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>
29 #include <linux/lockdep.h>
30 #include <linux/crc32c.h>
33 #include "print-tree.h"
37 #include "free-space-cache.h"
38 #include "free-space-tree.h"
42 #include "ref-verify.h"
44 #undef SCRAMBLE_DELAYED_REFS
47 * control flags for do_chunk_alloc's force field
48 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
49 * if we really need one.
51 * CHUNK_ALLOC_LIMITED means to only try and allocate one
52 * if we have very few chunks already allocated. This is
53 * used as part of the clustering code to help make sure
54 * we have a good pool of storage to cluster in, without
55 * filling the FS with empty chunks
57 * CHUNK_ALLOC_FORCE means it must try to allocate one
61 CHUNK_ALLOC_NO_FORCE = 0,
62 CHUNK_ALLOC_LIMITED = 1,
63 CHUNK_ALLOC_FORCE = 2,
66 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
67 struct btrfs_fs_info *fs_info,
68 struct btrfs_delayed_ref_node *node, u64 parent,
69 u64 root_objectid, u64 owner_objectid,
70 u64 owner_offset, int refs_to_drop,
71 struct btrfs_delayed_extent_op *extra_op);
72 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
73 struct extent_buffer *leaf,
74 struct btrfs_extent_item *ei);
75 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
76 struct btrfs_fs_info *fs_info,
77 u64 parent, u64 root_objectid,
78 u64 flags, u64 owner, u64 offset,
79 struct btrfs_key *ins, int ref_mod);
80 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
81 struct btrfs_fs_info *fs_info,
82 u64 parent, u64 root_objectid,
83 u64 flags, struct btrfs_disk_key *key,
84 int level, struct btrfs_key *ins);
85 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
86 struct btrfs_fs_info *fs_info, u64 flags,
88 static int find_next_key(struct btrfs_path *path, int level,
89 struct btrfs_key *key);
90 static void dump_space_info(struct btrfs_fs_info *fs_info,
91 struct btrfs_space_info *info, u64 bytes,
92 int dump_block_groups);
93 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
95 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
96 struct btrfs_space_info *space_info,
98 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
99 struct btrfs_space_info *space_info,
103 block_group_cache_done(struct btrfs_block_group_cache *cache)
106 return cache->cached == BTRFS_CACHE_FINISHED ||
107 cache->cached == BTRFS_CACHE_ERROR;
110 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
112 return (cache->flags & bits) == bits;
115 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
117 atomic_inc(&cache->count);
120 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
122 if (atomic_dec_and_test(&cache->count)) {
123 WARN_ON(cache->pinned > 0);
124 WARN_ON(cache->reserved > 0);
127 * If not empty, someone is still holding mutex of
128 * full_stripe_lock, which can only be released by caller.
129 * And it will definitely cause use-after-free when caller
130 * tries to release full stripe lock.
132 * No better way to resolve, but only to warn.
134 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
135 kfree(cache->free_space_ctl);
141 * this adds the block group to the fs_info rb tree for the block group
144 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
145 struct btrfs_block_group_cache *block_group)
148 struct rb_node *parent = NULL;
149 struct btrfs_block_group_cache *cache;
151 spin_lock(&info->block_group_cache_lock);
152 p = &info->block_group_cache_tree.rb_node;
156 cache = rb_entry(parent, struct btrfs_block_group_cache,
158 if (block_group->key.objectid < cache->key.objectid) {
160 } else if (block_group->key.objectid > cache->key.objectid) {
163 spin_unlock(&info->block_group_cache_lock);
168 rb_link_node(&block_group->cache_node, parent, p);
169 rb_insert_color(&block_group->cache_node,
170 &info->block_group_cache_tree);
172 if (info->first_logical_byte > block_group->key.objectid)
173 info->first_logical_byte = block_group->key.objectid;
175 spin_unlock(&info->block_group_cache_lock);
181 * This will return the block group at or after bytenr if contains is 0, else
182 * it will return the block group that contains the bytenr
184 static struct btrfs_block_group_cache *
185 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
188 struct btrfs_block_group_cache *cache, *ret = NULL;
192 spin_lock(&info->block_group_cache_lock);
193 n = info->block_group_cache_tree.rb_node;
196 cache = rb_entry(n, struct btrfs_block_group_cache,
198 end = cache->key.objectid + cache->key.offset - 1;
199 start = cache->key.objectid;
201 if (bytenr < start) {
202 if (!contains && (!ret || start < ret->key.objectid))
205 } else if (bytenr > start) {
206 if (contains && bytenr <= end) {
217 btrfs_get_block_group(ret);
218 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
219 info->first_logical_byte = ret->key.objectid;
221 spin_unlock(&info->block_group_cache_lock);
226 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
227 u64 start, u64 num_bytes)
229 u64 end = start + num_bytes - 1;
230 set_extent_bits(&fs_info->freed_extents[0],
231 start, end, EXTENT_UPTODATE);
232 set_extent_bits(&fs_info->freed_extents[1],
233 start, end, EXTENT_UPTODATE);
237 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
238 struct btrfs_block_group_cache *cache)
242 start = cache->key.objectid;
243 end = start + cache->key.offset - 1;
245 clear_extent_bits(&fs_info->freed_extents[0],
246 start, end, EXTENT_UPTODATE);
247 clear_extent_bits(&fs_info->freed_extents[1],
248 start, end, EXTENT_UPTODATE);
251 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
252 struct btrfs_block_group_cache *cache)
259 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
260 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
261 cache->bytes_super += stripe_len;
262 ret = add_excluded_extent(fs_info, cache->key.objectid,
268 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
269 bytenr = btrfs_sb_offset(i);
270 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
271 bytenr, 0, &logical, &nr, &stripe_len);
278 if (logical[nr] > cache->key.objectid +
282 if (logical[nr] + stripe_len <= cache->key.objectid)
286 if (start < cache->key.objectid) {
287 start = cache->key.objectid;
288 len = (logical[nr] + stripe_len) - start;
290 len = min_t(u64, stripe_len,
291 cache->key.objectid +
292 cache->key.offset - start);
295 cache->bytes_super += len;
296 ret = add_excluded_extent(fs_info, start, len);
308 static struct btrfs_caching_control *
309 get_caching_control(struct btrfs_block_group_cache *cache)
311 struct btrfs_caching_control *ctl;
313 spin_lock(&cache->lock);
314 if (!cache->caching_ctl) {
315 spin_unlock(&cache->lock);
319 ctl = cache->caching_ctl;
320 refcount_inc(&ctl->count);
321 spin_unlock(&cache->lock);
325 static void put_caching_control(struct btrfs_caching_control *ctl)
327 if (refcount_dec_and_test(&ctl->count))
331 #ifdef CONFIG_BTRFS_DEBUG
332 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
334 struct btrfs_fs_info *fs_info = block_group->fs_info;
335 u64 start = block_group->key.objectid;
336 u64 len = block_group->key.offset;
337 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
338 fs_info->nodesize : fs_info->sectorsize;
339 u64 step = chunk << 1;
341 while (len > chunk) {
342 btrfs_remove_free_space(block_group, start, chunk);
353 * this is only called by cache_block_group, since we could have freed extents
354 * we need to check the pinned_extents for any extents that can't be used yet
355 * since their free space will be released as soon as the transaction commits.
357 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
358 struct btrfs_fs_info *info, u64 start, u64 end)
360 u64 extent_start, extent_end, size, total_added = 0;
363 while (start < end) {
364 ret = find_first_extent_bit(info->pinned_extents, start,
365 &extent_start, &extent_end,
366 EXTENT_DIRTY | EXTENT_UPTODATE,
371 if (extent_start <= start) {
372 start = extent_end + 1;
373 } else if (extent_start > start && extent_start < end) {
374 size = extent_start - start;
376 ret = btrfs_add_free_space(block_group, start,
378 BUG_ON(ret); /* -ENOMEM or logic error */
379 start = extent_end + 1;
388 ret = btrfs_add_free_space(block_group, start, size);
389 BUG_ON(ret); /* -ENOMEM or logic error */
395 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
397 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
398 struct btrfs_fs_info *fs_info = block_group->fs_info;
399 struct btrfs_root *extent_root = fs_info->extent_root;
400 struct btrfs_path *path;
401 struct extent_buffer *leaf;
402 struct btrfs_key key;
409 path = btrfs_alloc_path();
413 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
415 #ifdef CONFIG_BTRFS_DEBUG
417 * If we're fragmenting we don't want to make anybody think we can
418 * allocate from this block group until we've had a chance to fragment
421 if (btrfs_should_fragment_free_space(block_group))
425 * We don't want to deadlock with somebody trying to allocate a new
426 * extent for the extent root while also trying to search the extent
427 * root to add free space. So we skip locking and search the commit
428 * root, since its read-only
430 path->skip_locking = 1;
431 path->search_commit_root = 1;
432 path->reada = READA_FORWARD;
436 key.type = BTRFS_EXTENT_ITEM_KEY;
439 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
443 leaf = path->nodes[0];
444 nritems = btrfs_header_nritems(leaf);
447 if (btrfs_fs_closing(fs_info) > 1) {
452 if (path->slots[0] < nritems) {
453 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
455 ret = find_next_key(path, 0, &key);
459 if (need_resched() ||
460 rwsem_is_contended(&fs_info->commit_root_sem)) {
462 caching_ctl->progress = last;
463 btrfs_release_path(path);
464 up_read(&fs_info->commit_root_sem);
465 mutex_unlock(&caching_ctl->mutex);
467 mutex_lock(&caching_ctl->mutex);
468 down_read(&fs_info->commit_root_sem);
472 ret = btrfs_next_leaf(extent_root, path);
477 leaf = path->nodes[0];
478 nritems = btrfs_header_nritems(leaf);
482 if (key.objectid < last) {
485 key.type = BTRFS_EXTENT_ITEM_KEY;
488 caching_ctl->progress = last;
489 btrfs_release_path(path);
493 if (key.objectid < block_group->key.objectid) {
498 if (key.objectid >= block_group->key.objectid +
499 block_group->key.offset)
502 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
503 key.type == BTRFS_METADATA_ITEM_KEY) {
504 total_found += add_new_free_space(block_group,
507 if (key.type == BTRFS_METADATA_ITEM_KEY)
508 last = key.objectid +
511 last = key.objectid + key.offset;
513 if (total_found > CACHING_CTL_WAKE_UP) {
516 wake_up(&caching_ctl->wait);
523 total_found += add_new_free_space(block_group, fs_info, last,
524 block_group->key.objectid +
525 block_group->key.offset);
526 caching_ctl->progress = (u64)-1;
529 btrfs_free_path(path);
533 static noinline void caching_thread(struct btrfs_work *work)
535 struct btrfs_block_group_cache *block_group;
536 struct btrfs_fs_info *fs_info;
537 struct btrfs_caching_control *caching_ctl;
540 caching_ctl = container_of(work, struct btrfs_caching_control, work);
541 block_group = caching_ctl->block_group;
542 fs_info = block_group->fs_info;
544 mutex_lock(&caching_ctl->mutex);
545 down_read(&fs_info->commit_root_sem);
547 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
548 ret = load_free_space_tree(caching_ctl);
550 ret = load_extent_tree_free(caching_ctl);
552 spin_lock(&block_group->lock);
553 block_group->caching_ctl = NULL;
554 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
555 spin_unlock(&block_group->lock);
557 #ifdef CONFIG_BTRFS_DEBUG
558 if (btrfs_should_fragment_free_space(block_group)) {
561 spin_lock(&block_group->space_info->lock);
562 spin_lock(&block_group->lock);
563 bytes_used = block_group->key.offset -
564 btrfs_block_group_used(&block_group->item);
565 block_group->space_info->bytes_used += bytes_used >> 1;
566 spin_unlock(&block_group->lock);
567 spin_unlock(&block_group->space_info->lock);
568 fragment_free_space(block_group);
572 caching_ctl->progress = (u64)-1;
574 up_read(&fs_info->commit_root_sem);
575 free_excluded_extents(fs_info, block_group);
576 mutex_unlock(&caching_ctl->mutex);
578 wake_up(&caching_ctl->wait);
580 put_caching_control(caching_ctl);
581 btrfs_put_block_group(block_group);
584 static int cache_block_group(struct btrfs_block_group_cache *cache,
588 struct btrfs_fs_info *fs_info = cache->fs_info;
589 struct btrfs_caching_control *caching_ctl;
592 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
596 INIT_LIST_HEAD(&caching_ctl->list);
597 mutex_init(&caching_ctl->mutex);
598 init_waitqueue_head(&caching_ctl->wait);
599 caching_ctl->block_group = cache;
600 caching_ctl->progress = cache->key.objectid;
601 refcount_set(&caching_ctl->count, 1);
602 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
603 caching_thread, NULL, NULL);
605 spin_lock(&cache->lock);
607 * This should be a rare occasion, but this could happen I think in the
608 * case where one thread starts to load the space cache info, and then
609 * some other thread starts a transaction commit which tries to do an
610 * allocation while the other thread is still loading the space cache
611 * info. The previous loop should have kept us from choosing this block
612 * group, but if we've moved to the state where we will wait on caching
613 * block groups we need to first check if we're doing a fast load here,
614 * so we can wait for it to finish, otherwise we could end up allocating
615 * from a block group who's cache gets evicted for one reason or
618 while (cache->cached == BTRFS_CACHE_FAST) {
619 struct btrfs_caching_control *ctl;
621 ctl = cache->caching_ctl;
622 refcount_inc(&ctl->count);
623 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
624 spin_unlock(&cache->lock);
628 finish_wait(&ctl->wait, &wait);
629 put_caching_control(ctl);
630 spin_lock(&cache->lock);
633 if (cache->cached != BTRFS_CACHE_NO) {
634 spin_unlock(&cache->lock);
638 WARN_ON(cache->caching_ctl);
639 cache->caching_ctl = caching_ctl;
640 cache->cached = BTRFS_CACHE_FAST;
641 spin_unlock(&cache->lock);
643 if (btrfs_test_opt(fs_info, SPACE_CACHE)) {
644 mutex_lock(&caching_ctl->mutex);
645 ret = load_free_space_cache(fs_info, cache);
647 spin_lock(&cache->lock);
649 cache->caching_ctl = NULL;
650 cache->cached = BTRFS_CACHE_FINISHED;
651 cache->last_byte_to_unpin = (u64)-1;
652 caching_ctl->progress = (u64)-1;
654 if (load_cache_only) {
655 cache->caching_ctl = NULL;
656 cache->cached = BTRFS_CACHE_NO;
658 cache->cached = BTRFS_CACHE_STARTED;
659 cache->has_caching_ctl = 1;
662 spin_unlock(&cache->lock);
663 #ifdef CONFIG_BTRFS_DEBUG
665 btrfs_should_fragment_free_space(cache)) {
668 spin_lock(&cache->space_info->lock);
669 spin_lock(&cache->lock);
670 bytes_used = cache->key.offset -
671 btrfs_block_group_used(&cache->item);
672 cache->space_info->bytes_used += bytes_used >> 1;
673 spin_unlock(&cache->lock);
674 spin_unlock(&cache->space_info->lock);
675 fragment_free_space(cache);
678 mutex_unlock(&caching_ctl->mutex);
680 wake_up(&caching_ctl->wait);
682 put_caching_control(caching_ctl);
683 free_excluded_extents(fs_info, cache);
688 * We're either using the free space tree or no caching at all.
689 * Set cached to the appropriate value and wakeup any waiters.
691 spin_lock(&cache->lock);
692 if (load_cache_only) {
693 cache->caching_ctl = NULL;
694 cache->cached = BTRFS_CACHE_NO;
696 cache->cached = BTRFS_CACHE_STARTED;
697 cache->has_caching_ctl = 1;
699 spin_unlock(&cache->lock);
700 wake_up(&caching_ctl->wait);
703 if (load_cache_only) {
704 put_caching_control(caching_ctl);
708 down_write(&fs_info->commit_root_sem);
709 refcount_inc(&caching_ctl->count);
710 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
711 up_write(&fs_info->commit_root_sem);
713 btrfs_get_block_group(cache);
715 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
721 * return the block group that starts at or after bytenr
723 static struct btrfs_block_group_cache *
724 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
726 return block_group_cache_tree_search(info, bytenr, 0);
730 * return the block group that contains the given bytenr
732 struct btrfs_block_group_cache *btrfs_lookup_block_group(
733 struct btrfs_fs_info *info,
736 return block_group_cache_tree_search(info, bytenr, 1);
739 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
742 struct list_head *head = &info->space_info;
743 struct btrfs_space_info *found;
745 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
748 list_for_each_entry_rcu(found, head, list) {
749 if (found->flags & flags) {
758 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes,
759 u64 owner, u64 root_objectid)
761 struct btrfs_space_info *space_info;
764 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
765 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
766 flags = BTRFS_BLOCK_GROUP_SYSTEM;
768 flags = BTRFS_BLOCK_GROUP_METADATA;
770 flags = BTRFS_BLOCK_GROUP_DATA;
773 space_info = __find_space_info(fs_info, flags);
775 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
779 * after adding space to the filesystem, we need to clear the full flags
780 * on all the space infos.
782 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
784 struct list_head *head = &info->space_info;
785 struct btrfs_space_info *found;
788 list_for_each_entry_rcu(found, head, list)
793 /* simple helper to search for an existing data extent at a given offset */
794 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
797 struct btrfs_key key;
798 struct btrfs_path *path;
800 path = btrfs_alloc_path();
804 key.objectid = start;
806 key.type = BTRFS_EXTENT_ITEM_KEY;
807 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
808 btrfs_free_path(path);
813 * helper function to lookup reference count and flags of a tree block.
815 * the head node for delayed ref is used to store the sum of all the
816 * reference count modifications queued up in the rbtree. the head
817 * node may also store the extent flags to set. This way you can check
818 * to see what the reference count and extent flags would be if all of
819 * the delayed refs are not processed.
821 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
822 struct btrfs_fs_info *fs_info, u64 bytenr,
823 u64 offset, int metadata, u64 *refs, u64 *flags)
825 struct btrfs_delayed_ref_head *head;
826 struct btrfs_delayed_ref_root *delayed_refs;
827 struct btrfs_path *path;
828 struct btrfs_extent_item *ei;
829 struct extent_buffer *leaf;
830 struct btrfs_key key;
837 * If we don't have skinny metadata, don't bother doing anything
840 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
841 offset = fs_info->nodesize;
845 path = btrfs_alloc_path();
850 path->skip_locking = 1;
851 path->search_commit_root = 1;
855 key.objectid = bytenr;
858 key.type = BTRFS_METADATA_ITEM_KEY;
860 key.type = BTRFS_EXTENT_ITEM_KEY;
862 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
866 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
867 if (path->slots[0]) {
869 btrfs_item_key_to_cpu(path->nodes[0], &key,
871 if (key.objectid == bytenr &&
872 key.type == BTRFS_EXTENT_ITEM_KEY &&
873 key.offset == fs_info->nodesize)
879 leaf = path->nodes[0];
880 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
881 if (item_size >= sizeof(*ei)) {
882 ei = btrfs_item_ptr(leaf, path->slots[0],
883 struct btrfs_extent_item);
884 num_refs = btrfs_extent_refs(leaf, ei);
885 extent_flags = btrfs_extent_flags(leaf, ei);
887 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
888 struct btrfs_extent_item_v0 *ei0;
889 BUG_ON(item_size != sizeof(*ei0));
890 ei0 = btrfs_item_ptr(leaf, path->slots[0],
891 struct btrfs_extent_item_v0);
892 num_refs = btrfs_extent_refs_v0(leaf, ei0);
893 /* FIXME: this isn't correct for data */
894 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
899 BUG_ON(num_refs == 0);
909 delayed_refs = &trans->transaction->delayed_refs;
910 spin_lock(&delayed_refs->lock);
911 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
913 if (!mutex_trylock(&head->mutex)) {
914 refcount_inc(&head->refs);
915 spin_unlock(&delayed_refs->lock);
917 btrfs_release_path(path);
920 * Mutex was contended, block until it's released and try
923 mutex_lock(&head->mutex);
924 mutex_unlock(&head->mutex);
925 btrfs_put_delayed_ref_head(head);
928 spin_lock(&head->lock);
929 if (head->extent_op && head->extent_op->update_flags)
930 extent_flags |= head->extent_op->flags_to_set;
932 BUG_ON(num_refs == 0);
934 num_refs += head->ref_mod;
935 spin_unlock(&head->lock);
936 mutex_unlock(&head->mutex);
938 spin_unlock(&delayed_refs->lock);
940 WARN_ON(num_refs == 0);
944 *flags = extent_flags;
946 btrfs_free_path(path);
951 * Back reference rules. Back refs have three main goals:
953 * 1) differentiate between all holders of references to an extent so that
954 * when a reference is dropped we can make sure it was a valid reference
955 * before freeing the extent.
957 * 2) Provide enough information to quickly find the holders of an extent
958 * if we notice a given block is corrupted or bad.
960 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
961 * maintenance. This is actually the same as #2, but with a slightly
962 * different use case.
964 * There are two kinds of back refs. The implicit back refs is optimized
965 * for pointers in non-shared tree blocks. For a given pointer in a block,
966 * back refs of this kind provide information about the block's owner tree
967 * and the pointer's key. These information allow us to find the block by
968 * b-tree searching. The full back refs is for pointers in tree blocks not
969 * referenced by their owner trees. The location of tree block is recorded
970 * in the back refs. Actually the full back refs is generic, and can be
971 * used in all cases the implicit back refs is used. The major shortcoming
972 * of the full back refs is its overhead. Every time a tree block gets
973 * COWed, we have to update back refs entry for all pointers in it.
975 * For a newly allocated tree block, we use implicit back refs for
976 * pointers in it. This means most tree related operations only involve
977 * implicit back refs. For a tree block created in old transaction, the
978 * only way to drop a reference to it is COW it. So we can detect the
979 * event that tree block loses its owner tree's reference and do the
980 * back refs conversion.
982 * When a tree block is COWed through a tree, there are four cases:
984 * The reference count of the block is one and the tree is the block's
985 * owner tree. Nothing to do in this case.
987 * The reference count of the block is one and the tree is not the
988 * block's owner tree. In this case, full back refs is used for pointers
989 * in the block. Remove these full back refs, add implicit back refs for
990 * every pointers in the new block.
992 * The reference count of the block is greater than one and the tree is
993 * the block's owner tree. In this case, implicit back refs is used for
994 * pointers in the block. Add full back refs for every pointers in the
995 * block, increase lower level extents' reference counts. The original
996 * implicit back refs are entailed to the new block.
998 * The reference count of the block is greater than one and the tree is
999 * not the block's owner tree. Add implicit back refs for every pointer in
1000 * the new block, increase lower level extents' reference count.
1002 * Back Reference Key composing:
1004 * The key objectid corresponds to the first byte in the extent,
1005 * The key type is used to differentiate between types of back refs.
1006 * There are different meanings of the key offset for different types
1009 * File extents can be referenced by:
1011 * - multiple snapshots, subvolumes, or different generations in one subvol
1012 * - different files inside a single subvolume
1013 * - different offsets inside a file (bookend extents in file.c)
1015 * The extent ref structure for the implicit back refs has fields for:
1017 * - Objectid of the subvolume root
1018 * - objectid of the file holding the reference
1019 * - original offset in the file
1020 * - how many bookend extents
1022 * The key offset for the implicit back refs is hash of the first
1025 * The extent ref structure for the full back refs has field for:
1027 * - number of pointers in the tree leaf
1029 * The key offset for the implicit back refs is the first byte of
1032 * When a file extent is allocated, The implicit back refs is used.
1033 * the fields are filled in:
1035 * (root_key.objectid, inode objectid, offset in file, 1)
1037 * When a file extent is removed file truncation, we find the
1038 * corresponding implicit back refs and check the following fields:
1040 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1042 * Btree extents can be referenced by:
1044 * - Different subvolumes
1046 * Both the implicit back refs and the full back refs for tree blocks
1047 * only consist of key. The key offset for the implicit back refs is
1048 * objectid of block's owner tree. The key offset for the full back refs
1049 * is the first byte of parent block.
1051 * When implicit back refs is used, information about the lowest key and
1052 * level of the tree block are required. These information are stored in
1053 * tree block info structure.
1056 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1057 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1058 struct btrfs_fs_info *fs_info,
1059 struct btrfs_path *path,
1060 u64 owner, u32 extra_size)
1062 struct btrfs_root *root = fs_info->extent_root;
1063 struct btrfs_extent_item *item;
1064 struct btrfs_extent_item_v0 *ei0;
1065 struct btrfs_extent_ref_v0 *ref0;
1066 struct btrfs_tree_block_info *bi;
1067 struct extent_buffer *leaf;
1068 struct btrfs_key key;
1069 struct btrfs_key found_key;
1070 u32 new_size = sizeof(*item);
1074 leaf = path->nodes[0];
1075 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1077 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1078 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1079 struct btrfs_extent_item_v0);
1080 refs = btrfs_extent_refs_v0(leaf, ei0);
1082 if (owner == (u64)-1) {
1084 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1085 ret = btrfs_next_leaf(root, path);
1088 BUG_ON(ret > 0); /* Corruption */
1089 leaf = path->nodes[0];
1091 btrfs_item_key_to_cpu(leaf, &found_key,
1093 BUG_ON(key.objectid != found_key.objectid);
1094 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1098 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1099 struct btrfs_extent_ref_v0);
1100 owner = btrfs_ref_objectid_v0(leaf, ref0);
1104 btrfs_release_path(path);
1106 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1107 new_size += sizeof(*bi);
1109 new_size -= sizeof(*ei0);
1110 ret = btrfs_search_slot(trans, root, &key, path,
1111 new_size + extra_size, 1);
1114 BUG_ON(ret); /* Corruption */
1116 btrfs_extend_item(fs_info, path, new_size);
1118 leaf = path->nodes[0];
1119 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1120 btrfs_set_extent_refs(leaf, item, refs);
1121 /* FIXME: get real generation */
1122 btrfs_set_extent_generation(leaf, item, 0);
1123 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1124 btrfs_set_extent_flags(leaf, item,
1125 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1126 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1127 bi = (struct btrfs_tree_block_info *)(item + 1);
1128 /* FIXME: get first key of the block */
1129 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1130 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1132 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1134 btrfs_mark_buffer_dirty(leaf);
1140 * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required,
1141 * is_data == BTRFS_REF_TYPE_DATA, data type is requried,
1142 * is_data == BTRFS_REF_TYPE_ANY, either type is OK.
1144 int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb,
1145 struct btrfs_extent_inline_ref *iref,
1146 enum btrfs_inline_ref_type is_data)
1148 int type = btrfs_extent_inline_ref_type(eb, iref);
1149 u64 offset = btrfs_extent_inline_ref_offset(eb, iref);
1151 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1152 type == BTRFS_SHARED_BLOCK_REF_KEY ||
1153 type == BTRFS_SHARED_DATA_REF_KEY ||
1154 type == BTRFS_EXTENT_DATA_REF_KEY) {
1155 if (is_data == BTRFS_REF_TYPE_BLOCK) {
1156 if (type == BTRFS_TREE_BLOCK_REF_KEY)
1158 if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1159 ASSERT(eb->fs_info);
1161 * Every shared one has parent tree
1162 * block, which must be aligned to
1166 IS_ALIGNED(offset, eb->fs_info->nodesize))
1169 } else if (is_data == BTRFS_REF_TYPE_DATA) {
1170 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1172 if (type == BTRFS_SHARED_DATA_REF_KEY) {
1173 ASSERT(eb->fs_info);
1175 * Every shared one has parent tree
1176 * block, which must be aligned to
1180 IS_ALIGNED(offset, eb->fs_info->nodesize))
1184 ASSERT(is_data == BTRFS_REF_TYPE_ANY);
1189 btrfs_print_leaf((struct extent_buffer *)eb);
1190 btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d",
1194 return BTRFS_REF_TYPE_INVALID;
1197 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1199 u32 high_crc = ~(u32)0;
1200 u32 low_crc = ~(u32)0;
1203 lenum = cpu_to_le64(root_objectid);
1204 high_crc = crc32c(high_crc, &lenum, sizeof(lenum));
1205 lenum = cpu_to_le64(owner);
1206 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1207 lenum = cpu_to_le64(offset);
1208 low_crc = crc32c(low_crc, &lenum, sizeof(lenum));
1210 return ((u64)high_crc << 31) ^ (u64)low_crc;
1213 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1214 struct btrfs_extent_data_ref *ref)
1216 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1217 btrfs_extent_data_ref_objectid(leaf, ref),
1218 btrfs_extent_data_ref_offset(leaf, ref));
1221 static int match_extent_data_ref(struct extent_buffer *leaf,
1222 struct btrfs_extent_data_ref *ref,
1223 u64 root_objectid, u64 owner, u64 offset)
1225 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1226 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1227 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1232 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1233 struct btrfs_fs_info *fs_info,
1234 struct btrfs_path *path,
1235 u64 bytenr, u64 parent,
1237 u64 owner, u64 offset)
1239 struct btrfs_root *root = fs_info->extent_root;
1240 struct btrfs_key key;
1241 struct btrfs_extent_data_ref *ref;
1242 struct extent_buffer *leaf;
1248 key.objectid = bytenr;
1250 key.type = BTRFS_SHARED_DATA_REF_KEY;
1251 key.offset = parent;
1253 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1254 key.offset = hash_extent_data_ref(root_objectid,
1259 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1268 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1269 key.type = BTRFS_EXTENT_REF_V0_KEY;
1270 btrfs_release_path(path);
1271 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1282 leaf = path->nodes[0];
1283 nritems = btrfs_header_nritems(leaf);
1285 if (path->slots[0] >= nritems) {
1286 ret = btrfs_next_leaf(root, path);
1292 leaf = path->nodes[0];
1293 nritems = btrfs_header_nritems(leaf);
1297 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1298 if (key.objectid != bytenr ||
1299 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1302 ref = btrfs_item_ptr(leaf, path->slots[0],
1303 struct btrfs_extent_data_ref);
1305 if (match_extent_data_ref(leaf, ref, root_objectid,
1308 btrfs_release_path(path);
1320 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1321 struct btrfs_fs_info *fs_info,
1322 struct btrfs_path *path,
1323 u64 bytenr, u64 parent,
1324 u64 root_objectid, u64 owner,
1325 u64 offset, int refs_to_add)
1327 struct btrfs_root *root = fs_info->extent_root;
1328 struct btrfs_key key;
1329 struct extent_buffer *leaf;
1334 key.objectid = bytenr;
1336 key.type = BTRFS_SHARED_DATA_REF_KEY;
1337 key.offset = parent;
1338 size = sizeof(struct btrfs_shared_data_ref);
1340 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1341 key.offset = hash_extent_data_ref(root_objectid,
1343 size = sizeof(struct btrfs_extent_data_ref);
1346 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1347 if (ret && ret != -EEXIST)
1350 leaf = path->nodes[0];
1352 struct btrfs_shared_data_ref *ref;
1353 ref = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_shared_data_ref);
1356 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1358 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1359 num_refs += refs_to_add;
1360 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1363 struct btrfs_extent_data_ref *ref;
1364 while (ret == -EEXIST) {
1365 ref = btrfs_item_ptr(leaf, path->slots[0],
1366 struct btrfs_extent_data_ref);
1367 if (match_extent_data_ref(leaf, ref, root_objectid,
1370 btrfs_release_path(path);
1372 ret = btrfs_insert_empty_item(trans, root, path, &key,
1374 if (ret && ret != -EEXIST)
1377 leaf = path->nodes[0];
1379 ref = btrfs_item_ptr(leaf, path->slots[0],
1380 struct btrfs_extent_data_ref);
1382 btrfs_set_extent_data_ref_root(leaf, ref,
1384 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1385 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1386 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1388 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1389 num_refs += refs_to_add;
1390 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1393 btrfs_mark_buffer_dirty(leaf);
1396 btrfs_release_path(path);
1400 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1401 struct btrfs_fs_info *fs_info,
1402 struct btrfs_path *path,
1403 int refs_to_drop, int *last_ref)
1405 struct btrfs_key key;
1406 struct btrfs_extent_data_ref *ref1 = NULL;
1407 struct btrfs_shared_data_ref *ref2 = NULL;
1408 struct extent_buffer *leaf;
1412 leaf = path->nodes[0];
1413 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1415 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1416 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1417 struct btrfs_extent_data_ref);
1418 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1419 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1420 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1421 struct btrfs_shared_data_ref);
1422 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1423 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1424 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1425 struct btrfs_extent_ref_v0 *ref0;
1426 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1427 struct btrfs_extent_ref_v0);
1428 num_refs = btrfs_ref_count_v0(leaf, ref0);
1434 BUG_ON(num_refs < refs_to_drop);
1435 num_refs -= refs_to_drop;
1437 if (num_refs == 0) {
1438 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1441 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1442 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1443 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1444 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1445 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1447 struct btrfs_extent_ref_v0 *ref0;
1448 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1449 struct btrfs_extent_ref_v0);
1450 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1453 btrfs_mark_buffer_dirty(leaf);
1458 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1459 struct btrfs_extent_inline_ref *iref)
1461 struct btrfs_key key;
1462 struct extent_buffer *leaf;
1463 struct btrfs_extent_data_ref *ref1;
1464 struct btrfs_shared_data_ref *ref2;
1468 leaf = path->nodes[0];
1469 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1472 * If type is invalid, we should have bailed out earlier than
1475 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
1476 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1477 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1478 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1479 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1481 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1482 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1484 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1485 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1486 struct btrfs_extent_data_ref);
1487 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1488 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1489 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1490 struct btrfs_shared_data_ref);
1491 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1492 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1493 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1494 struct btrfs_extent_ref_v0 *ref0;
1495 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1496 struct btrfs_extent_ref_v0);
1497 num_refs = btrfs_ref_count_v0(leaf, ref0);
1505 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1506 struct btrfs_fs_info *fs_info,
1507 struct btrfs_path *path,
1508 u64 bytenr, u64 parent,
1511 struct btrfs_root *root = fs_info->extent_root;
1512 struct btrfs_key key;
1515 key.objectid = bytenr;
1517 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1518 key.offset = parent;
1520 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1521 key.offset = root_objectid;
1524 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1527 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1528 if (ret == -ENOENT && parent) {
1529 btrfs_release_path(path);
1530 key.type = BTRFS_EXTENT_REF_V0_KEY;
1531 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1539 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1540 struct btrfs_fs_info *fs_info,
1541 struct btrfs_path *path,
1542 u64 bytenr, u64 parent,
1545 struct btrfs_key key;
1548 key.objectid = bytenr;
1550 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1551 key.offset = parent;
1553 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1554 key.offset = root_objectid;
1557 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1559 btrfs_release_path(path);
1563 static inline int extent_ref_type(u64 parent, u64 owner)
1566 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1568 type = BTRFS_SHARED_BLOCK_REF_KEY;
1570 type = BTRFS_TREE_BLOCK_REF_KEY;
1573 type = BTRFS_SHARED_DATA_REF_KEY;
1575 type = BTRFS_EXTENT_DATA_REF_KEY;
1580 static int find_next_key(struct btrfs_path *path, int level,
1581 struct btrfs_key *key)
1584 for (; level < BTRFS_MAX_LEVEL; level++) {
1585 if (!path->nodes[level])
1587 if (path->slots[level] + 1 >=
1588 btrfs_header_nritems(path->nodes[level]))
1591 btrfs_item_key_to_cpu(path->nodes[level], key,
1592 path->slots[level] + 1);
1594 btrfs_node_key_to_cpu(path->nodes[level], key,
1595 path->slots[level] + 1);
1602 * look for inline back ref. if back ref is found, *ref_ret is set
1603 * to the address of inline back ref, and 0 is returned.
1605 * if back ref isn't found, *ref_ret is set to the address where it
1606 * should be inserted, and -ENOENT is returned.
1608 * if insert is true and there are too many inline back refs, the path
1609 * points to the extent item, and -EAGAIN is returned.
1611 * NOTE: inline back refs are ordered in the same way that back ref
1612 * items in the tree are ordered.
1614 static noinline_for_stack
1615 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1616 struct btrfs_fs_info *fs_info,
1617 struct btrfs_path *path,
1618 struct btrfs_extent_inline_ref **ref_ret,
1619 u64 bytenr, u64 num_bytes,
1620 u64 parent, u64 root_objectid,
1621 u64 owner, u64 offset, int insert)
1623 struct btrfs_root *root = fs_info->extent_root;
1624 struct btrfs_key key;
1625 struct extent_buffer *leaf;
1626 struct btrfs_extent_item *ei;
1627 struct btrfs_extent_inline_ref *iref;
1637 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1640 key.objectid = bytenr;
1641 key.type = BTRFS_EXTENT_ITEM_KEY;
1642 key.offset = num_bytes;
1644 want = extent_ref_type(parent, owner);
1646 extra_size = btrfs_extent_inline_ref_size(want);
1647 path->keep_locks = 1;
1652 * Owner is our parent level, so we can just add one to get the level
1653 * for the block we are interested in.
1655 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1656 key.type = BTRFS_METADATA_ITEM_KEY;
1661 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1668 * We may be a newly converted file system which still has the old fat
1669 * extent entries for metadata, so try and see if we have one of those.
1671 if (ret > 0 && skinny_metadata) {
1672 skinny_metadata = false;
1673 if (path->slots[0]) {
1675 btrfs_item_key_to_cpu(path->nodes[0], &key,
1677 if (key.objectid == bytenr &&
1678 key.type == BTRFS_EXTENT_ITEM_KEY &&
1679 key.offset == num_bytes)
1683 key.objectid = bytenr;
1684 key.type = BTRFS_EXTENT_ITEM_KEY;
1685 key.offset = num_bytes;
1686 btrfs_release_path(path);
1691 if (ret && !insert) {
1694 } else if (WARN_ON(ret)) {
1699 leaf = path->nodes[0];
1700 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1701 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1702 if (item_size < sizeof(*ei)) {
1707 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1713 leaf = path->nodes[0];
1714 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1717 BUG_ON(item_size < sizeof(*ei));
1719 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1720 flags = btrfs_extent_flags(leaf, ei);
1722 ptr = (unsigned long)(ei + 1);
1723 end = (unsigned long)ei + item_size;
1725 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1726 ptr += sizeof(struct btrfs_tree_block_info);
1730 if (owner >= BTRFS_FIRST_FREE_OBJECTID)
1731 needed = BTRFS_REF_TYPE_DATA;
1733 needed = BTRFS_REF_TYPE_BLOCK;
1741 iref = (struct btrfs_extent_inline_ref *)ptr;
1742 type = btrfs_get_extent_inline_ref_type(leaf, iref, needed);
1743 if (type == BTRFS_REF_TYPE_INVALID) {
1751 ptr += btrfs_extent_inline_ref_size(type);
1755 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1756 struct btrfs_extent_data_ref *dref;
1757 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1758 if (match_extent_data_ref(leaf, dref, root_objectid,
1763 if (hash_extent_data_ref_item(leaf, dref) <
1764 hash_extent_data_ref(root_objectid, owner, offset))
1768 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1770 if (parent == ref_offset) {
1774 if (ref_offset < parent)
1777 if (root_objectid == ref_offset) {
1781 if (ref_offset < root_objectid)
1785 ptr += btrfs_extent_inline_ref_size(type);
1787 if (err == -ENOENT && insert) {
1788 if (item_size + extra_size >=
1789 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1794 * To add new inline back ref, we have to make sure
1795 * there is no corresponding back ref item.
1796 * For simplicity, we just do not add new inline back
1797 * ref if there is any kind of item for this block
1799 if (find_next_key(path, 0, &key) == 0 &&
1800 key.objectid == bytenr &&
1801 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1806 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1809 path->keep_locks = 0;
1810 btrfs_unlock_up_safe(path, 1);
1816 * helper to add new inline back ref
1818 static noinline_for_stack
1819 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1820 struct btrfs_path *path,
1821 struct btrfs_extent_inline_ref *iref,
1822 u64 parent, u64 root_objectid,
1823 u64 owner, u64 offset, int refs_to_add,
1824 struct btrfs_delayed_extent_op *extent_op)
1826 struct extent_buffer *leaf;
1827 struct btrfs_extent_item *ei;
1830 unsigned long item_offset;
1835 leaf = path->nodes[0];
1836 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1837 item_offset = (unsigned long)iref - (unsigned long)ei;
1839 type = extent_ref_type(parent, owner);
1840 size = btrfs_extent_inline_ref_size(type);
1842 btrfs_extend_item(fs_info, path, size);
1844 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1845 refs = btrfs_extent_refs(leaf, ei);
1846 refs += refs_to_add;
1847 btrfs_set_extent_refs(leaf, ei, refs);
1849 __run_delayed_extent_op(extent_op, leaf, ei);
1851 ptr = (unsigned long)ei + item_offset;
1852 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1853 if (ptr < end - size)
1854 memmove_extent_buffer(leaf, ptr + size, ptr,
1857 iref = (struct btrfs_extent_inline_ref *)ptr;
1858 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1859 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1860 struct btrfs_extent_data_ref *dref;
1861 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1862 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1863 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1864 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1865 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1866 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1867 struct btrfs_shared_data_ref *sref;
1868 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1869 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1870 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1871 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1872 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1874 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1876 btrfs_mark_buffer_dirty(leaf);
1879 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1880 struct btrfs_fs_info *fs_info,
1881 struct btrfs_path *path,
1882 struct btrfs_extent_inline_ref **ref_ret,
1883 u64 bytenr, u64 num_bytes, u64 parent,
1884 u64 root_objectid, u64 owner, u64 offset)
1888 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1889 bytenr, num_bytes, parent,
1890 root_objectid, owner, offset, 0);
1894 btrfs_release_path(path);
1897 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1898 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1899 parent, root_objectid);
1901 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1902 parent, root_objectid, owner,
1909 * helper to update/remove inline back ref
1911 static noinline_for_stack
1912 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1913 struct btrfs_path *path,
1914 struct btrfs_extent_inline_ref *iref,
1916 struct btrfs_delayed_extent_op *extent_op,
1919 struct extent_buffer *leaf;
1920 struct btrfs_extent_item *ei;
1921 struct btrfs_extent_data_ref *dref = NULL;
1922 struct btrfs_shared_data_ref *sref = NULL;
1930 leaf = path->nodes[0];
1931 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1932 refs = btrfs_extent_refs(leaf, ei);
1933 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1934 refs += refs_to_mod;
1935 btrfs_set_extent_refs(leaf, ei, refs);
1937 __run_delayed_extent_op(extent_op, leaf, ei);
1940 * If type is invalid, we should have bailed out after
1941 * lookup_inline_extent_backref().
1943 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY);
1944 ASSERT(type != BTRFS_REF_TYPE_INVALID);
1946 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1947 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1948 refs = btrfs_extent_data_ref_count(leaf, dref);
1949 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1950 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1951 refs = btrfs_shared_data_ref_count(leaf, sref);
1954 BUG_ON(refs_to_mod != -1);
1957 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1958 refs += refs_to_mod;
1961 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1962 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1964 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1967 size = btrfs_extent_inline_ref_size(type);
1968 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1969 ptr = (unsigned long)iref;
1970 end = (unsigned long)ei + item_size;
1971 if (ptr + size < end)
1972 memmove_extent_buffer(leaf, ptr, ptr + size,
1975 btrfs_truncate_item(fs_info, path, item_size, 1);
1977 btrfs_mark_buffer_dirty(leaf);
1980 static noinline_for_stack
1981 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1982 struct btrfs_fs_info *fs_info,
1983 struct btrfs_path *path,
1984 u64 bytenr, u64 num_bytes, u64 parent,
1985 u64 root_objectid, u64 owner,
1986 u64 offset, int refs_to_add,
1987 struct btrfs_delayed_extent_op *extent_op)
1989 struct btrfs_extent_inline_ref *iref;
1992 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1993 bytenr, num_bytes, parent,
1994 root_objectid, owner, offset, 1);
1996 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1997 update_inline_extent_backref(fs_info, path, iref,
1998 refs_to_add, extent_op, NULL);
1999 } else if (ret == -ENOENT) {
2000 setup_inline_extent_backref(fs_info, path, iref, parent,
2001 root_objectid, owner, offset,
2002 refs_to_add, extent_op);
2008 static int insert_extent_backref(struct btrfs_trans_handle *trans,
2009 struct btrfs_fs_info *fs_info,
2010 struct btrfs_path *path,
2011 u64 bytenr, u64 parent, u64 root_objectid,
2012 u64 owner, u64 offset, int refs_to_add)
2015 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2016 BUG_ON(refs_to_add != 1);
2017 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
2018 parent, root_objectid);
2020 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
2021 parent, root_objectid,
2022 owner, offset, refs_to_add);
2027 static int remove_extent_backref(struct btrfs_trans_handle *trans,
2028 struct btrfs_fs_info *fs_info,
2029 struct btrfs_path *path,
2030 struct btrfs_extent_inline_ref *iref,
2031 int refs_to_drop, int is_data, int *last_ref)
2035 BUG_ON(!is_data && refs_to_drop != 1);
2037 update_inline_extent_backref(fs_info, path, iref,
2038 -refs_to_drop, NULL, last_ref);
2039 } else if (is_data) {
2040 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
2044 ret = btrfs_del_item(trans, fs_info->extent_root, path);
2049 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
2050 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
2051 u64 *discarded_bytes)
2054 u64 bytes_left, end;
2055 u64 aligned_start = ALIGN(start, 1 << 9);
2057 if (WARN_ON(start != aligned_start)) {
2058 len -= aligned_start - start;
2059 len = round_down(len, 1 << 9);
2060 start = aligned_start;
2063 *discarded_bytes = 0;
2071 /* Skip any superblocks on this device. */
2072 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
2073 u64 sb_start = btrfs_sb_offset(j);
2074 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2075 u64 size = sb_start - start;
2077 if (!in_range(sb_start, start, bytes_left) &&
2078 !in_range(sb_end, start, bytes_left) &&
2079 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2083 * Superblock spans beginning of range. Adjust start and
2086 if (sb_start <= start) {
2087 start += sb_end - start;
2092 bytes_left = end - start;
2097 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2100 *discarded_bytes += size;
2101 else if (ret != -EOPNOTSUPP)
2110 bytes_left = end - start;
2114 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2117 *discarded_bytes += bytes_left;
2122 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2123 u64 num_bytes, u64 *actual_bytes)
2126 u64 discarded_bytes = 0;
2127 struct btrfs_bio *bbio = NULL;
2131 * Avoid races with device replace and make sure our bbio has devices
2132 * associated to its stripes that don't go away while we are discarding.
2134 btrfs_bio_counter_inc_blocked(fs_info);
2135 /* Tell the block device(s) that the sectors can be discarded */
2136 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2138 /* Error condition is -ENOMEM */
2140 struct btrfs_bio_stripe *stripe = bbio->stripes;
2144 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2146 struct request_queue *req_q;
2148 if (!stripe->dev->bdev) {
2149 ASSERT(btrfs_test_opt(fs_info, DEGRADED));
2152 req_q = bdev_get_queue(stripe->dev->bdev);
2153 if (!blk_queue_discard(req_q))
2156 ret = btrfs_issue_discard(stripe->dev->bdev,
2161 discarded_bytes += bytes;
2162 else if (ret != -EOPNOTSUPP)
2163 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2166 * Just in case we get back EOPNOTSUPP for some reason,
2167 * just ignore the return value so we don't screw up
2168 * people calling discard_extent.
2172 btrfs_put_bbio(bbio);
2174 btrfs_bio_counter_dec(fs_info);
2177 *actual_bytes = discarded_bytes;
2180 if (ret == -EOPNOTSUPP)
2185 /* Can return -ENOMEM */
2186 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2187 struct btrfs_root *root,
2188 u64 bytenr, u64 num_bytes, u64 parent,
2189 u64 root_objectid, u64 owner, u64 offset)
2191 struct btrfs_fs_info *fs_info = root->fs_info;
2192 int old_ref_mod, new_ref_mod;
2195 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2196 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2198 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid,
2199 owner, offset, BTRFS_ADD_DELAYED_REF);
2201 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2202 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2204 root_objectid, (int)owner,
2205 BTRFS_ADD_DELAYED_REF, NULL,
2206 &old_ref_mod, &new_ref_mod);
2208 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2210 root_objectid, owner, offset,
2211 0, BTRFS_ADD_DELAYED_REF,
2212 &old_ref_mod, &new_ref_mod);
2215 if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0)
2216 add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid);
2221 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2222 struct btrfs_fs_info *fs_info,
2223 struct btrfs_delayed_ref_node *node,
2224 u64 parent, u64 root_objectid,
2225 u64 owner, u64 offset, int refs_to_add,
2226 struct btrfs_delayed_extent_op *extent_op)
2228 struct btrfs_path *path;
2229 struct extent_buffer *leaf;
2230 struct btrfs_extent_item *item;
2231 struct btrfs_key key;
2232 u64 bytenr = node->bytenr;
2233 u64 num_bytes = node->num_bytes;
2237 path = btrfs_alloc_path();
2241 path->reada = READA_FORWARD;
2242 path->leave_spinning = 1;
2243 /* this will setup the path even if it fails to insert the back ref */
2244 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2245 num_bytes, parent, root_objectid,
2247 refs_to_add, extent_op);
2248 if ((ret < 0 && ret != -EAGAIN) || !ret)
2252 * Ok we had -EAGAIN which means we didn't have space to insert and
2253 * inline extent ref, so just update the reference count and add a
2256 leaf = path->nodes[0];
2257 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2258 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2259 refs = btrfs_extent_refs(leaf, item);
2260 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2262 __run_delayed_extent_op(extent_op, leaf, item);
2264 btrfs_mark_buffer_dirty(leaf);
2265 btrfs_release_path(path);
2267 path->reada = READA_FORWARD;
2268 path->leave_spinning = 1;
2269 /* now insert the actual backref */
2270 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2271 root_objectid, owner, offset, refs_to_add);
2273 btrfs_abort_transaction(trans, ret);
2275 btrfs_free_path(path);
2279 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2280 struct btrfs_fs_info *fs_info,
2281 struct btrfs_delayed_ref_node *node,
2282 struct btrfs_delayed_extent_op *extent_op,
2283 int insert_reserved)
2286 struct btrfs_delayed_data_ref *ref;
2287 struct btrfs_key ins;
2292 ins.objectid = node->bytenr;
2293 ins.offset = node->num_bytes;
2294 ins.type = BTRFS_EXTENT_ITEM_KEY;
2296 ref = btrfs_delayed_node_to_data_ref(node);
2297 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2299 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2300 parent = ref->parent;
2301 ref_root = ref->root;
2303 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2305 flags |= extent_op->flags_to_set;
2306 ret = alloc_reserved_file_extent(trans, fs_info,
2307 parent, ref_root, flags,
2308 ref->objectid, ref->offset,
2309 &ins, node->ref_mod);
2310 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2311 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2312 ref_root, ref->objectid,
2313 ref->offset, node->ref_mod,
2315 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2316 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2317 ref_root, ref->objectid,
2318 ref->offset, node->ref_mod,
2326 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2327 struct extent_buffer *leaf,
2328 struct btrfs_extent_item *ei)
2330 u64 flags = btrfs_extent_flags(leaf, ei);
2331 if (extent_op->update_flags) {
2332 flags |= extent_op->flags_to_set;
2333 btrfs_set_extent_flags(leaf, ei, flags);
2336 if (extent_op->update_key) {
2337 struct btrfs_tree_block_info *bi;
2338 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2339 bi = (struct btrfs_tree_block_info *)(ei + 1);
2340 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2344 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2345 struct btrfs_fs_info *fs_info,
2346 struct btrfs_delayed_ref_head *head,
2347 struct btrfs_delayed_extent_op *extent_op)
2349 struct btrfs_key key;
2350 struct btrfs_path *path;
2351 struct btrfs_extent_item *ei;
2352 struct extent_buffer *leaf;
2356 int metadata = !extent_op->is_data;
2361 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2364 path = btrfs_alloc_path();
2368 key.objectid = head->bytenr;
2371 key.type = BTRFS_METADATA_ITEM_KEY;
2372 key.offset = extent_op->level;
2374 key.type = BTRFS_EXTENT_ITEM_KEY;
2375 key.offset = head->num_bytes;
2379 path->reada = READA_FORWARD;
2380 path->leave_spinning = 1;
2381 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2388 if (path->slots[0] > 0) {
2390 btrfs_item_key_to_cpu(path->nodes[0], &key,
2392 if (key.objectid == head->bytenr &&
2393 key.type == BTRFS_EXTENT_ITEM_KEY &&
2394 key.offset == head->num_bytes)
2398 btrfs_release_path(path);
2401 key.objectid = head->bytenr;
2402 key.offset = head->num_bytes;
2403 key.type = BTRFS_EXTENT_ITEM_KEY;
2412 leaf = path->nodes[0];
2413 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2414 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2415 if (item_size < sizeof(*ei)) {
2416 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2421 leaf = path->nodes[0];
2422 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2425 BUG_ON(item_size < sizeof(*ei));
2426 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2427 __run_delayed_extent_op(extent_op, leaf, ei);
2429 btrfs_mark_buffer_dirty(leaf);
2431 btrfs_free_path(path);
2435 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2436 struct btrfs_fs_info *fs_info,
2437 struct btrfs_delayed_ref_node *node,
2438 struct btrfs_delayed_extent_op *extent_op,
2439 int insert_reserved)
2442 struct btrfs_delayed_tree_ref *ref;
2443 struct btrfs_key ins;
2446 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2448 ref = btrfs_delayed_node_to_tree_ref(node);
2449 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2451 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2452 parent = ref->parent;
2453 ref_root = ref->root;
2455 ins.objectid = node->bytenr;
2456 if (skinny_metadata) {
2457 ins.offset = ref->level;
2458 ins.type = BTRFS_METADATA_ITEM_KEY;
2460 ins.offset = node->num_bytes;
2461 ins.type = BTRFS_EXTENT_ITEM_KEY;
2464 if (node->ref_mod != 1) {
2466 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2467 node->bytenr, node->ref_mod, node->action, ref_root,
2471 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2472 BUG_ON(!extent_op || !extent_op->update_flags);
2473 ret = alloc_reserved_tree_block(trans, fs_info,
2475 extent_op->flags_to_set,
2478 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2479 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2483 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2484 ret = __btrfs_free_extent(trans, fs_info, node,
2486 ref->level, 0, 1, extent_op);
2493 /* helper function to actually process a single delayed ref entry */
2494 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2495 struct btrfs_fs_info *fs_info,
2496 struct btrfs_delayed_ref_node *node,
2497 struct btrfs_delayed_extent_op *extent_op,
2498 int insert_reserved)
2502 if (trans->aborted) {
2503 if (insert_reserved)
2504 btrfs_pin_extent(fs_info, node->bytenr,
2505 node->num_bytes, 1);
2509 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2510 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2511 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2513 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2514 node->type == BTRFS_SHARED_DATA_REF_KEY)
2515 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2522 static inline struct btrfs_delayed_ref_node *
2523 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2525 struct btrfs_delayed_ref_node *ref;
2527 if (RB_EMPTY_ROOT(&head->ref_tree))
2531 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2532 * This is to prevent a ref count from going down to zero, which deletes
2533 * the extent item from the extent tree, when there still are references
2534 * to add, which would fail because they would not find the extent item.
2536 if (!list_empty(&head->ref_add_list))
2537 return list_first_entry(&head->ref_add_list,
2538 struct btrfs_delayed_ref_node, add_list);
2540 ref = rb_entry(rb_first(&head->ref_tree),
2541 struct btrfs_delayed_ref_node, ref_node);
2542 ASSERT(list_empty(&ref->add_list));
2546 static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs,
2547 struct btrfs_delayed_ref_head *head)
2549 spin_lock(&delayed_refs->lock);
2550 head->processing = 0;
2551 delayed_refs->num_heads_ready++;
2552 spin_unlock(&delayed_refs->lock);
2553 btrfs_delayed_ref_unlock(head);
2556 static int cleanup_extent_op(struct btrfs_trans_handle *trans,
2557 struct btrfs_fs_info *fs_info,
2558 struct btrfs_delayed_ref_head *head)
2560 struct btrfs_delayed_extent_op *extent_op = head->extent_op;
2565 head->extent_op = NULL;
2566 if (head->must_insert_reserved) {
2567 btrfs_free_delayed_extent_op(extent_op);
2570 spin_unlock(&head->lock);
2571 ret = run_delayed_extent_op(trans, fs_info, head, extent_op);
2572 btrfs_free_delayed_extent_op(extent_op);
2573 return ret ? ret : 1;
2576 static int cleanup_ref_head(struct btrfs_trans_handle *trans,
2577 struct btrfs_fs_info *fs_info,
2578 struct btrfs_delayed_ref_head *head)
2580 struct btrfs_delayed_ref_root *delayed_refs;
2583 delayed_refs = &trans->transaction->delayed_refs;
2585 ret = cleanup_extent_op(trans, fs_info, head);
2587 unselect_delayed_ref_head(delayed_refs, head);
2588 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2595 * Need to drop our head ref lock and re-acquire the delayed ref lock
2596 * and then re-check to make sure nobody got added.
2598 spin_unlock(&head->lock);
2599 spin_lock(&delayed_refs->lock);
2600 spin_lock(&head->lock);
2601 if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) {
2602 spin_unlock(&head->lock);
2603 spin_unlock(&delayed_refs->lock);
2606 delayed_refs->num_heads--;
2607 rb_erase(&head->href_node, &delayed_refs->href_root);
2608 RB_CLEAR_NODE(&head->href_node);
2609 spin_unlock(&delayed_refs->lock);
2610 spin_unlock(&head->lock);
2611 atomic_dec(&delayed_refs->num_entries);
2613 trace_run_delayed_ref_head(fs_info, head, 0);
2615 if (head->total_ref_mod < 0) {
2616 struct btrfs_block_group_cache *cache;
2618 cache = btrfs_lookup_block_group(fs_info, head->bytenr);
2620 percpu_counter_add(&cache->space_info->total_bytes_pinned,
2622 btrfs_put_block_group(cache);
2624 if (head->is_data) {
2625 spin_lock(&delayed_refs->lock);
2626 delayed_refs->pending_csums -= head->num_bytes;
2627 spin_unlock(&delayed_refs->lock);
2631 if (head->must_insert_reserved) {
2632 btrfs_pin_extent(fs_info, head->bytenr,
2633 head->num_bytes, 1);
2634 if (head->is_data) {
2635 ret = btrfs_del_csums(trans, fs_info, head->bytenr,
2640 /* Also free its reserved qgroup space */
2641 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2642 head->qgroup_reserved);
2643 btrfs_delayed_ref_unlock(head);
2644 btrfs_put_delayed_ref_head(head);
2649 * Returns 0 on success or if called with an already aborted transaction.
2650 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2652 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2655 struct btrfs_fs_info *fs_info = trans->fs_info;
2656 struct btrfs_delayed_ref_root *delayed_refs;
2657 struct btrfs_delayed_ref_node *ref;
2658 struct btrfs_delayed_ref_head *locked_ref = NULL;
2659 struct btrfs_delayed_extent_op *extent_op;
2660 ktime_t start = ktime_get();
2662 unsigned long count = 0;
2663 unsigned long actual_count = 0;
2664 int must_insert_reserved = 0;
2666 delayed_refs = &trans->transaction->delayed_refs;
2672 spin_lock(&delayed_refs->lock);
2673 locked_ref = btrfs_select_ref_head(trans);
2675 spin_unlock(&delayed_refs->lock);
2679 /* grab the lock that says we are going to process
2680 * all the refs for this head */
2681 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2682 spin_unlock(&delayed_refs->lock);
2684 * we may have dropped the spin lock to get the head
2685 * mutex lock, and that might have given someone else
2686 * time to free the head. If that's true, it has been
2687 * removed from our list and we can move on.
2689 if (ret == -EAGAIN) {
2697 * We need to try and merge add/drops of the same ref since we
2698 * can run into issues with relocate dropping the implicit ref
2699 * and then it being added back again before the drop can
2700 * finish. If we merged anything we need to re-loop so we can
2702 * Or we can get node references of the same type that weren't
2703 * merged when created due to bumps in the tree mod seq, and
2704 * we need to merge them to prevent adding an inline extent
2705 * backref before dropping it (triggering a BUG_ON at
2706 * insert_inline_extent_backref()).
2708 spin_lock(&locked_ref->lock);
2709 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2713 * locked_ref is the head node, so we have to go one
2714 * node back for any delayed ref updates
2716 ref = select_delayed_ref(locked_ref);
2718 if (ref && ref->seq &&
2719 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2720 spin_unlock(&locked_ref->lock);
2721 unselect_delayed_ref_head(delayed_refs, locked_ref);
2729 * We're done processing refs in this ref_head, clean everything
2730 * up and move on to the next ref_head.
2733 ret = cleanup_ref_head(trans, fs_info, locked_ref);
2735 /* We dropped our lock, we need to loop. */
2748 rb_erase(&ref->ref_node, &locked_ref->ref_tree);
2749 RB_CLEAR_NODE(&ref->ref_node);
2750 if (!list_empty(&ref->add_list))
2751 list_del(&ref->add_list);
2753 * When we play the delayed ref, also correct the ref_mod on
2756 switch (ref->action) {
2757 case BTRFS_ADD_DELAYED_REF:
2758 case BTRFS_ADD_DELAYED_EXTENT:
2759 locked_ref->ref_mod -= ref->ref_mod;
2761 case BTRFS_DROP_DELAYED_REF:
2762 locked_ref->ref_mod += ref->ref_mod;
2767 atomic_dec(&delayed_refs->num_entries);
2770 * Record the must-insert_reserved flag before we drop the spin
2773 must_insert_reserved = locked_ref->must_insert_reserved;
2774 locked_ref->must_insert_reserved = 0;
2776 extent_op = locked_ref->extent_op;
2777 locked_ref->extent_op = NULL;
2778 spin_unlock(&locked_ref->lock);
2780 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2781 must_insert_reserved);
2783 btrfs_free_delayed_extent_op(extent_op);
2785 unselect_delayed_ref_head(delayed_refs, locked_ref);
2786 btrfs_put_delayed_ref(ref);
2787 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2792 btrfs_put_delayed_ref(ref);
2798 * We don't want to include ref heads since we can have empty ref heads
2799 * and those will drastically skew our runtime down since we just do
2800 * accounting, no actual extent tree updates.
2802 if (actual_count > 0) {
2803 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2807 * We weigh the current average higher than our current runtime
2808 * to avoid large swings in the average.
2810 spin_lock(&delayed_refs->lock);
2811 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2812 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2813 spin_unlock(&delayed_refs->lock);
2818 #ifdef SCRAMBLE_DELAYED_REFS
2820 * Normally delayed refs get processed in ascending bytenr order. This
2821 * correlates in most cases to the order added. To expose dependencies on this
2822 * order, we start to process the tree in the middle instead of the beginning
2824 static u64 find_middle(struct rb_root *root)
2826 struct rb_node *n = root->rb_node;
2827 struct btrfs_delayed_ref_node *entry;
2830 u64 first = 0, last = 0;
2834 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2835 first = entry->bytenr;
2839 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2840 last = entry->bytenr;
2845 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2846 WARN_ON(!entry->in_tree);
2848 middle = entry->bytenr;
2861 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2865 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2866 sizeof(struct btrfs_extent_inline_ref));
2867 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2868 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2871 * We don't ever fill up leaves all the way so multiply by 2 just to be
2872 * closer to what we're really going to want to use.
2874 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2878 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2879 * would require to store the csums for that many bytes.
2881 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2884 u64 num_csums_per_leaf;
2887 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2888 num_csums_per_leaf = div64_u64(csum_size,
2889 (u64)btrfs_super_csum_size(fs_info->super_copy));
2890 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2891 num_csums += num_csums_per_leaf - 1;
2892 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2896 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2897 struct btrfs_fs_info *fs_info)
2899 struct btrfs_block_rsv *global_rsv;
2900 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2901 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2902 unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs;
2903 u64 num_bytes, num_dirty_bgs_bytes;
2906 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2907 num_heads = heads_to_leaves(fs_info, num_heads);
2909 num_bytes += (num_heads - 1) * fs_info->nodesize;
2911 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2913 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2915 global_rsv = &fs_info->global_block_rsv;
2918 * If we can't allocate any more chunks lets make sure we have _lots_ of
2919 * wiggle room since running delayed refs can create more delayed refs.
2921 if (global_rsv->space_info->full) {
2922 num_dirty_bgs_bytes <<= 1;
2926 spin_lock(&global_rsv->lock);
2927 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2929 spin_unlock(&global_rsv->lock);
2933 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2934 struct btrfs_fs_info *fs_info)
2937 atomic_read(&trans->transaction->delayed_refs.num_entries);
2942 avg_runtime = fs_info->avg_delayed_ref_runtime;
2943 val = num_entries * avg_runtime;
2944 if (val >= NSEC_PER_SEC)
2946 if (val >= NSEC_PER_SEC / 2)
2949 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2952 struct async_delayed_refs {
2953 struct btrfs_root *root;
2958 struct completion wait;
2959 struct btrfs_work work;
2962 static inline struct async_delayed_refs *
2963 to_async_delayed_refs(struct btrfs_work *work)
2965 return container_of(work, struct async_delayed_refs, work);
2968 static void delayed_ref_async_start(struct btrfs_work *work)
2970 struct async_delayed_refs *async = to_async_delayed_refs(work);
2971 struct btrfs_trans_handle *trans;
2972 struct btrfs_fs_info *fs_info = async->root->fs_info;
2975 /* if the commit is already started, we don't need to wait here */
2976 if (btrfs_transaction_blocked(fs_info))
2979 trans = btrfs_join_transaction(async->root);
2980 if (IS_ERR(trans)) {
2981 async->error = PTR_ERR(trans);
2986 * trans->sync means that when we call end_transaction, we won't
2987 * wait on delayed refs
2991 /* Don't bother flushing if we got into a different transaction */
2992 if (trans->transid > async->transid)
2995 ret = btrfs_run_delayed_refs(trans, async->count);
2999 ret = btrfs_end_transaction(trans);
3000 if (ret && !async->error)
3004 complete(&async->wait);
3009 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
3010 unsigned long count, u64 transid, int wait)
3012 struct async_delayed_refs *async;
3015 async = kmalloc(sizeof(*async), GFP_NOFS);
3019 async->root = fs_info->tree_root;
3020 async->count = count;
3022 async->transid = transid;
3027 init_completion(&async->wait);
3029 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
3030 delayed_ref_async_start, NULL, NULL);
3032 btrfs_queue_work(fs_info->extent_workers, &async->work);
3035 wait_for_completion(&async->wait);
3044 * this starts processing the delayed reference count updates and
3045 * extent insertions we have queued up so far. count can be
3046 * 0, which means to process everything in the tree at the start
3047 * of the run (but not newly added entries), or it can be some target
3048 * number you'd like to process.
3050 * Returns 0 on success or if called with an aborted transaction
3051 * Returns <0 on error and aborts the transaction
3053 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
3054 unsigned long count)
3056 struct btrfs_fs_info *fs_info = trans->fs_info;
3057 struct rb_node *node;
3058 struct btrfs_delayed_ref_root *delayed_refs;
3059 struct btrfs_delayed_ref_head *head;
3061 int run_all = count == (unsigned long)-1;
3062 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
3064 /* We'll clean this up in btrfs_cleanup_transaction */
3068 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
3071 delayed_refs = &trans->transaction->delayed_refs;
3073 count = atomic_read(&delayed_refs->num_entries) * 2;
3076 #ifdef SCRAMBLE_DELAYED_REFS
3077 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
3079 trans->can_flush_pending_bgs = false;
3080 ret = __btrfs_run_delayed_refs(trans, count);
3082 btrfs_abort_transaction(trans, ret);
3087 if (!list_empty(&trans->new_bgs))
3088 btrfs_create_pending_block_groups(trans);
3090 spin_lock(&delayed_refs->lock);
3091 node = rb_first(&delayed_refs->href_root);
3093 spin_unlock(&delayed_refs->lock);
3096 head = rb_entry(node, struct btrfs_delayed_ref_head,
3098 refcount_inc(&head->refs);
3099 spin_unlock(&delayed_refs->lock);
3101 /* Mutex was contended, block until it's released and retry. */
3102 mutex_lock(&head->mutex);
3103 mutex_unlock(&head->mutex);
3105 btrfs_put_delayed_ref_head(head);
3110 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3114 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3115 struct btrfs_fs_info *fs_info,
3116 u64 bytenr, u64 num_bytes, u64 flags,
3117 int level, int is_data)
3119 struct btrfs_delayed_extent_op *extent_op;
3122 extent_op = btrfs_alloc_delayed_extent_op();
3126 extent_op->flags_to_set = flags;
3127 extent_op->update_flags = true;
3128 extent_op->update_key = false;
3129 extent_op->is_data = is_data ? true : false;
3130 extent_op->level = level;
3132 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3133 num_bytes, extent_op);
3135 btrfs_free_delayed_extent_op(extent_op);
3139 static noinline int check_delayed_ref(struct btrfs_root *root,
3140 struct btrfs_path *path,
3141 u64 objectid, u64 offset, u64 bytenr)
3143 struct btrfs_delayed_ref_head *head;
3144 struct btrfs_delayed_ref_node *ref;
3145 struct btrfs_delayed_data_ref *data_ref;
3146 struct btrfs_delayed_ref_root *delayed_refs;
3147 struct btrfs_transaction *cur_trans;
3148 struct rb_node *node;
3151 cur_trans = root->fs_info->running_transaction;
3155 delayed_refs = &cur_trans->delayed_refs;
3156 spin_lock(&delayed_refs->lock);
3157 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3159 spin_unlock(&delayed_refs->lock);
3163 if (!mutex_trylock(&head->mutex)) {
3164 refcount_inc(&head->refs);
3165 spin_unlock(&delayed_refs->lock);
3167 btrfs_release_path(path);
3170 * Mutex was contended, block until it's released and let
3173 mutex_lock(&head->mutex);
3174 mutex_unlock(&head->mutex);
3175 btrfs_put_delayed_ref_head(head);
3178 spin_unlock(&delayed_refs->lock);
3180 spin_lock(&head->lock);
3182 * XXX: We should replace this with a proper search function in the
3185 for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) {
3186 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node);
3187 /* If it's a shared ref we know a cross reference exists */
3188 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3193 data_ref = btrfs_delayed_node_to_data_ref(ref);
3196 * If our ref doesn't match the one we're currently looking at
3197 * then we have a cross reference.
3199 if (data_ref->root != root->root_key.objectid ||
3200 data_ref->objectid != objectid ||
3201 data_ref->offset != offset) {
3206 spin_unlock(&head->lock);
3207 mutex_unlock(&head->mutex);
3211 static noinline int check_committed_ref(struct btrfs_root *root,
3212 struct btrfs_path *path,
3213 u64 objectid, u64 offset, u64 bytenr)
3215 struct btrfs_fs_info *fs_info = root->fs_info;
3216 struct btrfs_root *extent_root = fs_info->extent_root;
3217 struct extent_buffer *leaf;
3218 struct btrfs_extent_data_ref *ref;
3219 struct btrfs_extent_inline_ref *iref;
3220 struct btrfs_extent_item *ei;
3221 struct btrfs_key key;
3226 key.objectid = bytenr;
3227 key.offset = (u64)-1;
3228 key.type = BTRFS_EXTENT_ITEM_KEY;
3230 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3233 BUG_ON(ret == 0); /* Corruption */
3236 if (path->slots[0] == 0)
3240 leaf = path->nodes[0];
3241 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3243 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3247 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3248 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3249 if (item_size < sizeof(*ei)) {
3250 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3254 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3256 if (item_size != sizeof(*ei) +
3257 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3260 if (btrfs_extent_generation(leaf, ei) <=
3261 btrfs_root_last_snapshot(&root->root_item))
3264 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3266 type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA);
3267 if (type != BTRFS_EXTENT_DATA_REF_KEY)
3270 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3271 if (btrfs_extent_refs(leaf, ei) !=
3272 btrfs_extent_data_ref_count(leaf, ref) ||
3273 btrfs_extent_data_ref_root(leaf, ref) !=
3274 root->root_key.objectid ||
3275 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3276 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3284 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3287 struct btrfs_path *path;
3291 path = btrfs_alloc_path();
3296 ret = check_committed_ref(root, path, objectid,
3298 if (ret && ret != -ENOENT)
3301 ret2 = check_delayed_ref(root, path, objectid,
3303 } while (ret2 == -EAGAIN);
3305 if (ret2 && ret2 != -ENOENT) {
3310 if (ret != -ENOENT || ret2 != -ENOENT)
3313 btrfs_free_path(path);
3314 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3319 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3320 struct btrfs_root *root,
3321 struct extent_buffer *buf,
3322 int full_backref, int inc)
3324 struct btrfs_fs_info *fs_info = root->fs_info;
3330 struct btrfs_key key;
3331 struct btrfs_file_extent_item *fi;
3335 int (*process_func)(struct btrfs_trans_handle *,
3336 struct btrfs_root *,
3337 u64, u64, u64, u64, u64, u64);
3340 if (btrfs_is_testing(fs_info))
3343 ref_root = btrfs_header_owner(buf);
3344 nritems = btrfs_header_nritems(buf);
3345 level = btrfs_header_level(buf);
3347 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3351 process_func = btrfs_inc_extent_ref;
3353 process_func = btrfs_free_extent;
3356 parent = buf->start;
3360 for (i = 0; i < nritems; i++) {
3362 btrfs_item_key_to_cpu(buf, &key, i);
3363 if (key.type != BTRFS_EXTENT_DATA_KEY)
3365 fi = btrfs_item_ptr(buf, i,
3366 struct btrfs_file_extent_item);
3367 if (btrfs_file_extent_type(buf, fi) ==
3368 BTRFS_FILE_EXTENT_INLINE)
3370 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3374 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3375 key.offset -= btrfs_file_extent_offset(buf, fi);
3376 ret = process_func(trans, root, bytenr, num_bytes,
3377 parent, ref_root, key.objectid,
3382 bytenr = btrfs_node_blockptr(buf, i);
3383 num_bytes = fs_info->nodesize;
3384 ret = process_func(trans, root, bytenr, num_bytes,
3385 parent, ref_root, level - 1, 0);
3395 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3396 struct extent_buffer *buf, int full_backref)
3398 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3401 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3402 struct extent_buffer *buf, int full_backref)
3404 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3407 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3408 struct btrfs_fs_info *fs_info,
3409 struct btrfs_path *path,
3410 struct btrfs_block_group_cache *cache)
3413 struct btrfs_root *extent_root = fs_info->extent_root;
3415 struct extent_buffer *leaf;
3417 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3424 leaf = path->nodes[0];
3425 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3426 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3427 btrfs_mark_buffer_dirty(leaf);
3429 btrfs_release_path(path);
3434 static struct btrfs_block_group_cache *
3435 next_block_group(struct btrfs_fs_info *fs_info,
3436 struct btrfs_block_group_cache *cache)
3438 struct rb_node *node;
3440 spin_lock(&fs_info->block_group_cache_lock);
3442 /* If our block group was removed, we need a full search. */
3443 if (RB_EMPTY_NODE(&cache->cache_node)) {
3444 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3446 spin_unlock(&fs_info->block_group_cache_lock);
3447 btrfs_put_block_group(cache);
3448 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3450 node = rb_next(&cache->cache_node);
3451 btrfs_put_block_group(cache);
3453 cache = rb_entry(node, struct btrfs_block_group_cache,
3455 btrfs_get_block_group(cache);
3458 spin_unlock(&fs_info->block_group_cache_lock);
3462 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3463 struct btrfs_trans_handle *trans,
3464 struct btrfs_path *path)
3466 struct btrfs_fs_info *fs_info = block_group->fs_info;
3467 struct btrfs_root *root = fs_info->tree_root;
3468 struct inode *inode = NULL;
3469 struct extent_changeset *data_reserved = NULL;
3471 int dcs = BTRFS_DC_ERROR;
3477 * If this block group is smaller than 100 megs don't bother caching the
3480 if (block_group->key.offset < (100 * SZ_1M)) {
3481 spin_lock(&block_group->lock);
3482 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3483 spin_unlock(&block_group->lock);
3490 inode = lookup_free_space_inode(fs_info, block_group, path);
3491 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3492 ret = PTR_ERR(inode);
3493 btrfs_release_path(path);
3497 if (IS_ERR(inode)) {
3501 if (block_group->ro)
3504 ret = create_free_space_inode(fs_info, trans, block_group,
3512 * We want to set the generation to 0, that way if anything goes wrong
3513 * from here on out we know not to trust this cache when we load up next
3516 BTRFS_I(inode)->generation = 0;
3517 ret = btrfs_update_inode(trans, root, inode);
3520 * So theoretically we could recover from this, simply set the
3521 * super cache generation to 0 so we know to invalidate the
3522 * cache, but then we'd have to keep track of the block groups
3523 * that fail this way so we know we _have_ to reset this cache
3524 * before the next commit or risk reading stale cache. So to
3525 * limit our exposure to horrible edge cases lets just abort the
3526 * transaction, this only happens in really bad situations
3529 btrfs_abort_transaction(trans, ret);
3534 /* We've already setup this transaction, go ahead and exit */
3535 if (block_group->cache_generation == trans->transid &&
3536 i_size_read(inode)) {
3537 dcs = BTRFS_DC_SETUP;
3541 if (i_size_read(inode) > 0) {
3542 ret = btrfs_check_trunc_cache_free_space(fs_info,
3543 &fs_info->global_block_rsv);
3547 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3552 spin_lock(&block_group->lock);
3553 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3554 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3556 * don't bother trying to write stuff out _if_
3557 * a) we're not cached,
3558 * b) we're with nospace_cache mount option,
3559 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3561 dcs = BTRFS_DC_WRITTEN;
3562 spin_unlock(&block_group->lock);
3565 spin_unlock(&block_group->lock);
3568 * We hit an ENOSPC when setting up the cache in this transaction, just
3569 * skip doing the setup, we've already cleared the cache so we're safe.
3571 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3577 * Try to preallocate enough space based on how big the block group is.
3578 * Keep in mind this has to include any pinned space which could end up
3579 * taking up quite a bit since it's not folded into the other space
3582 num_pages = div_u64(block_group->key.offset, SZ_256M);
3587 num_pages *= PAGE_SIZE;
3589 ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages);
3593 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3594 num_pages, num_pages,
3597 * Our cache requires contiguous chunks so that we don't modify a bunch
3598 * of metadata or split extents when writing the cache out, which means
3599 * we can enospc if we are heavily fragmented in addition to just normal
3600 * out of space conditions. So if we hit this just skip setting up any
3601 * other block groups for this transaction, maybe we'll unpin enough
3602 * space the next time around.
3605 dcs = BTRFS_DC_SETUP;
3606 else if (ret == -ENOSPC)
3607 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3612 btrfs_release_path(path);
3614 spin_lock(&block_group->lock);
3615 if (!ret && dcs == BTRFS_DC_SETUP)
3616 block_group->cache_generation = trans->transid;
3617 block_group->disk_cache_state = dcs;
3618 spin_unlock(&block_group->lock);
3620 extent_changeset_free(data_reserved);
3624 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3625 struct btrfs_fs_info *fs_info)
3627 struct btrfs_block_group_cache *cache, *tmp;
3628 struct btrfs_transaction *cur_trans = trans->transaction;
3629 struct btrfs_path *path;
3631 if (list_empty(&cur_trans->dirty_bgs) ||
3632 !btrfs_test_opt(fs_info, SPACE_CACHE))
3635 path = btrfs_alloc_path();
3639 /* Could add new block groups, use _safe just in case */
3640 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3642 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3643 cache_save_setup(cache, trans, path);
3646 btrfs_free_path(path);
3651 * transaction commit does final block group cache writeback during a
3652 * critical section where nothing is allowed to change the FS. This is
3653 * required in order for the cache to actually match the block group,
3654 * but can introduce a lot of latency into the commit.
3656 * So, btrfs_start_dirty_block_groups is here to kick off block group
3657 * cache IO. There's a chance we'll have to redo some of it if the
3658 * block group changes again during the commit, but it greatly reduces
3659 * the commit latency by getting rid of the easy block groups while
3660 * we're still allowing others to join the commit.
3662 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans)
3664 struct btrfs_fs_info *fs_info = trans->fs_info;
3665 struct btrfs_block_group_cache *cache;
3666 struct btrfs_transaction *cur_trans = trans->transaction;
3669 struct btrfs_path *path = NULL;
3671 struct list_head *io = &cur_trans->io_bgs;
3672 int num_started = 0;
3675 spin_lock(&cur_trans->dirty_bgs_lock);
3676 if (list_empty(&cur_trans->dirty_bgs)) {
3677 spin_unlock(&cur_trans->dirty_bgs_lock);
3680 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3681 spin_unlock(&cur_trans->dirty_bgs_lock);
3685 * make sure all the block groups on our dirty list actually
3688 btrfs_create_pending_block_groups(trans);
3691 path = btrfs_alloc_path();
3697 * cache_write_mutex is here only to save us from balance or automatic
3698 * removal of empty block groups deleting this block group while we are
3699 * writing out the cache
3701 mutex_lock(&trans->transaction->cache_write_mutex);
3702 while (!list_empty(&dirty)) {
3703 cache = list_first_entry(&dirty,
3704 struct btrfs_block_group_cache,
3707 * this can happen if something re-dirties a block
3708 * group that is already under IO. Just wait for it to
3709 * finish and then do it all again
3711 if (!list_empty(&cache->io_list)) {
3712 list_del_init(&cache->io_list);
3713 btrfs_wait_cache_io(trans, cache, path);
3714 btrfs_put_block_group(cache);
3719 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3720 * if it should update the cache_state. Don't delete
3721 * until after we wait.
3723 * Since we're not running in the commit critical section
3724 * we need the dirty_bgs_lock to protect from update_block_group
3726 spin_lock(&cur_trans->dirty_bgs_lock);
3727 list_del_init(&cache->dirty_list);
3728 spin_unlock(&cur_trans->dirty_bgs_lock);
3732 cache_save_setup(cache, trans, path);
3734 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3735 cache->io_ctl.inode = NULL;
3736 ret = btrfs_write_out_cache(fs_info, trans,
3738 if (ret == 0 && cache->io_ctl.inode) {
3743 * The cache_write_mutex is protecting the
3744 * io_list, also refer to the definition of
3745 * btrfs_transaction::io_bgs for more details
3747 list_add_tail(&cache->io_list, io);
3750 * if we failed to write the cache, the
3751 * generation will be bad and life goes on
3757 ret = write_one_cache_group(trans, fs_info,
3760 * Our block group might still be attached to the list
3761 * of new block groups in the transaction handle of some
3762 * other task (struct btrfs_trans_handle->new_bgs). This
3763 * means its block group item isn't yet in the extent
3764 * tree. If this happens ignore the error, as we will
3765 * try again later in the critical section of the
3766 * transaction commit.
3768 if (ret == -ENOENT) {
3770 spin_lock(&cur_trans->dirty_bgs_lock);
3771 if (list_empty(&cache->dirty_list)) {
3772 list_add_tail(&cache->dirty_list,
3773 &cur_trans->dirty_bgs);
3774 btrfs_get_block_group(cache);
3776 spin_unlock(&cur_trans->dirty_bgs_lock);
3778 btrfs_abort_transaction(trans, ret);
3782 /* if its not on the io list, we need to put the block group */
3784 btrfs_put_block_group(cache);
3790 * Avoid blocking other tasks for too long. It might even save
3791 * us from writing caches for block groups that are going to be
3794 mutex_unlock(&trans->transaction->cache_write_mutex);
3795 mutex_lock(&trans->transaction->cache_write_mutex);
3797 mutex_unlock(&trans->transaction->cache_write_mutex);
3800 * go through delayed refs for all the stuff we've just kicked off
3801 * and then loop back (just once)
3803 ret = btrfs_run_delayed_refs(trans, 0);
3804 if (!ret && loops == 0) {
3806 spin_lock(&cur_trans->dirty_bgs_lock);
3807 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3809 * dirty_bgs_lock protects us from concurrent block group
3810 * deletes too (not just cache_write_mutex).
3812 if (!list_empty(&dirty)) {
3813 spin_unlock(&cur_trans->dirty_bgs_lock);
3816 spin_unlock(&cur_trans->dirty_bgs_lock);
3817 } else if (ret < 0) {
3818 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3821 btrfs_free_path(path);
3825 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3826 struct btrfs_fs_info *fs_info)
3828 struct btrfs_block_group_cache *cache;
3829 struct btrfs_transaction *cur_trans = trans->transaction;
3832 struct btrfs_path *path;
3833 struct list_head *io = &cur_trans->io_bgs;
3834 int num_started = 0;
3836 path = btrfs_alloc_path();
3841 * Even though we are in the critical section of the transaction commit,
3842 * we can still have concurrent tasks adding elements to this
3843 * transaction's list of dirty block groups. These tasks correspond to
3844 * endio free space workers started when writeback finishes for a
3845 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3846 * allocate new block groups as a result of COWing nodes of the root
3847 * tree when updating the free space inode. The writeback for the space
3848 * caches is triggered by an earlier call to
3849 * btrfs_start_dirty_block_groups() and iterations of the following
3851 * Also we want to do the cache_save_setup first and then run the
3852 * delayed refs to make sure we have the best chance at doing this all
3855 spin_lock(&cur_trans->dirty_bgs_lock);
3856 while (!list_empty(&cur_trans->dirty_bgs)) {
3857 cache = list_first_entry(&cur_trans->dirty_bgs,
3858 struct btrfs_block_group_cache,
3862 * this can happen if cache_save_setup re-dirties a block
3863 * group that is already under IO. Just wait for it to
3864 * finish and then do it all again
3866 if (!list_empty(&cache->io_list)) {
3867 spin_unlock(&cur_trans->dirty_bgs_lock);
3868 list_del_init(&cache->io_list);
3869 btrfs_wait_cache_io(trans, cache, path);
3870 btrfs_put_block_group(cache);
3871 spin_lock(&cur_trans->dirty_bgs_lock);
3875 * don't remove from the dirty list until after we've waited
3878 list_del_init(&cache->dirty_list);
3879 spin_unlock(&cur_trans->dirty_bgs_lock);
3882 cache_save_setup(cache, trans, path);
3885 ret = btrfs_run_delayed_refs(trans,
3886 (unsigned long) -1);
3888 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3889 cache->io_ctl.inode = NULL;
3890 ret = btrfs_write_out_cache(fs_info, trans,
3892 if (ret == 0 && cache->io_ctl.inode) {
3895 list_add_tail(&cache->io_list, io);
3898 * if we failed to write the cache, the
3899 * generation will be bad and life goes on
3905 ret = write_one_cache_group(trans, fs_info,
3908 * One of the free space endio workers might have
3909 * created a new block group while updating a free space
3910 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3911 * and hasn't released its transaction handle yet, in
3912 * which case the new block group is still attached to
3913 * its transaction handle and its creation has not
3914 * finished yet (no block group item in the extent tree
3915 * yet, etc). If this is the case, wait for all free
3916 * space endio workers to finish and retry. This is a
3917 * a very rare case so no need for a more efficient and
3920 if (ret == -ENOENT) {
3921 wait_event(cur_trans->writer_wait,
3922 atomic_read(&cur_trans->num_writers) == 1);
3923 ret = write_one_cache_group(trans, fs_info,
3927 btrfs_abort_transaction(trans, ret);
3930 /* if its not on the io list, we need to put the block group */
3932 btrfs_put_block_group(cache);
3933 spin_lock(&cur_trans->dirty_bgs_lock);
3935 spin_unlock(&cur_trans->dirty_bgs_lock);
3938 * Refer to the definition of io_bgs member for details why it's safe
3939 * to use it without any locking
3941 while (!list_empty(io)) {
3942 cache = list_first_entry(io, struct btrfs_block_group_cache,
3944 list_del_init(&cache->io_list);
3945 btrfs_wait_cache_io(trans, cache, path);
3946 btrfs_put_block_group(cache);
3949 btrfs_free_path(path);
3953 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3955 struct btrfs_block_group_cache *block_group;
3958 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3959 if (!block_group || block_group->ro)
3962 btrfs_put_block_group(block_group);
3966 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3968 struct btrfs_block_group_cache *bg;
3971 bg = btrfs_lookup_block_group(fs_info, bytenr);
3975 spin_lock(&bg->lock);
3979 atomic_inc(&bg->nocow_writers);
3980 spin_unlock(&bg->lock);
3982 /* no put on block group, done by btrfs_dec_nocow_writers */
3984 btrfs_put_block_group(bg);
3990 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3992 struct btrfs_block_group_cache *bg;
3994 bg = btrfs_lookup_block_group(fs_info, bytenr);
3996 if (atomic_dec_and_test(&bg->nocow_writers))
3997 wake_up_var(&bg->nocow_writers);
3999 * Once for our lookup and once for the lookup done by a previous call
4000 * to btrfs_inc_nocow_writers()
4002 btrfs_put_block_group(bg);
4003 btrfs_put_block_group(bg);
4006 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
4008 wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers));
4011 static const char *alloc_name(u64 flags)
4014 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
4016 case BTRFS_BLOCK_GROUP_METADATA:
4018 case BTRFS_BLOCK_GROUP_DATA:
4020 case BTRFS_BLOCK_GROUP_SYSTEM:
4024 return "invalid-combination";
4028 static int create_space_info(struct btrfs_fs_info *info, u64 flags,
4029 struct btrfs_space_info **new)
4032 struct btrfs_space_info *space_info;
4036 space_info = kzalloc(sizeof(*space_info), GFP_NOFS);
4040 ret = percpu_counter_init(&space_info->total_bytes_pinned, 0,
4047 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
4048 INIT_LIST_HEAD(&space_info->block_groups[i]);
4049 init_rwsem(&space_info->groups_sem);
4050 spin_lock_init(&space_info->lock);
4051 space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
4052 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4053 init_waitqueue_head(&space_info->wait);
4054 INIT_LIST_HEAD(&space_info->ro_bgs);
4055 INIT_LIST_HEAD(&space_info->tickets);
4056 INIT_LIST_HEAD(&space_info->priority_tickets);
4058 ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype,
4059 info->space_info_kobj, "%s",
4060 alloc_name(space_info->flags));
4062 percpu_counter_destroy(&space_info->total_bytes_pinned);
4068 list_add_rcu(&space_info->list, &info->space_info);
4069 if (flags & BTRFS_BLOCK_GROUP_DATA)
4070 info->data_sinfo = space_info;
4075 static void update_space_info(struct btrfs_fs_info *info, u64 flags,
4076 u64 total_bytes, u64 bytes_used,
4078 struct btrfs_space_info **space_info)
4080 struct btrfs_space_info *found;
4083 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
4084 BTRFS_BLOCK_GROUP_RAID10))
4089 found = __find_space_info(info, flags);
4091 spin_lock(&found->lock);
4092 found->total_bytes += total_bytes;
4093 found->disk_total += total_bytes * factor;
4094 found->bytes_used += bytes_used;
4095 found->disk_used += bytes_used * factor;
4096 found->bytes_readonly += bytes_readonly;
4097 if (total_bytes > 0)
4099 space_info_add_new_bytes(info, found, total_bytes -
4100 bytes_used - bytes_readonly);
4101 spin_unlock(&found->lock);
4102 *space_info = found;
4105 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4107 u64 extra_flags = chunk_to_extended(flags) &
4108 BTRFS_EXTENDED_PROFILE_MASK;
4110 write_seqlock(&fs_info->profiles_lock);
4111 if (flags & BTRFS_BLOCK_GROUP_DATA)
4112 fs_info->avail_data_alloc_bits |= extra_flags;
4113 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4114 fs_info->avail_metadata_alloc_bits |= extra_flags;
4115 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4116 fs_info->avail_system_alloc_bits |= extra_flags;
4117 write_sequnlock(&fs_info->profiles_lock);
4121 * returns target flags in extended format or 0 if restripe for this
4122 * chunk_type is not in progress
4124 * should be called with either volume_mutex or balance_lock held
4126 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4128 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4134 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4135 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4136 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4137 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4138 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4139 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4140 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4141 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4142 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4149 * @flags: available profiles in extended format (see ctree.h)
4151 * Returns reduced profile in chunk format. If profile changing is in
4152 * progress (either running or paused) picks the target profile (if it's
4153 * already available), otherwise falls back to plain reducing.
4155 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4157 u64 num_devices = fs_info->fs_devices->rw_devices;
4163 * see if restripe for this chunk_type is in progress, if so
4164 * try to reduce to the target profile
4166 spin_lock(&fs_info->balance_lock);
4167 target = get_restripe_target(fs_info, flags);
4169 /* pick target profile only if it's already available */
4170 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4171 spin_unlock(&fs_info->balance_lock);
4172 return extended_to_chunk(target);
4175 spin_unlock(&fs_info->balance_lock);
4177 /* First, mask out the RAID levels which aren't possible */
4178 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4179 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4180 allowed |= btrfs_raid_group[raid_type];
4184 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4185 allowed = BTRFS_BLOCK_GROUP_RAID6;
4186 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4187 allowed = BTRFS_BLOCK_GROUP_RAID5;
4188 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4189 allowed = BTRFS_BLOCK_GROUP_RAID10;
4190 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4191 allowed = BTRFS_BLOCK_GROUP_RAID1;
4192 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4193 allowed = BTRFS_BLOCK_GROUP_RAID0;
4195 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4197 return extended_to_chunk(flags | allowed);
4200 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4207 seq = read_seqbegin(&fs_info->profiles_lock);
4209 if (flags & BTRFS_BLOCK_GROUP_DATA)
4210 flags |= fs_info->avail_data_alloc_bits;
4211 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4212 flags |= fs_info->avail_system_alloc_bits;
4213 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4214 flags |= fs_info->avail_metadata_alloc_bits;
4215 } while (read_seqretry(&fs_info->profiles_lock, seq));
4217 return btrfs_reduce_alloc_profile(fs_info, flags);
4220 static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data)
4222 struct btrfs_fs_info *fs_info = root->fs_info;
4227 flags = BTRFS_BLOCK_GROUP_DATA;
4228 else if (root == fs_info->chunk_root)
4229 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4231 flags = BTRFS_BLOCK_GROUP_METADATA;
4233 ret = get_alloc_profile(fs_info, flags);
4237 u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info)
4239 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA);
4242 u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info)
4244 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4247 u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info)
4249 return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4252 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4253 bool may_use_included)
4256 return s_info->bytes_used + s_info->bytes_reserved +
4257 s_info->bytes_pinned + s_info->bytes_readonly +
4258 (may_use_included ? s_info->bytes_may_use : 0);
4261 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4263 struct btrfs_root *root = inode->root;
4264 struct btrfs_fs_info *fs_info = root->fs_info;
4265 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo;
4268 int need_commit = 2;
4269 int have_pinned_space;
4271 /* make sure bytes are sectorsize aligned */
4272 bytes = ALIGN(bytes, fs_info->sectorsize);
4274 if (btrfs_is_free_space_inode(inode)) {
4276 ASSERT(current->journal_info);
4280 /* make sure we have enough space to handle the data first */
4281 spin_lock(&data_sinfo->lock);
4282 used = btrfs_space_info_used(data_sinfo, true);
4284 if (used + bytes > data_sinfo->total_bytes) {
4285 struct btrfs_trans_handle *trans;
4288 * if we don't have enough free bytes in this space then we need
4289 * to alloc a new chunk.
4291 if (!data_sinfo->full) {
4294 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4295 spin_unlock(&data_sinfo->lock);
4297 alloc_target = btrfs_data_alloc_profile(fs_info);
4299 * It is ugly that we don't call nolock join
4300 * transaction for the free space inode case here.
4301 * But it is safe because we only do the data space
4302 * reservation for the free space cache in the
4303 * transaction context, the common join transaction
4304 * just increase the counter of the current transaction
4305 * handler, doesn't try to acquire the trans_lock of
4308 trans = btrfs_join_transaction(root);
4310 return PTR_ERR(trans);
4312 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4313 CHUNK_ALLOC_NO_FORCE);
4314 btrfs_end_transaction(trans);
4319 have_pinned_space = 1;
4328 * If we don't have enough pinned space to deal with this
4329 * allocation, and no removed chunk in current transaction,
4330 * don't bother committing the transaction.
4332 have_pinned_space = percpu_counter_compare(
4333 &data_sinfo->total_bytes_pinned,
4334 used + bytes - data_sinfo->total_bytes);
4335 spin_unlock(&data_sinfo->lock);
4337 /* commit the current transaction and try again */
4342 if (need_commit > 0) {
4343 btrfs_start_delalloc_roots(fs_info, 0, -1);
4344 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0,
4348 trans = btrfs_join_transaction(root);
4350 return PTR_ERR(trans);
4351 if (have_pinned_space >= 0 ||
4352 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4353 &trans->transaction->flags) ||
4355 ret = btrfs_commit_transaction(trans);
4359 * The cleaner kthread might still be doing iput
4360 * operations. Wait for it to finish so that
4361 * more space is released.
4363 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4364 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4367 btrfs_end_transaction(trans);
4371 trace_btrfs_space_reservation(fs_info,
4372 "space_info:enospc",
4373 data_sinfo->flags, bytes, 1);
4376 data_sinfo->bytes_may_use += bytes;
4377 trace_btrfs_space_reservation(fs_info, "space_info",
4378 data_sinfo->flags, bytes, 1);
4379 spin_unlock(&data_sinfo->lock);
4384 int btrfs_check_data_free_space(struct inode *inode,
4385 struct extent_changeset **reserved, u64 start, u64 len)
4387 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4390 /* align the range */
4391 len = round_up(start + len, fs_info->sectorsize) -
4392 round_down(start, fs_info->sectorsize);
4393 start = round_down(start, fs_info->sectorsize);
4395 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4399 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4400 ret = btrfs_qgroup_reserve_data(inode, reserved, start, len);
4402 btrfs_free_reserved_data_space_noquota(inode, start, len);
4409 * Called if we need to clear a data reservation for this inode
4410 * Normally in a error case.
4412 * This one will *NOT* use accurate qgroup reserved space API, just for case
4413 * which we can't sleep and is sure it won't affect qgroup reserved space.
4414 * Like clear_bit_hook().
4416 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4419 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4420 struct btrfs_space_info *data_sinfo;
4422 /* Make sure the range is aligned to sectorsize */
4423 len = round_up(start + len, fs_info->sectorsize) -
4424 round_down(start, fs_info->sectorsize);
4425 start = round_down(start, fs_info->sectorsize);
4427 data_sinfo = fs_info->data_sinfo;
4428 spin_lock(&data_sinfo->lock);
4429 if (WARN_ON(data_sinfo->bytes_may_use < len))
4430 data_sinfo->bytes_may_use = 0;
4432 data_sinfo->bytes_may_use -= len;
4433 trace_btrfs_space_reservation(fs_info, "space_info",
4434 data_sinfo->flags, len, 0);
4435 spin_unlock(&data_sinfo->lock);
4439 * Called if we need to clear a data reservation for this inode
4440 * Normally in a error case.
4442 * This one will handle the per-inode data rsv map for accurate reserved
4445 void btrfs_free_reserved_data_space(struct inode *inode,
4446 struct extent_changeset *reserved, u64 start, u64 len)
4448 struct btrfs_root *root = BTRFS_I(inode)->root;
4450 /* Make sure the range is aligned to sectorsize */
4451 len = round_up(start + len, root->fs_info->sectorsize) -
4452 round_down(start, root->fs_info->sectorsize);
4453 start = round_down(start, root->fs_info->sectorsize);
4455 btrfs_free_reserved_data_space_noquota(inode, start, len);
4456 btrfs_qgroup_free_data(inode, reserved, start, len);
4459 static void force_metadata_allocation(struct btrfs_fs_info *info)
4461 struct list_head *head = &info->space_info;
4462 struct btrfs_space_info *found;
4465 list_for_each_entry_rcu(found, head, list) {
4466 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4467 found->force_alloc = CHUNK_ALLOC_FORCE;
4472 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4474 return (global->size << 1);
4477 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4478 struct btrfs_space_info *sinfo, int force)
4480 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4481 u64 bytes_used = btrfs_space_info_used(sinfo, false);
4484 if (force == CHUNK_ALLOC_FORCE)
4488 * We need to take into account the global rsv because for all intents
4489 * and purposes it's used space. Don't worry about locking the
4490 * global_rsv, it doesn't change except when the transaction commits.
4492 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4493 bytes_used += calc_global_rsv_need_space(global_rsv);
4496 * in limited mode, we want to have some free space up to
4497 * about 1% of the FS size.
4499 if (force == CHUNK_ALLOC_LIMITED) {
4500 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4501 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4503 if (sinfo->total_bytes - bytes_used < thresh)
4507 if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8))
4512 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4516 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4517 BTRFS_BLOCK_GROUP_RAID0 |
4518 BTRFS_BLOCK_GROUP_RAID5 |
4519 BTRFS_BLOCK_GROUP_RAID6))
4520 num_dev = fs_info->fs_devices->rw_devices;
4521 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4524 num_dev = 1; /* DUP or single */
4530 * If @is_allocation is true, reserve space in the system space info necessary
4531 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4534 void check_system_chunk(struct btrfs_trans_handle *trans,
4535 struct btrfs_fs_info *fs_info, u64 type)
4537 struct btrfs_space_info *info;
4544 * Needed because we can end up allocating a system chunk and for an
4545 * atomic and race free space reservation in the chunk block reserve.
4547 lockdep_assert_held(&fs_info->chunk_mutex);
4549 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4550 spin_lock(&info->lock);
4551 left = info->total_bytes - btrfs_space_info_used(info, true);
4552 spin_unlock(&info->lock);
4554 num_devs = get_profile_num_devs(fs_info, type);
4556 /* num_devs device items to update and 1 chunk item to add or remove */
4557 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4558 btrfs_calc_trans_metadata_size(fs_info, 1);
4560 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4561 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4562 left, thresh, type);
4563 dump_space_info(fs_info, info, 0, 0);
4566 if (left < thresh) {
4567 u64 flags = btrfs_system_alloc_profile(fs_info);
4570 * Ignore failure to create system chunk. We might end up not
4571 * needing it, as we might not need to COW all nodes/leafs from
4572 * the paths we visit in the chunk tree (they were already COWed
4573 * or created in the current transaction for example).
4575 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4579 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4580 &fs_info->chunk_block_rsv,
4581 thresh, BTRFS_RESERVE_NO_FLUSH);
4583 trans->chunk_bytes_reserved += thresh;
4588 * If force is CHUNK_ALLOC_FORCE:
4589 * - return 1 if it successfully allocates a chunk,
4590 * - return errors including -ENOSPC otherwise.
4591 * If force is NOT CHUNK_ALLOC_FORCE:
4592 * - return 0 if it doesn't need to allocate a new chunk,
4593 * - return 1 if it successfully allocates a chunk,
4594 * - return errors including -ENOSPC otherwise.
4596 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4597 struct btrfs_fs_info *fs_info, u64 flags, int force)
4599 struct btrfs_space_info *space_info;
4600 int wait_for_alloc = 0;
4603 /* Don't re-enter if we're already allocating a chunk */
4604 if (trans->allocating_chunk)
4607 space_info = __find_space_info(fs_info, flags);
4611 spin_lock(&space_info->lock);
4612 if (force < space_info->force_alloc)
4613 force = space_info->force_alloc;
4614 if (space_info->full) {
4615 if (should_alloc_chunk(fs_info, space_info, force))
4619 spin_unlock(&space_info->lock);
4623 if (!should_alloc_chunk(fs_info, space_info, force)) {
4624 spin_unlock(&space_info->lock);
4626 } else if (space_info->chunk_alloc) {
4629 space_info->chunk_alloc = 1;
4632 spin_unlock(&space_info->lock);
4634 mutex_lock(&fs_info->chunk_mutex);
4637 * The chunk_mutex is held throughout the entirety of a chunk
4638 * allocation, so once we've acquired the chunk_mutex we know that the
4639 * other guy is done and we need to recheck and see if we should
4642 if (wait_for_alloc) {
4643 mutex_unlock(&fs_info->chunk_mutex);
4648 trans->allocating_chunk = true;
4651 * If we have mixed data/metadata chunks we want to make sure we keep
4652 * allocating mixed chunks instead of individual chunks.
4654 if (btrfs_mixed_space_info(space_info))
4655 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4658 * if we're doing a data chunk, go ahead and make sure that
4659 * we keep a reasonable number of metadata chunks allocated in the
4662 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4663 fs_info->data_chunk_allocations++;
4664 if (!(fs_info->data_chunk_allocations %
4665 fs_info->metadata_ratio))
4666 force_metadata_allocation(fs_info);
4670 * Check if we have enough space in SYSTEM chunk because we may need
4671 * to update devices.
4673 check_system_chunk(trans, fs_info, flags);
4675 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4676 trans->allocating_chunk = false;
4678 spin_lock(&space_info->lock);
4679 if (ret < 0 && ret != -ENOSPC)
4682 space_info->full = 1;
4686 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4688 space_info->chunk_alloc = 0;
4689 spin_unlock(&space_info->lock);
4690 mutex_unlock(&fs_info->chunk_mutex);
4692 * When we allocate a new chunk we reserve space in the chunk block
4693 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4694 * add new nodes/leafs to it if we end up needing to do it when
4695 * inserting the chunk item and updating device items as part of the
4696 * second phase of chunk allocation, performed by
4697 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4698 * large number of new block groups to create in our transaction
4699 * handle's new_bgs list to avoid exhausting the chunk block reserve
4700 * in extreme cases - like having a single transaction create many new
4701 * block groups when starting to write out the free space caches of all
4702 * the block groups that were made dirty during the lifetime of the
4705 if (trans->can_flush_pending_bgs &&
4706 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4707 btrfs_create_pending_block_groups(trans);
4708 btrfs_trans_release_chunk_metadata(trans);
4713 static int can_overcommit(struct btrfs_fs_info *fs_info,
4714 struct btrfs_space_info *space_info, u64 bytes,
4715 enum btrfs_reserve_flush_enum flush,
4718 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4724 /* Don't overcommit when in mixed mode. */
4725 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4729 profile = btrfs_system_alloc_profile(fs_info);
4731 profile = btrfs_metadata_alloc_profile(fs_info);
4733 used = btrfs_space_info_used(space_info, false);
4736 * We only want to allow over committing if we have lots of actual space
4737 * free, but if we don't have enough space to handle the global reserve
4738 * space then we could end up having a real enospc problem when trying
4739 * to allocate a chunk or some other such important allocation.
4741 spin_lock(&global_rsv->lock);
4742 space_size = calc_global_rsv_need_space(global_rsv);
4743 spin_unlock(&global_rsv->lock);
4744 if (used + space_size >= space_info->total_bytes)
4747 used += space_info->bytes_may_use;
4749 avail = atomic64_read(&fs_info->free_chunk_space);
4752 * If we have dup, raid1 or raid10 then only half of the free
4753 * space is actually useable. For raid56, the space info used
4754 * doesn't include the parity drive, so we don't have to
4757 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4758 BTRFS_BLOCK_GROUP_RAID1 |
4759 BTRFS_BLOCK_GROUP_RAID10))
4763 * If we aren't flushing all things, let us overcommit up to
4764 * 1/2th of the space. If we can flush, don't let us overcommit
4765 * too much, let it overcommit up to 1/8 of the space.
4767 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4772 if (used + bytes < space_info->total_bytes + avail)
4777 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4778 unsigned long nr_pages, int nr_items)
4780 struct super_block *sb = fs_info->sb;
4782 if (down_read_trylock(&sb->s_umount)) {
4783 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4784 up_read(&sb->s_umount);
4787 * We needn't worry the filesystem going from r/w to r/o though
4788 * we don't acquire ->s_umount mutex, because the filesystem
4789 * should guarantee the delalloc inodes list be empty after
4790 * the filesystem is readonly(all dirty pages are written to
4793 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4794 if (!current->journal_info)
4795 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4799 static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4805 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4806 nr = div64_u64(to_reclaim, bytes);
4812 #define EXTENT_SIZE_PER_ITEM SZ_256K
4815 * shrink metadata reservation for delalloc
4817 static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim,
4818 u64 orig, bool wait_ordered)
4820 struct btrfs_space_info *space_info;
4821 struct btrfs_trans_handle *trans;
4826 unsigned long nr_pages;
4829 /* Calc the number of the pages we need flush for space reservation */
4830 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4831 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4833 trans = (struct btrfs_trans_handle *)current->journal_info;
4834 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4836 delalloc_bytes = percpu_counter_sum_positive(
4837 &fs_info->delalloc_bytes);
4838 if (delalloc_bytes == 0) {
4842 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4847 while (delalloc_bytes && loops < 3) {
4848 max_reclaim = min(delalloc_bytes, to_reclaim);
4849 nr_pages = max_reclaim >> PAGE_SHIFT;
4850 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4852 * We need to wait for the async pages to actually start before
4855 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4859 if (max_reclaim <= nr_pages)
4862 max_reclaim -= nr_pages;
4864 wait_event(fs_info->async_submit_wait,
4865 atomic_read(&fs_info->async_delalloc_pages) <=
4868 spin_lock(&space_info->lock);
4869 if (list_empty(&space_info->tickets) &&
4870 list_empty(&space_info->priority_tickets)) {
4871 spin_unlock(&space_info->lock);
4874 spin_unlock(&space_info->lock);
4877 if (wait_ordered && !trans) {
4878 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4880 time_left = schedule_timeout_killable(1);
4884 delalloc_bytes = percpu_counter_sum_positive(
4885 &fs_info->delalloc_bytes);
4889 struct reserve_ticket {
4892 struct list_head list;
4893 wait_queue_head_t wait;
4897 * maybe_commit_transaction - possibly commit the transaction if its ok to
4898 * @root - the root we're allocating for
4899 * @bytes - the number of bytes we want to reserve
4900 * @force - force the commit
4902 * This will check to make sure that committing the transaction will actually
4903 * get us somewhere and then commit the transaction if it does. Otherwise it
4904 * will return -ENOSPC.
4906 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4907 struct btrfs_space_info *space_info)
4909 struct reserve_ticket *ticket = NULL;
4910 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4911 struct btrfs_trans_handle *trans;
4914 trans = (struct btrfs_trans_handle *)current->journal_info;
4918 spin_lock(&space_info->lock);
4919 if (!list_empty(&space_info->priority_tickets))
4920 ticket = list_first_entry(&space_info->priority_tickets,
4921 struct reserve_ticket, list);
4922 else if (!list_empty(&space_info->tickets))
4923 ticket = list_first_entry(&space_info->tickets,
4924 struct reserve_ticket, list);
4925 bytes = (ticket) ? ticket->bytes : 0;
4926 spin_unlock(&space_info->lock);
4931 /* See if there is enough pinned space to make this reservation */
4932 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4937 * See if there is some space in the delayed insertion reservation for
4940 if (space_info != delayed_rsv->space_info)
4943 spin_lock(&delayed_rsv->lock);
4944 if (delayed_rsv->size > bytes)
4947 bytes -= delayed_rsv->size;
4948 spin_unlock(&delayed_rsv->lock);
4950 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4956 trans = btrfs_join_transaction(fs_info->extent_root);
4960 return btrfs_commit_transaction(trans);
4964 * Try to flush some data based on policy set by @state. This is only advisory
4965 * and may fail for various reasons. The caller is supposed to examine the
4966 * state of @space_info to detect the outcome.
4968 static void flush_space(struct btrfs_fs_info *fs_info,
4969 struct btrfs_space_info *space_info, u64 num_bytes,
4972 struct btrfs_root *root = fs_info->extent_root;
4973 struct btrfs_trans_handle *trans;
4978 case FLUSH_DELAYED_ITEMS_NR:
4979 case FLUSH_DELAYED_ITEMS:
4980 if (state == FLUSH_DELAYED_ITEMS_NR)
4981 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4985 trans = btrfs_join_transaction(root);
4986 if (IS_ERR(trans)) {
4987 ret = PTR_ERR(trans);
4990 ret = btrfs_run_delayed_items_nr(trans, nr);
4991 btrfs_end_transaction(trans);
4993 case FLUSH_DELALLOC:
4994 case FLUSH_DELALLOC_WAIT:
4995 shrink_delalloc(fs_info, num_bytes * 2, num_bytes,
4996 state == FLUSH_DELALLOC_WAIT);
4999 trans = btrfs_join_transaction(root);
5000 if (IS_ERR(trans)) {
5001 ret = PTR_ERR(trans);
5004 ret = do_chunk_alloc(trans, fs_info,
5005 btrfs_metadata_alloc_profile(fs_info),
5006 CHUNK_ALLOC_NO_FORCE);
5007 btrfs_end_transaction(trans);
5008 if (ret > 0 || ret == -ENOSPC)
5012 ret = may_commit_transaction(fs_info, space_info);
5019 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state,
5025 btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info,
5026 struct btrfs_space_info *space_info,
5029 struct reserve_ticket *ticket;
5034 list_for_each_entry(ticket, &space_info->tickets, list)
5035 to_reclaim += ticket->bytes;
5036 list_for_each_entry(ticket, &space_info->priority_tickets, list)
5037 to_reclaim += ticket->bytes;
5041 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
5042 if (can_overcommit(fs_info, space_info, to_reclaim,
5043 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5046 used = btrfs_space_info_used(space_info, true);
5048 if (can_overcommit(fs_info, space_info, SZ_1M,
5049 BTRFS_RESERVE_FLUSH_ALL, system_chunk))
5050 expected = div_factor_fine(space_info->total_bytes, 95);
5052 expected = div_factor_fine(space_info->total_bytes, 90);
5054 if (used > expected)
5055 to_reclaim = used - expected;
5058 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
5059 space_info->bytes_reserved);
5063 static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info,
5064 struct btrfs_space_info *space_info,
5065 u64 used, bool system_chunk)
5067 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
5069 /* If we're just plain full then async reclaim just slows us down. */
5070 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
5073 if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5077 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
5078 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
5081 static void wake_all_tickets(struct list_head *head)
5083 struct reserve_ticket *ticket;
5085 while (!list_empty(head)) {
5086 ticket = list_first_entry(head, struct reserve_ticket, list);
5087 list_del_init(&ticket->list);
5088 ticket->error = -ENOSPC;
5089 wake_up(&ticket->wait);
5094 * This is for normal flushers, we can wait all goddamned day if we want to. We
5095 * will loop and continuously try to flush as long as we are making progress.
5096 * We count progress as clearing off tickets each time we have to loop.
5098 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
5100 struct btrfs_fs_info *fs_info;
5101 struct btrfs_space_info *space_info;
5104 int commit_cycles = 0;
5105 u64 last_tickets_id;
5107 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5108 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5110 spin_lock(&space_info->lock);
5111 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5114 space_info->flush = 0;
5115 spin_unlock(&space_info->lock);
5118 last_tickets_id = space_info->tickets_id;
5119 spin_unlock(&space_info->lock);
5121 flush_state = FLUSH_DELAYED_ITEMS_NR;
5123 flush_space(fs_info, space_info, to_reclaim, flush_state);
5124 spin_lock(&space_info->lock);
5125 if (list_empty(&space_info->tickets)) {
5126 space_info->flush = 0;
5127 spin_unlock(&space_info->lock);
5130 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info,
5133 if (last_tickets_id == space_info->tickets_id) {
5136 last_tickets_id = space_info->tickets_id;
5137 flush_state = FLUSH_DELAYED_ITEMS_NR;
5142 if (flush_state > COMMIT_TRANS) {
5144 if (commit_cycles > 2) {
5145 wake_all_tickets(&space_info->tickets);
5146 space_info->flush = 0;
5148 flush_state = FLUSH_DELAYED_ITEMS_NR;
5151 spin_unlock(&space_info->lock);
5152 } while (flush_state <= COMMIT_TRANS);
5155 void btrfs_init_async_reclaim_work(struct work_struct *work)
5157 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5160 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5161 struct btrfs_space_info *space_info,
5162 struct reserve_ticket *ticket)
5165 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5167 spin_lock(&space_info->lock);
5168 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info,
5171 spin_unlock(&space_info->lock);
5174 spin_unlock(&space_info->lock);
5177 flush_space(fs_info, space_info, to_reclaim, flush_state);
5179 spin_lock(&space_info->lock);
5180 if (ticket->bytes == 0) {
5181 spin_unlock(&space_info->lock);
5184 spin_unlock(&space_info->lock);
5187 * Priority flushers can't wait on delalloc without
5190 if (flush_state == FLUSH_DELALLOC ||
5191 flush_state == FLUSH_DELALLOC_WAIT)
5192 flush_state = ALLOC_CHUNK;
5193 } while (flush_state < COMMIT_TRANS);
5196 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5197 struct btrfs_space_info *space_info,
5198 struct reserve_ticket *ticket, u64 orig_bytes)
5204 spin_lock(&space_info->lock);
5205 while (ticket->bytes > 0 && ticket->error == 0) {
5206 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5211 spin_unlock(&space_info->lock);
5215 finish_wait(&ticket->wait, &wait);
5216 spin_lock(&space_info->lock);
5219 ret = ticket->error;
5220 if (!list_empty(&ticket->list))
5221 list_del_init(&ticket->list);
5222 if (ticket->bytes && ticket->bytes < orig_bytes) {
5223 u64 num_bytes = orig_bytes - ticket->bytes;
5224 space_info->bytes_may_use -= num_bytes;
5225 trace_btrfs_space_reservation(fs_info, "space_info",
5226 space_info->flags, num_bytes, 0);
5228 spin_unlock(&space_info->lock);
5234 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5235 * @root - the root we're allocating for
5236 * @space_info - the space info we want to allocate from
5237 * @orig_bytes - the number of bytes we want
5238 * @flush - whether or not we can flush to make our reservation
5240 * This will reserve orig_bytes number of bytes from the space info associated
5241 * with the block_rsv. If there is not enough space it will make an attempt to
5242 * flush out space to make room. It will do this by flushing delalloc if
5243 * possible or committing the transaction. If flush is 0 then no attempts to
5244 * regain reservations will be made and this will fail if there is not enough
5247 static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info,
5248 struct btrfs_space_info *space_info,
5250 enum btrfs_reserve_flush_enum flush,
5253 struct reserve_ticket ticket;
5258 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5260 spin_lock(&space_info->lock);
5262 used = btrfs_space_info_used(space_info, true);
5265 * If we have enough space then hooray, make our reservation and carry
5266 * on. If not see if we can overcommit, and if we can, hooray carry on.
5267 * If not things get more complicated.
5269 if (used + orig_bytes <= space_info->total_bytes) {
5270 space_info->bytes_may_use += orig_bytes;
5271 trace_btrfs_space_reservation(fs_info, "space_info",
5272 space_info->flags, orig_bytes, 1);
5274 } else if (can_overcommit(fs_info, space_info, orig_bytes, flush,
5276 space_info->bytes_may_use += orig_bytes;
5277 trace_btrfs_space_reservation(fs_info, "space_info",
5278 space_info->flags, orig_bytes, 1);
5283 * If we couldn't make a reservation then setup our reservation ticket
5284 * and kick the async worker if it's not already running.
5286 * If we are a priority flusher then we just need to add our ticket to
5287 * the list and we will do our own flushing further down.
5289 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5290 ticket.bytes = orig_bytes;
5292 init_waitqueue_head(&ticket.wait);
5293 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5294 list_add_tail(&ticket.list, &space_info->tickets);
5295 if (!space_info->flush) {
5296 space_info->flush = 1;
5297 trace_btrfs_trigger_flush(fs_info,
5301 queue_work(system_unbound_wq,
5302 &fs_info->async_reclaim_work);
5305 list_add_tail(&ticket.list,
5306 &space_info->priority_tickets);
5308 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5311 * We will do the space reservation dance during log replay,
5312 * which means we won't have fs_info->fs_root set, so don't do
5313 * the async reclaim as we will panic.
5315 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5316 need_do_async_reclaim(fs_info, space_info,
5317 used, system_chunk) &&
5318 !work_busy(&fs_info->async_reclaim_work)) {
5319 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5320 orig_bytes, flush, "preempt");
5321 queue_work(system_unbound_wq,
5322 &fs_info->async_reclaim_work);
5325 spin_unlock(&space_info->lock);
5326 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5329 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5330 return wait_reserve_ticket(fs_info, space_info, &ticket,
5334 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5335 spin_lock(&space_info->lock);
5337 if (ticket.bytes < orig_bytes) {
5338 u64 num_bytes = orig_bytes - ticket.bytes;
5339 space_info->bytes_may_use -= num_bytes;
5340 trace_btrfs_space_reservation(fs_info, "space_info",
5345 list_del_init(&ticket.list);
5348 spin_unlock(&space_info->lock);
5349 ASSERT(list_empty(&ticket.list));
5354 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5355 * @root - the root we're allocating for
5356 * @block_rsv - the block_rsv we're allocating for
5357 * @orig_bytes - the number of bytes we want
5358 * @flush - whether or not we can flush to make our reservation
5360 * This will reserve orgi_bytes number of bytes from the space info associated
5361 * with the block_rsv. If there is not enough space it will make an attempt to
5362 * flush out space to make room. It will do this by flushing delalloc if
5363 * possible or committing the transaction. If flush is 0 then no attempts to
5364 * regain reservations will be made and this will fail if there is not enough
5367 static int reserve_metadata_bytes(struct btrfs_root *root,
5368 struct btrfs_block_rsv *block_rsv,
5370 enum btrfs_reserve_flush_enum flush)
5372 struct btrfs_fs_info *fs_info = root->fs_info;
5373 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5375 bool system_chunk = (root == fs_info->chunk_root);
5377 ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info,
5378 orig_bytes, flush, system_chunk);
5379 if (ret == -ENOSPC &&
5380 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5381 if (block_rsv != global_rsv &&
5382 !block_rsv_use_bytes(global_rsv, orig_bytes))
5385 if (ret == -ENOSPC) {
5386 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5387 block_rsv->space_info->flags,
5390 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG))
5391 dump_space_info(fs_info, block_rsv->space_info,
5397 static struct btrfs_block_rsv *get_block_rsv(
5398 const struct btrfs_trans_handle *trans,
5399 const struct btrfs_root *root)
5401 struct btrfs_fs_info *fs_info = root->fs_info;
5402 struct btrfs_block_rsv *block_rsv = NULL;
5404 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5405 (root == fs_info->csum_root && trans->adding_csums) ||
5406 (root == fs_info->uuid_root))
5407 block_rsv = trans->block_rsv;
5410 block_rsv = root->block_rsv;
5413 block_rsv = &fs_info->empty_block_rsv;
5418 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5422 spin_lock(&block_rsv->lock);
5423 if (block_rsv->reserved >= num_bytes) {
5424 block_rsv->reserved -= num_bytes;
5425 if (block_rsv->reserved < block_rsv->size)
5426 block_rsv->full = 0;
5429 spin_unlock(&block_rsv->lock);
5433 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5434 u64 num_bytes, int update_size)
5436 spin_lock(&block_rsv->lock);
5437 block_rsv->reserved += num_bytes;
5439 block_rsv->size += num_bytes;
5440 else if (block_rsv->reserved >= block_rsv->size)
5441 block_rsv->full = 1;
5442 spin_unlock(&block_rsv->lock);
5445 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5446 struct btrfs_block_rsv *dest, u64 num_bytes,
5449 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5452 if (global_rsv->space_info != dest->space_info)
5455 spin_lock(&global_rsv->lock);
5456 min_bytes = div_factor(global_rsv->size, min_factor);
5457 if (global_rsv->reserved < min_bytes + num_bytes) {
5458 spin_unlock(&global_rsv->lock);
5461 global_rsv->reserved -= num_bytes;
5462 if (global_rsv->reserved < global_rsv->size)
5463 global_rsv->full = 0;
5464 spin_unlock(&global_rsv->lock);
5466 block_rsv_add_bytes(dest, num_bytes, 1);
5471 * This is for space we already have accounted in space_info->bytes_may_use, so
5472 * basically when we're returning space from block_rsv's.
5474 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5475 struct btrfs_space_info *space_info,
5478 struct reserve_ticket *ticket;
5479 struct list_head *head;
5481 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5482 bool check_overcommit = false;
5484 spin_lock(&space_info->lock);
5485 head = &space_info->priority_tickets;
5488 * If we are over our limit then we need to check and see if we can
5489 * overcommit, and if we can't then we just need to free up our space
5490 * and not satisfy any requests.
5492 used = btrfs_space_info_used(space_info, true);
5493 if (used - num_bytes >= space_info->total_bytes)
5494 check_overcommit = true;
5496 while (!list_empty(head) && num_bytes) {
5497 ticket = list_first_entry(head, struct reserve_ticket,
5500 * We use 0 bytes because this space is already reserved, so
5501 * adding the ticket space would be a double count.
5503 if (check_overcommit &&
5504 !can_overcommit(fs_info, space_info, 0, flush, false))
5506 if (num_bytes >= ticket->bytes) {
5507 list_del_init(&ticket->list);
5508 num_bytes -= ticket->bytes;
5510 space_info->tickets_id++;
5511 wake_up(&ticket->wait);
5513 ticket->bytes -= num_bytes;
5518 if (num_bytes && head == &space_info->priority_tickets) {
5519 head = &space_info->tickets;
5520 flush = BTRFS_RESERVE_FLUSH_ALL;
5523 space_info->bytes_may_use -= num_bytes;
5524 trace_btrfs_space_reservation(fs_info, "space_info",
5525 space_info->flags, num_bytes, 0);
5526 spin_unlock(&space_info->lock);
5530 * This is for newly allocated space that isn't accounted in
5531 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5532 * we use this helper.
5534 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5535 struct btrfs_space_info *space_info,
5538 struct reserve_ticket *ticket;
5539 struct list_head *head = &space_info->priority_tickets;
5542 while (!list_empty(head) && num_bytes) {
5543 ticket = list_first_entry(head, struct reserve_ticket,
5545 if (num_bytes >= ticket->bytes) {
5546 trace_btrfs_space_reservation(fs_info, "space_info",
5549 list_del_init(&ticket->list);
5550 num_bytes -= ticket->bytes;
5551 space_info->bytes_may_use += ticket->bytes;
5553 space_info->tickets_id++;
5554 wake_up(&ticket->wait);
5556 trace_btrfs_space_reservation(fs_info, "space_info",
5559 space_info->bytes_may_use += num_bytes;
5560 ticket->bytes -= num_bytes;
5565 if (num_bytes && head == &space_info->priority_tickets) {
5566 head = &space_info->tickets;
5571 static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5572 struct btrfs_block_rsv *block_rsv,
5573 struct btrfs_block_rsv *dest, u64 num_bytes)
5575 struct btrfs_space_info *space_info = block_rsv->space_info;
5578 spin_lock(&block_rsv->lock);
5579 if (num_bytes == (u64)-1)
5580 num_bytes = block_rsv->size;
5581 block_rsv->size -= num_bytes;
5582 if (block_rsv->reserved >= block_rsv->size) {
5583 num_bytes = block_rsv->reserved - block_rsv->size;
5584 block_rsv->reserved = block_rsv->size;
5585 block_rsv->full = 1;
5589 spin_unlock(&block_rsv->lock);
5592 if (num_bytes > 0) {
5594 spin_lock(&dest->lock);
5598 bytes_to_add = dest->size - dest->reserved;
5599 bytes_to_add = min(num_bytes, bytes_to_add);
5600 dest->reserved += bytes_to_add;
5601 if (dest->reserved >= dest->size)
5603 num_bytes -= bytes_to_add;
5605 spin_unlock(&dest->lock);
5608 space_info_add_old_bytes(fs_info, space_info,
5614 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5615 struct btrfs_block_rsv *dst, u64 num_bytes,
5620 ret = block_rsv_use_bytes(src, num_bytes);
5624 block_rsv_add_bytes(dst, num_bytes, update_size);
5628 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5630 memset(rsv, 0, sizeof(*rsv));
5631 spin_lock_init(&rsv->lock);
5635 void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info,
5636 struct btrfs_block_rsv *rsv,
5637 unsigned short type)
5639 btrfs_init_block_rsv(rsv, type);
5640 rsv->space_info = __find_space_info(fs_info,
5641 BTRFS_BLOCK_GROUP_METADATA);
5644 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5645 unsigned short type)
5647 struct btrfs_block_rsv *block_rsv;
5649 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5653 btrfs_init_metadata_block_rsv(fs_info, block_rsv, type);
5657 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5658 struct btrfs_block_rsv *rsv)
5662 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5666 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5671 int btrfs_block_rsv_add(struct btrfs_root *root,
5672 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5673 enum btrfs_reserve_flush_enum flush)
5680 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5682 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5689 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5697 spin_lock(&block_rsv->lock);
5698 num_bytes = div_factor(block_rsv->size, min_factor);
5699 if (block_rsv->reserved >= num_bytes)
5701 spin_unlock(&block_rsv->lock);
5706 int btrfs_block_rsv_refill(struct btrfs_root *root,
5707 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5708 enum btrfs_reserve_flush_enum flush)
5716 spin_lock(&block_rsv->lock);
5717 num_bytes = min_reserved;
5718 if (block_rsv->reserved >= num_bytes)
5721 num_bytes -= block_rsv->reserved;
5722 spin_unlock(&block_rsv->lock);
5727 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5729 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5737 * btrfs_inode_rsv_refill - refill the inode block rsv.
5738 * @inode - the inode we are refilling.
5739 * @flush - the flusing restriction.
5741 * Essentially the same as btrfs_block_rsv_refill, except it uses the
5742 * block_rsv->size as the minimum size. We'll either refill the missing amount
5743 * or return if we already have enough space. This will also handle the resreve
5744 * tracepoint for the reserved amount.
5746 static int btrfs_inode_rsv_refill(struct btrfs_inode *inode,
5747 enum btrfs_reserve_flush_enum flush)
5749 struct btrfs_root *root = inode->root;
5750 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5754 spin_lock(&block_rsv->lock);
5755 if (block_rsv->reserved < block_rsv->size)
5756 num_bytes = block_rsv->size - block_rsv->reserved;
5757 spin_unlock(&block_rsv->lock);
5762 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
5765 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5767 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5768 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5769 btrfs_ino(inode), num_bytes, 1);
5775 * btrfs_inode_rsv_release - release any excessive reservation.
5776 * @inode - the inode we need to release from.
5777 * @qgroup_free - free or convert qgroup meta.
5778 * Unlike normal operation, qgroup meta reservation needs to know if we are
5779 * freeing qgroup reservation or just converting it into per-trans. Normally
5780 * @qgroup_free is true for error handling, and false for normal release.
5782 * This is the same as btrfs_block_rsv_release, except that it handles the
5783 * tracepoint for the reservation.
5785 static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free)
5787 struct btrfs_fs_info *fs_info = inode->root->fs_info;
5788 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5789 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
5793 * Since we statically set the block_rsv->size we just want to say we
5794 * are releasing 0 bytes, and then we'll just get the reservation over
5797 released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0);
5799 trace_btrfs_space_reservation(fs_info, "delalloc",
5800 btrfs_ino(inode), released, 0);
5802 btrfs_qgroup_free_meta_prealloc(inode->root, released);
5804 btrfs_qgroup_convert_reserved_meta(inode->root, released);
5807 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5808 struct btrfs_block_rsv *block_rsv,
5811 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5813 if (global_rsv == block_rsv ||
5814 block_rsv->space_info != global_rsv->space_info)
5816 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5819 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5821 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5822 struct btrfs_space_info *sinfo = block_rsv->space_info;
5826 * The global block rsv is based on the size of the extent tree, the
5827 * checksum tree and the root tree. If the fs is empty we want to set
5828 * it to a minimal amount for safety.
5830 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5831 btrfs_root_used(&fs_info->csum_root->root_item) +
5832 btrfs_root_used(&fs_info->tree_root->root_item);
5833 num_bytes = max_t(u64, num_bytes, SZ_16M);
5835 spin_lock(&sinfo->lock);
5836 spin_lock(&block_rsv->lock);
5838 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5840 if (block_rsv->reserved < block_rsv->size) {
5841 num_bytes = btrfs_space_info_used(sinfo, true);
5842 if (sinfo->total_bytes > num_bytes) {
5843 num_bytes = sinfo->total_bytes - num_bytes;
5844 num_bytes = min(num_bytes,
5845 block_rsv->size - block_rsv->reserved);
5846 block_rsv->reserved += num_bytes;
5847 sinfo->bytes_may_use += num_bytes;
5848 trace_btrfs_space_reservation(fs_info, "space_info",
5849 sinfo->flags, num_bytes,
5852 } else if (block_rsv->reserved > block_rsv->size) {
5853 num_bytes = block_rsv->reserved - block_rsv->size;
5854 sinfo->bytes_may_use -= num_bytes;
5855 trace_btrfs_space_reservation(fs_info, "space_info",
5856 sinfo->flags, num_bytes, 0);
5857 block_rsv->reserved = block_rsv->size;
5860 if (block_rsv->reserved == block_rsv->size)
5861 block_rsv->full = 1;
5863 block_rsv->full = 0;
5865 spin_unlock(&block_rsv->lock);
5866 spin_unlock(&sinfo->lock);
5869 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5871 struct btrfs_space_info *space_info;
5873 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5874 fs_info->chunk_block_rsv.space_info = space_info;
5876 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5877 fs_info->global_block_rsv.space_info = space_info;
5878 fs_info->trans_block_rsv.space_info = space_info;
5879 fs_info->empty_block_rsv.space_info = space_info;
5880 fs_info->delayed_block_rsv.space_info = space_info;
5882 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5883 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5884 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5885 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5886 if (fs_info->quota_root)
5887 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5888 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5890 update_global_block_rsv(fs_info);
5893 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5895 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5897 WARN_ON(fs_info->trans_block_rsv.size > 0);
5898 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5899 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5900 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5901 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5902 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5907 * To be called after all the new block groups attached to the transaction
5908 * handle have been created (btrfs_create_pending_block_groups()).
5910 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5912 struct btrfs_fs_info *fs_info = trans->fs_info;
5914 if (!trans->chunk_bytes_reserved)
5917 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5919 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5920 trans->chunk_bytes_reserved);
5921 trans->chunk_bytes_reserved = 0;
5924 /* Can only return 0 or -ENOSPC */
5925 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5926 struct btrfs_inode *inode)
5928 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5929 struct btrfs_root *root = inode->root;
5931 * We always use trans->block_rsv here as we will have reserved space
5932 * for our orphan when starting the transaction, using get_block_rsv()
5933 * here will sometimes make us choose the wrong block rsv as we could be
5934 * doing a reloc inode for a non refcounted root.
5936 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5937 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5940 * We need to hold space in order to delete our orphan item once we've
5941 * added it, so this takes the reservation so we can release it later
5942 * when we are truly done with the orphan item.
5944 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5946 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5948 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5951 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5953 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5954 struct btrfs_root *root = inode->root;
5955 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5957 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5959 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5963 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5964 * root: the root of the parent directory
5965 * rsv: block reservation
5966 * items: the number of items that we need do reservation
5967 * qgroup_reserved: used to return the reserved size in qgroup
5969 * This function is used to reserve the space for snapshot/subvolume
5970 * creation and deletion. Those operations are different with the
5971 * common file/directory operations, they change two fs/file trees
5972 * and root tree, the number of items that the qgroup reserves is
5973 * different with the free space reservation. So we can not use
5974 * the space reservation mechanism in start_transaction().
5976 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5977 struct btrfs_block_rsv *rsv,
5979 u64 *qgroup_reserved,
5980 bool use_global_rsv)
5984 struct btrfs_fs_info *fs_info = root->fs_info;
5985 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5987 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5988 /* One for parent inode, two for dir entries */
5989 num_bytes = 3 * fs_info->nodesize;
5990 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
5997 *qgroup_reserved = num_bytes;
5999 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
6000 rsv->space_info = __find_space_info(fs_info,
6001 BTRFS_BLOCK_GROUP_METADATA);
6002 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
6003 BTRFS_RESERVE_FLUSH_ALL);
6005 if (ret == -ENOSPC && use_global_rsv)
6006 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
6008 if (ret && *qgroup_reserved)
6009 btrfs_qgroup_free_meta_prealloc(root, *qgroup_reserved);
6014 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
6015 struct btrfs_block_rsv *rsv)
6017 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
6020 static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info,
6021 struct btrfs_inode *inode)
6023 struct btrfs_block_rsv *block_rsv = &inode->block_rsv;
6024 u64 reserve_size = 0;
6026 unsigned outstanding_extents;
6028 lockdep_assert_held(&inode->lock);
6029 outstanding_extents = inode->outstanding_extents;
6030 if (outstanding_extents)
6031 reserve_size = btrfs_calc_trans_metadata_size(fs_info,
6032 outstanding_extents + 1);
6033 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info,
6035 reserve_size += btrfs_calc_trans_metadata_size(fs_info,
6038 spin_lock(&block_rsv->lock);
6039 block_rsv->size = reserve_size;
6040 spin_unlock(&block_rsv->lock);
6043 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
6045 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6046 unsigned nr_extents;
6047 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
6049 bool delalloc_lock = true;
6051 /* If we are a free space inode we need to not flush since we will be in
6052 * the middle of a transaction commit. We also don't need the delalloc
6053 * mutex since we won't race with anybody. We need this mostly to make
6054 * lockdep shut its filthy mouth.
6056 * If we have a transaction open (can happen if we call truncate_block
6057 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
6059 if (btrfs_is_free_space_inode(inode)) {
6060 flush = BTRFS_RESERVE_NO_FLUSH;
6061 delalloc_lock = false;
6063 if (current->journal_info)
6064 flush = BTRFS_RESERVE_FLUSH_LIMIT;
6066 if (btrfs_transaction_in_commit(fs_info))
6067 schedule_timeout(1);
6071 mutex_lock(&inode->delalloc_mutex);
6073 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6075 /* Add our new extents and calculate the new rsv size. */
6076 spin_lock(&inode->lock);
6077 nr_extents = count_max_extents(num_bytes);
6078 btrfs_mod_outstanding_extents(inode, nr_extents);
6079 inode->csum_bytes += num_bytes;
6080 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6081 spin_unlock(&inode->lock);
6083 ret = btrfs_inode_rsv_refill(inode, flush);
6088 mutex_unlock(&inode->delalloc_mutex);
6092 spin_lock(&inode->lock);
6093 nr_extents = count_max_extents(num_bytes);
6094 btrfs_mod_outstanding_extents(inode, -nr_extents);
6095 inode->csum_bytes -= num_bytes;
6096 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6097 spin_unlock(&inode->lock);
6099 btrfs_inode_rsv_release(inode, true);
6101 mutex_unlock(&inode->delalloc_mutex);
6106 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6107 * @inode: the inode to release the reservation for.
6108 * @num_bytes: the number of bytes we are releasing.
6109 * @qgroup_free: free qgroup reservation or convert it to per-trans reservation
6111 * This will release the metadata reservation for an inode. This can be called
6112 * once we complete IO for a given set of bytes to release their metadata
6113 * reservations, or on error for the same reason.
6115 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes,
6118 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6120 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6121 spin_lock(&inode->lock);
6122 inode->csum_bytes -= num_bytes;
6123 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6124 spin_unlock(&inode->lock);
6126 if (btrfs_is_testing(fs_info))
6129 btrfs_inode_rsv_release(inode, qgroup_free);
6133 * btrfs_delalloc_release_extents - release our outstanding_extents
6134 * @inode: the inode to balance the reservation for.
6135 * @num_bytes: the number of bytes we originally reserved with
6136 * @qgroup_free: do we need to free qgroup meta reservation or convert them.
6138 * When we reserve space we increase outstanding_extents for the extents we may
6139 * add. Once we've set the range as delalloc or created our ordered extents we
6140 * have outstanding_extents to track the real usage, so we use this to free our
6141 * temporarily tracked outstanding_extents. This _must_ be used in conjunction
6142 * with btrfs_delalloc_reserve_metadata.
6144 void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes,
6147 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6148 unsigned num_extents;
6150 spin_lock(&inode->lock);
6151 num_extents = count_max_extents(num_bytes);
6152 btrfs_mod_outstanding_extents(inode, -num_extents);
6153 btrfs_calculate_inode_block_rsv_size(fs_info, inode);
6154 spin_unlock(&inode->lock);
6156 if (btrfs_is_testing(fs_info))
6159 btrfs_inode_rsv_release(inode, qgroup_free);
6163 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6165 * @inode: inode we're writing to
6166 * @start: start range we are writing to
6167 * @len: how long the range we are writing to
6168 * @reserved: mandatory parameter, record actually reserved qgroup ranges of
6169 * current reservation.
6171 * This will do the following things
6173 * o reserve space in data space info for num bytes
6174 * and reserve precious corresponding qgroup space
6175 * (Done in check_data_free_space)
6177 * o reserve space for metadata space, based on the number of outstanding
6178 * extents and how much csums will be needed
6179 * also reserve metadata space in a per root over-reserve method.
6180 * o add to the inodes->delalloc_bytes
6181 * o add it to the fs_info's delalloc inodes list.
6182 * (Above 3 all done in delalloc_reserve_metadata)
6184 * Return 0 for success
6185 * Return <0 for error(-ENOSPC or -EQUOT)
6187 int btrfs_delalloc_reserve_space(struct inode *inode,
6188 struct extent_changeset **reserved, u64 start, u64 len)
6192 ret = btrfs_check_data_free_space(inode, reserved, start, len);
6195 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6197 btrfs_free_reserved_data_space(inode, *reserved, start, len);
6202 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6203 * @inode: inode we're releasing space for
6204 * @start: start position of the space already reserved
6205 * @len: the len of the space already reserved
6206 * @release_bytes: the len of the space we consumed or didn't use
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,
6215 u64 start, u64 len, bool qgroup_free)
6217 btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free);
6218 btrfs_free_reserved_data_space(inode, reserved, start, len);
6221 static int update_block_group(struct btrfs_trans_handle *trans,
6222 struct btrfs_fs_info *info, u64 bytenr,
6223 u64 num_bytes, int alloc)
6225 struct btrfs_block_group_cache *cache = NULL;
6226 u64 total = num_bytes;
6231 /* block accounting for super block */
6232 spin_lock(&info->delalloc_root_lock);
6233 old_val = btrfs_super_bytes_used(info->super_copy);
6235 old_val += num_bytes;
6237 old_val -= num_bytes;
6238 btrfs_set_super_bytes_used(info->super_copy, old_val);
6239 spin_unlock(&info->delalloc_root_lock);
6242 cache = btrfs_lookup_block_group(info, bytenr);
6245 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6246 BTRFS_BLOCK_GROUP_RAID1 |
6247 BTRFS_BLOCK_GROUP_RAID10))
6252 * If this block group has free space cache written out, we
6253 * need to make sure to load it if we are removing space. This
6254 * is because we need the unpinning stage to actually add the
6255 * space back to the block group, otherwise we will leak space.
6257 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6258 cache_block_group(cache, 1);
6260 byte_in_group = bytenr - cache->key.objectid;
6261 WARN_ON(byte_in_group > cache->key.offset);
6263 spin_lock(&cache->space_info->lock);
6264 spin_lock(&cache->lock);
6266 if (btrfs_test_opt(info, SPACE_CACHE) &&
6267 cache->disk_cache_state < BTRFS_DC_CLEAR)
6268 cache->disk_cache_state = BTRFS_DC_CLEAR;
6270 old_val = btrfs_block_group_used(&cache->item);
6271 num_bytes = min(total, cache->key.offset - byte_in_group);
6273 old_val += num_bytes;
6274 btrfs_set_block_group_used(&cache->item, old_val);
6275 cache->reserved -= num_bytes;
6276 cache->space_info->bytes_reserved -= num_bytes;
6277 cache->space_info->bytes_used += num_bytes;
6278 cache->space_info->disk_used += num_bytes * factor;
6279 spin_unlock(&cache->lock);
6280 spin_unlock(&cache->space_info->lock);
6282 old_val -= num_bytes;
6283 btrfs_set_block_group_used(&cache->item, old_val);
6284 cache->pinned += num_bytes;
6285 cache->space_info->bytes_pinned += num_bytes;
6286 cache->space_info->bytes_used -= num_bytes;
6287 cache->space_info->disk_used -= num_bytes * factor;
6288 spin_unlock(&cache->lock);
6289 spin_unlock(&cache->space_info->lock);
6291 trace_btrfs_space_reservation(info, "pinned",
6292 cache->space_info->flags,
6294 percpu_counter_add(&cache->space_info->total_bytes_pinned,
6296 set_extent_dirty(info->pinned_extents,
6297 bytenr, bytenr + num_bytes - 1,
6298 GFP_NOFS | __GFP_NOFAIL);
6301 spin_lock(&trans->transaction->dirty_bgs_lock);
6302 if (list_empty(&cache->dirty_list)) {
6303 list_add_tail(&cache->dirty_list,
6304 &trans->transaction->dirty_bgs);
6305 trans->transaction->num_dirty_bgs++;
6306 btrfs_get_block_group(cache);
6308 spin_unlock(&trans->transaction->dirty_bgs_lock);
6311 * No longer have used bytes in this block group, queue it for
6312 * deletion. We do this after adding the block group to the
6313 * dirty list to avoid races between cleaner kthread and space
6316 if (!alloc && old_val == 0) {
6317 spin_lock(&info->unused_bgs_lock);
6318 if (list_empty(&cache->bg_list)) {
6319 btrfs_get_block_group(cache);
6320 list_add_tail(&cache->bg_list,
6323 spin_unlock(&info->unused_bgs_lock);
6326 btrfs_put_block_group(cache);
6328 bytenr += num_bytes;
6333 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6335 struct btrfs_block_group_cache *cache;
6338 spin_lock(&fs_info->block_group_cache_lock);
6339 bytenr = fs_info->first_logical_byte;
6340 spin_unlock(&fs_info->block_group_cache_lock);
6342 if (bytenr < (u64)-1)
6345 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6349 bytenr = cache->key.objectid;
6350 btrfs_put_block_group(cache);
6355 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6356 struct btrfs_block_group_cache *cache,
6357 u64 bytenr, u64 num_bytes, int reserved)
6359 spin_lock(&cache->space_info->lock);
6360 spin_lock(&cache->lock);
6361 cache->pinned += num_bytes;
6362 cache->space_info->bytes_pinned += num_bytes;
6364 cache->reserved -= num_bytes;
6365 cache->space_info->bytes_reserved -= num_bytes;
6367 spin_unlock(&cache->lock);
6368 spin_unlock(&cache->space_info->lock);
6370 trace_btrfs_space_reservation(fs_info, "pinned",
6371 cache->space_info->flags, num_bytes, 1);
6372 percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes);
6373 set_extent_dirty(fs_info->pinned_extents, bytenr,
6374 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6379 * this function must be called within transaction
6381 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6382 u64 bytenr, u64 num_bytes, int reserved)
6384 struct btrfs_block_group_cache *cache;
6386 cache = btrfs_lookup_block_group(fs_info, bytenr);
6387 BUG_ON(!cache); /* Logic error */
6389 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6391 btrfs_put_block_group(cache);
6396 * this function must be called within transaction
6398 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6399 u64 bytenr, u64 num_bytes)
6401 struct btrfs_block_group_cache *cache;
6404 cache = btrfs_lookup_block_group(fs_info, bytenr);
6409 * pull in the free space cache (if any) so that our pin
6410 * removes the free space from the cache. We have load_only set
6411 * to one because the slow code to read in the free extents does check
6412 * the pinned extents.
6414 cache_block_group(cache, 1);
6416 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6418 /* remove us from the free space cache (if we're there at all) */
6419 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6420 btrfs_put_block_group(cache);
6424 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6425 u64 start, u64 num_bytes)
6428 struct btrfs_block_group_cache *block_group;
6429 struct btrfs_caching_control *caching_ctl;
6431 block_group = btrfs_lookup_block_group(fs_info, start);
6435 cache_block_group(block_group, 0);
6436 caching_ctl = get_caching_control(block_group);
6440 BUG_ON(!block_group_cache_done(block_group));
6441 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6443 mutex_lock(&caching_ctl->mutex);
6445 if (start >= caching_ctl->progress) {
6446 ret = add_excluded_extent(fs_info, start, num_bytes);
6447 } else if (start + num_bytes <= caching_ctl->progress) {
6448 ret = btrfs_remove_free_space(block_group,
6451 num_bytes = caching_ctl->progress - start;
6452 ret = btrfs_remove_free_space(block_group,
6457 num_bytes = (start + num_bytes) -
6458 caching_ctl->progress;
6459 start = caching_ctl->progress;
6460 ret = add_excluded_extent(fs_info, start, num_bytes);
6463 mutex_unlock(&caching_ctl->mutex);
6464 put_caching_control(caching_ctl);
6466 btrfs_put_block_group(block_group);
6470 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6471 struct extent_buffer *eb)
6473 struct btrfs_file_extent_item *item;
6474 struct btrfs_key key;
6478 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6481 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6482 btrfs_item_key_to_cpu(eb, &key, i);
6483 if (key.type != BTRFS_EXTENT_DATA_KEY)
6485 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6486 found_type = btrfs_file_extent_type(eb, item);
6487 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6489 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6491 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6492 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6493 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6500 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6502 atomic_inc(&bg->reservations);
6505 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6508 struct btrfs_block_group_cache *bg;
6510 bg = btrfs_lookup_block_group(fs_info, start);
6512 if (atomic_dec_and_test(&bg->reservations))
6513 wake_up_var(&bg->reservations);
6514 btrfs_put_block_group(bg);
6517 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6519 struct btrfs_space_info *space_info = bg->space_info;
6523 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6527 * Our block group is read only but before we set it to read only,
6528 * some task might have had allocated an extent from it already, but it
6529 * has not yet created a respective ordered extent (and added it to a
6530 * root's list of ordered extents).
6531 * Therefore wait for any task currently allocating extents, since the
6532 * block group's reservations counter is incremented while a read lock
6533 * on the groups' semaphore is held and decremented after releasing
6534 * the read access on that semaphore and creating the ordered extent.
6536 down_write(&space_info->groups_sem);
6537 up_write(&space_info->groups_sem);
6539 wait_var_event(&bg->reservations, !atomic_read(&bg->reservations));
6543 * btrfs_add_reserved_bytes - update the block_group and space info counters
6544 * @cache: The cache we are manipulating
6545 * @ram_bytes: The number of bytes of file content, and will be same to
6546 * @num_bytes except for the compress path.
6547 * @num_bytes: The number of bytes in question
6548 * @delalloc: The blocks are allocated for the delalloc write
6550 * This is called by the allocator when it reserves space. If this is a
6551 * reservation and the block group has become read only we cannot make the
6552 * reservation and return -EAGAIN, otherwise this function always succeeds.
6554 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6555 u64 ram_bytes, u64 num_bytes, int delalloc)
6557 struct btrfs_space_info *space_info = cache->space_info;
6560 spin_lock(&space_info->lock);
6561 spin_lock(&cache->lock);
6565 cache->reserved += num_bytes;
6566 space_info->bytes_reserved += num_bytes;
6568 trace_btrfs_space_reservation(cache->fs_info,
6569 "space_info", space_info->flags,
6571 space_info->bytes_may_use -= ram_bytes;
6573 cache->delalloc_bytes += num_bytes;
6575 spin_unlock(&cache->lock);
6576 spin_unlock(&space_info->lock);
6581 * btrfs_free_reserved_bytes - update the block_group and space info counters
6582 * @cache: The cache we are manipulating
6583 * @num_bytes: The number of bytes in question
6584 * @delalloc: The blocks are allocated for the delalloc write
6586 * This is called by somebody who is freeing space that was never actually used
6587 * on disk. For example if you reserve some space for a new leaf in transaction
6588 * A and before transaction A commits you free that leaf, you call this with
6589 * reserve set to 0 in order to clear the reservation.
6592 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6593 u64 num_bytes, int delalloc)
6595 struct btrfs_space_info *space_info = cache->space_info;
6598 spin_lock(&space_info->lock);
6599 spin_lock(&cache->lock);
6601 space_info->bytes_readonly += num_bytes;
6602 cache->reserved -= num_bytes;
6603 space_info->bytes_reserved -= num_bytes;
6606 cache->delalloc_bytes -= num_bytes;
6607 spin_unlock(&cache->lock);
6608 spin_unlock(&space_info->lock);
6611 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6613 struct btrfs_caching_control *next;
6614 struct btrfs_caching_control *caching_ctl;
6615 struct btrfs_block_group_cache *cache;
6617 down_write(&fs_info->commit_root_sem);
6619 list_for_each_entry_safe(caching_ctl, next,
6620 &fs_info->caching_block_groups, list) {
6621 cache = caching_ctl->block_group;
6622 if (block_group_cache_done(cache)) {
6623 cache->last_byte_to_unpin = (u64)-1;
6624 list_del_init(&caching_ctl->list);
6625 put_caching_control(caching_ctl);
6627 cache->last_byte_to_unpin = caching_ctl->progress;
6631 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6632 fs_info->pinned_extents = &fs_info->freed_extents[1];
6634 fs_info->pinned_extents = &fs_info->freed_extents[0];
6636 up_write(&fs_info->commit_root_sem);
6638 update_global_block_rsv(fs_info);
6642 * Returns the free cluster for the given space info and sets empty_cluster to
6643 * what it should be based on the mount options.
6645 static struct btrfs_free_cluster *
6646 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6647 struct btrfs_space_info *space_info, u64 *empty_cluster)
6649 struct btrfs_free_cluster *ret = NULL;
6652 if (btrfs_mixed_space_info(space_info))
6655 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6656 ret = &fs_info->meta_alloc_cluster;
6657 if (btrfs_test_opt(fs_info, SSD))
6658 *empty_cluster = SZ_2M;
6660 *empty_cluster = SZ_64K;
6661 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) &&
6662 btrfs_test_opt(fs_info, SSD_SPREAD)) {
6663 *empty_cluster = SZ_2M;
6664 ret = &fs_info->data_alloc_cluster;
6670 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6672 const bool return_free_space)
6674 struct btrfs_block_group_cache *cache = NULL;
6675 struct btrfs_space_info *space_info;
6676 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6677 struct btrfs_free_cluster *cluster = NULL;
6679 u64 total_unpinned = 0;
6680 u64 empty_cluster = 0;
6683 while (start <= end) {
6686 start >= cache->key.objectid + cache->key.offset) {
6688 btrfs_put_block_group(cache);
6690 cache = btrfs_lookup_block_group(fs_info, start);
6691 BUG_ON(!cache); /* Logic error */
6693 cluster = fetch_cluster_info(fs_info,
6696 empty_cluster <<= 1;
6699 len = cache->key.objectid + cache->key.offset - start;
6700 len = min(len, end + 1 - start);
6702 if (start < cache->last_byte_to_unpin) {
6703 len = min(len, cache->last_byte_to_unpin - start);
6704 if (return_free_space)
6705 btrfs_add_free_space(cache, start, len);
6709 total_unpinned += len;
6710 space_info = cache->space_info;
6713 * If this space cluster has been marked as fragmented and we've
6714 * unpinned enough in this block group to potentially allow a
6715 * cluster to be created inside of it go ahead and clear the
6718 if (cluster && cluster->fragmented &&
6719 total_unpinned > empty_cluster) {
6720 spin_lock(&cluster->lock);
6721 cluster->fragmented = 0;
6722 spin_unlock(&cluster->lock);
6725 spin_lock(&space_info->lock);
6726 spin_lock(&cache->lock);
6727 cache->pinned -= len;
6728 space_info->bytes_pinned -= len;
6730 trace_btrfs_space_reservation(fs_info, "pinned",
6731 space_info->flags, len, 0);
6732 space_info->max_extent_size = 0;
6733 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6735 space_info->bytes_readonly += len;
6738 spin_unlock(&cache->lock);
6739 if (!readonly && return_free_space &&
6740 global_rsv->space_info == space_info) {
6743 spin_lock(&global_rsv->lock);
6744 if (!global_rsv->full) {
6745 to_add = min(len, global_rsv->size -
6746 global_rsv->reserved);
6747 global_rsv->reserved += to_add;
6748 space_info->bytes_may_use += to_add;
6749 if (global_rsv->reserved >= global_rsv->size)
6750 global_rsv->full = 1;
6751 trace_btrfs_space_reservation(fs_info,
6757 spin_unlock(&global_rsv->lock);
6758 /* Add to any tickets we may have */
6760 space_info_add_new_bytes(fs_info, space_info,
6763 spin_unlock(&space_info->lock);
6767 btrfs_put_block_group(cache);
6771 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans)
6773 struct btrfs_fs_info *fs_info = trans->fs_info;
6774 struct btrfs_block_group_cache *block_group, *tmp;
6775 struct list_head *deleted_bgs;
6776 struct extent_io_tree *unpin;
6781 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6782 unpin = &fs_info->freed_extents[1];
6784 unpin = &fs_info->freed_extents[0];
6786 while (!trans->aborted) {
6787 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6788 ret = find_first_extent_bit(unpin, 0, &start, &end,
6789 EXTENT_DIRTY, NULL);
6791 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6795 if (btrfs_test_opt(fs_info, DISCARD))
6796 ret = btrfs_discard_extent(fs_info, start,
6797 end + 1 - start, NULL);
6799 clear_extent_dirty(unpin, start, end);
6800 unpin_extent_range(fs_info, start, end, true);
6801 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6806 * Transaction is finished. We don't need the lock anymore. We
6807 * do need to clean up the block groups in case of a transaction
6810 deleted_bgs = &trans->transaction->deleted_bgs;
6811 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6815 if (!trans->aborted)
6816 ret = btrfs_discard_extent(fs_info,
6817 block_group->key.objectid,
6818 block_group->key.offset,
6821 list_del_init(&block_group->bg_list);
6822 btrfs_put_block_group_trimming(block_group);
6823 btrfs_put_block_group(block_group);
6826 const char *errstr = btrfs_decode_error(ret);
6828 "discard failed while removing blockgroup: errno=%d %s",
6836 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6837 struct btrfs_fs_info *info,
6838 struct btrfs_delayed_ref_node *node, u64 parent,
6839 u64 root_objectid, u64 owner_objectid,
6840 u64 owner_offset, int refs_to_drop,
6841 struct btrfs_delayed_extent_op *extent_op)
6843 struct btrfs_key key;
6844 struct btrfs_path *path;
6845 struct btrfs_root *extent_root = info->extent_root;
6846 struct extent_buffer *leaf;
6847 struct btrfs_extent_item *ei;
6848 struct btrfs_extent_inline_ref *iref;
6851 int extent_slot = 0;
6852 int found_extent = 0;
6856 u64 bytenr = node->bytenr;
6857 u64 num_bytes = node->num_bytes;
6859 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6861 path = btrfs_alloc_path();
6865 path->reada = READA_FORWARD;
6866 path->leave_spinning = 1;
6868 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6869 BUG_ON(!is_data && refs_to_drop != 1);
6872 skinny_metadata = false;
6874 ret = lookup_extent_backref(trans, info, path, &iref,
6875 bytenr, num_bytes, parent,
6876 root_objectid, owner_objectid,
6879 extent_slot = path->slots[0];
6880 while (extent_slot >= 0) {
6881 btrfs_item_key_to_cpu(path->nodes[0], &key,
6883 if (key.objectid != bytenr)
6885 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6886 key.offset == num_bytes) {
6890 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6891 key.offset == owner_objectid) {
6895 if (path->slots[0] - extent_slot > 5)
6899 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6900 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6901 if (found_extent && item_size < sizeof(*ei))
6904 if (!found_extent) {
6906 ret = remove_extent_backref(trans, info, path, NULL,
6908 is_data, &last_ref);
6910 btrfs_abort_transaction(trans, ret);
6913 btrfs_release_path(path);
6914 path->leave_spinning = 1;
6916 key.objectid = bytenr;
6917 key.type = BTRFS_EXTENT_ITEM_KEY;
6918 key.offset = num_bytes;
6920 if (!is_data && skinny_metadata) {
6921 key.type = BTRFS_METADATA_ITEM_KEY;
6922 key.offset = owner_objectid;
6925 ret = btrfs_search_slot(trans, extent_root,
6927 if (ret > 0 && skinny_metadata && path->slots[0]) {
6929 * Couldn't find our skinny metadata item,
6930 * see if we have ye olde extent item.
6933 btrfs_item_key_to_cpu(path->nodes[0], &key,
6935 if (key.objectid == bytenr &&
6936 key.type == BTRFS_EXTENT_ITEM_KEY &&
6937 key.offset == num_bytes)
6941 if (ret > 0 && skinny_metadata) {
6942 skinny_metadata = false;
6943 key.objectid = bytenr;
6944 key.type = BTRFS_EXTENT_ITEM_KEY;
6945 key.offset = num_bytes;
6946 btrfs_release_path(path);
6947 ret = btrfs_search_slot(trans, extent_root,
6953 "umm, got %d back from search, was looking for %llu",
6956 btrfs_print_leaf(path->nodes[0]);
6959 btrfs_abort_transaction(trans, ret);
6962 extent_slot = path->slots[0];
6964 } else if (WARN_ON(ret == -ENOENT)) {
6965 btrfs_print_leaf(path->nodes[0]);
6967 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6968 bytenr, parent, root_objectid, owner_objectid,
6970 btrfs_abort_transaction(trans, ret);
6973 btrfs_abort_transaction(trans, ret);
6977 leaf = path->nodes[0];
6978 item_size = btrfs_item_size_nr(leaf, extent_slot);
6979 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6980 if (item_size < sizeof(*ei)) {
6981 BUG_ON(found_extent || extent_slot != path->slots[0]);
6982 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
6985 btrfs_abort_transaction(trans, ret);
6989 btrfs_release_path(path);
6990 path->leave_spinning = 1;
6992 key.objectid = bytenr;
6993 key.type = BTRFS_EXTENT_ITEM_KEY;
6994 key.offset = num_bytes;
6996 ret = btrfs_search_slot(trans, extent_root, &key, path,
7000 "umm, got %d back from search, was looking for %llu",
7002 btrfs_print_leaf(path->nodes[0]);
7005 btrfs_abort_transaction(trans, ret);
7009 extent_slot = path->slots[0];
7010 leaf = path->nodes[0];
7011 item_size = btrfs_item_size_nr(leaf, extent_slot);
7014 BUG_ON(item_size < sizeof(*ei));
7015 ei = btrfs_item_ptr(leaf, extent_slot,
7016 struct btrfs_extent_item);
7017 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7018 key.type == BTRFS_EXTENT_ITEM_KEY) {
7019 struct btrfs_tree_block_info *bi;
7020 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7021 bi = (struct btrfs_tree_block_info *)(ei + 1);
7022 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7025 refs = btrfs_extent_refs(leaf, ei);
7026 if (refs < refs_to_drop) {
7028 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7029 refs_to_drop, refs, bytenr);
7031 btrfs_abort_transaction(trans, ret);
7034 refs -= refs_to_drop;
7038 __run_delayed_extent_op(extent_op, leaf, ei);
7040 * In the case of inline back ref, reference count will
7041 * be updated by remove_extent_backref
7044 BUG_ON(!found_extent);
7046 btrfs_set_extent_refs(leaf, ei, refs);
7047 btrfs_mark_buffer_dirty(leaf);
7050 ret = remove_extent_backref(trans, info, path,
7052 is_data, &last_ref);
7054 btrfs_abort_transaction(trans, ret);
7060 BUG_ON(is_data && refs_to_drop !=
7061 extent_data_ref_count(path, iref));
7063 BUG_ON(path->slots[0] != extent_slot);
7065 BUG_ON(path->slots[0] != extent_slot + 1);
7066 path->slots[0] = extent_slot;
7072 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7075 btrfs_abort_transaction(trans, ret);
7078 btrfs_release_path(path);
7081 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7083 btrfs_abort_transaction(trans, ret);
7088 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7090 btrfs_abort_transaction(trans, ret);
7094 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7096 btrfs_abort_transaction(trans, ret);
7100 btrfs_release_path(path);
7103 btrfs_free_path(path);
7108 * when we free an block, it is possible (and likely) that we free the last
7109 * delayed ref for that extent as well. This searches the delayed ref tree for
7110 * a given extent, and if there are no other delayed refs to be processed, it
7111 * removes it from the tree.
7113 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7116 struct btrfs_delayed_ref_head *head;
7117 struct btrfs_delayed_ref_root *delayed_refs;
7120 delayed_refs = &trans->transaction->delayed_refs;
7121 spin_lock(&delayed_refs->lock);
7122 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7124 goto out_delayed_unlock;
7126 spin_lock(&head->lock);
7127 if (!RB_EMPTY_ROOT(&head->ref_tree))
7130 if (head->extent_op) {
7131 if (!head->must_insert_reserved)
7133 btrfs_free_delayed_extent_op(head->extent_op);
7134 head->extent_op = NULL;
7138 * waiting for the lock here would deadlock. If someone else has it
7139 * locked they are already in the process of dropping it anyway
7141 if (!mutex_trylock(&head->mutex))
7145 * at this point we have a head with no other entries. Go
7146 * ahead and process it.
7148 rb_erase(&head->href_node, &delayed_refs->href_root);
7149 RB_CLEAR_NODE(&head->href_node);
7150 atomic_dec(&delayed_refs->num_entries);
7153 * we don't take a ref on the node because we're removing it from the
7154 * tree, so we just steal the ref the tree was holding.
7156 delayed_refs->num_heads--;
7157 if (head->processing == 0)
7158 delayed_refs->num_heads_ready--;
7159 head->processing = 0;
7160 spin_unlock(&head->lock);
7161 spin_unlock(&delayed_refs->lock);
7163 BUG_ON(head->extent_op);
7164 if (head->must_insert_reserved)
7167 mutex_unlock(&head->mutex);
7168 btrfs_put_delayed_ref_head(head);
7171 spin_unlock(&head->lock);
7174 spin_unlock(&delayed_refs->lock);
7178 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7179 struct btrfs_root *root,
7180 struct extent_buffer *buf,
7181 u64 parent, int last_ref)
7183 struct btrfs_fs_info *fs_info = root->fs_info;
7187 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7188 int old_ref_mod, new_ref_mod;
7190 btrfs_ref_tree_mod(root, buf->start, buf->len, parent,
7191 root->root_key.objectid,
7192 btrfs_header_level(buf), 0,
7193 BTRFS_DROP_DELAYED_REF);
7194 ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start,
7196 root->root_key.objectid,
7197 btrfs_header_level(buf),
7198 BTRFS_DROP_DELAYED_REF, NULL,
7199 &old_ref_mod, &new_ref_mod);
7200 BUG_ON(ret); /* -ENOMEM */
7201 pin = old_ref_mod >= 0 && new_ref_mod < 0;
7204 if (last_ref && btrfs_header_generation(buf) == trans->transid) {
7205 struct btrfs_block_group_cache *cache;
7207 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7208 ret = check_ref_cleanup(trans, buf->start);
7214 cache = btrfs_lookup_block_group(fs_info, buf->start);
7216 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7217 pin_down_extent(fs_info, cache, buf->start,
7219 btrfs_put_block_group(cache);
7223 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7225 btrfs_add_free_space(cache, buf->start, buf->len);
7226 btrfs_free_reserved_bytes(cache, buf->len, 0);
7227 btrfs_put_block_group(cache);
7228 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7232 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7233 root->root_key.objectid);
7237 * Deleting the buffer, clear the corrupt flag since it doesn't
7240 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7244 /* Can return -ENOMEM */
7245 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7246 struct btrfs_root *root,
7247 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7248 u64 owner, u64 offset)
7250 struct btrfs_fs_info *fs_info = root->fs_info;
7251 int old_ref_mod, new_ref_mod;
7254 if (btrfs_is_testing(fs_info))
7257 if (root_objectid != BTRFS_TREE_LOG_OBJECTID)
7258 btrfs_ref_tree_mod(root, bytenr, num_bytes, parent,
7259 root_objectid, owner, offset,
7260 BTRFS_DROP_DELAYED_REF);
7263 * tree log blocks never actually go into the extent allocation
7264 * tree, just update pinning info and exit early.
7266 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7267 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7268 /* unlocks the pinned mutex */
7269 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7270 old_ref_mod = new_ref_mod = 0;
7272 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7273 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7275 root_objectid, (int)owner,
7276 BTRFS_DROP_DELAYED_REF, NULL,
7277 &old_ref_mod, &new_ref_mod);
7279 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7281 root_objectid, owner, offset,
7282 0, BTRFS_DROP_DELAYED_REF,
7283 &old_ref_mod, &new_ref_mod);
7286 if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0)
7287 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7293 * when we wait for progress in the block group caching, its because
7294 * our allocation attempt failed at least once. So, we must sleep
7295 * and let some progress happen before we try again.
7297 * This function will sleep at least once waiting for new free space to
7298 * show up, and then it will check the block group free space numbers
7299 * for our min num_bytes. Another option is to have it go ahead
7300 * and look in the rbtree for a free extent of a given size, but this
7303 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7304 * any of the information in this block group.
7306 static noinline void
7307 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7310 struct btrfs_caching_control *caching_ctl;
7312 caching_ctl = get_caching_control(cache);
7316 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7317 (cache->free_space_ctl->free_space >= num_bytes));
7319 put_caching_control(caching_ctl);
7323 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7325 struct btrfs_caching_control *caching_ctl;
7328 caching_ctl = get_caching_control(cache);
7330 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7332 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7333 if (cache->cached == BTRFS_CACHE_ERROR)
7335 put_caching_control(caching_ctl);
7339 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7340 [BTRFS_RAID_RAID10] = "raid10",
7341 [BTRFS_RAID_RAID1] = "raid1",
7342 [BTRFS_RAID_DUP] = "dup",
7343 [BTRFS_RAID_RAID0] = "raid0",
7344 [BTRFS_RAID_SINGLE] = "single",
7345 [BTRFS_RAID_RAID5] = "raid5",
7346 [BTRFS_RAID_RAID6] = "raid6",
7349 static const char *get_raid_name(enum btrfs_raid_types type)
7351 if (type >= BTRFS_NR_RAID_TYPES)
7354 return btrfs_raid_type_names[type];
7357 enum btrfs_loop_type {
7358 LOOP_CACHING_NOWAIT = 0,
7359 LOOP_CACHING_WAIT = 1,
7360 LOOP_ALLOC_CHUNK = 2,
7361 LOOP_NO_EMPTY_SIZE = 3,
7365 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7369 down_read(&cache->data_rwsem);
7373 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7376 btrfs_get_block_group(cache);
7378 down_read(&cache->data_rwsem);
7381 static struct btrfs_block_group_cache *
7382 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7383 struct btrfs_free_cluster *cluster,
7386 struct btrfs_block_group_cache *used_bg = NULL;
7388 spin_lock(&cluster->refill_lock);
7390 used_bg = cluster->block_group;
7394 if (used_bg == block_group)
7397 btrfs_get_block_group(used_bg);
7402 if (down_read_trylock(&used_bg->data_rwsem))
7405 spin_unlock(&cluster->refill_lock);
7407 /* We should only have one-level nested. */
7408 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7410 spin_lock(&cluster->refill_lock);
7411 if (used_bg == cluster->block_group)
7414 up_read(&used_bg->data_rwsem);
7415 btrfs_put_block_group(used_bg);
7420 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7424 up_read(&cache->data_rwsem);
7425 btrfs_put_block_group(cache);
7429 * walks the btree of allocated extents and find a hole of a given size.
7430 * The key ins is changed to record the hole:
7431 * ins->objectid == start position
7432 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7433 * ins->offset == the size of the hole.
7434 * Any available blocks before search_start are skipped.
7436 * If there is no suitable free space, we will record the max size of
7437 * the free space extent currently.
7439 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7440 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7441 u64 hint_byte, struct btrfs_key *ins,
7442 u64 flags, int delalloc)
7445 struct btrfs_root *root = fs_info->extent_root;
7446 struct btrfs_free_cluster *last_ptr = NULL;
7447 struct btrfs_block_group_cache *block_group = NULL;
7448 u64 search_start = 0;
7449 u64 max_extent_size = 0;
7450 u64 empty_cluster = 0;
7451 struct btrfs_space_info *space_info;
7453 int index = btrfs_bg_flags_to_raid_index(flags);
7454 bool failed_cluster_refill = false;
7455 bool failed_alloc = false;
7456 bool use_cluster = true;
7457 bool have_caching_bg = false;
7458 bool orig_have_caching_bg = false;
7459 bool full_search = false;
7461 WARN_ON(num_bytes < fs_info->sectorsize);
7462 ins->type = BTRFS_EXTENT_ITEM_KEY;
7466 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7468 space_info = __find_space_info(fs_info, flags);
7470 btrfs_err(fs_info, "No space info for %llu", flags);
7475 * If our free space is heavily fragmented we may not be able to make
7476 * big contiguous allocations, so instead of doing the expensive search
7477 * for free space, simply return ENOSPC with our max_extent_size so we
7478 * can go ahead and search for a more manageable chunk.
7480 * If our max_extent_size is large enough for our allocation simply
7481 * disable clustering since we will likely not be able to find enough
7482 * space to create a cluster and induce latency trying.
7484 if (unlikely(space_info->max_extent_size)) {
7485 spin_lock(&space_info->lock);
7486 if (space_info->max_extent_size &&
7487 num_bytes > space_info->max_extent_size) {
7488 ins->offset = space_info->max_extent_size;
7489 spin_unlock(&space_info->lock);
7491 } else if (space_info->max_extent_size) {
7492 use_cluster = false;
7494 spin_unlock(&space_info->lock);
7497 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7499 spin_lock(&last_ptr->lock);
7500 if (last_ptr->block_group)
7501 hint_byte = last_ptr->window_start;
7502 if (last_ptr->fragmented) {
7504 * We still set window_start so we can keep track of the
7505 * last place we found an allocation to try and save
7508 hint_byte = last_ptr->window_start;
7509 use_cluster = false;
7511 spin_unlock(&last_ptr->lock);
7514 search_start = max(search_start, first_logical_byte(fs_info, 0));
7515 search_start = max(search_start, hint_byte);
7516 if (search_start == hint_byte) {
7517 block_group = btrfs_lookup_block_group(fs_info, search_start);
7519 * we don't want to use the block group if it doesn't match our
7520 * allocation bits, or if its not cached.
7522 * However if we are re-searching with an ideal block group
7523 * picked out then we don't care that the block group is cached.
7525 if (block_group && block_group_bits(block_group, flags) &&
7526 block_group->cached != BTRFS_CACHE_NO) {
7527 down_read(&space_info->groups_sem);
7528 if (list_empty(&block_group->list) ||
7531 * someone is removing this block group,
7532 * we can't jump into the have_block_group
7533 * target because our list pointers are not
7536 btrfs_put_block_group(block_group);
7537 up_read(&space_info->groups_sem);
7539 index = btrfs_bg_flags_to_raid_index(
7540 block_group->flags);
7541 btrfs_lock_block_group(block_group, delalloc);
7542 goto have_block_group;
7544 } else if (block_group) {
7545 btrfs_put_block_group(block_group);
7549 have_caching_bg = false;
7550 if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags))
7552 down_read(&space_info->groups_sem);
7553 list_for_each_entry(block_group, &space_info->block_groups[index],
7558 /* If the block group is read-only, we can skip it entirely. */
7559 if (unlikely(block_group->ro))
7562 btrfs_grab_block_group(block_group, delalloc);
7563 search_start = block_group->key.objectid;
7566 * this can happen if we end up cycling through all the
7567 * raid types, but we want to make sure we only allocate
7568 * for the proper type.
7570 if (!block_group_bits(block_group, flags)) {
7571 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7572 BTRFS_BLOCK_GROUP_RAID1 |
7573 BTRFS_BLOCK_GROUP_RAID5 |
7574 BTRFS_BLOCK_GROUP_RAID6 |
7575 BTRFS_BLOCK_GROUP_RAID10;
7578 * if they asked for extra copies and this block group
7579 * doesn't provide them, bail. This does allow us to
7580 * fill raid0 from raid1.
7582 if ((flags & extra) && !(block_group->flags & extra))
7587 cached = block_group_cache_done(block_group);
7588 if (unlikely(!cached)) {
7589 have_caching_bg = true;
7590 ret = cache_block_group(block_group, 0);
7595 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7599 * Ok we want to try and use the cluster allocator, so
7602 if (last_ptr && use_cluster) {
7603 struct btrfs_block_group_cache *used_block_group;
7604 unsigned long aligned_cluster;
7606 * the refill lock keeps out other
7607 * people trying to start a new cluster
7609 used_block_group = btrfs_lock_cluster(block_group,
7612 if (!used_block_group)
7613 goto refill_cluster;
7615 if (used_block_group != block_group &&
7616 (used_block_group->ro ||
7617 !block_group_bits(used_block_group, flags)))
7618 goto release_cluster;
7620 offset = btrfs_alloc_from_cluster(used_block_group,
7623 used_block_group->key.objectid,
7626 /* we have a block, we're done */
7627 spin_unlock(&last_ptr->refill_lock);
7628 trace_btrfs_reserve_extent_cluster(fs_info,
7630 search_start, num_bytes);
7631 if (used_block_group != block_group) {
7632 btrfs_release_block_group(block_group,
7634 block_group = used_block_group;
7639 WARN_ON(last_ptr->block_group != used_block_group);
7641 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7642 * set up a new clusters, so lets just skip it
7643 * and let the allocator find whatever block
7644 * it can find. If we reach this point, we
7645 * will have tried the cluster allocator
7646 * plenty of times and not have found
7647 * anything, so we are likely way too
7648 * fragmented for the clustering stuff to find
7651 * However, if the cluster is taken from the
7652 * current block group, release the cluster
7653 * first, so that we stand a better chance of
7654 * succeeding in the unclustered
7656 if (loop >= LOOP_NO_EMPTY_SIZE &&
7657 used_block_group != block_group) {
7658 spin_unlock(&last_ptr->refill_lock);
7659 btrfs_release_block_group(used_block_group,
7661 goto unclustered_alloc;
7665 * this cluster didn't work out, free it and
7668 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7670 if (used_block_group != block_group)
7671 btrfs_release_block_group(used_block_group,
7674 if (loop >= LOOP_NO_EMPTY_SIZE) {
7675 spin_unlock(&last_ptr->refill_lock);
7676 goto unclustered_alloc;
7679 aligned_cluster = max_t(unsigned long,
7680 empty_cluster + empty_size,
7681 block_group->full_stripe_len);
7683 /* allocate a cluster in this block group */
7684 ret = btrfs_find_space_cluster(fs_info, block_group,
7685 last_ptr, search_start,
7690 * now pull our allocation out of this
7693 offset = btrfs_alloc_from_cluster(block_group,
7699 /* we found one, proceed */
7700 spin_unlock(&last_ptr->refill_lock);
7701 trace_btrfs_reserve_extent_cluster(fs_info,
7702 block_group, search_start,
7706 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7707 && !failed_cluster_refill) {
7708 spin_unlock(&last_ptr->refill_lock);
7710 failed_cluster_refill = true;
7711 wait_block_group_cache_progress(block_group,
7712 num_bytes + empty_cluster + empty_size);
7713 goto have_block_group;
7717 * at this point we either didn't find a cluster
7718 * or we weren't able to allocate a block from our
7719 * cluster. Free the cluster we've been trying
7720 * to use, and go to the next block group
7722 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7723 spin_unlock(&last_ptr->refill_lock);
7729 * We are doing an unclustered alloc, set the fragmented flag so
7730 * we don't bother trying to setup a cluster again until we get
7733 if (unlikely(last_ptr)) {
7734 spin_lock(&last_ptr->lock);
7735 last_ptr->fragmented = 1;
7736 spin_unlock(&last_ptr->lock);
7739 struct btrfs_free_space_ctl *ctl =
7740 block_group->free_space_ctl;
7742 spin_lock(&ctl->tree_lock);
7743 if (ctl->free_space <
7744 num_bytes + empty_cluster + empty_size) {
7745 if (ctl->free_space > max_extent_size)
7746 max_extent_size = ctl->free_space;
7747 spin_unlock(&ctl->tree_lock);
7750 spin_unlock(&ctl->tree_lock);
7753 offset = btrfs_find_space_for_alloc(block_group, search_start,
7754 num_bytes, empty_size,
7757 * If we didn't find a chunk, and we haven't failed on this
7758 * block group before, and this block group is in the middle of
7759 * caching and we are ok with waiting, then go ahead and wait
7760 * for progress to be made, and set failed_alloc to true.
7762 * If failed_alloc is true then we've already waited on this
7763 * block group once and should move on to the next block group.
7765 if (!offset && !failed_alloc && !cached &&
7766 loop > LOOP_CACHING_NOWAIT) {
7767 wait_block_group_cache_progress(block_group,
7768 num_bytes + empty_size);
7769 failed_alloc = true;
7770 goto have_block_group;
7771 } else if (!offset) {
7775 search_start = ALIGN(offset, fs_info->stripesize);
7777 /* move on to the next group */
7778 if (search_start + num_bytes >
7779 block_group->key.objectid + block_group->key.offset) {
7780 btrfs_add_free_space(block_group, offset, num_bytes);
7784 if (offset < search_start)
7785 btrfs_add_free_space(block_group, offset,
7786 search_start - offset);
7787 BUG_ON(offset > search_start);
7789 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7790 num_bytes, delalloc);
7791 if (ret == -EAGAIN) {
7792 btrfs_add_free_space(block_group, offset, num_bytes);
7795 btrfs_inc_block_group_reservations(block_group);
7797 /* we are all good, lets return */
7798 ins->objectid = search_start;
7799 ins->offset = num_bytes;
7801 trace_btrfs_reserve_extent(fs_info, block_group,
7802 search_start, num_bytes);
7803 btrfs_release_block_group(block_group, delalloc);
7806 failed_cluster_refill = false;
7807 failed_alloc = false;
7808 BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) !=
7810 btrfs_release_block_group(block_group, delalloc);
7813 up_read(&space_info->groups_sem);
7815 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7816 && !orig_have_caching_bg)
7817 orig_have_caching_bg = true;
7819 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7822 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7826 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7827 * caching kthreads as we move along
7828 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7829 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7830 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7833 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7835 if (loop == LOOP_CACHING_NOWAIT) {
7837 * We want to skip the LOOP_CACHING_WAIT step if we
7838 * don't have any uncached bgs and we've already done a
7839 * full search through.
7841 if (orig_have_caching_bg || !full_search)
7842 loop = LOOP_CACHING_WAIT;
7844 loop = LOOP_ALLOC_CHUNK;
7849 if (loop == LOOP_ALLOC_CHUNK) {
7850 struct btrfs_trans_handle *trans;
7853 trans = current->journal_info;
7857 trans = btrfs_join_transaction(root);
7859 if (IS_ERR(trans)) {
7860 ret = PTR_ERR(trans);
7864 ret = do_chunk_alloc(trans, fs_info, flags,
7868 * If we can't allocate a new chunk we've already looped
7869 * through at least once, move on to the NO_EMPTY_SIZE
7873 loop = LOOP_NO_EMPTY_SIZE;
7876 * Do not bail out on ENOSPC since we
7877 * can do more things.
7879 if (ret < 0 && ret != -ENOSPC)
7880 btrfs_abort_transaction(trans, ret);
7884 btrfs_end_transaction(trans);
7889 if (loop == LOOP_NO_EMPTY_SIZE) {
7891 * Don't loop again if we already have no empty_size and
7894 if (empty_size == 0 &&
7895 empty_cluster == 0) {
7904 } else if (!ins->objectid) {
7906 } else if (ins->objectid) {
7907 if (!use_cluster && last_ptr) {
7908 spin_lock(&last_ptr->lock);
7909 last_ptr->window_start = ins->objectid;
7910 spin_unlock(&last_ptr->lock);
7915 if (ret == -ENOSPC) {
7916 spin_lock(&space_info->lock);
7917 space_info->max_extent_size = max_extent_size;
7918 spin_unlock(&space_info->lock);
7919 ins->offset = max_extent_size;
7924 static void dump_space_info(struct btrfs_fs_info *fs_info,
7925 struct btrfs_space_info *info, u64 bytes,
7926 int dump_block_groups)
7928 struct btrfs_block_group_cache *cache;
7931 spin_lock(&info->lock);
7932 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7934 info->total_bytes - btrfs_space_info_used(info, true),
7935 info->full ? "" : "not ");
7937 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7938 info->total_bytes, info->bytes_used, info->bytes_pinned,
7939 info->bytes_reserved, info->bytes_may_use,
7940 info->bytes_readonly);
7941 spin_unlock(&info->lock);
7943 if (!dump_block_groups)
7946 down_read(&info->groups_sem);
7948 list_for_each_entry(cache, &info->block_groups[index], list) {
7949 spin_lock(&cache->lock);
7951 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7952 cache->key.objectid, cache->key.offset,
7953 btrfs_block_group_used(&cache->item), cache->pinned,
7954 cache->reserved, cache->ro ? "[readonly]" : "");
7955 btrfs_dump_free_space(cache, bytes);
7956 spin_unlock(&cache->lock);
7958 if (++index < BTRFS_NR_RAID_TYPES)
7960 up_read(&info->groups_sem);
7964 * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a
7965 * hole that is at least as big as @num_bytes.
7967 * @root - The root that will contain this extent
7969 * @ram_bytes - The amount of space in ram that @num_bytes take. This
7970 * is used for accounting purposes. This value differs
7971 * from @num_bytes only in the case of compressed extents.
7973 * @num_bytes - Number of bytes to allocate on-disk.
7975 * @min_alloc_size - Indicates the minimum amount of space that the
7976 * allocator should try to satisfy. In some cases
7977 * @num_bytes may be larger than what is required and if
7978 * the filesystem is fragmented then allocation fails.
7979 * However, the presence of @min_alloc_size gives a
7980 * chance to try and satisfy the smaller allocation.
7982 * @empty_size - A hint that you plan on doing more COW. This is the
7983 * size in bytes the allocator should try to find free
7984 * next to the block it returns. This is just a hint and
7985 * may be ignored by the allocator.
7987 * @hint_byte - Hint to the allocator to start searching above the byte
7988 * address passed. It might be ignored.
7990 * @ins - This key is modified to record the found hole. It will
7991 * have the following values:
7992 * ins->objectid == start position
7993 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7994 * ins->offset == the size of the hole.
7996 * @is_data - Boolean flag indicating whether an extent is
7997 * allocated for data (true) or metadata (false)
7999 * @delalloc - Boolean flag indicating whether this allocation is for
8000 * delalloc or not. If 'true' data_rwsem of block groups
8001 * is going to be acquired.
8004 * Returns 0 when an allocation succeeded or < 0 when an error occurred. In
8005 * case -ENOSPC is returned then @ins->offset will contain the size of the
8006 * largest available hole the allocator managed to find.
8008 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
8009 u64 num_bytes, u64 min_alloc_size,
8010 u64 empty_size, u64 hint_byte,
8011 struct btrfs_key *ins, int is_data, int delalloc)
8013 struct btrfs_fs_info *fs_info = root->fs_info;
8014 bool final_tried = num_bytes == min_alloc_size;
8018 flags = get_alloc_profile_by_root(root, is_data);
8020 WARN_ON(num_bytes < fs_info->sectorsize);
8021 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
8022 hint_byte, ins, flags, delalloc);
8023 if (!ret && !is_data) {
8024 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
8025 } else if (ret == -ENOSPC) {
8026 if (!final_tried && ins->offset) {
8027 num_bytes = min(num_bytes >> 1, ins->offset);
8028 num_bytes = round_down(num_bytes,
8029 fs_info->sectorsize);
8030 num_bytes = max(num_bytes, min_alloc_size);
8031 ram_bytes = num_bytes;
8032 if (num_bytes == min_alloc_size)
8035 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8036 struct btrfs_space_info *sinfo;
8038 sinfo = __find_space_info(fs_info, flags);
8040 "allocation failed flags %llu, wanted %llu",
8043 dump_space_info(fs_info, sinfo, num_bytes, 1);
8050 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8052 int pin, int delalloc)
8054 struct btrfs_block_group_cache *cache;
8057 cache = btrfs_lookup_block_group(fs_info, start);
8059 btrfs_err(fs_info, "Unable to find block group for %llu",
8065 pin_down_extent(fs_info, cache, start, len, 1);
8067 if (btrfs_test_opt(fs_info, DISCARD))
8068 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8069 btrfs_add_free_space(cache, start, len);
8070 btrfs_free_reserved_bytes(cache, len, delalloc);
8071 trace_btrfs_reserved_extent_free(fs_info, start, len);
8074 btrfs_put_block_group(cache);
8078 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8079 u64 start, u64 len, int delalloc)
8081 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8084 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8087 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8090 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8091 struct btrfs_fs_info *fs_info,
8092 u64 parent, u64 root_objectid,
8093 u64 flags, u64 owner, u64 offset,
8094 struct btrfs_key *ins, int ref_mod)
8097 struct btrfs_extent_item *extent_item;
8098 struct btrfs_extent_inline_ref *iref;
8099 struct btrfs_path *path;
8100 struct extent_buffer *leaf;
8105 type = BTRFS_SHARED_DATA_REF_KEY;
8107 type = BTRFS_EXTENT_DATA_REF_KEY;
8109 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8111 path = btrfs_alloc_path();
8115 path->leave_spinning = 1;
8116 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8119 btrfs_free_path(path);
8123 leaf = path->nodes[0];
8124 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8125 struct btrfs_extent_item);
8126 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8127 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8128 btrfs_set_extent_flags(leaf, extent_item,
8129 flags | BTRFS_EXTENT_FLAG_DATA);
8131 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8132 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8134 struct btrfs_shared_data_ref *ref;
8135 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8136 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8137 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8139 struct btrfs_extent_data_ref *ref;
8140 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8141 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8142 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8143 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8144 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8147 btrfs_mark_buffer_dirty(path->nodes[0]);
8148 btrfs_free_path(path);
8150 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8155 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8156 if (ret) { /* -ENOENT, logic error */
8157 btrfs_err(fs_info, "update block group failed for %llu %llu",
8158 ins->objectid, ins->offset);
8161 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8165 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8166 struct btrfs_fs_info *fs_info,
8167 u64 parent, u64 root_objectid,
8168 u64 flags, struct btrfs_disk_key *key,
8169 int level, struct btrfs_key *ins)
8172 struct btrfs_extent_item *extent_item;
8173 struct btrfs_tree_block_info *block_info;
8174 struct btrfs_extent_inline_ref *iref;
8175 struct btrfs_path *path;
8176 struct extent_buffer *leaf;
8177 u32 size = sizeof(*extent_item) + sizeof(*iref);
8178 u64 num_bytes = ins->offset;
8179 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8181 if (!skinny_metadata)
8182 size += sizeof(*block_info);
8184 path = btrfs_alloc_path();
8186 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8191 path->leave_spinning = 1;
8192 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8195 btrfs_free_path(path);
8196 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8201 leaf = path->nodes[0];
8202 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8203 struct btrfs_extent_item);
8204 btrfs_set_extent_refs(leaf, extent_item, 1);
8205 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8206 btrfs_set_extent_flags(leaf, extent_item,
8207 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8209 if (skinny_metadata) {
8210 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8211 num_bytes = fs_info->nodesize;
8213 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8214 btrfs_set_tree_block_key(leaf, block_info, key);
8215 btrfs_set_tree_block_level(leaf, block_info, level);
8216 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8220 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8221 btrfs_set_extent_inline_ref_type(leaf, iref,
8222 BTRFS_SHARED_BLOCK_REF_KEY);
8223 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8225 btrfs_set_extent_inline_ref_type(leaf, iref,
8226 BTRFS_TREE_BLOCK_REF_KEY);
8227 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8230 btrfs_mark_buffer_dirty(leaf);
8231 btrfs_free_path(path);
8233 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8238 ret = update_block_group(trans, fs_info, ins->objectid,
8239 fs_info->nodesize, 1);
8240 if (ret) { /* -ENOENT, logic error */
8241 btrfs_err(fs_info, "update block group failed for %llu %llu",
8242 ins->objectid, ins->offset);
8246 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8251 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8252 struct btrfs_root *root, u64 owner,
8253 u64 offset, u64 ram_bytes,
8254 struct btrfs_key *ins)
8256 struct btrfs_fs_info *fs_info = root->fs_info;
8259 BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID);
8261 btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0,
8262 root->root_key.objectid, owner, offset,
8263 BTRFS_ADD_DELAYED_EXTENT);
8265 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8267 root->root_key.objectid, owner,
8269 BTRFS_ADD_DELAYED_EXTENT, NULL, NULL);
8274 * this is used by the tree logging recovery code. It records that
8275 * an extent has been allocated and makes sure to clear the free
8276 * space cache bits as well
8278 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8279 struct btrfs_fs_info *fs_info,
8280 u64 root_objectid, u64 owner, u64 offset,
8281 struct btrfs_key *ins)
8284 struct btrfs_block_group_cache *block_group;
8285 struct btrfs_space_info *space_info;
8288 * Mixed block groups will exclude before processing the log so we only
8289 * need to do the exclude dance if this fs isn't mixed.
8291 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8292 ret = __exclude_logged_extent(fs_info, ins->objectid,
8298 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8302 space_info = block_group->space_info;
8303 spin_lock(&space_info->lock);
8304 spin_lock(&block_group->lock);
8305 space_info->bytes_reserved += ins->offset;
8306 block_group->reserved += ins->offset;
8307 spin_unlock(&block_group->lock);
8308 spin_unlock(&space_info->lock);
8310 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8311 0, owner, offset, ins, 1);
8312 btrfs_put_block_group(block_group);
8316 static struct extent_buffer *
8317 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8318 u64 bytenr, int level)
8320 struct btrfs_fs_info *fs_info = root->fs_info;
8321 struct extent_buffer *buf;
8323 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8327 btrfs_set_header_generation(buf, trans->transid);
8328 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8329 btrfs_tree_lock(buf);
8330 clean_tree_block(fs_info, buf);
8331 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8333 btrfs_set_lock_blocking(buf);
8334 set_extent_buffer_uptodate(buf);
8336 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8337 buf->log_index = root->log_transid % 2;
8339 * we allow two log transactions at a time, use different
8340 * EXENT bit to differentiate dirty pages.
8342 if (buf->log_index == 0)
8343 set_extent_dirty(&root->dirty_log_pages, buf->start,
8344 buf->start + buf->len - 1, GFP_NOFS);
8346 set_extent_new(&root->dirty_log_pages, buf->start,
8347 buf->start + buf->len - 1);
8349 buf->log_index = -1;
8350 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8351 buf->start + buf->len - 1, GFP_NOFS);
8353 trans->dirty = true;
8354 /* this returns a buffer locked for blocking */
8358 static struct btrfs_block_rsv *
8359 use_block_rsv(struct btrfs_trans_handle *trans,
8360 struct btrfs_root *root, u32 blocksize)
8362 struct btrfs_fs_info *fs_info = root->fs_info;
8363 struct btrfs_block_rsv *block_rsv;
8364 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8366 bool global_updated = false;
8368 block_rsv = get_block_rsv(trans, root);
8370 if (unlikely(block_rsv->size == 0))
8373 ret = block_rsv_use_bytes(block_rsv, blocksize);
8377 if (block_rsv->failfast)
8378 return ERR_PTR(ret);
8380 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8381 global_updated = true;
8382 update_global_block_rsv(fs_info);
8386 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8387 static DEFINE_RATELIMIT_STATE(_rs,
8388 DEFAULT_RATELIMIT_INTERVAL * 10,
8389 /*DEFAULT_RATELIMIT_BURST*/ 1);
8390 if (__ratelimit(&_rs))
8392 "BTRFS: block rsv returned %d\n", ret);
8395 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8396 BTRFS_RESERVE_NO_FLUSH);
8400 * If we couldn't reserve metadata bytes try and use some from
8401 * the global reserve if its space type is the same as the global
8404 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8405 block_rsv->space_info == global_rsv->space_info) {
8406 ret = block_rsv_use_bytes(global_rsv, blocksize);
8410 return ERR_PTR(ret);
8413 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8414 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8416 block_rsv_add_bytes(block_rsv, blocksize, 0);
8417 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8421 * finds a free extent and does all the dirty work required for allocation
8422 * returns the tree buffer or an ERR_PTR on error.
8424 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8425 struct btrfs_root *root,
8426 u64 parent, u64 root_objectid,
8427 const struct btrfs_disk_key *key,
8428 int level, u64 hint,
8431 struct btrfs_fs_info *fs_info = root->fs_info;
8432 struct btrfs_key ins;
8433 struct btrfs_block_rsv *block_rsv;
8434 struct extent_buffer *buf;
8435 struct btrfs_delayed_extent_op *extent_op;
8438 u32 blocksize = fs_info->nodesize;
8439 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8441 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8442 if (btrfs_is_testing(fs_info)) {
8443 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8446 root->alloc_bytenr += blocksize;
8451 block_rsv = use_block_rsv(trans, root, blocksize);
8452 if (IS_ERR(block_rsv))
8453 return ERR_CAST(block_rsv);
8455 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8456 empty_size, hint, &ins, 0, 0);
8460 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8463 goto out_free_reserved;
8466 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8468 parent = ins.objectid;
8469 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8473 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8474 extent_op = btrfs_alloc_delayed_extent_op();
8480 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8482 memset(&extent_op->key, 0, sizeof(extent_op->key));
8483 extent_op->flags_to_set = flags;
8484 extent_op->update_key = skinny_metadata ? false : true;
8485 extent_op->update_flags = true;
8486 extent_op->is_data = false;
8487 extent_op->level = level;
8489 btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent,
8490 root_objectid, level, 0,
8491 BTRFS_ADD_DELAYED_EXTENT);
8492 ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid,
8494 root_objectid, level,
8495 BTRFS_ADD_DELAYED_EXTENT,
8496 extent_op, NULL, NULL);
8498 goto out_free_delayed;
8503 btrfs_free_delayed_extent_op(extent_op);
8505 free_extent_buffer(buf);
8507 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8509 unuse_block_rsv(fs_info, block_rsv, blocksize);
8510 return ERR_PTR(ret);
8513 struct walk_control {
8514 u64 refs[BTRFS_MAX_LEVEL];
8515 u64 flags[BTRFS_MAX_LEVEL];
8516 struct btrfs_key update_progress;
8527 #define DROP_REFERENCE 1
8528 #define UPDATE_BACKREF 2
8530 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8531 struct btrfs_root *root,
8532 struct walk_control *wc,
8533 struct btrfs_path *path)
8535 struct btrfs_fs_info *fs_info = root->fs_info;
8541 struct btrfs_key key;
8542 struct extent_buffer *eb;
8547 if (path->slots[wc->level] < wc->reada_slot) {
8548 wc->reada_count = wc->reada_count * 2 / 3;
8549 wc->reada_count = max(wc->reada_count, 2);
8551 wc->reada_count = wc->reada_count * 3 / 2;
8552 wc->reada_count = min_t(int, wc->reada_count,
8553 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8556 eb = path->nodes[wc->level];
8557 nritems = btrfs_header_nritems(eb);
8559 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8560 if (nread >= wc->reada_count)
8564 bytenr = btrfs_node_blockptr(eb, slot);
8565 generation = btrfs_node_ptr_generation(eb, slot);
8567 if (slot == path->slots[wc->level])
8570 if (wc->stage == UPDATE_BACKREF &&
8571 generation <= root->root_key.offset)
8574 /* We don't lock the tree block, it's OK to be racy here */
8575 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8576 wc->level - 1, 1, &refs,
8578 /* We don't care about errors in readahead. */
8583 if (wc->stage == DROP_REFERENCE) {
8587 if (wc->level == 1 &&
8588 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8590 if (!wc->update_ref ||
8591 generation <= root->root_key.offset)
8593 btrfs_node_key_to_cpu(eb, &key, slot);
8594 ret = btrfs_comp_cpu_keys(&key,
8595 &wc->update_progress);
8599 if (wc->level == 1 &&
8600 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8604 readahead_tree_block(fs_info, bytenr);
8607 wc->reada_slot = slot;
8611 * helper to process tree block while walking down the tree.
8613 * when wc->stage == UPDATE_BACKREF, this function updates
8614 * back refs for pointers in the block.
8616 * NOTE: return value 1 means we should stop walking down.
8618 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8619 struct btrfs_root *root,
8620 struct btrfs_path *path,
8621 struct walk_control *wc, int lookup_info)
8623 struct btrfs_fs_info *fs_info = root->fs_info;
8624 int level = wc->level;
8625 struct extent_buffer *eb = path->nodes[level];
8626 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8629 if (wc->stage == UPDATE_BACKREF &&
8630 btrfs_header_owner(eb) != root->root_key.objectid)
8634 * when reference count of tree block is 1, it won't increase
8635 * again. once full backref flag is set, we never clear it.
8638 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8639 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8640 BUG_ON(!path->locks[level]);
8641 ret = btrfs_lookup_extent_info(trans, fs_info,
8642 eb->start, level, 1,
8645 BUG_ON(ret == -ENOMEM);
8648 BUG_ON(wc->refs[level] == 0);
8651 if (wc->stage == DROP_REFERENCE) {
8652 if (wc->refs[level] > 1)
8655 if (path->locks[level] && !wc->keep_locks) {
8656 btrfs_tree_unlock_rw(eb, path->locks[level]);
8657 path->locks[level] = 0;
8662 /* wc->stage == UPDATE_BACKREF */
8663 if (!(wc->flags[level] & flag)) {
8664 BUG_ON(!path->locks[level]);
8665 ret = btrfs_inc_ref(trans, root, eb, 1);
8666 BUG_ON(ret); /* -ENOMEM */
8667 ret = btrfs_dec_ref(trans, root, eb, 0);
8668 BUG_ON(ret); /* -ENOMEM */
8669 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8671 btrfs_header_level(eb), 0);
8672 BUG_ON(ret); /* -ENOMEM */
8673 wc->flags[level] |= flag;
8677 * the block is shared by multiple trees, so it's not good to
8678 * keep the tree lock
8680 if (path->locks[level] && level > 0) {
8681 btrfs_tree_unlock_rw(eb, path->locks[level]);
8682 path->locks[level] = 0;
8688 * helper to process tree block pointer.
8690 * when wc->stage == DROP_REFERENCE, this function checks
8691 * reference count of the block pointed to. if the block
8692 * is shared and we need update back refs for the subtree
8693 * rooted at the block, this function changes wc->stage to
8694 * UPDATE_BACKREF. if the block is shared and there is no
8695 * need to update back, this function drops the reference
8698 * NOTE: return value 1 means we should stop walking down.
8700 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8701 struct btrfs_root *root,
8702 struct btrfs_path *path,
8703 struct walk_control *wc, int *lookup_info)
8705 struct btrfs_fs_info *fs_info = root->fs_info;
8710 struct btrfs_key key;
8711 struct btrfs_key first_key;
8712 struct extent_buffer *next;
8713 int level = wc->level;
8716 bool need_account = false;
8718 generation = btrfs_node_ptr_generation(path->nodes[level],
8719 path->slots[level]);
8721 * if the lower level block was created before the snapshot
8722 * was created, we know there is no need to update back refs
8725 if (wc->stage == UPDATE_BACKREF &&
8726 generation <= root->root_key.offset) {
8731 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8732 btrfs_node_key_to_cpu(path->nodes[level], &first_key,
8733 path->slots[level]);
8734 blocksize = fs_info->nodesize;
8736 next = find_extent_buffer(fs_info, bytenr);
8738 next = btrfs_find_create_tree_block(fs_info, bytenr);
8740 return PTR_ERR(next);
8742 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8746 btrfs_tree_lock(next);
8747 btrfs_set_lock_blocking(next);
8749 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8750 &wc->refs[level - 1],
8751 &wc->flags[level - 1]);
8755 if (unlikely(wc->refs[level - 1] == 0)) {
8756 btrfs_err(fs_info, "Missing references.");
8762 if (wc->stage == DROP_REFERENCE) {
8763 if (wc->refs[level - 1] > 1) {
8764 need_account = true;
8766 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8769 if (!wc->update_ref ||
8770 generation <= root->root_key.offset)
8773 btrfs_node_key_to_cpu(path->nodes[level], &key,
8774 path->slots[level]);
8775 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8779 wc->stage = UPDATE_BACKREF;
8780 wc->shared_level = level - 1;
8784 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8788 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8789 btrfs_tree_unlock(next);
8790 free_extent_buffer(next);
8796 if (reada && level == 1)
8797 reada_walk_down(trans, root, wc, path);
8798 next = read_tree_block(fs_info, bytenr, generation, level - 1,
8801 return PTR_ERR(next);
8802 } else if (!extent_buffer_uptodate(next)) {
8803 free_extent_buffer(next);
8806 btrfs_tree_lock(next);
8807 btrfs_set_lock_blocking(next);
8811 ASSERT(level == btrfs_header_level(next));
8812 if (level != btrfs_header_level(next)) {
8813 btrfs_err(root->fs_info, "mismatched level");
8817 path->nodes[level] = next;
8818 path->slots[level] = 0;
8819 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8825 wc->refs[level - 1] = 0;
8826 wc->flags[level - 1] = 0;
8827 if (wc->stage == DROP_REFERENCE) {
8828 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8829 parent = path->nodes[level]->start;
8831 ASSERT(root->root_key.objectid ==
8832 btrfs_header_owner(path->nodes[level]));
8833 if (root->root_key.objectid !=
8834 btrfs_header_owner(path->nodes[level])) {
8835 btrfs_err(root->fs_info,
8836 "mismatched block owner");
8844 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8845 generation, level - 1);
8847 btrfs_err_rl(fs_info,
8848 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8852 ret = btrfs_free_extent(trans, root, bytenr, blocksize,
8853 parent, root->root_key.objectid,
8863 btrfs_tree_unlock(next);
8864 free_extent_buffer(next);
8870 * helper to process tree block while walking up the tree.
8872 * when wc->stage == DROP_REFERENCE, this function drops
8873 * reference count on the block.
8875 * when wc->stage == UPDATE_BACKREF, this function changes
8876 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8877 * to UPDATE_BACKREF previously while processing the block.
8879 * NOTE: return value 1 means we should stop walking up.
8881 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8882 struct btrfs_root *root,
8883 struct btrfs_path *path,
8884 struct walk_control *wc)
8886 struct btrfs_fs_info *fs_info = root->fs_info;
8888 int level = wc->level;
8889 struct extent_buffer *eb = path->nodes[level];
8892 if (wc->stage == UPDATE_BACKREF) {
8893 BUG_ON(wc->shared_level < level);
8894 if (level < wc->shared_level)
8897 ret = find_next_key(path, level + 1, &wc->update_progress);
8901 wc->stage = DROP_REFERENCE;
8902 wc->shared_level = -1;
8903 path->slots[level] = 0;
8906 * check reference count again if the block isn't locked.
8907 * we should start walking down the tree again if reference
8910 if (!path->locks[level]) {
8912 btrfs_tree_lock(eb);
8913 btrfs_set_lock_blocking(eb);
8914 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8916 ret = btrfs_lookup_extent_info(trans, fs_info,
8917 eb->start, level, 1,
8921 btrfs_tree_unlock_rw(eb, path->locks[level]);
8922 path->locks[level] = 0;
8925 BUG_ON(wc->refs[level] == 0);
8926 if (wc->refs[level] == 1) {
8927 btrfs_tree_unlock_rw(eb, path->locks[level]);
8928 path->locks[level] = 0;
8934 /* wc->stage == DROP_REFERENCE */
8935 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8937 if (wc->refs[level] == 1) {
8939 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8940 ret = btrfs_dec_ref(trans, root, eb, 1);
8942 ret = btrfs_dec_ref(trans, root, eb, 0);
8943 BUG_ON(ret); /* -ENOMEM */
8944 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8946 btrfs_err_rl(fs_info,
8947 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8951 /* make block locked assertion in clean_tree_block happy */
8952 if (!path->locks[level] &&
8953 btrfs_header_generation(eb) == trans->transid) {
8954 btrfs_tree_lock(eb);
8955 btrfs_set_lock_blocking(eb);
8956 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8958 clean_tree_block(fs_info, eb);
8961 if (eb == root->node) {
8962 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8965 BUG_ON(root->root_key.objectid !=
8966 btrfs_header_owner(eb));
8968 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8969 parent = path->nodes[level + 1]->start;
8971 BUG_ON(root->root_key.objectid !=
8972 btrfs_header_owner(path->nodes[level + 1]));
8975 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8977 wc->refs[level] = 0;
8978 wc->flags[level] = 0;
8982 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8983 struct btrfs_root *root,
8984 struct btrfs_path *path,
8985 struct walk_control *wc)
8987 int level = wc->level;
8988 int lookup_info = 1;
8991 while (level >= 0) {
8992 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8999 if (path->slots[level] >=
9000 btrfs_header_nritems(path->nodes[level]))
9003 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9005 path->slots[level]++;
9014 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9015 struct btrfs_root *root,
9016 struct btrfs_path *path,
9017 struct walk_control *wc, int max_level)
9019 int level = wc->level;
9022 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9023 while (level < max_level && path->nodes[level]) {
9025 if (path->slots[level] + 1 <
9026 btrfs_header_nritems(path->nodes[level])) {
9027 path->slots[level]++;
9030 ret = walk_up_proc(trans, root, path, wc);
9034 if (path->locks[level]) {
9035 btrfs_tree_unlock_rw(path->nodes[level],
9036 path->locks[level]);
9037 path->locks[level] = 0;
9039 free_extent_buffer(path->nodes[level]);
9040 path->nodes[level] = NULL;
9048 * drop a subvolume tree.
9050 * this function traverses the tree freeing any blocks that only
9051 * referenced by the tree.
9053 * when a shared tree block is found. this function decreases its
9054 * reference count by one. if update_ref is true, this function
9055 * also make sure backrefs for the shared block and all lower level
9056 * blocks are properly updated.
9058 * If called with for_reloc == 0, may exit early with -EAGAIN
9060 int btrfs_drop_snapshot(struct btrfs_root *root,
9061 struct btrfs_block_rsv *block_rsv, int update_ref,
9064 struct btrfs_fs_info *fs_info = root->fs_info;
9065 struct btrfs_path *path;
9066 struct btrfs_trans_handle *trans;
9067 struct btrfs_root *tree_root = fs_info->tree_root;
9068 struct btrfs_root_item *root_item = &root->root_item;
9069 struct walk_control *wc;
9070 struct btrfs_key key;
9074 bool root_dropped = false;
9076 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9078 path = btrfs_alloc_path();
9084 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9086 btrfs_free_path(path);
9091 trans = btrfs_start_transaction(tree_root, 0);
9092 if (IS_ERR(trans)) {
9093 err = PTR_ERR(trans);
9098 trans->block_rsv = block_rsv;
9100 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9101 level = btrfs_header_level(root->node);
9102 path->nodes[level] = btrfs_lock_root_node(root);
9103 btrfs_set_lock_blocking(path->nodes[level]);
9104 path->slots[level] = 0;
9105 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9106 memset(&wc->update_progress, 0,
9107 sizeof(wc->update_progress));
9109 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9110 memcpy(&wc->update_progress, &key,
9111 sizeof(wc->update_progress));
9113 level = root_item->drop_level;
9115 path->lowest_level = level;
9116 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9117 path->lowest_level = 0;
9125 * unlock our path, this is safe because only this
9126 * function is allowed to delete this snapshot
9128 btrfs_unlock_up_safe(path, 0);
9130 level = btrfs_header_level(root->node);
9132 btrfs_tree_lock(path->nodes[level]);
9133 btrfs_set_lock_blocking(path->nodes[level]);
9134 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9136 ret = btrfs_lookup_extent_info(trans, fs_info,
9137 path->nodes[level]->start,
9138 level, 1, &wc->refs[level],
9144 BUG_ON(wc->refs[level] == 0);
9146 if (level == root_item->drop_level)
9149 btrfs_tree_unlock(path->nodes[level]);
9150 path->locks[level] = 0;
9151 WARN_ON(wc->refs[level] != 1);
9157 wc->shared_level = -1;
9158 wc->stage = DROP_REFERENCE;
9159 wc->update_ref = update_ref;
9161 wc->for_reloc = for_reloc;
9162 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9166 ret = walk_down_tree(trans, root, path, wc);
9172 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9179 BUG_ON(wc->stage != DROP_REFERENCE);
9183 if (wc->stage == DROP_REFERENCE) {
9185 btrfs_node_key(path->nodes[level],
9186 &root_item->drop_progress,
9187 path->slots[level]);
9188 root_item->drop_level = level;
9191 BUG_ON(wc->level == 0);
9192 if (btrfs_should_end_transaction(trans) ||
9193 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9194 ret = btrfs_update_root(trans, tree_root,
9198 btrfs_abort_transaction(trans, ret);
9203 btrfs_end_transaction_throttle(trans);
9204 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9205 btrfs_debug(fs_info,
9206 "drop snapshot early exit");
9211 trans = btrfs_start_transaction(tree_root, 0);
9212 if (IS_ERR(trans)) {
9213 err = PTR_ERR(trans);
9217 trans->block_rsv = block_rsv;
9220 btrfs_release_path(path);
9224 ret = btrfs_del_root(trans, fs_info, &root->root_key);
9226 btrfs_abort_transaction(trans, ret);
9231 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9232 ret = btrfs_find_root(tree_root, &root->root_key, path,
9235 btrfs_abort_transaction(trans, ret);
9238 } else if (ret > 0) {
9239 /* if we fail to delete the orphan item this time
9240 * around, it'll get picked up the next time.
9242 * The most common failure here is just -ENOENT.
9244 btrfs_del_orphan_item(trans, tree_root,
9245 root->root_key.objectid);
9249 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9250 btrfs_add_dropped_root(trans, root);
9252 free_extent_buffer(root->node);
9253 free_extent_buffer(root->commit_root);
9254 btrfs_put_fs_root(root);
9256 root_dropped = true;
9258 btrfs_end_transaction_throttle(trans);
9261 btrfs_free_path(path);
9264 * So if we need to stop dropping the snapshot for whatever reason we
9265 * need to make sure to add it back to the dead root list so that we
9266 * keep trying to do the work later. This also cleans up roots if we
9267 * don't have it in the radix (like when we recover after a power fail
9268 * or unmount) so we don't leak memory.
9270 if (!for_reloc && !root_dropped)
9271 btrfs_add_dead_root(root);
9272 if (err && err != -EAGAIN)
9273 btrfs_handle_fs_error(fs_info, err, NULL);
9278 * drop subtree rooted at tree block 'node'.
9280 * NOTE: this function will unlock and release tree block 'node'
9281 * only used by relocation code
9283 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9284 struct btrfs_root *root,
9285 struct extent_buffer *node,
9286 struct extent_buffer *parent)
9288 struct btrfs_fs_info *fs_info = root->fs_info;
9289 struct btrfs_path *path;
9290 struct walk_control *wc;
9296 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9298 path = btrfs_alloc_path();
9302 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9304 btrfs_free_path(path);
9308 btrfs_assert_tree_locked(parent);
9309 parent_level = btrfs_header_level(parent);
9310 extent_buffer_get(parent);
9311 path->nodes[parent_level] = parent;
9312 path->slots[parent_level] = btrfs_header_nritems(parent);
9314 btrfs_assert_tree_locked(node);
9315 level = btrfs_header_level(node);
9316 path->nodes[level] = node;
9317 path->slots[level] = 0;
9318 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9320 wc->refs[parent_level] = 1;
9321 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9323 wc->shared_level = -1;
9324 wc->stage = DROP_REFERENCE;
9328 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9331 wret = walk_down_tree(trans, root, path, wc);
9337 wret = walk_up_tree(trans, root, path, wc, parent_level);
9345 btrfs_free_path(path);
9349 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9355 * if restripe for this chunk_type is on pick target profile and
9356 * return, otherwise do the usual balance
9358 stripped = get_restripe_target(fs_info, flags);
9360 return extended_to_chunk(stripped);
9362 num_devices = fs_info->fs_devices->rw_devices;
9364 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9365 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9366 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9368 if (num_devices == 1) {
9369 stripped |= BTRFS_BLOCK_GROUP_DUP;
9370 stripped = flags & ~stripped;
9372 /* turn raid0 into single device chunks */
9373 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9376 /* turn mirroring into duplication */
9377 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9378 BTRFS_BLOCK_GROUP_RAID10))
9379 return stripped | BTRFS_BLOCK_GROUP_DUP;
9381 /* they already had raid on here, just return */
9382 if (flags & stripped)
9385 stripped |= BTRFS_BLOCK_GROUP_DUP;
9386 stripped = flags & ~stripped;
9388 /* switch duplicated blocks with raid1 */
9389 if (flags & BTRFS_BLOCK_GROUP_DUP)
9390 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9392 /* this is drive concat, leave it alone */
9398 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9400 struct btrfs_space_info *sinfo = cache->space_info;
9402 u64 min_allocable_bytes;
9406 * We need some metadata space and system metadata space for
9407 * allocating chunks in some corner cases until we force to set
9408 * it to be readonly.
9411 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9413 min_allocable_bytes = SZ_1M;
9415 min_allocable_bytes = 0;
9417 spin_lock(&sinfo->lock);
9418 spin_lock(&cache->lock);
9426 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9427 cache->bytes_super - btrfs_block_group_used(&cache->item);
9429 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9430 min_allocable_bytes <= sinfo->total_bytes) {
9431 sinfo->bytes_readonly += num_bytes;
9433 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9437 spin_unlock(&cache->lock);
9438 spin_unlock(&sinfo->lock);
9442 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9443 struct btrfs_block_group_cache *cache)
9446 struct btrfs_trans_handle *trans;
9451 trans = btrfs_join_transaction(fs_info->extent_root);
9453 return PTR_ERR(trans);
9456 * we're not allowed to set block groups readonly after the dirty
9457 * block groups cache has started writing. If it already started,
9458 * back off and let this transaction commit
9460 mutex_lock(&fs_info->ro_block_group_mutex);
9461 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9462 u64 transid = trans->transid;
9464 mutex_unlock(&fs_info->ro_block_group_mutex);
9465 btrfs_end_transaction(trans);
9467 ret = btrfs_wait_for_commit(fs_info, transid);
9474 * if we are changing raid levels, try to allocate a corresponding
9475 * block group with the new raid level.
9477 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9478 if (alloc_flags != cache->flags) {
9479 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9482 * ENOSPC is allowed here, we may have enough space
9483 * already allocated at the new raid level to
9492 ret = inc_block_group_ro(cache, 0);
9495 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9496 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9500 ret = inc_block_group_ro(cache, 0);
9502 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9503 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9504 mutex_lock(&fs_info->chunk_mutex);
9505 check_system_chunk(trans, fs_info, alloc_flags);
9506 mutex_unlock(&fs_info->chunk_mutex);
9508 mutex_unlock(&fs_info->ro_block_group_mutex);
9510 btrfs_end_transaction(trans);
9514 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9515 struct btrfs_fs_info *fs_info, u64 type)
9517 u64 alloc_flags = get_alloc_profile(fs_info, type);
9519 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9523 * helper to account the unused space of all the readonly block group in the
9524 * space_info. takes mirrors into account.
9526 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9528 struct btrfs_block_group_cache *block_group;
9532 /* It's df, we don't care if it's racy */
9533 if (list_empty(&sinfo->ro_bgs))
9536 spin_lock(&sinfo->lock);
9537 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9538 spin_lock(&block_group->lock);
9540 if (!block_group->ro) {
9541 spin_unlock(&block_group->lock);
9545 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9546 BTRFS_BLOCK_GROUP_RAID10 |
9547 BTRFS_BLOCK_GROUP_DUP))
9552 free_bytes += (block_group->key.offset -
9553 btrfs_block_group_used(&block_group->item)) *
9556 spin_unlock(&block_group->lock);
9558 spin_unlock(&sinfo->lock);
9563 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9565 struct btrfs_space_info *sinfo = cache->space_info;
9570 spin_lock(&sinfo->lock);
9571 spin_lock(&cache->lock);
9573 num_bytes = cache->key.offset - cache->reserved -
9574 cache->pinned - cache->bytes_super -
9575 btrfs_block_group_used(&cache->item);
9576 sinfo->bytes_readonly -= num_bytes;
9577 list_del_init(&cache->ro_list);
9579 spin_unlock(&cache->lock);
9580 spin_unlock(&sinfo->lock);
9584 * checks to see if its even possible to relocate this block group.
9586 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9587 * ok to go ahead and try.
9589 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9591 struct btrfs_root *root = fs_info->extent_root;
9592 struct btrfs_block_group_cache *block_group;
9593 struct btrfs_space_info *space_info;
9594 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9595 struct btrfs_device *device;
9596 struct btrfs_trans_handle *trans;
9606 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9608 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9610 /* odd, couldn't find the block group, leave it alone */
9614 "can't find block group for bytenr %llu",
9619 min_free = btrfs_block_group_used(&block_group->item);
9621 /* no bytes used, we're good */
9625 space_info = block_group->space_info;
9626 spin_lock(&space_info->lock);
9628 full = space_info->full;
9631 * if this is the last block group we have in this space, we can't
9632 * relocate it unless we're able to allocate a new chunk below.
9634 * Otherwise, we need to make sure we have room in the space to handle
9635 * all of the extents from this block group. If we can, we're good
9637 if ((space_info->total_bytes != block_group->key.offset) &&
9638 (btrfs_space_info_used(space_info, false) + min_free <
9639 space_info->total_bytes)) {
9640 spin_unlock(&space_info->lock);
9643 spin_unlock(&space_info->lock);
9646 * ok we don't have enough space, but maybe we have free space on our
9647 * devices to allocate new chunks for relocation, so loop through our
9648 * alloc devices and guess if we have enough space. if this block
9649 * group is going to be restriped, run checks against the target
9650 * profile instead of the current one.
9662 target = get_restripe_target(fs_info, block_group->flags);
9664 index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target));
9667 * this is just a balance, so if we were marked as full
9668 * we know there is no space for a new chunk
9673 "no space to alloc new chunk for block group %llu",
9674 block_group->key.objectid);
9678 index = btrfs_bg_flags_to_raid_index(block_group->flags);
9681 if (index == BTRFS_RAID_RAID10) {
9685 } else if (index == BTRFS_RAID_RAID1) {
9687 } else if (index == BTRFS_RAID_DUP) {
9690 } else if (index == BTRFS_RAID_RAID0) {
9691 dev_min = fs_devices->rw_devices;
9692 min_free = div64_u64(min_free, dev_min);
9695 /* We need to do this so that we can look at pending chunks */
9696 trans = btrfs_join_transaction(root);
9697 if (IS_ERR(trans)) {
9698 ret = PTR_ERR(trans);
9702 mutex_lock(&fs_info->chunk_mutex);
9703 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9707 * check to make sure we can actually find a chunk with enough
9708 * space to fit our block group in.
9710 if (device->total_bytes > device->bytes_used + min_free &&
9711 !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) {
9712 ret = find_free_dev_extent(trans, device, min_free,
9717 if (dev_nr >= dev_min)
9723 if (debug && ret == -1)
9725 "no space to allocate a new chunk for block group %llu",
9726 block_group->key.objectid);
9727 mutex_unlock(&fs_info->chunk_mutex);
9728 btrfs_end_transaction(trans);
9730 btrfs_put_block_group(block_group);
9734 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9735 struct btrfs_path *path,
9736 struct btrfs_key *key)
9738 struct btrfs_root *root = fs_info->extent_root;
9740 struct btrfs_key found_key;
9741 struct extent_buffer *leaf;
9744 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9749 slot = path->slots[0];
9750 leaf = path->nodes[0];
9751 if (slot >= btrfs_header_nritems(leaf)) {
9752 ret = btrfs_next_leaf(root, path);
9759 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9761 if (found_key.objectid >= key->objectid &&
9762 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9763 struct extent_map_tree *em_tree;
9764 struct extent_map *em;
9766 em_tree = &root->fs_info->mapping_tree.map_tree;
9767 read_lock(&em_tree->lock);
9768 em = lookup_extent_mapping(em_tree, found_key.objectid,
9770 read_unlock(&em_tree->lock);
9773 "logical %llu len %llu found bg but no related chunk",
9774 found_key.objectid, found_key.offset);
9779 free_extent_map(em);
9788 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9790 struct btrfs_block_group_cache *block_group;
9794 struct inode *inode;
9796 block_group = btrfs_lookup_first_block_group(info, last);
9797 while (block_group) {
9798 spin_lock(&block_group->lock);
9799 if (block_group->iref)
9801 spin_unlock(&block_group->lock);
9802 block_group = next_block_group(info, block_group);
9811 inode = block_group->inode;
9812 block_group->iref = 0;
9813 block_group->inode = NULL;
9814 spin_unlock(&block_group->lock);
9815 ASSERT(block_group->io_ctl.inode == NULL);
9817 last = block_group->key.objectid + block_group->key.offset;
9818 btrfs_put_block_group(block_group);
9823 * Must be called only after stopping all workers, since we could have block
9824 * group caching kthreads running, and therefore they could race with us if we
9825 * freed the block groups before stopping them.
9827 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9829 struct btrfs_block_group_cache *block_group;
9830 struct btrfs_space_info *space_info;
9831 struct btrfs_caching_control *caching_ctl;
9834 down_write(&info->commit_root_sem);
9835 while (!list_empty(&info->caching_block_groups)) {
9836 caching_ctl = list_entry(info->caching_block_groups.next,
9837 struct btrfs_caching_control, list);
9838 list_del(&caching_ctl->list);
9839 put_caching_control(caching_ctl);
9841 up_write(&info->commit_root_sem);
9843 spin_lock(&info->unused_bgs_lock);
9844 while (!list_empty(&info->unused_bgs)) {
9845 block_group = list_first_entry(&info->unused_bgs,
9846 struct btrfs_block_group_cache,
9848 list_del_init(&block_group->bg_list);
9849 btrfs_put_block_group(block_group);
9851 spin_unlock(&info->unused_bgs_lock);
9853 spin_lock(&info->block_group_cache_lock);
9854 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9855 block_group = rb_entry(n, struct btrfs_block_group_cache,
9857 rb_erase(&block_group->cache_node,
9858 &info->block_group_cache_tree);
9859 RB_CLEAR_NODE(&block_group->cache_node);
9860 spin_unlock(&info->block_group_cache_lock);
9862 down_write(&block_group->space_info->groups_sem);
9863 list_del(&block_group->list);
9864 up_write(&block_group->space_info->groups_sem);
9867 * We haven't cached this block group, which means we could
9868 * possibly have excluded extents on this block group.
9870 if (block_group->cached == BTRFS_CACHE_NO ||
9871 block_group->cached == BTRFS_CACHE_ERROR)
9872 free_excluded_extents(info, block_group);
9874 btrfs_remove_free_space_cache(block_group);
9875 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9876 ASSERT(list_empty(&block_group->dirty_list));
9877 ASSERT(list_empty(&block_group->io_list));
9878 ASSERT(list_empty(&block_group->bg_list));
9879 ASSERT(atomic_read(&block_group->count) == 1);
9880 btrfs_put_block_group(block_group);
9882 spin_lock(&info->block_group_cache_lock);
9884 spin_unlock(&info->block_group_cache_lock);
9886 /* now that all the block groups are freed, go through and
9887 * free all the space_info structs. This is only called during
9888 * the final stages of unmount, and so we know nobody is
9889 * using them. We call synchronize_rcu() once before we start,
9890 * just to be on the safe side.
9894 release_global_block_rsv(info);
9896 while (!list_empty(&info->space_info)) {
9899 space_info = list_entry(info->space_info.next,
9900 struct btrfs_space_info,
9904 * Do not hide this behind enospc_debug, this is actually
9905 * important and indicates a real bug if this happens.
9907 if (WARN_ON(space_info->bytes_pinned > 0 ||
9908 space_info->bytes_reserved > 0 ||
9909 space_info->bytes_may_use > 0))
9910 dump_space_info(info, space_info, 0, 0);
9911 list_del(&space_info->list);
9912 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9913 struct kobject *kobj;
9914 kobj = space_info->block_group_kobjs[i];
9915 space_info->block_group_kobjs[i] = NULL;
9921 kobject_del(&space_info->kobj);
9922 kobject_put(&space_info->kobj);
9927 /* link_block_group will queue up kobjects to add when we're reclaim-safe */
9928 void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info)
9930 struct btrfs_space_info *space_info;
9931 struct raid_kobject *rkobj;
9936 spin_lock(&fs_info->pending_raid_kobjs_lock);
9937 list_splice_init(&fs_info->pending_raid_kobjs, &list);
9938 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9940 list_for_each_entry(rkobj, &list, list) {
9941 space_info = __find_space_info(fs_info, rkobj->flags);
9942 index = btrfs_bg_flags_to_raid_index(rkobj->flags);
9944 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9945 "%s", get_raid_name(index));
9947 kobject_put(&rkobj->kobj);
9953 "failed to add kobject for block cache, ignoring");
9956 static void link_block_group(struct btrfs_block_group_cache *cache)
9958 struct btrfs_space_info *space_info = cache->space_info;
9959 struct btrfs_fs_info *fs_info = cache->fs_info;
9960 int index = btrfs_bg_flags_to_raid_index(cache->flags);
9963 down_write(&space_info->groups_sem);
9964 if (list_empty(&space_info->block_groups[index]))
9966 list_add_tail(&cache->list, &space_info->block_groups[index]);
9967 up_write(&space_info->groups_sem);
9970 struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9972 btrfs_warn(cache->fs_info,
9973 "couldn't alloc memory for raid level kobject");
9976 rkobj->flags = cache->flags;
9977 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9979 spin_lock(&fs_info->pending_raid_kobjs_lock);
9980 list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs);
9981 spin_unlock(&fs_info->pending_raid_kobjs_lock);
9982 space_info->block_group_kobjs[index] = &rkobj->kobj;
9986 static struct btrfs_block_group_cache *
9987 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9988 u64 start, u64 size)
9990 struct btrfs_block_group_cache *cache;
9992 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9996 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9998 if (!cache->free_space_ctl) {
10003 cache->key.objectid = start;
10004 cache->key.offset = size;
10005 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10007 cache->fs_info = fs_info;
10008 cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start);
10009 set_free_space_tree_thresholds(cache);
10011 atomic_set(&cache->count, 1);
10012 spin_lock_init(&cache->lock);
10013 init_rwsem(&cache->data_rwsem);
10014 INIT_LIST_HEAD(&cache->list);
10015 INIT_LIST_HEAD(&cache->cluster_list);
10016 INIT_LIST_HEAD(&cache->bg_list);
10017 INIT_LIST_HEAD(&cache->ro_list);
10018 INIT_LIST_HEAD(&cache->dirty_list);
10019 INIT_LIST_HEAD(&cache->io_list);
10020 btrfs_init_free_space_ctl(cache);
10021 atomic_set(&cache->trimming, 0);
10022 mutex_init(&cache->free_space_lock);
10023 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
10028 int btrfs_read_block_groups(struct btrfs_fs_info *info)
10030 struct btrfs_path *path;
10032 struct btrfs_block_group_cache *cache;
10033 struct btrfs_space_info *space_info;
10034 struct btrfs_key key;
10035 struct btrfs_key found_key;
10036 struct extent_buffer *leaf;
10037 int need_clear = 0;
10042 feature = btrfs_super_incompat_flags(info->super_copy);
10043 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
10047 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10048 path = btrfs_alloc_path();
10051 path->reada = READA_FORWARD;
10053 cache_gen = btrfs_super_cache_generation(info->super_copy);
10054 if (btrfs_test_opt(info, SPACE_CACHE) &&
10055 btrfs_super_generation(info->super_copy) != cache_gen)
10057 if (btrfs_test_opt(info, CLEAR_CACHE))
10061 ret = find_first_block_group(info, path, &key);
10067 leaf = path->nodes[0];
10068 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10070 cache = btrfs_create_block_group_cache(info, found_key.objectid,
10079 * When we mount with old space cache, we need to
10080 * set BTRFS_DC_CLEAR and set dirty flag.
10082 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10083 * truncate the old free space cache inode and
10085 * b) Setting 'dirty flag' makes sure that we flush
10086 * the new space cache info onto disk.
10088 if (btrfs_test_opt(info, SPACE_CACHE))
10089 cache->disk_cache_state = BTRFS_DC_CLEAR;
10092 read_extent_buffer(leaf, &cache->item,
10093 btrfs_item_ptr_offset(leaf, path->slots[0]),
10094 sizeof(cache->item));
10095 cache->flags = btrfs_block_group_flags(&cache->item);
10097 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10098 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10100 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10101 cache->key.objectid);
10106 key.objectid = found_key.objectid + found_key.offset;
10107 btrfs_release_path(path);
10110 * We need to exclude the super stripes now so that the space
10111 * info has super bytes accounted for, otherwise we'll think
10112 * we have more space than we actually do.
10114 ret = exclude_super_stripes(info, cache);
10117 * We may have excluded something, so call this just in
10120 free_excluded_extents(info, cache);
10121 btrfs_put_block_group(cache);
10126 * check for two cases, either we are full, and therefore
10127 * don't need to bother with the caching work since we won't
10128 * find any space, or we are empty, and we can just add all
10129 * the space in and be done with it. This saves us _alot_ of
10130 * time, particularly in the full case.
10132 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10133 cache->last_byte_to_unpin = (u64)-1;
10134 cache->cached = BTRFS_CACHE_FINISHED;
10135 free_excluded_extents(info, cache);
10136 } else if (btrfs_block_group_used(&cache->item) == 0) {
10137 cache->last_byte_to_unpin = (u64)-1;
10138 cache->cached = BTRFS_CACHE_FINISHED;
10139 add_new_free_space(cache, info,
10140 found_key.objectid,
10141 found_key.objectid +
10143 free_excluded_extents(info, cache);
10146 ret = btrfs_add_block_group_cache(info, cache);
10148 btrfs_remove_free_space_cache(cache);
10149 btrfs_put_block_group(cache);
10153 trace_btrfs_add_block_group(info, cache, 0);
10154 update_space_info(info, cache->flags, found_key.offset,
10155 btrfs_block_group_used(&cache->item),
10156 cache->bytes_super, &space_info);
10158 cache->space_info = space_info;
10160 link_block_group(cache);
10162 set_avail_alloc_bits(info, cache->flags);
10163 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10164 inc_block_group_ro(cache, 1);
10165 } else if (btrfs_block_group_used(&cache->item) == 0) {
10166 spin_lock(&info->unused_bgs_lock);
10167 /* Should always be true but just in case. */
10168 if (list_empty(&cache->bg_list)) {
10169 btrfs_get_block_group(cache);
10170 list_add_tail(&cache->bg_list,
10171 &info->unused_bgs);
10173 spin_unlock(&info->unused_bgs_lock);
10177 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10178 if (!(get_alloc_profile(info, space_info->flags) &
10179 (BTRFS_BLOCK_GROUP_RAID10 |
10180 BTRFS_BLOCK_GROUP_RAID1 |
10181 BTRFS_BLOCK_GROUP_RAID5 |
10182 BTRFS_BLOCK_GROUP_RAID6 |
10183 BTRFS_BLOCK_GROUP_DUP)))
10186 * avoid allocating from un-mirrored block group if there are
10187 * mirrored block groups.
10189 list_for_each_entry(cache,
10190 &space_info->block_groups[BTRFS_RAID_RAID0],
10192 inc_block_group_ro(cache, 1);
10193 list_for_each_entry(cache,
10194 &space_info->block_groups[BTRFS_RAID_SINGLE],
10196 inc_block_group_ro(cache, 1);
10199 btrfs_add_raid_kobjects(info);
10200 init_global_block_rsv(info);
10203 btrfs_free_path(path);
10207 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans)
10209 struct btrfs_fs_info *fs_info = trans->fs_info;
10210 struct btrfs_block_group_cache *block_group, *tmp;
10211 struct btrfs_root *extent_root = fs_info->extent_root;
10212 struct btrfs_block_group_item item;
10213 struct btrfs_key key;
10215 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10217 trans->can_flush_pending_bgs = false;
10218 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10222 spin_lock(&block_group->lock);
10223 memcpy(&item, &block_group->item, sizeof(item));
10224 memcpy(&key, &block_group->key, sizeof(key));
10225 spin_unlock(&block_group->lock);
10227 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10230 btrfs_abort_transaction(trans, ret);
10231 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10234 btrfs_abort_transaction(trans, ret);
10235 add_block_group_free_space(trans, fs_info, block_group);
10236 /* already aborted the transaction if it failed. */
10238 list_del_init(&block_group->bg_list);
10240 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10243 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10244 struct btrfs_fs_info *fs_info, u64 bytes_used,
10245 u64 type, u64 chunk_offset, u64 size)
10247 struct btrfs_block_group_cache *cache;
10250 btrfs_set_log_full_commit(fs_info, trans);
10252 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10256 btrfs_set_block_group_used(&cache->item, bytes_used);
10257 btrfs_set_block_group_chunk_objectid(&cache->item,
10258 BTRFS_FIRST_CHUNK_TREE_OBJECTID);
10259 btrfs_set_block_group_flags(&cache->item, type);
10261 cache->flags = type;
10262 cache->last_byte_to_unpin = (u64)-1;
10263 cache->cached = BTRFS_CACHE_FINISHED;
10264 cache->needs_free_space = 1;
10265 ret = exclude_super_stripes(fs_info, cache);
10268 * We may have excluded something, so call this just in
10271 free_excluded_extents(fs_info, cache);
10272 btrfs_put_block_group(cache);
10276 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10278 free_excluded_extents(fs_info, cache);
10280 #ifdef CONFIG_BTRFS_DEBUG
10281 if (btrfs_should_fragment_free_space(cache)) {
10282 u64 new_bytes_used = size - bytes_used;
10284 bytes_used += new_bytes_used >> 1;
10285 fragment_free_space(cache);
10289 * Ensure the corresponding space_info object is created and
10290 * assigned to our block group. We want our bg to be added to the rbtree
10291 * with its ->space_info set.
10293 cache->space_info = __find_space_info(fs_info, cache->flags);
10294 ASSERT(cache->space_info);
10296 ret = btrfs_add_block_group_cache(fs_info, cache);
10298 btrfs_remove_free_space_cache(cache);
10299 btrfs_put_block_group(cache);
10304 * Now that our block group has its ->space_info set and is inserted in
10305 * the rbtree, update the space info's counters.
10307 trace_btrfs_add_block_group(fs_info, cache, 1);
10308 update_space_info(fs_info, cache->flags, size, bytes_used,
10309 cache->bytes_super, &cache->space_info);
10310 update_global_block_rsv(fs_info);
10312 link_block_group(cache);
10314 list_add_tail(&cache->bg_list, &trans->new_bgs);
10316 set_avail_alloc_bits(fs_info, type);
10320 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10322 u64 extra_flags = chunk_to_extended(flags) &
10323 BTRFS_EXTENDED_PROFILE_MASK;
10325 write_seqlock(&fs_info->profiles_lock);
10326 if (flags & BTRFS_BLOCK_GROUP_DATA)
10327 fs_info->avail_data_alloc_bits &= ~extra_flags;
10328 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10329 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10330 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10331 fs_info->avail_system_alloc_bits &= ~extra_flags;
10332 write_sequnlock(&fs_info->profiles_lock);
10335 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10336 struct btrfs_fs_info *fs_info, u64 group_start,
10337 struct extent_map *em)
10339 struct btrfs_root *root = fs_info->extent_root;
10340 struct btrfs_path *path;
10341 struct btrfs_block_group_cache *block_group;
10342 struct btrfs_free_cluster *cluster;
10343 struct btrfs_root *tree_root = fs_info->tree_root;
10344 struct btrfs_key key;
10345 struct inode *inode;
10346 struct kobject *kobj = NULL;
10350 struct btrfs_caching_control *caching_ctl = NULL;
10353 block_group = btrfs_lookup_block_group(fs_info, group_start);
10354 BUG_ON(!block_group);
10355 BUG_ON(!block_group->ro);
10358 * Free the reserved super bytes from this block group before
10361 free_excluded_extents(fs_info, block_group);
10362 btrfs_free_ref_tree_range(fs_info, block_group->key.objectid,
10363 block_group->key.offset);
10365 memcpy(&key, &block_group->key, sizeof(key));
10366 index = btrfs_bg_flags_to_raid_index(block_group->flags);
10367 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10368 BTRFS_BLOCK_GROUP_RAID1 |
10369 BTRFS_BLOCK_GROUP_RAID10))
10374 /* make sure this block group isn't part of an allocation cluster */
10375 cluster = &fs_info->data_alloc_cluster;
10376 spin_lock(&cluster->refill_lock);
10377 btrfs_return_cluster_to_free_space(block_group, cluster);
10378 spin_unlock(&cluster->refill_lock);
10381 * make sure this block group isn't part of a metadata
10382 * allocation cluster
10384 cluster = &fs_info->meta_alloc_cluster;
10385 spin_lock(&cluster->refill_lock);
10386 btrfs_return_cluster_to_free_space(block_group, cluster);
10387 spin_unlock(&cluster->refill_lock);
10389 path = btrfs_alloc_path();
10396 * get the inode first so any iput calls done for the io_list
10397 * aren't the final iput (no unlinks allowed now)
10399 inode = lookup_free_space_inode(fs_info, block_group, path);
10401 mutex_lock(&trans->transaction->cache_write_mutex);
10403 * make sure our free spache cache IO is done before remove the
10406 spin_lock(&trans->transaction->dirty_bgs_lock);
10407 if (!list_empty(&block_group->io_list)) {
10408 list_del_init(&block_group->io_list);
10410 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10412 spin_unlock(&trans->transaction->dirty_bgs_lock);
10413 btrfs_wait_cache_io(trans, block_group, path);
10414 btrfs_put_block_group(block_group);
10415 spin_lock(&trans->transaction->dirty_bgs_lock);
10418 if (!list_empty(&block_group->dirty_list)) {
10419 list_del_init(&block_group->dirty_list);
10420 btrfs_put_block_group(block_group);
10422 spin_unlock(&trans->transaction->dirty_bgs_lock);
10423 mutex_unlock(&trans->transaction->cache_write_mutex);
10425 if (!IS_ERR(inode)) {
10426 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10428 btrfs_add_delayed_iput(inode);
10431 clear_nlink(inode);
10432 /* One for the block groups ref */
10433 spin_lock(&block_group->lock);
10434 if (block_group->iref) {
10435 block_group->iref = 0;
10436 block_group->inode = NULL;
10437 spin_unlock(&block_group->lock);
10440 spin_unlock(&block_group->lock);
10442 /* One for our lookup ref */
10443 btrfs_add_delayed_iput(inode);
10446 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10447 key.offset = block_group->key.objectid;
10450 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10454 btrfs_release_path(path);
10456 ret = btrfs_del_item(trans, tree_root, path);
10459 btrfs_release_path(path);
10462 spin_lock(&fs_info->block_group_cache_lock);
10463 rb_erase(&block_group->cache_node,
10464 &fs_info->block_group_cache_tree);
10465 RB_CLEAR_NODE(&block_group->cache_node);
10467 if (fs_info->first_logical_byte == block_group->key.objectid)
10468 fs_info->first_logical_byte = (u64)-1;
10469 spin_unlock(&fs_info->block_group_cache_lock);
10471 down_write(&block_group->space_info->groups_sem);
10473 * we must use list_del_init so people can check to see if they
10474 * are still on the list after taking the semaphore
10476 list_del_init(&block_group->list);
10477 if (list_empty(&block_group->space_info->block_groups[index])) {
10478 kobj = block_group->space_info->block_group_kobjs[index];
10479 block_group->space_info->block_group_kobjs[index] = NULL;
10480 clear_avail_alloc_bits(fs_info, block_group->flags);
10482 up_write(&block_group->space_info->groups_sem);
10488 if (block_group->has_caching_ctl)
10489 caching_ctl = get_caching_control(block_group);
10490 if (block_group->cached == BTRFS_CACHE_STARTED)
10491 wait_block_group_cache_done(block_group);
10492 if (block_group->has_caching_ctl) {
10493 down_write(&fs_info->commit_root_sem);
10494 if (!caching_ctl) {
10495 struct btrfs_caching_control *ctl;
10497 list_for_each_entry(ctl,
10498 &fs_info->caching_block_groups, list)
10499 if (ctl->block_group == block_group) {
10501 refcount_inc(&caching_ctl->count);
10506 list_del_init(&caching_ctl->list);
10507 up_write(&fs_info->commit_root_sem);
10509 /* Once for the caching bgs list and once for us. */
10510 put_caching_control(caching_ctl);
10511 put_caching_control(caching_ctl);
10515 spin_lock(&trans->transaction->dirty_bgs_lock);
10516 if (!list_empty(&block_group->dirty_list)) {
10519 if (!list_empty(&block_group->io_list)) {
10522 spin_unlock(&trans->transaction->dirty_bgs_lock);
10523 btrfs_remove_free_space_cache(block_group);
10525 spin_lock(&block_group->space_info->lock);
10526 list_del_init(&block_group->ro_list);
10528 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10529 WARN_ON(block_group->space_info->total_bytes
10530 < block_group->key.offset);
10531 WARN_ON(block_group->space_info->bytes_readonly
10532 < block_group->key.offset);
10533 WARN_ON(block_group->space_info->disk_total
10534 < block_group->key.offset * factor);
10536 block_group->space_info->total_bytes -= block_group->key.offset;
10537 block_group->space_info->bytes_readonly -= block_group->key.offset;
10538 block_group->space_info->disk_total -= block_group->key.offset * factor;
10540 spin_unlock(&block_group->space_info->lock);
10542 memcpy(&key, &block_group->key, sizeof(key));
10544 mutex_lock(&fs_info->chunk_mutex);
10545 if (!list_empty(&em->list)) {
10546 /* We're in the transaction->pending_chunks list. */
10547 free_extent_map(em);
10549 spin_lock(&block_group->lock);
10550 block_group->removed = 1;
10552 * At this point trimming can't start on this block group, because we
10553 * removed the block group from the tree fs_info->block_group_cache_tree
10554 * so no one can't find it anymore and even if someone already got this
10555 * block group before we removed it from the rbtree, they have already
10556 * incremented block_group->trimming - if they didn't, they won't find
10557 * any free space entries because we already removed them all when we
10558 * called btrfs_remove_free_space_cache().
10560 * And we must not remove the extent map from the fs_info->mapping_tree
10561 * to prevent the same logical address range and physical device space
10562 * ranges from being reused for a new block group. This is because our
10563 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10564 * completely transactionless, so while it is trimming a range the
10565 * currently running transaction might finish and a new one start,
10566 * allowing for new block groups to be created that can reuse the same
10567 * physical device locations unless we take this special care.
10569 * There may also be an implicit trim operation if the file system
10570 * is mounted with -odiscard. The same protections must remain
10571 * in place until the extents have been discarded completely when
10572 * the transaction commit has completed.
10574 remove_em = (atomic_read(&block_group->trimming) == 0);
10576 * Make sure a trimmer task always sees the em in the pinned_chunks list
10577 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10578 * before checking block_group->removed).
10582 * Our em might be in trans->transaction->pending_chunks which
10583 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10584 * and so is the fs_info->pinned_chunks list.
10586 * So at this point we must be holding the chunk_mutex to avoid
10587 * any races with chunk allocation (more specifically at
10588 * volumes.c:contains_pending_extent()), to ensure it always
10589 * sees the em, either in the pending_chunks list or in the
10590 * pinned_chunks list.
10592 list_move_tail(&em->list, &fs_info->pinned_chunks);
10594 spin_unlock(&block_group->lock);
10597 struct extent_map_tree *em_tree;
10599 em_tree = &fs_info->mapping_tree.map_tree;
10600 write_lock(&em_tree->lock);
10602 * The em might be in the pending_chunks list, so make sure the
10603 * chunk mutex is locked, since remove_extent_mapping() will
10604 * delete us from that list.
10606 remove_extent_mapping(em_tree, em);
10607 write_unlock(&em_tree->lock);
10608 /* once for the tree */
10609 free_extent_map(em);
10612 mutex_unlock(&fs_info->chunk_mutex);
10614 ret = remove_block_group_free_space(trans, fs_info, block_group);
10618 btrfs_put_block_group(block_group);
10619 btrfs_put_block_group(block_group);
10621 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10627 ret = btrfs_del_item(trans, root, path);
10629 btrfs_free_path(path);
10633 struct btrfs_trans_handle *
10634 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10635 const u64 chunk_offset)
10637 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10638 struct extent_map *em;
10639 struct map_lookup *map;
10640 unsigned int num_items;
10642 read_lock(&em_tree->lock);
10643 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10644 read_unlock(&em_tree->lock);
10645 ASSERT(em && em->start == chunk_offset);
10648 * We need to reserve 3 + N units from the metadata space info in order
10649 * to remove a block group (done at btrfs_remove_chunk() and at
10650 * btrfs_remove_block_group()), which are used for:
10652 * 1 unit for adding the free space inode's orphan (located in the tree
10654 * 1 unit for deleting the block group item (located in the extent
10656 * 1 unit for deleting the free space item (located in tree of tree
10658 * N units for deleting N device extent items corresponding to each
10659 * stripe (located in the device tree).
10661 * In order to remove a block group we also need to reserve units in the
10662 * system space info in order to update the chunk tree (update one or
10663 * more device items and remove one chunk item), but this is done at
10664 * btrfs_remove_chunk() through a call to check_system_chunk().
10666 map = em->map_lookup;
10667 num_items = 3 + map->num_stripes;
10668 free_extent_map(em);
10670 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10675 * Process the unused_bgs list and remove any that don't have any allocated
10676 * space inside of them.
10678 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10680 struct btrfs_block_group_cache *block_group;
10681 struct btrfs_space_info *space_info;
10682 struct btrfs_trans_handle *trans;
10685 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10688 spin_lock(&fs_info->unused_bgs_lock);
10689 while (!list_empty(&fs_info->unused_bgs)) {
10693 block_group = list_first_entry(&fs_info->unused_bgs,
10694 struct btrfs_block_group_cache,
10696 list_del_init(&block_group->bg_list);
10698 space_info = block_group->space_info;
10700 if (ret || btrfs_mixed_space_info(space_info)) {
10701 btrfs_put_block_group(block_group);
10704 spin_unlock(&fs_info->unused_bgs_lock);
10706 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10708 /* Don't want to race with allocators so take the groups_sem */
10709 down_write(&space_info->groups_sem);
10710 spin_lock(&block_group->lock);
10711 if (block_group->reserved ||
10712 btrfs_block_group_used(&block_group->item) ||
10714 list_is_singular(&block_group->list)) {
10716 * We want to bail if we made new allocations or have
10717 * outstanding allocations in this block group. We do
10718 * the ro check in case balance is currently acting on
10719 * this block group.
10721 spin_unlock(&block_group->lock);
10722 up_write(&space_info->groups_sem);
10725 spin_unlock(&block_group->lock);
10727 /* We don't want to force the issue, only flip if it's ok. */
10728 ret = inc_block_group_ro(block_group, 0);
10729 up_write(&space_info->groups_sem);
10736 * Want to do this before we do anything else so we can recover
10737 * properly if we fail to join the transaction.
10739 trans = btrfs_start_trans_remove_block_group(fs_info,
10740 block_group->key.objectid);
10741 if (IS_ERR(trans)) {
10742 btrfs_dec_block_group_ro(block_group);
10743 ret = PTR_ERR(trans);
10748 * We could have pending pinned extents for this block group,
10749 * just delete them, we don't care about them anymore.
10751 start = block_group->key.objectid;
10752 end = start + block_group->key.offset - 1;
10754 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10755 * btrfs_finish_extent_commit(). If we are at transaction N,
10756 * another task might be running finish_extent_commit() for the
10757 * previous transaction N - 1, and have seen a range belonging
10758 * to the block group in freed_extents[] before we were able to
10759 * clear the whole block group range from freed_extents[]. This
10760 * means that task can lookup for the block group after we
10761 * unpinned it from freed_extents[] and removed it, leading to
10762 * a BUG_ON() at btrfs_unpin_extent_range().
10764 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10765 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10768 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10769 btrfs_dec_block_group_ro(block_group);
10772 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10775 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10776 btrfs_dec_block_group_ro(block_group);
10779 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10781 /* Reset pinned so btrfs_put_block_group doesn't complain */
10782 spin_lock(&space_info->lock);
10783 spin_lock(&block_group->lock);
10785 space_info->bytes_pinned -= block_group->pinned;
10786 space_info->bytes_readonly += block_group->pinned;
10787 percpu_counter_add(&space_info->total_bytes_pinned,
10788 -block_group->pinned);
10789 block_group->pinned = 0;
10791 spin_unlock(&block_group->lock);
10792 spin_unlock(&space_info->lock);
10794 /* DISCARD can flip during remount */
10795 trimming = btrfs_test_opt(fs_info, DISCARD);
10797 /* Implicit trim during transaction commit. */
10799 btrfs_get_block_group_trimming(block_group);
10802 * Btrfs_remove_chunk will abort the transaction if things go
10805 ret = btrfs_remove_chunk(trans, fs_info,
10806 block_group->key.objectid);
10810 btrfs_put_block_group_trimming(block_group);
10815 * If we're not mounted with -odiscard, we can just forget
10816 * about this block group. Otherwise we'll need to wait
10817 * until transaction commit to do the actual discard.
10820 spin_lock(&fs_info->unused_bgs_lock);
10822 * A concurrent scrub might have added us to the list
10823 * fs_info->unused_bgs, so use a list_move operation
10824 * to add the block group to the deleted_bgs list.
10826 list_move(&block_group->bg_list,
10827 &trans->transaction->deleted_bgs);
10828 spin_unlock(&fs_info->unused_bgs_lock);
10829 btrfs_get_block_group(block_group);
10832 btrfs_end_transaction(trans);
10834 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10835 btrfs_put_block_group(block_group);
10836 spin_lock(&fs_info->unused_bgs_lock);
10838 spin_unlock(&fs_info->unused_bgs_lock);
10841 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10843 struct btrfs_space_info *space_info;
10844 struct btrfs_super_block *disk_super;
10850 disk_super = fs_info->super_copy;
10851 if (!btrfs_super_root(disk_super))
10854 features = btrfs_super_incompat_flags(disk_super);
10855 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10858 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10859 ret = create_space_info(fs_info, flags, &space_info);
10864 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10865 ret = create_space_info(fs_info, flags, &space_info);
10867 flags = BTRFS_BLOCK_GROUP_METADATA;
10868 ret = create_space_info(fs_info, flags, &space_info);
10872 flags = BTRFS_BLOCK_GROUP_DATA;
10873 ret = create_space_info(fs_info, flags, &space_info);
10879 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10880 u64 start, u64 end)
10882 return unpin_extent_range(fs_info, start, end, false);
10886 * It used to be that old block groups would be left around forever.
10887 * Iterating over them would be enough to trim unused space. Since we
10888 * now automatically remove them, we also need to iterate over unallocated
10891 * We don't want a transaction for this since the discard may take a
10892 * substantial amount of time. We don't require that a transaction be
10893 * running, but we do need to take a running transaction into account
10894 * to ensure that we're not discarding chunks that were released in
10895 * the current transaction.
10897 * Holding the chunks lock will prevent other threads from allocating
10898 * or releasing chunks, but it won't prevent a running transaction
10899 * from committing and releasing the memory that the pending chunks
10900 * list head uses. For that, we need to take a reference to the
10903 static int btrfs_trim_free_extents(struct btrfs_device *device,
10904 u64 minlen, u64 *trimmed)
10906 u64 start = 0, len = 0;
10911 /* Not writeable = nothing to do. */
10912 if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state))
10915 /* No free space = nothing to do. */
10916 if (device->total_bytes <= device->bytes_used)
10922 struct btrfs_fs_info *fs_info = device->fs_info;
10923 struct btrfs_transaction *trans;
10926 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10930 down_read(&fs_info->commit_root_sem);
10932 spin_lock(&fs_info->trans_lock);
10933 trans = fs_info->running_transaction;
10935 refcount_inc(&trans->use_count);
10936 spin_unlock(&fs_info->trans_lock);
10938 ret = find_free_dev_extent_start(trans, device, minlen, start,
10941 btrfs_put_transaction(trans);
10944 up_read(&fs_info->commit_root_sem);
10945 mutex_unlock(&fs_info->chunk_mutex);
10946 if (ret == -ENOSPC)
10951 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10952 up_read(&fs_info->commit_root_sem);
10953 mutex_unlock(&fs_info->chunk_mutex);
10961 if (fatal_signal_pending(current)) {
10962 ret = -ERESTARTSYS;
10972 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10974 struct btrfs_block_group_cache *cache = NULL;
10975 struct btrfs_device *device;
10976 struct list_head *devices;
10981 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10985 * try to trim all FS space, our block group may start from non-zero.
10987 if (range->len == total_bytes)
10988 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10990 cache = btrfs_lookup_block_group(fs_info, range->start);
10993 if (cache->key.objectid >= (range->start + range->len)) {
10994 btrfs_put_block_group(cache);
10998 start = max(range->start, cache->key.objectid);
10999 end = min(range->start + range->len,
11000 cache->key.objectid + cache->key.offset);
11002 if (end - start >= range->minlen) {
11003 if (!block_group_cache_done(cache)) {
11004 ret = cache_block_group(cache, 0);
11006 btrfs_put_block_group(cache);
11009 ret = wait_block_group_cache_done(cache);
11011 btrfs_put_block_group(cache);
11015 ret = btrfs_trim_block_group(cache,
11021 trimmed += group_trimmed;
11023 btrfs_put_block_group(cache);
11028 cache = next_block_group(fs_info, cache);
11031 mutex_lock(&fs_info->fs_devices->device_list_mutex);
11032 devices = &fs_info->fs_devices->alloc_list;
11033 list_for_each_entry(device, devices, dev_alloc_list) {
11034 ret = btrfs_trim_free_extents(device, range->minlen,
11039 trimmed += group_trimmed;
11041 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
11043 range->len = trimmed;
11048 * btrfs_{start,end}_write_no_snapshotting() are similar to
11049 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11050 * data into the page cache through nocow before the subvolume is snapshoted,
11051 * but flush the data into disk after the snapshot creation, or to prevent
11052 * operations while snapshotting is ongoing and that cause the snapshot to be
11053 * inconsistent (writes followed by expanding truncates for example).
11055 void btrfs_end_write_no_snapshotting(struct btrfs_root *root)
11057 percpu_counter_dec(&root->subv_writers->counter);
11059 * Make sure counter is updated before we wake up waiters.
11062 if (waitqueue_active(&root->subv_writers->wait))
11063 wake_up(&root->subv_writers->wait);
11066 int btrfs_start_write_no_snapshotting(struct btrfs_root *root)
11068 if (atomic_read(&root->will_be_snapshotted))
11071 percpu_counter_inc(&root->subv_writers->counter);
11073 * Make sure counter is updated before we check for snapshot creation.
11076 if (atomic_read(&root->will_be_snapshotted)) {
11077 btrfs_end_write_no_snapshotting(root);
11083 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11088 ret = btrfs_start_write_no_snapshotting(root);
11091 wait_var_event(&root->will_be_snapshotted,
11092 !atomic_read(&root->will_be_snapshotted));