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 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
129 atomic_inc(&cache->count);
132 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
134 if (atomic_dec_and_test(&cache->count)) {
135 WARN_ON(cache->pinned > 0);
136 WARN_ON(cache->reserved > 0);
137 kfree(cache->free_space_ctl);
143 * this adds the block group to the fs_info rb tree for the block group
146 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
147 struct btrfs_block_group_cache *block_group)
150 struct rb_node *parent = NULL;
151 struct btrfs_block_group_cache *cache;
153 spin_lock(&info->block_group_cache_lock);
154 p = &info->block_group_cache_tree.rb_node;
158 cache = rb_entry(parent, struct btrfs_block_group_cache,
160 if (block_group->key.objectid < cache->key.objectid) {
162 } else if (block_group->key.objectid > cache->key.objectid) {
165 spin_unlock(&info->block_group_cache_lock);
170 rb_link_node(&block_group->cache_node, parent, p);
171 rb_insert_color(&block_group->cache_node,
172 &info->block_group_cache_tree);
174 if (info->first_logical_byte > block_group->key.objectid)
175 info->first_logical_byte = block_group->key.objectid;
177 spin_unlock(&info->block_group_cache_lock);
183 * This will return the block group at or after bytenr if contains is 0, else
184 * it will return the block group that contains the bytenr
186 static struct btrfs_block_group_cache *
187 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
190 struct btrfs_block_group_cache *cache, *ret = NULL;
194 spin_lock(&info->block_group_cache_lock);
195 n = info->block_group_cache_tree.rb_node;
198 cache = rb_entry(n, struct btrfs_block_group_cache,
200 end = cache->key.objectid + cache->key.offset - 1;
201 start = cache->key.objectid;
203 if (bytenr < start) {
204 if (!contains && (!ret || start < ret->key.objectid))
207 } else if (bytenr > start) {
208 if (contains && bytenr <= end) {
219 btrfs_get_block_group(ret);
220 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
221 info->first_logical_byte = ret->key.objectid;
223 spin_unlock(&info->block_group_cache_lock);
228 static int add_excluded_extent(struct btrfs_root *root,
229 u64 start, u64 num_bytes)
231 u64 end = start + num_bytes - 1;
232 set_extent_bits(&root->fs_info->freed_extents[0],
233 start, end, EXTENT_UPTODATE, GFP_NOFS);
234 set_extent_bits(&root->fs_info->freed_extents[1],
235 start, end, EXTENT_UPTODATE, GFP_NOFS);
239 static void free_excluded_extents(struct btrfs_root *root,
240 struct btrfs_block_group_cache *cache)
244 start = cache->key.objectid;
245 end = start + cache->key.offset - 1;
247 clear_extent_bits(&root->fs_info->freed_extents[0],
248 start, end, EXTENT_UPTODATE, GFP_NOFS);
249 clear_extent_bits(&root->fs_info->freed_extents[1],
250 start, end, EXTENT_UPTODATE, GFP_NOFS);
253 static int exclude_super_stripes(struct btrfs_root *root,
254 struct btrfs_block_group_cache *cache)
261 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
262 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
263 cache->bytes_super += stripe_len;
264 ret = add_excluded_extent(root, cache->key.objectid,
270 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
271 bytenr = btrfs_sb_offset(i);
272 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
273 cache->key.objectid, bytenr,
274 0, &logical, &nr, &stripe_len);
281 if (logical[nr] > cache->key.objectid +
285 if (logical[nr] + stripe_len <= cache->key.objectid)
289 if (start < cache->key.objectid) {
290 start = cache->key.objectid;
291 len = (logical[nr] + stripe_len) - start;
293 len = min_t(u64, stripe_len,
294 cache->key.objectid +
295 cache->key.offset - start);
298 cache->bytes_super += len;
299 ret = add_excluded_extent(root, start, len);
311 static struct btrfs_caching_control *
312 get_caching_control(struct btrfs_block_group_cache *cache)
314 struct btrfs_caching_control *ctl;
316 spin_lock(&cache->lock);
317 if (!cache->caching_ctl) {
318 spin_unlock(&cache->lock);
322 ctl = cache->caching_ctl;
323 atomic_inc(&ctl->count);
324 spin_unlock(&cache->lock);
328 static void put_caching_control(struct btrfs_caching_control *ctl)
330 if (atomic_dec_and_test(&ctl->count))
334 #ifdef CONFIG_BTRFS_DEBUG
335 static void fragment_free_space(struct btrfs_root *root,
336 struct btrfs_block_group_cache *block_group)
338 u64 start = block_group->key.objectid;
339 u64 len = block_group->key.offset;
340 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
341 root->nodesize : root->sectorsize;
342 u64 step = chunk << 1;
344 while (len > chunk) {
345 btrfs_remove_free_space(block_group, start, chunk);
356 * this is only called by cache_block_group, since we could have freed extents
357 * we need to check the pinned_extents for any extents that can't be used yet
358 * since their free space will be released as soon as the transaction commits.
360 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
361 struct btrfs_fs_info *info, u64 start, u64 end)
363 u64 extent_start, extent_end, size, total_added = 0;
366 while (start < end) {
367 ret = find_first_extent_bit(info->pinned_extents, start,
368 &extent_start, &extent_end,
369 EXTENT_DIRTY | EXTENT_UPTODATE,
374 if (extent_start <= start) {
375 start = extent_end + 1;
376 } else if (extent_start > start && extent_start < end) {
377 size = extent_start - start;
379 ret = btrfs_add_free_space(block_group, start,
381 BUG_ON(ret); /* -ENOMEM or logic error */
382 start = extent_end + 1;
391 ret = btrfs_add_free_space(block_group, start, size);
392 BUG_ON(ret); /* -ENOMEM or logic error */
398 static noinline void caching_thread(struct btrfs_work *work)
400 struct btrfs_block_group_cache *block_group;
401 struct btrfs_fs_info *fs_info;
402 struct btrfs_caching_control *caching_ctl;
403 struct btrfs_root *extent_root;
404 struct btrfs_path *path;
405 struct extent_buffer *leaf;
406 struct btrfs_key key;
413 caching_ctl = container_of(work, struct btrfs_caching_control, work);
414 block_group = caching_ctl->block_group;
415 fs_info = block_group->fs_info;
416 extent_root = fs_info->extent_root;
418 path = btrfs_alloc_path();
422 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
424 #ifdef CONFIG_BTRFS_DEBUG
426 * If we're fragmenting we don't want to make anybody think we can
427 * allocate from this block group until we've had a chance to fragment
430 if (btrfs_should_fragment_free_space(extent_root, block_group))
434 * We don't want to deadlock with somebody trying to allocate a new
435 * extent for the extent root while also trying to search the extent
436 * root to add free space. So we skip locking and search the commit
437 * root, since its read-only
439 path->skip_locking = 1;
440 path->search_commit_root = 1;
445 key.type = BTRFS_EXTENT_ITEM_KEY;
447 mutex_lock(&caching_ctl->mutex);
448 /* need to make sure the commit_root doesn't disappear */
449 down_read(&fs_info->commit_root_sem);
452 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
456 leaf = path->nodes[0];
457 nritems = btrfs_header_nritems(leaf);
460 if (btrfs_fs_closing(fs_info) > 1) {
465 if (path->slots[0] < nritems) {
466 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
468 ret = find_next_key(path, 0, &key);
472 if (need_resched() ||
473 rwsem_is_contended(&fs_info->commit_root_sem)) {
475 caching_ctl->progress = last;
476 btrfs_release_path(path);
477 up_read(&fs_info->commit_root_sem);
478 mutex_unlock(&caching_ctl->mutex);
483 ret = btrfs_next_leaf(extent_root, path);
488 leaf = path->nodes[0];
489 nritems = btrfs_header_nritems(leaf);
493 if (key.objectid < last) {
496 key.type = BTRFS_EXTENT_ITEM_KEY;
499 caching_ctl->progress = last;
500 btrfs_release_path(path);
504 if (key.objectid < block_group->key.objectid) {
509 if (key.objectid >= block_group->key.objectid +
510 block_group->key.offset)
513 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
514 key.type == BTRFS_METADATA_ITEM_KEY) {
515 total_found += add_new_free_space(block_group,
518 if (key.type == BTRFS_METADATA_ITEM_KEY)
519 last = key.objectid +
520 fs_info->tree_root->nodesize;
522 last = key.objectid + key.offset;
524 if (total_found > (1024 * 1024 * 2)) {
527 wake_up(&caching_ctl->wait);
534 total_found += add_new_free_space(block_group, fs_info, last,
535 block_group->key.objectid +
536 block_group->key.offset);
537 spin_lock(&block_group->lock);
538 block_group->caching_ctl = NULL;
539 block_group->cached = BTRFS_CACHE_FINISHED;
540 spin_unlock(&block_group->lock);
542 #ifdef CONFIG_BTRFS_DEBUG
543 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
546 spin_lock(&block_group->space_info->lock);
547 spin_lock(&block_group->lock);
548 bytes_used = block_group->key.offset -
549 btrfs_block_group_used(&block_group->item);
550 block_group->space_info->bytes_used += bytes_used >> 1;
551 spin_unlock(&block_group->lock);
552 spin_unlock(&block_group->space_info->lock);
553 fragment_free_space(extent_root, block_group);
557 caching_ctl->progress = (u64)-1;
559 btrfs_free_path(path);
560 up_read(&fs_info->commit_root_sem);
562 free_excluded_extents(extent_root, block_group);
564 mutex_unlock(&caching_ctl->mutex);
567 spin_lock(&block_group->lock);
568 block_group->caching_ctl = NULL;
569 block_group->cached = BTRFS_CACHE_ERROR;
570 spin_unlock(&block_group->lock);
572 wake_up(&caching_ctl->wait);
574 put_caching_control(caching_ctl);
575 btrfs_put_block_group(block_group);
578 static int cache_block_group(struct btrfs_block_group_cache *cache,
582 struct btrfs_fs_info *fs_info = cache->fs_info;
583 struct btrfs_caching_control *caching_ctl;
586 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
590 INIT_LIST_HEAD(&caching_ctl->list);
591 mutex_init(&caching_ctl->mutex);
592 init_waitqueue_head(&caching_ctl->wait);
593 caching_ctl->block_group = cache;
594 caching_ctl->progress = cache->key.objectid;
595 atomic_set(&caching_ctl->count, 1);
596 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
597 caching_thread, NULL, NULL);
599 spin_lock(&cache->lock);
601 * This should be a rare occasion, but this could happen I think in the
602 * case where one thread starts to load the space cache info, and then
603 * some other thread starts a transaction commit which tries to do an
604 * allocation while the other thread is still loading the space cache
605 * info. The previous loop should have kept us from choosing this block
606 * group, but if we've moved to the state where we will wait on caching
607 * block groups we need to first check if we're doing a fast load here,
608 * so we can wait for it to finish, otherwise we could end up allocating
609 * from a block group who's cache gets evicted for one reason or
612 while (cache->cached == BTRFS_CACHE_FAST) {
613 struct btrfs_caching_control *ctl;
615 ctl = cache->caching_ctl;
616 atomic_inc(&ctl->count);
617 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
618 spin_unlock(&cache->lock);
622 finish_wait(&ctl->wait, &wait);
623 put_caching_control(ctl);
624 spin_lock(&cache->lock);
627 if (cache->cached != BTRFS_CACHE_NO) {
628 spin_unlock(&cache->lock);
632 WARN_ON(cache->caching_ctl);
633 cache->caching_ctl = caching_ctl;
634 cache->cached = BTRFS_CACHE_FAST;
635 spin_unlock(&cache->lock);
637 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
638 mutex_lock(&caching_ctl->mutex);
639 ret = load_free_space_cache(fs_info, cache);
641 spin_lock(&cache->lock);
643 cache->caching_ctl = NULL;
644 cache->cached = BTRFS_CACHE_FINISHED;
645 cache->last_byte_to_unpin = (u64)-1;
646 caching_ctl->progress = (u64)-1;
648 if (load_cache_only) {
649 cache->caching_ctl = NULL;
650 cache->cached = BTRFS_CACHE_NO;
652 cache->cached = BTRFS_CACHE_STARTED;
653 cache->has_caching_ctl = 1;
656 spin_unlock(&cache->lock);
657 #ifdef CONFIG_BTRFS_DEBUG
659 btrfs_should_fragment_free_space(fs_info->extent_root,
663 spin_lock(&cache->space_info->lock);
664 spin_lock(&cache->lock);
665 bytes_used = cache->key.offset -
666 btrfs_block_group_used(&cache->item);
667 cache->space_info->bytes_used += bytes_used >> 1;
668 spin_unlock(&cache->lock);
669 spin_unlock(&cache->space_info->lock);
670 fragment_free_space(fs_info->extent_root, cache);
673 mutex_unlock(&caching_ctl->mutex);
675 wake_up(&caching_ctl->wait);
677 put_caching_control(caching_ctl);
678 free_excluded_extents(fs_info->extent_root, cache);
683 * We are not going to do the fast caching, set cached to the
684 * appropriate value and wakeup any waiters.
686 spin_lock(&cache->lock);
687 if (load_cache_only) {
688 cache->caching_ctl = NULL;
689 cache->cached = BTRFS_CACHE_NO;
691 cache->cached = BTRFS_CACHE_STARTED;
692 cache->has_caching_ctl = 1;
694 spin_unlock(&cache->lock);
695 wake_up(&caching_ctl->wait);
698 if (load_cache_only) {
699 put_caching_control(caching_ctl);
703 down_write(&fs_info->commit_root_sem);
704 atomic_inc(&caching_ctl->count);
705 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
706 up_write(&fs_info->commit_root_sem);
708 btrfs_get_block_group(cache);
710 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
716 * return the block group that starts at or after bytenr
718 static struct btrfs_block_group_cache *
719 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
721 struct btrfs_block_group_cache *cache;
723 cache = block_group_cache_tree_search(info, bytenr, 0);
729 * return the block group that contains the given bytenr
731 struct btrfs_block_group_cache *btrfs_lookup_block_group(
732 struct btrfs_fs_info *info,
735 struct btrfs_block_group_cache *cache;
737 cache = block_group_cache_tree_search(info, bytenr, 1);
742 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
745 struct list_head *head = &info->space_info;
746 struct btrfs_space_info *found;
748 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
751 list_for_each_entry_rcu(found, head, list) {
752 if (found->flags & flags) {
762 * after adding space to the filesystem, we need to clear the full flags
763 * on all the space infos.
765 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
767 struct list_head *head = &info->space_info;
768 struct btrfs_space_info *found;
771 list_for_each_entry_rcu(found, head, list)
776 /* simple helper to search for an existing data extent at a given offset */
777 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
780 struct btrfs_key key;
781 struct btrfs_path *path;
783 path = btrfs_alloc_path();
787 key.objectid = start;
789 key.type = BTRFS_EXTENT_ITEM_KEY;
790 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
792 btrfs_free_path(path);
797 * helper function to lookup reference count and flags of a tree block.
799 * the head node for delayed ref is used to store the sum of all the
800 * reference count modifications queued up in the rbtree. the head
801 * node may also store the extent flags to set. This way you can check
802 * to see what the reference count and extent flags would be if all of
803 * the delayed refs are not processed.
805 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
806 struct btrfs_root *root, u64 bytenr,
807 u64 offset, int metadata, u64 *refs, u64 *flags)
809 struct btrfs_delayed_ref_head *head;
810 struct btrfs_delayed_ref_root *delayed_refs;
811 struct btrfs_path *path;
812 struct btrfs_extent_item *ei;
813 struct extent_buffer *leaf;
814 struct btrfs_key key;
821 * If we don't have skinny metadata, don't bother doing anything
824 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
825 offset = root->nodesize;
829 path = btrfs_alloc_path();
834 path->skip_locking = 1;
835 path->search_commit_root = 1;
839 key.objectid = bytenr;
842 key.type = BTRFS_METADATA_ITEM_KEY;
844 key.type = BTRFS_EXTENT_ITEM_KEY;
846 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
851 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
852 if (path->slots[0]) {
854 btrfs_item_key_to_cpu(path->nodes[0], &key,
856 if (key.objectid == bytenr &&
857 key.type == BTRFS_EXTENT_ITEM_KEY &&
858 key.offset == root->nodesize)
864 leaf = path->nodes[0];
865 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
866 if (item_size >= sizeof(*ei)) {
867 ei = btrfs_item_ptr(leaf, path->slots[0],
868 struct btrfs_extent_item);
869 num_refs = btrfs_extent_refs(leaf, ei);
870 extent_flags = btrfs_extent_flags(leaf, ei);
872 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
873 struct btrfs_extent_item_v0 *ei0;
874 BUG_ON(item_size != sizeof(*ei0));
875 ei0 = btrfs_item_ptr(leaf, path->slots[0],
876 struct btrfs_extent_item_v0);
877 num_refs = btrfs_extent_refs_v0(leaf, ei0);
878 /* FIXME: this isn't correct for data */
879 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
884 BUG_ON(num_refs == 0);
894 delayed_refs = &trans->transaction->delayed_refs;
895 spin_lock(&delayed_refs->lock);
896 head = btrfs_find_delayed_ref_head(trans, bytenr);
898 if (!mutex_trylock(&head->mutex)) {
899 atomic_inc(&head->node.refs);
900 spin_unlock(&delayed_refs->lock);
902 btrfs_release_path(path);
905 * Mutex was contended, block until it's released and try
908 mutex_lock(&head->mutex);
909 mutex_unlock(&head->mutex);
910 btrfs_put_delayed_ref(&head->node);
913 spin_lock(&head->lock);
914 if (head->extent_op && head->extent_op->update_flags)
915 extent_flags |= head->extent_op->flags_to_set;
917 BUG_ON(num_refs == 0);
919 num_refs += head->node.ref_mod;
920 spin_unlock(&head->lock);
921 mutex_unlock(&head->mutex);
923 spin_unlock(&delayed_refs->lock);
925 WARN_ON(num_refs == 0);
929 *flags = extent_flags;
931 btrfs_free_path(path);
936 * Back reference rules. Back refs have three main goals:
938 * 1) differentiate between all holders of references to an extent so that
939 * when a reference is dropped we can make sure it was a valid reference
940 * before freeing the extent.
942 * 2) Provide enough information to quickly find the holders of an extent
943 * if we notice a given block is corrupted or bad.
945 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
946 * maintenance. This is actually the same as #2, but with a slightly
947 * different use case.
949 * There are two kinds of back refs. The implicit back refs is optimized
950 * for pointers in non-shared tree blocks. For a given pointer in a block,
951 * back refs of this kind provide information about the block's owner tree
952 * and the pointer's key. These information allow us to find the block by
953 * b-tree searching. The full back refs is for pointers in tree blocks not
954 * referenced by their owner trees. The location of tree block is recorded
955 * in the back refs. Actually the full back refs is generic, and can be
956 * used in all cases the implicit back refs is used. The major shortcoming
957 * of the full back refs is its overhead. Every time a tree block gets
958 * COWed, we have to update back refs entry for all pointers in it.
960 * For a newly allocated tree block, we use implicit back refs for
961 * pointers in it. This means most tree related operations only involve
962 * implicit back refs. For a tree block created in old transaction, the
963 * only way to drop a reference to it is COW it. So we can detect the
964 * event that tree block loses its owner tree's reference and do the
965 * back refs conversion.
967 * When a tree block is COW'd through a tree, there are four cases:
969 * The reference count of the block is one and the tree is the block's
970 * owner tree. Nothing to do in this case.
972 * The reference count of the block is one and the tree is not the
973 * block's owner tree. In this case, full back refs is used for pointers
974 * in the block. Remove these full back refs, add implicit back refs for
975 * every pointers in the new block.
977 * The reference count of the block is greater than one and the tree is
978 * the block's owner tree. In this case, implicit back refs is used for
979 * pointers in the block. Add full back refs for every pointers in the
980 * block, increase lower level extents' reference counts. The original
981 * implicit back refs are entailed to the new block.
983 * The reference count of the block is greater than one and the tree is
984 * not the block's owner tree. Add implicit back refs for every pointer in
985 * the new block, increase lower level extents' reference count.
987 * Back Reference Key composing:
989 * The key objectid corresponds to the first byte in the extent,
990 * The key type is used to differentiate between types of back refs.
991 * There are different meanings of the key offset for different types
994 * File extents can be referenced by:
996 * - multiple snapshots, subvolumes, or different generations in one subvol
997 * - different files inside a single subvolume
998 * - different offsets inside a file (bookend extents in file.c)
1000 * The extent ref structure for the implicit back refs has fields for:
1002 * - Objectid of the subvolume root
1003 * - objectid of the file holding the reference
1004 * - original offset in the file
1005 * - how many bookend extents
1007 * The key offset for the implicit back refs is hash of the first
1010 * The extent ref structure for the full back refs has field for:
1012 * - number of pointers in the tree leaf
1014 * The key offset for the implicit back refs is the first byte of
1017 * When a file extent is allocated, The implicit back refs is used.
1018 * the fields are filled in:
1020 * (root_key.objectid, inode objectid, offset in file, 1)
1022 * When a file extent is removed file truncation, we find the
1023 * corresponding implicit back refs and check the following fields:
1025 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1027 * Btree extents can be referenced by:
1029 * - Different subvolumes
1031 * Both the implicit back refs and the full back refs for tree blocks
1032 * only consist of key. The key offset for the implicit back refs is
1033 * objectid of block's owner tree. The key offset for the full back refs
1034 * is the first byte of parent block.
1036 * When implicit back refs is used, information about the lowest key and
1037 * level of the tree block are required. These information are stored in
1038 * tree block info structure.
1041 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1042 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1043 struct btrfs_root *root,
1044 struct btrfs_path *path,
1045 u64 owner, u32 extra_size)
1047 struct btrfs_extent_item *item;
1048 struct btrfs_extent_item_v0 *ei0;
1049 struct btrfs_extent_ref_v0 *ref0;
1050 struct btrfs_tree_block_info *bi;
1051 struct extent_buffer *leaf;
1052 struct btrfs_key key;
1053 struct btrfs_key found_key;
1054 u32 new_size = sizeof(*item);
1058 leaf = path->nodes[0];
1059 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1061 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1062 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_extent_item_v0);
1064 refs = btrfs_extent_refs_v0(leaf, ei0);
1066 if (owner == (u64)-1) {
1068 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1069 ret = btrfs_next_leaf(root, path);
1072 BUG_ON(ret > 0); /* Corruption */
1073 leaf = path->nodes[0];
1075 btrfs_item_key_to_cpu(leaf, &found_key,
1077 BUG_ON(key.objectid != found_key.objectid);
1078 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1082 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1083 struct btrfs_extent_ref_v0);
1084 owner = btrfs_ref_objectid_v0(leaf, ref0);
1088 btrfs_release_path(path);
1090 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1091 new_size += sizeof(*bi);
1093 new_size -= sizeof(*ei0);
1094 ret = btrfs_search_slot(trans, root, &key, path,
1095 new_size + extra_size, 1);
1098 BUG_ON(ret); /* Corruption */
1100 btrfs_extend_item(root, path, new_size);
1102 leaf = path->nodes[0];
1103 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1104 btrfs_set_extent_refs(leaf, item, refs);
1105 /* FIXME: get real generation */
1106 btrfs_set_extent_generation(leaf, item, 0);
1107 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1108 btrfs_set_extent_flags(leaf, item,
1109 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1110 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1111 bi = (struct btrfs_tree_block_info *)(item + 1);
1112 /* FIXME: get first key of the block */
1113 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1114 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1116 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1118 btrfs_mark_buffer_dirty(leaf);
1123 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1125 u32 high_crc = ~(u32)0;
1126 u32 low_crc = ~(u32)0;
1129 lenum = cpu_to_le64(root_objectid);
1130 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1131 lenum = cpu_to_le64(owner);
1132 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1133 lenum = cpu_to_le64(offset);
1134 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1136 return ((u64)high_crc << 31) ^ (u64)low_crc;
1139 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1140 struct btrfs_extent_data_ref *ref)
1142 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1143 btrfs_extent_data_ref_objectid(leaf, ref),
1144 btrfs_extent_data_ref_offset(leaf, ref));
1147 static int match_extent_data_ref(struct extent_buffer *leaf,
1148 struct btrfs_extent_data_ref *ref,
1149 u64 root_objectid, u64 owner, u64 offset)
1151 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1152 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1153 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1158 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1159 struct btrfs_root *root,
1160 struct btrfs_path *path,
1161 u64 bytenr, u64 parent,
1163 u64 owner, u64 offset)
1165 struct btrfs_key key;
1166 struct btrfs_extent_data_ref *ref;
1167 struct extent_buffer *leaf;
1173 key.objectid = bytenr;
1175 key.type = BTRFS_SHARED_DATA_REF_KEY;
1176 key.offset = parent;
1178 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1179 key.offset = hash_extent_data_ref(root_objectid,
1184 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1193 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1194 key.type = BTRFS_EXTENT_REF_V0_KEY;
1195 btrfs_release_path(path);
1196 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1207 leaf = path->nodes[0];
1208 nritems = btrfs_header_nritems(leaf);
1210 if (path->slots[0] >= nritems) {
1211 ret = btrfs_next_leaf(root, path);
1217 leaf = path->nodes[0];
1218 nritems = btrfs_header_nritems(leaf);
1222 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1223 if (key.objectid != bytenr ||
1224 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1227 ref = btrfs_item_ptr(leaf, path->slots[0],
1228 struct btrfs_extent_data_ref);
1230 if (match_extent_data_ref(leaf, ref, root_objectid,
1233 btrfs_release_path(path);
1245 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1246 struct btrfs_root *root,
1247 struct btrfs_path *path,
1248 u64 bytenr, u64 parent,
1249 u64 root_objectid, u64 owner,
1250 u64 offset, int refs_to_add)
1252 struct btrfs_key key;
1253 struct extent_buffer *leaf;
1258 key.objectid = bytenr;
1260 key.type = BTRFS_SHARED_DATA_REF_KEY;
1261 key.offset = parent;
1262 size = sizeof(struct btrfs_shared_data_ref);
1264 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1265 key.offset = hash_extent_data_ref(root_objectid,
1267 size = sizeof(struct btrfs_extent_data_ref);
1270 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1271 if (ret && ret != -EEXIST)
1274 leaf = path->nodes[0];
1276 struct btrfs_shared_data_ref *ref;
1277 ref = btrfs_item_ptr(leaf, path->slots[0],
1278 struct btrfs_shared_data_ref);
1280 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1282 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1283 num_refs += refs_to_add;
1284 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1287 struct btrfs_extent_data_ref *ref;
1288 while (ret == -EEXIST) {
1289 ref = btrfs_item_ptr(leaf, path->slots[0],
1290 struct btrfs_extent_data_ref);
1291 if (match_extent_data_ref(leaf, ref, root_objectid,
1294 btrfs_release_path(path);
1296 ret = btrfs_insert_empty_item(trans, root, path, &key,
1298 if (ret && ret != -EEXIST)
1301 leaf = path->nodes[0];
1303 ref = btrfs_item_ptr(leaf, path->slots[0],
1304 struct btrfs_extent_data_ref);
1306 btrfs_set_extent_data_ref_root(leaf, ref,
1308 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1309 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1310 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1312 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1313 num_refs += refs_to_add;
1314 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1317 btrfs_mark_buffer_dirty(leaf);
1320 btrfs_release_path(path);
1324 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1325 struct btrfs_root *root,
1326 struct btrfs_path *path,
1327 int refs_to_drop, int *last_ref)
1329 struct btrfs_key key;
1330 struct btrfs_extent_data_ref *ref1 = NULL;
1331 struct btrfs_shared_data_ref *ref2 = NULL;
1332 struct extent_buffer *leaf;
1336 leaf = path->nodes[0];
1337 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1339 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1340 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1341 struct btrfs_extent_data_ref);
1342 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1343 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1344 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1345 struct btrfs_shared_data_ref);
1346 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1347 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1348 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1349 struct btrfs_extent_ref_v0 *ref0;
1350 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1351 struct btrfs_extent_ref_v0);
1352 num_refs = btrfs_ref_count_v0(leaf, ref0);
1358 BUG_ON(num_refs < refs_to_drop);
1359 num_refs -= refs_to_drop;
1361 if (num_refs == 0) {
1362 ret = btrfs_del_item(trans, root, path);
1365 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1366 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1367 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1368 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1369 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1371 struct btrfs_extent_ref_v0 *ref0;
1372 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1373 struct btrfs_extent_ref_v0);
1374 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1377 btrfs_mark_buffer_dirty(leaf);
1382 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1383 struct btrfs_extent_inline_ref *iref)
1385 struct btrfs_key key;
1386 struct extent_buffer *leaf;
1387 struct btrfs_extent_data_ref *ref1;
1388 struct btrfs_shared_data_ref *ref2;
1391 leaf = path->nodes[0];
1392 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1394 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1395 BTRFS_EXTENT_DATA_REF_KEY) {
1396 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1397 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1399 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1400 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1402 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1403 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1404 struct btrfs_extent_data_ref);
1405 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1406 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1407 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1408 struct btrfs_shared_data_ref);
1409 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1410 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1411 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1412 struct btrfs_extent_ref_v0 *ref0;
1413 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1414 struct btrfs_extent_ref_v0);
1415 num_refs = btrfs_ref_count_v0(leaf, ref0);
1423 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1424 struct btrfs_root *root,
1425 struct btrfs_path *path,
1426 u64 bytenr, u64 parent,
1429 struct btrfs_key key;
1432 key.objectid = bytenr;
1434 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1435 key.offset = parent;
1437 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1438 key.offset = root_objectid;
1441 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1444 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1445 if (ret == -ENOENT && parent) {
1446 btrfs_release_path(path);
1447 key.type = BTRFS_EXTENT_REF_V0_KEY;
1448 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1456 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1457 struct btrfs_root *root,
1458 struct btrfs_path *path,
1459 u64 bytenr, u64 parent,
1462 struct btrfs_key key;
1465 key.objectid = bytenr;
1467 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1468 key.offset = parent;
1470 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1471 key.offset = root_objectid;
1474 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1475 btrfs_release_path(path);
1479 static inline int extent_ref_type(u64 parent, u64 owner)
1482 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1484 type = BTRFS_SHARED_BLOCK_REF_KEY;
1486 type = BTRFS_TREE_BLOCK_REF_KEY;
1489 type = BTRFS_SHARED_DATA_REF_KEY;
1491 type = BTRFS_EXTENT_DATA_REF_KEY;
1496 static int find_next_key(struct btrfs_path *path, int level,
1497 struct btrfs_key *key)
1500 for (; level < BTRFS_MAX_LEVEL; level++) {
1501 if (!path->nodes[level])
1503 if (path->slots[level] + 1 >=
1504 btrfs_header_nritems(path->nodes[level]))
1507 btrfs_item_key_to_cpu(path->nodes[level], key,
1508 path->slots[level] + 1);
1510 btrfs_node_key_to_cpu(path->nodes[level], key,
1511 path->slots[level] + 1);
1518 * look for inline back ref. if back ref is found, *ref_ret is set
1519 * to the address of inline back ref, and 0 is returned.
1521 * if back ref isn't found, *ref_ret is set to the address where it
1522 * should be inserted, and -ENOENT is returned.
1524 * if insert is true and there are too many inline back refs, the path
1525 * points to the extent item, and -EAGAIN is returned.
1527 * NOTE: inline back refs are ordered in the same way that back ref
1528 * items in the tree are ordered.
1530 static noinline_for_stack
1531 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1532 struct btrfs_root *root,
1533 struct btrfs_path *path,
1534 struct btrfs_extent_inline_ref **ref_ret,
1535 u64 bytenr, u64 num_bytes,
1536 u64 parent, u64 root_objectid,
1537 u64 owner, u64 offset, int insert)
1539 struct btrfs_key key;
1540 struct extent_buffer *leaf;
1541 struct btrfs_extent_item *ei;
1542 struct btrfs_extent_inline_ref *iref;
1552 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1555 key.objectid = bytenr;
1556 key.type = BTRFS_EXTENT_ITEM_KEY;
1557 key.offset = num_bytes;
1559 want = extent_ref_type(parent, owner);
1561 extra_size = btrfs_extent_inline_ref_size(want);
1562 path->keep_locks = 1;
1567 * Owner is our parent level, so we can just add one to get the level
1568 * for the block we are interested in.
1570 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1571 key.type = BTRFS_METADATA_ITEM_KEY;
1576 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1583 * We may be a newly converted file system which still has the old fat
1584 * extent entries for metadata, so try and see if we have one of those.
1586 if (ret > 0 && skinny_metadata) {
1587 skinny_metadata = false;
1588 if (path->slots[0]) {
1590 btrfs_item_key_to_cpu(path->nodes[0], &key,
1592 if (key.objectid == bytenr &&
1593 key.type == BTRFS_EXTENT_ITEM_KEY &&
1594 key.offset == num_bytes)
1598 key.objectid = bytenr;
1599 key.type = BTRFS_EXTENT_ITEM_KEY;
1600 key.offset = num_bytes;
1601 btrfs_release_path(path);
1606 if (ret && !insert) {
1609 } else if (WARN_ON(ret)) {
1614 leaf = path->nodes[0];
1615 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1616 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1617 if (item_size < sizeof(*ei)) {
1622 ret = convert_extent_item_v0(trans, root, path, owner,
1628 leaf = path->nodes[0];
1629 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1632 BUG_ON(item_size < sizeof(*ei));
1634 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1635 flags = btrfs_extent_flags(leaf, ei);
1637 ptr = (unsigned long)(ei + 1);
1638 end = (unsigned long)ei + item_size;
1640 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1641 ptr += sizeof(struct btrfs_tree_block_info);
1651 iref = (struct btrfs_extent_inline_ref *)ptr;
1652 type = btrfs_extent_inline_ref_type(leaf, iref);
1656 ptr += btrfs_extent_inline_ref_size(type);
1660 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1661 struct btrfs_extent_data_ref *dref;
1662 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1663 if (match_extent_data_ref(leaf, dref, root_objectid,
1668 if (hash_extent_data_ref_item(leaf, dref) <
1669 hash_extent_data_ref(root_objectid, owner, offset))
1673 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1675 if (parent == ref_offset) {
1679 if (ref_offset < parent)
1682 if (root_objectid == ref_offset) {
1686 if (ref_offset < root_objectid)
1690 ptr += btrfs_extent_inline_ref_size(type);
1692 if (err == -ENOENT && insert) {
1693 if (item_size + extra_size >=
1694 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1699 * To add new inline back ref, we have to make sure
1700 * there is no corresponding back ref item.
1701 * For simplicity, we just do not add new inline back
1702 * ref if there is any kind of item for this block
1704 if (find_next_key(path, 0, &key) == 0 &&
1705 key.objectid == bytenr &&
1706 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1711 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1714 path->keep_locks = 0;
1715 btrfs_unlock_up_safe(path, 1);
1721 * helper to add new inline back ref
1723 static noinline_for_stack
1724 void setup_inline_extent_backref(struct btrfs_root *root,
1725 struct btrfs_path *path,
1726 struct btrfs_extent_inline_ref *iref,
1727 u64 parent, u64 root_objectid,
1728 u64 owner, u64 offset, int refs_to_add,
1729 struct btrfs_delayed_extent_op *extent_op)
1731 struct extent_buffer *leaf;
1732 struct btrfs_extent_item *ei;
1735 unsigned long item_offset;
1740 leaf = path->nodes[0];
1741 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1742 item_offset = (unsigned long)iref - (unsigned long)ei;
1744 type = extent_ref_type(parent, owner);
1745 size = btrfs_extent_inline_ref_size(type);
1747 btrfs_extend_item(root, path, size);
1749 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1750 refs = btrfs_extent_refs(leaf, ei);
1751 refs += refs_to_add;
1752 btrfs_set_extent_refs(leaf, ei, refs);
1754 __run_delayed_extent_op(extent_op, leaf, ei);
1756 ptr = (unsigned long)ei + item_offset;
1757 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1758 if (ptr < end - size)
1759 memmove_extent_buffer(leaf, ptr + size, ptr,
1762 iref = (struct btrfs_extent_inline_ref *)ptr;
1763 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1764 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1765 struct btrfs_extent_data_ref *dref;
1766 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1767 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1768 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1769 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1770 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1771 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1772 struct btrfs_shared_data_ref *sref;
1773 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1774 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1775 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1776 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1777 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1779 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1781 btrfs_mark_buffer_dirty(leaf);
1784 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1785 struct btrfs_root *root,
1786 struct btrfs_path *path,
1787 struct btrfs_extent_inline_ref **ref_ret,
1788 u64 bytenr, u64 num_bytes, u64 parent,
1789 u64 root_objectid, u64 owner, u64 offset)
1793 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1794 bytenr, num_bytes, parent,
1795 root_objectid, owner, offset, 0);
1799 btrfs_release_path(path);
1802 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1803 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1806 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1807 root_objectid, owner, offset);
1813 * helper to update/remove inline back ref
1815 static noinline_for_stack
1816 void update_inline_extent_backref(struct btrfs_root *root,
1817 struct btrfs_path *path,
1818 struct btrfs_extent_inline_ref *iref,
1820 struct btrfs_delayed_extent_op *extent_op,
1823 struct extent_buffer *leaf;
1824 struct btrfs_extent_item *ei;
1825 struct btrfs_extent_data_ref *dref = NULL;
1826 struct btrfs_shared_data_ref *sref = NULL;
1834 leaf = path->nodes[0];
1835 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1836 refs = btrfs_extent_refs(leaf, ei);
1837 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1838 refs += refs_to_mod;
1839 btrfs_set_extent_refs(leaf, ei, refs);
1841 __run_delayed_extent_op(extent_op, leaf, ei);
1843 type = btrfs_extent_inline_ref_type(leaf, iref);
1845 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1846 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1847 refs = btrfs_extent_data_ref_count(leaf, dref);
1848 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1849 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1850 refs = btrfs_shared_data_ref_count(leaf, sref);
1853 BUG_ON(refs_to_mod != -1);
1856 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1857 refs += refs_to_mod;
1860 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1861 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1863 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1866 size = btrfs_extent_inline_ref_size(type);
1867 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1868 ptr = (unsigned long)iref;
1869 end = (unsigned long)ei + item_size;
1870 if (ptr + size < end)
1871 memmove_extent_buffer(leaf, ptr, ptr + size,
1874 btrfs_truncate_item(root, path, item_size, 1);
1876 btrfs_mark_buffer_dirty(leaf);
1879 static noinline_for_stack
1880 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1881 struct btrfs_root *root,
1882 struct btrfs_path *path,
1883 u64 bytenr, u64 num_bytes, u64 parent,
1884 u64 root_objectid, u64 owner,
1885 u64 offset, int refs_to_add,
1886 struct btrfs_delayed_extent_op *extent_op)
1888 struct btrfs_extent_inline_ref *iref;
1891 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1892 bytenr, num_bytes, parent,
1893 root_objectid, owner, offset, 1);
1895 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1896 update_inline_extent_backref(root, path, iref,
1897 refs_to_add, extent_op, NULL);
1898 } else if (ret == -ENOENT) {
1899 setup_inline_extent_backref(root, path, iref, parent,
1900 root_objectid, owner, offset,
1901 refs_to_add, extent_op);
1907 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1908 struct btrfs_root *root,
1909 struct btrfs_path *path,
1910 u64 bytenr, u64 parent, u64 root_objectid,
1911 u64 owner, u64 offset, int refs_to_add)
1914 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1915 BUG_ON(refs_to_add != 1);
1916 ret = insert_tree_block_ref(trans, root, path, bytenr,
1917 parent, root_objectid);
1919 ret = insert_extent_data_ref(trans, root, path, bytenr,
1920 parent, root_objectid,
1921 owner, offset, refs_to_add);
1926 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1927 struct btrfs_root *root,
1928 struct btrfs_path *path,
1929 struct btrfs_extent_inline_ref *iref,
1930 int refs_to_drop, int is_data, int *last_ref)
1934 BUG_ON(!is_data && refs_to_drop != 1);
1936 update_inline_extent_backref(root, path, iref,
1937 -refs_to_drop, NULL, last_ref);
1938 } else if (is_data) {
1939 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1943 ret = btrfs_del_item(trans, root, path);
1948 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1949 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1950 u64 *discarded_bytes)
1953 u64 bytes_left, end;
1954 u64 aligned_start = ALIGN(start, 1 << 9);
1956 if (WARN_ON(start != aligned_start)) {
1957 len -= aligned_start - start;
1958 len = round_down(len, 1 << 9);
1959 start = aligned_start;
1962 *discarded_bytes = 0;
1970 /* Skip any superblocks on this device. */
1971 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1972 u64 sb_start = btrfs_sb_offset(j);
1973 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1974 u64 size = sb_start - start;
1976 if (!in_range(sb_start, start, bytes_left) &&
1977 !in_range(sb_end, start, bytes_left) &&
1978 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1982 * Superblock spans beginning of range. Adjust start and
1985 if (sb_start <= start) {
1986 start += sb_end - start;
1991 bytes_left = end - start;
1996 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
1999 *discarded_bytes += size;
2000 else if (ret != -EOPNOTSUPP)
2009 bytes_left = end - start;
2013 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2016 *discarded_bytes += bytes_left;
2021 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2022 u64 num_bytes, u64 *actual_bytes)
2025 u64 discarded_bytes = 0;
2026 struct btrfs_bio *bbio = NULL;
2029 /* Tell the block device(s) that the sectors can be discarded */
2030 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2031 bytenr, &num_bytes, &bbio, 0);
2032 /* Error condition is -ENOMEM */
2034 struct btrfs_bio_stripe *stripe = bbio->stripes;
2038 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2040 if (!stripe->dev->can_discard)
2043 ret = btrfs_issue_discard(stripe->dev->bdev,
2048 discarded_bytes += bytes;
2049 else if (ret != -EOPNOTSUPP)
2050 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2053 * Just in case we get back EOPNOTSUPP for some reason,
2054 * just ignore the return value so we don't screw up
2055 * people calling discard_extent.
2059 btrfs_put_bbio(bbio);
2063 *actual_bytes = discarded_bytes;
2066 if (ret == -EOPNOTSUPP)
2071 /* Can return -ENOMEM */
2072 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2073 struct btrfs_root *root,
2074 u64 bytenr, u64 num_bytes, u64 parent,
2075 u64 root_objectid, u64 owner, u64 offset)
2078 struct btrfs_fs_info *fs_info = root->fs_info;
2080 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2081 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2083 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2084 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2086 parent, root_objectid, (int)owner,
2087 BTRFS_ADD_DELAYED_REF, NULL);
2089 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2090 num_bytes, parent, root_objectid,
2092 BTRFS_ADD_DELAYED_REF, NULL);
2097 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2098 struct btrfs_root *root,
2099 struct btrfs_delayed_ref_node *node,
2100 u64 parent, u64 root_objectid,
2101 u64 owner, u64 offset, int refs_to_add,
2102 struct btrfs_delayed_extent_op *extent_op)
2104 struct btrfs_fs_info *fs_info = root->fs_info;
2105 struct btrfs_path *path;
2106 struct extent_buffer *leaf;
2107 struct btrfs_extent_item *item;
2108 struct btrfs_key key;
2109 u64 bytenr = node->bytenr;
2110 u64 num_bytes = node->num_bytes;
2114 path = btrfs_alloc_path();
2119 path->leave_spinning = 1;
2120 /* this will setup the path even if it fails to insert the back ref */
2121 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2122 bytenr, num_bytes, parent,
2123 root_objectid, owner, offset,
2124 refs_to_add, extent_op);
2125 if ((ret < 0 && ret != -EAGAIN) || !ret)
2129 * Ok we had -EAGAIN which means we didn't have space to insert and
2130 * inline extent ref, so just update the reference count and add a
2133 leaf = path->nodes[0];
2134 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2135 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2136 refs = btrfs_extent_refs(leaf, item);
2137 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2139 __run_delayed_extent_op(extent_op, leaf, item);
2141 btrfs_mark_buffer_dirty(leaf);
2142 btrfs_release_path(path);
2145 path->leave_spinning = 1;
2146 /* now insert the actual backref */
2147 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2148 path, bytenr, parent, root_objectid,
2149 owner, offset, refs_to_add);
2151 btrfs_abort_transaction(trans, root, ret);
2153 btrfs_free_path(path);
2157 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2158 struct btrfs_root *root,
2159 struct btrfs_delayed_ref_node *node,
2160 struct btrfs_delayed_extent_op *extent_op,
2161 int insert_reserved)
2164 struct btrfs_delayed_data_ref *ref;
2165 struct btrfs_key ins;
2170 ins.objectid = node->bytenr;
2171 ins.offset = node->num_bytes;
2172 ins.type = BTRFS_EXTENT_ITEM_KEY;
2174 ref = btrfs_delayed_node_to_data_ref(node);
2175 trace_run_delayed_data_ref(node, ref, node->action);
2177 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2178 parent = ref->parent;
2179 ref_root = ref->root;
2181 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2183 flags |= extent_op->flags_to_set;
2184 ret = alloc_reserved_file_extent(trans, root,
2185 parent, ref_root, flags,
2186 ref->objectid, ref->offset,
2187 &ins, node->ref_mod);
2188 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2189 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2190 ref_root, ref->objectid,
2191 ref->offset, node->ref_mod,
2193 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2194 ret = __btrfs_free_extent(trans, root, node, parent,
2195 ref_root, ref->objectid,
2196 ref->offset, node->ref_mod,
2204 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2205 struct extent_buffer *leaf,
2206 struct btrfs_extent_item *ei)
2208 u64 flags = btrfs_extent_flags(leaf, ei);
2209 if (extent_op->update_flags) {
2210 flags |= extent_op->flags_to_set;
2211 btrfs_set_extent_flags(leaf, ei, flags);
2214 if (extent_op->update_key) {
2215 struct btrfs_tree_block_info *bi;
2216 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2217 bi = (struct btrfs_tree_block_info *)(ei + 1);
2218 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2222 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2223 struct btrfs_root *root,
2224 struct btrfs_delayed_ref_node *node,
2225 struct btrfs_delayed_extent_op *extent_op)
2227 struct btrfs_key key;
2228 struct btrfs_path *path;
2229 struct btrfs_extent_item *ei;
2230 struct extent_buffer *leaf;
2234 int metadata = !extent_op->is_data;
2239 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2242 path = btrfs_alloc_path();
2246 key.objectid = node->bytenr;
2249 key.type = BTRFS_METADATA_ITEM_KEY;
2250 key.offset = extent_op->level;
2252 key.type = BTRFS_EXTENT_ITEM_KEY;
2253 key.offset = node->num_bytes;
2258 path->leave_spinning = 1;
2259 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2267 if (path->slots[0] > 0) {
2269 btrfs_item_key_to_cpu(path->nodes[0], &key,
2271 if (key.objectid == node->bytenr &&
2272 key.type == BTRFS_EXTENT_ITEM_KEY &&
2273 key.offset == node->num_bytes)
2277 btrfs_release_path(path);
2280 key.objectid = node->bytenr;
2281 key.offset = node->num_bytes;
2282 key.type = BTRFS_EXTENT_ITEM_KEY;
2291 leaf = path->nodes[0];
2292 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2293 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2294 if (item_size < sizeof(*ei)) {
2295 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2301 leaf = path->nodes[0];
2302 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2305 BUG_ON(item_size < sizeof(*ei));
2306 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2307 __run_delayed_extent_op(extent_op, leaf, ei);
2309 btrfs_mark_buffer_dirty(leaf);
2311 btrfs_free_path(path);
2315 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2316 struct btrfs_root *root,
2317 struct btrfs_delayed_ref_node *node,
2318 struct btrfs_delayed_extent_op *extent_op,
2319 int insert_reserved)
2322 struct btrfs_delayed_tree_ref *ref;
2323 struct btrfs_key ins;
2326 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2329 ref = btrfs_delayed_node_to_tree_ref(node);
2330 trace_run_delayed_tree_ref(node, ref, node->action);
2332 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2333 parent = ref->parent;
2334 ref_root = ref->root;
2336 ins.objectid = node->bytenr;
2337 if (skinny_metadata) {
2338 ins.offset = ref->level;
2339 ins.type = BTRFS_METADATA_ITEM_KEY;
2341 ins.offset = node->num_bytes;
2342 ins.type = BTRFS_EXTENT_ITEM_KEY;
2345 BUG_ON(node->ref_mod != 1);
2346 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2347 BUG_ON(!extent_op || !extent_op->update_flags);
2348 ret = alloc_reserved_tree_block(trans, root,
2350 extent_op->flags_to_set,
2353 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2354 ret = __btrfs_inc_extent_ref(trans, root, node,
2358 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2359 ret = __btrfs_free_extent(trans, root, node,
2361 ref->level, 0, 1, extent_op);
2368 /* helper function to actually process a single delayed ref entry */
2369 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2370 struct btrfs_root *root,
2371 struct btrfs_delayed_ref_node *node,
2372 struct btrfs_delayed_extent_op *extent_op,
2373 int insert_reserved)
2377 if (trans->aborted) {
2378 if (insert_reserved)
2379 btrfs_pin_extent(root, node->bytenr,
2380 node->num_bytes, 1);
2384 if (btrfs_delayed_ref_is_head(node)) {
2385 struct btrfs_delayed_ref_head *head;
2387 * we've hit the end of the chain and we were supposed
2388 * to insert this extent into the tree. But, it got
2389 * deleted before we ever needed to insert it, so all
2390 * we have to do is clean up the accounting
2393 head = btrfs_delayed_node_to_head(node);
2394 trace_run_delayed_ref_head(node, head, node->action);
2396 if (insert_reserved) {
2397 btrfs_pin_extent(root, node->bytenr,
2398 node->num_bytes, 1);
2399 if (head->is_data) {
2400 ret = btrfs_del_csums(trans, root,
2406 /* Also free its reserved qgroup space */
2407 btrfs_qgroup_free_delayed_ref(root->fs_info,
2408 head->qgroup_ref_root,
2409 head->qgroup_reserved);
2413 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2414 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2415 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2417 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2418 node->type == BTRFS_SHARED_DATA_REF_KEY)
2419 ret = run_delayed_data_ref(trans, root, node, extent_op,
2426 static inline struct btrfs_delayed_ref_node *
2427 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2429 struct btrfs_delayed_ref_node *ref;
2431 if (list_empty(&head->ref_list))
2435 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2436 * This is to prevent a ref count from going down to zero, which deletes
2437 * the extent item from the extent tree, when there still are references
2438 * to add, which would fail because they would not find the extent item.
2440 list_for_each_entry(ref, &head->ref_list, list) {
2441 if (ref->action == BTRFS_ADD_DELAYED_REF)
2445 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2450 * Returns 0 on success or if called with an already aborted transaction.
2451 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2453 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2454 struct btrfs_root *root,
2457 struct btrfs_delayed_ref_root *delayed_refs;
2458 struct btrfs_delayed_ref_node *ref;
2459 struct btrfs_delayed_ref_head *locked_ref = NULL;
2460 struct btrfs_delayed_extent_op *extent_op;
2461 struct btrfs_fs_info *fs_info = root->fs_info;
2462 ktime_t start = ktime_get();
2464 unsigned long count = 0;
2465 unsigned long actual_count = 0;
2466 int must_insert_reserved = 0;
2468 delayed_refs = &trans->transaction->delayed_refs;
2474 spin_lock(&delayed_refs->lock);
2475 locked_ref = btrfs_select_ref_head(trans);
2477 spin_unlock(&delayed_refs->lock);
2481 /* grab the lock that says we are going to process
2482 * all the refs for this head */
2483 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2484 spin_unlock(&delayed_refs->lock);
2486 * we may have dropped the spin lock to get the head
2487 * mutex lock, and that might have given someone else
2488 * time to free the head. If that's true, it has been
2489 * removed from our list and we can move on.
2491 if (ret == -EAGAIN) {
2499 * We need to try and merge add/drops of the same ref since we
2500 * can run into issues with relocate dropping the implicit ref
2501 * and then it being added back again before the drop can
2502 * finish. If we merged anything we need to re-loop so we can
2504 * Or we can get node references of the same type that weren't
2505 * merged when created due to bumps in the tree mod seq, and
2506 * we need to merge them to prevent adding an inline extent
2507 * backref before dropping it (triggering a BUG_ON at
2508 * insert_inline_extent_backref()).
2510 spin_lock(&locked_ref->lock);
2511 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2515 * locked_ref is the head node, so we have to go one
2516 * node back for any delayed ref updates
2518 ref = select_delayed_ref(locked_ref);
2520 if (ref && ref->seq &&
2521 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2522 spin_unlock(&locked_ref->lock);
2523 btrfs_delayed_ref_unlock(locked_ref);
2524 spin_lock(&delayed_refs->lock);
2525 locked_ref->processing = 0;
2526 delayed_refs->num_heads_ready++;
2527 spin_unlock(&delayed_refs->lock);
2535 * record the must insert reserved flag before we
2536 * drop the spin lock.
2538 must_insert_reserved = locked_ref->must_insert_reserved;
2539 locked_ref->must_insert_reserved = 0;
2541 extent_op = locked_ref->extent_op;
2542 locked_ref->extent_op = NULL;
2547 /* All delayed refs have been processed, Go ahead
2548 * and send the head node to run_one_delayed_ref,
2549 * so that any accounting fixes can happen
2551 ref = &locked_ref->node;
2553 if (extent_op && must_insert_reserved) {
2554 btrfs_free_delayed_extent_op(extent_op);
2559 spin_unlock(&locked_ref->lock);
2560 ret = run_delayed_extent_op(trans, root,
2562 btrfs_free_delayed_extent_op(extent_op);
2566 * Need to reset must_insert_reserved if
2567 * there was an error so the abort stuff
2568 * can cleanup the reserved space
2571 if (must_insert_reserved)
2572 locked_ref->must_insert_reserved = 1;
2573 locked_ref->processing = 0;
2574 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2575 btrfs_delayed_ref_unlock(locked_ref);
2582 * Need to drop our head ref lock and re-aqcuire the
2583 * delayed ref lock and then re-check to make sure
2586 spin_unlock(&locked_ref->lock);
2587 spin_lock(&delayed_refs->lock);
2588 spin_lock(&locked_ref->lock);
2589 if (!list_empty(&locked_ref->ref_list) ||
2590 locked_ref->extent_op) {
2591 spin_unlock(&locked_ref->lock);
2592 spin_unlock(&delayed_refs->lock);
2596 delayed_refs->num_heads--;
2597 rb_erase(&locked_ref->href_node,
2598 &delayed_refs->href_root);
2599 spin_unlock(&delayed_refs->lock);
2603 list_del(&ref->list);
2605 atomic_dec(&delayed_refs->num_entries);
2607 if (!btrfs_delayed_ref_is_head(ref)) {
2609 * when we play the delayed ref, also correct the
2612 switch (ref->action) {
2613 case BTRFS_ADD_DELAYED_REF:
2614 case BTRFS_ADD_DELAYED_EXTENT:
2615 locked_ref->node.ref_mod -= ref->ref_mod;
2617 case BTRFS_DROP_DELAYED_REF:
2618 locked_ref->node.ref_mod += ref->ref_mod;
2624 spin_unlock(&locked_ref->lock);
2626 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2627 must_insert_reserved);
2629 btrfs_free_delayed_extent_op(extent_op);
2631 locked_ref->processing = 0;
2632 btrfs_delayed_ref_unlock(locked_ref);
2633 btrfs_put_delayed_ref(ref);
2634 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2639 * If this node is a head, that means all the refs in this head
2640 * have been dealt with, and we will pick the next head to deal
2641 * with, so we must unlock the head and drop it from the cluster
2642 * list before we release it.
2644 if (btrfs_delayed_ref_is_head(ref)) {
2645 if (locked_ref->is_data &&
2646 locked_ref->total_ref_mod < 0) {
2647 spin_lock(&delayed_refs->lock);
2648 delayed_refs->pending_csums -= ref->num_bytes;
2649 spin_unlock(&delayed_refs->lock);
2651 btrfs_delayed_ref_unlock(locked_ref);
2654 btrfs_put_delayed_ref(ref);
2660 * We don't want to include ref heads since we can have empty ref heads
2661 * and those will drastically skew our runtime down since we just do
2662 * accounting, no actual extent tree updates.
2664 if (actual_count > 0) {
2665 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2669 * We weigh the current average higher than our current runtime
2670 * to avoid large swings in the average.
2672 spin_lock(&delayed_refs->lock);
2673 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2674 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2675 spin_unlock(&delayed_refs->lock);
2680 #ifdef SCRAMBLE_DELAYED_REFS
2682 * Normally delayed refs get processed in ascending bytenr order. This
2683 * correlates in most cases to the order added. To expose dependencies on this
2684 * order, we start to process the tree in the middle instead of the beginning
2686 static u64 find_middle(struct rb_root *root)
2688 struct rb_node *n = root->rb_node;
2689 struct btrfs_delayed_ref_node *entry;
2692 u64 first = 0, last = 0;
2696 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2697 first = entry->bytenr;
2701 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2702 last = entry->bytenr;
2707 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2708 WARN_ON(!entry->in_tree);
2710 middle = entry->bytenr;
2723 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2727 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2728 sizeof(struct btrfs_extent_inline_ref));
2729 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2730 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2733 * We don't ever fill up leaves all the way so multiply by 2 just to be
2734 * closer to what we're really going to want to ouse.
2736 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2740 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2741 * would require to store the csums for that many bytes.
2743 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2746 u64 num_csums_per_leaf;
2749 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2750 num_csums_per_leaf = div64_u64(csum_size,
2751 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2752 num_csums = div64_u64(csum_bytes, root->sectorsize);
2753 num_csums += num_csums_per_leaf - 1;
2754 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2758 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2759 struct btrfs_root *root)
2761 struct btrfs_block_rsv *global_rsv;
2762 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2763 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2764 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2765 u64 num_bytes, num_dirty_bgs_bytes;
2768 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2769 num_heads = heads_to_leaves(root, num_heads);
2771 num_bytes += (num_heads - 1) * root->nodesize;
2773 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2774 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2776 global_rsv = &root->fs_info->global_block_rsv;
2779 * If we can't allocate any more chunks lets make sure we have _lots_ of
2780 * wiggle room since running delayed refs can create more delayed refs.
2782 if (global_rsv->space_info->full) {
2783 num_dirty_bgs_bytes <<= 1;
2787 spin_lock(&global_rsv->lock);
2788 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2790 spin_unlock(&global_rsv->lock);
2794 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2795 struct btrfs_root *root)
2797 struct btrfs_fs_info *fs_info = root->fs_info;
2799 atomic_read(&trans->transaction->delayed_refs.num_entries);
2804 avg_runtime = fs_info->avg_delayed_ref_runtime;
2805 val = num_entries * avg_runtime;
2806 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2808 if (val >= NSEC_PER_SEC / 2)
2811 return btrfs_check_space_for_delayed_refs(trans, root);
2814 struct async_delayed_refs {
2815 struct btrfs_root *root;
2819 struct completion wait;
2820 struct btrfs_work work;
2823 static void delayed_ref_async_start(struct btrfs_work *work)
2825 struct async_delayed_refs *async;
2826 struct btrfs_trans_handle *trans;
2829 async = container_of(work, struct async_delayed_refs, work);
2831 trans = btrfs_join_transaction(async->root);
2832 if (IS_ERR(trans)) {
2833 async->error = PTR_ERR(trans);
2838 * trans->sync means that when we call end_transaciton, we won't
2839 * wait on delayed refs
2842 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2846 ret = btrfs_end_transaction(trans, async->root);
2847 if (ret && !async->error)
2851 complete(&async->wait);
2856 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2857 unsigned long count, int wait)
2859 struct async_delayed_refs *async;
2862 async = kmalloc(sizeof(*async), GFP_NOFS);
2866 async->root = root->fs_info->tree_root;
2867 async->count = count;
2873 init_completion(&async->wait);
2875 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2876 delayed_ref_async_start, NULL, NULL);
2878 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2881 wait_for_completion(&async->wait);
2890 * this starts processing the delayed reference count updates and
2891 * extent insertions we have queued up so far. count can be
2892 * 0, which means to process everything in the tree at the start
2893 * of the run (but not newly added entries), or it can be some target
2894 * number you'd like to process.
2896 * Returns 0 on success or if called with an aborted transaction
2897 * Returns <0 on error and aborts the transaction
2899 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2900 struct btrfs_root *root, unsigned long count)
2902 struct rb_node *node;
2903 struct btrfs_delayed_ref_root *delayed_refs;
2904 struct btrfs_delayed_ref_head *head;
2906 int run_all = count == (unsigned long)-1;
2907 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2909 /* We'll clean this up in btrfs_cleanup_transaction */
2913 if (root == root->fs_info->extent_root)
2914 root = root->fs_info->tree_root;
2916 delayed_refs = &trans->transaction->delayed_refs;
2918 count = atomic_read(&delayed_refs->num_entries) * 2;
2921 #ifdef SCRAMBLE_DELAYED_REFS
2922 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2924 trans->can_flush_pending_bgs = false;
2925 ret = __btrfs_run_delayed_refs(trans, root, count);
2927 btrfs_abort_transaction(trans, root, ret);
2932 if (!list_empty(&trans->new_bgs))
2933 btrfs_create_pending_block_groups(trans, root);
2935 spin_lock(&delayed_refs->lock);
2936 node = rb_first(&delayed_refs->href_root);
2938 spin_unlock(&delayed_refs->lock);
2941 count = (unsigned long)-1;
2944 head = rb_entry(node, struct btrfs_delayed_ref_head,
2946 if (btrfs_delayed_ref_is_head(&head->node)) {
2947 struct btrfs_delayed_ref_node *ref;
2950 atomic_inc(&ref->refs);
2952 spin_unlock(&delayed_refs->lock);
2954 * Mutex was contended, block until it's
2955 * released and try again
2957 mutex_lock(&head->mutex);
2958 mutex_unlock(&head->mutex);
2960 btrfs_put_delayed_ref(ref);
2966 node = rb_next(node);
2968 spin_unlock(&delayed_refs->lock);
2973 assert_qgroups_uptodate(trans);
2974 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2978 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2979 struct btrfs_root *root,
2980 u64 bytenr, u64 num_bytes, u64 flags,
2981 int level, int is_data)
2983 struct btrfs_delayed_extent_op *extent_op;
2986 extent_op = btrfs_alloc_delayed_extent_op();
2990 extent_op->flags_to_set = flags;
2991 extent_op->update_flags = 1;
2992 extent_op->update_key = 0;
2993 extent_op->is_data = is_data ? 1 : 0;
2994 extent_op->level = level;
2996 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2997 num_bytes, extent_op);
2999 btrfs_free_delayed_extent_op(extent_op);
3003 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3004 struct btrfs_root *root,
3005 struct btrfs_path *path,
3006 u64 objectid, u64 offset, u64 bytenr)
3008 struct btrfs_delayed_ref_head *head;
3009 struct btrfs_delayed_ref_node *ref;
3010 struct btrfs_delayed_data_ref *data_ref;
3011 struct btrfs_delayed_ref_root *delayed_refs;
3014 delayed_refs = &trans->transaction->delayed_refs;
3015 spin_lock(&delayed_refs->lock);
3016 head = btrfs_find_delayed_ref_head(trans, bytenr);
3018 spin_unlock(&delayed_refs->lock);
3022 if (!mutex_trylock(&head->mutex)) {
3023 atomic_inc(&head->node.refs);
3024 spin_unlock(&delayed_refs->lock);
3026 btrfs_release_path(path);
3029 * Mutex was contended, block until it's released and let
3032 mutex_lock(&head->mutex);
3033 mutex_unlock(&head->mutex);
3034 btrfs_put_delayed_ref(&head->node);
3037 spin_unlock(&delayed_refs->lock);
3039 spin_lock(&head->lock);
3040 list_for_each_entry(ref, &head->ref_list, list) {
3041 /* If it's a shared ref we know a cross reference exists */
3042 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3047 data_ref = btrfs_delayed_node_to_data_ref(ref);
3050 * If our ref doesn't match the one we're currently looking at
3051 * then we have a cross reference.
3053 if (data_ref->root != root->root_key.objectid ||
3054 data_ref->objectid != objectid ||
3055 data_ref->offset != offset) {
3060 spin_unlock(&head->lock);
3061 mutex_unlock(&head->mutex);
3065 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3066 struct btrfs_root *root,
3067 struct btrfs_path *path,
3068 u64 objectid, u64 offset, u64 bytenr)
3070 struct btrfs_root *extent_root = root->fs_info->extent_root;
3071 struct extent_buffer *leaf;
3072 struct btrfs_extent_data_ref *ref;
3073 struct btrfs_extent_inline_ref *iref;
3074 struct btrfs_extent_item *ei;
3075 struct btrfs_key key;
3079 key.objectid = bytenr;
3080 key.offset = (u64)-1;
3081 key.type = BTRFS_EXTENT_ITEM_KEY;
3083 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3086 BUG_ON(ret == 0); /* Corruption */
3089 if (path->slots[0] == 0)
3093 leaf = path->nodes[0];
3094 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3096 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3100 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3101 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3102 if (item_size < sizeof(*ei)) {
3103 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3107 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3109 if (item_size != sizeof(*ei) +
3110 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3113 if (btrfs_extent_generation(leaf, ei) <=
3114 btrfs_root_last_snapshot(&root->root_item))
3117 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3118 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3119 BTRFS_EXTENT_DATA_REF_KEY)
3122 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3123 if (btrfs_extent_refs(leaf, ei) !=
3124 btrfs_extent_data_ref_count(leaf, ref) ||
3125 btrfs_extent_data_ref_root(leaf, ref) !=
3126 root->root_key.objectid ||
3127 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3128 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3136 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3137 struct btrfs_root *root,
3138 u64 objectid, u64 offset, u64 bytenr)
3140 struct btrfs_path *path;
3144 path = btrfs_alloc_path();
3149 ret = check_committed_ref(trans, root, path, objectid,
3151 if (ret && ret != -ENOENT)
3154 ret2 = check_delayed_ref(trans, root, path, objectid,
3156 } while (ret2 == -EAGAIN);
3158 if (ret2 && ret2 != -ENOENT) {
3163 if (ret != -ENOENT || ret2 != -ENOENT)
3166 btrfs_free_path(path);
3167 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3172 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3173 struct btrfs_root *root,
3174 struct extent_buffer *buf,
3175 int full_backref, int inc)
3182 struct btrfs_key key;
3183 struct btrfs_file_extent_item *fi;
3187 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3188 u64, u64, u64, u64, u64, u64);
3191 if (btrfs_test_is_dummy_root(root))
3194 ref_root = btrfs_header_owner(buf);
3195 nritems = btrfs_header_nritems(buf);
3196 level = btrfs_header_level(buf);
3198 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3202 process_func = btrfs_inc_extent_ref;
3204 process_func = btrfs_free_extent;
3207 parent = buf->start;
3211 for (i = 0; i < nritems; i++) {
3213 btrfs_item_key_to_cpu(buf, &key, i);
3214 if (key.type != BTRFS_EXTENT_DATA_KEY)
3216 fi = btrfs_item_ptr(buf, i,
3217 struct btrfs_file_extent_item);
3218 if (btrfs_file_extent_type(buf, fi) ==
3219 BTRFS_FILE_EXTENT_INLINE)
3221 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3225 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3226 key.offset -= btrfs_file_extent_offset(buf, fi);
3227 ret = process_func(trans, root, bytenr, num_bytes,
3228 parent, ref_root, key.objectid,
3233 bytenr = btrfs_node_blockptr(buf, i);
3234 num_bytes = root->nodesize;
3235 ret = process_func(trans, root, bytenr, num_bytes,
3236 parent, ref_root, level - 1, 0);
3246 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3247 struct extent_buffer *buf, int full_backref)
3249 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3252 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3253 struct extent_buffer *buf, int full_backref)
3255 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3258 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3259 struct btrfs_root *root,
3260 struct btrfs_path *path,
3261 struct btrfs_block_group_cache *cache)
3264 struct btrfs_root *extent_root = root->fs_info->extent_root;
3266 struct extent_buffer *leaf;
3268 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3275 leaf = path->nodes[0];
3276 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3277 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3278 btrfs_mark_buffer_dirty(leaf);
3280 btrfs_release_path(path);
3285 static struct btrfs_block_group_cache *
3286 next_block_group(struct btrfs_root *root,
3287 struct btrfs_block_group_cache *cache)
3289 struct rb_node *node;
3291 spin_lock(&root->fs_info->block_group_cache_lock);
3293 /* If our block group was removed, we need a full search. */
3294 if (RB_EMPTY_NODE(&cache->cache_node)) {
3295 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3297 spin_unlock(&root->fs_info->block_group_cache_lock);
3298 btrfs_put_block_group(cache);
3299 cache = btrfs_lookup_first_block_group(root->fs_info,
3303 node = rb_next(&cache->cache_node);
3304 btrfs_put_block_group(cache);
3306 cache = rb_entry(node, struct btrfs_block_group_cache,
3308 btrfs_get_block_group(cache);
3311 spin_unlock(&root->fs_info->block_group_cache_lock);
3315 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3316 struct btrfs_trans_handle *trans,
3317 struct btrfs_path *path)
3319 struct btrfs_root *root = block_group->fs_info->tree_root;
3320 struct inode *inode = NULL;
3322 int dcs = BTRFS_DC_ERROR;
3328 * If this block group is smaller than 100 megs don't bother caching the
3331 if (block_group->key.offset < (100 * 1024 * 1024)) {
3332 spin_lock(&block_group->lock);
3333 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3334 spin_unlock(&block_group->lock);
3341 inode = lookup_free_space_inode(root, block_group, path);
3342 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3343 ret = PTR_ERR(inode);
3344 btrfs_release_path(path);
3348 if (IS_ERR(inode)) {
3352 if (block_group->ro)
3355 ret = create_free_space_inode(root, trans, block_group, path);
3361 /* We've already setup this transaction, go ahead and exit */
3362 if (block_group->cache_generation == trans->transid &&
3363 i_size_read(inode)) {
3364 dcs = BTRFS_DC_SETUP;
3369 * We want to set the generation to 0, that way if anything goes wrong
3370 * from here on out we know not to trust this cache when we load up next
3373 BTRFS_I(inode)->generation = 0;
3374 ret = btrfs_update_inode(trans, root, inode);
3377 * So theoretically we could recover from this, simply set the
3378 * super cache generation to 0 so we know to invalidate the
3379 * cache, but then we'd have to keep track of the block groups
3380 * that fail this way so we know we _have_ to reset this cache
3381 * before the next commit or risk reading stale cache. So to
3382 * limit our exposure to horrible edge cases lets just abort the
3383 * transaction, this only happens in really bad situations
3386 btrfs_abort_transaction(trans, root, ret);
3391 if (i_size_read(inode) > 0) {
3392 ret = btrfs_check_trunc_cache_free_space(root,
3393 &root->fs_info->global_block_rsv);
3397 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3402 spin_lock(&block_group->lock);
3403 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3404 !btrfs_test_opt(root, SPACE_CACHE)) {
3406 * don't bother trying to write stuff out _if_
3407 * a) we're not cached,
3408 * b) we're with nospace_cache mount option.
3410 dcs = BTRFS_DC_WRITTEN;
3411 spin_unlock(&block_group->lock);
3414 spin_unlock(&block_group->lock);
3417 * We hit an ENOSPC when setting up the cache in this transaction, just
3418 * skip doing the setup, we've already cleared the cache so we're safe.
3420 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3426 * Try to preallocate enough space based on how big the block group is.
3427 * Keep in mind this has to include any pinned space which could end up
3428 * taking up quite a bit since it's not folded into the other space
3431 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3436 num_pages *= PAGE_CACHE_SIZE;
3438 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3442 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3443 num_pages, num_pages,
3446 * Our cache requires contiguous chunks so that we don't modify a bunch
3447 * of metadata or split extents when writing the cache out, which means
3448 * we can enospc if we are heavily fragmented in addition to just normal
3449 * out of space conditions. So if we hit this just skip setting up any
3450 * other block groups for this transaction, maybe we'll unpin enough
3451 * space the next time around.
3454 dcs = BTRFS_DC_SETUP;
3455 else if (ret == -ENOSPC)
3456 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3457 btrfs_free_reserved_data_space(inode, 0, num_pages);
3462 btrfs_release_path(path);
3464 spin_lock(&block_group->lock);
3465 if (!ret && dcs == BTRFS_DC_SETUP)
3466 block_group->cache_generation = trans->transid;
3467 block_group->disk_cache_state = dcs;
3468 spin_unlock(&block_group->lock);
3473 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3474 struct btrfs_root *root)
3476 struct btrfs_block_group_cache *cache, *tmp;
3477 struct btrfs_transaction *cur_trans = trans->transaction;
3478 struct btrfs_path *path;
3480 if (list_empty(&cur_trans->dirty_bgs) ||
3481 !btrfs_test_opt(root, SPACE_CACHE))
3484 path = btrfs_alloc_path();
3488 /* Could add new block groups, use _safe just in case */
3489 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3491 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3492 cache_save_setup(cache, trans, path);
3495 btrfs_free_path(path);
3500 * transaction commit does final block group cache writeback during a
3501 * critical section where nothing is allowed to change the FS. This is
3502 * required in order for the cache to actually match the block group,
3503 * but can introduce a lot of latency into the commit.
3505 * So, btrfs_start_dirty_block_groups is here to kick off block group
3506 * cache IO. There's a chance we'll have to redo some of it if the
3507 * block group changes again during the commit, but it greatly reduces
3508 * the commit latency by getting rid of the easy block groups while
3509 * we're still allowing others to join the commit.
3511 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3512 struct btrfs_root *root)
3514 struct btrfs_block_group_cache *cache;
3515 struct btrfs_transaction *cur_trans = trans->transaction;
3518 struct btrfs_path *path = NULL;
3520 struct list_head *io = &cur_trans->io_bgs;
3521 int num_started = 0;
3524 spin_lock(&cur_trans->dirty_bgs_lock);
3525 if (list_empty(&cur_trans->dirty_bgs)) {
3526 spin_unlock(&cur_trans->dirty_bgs_lock);
3529 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3530 spin_unlock(&cur_trans->dirty_bgs_lock);
3534 * make sure all the block groups on our dirty list actually
3537 btrfs_create_pending_block_groups(trans, root);
3540 path = btrfs_alloc_path();
3546 * cache_write_mutex is here only to save us from balance or automatic
3547 * removal of empty block groups deleting this block group while we are
3548 * writing out the cache
3550 mutex_lock(&trans->transaction->cache_write_mutex);
3551 while (!list_empty(&dirty)) {
3552 cache = list_first_entry(&dirty,
3553 struct btrfs_block_group_cache,
3556 * this can happen if something re-dirties a block
3557 * group that is already under IO. Just wait for it to
3558 * finish and then do it all again
3560 if (!list_empty(&cache->io_list)) {
3561 list_del_init(&cache->io_list);
3562 btrfs_wait_cache_io(root, trans, cache,
3563 &cache->io_ctl, path,
3564 cache->key.objectid);
3565 btrfs_put_block_group(cache);
3570 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3571 * if it should update the cache_state. Don't delete
3572 * until after we wait.
3574 * Since we're not running in the commit critical section
3575 * we need the dirty_bgs_lock to protect from update_block_group
3577 spin_lock(&cur_trans->dirty_bgs_lock);
3578 list_del_init(&cache->dirty_list);
3579 spin_unlock(&cur_trans->dirty_bgs_lock);
3583 cache_save_setup(cache, trans, path);
3585 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3586 cache->io_ctl.inode = NULL;
3587 ret = btrfs_write_out_cache(root, trans, cache, path);
3588 if (ret == 0 && cache->io_ctl.inode) {
3593 * the cache_write_mutex is protecting
3596 list_add_tail(&cache->io_list, io);
3599 * if we failed to write the cache, the
3600 * generation will be bad and life goes on
3606 ret = write_one_cache_group(trans, root, path, cache);
3608 * Our block group might still be attached to the list
3609 * of new block groups in the transaction handle of some
3610 * other task (struct btrfs_trans_handle->new_bgs). This
3611 * means its block group item isn't yet in the extent
3612 * tree. If this happens ignore the error, as we will
3613 * try again later in the critical section of the
3614 * transaction commit.
3616 if (ret == -ENOENT) {
3618 spin_lock(&cur_trans->dirty_bgs_lock);
3619 if (list_empty(&cache->dirty_list)) {
3620 list_add_tail(&cache->dirty_list,
3621 &cur_trans->dirty_bgs);
3622 btrfs_get_block_group(cache);
3624 spin_unlock(&cur_trans->dirty_bgs_lock);
3626 btrfs_abort_transaction(trans, root, ret);
3630 /* if its not on the io list, we need to put the block group */
3632 btrfs_put_block_group(cache);
3638 * Avoid blocking other tasks for too long. It might even save
3639 * us from writing caches for block groups that are going to be
3642 mutex_unlock(&trans->transaction->cache_write_mutex);
3643 mutex_lock(&trans->transaction->cache_write_mutex);
3645 mutex_unlock(&trans->transaction->cache_write_mutex);
3648 * go through delayed refs for all the stuff we've just kicked off
3649 * and then loop back (just once)
3651 ret = btrfs_run_delayed_refs(trans, root, 0);
3652 if (!ret && loops == 0) {
3654 spin_lock(&cur_trans->dirty_bgs_lock);
3655 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3657 * dirty_bgs_lock protects us from concurrent block group
3658 * deletes too (not just cache_write_mutex).
3660 if (!list_empty(&dirty)) {
3661 spin_unlock(&cur_trans->dirty_bgs_lock);
3664 spin_unlock(&cur_trans->dirty_bgs_lock);
3667 btrfs_free_path(path);
3671 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3672 struct btrfs_root *root)
3674 struct btrfs_block_group_cache *cache;
3675 struct btrfs_transaction *cur_trans = trans->transaction;
3678 struct btrfs_path *path;
3679 struct list_head *io = &cur_trans->io_bgs;
3680 int num_started = 0;
3682 path = btrfs_alloc_path();
3687 * We don't need the lock here since we are protected by the transaction
3688 * commit. We want to do the cache_save_setup first and then run the
3689 * delayed refs to make sure we have the best chance at doing this all
3692 while (!list_empty(&cur_trans->dirty_bgs)) {
3693 cache = list_first_entry(&cur_trans->dirty_bgs,
3694 struct btrfs_block_group_cache,
3698 * this can happen if cache_save_setup re-dirties a block
3699 * group that is already under IO. Just wait for it to
3700 * finish and then do it all again
3702 if (!list_empty(&cache->io_list)) {
3703 list_del_init(&cache->io_list);
3704 btrfs_wait_cache_io(root, trans, cache,
3705 &cache->io_ctl, path,
3706 cache->key.objectid);
3707 btrfs_put_block_group(cache);
3711 * don't remove from the dirty list until after we've waited
3714 list_del_init(&cache->dirty_list);
3717 cache_save_setup(cache, trans, path);
3720 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3722 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3723 cache->io_ctl.inode = NULL;
3724 ret = btrfs_write_out_cache(root, trans, cache, path);
3725 if (ret == 0 && cache->io_ctl.inode) {
3728 list_add_tail(&cache->io_list, io);
3731 * if we failed to write the cache, the
3732 * generation will be bad and life goes on
3738 ret = write_one_cache_group(trans, root, path, cache);
3740 btrfs_abort_transaction(trans, root, ret);
3743 /* if its not on the io list, we need to put the block group */
3745 btrfs_put_block_group(cache);
3748 while (!list_empty(io)) {
3749 cache = list_first_entry(io, struct btrfs_block_group_cache,
3751 list_del_init(&cache->io_list);
3752 btrfs_wait_cache_io(root, trans, cache,
3753 &cache->io_ctl, path, cache->key.objectid);
3754 btrfs_put_block_group(cache);
3757 btrfs_free_path(path);
3761 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3763 struct btrfs_block_group_cache *block_group;
3766 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3767 if (!block_group || block_group->ro)
3770 btrfs_put_block_group(block_group);
3774 static const char *alloc_name(u64 flags)
3777 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3779 case BTRFS_BLOCK_GROUP_METADATA:
3781 case BTRFS_BLOCK_GROUP_DATA:
3783 case BTRFS_BLOCK_GROUP_SYSTEM:
3787 return "invalid-combination";
3791 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3792 u64 total_bytes, u64 bytes_used,
3793 struct btrfs_space_info **space_info)
3795 struct btrfs_space_info *found;
3800 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3801 BTRFS_BLOCK_GROUP_RAID10))
3806 found = __find_space_info(info, flags);
3808 spin_lock(&found->lock);
3809 found->total_bytes += total_bytes;
3810 found->disk_total += total_bytes * factor;
3811 found->bytes_used += bytes_used;
3812 found->disk_used += bytes_used * factor;
3813 if (total_bytes > 0)
3815 spin_unlock(&found->lock);
3816 *space_info = found;
3819 found = kzalloc(sizeof(*found), GFP_NOFS);
3823 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3829 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3830 INIT_LIST_HEAD(&found->block_groups[i]);
3831 init_rwsem(&found->groups_sem);
3832 spin_lock_init(&found->lock);
3833 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3834 found->total_bytes = total_bytes;
3835 found->disk_total = total_bytes * factor;
3836 found->bytes_used = bytes_used;
3837 found->disk_used = bytes_used * factor;
3838 found->bytes_pinned = 0;
3839 found->bytes_reserved = 0;
3840 found->bytes_readonly = 0;
3841 found->bytes_may_use = 0;
3843 found->max_extent_size = 0;
3844 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3845 found->chunk_alloc = 0;
3847 init_waitqueue_head(&found->wait);
3848 INIT_LIST_HEAD(&found->ro_bgs);
3850 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3851 info->space_info_kobj, "%s",
3852 alloc_name(found->flags));
3858 *space_info = found;
3859 list_add_rcu(&found->list, &info->space_info);
3860 if (flags & BTRFS_BLOCK_GROUP_DATA)
3861 info->data_sinfo = found;
3866 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3868 u64 extra_flags = chunk_to_extended(flags) &
3869 BTRFS_EXTENDED_PROFILE_MASK;
3871 write_seqlock(&fs_info->profiles_lock);
3872 if (flags & BTRFS_BLOCK_GROUP_DATA)
3873 fs_info->avail_data_alloc_bits |= extra_flags;
3874 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3875 fs_info->avail_metadata_alloc_bits |= extra_flags;
3876 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3877 fs_info->avail_system_alloc_bits |= extra_flags;
3878 write_sequnlock(&fs_info->profiles_lock);
3882 * returns target flags in extended format or 0 if restripe for this
3883 * chunk_type is not in progress
3885 * should be called with either volume_mutex or balance_lock held
3887 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3889 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3895 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3896 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3897 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3898 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3899 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3900 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3901 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3902 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3903 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3910 * @flags: available profiles in extended format (see ctree.h)
3912 * Returns reduced profile in chunk format. If profile changing is in
3913 * progress (either running or paused) picks the target profile (if it's
3914 * already available), otherwise falls back to plain reducing.
3916 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3918 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3924 * see if restripe for this chunk_type is in progress, if so
3925 * try to reduce to the target profile
3927 spin_lock(&root->fs_info->balance_lock);
3928 target = get_restripe_target(root->fs_info, flags);
3930 /* pick target profile only if it's already available */
3931 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3932 spin_unlock(&root->fs_info->balance_lock);
3933 return extended_to_chunk(target);
3936 spin_unlock(&root->fs_info->balance_lock);
3938 /* First, mask out the RAID levels which aren't possible */
3939 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3940 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
3941 allowed |= btrfs_raid_group[raid_type];
3945 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
3946 allowed = BTRFS_BLOCK_GROUP_RAID6;
3947 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
3948 allowed = BTRFS_BLOCK_GROUP_RAID5;
3949 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
3950 allowed = BTRFS_BLOCK_GROUP_RAID10;
3951 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
3952 allowed = BTRFS_BLOCK_GROUP_RAID1;
3953 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
3954 allowed = BTRFS_BLOCK_GROUP_RAID0;
3956 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
3958 return extended_to_chunk(flags | allowed);
3961 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3968 seq = read_seqbegin(&root->fs_info->profiles_lock);
3970 if (flags & BTRFS_BLOCK_GROUP_DATA)
3971 flags |= root->fs_info->avail_data_alloc_bits;
3972 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3973 flags |= root->fs_info->avail_system_alloc_bits;
3974 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3975 flags |= root->fs_info->avail_metadata_alloc_bits;
3976 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3978 return btrfs_reduce_alloc_profile(root, flags);
3981 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3987 flags = BTRFS_BLOCK_GROUP_DATA;
3988 else if (root == root->fs_info->chunk_root)
3989 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3991 flags = BTRFS_BLOCK_GROUP_METADATA;
3993 ret = get_alloc_profile(root, flags);
3997 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
3999 struct btrfs_space_info *data_sinfo;
4000 struct btrfs_root *root = BTRFS_I(inode)->root;
4001 struct btrfs_fs_info *fs_info = root->fs_info;
4004 int need_commit = 2;
4005 int have_pinned_space;
4007 /* make sure bytes are sectorsize aligned */
4008 bytes = ALIGN(bytes, root->sectorsize);
4010 if (btrfs_is_free_space_inode(inode)) {
4012 ASSERT(current->journal_info);
4015 data_sinfo = fs_info->data_sinfo;
4020 /* make sure we have enough space to handle the data first */
4021 spin_lock(&data_sinfo->lock);
4022 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4023 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4024 data_sinfo->bytes_may_use;
4026 if (used + bytes > data_sinfo->total_bytes) {
4027 struct btrfs_trans_handle *trans;
4030 * if we don't have enough free bytes in this space then we need
4031 * to alloc a new chunk.
4033 if (!data_sinfo->full) {
4036 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4037 spin_unlock(&data_sinfo->lock);
4039 alloc_target = btrfs_get_alloc_profile(root, 1);
4041 * It is ugly that we don't call nolock join
4042 * transaction for the free space inode case here.
4043 * But it is safe because we only do the data space
4044 * reservation for the free space cache in the
4045 * transaction context, the common join transaction
4046 * just increase the counter of the current transaction
4047 * handler, doesn't try to acquire the trans_lock of
4050 trans = btrfs_join_transaction(root);
4052 return PTR_ERR(trans);
4054 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4056 CHUNK_ALLOC_NO_FORCE);
4057 btrfs_end_transaction(trans, root);
4062 have_pinned_space = 1;
4068 data_sinfo = fs_info->data_sinfo;
4074 * If we don't have enough pinned space to deal with this
4075 * allocation, and no removed chunk in current transaction,
4076 * don't bother committing the transaction.
4078 have_pinned_space = percpu_counter_compare(
4079 &data_sinfo->total_bytes_pinned,
4080 used + bytes - data_sinfo->total_bytes);
4081 spin_unlock(&data_sinfo->lock);
4083 /* commit the current transaction and try again */
4086 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4089 if (need_commit > 0)
4090 btrfs_wait_ordered_roots(fs_info, -1);
4092 trans = btrfs_join_transaction(root);
4094 return PTR_ERR(trans);
4095 if (have_pinned_space >= 0 ||
4096 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4097 &trans->transaction->flags) ||
4099 ret = btrfs_commit_transaction(trans, root);
4103 * make sure that all running delayed iput are
4106 down_write(&root->fs_info->delayed_iput_sem);
4107 up_write(&root->fs_info->delayed_iput_sem);
4110 btrfs_end_transaction(trans, root);
4114 trace_btrfs_space_reservation(root->fs_info,
4115 "space_info:enospc",
4116 data_sinfo->flags, bytes, 1);
4119 data_sinfo->bytes_may_use += bytes;
4120 trace_btrfs_space_reservation(root->fs_info, "space_info",
4121 data_sinfo->flags, bytes, 1);
4122 spin_unlock(&data_sinfo->lock);
4128 * New check_data_free_space() with ability for precious data reservation
4129 * Will replace old btrfs_check_data_free_space(), but for patch split,
4130 * add a new function first and then replace it.
4132 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4134 struct btrfs_root *root = BTRFS_I(inode)->root;
4137 /* align the range */
4138 len = round_up(start + len, root->sectorsize) -
4139 round_down(start, root->sectorsize);
4140 start = round_down(start, root->sectorsize);
4142 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4147 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4149 * TODO: Find a good method to avoid reserve data space for NOCOW
4150 * range, but don't impact performance on quota disable case.
4152 ret = btrfs_qgroup_reserve_data(inode, start, len);
4157 * Called if we need to clear a data reservation for this inode
4158 * Normally in a error case.
4160 * This one will *NOT* use accurate qgroup reserved space API, just for case
4161 * which we can't sleep and is sure it won't affect qgroup reserved space.
4162 * Like clear_bit_hook().
4164 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4167 struct btrfs_root *root = BTRFS_I(inode)->root;
4168 struct btrfs_space_info *data_sinfo;
4170 /* Make sure the range is aligned to sectorsize */
4171 len = round_up(start + len, root->sectorsize) -
4172 round_down(start, root->sectorsize);
4173 start = round_down(start, root->sectorsize);
4175 data_sinfo = root->fs_info->data_sinfo;
4176 spin_lock(&data_sinfo->lock);
4177 if (WARN_ON(data_sinfo->bytes_may_use < len))
4178 data_sinfo->bytes_may_use = 0;
4180 data_sinfo->bytes_may_use -= len;
4181 trace_btrfs_space_reservation(root->fs_info, "space_info",
4182 data_sinfo->flags, len, 0);
4183 spin_unlock(&data_sinfo->lock);
4187 * Called if we need to clear a data reservation for this inode
4188 * Normally in a error case.
4190 * This one will handle the per-indoe data rsv map for accurate reserved
4193 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4195 btrfs_free_reserved_data_space_noquota(inode, start, len);
4196 btrfs_qgroup_free_data(inode, start, len);
4199 static void force_metadata_allocation(struct btrfs_fs_info *info)
4201 struct list_head *head = &info->space_info;
4202 struct btrfs_space_info *found;
4205 list_for_each_entry_rcu(found, head, list) {
4206 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4207 found->force_alloc = CHUNK_ALLOC_FORCE;
4212 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4214 return (global->size << 1);
4217 static int should_alloc_chunk(struct btrfs_root *root,
4218 struct btrfs_space_info *sinfo, int force)
4220 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4221 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4222 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4225 if (force == CHUNK_ALLOC_FORCE)
4229 * We need to take into account the global rsv because for all intents
4230 * and purposes it's used space. Don't worry about locking the
4231 * global_rsv, it doesn't change except when the transaction commits.
4233 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4234 num_allocated += calc_global_rsv_need_space(global_rsv);
4237 * in limited mode, we want to have some free space up to
4238 * about 1% of the FS size.
4240 if (force == CHUNK_ALLOC_LIMITED) {
4241 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4242 thresh = max_t(u64, 64 * 1024 * 1024,
4243 div_factor_fine(thresh, 1));
4245 if (num_bytes - num_allocated < thresh)
4249 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4254 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4258 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4259 BTRFS_BLOCK_GROUP_RAID0 |
4260 BTRFS_BLOCK_GROUP_RAID5 |
4261 BTRFS_BLOCK_GROUP_RAID6))
4262 num_dev = root->fs_info->fs_devices->rw_devices;
4263 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4266 num_dev = 1; /* DUP or single */
4272 * If @is_allocation is true, reserve space in the system space info necessary
4273 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4276 void check_system_chunk(struct btrfs_trans_handle *trans,
4277 struct btrfs_root *root,
4280 struct btrfs_space_info *info;
4287 * Needed because we can end up allocating a system chunk and for an
4288 * atomic and race free space reservation in the chunk block reserve.
4290 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4292 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4293 spin_lock(&info->lock);
4294 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4295 info->bytes_reserved - info->bytes_readonly -
4296 info->bytes_may_use;
4297 spin_unlock(&info->lock);
4299 num_devs = get_profile_num_devs(root, type);
4301 /* num_devs device items to update and 1 chunk item to add or remove */
4302 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4303 btrfs_calc_trans_metadata_size(root, 1);
4305 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4306 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4307 left, thresh, type);
4308 dump_space_info(info, 0, 0);
4311 if (left < thresh) {
4314 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4316 * Ignore failure to create system chunk. We might end up not
4317 * needing it, as we might not need to COW all nodes/leafs from
4318 * the paths we visit in the chunk tree (they were already COWed
4319 * or created in the current transaction for example).
4321 ret = btrfs_alloc_chunk(trans, root, flags);
4325 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4326 &root->fs_info->chunk_block_rsv,
4327 thresh, BTRFS_RESERVE_NO_FLUSH);
4329 trans->chunk_bytes_reserved += thresh;
4333 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4334 struct btrfs_root *extent_root, u64 flags, int force)
4336 struct btrfs_space_info *space_info;
4337 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4338 int wait_for_alloc = 0;
4341 /* Don't re-enter if we're already allocating a chunk */
4342 if (trans->allocating_chunk)
4345 space_info = __find_space_info(extent_root->fs_info, flags);
4347 ret = update_space_info(extent_root->fs_info, flags,
4349 BUG_ON(ret); /* -ENOMEM */
4351 BUG_ON(!space_info); /* Logic error */
4354 spin_lock(&space_info->lock);
4355 if (force < space_info->force_alloc)
4356 force = space_info->force_alloc;
4357 if (space_info->full) {
4358 if (should_alloc_chunk(extent_root, space_info, force))
4362 spin_unlock(&space_info->lock);
4366 if (!should_alloc_chunk(extent_root, space_info, force)) {
4367 spin_unlock(&space_info->lock);
4369 } else if (space_info->chunk_alloc) {
4372 space_info->chunk_alloc = 1;
4375 spin_unlock(&space_info->lock);
4377 mutex_lock(&fs_info->chunk_mutex);
4380 * The chunk_mutex is held throughout the entirety of a chunk
4381 * allocation, so once we've acquired the chunk_mutex we know that the
4382 * other guy is done and we need to recheck and see if we should
4385 if (wait_for_alloc) {
4386 mutex_unlock(&fs_info->chunk_mutex);
4391 trans->allocating_chunk = true;
4394 * If we have mixed data/metadata chunks we want to make sure we keep
4395 * allocating mixed chunks instead of individual chunks.
4397 if (btrfs_mixed_space_info(space_info))
4398 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4401 * if we're doing a data chunk, go ahead and make sure that
4402 * we keep a reasonable number of metadata chunks allocated in the
4405 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4406 fs_info->data_chunk_allocations++;
4407 if (!(fs_info->data_chunk_allocations %
4408 fs_info->metadata_ratio))
4409 force_metadata_allocation(fs_info);
4413 * Check if we have enough space in SYSTEM chunk because we may need
4414 * to update devices.
4416 check_system_chunk(trans, extent_root, flags);
4418 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4419 trans->allocating_chunk = false;
4421 spin_lock(&space_info->lock);
4422 if (ret < 0 && ret != -ENOSPC)
4425 space_info->full = 1;
4429 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4431 space_info->chunk_alloc = 0;
4432 spin_unlock(&space_info->lock);
4433 mutex_unlock(&fs_info->chunk_mutex);
4435 * When we allocate a new chunk we reserve space in the chunk block
4436 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4437 * add new nodes/leafs to it if we end up needing to do it when
4438 * inserting the chunk item and updating device items as part of the
4439 * second phase of chunk allocation, performed by
4440 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4441 * large number of new block groups to create in our transaction
4442 * handle's new_bgs list to avoid exhausting the chunk block reserve
4443 * in extreme cases - like having a single transaction create many new
4444 * block groups when starting to write out the free space caches of all
4445 * the block groups that were made dirty during the lifetime of the
4448 if (trans->can_flush_pending_bgs &&
4449 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4450 btrfs_create_pending_block_groups(trans, trans->root);
4451 btrfs_trans_release_chunk_metadata(trans);
4456 static int can_overcommit(struct btrfs_root *root,
4457 struct btrfs_space_info *space_info, u64 bytes,
4458 enum btrfs_reserve_flush_enum flush)
4460 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4461 u64 profile = btrfs_get_alloc_profile(root, 0);
4466 used = space_info->bytes_used + space_info->bytes_reserved +
4467 space_info->bytes_pinned + space_info->bytes_readonly;
4470 * We only want to allow over committing if we have lots of actual space
4471 * free, but if we don't have enough space to handle the global reserve
4472 * space then we could end up having a real enospc problem when trying
4473 * to allocate a chunk or some other such important allocation.
4475 spin_lock(&global_rsv->lock);
4476 space_size = calc_global_rsv_need_space(global_rsv);
4477 spin_unlock(&global_rsv->lock);
4478 if (used + space_size >= space_info->total_bytes)
4481 used += space_info->bytes_may_use;
4483 spin_lock(&root->fs_info->free_chunk_lock);
4484 avail = root->fs_info->free_chunk_space;
4485 spin_unlock(&root->fs_info->free_chunk_lock);
4488 * If we have dup, raid1 or raid10 then only half of the free
4489 * space is actually useable. For raid56, the space info used
4490 * doesn't include the parity drive, so we don't have to
4493 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4494 BTRFS_BLOCK_GROUP_RAID1 |
4495 BTRFS_BLOCK_GROUP_RAID10))
4499 * If we aren't flushing all things, let us overcommit up to
4500 * 1/2th of the space. If we can flush, don't let us overcommit
4501 * too much, let it overcommit up to 1/8 of the space.
4503 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4508 if (used + bytes < space_info->total_bytes + avail)
4513 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4514 unsigned long nr_pages, int nr_items)
4516 struct super_block *sb = root->fs_info->sb;
4518 if (down_read_trylock(&sb->s_umount)) {
4519 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4520 up_read(&sb->s_umount);
4523 * We needn't worry the filesystem going from r/w to r/o though
4524 * we don't acquire ->s_umount mutex, because the filesystem
4525 * should guarantee the delalloc inodes list be empty after
4526 * the filesystem is readonly(all dirty pages are written to
4529 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4530 if (!current->journal_info)
4531 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4535 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4540 bytes = btrfs_calc_trans_metadata_size(root, 1);
4541 nr = (int)div64_u64(to_reclaim, bytes);
4547 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4550 * shrink metadata reservation for delalloc
4552 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4555 struct btrfs_block_rsv *block_rsv;
4556 struct btrfs_space_info *space_info;
4557 struct btrfs_trans_handle *trans;
4561 unsigned long nr_pages;
4564 enum btrfs_reserve_flush_enum flush;
4566 /* Calc the number of the pages we need flush for space reservation */
4567 items = calc_reclaim_items_nr(root, to_reclaim);
4568 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4570 trans = (struct btrfs_trans_handle *)current->journal_info;
4571 block_rsv = &root->fs_info->delalloc_block_rsv;
4572 space_info = block_rsv->space_info;
4574 delalloc_bytes = percpu_counter_sum_positive(
4575 &root->fs_info->delalloc_bytes);
4576 if (delalloc_bytes == 0) {
4580 btrfs_wait_ordered_roots(root->fs_info, items);
4585 while (delalloc_bytes && loops < 3) {
4586 max_reclaim = min(delalloc_bytes, to_reclaim);
4587 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4588 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4590 * We need to wait for the async pages to actually start before
4593 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4597 if (max_reclaim <= nr_pages)
4600 max_reclaim -= nr_pages;
4602 wait_event(root->fs_info->async_submit_wait,
4603 atomic_read(&root->fs_info->async_delalloc_pages) <=
4607 flush = BTRFS_RESERVE_FLUSH_ALL;
4609 flush = BTRFS_RESERVE_NO_FLUSH;
4610 spin_lock(&space_info->lock);
4611 if (can_overcommit(root, space_info, orig, flush)) {
4612 spin_unlock(&space_info->lock);
4615 spin_unlock(&space_info->lock);
4618 if (wait_ordered && !trans) {
4619 btrfs_wait_ordered_roots(root->fs_info, items);
4621 time_left = schedule_timeout_killable(1);
4625 delalloc_bytes = percpu_counter_sum_positive(
4626 &root->fs_info->delalloc_bytes);
4631 * maybe_commit_transaction - possibly commit the transaction if its ok to
4632 * @root - the root we're allocating for
4633 * @bytes - the number of bytes we want to reserve
4634 * @force - force the commit
4636 * This will check to make sure that committing the transaction will actually
4637 * get us somewhere and then commit the transaction if it does. Otherwise it
4638 * will return -ENOSPC.
4640 static int may_commit_transaction(struct btrfs_root *root,
4641 struct btrfs_space_info *space_info,
4642 u64 bytes, int force)
4644 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4645 struct btrfs_trans_handle *trans;
4647 trans = (struct btrfs_trans_handle *)current->journal_info;
4654 /* See if there is enough pinned space to make this reservation */
4655 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4660 * See if there is some space in the delayed insertion reservation for
4663 if (space_info != delayed_rsv->space_info)
4666 spin_lock(&delayed_rsv->lock);
4667 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4668 bytes - delayed_rsv->size) >= 0) {
4669 spin_unlock(&delayed_rsv->lock);
4672 spin_unlock(&delayed_rsv->lock);
4675 trans = btrfs_join_transaction(root);
4679 return btrfs_commit_transaction(trans, root);
4683 FLUSH_DELAYED_ITEMS_NR = 1,
4684 FLUSH_DELAYED_ITEMS = 2,
4686 FLUSH_DELALLOC_WAIT = 4,
4691 static int flush_space(struct btrfs_root *root,
4692 struct btrfs_space_info *space_info, u64 num_bytes,
4693 u64 orig_bytes, int state)
4695 struct btrfs_trans_handle *trans;
4700 case FLUSH_DELAYED_ITEMS_NR:
4701 case FLUSH_DELAYED_ITEMS:
4702 if (state == FLUSH_DELAYED_ITEMS_NR)
4703 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4707 trans = btrfs_join_transaction(root);
4708 if (IS_ERR(trans)) {
4709 ret = PTR_ERR(trans);
4712 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4713 btrfs_end_transaction(trans, root);
4715 case FLUSH_DELALLOC:
4716 case FLUSH_DELALLOC_WAIT:
4717 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4718 state == FLUSH_DELALLOC_WAIT);
4721 trans = btrfs_join_transaction(root);
4722 if (IS_ERR(trans)) {
4723 ret = PTR_ERR(trans);
4726 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4727 btrfs_get_alloc_profile(root, 0),
4728 CHUNK_ALLOC_NO_FORCE);
4729 btrfs_end_transaction(trans, root);
4734 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4745 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4746 struct btrfs_space_info *space_info)
4752 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4754 spin_lock(&space_info->lock);
4755 if (can_overcommit(root, space_info, to_reclaim,
4756 BTRFS_RESERVE_FLUSH_ALL)) {
4761 used = space_info->bytes_used + space_info->bytes_reserved +
4762 space_info->bytes_pinned + space_info->bytes_readonly +
4763 space_info->bytes_may_use;
4764 if (can_overcommit(root, space_info, 1024 * 1024,
4765 BTRFS_RESERVE_FLUSH_ALL))
4766 expected = div_factor_fine(space_info->total_bytes, 95);
4768 expected = div_factor_fine(space_info->total_bytes, 90);
4770 if (used > expected)
4771 to_reclaim = used - expected;
4774 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4775 space_info->bytes_reserved);
4777 spin_unlock(&space_info->lock);
4782 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4783 struct btrfs_fs_info *fs_info, u64 used)
4785 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4787 /* If we're just plain full then async reclaim just slows us down. */
4788 if (space_info->bytes_used >= thresh)
4791 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4792 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4795 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4796 struct btrfs_fs_info *fs_info,
4801 spin_lock(&space_info->lock);
4803 * We run out of space and have not got any free space via flush_space,
4804 * so don't bother doing async reclaim.
4806 if (flush_state > COMMIT_TRANS && space_info->full) {
4807 spin_unlock(&space_info->lock);
4811 used = space_info->bytes_used + space_info->bytes_reserved +
4812 space_info->bytes_pinned + space_info->bytes_readonly +
4813 space_info->bytes_may_use;
4814 if (need_do_async_reclaim(space_info, fs_info, used)) {
4815 spin_unlock(&space_info->lock);
4818 spin_unlock(&space_info->lock);
4823 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4825 struct btrfs_fs_info *fs_info;
4826 struct btrfs_space_info *space_info;
4830 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4831 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4833 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4838 flush_state = FLUSH_DELAYED_ITEMS_NR;
4840 flush_space(fs_info->fs_root, space_info, to_reclaim,
4841 to_reclaim, flush_state);
4843 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4846 } while (flush_state < COMMIT_TRANS);
4849 void btrfs_init_async_reclaim_work(struct work_struct *work)
4851 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4855 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4856 * @root - the root we're allocating for
4857 * @block_rsv - the block_rsv we're allocating for
4858 * @orig_bytes - the number of bytes we want
4859 * @flush - whether or not we can flush to make our reservation
4861 * This will reserve orgi_bytes number of bytes from the space info associated
4862 * with the block_rsv. If there is not enough space it will make an attempt to
4863 * flush out space to make room. It will do this by flushing delalloc if
4864 * possible or committing the transaction. If flush is 0 then no attempts to
4865 * regain reservations will be made and this will fail if there is not enough
4868 static int reserve_metadata_bytes(struct btrfs_root *root,
4869 struct btrfs_block_rsv *block_rsv,
4871 enum btrfs_reserve_flush_enum flush)
4873 struct btrfs_space_info *space_info = block_rsv->space_info;
4875 u64 num_bytes = orig_bytes;
4876 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4878 bool flushing = false;
4882 spin_lock(&space_info->lock);
4884 * We only want to wait if somebody other than us is flushing and we
4885 * are actually allowed to flush all things.
4887 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4888 space_info->flush) {
4889 spin_unlock(&space_info->lock);
4891 * If we have a trans handle we can't wait because the flusher
4892 * may have to commit the transaction, which would mean we would
4893 * deadlock since we are waiting for the flusher to finish, but
4894 * hold the current transaction open.
4896 if (current->journal_info)
4898 ret = wait_event_killable(space_info->wait, !space_info->flush);
4899 /* Must have been killed, return */
4903 spin_lock(&space_info->lock);
4907 used = space_info->bytes_used + space_info->bytes_reserved +
4908 space_info->bytes_pinned + space_info->bytes_readonly +
4909 space_info->bytes_may_use;
4912 * The idea here is that we've not already over-reserved the block group
4913 * then we can go ahead and save our reservation first and then start
4914 * flushing if we need to. Otherwise if we've already overcommitted
4915 * lets start flushing stuff first and then come back and try to make
4918 if (used <= space_info->total_bytes) {
4919 if (used + orig_bytes <= space_info->total_bytes) {
4920 space_info->bytes_may_use += orig_bytes;
4921 trace_btrfs_space_reservation(root->fs_info,
4922 "space_info", space_info->flags, orig_bytes, 1);
4926 * Ok set num_bytes to orig_bytes since we aren't
4927 * overocmmitted, this way we only try and reclaim what
4930 num_bytes = orig_bytes;
4934 * Ok we're over committed, set num_bytes to the overcommitted
4935 * amount plus the amount of bytes that we need for this
4938 num_bytes = used - space_info->total_bytes +
4942 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4943 space_info->bytes_may_use += orig_bytes;
4944 trace_btrfs_space_reservation(root->fs_info, "space_info",
4945 space_info->flags, orig_bytes,
4951 * Couldn't make our reservation, save our place so while we're trying
4952 * to reclaim space we can actually use it instead of somebody else
4953 * stealing it from us.
4955 * We make the other tasks wait for the flush only when we can flush
4958 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4960 space_info->flush = 1;
4961 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4964 * We will do the space reservation dance during log replay,
4965 * which means we won't have fs_info->fs_root set, so don't do
4966 * the async reclaim as we will panic.
4968 if (!root->fs_info->log_root_recovering &&
4969 need_do_async_reclaim(space_info, root->fs_info, used) &&
4970 !work_busy(&root->fs_info->async_reclaim_work))
4971 queue_work(system_unbound_wq,
4972 &root->fs_info->async_reclaim_work);
4974 spin_unlock(&space_info->lock);
4976 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4979 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4984 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4985 * would happen. So skip delalloc flush.
4987 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4988 (flush_state == FLUSH_DELALLOC ||
4989 flush_state == FLUSH_DELALLOC_WAIT))
4990 flush_state = ALLOC_CHUNK;
4994 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4995 flush_state < COMMIT_TRANS)
4997 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4998 flush_state <= COMMIT_TRANS)
5002 if (ret == -ENOSPC &&
5003 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5004 struct btrfs_block_rsv *global_rsv =
5005 &root->fs_info->global_block_rsv;
5007 if (block_rsv != global_rsv &&
5008 !block_rsv_use_bytes(global_rsv, orig_bytes))
5012 trace_btrfs_space_reservation(root->fs_info,
5013 "space_info:enospc",
5014 space_info->flags, orig_bytes, 1);
5016 spin_lock(&space_info->lock);
5017 space_info->flush = 0;
5018 wake_up_all(&space_info->wait);
5019 spin_unlock(&space_info->lock);
5024 static struct btrfs_block_rsv *get_block_rsv(
5025 const struct btrfs_trans_handle *trans,
5026 const struct btrfs_root *root)
5028 struct btrfs_block_rsv *block_rsv = NULL;
5030 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5031 (root == root->fs_info->csum_root && trans->adding_csums) ||
5032 (root == root->fs_info->uuid_root))
5033 block_rsv = trans->block_rsv;
5036 block_rsv = root->block_rsv;
5039 block_rsv = &root->fs_info->empty_block_rsv;
5044 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5048 spin_lock(&block_rsv->lock);
5049 if (block_rsv->reserved >= num_bytes) {
5050 block_rsv->reserved -= num_bytes;
5051 if (block_rsv->reserved < block_rsv->size)
5052 block_rsv->full = 0;
5055 spin_unlock(&block_rsv->lock);
5059 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5060 u64 num_bytes, int update_size)
5062 spin_lock(&block_rsv->lock);
5063 block_rsv->reserved += num_bytes;
5065 block_rsv->size += num_bytes;
5066 else if (block_rsv->reserved >= block_rsv->size)
5067 block_rsv->full = 1;
5068 spin_unlock(&block_rsv->lock);
5071 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5072 struct btrfs_block_rsv *dest, u64 num_bytes,
5075 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5078 if (global_rsv->space_info != dest->space_info)
5081 spin_lock(&global_rsv->lock);
5082 min_bytes = div_factor(global_rsv->size, min_factor);
5083 if (global_rsv->reserved < min_bytes + num_bytes) {
5084 spin_unlock(&global_rsv->lock);
5087 global_rsv->reserved -= num_bytes;
5088 if (global_rsv->reserved < global_rsv->size)
5089 global_rsv->full = 0;
5090 spin_unlock(&global_rsv->lock);
5092 block_rsv_add_bytes(dest, num_bytes, 1);
5096 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5097 struct btrfs_block_rsv *block_rsv,
5098 struct btrfs_block_rsv *dest, u64 num_bytes)
5100 struct btrfs_space_info *space_info = block_rsv->space_info;
5102 spin_lock(&block_rsv->lock);
5103 if (num_bytes == (u64)-1)
5104 num_bytes = block_rsv->size;
5105 block_rsv->size -= num_bytes;
5106 if (block_rsv->reserved >= block_rsv->size) {
5107 num_bytes = block_rsv->reserved - block_rsv->size;
5108 block_rsv->reserved = block_rsv->size;
5109 block_rsv->full = 1;
5113 spin_unlock(&block_rsv->lock);
5115 if (num_bytes > 0) {
5117 spin_lock(&dest->lock);
5121 bytes_to_add = dest->size - dest->reserved;
5122 bytes_to_add = min(num_bytes, bytes_to_add);
5123 dest->reserved += bytes_to_add;
5124 if (dest->reserved >= dest->size)
5126 num_bytes -= bytes_to_add;
5128 spin_unlock(&dest->lock);
5131 spin_lock(&space_info->lock);
5132 space_info->bytes_may_use -= num_bytes;
5133 trace_btrfs_space_reservation(fs_info, "space_info",
5134 space_info->flags, num_bytes, 0);
5135 spin_unlock(&space_info->lock);
5140 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5141 struct btrfs_block_rsv *dst, u64 num_bytes)
5145 ret = block_rsv_use_bytes(src, num_bytes);
5149 block_rsv_add_bytes(dst, num_bytes, 1);
5153 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5155 memset(rsv, 0, sizeof(*rsv));
5156 spin_lock_init(&rsv->lock);
5160 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5161 unsigned short type)
5163 struct btrfs_block_rsv *block_rsv;
5164 struct btrfs_fs_info *fs_info = root->fs_info;
5166 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5170 btrfs_init_block_rsv(block_rsv, type);
5171 block_rsv->space_info = __find_space_info(fs_info,
5172 BTRFS_BLOCK_GROUP_METADATA);
5176 void btrfs_free_block_rsv(struct btrfs_root *root,
5177 struct btrfs_block_rsv *rsv)
5181 btrfs_block_rsv_release(root, rsv, (u64)-1);
5185 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5190 int btrfs_block_rsv_add(struct btrfs_root *root,
5191 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5192 enum btrfs_reserve_flush_enum flush)
5199 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5201 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5208 int btrfs_block_rsv_check(struct btrfs_root *root,
5209 struct btrfs_block_rsv *block_rsv, int min_factor)
5217 spin_lock(&block_rsv->lock);
5218 num_bytes = div_factor(block_rsv->size, min_factor);
5219 if (block_rsv->reserved >= num_bytes)
5221 spin_unlock(&block_rsv->lock);
5226 int btrfs_block_rsv_refill(struct btrfs_root *root,
5227 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5228 enum btrfs_reserve_flush_enum flush)
5236 spin_lock(&block_rsv->lock);
5237 num_bytes = min_reserved;
5238 if (block_rsv->reserved >= num_bytes)
5241 num_bytes -= block_rsv->reserved;
5242 spin_unlock(&block_rsv->lock);
5247 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5249 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5256 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5257 struct btrfs_block_rsv *dst_rsv,
5260 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5263 void btrfs_block_rsv_release(struct btrfs_root *root,
5264 struct btrfs_block_rsv *block_rsv,
5267 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5268 if (global_rsv == block_rsv ||
5269 block_rsv->space_info != global_rsv->space_info)
5271 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5276 * helper to calculate size of global block reservation.
5277 * the desired value is sum of space used by extent tree,
5278 * checksum tree and root tree
5280 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5282 struct btrfs_space_info *sinfo;
5286 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5288 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5289 spin_lock(&sinfo->lock);
5290 data_used = sinfo->bytes_used;
5291 spin_unlock(&sinfo->lock);
5293 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5294 spin_lock(&sinfo->lock);
5295 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5297 meta_used = sinfo->bytes_used;
5298 spin_unlock(&sinfo->lock);
5300 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5302 num_bytes += div_u64(data_used + meta_used, 50);
5304 if (num_bytes * 3 > meta_used)
5305 num_bytes = div_u64(meta_used, 3);
5307 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5310 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5312 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5313 struct btrfs_space_info *sinfo = block_rsv->space_info;
5316 num_bytes = calc_global_metadata_size(fs_info);
5318 spin_lock(&sinfo->lock);
5319 spin_lock(&block_rsv->lock);
5321 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5323 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5324 sinfo->bytes_reserved + sinfo->bytes_readonly +
5325 sinfo->bytes_may_use;
5327 if (sinfo->total_bytes > num_bytes) {
5328 num_bytes = sinfo->total_bytes - num_bytes;
5329 block_rsv->reserved += num_bytes;
5330 sinfo->bytes_may_use += num_bytes;
5331 trace_btrfs_space_reservation(fs_info, "space_info",
5332 sinfo->flags, num_bytes, 1);
5335 if (block_rsv->reserved >= block_rsv->size) {
5336 num_bytes = block_rsv->reserved - block_rsv->size;
5337 sinfo->bytes_may_use -= num_bytes;
5338 trace_btrfs_space_reservation(fs_info, "space_info",
5339 sinfo->flags, num_bytes, 0);
5340 block_rsv->reserved = block_rsv->size;
5341 block_rsv->full = 1;
5344 spin_unlock(&block_rsv->lock);
5345 spin_unlock(&sinfo->lock);
5348 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5350 struct btrfs_space_info *space_info;
5352 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5353 fs_info->chunk_block_rsv.space_info = space_info;
5355 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5356 fs_info->global_block_rsv.space_info = space_info;
5357 fs_info->delalloc_block_rsv.space_info = space_info;
5358 fs_info->trans_block_rsv.space_info = space_info;
5359 fs_info->empty_block_rsv.space_info = space_info;
5360 fs_info->delayed_block_rsv.space_info = space_info;
5362 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5363 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5364 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5365 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5366 if (fs_info->quota_root)
5367 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5368 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5370 update_global_block_rsv(fs_info);
5373 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5375 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5377 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5378 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5379 WARN_ON(fs_info->trans_block_rsv.size > 0);
5380 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5381 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5382 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5383 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5384 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5387 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5388 struct btrfs_root *root)
5390 if (!trans->block_rsv)
5393 if (!trans->bytes_reserved)
5396 trace_btrfs_space_reservation(root->fs_info, "transaction",
5397 trans->transid, trans->bytes_reserved, 0);
5398 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5399 trans->bytes_reserved = 0;
5403 * To be called after all the new block groups attached to the transaction
5404 * handle have been created (btrfs_create_pending_block_groups()).
5406 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5408 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5410 if (!trans->chunk_bytes_reserved)
5413 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5415 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5416 trans->chunk_bytes_reserved);
5417 trans->chunk_bytes_reserved = 0;
5420 /* Can only return 0 or -ENOSPC */
5421 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5422 struct inode *inode)
5424 struct btrfs_root *root = BTRFS_I(inode)->root;
5425 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5426 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5429 * We need to hold space in order to delete our orphan item once we've
5430 * added it, so this takes the reservation so we can release it later
5431 * when we are truly done with the orphan item.
5433 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5434 trace_btrfs_space_reservation(root->fs_info, "orphan",
5435 btrfs_ino(inode), num_bytes, 1);
5436 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5439 void btrfs_orphan_release_metadata(struct inode *inode)
5441 struct btrfs_root *root = BTRFS_I(inode)->root;
5442 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5443 trace_btrfs_space_reservation(root->fs_info, "orphan",
5444 btrfs_ino(inode), num_bytes, 0);
5445 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5449 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5450 * root: the root of the parent directory
5451 * rsv: block reservation
5452 * items: the number of items that we need do reservation
5453 * qgroup_reserved: used to return the reserved size in qgroup
5455 * This function is used to reserve the space for snapshot/subvolume
5456 * creation and deletion. Those operations are different with the
5457 * common file/directory operations, they change two fs/file trees
5458 * and root tree, the number of items that the qgroup reserves is
5459 * different with the free space reservation. So we can not use
5460 * the space reseravtion mechanism in start_transaction().
5462 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5463 struct btrfs_block_rsv *rsv,
5465 u64 *qgroup_reserved,
5466 bool use_global_rsv)
5470 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5472 if (root->fs_info->quota_enabled) {
5473 /* One for parent inode, two for dir entries */
5474 num_bytes = 3 * root->nodesize;
5475 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5482 *qgroup_reserved = num_bytes;
5484 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5485 rsv->space_info = __find_space_info(root->fs_info,
5486 BTRFS_BLOCK_GROUP_METADATA);
5487 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5488 BTRFS_RESERVE_FLUSH_ALL);
5490 if (ret == -ENOSPC && use_global_rsv)
5491 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5493 if (ret && *qgroup_reserved)
5494 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5499 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5500 struct btrfs_block_rsv *rsv,
5501 u64 qgroup_reserved)
5503 btrfs_block_rsv_release(root, rsv, (u64)-1);
5507 * drop_outstanding_extent - drop an outstanding extent
5508 * @inode: the inode we're dropping the extent for
5509 * @num_bytes: the number of bytes we're relaseing.
5511 * This is called when we are freeing up an outstanding extent, either called
5512 * after an error or after an extent is written. This will return the number of
5513 * reserved extents that need to be freed. This must be called with
5514 * BTRFS_I(inode)->lock held.
5516 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5518 unsigned drop_inode_space = 0;
5519 unsigned dropped_extents = 0;
5520 unsigned num_extents = 0;
5522 num_extents = (unsigned)div64_u64(num_bytes +
5523 BTRFS_MAX_EXTENT_SIZE - 1,
5524 BTRFS_MAX_EXTENT_SIZE);
5525 ASSERT(num_extents);
5526 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5527 BTRFS_I(inode)->outstanding_extents -= num_extents;
5529 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5530 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5531 &BTRFS_I(inode)->runtime_flags))
5532 drop_inode_space = 1;
5535 * If we have more or the same amount of outsanding extents than we have
5536 * reserved then we need to leave the reserved extents count alone.
5538 if (BTRFS_I(inode)->outstanding_extents >=
5539 BTRFS_I(inode)->reserved_extents)
5540 return drop_inode_space;
5542 dropped_extents = BTRFS_I(inode)->reserved_extents -
5543 BTRFS_I(inode)->outstanding_extents;
5544 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5545 return dropped_extents + drop_inode_space;
5549 * calc_csum_metadata_size - return the amount of metada space that must be
5550 * reserved/free'd for the given bytes.
5551 * @inode: the inode we're manipulating
5552 * @num_bytes: the number of bytes in question
5553 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5555 * This adjusts the number of csum_bytes in the inode and then returns the
5556 * correct amount of metadata that must either be reserved or freed. We
5557 * calculate how many checksums we can fit into one leaf and then divide the
5558 * number of bytes that will need to be checksumed by this value to figure out
5559 * how many checksums will be required. If we are adding bytes then the number
5560 * may go up and we will return the number of additional bytes that must be
5561 * reserved. If it is going down we will return the number of bytes that must
5564 * This must be called with BTRFS_I(inode)->lock held.
5566 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5569 struct btrfs_root *root = BTRFS_I(inode)->root;
5570 u64 old_csums, num_csums;
5572 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5573 BTRFS_I(inode)->csum_bytes == 0)
5576 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5578 BTRFS_I(inode)->csum_bytes += num_bytes;
5580 BTRFS_I(inode)->csum_bytes -= num_bytes;
5581 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5583 /* No change, no need to reserve more */
5584 if (old_csums == num_csums)
5588 return btrfs_calc_trans_metadata_size(root,
5589 num_csums - old_csums);
5591 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5594 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5596 struct btrfs_root *root = BTRFS_I(inode)->root;
5597 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5600 unsigned nr_extents = 0;
5601 int extra_reserve = 0;
5602 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5604 bool delalloc_lock = true;
5608 /* If we are a free space inode we need to not flush since we will be in
5609 * the middle of a transaction commit. We also don't need the delalloc
5610 * mutex since we won't race with anybody. We need this mostly to make
5611 * lockdep shut its filthy mouth.
5613 if (btrfs_is_free_space_inode(inode)) {
5614 flush = BTRFS_RESERVE_NO_FLUSH;
5615 delalloc_lock = false;
5618 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5619 btrfs_transaction_in_commit(root->fs_info))
5620 schedule_timeout(1);
5623 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5625 num_bytes = ALIGN(num_bytes, root->sectorsize);
5627 spin_lock(&BTRFS_I(inode)->lock);
5628 nr_extents = (unsigned)div64_u64(num_bytes +
5629 BTRFS_MAX_EXTENT_SIZE - 1,
5630 BTRFS_MAX_EXTENT_SIZE);
5631 BTRFS_I(inode)->outstanding_extents += nr_extents;
5634 if (BTRFS_I(inode)->outstanding_extents >
5635 BTRFS_I(inode)->reserved_extents)
5636 nr_extents = BTRFS_I(inode)->outstanding_extents -
5637 BTRFS_I(inode)->reserved_extents;
5640 * Add an item to reserve for updating the inode when we complete the
5643 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5644 &BTRFS_I(inode)->runtime_flags)) {
5649 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5650 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5651 csum_bytes = BTRFS_I(inode)->csum_bytes;
5652 spin_unlock(&BTRFS_I(inode)->lock);
5654 if (root->fs_info->quota_enabled) {
5655 ret = btrfs_qgroup_reserve_meta(root,
5656 nr_extents * root->nodesize);
5661 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5662 if (unlikely(ret)) {
5663 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5667 spin_lock(&BTRFS_I(inode)->lock);
5668 if (extra_reserve) {
5669 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5670 &BTRFS_I(inode)->runtime_flags);
5673 BTRFS_I(inode)->reserved_extents += nr_extents;
5674 spin_unlock(&BTRFS_I(inode)->lock);
5677 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5680 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5681 btrfs_ino(inode), to_reserve, 1);
5682 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5687 spin_lock(&BTRFS_I(inode)->lock);
5688 dropped = drop_outstanding_extent(inode, num_bytes);
5690 * If the inodes csum_bytes is the same as the original
5691 * csum_bytes then we know we haven't raced with any free()ers
5692 * so we can just reduce our inodes csum bytes and carry on.
5694 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5695 calc_csum_metadata_size(inode, num_bytes, 0);
5697 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5701 * This is tricky, but first we need to figure out how much we
5702 * free'd from any free-ers that occured during this
5703 * reservation, so we reset ->csum_bytes to the csum_bytes
5704 * before we dropped our lock, and then call the free for the
5705 * number of bytes that were freed while we were trying our
5708 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5709 BTRFS_I(inode)->csum_bytes = csum_bytes;
5710 to_free = calc_csum_metadata_size(inode, bytes, 0);
5714 * Now we need to see how much we would have freed had we not
5715 * been making this reservation and our ->csum_bytes were not
5716 * artificially inflated.
5718 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5719 bytes = csum_bytes - orig_csum_bytes;
5720 bytes = calc_csum_metadata_size(inode, bytes, 0);
5723 * Now reset ->csum_bytes to what it should be. If bytes is
5724 * more than to_free then we would have free'd more space had we
5725 * not had an artificially high ->csum_bytes, so we need to free
5726 * the remainder. If bytes is the same or less then we don't
5727 * need to do anything, the other free-ers did the correct
5730 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5731 if (bytes > to_free)
5732 to_free = bytes - to_free;
5736 spin_unlock(&BTRFS_I(inode)->lock);
5738 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5741 btrfs_block_rsv_release(root, block_rsv, to_free);
5742 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5743 btrfs_ino(inode), to_free, 0);
5746 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5751 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5752 * @inode: the inode to release the reservation for
5753 * @num_bytes: the number of bytes we're releasing
5755 * This will release the metadata reservation for an inode. This can be called
5756 * once we complete IO for a given set of bytes to release their metadata
5759 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5761 struct btrfs_root *root = BTRFS_I(inode)->root;
5765 num_bytes = ALIGN(num_bytes, root->sectorsize);
5766 spin_lock(&BTRFS_I(inode)->lock);
5767 dropped = drop_outstanding_extent(inode, num_bytes);
5770 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5771 spin_unlock(&BTRFS_I(inode)->lock);
5773 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5775 if (btrfs_test_is_dummy_root(root))
5778 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5779 btrfs_ino(inode), to_free, 0);
5781 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5786 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5788 * @inode: inode we're writing to
5789 * @start: start range we are writing to
5790 * @len: how long the range we are writing to
5792 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5794 * This will do the following things
5796 * o reserve space in data space info for num bytes
5797 * and reserve precious corresponding qgroup space
5798 * (Done in check_data_free_space)
5800 * o reserve space for metadata space, based on the number of outstanding
5801 * extents and how much csums will be needed
5802 * also reserve metadata space in a per root over-reserve method.
5803 * o add to the inodes->delalloc_bytes
5804 * o add it to the fs_info's delalloc inodes list.
5805 * (Above 3 all done in delalloc_reserve_metadata)
5807 * Return 0 for success
5808 * Return <0 for error(-ENOSPC or -EQUOT)
5810 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5814 ret = btrfs_check_data_free_space(inode, start, len);
5817 ret = btrfs_delalloc_reserve_metadata(inode, len);
5819 btrfs_free_reserved_data_space(inode, start, len);
5824 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5825 * @inode: inode we're releasing space for
5826 * @start: start position of the space already reserved
5827 * @len: the len of the space already reserved
5829 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5830 * called in the case that we don't need the metadata AND data reservations
5831 * anymore. So if there is an error or we insert an inline extent.
5833 * This function will release the metadata space that was not used and will
5834 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5835 * list if there are no delalloc bytes left.
5836 * Also it will handle the qgroup reserved space.
5838 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5840 btrfs_delalloc_release_metadata(inode, len);
5841 btrfs_free_reserved_data_space(inode, start, len);
5844 static int update_block_group(struct btrfs_trans_handle *trans,
5845 struct btrfs_root *root, u64 bytenr,
5846 u64 num_bytes, int alloc)
5848 struct btrfs_block_group_cache *cache = NULL;
5849 struct btrfs_fs_info *info = root->fs_info;
5850 u64 total = num_bytes;
5855 /* block accounting for super block */
5856 spin_lock(&info->delalloc_root_lock);
5857 old_val = btrfs_super_bytes_used(info->super_copy);
5859 old_val += num_bytes;
5861 old_val -= num_bytes;
5862 btrfs_set_super_bytes_used(info->super_copy, old_val);
5863 spin_unlock(&info->delalloc_root_lock);
5866 cache = btrfs_lookup_block_group(info, bytenr);
5869 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5870 BTRFS_BLOCK_GROUP_RAID1 |
5871 BTRFS_BLOCK_GROUP_RAID10))
5876 * If this block group has free space cache written out, we
5877 * need to make sure to load it if we are removing space. This
5878 * is because we need the unpinning stage to actually add the
5879 * space back to the block group, otherwise we will leak space.
5881 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5882 cache_block_group(cache, 1);
5884 byte_in_group = bytenr - cache->key.objectid;
5885 WARN_ON(byte_in_group > cache->key.offset);
5887 spin_lock(&cache->space_info->lock);
5888 spin_lock(&cache->lock);
5890 if (btrfs_test_opt(root, SPACE_CACHE) &&
5891 cache->disk_cache_state < BTRFS_DC_CLEAR)
5892 cache->disk_cache_state = BTRFS_DC_CLEAR;
5894 old_val = btrfs_block_group_used(&cache->item);
5895 num_bytes = min(total, cache->key.offset - byte_in_group);
5897 old_val += num_bytes;
5898 btrfs_set_block_group_used(&cache->item, old_val);
5899 cache->reserved -= num_bytes;
5900 cache->space_info->bytes_reserved -= num_bytes;
5901 cache->space_info->bytes_used += num_bytes;
5902 cache->space_info->disk_used += num_bytes * factor;
5903 spin_unlock(&cache->lock);
5904 spin_unlock(&cache->space_info->lock);
5906 old_val -= num_bytes;
5907 btrfs_set_block_group_used(&cache->item, old_val);
5908 cache->pinned += num_bytes;
5909 cache->space_info->bytes_pinned += num_bytes;
5910 cache->space_info->bytes_used -= num_bytes;
5911 cache->space_info->disk_used -= num_bytes * factor;
5912 spin_unlock(&cache->lock);
5913 spin_unlock(&cache->space_info->lock);
5915 set_extent_dirty(info->pinned_extents,
5916 bytenr, bytenr + num_bytes - 1,
5917 GFP_NOFS | __GFP_NOFAIL);
5919 * No longer have used bytes in this block group, queue
5923 spin_lock(&info->unused_bgs_lock);
5924 if (list_empty(&cache->bg_list)) {
5925 btrfs_get_block_group(cache);
5926 list_add_tail(&cache->bg_list,
5929 spin_unlock(&info->unused_bgs_lock);
5933 spin_lock(&trans->transaction->dirty_bgs_lock);
5934 if (list_empty(&cache->dirty_list)) {
5935 list_add_tail(&cache->dirty_list,
5936 &trans->transaction->dirty_bgs);
5937 trans->transaction->num_dirty_bgs++;
5938 btrfs_get_block_group(cache);
5940 spin_unlock(&trans->transaction->dirty_bgs_lock);
5942 btrfs_put_block_group(cache);
5944 bytenr += num_bytes;
5949 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5951 struct btrfs_block_group_cache *cache;
5954 spin_lock(&root->fs_info->block_group_cache_lock);
5955 bytenr = root->fs_info->first_logical_byte;
5956 spin_unlock(&root->fs_info->block_group_cache_lock);
5958 if (bytenr < (u64)-1)
5961 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5965 bytenr = cache->key.objectid;
5966 btrfs_put_block_group(cache);
5971 static int pin_down_extent(struct btrfs_root *root,
5972 struct btrfs_block_group_cache *cache,
5973 u64 bytenr, u64 num_bytes, int reserved)
5975 spin_lock(&cache->space_info->lock);
5976 spin_lock(&cache->lock);
5977 cache->pinned += num_bytes;
5978 cache->space_info->bytes_pinned += num_bytes;
5980 cache->reserved -= num_bytes;
5981 cache->space_info->bytes_reserved -= num_bytes;
5983 spin_unlock(&cache->lock);
5984 spin_unlock(&cache->space_info->lock);
5986 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5987 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5989 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5994 * this function must be called within transaction
5996 int btrfs_pin_extent(struct btrfs_root *root,
5997 u64 bytenr, u64 num_bytes, int reserved)
5999 struct btrfs_block_group_cache *cache;
6001 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6002 BUG_ON(!cache); /* Logic error */
6004 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6006 btrfs_put_block_group(cache);
6011 * this function must be called within transaction
6013 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6014 u64 bytenr, u64 num_bytes)
6016 struct btrfs_block_group_cache *cache;
6019 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6024 * pull in the free space cache (if any) so that our pin
6025 * removes the free space from the cache. We have load_only set
6026 * to one because the slow code to read in the free extents does check
6027 * the pinned extents.
6029 cache_block_group(cache, 1);
6031 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6033 /* remove us from the free space cache (if we're there at all) */
6034 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6035 btrfs_put_block_group(cache);
6039 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6042 struct btrfs_block_group_cache *block_group;
6043 struct btrfs_caching_control *caching_ctl;
6045 block_group = btrfs_lookup_block_group(root->fs_info, start);
6049 cache_block_group(block_group, 0);
6050 caching_ctl = get_caching_control(block_group);
6054 BUG_ON(!block_group_cache_done(block_group));
6055 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6057 mutex_lock(&caching_ctl->mutex);
6059 if (start >= caching_ctl->progress) {
6060 ret = add_excluded_extent(root, start, num_bytes);
6061 } else if (start + num_bytes <= caching_ctl->progress) {
6062 ret = btrfs_remove_free_space(block_group,
6065 num_bytes = caching_ctl->progress - start;
6066 ret = btrfs_remove_free_space(block_group,
6071 num_bytes = (start + num_bytes) -
6072 caching_ctl->progress;
6073 start = caching_ctl->progress;
6074 ret = add_excluded_extent(root, start, num_bytes);
6077 mutex_unlock(&caching_ctl->mutex);
6078 put_caching_control(caching_ctl);
6080 btrfs_put_block_group(block_group);
6084 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6085 struct extent_buffer *eb)
6087 struct btrfs_file_extent_item *item;
6088 struct btrfs_key key;
6092 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6095 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6096 btrfs_item_key_to_cpu(eb, &key, i);
6097 if (key.type != BTRFS_EXTENT_DATA_KEY)
6099 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6100 found_type = btrfs_file_extent_type(eb, item);
6101 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6103 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6105 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6106 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6107 __exclude_logged_extent(log, key.objectid, key.offset);
6114 * btrfs_update_reserved_bytes - update the block_group and space info counters
6115 * @cache: The cache we are manipulating
6116 * @num_bytes: The number of bytes in question
6117 * @reserve: One of the reservation enums
6118 * @delalloc: The blocks are allocated for the delalloc write
6120 * This is called by the allocator when it reserves space, or by somebody who is
6121 * freeing space that was never actually used on disk. For example if you
6122 * reserve some space for a new leaf in transaction A and before transaction A
6123 * commits you free that leaf, you call this with reserve set to 0 in order to
6124 * clear the reservation.
6126 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6127 * ENOSPC accounting. For data we handle the reservation through clearing the
6128 * delalloc bits in the io_tree. We have to do this since we could end up
6129 * allocating less disk space for the amount of data we have reserved in the
6130 * case of compression.
6132 * If this is a reservation and the block group has become read only we cannot
6133 * make the reservation and return -EAGAIN, otherwise this function always
6136 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6137 u64 num_bytes, int reserve, int delalloc)
6139 struct btrfs_space_info *space_info = cache->space_info;
6142 spin_lock(&space_info->lock);
6143 spin_lock(&cache->lock);
6144 if (reserve != RESERVE_FREE) {
6148 cache->reserved += num_bytes;
6149 space_info->bytes_reserved += num_bytes;
6150 if (reserve == RESERVE_ALLOC) {
6151 trace_btrfs_space_reservation(cache->fs_info,
6152 "space_info", space_info->flags,
6154 space_info->bytes_may_use -= num_bytes;
6158 cache->delalloc_bytes += num_bytes;
6162 space_info->bytes_readonly += num_bytes;
6163 cache->reserved -= num_bytes;
6164 space_info->bytes_reserved -= num_bytes;
6167 cache->delalloc_bytes -= num_bytes;
6169 spin_unlock(&cache->lock);
6170 spin_unlock(&space_info->lock);
6174 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6175 struct btrfs_root *root)
6177 struct btrfs_fs_info *fs_info = root->fs_info;
6178 struct btrfs_caching_control *next;
6179 struct btrfs_caching_control *caching_ctl;
6180 struct btrfs_block_group_cache *cache;
6182 down_write(&fs_info->commit_root_sem);
6184 list_for_each_entry_safe(caching_ctl, next,
6185 &fs_info->caching_block_groups, list) {
6186 cache = caching_ctl->block_group;
6187 if (block_group_cache_done(cache)) {
6188 cache->last_byte_to_unpin = (u64)-1;
6189 list_del_init(&caching_ctl->list);
6190 put_caching_control(caching_ctl);
6192 cache->last_byte_to_unpin = caching_ctl->progress;
6196 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6197 fs_info->pinned_extents = &fs_info->freed_extents[1];
6199 fs_info->pinned_extents = &fs_info->freed_extents[0];
6201 up_write(&fs_info->commit_root_sem);
6203 update_global_block_rsv(fs_info);
6207 * Returns the free cluster for the given space info and sets empty_cluster to
6208 * what it should be based on the mount options.
6210 static struct btrfs_free_cluster *
6211 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6214 struct btrfs_free_cluster *ret = NULL;
6215 bool ssd = btrfs_test_opt(root, SSD);
6218 if (btrfs_mixed_space_info(space_info))
6222 *empty_cluster = 2 * 1024 * 1024;
6223 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6224 ret = &root->fs_info->meta_alloc_cluster;
6226 *empty_cluster = 64 * 1024;
6227 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6228 ret = &root->fs_info->data_alloc_cluster;
6234 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6235 const bool return_free_space)
6237 struct btrfs_fs_info *fs_info = root->fs_info;
6238 struct btrfs_block_group_cache *cache = NULL;
6239 struct btrfs_space_info *space_info;
6240 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6241 struct btrfs_free_cluster *cluster = NULL;
6243 u64 total_unpinned = 0;
6244 u64 empty_cluster = 0;
6247 while (start <= end) {
6250 start >= cache->key.objectid + cache->key.offset) {
6252 btrfs_put_block_group(cache);
6254 cache = btrfs_lookup_block_group(fs_info, start);
6255 BUG_ON(!cache); /* Logic error */
6257 cluster = fetch_cluster_info(root,
6260 empty_cluster <<= 1;
6263 len = cache->key.objectid + cache->key.offset - start;
6264 len = min(len, end + 1 - start);
6266 if (start < cache->last_byte_to_unpin) {
6267 len = min(len, cache->last_byte_to_unpin - start);
6268 if (return_free_space)
6269 btrfs_add_free_space(cache, start, len);
6273 total_unpinned += len;
6274 space_info = cache->space_info;
6277 * If this space cluster has been marked as fragmented and we've
6278 * unpinned enough in this block group to potentially allow a
6279 * cluster to be created inside of it go ahead and clear the
6282 if (cluster && cluster->fragmented &&
6283 total_unpinned > empty_cluster) {
6284 spin_lock(&cluster->lock);
6285 cluster->fragmented = 0;
6286 spin_unlock(&cluster->lock);
6289 spin_lock(&space_info->lock);
6290 spin_lock(&cache->lock);
6291 cache->pinned -= len;
6292 space_info->bytes_pinned -= len;
6293 space_info->max_extent_size = 0;
6294 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6296 space_info->bytes_readonly += len;
6299 spin_unlock(&cache->lock);
6300 if (!readonly && global_rsv->space_info == space_info) {
6301 spin_lock(&global_rsv->lock);
6302 if (!global_rsv->full) {
6303 len = min(len, global_rsv->size -
6304 global_rsv->reserved);
6305 global_rsv->reserved += len;
6306 space_info->bytes_may_use += len;
6307 if (global_rsv->reserved >= global_rsv->size)
6308 global_rsv->full = 1;
6310 spin_unlock(&global_rsv->lock);
6312 spin_unlock(&space_info->lock);
6316 btrfs_put_block_group(cache);
6320 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6321 struct btrfs_root *root)
6323 struct btrfs_fs_info *fs_info = root->fs_info;
6324 struct btrfs_block_group_cache *block_group, *tmp;
6325 struct list_head *deleted_bgs;
6326 struct extent_io_tree *unpin;
6331 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6332 unpin = &fs_info->freed_extents[1];
6334 unpin = &fs_info->freed_extents[0];
6336 while (!trans->aborted) {
6337 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6338 ret = find_first_extent_bit(unpin, 0, &start, &end,
6339 EXTENT_DIRTY, NULL);
6341 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6345 if (btrfs_test_opt(root, DISCARD))
6346 ret = btrfs_discard_extent(root, start,
6347 end + 1 - start, NULL);
6349 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6350 unpin_extent_range(root, start, end, true);
6351 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6356 * Transaction is finished. We don't need the lock anymore. We
6357 * do need to clean up the block groups in case of a transaction
6360 deleted_bgs = &trans->transaction->deleted_bgs;
6361 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6365 if (!trans->aborted)
6366 ret = btrfs_discard_extent(root,
6367 block_group->key.objectid,
6368 block_group->key.offset,
6371 list_del_init(&block_group->bg_list);
6372 btrfs_put_block_group_trimming(block_group);
6373 btrfs_put_block_group(block_group);
6376 const char *errstr = btrfs_decode_error(ret);
6378 "Discard failed while removing blockgroup: errno=%d %s\n",
6386 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6387 u64 owner, u64 root_objectid)
6389 struct btrfs_space_info *space_info;
6392 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6393 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6394 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6396 flags = BTRFS_BLOCK_GROUP_METADATA;
6398 flags = BTRFS_BLOCK_GROUP_DATA;
6401 space_info = __find_space_info(fs_info, flags);
6402 BUG_ON(!space_info); /* Logic bug */
6403 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6407 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6408 struct btrfs_root *root,
6409 struct btrfs_delayed_ref_node *node, u64 parent,
6410 u64 root_objectid, u64 owner_objectid,
6411 u64 owner_offset, int refs_to_drop,
6412 struct btrfs_delayed_extent_op *extent_op)
6414 struct btrfs_key key;
6415 struct btrfs_path *path;
6416 struct btrfs_fs_info *info = root->fs_info;
6417 struct btrfs_root *extent_root = info->extent_root;
6418 struct extent_buffer *leaf;
6419 struct btrfs_extent_item *ei;
6420 struct btrfs_extent_inline_ref *iref;
6423 int extent_slot = 0;
6424 int found_extent = 0;
6428 u64 bytenr = node->bytenr;
6429 u64 num_bytes = node->num_bytes;
6431 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6434 path = btrfs_alloc_path();
6439 path->leave_spinning = 1;
6441 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6442 BUG_ON(!is_data && refs_to_drop != 1);
6445 skinny_metadata = 0;
6447 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6448 bytenr, num_bytes, parent,
6449 root_objectid, owner_objectid,
6452 extent_slot = path->slots[0];
6453 while (extent_slot >= 0) {
6454 btrfs_item_key_to_cpu(path->nodes[0], &key,
6456 if (key.objectid != bytenr)
6458 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6459 key.offset == num_bytes) {
6463 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6464 key.offset == owner_objectid) {
6468 if (path->slots[0] - extent_slot > 5)
6472 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6473 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6474 if (found_extent && item_size < sizeof(*ei))
6477 if (!found_extent) {
6479 ret = remove_extent_backref(trans, extent_root, path,
6481 is_data, &last_ref);
6483 btrfs_abort_transaction(trans, extent_root, ret);
6486 btrfs_release_path(path);
6487 path->leave_spinning = 1;
6489 key.objectid = bytenr;
6490 key.type = BTRFS_EXTENT_ITEM_KEY;
6491 key.offset = num_bytes;
6493 if (!is_data && skinny_metadata) {
6494 key.type = BTRFS_METADATA_ITEM_KEY;
6495 key.offset = owner_objectid;
6498 ret = btrfs_search_slot(trans, extent_root,
6500 if (ret > 0 && skinny_metadata && path->slots[0]) {
6502 * Couldn't find our skinny metadata item,
6503 * see if we have ye olde extent item.
6506 btrfs_item_key_to_cpu(path->nodes[0], &key,
6508 if (key.objectid == bytenr &&
6509 key.type == BTRFS_EXTENT_ITEM_KEY &&
6510 key.offset == num_bytes)
6514 if (ret > 0 && skinny_metadata) {
6515 skinny_metadata = false;
6516 key.objectid = bytenr;
6517 key.type = BTRFS_EXTENT_ITEM_KEY;
6518 key.offset = num_bytes;
6519 btrfs_release_path(path);
6520 ret = btrfs_search_slot(trans, extent_root,
6525 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6528 btrfs_print_leaf(extent_root,
6532 btrfs_abort_transaction(trans, extent_root, ret);
6535 extent_slot = path->slots[0];
6537 } else if (WARN_ON(ret == -ENOENT)) {
6538 btrfs_print_leaf(extent_root, path->nodes[0]);
6540 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6541 bytenr, parent, root_objectid, owner_objectid,
6543 btrfs_abort_transaction(trans, extent_root, ret);
6546 btrfs_abort_transaction(trans, extent_root, ret);
6550 leaf = path->nodes[0];
6551 item_size = btrfs_item_size_nr(leaf, extent_slot);
6552 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6553 if (item_size < sizeof(*ei)) {
6554 BUG_ON(found_extent || extent_slot != path->slots[0]);
6555 ret = convert_extent_item_v0(trans, extent_root, path,
6558 btrfs_abort_transaction(trans, extent_root, ret);
6562 btrfs_release_path(path);
6563 path->leave_spinning = 1;
6565 key.objectid = bytenr;
6566 key.type = BTRFS_EXTENT_ITEM_KEY;
6567 key.offset = num_bytes;
6569 ret = btrfs_search_slot(trans, extent_root, &key, path,
6572 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6574 btrfs_print_leaf(extent_root, path->nodes[0]);
6577 btrfs_abort_transaction(trans, extent_root, ret);
6581 extent_slot = path->slots[0];
6582 leaf = path->nodes[0];
6583 item_size = btrfs_item_size_nr(leaf, extent_slot);
6586 BUG_ON(item_size < sizeof(*ei));
6587 ei = btrfs_item_ptr(leaf, extent_slot,
6588 struct btrfs_extent_item);
6589 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6590 key.type == BTRFS_EXTENT_ITEM_KEY) {
6591 struct btrfs_tree_block_info *bi;
6592 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6593 bi = (struct btrfs_tree_block_info *)(ei + 1);
6594 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6597 refs = btrfs_extent_refs(leaf, ei);
6598 if (refs < refs_to_drop) {
6599 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6600 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6602 btrfs_abort_transaction(trans, extent_root, ret);
6605 refs -= refs_to_drop;
6609 __run_delayed_extent_op(extent_op, leaf, ei);
6611 * In the case of inline back ref, reference count will
6612 * be updated by remove_extent_backref
6615 BUG_ON(!found_extent);
6617 btrfs_set_extent_refs(leaf, ei, refs);
6618 btrfs_mark_buffer_dirty(leaf);
6621 ret = remove_extent_backref(trans, extent_root, path,
6623 is_data, &last_ref);
6625 btrfs_abort_transaction(trans, extent_root, ret);
6629 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6633 BUG_ON(is_data && refs_to_drop !=
6634 extent_data_ref_count(path, iref));
6636 BUG_ON(path->slots[0] != extent_slot);
6638 BUG_ON(path->slots[0] != extent_slot + 1);
6639 path->slots[0] = extent_slot;
6645 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6648 btrfs_abort_transaction(trans, extent_root, ret);
6651 btrfs_release_path(path);
6654 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6656 btrfs_abort_transaction(trans, extent_root, ret);
6661 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6663 btrfs_abort_transaction(trans, extent_root, ret);
6667 btrfs_release_path(path);
6670 btrfs_free_path(path);
6675 * when we free an block, it is possible (and likely) that we free the last
6676 * delayed ref for that extent as well. This searches the delayed ref tree for
6677 * a given extent, and if there are no other delayed refs to be processed, it
6678 * removes it from the tree.
6680 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6681 struct btrfs_root *root, u64 bytenr)
6683 struct btrfs_delayed_ref_head *head;
6684 struct btrfs_delayed_ref_root *delayed_refs;
6687 delayed_refs = &trans->transaction->delayed_refs;
6688 spin_lock(&delayed_refs->lock);
6689 head = btrfs_find_delayed_ref_head(trans, bytenr);
6691 goto out_delayed_unlock;
6693 spin_lock(&head->lock);
6694 if (!list_empty(&head->ref_list))
6697 if (head->extent_op) {
6698 if (!head->must_insert_reserved)
6700 btrfs_free_delayed_extent_op(head->extent_op);
6701 head->extent_op = NULL;
6705 * waiting for the lock here would deadlock. If someone else has it
6706 * locked they are already in the process of dropping it anyway
6708 if (!mutex_trylock(&head->mutex))
6712 * at this point we have a head with no other entries. Go
6713 * ahead and process it.
6715 head->node.in_tree = 0;
6716 rb_erase(&head->href_node, &delayed_refs->href_root);
6718 atomic_dec(&delayed_refs->num_entries);
6721 * we don't take a ref on the node because we're removing it from the
6722 * tree, so we just steal the ref the tree was holding.
6724 delayed_refs->num_heads--;
6725 if (head->processing == 0)
6726 delayed_refs->num_heads_ready--;
6727 head->processing = 0;
6728 spin_unlock(&head->lock);
6729 spin_unlock(&delayed_refs->lock);
6731 BUG_ON(head->extent_op);
6732 if (head->must_insert_reserved)
6735 mutex_unlock(&head->mutex);
6736 btrfs_put_delayed_ref(&head->node);
6739 spin_unlock(&head->lock);
6742 spin_unlock(&delayed_refs->lock);
6746 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6747 struct btrfs_root *root,
6748 struct extent_buffer *buf,
6749 u64 parent, int last_ref)
6754 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6755 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6756 buf->start, buf->len,
6757 parent, root->root_key.objectid,
6758 btrfs_header_level(buf),
6759 BTRFS_DROP_DELAYED_REF, NULL);
6760 BUG_ON(ret); /* -ENOMEM */
6766 if (btrfs_header_generation(buf) == trans->transid) {
6767 struct btrfs_block_group_cache *cache;
6769 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6770 ret = check_ref_cleanup(trans, root, buf->start);
6775 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6777 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6778 pin_down_extent(root, cache, buf->start, buf->len, 1);
6779 btrfs_put_block_group(cache);
6783 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6785 btrfs_add_free_space(cache, buf->start, buf->len);
6786 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6787 btrfs_put_block_group(cache);
6788 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6793 add_pinned_bytes(root->fs_info, buf->len,
6794 btrfs_header_level(buf),
6795 root->root_key.objectid);
6798 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6801 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6804 /* Can return -ENOMEM */
6805 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6806 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6807 u64 owner, u64 offset)
6810 struct btrfs_fs_info *fs_info = root->fs_info;
6812 if (btrfs_test_is_dummy_root(root))
6815 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6818 * tree log blocks never actually go into the extent allocation
6819 * tree, just update pinning info and exit early.
6821 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6822 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6823 /* unlocks the pinned mutex */
6824 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6826 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6827 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6829 parent, root_objectid, (int)owner,
6830 BTRFS_DROP_DELAYED_REF, NULL);
6832 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6834 parent, root_objectid, owner,
6836 BTRFS_DROP_DELAYED_REF, NULL);
6842 * when we wait for progress in the block group caching, its because
6843 * our allocation attempt failed at least once. So, we must sleep
6844 * and let some progress happen before we try again.
6846 * This function will sleep at least once waiting for new free space to
6847 * show up, and then it will check the block group free space numbers
6848 * for our min num_bytes. Another option is to have it go ahead
6849 * and look in the rbtree for a free extent of a given size, but this
6852 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6853 * any of the information in this block group.
6855 static noinline void
6856 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6859 struct btrfs_caching_control *caching_ctl;
6861 caching_ctl = get_caching_control(cache);
6865 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6866 (cache->free_space_ctl->free_space >= num_bytes));
6868 put_caching_control(caching_ctl);
6872 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6874 struct btrfs_caching_control *caching_ctl;
6877 caching_ctl = get_caching_control(cache);
6879 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6881 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6882 if (cache->cached == BTRFS_CACHE_ERROR)
6884 put_caching_control(caching_ctl);
6888 int __get_raid_index(u64 flags)
6890 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6891 return BTRFS_RAID_RAID10;
6892 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6893 return BTRFS_RAID_RAID1;
6894 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6895 return BTRFS_RAID_DUP;
6896 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6897 return BTRFS_RAID_RAID0;
6898 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6899 return BTRFS_RAID_RAID5;
6900 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6901 return BTRFS_RAID_RAID6;
6903 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6906 int get_block_group_index(struct btrfs_block_group_cache *cache)
6908 return __get_raid_index(cache->flags);
6911 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6912 [BTRFS_RAID_RAID10] = "raid10",
6913 [BTRFS_RAID_RAID1] = "raid1",
6914 [BTRFS_RAID_DUP] = "dup",
6915 [BTRFS_RAID_RAID0] = "raid0",
6916 [BTRFS_RAID_SINGLE] = "single",
6917 [BTRFS_RAID_RAID5] = "raid5",
6918 [BTRFS_RAID_RAID6] = "raid6",
6921 static const char *get_raid_name(enum btrfs_raid_types type)
6923 if (type >= BTRFS_NR_RAID_TYPES)
6926 return btrfs_raid_type_names[type];
6929 enum btrfs_loop_type {
6930 LOOP_CACHING_NOWAIT = 0,
6931 LOOP_CACHING_WAIT = 1,
6932 LOOP_ALLOC_CHUNK = 2,
6933 LOOP_NO_EMPTY_SIZE = 3,
6937 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6941 down_read(&cache->data_rwsem);
6945 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6948 btrfs_get_block_group(cache);
6950 down_read(&cache->data_rwsem);
6953 static struct btrfs_block_group_cache *
6954 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6955 struct btrfs_free_cluster *cluster,
6958 struct btrfs_block_group_cache *used_bg;
6959 bool locked = false;
6961 spin_lock(&cluster->refill_lock);
6963 if (used_bg == cluster->block_group)
6966 up_read(&used_bg->data_rwsem);
6967 btrfs_put_block_group(used_bg);
6970 used_bg = cluster->block_group;
6974 if (used_bg == block_group)
6977 btrfs_get_block_group(used_bg);
6982 if (down_read_trylock(&used_bg->data_rwsem))
6985 spin_unlock(&cluster->refill_lock);
6986 down_read(&used_bg->data_rwsem);
6992 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6996 up_read(&cache->data_rwsem);
6997 btrfs_put_block_group(cache);
7001 * walks the btree of allocated extents and find a hole of a given size.
7002 * The key ins is changed to record the hole:
7003 * ins->objectid == start position
7004 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7005 * ins->offset == the size of the hole.
7006 * Any available blocks before search_start are skipped.
7008 * If there is no suitable free space, we will record the max size of
7009 * the free space extent currently.
7011 static noinline int find_free_extent(struct btrfs_root *orig_root,
7012 u64 num_bytes, u64 empty_size,
7013 u64 hint_byte, struct btrfs_key *ins,
7014 u64 flags, int delalloc)
7017 struct btrfs_root *root = orig_root->fs_info->extent_root;
7018 struct btrfs_free_cluster *last_ptr = NULL;
7019 struct btrfs_block_group_cache *block_group = NULL;
7020 u64 search_start = 0;
7021 u64 max_extent_size = 0;
7022 u64 empty_cluster = 0;
7023 struct btrfs_space_info *space_info;
7025 int index = __get_raid_index(flags);
7026 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7027 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7028 bool failed_cluster_refill = false;
7029 bool failed_alloc = false;
7030 bool use_cluster = true;
7031 bool have_caching_bg = false;
7032 bool orig_have_caching_bg = false;
7033 bool full_search = false;
7035 WARN_ON(num_bytes < root->sectorsize);
7036 ins->type = BTRFS_EXTENT_ITEM_KEY;
7040 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7042 space_info = __find_space_info(root->fs_info, flags);
7044 btrfs_err(root->fs_info, "No space info for %llu", flags);
7049 * If our free space is heavily fragmented we may not be able to make
7050 * big contiguous allocations, so instead of doing the expensive search
7051 * for free space, simply return ENOSPC with our max_extent_size so we
7052 * can go ahead and search for a more manageable chunk.
7054 * If our max_extent_size is large enough for our allocation simply
7055 * disable clustering since we will likely not be able to find enough
7056 * space to create a cluster and induce latency trying.
7058 if (unlikely(space_info->max_extent_size)) {
7059 spin_lock(&space_info->lock);
7060 if (space_info->max_extent_size &&
7061 num_bytes > space_info->max_extent_size) {
7062 ins->offset = space_info->max_extent_size;
7063 spin_unlock(&space_info->lock);
7065 } else if (space_info->max_extent_size) {
7066 use_cluster = false;
7068 spin_unlock(&space_info->lock);
7071 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7073 spin_lock(&last_ptr->lock);
7074 if (last_ptr->block_group)
7075 hint_byte = last_ptr->window_start;
7076 if (last_ptr->fragmented) {
7078 * We still set window_start so we can keep track of the
7079 * last place we found an allocation to try and save
7082 hint_byte = last_ptr->window_start;
7083 use_cluster = false;
7085 spin_unlock(&last_ptr->lock);
7088 search_start = max(search_start, first_logical_byte(root, 0));
7089 search_start = max(search_start, hint_byte);
7090 if (search_start == hint_byte) {
7091 block_group = btrfs_lookup_block_group(root->fs_info,
7094 * we don't want to use the block group if it doesn't match our
7095 * allocation bits, or if its not cached.
7097 * However if we are re-searching with an ideal block group
7098 * picked out then we don't care that the block group is cached.
7100 if (block_group && block_group_bits(block_group, flags) &&
7101 block_group->cached != BTRFS_CACHE_NO) {
7102 down_read(&space_info->groups_sem);
7103 if (list_empty(&block_group->list) ||
7106 * someone is removing this block group,
7107 * we can't jump into the have_block_group
7108 * target because our list pointers are not
7111 btrfs_put_block_group(block_group);
7112 up_read(&space_info->groups_sem);
7114 index = get_block_group_index(block_group);
7115 btrfs_lock_block_group(block_group, delalloc);
7116 goto have_block_group;
7118 } else if (block_group) {
7119 btrfs_put_block_group(block_group);
7123 have_caching_bg = false;
7124 if (index == 0 || index == __get_raid_index(flags))
7126 down_read(&space_info->groups_sem);
7127 list_for_each_entry(block_group, &space_info->block_groups[index],
7132 btrfs_grab_block_group(block_group, delalloc);
7133 search_start = block_group->key.objectid;
7136 * this can happen if we end up cycling through all the
7137 * raid types, but we want to make sure we only allocate
7138 * for the proper type.
7140 if (!block_group_bits(block_group, flags)) {
7141 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7142 BTRFS_BLOCK_GROUP_RAID1 |
7143 BTRFS_BLOCK_GROUP_RAID5 |
7144 BTRFS_BLOCK_GROUP_RAID6 |
7145 BTRFS_BLOCK_GROUP_RAID10;
7148 * if they asked for extra copies and this block group
7149 * doesn't provide them, bail. This does allow us to
7150 * fill raid0 from raid1.
7152 if ((flags & extra) && !(block_group->flags & extra))
7157 cached = block_group_cache_done(block_group);
7158 if (unlikely(!cached)) {
7159 have_caching_bg = true;
7160 ret = cache_block_group(block_group, 0);
7165 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7167 if (unlikely(block_group->ro))
7171 * Ok we want to try and use the cluster allocator, so
7174 if (last_ptr && use_cluster) {
7175 struct btrfs_block_group_cache *used_block_group;
7176 unsigned long aligned_cluster;
7178 * the refill lock keeps out other
7179 * people trying to start a new cluster
7181 used_block_group = btrfs_lock_cluster(block_group,
7184 if (!used_block_group)
7185 goto refill_cluster;
7187 if (used_block_group != block_group &&
7188 (used_block_group->ro ||
7189 !block_group_bits(used_block_group, flags)))
7190 goto release_cluster;
7192 offset = btrfs_alloc_from_cluster(used_block_group,
7195 used_block_group->key.objectid,
7198 /* we have a block, we're done */
7199 spin_unlock(&last_ptr->refill_lock);
7200 trace_btrfs_reserve_extent_cluster(root,
7202 search_start, num_bytes);
7203 if (used_block_group != block_group) {
7204 btrfs_release_block_group(block_group,
7206 block_group = used_block_group;
7211 WARN_ON(last_ptr->block_group != used_block_group);
7213 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7214 * set up a new clusters, so lets just skip it
7215 * and let the allocator find whatever block
7216 * it can find. If we reach this point, we
7217 * will have tried the cluster allocator
7218 * plenty of times and not have found
7219 * anything, so we are likely way too
7220 * fragmented for the clustering stuff to find
7223 * However, if the cluster is taken from the
7224 * current block group, release the cluster
7225 * first, so that we stand a better chance of
7226 * succeeding in the unclustered
7228 if (loop >= LOOP_NO_EMPTY_SIZE &&
7229 used_block_group != block_group) {
7230 spin_unlock(&last_ptr->refill_lock);
7231 btrfs_release_block_group(used_block_group,
7233 goto unclustered_alloc;
7237 * this cluster didn't work out, free it and
7240 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7242 if (used_block_group != block_group)
7243 btrfs_release_block_group(used_block_group,
7246 if (loop >= LOOP_NO_EMPTY_SIZE) {
7247 spin_unlock(&last_ptr->refill_lock);
7248 goto unclustered_alloc;
7251 aligned_cluster = max_t(unsigned long,
7252 empty_cluster + empty_size,
7253 block_group->full_stripe_len);
7255 /* allocate a cluster in this block group */
7256 ret = btrfs_find_space_cluster(root, block_group,
7257 last_ptr, search_start,
7262 * now pull our allocation out of this
7265 offset = btrfs_alloc_from_cluster(block_group,
7271 /* we found one, proceed */
7272 spin_unlock(&last_ptr->refill_lock);
7273 trace_btrfs_reserve_extent_cluster(root,
7274 block_group, search_start,
7278 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7279 && !failed_cluster_refill) {
7280 spin_unlock(&last_ptr->refill_lock);
7282 failed_cluster_refill = true;
7283 wait_block_group_cache_progress(block_group,
7284 num_bytes + empty_cluster + empty_size);
7285 goto have_block_group;
7289 * at this point we either didn't find a cluster
7290 * or we weren't able to allocate a block from our
7291 * cluster. Free the cluster we've been trying
7292 * to use, and go to the next block group
7294 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7295 spin_unlock(&last_ptr->refill_lock);
7301 * We are doing an unclustered alloc, set the fragmented flag so
7302 * we don't bother trying to setup a cluster again until we get
7305 if (unlikely(last_ptr)) {
7306 spin_lock(&last_ptr->lock);
7307 last_ptr->fragmented = 1;
7308 spin_unlock(&last_ptr->lock);
7310 spin_lock(&block_group->free_space_ctl->tree_lock);
7312 block_group->free_space_ctl->free_space <
7313 num_bytes + empty_cluster + empty_size) {
7314 if (block_group->free_space_ctl->free_space >
7317 block_group->free_space_ctl->free_space;
7318 spin_unlock(&block_group->free_space_ctl->tree_lock);
7321 spin_unlock(&block_group->free_space_ctl->tree_lock);
7323 offset = btrfs_find_space_for_alloc(block_group, search_start,
7324 num_bytes, empty_size,
7327 * If we didn't find a chunk, and we haven't failed on this
7328 * block group before, and this block group is in the middle of
7329 * caching and we are ok with waiting, then go ahead and wait
7330 * for progress to be made, and set failed_alloc to true.
7332 * If failed_alloc is true then we've already waited on this
7333 * block group once and should move on to the next block group.
7335 if (!offset && !failed_alloc && !cached &&
7336 loop > LOOP_CACHING_NOWAIT) {
7337 wait_block_group_cache_progress(block_group,
7338 num_bytes + empty_size);
7339 failed_alloc = true;
7340 goto have_block_group;
7341 } else if (!offset) {
7345 search_start = ALIGN(offset, root->stripesize);
7347 /* move on to the next group */
7348 if (search_start + num_bytes >
7349 block_group->key.objectid + block_group->key.offset) {
7350 btrfs_add_free_space(block_group, offset, num_bytes);
7354 if (offset < search_start)
7355 btrfs_add_free_space(block_group, offset,
7356 search_start - offset);
7357 BUG_ON(offset > search_start);
7359 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7360 alloc_type, delalloc);
7361 if (ret == -EAGAIN) {
7362 btrfs_add_free_space(block_group, offset, num_bytes);
7366 /* we are all good, lets return */
7367 ins->objectid = search_start;
7368 ins->offset = num_bytes;
7370 trace_btrfs_reserve_extent(orig_root, block_group,
7371 search_start, num_bytes);
7372 btrfs_release_block_group(block_group, delalloc);
7375 failed_cluster_refill = false;
7376 failed_alloc = false;
7377 BUG_ON(index != get_block_group_index(block_group));
7378 btrfs_release_block_group(block_group, delalloc);
7380 up_read(&space_info->groups_sem);
7382 if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg
7383 && !orig_have_caching_bg)
7384 orig_have_caching_bg = true;
7386 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7389 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7393 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7394 * caching kthreads as we move along
7395 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7396 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7397 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7400 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7402 if (loop == LOOP_CACHING_NOWAIT) {
7404 * We want to skip the LOOP_CACHING_WAIT step if we
7405 * don't have any unached bgs and we've alrelady done a
7406 * full search through.
7408 if (orig_have_caching_bg || !full_search)
7409 loop = LOOP_CACHING_WAIT;
7411 loop = LOOP_ALLOC_CHUNK;
7416 if (loop == LOOP_ALLOC_CHUNK) {
7417 struct btrfs_trans_handle *trans;
7420 trans = current->journal_info;
7424 trans = btrfs_join_transaction(root);
7426 if (IS_ERR(trans)) {
7427 ret = PTR_ERR(trans);
7431 ret = do_chunk_alloc(trans, root, flags,
7435 * If we can't allocate a new chunk we've already looped
7436 * through at least once, move on to the NO_EMPTY_SIZE
7440 loop = LOOP_NO_EMPTY_SIZE;
7443 * Do not bail out on ENOSPC since we
7444 * can do more things.
7446 if (ret < 0 && ret != -ENOSPC)
7447 btrfs_abort_transaction(trans,
7452 btrfs_end_transaction(trans, root);
7457 if (loop == LOOP_NO_EMPTY_SIZE) {
7459 * Don't loop again if we already have no empty_size and
7462 if (empty_size == 0 &&
7463 empty_cluster == 0) {
7472 } else if (!ins->objectid) {
7474 } else if (ins->objectid) {
7475 if (!use_cluster && last_ptr) {
7476 spin_lock(&last_ptr->lock);
7477 last_ptr->window_start = ins->objectid;
7478 spin_unlock(&last_ptr->lock);
7483 if (ret == -ENOSPC) {
7484 spin_lock(&space_info->lock);
7485 space_info->max_extent_size = max_extent_size;
7486 spin_unlock(&space_info->lock);
7487 ins->offset = max_extent_size;
7492 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7493 int dump_block_groups)
7495 struct btrfs_block_group_cache *cache;
7498 spin_lock(&info->lock);
7499 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7501 info->total_bytes - info->bytes_used - info->bytes_pinned -
7502 info->bytes_reserved - info->bytes_readonly,
7503 (info->full) ? "" : "not ");
7504 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7505 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7506 info->total_bytes, info->bytes_used, info->bytes_pinned,
7507 info->bytes_reserved, info->bytes_may_use,
7508 info->bytes_readonly);
7509 spin_unlock(&info->lock);
7511 if (!dump_block_groups)
7514 down_read(&info->groups_sem);
7516 list_for_each_entry(cache, &info->block_groups[index], list) {
7517 spin_lock(&cache->lock);
7518 printk(KERN_INFO "BTRFS: "
7519 "block group %llu has %llu bytes, "
7520 "%llu used %llu pinned %llu reserved %s\n",
7521 cache->key.objectid, cache->key.offset,
7522 btrfs_block_group_used(&cache->item), cache->pinned,
7523 cache->reserved, cache->ro ? "[readonly]" : "");
7524 btrfs_dump_free_space(cache, bytes);
7525 spin_unlock(&cache->lock);
7527 if (++index < BTRFS_NR_RAID_TYPES)
7529 up_read(&info->groups_sem);
7532 int btrfs_reserve_extent(struct btrfs_root *root,
7533 u64 num_bytes, u64 min_alloc_size,
7534 u64 empty_size, u64 hint_byte,
7535 struct btrfs_key *ins, int is_data, int delalloc)
7537 bool final_tried = num_bytes == min_alloc_size;
7541 flags = btrfs_get_alloc_profile(root, is_data);
7543 WARN_ON(num_bytes < root->sectorsize);
7544 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7547 if (ret == -ENOSPC) {
7548 if (!final_tried && ins->offset) {
7549 num_bytes = min(num_bytes >> 1, ins->offset);
7550 num_bytes = round_down(num_bytes, root->sectorsize);
7551 num_bytes = max(num_bytes, min_alloc_size);
7552 if (num_bytes == min_alloc_size)
7555 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7556 struct btrfs_space_info *sinfo;
7558 sinfo = __find_space_info(root->fs_info, flags);
7559 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7562 dump_space_info(sinfo, num_bytes, 1);
7569 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7571 int pin, int delalloc)
7573 struct btrfs_block_group_cache *cache;
7576 cache = btrfs_lookup_block_group(root->fs_info, start);
7578 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7584 pin_down_extent(root, cache, start, len, 1);
7586 if (btrfs_test_opt(root, DISCARD))
7587 ret = btrfs_discard_extent(root, start, len, NULL);
7588 btrfs_add_free_space(cache, start, len);
7589 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7592 btrfs_put_block_group(cache);
7594 trace_btrfs_reserved_extent_free(root, start, len);
7599 int btrfs_free_reserved_extent(struct btrfs_root *root,
7600 u64 start, u64 len, int delalloc)
7602 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7605 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7608 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7611 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7612 struct btrfs_root *root,
7613 u64 parent, u64 root_objectid,
7614 u64 flags, u64 owner, u64 offset,
7615 struct btrfs_key *ins, int ref_mod)
7618 struct btrfs_fs_info *fs_info = root->fs_info;
7619 struct btrfs_extent_item *extent_item;
7620 struct btrfs_extent_inline_ref *iref;
7621 struct btrfs_path *path;
7622 struct extent_buffer *leaf;
7627 type = BTRFS_SHARED_DATA_REF_KEY;
7629 type = BTRFS_EXTENT_DATA_REF_KEY;
7631 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7633 path = btrfs_alloc_path();
7637 path->leave_spinning = 1;
7638 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7641 btrfs_free_path(path);
7645 leaf = path->nodes[0];
7646 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7647 struct btrfs_extent_item);
7648 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7649 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7650 btrfs_set_extent_flags(leaf, extent_item,
7651 flags | BTRFS_EXTENT_FLAG_DATA);
7653 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7654 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7656 struct btrfs_shared_data_ref *ref;
7657 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7658 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7659 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7661 struct btrfs_extent_data_ref *ref;
7662 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7663 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7664 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7665 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7666 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7669 btrfs_mark_buffer_dirty(path->nodes[0]);
7670 btrfs_free_path(path);
7672 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7673 if (ret) { /* -ENOENT, logic error */
7674 btrfs_err(fs_info, "update block group failed for %llu %llu",
7675 ins->objectid, ins->offset);
7678 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7682 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7683 struct btrfs_root *root,
7684 u64 parent, u64 root_objectid,
7685 u64 flags, struct btrfs_disk_key *key,
7686 int level, struct btrfs_key *ins)
7689 struct btrfs_fs_info *fs_info = root->fs_info;
7690 struct btrfs_extent_item *extent_item;
7691 struct btrfs_tree_block_info *block_info;
7692 struct btrfs_extent_inline_ref *iref;
7693 struct btrfs_path *path;
7694 struct extent_buffer *leaf;
7695 u32 size = sizeof(*extent_item) + sizeof(*iref);
7696 u64 num_bytes = ins->offset;
7697 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7700 if (!skinny_metadata)
7701 size += sizeof(*block_info);
7703 path = btrfs_alloc_path();
7705 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7710 path->leave_spinning = 1;
7711 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7714 btrfs_free_path(path);
7715 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7720 leaf = path->nodes[0];
7721 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7722 struct btrfs_extent_item);
7723 btrfs_set_extent_refs(leaf, extent_item, 1);
7724 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7725 btrfs_set_extent_flags(leaf, extent_item,
7726 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7728 if (skinny_metadata) {
7729 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7730 num_bytes = root->nodesize;
7732 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7733 btrfs_set_tree_block_key(leaf, block_info, key);
7734 btrfs_set_tree_block_level(leaf, block_info, level);
7735 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7739 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7740 btrfs_set_extent_inline_ref_type(leaf, iref,
7741 BTRFS_SHARED_BLOCK_REF_KEY);
7742 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7744 btrfs_set_extent_inline_ref_type(leaf, iref,
7745 BTRFS_TREE_BLOCK_REF_KEY);
7746 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7749 btrfs_mark_buffer_dirty(leaf);
7750 btrfs_free_path(path);
7752 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7754 if (ret) { /* -ENOENT, logic error */
7755 btrfs_err(fs_info, "update block group failed for %llu %llu",
7756 ins->objectid, ins->offset);
7760 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7764 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7765 struct btrfs_root *root,
7766 u64 root_objectid, u64 owner,
7767 u64 offset, u64 ram_bytes,
7768 struct btrfs_key *ins)
7772 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7774 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7776 root_objectid, owner, offset,
7777 ram_bytes, BTRFS_ADD_DELAYED_EXTENT,
7783 * this is used by the tree logging recovery code. It records that
7784 * an extent has been allocated and makes sure to clear the free
7785 * space cache bits as well
7787 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7788 struct btrfs_root *root,
7789 u64 root_objectid, u64 owner, u64 offset,
7790 struct btrfs_key *ins)
7793 struct btrfs_block_group_cache *block_group;
7796 * Mixed block groups will exclude before processing the log so we only
7797 * need to do the exlude dance if this fs isn't mixed.
7799 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7800 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7805 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7809 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7810 RESERVE_ALLOC_NO_ACCOUNT, 0);
7811 BUG_ON(ret); /* logic error */
7812 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7813 0, owner, offset, ins, 1);
7814 btrfs_put_block_group(block_group);
7818 static struct extent_buffer *
7819 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7820 u64 bytenr, int level)
7822 struct extent_buffer *buf;
7824 buf = btrfs_find_create_tree_block(root, bytenr);
7826 return ERR_PTR(-ENOMEM);
7827 btrfs_set_header_generation(buf, trans->transid);
7828 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7829 btrfs_tree_lock(buf);
7830 clean_tree_block(trans, root->fs_info, buf);
7831 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7833 btrfs_set_lock_blocking(buf);
7834 btrfs_set_buffer_uptodate(buf);
7836 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7837 buf->log_index = root->log_transid % 2;
7839 * we allow two log transactions at a time, use different
7840 * EXENT bit to differentiate dirty pages.
7842 if (buf->log_index == 0)
7843 set_extent_dirty(&root->dirty_log_pages, buf->start,
7844 buf->start + buf->len - 1, GFP_NOFS);
7846 set_extent_new(&root->dirty_log_pages, buf->start,
7847 buf->start + buf->len - 1, GFP_NOFS);
7849 buf->log_index = -1;
7850 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7851 buf->start + buf->len - 1, GFP_NOFS);
7853 trans->blocks_used++;
7854 /* this returns a buffer locked for blocking */
7858 static struct btrfs_block_rsv *
7859 use_block_rsv(struct btrfs_trans_handle *trans,
7860 struct btrfs_root *root, u32 blocksize)
7862 struct btrfs_block_rsv *block_rsv;
7863 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7865 bool global_updated = false;
7867 block_rsv = get_block_rsv(trans, root);
7869 if (unlikely(block_rsv->size == 0))
7872 ret = block_rsv_use_bytes(block_rsv, blocksize);
7876 if (block_rsv->failfast)
7877 return ERR_PTR(ret);
7879 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7880 global_updated = true;
7881 update_global_block_rsv(root->fs_info);
7885 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7886 static DEFINE_RATELIMIT_STATE(_rs,
7887 DEFAULT_RATELIMIT_INTERVAL * 10,
7888 /*DEFAULT_RATELIMIT_BURST*/ 1);
7889 if (__ratelimit(&_rs))
7891 "BTRFS: block rsv returned %d\n", ret);
7894 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7895 BTRFS_RESERVE_NO_FLUSH);
7899 * If we couldn't reserve metadata bytes try and use some from
7900 * the global reserve if its space type is the same as the global
7903 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7904 block_rsv->space_info == global_rsv->space_info) {
7905 ret = block_rsv_use_bytes(global_rsv, blocksize);
7909 return ERR_PTR(ret);
7912 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7913 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7915 block_rsv_add_bytes(block_rsv, blocksize, 0);
7916 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7920 * finds a free extent and does all the dirty work required for allocation
7921 * returns the tree buffer or an ERR_PTR on error.
7923 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7924 struct btrfs_root *root,
7925 u64 parent, u64 root_objectid,
7926 struct btrfs_disk_key *key, int level,
7927 u64 hint, u64 empty_size)
7929 struct btrfs_key ins;
7930 struct btrfs_block_rsv *block_rsv;
7931 struct extent_buffer *buf;
7932 struct btrfs_delayed_extent_op *extent_op;
7935 u32 blocksize = root->nodesize;
7936 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7939 if (btrfs_test_is_dummy_root(root)) {
7940 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7943 root->alloc_bytenr += blocksize;
7947 block_rsv = use_block_rsv(trans, root, blocksize);
7948 if (IS_ERR(block_rsv))
7949 return ERR_CAST(block_rsv);
7951 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7952 empty_size, hint, &ins, 0, 0);
7956 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7959 goto out_free_reserved;
7962 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7964 parent = ins.objectid;
7965 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7969 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7970 extent_op = btrfs_alloc_delayed_extent_op();
7976 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7978 memset(&extent_op->key, 0, sizeof(extent_op->key));
7979 extent_op->flags_to_set = flags;
7980 if (skinny_metadata)
7981 extent_op->update_key = 0;
7983 extent_op->update_key = 1;
7984 extent_op->update_flags = 1;
7985 extent_op->is_data = 0;
7986 extent_op->level = level;
7988 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7989 ins.objectid, ins.offset,
7990 parent, root_objectid, level,
7991 BTRFS_ADD_DELAYED_EXTENT,
7994 goto out_free_delayed;
7999 btrfs_free_delayed_extent_op(extent_op);
8001 free_extent_buffer(buf);
8003 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8005 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8006 return ERR_PTR(ret);
8009 struct walk_control {
8010 u64 refs[BTRFS_MAX_LEVEL];
8011 u64 flags[BTRFS_MAX_LEVEL];
8012 struct btrfs_key update_progress;
8023 #define DROP_REFERENCE 1
8024 #define UPDATE_BACKREF 2
8026 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8027 struct btrfs_root *root,
8028 struct walk_control *wc,
8029 struct btrfs_path *path)
8037 struct btrfs_key key;
8038 struct extent_buffer *eb;
8043 if (path->slots[wc->level] < wc->reada_slot) {
8044 wc->reada_count = wc->reada_count * 2 / 3;
8045 wc->reada_count = max(wc->reada_count, 2);
8047 wc->reada_count = wc->reada_count * 3 / 2;
8048 wc->reada_count = min_t(int, wc->reada_count,
8049 BTRFS_NODEPTRS_PER_BLOCK(root));
8052 eb = path->nodes[wc->level];
8053 nritems = btrfs_header_nritems(eb);
8054 blocksize = root->nodesize;
8056 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8057 if (nread >= wc->reada_count)
8061 bytenr = btrfs_node_blockptr(eb, slot);
8062 generation = btrfs_node_ptr_generation(eb, slot);
8064 if (slot == path->slots[wc->level])
8067 if (wc->stage == UPDATE_BACKREF &&
8068 generation <= root->root_key.offset)
8071 /* We don't lock the tree block, it's OK to be racy here */
8072 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8073 wc->level - 1, 1, &refs,
8075 /* We don't care about errors in readahead. */
8080 if (wc->stage == DROP_REFERENCE) {
8084 if (wc->level == 1 &&
8085 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8087 if (!wc->update_ref ||
8088 generation <= root->root_key.offset)
8090 btrfs_node_key_to_cpu(eb, &key, slot);
8091 ret = btrfs_comp_cpu_keys(&key,
8092 &wc->update_progress);
8096 if (wc->level == 1 &&
8097 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8101 readahead_tree_block(root, bytenr);
8104 wc->reada_slot = slot;
8108 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
8109 * for later qgroup accounting.
8111 * Current, this function does nothing.
8113 static int account_leaf_items(struct btrfs_trans_handle *trans,
8114 struct btrfs_root *root,
8115 struct extent_buffer *eb)
8117 int nr = btrfs_header_nritems(eb);
8119 struct btrfs_key key;
8120 struct btrfs_file_extent_item *fi;
8121 u64 bytenr, num_bytes;
8123 for (i = 0; i < nr; i++) {
8124 btrfs_item_key_to_cpu(eb, &key, i);
8126 if (key.type != BTRFS_EXTENT_DATA_KEY)
8129 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8130 /* filter out non qgroup-accountable extents */
8131 extent_type = btrfs_file_extent_type(eb, fi);
8133 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8136 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8140 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8146 * Walk up the tree from the bottom, freeing leaves and any interior
8147 * nodes which have had all slots visited. If a node (leaf or
8148 * interior) is freed, the node above it will have it's slot
8149 * incremented. The root node will never be freed.
8151 * At the end of this function, we should have a path which has all
8152 * slots incremented to the next position for a search. If we need to
8153 * read a new node it will be NULL and the node above it will have the
8154 * correct slot selected for a later read.
8156 * If we increment the root nodes slot counter past the number of
8157 * elements, 1 is returned to signal completion of the search.
8159 static int adjust_slots_upwards(struct btrfs_root *root,
8160 struct btrfs_path *path, int root_level)
8164 struct extent_buffer *eb;
8166 if (root_level == 0)
8169 while (level <= root_level) {
8170 eb = path->nodes[level];
8171 nr = btrfs_header_nritems(eb);
8172 path->slots[level]++;
8173 slot = path->slots[level];
8174 if (slot >= nr || level == 0) {
8176 * Don't free the root - we will detect this
8177 * condition after our loop and return a
8178 * positive value for caller to stop walking the tree.
8180 if (level != root_level) {
8181 btrfs_tree_unlock_rw(eb, path->locks[level]);
8182 path->locks[level] = 0;
8184 free_extent_buffer(eb);
8185 path->nodes[level] = NULL;
8186 path->slots[level] = 0;
8190 * We have a valid slot to walk back down
8191 * from. Stop here so caller can process these
8200 eb = path->nodes[root_level];
8201 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8208 * root_eb is the subtree root and is locked before this function is called.
8209 * TODO: Modify this function to mark all (including complete shared node)
8210 * to dirty_extent_root to allow it get accounted in qgroup.
8212 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8213 struct btrfs_root *root,
8214 struct extent_buffer *root_eb,
8220 struct extent_buffer *eb = root_eb;
8221 struct btrfs_path *path = NULL;
8223 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8224 BUG_ON(root_eb == NULL);
8226 if (!root->fs_info->quota_enabled)
8229 if (!extent_buffer_uptodate(root_eb)) {
8230 ret = btrfs_read_buffer(root_eb, root_gen);
8235 if (root_level == 0) {
8236 ret = account_leaf_items(trans, root, root_eb);
8240 path = btrfs_alloc_path();
8245 * Walk down the tree. Missing extent blocks are filled in as
8246 * we go. Metadata is accounted every time we read a new
8249 * When we reach a leaf, we account for file extent items in it,
8250 * walk back up the tree (adjusting slot pointers as we go)
8251 * and restart the search process.
8253 extent_buffer_get(root_eb); /* For path */
8254 path->nodes[root_level] = root_eb;
8255 path->slots[root_level] = 0;
8256 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8259 while (level >= 0) {
8260 if (path->nodes[level] == NULL) {
8265 /* We need to get child blockptr/gen from
8266 * parent before we can read it. */
8267 eb = path->nodes[level + 1];
8268 parent_slot = path->slots[level + 1];
8269 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8270 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8272 eb = read_tree_block(root, child_bytenr, child_gen);
8276 } else if (!extent_buffer_uptodate(eb)) {
8277 free_extent_buffer(eb);
8282 path->nodes[level] = eb;
8283 path->slots[level] = 0;
8285 btrfs_tree_read_lock(eb);
8286 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8287 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8291 ret = account_leaf_items(trans, root, path->nodes[level]);
8295 /* Nonzero return here means we completed our search */
8296 ret = adjust_slots_upwards(root, path, root_level);
8300 /* Restart search with new slots */
8309 btrfs_free_path(path);
8315 * helper to process tree block while walking down the tree.
8317 * when wc->stage == UPDATE_BACKREF, this function updates
8318 * back refs for pointers in the block.
8320 * NOTE: return value 1 means we should stop walking down.
8322 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8323 struct btrfs_root *root,
8324 struct btrfs_path *path,
8325 struct walk_control *wc, int lookup_info)
8327 int level = wc->level;
8328 struct extent_buffer *eb = path->nodes[level];
8329 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8332 if (wc->stage == UPDATE_BACKREF &&
8333 btrfs_header_owner(eb) != root->root_key.objectid)
8337 * when reference count of tree block is 1, it won't increase
8338 * again. once full backref flag is set, we never clear it.
8341 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8342 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8343 BUG_ON(!path->locks[level]);
8344 ret = btrfs_lookup_extent_info(trans, root,
8345 eb->start, level, 1,
8348 BUG_ON(ret == -ENOMEM);
8351 BUG_ON(wc->refs[level] == 0);
8354 if (wc->stage == DROP_REFERENCE) {
8355 if (wc->refs[level] > 1)
8358 if (path->locks[level] && !wc->keep_locks) {
8359 btrfs_tree_unlock_rw(eb, path->locks[level]);
8360 path->locks[level] = 0;
8365 /* wc->stage == UPDATE_BACKREF */
8366 if (!(wc->flags[level] & flag)) {
8367 BUG_ON(!path->locks[level]);
8368 ret = btrfs_inc_ref(trans, root, eb, 1);
8369 BUG_ON(ret); /* -ENOMEM */
8370 ret = btrfs_dec_ref(trans, root, eb, 0);
8371 BUG_ON(ret); /* -ENOMEM */
8372 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8374 btrfs_header_level(eb), 0);
8375 BUG_ON(ret); /* -ENOMEM */
8376 wc->flags[level] |= flag;
8380 * the block is shared by multiple trees, so it's not good to
8381 * keep the tree lock
8383 if (path->locks[level] && level > 0) {
8384 btrfs_tree_unlock_rw(eb, path->locks[level]);
8385 path->locks[level] = 0;
8391 * helper to process tree block pointer.
8393 * when wc->stage == DROP_REFERENCE, this function checks
8394 * reference count of the block pointed to. if the block
8395 * is shared and we need update back refs for the subtree
8396 * rooted at the block, this function changes wc->stage to
8397 * UPDATE_BACKREF. if the block is shared and there is no
8398 * need to update back, this function drops the reference
8401 * NOTE: return value 1 means we should stop walking down.
8403 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8404 struct btrfs_root *root,
8405 struct btrfs_path *path,
8406 struct walk_control *wc, int *lookup_info)
8412 struct btrfs_key key;
8413 struct extent_buffer *next;
8414 int level = wc->level;
8417 bool need_account = false;
8419 generation = btrfs_node_ptr_generation(path->nodes[level],
8420 path->slots[level]);
8422 * if the lower level block was created before the snapshot
8423 * was created, we know there is no need to update back refs
8426 if (wc->stage == UPDATE_BACKREF &&
8427 generation <= root->root_key.offset) {
8432 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8433 blocksize = root->nodesize;
8435 next = btrfs_find_tree_block(root->fs_info, bytenr);
8437 next = btrfs_find_create_tree_block(root, bytenr);
8440 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8444 btrfs_tree_lock(next);
8445 btrfs_set_lock_blocking(next);
8447 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8448 &wc->refs[level - 1],
8449 &wc->flags[level - 1]);
8451 btrfs_tree_unlock(next);
8455 if (unlikely(wc->refs[level - 1] == 0)) {
8456 btrfs_err(root->fs_info, "Missing references.");
8461 if (wc->stage == DROP_REFERENCE) {
8462 if (wc->refs[level - 1] > 1) {
8463 need_account = true;
8465 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8468 if (!wc->update_ref ||
8469 generation <= root->root_key.offset)
8472 btrfs_node_key_to_cpu(path->nodes[level], &key,
8473 path->slots[level]);
8474 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8478 wc->stage = UPDATE_BACKREF;
8479 wc->shared_level = level - 1;
8483 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8487 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8488 btrfs_tree_unlock(next);
8489 free_extent_buffer(next);
8495 if (reada && level == 1)
8496 reada_walk_down(trans, root, wc, path);
8497 next = read_tree_block(root, bytenr, generation);
8499 return PTR_ERR(next);
8500 } else if (!extent_buffer_uptodate(next)) {
8501 free_extent_buffer(next);
8504 btrfs_tree_lock(next);
8505 btrfs_set_lock_blocking(next);
8509 BUG_ON(level != btrfs_header_level(next));
8510 path->nodes[level] = next;
8511 path->slots[level] = 0;
8512 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8518 wc->refs[level - 1] = 0;
8519 wc->flags[level - 1] = 0;
8520 if (wc->stage == DROP_REFERENCE) {
8521 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8522 parent = path->nodes[level]->start;
8524 BUG_ON(root->root_key.objectid !=
8525 btrfs_header_owner(path->nodes[level]));
8530 ret = account_shared_subtree(trans, root, next,
8531 generation, level - 1);
8533 btrfs_err_rl(root->fs_info,
8535 "%d accounting shared subtree. Quota "
8536 "is out of sync, rescan required.",
8540 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8541 root->root_key.objectid, level - 1, 0);
8542 BUG_ON(ret); /* -ENOMEM */
8544 btrfs_tree_unlock(next);
8545 free_extent_buffer(next);
8551 * helper to process tree block while walking up the tree.
8553 * when wc->stage == DROP_REFERENCE, this function drops
8554 * reference count on the block.
8556 * when wc->stage == UPDATE_BACKREF, this function changes
8557 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8558 * to UPDATE_BACKREF previously while processing the block.
8560 * NOTE: return value 1 means we should stop walking up.
8562 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8563 struct btrfs_root *root,
8564 struct btrfs_path *path,
8565 struct walk_control *wc)
8568 int level = wc->level;
8569 struct extent_buffer *eb = path->nodes[level];
8572 if (wc->stage == UPDATE_BACKREF) {
8573 BUG_ON(wc->shared_level < level);
8574 if (level < wc->shared_level)
8577 ret = find_next_key(path, level + 1, &wc->update_progress);
8581 wc->stage = DROP_REFERENCE;
8582 wc->shared_level = -1;
8583 path->slots[level] = 0;
8586 * check reference count again if the block isn't locked.
8587 * we should start walking down the tree again if reference
8590 if (!path->locks[level]) {
8592 btrfs_tree_lock(eb);
8593 btrfs_set_lock_blocking(eb);
8594 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8596 ret = btrfs_lookup_extent_info(trans, root,
8597 eb->start, level, 1,
8601 btrfs_tree_unlock_rw(eb, path->locks[level]);
8602 path->locks[level] = 0;
8605 BUG_ON(wc->refs[level] == 0);
8606 if (wc->refs[level] == 1) {
8607 btrfs_tree_unlock_rw(eb, path->locks[level]);
8608 path->locks[level] = 0;
8614 /* wc->stage == DROP_REFERENCE */
8615 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8617 if (wc->refs[level] == 1) {
8619 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8620 ret = btrfs_dec_ref(trans, root, eb, 1);
8622 ret = btrfs_dec_ref(trans, root, eb, 0);
8623 BUG_ON(ret); /* -ENOMEM */
8624 ret = account_leaf_items(trans, root, eb);
8626 btrfs_err_rl(root->fs_info,
8628 "%d accounting leaf items. Quota "
8629 "is out of sync, rescan required.",
8633 /* make block locked assertion in clean_tree_block happy */
8634 if (!path->locks[level] &&
8635 btrfs_header_generation(eb) == trans->transid) {
8636 btrfs_tree_lock(eb);
8637 btrfs_set_lock_blocking(eb);
8638 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8640 clean_tree_block(trans, root->fs_info, eb);
8643 if (eb == root->node) {
8644 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8647 BUG_ON(root->root_key.objectid !=
8648 btrfs_header_owner(eb));
8650 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8651 parent = path->nodes[level + 1]->start;
8653 BUG_ON(root->root_key.objectid !=
8654 btrfs_header_owner(path->nodes[level + 1]));
8657 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8659 wc->refs[level] = 0;
8660 wc->flags[level] = 0;
8664 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8665 struct btrfs_root *root,
8666 struct btrfs_path *path,
8667 struct walk_control *wc)
8669 int level = wc->level;
8670 int lookup_info = 1;
8673 while (level >= 0) {
8674 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8681 if (path->slots[level] >=
8682 btrfs_header_nritems(path->nodes[level]))
8685 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8687 path->slots[level]++;
8696 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8697 struct btrfs_root *root,
8698 struct btrfs_path *path,
8699 struct walk_control *wc, int max_level)
8701 int level = wc->level;
8704 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8705 while (level < max_level && path->nodes[level]) {
8707 if (path->slots[level] + 1 <
8708 btrfs_header_nritems(path->nodes[level])) {
8709 path->slots[level]++;
8712 ret = walk_up_proc(trans, root, path, wc);
8716 if (path->locks[level]) {
8717 btrfs_tree_unlock_rw(path->nodes[level],
8718 path->locks[level]);
8719 path->locks[level] = 0;
8721 free_extent_buffer(path->nodes[level]);
8722 path->nodes[level] = NULL;
8730 * drop a subvolume tree.
8732 * this function traverses the tree freeing any blocks that only
8733 * referenced by the tree.
8735 * when a shared tree block is found. this function decreases its
8736 * reference count by one. if update_ref is true, this function
8737 * also make sure backrefs for the shared block and all lower level
8738 * blocks are properly updated.
8740 * If called with for_reloc == 0, may exit early with -EAGAIN
8742 int btrfs_drop_snapshot(struct btrfs_root *root,
8743 struct btrfs_block_rsv *block_rsv, int update_ref,
8746 struct btrfs_path *path;
8747 struct btrfs_trans_handle *trans;
8748 struct btrfs_root *tree_root = root->fs_info->tree_root;
8749 struct btrfs_root_item *root_item = &root->root_item;
8750 struct walk_control *wc;
8751 struct btrfs_key key;
8755 bool root_dropped = false;
8757 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8759 path = btrfs_alloc_path();
8765 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8767 btrfs_free_path(path);
8772 trans = btrfs_start_transaction(tree_root, 0);
8773 if (IS_ERR(trans)) {
8774 err = PTR_ERR(trans);
8779 trans->block_rsv = block_rsv;
8781 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8782 level = btrfs_header_level(root->node);
8783 path->nodes[level] = btrfs_lock_root_node(root);
8784 btrfs_set_lock_blocking(path->nodes[level]);
8785 path->slots[level] = 0;
8786 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8787 memset(&wc->update_progress, 0,
8788 sizeof(wc->update_progress));
8790 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8791 memcpy(&wc->update_progress, &key,
8792 sizeof(wc->update_progress));
8794 level = root_item->drop_level;
8796 path->lowest_level = level;
8797 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8798 path->lowest_level = 0;
8806 * unlock our path, this is safe because only this
8807 * function is allowed to delete this snapshot
8809 btrfs_unlock_up_safe(path, 0);
8811 level = btrfs_header_level(root->node);
8813 btrfs_tree_lock(path->nodes[level]);
8814 btrfs_set_lock_blocking(path->nodes[level]);
8815 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8817 ret = btrfs_lookup_extent_info(trans, root,
8818 path->nodes[level]->start,
8819 level, 1, &wc->refs[level],
8825 BUG_ON(wc->refs[level] == 0);
8827 if (level == root_item->drop_level)
8830 btrfs_tree_unlock(path->nodes[level]);
8831 path->locks[level] = 0;
8832 WARN_ON(wc->refs[level] != 1);
8838 wc->shared_level = -1;
8839 wc->stage = DROP_REFERENCE;
8840 wc->update_ref = update_ref;
8842 wc->for_reloc = for_reloc;
8843 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8847 ret = walk_down_tree(trans, root, path, wc);
8853 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8860 BUG_ON(wc->stage != DROP_REFERENCE);
8864 if (wc->stage == DROP_REFERENCE) {
8866 btrfs_node_key(path->nodes[level],
8867 &root_item->drop_progress,
8868 path->slots[level]);
8869 root_item->drop_level = level;
8872 BUG_ON(wc->level == 0);
8873 if (btrfs_should_end_transaction(trans, tree_root) ||
8874 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8875 ret = btrfs_update_root(trans, tree_root,
8879 btrfs_abort_transaction(trans, tree_root, ret);
8884 btrfs_end_transaction_throttle(trans, tree_root);
8885 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8886 pr_debug("BTRFS: drop snapshot early exit\n");
8891 trans = btrfs_start_transaction(tree_root, 0);
8892 if (IS_ERR(trans)) {
8893 err = PTR_ERR(trans);
8897 trans->block_rsv = block_rsv;
8900 btrfs_release_path(path);
8904 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8906 btrfs_abort_transaction(trans, tree_root, ret);
8910 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8911 ret = btrfs_find_root(tree_root, &root->root_key, path,
8914 btrfs_abort_transaction(trans, tree_root, ret);
8917 } else if (ret > 0) {
8918 /* if we fail to delete the orphan item this time
8919 * around, it'll get picked up the next time.
8921 * The most common failure here is just -ENOENT.
8923 btrfs_del_orphan_item(trans, tree_root,
8924 root->root_key.objectid);
8928 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8929 btrfs_add_dropped_root(trans, root);
8931 free_extent_buffer(root->node);
8932 free_extent_buffer(root->commit_root);
8933 btrfs_put_fs_root(root);
8935 root_dropped = true;
8937 btrfs_end_transaction_throttle(trans, tree_root);
8940 btrfs_free_path(path);
8943 * So if we need to stop dropping the snapshot for whatever reason we
8944 * need to make sure to add it back to the dead root list so that we
8945 * keep trying to do the work later. This also cleans up roots if we
8946 * don't have it in the radix (like when we recover after a power fail
8947 * or unmount) so we don't leak memory.
8949 if (!for_reloc && root_dropped == false)
8950 btrfs_add_dead_root(root);
8951 if (err && err != -EAGAIN)
8952 btrfs_std_error(root->fs_info, err, NULL);
8957 * drop subtree rooted at tree block 'node'.
8959 * NOTE: this function will unlock and release tree block 'node'
8960 * only used by relocation code
8962 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8963 struct btrfs_root *root,
8964 struct extent_buffer *node,
8965 struct extent_buffer *parent)
8967 struct btrfs_path *path;
8968 struct walk_control *wc;
8974 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8976 path = btrfs_alloc_path();
8980 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8982 btrfs_free_path(path);
8986 btrfs_assert_tree_locked(parent);
8987 parent_level = btrfs_header_level(parent);
8988 extent_buffer_get(parent);
8989 path->nodes[parent_level] = parent;
8990 path->slots[parent_level] = btrfs_header_nritems(parent);
8992 btrfs_assert_tree_locked(node);
8993 level = btrfs_header_level(node);
8994 path->nodes[level] = node;
8995 path->slots[level] = 0;
8996 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8998 wc->refs[parent_level] = 1;
8999 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9001 wc->shared_level = -1;
9002 wc->stage = DROP_REFERENCE;
9006 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9009 wret = walk_down_tree(trans, root, path, wc);
9015 wret = walk_up_tree(trans, root, path, wc, parent_level);
9023 btrfs_free_path(path);
9027 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9033 * if restripe for this chunk_type is on pick target profile and
9034 * return, otherwise do the usual balance
9036 stripped = get_restripe_target(root->fs_info, flags);
9038 return extended_to_chunk(stripped);
9040 num_devices = root->fs_info->fs_devices->rw_devices;
9042 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9043 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9044 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9046 if (num_devices == 1) {
9047 stripped |= BTRFS_BLOCK_GROUP_DUP;
9048 stripped = flags & ~stripped;
9050 /* turn raid0 into single device chunks */
9051 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9054 /* turn mirroring into duplication */
9055 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9056 BTRFS_BLOCK_GROUP_RAID10))
9057 return stripped | BTRFS_BLOCK_GROUP_DUP;
9059 /* they already had raid on here, just return */
9060 if (flags & stripped)
9063 stripped |= BTRFS_BLOCK_GROUP_DUP;
9064 stripped = flags & ~stripped;
9066 /* switch duplicated blocks with raid1 */
9067 if (flags & BTRFS_BLOCK_GROUP_DUP)
9068 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9070 /* this is drive concat, leave it alone */
9076 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9078 struct btrfs_space_info *sinfo = cache->space_info;
9080 u64 min_allocable_bytes;
9084 * We need some metadata space and system metadata space for
9085 * allocating chunks in some corner cases until we force to set
9086 * it to be readonly.
9089 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9091 min_allocable_bytes = 1 * 1024 * 1024;
9093 min_allocable_bytes = 0;
9095 spin_lock(&sinfo->lock);
9096 spin_lock(&cache->lock);
9104 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9105 cache->bytes_super - btrfs_block_group_used(&cache->item);
9107 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9108 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9109 min_allocable_bytes <= sinfo->total_bytes) {
9110 sinfo->bytes_readonly += num_bytes;
9112 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9116 spin_unlock(&cache->lock);
9117 spin_unlock(&sinfo->lock);
9121 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9122 struct btrfs_block_group_cache *cache)
9125 struct btrfs_trans_handle *trans;
9130 trans = btrfs_join_transaction(root);
9132 return PTR_ERR(trans);
9135 * we're not allowed to set block groups readonly after the dirty
9136 * block groups cache has started writing. If it already started,
9137 * back off and let this transaction commit
9139 mutex_lock(&root->fs_info->ro_block_group_mutex);
9140 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9141 u64 transid = trans->transid;
9143 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9144 btrfs_end_transaction(trans, root);
9146 ret = btrfs_wait_for_commit(root, transid);
9153 * if we are changing raid levels, try to allocate a corresponding
9154 * block group with the new raid level.
9156 alloc_flags = update_block_group_flags(root, cache->flags);
9157 if (alloc_flags != cache->flags) {
9158 ret = do_chunk_alloc(trans, root, alloc_flags,
9161 * ENOSPC is allowed here, we may have enough space
9162 * already allocated at the new raid level to
9171 ret = inc_block_group_ro(cache, 0);
9174 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9175 ret = do_chunk_alloc(trans, root, alloc_flags,
9179 ret = inc_block_group_ro(cache, 0);
9181 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9182 alloc_flags = update_block_group_flags(root, cache->flags);
9183 lock_chunks(root->fs_info->chunk_root);
9184 check_system_chunk(trans, root, alloc_flags);
9185 unlock_chunks(root->fs_info->chunk_root);
9187 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9189 btrfs_end_transaction(trans, root);
9193 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9194 struct btrfs_root *root, u64 type)
9196 u64 alloc_flags = get_alloc_profile(root, type);
9197 return do_chunk_alloc(trans, root, alloc_flags,
9202 * helper to account the unused space of all the readonly block group in the
9203 * space_info. takes mirrors into account.
9205 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9207 struct btrfs_block_group_cache *block_group;
9211 /* It's df, we don't care if it's racey */
9212 if (list_empty(&sinfo->ro_bgs))
9215 spin_lock(&sinfo->lock);
9216 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9217 spin_lock(&block_group->lock);
9219 if (!block_group->ro) {
9220 spin_unlock(&block_group->lock);
9224 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9225 BTRFS_BLOCK_GROUP_RAID10 |
9226 BTRFS_BLOCK_GROUP_DUP))
9231 free_bytes += (block_group->key.offset -
9232 btrfs_block_group_used(&block_group->item)) *
9235 spin_unlock(&block_group->lock);
9237 spin_unlock(&sinfo->lock);
9242 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9243 struct btrfs_block_group_cache *cache)
9245 struct btrfs_space_info *sinfo = cache->space_info;
9250 spin_lock(&sinfo->lock);
9251 spin_lock(&cache->lock);
9253 num_bytes = cache->key.offset - cache->reserved -
9254 cache->pinned - cache->bytes_super -
9255 btrfs_block_group_used(&cache->item);
9256 sinfo->bytes_readonly -= num_bytes;
9257 list_del_init(&cache->ro_list);
9259 spin_unlock(&cache->lock);
9260 spin_unlock(&sinfo->lock);
9264 * checks to see if its even possible to relocate this block group.
9266 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9267 * ok to go ahead and try.
9269 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9271 struct btrfs_block_group_cache *block_group;
9272 struct btrfs_space_info *space_info;
9273 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9274 struct btrfs_device *device;
9275 struct btrfs_trans_handle *trans;
9284 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9286 /* odd, couldn't find the block group, leave it alone */
9290 min_free = btrfs_block_group_used(&block_group->item);
9292 /* no bytes used, we're good */
9296 space_info = block_group->space_info;
9297 spin_lock(&space_info->lock);
9299 full = space_info->full;
9302 * if this is the last block group we have in this space, we can't
9303 * relocate it unless we're able to allocate a new chunk below.
9305 * Otherwise, we need to make sure we have room in the space to handle
9306 * all of the extents from this block group. If we can, we're good
9308 if ((space_info->total_bytes != block_group->key.offset) &&
9309 (space_info->bytes_used + space_info->bytes_reserved +
9310 space_info->bytes_pinned + space_info->bytes_readonly +
9311 min_free < space_info->total_bytes)) {
9312 spin_unlock(&space_info->lock);
9315 spin_unlock(&space_info->lock);
9318 * ok we don't have enough space, but maybe we have free space on our
9319 * devices to allocate new chunks for relocation, so loop through our
9320 * alloc devices and guess if we have enough space. if this block
9321 * group is going to be restriped, run checks against the target
9322 * profile instead of the current one.
9334 target = get_restripe_target(root->fs_info, block_group->flags);
9336 index = __get_raid_index(extended_to_chunk(target));
9339 * this is just a balance, so if we were marked as full
9340 * we know there is no space for a new chunk
9345 index = get_block_group_index(block_group);
9348 if (index == BTRFS_RAID_RAID10) {
9352 } else if (index == BTRFS_RAID_RAID1) {
9354 } else if (index == BTRFS_RAID_DUP) {
9357 } else if (index == BTRFS_RAID_RAID0) {
9358 dev_min = fs_devices->rw_devices;
9359 min_free = div64_u64(min_free, dev_min);
9362 /* We need to do this so that we can look at pending chunks */
9363 trans = btrfs_join_transaction(root);
9364 if (IS_ERR(trans)) {
9365 ret = PTR_ERR(trans);
9369 mutex_lock(&root->fs_info->chunk_mutex);
9370 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9374 * check to make sure we can actually find a chunk with enough
9375 * space to fit our block group in.
9377 if (device->total_bytes > device->bytes_used + min_free &&
9378 !device->is_tgtdev_for_dev_replace) {
9379 ret = find_free_dev_extent(trans, device, min_free,
9384 if (dev_nr >= dev_min)
9390 mutex_unlock(&root->fs_info->chunk_mutex);
9391 btrfs_end_transaction(trans, root);
9393 btrfs_put_block_group(block_group);
9397 static int find_first_block_group(struct btrfs_root *root,
9398 struct btrfs_path *path, struct btrfs_key *key)
9401 struct btrfs_key found_key;
9402 struct extent_buffer *leaf;
9405 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9410 slot = path->slots[0];
9411 leaf = path->nodes[0];
9412 if (slot >= btrfs_header_nritems(leaf)) {
9413 ret = btrfs_next_leaf(root, path);
9420 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9422 if (found_key.objectid >= key->objectid &&
9423 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9433 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9435 struct btrfs_block_group_cache *block_group;
9439 struct inode *inode;
9441 block_group = btrfs_lookup_first_block_group(info, last);
9442 while (block_group) {
9443 spin_lock(&block_group->lock);
9444 if (block_group->iref)
9446 spin_unlock(&block_group->lock);
9447 block_group = next_block_group(info->tree_root,
9457 inode = block_group->inode;
9458 block_group->iref = 0;
9459 block_group->inode = NULL;
9460 spin_unlock(&block_group->lock);
9462 last = block_group->key.objectid + block_group->key.offset;
9463 btrfs_put_block_group(block_group);
9467 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9469 struct btrfs_block_group_cache *block_group;
9470 struct btrfs_space_info *space_info;
9471 struct btrfs_caching_control *caching_ctl;
9474 down_write(&info->commit_root_sem);
9475 while (!list_empty(&info->caching_block_groups)) {
9476 caching_ctl = list_entry(info->caching_block_groups.next,
9477 struct btrfs_caching_control, list);
9478 list_del(&caching_ctl->list);
9479 put_caching_control(caching_ctl);
9481 up_write(&info->commit_root_sem);
9483 spin_lock(&info->unused_bgs_lock);
9484 while (!list_empty(&info->unused_bgs)) {
9485 block_group = list_first_entry(&info->unused_bgs,
9486 struct btrfs_block_group_cache,
9488 list_del_init(&block_group->bg_list);
9489 btrfs_put_block_group(block_group);
9491 spin_unlock(&info->unused_bgs_lock);
9493 spin_lock(&info->block_group_cache_lock);
9494 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9495 block_group = rb_entry(n, struct btrfs_block_group_cache,
9497 rb_erase(&block_group->cache_node,
9498 &info->block_group_cache_tree);
9499 RB_CLEAR_NODE(&block_group->cache_node);
9500 spin_unlock(&info->block_group_cache_lock);
9502 down_write(&block_group->space_info->groups_sem);
9503 list_del(&block_group->list);
9504 up_write(&block_group->space_info->groups_sem);
9506 if (block_group->cached == BTRFS_CACHE_STARTED)
9507 wait_block_group_cache_done(block_group);
9510 * We haven't cached this block group, which means we could
9511 * possibly have excluded extents on this block group.
9513 if (block_group->cached == BTRFS_CACHE_NO ||
9514 block_group->cached == BTRFS_CACHE_ERROR)
9515 free_excluded_extents(info->extent_root, block_group);
9517 btrfs_remove_free_space_cache(block_group);
9518 btrfs_put_block_group(block_group);
9520 spin_lock(&info->block_group_cache_lock);
9522 spin_unlock(&info->block_group_cache_lock);
9524 /* now that all the block groups are freed, go through and
9525 * free all the space_info structs. This is only called during
9526 * the final stages of unmount, and so we know nobody is
9527 * using them. We call synchronize_rcu() once before we start,
9528 * just to be on the safe side.
9532 release_global_block_rsv(info);
9534 while (!list_empty(&info->space_info)) {
9537 space_info = list_entry(info->space_info.next,
9538 struct btrfs_space_info,
9540 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9541 if (WARN_ON(space_info->bytes_pinned > 0 ||
9542 space_info->bytes_reserved > 0 ||
9543 space_info->bytes_may_use > 0)) {
9544 dump_space_info(space_info, 0, 0);
9547 list_del(&space_info->list);
9548 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9549 struct kobject *kobj;
9550 kobj = space_info->block_group_kobjs[i];
9551 space_info->block_group_kobjs[i] = NULL;
9557 kobject_del(&space_info->kobj);
9558 kobject_put(&space_info->kobj);
9563 static void __link_block_group(struct btrfs_space_info *space_info,
9564 struct btrfs_block_group_cache *cache)
9566 int index = get_block_group_index(cache);
9569 down_write(&space_info->groups_sem);
9570 if (list_empty(&space_info->block_groups[index]))
9572 list_add_tail(&cache->list, &space_info->block_groups[index]);
9573 up_write(&space_info->groups_sem);
9576 struct raid_kobject *rkobj;
9579 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9582 rkobj->raid_type = index;
9583 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9584 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9585 "%s", get_raid_name(index));
9587 kobject_put(&rkobj->kobj);
9590 space_info->block_group_kobjs[index] = &rkobj->kobj;
9595 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9598 static struct btrfs_block_group_cache *
9599 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9601 struct btrfs_block_group_cache *cache;
9603 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9607 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9609 if (!cache->free_space_ctl) {
9614 cache->key.objectid = start;
9615 cache->key.offset = size;
9616 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9618 cache->sectorsize = root->sectorsize;
9619 cache->fs_info = root->fs_info;
9620 cache->full_stripe_len = btrfs_full_stripe_len(root,
9621 &root->fs_info->mapping_tree,
9623 atomic_set(&cache->count, 1);
9624 spin_lock_init(&cache->lock);
9625 init_rwsem(&cache->data_rwsem);
9626 INIT_LIST_HEAD(&cache->list);
9627 INIT_LIST_HEAD(&cache->cluster_list);
9628 INIT_LIST_HEAD(&cache->bg_list);
9629 INIT_LIST_HEAD(&cache->ro_list);
9630 INIT_LIST_HEAD(&cache->dirty_list);
9631 INIT_LIST_HEAD(&cache->io_list);
9632 btrfs_init_free_space_ctl(cache);
9633 atomic_set(&cache->trimming, 0);
9638 int btrfs_read_block_groups(struct btrfs_root *root)
9640 struct btrfs_path *path;
9642 struct btrfs_block_group_cache *cache;
9643 struct btrfs_fs_info *info = root->fs_info;
9644 struct btrfs_space_info *space_info;
9645 struct btrfs_key key;
9646 struct btrfs_key found_key;
9647 struct extent_buffer *leaf;
9651 root = info->extent_root;
9654 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9655 path = btrfs_alloc_path();
9660 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9661 if (btrfs_test_opt(root, SPACE_CACHE) &&
9662 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9664 if (btrfs_test_opt(root, CLEAR_CACHE))
9668 ret = find_first_block_group(root, path, &key);
9674 leaf = path->nodes[0];
9675 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9677 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9686 * When we mount with old space cache, we need to
9687 * set BTRFS_DC_CLEAR and set dirty flag.
9689 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9690 * truncate the old free space cache inode and
9692 * b) Setting 'dirty flag' makes sure that we flush
9693 * the new space cache info onto disk.
9695 if (btrfs_test_opt(root, SPACE_CACHE))
9696 cache->disk_cache_state = BTRFS_DC_CLEAR;
9699 read_extent_buffer(leaf, &cache->item,
9700 btrfs_item_ptr_offset(leaf, path->slots[0]),
9701 sizeof(cache->item));
9702 cache->flags = btrfs_block_group_flags(&cache->item);
9704 key.objectid = found_key.objectid + found_key.offset;
9705 btrfs_release_path(path);
9708 * We need to exclude the super stripes now so that the space
9709 * info has super bytes accounted for, otherwise we'll think
9710 * we have more space than we actually do.
9712 ret = exclude_super_stripes(root, cache);
9715 * We may have excluded something, so call this just in
9718 free_excluded_extents(root, cache);
9719 btrfs_put_block_group(cache);
9724 * check for two cases, either we are full, and therefore
9725 * don't need to bother with the caching work since we won't
9726 * find any space, or we are empty, and we can just add all
9727 * the space in and be done with it. This saves us _alot_ of
9728 * time, particularly in the full case.
9730 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9731 cache->last_byte_to_unpin = (u64)-1;
9732 cache->cached = BTRFS_CACHE_FINISHED;
9733 free_excluded_extents(root, cache);
9734 } else if (btrfs_block_group_used(&cache->item) == 0) {
9735 cache->last_byte_to_unpin = (u64)-1;
9736 cache->cached = BTRFS_CACHE_FINISHED;
9737 add_new_free_space(cache, root->fs_info,
9739 found_key.objectid +
9741 free_excluded_extents(root, cache);
9744 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9746 btrfs_remove_free_space_cache(cache);
9747 btrfs_put_block_group(cache);
9751 ret = update_space_info(info, cache->flags, found_key.offset,
9752 btrfs_block_group_used(&cache->item),
9755 btrfs_remove_free_space_cache(cache);
9756 spin_lock(&info->block_group_cache_lock);
9757 rb_erase(&cache->cache_node,
9758 &info->block_group_cache_tree);
9759 RB_CLEAR_NODE(&cache->cache_node);
9760 spin_unlock(&info->block_group_cache_lock);
9761 btrfs_put_block_group(cache);
9765 cache->space_info = space_info;
9766 spin_lock(&cache->space_info->lock);
9767 cache->space_info->bytes_readonly += cache->bytes_super;
9768 spin_unlock(&cache->space_info->lock);
9770 __link_block_group(space_info, cache);
9772 set_avail_alloc_bits(root->fs_info, cache->flags);
9773 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9774 inc_block_group_ro(cache, 1);
9775 } else if (btrfs_block_group_used(&cache->item) == 0) {
9776 spin_lock(&info->unused_bgs_lock);
9777 /* Should always be true but just in case. */
9778 if (list_empty(&cache->bg_list)) {
9779 btrfs_get_block_group(cache);
9780 list_add_tail(&cache->bg_list,
9783 spin_unlock(&info->unused_bgs_lock);
9787 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9788 if (!(get_alloc_profile(root, space_info->flags) &
9789 (BTRFS_BLOCK_GROUP_RAID10 |
9790 BTRFS_BLOCK_GROUP_RAID1 |
9791 BTRFS_BLOCK_GROUP_RAID5 |
9792 BTRFS_BLOCK_GROUP_RAID6 |
9793 BTRFS_BLOCK_GROUP_DUP)))
9796 * avoid allocating from un-mirrored block group if there are
9797 * mirrored block groups.
9799 list_for_each_entry(cache,
9800 &space_info->block_groups[BTRFS_RAID_RAID0],
9802 inc_block_group_ro(cache, 1);
9803 list_for_each_entry(cache,
9804 &space_info->block_groups[BTRFS_RAID_SINGLE],
9806 inc_block_group_ro(cache, 1);
9809 init_global_block_rsv(info);
9812 btrfs_free_path(path);
9816 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9817 struct btrfs_root *root)
9819 struct btrfs_block_group_cache *block_group, *tmp;
9820 struct btrfs_root *extent_root = root->fs_info->extent_root;
9821 struct btrfs_block_group_item item;
9822 struct btrfs_key key;
9824 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9826 trans->can_flush_pending_bgs = false;
9827 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9831 spin_lock(&block_group->lock);
9832 memcpy(&item, &block_group->item, sizeof(item));
9833 memcpy(&key, &block_group->key, sizeof(key));
9834 spin_unlock(&block_group->lock);
9836 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9839 btrfs_abort_transaction(trans, extent_root, ret);
9840 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9841 key.objectid, key.offset);
9843 btrfs_abort_transaction(trans, extent_root, ret);
9845 list_del_init(&block_group->bg_list);
9847 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9850 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9851 struct btrfs_root *root, u64 bytes_used,
9852 u64 type, u64 chunk_objectid, u64 chunk_offset,
9856 struct btrfs_root *extent_root;
9857 struct btrfs_block_group_cache *cache;
9859 extent_root = root->fs_info->extent_root;
9861 btrfs_set_log_full_commit(root->fs_info, trans);
9863 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9867 btrfs_set_block_group_used(&cache->item, bytes_used);
9868 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9869 btrfs_set_block_group_flags(&cache->item, type);
9871 cache->flags = type;
9872 cache->last_byte_to_unpin = (u64)-1;
9873 cache->cached = BTRFS_CACHE_FINISHED;
9874 ret = exclude_super_stripes(root, cache);
9877 * We may have excluded something, so call this just in
9880 free_excluded_extents(root, cache);
9881 btrfs_put_block_group(cache);
9885 add_new_free_space(cache, root->fs_info, chunk_offset,
9886 chunk_offset + size);
9888 free_excluded_extents(root, cache);
9890 #ifdef CONFIG_BTRFS_DEBUG
9891 if (btrfs_should_fragment_free_space(root, cache)) {
9892 u64 new_bytes_used = size - bytes_used;
9894 bytes_used += new_bytes_used >> 1;
9895 fragment_free_space(root, cache);
9899 * Call to ensure the corresponding space_info object is created and
9900 * assigned to our block group, but don't update its counters just yet.
9901 * We want our bg to be added to the rbtree with its ->space_info set.
9903 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9904 &cache->space_info);
9906 btrfs_remove_free_space_cache(cache);
9907 btrfs_put_block_group(cache);
9911 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9913 btrfs_remove_free_space_cache(cache);
9914 btrfs_put_block_group(cache);
9919 * Now that our block group has its ->space_info set and is inserted in
9920 * the rbtree, update the space info's counters.
9922 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9923 &cache->space_info);
9925 btrfs_remove_free_space_cache(cache);
9926 spin_lock(&root->fs_info->block_group_cache_lock);
9927 rb_erase(&cache->cache_node,
9928 &root->fs_info->block_group_cache_tree);
9929 RB_CLEAR_NODE(&cache->cache_node);
9930 spin_unlock(&root->fs_info->block_group_cache_lock);
9931 btrfs_put_block_group(cache);
9934 update_global_block_rsv(root->fs_info);
9936 spin_lock(&cache->space_info->lock);
9937 cache->space_info->bytes_readonly += cache->bytes_super;
9938 spin_unlock(&cache->space_info->lock);
9940 __link_block_group(cache->space_info, cache);
9942 list_add_tail(&cache->bg_list, &trans->new_bgs);
9944 set_avail_alloc_bits(extent_root->fs_info, type);
9949 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9951 u64 extra_flags = chunk_to_extended(flags) &
9952 BTRFS_EXTENDED_PROFILE_MASK;
9954 write_seqlock(&fs_info->profiles_lock);
9955 if (flags & BTRFS_BLOCK_GROUP_DATA)
9956 fs_info->avail_data_alloc_bits &= ~extra_flags;
9957 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9958 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9959 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9960 fs_info->avail_system_alloc_bits &= ~extra_flags;
9961 write_sequnlock(&fs_info->profiles_lock);
9964 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9965 struct btrfs_root *root, u64 group_start,
9966 struct extent_map *em)
9968 struct btrfs_path *path;
9969 struct btrfs_block_group_cache *block_group;
9970 struct btrfs_free_cluster *cluster;
9971 struct btrfs_root *tree_root = root->fs_info->tree_root;
9972 struct btrfs_key key;
9973 struct inode *inode;
9974 struct kobject *kobj = NULL;
9978 struct btrfs_caching_control *caching_ctl = NULL;
9981 root = root->fs_info->extent_root;
9983 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9984 BUG_ON(!block_group);
9985 BUG_ON(!block_group->ro);
9988 * Free the reserved super bytes from this block group before
9991 free_excluded_extents(root, block_group);
9993 memcpy(&key, &block_group->key, sizeof(key));
9994 index = get_block_group_index(block_group);
9995 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9996 BTRFS_BLOCK_GROUP_RAID1 |
9997 BTRFS_BLOCK_GROUP_RAID10))
10002 /* make sure this block group isn't part of an allocation cluster */
10003 cluster = &root->fs_info->data_alloc_cluster;
10004 spin_lock(&cluster->refill_lock);
10005 btrfs_return_cluster_to_free_space(block_group, cluster);
10006 spin_unlock(&cluster->refill_lock);
10009 * make sure this block group isn't part of a metadata
10010 * allocation cluster
10012 cluster = &root->fs_info->meta_alloc_cluster;
10013 spin_lock(&cluster->refill_lock);
10014 btrfs_return_cluster_to_free_space(block_group, cluster);
10015 spin_unlock(&cluster->refill_lock);
10017 path = btrfs_alloc_path();
10024 * get the inode first so any iput calls done for the io_list
10025 * aren't the final iput (no unlinks allowed now)
10027 inode = lookup_free_space_inode(tree_root, block_group, path);
10029 mutex_lock(&trans->transaction->cache_write_mutex);
10031 * make sure our free spache cache IO is done before remove the
10034 spin_lock(&trans->transaction->dirty_bgs_lock);
10035 if (!list_empty(&block_group->io_list)) {
10036 list_del_init(&block_group->io_list);
10038 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10040 spin_unlock(&trans->transaction->dirty_bgs_lock);
10041 btrfs_wait_cache_io(root, trans, block_group,
10042 &block_group->io_ctl, path,
10043 block_group->key.objectid);
10044 btrfs_put_block_group(block_group);
10045 spin_lock(&trans->transaction->dirty_bgs_lock);
10048 if (!list_empty(&block_group->dirty_list)) {
10049 list_del_init(&block_group->dirty_list);
10050 btrfs_put_block_group(block_group);
10052 spin_unlock(&trans->transaction->dirty_bgs_lock);
10053 mutex_unlock(&trans->transaction->cache_write_mutex);
10055 if (!IS_ERR(inode)) {
10056 ret = btrfs_orphan_add(trans, inode);
10058 btrfs_add_delayed_iput(inode);
10061 clear_nlink(inode);
10062 /* One for the block groups ref */
10063 spin_lock(&block_group->lock);
10064 if (block_group->iref) {
10065 block_group->iref = 0;
10066 block_group->inode = NULL;
10067 spin_unlock(&block_group->lock);
10070 spin_unlock(&block_group->lock);
10072 /* One for our lookup ref */
10073 btrfs_add_delayed_iput(inode);
10076 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10077 key.offset = block_group->key.objectid;
10080 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10084 btrfs_release_path(path);
10086 ret = btrfs_del_item(trans, tree_root, path);
10089 btrfs_release_path(path);
10092 spin_lock(&root->fs_info->block_group_cache_lock);
10093 rb_erase(&block_group->cache_node,
10094 &root->fs_info->block_group_cache_tree);
10095 RB_CLEAR_NODE(&block_group->cache_node);
10097 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10098 root->fs_info->first_logical_byte = (u64)-1;
10099 spin_unlock(&root->fs_info->block_group_cache_lock);
10101 down_write(&block_group->space_info->groups_sem);
10103 * we must use list_del_init so people can check to see if they
10104 * are still on the list after taking the semaphore
10106 list_del_init(&block_group->list);
10107 if (list_empty(&block_group->space_info->block_groups[index])) {
10108 kobj = block_group->space_info->block_group_kobjs[index];
10109 block_group->space_info->block_group_kobjs[index] = NULL;
10110 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10112 up_write(&block_group->space_info->groups_sem);
10118 if (block_group->has_caching_ctl)
10119 caching_ctl = get_caching_control(block_group);
10120 if (block_group->cached == BTRFS_CACHE_STARTED)
10121 wait_block_group_cache_done(block_group);
10122 if (block_group->has_caching_ctl) {
10123 down_write(&root->fs_info->commit_root_sem);
10124 if (!caching_ctl) {
10125 struct btrfs_caching_control *ctl;
10127 list_for_each_entry(ctl,
10128 &root->fs_info->caching_block_groups, list)
10129 if (ctl->block_group == block_group) {
10131 atomic_inc(&caching_ctl->count);
10136 list_del_init(&caching_ctl->list);
10137 up_write(&root->fs_info->commit_root_sem);
10139 /* Once for the caching bgs list and once for us. */
10140 put_caching_control(caching_ctl);
10141 put_caching_control(caching_ctl);
10145 spin_lock(&trans->transaction->dirty_bgs_lock);
10146 if (!list_empty(&block_group->dirty_list)) {
10149 if (!list_empty(&block_group->io_list)) {
10152 spin_unlock(&trans->transaction->dirty_bgs_lock);
10153 btrfs_remove_free_space_cache(block_group);
10155 spin_lock(&block_group->space_info->lock);
10156 list_del_init(&block_group->ro_list);
10158 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10159 WARN_ON(block_group->space_info->total_bytes
10160 < block_group->key.offset);
10161 WARN_ON(block_group->space_info->bytes_readonly
10162 < block_group->key.offset);
10163 WARN_ON(block_group->space_info->disk_total
10164 < block_group->key.offset * factor);
10166 block_group->space_info->total_bytes -= block_group->key.offset;
10167 block_group->space_info->bytes_readonly -= block_group->key.offset;
10168 block_group->space_info->disk_total -= block_group->key.offset * factor;
10170 spin_unlock(&block_group->space_info->lock);
10172 memcpy(&key, &block_group->key, sizeof(key));
10175 if (!list_empty(&em->list)) {
10176 /* We're in the transaction->pending_chunks list. */
10177 free_extent_map(em);
10179 spin_lock(&block_group->lock);
10180 block_group->removed = 1;
10182 * At this point trimming can't start on this block group, because we
10183 * removed the block group from the tree fs_info->block_group_cache_tree
10184 * so no one can't find it anymore and even if someone already got this
10185 * block group before we removed it from the rbtree, they have already
10186 * incremented block_group->trimming - if they didn't, they won't find
10187 * any free space entries because we already removed them all when we
10188 * called btrfs_remove_free_space_cache().
10190 * And we must not remove the extent map from the fs_info->mapping_tree
10191 * to prevent the same logical address range and physical device space
10192 * ranges from being reused for a new block group. This is because our
10193 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10194 * completely transactionless, so while it is trimming a range the
10195 * currently running transaction might finish and a new one start,
10196 * allowing for new block groups to be created that can reuse the same
10197 * physical device locations unless we take this special care.
10199 * There may also be an implicit trim operation if the file system
10200 * is mounted with -odiscard. The same protections must remain
10201 * in place until the extents have been discarded completely when
10202 * the transaction commit has completed.
10204 remove_em = (atomic_read(&block_group->trimming) == 0);
10206 * Make sure a trimmer task always sees the em in the pinned_chunks list
10207 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10208 * before checking block_group->removed).
10212 * Our em might be in trans->transaction->pending_chunks which
10213 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10214 * and so is the fs_info->pinned_chunks list.
10216 * So at this point we must be holding the chunk_mutex to avoid
10217 * any races with chunk allocation (more specifically at
10218 * volumes.c:contains_pending_extent()), to ensure it always
10219 * sees the em, either in the pending_chunks list or in the
10220 * pinned_chunks list.
10222 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10224 spin_unlock(&block_group->lock);
10227 struct extent_map_tree *em_tree;
10229 em_tree = &root->fs_info->mapping_tree.map_tree;
10230 write_lock(&em_tree->lock);
10232 * The em might be in the pending_chunks list, so make sure the
10233 * chunk mutex is locked, since remove_extent_mapping() will
10234 * delete us from that list.
10236 remove_extent_mapping(em_tree, em);
10237 write_unlock(&em_tree->lock);
10238 /* once for the tree */
10239 free_extent_map(em);
10242 unlock_chunks(root);
10244 btrfs_put_block_group(block_group);
10245 btrfs_put_block_group(block_group);
10247 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10253 ret = btrfs_del_item(trans, root, path);
10255 btrfs_free_path(path);
10260 * Process the unused_bgs list and remove any that don't have any allocated
10261 * space inside of them.
10263 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10265 struct btrfs_block_group_cache *block_group;
10266 struct btrfs_space_info *space_info;
10267 struct btrfs_root *root = fs_info->extent_root;
10268 struct btrfs_trans_handle *trans;
10271 if (!fs_info->open)
10274 spin_lock(&fs_info->unused_bgs_lock);
10275 while (!list_empty(&fs_info->unused_bgs)) {
10279 block_group = list_first_entry(&fs_info->unused_bgs,
10280 struct btrfs_block_group_cache,
10282 space_info = block_group->space_info;
10283 list_del_init(&block_group->bg_list);
10284 if (ret || btrfs_mixed_space_info(space_info)) {
10285 btrfs_put_block_group(block_group);
10288 spin_unlock(&fs_info->unused_bgs_lock);
10290 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
10292 /* Don't want to race with allocators so take the groups_sem */
10293 down_write(&space_info->groups_sem);
10294 spin_lock(&block_group->lock);
10295 if (block_group->reserved ||
10296 btrfs_block_group_used(&block_group->item) ||
10299 * We want to bail if we made new allocations or have
10300 * outstanding allocations in this block group. We do
10301 * the ro check in case balance is currently acting on
10302 * this block group.
10304 spin_unlock(&block_group->lock);
10305 up_write(&space_info->groups_sem);
10308 spin_unlock(&block_group->lock);
10310 /* We don't want to force the issue, only flip if it's ok. */
10311 ret = inc_block_group_ro(block_group, 0);
10312 up_write(&space_info->groups_sem);
10319 * Want to do this before we do anything else so we can recover
10320 * properly if we fail to join the transaction.
10322 /* 1 for btrfs_orphan_reserve_metadata() */
10323 trans = btrfs_start_transaction(root, 1);
10324 if (IS_ERR(trans)) {
10325 btrfs_dec_block_group_ro(root, block_group);
10326 ret = PTR_ERR(trans);
10331 * We could have pending pinned extents for this block group,
10332 * just delete them, we don't care about them anymore.
10334 start = block_group->key.objectid;
10335 end = start + block_group->key.offset - 1;
10337 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10338 * btrfs_finish_extent_commit(). If we are at transaction N,
10339 * another task might be running finish_extent_commit() for the
10340 * previous transaction N - 1, and have seen a range belonging
10341 * to the block group in freed_extents[] before we were able to
10342 * clear the whole block group range from freed_extents[]. This
10343 * means that task can lookup for the block group after we
10344 * unpinned it from freed_extents[] and removed it, leading to
10345 * a BUG_ON() at btrfs_unpin_extent_range().
10347 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10348 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10349 EXTENT_DIRTY, GFP_NOFS);
10351 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10352 btrfs_dec_block_group_ro(root, block_group);
10355 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10356 EXTENT_DIRTY, GFP_NOFS);
10358 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10359 btrfs_dec_block_group_ro(root, block_group);
10362 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10364 /* Reset pinned so btrfs_put_block_group doesn't complain */
10365 spin_lock(&space_info->lock);
10366 spin_lock(&block_group->lock);
10368 space_info->bytes_pinned -= block_group->pinned;
10369 space_info->bytes_readonly += block_group->pinned;
10370 percpu_counter_add(&space_info->total_bytes_pinned,
10371 -block_group->pinned);
10372 block_group->pinned = 0;
10374 spin_unlock(&block_group->lock);
10375 spin_unlock(&space_info->lock);
10377 /* DISCARD can flip during remount */
10378 trimming = btrfs_test_opt(root, DISCARD);
10380 /* Implicit trim during transaction commit. */
10382 btrfs_get_block_group_trimming(block_group);
10385 * Btrfs_remove_chunk will abort the transaction if things go
10388 ret = btrfs_remove_chunk(trans, root,
10389 block_group->key.objectid);
10393 btrfs_put_block_group_trimming(block_group);
10398 * If we're not mounted with -odiscard, we can just forget
10399 * about this block group. Otherwise we'll need to wait
10400 * until transaction commit to do the actual discard.
10403 WARN_ON(!list_empty(&block_group->bg_list));
10404 spin_lock(&trans->transaction->deleted_bgs_lock);
10405 list_move(&block_group->bg_list,
10406 &trans->transaction->deleted_bgs);
10407 spin_unlock(&trans->transaction->deleted_bgs_lock);
10408 btrfs_get_block_group(block_group);
10411 btrfs_end_transaction(trans, root);
10413 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10414 btrfs_put_block_group(block_group);
10415 spin_lock(&fs_info->unused_bgs_lock);
10417 spin_unlock(&fs_info->unused_bgs_lock);
10420 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10422 struct btrfs_space_info *space_info;
10423 struct btrfs_super_block *disk_super;
10429 disk_super = fs_info->super_copy;
10430 if (!btrfs_super_root(disk_super))
10433 features = btrfs_super_incompat_flags(disk_super);
10434 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10437 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10438 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10443 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10444 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10446 flags = BTRFS_BLOCK_GROUP_METADATA;
10447 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10451 flags = BTRFS_BLOCK_GROUP_DATA;
10452 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10458 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10460 return unpin_extent_range(root, start, end, false);
10464 * It used to be that old block groups would be left around forever.
10465 * Iterating over them would be enough to trim unused space. Since we
10466 * now automatically remove them, we also need to iterate over unallocated
10469 * We don't want a transaction for this since the discard may take a
10470 * substantial amount of time. We don't require that a transaction be
10471 * running, but we do need to take a running transaction into account
10472 * to ensure that we're not discarding chunks that were released in
10473 * the current transaction.
10475 * Holding the chunks lock will prevent other threads from allocating
10476 * or releasing chunks, but it won't prevent a running transaction
10477 * from committing and releasing the memory that the pending chunks
10478 * list head uses. For that, we need to take a reference to the
10481 static int btrfs_trim_free_extents(struct btrfs_device *device,
10482 u64 minlen, u64 *trimmed)
10484 u64 start = 0, len = 0;
10489 /* Not writeable = nothing to do. */
10490 if (!device->writeable)
10493 /* No free space = nothing to do. */
10494 if (device->total_bytes <= device->bytes_used)
10500 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10501 struct btrfs_transaction *trans;
10504 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10508 down_read(&fs_info->commit_root_sem);
10510 spin_lock(&fs_info->trans_lock);
10511 trans = fs_info->running_transaction;
10513 atomic_inc(&trans->use_count);
10514 spin_unlock(&fs_info->trans_lock);
10516 ret = find_free_dev_extent_start(trans, device, minlen, start,
10519 btrfs_put_transaction(trans);
10522 up_read(&fs_info->commit_root_sem);
10523 mutex_unlock(&fs_info->chunk_mutex);
10524 if (ret == -ENOSPC)
10529 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10530 up_read(&fs_info->commit_root_sem);
10531 mutex_unlock(&fs_info->chunk_mutex);
10539 if (fatal_signal_pending(current)) {
10540 ret = -ERESTARTSYS;
10550 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10552 struct btrfs_fs_info *fs_info = root->fs_info;
10553 struct btrfs_block_group_cache *cache = NULL;
10554 struct btrfs_device *device;
10555 struct list_head *devices;
10560 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10564 * try to trim all FS space, our block group may start from non-zero.
10566 if (range->len == total_bytes)
10567 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10569 cache = btrfs_lookup_block_group(fs_info, range->start);
10572 if (cache->key.objectid >= (range->start + range->len)) {
10573 btrfs_put_block_group(cache);
10577 start = max(range->start, cache->key.objectid);
10578 end = min(range->start + range->len,
10579 cache->key.objectid + cache->key.offset);
10581 if (end - start >= range->minlen) {
10582 if (!block_group_cache_done(cache)) {
10583 ret = cache_block_group(cache, 0);
10585 btrfs_put_block_group(cache);
10588 ret = wait_block_group_cache_done(cache);
10590 btrfs_put_block_group(cache);
10594 ret = btrfs_trim_block_group(cache,
10600 trimmed += group_trimmed;
10602 btrfs_put_block_group(cache);
10607 cache = next_block_group(fs_info->tree_root, cache);
10610 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10611 devices = &root->fs_info->fs_devices->alloc_list;
10612 list_for_each_entry(device, devices, dev_alloc_list) {
10613 ret = btrfs_trim_free_extents(device, range->minlen,
10618 trimmed += group_trimmed;
10620 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10622 range->len = trimmed;
10627 * btrfs_{start,end}_write_no_snapshoting() are similar to
10628 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10629 * data into the page cache through nocow before the subvolume is snapshoted,
10630 * but flush the data into disk after the snapshot creation, or to prevent
10631 * operations while snapshoting is ongoing and that cause the snapshot to be
10632 * inconsistent (writes followed by expanding truncates for example).
10634 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10636 percpu_counter_dec(&root->subv_writers->counter);
10638 * Make sure counter is updated before we wake up waiters.
10641 if (waitqueue_active(&root->subv_writers->wait))
10642 wake_up(&root->subv_writers->wait);
10645 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10647 if (atomic_read(&root->will_be_snapshoted))
10650 percpu_counter_inc(&root->subv_writers->counter);
10652 * Make sure counter is updated before we check for snapshot creation.
10655 if (atomic_read(&root->will_be_snapshoted)) {
10656 btrfs_end_write_no_snapshoting(root);