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/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
36 #include "free-space-tree.h"
41 #undef SCRAMBLE_DELAYED_REFS
44 * control flags for do_chunk_alloc's force field
45 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
46 * if we really need one.
48 * CHUNK_ALLOC_LIMITED means to only try and allocate one
49 * if we have very few chunks already allocated. This is
50 * used as part of the clustering code to help make sure
51 * we have a good pool of storage to cluster in, without
52 * filling the FS with empty chunks
54 * CHUNK_ALLOC_FORCE means it must try to allocate one
58 CHUNK_ALLOC_NO_FORCE = 0,
59 CHUNK_ALLOC_LIMITED = 1,
60 CHUNK_ALLOC_FORCE = 2,
64 * Control how reservations are dealt with.
66 * RESERVE_FREE - freeing a reservation.
67 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
69 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
70 * bytes_may_use as the ENOSPC accounting is done elsewhere
75 RESERVE_ALLOC_NO_ACCOUNT = 2,
78 static int update_block_group(struct btrfs_trans_handle *trans,
79 struct btrfs_root *root, u64 bytenr,
80 u64 num_bytes, int alloc);
81 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
82 struct btrfs_root *root,
83 struct btrfs_delayed_ref_node *node, u64 parent,
84 u64 root_objectid, u64 owner_objectid,
85 u64 owner_offset, int refs_to_drop,
86 struct btrfs_delayed_extent_op *extra_op);
87 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
88 struct extent_buffer *leaf,
89 struct btrfs_extent_item *ei);
90 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
91 struct btrfs_root *root,
92 u64 parent, u64 root_objectid,
93 u64 flags, u64 owner, u64 offset,
94 struct btrfs_key *ins, int ref_mod);
95 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
96 struct btrfs_root *root,
97 u64 parent, u64 root_objectid,
98 u64 flags, struct btrfs_disk_key *key,
99 int level, struct btrfs_key *ins);
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
114 static int __reserve_metadata_bytes(struct btrfs_root *root,
115 struct btrfs_space_info *space_info,
117 enum btrfs_reserve_flush_enum flush);
118 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
119 struct btrfs_space_info *space_info,
121 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
122 struct btrfs_space_info *space_info,
126 block_group_cache_done(struct btrfs_block_group_cache *cache)
129 return cache->cached == BTRFS_CACHE_FINISHED ||
130 cache->cached == BTRFS_CACHE_ERROR;
133 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
135 return (cache->flags & bits) == bits;
138 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
140 atomic_inc(&cache->count);
143 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
145 if (atomic_dec_and_test(&cache->count)) {
146 WARN_ON(cache->pinned > 0);
147 WARN_ON(cache->reserved > 0);
148 kfree(cache->free_space_ctl);
154 * this adds the block group to the fs_info rb tree for the block group
157 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
158 struct btrfs_block_group_cache *block_group)
161 struct rb_node *parent = NULL;
162 struct btrfs_block_group_cache *cache;
164 spin_lock(&info->block_group_cache_lock);
165 p = &info->block_group_cache_tree.rb_node;
169 cache = rb_entry(parent, struct btrfs_block_group_cache,
171 if (block_group->key.objectid < cache->key.objectid) {
173 } else if (block_group->key.objectid > cache->key.objectid) {
176 spin_unlock(&info->block_group_cache_lock);
181 rb_link_node(&block_group->cache_node, parent, p);
182 rb_insert_color(&block_group->cache_node,
183 &info->block_group_cache_tree);
185 if (info->first_logical_byte > block_group->key.objectid)
186 info->first_logical_byte = block_group->key.objectid;
188 spin_unlock(&info->block_group_cache_lock);
194 * This will return the block group at or after bytenr if contains is 0, else
195 * it will return the block group that contains the bytenr
197 static struct btrfs_block_group_cache *
198 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
201 struct btrfs_block_group_cache *cache, *ret = NULL;
205 spin_lock(&info->block_group_cache_lock);
206 n = info->block_group_cache_tree.rb_node;
209 cache = rb_entry(n, struct btrfs_block_group_cache,
211 end = cache->key.objectid + cache->key.offset - 1;
212 start = cache->key.objectid;
214 if (bytenr < start) {
215 if (!contains && (!ret || start < ret->key.objectid))
218 } else if (bytenr > start) {
219 if (contains && bytenr <= end) {
230 btrfs_get_block_group(ret);
231 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
232 info->first_logical_byte = ret->key.objectid;
234 spin_unlock(&info->block_group_cache_lock);
239 static int add_excluded_extent(struct btrfs_root *root,
240 u64 start, u64 num_bytes)
242 u64 end = start + num_bytes - 1;
243 set_extent_bits(&root->fs_info->freed_extents[0],
244 start, end, EXTENT_UPTODATE);
245 set_extent_bits(&root->fs_info->freed_extents[1],
246 start, end, EXTENT_UPTODATE);
250 static void free_excluded_extents(struct btrfs_root *root,
251 struct btrfs_block_group_cache *cache)
255 start = cache->key.objectid;
256 end = start + cache->key.offset - 1;
258 clear_extent_bits(&root->fs_info->freed_extents[0],
259 start, end, EXTENT_UPTODATE);
260 clear_extent_bits(&root->fs_info->freed_extents[1],
261 start, end, EXTENT_UPTODATE);
264 static int exclude_super_stripes(struct btrfs_root *root,
265 struct btrfs_block_group_cache *cache)
272 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
273 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
274 cache->bytes_super += stripe_len;
275 ret = add_excluded_extent(root, cache->key.objectid,
281 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
282 bytenr = btrfs_sb_offset(i);
283 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
284 cache->key.objectid, bytenr,
285 0, &logical, &nr, &stripe_len);
292 if (logical[nr] > cache->key.objectid +
296 if (logical[nr] + stripe_len <= cache->key.objectid)
300 if (start < cache->key.objectid) {
301 start = cache->key.objectid;
302 len = (logical[nr] + stripe_len) - start;
304 len = min_t(u64, stripe_len,
305 cache->key.objectid +
306 cache->key.offset - start);
309 cache->bytes_super += len;
310 ret = add_excluded_extent(root, start, len);
322 static struct btrfs_caching_control *
323 get_caching_control(struct btrfs_block_group_cache *cache)
325 struct btrfs_caching_control *ctl;
327 spin_lock(&cache->lock);
328 if (!cache->caching_ctl) {
329 spin_unlock(&cache->lock);
333 ctl = cache->caching_ctl;
334 atomic_inc(&ctl->count);
335 spin_unlock(&cache->lock);
339 static void put_caching_control(struct btrfs_caching_control *ctl)
341 if (atomic_dec_and_test(&ctl->count))
345 #ifdef CONFIG_BTRFS_DEBUG
346 static void fragment_free_space(struct btrfs_root *root,
347 struct btrfs_block_group_cache *block_group)
349 u64 start = block_group->key.objectid;
350 u64 len = block_group->key.offset;
351 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
352 root->nodesize : root->sectorsize;
353 u64 step = chunk << 1;
355 while (len > chunk) {
356 btrfs_remove_free_space(block_group, start, chunk);
367 * this is only called by cache_block_group, since we could have freed extents
368 * we need to check the pinned_extents for any extents that can't be used yet
369 * since their free space will be released as soon as the transaction commits.
371 u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
372 struct btrfs_fs_info *info, u64 start, u64 end)
374 u64 extent_start, extent_end, size, total_added = 0;
377 while (start < end) {
378 ret = find_first_extent_bit(info->pinned_extents, start,
379 &extent_start, &extent_end,
380 EXTENT_DIRTY | EXTENT_UPTODATE,
385 if (extent_start <= start) {
386 start = extent_end + 1;
387 } else if (extent_start > start && extent_start < end) {
388 size = extent_start - start;
390 ret = btrfs_add_free_space(block_group, start,
392 BUG_ON(ret); /* -ENOMEM or logic error */
393 start = extent_end + 1;
402 ret = btrfs_add_free_space(block_group, start, size);
403 BUG_ON(ret); /* -ENOMEM or logic error */
409 static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl)
411 struct btrfs_block_group_cache *block_group;
412 struct btrfs_fs_info *fs_info;
413 struct btrfs_root *extent_root;
414 struct btrfs_path *path;
415 struct extent_buffer *leaf;
416 struct btrfs_key key;
423 block_group = caching_ctl->block_group;
424 fs_info = block_group->fs_info;
425 extent_root = fs_info->extent_root;
427 path = btrfs_alloc_path();
431 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
433 #ifdef CONFIG_BTRFS_DEBUG
435 * If we're fragmenting we don't want to make anybody think we can
436 * allocate from this block group until we've had a chance to fragment
439 if (btrfs_should_fragment_free_space(extent_root, block_group))
443 * We don't want to deadlock with somebody trying to allocate a new
444 * extent for the extent root while also trying to search the extent
445 * root to add free space. So we skip locking and search the commit
446 * root, since its read-only
448 path->skip_locking = 1;
449 path->search_commit_root = 1;
450 path->reada = READA_FORWARD;
454 key.type = BTRFS_EXTENT_ITEM_KEY;
457 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
461 leaf = path->nodes[0];
462 nritems = btrfs_header_nritems(leaf);
465 if (btrfs_fs_closing(fs_info) > 1) {
470 if (path->slots[0] < nritems) {
471 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
473 ret = find_next_key(path, 0, &key);
477 if (need_resched() ||
478 rwsem_is_contended(&fs_info->commit_root_sem)) {
480 caching_ctl->progress = last;
481 btrfs_release_path(path);
482 up_read(&fs_info->commit_root_sem);
483 mutex_unlock(&caching_ctl->mutex);
485 mutex_lock(&caching_ctl->mutex);
486 down_read(&fs_info->commit_root_sem);
490 ret = btrfs_next_leaf(extent_root, path);
495 leaf = path->nodes[0];
496 nritems = btrfs_header_nritems(leaf);
500 if (key.objectid < last) {
503 key.type = BTRFS_EXTENT_ITEM_KEY;
506 caching_ctl->progress = last;
507 btrfs_release_path(path);
511 if (key.objectid < block_group->key.objectid) {
516 if (key.objectid >= block_group->key.objectid +
517 block_group->key.offset)
520 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
521 key.type == BTRFS_METADATA_ITEM_KEY) {
522 total_found += add_new_free_space(block_group,
525 if (key.type == BTRFS_METADATA_ITEM_KEY)
526 last = key.objectid +
527 fs_info->tree_root->nodesize;
529 last = key.objectid + key.offset;
531 if (total_found > CACHING_CTL_WAKE_UP) {
534 wake_up(&caching_ctl->wait);
541 total_found += add_new_free_space(block_group, fs_info, last,
542 block_group->key.objectid +
543 block_group->key.offset);
544 caching_ctl->progress = (u64)-1;
547 btrfs_free_path(path);
551 static noinline void caching_thread(struct btrfs_work *work)
553 struct btrfs_block_group_cache *block_group;
554 struct btrfs_fs_info *fs_info;
555 struct btrfs_caching_control *caching_ctl;
556 struct btrfs_root *extent_root;
559 caching_ctl = container_of(work, struct btrfs_caching_control, work);
560 block_group = caching_ctl->block_group;
561 fs_info = block_group->fs_info;
562 extent_root = fs_info->extent_root;
564 mutex_lock(&caching_ctl->mutex);
565 down_read(&fs_info->commit_root_sem);
567 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE))
568 ret = load_free_space_tree(caching_ctl);
570 ret = load_extent_tree_free(caching_ctl);
572 spin_lock(&block_group->lock);
573 block_group->caching_ctl = NULL;
574 block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED;
575 spin_unlock(&block_group->lock);
577 #ifdef CONFIG_BTRFS_DEBUG
578 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
581 spin_lock(&block_group->space_info->lock);
582 spin_lock(&block_group->lock);
583 bytes_used = block_group->key.offset -
584 btrfs_block_group_used(&block_group->item);
585 block_group->space_info->bytes_used += bytes_used >> 1;
586 spin_unlock(&block_group->lock);
587 spin_unlock(&block_group->space_info->lock);
588 fragment_free_space(extent_root, block_group);
592 caching_ctl->progress = (u64)-1;
594 up_read(&fs_info->commit_root_sem);
595 free_excluded_extents(fs_info->extent_root, block_group);
596 mutex_unlock(&caching_ctl->mutex);
598 wake_up(&caching_ctl->wait);
600 put_caching_control(caching_ctl);
601 btrfs_put_block_group(block_group);
604 static int cache_block_group(struct btrfs_block_group_cache *cache,
608 struct btrfs_fs_info *fs_info = cache->fs_info;
609 struct btrfs_caching_control *caching_ctl;
612 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
616 INIT_LIST_HEAD(&caching_ctl->list);
617 mutex_init(&caching_ctl->mutex);
618 init_waitqueue_head(&caching_ctl->wait);
619 caching_ctl->block_group = cache;
620 caching_ctl->progress = cache->key.objectid;
621 atomic_set(&caching_ctl->count, 1);
622 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
623 caching_thread, NULL, NULL);
625 spin_lock(&cache->lock);
627 * This should be a rare occasion, but this could happen I think in the
628 * case where one thread starts to load the space cache info, and then
629 * some other thread starts a transaction commit which tries to do an
630 * allocation while the other thread is still loading the space cache
631 * info. The previous loop should have kept us from choosing this block
632 * group, but if we've moved to the state where we will wait on caching
633 * block groups we need to first check if we're doing a fast load here,
634 * so we can wait for it to finish, otherwise we could end up allocating
635 * from a block group who's cache gets evicted for one reason or
638 while (cache->cached == BTRFS_CACHE_FAST) {
639 struct btrfs_caching_control *ctl;
641 ctl = cache->caching_ctl;
642 atomic_inc(&ctl->count);
643 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
644 spin_unlock(&cache->lock);
648 finish_wait(&ctl->wait, &wait);
649 put_caching_control(ctl);
650 spin_lock(&cache->lock);
653 if (cache->cached != BTRFS_CACHE_NO) {
654 spin_unlock(&cache->lock);
658 WARN_ON(cache->caching_ctl);
659 cache->caching_ctl = caching_ctl;
660 cache->cached = BTRFS_CACHE_FAST;
661 spin_unlock(&cache->lock);
663 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
664 mutex_lock(&caching_ctl->mutex);
665 ret = load_free_space_cache(fs_info, cache);
667 spin_lock(&cache->lock);
669 cache->caching_ctl = NULL;
670 cache->cached = BTRFS_CACHE_FINISHED;
671 cache->last_byte_to_unpin = (u64)-1;
672 caching_ctl->progress = (u64)-1;
674 if (load_cache_only) {
675 cache->caching_ctl = NULL;
676 cache->cached = BTRFS_CACHE_NO;
678 cache->cached = BTRFS_CACHE_STARTED;
679 cache->has_caching_ctl = 1;
682 spin_unlock(&cache->lock);
683 #ifdef CONFIG_BTRFS_DEBUG
685 btrfs_should_fragment_free_space(fs_info->extent_root,
689 spin_lock(&cache->space_info->lock);
690 spin_lock(&cache->lock);
691 bytes_used = cache->key.offset -
692 btrfs_block_group_used(&cache->item);
693 cache->space_info->bytes_used += bytes_used >> 1;
694 spin_unlock(&cache->lock);
695 spin_unlock(&cache->space_info->lock);
696 fragment_free_space(fs_info->extent_root, cache);
699 mutex_unlock(&caching_ctl->mutex);
701 wake_up(&caching_ctl->wait);
703 put_caching_control(caching_ctl);
704 free_excluded_extents(fs_info->extent_root, cache);
709 * We're either using the free space tree or no caching at all.
710 * Set cached to the appropriate value and wakeup any waiters.
712 spin_lock(&cache->lock);
713 if (load_cache_only) {
714 cache->caching_ctl = NULL;
715 cache->cached = BTRFS_CACHE_NO;
717 cache->cached = BTRFS_CACHE_STARTED;
718 cache->has_caching_ctl = 1;
720 spin_unlock(&cache->lock);
721 wake_up(&caching_ctl->wait);
724 if (load_cache_only) {
725 put_caching_control(caching_ctl);
729 down_write(&fs_info->commit_root_sem);
730 atomic_inc(&caching_ctl->count);
731 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
732 up_write(&fs_info->commit_root_sem);
734 btrfs_get_block_group(cache);
736 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
742 * return the block group that starts at or after bytenr
744 static struct btrfs_block_group_cache *
745 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
747 struct btrfs_block_group_cache *cache;
749 cache = block_group_cache_tree_search(info, bytenr, 0);
755 * return the block group that contains the given bytenr
757 struct btrfs_block_group_cache *btrfs_lookup_block_group(
758 struct btrfs_fs_info *info,
761 struct btrfs_block_group_cache *cache;
763 cache = block_group_cache_tree_search(info, bytenr, 1);
768 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
771 struct list_head *head = &info->space_info;
772 struct btrfs_space_info *found;
774 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
777 list_for_each_entry_rcu(found, head, list) {
778 if (found->flags & flags) {
788 * after adding space to the filesystem, we need to clear the full flags
789 * on all the space infos.
791 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
793 struct list_head *head = &info->space_info;
794 struct btrfs_space_info *found;
797 list_for_each_entry_rcu(found, head, list)
802 /* simple helper to search for an existing data extent at a given offset */
803 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
806 struct btrfs_key key;
807 struct btrfs_path *path;
809 path = btrfs_alloc_path();
813 key.objectid = start;
815 key.type = BTRFS_EXTENT_ITEM_KEY;
816 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
818 btrfs_free_path(path);
823 * helper function to lookup reference count and flags of a tree block.
825 * the head node for delayed ref is used to store the sum of all the
826 * reference count modifications queued up in the rbtree. the head
827 * node may also store the extent flags to set. This way you can check
828 * to see what the reference count and extent flags would be if all of
829 * the delayed refs are not processed.
831 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
832 struct btrfs_root *root, u64 bytenr,
833 u64 offset, int metadata, u64 *refs, u64 *flags)
835 struct btrfs_delayed_ref_head *head;
836 struct btrfs_delayed_ref_root *delayed_refs;
837 struct btrfs_path *path;
838 struct btrfs_extent_item *ei;
839 struct extent_buffer *leaf;
840 struct btrfs_key key;
847 * If we don't have skinny metadata, don't bother doing anything
850 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
851 offset = root->nodesize;
855 path = btrfs_alloc_path();
860 path->skip_locking = 1;
861 path->search_commit_root = 1;
865 key.objectid = bytenr;
868 key.type = BTRFS_METADATA_ITEM_KEY;
870 key.type = BTRFS_EXTENT_ITEM_KEY;
872 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
877 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
878 if (path->slots[0]) {
880 btrfs_item_key_to_cpu(path->nodes[0], &key,
882 if (key.objectid == bytenr &&
883 key.type == BTRFS_EXTENT_ITEM_KEY &&
884 key.offset == root->nodesize)
890 leaf = path->nodes[0];
891 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
892 if (item_size >= sizeof(*ei)) {
893 ei = btrfs_item_ptr(leaf, path->slots[0],
894 struct btrfs_extent_item);
895 num_refs = btrfs_extent_refs(leaf, ei);
896 extent_flags = btrfs_extent_flags(leaf, ei);
898 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
899 struct btrfs_extent_item_v0 *ei0;
900 BUG_ON(item_size != sizeof(*ei0));
901 ei0 = btrfs_item_ptr(leaf, path->slots[0],
902 struct btrfs_extent_item_v0);
903 num_refs = btrfs_extent_refs_v0(leaf, ei0);
904 /* FIXME: this isn't correct for data */
905 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
910 BUG_ON(num_refs == 0);
920 delayed_refs = &trans->transaction->delayed_refs;
921 spin_lock(&delayed_refs->lock);
922 head = btrfs_find_delayed_ref_head(trans, bytenr);
924 if (!mutex_trylock(&head->mutex)) {
925 atomic_inc(&head->node.refs);
926 spin_unlock(&delayed_refs->lock);
928 btrfs_release_path(path);
931 * Mutex was contended, block until it's released and try
934 mutex_lock(&head->mutex);
935 mutex_unlock(&head->mutex);
936 btrfs_put_delayed_ref(&head->node);
939 spin_lock(&head->lock);
940 if (head->extent_op && head->extent_op->update_flags)
941 extent_flags |= head->extent_op->flags_to_set;
943 BUG_ON(num_refs == 0);
945 num_refs += head->node.ref_mod;
946 spin_unlock(&head->lock);
947 mutex_unlock(&head->mutex);
949 spin_unlock(&delayed_refs->lock);
951 WARN_ON(num_refs == 0);
955 *flags = extent_flags;
957 btrfs_free_path(path);
962 * Back reference rules. Back refs have three main goals:
964 * 1) differentiate between all holders of references to an extent so that
965 * when a reference is dropped we can make sure it was a valid reference
966 * before freeing the extent.
968 * 2) Provide enough information to quickly find the holders of an extent
969 * if we notice a given block is corrupted or bad.
971 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
972 * maintenance. This is actually the same as #2, but with a slightly
973 * different use case.
975 * There are two kinds of back refs. The implicit back refs is optimized
976 * for pointers in non-shared tree blocks. For a given pointer in a block,
977 * back refs of this kind provide information about the block's owner tree
978 * and the pointer's key. These information allow us to find the block by
979 * b-tree searching. The full back refs is for pointers in tree blocks not
980 * referenced by their owner trees. The location of tree block is recorded
981 * in the back refs. Actually the full back refs is generic, and can be
982 * used in all cases the implicit back refs is used. The major shortcoming
983 * of the full back refs is its overhead. Every time a tree block gets
984 * COWed, we have to update back refs entry for all pointers in it.
986 * For a newly allocated tree block, we use implicit back refs for
987 * pointers in it. This means most tree related operations only involve
988 * implicit back refs. For a tree block created in old transaction, the
989 * only way to drop a reference to it is COW it. So we can detect the
990 * event that tree block loses its owner tree's reference and do the
991 * back refs conversion.
993 * When a tree block is COWed through a tree, there are four cases:
995 * The reference count of the block is one and the tree is the block's
996 * owner tree. Nothing to do in this case.
998 * The reference count of the block is one and the tree is not the
999 * block's owner tree. In this case, full back refs is used for pointers
1000 * in the block. Remove these full back refs, add implicit back refs for
1001 * every pointers in the new block.
1003 * The reference count of the block is greater than one and the tree is
1004 * the block's owner tree. In this case, implicit back refs is used for
1005 * pointers in the block. Add full back refs for every pointers in the
1006 * block, increase lower level extents' reference counts. The original
1007 * implicit back refs are entailed to the new block.
1009 * The reference count of the block is greater than one and the tree is
1010 * not the block's owner tree. Add implicit back refs for every pointer in
1011 * the new block, increase lower level extents' reference count.
1013 * Back Reference Key composing:
1015 * The key objectid corresponds to the first byte in the extent,
1016 * The key type is used to differentiate between types of back refs.
1017 * There are different meanings of the key offset for different types
1020 * File extents can be referenced by:
1022 * - multiple snapshots, subvolumes, or different generations in one subvol
1023 * - different files inside a single subvolume
1024 * - different offsets inside a file (bookend extents in file.c)
1026 * The extent ref structure for the implicit back refs has fields for:
1028 * - Objectid of the subvolume root
1029 * - objectid of the file holding the reference
1030 * - original offset in the file
1031 * - how many bookend extents
1033 * The key offset for the implicit back refs is hash of the first
1036 * The extent ref structure for the full back refs has field for:
1038 * - number of pointers in the tree leaf
1040 * The key offset for the implicit back refs is the first byte of
1043 * When a file extent is allocated, The implicit back refs is used.
1044 * the fields are filled in:
1046 * (root_key.objectid, inode objectid, offset in file, 1)
1048 * When a file extent is removed file truncation, we find the
1049 * corresponding implicit back refs and check the following fields:
1051 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1053 * Btree extents can be referenced by:
1055 * - Different subvolumes
1057 * Both the implicit back refs and the full back refs for tree blocks
1058 * only consist of key. The key offset for the implicit back refs is
1059 * objectid of block's owner tree. The key offset for the full back refs
1060 * is the first byte of parent block.
1062 * When implicit back refs is used, information about the lowest key and
1063 * level of the tree block are required. These information are stored in
1064 * tree block info structure.
1067 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1068 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1069 struct btrfs_root *root,
1070 struct btrfs_path *path,
1071 u64 owner, u32 extra_size)
1073 struct btrfs_extent_item *item;
1074 struct btrfs_extent_item_v0 *ei0;
1075 struct btrfs_extent_ref_v0 *ref0;
1076 struct btrfs_tree_block_info *bi;
1077 struct extent_buffer *leaf;
1078 struct btrfs_key key;
1079 struct btrfs_key found_key;
1080 u32 new_size = sizeof(*item);
1084 leaf = path->nodes[0];
1085 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1087 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1088 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1089 struct btrfs_extent_item_v0);
1090 refs = btrfs_extent_refs_v0(leaf, ei0);
1092 if (owner == (u64)-1) {
1094 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1095 ret = btrfs_next_leaf(root, path);
1098 BUG_ON(ret > 0); /* Corruption */
1099 leaf = path->nodes[0];
1101 btrfs_item_key_to_cpu(leaf, &found_key,
1103 BUG_ON(key.objectid != found_key.objectid);
1104 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1108 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1109 struct btrfs_extent_ref_v0);
1110 owner = btrfs_ref_objectid_v0(leaf, ref0);
1114 btrfs_release_path(path);
1116 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1117 new_size += sizeof(*bi);
1119 new_size -= sizeof(*ei0);
1120 ret = btrfs_search_slot(trans, root, &key, path,
1121 new_size + extra_size, 1);
1124 BUG_ON(ret); /* Corruption */
1126 btrfs_extend_item(root, path, new_size);
1128 leaf = path->nodes[0];
1129 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1130 btrfs_set_extent_refs(leaf, item, refs);
1131 /* FIXME: get real generation */
1132 btrfs_set_extent_generation(leaf, item, 0);
1133 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1134 btrfs_set_extent_flags(leaf, item,
1135 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1136 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1137 bi = (struct btrfs_tree_block_info *)(item + 1);
1138 /* FIXME: get first key of the block */
1139 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1140 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1142 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1144 btrfs_mark_buffer_dirty(leaf);
1149 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1151 u32 high_crc = ~(u32)0;
1152 u32 low_crc = ~(u32)0;
1155 lenum = cpu_to_le64(root_objectid);
1156 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1157 lenum = cpu_to_le64(owner);
1158 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1159 lenum = cpu_to_le64(offset);
1160 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1162 return ((u64)high_crc << 31) ^ (u64)low_crc;
1165 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1166 struct btrfs_extent_data_ref *ref)
1168 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1169 btrfs_extent_data_ref_objectid(leaf, ref),
1170 btrfs_extent_data_ref_offset(leaf, ref));
1173 static int match_extent_data_ref(struct extent_buffer *leaf,
1174 struct btrfs_extent_data_ref *ref,
1175 u64 root_objectid, u64 owner, u64 offset)
1177 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1178 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1179 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1184 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1185 struct btrfs_root *root,
1186 struct btrfs_path *path,
1187 u64 bytenr, u64 parent,
1189 u64 owner, u64 offset)
1191 struct btrfs_key key;
1192 struct btrfs_extent_data_ref *ref;
1193 struct extent_buffer *leaf;
1199 key.objectid = bytenr;
1201 key.type = BTRFS_SHARED_DATA_REF_KEY;
1202 key.offset = parent;
1204 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1205 key.offset = hash_extent_data_ref(root_objectid,
1210 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1219 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1220 key.type = BTRFS_EXTENT_REF_V0_KEY;
1221 btrfs_release_path(path);
1222 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1233 leaf = path->nodes[0];
1234 nritems = btrfs_header_nritems(leaf);
1236 if (path->slots[0] >= nritems) {
1237 ret = btrfs_next_leaf(root, path);
1243 leaf = path->nodes[0];
1244 nritems = btrfs_header_nritems(leaf);
1248 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1249 if (key.objectid != bytenr ||
1250 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1253 ref = btrfs_item_ptr(leaf, path->slots[0],
1254 struct btrfs_extent_data_ref);
1256 if (match_extent_data_ref(leaf, ref, root_objectid,
1259 btrfs_release_path(path);
1271 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1272 struct btrfs_root *root,
1273 struct btrfs_path *path,
1274 u64 bytenr, u64 parent,
1275 u64 root_objectid, u64 owner,
1276 u64 offset, int refs_to_add)
1278 struct btrfs_key key;
1279 struct extent_buffer *leaf;
1284 key.objectid = bytenr;
1286 key.type = BTRFS_SHARED_DATA_REF_KEY;
1287 key.offset = parent;
1288 size = sizeof(struct btrfs_shared_data_ref);
1290 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1291 key.offset = hash_extent_data_ref(root_objectid,
1293 size = sizeof(struct btrfs_extent_data_ref);
1296 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1297 if (ret && ret != -EEXIST)
1300 leaf = path->nodes[0];
1302 struct btrfs_shared_data_ref *ref;
1303 ref = btrfs_item_ptr(leaf, path->slots[0],
1304 struct btrfs_shared_data_ref);
1306 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1308 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1309 num_refs += refs_to_add;
1310 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1313 struct btrfs_extent_data_ref *ref;
1314 while (ret == -EEXIST) {
1315 ref = btrfs_item_ptr(leaf, path->slots[0],
1316 struct btrfs_extent_data_ref);
1317 if (match_extent_data_ref(leaf, ref, root_objectid,
1320 btrfs_release_path(path);
1322 ret = btrfs_insert_empty_item(trans, root, path, &key,
1324 if (ret && ret != -EEXIST)
1327 leaf = path->nodes[0];
1329 ref = btrfs_item_ptr(leaf, path->slots[0],
1330 struct btrfs_extent_data_ref);
1332 btrfs_set_extent_data_ref_root(leaf, ref,
1334 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1335 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1336 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1338 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1339 num_refs += refs_to_add;
1340 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1343 btrfs_mark_buffer_dirty(leaf);
1346 btrfs_release_path(path);
1350 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1351 struct btrfs_root *root,
1352 struct btrfs_path *path,
1353 int refs_to_drop, int *last_ref)
1355 struct btrfs_key key;
1356 struct btrfs_extent_data_ref *ref1 = NULL;
1357 struct btrfs_shared_data_ref *ref2 = NULL;
1358 struct extent_buffer *leaf;
1362 leaf = path->nodes[0];
1363 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1365 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1366 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1367 struct btrfs_extent_data_ref);
1368 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1369 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1370 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1371 struct btrfs_shared_data_ref);
1372 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1373 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1374 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1375 struct btrfs_extent_ref_v0 *ref0;
1376 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1377 struct btrfs_extent_ref_v0);
1378 num_refs = btrfs_ref_count_v0(leaf, ref0);
1384 BUG_ON(num_refs < refs_to_drop);
1385 num_refs -= refs_to_drop;
1387 if (num_refs == 0) {
1388 ret = btrfs_del_item(trans, root, path);
1391 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1392 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1393 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1394 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1395 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1397 struct btrfs_extent_ref_v0 *ref0;
1398 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1399 struct btrfs_extent_ref_v0);
1400 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1403 btrfs_mark_buffer_dirty(leaf);
1408 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1409 struct btrfs_extent_inline_ref *iref)
1411 struct btrfs_key key;
1412 struct extent_buffer *leaf;
1413 struct btrfs_extent_data_ref *ref1;
1414 struct btrfs_shared_data_ref *ref2;
1417 leaf = path->nodes[0];
1418 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1420 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1421 BTRFS_EXTENT_DATA_REF_KEY) {
1422 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1423 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1425 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1426 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1428 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1429 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1430 struct btrfs_extent_data_ref);
1431 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1432 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1433 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1434 struct btrfs_shared_data_ref);
1435 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1436 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1437 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1438 struct btrfs_extent_ref_v0 *ref0;
1439 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1440 struct btrfs_extent_ref_v0);
1441 num_refs = btrfs_ref_count_v0(leaf, ref0);
1449 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1450 struct btrfs_root *root,
1451 struct btrfs_path *path,
1452 u64 bytenr, u64 parent,
1455 struct btrfs_key key;
1458 key.objectid = bytenr;
1460 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1461 key.offset = parent;
1463 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1464 key.offset = root_objectid;
1467 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1470 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1471 if (ret == -ENOENT && parent) {
1472 btrfs_release_path(path);
1473 key.type = BTRFS_EXTENT_REF_V0_KEY;
1474 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1482 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1483 struct btrfs_root *root,
1484 struct btrfs_path *path,
1485 u64 bytenr, u64 parent,
1488 struct btrfs_key key;
1491 key.objectid = bytenr;
1493 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1494 key.offset = parent;
1496 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1497 key.offset = root_objectid;
1500 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1501 btrfs_release_path(path);
1505 static inline int extent_ref_type(u64 parent, u64 owner)
1508 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1510 type = BTRFS_SHARED_BLOCK_REF_KEY;
1512 type = BTRFS_TREE_BLOCK_REF_KEY;
1515 type = BTRFS_SHARED_DATA_REF_KEY;
1517 type = BTRFS_EXTENT_DATA_REF_KEY;
1522 static int find_next_key(struct btrfs_path *path, int level,
1523 struct btrfs_key *key)
1526 for (; level < BTRFS_MAX_LEVEL; level++) {
1527 if (!path->nodes[level])
1529 if (path->slots[level] + 1 >=
1530 btrfs_header_nritems(path->nodes[level]))
1533 btrfs_item_key_to_cpu(path->nodes[level], key,
1534 path->slots[level] + 1);
1536 btrfs_node_key_to_cpu(path->nodes[level], key,
1537 path->slots[level] + 1);
1544 * look for inline back ref. if back ref is found, *ref_ret is set
1545 * to the address of inline back ref, and 0 is returned.
1547 * if back ref isn't found, *ref_ret is set to the address where it
1548 * should be inserted, and -ENOENT is returned.
1550 * if insert is true and there are too many inline back refs, the path
1551 * points to the extent item, and -EAGAIN is returned.
1553 * NOTE: inline back refs are ordered in the same way that back ref
1554 * items in the tree are ordered.
1556 static noinline_for_stack
1557 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1558 struct btrfs_root *root,
1559 struct btrfs_path *path,
1560 struct btrfs_extent_inline_ref **ref_ret,
1561 u64 bytenr, u64 num_bytes,
1562 u64 parent, u64 root_objectid,
1563 u64 owner, u64 offset, int insert)
1565 struct btrfs_key key;
1566 struct extent_buffer *leaf;
1567 struct btrfs_extent_item *ei;
1568 struct btrfs_extent_inline_ref *iref;
1578 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1581 key.objectid = bytenr;
1582 key.type = BTRFS_EXTENT_ITEM_KEY;
1583 key.offset = num_bytes;
1585 want = extent_ref_type(parent, owner);
1587 extra_size = btrfs_extent_inline_ref_size(want);
1588 path->keep_locks = 1;
1593 * Owner is our parent level, so we can just add one to get the level
1594 * for the block we are interested in.
1596 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1597 key.type = BTRFS_METADATA_ITEM_KEY;
1602 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1609 * We may be a newly converted file system which still has the old fat
1610 * extent entries for metadata, so try and see if we have one of those.
1612 if (ret > 0 && skinny_metadata) {
1613 skinny_metadata = false;
1614 if (path->slots[0]) {
1616 btrfs_item_key_to_cpu(path->nodes[0], &key,
1618 if (key.objectid == bytenr &&
1619 key.type == BTRFS_EXTENT_ITEM_KEY &&
1620 key.offset == num_bytes)
1624 key.objectid = bytenr;
1625 key.type = BTRFS_EXTENT_ITEM_KEY;
1626 key.offset = num_bytes;
1627 btrfs_release_path(path);
1632 if (ret && !insert) {
1635 } else if (WARN_ON(ret)) {
1640 leaf = path->nodes[0];
1641 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1642 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1643 if (item_size < sizeof(*ei)) {
1648 ret = convert_extent_item_v0(trans, root, path, owner,
1654 leaf = path->nodes[0];
1655 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1658 BUG_ON(item_size < sizeof(*ei));
1660 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1661 flags = btrfs_extent_flags(leaf, ei);
1663 ptr = (unsigned long)(ei + 1);
1664 end = (unsigned long)ei + item_size;
1666 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1667 ptr += sizeof(struct btrfs_tree_block_info);
1677 iref = (struct btrfs_extent_inline_ref *)ptr;
1678 type = btrfs_extent_inline_ref_type(leaf, iref);
1682 ptr += btrfs_extent_inline_ref_size(type);
1686 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1687 struct btrfs_extent_data_ref *dref;
1688 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1689 if (match_extent_data_ref(leaf, dref, root_objectid,
1694 if (hash_extent_data_ref_item(leaf, dref) <
1695 hash_extent_data_ref(root_objectid, owner, offset))
1699 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1701 if (parent == ref_offset) {
1705 if (ref_offset < parent)
1708 if (root_objectid == ref_offset) {
1712 if (ref_offset < root_objectid)
1716 ptr += btrfs_extent_inline_ref_size(type);
1718 if (err == -ENOENT && insert) {
1719 if (item_size + extra_size >=
1720 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1725 * To add new inline back ref, we have to make sure
1726 * there is no corresponding back ref item.
1727 * For simplicity, we just do not add new inline back
1728 * ref if there is any kind of item for this block
1730 if (find_next_key(path, 0, &key) == 0 &&
1731 key.objectid == bytenr &&
1732 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1737 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1740 path->keep_locks = 0;
1741 btrfs_unlock_up_safe(path, 1);
1747 * helper to add new inline back ref
1749 static noinline_for_stack
1750 void setup_inline_extent_backref(struct btrfs_root *root,
1751 struct btrfs_path *path,
1752 struct btrfs_extent_inline_ref *iref,
1753 u64 parent, u64 root_objectid,
1754 u64 owner, u64 offset, int refs_to_add,
1755 struct btrfs_delayed_extent_op *extent_op)
1757 struct extent_buffer *leaf;
1758 struct btrfs_extent_item *ei;
1761 unsigned long item_offset;
1766 leaf = path->nodes[0];
1767 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1768 item_offset = (unsigned long)iref - (unsigned long)ei;
1770 type = extent_ref_type(parent, owner);
1771 size = btrfs_extent_inline_ref_size(type);
1773 btrfs_extend_item(root, path, size);
1775 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1776 refs = btrfs_extent_refs(leaf, ei);
1777 refs += refs_to_add;
1778 btrfs_set_extent_refs(leaf, ei, refs);
1780 __run_delayed_extent_op(extent_op, leaf, ei);
1782 ptr = (unsigned long)ei + item_offset;
1783 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1784 if (ptr < end - size)
1785 memmove_extent_buffer(leaf, ptr + size, ptr,
1788 iref = (struct btrfs_extent_inline_ref *)ptr;
1789 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1790 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1791 struct btrfs_extent_data_ref *dref;
1792 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1793 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1794 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1795 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1796 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1797 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1798 struct btrfs_shared_data_ref *sref;
1799 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1800 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1801 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1802 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1803 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1805 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1807 btrfs_mark_buffer_dirty(leaf);
1810 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1811 struct btrfs_root *root,
1812 struct btrfs_path *path,
1813 struct btrfs_extent_inline_ref **ref_ret,
1814 u64 bytenr, u64 num_bytes, u64 parent,
1815 u64 root_objectid, u64 owner, u64 offset)
1819 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1820 bytenr, num_bytes, parent,
1821 root_objectid, owner, offset, 0);
1825 btrfs_release_path(path);
1828 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1829 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1832 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1833 root_objectid, owner, offset);
1839 * helper to update/remove inline back ref
1841 static noinline_for_stack
1842 void update_inline_extent_backref(struct btrfs_root *root,
1843 struct btrfs_path *path,
1844 struct btrfs_extent_inline_ref *iref,
1846 struct btrfs_delayed_extent_op *extent_op,
1849 struct extent_buffer *leaf;
1850 struct btrfs_extent_item *ei;
1851 struct btrfs_extent_data_ref *dref = NULL;
1852 struct btrfs_shared_data_ref *sref = NULL;
1860 leaf = path->nodes[0];
1861 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1862 refs = btrfs_extent_refs(leaf, ei);
1863 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1864 refs += refs_to_mod;
1865 btrfs_set_extent_refs(leaf, ei, refs);
1867 __run_delayed_extent_op(extent_op, leaf, ei);
1869 type = btrfs_extent_inline_ref_type(leaf, iref);
1871 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1872 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1873 refs = btrfs_extent_data_ref_count(leaf, dref);
1874 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1875 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1876 refs = btrfs_shared_data_ref_count(leaf, sref);
1879 BUG_ON(refs_to_mod != -1);
1882 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1883 refs += refs_to_mod;
1886 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1887 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1889 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1892 size = btrfs_extent_inline_ref_size(type);
1893 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1894 ptr = (unsigned long)iref;
1895 end = (unsigned long)ei + item_size;
1896 if (ptr + size < end)
1897 memmove_extent_buffer(leaf, ptr, ptr + size,
1900 btrfs_truncate_item(root, path, item_size, 1);
1902 btrfs_mark_buffer_dirty(leaf);
1905 static noinline_for_stack
1906 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1907 struct btrfs_root *root,
1908 struct btrfs_path *path,
1909 u64 bytenr, u64 num_bytes, u64 parent,
1910 u64 root_objectid, u64 owner,
1911 u64 offset, int refs_to_add,
1912 struct btrfs_delayed_extent_op *extent_op)
1914 struct btrfs_extent_inline_ref *iref;
1917 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1918 bytenr, num_bytes, parent,
1919 root_objectid, owner, offset, 1);
1921 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1922 update_inline_extent_backref(root, path, iref,
1923 refs_to_add, extent_op, NULL);
1924 } else if (ret == -ENOENT) {
1925 setup_inline_extent_backref(root, path, iref, parent,
1926 root_objectid, owner, offset,
1927 refs_to_add, extent_op);
1933 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1934 struct btrfs_root *root,
1935 struct btrfs_path *path,
1936 u64 bytenr, u64 parent, u64 root_objectid,
1937 u64 owner, u64 offset, int refs_to_add)
1940 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1941 BUG_ON(refs_to_add != 1);
1942 ret = insert_tree_block_ref(trans, root, path, bytenr,
1943 parent, root_objectid);
1945 ret = insert_extent_data_ref(trans, root, path, bytenr,
1946 parent, root_objectid,
1947 owner, offset, refs_to_add);
1952 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1953 struct btrfs_root *root,
1954 struct btrfs_path *path,
1955 struct btrfs_extent_inline_ref *iref,
1956 int refs_to_drop, int is_data, int *last_ref)
1960 BUG_ON(!is_data && refs_to_drop != 1);
1962 update_inline_extent_backref(root, path, iref,
1963 -refs_to_drop, NULL, last_ref);
1964 } else if (is_data) {
1965 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1969 ret = btrfs_del_item(trans, root, path);
1974 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1975 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1976 u64 *discarded_bytes)
1979 u64 bytes_left, end;
1980 u64 aligned_start = ALIGN(start, 1 << 9);
1982 if (WARN_ON(start != aligned_start)) {
1983 len -= aligned_start - start;
1984 len = round_down(len, 1 << 9);
1985 start = aligned_start;
1988 *discarded_bytes = 0;
1996 /* Skip any superblocks on this device. */
1997 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1998 u64 sb_start = btrfs_sb_offset(j);
1999 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
2000 u64 size = sb_start - start;
2002 if (!in_range(sb_start, start, bytes_left) &&
2003 !in_range(sb_end, start, bytes_left) &&
2004 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
2008 * Superblock spans beginning of range. Adjust start and
2011 if (sb_start <= start) {
2012 start += sb_end - start;
2017 bytes_left = end - start;
2022 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2025 *discarded_bytes += size;
2026 else if (ret != -EOPNOTSUPP)
2035 bytes_left = end - start;
2039 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2042 *discarded_bytes += bytes_left;
2047 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2048 u64 num_bytes, u64 *actual_bytes)
2051 u64 discarded_bytes = 0;
2052 struct btrfs_bio *bbio = NULL;
2056 * Avoid races with device replace and make sure our bbio has devices
2057 * associated to its stripes that don't go away while we are discarding.
2059 btrfs_bio_counter_inc_blocked(root->fs_info);
2060 /* Tell the block device(s) that the sectors can be discarded */
2061 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2062 bytenr, &num_bytes, &bbio, 0);
2063 /* Error condition is -ENOMEM */
2065 struct btrfs_bio_stripe *stripe = bbio->stripes;
2069 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2071 if (!stripe->dev->can_discard)
2074 ret = btrfs_issue_discard(stripe->dev->bdev,
2079 discarded_bytes += bytes;
2080 else if (ret != -EOPNOTSUPP)
2081 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2084 * Just in case we get back EOPNOTSUPP for some reason,
2085 * just ignore the return value so we don't screw up
2086 * people calling discard_extent.
2090 btrfs_put_bbio(bbio);
2092 btrfs_bio_counter_dec(root->fs_info);
2095 *actual_bytes = discarded_bytes;
2098 if (ret == -EOPNOTSUPP)
2103 /* Can return -ENOMEM */
2104 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2105 struct btrfs_root *root,
2106 u64 bytenr, u64 num_bytes, u64 parent,
2107 u64 root_objectid, u64 owner, u64 offset)
2110 struct btrfs_fs_info *fs_info = root->fs_info;
2112 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2113 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2115 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2116 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2118 parent, root_objectid, (int)owner,
2119 BTRFS_ADD_DELAYED_REF, NULL);
2121 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2122 num_bytes, parent, root_objectid,
2124 BTRFS_ADD_DELAYED_REF, NULL);
2129 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2130 struct btrfs_root *root,
2131 struct btrfs_delayed_ref_node *node,
2132 u64 parent, u64 root_objectid,
2133 u64 owner, u64 offset, int refs_to_add,
2134 struct btrfs_delayed_extent_op *extent_op)
2136 struct btrfs_fs_info *fs_info = root->fs_info;
2137 struct btrfs_path *path;
2138 struct extent_buffer *leaf;
2139 struct btrfs_extent_item *item;
2140 struct btrfs_key key;
2141 u64 bytenr = node->bytenr;
2142 u64 num_bytes = node->num_bytes;
2146 path = btrfs_alloc_path();
2150 path->reada = READA_FORWARD;
2151 path->leave_spinning = 1;
2152 /* this will setup the path even if it fails to insert the back ref */
2153 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2154 bytenr, num_bytes, parent,
2155 root_objectid, owner, offset,
2156 refs_to_add, extent_op);
2157 if ((ret < 0 && ret != -EAGAIN) || !ret)
2161 * Ok we had -EAGAIN which means we didn't have space to insert and
2162 * inline extent ref, so just update the reference count and add a
2165 leaf = path->nodes[0];
2166 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2167 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2168 refs = btrfs_extent_refs(leaf, item);
2169 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2171 __run_delayed_extent_op(extent_op, leaf, item);
2173 btrfs_mark_buffer_dirty(leaf);
2174 btrfs_release_path(path);
2176 path->reada = READA_FORWARD;
2177 path->leave_spinning = 1;
2178 /* now insert the actual backref */
2179 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2180 path, bytenr, parent, root_objectid,
2181 owner, offset, refs_to_add);
2183 btrfs_abort_transaction(trans, root, ret);
2185 btrfs_free_path(path);
2189 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2190 struct btrfs_root *root,
2191 struct btrfs_delayed_ref_node *node,
2192 struct btrfs_delayed_extent_op *extent_op,
2193 int insert_reserved)
2196 struct btrfs_delayed_data_ref *ref;
2197 struct btrfs_key ins;
2202 ins.objectid = node->bytenr;
2203 ins.offset = node->num_bytes;
2204 ins.type = BTRFS_EXTENT_ITEM_KEY;
2206 ref = btrfs_delayed_node_to_data_ref(node);
2207 trace_run_delayed_data_ref(node, ref, node->action);
2209 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2210 parent = ref->parent;
2211 ref_root = ref->root;
2213 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2215 flags |= extent_op->flags_to_set;
2216 ret = alloc_reserved_file_extent(trans, root,
2217 parent, ref_root, flags,
2218 ref->objectid, ref->offset,
2219 &ins, node->ref_mod);
2220 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2221 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2222 ref_root, ref->objectid,
2223 ref->offset, node->ref_mod,
2225 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2226 ret = __btrfs_free_extent(trans, root, node, parent,
2227 ref_root, ref->objectid,
2228 ref->offset, node->ref_mod,
2236 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2237 struct extent_buffer *leaf,
2238 struct btrfs_extent_item *ei)
2240 u64 flags = btrfs_extent_flags(leaf, ei);
2241 if (extent_op->update_flags) {
2242 flags |= extent_op->flags_to_set;
2243 btrfs_set_extent_flags(leaf, ei, flags);
2246 if (extent_op->update_key) {
2247 struct btrfs_tree_block_info *bi;
2248 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2249 bi = (struct btrfs_tree_block_info *)(ei + 1);
2250 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2254 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2255 struct btrfs_root *root,
2256 struct btrfs_delayed_ref_node *node,
2257 struct btrfs_delayed_extent_op *extent_op)
2259 struct btrfs_key key;
2260 struct btrfs_path *path;
2261 struct btrfs_extent_item *ei;
2262 struct extent_buffer *leaf;
2266 int metadata = !extent_op->is_data;
2271 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2274 path = btrfs_alloc_path();
2278 key.objectid = node->bytenr;
2281 key.type = BTRFS_METADATA_ITEM_KEY;
2282 key.offset = extent_op->level;
2284 key.type = BTRFS_EXTENT_ITEM_KEY;
2285 key.offset = node->num_bytes;
2289 path->reada = READA_FORWARD;
2290 path->leave_spinning = 1;
2291 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2299 if (path->slots[0] > 0) {
2301 btrfs_item_key_to_cpu(path->nodes[0], &key,
2303 if (key.objectid == node->bytenr &&
2304 key.type == BTRFS_EXTENT_ITEM_KEY &&
2305 key.offset == node->num_bytes)
2309 btrfs_release_path(path);
2312 key.objectid = node->bytenr;
2313 key.offset = node->num_bytes;
2314 key.type = BTRFS_EXTENT_ITEM_KEY;
2323 leaf = path->nodes[0];
2324 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2325 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2326 if (item_size < sizeof(*ei)) {
2327 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2333 leaf = path->nodes[0];
2334 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2337 BUG_ON(item_size < sizeof(*ei));
2338 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2339 __run_delayed_extent_op(extent_op, leaf, ei);
2341 btrfs_mark_buffer_dirty(leaf);
2343 btrfs_free_path(path);
2347 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2348 struct btrfs_root *root,
2349 struct btrfs_delayed_ref_node *node,
2350 struct btrfs_delayed_extent_op *extent_op,
2351 int insert_reserved)
2354 struct btrfs_delayed_tree_ref *ref;
2355 struct btrfs_key ins;
2358 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2361 ref = btrfs_delayed_node_to_tree_ref(node);
2362 trace_run_delayed_tree_ref(node, ref, node->action);
2364 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2365 parent = ref->parent;
2366 ref_root = ref->root;
2368 ins.objectid = node->bytenr;
2369 if (skinny_metadata) {
2370 ins.offset = ref->level;
2371 ins.type = BTRFS_METADATA_ITEM_KEY;
2373 ins.offset = node->num_bytes;
2374 ins.type = BTRFS_EXTENT_ITEM_KEY;
2377 BUG_ON(node->ref_mod != 1);
2378 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2379 BUG_ON(!extent_op || !extent_op->update_flags);
2380 ret = alloc_reserved_tree_block(trans, root,
2382 extent_op->flags_to_set,
2385 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2386 ret = __btrfs_inc_extent_ref(trans, root, node,
2390 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2391 ret = __btrfs_free_extent(trans, root, node,
2393 ref->level, 0, 1, extent_op);
2400 /* helper function to actually process a single delayed ref entry */
2401 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2402 struct btrfs_root *root,
2403 struct btrfs_delayed_ref_node *node,
2404 struct btrfs_delayed_extent_op *extent_op,
2405 int insert_reserved)
2409 if (trans->aborted) {
2410 if (insert_reserved)
2411 btrfs_pin_extent(root, node->bytenr,
2412 node->num_bytes, 1);
2416 if (btrfs_delayed_ref_is_head(node)) {
2417 struct btrfs_delayed_ref_head *head;
2419 * we've hit the end of the chain and we were supposed
2420 * to insert this extent into the tree. But, it got
2421 * deleted before we ever needed to insert it, so all
2422 * we have to do is clean up the accounting
2425 head = btrfs_delayed_node_to_head(node);
2426 trace_run_delayed_ref_head(node, head, node->action);
2428 if (insert_reserved) {
2429 btrfs_pin_extent(root, node->bytenr,
2430 node->num_bytes, 1);
2431 if (head->is_data) {
2432 ret = btrfs_del_csums(trans, root,
2438 /* Also free its reserved qgroup space */
2439 btrfs_qgroup_free_delayed_ref(root->fs_info,
2440 head->qgroup_ref_root,
2441 head->qgroup_reserved);
2445 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2446 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2447 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2449 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2450 node->type == BTRFS_SHARED_DATA_REF_KEY)
2451 ret = run_delayed_data_ref(trans, root, node, extent_op,
2458 static inline struct btrfs_delayed_ref_node *
2459 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2461 struct btrfs_delayed_ref_node *ref;
2463 if (list_empty(&head->ref_list))
2467 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2468 * This is to prevent a ref count from going down to zero, which deletes
2469 * the extent item from the extent tree, when there still are references
2470 * to add, which would fail because they would not find the extent item.
2472 list_for_each_entry(ref, &head->ref_list, list) {
2473 if (ref->action == BTRFS_ADD_DELAYED_REF)
2477 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2482 * Returns 0 on success or if called with an already aborted transaction.
2483 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2485 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2486 struct btrfs_root *root,
2489 struct btrfs_delayed_ref_root *delayed_refs;
2490 struct btrfs_delayed_ref_node *ref;
2491 struct btrfs_delayed_ref_head *locked_ref = NULL;
2492 struct btrfs_delayed_extent_op *extent_op;
2493 struct btrfs_fs_info *fs_info = root->fs_info;
2494 ktime_t start = ktime_get();
2496 unsigned long count = 0;
2497 unsigned long actual_count = 0;
2498 int must_insert_reserved = 0;
2500 delayed_refs = &trans->transaction->delayed_refs;
2506 spin_lock(&delayed_refs->lock);
2507 locked_ref = btrfs_select_ref_head(trans);
2509 spin_unlock(&delayed_refs->lock);
2513 /* grab the lock that says we are going to process
2514 * all the refs for this head */
2515 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2516 spin_unlock(&delayed_refs->lock);
2518 * we may have dropped the spin lock to get the head
2519 * mutex lock, and that might have given someone else
2520 * time to free the head. If that's true, it has been
2521 * removed from our list and we can move on.
2523 if (ret == -EAGAIN) {
2531 * We need to try and merge add/drops of the same ref since we
2532 * can run into issues with relocate dropping the implicit ref
2533 * and then it being added back again before the drop can
2534 * finish. If we merged anything we need to re-loop so we can
2536 * Or we can get node references of the same type that weren't
2537 * merged when created due to bumps in the tree mod seq, and
2538 * we need to merge them to prevent adding an inline extent
2539 * backref before dropping it (triggering a BUG_ON at
2540 * insert_inline_extent_backref()).
2542 spin_lock(&locked_ref->lock);
2543 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2547 * locked_ref is the head node, so we have to go one
2548 * node back for any delayed ref updates
2550 ref = select_delayed_ref(locked_ref);
2552 if (ref && ref->seq &&
2553 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2554 spin_unlock(&locked_ref->lock);
2555 btrfs_delayed_ref_unlock(locked_ref);
2556 spin_lock(&delayed_refs->lock);
2557 locked_ref->processing = 0;
2558 delayed_refs->num_heads_ready++;
2559 spin_unlock(&delayed_refs->lock);
2567 * record the must insert reserved flag before we
2568 * drop the spin lock.
2570 must_insert_reserved = locked_ref->must_insert_reserved;
2571 locked_ref->must_insert_reserved = 0;
2573 extent_op = locked_ref->extent_op;
2574 locked_ref->extent_op = NULL;
2579 /* All delayed refs have been processed, Go ahead
2580 * and send the head node to run_one_delayed_ref,
2581 * so that any accounting fixes can happen
2583 ref = &locked_ref->node;
2585 if (extent_op && must_insert_reserved) {
2586 btrfs_free_delayed_extent_op(extent_op);
2591 spin_unlock(&locked_ref->lock);
2592 ret = run_delayed_extent_op(trans, root,
2594 btrfs_free_delayed_extent_op(extent_op);
2598 * Need to reset must_insert_reserved if
2599 * there was an error so the abort stuff
2600 * can cleanup the reserved space
2603 if (must_insert_reserved)
2604 locked_ref->must_insert_reserved = 1;
2605 locked_ref->processing = 0;
2606 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2607 btrfs_delayed_ref_unlock(locked_ref);
2614 * Need to drop our head ref lock and re-acquire the
2615 * delayed ref lock and then re-check to make sure
2618 spin_unlock(&locked_ref->lock);
2619 spin_lock(&delayed_refs->lock);
2620 spin_lock(&locked_ref->lock);
2621 if (!list_empty(&locked_ref->ref_list) ||
2622 locked_ref->extent_op) {
2623 spin_unlock(&locked_ref->lock);
2624 spin_unlock(&delayed_refs->lock);
2628 delayed_refs->num_heads--;
2629 rb_erase(&locked_ref->href_node,
2630 &delayed_refs->href_root);
2631 spin_unlock(&delayed_refs->lock);
2635 list_del(&ref->list);
2637 atomic_dec(&delayed_refs->num_entries);
2639 if (!btrfs_delayed_ref_is_head(ref)) {
2641 * when we play the delayed ref, also correct the
2644 switch (ref->action) {
2645 case BTRFS_ADD_DELAYED_REF:
2646 case BTRFS_ADD_DELAYED_EXTENT:
2647 locked_ref->node.ref_mod -= ref->ref_mod;
2649 case BTRFS_DROP_DELAYED_REF:
2650 locked_ref->node.ref_mod += ref->ref_mod;
2656 spin_unlock(&locked_ref->lock);
2658 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2659 must_insert_reserved);
2661 btrfs_free_delayed_extent_op(extent_op);
2663 locked_ref->processing = 0;
2664 btrfs_delayed_ref_unlock(locked_ref);
2665 btrfs_put_delayed_ref(ref);
2666 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2671 * If this node is a head, that means all the refs in this head
2672 * have been dealt with, and we will pick the next head to deal
2673 * with, so we must unlock the head and drop it from the cluster
2674 * list before we release it.
2676 if (btrfs_delayed_ref_is_head(ref)) {
2677 if (locked_ref->is_data &&
2678 locked_ref->total_ref_mod < 0) {
2679 spin_lock(&delayed_refs->lock);
2680 delayed_refs->pending_csums -= ref->num_bytes;
2681 spin_unlock(&delayed_refs->lock);
2683 btrfs_delayed_ref_unlock(locked_ref);
2686 btrfs_put_delayed_ref(ref);
2692 * We don't want to include ref heads since we can have empty ref heads
2693 * and those will drastically skew our runtime down since we just do
2694 * accounting, no actual extent tree updates.
2696 if (actual_count > 0) {
2697 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2701 * We weigh the current average higher than our current runtime
2702 * to avoid large swings in the average.
2704 spin_lock(&delayed_refs->lock);
2705 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2706 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2707 spin_unlock(&delayed_refs->lock);
2712 #ifdef SCRAMBLE_DELAYED_REFS
2714 * Normally delayed refs get processed in ascending bytenr order. This
2715 * correlates in most cases to the order added. To expose dependencies on this
2716 * order, we start to process the tree in the middle instead of the beginning
2718 static u64 find_middle(struct rb_root *root)
2720 struct rb_node *n = root->rb_node;
2721 struct btrfs_delayed_ref_node *entry;
2724 u64 first = 0, last = 0;
2728 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2729 first = entry->bytenr;
2733 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2734 last = entry->bytenr;
2739 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2740 WARN_ON(!entry->in_tree);
2742 middle = entry->bytenr;
2755 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2759 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2760 sizeof(struct btrfs_extent_inline_ref));
2761 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2762 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2765 * We don't ever fill up leaves all the way so multiply by 2 just to be
2766 * closer to what we're really going to want to use.
2768 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2772 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2773 * would require to store the csums for that many bytes.
2775 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2778 u64 num_csums_per_leaf;
2781 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2782 num_csums_per_leaf = div64_u64(csum_size,
2783 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2784 num_csums = div64_u64(csum_bytes, root->sectorsize);
2785 num_csums += num_csums_per_leaf - 1;
2786 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2790 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2791 struct btrfs_root *root)
2793 struct btrfs_block_rsv *global_rsv;
2794 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2795 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2796 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2797 u64 num_bytes, num_dirty_bgs_bytes;
2800 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2801 num_heads = heads_to_leaves(root, num_heads);
2803 num_bytes += (num_heads - 1) * root->nodesize;
2805 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2806 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2808 global_rsv = &root->fs_info->global_block_rsv;
2811 * If we can't allocate any more chunks lets make sure we have _lots_ of
2812 * wiggle room since running delayed refs can create more delayed refs.
2814 if (global_rsv->space_info->full) {
2815 num_dirty_bgs_bytes <<= 1;
2819 spin_lock(&global_rsv->lock);
2820 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2822 spin_unlock(&global_rsv->lock);
2826 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2827 struct btrfs_root *root)
2829 struct btrfs_fs_info *fs_info = root->fs_info;
2831 atomic_read(&trans->transaction->delayed_refs.num_entries);
2836 avg_runtime = fs_info->avg_delayed_ref_runtime;
2837 val = num_entries * avg_runtime;
2838 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2840 if (val >= NSEC_PER_SEC / 2)
2843 return btrfs_check_space_for_delayed_refs(trans, root);
2846 struct async_delayed_refs {
2847 struct btrfs_root *root;
2852 struct completion wait;
2853 struct btrfs_work work;
2856 static void delayed_ref_async_start(struct btrfs_work *work)
2858 struct async_delayed_refs *async;
2859 struct btrfs_trans_handle *trans;
2862 async = container_of(work, struct async_delayed_refs, work);
2864 /* if the commit is already started, we don't need to wait here */
2865 if (btrfs_transaction_blocked(async->root->fs_info))
2868 trans = btrfs_join_transaction(async->root);
2869 if (IS_ERR(trans)) {
2870 async->error = PTR_ERR(trans);
2875 * trans->sync means that when we call end_transaction, we won't
2876 * wait on delayed refs
2880 /* Don't bother flushing if we got into a different transaction */
2881 if (trans->transid > async->transid)
2884 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2888 ret = btrfs_end_transaction(trans, async->root);
2889 if (ret && !async->error)
2893 complete(&async->wait);
2898 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2899 unsigned long count, u64 transid, int wait)
2901 struct async_delayed_refs *async;
2904 async = kmalloc(sizeof(*async), GFP_NOFS);
2908 async->root = root->fs_info->tree_root;
2909 async->count = count;
2911 async->transid = transid;
2916 init_completion(&async->wait);
2918 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2919 delayed_ref_async_start, NULL, NULL);
2921 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2924 wait_for_completion(&async->wait);
2933 * this starts processing the delayed reference count updates and
2934 * extent insertions we have queued up so far. count can be
2935 * 0, which means to process everything in the tree at the start
2936 * of the run (but not newly added entries), or it can be some target
2937 * number you'd like to process.
2939 * Returns 0 on success or if called with an aborted transaction
2940 * Returns <0 on error and aborts the transaction
2942 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2943 struct btrfs_root *root, unsigned long count)
2945 struct rb_node *node;
2946 struct btrfs_delayed_ref_root *delayed_refs;
2947 struct btrfs_delayed_ref_head *head;
2949 int run_all = count == (unsigned long)-1;
2950 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2952 /* We'll clean this up in btrfs_cleanup_transaction */
2956 if (root->fs_info->creating_free_space_tree)
2959 if (root == root->fs_info->extent_root)
2960 root = root->fs_info->tree_root;
2962 delayed_refs = &trans->transaction->delayed_refs;
2964 count = atomic_read(&delayed_refs->num_entries) * 2;
2967 #ifdef SCRAMBLE_DELAYED_REFS
2968 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2970 trans->can_flush_pending_bgs = false;
2971 ret = __btrfs_run_delayed_refs(trans, root, count);
2973 btrfs_abort_transaction(trans, root, ret);
2978 if (!list_empty(&trans->new_bgs))
2979 btrfs_create_pending_block_groups(trans, root);
2981 spin_lock(&delayed_refs->lock);
2982 node = rb_first(&delayed_refs->href_root);
2984 spin_unlock(&delayed_refs->lock);
2987 count = (unsigned long)-1;
2990 head = rb_entry(node, struct btrfs_delayed_ref_head,
2992 if (btrfs_delayed_ref_is_head(&head->node)) {
2993 struct btrfs_delayed_ref_node *ref;
2996 atomic_inc(&ref->refs);
2998 spin_unlock(&delayed_refs->lock);
3000 * Mutex was contended, block until it's
3001 * released and try again
3003 mutex_lock(&head->mutex);
3004 mutex_unlock(&head->mutex);
3006 btrfs_put_delayed_ref(ref);
3012 node = rb_next(node);
3014 spin_unlock(&delayed_refs->lock);
3019 assert_qgroups_uptodate(trans);
3020 trans->can_flush_pending_bgs = can_flush_pending_bgs;
3024 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
3025 struct btrfs_root *root,
3026 u64 bytenr, u64 num_bytes, u64 flags,
3027 int level, int is_data)
3029 struct btrfs_delayed_extent_op *extent_op;
3032 extent_op = btrfs_alloc_delayed_extent_op();
3036 extent_op->flags_to_set = flags;
3037 extent_op->update_flags = true;
3038 extent_op->update_key = false;
3039 extent_op->is_data = is_data ? true : false;
3040 extent_op->level = level;
3042 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3043 num_bytes, extent_op);
3045 btrfs_free_delayed_extent_op(extent_op);
3049 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3050 struct btrfs_root *root,
3051 struct btrfs_path *path,
3052 u64 objectid, u64 offset, u64 bytenr)
3054 struct btrfs_delayed_ref_head *head;
3055 struct btrfs_delayed_ref_node *ref;
3056 struct btrfs_delayed_data_ref *data_ref;
3057 struct btrfs_delayed_ref_root *delayed_refs;
3060 delayed_refs = &trans->transaction->delayed_refs;
3061 spin_lock(&delayed_refs->lock);
3062 head = btrfs_find_delayed_ref_head(trans, bytenr);
3064 spin_unlock(&delayed_refs->lock);
3068 if (!mutex_trylock(&head->mutex)) {
3069 atomic_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_trans_handle *trans,
3112 struct btrfs_root *root,
3113 struct btrfs_path *path,
3114 u64 objectid, u64 offset, u64 bytenr)
3116 struct btrfs_root *extent_root = 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_trans_handle *trans,
3183 struct btrfs_root *root,
3184 u64 objectid, u64 offset, u64 bytenr)
3186 struct btrfs_path *path;
3190 path = btrfs_alloc_path();
3195 ret = check_committed_ref(trans, root, path, objectid,
3197 if (ret && ret != -ENOENT)
3200 ret2 = check_delayed_ref(trans, root, path, objectid,
3202 } while (ret2 == -EAGAIN);
3204 if (ret2 && ret2 != -ENOENT) {
3209 if (ret != -ENOENT || ret2 != -ENOENT)
3212 btrfs_free_path(path);
3213 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3218 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3219 struct btrfs_root *root,
3220 struct extent_buffer *buf,
3221 int full_backref, int inc)
3228 struct btrfs_key key;
3229 struct btrfs_file_extent_item *fi;
3233 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3234 u64, u64, u64, u64, u64, u64);
3237 if (btrfs_test_is_dummy_root(root))
3240 ref_root = btrfs_header_owner(buf);
3241 nritems = btrfs_header_nritems(buf);
3242 level = btrfs_header_level(buf);
3244 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3248 process_func = btrfs_inc_extent_ref;
3250 process_func = btrfs_free_extent;
3253 parent = buf->start;
3257 for (i = 0; i < nritems; i++) {
3259 btrfs_item_key_to_cpu(buf, &key, i);
3260 if (key.type != BTRFS_EXTENT_DATA_KEY)
3262 fi = btrfs_item_ptr(buf, i,
3263 struct btrfs_file_extent_item);
3264 if (btrfs_file_extent_type(buf, fi) ==
3265 BTRFS_FILE_EXTENT_INLINE)
3267 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3271 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3272 key.offset -= btrfs_file_extent_offset(buf, fi);
3273 ret = process_func(trans, root, bytenr, num_bytes,
3274 parent, ref_root, key.objectid,
3279 bytenr = btrfs_node_blockptr(buf, i);
3280 num_bytes = root->nodesize;
3281 ret = process_func(trans, root, bytenr, num_bytes,
3282 parent, ref_root, level - 1, 0);
3292 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3293 struct extent_buffer *buf, int full_backref)
3295 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3298 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3299 struct extent_buffer *buf, int full_backref)
3301 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3304 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3305 struct btrfs_root *root,
3306 struct btrfs_path *path,
3307 struct btrfs_block_group_cache *cache)
3310 struct btrfs_root *extent_root = root->fs_info->extent_root;
3312 struct extent_buffer *leaf;
3314 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3321 leaf = path->nodes[0];
3322 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3323 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3324 btrfs_mark_buffer_dirty(leaf);
3326 btrfs_release_path(path);
3331 static struct btrfs_block_group_cache *
3332 next_block_group(struct btrfs_root *root,
3333 struct btrfs_block_group_cache *cache)
3335 struct rb_node *node;
3337 spin_lock(&root->fs_info->block_group_cache_lock);
3339 /* If our block group was removed, we need a full search. */
3340 if (RB_EMPTY_NODE(&cache->cache_node)) {
3341 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3343 spin_unlock(&root->fs_info->block_group_cache_lock);
3344 btrfs_put_block_group(cache);
3345 cache = btrfs_lookup_first_block_group(root->fs_info,
3349 node = rb_next(&cache->cache_node);
3350 btrfs_put_block_group(cache);
3352 cache = rb_entry(node, struct btrfs_block_group_cache,
3354 btrfs_get_block_group(cache);
3357 spin_unlock(&root->fs_info->block_group_cache_lock);
3361 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3362 struct btrfs_trans_handle *trans,
3363 struct btrfs_path *path)
3365 struct btrfs_root *root = block_group->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(root, 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(root, trans, block_group, path);
3407 /* We've already setup this transaction, go ahead and exit */
3408 if (block_group->cache_generation == trans->transid &&
3409 i_size_read(inode)) {
3410 dcs = BTRFS_DC_SETUP;
3415 * We want to set the generation to 0, that way if anything goes wrong
3416 * from here on out we know not to trust this cache when we load up next
3419 BTRFS_I(inode)->generation = 0;
3420 ret = btrfs_update_inode(trans, root, inode);
3423 * So theoretically we could recover from this, simply set the
3424 * super cache generation to 0 so we know to invalidate the
3425 * cache, but then we'd have to keep track of the block groups
3426 * that fail this way so we know we _have_ to reset this cache
3427 * before the next commit or risk reading stale cache. So to
3428 * limit our exposure to horrible edge cases lets just abort the
3429 * transaction, this only happens in really bad situations
3432 btrfs_abort_transaction(trans, root, ret);
3437 if (i_size_read(inode) > 0) {
3438 ret = btrfs_check_trunc_cache_free_space(root,
3439 &root->fs_info->global_block_rsv);
3443 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3448 spin_lock(&block_group->lock);
3449 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3450 !btrfs_test_opt(root, SPACE_CACHE)) {
3452 * don't bother trying to write stuff out _if_
3453 * a) we're not cached,
3454 * b) we're with nospace_cache mount option.
3456 dcs = BTRFS_DC_WRITTEN;
3457 spin_unlock(&block_group->lock);
3460 spin_unlock(&block_group->lock);
3463 * We hit an ENOSPC when setting up the cache in this transaction, just
3464 * skip doing the setup, we've already cleared the cache so we're safe.
3466 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3472 * Try to preallocate enough space based on how big the block group is.
3473 * Keep in mind this has to include any pinned space which could end up
3474 * taking up quite a bit since it's not folded into the other space
3477 num_pages = div_u64(block_group->key.offset, SZ_256M);
3482 num_pages *= PAGE_SIZE;
3484 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3488 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3489 num_pages, num_pages,
3492 * Our cache requires contiguous chunks so that we don't modify a bunch
3493 * of metadata or split extents when writing the cache out, which means
3494 * we can enospc if we are heavily fragmented in addition to just normal
3495 * out of space conditions. So if we hit this just skip setting up any
3496 * other block groups for this transaction, maybe we'll unpin enough
3497 * space the next time around.
3500 dcs = BTRFS_DC_SETUP;
3501 else if (ret == -ENOSPC)
3502 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3503 btrfs_free_reserved_data_space(inode, 0, num_pages);
3508 btrfs_release_path(path);
3510 spin_lock(&block_group->lock);
3511 if (!ret && dcs == BTRFS_DC_SETUP)
3512 block_group->cache_generation = trans->transid;
3513 block_group->disk_cache_state = dcs;
3514 spin_unlock(&block_group->lock);
3519 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3520 struct btrfs_root *root)
3522 struct btrfs_block_group_cache *cache, *tmp;
3523 struct btrfs_transaction *cur_trans = trans->transaction;
3524 struct btrfs_path *path;
3526 if (list_empty(&cur_trans->dirty_bgs) ||
3527 !btrfs_test_opt(root, SPACE_CACHE))
3530 path = btrfs_alloc_path();
3534 /* Could add new block groups, use _safe just in case */
3535 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3537 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3538 cache_save_setup(cache, trans, path);
3541 btrfs_free_path(path);
3546 * transaction commit does final block group cache writeback during a
3547 * critical section where nothing is allowed to change the FS. This is
3548 * required in order for the cache to actually match the block group,
3549 * but can introduce a lot of latency into the commit.
3551 * So, btrfs_start_dirty_block_groups is here to kick off block group
3552 * cache IO. There's a chance we'll have to redo some of it if the
3553 * block group changes again during the commit, but it greatly reduces
3554 * the commit latency by getting rid of the easy block groups while
3555 * we're still allowing others to join the commit.
3557 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3558 struct btrfs_root *root)
3560 struct btrfs_block_group_cache *cache;
3561 struct btrfs_transaction *cur_trans = trans->transaction;
3564 struct btrfs_path *path = NULL;
3566 struct list_head *io = &cur_trans->io_bgs;
3567 int num_started = 0;
3570 spin_lock(&cur_trans->dirty_bgs_lock);
3571 if (list_empty(&cur_trans->dirty_bgs)) {
3572 spin_unlock(&cur_trans->dirty_bgs_lock);
3575 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3576 spin_unlock(&cur_trans->dirty_bgs_lock);
3580 * make sure all the block groups on our dirty list actually
3583 btrfs_create_pending_block_groups(trans, root);
3586 path = btrfs_alloc_path();
3592 * cache_write_mutex is here only to save us from balance or automatic
3593 * removal of empty block groups deleting this block group while we are
3594 * writing out the cache
3596 mutex_lock(&trans->transaction->cache_write_mutex);
3597 while (!list_empty(&dirty)) {
3598 cache = list_first_entry(&dirty,
3599 struct btrfs_block_group_cache,
3602 * this can happen if something re-dirties a block
3603 * group that is already under IO. Just wait for it to
3604 * finish and then do it all again
3606 if (!list_empty(&cache->io_list)) {
3607 list_del_init(&cache->io_list);
3608 btrfs_wait_cache_io(root, trans, cache,
3609 &cache->io_ctl, path,
3610 cache->key.objectid);
3611 btrfs_put_block_group(cache);
3616 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3617 * if it should update the cache_state. Don't delete
3618 * until after we wait.
3620 * Since we're not running in the commit critical section
3621 * we need the dirty_bgs_lock to protect from update_block_group
3623 spin_lock(&cur_trans->dirty_bgs_lock);
3624 list_del_init(&cache->dirty_list);
3625 spin_unlock(&cur_trans->dirty_bgs_lock);
3629 cache_save_setup(cache, trans, path);
3631 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3632 cache->io_ctl.inode = NULL;
3633 ret = btrfs_write_out_cache(root, trans, cache, path);
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, root, path, cache);
3654 * Our block group might still be attached to the list
3655 * of new block groups in the transaction handle of some
3656 * other task (struct btrfs_trans_handle->new_bgs). This
3657 * means its block group item isn't yet in the extent
3658 * tree. If this happens ignore the error, as we will
3659 * try again later in the critical section of the
3660 * transaction commit.
3662 if (ret == -ENOENT) {
3664 spin_lock(&cur_trans->dirty_bgs_lock);
3665 if (list_empty(&cache->dirty_list)) {
3666 list_add_tail(&cache->dirty_list,
3667 &cur_trans->dirty_bgs);
3668 btrfs_get_block_group(cache);
3670 spin_unlock(&cur_trans->dirty_bgs_lock);
3672 btrfs_abort_transaction(trans, root, ret);
3676 /* if its not on the io list, we need to put the block group */
3678 btrfs_put_block_group(cache);
3684 * Avoid blocking other tasks for too long. It might even save
3685 * us from writing caches for block groups that are going to be
3688 mutex_unlock(&trans->transaction->cache_write_mutex);
3689 mutex_lock(&trans->transaction->cache_write_mutex);
3691 mutex_unlock(&trans->transaction->cache_write_mutex);
3694 * go through delayed refs for all the stuff we've just kicked off
3695 * and then loop back (just once)
3697 ret = btrfs_run_delayed_refs(trans, root, 0);
3698 if (!ret && loops == 0) {
3700 spin_lock(&cur_trans->dirty_bgs_lock);
3701 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3703 * dirty_bgs_lock protects us from concurrent block group
3704 * deletes too (not just cache_write_mutex).
3706 if (!list_empty(&dirty)) {
3707 spin_unlock(&cur_trans->dirty_bgs_lock);
3710 spin_unlock(&cur_trans->dirty_bgs_lock);
3713 btrfs_free_path(path);
3717 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3718 struct btrfs_root *root)
3720 struct btrfs_block_group_cache *cache;
3721 struct btrfs_transaction *cur_trans = trans->transaction;
3724 struct btrfs_path *path;
3725 struct list_head *io = &cur_trans->io_bgs;
3726 int num_started = 0;
3728 path = btrfs_alloc_path();
3733 * Even though we are in the critical section of the transaction commit,
3734 * we can still have concurrent tasks adding elements to this
3735 * transaction's list of dirty block groups. These tasks correspond to
3736 * endio free space workers started when writeback finishes for a
3737 * space cache, which run inode.c:btrfs_finish_ordered_io(), and can
3738 * allocate new block groups as a result of COWing nodes of the root
3739 * tree when updating the free space inode. The writeback for the space
3740 * caches is triggered by an earlier call to
3741 * btrfs_start_dirty_block_groups() and iterations of the following
3743 * Also we want to do the cache_save_setup first and then run the
3744 * delayed refs to make sure we have the best chance at doing this all
3747 spin_lock(&cur_trans->dirty_bgs_lock);
3748 while (!list_empty(&cur_trans->dirty_bgs)) {
3749 cache = list_first_entry(&cur_trans->dirty_bgs,
3750 struct btrfs_block_group_cache,
3754 * this can happen if cache_save_setup re-dirties a block
3755 * group that is already under IO. Just wait for it to
3756 * finish and then do it all again
3758 if (!list_empty(&cache->io_list)) {
3759 spin_unlock(&cur_trans->dirty_bgs_lock);
3760 list_del_init(&cache->io_list);
3761 btrfs_wait_cache_io(root, trans, cache,
3762 &cache->io_ctl, path,
3763 cache->key.objectid);
3764 btrfs_put_block_group(cache);
3765 spin_lock(&cur_trans->dirty_bgs_lock);
3769 * don't remove from the dirty list until after we've waited
3772 list_del_init(&cache->dirty_list);
3773 spin_unlock(&cur_trans->dirty_bgs_lock);
3776 cache_save_setup(cache, trans, path);
3779 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3781 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3782 cache->io_ctl.inode = NULL;
3783 ret = btrfs_write_out_cache(root, trans, cache, path);
3784 if (ret == 0 && cache->io_ctl.inode) {
3787 list_add_tail(&cache->io_list, io);
3790 * if we failed to write the cache, the
3791 * generation will be bad and life goes on
3797 ret = write_one_cache_group(trans, root, path, cache);
3799 * One of the free space endio workers might have
3800 * created a new block group while updating a free space
3801 * cache's inode (at inode.c:btrfs_finish_ordered_io())
3802 * and hasn't released its transaction handle yet, in
3803 * which case the new block group is still attached to
3804 * its transaction handle and its creation has not
3805 * finished yet (no block group item in the extent tree
3806 * yet, etc). If this is the case, wait for all free
3807 * space endio workers to finish and retry. This is a
3808 * a very rare case so no need for a more efficient and
3811 if (ret == -ENOENT) {
3812 wait_event(cur_trans->writer_wait,
3813 atomic_read(&cur_trans->num_writers) == 1);
3814 ret = write_one_cache_group(trans, root, path,
3818 btrfs_abort_transaction(trans, root, ret);
3821 /* if its not on the io list, we need to put the block group */
3823 btrfs_put_block_group(cache);
3824 spin_lock(&cur_trans->dirty_bgs_lock);
3826 spin_unlock(&cur_trans->dirty_bgs_lock);
3828 while (!list_empty(io)) {
3829 cache = list_first_entry(io, struct btrfs_block_group_cache,
3831 list_del_init(&cache->io_list);
3832 btrfs_wait_cache_io(root, trans, cache,
3833 &cache->io_ctl, path, cache->key.objectid);
3834 btrfs_put_block_group(cache);
3837 btrfs_free_path(path);
3841 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3843 struct btrfs_block_group_cache *block_group;
3846 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3847 if (!block_group || block_group->ro)
3850 btrfs_put_block_group(block_group);
3854 bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3856 struct btrfs_block_group_cache *bg;
3859 bg = btrfs_lookup_block_group(fs_info, bytenr);
3863 spin_lock(&bg->lock);
3867 atomic_inc(&bg->nocow_writers);
3868 spin_unlock(&bg->lock);
3870 /* no put on block group, done by btrfs_dec_nocow_writers */
3872 btrfs_put_block_group(bg);
3878 void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr)
3880 struct btrfs_block_group_cache *bg;
3882 bg = btrfs_lookup_block_group(fs_info, bytenr);
3884 if (atomic_dec_and_test(&bg->nocow_writers))
3885 wake_up_atomic_t(&bg->nocow_writers);
3887 * Once for our lookup and once for the lookup done by a previous call
3888 * to btrfs_inc_nocow_writers()
3890 btrfs_put_block_group(bg);
3891 btrfs_put_block_group(bg);
3894 static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a)
3900 void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg)
3902 wait_on_atomic_t(&bg->nocow_writers,
3903 btrfs_wait_nocow_writers_atomic_t,
3904 TASK_UNINTERRUPTIBLE);
3907 static const char *alloc_name(u64 flags)
3910 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3912 case BTRFS_BLOCK_GROUP_METADATA:
3914 case BTRFS_BLOCK_GROUP_DATA:
3916 case BTRFS_BLOCK_GROUP_SYSTEM:
3920 return "invalid-combination";
3924 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3925 u64 total_bytes, u64 bytes_used,
3927 struct btrfs_space_info **space_info)
3929 struct btrfs_space_info *found;
3934 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3935 BTRFS_BLOCK_GROUP_RAID10))
3940 found = __find_space_info(info, flags);
3942 spin_lock(&found->lock);
3943 found->total_bytes += total_bytes;
3944 found->disk_total += total_bytes * factor;
3945 found->bytes_used += bytes_used;
3946 found->disk_used += bytes_used * factor;
3947 found->bytes_readonly += bytes_readonly;
3948 if (total_bytes > 0)
3950 space_info_add_new_bytes(info, found, total_bytes -
3951 bytes_used - bytes_readonly);
3952 spin_unlock(&found->lock);
3953 *space_info = found;
3956 found = kzalloc(sizeof(*found), GFP_NOFS);
3960 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3966 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3967 INIT_LIST_HEAD(&found->block_groups[i]);
3968 init_rwsem(&found->groups_sem);
3969 spin_lock_init(&found->lock);
3970 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3971 found->total_bytes = total_bytes;
3972 found->disk_total = total_bytes * factor;
3973 found->bytes_used = bytes_used;
3974 found->disk_used = bytes_used * factor;
3975 found->bytes_pinned = 0;
3976 found->bytes_reserved = 0;
3977 found->bytes_readonly = bytes_readonly;
3978 found->bytes_may_use = 0;
3980 found->max_extent_size = 0;
3981 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3982 found->chunk_alloc = 0;
3984 init_waitqueue_head(&found->wait);
3985 INIT_LIST_HEAD(&found->ro_bgs);
3986 INIT_LIST_HEAD(&found->tickets);
3987 INIT_LIST_HEAD(&found->priority_tickets);
3989 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3990 info->space_info_kobj, "%s",
3991 alloc_name(found->flags));
3997 *space_info = found;
3998 list_add_rcu(&found->list, &info->space_info);
3999 if (flags & BTRFS_BLOCK_GROUP_DATA)
4000 info->data_sinfo = found;
4005 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
4007 u64 extra_flags = chunk_to_extended(flags) &
4008 BTRFS_EXTENDED_PROFILE_MASK;
4010 write_seqlock(&fs_info->profiles_lock);
4011 if (flags & BTRFS_BLOCK_GROUP_DATA)
4012 fs_info->avail_data_alloc_bits |= extra_flags;
4013 if (flags & BTRFS_BLOCK_GROUP_METADATA)
4014 fs_info->avail_metadata_alloc_bits |= extra_flags;
4015 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4016 fs_info->avail_system_alloc_bits |= extra_flags;
4017 write_sequnlock(&fs_info->profiles_lock);
4021 * returns target flags in extended format or 0 if restripe for this
4022 * chunk_type is not in progress
4024 * should be called with either volume_mutex or balance_lock held
4026 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
4028 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
4034 if (flags & BTRFS_BLOCK_GROUP_DATA &&
4035 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4036 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
4037 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
4038 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4039 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
4040 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
4041 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
4042 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
4049 * @flags: available profiles in extended format (see ctree.h)
4051 * Returns reduced profile in chunk format. If profile changing is in
4052 * progress (either running or paused) picks the target profile (if it's
4053 * already available), otherwise falls back to plain reducing.
4055 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
4057 u64 num_devices = root->fs_info->fs_devices->rw_devices;
4063 * see if restripe for this chunk_type is in progress, if so
4064 * try to reduce to the target profile
4066 spin_lock(&root->fs_info->balance_lock);
4067 target = get_restripe_target(root->fs_info, flags);
4069 /* pick target profile only if it's already available */
4070 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
4071 spin_unlock(&root->fs_info->balance_lock);
4072 return extended_to_chunk(target);
4075 spin_unlock(&root->fs_info->balance_lock);
4077 /* First, mask out the RAID levels which aren't possible */
4078 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
4079 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
4080 allowed |= btrfs_raid_group[raid_type];
4084 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
4085 allowed = BTRFS_BLOCK_GROUP_RAID6;
4086 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
4087 allowed = BTRFS_BLOCK_GROUP_RAID5;
4088 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
4089 allowed = BTRFS_BLOCK_GROUP_RAID10;
4090 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
4091 allowed = BTRFS_BLOCK_GROUP_RAID1;
4092 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
4093 allowed = BTRFS_BLOCK_GROUP_RAID0;
4095 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
4097 return extended_to_chunk(flags | allowed);
4100 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
4107 seq = read_seqbegin(&root->fs_info->profiles_lock);
4109 if (flags & BTRFS_BLOCK_GROUP_DATA)
4110 flags |= root->fs_info->avail_data_alloc_bits;
4111 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
4112 flags |= root->fs_info->avail_system_alloc_bits;
4113 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
4114 flags |= root->fs_info->avail_metadata_alloc_bits;
4115 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
4117 return btrfs_reduce_alloc_profile(root, flags);
4120 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
4126 flags = BTRFS_BLOCK_GROUP_DATA;
4127 else if (root == root->fs_info->chunk_root)
4128 flags = BTRFS_BLOCK_GROUP_SYSTEM;
4130 flags = BTRFS_BLOCK_GROUP_METADATA;
4132 ret = get_alloc_profile(root, flags);
4136 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4138 struct btrfs_space_info *data_sinfo;
4139 struct btrfs_root *root = BTRFS_I(inode)->root;
4140 struct btrfs_fs_info *fs_info = root->fs_info;
4143 int need_commit = 2;
4144 int have_pinned_space;
4146 /* make sure bytes are sectorsize aligned */
4147 bytes = ALIGN(bytes, root->sectorsize);
4149 if (btrfs_is_free_space_inode(inode)) {
4151 ASSERT(current->journal_info);
4154 data_sinfo = fs_info->data_sinfo;
4159 /* make sure we have enough space to handle the data first */
4160 spin_lock(&data_sinfo->lock);
4161 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4162 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4163 data_sinfo->bytes_may_use;
4165 if (used + bytes > data_sinfo->total_bytes) {
4166 struct btrfs_trans_handle *trans;
4169 * if we don't have enough free bytes in this space then we need
4170 * to alloc a new chunk.
4172 if (!data_sinfo->full) {
4175 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4176 spin_unlock(&data_sinfo->lock);
4178 alloc_target = btrfs_get_alloc_profile(root, 1);
4180 * It is ugly that we don't call nolock join
4181 * transaction for the free space inode case here.
4182 * But it is safe because we only do the data space
4183 * reservation for the free space cache in the
4184 * transaction context, the common join transaction
4185 * just increase the counter of the current transaction
4186 * handler, doesn't try to acquire the trans_lock of
4189 trans = btrfs_join_transaction(root);
4191 return PTR_ERR(trans);
4193 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4195 CHUNK_ALLOC_NO_FORCE);
4196 btrfs_end_transaction(trans, root);
4201 have_pinned_space = 1;
4207 data_sinfo = fs_info->data_sinfo;
4213 * If we don't have enough pinned space to deal with this
4214 * allocation, and no removed chunk in current transaction,
4215 * don't bother committing the transaction.
4217 have_pinned_space = percpu_counter_compare(
4218 &data_sinfo->total_bytes_pinned,
4219 used + bytes - data_sinfo->total_bytes);
4220 spin_unlock(&data_sinfo->lock);
4222 /* commit the current transaction and try again */
4225 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4228 if (need_commit > 0) {
4229 btrfs_start_delalloc_roots(fs_info, 0, -1);
4230 btrfs_wait_ordered_roots(fs_info, -1, 0, (u64)-1);
4233 trans = btrfs_join_transaction(root);
4235 return PTR_ERR(trans);
4236 if (have_pinned_space >= 0 ||
4237 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4238 &trans->transaction->flags) ||
4240 ret = btrfs_commit_transaction(trans, root);
4244 * The cleaner kthread might still be doing iput
4245 * operations. Wait for it to finish so that
4246 * more space is released.
4248 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4249 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4252 btrfs_end_transaction(trans, root);
4256 trace_btrfs_space_reservation(root->fs_info,
4257 "space_info:enospc",
4258 data_sinfo->flags, bytes, 1);
4261 data_sinfo->bytes_may_use += bytes;
4262 trace_btrfs_space_reservation(root->fs_info, "space_info",
4263 data_sinfo->flags, bytes, 1);
4264 spin_unlock(&data_sinfo->lock);
4270 * New check_data_free_space() with ability for precious data reservation
4271 * Will replace old btrfs_check_data_free_space(), but for patch split,
4272 * add a new function first and then replace it.
4274 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4276 struct btrfs_root *root = BTRFS_I(inode)->root;
4279 /* align the range */
4280 len = round_up(start + len, root->sectorsize) -
4281 round_down(start, root->sectorsize);
4282 start = round_down(start, root->sectorsize);
4284 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4289 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4291 * TODO: Find a good method to avoid reserve data space for NOCOW
4292 * range, but don't impact performance on quota disable case.
4294 ret = btrfs_qgroup_reserve_data(inode, start, len);
4299 * Called if we need to clear a data reservation for this inode
4300 * Normally in a error case.
4302 * This one will *NOT* use accurate qgroup reserved space API, just for case
4303 * which we can't sleep and is sure it won't affect qgroup reserved space.
4304 * Like clear_bit_hook().
4306 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4309 struct btrfs_root *root = BTRFS_I(inode)->root;
4310 struct btrfs_space_info *data_sinfo;
4312 /* Make sure the range is aligned to sectorsize */
4313 len = round_up(start + len, root->sectorsize) -
4314 round_down(start, root->sectorsize);
4315 start = round_down(start, root->sectorsize);
4317 data_sinfo = root->fs_info->data_sinfo;
4318 spin_lock(&data_sinfo->lock);
4319 if (WARN_ON(data_sinfo->bytes_may_use < len))
4320 data_sinfo->bytes_may_use = 0;
4322 data_sinfo->bytes_may_use -= len;
4323 trace_btrfs_space_reservation(root->fs_info, "space_info",
4324 data_sinfo->flags, len, 0);
4325 spin_unlock(&data_sinfo->lock);
4329 * Called if we need to clear a data reservation for this inode
4330 * Normally in a error case.
4332 * This one will handle the per-inode data rsv map for accurate reserved
4335 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4337 btrfs_free_reserved_data_space_noquota(inode, start, len);
4338 btrfs_qgroup_free_data(inode, start, len);
4341 static void force_metadata_allocation(struct btrfs_fs_info *info)
4343 struct list_head *head = &info->space_info;
4344 struct btrfs_space_info *found;
4347 list_for_each_entry_rcu(found, head, list) {
4348 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4349 found->force_alloc = CHUNK_ALLOC_FORCE;
4354 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4356 return (global->size << 1);
4359 static int should_alloc_chunk(struct btrfs_root *root,
4360 struct btrfs_space_info *sinfo, int force)
4362 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4363 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4364 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4367 if (force == CHUNK_ALLOC_FORCE)
4371 * We need to take into account the global rsv because for all intents
4372 * and purposes it's used space. Don't worry about locking the
4373 * global_rsv, it doesn't change except when the transaction commits.
4375 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4376 num_allocated += calc_global_rsv_need_space(global_rsv);
4379 * in limited mode, we want to have some free space up to
4380 * about 1% of the FS size.
4382 if (force == CHUNK_ALLOC_LIMITED) {
4383 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4384 thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1));
4386 if (num_bytes - num_allocated < thresh)
4390 if (num_allocated + SZ_2M < div_factor(num_bytes, 8))
4395 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4399 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4400 BTRFS_BLOCK_GROUP_RAID0 |
4401 BTRFS_BLOCK_GROUP_RAID5 |
4402 BTRFS_BLOCK_GROUP_RAID6))
4403 num_dev = root->fs_info->fs_devices->rw_devices;
4404 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4407 num_dev = 1; /* DUP or single */
4413 * If @is_allocation is true, reserve space in the system space info necessary
4414 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4417 void check_system_chunk(struct btrfs_trans_handle *trans,
4418 struct btrfs_root *root,
4421 struct btrfs_space_info *info;
4428 * Needed because we can end up allocating a system chunk and for an
4429 * atomic and race free space reservation in the chunk block reserve.
4431 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4433 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4434 spin_lock(&info->lock);
4435 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4436 info->bytes_reserved - info->bytes_readonly -
4437 info->bytes_may_use;
4438 spin_unlock(&info->lock);
4440 num_devs = get_profile_num_devs(root, type);
4442 /* num_devs device items to update and 1 chunk item to add or remove */
4443 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4444 btrfs_calc_trans_metadata_size(root, 1);
4446 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4447 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4448 left, thresh, type);
4449 dump_space_info(info, 0, 0);
4452 if (left < thresh) {
4455 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4457 * Ignore failure to create system chunk. We might end up not
4458 * needing it, as we might not need to COW all nodes/leafs from
4459 * the paths we visit in the chunk tree (they were already COWed
4460 * or created in the current transaction for example).
4462 ret = btrfs_alloc_chunk(trans, root, flags);
4466 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4467 &root->fs_info->chunk_block_rsv,
4468 thresh, BTRFS_RESERVE_NO_FLUSH);
4470 trans->chunk_bytes_reserved += thresh;
4474 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4475 struct btrfs_root *extent_root, u64 flags, int force)
4477 struct btrfs_space_info *space_info;
4478 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4479 int wait_for_alloc = 0;
4482 /* Don't re-enter if we're already allocating a chunk */
4483 if (trans->allocating_chunk)
4486 space_info = __find_space_info(extent_root->fs_info, flags);
4488 ret = update_space_info(extent_root->fs_info, flags,
4489 0, 0, 0, &space_info);
4490 BUG_ON(ret); /* -ENOMEM */
4492 BUG_ON(!space_info); /* Logic error */
4495 spin_lock(&space_info->lock);
4496 if (force < space_info->force_alloc)
4497 force = space_info->force_alloc;
4498 if (space_info->full) {
4499 if (should_alloc_chunk(extent_root, space_info, force))
4503 spin_unlock(&space_info->lock);
4507 if (!should_alloc_chunk(extent_root, space_info, force)) {
4508 spin_unlock(&space_info->lock);
4510 } else if (space_info->chunk_alloc) {
4513 space_info->chunk_alloc = 1;
4516 spin_unlock(&space_info->lock);
4518 mutex_lock(&fs_info->chunk_mutex);
4521 * The chunk_mutex is held throughout the entirety of a chunk
4522 * allocation, so once we've acquired the chunk_mutex we know that the
4523 * other guy is done and we need to recheck and see if we should
4526 if (wait_for_alloc) {
4527 mutex_unlock(&fs_info->chunk_mutex);
4532 trans->allocating_chunk = true;
4535 * If we have mixed data/metadata chunks we want to make sure we keep
4536 * allocating mixed chunks instead of individual chunks.
4538 if (btrfs_mixed_space_info(space_info))
4539 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4542 * if we're doing a data chunk, go ahead and make sure that
4543 * we keep a reasonable number of metadata chunks allocated in the
4546 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4547 fs_info->data_chunk_allocations++;
4548 if (!(fs_info->data_chunk_allocations %
4549 fs_info->metadata_ratio))
4550 force_metadata_allocation(fs_info);
4554 * Check if we have enough space in SYSTEM chunk because we may need
4555 * to update devices.
4557 check_system_chunk(trans, extent_root, flags);
4559 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4560 trans->allocating_chunk = false;
4562 spin_lock(&space_info->lock);
4563 if (ret < 0 && ret != -ENOSPC)
4566 space_info->full = 1;
4570 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4572 space_info->chunk_alloc = 0;
4573 spin_unlock(&space_info->lock);
4574 mutex_unlock(&fs_info->chunk_mutex);
4576 * When we allocate a new chunk we reserve space in the chunk block
4577 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4578 * add new nodes/leafs to it if we end up needing to do it when
4579 * inserting the chunk item and updating device items as part of the
4580 * second phase of chunk allocation, performed by
4581 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4582 * large number of new block groups to create in our transaction
4583 * handle's new_bgs list to avoid exhausting the chunk block reserve
4584 * in extreme cases - like having a single transaction create many new
4585 * block groups when starting to write out the free space caches of all
4586 * the block groups that were made dirty during the lifetime of the
4589 if (trans->can_flush_pending_bgs &&
4590 trans->chunk_bytes_reserved >= (u64)SZ_2M) {
4591 btrfs_create_pending_block_groups(trans, trans->root);
4592 btrfs_trans_release_chunk_metadata(trans);
4597 static int can_overcommit(struct btrfs_root *root,
4598 struct btrfs_space_info *space_info, u64 bytes,
4599 enum btrfs_reserve_flush_enum flush)
4601 struct btrfs_block_rsv *global_rsv;
4607 /* Don't overcommit when in mixed mode. */
4608 if (space_info->flags & BTRFS_BLOCK_GROUP_DATA)
4611 BUG_ON(root->fs_info == NULL);
4612 global_rsv = &root->fs_info->global_block_rsv;
4613 profile = btrfs_get_alloc_profile(root, 0);
4614 used = space_info->bytes_used + space_info->bytes_reserved +
4615 space_info->bytes_pinned + space_info->bytes_readonly;
4618 * We only want to allow over committing if we have lots of actual space
4619 * free, but if we don't have enough space to handle the global reserve
4620 * space then we could end up having a real enospc problem when trying
4621 * to allocate a chunk or some other such important allocation.
4623 spin_lock(&global_rsv->lock);
4624 space_size = calc_global_rsv_need_space(global_rsv);
4625 spin_unlock(&global_rsv->lock);
4626 if (used + space_size >= space_info->total_bytes)
4629 used += space_info->bytes_may_use;
4631 spin_lock(&root->fs_info->free_chunk_lock);
4632 avail = root->fs_info->free_chunk_space;
4633 spin_unlock(&root->fs_info->free_chunk_lock);
4636 * If we have dup, raid1 or raid10 then only half of the free
4637 * space is actually useable. For raid56, the space info used
4638 * doesn't include the parity drive, so we don't have to
4641 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4642 BTRFS_BLOCK_GROUP_RAID1 |
4643 BTRFS_BLOCK_GROUP_RAID10))
4647 * If we aren't flushing all things, let us overcommit up to
4648 * 1/2th of the space. If we can flush, don't let us overcommit
4649 * too much, let it overcommit up to 1/8 of the space.
4651 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4656 if (used + bytes < space_info->total_bytes + avail)
4661 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4662 unsigned long nr_pages, int nr_items)
4664 struct super_block *sb = root->fs_info->sb;
4666 if (down_read_trylock(&sb->s_umount)) {
4667 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4668 up_read(&sb->s_umount);
4671 * We needn't worry the filesystem going from r/w to r/o though
4672 * we don't acquire ->s_umount mutex, because the filesystem
4673 * should guarantee the delalloc inodes list be empty after
4674 * the filesystem is readonly(all dirty pages are written to
4677 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4678 if (!current->journal_info)
4679 btrfs_wait_ordered_roots(root->fs_info, nr_items,
4684 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4689 bytes = btrfs_calc_trans_metadata_size(root, 1);
4690 nr = (int)div64_u64(to_reclaim, bytes);
4696 #define EXTENT_SIZE_PER_ITEM SZ_256K
4699 * shrink metadata reservation for delalloc
4701 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4704 struct btrfs_block_rsv *block_rsv;
4705 struct btrfs_space_info *space_info;
4706 struct btrfs_trans_handle *trans;
4710 unsigned long nr_pages;
4713 enum btrfs_reserve_flush_enum flush;
4715 /* Calc the number of the pages we need flush for space reservation */
4716 items = calc_reclaim_items_nr(root, to_reclaim);
4717 to_reclaim = (u64)items * EXTENT_SIZE_PER_ITEM;
4719 trans = (struct btrfs_trans_handle *)current->journal_info;
4720 block_rsv = &root->fs_info->delalloc_block_rsv;
4721 space_info = block_rsv->space_info;
4723 delalloc_bytes = percpu_counter_sum_positive(
4724 &root->fs_info->delalloc_bytes);
4725 if (delalloc_bytes == 0) {
4729 btrfs_wait_ordered_roots(root->fs_info, items,
4735 while (delalloc_bytes && loops < 3) {
4736 max_reclaim = min(delalloc_bytes, to_reclaim);
4737 nr_pages = max_reclaim >> PAGE_SHIFT;
4738 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4740 * We need to wait for the async pages to actually start before
4743 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4747 if (max_reclaim <= nr_pages)
4750 max_reclaim -= nr_pages;
4752 wait_event(root->fs_info->async_submit_wait,
4753 atomic_read(&root->fs_info->async_delalloc_pages) <=
4757 flush = BTRFS_RESERVE_FLUSH_ALL;
4759 flush = BTRFS_RESERVE_NO_FLUSH;
4760 spin_lock(&space_info->lock);
4761 if (can_overcommit(root, space_info, orig, flush)) {
4762 spin_unlock(&space_info->lock);
4765 if (list_empty(&space_info->tickets) &&
4766 list_empty(&space_info->priority_tickets)) {
4767 spin_unlock(&space_info->lock);
4770 spin_unlock(&space_info->lock);
4773 if (wait_ordered && !trans) {
4774 btrfs_wait_ordered_roots(root->fs_info, items,
4777 time_left = schedule_timeout_killable(1);
4781 delalloc_bytes = percpu_counter_sum_positive(
4782 &root->fs_info->delalloc_bytes);
4787 * maybe_commit_transaction - possibly commit the transaction if its ok to
4788 * @root - the root we're allocating for
4789 * @bytes - the number of bytes we want to reserve
4790 * @force - force the commit
4792 * This will check to make sure that committing the transaction will actually
4793 * get us somewhere and then commit the transaction if it does. Otherwise it
4794 * will return -ENOSPC.
4796 static int may_commit_transaction(struct btrfs_root *root,
4797 struct btrfs_space_info *space_info,
4798 u64 bytes, int force)
4800 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4801 struct btrfs_trans_handle *trans;
4803 trans = (struct btrfs_trans_handle *)current->journal_info;
4810 /* See if there is enough pinned space to make this reservation */
4811 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4816 * See if there is some space in the delayed insertion reservation for
4819 if (space_info != delayed_rsv->space_info)
4822 spin_lock(&delayed_rsv->lock);
4823 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4824 bytes - delayed_rsv->size) >= 0) {
4825 spin_unlock(&delayed_rsv->lock);
4828 spin_unlock(&delayed_rsv->lock);
4831 trans = btrfs_join_transaction(root);
4835 return btrfs_commit_transaction(trans, root);
4838 struct reserve_ticket {
4841 struct list_head list;
4842 wait_queue_head_t wait;
4845 static int flush_space(struct btrfs_root *root,
4846 struct btrfs_space_info *space_info, u64 num_bytes,
4847 u64 orig_bytes, int state)
4849 struct btrfs_trans_handle *trans;
4854 case FLUSH_DELAYED_ITEMS_NR:
4855 case FLUSH_DELAYED_ITEMS:
4856 if (state == FLUSH_DELAYED_ITEMS_NR)
4857 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4861 trans = btrfs_join_transaction(root);
4862 if (IS_ERR(trans)) {
4863 ret = PTR_ERR(trans);
4866 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4867 btrfs_end_transaction(trans, root);
4869 case FLUSH_DELALLOC:
4870 case FLUSH_DELALLOC_WAIT:
4871 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4872 state == FLUSH_DELALLOC_WAIT);
4875 trans = btrfs_join_transaction(root);
4876 if (IS_ERR(trans)) {
4877 ret = PTR_ERR(trans);
4880 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4881 btrfs_get_alloc_profile(root, 0),
4882 CHUNK_ALLOC_NO_FORCE);
4883 btrfs_end_transaction(trans, root);
4888 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4895 trace_btrfs_flush_space(root->fs_info, space_info->flags, num_bytes,
4896 orig_bytes, state, ret);
4901 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4902 struct btrfs_space_info *space_info)
4904 struct reserve_ticket *ticket;
4909 to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M);
4910 if (can_overcommit(root, space_info, to_reclaim,
4911 BTRFS_RESERVE_FLUSH_ALL))
4914 list_for_each_entry(ticket, &space_info->tickets, list)
4915 to_reclaim += ticket->bytes;
4916 list_for_each_entry(ticket, &space_info->priority_tickets, list)
4917 to_reclaim += ticket->bytes;
4921 used = space_info->bytes_used + space_info->bytes_reserved +
4922 space_info->bytes_pinned + space_info->bytes_readonly +
4923 space_info->bytes_may_use;
4924 if (can_overcommit(root, space_info, SZ_1M, BTRFS_RESERVE_FLUSH_ALL))
4925 expected = div_factor_fine(space_info->total_bytes, 95);
4927 expected = div_factor_fine(space_info->total_bytes, 90);
4929 if (used > expected)
4930 to_reclaim = used - expected;
4933 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4934 space_info->bytes_reserved);
4938 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4939 struct btrfs_root *root, u64 used)
4941 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4943 /* If we're just plain full then async reclaim just slows us down. */
4944 if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh)
4947 if (!btrfs_calc_reclaim_metadata_size(root, space_info))
4950 return (used >= thresh && !btrfs_fs_closing(root->fs_info) &&
4951 !test_bit(BTRFS_FS_STATE_REMOUNTING,
4952 &root->fs_info->fs_state));
4955 static void wake_all_tickets(struct list_head *head)
4957 struct reserve_ticket *ticket;
4959 while (!list_empty(head)) {
4960 ticket = list_first_entry(head, struct reserve_ticket, list);
4961 list_del_init(&ticket->list);
4962 ticket->error = -ENOSPC;
4963 wake_up(&ticket->wait);
4968 * This is for normal flushers, we can wait all goddamned day if we want to. We
4969 * will loop and continuously try to flush as long as we are making progress.
4970 * We count progress as clearing off tickets each time we have to loop.
4972 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4974 struct reserve_ticket *last_ticket = NULL;
4975 struct btrfs_fs_info *fs_info;
4976 struct btrfs_space_info *space_info;
4979 int commit_cycles = 0;
4981 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4982 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4984 spin_lock(&space_info->lock);
4985 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4988 space_info->flush = 0;
4989 spin_unlock(&space_info->lock);
4992 last_ticket = list_first_entry(&space_info->tickets,
4993 struct reserve_ticket, list);
4994 spin_unlock(&space_info->lock);
4996 flush_state = FLUSH_DELAYED_ITEMS_NR;
4998 struct reserve_ticket *ticket;
5001 ret = flush_space(fs_info->fs_root, space_info, to_reclaim,
5002 to_reclaim, flush_state);
5003 spin_lock(&space_info->lock);
5004 if (list_empty(&space_info->tickets)) {
5005 space_info->flush = 0;
5006 spin_unlock(&space_info->lock);
5009 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5011 ticket = list_first_entry(&space_info->tickets,
5012 struct reserve_ticket, list);
5013 if (last_ticket == ticket) {
5016 last_ticket = ticket;
5017 flush_state = FLUSH_DELAYED_ITEMS_NR;
5022 if (flush_state > COMMIT_TRANS) {
5024 if (commit_cycles > 2) {
5025 wake_all_tickets(&space_info->tickets);
5026 space_info->flush = 0;
5028 flush_state = FLUSH_DELAYED_ITEMS_NR;
5031 spin_unlock(&space_info->lock);
5032 } while (flush_state <= COMMIT_TRANS);
5035 void btrfs_init_async_reclaim_work(struct work_struct *work)
5037 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
5040 static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info,
5041 struct btrfs_space_info *space_info,
5042 struct reserve_ticket *ticket)
5045 int flush_state = FLUSH_DELAYED_ITEMS_NR;
5047 spin_lock(&space_info->lock);
5048 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
5051 spin_unlock(&space_info->lock);
5054 spin_unlock(&space_info->lock);
5057 flush_space(fs_info->fs_root, space_info, to_reclaim,
5058 to_reclaim, flush_state);
5060 spin_lock(&space_info->lock);
5061 if (ticket->bytes == 0) {
5062 spin_unlock(&space_info->lock);
5065 spin_unlock(&space_info->lock);
5068 * Priority flushers can't wait on delalloc without
5071 if (flush_state == FLUSH_DELALLOC ||
5072 flush_state == FLUSH_DELALLOC_WAIT)
5073 flush_state = ALLOC_CHUNK;
5074 } while (flush_state < COMMIT_TRANS);
5077 static int wait_reserve_ticket(struct btrfs_fs_info *fs_info,
5078 struct btrfs_space_info *space_info,
5079 struct reserve_ticket *ticket, u64 orig_bytes)
5085 spin_lock(&space_info->lock);
5086 while (ticket->bytes > 0 && ticket->error == 0) {
5087 ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE);
5092 spin_unlock(&space_info->lock);
5096 finish_wait(&ticket->wait, &wait);
5097 spin_lock(&space_info->lock);
5100 ret = ticket->error;
5101 if (!list_empty(&ticket->list))
5102 list_del_init(&ticket->list);
5103 if (ticket->bytes && ticket->bytes < orig_bytes) {
5104 u64 num_bytes = orig_bytes - ticket->bytes;
5105 space_info->bytes_may_use -= num_bytes;
5106 trace_btrfs_space_reservation(fs_info, "space_info",
5107 space_info->flags, num_bytes, 0);
5109 spin_unlock(&space_info->lock);
5115 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5116 * @root - the root we're allocating for
5117 * @space_info - the space info we want to allocate from
5118 * @orig_bytes - the number of bytes we want
5119 * @flush - whether or not we can flush to make our reservation
5121 * This will reserve orig_bytes number of bytes from the space info associated
5122 * with the block_rsv. If there is not enough space it will make an attempt to
5123 * flush out space to make room. It will do this by flushing delalloc if
5124 * possible or committing the transaction. If flush is 0 then no attempts to
5125 * regain reservations will be made and this will fail if there is not enough
5128 static int __reserve_metadata_bytes(struct btrfs_root *root,
5129 struct btrfs_space_info *space_info,
5131 enum btrfs_reserve_flush_enum flush)
5133 struct reserve_ticket ticket;
5138 spin_lock(&space_info->lock);
5140 used = space_info->bytes_used + space_info->bytes_reserved +
5141 space_info->bytes_pinned + space_info->bytes_readonly +
5142 space_info->bytes_may_use;
5145 * If we have enough space then hooray, make our reservation and carry
5146 * on. If not see if we can overcommit, and if we can, hooray carry on.
5147 * If not things get more complicated.
5149 if (used + orig_bytes <= space_info->total_bytes) {
5150 space_info->bytes_may_use += orig_bytes;
5151 trace_btrfs_space_reservation(root->fs_info, "space_info",
5152 space_info->flags, orig_bytes,
5155 } else if (can_overcommit(root, space_info, orig_bytes, flush)) {
5156 space_info->bytes_may_use += orig_bytes;
5157 trace_btrfs_space_reservation(root->fs_info, "space_info",
5158 space_info->flags, orig_bytes,
5164 * If we couldn't make a reservation then setup our reservation ticket
5165 * and kick the async worker if it's not already running.
5167 * If we are a priority flusher then we just need to add our ticket to
5168 * the list and we will do our own flushing further down.
5170 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
5171 ticket.bytes = orig_bytes;
5173 init_waitqueue_head(&ticket.wait);
5174 if (flush == BTRFS_RESERVE_FLUSH_ALL) {
5175 list_add_tail(&ticket.list, &space_info->tickets);
5176 if (!space_info->flush) {
5177 space_info->flush = 1;
5178 trace_btrfs_trigger_flush(root->fs_info,
5182 queue_work(system_unbound_wq,
5183 &root->fs_info->async_reclaim_work);
5186 list_add_tail(&ticket.list,
5187 &space_info->priority_tickets);
5189 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
5192 * We will do the space reservation dance during log replay,
5193 * which means we won't have fs_info->fs_root set, so don't do
5194 * the async reclaim as we will panic.
5196 if (!root->fs_info->log_root_recovering &&
5197 need_do_async_reclaim(space_info, root, used) &&
5198 !work_busy(&root->fs_info->async_reclaim_work)) {
5199 trace_btrfs_trigger_flush(root->fs_info,
5203 queue_work(system_unbound_wq,
5204 &root->fs_info->async_reclaim_work);
5207 spin_unlock(&space_info->lock);
5208 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
5211 if (flush == BTRFS_RESERVE_FLUSH_ALL)
5212 return wait_reserve_ticket(root->fs_info, space_info, &ticket,
5216 priority_reclaim_metadata_space(root->fs_info, space_info, &ticket);
5217 spin_lock(&space_info->lock);
5219 if (ticket.bytes < orig_bytes) {
5220 u64 num_bytes = orig_bytes - ticket.bytes;
5221 space_info->bytes_may_use -= num_bytes;
5222 trace_btrfs_space_reservation(root->fs_info,
5223 "space_info", space_info->flags,
5227 list_del_init(&ticket.list);
5230 spin_unlock(&space_info->lock);
5231 ASSERT(list_empty(&ticket.list));
5236 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
5237 * @root - the root we're allocating for
5238 * @block_rsv - the block_rsv we're allocating for
5239 * @orig_bytes - the number of bytes we want
5240 * @flush - whether or not we can flush to make our reservation
5242 * This will reserve orgi_bytes number of bytes from the space info associated
5243 * with the block_rsv. If there is not enough space it will make an attempt to
5244 * flush out space to make room. It will do this by flushing delalloc if
5245 * possible or committing the transaction. If flush is 0 then no attempts to
5246 * regain reservations will be made and this will fail if there is not enough
5249 static int reserve_metadata_bytes(struct btrfs_root *root,
5250 struct btrfs_block_rsv *block_rsv,
5252 enum btrfs_reserve_flush_enum flush)
5256 ret = __reserve_metadata_bytes(root, block_rsv->space_info, orig_bytes,
5258 if (ret == -ENOSPC &&
5259 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5260 struct btrfs_block_rsv *global_rsv =
5261 &root->fs_info->global_block_rsv;
5263 if (block_rsv != global_rsv &&
5264 !block_rsv_use_bytes(global_rsv, orig_bytes))
5268 trace_btrfs_space_reservation(root->fs_info,
5269 "space_info:enospc",
5270 block_rsv->space_info->flags,
5275 static struct btrfs_block_rsv *get_block_rsv(
5276 const struct btrfs_trans_handle *trans,
5277 const struct btrfs_root *root)
5279 struct btrfs_block_rsv *block_rsv = NULL;
5281 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5282 (root == root->fs_info->csum_root && trans->adding_csums) ||
5283 (root == root->fs_info->uuid_root))
5284 block_rsv = trans->block_rsv;
5287 block_rsv = root->block_rsv;
5290 block_rsv = &root->fs_info->empty_block_rsv;
5295 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5299 spin_lock(&block_rsv->lock);
5300 if (block_rsv->reserved >= num_bytes) {
5301 block_rsv->reserved -= num_bytes;
5302 if (block_rsv->reserved < block_rsv->size)
5303 block_rsv->full = 0;
5306 spin_unlock(&block_rsv->lock);
5310 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5311 u64 num_bytes, int update_size)
5313 spin_lock(&block_rsv->lock);
5314 block_rsv->reserved += num_bytes;
5316 block_rsv->size += num_bytes;
5317 else if (block_rsv->reserved >= block_rsv->size)
5318 block_rsv->full = 1;
5319 spin_unlock(&block_rsv->lock);
5322 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5323 struct btrfs_block_rsv *dest, u64 num_bytes,
5326 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5329 if (global_rsv->space_info != dest->space_info)
5332 spin_lock(&global_rsv->lock);
5333 min_bytes = div_factor(global_rsv->size, min_factor);
5334 if (global_rsv->reserved < min_bytes + num_bytes) {
5335 spin_unlock(&global_rsv->lock);
5338 global_rsv->reserved -= num_bytes;
5339 if (global_rsv->reserved < global_rsv->size)
5340 global_rsv->full = 0;
5341 spin_unlock(&global_rsv->lock);
5343 block_rsv_add_bytes(dest, num_bytes, 1);
5348 * This is for space we already have accounted in space_info->bytes_may_use, so
5349 * basically when we're returning space from block_rsv's.
5351 static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info,
5352 struct btrfs_space_info *space_info,
5355 struct reserve_ticket *ticket;
5356 struct list_head *head;
5358 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH;
5359 bool check_overcommit = false;
5361 spin_lock(&space_info->lock);
5362 head = &space_info->priority_tickets;
5365 * If we are over our limit then we need to check and see if we can
5366 * overcommit, and if we can't then we just need to free up our space
5367 * and not satisfy any requests.
5369 used = space_info->bytes_used + space_info->bytes_reserved +
5370 space_info->bytes_pinned + space_info->bytes_readonly +
5371 space_info->bytes_may_use;
5372 if (used - num_bytes >= space_info->total_bytes)
5373 check_overcommit = true;
5375 while (!list_empty(head) && num_bytes) {
5376 ticket = list_first_entry(head, struct reserve_ticket,
5379 * We use 0 bytes because this space is already reserved, so
5380 * adding the ticket space would be a double count.
5382 if (check_overcommit &&
5383 !can_overcommit(fs_info->extent_root, space_info, 0,
5386 if (num_bytes >= ticket->bytes) {
5387 list_del_init(&ticket->list);
5388 num_bytes -= ticket->bytes;
5390 wake_up(&ticket->wait);
5392 ticket->bytes -= num_bytes;
5397 if (num_bytes && head == &space_info->priority_tickets) {
5398 head = &space_info->tickets;
5399 flush = BTRFS_RESERVE_FLUSH_ALL;
5402 space_info->bytes_may_use -= num_bytes;
5403 trace_btrfs_space_reservation(fs_info, "space_info",
5404 space_info->flags, num_bytes, 0);
5405 spin_unlock(&space_info->lock);
5409 * This is for newly allocated space that isn't accounted in
5410 * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent
5411 * we use this helper.
5413 static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info,
5414 struct btrfs_space_info *space_info,
5417 struct reserve_ticket *ticket;
5418 struct list_head *head = &space_info->priority_tickets;
5421 while (!list_empty(head) && num_bytes) {
5422 ticket = list_first_entry(head, struct reserve_ticket,
5424 if (num_bytes >= ticket->bytes) {
5425 trace_btrfs_space_reservation(fs_info, "space_info",
5428 list_del_init(&ticket->list);
5429 num_bytes -= ticket->bytes;
5430 space_info->bytes_may_use += ticket->bytes;
5432 wake_up(&ticket->wait);
5434 trace_btrfs_space_reservation(fs_info, "space_info",
5437 space_info->bytes_may_use += num_bytes;
5438 ticket->bytes -= num_bytes;
5443 if (num_bytes && head == &space_info->priority_tickets) {
5444 head = &space_info->tickets;
5449 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5450 struct btrfs_block_rsv *block_rsv,
5451 struct btrfs_block_rsv *dest, u64 num_bytes)
5453 struct btrfs_space_info *space_info = block_rsv->space_info;
5455 spin_lock(&block_rsv->lock);
5456 if (num_bytes == (u64)-1)
5457 num_bytes = block_rsv->size;
5458 block_rsv->size -= num_bytes;
5459 if (block_rsv->reserved >= block_rsv->size) {
5460 num_bytes = block_rsv->reserved - block_rsv->size;
5461 block_rsv->reserved = block_rsv->size;
5462 block_rsv->full = 1;
5466 spin_unlock(&block_rsv->lock);
5468 if (num_bytes > 0) {
5470 spin_lock(&dest->lock);
5474 bytes_to_add = dest->size - dest->reserved;
5475 bytes_to_add = min(num_bytes, bytes_to_add);
5476 dest->reserved += bytes_to_add;
5477 if (dest->reserved >= dest->size)
5479 num_bytes -= bytes_to_add;
5481 spin_unlock(&dest->lock);
5484 space_info_add_old_bytes(fs_info, space_info,
5489 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src,
5490 struct btrfs_block_rsv *dst, u64 num_bytes,
5495 ret = block_rsv_use_bytes(src, num_bytes);
5499 block_rsv_add_bytes(dst, num_bytes, update_size);
5503 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5505 memset(rsv, 0, sizeof(*rsv));
5506 spin_lock_init(&rsv->lock);
5510 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5511 unsigned short type)
5513 struct btrfs_block_rsv *block_rsv;
5514 struct btrfs_fs_info *fs_info = root->fs_info;
5516 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5520 btrfs_init_block_rsv(block_rsv, type);
5521 block_rsv->space_info = __find_space_info(fs_info,
5522 BTRFS_BLOCK_GROUP_METADATA);
5526 void btrfs_free_block_rsv(struct btrfs_root *root,
5527 struct btrfs_block_rsv *rsv)
5531 btrfs_block_rsv_release(root, rsv, (u64)-1);
5535 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5540 int btrfs_block_rsv_add(struct btrfs_root *root,
5541 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5542 enum btrfs_reserve_flush_enum flush)
5549 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5551 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5558 int btrfs_block_rsv_check(struct btrfs_root *root,
5559 struct btrfs_block_rsv *block_rsv, int min_factor)
5567 spin_lock(&block_rsv->lock);
5568 num_bytes = div_factor(block_rsv->size, min_factor);
5569 if (block_rsv->reserved >= num_bytes)
5571 spin_unlock(&block_rsv->lock);
5576 int btrfs_block_rsv_refill(struct btrfs_root *root,
5577 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5578 enum btrfs_reserve_flush_enum flush)
5586 spin_lock(&block_rsv->lock);
5587 num_bytes = min_reserved;
5588 if (block_rsv->reserved >= num_bytes)
5591 num_bytes -= block_rsv->reserved;
5592 spin_unlock(&block_rsv->lock);
5597 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5599 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5606 void btrfs_block_rsv_release(struct btrfs_root *root,
5607 struct btrfs_block_rsv *block_rsv,
5610 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5611 if (global_rsv == block_rsv ||
5612 block_rsv->space_info != global_rsv->space_info)
5614 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5619 * helper to calculate size of global block reservation.
5620 * the desired value is sum of space used by extent tree,
5621 * checksum tree and root tree
5623 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5625 struct btrfs_space_info *sinfo;
5629 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5631 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5632 spin_lock(&sinfo->lock);
5633 data_used = sinfo->bytes_used;
5634 spin_unlock(&sinfo->lock);
5636 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5637 spin_lock(&sinfo->lock);
5638 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5640 meta_used = sinfo->bytes_used;
5641 spin_unlock(&sinfo->lock);
5643 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5645 num_bytes += div_u64(data_used + meta_used, 50);
5647 if (num_bytes * 3 > meta_used)
5648 num_bytes = div_u64(meta_used, 3);
5650 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5653 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5655 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5656 struct btrfs_space_info *sinfo = block_rsv->space_info;
5659 num_bytes = calc_global_metadata_size(fs_info);
5661 spin_lock(&sinfo->lock);
5662 spin_lock(&block_rsv->lock);
5664 block_rsv->size = min_t(u64, num_bytes, SZ_512M);
5666 if (block_rsv->reserved < block_rsv->size) {
5667 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5668 sinfo->bytes_reserved + sinfo->bytes_readonly +
5669 sinfo->bytes_may_use;
5670 if (sinfo->total_bytes > num_bytes) {
5671 num_bytes = sinfo->total_bytes - num_bytes;
5672 num_bytes = min(num_bytes,
5673 block_rsv->size - block_rsv->reserved);
5674 block_rsv->reserved += num_bytes;
5675 sinfo->bytes_may_use += num_bytes;
5676 trace_btrfs_space_reservation(fs_info, "space_info",
5677 sinfo->flags, num_bytes,
5680 } else if (block_rsv->reserved > block_rsv->size) {
5681 num_bytes = block_rsv->reserved - block_rsv->size;
5682 sinfo->bytes_may_use -= num_bytes;
5683 trace_btrfs_space_reservation(fs_info, "space_info",
5684 sinfo->flags, num_bytes, 0);
5685 block_rsv->reserved = block_rsv->size;
5688 if (block_rsv->reserved == block_rsv->size)
5689 block_rsv->full = 1;
5691 block_rsv->full = 0;
5693 spin_unlock(&block_rsv->lock);
5694 spin_unlock(&sinfo->lock);
5697 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5699 struct btrfs_space_info *space_info;
5701 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5702 fs_info->chunk_block_rsv.space_info = space_info;
5704 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5705 fs_info->global_block_rsv.space_info = space_info;
5706 fs_info->delalloc_block_rsv.space_info = space_info;
5707 fs_info->trans_block_rsv.space_info = space_info;
5708 fs_info->empty_block_rsv.space_info = space_info;
5709 fs_info->delayed_block_rsv.space_info = space_info;
5711 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5712 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5713 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5714 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5715 if (fs_info->quota_root)
5716 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5717 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5719 update_global_block_rsv(fs_info);
5722 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5724 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5726 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5727 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5728 WARN_ON(fs_info->trans_block_rsv.size > 0);
5729 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5730 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5731 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5732 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5733 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5736 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5737 struct btrfs_root *root)
5739 if (!trans->block_rsv)
5742 if (!trans->bytes_reserved)
5745 trace_btrfs_space_reservation(root->fs_info, "transaction",
5746 trans->transid, trans->bytes_reserved, 0);
5747 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5748 trans->bytes_reserved = 0;
5752 * To be called after all the new block groups attached to the transaction
5753 * handle have been created (btrfs_create_pending_block_groups()).
5755 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5757 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5759 if (!trans->chunk_bytes_reserved)
5762 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5764 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5765 trans->chunk_bytes_reserved);
5766 trans->chunk_bytes_reserved = 0;
5769 /* Can only return 0 or -ENOSPC */
5770 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5771 struct inode *inode)
5773 struct btrfs_root *root = BTRFS_I(inode)->root;
5774 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5775 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5778 * We need to hold space in order to delete our orphan item once we've
5779 * added it, so this takes the reservation so we can release it later
5780 * when we are truly done with the orphan item.
5782 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5783 trace_btrfs_space_reservation(root->fs_info, "orphan",
5784 btrfs_ino(inode), num_bytes, 1);
5785 return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
5788 void btrfs_orphan_release_metadata(struct inode *inode)
5790 struct btrfs_root *root = BTRFS_I(inode)->root;
5791 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5792 trace_btrfs_space_reservation(root->fs_info, "orphan",
5793 btrfs_ino(inode), num_bytes, 0);
5794 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5798 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5799 * root: the root of the parent directory
5800 * rsv: block reservation
5801 * items: the number of items that we need do reservation
5802 * qgroup_reserved: used to return the reserved size in qgroup
5804 * This function is used to reserve the space for snapshot/subvolume
5805 * creation and deletion. Those operations are different with the
5806 * common file/directory operations, they change two fs/file trees
5807 * and root tree, the number of items that the qgroup reserves is
5808 * different with the free space reservation. So we can not use
5809 * the space reservation mechanism in start_transaction().
5811 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5812 struct btrfs_block_rsv *rsv,
5814 u64 *qgroup_reserved,
5815 bool use_global_rsv)
5819 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5821 if (root->fs_info->quota_enabled) {
5822 /* One for parent inode, two for dir entries */
5823 num_bytes = 3 * root->nodesize;
5824 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5831 *qgroup_reserved = num_bytes;
5833 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5834 rsv->space_info = __find_space_info(root->fs_info,
5835 BTRFS_BLOCK_GROUP_METADATA);
5836 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5837 BTRFS_RESERVE_FLUSH_ALL);
5839 if (ret == -ENOSPC && use_global_rsv)
5840 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1);
5842 if (ret && *qgroup_reserved)
5843 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5848 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5849 struct btrfs_block_rsv *rsv,
5850 u64 qgroup_reserved)
5852 btrfs_block_rsv_release(root, rsv, (u64)-1);
5856 * drop_outstanding_extent - drop an outstanding extent
5857 * @inode: the inode we're dropping the extent for
5858 * @num_bytes: the number of bytes we're releasing.
5860 * This is called when we are freeing up an outstanding extent, either called
5861 * after an error or after an extent is written. This will return the number of
5862 * reserved extents that need to be freed. This must be called with
5863 * BTRFS_I(inode)->lock held.
5865 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5867 unsigned drop_inode_space = 0;
5868 unsigned dropped_extents = 0;
5869 unsigned num_extents = 0;
5871 num_extents = (unsigned)div64_u64(num_bytes +
5872 BTRFS_MAX_EXTENT_SIZE - 1,
5873 BTRFS_MAX_EXTENT_SIZE);
5874 ASSERT(num_extents);
5875 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5876 BTRFS_I(inode)->outstanding_extents -= num_extents;
5878 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5879 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5880 &BTRFS_I(inode)->runtime_flags))
5881 drop_inode_space = 1;
5884 * If we have more or the same amount of outstanding extents than we have
5885 * reserved then we need to leave the reserved extents count alone.
5887 if (BTRFS_I(inode)->outstanding_extents >=
5888 BTRFS_I(inode)->reserved_extents)
5889 return drop_inode_space;
5891 dropped_extents = BTRFS_I(inode)->reserved_extents -
5892 BTRFS_I(inode)->outstanding_extents;
5893 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5894 return dropped_extents + drop_inode_space;
5898 * calc_csum_metadata_size - return the amount of metadata space that must be
5899 * reserved/freed for the given bytes.
5900 * @inode: the inode we're manipulating
5901 * @num_bytes: the number of bytes in question
5902 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5904 * This adjusts the number of csum_bytes in the inode and then returns the
5905 * correct amount of metadata that must either be reserved or freed. We
5906 * calculate how many checksums we can fit into one leaf and then divide the
5907 * number of bytes that will need to be checksumed by this value to figure out
5908 * how many checksums will be required. If we are adding bytes then the number
5909 * may go up and we will return the number of additional bytes that must be
5910 * reserved. If it is going down we will return the number of bytes that must
5913 * This must be called with BTRFS_I(inode)->lock held.
5915 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5918 struct btrfs_root *root = BTRFS_I(inode)->root;
5919 u64 old_csums, num_csums;
5921 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5922 BTRFS_I(inode)->csum_bytes == 0)
5925 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5927 BTRFS_I(inode)->csum_bytes += num_bytes;
5929 BTRFS_I(inode)->csum_bytes -= num_bytes;
5930 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5932 /* No change, no need to reserve more */
5933 if (old_csums == num_csums)
5937 return btrfs_calc_trans_metadata_size(root,
5938 num_csums - old_csums);
5940 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5943 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5945 struct btrfs_root *root = BTRFS_I(inode)->root;
5946 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5949 unsigned nr_extents = 0;
5950 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5952 bool delalloc_lock = true;
5955 bool release_extra = false;
5957 /* If we are a free space inode we need to not flush since we will be in
5958 * the middle of a transaction commit. We also don't need the delalloc
5959 * mutex since we won't race with anybody. We need this mostly to make
5960 * lockdep shut its filthy mouth.
5962 if (btrfs_is_free_space_inode(inode)) {
5963 flush = BTRFS_RESERVE_NO_FLUSH;
5964 delalloc_lock = false;
5967 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5968 btrfs_transaction_in_commit(root->fs_info))
5969 schedule_timeout(1);
5972 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5974 num_bytes = ALIGN(num_bytes, root->sectorsize);
5976 spin_lock(&BTRFS_I(inode)->lock);
5977 nr_extents = (unsigned)div64_u64(num_bytes +
5978 BTRFS_MAX_EXTENT_SIZE - 1,
5979 BTRFS_MAX_EXTENT_SIZE);
5980 BTRFS_I(inode)->outstanding_extents += nr_extents;
5983 if (BTRFS_I(inode)->outstanding_extents >
5984 BTRFS_I(inode)->reserved_extents)
5985 nr_extents += BTRFS_I(inode)->outstanding_extents -
5986 BTRFS_I(inode)->reserved_extents;
5988 /* We always want to reserve a slot for updating the inode. */
5989 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents + 1);
5990 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5991 csum_bytes = BTRFS_I(inode)->csum_bytes;
5992 spin_unlock(&BTRFS_I(inode)->lock);
5994 if (root->fs_info->quota_enabled) {
5995 ret = btrfs_qgroup_reserve_meta(root,
5996 nr_extents * root->nodesize);
6001 ret = btrfs_block_rsv_add(root, block_rsv, to_reserve, flush);
6002 if (unlikely(ret)) {
6003 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
6007 spin_lock(&BTRFS_I(inode)->lock);
6008 if (test_and_set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
6009 &BTRFS_I(inode)->runtime_flags)) {
6010 to_reserve -= btrfs_calc_trans_metadata_size(root, 1);
6011 release_extra = true;
6013 BTRFS_I(inode)->reserved_extents += nr_extents;
6014 spin_unlock(&BTRFS_I(inode)->lock);
6017 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6020 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6021 btrfs_ino(inode), to_reserve, 1);
6023 btrfs_block_rsv_release(root, block_rsv,
6024 btrfs_calc_trans_metadata_size(root,
6029 spin_lock(&BTRFS_I(inode)->lock);
6030 dropped = drop_outstanding_extent(inode, num_bytes);
6032 * If the inodes csum_bytes is the same as the original
6033 * csum_bytes then we know we haven't raced with any free()ers
6034 * so we can just reduce our inodes csum bytes and carry on.
6036 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
6037 calc_csum_metadata_size(inode, num_bytes, 0);
6039 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
6043 * This is tricky, but first we need to figure out how much we
6044 * freed from any free-ers that occurred during this
6045 * reservation, so we reset ->csum_bytes to the csum_bytes
6046 * before we dropped our lock, and then call the free for the
6047 * number of bytes that were freed while we were trying our
6050 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
6051 BTRFS_I(inode)->csum_bytes = csum_bytes;
6052 to_free = calc_csum_metadata_size(inode, bytes, 0);
6056 * Now we need to see how much we would have freed had we not
6057 * been making this reservation and our ->csum_bytes were not
6058 * artificially inflated.
6060 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
6061 bytes = csum_bytes - orig_csum_bytes;
6062 bytes = calc_csum_metadata_size(inode, bytes, 0);
6065 * Now reset ->csum_bytes to what it should be. If bytes is
6066 * more than to_free then we would have freed more space had we
6067 * not had an artificially high ->csum_bytes, so we need to free
6068 * the remainder. If bytes is the same or less then we don't
6069 * need to do anything, the other free-ers did the correct
6072 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
6073 if (bytes > to_free)
6074 to_free = bytes - to_free;
6078 spin_unlock(&BTRFS_I(inode)->lock);
6080 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6083 btrfs_block_rsv_release(root, block_rsv, to_free);
6084 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6085 btrfs_ino(inode), to_free, 0);
6088 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
6093 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
6094 * @inode: the inode to release the reservation for
6095 * @num_bytes: the number of bytes we're releasing
6097 * This will release the metadata reservation for an inode. This can be called
6098 * once we complete IO for a given set of bytes to release their metadata
6101 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
6103 struct btrfs_root *root = BTRFS_I(inode)->root;
6107 num_bytes = ALIGN(num_bytes, root->sectorsize);
6108 spin_lock(&BTRFS_I(inode)->lock);
6109 dropped = drop_outstanding_extent(inode, num_bytes);
6112 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
6113 spin_unlock(&BTRFS_I(inode)->lock);
6115 to_free += btrfs_calc_trans_metadata_size(root, dropped);
6117 if (btrfs_test_is_dummy_root(root))
6120 trace_btrfs_space_reservation(root->fs_info, "delalloc",
6121 btrfs_ino(inode), to_free, 0);
6123 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
6128 * btrfs_delalloc_reserve_space - reserve data and metadata space for
6130 * @inode: inode we're writing to
6131 * @start: start range we are writing to
6132 * @len: how long the range we are writing to
6134 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
6136 * This will do the following things
6138 * o reserve space in data space info for num bytes
6139 * and reserve precious corresponding qgroup space
6140 * (Done in check_data_free_space)
6142 * o reserve space for metadata space, based on the number of outstanding
6143 * extents and how much csums will be needed
6144 * also reserve metadata space in a per root over-reserve method.
6145 * o add to the inodes->delalloc_bytes
6146 * o add it to the fs_info's delalloc inodes list.
6147 * (Above 3 all done in delalloc_reserve_metadata)
6149 * Return 0 for success
6150 * Return <0 for error(-ENOSPC or -EQUOT)
6152 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
6156 ret = btrfs_check_data_free_space(inode, start, len);
6159 ret = btrfs_delalloc_reserve_metadata(inode, len);
6161 btrfs_free_reserved_data_space(inode, start, len);
6166 * btrfs_delalloc_release_space - release data and metadata space for delalloc
6167 * @inode: inode we're releasing space for
6168 * @start: start position of the space already reserved
6169 * @len: the len of the space already reserved
6171 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
6172 * called in the case that we don't need the metadata AND data reservations
6173 * anymore. So if there is an error or we insert an inline extent.
6175 * This function will release the metadata space that was not used and will
6176 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
6177 * list if there are no delalloc bytes left.
6178 * Also it will handle the qgroup reserved space.
6180 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
6182 btrfs_delalloc_release_metadata(inode, len);
6183 btrfs_free_reserved_data_space(inode, start, len);
6186 static int update_block_group(struct btrfs_trans_handle *trans,
6187 struct btrfs_root *root, u64 bytenr,
6188 u64 num_bytes, int alloc)
6190 struct btrfs_block_group_cache *cache = NULL;
6191 struct btrfs_fs_info *info = root->fs_info;
6192 u64 total = num_bytes;
6197 /* block accounting for super block */
6198 spin_lock(&info->delalloc_root_lock);
6199 old_val = btrfs_super_bytes_used(info->super_copy);
6201 old_val += num_bytes;
6203 old_val -= num_bytes;
6204 btrfs_set_super_bytes_used(info->super_copy, old_val);
6205 spin_unlock(&info->delalloc_root_lock);
6208 cache = btrfs_lookup_block_group(info, bytenr);
6211 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
6212 BTRFS_BLOCK_GROUP_RAID1 |
6213 BTRFS_BLOCK_GROUP_RAID10))
6218 * If this block group has free space cache written out, we
6219 * need to make sure to load it if we are removing space. This
6220 * is because we need the unpinning stage to actually add the
6221 * space back to the block group, otherwise we will leak space.
6223 if (!alloc && cache->cached == BTRFS_CACHE_NO)
6224 cache_block_group(cache, 1);
6226 byte_in_group = bytenr - cache->key.objectid;
6227 WARN_ON(byte_in_group > cache->key.offset);
6229 spin_lock(&cache->space_info->lock);
6230 spin_lock(&cache->lock);
6232 if (btrfs_test_opt(root, SPACE_CACHE) &&
6233 cache->disk_cache_state < BTRFS_DC_CLEAR)
6234 cache->disk_cache_state = BTRFS_DC_CLEAR;
6236 old_val = btrfs_block_group_used(&cache->item);
6237 num_bytes = min(total, cache->key.offset - byte_in_group);
6239 old_val += num_bytes;
6240 btrfs_set_block_group_used(&cache->item, old_val);
6241 cache->reserved -= num_bytes;
6242 cache->space_info->bytes_reserved -= num_bytes;
6243 cache->space_info->bytes_used += num_bytes;
6244 cache->space_info->disk_used += num_bytes * factor;
6245 spin_unlock(&cache->lock);
6246 spin_unlock(&cache->space_info->lock);
6248 old_val -= num_bytes;
6249 btrfs_set_block_group_used(&cache->item, old_val);
6250 cache->pinned += num_bytes;
6251 cache->space_info->bytes_pinned += num_bytes;
6252 cache->space_info->bytes_used -= num_bytes;
6253 cache->space_info->disk_used -= num_bytes * factor;
6254 spin_unlock(&cache->lock);
6255 spin_unlock(&cache->space_info->lock);
6257 trace_btrfs_space_reservation(root->fs_info, "pinned",
6258 cache->space_info->flags,
6260 set_extent_dirty(info->pinned_extents,
6261 bytenr, bytenr + num_bytes - 1,
6262 GFP_NOFS | __GFP_NOFAIL);
6265 spin_lock(&trans->transaction->dirty_bgs_lock);
6266 if (list_empty(&cache->dirty_list)) {
6267 list_add_tail(&cache->dirty_list,
6268 &trans->transaction->dirty_bgs);
6269 trans->transaction->num_dirty_bgs++;
6270 btrfs_get_block_group(cache);
6272 spin_unlock(&trans->transaction->dirty_bgs_lock);
6275 * No longer have used bytes in this block group, queue it for
6276 * deletion. We do this after adding the block group to the
6277 * dirty list to avoid races between cleaner kthread and space
6280 if (!alloc && old_val == 0) {
6281 spin_lock(&info->unused_bgs_lock);
6282 if (list_empty(&cache->bg_list)) {
6283 btrfs_get_block_group(cache);
6284 list_add_tail(&cache->bg_list,
6287 spin_unlock(&info->unused_bgs_lock);
6290 btrfs_put_block_group(cache);
6292 bytenr += num_bytes;
6297 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
6299 struct btrfs_block_group_cache *cache;
6302 spin_lock(&root->fs_info->block_group_cache_lock);
6303 bytenr = root->fs_info->first_logical_byte;
6304 spin_unlock(&root->fs_info->block_group_cache_lock);
6306 if (bytenr < (u64)-1)
6309 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
6313 bytenr = cache->key.objectid;
6314 btrfs_put_block_group(cache);
6319 static int pin_down_extent(struct btrfs_root *root,
6320 struct btrfs_block_group_cache *cache,
6321 u64 bytenr, u64 num_bytes, int reserved)
6323 spin_lock(&cache->space_info->lock);
6324 spin_lock(&cache->lock);
6325 cache->pinned += num_bytes;
6326 cache->space_info->bytes_pinned += num_bytes;
6328 cache->reserved -= num_bytes;
6329 cache->space_info->bytes_reserved -= num_bytes;
6331 spin_unlock(&cache->lock);
6332 spin_unlock(&cache->space_info->lock);
6334 trace_btrfs_space_reservation(root->fs_info, "pinned",
6335 cache->space_info->flags, num_bytes, 1);
6336 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
6337 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
6342 * this function must be called within transaction
6344 int btrfs_pin_extent(struct btrfs_root *root,
6345 u64 bytenr, u64 num_bytes, int reserved)
6347 struct btrfs_block_group_cache *cache;
6349 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6350 BUG_ON(!cache); /* Logic error */
6352 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6354 btrfs_put_block_group(cache);
6359 * this function must be called within transaction
6361 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6362 u64 bytenr, u64 num_bytes)
6364 struct btrfs_block_group_cache *cache;
6367 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6372 * pull in the free space cache (if any) so that our pin
6373 * removes the free space from the cache. We have load_only set
6374 * to one because the slow code to read in the free extents does check
6375 * the pinned extents.
6377 cache_block_group(cache, 1);
6379 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6381 /* remove us from the free space cache (if we're there at all) */
6382 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6383 btrfs_put_block_group(cache);
6387 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6390 struct btrfs_block_group_cache *block_group;
6391 struct btrfs_caching_control *caching_ctl;
6393 block_group = btrfs_lookup_block_group(root->fs_info, start);
6397 cache_block_group(block_group, 0);
6398 caching_ctl = get_caching_control(block_group);
6402 BUG_ON(!block_group_cache_done(block_group));
6403 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6405 mutex_lock(&caching_ctl->mutex);
6407 if (start >= caching_ctl->progress) {
6408 ret = add_excluded_extent(root, start, num_bytes);
6409 } else if (start + num_bytes <= caching_ctl->progress) {
6410 ret = btrfs_remove_free_space(block_group,
6413 num_bytes = caching_ctl->progress - start;
6414 ret = btrfs_remove_free_space(block_group,
6419 num_bytes = (start + num_bytes) -
6420 caching_ctl->progress;
6421 start = caching_ctl->progress;
6422 ret = add_excluded_extent(root, start, num_bytes);
6425 mutex_unlock(&caching_ctl->mutex);
6426 put_caching_control(caching_ctl);
6428 btrfs_put_block_group(block_group);
6432 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6433 struct extent_buffer *eb)
6435 struct btrfs_file_extent_item *item;
6436 struct btrfs_key key;
6440 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6443 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6444 btrfs_item_key_to_cpu(eb, &key, i);
6445 if (key.type != BTRFS_EXTENT_DATA_KEY)
6447 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6448 found_type = btrfs_file_extent_type(eb, item);
6449 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6451 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6453 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6454 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6455 __exclude_logged_extent(log, key.objectid, key.offset);
6462 btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg)
6464 atomic_inc(&bg->reservations);
6467 void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info,
6470 struct btrfs_block_group_cache *bg;
6472 bg = btrfs_lookup_block_group(fs_info, start);
6474 if (atomic_dec_and_test(&bg->reservations))
6475 wake_up_atomic_t(&bg->reservations);
6476 btrfs_put_block_group(bg);
6479 static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a)
6485 void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg)
6487 struct btrfs_space_info *space_info = bg->space_info;
6491 if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA))
6495 * Our block group is read only but before we set it to read only,
6496 * some task might have had allocated an extent from it already, but it
6497 * has not yet created a respective ordered extent (and added it to a
6498 * root's list of ordered extents).
6499 * Therefore wait for any task currently allocating extents, since the
6500 * block group's reservations counter is incremented while a read lock
6501 * on the groups' semaphore is held and decremented after releasing
6502 * the read access on that semaphore and creating the ordered extent.
6504 down_write(&space_info->groups_sem);
6505 up_write(&space_info->groups_sem);
6507 wait_on_atomic_t(&bg->reservations,
6508 btrfs_wait_bg_reservations_atomic_t,
6509 TASK_UNINTERRUPTIBLE);
6513 * btrfs_update_reserved_bytes - update the block_group and space info counters
6514 * @cache: The cache we are manipulating
6515 * @num_bytes: The number of bytes in question
6516 * @reserve: One of the reservation enums
6517 * @delalloc: The blocks are allocated for the delalloc write
6519 * This is called by the allocator when it reserves space, or by somebody who is
6520 * freeing space that was never actually used on disk. For example if you
6521 * reserve some space for a new leaf in transaction A and before transaction A
6522 * commits you free that leaf, you call this with reserve set to 0 in order to
6523 * clear the reservation.
6525 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6526 * ENOSPC accounting. For data we handle the reservation through clearing the
6527 * delalloc bits in the io_tree. We have to do this since we could end up
6528 * allocating less disk space for the amount of data we have reserved in the
6529 * case of compression.
6531 * If this is a reservation and the block group has become read only we cannot
6532 * make the reservation and return -EAGAIN, otherwise this function always
6535 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6536 u64 num_bytes, int reserve, int delalloc)
6538 struct btrfs_space_info *space_info = cache->space_info;
6541 spin_lock(&space_info->lock);
6542 spin_lock(&cache->lock);
6543 if (reserve != RESERVE_FREE) {
6547 cache->reserved += num_bytes;
6548 space_info->bytes_reserved += num_bytes;
6549 if (reserve == RESERVE_ALLOC) {
6550 trace_btrfs_space_reservation(cache->fs_info,
6551 "space_info", space_info->flags,
6553 space_info->bytes_may_use -= num_bytes;
6557 cache->delalloc_bytes += num_bytes;
6561 space_info->bytes_readonly += num_bytes;
6562 cache->reserved -= num_bytes;
6563 space_info->bytes_reserved -= num_bytes;
6566 cache->delalloc_bytes -= num_bytes;
6568 spin_unlock(&cache->lock);
6569 spin_unlock(&space_info->lock);
6573 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6574 struct btrfs_root *root)
6576 struct btrfs_fs_info *fs_info = root->fs_info;
6577 struct btrfs_caching_control *next;
6578 struct btrfs_caching_control *caching_ctl;
6579 struct btrfs_block_group_cache *cache;
6581 down_write(&fs_info->commit_root_sem);
6583 list_for_each_entry_safe(caching_ctl, next,
6584 &fs_info->caching_block_groups, list) {
6585 cache = caching_ctl->block_group;
6586 if (block_group_cache_done(cache)) {
6587 cache->last_byte_to_unpin = (u64)-1;
6588 list_del_init(&caching_ctl->list);
6589 put_caching_control(caching_ctl);
6591 cache->last_byte_to_unpin = caching_ctl->progress;
6595 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6596 fs_info->pinned_extents = &fs_info->freed_extents[1];
6598 fs_info->pinned_extents = &fs_info->freed_extents[0];
6600 up_write(&fs_info->commit_root_sem);
6602 update_global_block_rsv(fs_info);
6606 * Returns the free cluster for the given space info and sets empty_cluster to
6607 * what it should be based on the mount options.
6609 static struct btrfs_free_cluster *
6610 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6613 struct btrfs_free_cluster *ret = NULL;
6614 bool ssd = btrfs_test_opt(root, SSD);
6617 if (btrfs_mixed_space_info(space_info))
6621 *empty_cluster = SZ_2M;
6622 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6623 ret = &root->fs_info->meta_alloc_cluster;
6625 *empty_cluster = SZ_64K;
6626 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6627 ret = &root->fs_info->data_alloc_cluster;
6633 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6634 const bool return_free_space)
6636 struct btrfs_fs_info *fs_info = root->fs_info;
6637 struct btrfs_block_group_cache *cache = NULL;
6638 struct btrfs_space_info *space_info;
6639 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6640 struct btrfs_free_cluster *cluster = NULL;
6642 u64 total_unpinned = 0;
6643 u64 empty_cluster = 0;
6646 while (start <= end) {
6649 start >= cache->key.objectid + cache->key.offset) {
6651 btrfs_put_block_group(cache);
6653 cache = btrfs_lookup_block_group(fs_info, start);
6654 BUG_ON(!cache); /* Logic error */
6656 cluster = fetch_cluster_info(root,
6659 empty_cluster <<= 1;
6662 len = cache->key.objectid + cache->key.offset - start;
6663 len = min(len, end + 1 - start);
6665 if (start < cache->last_byte_to_unpin) {
6666 len = min(len, cache->last_byte_to_unpin - start);
6667 if (return_free_space)
6668 btrfs_add_free_space(cache, start, len);
6672 total_unpinned += len;
6673 space_info = cache->space_info;
6676 * If this space cluster has been marked as fragmented and we've
6677 * unpinned enough in this block group to potentially allow a
6678 * cluster to be created inside of it go ahead and clear the
6681 if (cluster && cluster->fragmented &&
6682 total_unpinned > empty_cluster) {
6683 spin_lock(&cluster->lock);
6684 cluster->fragmented = 0;
6685 spin_unlock(&cluster->lock);
6688 spin_lock(&space_info->lock);
6689 spin_lock(&cache->lock);
6690 cache->pinned -= len;
6691 space_info->bytes_pinned -= len;
6693 trace_btrfs_space_reservation(fs_info, "pinned",
6694 space_info->flags, len, 0);
6695 space_info->max_extent_size = 0;
6696 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6698 space_info->bytes_readonly += len;
6701 spin_unlock(&cache->lock);
6702 if (!readonly && return_free_space &&
6703 global_rsv->space_info == space_info) {
6705 WARN_ON(!return_free_space);
6706 spin_lock(&global_rsv->lock);
6707 if (!global_rsv->full) {
6708 to_add = min(len, global_rsv->size -
6709 global_rsv->reserved);
6710 global_rsv->reserved += to_add;
6711 space_info->bytes_may_use += to_add;
6712 if (global_rsv->reserved >= global_rsv->size)
6713 global_rsv->full = 1;
6714 trace_btrfs_space_reservation(fs_info,
6720 spin_unlock(&global_rsv->lock);
6721 /* Add to any tickets we may have */
6723 space_info_add_new_bytes(fs_info, space_info,
6726 spin_unlock(&space_info->lock);
6730 btrfs_put_block_group(cache);
6734 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6735 struct btrfs_root *root)
6737 struct btrfs_fs_info *fs_info = root->fs_info;
6738 struct btrfs_block_group_cache *block_group, *tmp;
6739 struct list_head *deleted_bgs;
6740 struct extent_io_tree *unpin;
6745 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6746 unpin = &fs_info->freed_extents[1];
6748 unpin = &fs_info->freed_extents[0];
6750 while (!trans->aborted) {
6751 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6752 ret = find_first_extent_bit(unpin, 0, &start, &end,
6753 EXTENT_DIRTY, NULL);
6755 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6759 if (btrfs_test_opt(root, DISCARD))
6760 ret = btrfs_discard_extent(root, start,
6761 end + 1 - start, NULL);
6763 clear_extent_dirty(unpin, start, end);
6764 unpin_extent_range(root, start, end, true);
6765 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6770 * Transaction is finished. We don't need the lock anymore. We
6771 * do need to clean up the block groups in case of a transaction
6774 deleted_bgs = &trans->transaction->deleted_bgs;
6775 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6779 if (!trans->aborted)
6780 ret = btrfs_discard_extent(root,
6781 block_group->key.objectid,
6782 block_group->key.offset,
6785 list_del_init(&block_group->bg_list);
6786 btrfs_put_block_group_trimming(block_group);
6787 btrfs_put_block_group(block_group);
6790 const char *errstr = btrfs_decode_error(ret);
6792 "Discard failed while removing blockgroup: errno=%d %s\n",
6800 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6801 u64 owner, u64 root_objectid)
6803 struct btrfs_space_info *space_info;
6806 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6807 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6808 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6810 flags = BTRFS_BLOCK_GROUP_METADATA;
6812 flags = BTRFS_BLOCK_GROUP_DATA;
6815 space_info = __find_space_info(fs_info, flags);
6816 BUG_ON(!space_info); /* Logic bug */
6817 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6821 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6822 struct btrfs_root *root,
6823 struct btrfs_delayed_ref_node *node, u64 parent,
6824 u64 root_objectid, u64 owner_objectid,
6825 u64 owner_offset, int refs_to_drop,
6826 struct btrfs_delayed_extent_op *extent_op)
6828 struct btrfs_key key;
6829 struct btrfs_path *path;
6830 struct btrfs_fs_info *info = root->fs_info;
6831 struct btrfs_root *extent_root = info->extent_root;
6832 struct extent_buffer *leaf;
6833 struct btrfs_extent_item *ei;
6834 struct btrfs_extent_inline_ref *iref;
6837 int extent_slot = 0;
6838 int found_extent = 0;
6842 u64 bytenr = node->bytenr;
6843 u64 num_bytes = node->num_bytes;
6845 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6848 path = btrfs_alloc_path();
6852 path->reada = READA_FORWARD;
6853 path->leave_spinning = 1;
6855 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6856 BUG_ON(!is_data && refs_to_drop != 1);
6859 skinny_metadata = 0;
6861 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6862 bytenr, num_bytes, parent,
6863 root_objectid, owner_objectid,
6866 extent_slot = path->slots[0];
6867 while (extent_slot >= 0) {
6868 btrfs_item_key_to_cpu(path->nodes[0], &key,
6870 if (key.objectid != bytenr)
6872 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6873 key.offset == num_bytes) {
6877 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6878 key.offset == owner_objectid) {
6882 if (path->slots[0] - extent_slot > 5)
6886 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6887 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6888 if (found_extent && item_size < sizeof(*ei))
6891 if (!found_extent) {
6893 ret = remove_extent_backref(trans, extent_root, path,
6895 is_data, &last_ref);
6897 btrfs_abort_transaction(trans, extent_root, ret);
6900 btrfs_release_path(path);
6901 path->leave_spinning = 1;
6903 key.objectid = bytenr;
6904 key.type = BTRFS_EXTENT_ITEM_KEY;
6905 key.offset = num_bytes;
6907 if (!is_data && skinny_metadata) {
6908 key.type = BTRFS_METADATA_ITEM_KEY;
6909 key.offset = owner_objectid;
6912 ret = btrfs_search_slot(trans, extent_root,
6914 if (ret > 0 && skinny_metadata && path->slots[0]) {
6916 * Couldn't find our skinny metadata item,
6917 * see if we have ye olde extent item.
6920 btrfs_item_key_to_cpu(path->nodes[0], &key,
6922 if (key.objectid == bytenr &&
6923 key.type == BTRFS_EXTENT_ITEM_KEY &&
6924 key.offset == num_bytes)
6928 if (ret > 0 && skinny_metadata) {
6929 skinny_metadata = false;
6930 key.objectid = bytenr;
6931 key.type = BTRFS_EXTENT_ITEM_KEY;
6932 key.offset = num_bytes;
6933 btrfs_release_path(path);
6934 ret = btrfs_search_slot(trans, extent_root,
6939 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6942 btrfs_print_leaf(extent_root,
6946 btrfs_abort_transaction(trans, extent_root, ret);
6949 extent_slot = path->slots[0];
6951 } else if (WARN_ON(ret == -ENOENT)) {
6952 btrfs_print_leaf(extent_root, path->nodes[0]);
6954 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6955 bytenr, parent, root_objectid, owner_objectid,
6957 btrfs_abort_transaction(trans, extent_root, ret);
6960 btrfs_abort_transaction(trans, extent_root, ret);
6964 leaf = path->nodes[0];
6965 item_size = btrfs_item_size_nr(leaf, extent_slot);
6966 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6967 if (item_size < sizeof(*ei)) {
6968 BUG_ON(found_extent || extent_slot != path->slots[0]);
6969 ret = convert_extent_item_v0(trans, extent_root, path,
6972 btrfs_abort_transaction(trans, extent_root, ret);
6976 btrfs_release_path(path);
6977 path->leave_spinning = 1;
6979 key.objectid = bytenr;
6980 key.type = BTRFS_EXTENT_ITEM_KEY;
6981 key.offset = num_bytes;
6983 ret = btrfs_search_slot(trans, extent_root, &key, path,
6986 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6988 btrfs_print_leaf(extent_root, path->nodes[0]);
6991 btrfs_abort_transaction(trans, extent_root, ret);
6995 extent_slot = path->slots[0];
6996 leaf = path->nodes[0];
6997 item_size = btrfs_item_size_nr(leaf, extent_slot);
7000 BUG_ON(item_size < sizeof(*ei));
7001 ei = btrfs_item_ptr(leaf, extent_slot,
7002 struct btrfs_extent_item);
7003 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
7004 key.type == BTRFS_EXTENT_ITEM_KEY) {
7005 struct btrfs_tree_block_info *bi;
7006 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
7007 bi = (struct btrfs_tree_block_info *)(ei + 1);
7008 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
7011 refs = btrfs_extent_refs(leaf, ei);
7012 if (refs < refs_to_drop) {
7013 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
7014 "for bytenr %Lu", refs_to_drop, refs, bytenr);
7016 btrfs_abort_transaction(trans, extent_root, ret);
7019 refs -= refs_to_drop;
7023 __run_delayed_extent_op(extent_op, leaf, ei);
7025 * In the case of inline back ref, reference count will
7026 * be updated by remove_extent_backref
7029 BUG_ON(!found_extent);
7031 btrfs_set_extent_refs(leaf, ei, refs);
7032 btrfs_mark_buffer_dirty(leaf);
7035 ret = remove_extent_backref(trans, extent_root, path,
7037 is_data, &last_ref);
7039 btrfs_abort_transaction(trans, extent_root, ret);
7043 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
7047 BUG_ON(is_data && refs_to_drop !=
7048 extent_data_ref_count(path, iref));
7050 BUG_ON(path->slots[0] != extent_slot);
7052 BUG_ON(path->slots[0] != extent_slot + 1);
7053 path->slots[0] = extent_slot;
7059 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
7062 btrfs_abort_transaction(trans, extent_root, ret);
7065 btrfs_release_path(path);
7068 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
7070 btrfs_abort_transaction(trans, extent_root, ret);
7075 ret = add_to_free_space_tree(trans, root->fs_info, bytenr,
7078 btrfs_abort_transaction(trans, extent_root, ret);
7082 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
7084 btrfs_abort_transaction(trans, extent_root, ret);
7088 btrfs_release_path(path);
7091 btrfs_free_path(path);
7096 * when we free an block, it is possible (and likely) that we free the last
7097 * delayed ref for that extent as well. This searches the delayed ref tree for
7098 * a given extent, and if there are no other delayed refs to be processed, it
7099 * removes it from the tree.
7101 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
7102 struct btrfs_root *root, u64 bytenr)
7104 struct btrfs_delayed_ref_head *head;
7105 struct btrfs_delayed_ref_root *delayed_refs;
7108 delayed_refs = &trans->transaction->delayed_refs;
7109 spin_lock(&delayed_refs->lock);
7110 head = btrfs_find_delayed_ref_head(trans, bytenr);
7112 goto out_delayed_unlock;
7114 spin_lock(&head->lock);
7115 if (!list_empty(&head->ref_list))
7118 if (head->extent_op) {
7119 if (!head->must_insert_reserved)
7121 btrfs_free_delayed_extent_op(head->extent_op);
7122 head->extent_op = NULL;
7126 * waiting for the lock here would deadlock. If someone else has it
7127 * locked they are already in the process of dropping it anyway
7129 if (!mutex_trylock(&head->mutex))
7133 * at this point we have a head with no other entries. Go
7134 * ahead and process it.
7136 head->node.in_tree = 0;
7137 rb_erase(&head->href_node, &delayed_refs->href_root);
7139 atomic_dec(&delayed_refs->num_entries);
7142 * we don't take a ref on the node because we're removing it from the
7143 * tree, so we just steal the ref the tree was holding.
7145 delayed_refs->num_heads--;
7146 if (head->processing == 0)
7147 delayed_refs->num_heads_ready--;
7148 head->processing = 0;
7149 spin_unlock(&head->lock);
7150 spin_unlock(&delayed_refs->lock);
7152 BUG_ON(head->extent_op);
7153 if (head->must_insert_reserved)
7156 mutex_unlock(&head->mutex);
7157 btrfs_put_delayed_ref(&head->node);
7160 spin_unlock(&head->lock);
7163 spin_unlock(&delayed_refs->lock);
7167 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
7168 struct btrfs_root *root,
7169 struct extent_buffer *buf,
7170 u64 parent, int last_ref)
7175 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7176 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7177 buf->start, buf->len,
7178 parent, root->root_key.objectid,
7179 btrfs_header_level(buf),
7180 BTRFS_DROP_DELAYED_REF, NULL);
7181 BUG_ON(ret); /* -ENOMEM */
7187 if (btrfs_header_generation(buf) == trans->transid) {
7188 struct btrfs_block_group_cache *cache;
7190 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
7191 ret = check_ref_cleanup(trans, root, buf->start);
7196 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
7198 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
7199 pin_down_extent(root, cache, buf->start, buf->len, 1);
7200 btrfs_put_block_group(cache);
7204 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
7206 btrfs_add_free_space(cache, buf->start, buf->len);
7207 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
7208 btrfs_put_block_group(cache);
7209 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
7214 add_pinned_bytes(root->fs_info, buf->len,
7215 btrfs_header_level(buf),
7216 root->root_key.objectid);
7219 * Deleting the buffer, clear the corrupt flag since it doesn't matter
7222 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
7225 /* Can return -ENOMEM */
7226 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7227 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
7228 u64 owner, u64 offset)
7231 struct btrfs_fs_info *fs_info = root->fs_info;
7233 if (btrfs_test_is_dummy_root(root))
7236 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
7239 * tree log blocks never actually go into the extent allocation
7240 * tree, just update pinning info and exit early.
7242 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
7243 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
7244 /* unlocks the pinned mutex */
7245 btrfs_pin_extent(root, bytenr, num_bytes, 1);
7247 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
7248 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
7250 parent, root_objectid, (int)owner,
7251 BTRFS_DROP_DELAYED_REF, NULL);
7253 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
7255 parent, root_objectid, owner,
7257 BTRFS_DROP_DELAYED_REF, NULL);
7263 * when we wait for progress in the block group caching, its because
7264 * our allocation attempt failed at least once. So, we must sleep
7265 * and let some progress happen before we try again.
7267 * This function will sleep at least once waiting for new free space to
7268 * show up, and then it will check the block group free space numbers
7269 * for our min num_bytes. Another option is to have it go ahead
7270 * and look in the rbtree for a free extent of a given size, but this
7273 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
7274 * any of the information in this block group.
7276 static noinline void
7277 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
7280 struct btrfs_caching_control *caching_ctl;
7282 caching_ctl = get_caching_control(cache);
7286 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
7287 (cache->free_space_ctl->free_space >= num_bytes));
7289 put_caching_control(caching_ctl);
7293 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
7295 struct btrfs_caching_control *caching_ctl;
7298 caching_ctl = get_caching_control(cache);
7300 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
7302 wait_event(caching_ctl->wait, block_group_cache_done(cache));
7303 if (cache->cached == BTRFS_CACHE_ERROR)
7305 put_caching_control(caching_ctl);
7309 int __get_raid_index(u64 flags)
7311 if (flags & BTRFS_BLOCK_GROUP_RAID10)
7312 return BTRFS_RAID_RAID10;
7313 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
7314 return BTRFS_RAID_RAID1;
7315 else if (flags & BTRFS_BLOCK_GROUP_DUP)
7316 return BTRFS_RAID_DUP;
7317 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
7318 return BTRFS_RAID_RAID0;
7319 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
7320 return BTRFS_RAID_RAID5;
7321 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
7322 return BTRFS_RAID_RAID6;
7324 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
7327 int get_block_group_index(struct btrfs_block_group_cache *cache)
7329 return __get_raid_index(cache->flags);
7332 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
7333 [BTRFS_RAID_RAID10] = "raid10",
7334 [BTRFS_RAID_RAID1] = "raid1",
7335 [BTRFS_RAID_DUP] = "dup",
7336 [BTRFS_RAID_RAID0] = "raid0",
7337 [BTRFS_RAID_SINGLE] = "single",
7338 [BTRFS_RAID_RAID5] = "raid5",
7339 [BTRFS_RAID_RAID6] = "raid6",
7342 static const char *get_raid_name(enum btrfs_raid_types type)
7344 if (type >= BTRFS_NR_RAID_TYPES)
7347 return btrfs_raid_type_names[type];
7350 enum btrfs_loop_type {
7351 LOOP_CACHING_NOWAIT = 0,
7352 LOOP_CACHING_WAIT = 1,
7353 LOOP_ALLOC_CHUNK = 2,
7354 LOOP_NO_EMPTY_SIZE = 3,
7358 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
7362 down_read(&cache->data_rwsem);
7366 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
7369 btrfs_get_block_group(cache);
7371 down_read(&cache->data_rwsem);
7374 static struct btrfs_block_group_cache *
7375 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
7376 struct btrfs_free_cluster *cluster,
7379 struct btrfs_block_group_cache *used_bg = NULL;
7381 spin_lock(&cluster->refill_lock);
7383 used_bg = cluster->block_group;
7387 if (used_bg == block_group)
7390 btrfs_get_block_group(used_bg);
7395 if (down_read_trylock(&used_bg->data_rwsem))
7398 spin_unlock(&cluster->refill_lock);
7400 down_read(&used_bg->data_rwsem);
7402 spin_lock(&cluster->refill_lock);
7403 if (used_bg == cluster->block_group)
7406 up_read(&used_bg->data_rwsem);
7407 btrfs_put_block_group(used_bg);
7412 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7416 up_read(&cache->data_rwsem);
7417 btrfs_put_block_group(cache);
7421 * walks the btree of allocated extents and find a hole of a given size.
7422 * The key ins is changed to record the hole:
7423 * ins->objectid == start position
7424 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7425 * ins->offset == the size of the hole.
7426 * Any available blocks before search_start are skipped.
7428 * If there is no suitable free space, we will record the max size of
7429 * the free space extent currently.
7431 static noinline int find_free_extent(struct btrfs_root *orig_root,
7432 u64 num_bytes, u64 empty_size,
7433 u64 hint_byte, struct btrfs_key *ins,
7434 u64 flags, int delalloc)
7437 struct btrfs_root *root = orig_root->fs_info->extent_root;
7438 struct btrfs_free_cluster *last_ptr = NULL;
7439 struct btrfs_block_group_cache *block_group = NULL;
7440 u64 search_start = 0;
7441 u64 max_extent_size = 0;
7442 u64 empty_cluster = 0;
7443 struct btrfs_space_info *space_info;
7445 int index = __get_raid_index(flags);
7446 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7447 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7448 bool failed_cluster_refill = false;
7449 bool failed_alloc = false;
7450 bool use_cluster = true;
7451 bool have_caching_bg = false;
7452 bool orig_have_caching_bg = false;
7453 bool full_search = false;
7455 WARN_ON(num_bytes < root->sectorsize);
7456 ins->type = BTRFS_EXTENT_ITEM_KEY;
7460 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7462 space_info = __find_space_info(root->fs_info, flags);
7464 btrfs_err(root->fs_info, "No space info for %llu", flags);
7469 * If our free space is heavily fragmented we may not be able to make
7470 * big contiguous allocations, so instead of doing the expensive search
7471 * for free space, simply return ENOSPC with our max_extent_size so we
7472 * can go ahead and search for a more manageable chunk.
7474 * If our max_extent_size is large enough for our allocation simply
7475 * disable clustering since we will likely not be able to find enough
7476 * space to create a cluster and induce latency trying.
7478 if (unlikely(space_info->max_extent_size)) {
7479 spin_lock(&space_info->lock);
7480 if (space_info->max_extent_size &&
7481 num_bytes > space_info->max_extent_size) {
7482 ins->offset = space_info->max_extent_size;
7483 spin_unlock(&space_info->lock);
7485 } else if (space_info->max_extent_size) {
7486 use_cluster = false;
7488 spin_unlock(&space_info->lock);
7491 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7493 spin_lock(&last_ptr->lock);
7494 if (last_ptr->block_group)
7495 hint_byte = last_ptr->window_start;
7496 if (last_ptr->fragmented) {
7498 * We still set window_start so we can keep track of the
7499 * last place we found an allocation to try and save
7502 hint_byte = last_ptr->window_start;
7503 use_cluster = false;
7505 spin_unlock(&last_ptr->lock);
7508 search_start = max(search_start, first_logical_byte(root, 0));
7509 search_start = max(search_start, hint_byte);
7510 if (search_start == hint_byte) {
7511 block_group = btrfs_lookup_block_group(root->fs_info,
7514 * we don't want to use the block group if it doesn't match our
7515 * allocation bits, or if its not cached.
7517 * However if we are re-searching with an ideal block group
7518 * picked out then we don't care that the block group is cached.
7520 if (block_group && block_group_bits(block_group, flags) &&
7521 block_group->cached != BTRFS_CACHE_NO) {
7522 down_read(&space_info->groups_sem);
7523 if (list_empty(&block_group->list) ||
7526 * someone is removing this block group,
7527 * we can't jump into the have_block_group
7528 * target because our list pointers are not
7531 btrfs_put_block_group(block_group);
7532 up_read(&space_info->groups_sem);
7534 index = get_block_group_index(block_group);
7535 btrfs_lock_block_group(block_group, delalloc);
7536 goto have_block_group;
7538 } else if (block_group) {
7539 btrfs_put_block_group(block_group);
7543 have_caching_bg = false;
7544 if (index == 0 || index == __get_raid_index(flags))
7546 down_read(&space_info->groups_sem);
7547 list_for_each_entry(block_group, &space_info->block_groups[index],
7552 btrfs_grab_block_group(block_group, delalloc);
7553 search_start = block_group->key.objectid;
7556 * this can happen if we end up cycling through all the
7557 * raid types, but we want to make sure we only allocate
7558 * for the proper type.
7560 if (!block_group_bits(block_group, flags)) {
7561 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7562 BTRFS_BLOCK_GROUP_RAID1 |
7563 BTRFS_BLOCK_GROUP_RAID5 |
7564 BTRFS_BLOCK_GROUP_RAID6 |
7565 BTRFS_BLOCK_GROUP_RAID10;
7568 * if they asked for extra copies and this block group
7569 * doesn't provide them, bail. This does allow us to
7570 * fill raid0 from raid1.
7572 if ((flags & extra) && !(block_group->flags & extra))
7577 cached = block_group_cache_done(block_group);
7578 if (unlikely(!cached)) {
7579 have_caching_bg = true;
7580 ret = cache_block_group(block_group, 0);
7585 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7587 if (unlikely(block_group->ro))
7591 * Ok we want to try and use the cluster allocator, so
7594 if (last_ptr && use_cluster) {
7595 struct btrfs_block_group_cache *used_block_group;
7596 unsigned long aligned_cluster;
7598 * the refill lock keeps out other
7599 * people trying to start a new cluster
7601 used_block_group = btrfs_lock_cluster(block_group,
7604 if (!used_block_group)
7605 goto refill_cluster;
7607 if (used_block_group != block_group &&
7608 (used_block_group->ro ||
7609 !block_group_bits(used_block_group, flags)))
7610 goto release_cluster;
7612 offset = btrfs_alloc_from_cluster(used_block_group,
7615 used_block_group->key.objectid,
7618 /* we have a block, we're done */
7619 spin_unlock(&last_ptr->refill_lock);
7620 trace_btrfs_reserve_extent_cluster(root,
7622 search_start, num_bytes);
7623 if (used_block_group != block_group) {
7624 btrfs_release_block_group(block_group,
7626 block_group = used_block_group;
7631 WARN_ON(last_ptr->block_group != used_block_group);
7633 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7634 * set up a new clusters, so lets just skip it
7635 * and let the allocator find whatever block
7636 * it can find. If we reach this point, we
7637 * will have tried the cluster allocator
7638 * plenty of times and not have found
7639 * anything, so we are likely way too
7640 * fragmented for the clustering stuff to find
7643 * However, if the cluster is taken from the
7644 * current block group, release the cluster
7645 * first, so that we stand a better chance of
7646 * succeeding in the unclustered
7648 if (loop >= LOOP_NO_EMPTY_SIZE &&
7649 used_block_group != block_group) {
7650 spin_unlock(&last_ptr->refill_lock);
7651 btrfs_release_block_group(used_block_group,
7653 goto unclustered_alloc;
7657 * this cluster didn't work out, free it and
7660 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7662 if (used_block_group != block_group)
7663 btrfs_release_block_group(used_block_group,
7666 if (loop >= LOOP_NO_EMPTY_SIZE) {
7667 spin_unlock(&last_ptr->refill_lock);
7668 goto unclustered_alloc;
7671 aligned_cluster = max_t(unsigned long,
7672 empty_cluster + empty_size,
7673 block_group->full_stripe_len);
7675 /* allocate a cluster in this block group */
7676 ret = btrfs_find_space_cluster(root, block_group,
7677 last_ptr, search_start,
7682 * now pull our allocation out of this
7685 offset = btrfs_alloc_from_cluster(block_group,
7691 /* we found one, proceed */
7692 spin_unlock(&last_ptr->refill_lock);
7693 trace_btrfs_reserve_extent_cluster(root,
7694 block_group, search_start,
7698 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7699 && !failed_cluster_refill) {
7700 spin_unlock(&last_ptr->refill_lock);
7702 failed_cluster_refill = true;
7703 wait_block_group_cache_progress(block_group,
7704 num_bytes + empty_cluster + empty_size);
7705 goto have_block_group;
7709 * at this point we either didn't find a cluster
7710 * or we weren't able to allocate a block from our
7711 * cluster. Free the cluster we've been trying
7712 * to use, and go to the next block group
7714 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7715 spin_unlock(&last_ptr->refill_lock);
7721 * We are doing an unclustered alloc, set the fragmented flag so
7722 * we don't bother trying to setup a cluster again until we get
7725 if (unlikely(last_ptr)) {
7726 spin_lock(&last_ptr->lock);
7727 last_ptr->fragmented = 1;
7728 spin_unlock(&last_ptr->lock);
7730 spin_lock(&block_group->free_space_ctl->tree_lock);
7732 block_group->free_space_ctl->free_space <
7733 num_bytes + empty_cluster + empty_size) {
7734 if (block_group->free_space_ctl->free_space >
7737 block_group->free_space_ctl->free_space;
7738 spin_unlock(&block_group->free_space_ctl->tree_lock);
7741 spin_unlock(&block_group->free_space_ctl->tree_lock);
7743 offset = btrfs_find_space_for_alloc(block_group, search_start,
7744 num_bytes, empty_size,
7747 * If we didn't find a chunk, and we haven't failed on this
7748 * block group before, and this block group is in the middle of
7749 * caching and we are ok with waiting, then go ahead and wait
7750 * for progress to be made, and set failed_alloc to true.
7752 * If failed_alloc is true then we've already waited on this
7753 * block group once and should move on to the next block group.
7755 if (!offset && !failed_alloc && !cached &&
7756 loop > LOOP_CACHING_NOWAIT) {
7757 wait_block_group_cache_progress(block_group,
7758 num_bytes + empty_size);
7759 failed_alloc = true;
7760 goto have_block_group;
7761 } else if (!offset) {
7765 search_start = ALIGN(offset, root->stripesize);
7767 /* move on to the next group */
7768 if (search_start + num_bytes >
7769 block_group->key.objectid + block_group->key.offset) {
7770 btrfs_add_free_space(block_group, offset, num_bytes);
7774 if (offset < search_start)
7775 btrfs_add_free_space(block_group, offset,
7776 search_start - offset);
7777 BUG_ON(offset > search_start);
7779 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7780 alloc_type, delalloc);
7781 if (ret == -EAGAIN) {
7782 btrfs_add_free_space(block_group, offset, num_bytes);
7785 btrfs_inc_block_group_reservations(block_group);
7787 /* we are all good, lets return */
7788 ins->objectid = search_start;
7789 ins->offset = num_bytes;
7791 trace_btrfs_reserve_extent(orig_root, block_group,
7792 search_start, num_bytes);
7793 btrfs_release_block_group(block_group, delalloc);
7796 failed_cluster_refill = false;
7797 failed_alloc = false;
7798 BUG_ON(index != get_block_group_index(block_group));
7799 btrfs_release_block_group(block_group, delalloc);
7801 up_read(&space_info->groups_sem);
7803 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7804 && !orig_have_caching_bg)
7805 orig_have_caching_bg = true;
7807 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7810 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7814 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7815 * caching kthreads as we move along
7816 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7817 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7818 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7821 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7823 if (loop == LOOP_CACHING_NOWAIT) {
7825 * We want to skip the LOOP_CACHING_WAIT step if we
7826 * don't have any uncached bgs and we've already done a
7827 * full search through.
7829 if (orig_have_caching_bg || !full_search)
7830 loop = LOOP_CACHING_WAIT;
7832 loop = LOOP_ALLOC_CHUNK;
7837 if (loop == LOOP_ALLOC_CHUNK) {
7838 struct btrfs_trans_handle *trans;
7841 trans = current->journal_info;
7845 trans = btrfs_join_transaction(root);
7847 if (IS_ERR(trans)) {
7848 ret = PTR_ERR(trans);
7852 ret = do_chunk_alloc(trans, root, flags,
7856 * If we can't allocate a new chunk we've already looped
7857 * through at least once, move on to the NO_EMPTY_SIZE
7861 loop = LOOP_NO_EMPTY_SIZE;
7864 * Do not bail out on ENOSPC since we
7865 * can do more things.
7867 if (ret < 0 && ret != -ENOSPC)
7868 btrfs_abort_transaction(trans,
7873 btrfs_end_transaction(trans, root);
7878 if (loop == LOOP_NO_EMPTY_SIZE) {
7880 * Don't loop again if we already have no empty_size and
7883 if (empty_size == 0 &&
7884 empty_cluster == 0) {
7893 } else if (!ins->objectid) {
7895 } else if (ins->objectid) {
7896 if (!use_cluster && last_ptr) {
7897 spin_lock(&last_ptr->lock);
7898 last_ptr->window_start = ins->objectid;
7899 spin_unlock(&last_ptr->lock);
7904 if (ret == -ENOSPC) {
7905 spin_lock(&space_info->lock);
7906 space_info->max_extent_size = max_extent_size;
7907 spin_unlock(&space_info->lock);
7908 ins->offset = max_extent_size;
7913 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7914 int dump_block_groups)
7916 struct btrfs_block_group_cache *cache;
7919 spin_lock(&info->lock);
7920 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7922 info->total_bytes - info->bytes_used - info->bytes_pinned -
7923 info->bytes_reserved - info->bytes_readonly,
7924 (info->full) ? "" : "not ");
7925 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7926 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7927 info->total_bytes, info->bytes_used, info->bytes_pinned,
7928 info->bytes_reserved, info->bytes_may_use,
7929 info->bytes_readonly);
7930 spin_unlock(&info->lock);
7932 if (!dump_block_groups)
7935 down_read(&info->groups_sem);
7937 list_for_each_entry(cache, &info->block_groups[index], list) {
7938 spin_lock(&cache->lock);
7939 printk(KERN_INFO "BTRFS: "
7940 "block group %llu has %llu bytes, "
7941 "%llu used %llu pinned %llu reserved %s\n",
7942 cache->key.objectid, cache->key.offset,
7943 btrfs_block_group_used(&cache->item), cache->pinned,
7944 cache->reserved, cache->ro ? "[readonly]" : "");
7945 btrfs_dump_free_space(cache, bytes);
7946 spin_unlock(&cache->lock);
7948 if (++index < BTRFS_NR_RAID_TYPES)
7950 up_read(&info->groups_sem);
7953 int btrfs_reserve_extent(struct btrfs_root *root,
7954 u64 num_bytes, u64 min_alloc_size,
7955 u64 empty_size, u64 hint_byte,
7956 struct btrfs_key *ins, int is_data, int delalloc)
7958 bool final_tried = num_bytes == min_alloc_size;
7962 flags = btrfs_get_alloc_profile(root, is_data);
7964 WARN_ON(num_bytes < root->sectorsize);
7965 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7967 if (!ret && !is_data) {
7968 btrfs_dec_block_group_reservations(root->fs_info,
7970 } else if (ret == -ENOSPC) {
7971 if (!final_tried && ins->offset) {
7972 num_bytes = min(num_bytes >> 1, ins->offset);
7973 num_bytes = round_down(num_bytes, root->sectorsize);
7974 num_bytes = max(num_bytes, min_alloc_size);
7975 if (num_bytes == min_alloc_size)
7978 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7979 struct btrfs_space_info *sinfo;
7981 sinfo = __find_space_info(root->fs_info, flags);
7982 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7985 dump_space_info(sinfo, num_bytes, 1);
7992 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7994 int pin, int delalloc)
7996 struct btrfs_block_group_cache *cache;
7999 cache = btrfs_lookup_block_group(root->fs_info, start);
8001 btrfs_err(root->fs_info, "Unable to find block group for %llu",
8007 pin_down_extent(root, cache, start, len, 1);
8009 if (btrfs_test_opt(root, DISCARD))
8010 ret = btrfs_discard_extent(root, start, len, NULL);
8011 btrfs_add_free_space(cache, start, len);
8012 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
8013 trace_btrfs_reserved_extent_free(root, start, len);
8016 btrfs_put_block_group(cache);
8020 int btrfs_free_reserved_extent(struct btrfs_root *root,
8021 u64 start, u64 len, int delalloc)
8023 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
8026 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
8029 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
8032 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8033 struct btrfs_root *root,
8034 u64 parent, u64 root_objectid,
8035 u64 flags, u64 owner, u64 offset,
8036 struct btrfs_key *ins, int ref_mod)
8039 struct btrfs_fs_info *fs_info = root->fs_info;
8040 struct btrfs_extent_item *extent_item;
8041 struct btrfs_extent_inline_ref *iref;
8042 struct btrfs_path *path;
8043 struct extent_buffer *leaf;
8048 type = BTRFS_SHARED_DATA_REF_KEY;
8050 type = BTRFS_EXTENT_DATA_REF_KEY;
8052 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
8054 path = btrfs_alloc_path();
8058 path->leave_spinning = 1;
8059 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8062 btrfs_free_path(path);
8066 leaf = path->nodes[0];
8067 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8068 struct btrfs_extent_item);
8069 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
8070 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8071 btrfs_set_extent_flags(leaf, extent_item,
8072 flags | BTRFS_EXTENT_FLAG_DATA);
8074 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8075 btrfs_set_extent_inline_ref_type(leaf, iref, type);
8077 struct btrfs_shared_data_ref *ref;
8078 ref = (struct btrfs_shared_data_ref *)(iref + 1);
8079 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8080 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
8082 struct btrfs_extent_data_ref *ref;
8083 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
8084 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
8085 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
8086 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
8087 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
8090 btrfs_mark_buffer_dirty(path->nodes[0]);
8091 btrfs_free_path(path);
8093 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8098 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
8099 if (ret) { /* -ENOENT, logic error */
8100 btrfs_err(fs_info, "update block group failed for %llu %llu",
8101 ins->objectid, ins->offset);
8104 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
8108 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
8109 struct btrfs_root *root,
8110 u64 parent, u64 root_objectid,
8111 u64 flags, struct btrfs_disk_key *key,
8112 int level, struct btrfs_key *ins)
8115 struct btrfs_fs_info *fs_info = root->fs_info;
8116 struct btrfs_extent_item *extent_item;
8117 struct btrfs_tree_block_info *block_info;
8118 struct btrfs_extent_inline_ref *iref;
8119 struct btrfs_path *path;
8120 struct extent_buffer *leaf;
8121 u32 size = sizeof(*extent_item) + sizeof(*iref);
8122 u64 num_bytes = ins->offset;
8123 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8126 if (!skinny_metadata)
8127 size += sizeof(*block_info);
8129 path = btrfs_alloc_path();
8131 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8136 path->leave_spinning = 1;
8137 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
8140 btrfs_free_path(path);
8141 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
8146 leaf = path->nodes[0];
8147 extent_item = btrfs_item_ptr(leaf, path->slots[0],
8148 struct btrfs_extent_item);
8149 btrfs_set_extent_refs(leaf, extent_item, 1);
8150 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
8151 btrfs_set_extent_flags(leaf, extent_item,
8152 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
8154 if (skinny_metadata) {
8155 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
8156 num_bytes = root->nodesize;
8158 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
8159 btrfs_set_tree_block_key(leaf, block_info, key);
8160 btrfs_set_tree_block_level(leaf, block_info, level);
8161 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
8165 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
8166 btrfs_set_extent_inline_ref_type(leaf, iref,
8167 BTRFS_SHARED_BLOCK_REF_KEY);
8168 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
8170 btrfs_set_extent_inline_ref_type(leaf, iref,
8171 BTRFS_TREE_BLOCK_REF_KEY);
8172 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
8175 btrfs_mark_buffer_dirty(leaf);
8176 btrfs_free_path(path);
8178 ret = remove_from_free_space_tree(trans, fs_info, ins->objectid,
8183 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
8185 if (ret) { /* -ENOENT, logic error */
8186 btrfs_err(fs_info, "update block group failed for %llu %llu",
8187 ins->objectid, ins->offset);
8191 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
8195 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
8196 struct btrfs_root *root,
8197 u64 root_objectid, u64 owner,
8198 u64 offset, u64 ram_bytes,
8199 struct btrfs_key *ins)
8203 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
8205 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
8207 root_objectid, owner, offset,
8208 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
8214 * this is used by the tree logging recovery code. It records that
8215 * an extent has been allocated and makes sure to clear the free
8216 * space cache bits as well
8218 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
8219 struct btrfs_root *root,
8220 u64 root_objectid, u64 owner, u64 offset,
8221 struct btrfs_key *ins)
8224 struct btrfs_block_group_cache *block_group;
8227 * Mixed block groups will exclude before processing the log so we only
8228 * need to do the exclude dance if this fs isn't mixed.
8230 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
8231 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
8236 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
8240 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
8241 RESERVE_ALLOC_NO_ACCOUNT, 0);
8242 BUG_ON(ret); /* logic error */
8243 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
8244 0, owner, offset, ins, 1);
8245 btrfs_put_block_group(block_group);
8249 static struct extent_buffer *
8250 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
8251 u64 bytenr, int level)
8253 struct extent_buffer *buf;
8255 buf = btrfs_find_create_tree_block(root, bytenr);
8259 btrfs_set_header_generation(buf, trans->transid);
8260 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
8261 btrfs_tree_lock(buf);
8262 clean_tree_block(trans, root->fs_info, buf);
8263 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
8265 btrfs_set_lock_blocking(buf);
8266 set_extent_buffer_uptodate(buf);
8268 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
8269 buf->log_index = root->log_transid % 2;
8271 * we allow two log transactions at a time, use different
8272 * EXENT bit to differentiate dirty pages.
8274 if (buf->log_index == 0)
8275 set_extent_dirty(&root->dirty_log_pages, buf->start,
8276 buf->start + buf->len - 1, GFP_NOFS);
8278 set_extent_new(&root->dirty_log_pages, buf->start,
8279 buf->start + buf->len - 1);
8281 buf->log_index = -1;
8282 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
8283 buf->start + buf->len - 1, GFP_NOFS);
8285 trans->dirty = true;
8286 /* this returns a buffer locked for blocking */
8290 static struct btrfs_block_rsv *
8291 use_block_rsv(struct btrfs_trans_handle *trans,
8292 struct btrfs_root *root, u32 blocksize)
8294 struct btrfs_block_rsv *block_rsv;
8295 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
8297 bool global_updated = false;
8299 block_rsv = get_block_rsv(trans, root);
8301 if (unlikely(block_rsv->size == 0))
8304 ret = block_rsv_use_bytes(block_rsv, blocksize);
8308 if (block_rsv->failfast)
8309 return ERR_PTR(ret);
8311 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
8312 global_updated = true;
8313 update_global_block_rsv(root->fs_info);
8317 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
8318 static DEFINE_RATELIMIT_STATE(_rs,
8319 DEFAULT_RATELIMIT_INTERVAL * 10,
8320 /*DEFAULT_RATELIMIT_BURST*/ 1);
8321 if (__ratelimit(&_rs))
8323 "BTRFS: block rsv returned %d\n", ret);
8326 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
8327 BTRFS_RESERVE_NO_FLUSH);
8331 * If we couldn't reserve metadata bytes try and use some from
8332 * the global reserve if its space type is the same as the global
8335 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
8336 block_rsv->space_info == global_rsv->space_info) {
8337 ret = block_rsv_use_bytes(global_rsv, blocksize);
8341 return ERR_PTR(ret);
8344 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
8345 struct btrfs_block_rsv *block_rsv, u32 blocksize)
8347 block_rsv_add_bytes(block_rsv, blocksize, 0);
8348 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
8352 * finds a free extent and does all the dirty work required for allocation
8353 * returns the tree buffer or an ERR_PTR on error.
8355 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
8356 struct btrfs_root *root,
8357 u64 parent, u64 root_objectid,
8358 struct btrfs_disk_key *key, int level,
8359 u64 hint, u64 empty_size)
8361 struct btrfs_key ins;
8362 struct btrfs_block_rsv *block_rsv;
8363 struct extent_buffer *buf;
8364 struct btrfs_delayed_extent_op *extent_op;
8367 u32 blocksize = root->nodesize;
8368 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
8371 if (btrfs_test_is_dummy_root(root)) {
8372 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
8375 root->alloc_bytenr += blocksize;
8379 block_rsv = use_block_rsv(trans, root, blocksize);
8380 if (IS_ERR(block_rsv))
8381 return ERR_CAST(block_rsv);
8383 ret = btrfs_reserve_extent(root, blocksize, blocksize,
8384 empty_size, hint, &ins, 0, 0);
8388 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
8391 goto out_free_reserved;
8394 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
8396 parent = ins.objectid;
8397 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
8401 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
8402 extent_op = btrfs_alloc_delayed_extent_op();
8408 memcpy(&extent_op->key, key, sizeof(extent_op->key));
8410 memset(&extent_op->key, 0, sizeof(extent_op->key));
8411 extent_op->flags_to_set = flags;
8412 extent_op->update_key = skinny_metadata ? false : true;
8413 extent_op->update_flags = true;
8414 extent_op->is_data = false;
8415 extent_op->level = level;
8417 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
8418 ins.objectid, ins.offset,
8419 parent, root_objectid, level,
8420 BTRFS_ADD_DELAYED_EXTENT,
8423 goto out_free_delayed;
8428 btrfs_free_delayed_extent_op(extent_op);
8430 free_extent_buffer(buf);
8432 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8434 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8435 return ERR_PTR(ret);
8438 struct walk_control {
8439 u64 refs[BTRFS_MAX_LEVEL];
8440 u64 flags[BTRFS_MAX_LEVEL];
8441 struct btrfs_key update_progress;
8452 #define DROP_REFERENCE 1
8453 #define UPDATE_BACKREF 2
8455 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8456 struct btrfs_root *root,
8457 struct walk_control *wc,
8458 struct btrfs_path *path)
8466 struct btrfs_key key;
8467 struct extent_buffer *eb;
8472 if (path->slots[wc->level] < wc->reada_slot) {
8473 wc->reada_count = wc->reada_count * 2 / 3;
8474 wc->reada_count = max(wc->reada_count, 2);
8476 wc->reada_count = wc->reada_count * 3 / 2;
8477 wc->reada_count = min_t(int, wc->reada_count,
8478 BTRFS_NODEPTRS_PER_BLOCK(root));
8481 eb = path->nodes[wc->level];
8482 nritems = btrfs_header_nritems(eb);
8483 blocksize = root->nodesize;
8485 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8486 if (nread >= wc->reada_count)
8490 bytenr = btrfs_node_blockptr(eb, slot);
8491 generation = btrfs_node_ptr_generation(eb, slot);
8493 if (slot == path->slots[wc->level])
8496 if (wc->stage == UPDATE_BACKREF &&
8497 generation <= root->root_key.offset)
8500 /* We don't lock the tree block, it's OK to be racy here */
8501 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8502 wc->level - 1, 1, &refs,
8504 /* We don't care about errors in readahead. */
8509 if (wc->stage == DROP_REFERENCE) {
8513 if (wc->level == 1 &&
8514 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8516 if (!wc->update_ref ||
8517 generation <= root->root_key.offset)
8519 btrfs_node_key_to_cpu(eb, &key, slot);
8520 ret = btrfs_comp_cpu_keys(&key,
8521 &wc->update_progress);
8525 if (wc->level == 1 &&
8526 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8530 readahead_tree_block(root, bytenr);
8533 wc->reada_slot = slot;
8537 * These may not be seen by the usual inc/dec ref code so we have to
8540 static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8541 struct btrfs_root *root, u64 bytenr,
8544 struct btrfs_qgroup_extent_record *qrecord;
8545 struct btrfs_delayed_ref_root *delayed_refs;
8547 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8551 qrecord->bytenr = bytenr;
8552 qrecord->num_bytes = num_bytes;
8553 qrecord->old_roots = NULL;
8555 delayed_refs = &trans->transaction->delayed_refs;
8556 spin_lock(&delayed_refs->lock);
8557 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8559 spin_unlock(&delayed_refs->lock);
8564 static int account_leaf_items(struct btrfs_trans_handle *trans,
8565 struct btrfs_root *root,
8566 struct extent_buffer *eb)
8568 int nr = btrfs_header_nritems(eb);
8569 int i, extent_type, ret;
8570 struct btrfs_key key;
8571 struct btrfs_file_extent_item *fi;
8572 u64 bytenr, num_bytes;
8574 /* We can be called directly from walk_up_proc() */
8575 if (!root->fs_info->quota_enabled)
8578 for (i = 0; i < nr; i++) {
8579 btrfs_item_key_to_cpu(eb, &key, i);
8581 if (key.type != BTRFS_EXTENT_DATA_KEY)
8584 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8585 /* filter out non qgroup-accountable extents */
8586 extent_type = btrfs_file_extent_type(eb, fi);
8588 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8591 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8595 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8597 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8605 * Walk up the tree from the bottom, freeing leaves and any interior
8606 * nodes which have had all slots visited. If a node (leaf or
8607 * interior) is freed, the node above it will have it's slot
8608 * incremented. The root node will never be freed.
8610 * At the end of this function, we should have a path which has all
8611 * slots incremented to the next position for a search. If we need to
8612 * read a new node it will be NULL and the node above it will have the
8613 * correct slot selected for a later read.
8615 * If we increment the root nodes slot counter past the number of
8616 * elements, 1 is returned to signal completion of the search.
8618 static int adjust_slots_upwards(struct btrfs_root *root,
8619 struct btrfs_path *path, int root_level)
8623 struct extent_buffer *eb;
8625 if (root_level == 0)
8628 while (level <= root_level) {
8629 eb = path->nodes[level];
8630 nr = btrfs_header_nritems(eb);
8631 path->slots[level]++;
8632 slot = path->slots[level];
8633 if (slot >= nr || level == 0) {
8635 * Don't free the root - we will detect this
8636 * condition after our loop and return a
8637 * positive value for caller to stop walking the tree.
8639 if (level != root_level) {
8640 btrfs_tree_unlock_rw(eb, path->locks[level]);
8641 path->locks[level] = 0;
8643 free_extent_buffer(eb);
8644 path->nodes[level] = NULL;
8645 path->slots[level] = 0;
8649 * We have a valid slot to walk back down
8650 * from. Stop here so caller can process these
8659 eb = path->nodes[root_level];
8660 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8667 * root_eb is the subtree root and is locked before this function is called.
8669 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8670 struct btrfs_root *root,
8671 struct extent_buffer *root_eb,
8677 struct extent_buffer *eb = root_eb;
8678 struct btrfs_path *path = NULL;
8680 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8681 BUG_ON(root_eb == NULL);
8683 if (!root->fs_info->quota_enabled)
8686 if (!extent_buffer_uptodate(root_eb)) {
8687 ret = btrfs_read_buffer(root_eb, root_gen);
8692 if (root_level == 0) {
8693 ret = account_leaf_items(trans, root, root_eb);
8697 path = btrfs_alloc_path();
8702 * Walk down the tree. Missing extent blocks are filled in as
8703 * we go. Metadata is accounted every time we read a new
8706 * When we reach a leaf, we account for file extent items in it,
8707 * walk back up the tree (adjusting slot pointers as we go)
8708 * and restart the search process.
8710 extent_buffer_get(root_eb); /* For path */
8711 path->nodes[root_level] = root_eb;
8712 path->slots[root_level] = 0;
8713 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8716 while (level >= 0) {
8717 if (path->nodes[level] == NULL) {
8722 /* We need to get child blockptr/gen from
8723 * parent before we can read it. */
8724 eb = path->nodes[level + 1];
8725 parent_slot = path->slots[level + 1];
8726 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8727 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8729 eb = read_tree_block(root, child_bytenr, child_gen);
8733 } else if (!extent_buffer_uptodate(eb)) {
8734 free_extent_buffer(eb);
8739 path->nodes[level] = eb;
8740 path->slots[level] = 0;
8742 btrfs_tree_read_lock(eb);
8743 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8744 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8746 ret = record_one_subtree_extent(trans, root, child_bytenr,
8753 ret = account_leaf_items(trans, root, path->nodes[level]);
8757 /* Nonzero return here means we completed our search */
8758 ret = adjust_slots_upwards(root, path, root_level);
8762 /* Restart search with new slots */
8771 btrfs_free_path(path);
8777 * helper to process tree block while walking down the tree.
8779 * when wc->stage == UPDATE_BACKREF, this function updates
8780 * back refs for pointers in the block.
8782 * NOTE: return value 1 means we should stop walking down.
8784 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8785 struct btrfs_root *root,
8786 struct btrfs_path *path,
8787 struct walk_control *wc, int lookup_info)
8789 int level = wc->level;
8790 struct extent_buffer *eb = path->nodes[level];
8791 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8794 if (wc->stage == UPDATE_BACKREF &&
8795 btrfs_header_owner(eb) != root->root_key.objectid)
8799 * when reference count of tree block is 1, it won't increase
8800 * again. once full backref flag is set, we never clear it.
8803 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8804 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8805 BUG_ON(!path->locks[level]);
8806 ret = btrfs_lookup_extent_info(trans, root,
8807 eb->start, level, 1,
8810 BUG_ON(ret == -ENOMEM);
8813 BUG_ON(wc->refs[level] == 0);
8816 if (wc->stage == DROP_REFERENCE) {
8817 if (wc->refs[level] > 1)
8820 if (path->locks[level] && !wc->keep_locks) {
8821 btrfs_tree_unlock_rw(eb, path->locks[level]);
8822 path->locks[level] = 0;
8827 /* wc->stage == UPDATE_BACKREF */
8828 if (!(wc->flags[level] & flag)) {
8829 BUG_ON(!path->locks[level]);
8830 ret = btrfs_inc_ref(trans, root, eb, 1);
8831 BUG_ON(ret); /* -ENOMEM */
8832 ret = btrfs_dec_ref(trans, root, eb, 0);
8833 BUG_ON(ret); /* -ENOMEM */
8834 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8836 btrfs_header_level(eb), 0);
8837 BUG_ON(ret); /* -ENOMEM */
8838 wc->flags[level] |= flag;
8842 * the block is shared by multiple trees, so it's not good to
8843 * keep the tree lock
8845 if (path->locks[level] && level > 0) {
8846 btrfs_tree_unlock_rw(eb, path->locks[level]);
8847 path->locks[level] = 0;
8853 * helper to process tree block pointer.
8855 * when wc->stage == DROP_REFERENCE, this function checks
8856 * reference count of the block pointed to. if the block
8857 * is shared and we need update back refs for the subtree
8858 * rooted at the block, this function changes wc->stage to
8859 * UPDATE_BACKREF. if the block is shared and there is no
8860 * need to update back, this function drops the reference
8863 * NOTE: return value 1 means we should stop walking down.
8865 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8866 struct btrfs_root *root,
8867 struct btrfs_path *path,
8868 struct walk_control *wc, int *lookup_info)
8874 struct btrfs_key key;
8875 struct extent_buffer *next;
8876 int level = wc->level;
8879 bool need_account = false;
8881 generation = btrfs_node_ptr_generation(path->nodes[level],
8882 path->slots[level]);
8884 * if the lower level block was created before the snapshot
8885 * was created, we know there is no need to update back refs
8888 if (wc->stage == UPDATE_BACKREF &&
8889 generation <= root->root_key.offset) {
8894 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8895 blocksize = root->nodesize;
8897 next = btrfs_find_tree_block(root->fs_info, bytenr);
8899 next = btrfs_find_create_tree_block(root, bytenr);
8901 return PTR_ERR(next);
8903 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8907 btrfs_tree_lock(next);
8908 btrfs_set_lock_blocking(next);
8910 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8911 &wc->refs[level - 1],
8912 &wc->flags[level - 1]);
8914 btrfs_tree_unlock(next);
8918 if (unlikely(wc->refs[level - 1] == 0)) {
8919 btrfs_err(root->fs_info, "Missing references.");
8924 if (wc->stage == DROP_REFERENCE) {
8925 if (wc->refs[level - 1] > 1) {
8926 need_account = true;
8928 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8931 if (!wc->update_ref ||
8932 generation <= root->root_key.offset)
8935 btrfs_node_key_to_cpu(path->nodes[level], &key,
8936 path->slots[level]);
8937 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8941 wc->stage = UPDATE_BACKREF;
8942 wc->shared_level = level - 1;
8946 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8950 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8951 btrfs_tree_unlock(next);
8952 free_extent_buffer(next);
8958 if (reada && level == 1)
8959 reada_walk_down(trans, root, wc, path);
8960 next = read_tree_block(root, bytenr, generation);
8962 return PTR_ERR(next);
8963 } else if (!extent_buffer_uptodate(next)) {
8964 free_extent_buffer(next);
8967 btrfs_tree_lock(next);
8968 btrfs_set_lock_blocking(next);
8972 BUG_ON(level != btrfs_header_level(next));
8973 path->nodes[level] = next;
8974 path->slots[level] = 0;
8975 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8981 wc->refs[level - 1] = 0;
8982 wc->flags[level - 1] = 0;
8983 if (wc->stage == DROP_REFERENCE) {
8984 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8985 parent = path->nodes[level]->start;
8987 BUG_ON(root->root_key.objectid !=
8988 btrfs_header_owner(path->nodes[level]));
8993 ret = account_shared_subtree(trans, root, next,
8994 generation, level - 1);
8996 btrfs_err_rl(root->fs_info,
8998 "%d accounting shared subtree. Quota "
8999 "is out of sync, rescan required.",
9003 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
9004 root->root_key.objectid, level - 1, 0);
9005 BUG_ON(ret); /* -ENOMEM */
9007 btrfs_tree_unlock(next);
9008 free_extent_buffer(next);
9014 * helper to process tree block while walking up the tree.
9016 * when wc->stage == DROP_REFERENCE, this function drops
9017 * reference count on the block.
9019 * when wc->stage == UPDATE_BACKREF, this function changes
9020 * wc->stage back to DROP_REFERENCE if we changed wc->stage
9021 * to UPDATE_BACKREF previously while processing the block.
9023 * NOTE: return value 1 means we should stop walking up.
9025 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
9026 struct btrfs_root *root,
9027 struct btrfs_path *path,
9028 struct walk_control *wc)
9031 int level = wc->level;
9032 struct extent_buffer *eb = path->nodes[level];
9035 if (wc->stage == UPDATE_BACKREF) {
9036 BUG_ON(wc->shared_level < level);
9037 if (level < wc->shared_level)
9040 ret = find_next_key(path, level + 1, &wc->update_progress);
9044 wc->stage = DROP_REFERENCE;
9045 wc->shared_level = -1;
9046 path->slots[level] = 0;
9049 * check reference count again if the block isn't locked.
9050 * we should start walking down the tree again if reference
9053 if (!path->locks[level]) {
9055 btrfs_tree_lock(eb);
9056 btrfs_set_lock_blocking(eb);
9057 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9059 ret = btrfs_lookup_extent_info(trans, root,
9060 eb->start, level, 1,
9064 btrfs_tree_unlock_rw(eb, path->locks[level]);
9065 path->locks[level] = 0;
9068 BUG_ON(wc->refs[level] == 0);
9069 if (wc->refs[level] == 1) {
9070 btrfs_tree_unlock_rw(eb, path->locks[level]);
9071 path->locks[level] = 0;
9077 /* wc->stage == DROP_REFERENCE */
9078 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
9080 if (wc->refs[level] == 1) {
9082 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9083 ret = btrfs_dec_ref(trans, root, eb, 1);
9085 ret = btrfs_dec_ref(trans, root, eb, 0);
9086 BUG_ON(ret); /* -ENOMEM */
9087 ret = account_leaf_items(trans, root, eb);
9089 btrfs_err_rl(root->fs_info,
9091 "%d accounting leaf items. Quota "
9092 "is out of sync, rescan required.",
9096 /* make block locked assertion in clean_tree_block happy */
9097 if (!path->locks[level] &&
9098 btrfs_header_generation(eb) == trans->transid) {
9099 btrfs_tree_lock(eb);
9100 btrfs_set_lock_blocking(eb);
9101 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9103 clean_tree_block(trans, root->fs_info, eb);
9106 if (eb == root->node) {
9107 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9110 BUG_ON(root->root_key.objectid !=
9111 btrfs_header_owner(eb));
9113 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
9114 parent = path->nodes[level + 1]->start;
9116 BUG_ON(root->root_key.objectid !=
9117 btrfs_header_owner(path->nodes[level + 1]));
9120 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
9122 wc->refs[level] = 0;
9123 wc->flags[level] = 0;
9127 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
9128 struct btrfs_root *root,
9129 struct btrfs_path *path,
9130 struct walk_control *wc)
9132 int level = wc->level;
9133 int lookup_info = 1;
9136 while (level >= 0) {
9137 ret = walk_down_proc(trans, root, path, wc, lookup_info);
9144 if (path->slots[level] >=
9145 btrfs_header_nritems(path->nodes[level]))
9148 ret = do_walk_down(trans, root, path, wc, &lookup_info);
9150 path->slots[level]++;
9159 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
9160 struct btrfs_root *root,
9161 struct btrfs_path *path,
9162 struct walk_control *wc, int max_level)
9164 int level = wc->level;
9167 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
9168 while (level < max_level && path->nodes[level]) {
9170 if (path->slots[level] + 1 <
9171 btrfs_header_nritems(path->nodes[level])) {
9172 path->slots[level]++;
9175 ret = walk_up_proc(trans, root, path, wc);
9179 if (path->locks[level]) {
9180 btrfs_tree_unlock_rw(path->nodes[level],
9181 path->locks[level]);
9182 path->locks[level] = 0;
9184 free_extent_buffer(path->nodes[level]);
9185 path->nodes[level] = NULL;
9193 * drop a subvolume tree.
9195 * this function traverses the tree freeing any blocks that only
9196 * referenced by the tree.
9198 * when a shared tree block is found. this function decreases its
9199 * reference count by one. if update_ref is true, this function
9200 * also make sure backrefs for the shared block and all lower level
9201 * blocks are properly updated.
9203 * If called with for_reloc == 0, may exit early with -EAGAIN
9205 int btrfs_drop_snapshot(struct btrfs_root *root,
9206 struct btrfs_block_rsv *block_rsv, int update_ref,
9209 struct btrfs_path *path;
9210 struct btrfs_trans_handle *trans;
9211 struct btrfs_root *tree_root = root->fs_info->tree_root;
9212 struct btrfs_root_item *root_item = &root->root_item;
9213 struct walk_control *wc;
9214 struct btrfs_key key;
9218 bool root_dropped = false;
9220 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
9222 path = btrfs_alloc_path();
9228 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9230 btrfs_free_path(path);
9235 trans = btrfs_start_transaction(tree_root, 0);
9236 if (IS_ERR(trans)) {
9237 err = PTR_ERR(trans);
9242 trans->block_rsv = block_rsv;
9244 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
9245 level = btrfs_header_level(root->node);
9246 path->nodes[level] = btrfs_lock_root_node(root);
9247 btrfs_set_lock_blocking(path->nodes[level]);
9248 path->slots[level] = 0;
9249 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9250 memset(&wc->update_progress, 0,
9251 sizeof(wc->update_progress));
9253 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
9254 memcpy(&wc->update_progress, &key,
9255 sizeof(wc->update_progress));
9257 level = root_item->drop_level;
9259 path->lowest_level = level;
9260 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
9261 path->lowest_level = 0;
9269 * unlock our path, this is safe because only this
9270 * function is allowed to delete this snapshot
9272 btrfs_unlock_up_safe(path, 0);
9274 level = btrfs_header_level(root->node);
9276 btrfs_tree_lock(path->nodes[level]);
9277 btrfs_set_lock_blocking(path->nodes[level]);
9278 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9280 ret = btrfs_lookup_extent_info(trans, root,
9281 path->nodes[level]->start,
9282 level, 1, &wc->refs[level],
9288 BUG_ON(wc->refs[level] == 0);
9290 if (level == root_item->drop_level)
9293 btrfs_tree_unlock(path->nodes[level]);
9294 path->locks[level] = 0;
9295 WARN_ON(wc->refs[level] != 1);
9301 wc->shared_level = -1;
9302 wc->stage = DROP_REFERENCE;
9303 wc->update_ref = update_ref;
9305 wc->for_reloc = for_reloc;
9306 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9310 ret = walk_down_tree(trans, root, path, wc);
9316 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
9323 BUG_ON(wc->stage != DROP_REFERENCE);
9327 if (wc->stage == DROP_REFERENCE) {
9329 btrfs_node_key(path->nodes[level],
9330 &root_item->drop_progress,
9331 path->slots[level]);
9332 root_item->drop_level = level;
9335 BUG_ON(wc->level == 0);
9336 if (btrfs_should_end_transaction(trans, tree_root) ||
9337 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
9338 ret = btrfs_update_root(trans, tree_root,
9342 btrfs_abort_transaction(trans, tree_root, ret);
9347 btrfs_end_transaction_throttle(trans, tree_root);
9348 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
9349 pr_debug("BTRFS: drop snapshot early exit\n");
9354 trans = btrfs_start_transaction(tree_root, 0);
9355 if (IS_ERR(trans)) {
9356 err = PTR_ERR(trans);
9360 trans->block_rsv = block_rsv;
9363 btrfs_release_path(path);
9367 ret = btrfs_del_root(trans, tree_root, &root->root_key);
9369 btrfs_abort_transaction(trans, tree_root, ret);
9373 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
9374 ret = btrfs_find_root(tree_root, &root->root_key, path,
9377 btrfs_abort_transaction(trans, tree_root, ret);
9380 } else if (ret > 0) {
9381 /* if we fail to delete the orphan item this time
9382 * around, it'll get picked up the next time.
9384 * The most common failure here is just -ENOENT.
9386 btrfs_del_orphan_item(trans, tree_root,
9387 root->root_key.objectid);
9391 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
9392 btrfs_add_dropped_root(trans, root);
9394 free_extent_buffer(root->node);
9395 free_extent_buffer(root->commit_root);
9396 btrfs_put_fs_root(root);
9398 root_dropped = true;
9400 btrfs_end_transaction_throttle(trans, tree_root);
9403 btrfs_free_path(path);
9406 * So if we need to stop dropping the snapshot for whatever reason we
9407 * need to make sure to add it back to the dead root list so that we
9408 * keep trying to do the work later. This also cleans up roots if we
9409 * don't have it in the radix (like when we recover after a power fail
9410 * or unmount) so we don't leak memory.
9412 if (!for_reloc && root_dropped == false)
9413 btrfs_add_dead_root(root);
9414 if (err && err != -EAGAIN)
9415 btrfs_handle_fs_error(root->fs_info, err, NULL);
9420 * drop subtree rooted at tree block 'node'.
9422 * NOTE: this function will unlock and release tree block 'node'
9423 * only used by relocation code
9425 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9426 struct btrfs_root *root,
9427 struct extent_buffer *node,
9428 struct extent_buffer *parent)
9430 struct btrfs_path *path;
9431 struct walk_control *wc;
9437 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9439 path = btrfs_alloc_path();
9443 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9445 btrfs_free_path(path);
9449 btrfs_assert_tree_locked(parent);
9450 parent_level = btrfs_header_level(parent);
9451 extent_buffer_get(parent);
9452 path->nodes[parent_level] = parent;
9453 path->slots[parent_level] = btrfs_header_nritems(parent);
9455 btrfs_assert_tree_locked(node);
9456 level = btrfs_header_level(node);
9457 path->nodes[level] = node;
9458 path->slots[level] = 0;
9459 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9461 wc->refs[parent_level] = 1;
9462 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9464 wc->shared_level = -1;
9465 wc->stage = DROP_REFERENCE;
9469 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9472 wret = walk_down_tree(trans, root, path, wc);
9478 wret = walk_up_tree(trans, root, path, wc, parent_level);
9486 btrfs_free_path(path);
9490 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9496 * if restripe for this chunk_type is on pick target profile and
9497 * return, otherwise do the usual balance
9499 stripped = get_restripe_target(root->fs_info, flags);
9501 return extended_to_chunk(stripped);
9503 num_devices = root->fs_info->fs_devices->rw_devices;
9505 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9506 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9507 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9509 if (num_devices == 1) {
9510 stripped |= BTRFS_BLOCK_GROUP_DUP;
9511 stripped = flags & ~stripped;
9513 /* turn raid0 into single device chunks */
9514 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9517 /* turn mirroring into duplication */
9518 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9519 BTRFS_BLOCK_GROUP_RAID10))
9520 return stripped | BTRFS_BLOCK_GROUP_DUP;
9522 /* they already had raid on here, just return */
9523 if (flags & stripped)
9526 stripped |= BTRFS_BLOCK_GROUP_DUP;
9527 stripped = flags & ~stripped;
9529 /* switch duplicated blocks with raid1 */
9530 if (flags & BTRFS_BLOCK_GROUP_DUP)
9531 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9533 /* this is drive concat, leave it alone */
9539 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9541 struct btrfs_space_info *sinfo = cache->space_info;
9543 u64 min_allocable_bytes;
9547 * We need some metadata space and system metadata space for
9548 * allocating chunks in some corner cases until we force to set
9549 * it to be readonly.
9552 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9554 min_allocable_bytes = SZ_1M;
9556 min_allocable_bytes = 0;
9558 spin_lock(&sinfo->lock);
9559 spin_lock(&cache->lock);
9567 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9568 cache->bytes_super - btrfs_block_group_used(&cache->item);
9570 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9571 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9572 min_allocable_bytes <= sinfo->total_bytes) {
9573 sinfo->bytes_readonly += num_bytes;
9575 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9579 spin_unlock(&cache->lock);
9580 spin_unlock(&sinfo->lock);
9584 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9585 struct btrfs_block_group_cache *cache)
9588 struct btrfs_trans_handle *trans;
9593 trans = btrfs_join_transaction(root);
9595 return PTR_ERR(trans);
9598 * we're not allowed to set block groups readonly after the dirty
9599 * block groups cache has started writing. If it already started,
9600 * back off and let this transaction commit
9602 mutex_lock(&root->fs_info->ro_block_group_mutex);
9603 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9604 u64 transid = trans->transid;
9606 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9607 btrfs_end_transaction(trans, root);
9609 ret = btrfs_wait_for_commit(root, transid);
9616 * if we are changing raid levels, try to allocate a corresponding
9617 * block group with the new raid level.
9619 alloc_flags = update_block_group_flags(root, cache->flags);
9620 if (alloc_flags != cache->flags) {
9621 ret = do_chunk_alloc(trans, root, alloc_flags,
9624 * ENOSPC is allowed here, we may have enough space
9625 * already allocated at the new raid level to
9634 ret = inc_block_group_ro(cache, 0);
9637 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9638 ret = do_chunk_alloc(trans, root, alloc_flags,
9642 ret = inc_block_group_ro(cache, 0);
9644 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9645 alloc_flags = update_block_group_flags(root, cache->flags);
9646 lock_chunks(root->fs_info->chunk_root);
9647 check_system_chunk(trans, root, alloc_flags);
9648 unlock_chunks(root->fs_info->chunk_root);
9650 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9652 btrfs_end_transaction(trans, root);
9656 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9657 struct btrfs_root *root, u64 type)
9659 u64 alloc_flags = get_alloc_profile(root, type);
9660 return do_chunk_alloc(trans, root, alloc_flags,
9665 * helper to account the unused space of all the readonly block group in the
9666 * space_info. takes mirrors into account.
9668 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9670 struct btrfs_block_group_cache *block_group;
9674 /* It's df, we don't care if it's racy */
9675 if (list_empty(&sinfo->ro_bgs))
9678 spin_lock(&sinfo->lock);
9679 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9680 spin_lock(&block_group->lock);
9682 if (!block_group->ro) {
9683 spin_unlock(&block_group->lock);
9687 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9688 BTRFS_BLOCK_GROUP_RAID10 |
9689 BTRFS_BLOCK_GROUP_DUP))
9694 free_bytes += (block_group->key.offset -
9695 btrfs_block_group_used(&block_group->item)) *
9698 spin_unlock(&block_group->lock);
9700 spin_unlock(&sinfo->lock);
9705 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9706 struct btrfs_block_group_cache *cache)
9708 struct btrfs_space_info *sinfo = cache->space_info;
9713 spin_lock(&sinfo->lock);
9714 spin_lock(&cache->lock);
9716 num_bytes = cache->key.offset - cache->reserved -
9717 cache->pinned - cache->bytes_super -
9718 btrfs_block_group_used(&cache->item);
9719 sinfo->bytes_readonly -= num_bytes;
9720 list_del_init(&cache->ro_list);
9722 spin_unlock(&cache->lock);
9723 spin_unlock(&sinfo->lock);
9727 * checks to see if its even possible to relocate this block group.
9729 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9730 * ok to go ahead and try.
9732 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9734 struct btrfs_block_group_cache *block_group;
9735 struct btrfs_space_info *space_info;
9736 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9737 struct btrfs_device *device;
9738 struct btrfs_trans_handle *trans;
9748 debug = btrfs_test_opt(root, ENOSPC_DEBUG);
9750 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9752 /* odd, couldn't find the block group, leave it alone */
9755 btrfs_warn(root->fs_info,
9756 "can't find block group for bytenr %llu",
9761 min_free = btrfs_block_group_used(&block_group->item);
9763 /* no bytes used, we're good */
9767 space_info = block_group->space_info;
9768 spin_lock(&space_info->lock);
9770 full = space_info->full;
9773 * if this is the last block group we have in this space, we can't
9774 * relocate it unless we're able to allocate a new chunk below.
9776 * Otherwise, we need to make sure we have room in the space to handle
9777 * all of the extents from this block group. If we can, we're good
9779 if ((space_info->total_bytes != block_group->key.offset) &&
9780 (space_info->bytes_used + space_info->bytes_reserved +
9781 space_info->bytes_pinned + space_info->bytes_readonly +
9782 min_free < space_info->total_bytes)) {
9783 spin_unlock(&space_info->lock);
9786 spin_unlock(&space_info->lock);
9789 * ok we don't have enough space, but maybe we have free space on our
9790 * devices to allocate new chunks for relocation, so loop through our
9791 * alloc devices and guess if we have enough space. if this block
9792 * group is going to be restriped, run checks against the target
9793 * profile instead of the current one.
9805 target = get_restripe_target(root->fs_info, block_group->flags);
9807 index = __get_raid_index(extended_to_chunk(target));
9810 * this is just a balance, so if we were marked as full
9811 * we know there is no space for a new chunk
9815 btrfs_warn(root->fs_info,
9816 "no space to alloc new chunk for block group %llu",
9817 block_group->key.objectid);
9821 index = get_block_group_index(block_group);
9824 if (index == BTRFS_RAID_RAID10) {
9828 } else if (index == BTRFS_RAID_RAID1) {
9830 } else if (index == BTRFS_RAID_DUP) {
9833 } else if (index == BTRFS_RAID_RAID0) {
9834 dev_min = fs_devices->rw_devices;
9835 min_free = div64_u64(min_free, dev_min);
9838 /* We need to do this so that we can look at pending chunks */
9839 trans = btrfs_join_transaction(root);
9840 if (IS_ERR(trans)) {
9841 ret = PTR_ERR(trans);
9845 mutex_lock(&root->fs_info->chunk_mutex);
9846 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9850 * check to make sure we can actually find a chunk with enough
9851 * space to fit our block group in.
9853 if (device->total_bytes > device->bytes_used + min_free &&
9854 !device->is_tgtdev_for_dev_replace) {
9855 ret = find_free_dev_extent(trans, device, min_free,
9860 if (dev_nr >= dev_min)
9866 if (debug && ret == -1)
9867 btrfs_warn(root->fs_info,
9868 "no space to allocate a new chunk for block group %llu",
9869 block_group->key.objectid);
9870 mutex_unlock(&root->fs_info->chunk_mutex);
9871 btrfs_end_transaction(trans, root);
9873 btrfs_put_block_group(block_group);
9877 static int find_first_block_group(struct btrfs_root *root,
9878 struct btrfs_path *path, struct btrfs_key *key)
9881 struct btrfs_key found_key;
9882 struct extent_buffer *leaf;
9885 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9890 slot = path->slots[0];
9891 leaf = path->nodes[0];
9892 if (slot >= btrfs_header_nritems(leaf)) {
9893 ret = btrfs_next_leaf(root, path);
9900 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9902 if (found_key.objectid >= key->objectid &&
9903 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9913 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9915 struct btrfs_block_group_cache *block_group;
9919 struct inode *inode;
9921 block_group = btrfs_lookup_first_block_group(info, last);
9922 while (block_group) {
9923 spin_lock(&block_group->lock);
9924 if (block_group->iref)
9926 spin_unlock(&block_group->lock);
9927 block_group = next_block_group(info->tree_root,
9937 inode = block_group->inode;
9938 block_group->iref = 0;
9939 block_group->inode = NULL;
9940 spin_unlock(&block_group->lock);
9942 last = block_group->key.objectid + block_group->key.offset;
9943 btrfs_put_block_group(block_group);
9947 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9949 struct btrfs_block_group_cache *block_group;
9950 struct btrfs_space_info *space_info;
9951 struct btrfs_caching_control *caching_ctl;
9954 down_write(&info->commit_root_sem);
9955 while (!list_empty(&info->caching_block_groups)) {
9956 caching_ctl = list_entry(info->caching_block_groups.next,
9957 struct btrfs_caching_control, list);
9958 list_del(&caching_ctl->list);
9959 put_caching_control(caching_ctl);
9961 up_write(&info->commit_root_sem);
9963 spin_lock(&info->unused_bgs_lock);
9964 while (!list_empty(&info->unused_bgs)) {
9965 block_group = list_first_entry(&info->unused_bgs,
9966 struct btrfs_block_group_cache,
9968 list_del_init(&block_group->bg_list);
9969 btrfs_put_block_group(block_group);
9971 spin_unlock(&info->unused_bgs_lock);
9973 spin_lock(&info->block_group_cache_lock);
9974 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9975 block_group = rb_entry(n, struct btrfs_block_group_cache,
9977 rb_erase(&block_group->cache_node,
9978 &info->block_group_cache_tree);
9979 RB_CLEAR_NODE(&block_group->cache_node);
9980 spin_unlock(&info->block_group_cache_lock);
9982 down_write(&block_group->space_info->groups_sem);
9983 list_del(&block_group->list);
9984 up_write(&block_group->space_info->groups_sem);
9986 if (block_group->cached == BTRFS_CACHE_STARTED)
9987 wait_block_group_cache_done(block_group);
9990 * We haven't cached this block group, which means we could
9991 * possibly have excluded extents on this block group.
9993 if (block_group->cached == BTRFS_CACHE_NO ||
9994 block_group->cached == BTRFS_CACHE_ERROR)
9995 free_excluded_extents(info->extent_root, block_group);
9997 btrfs_remove_free_space_cache(block_group);
9998 btrfs_put_block_group(block_group);
10000 spin_lock(&info->block_group_cache_lock);
10002 spin_unlock(&info->block_group_cache_lock);
10004 /* now that all the block groups are freed, go through and
10005 * free all the space_info structs. This is only called during
10006 * the final stages of unmount, and so we know nobody is
10007 * using them. We call synchronize_rcu() once before we start,
10008 * just to be on the safe side.
10012 release_global_block_rsv(info);
10014 while (!list_empty(&info->space_info)) {
10017 space_info = list_entry(info->space_info.next,
10018 struct btrfs_space_info,
10022 * Do not hide this behind enospc_debug, this is actually
10023 * important and indicates a real bug if this happens.
10025 if (WARN_ON(space_info->bytes_pinned > 0 ||
10026 space_info->bytes_reserved > 0 ||
10027 space_info->bytes_may_use > 0))
10028 dump_space_info(space_info, 0, 0);
10029 list_del(&space_info->list);
10030 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
10031 struct kobject *kobj;
10032 kobj = space_info->block_group_kobjs[i];
10033 space_info->block_group_kobjs[i] = NULL;
10039 kobject_del(&space_info->kobj);
10040 kobject_put(&space_info->kobj);
10045 static void __link_block_group(struct btrfs_space_info *space_info,
10046 struct btrfs_block_group_cache *cache)
10048 int index = get_block_group_index(cache);
10049 bool first = false;
10051 down_write(&space_info->groups_sem);
10052 if (list_empty(&space_info->block_groups[index]))
10054 list_add_tail(&cache->list, &space_info->block_groups[index]);
10055 up_write(&space_info->groups_sem);
10058 struct raid_kobject *rkobj;
10061 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
10064 rkobj->raid_type = index;
10065 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
10066 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
10067 "%s", get_raid_name(index));
10069 kobject_put(&rkobj->kobj);
10072 space_info->block_group_kobjs[index] = &rkobj->kobj;
10077 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
10080 static struct btrfs_block_group_cache *
10081 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
10083 struct btrfs_block_group_cache *cache;
10085 cache = kzalloc(sizeof(*cache), GFP_NOFS);
10089 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
10091 if (!cache->free_space_ctl) {
10096 cache->key.objectid = start;
10097 cache->key.offset = size;
10098 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10100 cache->sectorsize = root->sectorsize;
10101 cache->fs_info = root->fs_info;
10102 cache->full_stripe_len = btrfs_full_stripe_len(root,
10103 &root->fs_info->mapping_tree,
10105 set_free_space_tree_thresholds(cache);
10107 atomic_set(&cache->count, 1);
10108 spin_lock_init(&cache->lock);
10109 init_rwsem(&cache->data_rwsem);
10110 INIT_LIST_HEAD(&cache->list);
10111 INIT_LIST_HEAD(&cache->cluster_list);
10112 INIT_LIST_HEAD(&cache->bg_list);
10113 INIT_LIST_HEAD(&cache->ro_list);
10114 INIT_LIST_HEAD(&cache->dirty_list);
10115 INIT_LIST_HEAD(&cache->io_list);
10116 btrfs_init_free_space_ctl(cache);
10117 atomic_set(&cache->trimming, 0);
10118 mutex_init(&cache->free_space_lock);
10123 int btrfs_read_block_groups(struct btrfs_root *root)
10125 struct btrfs_path *path;
10127 struct btrfs_block_group_cache *cache;
10128 struct btrfs_fs_info *info = root->fs_info;
10129 struct btrfs_space_info *space_info;
10130 struct btrfs_key key;
10131 struct btrfs_key found_key;
10132 struct extent_buffer *leaf;
10133 int need_clear = 0;
10136 root = info->extent_root;
10139 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
10140 path = btrfs_alloc_path();
10143 path->reada = READA_FORWARD;
10145 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
10146 if (btrfs_test_opt(root, SPACE_CACHE) &&
10147 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
10149 if (btrfs_test_opt(root, CLEAR_CACHE))
10153 ret = find_first_block_group(root, path, &key);
10159 leaf = path->nodes[0];
10160 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
10162 cache = btrfs_create_block_group_cache(root, found_key.objectid,
10171 * When we mount with old space cache, we need to
10172 * set BTRFS_DC_CLEAR and set dirty flag.
10174 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
10175 * truncate the old free space cache inode and
10177 * b) Setting 'dirty flag' makes sure that we flush
10178 * the new space cache info onto disk.
10180 if (btrfs_test_opt(root, SPACE_CACHE))
10181 cache->disk_cache_state = BTRFS_DC_CLEAR;
10184 read_extent_buffer(leaf, &cache->item,
10185 btrfs_item_ptr_offset(leaf, path->slots[0]),
10186 sizeof(cache->item));
10187 cache->flags = btrfs_block_group_flags(&cache->item);
10189 key.objectid = found_key.objectid + found_key.offset;
10190 btrfs_release_path(path);
10193 * We need to exclude the super stripes now so that the space
10194 * info has super bytes accounted for, otherwise we'll think
10195 * we have more space than we actually do.
10197 ret = exclude_super_stripes(root, cache);
10200 * We may have excluded something, so call this just in
10203 free_excluded_extents(root, cache);
10204 btrfs_put_block_group(cache);
10209 * check for two cases, either we are full, and therefore
10210 * don't need to bother with the caching work since we won't
10211 * find any space, or we are empty, and we can just add all
10212 * the space in and be done with it. This saves us _alot_ of
10213 * time, particularly in the full case.
10215 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
10216 cache->last_byte_to_unpin = (u64)-1;
10217 cache->cached = BTRFS_CACHE_FINISHED;
10218 free_excluded_extents(root, cache);
10219 } else if (btrfs_block_group_used(&cache->item) == 0) {
10220 cache->last_byte_to_unpin = (u64)-1;
10221 cache->cached = BTRFS_CACHE_FINISHED;
10222 add_new_free_space(cache, root->fs_info,
10223 found_key.objectid,
10224 found_key.objectid +
10226 free_excluded_extents(root, cache);
10229 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10231 btrfs_remove_free_space_cache(cache);
10232 btrfs_put_block_group(cache);
10236 trace_btrfs_add_block_group(root->fs_info, cache, 0);
10237 ret = update_space_info(info, cache->flags, found_key.offset,
10238 btrfs_block_group_used(&cache->item),
10239 cache->bytes_super, &space_info);
10241 btrfs_remove_free_space_cache(cache);
10242 spin_lock(&info->block_group_cache_lock);
10243 rb_erase(&cache->cache_node,
10244 &info->block_group_cache_tree);
10245 RB_CLEAR_NODE(&cache->cache_node);
10246 spin_unlock(&info->block_group_cache_lock);
10247 btrfs_put_block_group(cache);
10251 cache->space_info = space_info;
10253 __link_block_group(space_info, cache);
10255 set_avail_alloc_bits(root->fs_info, cache->flags);
10256 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
10257 inc_block_group_ro(cache, 1);
10258 } else if (btrfs_block_group_used(&cache->item) == 0) {
10259 spin_lock(&info->unused_bgs_lock);
10260 /* Should always be true but just in case. */
10261 if (list_empty(&cache->bg_list)) {
10262 btrfs_get_block_group(cache);
10263 list_add_tail(&cache->bg_list,
10264 &info->unused_bgs);
10266 spin_unlock(&info->unused_bgs_lock);
10270 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
10271 if (!(get_alloc_profile(root, space_info->flags) &
10272 (BTRFS_BLOCK_GROUP_RAID10 |
10273 BTRFS_BLOCK_GROUP_RAID1 |
10274 BTRFS_BLOCK_GROUP_RAID5 |
10275 BTRFS_BLOCK_GROUP_RAID6 |
10276 BTRFS_BLOCK_GROUP_DUP)))
10279 * avoid allocating from un-mirrored block group if there are
10280 * mirrored block groups.
10282 list_for_each_entry(cache,
10283 &space_info->block_groups[BTRFS_RAID_RAID0],
10285 inc_block_group_ro(cache, 1);
10286 list_for_each_entry(cache,
10287 &space_info->block_groups[BTRFS_RAID_SINGLE],
10289 inc_block_group_ro(cache, 1);
10292 init_global_block_rsv(info);
10295 btrfs_free_path(path);
10299 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
10300 struct btrfs_root *root)
10302 struct btrfs_block_group_cache *block_group, *tmp;
10303 struct btrfs_root *extent_root = root->fs_info->extent_root;
10304 struct btrfs_block_group_item item;
10305 struct btrfs_key key;
10307 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
10309 trans->can_flush_pending_bgs = false;
10310 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
10314 spin_lock(&block_group->lock);
10315 memcpy(&item, &block_group->item, sizeof(item));
10316 memcpy(&key, &block_group->key, sizeof(key));
10317 spin_unlock(&block_group->lock);
10319 ret = btrfs_insert_item(trans, extent_root, &key, &item,
10322 btrfs_abort_transaction(trans, extent_root, ret);
10323 ret = btrfs_finish_chunk_alloc(trans, extent_root,
10324 key.objectid, key.offset);
10326 btrfs_abort_transaction(trans, extent_root, ret);
10327 add_block_group_free_space(trans, root->fs_info, block_group);
10328 /* already aborted the transaction if it failed. */
10330 list_del_init(&block_group->bg_list);
10332 trans->can_flush_pending_bgs = can_flush_pending_bgs;
10335 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
10336 struct btrfs_root *root, u64 bytes_used,
10337 u64 type, u64 chunk_objectid, u64 chunk_offset,
10341 struct btrfs_root *extent_root;
10342 struct btrfs_block_group_cache *cache;
10343 extent_root = root->fs_info->extent_root;
10345 btrfs_set_log_full_commit(root->fs_info, trans);
10347 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
10351 btrfs_set_block_group_used(&cache->item, bytes_used);
10352 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
10353 btrfs_set_block_group_flags(&cache->item, type);
10355 cache->flags = type;
10356 cache->last_byte_to_unpin = (u64)-1;
10357 cache->cached = BTRFS_CACHE_FINISHED;
10358 cache->needs_free_space = 1;
10359 ret = exclude_super_stripes(root, cache);
10362 * We may have excluded something, so call this just in
10365 free_excluded_extents(root, cache);
10366 btrfs_put_block_group(cache);
10370 add_new_free_space(cache, root->fs_info, chunk_offset,
10371 chunk_offset + size);
10373 free_excluded_extents(root, cache);
10375 #ifdef CONFIG_BTRFS_DEBUG
10376 if (btrfs_should_fragment_free_space(root, cache)) {
10377 u64 new_bytes_used = size - bytes_used;
10379 bytes_used += new_bytes_used >> 1;
10380 fragment_free_space(root, cache);
10384 * Call to ensure the corresponding space_info object is created and
10385 * assigned to our block group, but don't update its counters just yet.
10386 * We want our bg to be added to the rbtree with its ->space_info set.
10388 ret = update_space_info(root->fs_info, cache->flags, 0, 0, 0,
10389 &cache->space_info);
10391 btrfs_remove_free_space_cache(cache);
10392 btrfs_put_block_group(cache);
10396 ret = btrfs_add_block_group_cache(root->fs_info, cache);
10398 btrfs_remove_free_space_cache(cache);
10399 btrfs_put_block_group(cache);
10404 * Now that our block group has its ->space_info set and is inserted in
10405 * the rbtree, update the space info's counters.
10407 trace_btrfs_add_block_group(root->fs_info, cache, 1);
10408 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
10409 cache->bytes_super, &cache->space_info);
10411 btrfs_remove_free_space_cache(cache);
10412 spin_lock(&root->fs_info->block_group_cache_lock);
10413 rb_erase(&cache->cache_node,
10414 &root->fs_info->block_group_cache_tree);
10415 RB_CLEAR_NODE(&cache->cache_node);
10416 spin_unlock(&root->fs_info->block_group_cache_lock);
10417 btrfs_put_block_group(cache);
10420 update_global_block_rsv(root->fs_info);
10422 __link_block_group(cache->space_info, cache);
10424 list_add_tail(&cache->bg_list, &trans->new_bgs);
10426 set_avail_alloc_bits(extent_root->fs_info, type);
10430 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
10432 u64 extra_flags = chunk_to_extended(flags) &
10433 BTRFS_EXTENDED_PROFILE_MASK;
10435 write_seqlock(&fs_info->profiles_lock);
10436 if (flags & BTRFS_BLOCK_GROUP_DATA)
10437 fs_info->avail_data_alloc_bits &= ~extra_flags;
10438 if (flags & BTRFS_BLOCK_GROUP_METADATA)
10439 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
10440 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
10441 fs_info->avail_system_alloc_bits &= ~extra_flags;
10442 write_sequnlock(&fs_info->profiles_lock);
10445 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10446 struct btrfs_root *root, u64 group_start,
10447 struct extent_map *em)
10449 struct btrfs_path *path;
10450 struct btrfs_block_group_cache *block_group;
10451 struct btrfs_free_cluster *cluster;
10452 struct btrfs_root *tree_root = root->fs_info->tree_root;
10453 struct btrfs_key key;
10454 struct inode *inode;
10455 struct kobject *kobj = NULL;
10459 struct btrfs_caching_control *caching_ctl = NULL;
10462 root = root->fs_info->extent_root;
10464 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10465 BUG_ON(!block_group);
10466 BUG_ON(!block_group->ro);
10469 * Free the reserved super bytes from this block group before
10472 free_excluded_extents(root, block_group);
10474 memcpy(&key, &block_group->key, sizeof(key));
10475 index = get_block_group_index(block_group);
10476 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10477 BTRFS_BLOCK_GROUP_RAID1 |
10478 BTRFS_BLOCK_GROUP_RAID10))
10483 /* make sure this block group isn't part of an allocation cluster */
10484 cluster = &root->fs_info->data_alloc_cluster;
10485 spin_lock(&cluster->refill_lock);
10486 btrfs_return_cluster_to_free_space(block_group, cluster);
10487 spin_unlock(&cluster->refill_lock);
10490 * make sure this block group isn't part of a metadata
10491 * allocation cluster
10493 cluster = &root->fs_info->meta_alloc_cluster;
10494 spin_lock(&cluster->refill_lock);
10495 btrfs_return_cluster_to_free_space(block_group, cluster);
10496 spin_unlock(&cluster->refill_lock);
10498 path = btrfs_alloc_path();
10505 * get the inode first so any iput calls done for the io_list
10506 * aren't the final iput (no unlinks allowed now)
10508 inode = lookup_free_space_inode(tree_root, block_group, path);
10510 mutex_lock(&trans->transaction->cache_write_mutex);
10512 * make sure our free spache cache IO is done before remove the
10515 spin_lock(&trans->transaction->dirty_bgs_lock);
10516 if (!list_empty(&block_group->io_list)) {
10517 list_del_init(&block_group->io_list);
10519 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10521 spin_unlock(&trans->transaction->dirty_bgs_lock);
10522 btrfs_wait_cache_io(root, trans, block_group,
10523 &block_group->io_ctl, path,
10524 block_group->key.objectid);
10525 btrfs_put_block_group(block_group);
10526 spin_lock(&trans->transaction->dirty_bgs_lock);
10529 if (!list_empty(&block_group->dirty_list)) {
10530 list_del_init(&block_group->dirty_list);
10531 btrfs_put_block_group(block_group);
10533 spin_unlock(&trans->transaction->dirty_bgs_lock);
10534 mutex_unlock(&trans->transaction->cache_write_mutex);
10536 if (!IS_ERR(inode)) {
10537 ret = btrfs_orphan_add(trans, inode);
10539 btrfs_add_delayed_iput(inode);
10542 clear_nlink(inode);
10543 /* One for the block groups ref */
10544 spin_lock(&block_group->lock);
10545 if (block_group->iref) {
10546 block_group->iref = 0;
10547 block_group->inode = NULL;
10548 spin_unlock(&block_group->lock);
10551 spin_unlock(&block_group->lock);
10553 /* One for our lookup ref */
10554 btrfs_add_delayed_iput(inode);
10557 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10558 key.offset = block_group->key.objectid;
10561 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10565 btrfs_release_path(path);
10567 ret = btrfs_del_item(trans, tree_root, path);
10570 btrfs_release_path(path);
10573 spin_lock(&root->fs_info->block_group_cache_lock);
10574 rb_erase(&block_group->cache_node,
10575 &root->fs_info->block_group_cache_tree);
10576 RB_CLEAR_NODE(&block_group->cache_node);
10578 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10579 root->fs_info->first_logical_byte = (u64)-1;
10580 spin_unlock(&root->fs_info->block_group_cache_lock);
10582 down_write(&block_group->space_info->groups_sem);
10584 * we must use list_del_init so people can check to see if they
10585 * are still on the list after taking the semaphore
10587 list_del_init(&block_group->list);
10588 if (list_empty(&block_group->space_info->block_groups[index])) {
10589 kobj = block_group->space_info->block_group_kobjs[index];
10590 block_group->space_info->block_group_kobjs[index] = NULL;
10591 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10593 up_write(&block_group->space_info->groups_sem);
10599 if (block_group->has_caching_ctl)
10600 caching_ctl = get_caching_control(block_group);
10601 if (block_group->cached == BTRFS_CACHE_STARTED)
10602 wait_block_group_cache_done(block_group);
10603 if (block_group->has_caching_ctl) {
10604 down_write(&root->fs_info->commit_root_sem);
10605 if (!caching_ctl) {
10606 struct btrfs_caching_control *ctl;
10608 list_for_each_entry(ctl,
10609 &root->fs_info->caching_block_groups, list)
10610 if (ctl->block_group == block_group) {
10612 atomic_inc(&caching_ctl->count);
10617 list_del_init(&caching_ctl->list);
10618 up_write(&root->fs_info->commit_root_sem);
10620 /* Once for the caching bgs list and once for us. */
10621 put_caching_control(caching_ctl);
10622 put_caching_control(caching_ctl);
10626 spin_lock(&trans->transaction->dirty_bgs_lock);
10627 if (!list_empty(&block_group->dirty_list)) {
10630 if (!list_empty(&block_group->io_list)) {
10633 spin_unlock(&trans->transaction->dirty_bgs_lock);
10634 btrfs_remove_free_space_cache(block_group);
10636 spin_lock(&block_group->space_info->lock);
10637 list_del_init(&block_group->ro_list);
10639 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10640 WARN_ON(block_group->space_info->total_bytes
10641 < block_group->key.offset);
10642 WARN_ON(block_group->space_info->bytes_readonly
10643 < block_group->key.offset);
10644 WARN_ON(block_group->space_info->disk_total
10645 < block_group->key.offset * factor);
10647 block_group->space_info->total_bytes -= block_group->key.offset;
10648 block_group->space_info->bytes_readonly -= block_group->key.offset;
10649 block_group->space_info->disk_total -= block_group->key.offset * factor;
10651 spin_unlock(&block_group->space_info->lock);
10653 memcpy(&key, &block_group->key, sizeof(key));
10656 if (!list_empty(&em->list)) {
10657 /* We're in the transaction->pending_chunks list. */
10658 free_extent_map(em);
10660 spin_lock(&block_group->lock);
10661 block_group->removed = 1;
10663 * At this point trimming can't start on this block group, because we
10664 * removed the block group from the tree fs_info->block_group_cache_tree
10665 * so no one can't find it anymore and even if someone already got this
10666 * block group before we removed it from the rbtree, they have already
10667 * incremented block_group->trimming - if they didn't, they won't find
10668 * any free space entries because we already removed them all when we
10669 * called btrfs_remove_free_space_cache().
10671 * And we must not remove the extent map from the fs_info->mapping_tree
10672 * to prevent the same logical address range and physical device space
10673 * ranges from being reused for a new block group. This is because our
10674 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10675 * completely transactionless, so while it is trimming a range the
10676 * currently running transaction might finish and a new one start,
10677 * allowing for new block groups to be created that can reuse the same
10678 * physical device locations unless we take this special care.
10680 * There may also be an implicit trim operation if the file system
10681 * is mounted with -odiscard. The same protections must remain
10682 * in place until the extents have been discarded completely when
10683 * the transaction commit has completed.
10685 remove_em = (atomic_read(&block_group->trimming) == 0);
10687 * Make sure a trimmer task always sees the em in the pinned_chunks list
10688 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10689 * before checking block_group->removed).
10693 * Our em might be in trans->transaction->pending_chunks which
10694 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10695 * and so is the fs_info->pinned_chunks list.
10697 * So at this point we must be holding the chunk_mutex to avoid
10698 * any races with chunk allocation (more specifically at
10699 * volumes.c:contains_pending_extent()), to ensure it always
10700 * sees the em, either in the pending_chunks list or in the
10701 * pinned_chunks list.
10703 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10705 spin_unlock(&block_group->lock);
10708 struct extent_map_tree *em_tree;
10710 em_tree = &root->fs_info->mapping_tree.map_tree;
10711 write_lock(&em_tree->lock);
10713 * The em might be in the pending_chunks list, so make sure the
10714 * chunk mutex is locked, since remove_extent_mapping() will
10715 * delete us from that list.
10717 remove_extent_mapping(em_tree, em);
10718 write_unlock(&em_tree->lock);
10719 /* once for the tree */
10720 free_extent_map(em);
10723 unlock_chunks(root);
10725 ret = remove_block_group_free_space(trans, root->fs_info, block_group);
10729 btrfs_put_block_group(block_group);
10730 btrfs_put_block_group(block_group);
10732 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10738 ret = btrfs_del_item(trans, root, path);
10740 btrfs_free_path(path);
10744 struct btrfs_trans_handle *
10745 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10746 const u64 chunk_offset)
10748 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10749 struct extent_map *em;
10750 struct map_lookup *map;
10751 unsigned int num_items;
10753 read_lock(&em_tree->lock);
10754 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10755 read_unlock(&em_tree->lock);
10756 ASSERT(em && em->start == chunk_offset);
10759 * We need to reserve 3 + N units from the metadata space info in order
10760 * to remove a block group (done at btrfs_remove_chunk() and at
10761 * btrfs_remove_block_group()), which are used for:
10763 * 1 unit for adding the free space inode's orphan (located in the tree
10765 * 1 unit for deleting the block group item (located in the extent
10767 * 1 unit for deleting the free space item (located in tree of tree
10769 * N units for deleting N device extent items corresponding to each
10770 * stripe (located in the device tree).
10772 * In order to remove a block group we also need to reserve units in the
10773 * system space info in order to update the chunk tree (update one or
10774 * more device items and remove one chunk item), but this is done at
10775 * btrfs_remove_chunk() through a call to check_system_chunk().
10777 map = em->map_lookup;
10778 num_items = 3 + map->num_stripes;
10779 free_extent_map(em);
10781 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10786 * Process the unused_bgs list and remove any that don't have any allocated
10787 * space inside of them.
10789 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10791 struct btrfs_block_group_cache *block_group;
10792 struct btrfs_space_info *space_info;
10793 struct btrfs_root *root = fs_info->extent_root;
10794 struct btrfs_trans_handle *trans;
10797 if (!fs_info->open)
10800 spin_lock(&fs_info->unused_bgs_lock);
10801 while (!list_empty(&fs_info->unused_bgs)) {
10805 block_group = list_first_entry(&fs_info->unused_bgs,
10806 struct btrfs_block_group_cache,
10808 list_del_init(&block_group->bg_list);
10810 space_info = block_group->space_info;
10812 if (ret || btrfs_mixed_space_info(space_info)) {
10813 btrfs_put_block_group(block_group);
10816 spin_unlock(&fs_info->unused_bgs_lock);
10818 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10820 /* Don't want to race with allocators so take the groups_sem */
10821 down_write(&space_info->groups_sem);
10822 spin_lock(&block_group->lock);
10823 if (block_group->reserved ||
10824 btrfs_block_group_used(&block_group->item) ||
10826 list_is_singular(&block_group->list)) {
10828 * We want to bail if we made new allocations or have
10829 * outstanding allocations in this block group. We do
10830 * the ro check in case balance is currently acting on
10831 * this block group.
10833 spin_unlock(&block_group->lock);
10834 up_write(&space_info->groups_sem);
10837 spin_unlock(&block_group->lock);
10839 /* We don't want to force the issue, only flip if it's ok. */
10840 ret = inc_block_group_ro(block_group, 0);
10841 up_write(&space_info->groups_sem);
10848 * Want to do this before we do anything else so we can recover
10849 * properly if we fail to join the transaction.
10851 trans = btrfs_start_trans_remove_block_group(fs_info,
10852 block_group->key.objectid);
10853 if (IS_ERR(trans)) {
10854 btrfs_dec_block_group_ro(root, block_group);
10855 ret = PTR_ERR(trans);
10860 * We could have pending pinned extents for this block group,
10861 * just delete them, we don't care about them anymore.
10863 start = block_group->key.objectid;
10864 end = start + block_group->key.offset - 1;
10866 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10867 * btrfs_finish_extent_commit(). If we are at transaction N,
10868 * another task might be running finish_extent_commit() for the
10869 * previous transaction N - 1, and have seen a range belonging
10870 * to the block group in freed_extents[] before we were able to
10871 * clear the whole block group range from freed_extents[]. This
10872 * means that task can lookup for the block group after we
10873 * unpinned it from freed_extents[] and removed it, leading to
10874 * a BUG_ON() at btrfs_unpin_extent_range().
10876 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10877 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10880 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10881 btrfs_dec_block_group_ro(root, block_group);
10884 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10887 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10888 btrfs_dec_block_group_ro(root, block_group);
10891 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10893 /* Reset pinned so btrfs_put_block_group doesn't complain */
10894 spin_lock(&space_info->lock);
10895 spin_lock(&block_group->lock);
10897 space_info->bytes_pinned -= block_group->pinned;
10898 space_info->bytes_readonly += block_group->pinned;
10899 percpu_counter_add(&space_info->total_bytes_pinned,
10900 -block_group->pinned);
10901 block_group->pinned = 0;
10903 spin_unlock(&block_group->lock);
10904 spin_unlock(&space_info->lock);
10906 /* DISCARD can flip during remount */
10907 trimming = btrfs_test_opt(root, DISCARD);
10909 /* Implicit trim during transaction commit. */
10911 btrfs_get_block_group_trimming(block_group);
10914 * Btrfs_remove_chunk will abort the transaction if things go
10917 ret = btrfs_remove_chunk(trans, root,
10918 block_group->key.objectid);
10922 btrfs_put_block_group_trimming(block_group);
10927 * If we're not mounted with -odiscard, we can just forget
10928 * about this block group. Otherwise we'll need to wait
10929 * until transaction commit to do the actual discard.
10932 spin_lock(&fs_info->unused_bgs_lock);
10934 * A concurrent scrub might have added us to the list
10935 * fs_info->unused_bgs, so use a list_move operation
10936 * to add the block group to the deleted_bgs list.
10938 list_move(&block_group->bg_list,
10939 &trans->transaction->deleted_bgs);
10940 spin_unlock(&fs_info->unused_bgs_lock);
10941 btrfs_get_block_group(block_group);
10944 btrfs_end_transaction(trans, root);
10946 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10947 btrfs_put_block_group(block_group);
10948 spin_lock(&fs_info->unused_bgs_lock);
10950 spin_unlock(&fs_info->unused_bgs_lock);
10953 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10955 struct btrfs_space_info *space_info;
10956 struct btrfs_super_block *disk_super;
10962 disk_super = fs_info->super_copy;
10963 if (!btrfs_super_root(disk_super))
10966 features = btrfs_super_incompat_flags(disk_super);
10967 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10970 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10971 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10976 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10977 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10979 flags = BTRFS_BLOCK_GROUP_METADATA;
10980 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10984 flags = BTRFS_BLOCK_GROUP_DATA;
10985 ret = update_space_info(fs_info, flags, 0, 0, 0, &space_info);
10991 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10993 return unpin_extent_range(root, start, end, false);
10997 * It used to be that old block groups would be left around forever.
10998 * Iterating over them would be enough to trim unused space. Since we
10999 * now automatically remove them, we also need to iterate over unallocated
11002 * We don't want a transaction for this since the discard may take a
11003 * substantial amount of time. We don't require that a transaction be
11004 * running, but we do need to take a running transaction into account
11005 * to ensure that we're not discarding chunks that were released in
11006 * the current transaction.
11008 * Holding the chunks lock will prevent other threads from allocating
11009 * or releasing chunks, but it won't prevent a running transaction
11010 * from committing and releasing the memory that the pending chunks
11011 * list head uses. For that, we need to take a reference to the
11014 static int btrfs_trim_free_extents(struct btrfs_device *device,
11015 u64 minlen, u64 *trimmed)
11017 u64 start = 0, len = 0;
11022 /* Not writeable = nothing to do. */
11023 if (!device->writeable)
11026 /* No free space = nothing to do. */
11027 if (device->total_bytes <= device->bytes_used)
11033 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
11034 struct btrfs_transaction *trans;
11037 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
11041 down_read(&fs_info->commit_root_sem);
11043 spin_lock(&fs_info->trans_lock);
11044 trans = fs_info->running_transaction;
11046 atomic_inc(&trans->use_count);
11047 spin_unlock(&fs_info->trans_lock);
11049 ret = find_free_dev_extent_start(trans, device, minlen, start,
11052 btrfs_put_transaction(trans);
11055 up_read(&fs_info->commit_root_sem);
11056 mutex_unlock(&fs_info->chunk_mutex);
11057 if (ret == -ENOSPC)
11062 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
11063 up_read(&fs_info->commit_root_sem);
11064 mutex_unlock(&fs_info->chunk_mutex);
11072 if (fatal_signal_pending(current)) {
11073 ret = -ERESTARTSYS;
11083 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
11085 struct btrfs_fs_info *fs_info = root->fs_info;
11086 struct btrfs_block_group_cache *cache = NULL;
11087 struct btrfs_device *device;
11088 struct list_head *devices;
11093 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
11097 * try to trim all FS space, our block group may start from non-zero.
11099 if (range->len == total_bytes)
11100 cache = btrfs_lookup_first_block_group(fs_info, range->start);
11102 cache = btrfs_lookup_block_group(fs_info, range->start);
11105 if (cache->key.objectid >= (range->start + range->len)) {
11106 btrfs_put_block_group(cache);
11110 start = max(range->start, cache->key.objectid);
11111 end = min(range->start + range->len,
11112 cache->key.objectid + cache->key.offset);
11114 if (end - start >= range->minlen) {
11115 if (!block_group_cache_done(cache)) {
11116 ret = cache_block_group(cache, 0);
11118 btrfs_put_block_group(cache);
11121 ret = wait_block_group_cache_done(cache);
11123 btrfs_put_block_group(cache);
11127 ret = btrfs_trim_block_group(cache,
11133 trimmed += group_trimmed;
11135 btrfs_put_block_group(cache);
11140 cache = next_block_group(fs_info->tree_root, cache);
11143 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
11144 devices = &root->fs_info->fs_devices->alloc_list;
11145 list_for_each_entry(device, devices, dev_alloc_list) {
11146 ret = btrfs_trim_free_extents(device, range->minlen,
11151 trimmed += group_trimmed;
11153 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
11155 range->len = trimmed;
11160 * btrfs_{start,end}_write_no_snapshoting() are similar to
11161 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
11162 * data into the page cache through nocow before the subvolume is snapshoted,
11163 * but flush the data into disk after the snapshot creation, or to prevent
11164 * operations while snapshoting is ongoing and that cause the snapshot to be
11165 * inconsistent (writes followed by expanding truncates for example).
11167 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
11169 percpu_counter_dec(&root->subv_writers->counter);
11171 * Make sure counter is updated before we wake up waiters.
11174 if (waitqueue_active(&root->subv_writers->wait))
11175 wake_up(&root->subv_writers->wait);
11178 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
11180 if (atomic_read(&root->will_be_snapshoted))
11183 percpu_counter_inc(&root->subv_writers->counter);
11185 * Make sure counter is updated before we check for snapshot creation.
11188 if (atomic_read(&root->will_be_snapshoted)) {
11189 btrfs_end_write_no_snapshoting(root);
11195 static int wait_snapshoting_atomic_t(atomic_t *a)
11201 void btrfs_wait_for_snapshot_creation(struct btrfs_root *root)
11206 ret = btrfs_start_write_no_snapshoting(root);
11209 wait_on_atomic_t(&root->will_be_snapshoted,
11210 wait_snapshoting_atomic_t,
11211 TASK_UNINTERRUPTIBLE);