2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/sched/signal.h>
20 #include <linux/pagemap.h>
21 #include <linux/writeback.h>
22 #include <linux/blkdev.h>
23 #include <linux/sort.h>
24 #include <linux/rcupdate.h>
25 #include <linux/kthread.h>
26 #include <linux/slab.h>
27 #include <linux/ratelimit.h>
28 #include <linux/percpu_counter.h>
32 #include "print-tree.h"
36 #include "free-space-cache.h"
37 #include "free-space-tree.h"
42 #undef SCRAMBLE_DELAYED_REFS
45 * control flags for do_chunk_alloc's force field
46 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
47 * if we really need one.
49 * CHUNK_ALLOC_LIMITED means to only try and allocate one
50 * if we have very few chunks already allocated. This is
51 * used as part of the clustering code to help make sure
52 * we have a good pool of storage to cluster in, without
53 * filling the FS with empty chunks
55 * CHUNK_ALLOC_FORCE means it must try to allocate one
59 CHUNK_ALLOC_NO_FORCE = 0,
60 CHUNK_ALLOC_LIMITED = 1,
61 CHUNK_ALLOC_FORCE = 2,
64 static int update_block_group(struct btrfs_trans_handle *trans,
65 struct btrfs_fs_info *fs_info, u64 bytenr,
66 u64 num_bytes, int alloc);
67 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
68 struct btrfs_fs_info *fs_info,
69 struct btrfs_delayed_ref_node *node, u64 parent,
70 u64 root_objectid, u64 owner_objectid,
71 u64 owner_offset, int refs_to_drop,
72 struct btrfs_delayed_extent_op *extra_op);
73 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
74 struct extent_buffer *leaf,
75 struct btrfs_extent_item *ei);
76 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
77 struct btrfs_fs_info *fs_info,
78 u64 parent, u64 root_objectid,
79 u64 flags, u64 owner, u64 offset,
80 struct btrfs_key *ins, int ref_mod);
81 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
82 struct btrfs_fs_info *fs_info,
83 u64 parent, u64 root_objectid,
84 u64 flags, struct btrfs_disk_key *key,
85 int level, struct btrfs_key *ins);
86 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
87 struct btrfs_fs_info *fs_info, u64 flags,
89 static int find_next_key(struct btrfs_path *path, int level,
90 struct btrfs_key *key);
91 static void dump_space_info(struct btrfs_fs_info *fs_info,
92 struct btrfs_space_info *info, u64 bytes,
93 int dump_block_groups);
94 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
95 u64 ram_bytes, u64 num_bytes, int delalloc);
96 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
97 u64 num_bytes, int delalloc);
98 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
100 static int __reserve_metadata_bytes(struct btrfs_root *root,
101 struct btrfs_space_info *space_info,
103 enum btrfs_reserve_flush_enum flush);
104 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
105 struct btrfs_space_info *space_info,
107 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
108 struct btrfs_space_info *space_info,
112 block_group_cache_done(struct btrfs_block_group_cache *cache)
115 return cache->cached == BTRFS_CACHE_FINISHED ||
116 cache->cached == BTRFS_CACHE_ERROR;
119 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
121 return (cache->flags & bits) == bits;
124 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
126 atomic_inc(&cache->count);
129 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
131 if (atomic_dec_and_test(&cache->count)) {
132 WARN_ON(cache->pinned > 0);
133 WARN_ON(cache->reserved > 0);
136 * If not empty, someone is still holding mutex of
137 * full_stripe_lock, which can only be released by caller.
138 * And it will definitely cause use-after-free when caller
139 * tries to release full stripe lock.
141 * No better way to resolve, but only to warn.
143 WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root));
144 kfree(cache->free_space_ctl);
150 * this adds the block group to the fs_info rb tree for the block group
153 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
154 struct btrfs_block_group_cache *block_group)
157 struct rb_node *parent = NULL;
158 struct btrfs_block_group_cache *cache;
160 spin_lock(&info->block_group_cache_lock);
161 p = &info->block_group_cache_tree.rb_node;
165 cache = rb_entry(parent, struct btrfs_block_group_cache,
167 if (block_group->key.objectid < cache->key.objectid) {
169 } else if (block_group->key.objectid > cache->key.objectid) {
172 spin_unlock(&info->block_group_cache_lock);
177 rb_link_node(&block_group->cache_node, parent, p);
178 rb_insert_color(&block_group->cache_node,
179 &info->block_group_cache_tree);
181 if (info->first_logical_byte > block_group->key.objectid)
182 info->first_logical_byte = block_group->key.objectid;
184 spin_unlock(&info->block_group_cache_lock);
190 * This will return the block group at or after bytenr if contains is 0, else
191 * it will return the block group that contains the bytenr
193 static struct btrfs_block_group_cache *
194 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
197 struct btrfs_block_group_cache *cache, *ret = NULL;
201 spin_lock(&info->block_group_cache_lock);
202 n = info->block_group_cache_tree.rb_node;
205 cache = rb_entry(n, struct btrfs_block_group_cache,
207 end = cache->key.objectid + cache->key.offset - 1;
208 start = cache->key.objectid;
210 if (bytenr < start) {
211 if (!contains && (!ret || start < ret->key.objectid))
214 } else if (bytenr > start) {
215 if (contains && bytenr <= end) {
226 btrfs_get_block_group(ret);
227 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
228 info->first_logical_byte = ret->key.objectid;
230 spin_unlock(&info->block_group_cache_lock);
235 static int add_excluded_extent(struct btrfs_fs_info *fs_info,
236 u64 start, u64 num_bytes)
238 u64 end = start + num_bytes - 1;
239 set_extent_bits(&fs_info->freed_extents[0],
240 start, end, EXTENT_UPTODATE);
241 set_extent_bits(&fs_info->freed_extents[1],
242 start, end, EXTENT_UPTODATE);
246 static void free_excluded_extents(struct btrfs_fs_info *fs_info,
247 struct btrfs_block_group_cache *cache)
251 start = cache->key.objectid;
252 end = start + cache->key.offset - 1;
254 clear_extent_bits(&fs_info->freed_extents[0],
255 start, end, EXTENT_UPTODATE);
256 clear_extent_bits(&fs_info->freed_extents[1],
257 start, end, EXTENT_UPTODATE);
260 static int exclude_super_stripes(struct btrfs_fs_info *fs_info,
261 struct btrfs_block_group_cache *cache)
268 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
269 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
270 cache->bytes_super += stripe_len;
271 ret = add_excluded_extent(fs_info, cache->key.objectid,
277 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
278 bytenr = btrfs_sb_offset(i);
279 ret = btrfs_rmap_block(fs_info, cache->key.objectid,
280 bytenr, 0, &logical, &nr, &stripe_len);
287 if (logical[nr] > cache->key.objectid +
291 if (logical[nr] + stripe_len <= cache->key.objectid)
295 if (start < cache->key.objectid) {
296 start = cache->key.objectid;
297 len = (logical[nr] + stripe_len) - start;
299 len = min_t(u64, stripe_len,
300 cache->key.objectid +
301 cache->key.offset - start);
304 cache->bytes_super += len;
305 ret = add_excluded_extent(fs_info, start, len);
317 static struct btrfs_caching_control *
318 get_caching_control(struct btrfs_block_group_cache *cache)
320 struct btrfs_caching_control *ctl;
322 spin_lock(&cache->lock);
323 if (!cache->caching_ctl) {
324 spin_unlock(&cache->lock);
328 ctl = cache->caching_ctl;
329 refcount_inc(&ctl->count);
330 spin_unlock(&cache->lock);
334 static void put_caching_control(struct btrfs_caching_control *ctl)
336 if (refcount_dec_and_test(&ctl->count))
340 #ifdef CONFIG_BTRFS_DEBUG
341 static void fragment_free_space(struct btrfs_block_group_cache *block_group)
343 struct btrfs_fs_info *fs_info = block_group->fs_info;
344 u64 start = block_group->key.objectid;
345 u64 len = block_group->key.offset;
346 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
347 fs_info->nodesize : fs_info->sectorsize;
348 u64 step = chunk << 1;
350 while (len > chunk) {
351 btrfs_remove_free_space(block_group, start, chunk);
362 * this is only called by cache_block_group, since we could have freed extents
363 * we need to check the pinned_extents for any extents that can't be used yet
364 * since their free space will be released as soon as the transaction commits.
366 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
367 struct btrfs_fs_info *info, u64 start, u64 end)
369 u64 extent_start, extent_end, size, total_added = 0;
372 while (start < end) {
373 ret = find_first_extent_bit(info->pinned_extents, start,
374 &extent_start, &extent_end,
375 EXTENT_DIRTY | EXTENT_UPTODATE,
380 if (extent_start <= start) {
381 start = extent_end + 1;
382 } else if (extent_start > start && extent_start < end) {
383 size = extent_start - start;
385 ret = btrfs_add_free_space(block_group, start,
387 BUG_ON(ret); /* -ENOMEM or logic error */
388 start = extent_end + 1;
397 ret = btrfs_add_free_space(block_group, start, size);
398 BUG_ON(ret); /* -ENOMEM or logic error */
404 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
406 struct btrfs_block_group_cache *block_group = caching_ctl->block_group;
407 struct btrfs_fs_info *fs_info = block_group->fs_info;
408 struct btrfs_root *extent_root = fs_info->extent_root;
409 struct btrfs_path *path;
410 struct extent_buffer *leaf;
411 struct btrfs_key key;
418 path = btrfs_alloc_path();
422 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
424 #ifdef CONFIG_BTRFS_DEBUG
426 * If we're fragmenting we don't want to make anybody think we can
427 * allocate from this block group until we've had a chance to fragment
430 if (btrfs_should_fragment_free_space(block_group))
434 * We don't want to deadlock with somebody trying to allocate a new
435 * extent for the extent root while also trying to search the extent
436 * root to add free space. So we skip locking and search the commit
437 * root, since its read-only
439 path->skip_locking = 1;
440 path->search_commit_root = 1;
441 path->reada = READA_FORWARD;
445 key.type = BTRFS_EXTENT_ITEM_KEY;
448 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
452 leaf = path->nodes[0];
453 nritems = btrfs_header_nritems(leaf);
456 if (btrfs_fs_closing(fs_info) > 1) {
461 if (path->slots[0] < nritems) {
462 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
464 ret = find_next_key(path, 0, &key);
468 if (need_resched() ||
469 rwsem_is_contended(&fs_info->commit_root_sem)) {
471 caching_ctl->progress = last;
472 btrfs_release_path(path);
473 up_read(&fs_info->commit_root_sem);
474 mutex_unlock(&caching_ctl->mutex);
476 mutex_lock(&caching_ctl->mutex);
477 down_read(&fs_info->commit_root_sem);
481 ret = btrfs_next_leaf(extent_root, path);
486 leaf = path->nodes[0];
487 nritems = btrfs_header_nritems(leaf);
491 if (key.objectid < last) {
494 key.type = BTRFS_EXTENT_ITEM_KEY;
497 caching_ctl->progress = last;
498 btrfs_release_path(path);
502 if (key.objectid < block_group->key.objectid) {
507 if (key.objectid >= block_group->key.objectid +
508 block_group->key.offset)
511 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
512 key.type == BTRFS_METADATA_ITEM_KEY) {
513 total_found += add_new_free_space(block_group,
516 if (key.type == BTRFS_METADATA_ITEM_KEY)
517 last = key.objectid +
520 last = key.objectid + key.offset;
522 if (total_found > CACHING_CTL_WAKE_UP) {
525 wake_up(&caching_ctl->wait);
532 total_found += add_new_free_space(block_group, fs_info, last,
533 block_group->key.objectid +
534 block_group->key.offset);
535 caching_ctl->progress = (u64)-1;
538 btrfs_free_path(path);
542 static noinline void caching_thread(struct btrfs_work *work)
544 struct btrfs_block_group_cache *block_group;
545 struct btrfs_fs_info *fs_info;
546 struct btrfs_caching_control *caching_ctl;
547 struct btrfs_root *extent_root;
550 caching_ctl = container_of(work, struct btrfs_caching_control, work);
551 block_group = caching_ctl->block_group;
552 fs_info = block_group->fs_info;
553 extent_root = fs_info->extent_root;
555 mutex_lock(&caching_ctl->mutex);
556 down_read(&fs_info->commit_root_sem);
558 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
559 ret = load_free_space_tree(caching_ctl);
561 ret = load_extent_tree_free(caching_ctl);
563 spin_lock(&block_group->lock);
564 block_group->caching_ctl = NULL;
565 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
566 spin_unlock(&block_group->lock);
568 #ifdef CONFIG_BTRFS_DEBUG
569 if (btrfs_should_fragment_free_space(block_group)) {
572 spin_lock(&block_group->space_info->lock);
573 spin_lock(&block_group->lock);
574 bytes_used = block_group->key.offset -
575 btrfs_block_group_used(&block_group->item);
576 block_group->space_info->bytes_used += bytes_used >> 1;
577 spin_unlock(&block_group->lock);
578 spin_unlock(&block_group->space_info->lock);
579 fragment_free_space(block_group);
583 caching_ctl->progress = (u64)-1;
585 up_read(&fs_info->commit_root_sem);
586 free_excluded_extents(fs_info, block_group);
587 mutex_unlock(&caching_ctl->mutex);
589 wake_up(&caching_ctl->wait);
591 put_caching_control(caching_ctl);
592 btrfs_put_block_group(block_group);
595 static int cache_block_group(struct btrfs_block_group_cache *cache,
599 struct btrfs_fs_info *fs_info = cache->fs_info;
600 struct btrfs_caching_control *caching_ctl;
603 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
607 INIT_LIST_HEAD(&caching_ctl->list);
608 mutex_init(&caching_ctl->mutex);
609 init_waitqueue_head(&caching_ctl->wait);
610 caching_ctl->block_group = cache;
611 caching_ctl->progress = cache->key.objectid;
612 refcount_set(&caching_ctl->count, 1);
613 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
614 caching_thread, NULL, NULL);
616 spin_lock(&cache->lock);
618 * This should be a rare occasion, but this could happen I think in the
619 * case where one thread starts to load the space cache info, and then
620 * some other thread starts a transaction commit which tries to do an
621 * allocation while the other thread is still loading the space cache
622 * info. The previous loop should have kept us from choosing this block
623 * group, but if we've moved to the state where we will wait on caching
624 * block groups we need to first check if we're doing a fast load here,
625 * so we can wait for it to finish, otherwise we could end up allocating
626 * from a block group who's cache gets evicted for one reason or
629 while (cache->cached == BTRFS_CACHE_FAST) {
630 struct btrfs_caching_control *ctl;
632 ctl = cache->caching_ctl;
633 refcount_inc(&ctl->count);
634 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
635 spin_unlock(&cache->lock);
639 finish_wait(&ctl->wait, &wait);
640 put_caching_control(ctl);
641 spin_lock(&cache->lock);
644 if (cache->cached != BTRFS_CACHE_NO) {
645 spin_unlock(&cache->lock);
649 WARN_ON(cache->caching_ctl);
650 cache->caching_ctl = caching_ctl;
651 cache->cached = BTRFS_CACHE_FAST;
652 spin_unlock(&cache->lock);
654 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
655 mutex_lock(&caching_ctl->mutex);
656 ret = load_free_space_cache(fs_info, cache);
658 spin_lock(&cache->lock);
660 cache->caching_ctl = NULL;
661 cache->cached = BTRFS_CACHE_FINISHED;
662 cache->last_byte_to_unpin = (u64)-1;
663 caching_ctl->progress = (u64)-1;
665 if (load_cache_only) {
666 cache->caching_ctl = NULL;
667 cache->cached = BTRFS_CACHE_NO;
669 cache->cached = BTRFS_CACHE_STARTED;
670 cache->has_caching_ctl = 1;
673 spin_unlock(&cache->lock);
674 #ifdef CONFIG_BTRFS_DEBUG
676 btrfs_should_fragment_free_space(cache)) {
679 spin_lock(&cache->space_info->lock);
680 spin_lock(&cache->lock);
681 bytes_used = cache->key.offset -
682 btrfs_block_group_used(&cache->item);
683 cache->space_info->bytes_used += bytes_used >> 1;
684 spin_unlock(&cache->lock);
685 spin_unlock(&cache->space_info->lock);
686 fragment_free_space(cache);
689 mutex_unlock(&caching_ctl->mutex);
691 wake_up(&caching_ctl->wait);
693 put_caching_control(caching_ctl);
694 free_excluded_extents(fs_info, cache);
699 * We're either using the free space tree or no caching at all.
700 * Set cached to the appropriate value and wakeup any waiters.
702 spin_lock(&cache->lock);
703 if (load_cache_only) {
704 cache->caching_ctl = NULL;
705 cache->cached = BTRFS_CACHE_NO;
707 cache->cached = BTRFS_CACHE_STARTED;
708 cache->has_caching_ctl = 1;
710 spin_unlock(&cache->lock);
711 wake_up(&caching_ctl->wait);
714 if (load_cache_only) {
715 put_caching_control(caching_ctl);
719 down_write(&fs_info->commit_root_sem);
720 refcount_inc(&caching_ctl->count);
721 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
722 up_write(&fs_info->commit_root_sem);
724 btrfs_get_block_group(cache);
726 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
732 * return the block group that starts at or after bytenr
734 static struct btrfs_block_group_cache *
735 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
737 return block_group_cache_tree_search(info, bytenr, 0);
741 * return the block group that contains the given bytenr
743 struct btrfs_block_group_cache *btrfs_lookup_block_group(
744 struct btrfs_fs_info *info,
747 return block_group_cache_tree_search(info, bytenr, 1);
750 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
753 struct list_head *head = &info->space_info;
754 struct btrfs_space_info *found;
756 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
759 list_for_each_entry_rcu(found, head, list) {
760 if (found->flags & flags) {
770 * after adding space to the filesystem, we need to clear the full flags
771 * on all the space infos.
773 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
775 struct list_head *head = &info->space_info;
776 struct btrfs_space_info *found;
779 list_for_each_entry_rcu(found, head, list)
784 /* simple helper to search for an existing data extent at a given offset */
785 int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len)
788 struct btrfs_key key;
789 struct btrfs_path *path;
791 path = btrfs_alloc_path();
795 key.objectid = start;
797 key.type = BTRFS_EXTENT_ITEM_KEY;
798 ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
799 btrfs_free_path(path);
804 * helper function to lookup reference count and flags of a tree block.
806 * the head node for delayed ref is used to store the sum of all the
807 * reference count modifications queued up in the rbtree. the head
808 * node may also store the extent flags to set. This way you can check
809 * to see what the reference count and extent flags would be if all of
810 * the delayed refs are not processed.
812 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
813 struct btrfs_fs_info *fs_info, u64 bytenr,
814 u64 offset, int metadata, u64 *refs, u64 *flags)
816 struct btrfs_delayed_ref_head *head;
817 struct btrfs_delayed_ref_root *delayed_refs;
818 struct btrfs_path *path;
819 struct btrfs_extent_item *ei;
820 struct extent_buffer *leaf;
821 struct btrfs_key key;
828 * If we don't have skinny metadata, don't bother doing anything
831 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) {
832 offset = fs_info->nodesize;
836 path = btrfs_alloc_path();
841 path->skip_locking = 1;
842 path->search_commit_root = 1;
846 key.objectid = bytenr;
849 key.type = BTRFS_METADATA_ITEM_KEY;
851 key.type = BTRFS_EXTENT_ITEM_KEY;
853 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
857 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
858 if (path->slots[0]) {
860 btrfs_item_key_to_cpu(path->nodes[0], &key,
862 if (key.objectid == bytenr &&
863 key.type == BTRFS_EXTENT_ITEM_KEY &&
864 key.offset == fs_info->nodesize)
870 leaf = path->nodes[0];
871 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
872 if (item_size >= sizeof(*ei)) {
873 ei = btrfs_item_ptr(leaf, path->slots[0],
874 struct btrfs_extent_item);
875 num_refs = btrfs_extent_refs(leaf, ei);
876 extent_flags = btrfs_extent_flags(leaf, ei);
878 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
879 struct btrfs_extent_item_v0 *ei0;
880 BUG_ON(item_size != sizeof(*ei0));
881 ei0 = btrfs_item_ptr(leaf, path->slots[0],
882 struct btrfs_extent_item_v0);
883 num_refs = btrfs_extent_refs_v0(leaf, ei0);
884 /* FIXME: this isn't correct for data */
885 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
890 BUG_ON(num_refs == 0);
900 delayed_refs = &trans->transaction->delayed_refs;
901 spin_lock(&delayed_refs->lock);
902 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
904 if (!mutex_trylock(&head->mutex)) {
905 refcount_inc(&head->node.refs);
906 spin_unlock(&delayed_refs->lock);
908 btrfs_release_path(path);
911 * Mutex was contended, block until it's released and try
914 mutex_lock(&head->mutex);
915 mutex_unlock(&head->mutex);
916 btrfs_put_delayed_ref(&head->node);
919 spin_lock(&head->lock);
920 if (head->extent_op && head->extent_op->update_flags)
921 extent_flags |= head->extent_op->flags_to_set;
923 BUG_ON(num_refs == 0);
925 num_refs += head->node.ref_mod;
926 spin_unlock(&head->lock);
927 mutex_unlock(&head->mutex);
929 spin_unlock(&delayed_refs->lock);
931 WARN_ON(num_refs == 0);
935 *flags = extent_flags;
937 btrfs_free_path(path);
942 * Back reference rules. Back refs have three main goals:
944 * 1) differentiate between all holders of references to an extent so that
945 * when a reference is dropped we can make sure it was a valid reference
946 * before freeing the extent.
948 * 2) Provide enough information to quickly find the holders of an extent
949 * if we notice a given block is corrupted or bad.
951 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
952 * maintenance. This is actually the same as #2, but with a slightly
953 * different use case.
955 * There are two kinds of back refs. The implicit back refs is optimized
956 * for pointers in non-shared tree blocks. For a given pointer in a block,
957 * back refs of this kind provide information about the block's owner tree
958 * and the pointer's key. These information allow us to find the block by
959 * b-tree searching. The full back refs is for pointers in tree blocks not
960 * referenced by their owner trees. The location of tree block is recorded
961 * in the back refs. Actually the full back refs is generic, and can be
962 * used in all cases the implicit back refs is used. The major shortcoming
963 * of the full back refs is its overhead. Every time a tree block gets
964 * COWed, we have to update back refs entry for all pointers in it.
966 * For a newly allocated tree block, we use implicit back refs for
967 * pointers in it. This means most tree related operations only involve
968 * implicit back refs. For a tree block created in old transaction, the
969 * only way to drop a reference to it is COW it. So we can detect the
970 * event that tree block loses its owner tree's reference and do the
971 * back refs conversion.
973 * When a tree block is COWed through a tree, there are four cases:
975 * The reference count of the block is one and the tree is the block's
976 * owner tree. Nothing to do in this case.
978 * The reference count of the block is one and the tree is not the
979 * block's owner tree. In this case, full back refs is used for pointers
980 * in the block. Remove these full back refs, add implicit back refs for
981 * every pointers in the new block.
983 * The reference count of the block is greater than one and the tree is
984 * the block's owner tree. In this case, implicit back refs is used for
985 * pointers in the block. Add full back refs for every pointers in the
986 * block, increase lower level extents' reference counts. The original
987 * implicit back refs are entailed to the new block.
989 * The reference count of the block is greater than one and the tree is
990 * not the block's owner tree. Add implicit back refs for every pointer in
991 * the new block, increase lower level extents' reference count.
993 * Back Reference Key composing:
995 * The key objectid corresponds to the first byte in the extent,
996 * The key type is used to differentiate between types of back refs.
997 * There are different meanings of the key offset for different types
1000 * File extents can be referenced by:
1002 * - multiple snapshots, subvolumes, or different generations in one subvol
1003 * - different files inside a single subvolume
1004 * - different offsets inside a file (bookend extents in file.c)
1006 * The extent ref structure for the implicit back refs has fields for:
1008 * - Objectid of the subvolume root
1009 * - objectid of the file holding the reference
1010 * - original offset in the file
1011 * - how many bookend extents
1013 * The key offset for the implicit back refs is hash of the first
1016 * The extent ref structure for the full back refs has field for:
1018 * - number of pointers in the tree leaf
1020 * The key offset for the implicit back refs is the first byte of
1023 * When a file extent is allocated, The implicit back refs is used.
1024 * the fields are filled in:
1026 * (root_key.objectid, inode objectid, offset in file, 1)
1028 * When a file extent is removed file truncation, we find the
1029 * corresponding implicit back refs and check the following fields:
1031 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1033 * Btree extents can be referenced by:
1035 * - Different subvolumes
1037 * Both the implicit back refs and the full back refs for tree blocks
1038 * only consist of key. The key offset for the implicit back refs is
1039 * objectid of block's owner tree. The key offset for the full back refs
1040 * is the first byte of parent block.
1042 * When implicit back refs is used, information about the lowest key and
1043 * level of the tree block are required. These information are stored in
1044 * tree block info structure.
1047 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1048 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1049 struct btrfs_fs_info *fs_info,
1050 struct btrfs_path *path,
1051 u64 owner, u32 extra_size)
1053 struct btrfs_root *root = fs_info->extent_root;
1054 struct btrfs_extent_item *item;
1055 struct btrfs_extent_item_v0 *ei0;
1056 struct btrfs_extent_ref_v0 *ref0;
1057 struct btrfs_tree_block_info *bi;
1058 struct extent_buffer *leaf;
1059 struct btrfs_key key;
1060 struct btrfs_key found_key;
1061 u32 new_size = sizeof(*item);
1065 leaf = path->nodes[0];
1066 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1068 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1069 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1070 struct btrfs_extent_item_v0);
1071 refs = btrfs_extent_refs_v0(leaf, ei0);
1073 if (owner == (u64)-1) {
1075 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1076 ret = btrfs_next_leaf(root, path);
1079 BUG_ON(ret > 0); /* Corruption */
1080 leaf = path->nodes[0];
1082 btrfs_item_key_to_cpu(leaf, &found_key,
1084 BUG_ON(key.objectid != found_key.objectid);
1085 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1089 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1090 struct btrfs_extent_ref_v0);
1091 owner = btrfs_ref_objectid_v0(leaf, ref0);
1095 btrfs_release_path(path);
1097 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1098 new_size += sizeof(*bi);
1100 new_size -= sizeof(*ei0);
1101 ret = btrfs_search_slot(trans, root, &key, path,
1102 new_size + extra_size, 1);
1105 BUG_ON(ret); /* Corruption */
1107 btrfs_extend_item(fs_info, path, new_size);
1109 leaf = path->nodes[0];
1110 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1111 btrfs_set_extent_refs(leaf, item, refs);
1112 /* FIXME: get real generation */
1113 btrfs_set_extent_generation(leaf, item, 0);
1114 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1115 btrfs_set_extent_flags(leaf, item,
1116 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1117 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1118 bi = (struct btrfs_tree_block_info *)(item + 1);
1119 /* FIXME: get first key of the block */
1120 memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi));
1121 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1123 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1125 btrfs_mark_buffer_dirty(leaf);
1130 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1132 u32 high_crc = ~(u32)0;
1133 u32 low_crc = ~(u32)0;
1136 lenum = cpu_to_le64(root_objectid);
1137 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1138 lenum = cpu_to_le64(owner);
1139 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1140 lenum = cpu_to_le64(offset);
1141 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1143 return ((u64)high_crc << 31) ^ (u64)low_crc;
1146 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1147 struct btrfs_extent_data_ref *ref)
1149 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1150 btrfs_extent_data_ref_objectid(leaf, ref),
1151 btrfs_extent_data_ref_offset(leaf, ref));
1154 static int match_extent_data_ref(struct extent_buffer *leaf,
1155 struct btrfs_extent_data_ref *ref,
1156 u64 root_objectid, u64 owner, u64 offset)
1158 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1159 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1160 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1165 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1166 struct btrfs_fs_info *fs_info,
1167 struct btrfs_path *path,
1168 u64 bytenr, u64 parent,
1170 u64 owner, u64 offset)
1172 struct btrfs_root *root = fs_info->extent_root;
1173 struct btrfs_key key;
1174 struct btrfs_extent_data_ref *ref;
1175 struct extent_buffer *leaf;
1181 key.objectid = bytenr;
1183 key.type = BTRFS_SHARED_DATA_REF_KEY;
1184 key.offset = parent;
1186 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1187 key.offset = hash_extent_data_ref(root_objectid,
1192 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1201 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1202 key.type = BTRFS_EXTENT_REF_V0_KEY;
1203 btrfs_release_path(path);
1204 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1215 leaf = path->nodes[0];
1216 nritems = btrfs_header_nritems(leaf);
1218 if (path->slots[0] >= nritems) {
1219 ret = btrfs_next_leaf(root, path);
1225 leaf = path->nodes[0];
1226 nritems = btrfs_header_nritems(leaf);
1230 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1231 if (key.objectid != bytenr ||
1232 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1235 ref = btrfs_item_ptr(leaf, path->slots[0],
1236 struct btrfs_extent_data_ref);
1238 if (match_extent_data_ref(leaf, ref, root_objectid,
1241 btrfs_release_path(path);
1253 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1254 struct btrfs_fs_info *fs_info,
1255 struct btrfs_path *path,
1256 u64 bytenr, u64 parent,
1257 u64 root_objectid, u64 owner,
1258 u64 offset, int refs_to_add)
1260 struct btrfs_root *root = fs_info->extent_root;
1261 struct btrfs_key key;
1262 struct extent_buffer *leaf;
1267 key.objectid = bytenr;
1269 key.type = BTRFS_SHARED_DATA_REF_KEY;
1270 key.offset = parent;
1271 size = sizeof(struct btrfs_shared_data_ref);
1273 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1274 key.offset = hash_extent_data_ref(root_objectid,
1276 size = sizeof(struct btrfs_extent_data_ref);
1279 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1280 if (ret && ret != -EEXIST)
1283 leaf = path->nodes[0];
1285 struct btrfs_shared_data_ref *ref;
1286 ref = btrfs_item_ptr(leaf, path->slots[0],
1287 struct btrfs_shared_data_ref);
1289 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1291 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1292 num_refs += refs_to_add;
1293 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1296 struct btrfs_extent_data_ref *ref;
1297 while (ret == -EEXIST) {
1298 ref = btrfs_item_ptr(leaf, path->slots[0],
1299 struct btrfs_extent_data_ref);
1300 if (match_extent_data_ref(leaf, ref, root_objectid,
1303 btrfs_release_path(path);
1305 ret = btrfs_insert_empty_item(trans, root, path, &key,
1307 if (ret && ret != -EEXIST)
1310 leaf = path->nodes[0];
1312 ref = btrfs_item_ptr(leaf, path->slots[0],
1313 struct btrfs_extent_data_ref);
1315 btrfs_set_extent_data_ref_root(leaf, ref,
1317 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1318 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1319 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1321 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1322 num_refs += refs_to_add;
1323 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1326 btrfs_mark_buffer_dirty(leaf);
1329 btrfs_release_path(path);
1333 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1334 struct btrfs_fs_info *fs_info,
1335 struct btrfs_path *path,
1336 int refs_to_drop, int *last_ref)
1338 struct btrfs_key key;
1339 struct btrfs_extent_data_ref *ref1 = NULL;
1340 struct btrfs_shared_data_ref *ref2 = NULL;
1341 struct extent_buffer *leaf;
1345 leaf = path->nodes[0];
1346 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1348 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1349 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1350 struct btrfs_extent_data_ref);
1351 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1352 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1353 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1354 struct btrfs_shared_data_ref);
1355 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1356 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1357 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1358 struct btrfs_extent_ref_v0 *ref0;
1359 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1360 struct btrfs_extent_ref_v0);
1361 num_refs = btrfs_ref_count_v0(leaf, ref0);
1367 BUG_ON(num_refs < refs_to_drop);
1368 num_refs -= refs_to_drop;
1370 if (num_refs == 0) {
1371 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1374 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1375 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1376 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1377 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1378 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1380 struct btrfs_extent_ref_v0 *ref0;
1381 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1382 struct btrfs_extent_ref_v0);
1383 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1386 btrfs_mark_buffer_dirty(leaf);
1391 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1392 struct btrfs_extent_inline_ref *iref)
1394 struct btrfs_key key;
1395 struct extent_buffer *leaf;
1396 struct btrfs_extent_data_ref *ref1;
1397 struct btrfs_shared_data_ref *ref2;
1400 leaf = path->nodes[0];
1401 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1403 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1404 BTRFS_EXTENT_DATA_REF_KEY) {
1405 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1406 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1408 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1409 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1411 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1412 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1413 struct btrfs_extent_data_ref);
1414 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1415 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1416 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1417 struct btrfs_shared_data_ref);
1418 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1419 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1420 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1421 struct btrfs_extent_ref_v0 *ref0;
1422 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1423 struct btrfs_extent_ref_v0);
1424 num_refs = btrfs_ref_count_v0(leaf, ref0);
1432 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1433 struct btrfs_fs_info *fs_info,
1434 struct btrfs_path *path,
1435 u64 bytenr, u64 parent,
1438 struct btrfs_root *root = fs_info->extent_root;
1439 struct btrfs_key key;
1442 key.objectid = bytenr;
1444 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1445 key.offset = parent;
1447 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1448 key.offset = root_objectid;
1451 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1454 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1455 if (ret == -ENOENT && parent) {
1456 btrfs_release_path(path);
1457 key.type = BTRFS_EXTENT_REF_V0_KEY;
1458 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1466 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1467 struct btrfs_fs_info *fs_info,
1468 struct btrfs_path *path,
1469 u64 bytenr, u64 parent,
1472 struct btrfs_key key;
1475 key.objectid = bytenr;
1477 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1478 key.offset = parent;
1480 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1481 key.offset = root_objectid;
1484 ret = btrfs_insert_empty_item(trans, fs_info->extent_root,
1486 btrfs_release_path(path);
1490 static inline int extent_ref_type(u64 parent, u64 owner)
1493 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1495 type = BTRFS_SHARED_BLOCK_REF_KEY;
1497 type = BTRFS_TREE_BLOCK_REF_KEY;
1500 type = BTRFS_SHARED_DATA_REF_KEY;
1502 type = BTRFS_EXTENT_DATA_REF_KEY;
1507 static int find_next_key(struct btrfs_path *path, int level,
1508 struct btrfs_key *key)
1511 for (; level < BTRFS_MAX_LEVEL; level++) {
1512 if (!path->nodes[level])
1514 if (path->slots[level] + 1 >=
1515 btrfs_header_nritems(path->nodes[level]))
1518 btrfs_item_key_to_cpu(path->nodes[level], key,
1519 path->slots[level] + 1);
1521 btrfs_node_key_to_cpu(path->nodes[level], key,
1522 path->slots[level] + 1);
1529 * look for inline back ref. if back ref is found, *ref_ret is set
1530 * to the address of inline back ref, and 0 is returned.
1532 * if back ref isn't found, *ref_ret is set to the address where it
1533 * should be inserted, and -ENOENT is returned.
1535 * if insert is true and there are too many inline back refs, the path
1536 * points to the extent item, and -EAGAIN is returned.
1538 * NOTE: inline back refs are ordered in the same way that back ref
1539 * items in the tree are ordered.
1541 static noinline_for_stack
1542 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1543 struct btrfs_fs_info *fs_info,
1544 struct btrfs_path *path,
1545 struct btrfs_extent_inline_ref **ref_ret,
1546 u64 bytenr, u64 num_bytes,
1547 u64 parent, u64 root_objectid,
1548 u64 owner, u64 offset, int insert)
1550 struct btrfs_root *root = fs_info->extent_root;
1551 struct btrfs_key key;
1552 struct extent_buffer *leaf;
1553 struct btrfs_extent_item *ei;
1554 struct btrfs_extent_inline_ref *iref;
1564 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
1566 key.objectid = bytenr;
1567 key.type = BTRFS_EXTENT_ITEM_KEY;
1568 key.offset = num_bytes;
1570 want = extent_ref_type(parent, owner);
1572 extra_size = btrfs_extent_inline_ref_size(want);
1573 path->keep_locks = 1;
1578 * Owner is our parent level, so we can just add one to get the level
1579 * for the block we are interested in.
1581 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1582 key.type = BTRFS_METADATA_ITEM_KEY;
1587 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1594 * We may be a newly converted file system which still has the old fat
1595 * extent entries for metadata, so try and see if we have one of those.
1597 if (ret > 0 && skinny_metadata) {
1598 skinny_metadata = false;
1599 if (path->slots[0]) {
1601 btrfs_item_key_to_cpu(path->nodes[0], &key,
1603 if (key.objectid == bytenr &&
1604 key.type == BTRFS_EXTENT_ITEM_KEY &&
1605 key.offset == num_bytes)
1609 key.objectid = bytenr;
1610 key.type = BTRFS_EXTENT_ITEM_KEY;
1611 key.offset = num_bytes;
1612 btrfs_release_path(path);
1617 if (ret && !insert) {
1620 } else if (WARN_ON(ret)) {
1625 leaf = path->nodes[0];
1626 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1627 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1628 if (item_size < sizeof(*ei)) {
1633 ret = convert_extent_item_v0(trans, fs_info, path, owner,
1639 leaf = path->nodes[0];
1640 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1643 BUG_ON(item_size < sizeof(*ei));
1645 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1646 flags = btrfs_extent_flags(leaf, ei);
1648 ptr = (unsigned long)(ei + 1);
1649 end = (unsigned long)ei + item_size;
1651 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1652 ptr += sizeof(struct btrfs_tree_block_info);
1662 iref = (struct btrfs_extent_inline_ref *)ptr;
1663 type = btrfs_extent_inline_ref_type(leaf, iref);
1667 ptr += btrfs_extent_inline_ref_size(type);
1671 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1672 struct btrfs_extent_data_ref *dref;
1673 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1674 if (match_extent_data_ref(leaf, dref, root_objectid,
1679 if (hash_extent_data_ref_item(leaf, dref) <
1680 hash_extent_data_ref(root_objectid, owner, offset))
1684 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1686 if (parent == ref_offset) {
1690 if (ref_offset < parent)
1693 if (root_objectid == ref_offset) {
1697 if (ref_offset < root_objectid)
1701 ptr += btrfs_extent_inline_ref_size(type);
1703 if (err == -ENOENT && insert) {
1704 if (item_size + extra_size >=
1705 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1710 * To add new inline back ref, we have to make sure
1711 * there is no corresponding back ref item.
1712 * For simplicity, we just do not add new inline back
1713 * ref if there is any kind of item for this block
1715 if (find_next_key(path, 0, &key) == 0 &&
1716 key.objectid == bytenr &&
1717 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1722 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1725 path->keep_locks = 0;
1726 btrfs_unlock_up_safe(path, 1);
1732 * helper to add new inline back ref
1734 static noinline_for_stack
1735 void setup_inline_extent_backref(struct btrfs_fs_info *fs_info,
1736 struct btrfs_path *path,
1737 struct btrfs_extent_inline_ref *iref,
1738 u64 parent, u64 root_objectid,
1739 u64 owner, u64 offset, int refs_to_add,
1740 struct btrfs_delayed_extent_op *extent_op)
1742 struct extent_buffer *leaf;
1743 struct btrfs_extent_item *ei;
1746 unsigned long item_offset;
1751 leaf = path->nodes[0];
1752 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1753 item_offset = (unsigned long)iref - (unsigned long)ei;
1755 type = extent_ref_type(parent, owner);
1756 size = btrfs_extent_inline_ref_size(type);
1758 btrfs_extend_item(fs_info, path, size);
1760 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1761 refs = btrfs_extent_refs(leaf, ei);
1762 refs += refs_to_add;
1763 btrfs_set_extent_refs(leaf, ei, refs);
1765 __run_delayed_extent_op(extent_op, leaf, ei);
1767 ptr = (unsigned long)ei + item_offset;
1768 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1769 if (ptr < end - size)
1770 memmove_extent_buffer(leaf, ptr + size, ptr,
1773 iref = (struct btrfs_extent_inline_ref *)ptr;
1774 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1775 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1776 struct btrfs_extent_data_ref *dref;
1777 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1778 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1779 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1780 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1781 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1782 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1783 struct btrfs_shared_data_ref *sref;
1784 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1785 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1786 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1787 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1788 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1790 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1792 btrfs_mark_buffer_dirty(leaf);
1795 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1796 struct btrfs_fs_info *fs_info,
1797 struct btrfs_path *path,
1798 struct btrfs_extent_inline_ref **ref_ret,
1799 u64 bytenr, u64 num_bytes, u64 parent,
1800 u64 root_objectid, u64 owner, u64 offset)
1804 ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret,
1805 bytenr, num_bytes, parent,
1806 root_objectid, owner, offset, 0);
1810 btrfs_release_path(path);
1813 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1814 ret = lookup_tree_block_ref(trans, fs_info, path, bytenr,
1815 parent, root_objectid);
1817 ret = lookup_extent_data_ref(trans, fs_info, path, bytenr,
1818 parent, root_objectid, owner,
1825 * helper to update/remove inline back ref
1827 static noinline_for_stack
1828 void update_inline_extent_backref(struct btrfs_fs_info *fs_info,
1829 struct btrfs_path *path,
1830 struct btrfs_extent_inline_ref *iref,
1832 struct btrfs_delayed_extent_op *extent_op,
1835 struct extent_buffer *leaf;
1836 struct btrfs_extent_item *ei;
1837 struct btrfs_extent_data_ref *dref = NULL;
1838 struct btrfs_shared_data_ref *sref = NULL;
1846 leaf = path->nodes[0];
1847 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1848 refs = btrfs_extent_refs(leaf, ei);
1849 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1850 refs += refs_to_mod;
1851 btrfs_set_extent_refs(leaf, ei, refs);
1853 __run_delayed_extent_op(extent_op, leaf, ei);
1855 type = btrfs_extent_inline_ref_type(leaf, iref);
1857 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1858 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1859 refs = btrfs_extent_data_ref_count(leaf, dref);
1860 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1861 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1862 refs = btrfs_shared_data_ref_count(leaf, sref);
1865 BUG_ON(refs_to_mod != -1);
1868 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1869 refs += refs_to_mod;
1872 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1873 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1875 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1878 size = btrfs_extent_inline_ref_size(type);
1879 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1880 ptr = (unsigned long)iref;
1881 end = (unsigned long)ei + item_size;
1882 if (ptr + size < end)
1883 memmove_extent_buffer(leaf, ptr, ptr + size,
1886 btrfs_truncate_item(fs_info, path, item_size, 1);
1888 btrfs_mark_buffer_dirty(leaf);
1891 static noinline_for_stack
1892 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1893 struct btrfs_fs_info *fs_info,
1894 struct btrfs_path *path,
1895 u64 bytenr, u64 num_bytes, u64 parent,
1896 u64 root_objectid, u64 owner,
1897 u64 offset, int refs_to_add,
1898 struct btrfs_delayed_extent_op *extent_op)
1900 struct btrfs_extent_inline_ref *iref;
1903 ret = lookup_inline_extent_backref(trans, fs_info, path, &iref,
1904 bytenr, num_bytes, parent,
1905 root_objectid, owner, offset, 1);
1907 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1908 update_inline_extent_backref(fs_info, path, iref,
1909 refs_to_add, extent_op, NULL);
1910 } else if (ret == -ENOENT) {
1911 setup_inline_extent_backref(fs_info, path, iref, parent,
1912 root_objectid, owner, offset,
1913 refs_to_add, extent_op);
1919 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1920 struct btrfs_fs_info *fs_info,
1921 struct btrfs_path *path,
1922 u64 bytenr, u64 parent, u64 root_objectid,
1923 u64 owner, u64 offset, int refs_to_add)
1926 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1927 BUG_ON(refs_to_add != 1);
1928 ret = insert_tree_block_ref(trans, fs_info, path, bytenr,
1929 parent, root_objectid);
1931 ret = insert_extent_data_ref(trans, fs_info, path, bytenr,
1932 parent, root_objectid,
1933 owner, offset, refs_to_add);
1938 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1939 struct btrfs_fs_info *fs_info,
1940 struct btrfs_path *path,
1941 struct btrfs_extent_inline_ref *iref,
1942 int refs_to_drop, int is_data, int *last_ref)
1946 BUG_ON(!is_data && refs_to_drop != 1);
1948 update_inline_extent_backref(fs_info, path, iref,
1949 -refs_to_drop, NULL, last_ref);
1950 } else if (is_data) {
1951 ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop,
1955 ret = btrfs_del_item(trans, fs_info->extent_root, path);
1960 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1961 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1962 u64 *discarded_bytes)
1965 u64 bytes_left, end;
1966 u64 aligned_start = ALIGN(start, 1 << 9);
1968 if (WARN_ON(start != aligned_start)) {
1969 len -= aligned_start - start;
1970 len = round_down(len, 1 << 9);
1971 start = aligned_start;
1974 *discarded_bytes = 0;
1982 /* Skip any superblocks on this device. */
1983 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1984 u64 sb_start = btrfs_sb_offset(j);
1985 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1986 u64 size = sb_start - start;
1988 if (!in_range(sb_start, start, bytes_left) &&
1989 !in_range(sb_end, start, bytes_left) &&
1990 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1994 * Superblock spans beginning of range. Adjust start and
1997 if (sb_start <= start) {
1998 start += sb_end - start;
2003 bytes_left = end - start;
2008 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2011 *discarded_bytes += size;
2012 else if (ret != -EOPNOTSUPP)
2021 bytes_left = end - start;
2025 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2028 *discarded_bytes += bytes_left;
2033 int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr,
2034 u64 num_bytes, u64 *actual_bytes)
2037 u64 discarded_bytes = 0;
2038 struct btrfs_bio *bbio = NULL;
2042 * Avoid races with device replace and make sure our bbio has devices
2043 * associated to its stripes that don't go away while we are discarding.
2045 btrfs_bio_counter_inc_blocked(fs_info);
2046 /* Tell the block device(s) that the sectors can be discarded */
2047 ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes,
2049 /* Error condition is -ENOMEM */
2051 struct btrfs_bio_stripe *stripe = bbio->stripes;
2055 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2057 if (!stripe->dev->can_discard)
2060 ret = btrfs_issue_discard(stripe->dev->bdev,
2065 discarded_bytes += bytes;
2066 else if (ret != -EOPNOTSUPP)
2067 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2070 * Just in case we get back EOPNOTSUPP for some reason,
2071 * just ignore the return value so we don't screw up
2072 * people calling discard_extent.
2076 btrfs_put_bbio(bbio);
2078 btrfs_bio_counter_dec(fs_info);
2081 *actual_bytes = discarded_bytes;
2084 if (ret == -EOPNOTSUPP)
2089 /* Can return -ENOMEM */
2090 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2091 struct btrfs_fs_info *fs_info,
2092 u64 bytenr, u64 num_bytes, u64 parent,
2093 u64 root_objectid, u64 owner, u64 offset)
2097 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2098 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2100 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2101 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2103 parent, root_objectid, (int)owner,
2104 BTRFS_ADD_DELAYED_REF, NULL);
2106 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2107 num_bytes, parent, root_objectid,
2109 BTRFS_ADD_DELAYED_REF);
2114 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2115 struct btrfs_fs_info *fs_info,
2116 struct btrfs_delayed_ref_node *node,
2117 u64 parent, u64 root_objectid,
2118 u64 owner, u64 offset, int refs_to_add,
2119 struct btrfs_delayed_extent_op *extent_op)
2121 struct btrfs_path *path;
2122 struct extent_buffer *leaf;
2123 struct btrfs_extent_item *item;
2124 struct btrfs_key key;
2125 u64 bytenr = node->bytenr;
2126 u64 num_bytes = node->num_bytes;
2130 path = btrfs_alloc_path();
2134 path->reada = READA_FORWARD;
2135 path->leave_spinning = 1;
2136 /* this will setup the path even if it fails to insert the back ref */
2137 ret = insert_inline_extent_backref(trans, fs_info, path, bytenr,
2138 num_bytes, parent, root_objectid,
2140 refs_to_add, extent_op);
2141 if ((ret < 0 && ret != -EAGAIN) || !ret)
2145 * Ok we had -EAGAIN which means we didn't have space to insert and
2146 * inline extent ref, so just update the reference count and add a
2149 leaf = path->nodes[0];
2150 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2151 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2152 refs = btrfs_extent_refs(leaf, item);
2153 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2155 __run_delayed_extent_op(extent_op, leaf, item);
2157 btrfs_mark_buffer_dirty(leaf);
2158 btrfs_release_path(path);
2160 path->reada = READA_FORWARD;
2161 path->leave_spinning = 1;
2162 /* now insert the actual backref */
2163 ret = insert_extent_backref(trans, fs_info, path, bytenr, parent,
2164 root_objectid, owner, offset, refs_to_add);
2166 btrfs_abort_transaction(trans, ret);
2168 btrfs_free_path(path);
2172 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2173 struct btrfs_fs_info *fs_info,
2174 struct btrfs_delayed_ref_node *node,
2175 struct btrfs_delayed_extent_op *extent_op,
2176 int insert_reserved)
2179 struct btrfs_delayed_data_ref *ref;
2180 struct btrfs_key ins;
2185 ins.objectid = node->bytenr;
2186 ins.offset = node->num_bytes;
2187 ins.type = BTRFS_EXTENT_ITEM_KEY;
2189 ref = btrfs_delayed_node_to_data_ref(node);
2190 trace_run_delayed_data_ref(fs_info, node, ref, node->action);
2192 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2193 parent = ref->parent;
2194 ref_root = ref->root;
2196 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2198 flags |= extent_op->flags_to_set;
2199 ret = alloc_reserved_file_extent(trans, fs_info,
2200 parent, ref_root, flags,
2201 ref->objectid, ref->offset,
2202 &ins, node->ref_mod);
2203 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2204 ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent,
2205 ref_root, ref->objectid,
2206 ref->offset, node->ref_mod,
2208 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2209 ret = __btrfs_free_extent(trans, fs_info, node, parent,
2210 ref_root, ref->objectid,
2211 ref->offset, node->ref_mod,
2219 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2220 struct extent_buffer *leaf,
2221 struct btrfs_extent_item *ei)
2223 u64 flags = btrfs_extent_flags(leaf, ei);
2224 if (extent_op->update_flags) {
2225 flags |= extent_op->flags_to_set;
2226 btrfs_set_extent_flags(leaf, ei, flags);
2229 if (extent_op->update_key) {
2230 struct btrfs_tree_block_info *bi;
2231 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2232 bi = (struct btrfs_tree_block_info *)(ei + 1);
2233 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2237 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2238 struct btrfs_fs_info *fs_info,
2239 struct btrfs_delayed_ref_node *node,
2240 struct btrfs_delayed_extent_op *extent_op)
2242 struct btrfs_key key;
2243 struct btrfs_path *path;
2244 struct btrfs_extent_item *ei;
2245 struct extent_buffer *leaf;
2249 int metadata = !extent_op->is_data;
2254 if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2257 path = btrfs_alloc_path();
2261 key.objectid = node->bytenr;
2264 key.type = BTRFS_METADATA_ITEM_KEY;
2265 key.offset = extent_op->level;
2267 key.type = BTRFS_EXTENT_ITEM_KEY;
2268 key.offset = node->num_bytes;
2272 path->reada = READA_FORWARD;
2273 path->leave_spinning = 1;
2274 ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1);
2281 if (path->slots[0] > 0) {
2283 btrfs_item_key_to_cpu(path->nodes[0], &key,
2285 if (key.objectid == node->bytenr &&
2286 key.type == BTRFS_EXTENT_ITEM_KEY &&
2287 key.offset == node->num_bytes)
2291 btrfs_release_path(path);
2294 key.objectid = node->bytenr;
2295 key.offset = node->num_bytes;
2296 key.type = BTRFS_EXTENT_ITEM_KEY;
2305 leaf = path->nodes[0];
2306 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2307 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2308 if (item_size < sizeof(*ei)) {
2309 ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0);
2314 leaf = path->nodes[0];
2315 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2318 BUG_ON(item_size < sizeof(*ei));
2319 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2320 __run_delayed_extent_op(extent_op, leaf, ei);
2322 btrfs_mark_buffer_dirty(leaf);
2324 btrfs_free_path(path);
2328 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2329 struct btrfs_fs_info *fs_info,
2330 struct btrfs_delayed_ref_node *node,
2331 struct btrfs_delayed_extent_op *extent_op,
2332 int insert_reserved)
2335 struct btrfs_delayed_tree_ref *ref;
2336 struct btrfs_key ins;
2339 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
2341 ref = btrfs_delayed_node_to_tree_ref(node);
2342 trace_run_delayed_tree_ref(fs_info, node, ref, node->action);
2344 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2345 parent = ref->parent;
2346 ref_root = ref->root;
2348 ins.objectid = node->bytenr;
2349 if (skinny_metadata) {
2350 ins.offset = ref->level;
2351 ins.type = BTRFS_METADATA_ITEM_KEY;
2353 ins.offset = node->num_bytes;
2354 ins.type = BTRFS_EXTENT_ITEM_KEY;
2357 if (node->ref_mod != 1) {
2359 "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu",
2360 node->bytenr, node->ref_mod, node->action, ref_root,
2364 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2365 BUG_ON(!extent_op || !extent_op->update_flags);
2366 ret = alloc_reserved_tree_block(trans, fs_info,
2368 extent_op->flags_to_set,
2371 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2372 ret = __btrfs_inc_extent_ref(trans, fs_info, node,
2376 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2377 ret = __btrfs_free_extent(trans, fs_info, node,
2379 ref->level, 0, 1, extent_op);
2386 /* helper function to actually process a single delayed ref entry */
2387 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2388 struct btrfs_fs_info *fs_info,
2389 struct btrfs_delayed_ref_node *node,
2390 struct btrfs_delayed_extent_op *extent_op,
2391 int insert_reserved)
2395 if (trans->aborted) {
2396 if (insert_reserved)
2397 btrfs_pin_extent(fs_info, node->bytenr,
2398 node->num_bytes, 1);
2402 if (btrfs_delayed_ref_is_head(node)) {
2403 struct btrfs_delayed_ref_head *head;
2405 * we've hit the end of the chain and we were supposed
2406 * to insert this extent into the tree. But, it got
2407 * deleted before we ever needed to insert it, so all
2408 * we have to do is clean up the accounting
2411 head = btrfs_delayed_node_to_head(node);
2412 trace_run_delayed_ref_head(fs_info, node, head, node->action);
2414 if (insert_reserved) {
2415 btrfs_pin_extent(fs_info, node->bytenr,
2416 node->num_bytes, 1);
2417 if (head->is_data) {
2418 ret = btrfs_del_csums(trans, fs_info,
2424 /* Also free its reserved qgroup space */
2425 btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root,
2426 head->qgroup_reserved);
2430 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2431 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2432 ret = run_delayed_tree_ref(trans, fs_info, node, extent_op,
2434 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2435 node->type == BTRFS_SHARED_DATA_REF_KEY)
2436 ret = run_delayed_data_ref(trans, fs_info, node, extent_op,
2443 static inline struct btrfs_delayed_ref_node *
2444 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2446 struct btrfs_delayed_ref_node *ref;
2448 if (list_empty(&head->ref_list))
2452 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2453 * This is to prevent a ref count from going down to zero, which deletes
2454 * the extent item from the extent tree, when there still are references
2455 * to add, which would fail because they would not find the extent item.
2457 if (!list_empty(&head->ref_add_list))
2458 return list_first_entry(&head->ref_add_list,
2459 struct btrfs_delayed_ref_node, add_list);
2461 ref = list_first_entry(&head->ref_list, struct btrfs_delayed_ref_node,
2463 ASSERT(list_empty(&ref->add_list));
2468 * Returns 0 on success or if called with an already aborted transaction.
2469 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2471 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2472 struct btrfs_fs_info *fs_info,
2475 struct btrfs_delayed_ref_root *delayed_refs;
2476 struct btrfs_delayed_ref_node *ref;
2477 struct btrfs_delayed_ref_head *locked_ref = NULL;
2478 struct btrfs_delayed_extent_op *extent_op;
2479 ktime_t start = ktime_get();
2481 unsigned long count = 0;
2482 unsigned long actual_count = 0;
2483 int must_insert_reserved = 0;
2485 delayed_refs = &trans->transaction->delayed_refs;
2491 spin_lock(&delayed_refs->lock);
2492 locked_ref = btrfs_select_ref_head(trans);
2494 spin_unlock(&delayed_refs->lock);
2498 /* grab the lock that says we are going to process
2499 * all the refs for this head */
2500 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2501 spin_unlock(&delayed_refs->lock);
2503 * we may have dropped the spin lock to get the head
2504 * mutex lock, and that might have given someone else
2505 * time to free the head. If that's true, it has been
2506 * removed from our list and we can move on.
2508 if (ret == -EAGAIN) {
2516 * We need to try and merge add/drops of the same ref since we
2517 * can run into issues with relocate dropping the implicit ref
2518 * and then it being added back again before the drop can
2519 * finish. If we merged anything we need to re-loop so we can
2521 * Or we can get node references of the same type that weren't
2522 * merged when created due to bumps in the tree mod seq, and
2523 * we need to merge them to prevent adding an inline extent
2524 * backref before dropping it (triggering a BUG_ON at
2525 * insert_inline_extent_backref()).
2527 spin_lock(&locked_ref->lock);
2528 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2532 * locked_ref is the head node, so we have to go one
2533 * node back for any delayed ref updates
2535 ref = select_delayed_ref(locked_ref);
2537 if (ref && ref->seq &&
2538 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2539 spin_unlock(&locked_ref->lock);
2540 spin_lock(&delayed_refs->lock);
2541 locked_ref->processing = 0;
2542 delayed_refs->num_heads_ready++;
2543 spin_unlock(&delayed_refs->lock);
2544 btrfs_delayed_ref_unlock(locked_ref);
2552 * record the must insert reserved flag before we
2553 * drop the spin lock.
2555 must_insert_reserved = locked_ref->must_insert_reserved;
2556 locked_ref->must_insert_reserved = 0;
2558 extent_op = locked_ref->extent_op;
2559 locked_ref->extent_op = NULL;
2564 /* All delayed refs have been processed, Go ahead
2565 * and send the head node to run_one_delayed_ref,
2566 * so that any accounting fixes can happen
2568 ref = &locked_ref->node;
2570 if (extent_op && must_insert_reserved) {
2571 btrfs_free_delayed_extent_op(extent_op);
2576 spin_unlock(&locked_ref->lock);
2577 ret = run_delayed_extent_op(trans, fs_info,
2579 btrfs_free_delayed_extent_op(extent_op);
2583 * Need to reset must_insert_reserved if
2584 * there was an error so the abort stuff
2585 * can cleanup the reserved space
2588 if (must_insert_reserved)
2589 locked_ref->must_insert_reserved = 1;
2590 spin_lock(&delayed_refs->lock);
2591 locked_ref->processing = 0;
2592 delayed_refs->num_heads_ready++;
2593 spin_unlock(&delayed_refs->lock);
2594 btrfs_debug(fs_info,
2595 "run_delayed_extent_op returned %d",
2597 btrfs_delayed_ref_unlock(locked_ref);
2604 * Need to drop our head ref lock and re-acquire the
2605 * delayed ref lock and then re-check to make sure
2608 spin_unlock(&locked_ref->lock);
2609 spin_lock(&delayed_refs->lock);
2610 spin_lock(&locked_ref->lock);
2611 if (!list_empty(&locked_ref->ref_list) ||
2612 locked_ref->extent_op) {
2613 spin_unlock(&locked_ref->lock);
2614 spin_unlock(&delayed_refs->lock);
2618 delayed_refs->num_heads--;
2619 rb_erase(&locked_ref->href_node,
2620 &delayed_refs->href_root);
2621 spin_unlock(&delayed_refs->lock);
2625 list_del(&ref->list);
2626 if (!list_empty(&ref->add_list))
2627 list_del(&ref->add_list);
2629 atomic_dec(&delayed_refs->num_entries);
2631 if (!btrfs_delayed_ref_is_head(ref)) {
2633 * when we play the delayed ref, also correct the
2636 switch (ref->action) {
2637 case BTRFS_ADD_DELAYED_REF:
2638 case BTRFS_ADD_DELAYED_EXTENT:
2639 locked_ref->node.ref_mod -= ref->ref_mod;
2641 case BTRFS_DROP_DELAYED_REF:
2642 locked_ref->node.ref_mod += ref->ref_mod;
2648 spin_unlock(&locked_ref->lock);
2650 ret = run_one_delayed_ref(trans, fs_info, ref, extent_op,
2651 must_insert_reserved);
2653 btrfs_free_delayed_extent_op(extent_op);
2655 spin_lock(&delayed_refs->lock);
2656 locked_ref->processing = 0;
2657 delayed_refs->num_heads_ready++;
2658 spin_unlock(&delayed_refs->lock);
2659 btrfs_delayed_ref_unlock(locked_ref);
2660 btrfs_put_delayed_ref(ref);
2661 btrfs_debug(fs_info, "run_one_delayed_ref returned %d",
2667 * If this node is a head, that means all the refs in this head
2668 * have been dealt with, and we will pick the next head to deal
2669 * with, so we must unlock the head and drop it from the cluster
2670 * list before we release it.
2672 if (btrfs_delayed_ref_is_head(ref)) {
2673 if (locked_ref->is_data &&
2674 locked_ref->total_ref_mod < 0) {
2675 spin_lock(&delayed_refs->lock);
2676 delayed_refs->pending_csums -= ref->num_bytes;
2677 spin_unlock(&delayed_refs->lock);
2679 btrfs_delayed_ref_unlock(locked_ref);
2682 btrfs_put_delayed_ref(ref);
2688 * We don't want to include ref heads since we can have empty ref heads
2689 * and those will drastically skew our runtime down since we just do
2690 * accounting, no actual extent tree updates.
2692 if (actual_count > 0) {
2693 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2697 * We weigh the current average higher than our current runtime
2698 * to avoid large swings in the average.
2700 spin_lock(&delayed_refs->lock);
2701 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2702 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2703 spin_unlock(&delayed_refs->lock);
2708 #ifdef SCRAMBLE_DELAYED_REFS
2710 * Normally delayed refs get processed in ascending bytenr order. This
2711 * correlates in most cases to the order added. To expose dependencies on this
2712 * order, we start to process the tree in the middle instead of the beginning
2714 static u64 find_middle(struct rb_root *root)
2716 struct rb_node *n = root->rb_node;
2717 struct btrfs_delayed_ref_node *entry;
2720 u64 first = 0, last = 0;
2724 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2725 first = entry->bytenr;
2729 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2730 last = entry->bytenr;
2735 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2736 WARN_ON(!entry->in_tree);
2738 middle = entry->bytenr;
2751 static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads)
2755 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2756 sizeof(struct btrfs_extent_inline_ref));
2757 if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2758 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2761 * We don't ever fill up leaves all the way so multiply by 2 just to be
2762 * closer to what we're really going to want to use.
2764 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info));
2768 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2769 * would require to store the csums for that many bytes.
2771 u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes)
2774 u64 num_csums_per_leaf;
2777 csum_size = BTRFS_MAX_ITEM_SIZE(fs_info);
2778 num_csums_per_leaf = div64_u64(csum_size,
2779 (u64)btrfs_super_csum_size(fs_info->super_copy));
2780 num_csums = div64_u64(csum_bytes, fs_info->sectorsize);
2781 num_csums += num_csums_per_leaf - 1;
2782 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2786 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2787 struct btrfs_fs_info *fs_info)
2789 struct btrfs_block_rsv *global_rsv;
2790 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2791 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2792 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2793 u64 num_bytes, num_dirty_bgs_bytes;
2796 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
2797 num_heads = heads_to_leaves(fs_info, num_heads);
2799 num_bytes += (num_heads - 1) * fs_info->nodesize;
2801 num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) *
2803 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info,
2805 global_rsv = &fs_info->global_block_rsv;
2808 * If we can't allocate any more chunks lets make sure we have _lots_ of
2809 * wiggle room since running delayed refs can create more delayed refs.
2811 if (global_rsv->space_info->full) {
2812 num_dirty_bgs_bytes <<= 1;
2816 spin_lock(&global_rsv->lock);
2817 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2819 spin_unlock(&global_rsv->lock);
2823 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2824 struct btrfs_fs_info *fs_info)
2827 atomic_read(&trans->transaction->delayed_refs.num_entries);
2832 avg_runtime = fs_info->avg_delayed_ref_runtime;
2833 val = num_entries * avg_runtime;
2834 if (val >= NSEC_PER_SEC)
2836 if (val >= NSEC_PER_SEC / 2)
2839 return btrfs_check_space_for_delayed_refs(trans, fs_info);
2842 struct async_delayed_refs {
2843 struct btrfs_root *root;
2848 struct completion wait;
2849 struct btrfs_work work;
2852 static inline struct async_delayed_refs *
2853 to_async_delayed_refs(struct btrfs_work *work)
2855 return container_of(work, struct async_delayed_refs, work);
2858 static void delayed_ref_async_start(struct btrfs_work *work)
2860 struct async_delayed_refs *async = to_async_delayed_refs(work);
2861 struct btrfs_trans_handle *trans;
2862 struct btrfs_fs_info *fs_info = async->root->fs_info;
2865 /* if the commit is already started, we don't need to wait here */
2866 if (btrfs_transaction_blocked(fs_info))
2869 trans = btrfs_join_transaction(async->root);
2870 if (IS_ERR(trans)) {
2871 async->error = PTR_ERR(trans);
2876 * trans->sync means that when we call end_transaction, we won't
2877 * wait on delayed refs
2881 /* Don't bother flushing if we got into a different transaction */
2882 if (trans->transid > async->transid)
2885 ret = btrfs_run_delayed_refs(trans, fs_info, async->count);
2889 ret = btrfs_end_transaction(trans);
2890 if (ret && !async->error)
2894 complete(&async->wait);
2899 int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info,
2900 unsigned long count, u64 transid, int wait)
2902 struct async_delayed_refs *async;
2905 async = kmalloc(sizeof(*async), GFP_NOFS);
2909 async->root = fs_info->tree_root;
2910 async->count = count;
2912 async->transid = transid;
2917 init_completion(&async->wait);
2919 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2920 delayed_ref_async_start, NULL, NULL);
2922 btrfs_queue_work(fs_info->extent_workers, &async->work);
2925 wait_for_completion(&async->wait);
2934 * this starts processing the delayed reference count updates and
2935 * extent insertions we have queued up so far. count can be
2936 * 0, which means to process everything in the tree at the start
2937 * of the run (but not newly added entries), or it can be some target
2938 * number you'd like to process.
2940 * Returns 0 on success or if called with an aborted transaction
2941 * Returns <0 on error and aborts the transaction
2943 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2944 struct btrfs_fs_info *fs_info, unsigned long count)
2946 struct rb_node *node;
2947 struct btrfs_delayed_ref_root *delayed_refs;
2948 struct btrfs_delayed_ref_head *head;
2950 int run_all = count == (unsigned long)-1;
2951 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2953 /* We'll clean this up in btrfs_cleanup_transaction */
2957 if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags))
2960 delayed_refs = &trans->transaction->delayed_refs;
2962 count = atomic_read(&delayed_refs->num_entries) * 2;
2965 #ifdef SCRAMBLE_DELAYED_REFS
2966 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2968 trans->can_flush_pending_bgs = false;
2969 ret = __btrfs_run_delayed_refs(trans, fs_info, count);
2971 btrfs_abort_transaction(trans, ret);
2976 if (!list_empty(&trans->new_bgs))
2977 btrfs_create_pending_block_groups(trans, fs_info);
2979 spin_lock(&delayed_refs->lock);
2980 node = rb_first(&delayed_refs->href_root);
2982 spin_unlock(&delayed_refs->lock);
2987 head = rb_entry(node, struct btrfs_delayed_ref_head,
2989 if (btrfs_delayed_ref_is_head(&head->node)) {
2990 struct btrfs_delayed_ref_node *ref;
2993 refcount_inc(&ref->refs);
2995 spin_unlock(&delayed_refs->lock);
2997 * Mutex was contended, block until it's
2998 * released and try again
3000 mutex_lock(&head->mutex);
3001 mutex_unlock(&head->mutex);
3003 btrfs_put_delayed_ref(ref);
3009 node = rb_next(node);
3011 spin_unlock(&delayed_refs->lock);
3016 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3020 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3021 struct btrfs_fs_info *fs_info,
3022 u64 bytenr, u64 num_bytes, u64 flags,
3023 int level, int is_data)
3025 struct btrfs_delayed_extent_op *extent_op;
3028 extent_op = btrfs_alloc_delayed_extent_op();
3032 extent_op->flags_to_set = flags;
3033 extent_op->update_flags = true;
3034 extent_op->update_key = false;
3035 extent_op->is_data = is_data ? true : false;
3036 extent_op->level = level;
3038 ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr,
3039 num_bytes, extent_op);
3041 btrfs_free_delayed_extent_op(extent_op);
3045 static noinline int check_delayed_ref(struct btrfs_root *root,
3046 struct btrfs_path *path,
3047 u64 objectid, u64 offset, u64 bytenr)
3049 struct btrfs_delayed_ref_head *head;
3050 struct btrfs_delayed_ref_node *ref;
3051 struct btrfs_delayed_data_ref *data_ref;
3052 struct btrfs_delayed_ref_root *delayed_refs;
3053 struct btrfs_transaction *cur_trans;
3056 cur_trans = root->fs_info->running_transaction;
3060 delayed_refs = &cur_trans->delayed_refs;
3061 spin_lock(&delayed_refs->lock);
3062 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
3064 spin_unlock(&delayed_refs->lock);
3068 if (!mutex_trylock(&head->mutex)) {
3069 refcount_inc(&head->node.refs);
3070 spin_unlock(&delayed_refs->lock);
3072 btrfs_release_path(path);
3075 * Mutex was contended, block until it's released and let
3078 mutex_lock(&head->mutex);
3079 mutex_unlock(&head->mutex);
3080 btrfs_put_delayed_ref(&head->node);
3083 spin_unlock(&delayed_refs->lock);
3085 spin_lock(&head->lock);
3086 list_for_each_entry(ref, &head->ref_list, list) {
3087 /* If it's a shared ref we know a cross reference exists */
3088 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3093 data_ref = btrfs_delayed_node_to_data_ref(ref);
3096 * If our ref doesn't match the one we're currently looking at
3097 * then we have a cross reference.
3099 if (data_ref->root != root->root_key.objectid ||
3100 data_ref->objectid != objectid ||
3101 data_ref->offset != offset) {
3106 spin_unlock(&head->lock);
3107 mutex_unlock(&head->mutex);
3111 static noinline int check_committed_ref(struct btrfs_root *root,
3112 struct btrfs_path *path,
3113 u64 objectid, u64 offset, u64 bytenr)
3115 struct btrfs_fs_info *fs_info = root->fs_info;
3116 struct btrfs_root *extent_root = fs_info->extent_root;
3117 struct extent_buffer *leaf;
3118 struct btrfs_extent_data_ref *ref;
3119 struct btrfs_extent_inline_ref *iref;
3120 struct btrfs_extent_item *ei;
3121 struct btrfs_key key;
3125 key.objectid = bytenr;
3126 key.offset = (u64)-1;
3127 key.type = BTRFS_EXTENT_ITEM_KEY;
3129 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3132 BUG_ON(ret == 0); /* Corruption */
3135 if (path->slots[0] == 0)
3139 leaf = path->nodes[0];
3140 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3142 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3146 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3147 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3148 if (item_size < sizeof(*ei)) {
3149 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3153 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3155 if (item_size != sizeof(*ei) +
3156 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3159 if (btrfs_extent_generation(leaf, ei) <=
3160 btrfs_root_last_snapshot(&root->root_item))
3163 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3164 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3165 BTRFS_EXTENT_DATA_REF_KEY)
3168 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3169 if (btrfs_extent_refs(leaf, ei) !=
3170 btrfs_extent_data_ref_count(leaf, ref) ||
3171 btrfs_extent_data_ref_root(leaf, ref) !=
3172 root->root_key.objectid ||
3173 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3174 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3182 int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset,
3185 struct btrfs_path *path;
3189 path = btrfs_alloc_path();
3194 ret = check_committed_ref(root, path, objectid,
3196 if (ret && ret != -ENOENT)
3199 ret2 = check_delayed_ref(root, path, objectid,
3201 } while (ret2 == -EAGAIN);
3203 if (ret2 && ret2 != -ENOENT) {
3208 if (ret != -ENOENT || ret2 != -ENOENT)
3211 btrfs_free_path(path);
3212 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3217 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3218 struct btrfs_root *root,
3219 struct extent_buffer *buf,
3220 int full_backref, int inc)
3222 struct btrfs_fs_info *fs_info = root->fs_info;
3228 struct btrfs_key key;
3229 struct btrfs_file_extent_item *fi;
3233 int (*process_func)(struct btrfs_trans_handle *,
3234 struct btrfs_fs_info *,
3235 u64, u64, u64, u64, u64, u64);
3238 if (btrfs_is_testing(fs_info))
3241 ref_root = btrfs_header_owner(buf);
3242 nritems = btrfs_header_nritems(buf);
3243 level = btrfs_header_level(buf);
3245 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3249 process_func = btrfs_inc_extent_ref;
3251 process_func = btrfs_free_extent;
3254 parent = buf->start;
3258 for (i = 0; i < nritems; i++) {
3260 btrfs_item_key_to_cpu(buf, &key, i);
3261 if (key.type != BTRFS_EXTENT_DATA_KEY)
3263 fi = btrfs_item_ptr(buf, i,
3264 struct btrfs_file_extent_item);
3265 if (btrfs_file_extent_type(buf, fi) ==
3266 BTRFS_FILE_EXTENT_INLINE)
3268 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3272 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3273 key.offset -= btrfs_file_extent_offset(buf, fi);
3274 ret = process_func(trans, fs_info, bytenr, num_bytes,
3275 parent, ref_root, key.objectid,
3280 bytenr = btrfs_node_blockptr(buf, i);
3281 num_bytes = fs_info->nodesize;
3282 ret = process_func(trans, fs_info, bytenr, num_bytes,
3283 parent, ref_root, level - 1, 0);
3293 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3294 struct extent_buffer *buf, int full_backref)
3296 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3299 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3300 struct extent_buffer *buf, int full_backref)
3302 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3305 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3306 struct btrfs_fs_info *fs_info,
3307 struct btrfs_path *path,
3308 struct btrfs_block_group_cache *cache)
3311 struct btrfs_root *extent_root = fs_info->extent_root;
3313 struct extent_buffer *leaf;
3315 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3322 leaf = path->nodes[0];
3323 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3324 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3325 btrfs_mark_buffer_dirty(leaf);
3327 btrfs_release_path(path);
3332 static struct btrfs_block_group_cache *
3333 next_block_group(struct btrfs_fs_info *fs_info,
3334 struct btrfs_block_group_cache *cache)
3336 struct rb_node *node;
3338 spin_lock(&fs_info->block_group_cache_lock);
3340 /* If our block group was removed, we need a full search. */
3341 if (RB_EMPTY_NODE(&cache->cache_node)) {
3342 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3344 spin_unlock(&fs_info->block_group_cache_lock);
3345 btrfs_put_block_group(cache);
3346 cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache;
3348 node = rb_next(&cache->cache_node);
3349 btrfs_put_block_group(cache);
3351 cache = rb_entry(node, struct btrfs_block_group_cache,
3353 btrfs_get_block_group(cache);
3356 spin_unlock(&fs_info->block_group_cache_lock);
3360 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3361 struct btrfs_trans_handle *trans,
3362 struct btrfs_path *path)
3364 struct btrfs_fs_info *fs_info = block_group->fs_info;
3365 struct btrfs_root *root = fs_info->tree_root;
3366 struct inode *inode = NULL;
3368 int dcs = BTRFS_DC_ERROR;
3374 * If this block group is smaller than 100 megs don't bother caching the
3377 if (block_group->key.offset < (100 * SZ_1M)) {
3378 spin_lock(&block_group->lock);
3379 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3380 spin_unlock(&block_group->lock);
3387 inode = lookup_free_space_inode(fs_info, block_group, path);
3388 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3389 ret = PTR_ERR(inode);
3390 btrfs_release_path(path);
3394 if (IS_ERR(inode)) {
3398 if (block_group->ro)
3401 ret = create_free_space_inode(fs_info, trans, block_group,
3408 /* We've already setup this transaction, go ahead and exit */
3409 if (block_group->cache_generation == trans->transid &&
3410 i_size_read(inode)) {
3411 dcs = BTRFS_DC_SETUP;
3416 * We want to set the generation to 0, that way if anything goes wrong
3417 * from here on out we know not to trust this cache when we load up next
3420 BTRFS_I(inode)->generation = 0;
3421 ret = btrfs_update_inode(trans, root, inode);
3424 * So theoretically we could recover from this, simply set the
3425 * super cache generation to 0 so we know to invalidate the
3426 * cache, but then we'd have to keep track of the block groups
3427 * that fail this way so we know we _have_ to reset this cache
3428 * before the next commit or risk reading stale cache. So to
3429 * limit our exposure to horrible edge cases lets just abort the
3430 * transaction, this only happens in really bad situations
3433 btrfs_abort_transaction(trans, ret);
3438 if (i_size_read(inode) > 0) {
3439 ret = btrfs_check_trunc_cache_free_space(fs_info,
3440 &fs_info->global_block_rsv);
3444 ret = btrfs_truncate_free_space_cache(trans, NULL, inode);
3449 spin_lock(&block_group->lock);
3450 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3451 !btrfs_test_opt(fs_info, SPACE_CACHE)) {
3453 * don't bother trying to write stuff out _if_
3454 * a) we're not cached,
3455 * b) we're with nospace_cache mount option,
3456 * c) we're with v2 space_cache (FREE_SPACE_TREE).
3458 dcs = BTRFS_DC_WRITTEN;
3459 spin_unlock(&block_group->lock);
3462 spin_unlock(&block_group->lock);
3465 * We hit an ENOSPC when setting up the cache in this transaction, just
3466 * skip doing the setup, we've already cleared the cache so we're safe.
3468 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3474 * Try to preallocate enough space based on how big the block group is.
3475 * Keep in mind this has to include any pinned space which could end up
3476 * taking up quite a bit since it's not folded into the other space
3479 num_pages = div_u64(block_group->key.offset, SZ_256M);
3484 num_pages *= PAGE_SIZE;
3486 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3490 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3491 num_pages, num_pages,
3494 * Our cache requires contiguous chunks so that we don't modify a bunch
3495 * of metadata or split extents when writing the cache out, which means
3496 * we can enospc if we are heavily fragmented in addition to just normal
3497 * out of space conditions. So if we hit this just skip setting up any
3498 * other block groups for this transaction, maybe we'll unpin enough
3499 * space the next time around.
3502 dcs = BTRFS_DC_SETUP;
3503 else if (ret == -ENOSPC)
3504 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3509 btrfs_release_path(path);
3511 spin_lock(&block_group->lock);
3512 if (!ret && dcs == BTRFS_DC_SETUP)
3513 block_group->cache_generation = trans->transid;
3514 block_group->disk_cache_state = dcs;
3515 spin_unlock(&block_group->lock);
3520 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3521 struct btrfs_fs_info *fs_info)
3523 struct btrfs_block_group_cache *cache, *tmp;
3524 struct btrfs_transaction *cur_trans = trans->transaction;
3525 struct btrfs_path *path;
3527 if (list_empty(&cur_trans->dirty_bgs) ||
3528 !btrfs_test_opt(fs_info, SPACE_CACHE))
3531 path = btrfs_alloc_path();
3535 /* Could add new block groups, use _safe just in case */
3536 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3538 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3539 cache_save_setup(cache, trans, path);
3542 btrfs_free_path(path);
3547 * transaction commit does final block group cache writeback during a
3548 * critical section where nothing is allowed to change the FS. This is
3549 * required in order for the cache to actually match the block group,
3550 * but can introduce a lot of latency into the commit.
3552 * So, btrfs_start_dirty_block_groups is here to kick off block group
3553 * cache IO. There's a chance we'll have to redo some of it if the
3554 * block group changes again during the commit, but it greatly reduces
3555 * the commit latency by getting rid of the easy block groups while
3556 * we're still allowing others to join the commit.
3558 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3559 struct btrfs_fs_info *fs_info)
3561 struct btrfs_block_group_cache *cache;
3562 struct btrfs_transaction *cur_trans = trans->transaction;
3565 struct btrfs_path *path = NULL;
3567 struct list_head *io = &cur_trans->io_bgs;
3568 int num_started = 0;
3571 spin_lock(&cur_trans->dirty_bgs_lock);
3572 if (list_empty(&cur_trans->dirty_bgs)) {
3573 spin_unlock(&cur_trans->dirty_bgs_lock);
3576 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3577 spin_unlock(&cur_trans->dirty_bgs_lock);
3581 * make sure all the block groups on our dirty list actually
3584 btrfs_create_pending_block_groups(trans, fs_info);
3587 path = btrfs_alloc_path();
3593 * cache_write_mutex is here only to save us from balance or automatic
3594 * removal of empty block groups deleting this block group while we are
3595 * writing out the cache
3597 mutex_lock(&trans->transaction->cache_write_mutex);
3598 while (!list_empty(&dirty)) {
3599 cache = list_first_entry(&dirty,
3600 struct btrfs_block_group_cache,
3603 * this can happen if something re-dirties a block
3604 * group that is already under IO. Just wait for it to
3605 * finish and then do it all again
3607 if (!list_empty(&cache->io_list)) {
3608 list_del_init(&cache->io_list);
3609 btrfs_wait_cache_io(trans, cache, path);
3610 btrfs_put_block_group(cache);
3615 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3616 * if it should update the cache_state. Don't delete
3617 * until after we wait.
3619 * Since we're not running in the commit critical section
3620 * we need the dirty_bgs_lock to protect from update_block_group
3622 spin_lock(&cur_trans->dirty_bgs_lock);
3623 list_del_init(&cache->dirty_list);
3624 spin_unlock(&cur_trans->dirty_bgs_lock);
3628 cache_save_setup(cache, trans, path);
3630 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3631 cache->io_ctl.inode = NULL;
3632 ret = btrfs_write_out_cache(fs_info, trans,
3634 if (ret == 0 && cache->io_ctl.inode) {
3639 * the cache_write_mutex is protecting
3642 list_add_tail(&cache->io_list, io);
3645 * if we failed to write the cache, the
3646 * generation will be bad and life goes on
3652 ret = write_one_cache_group(trans, fs_info,
3655 * Our block group might still be attached to the list
3656 * of new block groups in the transaction handle of some
3657 * other task (struct btrfs_trans_handle->new_bgs). This
3658 * means its block group item isn't yet in the extent
3659 * tree. If this happens ignore the error, as we will
3660 * try again later in the critical section of the
3661 * transaction commit.
3663 if (ret == -ENOENT) {
3665 spin_lock(&cur_trans->dirty_bgs_lock);
3666 if (list_empty(&cache->dirty_list)) {
3667 list_add_tail(&cache->dirty_list,
3668 &cur_trans->dirty_bgs);
3669 btrfs_get_block_group(cache);
3671 spin_unlock(&cur_trans->dirty_bgs_lock);
3673 btrfs_abort_transaction(trans, ret);
3677 /* if its not on the io list, we need to put the block group */
3679 btrfs_put_block_group(cache);
3685 * Avoid blocking other tasks for too long. It might even save
3686 * us from writing caches for block groups that are going to be
3689 mutex_unlock(&trans->transaction->cache_write_mutex);
3690 mutex_lock(&trans->transaction->cache_write_mutex);
3692 mutex_unlock(&trans->transaction->cache_write_mutex);
3695 * go through delayed refs for all the stuff we've just kicked off
3696 * and then loop back (just once)
3698 ret = btrfs_run_delayed_refs(trans, fs_info, 0);
3699 if (!ret && loops == 0) {
3701 spin_lock(&cur_trans->dirty_bgs_lock);
3702 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3704 * dirty_bgs_lock protects us from concurrent block group
3705 * deletes too (not just cache_write_mutex).
3707 if (!list_empty(&dirty)) {
3708 spin_unlock(&cur_trans->dirty_bgs_lock);
3711 spin_unlock(&cur_trans->dirty_bgs_lock);
3712 } else if (ret < 0) {
3713 btrfs_cleanup_dirty_bgs(cur_trans, fs_info);
3716 btrfs_free_path(path);
3720 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3721 struct btrfs_fs_info *fs_info)
3723 struct btrfs_block_group_cache *cache;
3724 struct btrfs_transaction *cur_trans = trans->transaction;
3727 struct btrfs_path *path;
3728 struct list_head *io = &cur_trans->io_bgs;
3729 int num_started = 0;
3731 path = btrfs_alloc_path();
3736 * Even though we are in the critical section of the transaction commit,
3737 * we can still have concurrent tasks adding elements to this
3738 * transaction's list of dirty block groups. These tasks correspond to
3739 * endio free space workers started when writeback finishes for a
3740 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3741 * allocate new block groups as a result of COWing nodes of the root
3742 * tree when updating the free space inode. The writeback for the space
3743 * caches is triggered by an earlier call to
3744 * btrfs_start_dirty_block_groups() and iterations of the following
3746 * Also we want to do the cache_save_setup first and then run the
3747 * delayed refs to make sure we have the best chance at doing this all
3750 spin_lock(&cur_trans->dirty_bgs_lock);
3751 while (!list_empty(&cur_trans->dirty_bgs)) {
3752 cache = list_first_entry(&cur_trans->dirty_bgs,
3753 struct btrfs_block_group_cache,
3757 * this can happen if cache_save_setup re-dirties a block
3758 * group that is already under IO. Just wait for it to
3759 * finish and then do it all again
3761 if (!list_empty(&cache->io_list)) {
3762 spin_unlock(&cur_trans->dirty_bgs_lock);
3763 list_del_init(&cache->io_list);
3764 btrfs_wait_cache_io(trans, cache, path);
3765 btrfs_put_block_group(cache);
3766 spin_lock(&cur_trans->dirty_bgs_lock);
3770 * don't remove from the dirty list until after we've waited
3773 list_del_init(&cache->dirty_list);
3774 spin_unlock(&cur_trans->dirty_bgs_lock);
3777 cache_save_setup(cache, trans, path);
3780 ret = btrfs_run_delayed_refs(trans, fs_info,
3781 (unsigned long) -1);
3783 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3784 cache->io_ctl.inode = NULL;
3785 ret = btrfs_write_out_cache(fs_info, trans,
3787 if (ret == 0 && cache->io_ctl.inode) {
3790 list_add_tail(&cache->io_list, io);
3793 * if we failed to write the cache, the
3794 * generation will be bad and life goes on
3800 ret = write_one_cache_group(trans, fs_info,
3803 * One of the free space endio workers might have
3804 * created a new block group while updating a free space
3805 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3806 * and hasn't released its transaction handle yet, in
3807 * which case the new block group is still attached to
3808 * its transaction handle and its creation has not
3809 * finished yet (no block group item in the extent tree
3810 * yet, etc). If this is the case, wait for all free
3811 * space endio workers to finish and retry. This is a
3812 * a very rare case so no need for a more efficient and
3815 if (ret == -ENOENT) {
3816 wait_event(cur_trans->writer_wait,
3817 atomic_read(&cur_trans->num_writers) == 1);
3818 ret = write_one_cache_group(trans, fs_info,
3822 btrfs_abort_transaction(trans, ret);
3825 /* if its not on the io list, we need to put the block group */
3827 btrfs_put_block_group(cache);
3828 spin_lock(&cur_trans->dirty_bgs_lock);
3830 spin_unlock(&cur_trans->dirty_bgs_lock);
3832 while (!list_empty(io)) {
3833 cache = list_first_entry(io, struct btrfs_block_group_cache,
3835 list_del_init(&cache->io_list);
3836 btrfs_wait_cache_io(trans, cache, path);
3837 btrfs_put_block_group(cache);
3840 btrfs_free_path(path);
3844 int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr)
3846 struct btrfs_block_group_cache *block_group;
3849 block_group = btrfs_lookup_block_group(fs_info, bytenr);
3850 if (!block_group || block_group->ro)
3853 btrfs_put_block_group(block_group);
3857 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3859 struct btrfs_block_group_cache *bg;
3862 bg = btrfs_lookup_block_group(fs_info, bytenr);
3866 spin_lock(&bg->lock);
3870 atomic_inc(&bg->nocow_writers);
3871 spin_unlock(&bg->lock);
3873 /* no put on block group, done by btrfs_dec_nocow_writers */
3875 btrfs_put_block_group(bg);
3881 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3883 struct btrfs_block_group_cache *bg;
3885 bg = btrfs_lookup_block_group(fs_info, bytenr);
3887 if (atomic_dec_and_test(&bg->nocow_writers))
3888 wake_up_atomic_t(&bg->nocow_writers);
3890 * Once for our lookup and once for the lookup done by a previous call
3891 * to btrfs_inc_nocow_writers()
3893 btrfs_put_block_group(bg);
3894 btrfs_put_block_group(bg);
3897 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3903 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3905 wait_on_atomic_t(&bg->nocow_writers,
3906 btrfs_wait_nocow_writers_atomic_t,
3907 TASK_UNINTERRUPTIBLE);
3910 static const char *alloc_name(u64 flags)
3913 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3915 case BTRFS_BLOCK_GROUP_METADATA:
3917 case BTRFS_BLOCK_GROUP_DATA:
3919 case BTRFS_BLOCK_GROUP_SYSTEM:
3923 return "invalid-combination";
3927 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3928 u64 total_bytes, u64 bytes_used,
3930 struct btrfs_space_info **space_info)
3932 struct btrfs_space_info *found;
3937 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3938 BTRFS_BLOCK_GROUP_RAID10))
3943 found = __find_space_info(info, flags);
3945 spin_lock(&found->lock);
3946 found->total_bytes += total_bytes;
3947 found->disk_total += total_bytes * factor;
3948 found->bytes_used += bytes_used;
3949 found->disk_used += bytes_used * factor;
3950 found->bytes_readonly += bytes_readonly;
3951 if (total_bytes > 0)
3953 space_info_add_new_bytes(info, found, total_bytes -
3954 bytes_used - bytes_readonly);
3955 spin_unlock(&found->lock);
3956 *space_info = found;
3959 found = kzalloc(sizeof(*found), GFP_NOFS);
3963 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3969 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3970 INIT_LIST_HEAD(&found->block_groups[i]);
3971 init_rwsem(&found->groups_sem);
3972 spin_lock_init(&found->lock);
3973 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3974 found->total_bytes = total_bytes;
3975 found->disk_total = total_bytes * factor;
3976 found->bytes_used = bytes_used;
3977 found->disk_used = bytes_used * factor;
3978 found->bytes_pinned = 0;
3979 found->bytes_reserved = 0;
3980 found->bytes_readonly = bytes_readonly;
3981 found->bytes_may_use = 0;
3983 found->max_extent_size = 0;
3984 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3985 found->chunk_alloc = 0;
3987 init_waitqueue_head(&found->wait);
3988 INIT_LIST_HEAD(&found->ro_bgs);
3989 INIT_LIST_HEAD(&found->tickets);
3990 INIT_LIST_HEAD(&found->priority_tickets);
3992 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3993 info->space_info_kobj, "%s",
3994 alloc_name(found->flags));
4000 *space_info = found;
4001 list_add_rcu(&found->list, &info->space_info);
4002 if (flags & BTRFS_BLOCK_GROUP_DATA)
4003 info->data_sinfo = found;
4008 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4010 u64 extra_flags = chunk_to_extended(flags) &
4011 BTRFS_EXTENDED_PROFILE_MASK;
4013 write_seqlock(&fs_info->profiles_lock);
4014 if (flags & BTRFS_BLOCK_GROUP_DATA)
4015 fs_info->avail_data_alloc_bits |= extra_flags;
4016 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4017 fs_info->avail_metadata_alloc_bits |= extra_flags;
4018 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4019 fs_info->avail_system_alloc_bits |= extra_flags;
4020 write_sequnlock(&fs_info->profiles_lock);
4024 * returns target flags in extended format or 0 if restripe for this
4025 * chunk_type is not in progress
4027 * should be called with either volume_mutex or balance_lock held
4029 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4031 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4037 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4038 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4039 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4040 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4041 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4042 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4043 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4044 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4045 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4052 * @flags: available profiles in extended format (see ctree.h)
4054 * Returns reduced profile in chunk format. If profile changing is in
4055 * progress (either running or paused) picks the target profile (if it's
4056 * already available), otherwise falls back to plain reducing.
4058 static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags)
4060 u64 num_devices = fs_info->fs_devices->rw_devices;
4066 * see if restripe for this chunk_type is in progress, if so
4067 * try to reduce to the target profile
4069 spin_lock(&fs_info->balance_lock);
4070 target = get_restripe_target(fs_info, flags);
4072 /* pick target profile only if it's already available */
4073 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4074 spin_unlock(&fs_info->balance_lock);
4075 return extended_to_chunk(target);
4078 spin_unlock(&fs_info->balance_lock);
4080 /* First, mask out the RAID levels which aren't possible */
4081 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4082 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4083 allowed |= btrfs_raid_group[raid_type];
4087 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4088 allowed = BTRFS_BLOCK_GROUP_RAID6;
4089 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4090 allowed = BTRFS_BLOCK_GROUP_RAID5;
4091 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4092 allowed = BTRFS_BLOCK_GROUP_RAID10;
4093 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4094 allowed = BTRFS_BLOCK_GROUP_RAID1;
4095 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4096 allowed = BTRFS_BLOCK_GROUP_RAID0;
4098 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4100 return extended_to_chunk(flags | allowed);
4103 static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags)
4110 seq = read_seqbegin(&fs_info->profiles_lock);
4112 if (flags & BTRFS_BLOCK_GROUP_DATA)
4113 flags |= fs_info->avail_data_alloc_bits;
4114 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4115 flags |= fs_info->avail_system_alloc_bits;
4116 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4117 flags |= fs_info->avail_metadata_alloc_bits;
4118 } while (read_seqretry(&fs_info->profiles_lock, seq));
4120 return btrfs_reduce_alloc_profile(fs_info, flags);
4123 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4125 struct btrfs_fs_info *fs_info = root->fs_info;
4130 flags = BTRFS_BLOCK_GROUP_DATA;
4131 else if (root == fs_info->chunk_root)
4132 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4134 flags = BTRFS_BLOCK_GROUP_METADATA;
4136 ret = get_alloc_profile(fs_info, flags);
4140 static u64 btrfs_space_info_used(struct btrfs_space_info *s_info,
4141 bool may_use_included)
4144 return s_info->bytes_used + s_info->bytes_reserved +
4145 s_info->bytes_pinned + s_info->bytes_readonly +
4146 (may_use_included ? s_info->bytes_may_use : 0);
4149 int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes)
4151 struct btrfs_space_info *data_sinfo;
4152 struct btrfs_root *root = inode->root;
4153 struct btrfs_fs_info *fs_info = root->fs_info;
4156 int need_commit = 2;
4157 int have_pinned_space;
4159 /* make sure bytes are sectorsize aligned */
4160 bytes = ALIGN(bytes, fs_info->sectorsize);
4162 if (btrfs_is_free_space_inode(inode)) {
4164 ASSERT(current->journal_info);
4167 data_sinfo = fs_info->data_sinfo;
4172 /* make sure we have enough space to handle the data first */
4173 spin_lock(&data_sinfo->lock);
4174 used = btrfs_space_info_used(data_sinfo, true);
4176 if (used + bytes > data_sinfo->total_bytes) {
4177 struct btrfs_trans_handle *trans;
4180 * if we don't have enough free bytes in this space then we need
4181 * to alloc a new chunk.
4183 if (!data_sinfo->full) {
4186 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4187 spin_unlock(&data_sinfo->lock);
4189 alloc_target = btrfs_get_alloc_profile(root, 1);
4191 * It is ugly that we don't call nolock join
4192 * transaction for the free space inode case here.
4193 * But it is safe because we only do the data space
4194 * reservation for the free space cache in the
4195 * transaction context, the common join transaction
4196 * just increase the counter of the current transaction
4197 * handler, doesn't try to acquire the trans_lock of
4200 trans = btrfs_join_transaction(root);
4202 return PTR_ERR(trans);
4204 ret = do_chunk_alloc(trans, fs_info, alloc_target,
4205 CHUNK_ALLOC_NO_FORCE);
4206 btrfs_end_transaction(trans);
4211 have_pinned_space = 1;
4217 data_sinfo = fs_info->data_sinfo;
4223 * If we don't have enough pinned space to deal with this
4224 * allocation, and no removed chunk in current transaction,
4225 * don't bother committing the transaction.
4227 have_pinned_space = percpu_counter_compare(
4228 &data_sinfo->total_bytes_pinned,
4229 used + bytes - data_sinfo->total_bytes);
4230 spin_unlock(&data_sinfo->lock);
4232 /* commit the current transaction and try again */
4235 !atomic_read(&fs_info->open_ioctl_trans)) {
4238 if (need_commit > 0) {
4239 btrfs_start_delalloc_roots(fs_info, 0, -1);
4240 btrfs_wait_ordered_roots(fs_info, -1, 0,
4244 trans = btrfs_join_transaction(root);
4246 return PTR_ERR(trans);
4247 if (have_pinned_space >= 0 ||
4248 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4249 &trans->transaction->flags) ||
4251 ret = btrfs_commit_transaction(trans);
4255 * The cleaner kthread might still be doing iput
4256 * operations. Wait for it to finish so that
4257 * more space is released.
4259 mutex_lock(&fs_info->cleaner_delayed_iput_mutex);
4260 mutex_unlock(&fs_info->cleaner_delayed_iput_mutex);
4263 btrfs_end_transaction(trans);
4267 trace_btrfs_space_reservation(fs_info,
4268 "space_info:enospc",
4269 data_sinfo->flags, bytes, 1);
4272 data_sinfo->bytes_may_use += bytes;
4273 trace_btrfs_space_reservation(fs_info, "space_info",
4274 data_sinfo->flags, bytes, 1);
4275 spin_unlock(&data_sinfo->lock);
4281 * New check_data_free_space() with ability for precious data reservation
4282 * Will replace old btrfs_check_data_free_space(), but for patch split,
4283 * add a new function first and then replace it.
4285 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4287 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4290 /* align the range */
4291 len = round_up(start + len, fs_info->sectorsize) -
4292 round_down(start, fs_info->sectorsize);
4293 start = round_down(start, fs_info->sectorsize);
4295 ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len);
4299 /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */
4300 ret = btrfs_qgroup_reserve_data(inode, start, len);
4302 btrfs_free_reserved_data_space_noquota(inode, start, len);
4307 * Called if we need to clear a data reservation for this inode
4308 * Normally in a error case.
4310 * This one will *NOT* use accurate qgroup reserved space API, just for case
4311 * which we can't sleep and is sure it won't affect qgroup reserved space.
4312 * Like clear_bit_hook().
4314 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4317 struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4318 struct btrfs_space_info *data_sinfo;
4320 /* Make sure the range is aligned to sectorsize */
4321 len = round_up(start + len, fs_info->sectorsize) -
4322 round_down(start, fs_info->sectorsize);
4323 start = round_down(start, fs_info->sectorsize);
4325 data_sinfo = fs_info->data_sinfo;
4326 spin_lock(&data_sinfo->lock);
4327 if (WARN_ON(data_sinfo->bytes_may_use < len))
4328 data_sinfo->bytes_may_use = 0;
4330 data_sinfo->bytes_may_use -= len;
4331 trace_btrfs_space_reservation(fs_info, "space_info",
4332 data_sinfo->flags, len, 0);
4333 spin_unlock(&data_sinfo->lock);
4337 * Called if we need to clear a data reservation for this inode
4338 * Normally in a error case.
4340 * This one will handle the per-inode data rsv map for accurate reserved
4343 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4345 struct btrfs_root *root = BTRFS_I(inode)->root;
4347 /* Make sure the range is aligned to sectorsize */
4348 len = round_up(start + len, root->fs_info->sectorsize) -
4349 round_down(start, root->fs_info->sectorsize);
4350 start = round_down(start, root->fs_info->sectorsize);
4352 btrfs_free_reserved_data_space_noquota(inode, start, len);
4353 btrfs_qgroup_free_data(inode, start, len);
4356 static void force_metadata_allocation(struct btrfs_fs_info *info)
4358 struct list_head *head = &info->space_info;
4359 struct btrfs_space_info *found;
4362 list_for_each_entry_rcu(found, head, list) {
4363 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4364 found->force_alloc = CHUNK_ALLOC_FORCE;
4369 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4371 return (global->size << 1);
4374 static int should_alloc_chunk(struct btrfs_fs_info *fs_info,
4375 struct btrfs_space_info *sinfo, int force)
4377 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4378 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4379 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4382 if (force == CHUNK_ALLOC_FORCE)
4386 * We need to take into account the global rsv because for all intents
4387 * and purposes it's used space. Don't worry about locking the
4388 * global_rsv, it doesn't change except when the transaction commits.
4390 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4391 num_allocated += calc_global_rsv_need_space(global_rsv);
4394 * in limited mode, we want to have some free space up to
4395 * about 1% of the FS size.
4397 if (force == CHUNK_ALLOC_LIMITED) {
4398 thresh = btrfs_super_total_bytes(fs_info->super_copy);
4399 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4401 if (num_bytes - num_allocated < thresh)
4405 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4410 static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type)
4414 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4415 BTRFS_BLOCK_GROUP_RAID0 |
4416 BTRFS_BLOCK_GROUP_RAID5 |
4417 BTRFS_BLOCK_GROUP_RAID6))
4418 num_dev = fs_info->fs_devices->rw_devices;
4419 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4422 num_dev = 1; /* DUP or single */
4428 * If @is_allocation is true, reserve space in the system space info necessary
4429 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4432 void check_system_chunk(struct btrfs_trans_handle *trans,
4433 struct btrfs_fs_info *fs_info, u64 type)
4435 struct btrfs_space_info *info;
4442 * Needed because we can end up allocating a system chunk and for an
4443 * atomic and race free space reservation in the chunk block reserve.
4445 ASSERT(mutex_is_locked(&fs_info->chunk_mutex));
4447 info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4448 spin_lock(&info->lock);
4449 left = info->total_bytes - btrfs_space_info_used(info, true);
4450 spin_unlock(&info->lock);
4452 num_devs = get_profile_num_devs(fs_info, type);
4454 /* num_devs device items to update and 1 chunk item to add or remove */
4455 thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) +
4456 btrfs_calc_trans_metadata_size(fs_info, 1);
4458 if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
4459 btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu",
4460 left, thresh, type);
4461 dump_space_info(fs_info, info, 0, 0);
4464 if (left < thresh) {
4467 flags = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4469 * Ignore failure to create system chunk. We might end up not
4470 * needing it, as we might not need to COW all nodes/leafs from
4471 * the paths we visit in the chunk tree (they were already COWed
4472 * or created in the current transaction for example).
4474 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4478 ret = btrfs_block_rsv_add(fs_info->chunk_root,
4479 &fs_info->chunk_block_rsv,
4480 thresh, BTRFS_RESERVE_NO_FLUSH);
4482 trans->chunk_bytes_reserved += thresh;
4487 * If force is CHUNK_ALLOC_FORCE:
4488 * - return 1 if it successfully allocates a chunk,
4489 * - return errors including -ENOSPC otherwise.
4490 * If force is NOT CHUNK_ALLOC_FORCE:
4491 * - return 0 if it doesn't need to allocate a new chunk,
4492 * - return 1 if it successfully allocates a chunk,
4493 * - return errors including -ENOSPC otherwise.
4495 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4496 struct btrfs_fs_info *fs_info, u64 flags, int force)
4498 struct btrfs_space_info *space_info;
4499 int wait_for_alloc = 0;
4502 /* Don't re-enter if we're already allocating a chunk */
4503 if (trans->allocating_chunk)
4506 space_info = __find_space_info(fs_info, flags);
4508 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
4509 BUG_ON(ret); /* -ENOMEM */
4511 BUG_ON(!space_info); /* Logic error */
4514 spin_lock(&space_info->lock);
4515 if (force < space_info->force_alloc)
4516 force = space_info->force_alloc;
4517 if (space_info->full) {
4518 if (should_alloc_chunk(fs_info, space_info, force))
4522 spin_unlock(&space_info->lock);
4526 if (!should_alloc_chunk(fs_info, space_info, force)) {
4527 spin_unlock(&space_info->lock);
4529 } else if (space_info->chunk_alloc) {
4532 space_info->chunk_alloc = 1;
4535 spin_unlock(&space_info->lock);
4537 mutex_lock(&fs_info->chunk_mutex);
4540 * The chunk_mutex is held throughout the entirety of a chunk
4541 * allocation, so once we've acquired the chunk_mutex we know that the
4542 * other guy is done and we need to recheck and see if we should
4545 if (wait_for_alloc) {
4546 mutex_unlock(&fs_info->chunk_mutex);
4551 trans->allocating_chunk = true;
4554 * If we have mixed data/metadata chunks we want to make sure we keep
4555 * allocating mixed chunks instead of individual chunks.
4557 if (btrfs_mixed_space_info(space_info))
4558 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4561 * if we're doing a data chunk, go ahead and make sure that
4562 * we keep a reasonable number of metadata chunks allocated in the
4565 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4566 fs_info->data_chunk_allocations++;
4567 if (!(fs_info->data_chunk_allocations %
4568 fs_info->metadata_ratio))
4569 force_metadata_allocation(fs_info);
4573 * Check if we have enough space in SYSTEM chunk because we may need
4574 * to update devices.
4576 check_system_chunk(trans, fs_info, flags);
4578 ret = btrfs_alloc_chunk(trans, fs_info, flags);
4579 trans->allocating_chunk = false;
4581 spin_lock(&space_info->lock);
4582 if (ret < 0 && ret != -ENOSPC)
4585 space_info->full = 1;
4589 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4591 space_info->chunk_alloc = 0;
4592 spin_unlock(&space_info->lock);
4593 mutex_unlock(&fs_info->chunk_mutex);
4595 * When we allocate a new chunk we reserve space in the chunk block
4596 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4597 * add new nodes/leafs to it if we end up needing to do it when
4598 * inserting the chunk item and updating device items as part of the
4599 * second phase of chunk allocation, performed by
4600 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4601 * large number of new block groups to create in our transaction
4602 * handle's new_bgs list to avoid exhausting the chunk block reserve
4603 * in extreme cases - like having a single transaction create many new
4604 * block groups when starting to write out the free space caches of all
4605 * the block groups that were made dirty during the lifetime of the
4608 if (trans->can_flush_pending_bgs &&
4609 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4610 btrfs_create_pending_block_groups(trans, fs_info);
4611 btrfs_trans_release_chunk_metadata(trans);
4616 static int can_overcommit(struct btrfs_root *root,
4617 struct btrfs_space_info *space_info, u64 bytes,
4618 enum btrfs_reserve_flush_enum flush)
4620 struct btrfs_fs_info *fs_info = root->fs_info;
4621 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4627 /* Don't overcommit when in mixed mode. */
4628 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4631 profile = btrfs_get_alloc_profile(root, 0);
4632 used = btrfs_space_info_used(space_info, false);
4635 * We only want to allow over committing if we have lots of actual space
4636 * free, but if we don't have enough space to handle the global reserve
4637 * space then we could end up having a real enospc problem when trying
4638 * to allocate a chunk or some other such important allocation.
4640 spin_lock(&global_rsv->lock);
4641 space_size = calc_global_rsv_need_space(global_rsv);
4642 spin_unlock(&global_rsv->lock);
4643 if (used + space_size >= space_info->total_bytes)
4646 used += space_info->bytes_may_use;
4648 spin_lock(&fs_info->free_chunk_lock);
4649 avail = fs_info->free_chunk_space;
4650 spin_unlock(&fs_info->free_chunk_lock);
4653 * If we have dup, raid1 or raid10 then only half of the free
4654 * space is actually useable. For raid56, the space info used
4655 * doesn't include the parity drive, so we don't have to
4658 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4659 BTRFS_BLOCK_GROUP_RAID1 |
4660 BTRFS_BLOCK_GROUP_RAID10))
4664 * If we aren't flushing all things, let us overcommit up to
4665 * 1/2th of the space. If we can flush, don't let us overcommit
4666 * too much, let it overcommit up to 1/8 of the space.
4668 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4673 if (used + bytes < space_info->total_bytes + avail)
4678 static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info,
4679 unsigned long nr_pages, int nr_items)
4681 struct super_block *sb = fs_info->sb;
4683 if (down_read_trylock(&sb->s_umount)) {
4684 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4685 up_read(&sb->s_umount);
4688 * We needn't worry the filesystem going from r/w to r/o though
4689 * we don't acquire ->s_umount mutex, because the filesystem
4690 * should guarantee the delalloc inodes list be empty after
4691 * the filesystem is readonly(all dirty pages are written to
4694 btrfs_start_delalloc_roots(fs_info, 0, nr_items);
4695 if (!current->journal_info)
4696 btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1);
4700 static inline int calc_reclaim_items_nr(struct btrfs_fs_info *fs_info,
4706 bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
4707 nr = (int)div64_u64(to_reclaim, bytes);
4713 #define EXTENT_SIZE_PER_ITEM SZ_256K
4716 * shrink metadata reservation for delalloc
4718 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4721 struct btrfs_fs_info *fs_info = root->fs_info;
4722 struct btrfs_block_rsv *block_rsv;
4723 struct btrfs_space_info *space_info;
4724 struct btrfs_trans_handle *trans;
4728 unsigned long nr_pages;
4731 enum btrfs_reserve_flush_enum flush;
4733 /* Calc the number of the pages we need flush for space reservation */
4734 items = calc_reclaim_items_nr(fs_info, to_reclaim);
4735 to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM;
4737 trans = (struct btrfs_trans_handle *)current->journal_info;
4738 block_rsv = &fs_info->delalloc_block_rsv;
4739 space_info = block_rsv->space_info;
4741 delalloc_bytes = percpu_counter_sum_positive(
4742 &fs_info->delalloc_bytes);
4743 if (delalloc_bytes == 0) {
4747 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4752 while (delalloc_bytes && loops < 3) {
4753 max_reclaim = min(delalloc_bytes, to_reclaim);
4754 nr_pages = max_reclaim >> PAGE_SHIFT;
4755 btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items);
4757 * We need to wait for the async pages to actually start before
4760 max_reclaim = atomic_read(&fs_info->async_delalloc_pages);
4764 if (max_reclaim <= nr_pages)
4767 max_reclaim -= nr_pages;
4769 wait_event(fs_info->async_submit_wait,
4770 atomic_read(&fs_info->async_delalloc_pages) <=
4774 flush = BTRFS_RESERVE_FLUSH_ALL;
4776 flush = BTRFS_RESERVE_NO_FLUSH;
4777 spin_lock(&space_info->lock);
4778 if (can_overcommit(root, space_info, orig, flush)) {
4779 spin_unlock(&space_info->lock);
4782 if (list_empty(&space_info->tickets) &&
4783 list_empty(&space_info->priority_tickets)) {
4784 spin_unlock(&space_info->lock);
4787 spin_unlock(&space_info->lock);
4790 if (wait_ordered && !trans) {
4791 btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1);
4793 time_left = schedule_timeout_killable(1);
4797 delalloc_bytes = percpu_counter_sum_positive(
4798 &fs_info->delalloc_bytes);
4803 * maybe_commit_transaction - possibly commit the transaction if its ok to
4804 * @root - the root we're allocating for
4805 * @bytes - the number of bytes we want to reserve
4806 * @force - force the commit
4808 * This will check to make sure that committing the transaction will actually
4809 * get us somewhere and then commit the transaction if it does. Otherwise it
4810 * will return -ENOSPC.
4812 static int may_commit_transaction(struct btrfs_fs_info *fs_info,
4813 struct btrfs_space_info *space_info,
4814 u64 bytes, int force)
4816 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv;
4817 struct btrfs_trans_handle *trans;
4819 trans = (struct btrfs_trans_handle *)current->journal_info;
4826 /* See if there is enough pinned space to make this reservation */
4827 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4832 * See if there is some space in the delayed insertion reservation for
4835 if (space_info != delayed_rsv->space_info)
4838 spin_lock(&delayed_rsv->lock);
4839 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4840 bytes - delayed_rsv->size) >= 0) {
4841 spin_unlock(&delayed_rsv->lock);
4844 spin_unlock(&delayed_rsv->lock);
4847 trans = btrfs_join_transaction(fs_info->fs_root);
4851 return btrfs_commit_transaction(trans);
4854 struct reserve_ticket {
4857 struct list_head list;
4858 wait_queue_head_t wait;
4861 static int flush_space(struct btrfs_fs_info *fs_info,
4862 struct btrfs_space_info *space_info, u64 num_bytes,
4863 u64 orig_bytes, int state)
4865 struct btrfs_root *root = fs_info->fs_root;
4866 struct btrfs_trans_handle *trans;
4871 case FLUSH_DELAYED_ITEMS_NR:
4872 case FLUSH_DELAYED_ITEMS:
4873 if (state == FLUSH_DELAYED_ITEMS_NR)
4874 nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2;
4878 trans = btrfs_join_transaction(root);
4879 if (IS_ERR(trans)) {
4880 ret = PTR_ERR(trans);
4883 ret = btrfs_run_delayed_items_nr(trans, fs_info, nr);
4884 btrfs_end_transaction(trans);
4886 case FLUSH_DELALLOC:
4887 case FLUSH_DELALLOC_WAIT:
4888 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4889 state == FLUSH_DELALLOC_WAIT);
4892 trans = btrfs_join_transaction(root);
4893 if (IS_ERR(trans)) {
4894 ret = PTR_ERR(trans);
4897 ret = do_chunk_alloc(trans, fs_info,
4898 btrfs_get_alloc_profile(root, 0),
4899 CHUNK_ALLOC_NO_FORCE);
4900 btrfs_end_transaction(trans);
4901 if (ret > 0 || ret == -ENOSPC)
4905 ret = may_commit_transaction(fs_info, space_info,
4913 trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes,
4914 orig_bytes, state, ret);
4919 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4920 struct btrfs_space_info *space_info)
4922 struct reserve_ticket *ticket;
4927 list_for_each_entry(ticket, &space_info->tickets, list)
4928 to_reclaim += ticket->bytes;
4929 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4930 to_reclaim += ticket->bytes;
4934 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4935 if (can_overcommit(root, space_info, to_reclaim,
4936 BTRFS_RESERVE_FLUSH_ALL))
4939 used = space_info->bytes_used + space_info->bytes_reserved +
4940 space_info->bytes_pinned + space_info->bytes_readonly +
4941 space_info->bytes_may_use;
4942 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4943 expected = div_factor_fine(space_info->total_bytes, 95);
4945 expected = div_factor_fine(space_info->total_bytes, 90);
4947 if (used > expected)
4948 to_reclaim = used - expected;
4951 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4952 space_info->bytes_reserved);
4956 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4957 struct btrfs_root *root, u64 used)
4959 struct btrfs_fs_info *fs_info = root->fs_info;
4960 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4962 /* If we're just plain full then async reclaim just slows us down. */
4963 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4966 if (!btrfs_calc_reclaim_metadata_size(root, space_info))
4969 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4970 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4973 static void wake_all_tickets(struct list_head *head)
4975 struct reserve_ticket *ticket;
4977 while (!list_empty(head)) {
4978 ticket = list_first_entry(head, struct reserve_ticket, list);
4979 list_del_init(&ticket->list);
4980 ticket->error = -ENOSPC;
4981 wake_up(&ticket->wait);
4986 * This is for normal flushers, we can wait all goddamned day if we want to. We
4987 * will loop and continuously try to flush as long as we are making progress.
4988 * We count progress as clearing off tickets each time we have to loop.
4990 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4992 struct btrfs_fs_info *fs_info;
4993 struct btrfs_space_info *space_info;
4996 int commit_cycles = 0;
4997 u64 last_tickets_id;
4999 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
5000 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5002 spin_lock(&space_info->lock);
5003 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5006 space_info->flush = 0;
5007 spin_unlock(&space_info->lock);
5010 last_tickets_id = space_info->tickets_id;
5011 spin_unlock(&space_info->lock);
5013 flush_state = FLUSH_DELAYED_ITEMS_NR;
5015 struct reserve_ticket *ticket;
5018 ret = flush_space(fs_info, space_info, to_reclaim, to_reclaim,
5020 spin_lock(&space_info->lock);
5021 if (list_empty(&space_info->tickets)) {
5022 space_info->flush = 0;
5023 spin_unlock(&space_info->lock);
5026 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5028 ticket = list_first_entry(&space_info->tickets,
5029 struct reserve_ticket, list);
5030 if (last_tickets_id == space_info->tickets_id) {
5033 last_tickets_id = space_info->tickets_id;
5034 flush_state = FLUSH_DELAYED_ITEMS_NR;
5039 if (flush_state > COMMIT_TRANS) {
5041 if (commit_cycles > 2) {
5042 wake_all_tickets(&space_info->tickets);
5043 space_info->flush = 0;
5045 flush_state = FLUSH_DELAYED_ITEMS_NR;
5048 spin_unlock(&space_info->lock);
5049 } while (flush_state <= COMMIT_TRANS);
5052 void btrfs_init_async_reclaim_work(struct work_struct *work)
5054 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5057 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5058 struct btrfs_space_info *space_info,
5059 struct reserve_ticket *ticket)
5062 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5064 spin_lock(&space_info->lock);
5065 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5068 spin_unlock(&space_info->lock);
5071 spin_unlock(&space_info->lock);
5074 flush_space(fs_info, space_info, to_reclaim, to_reclaim,
5077 spin_lock(&space_info->lock);
5078 if (ticket->bytes == 0) {
5079 spin_unlock(&space_info->lock);
5082 spin_unlock(&space_info->lock);
5085 * Priority flushers can't wait on delalloc without
5088 if (flush_state == FLUSH_DELALLOC ||
5089 flush_state == FLUSH_DELALLOC_WAIT)
5090 flush_state = ALLOC_CHUNK;
5091 } while (flush_state < COMMIT_TRANS);
5094 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5095 struct btrfs_space_info *space_info,
5096 struct reserve_ticket *ticket, u64 orig_bytes)
5102 spin_lock(&space_info->lock);
5103 while (ticket->bytes > 0 && ticket->error == 0) {
5104 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5109 spin_unlock(&space_info->lock);
5113 finish_wait(&ticket->wait, &wait);
5114 spin_lock(&space_info->lock);
5117 ret = ticket->error;
5118 if (!list_empty(&ticket->list))
5119 list_del_init(&ticket->list);
5120 if (ticket->bytes && ticket->bytes < orig_bytes) {
5121 u64 num_bytes = orig_bytes - ticket->bytes;
5122 space_info->bytes_may_use -= num_bytes;
5123 trace_btrfs_space_reservation(fs_info, "space_info",
5124 space_info->flags, num_bytes, 0);
5126 spin_unlock(&space_info->lock);
5132 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5133 * @root - the root we're allocating for
5134 * @space_info - the space info we want to allocate from
5135 * @orig_bytes - the number of bytes we want
5136 * @flush - whether or not we can flush to make our reservation
5138 * This will reserve orig_bytes number of bytes from the space info associated
5139 * with the block_rsv. If there is not enough space it will make an attempt to
5140 * flush out space to make room. It will do this by flushing delalloc if
5141 * possible or committing the transaction. If flush is 0 then no attempts to
5142 * regain reservations will be made and this will fail if there is not enough
5145 static int __reserve_metadata_bytes(struct btrfs_root *root,
5146 struct btrfs_space_info *space_info,
5148 enum btrfs_reserve_flush_enum flush)
5150 struct btrfs_fs_info *fs_info = root->fs_info;
5151 struct reserve_ticket ticket;
5156 ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL);
5158 spin_lock(&space_info->lock);
5160 used = btrfs_space_info_used(space_info, true);
5163 * If we have enough space then hooray, make our reservation and carry
5164 * on. If not see if we can overcommit, and if we can, hooray carry on.
5165 * If not things get more complicated.
5167 if (used + orig_bytes <= space_info->total_bytes) {
5168 space_info->bytes_may_use += orig_bytes;
5169 trace_btrfs_space_reservation(fs_info, "space_info",
5170 space_info->flags, orig_bytes, 1);
5172 } else if (can_overcommit(root, space_info, orig_bytes, flush)) {
5173 space_info->bytes_may_use += orig_bytes;
5174 trace_btrfs_space_reservation(fs_info, "space_info",
5175 space_info->flags, orig_bytes, 1);
5180 * If we couldn't make a reservation then setup our reservation ticket
5181 * and kick the async worker if it's not already running.
5183 * If we are a priority flusher then we just need to add our ticket to
5184 * the list and we will do our own flushing further down.
5186 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5187 ticket.bytes = orig_bytes;
5189 init_waitqueue_head(&ticket.wait);
5190 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5191 list_add_tail(&ticket.list, &space_info->tickets);
5192 if (!space_info->flush) {
5193 space_info->flush = 1;
5194 trace_btrfs_trigger_flush(fs_info,
5198 queue_work(system_unbound_wq,
5199 &root->fs_info->async_reclaim_work);
5202 list_add_tail(&ticket.list,
5203 &space_info->priority_tickets);
5205 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5208 * We will do the space reservation dance during log replay,
5209 * which means we won't have fs_info->fs_root set, so don't do
5210 * the async reclaim as we will panic.
5212 if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) &&
5213 need_do_async_reclaim(space_info, root, used) &&
5214 !work_busy(&fs_info->async_reclaim_work)) {
5215 trace_btrfs_trigger_flush(fs_info, space_info->flags,
5216 orig_bytes, flush, "preempt");
5217 queue_work(system_unbound_wq,
5218 &fs_info->async_reclaim_work);
5221 spin_unlock(&space_info->lock);
5222 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5225 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5226 return wait_reserve_ticket(fs_info, space_info, &ticket,
5230 priority_reclaim_metadata_space(fs_info, space_info, &ticket);
5231 spin_lock(&space_info->lock);
5233 if (ticket.bytes < orig_bytes) {
5234 u64 num_bytes = orig_bytes - ticket.bytes;
5235 space_info->bytes_may_use -= num_bytes;
5236 trace_btrfs_space_reservation(fs_info, "space_info",
5241 list_del_init(&ticket.list);
5244 spin_unlock(&space_info->lock);
5245 ASSERT(list_empty(&ticket.list));
5250 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5251 * @root - the root we're allocating for
5252 * @block_rsv - the block_rsv we're allocating for
5253 * @orig_bytes - the number of bytes we want
5254 * @flush - whether or not we can flush to make our reservation
5256 * This will reserve orgi_bytes number of bytes from the space info associated
5257 * with the block_rsv. If there is not enough space it will make an attempt to
5258 * flush out space to make room. It will do this by flushing delalloc if
5259 * possible or committing the transaction. If flush is 0 then no attempts to
5260 * regain reservations will be made and this will fail if there is not enough
5263 static int reserve_metadata_bytes(struct btrfs_root *root,
5264 struct btrfs_block_rsv *block_rsv,
5266 enum btrfs_reserve_flush_enum flush)
5268 struct btrfs_fs_info *fs_info = root->fs_info;
5269 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5272 ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes,
5274 if (ret == -ENOSPC &&
5275 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5276 if (block_rsv != global_rsv &&
5277 !block_rsv_use_bytes(global_rsv, orig_bytes))
5281 trace_btrfs_space_reservation(fs_info, "space_info:enospc",
5282 block_rsv->space_info->flags,
5287 static struct btrfs_block_rsv *get_block_rsv(
5288 const struct btrfs_trans_handle *trans,
5289 const struct btrfs_root *root)
5291 struct btrfs_fs_info *fs_info = root->fs_info;
5292 struct btrfs_block_rsv *block_rsv = NULL;
5294 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5295 (root == fs_info->csum_root && trans->adding_csums) ||
5296 (root == fs_info->uuid_root))
5297 block_rsv = trans->block_rsv;
5300 block_rsv = root->block_rsv;
5303 block_rsv = &fs_info->empty_block_rsv;
5308 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5312 spin_lock(&block_rsv->lock);
5313 if (block_rsv->reserved >= num_bytes) {
5314 block_rsv->reserved -= num_bytes;
5315 if (block_rsv->reserved < block_rsv->size)
5316 block_rsv->full = 0;
5319 spin_unlock(&block_rsv->lock);
5323 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5324 u64 num_bytes, int update_size)
5326 spin_lock(&block_rsv->lock);
5327 block_rsv->reserved += num_bytes;
5329 block_rsv->size += num_bytes;
5330 else if (block_rsv->reserved >= block_rsv->size)
5331 block_rsv->full = 1;
5332 spin_unlock(&block_rsv->lock);
5335 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5336 struct btrfs_block_rsv *dest, u64 num_bytes,
5339 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5342 if (global_rsv->space_info != dest->space_info)
5345 spin_lock(&global_rsv->lock);
5346 min_bytes = div_factor(global_rsv->size, min_factor);
5347 if (global_rsv->reserved < min_bytes + num_bytes) {
5348 spin_unlock(&global_rsv->lock);
5351 global_rsv->reserved -= num_bytes;
5352 if (global_rsv->reserved < global_rsv->size)
5353 global_rsv->full = 0;
5354 spin_unlock(&global_rsv->lock);
5356 block_rsv_add_bytes(dest, num_bytes, 1);
5361 * This is for space we already have accounted in space_info->bytes_may_use, so
5362 * basically when we're returning space from block_rsv's.
5364 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5365 struct btrfs_space_info *space_info,
5368 struct reserve_ticket *ticket;
5369 struct list_head *head;
5371 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5372 bool check_overcommit = false;
5374 spin_lock(&space_info->lock);
5375 head = &space_info->priority_tickets;
5378 * If we are over our limit then we need to check and see if we can
5379 * overcommit, and if we can't then we just need to free up our space
5380 * and not satisfy any requests.
5382 used = space_info->bytes_used + space_info->bytes_reserved +
5383 space_info->bytes_pinned + space_info->bytes_readonly +
5384 space_info->bytes_may_use;
5385 if (used - num_bytes >= space_info->total_bytes)
5386 check_overcommit = true;
5388 while (!list_empty(head) && num_bytes) {
5389 ticket = list_first_entry(head, struct reserve_ticket,
5392 * We use 0 bytes because this space is already reserved, so
5393 * adding the ticket space would be a double count.
5395 if (check_overcommit &&
5396 !can_overcommit(fs_info->extent_root, space_info, 0,
5399 if (num_bytes >= ticket->bytes) {
5400 list_del_init(&ticket->list);
5401 num_bytes -= ticket->bytes;
5403 space_info->tickets_id++;
5404 wake_up(&ticket->wait);
5406 ticket->bytes -= num_bytes;
5411 if (num_bytes && head == &space_info->priority_tickets) {
5412 head = &space_info->tickets;
5413 flush = BTRFS_RESERVE_FLUSH_ALL;
5416 space_info->bytes_may_use -= num_bytes;
5417 trace_btrfs_space_reservation(fs_info, "space_info",
5418 space_info->flags, num_bytes, 0);
5419 spin_unlock(&space_info->lock);
5423 * This is for newly allocated space that isn't accounted in
5424 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5425 * we use this helper.
5427 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5428 struct btrfs_space_info *space_info,
5431 struct reserve_ticket *ticket;
5432 struct list_head *head = &space_info->priority_tickets;
5435 while (!list_empty(head) && num_bytes) {
5436 ticket = list_first_entry(head, struct reserve_ticket,
5438 if (num_bytes >= ticket->bytes) {
5439 trace_btrfs_space_reservation(fs_info, "space_info",
5442 list_del_init(&ticket->list);
5443 num_bytes -= ticket->bytes;
5444 space_info->bytes_may_use += ticket->bytes;
5446 space_info->tickets_id++;
5447 wake_up(&ticket->wait);
5449 trace_btrfs_space_reservation(fs_info, "space_info",
5452 space_info->bytes_may_use += num_bytes;
5453 ticket->bytes -= num_bytes;
5458 if (num_bytes && head == &space_info->priority_tickets) {
5459 head = &space_info->tickets;
5464 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5465 struct btrfs_block_rsv *block_rsv,
5466 struct btrfs_block_rsv *dest, u64 num_bytes)
5468 struct btrfs_space_info *space_info = block_rsv->space_info;
5470 spin_lock(&block_rsv->lock);
5471 if (num_bytes == (u64)-1)
5472 num_bytes = block_rsv->size;
5473 block_rsv->size -= num_bytes;
5474 if (block_rsv->reserved >= block_rsv->size) {
5475 num_bytes = block_rsv->reserved - block_rsv->size;
5476 block_rsv->reserved = block_rsv->size;
5477 block_rsv->full = 1;
5481 spin_unlock(&block_rsv->lock);
5483 if (num_bytes > 0) {
5485 spin_lock(&dest->lock);
5489 bytes_to_add = dest->size - dest->reserved;
5490 bytes_to_add = min(num_bytes, bytes_to_add);
5491 dest->reserved += bytes_to_add;
5492 if (dest->reserved >= dest->size)
5494 num_bytes -= bytes_to_add;
5496 spin_unlock(&dest->lock);
5499 space_info_add_old_bytes(fs_info, space_info,
5504 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5505 struct btrfs_block_rsv *dst, u64 num_bytes,
5510 ret = block_rsv_use_bytes(src, num_bytes);
5514 block_rsv_add_bytes(dst, num_bytes, update_size);
5518 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5520 memset(rsv, 0, sizeof(*rsv));
5521 spin_lock_init(&rsv->lock);
5525 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info,
5526 unsigned short type)
5528 struct btrfs_block_rsv *block_rsv;
5530 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5534 btrfs_init_block_rsv(block_rsv, type);
5535 block_rsv->space_info = __find_space_info(fs_info,
5536 BTRFS_BLOCK_GROUP_METADATA);
5540 void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info,
5541 struct btrfs_block_rsv *rsv)
5545 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5549 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5554 int btrfs_block_rsv_add(struct btrfs_root *root,
5555 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5556 enum btrfs_reserve_flush_enum flush)
5563 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5565 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5572 int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor)
5580 spin_lock(&block_rsv->lock);
5581 num_bytes = div_factor(block_rsv->size, min_factor);
5582 if (block_rsv->reserved >= num_bytes)
5584 spin_unlock(&block_rsv->lock);
5589 int btrfs_block_rsv_refill(struct btrfs_root *root,
5590 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5591 enum btrfs_reserve_flush_enum flush)
5599 spin_lock(&block_rsv->lock);
5600 num_bytes = min_reserved;
5601 if (block_rsv->reserved >= num_bytes)
5604 num_bytes -= block_rsv->reserved;
5605 spin_unlock(&block_rsv->lock);
5610 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5612 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5619 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
5620 struct btrfs_block_rsv *block_rsv,
5623 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5625 if (global_rsv == block_rsv ||
5626 block_rsv->space_info != global_rsv->space_info)
5628 block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes);
5631 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5633 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5634 struct btrfs_space_info *sinfo = block_rsv->space_info;
5638 * The global block rsv is based on the size of the extent tree, the
5639 * checksum tree and the root tree. If the fs is empty we want to set
5640 * it to a minimal amount for safety.
5642 num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) +
5643 btrfs_root_used(&fs_info->csum_root->root_item) +
5644 btrfs_root_used(&fs_info->tree_root->root_item);
5645 num_bytes = max_t(u64, num_bytes, SZ_16M);
5647 spin_lock(&sinfo->lock);
5648 spin_lock(&block_rsv->lock);
5650 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5652 if (block_rsv->reserved < block_rsv->size) {
5653 num_bytes = btrfs_space_info_used(sinfo, true);
5654 if (sinfo->total_bytes > num_bytes) {
5655 num_bytes = sinfo->total_bytes - num_bytes;
5656 num_bytes = min(num_bytes,
5657 block_rsv->size - block_rsv->reserved);
5658 block_rsv->reserved += num_bytes;
5659 sinfo->bytes_may_use += num_bytes;
5660 trace_btrfs_space_reservation(fs_info, "space_info",
5661 sinfo->flags, num_bytes,
5664 } else if (block_rsv->reserved > block_rsv->size) {
5665 num_bytes = block_rsv->reserved - block_rsv->size;
5666 sinfo->bytes_may_use -= num_bytes;
5667 trace_btrfs_space_reservation(fs_info, "space_info",
5668 sinfo->flags, num_bytes, 0);
5669 block_rsv->reserved = block_rsv->size;
5672 if (block_rsv->reserved == block_rsv->size)
5673 block_rsv->full = 1;
5675 block_rsv->full = 0;
5677 spin_unlock(&block_rsv->lock);
5678 spin_unlock(&sinfo->lock);
5681 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5683 struct btrfs_space_info *space_info;
5685 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5686 fs_info->chunk_block_rsv.space_info = space_info;
5688 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5689 fs_info->global_block_rsv.space_info = space_info;
5690 fs_info->delalloc_block_rsv.space_info = space_info;
5691 fs_info->trans_block_rsv.space_info = space_info;
5692 fs_info->empty_block_rsv.space_info = space_info;
5693 fs_info->delayed_block_rsv.space_info = space_info;
5695 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5696 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5697 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5698 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5699 if (fs_info->quota_root)
5700 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5701 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5703 update_global_block_rsv(fs_info);
5706 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5708 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5710 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5711 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5712 WARN_ON(fs_info->trans_block_rsv.size > 0);
5713 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5714 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5715 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5716 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5717 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5720 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5721 struct btrfs_fs_info *fs_info)
5723 if (!trans->block_rsv)
5726 if (!trans->bytes_reserved)
5729 trace_btrfs_space_reservation(fs_info, "transaction",
5730 trans->transid, trans->bytes_reserved, 0);
5731 btrfs_block_rsv_release(fs_info, trans->block_rsv,
5732 trans->bytes_reserved);
5733 trans->bytes_reserved = 0;
5737 * To be called after all the new block groups attached to the transaction
5738 * handle have been created (btrfs_create_pending_block_groups()).
5740 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5742 struct btrfs_fs_info *fs_info = trans->fs_info;
5744 if (!trans->chunk_bytes_reserved)
5747 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5749 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5750 trans->chunk_bytes_reserved);
5751 trans->chunk_bytes_reserved = 0;
5754 /* Can only return 0 or -ENOSPC */
5755 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5756 struct btrfs_inode *inode)
5758 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5759 struct btrfs_root *root = inode->root;
5761 * We always use trans->block_rsv here as we will have reserved space
5762 * for our orphan when starting the transaction, using get_block_rsv()
5763 * here will sometimes make us choose the wrong block rsv as we could be
5764 * doing a reloc inode for a non refcounted root.
5766 struct btrfs_block_rsv *src_rsv = trans->block_rsv;
5767 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5770 * We need to hold space in order to delete our orphan item once we've
5771 * added it, so this takes the reservation so we can release it later
5772 * when we are truly done with the orphan item.
5774 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5776 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5778 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5781 void btrfs_orphan_release_metadata(struct btrfs_inode *inode)
5783 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5784 struct btrfs_root *root = inode->root;
5785 u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
5787 trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode),
5789 btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes);
5793 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5794 * root: the root of the parent directory
5795 * rsv: block reservation
5796 * items: the number of items that we need do reservation
5797 * qgroup_reserved: used to return the reserved size in qgroup
5799 * This function is used to reserve the space for snapshot/subvolume
5800 * creation and deletion. Those operations are different with the
5801 * common file/directory operations, they change two fs/file trees
5802 * and root tree, the number of items that the qgroup reserves is
5803 * different with the free space reservation. So we can not use
5804 * the space reservation mechanism in start_transaction().
5806 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5807 struct btrfs_block_rsv *rsv,
5809 u64 *qgroup_reserved,
5810 bool use_global_rsv)
5814 struct btrfs_fs_info *fs_info = root->fs_info;
5815 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5817 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5818 /* One for parent inode, two for dir entries */
5819 num_bytes = 3 * fs_info->nodesize;
5820 ret = btrfs_qgroup_reserve_meta(root, num_bytes, true);
5827 *qgroup_reserved = num_bytes;
5829 num_bytes = btrfs_calc_trans_metadata_size(fs_info, items);
5830 rsv->space_info = __find_space_info(fs_info,
5831 BTRFS_BLOCK_GROUP_METADATA);
5832 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5833 BTRFS_RESERVE_FLUSH_ALL);
5835 if (ret == -ENOSPC && use_global_rsv)
5836 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5838 if (ret && *qgroup_reserved)
5839 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5844 void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info,
5845 struct btrfs_block_rsv *rsv)
5847 btrfs_block_rsv_release(fs_info, rsv, (u64)-1);
5851 * drop_outstanding_extent - drop an outstanding extent
5852 * @inode: the inode we're dropping the extent for
5853 * @num_bytes: the number of bytes we're releasing.
5855 * This is called when we are freeing up an outstanding extent, either called
5856 * after an error or after an extent is written. This will return the number of
5857 * reserved extents that need to be freed. This must be called with
5858 * BTRFS_I(inode)->lock held.
5860 static unsigned drop_outstanding_extent(struct btrfs_inode *inode,
5863 unsigned drop_inode_space = 0;
5864 unsigned dropped_extents = 0;
5865 unsigned num_extents;
5867 num_extents = count_max_extents(num_bytes);
5868 ASSERT(num_extents);
5869 ASSERT(inode->outstanding_extents >= num_extents);
5870 inode->outstanding_extents -= num_extents;
5872 if (inode->outstanding_extents == 0 &&
5873 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5874 &inode->runtime_flags))
5875 drop_inode_space = 1;
5878 * If we have more or the same amount of outstanding extents than we have
5879 * reserved then we need to leave the reserved extents count alone.
5881 if (inode->outstanding_extents >= inode->reserved_extents)
5882 return drop_inode_space;
5884 dropped_extents = inode->reserved_extents - inode->outstanding_extents;
5885 inode->reserved_extents -= dropped_extents;
5886 return dropped_extents + drop_inode_space;
5890 * calc_csum_metadata_size - return the amount of metadata space that must be
5891 * reserved/freed for the given bytes.
5892 * @inode: the inode we're manipulating
5893 * @num_bytes: the number of bytes in question
5894 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5896 * This adjusts the number of csum_bytes in the inode and then returns the
5897 * correct amount of metadata that must either be reserved or freed. We
5898 * calculate how many checksums we can fit into one leaf and then divide the
5899 * number of bytes that will need to be checksumed by this value to figure out
5900 * how many checksums will be required. If we are adding bytes then the number
5901 * may go up and we will return the number of additional bytes that must be
5902 * reserved. If it is going down we will return the number of bytes that must
5905 * This must be called with BTRFS_I(inode)->lock held.
5907 static u64 calc_csum_metadata_size(struct btrfs_inode *inode, u64 num_bytes,
5910 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5911 u64 old_csums, num_csums;
5913 if (inode->flags & BTRFS_INODE_NODATASUM && inode->csum_bytes == 0)
5916 old_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
5918 inode->csum_bytes += num_bytes;
5920 inode->csum_bytes -= num_bytes;
5921 num_csums = btrfs_csum_bytes_to_leaves(fs_info, inode->csum_bytes);
5923 /* No change, no need to reserve more */
5924 if (old_csums == num_csums)
5928 return btrfs_calc_trans_metadata_size(fs_info,
5929 num_csums - old_csums);
5931 return btrfs_calc_trans_metadata_size(fs_info, old_csums - num_csums);
5934 int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes)
5936 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
5937 struct btrfs_root *root = inode->root;
5938 struct btrfs_block_rsv *block_rsv = &fs_info->delalloc_block_rsv;
5941 unsigned nr_extents;
5942 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5944 bool delalloc_lock = true;
5947 bool release_extra = false;
5949 /* If we are a free space inode we need to not flush since we will be in
5950 * the middle of a transaction commit. We also don't need the delalloc
5951 * mutex since we won't race with anybody. We need this mostly to make
5952 * lockdep shut its filthy mouth.
5954 * If we have a transaction open (can happen if we call truncate_block
5955 * from truncate), then we need FLUSH_LIMIT so we don't deadlock.
5957 if (btrfs_is_free_space_inode(inode)) {
5958 flush = BTRFS_RESERVE_NO_FLUSH;
5959 delalloc_lock = false;
5960 } else if (current->journal_info) {
5961 flush = BTRFS_RESERVE_FLUSH_LIMIT;
5964 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5965 btrfs_transaction_in_commit(fs_info))
5966 schedule_timeout(1);
5969 mutex_lock(&inode->delalloc_mutex);
5971 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
5973 spin_lock(&inode->lock);
5974 nr_extents = count_max_extents(num_bytes);
5975 inode->outstanding_extents += nr_extents;
5978 if (inode->outstanding_extents > inode->reserved_extents)
5979 nr_extents += inode->outstanding_extents -
5980 inode->reserved_extents;
5982 /* We always want to reserve a slot for updating the inode. */
5983 to_reserve = btrfs_calc_trans_metadata_size(fs_info, nr_extents + 1);
5984 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5985 csum_bytes = inode->csum_bytes;
5986 spin_unlock(&inode->lock);
5988 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) {
5989 ret = btrfs_qgroup_reserve_meta(root,
5990 nr_extents * fs_info->nodesize, true);
5995 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
5996 if (unlikely(ret)) {
5997 btrfs_qgroup_free_meta(root,
5998 nr_extents * fs_info->nodesize);
6002 spin_lock(&inode->lock);
6003 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6004 &inode->runtime_flags)) {
6005 to_reserve -= btrfs_calc_trans_metadata_size(fs_info, 1);
6006 release_extra = true;
6008 inode->reserved_extents += nr_extents;
6009 spin_unlock(&inode->lock);
6012 mutex_unlock(&inode->delalloc_mutex);
6015 trace_btrfs_space_reservation(fs_info, "delalloc",
6016 btrfs_ino(inode), to_reserve, 1);
6018 btrfs_block_rsv_release(fs_info, block_rsv,
6019 btrfs_calc_trans_metadata_size(fs_info, 1));
6023 spin_lock(&inode->lock);
6024 dropped = drop_outstanding_extent(inode, num_bytes);
6026 * If the inodes csum_bytes is the same as the original
6027 * csum_bytes then we know we haven't raced with any free()ers
6028 * so we can just reduce our inodes csum bytes and carry on.
6030 if (inode->csum_bytes == csum_bytes) {
6031 calc_csum_metadata_size(inode, num_bytes, 0);
6033 u64 orig_csum_bytes = inode->csum_bytes;
6037 * This is tricky, but first we need to figure out how much we
6038 * freed from any free-ers that occurred during this
6039 * reservation, so we reset ->csum_bytes to the csum_bytes
6040 * before we dropped our lock, and then call the free for the
6041 * number of bytes that were freed while we were trying our
6044 bytes = csum_bytes - inode->csum_bytes;
6045 inode->csum_bytes = csum_bytes;
6046 to_free = calc_csum_metadata_size(inode, bytes, 0);
6050 * Now we need to see how much we would have freed had we not
6051 * been making this reservation and our ->csum_bytes were not
6052 * artificially inflated.
6054 inode->csum_bytes = csum_bytes - num_bytes;
6055 bytes = csum_bytes - orig_csum_bytes;
6056 bytes = calc_csum_metadata_size(inode, bytes, 0);
6059 * Now reset ->csum_bytes to what it should be. If bytes is
6060 * more than to_free then we would have freed more space had we
6061 * not had an artificially high ->csum_bytes, so we need to free
6062 * the remainder. If bytes is the same or less then we don't
6063 * need to do anything, the other free-ers did the correct
6066 inode->csum_bytes = orig_csum_bytes - num_bytes;
6067 if (bytes > to_free)
6068 to_free = bytes - to_free;
6072 spin_unlock(&inode->lock);
6074 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6077 btrfs_block_rsv_release(fs_info, block_rsv, to_free);
6078 trace_btrfs_space_reservation(fs_info, "delalloc",
6079 btrfs_ino(inode), to_free, 0);
6082 mutex_unlock(&inode->delalloc_mutex);
6087 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6088 * @inode: the inode to release the reservation for
6089 * @num_bytes: the number of bytes we're releasing
6091 * This will release the metadata reservation for an inode. This can be called
6092 * once we complete IO for a given set of bytes to release their metadata
6095 void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes)
6097 struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
6101 num_bytes = ALIGN(num_bytes, fs_info->sectorsize);
6102 spin_lock(&inode->lock);
6103 dropped = drop_outstanding_extent(inode, num_bytes);
6106 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6107 spin_unlock(&inode->lock);
6109 to_free += btrfs_calc_trans_metadata_size(fs_info, dropped);
6111 if (btrfs_is_testing(fs_info))
6114 trace_btrfs_space_reservation(fs_info, "delalloc", btrfs_ino(inode),
6117 btrfs_block_rsv_release(fs_info, &fs_info->delalloc_block_rsv, to_free);
6121 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6123 * @inode: inode we're writing to
6124 * @start: start range we are writing to
6125 * @len: how long the range we are writing to
6127 * This will do the following things
6129 * o reserve space in data space info for num bytes
6130 * and reserve precious corresponding qgroup space
6131 * (Done in check_data_free_space)
6133 * o reserve space for metadata space, based on the number of outstanding
6134 * extents and how much csums will be needed
6135 * also reserve metadata space in a per root over-reserve method.
6136 * o add to the inodes->delalloc_bytes
6137 * o add it to the fs_info's delalloc inodes list.
6138 * (Above 3 all done in delalloc_reserve_metadata)
6140 * Return 0 for success
6141 * Return <0 for error(-ENOSPC or -EQUOT)
6143 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
6147 ret = btrfs_check_data_free_space(inode, start, len);
6150 ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len);
6152 btrfs_free_reserved_data_space(inode, start, len);
6157 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6158 * @inode: inode we're releasing space for
6159 * @start: start position of the space already reserved
6160 * @len: the len of the space already reserved
6162 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6163 * called in the case that we don't need the metadata AND data reservations
6164 * anymore. So if there is an error or we insert an inline extent.
6166 * This function will release the metadata space that was not used and will
6167 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6168 * list if there are no delalloc bytes left.
6169 * Also it will handle the qgroup reserved space.
6171 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
6173 btrfs_delalloc_release_metadata(BTRFS_I(inode), len);
6174 btrfs_free_reserved_data_space(inode, start, len);
6177 static int update_block_group(struct btrfs_trans_handle *trans,
6178 struct btrfs_fs_info *info, u64 bytenr,
6179 u64 num_bytes, int alloc)
6181 struct btrfs_block_group_cache *cache = NULL;
6182 u64 total = num_bytes;
6187 /* block accounting for super block */
6188 spin_lock(&info->delalloc_root_lock);
6189 old_val = btrfs_super_bytes_used(info->super_copy);
6191 old_val += num_bytes;
6193 old_val -= num_bytes;
6194 btrfs_set_super_bytes_used(info->super_copy, old_val);
6195 spin_unlock(&info->delalloc_root_lock);
6198 cache = btrfs_lookup_block_group(info, bytenr);
6201 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6202 BTRFS_BLOCK_GROUP_RAID1 |
6203 BTRFS_BLOCK_GROUP_RAID10))
6208 * If this block group has free space cache written out, we
6209 * need to make sure to load it if we are removing space. This
6210 * is because we need the unpinning stage to actually add the
6211 * space back to the block group, otherwise we will leak space.
6213 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6214 cache_block_group(cache, 1);
6216 byte_in_group = bytenr - cache->key.objectid;
6217 WARN_ON(byte_in_group > cache->key.offset);
6219 spin_lock(&cache->space_info->lock);
6220 spin_lock(&cache->lock);
6222 if (btrfs_test_opt(info, SPACE_CACHE) &&
6223 cache->disk_cache_state < BTRFS_DC_CLEAR)
6224 cache->disk_cache_state = BTRFS_DC_CLEAR;
6226 old_val = btrfs_block_group_used(&cache->item);
6227 num_bytes = min(total, cache->key.offset - byte_in_group);
6229 old_val += num_bytes;
6230 btrfs_set_block_group_used(&cache->item, old_val);
6231 cache->reserved -= num_bytes;
6232 cache->space_info->bytes_reserved -= num_bytes;
6233 cache->space_info->bytes_used += num_bytes;
6234 cache->space_info->disk_used += num_bytes * factor;
6235 spin_unlock(&cache->lock);
6236 spin_unlock(&cache->space_info->lock);
6238 old_val -= num_bytes;
6239 btrfs_set_block_group_used(&cache->item, old_val);
6240 cache->pinned += num_bytes;
6241 cache->space_info->bytes_pinned += num_bytes;
6242 cache->space_info->bytes_used -= num_bytes;
6243 cache->space_info->disk_used -= num_bytes * factor;
6244 spin_unlock(&cache->lock);
6245 spin_unlock(&cache->space_info->lock);
6247 trace_btrfs_space_reservation(info, "pinned",
6248 cache->space_info->flags,
6250 set_extent_dirty(info->pinned_extents,
6251 bytenr, bytenr + num_bytes - 1,
6252 GFP_NOFS | __GFP_NOFAIL);
6255 spin_lock(&trans->transaction->dirty_bgs_lock);
6256 if (list_empty(&cache->dirty_list)) {
6257 list_add_tail(&cache->dirty_list,
6258 &trans->transaction->dirty_bgs);
6259 trans->transaction->num_dirty_bgs++;
6260 btrfs_get_block_group(cache);
6262 spin_unlock(&trans->transaction->dirty_bgs_lock);
6265 * No longer have used bytes in this block group, queue it for
6266 * deletion. We do this after adding the block group to the
6267 * dirty list to avoid races between cleaner kthread and space
6270 if (!alloc && old_val == 0) {
6271 spin_lock(&info->unused_bgs_lock);
6272 if (list_empty(&cache->bg_list)) {
6273 btrfs_get_block_group(cache);
6274 list_add_tail(&cache->bg_list,
6277 spin_unlock(&info->unused_bgs_lock);
6280 btrfs_put_block_group(cache);
6282 bytenr += num_bytes;
6287 static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start)
6289 struct btrfs_block_group_cache *cache;
6292 spin_lock(&fs_info->block_group_cache_lock);
6293 bytenr = fs_info->first_logical_byte;
6294 spin_unlock(&fs_info->block_group_cache_lock);
6296 if (bytenr < (u64)-1)
6299 cache = btrfs_lookup_first_block_group(fs_info, search_start);
6303 bytenr = cache->key.objectid;
6304 btrfs_put_block_group(cache);
6309 static int pin_down_extent(struct btrfs_fs_info *fs_info,
6310 struct btrfs_block_group_cache *cache,
6311 u64 bytenr, u64 num_bytes, int reserved)
6313 spin_lock(&cache->space_info->lock);
6314 spin_lock(&cache->lock);
6315 cache->pinned += num_bytes;
6316 cache->space_info->bytes_pinned += num_bytes;
6318 cache->reserved -= num_bytes;
6319 cache->space_info->bytes_reserved -= num_bytes;
6321 spin_unlock(&cache->lock);
6322 spin_unlock(&cache->space_info->lock);
6324 trace_btrfs_space_reservation(fs_info, "pinned",
6325 cache->space_info->flags, num_bytes, 1);
6326 set_extent_dirty(fs_info->pinned_extents, bytenr,
6327 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6332 * this function must be called within transaction
6334 int btrfs_pin_extent(struct btrfs_fs_info *fs_info,
6335 u64 bytenr, u64 num_bytes, int reserved)
6337 struct btrfs_block_group_cache *cache;
6339 cache = btrfs_lookup_block_group(fs_info, bytenr);
6340 BUG_ON(!cache); /* Logic error */
6342 pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved);
6344 btrfs_put_block_group(cache);
6349 * this function must be called within transaction
6351 int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info,
6352 u64 bytenr, u64 num_bytes)
6354 struct btrfs_block_group_cache *cache;
6357 cache = btrfs_lookup_block_group(fs_info, bytenr);
6362 * pull in the free space cache (if any) so that our pin
6363 * removes the free space from the cache. We have load_only set
6364 * to one because the slow code to read in the free extents does check
6365 * the pinned extents.
6367 cache_block_group(cache, 1);
6369 pin_down_extent(fs_info, cache, bytenr, num_bytes, 0);
6371 /* remove us from the free space cache (if we're there at all) */
6372 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6373 btrfs_put_block_group(cache);
6377 static int __exclude_logged_extent(struct btrfs_fs_info *fs_info,
6378 u64 start, u64 num_bytes)
6381 struct btrfs_block_group_cache *block_group;
6382 struct btrfs_caching_control *caching_ctl;
6384 block_group = btrfs_lookup_block_group(fs_info, start);
6388 cache_block_group(block_group, 0);
6389 caching_ctl = get_caching_control(block_group);
6393 BUG_ON(!block_group_cache_done(block_group));
6394 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6396 mutex_lock(&caching_ctl->mutex);
6398 if (start >= caching_ctl->progress) {
6399 ret = add_excluded_extent(fs_info, start, num_bytes);
6400 } else if (start + num_bytes <= caching_ctl->progress) {
6401 ret = btrfs_remove_free_space(block_group,
6404 num_bytes = caching_ctl->progress - start;
6405 ret = btrfs_remove_free_space(block_group,
6410 num_bytes = (start + num_bytes) -
6411 caching_ctl->progress;
6412 start = caching_ctl->progress;
6413 ret = add_excluded_extent(fs_info, start, num_bytes);
6416 mutex_unlock(&caching_ctl->mutex);
6417 put_caching_control(caching_ctl);
6419 btrfs_put_block_group(block_group);
6423 int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info,
6424 struct extent_buffer *eb)
6426 struct btrfs_file_extent_item *item;
6427 struct btrfs_key key;
6431 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS))
6434 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6435 btrfs_item_key_to_cpu(eb, &key, i);
6436 if (key.type != BTRFS_EXTENT_DATA_KEY)
6438 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6439 found_type = btrfs_file_extent_type(eb, item);
6440 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6442 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6444 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6445 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6446 __exclude_logged_extent(fs_info, key.objectid, key.offset);
6453 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6455 atomic_inc(&bg->reservations);
6458 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6461 struct btrfs_block_group_cache *bg;
6463 bg = btrfs_lookup_block_group(fs_info, start);
6465 if (atomic_dec_and_test(&bg->reservations))
6466 wake_up_atomic_t(&bg->reservations);
6467 btrfs_put_block_group(bg);
6470 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6476 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6478 struct btrfs_space_info *space_info = bg->space_info;
6482 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6486 * Our block group is read only but before we set it to read only,
6487 * some task might have had allocated an extent from it already, but it
6488 * has not yet created a respective ordered extent (and added it to a
6489 * root's list of ordered extents).
6490 * Therefore wait for any task currently allocating extents, since the
6491 * block group's reservations counter is incremented while a read lock
6492 * on the groups' semaphore is held and decremented after releasing
6493 * the read access on that semaphore and creating the ordered extent.
6495 down_write(&space_info->groups_sem);
6496 up_write(&space_info->groups_sem);
6498 wait_on_atomic_t(&bg->reservations,
6499 btrfs_wait_bg_reservations_atomic_t,
6500 TASK_UNINTERRUPTIBLE);
6504 * btrfs_add_reserved_bytes - update the block_group and space info counters
6505 * @cache: The cache we are manipulating
6506 * @ram_bytes: The number of bytes of file content, and will be same to
6507 * @num_bytes except for the compress path.
6508 * @num_bytes: The number of bytes in question
6509 * @delalloc: The blocks are allocated for the delalloc write
6511 * This is called by the allocator when it reserves space. If this is a
6512 * reservation and the block group has become read only we cannot make the
6513 * reservation and return -EAGAIN, otherwise this function always succeeds.
6515 static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache,
6516 u64 ram_bytes, u64 num_bytes, int delalloc)
6518 struct btrfs_space_info *space_info = cache->space_info;
6521 spin_lock(&space_info->lock);
6522 spin_lock(&cache->lock);
6526 cache->reserved += num_bytes;
6527 space_info->bytes_reserved += num_bytes;
6529 trace_btrfs_space_reservation(cache->fs_info,
6530 "space_info", space_info->flags,
6532 space_info->bytes_may_use -= ram_bytes;
6534 cache->delalloc_bytes += num_bytes;
6536 spin_unlock(&cache->lock);
6537 spin_unlock(&space_info->lock);
6542 * btrfs_free_reserved_bytes - update the block_group and space info counters
6543 * @cache: The cache we are manipulating
6544 * @num_bytes: The number of bytes in question
6545 * @delalloc: The blocks are allocated for the delalloc write
6547 * This is called by somebody who is freeing space that was never actually used
6548 * on disk. For example if you reserve some space for a new leaf in transaction
6549 * A and before transaction A commits you free that leaf, you call this with
6550 * reserve set to 0 in order to clear the reservation.
6553 static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache,
6554 u64 num_bytes, int delalloc)
6556 struct btrfs_space_info *space_info = cache->space_info;
6559 spin_lock(&space_info->lock);
6560 spin_lock(&cache->lock);
6562 space_info->bytes_readonly += num_bytes;
6563 cache->reserved -= num_bytes;
6564 space_info->bytes_reserved -= num_bytes;
6567 cache->delalloc_bytes -= num_bytes;
6568 spin_unlock(&cache->lock);
6569 spin_unlock(&space_info->lock);
6572 void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info)
6574 struct btrfs_caching_control *next;
6575 struct btrfs_caching_control *caching_ctl;
6576 struct btrfs_block_group_cache *cache;
6578 down_write(&fs_info->commit_root_sem);
6580 list_for_each_entry_safe(caching_ctl, next,
6581 &fs_info->caching_block_groups, list) {
6582 cache = caching_ctl->block_group;
6583 if (block_group_cache_done(cache)) {
6584 cache->last_byte_to_unpin = (u64)-1;
6585 list_del_init(&caching_ctl->list);
6586 put_caching_control(caching_ctl);
6588 cache->last_byte_to_unpin = caching_ctl->progress;
6592 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6593 fs_info->pinned_extents = &fs_info->freed_extents[1];
6595 fs_info->pinned_extents = &fs_info->freed_extents[0];
6597 up_write(&fs_info->commit_root_sem);
6599 update_global_block_rsv(fs_info);
6603 * Returns the free cluster for the given space info and sets empty_cluster to
6604 * what it should be based on the mount options.
6606 static struct btrfs_free_cluster *
6607 fetch_cluster_info(struct btrfs_fs_info *fs_info,
6608 struct btrfs_space_info *space_info, u64 *empty_cluster)
6610 struct btrfs_free_cluster *ret = NULL;
6611 bool ssd = btrfs_test_opt(fs_info, SSD);
6614 if (btrfs_mixed_space_info(space_info))
6618 *empty_cluster = SZ_2M;
6619 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6620 ret = &fs_info->meta_alloc_cluster;
6622 *empty_cluster = SZ_64K;
6623 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6624 ret = &fs_info->data_alloc_cluster;
6630 static int unpin_extent_range(struct btrfs_fs_info *fs_info,
6632 const bool return_free_space)
6634 struct btrfs_block_group_cache *cache = NULL;
6635 struct btrfs_space_info *space_info;
6636 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6637 struct btrfs_free_cluster *cluster = NULL;
6639 u64 total_unpinned = 0;
6640 u64 empty_cluster = 0;
6643 while (start <= end) {
6646 start >= cache->key.objectid + cache->key.offset) {
6648 btrfs_put_block_group(cache);
6650 cache = btrfs_lookup_block_group(fs_info, start);
6651 BUG_ON(!cache); /* Logic error */
6653 cluster = fetch_cluster_info(fs_info,
6656 empty_cluster <<= 1;
6659 len = cache->key.objectid + cache->key.offset - start;
6660 len = min(len, end + 1 - start);
6662 if (start < cache->last_byte_to_unpin) {
6663 len = min(len, cache->last_byte_to_unpin - start);
6664 if (return_free_space)
6665 btrfs_add_free_space(cache, start, len);
6669 total_unpinned += len;
6670 space_info = cache->space_info;
6673 * If this space cluster has been marked as fragmented and we've
6674 * unpinned enough in this block group to potentially allow a
6675 * cluster to be created inside of it go ahead and clear the
6678 if (cluster && cluster->fragmented &&
6679 total_unpinned > empty_cluster) {
6680 spin_lock(&cluster->lock);
6681 cluster->fragmented = 0;
6682 spin_unlock(&cluster->lock);
6685 spin_lock(&space_info->lock);
6686 spin_lock(&cache->lock);
6687 cache->pinned -= len;
6688 space_info->bytes_pinned -= len;
6690 trace_btrfs_space_reservation(fs_info, "pinned",
6691 space_info->flags, len, 0);
6692 space_info->max_extent_size = 0;
6693 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6695 space_info->bytes_readonly += len;
6698 spin_unlock(&cache->lock);
6699 if (!readonly && return_free_space &&
6700 global_rsv->space_info == space_info) {
6702 WARN_ON(!return_free_space);
6703 spin_lock(&global_rsv->lock);
6704 if (!global_rsv->full) {
6705 to_add = min(len, global_rsv->size -
6706 global_rsv->reserved);
6707 global_rsv->reserved += to_add;
6708 space_info->bytes_may_use += to_add;
6709 if (global_rsv->reserved >= global_rsv->size)
6710 global_rsv->full = 1;
6711 trace_btrfs_space_reservation(fs_info,
6717 spin_unlock(&global_rsv->lock);
6718 /* Add to any tickets we may have */
6720 space_info_add_new_bytes(fs_info, space_info,
6723 spin_unlock(&space_info->lock);
6727 btrfs_put_block_group(cache);
6731 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6732 struct btrfs_fs_info *fs_info)
6734 struct btrfs_block_group_cache *block_group, *tmp;
6735 struct list_head *deleted_bgs;
6736 struct extent_io_tree *unpin;
6741 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6742 unpin = &fs_info->freed_extents[1];
6744 unpin = &fs_info->freed_extents[0];
6746 while (!trans->aborted) {
6747 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6748 ret = find_first_extent_bit(unpin, 0, &start, &end,
6749 EXTENT_DIRTY, NULL);
6751 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6755 if (btrfs_test_opt(fs_info, DISCARD))
6756 ret = btrfs_discard_extent(fs_info, start,
6757 end + 1 - start, NULL);
6759 clear_extent_dirty(unpin, start, end);
6760 unpin_extent_range(fs_info, start, end, true);
6761 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6766 * Transaction is finished. We don't need the lock anymore. We
6767 * do need to clean up the block groups in case of a transaction
6770 deleted_bgs = &trans->transaction->deleted_bgs;
6771 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6775 if (!trans->aborted)
6776 ret = btrfs_discard_extent(fs_info,
6777 block_group->key.objectid,
6778 block_group->key.offset,
6781 list_del_init(&block_group->bg_list);
6782 btrfs_put_block_group_trimming(block_group);
6783 btrfs_put_block_group(block_group);
6786 const char *errstr = btrfs_decode_error(ret);
6788 "Discard failed while removing blockgroup: errno=%d %s\n",
6796 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6797 u64 owner, u64 root_objectid)
6799 struct btrfs_space_info *space_info;
6802 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6803 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6804 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6806 flags = BTRFS_BLOCK_GROUP_METADATA;
6808 flags = BTRFS_BLOCK_GROUP_DATA;
6811 space_info = __find_space_info(fs_info, flags);
6812 BUG_ON(!space_info); /* Logic bug */
6813 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6817 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6818 struct btrfs_fs_info *info,
6819 struct btrfs_delayed_ref_node *node, u64 parent,
6820 u64 root_objectid, u64 owner_objectid,
6821 u64 owner_offset, int refs_to_drop,
6822 struct btrfs_delayed_extent_op *extent_op)
6824 struct btrfs_key key;
6825 struct btrfs_path *path;
6826 struct btrfs_root *extent_root = info->extent_root;
6827 struct extent_buffer *leaf;
6828 struct btrfs_extent_item *ei;
6829 struct btrfs_extent_inline_ref *iref;
6832 int extent_slot = 0;
6833 int found_extent = 0;
6837 u64 bytenr = node->bytenr;
6838 u64 num_bytes = node->num_bytes;
6840 bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA);
6842 path = btrfs_alloc_path();
6846 path->reada = READA_FORWARD;
6847 path->leave_spinning = 1;
6849 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6850 BUG_ON(!is_data && refs_to_drop != 1);
6853 skinny_metadata = 0;
6855 ret = lookup_extent_backref(trans, info, path, &iref,
6856 bytenr, num_bytes, parent,
6857 root_objectid, owner_objectid,
6860 extent_slot = path->slots[0];
6861 while (extent_slot >= 0) {
6862 btrfs_item_key_to_cpu(path->nodes[0], &key,
6864 if (key.objectid != bytenr)
6866 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6867 key.offset == num_bytes) {
6871 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6872 key.offset == owner_objectid) {
6876 if (path->slots[0] - extent_slot > 5)
6880 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6881 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6882 if (found_extent && item_size < sizeof(*ei))
6885 if (!found_extent) {
6887 ret = remove_extent_backref(trans, info, path, NULL,
6889 is_data, &last_ref);
6891 btrfs_abort_transaction(trans, ret);
6894 btrfs_release_path(path);
6895 path->leave_spinning = 1;
6897 key.objectid = bytenr;
6898 key.type = BTRFS_EXTENT_ITEM_KEY;
6899 key.offset = num_bytes;
6901 if (!is_data && skinny_metadata) {
6902 key.type = BTRFS_METADATA_ITEM_KEY;
6903 key.offset = owner_objectid;
6906 ret = btrfs_search_slot(trans, extent_root,
6908 if (ret > 0 && skinny_metadata && path->slots[0]) {
6910 * Couldn't find our skinny metadata item,
6911 * see if we have ye olde extent item.
6914 btrfs_item_key_to_cpu(path->nodes[0], &key,
6916 if (key.objectid == bytenr &&
6917 key.type == BTRFS_EXTENT_ITEM_KEY &&
6918 key.offset == num_bytes)
6922 if (ret > 0 && skinny_metadata) {
6923 skinny_metadata = false;
6924 key.objectid = bytenr;
6925 key.type = BTRFS_EXTENT_ITEM_KEY;
6926 key.offset = num_bytes;
6927 btrfs_release_path(path);
6928 ret = btrfs_search_slot(trans, extent_root,
6934 "umm, got %d back from search, was looking for %llu",
6937 btrfs_print_leaf(info, path->nodes[0]);
6940 btrfs_abort_transaction(trans, ret);
6943 extent_slot = path->slots[0];
6945 } else if (WARN_ON(ret == -ENOENT)) {
6946 btrfs_print_leaf(info, path->nodes[0]);
6948 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6949 bytenr, parent, root_objectid, owner_objectid,
6951 btrfs_abort_transaction(trans, ret);
6954 btrfs_abort_transaction(trans, ret);
6958 leaf = path->nodes[0];
6959 item_size = btrfs_item_size_nr(leaf, extent_slot);
6960 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6961 if (item_size < sizeof(*ei)) {
6962 BUG_ON(found_extent || extent_slot != path->slots[0]);
6963 ret = convert_extent_item_v0(trans, info, path, owner_objectid,
6966 btrfs_abort_transaction(trans, ret);
6970 btrfs_release_path(path);
6971 path->leave_spinning = 1;
6973 key.objectid = bytenr;
6974 key.type = BTRFS_EXTENT_ITEM_KEY;
6975 key.offset = num_bytes;
6977 ret = btrfs_search_slot(trans, extent_root, &key, path,
6981 "umm, got %d back from search, was looking for %llu",
6983 btrfs_print_leaf(info, path->nodes[0]);
6986 btrfs_abort_transaction(trans, ret);
6990 extent_slot = path->slots[0];
6991 leaf = path->nodes[0];
6992 item_size = btrfs_item_size_nr(leaf, extent_slot);
6995 BUG_ON(item_size < sizeof(*ei));
6996 ei = btrfs_item_ptr(leaf, extent_slot,
6997 struct btrfs_extent_item);
6998 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6999 key.type == BTRFS_EXTENT_ITEM_KEY) {
7000 struct btrfs_tree_block_info *bi;
7001 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7002 bi = (struct btrfs_tree_block_info *)(ei + 1);
7003 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7006 refs = btrfs_extent_refs(leaf, ei);
7007 if (refs < refs_to_drop) {
7009 "trying to drop %d refs but we only have %Lu for bytenr %Lu",
7010 refs_to_drop, refs, bytenr);
7012 btrfs_abort_transaction(trans, ret);
7015 refs -= refs_to_drop;
7019 __run_delayed_extent_op(extent_op, leaf, ei);
7021 * In the case of inline back ref, reference count will
7022 * be updated by remove_extent_backref
7025 BUG_ON(!found_extent);
7027 btrfs_set_extent_refs(leaf, ei, refs);
7028 btrfs_mark_buffer_dirty(leaf);
7031 ret = remove_extent_backref(trans, info, path,
7033 is_data, &last_ref);
7035 btrfs_abort_transaction(trans, ret);
7039 add_pinned_bytes(info, -num_bytes, owner_objectid,
7043 BUG_ON(is_data && refs_to_drop !=
7044 extent_data_ref_count(path, iref));
7046 BUG_ON(path->slots[0] != extent_slot);
7048 BUG_ON(path->slots[0] != extent_slot + 1);
7049 path->slots[0] = extent_slot;
7055 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7058 btrfs_abort_transaction(trans, ret);
7061 btrfs_release_path(path);
7064 ret = btrfs_del_csums(trans, info, bytenr, num_bytes);
7066 btrfs_abort_transaction(trans, ret);
7071 ret = add_to_free_space_tree(trans, info, bytenr, num_bytes);
7073 btrfs_abort_transaction(trans, ret);
7077 ret = update_block_group(trans, info, bytenr, num_bytes, 0);
7079 btrfs_abort_transaction(trans, ret);
7083 btrfs_release_path(path);
7086 btrfs_free_path(path);
7091 * when we free an block, it is possible (and likely) that we free the last
7092 * delayed ref for that extent as well. This searches the delayed ref tree for
7093 * a given extent, and if there are no other delayed refs to be processed, it
7094 * removes it from the tree.
7096 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7099 struct btrfs_delayed_ref_head *head;
7100 struct btrfs_delayed_ref_root *delayed_refs;
7103 delayed_refs = &trans->transaction->delayed_refs;
7104 spin_lock(&delayed_refs->lock);
7105 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
7107 goto out_delayed_unlock;
7109 spin_lock(&head->lock);
7110 if (!list_empty(&head->ref_list))
7113 if (head->extent_op) {
7114 if (!head->must_insert_reserved)
7116 btrfs_free_delayed_extent_op(head->extent_op);
7117 head->extent_op = NULL;
7121 * waiting for the lock here would deadlock. If someone else has it
7122 * locked they are already in the process of dropping it anyway
7124 if (!mutex_trylock(&head->mutex))
7128 * at this point we have a head with no other entries. Go
7129 * ahead and process it.
7131 head->node.in_tree = 0;
7132 rb_erase(&head->href_node, &delayed_refs->href_root);
7134 atomic_dec(&delayed_refs->num_entries);
7137 * we don't take a ref on the node because we're removing it from the
7138 * tree, so we just steal the ref the tree was holding.
7140 delayed_refs->num_heads--;
7141 if (head->processing == 0)
7142 delayed_refs->num_heads_ready--;
7143 head->processing = 0;
7144 spin_unlock(&head->lock);
7145 spin_unlock(&delayed_refs->lock);
7147 BUG_ON(head->extent_op);
7148 if (head->must_insert_reserved)
7151 mutex_unlock(&head->mutex);
7152 btrfs_put_delayed_ref(&head->node);
7155 spin_unlock(&head->lock);
7158 spin_unlock(&delayed_refs->lock);
7162 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7163 struct btrfs_root *root,
7164 struct extent_buffer *buf,
7165 u64 parent, int last_ref)
7167 struct btrfs_fs_info *fs_info = root->fs_info;
7171 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7172 ret = btrfs_add_delayed_tree_ref(fs_info, trans,
7173 buf->start, buf->len,
7175 root->root_key.objectid,
7176 btrfs_header_level(buf),
7177 BTRFS_DROP_DELAYED_REF, NULL);
7178 BUG_ON(ret); /* -ENOMEM */
7184 if (btrfs_header_generation(buf) == trans->transid) {
7185 struct btrfs_block_group_cache *cache;
7187 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7188 ret = check_ref_cleanup(trans, buf->start);
7193 cache = btrfs_lookup_block_group(fs_info, buf->start);
7195 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7196 pin_down_extent(fs_info, cache, buf->start,
7198 btrfs_put_block_group(cache);
7202 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7204 btrfs_add_free_space(cache, buf->start, buf->len);
7205 btrfs_free_reserved_bytes(cache, buf->len, 0);
7206 btrfs_put_block_group(cache);
7207 trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len);
7212 add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf),
7213 root->root_key.objectid);
7216 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7219 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7222 /* Can return -ENOMEM */
7223 int btrfs_free_extent(struct btrfs_trans_handle *trans,
7224 struct btrfs_fs_info *fs_info,
7225 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7226 u64 owner, u64 offset)
7230 if (btrfs_is_testing(fs_info))
7233 add_pinned_bytes(fs_info, num_bytes, owner, root_objectid);
7236 * tree log blocks never actually go into the extent allocation
7237 * tree, just update pinning info and exit early.
7239 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7240 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7241 /* unlocks the pinned mutex */
7242 btrfs_pin_extent(fs_info, bytenr, num_bytes, 1);
7244 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7245 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7247 parent, root_objectid, (int)owner,
7248 BTRFS_DROP_DELAYED_REF, NULL);
7250 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7252 parent, root_objectid, owner,
7254 BTRFS_DROP_DELAYED_REF);
7260 * when we wait for progress in the block group caching, its because
7261 * our allocation attempt failed at least once. So, we must sleep
7262 * and let some progress happen before we try again.
7264 * This function will sleep at least once waiting for new free space to
7265 * show up, and then it will check the block group free space numbers
7266 * for our min num_bytes. Another option is to have it go ahead
7267 * and look in the rbtree for a free extent of a given size, but this
7270 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7271 * any of the information in this block group.
7273 static noinline void
7274 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7277 struct btrfs_caching_control *caching_ctl;
7279 caching_ctl = get_caching_control(cache);
7283 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7284 (cache->free_space_ctl->free_space >= num_bytes));
7286 put_caching_control(caching_ctl);
7290 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7292 struct btrfs_caching_control *caching_ctl;
7295 caching_ctl = get_caching_control(cache);
7297 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7299 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7300 if (cache->cached == BTRFS_CACHE_ERROR)
7302 put_caching_control(caching_ctl);
7306 int __get_raid_index(u64 flags)
7308 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7309 return BTRFS_RAID_RAID10;
7310 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7311 return BTRFS_RAID_RAID1;
7312 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7313 return BTRFS_RAID_DUP;
7314 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7315 return BTRFS_RAID_RAID0;
7316 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7317 return BTRFS_RAID_RAID5;
7318 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7319 return BTRFS_RAID_RAID6;
7321 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7324 int get_block_group_index(struct btrfs_block_group_cache *cache)
7326 return __get_raid_index(cache->flags);
7329 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7330 [BTRFS_RAID_RAID10] = "raid10",
7331 [BTRFS_RAID_RAID1] = "raid1",
7332 [BTRFS_RAID_DUP] = "dup",
7333 [BTRFS_RAID_RAID0] = "raid0",
7334 [BTRFS_RAID_SINGLE] = "single",
7335 [BTRFS_RAID_RAID5] = "raid5",
7336 [BTRFS_RAID_RAID6] = "raid6",
7339 static const char *get_raid_name(enum btrfs_raid_types type)
7341 if (type >= BTRFS_NR_RAID_TYPES)
7344 return btrfs_raid_type_names[type];
7347 enum btrfs_loop_type {
7348 LOOP_CACHING_NOWAIT = 0,
7349 LOOP_CACHING_WAIT = 1,
7350 LOOP_ALLOC_CHUNK = 2,
7351 LOOP_NO_EMPTY_SIZE = 3,
7355 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7359 down_read(&cache->data_rwsem);
7363 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7366 btrfs_get_block_group(cache);
7368 down_read(&cache->data_rwsem);
7371 static struct btrfs_block_group_cache *
7372 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7373 struct btrfs_free_cluster *cluster,
7376 struct btrfs_block_group_cache *used_bg = NULL;
7378 spin_lock(&cluster->refill_lock);
7380 used_bg = cluster->block_group;
7384 if (used_bg == block_group)
7387 btrfs_get_block_group(used_bg);
7392 if (down_read_trylock(&used_bg->data_rwsem))
7395 spin_unlock(&cluster->refill_lock);
7397 /* We should only have one-level nested. */
7398 down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING);
7400 spin_lock(&cluster->refill_lock);
7401 if (used_bg == cluster->block_group)
7404 up_read(&used_bg->data_rwsem);
7405 btrfs_put_block_group(used_bg);
7410 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7414 up_read(&cache->data_rwsem);
7415 btrfs_put_block_group(cache);
7419 * walks the btree of allocated extents and find a hole of a given size.
7420 * The key ins is changed to record the hole:
7421 * ins->objectid == start position
7422 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7423 * ins->offset == the size of the hole.
7424 * Any available blocks before search_start are skipped.
7426 * If there is no suitable free space, we will record the max size of
7427 * the free space extent currently.
7429 static noinline int find_free_extent(struct btrfs_fs_info *fs_info,
7430 u64 ram_bytes, u64 num_bytes, u64 empty_size,
7431 u64 hint_byte, struct btrfs_key *ins,
7432 u64 flags, int delalloc)
7435 struct btrfs_root *root = fs_info->extent_root;
7436 struct btrfs_free_cluster *last_ptr = NULL;
7437 struct btrfs_block_group_cache *block_group = NULL;
7438 u64 search_start = 0;
7439 u64 max_extent_size = 0;
7440 u64 empty_cluster = 0;
7441 struct btrfs_space_info *space_info;
7443 int index = __get_raid_index(flags);
7444 bool failed_cluster_refill = false;
7445 bool failed_alloc = false;
7446 bool use_cluster = true;
7447 bool have_caching_bg = false;
7448 bool orig_have_caching_bg = false;
7449 bool full_search = false;
7451 WARN_ON(num_bytes < fs_info->sectorsize);
7452 ins->type = BTRFS_EXTENT_ITEM_KEY;
7456 trace_find_free_extent(fs_info, num_bytes, empty_size, flags);
7458 space_info = __find_space_info(fs_info, flags);
7460 btrfs_err(fs_info, "No space info for %llu", flags);
7465 * If our free space is heavily fragmented we may not be able to make
7466 * big contiguous allocations, so instead of doing the expensive search
7467 * for free space, simply return ENOSPC with our max_extent_size so we
7468 * can go ahead and search for a more manageable chunk.
7470 * If our max_extent_size is large enough for our allocation simply
7471 * disable clustering since we will likely not be able to find enough
7472 * space to create a cluster and induce latency trying.
7474 if (unlikely(space_info->max_extent_size)) {
7475 spin_lock(&space_info->lock);
7476 if (space_info->max_extent_size &&
7477 num_bytes > space_info->max_extent_size) {
7478 ins->offset = space_info->max_extent_size;
7479 spin_unlock(&space_info->lock);
7481 } else if (space_info->max_extent_size) {
7482 use_cluster = false;
7484 spin_unlock(&space_info->lock);
7487 last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster);
7489 spin_lock(&last_ptr->lock);
7490 if (last_ptr->block_group)
7491 hint_byte = last_ptr->window_start;
7492 if (last_ptr->fragmented) {
7494 * We still set window_start so we can keep track of the
7495 * last place we found an allocation to try and save
7498 hint_byte = last_ptr->window_start;
7499 use_cluster = false;
7501 spin_unlock(&last_ptr->lock);
7504 search_start = max(search_start, first_logical_byte(fs_info, 0));
7505 search_start = max(search_start, hint_byte);
7506 if (search_start == hint_byte) {
7507 block_group = btrfs_lookup_block_group(fs_info, search_start);
7509 * we don't want to use the block group if it doesn't match our
7510 * allocation bits, or if its not cached.
7512 * However if we are re-searching with an ideal block group
7513 * picked out then we don't care that the block group is cached.
7515 if (block_group && block_group_bits(block_group, flags) &&
7516 block_group->cached != BTRFS_CACHE_NO) {
7517 down_read(&space_info->groups_sem);
7518 if (list_empty(&block_group->list) ||
7521 * someone is removing this block group,
7522 * we can't jump into the have_block_group
7523 * target because our list pointers are not
7526 btrfs_put_block_group(block_group);
7527 up_read(&space_info->groups_sem);
7529 index = get_block_group_index(block_group);
7530 btrfs_lock_block_group(block_group, delalloc);
7531 goto have_block_group;
7533 } else if (block_group) {
7534 btrfs_put_block_group(block_group);
7538 have_caching_bg = false;
7539 if (index == 0 || index == __get_raid_index(flags))
7541 down_read(&space_info->groups_sem);
7542 list_for_each_entry(block_group, &space_info->block_groups[index],
7547 btrfs_grab_block_group(block_group, delalloc);
7548 search_start = block_group->key.objectid;
7551 * this can happen if we end up cycling through all the
7552 * raid types, but we want to make sure we only allocate
7553 * for the proper type.
7555 if (!block_group_bits(block_group, flags)) {
7556 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7557 BTRFS_BLOCK_GROUP_RAID1 |
7558 BTRFS_BLOCK_GROUP_RAID5 |
7559 BTRFS_BLOCK_GROUP_RAID6 |
7560 BTRFS_BLOCK_GROUP_RAID10;
7563 * if they asked for extra copies and this block group
7564 * doesn't provide them, bail. This does allow us to
7565 * fill raid0 from raid1.
7567 if ((flags & extra) && !(block_group->flags & extra))
7572 cached = block_group_cache_done(block_group);
7573 if (unlikely(!cached)) {
7574 have_caching_bg = true;
7575 ret = cache_block_group(block_group, 0);
7580 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7582 if (unlikely(block_group->ro))
7586 * Ok we want to try and use the cluster allocator, so
7589 if (last_ptr && use_cluster) {
7590 struct btrfs_block_group_cache *used_block_group;
7591 unsigned long aligned_cluster;
7593 * the refill lock keeps out other
7594 * people trying to start a new cluster
7596 used_block_group = btrfs_lock_cluster(block_group,
7599 if (!used_block_group)
7600 goto refill_cluster;
7602 if (used_block_group != block_group &&
7603 (used_block_group->ro ||
7604 !block_group_bits(used_block_group, flags)))
7605 goto release_cluster;
7607 offset = btrfs_alloc_from_cluster(used_block_group,
7610 used_block_group->key.objectid,
7613 /* we have a block, we're done */
7614 spin_unlock(&last_ptr->refill_lock);
7615 trace_btrfs_reserve_extent_cluster(fs_info,
7617 search_start, num_bytes);
7618 if (used_block_group != block_group) {
7619 btrfs_release_block_group(block_group,
7621 block_group = used_block_group;
7626 WARN_ON(last_ptr->block_group != used_block_group);
7628 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7629 * set up a new clusters, so lets just skip it
7630 * and let the allocator find whatever block
7631 * it can find. If we reach this point, we
7632 * will have tried the cluster allocator
7633 * plenty of times and not have found
7634 * anything, so we are likely way too
7635 * fragmented for the clustering stuff to find
7638 * However, if the cluster is taken from the
7639 * current block group, release the cluster
7640 * first, so that we stand a better chance of
7641 * succeeding in the unclustered
7643 if (loop >= LOOP_NO_EMPTY_SIZE &&
7644 used_block_group != block_group) {
7645 spin_unlock(&last_ptr->refill_lock);
7646 btrfs_release_block_group(used_block_group,
7648 goto unclustered_alloc;
7652 * this cluster didn't work out, free it and
7655 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7657 if (used_block_group != block_group)
7658 btrfs_release_block_group(used_block_group,
7661 if (loop >= LOOP_NO_EMPTY_SIZE) {
7662 spin_unlock(&last_ptr->refill_lock);
7663 goto unclustered_alloc;
7666 aligned_cluster = max_t(unsigned long,
7667 empty_cluster + empty_size,
7668 block_group->full_stripe_len);
7670 /* allocate a cluster in this block group */
7671 ret = btrfs_find_space_cluster(fs_info, block_group,
7672 last_ptr, search_start,
7677 * now pull our allocation out of this
7680 offset = btrfs_alloc_from_cluster(block_group,
7686 /* we found one, proceed */
7687 spin_unlock(&last_ptr->refill_lock);
7688 trace_btrfs_reserve_extent_cluster(fs_info,
7689 block_group, search_start,
7693 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7694 && !failed_cluster_refill) {
7695 spin_unlock(&last_ptr->refill_lock);
7697 failed_cluster_refill = true;
7698 wait_block_group_cache_progress(block_group,
7699 num_bytes + empty_cluster + empty_size);
7700 goto have_block_group;
7704 * at this point we either didn't find a cluster
7705 * or we weren't able to allocate a block from our
7706 * cluster. Free the cluster we've been trying
7707 * to use, and go to the next block group
7709 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7710 spin_unlock(&last_ptr->refill_lock);
7716 * We are doing an unclustered alloc, set the fragmented flag so
7717 * we don't bother trying to setup a cluster again until we get
7720 if (unlikely(last_ptr)) {
7721 spin_lock(&last_ptr->lock);
7722 last_ptr->fragmented = 1;
7723 spin_unlock(&last_ptr->lock);
7726 struct btrfs_free_space_ctl *ctl =
7727 block_group->free_space_ctl;
7729 spin_lock(&ctl->tree_lock);
7730 if (ctl->free_space <
7731 num_bytes + empty_cluster + empty_size) {
7732 if (ctl->free_space > max_extent_size)
7733 max_extent_size = ctl->free_space;
7734 spin_unlock(&ctl->tree_lock);
7737 spin_unlock(&ctl->tree_lock);
7740 offset = btrfs_find_space_for_alloc(block_group, search_start,
7741 num_bytes, empty_size,
7744 * If we didn't find a chunk, and we haven't failed on this
7745 * block group before, and this block group is in the middle of
7746 * caching and we are ok with waiting, then go ahead and wait
7747 * for progress to be made, and set failed_alloc to true.
7749 * If failed_alloc is true then we've already waited on this
7750 * block group once and should move on to the next block group.
7752 if (!offset && !failed_alloc && !cached &&
7753 loop > LOOP_CACHING_NOWAIT) {
7754 wait_block_group_cache_progress(block_group,
7755 num_bytes + empty_size);
7756 failed_alloc = true;
7757 goto have_block_group;
7758 } else if (!offset) {
7762 search_start = ALIGN(offset, fs_info->stripesize);
7764 /* move on to the next group */
7765 if (search_start + num_bytes >
7766 block_group->key.objectid + block_group->key.offset) {
7767 btrfs_add_free_space(block_group, offset, num_bytes);
7771 if (offset < search_start)
7772 btrfs_add_free_space(block_group, offset,
7773 search_start - offset);
7774 BUG_ON(offset > search_start);
7776 ret = btrfs_add_reserved_bytes(block_group, ram_bytes,
7777 num_bytes, delalloc);
7778 if (ret == -EAGAIN) {
7779 btrfs_add_free_space(block_group, offset, num_bytes);
7782 btrfs_inc_block_group_reservations(block_group);
7784 /* we are all good, lets return */
7785 ins->objectid = search_start;
7786 ins->offset = num_bytes;
7788 trace_btrfs_reserve_extent(fs_info, block_group,
7789 search_start, num_bytes);
7790 btrfs_release_block_group(block_group, delalloc);
7793 failed_cluster_refill = false;
7794 failed_alloc = false;
7795 BUG_ON(index != get_block_group_index(block_group));
7796 btrfs_release_block_group(block_group, delalloc);
7798 up_read(&space_info->groups_sem);
7800 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7801 && !orig_have_caching_bg)
7802 orig_have_caching_bg = true;
7804 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7807 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7811 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7812 * caching kthreads as we move along
7813 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7814 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7815 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7818 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7820 if (loop == LOOP_CACHING_NOWAIT) {
7822 * We want to skip the LOOP_CACHING_WAIT step if we
7823 * don't have any uncached bgs and we've already done a
7824 * full search through.
7826 if (orig_have_caching_bg || !full_search)
7827 loop = LOOP_CACHING_WAIT;
7829 loop = LOOP_ALLOC_CHUNK;
7834 if (loop == LOOP_ALLOC_CHUNK) {
7835 struct btrfs_trans_handle *trans;
7838 trans = current->journal_info;
7842 trans = btrfs_join_transaction(root);
7844 if (IS_ERR(trans)) {
7845 ret = PTR_ERR(trans);
7849 ret = do_chunk_alloc(trans, fs_info, flags,
7853 * If we can't allocate a new chunk we've already looped
7854 * through at least once, move on to the NO_EMPTY_SIZE
7858 loop = LOOP_NO_EMPTY_SIZE;
7861 * Do not bail out on ENOSPC since we
7862 * can do more things.
7864 if (ret < 0 && ret != -ENOSPC)
7865 btrfs_abort_transaction(trans, ret);
7869 btrfs_end_transaction(trans);
7874 if (loop == LOOP_NO_EMPTY_SIZE) {
7876 * Don't loop again if we already have no empty_size and
7879 if (empty_size == 0 &&
7880 empty_cluster == 0) {
7889 } else if (!ins->objectid) {
7891 } else if (ins->objectid) {
7892 if (!use_cluster && last_ptr) {
7893 spin_lock(&last_ptr->lock);
7894 last_ptr->window_start = ins->objectid;
7895 spin_unlock(&last_ptr->lock);
7900 if (ret == -ENOSPC) {
7901 spin_lock(&space_info->lock);
7902 space_info->max_extent_size = max_extent_size;
7903 spin_unlock(&space_info->lock);
7904 ins->offset = max_extent_size;
7909 static void dump_space_info(struct btrfs_fs_info *fs_info,
7910 struct btrfs_space_info *info, u64 bytes,
7911 int dump_block_groups)
7913 struct btrfs_block_group_cache *cache;
7916 spin_lock(&info->lock);
7917 btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull",
7919 info->total_bytes - btrfs_space_info_used(info, true),
7920 info->full ? "" : "not ");
7922 "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu",
7923 info->total_bytes, info->bytes_used, info->bytes_pinned,
7924 info->bytes_reserved, info->bytes_may_use,
7925 info->bytes_readonly);
7926 spin_unlock(&info->lock);
7928 if (!dump_block_groups)
7931 down_read(&info->groups_sem);
7933 list_for_each_entry(cache, &info->block_groups[index], list) {
7934 spin_lock(&cache->lock);
7936 "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s",
7937 cache->key.objectid, cache->key.offset,
7938 btrfs_block_group_used(&cache->item), cache->pinned,
7939 cache->reserved, cache->ro ? "[readonly]" : "");
7940 btrfs_dump_free_space(cache, bytes);
7941 spin_unlock(&cache->lock);
7943 if (++index < BTRFS_NR_RAID_TYPES)
7945 up_read(&info->groups_sem);
7948 int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes,
7949 u64 num_bytes, u64 min_alloc_size,
7950 u64 empty_size, u64 hint_byte,
7951 struct btrfs_key *ins, int is_data, int delalloc)
7953 struct btrfs_fs_info *fs_info = root->fs_info;
7954 bool final_tried = num_bytes == min_alloc_size;
7958 flags = btrfs_get_alloc_profile(root, is_data);
7960 WARN_ON(num_bytes < fs_info->sectorsize);
7961 ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size,
7962 hint_byte, ins, flags, delalloc);
7963 if (!ret && !is_data) {
7964 btrfs_dec_block_group_reservations(fs_info, ins->objectid);
7965 } else if (ret == -ENOSPC) {
7966 if (!final_tried && ins->offset) {
7967 num_bytes = min(num_bytes >> 1, ins->offset);
7968 num_bytes = round_down(num_bytes,
7969 fs_info->sectorsize);
7970 num_bytes = max(num_bytes, min_alloc_size);
7971 ram_bytes = num_bytes;
7972 if (num_bytes == min_alloc_size)
7975 } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
7976 struct btrfs_space_info *sinfo;
7978 sinfo = __find_space_info(fs_info, flags);
7980 "allocation failed flags %llu, wanted %llu",
7983 dump_space_info(fs_info, sinfo, num_bytes, 1);
7990 static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
7992 int pin, int delalloc)
7994 struct btrfs_block_group_cache *cache;
7997 cache = btrfs_lookup_block_group(fs_info, start);
7999 btrfs_err(fs_info, "Unable to find block group for %llu",
8005 pin_down_extent(fs_info, cache, start, len, 1);
8007 if (btrfs_test_opt(fs_info, DISCARD))
8008 ret = btrfs_discard_extent(fs_info, start, len, NULL);
8009 btrfs_add_free_space(cache, start, len);
8010 btrfs_free_reserved_bytes(cache, len, delalloc);
8011 trace_btrfs_reserved_extent_free(fs_info, start, len);
8014 btrfs_put_block_group(cache);
8018 int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info,
8019 u64 start, u64 len, int delalloc)
8021 return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc);
8024 int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info,
8027 return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0);
8030 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8031 struct btrfs_fs_info *fs_info,
8032 u64 parent, u64 root_objectid,
8033 u64 flags, u64 owner, u64 offset,
8034 struct btrfs_key *ins, int ref_mod)
8037 struct btrfs_extent_item *extent_item;
8038 struct btrfs_extent_inline_ref *iref;
8039 struct btrfs_path *path;
8040 struct extent_buffer *leaf;
8045 type = BTRFS_SHARED_DATA_REF_KEY;
8047 type = BTRFS_EXTENT_DATA_REF_KEY;
8049 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8051 path = btrfs_alloc_path();
8055 path->leave_spinning = 1;
8056 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8059 btrfs_free_path(path);
8063 leaf = path->nodes[0];
8064 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8065 struct btrfs_extent_item);
8066 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8067 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8068 btrfs_set_extent_flags(leaf, extent_item,
8069 flags | BTRFS_EXTENT_FLAG_DATA);
8071 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8072 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8074 struct btrfs_shared_data_ref *ref;
8075 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8076 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8077 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8079 struct btrfs_extent_data_ref *ref;
8080 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8081 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8082 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8083 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8084 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8087 btrfs_mark_buffer_dirty(path->nodes[0]);
8088 btrfs_free_path(path);
8090 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8095 ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1);
8096 if (ret) { /* -ENOENT, logic error */
8097 btrfs_err(fs_info, "update block group failed for %llu %llu",
8098 ins->objectid, ins->offset);
8101 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset);
8105 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8106 struct btrfs_fs_info *fs_info,
8107 u64 parent, u64 root_objectid,
8108 u64 flags, struct btrfs_disk_key *key,
8109 int level, struct btrfs_key *ins)
8112 struct btrfs_extent_item *extent_item;
8113 struct btrfs_tree_block_info *block_info;
8114 struct btrfs_extent_inline_ref *iref;
8115 struct btrfs_path *path;
8116 struct extent_buffer *leaf;
8117 u32 size = sizeof(*extent_item) + sizeof(*iref);
8118 u64 num_bytes = ins->offset;
8119 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8121 if (!skinny_metadata)
8122 size += sizeof(*block_info);
8124 path = btrfs_alloc_path();
8126 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8131 path->leave_spinning = 1;
8132 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8135 btrfs_free_path(path);
8136 btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid,
8141 leaf = path->nodes[0];
8142 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8143 struct btrfs_extent_item);
8144 btrfs_set_extent_refs(leaf, extent_item, 1);
8145 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8146 btrfs_set_extent_flags(leaf, extent_item,
8147 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8149 if (skinny_metadata) {
8150 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8151 num_bytes = fs_info->nodesize;
8153 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8154 btrfs_set_tree_block_key(leaf, block_info, key);
8155 btrfs_set_tree_block_level(leaf, block_info, level);
8156 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8160 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8161 btrfs_set_extent_inline_ref_type(leaf, iref,
8162 BTRFS_SHARED_BLOCK_REF_KEY);
8163 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8165 btrfs_set_extent_inline_ref_type(leaf, iref,
8166 BTRFS_TREE_BLOCK_REF_KEY);
8167 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8170 btrfs_mark_buffer_dirty(leaf);
8171 btrfs_free_path(path);
8173 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8178 ret = update_block_group(trans, fs_info, ins->objectid,
8179 fs_info->nodesize, 1);
8180 if (ret) { /* -ENOENT, logic error */
8181 btrfs_err(fs_info, "update block group failed for %llu %llu",
8182 ins->objectid, ins->offset);
8186 trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid,
8191 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8192 u64 root_objectid, u64 owner,
8193 u64 offset, u64 ram_bytes,
8194 struct btrfs_key *ins)
8196 struct btrfs_fs_info *fs_info = trans->fs_info;
8199 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8201 ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid,
8203 root_objectid, owner, offset,
8204 ram_bytes, BTRFS_ADD_DELAYED_EXTENT);
8209 * this is used by the tree logging recovery code. It records that
8210 * an extent has been allocated and makes sure to clear the free
8211 * space cache bits as well
8213 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8214 struct btrfs_fs_info *fs_info,
8215 u64 root_objectid, u64 owner, u64 offset,
8216 struct btrfs_key *ins)
8219 struct btrfs_block_group_cache *block_group;
8220 struct btrfs_space_info *space_info;
8223 * Mixed block groups will exclude before processing the log so we only
8224 * need to do the exclude dance if this fs isn't mixed.
8226 if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
8227 ret = __exclude_logged_extent(fs_info, ins->objectid,
8233 block_group = btrfs_lookup_block_group(fs_info, ins->objectid);
8237 space_info = block_group->space_info;
8238 spin_lock(&space_info->lock);
8239 spin_lock(&block_group->lock);
8240 space_info->bytes_reserved += ins->offset;
8241 block_group->reserved += ins->offset;
8242 spin_unlock(&block_group->lock);
8243 spin_unlock(&space_info->lock);
8245 ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid,
8246 0, owner, offset, ins, 1);
8247 btrfs_put_block_group(block_group);
8251 static struct extent_buffer *
8252 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8253 u64 bytenr, int level)
8255 struct btrfs_fs_info *fs_info = root->fs_info;
8256 struct extent_buffer *buf;
8258 buf = btrfs_find_create_tree_block(fs_info, bytenr);
8262 btrfs_set_header_generation(buf, trans->transid);
8263 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8264 btrfs_tree_lock(buf);
8265 clean_tree_block(fs_info, buf);
8266 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8268 btrfs_set_lock_blocking(buf);
8269 set_extent_buffer_uptodate(buf);
8271 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8272 buf->log_index = root->log_transid % 2;
8274 * we allow two log transactions at a time, use different
8275 * EXENT bit to differentiate dirty pages.
8277 if (buf->log_index == 0)
8278 set_extent_dirty(&root->dirty_log_pages, buf->start,
8279 buf->start + buf->len - 1, GFP_NOFS);
8281 set_extent_new(&root->dirty_log_pages, buf->start,
8282 buf->start + buf->len - 1);
8284 buf->log_index = -1;
8285 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8286 buf->start + buf->len - 1, GFP_NOFS);
8288 trans->dirty = true;
8289 /* this returns a buffer locked for blocking */
8293 static struct btrfs_block_rsv *
8294 use_block_rsv(struct btrfs_trans_handle *trans,
8295 struct btrfs_root *root, u32 blocksize)
8297 struct btrfs_fs_info *fs_info = root->fs_info;
8298 struct btrfs_block_rsv *block_rsv;
8299 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
8301 bool global_updated = false;
8303 block_rsv = get_block_rsv(trans, root);
8305 if (unlikely(block_rsv->size == 0))
8308 ret = block_rsv_use_bytes(block_rsv, blocksize);
8312 if (block_rsv->failfast)
8313 return ERR_PTR(ret);
8315 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8316 global_updated = true;
8317 update_global_block_rsv(fs_info);
8321 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
8322 static DEFINE_RATELIMIT_STATE(_rs,
8323 DEFAULT_RATELIMIT_INTERVAL * 10,
8324 /*DEFAULT_RATELIMIT_BURST*/ 1);
8325 if (__ratelimit(&_rs))
8327 "BTRFS: block rsv returned %d\n", ret);
8330 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8331 BTRFS_RESERVE_NO_FLUSH);
8335 * If we couldn't reserve metadata bytes try and use some from
8336 * the global reserve if its space type is the same as the global
8339 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8340 block_rsv->space_info == global_rsv->space_info) {
8341 ret = block_rsv_use_bytes(global_rsv, blocksize);
8345 return ERR_PTR(ret);
8348 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8349 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8351 block_rsv_add_bytes(block_rsv, blocksize, 0);
8352 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8356 * finds a free extent and does all the dirty work required for allocation
8357 * returns the tree buffer or an ERR_PTR on error.
8359 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8360 struct btrfs_root *root,
8361 u64 parent, u64 root_objectid,
8362 const struct btrfs_disk_key *key,
8363 int level, u64 hint,
8366 struct btrfs_fs_info *fs_info = root->fs_info;
8367 struct btrfs_key ins;
8368 struct btrfs_block_rsv *block_rsv;
8369 struct extent_buffer *buf;
8370 struct btrfs_delayed_extent_op *extent_op;
8373 u32 blocksize = fs_info->nodesize;
8374 bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
8376 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8377 if (btrfs_is_testing(fs_info)) {
8378 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8381 root->alloc_bytenr += blocksize;
8386 block_rsv = use_block_rsv(trans, root, blocksize);
8387 if (IS_ERR(block_rsv))
8388 return ERR_CAST(block_rsv);
8390 ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize,
8391 empty_size, hint, &ins, 0, 0);
8395 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8398 goto out_free_reserved;
8401 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8403 parent = ins.objectid;
8404 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8408 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8409 extent_op = btrfs_alloc_delayed_extent_op();
8415 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8417 memset(&extent_op->key, 0, sizeof(extent_op->key));
8418 extent_op->flags_to_set = flags;
8419 extent_op->update_key = skinny_metadata ? false : true;
8420 extent_op->update_flags = true;
8421 extent_op->is_data = false;
8422 extent_op->level = level;
8424 ret = btrfs_add_delayed_tree_ref(fs_info, trans,
8425 ins.objectid, ins.offset,
8426 parent, root_objectid, level,
8427 BTRFS_ADD_DELAYED_EXTENT,
8430 goto out_free_delayed;
8435 btrfs_free_delayed_extent_op(extent_op);
8437 free_extent_buffer(buf);
8439 btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0);
8441 unuse_block_rsv(fs_info, block_rsv, blocksize);
8442 return ERR_PTR(ret);
8445 struct walk_control {
8446 u64 refs[BTRFS_MAX_LEVEL];
8447 u64 flags[BTRFS_MAX_LEVEL];
8448 struct btrfs_key update_progress;
8459 #define DROP_REFERENCE 1
8460 #define UPDATE_BACKREF 2
8462 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8463 struct btrfs_root *root,
8464 struct walk_control *wc,
8465 struct btrfs_path *path)
8467 struct btrfs_fs_info *fs_info = root->fs_info;
8473 struct btrfs_key key;
8474 struct extent_buffer *eb;
8479 if (path->slots[wc->level] < wc->reada_slot) {
8480 wc->reada_count = wc->reada_count * 2 / 3;
8481 wc->reada_count = max(wc->reada_count, 2);
8483 wc->reada_count = wc->reada_count * 3 / 2;
8484 wc->reada_count = min_t(int, wc->reada_count,
8485 BTRFS_NODEPTRS_PER_BLOCK(fs_info));
8488 eb = path->nodes[wc->level];
8489 nritems = btrfs_header_nritems(eb);
8491 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8492 if (nread >= wc->reada_count)
8496 bytenr = btrfs_node_blockptr(eb, slot);
8497 generation = btrfs_node_ptr_generation(eb, slot);
8499 if (slot == path->slots[wc->level])
8502 if (wc->stage == UPDATE_BACKREF &&
8503 generation <= root->root_key.offset)
8506 /* We don't lock the tree block, it's OK to be racy here */
8507 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr,
8508 wc->level - 1, 1, &refs,
8510 /* We don't care about errors in readahead. */
8515 if (wc->stage == DROP_REFERENCE) {
8519 if (wc->level == 1 &&
8520 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8522 if (!wc->update_ref ||
8523 generation <= root->root_key.offset)
8525 btrfs_node_key_to_cpu(eb, &key, slot);
8526 ret = btrfs_comp_cpu_keys(&key,
8527 &wc->update_progress);
8531 if (wc->level == 1 &&
8532 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8536 readahead_tree_block(fs_info, bytenr);
8539 wc->reada_slot = slot;
8543 * helper to process tree block while walking down the tree.
8545 * when wc->stage == UPDATE_BACKREF, this function updates
8546 * back refs for pointers in the block.
8548 * NOTE: return value 1 means we should stop walking down.
8550 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8551 struct btrfs_root *root,
8552 struct btrfs_path *path,
8553 struct walk_control *wc, int lookup_info)
8555 struct btrfs_fs_info *fs_info = root->fs_info;
8556 int level = wc->level;
8557 struct extent_buffer *eb = path->nodes[level];
8558 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8561 if (wc->stage == UPDATE_BACKREF &&
8562 btrfs_header_owner(eb) != root->root_key.objectid)
8566 * when reference count of tree block is 1, it won't increase
8567 * again. once full backref flag is set, we never clear it.
8570 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8571 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8572 BUG_ON(!path->locks[level]);
8573 ret = btrfs_lookup_extent_info(trans, fs_info,
8574 eb->start, level, 1,
8577 BUG_ON(ret == -ENOMEM);
8580 BUG_ON(wc->refs[level] == 0);
8583 if (wc->stage == DROP_REFERENCE) {
8584 if (wc->refs[level] > 1)
8587 if (path->locks[level] && !wc->keep_locks) {
8588 btrfs_tree_unlock_rw(eb, path->locks[level]);
8589 path->locks[level] = 0;
8594 /* wc->stage == UPDATE_BACKREF */
8595 if (!(wc->flags[level] & flag)) {
8596 BUG_ON(!path->locks[level]);
8597 ret = btrfs_inc_ref(trans, root, eb, 1);
8598 BUG_ON(ret); /* -ENOMEM */
8599 ret = btrfs_dec_ref(trans, root, eb, 0);
8600 BUG_ON(ret); /* -ENOMEM */
8601 ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start,
8603 btrfs_header_level(eb), 0);
8604 BUG_ON(ret); /* -ENOMEM */
8605 wc->flags[level] |= flag;
8609 * the block is shared by multiple trees, so it's not good to
8610 * keep the tree lock
8612 if (path->locks[level] && level > 0) {
8613 btrfs_tree_unlock_rw(eb, path->locks[level]);
8614 path->locks[level] = 0;
8620 * helper to process tree block pointer.
8622 * when wc->stage == DROP_REFERENCE, this function checks
8623 * reference count of the block pointed to. if the block
8624 * is shared and we need update back refs for the subtree
8625 * rooted at the block, this function changes wc->stage to
8626 * UPDATE_BACKREF. if the block is shared and there is no
8627 * need to update back, this function drops the reference
8630 * NOTE: return value 1 means we should stop walking down.
8632 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8633 struct btrfs_root *root,
8634 struct btrfs_path *path,
8635 struct walk_control *wc, int *lookup_info)
8637 struct btrfs_fs_info *fs_info = root->fs_info;
8642 struct btrfs_key key;
8643 struct extent_buffer *next;
8644 int level = wc->level;
8647 bool need_account = false;
8649 generation = btrfs_node_ptr_generation(path->nodes[level],
8650 path->slots[level]);
8652 * if the lower level block was created before the snapshot
8653 * was created, we know there is no need to update back refs
8656 if (wc->stage == UPDATE_BACKREF &&
8657 generation <= root->root_key.offset) {
8662 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8663 blocksize = fs_info->nodesize;
8665 next = find_extent_buffer(fs_info, bytenr);
8667 next = btrfs_find_create_tree_block(fs_info, bytenr);
8669 return PTR_ERR(next);
8671 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8675 btrfs_tree_lock(next);
8676 btrfs_set_lock_blocking(next);
8678 ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1,
8679 &wc->refs[level - 1],
8680 &wc->flags[level - 1]);
8684 if (unlikely(wc->refs[level - 1] == 0)) {
8685 btrfs_err(fs_info, "Missing references.");
8691 if (wc->stage == DROP_REFERENCE) {
8692 if (wc->refs[level - 1] > 1) {
8693 need_account = true;
8695 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8698 if (!wc->update_ref ||
8699 generation <= root->root_key.offset)
8702 btrfs_node_key_to_cpu(path->nodes[level], &key,
8703 path->slots[level]);
8704 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8708 wc->stage = UPDATE_BACKREF;
8709 wc->shared_level = level - 1;
8713 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8717 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8718 btrfs_tree_unlock(next);
8719 free_extent_buffer(next);
8725 if (reada && level == 1)
8726 reada_walk_down(trans, root, wc, path);
8727 next = read_tree_block(fs_info, bytenr, generation);
8729 return PTR_ERR(next);
8730 } else if (!extent_buffer_uptodate(next)) {
8731 free_extent_buffer(next);
8734 btrfs_tree_lock(next);
8735 btrfs_set_lock_blocking(next);
8739 ASSERT(level == btrfs_header_level(next));
8740 if (level != btrfs_header_level(next)) {
8741 btrfs_err(root->fs_info, "mismatched level");
8745 path->nodes[level] = next;
8746 path->slots[level] = 0;
8747 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8753 wc->refs[level - 1] = 0;
8754 wc->flags[level - 1] = 0;
8755 if (wc->stage == DROP_REFERENCE) {
8756 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8757 parent = path->nodes[level]->start;
8759 ASSERT(root->root_key.objectid ==
8760 btrfs_header_owner(path->nodes[level]));
8761 if (root->root_key.objectid !=
8762 btrfs_header_owner(path->nodes[level])) {
8763 btrfs_err(root->fs_info,
8764 "mismatched block owner");
8772 ret = btrfs_qgroup_trace_subtree(trans, root, next,
8773 generation, level - 1);
8775 btrfs_err_rl(fs_info,
8776 "Error %d accounting shared subtree. Quota is out of sync, rescan required.",
8780 ret = btrfs_free_extent(trans, fs_info, bytenr, blocksize,
8781 parent, root->root_key.objectid,
8791 btrfs_tree_unlock(next);
8792 free_extent_buffer(next);
8798 * helper to process tree block while walking up the tree.
8800 * when wc->stage == DROP_REFERENCE, this function drops
8801 * reference count on the block.
8803 * when wc->stage == UPDATE_BACKREF, this function changes
8804 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8805 * to UPDATE_BACKREF previously while processing the block.
8807 * NOTE: return value 1 means we should stop walking up.
8809 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8810 struct btrfs_root *root,
8811 struct btrfs_path *path,
8812 struct walk_control *wc)
8814 struct btrfs_fs_info *fs_info = root->fs_info;
8816 int level = wc->level;
8817 struct extent_buffer *eb = path->nodes[level];
8820 if (wc->stage == UPDATE_BACKREF) {
8821 BUG_ON(wc->shared_level < level);
8822 if (level < wc->shared_level)
8825 ret = find_next_key(path, level + 1, &wc->update_progress);
8829 wc->stage = DROP_REFERENCE;
8830 wc->shared_level = -1;
8831 path->slots[level] = 0;
8834 * check reference count again if the block isn't locked.
8835 * we should start walking down the tree again if reference
8838 if (!path->locks[level]) {
8840 btrfs_tree_lock(eb);
8841 btrfs_set_lock_blocking(eb);
8842 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8844 ret = btrfs_lookup_extent_info(trans, fs_info,
8845 eb->start, level, 1,
8849 btrfs_tree_unlock_rw(eb, path->locks[level]);
8850 path->locks[level] = 0;
8853 BUG_ON(wc->refs[level] == 0);
8854 if (wc->refs[level] == 1) {
8855 btrfs_tree_unlock_rw(eb, path->locks[level]);
8856 path->locks[level] = 0;
8862 /* wc->stage == DROP_REFERENCE */
8863 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8865 if (wc->refs[level] == 1) {
8867 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8868 ret = btrfs_dec_ref(trans, root, eb, 1);
8870 ret = btrfs_dec_ref(trans, root, eb, 0);
8871 BUG_ON(ret); /* -ENOMEM */
8872 ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb);
8874 btrfs_err_rl(fs_info,
8875 "error %d accounting leaf items. Quota is out of sync, rescan required.",
8879 /* make block locked assertion in clean_tree_block happy */
8880 if (!path->locks[level] &&
8881 btrfs_header_generation(eb) == trans->transid) {
8882 btrfs_tree_lock(eb);
8883 btrfs_set_lock_blocking(eb);
8884 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8886 clean_tree_block(fs_info, eb);
8889 if (eb == root->node) {
8890 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8893 BUG_ON(root->root_key.objectid !=
8894 btrfs_header_owner(eb));
8896 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8897 parent = path->nodes[level + 1]->start;
8899 BUG_ON(root->root_key.objectid !=
8900 btrfs_header_owner(path->nodes[level + 1]));
8903 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8905 wc->refs[level] = 0;
8906 wc->flags[level] = 0;
8910 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8911 struct btrfs_root *root,
8912 struct btrfs_path *path,
8913 struct walk_control *wc)
8915 int level = wc->level;
8916 int lookup_info = 1;
8919 while (level >= 0) {
8920 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8927 if (path->slots[level] >=
8928 btrfs_header_nritems(path->nodes[level]))
8931 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8933 path->slots[level]++;
8942 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8943 struct btrfs_root *root,
8944 struct btrfs_path *path,
8945 struct walk_control *wc, int max_level)
8947 int level = wc->level;
8950 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8951 while (level < max_level && path->nodes[level]) {
8953 if (path->slots[level] + 1 <
8954 btrfs_header_nritems(path->nodes[level])) {
8955 path->slots[level]++;
8958 ret = walk_up_proc(trans, root, path, wc);
8962 if (path->locks[level]) {
8963 btrfs_tree_unlock_rw(path->nodes[level],
8964 path->locks[level]);
8965 path->locks[level] = 0;
8967 free_extent_buffer(path->nodes[level]);
8968 path->nodes[level] = NULL;
8976 * drop a subvolume tree.
8978 * this function traverses the tree freeing any blocks that only
8979 * referenced by the tree.
8981 * when a shared tree block is found. this function decreases its
8982 * reference count by one. if update_ref is true, this function
8983 * also make sure backrefs for the shared block and all lower level
8984 * blocks are properly updated.
8986 * If called with for_reloc == 0, may exit early with -EAGAIN
8988 int btrfs_drop_snapshot(struct btrfs_root *root,
8989 struct btrfs_block_rsv *block_rsv, int update_ref,
8992 struct btrfs_fs_info *fs_info = root->fs_info;
8993 struct btrfs_path *path;
8994 struct btrfs_trans_handle *trans;
8995 struct btrfs_root *tree_root = fs_info->tree_root;
8996 struct btrfs_root_item *root_item = &root->root_item;
8997 struct walk_control *wc;
8998 struct btrfs_key key;
9002 bool root_dropped = false;
9004 btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid);
9006 path = btrfs_alloc_path();
9012 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9014 btrfs_free_path(path);
9019 trans = btrfs_start_transaction(tree_root, 0);
9020 if (IS_ERR(trans)) {
9021 err = PTR_ERR(trans);
9026 trans->block_rsv = block_rsv;
9028 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9029 level = btrfs_header_level(root->node);
9030 path->nodes[level] = btrfs_lock_root_node(root);
9031 btrfs_set_lock_blocking(path->nodes[level]);
9032 path->slots[level] = 0;
9033 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9034 memset(&wc->update_progress, 0,
9035 sizeof(wc->update_progress));
9037 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9038 memcpy(&wc->update_progress, &key,
9039 sizeof(wc->update_progress));
9041 level = root_item->drop_level;
9043 path->lowest_level = level;
9044 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9045 path->lowest_level = 0;
9053 * unlock our path, this is safe because only this
9054 * function is allowed to delete this snapshot
9056 btrfs_unlock_up_safe(path, 0);
9058 level = btrfs_header_level(root->node);
9060 btrfs_tree_lock(path->nodes[level]);
9061 btrfs_set_lock_blocking(path->nodes[level]);
9062 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9064 ret = btrfs_lookup_extent_info(trans, fs_info,
9065 path->nodes[level]->start,
9066 level, 1, &wc->refs[level],
9072 BUG_ON(wc->refs[level] == 0);
9074 if (level == root_item->drop_level)
9077 btrfs_tree_unlock(path->nodes[level]);
9078 path->locks[level] = 0;
9079 WARN_ON(wc->refs[level] != 1);
9085 wc->shared_level = -1;
9086 wc->stage = DROP_REFERENCE;
9087 wc->update_ref = update_ref;
9089 wc->for_reloc = for_reloc;
9090 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9094 ret = walk_down_tree(trans, root, path, wc);
9100 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9107 BUG_ON(wc->stage != DROP_REFERENCE);
9111 if (wc->stage == DROP_REFERENCE) {
9113 btrfs_node_key(path->nodes[level],
9114 &root_item->drop_progress,
9115 path->slots[level]);
9116 root_item->drop_level = level;
9119 BUG_ON(wc->level == 0);
9120 if (btrfs_should_end_transaction(trans) ||
9121 (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) {
9122 ret = btrfs_update_root(trans, tree_root,
9126 btrfs_abort_transaction(trans, ret);
9131 btrfs_end_transaction_throttle(trans);
9132 if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) {
9133 btrfs_debug(fs_info,
9134 "drop snapshot early exit");
9139 trans = btrfs_start_transaction(tree_root, 0);
9140 if (IS_ERR(trans)) {
9141 err = PTR_ERR(trans);
9145 trans->block_rsv = block_rsv;
9148 btrfs_release_path(path);
9152 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9154 btrfs_abort_transaction(trans, ret);
9158 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9159 ret = btrfs_find_root(tree_root, &root->root_key, path,
9162 btrfs_abort_transaction(trans, ret);
9165 } else if (ret > 0) {
9166 /* if we fail to delete the orphan item this time
9167 * around, it'll get picked up the next time.
9169 * The most common failure here is just -ENOENT.
9171 btrfs_del_orphan_item(trans, tree_root,
9172 root->root_key.objectid);
9176 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9177 btrfs_add_dropped_root(trans, root);
9179 free_extent_buffer(root->node);
9180 free_extent_buffer(root->commit_root);
9181 btrfs_put_fs_root(root);
9183 root_dropped = true;
9185 btrfs_end_transaction_throttle(trans);
9188 btrfs_free_path(path);
9191 * So if we need to stop dropping the snapshot for whatever reason we
9192 * need to make sure to add it back to the dead root list so that we
9193 * keep trying to do the work later. This also cleans up roots if we
9194 * don't have it in the radix (like when we recover after a power fail
9195 * or unmount) so we don't leak memory.
9197 if (!for_reloc && root_dropped == false)
9198 btrfs_add_dead_root(root);
9199 if (err && err != -EAGAIN)
9200 btrfs_handle_fs_error(fs_info, err, NULL);
9205 * drop subtree rooted at tree block 'node'.
9207 * NOTE: this function will unlock and release tree block 'node'
9208 * only used by relocation code
9210 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9211 struct btrfs_root *root,
9212 struct extent_buffer *node,
9213 struct extent_buffer *parent)
9215 struct btrfs_fs_info *fs_info = root->fs_info;
9216 struct btrfs_path *path;
9217 struct walk_control *wc;
9223 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9225 path = btrfs_alloc_path();
9229 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9231 btrfs_free_path(path);
9235 btrfs_assert_tree_locked(parent);
9236 parent_level = btrfs_header_level(parent);
9237 extent_buffer_get(parent);
9238 path->nodes[parent_level] = parent;
9239 path->slots[parent_level] = btrfs_header_nritems(parent);
9241 btrfs_assert_tree_locked(node);
9242 level = btrfs_header_level(node);
9243 path->nodes[level] = node;
9244 path->slots[level] = 0;
9245 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9247 wc->refs[parent_level] = 1;
9248 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9250 wc->shared_level = -1;
9251 wc->stage = DROP_REFERENCE;
9255 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info);
9258 wret = walk_down_tree(trans, root, path, wc);
9264 wret = walk_up_tree(trans, root, path, wc, parent_level);
9272 btrfs_free_path(path);
9276 static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags)
9282 * if restripe for this chunk_type is on pick target profile and
9283 * return, otherwise do the usual balance
9285 stripped = get_restripe_target(fs_info, flags);
9287 return extended_to_chunk(stripped);
9289 num_devices = fs_info->fs_devices->rw_devices;
9291 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9292 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9293 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9295 if (num_devices == 1) {
9296 stripped |= BTRFS_BLOCK_GROUP_DUP;
9297 stripped = flags & ~stripped;
9299 /* turn raid0 into single device chunks */
9300 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9303 /* turn mirroring into duplication */
9304 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9305 BTRFS_BLOCK_GROUP_RAID10))
9306 return stripped | BTRFS_BLOCK_GROUP_DUP;
9308 /* they already had raid on here, just return */
9309 if (flags & stripped)
9312 stripped |= BTRFS_BLOCK_GROUP_DUP;
9313 stripped = flags & ~stripped;
9315 /* switch duplicated blocks with raid1 */
9316 if (flags & BTRFS_BLOCK_GROUP_DUP)
9317 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9319 /* this is drive concat, leave it alone */
9325 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9327 struct btrfs_space_info *sinfo = cache->space_info;
9329 u64 min_allocable_bytes;
9333 * We need some metadata space and system metadata space for
9334 * allocating chunks in some corner cases until we force to set
9335 * it to be readonly.
9338 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9340 min_allocable_bytes = SZ_1M;
9342 min_allocable_bytes = 0;
9344 spin_lock(&sinfo->lock);
9345 spin_lock(&cache->lock);
9353 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9354 cache->bytes_super - btrfs_block_group_used(&cache->item);
9356 if (btrfs_space_info_used(sinfo, true) + num_bytes +
9357 min_allocable_bytes <= sinfo->total_bytes) {
9358 sinfo->bytes_readonly += num_bytes;
9360 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9364 spin_unlock(&cache->lock);
9365 spin_unlock(&sinfo->lock);
9369 int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info,
9370 struct btrfs_block_group_cache *cache)
9373 struct btrfs_trans_handle *trans;
9378 trans = btrfs_join_transaction(fs_info->extent_root);
9380 return PTR_ERR(trans);
9383 * we're not allowed to set block groups readonly after the dirty
9384 * block groups cache has started writing. If it already started,
9385 * back off and let this transaction commit
9387 mutex_lock(&fs_info->ro_block_group_mutex);
9388 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9389 u64 transid = trans->transid;
9391 mutex_unlock(&fs_info->ro_block_group_mutex);
9392 btrfs_end_transaction(trans);
9394 ret = btrfs_wait_for_commit(fs_info, transid);
9401 * if we are changing raid levels, try to allocate a corresponding
9402 * block group with the new raid level.
9404 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9405 if (alloc_flags != cache->flags) {
9406 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9409 * ENOSPC is allowed here, we may have enough space
9410 * already allocated at the new raid level to
9419 ret = inc_block_group_ro(cache, 0);
9422 alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags);
9423 ret = do_chunk_alloc(trans, fs_info, alloc_flags,
9427 ret = inc_block_group_ro(cache, 0);
9429 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9430 alloc_flags = update_block_group_flags(fs_info, cache->flags);
9431 mutex_lock(&fs_info->chunk_mutex);
9432 check_system_chunk(trans, fs_info, alloc_flags);
9433 mutex_unlock(&fs_info->chunk_mutex);
9435 mutex_unlock(&fs_info->ro_block_group_mutex);
9437 btrfs_end_transaction(trans);
9441 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9442 struct btrfs_fs_info *fs_info, u64 type)
9444 u64 alloc_flags = get_alloc_profile(fs_info, type);
9446 return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE);
9450 * helper to account the unused space of all the readonly block group in the
9451 * space_info. takes mirrors into account.
9453 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9455 struct btrfs_block_group_cache *block_group;
9459 /* It's df, we don't care if it's racy */
9460 if (list_empty(&sinfo->ro_bgs))
9463 spin_lock(&sinfo->lock);
9464 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9465 spin_lock(&block_group->lock);
9467 if (!block_group->ro) {
9468 spin_unlock(&block_group->lock);
9472 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9473 BTRFS_BLOCK_GROUP_RAID10 |
9474 BTRFS_BLOCK_GROUP_DUP))
9479 free_bytes += (block_group->key.offset -
9480 btrfs_block_group_used(&block_group->item)) *
9483 spin_unlock(&block_group->lock);
9485 spin_unlock(&sinfo->lock);
9490 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache)
9492 struct btrfs_space_info *sinfo = cache->space_info;
9497 spin_lock(&sinfo->lock);
9498 spin_lock(&cache->lock);
9500 num_bytes = cache->key.offset - cache->reserved -
9501 cache->pinned - cache->bytes_super -
9502 btrfs_block_group_used(&cache->item);
9503 sinfo->bytes_readonly -= num_bytes;
9504 list_del_init(&cache->ro_list);
9506 spin_unlock(&cache->lock);
9507 spin_unlock(&sinfo->lock);
9511 * checks to see if its even possible to relocate this block group.
9513 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9514 * ok to go ahead and try.
9516 int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr)
9518 struct btrfs_root *root = fs_info->extent_root;
9519 struct btrfs_block_group_cache *block_group;
9520 struct btrfs_space_info *space_info;
9521 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
9522 struct btrfs_device *device;
9523 struct btrfs_trans_handle *trans;
9533 debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG);
9535 block_group = btrfs_lookup_block_group(fs_info, bytenr);
9537 /* odd, couldn't find the block group, leave it alone */
9541 "can't find block group for bytenr %llu",
9546 min_free = btrfs_block_group_used(&block_group->item);
9548 /* no bytes used, we're good */
9552 space_info = block_group->space_info;
9553 spin_lock(&space_info->lock);
9555 full = space_info->full;
9558 * if this is the last block group we have in this space, we can't
9559 * relocate it unless we're able to allocate a new chunk below.
9561 * Otherwise, we need to make sure we have room in the space to handle
9562 * all of the extents from this block group. If we can, we're good
9564 if ((space_info->total_bytes != block_group->key.offset) &&
9565 (btrfs_space_info_used(space_info, false) + min_free <
9566 space_info->total_bytes)) {
9567 spin_unlock(&space_info->lock);
9570 spin_unlock(&space_info->lock);
9573 * ok we don't have enough space, but maybe we have free space on our
9574 * devices to allocate new chunks for relocation, so loop through our
9575 * alloc devices and guess if we have enough space. if this block
9576 * group is going to be restriped, run checks against the target
9577 * profile instead of the current one.
9589 target = get_restripe_target(fs_info, block_group->flags);
9591 index = __get_raid_index(extended_to_chunk(target));
9594 * this is just a balance, so if we were marked as full
9595 * we know there is no space for a new chunk
9600 "no space to alloc new chunk for block group %llu",
9601 block_group->key.objectid);
9605 index = get_block_group_index(block_group);
9608 if (index == BTRFS_RAID_RAID10) {
9612 } else if (index == BTRFS_RAID_RAID1) {
9614 } else if (index == BTRFS_RAID_DUP) {
9617 } else if (index == BTRFS_RAID_RAID0) {
9618 dev_min = fs_devices->rw_devices;
9619 min_free = div64_u64(min_free, dev_min);
9622 /* We need to do this so that we can look at pending chunks */
9623 trans = btrfs_join_transaction(root);
9624 if (IS_ERR(trans)) {
9625 ret = PTR_ERR(trans);
9629 mutex_lock(&fs_info->chunk_mutex);
9630 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9634 * check to make sure we can actually find a chunk with enough
9635 * space to fit our block group in.
9637 if (device->total_bytes > device->bytes_used + min_free &&
9638 !device->is_tgtdev_for_dev_replace) {
9639 ret = find_free_dev_extent(trans, device, min_free,
9644 if (dev_nr >= dev_min)
9650 if (debug && ret == -1)
9652 "no space to allocate a new chunk for block group %llu",
9653 block_group->key.objectid);
9654 mutex_unlock(&fs_info->chunk_mutex);
9655 btrfs_end_transaction(trans);
9657 btrfs_put_block_group(block_group);
9661 static int find_first_block_group(struct btrfs_fs_info *fs_info,
9662 struct btrfs_path *path,
9663 struct btrfs_key *key)
9665 struct btrfs_root *root = fs_info->extent_root;
9667 struct btrfs_key found_key;
9668 struct extent_buffer *leaf;
9671 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9676 slot = path->slots[0];
9677 leaf = path->nodes[0];
9678 if (slot >= btrfs_header_nritems(leaf)) {
9679 ret = btrfs_next_leaf(root, path);
9686 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9688 if (found_key.objectid >= key->objectid &&
9689 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9690 struct extent_map_tree *em_tree;
9691 struct extent_map *em;
9693 em_tree = &root->fs_info->mapping_tree.map_tree;
9694 read_lock(&em_tree->lock);
9695 em = lookup_extent_mapping(em_tree, found_key.objectid,
9697 read_unlock(&em_tree->lock);
9700 "logical %llu len %llu found bg but no related chunk",
9701 found_key.objectid, found_key.offset);
9706 free_extent_map(em);
9715 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9717 struct btrfs_block_group_cache *block_group;
9721 struct inode *inode;
9723 block_group = btrfs_lookup_first_block_group(info, last);
9724 while (block_group) {
9725 spin_lock(&block_group->lock);
9726 if (block_group->iref)
9728 spin_unlock(&block_group->lock);
9729 block_group = next_block_group(info, block_group);
9738 inode = block_group->inode;
9739 block_group->iref = 0;
9740 block_group->inode = NULL;
9741 spin_unlock(&block_group->lock);
9742 ASSERT(block_group->io_ctl.inode == NULL);
9744 last = block_group->key.objectid + block_group->key.offset;
9745 btrfs_put_block_group(block_group);
9750 * Must be called only after stopping all workers, since we could have block
9751 * group caching kthreads running, and therefore they could race with us if we
9752 * freed the block groups before stopping them.
9754 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9756 struct btrfs_block_group_cache *block_group;
9757 struct btrfs_space_info *space_info;
9758 struct btrfs_caching_control *caching_ctl;
9761 down_write(&info->commit_root_sem);
9762 while (!list_empty(&info->caching_block_groups)) {
9763 caching_ctl = list_entry(info->caching_block_groups.next,
9764 struct btrfs_caching_control, list);
9765 list_del(&caching_ctl->list);
9766 put_caching_control(caching_ctl);
9768 up_write(&info->commit_root_sem);
9770 spin_lock(&info->unused_bgs_lock);
9771 while (!list_empty(&info->unused_bgs)) {
9772 block_group = list_first_entry(&info->unused_bgs,
9773 struct btrfs_block_group_cache,
9775 list_del_init(&block_group->bg_list);
9776 btrfs_put_block_group(block_group);
9778 spin_unlock(&info->unused_bgs_lock);
9780 spin_lock(&info->block_group_cache_lock);
9781 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9782 block_group = rb_entry(n, struct btrfs_block_group_cache,
9784 rb_erase(&block_group->cache_node,
9785 &info->block_group_cache_tree);
9786 RB_CLEAR_NODE(&block_group->cache_node);
9787 spin_unlock(&info->block_group_cache_lock);
9789 down_write(&block_group->space_info->groups_sem);
9790 list_del(&block_group->list);
9791 up_write(&block_group->space_info->groups_sem);
9794 * We haven't cached this block group, which means we could
9795 * possibly have excluded extents on this block group.
9797 if (block_group->cached == BTRFS_CACHE_NO ||
9798 block_group->cached == BTRFS_CACHE_ERROR)
9799 free_excluded_extents(info, block_group);
9801 btrfs_remove_free_space_cache(block_group);
9802 ASSERT(block_group->cached != BTRFS_CACHE_STARTED);
9803 ASSERT(list_empty(&block_group->dirty_list));
9804 ASSERT(list_empty(&block_group->io_list));
9805 ASSERT(list_empty(&block_group->bg_list));
9806 ASSERT(atomic_read(&block_group->count) == 1);
9807 btrfs_put_block_group(block_group);
9809 spin_lock(&info->block_group_cache_lock);
9811 spin_unlock(&info->block_group_cache_lock);
9813 /* now that all the block groups are freed, go through and
9814 * free all the space_info structs. This is only called during
9815 * the final stages of unmount, and so we know nobody is
9816 * using them. We call synchronize_rcu() once before we start,
9817 * just to be on the safe side.
9821 release_global_block_rsv(info);
9823 while (!list_empty(&info->space_info)) {
9826 space_info = list_entry(info->space_info.next,
9827 struct btrfs_space_info,
9831 * Do not hide this behind enospc_debug, this is actually
9832 * important and indicates a real bug if this happens.
9834 if (WARN_ON(space_info->bytes_pinned > 0 ||
9835 space_info->bytes_reserved > 0 ||
9836 space_info->bytes_may_use > 0))
9837 dump_space_info(info, space_info, 0, 0);
9838 list_del(&space_info->list);
9839 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9840 struct kobject *kobj;
9841 kobj = space_info->block_group_kobjs[i];
9842 space_info->block_group_kobjs[i] = NULL;
9848 kobject_del(&space_info->kobj);
9849 kobject_put(&space_info->kobj);
9854 static void __link_block_group(struct btrfs_space_info *space_info,
9855 struct btrfs_block_group_cache *cache)
9857 int index = get_block_group_index(cache);
9860 down_write(&space_info->groups_sem);
9861 if (list_empty(&space_info->block_groups[index]))
9863 list_add_tail(&cache->list, &space_info->block_groups[index]);
9864 up_write(&space_info->groups_sem);
9867 struct raid_kobject *rkobj;
9870 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9873 rkobj->raid_type = index;
9874 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9875 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9876 "%s", get_raid_name(index));
9878 kobject_put(&rkobj->kobj);
9881 space_info->block_group_kobjs[index] = &rkobj->kobj;
9886 btrfs_warn(cache->fs_info,
9887 "failed to add kobject for block cache, ignoring");
9890 static struct btrfs_block_group_cache *
9891 btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info,
9892 u64 start, u64 size)
9894 struct btrfs_block_group_cache *cache;
9896 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9900 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9902 if (!cache->free_space_ctl) {
9907 cache->key.objectid = start;
9908 cache->key.offset = size;
9909 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9911 cache->sectorsize = fs_info->sectorsize;
9912 cache->fs_info = fs_info;
9913 cache->full_stripe_len = btrfs_full_stripe_len(fs_info,
9914 &fs_info->mapping_tree,
9916 set_free_space_tree_thresholds(cache);
9918 atomic_set(&cache->count, 1);
9919 spin_lock_init(&cache->lock);
9920 init_rwsem(&cache->data_rwsem);
9921 INIT_LIST_HEAD(&cache->list);
9922 INIT_LIST_HEAD(&cache->cluster_list);
9923 INIT_LIST_HEAD(&cache->bg_list);
9924 INIT_LIST_HEAD(&cache->ro_list);
9925 INIT_LIST_HEAD(&cache->dirty_list);
9926 INIT_LIST_HEAD(&cache->io_list);
9927 btrfs_init_free_space_ctl(cache);
9928 atomic_set(&cache->trimming, 0);
9929 mutex_init(&cache->free_space_lock);
9930 btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root);
9935 int btrfs_read_block_groups(struct btrfs_fs_info *info)
9937 struct btrfs_path *path;
9939 struct btrfs_block_group_cache *cache;
9940 struct btrfs_space_info *space_info;
9941 struct btrfs_key key;
9942 struct btrfs_key found_key;
9943 struct extent_buffer *leaf;
9949 feature = btrfs_super_incompat_flags(info->super_copy);
9950 mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS);
9954 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9955 path = btrfs_alloc_path();
9958 path->reada = READA_FORWARD;
9960 cache_gen = btrfs_super_cache_generation(info->super_copy);
9961 if (btrfs_test_opt(info, SPACE_CACHE) &&
9962 btrfs_super_generation(info->super_copy) != cache_gen)
9964 if (btrfs_test_opt(info, CLEAR_CACHE))
9968 ret = find_first_block_group(info, path, &key);
9974 leaf = path->nodes[0];
9975 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9977 cache = btrfs_create_block_group_cache(info, found_key.objectid,
9986 * When we mount with old space cache, we need to
9987 * set BTRFS_DC_CLEAR and set dirty flag.
9989 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9990 * truncate the old free space cache inode and
9992 * b) Setting 'dirty flag' makes sure that we flush
9993 * the new space cache info onto disk.
9995 if (btrfs_test_opt(info, SPACE_CACHE))
9996 cache->disk_cache_state = BTRFS_DC_CLEAR;
9999 read_extent_buffer(leaf, &cache->item,
10000 btrfs_item_ptr_offset(leaf, path->slots[0]),
10001 sizeof(cache->item));
10002 cache->flags = btrfs_block_group_flags(&cache->item);
10004 ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) &&
10005 (cache->flags & BTRFS_BLOCK_GROUP_DATA))) {
10007 "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups",
10008 cache->key.objectid);
10013 key.objectid = found_key.objectid + found_key.offset;
10014 btrfs_release_path(path);
10017 * We need to exclude the super stripes now so that the space
10018 * info has super bytes accounted for, otherwise we'll think
10019 * we have more space than we actually do.
10021 ret = exclude_super_stripes(info, cache);
10024 * We may have excluded something, so call this just in
10027 free_excluded_extents(info, cache);
10028 btrfs_put_block_group(cache);
10033 * check for two cases, either we are full, and therefore
10034 * don't need to bother with the caching work since we won't
10035 * find any space, or we are empty, and we can just add all
10036 * the space in and be done with it. This saves us _alot_ of
10037 * time, particularly in the full case.
10039 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10040 cache->last_byte_to_unpin = (u64)-1;
10041 cache->cached = BTRFS_CACHE_FINISHED;
10042 free_excluded_extents(info, cache);
10043 } else if (btrfs_block_group_used(&cache->item) == 0) {
10044 cache->last_byte_to_unpin = (u64)-1;
10045 cache->cached = BTRFS_CACHE_FINISHED;
10046 add_new_free_space(cache, info,
10047 found_key.objectid,
10048 found_key.objectid +
10050 free_excluded_extents(info, cache);
10053 ret = btrfs_add_block_group_cache(info, cache);
10055 btrfs_remove_free_space_cache(cache);
10056 btrfs_put_block_group(cache);
10060 trace_btrfs_add_block_group(info, cache, 0);
10061 ret = update_space_info(info, cache->flags, found_key.offset,
10062 btrfs_block_group_used(&cache->item),
10063 cache->bytes_super, &space_info);
10065 btrfs_remove_free_space_cache(cache);
10066 spin_lock(&info->block_group_cache_lock);
10067 rb_erase(&cache->cache_node,
10068 &info->block_group_cache_tree);
10069 RB_CLEAR_NODE(&cache->cache_node);
10070 spin_unlock(&info->block_group_cache_lock);
10071 btrfs_put_block_group(cache);
10075 cache->space_info = space_info;
10077 __link_block_group(space_info, cache);
10079 set_avail_alloc_bits(info, cache->flags);
10080 if (btrfs_chunk_readonly(info, cache->key.objectid)) {
10081 inc_block_group_ro(cache, 1);
10082 } else if (btrfs_block_group_used(&cache->item) == 0) {
10083 spin_lock(&info->unused_bgs_lock);
10084 /* Should always be true but just in case. */
10085 if (list_empty(&cache->bg_list)) {
10086 btrfs_get_block_group(cache);
10087 list_add_tail(&cache->bg_list,
10088 &info->unused_bgs);
10090 spin_unlock(&info->unused_bgs_lock);
10094 list_for_each_entry_rcu(space_info, &info->space_info, list) {
10095 if (!(get_alloc_profile(info, space_info->flags) &
10096 (BTRFS_BLOCK_GROUP_RAID10 |
10097 BTRFS_BLOCK_GROUP_RAID1 |
10098 BTRFS_BLOCK_GROUP_RAID5 |
10099 BTRFS_BLOCK_GROUP_RAID6 |
10100 BTRFS_BLOCK_GROUP_DUP)))
10103 * avoid allocating from un-mirrored block group if there are
10104 * mirrored block groups.
10106 list_for_each_entry(cache,
10107 &space_info->block_groups[BTRFS_RAID_RAID0],
10109 inc_block_group_ro(cache, 1);
10110 list_for_each_entry(cache,
10111 &space_info->block_groups[BTRFS_RAID_SINGLE],
10113 inc_block_group_ro(cache, 1);
10116 init_global_block_rsv(info);
10119 btrfs_free_path(path);
10123 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10124 struct btrfs_fs_info *fs_info)
10126 struct btrfs_block_group_cache *block_group, *tmp;
10127 struct btrfs_root *extent_root = fs_info->extent_root;
10128 struct btrfs_block_group_item item;
10129 struct btrfs_key key;
10131 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10133 trans->can_flush_pending_bgs = false;
10134 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10138 spin_lock(&block_group->lock);
10139 memcpy(&item, &block_group->item, sizeof(item));
10140 memcpy(&key, &block_group->key, sizeof(key));
10141 spin_unlock(&block_group->lock);
10143 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10146 btrfs_abort_transaction(trans, ret);
10147 ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid,
10150 btrfs_abort_transaction(trans, ret);
10151 add_block_group_free_space(trans, fs_info, block_group);
10152 /* already aborted the transaction if it failed. */
10154 list_del_init(&block_group->bg_list);
10156 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10159 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10160 struct btrfs_fs_info *fs_info, u64 bytes_used,
10161 u64 type, u64 chunk_objectid, u64 chunk_offset,
10164 struct btrfs_block_group_cache *cache;
10167 btrfs_set_log_full_commit(fs_info, trans);
10169 cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size);
10173 btrfs_set_block_group_used(&cache->item, bytes_used);
10174 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10175 btrfs_set_block_group_flags(&cache->item, type);
10177 cache->flags = type;
10178 cache->last_byte_to_unpin = (u64)-1;
10179 cache->cached = BTRFS_CACHE_FINISHED;
10180 cache->needs_free_space = 1;
10181 ret = exclude_super_stripes(fs_info, cache);
10184 * We may have excluded something, so call this just in
10187 free_excluded_extents(fs_info, cache);
10188 btrfs_put_block_group(cache);
10192 add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size);
10194 free_excluded_extents(fs_info, cache);
10196 #ifdef CONFIG_BTRFS_DEBUG
10197 if (btrfs_should_fragment_free_space(cache)) {
10198 u64 new_bytes_used = size - bytes_used;
10200 bytes_used += new_bytes_used >> 1;
10201 fragment_free_space(cache);
10205 * Call to ensure the corresponding space_info object is created and
10206 * assigned to our block group, but don't update its counters just yet.
10207 * We want our bg to be added to the rbtree with its ->space_info set.
10209 ret = update_space_info(fs_info, cache->flags, 0, 0, 0,
10210 &cache->space_info);
10212 btrfs_remove_free_space_cache(cache);
10213 btrfs_put_block_group(cache);
10217 ret = btrfs_add_block_group_cache(fs_info, cache);
10219 btrfs_remove_free_space_cache(cache);
10220 btrfs_put_block_group(cache);
10225 * Now that our block group has its ->space_info set and is inserted in
10226 * the rbtree, update the space info's counters.
10228 trace_btrfs_add_block_group(fs_info, cache, 1);
10229 ret = update_space_info(fs_info, cache->flags, size, bytes_used,
10230 cache->bytes_super, &cache->space_info);
10232 btrfs_remove_free_space_cache(cache);
10233 spin_lock(&fs_info->block_group_cache_lock);
10234 rb_erase(&cache->cache_node,
10235 &fs_info->block_group_cache_tree);
10236 RB_CLEAR_NODE(&cache->cache_node);
10237 spin_unlock(&fs_info->block_group_cache_lock);
10238 btrfs_put_block_group(cache);
10241 update_global_block_rsv(fs_info);
10243 __link_block_group(cache->space_info, cache);
10245 list_add_tail(&cache->bg_list, &trans->new_bgs);
10247 set_avail_alloc_bits(fs_info, type);
10251 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10253 u64 extra_flags = chunk_to_extended(flags) &
10254 BTRFS_EXTENDED_PROFILE_MASK;
10256 write_seqlock(&fs_info->profiles_lock);
10257 if (flags & BTRFS_BLOCK_GROUP_DATA)
10258 fs_info->avail_data_alloc_bits &= ~extra_flags;
10259 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10260 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10261 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10262 fs_info->avail_system_alloc_bits &= ~extra_flags;
10263 write_sequnlock(&fs_info->profiles_lock);
10266 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10267 struct btrfs_fs_info *fs_info, u64 group_start,
10268 struct extent_map *em)
10270 struct btrfs_root *root = fs_info->extent_root;
10271 struct btrfs_path *path;
10272 struct btrfs_block_group_cache *block_group;
10273 struct btrfs_free_cluster *cluster;
10274 struct btrfs_root *tree_root = fs_info->tree_root;
10275 struct btrfs_key key;
10276 struct inode *inode;
10277 struct kobject *kobj = NULL;
10281 struct btrfs_caching_control *caching_ctl = NULL;
10284 block_group = btrfs_lookup_block_group(fs_info, group_start);
10285 BUG_ON(!block_group);
10286 BUG_ON(!block_group->ro);
10289 * Free the reserved super bytes from this block group before
10292 free_excluded_extents(fs_info, block_group);
10294 memcpy(&key, &block_group->key, sizeof(key));
10295 index = get_block_group_index(block_group);
10296 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10297 BTRFS_BLOCK_GROUP_RAID1 |
10298 BTRFS_BLOCK_GROUP_RAID10))
10303 /* make sure this block group isn't part of an allocation cluster */
10304 cluster = &fs_info->data_alloc_cluster;
10305 spin_lock(&cluster->refill_lock);
10306 btrfs_return_cluster_to_free_space(block_group, cluster);
10307 spin_unlock(&cluster->refill_lock);
10310 * make sure this block group isn't part of a metadata
10311 * allocation cluster
10313 cluster = &fs_info->meta_alloc_cluster;
10314 spin_lock(&cluster->refill_lock);
10315 btrfs_return_cluster_to_free_space(block_group, cluster);
10316 spin_unlock(&cluster->refill_lock);
10318 path = btrfs_alloc_path();
10325 * get the inode first so any iput calls done for the io_list
10326 * aren't the final iput (no unlinks allowed now)
10328 inode = lookup_free_space_inode(fs_info, block_group, path);
10330 mutex_lock(&trans->transaction->cache_write_mutex);
10332 * make sure our free spache cache IO is done before remove the
10335 spin_lock(&trans->transaction->dirty_bgs_lock);
10336 if (!list_empty(&block_group->io_list)) {
10337 list_del_init(&block_group->io_list);
10339 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10341 spin_unlock(&trans->transaction->dirty_bgs_lock);
10342 btrfs_wait_cache_io(trans, block_group, path);
10343 btrfs_put_block_group(block_group);
10344 spin_lock(&trans->transaction->dirty_bgs_lock);
10347 if (!list_empty(&block_group->dirty_list)) {
10348 list_del_init(&block_group->dirty_list);
10349 btrfs_put_block_group(block_group);
10351 spin_unlock(&trans->transaction->dirty_bgs_lock);
10352 mutex_unlock(&trans->transaction->cache_write_mutex);
10354 if (!IS_ERR(inode)) {
10355 ret = btrfs_orphan_add(trans, BTRFS_I(inode));
10357 btrfs_add_delayed_iput(inode);
10360 clear_nlink(inode);
10361 /* One for the block groups ref */
10362 spin_lock(&block_group->lock);
10363 if (block_group->iref) {
10364 block_group->iref = 0;
10365 block_group->inode = NULL;
10366 spin_unlock(&block_group->lock);
10369 spin_unlock(&block_group->lock);
10371 /* One for our lookup ref */
10372 btrfs_add_delayed_iput(inode);
10375 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10376 key.offset = block_group->key.objectid;
10379 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10383 btrfs_release_path(path);
10385 ret = btrfs_del_item(trans, tree_root, path);
10388 btrfs_release_path(path);
10391 spin_lock(&fs_info->block_group_cache_lock);
10392 rb_erase(&block_group->cache_node,
10393 &fs_info->block_group_cache_tree);
10394 RB_CLEAR_NODE(&block_group->cache_node);
10396 if (fs_info->first_logical_byte == block_group->key.objectid)
10397 fs_info->first_logical_byte = (u64)-1;
10398 spin_unlock(&fs_info->block_group_cache_lock);
10400 down_write(&block_group->space_info->groups_sem);
10402 * we must use list_del_init so people can check to see if they
10403 * are still on the list after taking the semaphore
10405 list_del_init(&block_group->list);
10406 if (list_empty(&block_group->space_info->block_groups[index])) {
10407 kobj = block_group->space_info->block_group_kobjs[index];
10408 block_group->space_info->block_group_kobjs[index] = NULL;
10409 clear_avail_alloc_bits(fs_info, block_group->flags);
10411 up_write(&block_group->space_info->groups_sem);
10417 if (block_group->has_caching_ctl)
10418 caching_ctl = get_caching_control(block_group);
10419 if (block_group->cached == BTRFS_CACHE_STARTED)
10420 wait_block_group_cache_done(block_group);
10421 if (block_group->has_caching_ctl) {
10422 down_write(&fs_info->commit_root_sem);
10423 if (!caching_ctl) {
10424 struct btrfs_caching_control *ctl;
10426 list_for_each_entry(ctl,
10427 &fs_info->caching_block_groups, list)
10428 if (ctl->block_group == block_group) {
10430 refcount_inc(&caching_ctl->count);
10435 list_del_init(&caching_ctl->list);
10436 up_write(&fs_info->commit_root_sem);
10438 /* Once for the caching bgs list and once for us. */
10439 put_caching_control(caching_ctl);
10440 put_caching_control(caching_ctl);
10444 spin_lock(&trans->transaction->dirty_bgs_lock);
10445 if (!list_empty(&block_group->dirty_list)) {
10448 if (!list_empty(&block_group->io_list)) {
10451 spin_unlock(&trans->transaction->dirty_bgs_lock);
10452 btrfs_remove_free_space_cache(block_group);
10454 spin_lock(&block_group->space_info->lock);
10455 list_del_init(&block_group->ro_list);
10457 if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) {
10458 WARN_ON(block_group->space_info->total_bytes
10459 < block_group->key.offset);
10460 WARN_ON(block_group->space_info->bytes_readonly
10461 < block_group->key.offset);
10462 WARN_ON(block_group->space_info->disk_total
10463 < block_group->key.offset * factor);
10465 block_group->space_info->total_bytes -= block_group->key.offset;
10466 block_group->space_info->bytes_readonly -= block_group->key.offset;
10467 block_group->space_info->disk_total -= block_group->key.offset * factor;
10469 spin_unlock(&block_group->space_info->lock);
10471 memcpy(&key, &block_group->key, sizeof(key));
10473 mutex_lock(&fs_info->chunk_mutex);
10474 if (!list_empty(&em->list)) {
10475 /* We're in the transaction->pending_chunks list. */
10476 free_extent_map(em);
10478 spin_lock(&block_group->lock);
10479 block_group->removed = 1;
10481 * At this point trimming can't start on this block group, because we
10482 * removed the block group from the tree fs_info->block_group_cache_tree
10483 * so no one can't find it anymore and even if someone already got this
10484 * block group before we removed it from the rbtree, they have already
10485 * incremented block_group->trimming - if they didn't, they won't find
10486 * any free space entries because we already removed them all when we
10487 * called btrfs_remove_free_space_cache().
10489 * And we must not remove the extent map from the fs_info->mapping_tree
10490 * to prevent the same logical address range and physical device space
10491 * ranges from being reused for a new block group. This is because our
10492 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10493 * completely transactionless, so while it is trimming a range the
10494 * currently running transaction might finish and a new one start,
10495 * allowing for new block groups to be created that can reuse the same
10496 * physical device locations unless we take this special care.
10498 * There may also be an implicit trim operation if the file system
10499 * is mounted with -odiscard. The same protections must remain
10500 * in place until the extents have been discarded completely when
10501 * the transaction commit has completed.
10503 remove_em = (atomic_read(&block_group->trimming) == 0);
10505 * Make sure a trimmer task always sees the em in the pinned_chunks list
10506 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10507 * before checking block_group->removed).
10511 * Our em might be in trans->transaction->pending_chunks which
10512 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10513 * and so is the fs_info->pinned_chunks list.
10515 * So at this point we must be holding the chunk_mutex to avoid
10516 * any races with chunk allocation (more specifically at
10517 * volumes.c:contains_pending_extent()), to ensure it always
10518 * sees the em, either in the pending_chunks list or in the
10519 * pinned_chunks list.
10521 list_move_tail(&em->list, &fs_info->pinned_chunks);
10523 spin_unlock(&block_group->lock);
10526 struct extent_map_tree *em_tree;
10528 em_tree = &fs_info->mapping_tree.map_tree;
10529 write_lock(&em_tree->lock);
10531 * The em might be in the pending_chunks list, so make sure the
10532 * chunk mutex is locked, since remove_extent_mapping() will
10533 * delete us from that list.
10535 remove_extent_mapping(em_tree, em);
10536 write_unlock(&em_tree->lock);
10537 /* once for the tree */
10538 free_extent_map(em);
10541 mutex_unlock(&fs_info->chunk_mutex);
10543 ret = remove_block_group_free_space(trans, fs_info, block_group);
10547 btrfs_put_block_group(block_group);
10548 btrfs_put_block_group(block_group);
10550 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10556 ret = btrfs_del_item(trans, root, path);
10558 btrfs_free_path(path);
10562 struct btrfs_trans_handle *
10563 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10564 const u64 chunk_offset)
10566 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10567 struct extent_map *em;
10568 struct map_lookup *map;
10569 unsigned int num_items;
10571 read_lock(&em_tree->lock);
10572 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10573 read_unlock(&em_tree->lock);
10574 ASSERT(em && em->start == chunk_offset);
10577 * We need to reserve 3 + N units from the metadata space info in order
10578 * to remove a block group (done at btrfs_remove_chunk() and at
10579 * btrfs_remove_block_group()), which are used for:
10581 * 1 unit for adding the free space inode's orphan (located in the tree
10583 * 1 unit for deleting the block group item (located in the extent
10585 * 1 unit for deleting the free space item (located in tree of tree
10587 * N units for deleting N device extent items corresponding to each
10588 * stripe (located in the device tree).
10590 * In order to remove a block group we also need to reserve units in the
10591 * system space info in order to update the chunk tree (update one or
10592 * more device items and remove one chunk item), but this is done at
10593 * btrfs_remove_chunk() through a call to check_system_chunk().
10595 map = em->map_lookup;
10596 num_items = 3 + map->num_stripes;
10597 free_extent_map(em);
10599 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10604 * Process the unused_bgs list and remove any that don't have any allocated
10605 * space inside of them.
10607 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10609 struct btrfs_block_group_cache *block_group;
10610 struct btrfs_space_info *space_info;
10611 struct btrfs_trans_handle *trans;
10614 if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags))
10617 spin_lock(&fs_info->unused_bgs_lock);
10618 while (!list_empty(&fs_info->unused_bgs)) {
10622 block_group = list_first_entry(&fs_info->unused_bgs,
10623 struct btrfs_block_group_cache,
10625 list_del_init(&block_group->bg_list);
10627 space_info = block_group->space_info;
10629 if (ret || btrfs_mixed_space_info(space_info)) {
10630 btrfs_put_block_group(block_group);
10633 spin_unlock(&fs_info->unused_bgs_lock);
10635 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10637 /* Don't want to race with allocators so take the groups_sem */
10638 down_write(&space_info->groups_sem);
10639 spin_lock(&block_group->lock);
10640 if (block_group->reserved ||
10641 btrfs_block_group_used(&block_group->item) ||
10643 list_is_singular(&block_group->list)) {
10645 * We want to bail if we made new allocations or have
10646 * outstanding allocations in this block group. We do
10647 * the ro check in case balance is currently acting on
10648 * this block group.
10650 spin_unlock(&block_group->lock);
10651 up_write(&space_info->groups_sem);
10654 spin_unlock(&block_group->lock);
10656 /* We don't want to force the issue, only flip if it's ok. */
10657 ret = inc_block_group_ro(block_group, 0);
10658 up_write(&space_info->groups_sem);
10665 * Want to do this before we do anything else so we can recover
10666 * properly if we fail to join the transaction.
10668 trans = btrfs_start_trans_remove_block_group(fs_info,
10669 block_group->key.objectid);
10670 if (IS_ERR(trans)) {
10671 btrfs_dec_block_group_ro(block_group);
10672 ret = PTR_ERR(trans);
10677 * We could have pending pinned extents for this block group,
10678 * just delete them, we don't care about them anymore.
10680 start = block_group->key.objectid;
10681 end = start + block_group->key.offset - 1;
10683 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10684 * btrfs_finish_extent_commit(). If we are at transaction N,
10685 * another task might be running finish_extent_commit() for the
10686 * previous transaction N - 1, and have seen a range belonging
10687 * to the block group in freed_extents[] before we were able to
10688 * clear the whole block group range from freed_extents[]. This
10689 * means that task can lookup for the block group after we
10690 * unpinned it from freed_extents[] and removed it, leading to
10691 * a BUG_ON() at btrfs_unpin_extent_range().
10693 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10694 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10697 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10698 btrfs_dec_block_group_ro(block_group);
10701 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10704 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10705 btrfs_dec_block_group_ro(block_group);
10708 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10710 /* Reset pinned so btrfs_put_block_group doesn't complain */
10711 spin_lock(&space_info->lock);
10712 spin_lock(&block_group->lock);
10714 space_info->bytes_pinned -= block_group->pinned;
10715 space_info->bytes_readonly += block_group->pinned;
10716 percpu_counter_add(&space_info->total_bytes_pinned,
10717 -block_group->pinned);
10718 block_group->pinned = 0;
10720 spin_unlock(&block_group->lock);
10721 spin_unlock(&space_info->lock);
10723 /* DISCARD can flip during remount */
10724 trimming = btrfs_test_opt(fs_info, DISCARD);
10726 /* Implicit trim during transaction commit. */
10728 btrfs_get_block_group_trimming(block_group);
10731 * Btrfs_remove_chunk will abort the transaction if things go
10734 ret = btrfs_remove_chunk(trans, fs_info,
10735 block_group->key.objectid);
10739 btrfs_put_block_group_trimming(block_group);
10744 * If we're not mounted with -odiscard, we can just forget
10745 * about this block group. Otherwise we'll need to wait
10746 * until transaction commit to do the actual discard.
10749 spin_lock(&fs_info->unused_bgs_lock);
10751 * A concurrent scrub might have added us to the list
10752 * fs_info->unused_bgs, so use a list_move operation
10753 * to add the block group to the deleted_bgs list.
10755 list_move(&block_group->bg_list,
10756 &trans->transaction->deleted_bgs);
10757 spin_unlock(&fs_info->unused_bgs_lock);
10758 btrfs_get_block_group(block_group);
10761 btrfs_end_transaction(trans);
10763 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10764 btrfs_put_block_group(block_group);
10765 spin_lock(&fs_info->unused_bgs_lock);
10767 spin_unlock(&fs_info->unused_bgs_lock);
10770 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10772 struct btrfs_space_info *space_info;
10773 struct btrfs_super_block *disk_super;
10779 disk_super = fs_info->super_copy;
10780 if (!btrfs_super_root(disk_super))
10783 features = btrfs_super_incompat_flags(disk_super);
10784 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10787 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10788 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10793 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10794 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10796 flags = BTRFS_BLOCK_GROUP_METADATA;
10797 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10801 flags = BTRFS_BLOCK_GROUP_DATA;
10802 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10808 int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info,
10809 u64 start, u64 end)
10811 return unpin_extent_range(fs_info, start, end, false);
10815 * It used to be that old block groups would be left around forever.
10816 * Iterating over them would be enough to trim unused space. Since we
10817 * now automatically remove them, we also need to iterate over unallocated
10820 * We don't want a transaction for this since the discard may take a
10821 * substantial amount of time. We don't require that a transaction be
10822 * running, but we do need to take a running transaction into account
10823 * to ensure that we're not discarding chunks that were released in
10824 * the current transaction.
10826 * Holding the chunks lock will prevent other threads from allocating
10827 * or releasing chunks, but it won't prevent a running transaction
10828 * from committing and releasing the memory that the pending chunks
10829 * list head uses. For that, we need to take a reference to the
10832 static int btrfs_trim_free_extents(struct btrfs_device *device,
10833 u64 minlen, u64 *trimmed)
10835 u64 start = 0, len = 0;
10840 /* Not writeable = nothing to do. */
10841 if (!device->writeable)
10844 /* No free space = nothing to do. */
10845 if (device->total_bytes <= device->bytes_used)
10851 struct btrfs_fs_info *fs_info = device->fs_info;
10852 struct btrfs_transaction *trans;
10855 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10859 down_read(&fs_info->commit_root_sem);
10861 spin_lock(&fs_info->trans_lock);
10862 trans = fs_info->running_transaction;
10864 refcount_inc(&trans->use_count);
10865 spin_unlock(&fs_info->trans_lock);
10867 ret = find_free_dev_extent_start(trans, device, minlen, start,
10870 btrfs_put_transaction(trans);
10873 up_read(&fs_info->commit_root_sem);
10874 mutex_unlock(&fs_info->chunk_mutex);
10875 if (ret == -ENOSPC)
10880 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10881 up_read(&fs_info->commit_root_sem);
10882 mutex_unlock(&fs_info->chunk_mutex);
10890 if (fatal_signal_pending(current)) {
10891 ret = -ERESTARTSYS;
10901 int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range)
10903 struct btrfs_block_group_cache *cache = NULL;
10904 struct btrfs_device *device;
10905 struct list_head *devices;
10910 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10914 * try to trim all FS space, our block group may start from non-zero.
10916 if (range->len == total_bytes)
10917 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10919 cache = btrfs_lookup_block_group(fs_info, range->start);
10922 if (cache->key.objectid >= (range->start + range->len)) {
10923 btrfs_put_block_group(cache);
10927 start = max(range->start, cache->key.objectid);
10928 end = min(range->start + range->len,
10929 cache->key.objectid + cache->key.offset);
10931 if (end - start >= range->minlen) {
10932 if (!block_group_cache_done(cache)) {
10933 ret = cache_block_group(cache, 0);
10935 btrfs_put_block_group(cache);
10938 ret = wait_block_group_cache_done(cache);
10940 btrfs_put_block_group(cache);
10944 ret = btrfs_trim_block_group(cache,
10950 trimmed += group_trimmed;
10952 btrfs_put_block_group(cache);
10957 cache = next_block_group(fs_info, cache);
10960 mutex_lock(&fs_info->fs_devices->device_list_mutex);
10961 devices = &fs_info->fs_devices->alloc_list;
10962 list_for_each_entry(device, devices, dev_alloc_list) {
10963 ret = btrfs_trim_free_extents(device, range->minlen,
10968 trimmed += group_trimmed;
10970 mutex_unlock(&fs_info->fs_devices->device_list_mutex);
10972 range->len = trimmed;
10977 * btrfs_{start,end}_write_no_snapshoting() are similar to
10978 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10979 * data into the page cache through nocow before the subvolume is snapshoted,
10980 * but flush the data into disk after the snapshot creation, or to prevent
10981 * operations while snapshoting is ongoing and that cause the snapshot to be
10982 * inconsistent (writes followed by expanding truncates for example).
10984 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10986 percpu_counter_dec(&root->subv_writers->counter);
10988 * Make sure counter is updated before we wake up waiters.
10991 if (waitqueue_active(&root->subv_writers->wait))
10992 wake_up(&root->subv_writers->wait);
10995 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10997 if (atomic_read(&root->will_be_snapshoted))
11000 percpu_counter_inc(&root->subv_writers->counter);
11002 * Make sure counter is updated before we check for snapshot creation.
11005 if (atomic_read(&root->will_be_snapshoted)) {
11006 btrfs_end_write_no_snapshoting(root);
11012 static int wait_snapshoting_atomic_t(atomic_t *a)
11018 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11023 ret = btrfs_start_write_no_snapshoting(root);
11026 wait_on_atomic_t(&root->will_be_snapshoted,
11027 wait_snapshoting_atomic_t,
11028 TASK_UNINTERRUPTIBLE);