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"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 struct btrfs_delayed_ref_node *node, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
87 struct extent_buffer *leaf,
88 struct btrfs_extent_item *ei);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, u64 owner, u64 offset,
93 struct btrfs_key *ins, int ref_mod);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 u64 parent, u64 root_objectid,
97 u64 flags, struct btrfs_disk_key *key,
98 int level, struct btrfs_key *ins);
99 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
100 struct btrfs_root *extent_root, u64 flags,
102 static int find_next_key(struct btrfs_path *path, int level,
103 struct btrfs_key *key);
104 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
105 int dump_block_groups);
106 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
107 u64 num_bytes, int reserve,
109 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
111 int btrfs_pin_extent(struct btrfs_root *root,
112 u64 bytenr, u64 num_bytes, int reserved);
115 block_group_cache_done(struct btrfs_block_group_cache *cache)
118 return cache->cached == BTRFS_CACHE_FINISHED ||
119 cache->cached == BTRFS_CACHE_ERROR;
122 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
124 return (cache->flags & bits) == bits;
127 void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
129 atomic_inc(&cache->count);
132 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
134 if (atomic_dec_and_test(&cache->count)) {
135 WARN_ON(cache->pinned > 0);
136 WARN_ON(cache->reserved > 0);
137 kfree(cache->free_space_ctl);
143 * this adds the block group to the fs_info rb tree for the block group
146 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
147 struct btrfs_block_group_cache *block_group)
150 struct rb_node *parent = NULL;
151 struct btrfs_block_group_cache *cache;
153 spin_lock(&info->block_group_cache_lock);
154 p = &info->block_group_cache_tree.rb_node;
158 cache = rb_entry(parent, struct btrfs_block_group_cache,
160 if (block_group->key.objectid < cache->key.objectid) {
162 } else if (block_group->key.objectid > cache->key.objectid) {
165 spin_unlock(&info->block_group_cache_lock);
170 rb_link_node(&block_group->cache_node, parent, p);
171 rb_insert_color(&block_group->cache_node,
172 &info->block_group_cache_tree);
174 if (info->first_logical_byte > block_group->key.objectid)
175 info->first_logical_byte = block_group->key.objectid;
177 spin_unlock(&info->block_group_cache_lock);
183 * This will return the block group at or after bytenr if contains is 0, else
184 * it will return the block group that contains the bytenr
186 static struct btrfs_block_group_cache *
187 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
190 struct btrfs_block_group_cache *cache, *ret = NULL;
194 spin_lock(&info->block_group_cache_lock);
195 n = info->block_group_cache_tree.rb_node;
198 cache = rb_entry(n, struct btrfs_block_group_cache,
200 end = cache->key.objectid + cache->key.offset - 1;
201 start = cache->key.objectid;
203 if (bytenr < start) {
204 if (!contains && (!ret || start < ret->key.objectid))
207 } else if (bytenr > start) {
208 if (contains && bytenr <= end) {
219 btrfs_get_block_group(ret);
220 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
221 info->first_logical_byte = ret->key.objectid;
223 spin_unlock(&info->block_group_cache_lock);
228 static int add_excluded_extent(struct btrfs_root *root,
229 u64 start, u64 num_bytes)
231 u64 end = start + num_bytes - 1;
232 set_extent_bits(&root->fs_info->freed_extents[0],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 set_extent_bits(&root->fs_info->freed_extents[1],
235 start, end, EXTENT_UPTODATE, GFP_NOFS);
239 static void free_excluded_extents(struct btrfs_root *root,
240 struct btrfs_block_group_cache *cache)
244 start = cache->key.objectid;
245 end = start + cache->key.offset - 1;
247 clear_extent_bits(&root->fs_info->freed_extents[0],
248 start, end, EXTENT_UPTODATE, GFP_NOFS);
249 clear_extent_bits(&root->fs_info->freed_extents[1],
250 start, end, EXTENT_UPTODATE, GFP_NOFS);
253 static int exclude_super_stripes(struct btrfs_root *root,
254 struct btrfs_block_group_cache *cache)
261 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
262 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
263 cache->bytes_super += stripe_len;
264 ret = add_excluded_extent(root, cache->key.objectid,
270 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
271 bytenr = btrfs_sb_offset(i);
272 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
273 cache->key.objectid, bytenr,
274 0, &logical, &nr, &stripe_len);
281 if (logical[nr] > cache->key.objectid +
285 if (logical[nr] + stripe_len <= cache->key.objectid)
289 if (start < cache->key.objectid) {
290 start = cache->key.objectid;
291 len = (logical[nr] + stripe_len) - start;
293 len = min_t(u64, stripe_len,
294 cache->key.objectid +
295 cache->key.offset - start);
298 cache->bytes_super += len;
299 ret = add_excluded_extent(root, start, len);
311 static struct btrfs_caching_control *
312 get_caching_control(struct btrfs_block_group_cache *cache)
314 struct btrfs_caching_control *ctl;
316 spin_lock(&cache->lock);
317 if (!cache->caching_ctl) {
318 spin_unlock(&cache->lock);
322 ctl = cache->caching_ctl;
323 atomic_inc(&ctl->count);
324 spin_unlock(&cache->lock);
328 static void put_caching_control(struct btrfs_caching_control *ctl)
330 if (atomic_dec_and_test(&ctl->count))
334 #ifdef CONFIG_BTRFS_DEBUG
335 static void fragment_free_space(struct btrfs_root *root,
336 struct btrfs_block_group_cache *block_group)
338 u64 start = block_group->key.objectid;
339 u64 len = block_group->key.offset;
340 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
341 root->nodesize : root->sectorsize;
342 u64 step = chunk << 1;
344 while (len > chunk) {
345 btrfs_remove_free_space(block_group, start, chunk);
356 * this is only called by cache_block_group, since we could have freed extents
357 * we need to check the pinned_extents for any extents that can't be used yet
358 * since their free space will be released as soon as the transaction commits.
360 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
361 struct btrfs_fs_info *info, u64 start, u64 end)
363 u64 extent_start, extent_end, size, total_added = 0;
366 while (start < end) {
367 ret = find_first_extent_bit(info->pinned_extents, start,
368 &extent_start, &extent_end,
369 EXTENT_DIRTY | EXTENT_UPTODATE,
374 if (extent_start <= start) {
375 start = extent_end + 1;
376 } else if (extent_start > start && extent_start < end) {
377 size = extent_start - start;
379 ret = btrfs_add_free_space(block_group, start,
381 BUG_ON(ret); /* -ENOMEM or logic error */
382 start = extent_end + 1;
391 ret = btrfs_add_free_space(block_group, start, size);
392 BUG_ON(ret); /* -ENOMEM or logic error */
398 static noinline void caching_thread(struct btrfs_work *work)
400 struct btrfs_block_group_cache *block_group;
401 struct btrfs_fs_info *fs_info;
402 struct btrfs_caching_control *caching_ctl;
403 struct btrfs_root *extent_root;
404 struct btrfs_path *path;
405 struct extent_buffer *leaf;
406 struct btrfs_key key;
413 caching_ctl = container_of(work, struct btrfs_caching_control, work);
414 block_group = caching_ctl->block_group;
415 fs_info = block_group->fs_info;
416 extent_root = fs_info->extent_root;
418 path = btrfs_alloc_path();
422 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
424 #ifdef CONFIG_BTRFS_DEBUG
426 * If we're fragmenting we don't want to make anybody think we can
427 * allocate from this block group until we've had a chance to fragment
430 if (btrfs_should_fragment_free_space(extent_root, block_group))
434 * We don't want to deadlock with somebody trying to allocate a new
435 * extent for the extent root while also trying to search the extent
436 * root to add free space. So we skip locking and search the commit
437 * root, since its read-only
439 path->skip_locking = 1;
440 path->search_commit_root = 1;
445 key.type = BTRFS_EXTENT_ITEM_KEY;
447 mutex_lock(&caching_ctl->mutex);
448 /* need to make sure the commit_root doesn't disappear */
449 down_read(&fs_info->commit_root_sem);
452 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
456 leaf = path->nodes[0];
457 nritems = btrfs_header_nritems(leaf);
460 if (btrfs_fs_closing(fs_info) > 1) {
465 if (path->slots[0] < nritems) {
466 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
468 ret = find_next_key(path, 0, &key);
472 if (need_resched() ||
473 rwsem_is_contended(&fs_info->commit_root_sem)) {
475 caching_ctl->progress = last;
476 btrfs_release_path(path);
477 up_read(&fs_info->commit_root_sem);
478 mutex_unlock(&caching_ctl->mutex);
483 ret = btrfs_next_leaf(extent_root, path);
488 leaf = path->nodes[0];
489 nritems = btrfs_header_nritems(leaf);
493 if (key.objectid < last) {
496 key.type = BTRFS_EXTENT_ITEM_KEY;
499 caching_ctl->progress = last;
500 btrfs_release_path(path);
504 if (key.objectid < block_group->key.objectid) {
509 if (key.objectid >= block_group->key.objectid +
510 block_group->key.offset)
513 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
514 key.type == BTRFS_METADATA_ITEM_KEY) {
515 total_found += add_new_free_space(block_group,
518 if (key.type == BTRFS_METADATA_ITEM_KEY)
519 last = key.objectid +
520 fs_info->tree_root->nodesize;
522 last = key.objectid + key.offset;
524 if (total_found > (1024 * 1024 * 2)) {
527 wake_up(&caching_ctl->wait);
534 total_found += add_new_free_space(block_group, fs_info, last,
535 block_group->key.objectid +
536 block_group->key.offset);
537 spin_lock(&block_group->lock);
538 block_group->caching_ctl = NULL;
539 block_group->cached = BTRFS_CACHE_FINISHED;
540 spin_unlock(&block_group->lock);
542 #ifdef CONFIG_BTRFS_DEBUG
543 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
546 spin_lock(&block_group->space_info->lock);
547 spin_lock(&block_group->lock);
548 bytes_used = block_group->key.offset -
549 btrfs_block_group_used(&block_group->item);
550 block_group->space_info->bytes_used += bytes_used >> 1;
551 spin_unlock(&block_group->lock);
552 spin_unlock(&block_group->space_info->lock);
553 fragment_free_space(extent_root, block_group);
557 caching_ctl->progress = (u64)-1;
559 btrfs_free_path(path);
560 up_read(&fs_info->commit_root_sem);
562 free_excluded_extents(extent_root, block_group);
564 mutex_unlock(&caching_ctl->mutex);
567 spin_lock(&block_group->lock);
568 block_group->caching_ctl = NULL;
569 block_group->cached = BTRFS_CACHE_ERROR;
570 spin_unlock(&block_group->lock);
572 wake_up(&caching_ctl->wait);
574 put_caching_control(caching_ctl);
575 btrfs_put_block_group(block_group);
578 static int cache_block_group(struct btrfs_block_group_cache *cache,
582 struct btrfs_fs_info *fs_info = cache->fs_info;
583 struct btrfs_caching_control *caching_ctl;
586 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
590 INIT_LIST_HEAD(&caching_ctl->list);
591 mutex_init(&caching_ctl->mutex);
592 init_waitqueue_head(&caching_ctl->wait);
593 caching_ctl->block_group = cache;
594 caching_ctl->progress = cache->key.objectid;
595 atomic_set(&caching_ctl->count, 1);
596 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
597 caching_thread, NULL, NULL);
599 spin_lock(&cache->lock);
601 * This should be a rare occasion, but this could happen I think in the
602 * case where one thread starts to load the space cache info, and then
603 * some other thread starts a transaction commit which tries to do an
604 * allocation while the other thread is still loading the space cache
605 * info. The previous loop should have kept us from choosing this block
606 * group, but if we've moved to the state where we will wait on caching
607 * block groups we need to first check if we're doing a fast load here,
608 * so we can wait for it to finish, otherwise we could end up allocating
609 * from a block group who's cache gets evicted for one reason or
612 while (cache->cached == BTRFS_CACHE_FAST) {
613 struct btrfs_caching_control *ctl;
615 ctl = cache->caching_ctl;
616 atomic_inc(&ctl->count);
617 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
618 spin_unlock(&cache->lock);
622 finish_wait(&ctl->wait, &wait);
623 put_caching_control(ctl);
624 spin_lock(&cache->lock);
627 if (cache->cached != BTRFS_CACHE_NO) {
628 spin_unlock(&cache->lock);
632 WARN_ON(cache->caching_ctl);
633 cache->caching_ctl = caching_ctl;
634 cache->cached = BTRFS_CACHE_FAST;
635 spin_unlock(&cache->lock);
637 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
638 mutex_lock(&caching_ctl->mutex);
639 ret = load_free_space_cache(fs_info, cache);
641 spin_lock(&cache->lock);
643 cache->caching_ctl = NULL;
644 cache->cached = BTRFS_CACHE_FINISHED;
645 cache->last_byte_to_unpin = (u64)-1;
646 caching_ctl->progress = (u64)-1;
648 if (load_cache_only) {
649 cache->caching_ctl = NULL;
650 cache->cached = BTRFS_CACHE_NO;
652 cache->cached = BTRFS_CACHE_STARTED;
653 cache->has_caching_ctl = 1;
656 spin_unlock(&cache->lock);
657 #ifdef CONFIG_BTRFS_DEBUG
659 btrfs_should_fragment_free_space(fs_info->extent_root,
663 spin_lock(&cache->space_info->lock);
664 spin_lock(&cache->lock);
665 bytes_used = cache->key.offset -
666 btrfs_block_group_used(&cache->item);
667 cache->space_info->bytes_used += bytes_used >> 1;
668 spin_unlock(&cache->lock);
669 spin_unlock(&cache->space_info->lock);
670 fragment_free_space(fs_info->extent_root, cache);
673 mutex_unlock(&caching_ctl->mutex);
675 wake_up(&caching_ctl->wait);
677 put_caching_control(caching_ctl);
678 free_excluded_extents(fs_info->extent_root, cache);
683 * We are not going to do the fast caching, set cached to the
684 * appropriate value and wakeup any waiters.
686 spin_lock(&cache->lock);
687 if (load_cache_only) {
688 cache->caching_ctl = NULL;
689 cache->cached = BTRFS_CACHE_NO;
691 cache->cached = BTRFS_CACHE_STARTED;
692 cache->has_caching_ctl = 1;
694 spin_unlock(&cache->lock);
695 wake_up(&caching_ctl->wait);
698 if (load_cache_only) {
699 put_caching_control(caching_ctl);
703 down_write(&fs_info->commit_root_sem);
704 atomic_inc(&caching_ctl->count);
705 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
706 up_write(&fs_info->commit_root_sem);
708 btrfs_get_block_group(cache);
710 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
716 * return the block group that starts at or after bytenr
718 static struct btrfs_block_group_cache *
719 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
721 struct btrfs_block_group_cache *cache;
723 cache = block_group_cache_tree_search(info, bytenr, 0);
729 * return the block group that contains the given bytenr
731 struct btrfs_block_group_cache *btrfs_lookup_block_group(
732 struct btrfs_fs_info *info,
735 struct btrfs_block_group_cache *cache;
737 cache = block_group_cache_tree_search(info, bytenr, 1);
742 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
745 struct list_head *head = &info->space_info;
746 struct btrfs_space_info *found;
748 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
751 list_for_each_entry_rcu(found, head, list) {
752 if (found->flags & flags) {
762 * after adding space to the filesystem, we need to clear the full flags
763 * on all the space infos.
765 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
767 struct list_head *head = &info->space_info;
768 struct btrfs_space_info *found;
771 list_for_each_entry_rcu(found, head, list)
776 /* simple helper to search for an existing data extent at a given offset */
777 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
780 struct btrfs_key key;
781 struct btrfs_path *path;
783 path = btrfs_alloc_path();
787 key.objectid = start;
789 key.type = BTRFS_EXTENT_ITEM_KEY;
790 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
792 btrfs_free_path(path);
797 * helper function to lookup reference count and flags of a tree block.
799 * the head node for delayed ref is used to store the sum of all the
800 * reference count modifications queued up in the rbtree. the head
801 * node may also store the extent flags to set. This way you can check
802 * to see what the reference count and extent flags would be if all of
803 * the delayed refs are not processed.
805 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
806 struct btrfs_root *root, u64 bytenr,
807 u64 offset, int metadata, u64 *refs, u64 *flags)
809 struct btrfs_delayed_ref_head *head;
810 struct btrfs_delayed_ref_root *delayed_refs;
811 struct btrfs_path *path;
812 struct btrfs_extent_item *ei;
813 struct extent_buffer *leaf;
814 struct btrfs_key key;
821 * If we don't have skinny metadata, don't bother doing anything
824 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
825 offset = root->nodesize;
829 path = btrfs_alloc_path();
834 path->skip_locking = 1;
835 path->search_commit_root = 1;
839 key.objectid = bytenr;
842 key.type = BTRFS_METADATA_ITEM_KEY;
844 key.type = BTRFS_EXTENT_ITEM_KEY;
846 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
851 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
852 if (path->slots[0]) {
854 btrfs_item_key_to_cpu(path->nodes[0], &key,
856 if (key.objectid == bytenr &&
857 key.type == BTRFS_EXTENT_ITEM_KEY &&
858 key.offset == root->nodesize)
864 leaf = path->nodes[0];
865 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
866 if (item_size >= sizeof(*ei)) {
867 ei = btrfs_item_ptr(leaf, path->slots[0],
868 struct btrfs_extent_item);
869 num_refs = btrfs_extent_refs(leaf, ei);
870 extent_flags = btrfs_extent_flags(leaf, ei);
872 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
873 struct btrfs_extent_item_v0 *ei0;
874 BUG_ON(item_size != sizeof(*ei0));
875 ei0 = btrfs_item_ptr(leaf, path->slots[0],
876 struct btrfs_extent_item_v0);
877 num_refs = btrfs_extent_refs_v0(leaf, ei0);
878 /* FIXME: this isn't correct for data */
879 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
884 BUG_ON(num_refs == 0);
894 delayed_refs = &trans->transaction->delayed_refs;
895 spin_lock(&delayed_refs->lock);
896 head = btrfs_find_delayed_ref_head(trans, bytenr);
898 if (!mutex_trylock(&head->mutex)) {
899 atomic_inc(&head->node.refs);
900 spin_unlock(&delayed_refs->lock);
902 btrfs_release_path(path);
905 * Mutex was contended, block until it's released and try
908 mutex_lock(&head->mutex);
909 mutex_unlock(&head->mutex);
910 btrfs_put_delayed_ref(&head->node);
913 spin_lock(&head->lock);
914 if (head->extent_op && head->extent_op->update_flags)
915 extent_flags |= head->extent_op->flags_to_set;
917 BUG_ON(num_refs == 0);
919 num_refs += head->node.ref_mod;
920 spin_unlock(&head->lock);
921 mutex_unlock(&head->mutex);
923 spin_unlock(&delayed_refs->lock);
925 WARN_ON(num_refs == 0);
929 *flags = extent_flags;
931 btrfs_free_path(path);
936 * Back reference rules. Back refs have three main goals:
938 * 1) differentiate between all holders of references to an extent so that
939 * when a reference is dropped we can make sure it was a valid reference
940 * before freeing the extent.
942 * 2) Provide enough information to quickly find the holders of an extent
943 * if we notice a given block is corrupted or bad.
945 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
946 * maintenance. This is actually the same as #2, but with a slightly
947 * different use case.
949 * There are two kinds of back refs. The implicit back refs is optimized
950 * for pointers in non-shared tree blocks. For a given pointer in a block,
951 * back refs of this kind provide information about the block's owner tree
952 * and the pointer's key. These information allow us to find the block by
953 * b-tree searching. The full back refs is for pointers in tree blocks not
954 * referenced by their owner trees. The location of tree block is recorded
955 * in the back refs. Actually the full back refs is generic, and can be
956 * used in all cases the implicit back refs is used. The major shortcoming
957 * of the full back refs is its overhead. Every time a tree block gets
958 * COWed, we have to update back refs entry for all pointers in it.
960 * For a newly allocated tree block, we use implicit back refs for
961 * pointers in it. This means most tree related operations only involve
962 * implicit back refs. For a tree block created in old transaction, the
963 * only way to drop a reference to it is COW it. So we can detect the
964 * event that tree block loses its owner tree's reference and do the
965 * back refs conversion.
967 * When a tree block is COW'd through a tree, there are four cases:
969 * The reference count of the block is one and the tree is the block's
970 * owner tree. Nothing to do in this case.
972 * The reference count of the block is one and the tree is not the
973 * block's owner tree. In this case, full back refs is used for pointers
974 * in the block. Remove these full back refs, add implicit back refs for
975 * every pointers in the new block.
977 * The reference count of the block is greater than one and the tree is
978 * the block's owner tree. In this case, implicit back refs is used for
979 * pointers in the block. Add full back refs for every pointers in the
980 * block, increase lower level extents' reference counts. The original
981 * implicit back refs are entailed to the new block.
983 * The reference count of the block is greater than one and the tree is
984 * not the block's owner tree. Add implicit back refs for every pointer in
985 * the new block, increase lower level extents' reference count.
987 * Back Reference Key composing:
989 * The key objectid corresponds to the first byte in the extent,
990 * The key type is used to differentiate between types of back refs.
991 * There are different meanings of the key offset for different types
994 * File extents can be referenced by:
996 * - multiple snapshots, subvolumes, or different generations in one subvol
997 * - different files inside a single subvolume
998 * - different offsets inside a file (bookend extents in file.c)
1000 * The extent ref structure for the implicit back refs has fields for:
1002 * - Objectid of the subvolume root
1003 * - objectid of the file holding the reference
1004 * - original offset in the file
1005 * - how many bookend extents
1007 * The key offset for the implicit back refs is hash of the first
1010 * The extent ref structure for the full back refs has field for:
1012 * - number of pointers in the tree leaf
1014 * The key offset for the implicit back refs is the first byte of
1017 * When a file extent is allocated, The implicit back refs is used.
1018 * the fields are filled in:
1020 * (root_key.objectid, inode objectid, offset in file, 1)
1022 * When a file extent is removed file truncation, we find the
1023 * corresponding implicit back refs and check the following fields:
1025 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1027 * Btree extents can be referenced by:
1029 * - Different subvolumes
1031 * Both the implicit back refs and the full back refs for tree blocks
1032 * only consist of key. The key offset for the implicit back refs is
1033 * objectid of block's owner tree. The key offset for the full back refs
1034 * is the first byte of parent block.
1036 * When implicit back refs is used, information about the lowest key and
1037 * level of the tree block are required. These information are stored in
1038 * tree block info structure.
1041 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1042 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1043 struct btrfs_root *root,
1044 struct btrfs_path *path,
1045 u64 owner, u32 extra_size)
1047 struct btrfs_extent_item *item;
1048 struct btrfs_extent_item_v0 *ei0;
1049 struct btrfs_extent_ref_v0 *ref0;
1050 struct btrfs_tree_block_info *bi;
1051 struct extent_buffer *leaf;
1052 struct btrfs_key key;
1053 struct btrfs_key found_key;
1054 u32 new_size = sizeof(*item);
1058 leaf = path->nodes[0];
1059 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1061 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1062 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_extent_item_v0);
1064 refs = btrfs_extent_refs_v0(leaf, ei0);
1066 if (owner == (u64)-1) {
1068 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1069 ret = btrfs_next_leaf(root, path);
1072 BUG_ON(ret > 0); /* Corruption */
1073 leaf = path->nodes[0];
1075 btrfs_item_key_to_cpu(leaf, &found_key,
1077 BUG_ON(key.objectid != found_key.objectid);
1078 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1082 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1083 struct btrfs_extent_ref_v0);
1084 owner = btrfs_ref_objectid_v0(leaf, ref0);
1088 btrfs_release_path(path);
1090 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1091 new_size += sizeof(*bi);
1093 new_size -= sizeof(*ei0);
1094 ret = btrfs_search_slot(trans, root, &key, path,
1095 new_size + extra_size, 1);
1098 BUG_ON(ret); /* Corruption */
1100 btrfs_extend_item(root, path, new_size);
1102 leaf = path->nodes[0];
1103 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1104 btrfs_set_extent_refs(leaf, item, refs);
1105 /* FIXME: get real generation */
1106 btrfs_set_extent_generation(leaf, item, 0);
1107 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1108 btrfs_set_extent_flags(leaf, item,
1109 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1110 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1111 bi = (struct btrfs_tree_block_info *)(item + 1);
1112 /* FIXME: get first key of the block */
1113 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1114 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1116 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1118 btrfs_mark_buffer_dirty(leaf);
1123 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1125 u32 high_crc = ~(u32)0;
1126 u32 low_crc = ~(u32)0;
1129 lenum = cpu_to_le64(root_objectid);
1130 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1131 lenum = cpu_to_le64(owner);
1132 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1133 lenum = cpu_to_le64(offset);
1134 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1136 return ((u64)high_crc << 31) ^ (u64)low_crc;
1139 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1140 struct btrfs_extent_data_ref *ref)
1142 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1143 btrfs_extent_data_ref_objectid(leaf, ref),
1144 btrfs_extent_data_ref_offset(leaf, ref));
1147 static int match_extent_data_ref(struct extent_buffer *leaf,
1148 struct btrfs_extent_data_ref *ref,
1149 u64 root_objectid, u64 owner, u64 offset)
1151 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1152 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1153 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1158 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1159 struct btrfs_root *root,
1160 struct btrfs_path *path,
1161 u64 bytenr, u64 parent,
1163 u64 owner, u64 offset)
1165 struct btrfs_key key;
1166 struct btrfs_extent_data_ref *ref;
1167 struct extent_buffer *leaf;
1173 key.objectid = bytenr;
1175 key.type = BTRFS_SHARED_DATA_REF_KEY;
1176 key.offset = parent;
1178 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1179 key.offset = hash_extent_data_ref(root_objectid,
1184 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1193 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1194 key.type = BTRFS_EXTENT_REF_V0_KEY;
1195 btrfs_release_path(path);
1196 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1207 leaf = path->nodes[0];
1208 nritems = btrfs_header_nritems(leaf);
1210 if (path->slots[0] >= nritems) {
1211 ret = btrfs_next_leaf(root, path);
1217 leaf = path->nodes[0];
1218 nritems = btrfs_header_nritems(leaf);
1222 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1223 if (key.objectid != bytenr ||
1224 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1227 ref = btrfs_item_ptr(leaf, path->slots[0],
1228 struct btrfs_extent_data_ref);
1230 if (match_extent_data_ref(leaf, ref, root_objectid,
1233 btrfs_release_path(path);
1245 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1246 struct btrfs_root *root,
1247 struct btrfs_path *path,
1248 u64 bytenr, u64 parent,
1249 u64 root_objectid, u64 owner,
1250 u64 offset, int refs_to_add)
1252 struct btrfs_key key;
1253 struct extent_buffer *leaf;
1258 key.objectid = bytenr;
1260 key.type = BTRFS_SHARED_DATA_REF_KEY;
1261 key.offset = parent;
1262 size = sizeof(struct btrfs_shared_data_ref);
1264 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1265 key.offset = hash_extent_data_ref(root_objectid,
1267 size = sizeof(struct btrfs_extent_data_ref);
1270 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1271 if (ret && ret != -EEXIST)
1274 leaf = path->nodes[0];
1276 struct btrfs_shared_data_ref *ref;
1277 ref = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_shared_data_ref);
1280 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1282 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1283 num_refs += refs_to_add;
1284 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1287 struct btrfs_extent_data_ref *ref;
1288 while (ret == -EEXIST) {
1289 ref = btrfs_item_ptr(leaf, path->slots[0],
1290 struct btrfs_extent_data_ref);
1291 if (match_extent_data_ref(leaf, ref, root_objectid,
1294 btrfs_release_path(path);
1296 ret = btrfs_insert_empty_item(trans, root, path, &key,
1298 if (ret && ret != -EEXIST)
1301 leaf = path->nodes[0];
1303 ref = btrfs_item_ptr(leaf, path->slots[0],
1304 struct btrfs_extent_data_ref);
1306 btrfs_set_extent_data_ref_root(leaf, ref,
1308 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1309 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1310 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1312 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1313 num_refs += refs_to_add;
1314 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1317 btrfs_mark_buffer_dirty(leaf);
1320 btrfs_release_path(path);
1324 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1325 struct btrfs_root *root,
1326 struct btrfs_path *path,
1327 int refs_to_drop, int *last_ref)
1329 struct btrfs_key key;
1330 struct btrfs_extent_data_ref *ref1 = NULL;
1331 struct btrfs_shared_data_ref *ref2 = NULL;
1332 struct extent_buffer *leaf;
1336 leaf = path->nodes[0];
1337 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1339 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1340 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1341 struct btrfs_extent_data_ref);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1343 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1344 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1345 struct btrfs_shared_data_ref);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1348 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1349 struct btrfs_extent_ref_v0 *ref0;
1350 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_extent_ref_v0);
1352 num_refs = btrfs_ref_count_v0(leaf, ref0);
1358 BUG_ON(num_refs < refs_to_drop);
1359 num_refs -= refs_to_drop;
1361 if (num_refs == 0) {
1362 ret = btrfs_del_item(trans, root, path);
1365 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1366 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1367 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1368 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1369 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1371 struct btrfs_extent_ref_v0 *ref0;
1372 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1373 struct btrfs_extent_ref_v0);
1374 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1377 btrfs_mark_buffer_dirty(leaf);
1382 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1383 struct btrfs_extent_inline_ref *iref)
1385 struct btrfs_key key;
1386 struct extent_buffer *leaf;
1387 struct btrfs_extent_data_ref *ref1;
1388 struct btrfs_shared_data_ref *ref2;
1391 leaf = path->nodes[0];
1392 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1394 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1395 BTRFS_EXTENT_DATA_REF_KEY) {
1396 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1397 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1399 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1400 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1402 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1403 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1404 struct btrfs_extent_data_ref);
1405 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1406 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1407 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1408 struct btrfs_shared_data_ref);
1409 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1410 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1411 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1412 struct btrfs_extent_ref_v0 *ref0;
1413 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1414 struct btrfs_extent_ref_v0);
1415 num_refs = btrfs_ref_count_v0(leaf, ref0);
1423 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1424 struct btrfs_root *root,
1425 struct btrfs_path *path,
1426 u64 bytenr, u64 parent,
1429 struct btrfs_key key;
1432 key.objectid = bytenr;
1434 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1435 key.offset = parent;
1437 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1438 key.offset = root_objectid;
1441 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1444 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1445 if (ret == -ENOENT && parent) {
1446 btrfs_release_path(path);
1447 key.type = BTRFS_EXTENT_REF_V0_KEY;
1448 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1456 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1457 struct btrfs_root *root,
1458 struct btrfs_path *path,
1459 u64 bytenr, u64 parent,
1462 struct btrfs_key key;
1465 key.objectid = bytenr;
1467 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1468 key.offset = parent;
1470 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1471 key.offset = root_objectid;
1474 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1475 btrfs_release_path(path);
1479 static inline int extent_ref_type(u64 parent, u64 owner)
1482 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1484 type = BTRFS_SHARED_BLOCK_REF_KEY;
1486 type = BTRFS_TREE_BLOCK_REF_KEY;
1489 type = BTRFS_SHARED_DATA_REF_KEY;
1491 type = BTRFS_EXTENT_DATA_REF_KEY;
1496 static int find_next_key(struct btrfs_path *path, int level,
1497 struct btrfs_key *key)
1500 for (; level < BTRFS_MAX_LEVEL; level++) {
1501 if (!path->nodes[level])
1503 if (path->slots[level] + 1 >=
1504 btrfs_header_nritems(path->nodes[level]))
1507 btrfs_item_key_to_cpu(path->nodes[level], key,
1508 path->slots[level] + 1);
1510 btrfs_node_key_to_cpu(path->nodes[level], key,
1511 path->slots[level] + 1);
1518 * look for inline back ref. if back ref is found, *ref_ret is set
1519 * to the address of inline back ref, and 0 is returned.
1521 * if back ref isn't found, *ref_ret is set to the address where it
1522 * should be inserted, and -ENOENT is returned.
1524 * if insert is true and there are too many inline back refs, the path
1525 * points to the extent item, and -EAGAIN is returned.
1527 * NOTE: inline back refs are ordered in the same way that back ref
1528 * items in the tree are ordered.
1530 static noinline_for_stack
1531 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1532 struct btrfs_root *root,
1533 struct btrfs_path *path,
1534 struct btrfs_extent_inline_ref **ref_ret,
1535 u64 bytenr, u64 num_bytes,
1536 u64 parent, u64 root_objectid,
1537 u64 owner, u64 offset, int insert)
1539 struct btrfs_key key;
1540 struct extent_buffer *leaf;
1541 struct btrfs_extent_item *ei;
1542 struct btrfs_extent_inline_ref *iref;
1552 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1555 key.objectid = bytenr;
1556 key.type = BTRFS_EXTENT_ITEM_KEY;
1557 key.offset = num_bytes;
1559 want = extent_ref_type(parent, owner);
1561 extra_size = btrfs_extent_inline_ref_size(want);
1562 path->keep_locks = 1;
1567 * Owner is our parent level, so we can just add one to get the level
1568 * for the block we are interested in.
1570 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1571 key.type = BTRFS_METADATA_ITEM_KEY;
1576 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1583 * We may be a newly converted file system which still has the old fat
1584 * extent entries for metadata, so try and see if we have one of those.
1586 if (ret > 0 && skinny_metadata) {
1587 skinny_metadata = false;
1588 if (path->slots[0]) {
1590 btrfs_item_key_to_cpu(path->nodes[0], &key,
1592 if (key.objectid == bytenr &&
1593 key.type == BTRFS_EXTENT_ITEM_KEY &&
1594 key.offset == num_bytes)
1598 key.objectid = bytenr;
1599 key.type = BTRFS_EXTENT_ITEM_KEY;
1600 key.offset = num_bytes;
1601 btrfs_release_path(path);
1606 if (ret && !insert) {
1609 } else if (WARN_ON(ret)) {
1614 leaf = path->nodes[0];
1615 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1616 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1617 if (item_size < sizeof(*ei)) {
1622 ret = convert_extent_item_v0(trans, root, path, owner,
1628 leaf = path->nodes[0];
1629 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1632 BUG_ON(item_size < sizeof(*ei));
1634 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1635 flags = btrfs_extent_flags(leaf, ei);
1637 ptr = (unsigned long)(ei + 1);
1638 end = (unsigned long)ei + item_size;
1640 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1641 ptr += sizeof(struct btrfs_tree_block_info);
1651 iref = (struct btrfs_extent_inline_ref *)ptr;
1652 type = btrfs_extent_inline_ref_type(leaf, iref);
1656 ptr += btrfs_extent_inline_ref_size(type);
1660 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1661 struct btrfs_extent_data_ref *dref;
1662 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1663 if (match_extent_data_ref(leaf, dref, root_objectid,
1668 if (hash_extent_data_ref_item(leaf, dref) <
1669 hash_extent_data_ref(root_objectid, owner, offset))
1673 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1675 if (parent == ref_offset) {
1679 if (ref_offset < parent)
1682 if (root_objectid == ref_offset) {
1686 if (ref_offset < root_objectid)
1690 ptr += btrfs_extent_inline_ref_size(type);
1692 if (err == -ENOENT && insert) {
1693 if (item_size + extra_size >=
1694 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1699 * To add new inline back ref, we have to make sure
1700 * there is no corresponding back ref item.
1701 * For simplicity, we just do not add new inline back
1702 * ref if there is any kind of item for this block
1704 if (find_next_key(path, 0, &key) == 0 &&
1705 key.objectid == bytenr &&
1706 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1711 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1714 path->keep_locks = 0;
1715 btrfs_unlock_up_safe(path, 1);
1721 * helper to add new inline back ref
1723 static noinline_for_stack
1724 void setup_inline_extent_backref(struct btrfs_root *root,
1725 struct btrfs_path *path,
1726 struct btrfs_extent_inline_ref *iref,
1727 u64 parent, u64 root_objectid,
1728 u64 owner, u64 offset, int refs_to_add,
1729 struct btrfs_delayed_extent_op *extent_op)
1731 struct extent_buffer *leaf;
1732 struct btrfs_extent_item *ei;
1735 unsigned long item_offset;
1740 leaf = path->nodes[0];
1741 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1742 item_offset = (unsigned long)iref - (unsigned long)ei;
1744 type = extent_ref_type(parent, owner);
1745 size = btrfs_extent_inline_ref_size(type);
1747 btrfs_extend_item(root, path, size);
1749 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1750 refs = btrfs_extent_refs(leaf, ei);
1751 refs += refs_to_add;
1752 btrfs_set_extent_refs(leaf, ei, refs);
1754 __run_delayed_extent_op(extent_op, leaf, ei);
1756 ptr = (unsigned long)ei + item_offset;
1757 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1758 if (ptr < end - size)
1759 memmove_extent_buffer(leaf, ptr + size, ptr,
1762 iref = (struct btrfs_extent_inline_ref *)ptr;
1763 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1764 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1765 struct btrfs_extent_data_ref *dref;
1766 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1767 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1768 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1769 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1770 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1771 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1772 struct btrfs_shared_data_ref *sref;
1773 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1774 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1775 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1776 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1777 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1779 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1781 btrfs_mark_buffer_dirty(leaf);
1784 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1785 struct btrfs_root *root,
1786 struct btrfs_path *path,
1787 struct btrfs_extent_inline_ref **ref_ret,
1788 u64 bytenr, u64 num_bytes, u64 parent,
1789 u64 root_objectid, u64 owner, u64 offset)
1793 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1794 bytenr, num_bytes, parent,
1795 root_objectid, owner, offset, 0);
1799 btrfs_release_path(path);
1802 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1803 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1806 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1807 root_objectid, owner, offset);
1813 * helper to update/remove inline back ref
1815 static noinline_for_stack
1816 void update_inline_extent_backref(struct btrfs_root *root,
1817 struct btrfs_path *path,
1818 struct btrfs_extent_inline_ref *iref,
1820 struct btrfs_delayed_extent_op *extent_op,
1823 struct extent_buffer *leaf;
1824 struct btrfs_extent_item *ei;
1825 struct btrfs_extent_data_ref *dref = NULL;
1826 struct btrfs_shared_data_ref *sref = NULL;
1834 leaf = path->nodes[0];
1835 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1836 refs = btrfs_extent_refs(leaf, ei);
1837 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1838 refs += refs_to_mod;
1839 btrfs_set_extent_refs(leaf, ei, refs);
1841 __run_delayed_extent_op(extent_op, leaf, ei);
1843 type = btrfs_extent_inline_ref_type(leaf, iref);
1845 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1846 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1847 refs = btrfs_extent_data_ref_count(leaf, dref);
1848 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1849 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1850 refs = btrfs_shared_data_ref_count(leaf, sref);
1853 BUG_ON(refs_to_mod != -1);
1856 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1857 refs += refs_to_mod;
1860 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1861 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1863 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1866 size = btrfs_extent_inline_ref_size(type);
1867 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1868 ptr = (unsigned long)iref;
1869 end = (unsigned long)ei + item_size;
1870 if (ptr + size < end)
1871 memmove_extent_buffer(leaf, ptr, ptr + size,
1874 btrfs_truncate_item(root, path, item_size, 1);
1876 btrfs_mark_buffer_dirty(leaf);
1879 static noinline_for_stack
1880 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1881 struct btrfs_root *root,
1882 struct btrfs_path *path,
1883 u64 bytenr, u64 num_bytes, u64 parent,
1884 u64 root_objectid, u64 owner,
1885 u64 offset, int refs_to_add,
1886 struct btrfs_delayed_extent_op *extent_op)
1888 struct btrfs_extent_inline_ref *iref;
1891 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1892 bytenr, num_bytes, parent,
1893 root_objectid, owner, offset, 1);
1895 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1896 update_inline_extent_backref(root, path, iref,
1897 refs_to_add, extent_op, NULL);
1898 } else if (ret == -ENOENT) {
1899 setup_inline_extent_backref(root, path, iref, parent,
1900 root_objectid, owner, offset,
1901 refs_to_add, extent_op);
1907 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1908 struct btrfs_root *root,
1909 struct btrfs_path *path,
1910 u64 bytenr, u64 parent, u64 root_objectid,
1911 u64 owner, u64 offset, int refs_to_add)
1914 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1915 BUG_ON(refs_to_add != 1);
1916 ret = insert_tree_block_ref(trans, root, path, bytenr,
1917 parent, root_objectid);
1919 ret = insert_extent_data_ref(trans, root, path, bytenr,
1920 parent, root_objectid,
1921 owner, offset, refs_to_add);
1926 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1927 struct btrfs_root *root,
1928 struct btrfs_path *path,
1929 struct btrfs_extent_inline_ref *iref,
1930 int refs_to_drop, int is_data, int *last_ref)
1934 BUG_ON(!is_data && refs_to_drop != 1);
1936 update_inline_extent_backref(root, path, iref,
1937 -refs_to_drop, NULL, last_ref);
1938 } else if (is_data) {
1939 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1943 ret = btrfs_del_item(trans, root, path);
1948 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1949 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1950 u64 *discarded_bytes)
1953 u64 bytes_left, end;
1954 u64 aligned_start = ALIGN(start, 1 << 9);
1956 if (WARN_ON(start != aligned_start)) {
1957 len -= aligned_start - start;
1958 len = round_down(len, 1 << 9);
1959 start = aligned_start;
1962 *discarded_bytes = 0;
1970 /* Skip any superblocks on this device. */
1971 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1972 u64 sb_start = btrfs_sb_offset(j);
1973 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1974 u64 size = sb_start - start;
1976 if (!in_range(sb_start, start, bytes_left) &&
1977 !in_range(sb_end, start, bytes_left) &&
1978 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1982 * Superblock spans beginning of range. Adjust start and
1985 if (sb_start <= start) {
1986 start += sb_end - start;
1991 bytes_left = end - start;
1996 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1999 *discarded_bytes += size;
2000 else if (ret != -EOPNOTSUPP)
2009 bytes_left = end - start;
2013 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2016 *discarded_bytes += bytes_left;
2021 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2022 u64 num_bytes, u64 *actual_bytes)
2025 u64 discarded_bytes = 0;
2026 struct btrfs_bio *bbio = NULL;
2029 /* Tell the block device(s) that the sectors can be discarded */
2030 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2031 bytenr, &num_bytes, &bbio, 0);
2032 /* Error condition is -ENOMEM */
2034 struct btrfs_bio_stripe *stripe = bbio->stripes;
2038 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2040 if (!stripe->dev->can_discard)
2043 ret = btrfs_issue_discard(stripe->dev->bdev,
2048 discarded_bytes += bytes;
2049 else if (ret != -EOPNOTSUPP)
2050 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2053 * Just in case we get back EOPNOTSUPP for some reason,
2054 * just ignore the return value so we don't screw up
2055 * people calling discard_extent.
2059 btrfs_put_bbio(bbio);
2063 *actual_bytes = discarded_bytes;
2066 if (ret == -EOPNOTSUPP)
2071 /* Can return -ENOMEM */
2072 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2073 struct btrfs_root *root,
2074 u64 bytenr, u64 num_bytes, u64 parent,
2075 u64 root_objectid, u64 owner, u64 offset)
2078 struct btrfs_fs_info *fs_info = root->fs_info;
2080 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2081 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2083 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2084 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2086 parent, root_objectid, (int)owner,
2087 BTRFS_ADD_DELAYED_REF, NULL);
2089 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2090 num_bytes, parent, root_objectid,
2092 BTRFS_ADD_DELAYED_REF, NULL);
2097 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2098 struct btrfs_root *root,
2099 struct btrfs_delayed_ref_node *node,
2100 u64 parent, u64 root_objectid,
2101 u64 owner, u64 offset, int refs_to_add,
2102 struct btrfs_delayed_extent_op *extent_op)
2104 struct btrfs_fs_info *fs_info = root->fs_info;
2105 struct btrfs_path *path;
2106 struct extent_buffer *leaf;
2107 struct btrfs_extent_item *item;
2108 struct btrfs_key key;
2109 u64 bytenr = node->bytenr;
2110 u64 num_bytes = node->num_bytes;
2114 path = btrfs_alloc_path();
2119 path->leave_spinning = 1;
2120 /* this will setup the path even if it fails to insert the back ref */
2121 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2122 bytenr, num_bytes, parent,
2123 root_objectid, owner, offset,
2124 refs_to_add, extent_op);
2125 if ((ret < 0 && ret != -EAGAIN) || !ret)
2129 * Ok we had -EAGAIN which means we didn't have space to insert and
2130 * inline extent ref, so just update the reference count and add a
2133 leaf = path->nodes[0];
2134 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2135 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2136 refs = btrfs_extent_refs(leaf, item);
2137 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2139 __run_delayed_extent_op(extent_op, leaf, item);
2141 btrfs_mark_buffer_dirty(leaf);
2142 btrfs_release_path(path);
2145 path->leave_spinning = 1;
2146 /* now insert the actual backref */
2147 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2148 path, bytenr, parent, root_objectid,
2149 owner, offset, refs_to_add);
2151 btrfs_abort_transaction(trans, root, ret);
2153 btrfs_free_path(path);
2157 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2158 struct btrfs_root *root,
2159 struct btrfs_delayed_ref_node *node,
2160 struct btrfs_delayed_extent_op *extent_op,
2161 int insert_reserved)
2164 struct btrfs_delayed_data_ref *ref;
2165 struct btrfs_key ins;
2170 ins.objectid = node->bytenr;
2171 ins.offset = node->num_bytes;
2172 ins.type = BTRFS_EXTENT_ITEM_KEY;
2174 ref = btrfs_delayed_node_to_data_ref(node);
2175 trace_run_delayed_data_ref(node, ref, node->action);
2177 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2178 parent = ref->parent;
2179 ref_root = ref->root;
2181 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2183 flags |= extent_op->flags_to_set;
2184 ret = alloc_reserved_file_extent(trans, root,
2185 parent, ref_root, flags,
2186 ref->objectid, ref->offset,
2187 &ins, node->ref_mod);
2188 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2189 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2190 ref_root, ref->objectid,
2191 ref->offset, node->ref_mod,
2193 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2194 ret = __btrfs_free_extent(trans, root, node, parent,
2195 ref_root, ref->objectid,
2196 ref->offset, node->ref_mod,
2204 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2205 struct extent_buffer *leaf,
2206 struct btrfs_extent_item *ei)
2208 u64 flags = btrfs_extent_flags(leaf, ei);
2209 if (extent_op->update_flags) {
2210 flags |= extent_op->flags_to_set;
2211 btrfs_set_extent_flags(leaf, ei, flags);
2214 if (extent_op->update_key) {
2215 struct btrfs_tree_block_info *bi;
2216 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2217 bi = (struct btrfs_tree_block_info *)(ei + 1);
2218 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2222 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2223 struct btrfs_root *root,
2224 struct btrfs_delayed_ref_node *node,
2225 struct btrfs_delayed_extent_op *extent_op)
2227 struct btrfs_key key;
2228 struct btrfs_path *path;
2229 struct btrfs_extent_item *ei;
2230 struct extent_buffer *leaf;
2234 int metadata = !extent_op->is_data;
2239 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2242 path = btrfs_alloc_path();
2246 key.objectid = node->bytenr;
2249 key.type = BTRFS_METADATA_ITEM_KEY;
2250 key.offset = extent_op->level;
2252 key.type = BTRFS_EXTENT_ITEM_KEY;
2253 key.offset = node->num_bytes;
2258 path->leave_spinning = 1;
2259 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2267 if (path->slots[0] > 0) {
2269 btrfs_item_key_to_cpu(path->nodes[0], &key,
2271 if (key.objectid == node->bytenr &&
2272 key.type == BTRFS_EXTENT_ITEM_KEY &&
2273 key.offset == node->num_bytes)
2277 btrfs_release_path(path);
2280 key.objectid = node->bytenr;
2281 key.offset = node->num_bytes;
2282 key.type = BTRFS_EXTENT_ITEM_KEY;
2291 leaf = path->nodes[0];
2292 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2293 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2294 if (item_size < sizeof(*ei)) {
2295 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2301 leaf = path->nodes[0];
2302 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2305 BUG_ON(item_size < sizeof(*ei));
2306 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2307 __run_delayed_extent_op(extent_op, leaf, ei);
2309 btrfs_mark_buffer_dirty(leaf);
2311 btrfs_free_path(path);
2315 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2316 struct btrfs_root *root,
2317 struct btrfs_delayed_ref_node *node,
2318 struct btrfs_delayed_extent_op *extent_op,
2319 int insert_reserved)
2322 struct btrfs_delayed_tree_ref *ref;
2323 struct btrfs_key ins;
2326 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2329 ref = btrfs_delayed_node_to_tree_ref(node);
2330 trace_run_delayed_tree_ref(node, ref, node->action);
2332 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2333 parent = ref->parent;
2334 ref_root = ref->root;
2336 ins.objectid = node->bytenr;
2337 if (skinny_metadata) {
2338 ins.offset = ref->level;
2339 ins.type = BTRFS_METADATA_ITEM_KEY;
2341 ins.offset = node->num_bytes;
2342 ins.type = BTRFS_EXTENT_ITEM_KEY;
2345 BUG_ON(node->ref_mod != 1);
2346 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2347 BUG_ON(!extent_op || !extent_op->update_flags);
2348 ret = alloc_reserved_tree_block(trans, root,
2350 extent_op->flags_to_set,
2353 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2354 ret = __btrfs_inc_extent_ref(trans, root, node,
2358 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2359 ret = __btrfs_free_extent(trans, root, node,
2361 ref->level, 0, 1, extent_op);
2368 /* helper function to actually process a single delayed ref entry */
2369 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2370 struct btrfs_root *root,
2371 struct btrfs_delayed_ref_node *node,
2372 struct btrfs_delayed_extent_op *extent_op,
2373 int insert_reserved)
2377 if (trans->aborted) {
2378 if (insert_reserved)
2379 btrfs_pin_extent(root, node->bytenr,
2380 node->num_bytes, 1);
2384 if (btrfs_delayed_ref_is_head(node)) {
2385 struct btrfs_delayed_ref_head *head;
2387 * we've hit the end of the chain and we were supposed
2388 * to insert this extent into the tree. But, it got
2389 * deleted before we ever needed to insert it, so all
2390 * we have to do is clean up the accounting
2393 head = btrfs_delayed_node_to_head(node);
2394 trace_run_delayed_ref_head(node, head, node->action);
2396 if (insert_reserved) {
2397 btrfs_pin_extent(root, node->bytenr,
2398 node->num_bytes, 1);
2399 if (head->is_data) {
2400 ret = btrfs_del_csums(trans, root,
2406 /* Also free its reserved qgroup space */
2407 btrfs_qgroup_free_delayed_ref(root->fs_info,
2408 head->qgroup_ref_root,
2409 head->qgroup_reserved);
2413 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2414 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2415 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2417 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2418 node->type == BTRFS_SHARED_DATA_REF_KEY)
2419 ret = run_delayed_data_ref(trans, root, node, extent_op,
2426 static inline struct btrfs_delayed_ref_node *
2427 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2429 struct btrfs_delayed_ref_node *ref;
2431 if (list_empty(&head->ref_list))
2435 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2436 * This is to prevent a ref count from going down to zero, which deletes
2437 * the extent item from the extent tree, when there still are references
2438 * to add, which would fail because they would not find the extent item.
2440 list_for_each_entry(ref, &head->ref_list, list) {
2441 if (ref->action == BTRFS_ADD_DELAYED_REF)
2445 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2450 * Returns 0 on success or if called with an already aborted transaction.
2451 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2453 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2454 struct btrfs_root *root,
2457 struct btrfs_delayed_ref_root *delayed_refs;
2458 struct btrfs_delayed_ref_node *ref;
2459 struct btrfs_delayed_ref_head *locked_ref = NULL;
2460 struct btrfs_delayed_extent_op *extent_op;
2461 struct btrfs_fs_info *fs_info = root->fs_info;
2462 ktime_t start = ktime_get();
2464 unsigned long count = 0;
2465 unsigned long actual_count = 0;
2466 int must_insert_reserved = 0;
2468 delayed_refs = &trans->transaction->delayed_refs;
2474 spin_lock(&delayed_refs->lock);
2475 locked_ref = btrfs_select_ref_head(trans);
2477 spin_unlock(&delayed_refs->lock);
2481 /* grab the lock that says we are going to process
2482 * all the refs for this head */
2483 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2484 spin_unlock(&delayed_refs->lock);
2486 * we may have dropped the spin lock to get the head
2487 * mutex lock, and that might have given someone else
2488 * time to free the head. If that's true, it has been
2489 * removed from our list and we can move on.
2491 if (ret == -EAGAIN) {
2499 * We need to try and merge add/drops of the same ref since we
2500 * can run into issues with relocate dropping the implicit ref
2501 * and then it being added back again before the drop can
2502 * finish. If we merged anything we need to re-loop so we can
2504 * Or we can get node references of the same type that weren't
2505 * merged when created due to bumps in the tree mod seq, and
2506 * we need to merge them to prevent adding an inline extent
2507 * backref before dropping it (triggering a BUG_ON at
2508 * insert_inline_extent_backref()).
2510 spin_lock(&locked_ref->lock);
2511 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2515 * locked_ref is the head node, so we have to go one
2516 * node back for any delayed ref updates
2518 ref = select_delayed_ref(locked_ref);
2520 if (ref && ref->seq &&
2521 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2522 spin_unlock(&locked_ref->lock);
2523 btrfs_delayed_ref_unlock(locked_ref);
2524 spin_lock(&delayed_refs->lock);
2525 locked_ref->processing = 0;
2526 delayed_refs->num_heads_ready++;
2527 spin_unlock(&delayed_refs->lock);
2535 * record the must insert reserved flag before we
2536 * drop the spin lock.
2538 must_insert_reserved = locked_ref->must_insert_reserved;
2539 locked_ref->must_insert_reserved = 0;
2541 extent_op = locked_ref->extent_op;
2542 locked_ref->extent_op = NULL;
2547 /* All delayed refs have been processed, Go ahead
2548 * and send the head node to run_one_delayed_ref,
2549 * so that any accounting fixes can happen
2551 ref = &locked_ref->node;
2553 if (extent_op && must_insert_reserved) {
2554 btrfs_free_delayed_extent_op(extent_op);
2559 spin_unlock(&locked_ref->lock);
2560 ret = run_delayed_extent_op(trans, root,
2562 btrfs_free_delayed_extent_op(extent_op);
2566 * Need to reset must_insert_reserved if
2567 * there was an error so the abort stuff
2568 * can cleanup the reserved space
2571 if (must_insert_reserved)
2572 locked_ref->must_insert_reserved = 1;
2573 locked_ref->processing = 0;
2574 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2575 btrfs_delayed_ref_unlock(locked_ref);
2582 * Need to drop our head ref lock and re-aqcuire the
2583 * delayed ref lock and then re-check to make sure
2586 spin_unlock(&locked_ref->lock);
2587 spin_lock(&delayed_refs->lock);
2588 spin_lock(&locked_ref->lock);
2589 if (!list_empty(&locked_ref->ref_list) ||
2590 locked_ref->extent_op) {
2591 spin_unlock(&locked_ref->lock);
2592 spin_unlock(&delayed_refs->lock);
2596 delayed_refs->num_heads--;
2597 rb_erase(&locked_ref->href_node,
2598 &delayed_refs->href_root);
2599 spin_unlock(&delayed_refs->lock);
2603 list_del(&ref->list);
2605 atomic_dec(&delayed_refs->num_entries);
2607 if (!btrfs_delayed_ref_is_head(ref)) {
2609 * when we play the delayed ref, also correct the
2612 switch (ref->action) {
2613 case BTRFS_ADD_DELAYED_REF:
2614 case BTRFS_ADD_DELAYED_EXTENT:
2615 locked_ref->node.ref_mod -= ref->ref_mod;
2617 case BTRFS_DROP_DELAYED_REF:
2618 locked_ref->node.ref_mod += ref->ref_mod;
2624 spin_unlock(&locked_ref->lock);
2626 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2627 must_insert_reserved);
2629 btrfs_free_delayed_extent_op(extent_op);
2631 locked_ref->processing = 0;
2632 btrfs_delayed_ref_unlock(locked_ref);
2633 btrfs_put_delayed_ref(ref);
2634 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2639 * If this node is a head, that means all the refs in this head
2640 * have been dealt with, and we will pick the next head to deal
2641 * with, so we must unlock the head and drop it from the cluster
2642 * list before we release it.
2644 if (btrfs_delayed_ref_is_head(ref)) {
2645 if (locked_ref->is_data &&
2646 locked_ref->total_ref_mod < 0) {
2647 spin_lock(&delayed_refs->lock);
2648 delayed_refs->pending_csums -= ref->num_bytes;
2649 spin_unlock(&delayed_refs->lock);
2651 btrfs_delayed_ref_unlock(locked_ref);
2654 btrfs_put_delayed_ref(ref);
2660 * We don't want to include ref heads since we can have empty ref heads
2661 * and those will drastically skew our runtime down since we just do
2662 * accounting, no actual extent tree updates.
2664 if (actual_count > 0) {
2665 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2669 * We weigh the current average higher than our current runtime
2670 * to avoid large swings in the average.
2672 spin_lock(&delayed_refs->lock);
2673 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2674 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2675 spin_unlock(&delayed_refs->lock);
2680 #ifdef SCRAMBLE_DELAYED_REFS
2682 * Normally delayed refs get processed in ascending bytenr order. This
2683 * correlates in most cases to the order added. To expose dependencies on this
2684 * order, we start to process the tree in the middle instead of the beginning
2686 static u64 find_middle(struct rb_root *root)
2688 struct rb_node *n = root->rb_node;
2689 struct btrfs_delayed_ref_node *entry;
2692 u64 first = 0, last = 0;
2696 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2697 first = entry->bytenr;
2701 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2702 last = entry->bytenr;
2707 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2708 WARN_ON(!entry->in_tree);
2710 middle = entry->bytenr;
2723 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2727 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2728 sizeof(struct btrfs_extent_inline_ref));
2729 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2730 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2733 * We don't ever fill up leaves all the way so multiply by 2 just to be
2734 * closer to what we're really going to want to ouse.
2736 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2740 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2741 * would require to store the csums for that many bytes.
2743 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2746 u64 num_csums_per_leaf;
2749 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2750 num_csums_per_leaf = div64_u64(csum_size,
2751 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2752 num_csums = div64_u64(csum_bytes, root->sectorsize);
2753 num_csums += num_csums_per_leaf - 1;
2754 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2758 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2759 struct btrfs_root *root)
2761 struct btrfs_block_rsv *global_rsv;
2762 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2763 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2764 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2765 u64 num_bytes, num_dirty_bgs_bytes;
2768 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2769 num_heads = heads_to_leaves(root, num_heads);
2771 num_bytes += (num_heads - 1) * root->nodesize;
2773 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2774 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2776 global_rsv = &root->fs_info->global_block_rsv;
2779 * If we can't allocate any more chunks lets make sure we have _lots_ of
2780 * wiggle room since running delayed refs can create more delayed refs.
2782 if (global_rsv->space_info->full) {
2783 num_dirty_bgs_bytes <<= 1;
2787 spin_lock(&global_rsv->lock);
2788 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2790 spin_unlock(&global_rsv->lock);
2794 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2795 struct btrfs_root *root)
2797 struct btrfs_fs_info *fs_info = root->fs_info;
2799 atomic_read(&trans->transaction->delayed_refs.num_entries);
2804 avg_runtime = fs_info->avg_delayed_ref_runtime;
2805 val = num_entries * avg_runtime;
2806 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2808 if (val >= NSEC_PER_SEC / 2)
2811 return btrfs_check_space_for_delayed_refs(trans, root);
2814 struct async_delayed_refs {
2815 struct btrfs_root *root;
2819 struct completion wait;
2820 struct btrfs_work work;
2823 static void delayed_ref_async_start(struct btrfs_work *work)
2825 struct async_delayed_refs *async;
2826 struct btrfs_trans_handle *trans;
2829 async = container_of(work, struct async_delayed_refs, work);
2831 trans = btrfs_join_transaction(async->root);
2832 if (IS_ERR(trans)) {
2833 async->error = PTR_ERR(trans);
2838 * trans->sync means that when we call end_transaciton, we won't
2839 * wait on delayed refs
2842 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2846 ret = btrfs_end_transaction(trans, async->root);
2847 if (ret && !async->error)
2851 complete(&async->wait);
2856 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2857 unsigned long count, int wait)
2859 struct async_delayed_refs *async;
2862 async = kmalloc(sizeof(*async), GFP_NOFS);
2866 async->root = root->fs_info->tree_root;
2867 async->count = count;
2873 init_completion(&async->wait);
2875 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2876 delayed_ref_async_start, NULL, NULL);
2878 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2881 wait_for_completion(&async->wait);
2890 * this starts processing the delayed reference count updates and
2891 * extent insertions we have queued up so far. count can be
2892 * 0, which means to process everything in the tree at the start
2893 * of the run (but not newly added entries), or it can be some target
2894 * number you'd like to process.
2896 * Returns 0 on success or if called with an aborted transaction
2897 * Returns <0 on error and aborts the transaction
2899 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2900 struct btrfs_root *root, unsigned long count)
2902 struct rb_node *node;
2903 struct btrfs_delayed_ref_root *delayed_refs;
2904 struct btrfs_delayed_ref_head *head;
2906 int run_all = count == (unsigned long)-1;
2907 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2909 /* We'll clean this up in btrfs_cleanup_transaction */
2913 if (root == root->fs_info->extent_root)
2914 root = root->fs_info->tree_root;
2916 delayed_refs = &trans->transaction->delayed_refs;
2918 count = atomic_read(&delayed_refs->num_entries) * 2;
2921 #ifdef SCRAMBLE_DELAYED_REFS
2922 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2924 trans->can_flush_pending_bgs = false;
2925 ret = __btrfs_run_delayed_refs(trans, root, count);
2927 btrfs_abort_transaction(trans, root, ret);
2932 if (!list_empty(&trans->new_bgs))
2933 btrfs_create_pending_block_groups(trans, root);
2935 spin_lock(&delayed_refs->lock);
2936 node = rb_first(&delayed_refs->href_root);
2938 spin_unlock(&delayed_refs->lock);
2941 count = (unsigned long)-1;
2944 head = rb_entry(node, struct btrfs_delayed_ref_head,
2946 if (btrfs_delayed_ref_is_head(&head->node)) {
2947 struct btrfs_delayed_ref_node *ref;
2950 atomic_inc(&ref->refs);
2952 spin_unlock(&delayed_refs->lock);
2954 * Mutex was contended, block until it's
2955 * released and try again
2957 mutex_lock(&head->mutex);
2958 mutex_unlock(&head->mutex);
2960 btrfs_put_delayed_ref(ref);
2966 node = rb_next(node);
2968 spin_unlock(&delayed_refs->lock);
2973 assert_qgroups_uptodate(trans);
2974 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2978 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2979 struct btrfs_root *root,
2980 u64 bytenr, u64 num_bytes, u64 flags,
2981 int level, int is_data)
2983 struct btrfs_delayed_extent_op *extent_op;
2986 extent_op = btrfs_alloc_delayed_extent_op();
2990 extent_op->flags_to_set = flags;
2991 extent_op->update_flags = 1;
2992 extent_op->update_key = 0;
2993 extent_op->is_data = is_data ? 1 : 0;
2994 extent_op->level = level;
2996 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2997 num_bytes, extent_op);
2999 btrfs_free_delayed_extent_op(extent_op);
3003 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3004 struct btrfs_root *root,
3005 struct btrfs_path *path,
3006 u64 objectid, u64 offset, u64 bytenr)
3008 struct btrfs_delayed_ref_head *head;
3009 struct btrfs_delayed_ref_node *ref;
3010 struct btrfs_delayed_data_ref *data_ref;
3011 struct btrfs_delayed_ref_root *delayed_refs;
3014 delayed_refs = &trans->transaction->delayed_refs;
3015 spin_lock(&delayed_refs->lock);
3016 head = btrfs_find_delayed_ref_head(trans, bytenr);
3018 spin_unlock(&delayed_refs->lock);
3022 if (!mutex_trylock(&head->mutex)) {
3023 atomic_inc(&head->node.refs);
3024 spin_unlock(&delayed_refs->lock);
3026 btrfs_release_path(path);
3029 * Mutex was contended, block until it's released and let
3032 mutex_lock(&head->mutex);
3033 mutex_unlock(&head->mutex);
3034 btrfs_put_delayed_ref(&head->node);
3037 spin_unlock(&delayed_refs->lock);
3039 spin_lock(&head->lock);
3040 list_for_each_entry(ref, &head->ref_list, list) {
3041 /* If it's a shared ref we know a cross reference exists */
3042 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3047 data_ref = btrfs_delayed_node_to_data_ref(ref);
3050 * If our ref doesn't match the one we're currently looking at
3051 * then we have a cross reference.
3053 if (data_ref->root != root->root_key.objectid ||
3054 data_ref->objectid != objectid ||
3055 data_ref->offset != offset) {
3060 spin_unlock(&head->lock);
3061 mutex_unlock(&head->mutex);
3065 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3066 struct btrfs_root *root,
3067 struct btrfs_path *path,
3068 u64 objectid, u64 offset, u64 bytenr)
3070 struct btrfs_root *extent_root = root->fs_info->extent_root;
3071 struct extent_buffer *leaf;
3072 struct btrfs_extent_data_ref *ref;
3073 struct btrfs_extent_inline_ref *iref;
3074 struct btrfs_extent_item *ei;
3075 struct btrfs_key key;
3079 key.objectid = bytenr;
3080 key.offset = (u64)-1;
3081 key.type = BTRFS_EXTENT_ITEM_KEY;
3083 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3086 BUG_ON(ret == 0); /* Corruption */
3089 if (path->slots[0] == 0)
3093 leaf = path->nodes[0];
3094 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3096 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3100 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3101 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3102 if (item_size < sizeof(*ei)) {
3103 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3107 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3109 if (item_size != sizeof(*ei) +
3110 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3113 if (btrfs_extent_generation(leaf, ei) <=
3114 btrfs_root_last_snapshot(&root->root_item))
3117 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3118 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3119 BTRFS_EXTENT_DATA_REF_KEY)
3122 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3123 if (btrfs_extent_refs(leaf, ei) !=
3124 btrfs_extent_data_ref_count(leaf, ref) ||
3125 btrfs_extent_data_ref_root(leaf, ref) !=
3126 root->root_key.objectid ||
3127 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3128 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3136 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3137 struct btrfs_root *root,
3138 u64 objectid, u64 offset, u64 bytenr)
3140 struct btrfs_path *path;
3144 path = btrfs_alloc_path();
3149 ret = check_committed_ref(trans, root, path, objectid,
3151 if (ret && ret != -ENOENT)
3154 ret2 = check_delayed_ref(trans, root, path, objectid,
3156 } while (ret2 == -EAGAIN);
3158 if (ret2 && ret2 != -ENOENT) {
3163 if (ret != -ENOENT || ret2 != -ENOENT)
3166 btrfs_free_path(path);
3167 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3172 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3173 struct btrfs_root *root,
3174 struct extent_buffer *buf,
3175 int full_backref, int inc)
3182 struct btrfs_key key;
3183 struct btrfs_file_extent_item *fi;
3187 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3188 u64, u64, u64, u64, u64, u64);
3191 if (btrfs_test_is_dummy_root(root))
3194 ref_root = btrfs_header_owner(buf);
3195 nritems = btrfs_header_nritems(buf);
3196 level = btrfs_header_level(buf);
3198 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3202 process_func = btrfs_inc_extent_ref;
3204 process_func = btrfs_free_extent;
3207 parent = buf->start;
3211 for (i = 0; i < nritems; i++) {
3213 btrfs_item_key_to_cpu(buf, &key, i);
3214 if (key.type != BTRFS_EXTENT_DATA_KEY)
3216 fi = btrfs_item_ptr(buf, i,
3217 struct btrfs_file_extent_item);
3218 if (btrfs_file_extent_type(buf, fi) ==
3219 BTRFS_FILE_EXTENT_INLINE)
3221 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3225 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3226 key.offset -= btrfs_file_extent_offset(buf, fi);
3227 ret = process_func(trans, root, bytenr, num_bytes,
3228 parent, ref_root, key.objectid,
3233 bytenr = btrfs_node_blockptr(buf, i);
3234 num_bytes = root->nodesize;
3235 ret = process_func(trans, root, bytenr, num_bytes,
3236 parent, ref_root, level - 1, 0);
3246 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3247 struct extent_buffer *buf, int full_backref)
3249 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3252 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3253 struct extent_buffer *buf, int full_backref)
3255 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3258 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3259 struct btrfs_root *root,
3260 struct btrfs_path *path,
3261 struct btrfs_block_group_cache *cache)
3264 struct btrfs_root *extent_root = root->fs_info->extent_root;
3266 struct extent_buffer *leaf;
3268 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3275 leaf = path->nodes[0];
3276 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3277 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3278 btrfs_mark_buffer_dirty(leaf);
3280 btrfs_release_path(path);
3285 static struct btrfs_block_group_cache *
3286 next_block_group(struct btrfs_root *root,
3287 struct btrfs_block_group_cache *cache)
3289 struct rb_node *node;
3291 spin_lock(&root->fs_info->block_group_cache_lock);
3293 /* If our block group was removed, we need a full search. */
3294 if (RB_EMPTY_NODE(&cache->cache_node)) {
3295 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3297 spin_unlock(&root->fs_info->block_group_cache_lock);
3298 btrfs_put_block_group(cache);
3299 cache = btrfs_lookup_first_block_group(root->fs_info,
3303 node = rb_next(&cache->cache_node);
3304 btrfs_put_block_group(cache);
3306 cache = rb_entry(node, struct btrfs_block_group_cache,
3308 btrfs_get_block_group(cache);
3311 spin_unlock(&root->fs_info->block_group_cache_lock);
3315 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3316 struct btrfs_trans_handle *trans,
3317 struct btrfs_path *path)
3319 struct btrfs_root *root = block_group->fs_info->tree_root;
3320 struct inode *inode = NULL;
3322 int dcs = BTRFS_DC_ERROR;
3328 * If this block group is smaller than 100 megs don't bother caching the
3331 if (block_group->key.offset < (100 * 1024 * 1024)) {
3332 spin_lock(&block_group->lock);
3333 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3334 spin_unlock(&block_group->lock);
3341 inode = lookup_free_space_inode(root, block_group, path);
3342 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3343 ret = PTR_ERR(inode);
3344 btrfs_release_path(path);
3348 if (IS_ERR(inode)) {
3352 if (block_group->ro)
3355 ret = create_free_space_inode(root, trans, block_group, path);
3361 /* We've already setup this transaction, go ahead and exit */
3362 if (block_group->cache_generation == trans->transid &&
3363 i_size_read(inode)) {
3364 dcs = BTRFS_DC_SETUP;
3369 * We want to set the generation to 0, that way if anything goes wrong
3370 * from here on out we know not to trust this cache when we load up next
3373 BTRFS_I(inode)->generation = 0;
3374 ret = btrfs_update_inode(trans, root, inode);
3377 * So theoretically we could recover from this, simply set the
3378 * super cache generation to 0 so we know to invalidate the
3379 * cache, but then we'd have to keep track of the block groups
3380 * that fail this way so we know we _have_ to reset this cache
3381 * before the next commit or risk reading stale cache. So to
3382 * limit our exposure to horrible edge cases lets just abort the
3383 * transaction, this only happens in really bad situations
3386 btrfs_abort_transaction(trans, root, ret);
3391 if (i_size_read(inode) > 0) {
3392 ret = btrfs_check_trunc_cache_free_space(root,
3393 &root->fs_info->global_block_rsv);
3397 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3402 spin_lock(&block_group->lock);
3403 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3404 !btrfs_test_opt(root, SPACE_CACHE)) {
3406 * don't bother trying to write stuff out _if_
3407 * a) we're not cached,
3408 * b) we're with nospace_cache mount option.
3410 dcs = BTRFS_DC_WRITTEN;
3411 spin_unlock(&block_group->lock);
3414 spin_unlock(&block_group->lock);
3417 * We hit an ENOSPC when setting up the cache in this transaction, just
3418 * skip doing the setup, we've already cleared the cache so we're safe.
3420 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3426 * Try to preallocate enough space based on how big the block group is.
3427 * Keep in mind this has to include any pinned space which could end up
3428 * taking up quite a bit since it's not folded into the other space
3431 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3436 num_pages *= PAGE_CACHE_SIZE;
3438 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3442 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3443 num_pages, num_pages,
3446 * Our cache requires contiguous chunks so that we don't modify a bunch
3447 * of metadata or split extents when writing the cache out, which means
3448 * we can enospc if we are heavily fragmented in addition to just normal
3449 * out of space conditions. So if we hit this just skip setting up any
3450 * other block groups for this transaction, maybe we'll unpin enough
3451 * space the next time around.
3454 dcs = BTRFS_DC_SETUP;
3455 else if (ret == -ENOSPC)
3456 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3457 btrfs_free_reserved_data_space(inode, 0, num_pages);
3462 btrfs_release_path(path);
3464 spin_lock(&block_group->lock);
3465 if (!ret && dcs == BTRFS_DC_SETUP)
3466 block_group->cache_generation = trans->transid;
3467 block_group->disk_cache_state = dcs;
3468 spin_unlock(&block_group->lock);
3473 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3474 struct btrfs_root *root)
3476 struct btrfs_block_group_cache *cache, *tmp;
3477 struct btrfs_transaction *cur_trans = trans->transaction;
3478 struct btrfs_path *path;
3480 if (list_empty(&cur_trans->dirty_bgs) ||
3481 !btrfs_test_opt(root, SPACE_CACHE))
3484 path = btrfs_alloc_path();
3488 /* Could add new block groups, use _safe just in case */
3489 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3491 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3492 cache_save_setup(cache, trans, path);
3495 btrfs_free_path(path);
3500 * transaction commit does final block group cache writeback during a
3501 * critical section where nothing is allowed to change the FS. This is
3502 * required in order for the cache to actually match the block group,
3503 * but can introduce a lot of latency into the commit.
3505 * So, btrfs_start_dirty_block_groups is here to kick off block group
3506 * cache IO. There's a chance we'll have to redo some of it if the
3507 * block group changes again during the commit, but it greatly reduces
3508 * the commit latency by getting rid of the easy block groups while
3509 * we're still allowing others to join the commit.
3511 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3512 struct btrfs_root *root)
3514 struct btrfs_block_group_cache *cache;
3515 struct btrfs_transaction *cur_trans = trans->transaction;
3518 struct btrfs_path *path = NULL;
3520 struct list_head *io = &cur_trans->io_bgs;
3521 int num_started = 0;
3524 spin_lock(&cur_trans->dirty_bgs_lock);
3525 if (list_empty(&cur_trans->dirty_bgs)) {
3526 spin_unlock(&cur_trans->dirty_bgs_lock);
3529 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3530 spin_unlock(&cur_trans->dirty_bgs_lock);
3534 * make sure all the block groups on our dirty list actually
3537 btrfs_create_pending_block_groups(trans, root);
3540 path = btrfs_alloc_path();
3546 * cache_write_mutex is here only to save us from balance or automatic
3547 * removal of empty block groups deleting this block group while we are
3548 * writing out the cache
3550 mutex_lock(&trans->transaction->cache_write_mutex);
3551 while (!list_empty(&dirty)) {
3552 cache = list_first_entry(&dirty,
3553 struct btrfs_block_group_cache,
3556 * this can happen if something re-dirties a block
3557 * group that is already under IO. Just wait for it to
3558 * finish and then do it all again
3560 if (!list_empty(&cache->io_list)) {
3561 list_del_init(&cache->io_list);
3562 btrfs_wait_cache_io(root, trans, cache,
3563 &cache->io_ctl, path,
3564 cache->key.objectid);
3565 btrfs_put_block_group(cache);
3570 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3571 * if it should update the cache_state. Don't delete
3572 * until after we wait.
3574 * Since we're not running in the commit critical section
3575 * we need the dirty_bgs_lock to protect from update_block_group
3577 spin_lock(&cur_trans->dirty_bgs_lock);
3578 list_del_init(&cache->dirty_list);
3579 spin_unlock(&cur_trans->dirty_bgs_lock);
3583 cache_save_setup(cache, trans, path);
3585 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3586 cache->io_ctl.inode = NULL;
3587 ret = btrfs_write_out_cache(root, trans, cache, path);
3588 if (ret == 0 && cache->io_ctl.inode) {
3593 * the cache_write_mutex is protecting
3596 list_add_tail(&cache->io_list, io);
3599 * if we failed to write the cache, the
3600 * generation will be bad and life goes on
3606 ret = write_one_cache_group(trans, root, path, cache);
3608 * Our block group might still be attached to the list
3609 * of new block groups in the transaction handle of some
3610 * other task (struct btrfs_trans_handle->new_bgs). This
3611 * means its block group item isn't yet in the extent
3612 * tree. If this happens ignore the error, as we will
3613 * try again later in the critical section of the
3614 * transaction commit.
3616 if (ret == -ENOENT) {
3618 spin_lock(&cur_trans->dirty_bgs_lock);
3619 if (list_empty(&cache->dirty_list)) {
3620 list_add_tail(&cache->dirty_list,
3621 &cur_trans->dirty_bgs);
3622 btrfs_get_block_group(cache);
3624 spin_unlock(&cur_trans->dirty_bgs_lock);
3626 btrfs_abort_transaction(trans, root, ret);
3630 /* if its not on the io list, we need to put the block group */
3632 btrfs_put_block_group(cache);
3638 * Avoid blocking other tasks for too long. It might even save
3639 * us from writing caches for block groups that are going to be
3642 mutex_unlock(&trans->transaction->cache_write_mutex);
3643 mutex_lock(&trans->transaction->cache_write_mutex);
3645 mutex_unlock(&trans->transaction->cache_write_mutex);
3648 * go through delayed refs for all the stuff we've just kicked off
3649 * and then loop back (just once)
3651 ret = btrfs_run_delayed_refs(trans, root, 0);
3652 if (!ret && loops == 0) {
3654 spin_lock(&cur_trans->dirty_bgs_lock);
3655 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3657 * dirty_bgs_lock protects us from concurrent block group
3658 * deletes too (not just cache_write_mutex).
3660 if (!list_empty(&dirty)) {
3661 spin_unlock(&cur_trans->dirty_bgs_lock);
3664 spin_unlock(&cur_trans->dirty_bgs_lock);
3667 btrfs_free_path(path);
3671 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3672 struct btrfs_root *root)
3674 struct btrfs_block_group_cache *cache;
3675 struct btrfs_transaction *cur_trans = trans->transaction;
3678 struct btrfs_path *path;
3679 struct list_head *io = &cur_trans->io_bgs;
3680 int num_started = 0;
3682 path = btrfs_alloc_path();
3687 * We don't need the lock here since we are protected by the transaction
3688 * commit. We want to do the cache_save_setup first and then run the
3689 * delayed refs to make sure we have the best chance at doing this all
3692 while (!list_empty(&cur_trans->dirty_bgs)) {
3693 cache = list_first_entry(&cur_trans->dirty_bgs,
3694 struct btrfs_block_group_cache,
3698 * this can happen if cache_save_setup re-dirties a block
3699 * group that is already under IO. Just wait for it to
3700 * finish and then do it all again
3702 if (!list_empty(&cache->io_list)) {
3703 list_del_init(&cache->io_list);
3704 btrfs_wait_cache_io(root, trans, cache,
3705 &cache->io_ctl, path,
3706 cache->key.objectid);
3707 btrfs_put_block_group(cache);
3711 * don't remove from the dirty list until after we've waited
3714 list_del_init(&cache->dirty_list);
3717 cache_save_setup(cache, trans, path);
3720 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3722 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3723 cache->io_ctl.inode = NULL;
3724 ret = btrfs_write_out_cache(root, trans, cache, path);
3725 if (ret == 0 && cache->io_ctl.inode) {
3728 list_add_tail(&cache->io_list, io);
3731 * if we failed to write the cache, the
3732 * generation will be bad and life goes on
3738 ret = write_one_cache_group(trans, root, path, cache);
3740 btrfs_abort_transaction(trans, root, ret);
3743 /* if its not on the io list, we need to put the block group */
3745 btrfs_put_block_group(cache);
3748 while (!list_empty(io)) {
3749 cache = list_first_entry(io, struct btrfs_block_group_cache,
3751 list_del_init(&cache->io_list);
3752 btrfs_wait_cache_io(root, trans, cache,
3753 &cache->io_ctl, path, cache->key.objectid);
3754 btrfs_put_block_group(cache);
3757 btrfs_free_path(path);
3761 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3763 struct btrfs_block_group_cache *block_group;
3766 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3767 if (!block_group || block_group->ro)
3770 btrfs_put_block_group(block_group);
3774 static const char *alloc_name(u64 flags)
3777 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3779 case BTRFS_BLOCK_GROUP_METADATA:
3781 case BTRFS_BLOCK_GROUP_DATA:
3783 case BTRFS_BLOCK_GROUP_SYSTEM:
3787 return "invalid-combination";
3791 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3792 u64 total_bytes, u64 bytes_used,
3793 struct btrfs_space_info **space_info)
3795 struct btrfs_space_info *found;
3800 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3801 BTRFS_BLOCK_GROUP_RAID10))
3806 found = __find_space_info(info, flags);
3808 spin_lock(&found->lock);
3809 found->total_bytes += total_bytes;
3810 found->disk_total += total_bytes * factor;
3811 found->bytes_used += bytes_used;
3812 found->disk_used += bytes_used * factor;
3813 if (total_bytes > 0)
3815 spin_unlock(&found->lock);
3816 *space_info = found;
3819 found = kzalloc(sizeof(*found), GFP_NOFS);
3823 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3829 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3830 INIT_LIST_HEAD(&found->block_groups[i]);
3831 init_rwsem(&found->groups_sem);
3832 spin_lock_init(&found->lock);
3833 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3834 found->total_bytes = total_bytes;
3835 found->disk_total = total_bytes * factor;
3836 found->bytes_used = bytes_used;
3837 found->disk_used = bytes_used * factor;
3838 found->bytes_pinned = 0;
3839 found->bytes_reserved = 0;
3840 found->bytes_readonly = 0;
3841 found->bytes_may_use = 0;
3843 found->max_extent_size = 0;
3844 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3845 found->chunk_alloc = 0;
3847 init_waitqueue_head(&found->wait);
3848 INIT_LIST_HEAD(&found->ro_bgs);
3850 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3851 info->space_info_kobj, "%s",
3852 alloc_name(found->flags));
3858 *space_info = found;
3859 list_add_rcu(&found->list, &info->space_info);
3860 if (flags & BTRFS_BLOCK_GROUP_DATA)
3861 info->data_sinfo = found;
3866 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3868 u64 extra_flags = chunk_to_extended(flags) &
3869 BTRFS_EXTENDED_PROFILE_MASK;
3871 write_seqlock(&fs_info->profiles_lock);
3872 if (flags & BTRFS_BLOCK_GROUP_DATA)
3873 fs_info->avail_data_alloc_bits |= extra_flags;
3874 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3875 fs_info->avail_metadata_alloc_bits |= extra_flags;
3876 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3877 fs_info->avail_system_alloc_bits |= extra_flags;
3878 write_sequnlock(&fs_info->profiles_lock);
3882 * returns target flags in extended format or 0 if restripe for this
3883 * chunk_type is not in progress
3885 * should be called with either volume_mutex or balance_lock held
3887 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3889 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3895 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3896 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3897 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3898 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3899 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3900 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3901 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3902 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3903 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3910 * @flags: available profiles in extended format (see ctree.h)
3912 * Returns reduced profile in chunk format. If profile changing is in
3913 * progress (either running or paused) picks the target profile (if it's
3914 * already available), otherwise falls back to plain reducing.
3916 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3918 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3924 * see if restripe for this chunk_type is in progress, if so
3925 * try to reduce to the target profile
3927 spin_lock(&root->fs_info->balance_lock);
3928 target = get_restripe_target(root->fs_info, flags);
3930 /* pick target profile only if it's already available */
3931 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3932 spin_unlock(&root->fs_info->balance_lock);
3933 return extended_to_chunk(target);
3936 spin_unlock(&root->fs_info->balance_lock);
3938 /* First, mask out the RAID levels which aren't possible */
3939 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3940 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
3941 allowed |= btrfs_raid_group[raid_type];
3945 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
3946 allowed = BTRFS_BLOCK_GROUP_RAID6;
3947 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
3948 allowed = BTRFS_BLOCK_GROUP_RAID5;
3949 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
3950 allowed = BTRFS_BLOCK_GROUP_RAID10;
3951 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
3952 allowed = BTRFS_BLOCK_GROUP_RAID1;
3953 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
3954 allowed = BTRFS_BLOCK_GROUP_RAID0;
3956 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
3958 return extended_to_chunk(flags | allowed);
3961 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3968 seq = read_seqbegin(&root->fs_info->profiles_lock);
3970 if (flags & BTRFS_BLOCK_GROUP_DATA)
3971 flags |= root->fs_info->avail_data_alloc_bits;
3972 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3973 flags |= root->fs_info->avail_system_alloc_bits;
3974 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3975 flags |= root->fs_info->avail_metadata_alloc_bits;
3976 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3978 return btrfs_reduce_alloc_profile(root, flags);
3981 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3987 flags = BTRFS_BLOCK_GROUP_DATA;
3988 else if (root == root->fs_info->chunk_root)
3989 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3991 flags = BTRFS_BLOCK_GROUP_METADATA;
3993 ret = get_alloc_profile(root, flags);
3997 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
3999 struct btrfs_space_info *data_sinfo;
4000 struct btrfs_root *root = BTRFS_I(inode)->root;
4001 struct btrfs_fs_info *fs_info = root->fs_info;
4004 int need_commit = 2;
4005 int have_pinned_space;
4007 /* make sure bytes are sectorsize aligned */
4008 bytes = ALIGN(bytes, root->sectorsize);
4010 if (btrfs_is_free_space_inode(inode)) {
4012 ASSERT(current->journal_info);
4015 data_sinfo = fs_info->data_sinfo;
4020 /* make sure we have enough space to handle the data first */
4021 spin_lock(&data_sinfo->lock);
4022 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4023 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4024 data_sinfo->bytes_may_use;
4026 if (used + bytes > data_sinfo->total_bytes) {
4027 struct btrfs_trans_handle *trans;
4030 * if we don't have enough free bytes in this space then we need
4031 * to alloc a new chunk.
4033 if (!data_sinfo->full) {
4036 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4037 spin_unlock(&data_sinfo->lock);
4039 alloc_target = btrfs_get_alloc_profile(root, 1);
4041 * It is ugly that we don't call nolock join
4042 * transaction for the free space inode case here.
4043 * But it is safe because we only do the data space
4044 * reservation for the free space cache in the
4045 * transaction context, the common join transaction
4046 * just increase the counter of the current transaction
4047 * handler, doesn't try to acquire the trans_lock of
4050 trans = btrfs_join_transaction(root);
4052 return PTR_ERR(trans);
4054 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4056 CHUNK_ALLOC_NO_FORCE);
4057 btrfs_end_transaction(trans, root);
4062 have_pinned_space = 1;
4068 data_sinfo = fs_info->data_sinfo;
4074 * If we don't have enough pinned space to deal with this
4075 * allocation, and no removed chunk in current transaction,
4076 * don't bother committing the transaction.
4078 have_pinned_space = percpu_counter_compare(
4079 &data_sinfo->total_bytes_pinned,
4080 used + bytes - data_sinfo->total_bytes);
4081 spin_unlock(&data_sinfo->lock);
4083 /* commit the current transaction and try again */
4086 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4089 if (need_commit > 0)
4090 btrfs_wait_ordered_roots(fs_info, -1);
4092 trans = btrfs_join_transaction(root);
4094 return PTR_ERR(trans);
4095 if (have_pinned_space >= 0 ||
4096 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4097 &trans->transaction->flags) ||
4099 ret = btrfs_commit_transaction(trans, root);
4103 * The cleaner kthread might still be doing iput
4104 * operations. Wait for it to finish so that
4105 * more space is released.
4107 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4108 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4111 btrfs_end_transaction(trans, root);
4115 trace_btrfs_space_reservation(root->fs_info,
4116 "space_info:enospc",
4117 data_sinfo->flags, bytes, 1);
4120 data_sinfo->bytes_may_use += bytes;
4121 trace_btrfs_space_reservation(root->fs_info, "space_info",
4122 data_sinfo->flags, bytes, 1);
4123 spin_unlock(&data_sinfo->lock);
4129 * New check_data_free_space() with ability for precious data reservation
4130 * Will replace old btrfs_check_data_free_space(), but for patch split,
4131 * add a new function first and then replace it.
4133 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4135 struct btrfs_root *root = BTRFS_I(inode)->root;
4138 /* align the range */
4139 len = round_up(start + len, root->sectorsize) -
4140 round_down(start, root->sectorsize);
4141 start = round_down(start, root->sectorsize);
4143 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4148 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4150 * TODO: Find a good method to avoid reserve data space for NOCOW
4151 * range, but don't impact performance on quota disable case.
4153 ret = btrfs_qgroup_reserve_data(inode, start, len);
4158 * Called if we need to clear a data reservation for this inode
4159 * Normally in a error case.
4161 * This one will *NOT* use accurate qgroup reserved space API, just for case
4162 * which we can't sleep and is sure it won't affect qgroup reserved space.
4163 * Like clear_bit_hook().
4165 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4168 struct btrfs_root *root = BTRFS_I(inode)->root;
4169 struct btrfs_space_info *data_sinfo;
4171 /* Make sure the range is aligned to sectorsize */
4172 len = round_up(start + len, root->sectorsize) -
4173 round_down(start, root->sectorsize);
4174 start = round_down(start, root->sectorsize);
4176 data_sinfo = root->fs_info->data_sinfo;
4177 spin_lock(&data_sinfo->lock);
4178 if (WARN_ON(data_sinfo->bytes_may_use < len))
4179 data_sinfo->bytes_may_use = 0;
4181 data_sinfo->bytes_may_use -= len;
4182 trace_btrfs_space_reservation(root->fs_info, "space_info",
4183 data_sinfo->flags, len, 0);
4184 spin_unlock(&data_sinfo->lock);
4188 * Called if we need to clear a data reservation for this inode
4189 * Normally in a error case.
4191 * This one will handle the per-indoe data rsv map for accurate reserved
4194 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4196 btrfs_free_reserved_data_space_noquota(inode, start, len);
4197 btrfs_qgroup_free_data(inode, start, len);
4200 static void force_metadata_allocation(struct btrfs_fs_info *info)
4202 struct list_head *head = &info->space_info;
4203 struct btrfs_space_info *found;
4206 list_for_each_entry_rcu(found, head, list) {
4207 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4208 found->force_alloc = CHUNK_ALLOC_FORCE;
4213 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4215 return (global->size << 1);
4218 static int should_alloc_chunk(struct btrfs_root *root,
4219 struct btrfs_space_info *sinfo, int force)
4221 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4222 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4223 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4226 if (force == CHUNK_ALLOC_FORCE)
4230 * We need to take into account the global rsv because for all intents
4231 * and purposes it's used space. Don't worry about locking the
4232 * global_rsv, it doesn't change except when the transaction commits.
4234 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4235 num_allocated += calc_global_rsv_need_space(global_rsv);
4238 * in limited mode, we want to have some free space up to
4239 * about 1% of the FS size.
4241 if (force == CHUNK_ALLOC_LIMITED) {
4242 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4243 thresh = max_t(u64, 64 * 1024 * 1024,
4244 div_factor_fine(thresh, 1));
4246 if (num_bytes - num_allocated < thresh)
4250 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4255 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4259 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4260 BTRFS_BLOCK_GROUP_RAID0 |
4261 BTRFS_BLOCK_GROUP_RAID5 |
4262 BTRFS_BLOCK_GROUP_RAID6))
4263 num_dev = root->fs_info->fs_devices->rw_devices;
4264 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4267 num_dev = 1; /* DUP or single */
4273 * If @is_allocation is true, reserve space in the system space info necessary
4274 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4277 void check_system_chunk(struct btrfs_trans_handle *trans,
4278 struct btrfs_root *root,
4281 struct btrfs_space_info *info;
4288 * Needed because we can end up allocating a system chunk and for an
4289 * atomic and race free space reservation in the chunk block reserve.
4291 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4293 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4294 spin_lock(&info->lock);
4295 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4296 info->bytes_reserved - info->bytes_readonly -
4297 info->bytes_may_use;
4298 spin_unlock(&info->lock);
4300 num_devs = get_profile_num_devs(root, type);
4302 /* num_devs device items to update and 1 chunk item to add or remove */
4303 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4304 btrfs_calc_trans_metadata_size(root, 1);
4306 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4307 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4308 left, thresh, type);
4309 dump_space_info(info, 0, 0);
4312 if (left < thresh) {
4315 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4317 * Ignore failure to create system chunk. We might end up not
4318 * needing it, as we might not need to COW all nodes/leafs from
4319 * the paths we visit in the chunk tree (they were already COWed
4320 * or created in the current transaction for example).
4322 ret = btrfs_alloc_chunk(trans, root, flags);
4326 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4327 &root->fs_info->chunk_block_rsv,
4328 thresh, BTRFS_RESERVE_NO_FLUSH);
4330 trans->chunk_bytes_reserved += thresh;
4334 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4335 struct btrfs_root *extent_root, u64 flags, int force)
4337 struct btrfs_space_info *space_info;
4338 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4339 int wait_for_alloc = 0;
4342 /* Don't re-enter if we're already allocating a chunk */
4343 if (trans->allocating_chunk)
4346 space_info = __find_space_info(extent_root->fs_info, flags);
4348 ret = update_space_info(extent_root->fs_info, flags,
4350 BUG_ON(ret); /* -ENOMEM */
4352 BUG_ON(!space_info); /* Logic error */
4355 spin_lock(&space_info->lock);
4356 if (force < space_info->force_alloc)
4357 force = space_info->force_alloc;
4358 if (space_info->full) {
4359 if (should_alloc_chunk(extent_root, space_info, force))
4363 spin_unlock(&space_info->lock);
4367 if (!should_alloc_chunk(extent_root, space_info, force)) {
4368 spin_unlock(&space_info->lock);
4370 } else if (space_info->chunk_alloc) {
4373 space_info->chunk_alloc = 1;
4376 spin_unlock(&space_info->lock);
4378 mutex_lock(&fs_info->chunk_mutex);
4381 * The chunk_mutex is held throughout the entirety of a chunk
4382 * allocation, so once we've acquired the chunk_mutex we know that the
4383 * other guy is done and we need to recheck and see if we should
4386 if (wait_for_alloc) {
4387 mutex_unlock(&fs_info->chunk_mutex);
4392 trans->allocating_chunk = true;
4395 * If we have mixed data/metadata chunks we want to make sure we keep
4396 * allocating mixed chunks instead of individual chunks.
4398 if (btrfs_mixed_space_info(space_info))
4399 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4402 * if we're doing a data chunk, go ahead and make sure that
4403 * we keep a reasonable number of metadata chunks allocated in the
4406 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4407 fs_info->data_chunk_allocations++;
4408 if (!(fs_info->data_chunk_allocations %
4409 fs_info->metadata_ratio))
4410 force_metadata_allocation(fs_info);
4414 * Check if we have enough space in SYSTEM chunk because we may need
4415 * to update devices.
4417 check_system_chunk(trans, extent_root, flags);
4419 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4420 trans->allocating_chunk = false;
4422 spin_lock(&space_info->lock);
4423 if (ret < 0 && ret != -ENOSPC)
4426 space_info->full = 1;
4430 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4432 space_info->chunk_alloc = 0;
4433 spin_unlock(&space_info->lock);
4434 mutex_unlock(&fs_info->chunk_mutex);
4436 * When we allocate a new chunk we reserve space in the chunk block
4437 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4438 * add new nodes/leafs to it if we end up needing to do it when
4439 * inserting the chunk item and updating device items as part of the
4440 * second phase of chunk allocation, performed by
4441 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4442 * large number of new block groups to create in our transaction
4443 * handle's new_bgs list to avoid exhausting the chunk block reserve
4444 * in extreme cases - like having a single transaction create many new
4445 * block groups when starting to write out the free space caches of all
4446 * the block groups that were made dirty during the lifetime of the
4449 if (trans->can_flush_pending_bgs &&
4450 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4451 btrfs_create_pending_block_groups(trans, trans->root);
4452 btrfs_trans_release_chunk_metadata(trans);
4457 static int can_overcommit(struct btrfs_root *root,
4458 struct btrfs_space_info *space_info, u64 bytes,
4459 enum btrfs_reserve_flush_enum flush)
4461 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4462 u64 profile = btrfs_get_alloc_profile(root, 0);
4467 used = space_info->bytes_used + space_info->bytes_reserved +
4468 space_info->bytes_pinned + space_info->bytes_readonly;
4471 * We only want to allow over committing if we have lots of actual space
4472 * free, but if we don't have enough space to handle the global reserve
4473 * space then we could end up having a real enospc problem when trying
4474 * to allocate a chunk or some other such important allocation.
4476 spin_lock(&global_rsv->lock);
4477 space_size = calc_global_rsv_need_space(global_rsv);
4478 spin_unlock(&global_rsv->lock);
4479 if (used + space_size >= space_info->total_bytes)
4482 used += space_info->bytes_may_use;
4484 spin_lock(&root->fs_info->free_chunk_lock);
4485 avail = root->fs_info->free_chunk_space;
4486 spin_unlock(&root->fs_info->free_chunk_lock);
4489 * If we have dup, raid1 or raid10 then only half of the free
4490 * space is actually useable. For raid56, the space info used
4491 * doesn't include the parity drive, so we don't have to
4494 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4495 BTRFS_BLOCK_GROUP_RAID1 |
4496 BTRFS_BLOCK_GROUP_RAID10))
4500 * If we aren't flushing all things, let us overcommit up to
4501 * 1/2th of the space. If we can flush, don't let us overcommit
4502 * too much, let it overcommit up to 1/8 of the space.
4504 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4509 if (used + bytes < space_info->total_bytes + avail)
4514 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4515 unsigned long nr_pages, int nr_items)
4517 struct super_block *sb = root->fs_info->sb;
4519 if (down_read_trylock(&sb->s_umount)) {
4520 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4521 up_read(&sb->s_umount);
4524 * We needn't worry the filesystem going from r/w to r/o though
4525 * we don't acquire ->s_umount mutex, because the filesystem
4526 * should guarantee the delalloc inodes list be empty after
4527 * the filesystem is readonly(all dirty pages are written to
4530 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4531 if (!current->journal_info)
4532 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4536 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4541 bytes = btrfs_calc_trans_metadata_size(root, 1);
4542 nr = (int)div64_u64(to_reclaim, bytes);
4548 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4551 * shrink metadata reservation for delalloc
4553 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4556 struct btrfs_block_rsv *block_rsv;
4557 struct btrfs_space_info *space_info;
4558 struct btrfs_trans_handle *trans;
4562 unsigned long nr_pages;
4565 enum btrfs_reserve_flush_enum flush;
4567 /* Calc the number of the pages we need flush for space reservation */
4568 items = calc_reclaim_items_nr(root, to_reclaim);
4569 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4571 trans = (struct btrfs_trans_handle *)current->journal_info;
4572 block_rsv = &root->fs_info->delalloc_block_rsv;
4573 space_info = block_rsv->space_info;
4575 delalloc_bytes = percpu_counter_sum_positive(
4576 &root->fs_info->delalloc_bytes);
4577 if (delalloc_bytes == 0) {
4581 btrfs_wait_ordered_roots(root->fs_info, items);
4586 while (delalloc_bytes && loops < 3) {
4587 max_reclaim = min(delalloc_bytes, to_reclaim);
4588 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4589 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4591 * We need to wait for the async pages to actually start before
4594 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4598 if (max_reclaim <= nr_pages)
4601 max_reclaim -= nr_pages;
4603 wait_event(root->fs_info->async_submit_wait,
4604 atomic_read(&root->fs_info->async_delalloc_pages) <=
4608 flush = BTRFS_RESERVE_FLUSH_ALL;
4610 flush = BTRFS_RESERVE_NO_FLUSH;
4611 spin_lock(&space_info->lock);
4612 if (can_overcommit(root, space_info, orig, flush)) {
4613 spin_unlock(&space_info->lock);
4616 spin_unlock(&space_info->lock);
4619 if (wait_ordered && !trans) {
4620 btrfs_wait_ordered_roots(root->fs_info, items);
4622 time_left = schedule_timeout_killable(1);
4626 delalloc_bytes = percpu_counter_sum_positive(
4627 &root->fs_info->delalloc_bytes);
4632 * maybe_commit_transaction - possibly commit the transaction if its ok to
4633 * @root - the root we're allocating for
4634 * @bytes - the number of bytes we want to reserve
4635 * @force - force the commit
4637 * This will check to make sure that committing the transaction will actually
4638 * get us somewhere and then commit the transaction if it does. Otherwise it
4639 * will return -ENOSPC.
4641 static int may_commit_transaction(struct btrfs_root *root,
4642 struct btrfs_space_info *space_info,
4643 u64 bytes, int force)
4645 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4646 struct btrfs_trans_handle *trans;
4648 trans = (struct btrfs_trans_handle *)current->journal_info;
4655 /* See if there is enough pinned space to make this reservation */
4656 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4661 * See if there is some space in the delayed insertion reservation for
4664 if (space_info != delayed_rsv->space_info)
4667 spin_lock(&delayed_rsv->lock);
4668 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4669 bytes - delayed_rsv->size) >= 0) {
4670 spin_unlock(&delayed_rsv->lock);
4673 spin_unlock(&delayed_rsv->lock);
4676 trans = btrfs_join_transaction(root);
4680 return btrfs_commit_transaction(trans, root);
4684 FLUSH_DELAYED_ITEMS_NR = 1,
4685 FLUSH_DELAYED_ITEMS = 2,
4687 FLUSH_DELALLOC_WAIT = 4,
4692 static int flush_space(struct btrfs_root *root,
4693 struct btrfs_space_info *space_info, u64 num_bytes,
4694 u64 orig_bytes, int state)
4696 struct btrfs_trans_handle *trans;
4701 case FLUSH_DELAYED_ITEMS_NR:
4702 case FLUSH_DELAYED_ITEMS:
4703 if (state == FLUSH_DELAYED_ITEMS_NR)
4704 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4708 trans = btrfs_join_transaction(root);
4709 if (IS_ERR(trans)) {
4710 ret = PTR_ERR(trans);
4713 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4714 btrfs_end_transaction(trans, root);
4716 case FLUSH_DELALLOC:
4717 case FLUSH_DELALLOC_WAIT:
4718 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4719 state == FLUSH_DELALLOC_WAIT);
4722 trans = btrfs_join_transaction(root);
4723 if (IS_ERR(trans)) {
4724 ret = PTR_ERR(trans);
4727 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4728 btrfs_get_alloc_profile(root, 0),
4729 CHUNK_ALLOC_NO_FORCE);
4730 btrfs_end_transaction(trans, root);
4735 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4746 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4747 struct btrfs_space_info *space_info)
4753 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4755 spin_lock(&space_info->lock);
4756 if (can_overcommit(root, space_info, to_reclaim,
4757 BTRFS_RESERVE_FLUSH_ALL)) {
4762 used = space_info->bytes_used + space_info->bytes_reserved +
4763 space_info->bytes_pinned + space_info->bytes_readonly +
4764 space_info->bytes_may_use;
4765 if (can_overcommit(root, space_info, 1024 * 1024,
4766 BTRFS_RESERVE_FLUSH_ALL))
4767 expected = div_factor_fine(space_info->total_bytes, 95);
4769 expected = div_factor_fine(space_info->total_bytes, 90);
4771 if (used > expected)
4772 to_reclaim = used - expected;
4775 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4776 space_info->bytes_reserved);
4778 spin_unlock(&space_info->lock);
4783 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4784 struct btrfs_fs_info *fs_info, u64 used)
4786 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4788 /* If we're just plain full then async reclaim just slows us down. */
4789 if (space_info->bytes_used >= thresh)
4792 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4793 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4796 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4797 struct btrfs_fs_info *fs_info,
4802 spin_lock(&space_info->lock);
4804 * We run out of space and have not got any free space via flush_space,
4805 * so don't bother doing async reclaim.
4807 if (flush_state > COMMIT_TRANS && space_info->full) {
4808 spin_unlock(&space_info->lock);
4812 used = space_info->bytes_used + space_info->bytes_reserved +
4813 space_info->bytes_pinned + space_info->bytes_readonly +
4814 space_info->bytes_may_use;
4815 if (need_do_async_reclaim(space_info, fs_info, used)) {
4816 spin_unlock(&space_info->lock);
4819 spin_unlock(&space_info->lock);
4824 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4826 struct btrfs_fs_info *fs_info;
4827 struct btrfs_space_info *space_info;
4831 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4832 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4834 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4839 flush_state = FLUSH_DELAYED_ITEMS_NR;
4841 flush_space(fs_info->fs_root, space_info, to_reclaim,
4842 to_reclaim, flush_state);
4844 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4847 } while (flush_state < COMMIT_TRANS);
4850 void btrfs_init_async_reclaim_work(struct work_struct *work)
4852 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4856 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4857 * @root - the root we're allocating for
4858 * @block_rsv - the block_rsv we're allocating for
4859 * @orig_bytes - the number of bytes we want
4860 * @flush - whether or not we can flush to make our reservation
4862 * This will reserve orgi_bytes number of bytes from the space info associated
4863 * with the block_rsv. If there is not enough space it will make an attempt to
4864 * flush out space to make room. It will do this by flushing delalloc if
4865 * possible or committing the transaction. If flush is 0 then no attempts to
4866 * regain reservations will be made and this will fail if there is not enough
4869 static int reserve_metadata_bytes(struct btrfs_root *root,
4870 struct btrfs_block_rsv *block_rsv,
4872 enum btrfs_reserve_flush_enum flush)
4874 struct btrfs_space_info *space_info = block_rsv->space_info;
4876 u64 num_bytes = orig_bytes;
4877 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4879 bool flushing = false;
4883 spin_lock(&space_info->lock);
4885 * We only want to wait if somebody other than us is flushing and we
4886 * are actually allowed to flush all things.
4888 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4889 space_info->flush) {
4890 spin_unlock(&space_info->lock);
4892 * If we have a trans handle we can't wait because the flusher
4893 * may have to commit the transaction, which would mean we would
4894 * deadlock since we are waiting for the flusher to finish, but
4895 * hold the current transaction open.
4897 if (current->journal_info)
4899 ret = wait_event_killable(space_info->wait, !space_info->flush);
4900 /* Must have been killed, return */
4904 spin_lock(&space_info->lock);
4908 used = space_info->bytes_used + space_info->bytes_reserved +
4909 space_info->bytes_pinned + space_info->bytes_readonly +
4910 space_info->bytes_may_use;
4913 * The idea here is that we've not already over-reserved the block group
4914 * then we can go ahead and save our reservation first and then start
4915 * flushing if we need to. Otherwise if we've already overcommitted
4916 * lets start flushing stuff first and then come back and try to make
4919 if (used <= space_info->total_bytes) {
4920 if (used + orig_bytes <= space_info->total_bytes) {
4921 space_info->bytes_may_use += orig_bytes;
4922 trace_btrfs_space_reservation(root->fs_info,
4923 "space_info", space_info->flags, orig_bytes, 1);
4927 * Ok set num_bytes to orig_bytes since we aren't
4928 * overocmmitted, this way we only try and reclaim what
4931 num_bytes = orig_bytes;
4935 * Ok we're over committed, set num_bytes to the overcommitted
4936 * amount plus the amount of bytes that we need for this
4939 num_bytes = used - space_info->total_bytes +
4943 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4944 space_info->bytes_may_use += orig_bytes;
4945 trace_btrfs_space_reservation(root->fs_info, "space_info",
4946 space_info->flags, orig_bytes,
4952 * Couldn't make our reservation, save our place so while we're trying
4953 * to reclaim space we can actually use it instead of somebody else
4954 * stealing it from us.
4956 * We make the other tasks wait for the flush only when we can flush
4959 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4961 space_info->flush = 1;
4962 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4965 * We will do the space reservation dance during log replay,
4966 * which means we won't have fs_info->fs_root set, so don't do
4967 * the async reclaim as we will panic.
4969 if (!root->fs_info->log_root_recovering &&
4970 need_do_async_reclaim(space_info, root->fs_info, used) &&
4971 !work_busy(&root->fs_info->async_reclaim_work))
4972 queue_work(system_unbound_wq,
4973 &root->fs_info->async_reclaim_work);
4975 spin_unlock(&space_info->lock);
4977 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4980 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4985 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4986 * would happen. So skip delalloc flush.
4988 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4989 (flush_state == FLUSH_DELALLOC ||
4990 flush_state == FLUSH_DELALLOC_WAIT))
4991 flush_state = ALLOC_CHUNK;
4995 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4996 flush_state < COMMIT_TRANS)
4998 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4999 flush_state <= COMMIT_TRANS)
5003 if (ret == -ENOSPC &&
5004 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5005 struct btrfs_block_rsv *global_rsv =
5006 &root->fs_info->global_block_rsv;
5008 if (block_rsv != global_rsv &&
5009 !block_rsv_use_bytes(global_rsv, orig_bytes))
5013 trace_btrfs_space_reservation(root->fs_info,
5014 "space_info:enospc",
5015 space_info->flags, orig_bytes, 1);
5017 spin_lock(&space_info->lock);
5018 space_info->flush = 0;
5019 wake_up_all(&space_info->wait);
5020 spin_unlock(&space_info->lock);
5025 static struct btrfs_block_rsv *get_block_rsv(
5026 const struct btrfs_trans_handle *trans,
5027 const struct btrfs_root *root)
5029 struct btrfs_block_rsv *block_rsv = NULL;
5031 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5032 (root == root->fs_info->csum_root && trans->adding_csums) ||
5033 (root == root->fs_info->uuid_root))
5034 block_rsv = trans->block_rsv;
5037 block_rsv = root->block_rsv;
5040 block_rsv = &root->fs_info->empty_block_rsv;
5045 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5049 spin_lock(&block_rsv->lock);
5050 if (block_rsv->reserved >= num_bytes) {
5051 block_rsv->reserved -= num_bytes;
5052 if (block_rsv->reserved < block_rsv->size)
5053 block_rsv->full = 0;
5056 spin_unlock(&block_rsv->lock);
5060 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5061 u64 num_bytes, int update_size)
5063 spin_lock(&block_rsv->lock);
5064 block_rsv->reserved += num_bytes;
5066 block_rsv->size += num_bytes;
5067 else if (block_rsv->reserved >= block_rsv->size)
5068 block_rsv->full = 1;
5069 spin_unlock(&block_rsv->lock);
5072 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5073 struct btrfs_block_rsv *dest, u64 num_bytes,
5076 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5079 if (global_rsv->space_info != dest->space_info)
5082 spin_lock(&global_rsv->lock);
5083 min_bytes = div_factor(global_rsv->size, min_factor);
5084 if (global_rsv->reserved < min_bytes + num_bytes) {
5085 spin_unlock(&global_rsv->lock);
5088 global_rsv->reserved -= num_bytes;
5089 if (global_rsv->reserved < global_rsv->size)
5090 global_rsv->full = 0;
5091 spin_unlock(&global_rsv->lock);
5093 block_rsv_add_bytes(dest, num_bytes, 1);
5097 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5098 struct btrfs_block_rsv *block_rsv,
5099 struct btrfs_block_rsv *dest, u64 num_bytes)
5101 struct btrfs_space_info *space_info = block_rsv->space_info;
5103 spin_lock(&block_rsv->lock);
5104 if (num_bytes == (u64)-1)
5105 num_bytes = block_rsv->size;
5106 block_rsv->size -= num_bytes;
5107 if (block_rsv->reserved >= block_rsv->size) {
5108 num_bytes = block_rsv->reserved - block_rsv->size;
5109 block_rsv->reserved = block_rsv->size;
5110 block_rsv->full = 1;
5114 spin_unlock(&block_rsv->lock);
5116 if (num_bytes > 0) {
5118 spin_lock(&dest->lock);
5122 bytes_to_add = dest->size - dest->reserved;
5123 bytes_to_add = min(num_bytes, bytes_to_add);
5124 dest->reserved += bytes_to_add;
5125 if (dest->reserved >= dest->size)
5127 num_bytes -= bytes_to_add;
5129 spin_unlock(&dest->lock);
5132 spin_lock(&space_info->lock);
5133 space_info->bytes_may_use -= num_bytes;
5134 trace_btrfs_space_reservation(fs_info, "space_info",
5135 space_info->flags, num_bytes, 0);
5136 spin_unlock(&space_info->lock);
5141 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5142 struct btrfs_block_rsv *dst, u64 num_bytes)
5146 ret = block_rsv_use_bytes(src, num_bytes);
5150 block_rsv_add_bytes(dst, num_bytes, 1);
5154 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5156 memset(rsv, 0, sizeof(*rsv));
5157 spin_lock_init(&rsv->lock);
5161 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5162 unsigned short type)
5164 struct btrfs_block_rsv *block_rsv;
5165 struct btrfs_fs_info *fs_info = root->fs_info;
5167 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5171 btrfs_init_block_rsv(block_rsv, type);
5172 block_rsv->space_info = __find_space_info(fs_info,
5173 BTRFS_BLOCK_GROUP_METADATA);
5177 void btrfs_free_block_rsv(struct btrfs_root *root,
5178 struct btrfs_block_rsv *rsv)
5182 btrfs_block_rsv_release(root, rsv, (u64)-1);
5186 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5191 int btrfs_block_rsv_add(struct btrfs_root *root,
5192 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5193 enum btrfs_reserve_flush_enum flush)
5200 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5202 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5209 int btrfs_block_rsv_check(struct btrfs_root *root,
5210 struct btrfs_block_rsv *block_rsv, int min_factor)
5218 spin_lock(&block_rsv->lock);
5219 num_bytes = div_factor(block_rsv->size, min_factor);
5220 if (block_rsv->reserved >= num_bytes)
5222 spin_unlock(&block_rsv->lock);
5227 int btrfs_block_rsv_refill(struct btrfs_root *root,
5228 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5229 enum btrfs_reserve_flush_enum flush)
5237 spin_lock(&block_rsv->lock);
5238 num_bytes = min_reserved;
5239 if (block_rsv->reserved >= num_bytes)
5242 num_bytes -= block_rsv->reserved;
5243 spin_unlock(&block_rsv->lock);
5248 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5250 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5257 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5258 struct btrfs_block_rsv *dst_rsv,
5261 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5264 void btrfs_block_rsv_release(struct btrfs_root *root,
5265 struct btrfs_block_rsv *block_rsv,
5268 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5269 if (global_rsv == block_rsv ||
5270 block_rsv->space_info != global_rsv->space_info)
5272 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5277 * helper to calculate size of global block reservation.
5278 * the desired value is sum of space used by extent tree,
5279 * checksum tree and root tree
5281 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5283 struct btrfs_space_info *sinfo;
5287 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5289 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5290 spin_lock(&sinfo->lock);
5291 data_used = sinfo->bytes_used;
5292 spin_unlock(&sinfo->lock);
5294 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5295 spin_lock(&sinfo->lock);
5296 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5298 meta_used = sinfo->bytes_used;
5299 spin_unlock(&sinfo->lock);
5301 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5303 num_bytes += div_u64(data_used + meta_used, 50);
5305 if (num_bytes * 3 > meta_used)
5306 num_bytes = div_u64(meta_used, 3);
5308 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5311 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5313 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5314 struct btrfs_space_info *sinfo = block_rsv->space_info;
5317 num_bytes = calc_global_metadata_size(fs_info);
5319 spin_lock(&sinfo->lock);
5320 spin_lock(&block_rsv->lock);
5322 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5324 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5325 sinfo->bytes_reserved + sinfo->bytes_readonly +
5326 sinfo->bytes_may_use;
5328 if (sinfo->total_bytes > num_bytes) {
5329 num_bytes = sinfo->total_bytes - num_bytes;
5330 block_rsv->reserved += num_bytes;
5331 sinfo->bytes_may_use += num_bytes;
5332 trace_btrfs_space_reservation(fs_info, "space_info",
5333 sinfo->flags, num_bytes, 1);
5336 if (block_rsv->reserved >= block_rsv->size) {
5337 num_bytes = block_rsv->reserved - block_rsv->size;
5338 sinfo->bytes_may_use -= num_bytes;
5339 trace_btrfs_space_reservation(fs_info, "space_info",
5340 sinfo->flags, num_bytes, 0);
5341 block_rsv->reserved = block_rsv->size;
5342 block_rsv->full = 1;
5345 spin_unlock(&block_rsv->lock);
5346 spin_unlock(&sinfo->lock);
5349 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5351 struct btrfs_space_info *space_info;
5353 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5354 fs_info->chunk_block_rsv.space_info = space_info;
5356 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5357 fs_info->global_block_rsv.space_info = space_info;
5358 fs_info->delalloc_block_rsv.space_info = space_info;
5359 fs_info->trans_block_rsv.space_info = space_info;
5360 fs_info->empty_block_rsv.space_info = space_info;
5361 fs_info->delayed_block_rsv.space_info = space_info;
5363 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5364 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5365 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5366 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5367 if (fs_info->quota_root)
5368 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5369 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5371 update_global_block_rsv(fs_info);
5374 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5376 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5378 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5379 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5380 WARN_ON(fs_info->trans_block_rsv.size > 0);
5381 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5382 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5383 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5384 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5385 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5388 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5389 struct btrfs_root *root)
5391 if (!trans->block_rsv)
5394 if (!trans->bytes_reserved)
5397 trace_btrfs_space_reservation(root->fs_info, "transaction",
5398 trans->transid, trans->bytes_reserved, 0);
5399 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5400 trans->bytes_reserved = 0;
5404 * To be called after all the new block groups attached to the transaction
5405 * handle have been created (btrfs_create_pending_block_groups()).
5407 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5409 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5411 if (!trans->chunk_bytes_reserved)
5414 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5416 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5417 trans->chunk_bytes_reserved);
5418 trans->chunk_bytes_reserved = 0;
5421 /* Can only return 0 or -ENOSPC */
5422 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5423 struct inode *inode)
5425 struct btrfs_root *root = BTRFS_I(inode)->root;
5426 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5427 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5430 * We need to hold space in order to delete our orphan item once we've
5431 * added it, so this takes the reservation so we can release it later
5432 * when we are truly done with the orphan item.
5434 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5435 trace_btrfs_space_reservation(root->fs_info, "orphan",
5436 btrfs_ino(inode), num_bytes, 1);
5437 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5440 void btrfs_orphan_release_metadata(struct inode *inode)
5442 struct btrfs_root *root = BTRFS_I(inode)->root;
5443 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5444 trace_btrfs_space_reservation(root->fs_info, "orphan",
5445 btrfs_ino(inode), num_bytes, 0);
5446 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5450 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5451 * root: the root of the parent directory
5452 * rsv: block reservation
5453 * items: the number of items that we need do reservation
5454 * qgroup_reserved: used to return the reserved size in qgroup
5456 * This function is used to reserve the space for snapshot/subvolume
5457 * creation and deletion. Those operations are different with the
5458 * common file/directory operations, they change two fs/file trees
5459 * and root tree, the number of items that the qgroup reserves is
5460 * different with the free space reservation. So we can not use
5461 * the space reseravtion mechanism in start_transaction().
5463 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5464 struct btrfs_block_rsv *rsv,
5466 u64 *qgroup_reserved,
5467 bool use_global_rsv)
5471 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5473 if (root->fs_info->quota_enabled) {
5474 /* One for parent inode, two for dir entries */
5475 num_bytes = 3 * root->nodesize;
5476 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5483 *qgroup_reserved = num_bytes;
5485 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5486 rsv->space_info = __find_space_info(root->fs_info,
5487 BTRFS_BLOCK_GROUP_METADATA);
5488 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5489 BTRFS_RESERVE_FLUSH_ALL);
5491 if (ret == -ENOSPC && use_global_rsv)
5492 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5494 if (ret && *qgroup_reserved)
5495 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5500 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5501 struct btrfs_block_rsv *rsv,
5502 u64 qgroup_reserved)
5504 btrfs_block_rsv_release(root, rsv, (u64)-1);
5508 * drop_outstanding_extent - drop an outstanding extent
5509 * @inode: the inode we're dropping the extent for
5510 * @num_bytes: the number of bytes we're relaseing.
5512 * This is called when we are freeing up an outstanding extent, either called
5513 * after an error or after an extent is written. This will return the number of
5514 * reserved extents that need to be freed. This must be called with
5515 * BTRFS_I(inode)->lock held.
5517 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5519 unsigned drop_inode_space = 0;
5520 unsigned dropped_extents = 0;
5521 unsigned num_extents = 0;
5523 num_extents = (unsigned)div64_u64(num_bytes +
5524 BTRFS_MAX_EXTENT_SIZE - 1,
5525 BTRFS_MAX_EXTENT_SIZE);
5526 ASSERT(num_extents);
5527 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5528 BTRFS_I(inode)->outstanding_extents -= num_extents;
5530 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5531 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5532 &BTRFS_I(inode)->runtime_flags))
5533 drop_inode_space = 1;
5536 * If we have more or the same amount of outsanding extents than we have
5537 * reserved then we need to leave the reserved extents count alone.
5539 if (BTRFS_I(inode)->outstanding_extents >=
5540 BTRFS_I(inode)->reserved_extents)
5541 return drop_inode_space;
5543 dropped_extents = BTRFS_I(inode)->reserved_extents -
5544 BTRFS_I(inode)->outstanding_extents;
5545 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5546 return dropped_extents + drop_inode_space;
5550 * calc_csum_metadata_size - return the amount of metada space that must be
5551 * reserved/free'd for the given bytes.
5552 * @inode: the inode we're manipulating
5553 * @num_bytes: the number of bytes in question
5554 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5556 * This adjusts the number of csum_bytes in the inode and then returns the
5557 * correct amount of metadata that must either be reserved or freed. We
5558 * calculate how many checksums we can fit into one leaf and then divide the
5559 * number of bytes that will need to be checksumed by this value to figure out
5560 * how many checksums will be required. If we are adding bytes then the number
5561 * may go up and we will return the number of additional bytes that must be
5562 * reserved. If it is going down we will return the number of bytes that must
5565 * This must be called with BTRFS_I(inode)->lock held.
5567 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5570 struct btrfs_root *root = BTRFS_I(inode)->root;
5571 u64 old_csums, num_csums;
5573 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5574 BTRFS_I(inode)->csum_bytes == 0)
5577 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5579 BTRFS_I(inode)->csum_bytes += num_bytes;
5581 BTRFS_I(inode)->csum_bytes -= num_bytes;
5582 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5584 /* No change, no need to reserve more */
5585 if (old_csums == num_csums)
5589 return btrfs_calc_trans_metadata_size(root,
5590 num_csums - old_csums);
5592 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5595 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5597 struct btrfs_root *root = BTRFS_I(inode)->root;
5598 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5601 unsigned nr_extents = 0;
5602 int extra_reserve = 0;
5603 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5605 bool delalloc_lock = true;
5609 /* If we are a free space inode we need to not flush since we will be in
5610 * the middle of a transaction commit. We also don't need the delalloc
5611 * mutex since we won't race with anybody. We need this mostly to make
5612 * lockdep shut its filthy mouth.
5614 if (btrfs_is_free_space_inode(inode)) {
5615 flush = BTRFS_RESERVE_NO_FLUSH;
5616 delalloc_lock = false;
5619 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5620 btrfs_transaction_in_commit(root->fs_info))
5621 schedule_timeout(1);
5624 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5626 num_bytes = ALIGN(num_bytes, root->sectorsize);
5628 spin_lock(&BTRFS_I(inode)->lock);
5629 nr_extents = (unsigned)div64_u64(num_bytes +
5630 BTRFS_MAX_EXTENT_SIZE - 1,
5631 BTRFS_MAX_EXTENT_SIZE);
5632 BTRFS_I(inode)->outstanding_extents += nr_extents;
5635 if (BTRFS_I(inode)->outstanding_extents >
5636 BTRFS_I(inode)->reserved_extents)
5637 nr_extents = BTRFS_I(inode)->outstanding_extents -
5638 BTRFS_I(inode)->reserved_extents;
5641 * Add an item to reserve for updating the inode when we complete the
5644 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5645 &BTRFS_I(inode)->runtime_flags)) {
5650 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5651 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5652 csum_bytes = BTRFS_I(inode)->csum_bytes;
5653 spin_unlock(&BTRFS_I(inode)->lock);
5655 if (root->fs_info->quota_enabled) {
5656 ret = btrfs_qgroup_reserve_meta(root,
5657 nr_extents * root->nodesize);
5662 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5663 if (unlikely(ret)) {
5664 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5668 spin_lock(&BTRFS_I(inode)->lock);
5669 if (extra_reserve) {
5670 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5671 &BTRFS_I(inode)->runtime_flags);
5674 BTRFS_I(inode)->reserved_extents += nr_extents;
5675 spin_unlock(&BTRFS_I(inode)->lock);
5678 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5681 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5682 btrfs_ino(inode), to_reserve, 1);
5683 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5688 spin_lock(&BTRFS_I(inode)->lock);
5689 dropped = drop_outstanding_extent(inode, num_bytes);
5691 * If the inodes csum_bytes is the same as the original
5692 * csum_bytes then we know we haven't raced with any free()ers
5693 * so we can just reduce our inodes csum bytes and carry on.
5695 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5696 calc_csum_metadata_size(inode, num_bytes, 0);
5698 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5702 * This is tricky, but first we need to figure out how much we
5703 * free'd from any free-ers that occured during this
5704 * reservation, so we reset ->csum_bytes to the csum_bytes
5705 * before we dropped our lock, and then call the free for the
5706 * number of bytes that were freed while we were trying our
5709 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5710 BTRFS_I(inode)->csum_bytes = csum_bytes;
5711 to_free = calc_csum_metadata_size(inode, bytes, 0);
5715 * Now we need to see how much we would have freed had we not
5716 * been making this reservation and our ->csum_bytes were not
5717 * artificially inflated.
5719 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5720 bytes = csum_bytes - orig_csum_bytes;
5721 bytes = calc_csum_metadata_size(inode, bytes, 0);
5724 * Now reset ->csum_bytes to what it should be. If bytes is
5725 * more than to_free then we would have free'd more space had we
5726 * not had an artificially high ->csum_bytes, so we need to free
5727 * the remainder. If bytes is the same or less then we don't
5728 * need to do anything, the other free-ers did the correct
5731 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5732 if (bytes > to_free)
5733 to_free = bytes - to_free;
5737 spin_unlock(&BTRFS_I(inode)->lock);
5739 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5742 btrfs_block_rsv_release(root, block_rsv, to_free);
5743 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5744 btrfs_ino(inode), to_free, 0);
5747 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5752 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5753 * @inode: the inode to release the reservation for
5754 * @num_bytes: the number of bytes we're releasing
5756 * This will release the metadata reservation for an inode. This can be called
5757 * once we complete IO for a given set of bytes to release their metadata
5760 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5762 struct btrfs_root *root = BTRFS_I(inode)->root;
5766 num_bytes = ALIGN(num_bytes, root->sectorsize);
5767 spin_lock(&BTRFS_I(inode)->lock);
5768 dropped = drop_outstanding_extent(inode, num_bytes);
5771 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5772 spin_unlock(&BTRFS_I(inode)->lock);
5774 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5776 if (btrfs_test_is_dummy_root(root))
5779 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5780 btrfs_ino(inode), to_free, 0);
5782 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5787 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5789 * @inode: inode we're writing to
5790 * @start: start range we are writing to
5791 * @len: how long the range we are writing to
5793 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5795 * This will do the following things
5797 * o reserve space in data space info for num bytes
5798 * and reserve precious corresponding qgroup space
5799 * (Done in check_data_free_space)
5801 * o reserve space for metadata space, based on the number of outstanding
5802 * extents and how much csums will be needed
5803 * also reserve metadata space in a per root over-reserve method.
5804 * o add to the inodes->delalloc_bytes
5805 * o add it to the fs_info's delalloc inodes list.
5806 * (Above 3 all done in delalloc_reserve_metadata)
5808 * Return 0 for success
5809 * Return <0 for error(-ENOSPC or -EQUOT)
5811 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5815 ret = btrfs_check_data_free_space(inode, start, len);
5818 ret = btrfs_delalloc_reserve_metadata(inode, len);
5820 btrfs_free_reserved_data_space(inode, start, len);
5825 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5826 * @inode: inode we're releasing space for
5827 * @start: start position of the space already reserved
5828 * @len: the len of the space already reserved
5830 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5831 * called in the case that we don't need the metadata AND data reservations
5832 * anymore. So if there is an error or we insert an inline extent.
5834 * This function will release the metadata space that was not used and will
5835 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5836 * list if there are no delalloc bytes left.
5837 * Also it will handle the qgroup reserved space.
5839 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5841 btrfs_delalloc_release_metadata(inode, len);
5842 btrfs_free_reserved_data_space(inode, start, len);
5845 static int update_block_group(struct btrfs_trans_handle *trans,
5846 struct btrfs_root *root, u64 bytenr,
5847 u64 num_bytes, int alloc)
5849 struct btrfs_block_group_cache *cache = NULL;
5850 struct btrfs_fs_info *info = root->fs_info;
5851 u64 total = num_bytes;
5856 /* block accounting for super block */
5857 spin_lock(&info->delalloc_root_lock);
5858 old_val = btrfs_super_bytes_used(info->super_copy);
5860 old_val += num_bytes;
5862 old_val -= num_bytes;
5863 btrfs_set_super_bytes_used(info->super_copy, old_val);
5864 spin_unlock(&info->delalloc_root_lock);
5867 cache = btrfs_lookup_block_group(info, bytenr);
5870 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5871 BTRFS_BLOCK_GROUP_RAID1 |
5872 BTRFS_BLOCK_GROUP_RAID10))
5877 * If this block group has free space cache written out, we
5878 * need to make sure to load it if we are removing space. This
5879 * is because we need the unpinning stage to actually add the
5880 * space back to the block group, otherwise we will leak space.
5882 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5883 cache_block_group(cache, 1);
5885 byte_in_group = bytenr - cache->key.objectid;
5886 WARN_ON(byte_in_group > cache->key.offset);
5888 spin_lock(&cache->space_info->lock);
5889 spin_lock(&cache->lock);
5891 if (btrfs_test_opt(root, SPACE_CACHE) &&
5892 cache->disk_cache_state < BTRFS_DC_CLEAR)
5893 cache->disk_cache_state = BTRFS_DC_CLEAR;
5895 old_val = btrfs_block_group_used(&cache->item);
5896 num_bytes = min(total, cache->key.offset - byte_in_group);
5898 old_val += num_bytes;
5899 btrfs_set_block_group_used(&cache->item, old_val);
5900 cache->reserved -= num_bytes;
5901 cache->space_info->bytes_reserved -= num_bytes;
5902 cache->space_info->bytes_used += num_bytes;
5903 cache->space_info->disk_used += num_bytes * factor;
5904 spin_unlock(&cache->lock);
5905 spin_unlock(&cache->space_info->lock);
5907 old_val -= num_bytes;
5908 btrfs_set_block_group_used(&cache->item, old_val);
5909 cache->pinned += num_bytes;
5910 cache->space_info->bytes_pinned += num_bytes;
5911 cache->space_info->bytes_used -= num_bytes;
5912 cache->space_info->disk_used -= num_bytes * factor;
5913 spin_unlock(&cache->lock);
5914 spin_unlock(&cache->space_info->lock);
5916 set_extent_dirty(info->pinned_extents,
5917 bytenr, bytenr + num_bytes - 1,
5918 GFP_NOFS | __GFP_NOFAIL);
5921 spin_lock(&trans->transaction->dirty_bgs_lock);
5922 if (list_empty(&cache->dirty_list)) {
5923 list_add_tail(&cache->dirty_list,
5924 &trans->transaction->dirty_bgs);
5925 trans->transaction->num_dirty_bgs++;
5926 btrfs_get_block_group(cache);
5928 spin_unlock(&trans->transaction->dirty_bgs_lock);
5931 * No longer have used bytes in this block group, queue it for
5932 * deletion. We do this after adding the block group to the
5933 * dirty list to avoid races between cleaner kthread and space
5936 if (!alloc && old_val == 0) {
5937 spin_lock(&info->unused_bgs_lock);
5938 if (list_empty(&cache->bg_list)) {
5939 btrfs_get_block_group(cache);
5940 list_add_tail(&cache->bg_list,
5943 spin_unlock(&info->unused_bgs_lock);
5946 btrfs_put_block_group(cache);
5948 bytenr += num_bytes;
5953 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5955 struct btrfs_block_group_cache *cache;
5958 spin_lock(&root->fs_info->block_group_cache_lock);
5959 bytenr = root->fs_info->first_logical_byte;
5960 spin_unlock(&root->fs_info->block_group_cache_lock);
5962 if (bytenr < (u64)-1)
5965 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5969 bytenr = cache->key.objectid;
5970 btrfs_put_block_group(cache);
5975 static int pin_down_extent(struct btrfs_root *root,
5976 struct btrfs_block_group_cache *cache,
5977 u64 bytenr, u64 num_bytes, int reserved)
5979 spin_lock(&cache->space_info->lock);
5980 spin_lock(&cache->lock);
5981 cache->pinned += num_bytes;
5982 cache->space_info->bytes_pinned += num_bytes;
5984 cache->reserved -= num_bytes;
5985 cache->space_info->bytes_reserved -= num_bytes;
5987 spin_unlock(&cache->lock);
5988 spin_unlock(&cache->space_info->lock);
5990 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5991 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5993 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5998 * this function must be called within transaction
6000 int btrfs_pin_extent(struct btrfs_root *root,
6001 u64 bytenr, u64 num_bytes, int reserved)
6003 struct btrfs_block_group_cache *cache;
6005 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6006 BUG_ON(!cache); /* Logic error */
6008 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6010 btrfs_put_block_group(cache);
6015 * this function must be called within transaction
6017 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6018 u64 bytenr, u64 num_bytes)
6020 struct btrfs_block_group_cache *cache;
6023 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6028 * pull in the free space cache (if any) so that our pin
6029 * removes the free space from the cache. We have load_only set
6030 * to one because the slow code to read in the free extents does check
6031 * the pinned extents.
6033 cache_block_group(cache, 1);
6035 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6037 /* remove us from the free space cache (if we're there at all) */
6038 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6039 btrfs_put_block_group(cache);
6043 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6046 struct btrfs_block_group_cache *block_group;
6047 struct btrfs_caching_control *caching_ctl;
6049 block_group = btrfs_lookup_block_group(root->fs_info, start);
6053 cache_block_group(block_group, 0);
6054 caching_ctl = get_caching_control(block_group);
6058 BUG_ON(!block_group_cache_done(block_group));
6059 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6061 mutex_lock(&caching_ctl->mutex);
6063 if (start >= caching_ctl->progress) {
6064 ret = add_excluded_extent(root, start, num_bytes);
6065 } else if (start + num_bytes <= caching_ctl->progress) {
6066 ret = btrfs_remove_free_space(block_group,
6069 num_bytes = caching_ctl->progress - start;
6070 ret = btrfs_remove_free_space(block_group,
6075 num_bytes = (start + num_bytes) -
6076 caching_ctl->progress;
6077 start = caching_ctl->progress;
6078 ret = add_excluded_extent(root, start, num_bytes);
6081 mutex_unlock(&caching_ctl->mutex);
6082 put_caching_control(caching_ctl);
6084 btrfs_put_block_group(block_group);
6088 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6089 struct extent_buffer *eb)
6091 struct btrfs_file_extent_item *item;
6092 struct btrfs_key key;
6096 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6099 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6100 btrfs_item_key_to_cpu(eb, &key, i);
6101 if (key.type != BTRFS_EXTENT_DATA_KEY)
6103 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6104 found_type = btrfs_file_extent_type(eb, item);
6105 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6107 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6109 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6110 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6111 __exclude_logged_extent(log, key.objectid, key.offset);
6118 * btrfs_update_reserved_bytes - update the block_group and space info counters
6119 * @cache: The cache we are manipulating
6120 * @num_bytes: The number of bytes in question
6121 * @reserve: One of the reservation enums
6122 * @delalloc: The blocks are allocated for the delalloc write
6124 * This is called by the allocator when it reserves space, or by somebody who is
6125 * freeing space that was never actually used on disk. For example if you
6126 * reserve some space for a new leaf in transaction A and before transaction A
6127 * commits you free that leaf, you call this with reserve set to 0 in order to
6128 * clear the reservation.
6130 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6131 * ENOSPC accounting. For data we handle the reservation through clearing the
6132 * delalloc bits in the io_tree. We have to do this since we could end up
6133 * allocating less disk space for the amount of data we have reserved in the
6134 * case of compression.
6136 * If this is a reservation and the block group has become read only we cannot
6137 * make the reservation and return -EAGAIN, otherwise this function always
6140 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6141 u64 num_bytes, int reserve, int delalloc)
6143 struct btrfs_space_info *space_info = cache->space_info;
6146 spin_lock(&space_info->lock);
6147 spin_lock(&cache->lock);
6148 if (reserve != RESERVE_FREE) {
6152 cache->reserved += num_bytes;
6153 space_info->bytes_reserved += num_bytes;
6154 if (reserve == RESERVE_ALLOC) {
6155 trace_btrfs_space_reservation(cache->fs_info,
6156 "space_info", space_info->flags,
6158 space_info->bytes_may_use -= num_bytes;
6162 cache->delalloc_bytes += num_bytes;
6166 space_info->bytes_readonly += num_bytes;
6167 cache->reserved -= num_bytes;
6168 space_info->bytes_reserved -= num_bytes;
6171 cache->delalloc_bytes -= num_bytes;
6173 spin_unlock(&cache->lock);
6174 spin_unlock(&space_info->lock);
6178 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6179 struct btrfs_root *root)
6181 struct btrfs_fs_info *fs_info = root->fs_info;
6182 struct btrfs_caching_control *next;
6183 struct btrfs_caching_control *caching_ctl;
6184 struct btrfs_block_group_cache *cache;
6186 down_write(&fs_info->commit_root_sem);
6188 list_for_each_entry_safe(caching_ctl, next,
6189 &fs_info->caching_block_groups, list) {
6190 cache = caching_ctl->block_group;
6191 if (block_group_cache_done(cache)) {
6192 cache->last_byte_to_unpin = (u64)-1;
6193 list_del_init(&caching_ctl->list);
6194 put_caching_control(caching_ctl);
6196 cache->last_byte_to_unpin = caching_ctl->progress;
6200 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6201 fs_info->pinned_extents = &fs_info->freed_extents[1];
6203 fs_info->pinned_extents = &fs_info->freed_extents[0];
6205 up_write(&fs_info->commit_root_sem);
6207 update_global_block_rsv(fs_info);
6211 * Returns the free cluster for the given space info and sets empty_cluster to
6212 * what it should be based on the mount options.
6214 static struct btrfs_free_cluster *
6215 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6218 struct btrfs_free_cluster *ret = NULL;
6219 bool ssd = btrfs_test_opt(root, SSD);
6222 if (btrfs_mixed_space_info(space_info))
6226 *empty_cluster = 2 * 1024 * 1024;
6227 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6228 ret = &root->fs_info->meta_alloc_cluster;
6230 *empty_cluster = 64 * 1024;
6231 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6232 ret = &root->fs_info->data_alloc_cluster;
6238 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6239 const bool return_free_space)
6241 struct btrfs_fs_info *fs_info = root->fs_info;
6242 struct btrfs_block_group_cache *cache = NULL;
6243 struct btrfs_space_info *space_info;
6244 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6245 struct btrfs_free_cluster *cluster = NULL;
6247 u64 total_unpinned = 0;
6248 u64 empty_cluster = 0;
6251 while (start <= end) {
6254 start >= cache->key.objectid + cache->key.offset) {
6256 btrfs_put_block_group(cache);
6258 cache = btrfs_lookup_block_group(fs_info, start);
6259 BUG_ON(!cache); /* Logic error */
6261 cluster = fetch_cluster_info(root,
6264 empty_cluster <<= 1;
6267 len = cache->key.objectid + cache->key.offset - start;
6268 len = min(len, end + 1 - start);
6270 if (start < cache->last_byte_to_unpin) {
6271 len = min(len, cache->last_byte_to_unpin - start);
6272 if (return_free_space)
6273 btrfs_add_free_space(cache, start, len);
6277 total_unpinned += len;
6278 space_info = cache->space_info;
6281 * If this space cluster has been marked as fragmented and we've
6282 * unpinned enough in this block group to potentially allow a
6283 * cluster to be created inside of it go ahead and clear the
6286 if (cluster && cluster->fragmented &&
6287 total_unpinned > empty_cluster) {
6288 spin_lock(&cluster->lock);
6289 cluster->fragmented = 0;
6290 spin_unlock(&cluster->lock);
6293 spin_lock(&space_info->lock);
6294 spin_lock(&cache->lock);
6295 cache->pinned -= len;
6296 space_info->bytes_pinned -= len;
6297 space_info->max_extent_size = 0;
6298 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6300 space_info->bytes_readonly += len;
6303 spin_unlock(&cache->lock);
6304 if (!readonly && global_rsv->space_info == space_info) {
6305 spin_lock(&global_rsv->lock);
6306 if (!global_rsv->full) {
6307 len = min(len, global_rsv->size -
6308 global_rsv->reserved);
6309 global_rsv->reserved += len;
6310 space_info->bytes_may_use += len;
6311 if (global_rsv->reserved >= global_rsv->size)
6312 global_rsv->full = 1;
6314 spin_unlock(&global_rsv->lock);
6316 spin_unlock(&space_info->lock);
6320 btrfs_put_block_group(cache);
6324 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6325 struct btrfs_root *root)
6327 struct btrfs_fs_info *fs_info = root->fs_info;
6328 struct btrfs_block_group_cache *block_group, *tmp;
6329 struct list_head *deleted_bgs;
6330 struct extent_io_tree *unpin;
6335 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6336 unpin = &fs_info->freed_extents[1];
6338 unpin = &fs_info->freed_extents[0];
6340 while (!trans->aborted) {
6341 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6342 ret = find_first_extent_bit(unpin, 0, &start, &end,
6343 EXTENT_DIRTY, NULL);
6345 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6349 if (btrfs_test_opt(root, DISCARD))
6350 ret = btrfs_discard_extent(root, start,
6351 end + 1 - start, NULL);
6353 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6354 unpin_extent_range(root, start, end, true);
6355 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6360 * Transaction is finished. We don't need the lock anymore. We
6361 * do need to clean up the block groups in case of a transaction
6364 deleted_bgs = &trans->transaction->deleted_bgs;
6365 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6369 if (!trans->aborted)
6370 ret = btrfs_discard_extent(root,
6371 block_group->key.objectid,
6372 block_group->key.offset,
6375 list_del_init(&block_group->bg_list);
6376 btrfs_put_block_group_trimming(block_group);
6377 btrfs_put_block_group(block_group);
6380 const char *errstr = btrfs_decode_error(ret);
6382 "Discard failed while removing blockgroup: errno=%d %s\n",
6390 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6391 u64 owner, u64 root_objectid)
6393 struct btrfs_space_info *space_info;
6396 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6397 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6398 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6400 flags = BTRFS_BLOCK_GROUP_METADATA;
6402 flags = BTRFS_BLOCK_GROUP_DATA;
6405 space_info = __find_space_info(fs_info, flags);
6406 BUG_ON(!space_info); /* Logic bug */
6407 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6411 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6412 struct btrfs_root *root,
6413 struct btrfs_delayed_ref_node *node, u64 parent,
6414 u64 root_objectid, u64 owner_objectid,
6415 u64 owner_offset, int refs_to_drop,
6416 struct btrfs_delayed_extent_op *extent_op)
6418 struct btrfs_key key;
6419 struct btrfs_path *path;
6420 struct btrfs_fs_info *info = root->fs_info;
6421 struct btrfs_root *extent_root = info->extent_root;
6422 struct extent_buffer *leaf;
6423 struct btrfs_extent_item *ei;
6424 struct btrfs_extent_inline_ref *iref;
6427 int extent_slot = 0;
6428 int found_extent = 0;
6432 u64 bytenr = node->bytenr;
6433 u64 num_bytes = node->num_bytes;
6435 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6438 path = btrfs_alloc_path();
6443 path->leave_spinning = 1;
6445 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6446 BUG_ON(!is_data && refs_to_drop != 1);
6449 skinny_metadata = 0;
6451 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6452 bytenr, num_bytes, parent,
6453 root_objectid, owner_objectid,
6456 extent_slot = path->slots[0];
6457 while (extent_slot >= 0) {
6458 btrfs_item_key_to_cpu(path->nodes[0], &key,
6460 if (key.objectid != bytenr)
6462 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6463 key.offset == num_bytes) {
6467 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6468 key.offset == owner_objectid) {
6472 if (path->slots[0] - extent_slot > 5)
6476 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6477 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6478 if (found_extent && item_size < sizeof(*ei))
6481 if (!found_extent) {
6483 ret = remove_extent_backref(trans, extent_root, path,
6485 is_data, &last_ref);
6487 btrfs_abort_transaction(trans, extent_root, ret);
6490 btrfs_release_path(path);
6491 path->leave_spinning = 1;
6493 key.objectid = bytenr;
6494 key.type = BTRFS_EXTENT_ITEM_KEY;
6495 key.offset = num_bytes;
6497 if (!is_data && skinny_metadata) {
6498 key.type = BTRFS_METADATA_ITEM_KEY;
6499 key.offset = owner_objectid;
6502 ret = btrfs_search_slot(trans, extent_root,
6504 if (ret > 0 && skinny_metadata && path->slots[0]) {
6506 * Couldn't find our skinny metadata item,
6507 * see if we have ye olde extent item.
6510 btrfs_item_key_to_cpu(path->nodes[0], &key,
6512 if (key.objectid == bytenr &&
6513 key.type == BTRFS_EXTENT_ITEM_KEY &&
6514 key.offset == num_bytes)
6518 if (ret > 0 && skinny_metadata) {
6519 skinny_metadata = false;
6520 key.objectid = bytenr;
6521 key.type = BTRFS_EXTENT_ITEM_KEY;
6522 key.offset = num_bytes;
6523 btrfs_release_path(path);
6524 ret = btrfs_search_slot(trans, extent_root,
6529 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6532 btrfs_print_leaf(extent_root,
6536 btrfs_abort_transaction(trans, extent_root, ret);
6539 extent_slot = path->slots[0];
6541 } else if (WARN_ON(ret == -ENOENT)) {
6542 btrfs_print_leaf(extent_root, path->nodes[0]);
6544 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6545 bytenr, parent, root_objectid, owner_objectid,
6547 btrfs_abort_transaction(trans, extent_root, ret);
6550 btrfs_abort_transaction(trans, extent_root, ret);
6554 leaf = path->nodes[0];
6555 item_size = btrfs_item_size_nr(leaf, extent_slot);
6556 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6557 if (item_size < sizeof(*ei)) {
6558 BUG_ON(found_extent || extent_slot != path->slots[0]);
6559 ret = convert_extent_item_v0(trans, extent_root, path,
6562 btrfs_abort_transaction(trans, extent_root, ret);
6566 btrfs_release_path(path);
6567 path->leave_spinning = 1;
6569 key.objectid = bytenr;
6570 key.type = BTRFS_EXTENT_ITEM_KEY;
6571 key.offset = num_bytes;
6573 ret = btrfs_search_slot(trans, extent_root, &key, path,
6576 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6578 btrfs_print_leaf(extent_root, path->nodes[0]);
6581 btrfs_abort_transaction(trans, extent_root, ret);
6585 extent_slot = path->slots[0];
6586 leaf = path->nodes[0];
6587 item_size = btrfs_item_size_nr(leaf, extent_slot);
6590 BUG_ON(item_size < sizeof(*ei));
6591 ei = btrfs_item_ptr(leaf, extent_slot,
6592 struct btrfs_extent_item);
6593 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6594 key.type == BTRFS_EXTENT_ITEM_KEY) {
6595 struct btrfs_tree_block_info *bi;
6596 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6597 bi = (struct btrfs_tree_block_info *)(ei + 1);
6598 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6601 refs = btrfs_extent_refs(leaf, ei);
6602 if (refs < refs_to_drop) {
6603 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6604 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6606 btrfs_abort_transaction(trans, extent_root, ret);
6609 refs -= refs_to_drop;
6613 __run_delayed_extent_op(extent_op, leaf, ei);
6615 * In the case of inline back ref, reference count will
6616 * be updated by remove_extent_backref
6619 BUG_ON(!found_extent);
6621 btrfs_set_extent_refs(leaf, ei, refs);
6622 btrfs_mark_buffer_dirty(leaf);
6625 ret = remove_extent_backref(trans, extent_root, path,
6627 is_data, &last_ref);
6629 btrfs_abort_transaction(trans, extent_root, ret);
6633 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6637 BUG_ON(is_data && refs_to_drop !=
6638 extent_data_ref_count(path, iref));
6640 BUG_ON(path->slots[0] != extent_slot);
6642 BUG_ON(path->slots[0] != extent_slot + 1);
6643 path->slots[0] = extent_slot;
6649 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6652 btrfs_abort_transaction(trans, extent_root, ret);
6655 btrfs_release_path(path);
6658 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6660 btrfs_abort_transaction(trans, extent_root, ret);
6665 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6667 btrfs_abort_transaction(trans, extent_root, ret);
6671 btrfs_release_path(path);
6674 btrfs_free_path(path);
6679 * when we free an block, it is possible (and likely) that we free the last
6680 * delayed ref for that extent as well. This searches the delayed ref tree for
6681 * a given extent, and if there are no other delayed refs to be processed, it
6682 * removes it from the tree.
6684 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6685 struct btrfs_root *root, u64 bytenr)
6687 struct btrfs_delayed_ref_head *head;
6688 struct btrfs_delayed_ref_root *delayed_refs;
6691 delayed_refs = &trans->transaction->delayed_refs;
6692 spin_lock(&delayed_refs->lock);
6693 head = btrfs_find_delayed_ref_head(trans, bytenr);
6695 goto out_delayed_unlock;
6697 spin_lock(&head->lock);
6698 if (!list_empty(&head->ref_list))
6701 if (head->extent_op) {
6702 if (!head->must_insert_reserved)
6704 btrfs_free_delayed_extent_op(head->extent_op);
6705 head->extent_op = NULL;
6709 * waiting for the lock here would deadlock. If someone else has it
6710 * locked they are already in the process of dropping it anyway
6712 if (!mutex_trylock(&head->mutex))
6716 * at this point we have a head with no other entries. Go
6717 * ahead and process it.
6719 head->node.in_tree = 0;
6720 rb_erase(&head->href_node, &delayed_refs->href_root);
6722 atomic_dec(&delayed_refs->num_entries);
6725 * we don't take a ref on the node because we're removing it from the
6726 * tree, so we just steal the ref the tree was holding.
6728 delayed_refs->num_heads--;
6729 if (head->processing == 0)
6730 delayed_refs->num_heads_ready--;
6731 head->processing = 0;
6732 spin_unlock(&head->lock);
6733 spin_unlock(&delayed_refs->lock);
6735 BUG_ON(head->extent_op);
6736 if (head->must_insert_reserved)
6739 mutex_unlock(&head->mutex);
6740 btrfs_put_delayed_ref(&head->node);
6743 spin_unlock(&head->lock);
6746 spin_unlock(&delayed_refs->lock);
6750 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6751 struct btrfs_root *root,
6752 struct extent_buffer *buf,
6753 u64 parent, int last_ref)
6758 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6759 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6760 buf->start, buf->len,
6761 parent, root->root_key.objectid,
6762 btrfs_header_level(buf),
6763 BTRFS_DROP_DELAYED_REF, NULL);
6764 BUG_ON(ret); /* -ENOMEM */
6770 if (btrfs_header_generation(buf) == trans->transid) {
6771 struct btrfs_block_group_cache *cache;
6773 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6774 ret = check_ref_cleanup(trans, root, buf->start);
6779 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6781 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6782 pin_down_extent(root, cache, buf->start, buf->len, 1);
6783 btrfs_put_block_group(cache);
6787 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6789 btrfs_add_free_space(cache, buf->start, buf->len);
6790 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6791 btrfs_put_block_group(cache);
6792 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6797 add_pinned_bytes(root->fs_info, buf->len,
6798 btrfs_header_level(buf),
6799 root->root_key.objectid);
6802 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6805 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6808 /* Can return -ENOMEM */
6809 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6810 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6811 u64 owner, u64 offset)
6814 struct btrfs_fs_info *fs_info = root->fs_info;
6816 if (btrfs_test_is_dummy_root(root))
6819 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6822 * tree log blocks never actually go into the extent allocation
6823 * tree, just update pinning info and exit early.
6825 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6826 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6827 /* unlocks the pinned mutex */
6828 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6830 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6831 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6833 parent, root_objectid, (int)owner,
6834 BTRFS_DROP_DELAYED_REF, NULL);
6836 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6838 parent, root_objectid, owner,
6840 BTRFS_DROP_DELAYED_REF, NULL);
6846 * when we wait for progress in the block group caching, its because
6847 * our allocation attempt failed at least once. So, we must sleep
6848 * and let some progress happen before we try again.
6850 * This function will sleep at least once waiting for new free space to
6851 * show up, and then it will check the block group free space numbers
6852 * for our min num_bytes. Another option is to have it go ahead
6853 * and look in the rbtree for a free extent of a given size, but this
6856 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6857 * any of the information in this block group.
6859 static noinline void
6860 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6863 struct btrfs_caching_control *caching_ctl;
6865 caching_ctl = get_caching_control(cache);
6869 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6870 (cache->free_space_ctl->free_space >= num_bytes));
6872 put_caching_control(caching_ctl);
6876 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6878 struct btrfs_caching_control *caching_ctl;
6881 caching_ctl = get_caching_control(cache);
6883 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6885 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6886 if (cache->cached == BTRFS_CACHE_ERROR)
6888 put_caching_control(caching_ctl);
6892 int __get_raid_index(u64 flags)
6894 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6895 return BTRFS_RAID_RAID10;
6896 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6897 return BTRFS_RAID_RAID1;
6898 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6899 return BTRFS_RAID_DUP;
6900 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6901 return BTRFS_RAID_RAID0;
6902 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6903 return BTRFS_RAID_RAID5;
6904 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6905 return BTRFS_RAID_RAID6;
6907 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6910 int get_block_group_index(struct btrfs_block_group_cache *cache)
6912 return __get_raid_index(cache->flags);
6915 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6916 [BTRFS_RAID_RAID10] = "raid10",
6917 [BTRFS_RAID_RAID1] = "raid1",
6918 [BTRFS_RAID_DUP] = "dup",
6919 [BTRFS_RAID_RAID0] = "raid0",
6920 [BTRFS_RAID_SINGLE] = "single",
6921 [BTRFS_RAID_RAID5] = "raid5",
6922 [BTRFS_RAID_RAID6] = "raid6",
6925 static const char *get_raid_name(enum btrfs_raid_types type)
6927 if (type >= BTRFS_NR_RAID_TYPES)
6930 return btrfs_raid_type_names[type];
6933 enum btrfs_loop_type {
6934 LOOP_CACHING_NOWAIT = 0,
6935 LOOP_CACHING_WAIT = 1,
6936 LOOP_ALLOC_CHUNK = 2,
6937 LOOP_NO_EMPTY_SIZE = 3,
6941 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6945 down_read(&cache->data_rwsem);
6949 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6952 btrfs_get_block_group(cache);
6954 down_read(&cache->data_rwsem);
6957 static struct btrfs_block_group_cache *
6958 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6959 struct btrfs_free_cluster *cluster,
6962 struct btrfs_block_group_cache *used_bg;
6963 bool locked = false;
6965 spin_lock(&cluster->refill_lock);
6967 if (used_bg == cluster->block_group)
6970 up_read(&used_bg->data_rwsem);
6971 btrfs_put_block_group(used_bg);
6974 used_bg = cluster->block_group;
6978 if (used_bg == block_group)
6981 btrfs_get_block_group(used_bg);
6986 if (down_read_trylock(&used_bg->data_rwsem))
6989 spin_unlock(&cluster->refill_lock);
6990 down_read(&used_bg->data_rwsem);
6996 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7000 up_read(&cache->data_rwsem);
7001 btrfs_put_block_group(cache);
7005 * walks the btree of allocated extents and find a hole of a given size.
7006 * The key ins is changed to record the hole:
7007 * ins->objectid == start position
7008 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7009 * ins->offset == the size of the hole.
7010 * Any available blocks before search_start are skipped.
7012 * If there is no suitable free space, we will record the max size of
7013 * the free space extent currently.
7015 static noinline int find_free_extent(struct btrfs_root *orig_root,
7016 u64 num_bytes, u64 empty_size,
7017 u64 hint_byte, struct btrfs_key *ins,
7018 u64 flags, int delalloc)
7021 struct btrfs_root *root = orig_root->fs_info->extent_root;
7022 struct btrfs_free_cluster *last_ptr = NULL;
7023 struct btrfs_block_group_cache *block_group = NULL;
7024 u64 search_start = 0;
7025 u64 max_extent_size = 0;
7026 u64 empty_cluster = 0;
7027 struct btrfs_space_info *space_info;
7029 int index = __get_raid_index(flags);
7030 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7031 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7032 bool failed_cluster_refill = false;
7033 bool failed_alloc = false;
7034 bool use_cluster = true;
7035 bool have_caching_bg = false;
7036 bool orig_have_caching_bg = false;
7037 bool full_search = false;
7039 WARN_ON(num_bytes < root->sectorsize);
7040 ins->type = BTRFS_EXTENT_ITEM_KEY;
7044 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7046 space_info = __find_space_info(root->fs_info, flags);
7048 btrfs_err(root->fs_info, "No space info for %llu", flags);
7053 * If our free space is heavily fragmented we may not be able to make
7054 * big contiguous allocations, so instead of doing the expensive search
7055 * for free space, simply return ENOSPC with our max_extent_size so we
7056 * can go ahead and search for a more manageable chunk.
7058 * If our max_extent_size is large enough for our allocation simply
7059 * disable clustering since we will likely not be able to find enough
7060 * space to create a cluster and induce latency trying.
7062 if (unlikely(space_info->max_extent_size)) {
7063 spin_lock(&space_info->lock);
7064 if (space_info->max_extent_size &&
7065 num_bytes > space_info->max_extent_size) {
7066 ins->offset = space_info->max_extent_size;
7067 spin_unlock(&space_info->lock);
7069 } else if (space_info->max_extent_size) {
7070 use_cluster = false;
7072 spin_unlock(&space_info->lock);
7075 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7077 spin_lock(&last_ptr->lock);
7078 if (last_ptr->block_group)
7079 hint_byte = last_ptr->window_start;
7080 if (last_ptr->fragmented) {
7082 * We still set window_start so we can keep track of the
7083 * last place we found an allocation to try and save
7086 hint_byte = last_ptr->window_start;
7087 use_cluster = false;
7089 spin_unlock(&last_ptr->lock);
7092 search_start = max(search_start, first_logical_byte(root, 0));
7093 search_start = max(search_start, hint_byte);
7094 if (search_start == hint_byte) {
7095 block_group = btrfs_lookup_block_group(root->fs_info,
7098 * we don't want to use the block group if it doesn't match our
7099 * allocation bits, or if its not cached.
7101 * However if we are re-searching with an ideal block group
7102 * picked out then we don't care that the block group is cached.
7104 if (block_group && block_group_bits(block_group, flags) &&
7105 block_group->cached != BTRFS_CACHE_NO) {
7106 down_read(&space_info->groups_sem);
7107 if (list_empty(&block_group->list) ||
7110 * someone is removing this block group,
7111 * we can't jump into the have_block_group
7112 * target because our list pointers are not
7115 btrfs_put_block_group(block_group);
7116 up_read(&space_info->groups_sem);
7118 index = get_block_group_index(block_group);
7119 btrfs_lock_block_group(block_group, delalloc);
7120 goto have_block_group;
7122 } else if (block_group) {
7123 btrfs_put_block_group(block_group);
7127 have_caching_bg = false;
7128 if (index == 0 || index == __get_raid_index(flags))
7130 down_read(&space_info->groups_sem);
7131 list_for_each_entry(block_group, &space_info->block_groups[index],
7136 btrfs_grab_block_group(block_group, delalloc);
7137 search_start = block_group->key.objectid;
7140 * this can happen if we end up cycling through all the
7141 * raid types, but we want to make sure we only allocate
7142 * for the proper type.
7144 if (!block_group_bits(block_group, flags)) {
7145 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7146 BTRFS_BLOCK_GROUP_RAID1 |
7147 BTRFS_BLOCK_GROUP_RAID5 |
7148 BTRFS_BLOCK_GROUP_RAID6 |
7149 BTRFS_BLOCK_GROUP_RAID10;
7152 * if they asked for extra copies and this block group
7153 * doesn't provide them, bail. This does allow us to
7154 * fill raid0 from raid1.
7156 if ((flags & extra) && !(block_group->flags & extra))
7161 cached = block_group_cache_done(block_group);
7162 if (unlikely(!cached)) {
7163 have_caching_bg = true;
7164 ret = cache_block_group(block_group, 0);
7169 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7171 if (unlikely(block_group->ro))
7175 * Ok we want to try and use the cluster allocator, so
7178 if (last_ptr && use_cluster) {
7179 struct btrfs_block_group_cache *used_block_group;
7180 unsigned long aligned_cluster;
7182 * the refill lock keeps out other
7183 * people trying to start a new cluster
7185 used_block_group = btrfs_lock_cluster(block_group,
7188 if (!used_block_group)
7189 goto refill_cluster;
7191 if (used_block_group != block_group &&
7192 (used_block_group->ro ||
7193 !block_group_bits(used_block_group, flags)))
7194 goto release_cluster;
7196 offset = btrfs_alloc_from_cluster(used_block_group,
7199 used_block_group->key.objectid,
7202 /* we have a block, we're done */
7203 spin_unlock(&last_ptr->refill_lock);
7204 trace_btrfs_reserve_extent_cluster(root,
7206 search_start, num_bytes);
7207 if (used_block_group != block_group) {
7208 btrfs_release_block_group(block_group,
7210 block_group = used_block_group;
7215 WARN_ON(last_ptr->block_group != used_block_group);
7217 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7218 * set up a new clusters, so lets just skip it
7219 * and let the allocator find whatever block
7220 * it can find. If we reach this point, we
7221 * will have tried the cluster allocator
7222 * plenty of times and not have found
7223 * anything, so we are likely way too
7224 * fragmented for the clustering stuff to find
7227 * However, if the cluster is taken from the
7228 * current block group, release the cluster
7229 * first, so that we stand a better chance of
7230 * succeeding in the unclustered
7232 if (loop >= LOOP_NO_EMPTY_SIZE &&
7233 used_block_group != block_group) {
7234 spin_unlock(&last_ptr->refill_lock);
7235 btrfs_release_block_group(used_block_group,
7237 goto unclustered_alloc;
7241 * this cluster didn't work out, free it and
7244 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7246 if (used_block_group != block_group)
7247 btrfs_release_block_group(used_block_group,
7250 if (loop >= LOOP_NO_EMPTY_SIZE) {
7251 spin_unlock(&last_ptr->refill_lock);
7252 goto unclustered_alloc;
7255 aligned_cluster = max_t(unsigned long,
7256 empty_cluster + empty_size,
7257 block_group->full_stripe_len);
7259 /* allocate a cluster in this block group */
7260 ret = btrfs_find_space_cluster(root, block_group,
7261 last_ptr, search_start,
7266 * now pull our allocation out of this
7269 offset = btrfs_alloc_from_cluster(block_group,
7275 /* we found one, proceed */
7276 spin_unlock(&last_ptr->refill_lock);
7277 trace_btrfs_reserve_extent_cluster(root,
7278 block_group, search_start,
7282 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7283 && !failed_cluster_refill) {
7284 spin_unlock(&last_ptr->refill_lock);
7286 failed_cluster_refill = true;
7287 wait_block_group_cache_progress(block_group,
7288 num_bytes + empty_cluster + empty_size);
7289 goto have_block_group;
7293 * at this point we either didn't find a cluster
7294 * or we weren't able to allocate a block from our
7295 * cluster. Free the cluster we've been trying
7296 * to use, and go to the next block group
7298 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7299 spin_unlock(&last_ptr->refill_lock);
7305 * We are doing an unclustered alloc, set the fragmented flag so
7306 * we don't bother trying to setup a cluster again until we get
7309 if (unlikely(last_ptr)) {
7310 spin_lock(&last_ptr->lock);
7311 last_ptr->fragmented = 1;
7312 spin_unlock(&last_ptr->lock);
7314 spin_lock(&block_group->free_space_ctl->tree_lock);
7316 block_group->free_space_ctl->free_space <
7317 num_bytes + empty_cluster + empty_size) {
7318 if (block_group->free_space_ctl->free_space >
7321 block_group->free_space_ctl->free_space;
7322 spin_unlock(&block_group->free_space_ctl->tree_lock);
7325 spin_unlock(&block_group->free_space_ctl->tree_lock);
7327 offset = btrfs_find_space_for_alloc(block_group, search_start,
7328 num_bytes, empty_size,
7331 * If we didn't find a chunk, and we haven't failed on this
7332 * block group before, and this block group is in the middle of
7333 * caching and we are ok with waiting, then go ahead and wait
7334 * for progress to be made, and set failed_alloc to true.
7336 * If failed_alloc is true then we've already waited on this
7337 * block group once and should move on to the next block group.
7339 if (!offset && !failed_alloc && !cached &&
7340 loop > LOOP_CACHING_NOWAIT) {
7341 wait_block_group_cache_progress(block_group,
7342 num_bytes + empty_size);
7343 failed_alloc = true;
7344 goto have_block_group;
7345 } else if (!offset) {
7349 search_start = ALIGN(offset, root->stripesize);
7351 /* move on to the next group */
7352 if (search_start + num_bytes >
7353 block_group->key.objectid + block_group->key.offset) {
7354 btrfs_add_free_space(block_group, offset, num_bytes);
7358 if (offset < search_start)
7359 btrfs_add_free_space(block_group, offset,
7360 search_start - offset);
7361 BUG_ON(offset > search_start);
7363 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7364 alloc_type, delalloc);
7365 if (ret == -EAGAIN) {
7366 btrfs_add_free_space(block_group, offset, num_bytes);
7370 /* we are all good, lets return */
7371 ins->objectid = search_start;
7372 ins->offset = num_bytes;
7374 trace_btrfs_reserve_extent(orig_root, block_group,
7375 search_start, num_bytes);
7376 btrfs_release_block_group(block_group, delalloc);
7379 failed_cluster_refill = false;
7380 failed_alloc = false;
7381 BUG_ON(index != get_block_group_index(block_group));
7382 btrfs_release_block_group(block_group, delalloc);
7384 up_read(&space_info->groups_sem);
7386 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7387 && !orig_have_caching_bg)
7388 orig_have_caching_bg = true;
7390 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7393 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7397 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7398 * caching kthreads as we move along
7399 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7400 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7401 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7404 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7406 if (loop == LOOP_CACHING_NOWAIT) {
7408 * We want to skip the LOOP_CACHING_WAIT step if we
7409 * don't have any unached bgs and we've alrelady done a
7410 * full search through.
7412 if (orig_have_caching_bg || !full_search)
7413 loop = LOOP_CACHING_WAIT;
7415 loop = LOOP_ALLOC_CHUNK;
7420 if (loop == LOOP_ALLOC_CHUNK) {
7421 struct btrfs_trans_handle *trans;
7424 trans = current->journal_info;
7428 trans = btrfs_join_transaction(root);
7430 if (IS_ERR(trans)) {
7431 ret = PTR_ERR(trans);
7435 ret = do_chunk_alloc(trans, root, flags,
7439 * If we can't allocate a new chunk we've already looped
7440 * through at least once, move on to the NO_EMPTY_SIZE
7444 loop = LOOP_NO_EMPTY_SIZE;
7447 * Do not bail out on ENOSPC since we
7448 * can do more things.
7450 if (ret < 0 && ret != -ENOSPC)
7451 btrfs_abort_transaction(trans,
7456 btrfs_end_transaction(trans, root);
7461 if (loop == LOOP_NO_EMPTY_SIZE) {
7463 * Don't loop again if we already have no empty_size and
7466 if (empty_size == 0 &&
7467 empty_cluster == 0) {
7476 } else if (!ins->objectid) {
7478 } else if (ins->objectid) {
7479 if (!use_cluster && last_ptr) {
7480 spin_lock(&last_ptr->lock);
7481 last_ptr->window_start = ins->objectid;
7482 spin_unlock(&last_ptr->lock);
7487 if (ret == -ENOSPC) {
7488 spin_lock(&space_info->lock);
7489 space_info->max_extent_size = max_extent_size;
7490 spin_unlock(&space_info->lock);
7491 ins->offset = max_extent_size;
7496 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7497 int dump_block_groups)
7499 struct btrfs_block_group_cache *cache;
7502 spin_lock(&info->lock);
7503 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7505 info->total_bytes - info->bytes_used - info->bytes_pinned -
7506 info->bytes_reserved - info->bytes_readonly,
7507 (info->full) ? "" : "not ");
7508 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7509 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7510 info->total_bytes, info->bytes_used, info->bytes_pinned,
7511 info->bytes_reserved, info->bytes_may_use,
7512 info->bytes_readonly);
7513 spin_unlock(&info->lock);
7515 if (!dump_block_groups)
7518 down_read(&info->groups_sem);
7520 list_for_each_entry(cache, &info->block_groups[index], list) {
7521 spin_lock(&cache->lock);
7522 printk(KERN_INFO "BTRFS: "
7523 "block group %llu has %llu bytes, "
7524 "%llu used %llu pinned %llu reserved %s\n",
7525 cache->key.objectid, cache->key.offset,
7526 btrfs_block_group_used(&cache->item), cache->pinned,
7527 cache->reserved, cache->ro ? "[readonly]" : "");
7528 btrfs_dump_free_space(cache, bytes);
7529 spin_unlock(&cache->lock);
7531 if (++index < BTRFS_NR_RAID_TYPES)
7533 up_read(&info->groups_sem);
7536 int btrfs_reserve_extent(struct btrfs_root *root,
7537 u64 num_bytes, u64 min_alloc_size,
7538 u64 empty_size, u64 hint_byte,
7539 struct btrfs_key *ins, int is_data, int delalloc)
7541 bool final_tried = num_bytes == min_alloc_size;
7545 flags = btrfs_get_alloc_profile(root, is_data);
7547 WARN_ON(num_bytes < root->sectorsize);
7548 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7551 if (ret == -ENOSPC) {
7552 if (!final_tried && ins->offset) {
7553 num_bytes = min(num_bytes >> 1, ins->offset);
7554 num_bytes = round_down(num_bytes, root->sectorsize);
7555 num_bytes = max(num_bytes, min_alloc_size);
7556 if (num_bytes == min_alloc_size)
7559 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7560 struct btrfs_space_info *sinfo;
7562 sinfo = __find_space_info(root->fs_info, flags);
7563 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7566 dump_space_info(sinfo, num_bytes, 1);
7573 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7575 int pin, int delalloc)
7577 struct btrfs_block_group_cache *cache;
7580 cache = btrfs_lookup_block_group(root->fs_info, start);
7582 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7588 pin_down_extent(root, cache, start, len, 1);
7590 if (btrfs_test_opt(root, DISCARD))
7591 ret = btrfs_discard_extent(root, start, len, NULL);
7592 btrfs_add_free_space(cache, start, len);
7593 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7596 btrfs_put_block_group(cache);
7598 trace_btrfs_reserved_extent_free(root, start, len);
7603 int btrfs_free_reserved_extent(struct btrfs_root *root,
7604 u64 start, u64 len, int delalloc)
7606 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7609 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7612 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7615 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7616 struct btrfs_root *root,
7617 u64 parent, u64 root_objectid,
7618 u64 flags, u64 owner, u64 offset,
7619 struct btrfs_key *ins, int ref_mod)
7622 struct btrfs_fs_info *fs_info = root->fs_info;
7623 struct btrfs_extent_item *extent_item;
7624 struct btrfs_extent_inline_ref *iref;
7625 struct btrfs_path *path;
7626 struct extent_buffer *leaf;
7631 type = BTRFS_SHARED_DATA_REF_KEY;
7633 type = BTRFS_EXTENT_DATA_REF_KEY;
7635 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7637 path = btrfs_alloc_path();
7641 path->leave_spinning = 1;
7642 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7645 btrfs_free_path(path);
7649 leaf = path->nodes[0];
7650 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7651 struct btrfs_extent_item);
7652 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7653 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7654 btrfs_set_extent_flags(leaf, extent_item,
7655 flags | BTRFS_EXTENT_FLAG_DATA);
7657 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7658 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7660 struct btrfs_shared_data_ref *ref;
7661 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7662 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7663 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7665 struct btrfs_extent_data_ref *ref;
7666 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7667 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7668 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7669 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7670 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7673 btrfs_mark_buffer_dirty(path->nodes[0]);
7674 btrfs_free_path(path);
7676 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7677 if (ret) { /* -ENOENT, logic error */
7678 btrfs_err(fs_info, "update block group failed for %llu %llu",
7679 ins->objectid, ins->offset);
7682 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7686 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7687 struct btrfs_root *root,
7688 u64 parent, u64 root_objectid,
7689 u64 flags, struct btrfs_disk_key *key,
7690 int level, struct btrfs_key *ins)
7693 struct btrfs_fs_info *fs_info = root->fs_info;
7694 struct btrfs_extent_item *extent_item;
7695 struct btrfs_tree_block_info *block_info;
7696 struct btrfs_extent_inline_ref *iref;
7697 struct btrfs_path *path;
7698 struct extent_buffer *leaf;
7699 u32 size = sizeof(*extent_item) + sizeof(*iref);
7700 u64 num_bytes = ins->offset;
7701 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7704 if (!skinny_metadata)
7705 size += sizeof(*block_info);
7707 path = btrfs_alloc_path();
7709 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7714 path->leave_spinning = 1;
7715 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7718 btrfs_free_path(path);
7719 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7724 leaf = path->nodes[0];
7725 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7726 struct btrfs_extent_item);
7727 btrfs_set_extent_refs(leaf, extent_item, 1);
7728 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7729 btrfs_set_extent_flags(leaf, extent_item,
7730 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7732 if (skinny_metadata) {
7733 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7734 num_bytes = root->nodesize;
7736 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7737 btrfs_set_tree_block_key(leaf, block_info, key);
7738 btrfs_set_tree_block_level(leaf, block_info, level);
7739 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7743 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7744 btrfs_set_extent_inline_ref_type(leaf, iref,
7745 BTRFS_SHARED_BLOCK_REF_KEY);
7746 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7748 btrfs_set_extent_inline_ref_type(leaf, iref,
7749 BTRFS_TREE_BLOCK_REF_KEY);
7750 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7753 btrfs_mark_buffer_dirty(leaf);
7754 btrfs_free_path(path);
7756 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7758 if (ret) { /* -ENOENT, logic error */
7759 btrfs_err(fs_info, "update block group failed for %llu %llu",
7760 ins->objectid, ins->offset);
7764 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7768 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7769 struct btrfs_root *root,
7770 u64 root_objectid, u64 owner,
7771 u64 offset, u64 ram_bytes,
7772 struct btrfs_key *ins)
7776 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7778 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7780 root_objectid, owner, offset,
7781 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
7787 * this is used by the tree logging recovery code. It records that
7788 * an extent has been allocated and makes sure to clear the free
7789 * space cache bits as well
7791 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7792 struct btrfs_root *root,
7793 u64 root_objectid, u64 owner, u64 offset,
7794 struct btrfs_key *ins)
7797 struct btrfs_block_group_cache *block_group;
7800 * Mixed block groups will exclude before processing the log so we only
7801 * need to do the exlude dance if this fs isn't mixed.
7803 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7804 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7809 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7813 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7814 RESERVE_ALLOC_NO_ACCOUNT, 0);
7815 BUG_ON(ret); /* logic error */
7816 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7817 0, owner, offset, ins, 1);
7818 btrfs_put_block_group(block_group);
7822 static struct extent_buffer *
7823 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7824 u64 bytenr, int level)
7826 struct extent_buffer *buf;
7828 buf = btrfs_find_create_tree_block(root, bytenr);
7830 return ERR_PTR(-ENOMEM);
7831 btrfs_set_header_generation(buf, trans->transid);
7832 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7833 btrfs_tree_lock(buf);
7834 clean_tree_block(trans, root->fs_info, buf);
7835 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7837 btrfs_set_lock_blocking(buf);
7838 btrfs_set_buffer_uptodate(buf);
7840 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7841 buf->log_index = root->log_transid % 2;
7843 * we allow two log transactions at a time, use different
7844 * EXENT bit to differentiate dirty pages.
7846 if (buf->log_index == 0)
7847 set_extent_dirty(&root->dirty_log_pages, buf->start,
7848 buf->start + buf->len - 1, GFP_NOFS);
7850 set_extent_new(&root->dirty_log_pages, buf->start,
7851 buf->start + buf->len - 1, GFP_NOFS);
7853 buf->log_index = -1;
7854 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7855 buf->start + buf->len - 1, GFP_NOFS);
7857 trans->blocks_used++;
7858 /* this returns a buffer locked for blocking */
7862 static struct btrfs_block_rsv *
7863 use_block_rsv(struct btrfs_trans_handle *trans,
7864 struct btrfs_root *root, u32 blocksize)
7866 struct btrfs_block_rsv *block_rsv;
7867 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7869 bool global_updated = false;
7871 block_rsv = get_block_rsv(trans, root);
7873 if (unlikely(block_rsv->size == 0))
7876 ret = block_rsv_use_bytes(block_rsv, blocksize);
7880 if (block_rsv->failfast)
7881 return ERR_PTR(ret);
7883 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7884 global_updated = true;
7885 update_global_block_rsv(root->fs_info);
7889 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7890 static DEFINE_RATELIMIT_STATE(_rs,
7891 DEFAULT_RATELIMIT_INTERVAL * 10,
7892 /*DEFAULT_RATELIMIT_BURST*/ 1);
7893 if (__ratelimit(&_rs))
7895 "BTRFS: block rsv returned %d\n", ret);
7898 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7899 BTRFS_RESERVE_NO_FLUSH);
7903 * If we couldn't reserve metadata bytes try and use some from
7904 * the global reserve if its space type is the same as the global
7907 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7908 block_rsv->space_info == global_rsv->space_info) {
7909 ret = block_rsv_use_bytes(global_rsv, blocksize);
7913 return ERR_PTR(ret);
7916 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7917 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7919 block_rsv_add_bytes(block_rsv, blocksize, 0);
7920 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7924 * finds a free extent and does all the dirty work required for allocation
7925 * returns the tree buffer or an ERR_PTR on error.
7927 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7928 struct btrfs_root *root,
7929 u64 parent, u64 root_objectid,
7930 struct btrfs_disk_key *key, int level,
7931 u64 hint, u64 empty_size)
7933 struct btrfs_key ins;
7934 struct btrfs_block_rsv *block_rsv;
7935 struct extent_buffer *buf;
7936 struct btrfs_delayed_extent_op *extent_op;
7939 u32 blocksize = root->nodesize;
7940 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7943 if (btrfs_test_is_dummy_root(root)) {
7944 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7947 root->alloc_bytenr += blocksize;
7951 block_rsv = use_block_rsv(trans, root, blocksize);
7952 if (IS_ERR(block_rsv))
7953 return ERR_CAST(block_rsv);
7955 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7956 empty_size, hint, &ins, 0, 0);
7960 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7963 goto out_free_reserved;
7966 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7968 parent = ins.objectid;
7969 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7973 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7974 extent_op = btrfs_alloc_delayed_extent_op();
7980 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7982 memset(&extent_op->key, 0, sizeof(extent_op->key));
7983 extent_op->flags_to_set = flags;
7984 if (skinny_metadata)
7985 extent_op->update_key = 0;
7987 extent_op->update_key = 1;
7988 extent_op->update_flags = 1;
7989 extent_op->is_data = 0;
7990 extent_op->level = level;
7992 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7993 ins.objectid, ins.offset,
7994 parent, root_objectid, level,
7995 BTRFS_ADD_DELAYED_EXTENT,
7998 goto out_free_delayed;
8003 btrfs_free_delayed_extent_op(extent_op);
8005 free_extent_buffer(buf);
8007 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8009 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8010 return ERR_PTR(ret);
8013 struct walk_control {
8014 u64 refs[BTRFS_MAX_LEVEL];
8015 u64 flags[BTRFS_MAX_LEVEL];
8016 struct btrfs_key update_progress;
8027 #define DROP_REFERENCE 1
8028 #define UPDATE_BACKREF 2
8030 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8031 struct btrfs_root *root,
8032 struct walk_control *wc,
8033 struct btrfs_path *path)
8041 struct btrfs_key key;
8042 struct extent_buffer *eb;
8047 if (path->slots[wc->level] < wc->reada_slot) {
8048 wc->reada_count = wc->reada_count * 2 / 3;
8049 wc->reada_count = max(wc->reada_count, 2);
8051 wc->reada_count = wc->reada_count * 3 / 2;
8052 wc->reada_count = min_t(int, wc->reada_count,
8053 BTRFS_NODEPTRS_PER_BLOCK(root));
8056 eb = path->nodes[wc->level];
8057 nritems = btrfs_header_nritems(eb);
8058 blocksize = root->nodesize;
8060 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8061 if (nread >= wc->reada_count)
8065 bytenr = btrfs_node_blockptr(eb, slot);
8066 generation = btrfs_node_ptr_generation(eb, slot);
8068 if (slot == path->slots[wc->level])
8071 if (wc->stage == UPDATE_BACKREF &&
8072 generation <= root->root_key.offset)
8075 /* We don't lock the tree block, it's OK to be racy here */
8076 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8077 wc->level - 1, 1, &refs,
8079 /* We don't care about errors in readahead. */
8084 if (wc->stage == DROP_REFERENCE) {
8088 if (wc->level == 1 &&
8089 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8091 if (!wc->update_ref ||
8092 generation <= root->root_key.offset)
8094 btrfs_node_key_to_cpu(eb, &key, slot);
8095 ret = btrfs_comp_cpu_keys(&key,
8096 &wc->update_progress);
8100 if (wc->level == 1 &&
8101 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8105 readahead_tree_block(root, bytenr);
8108 wc->reada_slot = slot;
8112 * These may not be seen by the usual inc/dec ref code so we have to
8115 static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8116 struct btrfs_root *root, u64 bytenr,
8119 struct btrfs_qgroup_extent_record *qrecord;
8120 struct btrfs_delayed_ref_root *delayed_refs;
8122 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8126 qrecord->bytenr = bytenr;
8127 qrecord->num_bytes = num_bytes;
8128 qrecord->old_roots = NULL;
8130 delayed_refs = &trans->transaction->delayed_refs;
8131 spin_lock(&delayed_refs->lock);
8132 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8134 spin_unlock(&delayed_refs->lock);
8139 static int account_leaf_items(struct btrfs_trans_handle *trans,
8140 struct btrfs_root *root,
8141 struct extent_buffer *eb)
8143 int nr = btrfs_header_nritems(eb);
8144 int i, extent_type, ret;
8145 struct btrfs_key key;
8146 struct btrfs_file_extent_item *fi;
8147 u64 bytenr, num_bytes;
8149 /* We can be called directly from walk_up_proc() */
8150 if (!root->fs_info->quota_enabled)
8153 for (i = 0; i < nr; i++) {
8154 btrfs_item_key_to_cpu(eb, &key, i);
8156 if (key.type != BTRFS_EXTENT_DATA_KEY)
8159 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8160 /* filter out non qgroup-accountable extents */
8161 extent_type = btrfs_file_extent_type(eb, fi);
8163 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8166 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8170 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8172 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8180 * Walk up the tree from the bottom, freeing leaves and any interior
8181 * nodes which have had all slots visited. If a node (leaf or
8182 * interior) is freed, the node above it will have it's slot
8183 * incremented. The root node will never be freed.
8185 * At the end of this function, we should have a path which has all
8186 * slots incremented to the next position for a search. If we need to
8187 * read a new node it will be NULL and the node above it will have the
8188 * correct slot selected for a later read.
8190 * If we increment the root nodes slot counter past the number of
8191 * elements, 1 is returned to signal completion of the search.
8193 static int adjust_slots_upwards(struct btrfs_root *root,
8194 struct btrfs_path *path, int root_level)
8198 struct extent_buffer *eb;
8200 if (root_level == 0)
8203 while (level <= root_level) {
8204 eb = path->nodes[level];
8205 nr = btrfs_header_nritems(eb);
8206 path->slots[level]++;
8207 slot = path->slots[level];
8208 if (slot >= nr || level == 0) {
8210 * Don't free the root - we will detect this
8211 * condition after our loop and return a
8212 * positive value for caller to stop walking the tree.
8214 if (level != root_level) {
8215 btrfs_tree_unlock_rw(eb, path->locks[level]);
8216 path->locks[level] = 0;
8218 free_extent_buffer(eb);
8219 path->nodes[level] = NULL;
8220 path->slots[level] = 0;
8224 * We have a valid slot to walk back down
8225 * from. Stop here so caller can process these
8234 eb = path->nodes[root_level];
8235 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8242 * root_eb is the subtree root and is locked before this function is called.
8244 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8245 struct btrfs_root *root,
8246 struct extent_buffer *root_eb,
8252 struct extent_buffer *eb = root_eb;
8253 struct btrfs_path *path = NULL;
8255 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8256 BUG_ON(root_eb == NULL);
8258 if (!root->fs_info->quota_enabled)
8261 if (!extent_buffer_uptodate(root_eb)) {
8262 ret = btrfs_read_buffer(root_eb, root_gen);
8267 if (root_level == 0) {
8268 ret = account_leaf_items(trans, root, root_eb);
8272 path = btrfs_alloc_path();
8277 * Walk down the tree. Missing extent blocks are filled in as
8278 * we go. Metadata is accounted every time we read a new
8281 * When we reach a leaf, we account for file extent items in it,
8282 * walk back up the tree (adjusting slot pointers as we go)
8283 * and restart the search process.
8285 extent_buffer_get(root_eb); /* For path */
8286 path->nodes[root_level] = root_eb;
8287 path->slots[root_level] = 0;
8288 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8291 while (level >= 0) {
8292 if (path->nodes[level] == NULL) {
8297 /* We need to get child blockptr/gen from
8298 * parent before we can read it. */
8299 eb = path->nodes[level + 1];
8300 parent_slot = path->slots[level + 1];
8301 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8302 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8304 eb = read_tree_block(root, child_bytenr, child_gen);
8308 } else if (!extent_buffer_uptodate(eb)) {
8309 free_extent_buffer(eb);
8314 path->nodes[level] = eb;
8315 path->slots[level] = 0;
8317 btrfs_tree_read_lock(eb);
8318 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8319 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8321 ret = record_one_subtree_extent(trans, root, child_bytenr,
8328 ret = account_leaf_items(trans, root, path->nodes[level]);
8332 /* Nonzero return here means we completed our search */
8333 ret = adjust_slots_upwards(root, path, root_level);
8337 /* Restart search with new slots */
8346 btrfs_free_path(path);
8352 * helper to process tree block while walking down the tree.
8354 * when wc->stage == UPDATE_BACKREF, this function updates
8355 * back refs for pointers in the block.
8357 * NOTE: return value 1 means we should stop walking down.
8359 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8360 struct btrfs_root *root,
8361 struct btrfs_path *path,
8362 struct walk_control *wc, int lookup_info)
8364 int level = wc->level;
8365 struct extent_buffer *eb = path->nodes[level];
8366 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8369 if (wc->stage == UPDATE_BACKREF &&
8370 btrfs_header_owner(eb) != root->root_key.objectid)
8374 * when reference count of tree block is 1, it won't increase
8375 * again. once full backref flag is set, we never clear it.
8378 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8379 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8380 BUG_ON(!path->locks[level]);
8381 ret = btrfs_lookup_extent_info(trans, root,
8382 eb->start, level, 1,
8385 BUG_ON(ret == -ENOMEM);
8388 BUG_ON(wc->refs[level] == 0);
8391 if (wc->stage == DROP_REFERENCE) {
8392 if (wc->refs[level] > 1)
8395 if (path->locks[level] && !wc->keep_locks) {
8396 btrfs_tree_unlock_rw(eb, path->locks[level]);
8397 path->locks[level] = 0;
8402 /* wc->stage == UPDATE_BACKREF */
8403 if (!(wc->flags[level] & flag)) {
8404 BUG_ON(!path->locks[level]);
8405 ret = btrfs_inc_ref(trans, root, eb, 1);
8406 BUG_ON(ret); /* -ENOMEM */
8407 ret = btrfs_dec_ref(trans, root, eb, 0);
8408 BUG_ON(ret); /* -ENOMEM */
8409 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8411 btrfs_header_level(eb), 0);
8412 BUG_ON(ret); /* -ENOMEM */
8413 wc->flags[level] |= flag;
8417 * the block is shared by multiple trees, so it's not good to
8418 * keep the tree lock
8420 if (path->locks[level] && level > 0) {
8421 btrfs_tree_unlock_rw(eb, path->locks[level]);
8422 path->locks[level] = 0;
8428 * helper to process tree block pointer.
8430 * when wc->stage == DROP_REFERENCE, this function checks
8431 * reference count of the block pointed to. if the block
8432 * is shared and we need update back refs for the subtree
8433 * rooted at the block, this function changes wc->stage to
8434 * UPDATE_BACKREF. if the block is shared and there is no
8435 * need to update back, this function drops the reference
8438 * NOTE: return value 1 means we should stop walking down.
8440 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8441 struct btrfs_root *root,
8442 struct btrfs_path *path,
8443 struct walk_control *wc, int *lookup_info)
8449 struct btrfs_key key;
8450 struct extent_buffer *next;
8451 int level = wc->level;
8454 bool need_account = false;
8456 generation = btrfs_node_ptr_generation(path->nodes[level],
8457 path->slots[level]);
8459 * if the lower level block was created before the snapshot
8460 * was created, we know there is no need to update back refs
8463 if (wc->stage == UPDATE_BACKREF &&
8464 generation <= root->root_key.offset) {
8469 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8470 blocksize = root->nodesize;
8472 next = btrfs_find_tree_block(root->fs_info, bytenr);
8474 next = btrfs_find_create_tree_block(root, bytenr);
8477 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8481 btrfs_tree_lock(next);
8482 btrfs_set_lock_blocking(next);
8484 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8485 &wc->refs[level - 1],
8486 &wc->flags[level - 1]);
8488 btrfs_tree_unlock(next);
8492 if (unlikely(wc->refs[level - 1] == 0)) {
8493 btrfs_err(root->fs_info, "Missing references.");
8498 if (wc->stage == DROP_REFERENCE) {
8499 if (wc->refs[level - 1] > 1) {
8500 need_account = true;
8502 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8505 if (!wc->update_ref ||
8506 generation <= root->root_key.offset)
8509 btrfs_node_key_to_cpu(path->nodes[level], &key,
8510 path->slots[level]);
8511 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8515 wc->stage = UPDATE_BACKREF;
8516 wc->shared_level = level - 1;
8520 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8524 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8525 btrfs_tree_unlock(next);
8526 free_extent_buffer(next);
8532 if (reada && level == 1)
8533 reada_walk_down(trans, root, wc, path);
8534 next = read_tree_block(root, bytenr, generation);
8536 return PTR_ERR(next);
8537 } else if (!extent_buffer_uptodate(next)) {
8538 free_extent_buffer(next);
8541 btrfs_tree_lock(next);
8542 btrfs_set_lock_blocking(next);
8546 BUG_ON(level != btrfs_header_level(next));
8547 path->nodes[level] = next;
8548 path->slots[level] = 0;
8549 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8555 wc->refs[level - 1] = 0;
8556 wc->flags[level - 1] = 0;
8557 if (wc->stage == DROP_REFERENCE) {
8558 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8559 parent = path->nodes[level]->start;
8561 BUG_ON(root->root_key.objectid !=
8562 btrfs_header_owner(path->nodes[level]));
8567 ret = account_shared_subtree(trans, root, next,
8568 generation, level - 1);
8570 btrfs_err_rl(root->fs_info,
8572 "%d accounting shared subtree. Quota "
8573 "is out of sync, rescan required.",
8577 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8578 root->root_key.objectid, level - 1, 0);
8579 BUG_ON(ret); /* -ENOMEM */
8581 btrfs_tree_unlock(next);
8582 free_extent_buffer(next);
8588 * helper to process tree block while walking up the tree.
8590 * when wc->stage == DROP_REFERENCE, this function drops
8591 * reference count on the block.
8593 * when wc->stage == UPDATE_BACKREF, this function changes
8594 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8595 * to UPDATE_BACKREF previously while processing the block.
8597 * NOTE: return value 1 means we should stop walking up.
8599 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8600 struct btrfs_root *root,
8601 struct btrfs_path *path,
8602 struct walk_control *wc)
8605 int level = wc->level;
8606 struct extent_buffer *eb = path->nodes[level];
8609 if (wc->stage == UPDATE_BACKREF) {
8610 BUG_ON(wc->shared_level < level);
8611 if (level < wc->shared_level)
8614 ret = find_next_key(path, level + 1, &wc->update_progress);
8618 wc->stage = DROP_REFERENCE;
8619 wc->shared_level = -1;
8620 path->slots[level] = 0;
8623 * check reference count again if the block isn't locked.
8624 * we should start walking down the tree again if reference
8627 if (!path->locks[level]) {
8629 btrfs_tree_lock(eb);
8630 btrfs_set_lock_blocking(eb);
8631 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8633 ret = btrfs_lookup_extent_info(trans, root,
8634 eb->start, level, 1,
8638 btrfs_tree_unlock_rw(eb, path->locks[level]);
8639 path->locks[level] = 0;
8642 BUG_ON(wc->refs[level] == 0);
8643 if (wc->refs[level] == 1) {
8644 btrfs_tree_unlock_rw(eb, path->locks[level]);
8645 path->locks[level] = 0;
8651 /* wc->stage == DROP_REFERENCE */
8652 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8654 if (wc->refs[level] == 1) {
8656 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8657 ret = btrfs_dec_ref(trans, root, eb, 1);
8659 ret = btrfs_dec_ref(trans, root, eb, 0);
8660 BUG_ON(ret); /* -ENOMEM */
8661 ret = account_leaf_items(trans, root, eb);
8663 btrfs_err_rl(root->fs_info,
8665 "%d accounting leaf items. Quota "
8666 "is out of sync, rescan required.",
8670 /* make block locked assertion in clean_tree_block happy */
8671 if (!path->locks[level] &&
8672 btrfs_header_generation(eb) == trans->transid) {
8673 btrfs_tree_lock(eb);
8674 btrfs_set_lock_blocking(eb);
8675 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8677 clean_tree_block(trans, root->fs_info, eb);
8680 if (eb == root->node) {
8681 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8684 BUG_ON(root->root_key.objectid !=
8685 btrfs_header_owner(eb));
8687 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8688 parent = path->nodes[level + 1]->start;
8690 BUG_ON(root->root_key.objectid !=
8691 btrfs_header_owner(path->nodes[level + 1]));
8694 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8696 wc->refs[level] = 0;
8697 wc->flags[level] = 0;
8701 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8702 struct btrfs_root *root,
8703 struct btrfs_path *path,
8704 struct walk_control *wc)
8706 int level = wc->level;
8707 int lookup_info = 1;
8710 while (level >= 0) {
8711 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8718 if (path->slots[level] >=
8719 btrfs_header_nritems(path->nodes[level]))
8722 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8724 path->slots[level]++;
8733 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8734 struct btrfs_root *root,
8735 struct btrfs_path *path,
8736 struct walk_control *wc, int max_level)
8738 int level = wc->level;
8741 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8742 while (level < max_level && path->nodes[level]) {
8744 if (path->slots[level] + 1 <
8745 btrfs_header_nritems(path->nodes[level])) {
8746 path->slots[level]++;
8749 ret = walk_up_proc(trans, root, path, wc);
8753 if (path->locks[level]) {
8754 btrfs_tree_unlock_rw(path->nodes[level],
8755 path->locks[level]);
8756 path->locks[level] = 0;
8758 free_extent_buffer(path->nodes[level]);
8759 path->nodes[level] = NULL;
8767 * drop a subvolume tree.
8769 * this function traverses the tree freeing any blocks that only
8770 * referenced by the tree.
8772 * when a shared tree block is found. this function decreases its
8773 * reference count by one. if update_ref is true, this function
8774 * also make sure backrefs for the shared block and all lower level
8775 * blocks are properly updated.
8777 * If called with for_reloc == 0, may exit early with -EAGAIN
8779 int btrfs_drop_snapshot(struct btrfs_root *root,
8780 struct btrfs_block_rsv *block_rsv, int update_ref,
8783 struct btrfs_path *path;
8784 struct btrfs_trans_handle *trans;
8785 struct btrfs_root *tree_root = root->fs_info->tree_root;
8786 struct btrfs_root_item *root_item = &root->root_item;
8787 struct walk_control *wc;
8788 struct btrfs_key key;
8792 bool root_dropped = false;
8794 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8796 path = btrfs_alloc_path();
8802 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8804 btrfs_free_path(path);
8809 trans = btrfs_start_transaction(tree_root, 0);
8810 if (IS_ERR(trans)) {
8811 err = PTR_ERR(trans);
8816 trans->block_rsv = block_rsv;
8818 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8819 level = btrfs_header_level(root->node);
8820 path->nodes[level] = btrfs_lock_root_node(root);
8821 btrfs_set_lock_blocking(path->nodes[level]);
8822 path->slots[level] = 0;
8823 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8824 memset(&wc->update_progress, 0,
8825 sizeof(wc->update_progress));
8827 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8828 memcpy(&wc->update_progress, &key,
8829 sizeof(wc->update_progress));
8831 level = root_item->drop_level;
8833 path->lowest_level = level;
8834 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8835 path->lowest_level = 0;
8843 * unlock our path, this is safe because only this
8844 * function is allowed to delete this snapshot
8846 btrfs_unlock_up_safe(path, 0);
8848 level = btrfs_header_level(root->node);
8850 btrfs_tree_lock(path->nodes[level]);
8851 btrfs_set_lock_blocking(path->nodes[level]);
8852 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8854 ret = btrfs_lookup_extent_info(trans, root,
8855 path->nodes[level]->start,
8856 level, 1, &wc->refs[level],
8862 BUG_ON(wc->refs[level] == 0);
8864 if (level == root_item->drop_level)
8867 btrfs_tree_unlock(path->nodes[level]);
8868 path->locks[level] = 0;
8869 WARN_ON(wc->refs[level] != 1);
8875 wc->shared_level = -1;
8876 wc->stage = DROP_REFERENCE;
8877 wc->update_ref = update_ref;
8879 wc->for_reloc = for_reloc;
8880 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8884 ret = walk_down_tree(trans, root, path, wc);
8890 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8897 BUG_ON(wc->stage != DROP_REFERENCE);
8901 if (wc->stage == DROP_REFERENCE) {
8903 btrfs_node_key(path->nodes[level],
8904 &root_item->drop_progress,
8905 path->slots[level]);
8906 root_item->drop_level = level;
8909 BUG_ON(wc->level == 0);
8910 if (btrfs_should_end_transaction(trans, tree_root) ||
8911 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8912 ret = btrfs_update_root(trans, tree_root,
8916 btrfs_abort_transaction(trans, tree_root, ret);
8921 btrfs_end_transaction_throttle(trans, tree_root);
8922 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8923 pr_debug("BTRFS: drop snapshot early exit\n");
8928 trans = btrfs_start_transaction(tree_root, 0);
8929 if (IS_ERR(trans)) {
8930 err = PTR_ERR(trans);
8934 trans->block_rsv = block_rsv;
8937 btrfs_release_path(path);
8941 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8943 btrfs_abort_transaction(trans, tree_root, ret);
8947 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8948 ret = btrfs_find_root(tree_root, &root->root_key, path,
8951 btrfs_abort_transaction(trans, tree_root, ret);
8954 } else if (ret > 0) {
8955 /* if we fail to delete the orphan item this time
8956 * around, it'll get picked up the next time.
8958 * The most common failure here is just -ENOENT.
8960 btrfs_del_orphan_item(trans, tree_root,
8961 root->root_key.objectid);
8965 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8966 btrfs_add_dropped_root(trans, root);
8968 free_extent_buffer(root->node);
8969 free_extent_buffer(root->commit_root);
8970 btrfs_put_fs_root(root);
8972 root_dropped = true;
8974 btrfs_end_transaction_throttle(trans, tree_root);
8977 btrfs_free_path(path);
8980 * So if we need to stop dropping the snapshot for whatever reason we
8981 * need to make sure to add it back to the dead root list so that we
8982 * keep trying to do the work later. This also cleans up roots if we
8983 * don't have it in the radix (like when we recover after a power fail
8984 * or unmount) so we don't leak memory.
8986 if (!for_reloc && root_dropped == false)
8987 btrfs_add_dead_root(root);
8988 if (err && err != -EAGAIN)
8989 btrfs_std_error(root->fs_info, err, NULL);
8994 * drop subtree rooted at tree block 'node'.
8996 * NOTE: this function will unlock and release tree block 'node'
8997 * only used by relocation code
8999 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9000 struct btrfs_root *root,
9001 struct extent_buffer *node,
9002 struct extent_buffer *parent)
9004 struct btrfs_path *path;
9005 struct walk_control *wc;
9011 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9013 path = btrfs_alloc_path();
9017 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9019 btrfs_free_path(path);
9023 btrfs_assert_tree_locked(parent);
9024 parent_level = btrfs_header_level(parent);
9025 extent_buffer_get(parent);
9026 path->nodes[parent_level] = parent;
9027 path->slots[parent_level] = btrfs_header_nritems(parent);
9029 btrfs_assert_tree_locked(node);
9030 level = btrfs_header_level(node);
9031 path->nodes[level] = node;
9032 path->slots[level] = 0;
9033 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9035 wc->refs[parent_level] = 1;
9036 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9038 wc->shared_level = -1;
9039 wc->stage = DROP_REFERENCE;
9043 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9046 wret = walk_down_tree(trans, root, path, wc);
9052 wret = walk_up_tree(trans, root, path, wc, parent_level);
9060 btrfs_free_path(path);
9064 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9070 * if restripe for this chunk_type is on pick target profile and
9071 * return, otherwise do the usual balance
9073 stripped = get_restripe_target(root->fs_info, flags);
9075 return extended_to_chunk(stripped);
9077 num_devices = root->fs_info->fs_devices->rw_devices;
9079 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9080 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9081 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9083 if (num_devices == 1) {
9084 stripped |= BTRFS_BLOCK_GROUP_DUP;
9085 stripped = flags & ~stripped;
9087 /* turn raid0 into single device chunks */
9088 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9091 /* turn mirroring into duplication */
9092 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9093 BTRFS_BLOCK_GROUP_RAID10))
9094 return stripped | BTRFS_BLOCK_GROUP_DUP;
9096 /* they already had raid on here, just return */
9097 if (flags & stripped)
9100 stripped |= BTRFS_BLOCK_GROUP_DUP;
9101 stripped = flags & ~stripped;
9103 /* switch duplicated blocks with raid1 */
9104 if (flags & BTRFS_BLOCK_GROUP_DUP)
9105 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9107 /* this is drive concat, leave it alone */
9113 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9115 struct btrfs_space_info *sinfo = cache->space_info;
9117 u64 min_allocable_bytes;
9121 * We need some metadata space and system metadata space for
9122 * allocating chunks in some corner cases until we force to set
9123 * it to be readonly.
9126 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9128 min_allocable_bytes = 1 * 1024 * 1024;
9130 min_allocable_bytes = 0;
9132 spin_lock(&sinfo->lock);
9133 spin_lock(&cache->lock);
9141 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9142 cache->bytes_super - btrfs_block_group_used(&cache->item);
9144 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9145 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9146 min_allocable_bytes <= sinfo->total_bytes) {
9147 sinfo->bytes_readonly += num_bytes;
9149 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9153 spin_unlock(&cache->lock);
9154 spin_unlock(&sinfo->lock);
9158 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9159 struct btrfs_block_group_cache *cache)
9162 struct btrfs_trans_handle *trans;
9167 trans = btrfs_join_transaction(root);
9169 return PTR_ERR(trans);
9172 * we're not allowed to set block groups readonly after the dirty
9173 * block groups cache has started writing. If it already started,
9174 * back off and let this transaction commit
9176 mutex_lock(&root->fs_info->ro_block_group_mutex);
9177 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9178 u64 transid = trans->transid;
9180 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9181 btrfs_end_transaction(trans, root);
9183 ret = btrfs_wait_for_commit(root, transid);
9190 * if we are changing raid levels, try to allocate a corresponding
9191 * block group with the new raid level.
9193 alloc_flags = update_block_group_flags(root, cache->flags);
9194 if (alloc_flags != cache->flags) {
9195 ret = do_chunk_alloc(trans, root, alloc_flags,
9198 * ENOSPC is allowed here, we may have enough space
9199 * already allocated at the new raid level to
9208 ret = inc_block_group_ro(cache, 0);
9211 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9212 ret = do_chunk_alloc(trans, root, alloc_flags,
9216 ret = inc_block_group_ro(cache, 0);
9218 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9219 alloc_flags = update_block_group_flags(root, cache->flags);
9220 lock_chunks(root->fs_info->chunk_root);
9221 check_system_chunk(trans, root, alloc_flags);
9222 unlock_chunks(root->fs_info->chunk_root);
9224 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9226 btrfs_end_transaction(trans, root);
9230 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9231 struct btrfs_root *root, u64 type)
9233 u64 alloc_flags = get_alloc_profile(root, type);
9234 return do_chunk_alloc(trans, root, alloc_flags,
9239 * helper to account the unused space of all the readonly block group in the
9240 * space_info. takes mirrors into account.
9242 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9244 struct btrfs_block_group_cache *block_group;
9248 /* It's df, we don't care if it's racey */
9249 if (list_empty(&sinfo->ro_bgs))
9252 spin_lock(&sinfo->lock);
9253 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9254 spin_lock(&block_group->lock);
9256 if (!block_group->ro) {
9257 spin_unlock(&block_group->lock);
9261 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9262 BTRFS_BLOCK_GROUP_RAID10 |
9263 BTRFS_BLOCK_GROUP_DUP))
9268 free_bytes += (block_group->key.offset -
9269 btrfs_block_group_used(&block_group->item)) *
9272 spin_unlock(&block_group->lock);
9274 spin_unlock(&sinfo->lock);
9279 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9280 struct btrfs_block_group_cache *cache)
9282 struct btrfs_space_info *sinfo = cache->space_info;
9287 spin_lock(&sinfo->lock);
9288 spin_lock(&cache->lock);
9290 num_bytes = cache->key.offset - cache->reserved -
9291 cache->pinned - cache->bytes_super -
9292 btrfs_block_group_used(&cache->item);
9293 sinfo->bytes_readonly -= num_bytes;
9294 list_del_init(&cache->ro_list);
9296 spin_unlock(&cache->lock);
9297 spin_unlock(&sinfo->lock);
9301 * checks to see if its even possible to relocate this block group.
9303 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9304 * ok to go ahead and try.
9306 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9308 struct btrfs_block_group_cache *block_group;
9309 struct btrfs_space_info *space_info;
9310 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9311 struct btrfs_device *device;
9312 struct btrfs_trans_handle *trans;
9321 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9323 /* odd, couldn't find the block group, leave it alone */
9327 min_free = btrfs_block_group_used(&block_group->item);
9329 /* no bytes used, we're good */
9333 space_info = block_group->space_info;
9334 spin_lock(&space_info->lock);
9336 full = space_info->full;
9339 * if this is the last block group we have in this space, we can't
9340 * relocate it unless we're able to allocate a new chunk below.
9342 * Otherwise, we need to make sure we have room in the space to handle
9343 * all of the extents from this block group. If we can, we're good
9345 if ((space_info->total_bytes != block_group->key.offset) &&
9346 (space_info->bytes_used + space_info->bytes_reserved +
9347 space_info->bytes_pinned + space_info->bytes_readonly +
9348 min_free < space_info->total_bytes)) {
9349 spin_unlock(&space_info->lock);
9352 spin_unlock(&space_info->lock);
9355 * ok we don't have enough space, but maybe we have free space on our
9356 * devices to allocate new chunks for relocation, so loop through our
9357 * alloc devices and guess if we have enough space. if this block
9358 * group is going to be restriped, run checks against the target
9359 * profile instead of the current one.
9371 target = get_restripe_target(root->fs_info, block_group->flags);
9373 index = __get_raid_index(extended_to_chunk(target));
9376 * this is just a balance, so if we were marked as full
9377 * we know there is no space for a new chunk
9382 index = get_block_group_index(block_group);
9385 if (index == BTRFS_RAID_RAID10) {
9389 } else if (index == BTRFS_RAID_RAID1) {
9391 } else if (index == BTRFS_RAID_DUP) {
9394 } else if (index == BTRFS_RAID_RAID0) {
9395 dev_min = fs_devices->rw_devices;
9396 min_free = div64_u64(min_free, dev_min);
9399 /* We need to do this so that we can look at pending chunks */
9400 trans = btrfs_join_transaction(root);
9401 if (IS_ERR(trans)) {
9402 ret = PTR_ERR(trans);
9406 mutex_lock(&root->fs_info->chunk_mutex);
9407 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9411 * check to make sure we can actually find a chunk with enough
9412 * space to fit our block group in.
9414 if (device->total_bytes > device->bytes_used + min_free &&
9415 !device->is_tgtdev_for_dev_replace) {
9416 ret = find_free_dev_extent(trans, device, min_free,
9421 if (dev_nr >= dev_min)
9427 mutex_unlock(&root->fs_info->chunk_mutex);
9428 btrfs_end_transaction(trans, root);
9430 btrfs_put_block_group(block_group);
9434 static int find_first_block_group(struct btrfs_root *root,
9435 struct btrfs_path *path, struct btrfs_key *key)
9438 struct btrfs_key found_key;
9439 struct extent_buffer *leaf;
9442 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9447 slot = path->slots[0];
9448 leaf = path->nodes[0];
9449 if (slot >= btrfs_header_nritems(leaf)) {
9450 ret = btrfs_next_leaf(root, path);
9457 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9459 if (found_key.objectid >= key->objectid &&
9460 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9470 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9472 struct btrfs_block_group_cache *block_group;
9476 struct inode *inode;
9478 block_group = btrfs_lookup_first_block_group(info, last);
9479 while (block_group) {
9480 spin_lock(&block_group->lock);
9481 if (block_group->iref)
9483 spin_unlock(&block_group->lock);
9484 block_group = next_block_group(info->tree_root,
9494 inode = block_group->inode;
9495 block_group->iref = 0;
9496 block_group->inode = NULL;
9497 spin_unlock(&block_group->lock);
9499 last = block_group->key.objectid + block_group->key.offset;
9500 btrfs_put_block_group(block_group);
9504 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9506 struct btrfs_block_group_cache *block_group;
9507 struct btrfs_space_info *space_info;
9508 struct btrfs_caching_control *caching_ctl;
9511 down_write(&info->commit_root_sem);
9512 while (!list_empty(&info->caching_block_groups)) {
9513 caching_ctl = list_entry(info->caching_block_groups.next,
9514 struct btrfs_caching_control, list);
9515 list_del(&caching_ctl->list);
9516 put_caching_control(caching_ctl);
9518 up_write(&info->commit_root_sem);
9520 spin_lock(&info->unused_bgs_lock);
9521 while (!list_empty(&info->unused_bgs)) {
9522 block_group = list_first_entry(&info->unused_bgs,
9523 struct btrfs_block_group_cache,
9525 list_del_init(&block_group->bg_list);
9526 btrfs_put_block_group(block_group);
9528 spin_unlock(&info->unused_bgs_lock);
9530 spin_lock(&info->block_group_cache_lock);
9531 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9532 block_group = rb_entry(n, struct btrfs_block_group_cache,
9534 rb_erase(&block_group->cache_node,
9535 &info->block_group_cache_tree);
9536 RB_CLEAR_NODE(&block_group->cache_node);
9537 spin_unlock(&info->block_group_cache_lock);
9539 down_write(&block_group->space_info->groups_sem);
9540 list_del(&block_group->list);
9541 up_write(&block_group->space_info->groups_sem);
9543 if (block_group->cached == BTRFS_CACHE_STARTED)
9544 wait_block_group_cache_done(block_group);
9547 * We haven't cached this block group, which means we could
9548 * possibly have excluded extents on this block group.
9550 if (block_group->cached == BTRFS_CACHE_NO ||
9551 block_group->cached == BTRFS_CACHE_ERROR)
9552 free_excluded_extents(info->extent_root, block_group);
9554 btrfs_remove_free_space_cache(block_group);
9555 btrfs_put_block_group(block_group);
9557 spin_lock(&info->block_group_cache_lock);
9559 spin_unlock(&info->block_group_cache_lock);
9561 /* now that all the block groups are freed, go through and
9562 * free all the space_info structs. This is only called during
9563 * the final stages of unmount, and so we know nobody is
9564 * using them. We call synchronize_rcu() once before we start,
9565 * just to be on the safe side.
9569 release_global_block_rsv(info);
9571 while (!list_empty(&info->space_info)) {
9574 space_info = list_entry(info->space_info.next,
9575 struct btrfs_space_info,
9577 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9578 if (WARN_ON(space_info->bytes_pinned > 0 ||
9579 space_info->bytes_reserved > 0 ||
9580 space_info->bytes_may_use > 0)) {
9581 dump_space_info(space_info, 0, 0);
9584 list_del(&space_info->list);
9585 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9586 struct kobject *kobj;
9587 kobj = space_info->block_group_kobjs[i];
9588 space_info->block_group_kobjs[i] = NULL;
9594 kobject_del(&space_info->kobj);
9595 kobject_put(&space_info->kobj);
9600 static void __link_block_group(struct btrfs_space_info *space_info,
9601 struct btrfs_block_group_cache *cache)
9603 int index = get_block_group_index(cache);
9606 down_write(&space_info->groups_sem);
9607 if (list_empty(&space_info->block_groups[index]))
9609 list_add_tail(&cache->list, &space_info->block_groups[index]);
9610 up_write(&space_info->groups_sem);
9613 struct raid_kobject *rkobj;
9616 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9619 rkobj->raid_type = index;
9620 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9621 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9622 "%s", get_raid_name(index));
9624 kobject_put(&rkobj->kobj);
9627 space_info->block_group_kobjs[index] = &rkobj->kobj;
9632 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9635 static struct btrfs_block_group_cache *
9636 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9638 struct btrfs_block_group_cache *cache;
9640 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9644 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9646 if (!cache->free_space_ctl) {
9651 cache->key.objectid = start;
9652 cache->key.offset = size;
9653 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9655 cache->sectorsize = root->sectorsize;
9656 cache->fs_info = root->fs_info;
9657 cache->full_stripe_len = btrfs_full_stripe_len(root,
9658 &root->fs_info->mapping_tree,
9660 atomic_set(&cache->count, 1);
9661 spin_lock_init(&cache->lock);
9662 init_rwsem(&cache->data_rwsem);
9663 INIT_LIST_HEAD(&cache->list);
9664 INIT_LIST_HEAD(&cache->cluster_list);
9665 INIT_LIST_HEAD(&cache->bg_list);
9666 INIT_LIST_HEAD(&cache->ro_list);
9667 INIT_LIST_HEAD(&cache->dirty_list);
9668 INIT_LIST_HEAD(&cache->io_list);
9669 btrfs_init_free_space_ctl(cache);
9670 atomic_set(&cache->trimming, 0);
9675 int btrfs_read_block_groups(struct btrfs_root *root)
9677 struct btrfs_path *path;
9679 struct btrfs_block_group_cache *cache;
9680 struct btrfs_fs_info *info = root->fs_info;
9681 struct btrfs_space_info *space_info;
9682 struct btrfs_key key;
9683 struct btrfs_key found_key;
9684 struct extent_buffer *leaf;
9688 root = info->extent_root;
9691 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9692 path = btrfs_alloc_path();
9697 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9698 if (btrfs_test_opt(root, SPACE_CACHE) &&
9699 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9701 if (btrfs_test_opt(root, CLEAR_CACHE))
9705 ret = find_first_block_group(root, path, &key);
9711 leaf = path->nodes[0];
9712 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9714 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9723 * When we mount with old space cache, we need to
9724 * set BTRFS_DC_CLEAR and set dirty flag.
9726 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9727 * truncate the old free space cache inode and
9729 * b) Setting 'dirty flag' makes sure that we flush
9730 * the new space cache info onto disk.
9732 if (btrfs_test_opt(root, SPACE_CACHE))
9733 cache->disk_cache_state = BTRFS_DC_CLEAR;
9736 read_extent_buffer(leaf, &cache->item,
9737 btrfs_item_ptr_offset(leaf, path->slots[0]),
9738 sizeof(cache->item));
9739 cache->flags = btrfs_block_group_flags(&cache->item);
9741 key.objectid = found_key.objectid + found_key.offset;
9742 btrfs_release_path(path);
9745 * We need to exclude the super stripes now so that the space
9746 * info has super bytes accounted for, otherwise we'll think
9747 * we have more space than we actually do.
9749 ret = exclude_super_stripes(root, cache);
9752 * We may have excluded something, so call this just in
9755 free_excluded_extents(root, cache);
9756 btrfs_put_block_group(cache);
9761 * check for two cases, either we are full, and therefore
9762 * don't need to bother with the caching work since we won't
9763 * find any space, or we are empty, and we can just add all
9764 * the space in and be done with it. This saves us _alot_ of
9765 * time, particularly in the full case.
9767 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9768 cache->last_byte_to_unpin = (u64)-1;
9769 cache->cached = BTRFS_CACHE_FINISHED;
9770 free_excluded_extents(root, cache);
9771 } else if (btrfs_block_group_used(&cache->item) == 0) {
9772 cache->last_byte_to_unpin = (u64)-1;
9773 cache->cached = BTRFS_CACHE_FINISHED;
9774 add_new_free_space(cache, root->fs_info,
9776 found_key.objectid +
9778 free_excluded_extents(root, cache);
9781 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9783 btrfs_remove_free_space_cache(cache);
9784 btrfs_put_block_group(cache);
9788 ret = update_space_info(info, cache->flags, found_key.offset,
9789 btrfs_block_group_used(&cache->item),
9792 btrfs_remove_free_space_cache(cache);
9793 spin_lock(&info->block_group_cache_lock);
9794 rb_erase(&cache->cache_node,
9795 &info->block_group_cache_tree);
9796 RB_CLEAR_NODE(&cache->cache_node);
9797 spin_unlock(&info->block_group_cache_lock);
9798 btrfs_put_block_group(cache);
9802 cache->space_info = space_info;
9803 spin_lock(&cache->space_info->lock);
9804 cache->space_info->bytes_readonly += cache->bytes_super;
9805 spin_unlock(&cache->space_info->lock);
9807 __link_block_group(space_info, cache);
9809 set_avail_alloc_bits(root->fs_info, cache->flags);
9810 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9811 inc_block_group_ro(cache, 1);
9812 } else if (btrfs_block_group_used(&cache->item) == 0) {
9813 spin_lock(&info->unused_bgs_lock);
9814 /* Should always be true but just in case. */
9815 if (list_empty(&cache->bg_list)) {
9816 btrfs_get_block_group(cache);
9817 list_add_tail(&cache->bg_list,
9820 spin_unlock(&info->unused_bgs_lock);
9824 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9825 if (!(get_alloc_profile(root, space_info->flags) &
9826 (BTRFS_BLOCK_GROUP_RAID10 |
9827 BTRFS_BLOCK_GROUP_RAID1 |
9828 BTRFS_BLOCK_GROUP_RAID5 |
9829 BTRFS_BLOCK_GROUP_RAID6 |
9830 BTRFS_BLOCK_GROUP_DUP)))
9833 * avoid allocating from un-mirrored block group if there are
9834 * mirrored block groups.
9836 list_for_each_entry(cache,
9837 &space_info->block_groups[BTRFS_RAID_RAID0],
9839 inc_block_group_ro(cache, 1);
9840 list_for_each_entry(cache,
9841 &space_info->block_groups[BTRFS_RAID_SINGLE],
9843 inc_block_group_ro(cache, 1);
9846 init_global_block_rsv(info);
9849 btrfs_free_path(path);
9853 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9854 struct btrfs_root *root)
9856 struct btrfs_block_group_cache *block_group, *tmp;
9857 struct btrfs_root *extent_root = root->fs_info->extent_root;
9858 struct btrfs_block_group_item item;
9859 struct btrfs_key key;
9861 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9863 trans->can_flush_pending_bgs = false;
9864 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9868 spin_lock(&block_group->lock);
9869 memcpy(&item, &block_group->item, sizeof(item));
9870 memcpy(&key, &block_group->key, sizeof(key));
9871 spin_unlock(&block_group->lock);
9873 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9876 btrfs_abort_transaction(trans, extent_root, ret);
9877 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9878 key.objectid, key.offset);
9880 btrfs_abort_transaction(trans, extent_root, ret);
9882 list_del_init(&block_group->bg_list);
9884 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9887 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9888 struct btrfs_root *root, u64 bytes_used,
9889 u64 type, u64 chunk_objectid, u64 chunk_offset,
9893 struct btrfs_root *extent_root;
9894 struct btrfs_block_group_cache *cache;
9896 extent_root = root->fs_info->extent_root;
9898 btrfs_set_log_full_commit(root->fs_info, trans);
9900 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9904 btrfs_set_block_group_used(&cache->item, bytes_used);
9905 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9906 btrfs_set_block_group_flags(&cache->item, type);
9908 cache->flags = type;
9909 cache->last_byte_to_unpin = (u64)-1;
9910 cache->cached = BTRFS_CACHE_FINISHED;
9911 ret = exclude_super_stripes(root, cache);
9914 * We may have excluded something, so call this just in
9917 free_excluded_extents(root, cache);
9918 btrfs_put_block_group(cache);
9922 add_new_free_space(cache, root->fs_info, chunk_offset,
9923 chunk_offset + size);
9925 free_excluded_extents(root, cache);
9927 #ifdef CONFIG_BTRFS_DEBUG
9928 if (btrfs_should_fragment_free_space(root, cache)) {
9929 u64 new_bytes_used = size - bytes_used;
9931 bytes_used += new_bytes_used >> 1;
9932 fragment_free_space(root, cache);
9936 * Call to ensure the corresponding space_info object is created and
9937 * assigned to our block group, but don't update its counters just yet.
9938 * We want our bg to be added to the rbtree with its ->space_info set.
9940 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9941 &cache->space_info);
9943 btrfs_remove_free_space_cache(cache);
9944 btrfs_put_block_group(cache);
9948 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9950 btrfs_remove_free_space_cache(cache);
9951 btrfs_put_block_group(cache);
9956 * Now that our block group has its ->space_info set and is inserted in
9957 * the rbtree, update the space info's counters.
9959 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9960 &cache->space_info);
9962 btrfs_remove_free_space_cache(cache);
9963 spin_lock(&root->fs_info->block_group_cache_lock);
9964 rb_erase(&cache->cache_node,
9965 &root->fs_info->block_group_cache_tree);
9966 RB_CLEAR_NODE(&cache->cache_node);
9967 spin_unlock(&root->fs_info->block_group_cache_lock);
9968 btrfs_put_block_group(cache);
9971 update_global_block_rsv(root->fs_info);
9973 spin_lock(&cache->space_info->lock);
9974 cache->space_info->bytes_readonly += cache->bytes_super;
9975 spin_unlock(&cache->space_info->lock);
9977 __link_block_group(cache->space_info, cache);
9979 list_add_tail(&cache->bg_list, &trans->new_bgs);
9981 set_avail_alloc_bits(extent_root->fs_info, type);
9986 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9988 u64 extra_flags = chunk_to_extended(flags) &
9989 BTRFS_EXTENDED_PROFILE_MASK;
9991 write_seqlock(&fs_info->profiles_lock);
9992 if (flags & BTRFS_BLOCK_GROUP_DATA)
9993 fs_info->avail_data_alloc_bits &= ~extra_flags;
9994 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9995 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9996 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9997 fs_info->avail_system_alloc_bits &= ~extra_flags;
9998 write_sequnlock(&fs_info->profiles_lock);
10001 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10002 struct btrfs_root *root, u64 group_start,
10003 struct extent_map *em)
10005 struct btrfs_path *path;
10006 struct btrfs_block_group_cache *block_group;
10007 struct btrfs_free_cluster *cluster;
10008 struct btrfs_root *tree_root = root->fs_info->tree_root;
10009 struct btrfs_key key;
10010 struct inode *inode;
10011 struct kobject *kobj = NULL;
10015 struct btrfs_caching_control *caching_ctl = NULL;
10018 root = root->fs_info->extent_root;
10020 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10021 BUG_ON(!block_group);
10022 BUG_ON(!block_group->ro);
10025 * Free the reserved super bytes from this block group before
10028 free_excluded_extents(root, block_group);
10030 memcpy(&key, &block_group->key, sizeof(key));
10031 index = get_block_group_index(block_group);
10032 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10033 BTRFS_BLOCK_GROUP_RAID1 |
10034 BTRFS_BLOCK_GROUP_RAID10))
10039 /* make sure this block group isn't part of an allocation cluster */
10040 cluster = &root->fs_info->data_alloc_cluster;
10041 spin_lock(&cluster->refill_lock);
10042 btrfs_return_cluster_to_free_space(block_group, cluster);
10043 spin_unlock(&cluster->refill_lock);
10046 * make sure this block group isn't part of a metadata
10047 * allocation cluster
10049 cluster = &root->fs_info->meta_alloc_cluster;
10050 spin_lock(&cluster->refill_lock);
10051 btrfs_return_cluster_to_free_space(block_group, cluster);
10052 spin_unlock(&cluster->refill_lock);
10054 path = btrfs_alloc_path();
10061 * get the inode first so any iput calls done for the io_list
10062 * aren't the final iput (no unlinks allowed now)
10064 inode = lookup_free_space_inode(tree_root, block_group, path);
10066 mutex_lock(&trans->transaction->cache_write_mutex);
10068 * make sure our free spache cache IO is done before remove the
10071 spin_lock(&trans->transaction->dirty_bgs_lock);
10072 if (!list_empty(&block_group->io_list)) {
10073 list_del_init(&block_group->io_list);
10075 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10077 spin_unlock(&trans->transaction->dirty_bgs_lock);
10078 btrfs_wait_cache_io(root, trans, block_group,
10079 &block_group->io_ctl, path,
10080 block_group->key.objectid);
10081 btrfs_put_block_group(block_group);
10082 spin_lock(&trans->transaction->dirty_bgs_lock);
10085 if (!list_empty(&block_group->dirty_list)) {
10086 list_del_init(&block_group->dirty_list);
10087 btrfs_put_block_group(block_group);
10089 spin_unlock(&trans->transaction->dirty_bgs_lock);
10090 mutex_unlock(&trans->transaction->cache_write_mutex);
10092 if (!IS_ERR(inode)) {
10093 ret = btrfs_orphan_add(trans, inode);
10095 btrfs_add_delayed_iput(inode);
10098 clear_nlink(inode);
10099 /* One for the block groups ref */
10100 spin_lock(&block_group->lock);
10101 if (block_group->iref) {
10102 block_group->iref = 0;
10103 block_group->inode = NULL;
10104 spin_unlock(&block_group->lock);
10107 spin_unlock(&block_group->lock);
10109 /* One for our lookup ref */
10110 btrfs_add_delayed_iput(inode);
10113 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10114 key.offset = block_group->key.objectid;
10117 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10121 btrfs_release_path(path);
10123 ret = btrfs_del_item(trans, tree_root, path);
10126 btrfs_release_path(path);
10129 spin_lock(&root->fs_info->block_group_cache_lock);
10130 rb_erase(&block_group->cache_node,
10131 &root->fs_info->block_group_cache_tree);
10132 RB_CLEAR_NODE(&block_group->cache_node);
10134 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10135 root->fs_info->first_logical_byte = (u64)-1;
10136 spin_unlock(&root->fs_info->block_group_cache_lock);
10138 down_write(&block_group->space_info->groups_sem);
10140 * we must use list_del_init so people can check to see if they
10141 * are still on the list after taking the semaphore
10143 list_del_init(&block_group->list);
10144 if (list_empty(&block_group->space_info->block_groups[index])) {
10145 kobj = block_group->space_info->block_group_kobjs[index];
10146 block_group->space_info->block_group_kobjs[index] = NULL;
10147 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10149 up_write(&block_group->space_info->groups_sem);
10155 if (block_group->has_caching_ctl)
10156 caching_ctl = get_caching_control(block_group);
10157 if (block_group->cached == BTRFS_CACHE_STARTED)
10158 wait_block_group_cache_done(block_group);
10159 if (block_group->has_caching_ctl) {
10160 down_write(&root->fs_info->commit_root_sem);
10161 if (!caching_ctl) {
10162 struct btrfs_caching_control *ctl;
10164 list_for_each_entry(ctl,
10165 &root->fs_info->caching_block_groups, list)
10166 if (ctl->block_group == block_group) {
10168 atomic_inc(&caching_ctl->count);
10173 list_del_init(&caching_ctl->list);
10174 up_write(&root->fs_info->commit_root_sem);
10176 /* Once for the caching bgs list and once for us. */
10177 put_caching_control(caching_ctl);
10178 put_caching_control(caching_ctl);
10182 spin_lock(&trans->transaction->dirty_bgs_lock);
10183 if (!list_empty(&block_group->dirty_list)) {
10186 if (!list_empty(&block_group->io_list)) {
10189 spin_unlock(&trans->transaction->dirty_bgs_lock);
10190 btrfs_remove_free_space_cache(block_group);
10192 spin_lock(&block_group->space_info->lock);
10193 list_del_init(&block_group->ro_list);
10195 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10196 WARN_ON(block_group->space_info->total_bytes
10197 < block_group->key.offset);
10198 WARN_ON(block_group->space_info->bytes_readonly
10199 < block_group->key.offset);
10200 WARN_ON(block_group->space_info->disk_total
10201 < block_group->key.offset * factor);
10203 block_group->space_info->total_bytes -= block_group->key.offset;
10204 block_group->space_info->bytes_readonly -= block_group->key.offset;
10205 block_group->space_info->disk_total -= block_group->key.offset * factor;
10207 spin_unlock(&block_group->space_info->lock);
10209 memcpy(&key, &block_group->key, sizeof(key));
10212 if (!list_empty(&em->list)) {
10213 /* We're in the transaction->pending_chunks list. */
10214 free_extent_map(em);
10216 spin_lock(&block_group->lock);
10217 block_group->removed = 1;
10219 * At this point trimming can't start on this block group, because we
10220 * removed the block group from the tree fs_info->block_group_cache_tree
10221 * so no one can't find it anymore and even if someone already got this
10222 * block group before we removed it from the rbtree, they have already
10223 * incremented block_group->trimming - if they didn't, they won't find
10224 * any free space entries because we already removed them all when we
10225 * called btrfs_remove_free_space_cache().
10227 * And we must not remove the extent map from the fs_info->mapping_tree
10228 * to prevent the same logical address range and physical device space
10229 * ranges from being reused for a new block group. This is because our
10230 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10231 * completely transactionless, so while it is trimming a range the
10232 * currently running transaction might finish and a new one start,
10233 * allowing for new block groups to be created that can reuse the same
10234 * physical device locations unless we take this special care.
10236 * There may also be an implicit trim operation if the file system
10237 * is mounted with -odiscard. The same protections must remain
10238 * in place until the extents have been discarded completely when
10239 * the transaction commit has completed.
10241 remove_em = (atomic_read(&block_group->trimming) == 0);
10243 * Make sure a trimmer task always sees the em in the pinned_chunks list
10244 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10245 * before checking block_group->removed).
10249 * Our em might be in trans->transaction->pending_chunks which
10250 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10251 * and so is the fs_info->pinned_chunks list.
10253 * So at this point we must be holding the chunk_mutex to avoid
10254 * any races with chunk allocation (more specifically at
10255 * volumes.c:contains_pending_extent()), to ensure it always
10256 * sees the em, either in the pending_chunks list or in the
10257 * pinned_chunks list.
10259 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10261 spin_unlock(&block_group->lock);
10264 struct extent_map_tree *em_tree;
10266 em_tree = &root->fs_info->mapping_tree.map_tree;
10267 write_lock(&em_tree->lock);
10269 * The em might be in the pending_chunks list, so make sure the
10270 * chunk mutex is locked, since remove_extent_mapping() will
10271 * delete us from that list.
10273 remove_extent_mapping(em_tree, em);
10274 write_unlock(&em_tree->lock);
10275 /* once for the tree */
10276 free_extent_map(em);
10279 unlock_chunks(root);
10281 btrfs_put_block_group(block_group);
10282 btrfs_put_block_group(block_group);
10284 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10290 ret = btrfs_del_item(trans, root, path);
10292 btrfs_free_path(path);
10296 struct btrfs_trans_handle *
10297 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10298 const u64 chunk_offset)
10300 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10301 struct extent_map *em;
10302 struct map_lookup *map;
10303 unsigned int num_items;
10305 read_lock(&em_tree->lock);
10306 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10307 read_unlock(&em_tree->lock);
10308 ASSERT(em && em->start == chunk_offset);
10311 * We need to reserve 3 + N units from the metadata space info in order
10312 * to remove a block group (done at btrfs_remove_chunk() and at
10313 * btrfs_remove_block_group()), which are used for:
10315 * 1 unit for adding the free space inode's orphan (located in the tree
10317 * 1 unit for deleting the block group item (located in the extent
10319 * 1 unit for deleting the free space item (located in tree of tree
10321 * N units for deleting N device extent items corresponding to each
10322 * stripe (located in the device tree).
10324 * In order to remove a block group we also need to reserve units in the
10325 * system space info in order to update the chunk tree (update one or
10326 * more device items and remove one chunk item), but this is done at
10327 * btrfs_remove_chunk() through a call to check_system_chunk().
10329 map = (struct map_lookup *)em->bdev;
10330 num_items = 3 + map->num_stripes;
10331 free_extent_map(em);
10333 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10338 * Process the unused_bgs list and remove any that don't have any allocated
10339 * space inside of them.
10341 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10343 struct btrfs_block_group_cache *block_group;
10344 struct btrfs_space_info *space_info;
10345 struct btrfs_root *root = fs_info->extent_root;
10346 struct btrfs_trans_handle *trans;
10349 if (!fs_info->open)
10352 spin_lock(&fs_info->unused_bgs_lock);
10353 while (!list_empty(&fs_info->unused_bgs)) {
10357 block_group = list_first_entry(&fs_info->unused_bgs,
10358 struct btrfs_block_group_cache,
10360 list_del_init(&block_group->bg_list);
10362 space_info = block_group->space_info;
10364 if (ret || btrfs_mixed_space_info(space_info)) {
10365 btrfs_put_block_group(block_group);
10368 spin_unlock(&fs_info->unused_bgs_lock);
10370 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10372 /* Don't want to race with allocators so take the groups_sem */
10373 down_write(&space_info->groups_sem);
10374 spin_lock(&block_group->lock);
10375 if (block_group->reserved ||
10376 btrfs_block_group_used(&block_group->item) ||
10378 list_is_singular(&block_group->list)) {
10380 * We want to bail if we made new allocations or have
10381 * outstanding allocations in this block group. We do
10382 * the ro check in case balance is currently acting on
10383 * this block group.
10385 spin_unlock(&block_group->lock);
10386 up_write(&space_info->groups_sem);
10389 spin_unlock(&block_group->lock);
10391 /* We don't want to force the issue, only flip if it's ok. */
10392 ret = inc_block_group_ro(block_group, 0);
10393 up_write(&space_info->groups_sem);
10400 * Want to do this before we do anything else so we can recover
10401 * properly if we fail to join the transaction.
10403 trans = btrfs_start_trans_remove_block_group(fs_info,
10404 block_group->key.objectid);
10405 if (IS_ERR(trans)) {
10406 btrfs_dec_block_group_ro(root, block_group);
10407 ret = PTR_ERR(trans);
10412 * We could have pending pinned extents for this block group,
10413 * just delete them, we don't care about them anymore.
10415 start = block_group->key.objectid;
10416 end = start + block_group->key.offset - 1;
10418 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10419 * btrfs_finish_extent_commit(). If we are at transaction N,
10420 * another task might be running finish_extent_commit() for the
10421 * previous transaction N - 1, and have seen a range belonging
10422 * to the block group in freed_extents[] before we were able to
10423 * clear the whole block group range from freed_extents[]. This
10424 * means that task can lookup for the block group after we
10425 * unpinned it from freed_extents[] and removed it, leading to
10426 * a BUG_ON() at btrfs_unpin_extent_range().
10428 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10429 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10430 EXTENT_DIRTY, GFP_NOFS);
10432 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10433 btrfs_dec_block_group_ro(root, block_group);
10436 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10437 EXTENT_DIRTY, GFP_NOFS);
10439 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10440 btrfs_dec_block_group_ro(root, block_group);
10443 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10445 /* Reset pinned so btrfs_put_block_group doesn't complain */
10446 spin_lock(&space_info->lock);
10447 spin_lock(&block_group->lock);
10449 space_info->bytes_pinned -= block_group->pinned;
10450 space_info->bytes_readonly += block_group->pinned;
10451 percpu_counter_add(&space_info->total_bytes_pinned,
10452 -block_group->pinned);
10453 block_group->pinned = 0;
10455 spin_unlock(&block_group->lock);
10456 spin_unlock(&space_info->lock);
10458 /* DISCARD can flip during remount */
10459 trimming = btrfs_test_opt(root, DISCARD);
10461 /* Implicit trim during transaction commit. */
10463 btrfs_get_block_group_trimming(block_group);
10466 * Btrfs_remove_chunk will abort the transaction if things go
10469 ret = btrfs_remove_chunk(trans, root,
10470 block_group->key.objectid);
10474 btrfs_put_block_group_trimming(block_group);
10479 * If we're not mounted with -odiscard, we can just forget
10480 * about this block group. Otherwise we'll need to wait
10481 * until transaction commit to do the actual discard.
10484 spin_lock(&fs_info->unused_bgs_lock);
10486 * A concurrent scrub might have added us to the list
10487 * fs_info->unused_bgs, so use a list_move operation
10488 * to add the block group to the deleted_bgs list.
10490 list_move(&block_group->bg_list,
10491 &trans->transaction->deleted_bgs);
10492 spin_unlock(&fs_info->unused_bgs_lock);
10493 btrfs_get_block_group(block_group);
10496 btrfs_end_transaction(trans, root);
10498 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10499 btrfs_put_block_group(block_group);
10500 spin_lock(&fs_info->unused_bgs_lock);
10502 spin_unlock(&fs_info->unused_bgs_lock);
10505 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10507 struct btrfs_space_info *space_info;
10508 struct btrfs_super_block *disk_super;
10514 disk_super = fs_info->super_copy;
10515 if (!btrfs_super_root(disk_super))
10518 features = btrfs_super_incompat_flags(disk_super);
10519 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10522 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10523 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10528 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10529 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10531 flags = BTRFS_BLOCK_GROUP_METADATA;
10532 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10536 flags = BTRFS_BLOCK_GROUP_DATA;
10537 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10543 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10545 return unpin_extent_range(root, start, end, false);
10549 * It used to be that old block groups would be left around forever.
10550 * Iterating over them would be enough to trim unused space. Since we
10551 * now automatically remove them, we also need to iterate over unallocated
10554 * We don't want a transaction for this since the discard may take a
10555 * substantial amount of time. We don't require that a transaction be
10556 * running, but we do need to take a running transaction into account
10557 * to ensure that we're not discarding chunks that were released in
10558 * the current transaction.
10560 * Holding the chunks lock will prevent other threads from allocating
10561 * or releasing chunks, but it won't prevent a running transaction
10562 * from committing and releasing the memory that the pending chunks
10563 * list head uses. For that, we need to take a reference to the
10566 static int btrfs_trim_free_extents(struct btrfs_device *device,
10567 u64 minlen, u64 *trimmed)
10569 u64 start = 0, len = 0;
10574 /* Not writeable = nothing to do. */
10575 if (!device->writeable)
10578 /* No free space = nothing to do. */
10579 if (device->total_bytes <= device->bytes_used)
10585 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10586 struct btrfs_transaction *trans;
10589 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10593 down_read(&fs_info->commit_root_sem);
10595 spin_lock(&fs_info->trans_lock);
10596 trans = fs_info->running_transaction;
10598 atomic_inc(&trans->use_count);
10599 spin_unlock(&fs_info->trans_lock);
10601 ret = find_free_dev_extent_start(trans, device, minlen, start,
10604 btrfs_put_transaction(trans);
10607 up_read(&fs_info->commit_root_sem);
10608 mutex_unlock(&fs_info->chunk_mutex);
10609 if (ret == -ENOSPC)
10614 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10615 up_read(&fs_info->commit_root_sem);
10616 mutex_unlock(&fs_info->chunk_mutex);
10624 if (fatal_signal_pending(current)) {
10625 ret = -ERESTARTSYS;
10635 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10637 struct btrfs_fs_info *fs_info = root->fs_info;
10638 struct btrfs_block_group_cache *cache = NULL;
10639 struct btrfs_device *device;
10640 struct list_head *devices;
10645 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10649 * try to trim all FS space, our block group may start from non-zero.
10651 if (range->len == total_bytes)
10652 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10654 cache = btrfs_lookup_block_group(fs_info, range->start);
10657 if (cache->key.objectid >= (range->start + range->len)) {
10658 btrfs_put_block_group(cache);
10662 start = max(range->start, cache->key.objectid);
10663 end = min(range->start + range->len,
10664 cache->key.objectid + cache->key.offset);
10666 if (end - start >= range->minlen) {
10667 if (!block_group_cache_done(cache)) {
10668 ret = cache_block_group(cache, 0);
10670 btrfs_put_block_group(cache);
10673 ret = wait_block_group_cache_done(cache);
10675 btrfs_put_block_group(cache);
10679 ret = btrfs_trim_block_group(cache,
10685 trimmed += group_trimmed;
10687 btrfs_put_block_group(cache);
10692 cache = next_block_group(fs_info->tree_root, cache);
10695 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10696 devices = &root->fs_info->fs_devices->alloc_list;
10697 list_for_each_entry(device, devices, dev_alloc_list) {
10698 ret = btrfs_trim_free_extents(device, range->minlen,
10703 trimmed += group_trimmed;
10705 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10707 range->len = trimmed;
10712 * btrfs_{start,end}_write_no_snapshoting() are similar to
10713 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10714 * data into the page cache through nocow before the subvolume is snapshoted,
10715 * but flush the data into disk after the snapshot creation, or to prevent
10716 * operations while snapshoting is ongoing and that cause the snapshot to be
10717 * inconsistent (writes followed by expanding truncates for example).
10719 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10721 percpu_counter_dec(&root->subv_writers->counter);
10723 * Make sure counter is updated before we wake up waiters.
10726 if (waitqueue_active(&root->subv_writers->wait))
10727 wake_up(&root->subv_writers->wait);
10730 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10732 if (atomic_read(&root->will_be_snapshoted))
10735 percpu_counter_inc(&root->subv_writers->counter);
10737 * Make sure counter is updated before we check for snapshot creation.
10740 if (atomic_read(&root->will_be_snapshoted)) {
10741 btrfs_end_write_no_snapshoting(root);