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_start_delalloc_roots(fs_info, 0, -1);
4091 btrfs_wait_ordered_roots(fs_info, -1);
4094 trans = btrfs_join_transaction(root);
4096 return PTR_ERR(trans);
4097 if (have_pinned_space >= 0 ||
4098 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4099 &trans->transaction->flags) ||
4101 ret = btrfs_commit_transaction(trans, root);
4105 * The cleaner kthread might still be doing iput
4106 * operations. Wait for it to finish so that
4107 * more space is released.
4109 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
4110 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
4113 btrfs_end_transaction(trans, root);
4117 trace_btrfs_space_reservation(root->fs_info,
4118 "space_info:enospc",
4119 data_sinfo->flags, bytes, 1);
4122 data_sinfo->bytes_may_use += bytes;
4123 trace_btrfs_space_reservation(root->fs_info, "space_info",
4124 data_sinfo->flags, bytes, 1);
4125 spin_unlock(&data_sinfo->lock);
4131 * New check_data_free_space() with ability for precious data reservation
4132 * Will replace old btrfs_check_data_free_space(), but for patch split,
4133 * add a new function first and then replace it.
4135 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4137 struct btrfs_root *root = BTRFS_I(inode)->root;
4140 /* align the range */
4141 len = round_up(start + len, root->sectorsize) -
4142 round_down(start, root->sectorsize);
4143 start = round_down(start, root->sectorsize);
4145 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4150 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4152 * TODO: Find a good method to avoid reserve data space for NOCOW
4153 * range, but don't impact performance on quota disable case.
4155 ret = btrfs_qgroup_reserve_data(inode, start, len);
4160 * Called if we need to clear a data reservation for this inode
4161 * Normally in a error case.
4163 * This one will *NOT* use accurate qgroup reserved space API, just for case
4164 * which we can't sleep and is sure it won't affect qgroup reserved space.
4165 * Like clear_bit_hook().
4167 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4170 struct btrfs_root *root = BTRFS_I(inode)->root;
4171 struct btrfs_space_info *data_sinfo;
4173 /* Make sure the range is aligned to sectorsize */
4174 len = round_up(start + len, root->sectorsize) -
4175 round_down(start, root->sectorsize);
4176 start = round_down(start, root->sectorsize);
4178 data_sinfo = root->fs_info->data_sinfo;
4179 spin_lock(&data_sinfo->lock);
4180 if (WARN_ON(data_sinfo->bytes_may_use < len))
4181 data_sinfo->bytes_may_use = 0;
4183 data_sinfo->bytes_may_use -= len;
4184 trace_btrfs_space_reservation(root->fs_info, "space_info",
4185 data_sinfo->flags, len, 0);
4186 spin_unlock(&data_sinfo->lock);
4190 * Called if we need to clear a data reservation for this inode
4191 * Normally in a error case.
4193 * This one will handle the per-indoe data rsv map for accurate reserved
4196 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4198 btrfs_free_reserved_data_space_noquota(inode, start, len);
4199 btrfs_qgroup_free_data(inode, start, len);
4202 static void force_metadata_allocation(struct btrfs_fs_info *info)
4204 struct list_head *head = &info->space_info;
4205 struct btrfs_space_info *found;
4208 list_for_each_entry_rcu(found, head, list) {
4209 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4210 found->force_alloc = CHUNK_ALLOC_FORCE;
4215 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4217 return (global->size << 1);
4220 static int should_alloc_chunk(struct btrfs_root *root,
4221 struct btrfs_space_info *sinfo, int force)
4223 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4224 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4225 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4228 if (force == CHUNK_ALLOC_FORCE)
4232 * We need to take into account the global rsv because for all intents
4233 * and purposes it's used space. Don't worry about locking the
4234 * global_rsv, it doesn't change except when the transaction commits.
4236 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4237 num_allocated += calc_global_rsv_need_space(global_rsv);
4240 * in limited mode, we want to have some free space up to
4241 * about 1% of the FS size.
4243 if (force == CHUNK_ALLOC_LIMITED) {
4244 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4245 thresh = max_t(u64, 64 * 1024 * 1024,
4246 div_factor_fine(thresh, 1));
4248 if (num_bytes - num_allocated < thresh)
4252 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4257 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4261 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4262 BTRFS_BLOCK_GROUP_RAID0 |
4263 BTRFS_BLOCK_GROUP_RAID5 |
4264 BTRFS_BLOCK_GROUP_RAID6))
4265 num_dev = root->fs_info->fs_devices->rw_devices;
4266 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4269 num_dev = 1; /* DUP or single */
4275 * If @is_allocation is true, reserve space in the system space info necessary
4276 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4279 void check_system_chunk(struct btrfs_trans_handle *trans,
4280 struct btrfs_root *root,
4283 struct btrfs_space_info *info;
4290 * Needed because we can end up allocating a system chunk and for an
4291 * atomic and race free space reservation in the chunk block reserve.
4293 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4295 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4296 spin_lock(&info->lock);
4297 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4298 info->bytes_reserved - info->bytes_readonly -
4299 info->bytes_may_use;
4300 spin_unlock(&info->lock);
4302 num_devs = get_profile_num_devs(root, type);
4304 /* num_devs device items to update and 1 chunk item to add or remove */
4305 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4306 btrfs_calc_trans_metadata_size(root, 1);
4308 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4309 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4310 left, thresh, type);
4311 dump_space_info(info, 0, 0);
4314 if (left < thresh) {
4317 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4319 * Ignore failure to create system chunk. We might end up not
4320 * needing it, as we might not need to COW all nodes/leafs from
4321 * the paths we visit in the chunk tree (they were already COWed
4322 * or created in the current transaction for example).
4324 ret = btrfs_alloc_chunk(trans, root, flags);
4328 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4329 &root->fs_info->chunk_block_rsv,
4330 thresh, BTRFS_RESERVE_NO_FLUSH);
4332 trans->chunk_bytes_reserved += thresh;
4336 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4337 struct btrfs_root *extent_root, u64 flags, int force)
4339 struct btrfs_space_info *space_info;
4340 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4341 int wait_for_alloc = 0;
4344 /* Don't re-enter if we're already allocating a chunk */
4345 if (trans->allocating_chunk)
4348 space_info = __find_space_info(extent_root->fs_info, flags);
4350 ret = update_space_info(extent_root->fs_info, flags,
4352 BUG_ON(ret); /* -ENOMEM */
4354 BUG_ON(!space_info); /* Logic error */
4357 spin_lock(&space_info->lock);
4358 if (force < space_info->force_alloc)
4359 force = space_info->force_alloc;
4360 if (space_info->full) {
4361 if (should_alloc_chunk(extent_root, space_info, force))
4365 spin_unlock(&space_info->lock);
4369 if (!should_alloc_chunk(extent_root, space_info, force)) {
4370 spin_unlock(&space_info->lock);
4372 } else if (space_info->chunk_alloc) {
4375 space_info->chunk_alloc = 1;
4378 spin_unlock(&space_info->lock);
4380 mutex_lock(&fs_info->chunk_mutex);
4383 * The chunk_mutex is held throughout the entirety of a chunk
4384 * allocation, so once we've acquired the chunk_mutex we know that the
4385 * other guy is done and we need to recheck and see if we should
4388 if (wait_for_alloc) {
4389 mutex_unlock(&fs_info->chunk_mutex);
4394 trans->allocating_chunk = true;
4397 * If we have mixed data/metadata chunks we want to make sure we keep
4398 * allocating mixed chunks instead of individual chunks.
4400 if (btrfs_mixed_space_info(space_info))
4401 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4404 * if we're doing a data chunk, go ahead and make sure that
4405 * we keep a reasonable number of metadata chunks allocated in the
4408 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4409 fs_info->data_chunk_allocations++;
4410 if (!(fs_info->data_chunk_allocations %
4411 fs_info->metadata_ratio))
4412 force_metadata_allocation(fs_info);
4416 * Check if we have enough space in SYSTEM chunk because we may need
4417 * to update devices.
4419 check_system_chunk(trans, extent_root, flags);
4421 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4422 trans->allocating_chunk = false;
4424 spin_lock(&space_info->lock);
4425 if (ret < 0 && ret != -ENOSPC)
4428 space_info->full = 1;
4432 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4434 space_info->chunk_alloc = 0;
4435 spin_unlock(&space_info->lock);
4436 mutex_unlock(&fs_info->chunk_mutex);
4438 * When we allocate a new chunk we reserve space in the chunk block
4439 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4440 * add new nodes/leafs to it if we end up needing to do it when
4441 * inserting the chunk item and updating device items as part of the
4442 * second phase of chunk allocation, performed by
4443 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4444 * large number of new block groups to create in our transaction
4445 * handle's new_bgs list to avoid exhausting the chunk block reserve
4446 * in extreme cases - like having a single transaction create many new
4447 * block groups when starting to write out the free space caches of all
4448 * the block groups that were made dirty during the lifetime of the
4451 if (trans->can_flush_pending_bgs &&
4452 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4453 btrfs_create_pending_block_groups(trans, trans->root);
4454 btrfs_trans_release_chunk_metadata(trans);
4459 static int can_overcommit(struct btrfs_root *root,
4460 struct btrfs_space_info *space_info, u64 bytes,
4461 enum btrfs_reserve_flush_enum flush)
4463 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4464 u64 profile = btrfs_get_alloc_profile(root, 0);
4469 used = space_info->bytes_used + space_info->bytes_reserved +
4470 space_info->bytes_pinned + space_info->bytes_readonly;
4473 * We only want to allow over committing if we have lots of actual space
4474 * free, but if we don't have enough space to handle the global reserve
4475 * space then we could end up having a real enospc problem when trying
4476 * to allocate a chunk or some other such important allocation.
4478 spin_lock(&global_rsv->lock);
4479 space_size = calc_global_rsv_need_space(global_rsv);
4480 spin_unlock(&global_rsv->lock);
4481 if (used + space_size >= space_info->total_bytes)
4484 used += space_info->bytes_may_use;
4486 spin_lock(&root->fs_info->free_chunk_lock);
4487 avail = root->fs_info->free_chunk_space;
4488 spin_unlock(&root->fs_info->free_chunk_lock);
4491 * If we have dup, raid1 or raid10 then only half of the free
4492 * space is actually useable. For raid56, the space info used
4493 * doesn't include the parity drive, so we don't have to
4496 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4497 BTRFS_BLOCK_GROUP_RAID1 |
4498 BTRFS_BLOCK_GROUP_RAID10))
4502 * If we aren't flushing all things, let us overcommit up to
4503 * 1/2th of the space. If we can flush, don't let us overcommit
4504 * too much, let it overcommit up to 1/8 of the space.
4506 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4511 if (used + bytes < space_info->total_bytes + avail)
4516 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4517 unsigned long nr_pages, int nr_items)
4519 struct super_block *sb = root->fs_info->sb;
4521 if (down_read_trylock(&sb->s_umount)) {
4522 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4523 up_read(&sb->s_umount);
4526 * We needn't worry the filesystem going from r/w to r/o though
4527 * we don't acquire ->s_umount mutex, because the filesystem
4528 * should guarantee the delalloc inodes list be empty after
4529 * the filesystem is readonly(all dirty pages are written to
4532 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4533 if (!current->journal_info)
4534 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4538 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4543 bytes = btrfs_calc_trans_metadata_size(root, 1);
4544 nr = (int)div64_u64(to_reclaim, bytes);
4550 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4553 * shrink metadata reservation for delalloc
4555 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4558 struct btrfs_block_rsv *block_rsv;
4559 struct btrfs_space_info *space_info;
4560 struct btrfs_trans_handle *trans;
4564 unsigned long nr_pages;
4567 enum btrfs_reserve_flush_enum flush;
4569 /* Calc the number of the pages we need flush for space reservation */
4570 items = calc_reclaim_items_nr(root, to_reclaim);
4571 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4573 trans = (struct btrfs_trans_handle *)current->journal_info;
4574 block_rsv = &root->fs_info->delalloc_block_rsv;
4575 space_info = block_rsv->space_info;
4577 delalloc_bytes = percpu_counter_sum_positive(
4578 &root->fs_info->delalloc_bytes);
4579 if (delalloc_bytes == 0) {
4583 btrfs_wait_ordered_roots(root->fs_info, items);
4588 while (delalloc_bytes && loops < 3) {
4589 max_reclaim = min(delalloc_bytes, to_reclaim);
4590 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4591 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4593 * We need to wait for the async pages to actually start before
4596 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4600 if (max_reclaim <= nr_pages)
4603 max_reclaim -= nr_pages;
4605 wait_event(root->fs_info->async_submit_wait,
4606 atomic_read(&root->fs_info->async_delalloc_pages) <=
4610 flush = BTRFS_RESERVE_FLUSH_ALL;
4612 flush = BTRFS_RESERVE_NO_FLUSH;
4613 spin_lock(&space_info->lock);
4614 if (can_overcommit(root, space_info, orig, flush)) {
4615 spin_unlock(&space_info->lock);
4618 spin_unlock(&space_info->lock);
4621 if (wait_ordered && !trans) {
4622 btrfs_wait_ordered_roots(root->fs_info, items);
4624 time_left = schedule_timeout_killable(1);
4628 delalloc_bytes = percpu_counter_sum_positive(
4629 &root->fs_info->delalloc_bytes);
4634 * maybe_commit_transaction - possibly commit the transaction if its ok to
4635 * @root - the root we're allocating for
4636 * @bytes - the number of bytes we want to reserve
4637 * @force - force the commit
4639 * This will check to make sure that committing the transaction will actually
4640 * get us somewhere and then commit the transaction if it does. Otherwise it
4641 * will return -ENOSPC.
4643 static int may_commit_transaction(struct btrfs_root *root,
4644 struct btrfs_space_info *space_info,
4645 u64 bytes, int force)
4647 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4648 struct btrfs_trans_handle *trans;
4650 trans = (struct btrfs_trans_handle *)current->journal_info;
4657 /* See if there is enough pinned space to make this reservation */
4658 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4663 * See if there is some space in the delayed insertion reservation for
4666 if (space_info != delayed_rsv->space_info)
4669 spin_lock(&delayed_rsv->lock);
4670 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4671 bytes - delayed_rsv->size) >= 0) {
4672 spin_unlock(&delayed_rsv->lock);
4675 spin_unlock(&delayed_rsv->lock);
4678 trans = btrfs_join_transaction(root);
4682 return btrfs_commit_transaction(trans, root);
4686 FLUSH_DELAYED_ITEMS_NR = 1,
4687 FLUSH_DELAYED_ITEMS = 2,
4689 FLUSH_DELALLOC_WAIT = 4,
4694 static int flush_space(struct btrfs_root *root,
4695 struct btrfs_space_info *space_info, u64 num_bytes,
4696 u64 orig_bytes, int state)
4698 struct btrfs_trans_handle *trans;
4703 case FLUSH_DELAYED_ITEMS_NR:
4704 case FLUSH_DELAYED_ITEMS:
4705 if (state == FLUSH_DELAYED_ITEMS_NR)
4706 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4710 trans = btrfs_join_transaction(root);
4711 if (IS_ERR(trans)) {
4712 ret = PTR_ERR(trans);
4715 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4716 btrfs_end_transaction(trans, root);
4718 case FLUSH_DELALLOC:
4719 case FLUSH_DELALLOC_WAIT:
4720 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4721 state == FLUSH_DELALLOC_WAIT);
4724 trans = btrfs_join_transaction(root);
4725 if (IS_ERR(trans)) {
4726 ret = PTR_ERR(trans);
4729 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4730 btrfs_get_alloc_profile(root, 0),
4731 CHUNK_ALLOC_NO_FORCE);
4732 btrfs_end_transaction(trans, root);
4737 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4748 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4749 struct btrfs_space_info *space_info)
4755 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4757 spin_lock(&space_info->lock);
4758 if (can_overcommit(root, space_info, to_reclaim,
4759 BTRFS_RESERVE_FLUSH_ALL)) {
4764 used = space_info->bytes_used + space_info->bytes_reserved +
4765 space_info->bytes_pinned + space_info->bytes_readonly +
4766 space_info->bytes_may_use;
4767 if (can_overcommit(root, space_info, 1024 * 1024,
4768 BTRFS_RESERVE_FLUSH_ALL))
4769 expected = div_factor_fine(space_info->total_bytes, 95);
4771 expected = div_factor_fine(space_info->total_bytes, 90);
4773 if (used > expected)
4774 to_reclaim = used - expected;
4777 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4778 space_info->bytes_reserved);
4780 spin_unlock(&space_info->lock);
4785 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4786 struct btrfs_fs_info *fs_info, u64 used)
4788 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4790 /* If we're just plain full then async reclaim just slows us down. */
4791 if (space_info->bytes_used >= thresh)
4794 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4795 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4798 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4799 struct btrfs_fs_info *fs_info,
4804 spin_lock(&space_info->lock);
4806 * We run out of space and have not got any free space via flush_space,
4807 * so don't bother doing async reclaim.
4809 if (flush_state > COMMIT_TRANS && space_info->full) {
4810 spin_unlock(&space_info->lock);
4814 used = space_info->bytes_used + space_info->bytes_reserved +
4815 space_info->bytes_pinned + space_info->bytes_readonly +
4816 space_info->bytes_may_use;
4817 if (need_do_async_reclaim(space_info, fs_info, used)) {
4818 spin_unlock(&space_info->lock);
4821 spin_unlock(&space_info->lock);
4826 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4828 struct btrfs_fs_info *fs_info;
4829 struct btrfs_space_info *space_info;
4833 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4834 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4836 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4841 flush_state = FLUSH_DELAYED_ITEMS_NR;
4843 flush_space(fs_info->fs_root, space_info, to_reclaim,
4844 to_reclaim, flush_state);
4846 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4849 } while (flush_state < COMMIT_TRANS);
4852 void btrfs_init_async_reclaim_work(struct work_struct *work)
4854 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4858 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4859 * @root - the root we're allocating for
4860 * @block_rsv - the block_rsv we're allocating for
4861 * @orig_bytes - the number of bytes we want
4862 * @flush - whether or not we can flush to make our reservation
4864 * This will reserve orgi_bytes number of bytes from the space info associated
4865 * with the block_rsv. If there is not enough space it will make an attempt to
4866 * flush out space to make room. It will do this by flushing delalloc if
4867 * possible or committing the transaction. If flush is 0 then no attempts to
4868 * regain reservations will be made and this will fail if there is not enough
4871 static int reserve_metadata_bytes(struct btrfs_root *root,
4872 struct btrfs_block_rsv *block_rsv,
4874 enum btrfs_reserve_flush_enum flush)
4876 struct btrfs_space_info *space_info = block_rsv->space_info;
4878 u64 num_bytes = orig_bytes;
4879 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4881 bool flushing = false;
4885 spin_lock(&space_info->lock);
4887 * We only want to wait if somebody other than us is flushing and we
4888 * are actually allowed to flush all things.
4890 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4891 space_info->flush) {
4892 spin_unlock(&space_info->lock);
4894 * If we have a trans handle we can't wait because the flusher
4895 * may have to commit the transaction, which would mean we would
4896 * deadlock since we are waiting for the flusher to finish, but
4897 * hold the current transaction open.
4899 if (current->journal_info)
4901 ret = wait_event_killable(space_info->wait, !space_info->flush);
4902 /* Must have been killed, return */
4906 spin_lock(&space_info->lock);
4910 used = space_info->bytes_used + space_info->bytes_reserved +
4911 space_info->bytes_pinned + space_info->bytes_readonly +
4912 space_info->bytes_may_use;
4915 * The idea here is that we've not already over-reserved the block group
4916 * then we can go ahead and save our reservation first and then start
4917 * flushing if we need to. Otherwise if we've already overcommitted
4918 * lets start flushing stuff first and then come back and try to make
4921 if (used <= space_info->total_bytes) {
4922 if (used + orig_bytes <= space_info->total_bytes) {
4923 space_info->bytes_may_use += orig_bytes;
4924 trace_btrfs_space_reservation(root->fs_info,
4925 "space_info", space_info->flags, orig_bytes, 1);
4929 * Ok set num_bytes to orig_bytes since we aren't
4930 * overocmmitted, this way we only try and reclaim what
4933 num_bytes = orig_bytes;
4937 * Ok we're over committed, set num_bytes to the overcommitted
4938 * amount plus the amount of bytes that we need for this
4941 num_bytes = used - space_info->total_bytes +
4945 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4946 space_info->bytes_may_use += orig_bytes;
4947 trace_btrfs_space_reservation(root->fs_info, "space_info",
4948 space_info->flags, orig_bytes,
4954 * Couldn't make our reservation, save our place so while we're trying
4955 * to reclaim space we can actually use it instead of somebody else
4956 * stealing it from us.
4958 * We make the other tasks wait for the flush only when we can flush
4961 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4963 space_info->flush = 1;
4964 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4967 * We will do the space reservation dance during log replay,
4968 * which means we won't have fs_info->fs_root set, so don't do
4969 * the async reclaim as we will panic.
4971 if (!root->fs_info->log_root_recovering &&
4972 need_do_async_reclaim(space_info, root->fs_info, used) &&
4973 !work_busy(&root->fs_info->async_reclaim_work))
4974 queue_work(system_unbound_wq,
4975 &root->fs_info->async_reclaim_work);
4977 spin_unlock(&space_info->lock);
4979 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4982 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4987 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4988 * would happen. So skip delalloc flush.
4990 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4991 (flush_state == FLUSH_DELALLOC ||
4992 flush_state == FLUSH_DELALLOC_WAIT))
4993 flush_state = ALLOC_CHUNK;
4997 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4998 flush_state < COMMIT_TRANS)
5000 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
5001 flush_state <= COMMIT_TRANS)
5005 if (ret == -ENOSPC &&
5006 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5007 struct btrfs_block_rsv *global_rsv =
5008 &root->fs_info->global_block_rsv;
5010 if (block_rsv != global_rsv &&
5011 !block_rsv_use_bytes(global_rsv, orig_bytes))
5015 trace_btrfs_space_reservation(root->fs_info,
5016 "space_info:enospc",
5017 space_info->flags, orig_bytes, 1);
5019 spin_lock(&space_info->lock);
5020 space_info->flush = 0;
5021 wake_up_all(&space_info->wait);
5022 spin_unlock(&space_info->lock);
5027 static struct btrfs_block_rsv *get_block_rsv(
5028 const struct btrfs_trans_handle *trans,
5029 const struct btrfs_root *root)
5031 struct btrfs_block_rsv *block_rsv = NULL;
5033 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5034 (root == root->fs_info->csum_root && trans->adding_csums) ||
5035 (root == root->fs_info->uuid_root))
5036 block_rsv = trans->block_rsv;
5039 block_rsv = root->block_rsv;
5042 block_rsv = &root->fs_info->empty_block_rsv;
5047 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5051 spin_lock(&block_rsv->lock);
5052 if (block_rsv->reserved >= num_bytes) {
5053 block_rsv->reserved -= num_bytes;
5054 if (block_rsv->reserved < block_rsv->size)
5055 block_rsv->full = 0;
5058 spin_unlock(&block_rsv->lock);
5062 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5063 u64 num_bytes, int update_size)
5065 spin_lock(&block_rsv->lock);
5066 block_rsv->reserved += num_bytes;
5068 block_rsv->size += num_bytes;
5069 else if (block_rsv->reserved >= block_rsv->size)
5070 block_rsv->full = 1;
5071 spin_unlock(&block_rsv->lock);
5074 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5075 struct btrfs_block_rsv *dest, u64 num_bytes,
5078 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5081 if (global_rsv->space_info != dest->space_info)
5084 spin_lock(&global_rsv->lock);
5085 min_bytes = div_factor(global_rsv->size, min_factor);
5086 if (global_rsv->reserved < min_bytes + num_bytes) {
5087 spin_unlock(&global_rsv->lock);
5090 global_rsv->reserved -= num_bytes;
5091 if (global_rsv->reserved < global_rsv->size)
5092 global_rsv->full = 0;
5093 spin_unlock(&global_rsv->lock);
5095 block_rsv_add_bytes(dest, num_bytes, 1);
5099 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5100 struct btrfs_block_rsv *block_rsv,
5101 struct btrfs_block_rsv *dest, u64 num_bytes)
5103 struct btrfs_space_info *space_info = block_rsv->space_info;
5105 spin_lock(&block_rsv->lock);
5106 if (num_bytes == (u64)-1)
5107 num_bytes = block_rsv->size;
5108 block_rsv->size -= num_bytes;
5109 if (block_rsv->reserved >= block_rsv->size) {
5110 num_bytes = block_rsv->reserved - block_rsv->size;
5111 block_rsv->reserved = block_rsv->size;
5112 block_rsv->full = 1;
5116 spin_unlock(&block_rsv->lock);
5118 if (num_bytes > 0) {
5120 spin_lock(&dest->lock);
5124 bytes_to_add = dest->size - dest->reserved;
5125 bytes_to_add = min(num_bytes, bytes_to_add);
5126 dest->reserved += bytes_to_add;
5127 if (dest->reserved >= dest->size)
5129 num_bytes -= bytes_to_add;
5131 spin_unlock(&dest->lock);
5134 spin_lock(&space_info->lock);
5135 space_info->bytes_may_use -= num_bytes;
5136 trace_btrfs_space_reservation(fs_info, "space_info",
5137 space_info->flags, num_bytes, 0);
5138 spin_unlock(&space_info->lock);
5143 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5144 struct btrfs_block_rsv *dst, u64 num_bytes)
5148 ret = block_rsv_use_bytes(src, num_bytes);
5152 block_rsv_add_bytes(dst, num_bytes, 1);
5156 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5158 memset(rsv, 0, sizeof(*rsv));
5159 spin_lock_init(&rsv->lock);
5163 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5164 unsigned short type)
5166 struct btrfs_block_rsv *block_rsv;
5167 struct btrfs_fs_info *fs_info = root->fs_info;
5169 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5173 btrfs_init_block_rsv(block_rsv, type);
5174 block_rsv->space_info = __find_space_info(fs_info,
5175 BTRFS_BLOCK_GROUP_METADATA);
5179 void btrfs_free_block_rsv(struct btrfs_root *root,
5180 struct btrfs_block_rsv *rsv)
5184 btrfs_block_rsv_release(root, rsv, (u64)-1);
5188 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5193 int btrfs_block_rsv_add(struct btrfs_root *root,
5194 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5195 enum btrfs_reserve_flush_enum flush)
5202 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5204 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5211 int btrfs_block_rsv_check(struct btrfs_root *root,
5212 struct btrfs_block_rsv *block_rsv, int min_factor)
5220 spin_lock(&block_rsv->lock);
5221 num_bytes = div_factor(block_rsv->size, min_factor);
5222 if (block_rsv->reserved >= num_bytes)
5224 spin_unlock(&block_rsv->lock);
5229 int btrfs_block_rsv_refill(struct btrfs_root *root,
5230 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5231 enum btrfs_reserve_flush_enum flush)
5239 spin_lock(&block_rsv->lock);
5240 num_bytes = min_reserved;
5241 if (block_rsv->reserved >= num_bytes)
5244 num_bytes -= block_rsv->reserved;
5245 spin_unlock(&block_rsv->lock);
5250 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5252 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5259 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5260 struct btrfs_block_rsv *dst_rsv,
5263 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5266 void btrfs_block_rsv_release(struct btrfs_root *root,
5267 struct btrfs_block_rsv *block_rsv,
5270 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5271 if (global_rsv == block_rsv ||
5272 block_rsv->space_info != global_rsv->space_info)
5274 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5279 * helper to calculate size of global block reservation.
5280 * the desired value is sum of space used by extent tree,
5281 * checksum tree and root tree
5283 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5285 struct btrfs_space_info *sinfo;
5289 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5291 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5292 spin_lock(&sinfo->lock);
5293 data_used = sinfo->bytes_used;
5294 spin_unlock(&sinfo->lock);
5296 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5297 spin_lock(&sinfo->lock);
5298 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5300 meta_used = sinfo->bytes_used;
5301 spin_unlock(&sinfo->lock);
5303 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5305 num_bytes += div_u64(data_used + meta_used, 50);
5307 if (num_bytes * 3 > meta_used)
5308 num_bytes = div_u64(meta_used, 3);
5310 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5313 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5315 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5316 struct btrfs_space_info *sinfo = block_rsv->space_info;
5319 num_bytes = calc_global_metadata_size(fs_info);
5321 spin_lock(&sinfo->lock);
5322 spin_lock(&block_rsv->lock);
5324 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5326 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5327 sinfo->bytes_reserved + sinfo->bytes_readonly +
5328 sinfo->bytes_may_use;
5330 if (sinfo->total_bytes > num_bytes) {
5331 num_bytes = sinfo->total_bytes - num_bytes;
5332 block_rsv->reserved += num_bytes;
5333 sinfo->bytes_may_use += num_bytes;
5334 trace_btrfs_space_reservation(fs_info, "space_info",
5335 sinfo->flags, num_bytes, 1);
5338 if (block_rsv->reserved >= block_rsv->size) {
5339 num_bytes = block_rsv->reserved - block_rsv->size;
5340 sinfo->bytes_may_use -= num_bytes;
5341 trace_btrfs_space_reservation(fs_info, "space_info",
5342 sinfo->flags, num_bytes, 0);
5343 block_rsv->reserved = block_rsv->size;
5344 block_rsv->full = 1;
5347 spin_unlock(&block_rsv->lock);
5348 spin_unlock(&sinfo->lock);
5351 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5353 struct btrfs_space_info *space_info;
5355 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5356 fs_info->chunk_block_rsv.space_info = space_info;
5358 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5359 fs_info->global_block_rsv.space_info = space_info;
5360 fs_info->delalloc_block_rsv.space_info = space_info;
5361 fs_info->trans_block_rsv.space_info = space_info;
5362 fs_info->empty_block_rsv.space_info = space_info;
5363 fs_info->delayed_block_rsv.space_info = space_info;
5365 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5366 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5367 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5368 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5369 if (fs_info->quota_root)
5370 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5371 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5373 update_global_block_rsv(fs_info);
5376 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5378 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5380 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5381 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5382 WARN_ON(fs_info->trans_block_rsv.size > 0);
5383 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5384 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5385 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5386 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5387 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5390 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5391 struct btrfs_root *root)
5393 if (!trans->block_rsv)
5396 if (!trans->bytes_reserved)
5399 trace_btrfs_space_reservation(root->fs_info, "transaction",
5400 trans->transid, trans->bytes_reserved, 0);
5401 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5402 trans->bytes_reserved = 0;
5406 * To be called after all the new block groups attached to the transaction
5407 * handle have been created (btrfs_create_pending_block_groups()).
5409 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5411 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5413 if (!trans->chunk_bytes_reserved)
5416 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5418 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5419 trans->chunk_bytes_reserved);
5420 trans->chunk_bytes_reserved = 0;
5423 /* Can only return 0 or -ENOSPC */
5424 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5425 struct inode *inode)
5427 struct btrfs_root *root = BTRFS_I(inode)->root;
5428 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5429 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5432 * We need to hold space in order to delete our orphan item once we've
5433 * added it, so this takes the reservation so we can release it later
5434 * when we are truly done with the orphan item.
5436 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5437 trace_btrfs_space_reservation(root->fs_info, "orphan",
5438 btrfs_ino(inode), num_bytes, 1);
5439 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5442 void btrfs_orphan_release_metadata(struct inode *inode)
5444 struct btrfs_root *root = BTRFS_I(inode)->root;
5445 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5446 trace_btrfs_space_reservation(root->fs_info, "orphan",
5447 btrfs_ino(inode), num_bytes, 0);
5448 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5452 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5453 * root: the root of the parent directory
5454 * rsv: block reservation
5455 * items: the number of items that we need do reservation
5456 * qgroup_reserved: used to return the reserved size in qgroup
5458 * This function is used to reserve the space for snapshot/subvolume
5459 * creation and deletion. Those operations are different with the
5460 * common file/directory operations, they change two fs/file trees
5461 * and root tree, the number of items that the qgroup reserves is
5462 * different with the free space reservation. So we can not use
5463 * the space reseravtion mechanism in start_transaction().
5465 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5466 struct btrfs_block_rsv *rsv,
5468 u64 *qgroup_reserved,
5469 bool use_global_rsv)
5473 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5475 if (root->fs_info->quota_enabled) {
5476 /* One for parent inode, two for dir entries */
5477 num_bytes = 3 * root->nodesize;
5478 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5485 *qgroup_reserved = num_bytes;
5487 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5488 rsv->space_info = __find_space_info(root->fs_info,
5489 BTRFS_BLOCK_GROUP_METADATA);
5490 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5491 BTRFS_RESERVE_FLUSH_ALL);
5493 if (ret == -ENOSPC && use_global_rsv)
5494 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5496 if (ret && *qgroup_reserved)
5497 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5502 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5503 struct btrfs_block_rsv *rsv,
5504 u64 qgroup_reserved)
5506 btrfs_block_rsv_release(root, rsv, (u64)-1);
5510 * drop_outstanding_extent - drop an outstanding extent
5511 * @inode: the inode we're dropping the extent for
5512 * @num_bytes: the number of bytes we're relaseing.
5514 * This is called when we are freeing up an outstanding extent, either called
5515 * after an error or after an extent is written. This will return the number of
5516 * reserved extents that need to be freed. This must be called with
5517 * BTRFS_I(inode)->lock held.
5519 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5521 unsigned drop_inode_space = 0;
5522 unsigned dropped_extents = 0;
5523 unsigned num_extents = 0;
5525 num_extents = (unsigned)div64_u64(num_bytes +
5526 BTRFS_MAX_EXTENT_SIZE - 1,
5527 BTRFS_MAX_EXTENT_SIZE);
5528 ASSERT(num_extents);
5529 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5530 BTRFS_I(inode)->outstanding_extents -= num_extents;
5532 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5533 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5534 &BTRFS_I(inode)->runtime_flags))
5535 drop_inode_space = 1;
5538 * If we have more or the same amount of outsanding extents than we have
5539 * reserved then we need to leave the reserved extents count alone.
5541 if (BTRFS_I(inode)->outstanding_extents >=
5542 BTRFS_I(inode)->reserved_extents)
5543 return drop_inode_space;
5545 dropped_extents = BTRFS_I(inode)->reserved_extents -
5546 BTRFS_I(inode)->outstanding_extents;
5547 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5548 return dropped_extents + drop_inode_space;
5552 * calc_csum_metadata_size - return the amount of metada space that must be
5553 * reserved/free'd for the given bytes.
5554 * @inode: the inode we're manipulating
5555 * @num_bytes: the number of bytes in question
5556 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5558 * This adjusts the number of csum_bytes in the inode and then returns the
5559 * correct amount of metadata that must either be reserved or freed. We
5560 * calculate how many checksums we can fit into one leaf and then divide the
5561 * number of bytes that will need to be checksumed by this value to figure out
5562 * how many checksums will be required. If we are adding bytes then the number
5563 * may go up and we will return the number of additional bytes that must be
5564 * reserved. If it is going down we will return the number of bytes that must
5567 * This must be called with BTRFS_I(inode)->lock held.
5569 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5572 struct btrfs_root *root = BTRFS_I(inode)->root;
5573 u64 old_csums, num_csums;
5575 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5576 BTRFS_I(inode)->csum_bytes == 0)
5579 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5581 BTRFS_I(inode)->csum_bytes += num_bytes;
5583 BTRFS_I(inode)->csum_bytes -= num_bytes;
5584 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5586 /* No change, no need to reserve more */
5587 if (old_csums == num_csums)
5591 return btrfs_calc_trans_metadata_size(root,
5592 num_csums - old_csums);
5594 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5597 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5599 struct btrfs_root *root = BTRFS_I(inode)->root;
5600 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5603 unsigned nr_extents = 0;
5604 int extra_reserve = 0;
5605 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5607 bool delalloc_lock = true;
5611 /* If we are a free space inode we need to not flush since we will be in
5612 * the middle of a transaction commit. We also don't need the delalloc
5613 * mutex since we won't race with anybody. We need this mostly to make
5614 * lockdep shut its filthy mouth.
5616 if (btrfs_is_free_space_inode(inode)) {
5617 flush = BTRFS_RESERVE_NO_FLUSH;
5618 delalloc_lock = false;
5621 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5622 btrfs_transaction_in_commit(root->fs_info))
5623 schedule_timeout(1);
5626 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5628 num_bytes = ALIGN(num_bytes, root->sectorsize);
5630 spin_lock(&BTRFS_I(inode)->lock);
5631 nr_extents = (unsigned)div64_u64(num_bytes +
5632 BTRFS_MAX_EXTENT_SIZE - 1,
5633 BTRFS_MAX_EXTENT_SIZE);
5634 BTRFS_I(inode)->outstanding_extents += nr_extents;
5637 if (BTRFS_I(inode)->outstanding_extents >
5638 BTRFS_I(inode)->reserved_extents)
5639 nr_extents = BTRFS_I(inode)->outstanding_extents -
5640 BTRFS_I(inode)->reserved_extents;
5643 * Add an item to reserve for updating the inode when we complete the
5646 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5647 &BTRFS_I(inode)->runtime_flags)) {
5652 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5653 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5654 csum_bytes = BTRFS_I(inode)->csum_bytes;
5655 spin_unlock(&BTRFS_I(inode)->lock);
5657 if (root->fs_info->quota_enabled) {
5658 ret = btrfs_qgroup_reserve_meta(root,
5659 nr_extents * root->nodesize);
5664 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5665 if (unlikely(ret)) {
5666 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5670 spin_lock(&BTRFS_I(inode)->lock);
5671 if (extra_reserve) {
5672 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5673 &BTRFS_I(inode)->runtime_flags);
5676 BTRFS_I(inode)->reserved_extents += nr_extents;
5677 spin_unlock(&BTRFS_I(inode)->lock);
5680 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5683 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5684 btrfs_ino(inode), to_reserve, 1);
5685 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5690 spin_lock(&BTRFS_I(inode)->lock);
5691 dropped = drop_outstanding_extent(inode, num_bytes);
5693 * If the inodes csum_bytes is the same as the original
5694 * csum_bytes then we know we haven't raced with any free()ers
5695 * so we can just reduce our inodes csum bytes and carry on.
5697 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5698 calc_csum_metadata_size(inode, num_bytes, 0);
5700 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5704 * This is tricky, but first we need to figure out how much we
5705 * free'd from any free-ers that occured during this
5706 * reservation, so we reset ->csum_bytes to the csum_bytes
5707 * before we dropped our lock, and then call the free for the
5708 * number of bytes that were freed while we were trying our
5711 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5712 BTRFS_I(inode)->csum_bytes = csum_bytes;
5713 to_free = calc_csum_metadata_size(inode, bytes, 0);
5717 * Now we need to see how much we would have freed had we not
5718 * been making this reservation and our ->csum_bytes were not
5719 * artificially inflated.
5721 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5722 bytes = csum_bytes - orig_csum_bytes;
5723 bytes = calc_csum_metadata_size(inode, bytes, 0);
5726 * Now reset ->csum_bytes to what it should be. If bytes is
5727 * more than to_free then we would have free'd more space had we
5728 * not had an artificially high ->csum_bytes, so we need to free
5729 * the remainder. If bytes is the same or less then we don't
5730 * need to do anything, the other free-ers did the correct
5733 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5734 if (bytes > to_free)
5735 to_free = bytes - to_free;
5739 spin_unlock(&BTRFS_I(inode)->lock);
5741 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5744 btrfs_block_rsv_release(root, block_rsv, to_free);
5745 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5746 btrfs_ino(inode), to_free, 0);
5749 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5754 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5755 * @inode: the inode to release the reservation for
5756 * @num_bytes: the number of bytes we're releasing
5758 * This will release the metadata reservation for an inode. This can be called
5759 * once we complete IO for a given set of bytes to release their metadata
5762 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5764 struct btrfs_root *root = BTRFS_I(inode)->root;
5768 num_bytes = ALIGN(num_bytes, root->sectorsize);
5769 spin_lock(&BTRFS_I(inode)->lock);
5770 dropped = drop_outstanding_extent(inode, num_bytes);
5773 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5774 spin_unlock(&BTRFS_I(inode)->lock);
5776 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5778 if (btrfs_test_is_dummy_root(root))
5781 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5782 btrfs_ino(inode), to_free, 0);
5784 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5789 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5791 * @inode: inode we're writing to
5792 * @start: start range we are writing to
5793 * @len: how long the range we are writing to
5795 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5797 * This will do the following things
5799 * o reserve space in data space info for num bytes
5800 * and reserve precious corresponding qgroup space
5801 * (Done in check_data_free_space)
5803 * o reserve space for metadata space, based on the number of outstanding
5804 * extents and how much csums will be needed
5805 * also reserve metadata space in a per root over-reserve method.
5806 * o add to the inodes->delalloc_bytes
5807 * o add it to the fs_info's delalloc inodes list.
5808 * (Above 3 all done in delalloc_reserve_metadata)
5810 * Return 0 for success
5811 * Return <0 for error(-ENOSPC or -EQUOT)
5813 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5817 ret = btrfs_check_data_free_space(inode, start, len);
5820 ret = btrfs_delalloc_reserve_metadata(inode, len);
5822 btrfs_free_reserved_data_space(inode, start, len);
5827 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5828 * @inode: inode we're releasing space for
5829 * @start: start position of the space already reserved
5830 * @len: the len of the space already reserved
5832 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5833 * called in the case that we don't need the metadata AND data reservations
5834 * anymore. So if there is an error or we insert an inline extent.
5836 * This function will release the metadata space that was not used and will
5837 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5838 * list if there are no delalloc bytes left.
5839 * Also it will handle the qgroup reserved space.
5841 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5843 btrfs_delalloc_release_metadata(inode, len);
5844 btrfs_free_reserved_data_space(inode, start, len);
5847 static int update_block_group(struct btrfs_trans_handle *trans,
5848 struct btrfs_root *root, u64 bytenr,
5849 u64 num_bytes, int alloc)
5851 struct btrfs_block_group_cache *cache = NULL;
5852 struct btrfs_fs_info *info = root->fs_info;
5853 u64 total = num_bytes;
5858 /* block accounting for super block */
5859 spin_lock(&info->delalloc_root_lock);
5860 old_val = btrfs_super_bytes_used(info->super_copy);
5862 old_val += num_bytes;
5864 old_val -= num_bytes;
5865 btrfs_set_super_bytes_used(info->super_copy, old_val);
5866 spin_unlock(&info->delalloc_root_lock);
5869 cache = btrfs_lookup_block_group(info, bytenr);
5872 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5873 BTRFS_BLOCK_GROUP_RAID1 |
5874 BTRFS_BLOCK_GROUP_RAID10))
5879 * If this block group has free space cache written out, we
5880 * need to make sure to load it if we are removing space. This
5881 * is because we need the unpinning stage to actually add the
5882 * space back to the block group, otherwise we will leak space.
5884 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5885 cache_block_group(cache, 1);
5887 byte_in_group = bytenr - cache->key.objectid;
5888 WARN_ON(byte_in_group > cache->key.offset);
5890 spin_lock(&cache->space_info->lock);
5891 spin_lock(&cache->lock);
5893 if (btrfs_test_opt(root, SPACE_CACHE) &&
5894 cache->disk_cache_state < BTRFS_DC_CLEAR)
5895 cache->disk_cache_state = BTRFS_DC_CLEAR;
5897 old_val = btrfs_block_group_used(&cache->item);
5898 num_bytes = min(total, cache->key.offset - byte_in_group);
5900 old_val += num_bytes;
5901 btrfs_set_block_group_used(&cache->item, old_val);
5902 cache->reserved -= num_bytes;
5903 cache->space_info->bytes_reserved -= num_bytes;
5904 cache->space_info->bytes_used += num_bytes;
5905 cache->space_info->disk_used += num_bytes * factor;
5906 spin_unlock(&cache->lock);
5907 spin_unlock(&cache->space_info->lock);
5909 old_val -= num_bytes;
5910 btrfs_set_block_group_used(&cache->item, old_val);
5911 cache->pinned += num_bytes;
5912 cache->space_info->bytes_pinned += num_bytes;
5913 cache->space_info->bytes_used -= num_bytes;
5914 cache->space_info->disk_used -= num_bytes * factor;
5915 spin_unlock(&cache->lock);
5916 spin_unlock(&cache->space_info->lock);
5918 set_extent_dirty(info->pinned_extents,
5919 bytenr, bytenr + num_bytes - 1,
5920 GFP_NOFS | __GFP_NOFAIL);
5923 spin_lock(&trans->transaction->dirty_bgs_lock);
5924 if (list_empty(&cache->dirty_list)) {
5925 list_add_tail(&cache->dirty_list,
5926 &trans->transaction->dirty_bgs);
5927 trans->transaction->num_dirty_bgs++;
5928 btrfs_get_block_group(cache);
5930 spin_unlock(&trans->transaction->dirty_bgs_lock);
5933 * No longer have used bytes in this block group, queue it for
5934 * deletion. We do this after adding the block group to the
5935 * dirty list to avoid races between cleaner kthread and space
5938 if (!alloc && old_val == 0) {
5939 spin_lock(&info->unused_bgs_lock);
5940 if (list_empty(&cache->bg_list)) {
5941 btrfs_get_block_group(cache);
5942 list_add_tail(&cache->bg_list,
5945 spin_unlock(&info->unused_bgs_lock);
5948 btrfs_put_block_group(cache);
5950 bytenr += num_bytes;
5955 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5957 struct btrfs_block_group_cache *cache;
5960 spin_lock(&root->fs_info->block_group_cache_lock);
5961 bytenr = root->fs_info->first_logical_byte;
5962 spin_unlock(&root->fs_info->block_group_cache_lock);
5964 if (bytenr < (u64)-1)
5967 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5971 bytenr = cache->key.objectid;
5972 btrfs_put_block_group(cache);
5977 static int pin_down_extent(struct btrfs_root *root,
5978 struct btrfs_block_group_cache *cache,
5979 u64 bytenr, u64 num_bytes, int reserved)
5981 spin_lock(&cache->space_info->lock);
5982 spin_lock(&cache->lock);
5983 cache->pinned += num_bytes;
5984 cache->space_info->bytes_pinned += num_bytes;
5986 cache->reserved -= num_bytes;
5987 cache->space_info->bytes_reserved -= num_bytes;
5989 spin_unlock(&cache->lock);
5990 spin_unlock(&cache->space_info->lock);
5992 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5993 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5995 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
6000 * this function must be called within transaction
6002 int btrfs_pin_extent(struct btrfs_root *root,
6003 u64 bytenr, u64 num_bytes, int reserved)
6005 struct btrfs_block_group_cache *cache;
6007 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6008 BUG_ON(!cache); /* Logic error */
6010 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6012 btrfs_put_block_group(cache);
6017 * this function must be called within transaction
6019 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6020 u64 bytenr, u64 num_bytes)
6022 struct btrfs_block_group_cache *cache;
6025 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6030 * pull in the free space cache (if any) so that our pin
6031 * removes the free space from the cache. We have load_only set
6032 * to one because the slow code to read in the free extents does check
6033 * the pinned extents.
6035 cache_block_group(cache, 1);
6037 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6039 /* remove us from the free space cache (if we're there at all) */
6040 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6041 btrfs_put_block_group(cache);
6045 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6048 struct btrfs_block_group_cache *block_group;
6049 struct btrfs_caching_control *caching_ctl;
6051 block_group = btrfs_lookup_block_group(root->fs_info, start);
6055 cache_block_group(block_group, 0);
6056 caching_ctl = get_caching_control(block_group);
6060 BUG_ON(!block_group_cache_done(block_group));
6061 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6063 mutex_lock(&caching_ctl->mutex);
6065 if (start >= caching_ctl->progress) {
6066 ret = add_excluded_extent(root, start, num_bytes);
6067 } else if (start + num_bytes <= caching_ctl->progress) {
6068 ret = btrfs_remove_free_space(block_group,
6071 num_bytes = caching_ctl->progress - start;
6072 ret = btrfs_remove_free_space(block_group,
6077 num_bytes = (start + num_bytes) -
6078 caching_ctl->progress;
6079 start = caching_ctl->progress;
6080 ret = add_excluded_extent(root, start, num_bytes);
6083 mutex_unlock(&caching_ctl->mutex);
6084 put_caching_control(caching_ctl);
6086 btrfs_put_block_group(block_group);
6090 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6091 struct extent_buffer *eb)
6093 struct btrfs_file_extent_item *item;
6094 struct btrfs_key key;
6098 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6101 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6102 btrfs_item_key_to_cpu(eb, &key, i);
6103 if (key.type != BTRFS_EXTENT_DATA_KEY)
6105 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6106 found_type = btrfs_file_extent_type(eb, item);
6107 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6109 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6111 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6112 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6113 __exclude_logged_extent(log, key.objectid, key.offset);
6120 * btrfs_update_reserved_bytes - update the block_group and space info counters
6121 * @cache: The cache we are manipulating
6122 * @num_bytes: The number of bytes in question
6123 * @reserve: One of the reservation enums
6124 * @delalloc: The blocks are allocated for the delalloc write
6126 * This is called by the allocator when it reserves space, or by somebody who is
6127 * freeing space that was never actually used on disk. For example if you
6128 * reserve some space for a new leaf in transaction A and before transaction A
6129 * commits you free that leaf, you call this with reserve set to 0 in order to
6130 * clear the reservation.
6132 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6133 * ENOSPC accounting. For data we handle the reservation through clearing the
6134 * delalloc bits in the io_tree. We have to do this since we could end up
6135 * allocating less disk space for the amount of data we have reserved in the
6136 * case of compression.
6138 * If this is a reservation and the block group has become read only we cannot
6139 * make the reservation and return -EAGAIN, otherwise this function always
6142 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6143 u64 num_bytes, int reserve, int delalloc)
6145 struct btrfs_space_info *space_info = cache->space_info;
6148 spin_lock(&space_info->lock);
6149 spin_lock(&cache->lock);
6150 if (reserve != RESERVE_FREE) {
6154 cache->reserved += num_bytes;
6155 space_info->bytes_reserved += num_bytes;
6156 if (reserve == RESERVE_ALLOC) {
6157 trace_btrfs_space_reservation(cache->fs_info,
6158 "space_info", space_info->flags,
6160 space_info->bytes_may_use -= num_bytes;
6164 cache->delalloc_bytes += num_bytes;
6168 space_info->bytes_readonly += num_bytes;
6169 cache->reserved -= num_bytes;
6170 space_info->bytes_reserved -= num_bytes;
6173 cache->delalloc_bytes -= num_bytes;
6175 spin_unlock(&cache->lock);
6176 spin_unlock(&space_info->lock);
6180 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6181 struct btrfs_root *root)
6183 struct btrfs_fs_info *fs_info = root->fs_info;
6184 struct btrfs_caching_control *next;
6185 struct btrfs_caching_control *caching_ctl;
6186 struct btrfs_block_group_cache *cache;
6188 down_write(&fs_info->commit_root_sem);
6190 list_for_each_entry_safe(caching_ctl, next,
6191 &fs_info->caching_block_groups, list) {
6192 cache = caching_ctl->block_group;
6193 if (block_group_cache_done(cache)) {
6194 cache->last_byte_to_unpin = (u64)-1;
6195 list_del_init(&caching_ctl->list);
6196 put_caching_control(caching_ctl);
6198 cache->last_byte_to_unpin = caching_ctl->progress;
6202 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6203 fs_info->pinned_extents = &fs_info->freed_extents[1];
6205 fs_info->pinned_extents = &fs_info->freed_extents[0];
6207 up_write(&fs_info->commit_root_sem);
6209 update_global_block_rsv(fs_info);
6213 * Returns the free cluster for the given space info and sets empty_cluster to
6214 * what it should be based on the mount options.
6216 static struct btrfs_free_cluster *
6217 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6220 struct btrfs_free_cluster *ret = NULL;
6221 bool ssd = btrfs_test_opt(root, SSD);
6224 if (btrfs_mixed_space_info(space_info))
6228 *empty_cluster = 2 * 1024 * 1024;
6229 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6230 ret = &root->fs_info->meta_alloc_cluster;
6232 *empty_cluster = 64 * 1024;
6233 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6234 ret = &root->fs_info->data_alloc_cluster;
6240 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6241 const bool return_free_space)
6243 struct btrfs_fs_info *fs_info = root->fs_info;
6244 struct btrfs_block_group_cache *cache = NULL;
6245 struct btrfs_space_info *space_info;
6246 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6247 struct btrfs_free_cluster *cluster = NULL;
6249 u64 total_unpinned = 0;
6250 u64 empty_cluster = 0;
6253 while (start <= end) {
6256 start >= cache->key.objectid + cache->key.offset) {
6258 btrfs_put_block_group(cache);
6260 cache = btrfs_lookup_block_group(fs_info, start);
6261 BUG_ON(!cache); /* Logic error */
6263 cluster = fetch_cluster_info(root,
6266 empty_cluster <<= 1;
6269 len = cache->key.objectid + cache->key.offset - start;
6270 len = min(len, end + 1 - start);
6272 if (start < cache->last_byte_to_unpin) {
6273 len = min(len, cache->last_byte_to_unpin - start);
6274 if (return_free_space)
6275 btrfs_add_free_space(cache, start, len);
6279 total_unpinned += len;
6280 space_info = cache->space_info;
6283 * If this space cluster has been marked as fragmented and we've
6284 * unpinned enough in this block group to potentially allow a
6285 * cluster to be created inside of it go ahead and clear the
6288 if (cluster && cluster->fragmented &&
6289 total_unpinned > empty_cluster) {
6290 spin_lock(&cluster->lock);
6291 cluster->fragmented = 0;
6292 spin_unlock(&cluster->lock);
6295 spin_lock(&space_info->lock);
6296 spin_lock(&cache->lock);
6297 cache->pinned -= len;
6298 space_info->bytes_pinned -= len;
6299 space_info->max_extent_size = 0;
6300 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6302 space_info->bytes_readonly += len;
6305 spin_unlock(&cache->lock);
6306 if (!readonly && global_rsv->space_info == space_info) {
6307 spin_lock(&global_rsv->lock);
6308 if (!global_rsv->full) {
6309 len = min(len, global_rsv->size -
6310 global_rsv->reserved);
6311 global_rsv->reserved += len;
6312 space_info->bytes_may_use += len;
6313 if (global_rsv->reserved >= global_rsv->size)
6314 global_rsv->full = 1;
6316 spin_unlock(&global_rsv->lock);
6318 spin_unlock(&space_info->lock);
6322 btrfs_put_block_group(cache);
6326 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6327 struct btrfs_root *root)
6329 struct btrfs_fs_info *fs_info = root->fs_info;
6330 struct btrfs_block_group_cache *block_group, *tmp;
6331 struct list_head *deleted_bgs;
6332 struct extent_io_tree *unpin;
6337 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6338 unpin = &fs_info->freed_extents[1];
6340 unpin = &fs_info->freed_extents[0];
6342 while (!trans->aborted) {
6343 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6344 ret = find_first_extent_bit(unpin, 0, &start, &end,
6345 EXTENT_DIRTY, NULL);
6347 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6351 if (btrfs_test_opt(root, DISCARD))
6352 ret = btrfs_discard_extent(root, start,
6353 end + 1 - start, NULL);
6355 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6356 unpin_extent_range(root, start, end, true);
6357 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6362 * Transaction is finished. We don't need the lock anymore. We
6363 * do need to clean up the block groups in case of a transaction
6366 deleted_bgs = &trans->transaction->deleted_bgs;
6367 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6371 if (!trans->aborted)
6372 ret = btrfs_discard_extent(root,
6373 block_group->key.objectid,
6374 block_group->key.offset,
6377 list_del_init(&block_group->bg_list);
6378 btrfs_put_block_group_trimming(block_group);
6379 btrfs_put_block_group(block_group);
6382 const char *errstr = btrfs_decode_error(ret);
6384 "Discard failed while removing blockgroup: errno=%d %s\n",
6392 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6393 u64 owner, u64 root_objectid)
6395 struct btrfs_space_info *space_info;
6398 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6399 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6400 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6402 flags = BTRFS_BLOCK_GROUP_METADATA;
6404 flags = BTRFS_BLOCK_GROUP_DATA;
6407 space_info = __find_space_info(fs_info, flags);
6408 BUG_ON(!space_info); /* Logic bug */
6409 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6413 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6414 struct btrfs_root *root,
6415 struct btrfs_delayed_ref_node *node, u64 parent,
6416 u64 root_objectid, u64 owner_objectid,
6417 u64 owner_offset, int refs_to_drop,
6418 struct btrfs_delayed_extent_op *extent_op)
6420 struct btrfs_key key;
6421 struct btrfs_path *path;
6422 struct btrfs_fs_info *info = root->fs_info;
6423 struct btrfs_root *extent_root = info->extent_root;
6424 struct extent_buffer *leaf;
6425 struct btrfs_extent_item *ei;
6426 struct btrfs_extent_inline_ref *iref;
6429 int extent_slot = 0;
6430 int found_extent = 0;
6434 u64 bytenr = node->bytenr;
6435 u64 num_bytes = node->num_bytes;
6437 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6440 path = btrfs_alloc_path();
6445 path->leave_spinning = 1;
6447 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6448 BUG_ON(!is_data && refs_to_drop != 1);
6451 skinny_metadata = 0;
6453 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6454 bytenr, num_bytes, parent,
6455 root_objectid, owner_objectid,
6458 extent_slot = path->slots[0];
6459 while (extent_slot >= 0) {
6460 btrfs_item_key_to_cpu(path->nodes[0], &key,
6462 if (key.objectid != bytenr)
6464 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6465 key.offset == num_bytes) {
6469 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6470 key.offset == owner_objectid) {
6474 if (path->slots[0] - extent_slot > 5)
6478 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6479 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6480 if (found_extent && item_size < sizeof(*ei))
6483 if (!found_extent) {
6485 ret = remove_extent_backref(trans, extent_root, path,
6487 is_data, &last_ref);
6489 btrfs_abort_transaction(trans, extent_root, ret);
6492 btrfs_release_path(path);
6493 path->leave_spinning = 1;
6495 key.objectid = bytenr;
6496 key.type = BTRFS_EXTENT_ITEM_KEY;
6497 key.offset = num_bytes;
6499 if (!is_data && skinny_metadata) {
6500 key.type = BTRFS_METADATA_ITEM_KEY;
6501 key.offset = owner_objectid;
6504 ret = btrfs_search_slot(trans, extent_root,
6506 if (ret > 0 && skinny_metadata && path->slots[0]) {
6508 * Couldn't find our skinny metadata item,
6509 * see if we have ye olde extent item.
6512 btrfs_item_key_to_cpu(path->nodes[0], &key,
6514 if (key.objectid == bytenr &&
6515 key.type == BTRFS_EXTENT_ITEM_KEY &&
6516 key.offset == num_bytes)
6520 if (ret > 0 && skinny_metadata) {
6521 skinny_metadata = false;
6522 key.objectid = bytenr;
6523 key.type = BTRFS_EXTENT_ITEM_KEY;
6524 key.offset = num_bytes;
6525 btrfs_release_path(path);
6526 ret = btrfs_search_slot(trans, extent_root,
6531 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6534 btrfs_print_leaf(extent_root,
6538 btrfs_abort_transaction(trans, extent_root, ret);
6541 extent_slot = path->slots[0];
6543 } else if (WARN_ON(ret == -ENOENT)) {
6544 btrfs_print_leaf(extent_root, path->nodes[0]);
6546 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6547 bytenr, parent, root_objectid, owner_objectid,
6549 btrfs_abort_transaction(trans, extent_root, ret);
6552 btrfs_abort_transaction(trans, extent_root, ret);
6556 leaf = path->nodes[0];
6557 item_size = btrfs_item_size_nr(leaf, extent_slot);
6558 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6559 if (item_size < sizeof(*ei)) {
6560 BUG_ON(found_extent || extent_slot != path->slots[0]);
6561 ret = convert_extent_item_v0(trans, extent_root, path,
6564 btrfs_abort_transaction(trans, extent_root, ret);
6568 btrfs_release_path(path);
6569 path->leave_spinning = 1;
6571 key.objectid = bytenr;
6572 key.type = BTRFS_EXTENT_ITEM_KEY;
6573 key.offset = num_bytes;
6575 ret = btrfs_search_slot(trans, extent_root, &key, path,
6578 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6580 btrfs_print_leaf(extent_root, path->nodes[0]);
6583 btrfs_abort_transaction(trans, extent_root, ret);
6587 extent_slot = path->slots[0];
6588 leaf = path->nodes[0];
6589 item_size = btrfs_item_size_nr(leaf, extent_slot);
6592 BUG_ON(item_size < sizeof(*ei));
6593 ei = btrfs_item_ptr(leaf, extent_slot,
6594 struct btrfs_extent_item);
6595 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6596 key.type == BTRFS_EXTENT_ITEM_KEY) {
6597 struct btrfs_tree_block_info *bi;
6598 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6599 bi = (struct btrfs_tree_block_info *)(ei + 1);
6600 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6603 refs = btrfs_extent_refs(leaf, ei);
6604 if (refs < refs_to_drop) {
6605 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6606 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6608 btrfs_abort_transaction(trans, extent_root, ret);
6611 refs -= refs_to_drop;
6615 __run_delayed_extent_op(extent_op, leaf, ei);
6617 * In the case of inline back ref, reference count will
6618 * be updated by remove_extent_backref
6621 BUG_ON(!found_extent);
6623 btrfs_set_extent_refs(leaf, ei, refs);
6624 btrfs_mark_buffer_dirty(leaf);
6627 ret = remove_extent_backref(trans, extent_root, path,
6629 is_data, &last_ref);
6631 btrfs_abort_transaction(trans, extent_root, ret);
6635 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6639 BUG_ON(is_data && refs_to_drop !=
6640 extent_data_ref_count(path, iref));
6642 BUG_ON(path->slots[0] != extent_slot);
6644 BUG_ON(path->slots[0] != extent_slot + 1);
6645 path->slots[0] = extent_slot;
6651 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6654 btrfs_abort_transaction(trans, extent_root, ret);
6657 btrfs_release_path(path);
6660 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6662 btrfs_abort_transaction(trans, extent_root, ret);
6667 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6669 btrfs_abort_transaction(trans, extent_root, ret);
6673 btrfs_release_path(path);
6676 btrfs_free_path(path);
6681 * when we free an block, it is possible (and likely) that we free the last
6682 * delayed ref for that extent as well. This searches the delayed ref tree for
6683 * a given extent, and if there are no other delayed refs to be processed, it
6684 * removes it from the tree.
6686 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6687 struct btrfs_root *root, u64 bytenr)
6689 struct btrfs_delayed_ref_head *head;
6690 struct btrfs_delayed_ref_root *delayed_refs;
6693 delayed_refs = &trans->transaction->delayed_refs;
6694 spin_lock(&delayed_refs->lock);
6695 head = btrfs_find_delayed_ref_head(trans, bytenr);
6697 goto out_delayed_unlock;
6699 spin_lock(&head->lock);
6700 if (!list_empty(&head->ref_list))
6703 if (head->extent_op) {
6704 if (!head->must_insert_reserved)
6706 btrfs_free_delayed_extent_op(head->extent_op);
6707 head->extent_op = NULL;
6711 * waiting for the lock here would deadlock. If someone else has it
6712 * locked they are already in the process of dropping it anyway
6714 if (!mutex_trylock(&head->mutex))
6718 * at this point we have a head with no other entries. Go
6719 * ahead and process it.
6721 head->node.in_tree = 0;
6722 rb_erase(&head->href_node, &delayed_refs->href_root);
6724 atomic_dec(&delayed_refs->num_entries);
6727 * we don't take a ref on the node because we're removing it from the
6728 * tree, so we just steal the ref the tree was holding.
6730 delayed_refs->num_heads--;
6731 if (head->processing == 0)
6732 delayed_refs->num_heads_ready--;
6733 head->processing = 0;
6734 spin_unlock(&head->lock);
6735 spin_unlock(&delayed_refs->lock);
6737 BUG_ON(head->extent_op);
6738 if (head->must_insert_reserved)
6741 mutex_unlock(&head->mutex);
6742 btrfs_put_delayed_ref(&head->node);
6745 spin_unlock(&head->lock);
6748 spin_unlock(&delayed_refs->lock);
6752 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6753 struct btrfs_root *root,
6754 struct extent_buffer *buf,
6755 u64 parent, int last_ref)
6760 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6761 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6762 buf->start, buf->len,
6763 parent, root->root_key.objectid,
6764 btrfs_header_level(buf),
6765 BTRFS_DROP_DELAYED_REF, NULL);
6766 BUG_ON(ret); /* -ENOMEM */
6772 if (btrfs_header_generation(buf) == trans->transid) {
6773 struct btrfs_block_group_cache *cache;
6775 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6776 ret = check_ref_cleanup(trans, root, buf->start);
6781 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6783 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6784 pin_down_extent(root, cache, buf->start, buf->len, 1);
6785 btrfs_put_block_group(cache);
6789 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6791 btrfs_add_free_space(cache, buf->start, buf->len);
6792 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6793 btrfs_put_block_group(cache);
6794 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6799 add_pinned_bytes(root->fs_info, buf->len,
6800 btrfs_header_level(buf),
6801 root->root_key.objectid);
6804 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6807 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6810 /* Can return -ENOMEM */
6811 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6812 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6813 u64 owner, u64 offset)
6816 struct btrfs_fs_info *fs_info = root->fs_info;
6818 if (btrfs_test_is_dummy_root(root))
6821 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6824 * tree log blocks never actually go into the extent allocation
6825 * tree, just update pinning info and exit early.
6827 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6828 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6829 /* unlocks the pinned mutex */
6830 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6832 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6833 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6835 parent, root_objectid, (int)owner,
6836 BTRFS_DROP_DELAYED_REF, NULL);
6838 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6840 parent, root_objectid, owner,
6842 BTRFS_DROP_DELAYED_REF, NULL);
6848 * when we wait for progress in the block group caching, its because
6849 * our allocation attempt failed at least once. So, we must sleep
6850 * and let some progress happen before we try again.
6852 * This function will sleep at least once waiting for new free space to
6853 * show up, and then it will check the block group free space numbers
6854 * for our min num_bytes. Another option is to have it go ahead
6855 * and look in the rbtree for a free extent of a given size, but this
6858 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6859 * any of the information in this block group.
6861 static noinline void
6862 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6865 struct btrfs_caching_control *caching_ctl;
6867 caching_ctl = get_caching_control(cache);
6871 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6872 (cache->free_space_ctl->free_space >= num_bytes));
6874 put_caching_control(caching_ctl);
6878 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6880 struct btrfs_caching_control *caching_ctl;
6883 caching_ctl = get_caching_control(cache);
6885 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6887 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6888 if (cache->cached == BTRFS_CACHE_ERROR)
6890 put_caching_control(caching_ctl);
6894 int __get_raid_index(u64 flags)
6896 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6897 return BTRFS_RAID_RAID10;
6898 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6899 return BTRFS_RAID_RAID1;
6900 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6901 return BTRFS_RAID_DUP;
6902 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6903 return BTRFS_RAID_RAID0;
6904 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6905 return BTRFS_RAID_RAID5;
6906 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6907 return BTRFS_RAID_RAID6;
6909 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6912 int get_block_group_index(struct btrfs_block_group_cache *cache)
6914 return __get_raid_index(cache->flags);
6917 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6918 [BTRFS_RAID_RAID10] = "raid10",
6919 [BTRFS_RAID_RAID1] = "raid1",
6920 [BTRFS_RAID_DUP] = "dup",
6921 [BTRFS_RAID_RAID0] = "raid0",
6922 [BTRFS_RAID_SINGLE] = "single",
6923 [BTRFS_RAID_RAID5] = "raid5",
6924 [BTRFS_RAID_RAID6] = "raid6",
6927 static const char *get_raid_name(enum btrfs_raid_types type)
6929 if (type >= BTRFS_NR_RAID_TYPES)
6932 return btrfs_raid_type_names[type];
6935 enum btrfs_loop_type {
6936 LOOP_CACHING_NOWAIT = 0,
6937 LOOP_CACHING_WAIT = 1,
6938 LOOP_ALLOC_CHUNK = 2,
6939 LOOP_NO_EMPTY_SIZE = 3,
6943 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6947 down_read(&cache->data_rwsem);
6951 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6954 btrfs_get_block_group(cache);
6956 down_read(&cache->data_rwsem);
6959 static struct btrfs_block_group_cache *
6960 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6961 struct btrfs_free_cluster *cluster,
6964 struct btrfs_block_group_cache *used_bg;
6965 bool locked = false;
6967 spin_lock(&cluster->refill_lock);
6969 if (used_bg == cluster->block_group)
6972 up_read(&used_bg->data_rwsem);
6973 btrfs_put_block_group(used_bg);
6976 used_bg = cluster->block_group;
6980 if (used_bg == block_group)
6983 btrfs_get_block_group(used_bg);
6988 if (down_read_trylock(&used_bg->data_rwsem))
6991 spin_unlock(&cluster->refill_lock);
6992 down_read(&used_bg->data_rwsem);
6998 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7002 up_read(&cache->data_rwsem);
7003 btrfs_put_block_group(cache);
7007 * walks the btree of allocated extents and find a hole of a given size.
7008 * The key ins is changed to record the hole:
7009 * ins->objectid == start position
7010 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7011 * ins->offset == the size of the hole.
7012 * Any available blocks before search_start are skipped.
7014 * If there is no suitable free space, we will record the max size of
7015 * the free space extent currently.
7017 static noinline int find_free_extent(struct btrfs_root *orig_root,
7018 u64 num_bytes, u64 empty_size,
7019 u64 hint_byte, struct btrfs_key *ins,
7020 u64 flags, int delalloc)
7023 struct btrfs_root *root = orig_root->fs_info->extent_root;
7024 struct btrfs_free_cluster *last_ptr = NULL;
7025 struct btrfs_block_group_cache *block_group = NULL;
7026 u64 search_start = 0;
7027 u64 max_extent_size = 0;
7028 u64 empty_cluster = 0;
7029 struct btrfs_space_info *space_info;
7031 int index = __get_raid_index(flags);
7032 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7033 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7034 bool failed_cluster_refill = false;
7035 bool failed_alloc = false;
7036 bool use_cluster = true;
7037 bool have_caching_bg = false;
7038 bool orig_have_caching_bg = false;
7039 bool full_search = false;
7041 WARN_ON(num_bytes < root->sectorsize);
7042 ins->type = BTRFS_EXTENT_ITEM_KEY;
7046 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7048 space_info = __find_space_info(root->fs_info, flags);
7050 btrfs_err(root->fs_info, "No space info for %llu", flags);
7055 * If our free space is heavily fragmented we may not be able to make
7056 * big contiguous allocations, so instead of doing the expensive search
7057 * for free space, simply return ENOSPC with our max_extent_size so we
7058 * can go ahead and search for a more manageable chunk.
7060 * If our max_extent_size is large enough for our allocation simply
7061 * disable clustering since we will likely not be able to find enough
7062 * space to create a cluster and induce latency trying.
7064 if (unlikely(space_info->max_extent_size)) {
7065 spin_lock(&space_info->lock);
7066 if (space_info->max_extent_size &&
7067 num_bytes > space_info->max_extent_size) {
7068 ins->offset = space_info->max_extent_size;
7069 spin_unlock(&space_info->lock);
7071 } else if (space_info->max_extent_size) {
7072 use_cluster = false;
7074 spin_unlock(&space_info->lock);
7077 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7079 spin_lock(&last_ptr->lock);
7080 if (last_ptr->block_group)
7081 hint_byte = last_ptr->window_start;
7082 if (last_ptr->fragmented) {
7084 * We still set window_start so we can keep track of the
7085 * last place we found an allocation to try and save
7088 hint_byte = last_ptr->window_start;
7089 use_cluster = false;
7091 spin_unlock(&last_ptr->lock);
7094 search_start = max(search_start, first_logical_byte(root, 0));
7095 search_start = max(search_start, hint_byte);
7096 if (search_start == hint_byte) {
7097 block_group = btrfs_lookup_block_group(root->fs_info,
7100 * we don't want to use the block group if it doesn't match our
7101 * allocation bits, or if its not cached.
7103 * However if we are re-searching with an ideal block group
7104 * picked out then we don't care that the block group is cached.
7106 if (block_group && block_group_bits(block_group, flags) &&
7107 block_group->cached != BTRFS_CACHE_NO) {
7108 down_read(&space_info->groups_sem);
7109 if (list_empty(&block_group->list) ||
7112 * someone is removing this block group,
7113 * we can't jump into the have_block_group
7114 * target because our list pointers are not
7117 btrfs_put_block_group(block_group);
7118 up_read(&space_info->groups_sem);
7120 index = get_block_group_index(block_group);
7121 btrfs_lock_block_group(block_group, delalloc);
7122 goto have_block_group;
7124 } else if (block_group) {
7125 btrfs_put_block_group(block_group);
7129 have_caching_bg = false;
7130 if (index == 0 || index == __get_raid_index(flags))
7132 down_read(&space_info->groups_sem);
7133 list_for_each_entry(block_group, &space_info->block_groups[index],
7138 btrfs_grab_block_group(block_group, delalloc);
7139 search_start = block_group->key.objectid;
7142 * this can happen if we end up cycling through all the
7143 * raid types, but we want to make sure we only allocate
7144 * for the proper type.
7146 if (!block_group_bits(block_group, flags)) {
7147 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7148 BTRFS_BLOCK_GROUP_RAID1 |
7149 BTRFS_BLOCK_GROUP_RAID5 |
7150 BTRFS_BLOCK_GROUP_RAID6 |
7151 BTRFS_BLOCK_GROUP_RAID10;
7154 * if they asked for extra copies and this block group
7155 * doesn't provide them, bail. This does allow us to
7156 * fill raid0 from raid1.
7158 if ((flags & extra) && !(block_group->flags & extra))
7163 cached = block_group_cache_done(block_group);
7164 if (unlikely(!cached)) {
7165 have_caching_bg = true;
7166 ret = cache_block_group(block_group, 0);
7171 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7173 if (unlikely(block_group->ro))
7177 * Ok we want to try and use the cluster allocator, so
7180 if (last_ptr && use_cluster) {
7181 struct btrfs_block_group_cache *used_block_group;
7182 unsigned long aligned_cluster;
7184 * the refill lock keeps out other
7185 * people trying to start a new cluster
7187 used_block_group = btrfs_lock_cluster(block_group,
7190 if (!used_block_group)
7191 goto refill_cluster;
7193 if (used_block_group != block_group &&
7194 (used_block_group->ro ||
7195 !block_group_bits(used_block_group, flags)))
7196 goto release_cluster;
7198 offset = btrfs_alloc_from_cluster(used_block_group,
7201 used_block_group->key.objectid,
7204 /* we have a block, we're done */
7205 spin_unlock(&last_ptr->refill_lock);
7206 trace_btrfs_reserve_extent_cluster(root,
7208 search_start, num_bytes);
7209 if (used_block_group != block_group) {
7210 btrfs_release_block_group(block_group,
7212 block_group = used_block_group;
7217 WARN_ON(last_ptr->block_group != used_block_group);
7219 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7220 * set up a new clusters, so lets just skip it
7221 * and let the allocator find whatever block
7222 * it can find. If we reach this point, we
7223 * will have tried the cluster allocator
7224 * plenty of times and not have found
7225 * anything, so we are likely way too
7226 * fragmented for the clustering stuff to find
7229 * However, if the cluster is taken from the
7230 * current block group, release the cluster
7231 * first, so that we stand a better chance of
7232 * succeeding in the unclustered
7234 if (loop >= LOOP_NO_EMPTY_SIZE &&
7235 used_block_group != block_group) {
7236 spin_unlock(&last_ptr->refill_lock);
7237 btrfs_release_block_group(used_block_group,
7239 goto unclustered_alloc;
7243 * this cluster didn't work out, free it and
7246 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7248 if (used_block_group != block_group)
7249 btrfs_release_block_group(used_block_group,
7252 if (loop >= LOOP_NO_EMPTY_SIZE) {
7253 spin_unlock(&last_ptr->refill_lock);
7254 goto unclustered_alloc;
7257 aligned_cluster = max_t(unsigned long,
7258 empty_cluster + empty_size,
7259 block_group->full_stripe_len);
7261 /* allocate a cluster in this block group */
7262 ret = btrfs_find_space_cluster(root, block_group,
7263 last_ptr, search_start,
7268 * now pull our allocation out of this
7271 offset = btrfs_alloc_from_cluster(block_group,
7277 /* we found one, proceed */
7278 spin_unlock(&last_ptr->refill_lock);
7279 trace_btrfs_reserve_extent_cluster(root,
7280 block_group, search_start,
7284 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7285 && !failed_cluster_refill) {
7286 spin_unlock(&last_ptr->refill_lock);
7288 failed_cluster_refill = true;
7289 wait_block_group_cache_progress(block_group,
7290 num_bytes + empty_cluster + empty_size);
7291 goto have_block_group;
7295 * at this point we either didn't find a cluster
7296 * or we weren't able to allocate a block from our
7297 * cluster. Free the cluster we've been trying
7298 * to use, and go to the next block group
7300 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7301 spin_unlock(&last_ptr->refill_lock);
7307 * We are doing an unclustered alloc, set the fragmented flag so
7308 * we don't bother trying to setup a cluster again until we get
7311 if (unlikely(last_ptr)) {
7312 spin_lock(&last_ptr->lock);
7313 last_ptr->fragmented = 1;
7314 spin_unlock(&last_ptr->lock);
7316 spin_lock(&block_group->free_space_ctl->tree_lock);
7318 block_group->free_space_ctl->free_space <
7319 num_bytes + empty_cluster + empty_size) {
7320 if (block_group->free_space_ctl->free_space >
7323 block_group->free_space_ctl->free_space;
7324 spin_unlock(&block_group->free_space_ctl->tree_lock);
7327 spin_unlock(&block_group->free_space_ctl->tree_lock);
7329 offset = btrfs_find_space_for_alloc(block_group, search_start,
7330 num_bytes, empty_size,
7333 * If we didn't find a chunk, and we haven't failed on this
7334 * block group before, and this block group is in the middle of
7335 * caching and we are ok with waiting, then go ahead and wait
7336 * for progress to be made, and set failed_alloc to true.
7338 * If failed_alloc is true then we've already waited on this
7339 * block group once and should move on to the next block group.
7341 if (!offset && !failed_alloc && !cached &&
7342 loop > LOOP_CACHING_NOWAIT) {
7343 wait_block_group_cache_progress(block_group,
7344 num_bytes + empty_size);
7345 failed_alloc = true;
7346 goto have_block_group;
7347 } else if (!offset) {
7351 search_start = ALIGN(offset, root->stripesize);
7353 /* move on to the next group */
7354 if (search_start + num_bytes >
7355 block_group->key.objectid + block_group->key.offset) {
7356 btrfs_add_free_space(block_group, offset, num_bytes);
7360 if (offset < search_start)
7361 btrfs_add_free_space(block_group, offset,
7362 search_start - offset);
7363 BUG_ON(offset > search_start);
7365 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7366 alloc_type, delalloc);
7367 if (ret == -EAGAIN) {
7368 btrfs_add_free_space(block_group, offset, num_bytes);
7372 /* we are all good, lets return */
7373 ins->objectid = search_start;
7374 ins->offset = num_bytes;
7376 trace_btrfs_reserve_extent(orig_root, block_group,
7377 search_start, num_bytes);
7378 btrfs_release_block_group(block_group, delalloc);
7381 failed_cluster_refill = false;
7382 failed_alloc = false;
7383 BUG_ON(index != get_block_group_index(block_group));
7384 btrfs_release_block_group(block_group, delalloc);
7386 up_read(&space_info->groups_sem);
7388 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7389 && !orig_have_caching_bg)
7390 orig_have_caching_bg = true;
7392 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7395 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7399 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7400 * caching kthreads as we move along
7401 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7402 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7403 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7406 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7408 if (loop == LOOP_CACHING_NOWAIT) {
7410 * We want to skip the LOOP_CACHING_WAIT step if we
7411 * don't have any unached bgs and we've alrelady done a
7412 * full search through.
7414 if (orig_have_caching_bg || !full_search)
7415 loop = LOOP_CACHING_WAIT;
7417 loop = LOOP_ALLOC_CHUNK;
7422 if (loop == LOOP_ALLOC_CHUNK) {
7423 struct btrfs_trans_handle *trans;
7426 trans = current->journal_info;
7430 trans = btrfs_join_transaction(root);
7432 if (IS_ERR(trans)) {
7433 ret = PTR_ERR(trans);
7437 ret = do_chunk_alloc(trans, root, flags,
7441 * If we can't allocate a new chunk we've already looped
7442 * through at least once, move on to the NO_EMPTY_SIZE
7446 loop = LOOP_NO_EMPTY_SIZE;
7449 * Do not bail out on ENOSPC since we
7450 * can do more things.
7452 if (ret < 0 && ret != -ENOSPC)
7453 btrfs_abort_transaction(trans,
7458 btrfs_end_transaction(trans, root);
7463 if (loop == LOOP_NO_EMPTY_SIZE) {
7465 * Don't loop again if we already have no empty_size and
7468 if (empty_size == 0 &&
7469 empty_cluster == 0) {
7478 } else if (!ins->objectid) {
7480 } else if (ins->objectid) {
7481 if (!use_cluster && last_ptr) {
7482 spin_lock(&last_ptr->lock);
7483 last_ptr->window_start = ins->objectid;
7484 spin_unlock(&last_ptr->lock);
7489 if (ret == -ENOSPC) {
7490 spin_lock(&space_info->lock);
7491 space_info->max_extent_size = max_extent_size;
7492 spin_unlock(&space_info->lock);
7493 ins->offset = max_extent_size;
7498 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7499 int dump_block_groups)
7501 struct btrfs_block_group_cache *cache;
7504 spin_lock(&info->lock);
7505 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7507 info->total_bytes - info->bytes_used - info->bytes_pinned -
7508 info->bytes_reserved - info->bytes_readonly,
7509 (info->full) ? "" : "not ");
7510 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7511 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7512 info->total_bytes, info->bytes_used, info->bytes_pinned,
7513 info->bytes_reserved, info->bytes_may_use,
7514 info->bytes_readonly);
7515 spin_unlock(&info->lock);
7517 if (!dump_block_groups)
7520 down_read(&info->groups_sem);
7522 list_for_each_entry(cache, &info->block_groups[index], list) {
7523 spin_lock(&cache->lock);
7524 printk(KERN_INFO "BTRFS: "
7525 "block group %llu has %llu bytes, "
7526 "%llu used %llu pinned %llu reserved %s\n",
7527 cache->key.objectid, cache->key.offset,
7528 btrfs_block_group_used(&cache->item), cache->pinned,
7529 cache->reserved, cache->ro ? "[readonly]" : "");
7530 btrfs_dump_free_space(cache, bytes);
7531 spin_unlock(&cache->lock);
7533 if (++index < BTRFS_NR_RAID_TYPES)
7535 up_read(&info->groups_sem);
7538 int btrfs_reserve_extent(struct btrfs_root *root,
7539 u64 num_bytes, u64 min_alloc_size,
7540 u64 empty_size, u64 hint_byte,
7541 struct btrfs_key *ins, int is_data, int delalloc)
7543 bool final_tried = num_bytes == min_alloc_size;
7547 flags = btrfs_get_alloc_profile(root, is_data);
7549 WARN_ON(num_bytes < root->sectorsize);
7550 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7553 if (ret == -ENOSPC) {
7554 if (!final_tried && ins->offset) {
7555 num_bytes = min(num_bytes >> 1, ins->offset);
7556 num_bytes = round_down(num_bytes, root->sectorsize);
7557 num_bytes = max(num_bytes, min_alloc_size);
7558 if (num_bytes == min_alloc_size)
7561 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7562 struct btrfs_space_info *sinfo;
7564 sinfo = __find_space_info(root->fs_info, flags);
7565 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7568 dump_space_info(sinfo, num_bytes, 1);
7575 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7577 int pin, int delalloc)
7579 struct btrfs_block_group_cache *cache;
7582 cache = btrfs_lookup_block_group(root->fs_info, start);
7584 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7590 pin_down_extent(root, cache, start, len, 1);
7592 if (btrfs_test_opt(root, DISCARD))
7593 ret = btrfs_discard_extent(root, start, len, NULL);
7594 btrfs_add_free_space(cache, start, len);
7595 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7598 btrfs_put_block_group(cache);
7600 trace_btrfs_reserved_extent_free(root, start, len);
7605 int btrfs_free_reserved_extent(struct btrfs_root *root,
7606 u64 start, u64 len, int delalloc)
7608 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7611 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7614 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7617 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7618 struct btrfs_root *root,
7619 u64 parent, u64 root_objectid,
7620 u64 flags, u64 owner, u64 offset,
7621 struct btrfs_key *ins, int ref_mod)
7624 struct btrfs_fs_info *fs_info = root->fs_info;
7625 struct btrfs_extent_item *extent_item;
7626 struct btrfs_extent_inline_ref *iref;
7627 struct btrfs_path *path;
7628 struct extent_buffer *leaf;
7633 type = BTRFS_SHARED_DATA_REF_KEY;
7635 type = BTRFS_EXTENT_DATA_REF_KEY;
7637 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7639 path = btrfs_alloc_path();
7643 path->leave_spinning = 1;
7644 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7647 btrfs_free_path(path);
7651 leaf = path->nodes[0];
7652 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7653 struct btrfs_extent_item);
7654 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7655 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7656 btrfs_set_extent_flags(leaf, extent_item,
7657 flags | BTRFS_EXTENT_FLAG_DATA);
7659 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7660 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7662 struct btrfs_shared_data_ref *ref;
7663 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7664 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7665 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7667 struct btrfs_extent_data_ref *ref;
7668 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7669 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7670 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7671 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7672 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7675 btrfs_mark_buffer_dirty(path->nodes[0]);
7676 btrfs_free_path(path);
7678 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7679 if (ret) { /* -ENOENT, logic error */
7680 btrfs_err(fs_info, "update block group failed for %llu %llu",
7681 ins->objectid, ins->offset);
7684 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7688 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7689 struct btrfs_root *root,
7690 u64 parent, u64 root_objectid,
7691 u64 flags, struct btrfs_disk_key *key,
7692 int level, struct btrfs_key *ins)
7695 struct btrfs_fs_info *fs_info = root->fs_info;
7696 struct btrfs_extent_item *extent_item;
7697 struct btrfs_tree_block_info *block_info;
7698 struct btrfs_extent_inline_ref *iref;
7699 struct btrfs_path *path;
7700 struct extent_buffer *leaf;
7701 u32 size = sizeof(*extent_item) + sizeof(*iref);
7702 u64 num_bytes = ins->offset;
7703 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7706 if (!skinny_metadata)
7707 size += sizeof(*block_info);
7709 path = btrfs_alloc_path();
7711 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7716 path->leave_spinning = 1;
7717 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7720 btrfs_free_path(path);
7721 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7726 leaf = path->nodes[0];
7727 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7728 struct btrfs_extent_item);
7729 btrfs_set_extent_refs(leaf, extent_item, 1);
7730 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7731 btrfs_set_extent_flags(leaf, extent_item,
7732 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7734 if (skinny_metadata) {
7735 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7736 num_bytes = root->nodesize;
7738 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7739 btrfs_set_tree_block_key(leaf, block_info, key);
7740 btrfs_set_tree_block_level(leaf, block_info, level);
7741 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7745 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7746 btrfs_set_extent_inline_ref_type(leaf, iref,
7747 BTRFS_SHARED_BLOCK_REF_KEY);
7748 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7750 btrfs_set_extent_inline_ref_type(leaf, iref,
7751 BTRFS_TREE_BLOCK_REF_KEY);
7752 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7755 btrfs_mark_buffer_dirty(leaf);
7756 btrfs_free_path(path);
7758 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7760 if (ret) { /* -ENOENT, logic error */
7761 btrfs_err(fs_info, "update block group failed for %llu %llu",
7762 ins->objectid, ins->offset);
7766 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7770 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7771 struct btrfs_root *root,
7772 u64 root_objectid, u64 owner,
7773 u64 offset, u64 ram_bytes,
7774 struct btrfs_key *ins)
7778 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7780 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7782 root_objectid, owner, offset,
7783 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
7789 * this is used by the tree logging recovery code. It records that
7790 * an extent has been allocated and makes sure to clear the free
7791 * space cache bits as well
7793 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7794 struct btrfs_root *root,
7795 u64 root_objectid, u64 owner, u64 offset,
7796 struct btrfs_key *ins)
7799 struct btrfs_block_group_cache *block_group;
7802 * Mixed block groups will exclude before processing the log so we only
7803 * need to do the exlude dance if this fs isn't mixed.
7805 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7806 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7811 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7815 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7816 RESERVE_ALLOC_NO_ACCOUNT, 0);
7817 BUG_ON(ret); /* logic error */
7818 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7819 0, owner, offset, ins, 1);
7820 btrfs_put_block_group(block_group);
7824 static struct extent_buffer *
7825 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7826 u64 bytenr, int level)
7828 struct extent_buffer *buf;
7830 buf = btrfs_find_create_tree_block(root, bytenr);
7832 return ERR_PTR(-ENOMEM);
7833 btrfs_set_header_generation(buf, trans->transid);
7834 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7835 btrfs_tree_lock(buf);
7836 clean_tree_block(trans, root->fs_info, buf);
7837 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7839 btrfs_set_lock_blocking(buf);
7840 btrfs_set_buffer_uptodate(buf);
7842 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7843 buf->log_index = root->log_transid % 2;
7845 * we allow two log transactions at a time, use different
7846 * EXENT bit to differentiate dirty pages.
7848 if (buf->log_index == 0)
7849 set_extent_dirty(&root->dirty_log_pages, buf->start,
7850 buf->start + buf->len - 1, GFP_NOFS);
7852 set_extent_new(&root->dirty_log_pages, buf->start,
7853 buf->start + buf->len - 1, GFP_NOFS);
7855 buf->log_index = -1;
7856 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7857 buf->start + buf->len - 1, GFP_NOFS);
7859 trans->blocks_used++;
7860 /* this returns a buffer locked for blocking */
7864 static struct btrfs_block_rsv *
7865 use_block_rsv(struct btrfs_trans_handle *trans,
7866 struct btrfs_root *root, u32 blocksize)
7868 struct btrfs_block_rsv *block_rsv;
7869 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7871 bool global_updated = false;
7873 block_rsv = get_block_rsv(trans, root);
7875 if (unlikely(block_rsv->size == 0))
7878 ret = block_rsv_use_bytes(block_rsv, blocksize);
7882 if (block_rsv->failfast)
7883 return ERR_PTR(ret);
7885 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7886 global_updated = true;
7887 update_global_block_rsv(root->fs_info);
7891 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7892 static DEFINE_RATELIMIT_STATE(_rs,
7893 DEFAULT_RATELIMIT_INTERVAL * 10,
7894 /*DEFAULT_RATELIMIT_BURST*/ 1);
7895 if (__ratelimit(&_rs))
7897 "BTRFS: block rsv returned %d\n", ret);
7900 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7901 BTRFS_RESERVE_NO_FLUSH);
7905 * If we couldn't reserve metadata bytes try and use some from
7906 * the global reserve if its space type is the same as the global
7909 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7910 block_rsv->space_info == global_rsv->space_info) {
7911 ret = block_rsv_use_bytes(global_rsv, blocksize);
7915 return ERR_PTR(ret);
7918 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7919 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7921 block_rsv_add_bytes(block_rsv, blocksize, 0);
7922 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7926 * finds a free extent and does all the dirty work required for allocation
7927 * returns the tree buffer or an ERR_PTR on error.
7929 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7930 struct btrfs_root *root,
7931 u64 parent, u64 root_objectid,
7932 struct btrfs_disk_key *key, int level,
7933 u64 hint, u64 empty_size)
7935 struct btrfs_key ins;
7936 struct btrfs_block_rsv *block_rsv;
7937 struct extent_buffer *buf;
7938 struct btrfs_delayed_extent_op *extent_op;
7941 u32 blocksize = root->nodesize;
7942 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7945 if (btrfs_test_is_dummy_root(root)) {
7946 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7949 root->alloc_bytenr += blocksize;
7953 block_rsv = use_block_rsv(trans, root, blocksize);
7954 if (IS_ERR(block_rsv))
7955 return ERR_CAST(block_rsv);
7957 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7958 empty_size, hint, &ins, 0, 0);
7962 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7965 goto out_free_reserved;
7968 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7970 parent = ins.objectid;
7971 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7975 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7976 extent_op = btrfs_alloc_delayed_extent_op();
7982 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7984 memset(&extent_op->key, 0, sizeof(extent_op->key));
7985 extent_op->flags_to_set = flags;
7986 if (skinny_metadata)
7987 extent_op->update_key = 0;
7989 extent_op->update_key = 1;
7990 extent_op->update_flags = 1;
7991 extent_op->is_data = 0;
7992 extent_op->level = level;
7994 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7995 ins.objectid, ins.offset,
7996 parent, root_objectid, level,
7997 BTRFS_ADD_DELAYED_EXTENT,
8000 goto out_free_delayed;
8005 btrfs_free_delayed_extent_op(extent_op);
8007 free_extent_buffer(buf);
8009 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8011 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8012 return ERR_PTR(ret);
8015 struct walk_control {
8016 u64 refs[BTRFS_MAX_LEVEL];
8017 u64 flags[BTRFS_MAX_LEVEL];
8018 struct btrfs_key update_progress;
8029 #define DROP_REFERENCE 1
8030 #define UPDATE_BACKREF 2
8032 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8033 struct btrfs_root *root,
8034 struct walk_control *wc,
8035 struct btrfs_path *path)
8043 struct btrfs_key key;
8044 struct extent_buffer *eb;
8049 if (path->slots[wc->level] < wc->reada_slot) {
8050 wc->reada_count = wc->reada_count * 2 / 3;
8051 wc->reada_count = max(wc->reada_count, 2);
8053 wc->reada_count = wc->reada_count * 3 / 2;
8054 wc->reada_count = min_t(int, wc->reada_count,
8055 BTRFS_NODEPTRS_PER_BLOCK(root));
8058 eb = path->nodes[wc->level];
8059 nritems = btrfs_header_nritems(eb);
8060 blocksize = root->nodesize;
8062 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8063 if (nread >= wc->reada_count)
8067 bytenr = btrfs_node_blockptr(eb, slot);
8068 generation = btrfs_node_ptr_generation(eb, slot);
8070 if (slot == path->slots[wc->level])
8073 if (wc->stage == UPDATE_BACKREF &&
8074 generation <= root->root_key.offset)
8077 /* We don't lock the tree block, it's OK to be racy here */
8078 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8079 wc->level - 1, 1, &refs,
8081 /* We don't care about errors in readahead. */
8086 if (wc->stage == DROP_REFERENCE) {
8090 if (wc->level == 1 &&
8091 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8093 if (!wc->update_ref ||
8094 generation <= root->root_key.offset)
8096 btrfs_node_key_to_cpu(eb, &key, slot);
8097 ret = btrfs_comp_cpu_keys(&key,
8098 &wc->update_progress);
8102 if (wc->level == 1 &&
8103 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8107 readahead_tree_block(root, bytenr);
8110 wc->reada_slot = slot;
8114 * These may not be seen by the usual inc/dec ref code so we have to
8117 static int record_one_subtree_extent(struct btrfs_trans_handle *trans,
8118 struct btrfs_root *root, u64 bytenr,
8121 struct btrfs_qgroup_extent_record *qrecord;
8122 struct btrfs_delayed_ref_root *delayed_refs;
8124 qrecord = kmalloc(sizeof(*qrecord), GFP_NOFS);
8128 qrecord->bytenr = bytenr;
8129 qrecord->num_bytes = num_bytes;
8130 qrecord->old_roots = NULL;
8132 delayed_refs = &trans->transaction->delayed_refs;
8133 spin_lock(&delayed_refs->lock);
8134 if (btrfs_qgroup_insert_dirty_extent(delayed_refs, qrecord))
8136 spin_unlock(&delayed_refs->lock);
8141 static int account_leaf_items(struct btrfs_trans_handle *trans,
8142 struct btrfs_root *root,
8143 struct extent_buffer *eb)
8145 int nr = btrfs_header_nritems(eb);
8146 int i, extent_type, ret;
8147 struct btrfs_key key;
8148 struct btrfs_file_extent_item *fi;
8149 u64 bytenr, num_bytes;
8151 /* We can be called directly from walk_up_proc() */
8152 if (!root->fs_info->quota_enabled)
8155 for (i = 0; i < nr; i++) {
8156 btrfs_item_key_to_cpu(eb, &key, i);
8158 if (key.type != BTRFS_EXTENT_DATA_KEY)
8161 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8162 /* filter out non qgroup-accountable extents */
8163 extent_type = btrfs_file_extent_type(eb, fi);
8165 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8168 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8172 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8174 ret = record_one_subtree_extent(trans, root, bytenr, num_bytes);
8182 * Walk up the tree from the bottom, freeing leaves and any interior
8183 * nodes which have had all slots visited. If a node (leaf or
8184 * interior) is freed, the node above it will have it's slot
8185 * incremented. The root node will never be freed.
8187 * At the end of this function, we should have a path which has all
8188 * slots incremented to the next position for a search. If we need to
8189 * read a new node it will be NULL and the node above it will have the
8190 * correct slot selected for a later read.
8192 * If we increment the root nodes slot counter past the number of
8193 * elements, 1 is returned to signal completion of the search.
8195 static int adjust_slots_upwards(struct btrfs_root *root,
8196 struct btrfs_path *path, int root_level)
8200 struct extent_buffer *eb;
8202 if (root_level == 0)
8205 while (level <= root_level) {
8206 eb = path->nodes[level];
8207 nr = btrfs_header_nritems(eb);
8208 path->slots[level]++;
8209 slot = path->slots[level];
8210 if (slot >= nr || level == 0) {
8212 * Don't free the root - we will detect this
8213 * condition after our loop and return a
8214 * positive value for caller to stop walking the tree.
8216 if (level != root_level) {
8217 btrfs_tree_unlock_rw(eb, path->locks[level]);
8218 path->locks[level] = 0;
8220 free_extent_buffer(eb);
8221 path->nodes[level] = NULL;
8222 path->slots[level] = 0;
8226 * We have a valid slot to walk back down
8227 * from. Stop here so caller can process these
8236 eb = path->nodes[root_level];
8237 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8244 * root_eb is the subtree root and is locked before this function is called.
8246 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8247 struct btrfs_root *root,
8248 struct extent_buffer *root_eb,
8254 struct extent_buffer *eb = root_eb;
8255 struct btrfs_path *path = NULL;
8257 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8258 BUG_ON(root_eb == NULL);
8260 if (!root->fs_info->quota_enabled)
8263 if (!extent_buffer_uptodate(root_eb)) {
8264 ret = btrfs_read_buffer(root_eb, root_gen);
8269 if (root_level == 0) {
8270 ret = account_leaf_items(trans, root, root_eb);
8274 path = btrfs_alloc_path();
8279 * Walk down the tree. Missing extent blocks are filled in as
8280 * we go. Metadata is accounted every time we read a new
8283 * When we reach a leaf, we account for file extent items in it,
8284 * walk back up the tree (adjusting slot pointers as we go)
8285 * and restart the search process.
8287 extent_buffer_get(root_eb); /* For path */
8288 path->nodes[root_level] = root_eb;
8289 path->slots[root_level] = 0;
8290 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8293 while (level >= 0) {
8294 if (path->nodes[level] == NULL) {
8299 /* We need to get child blockptr/gen from
8300 * parent before we can read it. */
8301 eb = path->nodes[level + 1];
8302 parent_slot = path->slots[level + 1];
8303 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8304 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8306 eb = read_tree_block(root, child_bytenr, child_gen);
8310 } else if (!extent_buffer_uptodate(eb)) {
8311 free_extent_buffer(eb);
8316 path->nodes[level] = eb;
8317 path->slots[level] = 0;
8319 btrfs_tree_read_lock(eb);
8320 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8321 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8323 ret = record_one_subtree_extent(trans, root, child_bytenr,
8330 ret = account_leaf_items(trans, root, path->nodes[level]);
8334 /* Nonzero return here means we completed our search */
8335 ret = adjust_slots_upwards(root, path, root_level);
8339 /* Restart search with new slots */
8348 btrfs_free_path(path);
8354 * helper to process tree block while walking down the tree.
8356 * when wc->stage == UPDATE_BACKREF, this function updates
8357 * back refs for pointers in the block.
8359 * NOTE: return value 1 means we should stop walking down.
8361 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8362 struct btrfs_root *root,
8363 struct btrfs_path *path,
8364 struct walk_control *wc, int lookup_info)
8366 int level = wc->level;
8367 struct extent_buffer *eb = path->nodes[level];
8368 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8371 if (wc->stage == UPDATE_BACKREF &&
8372 btrfs_header_owner(eb) != root->root_key.objectid)
8376 * when reference count of tree block is 1, it won't increase
8377 * again. once full backref flag is set, we never clear it.
8380 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8381 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8382 BUG_ON(!path->locks[level]);
8383 ret = btrfs_lookup_extent_info(trans, root,
8384 eb->start, level, 1,
8387 BUG_ON(ret == -ENOMEM);
8390 BUG_ON(wc->refs[level] == 0);
8393 if (wc->stage == DROP_REFERENCE) {
8394 if (wc->refs[level] > 1)
8397 if (path->locks[level] && !wc->keep_locks) {
8398 btrfs_tree_unlock_rw(eb, path->locks[level]);
8399 path->locks[level] = 0;
8404 /* wc->stage == UPDATE_BACKREF */
8405 if (!(wc->flags[level] & flag)) {
8406 BUG_ON(!path->locks[level]);
8407 ret = btrfs_inc_ref(trans, root, eb, 1);
8408 BUG_ON(ret); /* -ENOMEM */
8409 ret = btrfs_dec_ref(trans, root, eb, 0);
8410 BUG_ON(ret); /* -ENOMEM */
8411 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8413 btrfs_header_level(eb), 0);
8414 BUG_ON(ret); /* -ENOMEM */
8415 wc->flags[level] |= flag;
8419 * the block is shared by multiple trees, so it's not good to
8420 * keep the tree lock
8422 if (path->locks[level] && level > 0) {
8423 btrfs_tree_unlock_rw(eb, path->locks[level]);
8424 path->locks[level] = 0;
8430 * helper to process tree block pointer.
8432 * when wc->stage == DROP_REFERENCE, this function checks
8433 * reference count of the block pointed to. if the block
8434 * is shared and we need update back refs for the subtree
8435 * rooted at the block, this function changes wc->stage to
8436 * UPDATE_BACKREF. if the block is shared and there is no
8437 * need to update back, this function drops the reference
8440 * NOTE: return value 1 means we should stop walking down.
8442 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8443 struct btrfs_root *root,
8444 struct btrfs_path *path,
8445 struct walk_control *wc, int *lookup_info)
8451 struct btrfs_key key;
8452 struct extent_buffer *next;
8453 int level = wc->level;
8456 bool need_account = false;
8458 generation = btrfs_node_ptr_generation(path->nodes[level],
8459 path->slots[level]);
8461 * if the lower level block was created before the snapshot
8462 * was created, we know there is no need to update back refs
8465 if (wc->stage == UPDATE_BACKREF &&
8466 generation <= root->root_key.offset) {
8471 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8472 blocksize = root->nodesize;
8474 next = btrfs_find_tree_block(root->fs_info, bytenr);
8476 next = btrfs_find_create_tree_block(root, bytenr);
8479 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8483 btrfs_tree_lock(next);
8484 btrfs_set_lock_blocking(next);
8486 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8487 &wc->refs[level - 1],
8488 &wc->flags[level - 1]);
8490 btrfs_tree_unlock(next);
8494 if (unlikely(wc->refs[level - 1] == 0)) {
8495 btrfs_err(root->fs_info, "Missing references.");
8500 if (wc->stage == DROP_REFERENCE) {
8501 if (wc->refs[level - 1] > 1) {
8502 need_account = true;
8504 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8507 if (!wc->update_ref ||
8508 generation <= root->root_key.offset)
8511 btrfs_node_key_to_cpu(path->nodes[level], &key,
8512 path->slots[level]);
8513 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8517 wc->stage = UPDATE_BACKREF;
8518 wc->shared_level = level - 1;
8522 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8526 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8527 btrfs_tree_unlock(next);
8528 free_extent_buffer(next);
8534 if (reada && level == 1)
8535 reada_walk_down(trans, root, wc, path);
8536 next = read_tree_block(root, bytenr, generation);
8538 return PTR_ERR(next);
8539 } else if (!extent_buffer_uptodate(next)) {
8540 free_extent_buffer(next);
8543 btrfs_tree_lock(next);
8544 btrfs_set_lock_blocking(next);
8548 BUG_ON(level != btrfs_header_level(next));
8549 path->nodes[level] = next;
8550 path->slots[level] = 0;
8551 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8557 wc->refs[level - 1] = 0;
8558 wc->flags[level - 1] = 0;
8559 if (wc->stage == DROP_REFERENCE) {
8560 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8561 parent = path->nodes[level]->start;
8563 BUG_ON(root->root_key.objectid !=
8564 btrfs_header_owner(path->nodes[level]));
8569 ret = account_shared_subtree(trans, root, next,
8570 generation, level - 1);
8572 btrfs_err_rl(root->fs_info,
8574 "%d accounting shared subtree. Quota "
8575 "is out of sync, rescan required.",
8579 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8580 root->root_key.objectid, level - 1, 0);
8581 BUG_ON(ret); /* -ENOMEM */
8583 btrfs_tree_unlock(next);
8584 free_extent_buffer(next);
8590 * helper to process tree block while walking up the tree.
8592 * when wc->stage == DROP_REFERENCE, this function drops
8593 * reference count on the block.
8595 * when wc->stage == UPDATE_BACKREF, this function changes
8596 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8597 * to UPDATE_BACKREF previously while processing the block.
8599 * NOTE: return value 1 means we should stop walking up.
8601 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8602 struct btrfs_root *root,
8603 struct btrfs_path *path,
8604 struct walk_control *wc)
8607 int level = wc->level;
8608 struct extent_buffer *eb = path->nodes[level];
8611 if (wc->stage == UPDATE_BACKREF) {
8612 BUG_ON(wc->shared_level < level);
8613 if (level < wc->shared_level)
8616 ret = find_next_key(path, level + 1, &wc->update_progress);
8620 wc->stage = DROP_REFERENCE;
8621 wc->shared_level = -1;
8622 path->slots[level] = 0;
8625 * check reference count again if the block isn't locked.
8626 * we should start walking down the tree again if reference
8629 if (!path->locks[level]) {
8631 btrfs_tree_lock(eb);
8632 btrfs_set_lock_blocking(eb);
8633 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8635 ret = btrfs_lookup_extent_info(trans, root,
8636 eb->start, level, 1,
8640 btrfs_tree_unlock_rw(eb, path->locks[level]);
8641 path->locks[level] = 0;
8644 BUG_ON(wc->refs[level] == 0);
8645 if (wc->refs[level] == 1) {
8646 btrfs_tree_unlock_rw(eb, path->locks[level]);
8647 path->locks[level] = 0;
8653 /* wc->stage == DROP_REFERENCE */
8654 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8656 if (wc->refs[level] == 1) {
8658 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8659 ret = btrfs_dec_ref(trans, root, eb, 1);
8661 ret = btrfs_dec_ref(trans, root, eb, 0);
8662 BUG_ON(ret); /* -ENOMEM */
8663 ret = account_leaf_items(trans, root, eb);
8665 btrfs_err_rl(root->fs_info,
8667 "%d accounting leaf items. Quota "
8668 "is out of sync, rescan required.",
8672 /* make block locked assertion in clean_tree_block happy */
8673 if (!path->locks[level] &&
8674 btrfs_header_generation(eb) == trans->transid) {
8675 btrfs_tree_lock(eb);
8676 btrfs_set_lock_blocking(eb);
8677 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8679 clean_tree_block(trans, root->fs_info, eb);
8682 if (eb == root->node) {
8683 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8686 BUG_ON(root->root_key.objectid !=
8687 btrfs_header_owner(eb));
8689 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8690 parent = path->nodes[level + 1]->start;
8692 BUG_ON(root->root_key.objectid !=
8693 btrfs_header_owner(path->nodes[level + 1]));
8696 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8698 wc->refs[level] = 0;
8699 wc->flags[level] = 0;
8703 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8704 struct btrfs_root *root,
8705 struct btrfs_path *path,
8706 struct walk_control *wc)
8708 int level = wc->level;
8709 int lookup_info = 1;
8712 while (level >= 0) {
8713 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8720 if (path->slots[level] >=
8721 btrfs_header_nritems(path->nodes[level]))
8724 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8726 path->slots[level]++;
8735 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8736 struct btrfs_root *root,
8737 struct btrfs_path *path,
8738 struct walk_control *wc, int max_level)
8740 int level = wc->level;
8743 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8744 while (level < max_level && path->nodes[level]) {
8746 if (path->slots[level] + 1 <
8747 btrfs_header_nritems(path->nodes[level])) {
8748 path->slots[level]++;
8751 ret = walk_up_proc(trans, root, path, wc);
8755 if (path->locks[level]) {
8756 btrfs_tree_unlock_rw(path->nodes[level],
8757 path->locks[level]);
8758 path->locks[level] = 0;
8760 free_extent_buffer(path->nodes[level]);
8761 path->nodes[level] = NULL;
8769 * drop a subvolume tree.
8771 * this function traverses the tree freeing any blocks that only
8772 * referenced by the tree.
8774 * when a shared tree block is found. this function decreases its
8775 * reference count by one. if update_ref is true, this function
8776 * also make sure backrefs for the shared block and all lower level
8777 * blocks are properly updated.
8779 * If called with for_reloc == 0, may exit early with -EAGAIN
8781 int btrfs_drop_snapshot(struct btrfs_root *root,
8782 struct btrfs_block_rsv *block_rsv, int update_ref,
8785 struct btrfs_path *path;
8786 struct btrfs_trans_handle *trans;
8787 struct btrfs_root *tree_root = root->fs_info->tree_root;
8788 struct btrfs_root_item *root_item = &root->root_item;
8789 struct walk_control *wc;
8790 struct btrfs_key key;
8794 bool root_dropped = false;
8796 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8798 path = btrfs_alloc_path();
8804 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8806 btrfs_free_path(path);
8811 trans = btrfs_start_transaction(tree_root, 0);
8812 if (IS_ERR(trans)) {
8813 err = PTR_ERR(trans);
8818 trans->block_rsv = block_rsv;
8820 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8821 level = btrfs_header_level(root->node);
8822 path->nodes[level] = btrfs_lock_root_node(root);
8823 btrfs_set_lock_blocking(path->nodes[level]);
8824 path->slots[level] = 0;
8825 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8826 memset(&wc->update_progress, 0,
8827 sizeof(wc->update_progress));
8829 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8830 memcpy(&wc->update_progress, &key,
8831 sizeof(wc->update_progress));
8833 level = root_item->drop_level;
8835 path->lowest_level = level;
8836 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8837 path->lowest_level = 0;
8845 * unlock our path, this is safe because only this
8846 * function is allowed to delete this snapshot
8848 btrfs_unlock_up_safe(path, 0);
8850 level = btrfs_header_level(root->node);
8852 btrfs_tree_lock(path->nodes[level]);
8853 btrfs_set_lock_blocking(path->nodes[level]);
8854 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8856 ret = btrfs_lookup_extent_info(trans, root,
8857 path->nodes[level]->start,
8858 level, 1, &wc->refs[level],
8864 BUG_ON(wc->refs[level] == 0);
8866 if (level == root_item->drop_level)
8869 btrfs_tree_unlock(path->nodes[level]);
8870 path->locks[level] = 0;
8871 WARN_ON(wc->refs[level] != 1);
8877 wc->shared_level = -1;
8878 wc->stage = DROP_REFERENCE;
8879 wc->update_ref = update_ref;
8881 wc->for_reloc = for_reloc;
8882 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8886 ret = walk_down_tree(trans, root, path, wc);
8892 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8899 BUG_ON(wc->stage != DROP_REFERENCE);
8903 if (wc->stage == DROP_REFERENCE) {
8905 btrfs_node_key(path->nodes[level],
8906 &root_item->drop_progress,
8907 path->slots[level]);
8908 root_item->drop_level = level;
8911 BUG_ON(wc->level == 0);
8912 if (btrfs_should_end_transaction(trans, tree_root) ||
8913 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8914 ret = btrfs_update_root(trans, tree_root,
8918 btrfs_abort_transaction(trans, tree_root, ret);
8923 btrfs_end_transaction_throttle(trans, tree_root);
8924 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8925 pr_debug("BTRFS: drop snapshot early exit\n");
8930 trans = btrfs_start_transaction(tree_root, 0);
8931 if (IS_ERR(trans)) {
8932 err = PTR_ERR(trans);
8936 trans->block_rsv = block_rsv;
8939 btrfs_release_path(path);
8943 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8945 btrfs_abort_transaction(trans, tree_root, ret);
8949 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8950 ret = btrfs_find_root(tree_root, &root->root_key, path,
8953 btrfs_abort_transaction(trans, tree_root, ret);
8956 } else if (ret > 0) {
8957 /* if we fail to delete the orphan item this time
8958 * around, it'll get picked up the next time.
8960 * The most common failure here is just -ENOENT.
8962 btrfs_del_orphan_item(trans, tree_root,
8963 root->root_key.objectid);
8967 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8968 btrfs_add_dropped_root(trans, root);
8970 free_extent_buffer(root->node);
8971 free_extent_buffer(root->commit_root);
8972 btrfs_put_fs_root(root);
8974 root_dropped = true;
8976 btrfs_end_transaction_throttle(trans, tree_root);
8979 btrfs_free_path(path);
8982 * So if we need to stop dropping the snapshot for whatever reason we
8983 * need to make sure to add it back to the dead root list so that we
8984 * keep trying to do the work later. This also cleans up roots if we
8985 * don't have it in the radix (like when we recover after a power fail
8986 * or unmount) so we don't leak memory.
8988 if (!for_reloc && root_dropped == false)
8989 btrfs_add_dead_root(root);
8990 if (err && err != -EAGAIN)
8991 btrfs_std_error(root->fs_info, err, NULL);
8996 * drop subtree rooted at tree block 'node'.
8998 * NOTE: this function will unlock and release tree block 'node'
8999 * only used by relocation code
9001 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
9002 struct btrfs_root *root,
9003 struct extent_buffer *node,
9004 struct extent_buffer *parent)
9006 struct btrfs_path *path;
9007 struct walk_control *wc;
9013 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
9015 path = btrfs_alloc_path();
9019 wc = kzalloc(sizeof(*wc), GFP_NOFS);
9021 btrfs_free_path(path);
9025 btrfs_assert_tree_locked(parent);
9026 parent_level = btrfs_header_level(parent);
9027 extent_buffer_get(parent);
9028 path->nodes[parent_level] = parent;
9029 path->slots[parent_level] = btrfs_header_nritems(parent);
9031 btrfs_assert_tree_locked(node);
9032 level = btrfs_header_level(node);
9033 path->nodes[level] = node;
9034 path->slots[level] = 0;
9035 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9037 wc->refs[parent_level] = 1;
9038 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9040 wc->shared_level = -1;
9041 wc->stage = DROP_REFERENCE;
9045 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9048 wret = walk_down_tree(trans, root, path, wc);
9054 wret = walk_up_tree(trans, root, path, wc, parent_level);
9062 btrfs_free_path(path);
9066 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9072 * if restripe for this chunk_type is on pick target profile and
9073 * return, otherwise do the usual balance
9075 stripped = get_restripe_target(root->fs_info, flags);
9077 return extended_to_chunk(stripped);
9079 num_devices = root->fs_info->fs_devices->rw_devices;
9081 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9082 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9083 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9085 if (num_devices == 1) {
9086 stripped |= BTRFS_BLOCK_GROUP_DUP;
9087 stripped = flags & ~stripped;
9089 /* turn raid0 into single device chunks */
9090 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9093 /* turn mirroring into duplication */
9094 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9095 BTRFS_BLOCK_GROUP_RAID10))
9096 return stripped | BTRFS_BLOCK_GROUP_DUP;
9098 /* they already had raid on here, just return */
9099 if (flags & stripped)
9102 stripped |= BTRFS_BLOCK_GROUP_DUP;
9103 stripped = flags & ~stripped;
9105 /* switch duplicated blocks with raid1 */
9106 if (flags & BTRFS_BLOCK_GROUP_DUP)
9107 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9109 /* this is drive concat, leave it alone */
9115 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9117 struct btrfs_space_info *sinfo = cache->space_info;
9119 u64 min_allocable_bytes;
9123 * We need some metadata space and system metadata space for
9124 * allocating chunks in some corner cases until we force to set
9125 * it to be readonly.
9128 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9130 min_allocable_bytes = 1 * 1024 * 1024;
9132 min_allocable_bytes = 0;
9134 spin_lock(&sinfo->lock);
9135 spin_lock(&cache->lock);
9143 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9144 cache->bytes_super - btrfs_block_group_used(&cache->item);
9146 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9147 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9148 min_allocable_bytes <= sinfo->total_bytes) {
9149 sinfo->bytes_readonly += num_bytes;
9151 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9155 spin_unlock(&cache->lock);
9156 spin_unlock(&sinfo->lock);
9160 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9161 struct btrfs_block_group_cache *cache)
9164 struct btrfs_trans_handle *trans;
9169 trans = btrfs_join_transaction(root);
9171 return PTR_ERR(trans);
9174 * we're not allowed to set block groups readonly after the dirty
9175 * block groups cache has started writing. If it already started,
9176 * back off and let this transaction commit
9178 mutex_lock(&root->fs_info->ro_block_group_mutex);
9179 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9180 u64 transid = trans->transid;
9182 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9183 btrfs_end_transaction(trans, root);
9185 ret = btrfs_wait_for_commit(root, transid);
9192 * if we are changing raid levels, try to allocate a corresponding
9193 * block group with the new raid level.
9195 alloc_flags = update_block_group_flags(root, cache->flags);
9196 if (alloc_flags != cache->flags) {
9197 ret = do_chunk_alloc(trans, root, alloc_flags,
9200 * ENOSPC is allowed here, we may have enough space
9201 * already allocated at the new raid level to
9210 ret = inc_block_group_ro(cache, 0);
9213 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9214 ret = do_chunk_alloc(trans, root, alloc_flags,
9218 ret = inc_block_group_ro(cache, 0);
9220 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9221 alloc_flags = update_block_group_flags(root, cache->flags);
9222 lock_chunks(root->fs_info->chunk_root);
9223 check_system_chunk(trans, root, alloc_flags);
9224 unlock_chunks(root->fs_info->chunk_root);
9226 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9228 btrfs_end_transaction(trans, root);
9232 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9233 struct btrfs_root *root, u64 type)
9235 u64 alloc_flags = get_alloc_profile(root, type);
9236 return do_chunk_alloc(trans, root, alloc_flags,
9241 * helper to account the unused space of all the readonly block group in the
9242 * space_info. takes mirrors into account.
9244 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9246 struct btrfs_block_group_cache *block_group;
9250 /* It's df, we don't care if it's racey */
9251 if (list_empty(&sinfo->ro_bgs))
9254 spin_lock(&sinfo->lock);
9255 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9256 spin_lock(&block_group->lock);
9258 if (!block_group->ro) {
9259 spin_unlock(&block_group->lock);
9263 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9264 BTRFS_BLOCK_GROUP_RAID10 |
9265 BTRFS_BLOCK_GROUP_DUP))
9270 free_bytes += (block_group->key.offset -
9271 btrfs_block_group_used(&block_group->item)) *
9274 spin_unlock(&block_group->lock);
9276 spin_unlock(&sinfo->lock);
9281 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9282 struct btrfs_block_group_cache *cache)
9284 struct btrfs_space_info *sinfo = cache->space_info;
9289 spin_lock(&sinfo->lock);
9290 spin_lock(&cache->lock);
9292 num_bytes = cache->key.offset - cache->reserved -
9293 cache->pinned - cache->bytes_super -
9294 btrfs_block_group_used(&cache->item);
9295 sinfo->bytes_readonly -= num_bytes;
9296 list_del_init(&cache->ro_list);
9298 spin_unlock(&cache->lock);
9299 spin_unlock(&sinfo->lock);
9303 * checks to see if its even possible to relocate this block group.
9305 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9306 * ok to go ahead and try.
9308 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9310 struct btrfs_block_group_cache *block_group;
9311 struct btrfs_space_info *space_info;
9312 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9313 struct btrfs_device *device;
9314 struct btrfs_trans_handle *trans;
9323 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9325 /* odd, couldn't find the block group, leave it alone */
9329 min_free = btrfs_block_group_used(&block_group->item);
9331 /* no bytes used, we're good */
9335 space_info = block_group->space_info;
9336 spin_lock(&space_info->lock);
9338 full = space_info->full;
9341 * if this is the last block group we have in this space, we can't
9342 * relocate it unless we're able to allocate a new chunk below.
9344 * Otherwise, we need to make sure we have room in the space to handle
9345 * all of the extents from this block group. If we can, we're good
9347 if ((space_info->total_bytes != block_group->key.offset) &&
9348 (space_info->bytes_used + space_info->bytes_reserved +
9349 space_info->bytes_pinned + space_info->bytes_readonly +
9350 min_free < space_info->total_bytes)) {
9351 spin_unlock(&space_info->lock);
9354 spin_unlock(&space_info->lock);
9357 * ok we don't have enough space, but maybe we have free space on our
9358 * devices to allocate new chunks for relocation, so loop through our
9359 * alloc devices and guess if we have enough space. if this block
9360 * group is going to be restriped, run checks against the target
9361 * profile instead of the current one.
9373 target = get_restripe_target(root->fs_info, block_group->flags);
9375 index = __get_raid_index(extended_to_chunk(target));
9378 * this is just a balance, so if we were marked as full
9379 * we know there is no space for a new chunk
9384 index = get_block_group_index(block_group);
9387 if (index == BTRFS_RAID_RAID10) {
9391 } else if (index == BTRFS_RAID_RAID1) {
9393 } else if (index == BTRFS_RAID_DUP) {
9396 } else if (index == BTRFS_RAID_RAID0) {
9397 dev_min = fs_devices->rw_devices;
9398 min_free = div64_u64(min_free, dev_min);
9401 /* We need to do this so that we can look at pending chunks */
9402 trans = btrfs_join_transaction(root);
9403 if (IS_ERR(trans)) {
9404 ret = PTR_ERR(trans);
9408 mutex_lock(&root->fs_info->chunk_mutex);
9409 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9413 * check to make sure we can actually find a chunk with enough
9414 * space to fit our block group in.
9416 if (device->total_bytes > device->bytes_used + min_free &&
9417 !device->is_tgtdev_for_dev_replace) {
9418 ret = find_free_dev_extent(trans, device, min_free,
9423 if (dev_nr >= dev_min)
9429 mutex_unlock(&root->fs_info->chunk_mutex);
9430 btrfs_end_transaction(trans, root);
9432 btrfs_put_block_group(block_group);
9436 static int find_first_block_group(struct btrfs_root *root,
9437 struct btrfs_path *path, struct btrfs_key *key)
9440 struct btrfs_key found_key;
9441 struct extent_buffer *leaf;
9444 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9449 slot = path->slots[0];
9450 leaf = path->nodes[0];
9451 if (slot >= btrfs_header_nritems(leaf)) {
9452 ret = btrfs_next_leaf(root, path);
9459 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9461 if (found_key.objectid >= key->objectid &&
9462 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9472 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9474 struct btrfs_block_group_cache *block_group;
9478 struct inode *inode;
9480 block_group = btrfs_lookup_first_block_group(info, last);
9481 while (block_group) {
9482 spin_lock(&block_group->lock);
9483 if (block_group->iref)
9485 spin_unlock(&block_group->lock);
9486 block_group = next_block_group(info->tree_root,
9496 inode = block_group->inode;
9497 block_group->iref = 0;
9498 block_group->inode = NULL;
9499 spin_unlock(&block_group->lock);
9501 last = block_group->key.objectid + block_group->key.offset;
9502 btrfs_put_block_group(block_group);
9506 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9508 struct btrfs_block_group_cache *block_group;
9509 struct btrfs_space_info *space_info;
9510 struct btrfs_caching_control *caching_ctl;
9513 down_write(&info->commit_root_sem);
9514 while (!list_empty(&info->caching_block_groups)) {
9515 caching_ctl = list_entry(info->caching_block_groups.next,
9516 struct btrfs_caching_control, list);
9517 list_del(&caching_ctl->list);
9518 put_caching_control(caching_ctl);
9520 up_write(&info->commit_root_sem);
9522 spin_lock(&info->unused_bgs_lock);
9523 while (!list_empty(&info->unused_bgs)) {
9524 block_group = list_first_entry(&info->unused_bgs,
9525 struct btrfs_block_group_cache,
9527 list_del_init(&block_group->bg_list);
9528 btrfs_put_block_group(block_group);
9530 spin_unlock(&info->unused_bgs_lock);
9532 spin_lock(&info->block_group_cache_lock);
9533 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9534 block_group = rb_entry(n, struct btrfs_block_group_cache,
9536 rb_erase(&block_group->cache_node,
9537 &info->block_group_cache_tree);
9538 RB_CLEAR_NODE(&block_group->cache_node);
9539 spin_unlock(&info->block_group_cache_lock);
9541 down_write(&block_group->space_info->groups_sem);
9542 list_del(&block_group->list);
9543 up_write(&block_group->space_info->groups_sem);
9545 if (block_group->cached == BTRFS_CACHE_STARTED)
9546 wait_block_group_cache_done(block_group);
9549 * We haven't cached this block group, which means we could
9550 * possibly have excluded extents on this block group.
9552 if (block_group->cached == BTRFS_CACHE_NO ||
9553 block_group->cached == BTRFS_CACHE_ERROR)
9554 free_excluded_extents(info->extent_root, block_group);
9556 btrfs_remove_free_space_cache(block_group);
9557 btrfs_put_block_group(block_group);
9559 spin_lock(&info->block_group_cache_lock);
9561 spin_unlock(&info->block_group_cache_lock);
9563 /* now that all the block groups are freed, go through and
9564 * free all the space_info structs. This is only called during
9565 * the final stages of unmount, and so we know nobody is
9566 * using them. We call synchronize_rcu() once before we start,
9567 * just to be on the safe side.
9571 release_global_block_rsv(info);
9573 while (!list_empty(&info->space_info)) {
9576 space_info = list_entry(info->space_info.next,
9577 struct btrfs_space_info,
9579 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9580 if (WARN_ON(space_info->bytes_pinned > 0 ||
9581 space_info->bytes_reserved > 0 ||
9582 space_info->bytes_may_use > 0)) {
9583 dump_space_info(space_info, 0, 0);
9586 list_del(&space_info->list);
9587 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9588 struct kobject *kobj;
9589 kobj = space_info->block_group_kobjs[i];
9590 space_info->block_group_kobjs[i] = NULL;
9596 kobject_del(&space_info->kobj);
9597 kobject_put(&space_info->kobj);
9602 static void __link_block_group(struct btrfs_space_info *space_info,
9603 struct btrfs_block_group_cache *cache)
9605 int index = get_block_group_index(cache);
9608 down_write(&space_info->groups_sem);
9609 if (list_empty(&space_info->block_groups[index]))
9611 list_add_tail(&cache->list, &space_info->block_groups[index]);
9612 up_write(&space_info->groups_sem);
9615 struct raid_kobject *rkobj;
9618 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9621 rkobj->raid_type = index;
9622 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9623 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9624 "%s", get_raid_name(index));
9626 kobject_put(&rkobj->kobj);
9629 space_info->block_group_kobjs[index] = &rkobj->kobj;
9634 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9637 static struct btrfs_block_group_cache *
9638 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9640 struct btrfs_block_group_cache *cache;
9642 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9646 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9648 if (!cache->free_space_ctl) {
9653 cache->key.objectid = start;
9654 cache->key.offset = size;
9655 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9657 cache->sectorsize = root->sectorsize;
9658 cache->fs_info = root->fs_info;
9659 cache->full_stripe_len = btrfs_full_stripe_len(root,
9660 &root->fs_info->mapping_tree,
9662 atomic_set(&cache->count, 1);
9663 spin_lock_init(&cache->lock);
9664 init_rwsem(&cache->data_rwsem);
9665 INIT_LIST_HEAD(&cache->list);
9666 INIT_LIST_HEAD(&cache->cluster_list);
9667 INIT_LIST_HEAD(&cache->bg_list);
9668 INIT_LIST_HEAD(&cache->ro_list);
9669 INIT_LIST_HEAD(&cache->dirty_list);
9670 INIT_LIST_HEAD(&cache->io_list);
9671 btrfs_init_free_space_ctl(cache);
9672 atomic_set(&cache->trimming, 0);
9677 int btrfs_read_block_groups(struct btrfs_root *root)
9679 struct btrfs_path *path;
9681 struct btrfs_block_group_cache *cache;
9682 struct btrfs_fs_info *info = root->fs_info;
9683 struct btrfs_space_info *space_info;
9684 struct btrfs_key key;
9685 struct btrfs_key found_key;
9686 struct extent_buffer *leaf;
9690 root = info->extent_root;
9693 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9694 path = btrfs_alloc_path();
9699 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9700 if (btrfs_test_opt(root, SPACE_CACHE) &&
9701 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9703 if (btrfs_test_opt(root, CLEAR_CACHE))
9707 ret = find_first_block_group(root, path, &key);
9713 leaf = path->nodes[0];
9714 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9716 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9725 * When we mount with old space cache, we need to
9726 * set BTRFS_DC_CLEAR and set dirty flag.
9728 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9729 * truncate the old free space cache inode and
9731 * b) Setting 'dirty flag' makes sure that we flush
9732 * the new space cache info onto disk.
9734 if (btrfs_test_opt(root, SPACE_CACHE))
9735 cache->disk_cache_state = BTRFS_DC_CLEAR;
9738 read_extent_buffer(leaf, &cache->item,
9739 btrfs_item_ptr_offset(leaf, path->slots[0]),
9740 sizeof(cache->item));
9741 cache->flags = btrfs_block_group_flags(&cache->item);
9743 key.objectid = found_key.objectid + found_key.offset;
9744 btrfs_release_path(path);
9747 * We need to exclude the super stripes now so that the space
9748 * info has super bytes accounted for, otherwise we'll think
9749 * we have more space than we actually do.
9751 ret = exclude_super_stripes(root, cache);
9754 * We may have excluded something, so call this just in
9757 free_excluded_extents(root, cache);
9758 btrfs_put_block_group(cache);
9763 * check for two cases, either we are full, and therefore
9764 * don't need to bother with the caching work since we won't
9765 * find any space, or we are empty, and we can just add all
9766 * the space in and be done with it. This saves us _alot_ of
9767 * time, particularly in the full case.
9769 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9770 cache->last_byte_to_unpin = (u64)-1;
9771 cache->cached = BTRFS_CACHE_FINISHED;
9772 free_excluded_extents(root, cache);
9773 } else if (btrfs_block_group_used(&cache->item) == 0) {
9774 cache->last_byte_to_unpin = (u64)-1;
9775 cache->cached = BTRFS_CACHE_FINISHED;
9776 add_new_free_space(cache, root->fs_info,
9778 found_key.objectid +
9780 free_excluded_extents(root, cache);
9783 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9785 btrfs_remove_free_space_cache(cache);
9786 btrfs_put_block_group(cache);
9790 ret = update_space_info(info, cache->flags, found_key.offset,
9791 btrfs_block_group_used(&cache->item),
9794 btrfs_remove_free_space_cache(cache);
9795 spin_lock(&info->block_group_cache_lock);
9796 rb_erase(&cache->cache_node,
9797 &info->block_group_cache_tree);
9798 RB_CLEAR_NODE(&cache->cache_node);
9799 spin_unlock(&info->block_group_cache_lock);
9800 btrfs_put_block_group(cache);
9804 cache->space_info = space_info;
9805 spin_lock(&cache->space_info->lock);
9806 cache->space_info->bytes_readonly += cache->bytes_super;
9807 spin_unlock(&cache->space_info->lock);
9809 __link_block_group(space_info, cache);
9811 set_avail_alloc_bits(root->fs_info, cache->flags);
9812 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9813 inc_block_group_ro(cache, 1);
9814 } else if (btrfs_block_group_used(&cache->item) == 0) {
9815 spin_lock(&info->unused_bgs_lock);
9816 /* Should always be true but just in case. */
9817 if (list_empty(&cache->bg_list)) {
9818 btrfs_get_block_group(cache);
9819 list_add_tail(&cache->bg_list,
9822 spin_unlock(&info->unused_bgs_lock);
9826 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9827 if (!(get_alloc_profile(root, space_info->flags) &
9828 (BTRFS_BLOCK_GROUP_RAID10 |
9829 BTRFS_BLOCK_GROUP_RAID1 |
9830 BTRFS_BLOCK_GROUP_RAID5 |
9831 BTRFS_BLOCK_GROUP_RAID6 |
9832 BTRFS_BLOCK_GROUP_DUP)))
9835 * avoid allocating from un-mirrored block group if there are
9836 * mirrored block groups.
9838 list_for_each_entry(cache,
9839 &space_info->block_groups[BTRFS_RAID_RAID0],
9841 inc_block_group_ro(cache, 1);
9842 list_for_each_entry(cache,
9843 &space_info->block_groups[BTRFS_RAID_SINGLE],
9845 inc_block_group_ro(cache, 1);
9848 init_global_block_rsv(info);
9851 btrfs_free_path(path);
9855 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9856 struct btrfs_root *root)
9858 struct btrfs_block_group_cache *block_group, *tmp;
9859 struct btrfs_root *extent_root = root->fs_info->extent_root;
9860 struct btrfs_block_group_item item;
9861 struct btrfs_key key;
9863 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9865 trans->can_flush_pending_bgs = false;
9866 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9870 spin_lock(&block_group->lock);
9871 memcpy(&item, &block_group->item, sizeof(item));
9872 memcpy(&key, &block_group->key, sizeof(key));
9873 spin_unlock(&block_group->lock);
9875 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9878 btrfs_abort_transaction(trans, extent_root, ret);
9879 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9880 key.objectid, key.offset);
9882 btrfs_abort_transaction(trans, extent_root, ret);
9884 list_del_init(&block_group->bg_list);
9886 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9889 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9890 struct btrfs_root *root, u64 bytes_used,
9891 u64 type, u64 chunk_objectid, u64 chunk_offset,
9895 struct btrfs_root *extent_root;
9896 struct btrfs_block_group_cache *cache;
9898 extent_root = root->fs_info->extent_root;
9900 btrfs_set_log_full_commit(root->fs_info, trans);
9902 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9906 btrfs_set_block_group_used(&cache->item, bytes_used);
9907 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9908 btrfs_set_block_group_flags(&cache->item, type);
9910 cache->flags = type;
9911 cache->last_byte_to_unpin = (u64)-1;
9912 cache->cached = BTRFS_CACHE_FINISHED;
9913 ret = exclude_super_stripes(root, cache);
9916 * We may have excluded something, so call this just in
9919 free_excluded_extents(root, cache);
9920 btrfs_put_block_group(cache);
9924 add_new_free_space(cache, root->fs_info, chunk_offset,
9925 chunk_offset + size);
9927 free_excluded_extents(root, cache);
9929 #ifdef CONFIG_BTRFS_DEBUG
9930 if (btrfs_should_fragment_free_space(root, cache)) {
9931 u64 new_bytes_used = size - bytes_used;
9933 bytes_used += new_bytes_used >> 1;
9934 fragment_free_space(root, cache);
9938 * Call to ensure the corresponding space_info object is created and
9939 * assigned to our block group, but don't update its counters just yet.
9940 * We want our bg to be added to the rbtree with its ->space_info set.
9942 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9943 &cache->space_info);
9945 btrfs_remove_free_space_cache(cache);
9946 btrfs_put_block_group(cache);
9950 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9952 btrfs_remove_free_space_cache(cache);
9953 btrfs_put_block_group(cache);
9958 * Now that our block group has its ->space_info set and is inserted in
9959 * the rbtree, update the space info's counters.
9961 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9962 &cache->space_info);
9964 btrfs_remove_free_space_cache(cache);
9965 spin_lock(&root->fs_info->block_group_cache_lock);
9966 rb_erase(&cache->cache_node,
9967 &root->fs_info->block_group_cache_tree);
9968 RB_CLEAR_NODE(&cache->cache_node);
9969 spin_unlock(&root->fs_info->block_group_cache_lock);
9970 btrfs_put_block_group(cache);
9973 update_global_block_rsv(root->fs_info);
9975 spin_lock(&cache->space_info->lock);
9976 cache->space_info->bytes_readonly += cache->bytes_super;
9977 spin_unlock(&cache->space_info->lock);
9979 __link_block_group(cache->space_info, cache);
9981 list_add_tail(&cache->bg_list, &trans->new_bgs);
9983 set_avail_alloc_bits(extent_root->fs_info, type);
9988 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9990 u64 extra_flags = chunk_to_extended(flags) &
9991 BTRFS_EXTENDED_PROFILE_MASK;
9993 write_seqlock(&fs_info->profiles_lock);
9994 if (flags & BTRFS_BLOCK_GROUP_DATA)
9995 fs_info->avail_data_alloc_bits &= ~extra_flags;
9996 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9997 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9998 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9999 fs_info->avail_system_alloc_bits &= ~extra_flags;
10000 write_sequnlock(&fs_info->profiles_lock);
10003 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
10004 struct btrfs_root *root, u64 group_start,
10005 struct extent_map *em)
10007 struct btrfs_path *path;
10008 struct btrfs_block_group_cache *block_group;
10009 struct btrfs_free_cluster *cluster;
10010 struct btrfs_root *tree_root = root->fs_info->tree_root;
10011 struct btrfs_key key;
10012 struct inode *inode;
10013 struct kobject *kobj = NULL;
10017 struct btrfs_caching_control *caching_ctl = NULL;
10020 root = root->fs_info->extent_root;
10022 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
10023 BUG_ON(!block_group);
10024 BUG_ON(!block_group->ro);
10027 * Free the reserved super bytes from this block group before
10030 free_excluded_extents(root, block_group);
10032 memcpy(&key, &block_group->key, sizeof(key));
10033 index = get_block_group_index(block_group);
10034 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10035 BTRFS_BLOCK_GROUP_RAID1 |
10036 BTRFS_BLOCK_GROUP_RAID10))
10041 /* make sure this block group isn't part of an allocation cluster */
10042 cluster = &root->fs_info->data_alloc_cluster;
10043 spin_lock(&cluster->refill_lock);
10044 btrfs_return_cluster_to_free_space(block_group, cluster);
10045 spin_unlock(&cluster->refill_lock);
10048 * make sure this block group isn't part of a metadata
10049 * allocation cluster
10051 cluster = &root->fs_info->meta_alloc_cluster;
10052 spin_lock(&cluster->refill_lock);
10053 btrfs_return_cluster_to_free_space(block_group, cluster);
10054 spin_unlock(&cluster->refill_lock);
10056 path = btrfs_alloc_path();
10063 * get the inode first so any iput calls done for the io_list
10064 * aren't the final iput (no unlinks allowed now)
10066 inode = lookup_free_space_inode(tree_root, block_group, path);
10068 mutex_lock(&trans->transaction->cache_write_mutex);
10070 * make sure our free spache cache IO is done before remove the
10073 spin_lock(&trans->transaction->dirty_bgs_lock);
10074 if (!list_empty(&block_group->io_list)) {
10075 list_del_init(&block_group->io_list);
10077 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10079 spin_unlock(&trans->transaction->dirty_bgs_lock);
10080 btrfs_wait_cache_io(root, trans, block_group,
10081 &block_group->io_ctl, path,
10082 block_group->key.objectid);
10083 btrfs_put_block_group(block_group);
10084 spin_lock(&trans->transaction->dirty_bgs_lock);
10087 if (!list_empty(&block_group->dirty_list)) {
10088 list_del_init(&block_group->dirty_list);
10089 btrfs_put_block_group(block_group);
10091 spin_unlock(&trans->transaction->dirty_bgs_lock);
10092 mutex_unlock(&trans->transaction->cache_write_mutex);
10094 if (!IS_ERR(inode)) {
10095 ret = btrfs_orphan_add(trans, inode);
10097 btrfs_add_delayed_iput(inode);
10100 clear_nlink(inode);
10101 /* One for the block groups ref */
10102 spin_lock(&block_group->lock);
10103 if (block_group->iref) {
10104 block_group->iref = 0;
10105 block_group->inode = NULL;
10106 spin_unlock(&block_group->lock);
10109 spin_unlock(&block_group->lock);
10111 /* One for our lookup ref */
10112 btrfs_add_delayed_iput(inode);
10115 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10116 key.offset = block_group->key.objectid;
10119 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10123 btrfs_release_path(path);
10125 ret = btrfs_del_item(trans, tree_root, path);
10128 btrfs_release_path(path);
10131 spin_lock(&root->fs_info->block_group_cache_lock);
10132 rb_erase(&block_group->cache_node,
10133 &root->fs_info->block_group_cache_tree);
10134 RB_CLEAR_NODE(&block_group->cache_node);
10136 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10137 root->fs_info->first_logical_byte = (u64)-1;
10138 spin_unlock(&root->fs_info->block_group_cache_lock);
10140 down_write(&block_group->space_info->groups_sem);
10142 * we must use list_del_init so people can check to see if they
10143 * are still on the list after taking the semaphore
10145 list_del_init(&block_group->list);
10146 if (list_empty(&block_group->space_info->block_groups[index])) {
10147 kobj = block_group->space_info->block_group_kobjs[index];
10148 block_group->space_info->block_group_kobjs[index] = NULL;
10149 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10151 up_write(&block_group->space_info->groups_sem);
10157 if (block_group->has_caching_ctl)
10158 caching_ctl = get_caching_control(block_group);
10159 if (block_group->cached == BTRFS_CACHE_STARTED)
10160 wait_block_group_cache_done(block_group);
10161 if (block_group->has_caching_ctl) {
10162 down_write(&root->fs_info->commit_root_sem);
10163 if (!caching_ctl) {
10164 struct btrfs_caching_control *ctl;
10166 list_for_each_entry(ctl,
10167 &root->fs_info->caching_block_groups, list)
10168 if (ctl->block_group == block_group) {
10170 atomic_inc(&caching_ctl->count);
10175 list_del_init(&caching_ctl->list);
10176 up_write(&root->fs_info->commit_root_sem);
10178 /* Once for the caching bgs list and once for us. */
10179 put_caching_control(caching_ctl);
10180 put_caching_control(caching_ctl);
10184 spin_lock(&trans->transaction->dirty_bgs_lock);
10185 if (!list_empty(&block_group->dirty_list)) {
10188 if (!list_empty(&block_group->io_list)) {
10191 spin_unlock(&trans->transaction->dirty_bgs_lock);
10192 btrfs_remove_free_space_cache(block_group);
10194 spin_lock(&block_group->space_info->lock);
10195 list_del_init(&block_group->ro_list);
10197 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10198 WARN_ON(block_group->space_info->total_bytes
10199 < block_group->key.offset);
10200 WARN_ON(block_group->space_info->bytes_readonly
10201 < block_group->key.offset);
10202 WARN_ON(block_group->space_info->disk_total
10203 < block_group->key.offset * factor);
10205 block_group->space_info->total_bytes -= block_group->key.offset;
10206 block_group->space_info->bytes_readonly -= block_group->key.offset;
10207 block_group->space_info->disk_total -= block_group->key.offset * factor;
10209 spin_unlock(&block_group->space_info->lock);
10211 memcpy(&key, &block_group->key, sizeof(key));
10214 if (!list_empty(&em->list)) {
10215 /* We're in the transaction->pending_chunks list. */
10216 free_extent_map(em);
10218 spin_lock(&block_group->lock);
10219 block_group->removed = 1;
10221 * At this point trimming can't start on this block group, because we
10222 * removed the block group from the tree fs_info->block_group_cache_tree
10223 * so no one can't find it anymore and even if someone already got this
10224 * block group before we removed it from the rbtree, they have already
10225 * incremented block_group->trimming - if they didn't, they won't find
10226 * any free space entries because we already removed them all when we
10227 * called btrfs_remove_free_space_cache().
10229 * And we must not remove the extent map from the fs_info->mapping_tree
10230 * to prevent the same logical address range and physical device space
10231 * ranges from being reused for a new block group. This is because our
10232 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10233 * completely transactionless, so while it is trimming a range the
10234 * currently running transaction might finish and a new one start,
10235 * allowing for new block groups to be created that can reuse the same
10236 * physical device locations unless we take this special care.
10238 * There may also be an implicit trim operation if the file system
10239 * is mounted with -odiscard. The same protections must remain
10240 * in place until the extents have been discarded completely when
10241 * the transaction commit has completed.
10243 remove_em = (atomic_read(&block_group->trimming) == 0);
10245 * Make sure a trimmer task always sees the em in the pinned_chunks list
10246 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10247 * before checking block_group->removed).
10251 * Our em might be in trans->transaction->pending_chunks which
10252 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10253 * and so is the fs_info->pinned_chunks list.
10255 * So at this point we must be holding the chunk_mutex to avoid
10256 * any races with chunk allocation (more specifically at
10257 * volumes.c:contains_pending_extent()), to ensure it always
10258 * sees the em, either in the pending_chunks list or in the
10259 * pinned_chunks list.
10261 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10263 spin_unlock(&block_group->lock);
10266 struct extent_map_tree *em_tree;
10268 em_tree = &root->fs_info->mapping_tree.map_tree;
10269 write_lock(&em_tree->lock);
10271 * The em might be in the pending_chunks list, so make sure the
10272 * chunk mutex is locked, since remove_extent_mapping() will
10273 * delete us from that list.
10275 remove_extent_mapping(em_tree, em);
10276 write_unlock(&em_tree->lock);
10277 /* once for the tree */
10278 free_extent_map(em);
10281 unlock_chunks(root);
10283 btrfs_put_block_group(block_group);
10284 btrfs_put_block_group(block_group);
10286 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10292 ret = btrfs_del_item(trans, root, path);
10294 btrfs_free_path(path);
10298 struct btrfs_trans_handle *
10299 btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info,
10300 const u64 chunk_offset)
10302 struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree;
10303 struct extent_map *em;
10304 struct map_lookup *map;
10305 unsigned int num_items;
10307 read_lock(&em_tree->lock);
10308 em = lookup_extent_mapping(em_tree, chunk_offset, 1);
10309 read_unlock(&em_tree->lock);
10310 ASSERT(em && em->start == chunk_offset);
10313 * We need to reserve 3 + N units from the metadata space info in order
10314 * to remove a block group (done at btrfs_remove_chunk() and at
10315 * btrfs_remove_block_group()), which are used for:
10317 * 1 unit for adding the free space inode's orphan (located in the tree
10319 * 1 unit for deleting the block group item (located in the extent
10321 * 1 unit for deleting the free space item (located in tree of tree
10323 * N units for deleting N device extent items corresponding to each
10324 * stripe (located in the device tree).
10326 * In order to remove a block group we also need to reserve units in the
10327 * system space info in order to update the chunk tree (update one or
10328 * more device items and remove one chunk item), but this is done at
10329 * btrfs_remove_chunk() through a call to check_system_chunk().
10331 map = (struct map_lookup *)em->bdev;
10332 num_items = 3 + map->num_stripes;
10333 free_extent_map(em);
10335 return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root,
10340 * Process the unused_bgs list and remove any that don't have any allocated
10341 * space inside of them.
10343 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10345 struct btrfs_block_group_cache *block_group;
10346 struct btrfs_space_info *space_info;
10347 struct btrfs_root *root = fs_info->extent_root;
10348 struct btrfs_trans_handle *trans;
10351 if (!fs_info->open)
10354 spin_lock(&fs_info->unused_bgs_lock);
10355 while (!list_empty(&fs_info->unused_bgs)) {
10359 block_group = list_first_entry(&fs_info->unused_bgs,
10360 struct btrfs_block_group_cache,
10362 list_del_init(&block_group->bg_list);
10364 space_info = block_group->space_info;
10366 if (ret || btrfs_mixed_space_info(space_info)) {
10367 btrfs_put_block_group(block_group);
10370 spin_unlock(&fs_info->unused_bgs_lock);
10372 mutex_lock(&fs_info->delete_unused_bgs_mutex);
10374 /* Don't want to race with allocators so take the groups_sem */
10375 down_write(&space_info->groups_sem);
10376 spin_lock(&block_group->lock);
10377 if (block_group->reserved ||
10378 btrfs_block_group_used(&block_group->item) ||
10380 list_is_singular(&block_group->list)) {
10382 * We want to bail if we made new allocations or have
10383 * outstanding allocations in this block group. We do
10384 * the ro check in case balance is currently acting on
10385 * this block group.
10387 spin_unlock(&block_group->lock);
10388 up_write(&space_info->groups_sem);
10391 spin_unlock(&block_group->lock);
10393 /* We don't want to force the issue, only flip if it's ok. */
10394 ret = inc_block_group_ro(block_group, 0);
10395 up_write(&space_info->groups_sem);
10402 * Want to do this before we do anything else so we can recover
10403 * properly if we fail to join the transaction.
10405 trans = btrfs_start_trans_remove_block_group(fs_info,
10406 block_group->key.objectid);
10407 if (IS_ERR(trans)) {
10408 btrfs_dec_block_group_ro(root, block_group);
10409 ret = PTR_ERR(trans);
10414 * We could have pending pinned extents for this block group,
10415 * just delete them, we don't care about them anymore.
10417 start = block_group->key.objectid;
10418 end = start + block_group->key.offset - 1;
10420 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10421 * btrfs_finish_extent_commit(). If we are at transaction N,
10422 * another task might be running finish_extent_commit() for the
10423 * previous transaction N - 1, and have seen a range belonging
10424 * to the block group in freed_extents[] before we were able to
10425 * clear the whole block group range from freed_extents[]. This
10426 * means that task can lookup for the block group after we
10427 * unpinned it from freed_extents[] and removed it, leading to
10428 * a BUG_ON() at btrfs_unpin_extent_range().
10430 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10431 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10432 EXTENT_DIRTY, GFP_NOFS);
10434 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10435 btrfs_dec_block_group_ro(root, block_group);
10438 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10439 EXTENT_DIRTY, GFP_NOFS);
10441 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10442 btrfs_dec_block_group_ro(root, block_group);
10445 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10447 /* Reset pinned so btrfs_put_block_group doesn't complain */
10448 spin_lock(&space_info->lock);
10449 spin_lock(&block_group->lock);
10451 space_info->bytes_pinned -= block_group->pinned;
10452 space_info->bytes_readonly += block_group->pinned;
10453 percpu_counter_add(&space_info->total_bytes_pinned,
10454 -block_group->pinned);
10455 block_group->pinned = 0;
10457 spin_unlock(&block_group->lock);
10458 spin_unlock(&space_info->lock);
10460 /* DISCARD can flip during remount */
10461 trimming = btrfs_test_opt(root, DISCARD);
10463 /* Implicit trim during transaction commit. */
10465 btrfs_get_block_group_trimming(block_group);
10468 * Btrfs_remove_chunk will abort the transaction if things go
10471 ret = btrfs_remove_chunk(trans, root,
10472 block_group->key.objectid);
10476 btrfs_put_block_group_trimming(block_group);
10481 * If we're not mounted with -odiscard, we can just forget
10482 * about this block group. Otherwise we'll need to wait
10483 * until transaction commit to do the actual discard.
10486 spin_lock(&fs_info->unused_bgs_lock);
10488 * A concurrent scrub might have added us to the list
10489 * fs_info->unused_bgs, so use a list_move operation
10490 * to add the block group to the deleted_bgs list.
10492 list_move(&block_group->bg_list,
10493 &trans->transaction->deleted_bgs);
10494 spin_unlock(&fs_info->unused_bgs_lock);
10495 btrfs_get_block_group(block_group);
10498 btrfs_end_transaction(trans, root);
10500 mutex_unlock(&fs_info->delete_unused_bgs_mutex);
10501 btrfs_put_block_group(block_group);
10502 spin_lock(&fs_info->unused_bgs_lock);
10504 spin_unlock(&fs_info->unused_bgs_lock);
10507 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10509 struct btrfs_space_info *space_info;
10510 struct btrfs_super_block *disk_super;
10516 disk_super = fs_info->super_copy;
10517 if (!btrfs_super_root(disk_super))
10520 features = btrfs_super_incompat_flags(disk_super);
10521 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10524 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10525 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10530 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10531 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10533 flags = BTRFS_BLOCK_GROUP_METADATA;
10534 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10538 flags = BTRFS_BLOCK_GROUP_DATA;
10539 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10545 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10547 return unpin_extent_range(root, start, end, false);
10551 * It used to be that old block groups would be left around forever.
10552 * Iterating over them would be enough to trim unused space. Since we
10553 * now automatically remove them, we also need to iterate over unallocated
10556 * We don't want a transaction for this since the discard may take a
10557 * substantial amount of time. We don't require that a transaction be
10558 * running, but we do need to take a running transaction into account
10559 * to ensure that we're not discarding chunks that were released in
10560 * the current transaction.
10562 * Holding the chunks lock will prevent other threads from allocating
10563 * or releasing chunks, but it won't prevent a running transaction
10564 * from committing and releasing the memory that the pending chunks
10565 * list head uses. For that, we need to take a reference to the
10568 static int btrfs_trim_free_extents(struct btrfs_device *device,
10569 u64 minlen, u64 *trimmed)
10571 u64 start = 0, len = 0;
10576 /* Not writeable = nothing to do. */
10577 if (!device->writeable)
10580 /* No free space = nothing to do. */
10581 if (device->total_bytes <= device->bytes_used)
10587 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10588 struct btrfs_transaction *trans;
10591 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10595 down_read(&fs_info->commit_root_sem);
10597 spin_lock(&fs_info->trans_lock);
10598 trans = fs_info->running_transaction;
10600 atomic_inc(&trans->use_count);
10601 spin_unlock(&fs_info->trans_lock);
10603 ret = find_free_dev_extent_start(trans, device, minlen, start,
10606 btrfs_put_transaction(trans);
10609 up_read(&fs_info->commit_root_sem);
10610 mutex_unlock(&fs_info->chunk_mutex);
10611 if (ret == -ENOSPC)
10616 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10617 up_read(&fs_info->commit_root_sem);
10618 mutex_unlock(&fs_info->chunk_mutex);
10626 if (fatal_signal_pending(current)) {
10627 ret = -ERESTARTSYS;
10637 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10639 struct btrfs_fs_info *fs_info = root->fs_info;
10640 struct btrfs_block_group_cache *cache = NULL;
10641 struct btrfs_device *device;
10642 struct list_head *devices;
10647 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10651 * try to trim all FS space, our block group may start from non-zero.
10653 if (range->len == total_bytes)
10654 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10656 cache = btrfs_lookup_block_group(fs_info, range->start);
10659 if (cache->key.objectid >= (range->start + range->len)) {
10660 btrfs_put_block_group(cache);
10664 start = max(range->start, cache->key.objectid);
10665 end = min(range->start + range->len,
10666 cache->key.objectid + cache->key.offset);
10668 if (end - start >= range->minlen) {
10669 if (!block_group_cache_done(cache)) {
10670 ret = cache_block_group(cache, 0);
10672 btrfs_put_block_group(cache);
10675 ret = wait_block_group_cache_done(cache);
10677 btrfs_put_block_group(cache);
10681 ret = btrfs_trim_block_group(cache,
10687 trimmed += group_trimmed;
10689 btrfs_put_block_group(cache);
10694 cache = next_block_group(fs_info->tree_root, cache);
10697 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10698 devices = &root->fs_info->fs_devices->alloc_list;
10699 list_for_each_entry(device, devices, dev_alloc_list) {
10700 ret = btrfs_trim_free_extents(device, range->minlen,
10705 trimmed += group_trimmed;
10707 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10709 range->len = trimmed;
10714 * btrfs_{start,end}_write_no_snapshoting() are similar to
10715 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10716 * data into the page cache through nocow before the subvolume is snapshoted,
10717 * but flush the data into disk after the snapshot creation, or to prevent
10718 * operations while snapshoting is ongoing and that cause the snapshot to be
10719 * inconsistent (writes followed by expanding truncates for example).
10721 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10723 percpu_counter_dec(&root->subv_writers->counter);
10725 * Make sure counter is updated before we wake up waiters.
10728 if (waitqueue_active(&root->subv_writers->wait))
10729 wake_up(&root->subv_writers->wait);
10732 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10734 if (atomic_read(&root->will_be_snapshoted))
10737 percpu_counter_inc(&root->subv_writers->counter);
10739 * Make sure counter is updated before we check for snapshot creation.
10742 if (atomic_read(&root->will_be_snapshoted)) {
10743 btrfs_end_write_no_snapshoting(root);