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,
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 return (cache->flags & bits) == bits;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 atomic_inc(&cache->count);
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
161 if (block_group->key.objectid < cache->key.objectid) {
163 } else if (block_group->key.objectid > cache->key.objectid) {
166 spin_unlock(&info->block_group_cache_lock);
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
178 spin_unlock(&info->block_group_cache_lock);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
191 struct btrfs_block_group_cache *cache, *ret = NULL;
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
199 cache = rb_entry(n, struct btrfs_block_group_cache,
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
224 spin_unlock(&info->block_group_cache_lock);
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
282 if (logical[nr] > cache->key.objectid +
286 if (logical[nr] + stripe_len <= cache->key.objectid)
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
315 struct btrfs_caching_control *ctl;
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
329 static void put_caching_control(struct btrfs_caching_control *ctl)
331 if (atomic_dec_and_test(&ctl->count))
335 #ifdef CONFIG_BTRFS_DEBUG
336 static void fragment_free_space(struct btrfs_root *root,
337 struct btrfs_block_group_cache *block_group)
339 u64 start = block_group->key.objectid;
340 u64 len = block_group->key.offset;
341 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
342 root->nodesize : root->sectorsize;
343 u64 step = chunk << 1;
345 while (len > chunk) {
346 btrfs_remove_free_space(block_group, start, chunk);
357 * this is only called by cache_block_group, since we could have freed extents
358 * we need to check the pinned_extents for any extents that can't be used yet
359 * since their free space will be released as soon as the transaction commits.
361 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
362 struct btrfs_fs_info *info, u64 start, u64 end)
364 u64 extent_start, extent_end, size, total_added = 0;
367 while (start < end) {
368 ret = find_first_extent_bit(info->pinned_extents, start,
369 &extent_start, &extent_end,
370 EXTENT_DIRTY | EXTENT_UPTODATE,
375 if (extent_start <= start) {
376 start = extent_end + 1;
377 } else if (extent_start > start && extent_start < end) {
378 size = extent_start - start;
380 ret = btrfs_add_free_space(block_group, start,
382 BUG_ON(ret); /* -ENOMEM or logic error */
383 start = extent_end + 1;
392 ret = btrfs_add_free_space(block_group, start, size);
393 BUG_ON(ret); /* -ENOMEM or logic error */
399 static noinline void caching_thread(struct btrfs_work *work)
401 struct btrfs_block_group_cache *block_group;
402 struct btrfs_fs_info *fs_info;
403 struct btrfs_caching_control *caching_ctl;
404 struct btrfs_root *extent_root;
405 struct btrfs_path *path;
406 struct extent_buffer *leaf;
407 struct btrfs_key key;
414 caching_ctl = container_of(work, struct btrfs_caching_control, work);
415 block_group = caching_ctl->block_group;
416 fs_info = block_group->fs_info;
417 extent_root = fs_info->extent_root;
419 path = btrfs_alloc_path();
423 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
425 #ifdef CONFIG_BTRFS_DEBUG
427 * If we're fragmenting we don't want to make anybody think we can
428 * allocate from this block group until we've had a chance to fragment
431 if (btrfs_should_fragment_free_space(extent_root, block_group))
435 * We don't want to deadlock with somebody trying to allocate a new
436 * extent for the extent root while also trying to search the extent
437 * root to add free space. So we skip locking and search the commit
438 * root, since its read-only
440 path->skip_locking = 1;
441 path->search_commit_root = 1;
446 key.type = BTRFS_EXTENT_ITEM_KEY;
448 mutex_lock(&caching_ctl->mutex);
449 /* need to make sure the commit_root doesn't disappear */
450 down_read(&fs_info->commit_root_sem);
453 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
461 if (btrfs_fs_closing(fs_info) > 1) {
466 if (path->slots[0] < nritems) {
467 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
469 ret = find_next_key(path, 0, &key);
473 if (need_resched() ||
474 rwsem_is_contended(&fs_info->commit_root_sem)) {
476 caching_ctl->progress = last;
477 btrfs_release_path(path);
478 up_read(&fs_info->commit_root_sem);
479 mutex_unlock(&caching_ctl->mutex);
484 ret = btrfs_next_leaf(extent_root, path);
489 leaf = path->nodes[0];
490 nritems = btrfs_header_nritems(leaf);
494 if (key.objectid < last) {
497 key.type = BTRFS_EXTENT_ITEM_KEY;
500 caching_ctl->progress = last;
501 btrfs_release_path(path);
505 if (key.objectid < block_group->key.objectid) {
510 if (key.objectid >= block_group->key.objectid +
511 block_group->key.offset)
514 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
515 key.type == BTRFS_METADATA_ITEM_KEY) {
516 total_found += add_new_free_space(block_group,
519 if (key.type == BTRFS_METADATA_ITEM_KEY)
520 last = key.objectid +
521 fs_info->tree_root->nodesize;
523 last = key.objectid + key.offset;
525 if (total_found > (1024 * 1024 * 2)) {
528 wake_up(&caching_ctl->wait);
535 total_found += add_new_free_space(block_group, fs_info, last,
536 block_group->key.objectid +
537 block_group->key.offset);
538 spin_lock(&block_group->lock);
539 block_group->caching_ctl = NULL;
540 block_group->cached = BTRFS_CACHE_FINISHED;
541 spin_unlock(&block_group->lock);
543 #ifdef CONFIG_BTRFS_DEBUG
544 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
547 spin_lock(&block_group->space_info->lock);
548 spin_lock(&block_group->lock);
549 bytes_used = block_group->key.offset -
550 btrfs_block_group_used(&block_group->item);
551 block_group->space_info->bytes_used += bytes_used >> 1;
552 spin_unlock(&block_group->lock);
553 spin_unlock(&block_group->space_info->lock);
554 fragment_free_space(extent_root, block_group);
558 caching_ctl->progress = (u64)-1;
560 btrfs_free_path(path);
561 up_read(&fs_info->commit_root_sem);
563 free_excluded_extents(extent_root, block_group);
565 mutex_unlock(&caching_ctl->mutex);
568 spin_lock(&block_group->lock);
569 block_group->caching_ctl = NULL;
570 block_group->cached = BTRFS_CACHE_ERROR;
571 spin_unlock(&block_group->lock);
573 wake_up(&caching_ctl->wait);
575 put_caching_control(caching_ctl);
576 btrfs_put_block_group(block_group);
579 static int cache_block_group(struct btrfs_block_group_cache *cache,
583 struct btrfs_fs_info *fs_info = cache->fs_info;
584 struct btrfs_caching_control *caching_ctl;
587 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
591 INIT_LIST_HEAD(&caching_ctl->list);
592 mutex_init(&caching_ctl->mutex);
593 init_waitqueue_head(&caching_ctl->wait);
594 caching_ctl->block_group = cache;
595 caching_ctl->progress = cache->key.objectid;
596 atomic_set(&caching_ctl->count, 1);
597 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
598 caching_thread, NULL, NULL);
600 spin_lock(&cache->lock);
602 * This should be a rare occasion, but this could happen I think in the
603 * case where one thread starts to load the space cache info, and then
604 * some other thread starts a transaction commit which tries to do an
605 * allocation while the other thread is still loading the space cache
606 * info. The previous loop should have kept us from choosing this block
607 * group, but if we've moved to the state where we will wait on caching
608 * block groups we need to first check if we're doing a fast load here,
609 * so we can wait for it to finish, otherwise we could end up allocating
610 * from a block group who's cache gets evicted for one reason or
613 while (cache->cached == BTRFS_CACHE_FAST) {
614 struct btrfs_caching_control *ctl;
616 ctl = cache->caching_ctl;
617 atomic_inc(&ctl->count);
618 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
619 spin_unlock(&cache->lock);
623 finish_wait(&ctl->wait, &wait);
624 put_caching_control(ctl);
625 spin_lock(&cache->lock);
628 if (cache->cached != BTRFS_CACHE_NO) {
629 spin_unlock(&cache->lock);
633 WARN_ON(cache->caching_ctl);
634 cache->caching_ctl = caching_ctl;
635 cache->cached = BTRFS_CACHE_FAST;
636 spin_unlock(&cache->lock);
638 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
639 mutex_lock(&caching_ctl->mutex);
640 ret = load_free_space_cache(fs_info, cache);
642 spin_lock(&cache->lock);
644 cache->caching_ctl = NULL;
645 cache->cached = BTRFS_CACHE_FINISHED;
646 cache->last_byte_to_unpin = (u64)-1;
647 caching_ctl->progress = (u64)-1;
649 if (load_cache_only) {
650 cache->caching_ctl = NULL;
651 cache->cached = BTRFS_CACHE_NO;
653 cache->cached = BTRFS_CACHE_STARTED;
654 cache->has_caching_ctl = 1;
657 spin_unlock(&cache->lock);
658 #ifdef CONFIG_BTRFS_DEBUG
660 btrfs_should_fragment_free_space(fs_info->extent_root,
664 spin_lock(&cache->space_info->lock);
665 spin_lock(&cache->lock);
666 bytes_used = cache->key.offset -
667 btrfs_block_group_used(&cache->item);
668 cache->space_info->bytes_used += bytes_used >> 1;
669 spin_unlock(&cache->lock);
670 spin_unlock(&cache->space_info->lock);
671 fragment_free_space(fs_info->extent_root, cache);
674 mutex_unlock(&caching_ctl->mutex);
676 wake_up(&caching_ctl->wait);
678 put_caching_control(caching_ctl);
679 free_excluded_extents(fs_info->extent_root, cache);
684 * We are not going to do the fast caching, set cached to the
685 * appropriate value and wakeup any waiters.
687 spin_lock(&cache->lock);
688 if (load_cache_only) {
689 cache->caching_ctl = NULL;
690 cache->cached = BTRFS_CACHE_NO;
692 cache->cached = BTRFS_CACHE_STARTED;
693 cache->has_caching_ctl = 1;
695 spin_unlock(&cache->lock);
696 wake_up(&caching_ctl->wait);
699 if (load_cache_only) {
700 put_caching_control(caching_ctl);
704 down_write(&fs_info->commit_root_sem);
705 atomic_inc(&caching_ctl->count);
706 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
707 up_write(&fs_info->commit_root_sem);
709 btrfs_get_block_group(cache);
711 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
717 * return the block group that starts at or after bytenr
719 static struct btrfs_block_group_cache *
720 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
722 struct btrfs_block_group_cache *cache;
724 cache = block_group_cache_tree_search(info, bytenr, 0);
730 * return the block group that contains the given bytenr
732 struct btrfs_block_group_cache *btrfs_lookup_block_group(
733 struct btrfs_fs_info *info,
736 struct btrfs_block_group_cache *cache;
738 cache = block_group_cache_tree_search(info, bytenr, 1);
743 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
746 struct list_head *head = &info->space_info;
747 struct btrfs_space_info *found;
749 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
752 list_for_each_entry_rcu(found, head, list) {
753 if (found->flags & flags) {
763 * after adding space to the filesystem, we need to clear the full flags
764 * on all the space infos.
766 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
768 struct list_head *head = &info->space_info;
769 struct btrfs_space_info *found;
772 list_for_each_entry_rcu(found, head, list)
777 /* simple helper to search for an existing data extent at a given offset */
778 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
781 struct btrfs_key key;
782 struct btrfs_path *path;
784 path = btrfs_alloc_path();
788 key.objectid = start;
790 key.type = BTRFS_EXTENT_ITEM_KEY;
791 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
793 btrfs_free_path(path);
798 * helper function to lookup reference count and flags of a tree block.
800 * the head node for delayed ref is used to store the sum of all the
801 * reference count modifications queued up in the rbtree. the head
802 * node may also store the extent flags to set. This way you can check
803 * to see what the reference count and extent flags would be if all of
804 * the delayed refs are not processed.
806 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
807 struct btrfs_root *root, u64 bytenr,
808 u64 offset, int metadata, u64 *refs, u64 *flags)
810 struct btrfs_delayed_ref_head *head;
811 struct btrfs_delayed_ref_root *delayed_refs;
812 struct btrfs_path *path;
813 struct btrfs_extent_item *ei;
814 struct extent_buffer *leaf;
815 struct btrfs_key key;
822 * If we don't have skinny metadata, don't bother doing anything
825 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
826 offset = root->nodesize;
830 path = btrfs_alloc_path();
835 path->skip_locking = 1;
836 path->search_commit_root = 1;
840 key.objectid = bytenr;
843 key.type = BTRFS_METADATA_ITEM_KEY;
845 key.type = BTRFS_EXTENT_ITEM_KEY;
847 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
852 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
853 if (path->slots[0]) {
855 btrfs_item_key_to_cpu(path->nodes[0], &key,
857 if (key.objectid == bytenr &&
858 key.type == BTRFS_EXTENT_ITEM_KEY &&
859 key.offset == root->nodesize)
865 leaf = path->nodes[0];
866 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
867 if (item_size >= sizeof(*ei)) {
868 ei = btrfs_item_ptr(leaf, path->slots[0],
869 struct btrfs_extent_item);
870 num_refs = btrfs_extent_refs(leaf, ei);
871 extent_flags = btrfs_extent_flags(leaf, ei);
873 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
874 struct btrfs_extent_item_v0 *ei0;
875 BUG_ON(item_size != sizeof(*ei0));
876 ei0 = btrfs_item_ptr(leaf, path->slots[0],
877 struct btrfs_extent_item_v0);
878 num_refs = btrfs_extent_refs_v0(leaf, ei0);
879 /* FIXME: this isn't correct for data */
880 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
885 BUG_ON(num_refs == 0);
895 delayed_refs = &trans->transaction->delayed_refs;
896 spin_lock(&delayed_refs->lock);
897 head = btrfs_find_delayed_ref_head(trans, bytenr);
899 if (!mutex_trylock(&head->mutex)) {
900 atomic_inc(&head->node.refs);
901 spin_unlock(&delayed_refs->lock);
903 btrfs_release_path(path);
906 * Mutex was contended, block until it's released and try
909 mutex_lock(&head->mutex);
910 mutex_unlock(&head->mutex);
911 btrfs_put_delayed_ref(&head->node);
914 spin_lock(&head->lock);
915 if (head->extent_op && head->extent_op->update_flags)
916 extent_flags |= head->extent_op->flags_to_set;
918 BUG_ON(num_refs == 0);
920 num_refs += head->node.ref_mod;
921 spin_unlock(&head->lock);
922 mutex_unlock(&head->mutex);
924 spin_unlock(&delayed_refs->lock);
926 WARN_ON(num_refs == 0);
930 *flags = extent_flags;
932 btrfs_free_path(path);
937 * Back reference rules. Back refs have three main goals:
939 * 1) differentiate between all holders of references to an extent so that
940 * when a reference is dropped we can make sure it was a valid reference
941 * before freeing the extent.
943 * 2) Provide enough information to quickly find the holders of an extent
944 * if we notice a given block is corrupted or bad.
946 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
947 * maintenance. This is actually the same as #2, but with a slightly
948 * different use case.
950 * There are two kinds of back refs. The implicit back refs is optimized
951 * for pointers in non-shared tree blocks. For a given pointer in a block,
952 * back refs of this kind provide information about the block's owner tree
953 * and the pointer's key. These information allow us to find the block by
954 * b-tree searching. The full back refs is for pointers in tree blocks not
955 * referenced by their owner trees. The location of tree block is recorded
956 * in the back refs. Actually the full back refs is generic, and can be
957 * used in all cases the implicit back refs is used. The major shortcoming
958 * of the full back refs is its overhead. Every time a tree block gets
959 * COWed, we have to update back refs entry for all pointers in it.
961 * For a newly allocated tree block, we use implicit back refs for
962 * pointers in it. This means most tree related operations only involve
963 * implicit back refs. For a tree block created in old transaction, the
964 * only way to drop a reference to it is COW it. So we can detect the
965 * event that tree block loses its owner tree's reference and do the
966 * back refs conversion.
968 * When a tree block is COW'd through a tree, there are four cases:
970 * The reference count of the block is one and the tree is the block's
971 * owner tree. Nothing to do in this case.
973 * The reference count of the block is one and the tree is not the
974 * block's owner tree. In this case, full back refs is used for pointers
975 * in the block. Remove these full back refs, add implicit back refs for
976 * every pointers in the new block.
978 * The reference count of the block is greater than one and the tree is
979 * the block's owner tree. In this case, implicit back refs is used for
980 * pointers in the block. Add full back refs for every pointers in the
981 * block, increase lower level extents' reference counts. The original
982 * implicit back refs are entailed to the new block.
984 * The reference count of the block is greater than one and the tree is
985 * not the block's owner tree. Add implicit back refs for every pointer in
986 * the new block, increase lower level extents' reference count.
988 * Back Reference Key composing:
990 * The key objectid corresponds to the first byte in the extent,
991 * The key type is used to differentiate between types of back refs.
992 * There are different meanings of the key offset for different types
995 * File extents can be referenced by:
997 * - multiple snapshots, subvolumes, or different generations in one subvol
998 * - different files inside a single subvolume
999 * - different offsets inside a file (bookend extents in file.c)
1001 * The extent ref structure for the implicit back refs has fields for:
1003 * - Objectid of the subvolume root
1004 * - objectid of the file holding the reference
1005 * - original offset in the file
1006 * - how many bookend extents
1008 * The key offset for the implicit back refs is hash of the first
1011 * The extent ref structure for the full back refs has field for:
1013 * - number of pointers in the tree leaf
1015 * The key offset for the implicit back refs is the first byte of
1018 * When a file extent is allocated, The implicit back refs is used.
1019 * the fields are filled in:
1021 * (root_key.objectid, inode objectid, offset in file, 1)
1023 * When a file extent is removed file truncation, we find the
1024 * corresponding implicit back refs and check the following fields:
1026 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1028 * Btree extents can be referenced by:
1030 * - Different subvolumes
1032 * Both the implicit back refs and the full back refs for tree blocks
1033 * only consist of key. The key offset for the implicit back refs is
1034 * objectid of block's owner tree. The key offset for the full back refs
1035 * is the first byte of parent block.
1037 * When implicit back refs is used, information about the lowest key and
1038 * level of the tree block are required. These information are stored in
1039 * tree block info structure.
1042 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1043 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1044 struct btrfs_root *root,
1045 struct btrfs_path *path,
1046 u64 owner, u32 extra_size)
1048 struct btrfs_extent_item *item;
1049 struct btrfs_extent_item_v0 *ei0;
1050 struct btrfs_extent_ref_v0 *ref0;
1051 struct btrfs_tree_block_info *bi;
1052 struct extent_buffer *leaf;
1053 struct btrfs_key key;
1054 struct btrfs_key found_key;
1055 u32 new_size = sizeof(*item);
1059 leaf = path->nodes[0];
1060 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1062 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1063 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1064 struct btrfs_extent_item_v0);
1065 refs = btrfs_extent_refs_v0(leaf, ei0);
1067 if (owner == (u64)-1) {
1069 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1070 ret = btrfs_next_leaf(root, path);
1073 BUG_ON(ret > 0); /* Corruption */
1074 leaf = path->nodes[0];
1076 btrfs_item_key_to_cpu(leaf, &found_key,
1078 BUG_ON(key.objectid != found_key.objectid);
1079 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1083 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1084 struct btrfs_extent_ref_v0);
1085 owner = btrfs_ref_objectid_v0(leaf, ref0);
1089 btrfs_release_path(path);
1091 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1092 new_size += sizeof(*bi);
1094 new_size -= sizeof(*ei0);
1095 ret = btrfs_search_slot(trans, root, &key, path,
1096 new_size + extra_size, 1);
1099 BUG_ON(ret); /* Corruption */
1101 btrfs_extend_item(root, path, new_size);
1103 leaf = path->nodes[0];
1104 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1105 btrfs_set_extent_refs(leaf, item, refs);
1106 /* FIXME: get real generation */
1107 btrfs_set_extent_generation(leaf, item, 0);
1108 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1109 btrfs_set_extent_flags(leaf, item,
1110 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1111 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1112 bi = (struct btrfs_tree_block_info *)(item + 1);
1113 /* FIXME: get first key of the block */
1114 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1115 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1117 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1119 btrfs_mark_buffer_dirty(leaf);
1124 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1126 u32 high_crc = ~(u32)0;
1127 u32 low_crc = ~(u32)0;
1130 lenum = cpu_to_le64(root_objectid);
1131 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1132 lenum = cpu_to_le64(owner);
1133 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1134 lenum = cpu_to_le64(offset);
1135 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1137 return ((u64)high_crc << 31) ^ (u64)low_crc;
1140 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1141 struct btrfs_extent_data_ref *ref)
1143 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1144 btrfs_extent_data_ref_objectid(leaf, ref),
1145 btrfs_extent_data_ref_offset(leaf, ref));
1148 static int match_extent_data_ref(struct extent_buffer *leaf,
1149 struct btrfs_extent_data_ref *ref,
1150 u64 root_objectid, u64 owner, u64 offset)
1152 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1153 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1154 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1159 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 struct btrfs_path *path,
1162 u64 bytenr, u64 parent,
1164 u64 owner, u64 offset)
1166 struct btrfs_key key;
1167 struct btrfs_extent_data_ref *ref;
1168 struct extent_buffer *leaf;
1174 key.objectid = bytenr;
1176 key.type = BTRFS_SHARED_DATA_REF_KEY;
1177 key.offset = parent;
1179 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1180 key.offset = hash_extent_data_ref(root_objectid,
1185 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1194 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1195 key.type = BTRFS_EXTENT_REF_V0_KEY;
1196 btrfs_release_path(path);
1197 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1208 leaf = path->nodes[0];
1209 nritems = btrfs_header_nritems(leaf);
1211 if (path->slots[0] >= nritems) {
1212 ret = btrfs_next_leaf(root, path);
1218 leaf = path->nodes[0];
1219 nritems = btrfs_header_nritems(leaf);
1223 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1224 if (key.objectid != bytenr ||
1225 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1228 ref = btrfs_item_ptr(leaf, path->slots[0],
1229 struct btrfs_extent_data_ref);
1231 if (match_extent_data_ref(leaf, ref, root_objectid,
1234 btrfs_release_path(path);
1246 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1247 struct btrfs_root *root,
1248 struct btrfs_path *path,
1249 u64 bytenr, u64 parent,
1250 u64 root_objectid, u64 owner,
1251 u64 offset, int refs_to_add)
1253 struct btrfs_key key;
1254 struct extent_buffer *leaf;
1259 key.objectid = bytenr;
1261 key.type = BTRFS_SHARED_DATA_REF_KEY;
1262 key.offset = parent;
1263 size = sizeof(struct btrfs_shared_data_ref);
1265 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1266 key.offset = hash_extent_data_ref(root_objectid,
1268 size = sizeof(struct btrfs_extent_data_ref);
1271 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1272 if (ret && ret != -EEXIST)
1275 leaf = path->nodes[0];
1277 struct btrfs_shared_data_ref *ref;
1278 ref = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_shared_data_ref);
1281 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1283 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1284 num_refs += refs_to_add;
1285 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1288 struct btrfs_extent_data_ref *ref;
1289 while (ret == -EEXIST) {
1290 ref = btrfs_item_ptr(leaf, path->slots[0],
1291 struct btrfs_extent_data_ref);
1292 if (match_extent_data_ref(leaf, ref, root_objectid,
1295 btrfs_release_path(path);
1297 ret = btrfs_insert_empty_item(trans, root, path, &key,
1299 if (ret && ret != -EEXIST)
1302 leaf = path->nodes[0];
1304 ref = btrfs_item_ptr(leaf, path->slots[0],
1305 struct btrfs_extent_data_ref);
1307 btrfs_set_extent_data_ref_root(leaf, ref,
1309 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1310 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1311 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1313 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1314 num_refs += refs_to_add;
1315 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1318 btrfs_mark_buffer_dirty(leaf);
1321 btrfs_release_path(path);
1325 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1326 struct btrfs_root *root,
1327 struct btrfs_path *path,
1328 int refs_to_drop, int *last_ref)
1330 struct btrfs_key key;
1331 struct btrfs_extent_data_ref *ref1 = NULL;
1332 struct btrfs_shared_data_ref *ref2 = NULL;
1333 struct extent_buffer *leaf;
1337 leaf = path->nodes[0];
1338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1340 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_extent_data_ref);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_shared_data_ref);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350 struct btrfs_extent_ref_v0 *ref0;
1351 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_ref_v0);
1353 num_refs = btrfs_ref_count_v0(leaf, ref0);
1359 BUG_ON(num_refs < refs_to_drop);
1360 num_refs -= refs_to_drop;
1362 if (num_refs == 0) {
1363 ret = btrfs_del_item(trans, root, path);
1366 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1367 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1368 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1369 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1370 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1372 struct btrfs_extent_ref_v0 *ref0;
1373 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1374 struct btrfs_extent_ref_v0);
1375 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1378 btrfs_mark_buffer_dirty(leaf);
1383 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1384 struct btrfs_extent_inline_ref *iref)
1386 struct btrfs_key key;
1387 struct extent_buffer *leaf;
1388 struct btrfs_extent_data_ref *ref1;
1389 struct btrfs_shared_data_ref *ref2;
1392 leaf = path->nodes[0];
1393 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1395 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1396 BTRFS_EXTENT_DATA_REF_KEY) {
1397 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1398 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1400 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1401 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1403 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1404 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1405 struct btrfs_extent_data_ref);
1406 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1407 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1408 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1409 struct btrfs_shared_data_ref);
1410 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1411 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1412 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1413 struct btrfs_extent_ref_v0 *ref0;
1414 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1415 struct btrfs_extent_ref_v0);
1416 num_refs = btrfs_ref_count_v0(leaf, ref0);
1424 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1425 struct btrfs_root *root,
1426 struct btrfs_path *path,
1427 u64 bytenr, u64 parent,
1430 struct btrfs_key key;
1433 key.objectid = bytenr;
1435 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1436 key.offset = parent;
1438 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1439 key.offset = root_objectid;
1442 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1445 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1446 if (ret == -ENOENT && parent) {
1447 btrfs_release_path(path);
1448 key.type = BTRFS_EXTENT_REF_V0_KEY;
1449 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1457 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1458 struct btrfs_root *root,
1459 struct btrfs_path *path,
1460 u64 bytenr, u64 parent,
1463 struct btrfs_key key;
1466 key.objectid = bytenr;
1468 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1469 key.offset = parent;
1471 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1472 key.offset = root_objectid;
1475 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1476 btrfs_release_path(path);
1480 static inline int extent_ref_type(u64 parent, u64 owner)
1483 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1485 type = BTRFS_SHARED_BLOCK_REF_KEY;
1487 type = BTRFS_TREE_BLOCK_REF_KEY;
1490 type = BTRFS_SHARED_DATA_REF_KEY;
1492 type = BTRFS_EXTENT_DATA_REF_KEY;
1497 static int find_next_key(struct btrfs_path *path, int level,
1498 struct btrfs_key *key)
1501 for (; level < BTRFS_MAX_LEVEL; level++) {
1502 if (!path->nodes[level])
1504 if (path->slots[level] + 1 >=
1505 btrfs_header_nritems(path->nodes[level]))
1508 btrfs_item_key_to_cpu(path->nodes[level], key,
1509 path->slots[level] + 1);
1511 btrfs_node_key_to_cpu(path->nodes[level], key,
1512 path->slots[level] + 1);
1519 * look for inline back ref. if back ref is found, *ref_ret is set
1520 * to the address of inline back ref, and 0 is returned.
1522 * if back ref isn't found, *ref_ret is set to the address where it
1523 * should be inserted, and -ENOENT is returned.
1525 * if insert is true and there are too many inline back refs, the path
1526 * points to the extent item, and -EAGAIN is returned.
1528 * NOTE: inline back refs are ordered in the same way that back ref
1529 * items in the tree are ordered.
1531 static noinline_for_stack
1532 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1533 struct btrfs_root *root,
1534 struct btrfs_path *path,
1535 struct btrfs_extent_inline_ref **ref_ret,
1536 u64 bytenr, u64 num_bytes,
1537 u64 parent, u64 root_objectid,
1538 u64 owner, u64 offset, int insert)
1540 struct btrfs_key key;
1541 struct extent_buffer *leaf;
1542 struct btrfs_extent_item *ei;
1543 struct btrfs_extent_inline_ref *iref;
1553 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1556 key.objectid = bytenr;
1557 key.type = BTRFS_EXTENT_ITEM_KEY;
1558 key.offset = num_bytes;
1560 want = extent_ref_type(parent, owner);
1562 extra_size = btrfs_extent_inline_ref_size(want);
1563 path->keep_locks = 1;
1568 * Owner is our parent level, so we can just add one to get the level
1569 * for the block we are interested in.
1571 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1572 key.type = BTRFS_METADATA_ITEM_KEY;
1577 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1584 * We may be a newly converted file system which still has the old fat
1585 * extent entries for metadata, so try and see if we have one of those.
1587 if (ret > 0 && skinny_metadata) {
1588 skinny_metadata = false;
1589 if (path->slots[0]) {
1591 btrfs_item_key_to_cpu(path->nodes[0], &key,
1593 if (key.objectid == bytenr &&
1594 key.type == BTRFS_EXTENT_ITEM_KEY &&
1595 key.offset == num_bytes)
1599 key.objectid = bytenr;
1600 key.type = BTRFS_EXTENT_ITEM_KEY;
1601 key.offset = num_bytes;
1602 btrfs_release_path(path);
1607 if (ret && !insert) {
1610 } else if (WARN_ON(ret)) {
1615 leaf = path->nodes[0];
1616 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1617 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1618 if (item_size < sizeof(*ei)) {
1623 ret = convert_extent_item_v0(trans, root, path, owner,
1629 leaf = path->nodes[0];
1630 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1633 BUG_ON(item_size < sizeof(*ei));
1635 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1636 flags = btrfs_extent_flags(leaf, ei);
1638 ptr = (unsigned long)(ei + 1);
1639 end = (unsigned long)ei + item_size;
1641 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1642 ptr += sizeof(struct btrfs_tree_block_info);
1652 iref = (struct btrfs_extent_inline_ref *)ptr;
1653 type = btrfs_extent_inline_ref_type(leaf, iref);
1657 ptr += btrfs_extent_inline_ref_size(type);
1661 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1662 struct btrfs_extent_data_ref *dref;
1663 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1664 if (match_extent_data_ref(leaf, dref, root_objectid,
1669 if (hash_extent_data_ref_item(leaf, dref) <
1670 hash_extent_data_ref(root_objectid, owner, offset))
1674 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1676 if (parent == ref_offset) {
1680 if (ref_offset < parent)
1683 if (root_objectid == ref_offset) {
1687 if (ref_offset < root_objectid)
1691 ptr += btrfs_extent_inline_ref_size(type);
1693 if (err == -ENOENT && insert) {
1694 if (item_size + extra_size >=
1695 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1700 * To add new inline back ref, we have to make sure
1701 * there is no corresponding back ref item.
1702 * For simplicity, we just do not add new inline back
1703 * ref if there is any kind of item for this block
1705 if (find_next_key(path, 0, &key) == 0 &&
1706 key.objectid == bytenr &&
1707 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1712 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1715 path->keep_locks = 0;
1716 btrfs_unlock_up_safe(path, 1);
1722 * helper to add new inline back ref
1724 static noinline_for_stack
1725 void setup_inline_extent_backref(struct btrfs_root *root,
1726 struct btrfs_path *path,
1727 struct btrfs_extent_inline_ref *iref,
1728 u64 parent, u64 root_objectid,
1729 u64 owner, u64 offset, int refs_to_add,
1730 struct btrfs_delayed_extent_op *extent_op)
1732 struct extent_buffer *leaf;
1733 struct btrfs_extent_item *ei;
1736 unsigned long item_offset;
1741 leaf = path->nodes[0];
1742 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1743 item_offset = (unsigned long)iref - (unsigned long)ei;
1745 type = extent_ref_type(parent, owner);
1746 size = btrfs_extent_inline_ref_size(type);
1748 btrfs_extend_item(root, path, size);
1750 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1751 refs = btrfs_extent_refs(leaf, ei);
1752 refs += refs_to_add;
1753 btrfs_set_extent_refs(leaf, ei, refs);
1755 __run_delayed_extent_op(extent_op, leaf, ei);
1757 ptr = (unsigned long)ei + item_offset;
1758 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1759 if (ptr < end - size)
1760 memmove_extent_buffer(leaf, ptr + size, ptr,
1763 iref = (struct btrfs_extent_inline_ref *)ptr;
1764 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1765 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1766 struct btrfs_extent_data_ref *dref;
1767 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1768 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1769 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1770 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1771 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1772 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1773 struct btrfs_shared_data_ref *sref;
1774 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1775 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1776 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1777 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1778 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1780 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1782 btrfs_mark_buffer_dirty(leaf);
1785 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1786 struct btrfs_root *root,
1787 struct btrfs_path *path,
1788 struct btrfs_extent_inline_ref **ref_ret,
1789 u64 bytenr, u64 num_bytes, u64 parent,
1790 u64 root_objectid, u64 owner, u64 offset)
1794 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1795 bytenr, num_bytes, parent,
1796 root_objectid, owner, offset, 0);
1800 btrfs_release_path(path);
1803 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1804 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1807 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1808 root_objectid, owner, offset);
1814 * helper to update/remove inline back ref
1816 static noinline_for_stack
1817 void update_inline_extent_backref(struct btrfs_root *root,
1818 struct btrfs_path *path,
1819 struct btrfs_extent_inline_ref *iref,
1821 struct btrfs_delayed_extent_op *extent_op,
1824 struct extent_buffer *leaf;
1825 struct btrfs_extent_item *ei;
1826 struct btrfs_extent_data_ref *dref = NULL;
1827 struct btrfs_shared_data_ref *sref = NULL;
1835 leaf = path->nodes[0];
1836 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1837 refs = btrfs_extent_refs(leaf, ei);
1838 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1839 refs += refs_to_mod;
1840 btrfs_set_extent_refs(leaf, ei, refs);
1842 __run_delayed_extent_op(extent_op, leaf, ei);
1844 type = btrfs_extent_inline_ref_type(leaf, iref);
1846 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1847 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1848 refs = btrfs_extent_data_ref_count(leaf, dref);
1849 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1850 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1851 refs = btrfs_shared_data_ref_count(leaf, sref);
1854 BUG_ON(refs_to_mod != -1);
1857 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1858 refs += refs_to_mod;
1861 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1862 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1864 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1867 size = btrfs_extent_inline_ref_size(type);
1868 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1869 ptr = (unsigned long)iref;
1870 end = (unsigned long)ei + item_size;
1871 if (ptr + size < end)
1872 memmove_extent_buffer(leaf, ptr, ptr + size,
1875 btrfs_truncate_item(root, path, item_size, 1);
1877 btrfs_mark_buffer_dirty(leaf);
1880 static noinline_for_stack
1881 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1882 struct btrfs_root *root,
1883 struct btrfs_path *path,
1884 u64 bytenr, u64 num_bytes, u64 parent,
1885 u64 root_objectid, u64 owner,
1886 u64 offset, int refs_to_add,
1887 struct btrfs_delayed_extent_op *extent_op)
1889 struct btrfs_extent_inline_ref *iref;
1892 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1893 bytenr, num_bytes, parent,
1894 root_objectid, owner, offset, 1);
1896 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1897 update_inline_extent_backref(root, path, iref,
1898 refs_to_add, extent_op, NULL);
1899 } else if (ret == -ENOENT) {
1900 setup_inline_extent_backref(root, path, iref, parent,
1901 root_objectid, owner, offset,
1902 refs_to_add, extent_op);
1908 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1909 struct btrfs_root *root,
1910 struct btrfs_path *path,
1911 u64 bytenr, u64 parent, u64 root_objectid,
1912 u64 owner, u64 offset, int refs_to_add)
1915 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1916 BUG_ON(refs_to_add != 1);
1917 ret = insert_tree_block_ref(trans, root, path, bytenr,
1918 parent, root_objectid);
1920 ret = insert_extent_data_ref(trans, root, path, bytenr,
1921 parent, root_objectid,
1922 owner, offset, refs_to_add);
1927 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1928 struct btrfs_root *root,
1929 struct btrfs_path *path,
1930 struct btrfs_extent_inline_ref *iref,
1931 int refs_to_drop, int is_data, int *last_ref)
1935 BUG_ON(!is_data && refs_to_drop != 1);
1937 update_inline_extent_backref(root, path, iref,
1938 -refs_to_drop, NULL, last_ref);
1939 } else if (is_data) {
1940 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1944 ret = btrfs_del_item(trans, root, path);
1949 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1950 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1951 u64 *discarded_bytes)
1954 u64 bytes_left, end;
1955 u64 aligned_start = ALIGN(start, 1 << 9);
1957 if (WARN_ON(start != aligned_start)) {
1958 len -= aligned_start - start;
1959 len = round_down(len, 1 << 9);
1960 start = aligned_start;
1963 *discarded_bytes = 0;
1971 /* Skip any superblocks on this device. */
1972 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1973 u64 sb_start = btrfs_sb_offset(j);
1974 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1975 u64 size = sb_start - start;
1977 if (!in_range(sb_start, start, bytes_left) &&
1978 !in_range(sb_end, start, bytes_left) &&
1979 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1983 * Superblock spans beginning of range. Adjust start and
1986 if (sb_start <= start) {
1987 start += sb_end - start;
1992 bytes_left = end - start;
1997 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2000 *discarded_bytes += size;
2001 else if (ret != -EOPNOTSUPP)
2010 bytes_left = end - start;
2014 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2017 *discarded_bytes += bytes_left;
2022 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2023 u64 num_bytes, u64 *actual_bytes)
2026 u64 discarded_bytes = 0;
2027 struct btrfs_bio *bbio = NULL;
2030 /* Tell the block device(s) that the sectors can be discarded */
2031 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2032 bytenr, &num_bytes, &bbio, 0);
2033 /* Error condition is -ENOMEM */
2035 struct btrfs_bio_stripe *stripe = bbio->stripes;
2039 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2041 if (!stripe->dev->can_discard)
2044 ret = btrfs_issue_discard(stripe->dev->bdev,
2049 discarded_bytes += bytes;
2050 else if (ret != -EOPNOTSUPP)
2051 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2054 * Just in case we get back EOPNOTSUPP for some reason,
2055 * just ignore the return value so we don't screw up
2056 * people calling discard_extent.
2060 btrfs_put_bbio(bbio);
2064 *actual_bytes = discarded_bytes;
2067 if (ret == -EOPNOTSUPP)
2072 /* Can return -ENOMEM */
2073 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2074 struct btrfs_root *root,
2075 u64 bytenr, u64 num_bytes, u64 parent,
2076 u64 root_objectid, u64 owner, u64 offset,
2080 struct btrfs_fs_info *fs_info = root->fs_info;
2082 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2083 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2085 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2086 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2088 parent, root_objectid, (int)owner,
2089 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2091 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2093 parent, root_objectid, owner, offset,
2094 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2099 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct btrfs_delayed_ref_node *node,
2102 u64 parent, u64 root_objectid,
2103 u64 owner, u64 offset, int refs_to_add,
2104 struct btrfs_delayed_extent_op *extent_op)
2106 struct btrfs_fs_info *fs_info = root->fs_info;
2107 struct btrfs_path *path;
2108 struct extent_buffer *leaf;
2109 struct btrfs_extent_item *item;
2110 struct btrfs_key key;
2111 u64 bytenr = node->bytenr;
2112 u64 num_bytes = node->num_bytes;
2115 int no_quota = node->no_quota;
2117 path = btrfs_alloc_path();
2121 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2125 path->leave_spinning = 1;
2126 /* this will setup the path even if it fails to insert the back ref */
2127 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2128 bytenr, num_bytes, parent,
2129 root_objectid, owner, offset,
2130 refs_to_add, extent_op);
2131 if ((ret < 0 && ret != -EAGAIN) || !ret)
2135 * Ok we had -EAGAIN which means we didn't have space to insert and
2136 * inline extent ref, so just update the reference count and add a
2139 leaf = path->nodes[0];
2140 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2141 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2142 refs = btrfs_extent_refs(leaf, item);
2143 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2145 __run_delayed_extent_op(extent_op, leaf, item);
2147 btrfs_mark_buffer_dirty(leaf);
2148 btrfs_release_path(path);
2151 path->leave_spinning = 1;
2152 /* now insert the actual backref */
2153 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2154 path, bytenr, parent, root_objectid,
2155 owner, offset, refs_to_add);
2157 btrfs_abort_transaction(trans, root, ret);
2159 btrfs_free_path(path);
2163 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2164 struct btrfs_root *root,
2165 struct btrfs_delayed_ref_node *node,
2166 struct btrfs_delayed_extent_op *extent_op,
2167 int insert_reserved)
2170 struct btrfs_delayed_data_ref *ref;
2171 struct btrfs_key ins;
2176 ins.objectid = node->bytenr;
2177 ins.offset = node->num_bytes;
2178 ins.type = BTRFS_EXTENT_ITEM_KEY;
2180 ref = btrfs_delayed_node_to_data_ref(node);
2181 trace_run_delayed_data_ref(node, ref, node->action);
2183 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2184 parent = ref->parent;
2185 ref_root = ref->root;
2187 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2189 flags |= extent_op->flags_to_set;
2190 ret = alloc_reserved_file_extent(trans, root,
2191 parent, ref_root, flags,
2192 ref->objectid, ref->offset,
2193 &ins, node->ref_mod);
2194 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2195 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2196 ref_root, ref->objectid,
2197 ref->offset, node->ref_mod,
2199 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2200 ret = __btrfs_free_extent(trans, root, node, parent,
2201 ref_root, ref->objectid,
2202 ref->offset, node->ref_mod,
2210 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2211 struct extent_buffer *leaf,
2212 struct btrfs_extent_item *ei)
2214 u64 flags = btrfs_extent_flags(leaf, ei);
2215 if (extent_op->update_flags) {
2216 flags |= extent_op->flags_to_set;
2217 btrfs_set_extent_flags(leaf, ei, flags);
2220 if (extent_op->update_key) {
2221 struct btrfs_tree_block_info *bi;
2222 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2223 bi = (struct btrfs_tree_block_info *)(ei + 1);
2224 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2228 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2229 struct btrfs_root *root,
2230 struct btrfs_delayed_ref_node *node,
2231 struct btrfs_delayed_extent_op *extent_op)
2233 struct btrfs_key key;
2234 struct btrfs_path *path;
2235 struct btrfs_extent_item *ei;
2236 struct extent_buffer *leaf;
2240 int metadata = !extent_op->is_data;
2245 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2248 path = btrfs_alloc_path();
2252 key.objectid = node->bytenr;
2255 key.type = BTRFS_METADATA_ITEM_KEY;
2256 key.offset = extent_op->level;
2258 key.type = BTRFS_EXTENT_ITEM_KEY;
2259 key.offset = node->num_bytes;
2264 path->leave_spinning = 1;
2265 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2273 if (path->slots[0] > 0) {
2275 btrfs_item_key_to_cpu(path->nodes[0], &key,
2277 if (key.objectid == node->bytenr &&
2278 key.type == BTRFS_EXTENT_ITEM_KEY &&
2279 key.offset == node->num_bytes)
2283 btrfs_release_path(path);
2286 key.objectid = node->bytenr;
2287 key.offset = node->num_bytes;
2288 key.type = BTRFS_EXTENT_ITEM_KEY;
2297 leaf = path->nodes[0];
2298 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2299 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2300 if (item_size < sizeof(*ei)) {
2301 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2307 leaf = path->nodes[0];
2308 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2311 BUG_ON(item_size < sizeof(*ei));
2312 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2313 __run_delayed_extent_op(extent_op, leaf, ei);
2315 btrfs_mark_buffer_dirty(leaf);
2317 btrfs_free_path(path);
2321 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2322 struct btrfs_root *root,
2323 struct btrfs_delayed_ref_node *node,
2324 struct btrfs_delayed_extent_op *extent_op,
2325 int insert_reserved)
2328 struct btrfs_delayed_tree_ref *ref;
2329 struct btrfs_key ins;
2332 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2335 ref = btrfs_delayed_node_to_tree_ref(node);
2336 trace_run_delayed_tree_ref(node, ref, node->action);
2338 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2339 parent = ref->parent;
2340 ref_root = ref->root;
2342 ins.objectid = node->bytenr;
2343 if (skinny_metadata) {
2344 ins.offset = ref->level;
2345 ins.type = BTRFS_METADATA_ITEM_KEY;
2347 ins.offset = node->num_bytes;
2348 ins.type = BTRFS_EXTENT_ITEM_KEY;
2351 BUG_ON(node->ref_mod != 1);
2352 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2353 BUG_ON(!extent_op || !extent_op->update_flags);
2354 ret = alloc_reserved_tree_block(trans, root,
2356 extent_op->flags_to_set,
2360 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2361 ret = __btrfs_inc_extent_ref(trans, root, node,
2365 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2366 ret = __btrfs_free_extent(trans, root, node,
2368 ref->level, 0, 1, extent_op);
2375 /* helper function to actually process a single delayed ref entry */
2376 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2377 struct btrfs_root *root,
2378 struct btrfs_delayed_ref_node *node,
2379 struct btrfs_delayed_extent_op *extent_op,
2380 int insert_reserved)
2384 if (trans->aborted) {
2385 if (insert_reserved)
2386 btrfs_pin_extent(root, node->bytenr,
2387 node->num_bytes, 1);
2391 if (btrfs_delayed_ref_is_head(node)) {
2392 struct btrfs_delayed_ref_head *head;
2394 * we've hit the end of the chain and we were supposed
2395 * to insert this extent into the tree. But, it got
2396 * deleted before we ever needed to insert it, so all
2397 * we have to do is clean up the accounting
2400 head = btrfs_delayed_node_to_head(node);
2401 trace_run_delayed_ref_head(node, head, node->action);
2403 if (insert_reserved) {
2404 btrfs_pin_extent(root, node->bytenr,
2405 node->num_bytes, 1);
2406 if (head->is_data) {
2407 ret = btrfs_del_csums(trans, root,
2415 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2416 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2417 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2419 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2420 node->type == BTRFS_SHARED_DATA_REF_KEY)
2421 ret = run_delayed_data_ref(trans, root, node, extent_op,
2428 static inline struct btrfs_delayed_ref_node *
2429 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2431 struct btrfs_delayed_ref_node *ref;
2433 if (list_empty(&head->ref_list))
2437 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2438 * This is to prevent a ref count from going down to zero, which deletes
2439 * the extent item from the extent tree, when there still are references
2440 * to add, which would fail because they would not find the extent item.
2442 list_for_each_entry(ref, &head->ref_list, list) {
2443 if (ref->action == BTRFS_ADD_DELAYED_REF)
2447 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2452 * Returns 0 on success or if called with an already aborted transaction.
2453 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2455 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2456 struct btrfs_root *root,
2459 struct btrfs_delayed_ref_root *delayed_refs;
2460 struct btrfs_delayed_ref_node *ref;
2461 struct btrfs_delayed_ref_head *locked_ref = NULL;
2462 struct btrfs_delayed_extent_op *extent_op;
2463 struct btrfs_fs_info *fs_info = root->fs_info;
2464 ktime_t start = ktime_get();
2466 unsigned long count = 0;
2467 unsigned long actual_count = 0;
2468 int must_insert_reserved = 0;
2470 delayed_refs = &trans->transaction->delayed_refs;
2476 spin_lock(&delayed_refs->lock);
2477 locked_ref = btrfs_select_ref_head(trans);
2479 spin_unlock(&delayed_refs->lock);
2483 /* grab the lock that says we are going to process
2484 * all the refs for this head */
2485 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2486 spin_unlock(&delayed_refs->lock);
2488 * we may have dropped the spin lock to get the head
2489 * mutex lock, and that might have given someone else
2490 * time to free the head. If that's true, it has been
2491 * removed from our list and we can move on.
2493 if (ret == -EAGAIN) {
2500 spin_lock(&locked_ref->lock);
2503 * locked_ref is the head node, so we have to go one
2504 * node back for any delayed ref updates
2506 ref = select_delayed_ref(locked_ref);
2508 if (ref && ref->seq &&
2509 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2510 spin_unlock(&locked_ref->lock);
2511 btrfs_delayed_ref_unlock(locked_ref);
2512 spin_lock(&delayed_refs->lock);
2513 locked_ref->processing = 0;
2514 delayed_refs->num_heads_ready++;
2515 spin_unlock(&delayed_refs->lock);
2523 * record the must insert reserved flag before we
2524 * drop the spin lock.
2526 must_insert_reserved = locked_ref->must_insert_reserved;
2527 locked_ref->must_insert_reserved = 0;
2529 extent_op = locked_ref->extent_op;
2530 locked_ref->extent_op = NULL;
2535 /* All delayed refs have been processed, Go ahead
2536 * and send the head node to run_one_delayed_ref,
2537 * so that any accounting fixes can happen
2539 ref = &locked_ref->node;
2541 if (extent_op && must_insert_reserved) {
2542 btrfs_free_delayed_extent_op(extent_op);
2547 spin_unlock(&locked_ref->lock);
2548 ret = run_delayed_extent_op(trans, root,
2550 btrfs_free_delayed_extent_op(extent_op);
2554 * Need to reset must_insert_reserved if
2555 * there was an error so the abort stuff
2556 * can cleanup the reserved space
2559 if (must_insert_reserved)
2560 locked_ref->must_insert_reserved = 1;
2561 locked_ref->processing = 0;
2562 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2563 btrfs_delayed_ref_unlock(locked_ref);
2570 * Need to drop our head ref lock and re-aqcuire the
2571 * delayed ref lock and then re-check to make sure
2574 spin_unlock(&locked_ref->lock);
2575 spin_lock(&delayed_refs->lock);
2576 spin_lock(&locked_ref->lock);
2577 if (!list_empty(&locked_ref->ref_list) ||
2578 locked_ref->extent_op) {
2579 spin_unlock(&locked_ref->lock);
2580 spin_unlock(&delayed_refs->lock);
2584 delayed_refs->num_heads--;
2585 rb_erase(&locked_ref->href_node,
2586 &delayed_refs->href_root);
2587 spin_unlock(&delayed_refs->lock);
2591 list_del(&ref->list);
2593 atomic_dec(&delayed_refs->num_entries);
2595 if (!btrfs_delayed_ref_is_head(ref)) {
2597 * when we play the delayed ref, also correct the
2600 switch (ref->action) {
2601 case BTRFS_ADD_DELAYED_REF:
2602 case BTRFS_ADD_DELAYED_EXTENT:
2603 locked_ref->node.ref_mod -= ref->ref_mod;
2605 case BTRFS_DROP_DELAYED_REF:
2606 locked_ref->node.ref_mod += ref->ref_mod;
2612 spin_unlock(&locked_ref->lock);
2614 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2615 must_insert_reserved);
2617 btrfs_free_delayed_extent_op(extent_op);
2619 locked_ref->processing = 0;
2620 btrfs_delayed_ref_unlock(locked_ref);
2621 btrfs_put_delayed_ref(ref);
2622 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2627 * If this node is a head, that means all the refs in this head
2628 * have been dealt with, and we will pick the next head to deal
2629 * with, so we must unlock the head and drop it from the cluster
2630 * list before we release it.
2632 if (btrfs_delayed_ref_is_head(ref)) {
2633 if (locked_ref->is_data &&
2634 locked_ref->total_ref_mod < 0) {
2635 spin_lock(&delayed_refs->lock);
2636 delayed_refs->pending_csums -= ref->num_bytes;
2637 spin_unlock(&delayed_refs->lock);
2639 btrfs_delayed_ref_unlock(locked_ref);
2642 btrfs_put_delayed_ref(ref);
2648 * We don't want to include ref heads since we can have empty ref heads
2649 * and those will drastically skew our runtime down since we just do
2650 * accounting, no actual extent tree updates.
2652 if (actual_count > 0) {
2653 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2657 * We weigh the current average higher than our current runtime
2658 * to avoid large swings in the average.
2660 spin_lock(&delayed_refs->lock);
2661 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2662 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2663 spin_unlock(&delayed_refs->lock);
2668 #ifdef SCRAMBLE_DELAYED_REFS
2670 * Normally delayed refs get processed in ascending bytenr order. This
2671 * correlates in most cases to the order added. To expose dependencies on this
2672 * order, we start to process the tree in the middle instead of the beginning
2674 static u64 find_middle(struct rb_root *root)
2676 struct rb_node *n = root->rb_node;
2677 struct btrfs_delayed_ref_node *entry;
2680 u64 first = 0, last = 0;
2684 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2685 first = entry->bytenr;
2689 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2690 last = entry->bytenr;
2695 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2696 WARN_ON(!entry->in_tree);
2698 middle = entry->bytenr;
2711 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2715 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2716 sizeof(struct btrfs_extent_inline_ref));
2717 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2718 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2721 * We don't ever fill up leaves all the way so multiply by 2 just to be
2722 * closer to what we're really going to want to ouse.
2724 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2728 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2729 * would require to store the csums for that many bytes.
2731 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2734 u64 num_csums_per_leaf;
2737 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2738 num_csums_per_leaf = div64_u64(csum_size,
2739 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2740 num_csums = div64_u64(csum_bytes, root->sectorsize);
2741 num_csums += num_csums_per_leaf - 1;
2742 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2746 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2747 struct btrfs_root *root)
2749 struct btrfs_block_rsv *global_rsv;
2750 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2751 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2752 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2753 u64 num_bytes, num_dirty_bgs_bytes;
2756 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2757 num_heads = heads_to_leaves(root, num_heads);
2759 num_bytes += (num_heads - 1) * root->nodesize;
2761 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2762 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2764 global_rsv = &root->fs_info->global_block_rsv;
2767 * If we can't allocate any more chunks lets make sure we have _lots_ of
2768 * wiggle room since running delayed refs can create more delayed refs.
2770 if (global_rsv->space_info->full) {
2771 num_dirty_bgs_bytes <<= 1;
2775 spin_lock(&global_rsv->lock);
2776 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2778 spin_unlock(&global_rsv->lock);
2782 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2783 struct btrfs_root *root)
2785 struct btrfs_fs_info *fs_info = root->fs_info;
2787 atomic_read(&trans->transaction->delayed_refs.num_entries);
2792 avg_runtime = fs_info->avg_delayed_ref_runtime;
2793 val = num_entries * avg_runtime;
2794 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2796 if (val >= NSEC_PER_SEC / 2)
2799 return btrfs_check_space_for_delayed_refs(trans, root);
2802 struct async_delayed_refs {
2803 struct btrfs_root *root;
2807 struct completion wait;
2808 struct btrfs_work work;
2811 static void delayed_ref_async_start(struct btrfs_work *work)
2813 struct async_delayed_refs *async;
2814 struct btrfs_trans_handle *trans;
2817 async = container_of(work, struct async_delayed_refs, work);
2819 trans = btrfs_join_transaction(async->root);
2820 if (IS_ERR(trans)) {
2821 async->error = PTR_ERR(trans);
2826 * trans->sync means that when we call end_transaciton, we won't
2827 * wait on delayed refs
2830 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2834 ret = btrfs_end_transaction(trans, async->root);
2835 if (ret && !async->error)
2839 complete(&async->wait);
2844 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2845 unsigned long count, int wait)
2847 struct async_delayed_refs *async;
2850 async = kmalloc(sizeof(*async), GFP_NOFS);
2854 async->root = root->fs_info->tree_root;
2855 async->count = count;
2861 init_completion(&async->wait);
2863 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2864 delayed_ref_async_start, NULL, NULL);
2866 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2869 wait_for_completion(&async->wait);
2878 * this starts processing the delayed reference count updates and
2879 * extent insertions we have queued up so far. count can be
2880 * 0, which means to process everything in the tree at the start
2881 * of the run (but not newly added entries), or it can be some target
2882 * number you'd like to process.
2884 * Returns 0 on success or if called with an aborted transaction
2885 * Returns <0 on error and aborts the transaction
2887 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2888 struct btrfs_root *root, unsigned long count)
2890 struct rb_node *node;
2891 struct btrfs_delayed_ref_root *delayed_refs;
2892 struct btrfs_delayed_ref_head *head;
2894 int run_all = count == (unsigned long)-1;
2895 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2897 /* We'll clean this up in btrfs_cleanup_transaction */
2901 if (root == root->fs_info->extent_root)
2902 root = root->fs_info->tree_root;
2904 delayed_refs = &trans->transaction->delayed_refs;
2906 count = atomic_read(&delayed_refs->num_entries) * 2;
2909 #ifdef SCRAMBLE_DELAYED_REFS
2910 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2912 trans->can_flush_pending_bgs = false;
2913 ret = __btrfs_run_delayed_refs(trans, root, count);
2915 btrfs_abort_transaction(trans, root, ret);
2920 if (!list_empty(&trans->new_bgs))
2921 btrfs_create_pending_block_groups(trans, root);
2923 spin_lock(&delayed_refs->lock);
2924 node = rb_first(&delayed_refs->href_root);
2926 spin_unlock(&delayed_refs->lock);
2929 count = (unsigned long)-1;
2932 head = rb_entry(node, struct btrfs_delayed_ref_head,
2934 if (btrfs_delayed_ref_is_head(&head->node)) {
2935 struct btrfs_delayed_ref_node *ref;
2938 atomic_inc(&ref->refs);
2940 spin_unlock(&delayed_refs->lock);
2942 * Mutex was contended, block until it's
2943 * released and try again
2945 mutex_lock(&head->mutex);
2946 mutex_unlock(&head->mutex);
2948 btrfs_put_delayed_ref(ref);
2954 node = rb_next(node);
2956 spin_unlock(&delayed_refs->lock);
2961 assert_qgroups_uptodate(trans);
2962 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2966 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2967 struct btrfs_root *root,
2968 u64 bytenr, u64 num_bytes, u64 flags,
2969 int level, int is_data)
2971 struct btrfs_delayed_extent_op *extent_op;
2974 extent_op = btrfs_alloc_delayed_extent_op();
2978 extent_op->flags_to_set = flags;
2979 extent_op->update_flags = 1;
2980 extent_op->update_key = 0;
2981 extent_op->is_data = is_data ? 1 : 0;
2982 extent_op->level = level;
2984 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2985 num_bytes, extent_op);
2987 btrfs_free_delayed_extent_op(extent_op);
2991 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2992 struct btrfs_root *root,
2993 struct btrfs_path *path,
2994 u64 objectid, u64 offset, u64 bytenr)
2996 struct btrfs_delayed_ref_head *head;
2997 struct btrfs_delayed_ref_node *ref;
2998 struct btrfs_delayed_data_ref *data_ref;
2999 struct btrfs_delayed_ref_root *delayed_refs;
3002 delayed_refs = &trans->transaction->delayed_refs;
3003 spin_lock(&delayed_refs->lock);
3004 head = btrfs_find_delayed_ref_head(trans, bytenr);
3006 spin_unlock(&delayed_refs->lock);
3010 if (!mutex_trylock(&head->mutex)) {
3011 atomic_inc(&head->node.refs);
3012 spin_unlock(&delayed_refs->lock);
3014 btrfs_release_path(path);
3017 * Mutex was contended, block until it's released and let
3020 mutex_lock(&head->mutex);
3021 mutex_unlock(&head->mutex);
3022 btrfs_put_delayed_ref(&head->node);
3025 spin_unlock(&delayed_refs->lock);
3027 spin_lock(&head->lock);
3028 list_for_each_entry(ref, &head->ref_list, list) {
3029 /* If it's a shared ref we know a cross reference exists */
3030 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3035 data_ref = btrfs_delayed_node_to_data_ref(ref);
3038 * If our ref doesn't match the one we're currently looking at
3039 * then we have a cross reference.
3041 if (data_ref->root != root->root_key.objectid ||
3042 data_ref->objectid != objectid ||
3043 data_ref->offset != offset) {
3048 spin_unlock(&head->lock);
3049 mutex_unlock(&head->mutex);
3053 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3054 struct btrfs_root *root,
3055 struct btrfs_path *path,
3056 u64 objectid, u64 offset, u64 bytenr)
3058 struct btrfs_root *extent_root = root->fs_info->extent_root;
3059 struct extent_buffer *leaf;
3060 struct btrfs_extent_data_ref *ref;
3061 struct btrfs_extent_inline_ref *iref;
3062 struct btrfs_extent_item *ei;
3063 struct btrfs_key key;
3067 key.objectid = bytenr;
3068 key.offset = (u64)-1;
3069 key.type = BTRFS_EXTENT_ITEM_KEY;
3071 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3074 BUG_ON(ret == 0); /* Corruption */
3077 if (path->slots[0] == 0)
3081 leaf = path->nodes[0];
3082 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3084 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3088 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3089 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3090 if (item_size < sizeof(*ei)) {
3091 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3095 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3097 if (item_size != sizeof(*ei) +
3098 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3101 if (btrfs_extent_generation(leaf, ei) <=
3102 btrfs_root_last_snapshot(&root->root_item))
3105 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3106 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3107 BTRFS_EXTENT_DATA_REF_KEY)
3110 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3111 if (btrfs_extent_refs(leaf, ei) !=
3112 btrfs_extent_data_ref_count(leaf, ref) ||
3113 btrfs_extent_data_ref_root(leaf, ref) !=
3114 root->root_key.objectid ||
3115 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3116 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3124 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3125 struct btrfs_root *root,
3126 u64 objectid, u64 offset, u64 bytenr)
3128 struct btrfs_path *path;
3132 path = btrfs_alloc_path();
3137 ret = check_committed_ref(trans, root, path, objectid,
3139 if (ret && ret != -ENOENT)
3142 ret2 = check_delayed_ref(trans, root, path, objectid,
3144 } while (ret2 == -EAGAIN);
3146 if (ret2 && ret2 != -ENOENT) {
3151 if (ret != -ENOENT || ret2 != -ENOENT)
3154 btrfs_free_path(path);
3155 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3160 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3161 struct btrfs_root *root,
3162 struct extent_buffer *buf,
3163 int full_backref, int inc)
3170 struct btrfs_key key;
3171 struct btrfs_file_extent_item *fi;
3175 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3176 u64, u64, u64, u64, u64, u64, int);
3179 if (btrfs_test_is_dummy_root(root))
3182 ref_root = btrfs_header_owner(buf);
3183 nritems = btrfs_header_nritems(buf);
3184 level = btrfs_header_level(buf);
3186 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3190 process_func = btrfs_inc_extent_ref;
3192 process_func = btrfs_free_extent;
3195 parent = buf->start;
3199 for (i = 0; i < nritems; i++) {
3201 btrfs_item_key_to_cpu(buf, &key, i);
3202 if (key.type != BTRFS_EXTENT_DATA_KEY)
3204 fi = btrfs_item_ptr(buf, i,
3205 struct btrfs_file_extent_item);
3206 if (btrfs_file_extent_type(buf, fi) ==
3207 BTRFS_FILE_EXTENT_INLINE)
3209 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3213 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3214 key.offset -= btrfs_file_extent_offset(buf, fi);
3215 ret = process_func(trans, root, bytenr, num_bytes,
3216 parent, ref_root, key.objectid,
3221 bytenr = btrfs_node_blockptr(buf, i);
3222 num_bytes = root->nodesize;
3223 ret = process_func(trans, root, bytenr, num_bytes,
3224 parent, ref_root, level - 1, 0,
3235 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3236 struct extent_buffer *buf, int full_backref)
3238 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3241 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3242 struct extent_buffer *buf, int full_backref)
3244 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3247 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3248 struct btrfs_root *root,
3249 struct btrfs_path *path,
3250 struct btrfs_block_group_cache *cache)
3253 struct btrfs_root *extent_root = root->fs_info->extent_root;
3255 struct extent_buffer *leaf;
3257 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3264 leaf = path->nodes[0];
3265 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3266 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3267 btrfs_mark_buffer_dirty(leaf);
3269 btrfs_release_path(path);
3274 static struct btrfs_block_group_cache *
3275 next_block_group(struct btrfs_root *root,
3276 struct btrfs_block_group_cache *cache)
3278 struct rb_node *node;
3280 spin_lock(&root->fs_info->block_group_cache_lock);
3282 /* If our block group was removed, we need a full search. */
3283 if (RB_EMPTY_NODE(&cache->cache_node)) {
3284 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3286 spin_unlock(&root->fs_info->block_group_cache_lock);
3287 btrfs_put_block_group(cache);
3288 cache = btrfs_lookup_first_block_group(root->fs_info,
3292 node = rb_next(&cache->cache_node);
3293 btrfs_put_block_group(cache);
3295 cache = rb_entry(node, struct btrfs_block_group_cache,
3297 btrfs_get_block_group(cache);
3300 spin_unlock(&root->fs_info->block_group_cache_lock);
3304 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3305 struct btrfs_trans_handle *trans,
3306 struct btrfs_path *path)
3308 struct btrfs_root *root = block_group->fs_info->tree_root;
3309 struct inode *inode = NULL;
3311 int dcs = BTRFS_DC_ERROR;
3317 * If this block group is smaller than 100 megs don't bother caching the
3320 if (block_group->key.offset < (100 * 1024 * 1024)) {
3321 spin_lock(&block_group->lock);
3322 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3323 spin_unlock(&block_group->lock);
3330 inode = lookup_free_space_inode(root, block_group, path);
3331 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3332 ret = PTR_ERR(inode);
3333 btrfs_release_path(path);
3337 if (IS_ERR(inode)) {
3341 if (block_group->ro)
3344 ret = create_free_space_inode(root, trans, block_group, path);
3350 /* We've already setup this transaction, go ahead and exit */
3351 if (block_group->cache_generation == trans->transid &&
3352 i_size_read(inode)) {
3353 dcs = BTRFS_DC_SETUP;
3358 * We want to set the generation to 0, that way if anything goes wrong
3359 * from here on out we know not to trust this cache when we load up next
3362 BTRFS_I(inode)->generation = 0;
3363 ret = btrfs_update_inode(trans, root, inode);
3366 * So theoretically we could recover from this, simply set the
3367 * super cache generation to 0 so we know to invalidate the
3368 * cache, but then we'd have to keep track of the block groups
3369 * that fail this way so we know we _have_ to reset this cache
3370 * before the next commit or risk reading stale cache. So to
3371 * limit our exposure to horrible edge cases lets just abort the
3372 * transaction, this only happens in really bad situations
3375 btrfs_abort_transaction(trans, root, ret);
3380 if (i_size_read(inode) > 0) {
3381 ret = btrfs_check_trunc_cache_free_space(root,
3382 &root->fs_info->global_block_rsv);
3386 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3391 spin_lock(&block_group->lock);
3392 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3393 !btrfs_test_opt(root, SPACE_CACHE)) {
3395 * don't bother trying to write stuff out _if_
3396 * a) we're not cached,
3397 * b) we're with nospace_cache mount option.
3399 dcs = BTRFS_DC_WRITTEN;
3400 spin_unlock(&block_group->lock);
3403 spin_unlock(&block_group->lock);
3406 * We hit an ENOSPC when setting up the cache in this transaction, just
3407 * skip doing the setup, we've already cleared the cache so we're safe.
3409 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3415 * Try to preallocate enough space based on how big the block group is.
3416 * Keep in mind this has to include any pinned space which could end up
3417 * taking up quite a bit since it's not folded into the other space
3420 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3425 num_pages *= PAGE_CACHE_SIZE;
3427 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3431 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3432 num_pages, num_pages,
3435 * Our cache requires contiguous chunks so that we don't modify a bunch
3436 * of metadata or split extents when writing the cache out, which means
3437 * we can enospc if we are heavily fragmented in addition to just normal
3438 * out of space conditions. So if we hit this just skip setting up any
3439 * other block groups for this transaction, maybe we'll unpin enough
3440 * space the next time around.
3443 dcs = BTRFS_DC_SETUP;
3444 else if (ret == -ENOSPC)
3445 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3446 btrfs_free_reserved_data_space(inode, num_pages);
3451 btrfs_release_path(path);
3453 spin_lock(&block_group->lock);
3454 if (!ret && dcs == BTRFS_DC_SETUP)
3455 block_group->cache_generation = trans->transid;
3456 block_group->disk_cache_state = dcs;
3457 spin_unlock(&block_group->lock);
3462 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3463 struct btrfs_root *root)
3465 struct btrfs_block_group_cache *cache, *tmp;
3466 struct btrfs_transaction *cur_trans = trans->transaction;
3467 struct btrfs_path *path;
3469 if (list_empty(&cur_trans->dirty_bgs) ||
3470 !btrfs_test_opt(root, SPACE_CACHE))
3473 path = btrfs_alloc_path();
3477 /* Could add new block groups, use _safe just in case */
3478 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3480 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3481 cache_save_setup(cache, trans, path);
3484 btrfs_free_path(path);
3489 * transaction commit does final block group cache writeback during a
3490 * critical section where nothing is allowed to change the FS. This is
3491 * required in order for the cache to actually match the block group,
3492 * but can introduce a lot of latency into the commit.
3494 * So, btrfs_start_dirty_block_groups is here to kick off block group
3495 * cache IO. There's a chance we'll have to redo some of it if the
3496 * block group changes again during the commit, but it greatly reduces
3497 * the commit latency by getting rid of the easy block groups while
3498 * we're still allowing others to join the commit.
3500 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3501 struct btrfs_root *root)
3503 struct btrfs_block_group_cache *cache;
3504 struct btrfs_transaction *cur_trans = trans->transaction;
3507 struct btrfs_path *path = NULL;
3509 struct list_head *io = &cur_trans->io_bgs;
3510 int num_started = 0;
3513 spin_lock(&cur_trans->dirty_bgs_lock);
3514 if (list_empty(&cur_trans->dirty_bgs)) {
3515 spin_unlock(&cur_trans->dirty_bgs_lock);
3518 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3519 spin_unlock(&cur_trans->dirty_bgs_lock);
3523 * make sure all the block groups on our dirty list actually
3526 btrfs_create_pending_block_groups(trans, root);
3529 path = btrfs_alloc_path();
3535 * cache_write_mutex is here only to save us from balance or automatic
3536 * removal of empty block groups deleting this block group while we are
3537 * writing out the cache
3539 mutex_lock(&trans->transaction->cache_write_mutex);
3540 while (!list_empty(&dirty)) {
3541 cache = list_first_entry(&dirty,
3542 struct btrfs_block_group_cache,
3545 * this can happen if something re-dirties a block
3546 * group that is already under IO. Just wait for it to
3547 * finish and then do it all again
3549 if (!list_empty(&cache->io_list)) {
3550 list_del_init(&cache->io_list);
3551 btrfs_wait_cache_io(root, trans, cache,
3552 &cache->io_ctl, path,
3553 cache->key.objectid);
3554 btrfs_put_block_group(cache);
3559 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3560 * if it should update the cache_state. Don't delete
3561 * until after we wait.
3563 * Since we're not running in the commit critical section
3564 * we need the dirty_bgs_lock to protect from update_block_group
3566 spin_lock(&cur_trans->dirty_bgs_lock);
3567 list_del_init(&cache->dirty_list);
3568 spin_unlock(&cur_trans->dirty_bgs_lock);
3572 cache_save_setup(cache, trans, path);
3574 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3575 cache->io_ctl.inode = NULL;
3576 ret = btrfs_write_out_cache(root, trans, cache, path);
3577 if (ret == 0 && cache->io_ctl.inode) {
3582 * the cache_write_mutex is protecting
3585 list_add_tail(&cache->io_list, io);
3588 * if we failed to write the cache, the
3589 * generation will be bad and life goes on
3595 ret = write_one_cache_group(trans, root, path, cache);
3597 * Our block group might still be attached to the list
3598 * of new block groups in the transaction handle of some
3599 * other task (struct btrfs_trans_handle->new_bgs). This
3600 * means its block group item isn't yet in the extent
3601 * tree. If this happens ignore the error, as we will
3602 * try again later in the critical section of the
3603 * transaction commit.
3605 if (ret == -ENOENT) {
3607 spin_lock(&cur_trans->dirty_bgs_lock);
3608 if (list_empty(&cache->dirty_list)) {
3609 list_add_tail(&cache->dirty_list,
3610 &cur_trans->dirty_bgs);
3611 btrfs_get_block_group(cache);
3613 spin_unlock(&cur_trans->dirty_bgs_lock);
3615 btrfs_abort_transaction(trans, root, ret);
3619 /* if its not on the io list, we need to put the block group */
3621 btrfs_put_block_group(cache);
3627 * Avoid blocking other tasks for too long. It might even save
3628 * us from writing caches for block groups that are going to be
3631 mutex_unlock(&trans->transaction->cache_write_mutex);
3632 mutex_lock(&trans->transaction->cache_write_mutex);
3634 mutex_unlock(&trans->transaction->cache_write_mutex);
3637 * go through delayed refs for all the stuff we've just kicked off
3638 * and then loop back (just once)
3640 ret = btrfs_run_delayed_refs(trans, root, 0);
3641 if (!ret && loops == 0) {
3643 spin_lock(&cur_trans->dirty_bgs_lock);
3644 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3646 * dirty_bgs_lock protects us from concurrent block group
3647 * deletes too (not just cache_write_mutex).
3649 if (!list_empty(&dirty)) {
3650 spin_unlock(&cur_trans->dirty_bgs_lock);
3653 spin_unlock(&cur_trans->dirty_bgs_lock);
3656 btrfs_free_path(path);
3660 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3661 struct btrfs_root *root)
3663 struct btrfs_block_group_cache *cache;
3664 struct btrfs_transaction *cur_trans = trans->transaction;
3667 struct btrfs_path *path;
3668 struct list_head *io = &cur_trans->io_bgs;
3669 int num_started = 0;
3671 path = btrfs_alloc_path();
3676 * We don't need the lock here since we are protected by the transaction
3677 * commit. We want to do the cache_save_setup first and then run the
3678 * delayed refs to make sure we have the best chance at doing this all
3681 while (!list_empty(&cur_trans->dirty_bgs)) {
3682 cache = list_first_entry(&cur_trans->dirty_bgs,
3683 struct btrfs_block_group_cache,
3687 * this can happen if cache_save_setup re-dirties a block
3688 * group that is already under IO. Just wait for it to
3689 * finish and then do it all again
3691 if (!list_empty(&cache->io_list)) {
3692 list_del_init(&cache->io_list);
3693 btrfs_wait_cache_io(root, trans, cache,
3694 &cache->io_ctl, path,
3695 cache->key.objectid);
3696 btrfs_put_block_group(cache);
3700 * don't remove from the dirty list until after we've waited
3703 list_del_init(&cache->dirty_list);
3706 cache_save_setup(cache, trans, path);
3709 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3711 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3712 cache->io_ctl.inode = NULL;
3713 ret = btrfs_write_out_cache(root, trans, cache, path);
3714 if (ret == 0 && cache->io_ctl.inode) {
3717 list_add_tail(&cache->io_list, io);
3720 * if we failed to write the cache, the
3721 * generation will be bad and life goes on
3727 ret = write_one_cache_group(trans, root, path, cache);
3729 btrfs_abort_transaction(trans, root, ret);
3732 /* if its not on the io list, we need to put the block group */
3734 btrfs_put_block_group(cache);
3737 while (!list_empty(io)) {
3738 cache = list_first_entry(io, struct btrfs_block_group_cache,
3740 list_del_init(&cache->io_list);
3741 btrfs_wait_cache_io(root, trans, cache,
3742 &cache->io_ctl, path, cache->key.objectid);
3743 btrfs_put_block_group(cache);
3746 btrfs_free_path(path);
3750 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3752 struct btrfs_block_group_cache *block_group;
3755 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3756 if (!block_group || block_group->ro)
3759 btrfs_put_block_group(block_group);
3763 static const char *alloc_name(u64 flags)
3766 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3768 case BTRFS_BLOCK_GROUP_METADATA:
3770 case BTRFS_BLOCK_GROUP_DATA:
3772 case BTRFS_BLOCK_GROUP_SYSTEM:
3776 return "invalid-combination";
3780 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3781 u64 total_bytes, u64 bytes_used,
3782 struct btrfs_space_info **space_info)
3784 struct btrfs_space_info *found;
3789 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3790 BTRFS_BLOCK_GROUP_RAID10))
3795 found = __find_space_info(info, flags);
3797 spin_lock(&found->lock);
3798 found->total_bytes += total_bytes;
3799 found->disk_total += total_bytes * factor;
3800 found->bytes_used += bytes_used;
3801 found->disk_used += bytes_used * factor;
3802 if (total_bytes > 0)
3804 spin_unlock(&found->lock);
3805 *space_info = found;
3808 found = kzalloc(sizeof(*found), GFP_NOFS);
3812 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3818 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3819 INIT_LIST_HEAD(&found->block_groups[i]);
3820 init_rwsem(&found->groups_sem);
3821 spin_lock_init(&found->lock);
3822 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3823 found->total_bytes = total_bytes;
3824 found->disk_total = total_bytes * factor;
3825 found->bytes_used = bytes_used;
3826 found->disk_used = bytes_used * factor;
3827 found->bytes_pinned = 0;
3828 found->bytes_reserved = 0;
3829 found->bytes_readonly = 0;
3830 found->bytes_may_use = 0;
3832 found->max_extent_size = 0;
3833 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3834 found->chunk_alloc = 0;
3836 init_waitqueue_head(&found->wait);
3837 INIT_LIST_HEAD(&found->ro_bgs);
3839 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3840 info->space_info_kobj, "%s",
3841 alloc_name(found->flags));
3847 *space_info = found;
3848 list_add_rcu(&found->list, &info->space_info);
3849 if (flags & BTRFS_BLOCK_GROUP_DATA)
3850 info->data_sinfo = found;
3855 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3857 u64 extra_flags = chunk_to_extended(flags) &
3858 BTRFS_EXTENDED_PROFILE_MASK;
3860 write_seqlock(&fs_info->profiles_lock);
3861 if (flags & BTRFS_BLOCK_GROUP_DATA)
3862 fs_info->avail_data_alloc_bits |= extra_flags;
3863 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3864 fs_info->avail_metadata_alloc_bits |= extra_flags;
3865 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3866 fs_info->avail_system_alloc_bits |= extra_flags;
3867 write_sequnlock(&fs_info->profiles_lock);
3871 * returns target flags in extended format or 0 if restripe for this
3872 * chunk_type is not in progress
3874 * should be called with either volume_mutex or balance_lock held
3876 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3878 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3884 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3885 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3886 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3887 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3888 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3889 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3890 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3891 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3892 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3899 * @flags: available profiles in extended format (see ctree.h)
3901 * Returns reduced profile in chunk format. If profile changing is in
3902 * progress (either running or paused) picks the target profile (if it's
3903 * already available), otherwise falls back to plain reducing.
3905 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3907 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3913 * see if restripe for this chunk_type is in progress, if so
3914 * try to reduce to the target profile
3916 spin_lock(&root->fs_info->balance_lock);
3917 target = get_restripe_target(root->fs_info, flags);
3919 /* pick target profile only if it's already available */
3920 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3921 spin_unlock(&root->fs_info->balance_lock);
3922 return extended_to_chunk(target);
3925 spin_unlock(&root->fs_info->balance_lock);
3927 /* First, mask out the RAID levels which aren't possible */
3928 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3929 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
3930 allowed |= btrfs_raid_group[raid_type];
3934 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
3935 allowed = BTRFS_BLOCK_GROUP_RAID6;
3936 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
3937 allowed = BTRFS_BLOCK_GROUP_RAID5;
3938 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
3939 allowed = BTRFS_BLOCK_GROUP_RAID10;
3940 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
3941 allowed = BTRFS_BLOCK_GROUP_RAID1;
3942 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
3943 allowed = BTRFS_BLOCK_GROUP_RAID0;
3945 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
3947 return extended_to_chunk(flags | allowed);
3950 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3957 seq = read_seqbegin(&root->fs_info->profiles_lock);
3959 if (flags & BTRFS_BLOCK_GROUP_DATA)
3960 flags |= root->fs_info->avail_data_alloc_bits;
3961 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3962 flags |= root->fs_info->avail_system_alloc_bits;
3963 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3964 flags |= root->fs_info->avail_metadata_alloc_bits;
3965 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3967 return btrfs_reduce_alloc_profile(root, flags);
3970 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3976 flags = BTRFS_BLOCK_GROUP_DATA;
3977 else if (root == root->fs_info->chunk_root)
3978 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3980 flags = BTRFS_BLOCK_GROUP_METADATA;
3982 ret = get_alloc_profile(root, flags);
3987 * This will check the space that the inode allocates from to make sure we have
3988 * enough space for bytes.
3990 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3992 struct btrfs_space_info *data_sinfo;
3993 struct btrfs_root *root = BTRFS_I(inode)->root;
3994 struct btrfs_fs_info *fs_info = root->fs_info;
3997 int need_commit = 2;
3998 int have_pinned_space;
4000 /* make sure bytes are sectorsize aligned */
4001 bytes = ALIGN(bytes, root->sectorsize);
4003 if (btrfs_is_free_space_inode(inode)) {
4005 ASSERT(current->journal_info);
4008 data_sinfo = fs_info->data_sinfo;
4013 /* make sure we have enough space to handle the data first */
4014 spin_lock(&data_sinfo->lock);
4015 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4016 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4017 data_sinfo->bytes_may_use;
4019 if (used + bytes > data_sinfo->total_bytes) {
4020 struct btrfs_trans_handle *trans;
4023 * if we don't have enough free bytes in this space then we need
4024 * to alloc a new chunk.
4026 if (!data_sinfo->full) {
4029 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4030 spin_unlock(&data_sinfo->lock);
4032 alloc_target = btrfs_get_alloc_profile(root, 1);
4034 * It is ugly that we don't call nolock join
4035 * transaction for the free space inode case here.
4036 * But it is safe because we only do the data space
4037 * reservation for the free space cache in the
4038 * transaction context, the common join transaction
4039 * just increase the counter of the current transaction
4040 * handler, doesn't try to acquire the trans_lock of
4043 trans = btrfs_join_transaction(root);
4045 return PTR_ERR(trans);
4047 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4049 CHUNK_ALLOC_NO_FORCE);
4050 btrfs_end_transaction(trans, root);
4055 have_pinned_space = 1;
4061 data_sinfo = fs_info->data_sinfo;
4067 * If we don't have enough pinned space to deal with this
4068 * allocation, and no removed chunk in current transaction,
4069 * don't bother committing the transaction.
4071 have_pinned_space = percpu_counter_compare(
4072 &data_sinfo->total_bytes_pinned,
4073 used + bytes - data_sinfo->total_bytes);
4074 spin_unlock(&data_sinfo->lock);
4076 /* commit the current transaction and try again */
4079 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4082 if (need_commit > 0)
4083 btrfs_wait_ordered_roots(fs_info, -1);
4085 trans = btrfs_join_transaction(root);
4087 return PTR_ERR(trans);
4088 if (have_pinned_space >= 0 ||
4089 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4090 &trans->transaction->flags) ||
4092 ret = btrfs_commit_transaction(trans, root);
4096 * make sure that all running delayed iput are
4099 down_write(&root->fs_info->delayed_iput_sem);
4100 up_write(&root->fs_info->delayed_iput_sem);
4103 btrfs_end_transaction(trans, root);
4107 trace_btrfs_space_reservation(root->fs_info,
4108 "space_info:enospc",
4109 data_sinfo->flags, bytes, 1);
4112 ret = btrfs_qgroup_reserve(root, write_bytes);
4115 data_sinfo->bytes_may_use += bytes;
4116 trace_btrfs_space_reservation(root->fs_info, "space_info",
4117 data_sinfo->flags, bytes, 1);
4119 spin_unlock(&data_sinfo->lock);
4125 * Called if we need to clear a data reservation for this inode.
4127 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
4129 struct btrfs_root *root = BTRFS_I(inode)->root;
4130 struct btrfs_space_info *data_sinfo;
4132 /* make sure bytes are sectorsize aligned */
4133 bytes = ALIGN(bytes, root->sectorsize);
4135 data_sinfo = root->fs_info->data_sinfo;
4136 spin_lock(&data_sinfo->lock);
4137 WARN_ON(data_sinfo->bytes_may_use < bytes);
4138 data_sinfo->bytes_may_use -= bytes;
4139 trace_btrfs_space_reservation(root->fs_info, "space_info",
4140 data_sinfo->flags, bytes, 0);
4141 spin_unlock(&data_sinfo->lock);
4144 static void force_metadata_allocation(struct btrfs_fs_info *info)
4146 struct list_head *head = &info->space_info;
4147 struct btrfs_space_info *found;
4150 list_for_each_entry_rcu(found, head, list) {
4151 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4152 found->force_alloc = CHUNK_ALLOC_FORCE;
4157 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4159 return (global->size << 1);
4162 static int should_alloc_chunk(struct btrfs_root *root,
4163 struct btrfs_space_info *sinfo, int force)
4165 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4166 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4167 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4170 if (force == CHUNK_ALLOC_FORCE)
4174 * We need to take into account the global rsv because for all intents
4175 * and purposes it's used space. Don't worry about locking the
4176 * global_rsv, it doesn't change except when the transaction commits.
4178 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4179 num_allocated += calc_global_rsv_need_space(global_rsv);
4182 * in limited mode, we want to have some free space up to
4183 * about 1% of the FS size.
4185 if (force == CHUNK_ALLOC_LIMITED) {
4186 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4187 thresh = max_t(u64, 64 * 1024 * 1024,
4188 div_factor_fine(thresh, 1));
4190 if (num_bytes - num_allocated < thresh)
4194 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4199 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4203 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4204 BTRFS_BLOCK_GROUP_RAID0 |
4205 BTRFS_BLOCK_GROUP_RAID5 |
4206 BTRFS_BLOCK_GROUP_RAID6))
4207 num_dev = root->fs_info->fs_devices->rw_devices;
4208 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4211 num_dev = 1; /* DUP or single */
4217 * If @is_allocation is true, reserve space in the system space info necessary
4218 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4221 void check_system_chunk(struct btrfs_trans_handle *trans,
4222 struct btrfs_root *root,
4225 struct btrfs_space_info *info;
4232 * Needed because we can end up allocating a system chunk and for an
4233 * atomic and race free space reservation in the chunk block reserve.
4235 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4237 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4238 spin_lock(&info->lock);
4239 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4240 info->bytes_reserved - info->bytes_readonly -
4241 info->bytes_may_use;
4242 spin_unlock(&info->lock);
4244 num_devs = get_profile_num_devs(root, type);
4246 /* num_devs device items to update and 1 chunk item to add or remove */
4247 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4248 btrfs_calc_trans_metadata_size(root, 1);
4250 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4251 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4252 left, thresh, type);
4253 dump_space_info(info, 0, 0);
4256 if (left < thresh) {
4259 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4261 * Ignore failure to create system chunk. We might end up not
4262 * needing it, as we might not need to COW all nodes/leafs from
4263 * the paths we visit in the chunk tree (they were already COWed
4264 * or created in the current transaction for example).
4266 ret = btrfs_alloc_chunk(trans, root, flags);
4270 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4271 &root->fs_info->chunk_block_rsv,
4272 thresh, BTRFS_RESERVE_NO_FLUSH);
4274 trans->chunk_bytes_reserved += thresh;
4278 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4279 struct btrfs_root *extent_root, u64 flags, int force)
4281 struct btrfs_space_info *space_info;
4282 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4283 int wait_for_alloc = 0;
4286 /* Don't re-enter if we're already allocating a chunk */
4287 if (trans->allocating_chunk)
4290 space_info = __find_space_info(extent_root->fs_info, flags);
4292 ret = update_space_info(extent_root->fs_info, flags,
4294 BUG_ON(ret); /* -ENOMEM */
4296 BUG_ON(!space_info); /* Logic error */
4299 spin_lock(&space_info->lock);
4300 if (force < space_info->force_alloc)
4301 force = space_info->force_alloc;
4302 if (space_info->full) {
4303 if (should_alloc_chunk(extent_root, space_info, force))
4307 spin_unlock(&space_info->lock);
4311 if (!should_alloc_chunk(extent_root, space_info, force)) {
4312 spin_unlock(&space_info->lock);
4314 } else if (space_info->chunk_alloc) {
4317 space_info->chunk_alloc = 1;
4320 spin_unlock(&space_info->lock);
4322 mutex_lock(&fs_info->chunk_mutex);
4325 * The chunk_mutex is held throughout the entirety of a chunk
4326 * allocation, so once we've acquired the chunk_mutex we know that the
4327 * other guy is done and we need to recheck and see if we should
4330 if (wait_for_alloc) {
4331 mutex_unlock(&fs_info->chunk_mutex);
4336 trans->allocating_chunk = true;
4339 * If we have mixed data/metadata chunks we want to make sure we keep
4340 * allocating mixed chunks instead of individual chunks.
4342 if (btrfs_mixed_space_info(space_info))
4343 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4346 * if we're doing a data chunk, go ahead and make sure that
4347 * we keep a reasonable number of metadata chunks allocated in the
4350 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4351 fs_info->data_chunk_allocations++;
4352 if (!(fs_info->data_chunk_allocations %
4353 fs_info->metadata_ratio))
4354 force_metadata_allocation(fs_info);
4358 * Check if we have enough space in SYSTEM chunk because we may need
4359 * to update devices.
4361 check_system_chunk(trans, extent_root, flags);
4363 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4364 trans->allocating_chunk = false;
4366 spin_lock(&space_info->lock);
4367 if (ret < 0 && ret != -ENOSPC)
4370 space_info->full = 1;
4374 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4376 space_info->chunk_alloc = 0;
4377 spin_unlock(&space_info->lock);
4378 mutex_unlock(&fs_info->chunk_mutex);
4380 * When we allocate a new chunk we reserve space in the chunk block
4381 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4382 * add new nodes/leafs to it if we end up needing to do it when
4383 * inserting the chunk item and updating device items as part of the
4384 * second phase of chunk allocation, performed by
4385 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4386 * large number of new block groups to create in our transaction
4387 * handle's new_bgs list to avoid exhausting the chunk block reserve
4388 * in extreme cases - like having a single transaction create many new
4389 * block groups when starting to write out the free space caches of all
4390 * the block groups that were made dirty during the lifetime of the
4393 if (trans->can_flush_pending_bgs &&
4394 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4395 btrfs_create_pending_block_groups(trans, trans->root);
4396 btrfs_trans_release_chunk_metadata(trans);
4401 static int can_overcommit(struct btrfs_root *root,
4402 struct btrfs_space_info *space_info, u64 bytes,
4403 enum btrfs_reserve_flush_enum flush)
4405 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4406 u64 profile = btrfs_get_alloc_profile(root, 0);
4411 used = space_info->bytes_used + space_info->bytes_reserved +
4412 space_info->bytes_pinned + space_info->bytes_readonly;
4415 * We only want to allow over committing if we have lots of actual space
4416 * free, but if we don't have enough space to handle the global reserve
4417 * space then we could end up having a real enospc problem when trying
4418 * to allocate a chunk or some other such important allocation.
4420 spin_lock(&global_rsv->lock);
4421 space_size = calc_global_rsv_need_space(global_rsv);
4422 spin_unlock(&global_rsv->lock);
4423 if (used + space_size >= space_info->total_bytes)
4426 used += space_info->bytes_may_use;
4428 spin_lock(&root->fs_info->free_chunk_lock);
4429 avail = root->fs_info->free_chunk_space;
4430 spin_unlock(&root->fs_info->free_chunk_lock);
4433 * If we have dup, raid1 or raid10 then only half of the free
4434 * space is actually useable. For raid56, the space info used
4435 * doesn't include the parity drive, so we don't have to
4438 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4439 BTRFS_BLOCK_GROUP_RAID1 |
4440 BTRFS_BLOCK_GROUP_RAID10))
4444 * If we aren't flushing all things, let us overcommit up to
4445 * 1/2th of the space. If we can flush, don't let us overcommit
4446 * too much, let it overcommit up to 1/8 of the space.
4448 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4453 if (used + bytes < space_info->total_bytes + avail)
4458 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4459 unsigned long nr_pages, int nr_items)
4461 struct super_block *sb = root->fs_info->sb;
4463 if (down_read_trylock(&sb->s_umount)) {
4464 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4465 up_read(&sb->s_umount);
4468 * We needn't worry the filesystem going from r/w to r/o though
4469 * we don't acquire ->s_umount mutex, because the filesystem
4470 * should guarantee the delalloc inodes list be empty after
4471 * the filesystem is readonly(all dirty pages are written to
4474 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4475 if (!current->journal_info)
4476 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4480 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4485 bytes = btrfs_calc_trans_metadata_size(root, 1);
4486 nr = (int)div64_u64(to_reclaim, bytes);
4492 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4495 * shrink metadata reservation for delalloc
4497 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4500 struct btrfs_block_rsv *block_rsv;
4501 struct btrfs_space_info *space_info;
4502 struct btrfs_trans_handle *trans;
4506 unsigned long nr_pages;
4509 enum btrfs_reserve_flush_enum flush;
4511 /* Calc the number of the pages we need flush for space reservation */
4512 items = calc_reclaim_items_nr(root, to_reclaim);
4513 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4515 trans = (struct btrfs_trans_handle *)current->journal_info;
4516 block_rsv = &root->fs_info->delalloc_block_rsv;
4517 space_info = block_rsv->space_info;
4519 delalloc_bytes = percpu_counter_sum_positive(
4520 &root->fs_info->delalloc_bytes);
4521 if (delalloc_bytes == 0) {
4525 btrfs_wait_ordered_roots(root->fs_info, items);
4530 while (delalloc_bytes && loops < 3) {
4531 max_reclaim = min(delalloc_bytes, to_reclaim);
4532 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4533 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4535 * We need to wait for the async pages to actually start before
4538 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4542 if (max_reclaim <= nr_pages)
4545 max_reclaim -= nr_pages;
4547 wait_event(root->fs_info->async_submit_wait,
4548 atomic_read(&root->fs_info->async_delalloc_pages) <=
4552 flush = BTRFS_RESERVE_FLUSH_ALL;
4554 flush = BTRFS_RESERVE_NO_FLUSH;
4555 spin_lock(&space_info->lock);
4556 if (can_overcommit(root, space_info, orig, flush)) {
4557 spin_unlock(&space_info->lock);
4560 spin_unlock(&space_info->lock);
4563 if (wait_ordered && !trans) {
4564 btrfs_wait_ordered_roots(root->fs_info, items);
4566 time_left = schedule_timeout_killable(1);
4570 delalloc_bytes = percpu_counter_sum_positive(
4571 &root->fs_info->delalloc_bytes);
4576 * maybe_commit_transaction - possibly commit the transaction if its ok to
4577 * @root - the root we're allocating for
4578 * @bytes - the number of bytes we want to reserve
4579 * @force - force the commit
4581 * This will check to make sure that committing the transaction will actually
4582 * get us somewhere and then commit the transaction if it does. Otherwise it
4583 * will return -ENOSPC.
4585 static int may_commit_transaction(struct btrfs_root *root,
4586 struct btrfs_space_info *space_info,
4587 u64 bytes, int force)
4589 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4590 struct btrfs_trans_handle *trans;
4592 trans = (struct btrfs_trans_handle *)current->journal_info;
4599 /* See if there is enough pinned space to make this reservation */
4600 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4605 * See if there is some space in the delayed insertion reservation for
4608 if (space_info != delayed_rsv->space_info)
4611 spin_lock(&delayed_rsv->lock);
4612 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4613 bytes - delayed_rsv->size) >= 0) {
4614 spin_unlock(&delayed_rsv->lock);
4617 spin_unlock(&delayed_rsv->lock);
4620 trans = btrfs_join_transaction(root);
4624 return btrfs_commit_transaction(trans, root);
4628 FLUSH_DELAYED_ITEMS_NR = 1,
4629 FLUSH_DELAYED_ITEMS = 2,
4631 FLUSH_DELALLOC_WAIT = 4,
4636 static int flush_space(struct btrfs_root *root,
4637 struct btrfs_space_info *space_info, u64 num_bytes,
4638 u64 orig_bytes, int state)
4640 struct btrfs_trans_handle *trans;
4645 case FLUSH_DELAYED_ITEMS_NR:
4646 case FLUSH_DELAYED_ITEMS:
4647 if (state == FLUSH_DELAYED_ITEMS_NR)
4648 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4652 trans = btrfs_join_transaction(root);
4653 if (IS_ERR(trans)) {
4654 ret = PTR_ERR(trans);
4657 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4658 btrfs_end_transaction(trans, root);
4660 case FLUSH_DELALLOC:
4661 case FLUSH_DELALLOC_WAIT:
4662 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4663 state == FLUSH_DELALLOC_WAIT);
4666 trans = btrfs_join_transaction(root);
4667 if (IS_ERR(trans)) {
4668 ret = PTR_ERR(trans);
4671 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4672 btrfs_get_alloc_profile(root, 0),
4673 CHUNK_ALLOC_NO_FORCE);
4674 btrfs_end_transaction(trans, root);
4679 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4690 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4691 struct btrfs_space_info *space_info)
4697 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4699 spin_lock(&space_info->lock);
4700 if (can_overcommit(root, space_info, to_reclaim,
4701 BTRFS_RESERVE_FLUSH_ALL)) {
4706 used = space_info->bytes_used + space_info->bytes_reserved +
4707 space_info->bytes_pinned + space_info->bytes_readonly +
4708 space_info->bytes_may_use;
4709 if (can_overcommit(root, space_info, 1024 * 1024,
4710 BTRFS_RESERVE_FLUSH_ALL))
4711 expected = div_factor_fine(space_info->total_bytes, 95);
4713 expected = div_factor_fine(space_info->total_bytes, 90);
4715 if (used > expected)
4716 to_reclaim = used - expected;
4719 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4720 space_info->bytes_reserved);
4722 spin_unlock(&space_info->lock);
4727 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4728 struct btrfs_fs_info *fs_info, u64 used)
4730 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4732 /* If we're just plain full then async reclaim just slows us down. */
4733 if (space_info->bytes_used >= thresh)
4736 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4737 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4740 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4741 struct btrfs_fs_info *fs_info,
4746 spin_lock(&space_info->lock);
4748 * We run out of space and have not got any free space via flush_space,
4749 * so don't bother doing async reclaim.
4751 if (flush_state > COMMIT_TRANS && space_info->full) {
4752 spin_unlock(&space_info->lock);
4756 used = space_info->bytes_used + space_info->bytes_reserved +
4757 space_info->bytes_pinned + space_info->bytes_readonly +
4758 space_info->bytes_may_use;
4759 if (need_do_async_reclaim(space_info, fs_info, used)) {
4760 spin_unlock(&space_info->lock);
4763 spin_unlock(&space_info->lock);
4768 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4770 struct btrfs_fs_info *fs_info;
4771 struct btrfs_space_info *space_info;
4775 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4776 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4778 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4783 flush_state = FLUSH_DELAYED_ITEMS_NR;
4785 flush_space(fs_info->fs_root, space_info, to_reclaim,
4786 to_reclaim, flush_state);
4788 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4791 } while (flush_state < COMMIT_TRANS);
4794 void btrfs_init_async_reclaim_work(struct work_struct *work)
4796 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4800 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4801 * @root - the root we're allocating for
4802 * @block_rsv - the block_rsv we're allocating for
4803 * @orig_bytes - the number of bytes we want
4804 * @flush - whether or not we can flush to make our reservation
4806 * This will reserve orgi_bytes number of bytes from the space info associated
4807 * with the block_rsv. If there is not enough space it will make an attempt to
4808 * flush out space to make room. It will do this by flushing delalloc if
4809 * possible or committing the transaction. If flush is 0 then no attempts to
4810 * regain reservations will be made and this will fail if there is not enough
4813 static int reserve_metadata_bytes(struct btrfs_root *root,
4814 struct btrfs_block_rsv *block_rsv,
4816 enum btrfs_reserve_flush_enum flush)
4818 struct btrfs_space_info *space_info = block_rsv->space_info;
4820 u64 num_bytes = orig_bytes;
4821 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4823 bool flushing = false;
4827 spin_lock(&space_info->lock);
4829 * We only want to wait if somebody other than us is flushing and we
4830 * are actually allowed to flush all things.
4832 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4833 space_info->flush) {
4834 spin_unlock(&space_info->lock);
4836 * If we have a trans handle we can't wait because the flusher
4837 * may have to commit the transaction, which would mean we would
4838 * deadlock since we are waiting for the flusher to finish, but
4839 * hold the current transaction open.
4841 if (current->journal_info)
4843 ret = wait_event_killable(space_info->wait, !space_info->flush);
4844 /* Must have been killed, return */
4848 spin_lock(&space_info->lock);
4852 used = space_info->bytes_used + space_info->bytes_reserved +
4853 space_info->bytes_pinned + space_info->bytes_readonly +
4854 space_info->bytes_may_use;
4857 * The idea here is that we've not already over-reserved the block group
4858 * then we can go ahead and save our reservation first and then start
4859 * flushing if we need to. Otherwise if we've already overcommitted
4860 * lets start flushing stuff first and then come back and try to make
4863 if (used <= space_info->total_bytes) {
4864 if (used + orig_bytes <= space_info->total_bytes) {
4865 space_info->bytes_may_use += orig_bytes;
4866 trace_btrfs_space_reservation(root->fs_info,
4867 "space_info", space_info->flags, orig_bytes, 1);
4871 * Ok set num_bytes to orig_bytes since we aren't
4872 * overocmmitted, this way we only try and reclaim what
4875 num_bytes = orig_bytes;
4879 * Ok we're over committed, set num_bytes to the overcommitted
4880 * amount plus the amount of bytes that we need for this
4883 num_bytes = used - space_info->total_bytes +
4887 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4888 space_info->bytes_may_use += orig_bytes;
4889 trace_btrfs_space_reservation(root->fs_info, "space_info",
4890 space_info->flags, orig_bytes,
4896 * Couldn't make our reservation, save our place so while we're trying
4897 * to reclaim space we can actually use it instead of somebody else
4898 * stealing it from us.
4900 * We make the other tasks wait for the flush only when we can flush
4903 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4905 space_info->flush = 1;
4906 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4909 * We will do the space reservation dance during log replay,
4910 * which means we won't have fs_info->fs_root set, so don't do
4911 * the async reclaim as we will panic.
4913 if (!root->fs_info->log_root_recovering &&
4914 need_do_async_reclaim(space_info, root->fs_info, used) &&
4915 !work_busy(&root->fs_info->async_reclaim_work))
4916 queue_work(system_unbound_wq,
4917 &root->fs_info->async_reclaim_work);
4919 spin_unlock(&space_info->lock);
4921 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4924 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4929 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4930 * would happen. So skip delalloc flush.
4932 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4933 (flush_state == FLUSH_DELALLOC ||
4934 flush_state == FLUSH_DELALLOC_WAIT))
4935 flush_state = ALLOC_CHUNK;
4939 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4940 flush_state < COMMIT_TRANS)
4942 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4943 flush_state <= COMMIT_TRANS)
4947 if (ret == -ENOSPC &&
4948 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4949 struct btrfs_block_rsv *global_rsv =
4950 &root->fs_info->global_block_rsv;
4952 if (block_rsv != global_rsv &&
4953 !block_rsv_use_bytes(global_rsv, orig_bytes))
4957 trace_btrfs_space_reservation(root->fs_info,
4958 "space_info:enospc",
4959 space_info->flags, orig_bytes, 1);
4961 spin_lock(&space_info->lock);
4962 space_info->flush = 0;
4963 wake_up_all(&space_info->wait);
4964 spin_unlock(&space_info->lock);
4969 static struct btrfs_block_rsv *get_block_rsv(
4970 const struct btrfs_trans_handle *trans,
4971 const struct btrfs_root *root)
4973 struct btrfs_block_rsv *block_rsv = NULL;
4975 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4976 (root == root->fs_info->csum_root && trans->adding_csums) ||
4977 (root == root->fs_info->uuid_root))
4978 block_rsv = trans->block_rsv;
4981 block_rsv = root->block_rsv;
4984 block_rsv = &root->fs_info->empty_block_rsv;
4989 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4993 spin_lock(&block_rsv->lock);
4994 if (block_rsv->reserved >= num_bytes) {
4995 block_rsv->reserved -= num_bytes;
4996 if (block_rsv->reserved < block_rsv->size)
4997 block_rsv->full = 0;
5000 spin_unlock(&block_rsv->lock);
5004 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5005 u64 num_bytes, int update_size)
5007 spin_lock(&block_rsv->lock);
5008 block_rsv->reserved += num_bytes;
5010 block_rsv->size += num_bytes;
5011 else if (block_rsv->reserved >= block_rsv->size)
5012 block_rsv->full = 1;
5013 spin_unlock(&block_rsv->lock);
5016 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5017 struct btrfs_block_rsv *dest, u64 num_bytes,
5020 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5023 if (global_rsv->space_info != dest->space_info)
5026 spin_lock(&global_rsv->lock);
5027 min_bytes = div_factor(global_rsv->size, min_factor);
5028 if (global_rsv->reserved < min_bytes + num_bytes) {
5029 spin_unlock(&global_rsv->lock);
5032 global_rsv->reserved -= num_bytes;
5033 if (global_rsv->reserved < global_rsv->size)
5034 global_rsv->full = 0;
5035 spin_unlock(&global_rsv->lock);
5037 block_rsv_add_bytes(dest, num_bytes, 1);
5041 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5042 struct btrfs_block_rsv *block_rsv,
5043 struct btrfs_block_rsv *dest, u64 num_bytes)
5045 struct btrfs_space_info *space_info = block_rsv->space_info;
5047 spin_lock(&block_rsv->lock);
5048 if (num_bytes == (u64)-1)
5049 num_bytes = block_rsv->size;
5050 block_rsv->size -= num_bytes;
5051 if (block_rsv->reserved >= block_rsv->size) {
5052 num_bytes = block_rsv->reserved - block_rsv->size;
5053 block_rsv->reserved = block_rsv->size;
5054 block_rsv->full = 1;
5058 spin_unlock(&block_rsv->lock);
5060 if (num_bytes > 0) {
5062 spin_lock(&dest->lock);
5066 bytes_to_add = dest->size - dest->reserved;
5067 bytes_to_add = min(num_bytes, bytes_to_add);
5068 dest->reserved += bytes_to_add;
5069 if (dest->reserved >= dest->size)
5071 num_bytes -= bytes_to_add;
5073 spin_unlock(&dest->lock);
5076 spin_lock(&space_info->lock);
5077 space_info->bytes_may_use -= num_bytes;
5078 trace_btrfs_space_reservation(fs_info, "space_info",
5079 space_info->flags, num_bytes, 0);
5080 spin_unlock(&space_info->lock);
5085 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5086 struct btrfs_block_rsv *dst, u64 num_bytes)
5090 ret = block_rsv_use_bytes(src, num_bytes);
5094 block_rsv_add_bytes(dst, num_bytes, 1);
5098 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5100 memset(rsv, 0, sizeof(*rsv));
5101 spin_lock_init(&rsv->lock);
5105 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5106 unsigned short type)
5108 struct btrfs_block_rsv *block_rsv;
5109 struct btrfs_fs_info *fs_info = root->fs_info;
5111 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5115 btrfs_init_block_rsv(block_rsv, type);
5116 block_rsv->space_info = __find_space_info(fs_info,
5117 BTRFS_BLOCK_GROUP_METADATA);
5121 void btrfs_free_block_rsv(struct btrfs_root *root,
5122 struct btrfs_block_rsv *rsv)
5126 btrfs_block_rsv_release(root, rsv, (u64)-1);
5130 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5135 int btrfs_block_rsv_add(struct btrfs_root *root,
5136 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5137 enum btrfs_reserve_flush_enum flush)
5144 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5146 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5153 int btrfs_block_rsv_check(struct btrfs_root *root,
5154 struct btrfs_block_rsv *block_rsv, int min_factor)
5162 spin_lock(&block_rsv->lock);
5163 num_bytes = div_factor(block_rsv->size, min_factor);
5164 if (block_rsv->reserved >= num_bytes)
5166 spin_unlock(&block_rsv->lock);
5171 int btrfs_block_rsv_refill(struct btrfs_root *root,
5172 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5173 enum btrfs_reserve_flush_enum flush)
5181 spin_lock(&block_rsv->lock);
5182 num_bytes = min_reserved;
5183 if (block_rsv->reserved >= num_bytes)
5186 num_bytes -= block_rsv->reserved;
5187 spin_unlock(&block_rsv->lock);
5192 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5194 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5201 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5202 struct btrfs_block_rsv *dst_rsv,
5205 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5208 void btrfs_block_rsv_release(struct btrfs_root *root,
5209 struct btrfs_block_rsv *block_rsv,
5212 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5213 if (global_rsv == block_rsv ||
5214 block_rsv->space_info != global_rsv->space_info)
5216 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5221 * helper to calculate size of global block reservation.
5222 * the desired value is sum of space used by extent tree,
5223 * checksum tree and root tree
5225 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5227 struct btrfs_space_info *sinfo;
5231 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5233 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5234 spin_lock(&sinfo->lock);
5235 data_used = sinfo->bytes_used;
5236 spin_unlock(&sinfo->lock);
5238 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5239 spin_lock(&sinfo->lock);
5240 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5242 meta_used = sinfo->bytes_used;
5243 spin_unlock(&sinfo->lock);
5245 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5247 num_bytes += div_u64(data_used + meta_used, 50);
5249 if (num_bytes * 3 > meta_used)
5250 num_bytes = div_u64(meta_used, 3);
5252 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5255 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5257 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5258 struct btrfs_space_info *sinfo = block_rsv->space_info;
5261 num_bytes = calc_global_metadata_size(fs_info);
5263 spin_lock(&sinfo->lock);
5264 spin_lock(&block_rsv->lock);
5266 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5268 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5269 sinfo->bytes_reserved + sinfo->bytes_readonly +
5270 sinfo->bytes_may_use;
5272 if (sinfo->total_bytes > num_bytes) {
5273 num_bytes = sinfo->total_bytes - num_bytes;
5274 block_rsv->reserved += num_bytes;
5275 sinfo->bytes_may_use += num_bytes;
5276 trace_btrfs_space_reservation(fs_info, "space_info",
5277 sinfo->flags, num_bytes, 1);
5280 if (block_rsv->reserved >= block_rsv->size) {
5281 num_bytes = block_rsv->reserved - block_rsv->size;
5282 sinfo->bytes_may_use -= num_bytes;
5283 trace_btrfs_space_reservation(fs_info, "space_info",
5284 sinfo->flags, num_bytes, 0);
5285 block_rsv->reserved = block_rsv->size;
5286 block_rsv->full = 1;
5289 spin_unlock(&block_rsv->lock);
5290 spin_unlock(&sinfo->lock);
5293 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5295 struct btrfs_space_info *space_info;
5297 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5298 fs_info->chunk_block_rsv.space_info = space_info;
5300 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5301 fs_info->global_block_rsv.space_info = space_info;
5302 fs_info->delalloc_block_rsv.space_info = space_info;
5303 fs_info->trans_block_rsv.space_info = space_info;
5304 fs_info->empty_block_rsv.space_info = space_info;
5305 fs_info->delayed_block_rsv.space_info = space_info;
5307 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5308 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5309 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5310 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5311 if (fs_info->quota_root)
5312 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5313 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5315 update_global_block_rsv(fs_info);
5318 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5320 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5322 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5323 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5324 WARN_ON(fs_info->trans_block_rsv.size > 0);
5325 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5326 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5327 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5328 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5329 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5332 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5333 struct btrfs_root *root)
5335 if (!trans->block_rsv)
5338 if (!trans->bytes_reserved)
5341 trace_btrfs_space_reservation(root->fs_info, "transaction",
5342 trans->transid, trans->bytes_reserved, 0);
5343 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5344 trans->bytes_reserved = 0;
5348 * To be called after all the new block groups attached to the transaction
5349 * handle have been created (btrfs_create_pending_block_groups()).
5351 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5353 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5355 if (!trans->chunk_bytes_reserved)
5358 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5360 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5361 trans->chunk_bytes_reserved);
5362 trans->chunk_bytes_reserved = 0;
5365 /* Can only return 0 or -ENOSPC */
5366 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5367 struct inode *inode)
5369 struct btrfs_root *root = BTRFS_I(inode)->root;
5370 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5371 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5374 * We need to hold space in order to delete our orphan item once we've
5375 * added it, so this takes the reservation so we can release it later
5376 * when we are truly done with the orphan item.
5378 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5379 trace_btrfs_space_reservation(root->fs_info, "orphan",
5380 btrfs_ino(inode), num_bytes, 1);
5381 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5384 void btrfs_orphan_release_metadata(struct inode *inode)
5386 struct btrfs_root *root = BTRFS_I(inode)->root;
5387 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5388 trace_btrfs_space_reservation(root->fs_info, "orphan",
5389 btrfs_ino(inode), num_bytes, 0);
5390 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5394 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5395 * root: the root of the parent directory
5396 * rsv: block reservation
5397 * items: the number of items that we need do reservation
5398 * qgroup_reserved: used to return the reserved size in qgroup
5400 * This function is used to reserve the space for snapshot/subvolume
5401 * creation and deletion. Those operations are different with the
5402 * common file/directory operations, they change two fs/file trees
5403 * and root tree, the number of items that the qgroup reserves is
5404 * different with the free space reservation. So we can not use
5405 * the space reseravtion mechanism in start_transaction().
5407 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5408 struct btrfs_block_rsv *rsv,
5410 u64 *qgroup_reserved,
5411 bool use_global_rsv)
5415 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5417 if (root->fs_info->quota_enabled) {
5418 /* One for parent inode, two for dir entries */
5419 num_bytes = 3 * root->nodesize;
5420 ret = btrfs_qgroup_reserve(root, num_bytes);
5427 *qgroup_reserved = num_bytes;
5429 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5430 rsv->space_info = __find_space_info(root->fs_info,
5431 BTRFS_BLOCK_GROUP_METADATA);
5432 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5433 BTRFS_RESERVE_FLUSH_ALL);
5435 if (ret == -ENOSPC && use_global_rsv)
5436 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5439 if (*qgroup_reserved)
5440 btrfs_qgroup_free(root, *qgroup_reserved);
5446 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5447 struct btrfs_block_rsv *rsv,
5448 u64 qgroup_reserved)
5450 btrfs_block_rsv_release(root, rsv, (u64)-1);
5454 * drop_outstanding_extent - drop an outstanding extent
5455 * @inode: the inode we're dropping the extent for
5456 * @num_bytes: the number of bytes we're relaseing.
5458 * This is called when we are freeing up an outstanding extent, either called
5459 * after an error or after an extent is written. This will return the number of
5460 * reserved extents that need to be freed. This must be called with
5461 * BTRFS_I(inode)->lock held.
5463 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5465 unsigned drop_inode_space = 0;
5466 unsigned dropped_extents = 0;
5467 unsigned num_extents = 0;
5469 num_extents = (unsigned)div64_u64(num_bytes +
5470 BTRFS_MAX_EXTENT_SIZE - 1,
5471 BTRFS_MAX_EXTENT_SIZE);
5472 ASSERT(num_extents);
5473 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5474 BTRFS_I(inode)->outstanding_extents -= num_extents;
5476 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5477 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5478 &BTRFS_I(inode)->runtime_flags))
5479 drop_inode_space = 1;
5482 * If we have more or the same amount of outsanding extents than we have
5483 * reserved then we need to leave the reserved extents count alone.
5485 if (BTRFS_I(inode)->outstanding_extents >=
5486 BTRFS_I(inode)->reserved_extents)
5487 return drop_inode_space;
5489 dropped_extents = BTRFS_I(inode)->reserved_extents -
5490 BTRFS_I(inode)->outstanding_extents;
5491 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5492 return dropped_extents + drop_inode_space;
5496 * calc_csum_metadata_size - return the amount of metada space that must be
5497 * reserved/free'd for the given bytes.
5498 * @inode: the inode we're manipulating
5499 * @num_bytes: the number of bytes in question
5500 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5502 * This adjusts the number of csum_bytes in the inode and then returns the
5503 * correct amount of metadata that must either be reserved or freed. We
5504 * calculate how many checksums we can fit into one leaf and then divide the
5505 * number of bytes that will need to be checksumed by this value to figure out
5506 * how many checksums will be required. If we are adding bytes then the number
5507 * may go up and we will return the number of additional bytes that must be
5508 * reserved. If it is going down we will return the number of bytes that must
5511 * This must be called with BTRFS_I(inode)->lock held.
5513 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5516 struct btrfs_root *root = BTRFS_I(inode)->root;
5517 u64 old_csums, num_csums;
5519 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5520 BTRFS_I(inode)->csum_bytes == 0)
5523 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5525 BTRFS_I(inode)->csum_bytes += num_bytes;
5527 BTRFS_I(inode)->csum_bytes -= num_bytes;
5528 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5530 /* No change, no need to reserve more */
5531 if (old_csums == num_csums)
5535 return btrfs_calc_trans_metadata_size(root,
5536 num_csums - old_csums);
5538 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5541 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5543 struct btrfs_root *root = BTRFS_I(inode)->root;
5544 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5547 unsigned nr_extents = 0;
5548 int extra_reserve = 0;
5549 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5551 bool delalloc_lock = true;
5555 /* If we are a free space inode we need to not flush since we will be in
5556 * the middle of a transaction commit. We also don't need the delalloc
5557 * mutex since we won't race with anybody. We need this mostly to make
5558 * lockdep shut its filthy mouth.
5560 if (btrfs_is_free_space_inode(inode)) {
5561 flush = BTRFS_RESERVE_NO_FLUSH;
5562 delalloc_lock = false;
5565 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5566 btrfs_transaction_in_commit(root->fs_info))
5567 schedule_timeout(1);
5570 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5572 num_bytes = ALIGN(num_bytes, root->sectorsize);
5574 spin_lock(&BTRFS_I(inode)->lock);
5575 nr_extents = (unsigned)div64_u64(num_bytes +
5576 BTRFS_MAX_EXTENT_SIZE - 1,
5577 BTRFS_MAX_EXTENT_SIZE);
5578 BTRFS_I(inode)->outstanding_extents += nr_extents;
5581 if (BTRFS_I(inode)->outstanding_extents >
5582 BTRFS_I(inode)->reserved_extents)
5583 nr_extents = BTRFS_I(inode)->outstanding_extents -
5584 BTRFS_I(inode)->reserved_extents;
5587 * Add an item to reserve for updating the inode when we complete the
5590 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5591 &BTRFS_I(inode)->runtime_flags)) {
5596 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5597 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5598 csum_bytes = BTRFS_I(inode)->csum_bytes;
5599 spin_unlock(&BTRFS_I(inode)->lock);
5601 if (root->fs_info->quota_enabled) {
5602 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5607 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5608 if (unlikely(ret)) {
5609 if (root->fs_info->quota_enabled)
5610 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5614 spin_lock(&BTRFS_I(inode)->lock);
5615 if (extra_reserve) {
5616 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5617 &BTRFS_I(inode)->runtime_flags);
5620 BTRFS_I(inode)->reserved_extents += nr_extents;
5621 spin_unlock(&BTRFS_I(inode)->lock);
5624 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5627 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5628 btrfs_ino(inode), to_reserve, 1);
5629 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5634 spin_lock(&BTRFS_I(inode)->lock);
5635 dropped = drop_outstanding_extent(inode, num_bytes);
5637 * If the inodes csum_bytes is the same as the original
5638 * csum_bytes then we know we haven't raced with any free()ers
5639 * so we can just reduce our inodes csum bytes and carry on.
5641 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5642 calc_csum_metadata_size(inode, num_bytes, 0);
5644 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5648 * This is tricky, but first we need to figure out how much we
5649 * free'd from any free-ers that occured during this
5650 * reservation, so we reset ->csum_bytes to the csum_bytes
5651 * before we dropped our lock, and then call the free for the
5652 * number of bytes that were freed while we were trying our
5655 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5656 BTRFS_I(inode)->csum_bytes = csum_bytes;
5657 to_free = calc_csum_metadata_size(inode, bytes, 0);
5661 * Now we need to see how much we would have freed had we not
5662 * been making this reservation and our ->csum_bytes were not
5663 * artificially inflated.
5665 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5666 bytes = csum_bytes - orig_csum_bytes;
5667 bytes = calc_csum_metadata_size(inode, bytes, 0);
5670 * Now reset ->csum_bytes to what it should be. If bytes is
5671 * more than to_free then we would have free'd more space had we
5672 * not had an artificially high ->csum_bytes, so we need to free
5673 * the remainder. If bytes is the same or less then we don't
5674 * need to do anything, the other free-ers did the correct
5677 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5678 if (bytes > to_free)
5679 to_free = bytes - to_free;
5683 spin_unlock(&BTRFS_I(inode)->lock);
5685 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5688 btrfs_block_rsv_release(root, block_rsv, to_free);
5689 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5690 btrfs_ino(inode), to_free, 0);
5693 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5698 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5699 * @inode: the inode to release the reservation for
5700 * @num_bytes: the number of bytes we're releasing
5702 * This will release the metadata reservation for an inode. This can be called
5703 * once we complete IO for a given set of bytes to release their metadata
5706 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5708 struct btrfs_root *root = BTRFS_I(inode)->root;
5712 num_bytes = ALIGN(num_bytes, root->sectorsize);
5713 spin_lock(&BTRFS_I(inode)->lock);
5714 dropped = drop_outstanding_extent(inode, num_bytes);
5717 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5718 spin_unlock(&BTRFS_I(inode)->lock);
5720 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5722 if (btrfs_test_is_dummy_root(root))
5725 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5726 btrfs_ino(inode), to_free, 0);
5728 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5733 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5734 * @inode: inode we're writing to
5735 * @num_bytes: the number of bytes we want to allocate
5737 * This will do the following things
5739 * o reserve space in the data space info for num_bytes
5740 * o reserve space in the metadata space info based on number of outstanding
5741 * extents and how much csums will be needed
5742 * o add to the inodes ->delalloc_bytes
5743 * o add it to the fs_info's delalloc inodes list.
5745 * This will return 0 for success and -ENOSPC if there is no space left.
5747 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5751 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5755 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5757 btrfs_free_reserved_data_space(inode, num_bytes);
5765 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5766 * @inode: inode we're releasing space for
5767 * @num_bytes: the number of bytes we want to free up
5769 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5770 * called in the case that we don't need the metadata AND data reservations
5771 * anymore. So if there is an error or we insert an inline extent.
5773 * This function will release the metadata space that was not used and will
5774 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5775 * list if there are no delalloc bytes left.
5777 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5779 btrfs_delalloc_release_metadata(inode, num_bytes);
5780 btrfs_free_reserved_data_space(inode, num_bytes);
5783 static int update_block_group(struct btrfs_trans_handle *trans,
5784 struct btrfs_root *root, u64 bytenr,
5785 u64 num_bytes, int alloc)
5787 struct btrfs_block_group_cache *cache = NULL;
5788 struct btrfs_fs_info *info = root->fs_info;
5789 u64 total = num_bytes;
5794 /* block accounting for super block */
5795 spin_lock(&info->delalloc_root_lock);
5796 old_val = btrfs_super_bytes_used(info->super_copy);
5798 old_val += num_bytes;
5800 old_val -= num_bytes;
5801 btrfs_set_super_bytes_used(info->super_copy, old_val);
5802 spin_unlock(&info->delalloc_root_lock);
5805 cache = btrfs_lookup_block_group(info, bytenr);
5808 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5809 BTRFS_BLOCK_GROUP_RAID1 |
5810 BTRFS_BLOCK_GROUP_RAID10))
5815 * If this block group has free space cache written out, we
5816 * need to make sure to load it if we are removing space. This
5817 * is because we need the unpinning stage to actually add the
5818 * space back to the block group, otherwise we will leak space.
5820 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5821 cache_block_group(cache, 1);
5823 byte_in_group = bytenr - cache->key.objectid;
5824 WARN_ON(byte_in_group > cache->key.offset);
5826 spin_lock(&cache->space_info->lock);
5827 spin_lock(&cache->lock);
5829 if (btrfs_test_opt(root, SPACE_CACHE) &&
5830 cache->disk_cache_state < BTRFS_DC_CLEAR)
5831 cache->disk_cache_state = BTRFS_DC_CLEAR;
5833 old_val = btrfs_block_group_used(&cache->item);
5834 num_bytes = min(total, cache->key.offset - byte_in_group);
5836 old_val += num_bytes;
5837 btrfs_set_block_group_used(&cache->item, old_val);
5838 cache->reserved -= num_bytes;
5839 cache->space_info->bytes_reserved -= num_bytes;
5840 cache->space_info->bytes_used += num_bytes;
5841 cache->space_info->disk_used += num_bytes * factor;
5842 spin_unlock(&cache->lock);
5843 spin_unlock(&cache->space_info->lock);
5845 old_val -= num_bytes;
5846 btrfs_set_block_group_used(&cache->item, old_val);
5847 cache->pinned += num_bytes;
5848 cache->space_info->bytes_pinned += num_bytes;
5849 cache->space_info->bytes_used -= num_bytes;
5850 cache->space_info->disk_used -= num_bytes * factor;
5851 spin_unlock(&cache->lock);
5852 spin_unlock(&cache->space_info->lock);
5854 set_extent_dirty(info->pinned_extents,
5855 bytenr, bytenr + num_bytes - 1,
5856 GFP_NOFS | __GFP_NOFAIL);
5858 * No longer have used bytes in this block group, queue
5862 spin_lock(&info->unused_bgs_lock);
5863 if (list_empty(&cache->bg_list)) {
5864 btrfs_get_block_group(cache);
5865 list_add_tail(&cache->bg_list,
5868 spin_unlock(&info->unused_bgs_lock);
5872 spin_lock(&trans->transaction->dirty_bgs_lock);
5873 if (list_empty(&cache->dirty_list)) {
5874 list_add_tail(&cache->dirty_list,
5875 &trans->transaction->dirty_bgs);
5876 trans->transaction->num_dirty_bgs++;
5877 btrfs_get_block_group(cache);
5879 spin_unlock(&trans->transaction->dirty_bgs_lock);
5881 btrfs_put_block_group(cache);
5883 bytenr += num_bytes;
5888 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5890 struct btrfs_block_group_cache *cache;
5893 spin_lock(&root->fs_info->block_group_cache_lock);
5894 bytenr = root->fs_info->first_logical_byte;
5895 spin_unlock(&root->fs_info->block_group_cache_lock);
5897 if (bytenr < (u64)-1)
5900 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5904 bytenr = cache->key.objectid;
5905 btrfs_put_block_group(cache);
5910 static int pin_down_extent(struct btrfs_root *root,
5911 struct btrfs_block_group_cache *cache,
5912 u64 bytenr, u64 num_bytes, int reserved)
5914 spin_lock(&cache->space_info->lock);
5915 spin_lock(&cache->lock);
5916 cache->pinned += num_bytes;
5917 cache->space_info->bytes_pinned += num_bytes;
5919 cache->reserved -= num_bytes;
5920 cache->space_info->bytes_reserved -= num_bytes;
5922 spin_unlock(&cache->lock);
5923 spin_unlock(&cache->space_info->lock);
5925 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5926 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5928 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5933 * this function must be called within transaction
5935 int btrfs_pin_extent(struct btrfs_root *root,
5936 u64 bytenr, u64 num_bytes, int reserved)
5938 struct btrfs_block_group_cache *cache;
5940 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5941 BUG_ON(!cache); /* Logic error */
5943 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5945 btrfs_put_block_group(cache);
5950 * this function must be called within transaction
5952 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5953 u64 bytenr, u64 num_bytes)
5955 struct btrfs_block_group_cache *cache;
5958 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5963 * pull in the free space cache (if any) so that our pin
5964 * removes the free space from the cache. We have load_only set
5965 * to one because the slow code to read in the free extents does check
5966 * the pinned extents.
5968 cache_block_group(cache, 1);
5970 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5972 /* remove us from the free space cache (if we're there at all) */
5973 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5974 btrfs_put_block_group(cache);
5978 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5981 struct btrfs_block_group_cache *block_group;
5982 struct btrfs_caching_control *caching_ctl;
5984 block_group = btrfs_lookup_block_group(root->fs_info, start);
5988 cache_block_group(block_group, 0);
5989 caching_ctl = get_caching_control(block_group);
5993 BUG_ON(!block_group_cache_done(block_group));
5994 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5996 mutex_lock(&caching_ctl->mutex);
5998 if (start >= caching_ctl->progress) {
5999 ret = add_excluded_extent(root, start, num_bytes);
6000 } else if (start + num_bytes <= caching_ctl->progress) {
6001 ret = btrfs_remove_free_space(block_group,
6004 num_bytes = caching_ctl->progress - start;
6005 ret = btrfs_remove_free_space(block_group,
6010 num_bytes = (start + num_bytes) -
6011 caching_ctl->progress;
6012 start = caching_ctl->progress;
6013 ret = add_excluded_extent(root, start, num_bytes);
6016 mutex_unlock(&caching_ctl->mutex);
6017 put_caching_control(caching_ctl);
6019 btrfs_put_block_group(block_group);
6023 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6024 struct extent_buffer *eb)
6026 struct btrfs_file_extent_item *item;
6027 struct btrfs_key key;
6031 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6034 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6035 btrfs_item_key_to_cpu(eb, &key, i);
6036 if (key.type != BTRFS_EXTENT_DATA_KEY)
6038 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6039 found_type = btrfs_file_extent_type(eb, item);
6040 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6042 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6044 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6045 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6046 __exclude_logged_extent(log, key.objectid, key.offset);
6053 * btrfs_update_reserved_bytes - update the block_group and space info counters
6054 * @cache: The cache we are manipulating
6055 * @num_bytes: The number of bytes in question
6056 * @reserve: One of the reservation enums
6057 * @delalloc: The blocks are allocated for the delalloc write
6059 * This is called by the allocator when it reserves space, or by somebody who is
6060 * freeing space that was never actually used on disk. For example if you
6061 * reserve some space for a new leaf in transaction A and before transaction A
6062 * commits you free that leaf, you call this with reserve set to 0 in order to
6063 * clear the reservation.
6065 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6066 * ENOSPC accounting. For data we handle the reservation through clearing the
6067 * delalloc bits in the io_tree. We have to do this since we could end up
6068 * allocating less disk space for the amount of data we have reserved in the
6069 * case of compression.
6071 * If this is a reservation and the block group has become read only we cannot
6072 * make the reservation and return -EAGAIN, otherwise this function always
6075 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6076 u64 num_bytes, int reserve, int delalloc)
6078 struct btrfs_space_info *space_info = cache->space_info;
6081 spin_lock(&space_info->lock);
6082 spin_lock(&cache->lock);
6083 if (reserve != RESERVE_FREE) {
6087 cache->reserved += num_bytes;
6088 space_info->bytes_reserved += num_bytes;
6089 if (reserve == RESERVE_ALLOC) {
6090 trace_btrfs_space_reservation(cache->fs_info,
6091 "space_info", space_info->flags,
6093 space_info->bytes_may_use -= num_bytes;
6097 cache->delalloc_bytes += num_bytes;
6101 space_info->bytes_readonly += num_bytes;
6102 cache->reserved -= num_bytes;
6103 space_info->bytes_reserved -= num_bytes;
6106 cache->delalloc_bytes -= num_bytes;
6108 spin_unlock(&cache->lock);
6109 spin_unlock(&space_info->lock);
6113 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6114 struct btrfs_root *root)
6116 struct btrfs_fs_info *fs_info = root->fs_info;
6117 struct btrfs_caching_control *next;
6118 struct btrfs_caching_control *caching_ctl;
6119 struct btrfs_block_group_cache *cache;
6121 down_write(&fs_info->commit_root_sem);
6123 list_for_each_entry_safe(caching_ctl, next,
6124 &fs_info->caching_block_groups, list) {
6125 cache = caching_ctl->block_group;
6126 if (block_group_cache_done(cache)) {
6127 cache->last_byte_to_unpin = (u64)-1;
6128 list_del_init(&caching_ctl->list);
6129 put_caching_control(caching_ctl);
6131 cache->last_byte_to_unpin = caching_ctl->progress;
6135 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6136 fs_info->pinned_extents = &fs_info->freed_extents[1];
6138 fs_info->pinned_extents = &fs_info->freed_extents[0];
6140 up_write(&fs_info->commit_root_sem);
6142 update_global_block_rsv(fs_info);
6145 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6146 const bool return_free_space)
6148 struct btrfs_fs_info *fs_info = root->fs_info;
6149 struct btrfs_block_group_cache *cache = NULL;
6150 struct btrfs_space_info *space_info;
6151 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6155 while (start <= end) {
6158 start >= cache->key.objectid + cache->key.offset) {
6160 btrfs_put_block_group(cache);
6161 cache = btrfs_lookup_block_group(fs_info, start);
6162 BUG_ON(!cache); /* Logic error */
6165 len = cache->key.objectid + cache->key.offset - start;
6166 len = min(len, end + 1 - start);
6168 if (start < cache->last_byte_to_unpin) {
6169 len = min(len, cache->last_byte_to_unpin - start);
6170 if (return_free_space)
6171 btrfs_add_free_space(cache, start, len);
6175 space_info = cache->space_info;
6177 spin_lock(&space_info->lock);
6178 spin_lock(&cache->lock);
6179 cache->pinned -= len;
6180 space_info->bytes_pinned -= len;
6181 space_info->max_extent_size = 0;
6182 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6184 space_info->bytes_readonly += len;
6187 spin_unlock(&cache->lock);
6188 if (!readonly && global_rsv->space_info == space_info) {
6189 spin_lock(&global_rsv->lock);
6190 if (!global_rsv->full) {
6191 len = min(len, global_rsv->size -
6192 global_rsv->reserved);
6193 global_rsv->reserved += len;
6194 space_info->bytes_may_use += len;
6195 if (global_rsv->reserved >= global_rsv->size)
6196 global_rsv->full = 1;
6198 spin_unlock(&global_rsv->lock);
6200 spin_unlock(&space_info->lock);
6204 btrfs_put_block_group(cache);
6208 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6209 struct btrfs_root *root)
6211 struct btrfs_fs_info *fs_info = root->fs_info;
6212 struct btrfs_block_group_cache *block_group, *tmp;
6213 struct list_head *deleted_bgs;
6214 struct extent_io_tree *unpin;
6219 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6220 unpin = &fs_info->freed_extents[1];
6222 unpin = &fs_info->freed_extents[0];
6224 while (!trans->aborted) {
6225 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6226 ret = find_first_extent_bit(unpin, 0, &start, &end,
6227 EXTENT_DIRTY, NULL);
6229 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6233 if (btrfs_test_opt(root, DISCARD))
6234 ret = btrfs_discard_extent(root, start,
6235 end + 1 - start, NULL);
6237 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6238 unpin_extent_range(root, start, end, true);
6239 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6244 * Transaction is finished. We don't need the lock anymore. We
6245 * do need to clean up the block groups in case of a transaction
6248 deleted_bgs = &trans->transaction->deleted_bgs;
6249 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6253 if (!trans->aborted)
6254 ret = btrfs_discard_extent(root,
6255 block_group->key.objectid,
6256 block_group->key.offset,
6259 list_del_init(&block_group->bg_list);
6260 btrfs_put_block_group_trimming(block_group);
6261 btrfs_put_block_group(block_group);
6264 const char *errstr = btrfs_decode_error(ret);
6266 "Discard failed while removing blockgroup: errno=%d %s\n",
6274 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6275 u64 owner, u64 root_objectid)
6277 struct btrfs_space_info *space_info;
6280 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6281 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6282 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6284 flags = BTRFS_BLOCK_GROUP_METADATA;
6286 flags = BTRFS_BLOCK_GROUP_DATA;
6289 space_info = __find_space_info(fs_info, flags);
6290 BUG_ON(!space_info); /* Logic bug */
6291 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6295 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6296 struct btrfs_root *root,
6297 struct btrfs_delayed_ref_node *node, u64 parent,
6298 u64 root_objectid, u64 owner_objectid,
6299 u64 owner_offset, int refs_to_drop,
6300 struct btrfs_delayed_extent_op *extent_op)
6302 struct btrfs_key key;
6303 struct btrfs_path *path;
6304 struct btrfs_fs_info *info = root->fs_info;
6305 struct btrfs_root *extent_root = info->extent_root;
6306 struct extent_buffer *leaf;
6307 struct btrfs_extent_item *ei;
6308 struct btrfs_extent_inline_ref *iref;
6311 int extent_slot = 0;
6312 int found_extent = 0;
6314 int no_quota = node->no_quota;
6317 u64 bytenr = node->bytenr;
6318 u64 num_bytes = node->num_bytes;
6320 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6323 if (!info->quota_enabled || !is_fstree(root_objectid))
6326 path = btrfs_alloc_path();
6331 path->leave_spinning = 1;
6333 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6334 BUG_ON(!is_data && refs_to_drop != 1);
6337 skinny_metadata = 0;
6339 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6340 bytenr, num_bytes, parent,
6341 root_objectid, owner_objectid,
6344 extent_slot = path->slots[0];
6345 while (extent_slot >= 0) {
6346 btrfs_item_key_to_cpu(path->nodes[0], &key,
6348 if (key.objectid != bytenr)
6350 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6351 key.offset == num_bytes) {
6355 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6356 key.offset == owner_objectid) {
6360 if (path->slots[0] - extent_slot > 5)
6364 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6365 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6366 if (found_extent && item_size < sizeof(*ei))
6369 if (!found_extent) {
6371 ret = remove_extent_backref(trans, extent_root, path,
6373 is_data, &last_ref);
6375 btrfs_abort_transaction(trans, extent_root, ret);
6378 btrfs_release_path(path);
6379 path->leave_spinning = 1;
6381 key.objectid = bytenr;
6382 key.type = BTRFS_EXTENT_ITEM_KEY;
6383 key.offset = num_bytes;
6385 if (!is_data && skinny_metadata) {
6386 key.type = BTRFS_METADATA_ITEM_KEY;
6387 key.offset = owner_objectid;
6390 ret = btrfs_search_slot(trans, extent_root,
6392 if (ret > 0 && skinny_metadata && path->slots[0]) {
6394 * Couldn't find our skinny metadata item,
6395 * see if we have ye olde extent item.
6398 btrfs_item_key_to_cpu(path->nodes[0], &key,
6400 if (key.objectid == bytenr &&
6401 key.type == BTRFS_EXTENT_ITEM_KEY &&
6402 key.offset == num_bytes)
6406 if (ret > 0 && skinny_metadata) {
6407 skinny_metadata = false;
6408 key.objectid = bytenr;
6409 key.type = BTRFS_EXTENT_ITEM_KEY;
6410 key.offset = num_bytes;
6411 btrfs_release_path(path);
6412 ret = btrfs_search_slot(trans, extent_root,
6417 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6420 btrfs_print_leaf(extent_root,
6424 btrfs_abort_transaction(trans, extent_root, ret);
6427 extent_slot = path->slots[0];
6429 } else if (WARN_ON(ret == -ENOENT)) {
6430 btrfs_print_leaf(extent_root, path->nodes[0]);
6432 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6433 bytenr, parent, root_objectid, owner_objectid,
6435 btrfs_abort_transaction(trans, extent_root, ret);
6438 btrfs_abort_transaction(trans, extent_root, ret);
6442 leaf = path->nodes[0];
6443 item_size = btrfs_item_size_nr(leaf, extent_slot);
6444 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6445 if (item_size < sizeof(*ei)) {
6446 BUG_ON(found_extent || extent_slot != path->slots[0]);
6447 ret = convert_extent_item_v0(trans, extent_root, path,
6450 btrfs_abort_transaction(trans, extent_root, ret);
6454 btrfs_release_path(path);
6455 path->leave_spinning = 1;
6457 key.objectid = bytenr;
6458 key.type = BTRFS_EXTENT_ITEM_KEY;
6459 key.offset = num_bytes;
6461 ret = btrfs_search_slot(trans, extent_root, &key, path,
6464 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6466 btrfs_print_leaf(extent_root, path->nodes[0]);
6469 btrfs_abort_transaction(trans, extent_root, ret);
6473 extent_slot = path->slots[0];
6474 leaf = path->nodes[0];
6475 item_size = btrfs_item_size_nr(leaf, extent_slot);
6478 BUG_ON(item_size < sizeof(*ei));
6479 ei = btrfs_item_ptr(leaf, extent_slot,
6480 struct btrfs_extent_item);
6481 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6482 key.type == BTRFS_EXTENT_ITEM_KEY) {
6483 struct btrfs_tree_block_info *bi;
6484 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6485 bi = (struct btrfs_tree_block_info *)(ei + 1);
6486 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6489 refs = btrfs_extent_refs(leaf, ei);
6490 if (refs < refs_to_drop) {
6491 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6492 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6494 btrfs_abort_transaction(trans, extent_root, ret);
6497 refs -= refs_to_drop;
6501 __run_delayed_extent_op(extent_op, leaf, ei);
6503 * In the case of inline back ref, reference count will
6504 * be updated by remove_extent_backref
6507 BUG_ON(!found_extent);
6509 btrfs_set_extent_refs(leaf, ei, refs);
6510 btrfs_mark_buffer_dirty(leaf);
6513 ret = remove_extent_backref(trans, extent_root, path,
6515 is_data, &last_ref);
6517 btrfs_abort_transaction(trans, extent_root, ret);
6521 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6525 BUG_ON(is_data && refs_to_drop !=
6526 extent_data_ref_count(path, iref));
6528 BUG_ON(path->slots[0] != extent_slot);
6530 BUG_ON(path->slots[0] != extent_slot + 1);
6531 path->slots[0] = extent_slot;
6537 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6540 btrfs_abort_transaction(trans, extent_root, ret);
6543 btrfs_release_path(path);
6546 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6548 btrfs_abort_transaction(trans, extent_root, ret);
6553 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6555 btrfs_abort_transaction(trans, extent_root, ret);
6559 btrfs_release_path(path);
6562 btrfs_free_path(path);
6567 * when we free an block, it is possible (and likely) that we free the last
6568 * delayed ref for that extent as well. This searches the delayed ref tree for
6569 * a given extent, and if there are no other delayed refs to be processed, it
6570 * removes it from the tree.
6572 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6573 struct btrfs_root *root, u64 bytenr)
6575 struct btrfs_delayed_ref_head *head;
6576 struct btrfs_delayed_ref_root *delayed_refs;
6579 delayed_refs = &trans->transaction->delayed_refs;
6580 spin_lock(&delayed_refs->lock);
6581 head = btrfs_find_delayed_ref_head(trans, bytenr);
6583 goto out_delayed_unlock;
6585 spin_lock(&head->lock);
6586 if (!list_empty(&head->ref_list))
6589 if (head->extent_op) {
6590 if (!head->must_insert_reserved)
6592 btrfs_free_delayed_extent_op(head->extent_op);
6593 head->extent_op = NULL;
6597 * waiting for the lock here would deadlock. If someone else has it
6598 * locked they are already in the process of dropping it anyway
6600 if (!mutex_trylock(&head->mutex))
6604 * at this point we have a head with no other entries. Go
6605 * ahead and process it.
6607 head->node.in_tree = 0;
6608 rb_erase(&head->href_node, &delayed_refs->href_root);
6610 atomic_dec(&delayed_refs->num_entries);
6613 * we don't take a ref on the node because we're removing it from the
6614 * tree, so we just steal the ref the tree was holding.
6616 delayed_refs->num_heads--;
6617 if (head->processing == 0)
6618 delayed_refs->num_heads_ready--;
6619 head->processing = 0;
6620 spin_unlock(&head->lock);
6621 spin_unlock(&delayed_refs->lock);
6623 BUG_ON(head->extent_op);
6624 if (head->must_insert_reserved)
6627 mutex_unlock(&head->mutex);
6628 btrfs_put_delayed_ref(&head->node);
6631 spin_unlock(&head->lock);
6634 spin_unlock(&delayed_refs->lock);
6638 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6639 struct btrfs_root *root,
6640 struct extent_buffer *buf,
6641 u64 parent, int last_ref)
6646 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6647 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6648 buf->start, buf->len,
6649 parent, root->root_key.objectid,
6650 btrfs_header_level(buf),
6651 BTRFS_DROP_DELAYED_REF, NULL, 0);
6652 BUG_ON(ret); /* -ENOMEM */
6658 if (btrfs_header_generation(buf) == trans->transid) {
6659 struct btrfs_block_group_cache *cache;
6661 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6662 ret = check_ref_cleanup(trans, root, buf->start);
6667 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6669 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6670 pin_down_extent(root, cache, buf->start, buf->len, 1);
6671 btrfs_put_block_group(cache);
6675 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6677 btrfs_add_free_space(cache, buf->start, buf->len);
6678 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6679 btrfs_put_block_group(cache);
6680 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6685 add_pinned_bytes(root->fs_info, buf->len,
6686 btrfs_header_level(buf),
6687 root->root_key.objectid);
6690 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6693 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6696 /* Can return -ENOMEM */
6697 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6698 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6699 u64 owner, u64 offset, int no_quota)
6702 struct btrfs_fs_info *fs_info = root->fs_info;
6704 if (btrfs_test_is_dummy_root(root))
6707 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6710 * tree log blocks never actually go into the extent allocation
6711 * tree, just update pinning info and exit early.
6713 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6714 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6715 /* unlocks the pinned mutex */
6716 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6718 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6719 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6721 parent, root_objectid, (int)owner,
6722 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6724 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6726 parent, root_objectid, owner,
6727 offset, BTRFS_DROP_DELAYED_REF,
6734 * when we wait for progress in the block group caching, its because
6735 * our allocation attempt failed at least once. So, we must sleep
6736 * and let some progress happen before we try again.
6738 * This function will sleep at least once waiting for new free space to
6739 * show up, and then it will check the block group free space numbers
6740 * for our min num_bytes. Another option is to have it go ahead
6741 * and look in the rbtree for a free extent of a given size, but this
6744 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6745 * any of the information in this block group.
6747 static noinline void
6748 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6751 struct btrfs_caching_control *caching_ctl;
6753 caching_ctl = get_caching_control(cache);
6757 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6758 (cache->free_space_ctl->free_space >= num_bytes));
6760 put_caching_control(caching_ctl);
6764 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6766 struct btrfs_caching_control *caching_ctl;
6769 caching_ctl = get_caching_control(cache);
6771 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6773 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6774 if (cache->cached == BTRFS_CACHE_ERROR)
6776 put_caching_control(caching_ctl);
6780 int __get_raid_index(u64 flags)
6782 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6783 return BTRFS_RAID_RAID10;
6784 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6785 return BTRFS_RAID_RAID1;
6786 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6787 return BTRFS_RAID_DUP;
6788 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6789 return BTRFS_RAID_RAID0;
6790 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6791 return BTRFS_RAID_RAID5;
6792 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6793 return BTRFS_RAID_RAID6;
6795 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6798 int get_block_group_index(struct btrfs_block_group_cache *cache)
6800 return __get_raid_index(cache->flags);
6803 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6804 [BTRFS_RAID_RAID10] = "raid10",
6805 [BTRFS_RAID_RAID1] = "raid1",
6806 [BTRFS_RAID_DUP] = "dup",
6807 [BTRFS_RAID_RAID0] = "raid0",
6808 [BTRFS_RAID_SINGLE] = "single",
6809 [BTRFS_RAID_RAID5] = "raid5",
6810 [BTRFS_RAID_RAID6] = "raid6",
6813 static const char *get_raid_name(enum btrfs_raid_types type)
6815 if (type >= BTRFS_NR_RAID_TYPES)
6818 return btrfs_raid_type_names[type];
6821 enum btrfs_loop_type {
6822 LOOP_CACHING_NOWAIT = 0,
6823 LOOP_CACHING_WAIT = 1,
6824 LOOP_ALLOC_CHUNK = 2,
6825 LOOP_NO_EMPTY_SIZE = 3,
6829 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6833 down_read(&cache->data_rwsem);
6837 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6840 btrfs_get_block_group(cache);
6842 down_read(&cache->data_rwsem);
6845 static struct btrfs_block_group_cache *
6846 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6847 struct btrfs_free_cluster *cluster,
6850 struct btrfs_block_group_cache *used_bg;
6851 bool locked = false;
6853 spin_lock(&cluster->refill_lock);
6855 if (used_bg == cluster->block_group)
6858 up_read(&used_bg->data_rwsem);
6859 btrfs_put_block_group(used_bg);
6862 used_bg = cluster->block_group;
6866 if (used_bg == block_group)
6869 btrfs_get_block_group(used_bg);
6874 if (down_read_trylock(&used_bg->data_rwsem))
6877 spin_unlock(&cluster->refill_lock);
6878 down_read(&used_bg->data_rwsem);
6884 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6888 up_read(&cache->data_rwsem);
6889 btrfs_put_block_group(cache);
6893 * walks the btree of allocated extents and find a hole of a given size.
6894 * The key ins is changed to record the hole:
6895 * ins->objectid == start position
6896 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6897 * ins->offset == the size of the hole.
6898 * Any available blocks before search_start are skipped.
6900 * If there is no suitable free space, we will record the max size of
6901 * the free space extent currently.
6903 static noinline int find_free_extent(struct btrfs_root *orig_root,
6904 u64 num_bytes, u64 empty_size,
6905 u64 hint_byte, struct btrfs_key *ins,
6906 u64 flags, int delalloc)
6909 struct btrfs_root *root = orig_root->fs_info->extent_root;
6910 struct btrfs_free_cluster *last_ptr = NULL;
6911 struct btrfs_block_group_cache *block_group = NULL;
6912 u64 search_start = 0;
6913 u64 max_extent_size = 0;
6914 int empty_cluster = 2 * 1024 * 1024;
6915 struct btrfs_space_info *space_info;
6917 int index = __get_raid_index(flags);
6918 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6919 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6920 bool failed_cluster_refill = false;
6921 bool failed_alloc = false;
6922 bool use_cluster = true;
6923 bool have_caching_bg = false;
6924 bool full_search = false;
6926 WARN_ON(num_bytes < root->sectorsize);
6927 ins->type = BTRFS_EXTENT_ITEM_KEY;
6931 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6933 space_info = __find_space_info(root->fs_info, flags);
6935 btrfs_err(root->fs_info, "No space info for %llu", flags);
6940 * If our free space is heavily fragmented we may not be able to make
6941 * big contiguous allocations, so instead of doing the expensive search
6942 * for free space, simply return ENOSPC with our max_extent_size so we
6943 * can go ahead and search for a more manageable chunk.
6945 * If our max_extent_size is large enough for our allocation simply
6946 * disable clustering since we will likely not be able to find enough
6947 * space to create a cluster and induce latency trying.
6949 if (unlikely(space_info->max_extent_size)) {
6950 spin_lock(&space_info->lock);
6951 if (space_info->max_extent_size &&
6952 num_bytes > space_info->max_extent_size) {
6953 ins->offset = space_info->max_extent_size;
6954 spin_unlock(&space_info->lock);
6956 } else if (space_info->max_extent_size) {
6957 use_cluster = false;
6959 spin_unlock(&space_info->lock);
6963 * If the space info is for both data and metadata it means we have a
6964 * small filesystem and we can't use the clustering stuff.
6966 if (btrfs_mixed_space_info(space_info))
6967 use_cluster = false;
6969 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6970 last_ptr = &root->fs_info->meta_alloc_cluster;
6971 if (!btrfs_test_opt(root, SSD))
6972 empty_cluster = 64 * 1024;
6975 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6976 btrfs_test_opt(root, SSD)) {
6977 last_ptr = &root->fs_info->data_alloc_cluster;
6981 spin_lock(&last_ptr->lock);
6982 if (last_ptr->block_group)
6983 hint_byte = last_ptr->window_start;
6984 spin_unlock(&last_ptr->lock);
6987 search_start = max(search_start, first_logical_byte(root, 0));
6988 search_start = max(search_start, hint_byte);
6993 if (search_start == hint_byte) {
6994 block_group = btrfs_lookup_block_group(root->fs_info,
6997 * we don't want to use the block group if it doesn't match our
6998 * allocation bits, or if its not cached.
7000 * However if we are re-searching with an ideal block group
7001 * picked out then we don't care that the block group is cached.
7003 if (block_group && block_group_bits(block_group, flags) &&
7004 block_group->cached != BTRFS_CACHE_NO) {
7005 down_read(&space_info->groups_sem);
7006 if (list_empty(&block_group->list) ||
7009 * someone is removing this block group,
7010 * we can't jump into the have_block_group
7011 * target because our list pointers are not
7014 btrfs_put_block_group(block_group);
7015 up_read(&space_info->groups_sem);
7017 index = get_block_group_index(block_group);
7018 btrfs_lock_block_group(block_group, delalloc);
7019 goto have_block_group;
7021 } else if (block_group) {
7022 btrfs_put_block_group(block_group);
7026 have_caching_bg = false;
7027 if (index == 0 || index == __get_raid_index(flags))
7029 down_read(&space_info->groups_sem);
7030 list_for_each_entry(block_group, &space_info->block_groups[index],
7035 btrfs_grab_block_group(block_group, delalloc);
7036 search_start = block_group->key.objectid;
7039 * this can happen if we end up cycling through all the
7040 * raid types, but we want to make sure we only allocate
7041 * for the proper type.
7043 if (!block_group_bits(block_group, flags)) {
7044 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7045 BTRFS_BLOCK_GROUP_RAID1 |
7046 BTRFS_BLOCK_GROUP_RAID5 |
7047 BTRFS_BLOCK_GROUP_RAID6 |
7048 BTRFS_BLOCK_GROUP_RAID10;
7051 * if they asked for extra copies and this block group
7052 * doesn't provide them, bail. This does allow us to
7053 * fill raid0 from raid1.
7055 if ((flags & extra) && !(block_group->flags & extra))
7060 cached = block_group_cache_done(block_group);
7061 if (unlikely(!cached)) {
7062 have_caching_bg = true;
7063 ret = cache_block_group(block_group, 0);
7068 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7070 if (unlikely(block_group->ro))
7074 * Ok we want to try and use the cluster allocator, so
7078 struct btrfs_block_group_cache *used_block_group;
7079 unsigned long aligned_cluster;
7081 * the refill lock keeps out other
7082 * people trying to start a new cluster
7084 used_block_group = btrfs_lock_cluster(block_group,
7087 if (!used_block_group)
7088 goto refill_cluster;
7090 if (used_block_group != block_group &&
7091 (used_block_group->ro ||
7092 !block_group_bits(used_block_group, flags)))
7093 goto release_cluster;
7095 offset = btrfs_alloc_from_cluster(used_block_group,
7098 used_block_group->key.objectid,
7101 /* we have a block, we're done */
7102 spin_unlock(&last_ptr->refill_lock);
7103 trace_btrfs_reserve_extent_cluster(root,
7105 search_start, num_bytes);
7106 if (used_block_group != block_group) {
7107 btrfs_release_block_group(block_group,
7109 block_group = used_block_group;
7114 WARN_ON(last_ptr->block_group != used_block_group);
7116 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7117 * set up a new clusters, so lets just skip it
7118 * and let the allocator find whatever block
7119 * it can find. If we reach this point, we
7120 * will have tried the cluster allocator
7121 * plenty of times and not have found
7122 * anything, so we are likely way too
7123 * fragmented for the clustering stuff to find
7126 * However, if the cluster is taken from the
7127 * current block group, release the cluster
7128 * first, so that we stand a better chance of
7129 * succeeding in the unclustered
7131 if (loop >= LOOP_NO_EMPTY_SIZE &&
7132 used_block_group != block_group) {
7133 spin_unlock(&last_ptr->refill_lock);
7134 btrfs_release_block_group(used_block_group,
7136 goto unclustered_alloc;
7140 * this cluster didn't work out, free it and
7143 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7145 if (used_block_group != block_group)
7146 btrfs_release_block_group(used_block_group,
7149 if (loop >= LOOP_NO_EMPTY_SIZE) {
7150 spin_unlock(&last_ptr->refill_lock);
7151 goto unclustered_alloc;
7154 aligned_cluster = max_t(unsigned long,
7155 empty_cluster + empty_size,
7156 block_group->full_stripe_len);
7158 /* allocate a cluster in this block group */
7159 ret = btrfs_find_space_cluster(root, block_group,
7160 last_ptr, search_start,
7165 * now pull our allocation out of this
7168 offset = btrfs_alloc_from_cluster(block_group,
7174 /* we found one, proceed */
7175 spin_unlock(&last_ptr->refill_lock);
7176 trace_btrfs_reserve_extent_cluster(root,
7177 block_group, search_start,
7181 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7182 && !failed_cluster_refill) {
7183 spin_unlock(&last_ptr->refill_lock);
7185 failed_cluster_refill = true;
7186 wait_block_group_cache_progress(block_group,
7187 num_bytes + empty_cluster + empty_size);
7188 goto have_block_group;
7192 * at this point we either didn't find a cluster
7193 * or we weren't able to allocate a block from our
7194 * cluster. Free the cluster we've been trying
7195 * to use, and go to the next block group
7197 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7198 spin_unlock(&last_ptr->refill_lock);
7203 spin_lock(&block_group->free_space_ctl->tree_lock);
7205 block_group->free_space_ctl->free_space <
7206 num_bytes + empty_cluster + empty_size) {
7207 if (block_group->free_space_ctl->free_space >
7210 block_group->free_space_ctl->free_space;
7211 spin_unlock(&block_group->free_space_ctl->tree_lock);
7214 spin_unlock(&block_group->free_space_ctl->tree_lock);
7216 offset = btrfs_find_space_for_alloc(block_group, search_start,
7217 num_bytes, empty_size,
7220 * If we didn't find a chunk, and we haven't failed on this
7221 * block group before, and this block group is in the middle of
7222 * caching and we are ok with waiting, then go ahead and wait
7223 * for progress to be made, and set failed_alloc to true.
7225 * If failed_alloc is true then we've already waited on this
7226 * block group once and should move on to the next block group.
7228 if (!offset && !failed_alloc && !cached &&
7229 loop > LOOP_CACHING_NOWAIT) {
7230 wait_block_group_cache_progress(block_group,
7231 num_bytes + empty_size);
7232 failed_alloc = true;
7233 goto have_block_group;
7234 } else if (!offset) {
7238 search_start = ALIGN(offset, root->stripesize);
7240 /* move on to the next group */
7241 if (search_start + num_bytes >
7242 block_group->key.objectid + block_group->key.offset) {
7243 btrfs_add_free_space(block_group, offset, num_bytes);
7247 if (offset < search_start)
7248 btrfs_add_free_space(block_group, offset,
7249 search_start - offset);
7250 BUG_ON(offset > search_start);
7252 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7253 alloc_type, delalloc);
7254 if (ret == -EAGAIN) {
7255 btrfs_add_free_space(block_group, offset, num_bytes);
7259 /* we are all good, lets return */
7260 ins->objectid = search_start;
7261 ins->offset = num_bytes;
7263 trace_btrfs_reserve_extent(orig_root, block_group,
7264 search_start, num_bytes);
7265 btrfs_release_block_group(block_group, delalloc);
7268 failed_cluster_refill = false;
7269 failed_alloc = false;
7270 BUG_ON(index != get_block_group_index(block_group));
7271 btrfs_release_block_group(block_group, delalloc);
7273 up_read(&space_info->groups_sem);
7275 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7278 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7282 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7283 * caching kthreads as we move along
7284 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7285 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7286 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7289 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7291 if (loop == LOOP_CACHING_NOWAIT) {
7293 * We want to skip the LOOP_CACHING_WAIT step if we
7294 * don't have any unached bgs and we've alrelady done a
7295 * full search through.
7297 if (have_caching_bg || !full_search)
7298 loop = LOOP_CACHING_WAIT;
7300 loop = LOOP_ALLOC_CHUNK;
7305 if (loop == LOOP_ALLOC_CHUNK) {
7306 struct btrfs_trans_handle *trans;
7309 trans = current->journal_info;
7313 trans = btrfs_join_transaction(root);
7315 if (IS_ERR(trans)) {
7316 ret = PTR_ERR(trans);
7320 ret = do_chunk_alloc(trans, root, flags,
7324 * If we can't allocate a new chunk we've already looped
7325 * through at least once, move on to the NO_EMPTY_SIZE
7329 loop = LOOP_NO_EMPTY_SIZE;
7332 * Do not bail out on ENOSPC since we
7333 * can do more things.
7335 if (ret < 0 && ret != -ENOSPC)
7336 btrfs_abort_transaction(trans,
7341 btrfs_end_transaction(trans, root);
7346 if (loop == LOOP_NO_EMPTY_SIZE) {
7348 * Don't loop again if we already have no empty_size and
7351 if (empty_size == 0 &&
7352 empty_cluster == 0) {
7361 } else if (!ins->objectid) {
7363 } else if (ins->objectid) {
7367 if (ret == -ENOSPC) {
7368 spin_lock(&space_info->lock);
7369 space_info->max_extent_size = max_extent_size;
7370 spin_unlock(&space_info->lock);
7371 ins->offset = max_extent_size;
7376 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7377 int dump_block_groups)
7379 struct btrfs_block_group_cache *cache;
7382 spin_lock(&info->lock);
7383 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7385 info->total_bytes - info->bytes_used - info->bytes_pinned -
7386 info->bytes_reserved - info->bytes_readonly,
7387 (info->full) ? "" : "not ");
7388 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7389 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7390 info->total_bytes, info->bytes_used, info->bytes_pinned,
7391 info->bytes_reserved, info->bytes_may_use,
7392 info->bytes_readonly);
7393 spin_unlock(&info->lock);
7395 if (!dump_block_groups)
7398 down_read(&info->groups_sem);
7400 list_for_each_entry(cache, &info->block_groups[index], list) {
7401 spin_lock(&cache->lock);
7402 printk(KERN_INFO "BTRFS: "
7403 "block group %llu has %llu bytes, "
7404 "%llu used %llu pinned %llu reserved %s\n",
7405 cache->key.objectid, cache->key.offset,
7406 btrfs_block_group_used(&cache->item), cache->pinned,
7407 cache->reserved, cache->ro ? "[readonly]" : "");
7408 btrfs_dump_free_space(cache, bytes);
7409 spin_unlock(&cache->lock);
7411 if (++index < BTRFS_NR_RAID_TYPES)
7413 up_read(&info->groups_sem);
7416 int btrfs_reserve_extent(struct btrfs_root *root,
7417 u64 num_bytes, u64 min_alloc_size,
7418 u64 empty_size, u64 hint_byte,
7419 struct btrfs_key *ins, int is_data, int delalloc)
7421 bool final_tried = num_bytes == min_alloc_size;
7425 flags = btrfs_get_alloc_profile(root, is_data);
7427 WARN_ON(num_bytes < root->sectorsize);
7428 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7431 if (ret == -ENOSPC) {
7432 if (!final_tried && ins->offset) {
7433 num_bytes = min(num_bytes >> 1, ins->offset);
7434 num_bytes = round_down(num_bytes, root->sectorsize);
7435 num_bytes = max(num_bytes, min_alloc_size);
7436 if (num_bytes == min_alloc_size)
7439 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7440 struct btrfs_space_info *sinfo;
7442 sinfo = __find_space_info(root->fs_info, flags);
7443 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7446 dump_space_info(sinfo, num_bytes, 1);
7453 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7455 int pin, int delalloc)
7457 struct btrfs_block_group_cache *cache;
7460 cache = btrfs_lookup_block_group(root->fs_info, start);
7462 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7468 pin_down_extent(root, cache, start, len, 1);
7470 if (btrfs_test_opt(root, DISCARD))
7471 ret = btrfs_discard_extent(root, start, len, NULL);
7472 btrfs_add_free_space(cache, start, len);
7473 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7476 btrfs_put_block_group(cache);
7478 trace_btrfs_reserved_extent_free(root, start, len);
7483 int btrfs_free_reserved_extent(struct btrfs_root *root,
7484 u64 start, u64 len, int delalloc)
7486 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7489 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7492 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7495 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7496 struct btrfs_root *root,
7497 u64 parent, u64 root_objectid,
7498 u64 flags, u64 owner, u64 offset,
7499 struct btrfs_key *ins, int ref_mod)
7502 struct btrfs_fs_info *fs_info = root->fs_info;
7503 struct btrfs_extent_item *extent_item;
7504 struct btrfs_extent_inline_ref *iref;
7505 struct btrfs_path *path;
7506 struct extent_buffer *leaf;
7511 type = BTRFS_SHARED_DATA_REF_KEY;
7513 type = BTRFS_EXTENT_DATA_REF_KEY;
7515 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7517 path = btrfs_alloc_path();
7521 path->leave_spinning = 1;
7522 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7525 btrfs_free_path(path);
7529 leaf = path->nodes[0];
7530 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7531 struct btrfs_extent_item);
7532 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7533 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7534 btrfs_set_extent_flags(leaf, extent_item,
7535 flags | BTRFS_EXTENT_FLAG_DATA);
7537 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7538 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7540 struct btrfs_shared_data_ref *ref;
7541 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7542 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7543 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7545 struct btrfs_extent_data_ref *ref;
7546 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7547 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7548 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7549 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7550 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7553 btrfs_mark_buffer_dirty(path->nodes[0]);
7554 btrfs_free_path(path);
7556 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7557 if (ret) { /* -ENOENT, logic error */
7558 btrfs_err(fs_info, "update block group failed for %llu %llu",
7559 ins->objectid, ins->offset);
7562 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7566 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7567 struct btrfs_root *root,
7568 u64 parent, u64 root_objectid,
7569 u64 flags, struct btrfs_disk_key *key,
7570 int level, struct btrfs_key *ins,
7574 struct btrfs_fs_info *fs_info = root->fs_info;
7575 struct btrfs_extent_item *extent_item;
7576 struct btrfs_tree_block_info *block_info;
7577 struct btrfs_extent_inline_ref *iref;
7578 struct btrfs_path *path;
7579 struct extent_buffer *leaf;
7580 u32 size = sizeof(*extent_item) + sizeof(*iref);
7581 u64 num_bytes = ins->offset;
7582 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7585 if (!skinny_metadata)
7586 size += sizeof(*block_info);
7588 path = btrfs_alloc_path();
7590 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7595 path->leave_spinning = 1;
7596 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7599 btrfs_free_path(path);
7600 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7605 leaf = path->nodes[0];
7606 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7607 struct btrfs_extent_item);
7608 btrfs_set_extent_refs(leaf, extent_item, 1);
7609 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7610 btrfs_set_extent_flags(leaf, extent_item,
7611 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7613 if (skinny_metadata) {
7614 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7615 num_bytes = root->nodesize;
7617 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7618 btrfs_set_tree_block_key(leaf, block_info, key);
7619 btrfs_set_tree_block_level(leaf, block_info, level);
7620 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7624 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7625 btrfs_set_extent_inline_ref_type(leaf, iref,
7626 BTRFS_SHARED_BLOCK_REF_KEY);
7627 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7629 btrfs_set_extent_inline_ref_type(leaf, iref,
7630 BTRFS_TREE_BLOCK_REF_KEY);
7631 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7634 btrfs_mark_buffer_dirty(leaf);
7635 btrfs_free_path(path);
7637 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7639 if (ret) { /* -ENOENT, logic error */
7640 btrfs_err(fs_info, "update block group failed for %llu %llu",
7641 ins->objectid, ins->offset);
7645 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7649 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7650 struct btrfs_root *root,
7651 u64 root_objectid, u64 owner,
7652 u64 offset, struct btrfs_key *ins)
7656 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7658 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7660 root_objectid, owner, offset,
7661 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7666 * this is used by the tree logging recovery code. It records that
7667 * an extent has been allocated and makes sure to clear the free
7668 * space cache bits as well
7670 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7671 struct btrfs_root *root,
7672 u64 root_objectid, u64 owner, u64 offset,
7673 struct btrfs_key *ins)
7676 struct btrfs_block_group_cache *block_group;
7679 * Mixed block groups will exclude before processing the log so we only
7680 * need to do the exlude dance if this fs isn't mixed.
7682 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7683 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7688 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7692 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7693 RESERVE_ALLOC_NO_ACCOUNT, 0);
7694 BUG_ON(ret); /* logic error */
7695 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7696 0, owner, offset, ins, 1);
7697 btrfs_put_block_group(block_group);
7701 static struct extent_buffer *
7702 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7703 u64 bytenr, int level)
7705 struct extent_buffer *buf;
7707 buf = btrfs_find_create_tree_block(root, bytenr);
7709 return ERR_PTR(-ENOMEM);
7710 btrfs_set_header_generation(buf, trans->transid);
7711 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7712 btrfs_tree_lock(buf);
7713 clean_tree_block(trans, root->fs_info, buf);
7714 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7716 btrfs_set_lock_blocking(buf);
7717 btrfs_set_buffer_uptodate(buf);
7719 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7720 buf->log_index = root->log_transid % 2;
7722 * we allow two log transactions at a time, use different
7723 * EXENT bit to differentiate dirty pages.
7725 if (buf->log_index == 0)
7726 set_extent_dirty(&root->dirty_log_pages, buf->start,
7727 buf->start + buf->len - 1, GFP_NOFS);
7729 set_extent_new(&root->dirty_log_pages, buf->start,
7730 buf->start + buf->len - 1, GFP_NOFS);
7732 buf->log_index = -1;
7733 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7734 buf->start + buf->len - 1, GFP_NOFS);
7736 trans->blocks_used++;
7737 /* this returns a buffer locked for blocking */
7741 static struct btrfs_block_rsv *
7742 use_block_rsv(struct btrfs_trans_handle *trans,
7743 struct btrfs_root *root, u32 blocksize)
7745 struct btrfs_block_rsv *block_rsv;
7746 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7748 bool global_updated = false;
7750 block_rsv = get_block_rsv(trans, root);
7752 if (unlikely(block_rsv->size == 0))
7755 ret = block_rsv_use_bytes(block_rsv, blocksize);
7759 if (block_rsv->failfast)
7760 return ERR_PTR(ret);
7762 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7763 global_updated = true;
7764 update_global_block_rsv(root->fs_info);
7768 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7769 static DEFINE_RATELIMIT_STATE(_rs,
7770 DEFAULT_RATELIMIT_INTERVAL * 10,
7771 /*DEFAULT_RATELIMIT_BURST*/ 1);
7772 if (__ratelimit(&_rs))
7774 "BTRFS: block rsv returned %d\n", ret);
7777 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7778 BTRFS_RESERVE_NO_FLUSH);
7782 * If we couldn't reserve metadata bytes try and use some from
7783 * the global reserve if its space type is the same as the global
7786 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7787 block_rsv->space_info == global_rsv->space_info) {
7788 ret = block_rsv_use_bytes(global_rsv, blocksize);
7792 return ERR_PTR(ret);
7795 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7796 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7798 block_rsv_add_bytes(block_rsv, blocksize, 0);
7799 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7803 * finds a free extent and does all the dirty work required for allocation
7804 * returns the tree buffer or an ERR_PTR on error.
7806 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7807 struct btrfs_root *root,
7808 u64 parent, u64 root_objectid,
7809 struct btrfs_disk_key *key, int level,
7810 u64 hint, u64 empty_size)
7812 struct btrfs_key ins;
7813 struct btrfs_block_rsv *block_rsv;
7814 struct extent_buffer *buf;
7815 struct btrfs_delayed_extent_op *extent_op;
7818 u32 blocksize = root->nodesize;
7819 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7822 if (btrfs_test_is_dummy_root(root)) {
7823 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7826 root->alloc_bytenr += blocksize;
7830 block_rsv = use_block_rsv(trans, root, blocksize);
7831 if (IS_ERR(block_rsv))
7832 return ERR_CAST(block_rsv);
7834 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7835 empty_size, hint, &ins, 0, 0);
7839 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7842 goto out_free_reserved;
7845 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7847 parent = ins.objectid;
7848 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7852 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7853 extent_op = btrfs_alloc_delayed_extent_op();
7859 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7861 memset(&extent_op->key, 0, sizeof(extent_op->key));
7862 extent_op->flags_to_set = flags;
7863 if (skinny_metadata)
7864 extent_op->update_key = 0;
7866 extent_op->update_key = 1;
7867 extent_op->update_flags = 1;
7868 extent_op->is_data = 0;
7869 extent_op->level = level;
7871 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7872 ins.objectid, ins.offset,
7873 parent, root_objectid, level,
7874 BTRFS_ADD_DELAYED_EXTENT,
7877 goto out_free_delayed;
7882 btrfs_free_delayed_extent_op(extent_op);
7884 free_extent_buffer(buf);
7886 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7888 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7889 return ERR_PTR(ret);
7892 struct walk_control {
7893 u64 refs[BTRFS_MAX_LEVEL];
7894 u64 flags[BTRFS_MAX_LEVEL];
7895 struct btrfs_key update_progress;
7906 #define DROP_REFERENCE 1
7907 #define UPDATE_BACKREF 2
7909 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7910 struct btrfs_root *root,
7911 struct walk_control *wc,
7912 struct btrfs_path *path)
7920 struct btrfs_key key;
7921 struct extent_buffer *eb;
7926 if (path->slots[wc->level] < wc->reada_slot) {
7927 wc->reada_count = wc->reada_count * 2 / 3;
7928 wc->reada_count = max(wc->reada_count, 2);
7930 wc->reada_count = wc->reada_count * 3 / 2;
7931 wc->reada_count = min_t(int, wc->reada_count,
7932 BTRFS_NODEPTRS_PER_BLOCK(root));
7935 eb = path->nodes[wc->level];
7936 nritems = btrfs_header_nritems(eb);
7937 blocksize = root->nodesize;
7939 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7940 if (nread >= wc->reada_count)
7944 bytenr = btrfs_node_blockptr(eb, slot);
7945 generation = btrfs_node_ptr_generation(eb, slot);
7947 if (slot == path->slots[wc->level])
7950 if (wc->stage == UPDATE_BACKREF &&
7951 generation <= root->root_key.offset)
7954 /* We don't lock the tree block, it's OK to be racy here */
7955 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7956 wc->level - 1, 1, &refs,
7958 /* We don't care about errors in readahead. */
7963 if (wc->stage == DROP_REFERENCE) {
7967 if (wc->level == 1 &&
7968 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7970 if (!wc->update_ref ||
7971 generation <= root->root_key.offset)
7973 btrfs_node_key_to_cpu(eb, &key, slot);
7974 ret = btrfs_comp_cpu_keys(&key,
7975 &wc->update_progress);
7979 if (wc->level == 1 &&
7980 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7984 readahead_tree_block(root, bytenr);
7987 wc->reada_slot = slot;
7991 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7992 * for later qgroup accounting.
7994 * Current, this function does nothing.
7996 static int account_leaf_items(struct btrfs_trans_handle *trans,
7997 struct btrfs_root *root,
7998 struct extent_buffer *eb)
8000 int nr = btrfs_header_nritems(eb);
8002 struct btrfs_key key;
8003 struct btrfs_file_extent_item *fi;
8004 u64 bytenr, num_bytes;
8006 for (i = 0; i < nr; i++) {
8007 btrfs_item_key_to_cpu(eb, &key, i);
8009 if (key.type != BTRFS_EXTENT_DATA_KEY)
8012 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8013 /* filter out non qgroup-accountable extents */
8014 extent_type = btrfs_file_extent_type(eb, fi);
8016 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8019 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8023 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8029 * Walk up the tree from the bottom, freeing leaves and any interior
8030 * nodes which have had all slots visited. If a node (leaf or
8031 * interior) is freed, the node above it will have it's slot
8032 * incremented. The root node will never be freed.
8034 * At the end of this function, we should have a path which has all
8035 * slots incremented to the next position for a search. If we need to
8036 * read a new node it will be NULL and the node above it will have the
8037 * correct slot selected for a later read.
8039 * If we increment the root nodes slot counter past the number of
8040 * elements, 1 is returned to signal completion of the search.
8042 static int adjust_slots_upwards(struct btrfs_root *root,
8043 struct btrfs_path *path, int root_level)
8047 struct extent_buffer *eb;
8049 if (root_level == 0)
8052 while (level <= root_level) {
8053 eb = path->nodes[level];
8054 nr = btrfs_header_nritems(eb);
8055 path->slots[level]++;
8056 slot = path->slots[level];
8057 if (slot >= nr || level == 0) {
8059 * Don't free the root - we will detect this
8060 * condition after our loop and return a
8061 * positive value for caller to stop walking the tree.
8063 if (level != root_level) {
8064 btrfs_tree_unlock_rw(eb, path->locks[level]);
8065 path->locks[level] = 0;
8067 free_extent_buffer(eb);
8068 path->nodes[level] = NULL;
8069 path->slots[level] = 0;
8073 * We have a valid slot to walk back down
8074 * from. Stop here so caller can process these
8083 eb = path->nodes[root_level];
8084 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8091 * root_eb is the subtree root and is locked before this function is called.
8092 * TODO: Modify this function to mark all (including complete shared node)
8093 * to dirty_extent_root to allow it get accounted in qgroup.
8095 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8096 struct btrfs_root *root,
8097 struct extent_buffer *root_eb,
8103 struct extent_buffer *eb = root_eb;
8104 struct btrfs_path *path = NULL;
8106 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8107 BUG_ON(root_eb == NULL);
8109 if (!root->fs_info->quota_enabled)
8112 if (!extent_buffer_uptodate(root_eb)) {
8113 ret = btrfs_read_buffer(root_eb, root_gen);
8118 if (root_level == 0) {
8119 ret = account_leaf_items(trans, root, root_eb);
8123 path = btrfs_alloc_path();
8128 * Walk down the tree. Missing extent blocks are filled in as
8129 * we go. Metadata is accounted every time we read a new
8132 * When we reach a leaf, we account for file extent items in it,
8133 * walk back up the tree (adjusting slot pointers as we go)
8134 * and restart the search process.
8136 extent_buffer_get(root_eb); /* For path */
8137 path->nodes[root_level] = root_eb;
8138 path->slots[root_level] = 0;
8139 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8142 while (level >= 0) {
8143 if (path->nodes[level] == NULL) {
8148 /* We need to get child blockptr/gen from
8149 * parent before we can read it. */
8150 eb = path->nodes[level + 1];
8151 parent_slot = path->slots[level + 1];
8152 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8153 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8155 eb = read_tree_block(root, child_bytenr, child_gen);
8159 } else if (!extent_buffer_uptodate(eb)) {
8160 free_extent_buffer(eb);
8165 path->nodes[level] = eb;
8166 path->slots[level] = 0;
8168 btrfs_tree_read_lock(eb);
8169 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8170 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8174 ret = account_leaf_items(trans, root, path->nodes[level]);
8178 /* Nonzero return here means we completed our search */
8179 ret = adjust_slots_upwards(root, path, root_level);
8183 /* Restart search with new slots */
8192 btrfs_free_path(path);
8198 * helper to process tree block while walking down the tree.
8200 * when wc->stage == UPDATE_BACKREF, this function updates
8201 * back refs for pointers in the block.
8203 * NOTE: return value 1 means we should stop walking down.
8205 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8206 struct btrfs_root *root,
8207 struct btrfs_path *path,
8208 struct walk_control *wc, int lookup_info)
8210 int level = wc->level;
8211 struct extent_buffer *eb = path->nodes[level];
8212 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8215 if (wc->stage == UPDATE_BACKREF &&
8216 btrfs_header_owner(eb) != root->root_key.objectid)
8220 * when reference count of tree block is 1, it won't increase
8221 * again. once full backref flag is set, we never clear it.
8224 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8225 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8226 BUG_ON(!path->locks[level]);
8227 ret = btrfs_lookup_extent_info(trans, root,
8228 eb->start, level, 1,
8231 BUG_ON(ret == -ENOMEM);
8234 BUG_ON(wc->refs[level] == 0);
8237 if (wc->stage == DROP_REFERENCE) {
8238 if (wc->refs[level] > 1)
8241 if (path->locks[level] && !wc->keep_locks) {
8242 btrfs_tree_unlock_rw(eb, path->locks[level]);
8243 path->locks[level] = 0;
8248 /* wc->stage == UPDATE_BACKREF */
8249 if (!(wc->flags[level] & flag)) {
8250 BUG_ON(!path->locks[level]);
8251 ret = btrfs_inc_ref(trans, root, eb, 1);
8252 BUG_ON(ret); /* -ENOMEM */
8253 ret = btrfs_dec_ref(trans, root, eb, 0);
8254 BUG_ON(ret); /* -ENOMEM */
8255 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8257 btrfs_header_level(eb), 0);
8258 BUG_ON(ret); /* -ENOMEM */
8259 wc->flags[level] |= flag;
8263 * the block is shared by multiple trees, so it's not good to
8264 * keep the tree lock
8266 if (path->locks[level] && level > 0) {
8267 btrfs_tree_unlock_rw(eb, path->locks[level]);
8268 path->locks[level] = 0;
8274 * helper to process tree block pointer.
8276 * when wc->stage == DROP_REFERENCE, this function checks
8277 * reference count of the block pointed to. if the block
8278 * is shared and we need update back refs for the subtree
8279 * rooted at the block, this function changes wc->stage to
8280 * UPDATE_BACKREF. if the block is shared and there is no
8281 * need to update back, this function drops the reference
8284 * NOTE: return value 1 means we should stop walking down.
8286 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8287 struct btrfs_root *root,
8288 struct btrfs_path *path,
8289 struct walk_control *wc, int *lookup_info)
8295 struct btrfs_key key;
8296 struct extent_buffer *next;
8297 int level = wc->level;
8300 bool need_account = false;
8302 generation = btrfs_node_ptr_generation(path->nodes[level],
8303 path->slots[level]);
8305 * if the lower level block was created before the snapshot
8306 * was created, we know there is no need to update back refs
8309 if (wc->stage == UPDATE_BACKREF &&
8310 generation <= root->root_key.offset) {
8315 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8316 blocksize = root->nodesize;
8318 next = btrfs_find_tree_block(root->fs_info, bytenr);
8320 next = btrfs_find_create_tree_block(root, bytenr);
8323 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8327 btrfs_tree_lock(next);
8328 btrfs_set_lock_blocking(next);
8330 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8331 &wc->refs[level - 1],
8332 &wc->flags[level - 1]);
8334 btrfs_tree_unlock(next);
8338 if (unlikely(wc->refs[level - 1] == 0)) {
8339 btrfs_err(root->fs_info, "Missing references.");
8344 if (wc->stage == DROP_REFERENCE) {
8345 if (wc->refs[level - 1] > 1) {
8346 need_account = true;
8348 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8351 if (!wc->update_ref ||
8352 generation <= root->root_key.offset)
8355 btrfs_node_key_to_cpu(path->nodes[level], &key,
8356 path->slots[level]);
8357 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8361 wc->stage = UPDATE_BACKREF;
8362 wc->shared_level = level - 1;
8366 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8370 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8371 btrfs_tree_unlock(next);
8372 free_extent_buffer(next);
8378 if (reada && level == 1)
8379 reada_walk_down(trans, root, wc, path);
8380 next = read_tree_block(root, bytenr, generation);
8382 return PTR_ERR(next);
8383 } else if (!extent_buffer_uptodate(next)) {
8384 free_extent_buffer(next);
8387 btrfs_tree_lock(next);
8388 btrfs_set_lock_blocking(next);
8392 BUG_ON(level != btrfs_header_level(next));
8393 path->nodes[level] = next;
8394 path->slots[level] = 0;
8395 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8401 wc->refs[level - 1] = 0;
8402 wc->flags[level - 1] = 0;
8403 if (wc->stage == DROP_REFERENCE) {
8404 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8405 parent = path->nodes[level]->start;
8407 BUG_ON(root->root_key.objectid !=
8408 btrfs_header_owner(path->nodes[level]));
8413 ret = account_shared_subtree(trans, root, next,
8414 generation, level - 1);
8416 btrfs_err_rl(root->fs_info,
8418 "%d accounting shared subtree. Quota "
8419 "is out of sync, rescan required.",
8423 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8424 root->root_key.objectid, level - 1, 0, 0);
8425 BUG_ON(ret); /* -ENOMEM */
8427 btrfs_tree_unlock(next);
8428 free_extent_buffer(next);
8434 * helper to process tree block while walking up the tree.
8436 * when wc->stage == DROP_REFERENCE, this function drops
8437 * reference count on the block.
8439 * when wc->stage == UPDATE_BACKREF, this function changes
8440 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8441 * to UPDATE_BACKREF previously while processing the block.
8443 * NOTE: return value 1 means we should stop walking up.
8445 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8446 struct btrfs_root *root,
8447 struct btrfs_path *path,
8448 struct walk_control *wc)
8451 int level = wc->level;
8452 struct extent_buffer *eb = path->nodes[level];
8455 if (wc->stage == UPDATE_BACKREF) {
8456 BUG_ON(wc->shared_level < level);
8457 if (level < wc->shared_level)
8460 ret = find_next_key(path, level + 1, &wc->update_progress);
8464 wc->stage = DROP_REFERENCE;
8465 wc->shared_level = -1;
8466 path->slots[level] = 0;
8469 * check reference count again if the block isn't locked.
8470 * we should start walking down the tree again if reference
8473 if (!path->locks[level]) {
8475 btrfs_tree_lock(eb);
8476 btrfs_set_lock_blocking(eb);
8477 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8479 ret = btrfs_lookup_extent_info(trans, root,
8480 eb->start, level, 1,
8484 btrfs_tree_unlock_rw(eb, path->locks[level]);
8485 path->locks[level] = 0;
8488 BUG_ON(wc->refs[level] == 0);
8489 if (wc->refs[level] == 1) {
8490 btrfs_tree_unlock_rw(eb, path->locks[level]);
8491 path->locks[level] = 0;
8497 /* wc->stage == DROP_REFERENCE */
8498 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8500 if (wc->refs[level] == 1) {
8502 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8503 ret = btrfs_dec_ref(trans, root, eb, 1);
8505 ret = btrfs_dec_ref(trans, root, eb, 0);
8506 BUG_ON(ret); /* -ENOMEM */
8507 ret = account_leaf_items(trans, root, eb);
8509 btrfs_err_rl(root->fs_info,
8511 "%d accounting leaf items. Quota "
8512 "is out of sync, rescan required.",
8516 /* make block locked assertion in clean_tree_block happy */
8517 if (!path->locks[level] &&
8518 btrfs_header_generation(eb) == trans->transid) {
8519 btrfs_tree_lock(eb);
8520 btrfs_set_lock_blocking(eb);
8521 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8523 clean_tree_block(trans, root->fs_info, eb);
8526 if (eb == root->node) {
8527 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8530 BUG_ON(root->root_key.objectid !=
8531 btrfs_header_owner(eb));
8533 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8534 parent = path->nodes[level + 1]->start;
8536 BUG_ON(root->root_key.objectid !=
8537 btrfs_header_owner(path->nodes[level + 1]));
8540 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8542 wc->refs[level] = 0;
8543 wc->flags[level] = 0;
8547 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8548 struct btrfs_root *root,
8549 struct btrfs_path *path,
8550 struct walk_control *wc)
8552 int level = wc->level;
8553 int lookup_info = 1;
8556 while (level >= 0) {
8557 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8564 if (path->slots[level] >=
8565 btrfs_header_nritems(path->nodes[level]))
8568 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8570 path->slots[level]++;
8579 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8580 struct btrfs_root *root,
8581 struct btrfs_path *path,
8582 struct walk_control *wc, int max_level)
8584 int level = wc->level;
8587 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8588 while (level < max_level && path->nodes[level]) {
8590 if (path->slots[level] + 1 <
8591 btrfs_header_nritems(path->nodes[level])) {
8592 path->slots[level]++;
8595 ret = walk_up_proc(trans, root, path, wc);
8599 if (path->locks[level]) {
8600 btrfs_tree_unlock_rw(path->nodes[level],
8601 path->locks[level]);
8602 path->locks[level] = 0;
8604 free_extent_buffer(path->nodes[level]);
8605 path->nodes[level] = NULL;
8613 * drop a subvolume tree.
8615 * this function traverses the tree freeing any blocks that only
8616 * referenced by the tree.
8618 * when a shared tree block is found. this function decreases its
8619 * reference count by one. if update_ref is true, this function
8620 * also make sure backrefs for the shared block and all lower level
8621 * blocks are properly updated.
8623 * If called with for_reloc == 0, may exit early with -EAGAIN
8625 int btrfs_drop_snapshot(struct btrfs_root *root,
8626 struct btrfs_block_rsv *block_rsv, int update_ref,
8629 struct btrfs_path *path;
8630 struct btrfs_trans_handle *trans;
8631 struct btrfs_root *tree_root = root->fs_info->tree_root;
8632 struct btrfs_root_item *root_item = &root->root_item;
8633 struct walk_control *wc;
8634 struct btrfs_key key;
8638 bool root_dropped = false;
8640 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8642 path = btrfs_alloc_path();
8648 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8650 btrfs_free_path(path);
8655 trans = btrfs_start_transaction(tree_root, 0);
8656 if (IS_ERR(trans)) {
8657 err = PTR_ERR(trans);
8662 trans->block_rsv = block_rsv;
8664 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8665 level = btrfs_header_level(root->node);
8666 path->nodes[level] = btrfs_lock_root_node(root);
8667 btrfs_set_lock_blocking(path->nodes[level]);
8668 path->slots[level] = 0;
8669 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8670 memset(&wc->update_progress, 0,
8671 sizeof(wc->update_progress));
8673 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8674 memcpy(&wc->update_progress, &key,
8675 sizeof(wc->update_progress));
8677 level = root_item->drop_level;
8679 path->lowest_level = level;
8680 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8681 path->lowest_level = 0;
8689 * unlock our path, this is safe because only this
8690 * function is allowed to delete this snapshot
8692 btrfs_unlock_up_safe(path, 0);
8694 level = btrfs_header_level(root->node);
8696 btrfs_tree_lock(path->nodes[level]);
8697 btrfs_set_lock_blocking(path->nodes[level]);
8698 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8700 ret = btrfs_lookup_extent_info(trans, root,
8701 path->nodes[level]->start,
8702 level, 1, &wc->refs[level],
8708 BUG_ON(wc->refs[level] == 0);
8710 if (level == root_item->drop_level)
8713 btrfs_tree_unlock(path->nodes[level]);
8714 path->locks[level] = 0;
8715 WARN_ON(wc->refs[level] != 1);
8721 wc->shared_level = -1;
8722 wc->stage = DROP_REFERENCE;
8723 wc->update_ref = update_ref;
8725 wc->for_reloc = for_reloc;
8726 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8730 ret = walk_down_tree(trans, root, path, wc);
8736 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8743 BUG_ON(wc->stage != DROP_REFERENCE);
8747 if (wc->stage == DROP_REFERENCE) {
8749 btrfs_node_key(path->nodes[level],
8750 &root_item->drop_progress,
8751 path->slots[level]);
8752 root_item->drop_level = level;
8755 BUG_ON(wc->level == 0);
8756 if (btrfs_should_end_transaction(trans, tree_root) ||
8757 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8758 ret = btrfs_update_root(trans, tree_root,
8762 btrfs_abort_transaction(trans, tree_root, ret);
8767 btrfs_end_transaction_throttle(trans, tree_root);
8768 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8769 pr_debug("BTRFS: drop snapshot early exit\n");
8774 trans = btrfs_start_transaction(tree_root, 0);
8775 if (IS_ERR(trans)) {
8776 err = PTR_ERR(trans);
8780 trans->block_rsv = block_rsv;
8783 btrfs_release_path(path);
8787 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8789 btrfs_abort_transaction(trans, tree_root, ret);
8793 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8794 ret = btrfs_find_root(tree_root, &root->root_key, path,
8797 btrfs_abort_transaction(trans, tree_root, ret);
8800 } else if (ret > 0) {
8801 /* if we fail to delete the orphan item this time
8802 * around, it'll get picked up the next time.
8804 * The most common failure here is just -ENOENT.
8806 btrfs_del_orphan_item(trans, tree_root,
8807 root->root_key.objectid);
8811 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8812 btrfs_add_dropped_root(trans, root);
8814 free_extent_buffer(root->node);
8815 free_extent_buffer(root->commit_root);
8816 btrfs_put_fs_root(root);
8818 root_dropped = true;
8820 btrfs_end_transaction_throttle(trans, tree_root);
8823 btrfs_free_path(path);
8826 * So if we need to stop dropping the snapshot for whatever reason we
8827 * need to make sure to add it back to the dead root list so that we
8828 * keep trying to do the work later. This also cleans up roots if we
8829 * don't have it in the radix (like when we recover after a power fail
8830 * or unmount) so we don't leak memory.
8832 if (!for_reloc && root_dropped == false)
8833 btrfs_add_dead_root(root);
8834 if (err && err != -EAGAIN)
8835 btrfs_std_error(root->fs_info, err, NULL);
8840 * drop subtree rooted at tree block 'node'.
8842 * NOTE: this function will unlock and release tree block 'node'
8843 * only used by relocation code
8845 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8846 struct btrfs_root *root,
8847 struct extent_buffer *node,
8848 struct extent_buffer *parent)
8850 struct btrfs_path *path;
8851 struct walk_control *wc;
8857 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8859 path = btrfs_alloc_path();
8863 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8865 btrfs_free_path(path);
8869 btrfs_assert_tree_locked(parent);
8870 parent_level = btrfs_header_level(parent);
8871 extent_buffer_get(parent);
8872 path->nodes[parent_level] = parent;
8873 path->slots[parent_level] = btrfs_header_nritems(parent);
8875 btrfs_assert_tree_locked(node);
8876 level = btrfs_header_level(node);
8877 path->nodes[level] = node;
8878 path->slots[level] = 0;
8879 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8881 wc->refs[parent_level] = 1;
8882 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8884 wc->shared_level = -1;
8885 wc->stage = DROP_REFERENCE;
8889 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8892 wret = walk_down_tree(trans, root, path, wc);
8898 wret = walk_up_tree(trans, root, path, wc, parent_level);
8906 btrfs_free_path(path);
8910 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8916 * if restripe for this chunk_type is on pick target profile and
8917 * return, otherwise do the usual balance
8919 stripped = get_restripe_target(root->fs_info, flags);
8921 return extended_to_chunk(stripped);
8923 num_devices = root->fs_info->fs_devices->rw_devices;
8925 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8926 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8927 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8929 if (num_devices == 1) {
8930 stripped |= BTRFS_BLOCK_GROUP_DUP;
8931 stripped = flags & ~stripped;
8933 /* turn raid0 into single device chunks */
8934 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8937 /* turn mirroring into duplication */
8938 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8939 BTRFS_BLOCK_GROUP_RAID10))
8940 return stripped | BTRFS_BLOCK_GROUP_DUP;
8942 /* they already had raid on here, just return */
8943 if (flags & stripped)
8946 stripped |= BTRFS_BLOCK_GROUP_DUP;
8947 stripped = flags & ~stripped;
8949 /* switch duplicated blocks with raid1 */
8950 if (flags & BTRFS_BLOCK_GROUP_DUP)
8951 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8953 /* this is drive concat, leave it alone */
8959 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8961 struct btrfs_space_info *sinfo = cache->space_info;
8963 u64 min_allocable_bytes;
8967 * We need some metadata space and system metadata space for
8968 * allocating chunks in some corner cases until we force to set
8969 * it to be readonly.
8972 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8974 min_allocable_bytes = 1 * 1024 * 1024;
8976 min_allocable_bytes = 0;
8978 spin_lock(&sinfo->lock);
8979 spin_lock(&cache->lock);
8987 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8988 cache->bytes_super - btrfs_block_group_used(&cache->item);
8990 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8991 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8992 min_allocable_bytes <= sinfo->total_bytes) {
8993 sinfo->bytes_readonly += num_bytes;
8995 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8999 spin_unlock(&cache->lock);
9000 spin_unlock(&sinfo->lock);
9004 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9005 struct btrfs_block_group_cache *cache)
9008 struct btrfs_trans_handle *trans;
9013 trans = btrfs_join_transaction(root);
9015 return PTR_ERR(trans);
9018 * we're not allowed to set block groups readonly after the dirty
9019 * block groups cache has started writing. If it already started,
9020 * back off and let this transaction commit
9022 mutex_lock(&root->fs_info->ro_block_group_mutex);
9023 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9024 u64 transid = trans->transid;
9026 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9027 btrfs_end_transaction(trans, root);
9029 ret = btrfs_wait_for_commit(root, transid);
9036 * if we are changing raid levels, try to allocate a corresponding
9037 * block group with the new raid level.
9039 alloc_flags = update_block_group_flags(root, cache->flags);
9040 if (alloc_flags != cache->flags) {
9041 ret = do_chunk_alloc(trans, root, alloc_flags,
9044 * ENOSPC is allowed here, we may have enough space
9045 * already allocated at the new raid level to
9054 ret = inc_block_group_ro(cache, 0);
9057 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9058 ret = do_chunk_alloc(trans, root, alloc_flags,
9062 ret = inc_block_group_ro(cache, 0);
9064 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9065 alloc_flags = update_block_group_flags(root, cache->flags);
9066 lock_chunks(root->fs_info->chunk_root);
9067 check_system_chunk(trans, root, alloc_flags);
9068 unlock_chunks(root->fs_info->chunk_root);
9070 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9072 btrfs_end_transaction(trans, root);
9076 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9077 struct btrfs_root *root, u64 type)
9079 u64 alloc_flags = get_alloc_profile(root, type);
9080 return do_chunk_alloc(trans, root, alloc_flags,
9085 * helper to account the unused space of all the readonly block group in the
9086 * space_info. takes mirrors into account.
9088 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9090 struct btrfs_block_group_cache *block_group;
9094 /* It's df, we don't care if it's racey */
9095 if (list_empty(&sinfo->ro_bgs))
9098 spin_lock(&sinfo->lock);
9099 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9100 spin_lock(&block_group->lock);
9102 if (!block_group->ro) {
9103 spin_unlock(&block_group->lock);
9107 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9108 BTRFS_BLOCK_GROUP_RAID10 |
9109 BTRFS_BLOCK_GROUP_DUP))
9114 free_bytes += (block_group->key.offset -
9115 btrfs_block_group_used(&block_group->item)) *
9118 spin_unlock(&block_group->lock);
9120 spin_unlock(&sinfo->lock);
9125 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9126 struct btrfs_block_group_cache *cache)
9128 struct btrfs_space_info *sinfo = cache->space_info;
9133 spin_lock(&sinfo->lock);
9134 spin_lock(&cache->lock);
9136 num_bytes = cache->key.offset - cache->reserved -
9137 cache->pinned - cache->bytes_super -
9138 btrfs_block_group_used(&cache->item);
9139 sinfo->bytes_readonly -= num_bytes;
9140 list_del_init(&cache->ro_list);
9142 spin_unlock(&cache->lock);
9143 spin_unlock(&sinfo->lock);
9147 * checks to see if its even possible to relocate this block group.
9149 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9150 * ok to go ahead and try.
9152 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9154 struct btrfs_block_group_cache *block_group;
9155 struct btrfs_space_info *space_info;
9156 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9157 struct btrfs_device *device;
9158 struct btrfs_trans_handle *trans;
9167 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9169 /* odd, couldn't find the block group, leave it alone */
9173 min_free = btrfs_block_group_used(&block_group->item);
9175 /* no bytes used, we're good */
9179 space_info = block_group->space_info;
9180 spin_lock(&space_info->lock);
9182 full = space_info->full;
9185 * if this is the last block group we have in this space, we can't
9186 * relocate it unless we're able to allocate a new chunk below.
9188 * Otherwise, we need to make sure we have room in the space to handle
9189 * all of the extents from this block group. If we can, we're good
9191 if ((space_info->total_bytes != block_group->key.offset) &&
9192 (space_info->bytes_used + space_info->bytes_reserved +
9193 space_info->bytes_pinned + space_info->bytes_readonly +
9194 min_free < space_info->total_bytes)) {
9195 spin_unlock(&space_info->lock);
9198 spin_unlock(&space_info->lock);
9201 * ok we don't have enough space, but maybe we have free space on our
9202 * devices to allocate new chunks for relocation, so loop through our
9203 * alloc devices and guess if we have enough space. if this block
9204 * group is going to be restriped, run checks against the target
9205 * profile instead of the current one.
9217 target = get_restripe_target(root->fs_info, block_group->flags);
9219 index = __get_raid_index(extended_to_chunk(target));
9222 * this is just a balance, so if we were marked as full
9223 * we know there is no space for a new chunk
9228 index = get_block_group_index(block_group);
9231 if (index == BTRFS_RAID_RAID10) {
9235 } else if (index == BTRFS_RAID_RAID1) {
9237 } else if (index == BTRFS_RAID_DUP) {
9240 } else if (index == BTRFS_RAID_RAID0) {
9241 dev_min = fs_devices->rw_devices;
9242 min_free = div64_u64(min_free, dev_min);
9245 /* We need to do this so that we can look at pending chunks */
9246 trans = btrfs_join_transaction(root);
9247 if (IS_ERR(trans)) {
9248 ret = PTR_ERR(trans);
9252 mutex_lock(&root->fs_info->chunk_mutex);
9253 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9257 * check to make sure we can actually find a chunk with enough
9258 * space to fit our block group in.
9260 if (device->total_bytes > device->bytes_used + min_free &&
9261 !device->is_tgtdev_for_dev_replace) {
9262 ret = find_free_dev_extent(trans, device, min_free,
9267 if (dev_nr >= dev_min)
9273 mutex_unlock(&root->fs_info->chunk_mutex);
9274 btrfs_end_transaction(trans, root);
9276 btrfs_put_block_group(block_group);
9280 static int find_first_block_group(struct btrfs_root *root,
9281 struct btrfs_path *path, struct btrfs_key *key)
9284 struct btrfs_key found_key;
9285 struct extent_buffer *leaf;
9288 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9293 slot = path->slots[0];
9294 leaf = path->nodes[0];
9295 if (slot >= btrfs_header_nritems(leaf)) {
9296 ret = btrfs_next_leaf(root, path);
9303 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9305 if (found_key.objectid >= key->objectid &&
9306 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9316 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9318 struct btrfs_block_group_cache *block_group;
9322 struct inode *inode;
9324 block_group = btrfs_lookup_first_block_group(info, last);
9325 while (block_group) {
9326 spin_lock(&block_group->lock);
9327 if (block_group->iref)
9329 spin_unlock(&block_group->lock);
9330 block_group = next_block_group(info->tree_root,
9340 inode = block_group->inode;
9341 block_group->iref = 0;
9342 block_group->inode = NULL;
9343 spin_unlock(&block_group->lock);
9345 last = block_group->key.objectid + block_group->key.offset;
9346 btrfs_put_block_group(block_group);
9350 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9352 struct btrfs_block_group_cache *block_group;
9353 struct btrfs_space_info *space_info;
9354 struct btrfs_caching_control *caching_ctl;
9357 down_write(&info->commit_root_sem);
9358 while (!list_empty(&info->caching_block_groups)) {
9359 caching_ctl = list_entry(info->caching_block_groups.next,
9360 struct btrfs_caching_control, list);
9361 list_del(&caching_ctl->list);
9362 put_caching_control(caching_ctl);
9364 up_write(&info->commit_root_sem);
9366 spin_lock(&info->unused_bgs_lock);
9367 while (!list_empty(&info->unused_bgs)) {
9368 block_group = list_first_entry(&info->unused_bgs,
9369 struct btrfs_block_group_cache,
9371 list_del_init(&block_group->bg_list);
9372 btrfs_put_block_group(block_group);
9374 spin_unlock(&info->unused_bgs_lock);
9376 spin_lock(&info->block_group_cache_lock);
9377 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9378 block_group = rb_entry(n, struct btrfs_block_group_cache,
9380 rb_erase(&block_group->cache_node,
9381 &info->block_group_cache_tree);
9382 RB_CLEAR_NODE(&block_group->cache_node);
9383 spin_unlock(&info->block_group_cache_lock);
9385 down_write(&block_group->space_info->groups_sem);
9386 list_del(&block_group->list);
9387 up_write(&block_group->space_info->groups_sem);
9389 if (block_group->cached == BTRFS_CACHE_STARTED)
9390 wait_block_group_cache_done(block_group);
9393 * We haven't cached this block group, which means we could
9394 * possibly have excluded extents on this block group.
9396 if (block_group->cached == BTRFS_CACHE_NO ||
9397 block_group->cached == BTRFS_CACHE_ERROR)
9398 free_excluded_extents(info->extent_root, block_group);
9400 btrfs_remove_free_space_cache(block_group);
9401 btrfs_put_block_group(block_group);
9403 spin_lock(&info->block_group_cache_lock);
9405 spin_unlock(&info->block_group_cache_lock);
9407 /* now that all the block groups are freed, go through and
9408 * free all the space_info structs. This is only called during
9409 * the final stages of unmount, and so we know nobody is
9410 * using them. We call synchronize_rcu() once before we start,
9411 * just to be on the safe side.
9415 release_global_block_rsv(info);
9417 while (!list_empty(&info->space_info)) {
9420 space_info = list_entry(info->space_info.next,
9421 struct btrfs_space_info,
9423 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9424 if (WARN_ON(space_info->bytes_pinned > 0 ||
9425 space_info->bytes_reserved > 0 ||
9426 space_info->bytes_may_use > 0)) {
9427 dump_space_info(space_info, 0, 0);
9430 list_del(&space_info->list);
9431 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9432 struct kobject *kobj;
9433 kobj = space_info->block_group_kobjs[i];
9434 space_info->block_group_kobjs[i] = NULL;
9440 kobject_del(&space_info->kobj);
9441 kobject_put(&space_info->kobj);
9446 static void __link_block_group(struct btrfs_space_info *space_info,
9447 struct btrfs_block_group_cache *cache)
9449 int index = get_block_group_index(cache);
9452 down_write(&space_info->groups_sem);
9453 if (list_empty(&space_info->block_groups[index]))
9455 list_add_tail(&cache->list, &space_info->block_groups[index]);
9456 up_write(&space_info->groups_sem);
9459 struct raid_kobject *rkobj;
9462 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9465 rkobj->raid_type = index;
9466 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9467 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9468 "%s", get_raid_name(index));
9470 kobject_put(&rkobj->kobj);
9473 space_info->block_group_kobjs[index] = &rkobj->kobj;
9478 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9481 static struct btrfs_block_group_cache *
9482 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9484 struct btrfs_block_group_cache *cache;
9486 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9490 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9492 if (!cache->free_space_ctl) {
9497 cache->key.objectid = start;
9498 cache->key.offset = size;
9499 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9501 cache->sectorsize = root->sectorsize;
9502 cache->fs_info = root->fs_info;
9503 cache->full_stripe_len = btrfs_full_stripe_len(root,
9504 &root->fs_info->mapping_tree,
9506 atomic_set(&cache->count, 1);
9507 spin_lock_init(&cache->lock);
9508 init_rwsem(&cache->data_rwsem);
9509 INIT_LIST_HEAD(&cache->list);
9510 INIT_LIST_HEAD(&cache->cluster_list);
9511 INIT_LIST_HEAD(&cache->bg_list);
9512 INIT_LIST_HEAD(&cache->ro_list);
9513 INIT_LIST_HEAD(&cache->dirty_list);
9514 INIT_LIST_HEAD(&cache->io_list);
9515 btrfs_init_free_space_ctl(cache);
9516 atomic_set(&cache->trimming, 0);
9521 int btrfs_read_block_groups(struct btrfs_root *root)
9523 struct btrfs_path *path;
9525 struct btrfs_block_group_cache *cache;
9526 struct btrfs_fs_info *info = root->fs_info;
9527 struct btrfs_space_info *space_info;
9528 struct btrfs_key key;
9529 struct btrfs_key found_key;
9530 struct extent_buffer *leaf;
9534 root = info->extent_root;
9537 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9538 path = btrfs_alloc_path();
9543 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9544 if (btrfs_test_opt(root, SPACE_CACHE) &&
9545 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9547 if (btrfs_test_opt(root, CLEAR_CACHE))
9551 ret = find_first_block_group(root, path, &key);
9557 leaf = path->nodes[0];
9558 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9560 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9569 * When we mount with old space cache, we need to
9570 * set BTRFS_DC_CLEAR and set dirty flag.
9572 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9573 * truncate the old free space cache inode and
9575 * b) Setting 'dirty flag' makes sure that we flush
9576 * the new space cache info onto disk.
9578 if (btrfs_test_opt(root, SPACE_CACHE))
9579 cache->disk_cache_state = BTRFS_DC_CLEAR;
9582 read_extent_buffer(leaf, &cache->item,
9583 btrfs_item_ptr_offset(leaf, path->slots[0]),
9584 sizeof(cache->item));
9585 cache->flags = btrfs_block_group_flags(&cache->item);
9587 key.objectid = found_key.objectid + found_key.offset;
9588 btrfs_release_path(path);
9591 * We need to exclude the super stripes now so that the space
9592 * info has super bytes accounted for, otherwise we'll think
9593 * we have more space than we actually do.
9595 ret = exclude_super_stripes(root, cache);
9598 * We may have excluded something, so call this just in
9601 free_excluded_extents(root, cache);
9602 btrfs_put_block_group(cache);
9607 * check for two cases, either we are full, and therefore
9608 * don't need to bother with the caching work since we won't
9609 * find any space, or we are empty, and we can just add all
9610 * the space in and be done with it. This saves us _alot_ of
9611 * time, particularly in the full case.
9613 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9614 cache->last_byte_to_unpin = (u64)-1;
9615 cache->cached = BTRFS_CACHE_FINISHED;
9616 free_excluded_extents(root, cache);
9617 } else if (btrfs_block_group_used(&cache->item) == 0) {
9618 cache->last_byte_to_unpin = (u64)-1;
9619 cache->cached = BTRFS_CACHE_FINISHED;
9620 add_new_free_space(cache, root->fs_info,
9622 found_key.objectid +
9624 free_excluded_extents(root, cache);
9627 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9629 btrfs_remove_free_space_cache(cache);
9630 btrfs_put_block_group(cache);
9634 ret = update_space_info(info, cache->flags, found_key.offset,
9635 btrfs_block_group_used(&cache->item),
9638 btrfs_remove_free_space_cache(cache);
9639 spin_lock(&info->block_group_cache_lock);
9640 rb_erase(&cache->cache_node,
9641 &info->block_group_cache_tree);
9642 RB_CLEAR_NODE(&cache->cache_node);
9643 spin_unlock(&info->block_group_cache_lock);
9644 btrfs_put_block_group(cache);
9648 cache->space_info = space_info;
9649 spin_lock(&cache->space_info->lock);
9650 cache->space_info->bytes_readonly += cache->bytes_super;
9651 spin_unlock(&cache->space_info->lock);
9653 __link_block_group(space_info, cache);
9655 set_avail_alloc_bits(root->fs_info, cache->flags);
9656 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9657 inc_block_group_ro(cache, 1);
9658 } else if (btrfs_block_group_used(&cache->item) == 0) {
9659 spin_lock(&info->unused_bgs_lock);
9660 /* Should always be true but just in case. */
9661 if (list_empty(&cache->bg_list)) {
9662 btrfs_get_block_group(cache);
9663 list_add_tail(&cache->bg_list,
9666 spin_unlock(&info->unused_bgs_lock);
9670 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9671 if (!(get_alloc_profile(root, space_info->flags) &
9672 (BTRFS_BLOCK_GROUP_RAID10 |
9673 BTRFS_BLOCK_GROUP_RAID1 |
9674 BTRFS_BLOCK_GROUP_RAID5 |
9675 BTRFS_BLOCK_GROUP_RAID6 |
9676 BTRFS_BLOCK_GROUP_DUP)))
9679 * avoid allocating from un-mirrored block group if there are
9680 * mirrored block groups.
9682 list_for_each_entry(cache,
9683 &space_info->block_groups[BTRFS_RAID_RAID0],
9685 inc_block_group_ro(cache, 1);
9686 list_for_each_entry(cache,
9687 &space_info->block_groups[BTRFS_RAID_SINGLE],
9689 inc_block_group_ro(cache, 1);
9692 init_global_block_rsv(info);
9695 btrfs_free_path(path);
9699 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9700 struct btrfs_root *root)
9702 struct btrfs_block_group_cache *block_group, *tmp;
9703 struct btrfs_root *extent_root = root->fs_info->extent_root;
9704 struct btrfs_block_group_item item;
9705 struct btrfs_key key;
9707 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9709 trans->can_flush_pending_bgs = false;
9710 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9714 spin_lock(&block_group->lock);
9715 memcpy(&item, &block_group->item, sizeof(item));
9716 memcpy(&key, &block_group->key, sizeof(key));
9717 spin_unlock(&block_group->lock);
9719 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9722 btrfs_abort_transaction(trans, extent_root, ret);
9723 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9724 key.objectid, key.offset);
9726 btrfs_abort_transaction(trans, extent_root, ret);
9728 list_del_init(&block_group->bg_list);
9730 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9733 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9734 struct btrfs_root *root, u64 bytes_used,
9735 u64 type, u64 chunk_objectid, u64 chunk_offset,
9739 struct btrfs_root *extent_root;
9740 struct btrfs_block_group_cache *cache;
9742 extent_root = root->fs_info->extent_root;
9744 btrfs_set_log_full_commit(root->fs_info, trans);
9746 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9750 btrfs_set_block_group_used(&cache->item, bytes_used);
9751 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9752 btrfs_set_block_group_flags(&cache->item, type);
9754 cache->flags = type;
9755 cache->last_byte_to_unpin = (u64)-1;
9756 cache->cached = BTRFS_CACHE_FINISHED;
9757 ret = exclude_super_stripes(root, cache);
9760 * We may have excluded something, so call this just in
9763 free_excluded_extents(root, cache);
9764 btrfs_put_block_group(cache);
9768 add_new_free_space(cache, root->fs_info, chunk_offset,
9769 chunk_offset + size);
9771 free_excluded_extents(root, cache);
9773 #ifdef CONFIG_BTRFS_DEBUG
9774 if (btrfs_should_fragment_free_space(root, cache)) {
9775 u64 new_bytes_used = size - bytes_used;
9777 bytes_used += new_bytes_used >> 1;
9778 fragment_free_space(root, cache);
9782 * Call to ensure the corresponding space_info object is created and
9783 * assigned to our block group, but don't update its counters just yet.
9784 * We want our bg to be added to the rbtree with its ->space_info set.
9786 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9787 &cache->space_info);
9789 btrfs_remove_free_space_cache(cache);
9790 btrfs_put_block_group(cache);
9794 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9796 btrfs_remove_free_space_cache(cache);
9797 btrfs_put_block_group(cache);
9802 * Now that our block group has its ->space_info set and is inserted in
9803 * the rbtree, update the space info's counters.
9805 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9806 &cache->space_info);
9808 btrfs_remove_free_space_cache(cache);
9809 spin_lock(&root->fs_info->block_group_cache_lock);
9810 rb_erase(&cache->cache_node,
9811 &root->fs_info->block_group_cache_tree);
9812 RB_CLEAR_NODE(&cache->cache_node);
9813 spin_unlock(&root->fs_info->block_group_cache_lock);
9814 btrfs_put_block_group(cache);
9817 update_global_block_rsv(root->fs_info);
9819 spin_lock(&cache->space_info->lock);
9820 cache->space_info->bytes_readonly += cache->bytes_super;
9821 spin_unlock(&cache->space_info->lock);
9823 __link_block_group(cache->space_info, cache);
9825 list_add_tail(&cache->bg_list, &trans->new_bgs);
9827 set_avail_alloc_bits(extent_root->fs_info, type);
9832 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9834 u64 extra_flags = chunk_to_extended(flags) &
9835 BTRFS_EXTENDED_PROFILE_MASK;
9837 write_seqlock(&fs_info->profiles_lock);
9838 if (flags & BTRFS_BLOCK_GROUP_DATA)
9839 fs_info->avail_data_alloc_bits &= ~extra_flags;
9840 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9841 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9842 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9843 fs_info->avail_system_alloc_bits &= ~extra_flags;
9844 write_sequnlock(&fs_info->profiles_lock);
9847 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9848 struct btrfs_root *root, u64 group_start,
9849 struct extent_map *em)
9851 struct btrfs_path *path;
9852 struct btrfs_block_group_cache *block_group;
9853 struct btrfs_free_cluster *cluster;
9854 struct btrfs_root *tree_root = root->fs_info->tree_root;
9855 struct btrfs_key key;
9856 struct inode *inode;
9857 struct kobject *kobj = NULL;
9861 struct btrfs_caching_control *caching_ctl = NULL;
9864 root = root->fs_info->extent_root;
9866 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9867 BUG_ON(!block_group);
9868 BUG_ON(!block_group->ro);
9871 * Free the reserved super bytes from this block group before
9874 free_excluded_extents(root, block_group);
9876 memcpy(&key, &block_group->key, sizeof(key));
9877 index = get_block_group_index(block_group);
9878 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9879 BTRFS_BLOCK_GROUP_RAID1 |
9880 BTRFS_BLOCK_GROUP_RAID10))
9885 /* make sure this block group isn't part of an allocation cluster */
9886 cluster = &root->fs_info->data_alloc_cluster;
9887 spin_lock(&cluster->refill_lock);
9888 btrfs_return_cluster_to_free_space(block_group, cluster);
9889 spin_unlock(&cluster->refill_lock);
9892 * make sure this block group isn't part of a metadata
9893 * allocation cluster
9895 cluster = &root->fs_info->meta_alloc_cluster;
9896 spin_lock(&cluster->refill_lock);
9897 btrfs_return_cluster_to_free_space(block_group, cluster);
9898 spin_unlock(&cluster->refill_lock);
9900 path = btrfs_alloc_path();
9907 * get the inode first so any iput calls done for the io_list
9908 * aren't the final iput (no unlinks allowed now)
9910 inode = lookup_free_space_inode(tree_root, block_group, path);
9912 mutex_lock(&trans->transaction->cache_write_mutex);
9914 * make sure our free spache cache IO is done before remove the
9917 spin_lock(&trans->transaction->dirty_bgs_lock);
9918 if (!list_empty(&block_group->io_list)) {
9919 list_del_init(&block_group->io_list);
9921 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9923 spin_unlock(&trans->transaction->dirty_bgs_lock);
9924 btrfs_wait_cache_io(root, trans, block_group,
9925 &block_group->io_ctl, path,
9926 block_group->key.objectid);
9927 btrfs_put_block_group(block_group);
9928 spin_lock(&trans->transaction->dirty_bgs_lock);
9931 if (!list_empty(&block_group->dirty_list)) {
9932 list_del_init(&block_group->dirty_list);
9933 btrfs_put_block_group(block_group);
9935 spin_unlock(&trans->transaction->dirty_bgs_lock);
9936 mutex_unlock(&trans->transaction->cache_write_mutex);
9938 if (!IS_ERR(inode)) {
9939 ret = btrfs_orphan_add(trans, inode);
9941 btrfs_add_delayed_iput(inode);
9945 /* One for the block groups ref */
9946 spin_lock(&block_group->lock);
9947 if (block_group->iref) {
9948 block_group->iref = 0;
9949 block_group->inode = NULL;
9950 spin_unlock(&block_group->lock);
9953 spin_unlock(&block_group->lock);
9955 /* One for our lookup ref */
9956 btrfs_add_delayed_iput(inode);
9959 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9960 key.offset = block_group->key.objectid;
9963 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9967 btrfs_release_path(path);
9969 ret = btrfs_del_item(trans, tree_root, path);
9972 btrfs_release_path(path);
9975 spin_lock(&root->fs_info->block_group_cache_lock);
9976 rb_erase(&block_group->cache_node,
9977 &root->fs_info->block_group_cache_tree);
9978 RB_CLEAR_NODE(&block_group->cache_node);
9980 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9981 root->fs_info->first_logical_byte = (u64)-1;
9982 spin_unlock(&root->fs_info->block_group_cache_lock);
9984 down_write(&block_group->space_info->groups_sem);
9986 * we must use list_del_init so people can check to see if they
9987 * are still on the list after taking the semaphore
9989 list_del_init(&block_group->list);
9990 if (list_empty(&block_group->space_info->block_groups[index])) {
9991 kobj = block_group->space_info->block_group_kobjs[index];
9992 block_group->space_info->block_group_kobjs[index] = NULL;
9993 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9995 up_write(&block_group->space_info->groups_sem);
10001 if (block_group->has_caching_ctl)
10002 caching_ctl = get_caching_control(block_group);
10003 if (block_group->cached == BTRFS_CACHE_STARTED)
10004 wait_block_group_cache_done(block_group);
10005 if (block_group->has_caching_ctl) {
10006 down_write(&root->fs_info->commit_root_sem);
10007 if (!caching_ctl) {
10008 struct btrfs_caching_control *ctl;
10010 list_for_each_entry(ctl,
10011 &root->fs_info->caching_block_groups, list)
10012 if (ctl->block_group == block_group) {
10014 atomic_inc(&caching_ctl->count);
10019 list_del_init(&caching_ctl->list);
10020 up_write(&root->fs_info->commit_root_sem);
10022 /* Once for the caching bgs list and once for us. */
10023 put_caching_control(caching_ctl);
10024 put_caching_control(caching_ctl);
10028 spin_lock(&trans->transaction->dirty_bgs_lock);
10029 if (!list_empty(&block_group->dirty_list)) {
10032 if (!list_empty(&block_group->io_list)) {
10035 spin_unlock(&trans->transaction->dirty_bgs_lock);
10036 btrfs_remove_free_space_cache(block_group);
10038 spin_lock(&block_group->space_info->lock);
10039 list_del_init(&block_group->ro_list);
10041 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10042 WARN_ON(block_group->space_info->total_bytes
10043 < block_group->key.offset);
10044 WARN_ON(block_group->space_info->bytes_readonly
10045 < block_group->key.offset);
10046 WARN_ON(block_group->space_info->disk_total
10047 < block_group->key.offset * factor);
10049 block_group->space_info->total_bytes -= block_group->key.offset;
10050 block_group->space_info->bytes_readonly -= block_group->key.offset;
10051 block_group->space_info->disk_total -= block_group->key.offset * factor;
10053 spin_unlock(&block_group->space_info->lock);
10055 memcpy(&key, &block_group->key, sizeof(key));
10058 if (!list_empty(&em->list)) {
10059 /* We're in the transaction->pending_chunks list. */
10060 free_extent_map(em);
10062 spin_lock(&block_group->lock);
10063 block_group->removed = 1;
10065 * At this point trimming can't start on this block group, because we
10066 * removed the block group from the tree fs_info->block_group_cache_tree
10067 * so no one can't find it anymore and even if someone already got this
10068 * block group before we removed it from the rbtree, they have already
10069 * incremented block_group->trimming - if they didn't, they won't find
10070 * any free space entries because we already removed them all when we
10071 * called btrfs_remove_free_space_cache().
10073 * And we must not remove the extent map from the fs_info->mapping_tree
10074 * to prevent the same logical address range and physical device space
10075 * ranges from being reused for a new block group. This is because our
10076 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10077 * completely transactionless, so while it is trimming a range the
10078 * currently running transaction might finish and a new one start,
10079 * allowing for new block groups to be created that can reuse the same
10080 * physical device locations unless we take this special care.
10082 * There may also be an implicit trim operation if the file system
10083 * is mounted with -odiscard. The same protections must remain
10084 * in place until the extents have been discarded completely when
10085 * the transaction commit has completed.
10087 remove_em = (atomic_read(&block_group->trimming) == 0);
10089 * Make sure a trimmer task always sees the em in the pinned_chunks list
10090 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10091 * before checking block_group->removed).
10095 * Our em might be in trans->transaction->pending_chunks which
10096 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10097 * and so is the fs_info->pinned_chunks list.
10099 * So at this point we must be holding the chunk_mutex to avoid
10100 * any races with chunk allocation (more specifically at
10101 * volumes.c:contains_pending_extent()), to ensure it always
10102 * sees the em, either in the pending_chunks list or in the
10103 * pinned_chunks list.
10105 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10107 spin_unlock(&block_group->lock);
10110 struct extent_map_tree *em_tree;
10112 em_tree = &root->fs_info->mapping_tree.map_tree;
10113 write_lock(&em_tree->lock);
10115 * The em might be in the pending_chunks list, so make sure the
10116 * chunk mutex is locked, since remove_extent_mapping() will
10117 * delete us from that list.
10119 remove_extent_mapping(em_tree, em);
10120 write_unlock(&em_tree->lock);
10121 /* once for the tree */
10122 free_extent_map(em);
10125 unlock_chunks(root);
10127 btrfs_put_block_group(block_group);
10128 btrfs_put_block_group(block_group);
10130 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10136 ret = btrfs_del_item(trans, root, path);
10138 btrfs_free_path(path);
10143 * Process the unused_bgs list and remove any that don't have any allocated
10144 * space inside of them.
10146 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10148 struct btrfs_block_group_cache *block_group;
10149 struct btrfs_space_info *space_info;
10150 struct btrfs_root *root = fs_info->extent_root;
10151 struct btrfs_trans_handle *trans;
10154 if (!fs_info->open)
10157 spin_lock(&fs_info->unused_bgs_lock);
10158 while (!list_empty(&fs_info->unused_bgs)) {
10162 block_group = list_first_entry(&fs_info->unused_bgs,
10163 struct btrfs_block_group_cache,
10165 space_info = block_group->space_info;
10166 list_del_init(&block_group->bg_list);
10167 if (ret || btrfs_mixed_space_info(space_info)) {
10168 btrfs_put_block_group(block_group);
10171 spin_unlock(&fs_info->unused_bgs_lock);
10173 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
10175 /* Don't want to race with allocators so take the groups_sem */
10176 down_write(&space_info->groups_sem);
10177 spin_lock(&block_group->lock);
10178 if (block_group->reserved ||
10179 btrfs_block_group_used(&block_group->item) ||
10182 * We want to bail if we made new allocations or have
10183 * outstanding allocations in this block group. We do
10184 * the ro check in case balance is currently acting on
10185 * this block group.
10187 spin_unlock(&block_group->lock);
10188 up_write(&space_info->groups_sem);
10191 spin_unlock(&block_group->lock);
10193 /* We don't want to force the issue, only flip if it's ok. */
10194 ret = inc_block_group_ro(block_group, 0);
10195 up_write(&space_info->groups_sem);
10202 * Want to do this before we do anything else so we can recover
10203 * properly if we fail to join the transaction.
10205 /* 1 for btrfs_orphan_reserve_metadata() */
10206 trans = btrfs_start_transaction(root, 1);
10207 if (IS_ERR(trans)) {
10208 btrfs_dec_block_group_ro(root, block_group);
10209 ret = PTR_ERR(trans);
10214 * We could have pending pinned extents for this block group,
10215 * just delete them, we don't care about them anymore.
10217 start = block_group->key.objectid;
10218 end = start + block_group->key.offset - 1;
10220 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10221 * btrfs_finish_extent_commit(). If we are at transaction N,
10222 * another task might be running finish_extent_commit() for the
10223 * previous transaction N - 1, and have seen a range belonging
10224 * to the block group in freed_extents[] before we were able to
10225 * clear the whole block group range from freed_extents[]. This
10226 * means that task can lookup for the block group after we
10227 * unpinned it from freed_extents[] and removed it, leading to
10228 * a BUG_ON() at btrfs_unpin_extent_range().
10230 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10231 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10232 EXTENT_DIRTY, GFP_NOFS);
10234 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10235 btrfs_dec_block_group_ro(root, block_group);
10238 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10239 EXTENT_DIRTY, GFP_NOFS);
10241 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10242 btrfs_dec_block_group_ro(root, block_group);
10245 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10247 /* Reset pinned so btrfs_put_block_group doesn't complain */
10248 spin_lock(&space_info->lock);
10249 spin_lock(&block_group->lock);
10251 space_info->bytes_pinned -= block_group->pinned;
10252 space_info->bytes_readonly += block_group->pinned;
10253 percpu_counter_add(&space_info->total_bytes_pinned,
10254 -block_group->pinned);
10255 block_group->pinned = 0;
10257 spin_unlock(&block_group->lock);
10258 spin_unlock(&space_info->lock);
10260 /* DISCARD can flip during remount */
10261 trimming = btrfs_test_opt(root, DISCARD);
10263 /* Implicit trim during transaction commit. */
10265 btrfs_get_block_group_trimming(block_group);
10268 * Btrfs_remove_chunk will abort the transaction if things go
10271 ret = btrfs_remove_chunk(trans, root,
10272 block_group->key.objectid);
10276 btrfs_put_block_group_trimming(block_group);
10281 * If we're not mounted with -odiscard, we can just forget
10282 * about this block group. Otherwise we'll need to wait
10283 * until transaction commit to do the actual discard.
10286 WARN_ON(!list_empty(&block_group->bg_list));
10287 spin_lock(&trans->transaction->deleted_bgs_lock);
10288 list_move(&block_group->bg_list,
10289 &trans->transaction->deleted_bgs);
10290 spin_unlock(&trans->transaction->deleted_bgs_lock);
10291 btrfs_get_block_group(block_group);
10294 btrfs_end_transaction(trans, root);
10296 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10297 btrfs_put_block_group(block_group);
10298 spin_lock(&fs_info->unused_bgs_lock);
10300 spin_unlock(&fs_info->unused_bgs_lock);
10303 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10305 struct btrfs_space_info *space_info;
10306 struct btrfs_super_block *disk_super;
10312 disk_super = fs_info->super_copy;
10313 if (!btrfs_super_root(disk_super))
10316 features = btrfs_super_incompat_flags(disk_super);
10317 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10320 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10321 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10326 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10327 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10329 flags = BTRFS_BLOCK_GROUP_METADATA;
10330 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10334 flags = BTRFS_BLOCK_GROUP_DATA;
10335 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10341 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10343 return unpin_extent_range(root, start, end, false);
10347 * It used to be that old block groups would be left around forever.
10348 * Iterating over them would be enough to trim unused space. Since we
10349 * now automatically remove them, we also need to iterate over unallocated
10352 * We don't want a transaction for this since the discard may take a
10353 * substantial amount of time. We don't require that a transaction be
10354 * running, but we do need to take a running transaction into account
10355 * to ensure that we're not discarding chunks that were released in
10356 * the current transaction.
10358 * Holding the chunks lock will prevent other threads from allocating
10359 * or releasing chunks, but it won't prevent a running transaction
10360 * from committing and releasing the memory that the pending chunks
10361 * list head uses. For that, we need to take a reference to the
10364 static int btrfs_trim_free_extents(struct btrfs_device *device,
10365 u64 minlen, u64 *trimmed)
10367 u64 start = 0, len = 0;
10372 /* Not writeable = nothing to do. */
10373 if (!device->writeable)
10376 /* No free space = nothing to do. */
10377 if (device->total_bytes <= device->bytes_used)
10383 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10384 struct btrfs_transaction *trans;
10387 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10391 down_read(&fs_info->commit_root_sem);
10393 spin_lock(&fs_info->trans_lock);
10394 trans = fs_info->running_transaction;
10396 atomic_inc(&trans->use_count);
10397 spin_unlock(&fs_info->trans_lock);
10399 ret = find_free_dev_extent_start(trans, device, minlen, start,
10402 btrfs_put_transaction(trans);
10405 up_read(&fs_info->commit_root_sem);
10406 mutex_unlock(&fs_info->chunk_mutex);
10407 if (ret == -ENOSPC)
10412 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10413 up_read(&fs_info->commit_root_sem);
10414 mutex_unlock(&fs_info->chunk_mutex);
10422 if (fatal_signal_pending(current)) {
10423 ret = -ERESTARTSYS;
10433 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10435 struct btrfs_fs_info *fs_info = root->fs_info;
10436 struct btrfs_block_group_cache *cache = NULL;
10437 struct btrfs_device *device;
10438 struct list_head *devices;
10443 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10447 * try to trim all FS space, our block group may start from non-zero.
10449 if (range->len == total_bytes)
10450 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10452 cache = btrfs_lookup_block_group(fs_info, range->start);
10455 if (cache->key.objectid >= (range->start + range->len)) {
10456 btrfs_put_block_group(cache);
10460 start = max(range->start, cache->key.objectid);
10461 end = min(range->start + range->len,
10462 cache->key.objectid + cache->key.offset);
10464 if (end - start >= range->minlen) {
10465 if (!block_group_cache_done(cache)) {
10466 ret = cache_block_group(cache, 0);
10468 btrfs_put_block_group(cache);
10471 ret = wait_block_group_cache_done(cache);
10473 btrfs_put_block_group(cache);
10477 ret = btrfs_trim_block_group(cache,
10483 trimmed += group_trimmed;
10485 btrfs_put_block_group(cache);
10490 cache = next_block_group(fs_info->tree_root, cache);
10493 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10494 devices = &root->fs_info->fs_devices->alloc_list;
10495 list_for_each_entry(device, devices, dev_alloc_list) {
10496 ret = btrfs_trim_free_extents(device, range->minlen,
10501 trimmed += group_trimmed;
10503 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10505 range->len = trimmed;
10510 * btrfs_{start,end}_write_no_snapshoting() are similar to
10511 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10512 * data into the page cache through nocow before the subvolume is snapshoted,
10513 * but flush the data into disk after the snapshot creation, or to prevent
10514 * operations while snapshoting is ongoing and that cause the snapshot to be
10515 * inconsistent (writes followed by expanding truncates for example).
10517 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10519 percpu_counter_dec(&root->subv_writers->counter);
10521 * Make sure counter is updated before we wake up waiters.
10524 if (waitqueue_active(&root->subv_writers->wait))
10525 wake_up(&root->subv_writers->wait);
10528 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10530 if (atomic_read(&root->will_be_snapshoted))
10533 percpu_counter_inc(&root->subv_writers->counter);
10535 * Make sure counter is updated before we check for snapshot creation.
10538 if (atomic_read(&root->will_be_snapshoted)) {
10539 btrfs_end_write_no_snapshoting(root);