group->key.offset -
576 btrfs_block_group_used(&block_group->item);
577 block_group->space_info->bytes_used += bytes_used >> 1;
578 spin_unlock(&block_group->lock);
579 spin_unlock(&block_group->space_info->lock);
580 fragment_free_space(block_group);
584 caching_ctl->progress = (u64)-1;
586 up_read(&fs_info->commit_root_sem);
587 free_excluded_extents(fs_info, block_group);
588 mutex_unlock(&caching_ctl->mutex);
590 wake_up(&caching_ctl->wait);
592 put_caching_control(caching_ctl);
593 btrfs_put_block_group(block_group);
596 static int cache_block_group(struct btrfs_block_group_cache *cache,
600 struct btrfs_fs_info *fs_info = cache->fs_info;
601 struct btrfs_caching_control *caching_ctl;
604 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
608 INIT_LIST_HEAD(&caching_ctl->list);
609 mutex_init(&caching_ctl->mutex);
610 init_waitqueue_head(&caching_ctl->wait);
611 caching_ctl->block_group = cache;
612 caching_ctl->progress = cache->key.objectid;
613 refcount_set(&caching_ctl->count, 1);
614 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
615 caching_thread, NULL, NULL);
617 spin_lock(&cache->lock);
619 * This should be a rare occasion, but this could happen I think in the
620 * case where one thread starts to load the space cache info, and then
621 * some other thread starts a transaction commit which tries to do an
622 * allocation while the other thread is still loading the space cache
623 * info. The previous loop should have kept us from choosing this block
624 * group, but if we've moved to the state where we will wait on caching
625 * block groups we need to first check if we're doing a fast load here,
626 * so we can wait for it to finish, otherwise we could end up allocating
627 * from a block group who's cache gets evicted for one reason or
630 while (cache->cached == BTRFS_CACHE_FAST) {
631 struct btrfs_caching_control *ctl;
633 ctl = cache->caching_ctl;
634 refcount_inc(&ctl->count);
635 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
636 spin_unlock(&cache->lock);
640 finish_wait(&ctl->wait, &wait);
641 put_caching_control(ctl);
642 spin_lock(&cache->lock);
645 if (cache->cached != BTRFS_CACHE_NO) {
646 spin_unlock(&cache->lock);
650 WARN_ON(cache->caching_ctl);
651 cache->caching_ctl = caching_ctl;
652 cache->cached = BTRFS_CACHE_FAST;
653 spin_unlock(&cache->lock);
655 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
656 mutex_lock(&caching_ctl->mutex);
657 ret = load_free_space_cache(fs_info, cache);
659 spin_lock(&cache->lock);
661 cache->caching_ctl = NULL;
662 cache->cached = BTRFS_CACHE_FINISHED;
663 cache->last_byte_to_unpin = (u64)-1;
664 caching_ctl->progress = (u64)-1;
666 if (load_cache_only) {
667 cache->caching_ctl = NULL;
668 cache->cached = BTRFS_CACHE_NO;
670 cache->cached = BTRFS_CACHE_STARTED;
671 cache->has_caching_ctl = 1;
674 spin_unlock(&cache->lock);
675 #ifdef CONFIG_BTRFS_DEBUG
677 btrfs_should_fragment_free_space(cache)) {
680 spin_lock(&cache->space_info->lock);
681 spin_lock(&cache->lock);
682 bytes_used = cache->key.offset -
683 btrfs_block_group_used(&cache->item);
684 cache->space_info->bytes_used += bytes_used >> 1;
685 spin_unlock(&cache->lock);
686 spin_unlock(&cache->space_info->lock);
687 fragment_free_space(cache);
690 mutex_unlock(&caching_ctl->mutex);
692 wake_up(&caching_ctl->wait);
694 put_caching_control(caching_ctl);
695 free_excluded_extents(fs_info, cache);
700 * We're either using the free space tree or no caching at all.
701 * Set cached to the appropriate value and wakeup any waiters.
703 spin_lock(&cache->lock);
704 if (load_cache_only) {
705 cache->caching_ctl = NULL;
706 cache->cached = BTRFS_CACHE_NO;
708 cache->cached = BTRFS_CACHE_STARTED;
709 cache->has_caching_ctl = 1;
711 spin_unlock(&cache->lock);
712 wake_up(&caching_ctl->wait);
715 if (load_cache_only) {
716 put_caching_control(caching_ctl);
720 down_write(&fs_info->commit_root_sem);
721 refcount_inc(&caching_ctl->count);
722 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
723 up_write(&fs_info->commit_root_sem);
725 btrfs_get_block_group(cache);
727 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
733 * return the block group that starts at or after bytenr
735 static struct btrfs_block_group_cache *
736 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
738 return block_group_cache_tree_search(info, bytenr, 0);
742 * return the block group that contains the given bytenr
744 struct btrfs_block_group_cache *btrfs_lookup_block_group(
745 struct btrfs_fs_info *info,
748 return block_group_cache_tree_search(info, bytenr, 1);
751 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
754 struct list_head *head = &info->space_info;
755 struct btrfs_space_info *found;
757 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
760 list_for_each_entry_rcu(found, head, list) {
761 if (found->flags & flags) {
770 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
771 u64 owner, u64 root_objectid)
773 struct btrfs_space_info *space_info;
776 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
777 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
778 flags = BTRFS_BLOCK_GROUP_SYSTEM;
780 flags = BTRFS_BLOCK_GROUP_METADATA;
782 flags = BTRFS_BLOCK_GROUP_DATA;
785 space_info = __find_space_info(fs_info, flags);
787 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
791 * after adding space to the filesystem, we need to clear the full flags
792 * on all the space infos.
794 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
796 struct list_head *head = &info->space_info;
797 struct btrfs_space_info *found;
800 list_for_each_entry_rcu(found, head, list)
805 /* simple helper to search for an existing data extent at a given offset */
806 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
809 struct btrfs_key key;
810 struct btrfs_path *path;
812 path = btrfs_alloc_path();
816 key.objectid = start;
818 key.type = BTRFS_EXTENT_ITEM_KEY;
819 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
820 btrfs_free_path(path);
825 * helper function to lookup reference count and flags of a tree block.
827 * the head node for delayed ref is used to store the sum of all the
828 * reference count modifications queued up in the rbtree. the head
829 * node may also store the extent flags to set. This way you can check
830 * to see what the reference count and extent flags would be if all of
831 * the delayed refs are not processed.
833 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
834 struct btrfs_fs_info *fs_info, u64 bytenr,
835 u64 offset, int metadata, u64 *refs, u64 *flags)
837 struct btrfs_delayed_ref_head *head;
838 struct btrfs_delayed_ref_root *delayed_refs;
839 struct btrfs_path *path;
840 struct btrfs_extent_item *ei;
841 struct extent_buffer *leaf;
842 struct btrfs_key key;
849 * If we don't have skinny metadata, don't bother doing anything
852 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
853 offset = fs_info->nodesize;
857 path = btrfs_alloc_path();
862 path->skip_locking = 1;
863 path->search_commit_root = 1;
867 key.objectid = bytenr;
870 key.type = BTRFS_METADATA_ITEM_KEY;
872 key.type = BTRFS_EXTENT_ITEM_KEY;
874 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
878 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
879 if (path->slots[0]) {
881 btrfs_item_key_to_cpu(path->nodes[0], &key,
883 if (key.objectid == bytenr &&
884 key.type == BTRFS_EXTENT_ITEM_KEY &&
885 key.offset == fs_info->nodesize)
891 leaf = path->nodes[0];
892 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
893 if (item_size >= sizeof(*ei)) {
894 ei = btrfs_item_ptr(leaf, path->slots[0],
895 struct btrfs_extent_item);
896 num_refs = btrfs_extent_refs(leaf, ei);
897 extent_flags = btrfs_extent_flags(leaf, ei);
899 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
900 struct btrfs_extent_item_v0 *ei0;
901 BUG_ON(item_size != sizeof(*ei0));
902 ei0 = btrfs_item_ptr(leaf, path->slots[0],
903 struct btrfs_extent_item_v0);
904 num_refs = btrfs_extent_refs_v0(leaf, ei0);
905 /* FIXME: this isn't correct for data */
906 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
911 BUG_ON(num_refs == 0);
921 delayed_refs = &trans->transaction->delayed_refs;
922 spin_lock(&delayed_refs->lock);
923 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
925 if (!mutex_trylock(&head->mutex)) {
926 refcount_inc(&head->node.refs);
927 spin_unlock(&delayed_refs->lock);
929 btrfs_release_path(path);
932 * Mutex was contended, block until it's released and try
935 mutex_lock(&head->mutex);
936 mutex_unlock(&head->mutex);
937 btrfs_put_delayed_ref(&head->node);
940 spin_lock(&head->lock);
941 if (head->extent_op && head->extent_op->update_flags)
942 extent_flags |= head->extent_op->flags_to_set;
944 BUG_ON(num_refs == 0);
946 num_refs += head->node.ref_mod;
947 spin_unlock(&head->lock);
948 mutex_unlock(&head->mutex);
950 spin_unlock(&delayed_refs->lock);
952 WARN_ON(num_refs == 0);
956 *flags = extent_flags;
958 btrfs_free_path(path);
963 * Back reference rules. Back refs have three main goals:
965 * 1) differentiate between all holders of references to an extent so that
966 * when a reference is dropped we can make sure it was a valid reference
967 * before freeing the extent.
969 * 2) Provide enough information to quickly find the holders of an extent
970 * if we notice a given block is corrupted or bad.
972 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
973 * maintenance. This is actually the same as #2, but with a slightly
974 * different use case.
976 * There are two kinds of back refs. The implicit back refs is optimized
977 * for pointers in non-shared tree blocks. For a given pointer in a block,
978 * back refs of this kind provide information about the block's owner tree
979 * and the pointer's key. These information allow us to find the block by
980 * b-tree searching. The full back refs is for pointers in tree blocks not
981 * referenced by their owner trees. The location of tree block is recorded
982 * in the back refs. Actually the full back refs is generic, and can be
983 * used in all cases the implicit back refs is used. The major shortcoming
984 * of the full back refs is its overhead. Every time a tree block gets
985 * COWed, we have to update back refs entry for all pointers in it.
987 * For a newly allocated tree block, we use implicit back refs for
988 * pointers in it. This means most tree related operations only involve
989 * implicit back refs. For a tree block created in old transaction, the
990 * only way to drop a reference to it is COW it. So we can detect the
991 * event that tree block loses its owner tree's reference and do the
992 * back refs conversion.
994 * When a tree block is COWed through a tree, there are four cases:
996 * The reference count of the block is one and the tree is the block's
997 * owner tree. Nothing to do in this case.
999 * The reference count of the block is one and the tree is not the
1000 * block's owner tree. In this case, full back refs is used for pointers
1001 * in the block. Remove these full back refs, add implicit back refs for
1002 * every pointers in the new block.
1004 * The reference count of the block is greater than one and the tree is
1005 * the block's owner tree. In this case, implicit back refs is used for
1006 * pointers in the block. Add full back refs for every pointers in the
1007 * block, increase lower level extents' reference counts. The original
1008 * implicit back refs are entailed to the new block.
1010 * The reference count of the block is greater than one and the tree is
1011 * not the block's owner tree. Add implicit back refs for every pointer in
1012 * the new block, increase lower level extents' reference count.
1014 * Back Reference Key composing:
1016 * The key objectid corresponds to the first byte in the extent,
1017 * The key type is used to differentiate between types of back refs.
1018 * There are different meanings of the key offset for different types
1021 * File extents can be referenced by:
1023 * - multiple snapshots, subvolumes, or different generations in one subvol
1024 * - different files inside a single subvolume
1025 * - different offsets inside a file (bookend extents in file.c)
1027 * The extent ref structure for the implicit back refs has fields for:
1029 * - Objectid of the subvolume root
1030 * - objectid of the file holding the reference
1031 * - original offset in the file
1032 * - how many bookend extents
1034 * The key offset for the implicit back refs is hash of the first
1037 * The extent ref structure for the full back refs has field for:
1039 * - number of pointers in the tree leaf
1041 * The key offset for the implicit back refs is the first byte of
1044 * When a file extent is allocated, The implicit back refs is used.
1045 * the fields are filled in:
1047 * (root_key.objectid, inode objectid, offset in file, 1)
1049 * When a file extent is removed file truncation, we find the
1050 * corresponding implicit back refs and check the following fields:
1052 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1054 * Btree extents can be referenced by:
1056 * - Different subvolumes
1058 * Both the implicit back refs and the full back refs for tree blocks
1059 * only consist of key. The key offset for the implicit back refs is
1060 * objectid of block's owner tree. The key offset for the full back refs
1061 * is the first byte of parent block.
1063 * When implicit back refs is used, information about the lowest key and
1064 * level of the tree block are required. These information are stored in
1065 * tree block info structure.
1068 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1069 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1070 struct btrfs_fs_info *fs_info,
1071 struct btrfs_path *path,
1072 u64 owner, u32 extra_size)
1074 struct btrfs_root *root = fs_info->extent_root;
1075 struct btrfs_extent_item *item;
1076 struct btrfs_extent_item_v0 *ei0;
1077 struct btrfs_extent_ref_v0 *ref0;
1078 struct btrfs_tree_block_info *bi;
1079 struct extent_buffer *leaf;
1080 struct btrfs_key key;
1081 struct btrfs_key found_key;
1082 u32 new_size = sizeof(*item);
1086 leaf = path->nodes[0];
1087 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1089 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1090 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1091 struct btrfs_extent_item_v0);
1092 refs = btrfs_extent_refs_v0(leaf, ei0);
1094 if (owner == (u64)-1) {
1096 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1097 ret = btrfs_next_leaf(root, path);
1100 BUG_ON(ret > 0); /* Corruption */
1101 leaf = path->nodes[0];
1103 btrfs_item_key_to_cpu(leaf, &found_key,
1105 BUG_ON(key.objectid != found_key.objectid);
1106 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1110 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1111 struct btrfs_extent_ref_v0);
1112 owner = btrfs_ref_objectid_v0(leaf, ref0);
1116 btrfs_release_path(path);
1118 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1119 new_size += sizeof(*bi);
1121 new_size -= sizeof(*ei0);
1122 ret = btrfs_search_slot(trans, root, &key, path,
1123 new_size + extra_size, 1);
1126 BUG_ON(ret); /* Corruption */
1128 btrfs_extend_item(fs_info, path, new_size);
1130 leaf = path->nodes[0];
1131 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1132 btrfs_set_extent_refs(leaf, item, refs);
1133 /* FIXME: get real generation */
1134 btrfs_set_extent_generation(leaf, item, 0);
1135 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1136 btrfs_set_extent_flags(leaf, item,
1137 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1138 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1139 bi = (struct btrfs_tree_block_info *)(item + 1);
1140 /* FIXME: get first key of the block */
1141 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1142 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1144 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1146 btrfs_mark_buffer_dirty(leaf);
1151 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1153 u32 high_crc = ~(u32)0;
1154 u32 low_crc = ~(u32)0;
1157 lenum = cpu_to_le64(root_objectid);
1158 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1159 lenum = cpu_to_le64(owner);
1160 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1161 lenum = cpu_to_le64(offset);
1162 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1164 return ((u64)high_crc << 31) ^ (u64)low_crc;
1167 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1168 struct btrfs_extent_data_ref *ref)
1170 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1171 btrfs_extent_data_ref_objectid(leaf, ref),
1172 btrfs_extent_data_ref_offset(leaf, ref));
1175 static int match_extent_data_ref(struct extent_buffer *leaf,
1176 struct btrfs_extent_data_ref *ref,
1177 u64 root_objectid, u64 owner, u64 offset)
1179 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1180 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1181 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1186 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1187 struct btrfs_fs_info *fs_info,
1188 struct btrfs_path *path,
1189 u64 bytenr, u64 parent,
1191 u64 owner, u64 offset)
1193 struct btrfs_root *root = fs_info->extent_root;
1194 struct btrfs_key key;
1195 struct btrfs_extent_data_ref *ref;
1196 struct extent_buffer *leaf;
1202 key.objectid = bytenr;
1204 key.type = BTRFS_SHARED_DATA_REF_KEY;
1205 key.offset = parent;
1207 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1208 key.offset = hash_extent_data_ref(root_objectid,
1213 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1222 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1223 key.type = BTRFS_EXTENT_REF_V0_KEY;
1224 btrfs_release_path(path);
1225 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1236 leaf = path->nodes[0];
1237 nritems = btrfs_header_nritems(leaf);
1239 if (path->slots[0] >= nritems) {
1240 ret = btrfs_next_leaf(root, path);
1246 leaf = path->nodes[0];
1247 nritems = btrfs_header_nritems(leaf);
1251 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1252 if (key.objectid != bytenr ||
1253 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1256 ref = btrfs_item_ptr(leaf, path->slots[0],
1257 struct btrfs_extent_data_ref);
1259 if (match_extent_data_ref(leaf, ref, root_objectid,
1262 btrfs_release_path(path);
1274 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1275 struct btrfs_fs_info *fs_info,
1276 struct btrfs_path *path,
1277 u64 bytenr, u64 parent,
1278 u64 root_objectid, u64 owner,
1279 u64 offset, int refs_to_add)
1281 struct btrfs_root *root = fs_info->extent_root;
1282 struct btrfs_key key;
1283 struct extent_buffer *leaf;
1288 key.objectid = bytenr;
1290 key.type = BTRFS_SHARED_DATA_REF_KEY;
1291 key.offset = parent;
1292 size = sizeof(struct btrfs_shared_data_ref);
1294 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1295 key.offset = hash_extent_data_ref(root_objectid,
1297 size = sizeof(struct btrfs_extent_data_ref);
1300 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1301 if (ret && ret != -EEXIST)
1304 leaf = path->nodes[0];
1306 struct btrfs_shared_data_ref *ref;
1307 ref = btrfs_item_ptr(leaf, path->slots[0],
1308 struct btrfs_shared_data_ref);
1310 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1312 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1313 num_refs += refs_to_add;
1314 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1317 struct btrfs_extent_data_ref *ref;
1318 while (ret == -EEXIST) {
1319 ref = btrfs_item_ptr(leaf, path->slots[0],
1320 struct btrfs_extent_data_ref);
1321 if (match_extent_data_ref(leaf, ref, root_objectid,
1324 btrfs_release_path(path);
1326 ret = btrfs_insert_empty_item(trans, root, path, &key,
1328 if (ret && ret != -EEXIST)
1331 leaf = path->nodes[0];
1333 ref = btrfs_item_ptr(leaf, path->slots[0],
1334 struct btrfs_extent_data_ref);
1336 btrfs_set_extent_data_ref_root(leaf, ref,
1338 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1339 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1340 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1343 num_refs += refs_to_add;
1344 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1347 btrfs_mark_buffer_dirty(leaf);
1350 btrfs_release_path(path);
1354 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1355 struct btrfs_fs_info *fs_info,
1356 struct btrfs_path *path,
1357 int refs_to_drop, int *last_ref)
1359 struct btrfs_key key;
1360 struct btrfs_extent_data_ref *ref1 = NULL;
1361 struct btrfs_shared_data_ref *ref2 = NULL;
1362 struct extent_buffer *leaf;
1366 leaf = path->nodes[0];
1367 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1369 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1370 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1371 struct btrfs_extent_data_ref);
1372 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1373 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1374 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1375 struct btrfs_shared_data_ref);
1376 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1377 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1378 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1379 struct btrfs_extent_ref_v0 *ref0;
1380 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1381 struct btrfs_extent_ref_v0);
1382 num_refs = btrfs_ref_count_v0(leaf, ref0);
1388 BUG_ON(num_refs < refs_to_drop);
1389 num_refs -= refs_to_drop;
1391 if (num_refs == 0) {
1392 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1395 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1396 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1397 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1398 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1399 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1401 struct btrfs_extent_ref_v0 *ref0;
1402 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1403 struct btrfs_extent_ref_v0);
1404 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1407 btrfs_mark_buffer_dirty(leaf);
1412 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1413 struct btrfs_extent_inline_ref *iref)
1415 struct btrfs_key key;
1416 struct extent_buffer *leaf;
1417 struct btrfs_extent_data_ref *ref1;
1418 struct btrfs_shared_data_ref *ref2;
1421 leaf = path->nodes[0];
1422 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1424 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1425 BTRFS_EXTENT_DATA_REF_KEY) {
1426 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1427 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1429 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1430 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1432 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1433 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1434 struct btrfs_extent_data_ref);
1435 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1436 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1437 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1438 struct btrfs_shared_data_ref);
1439 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1440 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1441 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1442 struct btrfs_extent_ref_v0 *ref0;
1443 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1444 struct btrfs_extent_ref_v0);
1445 num_refs = btrfs_ref_count_v0(leaf, ref0);
1453 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1454 struct btrfs_fs_info *fs_info,
1455 struct btrfs_path *path,
1456 u64 bytenr, u64 parent,
1459 struct btrfs_root *root = fs_info->extent_root;
1460 struct btrfs_key key;
1463 key.objectid = bytenr;
1465 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1466 key.offset = parent;
1468 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1469 key.offset = root_objectid;
1472 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1475 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1476 if (ret == -ENOENT && parent) {
1477 btrfs_release_path(path);
1478 key.type = BTRFS_EXTENT_REF_V0_KEY;
1479 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1487 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1488 struct btrfs_fs_info *fs_info,
1489 struct btrfs_path *path,
1490 u64 bytenr, u64 parent,
1493 struct btrfs_key key;
1496 key.objectid = bytenr;
1498 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1499 key.offset = parent;
1501 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1502 key.offset = root_objectid;
1505 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1507 btrfs_release_path(path);
1511 static inline int extent_ref_type(u64 parent, u64 owner)
1514 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1516 type = BTRFS_SHARED_BLOCK_REF_KEY;
1518 type = BTRFS_TREE_BLOCK_REF_KEY;
1521 type = BTRFS_SHARED_DATA_REF_KEY;
1523 type = BTRFS_EXTENT_DATA_REF_KEY;
1528 static int find_next_key(struct btrfs_path *path, int level,
1529 struct btrfs_key *key)
1532 for (; level < BTRFS_MAX_LEVEL; level++) {
1533 if (!path->nodes[level])
1535 if (path->slots[level] + 1 >=
1536 btrfs_header_nritems(path->nodes[level]))
1539 btrfs_item_key_to_cpu(path->nodes[level], key,
1540 path->slots[level] + 1);
1542 btrfs_node_key_to_cpu(path->nodes[level], key,
1543 path->slots[level] + 1);
1550 * look for inline back ref. if back ref is found, *ref_ret is set
1551 * to the address of inline back ref, and 0 is returned.
1553 * if back ref isn't found, *ref_ret is set to the address where it
1554 * should be inserted, and -ENOENT is returned.
1556 * if insert is true and there are too many inline back refs, the path
1557 * points to the extent item, and -EAGAIN is returned.
1559 * NOTE: inline back refs are ordered in the same way that back ref
1560 * items in the tree are ordered.
1562 static noinline_for_stack
1563 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1564 struct btrfs_fs_info *fs_info,
1565 struct btrfs_path *path,
1566 struct btrfs_extent_inline_ref **ref_ret,
1567 u64 bytenr, u64 num_bytes,
1568 u64 parent, u64 root_objectid,
1569 u64 owner, u64 offset, int insert)
1571 struct btrfs_root *root = fs_info->extent_root;
1572 struct btrfs_key key;
1573 struct extent_buffer *leaf;
1574 struct btrfs_extent_item *ei;
1575 struct btrfs_extent_inline_ref *iref;
1585 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1587 key.objectid = bytenr;
1588 key.type = BTRFS_EXTENT_ITEM_KEY;
1589 key.offset = num_bytes;
1591 want = extent_ref_type(parent, owner);
1593 extra_size = btrfs_extent_inline_ref_size(want);
1594 path->keep_locks = 1;
1599 * Owner is our parent level, so we can just add one to get the level
1600 * for the block we are interested in.
1602 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1603 key.type = BTRFS_METADATA_ITEM_KEY;
1608 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1615 * We may be a newly converted file system which still has the old fat
1616 * extent entries for metadata, so try and see if we have one of those.
1618 if (ret > 0 && skinny_metadata) {
1619 skinny_metadata = false;
1620 if (path->slots[0]) {
1622 btrfs_item_key_to_cpu(path->nodes[0], &key,
1624 if (key.objectid == bytenr &&
1625 key.type == BTRFS_EXTENT_ITEM_KEY &&
1626 key.offset == num_bytes)
1630 key.objectid = bytenr;
1631 key.type = BTRFS_EXTENT_ITEM_KEY;
1632 key.offset = num_bytes;
1633 btrfs_release_path(path);
1638 if (ret && !insert) {
1641 } else if (WARN_ON(ret)) {
1646 leaf = path->nodes[0];
1647 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1648 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1649 if (item_size < sizeof(*ei)) {
1654 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1660 leaf = path->nodes[0];
1661 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1664 BUG_ON(item_size < sizeof(*ei));
1666 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1667 flags = btrfs_extent_flags(leaf, ei);
1669 ptr = (unsigned long)(ei + 1);
1670 end = (unsigned long)ei + item_size;
1672 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1673 ptr += sizeof(struct btrfs_tree_block_info);
1683 iref = (struct btrfs_extent_inline_ref *)ptr;
1684 type = btrfs_extent_inline_ref_type(leaf, iref);
1688 ptr += btrfs_extent_inline_ref_size(type);
1692 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1693 struct btrfs_extent_data_ref *dref;
1694 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1695 if (match_extent_data_ref(leaf, dref, root_objectid,
1700 if (hash_extent_data_ref_item(leaf, dref) <
1701 hash_extent_data_ref(root_objectid, owner, offset))
1705 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1707 if (parent == ref_offset) {
1711 if (ref_offset < parent)
1714 if (root_objectid == ref_offset) {
1718 if (ref_offset < root_objectid)
1722 ptr += btrfs_extent_inline_ref_size(type);
1724 if (err == -ENOENT && insert) {
1725 if (item_size + extra_size >=
1726 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1731 * To add new inline back ref, we have to make sure
1732 * there is no corresponding back ref item.
1733 * For simplicity, we just do not add new inline back
1734 * ref if there is any kind of item for this block
1736 if (find_next_key(path, 0, &key) == 0 &&
1737 key.objectid == bytenr &&
1738 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1743 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1746 path->keep_locks = 0;
1747 btrfs_unlock_up_safe(path, 1);
1753 * helper to add new inline back ref
1755 static noinline_for_stack
1756 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1757 struct btrfs_path *path,
1758 struct btrfs_extent_inline_ref *iref,
1759 u64 parent, u64 root_objectid,
1760 u64 owner, u64 offset, int refs_to_add,
1761 struct btrfs_delayed_extent_op *extent_op)
1763 struct extent_buffer *leaf;
1764 struct btrfs_extent_item *ei;
1767 unsigned long item_offset;
1772 leaf = path->nodes[0];
1773 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1774 item_offset = (unsigned long)iref - (unsigned long)ei;
1776 type = extent_ref_type(parent, owner);
1777 size = btrfs_extent_inline_ref_size(type);
1779 btrfs_extend_item(fs_info, path, size);
1781 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1782 refs = btrfs_extent_refs(leaf, ei);
1783 refs += refs_to_add;
1784 btrfs_set_extent_refs(leaf, ei, refs);
1786 __run_delayed_extent_op(extent_op, leaf, ei);
1788 ptr = (unsigned long)ei + item_offset;
1789 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1790 if (ptr < end - size)
1791 memmove_extent_buffer(leaf, ptr + size, ptr,
1794 iref = (struct btrfs_extent_inline_ref *)ptr;
1795 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1796 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1797 struct btrfs_extent_data_ref *dref;
1798 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1799 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1800 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1801 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1802 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1803 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1804 struct btrfs_shared_data_ref *sref;
1805 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1806 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1807 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1808 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1809 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1811 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1813 btrfs_mark_buffer_dirty(leaf);
1816 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1817 struct btrfs_fs_info *fs_info,
1818 struct btrfs_path *path,
1819 struct btrfs_extent_inline_ref **ref_ret,
1820 u64 bytenr, u64 num_bytes, u64 parent,
1821 u64 root_objectid, u64 owner, u64 offset)
1825 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1826 bytenr, num_bytes, parent,
1827 root_objectid, owner, offset, 0);
1831 btrfs_release_path(path);
1834 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1835 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1836 parent, root_objectid);
1838 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1839 parent, root_objectid, owner,
1846 * helper to update/remove inline back ref
1848 static noinline_for_stack
1849 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1850 struct btrfs_path *path,
1851 struct btrfs_extent_inline_ref *iref,
1853 struct btrfs_delayed_extent_op *extent_op,
1856 struct extent_buffer *leaf;
1857 struct btrfs_extent_item *ei;
1858 struct btrfs_extent_data_ref *dref = NULL;
1859 struct btrfs_shared_data_ref *sref = NULL;
1867 leaf = path->nodes[0];
1868 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1869 refs = btrfs_extent_refs(leaf, ei);
1870 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1871 refs += refs_to_mod;
1872 btrfs_set_extent_refs(leaf, ei, refs);
1874 __run_delayed_extent_op(extent_op, leaf, ei);
1876 type = btrfs_extent_inline_ref_type(leaf, iref);
1878 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1879 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1880 refs = btrfs_extent_data_ref_count(leaf, dref);
1881 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1882 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1883 refs = btrfs_shared_data_ref_count(leaf, sref);
1886 BUG_ON(refs_to_mod != -1);
1889 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1890 refs += refs_to_mod;
1893 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1894 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1896 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1899 size = btrfs_extent_inline_ref_size(type);
1900 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1901 ptr = (unsigned long)iref;
1902 end = (unsigned long)ei + item_size;
1903 if (ptr + size < end)
1904 memmove_extent_buffer(leaf, ptr, ptr + size,
1907 btrfs_truncate_item(fs_info, path, item_size, 1);
1909 btrfs_mark_buffer_dirty(leaf);
1912 static noinline_for_stack
1913 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1914 struct btrfs_fs_info *fs_info,
1915 struct btrfs_path *path,
1916 u64 bytenr, u64 num_bytes, u64 parent,
1917 u64 root_objectid, u64 owner,
1918 u64 offset, int refs_to_add,
1919 struct btrfs_delayed_extent_op *extent_op)
1921 struct btrfs_extent_inline_ref *iref;
1924 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1925 bytenr, num_bytes, parent,
1926 root_objectid, owner, offset, 1);
1928 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1929 update_inline_extent_backref(fs_info, path, iref,
1930 refs_to_add, extent_op, NULL);
1931 } else if (ret == -ENOENT) {
1932 setup_inline_extent_backref(fs_info, path, iref, parent,
1933 root_objectid, owner, offset,
1934 refs_to_add, extent_op);
1940 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1941 struct btrfs_fs_info *fs_info,
1942 struct btrfs_path *path,
1943 u64 bytenr, u64 parent, u64 root_objectid,
1944 u64 owner, u64 offset, int refs_to_add)
1947 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1948 BUG_ON(refs_to_add != 1);
1949 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
1950 parent, root_objectid);
1952 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
1953 parent, root_objectid,
1954 owner, offset, refs_to_add);
1959 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1960 struct btrfs_fs_info *fs_info,
1961 struct btrfs_path *path,
1962 struct btrfs_extent_inline_ref *iref,
1963 int refs_to_drop, int is_data, int *last_ref)
1967 BUG_ON(!is_data && refs_to_drop != 1);
1969 update_inline_extent_backref(fs_info, path, iref,
1970 -refs_to_drop, NULL, last_ref);
1971 } else if (is_data) {
1972 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
1976 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1981 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1982 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1983 u64 *discarded_bytes)
1986 u64 bytes_left, end;
1987 u64 aligned_start = ALIGN(start, 1 << 9);
1989 if (WARN_ON(start != aligned_start)) {
1990 len -= aligned_start - start;
1991 len = round_down(len, 1 << 9);
1992 start = aligned_start;
1995 *discarded_bytes = 0;
2003 /* Skip any superblocks on this device. */
2004 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2005 u64 sb_start = btrfs_sb_offset(j);
2006 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2007 u64 size = sb_start - start;
2009 if (!in_range(sb_start, start, bytes_left) &&
2010 !in_range(sb_end, start, bytes_left) &&
2011 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2015 * Superblock spans beginning of range. Adjust start and
2018 if (sb_start <= start) {
2019 start += sb_end - start;
2024 bytes_left = end - start;
2029 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2032 *discarded_bytes += size;
2033 else if (ret != -EOPNOTSUPP)
2042 bytes_left = end - start;
2046 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2049 *discarded_bytes += bytes_left;
2054 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2055 u64 num_bytes, u64 *actual_bytes)
2058 u64 discarded_bytes = 0;
2059 struct btrfs_bio *bbio = NULL;
2063 * Avoid races with device replace and make sure our bbio has devices
2064 * associated to its stripes that don't go away while we are discarding.
2066 btrfs_bio_counter_inc_blocked(fs_info);
2067 /* Tell the block device(s) that the sectors can be discarded */
2068 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2070 /* Error condition is -ENOMEM */
2072 struct btrfs_bio_stripe *stripe = bbio->stripes;
2076 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2078 if (!stripe->dev->can_discard)
2081 ret = btrfs_issue_discard(stripe->dev->bdev,
2086 discarded_bytes += bytes;
2087 else if (ret != -EOPNOTSUPP)
2088 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2091 * Just in case we get back EOPNOTSUPP for some reason,
2092 * just ignore the return value so we don't screw up
2093 * people calling discard_extent.
2097 btrfs_put_bbio(bbio);
2099 btrfs_bio_counter_dec(fs_info);
2102 *actual_bytes = discarded_bytes;
2105 if (ret == -EOPNOTSUPP)
2110 /* Can return -ENOMEM */
2111 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2112 struct btrfs_fs_info *fs_info,
2113 u64 bytenr, u64 num_bytes, u64 parent,
2114 u64 root_objectid, u64 owner, u64 offset)
2116 int old_ref_mod, new_ref_mod;
2119 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2120 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2122 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2123 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2125 root_objectid, (int)owner,
2126 BTRFS_ADD_DELAYED_REF, NULL,
2127 &old_ref_mod, &new_ref_mod);
2129 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2131 root_objectid, owner, offset,
2132 0, BTRFS_ADD_DELAYED_REF,
2133 &old_ref_mod, &new_ref_mod);
2136 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2137 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2142 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2143 struct btrfs_fs_info *fs_info,
2144 struct btrfs_delayed_ref_node *node,
2145 u64 parent, u64 root_objectid,
2146 u64 owner, u64 offset, int refs_to_add,
2147 struct btrfs_delayed_extent_op *extent_op)
2149 struct btrfs_path *path;
2150 struct extent_buffer *leaf;
2151 struct btrfs_extent_item *item;
2152 struct btrfs_key key;
2153 u64 bytenr = node->bytenr;
2154 u64 num_bytes = node->num_bytes;
2158 path = btrfs_alloc_path();
2162 path->reada = READA_FORWARD;
2163 path->leave_spinning = 1;
2164 /* this will setup the path even if it fails to insert the back ref */
2165 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2166 num_bytes, parent, root_objectid,
2168 refs_to_add, extent_op);
2169 if ((ret < 0 && ret != -EAGAIN) || !ret)
2173 * Ok we had -EAGAIN which means we didn't have space to insert and
2174 * inline extent ref, so just update the reference count and add a
2177 leaf = path->nodes[0];
2178 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2179 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2180 refs = btrfs_extent_refs(leaf, item);
2181 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2183 __run_delayed_extent_op(extent_op, leaf, item);
2185 btrfs_mark_buffer_dirty(leaf);
2186 btrfs_release_path(path);
2188 path->reada = READA_FORWARD;
2189 path->leave_spinning = 1;
2190 /* now insert the actual backref */
2191 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2192 root_objectid, owner, offset, refs_to_add);
2194 btrfs_abort_transaction(trans, ret);
2196 btrfs_free_path(path);
2200 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2201 struct btrfs_fs_info *fs_info,
2202 struct btrfs_delayed_ref_node *node,
2203 struct btrfs_delayed_extent_op *extent_op,
2204 int insert_reserved)
2207 struct btrfs_delayed_data_ref *ref;
2208 struct btrfs_key ins;
2213 ins.objectid = node->bytenr;
2214 ins.offset = node->num_bytes;
2215 ins.type = BTRFS_EXTENT_ITEM_KEY;
2217 ref = btrfs_delayed_node_to_data_ref(node);
2218 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2220 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2221 parent = ref->parent;
2222 ref_root = ref->root;
2224 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2226 flags |= extent_op->flags_to_set;
2227 ret = alloc_reserved_file_extent(trans, fs_info,
2228 parent, ref_root, flags,
2229 ref->objectid, ref->offset,
2230 &ins, node->ref_mod);
2231 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2232 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2233 ref_root, ref->objectid,
2234 ref->offset, node->ref_mod,
2236 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2237 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2238 ref_root, ref->objectid,
2239 ref->offset, node->ref_mod,
2247 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2248 struct extent_buffer *leaf,
2249 struct btrfs_extent_item *ei)
2251 u64 flags = btrfs_extent_flags(leaf, ei);
2252 if (extent_op->update_flags) {
2253 flags |= extent_op->flags_to_set;
2254 btrfs_set_extent_flags(leaf, ei, flags);
2257 if (extent_op->update_key) {
2258 struct btrfs_tree_block_info *bi;
2259 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2260 bi = (struct btrfs_tree_block_info *)(ei + 1);
2261 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2265 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2266 struct btrfs_fs_info *fs_info,
2267 struct btrfs_delayed_ref_node *node,
2268 struct btrfs_delayed_extent_op *extent_op)
2270 struct btrfs_key key;
2271 struct btrfs_path *path;
2272 struct btrfs_extent_item *ei;
2273 struct extent_buffer *leaf;
2277 int metadata = !extent_op->is_data;
2282 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2285 path = btrfs_alloc_path();
2289 key.objectid = node->bytenr;
2292 key.type = BTRFS_METADATA_ITEM_KEY;
2293 key.offset = extent_op->level;
2295 key.type = BTRFS_EXTENT_ITEM_KEY;
2296 key.offset = node->num_bytes;
2300 path->reada = READA_FORWARD;
2301 path->leave_spinning = 1;
2302 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2309 if (path->slots[0] > 0) {
2311 btrfs_item_key_to_cpu(path->nodes[0], &key,
2313 if (key.objectid == node->bytenr &&
2314 key.type == BTRFS_EXTENT_ITEM_KEY &&
2315 key.offset == node->num_bytes)
2319 btrfs_release_path(path);
2322 key.objectid = node->bytenr;
2323 key.offset = node->num_bytes;
2324 key.type = BTRFS_EXTENT_ITEM_KEY;
2333 leaf = path->nodes[0];
2334 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2335 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2336 if (item_size < sizeof(*ei)) {
2337 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2342 leaf = path->nodes[0];
2343 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2346 BUG_ON(item_size < sizeof(*ei));
2347 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2348 __run_delayed_extent_op(extent_op, leaf, ei);
2350 btrfs_mark_buffer_dirty(leaf);
2352 btrfs_free_path(path);
2356 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2357 struct btrfs_fs_info *fs_info,
2358 struct btrfs_delayed_ref_node *node,
2359 struct btrfs_delayed_extent_op *extent_op,
2360 int insert_reserved)
2363 struct btrfs_delayed_tree_ref *ref;
2364 struct btrfs_key ins;
2367 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2369 ref = btrfs_delayed_node_to_tree_ref(node);
2370 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2372 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2373 parent = ref->parent;
2374 ref_root = ref->root;
2376 ins.objectid = node->bytenr;
2377 if (skinny_metadata) {
2378 ins.offset = ref->level;
2379 ins.type = BTRFS_METADATA_ITEM_KEY;
2381 ins.offset = node->num_bytes;
2382 ins.type = BTRFS_EXTENT_ITEM_KEY;
2385 if (node->ref_mod != 1) {
2387 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2388 node->bytenr, node->ref_mod, node->action, ref_root,
2392 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2393 BUG_ON(!extent_op || !extent_op->update_flags);
2394 ret = alloc_reserved_tree_block(trans, fs_info,
2396 extent_op->flags_to_set,
2399 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2400 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2404 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2405 ret = __btrfs_free_extent(trans, fs_info, node,
2407 ref->level, 0, 1, extent_op);
2414 /* helper function to actually process a single delayed ref entry */
2415 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2416 struct btrfs_fs_info *fs_info,
2417 struct btrfs_delayed_ref_node *node,
2418 struct btrfs_delayed_extent_op *extent_op,
2419 int insert_reserved)
2423 if (trans->aborted) {
2424 if (insert_reserved)
2425 btrfs_pin_extent(fs_info, node->bytenr,
2426 node->num_bytes, 1);
2430 if (btrfs_delayed_ref_is_head(node)) {
2431 struct btrfs_delayed_ref_head *head;
2433 * we've hit the end of the chain and we were supposed
2434 * to insert this extent into the tree. But, it got
2435 * deleted before we ever needed to insert it, so all
2436 * we have to do is clean up the accounting
2439 head = btrfs_delayed_node_to_head(node);
2440 trace_run_delayed_ref_head(fs_info, node, head, node->action);
2442 if (head->total_ref_mod < 0) {
2443 struct btrfs_block_group_cache *cache;
2445 cache = btrfs_lookup_block_group(fs_info, node->bytenr);
2447 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2449 btrfs_put_block_group(cache);
2452 if (insert_reserved) {
2453 btrfs_pin_extent(fs_info, node->bytenr,
2454 node->num_bytes, 1);
2455 if (head->is_data) {
2456 ret = btrfs_del_csums(trans, fs_info,
2462 /* Also free its reserved qgroup space */
2463 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2464 head->qgroup_reserved);
2468 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2469 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2470 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2472 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2473 node->type == BTRFS_SHARED_DATA_REF_KEY)
2474 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2481 static inline struct btrfs_delayed_ref_node *
2482 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2484 struct btrfs_delayed_ref_node *ref;
2486 if (list_empty(&head->ref_list))
2490 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2491 * This is to prevent a ref count from going down to zero, which deletes
2492 * the extent item from the extent tree, when there still are references
2493 * to add, which would fail because they would not find the extent item.
2495 if (!list_empty(&head->ref_add_list))
2496 return list_first_entry(&head->ref_add_list,
2497 struct btrfs_delayed_ref_node, add_list);
2499 ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
2501 ASSERT(list_empty(&ref->add_list));
2506 * Returns 0 on success or if called with an already aborted transaction.
2507 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2509 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2510 struct btrfs_fs_info *fs_info,
2513 struct btrfs_delayed_ref_root *delayed_refs;
2514 struct btrfs_delayed_ref_node *ref;
2515 struct btrfs_delayed_ref_head *locked_ref = NULL;
2516 struct btrfs_delayed_extent_op *extent_op;
2517 ktime_t start = ktime_get();
2519 unsigned long count = 0;
2520 unsigned long actual_count = 0;
2521 int must_insert_reserved = 0;
2523 delayed_refs = &trans->transaction->delayed_refs;
2529 spin_lock(&delayed_refs->lock);
2530 locked_ref = btrfs_select_ref_head(trans);
2532 spin_unlock(&delayed_refs->lock);
2536 /* grab the lock that says we are going to process
2537 * all the refs for this head */
2538 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2539 spin_unlock(&delayed_refs->lock);
2541 * we may have dropped the spin lock to get the head
2542 * mutex lock, and that might have given someone else
2543 * time to free the head. If that's true, it has been
2544 * removed from our list and we can move on.
2546 if (ret == -EAGAIN) {
2554 * We need to try and merge add/drops of the same ref since we
2555 * can run into issues with relocate dropping the implicit ref
2556 * and then it being added back again before the drop can
2557 * finish. If we merged anything we need to re-loop so we can
2559 * Or we can get node references of the same type that weren't
2560 * merged when created due to bumps in the tree mod seq, and
2561 * we need to merge them to prevent adding an inline extent
2562 * backref before dropping it (triggering a BUG_ON at
2563 * insert_inline_extent_backref()).
2565 spin_lock(&locked_ref->lock);
2566 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2570 * locked_ref is the head node, so we have to go one
2571 * node back for any delayed ref updates
2573 ref = select_delayed_ref(locked_ref);
2575 if (ref && ref->seq &&
2576 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2577 spin_unlock(&locked_ref->lock);
2578 spin_lock(&delayed_refs->lock);
2579 locked_ref->processing = 0;
2580 delayed_refs->num_heads_ready++;
2581 spin_unlock(&delayed_refs->lock);
2582 btrfs_delayed_ref_unlock(locked_ref);
2590 * record the must insert reserved flag before we
2591 * drop the spin lock.
2593 must_insert_reserved = locked_ref->must_insert_reserved;
2594 locked_ref->must_insert_reserved = 0;
2596 extent_op = locked_ref->extent_op;
2597 locked_ref->extent_op = NULL;
2602 /* All delayed refs have been processed, Go ahead
2603 * and send the head node to run_one_delayed_ref,
2604 * so that any accounting fixes can happen
2606 ref = &locked_ref->node;
2608 if (extent_op && must_insert_reserved) {
2609 btrfs_free_delayed_extent_op(extent_op);
2614 spin_unlock(&locked_ref->lock);
2615 ret = run_delayed_extent_op(trans, fs_info,
2617 btrfs_free_delayed_extent_op(extent_op);
2621 * Need to reset must_insert_reserved if
2622 * there was an error so the abort stuff
2623 * can cleanup the reserved space
2626 if (must_insert_reserved)
2627 locked_ref->must_insert_reserved = 1;
2628 spin_lock(&delayed_refs->lock);
2629 locked_ref->processing = 0;
2630 delayed_refs->num_heads_ready++;
2631 spin_unlock(&delayed_refs->lock);
2632 btrfs_debug(fs_info,
2633 "run_delayed_extent_op returned %d",
2635 btrfs_delayed_ref_unlock(locked_ref);
2642 * Need to drop our head ref lock and re-acquire the
2643 * delayed ref lock and then re-check to make sure
2646 spin_unlock(&locked_ref->lock);
2647 spin_lock(&delayed_refs->lock);
2648 spin_lock(&locked_ref->lock);
2649 if (!list_empty(&locked_ref->ref_list) ||
2650 locked_ref->extent_op) {
2651 spin_unlock(&locked_ref->lock);
2652 spin_unlock(&delayed_refs->lock);
2656 delayed_refs->num_heads--;
2657 rb_erase(&locked_ref->href_node,
2658 &delayed_refs->href_root);
2659 spin_unlock(&delayed_refs->lock);
2663 list_del(&ref->list);
2664 if (!list_empty(&ref->add_list))
2665 list_del(&ref->add_list);
2667 atomic_dec(&delayed_refs->num_entries);
2669 if (!btrfs_delayed_ref_is_head(ref)) {
2671 * when we play the delayed ref, also correct the
2674 switch (ref->action) {
2675 case BTRFS_ADD_DELAYED_REF:
2676 case BTRFS_ADD_DELAYED_EXTENT:
2677 locked_ref->node.ref_mod -= ref->ref_mod;
2679 case BTRFS_DROP_DELAYED_REF:
2680 locked_ref->node.ref_mod += ref->ref_mod;
2686 spin_unlock(&locked_ref->lock);
2688 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2689 must_insert_reserved);
2691 btrfs_free_delayed_extent_op(extent_op);
2693 spin_lock(&delayed_refs->lock);
2694 locked_ref->processing = 0;
2695 delayed_refs->num_heads_ready++;
2696 spin_unlock(&delayed_refs->lock);
2697 btrfs_delayed_ref_unlock(locked_ref);
2698 btrfs_put_delayed_ref(ref);
2699 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2705 * If this node is a head, that means all the refs in this head
2706 * have been dealt with, and we will pick the next head to deal
2707 * with, so we must unlock the head and drop it from the cluster
2708 * list before we release it.
2710 if (btrfs_delayed_ref_is_head(ref)) {
2711 if (locked_ref->is_data &&
2712 locked_ref->total_ref_mod < 0) {
2713 spin_lock(&delayed_refs->lock);
2714 delayed_refs->pending_csums -= ref->num_bytes;
2715 spin_unlock(&delayed_refs->lock);
2717 btrfs_delayed_ref_unlock(locked_ref);
2720 btrfs_put_delayed_ref(ref);
2726 * We don't want to include ref heads since we can have empty ref heads
2727 * and those will drastically skew our runtime down since we just do
2728 * accounting, no actual extent tree updates.
2730 if (actual_count > 0) {
2731 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2735 * We weigh the current average higher than our current runtime
2736 * to avoid large swings in the average.
2738 spin_lock(&delayed_refs->lock);
2739 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2740 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2741 spin_unlock(&delayed_refs->lock);
2746 #ifdef SCRAMBLE_DELAYED_REFS
2748 * Normally delayed refs get processed in ascending bytenr order. This
2749 * correlates in most cases to the order added. To expose dependencies on this
2750 * order, we start to process the tree in the middle instead of the beginning
2752 static u64 find_middle(struct rb_root *root)
2754 struct rb_node *n = root->rb_node;
2755 struct btrfs_delayed_ref_node *entry;
2758 u64 first = 0, last = 0;
2762 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2763 first = entry->bytenr;
2767 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2768 last = entry->bytenr;
2773 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2774 WARN_ON(!entry->in_tree);
2776 middle = entry->bytenr;
2789 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2793 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2794 sizeof(struct btrfs_extent_inline_ref));
2795 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2796 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2799 * We don't ever fill up leaves all the way so multiply by 2 just to be
2800 * closer to what we're really going to want to use.
2802 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2806 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2807 * would require to store the csums for that many bytes.
2809 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2812 u64 num_csums_per_leaf;
2815 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2816 num_csums_per_leaf = div64_u64(csum_size,
2817 (u64)btrfs_super_csum_size(fs_info->super_copy));
2818 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2819 num_csums += num_csums_per_leaf - 1;
2820 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2824 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2825 struct btrfs_fs_info *fs_info)
2827 struct btrfs_block_rsv *global_rsv;
2828 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2829 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2830 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2831 u64 num_bytes, num_dirty_bgs_bytes;
2834 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2835 num_heads = heads_to_leaves(fs_info, num_heads);
2837 num_bytes += (num_heads - 1) * fs_info->nodesize;
2839 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2841 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2843 global_rsv = &fs_info->global_block_rsv;
2846 * If we can't allocate any more chunks lets make sure we have _lots_ of
2847 * wiggle room since running delayed refs can create more delayed refs.
2849 if (global_rsv->space_info->full) {
2850 num_dirty_bgs_bytes <<= 1;
2854 spin_lock(&global_rsv->lock);
2855 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2857 spin_unlock(&global_rsv->lock);
2861 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2862 struct btrfs_fs_info *fs_info)
2865 atomic_read(&trans->transaction->delayed_refs.num_entries);
2870 avg_runtime = fs_info->avg_delayed_ref_runtime;
2871 val = num_entries * avg_runtime;
2872 if (val >= NSEC_PER_SEC)
2874 if (val >= NSEC_PER_SEC / 2)
2877 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2880 struct async_delayed_refs {
2881 struct btrfs_root *root;
2886 struct completion wait;
2887 struct btrfs_work work;
2890 static inline struct async_delayed_refs *
2891 to_async_delayed_refs(struct btrfs_work *work)
2893 return container_of(work, struct async_delayed_refs, work);
2896 static void delayed_ref_async_start(struct btrfs_work *work)
2898 struct async_delayed_refs *async = to_async_delayed_refs(work);
2899 struct btrfs_trans_handle *trans;
2900 struct btrfs_fs_info *fs_info = async->root->fs_info;
2903 /* if the commit is already started, we don't need to wait here */
2904 if (btrfs_transaction_blocked(fs_info))
2907 trans = btrfs_join_transaction(async->root);
2908 if (IS_ERR(trans)) {
2909 async->error = PTR_ERR(trans);
2914 * trans->sync means that when we call end_transaction, we won't
2915 * wait on delayed refs
2919 /* Don't bother flushing if we got into a different transaction */
2920 if (trans->transid > async->transid)
2923 ret = btrfs_run_delayed_refs(trans, fs_info, async->count);
2927 ret = btrfs_end_transaction(trans);
2928 if (ret && !async->error)
2932 complete(&async->wait);
2937 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2938 unsigned long count, u64 transid, int wait)
2940 struct async_delayed_refs *async;
2943 async = kmalloc(sizeof(*async), GFP_NOFS);
2947 async->root = fs_info->tree_root;
2948 async->count = count;
2950 async->transid = transid;
2955 init_completion(&async->wait);
2957 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2958 delayed_ref_async_start, NULL, NULL);
2960 btrfs_queue_work(fs_info->extent_workers, &async->work);
2963 wait_for_completion(&async->wait);
2972 * this starts processing the delayed reference count updates and
2973 * extent insertions we have queued up so far. count can be
2974 * 0, which means to process everything in the tree at the start
2975 * of the run (but not newly added entries), or it can be some target
2976 * number you'd like to process.
2978 * Returns 0 on success or if called with an aborted transaction
2979 * Returns <0 on error and aborts the transaction
2981 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2982 struct btrfs_fs_info *fs_info, unsigned long count)
2984 struct rb_node *node;
2985 struct btrfs_delayed_ref_root *delayed_refs;
2986 struct btrfs_delayed_ref_head *head;
2988 int run_all = count == (unsigned long)-1;
2989 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2991 /* We'll clean this up in btrfs_cleanup_transaction */
2995 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2998 delayed_refs = &trans->transaction->delayed_refs;
3000 count = atomic_read(&delayed_refs->num_entries) * 2;
3003 #ifdef SCRAMBLE_DELAYED_REFS
3004 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3006 trans->can_flush_pending_bgs = false;
3007 ret = __btrfs_run_delayed_refs(trans, fs_info, count);
3009 btrfs_abort_transaction(trans, ret);
3014 if (!list_empty(&trans->new_bgs))
3015 btrfs_create_pending_block_groups(trans, fs_info);
3017 spin_lock(&delayed_refs->lock);
3018 node = rb_first(&delayed_refs->href_root);
3020 spin_unlock(&delayed_refs->lock);
3025 head = rb_entry(node, struct btrfs_delayed_ref_head,
3027 if (btrfs_delayed_ref_is_head(&head->node)) {
3028 struct btrfs_delayed_ref_node *ref;
3031 refcount_inc(&ref->refs);
3033 spin_unlock(&delayed_refs->lock);
3035 * Mutex was contended, block until it's
3036 * released and try again
3038 mutex_lock(&head->mutex);
3039 mutex_unlock(&head->mutex);
3041 btrfs_put_delayed_ref(ref);
3047 node = rb_next(node);
3049 spin_unlock(&delayed_refs->lock);
3054 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3058 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3059 struct btrfs_fs_info *fs_info,
3060 u64 bytenr, u64 num_bytes, u64 flags,
3061 int level, int is_data)
3063 struct btrfs_delayed_extent_op *extent_op;
3066 extent_op = btrfs_alloc_delayed_extent_op();
3070 extent_op->flags_to_set = flags;
3071 extent_op->update_flags = true;
3072 extent_op->update_key = false;
3073 extent_op->is_data = is_data ? true : false;
3074 extent_op->level = level;
3076 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3077 num_bytes, extent_op);
3079 btrfs_free_delayed_extent_op(extent_op);
3083 static noinline int check_delayed_ref(struct btrfs_root *root,
3084 struct btrfs_path *path,
3085 u64 objectid, u64 offset, u64 bytenr)
3087 struct btrfs_delayed_ref_head *head;
3088 struct btrfs_delayed_ref_node *ref;
3089 struct btrfs_delayed_data_ref *data_ref;
3090 struct btrfs_delayed_ref_root *delayed_refs;
3091 struct btrfs_transaction *cur_trans;
3094 cur_trans = root->fs_info->running_transaction;
3098 delayed_refs = &cur_trans->delayed_refs;
3099 spin_lock(&delayed_refs->lock);
3100 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3102 spin_unlock(&delayed_refs->lock);
3106 if (!mutex_trylock(&head->mutex)) {
3107 refcount_inc(&head->node.refs);
3108 spin_unlock(&delayed_refs->lock);
3110 btrfs_release_path(path);
3113 * Mutex was contended, block until it's released and let
3116 mutex_lock(&head->mutex);
3117 mutex_unlock(&head->mutex);
3118 btrfs_put_delayed_ref(&head->node);
3121 spin_unlock(&delayed_refs->lock);
3123 spin_lock(&head->lock);
3124 list_for_each_entry(ref, &head->ref_list, list) {
3125 /* If it's a shared ref we know a cross reference exists */
3126 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3131 data_ref = btrfs_delayed_node_to_data_ref(ref);
3134 * If our ref doesn't match the one we're currently looking at
3135 * then we have a cross reference.
3137 if (data_ref->root != root->root_key.objectid ||
3138 data_ref->objectid != objectid ||
3139 data_ref->offset != offset) {
3144 spin_unlock(&head->lock);
3145 mutex_unlock(&head->mutex);
3149 static noinline int check_committed_ref(struct btrfs_root *root,
3150 struct btrfs_path *path,
3151 u64 objectid, u64 offset, u64 bytenr)
3153 struct btrfs_fs_info *fs_info = root->fs_info;
3154 struct btrfs_root *extent_root = fs_info->extent_root;
3155 struct extent_buffer *leaf;
3156 struct btrfs_extent_data_ref *ref;
3157 struct btrfs_extent_inline_ref *iref;
3158 struct btrfs_extent_item *ei;
3159 struct btrfs_key key;
3163 key.objectid = bytenr;
3164 key.offset = (u64)-1;
3165 key.type = BTRFS_EXTENT_ITEM_KEY;
3167 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3170 BUG_ON(ret == 0); /* Corruption */
3173 if (path->slots[0] == 0)
3177 leaf = path->nodes[0];
3178 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3180 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3184 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3185 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3186 if (item_size < sizeof(*ei)) {
3187 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3191 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3193 if (item_size != sizeof(*ei) +
3194 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3197 if (btrfs_extent_generation(leaf, ei) <=
3198 btrfs_root_last_snapshot(&root->root_item))
3201 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3202 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3203 BTRFS_EXTENT_DATA_REF_KEY)
3206 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3207 if (btrfs_extent_refs(leaf, ei) !=
3208 btrfs_extent_data_ref_count(leaf, ref) ||
3209 btrfs_extent_data_ref_root(leaf, ref) !=
3210 root->root_key.objectid ||
3211 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3212 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3220 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3223 struct btrfs_path *path;
3227 path = btrfs_alloc_path();
3232 ret = check_committed_ref(root, path, objectid,
3234 if (ret && ret != -ENOENT)
3237 ret2 = check_delayed_ref(root, path, objectid,
3239 } while (ret2 == -EAGAIN);
3241 if (ret2 && ret2 != -ENOENT) {
3246 if (ret != -ENOENT || ret2 != -ENOENT)
3249 btrfs_free_path(path);
3250 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3255 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3256 struct btrfs_root *root,
3257 struct extent_buffer *buf,
3258 int full_backref, int inc)
3260 struct btrfs_fs_info *fs_info = root->fs_info;
3266 struct btrfs_key key;
3267 struct btrfs_file_extent_item *fi;
3271 int (*process_func)(struct btrfs_trans_handle *,
3272 struct btrfs_fs_info *,
3273 u64, u64, u64, u64, u64, u64);
3276 if (btrfs_is_testing(fs_info))
3279 ref_root = btrfs_header_owner(buf);
3280 nritems = btrfs_header_nritems(buf);
3281 level = btrfs_header_level(buf);
3283 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3287 process_func = btrfs_inc_extent_ref;
3289 process_func = btrfs_free_extent;
3292 parent = buf->start;
3296 for (i = 0; i < nritems; i++) {
3298 btrfs_item_key_to_cpu(buf, &key, i);
3299 if (key.type != BTRFS_EXTENT_DATA_KEY)
3301 fi = btrfs_item_ptr(buf, i,
3302 struct btrfs_file_extent_item);
3303 if (btrfs_file_extent_type(buf, fi) ==
3304 BTRFS_FILE_EXTENT_INLINE)
3306 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3310 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3311 key.offset -= btrfs_file_extent_offset(buf, fi);
3312 ret = process_func(trans, fs_info, bytenr, num_bytes,
3313 parent, ref_root, key.objectid,
3318 bytenr = btrfs_node_blockptr(buf, i);
3319 num_bytes = fs_info->nodesize;
3320 ret = process_func(trans, fs_info, bytenr, num_bytes,
3321 parent, ref_root, level - 1, 0);
3331 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3332 struct extent_buffer *buf, int full_backref)
3334 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3337 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3338 struct extent_buffer *buf, int full_backref)
3340 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3343 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3344 struct btrfs_fs_info *fs_info,
3345 struct btrfs_path *path,
3346 struct btrfs_block_group_cache *cache)
3349 struct btrfs_root *extent_root = fs_info->extent_root;
3351 struct extent_buffer *leaf;
3353 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3360 leaf = path->nodes[0];
3361 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3362 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3363 btrfs_mark_buffer_dirty(leaf);
3365 btrfs_release_path(path);
3370 static struct btrfs_block_group_cache *
3371 next_block_group(struct btrfs_fs_info *fs_info,
3372 struct btrfs_block_group_cache *cache)
3374 struct rb_node *node;
3376 spin_lock(&fs_info->block_group_cache_lock);
3378 /* If our block group was removed, we need a full search. */
3379 if (RB_EMPTY_NODE(&cache->cache_node)) {
3380 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3382 spin_unlock(&fs_info->block_group_cache_lock);
3383 btrfs_put_block_group(cache);
3384 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3386 node = rb_next(&cache->cache_node);
3387 btrfs_put_block_group(cache);
3389 cache = rb_entry(node, struct btrfs_block_group_cache,
3391 btrfs_get_block_group(cache);
3394 spin_unlock(&fs_info->block_group_cache_lock);
3398 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3399 struct btrfs_trans_handle *trans,
3400 struct btrfs_path *path)
3402 struct btrfs_fs_info *fs_info = block_group->fs_info;
3403 struct btrfs_root *root = fs_info->tree_root;
3404 struct inode *inode = NULL;
3405 struct extent_changeset *data_reserved = NULL;
3407 int dcs = BTRFS_DC_ERROR;
3413 * If this block group is smaller than 100 megs don't bother caching the
3416 if (block_group->key.offset < (100 * SZ_1M)) {
3417 spin_lock(&block_group->lock);
3418 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3419 spin_unlock(&block_group->lock);
3426 inode = lookup_free_space_inode(fs_info, block_group, path);
3427 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3428 ret = PTR_ERR(inode);
3429 btrfs_release_path(path);
3433 if (IS_ERR(inode)) {
3437 if (block_group->ro)
3440 ret = create_free_space_inode(fs_info, trans, block_group,
3447 /* We've already setup this transaction, go ahead and exit */
3448 if (block_group->cache_generation == trans->transid &&
3449 i_size_read(inode)) {
3450 dcs = BTRFS_DC_SETUP;
3455 * We want to set the generation to 0, that way if anything goes wrong
3456 * from here on out we know not to trust this cache when we load up next
3459 BTRFS_I(inode)->generation = 0;
3460 ret = btrfs_update_inode(trans, root, inode);
3463 * So theoretically we could recover from this, simply set the
3464 * super cache generation to 0 so we know to invalidate the
3465 * cache, but then we'd have to keep track of the block groups
3466 * that fail this way so we know we _have_ to reset this cache
3467 * before the next commit or risk reading stale cache. So to
3468 * limit our exposure to horrible edge cases lets just abort the
3469 * transaction, this only happens in really bad situations
3472 btrfs_abort_transaction(trans, ret);
3477 if (i_size_read(inode) > 0) {
3478 ret = btrfs_check_trunc_cache_free_space(fs_info,
3479 &fs_info->global_block_rsv);
3483 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3488 spin_lock(&block_group->lock);
3489 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3490 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3492 * don't bother trying to write stuff out _if_
3493 * a) we're not cached,
3494 * b) we're with nospace_cache mount option,
3495 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3497 dcs = BTRFS_DC_WRITTEN;
3498 spin_unlock(&block_group->lock);
3501 spin_unlock(&block_group->lock);
3504 * We hit an ENOSPC when setting up the cache in this transaction, just
3505 * skip doing the setup, we've already cleared the cache so we're safe.
3507 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3513 * Try to preallocate enough space based on how big the block group is.
3514 * Keep in mind this has to include any pinned space which could end up
3515 * taking up quite a bit since it's not folded into the other space
3518 num_pages = div_u64(block_group->key.offset, SZ_256M);
3523 num_pages *= PAGE_SIZE;
3525 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3529 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3530 num_pages, num_pages,
3533 * Our cache requires contiguous chunks so that we don't modify a bunch
3534 * of metadata or split extents when writing the cache out, which means
3535 * we can enospc if we are heavily fragmented in addition to just normal
3536 * out of space conditions. So if we hit this just skip setting up any
3537 * other block groups for this transaction, maybe we'll unpin enough
3538 * space the next time around.
3541 dcs = BTRFS_DC_SETUP;
3542 else if (ret == -ENOSPC)
3543 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3548 btrfs_release_path(path);
3550 spin_lock(&block_group->lock);
3551 if (!ret && dcs == BTRFS_DC_SETUP)
3552 block_group->cache_generation = trans->transid;
3553 block_group->disk_cache_state = dcs;
3554 spin_unlock(&block_group->lock);
3556 extent_changeset_free(data_reserved);
3560 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3561 struct btrfs_fs_info *fs_info)
3563 struct btrfs_block_group_cache *cache, *tmp;
3564 struct btrfs_transaction *cur_trans = trans->transaction;
3565 struct btrfs_path *path;
3567 if (list_empty(&cur_trans->dirty_bgs) ||
3568 !btrfs_test_opt(fs_info, SPACE_CACHE))
3571 path = btrfs_alloc_path();
3575 /* Could add new block groups, use _safe just in case */
3576 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3578 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3579 cache_save_setup(cache, trans, path);
3582 btrfs_free_path(path);
3587 * transaction commit does final block group cache writeback during a
3588 * critical section where nothing is allowed to change the FS. This is
3589 * required in order for the cache to actually match the block group,
3590 * but can introduce a lot of latency into the commit.
3592 * So, btrfs_start_dirty_block_groups is here to kick off block group
3593 * cache IO. There's a chance we'll have to redo some of it if the
3594 * block group changes again during the commit, but it greatly reduces
3595 * the commit latency by getting rid of the easy block groups while
3596 * we're still allowing others to join the commit.
3598 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3599 struct btrfs_fs_info *fs_info)
3601 struct btrfs_block_group_cache *cache;
3602 struct btrfs_transaction *cur_trans = trans->transaction;
3605 struct btrfs_path *path = NULL;
3607 struct list_head *io = &cur_trans->io_bgs;
3608 int num_started = 0;
3611 spin_lock(&cur_trans->dirty_bgs_lock);
3612 if (list_empty(&cur_trans->dirty_bgs)) {
3613 spin_unlock(&cur_trans->dirty_bgs_lock);
3616 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3617 spin_unlock(&cur_trans->dirty_bgs_lock);
3621 * make sure all the block groups on our dirty list actually
3624 btrfs_create_pending_block_groups(trans, fs_info);
3627 path = btrfs_alloc_path();
3633 * cache_write_mutex is here only to save us from balance or automatic
3634 * removal of empty block groups deleting this block group while we are
3635 * writing out the cache
3637 mutex_lock(&trans->transaction->cache_write_mutex);
3638 while (!list_empty(&dirty)) {
3639 cache = list_first_entry(&dirty,
3640 struct btrfs_block_group_cache,
3643 * this can happen if something re-dirties a block
3644 * group that is already under IO. Just wait for it to
3645 * finish and then do it all again
3647 if (!list_empty(&cache->io_list)) {
3648 list_del_init(&cache->io_list);
3649 btrfs_wait_cache_io(trans, cache, path);
3650 btrfs_put_block_group(cache);
3655 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3656 * if it should update the cache_state. Don't delete
3657 * until after we wait.
3659 * Since we're not running in the commit critical section
3660 * we need the dirty_bgs_lock to protect from update_block_group
3662 spin_lock(&cur_trans->dirty_bgs_lock);
3663 list_del_init(&cache->dirty_list);
3664 spin_unlock(&cur_trans->dirty_bgs_lock);
3668 cache_save_setup(cache, trans, path);
3670 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3671 cache->io_ctl.inode = NULL;
3672 ret = btrfs_write_out_cache(fs_info, trans,
3674 if (ret == 0 && cache->io_ctl.inode) {
3679 * the cache_write_mutex is protecting
3682 list_add_tail(&cache->io_list, io);
3685 * if we failed to write the cache, the
3686 * generation will be bad and life goes on
3692 ret = write_one_cache_group(trans, fs_info,
3695 * Our block group might still be attached to the list
3696 * of new block groups in the transaction handle of some
3697 * other task (struct btrfs_trans_handle->new_bgs). This
3698 * means its block group item isn't yet in the extent
3699 * tree. If this happens ignore the error, as we will
3700 * try again later in the critical section of the
3701 * transaction commit.
3703 if (ret == -ENOENT) {
3705 spin_lock(&cur_trans->dirty_bgs_lock);
3706 if (list_empty(&cache->dirty_list)) {
3707 list_add_tail(&cache->dirty_list,
3708 &cur_trans->dirty_bgs);
3709 btrfs_get_block_group(cache);
3711 spin_unlock(&cur_trans->dirty_bgs_lock);
3713 btrfs_abort_transaction(trans, ret);
3717 /* if its not on the io list, we need to put the block group */
3719 btrfs_put_block_group(cache);
3725 * Avoid blocking other tasks for too long. It might even save
3726 * us from writing caches for block groups that are going to be
3729 mutex_unlock(&trans->transaction->cache_write_mutex);
3730 mutex_lock(&trans->transaction->cache_write_mutex);
3732 mutex_unlock(&trans->transaction->cache_write_mutex);
3735 * go through delayed refs for all the stuff we've just kicked off
3736 * and then loop back (just once)
3738 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
3739 if (!ret && loops == 0) {
3741 spin_lock(&cur_trans->dirty_bgs_lock);
3742 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3744 * dirty_bgs_lock protects us from concurrent block group
3745 * deletes too (not just cache_write_mutex).
3747 if (!list_empty(&dirty)) {
3748 spin_unlock(&cur_trans->dirty_bgs_lock);
3751 spin_unlock(&cur_trans->dirty_bgs_lock);
3752 } else if (ret < 0) {
3753 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3756 btrfs_free_path(path);
3760 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3761 struct btrfs_fs_info *fs_info)
3763 struct btrfs_block_group_cache *cache;
3764 struct btrfs_transaction *cur_trans = trans->transaction;
3767 struct btrfs_path *path;
3768 struct list_head *io = &cur_trans->io_bgs;
3769 int num_started = 0;
3771 path = btrfs_alloc_path();
3776 * Even though we are in the critical section of the transaction commit,
3777 * we can still have concurrent tasks adding elements to this
3778 * transaction's list of dirty block groups. These tasks correspond to
3779 * endio free space workers started when writeback finishes for a
3780 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3781 * allocate new block groups as a result of COWing nodes of the root
3782 * tree when updating the free space inode. The writeback for the space
3783 * caches is triggered by an earlier call to
3784 * btrfs_start_dirty_block_groups() and iterations of the following
3786 * Also we want to do the cache_save_setup first and then run the
3787 * delayed refs to make sure we have the best chance at doing this all
3790 spin_lock(&cur_trans->dirty_bgs_lock);
3791 while (!list_empty(&cur_trans->dirty_bgs)) {
3792 cache = list_first_entry(&cur_trans->dirty_bgs,
3793 struct btrfs_block_group_cache,
3797 * this can happen if cache_save_setup re-dirties a block
3798 * group that is already under IO. Just wait for it to
3799 * finish and then do it all again
3801 if (!list_empty(&cache->io_list)) {
3802 spin_unlock(&cur_trans->dirty_bgs_lock);
3803 list_del_init(&cache->io_list);
3804 btrfs_wait_cache_io(trans, cache, path);
3805 btrfs_put_block_group(cache);
3806 spin_lock(&cur_trans->dirty_bgs_lock);
3810 * don't remove from the dirty list until after we've waited
3813 list_del_init(&cache->dirty_list);
3814 spin_unlock(&cur_trans->dirty_bgs_lock);
3817 cache_save_setup(cache, trans, path);
3820 ret = btrfs_run_delayed_refs(trans, fs_info,
3821 (unsigned long) -1);
3823 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3824 cache->io_ctl.inode = NULL;
3825 ret = btrfs_write_out_cache(fs_info, trans,
3827 if (ret == 0 && cache->io_ctl.inode) {
3830 list_add_tail(&cache->io_list, io);
3833 * if we failed to write the cache, the
3834 * generation will be bad and life goes on
3840 ret = write_one_cache_group(trans, fs_info,
3843 * One of the free space endio workers might have
3844 * created a new block group while updating a free space
3845 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3846 * and hasn't released its transaction handle yet, in
3847 * which case the new block group is still attached to
3848 * its transaction handle and its creation has not
3849 * finished yet (no block group item in the extent tree
3850 * yet, etc). If this is the case, wait for all free
3851 * space endio workers to finish and retry. This is a
3852 * a very rare case so no need for a more efficient and
3855 if (ret == -ENOENT) {
3856 wait_event(cur_trans->writer_wait,
3857 atomic_read(&cur_trans->num_writers) == 1);
3858 ret = write_one_cache_group(trans, fs_info,
3862 btrfs_abort_transaction(trans, ret);
3865 /* if its not on the io list, we need to put the block group */
3867 btrfs_put_block_group(cache);
3868 spin_lock(&cur_trans->dirty_bgs_lock);
3870 spin_unlock(&cur_trans->dirty_bgs_lock);
3872 while (!list_empty(io)) {
3873 cache = list_first_entry(io, struct btrfs_block_group_cache,
3875 list_del_init(&cache->io_list);
3876 btrfs_wait_cache_io(trans, cache, path);
3877 btrfs_put_block_group(cache);
3880 btrfs_free_path(path);
3884 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3886 struct btrfs_block_group_cache *block_group;
3889 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3890 if (!block_group || block_group->ro)
3893 btrfs_put_block_group(block_group);
3897 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3899 struct btrfs_block_group_cache *bg;
3902 bg = btrfs_lookup_block_group(fs_info, bytenr);
3906 spin_lock(&bg->lock);
3910 atomic_inc(&bg->nocow_writers);
3911 spin_unlock(&bg->lock);
3913 /* no put on block group, done by btrfs_dec_nocow_writers */
3915 btrfs_put_block_group(bg);
3921 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3923 struct btrfs_block_group_cache *bg;
3925 bg = btrfs_lookup_block_group(fs_info, bytenr);
3927 if (atomic_dec_and_test(&bg->nocow_writers))
3928 wake_up_atomic_t(&bg->nocow_writers);
3930 * Once for our lookup and once for the lookup done by a previous call
3931 * to btrfs_inc_nocow_writers()
3933 btrfs_put_block_group(bg);
3934 btrfs_put_block_group(bg);
3937 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3943 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3945 wait_on_atomic_t(&bg->nocow_writers,
3946 btrfs_wait_nocow_writers_atomic_t,
3947 TASK_UNINTERRUPTIBLE);
3950 static const char *alloc_name(u64 flags)
3953 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3955 case BTRFS_BLOCK_GROUP_METADATA:
3957 case BTRFS_BLOCK_GROUP_DATA:
3959 case BTRFS_BLOCK_GROUP_SYSTEM:
3963 return "invalid-combination";
3967 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
3968 struct btrfs_space_info **new)
3971 struct btrfs_space_info *space_info;
3975 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
3979 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
3986 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3987 INIT_LIST_HEAD(&space_info->block_groups[i]);
3988 init_rwsem(&space_info->groups_sem);
3989 spin_lock_init(&space_info->lock);
3990 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3991 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
3992 init_waitqueue_head(&space_info->wait);
3993 INIT_LIST_HEAD(&space_info->ro_bgs);
3994 INIT_LIST_HEAD(&space_info->tickets);
3995 INIT_LIST_HEAD(&space_info->priority_tickets);
3997 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
3998 info->space_info_kobj, "%s",
3999 alloc_name(space_info->flags));
4001 percpu_counter_destroy(&space_info->total_bytes_pinned);
4007 list_add_rcu(&space_info->list, &info->space_info);
4008 if (flags & BTRFS_BLOCK_GROUP_DATA)
4009 info->data_sinfo = space_info;
4014 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4015 u64 total_bytes, u64 bytes_used,
4017 struct btrfs_space_info **space_info)
4019 struct btrfs_space_info *found;
4022 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4023 BTRFS_BLOCK_GROUP_RAID10))
4028 found = __find_space_info(info, flags);
4030 spin_lock(&found->lock);
4031 found->total_bytes += total_bytes;
4032 found->disk_total += total_bytes * factor;
4033 found->bytes_used += bytes_used;
4034 found->disk_used += bytes_used * factor;
4035 found->bytes_readonly += bytes_readonly;
4036 if (total_bytes > 0)
4038 space_info_add_new_bytes(info, found, total_bytes -
4039 bytes_used - bytes_readonly);
4040 spin_unlock(&found->lock);
4041 *space_info = found;
4044 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4046 u64 extra_flags = chunk_to_extended(flags) &
4047 BTRFS_EXTENDED_PROFILE_MASK;
4049 write_seqlock(&fs_info->profiles_lock);
4050 if (flags & BTRFS_BLOCK_GROUP_DATA)
4051 fs_info->avail_data_alloc_bits |= extra_flags;
4052 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4053 fs_info->avail_metadata_alloc_bits |= extra_flags;
4054 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4055 fs_info->avail_system_alloc_bits |= extra_flags;
4056 write_sequnlock(&fs_info->profiles_lock);
4060 * returns target flags in extended format or 0 if restripe for this
4061 * chunk_type is not in progress
4063 * should be called with either volume_mutex or balance_lock held
4065 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4067 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4073 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4074 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4075 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4076 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4077 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4078 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4079 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4080 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4081 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4088 * @flags: available profiles in extended format (see ctree.h)
4090 * Returns reduced profile in chunk format. If profile changing is in
4091 * progress (either running or paused) picks the target profile (if it's
4092 * already available), otherwise falls back to plain reducing.
4094 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4096 u64 num_devices = fs_info->fs_devices->rw_devices;
4102 * see if restripe for this chunk_type is in progress, if so
4103 * try to reduce to the target profile
4105 spin_lock(&fs_info->balance_lock);
4106 target = get_restripe_target(fs_info, flags);
4108 /* pick target profile only if it's already available */
4109 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4110 spin_unlock(&fs_info->balance_lock);
4111 return extended_to_chunk(target);
4114 spin_unlock(&fs_info->balance_lock);
4116 /* First, mask out the RAID levels which aren't possible */
4117 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4118 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4119 allowed |= btrfs_raid_group[raid_type];
4123 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4124 allowed = BTRFS_BLOCK_GROUP_RAID6;
4125 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4126 allowed = BTRFS_BLOCK_GROUP_RAID5;
4127 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4128 allowed = BTRFS_BLOCK_GROUP_RAID10;
4129 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4130 allowed = BTRFS_BLOCK_GROUP_RAID1;
4131 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4132 allowed = BTRFS_BLOCK_GROUP_RAID0;
4134 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4136 return extended_to_chunk(flags | allowed);
4139 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4146 seq = read_seqbegin(&fs_info->profiles_lock);
4148 if (flags & BTRFS_BLOCK_GROUP_DATA)
4149 flags |= fs_info->avail_data_alloc_bits;
4150 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4151 flags |= fs_info->avail_system_alloc_bits;
4152 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4153 flags |= fs_info->avail_metadata_alloc_bits;
4154 } while (read_seqretry(&fs_info->profiles_lock, seq));
4156 return btrfs_reduce_alloc_profile(fs_info, flags);
4159 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4161 struct btrfs_fs_info *fs_info = root->fs_info;
4166 flags = BTRFS_BLOCK_GROUP_DATA;
4167 else if (root == fs_info->chunk_root)
4168 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4170 flags = BTRFS_BLOCK_GROUP_METADATA;
4172 ret = get_alloc_profile(fs_info, flags);
4176 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4178 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4181 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4183 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4186 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4188 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4191 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4192 bool may_use_included)
4195 return s_info->bytes_used + s_info->bytes_reserved +
4196 s_info->bytes_pinned + s_info->bytes_readonly +
4197 (may_use_included ? s_info->bytes_may_use : 0);
4200 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4202 struct btrfs_root *root = inode->root;
4203 struct btrfs_fs_info *fs_info = root->fs_info;
4204 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4207 int need_commit = 2;
4208 int have_pinned_space;
4210 /* make sure bytes are sectorsize aligned */
4211 bytes = ALIGN(bytes, fs_info->sectorsize);
4213 if (btrfs_is_free_space_inode(inode)) {
4215 ASSERT(current->journal_info);
4219 /* make sure we have enough space to handle the data first */
4220 spin_lock(&data_sinfo->lock);
4221 used = btrfs_space_info_used(data_sinfo, true);
4223 if (used + bytes > data_sinfo->total_bytes) {
4224 struct btrfs_trans_handle *trans;
4227 * if we don't have enough free bytes in this space then we need
4228 * to alloc a new chunk.
4230 if (!data_sinfo->full) {
4233 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4234 spin_unlock(&data_sinfo->lock);
4236 alloc_target = btrfs_data_alloc_profile(fs_info);
4238 * It is ugly that we don't call nolock join
4239 * transaction for the free space inode case here.
4240 * But it is safe because we only do the data space
4241 * reservation for the free space cache in the
4242 * transaction context, the common join transaction
4243 * just increase the counter of the current transaction
4244 * handler, doesn't try to acquire the trans_lock of
4247 trans = btrfs_join_transaction(root);
4249 return PTR_ERR(trans);
4251 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4252 CHUNK_ALLOC_NO_FORCE);
4253 btrfs_end_transaction(trans);
4258 have_pinned_space = 1;
4267 * If we don't have enough pinned space to deal with this
4268 * allocation, and no removed chunk in current transaction,
4269 * don't bother committing the transaction.
4271 have_pinned_space = percpu_counter_compare(
4272 &data_sinfo->total_bytes_pinned,
4273 used + bytes - data_sinfo->total_bytes);
4274 spin_unlock(&data_sinfo->lock);
4276 /* commit the current transaction and try again */
4279 !atomic_read(&fs_info->open_ioctl_trans)) {
4282 if (need_commit > 0) {
4283 btrfs_start_delalloc_roots(fs_info, 0, -1);
4284 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4288 trans = btrfs_join_transaction(root);
4290 return PTR_ERR(trans);
4291 if (have_pinned_space >= 0 ||
4292 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4293 &trans->transaction->flags) ||
4295 ret = btrfs_commit_transaction(trans);
4299 * The cleaner kthread might still be doing iput
4300 * operations. Wait for it to finish so that
4301 * more space is released.
4303 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4304 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4307 btrfs_end_transaction(trans);
4311 trace_btrfs_space_reservation(fs_info,
4312 "space_info:enospc",
4313 data_sinfo->flags, bytes, 1);
4316 data_sinfo->bytes_may_use += bytes;
4317 trace_btrfs_space_reservation(fs_info, "space_info",
4318 data_sinfo->flags, bytes, 1);
4319 spin_unlock(&data_sinfo->lock);
4324 int btrfs_check_data_free_space(struct inode *inode,
4325 struct extent_changeset **reserved, u64 start, u64 len)
4327 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4330 /* align the range */
4331 len = round_up(start + len, fs_info->sectorsize) -
4332 round_down(start, fs_info->sectorsize);
4333 start = round_down(start, fs_info->sectorsize);
4335 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4339 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4340 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4342 btrfs_free_reserved_data_space_noquota(inode, start, len);
4349 * Called if we need to clear a data reservation for this inode
4350 * Normally in a error case.
4352 * This one will *NOT* use accurate qgroup reserved space API, just for case
4353 * which we can't sleep and is sure it won't affect qgroup reserved space.
4354 * Like clear_bit_hook().
4356 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4359 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4360 struct btrfs_space_info *data_sinfo;
4362 /* Make sure the range is aligned to sectorsize */
4363 len = round_up(start + len, fs_info->sectorsize) -
4364 round_down(start, fs_info->sectorsize);
4365 start = round_down(start, fs_info->sectorsize);
4367 data_sinfo = fs_info->data_sinfo;
4368 spin_lock(&data_sinfo->lock);
4369 if (WARN_ON(data_sinfo->bytes_may_use < len))
4370 data_sinfo->bytes_may_use = 0;
4372 data_sinfo->bytes_may_use -= len;
4373 trace_btrfs_space_reservation(fs_info, "space_info",
4374 data_sinfo->flags, len, 0);
4375 spin_unlock(&data_sinfo->lock);
4379 * Called if we need to clear a data reservation for this inode
4380 * Normally in a error case.
4382 * This one will handle the per-inode data rsv map for accurate reserved
4385 void btrfs_free_reserved_data_space(struct inode *inode,
4386 struct extent_changeset *reserved, u64 start, u64 len)
4388 struct btrfs_root *root = BTRFS_I(inode)->root;
4390 /* Make sure the range is aligned to sectorsize */
4391 len = round_up(start + len, root->fs_info->sectorsize) -
4392 round_down(start, root->fs_info->sectorsize);
4393 start = round_down(start, root->fs_info->sectorsize);
4395 btrfs_free_reserved_data_space_noquota(inode, start, len);
4396 btrfs_qgroup_free_data(inode, reserved, start, len);
4399 static void force_metadata_allocation(struct btrfs_fs_info *info)
4401 struct list_head *head = &info->space_info;
4402 struct btrfs_space_info *found;
4405 list_for_each_entry_rcu(found, head, list) {
4406 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4407 found->force_alloc = CHUNK_ALLOC_FORCE;
4412 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4414 return (global->size << 1);
4417 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4418 struct btrfs_space_info *sinfo, int force)
4420 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4421 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4424 if (force == CHUNK_ALLOC_FORCE)
4428 * We need to take into account the global rsv because for all intents
4429 * and purposes it's used space. Don't worry about locking the
4430 * global_rsv, it doesn't change except when the transaction commits.
4432 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4433 bytes_used += calc_global_rsv_need_space(global_rsv);
4436 * in limited mode, we want to have some free space up to
4437 * about 1% of the FS size.
4439 if (force == CHUNK_ALLOC_LIMITED) {
4440 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4441 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4443 if (sinfo->total_bytes - bytes_used < thresh)
4447 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4452 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4456 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4457 BTRFS_BLOCK_GROUP_RAID0 |
4458 BTRFS_BLOCK_GROUP_RAID5 |
4459 BTRFS_BLOCK_GROUP_RAID6))
4460 num_dev = fs_info->fs_devices->rw_devices;
4461 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4464 num_dev = 1; /* DUP or single */
4470 * If @is_allocation is true, reserve space in the system space info necessary
4471 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4474 void check_system_chunk(struct btrfs_trans_handle *trans,
4475 struct btrfs_fs_info *fs_info, u64 type)
4477 struct btrfs_space_info *info;
4484 * Needed because we can end up allocating a system chunk and for an
4485 * atomic and race free space reservation in the chunk block reserve.
4487 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4489 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4490 spin_lock(&info->lock);
4491 left = info->total_bytes - btrfs_space_info_used(info, true);
4492 spin_unlock(&info->lock);
4494 num_devs = get_profile_num_devs(fs_info, type);
4496 /* num_devs device items to update and 1 chunk item to add or remove */
4497 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4498 btrfs_calc_trans_metadata_size(fs_info, 1);
4500 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4501 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4502 left, thresh, type);
4503 dump_space_info(fs_info, info, 0, 0);
4506 if (left < thresh) {
4507 u64 flags = btrfs_system_alloc_profile(fs_info);
4510 * Ignore failure to create system chunk. We might end up not
4511 * needing it, as we might not need to COW all nodes/leafs from
4512 * the paths we visit in the chunk tree (they were already COWed
4513 * or created in the current transaction for example).
4515 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4519 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4520 &fs_info->chunk_block_rsv,
4521 thresh, BTRFS_RESERVE_NO_FLUSH);
4523 trans->chunk_bytes_reserved += thresh;
4528 * If force is CHUNK_ALLOC_FORCE:
4529 * - return 1 if it successfully allocates a chunk,
4530 * - return errors including -ENOSPC otherwise.
4531 * If force is NOT CHUNK_ALLOC_FORCE:
4532 * - return 0 if it doesn't need to allocate a new chunk,
4533 * - return 1 if it successfully allocates a chunk,
4534 * - return errors including -ENOSPC otherwise.
4536 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4537 struct btrfs_fs_info *fs_info, u64 flags, int force)
4539 struct btrfs_space_info *space_info;
4540 int wait_for_alloc = 0;
4543 /* Don't re-enter if we're already allocating a chunk */
4544 if (trans->allocating_chunk)
4547 space_info = __find_space_info(fs_info, flags);
4549 ret = create_space_info(fs_info, flags, &space_info);
4555 spin_lock(&space_info->lock);
4556 if (force < space_info->force_alloc)
4557 force = space_info->force_alloc;
4558 if (space_info->full) {
4559 if (should_alloc_chunk(fs_info, space_info, force))
4563 spin_unlock(&space_info->lock);
4567 if (!should_alloc_chunk(fs_info, space_info, force)) {
4568 spin_unlock(&space_info->lock);
4570 } else if (space_info->chunk_alloc) {
4573 space_info->chunk_alloc = 1;
4576 spin_unlock(&space_info->lock);
4578 mutex_lock(&fs_info->chunk_mutex);
4581 * The chunk_mutex is held throughout the entirety of a chunk
4582 * allocation, so once we've acquired the chunk_mutex we know that the
4583 * other guy is done and we need to recheck and see if we should
4586 if (wait_for_alloc) {
4587 mutex_unlock(&fs_info->chunk_mutex);
4592 trans->allocating_chunk = true;
4595 * If we have mixed data/metadata chunks we want to make sure we keep
4596 * allocating mixed chunks instead of individual chunks.
4598 if (btrfs_mixed_space_info(space_info))
4599 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4602 * if we're doing a data chunk, go ahead and make sure that
4603 * we keep a reasonable number of metadata chunks allocated in the
4606 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4607 fs_info->data_chunk_allocations++;
4608 if (!(fs_info->data_chunk_allocations %
4609 fs_info->metadata_ratio))
4610 force_metadata_allocation(fs_info);
4614 * Check if we have enough space in SYSTEM chunk because we may need
4615 * to update devices.
4617 check_system_chunk(trans, fs_info, flags);
4619 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4620 trans->allocating_chunk = false;
4622 spin_lock(&space_info->lock);
4623 if (ret < 0 && ret != -ENOSPC)
4626 space_info->full = 1;
4630 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4632 space_info->chunk_alloc = 0;
4633 spin_unlock(&space_info->lock);
4634 mutex_unlock(&fs_info->chunk_mutex);
4636 * When we allocate a new chunk we reserve space in the chunk block
4637 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4638 * add new nodes/leafs to it if we end up needing to do it when
4639 * inserting the chunk item and updating device items as part of the
4640 * second phase of chunk allocation, performed by
4641 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4642 * large number of new block groups to create in our transaction
4643 * handle's new_bgs list to avoid exhausting the chunk block reserve
4644 * in extreme cases - like having a single transaction create many new
4645 * block groups when starting to write out the free space caches of all
4646 * the block groups that were made dirty during the lifetime of the
4649 if (trans->can_flush_pending_bgs &&
4650 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4651 btrfs_create_pending_block_groups(trans, fs_info);
4652 btrfs_trans_release_chunk_metadata(trans);
4657 static int can_overcommit(struct btrfs_fs_info *fs_info,
4658 struct btrfs_space_info *space_info, u64 bytes,
4659 enum btrfs_reserve_flush_enum flush,
4662 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4668 /* Don't overcommit when in mixed mode. */
4669 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4673 profile = btrfs_system_alloc_profile(fs_info);
4675 profile = btrfs_metadata_alloc_profile(fs_info);
4677 used = btrfs_space_info_used(space_info, false);
4680 * We only want to allow over committing if we have lots of actual space
4681 * free, but if we don't have enough space to handle the global reserve
4682 * space then we could end up having a real enospc problem when trying
4683 * to allocate a chunk or some other such important allocation.
4685 spin_lock(&global_rsv->lock);
4686 space_size = calc_global_rsv_need_space(global_rsv);
4687 spin_unlock(&global_rsv->lock);
4688 if (used + space_size >= space_info->total_bytes)
4691 used += space_info->bytes_may_use;
4693 avail = atomic64_read(&fs_info->free_chunk_space);
4696 * If we have dup, raid1 or raid10 then only half of the free
4697 * space is actually useable. For raid56, the space info used
4698 * doesn't include the parity drive, so we don't have to
4701 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4702 BTRFS_BLOCK_GROUP_RAID1 |
4703 BTRFS_BLOCK_GROUP_RAID10))
4707 * If we aren't flushing all things, let us overcommit up to
4708 * 1/2th of the space. If we can flush, don't let us overcommit
4709 * too much, let it overcommit up to 1/8 of the space.
4711 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4716 if (used + bytes < space_info->total_bytes + avail)
4721 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4722 unsigned long nr_pages, int nr_items)
4724 struct super_block *sb = fs_info->sb;
4726 if (down_read_trylock(&sb->s_umount)) {
4727 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4728 up_read(&sb->s_umount);
4731 * We needn't worry the filesystem going from r/w to r/o though
4732 * we don't acquire ->s_umount mutex, because the filesystem
4733 * should guarantee the delalloc inodes list be empty after
4734 * the filesystem is readonly(all dirty pages are written to
4737 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4738 if (!current->journal_info)
4739 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4743 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4749 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4750 nr = div64_u64(to_reclaim, bytes);
4756 #define EXTENT_SIZE_PER_ITEM SZ_256K
4759 * shrink metadata reservation for delalloc
4761 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4762 u64 orig, bool wait_ordered)
4764 struct btrfs_block_rsv *block_rsv;
4765 struct btrfs_space_info *space_info;
4766 struct btrfs_trans_handle *trans;
4771 unsigned long nr_pages;
4773 enum btrfs_reserve_flush_enum flush;
4775 /* Calc the number of the pages we need flush for space reservation */
4776 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4777 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4779 trans = (struct btrfs_trans_handle *)current->journal_info;
4780 block_rsv = &fs_info->delalloc_block_rsv;
4781 space_info = block_rsv->space_info;
4783 delalloc_bytes = percpu_counter_sum_positive(
4784 &fs_info->delalloc_bytes);
4785 if (delalloc_bytes == 0) {
4789 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4794 while (delalloc_bytes && loops < 3) {
4795 max_reclaim = min(delalloc_bytes, to_reclaim);
4796 nr_pages = max_reclaim >> PAGE_SHIFT;
4797 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4799 * We need to wait for the async pages to actually start before
4802 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4806 if (max_reclaim <= nr_pages)
4809 max_reclaim -= nr_pages;
4811 wait_event(fs_info->async_submit_wait,
4812 atomic_read(&fs_info->async_delalloc_pages) <=
4816 flush = BTRFS_RESERVE_FLUSH_ALL;
4818 flush = BTRFS_RESERVE_NO_FLUSH;
4819 spin_lock(&space_info->lock);
4820 if (list_empty(&space_info->tickets) &&
4821 list_empty(&space_info->priority_tickets)) {
4822 spin_unlock(&space_info->lock);
4825 spin_unlock(&space_info->lock);
4828 if (wait_ordered && !trans) {
4829 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4831 time_left = schedule_timeout_killable(1);
4835 delalloc_bytes = percpu_counter_sum_positive(
4836 &fs_info->delalloc_bytes);
4841 * maybe_commit_transaction - possibly commit the transaction if its ok to
4842 * @root - the root we're allocating for
4843 * @bytes - the number of bytes we want to reserve
4844 * @force - force the commit
4846 * This will check to make sure that committing the transaction will actually
4847 * get us somewhere and then commit the transaction if it does. Otherwise it
4848 * will return -ENOSPC.
4850 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4851 struct btrfs_space_info *space_info,
4852 u64 bytes, int force)
4854 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4855 struct btrfs_trans_handle *trans;
4857 trans = (struct btrfs_trans_handle *)current->journal_info;
4864 /* See if there is enough pinned space to make this reservation */
4865 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4870 * See if there is some space in the delayed insertion reservation for
4873 if (space_info != delayed_rsv->space_info)
4876 spin_lock(&delayed_rsv->lock);
4877 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4878 bytes - delayed_rsv->size) < 0) {
4879 spin_unlock(&delayed_rsv->lock);
4882 spin_unlock(&delayed_rsv->lock);
4885 trans = btrfs_join_transaction(fs_info->extent_root);
4889 return btrfs_commit_transaction(trans);
4892 struct reserve_ticket {
4895 struct list_head list;
4896 wait_queue_head_t wait;
4900 * Try to flush some data based on policy set by @state. This is only advisory
4901 * and may fail for various reasons. The caller is supposed to examine the
4902 * state of @space_info to detect the outcome.
4904 static void flush_space(struct btrfs_fs_info *fs_info,
4905 struct btrfs_space_info *space_info, u64 num_bytes,
4908 struct btrfs_root *root = fs_info->extent_root;
4909 struct btrfs_trans_handle *trans;
4914 case FLUSH_DELAYED_ITEMS_NR:
4915 case FLUSH_DELAYED_ITEMS:
4916 if (state == FLUSH_DELAYED_ITEMS_NR)
4917 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4921 trans = btrfs_join_transaction(root);
4922 if (IS_ERR(trans)) {
4923 ret = PTR_ERR(trans);
4926 ret = btrfs_run_delayed_items_nr(trans, fs_info, nr);
4927 btrfs_end_transaction(trans);
4929 case FLUSH_DELALLOC:
4930 case FLUSH_DELALLOC_WAIT:
4931 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4932 state == FLUSH_DELALLOC_WAIT);
4935 trans = btrfs_join_transaction(root);
4936 if (IS_ERR(trans)) {
4937 ret = PTR_ERR(trans);
4940 ret = do_chunk_alloc(trans, fs_info,
4941 btrfs_metadata_alloc_profile(fs_info),
4942 CHUNK_ALLOC_NO_FORCE);
4943 btrfs_end_transaction(trans);
4944 if (ret > 0 || ret == -ENOSPC)
4948 ret = may_commit_transaction(fs_info, space_info,
4956 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
4962 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
4963 struct btrfs_space_info *space_info,
4966 struct reserve_ticket *ticket;
4971 list_for_each_entry(ticket, &space_info->tickets, list)
4972 to_reclaim += ticket->bytes;
4973 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4974 to_reclaim += ticket->bytes;
4978 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4979 if (can_overcommit(fs_info, space_info, to_reclaim,
4980 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4983 used = btrfs_space_info_used(space_info, true);
4985 if (can_overcommit(fs_info, space_info, SZ_1M,
4986 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
4987 expected = div_factor_fine(space_info->total_bytes, 95);
4989 expected = div_factor_fine(space_info->total_bytes, 90);
4991 if (used > expected)
4992 to_reclaim = used - expected;
4995 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4996 space_info->bytes_reserved);
5000 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5001 struct btrfs_space_info *space_info,
5002 u64 used, bool system_chunk)
5004 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5006 /* If we're just plain full then async reclaim just slows us down. */
5007 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5010 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5014 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5015 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5018 static void wake_all_tickets(struct list_head *head)
5020 struct reserve_ticket *ticket;
5022 while (!list_empty(head)) {
5023 ticket = list_first_entry(head, struct reserve_ticket, list);
5024 list_del_init(&ticket->list);
5025 ticket->error = -ENOSPC;
5026 wake_up(&ticket->wait);
5031 * This is for normal flushers, we can wait all goddamned day if we want to. We
5032 * will loop and continuously try to flush as long as we are making progress.
5033 * We count progress as clearing off tickets each time we have to loop.
5035 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5037 struct btrfs_fs_info *fs_info;
5038 struct btrfs_space_info *space_info;
5041 int commit_cycles = 0;
5042 u64 last_tickets_id;
5044 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5045 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5047 spin_lock(&space_info->lock);
5048 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5051 space_info->flush = 0;
5052 spin_unlock(&space_info->lock);
5055 last_tickets_id = space_info->tickets_id;
5056 spin_unlock(&space_info->lock);
5058 flush_state = FLUSH_DELAYED_ITEMS_NR;
5060 flush_space(fs_info, space_info, to_reclaim, flush_state);
5061 spin_lock(&space_info->lock);
5062 if (list_empty(&space_info->tickets)) {
5063 space_info->flush = 0;
5064 spin_unlock(&space_info->lock);
5067 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5070 if (last_tickets_id == space_info->tickets_id) {
5073 last_tickets_id = space_info->tickets_id;
5074 flush_state = FLUSH_DELAYED_ITEMS_NR;
5079 if (flush_state > COMMIT_TRANS) {
5081 if (commit_cycles > 2) {
5082 wake_all_tickets(&space_info->tickets);
5083 space_info->flush = 0;
5085 flush_state = FLUSH_DELAYED_ITEMS_NR;
5088 spin_unlock(&space_info->lock);
5089 } while (flush_state <= COMMIT_TRANS);
5092 void btrfs_init_async_reclaim_work(struct work_struct *work)
5094 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5097 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5098 struct btrfs_space_info *space_info,
5099 struct reserve_ticket *ticket)
5102 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5104 spin_lock(&space_info->lock);
5105 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5108 spin_unlock(&space_info->lock);
5111 spin_unlock(&space_info->lock);
5114 flush_space(fs_info, space_info, to_reclaim, flush_state);
5116 spin_lock(&space_info->lock);
5117 if (ticket->bytes == 0) {
5118 spin_unlock(&space_info->lock);
5121 spin_unlock(&space_info->lock);
5124 * Priority flushers can't wait on delalloc without
5127 if (flush_state == FLUSH_DELALLOC ||
5128 flush_state == FLUSH_DELALLOC_WAIT)
5129 flush_state = ALLOC_CHUNK;
5130 } while (flush_state < COMMIT_TRANS);
5133 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5134 struct btrfs_space_info *space_info,
5135 struct reserve_ticket *ticket, u64 orig_bytes)
5141 spin_lock(&space_info->lock);
5142 while (ticket->bytes > 0 && ticket->error == 0) {
5143 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5148 spin_unlock(&space_info->lock);
5152 finish_wait(&ticket->wait, &wait);
5153 spin_lock(&space_info->lock);
5156 ret = ticket->error;
5157 if (!list_empty(&ticket->list))
5158 list_del_init(&ticket->list);
5159 if (ticket->bytes && ticket->bytes < orig_bytes) {
5160 u64 num_bytes = orig_bytes - ticket->bytes;
5161 space_info->bytes_may_use -= num_bytes;
5162 trace_btrfs_space_reservation(fs_info, "space_info",
5163 space_info->flags, num_bytes, 0);
5165 spin_unlock(&space_info->lock);
5171 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5172 * @root - the root we're allocating for
5173 * @space_info - the space info we want to allocate from
5174 * @orig_bytes - the number of bytes we want
5175 * @flush - whether or not we can flush to make our reservation
5177 * This will reserve orig_bytes number of bytes from the space info associated
5178 * with the block_rsv. If there is not enough space it will make an attempt to
5179 * flush out space to make room. It will do this by flushing delalloc if
5180 * possible or committing the transaction. If flush is 0 then no attempts to
5181 * regain reservations will be made and this will fail if there is not enough
5184 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5185 struct btrfs_space_info *space_info,
5187 enum btrfs_reserve_flush_enum flush,
5190 struct reserve_ticket ticket;
5195 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5197 spin_lock(&space_info->lock);
5199 used = btrfs_space_info_used(space_info, true);
5202 * If we have enough space then hooray, make our reservation and carry
5203 * on. If not see if we can overcommit, and if we can, hooray carry on.
5204 * If not things get more complicated.
5206 if (used + orig_bytes <= space_info->total_bytes) {
5207 space_info->bytes_may_use += orig_bytes;
5208 trace_btrfs_space_reservation(fs_info, "space_info",
5209 space_info->flags, orig_bytes, 1);
5211 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5213 space_info->bytes_may_use += orig_bytes;
5214 trace_btrfs_space_reservation(fs_info, "space_info",
5215 space_info->flags, orig_bytes, 1);
5220 * If we couldn't make a reservation then setup our reservation ticket
5221 * and kick the async worker if it's not already running.
5223 * If we are a priority flusher then we just need to add our ticket to
5224 * the list and we will do our own flushing further down.
5226 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5227 ticket.bytes = orig_bytes;
5229 init_waitqueue_head(&ticket.wait);
5230 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5231 list_add_tail(&ticket.list, &space_info->tickets);
5232 if (!space_info->flush) {
5233 space_info->flush = 1;
5234 trace_btrfs_trigger_flush(fs_info,
5238 queue_work(system_unbound_wq,
5239 &fs_info->async_reclaim_work);
5242 list_add_tail(&ticket.list,
5243 &space_info->priority_tickets);
5245 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5248 * We will do the space reservation dance during log replay,
5249 * which means we won't have fs_info->fs_root set, so don't do
5250 * the async reclaim as we will panic.
5252 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5253 need_do_async_reclaim(fs_info, space_info,
5254 used, system_chunk) &&
5255 !work_busy(&fs_info->async_reclaim_work)) {
5256 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5257 orig_bytes, flush, "preempt");
5258 queue_work(system_unbound_wq,
5259 &fs_info->async_reclaim_work);
5262 spin_unlock(&space_info->lock);
5263 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5266 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5267 return wait_reserve_ticket(fs_info, space_info, &ticket,
5271 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5272 spin_lock(&space_info->lock);
5274 if (ticket.bytes < orig_bytes) {
5275 u64 num_bytes = orig_bytes - ticket.bytes;
5276 space_info->bytes_may_use -= num_bytes;
5277 trace_btrfs_space_reservation(fs_info, "space_info",
5282 list_del_init(&ticket.list);
5285 spin_unlock(&space_info->lock);
5286 ASSERT(list_empty(&ticket.list));
5291 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5292 * @root - the root we're allocating for
5293 * @block_rsv - the block_rsv we're allocating for
5294 * @orig_bytes - the number of bytes we want
5295 * @flush - whether or not we can flush to make our reservation
5297 * This will reserve orgi_bytes number of bytes from the space info associated
5298 * with the block_rsv. If there is not enough space it will make an attempt to
5299 * flush out space to make room. It will do this by flushing delalloc if
5300 * possible or committing the transaction. If flush is 0 then no attempts to
5301 * regain reservations will be made and this will fail if there is not enough
5304 static int reserve_metadata_bytes(struct btrfs_root *root,
5305 struct btrfs_block_rsv *block_rsv,
5307 enum btrfs_reserve_flush_enum flush)
5309 struct btrfs_fs_info *fs_info = root->fs_info;
5310 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5312 bool system_chunk = (root == fs_info->chunk_root);
5314 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5315 orig_bytes, flush, system_chunk);
5316 if (ret == -ENOSPC &&
5317 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5318 if (block_rsv != global_rsv &&
5319 !block_rsv_use_bytes(global_rsv, orig_bytes))
5323 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5324 block_rsv->space_info->flags,
5329 static struct btrfs_block_rsv *get_block_rsv(
5330 const struct btrfs_trans_handle *trans,
5331 const struct btrfs_root *root)
5333 struct btrfs_fs_info *fs_info = root->fs_info;
5334 struct btrfs_block_rsv *block_rsv = NULL;
5336 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5337 (root == fs_info->csum_root && trans->adding_csums) ||
5338 (root == fs_info->uuid_root))
5339 block_rsv = trans->block_rsv;
5342 block_rsv = root->block_rsv;
5345 block_rsv = &fs_info->empty_block_rsv;
5350 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5354 spin_lock(&block_rsv->lock);
5355 if (block_rsv->reserved >= num_bytes) {
5356 block_rsv->reserved -= num_bytes;
5357 if (block_rsv->reserved < block_rsv->size)
5358 block_rsv->full = 0;
5361 spin_unlock(&block_rsv->lock);
5365 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5366 u64 num_bytes, int update_size)
5368 spin_lock(&block_rsv->lock);
5369 block_rsv->reserved += num_bytes;
5371 block_rsv->size += num_bytes;
5372 else if (block_rsv->reserved >= block_rsv->size)
5373 block_rsv->full = 1;
5374 spin_unlock(&block_rsv->lock);
5377 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5378 struct btrfs_block_rsv *dest, u64 num_bytes,
5381 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5384 if (global_rsv->space_info != dest->space_info)
5387 spin_lock(&global_rsv->lock);
5388 min_bytes = div_factor(global_rsv->size, min_factor);
5389 if (global_rsv->reserved < min_bytes + num_bytes) {
5390 spin_unlock(&global_rsv->lock);
5393 global_rsv->reserved -= num_bytes;
5394 if (global_rsv->reserved < global_rsv->size)
5395 global_rsv->full = 0;
5396 spin_unlock(&global_rsv->lock);
5398 block_rsv_add_bytes(dest, num_bytes, 1);
5403 * This is for space we already have accounted in space_info->bytes_may_use, so
5404 * basically when we're returning space from block_rsv's.
5406 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5407 struct btrfs_space_info *space_info,
5410 struct reserve_ticket *ticket;
5411 struct list_head *head;
5413 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5414 bool check_overcommit = false;
5416 spin_lock(&space_info->lock);
5417 head = &space_info->priority_tickets;
5420 * If we are over our limit then we need to check and see if we can
5421 * overcommit, and if we can't then we just need to free up our space
5422 * and not satisfy any requests.
5424 used = btrfs_space_info_used(space_info, true);
5425 if (used - num_bytes >= space_info->total_bytes)
5426 check_overcommit = true;
5428 while (!list_empty(head) && num_bytes) {
5429 ticket = list_first_entry(head, struct reserve_ticket,
5432 * We use 0 bytes because this space is already reserved, so
5433 * adding the ticket space would be a double count.
5435 if (check_overcommit &&
5436 !can_overcommit(fs_info, space_info, 0, flush, false))
5438 if (num_bytes >= ticket->bytes) {
5439 list_del_init(&ticket->list);
5440 num_bytes -= ticket->bytes;
5442 space_info->tickets_id++;
5443 wake_up(&ticket->wait);
5445 ticket->bytes -= num_bytes;
5450 if (num_bytes && head == &space_info->priority_tickets) {
5451 head = &space_info->tickets;
5452 flush = BTRFS_RESERVE_FLUSH_ALL;
5455 space_info->bytes_may_use -= num_bytes;
5456 trace_btrfs_space_reservation(fs_info, "space_info",
5457 space_info->flags, num_bytes, 0);
5458 spin_unlock(&space_info->lock);
5462 * This is for newly allocated space that isn't accounted in
5463 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5464 * we use this helper.
5466 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5467 struct btrfs_space_info *space_info,
5470 struct reserve_ticket *ticket;
5471 struct list_head *head = &space_info->priority_tickets;
5474 while (!list_empty(head) && num_bytes) {
5475 ticket = list_first_entry(head, struct reserve_ticket,
5477 if (num_bytes >= ticket->bytes) {
5478 trace_btrfs_space_reservation(fs_info, "space_info",
5481 list_del_init(&ticket->list);
5482 num_bytes -= ticket->bytes;
5483 space_info->bytes_may_use += ticket->bytes;
5485 space_info->tickets_id++;
5486 wake_up(&ticket->wait);
5488 trace_btrfs_space_reservation(fs_info, "space_info",
5491 space_info->bytes_may_use += num_bytes;
5492 ticket->bytes -= num_bytes;
5497 if (num_bytes && head == &space_info->priority_tickets) {
5498 head = &space_info->tickets;
5503 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5504 struct btrfs_block_rsv *block_rsv,
5505 struct btrfs_block_rsv *dest, u64 num_bytes)
5507 struct btrfs_space_info *space_info = block_rsv->space_info;
5509 spin_lock(&block_rsv->lock);
5510 if (num_bytes == (u64)-1)
5511 num_bytes = block_rsv->size;
5512 block_rsv->size -= num_bytes;
5513 if (block_rsv->reserved >= block_rsv->size) {
5514 num_bytes = block_rsv->reserved - block_rsv->size;
5515 block_rsv->reserved = block_rsv->size;
5516 block_rsv->full = 1;
5520 spin_unlock(&block_rsv->lock);
5522 if (num_bytes > 0) {
5524 spin_lock(&dest->lock);
5528 bytes_to_add = dest->size - dest->reserved;
5529 bytes_to_add = min(num_bytes, bytes_to_add);
5530 dest->reserved += bytes_to_add;
5531 if (dest->reserved >= dest->size)
5533 num_bytes -= bytes_to_add;
5535 spin_unlock(&dest->lock);
5538 space_info_add_old_bytes(fs_info, space_info,
5543 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5544 struct btrfs_block_rsv *dst, u64 num_bytes,
5549 ret = block_rsv_use_bytes(src, num_bytes);
5553 block_rsv_add_bytes(dst, num_bytes, update_size);
5557 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5559 memset(rsv, 0, sizeof(*rsv));
5560 spin_lock_init(&rsv->lock);
5564 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5565 unsigned short type)
5567 struct btrfs_block_rsv *block_rsv;
5569 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5573 btrfs_init_block_rsv(block_rsv, type);
5574 block_rsv->space_info = __find_space_info(fs_info,
5575 BTRFS_BLOCK_GROUP_METADATA);
5579 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5580 struct btrfs_block_rsv *rsv)
5584 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5588 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5593 int btrfs_block_rsv_add(struct btrfs_root *root,
5594 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5595 enum btrfs_reserve_flush_enum flush)
5602 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5604 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5611 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5619 spin_lock(&block_rsv->lock);
5620 num_bytes = div_factor(block_rsv->size, min_factor);
5621 if (block_rsv->reserved >= num_bytes)
5623 spin_unlock(&block_rsv->lock);
5628 int btrfs_block_rsv_refill(struct btrfs_root *root,
5629 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5630 enum btrfs_reserve_flush_enum flush)
5638 spin_lock(&block_rsv->lock);
5639 num_bytes = min_reserved;
5640 if (block_rsv->reserved >= num_bytes)
5643 num_bytes -= block_rsv->reserved;
5644 spin_unlock(&block_rsv->lock);
5649 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5651 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5658 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5659 struct btrfs_block_rsv *block_rsv,
5662 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5664 if (global_rsv == block_rsv ||
5665 block_rsv->space_info != global_rsv->space_info)
5667 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5670 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5672 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5673 struct btrfs_space_info *sinfo = block_rsv->space_info;
5677 * The global block rsv is based on the size of the extent tree, the
5678 * checksum tree and the root tree. If the fs is empty we want to set
5679 * it to a minimal amount for safety.
5681 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5682 btrfs_root_used(&fs_info->csum_root->root_item) +
5683 btrfs_root_used(&fs_info->tree_root->root_item);
5684 num_bytes = max_t(u64, num_bytes, SZ_16M);
5686 spin_lock(&sinfo->lock);
5687 spin_lock(&block_rsv->lock);
5689 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5691 if (block_rsv->reserved < block_rsv->size) {
5692 num_bytes = btrfs_space_info_used(sinfo, true);
5693 if (sinfo->total_bytes > num_bytes) {
5694 num_bytes = sinfo->total_bytes - num_bytes;
5695 num_bytes = min(num_bytes,
5696 block_rsv->size - block_rsv->reserved);
5697 block_rsv->reserved += num_bytes;
5698 sinfo->bytes_may_use += num_bytes;
5699 trace_btrfs_space_reservation(fs_info, "space_info",
5700 sinfo->flags, num_bytes,
5703 } else if (block_rsv->reserved > block_rsv->size) {
5704 num_bytes = block_rsv->reserved - block_rsv->size;
5705 sinfo->bytes_may_use -= num_bytes;
5706 trace_btrfs_space_reservation(fs_info, "space_info",
5707 sinfo->flags, num_bytes, 0);
5708 block_rsv->reserved = block_rsv->size;
5711 if (block_rsv->reserved == block_rsv->size)
5712 block_rsv->full = 1;
5714 block_rsv->full = 0;
5716 spin_unlock(&block_rsv->lock);
5717 spin_unlock(&sinfo->lock);
5720 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5722 struct btrfs_space_info *space_info;
5724 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5725 fs_info->chunk_block_rsv.space_info = space_info;
5727 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5728 fs_info->global_block_rsv.space_info = space_info;
5729 fs_info->delalloc_block_rsv.space_info = space_info;
5730 fs_info->trans_block_rsv.space_info = space_info;
5731 fs_info->empty_block_rsv.space_info = space_info;
5732 fs_info->delayed_block_rsv.space_info = space_info;
5734 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5735 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5736 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5737 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5738 if (fs_info->quota_root)
5739 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5740 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5742 update_global_block_rsv(fs_info);
5745 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5747 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5749 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5750 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5751 WARN_ON(fs_info->trans_block_rsv.size > 0);
5752 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5753 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5754 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5755 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5756 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5759 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5760 struct btrfs_fs_info *fs_info)
5762 if (!trans->block_rsv)
5765 if (!trans->bytes_reserved)
5768 trace_btrfs_space_reservation(fs_info, "transaction",
5769 trans->transid, trans->bytes_reserved, 0);
5770 btrfs_block_rsv_release(fs_info, trans->block_rsv,
5771 trans->bytes_reserved);
5772 trans->bytes_reserved = 0;
5776 * To be called after all the new block groups attached to the transaction
5777 * handle have been created (btrfs_create_pending_block_groups()).
5779 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5781 struct btrfs_fs_info *fs_info = trans->fs_info;
5783 if (!trans->chunk_bytes_reserved)
5786 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5788 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5789 trans->chunk_bytes_reserved);
5790 trans->chunk_bytes_reserved = 0;
5793 /* Can only return 0 or -ENOSPC */
5794 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5795 struct btrfs_inode *inode)
5797 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5798 struct btrfs_root *root = inode->root;
5800 * We always use trans->block_rsv here as we will have reserved space
5801 * for our orphan when starting the transaction, using get_block_rsv()
5802 * here will sometimes make us choose the wrong block rsv as we could be
5803 * doing a reloc inode for a non refcounted root.
5805 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5806 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5809 * We need to hold space in order to delete our orphan item once we've
5810 * added it, so this takes the reservation so we can release it later
5811 * when we are truly done with the orphan item.
5813 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5815 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5817 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5820 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5822 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5823 struct btrfs_root *root = inode->root;
5824 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5826 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5828 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5832 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5833 * root: the root of the parent directory
5834 * rsv: block reservation
5835 * items: the number of items that we need do reservation
5836 * qgroup_reserved: used to return the reserved size in qgroup
5838 * This function is used to reserve the space for snapshot/subvolume
5839 * creation and deletion. Those operations are different with the
5840 * common file/directory operations, they change two fs/file trees
5841 * and root tree, the number of items that the qgroup reserves is
5842 * different with the free space reservation. So we can not use
5843 * the space reservation mechanism in start_transaction().
5845 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5846 struct btrfs_block_rsv *rsv,
5848 u64 *qgroup_reserved,
5849 bool use_global_rsv)
5853 struct btrfs_fs_info *fs_info = root->fs_info;
5854 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5856 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5857 /* One for parent inode, two for dir entries */
5858 num_bytes = 3 * fs_info->nodesize;
5859 ret = btrfs_qgroup_reserve_meta(root, num_bytes, true);
5866 *qgroup_reserved = num_bytes;
5868 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5869 rsv->space_info = __find_space_info(fs_info,
5870 BTRFS_BLOCK_GROUP_METADATA);
5871 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5872 BTRFS_RESERVE_FLUSH_ALL);
5874 if (ret == -ENOSPC && use_global_rsv)
5875 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5877 if (ret && *qgroup_reserved)
5878 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5883 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5884 struct btrfs_block_rsv *rsv)
5886 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5890 * drop_outstanding_extent - drop an outstanding extent
5891 * @inode: the inode we're dropping the extent for
5892 * @num_bytes: the number of bytes we're releasing.
5894 * This is called when we are freeing up an outstanding extent, either called
5895 * after an error or after an extent is written. This will return the number of
5896 * reserved extents that need to be freed. This must be called with
5897 * BTRFS_I(inode)->lock held.
5899 static unsigned drop_outstanding_extent(struct btrfs_inode *inode,
5902 unsigned drop_inode_space = 0;
5903 unsigned dropped_extents = 0;
5904 unsigned num_extents;
5906 num_extents = count_max_extents(num_bytes);
5907 ASSERT(num_extents);
5908 ASSERT(inode->outstanding_extents >= num_extents);
5909 inode->outstanding_extents -= num_extents;
5911 if (inode->outstanding_extents == 0 &&
5912 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5913 &inode->runtime_flags))
5914 drop_inode_space = 1;
5917 * If we have more or the same amount of outstanding extents than we have
5918 * reserved then we need to leave the reserved extents count alone.
5920 if (inode->outstanding_extents >= inode->reserved_extents)
5921 return drop_inode_space;
5923 dropped_extents = inode->reserved_extents - inode->outstanding_extents;
5924 inode->reserved_extents -= dropped_extents;
5925 return dropped_extents + drop_inode_space;
5929 * calc_csum_metadata_size - return the amount of metadata space that must be
5930 * reserved/freed for the given bytes.
5931 * @inode: the inode we're manipulating
5932 * @num_bytes: the number of bytes in question
5933 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5935 * This adjusts the number of csum_bytes in the inode and then returns the
5936 * correct amount of metadata that must either be reserved or freed. We
5937 * calculate how many checksums we can fit into one leaf and then divide the
5938 * number of bytes that will need to be checksumed by this value to figure out
5939 * how many checksums will be required. If we are adding bytes then the number
5940 * may go up and we will return the number of additional bytes that must be
5941 * reserved. If it is going down we will return the number of bytes that must
5944 * This must be called with BTRFS_I(inode)->lock held.
5946 static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes,
5949 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5950 u64 old_csums, num_csums;
5952 if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0)
5955 old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
5957 inode->csum_bytes += num_bytes;
5959 inode->csum_bytes -= num_bytes;
5960 num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
5962 /* No change, no need to reserve more */
5963 if (old_csums == num_csums)
5967 return btrfs_calc_trans_metadata_size(fs_info,
5968 num_csums - old_csums);
5970 return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums);
5973 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
5975 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5976 struct btrfs_root *root = inode->root;
5977 struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv;
5980 unsigned nr_extents;
5981 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5983 bool delalloc_lock = true;
5986 bool release_extra = false;
5988 /* If we are a free space inode we need to not flush since we will be in
5989 * the middle of a transaction commit. We also don't need the delalloc
5990 * mutex since we won't race with anybody. We need this mostly to make
5991 * lockdep shut its filthy mouth.
5993 * If we have a transaction open (can happen if we call truncate_block
5994 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5996 if (btrfs_is_free_space_inode(inode)) {
5997 flush = BTRFS_RESERVE_NO_FLUSH;
5998 delalloc_lock = false;
5999 } else if (current->journal_info) {
6000 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6003 if (flush != BTRFS_RESERVE_NO_FLUSH &&
6004 btrfs_transaction_in_commit(fs_info))
6005 schedule_timeout(1);
6008 mutex_lock(&inode->delalloc_mutex);
6010 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6012 spin_lock(&inode->lock);
6013 nr_extents = count_max_extents(num_bytes);
6014 inode->outstanding_extents += nr_extents;
6017 if (inode->outstanding_extents > inode->reserved_extents)
6018 nr_extents += inode->outstanding_extents -
6019 inode->reserved_extents;
6021 /* We always want to reserve a slot for updating the inode. */
6022 to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1);
6023 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
6024 csum_bytes = inode->csum_bytes;
6025 spin_unlock(&inode->lock);
6027 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
6028 ret = btrfs_qgroup_reserve_meta(root,
6029 nr_extents * fs_info->nodesize, true);
6034 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
6035 if (unlikely(ret)) {
6036 btrfs_qgroup_free_meta(root,
6037 nr_extents * fs_info->nodesize);
6041 spin_lock(&inode->lock);
6042 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6043 &inode->runtime_flags)) {
6044 to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1);
6045 release_extra = true;
6047 inode->reserved_extents += nr_extents;
6048 spin_unlock(&inode->lock);
6051 mutex_unlock(&inode->delalloc_mutex);
6054 trace_btrfs_space_reservation(fs_info, "delalloc",
6055 btrfs_ino(inode), to_reserve, 1);
6057 btrfs_block_rsv_release(fs_info, block_rsv,
6058 btrfs_calc_trans_metadata_size(fs_info, 1));
6062 spin_lock(&inode->lock);
6063 dropped = drop_outstanding_extent(inode, num_bytes);
6065 * If the inodes csum_bytes is the same as the original
6066 * csum_bytes then we know we haven't raced with any free()ers
6067 * so we can just reduce our inodes csum bytes and carry on.
6069 if (inode->csum_bytes == csum_bytes) {
6070 calc_csum_metadata_size(inode, num_bytes, 0);
6072 u64 orig_csum_bytes = inode->csum_bytes;
6076 * This is tricky, but first we need to figure out how much we
6077 * freed from any free-ers that occurred during this
6078 * reservation, so we reset ->csum_bytes to the csum_bytes
6079 * before we dropped our lock, and then call the free for the
6080 * number of bytes that were freed while we were trying our
6083 bytes = csum_bytes - inode->csum_bytes;
6084 inode->csum_bytes = csum_bytes;
6085 to_free = calc_csum_metadata_size(inode, bytes, 0);
6089 * Now we need to see how much we would have freed had we not
6090 * been making this reservation and our ->csum_bytes were not
6091 * artificially inflated.
6093 inode->csum_bytes = csum_bytes - num_bytes;
6094 bytes = csum_bytes - orig_csum_bytes;
6095 bytes = calc_csum_metadata_size(inode, bytes, 0);
6098 * Now reset ->csum_bytes to what it should be. If bytes is
6099 * more than to_free then we would have freed more space had we
6100 * not had an artificially high ->csum_bytes, so we need to free
6101 * the remainder. If bytes is the same or less then we don't
6102 * need to do anything, the other free-ers did the correct
6105 inode->csum_bytes = orig_csum_bytes - num_bytes;
6106 if (bytes > to_free)
6107 to_free = bytes - to_free;
6111 spin_unlock(&inode->lock);
6113 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6116 btrfs_block_rsv_release(fs_info, block_rsv, to_free);
6117 trace_btrfs_space_reservation(fs_info, "delalloc",
6118 btrfs_ino(inode), to_free, 0);
6121 mutex_unlock(&inode->delalloc_mutex);
6126 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6127 * @inode: the inode to release the reservation for
6128 * @num_bytes: the number of bytes we're releasing
6130 * This will release the metadata reservation for an inode. This can be called
6131 * once we complete IO for a given set of bytes to release their metadata
6134 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes)
6136 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6140 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6141 spin_lock(&inode->lock);
6142 dropped = drop_outstanding_extent(inode, num_bytes);
6145 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6146 spin_unlock(&inode->lock);
6148 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6150 if (btrfs_is_testing(fs_info))
6153 trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode),
6156 btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free);
6160 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6162 * @inode: inode we're writing to
6163 * @start: start range we are writing to
6164 * @len: how long the range we are writing to
6165 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6166 * current reservation.
6168 * This will do the following things
6170 * o reserve space in data space info for num bytes
6171 * and reserve precious corresponding qgroup space
6172 * (Done in check_data_free_space)
6174 * o reserve space for metadata space, based on the number of outstanding
6175 * extents and how much csums will be needed
6176 * also reserve metadata space in a per root over-reserve method.
6177 * o add to the inodes->delalloc_bytes
6178 * o add it to the fs_info's delalloc inodes list.
6179 * (Above 3 all done in delalloc_reserve_metadata)
6181 * Return 0 for success
6182 * Return <0 for error(-ENOSPC or -EQUOT)
6184 int btrfs_delalloc_reserve_space(struct inode *inode,
6185 struct extent_changeset **reserved, u64 start, u64 len)
6189 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6192 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6194 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6199 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6200 * @inode: inode we're releasing space for
6201 * @start: start position of the space already reserved
6202 * @len: the len of the space already reserved
6204 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6205 * called in the case that we don't need the metadata AND data reservations
6206 * anymore. So if there is an error or we insert an inline extent.
6208 * This function will release the metadata space that was not used and will
6209 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6210 * list if there are no delalloc bytes left.
6211 * Also it will handle the qgroup reserved space.
6213 void btrfs_delalloc_release_space(struct inode *inode,
6214 struct extent_changeset *reserved, u64 start, u64 len)
6216 btrfs_delalloc_release_metadata(BTRFS_I(inode), len);
6217 btrfs_free_reserved_data_space(inode, reserved, start, len);
6220 static int update_block_group(struct btrfs_trans_handle *trans,
6221 struct btrfs_fs_info *info, u64 bytenr,
6222 u64 num_bytes, int alloc)
6224 struct btrfs_block_group_cache *cache = NULL;
6225 u64 total = num_bytes;
6230 /* block accounting for super block */
6231 spin_lock(&info->delalloc_root_lock);
6232 old_val = btrfs_super_bytes_used(info->super_copy);
6234 old_val += num_bytes;
6236 old_val -= num_bytes;
6237 btrfs_set_super_bytes_used(info->super_copy, old_val);
6238 spin_unlock(&info->delalloc_root_lock);
6241 cache = btrfs_lookup_block_group(info, bytenr);
6244 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6245 BTRFS_BLOCK_GROUP_RAID1 |
6246 BTRFS_BLOCK_GROUP_RAID10))
6251 * If this block group has free space cache written out, we
6252 * need to make sure to load it if we are removing space. This
6253 * is because we need the unpinning stage to actually add the
6254 * space back to the block group, otherwise we will leak space.
6256 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6257 cache_block_group(cache, 1);
6259 byte_in_group = bytenr - cache->key.objectid;
6260 WARN_ON(byte_in_group > cache->key.offset);
6262 spin_lock(&cache->space_info->lock);
6263 spin_lock(&cache->lock);
6265 if (btrfs_test_opt(info, SPACE_CACHE) &&
6266 cache->disk_cache_state < BTRFS_DC_CLEAR)
6267 cache->disk_cache_state = BTRFS_DC_CLEAR;
6269 old_val = btrfs_block_group_used(&cache->item);
6270 num_bytes = min(total, cache->key.offset - byte_in_group);
6272 old_val += num_bytes;
6273 btrfs_set_block_group_used(&cache->item, old_val);
6274 cache->reserved -= num_bytes;
6275 cache->space_info->bytes_reserved -= num_bytes;
6276 cache->space_info->bytes_used += num_bytes;
6277 cache->space_info->disk_used += num_bytes * factor;
6278 spin_unlock(&cache->lock);
6279 spin_unlock(&cache->space_info->lock);
6281 old_val -= num_bytes;
6282 btrfs_set_block_group_used(&cache->item, old_val);
6283 cache->pinned += num_bytes;
6284 cache->space_info->bytes_pinned += num_bytes;
6285 cache->space_info->bytes_used -= num_bytes;
6286 cache->space_info->disk_used -= num_bytes * factor;
6287 spin_unlock(&cache->lock);
6288 spin_unlock(&cache->space_info->lock);
6290 trace_btrfs_space_reservation(info, "pinned",
6291 cache->space_info->flags,
6293 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6295 set_extent_dirty(info->pinned_extents,
6296 bytenr, bytenr + num_bytes - 1,
6297 GFP_NOFS | __GFP_NOFAIL);
6300 spin_lock(&trans->transaction->dirty_bgs_lock);
6301 if (list_empty(&cache->dirty_list)) {
6302 list_add_tail(&cache->dirty_list,
6303 &trans->transaction->dirty_bgs);
6304 trans->transaction->num_dirty_bgs++;
6305 btrfs_get_block_group(cache);
6307 spin_unlock(&trans->transaction->dirty_bgs_lock);
6310 * No longer have used bytes in this block group, queue it for
6311 * deletion. We do this after adding the block group to the
6312 * dirty list to avoid races between cleaner kthread and space
6315 if (!alloc && old_val == 0) {
6316 spin_lock(&info->unused_bgs_lock);
6317 if (list_empty(&cache->bg_list)) {
6318 btrfs_get_block_group(cache);
6319 list_add_tail(&cache->bg_list,
6322 spin_unlock(&info->unused_bgs_lock);
6325 btrfs_put_block_group(cache);
6327 bytenr += num_bytes;
6332 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6334 struct btrfs_block_group_cache *cache;
6337 spin_lock(&fs_info->block_group_cache_lock);
6338 bytenr = fs_info->first_logical_byte;
6339 spin_unlock(&fs_info->block_group_cache_lock);
6341 if (bytenr < (u64)-1)
6344 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6348 bytenr = cache->key.objectid;
6349 btrfs_put_block_group(cache);
6354 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6355 struct btrfs_block_group_cache *cache,
6356 u64 bytenr, u64 num_bytes, int reserved)
6358 spin_lock(&cache->space_info->lock);
6359 spin_lock(&cache->lock);
6360 cache->pinned += num_bytes;
6361 cache->space_info->bytes_pinned += num_bytes;
6363 cache->reserved -= num_bytes;
6364 cache->space_info->bytes_reserved -= num_bytes;
6366 spin_unlock(&cache->lock);
6367 spin_unlock(&cache->space_info->lock);
6369 trace_btrfs_space_reservation(fs_info, "pinned",
6370 cache->space_info->flags, num_bytes, 1);
6371 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6372 set_extent_dirty(fs_info->pinned_extents, bytenr,
6373 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6378 * this function must be called within transaction
6380 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6381 u64 bytenr, u64 num_bytes, int reserved)
6383 struct btrfs_block_group_cache *cache;
6385 cache = btrfs_lookup_block_group(fs_info, bytenr);
6386 BUG_ON(!cache); /* Logic error */
6388 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6390 btrfs_put_block_group(cache);
6395 * this function must be called within transaction
6397 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6398 u64 bytenr, u64 num_bytes)
6400 struct btrfs_block_group_cache *cache;
6403 cache = btrfs_lookup_block_group(fs_info, bytenr);
6408 * pull in the free space cache (if any) so that our pin
6409 * removes the free space from the cache. We have load_only set
6410 * to one because the slow code to read in the free extents does check
6411 * the pinned extents.
6413 cache_block_group(cache, 1);
6415 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6417 /* remove us from the free space cache (if we're there at all) */
6418 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6419 btrfs_put_block_group(cache);
6423 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6424 u64 start, u64 num_bytes)
6427 struct btrfs_block_group_cache *block_group;
6428 struct btrfs_caching_control *caching_ctl;
6430 block_group = btrfs_lookup_block_group(fs_info, start);
6434 cache_block_group(block_group, 0);
6435 caching_ctl = get_caching_control(block_group);
6439 BUG_ON(!block_group_cache_done(block_group));
6440 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6442 mutex_lock(&caching_ctl->mutex);
6444 if (start >= caching_ctl->progress) {
6445 ret = add_excluded_extent(fs_info, start, num_bytes);
6446 } else if (start + num_bytes <= caching_ctl->progress) {
6447 ret = btrfs_remove_free_space(block_group,
6450 num_bytes = caching_ctl->progress - start;
6451 ret = btrfs_remove_free_space(block_group,
6456 num_bytes = (start + num_bytes) -
6457 caching_ctl->progress;
6458 start = caching_ctl->progress;
6459 ret = add_excluded_extent(fs_info, start, num_bytes);
6462 mutex_unlock(&caching_ctl->mutex);
6463 put_caching_control(caching_ctl);
6465 btrfs_put_block_group(block_group);
6469 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6470 struct extent_buffer *eb)
6472 struct btrfs_file_extent_item *item;
6473 struct btrfs_key key;
6477 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6480 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6481 btrfs_item_key_to_cpu(eb, &key, i);
6482 if (key.type != BTRFS_EXTENT_DATA_KEY)
6484 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6485 found_type = btrfs_file_extent_type(eb, item);
6486 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6488 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6490 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6491 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6492 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6499 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6501 atomic_inc(&bg->reservations);
6504 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6507 struct btrfs_block_group_cache *bg;
6509 bg = btrfs_lookup_block_group(fs_info, start);
6511 if (atomic_dec_and_test(&bg->reservations))
6512 wake_up_atomic_t(&bg->reservations);
6513 btrfs_put_block_group(bg);
6516 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6522 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6524 struct btrfs_space_info *space_info = bg->space_info;
6528 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6532 * Our block group is read only but before we set it to read only,
6533 * some task might have had allocated an extent from it already, but it
6534 * has not yet created a respective ordered extent (and added it to a
6535 * root's list of ordered extents).
6536 * Therefore wait for any task currently allocating extents, since the
6537 * block group's reservations counter is incremented while a read lock
6538 * on the groups' semaphore is held and decremented after releasing
6539 * the read access on that semaphore and creating the ordered extent.
6541 down_write(&space_info->groups_sem);
6542 up_write(&space_info->groups_sem);
6544 wait_on_atomic_t(&bg->reservations,
6545 btrfs_wait_bg_reservations_atomic_t,
6546 TASK_UNINTERRUPTIBLE);
6550 * btrfs_add_reserved_bytes - update the block_group and space info counters
6551 * @cache: The cache we are manipulating
6552 * @ram_bytes: The number of bytes of file content, and will be same to
6553 * @num_bytes except for the compress path.
6554 * @num_bytes: The number of bytes in question
6555 * @delalloc: The blocks are allocated for the delalloc write
6557 * This is called by the allocator when it reserves space. If this is a
6558 * reservation and the block group has become read only we cannot make the
6559 * reservation and return -EAGAIN, otherwise this function always succeeds.
6561 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6562 u64 ram_bytes, u64 num_bytes, int delalloc)
6564 struct btrfs_space_info *space_info = cache->space_info;
6567 spin_lock(&space_info->lock);
6568 spin_lock(&cache->lock);
6572 cache->reserved += num_bytes;
6573 space_info->bytes_reserved += num_bytes;
6575 trace_btrfs_space_reservation(cache->fs_info,
6576 "space_info", space_info->flags,
6578 space_info->bytes_may_use -= ram_bytes;
6580 cache->delalloc_bytes += num_bytes;
6582 spin_unlock(&cache->lock);
6583 spin_unlock(&space_info->lock);
6588 * btrfs_free_reserved_bytes - update the block_group and space info counters
6589 * @cache: The cache we are manipulating
6590 * @num_bytes: The number of bytes in question
6591 * @delalloc: The blocks are allocated for the delalloc write
6593 * This is called by somebody who is freeing space that was never actually used
6594 * on disk. For example if you reserve some space for a new leaf in transaction
6595 * A and before transaction A commits you free that leaf, you call this with
6596 * reserve set to 0 in order to clear the reservation.
6599 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6600 u64 num_bytes, int delalloc)
6602 struct btrfs_space_info *space_info = cache->space_info;
6605 spin_lock(&space_info->lock);
6606 spin_lock(&cache->lock);
6608 space_info->bytes_readonly += num_bytes;
6609 cache->reserved -= num_bytes;
6610 space_info->bytes_reserved -= num_bytes;
6613 cache->delalloc_bytes -= num_bytes;
6614 spin_unlock(&cache->lock);
6615 spin_unlock(&space_info->lock);
6618 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6620 struct btrfs_caching_control *next;
6621 struct btrfs_caching_control *caching_ctl;
6622 struct btrfs_block_group_cache *cache;
6624 down_write(&fs_info->commit_root_sem);
6626 list_for_each_entry_safe(caching_ctl, next,
6627 &fs_info->caching_block_groups, list) {
6628 cache = caching_ctl->block_group;
6629 if (block_group_cache_done(cache)) {
6630 cache->last_byte_to_unpin = (u64)-1;
6631 list_del_init(&caching_ctl->list);
6632 put_caching_control(caching_ctl);
6634 cache->last_byte_to_unpin = caching_ctl->progress;
6638 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6639 fs_info->pinned_extents = &fs_info->freed_extents[1];
6641 fs_info->pinned_extents = &fs_info->freed_extents[0];
6643 up_write(&fs_info->commit_root_sem);
6645 update_global_block_rsv(fs_info);
6649 * Returns the free cluster for the given space info and sets empty_cluster to
6650 * what it should be based on the mount options.
6652 static struct btrfs_free_cluster *
6653 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6654 struct btrfs_space_info *space_info, u64 *empty_cluster)
6656 struct btrfs_free_cluster *ret = NULL;
6659 if (btrfs_mixed_space_info(space_info))
6662 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6663 ret = &fs_info->meta_alloc_cluster;
6664 if (btrfs_test_opt(fs_info, SSD))
6665 *empty_cluster = SZ_2M;
6667 *empty_cluster = SZ_64K;
6668 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6669 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6670 *empty_cluster = SZ_2M;
6671 ret = &fs_info->data_alloc_cluster;
6677 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6679 const bool return_free_space)
6681 struct btrfs_block_group_cache *cache = NULL;
6682 struct btrfs_space_info *space_info;
6683 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6684 struct btrfs_free_cluster *cluster = NULL;
6686 u64 total_unpinned = 0;
6687 u64 empty_cluster = 0;
6690 while (start <= end) {
6693 start >= cache->key.objectid + cache->key.offset) {
6695 btrfs_put_block_group(cache);
6697 cache = btrfs_lookup_block_group(fs_info, start);
6698 BUG_ON(!cache); /* Logic error */
6700 cluster = fetch_cluster_info(fs_info,
6703 empty_cluster <<= 1;
6706 len = cache->key.objectid + cache->key.offset - start;
6707 len = min(len, end + 1 - start);
6709 if (start < cache->last_byte_to_unpin) {
6710 len = min(len, cache->last_byte_to_unpin - start);
6711 if (return_free_space)
6712 btrfs_add_free_space(cache, start, len);
6716 total_unpinned += len;
6717 space_info = cache->space_info;
6720 * If this space cluster has been marked as fragmented and we've
6721 * unpinned enough in this block group to potentially allow a
6722 * cluster to be created inside of it go ahead and clear the
6725 if (cluster && cluster->fragmented &&
6726 total_unpinned > empty_cluster) {
6727 spin_lock(&cluster->lock);
6728 cluster->fragmented = 0;
6729 spin_unlock(&cluster->lock);
6732 spin_lock(&space_info->lock);
6733 spin_lock(&cache->lock);
6734 cache->pinned -= len;
6735 space_info->bytes_pinned -= len;
6737 trace_btrfs_space_reservation(fs_info, "pinned",
6738 space_info->flags, len, 0);
6739 space_info->max_extent_size = 0;
6740 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6742 space_info->bytes_readonly += len;
6745 spin_unlock(&cache->lock);
6746 if (!readonly && return_free_space &&
6747 global_rsv->space_info == space_info) {
6750 spin_lock(&global_rsv->lock);
6751 if (!global_rsv->full) {
6752 to_add = min(len, global_rsv->size -
6753 global_rsv->reserved);
6754 global_rsv->reserved += to_add;
6755 space_info->bytes_may_use += to_add;
6756 if (global_rsv->reserved >= global_rsv->size)
6757 global_rsv->full = 1;
6758 trace_btrfs_space_reservation(fs_info,
6764 spin_unlock(&global_rsv->lock);
6765 /* Add to any tickets we may have */
6767 space_info_add_new_bytes(fs_info, space_info,
6770 spin_unlock(&space_info->lock);
6774 btrfs_put_block_group(cache);
6778 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6779 struct btrfs_fs_info *fs_info)
6781 struct btrfs_block_group_cache *block_group, *tmp;
6782 struct list_head *deleted_bgs;
6783 struct extent_io_tree *unpin;
6788 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6789 unpin = &fs_info->freed_extents[1];
6791 unpin = &fs_info->freed_extents[0];
6793 while (!trans->aborted) {
6794 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6795 ret = find_first_extent_bit(unpin, 0, &start, &end,
6796 EXTENT_DIRTY, NULL);
6798 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6802 if (btrfs_test_opt(fs_info, DISCARD))
6803 ret = btrfs_discard_extent(fs_info, start,
6804 end + 1 - start, NULL);
6806 clear_extent_dirty(unpin, start, end);
6807 unpin_extent_range(fs_info, start, end, true);
6808 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6813 * Transaction is finished. We don't need the lock anymore. We
6814 * do need to clean up the block groups in case of a transaction
6817 deleted_bgs = &trans->transaction->deleted_bgs;
6818 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6822 if (!trans->aborted)
6823 ret = btrfs_discard_extent(fs_info,
6824 block_group->key.objectid,
6825 block_group->key.offset,
6828 list_del_init(&block_group->bg_list);
6829 btrfs_put_block_group_trimming(block_group);
6830 btrfs_put_block_group(block_group);
6833 const char *errstr = btrfs_decode_error(ret);
6835 "discard failed while removing blockgroup: errno=%d %s",
6843 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6844 struct btrfs_fs_info *info,
6845 struct btrfs_delayed_ref_node *node, u64 parent,
6846 u64 root_objectid, u64 owner_objectid,
6847 u64 owner_offset, int refs_to_drop,
6848 struct btrfs_delayed_extent_op *extent_op)
6850 struct btrfs_key key;
6851 struct btrfs_path *path;
6852 struct btrfs_root *extent_root = info->extent_root;
6853 struct extent_buffer *leaf;
6854 struct btrfs_extent_item *ei;
6855 struct btrfs_extent_inline_ref *iref;
6858 int extent_slot = 0;
6859 int found_extent = 0;
6863 u64 bytenr = node->bytenr;
6864 u64 num_bytes = node->num_bytes;
6866 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6868 path = btrfs_alloc_path();
6872 path->reada = READA_FORWARD;
6873 path->leave_spinning = 1;
6875 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6876 BUG_ON(!is_data && refs_to_drop != 1);
6879 skinny_metadata = 0;
6881 ret = lookup_extent_backref(trans, info, path, &iref,
6882 bytenr, num_bytes, parent,
6883 root_objectid, owner_objectid,
6886 extent_slot = path->slots[0];
6887 while (extent_slot >= 0) {
6888 btrfs_item_key_to_cpu(path->nodes[0], &key,
6890 if (key.objectid != bytenr)
6892 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6893 key.offset == num_bytes) {
6897 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6898 key.offset == owner_objectid) {
6902 if (path->slots[0] - extent_slot > 5)
6906 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6907 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6908 if (found_extent && item_size < sizeof(*ei))
6911 if (!found_extent) {
6913 ret = remove_extent_backref(trans, info, path, NULL,
6915 is_data, &last_ref);
6917 btrfs_abort_transaction(trans, ret);
6920 btrfs_release_path(path);
6921 path->leave_spinning = 1;
6923 key.objectid = bytenr;
6924 key.type = BTRFS_EXTENT_ITEM_KEY;
6925 key.offset = num_bytes;
6927 if (!is_data && skinny_metadata) {
6928 key.type = BTRFS_METADATA_ITEM_KEY;
6929 key.offset = owner_objectid;
6932 ret = btrfs_search_slot(trans, extent_root,
6934 if (ret > 0 && skinny_metadata && path->slots[0]) {
6936 * Couldn't find our skinny metadata item,
6937 * see if we have ye olde extent item.
6940 btrfs_item_key_to_cpu(path->nodes[0], &key,
6942 if (key.objectid == bytenr &&
6943 key.type == BTRFS_EXTENT_ITEM_KEY &&
6944 key.offset == num_bytes)
6948 if (ret > 0 && skinny_metadata) {
6949 skinny_metadata = false;
6950 key.objectid = bytenr;
6951 key.type = BTRFS_EXTENT_ITEM_KEY;
6952 key.offset = num_bytes;
6953 btrfs_release_path(path);
6954 ret = btrfs_search_slot(trans, extent_root,
6960 "umm, got %d back from search, was looking for %llu",
6963 btrfs_print_leaf(path->nodes[0]);
6966 btrfs_abort_transaction(trans, ret);
6969 extent_slot = path->slots[0];
6971 } else if (WARN_ON(ret == -ENOENT)) {
6972 btrfs_print_leaf(path->nodes[0]);
6974 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6975 bytenr, parent, root_objectid, owner_objectid,
6977 btrfs_abort_transaction(trans, ret);
6980 btrfs_abort_transaction(trans, ret);
6984 leaf = path->nodes[0];
6985 item_size = btrfs_item_size_nr(leaf, extent_slot);
6986 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6987 if (item_size < sizeof(*ei)) {
6988 BUG_ON(found_extent || extent_slot != path->slots[0]);
6989 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
6992 btrfs_abort_transaction(trans, ret);
6996 btrfs_release_path(path);
6997 path->leave_spinning = 1;
6999 key.objectid = bytenr;
7000 key.type = BTRFS_EXTENT_ITEM_KEY;
7001 key.offset = num_bytes;
7003 ret = btrfs_search_slot(trans, extent_root, &key, path,
7007 "umm, got %d back from search, was looking for %llu",
7009 btrfs_print_leaf(path->nodes[0]);
7012 btrfs_abort_transaction(trans, ret);
7016 extent_slot = path->slots[0];
7017 leaf = path->nodes[0];
7018 item_size = btrfs_item_size_nr(leaf, extent_slot);
7021 BUG_ON(item_size < sizeof(*ei));
7022 ei = btrfs_item_ptr(leaf, extent_slot,
7023 struct btrfs_extent_item);
7024 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7025 key.type == BTRFS_EXTENT_ITEM_KEY) {
7026 struct btrfs_tree_block_info *bi;
7027 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7028 bi = (struct btrfs_tree_block_info *)(ei + 1);
7029 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7032 refs = btrfs_extent_refs(leaf, ei);
7033 if (refs < refs_to_drop) {
7035 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7036 refs_to_drop, refs, bytenr);
7038 btrfs_abort_transaction(trans, ret);
7041 refs -= refs_to_drop;
7045 __run_delayed_extent_op(extent_op, leaf, ei);
7047 * In the case of inline back ref, reference count will
7048 * be updated by remove_extent_backref
7051 BUG_ON(!found_extent);
7053 btrfs_set_extent_refs(leaf, ei, refs);
7054 btrfs_mark_buffer_dirty(leaf);
7057 ret = remove_extent_backref(trans, info, path,
7059 is_data, &last_ref);
7061 btrfs_abort_transaction(trans, ret);
7067 BUG_ON(is_data && refs_to_drop !=
7068 extent_data_ref_count(path, iref));
7070 BUG_ON(path->slots[0] != extent_slot);
7072 BUG_ON(path->slots[0] != extent_slot + 1);
7073 path->slots[0] = extent_slot;
7079 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7082 btrfs_abort_transaction(trans, ret);
7085 btrfs_release_path(path);
7088 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7090 btrfs_abort_transaction(trans, ret);
7095 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7097 btrfs_abort_transaction(trans, ret);
7101 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7103 btrfs_abort_transaction(trans, ret);
7107 btrfs_release_path(path);
7110 btrfs_free_path(path);
7115 * when we free an block, it is possible (and likely) that we free the last
7116 * delayed ref for that extent as well. This searches the delayed ref tree for
7117 * a given extent, and if there are no other delayed refs to be processed, it
7118 * removes it from the tree.
7120 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7123 struct btrfs_delayed_ref_head *head;
7124 struct btrfs_delayed_ref_root *delayed_refs;
7127 delayed_refs = &trans->transaction->delayed_refs;
7128 spin_lock(&delayed_refs->lock);
7129 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7131 goto out_delayed_unlock;
7133 spin_lock(&head->lock);
7134 if (!list_empty(&head->ref_list))
7137 if (head->extent_op) {
7138 if (!head->must_insert_reserved)
7140 btrfs_free_delayed_extent_op(head->extent_op);
7141 head->extent_op = NULL;
7145 * waiting for the lock here would deadlock. If someone else has it
7146 * locked they are already in the process of dropping it anyway
7148 if (!mutex_trylock(&head->mutex))
7152 * at this point we have a head with no other entries. Go
7153 * ahead and process it.
7155 head->node.in_tree = 0;
7156 rb_erase(&head->href_node, &delayed_refs->href_root);
7158 atomic_dec(&delayed_refs->num_entries);
7161 * we don't take a ref on the node because we're removing it from the
7162 * tree, so we just steal the ref the tree was holding.
7164 delayed_refs->num_heads--;
7165 if (head->processing == 0)
7166 delayed_refs->num_heads_ready--;
7167 head->processing = 0;
7168 spin_unlock(&head->lock);
7169 spin_unlock(&delayed_refs->lock);
7171 BUG_ON(head->extent_op);
7172 if (head->must_insert_reserved)
7175 mutex_unlock(&head->mutex);
7176 btrfs_put_delayed_ref(&head->node);
7179 spin_unlock(&head->lock);
7182 spin_unlock(&delayed_refs->lock);
7186 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7187 struct btrfs_root *root,
7188 struct extent_buffer *buf,
7189 u64 parent, int last_ref)
7191 struct btrfs_fs_info *fs_info = root->fs_info;
7195 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7196 int old_ref_mod, new_ref_mod;
7198 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7200 root->root_key.objectid,
7201 btrfs_header_level(buf),
7202 BTRFS_DROP_DELAYED_REF, NULL,
7203 &old_ref_mod, &new_ref_mod);
7204 BUG_ON(ret); /* -ENOMEM */
7205 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7208 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7209 struct btrfs_block_group_cache *cache;
7211 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7212 ret = check_ref_cleanup(trans, buf->start);
7218 cache = btrfs_lookup_block_group(fs_info, buf->start);
7220 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7221 pin_down_extent(fs_info, cache, buf->start,
7223 btrfs_put_block_group(cache);
7227 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7229 btrfs_add_free_space(cache, buf->start, buf->len);
7230 btrfs_free_reserved_bytes(cache, buf->len, 0);
7231 btrfs_put_block_group(cache);
7232 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7236 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7237 root->root_key.objectid);
7241 * Deleting the buffer, clear the corrupt flag since it doesn't
7244 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7248 /* Can return -ENOMEM */
7249 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7250 struct btrfs_fs_info *fs_info,
7251 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7252 u64 owner, u64 offset)
7254 int old_ref_mod, new_ref_mod;
7257 if (btrfs_is_testing(fs_info))
7262 * tree log blocks never actually go into the extent allocation
7263 * tree, just update pinning info and exit early.
7265 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7266 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7267 /* unlocks the pinned mutex */
7268 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7269 old_ref_mod = new_ref_mod = 0;
7271 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7272 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7274 root_objectid, (int)owner,
7275 BTRFS_DROP_DELAYED_REF, NULL,
7276 &old_ref_mod, &new_ref_mod);
7278 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7280 root_objectid, owner, offset,
7281 0, BTRFS_DROP_DELAYED_REF,
7282 &old_ref_mod, &new_ref_mod);
7285 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7286 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7292 * when we wait for progress in the block group caching, its because
7293 * our allocation attempt failed at least once. So, we must sleep
7294 * and let some progress happen before we try again.
7296 * This function will sleep at least once waiting for new free space to
7297 * show up, and then it will check the block group free space numbers
7298 * for our min num_bytes. Another option is to have it go ahead
7299 * and look in the rbtree for a free extent of a given size, but this
7302 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7303 * any of the information in this block group.
7305 static noinline void
7306 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7309 struct btrfs_caching_control *caching_ctl;
7311 caching_ctl = get_caching_control(cache);
7315 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7316 (cache->free_space_ctl->free_space >= num_bytes));
7318 put_caching_control(caching_ctl);
7322 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7324 struct btrfs_caching_control *caching_ctl;
7327 caching_ctl = get_caching_control(cache);
7329 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7331 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7332 if (cache->cached == BTRFS_CACHE_ERROR)
7334 put_caching_control(caching_ctl);
7338 int __get_raid_index(u64 flags)
7340 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7341 return BTRFS_RAID_RAID10;
7342 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7343 return BTRFS_RAID_RAID1;
7344 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7345 return BTRFS_RAID_DUP;
7346 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7347 return BTRFS_RAID_RAID0;
7348 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7349 return BTRFS_RAID_RAID5;
7350 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7351 return BTRFS_RAID_RAID6;
7353 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7356 int get_block_group_index(struct btrfs_block_group_cache *cache)
7358 return __get_raid_index(cache->flags);
7361 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7362 [BTRFS_RAID_RAID10] = "raid10",
7363 [BTRFS_RAID_RAID1] = "raid1",
7364 [BTRFS_RAID_DUP] = "dup",
7365 [BTRFS_RAID_RAID0] = "raid0",
7366 [BTRFS_RAID_SINGLE] = "single",
7367 [BTRFS_RAID_RAID5] = "raid5",
7368 [BTRFS_RAID_RAID6] = "raid6",
7371 static const char *get_raid_name(enum btrfs_raid_types type)
7373 if (type >= BTRFS_NR_RAID_TYPES)
7376 return btrfs_raid_type_names[type];
7379 enum btrfs_loop_type {
7380 LOOP_CACHING_NOWAIT = 0,
7381 LOOP_CACHING_WAIT = 1,
7382 LOOP_ALLOC_CHUNK = 2,
7383 LOOP_NO_EMPTY_SIZE = 3,
7387 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7391 down_read(&cache->data_rwsem);
7395 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7398 btrfs_get_block_group(cache);
7400 down_read(&cache->data_rwsem);
7403 static struct btrfs_block_group_cache *
7404 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7405 struct btrfs_free_cluster *cluster,
7408 struct btrfs_block_group_cache *used_bg = NULL;
7410 spin_lock(&cluster->refill_lock);
7412 used_bg = cluster->block_group;
7416 if (used_bg == block_group)
7419 btrfs_get_block_group(used_bg);
7424 if (down_read_trylock(&used_bg->data_rwsem))
7427 spin_unlock(&cluster->refill_lock);
7429 /* We should only have one-level nested. */
7430 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7432 spin_lock(&cluster->refill_lock);
7433 if (used_bg == cluster->block_group)
7436 up_read(&used_bg->data_rwsem);
7437 btrfs_put_block_group(used_bg);
7442 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7446 up_read(&cache->data_rwsem);
7447 btrfs_put_block_group(cache);
7451 * walks the btree of allocated extents and find a hole of a given size.
7452 * The key ins is changed to record the hole:
7453 * ins->objectid == start position
7454 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7455 * ins->offset == the size of the hole.
7456 * Any available blocks before search_start are skipped.
7458 * If there is no suitable free space, we will record the max size of
7459 * the free space extent currently.
7461 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7462 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7463 u64 hint_byte, struct btrfs_key *ins,
7464 u64 flags, int delalloc)
7467 struct btrfs_root *root = fs_info->extent_root;
7468 struct btrfs_free_cluster *last_ptr = NULL;
7469 struct btrfs_block_group_cache *block_group = NULL;
7470 u64 search_start = 0;
7471 u64 max_extent_size = 0;
7472 u64 empty_cluster = 0;
7473 struct btrfs_space_info *space_info;
7475 int index = __get_raid_index(flags);
7476 bool failed_cluster_refill = false;
7477 bool failed_alloc = false;
7478 bool use_cluster = true;
7479 bool have_caching_bg = false;
7480 bool orig_have_caching_bg = false;
7481 bool full_search = false;
7483 WARN_ON(num_bytes < fs_info->sectorsize);
7484 ins->type = BTRFS_EXTENT_ITEM_KEY;
7488 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7490 space_info = __find_space_info(fs_info, flags);
7492 btrfs_err(fs_info, "No space info for %llu", flags);
7497 * If our free space is heavily fragmented we may not be able to make
7498 * big contiguous allocations, so instead of doing the expensive search
7499 * for free space, simply return ENOSPC with our max_extent_size so we
7500 * can go ahead and search for a more manageable chunk.
7502 * If our max_extent_size is large enough for our allocation simply
7503 * disable clustering since we will likely not be able to find enough
7504 * space to create a cluster and induce latency trying.
7506 if (unlikely(space_info->max_extent_size)) {
7507 spin_lock(&space_info->lock);
7508 if (space_info->max_extent_size &&
7509 num_bytes > space_info->max_extent_size) {
7510 ins->offset = space_info->max_extent_size;
7511 spin_unlock(&space_info->lock);
7513 } else if (space_info->max_extent_size) {
7514 use_cluster = false;
7516 spin_unlock(&space_info->lock);
7519 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7521 spin_lock(&last_ptr->lock);
7522 if (last_ptr->block_group)
7523 hint_byte = last_ptr->window_start;
7524 if (last_ptr->fragmented) {
7526 * We still set window_start so we can keep track of the
7527 * last place we found an allocation to try and save
7530 hint_byte = last_ptr->window_start;
7531 use_cluster = false;
7533 spin_unlock(&last_ptr->lock);
7536 search_start = max(search_start, first_logical_byte(fs_info, 0));
7537 search_start = max(search_start, hint_byte);
7538 if (search_start == hint_byte) {
7539 block_group = btrfs_lookup_block_group(fs_info, search_start);
7541 * we don't want to use the block group if it doesn't match our
7542 * allocation bits, or if its not cached.
7544 * However if we are re-searching with an ideal block group
7545 * picked out then we don't care that the block group is cached.
7547 if (block_group && block_group_bits(block_group, flags) &&
7548 block_group->cached != BTRFS_CACHE_NO) {
7549 down_read(&space_info->groups_sem);
7550 if (list_empty(&block_group->list) ||
7553 * someone is removing this block group,
7554 * we can't jump into the have_block_group
7555 * target because our list pointers are not
7558 btrfs_put_block_group(block_group);
7559 up_read(&space_info->groups_sem);
7561 index = get_block_group_index(block_group);
7562 btrfs_lock_block_group(block_group, delalloc);
7563 goto have_block_group;
7565 } else if (block_group) {
7566 btrfs_put_block_group(block_group);
7570 have_caching_bg = false;
7571 if (index == 0 || index == __get_raid_index(flags))
7573 down_read(&space_info->groups_sem);
7574 list_for_each_entry(block_group, &space_info->block_groups[index],
7579 /* If the block group is read-only, we can skip it entirely. */
7580 if (unlikely(block_group->ro))
7583 btrfs_grab_block_group(block_group, delalloc);
7584 search_start = block_group->key.objectid;
7587 * this can happen if we end up cycling through all the
7588 * raid types, but we want to make sure we only allocate
7589 * for the proper type.
7591 if (!block_group_bits(block_group, flags)) {
7592 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7593 BTRFS_BLOCK_GROUP_RAID1 |
7594 BTRFS_BLOCK_GROUP_RAID5 |
7595 BTRFS_BLOCK_GROUP_RAID6 |
7596 BTRFS_BLOCK_GROUP_RAID10;
7599 * if they asked for extra copies and this block group
7600 * doesn't provide them, bail. This does allow us to
7601 * fill raid0 from raid1.
7603 if ((flags & extra) && !(block_group->flags & extra))
7608 cached = block_group_cache_done(block_group);
7609 if (unlikely(!cached)) {
7610 have_caching_bg = true;
7611 ret = cache_block_group(block_group, 0);
7616 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7620 * Ok we want to try and use the cluster allocator, so
7623 if (last_ptr && use_cluster) {
7624 struct btrfs_block_group_cache *used_block_group;
7625 unsigned long aligned_cluster;
7627 * the refill lock keeps out other
7628 * people trying to start a new cluster
7630 used_block_group = btrfs_lock_cluster(block_group,
7633 if (!used_block_group)
7634 goto refill_cluster;
7636 if (used_block_group != block_group &&
7637 (used_block_group->ro ||
7638 !block_group_bits(used_block_group, flags)))
7639 goto release_cluster;
7641 offset = btrfs_alloc_from_cluster(used_block_group,
7644 used_block_group->key.objectid,
7647 /* we have a block, we're done */
7648 spin_unlock(&last_ptr->refill_lock);
7649 trace_btrfs_reserve_extent_cluster(fs_info,
7651 search_start, num_bytes);
7652 if (used_block_group != block_group) {
7653 btrfs_release_block_group(block_group,
7655 block_group = used_block_group;
7660 WARN_ON(last_ptr->block_group != used_block_group);
7662 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7663 * set up a new clusters, so lets just skip it
7664 * and let the allocator find whatever block
7665 * it can find. If we reach this point, we
7666 * will have tried the cluster allocator
7667 * plenty of times and not have found
7668 * anything, so we are likely way too
7669 * fragmented for the clustering stuff to find
7672 * However, if the cluster is taken from the
7673 * current block group, release the cluster
7674 * first, so that we stand a better chance of
7675 * succeeding in the unclustered
7677 if (loop >= LOOP_NO_EMPTY_SIZE &&
7678 used_block_group != block_group) {
7679 spin_unlock(&last_ptr->refill_lock);
7680 btrfs_release_block_group(used_block_group,
7682 goto unclustered_alloc;
7686 * this cluster didn't work out, free it and
7689 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7691 if (used_block_group != block_group)
7692 btrfs_release_block_group(used_block_group,
7695 if (loop >= LOOP_NO_EMPTY_SIZE) {
7696 spin_unlock(&last_ptr->refill_lock);
7697 goto unclustered_alloc;
7700 aligned_cluster = max_t(unsigned long,
7701 empty_cluster + empty_size,
7702 block_group->full_stripe_len);
7704 /* allocate a cluster in this block group */
7705 ret = btrfs_find_space_cluster(fs_info, block_group,
7706 last_ptr, search_start,
7711 * now pull our allocation out of this
7714 offset = btrfs_alloc_from_cluster(block_group,
7720 /* we found one, proceed */
7721 spin_unlock(&last_ptr->refill_lock);
7722 trace_btrfs_reserve_extent_cluster(fs_info,
7723 block_group, search_start,
7727 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7728 && !failed_cluster_refill) {
7729 spin_unlock(&last_ptr->refill_lock);
7731 failed_cluster_refill = true;
7732 wait_block_group_cache_progress(block_group,
7733 num_bytes + empty_cluster + empty_size);
7734 goto have_block_group;
7738 * at this point we either didn't find a cluster
7739 * or we weren't able to allocate a block from our
7740 * cluster. Free the cluster we've been trying
7741 * to use, and go to the next block group
7743 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7744 spin_unlock(&last_ptr->refill_lock);
7750 * We are doing an unclustered alloc, set the fragmented flag so
7751 * we don't bother trying to setup a cluster again until we get
7754 if (unlikely(last_ptr)) {
7755 spin_lock(&last_ptr->lock);
7756 last_ptr->fragmented = 1;
7757 spin_unlock(&last_ptr->lock);
7760 struct btrfs_free_space_ctl *ctl =
7761 block_group->free_space_ctl;
7763 spin_lock(&ctl->tree_lock);
7764 if (ctl->free_space <
7765 num_bytes + empty_cluster + empty_size) {
7766 if (ctl->free_space > max_extent_size)
7767 max_extent_size = ctl->free_space;
7768 spin_unlock(&ctl->tree_lock);
7771 spin_unlock(&ctl->tree_lock);
7774 offset = btrfs_find_space_for_alloc(block_group, search_start,
7775 num_bytes, empty_size,
7778 * If we didn't find a chunk, and we haven't failed on this
7779 * block group before, and this block group is in the middle of
7780 * caching and we are ok with waiting, then go ahead and wait
7781 * for progress to be made, and set failed_alloc to true.
7783 * If failed_alloc is true then we've already waited on this
7784 * block group once and should move on to the next block group.
7786 if (!offset && !failed_alloc && !cached &&
7787 loop > LOOP_CACHING_NOWAIT) {
7788 wait_block_group_cache_progress(block_group,
7789 num_bytes + empty_size);
7790 failed_alloc = true;
7791 goto have_block_group;
7792 } else if (!offset) {
7796 search_start = ALIGN(offset, fs_info->stripesize);
7798 /* move on to the next group */
7799 if (search_start + num_bytes >
7800 block_group->key.objectid + block_group->key.offset) {
7801 btrfs_add_free_space(block_group, offset, num_bytes);
7805 if (offset < search_start)
7806 btrfs_add_free_space(block_group, offset,
7807 search_start - offset);
7808 BUG_ON(offset > search_start);
7810 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7811 num_bytes, delalloc);
7812 if (ret == -EAGAIN) {
7813 btrfs_add_free_space(block_group, offset, num_bytes);
7816 btrfs_inc_block_group_reservations(block_group);
7818 /* we are all good, lets return */
7819 ins->objectid = search_start;
7820 ins->offset = num_bytes;
7822 trace_btrfs_reserve_extent(fs_info, block_group,
7823 search_start, num_bytes);
7824 btrfs_release_block_group(block_group, delalloc);
7827 failed_cluster_refill = false;
7828 failed_alloc = false;
7829 BUG_ON(index != get_block_group_index(block_group));
7830 btrfs_release_block_group(block_group, delalloc);
7833 up_read(&space_info->groups_sem);
7835 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7836 && !orig_have_caching_bg)
7837 orig_have_caching_bg = true;
7839 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7842 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7846 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7847 * caching kthreads as we move along
7848 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7849 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7850 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7853 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7855 if (loop == LOOP_CACHING_NOWAIT) {
7857 * We want to skip the LOOP_CACHING_WAIT step if we
7858 * don't have any uncached bgs and we've already done a
7859 * full search through.
7861 if (orig_have_caching_bg || !full_search)
7862 loop = LOOP_CACHING_WAIT;
7864 loop = LOOP_ALLOC_CHUNK;
7869 if (loop == LOOP_ALLOC_CHUNK) {
7870 struct btrfs_trans_handle *trans;
7873 trans = current->journal_info;
7877 trans = btrfs_join_transaction(root);
7879 if (IS_ERR(trans)) {
7880 ret = PTR_ERR(trans);
7884 ret = do_chunk_alloc(trans, fs_info, flags,
7888 * If we can't allocate a new chunk we've already looped
7889 * through at least once, move on to the NO_EMPTY_SIZE
7893 loop = LOOP_NO_EMPTY_SIZE;
7896 * Do not bail out on ENOSPC since we
7897 * can do more things.
7899 if (ret < 0 && ret != -ENOSPC)
7900 btrfs_abort_transaction(trans, ret);
7904 btrfs_end_transaction(trans);
7909 if (loop == LOOP_NO_EMPTY_SIZE) {
7911 * Don't loop again if we already have no empty_size and
7914 if (empty_size == 0 &&
7915 empty_cluster == 0) {
7924 } else if (!ins->objectid) {
7926 } else if (ins->objectid) {
7927 if (!use_cluster && last_ptr) {
7928 spin_lock(&last_ptr->lock);
7929 last_ptr->window_start = ins->objectid;
7930 spin_unlock(&last_ptr->lock);
7935 if (ret == -ENOSPC) {
7936 spin_lock(&space_info->lock);
7937 space_info->max_extent_size = max_extent_size;
7938 spin_unlock(&space_info->lock);
7939 ins->offset = max_extent_size;
7944 static void dump_space_info(struct btrfs_fs_info *fs_info,
7945 struct btrfs_space_info *info, u64 bytes,
7946 int dump_block_groups)
7948 struct btrfs_block_group_cache *cache;
7951 spin_lock(&info->lock);
7952 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7954 info->total_bytes - btrfs_space_info_used(info, true),
7955 info->full ? "" : "not ");
7957 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7958 info->total_bytes, info->bytes_used, info->bytes_pinned,
7959 info->bytes_reserved, info->bytes_may_use,
7960 info->bytes_readonly);
7961 spin_unlock(&info->lock);
7963 if (!dump_block_groups)
7966 down_read(&info->groups_sem);
7968 list_for_each_entry(cache, &info->block_groups[index], list) {
7969 spin_lock(&cache->lock);
7971 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7972 cache->key.objectid, cache->key.offset,
7973 btrfs_block_group_used(&cache->item), cache->pinned,
7974 cache->reserved, cache->ro ? "[readonly]" : "");
7975 btrfs_dump_free_space(cache, bytes);
7976 spin_unlock(&cache->lock);
7978 if (++index < BTRFS_NR_RAID_TYPES)
7980 up_read(&info->groups_sem);
7983 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7984 u64 num_bytes, u64 min_alloc_size,
7985 u64 empty_size, u64 hint_byte,
7986 struct btrfs_key *ins, int is_data, int delalloc)
7988 struct btrfs_fs_info *fs_info = root->fs_info;
7989 bool final_tried = num_bytes == min_alloc_size;
7993 flags = get_alloc_profile_by_root(root, is_data);
7995 WARN_ON(num_bytes < fs_info->sectorsize);
7996 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
7997 hint_byte, ins, flags, delalloc);
7998 if (!ret && !is_data) {
7999 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8000 } else if (ret == -ENOSPC) {
8001 if (!final_tried && ins->offset) {
8002 num_bytes = min(num_bytes >> 1, ins->offset);
8003 num_bytes = round_down(num_bytes,
8004 fs_info->sectorsize);
8005 num_bytes = max(num_bytes, min_alloc_size);
8006 ram_bytes = num_bytes;
8007 if (num_bytes == min_alloc_size)
8010 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8011 struct btrfs_space_info *sinfo;
8013 sinfo = __find_space_info(fs_info, flags);
8015 "allocation failed flags %llu, wanted %llu",
8018 dump_space_info(fs_info, sinfo, num_bytes, 1);
8025 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8027 int pin, int delalloc)
8029 struct btrfs_block_group_cache *cache;
8032 cache = btrfs_lookup_block_group(fs_info, start);
8034 btrfs_err(fs_info, "Unable to find block group for %llu",
8040 pin_down_extent(fs_info, cache, start, len, 1);
8042 if (btrfs_test_opt(fs_info, DISCARD))
8043 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8044 btrfs_add_free_space(cache, start, len);
8045 btrfs_free_reserved_bytes(cache, len, delalloc);
8046 trace_btrfs_reserved_extent_free(fs_info, start, len);
8049 btrfs_put_block_group(cache);
8053 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8054 u64 start, u64 len, int delalloc)
8056 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8059 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8062 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8065 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8066 struct btrfs_fs_info *fs_info,
8067 u64 parent, u64 root_objectid,
8068 u64 flags, u64 owner, u64 offset,
8069 struct btrfs_key *ins, int ref_mod)
8072 struct btrfs_extent_item *extent_item;
8073 struct btrfs_extent_inline_ref *iref;
8074 struct btrfs_path *path;
8075 struct extent_buffer *leaf;
8080 type = BTRFS_SHARED_DATA_REF_KEY;
8082 type = BTRFS_EXTENT_DATA_REF_KEY;
8084 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8086 path = btrfs_alloc_path();
8090 path->leave_spinning = 1;
8091 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8094 btrfs_free_path(path);
8098 leaf = path->nodes[0];
8099 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8100 struct btrfs_extent_item);
8101 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8102 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8103 btrfs_set_extent_flags(leaf, extent_item,
8104 flags | BTRFS_EXTENT_FLAG_DATA);
8106 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8107 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8109 struct btrfs_shared_data_ref *ref;
8110 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8111 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8112 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8114 struct btrfs_extent_data_ref *ref;
8115 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8116 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8117 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8118 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8119 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8122 btrfs_mark_buffer_dirty(path->nodes[0]);
8123 btrfs_free_path(path);
8125 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8130 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8131 if (ret) { /* -ENOENT, logic error */
8132 btrfs_err(fs_info, "update block group failed for %llu %llu",
8133 ins->objectid, ins->offset);
8136 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8140 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8141 struct btrfs_fs_info *fs_info,
8142 u64 parent, u64 root_objectid,
8143 u64 flags, struct btrfs_disk_key *key,
8144 int level, struct btrfs_key *ins)
8147 struct btrfs_extent_item *extent_item;
8148 struct btrfs_tree_block_info *block_info;
8149 struct btrfs_extent_inline_ref *iref;
8150 struct btrfs_path *path;
8151 struct extent_buffer *leaf;
8152 u32 size = sizeof(*extent_item) + sizeof(*iref);
8153 u64 num_bytes = ins->offset;
8154 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8156 if (!skinny_metadata)
8157 size += sizeof(*block_info);
8159 path = btrfs_alloc_path();
8161 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8166 path->leave_spinning = 1;
8167 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8170 btrfs_free_path(path);
8171 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8176 leaf = path->nodes[0];
8177 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8178 struct btrfs_extent_item);
8179 btrfs_set_extent_refs(leaf, extent_item, 1);
8180 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8181 btrfs_set_extent_flags(leaf, extent_item,
8182 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8184 if (skinny_metadata) {
8185 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8186 num_bytes = fs_info->nodesize;
8188 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8189 btrfs_set_tree_block_key(leaf, block_info, key);
8190 btrfs_set_tree_block_level(leaf, block_info, level);
8191 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8195 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8196 btrfs_set_extent_inline_ref_type(leaf, iref,
8197 BTRFS_SHARED_BLOCK_REF_KEY);
8198 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8200 btrfs_set_extent_inline_ref_type(leaf, iref,
8201 BTRFS_TREE_BLOCK_REF_KEY);
8202 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8205 btrfs_mark_buffer_dirty(leaf);
8206 btrfs_free_path(path);
8208 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8213 ret = update_block_group(trans, fs_info, ins->objectid,
8214 fs_info->nodesize, 1);
8215 if (ret) { /* -ENOENT, logic error */
8216 btrfs_err(fs_info, "update block group failed for %llu %llu",
8217 ins->objectid, ins->offset);
8221 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8226 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8227 u64 root_objectid, u64 owner,
8228 u64 offset, u64 ram_bytes,
8229 struct btrfs_key *ins)
8231 struct btrfs_fs_info *fs_info = trans->fs_info;
8234 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8236 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8237 ins->offset, 0, root_objectid, owner,
8239 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8244 * this is used by the tree logging recovery code. It records that
8245 * an extent has been allocated and makes sure to clear the free
8246 * space cache bits as well
8248 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8249 struct btrfs_fs_info *fs_info,
8250 u64 root_objectid, u64 owner, u64 offset,
8251 struct btrfs_key *ins)
8254 struct btrfs_block_group_cache *block_group;
8255 struct btrfs_space_info *space_info;
8258 * Mixed block groups will exclude before processing the log so we only
8259 * need to do the exclude dance if this fs isn't mixed.
8261 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8262 ret = __exclude_logged_extent(fs_info, ins->objectid,
8268 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8272 space_info = block_group->space_info;
8273 spin_lock(&space_info->lock);
8274 spin_lock(&block_group->lock);
8275 space_info->bytes_reserved += ins->offset;
8276 block_group->reserved += ins->offset;
8277 spin_unlock(&block_group->lock);
8278 spin_unlock(&space_info->lock);
8280 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8281 0, owner, offset, ins, 1);
8282 btrfs_put_block_group(block_group);
8286 static struct extent_buffer *
8287 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8288 u64 bytenr, int level)
8290 struct btrfs_fs_info *fs_info = root->fs_info;
8291 struct extent_buffer *buf;
8293 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8297 btrfs_set_header_generation(buf, trans->transid);
8298 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8299 btrfs_tree_lock(buf);
8300 clean_tree_block(fs_info, buf);
8301 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8303 btrfs_set_lock_blocking(buf);
8304 set_extent_buffer_uptodate(buf);
8306 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8307 buf->log_index = root->log_transid % 2;
8309 * we allow two log transactions at a time, use different
8310 * EXENT bit to differentiate dirty pages.
8312 if (buf->log_index == 0)
8313 set_extent_dirty(&root->dirty_log_pages, buf->start,
8314 buf->start + buf->len - 1, GFP_NOFS);
8316 set_extent_new(&root->dirty_log_pages, buf->start,
8317 buf->start + buf->len - 1);
8319 buf->log_index = -1;
8320 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8321 buf->start + buf->len - 1, GFP_NOFS);
8323 trans->dirty = true;
8324 /* this returns a buffer locked for blocking */
8328 static struct btrfs_block_rsv *
8329 use_block_rsv(struct btrfs_trans_handle *trans,
8330 struct btrfs_root *root, u32 blocksize)
8332 struct btrfs_fs_info *fs_info = root->fs_info;
8333 struct btrfs_block_rsv *block_rsv;
8334 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8336 bool global_updated = false;
8338 block_rsv = get_block_rsv(trans, root);
8340 if (unlikely(block_rsv->size == 0))
8343 ret = block_rsv_use_bytes(block_rsv, blocksize);
8347 if (block_rsv->failfast)
8348 return ERR_PTR(ret);
8350 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8351 global_updated = true;
8352 update_global_block_rsv(fs_info);
8356 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8357 static DEFINE_RATELIMIT_STATE(_rs,
8358 DEFAULT_RATELIMIT_INTERVAL * 10,
8359 /*DEFAULT_RATELIMIT_BURST*/ 1);
8360 if (__ratelimit(&_rs))
8362 "BTRFS: block rsv returned %d\n", ret);
8365 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8366 BTRFS_RESERVE_NO_FLUSH);
8370 * If we couldn't reserve metadata bytes try and use some from
8371 * the global reserve if its space type is the same as the global
8374 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8375 block_rsv->space_info == global_rsv->space_info) {
8376 ret = block_rsv_use_bytes(global_rsv, blocksize);
8380 return ERR_PTR(ret);
8383 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8384 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8386 block_rsv_add_bytes(block_rsv, blocksize, 0);
8387 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8391 * finds a free extent and does all the dirty work required for allocation
8392 * returns the tree buffer or an ERR_PTR on error.
8394 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8395 struct btrfs_root *root,
8396 u64 parent, u64 root_objectid,
8397 const struct btrfs_disk_key *key,
8398 int level, u64 hint,
8401 struct btrfs_fs_info *fs_info = root->fs_info;
8402 struct btrfs_key ins;
8403 struct btrfs_block_rsv *block_rsv;
8404 struct extent_buffer *buf;
8405 struct btrfs_delayed_extent_op *extent_op;
8408 u32 blocksize = fs_info->nodesize;
8409 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8411 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8412 if (btrfs_is_testing(fs_info)) {
8413 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8416 root->alloc_bytenr += blocksize;
8421 block_rsv = use_block_rsv(trans, root, blocksize);
8422 if (IS_ERR(block_rsv))
8423 return ERR_CAST(block_rsv);
8425 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8426 empty_size, hint, &ins, 0, 0);
8430 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8433 goto out_free_reserved;
8436 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8438 parent = ins.objectid;
8439 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8443 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8444 extent_op = btrfs_alloc_delayed_extent_op();
8450 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8452 memset(&extent_op->key, 0, sizeof(extent_op->key));
8453 extent_op->flags_to_set = flags;
8454 extent_op->update_key = skinny_metadata ? false : true;
8455 extent_op->update_flags = true;
8456 extent_op->is_data = false;
8457 extent_op->level = level;
8459 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8461 root_objectid, level,
8462 BTRFS_ADD_DELAYED_EXTENT,
8463 extent_op, NULL, NULL);
8465 goto out_free_delayed;
8470 btrfs_free_delayed_extent_op(extent_op);
8472 free_extent_buffer(buf);
8474 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8476 unuse_block_rsv(fs_info, block_rsv, blocksize);
8477 return ERR_PTR(ret);
8480 struct walk_control {
8481 u64 refs[BTRFS_MAX_LEVEL];
8482 u64 flags[BTRFS_MAX_LEVEL];
8483 struct btrfs_key update_progress;
8494 #define DROP_REFERENCE 1
8495 #define UPDATE_BACKREF 2
8497 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8498 struct btrfs_root *root,
8499 struct walk_control *wc,
8500 struct btrfs_path *path)
8502 struct btrfs_fs_info *fs_info = root->fs_info;
8508 struct btrfs_key key;
8509 struct extent_buffer *eb;
8514 if (path->slots[wc->level] < wc->reada_slot) {
8515 wc->reada_count = wc->reada_count * 2 / 3;
8516 wc->reada_count = max(wc->reada_count, 2);
8518 wc->reada_count = wc->reada_count * 3 / 2;
8519 wc->reada_count = min_t(int, wc->reada_count,
8520 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8523 eb = path->nodes[wc->level];
8524 nritems = btrfs_header_nritems(eb);
8526 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8527 if (nread >= wc->reada_count)
8531 bytenr = btrfs_node_blockptr(eb, slot);
8532 generation = btrfs_node_ptr_generation(eb, slot);
8534 if (slot == path->slots[wc->level])
8537 if (wc->stage == UPDATE_BACKREF &&
8538 generation <= root->root_key.offset)
8541 /* We don't lock the tree block, it's OK to be racy here */
8542 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8543 wc->level - 1, 1, &refs,
8545 /* We don't care about errors in readahead. */
8550 if (wc->stage == DROP_REFERENCE) {
8554 if (wc->level == 1 &&
8555 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8557 if (!wc->update_ref ||
8558 generation <= root->root_key.offset)
8560 btrfs_node_key_to_cpu(eb, &key, slot);
8561 ret = btrfs_comp_cpu_keys(&key,
8562 &wc->update_progress);
8566 if (wc->level == 1 &&
8567 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8571 readahead_tree_block(fs_info, bytenr);
8574 wc->reada_slot = slot;
8578 * helper to process tree block while walking down the tree.
8580 * when wc->stage == UPDATE_BACKREF, this function updates
8581 * back refs for pointers in the block.
8583 * NOTE: return value 1 means we should stop walking down.
8585 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8586 struct btrfs_root *root,
8587 struct btrfs_path *path,
8588 struct walk_control *wc, int lookup_info)
8590 struct btrfs_fs_info *fs_info = root->fs_info;
8591 int level = wc->level;
8592 struct extent_buffer *eb = path->nodes[level];
8593 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8596 if (wc->stage == UPDATE_BACKREF &&
8597 btrfs_header_owner(eb) != root->root_key.objectid)
8601 * when reference count of tree block is 1, it won't increase
8602 * again. once full backref flag is set, we never clear it.
8605 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8606 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8607 BUG_ON(!path->locks[level]);
8608 ret = btrfs_lookup_extent_info(trans, fs_info,
8609 eb->start, level, 1,
8612 BUG_ON(ret == -ENOMEM);
8615 BUG_ON(wc->refs[level] == 0);
8618 if (wc->stage == DROP_REFERENCE) {
8619 if (wc->refs[level] > 1)
8622 if (path->locks[level] && !wc->keep_locks) {
8623 btrfs_tree_unlock_rw(eb, path->locks[level]);
8624 path->locks[level] = 0;
8629 /* wc->stage == UPDATE_BACKREF */
8630 if (!(wc->flags[level] & flag)) {
8631 BUG_ON(!path->locks[level]);
8632 ret = btrfs_inc_ref(trans, root, eb, 1);
8633 BUG_ON(ret); /* -ENOMEM */
8634 ret = btrfs_dec_ref(trans, root, eb, 0);
8635 BUG_ON(ret); /* -ENOMEM */
8636 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8638 btrfs_header_level(eb), 0);
8639 BUG_ON(ret); /* -ENOMEM */
8640 wc->flags[level] |= flag;
8644 * the block is shared by multiple trees, so it's not good to
8645 * keep the tree lock
8647 if (path->locks[level] && level > 0) {
8648 btrfs_tree_unlock_rw(eb, path->locks[level]);
8649 path->locks[level] = 0;
8655 * helper to process tree block pointer.
8657 * when wc->stage == DROP_REFERENCE, this function checks
8658 * reference count of the block pointed to. if the block
8659 * is shared and we need update back refs for the subtree
8660 * rooted at the block, this function changes wc->stage to
8661 * UPDATE_BACKREF. if the block is shared and there is no
8662 * need to update back, this function drops the reference
8665 * NOTE: return value 1 means we should stop walking down.
8667 static noinline int do_walk_down(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;
8677 struct btrfs_key key;
8678 struct extent_buffer *next;
8679 int level = wc->level;
8682 bool need_account = false;
8684 generation = btrfs_node_ptr_generation(path->nodes[level],
8685 path->slots[level]);
8687 * if the lower level block was created before the snapshot
8688 * was created, we know there is no need to update back refs
8691 if (wc->stage == UPDATE_BACKREF &&
8692 generation <= root->root_key.offset) {
8697 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8698 blocksize = fs_info->nodesize;
8700 next = find_extent_buffer(fs_info, bytenr);
8702 next = btrfs_find_create_tree_block(fs_info, bytenr);
8704 return PTR_ERR(next);
8706 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8710 btrfs_tree_lock(next);
8711 btrfs_set_lock_blocking(next);
8713 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8714 &wc->refs[level - 1],
8715 &wc->flags[level - 1]);
8719 if (unlikely(wc->refs[level - 1] == 0)) {
8720 btrfs_err(fs_info, "Missing references.");
8726 if (wc->stage == DROP_REFERENCE) {
8727 if (wc->refs[level - 1] > 1) {
8728 need_account = true;
8730 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8733 if (!wc->update_ref ||
8734 generation <= root->root_key.offset)
8737 btrfs_node_key_to_cpu(path->nodes[level], &key,
8738 path->slots[level]);
8739 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8743 wc->stage = UPDATE_BACKREF;
8744 wc->shared_level = level - 1;
8748 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8752 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8753 btrfs_tree_unlock(next);
8754 free_extent_buffer(next);
8760 if (reada && level == 1)
8761 reada_walk_down(trans, root, wc, path);
8762 next = read_tree_block(fs_info, bytenr, generation);
8764 return PTR_ERR(next);
8765 } else if (!extent_buffer_uptodate(next)) {
8766 free_extent_buffer(next);
8769 btrfs_tree_lock(next);
8770 btrfs_set_lock_blocking(next);
8774 ASSERT(level == btrfs_header_level(next));
8775 if (level != btrfs_header_level(next)) {
8776 btrfs_err(root->fs_info, "mismatched level");
8780 path->nodes[level] = next;
8781 path->slots[level] = 0;
8782 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8788 wc->refs[level - 1] = 0;
8789 wc->flags[level - 1] = 0;
8790 if (wc->stage == DROP_REFERENCE) {
8791 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8792 parent = path->nodes[level]->start;
8794 ASSERT(root->root_key.objectid ==
8795 btrfs_header_owner(path->nodes[level]));
8796 if (root->root_key.objectid !=
8797 btrfs_header_owner(path->nodes[level])) {
8798 btrfs_err(root->fs_info,
8799 "mismatched block owner");
8807 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8808 generation, level - 1);
8810 btrfs_err_rl(fs_info,
8811 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8815 ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize,
8816 parent, root->root_key.objectid,
8826 btrfs_tree_unlock(next);
8827 free_extent_buffer(next);
8833 * helper to process tree block while walking up the tree.
8835 * when wc->stage == DROP_REFERENCE, this function drops
8836 * reference count on the block.
8838 * when wc->stage == UPDATE_BACKREF, this function changes
8839 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8840 * to UPDATE_BACKREF previously while processing the block.
8842 * NOTE: return value 1 means we should stop walking up.
8844 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8845 struct btrfs_root *root,
8846 struct btrfs_path *path,
8847 struct walk_control *wc)
8849 struct btrfs_fs_info *fs_info = root->fs_info;
8851 int level = wc->level;
8852 struct extent_buffer *eb = path->nodes[level];
8855 if (wc->stage == UPDATE_BACKREF) {
8856 BUG_ON(wc->shared_level < level);
8857 if (level < wc->shared_level)
8860 ret = find_next_key(path, level + 1, &wc->update_progress);
8864 wc->stage = DROP_REFERENCE;
8865 wc->shared_level = -1;
8866 path->slots[level] = 0;
8869 * check reference count again if the block isn't locked.
8870 * we should start walking down the tree again if reference
8873 if (!path->locks[level]) {
8875 btrfs_tree_lock(eb);
8876 btrfs_set_lock_blocking(eb);
8877 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8879 ret = btrfs_lookup_extent_info(trans, fs_info,
8880 eb->start, level, 1,
8884 btrfs_tree_unlock_rw(eb, path->locks[level]);
8885 path->locks[level] = 0;
8888 BUG_ON(wc->refs[level] == 0);
8889 if (wc->refs[level] == 1) {
8890 btrfs_tree_unlock_rw(eb, path->locks[level]);
8891 path->locks[level] = 0;
8897 /* wc->stage == DROP_REFERENCE */
8898 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8900 if (wc->refs[level] == 1) {
8902 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8903 ret = btrfs_dec_ref(trans, root, eb, 1);
8905 ret = btrfs_dec_ref(trans, root, eb, 0);
8906 BUG_ON(ret); /* -ENOMEM */
8907 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8909 btrfs_err_rl(fs_info,
8910 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8914 /* make block locked assertion in clean_tree_block happy */
8915 if (!path->locks[level] &&
8916 btrfs_header_generation(eb) == trans->transid) {
8917 btrfs_tree_lock(eb);
8918 btrfs_set_lock_blocking(eb);
8919 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8921 clean_tree_block(fs_info, eb);
8924 if (eb == root->node) {
8925 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8928 BUG_ON(root->root_key.objectid !=
8929 btrfs_header_owner(eb));
8931 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8932 parent = path->nodes[level + 1]->start;
8934 BUG_ON(root->root_key.objectid !=
8935 btrfs_header_owner(path->nodes[level + 1]));
8938 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8940 wc->refs[level] = 0;
8941 wc->flags[level] = 0;
8945 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8946 struct btrfs_root *root,
8947 struct btrfs_path *path,
8948 struct walk_control *wc)
8950 int level = wc->level;
8951 int lookup_info = 1;
8954 while (level >= 0) {
8955 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8962 if (path->slots[level] >=
8963 btrfs_header_nritems(path->nodes[level]))
8966 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8968 path->slots[level]++;
8977 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8978 struct btrfs_root *root,
8979 struct btrfs_path *path,
8980 struct walk_control *wc, int max_level)
8982 int level = wc->level;
8985 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8986 while (level < max_level && path->nodes[level]) {
8988 if (path->slots[level] + 1 <
8989 btrfs_header_nritems(path->nodes[level])) {
8990 path->slots[level]++;
8993 ret = walk_up_proc(trans, root, path, wc);
8997 if (path->locks[level]) {
8998 btrfs_tree_unlock_rw(path->nodes[level],
8999 path->locks[level]);
9000 path->locks[level] = 0;
9002 free_extent_buffer(path->nodes[level]);
9003 path->nodes[level] = NULL;
9011 * drop a subvolume tree.
9013 * this function traverses the tree freeing any blocks that only
9014 * referenced by the tree.
9016 * when a shared tree block is found. this function decreases its
9017 * reference count by one. if update_ref is true, this function
9018 * also make sure backrefs for the shared block and all lower level
9019 * blocks are properly updated.
9021 * If called with for_reloc == 0, may exit early with -EAGAIN
9023 int btrfs_drop_snapshot(struct btrfs_root *root,
9024 struct btrfs_block_rsv *block_rsv, int update_ref,
9027 struct btrfs_fs_info *fs_info = root->fs_info;
9028 struct btrfs_path *path;
9029 struct btrfs_trans_handle *trans;
9030 struct btrfs_root *tree_root = fs_info->tree_root;
9031 struct btrfs_root_item *root_item = &root->root_item;
9032 struct walk_control *wc;
9033 struct btrfs_key key;
9037 bool root_dropped = false;
9039 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9041 path = btrfs_alloc_path();
9047 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9049 btrfs_free_path(path);
9054 trans = btrfs_start_transaction(tree_root, 0);
9055 if (IS_ERR(trans)) {
9056 err = PTR_ERR(trans);
9061 trans->block_rsv = block_rsv;
9063 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9064 level = btrfs_header_level(root->node);
9065 path->nodes[level] = btrfs_lock_root_node(root);
9066 btrfs_set_lock_blocking(path->nodes[level]);
9067 path->slots[level] = 0;
9068 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9069 memset(&wc->update_progress, 0,
9070 sizeof(wc->update_progress));
9072 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9073 memcpy(&wc->update_progress, &key,
9074 sizeof(wc->update_progress));
9076 level = root_item->drop_level;
9078 path->lowest_level = level;
9079 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9080 path->lowest_level = 0;
9088 * unlock our path, this is safe because only this
9089 * function is allowed to delete this snapshot
9091 btrfs_unlock_up_safe(path, 0);
9093 level = btrfs_header_level(root->node);
9095 btrfs_tree_lock(path->nodes[level]);
9096 btrfs_set_lock_blocking(path->nodes[level]);
9097 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9099 ret = btrfs_lookup_extent_info(trans, fs_info,
9100 path->nodes[level]->start,
9101 level, 1, &wc->refs[level],
9107 BUG_ON(wc->refs[level] == 0);
9109 if (level == root_item->drop_level)
9112 btrfs_tree_unlock(path->nodes[level]);
9113 path->locks[level] = 0;
9114 WARN_ON(wc->refs[level] != 1);
9120 wc->shared_level = -1;
9121 wc->stage = DROP_REFERENCE;
9122 wc->update_ref = update_ref;
9124 wc->for_reloc = for_reloc;
9125 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9129 ret = walk_down_tree(trans, root, path, wc);
9135 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9142 BUG_ON(wc->stage != DROP_REFERENCE);
9146 if (wc->stage == DROP_REFERENCE) {
9148 btrfs_node_key(path->nodes[level],
9149 &root_item->drop_progress,
9150 path->slots[level]);
9151 root_item->drop_level = level;
9154 BUG_ON(wc->level == 0);
9155 if (btrfs_should_end_transaction(trans) ||
9156 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9157 ret = btrfs_update_root(trans, tree_root,
9161 btrfs_abort_transaction(trans, ret);
9166 btrfs_end_transaction_throttle(trans);
9167 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9168 btrfs_debug(fs_info,
9169 "drop snapshot early exit");
9174 trans = btrfs_start_transaction(tree_root, 0);
9175 if (IS_ERR(trans)) {
9176 err = PTR_ERR(trans);
9180 trans->block_rsv = block_rsv;
9183 btrfs_release_path(path);
9187 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9189 btrfs_abort_transaction(trans, ret);
9193 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9194 ret = btrfs_find_root(tree_root, &root->root_key, path,
9197 btrfs_abort_transaction(trans, ret);
9200 } else if (ret > 0) {
9201 /* if we fail to delete the orphan item this time
9202 * around, it'll get picked up the next time.
9204 * The most common failure here is just -ENOENT.
9206 btrfs_del_orphan_item(trans, tree_root,
9207 root->root_key.objectid);
9211 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9212 btrfs_add_dropped_root(trans, root);
9214 free_extent_buffer(root->node);
9215 free_extent_buffer(root->commit_root);
9216 btrfs_put_fs_root(root);
9218 root_dropped = true;
9220 btrfs_end_transaction_throttle(trans);
9223 btrfs_free_path(path);
9226 * So if we need to stop dropping the snapshot for whatever reason we
9227 * need to make sure to add it back to the dead root list so that we
9228 * keep trying to do the work later. This also cleans up roots if we
9229 * don't have it in the radix (like when we recover after a power fail
9230 * or unmount) so we don't leak memory.
9232 if (!for_reloc && root_dropped == false)
9233 btrfs_add_dead_root(root);
9234 if (err && err != -EAGAIN)
9235 btrfs_handle_fs_error(fs_info, err, NULL);
9240 * drop subtree rooted at tree block 'node'.
9242 * NOTE: this function will unlock and release tree block 'node'
9243 * only used by relocation code
9245 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9246 struct btrfs_root *root,
9247 struct extent_buffer *node,
9248 struct extent_buffer *parent)
9250 struct btrfs_fs_info *fs_info = root->fs_info;
9251 struct btrfs_path *path;
9252 struct walk_control *wc;
9258 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9260 path = btrfs_alloc_path();
9264 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9266 btrfs_free_path(path);
9270 btrfs_assert_tree_locked(parent);
9271 parent_level = btrfs_header_level(parent);
9272 extent_buffer_get(parent);
9273 path->nodes[parent_level] = parent;
9274 path->slots[parent_level] = btrfs_header_nritems(parent);
9276 btrfs_assert_tree_locked(node);
9277 level = btrfs_header_level(node);
9278 path->nodes[level] = node;
9279 path->slots[level] = 0;
9280 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9282 wc->refs[parent_level] = 1;
9283 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9285 wc->shared_level = -1;
9286 wc->stage = DROP_REFERENCE;
9290 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9293 wret = walk_down_tree(trans, root, path, wc);
9299 wret = walk_up_tree(trans, root, path, wc, parent_level);
9307 btrfs_free_path(path);
9311 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9317 * if restripe for this chunk_type is on pick target profile and
9318 * return, otherwise do the usual balance
9320 stripped = get_restripe_target(fs_info, flags);
9322 return extended_to_chunk(stripped);
9324 num_devices = fs_info->fs_devices->rw_devices;
9326 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9327 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9328 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9330 if (num_devices == 1) {
9331 stripped |= BTRFS_BLOCK_GROUP_DUP;
9332 stripped = flags & ~stripped;
9334 /* turn raid0 into single device chunks */
9335 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9338 /* turn mirroring into duplication */
9339 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9340 BTRFS_BLOCK_GROUP_RAID10))
9341 return stripped | BTRFS_BLOCK_GROUP_DUP;
9343 /* they already had raid on here, just return */
9344 if (flags & stripped)
9347 stripped |= BTRFS_BLOCK_GROUP_DUP;
9348 stripped = flags & ~stripped;
9350 /* switch duplicated blocks with raid1 */
9351 if (flags & BTRFS_BLOCK_GROUP_DUP)
9352 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9354 /* this is drive concat, leave it alone */
9360 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9362 struct btrfs_space_info *sinfo = cache->space_info;
9364 u64 min_allocable_bytes;
9368 * We need some metadata space and system metadata space for
9369 * allocating chunks in some corner cases until we force to set
9370 * it to be readonly.
9373 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9375 min_allocable_bytes = SZ_1M;
9377 min_allocable_bytes = 0;
9379 spin_lock(&sinfo->lock);
9380 spin_lock(&cache->lock);
9388 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9389 cache->bytes_super - btrfs_block_group_used(&cache->item);
9391 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9392 min_allocable_bytes <= sinfo->total_bytes) {
9393 sinfo->bytes_readonly += num_bytes;
9395 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9399 spin_unlock(&cache->lock);
9400 spin_unlock(&sinfo->lock);
9404 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9405 struct btrfs_block_group_cache *cache)
9408 struct btrfs_trans_handle *trans;
9413 trans = btrfs_join_transaction(fs_info->extent_root);
9415 return PTR_ERR(trans);
9418 * we're not allowed to set block groups readonly after the dirty
9419 * block groups cache has started writing. If it already started,
9420 * back off and let this transaction commit
9422 mutex_lock(&fs_info->ro_block_group_mutex);
9423 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9424 u64 transid = trans->transid;
9426 mutex_unlock(&fs_info->ro_block_group_mutex);
9427 btrfs_end_transaction(trans);
9429 ret = btrfs_wait_for_commit(fs_info, transid);
9436 * if we are changing raid levels, try to allocate a corresponding
9437 * block group with the new raid level.
9439 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9440 if (alloc_flags != cache->flags) {
9441 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9444 * ENOSPC is allowed here, we may have enough space
9445 * already allocated at the new raid level to
9454 ret = inc_block_group_ro(cache, 0);
9457 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9458 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9462 ret = inc_block_group_ro(cache, 0);
9464 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9465 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9466 mutex_lock(&fs_info->chunk_mutex);
9467 check_system_chunk(trans, fs_info, alloc_flags);
9468 mutex_unlock(&fs_info->chunk_mutex);
9470 mutex_unlock(&fs_info->ro_block_group_mutex);
9472 btrfs_end_transaction(trans);
9476 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9477 struct btrfs_fs_info *fs_info, u64 type)
9479 u64 alloc_flags = get_alloc_profile(fs_info, type);
9481 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9485 * helper to account the unused space of all the readonly block group in the
9486 * space_info. takes mirrors into account.
9488 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9490 struct btrfs_block_group_cache *block_group;
9494 /* It's df, we don't care if it's racy */
9495 if (list_empty(&sinfo->ro_bgs))
9498 spin_lock(&sinfo->lock);
9499 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9500 spin_lock(&block_group->lock);
9502 if (!block_group->ro) {
9503 spin_unlock(&block_group->lock);
9507 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9508 BTRFS_BLOCK_GROUP_RAID10 |
9509 BTRFS_BLOCK_GROUP_DUP))
9514 free_bytes += (block_group->key.offset -
9515 btrfs_block_group_used(&block_group->item)) *
9518 spin_unlock(&block_group->lock);
9520 spin_unlock(&sinfo->lock);
9525 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9527 struct btrfs_space_info *sinfo = cache->space_info;
9532 spin_lock(&sinfo->lock);
9533 spin_lock(&cache->lock);
9535 num_bytes = cache->key.offset - cache->reserved -
9536 cache->pinned - cache->bytes_super -
9537 btrfs_block_group_used(&cache->item);
9538 sinfo->bytes_readonly -= num_bytes;
9539 list_del_init(&cache->ro_list);
9541 spin_unlock(&cache->lock);
9542 spin_unlock(&sinfo->lock);
9546 * checks to see if its even possible to relocate this block group.
9548 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9549 * ok to go ahead and try.
9551 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9553 struct btrfs_root *root = fs_info->extent_root;
9554 struct btrfs_block_group_cache *block_group;
9555 struct btrfs_space_info *space_info;
9556 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9557 struct btrfs_device *device;
9558 struct btrfs_trans_handle *trans;
9568 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9570 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9572 /* odd, couldn't find the block group, leave it alone */
9576 "can't find block group for bytenr %llu",
9581 min_free = btrfs_block_group_used(&block_group->item);
9583 /* no bytes used, we're good */
9587 space_info = block_group->space_info;
9588 spin_lock(&space_info->lock);
9590 full = space_info->full;
9593 * if this is the last block group we have in this space, we can't
9594 * relocate it unless we're able to allocate a new chunk below.
9596 * Otherwise, we need to make sure we have room in the space to handle
9597 * all of the extents from this block group. If we can, we're good
9599 if ((space_info->total_bytes != block_group->key.offset) &&
9600 (btrfs_space_info_used(space_info, false) + min_free <
9601 space_info->total_bytes)) {
9602 spin_unlock(&space_info->lock);
9605 spin_unlock(&space_info->lock);
9608 * ok we don't have enough space, but maybe we have free space on our
9609 * devices to allocate new chunks for relocation, so loop through our
9610 * alloc devices and guess if we have enough space. if this block
9611 * group is going to be restriped, run checks against the target
9612 * profile instead of the current one.
9624 target = get_restripe_target(fs_info, block_group->flags);
9626 index = __get_raid_index(extended_to_chunk(target));
9629 * this is just a balance, so if we were marked as full
9630 * we know there is no space for a new chunk
9635 "no space to alloc new chunk for block group %llu",
9636 block_group->key.objectid);
9640 index = get_block_group_index(block_group);
9643 if (index == BTRFS_RAID_RAID10) {
9647 } else if (index == BTRFS_RAID_RAID1) {
9649 } else if (index == BTRFS_RAID_DUP) {
9652 } else if (index == BTRFS_RAID_RAID0) {
9653 dev_min = fs_devices->rw_devices;
9654 min_free = div64_u64(min_free, dev_min);
9657 /* We need to do this so that we can look at pending chunks */
9658 trans = btrfs_join_transaction(root);
9659 if (IS_ERR(trans)) {
9660 ret = PTR_ERR(trans);
9664 mutex_lock(&fs_info->chunk_mutex);
9665 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9669 * check to make sure we can actually find a chunk with enough
9670 * space to fit our block group in.
9672 if (device->total_bytes > device->bytes_used + min_free &&
9673 !device->is_tgtdev_for_dev_replace) {
9674 ret = find_free_dev_extent(trans, device, min_free,
9679 if (dev_nr >= dev_min)
9685 if (debug && ret == -1)
9687 "no space to allocate a new chunk for block group %llu",
9688 block_group->key.objectid);
9689 mutex_unlock(&fs_info->chunk_mutex);
9690 btrfs_end_transaction(trans);
9692 btrfs_put_block_group(block_group);
9696 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9697 struct btrfs_path *path,
9698 struct btrfs_key *key)
9700 struct btrfs_root *root = fs_info->extent_root;
9702 struct btrfs_key found_key;
9703 struct extent_buffer *leaf;
9706 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9711 slot = path->slots[0];
9712 leaf = path->nodes[0];
9713 if (slot >= btrfs_header_nritems(leaf)) {
9714 ret = btrfs_next_leaf(root, path);
9721 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9723 if (found_key.objectid >= key->objectid &&
9724 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9725 struct extent_map_tree *em_tree;
9726 struct extent_map *em;
9728 em_tree = &root->fs_info->mapping_tree.map_tree;
9729 read_lock(&em_tree->lock);
9730 em = lookup_extent_mapping(em_tree, found_key.objectid,
9732 read_unlock(&em_tree->lock);
9735 "logical %llu len %llu found bg but no related chunk",
9736 found_key.objectid, found_key.offset);
9741 free_extent_map(em);
9750 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9752 struct btrfs_block_group_cache *block_group;
9756 struct inode *inode;
9758 block_group = btrfs_lookup_first_block_group(info, last);
9759 while (block_group) {
9760 spin_lock(&block_group->lock);
9761 if (block_group->iref)
9763 spin_unlock(&block_group->lock);
9764 block_group = next_block_group(info, block_group);
9773 inode = block_group->inode;
9774 block_group->iref = 0;
9775 block_group->inode = NULL;
9776 spin_unlock(&block_group->lock);
9777 ASSERT(block_group->io_ctl.inode == NULL);
9779 last = block_group->key.objectid + block_group->key.offset;
9780 btrfs_put_block_group(block_group);
9785 * Must be called only after stopping all workers, since we could have block
9786 * group caching kthreads running, and therefore they could race with us if we
9787 * freed the block groups before stopping them.
9789 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9791 struct btrfs_block_group_cache *block_group;
9792 struct btrfs_space_info *space_info;
9793 struct btrfs_caching_control *caching_ctl;
9796 down_write(&info->commit_root_sem);
9797 while (!list_empty(&info->caching_block_groups)) {
9798 caching_ctl = list_entry(info->caching_block_groups.next,
9799 struct btrfs_caching_control, list);
9800 list_del(&caching_ctl->list);
9801 put_caching_control(caching_ctl);
9803 up_write(&info->commit_root_sem);
9805 spin_lock(&info->unused_bgs_lock);
9806 while (!list_empty(&info->unused_bgs)) {
9807 block_group = list_first_entry(&info->unused_bgs,
9808 struct btrfs_block_group_cache,
9810 list_del_init(&block_group->bg_list);
9811 btrfs_put_block_group(block_group);
9813 spin_unlock(&info->unused_bgs_lock);
9815 spin_lock(&info->block_group_cache_lock);
9816 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9817 block_group = rb_entry(n, struct btrfs_block_group_cache,
9819 rb_erase(&block_group->cache_node,
9820 &info->block_group_cache_tree);
9821 RB_CLEAR_NODE(&block_group->cache_node);
9822 spin_unlock(&info->block_group_cache_lock);
9824 down_write(&block_group->space_info->groups_sem);
9825 list_del(&block_group->list);
9826 up_write(&block_group->space_info->groups_sem);
9829 * We haven't cached this block group, which means we could
9830 * possibly have excluded extents on this block group.
9832 if (block_group->cached == BTRFS_CACHE_NO ||
9833 block_group->cached == BTRFS_CACHE_ERROR)
9834 free_excluded_extents(info, block_group);
9836 btrfs_remove_free_space_cache(block_group);
9837 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9838 ASSERT(list_empty(&block_group->dirty_list));
9839 ASSERT(list_empty(&block_group->io_list));
9840 ASSERT(list_empty(&block_group->bg_list));
9841 ASSERT(atomic_read(&block_group->count) == 1);
9842 btrfs_put_block_group(block_group);
9844 spin_lock(&info->block_group_cache_lock);
9846 spin_unlock(&info->block_group_cache_lock);
9848 /* now that all the block groups are freed, go through and
9849 * free all the space_info structs. This is only called during
9850 * the final stages of unmount, and so we know nobody is
9851 * using them. We call synchronize_rcu() once before we start,
9852 * just to be on the safe side.
9856 release_global_block_rsv(info);
9858 while (!list_empty(&info->space_info)) {
9861 space_info = list_entry(info->space_info.next,
9862 struct btrfs_space_info,
9866 * Do not hide this behind enospc_debug, this is actually
9867 * important and indicates a real bug if this happens.
9869 if (WARN_ON(space_info->bytes_pinned > 0 ||
9870 space_info->bytes_reserved > 0 ||
9871 space_info->bytes_may_use > 0))
9872 dump_space_info(info, space_info, 0, 0);
9873 list_del(&space_info->list);
9874 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9875 struct kobject *kobj;
9876 kobj = space_info->block_group_kobjs[i];
9877 space_info->block_group_kobjs[i] = NULL;
9883 kobject_del(&space_info->kobj);
9884 kobject_put(&space_info->kobj);
9889 static void __link_block_group(struct btrfs_space_info *space_info,
9890 struct btrfs_block_group_cache *cache)
9892 int index = get_block_group_index(cache);
9895 down_write(&space_info->groups_sem);
9896 if (list_empty(&space_info->block_groups[index]))
9898 list_add_tail(&cache->list, &space_info->block_groups[index]);
9899 up_write(&space_info->groups_sem);
9902 struct raid_kobject *rkobj;
9905 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9908 rkobj->raid_type = index;
9909 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9910 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9911 "%s", get_raid_name(index));
9913 kobject_put(&rkobj->kobj);
9916 space_info->block_group_kobjs[index] = &rkobj->kobj;
9921 btrfs_warn(cache->fs_info,
9922 "failed to add kobject for block cache, ignoring");
9925 static struct btrfs_block_group_cache *
9926 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9927 u64 start, u64 size)
9929 struct btrfs_block_group_cache *cache;
9931 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9935 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9937 if (!cache->free_space_ctl) {
9942 cache->key.objectid = start;
9943 cache->key.offset = size;
9944 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9946 cache->fs_info = fs_info;
9947 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
9948 set_free_space_tree_thresholds(cache);
9950 atomic_set(&cache->count, 1);
9951 spin_lock_init(&cache->lock);
9952 init_rwsem(&cache->data_rwsem);
9953 INIT_LIST_HEAD(&cache->list);
9954 INIT_LIST_HEAD(&cache->cluster_list);
9955 INIT_LIST_HEAD(&cache->bg_list);
9956 INIT_LIST_HEAD(&cache->ro_list);
9957 INIT_LIST_HEAD(&cache->dirty_list);
9958 INIT_LIST_HEAD(&cache->io_list);
9959 btrfs_init_free_space_ctl(cache);
9960 atomic_set(&cache->trimming, 0);
9961 mutex_init(&cache->free_space_lock);
9962 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
9967 int btrfs_read_block_groups(struct btrfs_fs_info *info)
9969 struct btrfs_path *path;
9971 struct btrfs_block_group_cache *cache;
9972 struct btrfs_space_info *space_info;
9973 struct btrfs_key key;
9974 struct btrfs_key found_key;
9975 struct extent_buffer *leaf;
9981 feature = btrfs_super_incompat_flags(info->super_copy);
9982 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
9986 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9987 path = btrfs_alloc_path();
9990 path->reada = READA_FORWARD;
9992 cache_gen = btrfs_super_cache_generation(info->super_copy);
9993 if (btrfs_test_opt(info, SPACE_CACHE) &&
9994 btrfs_super_generation(info->super_copy) != cache_gen)
9996 if (btrfs_test_opt(info, CLEAR_CACHE))
10000 ret = find_first_block_group(info, path, &key);
10006 leaf = path->nodes[0];
10007 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10009 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10018 * When we mount with old space cache, we need to
10019 * set BTRFS_DC_CLEAR and set dirty flag.
10021 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10022 * truncate the old free space cache inode and
10024 * b) Setting 'dirty flag' makes sure that we flush
10025 * the new space cache info onto disk.
10027 if (btrfs_test_opt(info, SPACE_CACHE))
10028 cache->disk_cache_state = BTRFS_DC_CLEAR;
10031 read_extent_buffer(leaf, &cache->item,
10032 btrfs_item_ptr_offset(leaf, path->slots[0]),
10033 sizeof(cache->item));
10034 cache->flags = btrfs_block_group_flags(&cache->item);
10036 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10037 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10039 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10040 cache->key.objectid);
10045 key.objectid = found_key.objectid + found_key.offset;
10046 btrfs_release_path(path);
10049 * We need to exclude the super stripes now so that the space
10050 * info has super bytes accounted for, otherwise we'll think
10051 * we have more space than we actually do.
10053 ret = exclude_super_stripes(info, cache);
10056 * We may have excluded something, so call this just in
10059 free_excluded_extents(info, cache);
10060 btrfs_put_block_group(cache);
10065 * check for two cases, either we are full, and therefore
10066 * don't need to bother with the caching work since we won't
10067 * find any space, or we are empty, and we can just add all
10068 * the space in and be done with it. This saves us _alot_ of
10069 * time, particularly in the full case.
10071 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10072 cache->last_byte_to_unpin = (u64)-1;
10073 cache->cached = BTRFS_CACHE_FINISHED;
10074 free_excluded_extents(info, cache);
10075 } else if (btrfs_block_group_used(&cache->item) == 0) {
10076 cache->last_byte_to_unpin = (u64)-1;
10077 cache->cached = BTRFS_CACHE_FINISHED;
10078 add_new_free_space(cache, info,
10079 found_key.objectid,
10080 found_key.objectid +
10082 free_excluded_extents(info, cache);
10085 ret = btrfs_add_block_group_cache(info, cache);
10087 btrfs_remove_free_space_cache(cache);
10088 btrfs_put_block_group(cache);
10092 trace_btrfs_add_block_group(info, cache, 0);
10093 update_space_info(info, cache->flags, found_key.offset,
10094 btrfs_block_group_used(&cache->item),
10095 cache->bytes_super, &space_info);
10097 cache->space_info = space_info;
10099 __link_block_group(space_info, cache);
10101 set_avail_alloc_bits(info, cache->flags);
10102 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10103 inc_block_group_ro(cache, 1);
10104 } else if (btrfs_block_group_used(&cache->item) == 0) {
10105 spin_lock(&info->unused_bgs_lock);
10106 /* Should always be true but just in case. */
10107 if (list_empty(&cache->bg_list)) {
10108 btrfs_get_block_group(cache);
10109 list_add_tail(&cache->bg_list,
10110 &info->unused_bgs);
10112 spin_unlock(&info->unused_bgs_lock);
10116 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10117 if (!(get_alloc_profile(info, space_info->flags) &
10118 (BTRFS_BLOCK_GROUP_RAID10 |
10119 BTRFS_BLOCK_GROUP_RAID1 |
10120 BTRFS_BLOCK_GROUP_RAID5 |
10121 BTRFS_BLOCK_GROUP_RAID6 |
10122 BTRFS_BLOCK_GROUP_DUP)))
10125 * avoid allocating from un-mirrored block group if there are
10126 * mirrored block groups.
10128 list_for_each_entry(cache,
10129 &space_info->block_groups[BTRFS_RAID_RAID0],
10131 inc_block_group_ro(cache, 1);
10132 list_for_each_entry(cache,
10133 &space_info->block_groups[BTRFS_RAID_SINGLE],
10135 inc_block_group_ro(cache, 1);
10138 init_global_block_rsv(info);
10141 btrfs_free_path(path);
10145 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10146 struct btrfs_fs_info *fs_info)
10148 struct btrfs_block_group_cache *block_group, *tmp;
10149 struct btrfs_root *extent_root = fs_info->extent_root;
10150 struct btrfs_block_group_item item;
10151 struct btrfs_key key;
10153 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10155 trans->can_flush_pending_bgs = false;
10156 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10160 spin_lock(&block_group->lock);
10161 memcpy(&item, &block_group->item, sizeof(item));
10162 memcpy(&key, &block_group->key, sizeof(key));
10163 spin_unlock(&block_group->lock);
10165 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10168 btrfs_abort_transaction(trans, ret);
10169 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10172 btrfs_abort_transaction(trans, ret);
10173 add_block_group_free_space(trans, fs_info, block_group);
10174 /* already aborted the transaction if it failed. */
10176 list_del_init(&block_group->bg_list);
10178 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10181 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10182 struct btrfs_fs_info *fs_info, u64 bytes_used,
10183 u64 type, u64 chunk_offset, u64 size)
10185 struct btrfs_block_group_cache *cache;
10188 btrfs_set_log_full_commit(fs_info, trans);
10190 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10194 btrfs_set_block_group_used(&cache->item, bytes_used);
10195 btrfs_set_block_group_chunk_objectid(&cache->item,
10196 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10197 btrfs_set_block_group_flags(&cache->item, type);
10199 cache->flags = type;
10200 cache->last_byte_to_unpin = (u64)-1;
10201 cache->cached = BTRFS_CACHE_FINISHED;
10202 cache->needs_free_space = 1;
10203 ret = exclude_super_stripes(fs_info, cache);
10206 * We may have excluded something, so call this just in
10209 free_excluded_extents(fs_info, cache);
10210 btrfs_put_block_group(cache);
10214 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10216 free_excluded_extents(fs_info, cache);
10218 #ifdef CONFIG_BTRFS_DEBUG
10219 if (btrfs_should_fragment_free_space(cache)) {
10220 u64 new_bytes_used = size - bytes_used;
10222 bytes_used += new_bytes_used >> 1;
10223 fragment_free_space(cache);
10227 * Ensure the corresponding space_info object is created and
10228 * assigned to our block group. We want our bg to be added to the rbtree
10229 * with its ->space_info set.
10231 cache->space_info = __find_space_info(fs_info, cache->flags);
10232 if (!cache->space_info) {
10233 ret = create_space_info(fs_info, cache->flags,
10234 &cache->space_info);
10236 btrfs_remove_free_space_cache(cache);
10237 btrfs_put_block_group(cache);
10242 ret = btrfs_add_block_group_cache(fs_info, cache);
10244 btrfs_remove_free_space_cache(cache);
10245 btrfs_put_block_group(cache);
10250 * Now that our block group has its ->space_info set and is inserted in
10251 * the rbtree, update the space info's counters.
10253 trace_btrfs_add_block_group(fs_info, cache, 1);
10254 update_space_info(fs_info, cache->flags, size, bytes_used,
10255 cache->bytes_super, &cache->space_info);
10256 update_global_block_rsv(fs_info);
10258 __link_block_group(cache->space_info, cache);
10260 list_add_tail(&cache->bg_list, &trans->new_bgs);
10262 set_avail_alloc_bits(fs_info, type);
10266 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10268 u64 extra_flags = chunk_to_extended(flags) &
10269 BTRFS_EXTENDED_PROFILE_MASK;
10271 write_seqlock(&fs_info->profiles_lock);
10272 if (flags & BTRFS_BLOCK_GROUP_DATA)
10273 fs_info->avail_data_alloc_bits &= ~extra_flags;
10274 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10275 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10276 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10277 fs_info->avail_system_alloc_bits &= ~extra_flags;
10278 write_sequnlock(&fs_info->profiles_lock);
10281 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10282 struct btrfs_fs_info *fs_info, u64 group_start,
10283 struct extent_map *em)
10285 struct btrfs_root *root = fs_info->extent_root;
10286 struct btrfs_path *path;
10287 struct btrfs_block_group_cache *block_group;
10288 struct btrfs_free_cluster *cluster;
10289 struct btrfs_root *tree_root = fs_info->tree_root;
10290 struct btrfs_key key;
10291 struct inode *inode;
10292 struct kobject *kobj = NULL;
10296 struct btrfs_caching_control *caching_ctl = NULL;
10299 block_group = btrfs_lookup_block_group(fs_info, group_start);
10300 BUG_ON(!block_group);
10301 BUG_ON(!block_group->ro);
10304 * Free the reserved super bytes from this block group before
10307 free_excluded_extents(fs_info, block_group);
10309 memcpy(&key, &block_group->key, sizeof(key));
10310 index = get_block_group_index(block_group);
10311 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10312 BTRFS_BLOCK_GROUP_RAID1 |
10313 BTRFS_BLOCK_GROUP_RAID10))
10318 /* make sure this block group isn't part of an allocation cluster */
10319 cluster = &fs_info->data_alloc_cluster;
10320 spin_lock(&cluster->refill_lock);
10321 btrfs_return_cluster_to_free_space(block_group, cluster);
10322 spin_unlock(&cluster->refill_lock);
10325 * make sure this block group isn't part of a metadata
10326 * allocation cluster
10328 cluster = &fs_info->meta_alloc_cluster;
10329 spin_lock(&cluster->refill_lock);
10330 btrfs_return_cluster_to_free_space(block_group, cluster);
10331 spin_unlock(&cluster->refill_lock);
10333 path = btrfs_alloc_path();
10340 * get the inode first so any iput calls done for the io_list
10341 * aren't the final iput (no unlinks allowed now)
10343 inode = lookup_free_space_inode(fs_info, block_group, path);
10345 mutex_lock(&trans->transaction->cache_write_mutex);
10347 * make sure our free spache cache IO is done before remove the
10350 spin_lock(&trans->transaction->dirty_bgs_lock);
10351 if (!list_empty(&block_group->io_list)) {
10352 list_del_init(&block_group->io_list);
10354 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10356 spin_unlock(&trans->transaction->dirty_bgs_lock);
10357 btrfs_wait_cache_io(trans, block_group, path);
10358 btrfs_put_block_group(block_group);
10359 spin_lock(&trans->transaction->dirty_bgs_lock);
10362 if (!list_empty(&block_group->dirty_list)) {
10363 list_del_init(&block_group->dirty_list);
10364 btrfs_put_block_group(block_group);
10366 spin_unlock(&trans->transaction->dirty_bgs_lock);
10367 mutex_unlock(&trans->transaction->cache_write_mutex);
10369 if (!IS_ERR(inode)) {
10370 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10372 btrfs_add_delayed_iput(inode);
10375 clear_nlink(inode);
10376 /* One for the block groups ref */
10377 spin_lock(&block_group->lock);
10378 if (block_group->iref) {
10379 block_group->iref = 0;
10380 block_group->inode = NULL;
10381 spin_unlock(&block_group->lock);
10384 spin_unlock(&block_group->lock);
10386 /* One for our lookup ref */
10387 btrfs_add_delayed_iput(inode);
10390 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10391 key.offset = block_group->key.objectid;
10394 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10398 btrfs_release_path(path);
10400 ret = btrfs_del_item(trans, tree_root, path);
10403 btrfs_release_path(path);
10406 spin_lock(&fs_info->block_group_cache_lock);
10407 rb_erase(&block_group->cache_node,
10408 &fs_info->block_group_cache_tree);
10409 RB_CLEAR_NODE(&block_group->cache_node);
10411 if (fs_info->first_logical_byte == block_group->key.objectid)
10412 fs_info->first_logical_byte = (u64)-1;
10413 spin_unlock(&fs_info->block_group_cache_lock);
10415 down_write(&block_group->space_info->groups_sem);
10417 * we must use list_del_init so people can check to see if they
10418 * are still on the list after taking the semaphore
10420 list_del_init(&block_group->list);
10421 if (list_empty(&block_group->space_info->block_groups[index])) {
10422 kobj = block_group->space_info->block_group_kobjs[index];
10423 block_group->space_info->block_group_kobjs[index] = NULL;
10424 clear_avail_alloc_bits(fs_info, block_group->flags);
10426 up_write(&block_group->space_info->groups_sem);
10432 if (block_group->has_caching_ctl)
10433 caching_ctl = get_caching_control(block_group);
10434 if (block_group->cached == BTRFS_CACHE_STARTED)
10435 wait_block_group_cache_done(block_group);
10436 if (block_group->has_caching_ctl) {
10437 down_write(&fs_info->commit_root_sem);
10438 if (!caching_ctl) {
10439 struct btrfs_caching_control *ctl;
10441 list_for_each_entry(ctl,
10442 &fs_info->caching_block_groups, list)
10443 if (ctl->block_group == block_group) {
10445 refcount_inc(&caching_ctl->count);
10450 list_del_init(&caching_ctl->list);
10451 up_write(&fs_info->commit_root_sem);
10453 /* Once for the caching bgs list and once for us. */
10454 put_caching_control(caching_ctl);
10455 put_caching_control(caching_ctl);
10459 spin_lock(&trans->transaction->dirty_bgs_lock);
10460 if (!list_empty(&block_group->dirty_list)) {
10463 if (!list_empty(&block_group->io_list)) {
10466 spin_unlock(&trans->transaction->dirty_bgs_lock);
10467 btrfs_remove_free_space_cache(block_group);
10469 spin_lock(&block_group->space_info->lock);
10470 list_del_init(&block_group->ro_list);
10472 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10473 WARN_ON(block_group->space_info->total_bytes
10474 < block_group->key.offset);
10475 WARN_ON(block_group->space_info->bytes_readonly
10476 < block_group->key.offset);
10477 WARN_ON(block_group->space_info->disk_total
10478 < block_group->key.offset * factor);
10480 block_group->space_info->total_bytes -= block_group->key.offset;
10481 block_group->space_info->bytes_readonly -= block_group->key.offset;
10482 block_group->space_info->disk_total -= block_group->key.offset * factor;
10484 spin_unlock(&block_group->space_info->lock);
10486 memcpy(&key, &block_group->key, sizeof(key));
10488 mutex_lock(&fs_info->chunk_mutex);
10489 if (!list_empty(&em->list)) {
10490 /* We're in the transaction->pending_chunks list. */
10491 free_extent_map(em);
10493 spin_lock(&block_group->lock);
10494 block_group->removed = 1;
10496 * At this point trimming can't start on this block group, because we
10497 * removed the block group from the tree fs_info->block_group_cache_tree
10498 * so no one can't find it anymore and even if someone already got this
10499 * block group before we removed it from the rbtree, they have already
10500 * incremented block_group->trimming - if they didn't, they won't find
10501 * any free space entries because we already removed them all when we
10502 * called btrfs_remove_free_space_cache().
10504 * And we must not remove the extent map from the fs_info->mapping_tree
10505 * to prevent the same logical address range and physical device space
10506 * ranges from being reused for a new block group. This is because our
10507 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10508 * completely transactionless, so while it is trimming a range the
10509 * currently running transaction might finish and a new one start,
10510 * allowing for new block groups to be created that can reuse the same
10511 * physical device locations unless we take this special care.
10513 * There may also be an implicit trim operation if the file system
10514 * is mounted with -odiscard. The same protections must remain
10515 * in place until the extents have been discarded completely when
10516 * the transaction commit has completed.
10518 remove_em = (atomic_read(&block_group->trimming) == 0);
10520 * Make sure a trimmer task always sees the em in the pinned_chunks list
10521 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10522 * before checking block_group->removed).
10526 * Our em might be in trans->transaction->pending_chunks which
10527 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10528 * and so is the fs_info->pinned_chunks list.
10530 * So at this point we must be holding the chunk_mutex to avoid
10531 * any races with chunk allocation (more specifically at
10532 * volumes.c:contains_pending_extent()), to ensure it always
10533 * sees the em, either in the pending_chunks list or in the
10534 * pinned_chunks list.
10536 list_move_tail(&em->list, &fs_info->pinned_chunks);
10538 spin_unlock(&block_group->lock);
10541 struct extent_map_tree *em_tree;
10543 em_tree = &fs_info->mapping_tree.map_tree;
10544 write_lock(&em_tree->lock);
10546 * The em might be in the pending_chunks list, so make sure the
10547 * chunk mutex is locked, since remove_extent_mapping() will
10548 * delete us from that list.
10550 remove_extent_mapping(em_tree, em);
10551 write_unlock(&em_tree->lock);
10552 /* once for the tree */
10553 free_extent_map(em);
10556 mutex_unlock(&fs_info->chunk_mutex);
10558 ret = remove_block_group_free_space(trans, fs_info, block_group);
10562 btrfs_put_block_group(block_group);
10563 btrfs_put_block_group(block_group);
10565 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10571 ret = btrfs_del_item(trans, root, path);
10573 btrfs_free_path(path);
10577 struct btrfs_trans_handle *
10578 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10579 const u64 chunk_offset)
10581 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10582 struct extent_map *em;
10583 struct map_lookup *map;
10584 unsigned int num_items;
10586 read_lock(&em_tree->lock);
10587 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10588 read_unlock(&em_tree->lock);
10589 ASSERT(em && em->start == chunk_offset);
10592 * We need to reserve 3 + N units from the metadata space info in order
10593 * to remove a block group (done at btrfs_remove_chunk() and at
10594 * btrfs_remove_block_group()), which are used for:
10596 * 1 unit for adding the free space inode's orphan (located in the tree
10598 * 1 unit for deleting the block group item (located in the extent
10600 * 1 unit for deleting the free space item (located in tree of tree
10602 * N units for deleting N device extent items corresponding to each
10603 * stripe (located in the device tree).
10605 * In order to remove a block group we also need to reserve units in the
10606 * system space info in order to update the chunk tree (update one or
10607 * more device items and remove one chunk item), but this is done at
10608 * btrfs_remove_chunk() through a call to check_system_chunk().
10610 map = em->map_lookup;
10611 num_items = 3 + map->num_stripes;
10612 free_extent_map(em);
10614 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10619 * Process the unused_bgs list and remove any that don't have any allocated
10620 * space inside of them.
10622 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10624 struct btrfs_block_group_cache *block_group;
10625 struct btrfs_space_info *space_info;
10626 struct btrfs_trans_handle *trans;
10629 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10632 spin_lock(&fs_info->unused_bgs_lock);
10633 while (!list_empty(&fs_info->unused_bgs)) {
10637 block_group = list_first_entry(&fs_info->unused_bgs,
10638 struct btrfs_block_group_cache,
10640 list_del_init(&block_group->bg_list);
10642 space_info = block_group->space_info;
10644 if (ret || btrfs_mixed_space_info(space_info)) {
10645 btrfs_put_block_group(block_group);
10648 spin_unlock(&fs_info->unused_bgs_lock);
10650 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10652 /* Don't want to race with allocators so take the groups_sem */
10653 down_write(&space_info->groups_sem);
10654 spin_lock(&block_group->lock);
10655 if (block_group->reserved ||
10656 btrfs_block_group_used(&block_group->item) ||
10658 list_is_singular(&block_group->list)) {
10660 * We want to bail if we made new allocations or have
10661 * outstanding allocations in this block group. We do
10662 * the ro check in case balance is currently acting on
10663 * this block group.
10665 spin_unlock(&block_group->lock);
10666 up_write(&space_info->groups_sem);
10669 spin_unlock(&block_group->lock);
10671 /* We don't want to force the issue, only flip if it's ok. */
10672 ret = inc_block_group_ro(block_group, 0);
10673 up_write(&space_info->groups_sem);
10680 * Want to do this before we do anything else so we can recover
10681 * properly if we fail to join the transaction.
10683 trans = btrfs_start_trans_remove_block_group(fs_info,
10684 block_group->key.objectid);
10685 if (IS_ERR(trans)) {
10686 btrfs_dec_block_group_ro(block_group);
10687 ret = PTR_ERR(trans);
10692 * We could have pending pinned extents for this block group,
10693 * just delete them, we don't care about them anymore.
10695 start = block_group->key.objectid;
10696 end = start + block_group->key.offset - 1;
10698 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10699 * btrfs_finish_extent_commit(). If we are at transaction N,
10700 * another task might be running finish_extent_commit() for the
10701 * previous transaction N - 1, and have seen a range belonging
10702 * to the block group in freed_extents[] before we were able to
10703 * clear the whole block group range from freed_extents[]. This
10704 * means that task can lookup for the block group after we
10705 * unpinned it from freed_extents[] and removed it, leading to
10706 * a BUG_ON() at btrfs_unpin_extent_range().
10708 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10709 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10712 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10713 btrfs_dec_block_group_ro(block_group);
10716 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10719 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10720 btrfs_dec_block_group_ro(block_group);
10723 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10725 /* Reset pinned so btrfs_put_block_group doesn't complain */
10726 spin_lock(&space_info->lock);
10727 spin_lock(&block_group->lock);
10729 space_info->bytes_pinned -= block_group->pinned;
10730 space_info->bytes_readonly += block_group->pinned;
10731 percpu_counter_add(&space_info->total_bytes_pinned,
10732 -block_group->pinned);
10733 block_group->pinned = 0;
10735 spin_unlock(&block_group->lock);
10736 spin_unlock(&space_info->lock);
10738 /* DISCARD can flip during remount */
10739 trimming = btrfs_test_opt(fs_info, DISCARD);
10741 /* Implicit trim during transaction commit. */
10743 btrfs_get_block_group_trimming(block_group);
10746 * Btrfs_remove_chunk will abort the transaction if things go
10749 ret = btrfs_remove_chunk(trans, fs_info,
10750 block_group->key.objectid);
10754 btrfs_put_block_group_trimming(block_group);
10759 * If we're not mounted with -odiscard, we can just forget
10760 * about this block group. Otherwise we'll need to wait
10761 * until transaction commit to do the actual discard.
10764 spin_lock(&fs_info->unused_bgs_lock);
10766 * A concurrent scrub might have added us to the list
10767 * fs_info->unused_bgs, so use a list_move operation
10768 * to add the block group to the deleted_bgs list.
10770 list_move(&block_group->bg_list,
10771 &trans->transaction->deleted_bgs);
10772 spin_unlock(&fs_info->unused_bgs_lock);
10773 btrfs_get_block_group(block_group);
10776 btrfs_end_transaction(trans);
10778 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10779 btrfs_put_block_group(block_group);
10780 spin_lock(&fs_info->unused_bgs_lock);
10782 spin_unlock(&fs_info->unused_bgs_lock);
10785 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10787 struct btrfs_space_info *space_info;
10788 struct btrfs_super_block *disk_super;
10794 disk_super = fs_info->super_copy;
10795 if (!btrfs_super_root(disk_super))
10798 features = btrfs_super_incompat_flags(disk_super);
10799 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10802 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10803 ret = create_space_info(fs_info, flags, &space_info);
10808 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10809 ret = create_space_info(fs_info, flags, &space_info);
10811 flags = BTRFS_BLOCK_GROUP_METADATA;
10812 ret = create_space_info(fs_info, flags, &space_info);
10816 flags = BTRFS_BLOCK_GROUP_DATA;
10817 ret = create_space_info(fs_info, flags, &space_info);
10823 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10824 u64 start, u64 end)
10826 return unpin_extent_range(fs_info, start, end, false);
10830 * It used to be that old block groups would be left around forever.
10831 * Iterating over them would be enough to trim unused space. Since we
10832 * now automatically remove them, we also need to iterate over unallocated
10835 * We don't want a transaction for this since the discard may take a
10836 * substantial amount of time. We don't require that a transaction be
10837 * running, but we do need to take a running transaction into account
10838 * to ensure that we're not discarding chunks that were released in
10839 * the current transaction.
10841 * Holding the chunks lock will prevent other threads from allocating
10842 * or releasing chunks, but it won't prevent a running transaction
10843 * from committing and releasing the memory that the pending chunks
10844 * list head uses. For that, we need to take a reference to the
10847 static int btrfs_trim_free_extents(struct btrfs_device *device,
10848 u64 minlen, u64 *trimmed)
10850 u64 start = 0, len = 0;
10855 /* Not writeable = nothing to do. */
10856 if (!device->writeable)
10859 /* No free space = nothing to do. */
10860 if (device->total_bytes <= device->bytes_used)
10866 struct btrfs_fs_info *fs_info = device->fs_info;
10867 struct btrfs_transaction *trans;
10870 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10874 down_read(&fs_info->commit_root_sem);
10876 spin_lock(&fs_info->trans_lock);
10877 trans = fs_info->running_transaction;
10879 refcount_inc(&trans->use_count);
10880 spin_unlock(&fs_info->trans_lock);
10882 ret = find_free_dev_extent_start(trans, device, minlen, start,
10885 btrfs_put_transaction(trans);
10888 up_read(&fs_info->commit_root_sem);
10889 mutex_unlock(&fs_info->chunk_mutex);
10890 if (ret == -ENOSPC)
10895 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10896 up_read(&fs_info->commit_root_sem);
10897 mutex_unlock(&fs_info->chunk_mutex);
10905 if (fatal_signal_pending(current)) {
10906 ret = -ERESTARTSYS;
10916 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10918 struct btrfs_block_group_cache *cache = NULL;
10919 struct btrfs_device *device;
10920 struct list_head *devices;
10925 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10929 * try to trim all FS space, our block group may start from non-zero.
10931 if (range->len == total_bytes)
10932 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10934 cache = btrfs_lookup_block_group(fs_info, range->start);
10937 if (cache->key.objectid >= (range->start + range->len)) {
10938 btrfs_put_block_group(cache);
10942 start = max(range->start, cache->key.objectid);
10943 end = min(range->start + range->len,
10944 cache->key.objectid + cache->key.offset);
10946 if (end - start >= range->minlen) {
10947 if (!block_group_cache_done(cache)) {
10948 ret = cache_block_group(cache, 0);
10950 btrfs_put_block_group(cache);
10953 ret = wait_block_group_cache_done(cache);
10955 btrfs_put_block_group(cache);
10959 ret = btrfs_trim_block_group(cache,
10965 trimmed += group_trimmed;
10967 btrfs_put_block_group(cache);
10972 cache = next_block_group(fs_info, cache);
10975 mutex_lock(&fs_info->fs_devices->device_list_mutex);
10976 devices = &fs_info->fs_devices->alloc_list;
10977 list_for_each_entry(device, devices, dev_alloc_list) {
10978 ret = btrfs_trim_free_extents(device, range->minlen,
10983 trimmed += group_trimmed;
10985 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
10987 range->len = trimmed;
10992 * btrfs_{start,end}_write_no_snapshotting() are similar to
10993 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10994 * data into the page cache through nocow before the subvolume is snapshoted,
10995 * but flush the data into disk after the snapshot creation, or to prevent
10996 * operations while snapshotting is ongoing and that cause the snapshot to be
10997 * inconsistent (writes followed by expanding truncates for example).
10999 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11001 percpu_counter_dec(&root->subv_writers->counter);
11003 * Make sure counter is updated before we wake up waiters.
11006 if (waitqueue_active(&root->subv_writers->wait))
11007 wake_up(&root->subv_writers->wait);
11010 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11012 if (atomic_read(&root->will_be_snapshotted))
11015 percpu_counter_inc(&root->subv_writers->counter);
11017 * Make sure counter is updated before we check for snapshot creation.
11020 if (atomic_read(&root->will_be_snapshotted)) {
11021 btrfs_end_write_no_snapshotting(root);
11027 static int wait_snapshotting_atomic_t(atomic_t *a)
11033 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11038 ret = btrfs_start_write_no_snapshotting(root);
11041 wait_on_atomic_t(&root->will_be_snapshotted,
11042 wait_snapshotting_atomic_t,
11043 TASK_UNINTERRUPTIBLE);