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,
2413 /* Also free its reserved qgroup space */
2414 btrfs_qgroup_free_delayed_ref(root->fs_info,
2415 head->qgroup_ref_root,
2416 head->qgroup_reserved);
2420 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2421 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2422 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2424 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2425 node->type == BTRFS_SHARED_DATA_REF_KEY)
2426 ret = run_delayed_data_ref(trans, root, node, extent_op,
2433 static inline struct btrfs_delayed_ref_node *
2434 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2436 struct btrfs_delayed_ref_node *ref;
2438 if (list_empty(&head->ref_list))
2442 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2443 * This is to prevent a ref count from going down to zero, which deletes
2444 * the extent item from the extent tree, when there still are references
2445 * to add, which would fail because they would not find the extent item.
2447 list_for_each_entry(ref, &head->ref_list, list) {
2448 if (ref->action == BTRFS_ADD_DELAYED_REF)
2452 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2457 * Returns 0 on success or if called with an already aborted transaction.
2458 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2460 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2461 struct btrfs_root *root,
2464 struct btrfs_delayed_ref_root *delayed_refs;
2465 struct btrfs_delayed_ref_node *ref;
2466 struct btrfs_delayed_ref_head *locked_ref = NULL;
2467 struct btrfs_delayed_extent_op *extent_op;
2468 struct btrfs_fs_info *fs_info = root->fs_info;
2469 ktime_t start = ktime_get();
2471 unsigned long count = 0;
2472 unsigned long actual_count = 0;
2473 int must_insert_reserved = 0;
2475 delayed_refs = &trans->transaction->delayed_refs;
2481 spin_lock(&delayed_refs->lock);
2482 locked_ref = btrfs_select_ref_head(trans);
2484 spin_unlock(&delayed_refs->lock);
2488 /* grab the lock that says we are going to process
2489 * all the refs for this head */
2490 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2491 spin_unlock(&delayed_refs->lock);
2493 * we may have dropped the spin lock to get the head
2494 * mutex lock, and that might have given someone else
2495 * time to free the head. If that's true, it has been
2496 * removed from our list and we can move on.
2498 if (ret == -EAGAIN) {
2506 * We need to try and merge add/drops of the same ref since we
2507 * can run into issues with relocate dropping the implicit ref
2508 * and then it being added back again before the drop can
2509 * finish. If we merged anything we need to re-loop so we can
2511 * Or we can get node references of the same type that weren't
2512 * merged when created due to bumps in the tree mod seq, and
2513 * we need to merge them to prevent adding an inline extent
2514 * backref before dropping it (triggering a BUG_ON at
2515 * insert_inline_extent_backref()).
2517 spin_lock(&locked_ref->lock);
2518 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2522 * locked_ref is the head node, so we have to go one
2523 * node back for any delayed ref updates
2525 ref = select_delayed_ref(locked_ref);
2527 if (ref && ref->seq &&
2528 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2529 spin_unlock(&locked_ref->lock);
2530 btrfs_delayed_ref_unlock(locked_ref);
2531 spin_lock(&delayed_refs->lock);
2532 locked_ref->processing = 0;
2533 delayed_refs->num_heads_ready++;
2534 spin_unlock(&delayed_refs->lock);
2542 * record the must insert reserved flag before we
2543 * drop the spin lock.
2545 must_insert_reserved = locked_ref->must_insert_reserved;
2546 locked_ref->must_insert_reserved = 0;
2548 extent_op = locked_ref->extent_op;
2549 locked_ref->extent_op = NULL;
2554 /* All delayed refs have been processed, Go ahead
2555 * and send the head node to run_one_delayed_ref,
2556 * so that any accounting fixes can happen
2558 ref = &locked_ref->node;
2560 if (extent_op && must_insert_reserved) {
2561 btrfs_free_delayed_extent_op(extent_op);
2566 spin_unlock(&locked_ref->lock);
2567 ret = run_delayed_extent_op(trans, root,
2569 btrfs_free_delayed_extent_op(extent_op);
2573 * Need to reset must_insert_reserved if
2574 * there was an error so the abort stuff
2575 * can cleanup the reserved space
2578 if (must_insert_reserved)
2579 locked_ref->must_insert_reserved = 1;
2580 locked_ref->processing = 0;
2581 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2582 btrfs_delayed_ref_unlock(locked_ref);
2589 * Need to drop our head ref lock and re-aqcuire the
2590 * delayed ref lock and then re-check to make sure
2593 spin_unlock(&locked_ref->lock);
2594 spin_lock(&delayed_refs->lock);
2595 spin_lock(&locked_ref->lock);
2596 if (!list_empty(&locked_ref->ref_list) ||
2597 locked_ref->extent_op) {
2598 spin_unlock(&locked_ref->lock);
2599 spin_unlock(&delayed_refs->lock);
2603 delayed_refs->num_heads--;
2604 rb_erase(&locked_ref->href_node,
2605 &delayed_refs->href_root);
2606 spin_unlock(&delayed_refs->lock);
2610 list_del(&ref->list);
2612 atomic_dec(&delayed_refs->num_entries);
2614 if (!btrfs_delayed_ref_is_head(ref)) {
2616 * when we play the delayed ref, also correct the
2619 switch (ref->action) {
2620 case BTRFS_ADD_DELAYED_REF:
2621 case BTRFS_ADD_DELAYED_EXTENT:
2622 locked_ref->node.ref_mod -= ref->ref_mod;
2624 case BTRFS_DROP_DELAYED_REF:
2625 locked_ref->node.ref_mod += ref->ref_mod;
2631 spin_unlock(&locked_ref->lock);
2633 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2634 must_insert_reserved);
2636 btrfs_free_delayed_extent_op(extent_op);
2638 locked_ref->processing = 0;
2639 btrfs_delayed_ref_unlock(locked_ref);
2640 btrfs_put_delayed_ref(ref);
2641 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2646 * If this node is a head, that means all the refs in this head
2647 * have been dealt with, and we will pick the next head to deal
2648 * with, so we must unlock the head and drop it from the cluster
2649 * list before we release it.
2651 if (btrfs_delayed_ref_is_head(ref)) {
2652 if (locked_ref->is_data &&
2653 locked_ref->total_ref_mod < 0) {
2654 spin_lock(&delayed_refs->lock);
2655 delayed_refs->pending_csums -= ref->num_bytes;
2656 spin_unlock(&delayed_refs->lock);
2658 btrfs_delayed_ref_unlock(locked_ref);
2661 btrfs_put_delayed_ref(ref);
2667 * We don't want to include ref heads since we can have empty ref heads
2668 * and those will drastically skew our runtime down since we just do
2669 * accounting, no actual extent tree updates.
2671 if (actual_count > 0) {
2672 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2676 * We weigh the current average higher than our current runtime
2677 * to avoid large swings in the average.
2679 spin_lock(&delayed_refs->lock);
2680 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2681 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2682 spin_unlock(&delayed_refs->lock);
2687 #ifdef SCRAMBLE_DELAYED_REFS
2689 * Normally delayed refs get processed in ascending bytenr order. This
2690 * correlates in most cases to the order added. To expose dependencies on this
2691 * order, we start to process the tree in the middle instead of the beginning
2693 static u64 find_middle(struct rb_root *root)
2695 struct rb_node *n = root->rb_node;
2696 struct btrfs_delayed_ref_node *entry;
2699 u64 first = 0, last = 0;
2703 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2704 first = entry->bytenr;
2708 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2709 last = entry->bytenr;
2714 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2715 WARN_ON(!entry->in_tree);
2717 middle = entry->bytenr;
2730 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2734 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2735 sizeof(struct btrfs_extent_inline_ref));
2736 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2737 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2740 * We don't ever fill up leaves all the way so multiply by 2 just to be
2741 * closer to what we're really going to want to ouse.
2743 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2747 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2748 * would require to store the csums for that many bytes.
2750 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2753 u64 num_csums_per_leaf;
2756 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2757 num_csums_per_leaf = div64_u64(csum_size,
2758 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2759 num_csums = div64_u64(csum_bytes, root->sectorsize);
2760 num_csums += num_csums_per_leaf - 1;
2761 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2765 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2766 struct btrfs_root *root)
2768 struct btrfs_block_rsv *global_rsv;
2769 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2770 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2771 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2772 u64 num_bytes, num_dirty_bgs_bytes;
2775 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2776 num_heads = heads_to_leaves(root, num_heads);
2778 num_bytes += (num_heads - 1) * root->nodesize;
2780 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2781 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2783 global_rsv = &root->fs_info->global_block_rsv;
2786 * If we can't allocate any more chunks lets make sure we have _lots_ of
2787 * wiggle room since running delayed refs can create more delayed refs.
2789 if (global_rsv->space_info->full) {
2790 num_dirty_bgs_bytes <<= 1;
2794 spin_lock(&global_rsv->lock);
2795 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2797 spin_unlock(&global_rsv->lock);
2801 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2802 struct btrfs_root *root)
2804 struct btrfs_fs_info *fs_info = root->fs_info;
2806 atomic_read(&trans->transaction->delayed_refs.num_entries);
2811 avg_runtime = fs_info->avg_delayed_ref_runtime;
2812 val = num_entries * avg_runtime;
2813 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2815 if (val >= NSEC_PER_SEC / 2)
2818 return btrfs_check_space_for_delayed_refs(trans, root);
2821 struct async_delayed_refs {
2822 struct btrfs_root *root;
2826 struct completion wait;
2827 struct btrfs_work work;
2830 static void delayed_ref_async_start(struct btrfs_work *work)
2832 struct async_delayed_refs *async;
2833 struct btrfs_trans_handle *trans;
2836 async = container_of(work, struct async_delayed_refs, work);
2838 trans = btrfs_join_transaction(async->root);
2839 if (IS_ERR(trans)) {
2840 async->error = PTR_ERR(trans);
2845 * trans->sync means that when we call end_transaciton, we won't
2846 * wait on delayed refs
2849 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2853 ret = btrfs_end_transaction(trans, async->root);
2854 if (ret && !async->error)
2858 complete(&async->wait);
2863 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2864 unsigned long count, int wait)
2866 struct async_delayed_refs *async;
2869 async = kmalloc(sizeof(*async), GFP_NOFS);
2873 async->root = root->fs_info->tree_root;
2874 async->count = count;
2880 init_completion(&async->wait);
2882 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2883 delayed_ref_async_start, NULL, NULL);
2885 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2888 wait_for_completion(&async->wait);
2897 * this starts processing the delayed reference count updates and
2898 * extent insertions we have queued up so far. count can be
2899 * 0, which means to process everything in the tree at the start
2900 * of the run (but not newly added entries), or it can be some target
2901 * number you'd like to process.
2903 * Returns 0 on success or if called with an aborted transaction
2904 * Returns <0 on error and aborts the transaction
2906 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2907 struct btrfs_root *root, unsigned long count)
2909 struct rb_node *node;
2910 struct btrfs_delayed_ref_root *delayed_refs;
2911 struct btrfs_delayed_ref_head *head;
2913 int run_all = count == (unsigned long)-1;
2914 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2916 /* We'll clean this up in btrfs_cleanup_transaction */
2920 if (root == root->fs_info->extent_root)
2921 root = root->fs_info->tree_root;
2923 delayed_refs = &trans->transaction->delayed_refs;
2925 count = atomic_read(&delayed_refs->num_entries) * 2;
2928 #ifdef SCRAMBLE_DELAYED_REFS
2929 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2931 trans->can_flush_pending_bgs = false;
2932 ret = __btrfs_run_delayed_refs(trans, root, count);
2934 btrfs_abort_transaction(trans, root, ret);
2939 if (!list_empty(&trans->new_bgs))
2940 btrfs_create_pending_block_groups(trans, root);
2942 spin_lock(&delayed_refs->lock);
2943 node = rb_first(&delayed_refs->href_root);
2945 spin_unlock(&delayed_refs->lock);
2948 count = (unsigned long)-1;
2951 head = rb_entry(node, struct btrfs_delayed_ref_head,
2953 if (btrfs_delayed_ref_is_head(&head->node)) {
2954 struct btrfs_delayed_ref_node *ref;
2957 atomic_inc(&ref->refs);
2959 spin_unlock(&delayed_refs->lock);
2961 * Mutex was contended, block until it's
2962 * released and try again
2964 mutex_lock(&head->mutex);
2965 mutex_unlock(&head->mutex);
2967 btrfs_put_delayed_ref(ref);
2973 node = rb_next(node);
2975 spin_unlock(&delayed_refs->lock);
2980 assert_qgroups_uptodate(trans);
2981 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2985 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2986 struct btrfs_root *root,
2987 u64 bytenr, u64 num_bytes, u64 flags,
2988 int level, int is_data)
2990 struct btrfs_delayed_extent_op *extent_op;
2993 extent_op = btrfs_alloc_delayed_extent_op();
2997 extent_op->flags_to_set = flags;
2998 extent_op->update_flags = 1;
2999 extent_op->update_key = 0;
3000 extent_op->is_data = is_data ? 1 : 0;
3001 extent_op->level = level;
3003 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
3004 num_bytes, extent_op);
3006 btrfs_free_delayed_extent_op(extent_op);
3010 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
3011 struct btrfs_root *root,
3012 struct btrfs_path *path,
3013 u64 objectid, u64 offset, u64 bytenr)
3015 struct btrfs_delayed_ref_head *head;
3016 struct btrfs_delayed_ref_node *ref;
3017 struct btrfs_delayed_data_ref *data_ref;
3018 struct btrfs_delayed_ref_root *delayed_refs;
3021 delayed_refs = &trans->transaction->delayed_refs;
3022 spin_lock(&delayed_refs->lock);
3023 head = btrfs_find_delayed_ref_head(trans, bytenr);
3025 spin_unlock(&delayed_refs->lock);
3029 if (!mutex_trylock(&head->mutex)) {
3030 atomic_inc(&head->node.refs);
3031 spin_unlock(&delayed_refs->lock);
3033 btrfs_release_path(path);
3036 * Mutex was contended, block until it's released and let
3039 mutex_lock(&head->mutex);
3040 mutex_unlock(&head->mutex);
3041 btrfs_put_delayed_ref(&head->node);
3044 spin_unlock(&delayed_refs->lock);
3046 spin_lock(&head->lock);
3047 list_for_each_entry(ref, &head->ref_list, list) {
3048 /* If it's a shared ref we know a cross reference exists */
3049 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3054 data_ref = btrfs_delayed_node_to_data_ref(ref);
3057 * If our ref doesn't match the one we're currently looking at
3058 * then we have a cross reference.
3060 if (data_ref->root != root->root_key.objectid ||
3061 data_ref->objectid != objectid ||
3062 data_ref->offset != offset) {
3067 spin_unlock(&head->lock);
3068 mutex_unlock(&head->mutex);
3072 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3073 struct btrfs_root *root,
3074 struct btrfs_path *path,
3075 u64 objectid, u64 offset, u64 bytenr)
3077 struct btrfs_root *extent_root = root->fs_info->extent_root;
3078 struct extent_buffer *leaf;
3079 struct btrfs_extent_data_ref *ref;
3080 struct btrfs_extent_inline_ref *iref;
3081 struct btrfs_extent_item *ei;
3082 struct btrfs_key key;
3086 key.objectid = bytenr;
3087 key.offset = (u64)-1;
3088 key.type = BTRFS_EXTENT_ITEM_KEY;
3090 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3093 BUG_ON(ret == 0); /* Corruption */
3096 if (path->slots[0] == 0)
3100 leaf = path->nodes[0];
3101 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3103 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3107 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3108 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3109 if (item_size < sizeof(*ei)) {
3110 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3114 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3116 if (item_size != sizeof(*ei) +
3117 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3120 if (btrfs_extent_generation(leaf, ei) <=
3121 btrfs_root_last_snapshot(&root->root_item))
3124 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3125 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3126 BTRFS_EXTENT_DATA_REF_KEY)
3129 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3130 if (btrfs_extent_refs(leaf, ei) !=
3131 btrfs_extent_data_ref_count(leaf, ref) ||
3132 btrfs_extent_data_ref_root(leaf, ref) !=
3133 root->root_key.objectid ||
3134 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3135 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3143 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3144 struct btrfs_root *root,
3145 u64 objectid, u64 offset, u64 bytenr)
3147 struct btrfs_path *path;
3151 path = btrfs_alloc_path();
3156 ret = check_committed_ref(trans, root, path, objectid,
3158 if (ret && ret != -ENOENT)
3161 ret2 = check_delayed_ref(trans, root, path, objectid,
3163 } while (ret2 == -EAGAIN);
3165 if (ret2 && ret2 != -ENOENT) {
3170 if (ret != -ENOENT || ret2 != -ENOENT)
3173 btrfs_free_path(path);
3174 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3179 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3180 struct btrfs_root *root,
3181 struct extent_buffer *buf,
3182 int full_backref, int inc)
3189 struct btrfs_key key;
3190 struct btrfs_file_extent_item *fi;
3194 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3195 u64, u64, u64, u64, u64, u64, int);
3198 if (btrfs_test_is_dummy_root(root))
3201 ref_root = btrfs_header_owner(buf);
3202 nritems = btrfs_header_nritems(buf);
3203 level = btrfs_header_level(buf);
3205 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3209 process_func = btrfs_inc_extent_ref;
3211 process_func = btrfs_free_extent;
3214 parent = buf->start;
3218 for (i = 0; i < nritems; i++) {
3220 btrfs_item_key_to_cpu(buf, &key, i);
3221 if (key.type != BTRFS_EXTENT_DATA_KEY)
3223 fi = btrfs_item_ptr(buf, i,
3224 struct btrfs_file_extent_item);
3225 if (btrfs_file_extent_type(buf, fi) ==
3226 BTRFS_FILE_EXTENT_INLINE)
3228 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3232 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3233 key.offset -= btrfs_file_extent_offset(buf, fi);
3234 ret = process_func(trans, root, bytenr, num_bytes,
3235 parent, ref_root, key.objectid,
3240 bytenr = btrfs_node_blockptr(buf, i);
3241 num_bytes = root->nodesize;
3242 ret = process_func(trans, root, bytenr, num_bytes,
3243 parent, ref_root, level - 1, 0,
3254 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3255 struct extent_buffer *buf, int full_backref)
3257 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3260 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3261 struct extent_buffer *buf, int full_backref)
3263 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3266 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3267 struct btrfs_root *root,
3268 struct btrfs_path *path,
3269 struct btrfs_block_group_cache *cache)
3272 struct btrfs_root *extent_root = root->fs_info->extent_root;
3274 struct extent_buffer *leaf;
3276 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3283 leaf = path->nodes[0];
3284 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3285 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3286 btrfs_mark_buffer_dirty(leaf);
3288 btrfs_release_path(path);
3293 static struct btrfs_block_group_cache *
3294 next_block_group(struct btrfs_root *root,
3295 struct btrfs_block_group_cache *cache)
3297 struct rb_node *node;
3299 spin_lock(&root->fs_info->block_group_cache_lock);
3301 /* If our block group was removed, we need a full search. */
3302 if (RB_EMPTY_NODE(&cache->cache_node)) {
3303 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3305 spin_unlock(&root->fs_info->block_group_cache_lock);
3306 btrfs_put_block_group(cache);
3307 cache = btrfs_lookup_first_block_group(root->fs_info,
3311 node = rb_next(&cache->cache_node);
3312 btrfs_put_block_group(cache);
3314 cache = rb_entry(node, struct btrfs_block_group_cache,
3316 btrfs_get_block_group(cache);
3319 spin_unlock(&root->fs_info->block_group_cache_lock);
3323 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3324 struct btrfs_trans_handle *trans,
3325 struct btrfs_path *path)
3327 struct btrfs_root *root = block_group->fs_info->tree_root;
3328 struct inode *inode = NULL;
3330 int dcs = BTRFS_DC_ERROR;
3336 * If this block group is smaller than 100 megs don't bother caching the
3339 if (block_group->key.offset < (100 * 1024 * 1024)) {
3340 spin_lock(&block_group->lock);
3341 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3342 spin_unlock(&block_group->lock);
3349 inode = lookup_free_space_inode(root, block_group, path);
3350 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3351 ret = PTR_ERR(inode);
3352 btrfs_release_path(path);
3356 if (IS_ERR(inode)) {
3360 if (block_group->ro)
3363 ret = create_free_space_inode(root, trans, block_group, path);
3369 /* We've already setup this transaction, go ahead and exit */
3370 if (block_group->cache_generation == trans->transid &&
3371 i_size_read(inode)) {
3372 dcs = BTRFS_DC_SETUP;
3377 * We want to set the generation to 0, that way if anything goes wrong
3378 * from here on out we know not to trust this cache when we load up next
3381 BTRFS_I(inode)->generation = 0;
3382 ret = btrfs_update_inode(trans, root, inode);
3385 * So theoretically we could recover from this, simply set the
3386 * super cache generation to 0 so we know to invalidate the
3387 * cache, but then we'd have to keep track of the block groups
3388 * that fail this way so we know we _have_ to reset this cache
3389 * before the next commit or risk reading stale cache. So to
3390 * limit our exposure to horrible edge cases lets just abort the
3391 * transaction, this only happens in really bad situations
3394 btrfs_abort_transaction(trans, root, ret);
3399 if (i_size_read(inode) > 0) {
3400 ret = btrfs_check_trunc_cache_free_space(root,
3401 &root->fs_info->global_block_rsv);
3405 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3410 spin_lock(&block_group->lock);
3411 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3412 !btrfs_test_opt(root, SPACE_CACHE)) {
3414 * don't bother trying to write stuff out _if_
3415 * a) we're not cached,
3416 * b) we're with nospace_cache mount option.
3418 dcs = BTRFS_DC_WRITTEN;
3419 spin_unlock(&block_group->lock);
3422 spin_unlock(&block_group->lock);
3425 * We hit an ENOSPC when setting up the cache in this transaction, just
3426 * skip doing the setup, we've already cleared the cache so we're safe.
3428 if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) {
3434 * Try to preallocate enough space based on how big the block group is.
3435 * Keep in mind this has to include any pinned space which could end up
3436 * taking up quite a bit since it's not folded into the other space
3439 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3444 num_pages *= PAGE_CACHE_SIZE;
3446 ret = btrfs_check_data_free_space(inode, 0, num_pages);
3450 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3451 num_pages, num_pages,
3454 * Our cache requires contiguous chunks so that we don't modify a bunch
3455 * of metadata or split extents when writing the cache out, which means
3456 * we can enospc if we are heavily fragmented in addition to just normal
3457 * out of space conditions. So if we hit this just skip setting up any
3458 * other block groups for this transaction, maybe we'll unpin enough
3459 * space the next time around.
3462 dcs = BTRFS_DC_SETUP;
3463 else if (ret == -ENOSPC)
3464 set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags);
3465 btrfs_free_reserved_data_space(inode, 0, num_pages);
3470 btrfs_release_path(path);
3472 spin_lock(&block_group->lock);
3473 if (!ret && dcs == BTRFS_DC_SETUP)
3474 block_group->cache_generation = trans->transid;
3475 block_group->disk_cache_state = dcs;
3476 spin_unlock(&block_group->lock);
3481 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3482 struct btrfs_root *root)
3484 struct btrfs_block_group_cache *cache, *tmp;
3485 struct btrfs_transaction *cur_trans = trans->transaction;
3486 struct btrfs_path *path;
3488 if (list_empty(&cur_trans->dirty_bgs) ||
3489 !btrfs_test_opt(root, SPACE_CACHE))
3492 path = btrfs_alloc_path();
3496 /* Could add new block groups, use _safe just in case */
3497 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3499 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3500 cache_save_setup(cache, trans, path);
3503 btrfs_free_path(path);
3508 * transaction commit does final block group cache writeback during a
3509 * critical section where nothing is allowed to change the FS. This is
3510 * required in order for the cache to actually match the block group,
3511 * but can introduce a lot of latency into the commit.
3513 * So, btrfs_start_dirty_block_groups is here to kick off block group
3514 * cache IO. There's a chance we'll have to redo some of it if the
3515 * block group changes again during the commit, but it greatly reduces
3516 * the commit latency by getting rid of the easy block groups while
3517 * we're still allowing others to join the commit.
3519 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3520 struct btrfs_root *root)
3522 struct btrfs_block_group_cache *cache;
3523 struct btrfs_transaction *cur_trans = trans->transaction;
3526 struct btrfs_path *path = NULL;
3528 struct list_head *io = &cur_trans->io_bgs;
3529 int num_started = 0;
3532 spin_lock(&cur_trans->dirty_bgs_lock);
3533 if (list_empty(&cur_trans->dirty_bgs)) {
3534 spin_unlock(&cur_trans->dirty_bgs_lock);
3537 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3538 spin_unlock(&cur_trans->dirty_bgs_lock);
3542 * make sure all the block groups on our dirty list actually
3545 btrfs_create_pending_block_groups(trans, root);
3548 path = btrfs_alloc_path();
3554 * cache_write_mutex is here only to save us from balance or automatic
3555 * removal of empty block groups deleting this block group while we are
3556 * writing out the cache
3558 mutex_lock(&trans->transaction->cache_write_mutex);
3559 while (!list_empty(&dirty)) {
3560 cache = list_first_entry(&dirty,
3561 struct btrfs_block_group_cache,
3564 * this can happen if something re-dirties a block
3565 * group that is already under IO. Just wait for it to
3566 * finish and then do it all again
3568 if (!list_empty(&cache->io_list)) {
3569 list_del_init(&cache->io_list);
3570 btrfs_wait_cache_io(root, trans, cache,
3571 &cache->io_ctl, path,
3572 cache->key.objectid);
3573 btrfs_put_block_group(cache);
3578 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3579 * if it should update the cache_state. Don't delete
3580 * until after we wait.
3582 * Since we're not running in the commit critical section
3583 * we need the dirty_bgs_lock to protect from update_block_group
3585 spin_lock(&cur_trans->dirty_bgs_lock);
3586 list_del_init(&cache->dirty_list);
3587 spin_unlock(&cur_trans->dirty_bgs_lock);
3591 cache_save_setup(cache, trans, path);
3593 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3594 cache->io_ctl.inode = NULL;
3595 ret = btrfs_write_out_cache(root, trans, cache, path);
3596 if (ret == 0 && cache->io_ctl.inode) {
3601 * the cache_write_mutex is protecting
3604 list_add_tail(&cache->io_list, io);
3607 * if we failed to write the cache, the
3608 * generation will be bad and life goes on
3614 ret = write_one_cache_group(trans, root, path, cache);
3616 * Our block group might still be attached to the list
3617 * of new block groups in the transaction handle of some
3618 * other task (struct btrfs_trans_handle->new_bgs). This
3619 * means its block group item isn't yet in the extent
3620 * tree. If this happens ignore the error, as we will
3621 * try again later in the critical section of the
3622 * transaction commit.
3624 if (ret == -ENOENT) {
3626 spin_lock(&cur_trans->dirty_bgs_lock);
3627 if (list_empty(&cache->dirty_list)) {
3628 list_add_tail(&cache->dirty_list,
3629 &cur_trans->dirty_bgs);
3630 btrfs_get_block_group(cache);
3632 spin_unlock(&cur_trans->dirty_bgs_lock);
3634 btrfs_abort_transaction(trans, root, ret);
3638 /* if its not on the io list, we need to put the block group */
3640 btrfs_put_block_group(cache);
3646 * Avoid blocking other tasks for too long. It might even save
3647 * us from writing caches for block groups that are going to be
3650 mutex_unlock(&trans->transaction->cache_write_mutex);
3651 mutex_lock(&trans->transaction->cache_write_mutex);
3653 mutex_unlock(&trans->transaction->cache_write_mutex);
3656 * go through delayed refs for all the stuff we've just kicked off
3657 * and then loop back (just once)
3659 ret = btrfs_run_delayed_refs(trans, root, 0);
3660 if (!ret && loops == 0) {
3662 spin_lock(&cur_trans->dirty_bgs_lock);
3663 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3665 * dirty_bgs_lock protects us from concurrent block group
3666 * deletes too (not just cache_write_mutex).
3668 if (!list_empty(&dirty)) {
3669 spin_unlock(&cur_trans->dirty_bgs_lock);
3672 spin_unlock(&cur_trans->dirty_bgs_lock);
3675 btrfs_free_path(path);
3679 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3680 struct btrfs_root *root)
3682 struct btrfs_block_group_cache *cache;
3683 struct btrfs_transaction *cur_trans = trans->transaction;
3686 struct btrfs_path *path;
3687 struct list_head *io = &cur_trans->io_bgs;
3688 int num_started = 0;
3690 path = btrfs_alloc_path();
3695 * We don't need the lock here since we are protected by the transaction
3696 * commit. We want to do the cache_save_setup first and then run the
3697 * delayed refs to make sure we have the best chance at doing this all
3700 while (!list_empty(&cur_trans->dirty_bgs)) {
3701 cache = list_first_entry(&cur_trans->dirty_bgs,
3702 struct btrfs_block_group_cache,
3706 * this can happen if cache_save_setup re-dirties a block
3707 * group that is already under IO. Just wait for it to
3708 * finish and then do it all again
3710 if (!list_empty(&cache->io_list)) {
3711 list_del_init(&cache->io_list);
3712 btrfs_wait_cache_io(root, trans, cache,
3713 &cache->io_ctl, path,
3714 cache->key.objectid);
3715 btrfs_put_block_group(cache);
3719 * don't remove from the dirty list until after we've waited
3722 list_del_init(&cache->dirty_list);
3725 cache_save_setup(cache, trans, path);
3728 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3730 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3731 cache->io_ctl.inode = NULL;
3732 ret = btrfs_write_out_cache(root, trans, cache, path);
3733 if (ret == 0 && cache->io_ctl.inode) {
3736 list_add_tail(&cache->io_list, io);
3739 * if we failed to write the cache, the
3740 * generation will be bad and life goes on
3746 ret = write_one_cache_group(trans, root, path, cache);
3748 btrfs_abort_transaction(trans, root, ret);
3751 /* if its not on the io list, we need to put the block group */
3753 btrfs_put_block_group(cache);
3756 while (!list_empty(io)) {
3757 cache = list_first_entry(io, struct btrfs_block_group_cache,
3759 list_del_init(&cache->io_list);
3760 btrfs_wait_cache_io(root, trans, cache,
3761 &cache->io_ctl, path, cache->key.objectid);
3762 btrfs_put_block_group(cache);
3765 btrfs_free_path(path);
3769 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3771 struct btrfs_block_group_cache *block_group;
3774 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3775 if (!block_group || block_group->ro)
3778 btrfs_put_block_group(block_group);
3782 static const char *alloc_name(u64 flags)
3785 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3787 case BTRFS_BLOCK_GROUP_METADATA:
3789 case BTRFS_BLOCK_GROUP_DATA:
3791 case BTRFS_BLOCK_GROUP_SYSTEM:
3795 return "invalid-combination";
3799 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3800 u64 total_bytes, u64 bytes_used,
3801 struct btrfs_space_info **space_info)
3803 struct btrfs_space_info *found;
3808 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3809 BTRFS_BLOCK_GROUP_RAID10))
3814 found = __find_space_info(info, flags);
3816 spin_lock(&found->lock);
3817 found->total_bytes += total_bytes;
3818 found->disk_total += total_bytes * factor;
3819 found->bytes_used += bytes_used;
3820 found->disk_used += bytes_used * factor;
3821 if (total_bytes > 0)
3823 spin_unlock(&found->lock);
3824 *space_info = found;
3827 found = kzalloc(sizeof(*found), GFP_NOFS);
3831 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3837 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3838 INIT_LIST_HEAD(&found->block_groups[i]);
3839 init_rwsem(&found->groups_sem);
3840 spin_lock_init(&found->lock);
3841 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3842 found->total_bytes = total_bytes;
3843 found->disk_total = total_bytes * factor;
3844 found->bytes_used = bytes_used;
3845 found->disk_used = bytes_used * factor;
3846 found->bytes_pinned = 0;
3847 found->bytes_reserved = 0;
3848 found->bytes_readonly = 0;
3849 found->bytes_may_use = 0;
3851 found->max_extent_size = 0;
3852 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3853 found->chunk_alloc = 0;
3855 init_waitqueue_head(&found->wait);
3856 INIT_LIST_HEAD(&found->ro_bgs);
3858 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3859 info->space_info_kobj, "%s",
3860 alloc_name(found->flags));
3866 *space_info = found;
3867 list_add_rcu(&found->list, &info->space_info);
3868 if (flags & BTRFS_BLOCK_GROUP_DATA)
3869 info->data_sinfo = found;
3874 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3876 u64 extra_flags = chunk_to_extended(flags) &
3877 BTRFS_EXTENDED_PROFILE_MASK;
3879 write_seqlock(&fs_info->profiles_lock);
3880 if (flags & BTRFS_BLOCK_GROUP_DATA)
3881 fs_info->avail_data_alloc_bits |= extra_flags;
3882 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3883 fs_info->avail_metadata_alloc_bits |= extra_flags;
3884 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3885 fs_info->avail_system_alloc_bits |= extra_flags;
3886 write_sequnlock(&fs_info->profiles_lock);
3890 * returns target flags in extended format or 0 if restripe for this
3891 * chunk_type is not in progress
3893 * should be called with either volume_mutex or balance_lock held
3895 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3897 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3903 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3904 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3905 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3906 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3907 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3908 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3909 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3910 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3911 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3918 * @flags: available profiles in extended format (see ctree.h)
3920 * Returns reduced profile in chunk format. If profile changing is in
3921 * progress (either running or paused) picks the target profile (if it's
3922 * already available), otherwise falls back to plain reducing.
3924 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3926 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3932 * see if restripe for this chunk_type is in progress, if so
3933 * try to reduce to the target profile
3935 spin_lock(&root->fs_info->balance_lock);
3936 target = get_restripe_target(root->fs_info, flags);
3938 /* pick target profile only if it's already available */
3939 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3940 spin_unlock(&root->fs_info->balance_lock);
3941 return extended_to_chunk(target);
3944 spin_unlock(&root->fs_info->balance_lock);
3946 /* First, mask out the RAID levels which aren't possible */
3947 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3948 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
3949 allowed |= btrfs_raid_group[raid_type];
3953 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
3954 allowed = BTRFS_BLOCK_GROUP_RAID6;
3955 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
3956 allowed = BTRFS_BLOCK_GROUP_RAID5;
3957 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
3958 allowed = BTRFS_BLOCK_GROUP_RAID10;
3959 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
3960 allowed = BTRFS_BLOCK_GROUP_RAID1;
3961 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
3962 allowed = BTRFS_BLOCK_GROUP_RAID0;
3964 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
3966 return extended_to_chunk(flags | allowed);
3969 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3976 seq = read_seqbegin(&root->fs_info->profiles_lock);
3978 if (flags & BTRFS_BLOCK_GROUP_DATA)
3979 flags |= root->fs_info->avail_data_alloc_bits;
3980 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3981 flags |= root->fs_info->avail_system_alloc_bits;
3982 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3983 flags |= root->fs_info->avail_metadata_alloc_bits;
3984 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3986 return btrfs_reduce_alloc_profile(root, flags);
3989 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3995 flags = BTRFS_BLOCK_GROUP_DATA;
3996 else if (root == root->fs_info->chunk_root)
3997 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3999 flags = BTRFS_BLOCK_GROUP_METADATA;
4001 ret = get_alloc_profile(root, flags);
4005 int btrfs_alloc_data_chunk_ondemand(struct inode *inode, u64 bytes)
4007 struct btrfs_space_info *data_sinfo;
4008 struct btrfs_root *root = BTRFS_I(inode)->root;
4009 struct btrfs_fs_info *fs_info = root->fs_info;
4012 int need_commit = 2;
4013 int have_pinned_space;
4015 /* make sure bytes are sectorsize aligned */
4016 bytes = ALIGN(bytes, root->sectorsize);
4018 if (btrfs_is_free_space_inode(inode)) {
4020 ASSERT(current->journal_info);
4023 data_sinfo = fs_info->data_sinfo;
4028 /* make sure we have enough space to handle the data first */
4029 spin_lock(&data_sinfo->lock);
4030 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
4031 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
4032 data_sinfo->bytes_may_use;
4034 if (used + bytes > data_sinfo->total_bytes) {
4035 struct btrfs_trans_handle *trans;
4038 * if we don't have enough free bytes in this space then we need
4039 * to alloc a new chunk.
4041 if (!data_sinfo->full) {
4044 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4045 spin_unlock(&data_sinfo->lock);
4047 alloc_target = btrfs_get_alloc_profile(root, 1);
4049 * It is ugly that we don't call nolock join
4050 * transaction for the free space inode case here.
4051 * But it is safe because we only do the data space
4052 * reservation for the free space cache in the
4053 * transaction context, the common join transaction
4054 * just increase the counter of the current transaction
4055 * handler, doesn't try to acquire the trans_lock of
4058 trans = btrfs_join_transaction(root);
4060 return PTR_ERR(trans);
4062 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4064 CHUNK_ALLOC_NO_FORCE);
4065 btrfs_end_transaction(trans, root);
4070 have_pinned_space = 1;
4076 data_sinfo = fs_info->data_sinfo;
4082 * If we don't have enough pinned space to deal with this
4083 * allocation, and no removed chunk in current transaction,
4084 * don't bother committing the transaction.
4086 have_pinned_space = percpu_counter_compare(
4087 &data_sinfo->total_bytes_pinned,
4088 used + bytes - data_sinfo->total_bytes);
4089 spin_unlock(&data_sinfo->lock);
4091 /* commit the current transaction and try again */
4094 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4097 if (need_commit > 0)
4098 btrfs_wait_ordered_roots(fs_info, -1);
4100 trans = btrfs_join_transaction(root);
4102 return PTR_ERR(trans);
4103 if (have_pinned_space >= 0 ||
4104 test_bit(BTRFS_TRANS_HAVE_FREE_BGS,
4105 &trans->transaction->flags) ||
4107 ret = btrfs_commit_transaction(trans, root);
4111 * make sure that all running delayed iput are
4114 down_write(&root->fs_info->delayed_iput_sem);
4115 up_write(&root->fs_info->delayed_iput_sem);
4118 btrfs_end_transaction(trans, root);
4122 trace_btrfs_space_reservation(root->fs_info,
4123 "space_info:enospc",
4124 data_sinfo->flags, bytes, 1);
4127 data_sinfo->bytes_may_use += bytes;
4128 trace_btrfs_space_reservation(root->fs_info, "space_info",
4129 data_sinfo->flags, bytes, 1);
4130 spin_unlock(&data_sinfo->lock);
4136 * New check_data_free_space() with ability for precious data reservation
4137 * Will replace old btrfs_check_data_free_space(), but for patch split,
4138 * add a new function first and then replace it.
4140 int btrfs_check_data_free_space(struct inode *inode, u64 start, u64 len)
4142 struct btrfs_root *root = BTRFS_I(inode)->root;
4145 /* align the range */
4146 len = round_up(start + len, root->sectorsize) -
4147 round_down(start, root->sectorsize);
4148 start = round_down(start, root->sectorsize);
4150 ret = btrfs_alloc_data_chunk_ondemand(inode, len);
4155 * Use new btrfs_qgroup_reserve_data to reserve precious data space
4157 * TODO: Find a good method to avoid reserve data space for NOCOW
4158 * range, but don't impact performance on quota disable case.
4160 ret = btrfs_qgroup_reserve_data(inode, start, len);
4165 * Called if we need to clear a data reservation for this inode
4166 * Normally in a error case.
4168 * This one will *NOT* use accurate qgroup reserved space API, just for case
4169 * which we can't sleep and is sure it won't affect qgroup reserved space.
4170 * Like clear_bit_hook().
4172 void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start,
4175 struct btrfs_root *root = BTRFS_I(inode)->root;
4176 struct btrfs_space_info *data_sinfo;
4178 /* Make sure the range is aligned to sectorsize */
4179 len = round_up(start + len, root->sectorsize) -
4180 round_down(start, root->sectorsize);
4181 start = round_down(start, root->sectorsize);
4183 data_sinfo = root->fs_info->data_sinfo;
4184 spin_lock(&data_sinfo->lock);
4185 if (WARN_ON(data_sinfo->bytes_may_use < len))
4186 data_sinfo->bytes_may_use = 0;
4188 data_sinfo->bytes_may_use -= len;
4189 trace_btrfs_space_reservation(root->fs_info, "space_info",
4190 data_sinfo->flags, len, 0);
4191 spin_unlock(&data_sinfo->lock);
4195 * Called if we need to clear a data reservation for this inode
4196 * Normally in a error case.
4198 * This one will handle the per-indoe data rsv map for accurate reserved
4201 void btrfs_free_reserved_data_space(struct inode *inode, u64 start, u64 len)
4203 btrfs_free_reserved_data_space_noquota(inode, start, len);
4204 btrfs_qgroup_free_data(inode, start, len);
4207 static void force_metadata_allocation(struct btrfs_fs_info *info)
4209 struct list_head *head = &info->space_info;
4210 struct btrfs_space_info *found;
4213 list_for_each_entry_rcu(found, head, list) {
4214 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4215 found->force_alloc = CHUNK_ALLOC_FORCE;
4220 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4222 return (global->size << 1);
4225 static int should_alloc_chunk(struct btrfs_root *root,
4226 struct btrfs_space_info *sinfo, int force)
4228 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4229 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4230 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4233 if (force == CHUNK_ALLOC_FORCE)
4237 * We need to take into account the global rsv because for all intents
4238 * and purposes it's used space. Don't worry about locking the
4239 * global_rsv, it doesn't change except when the transaction commits.
4241 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4242 num_allocated += calc_global_rsv_need_space(global_rsv);
4245 * in limited mode, we want to have some free space up to
4246 * about 1% of the FS size.
4248 if (force == CHUNK_ALLOC_LIMITED) {
4249 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4250 thresh = max_t(u64, 64 * 1024 * 1024,
4251 div_factor_fine(thresh, 1));
4253 if (num_bytes - num_allocated < thresh)
4257 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4262 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4266 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4267 BTRFS_BLOCK_GROUP_RAID0 |
4268 BTRFS_BLOCK_GROUP_RAID5 |
4269 BTRFS_BLOCK_GROUP_RAID6))
4270 num_dev = root->fs_info->fs_devices->rw_devices;
4271 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4274 num_dev = 1; /* DUP or single */
4280 * If @is_allocation is true, reserve space in the system space info necessary
4281 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4284 void check_system_chunk(struct btrfs_trans_handle *trans,
4285 struct btrfs_root *root,
4288 struct btrfs_space_info *info;
4295 * Needed because we can end up allocating a system chunk and for an
4296 * atomic and race free space reservation in the chunk block reserve.
4298 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4300 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4301 spin_lock(&info->lock);
4302 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4303 info->bytes_reserved - info->bytes_readonly -
4304 info->bytes_may_use;
4305 spin_unlock(&info->lock);
4307 num_devs = get_profile_num_devs(root, type);
4309 /* num_devs device items to update and 1 chunk item to add or remove */
4310 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4311 btrfs_calc_trans_metadata_size(root, 1);
4313 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4314 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4315 left, thresh, type);
4316 dump_space_info(info, 0, 0);
4319 if (left < thresh) {
4322 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4324 * Ignore failure to create system chunk. We might end up not
4325 * needing it, as we might not need to COW all nodes/leafs from
4326 * the paths we visit in the chunk tree (they were already COWed
4327 * or created in the current transaction for example).
4329 ret = btrfs_alloc_chunk(trans, root, flags);
4333 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4334 &root->fs_info->chunk_block_rsv,
4335 thresh, BTRFS_RESERVE_NO_FLUSH);
4337 trans->chunk_bytes_reserved += thresh;
4341 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4342 struct btrfs_root *extent_root, u64 flags, int force)
4344 struct btrfs_space_info *space_info;
4345 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4346 int wait_for_alloc = 0;
4349 /* Don't re-enter if we're already allocating a chunk */
4350 if (trans->allocating_chunk)
4353 space_info = __find_space_info(extent_root->fs_info, flags);
4355 ret = update_space_info(extent_root->fs_info, flags,
4357 BUG_ON(ret); /* -ENOMEM */
4359 BUG_ON(!space_info); /* Logic error */
4362 spin_lock(&space_info->lock);
4363 if (force < space_info->force_alloc)
4364 force = space_info->force_alloc;
4365 if (space_info->full) {
4366 if (should_alloc_chunk(extent_root, space_info, force))
4370 spin_unlock(&space_info->lock);
4374 if (!should_alloc_chunk(extent_root, space_info, force)) {
4375 spin_unlock(&space_info->lock);
4377 } else if (space_info->chunk_alloc) {
4380 space_info->chunk_alloc = 1;
4383 spin_unlock(&space_info->lock);
4385 mutex_lock(&fs_info->chunk_mutex);
4388 * The chunk_mutex is held throughout the entirety of a chunk
4389 * allocation, so once we've acquired the chunk_mutex we know that the
4390 * other guy is done and we need to recheck and see if we should
4393 if (wait_for_alloc) {
4394 mutex_unlock(&fs_info->chunk_mutex);
4399 trans->allocating_chunk = true;
4402 * If we have mixed data/metadata chunks we want to make sure we keep
4403 * allocating mixed chunks instead of individual chunks.
4405 if (btrfs_mixed_space_info(space_info))
4406 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4409 * if we're doing a data chunk, go ahead and make sure that
4410 * we keep a reasonable number of metadata chunks allocated in the
4413 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4414 fs_info->data_chunk_allocations++;
4415 if (!(fs_info->data_chunk_allocations %
4416 fs_info->metadata_ratio))
4417 force_metadata_allocation(fs_info);
4421 * Check if we have enough space in SYSTEM chunk because we may need
4422 * to update devices.
4424 check_system_chunk(trans, extent_root, flags);
4426 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4427 trans->allocating_chunk = false;
4429 spin_lock(&space_info->lock);
4430 if (ret < 0 && ret != -ENOSPC)
4433 space_info->full = 1;
4437 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4439 space_info->chunk_alloc = 0;
4440 spin_unlock(&space_info->lock);
4441 mutex_unlock(&fs_info->chunk_mutex);
4443 * When we allocate a new chunk we reserve space in the chunk block
4444 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4445 * add new nodes/leafs to it if we end up needing to do it when
4446 * inserting the chunk item and updating device items as part of the
4447 * second phase of chunk allocation, performed by
4448 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4449 * large number of new block groups to create in our transaction
4450 * handle's new_bgs list to avoid exhausting the chunk block reserve
4451 * in extreme cases - like having a single transaction create many new
4452 * block groups when starting to write out the free space caches of all
4453 * the block groups that were made dirty during the lifetime of the
4456 if (trans->can_flush_pending_bgs &&
4457 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4458 btrfs_create_pending_block_groups(trans, trans->root);
4459 btrfs_trans_release_chunk_metadata(trans);
4464 static int can_overcommit(struct btrfs_root *root,
4465 struct btrfs_space_info *space_info, u64 bytes,
4466 enum btrfs_reserve_flush_enum flush)
4468 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4469 u64 profile = btrfs_get_alloc_profile(root, 0);
4474 used = space_info->bytes_used + space_info->bytes_reserved +
4475 space_info->bytes_pinned + space_info->bytes_readonly;
4478 * We only want to allow over committing if we have lots of actual space
4479 * free, but if we don't have enough space to handle the global reserve
4480 * space then we could end up having a real enospc problem when trying
4481 * to allocate a chunk or some other such important allocation.
4483 spin_lock(&global_rsv->lock);
4484 space_size = calc_global_rsv_need_space(global_rsv);
4485 spin_unlock(&global_rsv->lock);
4486 if (used + space_size >= space_info->total_bytes)
4489 used += space_info->bytes_may_use;
4491 spin_lock(&root->fs_info->free_chunk_lock);
4492 avail = root->fs_info->free_chunk_space;
4493 spin_unlock(&root->fs_info->free_chunk_lock);
4496 * If we have dup, raid1 or raid10 then only half of the free
4497 * space is actually useable. For raid56, the space info used
4498 * doesn't include the parity drive, so we don't have to
4501 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4502 BTRFS_BLOCK_GROUP_RAID1 |
4503 BTRFS_BLOCK_GROUP_RAID10))
4507 * If we aren't flushing all things, let us overcommit up to
4508 * 1/2th of the space. If we can flush, don't let us overcommit
4509 * too much, let it overcommit up to 1/8 of the space.
4511 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4516 if (used + bytes < space_info->total_bytes + avail)
4521 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4522 unsigned long nr_pages, int nr_items)
4524 struct super_block *sb = root->fs_info->sb;
4526 if (down_read_trylock(&sb->s_umount)) {
4527 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4528 up_read(&sb->s_umount);
4531 * We needn't worry the filesystem going from r/w to r/o though
4532 * we don't acquire ->s_umount mutex, because the filesystem
4533 * should guarantee the delalloc inodes list be empty after
4534 * the filesystem is readonly(all dirty pages are written to
4537 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4538 if (!current->journal_info)
4539 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4543 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4548 bytes = btrfs_calc_trans_metadata_size(root, 1);
4549 nr = (int)div64_u64(to_reclaim, bytes);
4555 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4558 * shrink metadata reservation for delalloc
4560 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4563 struct btrfs_block_rsv *block_rsv;
4564 struct btrfs_space_info *space_info;
4565 struct btrfs_trans_handle *trans;
4569 unsigned long nr_pages;
4572 enum btrfs_reserve_flush_enum flush;
4574 /* Calc the number of the pages we need flush for space reservation */
4575 items = calc_reclaim_items_nr(root, to_reclaim);
4576 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4578 trans = (struct btrfs_trans_handle *)current->journal_info;
4579 block_rsv = &root->fs_info->delalloc_block_rsv;
4580 space_info = block_rsv->space_info;
4582 delalloc_bytes = percpu_counter_sum_positive(
4583 &root->fs_info->delalloc_bytes);
4584 if (delalloc_bytes == 0) {
4588 btrfs_wait_ordered_roots(root->fs_info, items);
4593 while (delalloc_bytes && loops < 3) {
4594 max_reclaim = min(delalloc_bytes, to_reclaim);
4595 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4596 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4598 * We need to wait for the async pages to actually start before
4601 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4605 if (max_reclaim <= nr_pages)
4608 max_reclaim -= nr_pages;
4610 wait_event(root->fs_info->async_submit_wait,
4611 atomic_read(&root->fs_info->async_delalloc_pages) <=
4615 flush = BTRFS_RESERVE_FLUSH_ALL;
4617 flush = BTRFS_RESERVE_NO_FLUSH;
4618 spin_lock(&space_info->lock);
4619 if (can_overcommit(root, space_info, orig, flush)) {
4620 spin_unlock(&space_info->lock);
4623 spin_unlock(&space_info->lock);
4626 if (wait_ordered && !trans) {
4627 btrfs_wait_ordered_roots(root->fs_info, items);
4629 time_left = schedule_timeout_killable(1);
4633 delalloc_bytes = percpu_counter_sum_positive(
4634 &root->fs_info->delalloc_bytes);
4639 * maybe_commit_transaction - possibly commit the transaction if its ok to
4640 * @root - the root we're allocating for
4641 * @bytes - the number of bytes we want to reserve
4642 * @force - force the commit
4644 * This will check to make sure that committing the transaction will actually
4645 * get us somewhere and then commit the transaction if it does. Otherwise it
4646 * will return -ENOSPC.
4648 static int may_commit_transaction(struct btrfs_root *root,
4649 struct btrfs_space_info *space_info,
4650 u64 bytes, int force)
4652 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4653 struct btrfs_trans_handle *trans;
4655 trans = (struct btrfs_trans_handle *)current->journal_info;
4662 /* See if there is enough pinned space to make this reservation */
4663 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4668 * See if there is some space in the delayed insertion reservation for
4671 if (space_info != delayed_rsv->space_info)
4674 spin_lock(&delayed_rsv->lock);
4675 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4676 bytes - delayed_rsv->size) >= 0) {
4677 spin_unlock(&delayed_rsv->lock);
4680 spin_unlock(&delayed_rsv->lock);
4683 trans = btrfs_join_transaction(root);
4687 return btrfs_commit_transaction(trans, root);
4691 FLUSH_DELAYED_ITEMS_NR = 1,
4692 FLUSH_DELAYED_ITEMS = 2,
4694 FLUSH_DELALLOC_WAIT = 4,
4699 static int flush_space(struct btrfs_root *root,
4700 struct btrfs_space_info *space_info, u64 num_bytes,
4701 u64 orig_bytes, int state)
4703 struct btrfs_trans_handle *trans;
4708 case FLUSH_DELAYED_ITEMS_NR:
4709 case FLUSH_DELAYED_ITEMS:
4710 if (state == FLUSH_DELAYED_ITEMS_NR)
4711 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4715 trans = btrfs_join_transaction(root);
4716 if (IS_ERR(trans)) {
4717 ret = PTR_ERR(trans);
4720 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4721 btrfs_end_transaction(trans, root);
4723 case FLUSH_DELALLOC:
4724 case FLUSH_DELALLOC_WAIT:
4725 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4726 state == FLUSH_DELALLOC_WAIT);
4729 trans = btrfs_join_transaction(root);
4730 if (IS_ERR(trans)) {
4731 ret = PTR_ERR(trans);
4734 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4735 btrfs_get_alloc_profile(root, 0),
4736 CHUNK_ALLOC_NO_FORCE);
4737 btrfs_end_transaction(trans, root);
4742 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4753 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4754 struct btrfs_space_info *space_info)
4760 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4762 spin_lock(&space_info->lock);
4763 if (can_overcommit(root, space_info, to_reclaim,
4764 BTRFS_RESERVE_FLUSH_ALL)) {
4769 used = space_info->bytes_used + space_info->bytes_reserved +
4770 space_info->bytes_pinned + space_info->bytes_readonly +
4771 space_info->bytes_may_use;
4772 if (can_overcommit(root, space_info, 1024 * 1024,
4773 BTRFS_RESERVE_FLUSH_ALL))
4774 expected = div_factor_fine(space_info->total_bytes, 95);
4776 expected = div_factor_fine(space_info->total_bytes, 90);
4778 if (used > expected)
4779 to_reclaim = used - expected;
4782 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4783 space_info->bytes_reserved);
4785 spin_unlock(&space_info->lock);
4790 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4791 struct btrfs_fs_info *fs_info, u64 used)
4793 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4795 /* If we're just plain full then async reclaim just slows us down. */
4796 if (space_info->bytes_used >= thresh)
4799 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4800 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4803 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4804 struct btrfs_fs_info *fs_info,
4809 spin_lock(&space_info->lock);
4811 * We run out of space and have not got any free space via flush_space,
4812 * so don't bother doing async reclaim.
4814 if (flush_state > COMMIT_TRANS && space_info->full) {
4815 spin_unlock(&space_info->lock);
4819 used = space_info->bytes_used + space_info->bytes_reserved +
4820 space_info->bytes_pinned + space_info->bytes_readonly +
4821 space_info->bytes_may_use;
4822 if (need_do_async_reclaim(space_info, fs_info, used)) {
4823 spin_unlock(&space_info->lock);
4826 spin_unlock(&space_info->lock);
4831 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4833 struct btrfs_fs_info *fs_info;
4834 struct btrfs_space_info *space_info;
4838 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4839 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4841 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4846 flush_state = FLUSH_DELAYED_ITEMS_NR;
4848 flush_space(fs_info->fs_root, space_info, to_reclaim,
4849 to_reclaim, flush_state);
4851 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4854 } while (flush_state < COMMIT_TRANS);
4857 void btrfs_init_async_reclaim_work(struct work_struct *work)
4859 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4863 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4864 * @root - the root we're allocating for
4865 * @block_rsv - the block_rsv we're allocating for
4866 * @orig_bytes - the number of bytes we want
4867 * @flush - whether or not we can flush to make our reservation
4869 * This will reserve orgi_bytes number of bytes from the space info associated
4870 * with the block_rsv. If there is not enough space it will make an attempt to
4871 * flush out space to make room. It will do this by flushing delalloc if
4872 * possible or committing the transaction. If flush is 0 then no attempts to
4873 * regain reservations will be made and this will fail if there is not enough
4876 static int reserve_metadata_bytes(struct btrfs_root *root,
4877 struct btrfs_block_rsv *block_rsv,
4879 enum btrfs_reserve_flush_enum flush)
4881 struct btrfs_space_info *space_info = block_rsv->space_info;
4883 u64 num_bytes = orig_bytes;
4884 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4886 bool flushing = false;
4890 spin_lock(&space_info->lock);
4892 * We only want to wait if somebody other than us is flushing and we
4893 * are actually allowed to flush all things.
4895 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4896 space_info->flush) {
4897 spin_unlock(&space_info->lock);
4899 * If we have a trans handle we can't wait because the flusher
4900 * may have to commit the transaction, which would mean we would
4901 * deadlock since we are waiting for the flusher to finish, but
4902 * hold the current transaction open.
4904 if (current->journal_info)
4906 ret = wait_event_killable(space_info->wait, !space_info->flush);
4907 /* Must have been killed, return */
4911 spin_lock(&space_info->lock);
4915 used = space_info->bytes_used + space_info->bytes_reserved +
4916 space_info->bytes_pinned + space_info->bytes_readonly +
4917 space_info->bytes_may_use;
4920 * The idea here is that we've not already over-reserved the block group
4921 * then we can go ahead and save our reservation first and then start
4922 * flushing if we need to. Otherwise if we've already overcommitted
4923 * lets start flushing stuff first and then come back and try to make
4926 if (used <= space_info->total_bytes) {
4927 if (used + orig_bytes <= space_info->total_bytes) {
4928 space_info->bytes_may_use += orig_bytes;
4929 trace_btrfs_space_reservation(root->fs_info,
4930 "space_info", space_info->flags, orig_bytes, 1);
4934 * Ok set num_bytes to orig_bytes since we aren't
4935 * overocmmitted, this way we only try and reclaim what
4938 num_bytes = orig_bytes;
4942 * Ok we're over committed, set num_bytes to the overcommitted
4943 * amount plus the amount of bytes that we need for this
4946 num_bytes = used - space_info->total_bytes +
4950 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4951 space_info->bytes_may_use += orig_bytes;
4952 trace_btrfs_space_reservation(root->fs_info, "space_info",
4953 space_info->flags, orig_bytes,
4959 * Couldn't make our reservation, save our place so while we're trying
4960 * to reclaim space we can actually use it instead of somebody else
4961 * stealing it from us.
4963 * We make the other tasks wait for the flush only when we can flush
4966 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4968 space_info->flush = 1;
4969 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4972 * We will do the space reservation dance during log replay,
4973 * which means we won't have fs_info->fs_root set, so don't do
4974 * the async reclaim as we will panic.
4976 if (!root->fs_info->log_root_recovering &&
4977 need_do_async_reclaim(space_info, root->fs_info, used) &&
4978 !work_busy(&root->fs_info->async_reclaim_work))
4979 queue_work(system_unbound_wq,
4980 &root->fs_info->async_reclaim_work);
4982 spin_unlock(&space_info->lock);
4984 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4987 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4992 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4993 * would happen. So skip delalloc flush.
4995 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4996 (flush_state == FLUSH_DELALLOC ||
4997 flush_state == FLUSH_DELALLOC_WAIT))
4998 flush_state = ALLOC_CHUNK;
5002 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
5003 flush_state < COMMIT_TRANS)
5005 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
5006 flush_state <= COMMIT_TRANS)
5010 if (ret == -ENOSPC &&
5011 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
5012 struct btrfs_block_rsv *global_rsv =
5013 &root->fs_info->global_block_rsv;
5015 if (block_rsv != global_rsv &&
5016 !block_rsv_use_bytes(global_rsv, orig_bytes))
5020 trace_btrfs_space_reservation(root->fs_info,
5021 "space_info:enospc",
5022 space_info->flags, orig_bytes, 1);
5024 spin_lock(&space_info->lock);
5025 space_info->flush = 0;
5026 wake_up_all(&space_info->wait);
5027 spin_unlock(&space_info->lock);
5032 static struct btrfs_block_rsv *get_block_rsv(
5033 const struct btrfs_trans_handle *trans,
5034 const struct btrfs_root *root)
5036 struct btrfs_block_rsv *block_rsv = NULL;
5038 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
5039 (root == root->fs_info->csum_root && trans->adding_csums) ||
5040 (root == root->fs_info->uuid_root))
5041 block_rsv = trans->block_rsv;
5044 block_rsv = root->block_rsv;
5047 block_rsv = &root->fs_info->empty_block_rsv;
5052 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
5056 spin_lock(&block_rsv->lock);
5057 if (block_rsv->reserved >= num_bytes) {
5058 block_rsv->reserved -= num_bytes;
5059 if (block_rsv->reserved < block_rsv->size)
5060 block_rsv->full = 0;
5063 spin_unlock(&block_rsv->lock);
5067 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
5068 u64 num_bytes, int update_size)
5070 spin_lock(&block_rsv->lock);
5071 block_rsv->reserved += num_bytes;
5073 block_rsv->size += num_bytes;
5074 else if (block_rsv->reserved >= block_rsv->size)
5075 block_rsv->full = 1;
5076 spin_unlock(&block_rsv->lock);
5079 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
5080 struct btrfs_block_rsv *dest, u64 num_bytes,
5083 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5086 if (global_rsv->space_info != dest->space_info)
5089 spin_lock(&global_rsv->lock);
5090 min_bytes = div_factor(global_rsv->size, min_factor);
5091 if (global_rsv->reserved < min_bytes + num_bytes) {
5092 spin_unlock(&global_rsv->lock);
5095 global_rsv->reserved -= num_bytes;
5096 if (global_rsv->reserved < global_rsv->size)
5097 global_rsv->full = 0;
5098 spin_unlock(&global_rsv->lock);
5100 block_rsv_add_bytes(dest, num_bytes, 1);
5104 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5105 struct btrfs_block_rsv *block_rsv,
5106 struct btrfs_block_rsv *dest, u64 num_bytes)
5108 struct btrfs_space_info *space_info = block_rsv->space_info;
5110 spin_lock(&block_rsv->lock);
5111 if (num_bytes == (u64)-1)
5112 num_bytes = block_rsv->size;
5113 block_rsv->size -= num_bytes;
5114 if (block_rsv->reserved >= block_rsv->size) {
5115 num_bytes = block_rsv->reserved - block_rsv->size;
5116 block_rsv->reserved = block_rsv->size;
5117 block_rsv->full = 1;
5121 spin_unlock(&block_rsv->lock);
5123 if (num_bytes > 0) {
5125 spin_lock(&dest->lock);
5129 bytes_to_add = dest->size - dest->reserved;
5130 bytes_to_add = min(num_bytes, bytes_to_add);
5131 dest->reserved += bytes_to_add;
5132 if (dest->reserved >= dest->size)
5134 num_bytes -= bytes_to_add;
5136 spin_unlock(&dest->lock);
5139 spin_lock(&space_info->lock);
5140 space_info->bytes_may_use -= num_bytes;
5141 trace_btrfs_space_reservation(fs_info, "space_info",
5142 space_info->flags, num_bytes, 0);
5143 spin_unlock(&space_info->lock);
5148 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5149 struct btrfs_block_rsv *dst, u64 num_bytes)
5153 ret = block_rsv_use_bytes(src, num_bytes);
5157 block_rsv_add_bytes(dst, num_bytes, 1);
5161 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5163 memset(rsv, 0, sizeof(*rsv));
5164 spin_lock_init(&rsv->lock);
5168 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5169 unsigned short type)
5171 struct btrfs_block_rsv *block_rsv;
5172 struct btrfs_fs_info *fs_info = root->fs_info;
5174 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5178 btrfs_init_block_rsv(block_rsv, type);
5179 block_rsv->space_info = __find_space_info(fs_info,
5180 BTRFS_BLOCK_GROUP_METADATA);
5184 void btrfs_free_block_rsv(struct btrfs_root *root,
5185 struct btrfs_block_rsv *rsv)
5189 btrfs_block_rsv_release(root, rsv, (u64)-1);
5193 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5198 int btrfs_block_rsv_add(struct btrfs_root *root,
5199 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5200 enum btrfs_reserve_flush_enum flush)
5207 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5209 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5216 int btrfs_block_rsv_check(struct btrfs_root *root,
5217 struct btrfs_block_rsv *block_rsv, int min_factor)
5225 spin_lock(&block_rsv->lock);
5226 num_bytes = div_factor(block_rsv->size, min_factor);
5227 if (block_rsv->reserved >= num_bytes)
5229 spin_unlock(&block_rsv->lock);
5234 int btrfs_block_rsv_refill(struct btrfs_root *root,
5235 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5236 enum btrfs_reserve_flush_enum flush)
5244 spin_lock(&block_rsv->lock);
5245 num_bytes = min_reserved;
5246 if (block_rsv->reserved >= num_bytes)
5249 num_bytes -= block_rsv->reserved;
5250 spin_unlock(&block_rsv->lock);
5255 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5257 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5264 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5265 struct btrfs_block_rsv *dst_rsv,
5268 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5271 void btrfs_block_rsv_release(struct btrfs_root *root,
5272 struct btrfs_block_rsv *block_rsv,
5275 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5276 if (global_rsv == block_rsv ||
5277 block_rsv->space_info != global_rsv->space_info)
5279 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5284 * helper to calculate size of global block reservation.
5285 * the desired value is sum of space used by extent tree,
5286 * checksum tree and root tree
5288 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5290 struct btrfs_space_info *sinfo;
5294 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5296 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5297 spin_lock(&sinfo->lock);
5298 data_used = sinfo->bytes_used;
5299 spin_unlock(&sinfo->lock);
5301 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5302 spin_lock(&sinfo->lock);
5303 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5305 meta_used = sinfo->bytes_used;
5306 spin_unlock(&sinfo->lock);
5308 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5310 num_bytes += div_u64(data_used + meta_used, 50);
5312 if (num_bytes * 3 > meta_used)
5313 num_bytes = div_u64(meta_used, 3);
5315 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5318 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5320 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5321 struct btrfs_space_info *sinfo = block_rsv->space_info;
5324 num_bytes = calc_global_metadata_size(fs_info);
5326 spin_lock(&sinfo->lock);
5327 spin_lock(&block_rsv->lock);
5329 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5331 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5332 sinfo->bytes_reserved + sinfo->bytes_readonly +
5333 sinfo->bytes_may_use;
5335 if (sinfo->total_bytes > num_bytes) {
5336 num_bytes = sinfo->total_bytes - num_bytes;
5337 block_rsv->reserved += num_bytes;
5338 sinfo->bytes_may_use += num_bytes;
5339 trace_btrfs_space_reservation(fs_info, "space_info",
5340 sinfo->flags, num_bytes, 1);
5343 if (block_rsv->reserved >= block_rsv->size) {
5344 num_bytes = block_rsv->reserved - block_rsv->size;
5345 sinfo->bytes_may_use -= num_bytes;
5346 trace_btrfs_space_reservation(fs_info, "space_info",
5347 sinfo->flags, num_bytes, 0);
5348 block_rsv->reserved = block_rsv->size;
5349 block_rsv->full = 1;
5352 spin_unlock(&block_rsv->lock);
5353 spin_unlock(&sinfo->lock);
5356 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5358 struct btrfs_space_info *space_info;
5360 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5361 fs_info->chunk_block_rsv.space_info = space_info;
5363 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5364 fs_info->global_block_rsv.space_info = space_info;
5365 fs_info->delalloc_block_rsv.space_info = space_info;
5366 fs_info->trans_block_rsv.space_info = space_info;
5367 fs_info->empty_block_rsv.space_info = space_info;
5368 fs_info->delayed_block_rsv.space_info = space_info;
5370 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5371 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5372 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5373 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5374 if (fs_info->quota_root)
5375 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5376 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5378 update_global_block_rsv(fs_info);
5381 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5383 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5385 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5386 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5387 WARN_ON(fs_info->trans_block_rsv.size > 0);
5388 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5389 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5390 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5391 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5392 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5395 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5396 struct btrfs_root *root)
5398 if (!trans->block_rsv)
5401 if (!trans->bytes_reserved)
5404 trace_btrfs_space_reservation(root->fs_info, "transaction",
5405 trans->transid, trans->bytes_reserved, 0);
5406 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5407 trans->bytes_reserved = 0;
5411 * To be called after all the new block groups attached to the transaction
5412 * handle have been created (btrfs_create_pending_block_groups()).
5414 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5416 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5418 if (!trans->chunk_bytes_reserved)
5421 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5423 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5424 trans->chunk_bytes_reserved);
5425 trans->chunk_bytes_reserved = 0;
5428 /* Can only return 0 or -ENOSPC */
5429 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5430 struct inode *inode)
5432 struct btrfs_root *root = BTRFS_I(inode)->root;
5433 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5434 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5437 * We need to hold space in order to delete our orphan item once we've
5438 * added it, so this takes the reservation so we can release it later
5439 * when we are truly done with the orphan item.
5441 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5442 trace_btrfs_space_reservation(root->fs_info, "orphan",
5443 btrfs_ino(inode), num_bytes, 1);
5444 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5447 void btrfs_orphan_release_metadata(struct inode *inode)
5449 struct btrfs_root *root = BTRFS_I(inode)->root;
5450 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5451 trace_btrfs_space_reservation(root->fs_info, "orphan",
5452 btrfs_ino(inode), num_bytes, 0);
5453 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5457 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5458 * root: the root of the parent directory
5459 * rsv: block reservation
5460 * items: the number of items that we need do reservation
5461 * qgroup_reserved: used to return the reserved size in qgroup
5463 * This function is used to reserve the space for snapshot/subvolume
5464 * creation and deletion. Those operations are different with the
5465 * common file/directory operations, they change two fs/file trees
5466 * and root tree, the number of items that the qgroup reserves is
5467 * different with the free space reservation. So we can not use
5468 * the space reseravtion mechanism in start_transaction().
5470 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5471 struct btrfs_block_rsv *rsv,
5473 u64 *qgroup_reserved,
5474 bool use_global_rsv)
5478 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5480 if (root->fs_info->quota_enabled) {
5481 /* One for parent inode, two for dir entries */
5482 num_bytes = 3 * root->nodesize;
5483 ret = btrfs_qgroup_reserve_meta(root, num_bytes);
5490 *qgroup_reserved = num_bytes;
5492 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5493 rsv->space_info = __find_space_info(root->fs_info,
5494 BTRFS_BLOCK_GROUP_METADATA);
5495 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5496 BTRFS_RESERVE_FLUSH_ALL);
5498 if (ret == -ENOSPC && use_global_rsv)
5499 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5501 if (ret && *qgroup_reserved)
5502 btrfs_qgroup_free_meta(root, *qgroup_reserved);
5507 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5508 struct btrfs_block_rsv *rsv,
5509 u64 qgroup_reserved)
5511 btrfs_block_rsv_release(root, rsv, (u64)-1);
5515 * drop_outstanding_extent - drop an outstanding extent
5516 * @inode: the inode we're dropping the extent for
5517 * @num_bytes: the number of bytes we're relaseing.
5519 * This is called when we are freeing up an outstanding extent, either called
5520 * after an error or after an extent is written. This will return the number of
5521 * reserved extents that need to be freed. This must be called with
5522 * BTRFS_I(inode)->lock held.
5524 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5526 unsigned drop_inode_space = 0;
5527 unsigned dropped_extents = 0;
5528 unsigned num_extents = 0;
5530 num_extents = (unsigned)div64_u64(num_bytes +
5531 BTRFS_MAX_EXTENT_SIZE - 1,
5532 BTRFS_MAX_EXTENT_SIZE);
5533 ASSERT(num_extents);
5534 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5535 BTRFS_I(inode)->outstanding_extents -= num_extents;
5537 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5538 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5539 &BTRFS_I(inode)->runtime_flags))
5540 drop_inode_space = 1;
5543 * If we have more or the same amount of outsanding extents than we have
5544 * reserved then we need to leave the reserved extents count alone.
5546 if (BTRFS_I(inode)->outstanding_extents >=
5547 BTRFS_I(inode)->reserved_extents)
5548 return drop_inode_space;
5550 dropped_extents = BTRFS_I(inode)->reserved_extents -
5551 BTRFS_I(inode)->outstanding_extents;
5552 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5553 return dropped_extents + drop_inode_space;
5557 * calc_csum_metadata_size - return the amount of metada space that must be
5558 * reserved/free'd for the given bytes.
5559 * @inode: the inode we're manipulating
5560 * @num_bytes: the number of bytes in question
5561 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5563 * This adjusts the number of csum_bytes in the inode and then returns the
5564 * correct amount of metadata that must either be reserved or freed. We
5565 * calculate how many checksums we can fit into one leaf and then divide the
5566 * number of bytes that will need to be checksumed by this value to figure out
5567 * how many checksums will be required. If we are adding bytes then the number
5568 * may go up and we will return the number of additional bytes that must be
5569 * reserved. If it is going down we will return the number of bytes that must
5572 * This must be called with BTRFS_I(inode)->lock held.
5574 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5577 struct btrfs_root *root = BTRFS_I(inode)->root;
5578 u64 old_csums, num_csums;
5580 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5581 BTRFS_I(inode)->csum_bytes == 0)
5584 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5586 BTRFS_I(inode)->csum_bytes += num_bytes;
5588 BTRFS_I(inode)->csum_bytes -= num_bytes;
5589 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5591 /* No change, no need to reserve more */
5592 if (old_csums == num_csums)
5596 return btrfs_calc_trans_metadata_size(root,
5597 num_csums - old_csums);
5599 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5602 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5604 struct btrfs_root *root = BTRFS_I(inode)->root;
5605 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5608 unsigned nr_extents = 0;
5609 int extra_reserve = 0;
5610 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5612 bool delalloc_lock = true;
5616 /* If we are a free space inode we need to not flush since we will be in
5617 * the middle of a transaction commit. We also don't need the delalloc
5618 * mutex since we won't race with anybody. We need this mostly to make
5619 * lockdep shut its filthy mouth.
5621 if (btrfs_is_free_space_inode(inode)) {
5622 flush = BTRFS_RESERVE_NO_FLUSH;
5623 delalloc_lock = false;
5626 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5627 btrfs_transaction_in_commit(root->fs_info))
5628 schedule_timeout(1);
5631 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5633 num_bytes = ALIGN(num_bytes, root->sectorsize);
5635 spin_lock(&BTRFS_I(inode)->lock);
5636 nr_extents = (unsigned)div64_u64(num_bytes +
5637 BTRFS_MAX_EXTENT_SIZE - 1,
5638 BTRFS_MAX_EXTENT_SIZE);
5639 BTRFS_I(inode)->outstanding_extents += nr_extents;
5642 if (BTRFS_I(inode)->outstanding_extents >
5643 BTRFS_I(inode)->reserved_extents)
5644 nr_extents = BTRFS_I(inode)->outstanding_extents -
5645 BTRFS_I(inode)->reserved_extents;
5648 * Add an item to reserve for updating the inode when we complete the
5651 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5652 &BTRFS_I(inode)->runtime_flags)) {
5657 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5658 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5659 csum_bytes = BTRFS_I(inode)->csum_bytes;
5660 spin_unlock(&BTRFS_I(inode)->lock);
5662 if (root->fs_info->quota_enabled) {
5663 ret = btrfs_qgroup_reserve_meta(root,
5664 nr_extents * root->nodesize);
5669 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5670 if (unlikely(ret)) {
5671 btrfs_qgroup_free_meta(root, nr_extents * root->nodesize);
5675 spin_lock(&BTRFS_I(inode)->lock);
5676 if (extra_reserve) {
5677 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5678 &BTRFS_I(inode)->runtime_flags);
5681 BTRFS_I(inode)->reserved_extents += nr_extents;
5682 spin_unlock(&BTRFS_I(inode)->lock);
5685 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5688 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5689 btrfs_ino(inode), to_reserve, 1);
5690 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5695 spin_lock(&BTRFS_I(inode)->lock);
5696 dropped = drop_outstanding_extent(inode, num_bytes);
5698 * If the inodes csum_bytes is the same as the original
5699 * csum_bytes then we know we haven't raced with any free()ers
5700 * so we can just reduce our inodes csum bytes and carry on.
5702 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5703 calc_csum_metadata_size(inode, num_bytes, 0);
5705 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5709 * This is tricky, but first we need to figure out how much we
5710 * free'd from any free-ers that occured during this
5711 * reservation, so we reset ->csum_bytes to the csum_bytes
5712 * before we dropped our lock, and then call the free for the
5713 * number of bytes that were freed while we were trying our
5716 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5717 BTRFS_I(inode)->csum_bytes = csum_bytes;
5718 to_free = calc_csum_metadata_size(inode, bytes, 0);
5722 * Now we need to see how much we would have freed had we not
5723 * been making this reservation and our ->csum_bytes were not
5724 * artificially inflated.
5726 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5727 bytes = csum_bytes - orig_csum_bytes;
5728 bytes = calc_csum_metadata_size(inode, bytes, 0);
5731 * Now reset ->csum_bytes to what it should be. If bytes is
5732 * more than to_free then we would have free'd more space had we
5733 * not had an artificially high ->csum_bytes, so we need to free
5734 * the remainder. If bytes is the same or less then we don't
5735 * need to do anything, the other free-ers did the correct
5738 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5739 if (bytes > to_free)
5740 to_free = bytes - to_free;
5744 spin_unlock(&BTRFS_I(inode)->lock);
5746 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5749 btrfs_block_rsv_release(root, block_rsv, to_free);
5750 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5751 btrfs_ino(inode), to_free, 0);
5754 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5759 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5760 * @inode: the inode to release the reservation for
5761 * @num_bytes: the number of bytes we're releasing
5763 * This will release the metadata reservation for an inode. This can be called
5764 * once we complete IO for a given set of bytes to release their metadata
5767 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5769 struct btrfs_root *root = BTRFS_I(inode)->root;
5773 num_bytes = ALIGN(num_bytes, root->sectorsize);
5774 spin_lock(&BTRFS_I(inode)->lock);
5775 dropped = drop_outstanding_extent(inode, num_bytes);
5778 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5779 spin_unlock(&BTRFS_I(inode)->lock);
5781 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5783 if (btrfs_test_is_dummy_root(root))
5786 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5787 btrfs_ino(inode), to_free, 0);
5789 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5794 * btrfs_delalloc_reserve_space - reserve data and metadata space for
5796 * @inode: inode we're writing to
5797 * @start: start range we are writing to
5798 * @len: how long the range we are writing to
5800 * TODO: This function will finally replace old btrfs_delalloc_reserve_space()
5802 * This will do the following things
5804 * o reserve space in data space info for num bytes
5805 * and reserve precious corresponding qgroup space
5806 * (Done in check_data_free_space)
5808 * o reserve space for metadata space, based on the number of outstanding
5809 * extents and how much csums will be needed
5810 * also reserve metadata space in a per root over-reserve method.
5811 * o add to the inodes->delalloc_bytes
5812 * o add it to the fs_info's delalloc inodes list.
5813 * (Above 3 all done in delalloc_reserve_metadata)
5815 * Return 0 for success
5816 * Return <0 for error(-ENOSPC or -EQUOT)
5818 int btrfs_delalloc_reserve_space(struct inode *inode, u64 start, u64 len)
5822 ret = btrfs_check_data_free_space(inode, start, len);
5825 ret = btrfs_delalloc_reserve_metadata(inode, len);
5827 btrfs_free_reserved_data_space(inode, start, len);
5832 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5833 * @inode: inode we're releasing space for
5834 * @start: start position of the space already reserved
5835 * @len: the len of the space already reserved
5837 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5838 * called in the case that we don't need the metadata AND data reservations
5839 * anymore. So if there is an error or we insert an inline extent.
5841 * This function will release the metadata space that was not used and will
5842 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5843 * list if there are no delalloc bytes left.
5844 * Also it will handle the qgroup reserved space.
5846 void btrfs_delalloc_release_space(struct inode *inode, u64 start, u64 len)
5848 btrfs_delalloc_release_metadata(inode, len);
5849 btrfs_free_reserved_data_space(inode, start, len);
5852 static int update_block_group(struct btrfs_trans_handle *trans,
5853 struct btrfs_root *root, u64 bytenr,
5854 u64 num_bytes, int alloc)
5856 struct btrfs_block_group_cache *cache = NULL;
5857 struct btrfs_fs_info *info = root->fs_info;
5858 u64 total = num_bytes;
5863 /* block accounting for super block */
5864 spin_lock(&info->delalloc_root_lock);
5865 old_val = btrfs_super_bytes_used(info->super_copy);
5867 old_val += num_bytes;
5869 old_val -= num_bytes;
5870 btrfs_set_super_bytes_used(info->super_copy, old_val);
5871 spin_unlock(&info->delalloc_root_lock);
5874 cache = btrfs_lookup_block_group(info, bytenr);
5877 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5878 BTRFS_BLOCK_GROUP_RAID1 |
5879 BTRFS_BLOCK_GROUP_RAID10))
5884 * If this block group has free space cache written out, we
5885 * need to make sure to load it if we are removing space. This
5886 * is because we need the unpinning stage to actually add the
5887 * space back to the block group, otherwise we will leak space.
5889 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5890 cache_block_group(cache, 1);
5892 byte_in_group = bytenr - cache->key.objectid;
5893 WARN_ON(byte_in_group > cache->key.offset);
5895 spin_lock(&cache->space_info->lock);
5896 spin_lock(&cache->lock);
5898 if (btrfs_test_opt(root, SPACE_CACHE) &&
5899 cache->disk_cache_state < BTRFS_DC_CLEAR)
5900 cache->disk_cache_state = BTRFS_DC_CLEAR;
5902 old_val = btrfs_block_group_used(&cache->item);
5903 num_bytes = min(total, cache->key.offset - byte_in_group);
5905 old_val += num_bytes;
5906 btrfs_set_block_group_used(&cache->item, old_val);
5907 cache->reserved -= num_bytes;
5908 cache->space_info->bytes_reserved -= num_bytes;
5909 cache->space_info->bytes_used += num_bytes;
5910 cache->space_info->disk_used += num_bytes * factor;
5911 spin_unlock(&cache->lock);
5912 spin_unlock(&cache->space_info->lock);
5914 old_val -= num_bytes;
5915 btrfs_set_block_group_used(&cache->item, old_val);
5916 cache->pinned += num_bytes;
5917 cache->space_info->bytes_pinned += num_bytes;
5918 cache->space_info->bytes_used -= num_bytes;
5919 cache->space_info->disk_used -= num_bytes * factor;
5920 spin_unlock(&cache->lock);
5921 spin_unlock(&cache->space_info->lock);
5923 set_extent_dirty(info->pinned_extents,
5924 bytenr, bytenr + num_bytes - 1,
5925 GFP_NOFS | __GFP_NOFAIL);
5927 * No longer have used bytes in this block group, queue
5931 spin_lock(&info->unused_bgs_lock);
5932 if (list_empty(&cache->bg_list)) {
5933 btrfs_get_block_group(cache);
5934 list_add_tail(&cache->bg_list,
5937 spin_unlock(&info->unused_bgs_lock);
5941 spin_lock(&trans->transaction->dirty_bgs_lock);
5942 if (list_empty(&cache->dirty_list)) {
5943 list_add_tail(&cache->dirty_list,
5944 &trans->transaction->dirty_bgs);
5945 trans->transaction->num_dirty_bgs++;
5946 btrfs_get_block_group(cache);
5948 spin_unlock(&trans->transaction->dirty_bgs_lock);
5950 btrfs_put_block_group(cache);
5952 bytenr += num_bytes;
5957 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5959 struct btrfs_block_group_cache *cache;
5962 spin_lock(&root->fs_info->block_group_cache_lock);
5963 bytenr = root->fs_info->first_logical_byte;
5964 spin_unlock(&root->fs_info->block_group_cache_lock);
5966 if (bytenr < (u64)-1)
5969 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5973 bytenr = cache->key.objectid;
5974 btrfs_put_block_group(cache);
5979 static int pin_down_extent(struct btrfs_root *root,
5980 struct btrfs_block_group_cache *cache,
5981 u64 bytenr, u64 num_bytes, int reserved)
5983 spin_lock(&cache->space_info->lock);
5984 spin_lock(&cache->lock);
5985 cache->pinned += num_bytes;
5986 cache->space_info->bytes_pinned += num_bytes;
5988 cache->reserved -= num_bytes;
5989 cache->space_info->bytes_reserved -= num_bytes;
5991 spin_unlock(&cache->lock);
5992 spin_unlock(&cache->space_info->lock);
5994 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5995 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5997 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
6002 * this function must be called within transaction
6004 int btrfs_pin_extent(struct btrfs_root *root,
6005 u64 bytenr, u64 num_bytes, int reserved)
6007 struct btrfs_block_group_cache *cache;
6009 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6010 BUG_ON(!cache); /* Logic error */
6012 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
6014 btrfs_put_block_group(cache);
6019 * this function must be called within transaction
6021 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
6022 u64 bytenr, u64 num_bytes)
6024 struct btrfs_block_group_cache *cache;
6027 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
6032 * pull in the free space cache (if any) so that our pin
6033 * removes the free space from the cache. We have load_only set
6034 * to one because the slow code to read in the free extents does check
6035 * the pinned extents.
6037 cache_block_group(cache, 1);
6039 pin_down_extent(root, cache, bytenr, num_bytes, 0);
6041 /* remove us from the free space cache (if we're there at all) */
6042 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
6043 btrfs_put_block_group(cache);
6047 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
6050 struct btrfs_block_group_cache *block_group;
6051 struct btrfs_caching_control *caching_ctl;
6053 block_group = btrfs_lookup_block_group(root->fs_info, start);
6057 cache_block_group(block_group, 0);
6058 caching_ctl = get_caching_control(block_group);
6062 BUG_ON(!block_group_cache_done(block_group));
6063 ret = btrfs_remove_free_space(block_group, start, num_bytes);
6065 mutex_lock(&caching_ctl->mutex);
6067 if (start >= caching_ctl->progress) {
6068 ret = add_excluded_extent(root, start, num_bytes);
6069 } else if (start + num_bytes <= caching_ctl->progress) {
6070 ret = btrfs_remove_free_space(block_group,
6073 num_bytes = caching_ctl->progress - start;
6074 ret = btrfs_remove_free_space(block_group,
6079 num_bytes = (start + num_bytes) -
6080 caching_ctl->progress;
6081 start = caching_ctl->progress;
6082 ret = add_excluded_extent(root, start, num_bytes);
6085 mutex_unlock(&caching_ctl->mutex);
6086 put_caching_control(caching_ctl);
6088 btrfs_put_block_group(block_group);
6092 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6093 struct extent_buffer *eb)
6095 struct btrfs_file_extent_item *item;
6096 struct btrfs_key key;
6100 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6103 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6104 btrfs_item_key_to_cpu(eb, &key, i);
6105 if (key.type != BTRFS_EXTENT_DATA_KEY)
6107 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6108 found_type = btrfs_file_extent_type(eb, item);
6109 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6111 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6113 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6114 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6115 __exclude_logged_extent(log, key.objectid, key.offset);
6122 * btrfs_update_reserved_bytes - update the block_group and space info counters
6123 * @cache: The cache we are manipulating
6124 * @num_bytes: The number of bytes in question
6125 * @reserve: One of the reservation enums
6126 * @delalloc: The blocks are allocated for the delalloc write
6128 * This is called by the allocator when it reserves space, or by somebody who is
6129 * freeing space that was never actually used on disk. For example if you
6130 * reserve some space for a new leaf in transaction A and before transaction A
6131 * commits you free that leaf, you call this with reserve set to 0 in order to
6132 * clear the reservation.
6134 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6135 * ENOSPC accounting. For data we handle the reservation through clearing the
6136 * delalloc bits in the io_tree. We have to do this since we could end up
6137 * allocating less disk space for the amount of data we have reserved in the
6138 * case of compression.
6140 * If this is a reservation and the block group has become read only we cannot
6141 * make the reservation and return -EAGAIN, otherwise this function always
6144 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6145 u64 num_bytes, int reserve, int delalloc)
6147 struct btrfs_space_info *space_info = cache->space_info;
6150 spin_lock(&space_info->lock);
6151 spin_lock(&cache->lock);
6152 if (reserve != RESERVE_FREE) {
6156 cache->reserved += num_bytes;
6157 space_info->bytes_reserved += num_bytes;
6158 if (reserve == RESERVE_ALLOC) {
6159 trace_btrfs_space_reservation(cache->fs_info,
6160 "space_info", space_info->flags,
6162 space_info->bytes_may_use -= num_bytes;
6166 cache->delalloc_bytes += num_bytes;
6170 space_info->bytes_readonly += num_bytes;
6171 cache->reserved -= num_bytes;
6172 space_info->bytes_reserved -= num_bytes;
6175 cache->delalloc_bytes -= num_bytes;
6177 spin_unlock(&cache->lock);
6178 spin_unlock(&space_info->lock);
6182 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6183 struct btrfs_root *root)
6185 struct btrfs_fs_info *fs_info = root->fs_info;
6186 struct btrfs_caching_control *next;
6187 struct btrfs_caching_control *caching_ctl;
6188 struct btrfs_block_group_cache *cache;
6190 down_write(&fs_info->commit_root_sem);
6192 list_for_each_entry_safe(caching_ctl, next,
6193 &fs_info->caching_block_groups, list) {
6194 cache = caching_ctl->block_group;
6195 if (block_group_cache_done(cache)) {
6196 cache->last_byte_to_unpin = (u64)-1;
6197 list_del_init(&caching_ctl->list);
6198 put_caching_control(caching_ctl);
6200 cache->last_byte_to_unpin = caching_ctl->progress;
6204 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6205 fs_info->pinned_extents = &fs_info->freed_extents[1];
6207 fs_info->pinned_extents = &fs_info->freed_extents[0];
6209 up_write(&fs_info->commit_root_sem);
6211 update_global_block_rsv(fs_info);
6215 * Returns the free cluster for the given space info and sets empty_cluster to
6216 * what it should be based on the mount options.
6218 static struct btrfs_free_cluster *
6219 fetch_cluster_info(struct btrfs_root *root, struct btrfs_space_info *space_info,
6222 struct btrfs_free_cluster *ret = NULL;
6223 bool ssd = btrfs_test_opt(root, SSD);
6226 if (btrfs_mixed_space_info(space_info))
6230 *empty_cluster = 2 * 1024 * 1024;
6231 if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
6232 ret = &root->fs_info->meta_alloc_cluster;
6234 *empty_cluster = 64 * 1024;
6235 } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && ssd) {
6236 ret = &root->fs_info->data_alloc_cluster;
6242 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6243 const bool return_free_space)
6245 struct btrfs_fs_info *fs_info = root->fs_info;
6246 struct btrfs_block_group_cache *cache = NULL;
6247 struct btrfs_space_info *space_info;
6248 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6249 struct btrfs_free_cluster *cluster = NULL;
6251 u64 total_unpinned = 0;
6252 u64 empty_cluster = 0;
6255 while (start <= end) {
6258 start >= cache->key.objectid + cache->key.offset) {
6260 btrfs_put_block_group(cache);
6262 cache = btrfs_lookup_block_group(fs_info, start);
6263 BUG_ON(!cache); /* Logic error */
6265 cluster = fetch_cluster_info(root,
6268 empty_cluster <<= 1;
6271 len = cache->key.objectid + cache->key.offset - start;
6272 len = min(len, end + 1 - start);
6274 if (start < cache->last_byte_to_unpin) {
6275 len = min(len, cache->last_byte_to_unpin - start);
6276 if (return_free_space)
6277 btrfs_add_free_space(cache, start, len);
6281 total_unpinned += len;
6282 space_info = cache->space_info;
6285 * If this space cluster has been marked as fragmented and we've
6286 * unpinned enough in this block group to potentially allow a
6287 * cluster to be created inside of it go ahead and clear the
6290 if (cluster && cluster->fragmented &&
6291 total_unpinned > empty_cluster) {
6292 spin_lock(&cluster->lock);
6293 cluster->fragmented = 0;
6294 spin_unlock(&cluster->lock);
6297 spin_lock(&space_info->lock);
6298 spin_lock(&cache->lock);
6299 cache->pinned -= len;
6300 space_info->bytes_pinned -= len;
6301 space_info->max_extent_size = 0;
6302 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6304 space_info->bytes_readonly += len;
6307 spin_unlock(&cache->lock);
6308 if (!readonly && global_rsv->space_info == space_info) {
6309 spin_lock(&global_rsv->lock);
6310 if (!global_rsv->full) {
6311 len = min(len, global_rsv->size -
6312 global_rsv->reserved);
6313 global_rsv->reserved += len;
6314 space_info->bytes_may_use += len;
6315 if (global_rsv->reserved >= global_rsv->size)
6316 global_rsv->full = 1;
6318 spin_unlock(&global_rsv->lock);
6320 spin_unlock(&space_info->lock);
6324 btrfs_put_block_group(cache);
6328 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6329 struct btrfs_root *root)
6331 struct btrfs_fs_info *fs_info = root->fs_info;
6332 struct btrfs_block_group_cache *block_group, *tmp;
6333 struct list_head *deleted_bgs;
6334 struct extent_io_tree *unpin;
6339 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6340 unpin = &fs_info->freed_extents[1];
6342 unpin = &fs_info->freed_extents[0];
6344 while (!trans->aborted) {
6345 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6346 ret = find_first_extent_bit(unpin, 0, &start, &end,
6347 EXTENT_DIRTY, NULL);
6349 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6353 if (btrfs_test_opt(root, DISCARD))
6354 ret = btrfs_discard_extent(root, start,
6355 end + 1 - start, NULL);
6357 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6358 unpin_extent_range(root, start, end, true);
6359 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6364 * Transaction is finished. We don't need the lock anymore. We
6365 * do need to clean up the block groups in case of a transaction
6368 deleted_bgs = &trans->transaction->deleted_bgs;
6369 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6373 if (!trans->aborted)
6374 ret = btrfs_discard_extent(root,
6375 block_group->key.objectid,
6376 block_group->key.offset,
6379 list_del_init(&block_group->bg_list);
6380 btrfs_put_block_group_trimming(block_group);
6381 btrfs_put_block_group(block_group);
6384 const char *errstr = btrfs_decode_error(ret);
6386 "Discard failed while removing blockgroup: errno=%d %s\n",
6394 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6395 u64 owner, u64 root_objectid)
6397 struct btrfs_space_info *space_info;
6400 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6401 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6402 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6404 flags = BTRFS_BLOCK_GROUP_METADATA;
6406 flags = BTRFS_BLOCK_GROUP_DATA;
6409 space_info = __find_space_info(fs_info, flags);
6410 BUG_ON(!space_info); /* Logic bug */
6411 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6415 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6416 struct btrfs_root *root,
6417 struct btrfs_delayed_ref_node *node, u64 parent,
6418 u64 root_objectid, u64 owner_objectid,
6419 u64 owner_offset, int refs_to_drop,
6420 struct btrfs_delayed_extent_op *extent_op)
6422 struct btrfs_key key;
6423 struct btrfs_path *path;
6424 struct btrfs_fs_info *info = root->fs_info;
6425 struct btrfs_root *extent_root = info->extent_root;
6426 struct extent_buffer *leaf;
6427 struct btrfs_extent_item *ei;
6428 struct btrfs_extent_inline_ref *iref;
6431 int extent_slot = 0;
6432 int found_extent = 0;
6434 int no_quota = node->no_quota;
6437 u64 bytenr = node->bytenr;
6438 u64 num_bytes = node->num_bytes;
6440 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6443 if (!info->quota_enabled || !is_fstree(root_objectid))
6446 path = btrfs_alloc_path();
6451 path->leave_spinning = 1;
6453 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6454 BUG_ON(!is_data && refs_to_drop != 1);
6457 skinny_metadata = 0;
6459 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6460 bytenr, num_bytes, parent,
6461 root_objectid, owner_objectid,
6464 extent_slot = path->slots[0];
6465 while (extent_slot >= 0) {
6466 btrfs_item_key_to_cpu(path->nodes[0], &key,
6468 if (key.objectid != bytenr)
6470 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6471 key.offset == num_bytes) {
6475 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6476 key.offset == owner_objectid) {
6480 if (path->slots[0] - extent_slot > 5)
6484 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6485 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6486 if (found_extent && item_size < sizeof(*ei))
6489 if (!found_extent) {
6491 ret = remove_extent_backref(trans, extent_root, path,
6493 is_data, &last_ref);
6495 btrfs_abort_transaction(trans, extent_root, ret);
6498 btrfs_release_path(path);
6499 path->leave_spinning = 1;
6501 key.objectid = bytenr;
6502 key.type = BTRFS_EXTENT_ITEM_KEY;
6503 key.offset = num_bytes;
6505 if (!is_data && skinny_metadata) {
6506 key.type = BTRFS_METADATA_ITEM_KEY;
6507 key.offset = owner_objectid;
6510 ret = btrfs_search_slot(trans, extent_root,
6512 if (ret > 0 && skinny_metadata && path->slots[0]) {
6514 * Couldn't find our skinny metadata item,
6515 * see if we have ye olde extent item.
6518 btrfs_item_key_to_cpu(path->nodes[0], &key,
6520 if (key.objectid == bytenr &&
6521 key.type == BTRFS_EXTENT_ITEM_KEY &&
6522 key.offset == num_bytes)
6526 if (ret > 0 && skinny_metadata) {
6527 skinny_metadata = false;
6528 key.objectid = bytenr;
6529 key.type = BTRFS_EXTENT_ITEM_KEY;
6530 key.offset = num_bytes;
6531 btrfs_release_path(path);
6532 ret = btrfs_search_slot(trans, extent_root,
6537 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6540 btrfs_print_leaf(extent_root,
6544 btrfs_abort_transaction(trans, extent_root, ret);
6547 extent_slot = path->slots[0];
6549 } else if (WARN_ON(ret == -ENOENT)) {
6550 btrfs_print_leaf(extent_root, path->nodes[0]);
6552 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6553 bytenr, parent, root_objectid, owner_objectid,
6555 btrfs_abort_transaction(trans, extent_root, ret);
6558 btrfs_abort_transaction(trans, extent_root, ret);
6562 leaf = path->nodes[0];
6563 item_size = btrfs_item_size_nr(leaf, extent_slot);
6564 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6565 if (item_size < sizeof(*ei)) {
6566 BUG_ON(found_extent || extent_slot != path->slots[0]);
6567 ret = convert_extent_item_v0(trans, extent_root, path,
6570 btrfs_abort_transaction(trans, extent_root, ret);
6574 btrfs_release_path(path);
6575 path->leave_spinning = 1;
6577 key.objectid = bytenr;
6578 key.type = BTRFS_EXTENT_ITEM_KEY;
6579 key.offset = num_bytes;
6581 ret = btrfs_search_slot(trans, extent_root, &key, path,
6584 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6586 btrfs_print_leaf(extent_root, path->nodes[0]);
6589 btrfs_abort_transaction(trans, extent_root, ret);
6593 extent_slot = path->slots[0];
6594 leaf = path->nodes[0];
6595 item_size = btrfs_item_size_nr(leaf, extent_slot);
6598 BUG_ON(item_size < sizeof(*ei));
6599 ei = btrfs_item_ptr(leaf, extent_slot,
6600 struct btrfs_extent_item);
6601 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6602 key.type == BTRFS_EXTENT_ITEM_KEY) {
6603 struct btrfs_tree_block_info *bi;
6604 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6605 bi = (struct btrfs_tree_block_info *)(ei + 1);
6606 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6609 refs = btrfs_extent_refs(leaf, ei);
6610 if (refs < refs_to_drop) {
6611 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6612 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6614 btrfs_abort_transaction(trans, extent_root, ret);
6617 refs -= refs_to_drop;
6621 __run_delayed_extent_op(extent_op, leaf, ei);
6623 * In the case of inline back ref, reference count will
6624 * be updated by remove_extent_backref
6627 BUG_ON(!found_extent);
6629 btrfs_set_extent_refs(leaf, ei, refs);
6630 btrfs_mark_buffer_dirty(leaf);
6633 ret = remove_extent_backref(trans, extent_root, path,
6635 is_data, &last_ref);
6637 btrfs_abort_transaction(trans, extent_root, ret);
6641 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6645 BUG_ON(is_data && refs_to_drop !=
6646 extent_data_ref_count(path, iref));
6648 BUG_ON(path->slots[0] != extent_slot);
6650 BUG_ON(path->slots[0] != extent_slot + 1);
6651 path->slots[0] = extent_slot;
6657 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6660 btrfs_abort_transaction(trans, extent_root, ret);
6663 btrfs_release_path(path);
6666 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6668 btrfs_abort_transaction(trans, extent_root, ret);
6673 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6675 btrfs_abort_transaction(trans, extent_root, ret);
6679 btrfs_release_path(path);
6682 btrfs_free_path(path);
6687 * when we free an block, it is possible (and likely) that we free the last
6688 * delayed ref for that extent as well. This searches the delayed ref tree for
6689 * a given extent, and if there are no other delayed refs to be processed, it
6690 * removes it from the tree.
6692 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6693 struct btrfs_root *root, u64 bytenr)
6695 struct btrfs_delayed_ref_head *head;
6696 struct btrfs_delayed_ref_root *delayed_refs;
6699 delayed_refs = &trans->transaction->delayed_refs;
6700 spin_lock(&delayed_refs->lock);
6701 head = btrfs_find_delayed_ref_head(trans, bytenr);
6703 goto out_delayed_unlock;
6705 spin_lock(&head->lock);
6706 if (!list_empty(&head->ref_list))
6709 if (head->extent_op) {
6710 if (!head->must_insert_reserved)
6712 btrfs_free_delayed_extent_op(head->extent_op);
6713 head->extent_op = NULL;
6717 * waiting for the lock here would deadlock. If someone else has it
6718 * locked they are already in the process of dropping it anyway
6720 if (!mutex_trylock(&head->mutex))
6724 * at this point we have a head with no other entries. Go
6725 * ahead and process it.
6727 head->node.in_tree = 0;
6728 rb_erase(&head->href_node, &delayed_refs->href_root);
6730 atomic_dec(&delayed_refs->num_entries);
6733 * we don't take a ref on the node because we're removing it from the
6734 * tree, so we just steal the ref the tree was holding.
6736 delayed_refs->num_heads--;
6737 if (head->processing == 0)
6738 delayed_refs->num_heads_ready--;
6739 head->processing = 0;
6740 spin_unlock(&head->lock);
6741 spin_unlock(&delayed_refs->lock);
6743 BUG_ON(head->extent_op);
6744 if (head->must_insert_reserved)
6747 mutex_unlock(&head->mutex);
6748 btrfs_put_delayed_ref(&head->node);
6751 spin_unlock(&head->lock);
6754 spin_unlock(&delayed_refs->lock);
6758 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6759 struct btrfs_root *root,
6760 struct extent_buffer *buf,
6761 u64 parent, int last_ref)
6766 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6767 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6768 buf->start, buf->len,
6769 parent, root->root_key.objectid,
6770 btrfs_header_level(buf),
6771 BTRFS_DROP_DELAYED_REF, NULL, 0);
6772 BUG_ON(ret); /* -ENOMEM */
6778 if (btrfs_header_generation(buf) == trans->transid) {
6779 struct btrfs_block_group_cache *cache;
6781 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6782 ret = check_ref_cleanup(trans, root, buf->start);
6787 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6789 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6790 pin_down_extent(root, cache, buf->start, buf->len, 1);
6791 btrfs_put_block_group(cache);
6795 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6797 btrfs_add_free_space(cache, buf->start, buf->len);
6798 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6799 btrfs_put_block_group(cache);
6800 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6805 add_pinned_bytes(root->fs_info, buf->len,
6806 btrfs_header_level(buf),
6807 root->root_key.objectid);
6810 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6813 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6816 /* Can return -ENOMEM */
6817 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6818 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6819 u64 owner, u64 offset, int no_quota)
6822 struct btrfs_fs_info *fs_info = root->fs_info;
6824 if (btrfs_test_is_dummy_root(root))
6827 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6830 * tree log blocks never actually go into the extent allocation
6831 * tree, just update pinning info and exit early.
6833 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6834 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6835 /* unlocks the pinned mutex */
6836 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6838 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6839 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6841 parent, root_objectid, (int)owner,
6842 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6844 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6846 parent, root_objectid, owner,
6847 offset, BTRFS_DROP_DELAYED_REF,
6854 * when we wait for progress in the block group caching, its because
6855 * our allocation attempt failed at least once. So, we must sleep
6856 * and let some progress happen before we try again.
6858 * This function will sleep at least once waiting for new free space to
6859 * show up, and then it will check the block group free space numbers
6860 * for our min num_bytes. Another option is to have it go ahead
6861 * and look in the rbtree for a free extent of a given size, but this
6864 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6865 * any of the information in this block group.
6867 static noinline void
6868 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6871 struct btrfs_caching_control *caching_ctl;
6873 caching_ctl = get_caching_control(cache);
6877 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6878 (cache->free_space_ctl->free_space >= num_bytes));
6880 put_caching_control(caching_ctl);
6884 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6886 struct btrfs_caching_control *caching_ctl;
6889 caching_ctl = get_caching_control(cache);
6891 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6893 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6894 if (cache->cached == BTRFS_CACHE_ERROR)
6896 put_caching_control(caching_ctl);
6900 int __get_raid_index(u64 flags)
6902 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6903 return BTRFS_RAID_RAID10;
6904 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6905 return BTRFS_RAID_RAID1;
6906 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6907 return BTRFS_RAID_DUP;
6908 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6909 return BTRFS_RAID_RAID0;
6910 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6911 return BTRFS_RAID_RAID5;
6912 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6913 return BTRFS_RAID_RAID6;
6915 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6918 int get_block_group_index(struct btrfs_block_group_cache *cache)
6920 return __get_raid_index(cache->flags);
6923 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6924 [BTRFS_RAID_RAID10] = "raid10",
6925 [BTRFS_RAID_RAID1] = "raid1",
6926 [BTRFS_RAID_DUP] = "dup",
6927 [BTRFS_RAID_RAID0] = "raid0",
6928 [BTRFS_RAID_SINGLE] = "single",
6929 [BTRFS_RAID_RAID5] = "raid5",
6930 [BTRFS_RAID_RAID6] = "raid6",
6933 static const char *get_raid_name(enum btrfs_raid_types type)
6935 if (type >= BTRFS_NR_RAID_TYPES)
6938 return btrfs_raid_type_names[type];
6941 enum btrfs_loop_type {
6942 LOOP_CACHING_NOWAIT = 0,
6943 LOOP_CACHING_WAIT = 1,
6944 LOOP_ALLOC_CHUNK = 2,
6945 LOOP_NO_EMPTY_SIZE = 3,
6949 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6953 down_read(&cache->data_rwsem);
6957 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6960 btrfs_get_block_group(cache);
6962 down_read(&cache->data_rwsem);
6965 static struct btrfs_block_group_cache *
6966 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6967 struct btrfs_free_cluster *cluster,
6970 struct btrfs_block_group_cache *used_bg;
6971 bool locked = false;
6973 spin_lock(&cluster->refill_lock);
6975 if (used_bg == cluster->block_group)
6978 up_read(&used_bg->data_rwsem);
6979 btrfs_put_block_group(used_bg);
6982 used_bg = cluster->block_group;
6986 if (used_bg == block_group)
6989 btrfs_get_block_group(used_bg);
6994 if (down_read_trylock(&used_bg->data_rwsem))
6997 spin_unlock(&cluster->refill_lock);
6998 down_read(&used_bg->data_rwsem);
7004 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
7008 up_read(&cache->data_rwsem);
7009 btrfs_put_block_group(cache);
7013 * walks the btree of allocated extents and find a hole of a given size.
7014 * The key ins is changed to record the hole:
7015 * ins->objectid == start position
7016 * ins->flags = BTRFS_EXTENT_ITEM_KEY
7017 * ins->offset == the size of the hole.
7018 * Any available blocks before search_start are skipped.
7020 * If there is no suitable free space, we will record the max size of
7021 * the free space extent currently.
7023 static noinline int find_free_extent(struct btrfs_root *orig_root,
7024 u64 num_bytes, u64 empty_size,
7025 u64 hint_byte, struct btrfs_key *ins,
7026 u64 flags, int delalloc)
7029 struct btrfs_root *root = orig_root->fs_info->extent_root;
7030 struct btrfs_free_cluster *last_ptr = NULL;
7031 struct btrfs_block_group_cache *block_group = NULL;
7032 u64 search_start = 0;
7033 u64 max_extent_size = 0;
7034 u64 empty_cluster = 0;
7035 struct btrfs_space_info *space_info;
7037 int index = __get_raid_index(flags);
7038 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
7039 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
7040 bool failed_cluster_refill = false;
7041 bool failed_alloc = false;
7042 bool use_cluster = true;
7043 bool have_caching_bg = false;
7044 bool full_search = false;
7046 WARN_ON(num_bytes < root->sectorsize);
7047 ins->type = BTRFS_EXTENT_ITEM_KEY;
7051 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
7053 space_info = __find_space_info(root->fs_info, flags);
7055 btrfs_err(root->fs_info, "No space info for %llu", flags);
7060 * If our free space is heavily fragmented we may not be able to make
7061 * big contiguous allocations, so instead of doing the expensive search
7062 * for free space, simply return ENOSPC with our max_extent_size so we
7063 * can go ahead and search for a more manageable chunk.
7065 * If our max_extent_size is large enough for our allocation simply
7066 * disable clustering since we will likely not be able to find enough
7067 * space to create a cluster and induce latency trying.
7069 if (unlikely(space_info->max_extent_size)) {
7070 spin_lock(&space_info->lock);
7071 if (space_info->max_extent_size &&
7072 num_bytes > space_info->max_extent_size) {
7073 ins->offset = space_info->max_extent_size;
7074 spin_unlock(&space_info->lock);
7076 } else if (space_info->max_extent_size) {
7077 use_cluster = false;
7079 spin_unlock(&space_info->lock);
7082 last_ptr = fetch_cluster_info(orig_root, space_info, &empty_cluster);
7084 spin_lock(&last_ptr->lock);
7085 if (last_ptr->block_group)
7086 hint_byte = last_ptr->window_start;
7087 if (last_ptr->fragmented) {
7089 * We still set window_start so we can keep track of the
7090 * last place we found an allocation to try and save
7093 hint_byte = last_ptr->window_start;
7094 use_cluster = false;
7096 spin_unlock(&last_ptr->lock);
7099 search_start = max(search_start, first_logical_byte(root, 0));
7100 search_start = max(search_start, hint_byte);
7101 if (search_start == hint_byte) {
7102 block_group = btrfs_lookup_block_group(root->fs_info,
7105 * we don't want to use the block group if it doesn't match our
7106 * allocation bits, or if its not cached.
7108 * However if we are re-searching with an ideal block group
7109 * picked out then we don't care that the block group is cached.
7111 if (block_group && block_group_bits(block_group, flags) &&
7112 block_group->cached != BTRFS_CACHE_NO) {
7113 down_read(&space_info->groups_sem);
7114 if (list_empty(&block_group->list) ||
7117 * someone is removing this block group,
7118 * we can't jump into the have_block_group
7119 * target because our list pointers are not
7122 btrfs_put_block_group(block_group);
7123 up_read(&space_info->groups_sem);
7125 index = get_block_group_index(block_group);
7126 btrfs_lock_block_group(block_group, delalloc);
7127 goto have_block_group;
7129 } else if (block_group) {
7130 btrfs_put_block_group(block_group);
7134 have_caching_bg = false;
7135 if (index == 0 || index == __get_raid_index(flags))
7137 down_read(&space_info->groups_sem);
7138 list_for_each_entry(block_group, &space_info->block_groups[index],
7143 btrfs_grab_block_group(block_group, delalloc);
7144 search_start = block_group->key.objectid;
7147 * this can happen if we end up cycling through all the
7148 * raid types, but we want to make sure we only allocate
7149 * for the proper type.
7151 if (!block_group_bits(block_group, flags)) {
7152 u64 extra = BTRFS_BLOCK_GROUP_DUP |
7153 BTRFS_BLOCK_GROUP_RAID1 |
7154 BTRFS_BLOCK_GROUP_RAID5 |
7155 BTRFS_BLOCK_GROUP_RAID6 |
7156 BTRFS_BLOCK_GROUP_RAID10;
7159 * if they asked for extra copies and this block group
7160 * doesn't provide them, bail. This does allow us to
7161 * fill raid0 from raid1.
7163 if ((flags & extra) && !(block_group->flags & extra))
7168 cached = block_group_cache_done(block_group);
7169 if (unlikely(!cached)) {
7170 have_caching_bg = true;
7171 ret = cache_block_group(block_group, 0);
7176 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7178 if (unlikely(block_group->ro))
7182 * Ok we want to try and use the cluster allocator, so
7185 if (last_ptr && use_cluster) {
7186 struct btrfs_block_group_cache *used_block_group;
7187 unsigned long aligned_cluster;
7189 * the refill lock keeps out other
7190 * people trying to start a new cluster
7192 used_block_group = btrfs_lock_cluster(block_group,
7195 if (!used_block_group)
7196 goto refill_cluster;
7198 if (used_block_group != block_group &&
7199 (used_block_group->ro ||
7200 !block_group_bits(used_block_group, flags)))
7201 goto release_cluster;
7203 offset = btrfs_alloc_from_cluster(used_block_group,
7206 used_block_group->key.objectid,
7209 /* we have a block, we're done */
7210 spin_unlock(&last_ptr->refill_lock);
7211 trace_btrfs_reserve_extent_cluster(root,
7213 search_start, num_bytes);
7214 if (used_block_group != block_group) {
7215 btrfs_release_block_group(block_group,
7217 block_group = used_block_group;
7222 WARN_ON(last_ptr->block_group != used_block_group);
7224 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7225 * set up a new clusters, so lets just skip it
7226 * and let the allocator find whatever block
7227 * it can find. If we reach this point, we
7228 * will have tried the cluster allocator
7229 * plenty of times and not have found
7230 * anything, so we are likely way too
7231 * fragmented for the clustering stuff to find
7234 * However, if the cluster is taken from the
7235 * current block group, release the cluster
7236 * first, so that we stand a better chance of
7237 * succeeding in the unclustered
7239 if (loop >= LOOP_NO_EMPTY_SIZE &&
7240 used_block_group != block_group) {
7241 spin_unlock(&last_ptr->refill_lock);
7242 btrfs_release_block_group(used_block_group,
7244 goto unclustered_alloc;
7248 * this cluster didn't work out, free it and
7251 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7253 if (used_block_group != block_group)
7254 btrfs_release_block_group(used_block_group,
7257 if (loop >= LOOP_NO_EMPTY_SIZE) {
7258 spin_unlock(&last_ptr->refill_lock);
7259 goto unclustered_alloc;
7262 aligned_cluster = max_t(unsigned long,
7263 empty_cluster + empty_size,
7264 block_group->full_stripe_len);
7266 /* allocate a cluster in this block group */
7267 ret = btrfs_find_space_cluster(root, block_group,
7268 last_ptr, search_start,
7273 * now pull our allocation out of this
7276 offset = btrfs_alloc_from_cluster(block_group,
7282 /* we found one, proceed */
7283 spin_unlock(&last_ptr->refill_lock);
7284 trace_btrfs_reserve_extent_cluster(root,
7285 block_group, search_start,
7289 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7290 && !failed_cluster_refill) {
7291 spin_unlock(&last_ptr->refill_lock);
7293 failed_cluster_refill = true;
7294 wait_block_group_cache_progress(block_group,
7295 num_bytes + empty_cluster + empty_size);
7296 goto have_block_group;
7300 * at this point we either didn't find a cluster
7301 * or we weren't able to allocate a block from our
7302 * cluster. Free the cluster we've been trying
7303 * to use, and go to the next block group
7305 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7306 spin_unlock(&last_ptr->refill_lock);
7312 * We are doing an unclustered alloc, set the fragmented flag so
7313 * we don't bother trying to setup a cluster again until we get
7316 if (unlikely(last_ptr)) {
7317 spin_lock(&last_ptr->lock);
7318 last_ptr->fragmented = 1;
7319 spin_unlock(&last_ptr->lock);
7321 spin_lock(&block_group->free_space_ctl->tree_lock);
7323 block_group->free_space_ctl->free_space <
7324 num_bytes + empty_cluster + empty_size) {
7325 if (block_group->free_space_ctl->free_space >
7328 block_group->free_space_ctl->free_space;
7329 spin_unlock(&block_group->free_space_ctl->tree_lock);
7332 spin_unlock(&block_group->free_space_ctl->tree_lock);
7334 offset = btrfs_find_space_for_alloc(block_group, search_start,
7335 num_bytes, empty_size,
7338 * If we didn't find a chunk, and we haven't failed on this
7339 * block group before, and this block group is in the middle of
7340 * caching and we are ok with waiting, then go ahead and wait
7341 * for progress to be made, and set failed_alloc to true.
7343 * If failed_alloc is true then we've already waited on this
7344 * block group once and should move on to the next block group.
7346 if (!offset && !failed_alloc && !cached &&
7347 loop > LOOP_CACHING_NOWAIT) {
7348 wait_block_group_cache_progress(block_group,
7349 num_bytes + empty_size);
7350 failed_alloc = true;
7351 goto have_block_group;
7352 } else if (!offset) {
7356 search_start = ALIGN(offset, root->stripesize);
7358 /* move on to the next group */
7359 if (search_start + num_bytes >
7360 block_group->key.objectid + block_group->key.offset) {
7361 btrfs_add_free_space(block_group, offset, num_bytes);
7365 if (offset < search_start)
7366 btrfs_add_free_space(block_group, offset,
7367 search_start - offset);
7368 BUG_ON(offset > search_start);
7370 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7371 alloc_type, delalloc);
7372 if (ret == -EAGAIN) {
7373 btrfs_add_free_space(block_group, offset, num_bytes);
7377 /* we are all good, lets return */
7378 ins->objectid = search_start;
7379 ins->offset = num_bytes;
7381 trace_btrfs_reserve_extent(orig_root, block_group,
7382 search_start, num_bytes);
7383 btrfs_release_block_group(block_group, delalloc);
7386 failed_cluster_refill = false;
7387 failed_alloc = false;
7388 BUG_ON(index != get_block_group_index(block_group));
7389 btrfs_release_block_group(block_group, delalloc);
7391 up_read(&space_info->groups_sem);
7393 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7396 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7400 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7401 * caching kthreads as we move along
7402 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7403 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7404 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7407 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7409 if (loop == LOOP_CACHING_NOWAIT) {
7411 * We want to skip the LOOP_CACHING_WAIT step if we
7412 * don't have any unached bgs and we've alrelady done a
7413 * full search through.
7415 if (have_caching_bg || !full_search)
7416 loop = LOOP_CACHING_WAIT;
7418 loop = LOOP_ALLOC_CHUNK;
7423 if (loop == LOOP_ALLOC_CHUNK) {
7424 struct btrfs_trans_handle *trans;
7427 trans = current->journal_info;
7431 trans = btrfs_join_transaction(root);
7433 if (IS_ERR(trans)) {
7434 ret = PTR_ERR(trans);
7438 ret = do_chunk_alloc(trans, root, flags,
7442 * If we can't allocate a new chunk we've already looped
7443 * through at least once, move on to the NO_EMPTY_SIZE
7447 loop = LOOP_NO_EMPTY_SIZE;
7450 * Do not bail out on ENOSPC since we
7451 * can do more things.
7453 if (ret < 0 && ret != -ENOSPC)
7454 btrfs_abort_transaction(trans,
7459 btrfs_end_transaction(trans, root);
7464 if (loop == LOOP_NO_EMPTY_SIZE) {
7466 * Don't loop again if we already have no empty_size and
7469 if (empty_size == 0 &&
7470 empty_cluster == 0) {
7479 } else if (!ins->objectid) {
7481 } else if (ins->objectid) {
7482 if (!use_cluster && last_ptr) {
7483 spin_lock(&last_ptr->lock);
7484 last_ptr->window_start = ins->objectid;
7485 spin_unlock(&last_ptr->lock);
7490 if (ret == -ENOSPC) {
7491 spin_lock(&space_info->lock);
7492 space_info->max_extent_size = max_extent_size;
7493 spin_unlock(&space_info->lock);
7494 ins->offset = max_extent_size;
7499 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7500 int dump_block_groups)
7502 struct btrfs_block_group_cache *cache;
7505 spin_lock(&info->lock);
7506 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7508 info->total_bytes - info->bytes_used - info->bytes_pinned -
7509 info->bytes_reserved - info->bytes_readonly,
7510 (info->full) ? "" : "not ");
7511 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7512 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7513 info->total_bytes, info->bytes_used, info->bytes_pinned,
7514 info->bytes_reserved, info->bytes_may_use,
7515 info->bytes_readonly);
7516 spin_unlock(&info->lock);
7518 if (!dump_block_groups)
7521 down_read(&info->groups_sem);
7523 list_for_each_entry(cache, &info->block_groups[index], list) {
7524 spin_lock(&cache->lock);
7525 printk(KERN_INFO "BTRFS: "
7526 "block group %llu has %llu bytes, "
7527 "%llu used %llu pinned %llu reserved %s\n",
7528 cache->key.objectid, cache->key.offset,
7529 btrfs_block_group_used(&cache->item), cache->pinned,
7530 cache->reserved, cache->ro ? "[readonly]" : "");
7531 btrfs_dump_free_space(cache, bytes);
7532 spin_unlock(&cache->lock);
7534 if (++index < BTRFS_NR_RAID_TYPES)
7536 up_read(&info->groups_sem);
7539 int btrfs_reserve_extent(struct btrfs_root *root,
7540 u64 num_bytes, u64 min_alloc_size,
7541 u64 empty_size, u64 hint_byte,
7542 struct btrfs_key *ins, int is_data, int delalloc)
7544 bool final_tried = num_bytes == min_alloc_size;
7548 flags = btrfs_get_alloc_profile(root, is_data);
7550 WARN_ON(num_bytes < root->sectorsize);
7551 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7554 if (ret == -ENOSPC) {
7555 if (!final_tried && ins->offset) {
7556 num_bytes = min(num_bytes >> 1, ins->offset);
7557 num_bytes = round_down(num_bytes, root->sectorsize);
7558 num_bytes = max(num_bytes, min_alloc_size);
7559 if (num_bytes == min_alloc_size)
7562 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7563 struct btrfs_space_info *sinfo;
7565 sinfo = __find_space_info(root->fs_info, flags);
7566 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7569 dump_space_info(sinfo, num_bytes, 1);
7576 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7578 int pin, int delalloc)
7580 struct btrfs_block_group_cache *cache;
7583 cache = btrfs_lookup_block_group(root->fs_info, start);
7585 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7591 pin_down_extent(root, cache, start, len, 1);
7593 if (btrfs_test_opt(root, DISCARD))
7594 ret = btrfs_discard_extent(root, start, len, NULL);
7595 btrfs_add_free_space(cache, start, len);
7596 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7599 btrfs_put_block_group(cache);
7601 trace_btrfs_reserved_extent_free(root, start, len);
7606 int btrfs_free_reserved_extent(struct btrfs_root *root,
7607 u64 start, u64 len, int delalloc)
7609 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7612 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7615 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7618 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7619 struct btrfs_root *root,
7620 u64 parent, u64 root_objectid,
7621 u64 flags, u64 owner, u64 offset,
7622 struct btrfs_key *ins, int ref_mod)
7625 struct btrfs_fs_info *fs_info = root->fs_info;
7626 struct btrfs_extent_item *extent_item;
7627 struct btrfs_extent_inline_ref *iref;
7628 struct btrfs_path *path;
7629 struct extent_buffer *leaf;
7634 type = BTRFS_SHARED_DATA_REF_KEY;
7636 type = BTRFS_EXTENT_DATA_REF_KEY;
7638 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7640 path = btrfs_alloc_path();
7644 path->leave_spinning = 1;
7645 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7648 btrfs_free_path(path);
7652 leaf = path->nodes[0];
7653 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7654 struct btrfs_extent_item);
7655 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7656 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7657 btrfs_set_extent_flags(leaf, extent_item,
7658 flags | BTRFS_EXTENT_FLAG_DATA);
7660 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7661 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7663 struct btrfs_shared_data_ref *ref;
7664 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7665 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7666 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7668 struct btrfs_extent_data_ref *ref;
7669 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7670 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7671 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7672 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7673 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7676 btrfs_mark_buffer_dirty(path->nodes[0]);
7677 btrfs_free_path(path);
7679 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7680 if (ret) { /* -ENOENT, logic error */
7681 btrfs_err(fs_info, "update block group failed for %llu %llu",
7682 ins->objectid, ins->offset);
7685 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7689 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7690 struct btrfs_root *root,
7691 u64 parent, u64 root_objectid,
7692 u64 flags, struct btrfs_disk_key *key,
7693 int level, struct btrfs_key *ins,
7697 struct btrfs_fs_info *fs_info = root->fs_info;
7698 struct btrfs_extent_item *extent_item;
7699 struct btrfs_tree_block_info *block_info;
7700 struct btrfs_extent_inline_ref *iref;
7701 struct btrfs_path *path;
7702 struct extent_buffer *leaf;
7703 u32 size = sizeof(*extent_item) + sizeof(*iref);
7704 u64 num_bytes = ins->offset;
7705 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7708 if (!skinny_metadata)
7709 size += sizeof(*block_info);
7711 path = btrfs_alloc_path();
7713 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7718 path->leave_spinning = 1;
7719 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7722 btrfs_free_path(path);
7723 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7728 leaf = path->nodes[0];
7729 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7730 struct btrfs_extent_item);
7731 btrfs_set_extent_refs(leaf, extent_item, 1);
7732 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7733 btrfs_set_extent_flags(leaf, extent_item,
7734 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7736 if (skinny_metadata) {
7737 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7738 num_bytes = root->nodesize;
7740 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7741 btrfs_set_tree_block_key(leaf, block_info, key);
7742 btrfs_set_tree_block_level(leaf, block_info, level);
7743 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7747 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7748 btrfs_set_extent_inline_ref_type(leaf, iref,
7749 BTRFS_SHARED_BLOCK_REF_KEY);
7750 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7752 btrfs_set_extent_inline_ref_type(leaf, iref,
7753 BTRFS_TREE_BLOCK_REF_KEY);
7754 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7757 btrfs_mark_buffer_dirty(leaf);
7758 btrfs_free_path(path);
7760 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7762 if (ret) { /* -ENOENT, logic error */
7763 btrfs_err(fs_info, "update block group failed for %llu %llu",
7764 ins->objectid, ins->offset);
7768 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7772 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7773 struct btrfs_root *root,
7774 u64 root_objectid, u64 owner,
7775 u64 offset, struct btrfs_key *ins)
7779 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7781 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7783 root_objectid, owner, offset,
7784 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7789 * this is used by the tree logging recovery code. It records that
7790 * an extent has been allocated and makes sure to clear the free
7791 * space cache bits as well
7793 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7794 struct btrfs_root *root,
7795 u64 root_objectid, u64 owner, u64 offset,
7796 struct btrfs_key *ins)
7799 struct btrfs_block_group_cache *block_group;
7802 * Mixed block groups will exclude before processing the log so we only
7803 * need to do the exlude dance if this fs isn't mixed.
7805 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7806 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7811 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7815 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7816 RESERVE_ALLOC_NO_ACCOUNT, 0);
7817 BUG_ON(ret); /* logic error */
7818 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7819 0, owner, offset, ins, 1);
7820 btrfs_put_block_group(block_group);
7824 static struct extent_buffer *
7825 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7826 u64 bytenr, int level)
7828 struct extent_buffer *buf;
7830 buf = btrfs_find_create_tree_block(root, bytenr);
7832 return ERR_PTR(-ENOMEM);
7833 btrfs_set_header_generation(buf, trans->transid);
7834 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7835 btrfs_tree_lock(buf);
7836 clean_tree_block(trans, root->fs_info, buf);
7837 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7839 btrfs_set_lock_blocking(buf);
7840 btrfs_set_buffer_uptodate(buf);
7842 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7843 buf->log_index = root->log_transid % 2;
7845 * we allow two log transactions at a time, use different
7846 * EXENT bit to differentiate dirty pages.
7848 if (buf->log_index == 0)
7849 set_extent_dirty(&root->dirty_log_pages, buf->start,
7850 buf->start + buf->len - 1, GFP_NOFS);
7852 set_extent_new(&root->dirty_log_pages, buf->start,
7853 buf->start + buf->len - 1, GFP_NOFS);
7855 buf->log_index = -1;
7856 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7857 buf->start + buf->len - 1, GFP_NOFS);
7859 trans->blocks_used++;
7860 /* this returns a buffer locked for blocking */
7864 static struct btrfs_block_rsv *
7865 use_block_rsv(struct btrfs_trans_handle *trans,
7866 struct btrfs_root *root, u32 blocksize)
7868 struct btrfs_block_rsv *block_rsv;
7869 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7871 bool global_updated = false;
7873 block_rsv = get_block_rsv(trans, root);
7875 if (unlikely(block_rsv->size == 0))
7878 ret = block_rsv_use_bytes(block_rsv, blocksize);
7882 if (block_rsv->failfast)
7883 return ERR_PTR(ret);
7885 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7886 global_updated = true;
7887 update_global_block_rsv(root->fs_info);
7891 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7892 static DEFINE_RATELIMIT_STATE(_rs,
7893 DEFAULT_RATELIMIT_INTERVAL * 10,
7894 /*DEFAULT_RATELIMIT_BURST*/ 1);
7895 if (__ratelimit(&_rs))
7897 "BTRFS: block rsv returned %d\n", ret);
7900 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7901 BTRFS_RESERVE_NO_FLUSH);
7905 * If we couldn't reserve metadata bytes try and use some from
7906 * the global reserve if its space type is the same as the global
7909 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7910 block_rsv->space_info == global_rsv->space_info) {
7911 ret = block_rsv_use_bytes(global_rsv, blocksize);
7915 return ERR_PTR(ret);
7918 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7919 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7921 block_rsv_add_bytes(block_rsv, blocksize, 0);
7922 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7926 * finds a free extent and does all the dirty work required for allocation
7927 * returns the tree buffer or an ERR_PTR on error.
7929 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7930 struct btrfs_root *root,
7931 u64 parent, u64 root_objectid,
7932 struct btrfs_disk_key *key, int level,
7933 u64 hint, u64 empty_size)
7935 struct btrfs_key ins;
7936 struct btrfs_block_rsv *block_rsv;
7937 struct extent_buffer *buf;
7938 struct btrfs_delayed_extent_op *extent_op;
7941 u32 blocksize = root->nodesize;
7942 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7945 if (btrfs_test_is_dummy_root(root)) {
7946 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7949 root->alloc_bytenr += blocksize;
7953 block_rsv = use_block_rsv(trans, root, blocksize);
7954 if (IS_ERR(block_rsv))
7955 return ERR_CAST(block_rsv);
7957 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7958 empty_size, hint, &ins, 0, 0);
7962 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7965 goto out_free_reserved;
7968 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7970 parent = ins.objectid;
7971 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7975 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7976 extent_op = btrfs_alloc_delayed_extent_op();
7982 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7984 memset(&extent_op->key, 0, sizeof(extent_op->key));
7985 extent_op->flags_to_set = flags;
7986 if (skinny_metadata)
7987 extent_op->update_key = 0;
7989 extent_op->update_key = 1;
7990 extent_op->update_flags = 1;
7991 extent_op->is_data = 0;
7992 extent_op->level = level;
7994 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7995 ins.objectid, ins.offset,
7996 parent, root_objectid, level,
7997 BTRFS_ADD_DELAYED_EXTENT,
8000 goto out_free_delayed;
8005 btrfs_free_delayed_extent_op(extent_op);
8007 free_extent_buffer(buf);
8009 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
8011 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
8012 return ERR_PTR(ret);
8015 struct walk_control {
8016 u64 refs[BTRFS_MAX_LEVEL];
8017 u64 flags[BTRFS_MAX_LEVEL];
8018 struct btrfs_key update_progress;
8029 #define DROP_REFERENCE 1
8030 #define UPDATE_BACKREF 2
8032 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
8033 struct btrfs_root *root,
8034 struct walk_control *wc,
8035 struct btrfs_path *path)
8043 struct btrfs_key key;
8044 struct extent_buffer *eb;
8049 if (path->slots[wc->level] < wc->reada_slot) {
8050 wc->reada_count = wc->reada_count * 2 / 3;
8051 wc->reada_count = max(wc->reada_count, 2);
8053 wc->reada_count = wc->reada_count * 3 / 2;
8054 wc->reada_count = min_t(int, wc->reada_count,
8055 BTRFS_NODEPTRS_PER_BLOCK(root));
8058 eb = path->nodes[wc->level];
8059 nritems = btrfs_header_nritems(eb);
8060 blocksize = root->nodesize;
8062 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
8063 if (nread >= wc->reada_count)
8067 bytenr = btrfs_node_blockptr(eb, slot);
8068 generation = btrfs_node_ptr_generation(eb, slot);
8070 if (slot == path->slots[wc->level])
8073 if (wc->stage == UPDATE_BACKREF &&
8074 generation <= root->root_key.offset)
8077 /* We don't lock the tree block, it's OK to be racy here */
8078 ret = btrfs_lookup_extent_info(trans, root, bytenr,
8079 wc->level - 1, 1, &refs,
8081 /* We don't care about errors in readahead. */
8086 if (wc->stage == DROP_REFERENCE) {
8090 if (wc->level == 1 &&
8091 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8093 if (!wc->update_ref ||
8094 generation <= root->root_key.offset)
8096 btrfs_node_key_to_cpu(eb, &key, slot);
8097 ret = btrfs_comp_cpu_keys(&key,
8098 &wc->update_progress);
8102 if (wc->level == 1 &&
8103 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8107 readahead_tree_block(root, bytenr);
8110 wc->reada_slot = slot;
8114 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
8115 * for later qgroup accounting.
8117 * Current, this function does nothing.
8119 static int account_leaf_items(struct btrfs_trans_handle *trans,
8120 struct btrfs_root *root,
8121 struct extent_buffer *eb)
8123 int nr = btrfs_header_nritems(eb);
8125 struct btrfs_key key;
8126 struct btrfs_file_extent_item *fi;
8127 u64 bytenr, num_bytes;
8129 for (i = 0; i < nr; i++) {
8130 btrfs_item_key_to_cpu(eb, &key, i);
8132 if (key.type != BTRFS_EXTENT_DATA_KEY)
8135 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
8136 /* filter out non qgroup-accountable extents */
8137 extent_type = btrfs_file_extent_type(eb, fi);
8139 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
8142 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
8146 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
8152 * Walk up the tree from the bottom, freeing leaves and any interior
8153 * nodes which have had all slots visited. If a node (leaf or
8154 * interior) is freed, the node above it will have it's slot
8155 * incremented. The root node will never be freed.
8157 * At the end of this function, we should have a path which has all
8158 * slots incremented to the next position for a search. If we need to
8159 * read a new node it will be NULL and the node above it will have the
8160 * correct slot selected for a later read.
8162 * If we increment the root nodes slot counter past the number of
8163 * elements, 1 is returned to signal completion of the search.
8165 static int adjust_slots_upwards(struct btrfs_root *root,
8166 struct btrfs_path *path, int root_level)
8170 struct extent_buffer *eb;
8172 if (root_level == 0)
8175 while (level <= root_level) {
8176 eb = path->nodes[level];
8177 nr = btrfs_header_nritems(eb);
8178 path->slots[level]++;
8179 slot = path->slots[level];
8180 if (slot >= nr || level == 0) {
8182 * Don't free the root - we will detect this
8183 * condition after our loop and return a
8184 * positive value for caller to stop walking the tree.
8186 if (level != root_level) {
8187 btrfs_tree_unlock_rw(eb, path->locks[level]);
8188 path->locks[level] = 0;
8190 free_extent_buffer(eb);
8191 path->nodes[level] = NULL;
8192 path->slots[level] = 0;
8196 * We have a valid slot to walk back down
8197 * from. Stop here so caller can process these
8206 eb = path->nodes[root_level];
8207 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8214 * root_eb is the subtree root and is locked before this function is called.
8215 * TODO: Modify this function to mark all (including complete shared node)
8216 * to dirty_extent_root to allow it get accounted in qgroup.
8218 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8219 struct btrfs_root *root,
8220 struct extent_buffer *root_eb,
8226 struct extent_buffer *eb = root_eb;
8227 struct btrfs_path *path = NULL;
8229 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8230 BUG_ON(root_eb == NULL);
8232 if (!root->fs_info->quota_enabled)
8235 if (!extent_buffer_uptodate(root_eb)) {
8236 ret = btrfs_read_buffer(root_eb, root_gen);
8241 if (root_level == 0) {
8242 ret = account_leaf_items(trans, root, root_eb);
8246 path = btrfs_alloc_path();
8251 * Walk down the tree. Missing extent blocks are filled in as
8252 * we go. Metadata is accounted every time we read a new
8255 * When we reach a leaf, we account for file extent items in it,
8256 * walk back up the tree (adjusting slot pointers as we go)
8257 * and restart the search process.
8259 extent_buffer_get(root_eb); /* For path */
8260 path->nodes[root_level] = root_eb;
8261 path->slots[root_level] = 0;
8262 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8265 while (level >= 0) {
8266 if (path->nodes[level] == NULL) {
8271 /* We need to get child blockptr/gen from
8272 * parent before we can read it. */
8273 eb = path->nodes[level + 1];
8274 parent_slot = path->slots[level + 1];
8275 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8276 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8278 eb = read_tree_block(root, child_bytenr, child_gen);
8282 } else if (!extent_buffer_uptodate(eb)) {
8283 free_extent_buffer(eb);
8288 path->nodes[level] = eb;
8289 path->slots[level] = 0;
8291 btrfs_tree_read_lock(eb);
8292 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8293 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8297 ret = account_leaf_items(trans, root, path->nodes[level]);
8301 /* Nonzero return here means we completed our search */
8302 ret = adjust_slots_upwards(root, path, root_level);
8306 /* Restart search with new slots */
8315 btrfs_free_path(path);
8321 * helper to process tree block while walking down the tree.
8323 * when wc->stage == UPDATE_BACKREF, this function updates
8324 * back refs for pointers in the block.
8326 * NOTE: return value 1 means we should stop walking down.
8328 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8329 struct btrfs_root *root,
8330 struct btrfs_path *path,
8331 struct walk_control *wc, int lookup_info)
8333 int level = wc->level;
8334 struct extent_buffer *eb = path->nodes[level];
8335 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8338 if (wc->stage == UPDATE_BACKREF &&
8339 btrfs_header_owner(eb) != root->root_key.objectid)
8343 * when reference count of tree block is 1, it won't increase
8344 * again. once full backref flag is set, we never clear it.
8347 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8348 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8349 BUG_ON(!path->locks[level]);
8350 ret = btrfs_lookup_extent_info(trans, root,
8351 eb->start, level, 1,
8354 BUG_ON(ret == -ENOMEM);
8357 BUG_ON(wc->refs[level] == 0);
8360 if (wc->stage == DROP_REFERENCE) {
8361 if (wc->refs[level] > 1)
8364 if (path->locks[level] && !wc->keep_locks) {
8365 btrfs_tree_unlock_rw(eb, path->locks[level]);
8366 path->locks[level] = 0;
8371 /* wc->stage == UPDATE_BACKREF */
8372 if (!(wc->flags[level] & flag)) {
8373 BUG_ON(!path->locks[level]);
8374 ret = btrfs_inc_ref(trans, root, eb, 1);
8375 BUG_ON(ret); /* -ENOMEM */
8376 ret = btrfs_dec_ref(trans, root, eb, 0);
8377 BUG_ON(ret); /* -ENOMEM */
8378 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8380 btrfs_header_level(eb), 0);
8381 BUG_ON(ret); /* -ENOMEM */
8382 wc->flags[level] |= flag;
8386 * the block is shared by multiple trees, so it's not good to
8387 * keep the tree lock
8389 if (path->locks[level] && level > 0) {
8390 btrfs_tree_unlock_rw(eb, path->locks[level]);
8391 path->locks[level] = 0;
8397 * helper to process tree block pointer.
8399 * when wc->stage == DROP_REFERENCE, this function checks
8400 * reference count of the block pointed to. if the block
8401 * is shared and we need update back refs for the subtree
8402 * rooted at the block, this function changes wc->stage to
8403 * UPDATE_BACKREF. if the block is shared and there is no
8404 * need to update back, this function drops the reference
8407 * NOTE: return value 1 means we should stop walking down.
8409 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8410 struct btrfs_root *root,
8411 struct btrfs_path *path,
8412 struct walk_control *wc, int *lookup_info)
8418 struct btrfs_key key;
8419 struct extent_buffer *next;
8420 int level = wc->level;
8423 bool need_account = false;
8425 generation = btrfs_node_ptr_generation(path->nodes[level],
8426 path->slots[level]);
8428 * if the lower level block was created before the snapshot
8429 * was created, we know there is no need to update back refs
8432 if (wc->stage == UPDATE_BACKREF &&
8433 generation <= root->root_key.offset) {
8438 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8439 blocksize = root->nodesize;
8441 next = btrfs_find_tree_block(root->fs_info, bytenr);
8443 next = btrfs_find_create_tree_block(root, bytenr);
8446 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8450 btrfs_tree_lock(next);
8451 btrfs_set_lock_blocking(next);
8453 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8454 &wc->refs[level - 1],
8455 &wc->flags[level - 1]);
8457 btrfs_tree_unlock(next);
8461 if (unlikely(wc->refs[level - 1] == 0)) {
8462 btrfs_err(root->fs_info, "Missing references.");
8467 if (wc->stage == DROP_REFERENCE) {
8468 if (wc->refs[level - 1] > 1) {
8469 need_account = true;
8471 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8474 if (!wc->update_ref ||
8475 generation <= root->root_key.offset)
8478 btrfs_node_key_to_cpu(path->nodes[level], &key,
8479 path->slots[level]);
8480 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8484 wc->stage = UPDATE_BACKREF;
8485 wc->shared_level = level - 1;
8489 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8493 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8494 btrfs_tree_unlock(next);
8495 free_extent_buffer(next);
8501 if (reada && level == 1)
8502 reada_walk_down(trans, root, wc, path);
8503 next = read_tree_block(root, bytenr, generation);
8505 return PTR_ERR(next);
8506 } else if (!extent_buffer_uptodate(next)) {
8507 free_extent_buffer(next);
8510 btrfs_tree_lock(next);
8511 btrfs_set_lock_blocking(next);
8515 BUG_ON(level != btrfs_header_level(next));
8516 path->nodes[level] = next;
8517 path->slots[level] = 0;
8518 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8524 wc->refs[level - 1] = 0;
8525 wc->flags[level - 1] = 0;
8526 if (wc->stage == DROP_REFERENCE) {
8527 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8528 parent = path->nodes[level]->start;
8530 BUG_ON(root->root_key.objectid !=
8531 btrfs_header_owner(path->nodes[level]));
8536 ret = account_shared_subtree(trans, root, next,
8537 generation, level - 1);
8539 btrfs_err_rl(root->fs_info,
8541 "%d accounting shared subtree. Quota "
8542 "is out of sync, rescan required.",
8546 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8547 root->root_key.objectid, level - 1, 0, 0);
8548 BUG_ON(ret); /* -ENOMEM */
8550 btrfs_tree_unlock(next);
8551 free_extent_buffer(next);
8557 * helper to process tree block while walking up the tree.
8559 * when wc->stage == DROP_REFERENCE, this function drops
8560 * reference count on the block.
8562 * when wc->stage == UPDATE_BACKREF, this function changes
8563 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8564 * to UPDATE_BACKREF previously while processing the block.
8566 * NOTE: return value 1 means we should stop walking up.
8568 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8569 struct btrfs_root *root,
8570 struct btrfs_path *path,
8571 struct walk_control *wc)
8574 int level = wc->level;
8575 struct extent_buffer *eb = path->nodes[level];
8578 if (wc->stage == UPDATE_BACKREF) {
8579 BUG_ON(wc->shared_level < level);
8580 if (level < wc->shared_level)
8583 ret = find_next_key(path, level + 1, &wc->update_progress);
8587 wc->stage = DROP_REFERENCE;
8588 wc->shared_level = -1;
8589 path->slots[level] = 0;
8592 * check reference count again if the block isn't locked.
8593 * we should start walking down the tree again if reference
8596 if (!path->locks[level]) {
8598 btrfs_tree_lock(eb);
8599 btrfs_set_lock_blocking(eb);
8600 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8602 ret = btrfs_lookup_extent_info(trans, root,
8603 eb->start, level, 1,
8607 btrfs_tree_unlock_rw(eb, path->locks[level]);
8608 path->locks[level] = 0;
8611 BUG_ON(wc->refs[level] == 0);
8612 if (wc->refs[level] == 1) {
8613 btrfs_tree_unlock_rw(eb, path->locks[level]);
8614 path->locks[level] = 0;
8620 /* wc->stage == DROP_REFERENCE */
8621 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8623 if (wc->refs[level] == 1) {
8625 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8626 ret = btrfs_dec_ref(trans, root, eb, 1);
8628 ret = btrfs_dec_ref(trans, root, eb, 0);
8629 BUG_ON(ret); /* -ENOMEM */
8630 ret = account_leaf_items(trans, root, eb);
8632 btrfs_err_rl(root->fs_info,
8634 "%d accounting leaf items. Quota "
8635 "is out of sync, rescan required.",
8639 /* make block locked assertion in clean_tree_block happy */
8640 if (!path->locks[level] &&
8641 btrfs_header_generation(eb) == trans->transid) {
8642 btrfs_tree_lock(eb);
8643 btrfs_set_lock_blocking(eb);
8644 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8646 clean_tree_block(trans, root->fs_info, eb);
8649 if (eb == root->node) {
8650 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8653 BUG_ON(root->root_key.objectid !=
8654 btrfs_header_owner(eb));
8656 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8657 parent = path->nodes[level + 1]->start;
8659 BUG_ON(root->root_key.objectid !=
8660 btrfs_header_owner(path->nodes[level + 1]));
8663 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8665 wc->refs[level] = 0;
8666 wc->flags[level] = 0;
8670 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8671 struct btrfs_root *root,
8672 struct btrfs_path *path,
8673 struct walk_control *wc)
8675 int level = wc->level;
8676 int lookup_info = 1;
8679 while (level >= 0) {
8680 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8687 if (path->slots[level] >=
8688 btrfs_header_nritems(path->nodes[level]))
8691 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8693 path->slots[level]++;
8702 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8703 struct btrfs_root *root,
8704 struct btrfs_path *path,
8705 struct walk_control *wc, int max_level)
8707 int level = wc->level;
8710 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8711 while (level < max_level && path->nodes[level]) {
8713 if (path->slots[level] + 1 <
8714 btrfs_header_nritems(path->nodes[level])) {
8715 path->slots[level]++;
8718 ret = walk_up_proc(trans, root, path, wc);
8722 if (path->locks[level]) {
8723 btrfs_tree_unlock_rw(path->nodes[level],
8724 path->locks[level]);
8725 path->locks[level] = 0;
8727 free_extent_buffer(path->nodes[level]);
8728 path->nodes[level] = NULL;
8736 * drop a subvolume tree.
8738 * this function traverses the tree freeing any blocks that only
8739 * referenced by the tree.
8741 * when a shared tree block is found. this function decreases its
8742 * reference count by one. if update_ref is true, this function
8743 * also make sure backrefs for the shared block and all lower level
8744 * blocks are properly updated.
8746 * If called with for_reloc == 0, may exit early with -EAGAIN
8748 int btrfs_drop_snapshot(struct btrfs_root *root,
8749 struct btrfs_block_rsv *block_rsv, int update_ref,
8752 struct btrfs_path *path;
8753 struct btrfs_trans_handle *trans;
8754 struct btrfs_root *tree_root = root->fs_info->tree_root;
8755 struct btrfs_root_item *root_item = &root->root_item;
8756 struct walk_control *wc;
8757 struct btrfs_key key;
8761 bool root_dropped = false;
8763 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8765 path = btrfs_alloc_path();
8771 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8773 btrfs_free_path(path);
8778 trans = btrfs_start_transaction(tree_root, 0);
8779 if (IS_ERR(trans)) {
8780 err = PTR_ERR(trans);
8785 trans->block_rsv = block_rsv;
8787 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8788 level = btrfs_header_level(root->node);
8789 path->nodes[level] = btrfs_lock_root_node(root);
8790 btrfs_set_lock_blocking(path->nodes[level]);
8791 path->slots[level] = 0;
8792 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8793 memset(&wc->update_progress, 0,
8794 sizeof(wc->update_progress));
8796 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8797 memcpy(&wc->update_progress, &key,
8798 sizeof(wc->update_progress));
8800 level = root_item->drop_level;
8802 path->lowest_level = level;
8803 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8804 path->lowest_level = 0;
8812 * unlock our path, this is safe because only this
8813 * function is allowed to delete this snapshot
8815 btrfs_unlock_up_safe(path, 0);
8817 level = btrfs_header_level(root->node);
8819 btrfs_tree_lock(path->nodes[level]);
8820 btrfs_set_lock_blocking(path->nodes[level]);
8821 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8823 ret = btrfs_lookup_extent_info(trans, root,
8824 path->nodes[level]->start,
8825 level, 1, &wc->refs[level],
8831 BUG_ON(wc->refs[level] == 0);
8833 if (level == root_item->drop_level)
8836 btrfs_tree_unlock(path->nodes[level]);
8837 path->locks[level] = 0;
8838 WARN_ON(wc->refs[level] != 1);
8844 wc->shared_level = -1;
8845 wc->stage = DROP_REFERENCE;
8846 wc->update_ref = update_ref;
8848 wc->for_reloc = for_reloc;
8849 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8853 ret = walk_down_tree(trans, root, path, wc);
8859 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8866 BUG_ON(wc->stage != DROP_REFERENCE);
8870 if (wc->stage == DROP_REFERENCE) {
8872 btrfs_node_key(path->nodes[level],
8873 &root_item->drop_progress,
8874 path->slots[level]);
8875 root_item->drop_level = level;
8878 BUG_ON(wc->level == 0);
8879 if (btrfs_should_end_transaction(trans, tree_root) ||
8880 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8881 ret = btrfs_update_root(trans, tree_root,
8885 btrfs_abort_transaction(trans, tree_root, ret);
8890 btrfs_end_transaction_throttle(trans, tree_root);
8891 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8892 pr_debug("BTRFS: drop snapshot early exit\n");
8897 trans = btrfs_start_transaction(tree_root, 0);
8898 if (IS_ERR(trans)) {
8899 err = PTR_ERR(trans);
8903 trans->block_rsv = block_rsv;
8906 btrfs_release_path(path);
8910 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8912 btrfs_abort_transaction(trans, tree_root, ret);
8916 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8917 ret = btrfs_find_root(tree_root, &root->root_key, path,
8920 btrfs_abort_transaction(trans, tree_root, ret);
8923 } else if (ret > 0) {
8924 /* if we fail to delete the orphan item this time
8925 * around, it'll get picked up the next time.
8927 * The most common failure here is just -ENOENT.
8929 btrfs_del_orphan_item(trans, tree_root,
8930 root->root_key.objectid);
8934 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8935 btrfs_add_dropped_root(trans, root);
8937 free_extent_buffer(root->node);
8938 free_extent_buffer(root->commit_root);
8939 btrfs_put_fs_root(root);
8941 root_dropped = true;
8943 btrfs_end_transaction_throttle(trans, tree_root);
8946 btrfs_free_path(path);
8949 * So if we need to stop dropping the snapshot for whatever reason we
8950 * need to make sure to add it back to the dead root list so that we
8951 * keep trying to do the work later. This also cleans up roots if we
8952 * don't have it in the radix (like when we recover after a power fail
8953 * or unmount) so we don't leak memory.
8955 if (!for_reloc && root_dropped == false)
8956 btrfs_add_dead_root(root);
8957 if (err && err != -EAGAIN)
8958 btrfs_std_error(root->fs_info, err, NULL);
8963 * drop subtree rooted at tree block 'node'.
8965 * NOTE: this function will unlock and release tree block 'node'
8966 * only used by relocation code
8968 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8969 struct btrfs_root *root,
8970 struct extent_buffer *node,
8971 struct extent_buffer *parent)
8973 struct btrfs_path *path;
8974 struct walk_control *wc;
8980 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8982 path = btrfs_alloc_path();
8986 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8988 btrfs_free_path(path);
8992 btrfs_assert_tree_locked(parent);
8993 parent_level = btrfs_header_level(parent);
8994 extent_buffer_get(parent);
8995 path->nodes[parent_level] = parent;
8996 path->slots[parent_level] = btrfs_header_nritems(parent);
8998 btrfs_assert_tree_locked(node);
8999 level = btrfs_header_level(node);
9000 path->nodes[level] = node;
9001 path->slots[level] = 0;
9002 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
9004 wc->refs[parent_level] = 1;
9005 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
9007 wc->shared_level = -1;
9008 wc->stage = DROP_REFERENCE;
9012 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
9015 wret = walk_down_tree(trans, root, path, wc);
9021 wret = walk_up_tree(trans, root, path, wc, parent_level);
9029 btrfs_free_path(path);
9033 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
9039 * if restripe for this chunk_type is on pick target profile and
9040 * return, otherwise do the usual balance
9042 stripped = get_restripe_target(root->fs_info, flags);
9044 return extended_to_chunk(stripped);
9046 num_devices = root->fs_info->fs_devices->rw_devices;
9048 stripped = BTRFS_BLOCK_GROUP_RAID0 |
9049 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
9050 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
9052 if (num_devices == 1) {
9053 stripped |= BTRFS_BLOCK_GROUP_DUP;
9054 stripped = flags & ~stripped;
9056 /* turn raid0 into single device chunks */
9057 if (flags & BTRFS_BLOCK_GROUP_RAID0)
9060 /* turn mirroring into duplication */
9061 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
9062 BTRFS_BLOCK_GROUP_RAID10))
9063 return stripped | BTRFS_BLOCK_GROUP_DUP;
9065 /* they already had raid on here, just return */
9066 if (flags & stripped)
9069 stripped |= BTRFS_BLOCK_GROUP_DUP;
9070 stripped = flags & ~stripped;
9072 /* switch duplicated blocks with raid1 */
9073 if (flags & BTRFS_BLOCK_GROUP_DUP)
9074 return stripped | BTRFS_BLOCK_GROUP_RAID1;
9076 /* this is drive concat, leave it alone */
9082 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
9084 struct btrfs_space_info *sinfo = cache->space_info;
9086 u64 min_allocable_bytes;
9090 * We need some metadata space and system metadata space for
9091 * allocating chunks in some corner cases until we force to set
9092 * it to be readonly.
9095 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
9097 min_allocable_bytes = 1 * 1024 * 1024;
9099 min_allocable_bytes = 0;
9101 spin_lock(&sinfo->lock);
9102 spin_lock(&cache->lock);
9110 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
9111 cache->bytes_super - btrfs_block_group_used(&cache->item);
9113 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
9114 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
9115 min_allocable_bytes <= sinfo->total_bytes) {
9116 sinfo->bytes_readonly += num_bytes;
9118 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
9122 spin_unlock(&cache->lock);
9123 spin_unlock(&sinfo->lock);
9127 int btrfs_inc_block_group_ro(struct btrfs_root *root,
9128 struct btrfs_block_group_cache *cache)
9131 struct btrfs_trans_handle *trans;
9136 trans = btrfs_join_transaction(root);
9138 return PTR_ERR(trans);
9141 * we're not allowed to set block groups readonly after the dirty
9142 * block groups cache has started writing. If it already started,
9143 * back off and let this transaction commit
9145 mutex_lock(&root->fs_info->ro_block_group_mutex);
9146 if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) {
9147 u64 transid = trans->transid;
9149 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9150 btrfs_end_transaction(trans, root);
9152 ret = btrfs_wait_for_commit(root, transid);
9159 * if we are changing raid levels, try to allocate a corresponding
9160 * block group with the new raid level.
9162 alloc_flags = update_block_group_flags(root, cache->flags);
9163 if (alloc_flags != cache->flags) {
9164 ret = do_chunk_alloc(trans, root, alloc_flags,
9167 * ENOSPC is allowed here, we may have enough space
9168 * already allocated at the new raid level to
9177 ret = inc_block_group_ro(cache, 0);
9180 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
9181 ret = do_chunk_alloc(trans, root, alloc_flags,
9185 ret = inc_block_group_ro(cache, 0);
9187 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
9188 alloc_flags = update_block_group_flags(root, cache->flags);
9189 lock_chunks(root->fs_info->chunk_root);
9190 check_system_chunk(trans, root, alloc_flags);
9191 unlock_chunks(root->fs_info->chunk_root);
9193 mutex_unlock(&root->fs_info->ro_block_group_mutex);
9195 btrfs_end_transaction(trans, root);
9199 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
9200 struct btrfs_root *root, u64 type)
9202 u64 alloc_flags = get_alloc_profile(root, type);
9203 return do_chunk_alloc(trans, root, alloc_flags,
9208 * helper to account the unused space of all the readonly block group in the
9209 * space_info. takes mirrors into account.
9211 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9213 struct btrfs_block_group_cache *block_group;
9217 /* It's df, we don't care if it's racey */
9218 if (list_empty(&sinfo->ro_bgs))
9221 spin_lock(&sinfo->lock);
9222 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9223 spin_lock(&block_group->lock);
9225 if (!block_group->ro) {
9226 spin_unlock(&block_group->lock);
9230 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9231 BTRFS_BLOCK_GROUP_RAID10 |
9232 BTRFS_BLOCK_GROUP_DUP))
9237 free_bytes += (block_group->key.offset -
9238 btrfs_block_group_used(&block_group->item)) *
9241 spin_unlock(&block_group->lock);
9243 spin_unlock(&sinfo->lock);
9248 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9249 struct btrfs_block_group_cache *cache)
9251 struct btrfs_space_info *sinfo = cache->space_info;
9256 spin_lock(&sinfo->lock);
9257 spin_lock(&cache->lock);
9259 num_bytes = cache->key.offset - cache->reserved -
9260 cache->pinned - cache->bytes_super -
9261 btrfs_block_group_used(&cache->item);
9262 sinfo->bytes_readonly -= num_bytes;
9263 list_del_init(&cache->ro_list);
9265 spin_unlock(&cache->lock);
9266 spin_unlock(&sinfo->lock);
9270 * checks to see if its even possible to relocate this block group.
9272 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9273 * ok to go ahead and try.
9275 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9277 struct btrfs_block_group_cache *block_group;
9278 struct btrfs_space_info *space_info;
9279 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9280 struct btrfs_device *device;
9281 struct btrfs_trans_handle *trans;
9290 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9292 /* odd, couldn't find the block group, leave it alone */
9296 min_free = btrfs_block_group_used(&block_group->item);
9298 /* no bytes used, we're good */
9302 space_info = block_group->space_info;
9303 spin_lock(&space_info->lock);
9305 full = space_info->full;
9308 * if this is the last block group we have in this space, we can't
9309 * relocate it unless we're able to allocate a new chunk below.
9311 * Otherwise, we need to make sure we have room in the space to handle
9312 * all of the extents from this block group. If we can, we're good
9314 if ((space_info->total_bytes != block_group->key.offset) &&
9315 (space_info->bytes_used + space_info->bytes_reserved +
9316 space_info->bytes_pinned + space_info->bytes_readonly +
9317 min_free < space_info->total_bytes)) {
9318 spin_unlock(&space_info->lock);
9321 spin_unlock(&space_info->lock);
9324 * ok we don't have enough space, but maybe we have free space on our
9325 * devices to allocate new chunks for relocation, so loop through our
9326 * alloc devices and guess if we have enough space. if this block
9327 * group is going to be restriped, run checks against the target
9328 * profile instead of the current one.
9340 target = get_restripe_target(root->fs_info, block_group->flags);
9342 index = __get_raid_index(extended_to_chunk(target));
9345 * this is just a balance, so if we were marked as full
9346 * we know there is no space for a new chunk
9351 index = get_block_group_index(block_group);
9354 if (index == BTRFS_RAID_RAID10) {
9358 } else if (index == BTRFS_RAID_RAID1) {
9360 } else if (index == BTRFS_RAID_DUP) {
9363 } else if (index == BTRFS_RAID_RAID0) {
9364 dev_min = fs_devices->rw_devices;
9365 min_free = div64_u64(min_free, dev_min);
9368 /* We need to do this so that we can look at pending chunks */
9369 trans = btrfs_join_transaction(root);
9370 if (IS_ERR(trans)) {
9371 ret = PTR_ERR(trans);
9375 mutex_lock(&root->fs_info->chunk_mutex);
9376 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9380 * check to make sure we can actually find a chunk with enough
9381 * space to fit our block group in.
9383 if (device->total_bytes > device->bytes_used + min_free &&
9384 !device->is_tgtdev_for_dev_replace) {
9385 ret = find_free_dev_extent(trans, device, min_free,
9390 if (dev_nr >= dev_min)
9396 mutex_unlock(&root->fs_info->chunk_mutex);
9397 btrfs_end_transaction(trans, root);
9399 btrfs_put_block_group(block_group);
9403 static int find_first_block_group(struct btrfs_root *root,
9404 struct btrfs_path *path, struct btrfs_key *key)
9407 struct btrfs_key found_key;
9408 struct extent_buffer *leaf;
9411 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9416 slot = path->slots[0];
9417 leaf = path->nodes[0];
9418 if (slot >= btrfs_header_nritems(leaf)) {
9419 ret = btrfs_next_leaf(root, path);
9426 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9428 if (found_key.objectid >= key->objectid &&
9429 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9439 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9441 struct btrfs_block_group_cache *block_group;
9445 struct inode *inode;
9447 block_group = btrfs_lookup_first_block_group(info, last);
9448 while (block_group) {
9449 spin_lock(&block_group->lock);
9450 if (block_group->iref)
9452 spin_unlock(&block_group->lock);
9453 block_group = next_block_group(info->tree_root,
9463 inode = block_group->inode;
9464 block_group->iref = 0;
9465 block_group->inode = NULL;
9466 spin_unlock(&block_group->lock);
9468 last = block_group->key.objectid + block_group->key.offset;
9469 btrfs_put_block_group(block_group);
9473 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9475 struct btrfs_block_group_cache *block_group;
9476 struct btrfs_space_info *space_info;
9477 struct btrfs_caching_control *caching_ctl;
9480 down_write(&info->commit_root_sem);
9481 while (!list_empty(&info->caching_block_groups)) {
9482 caching_ctl = list_entry(info->caching_block_groups.next,
9483 struct btrfs_caching_control, list);
9484 list_del(&caching_ctl->list);
9485 put_caching_control(caching_ctl);
9487 up_write(&info->commit_root_sem);
9489 spin_lock(&info->unused_bgs_lock);
9490 while (!list_empty(&info->unused_bgs)) {
9491 block_group = list_first_entry(&info->unused_bgs,
9492 struct btrfs_block_group_cache,
9494 list_del_init(&block_group->bg_list);
9495 btrfs_put_block_group(block_group);
9497 spin_unlock(&info->unused_bgs_lock);
9499 spin_lock(&info->block_group_cache_lock);
9500 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9501 block_group = rb_entry(n, struct btrfs_block_group_cache,
9503 rb_erase(&block_group->cache_node,
9504 &info->block_group_cache_tree);
9505 RB_CLEAR_NODE(&block_group->cache_node);
9506 spin_unlock(&info->block_group_cache_lock);
9508 down_write(&block_group->space_info->groups_sem);
9509 list_del(&block_group->list);
9510 up_write(&block_group->space_info->groups_sem);
9512 if (block_group->cached == BTRFS_CACHE_STARTED)
9513 wait_block_group_cache_done(block_group);
9516 * We haven't cached this block group, which means we could
9517 * possibly have excluded extents on this block group.
9519 if (block_group->cached == BTRFS_CACHE_NO ||
9520 block_group->cached == BTRFS_CACHE_ERROR)
9521 free_excluded_extents(info->extent_root, block_group);
9523 btrfs_remove_free_space_cache(block_group);
9524 btrfs_put_block_group(block_group);
9526 spin_lock(&info->block_group_cache_lock);
9528 spin_unlock(&info->block_group_cache_lock);
9530 /* now that all the block groups are freed, go through and
9531 * free all the space_info structs. This is only called during
9532 * the final stages of unmount, and so we know nobody is
9533 * using them. We call synchronize_rcu() once before we start,
9534 * just to be on the safe side.
9538 release_global_block_rsv(info);
9540 while (!list_empty(&info->space_info)) {
9543 space_info = list_entry(info->space_info.next,
9544 struct btrfs_space_info,
9546 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9547 if (WARN_ON(space_info->bytes_pinned > 0 ||
9548 space_info->bytes_reserved > 0 ||
9549 space_info->bytes_may_use > 0)) {
9550 dump_space_info(space_info, 0, 0);
9553 list_del(&space_info->list);
9554 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9555 struct kobject *kobj;
9556 kobj = space_info->block_group_kobjs[i];
9557 space_info->block_group_kobjs[i] = NULL;
9563 kobject_del(&space_info->kobj);
9564 kobject_put(&space_info->kobj);
9569 static void __link_block_group(struct btrfs_space_info *space_info,
9570 struct btrfs_block_group_cache *cache)
9572 int index = get_block_group_index(cache);
9575 down_write(&space_info->groups_sem);
9576 if (list_empty(&space_info->block_groups[index]))
9578 list_add_tail(&cache->list, &space_info->block_groups[index]);
9579 up_write(&space_info->groups_sem);
9582 struct raid_kobject *rkobj;
9585 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9588 rkobj->raid_type = index;
9589 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9590 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9591 "%s", get_raid_name(index));
9593 kobject_put(&rkobj->kobj);
9596 space_info->block_group_kobjs[index] = &rkobj->kobj;
9601 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9604 static struct btrfs_block_group_cache *
9605 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9607 struct btrfs_block_group_cache *cache;
9609 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9613 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9615 if (!cache->free_space_ctl) {
9620 cache->key.objectid = start;
9621 cache->key.offset = size;
9622 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9624 cache->sectorsize = root->sectorsize;
9625 cache->fs_info = root->fs_info;
9626 cache->full_stripe_len = btrfs_full_stripe_len(root,
9627 &root->fs_info->mapping_tree,
9629 atomic_set(&cache->count, 1);
9630 spin_lock_init(&cache->lock);
9631 init_rwsem(&cache->data_rwsem);
9632 INIT_LIST_HEAD(&cache->list);
9633 INIT_LIST_HEAD(&cache->cluster_list);
9634 INIT_LIST_HEAD(&cache->bg_list);
9635 INIT_LIST_HEAD(&cache->ro_list);
9636 INIT_LIST_HEAD(&cache->dirty_list);
9637 INIT_LIST_HEAD(&cache->io_list);
9638 btrfs_init_free_space_ctl(cache);
9639 atomic_set(&cache->trimming, 0);
9644 int btrfs_read_block_groups(struct btrfs_root *root)
9646 struct btrfs_path *path;
9648 struct btrfs_block_group_cache *cache;
9649 struct btrfs_fs_info *info = root->fs_info;
9650 struct btrfs_space_info *space_info;
9651 struct btrfs_key key;
9652 struct btrfs_key found_key;
9653 struct extent_buffer *leaf;
9657 root = info->extent_root;
9660 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9661 path = btrfs_alloc_path();
9666 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9667 if (btrfs_test_opt(root, SPACE_CACHE) &&
9668 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9670 if (btrfs_test_opt(root, CLEAR_CACHE))
9674 ret = find_first_block_group(root, path, &key);
9680 leaf = path->nodes[0];
9681 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9683 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9692 * When we mount with old space cache, we need to
9693 * set BTRFS_DC_CLEAR and set dirty flag.
9695 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9696 * truncate the old free space cache inode and
9698 * b) Setting 'dirty flag' makes sure that we flush
9699 * the new space cache info onto disk.
9701 if (btrfs_test_opt(root, SPACE_CACHE))
9702 cache->disk_cache_state = BTRFS_DC_CLEAR;
9705 read_extent_buffer(leaf, &cache->item,
9706 btrfs_item_ptr_offset(leaf, path->slots[0]),
9707 sizeof(cache->item));
9708 cache->flags = btrfs_block_group_flags(&cache->item);
9710 key.objectid = found_key.objectid + found_key.offset;
9711 btrfs_release_path(path);
9714 * We need to exclude the super stripes now so that the space
9715 * info has super bytes accounted for, otherwise we'll think
9716 * we have more space than we actually do.
9718 ret = exclude_super_stripes(root, cache);
9721 * We may have excluded something, so call this just in
9724 free_excluded_extents(root, cache);
9725 btrfs_put_block_group(cache);
9730 * check for two cases, either we are full, and therefore
9731 * don't need to bother with the caching work since we won't
9732 * find any space, or we are empty, and we can just add all
9733 * the space in and be done with it. This saves us _alot_ of
9734 * time, particularly in the full case.
9736 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9737 cache->last_byte_to_unpin = (u64)-1;
9738 cache->cached = BTRFS_CACHE_FINISHED;
9739 free_excluded_extents(root, cache);
9740 } else if (btrfs_block_group_used(&cache->item) == 0) {
9741 cache->last_byte_to_unpin = (u64)-1;
9742 cache->cached = BTRFS_CACHE_FINISHED;
9743 add_new_free_space(cache, root->fs_info,
9745 found_key.objectid +
9747 free_excluded_extents(root, cache);
9750 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9752 btrfs_remove_free_space_cache(cache);
9753 btrfs_put_block_group(cache);
9757 ret = update_space_info(info, cache->flags, found_key.offset,
9758 btrfs_block_group_used(&cache->item),
9761 btrfs_remove_free_space_cache(cache);
9762 spin_lock(&info->block_group_cache_lock);
9763 rb_erase(&cache->cache_node,
9764 &info->block_group_cache_tree);
9765 RB_CLEAR_NODE(&cache->cache_node);
9766 spin_unlock(&info->block_group_cache_lock);
9767 btrfs_put_block_group(cache);
9771 cache->space_info = space_info;
9772 spin_lock(&cache->space_info->lock);
9773 cache->space_info->bytes_readonly += cache->bytes_super;
9774 spin_unlock(&cache->space_info->lock);
9776 __link_block_group(space_info, cache);
9778 set_avail_alloc_bits(root->fs_info, cache->flags);
9779 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9780 inc_block_group_ro(cache, 1);
9781 } else if (btrfs_block_group_used(&cache->item) == 0) {
9782 spin_lock(&info->unused_bgs_lock);
9783 /* Should always be true but just in case. */
9784 if (list_empty(&cache->bg_list)) {
9785 btrfs_get_block_group(cache);
9786 list_add_tail(&cache->bg_list,
9789 spin_unlock(&info->unused_bgs_lock);
9793 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9794 if (!(get_alloc_profile(root, space_info->flags) &
9795 (BTRFS_BLOCK_GROUP_RAID10 |
9796 BTRFS_BLOCK_GROUP_RAID1 |
9797 BTRFS_BLOCK_GROUP_RAID5 |
9798 BTRFS_BLOCK_GROUP_RAID6 |
9799 BTRFS_BLOCK_GROUP_DUP)))
9802 * avoid allocating from un-mirrored block group if there are
9803 * mirrored block groups.
9805 list_for_each_entry(cache,
9806 &space_info->block_groups[BTRFS_RAID_RAID0],
9808 inc_block_group_ro(cache, 1);
9809 list_for_each_entry(cache,
9810 &space_info->block_groups[BTRFS_RAID_SINGLE],
9812 inc_block_group_ro(cache, 1);
9815 init_global_block_rsv(info);
9818 btrfs_free_path(path);
9822 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9823 struct btrfs_root *root)
9825 struct btrfs_block_group_cache *block_group, *tmp;
9826 struct btrfs_root *extent_root = root->fs_info->extent_root;
9827 struct btrfs_block_group_item item;
9828 struct btrfs_key key;
9830 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9832 trans->can_flush_pending_bgs = false;
9833 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9837 spin_lock(&block_group->lock);
9838 memcpy(&item, &block_group->item, sizeof(item));
9839 memcpy(&key, &block_group->key, sizeof(key));
9840 spin_unlock(&block_group->lock);
9842 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9845 btrfs_abort_transaction(trans, extent_root, ret);
9846 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9847 key.objectid, key.offset);
9849 btrfs_abort_transaction(trans, extent_root, ret);
9851 list_del_init(&block_group->bg_list);
9853 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9856 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9857 struct btrfs_root *root, u64 bytes_used,
9858 u64 type, u64 chunk_objectid, u64 chunk_offset,
9862 struct btrfs_root *extent_root;
9863 struct btrfs_block_group_cache *cache;
9865 extent_root = root->fs_info->extent_root;
9867 btrfs_set_log_full_commit(root->fs_info, trans);
9869 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9873 btrfs_set_block_group_used(&cache->item, bytes_used);
9874 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9875 btrfs_set_block_group_flags(&cache->item, type);
9877 cache->flags = type;
9878 cache->last_byte_to_unpin = (u64)-1;
9879 cache->cached = BTRFS_CACHE_FINISHED;
9880 ret = exclude_super_stripes(root, cache);
9883 * We may have excluded something, so call this just in
9886 free_excluded_extents(root, cache);
9887 btrfs_put_block_group(cache);
9891 add_new_free_space(cache, root->fs_info, chunk_offset,
9892 chunk_offset + size);
9894 free_excluded_extents(root, cache);
9896 #ifdef CONFIG_BTRFS_DEBUG
9897 if (btrfs_should_fragment_free_space(root, cache)) {
9898 u64 new_bytes_used = size - bytes_used;
9900 bytes_used += new_bytes_used >> 1;
9901 fragment_free_space(root, cache);
9905 * Call to ensure the corresponding space_info object is created and
9906 * assigned to our block group, but don't update its counters just yet.
9907 * We want our bg to be added to the rbtree with its ->space_info set.
9909 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9910 &cache->space_info);
9912 btrfs_remove_free_space_cache(cache);
9913 btrfs_put_block_group(cache);
9917 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9919 btrfs_remove_free_space_cache(cache);
9920 btrfs_put_block_group(cache);
9925 * Now that our block group has its ->space_info set and is inserted in
9926 * the rbtree, update the space info's counters.
9928 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9929 &cache->space_info);
9931 btrfs_remove_free_space_cache(cache);
9932 spin_lock(&root->fs_info->block_group_cache_lock);
9933 rb_erase(&cache->cache_node,
9934 &root->fs_info->block_group_cache_tree);
9935 RB_CLEAR_NODE(&cache->cache_node);
9936 spin_unlock(&root->fs_info->block_group_cache_lock);
9937 btrfs_put_block_group(cache);
9940 update_global_block_rsv(root->fs_info);
9942 spin_lock(&cache->space_info->lock);
9943 cache->space_info->bytes_readonly += cache->bytes_super;
9944 spin_unlock(&cache->space_info->lock);
9946 __link_block_group(cache->space_info, cache);
9948 list_add_tail(&cache->bg_list, &trans->new_bgs);
9950 set_avail_alloc_bits(extent_root->fs_info, type);
9955 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9957 u64 extra_flags = chunk_to_extended(flags) &
9958 BTRFS_EXTENDED_PROFILE_MASK;
9960 write_seqlock(&fs_info->profiles_lock);
9961 if (flags & BTRFS_BLOCK_GROUP_DATA)
9962 fs_info->avail_data_alloc_bits &= ~extra_flags;
9963 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9964 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9965 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9966 fs_info->avail_system_alloc_bits &= ~extra_flags;
9967 write_sequnlock(&fs_info->profiles_lock);
9970 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9971 struct btrfs_root *root, u64 group_start,
9972 struct extent_map *em)
9974 struct btrfs_path *path;
9975 struct btrfs_block_group_cache *block_group;
9976 struct btrfs_free_cluster *cluster;
9977 struct btrfs_root *tree_root = root->fs_info->tree_root;
9978 struct btrfs_key key;
9979 struct inode *inode;
9980 struct kobject *kobj = NULL;
9984 struct btrfs_caching_control *caching_ctl = NULL;
9987 root = root->fs_info->extent_root;
9989 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9990 BUG_ON(!block_group);
9991 BUG_ON(!block_group->ro);
9994 * Free the reserved super bytes from this block group before
9997 free_excluded_extents(root, block_group);
9999 memcpy(&key, &block_group->key, sizeof(key));
10000 index = get_block_group_index(block_group);
10001 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
10002 BTRFS_BLOCK_GROUP_RAID1 |
10003 BTRFS_BLOCK_GROUP_RAID10))
10008 /* make sure this block group isn't part of an allocation cluster */
10009 cluster = &root->fs_info->data_alloc_cluster;
10010 spin_lock(&cluster->refill_lock);
10011 btrfs_return_cluster_to_free_space(block_group, cluster);
10012 spin_unlock(&cluster->refill_lock);
10015 * make sure this block group isn't part of a metadata
10016 * allocation cluster
10018 cluster = &root->fs_info->meta_alloc_cluster;
10019 spin_lock(&cluster->refill_lock);
10020 btrfs_return_cluster_to_free_space(block_group, cluster);
10021 spin_unlock(&cluster->refill_lock);
10023 path = btrfs_alloc_path();
10030 * get the inode first so any iput calls done for the io_list
10031 * aren't the final iput (no unlinks allowed now)
10033 inode = lookup_free_space_inode(tree_root, block_group, path);
10035 mutex_lock(&trans->transaction->cache_write_mutex);
10037 * make sure our free spache cache IO is done before remove the
10040 spin_lock(&trans->transaction->dirty_bgs_lock);
10041 if (!list_empty(&block_group->io_list)) {
10042 list_del_init(&block_group->io_list);
10044 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
10046 spin_unlock(&trans->transaction->dirty_bgs_lock);
10047 btrfs_wait_cache_io(root, trans, block_group,
10048 &block_group->io_ctl, path,
10049 block_group->key.objectid);
10050 btrfs_put_block_group(block_group);
10051 spin_lock(&trans->transaction->dirty_bgs_lock);
10054 if (!list_empty(&block_group->dirty_list)) {
10055 list_del_init(&block_group->dirty_list);
10056 btrfs_put_block_group(block_group);
10058 spin_unlock(&trans->transaction->dirty_bgs_lock);
10059 mutex_unlock(&trans->transaction->cache_write_mutex);
10061 if (!IS_ERR(inode)) {
10062 ret = btrfs_orphan_add(trans, inode);
10064 btrfs_add_delayed_iput(inode);
10067 clear_nlink(inode);
10068 /* One for the block groups ref */
10069 spin_lock(&block_group->lock);
10070 if (block_group->iref) {
10071 block_group->iref = 0;
10072 block_group->inode = NULL;
10073 spin_unlock(&block_group->lock);
10076 spin_unlock(&block_group->lock);
10078 /* One for our lookup ref */
10079 btrfs_add_delayed_iput(inode);
10082 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
10083 key.offset = block_group->key.objectid;
10086 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
10090 btrfs_release_path(path);
10092 ret = btrfs_del_item(trans, tree_root, path);
10095 btrfs_release_path(path);
10098 spin_lock(&root->fs_info->block_group_cache_lock);
10099 rb_erase(&block_group->cache_node,
10100 &root->fs_info->block_group_cache_tree);
10101 RB_CLEAR_NODE(&block_group->cache_node);
10103 if (root->fs_info->first_logical_byte == block_group->key.objectid)
10104 root->fs_info->first_logical_byte = (u64)-1;
10105 spin_unlock(&root->fs_info->block_group_cache_lock);
10107 down_write(&block_group->space_info->groups_sem);
10109 * we must use list_del_init so people can check to see if they
10110 * are still on the list after taking the semaphore
10112 list_del_init(&block_group->list);
10113 if (list_empty(&block_group->space_info->block_groups[index])) {
10114 kobj = block_group->space_info->block_group_kobjs[index];
10115 block_group->space_info->block_group_kobjs[index] = NULL;
10116 clear_avail_alloc_bits(root->fs_info, block_group->flags);
10118 up_write(&block_group->space_info->groups_sem);
10124 if (block_group->has_caching_ctl)
10125 caching_ctl = get_caching_control(block_group);
10126 if (block_group->cached == BTRFS_CACHE_STARTED)
10127 wait_block_group_cache_done(block_group);
10128 if (block_group->has_caching_ctl) {
10129 down_write(&root->fs_info->commit_root_sem);
10130 if (!caching_ctl) {
10131 struct btrfs_caching_control *ctl;
10133 list_for_each_entry(ctl,
10134 &root->fs_info->caching_block_groups, list)
10135 if (ctl->block_group == block_group) {
10137 atomic_inc(&caching_ctl->count);
10142 list_del_init(&caching_ctl->list);
10143 up_write(&root->fs_info->commit_root_sem);
10145 /* Once for the caching bgs list and once for us. */
10146 put_caching_control(caching_ctl);
10147 put_caching_control(caching_ctl);
10151 spin_lock(&trans->transaction->dirty_bgs_lock);
10152 if (!list_empty(&block_group->dirty_list)) {
10155 if (!list_empty(&block_group->io_list)) {
10158 spin_unlock(&trans->transaction->dirty_bgs_lock);
10159 btrfs_remove_free_space_cache(block_group);
10161 spin_lock(&block_group->space_info->lock);
10162 list_del_init(&block_group->ro_list);
10164 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
10165 WARN_ON(block_group->space_info->total_bytes
10166 < block_group->key.offset);
10167 WARN_ON(block_group->space_info->bytes_readonly
10168 < block_group->key.offset);
10169 WARN_ON(block_group->space_info->disk_total
10170 < block_group->key.offset * factor);
10172 block_group->space_info->total_bytes -= block_group->key.offset;
10173 block_group->space_info->bytes_readonly -= block_group->key.offset;
10174 block_group->space_info->disk_total -= block_group->key.offset * factor;
10176 spin_unlock(&block_group->space_info->lock);
10178 memcpy(&key, &block_group->key, sizeof(key));
10181 if (!list_empty(&em->list)) {
10182 /* We're in the transaction->pending_chunks list. */
10183 free_extent_map(em);
10185 spin_lock(&block_group->lock);
10186 block_group->removed = 1;
10188 * At this point trimming can't start on this block group, because we
10189 * removed the block group from the tree fs_info->block_group_cache_tree
10190 * so no one can't find it anymore and even if someone already got this
10191 * block group before we removed it from the rbtree, they have already
10192 * incremented block_group->trimming - if they didn't, they won't find
10193 * any free space entries because we already removed them all when we
10194 * called btrfs_remove_free_space_cache().
10196 * And we must not remove the extent map from the fs_info->mapping_tree
10197 * to prevent the same logical address range and physical device space
10198 * ranges from being reused for a new block group. This is because our
10199 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
10200 * completely transactionless, so while it is trimming a range the
10201 * currently running transaction might finish and a new one start,
10202 * allowing for new block groups to be created that can reuse the same
10203 * physical device locations unless we take this special care.
10205 * There may also be an implicit trim operation if the file system
10206 * is mounted with -odiscard. The same protections must remain
10207 * in place until the extents have been discarded completely when
10208 * the transaction commit has completed.
10210 remove_em = (atomic_read(&block_group->trimming) == 0);
10212 * Make sure a trimmer task always sees the em in the pinned_chunks list
10213 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10214 * before checking block_group->removed).
10218 * Our em might be in trans->transaction->pending_chunks which
10219 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10220 * and so is the fs_info->pinned_chunks list.
10222 * So at this point we must be holding the chunk_mutex to avoid
10223 * any races with chunk allocation (more specifically at
10224 * volumes.c:contains_pending_extent()), to ensure it always
10225 * sees the em, either in the pending_chunks list or in the
10226 * pinned_chunks list.
10228 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10230 spin_unlock(&block_group->lock);
10233 struct extent_map_tree *em_tree;
10235 em_tree = &root->fs_info->mapping_tree.map_tree;
10236 write_lock(&em_tree->lock);
10238 * The em might be in the pending_chunks list, so make sure the
10239 * chunk mutex is locked, since remove_extent_mapping() will
10240 * delete us from that list.
10242 remove_extent_mapping(em_tree, em);
10243 write_unlock(&em_tree->lock);
10244 /* once for the tree */
10245 free_extent_map(em);
10248 unlock_chunks(root);
10250 btrfs_put_block_group(block_group);
10251 btrfs_put_block_group(block_group);
10253 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10259 ret = btrfs_del_item(trans, root, path);
10261 btrfs_free_path(path);
10266 * Process the unused_bgs list and remove any that don't have any allocated
10267 * space inside of them.
10269 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10271 struct btrfs_block_group_cache *block_group;
10272 struct btrfs_space_info *space_info;
10273 struct btrfs_root *root = fs_info->extent_root;
10274 struct btrfs_trans_handle *trans;
10277 if (!fs_info->open)
10280 spin_lock(&fs_info->unused_bgs_lock);
10281 while (!list_empty(&fs_info->unused_bgs)) {
10285 block_group = list_first_entry(&fs_info->unused_bgs,
10286 struct btrfs_block_group_cache,
10288 space_info = block_group->space_info;
10289 list_del_init(&block_group->bg_list);
10290 if (ret || btrfs_mixed_space_info(space_info)) {
10291 btrfs_put_block_group(block_group);
10294 spin_unlock(&fs_info->unused_bgs_lock);
10296 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
10298 /* Don't want to race with allocators so take the groups_sem */
10299 down_write(&space_info->groups_sem);
10300 spin_lock(&block_group->lock);
10301 if (block_group->reserved ||
10302 btrfs_block_group_used(&block_group->item) ||
10305 * We want to bail if we made new allocations or have
10306 * outstanding allocations in this block group. We do
10307 * the ro check in case balance is currently acting on
10308 * this block group.
10310 spin_unlock(&block_group->lock);
10311 up_write(&space_info->groups_sem);
10314 spin_unlock(&block_group->lock);
10316 /* We don't want to force the issue, only flip if it's ok. */
10317 ret = inc_block_group_ro(block_group, 0);
10318 up_write(&space_info->groups_sem);
10325 * Want to do this before we do anything else so we can recover
10326 * properly if we fail to join the transaction.
10328 /* 1 for btrfs_orphan_reserve_metadata() */
10329 trans = btrfs_start_transaction(root, 1);
10330 if (IS_ERR(trans)) {
10331 btrfs_dec_block_group_ro(root, block_group);
10332 ret = PTR_ERR(trans);
10337 * We could have pending pinned extents for this block group,
10338 * just delete them, we don't care about them anymore.
10340 start = block_group->key.objectid;
10341 end = start + block_group->key.offset - 1;
10343 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10344 * btrfs_finish_extent_commit(). If we are at transaction N,
10345 * another task might be running finish_extent_commit() for the
10346 * previous transaction N - 1, and have seen a range belonging
10347 * to the block group in freed_extents[] before we were able to
10348 * clear the whole block group range from freed_extents[]. This
10349 * means that task can lookup for the block group after we
10350 * unpinned it from freed_extents[] and removed it, leading to
10351 * a BUG_ON() at btrfs_unpin_extent_range().
10353 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10354 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10355 EXTENT_DIRTY, GFP_NOFS);
10357 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10358 btrfs_dec_block_group_ro(root, block_group);
10361 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10362 EXTENT_DIRTY, GFP_NOFS);
10364 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10365 btrfs_dec_block_group_ro(root, block_group);
10368 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10370 /* Reset pinned so btrfs_put_block_group doesn't complain */
10371 spin_lock(&space_info->lock);
10372 spin_lock(&block_group->lock);
10374 space_info->bytes_pinned -= block_group->pinned;
10375 space_info->bytes_readonly += block_group->pinned;
10376 percpu_counter_add(&space_info->total_bytes_pinned,
10377 -block_group->pinned);
10378 block_group->pinned = 0;
10380 spin_unlock(&block_group->lock);
10381 spin_unlock(&space_info->lock);
10383 /* DISCARD can flip during remount */
10384 trimming = btrfs_test_opt(root, DISCARD);
10386 /* Implicit trim during transaction commit. */
10388 btrfs_get_block_group_trimming(block_group);
10391 * Btrfs_remove_chunk will abort the transaction if things go
10394 ret = btrfs_remove_chunk(trans, root,
10395 block_group->key.objectid);
10399 btrfs_put_block_group_trimming(block_group);
10404 * If we're not mounted with -odiscard, we can just forget
10405 * about this block group. Otherwise we'll need to wait
10406 * until transaction commit to do the actual discard.
10409 WARN_ON(!list_empty(&block_group->bg_list));
10410 spin_lock(&trans->transaction->deleted_bgs_lock);
10411 list_move(&block_group->bg_list,
10412 &trans->transaction->deleted_bgs);
10413 spin_unlock(&trans->transaction->deleted_bgs_lock);
10414 btrfs_get_block_group(block_group);
10417 btrfs_end_transaction(trans, root);
10419 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10420 btrfs_put_block_group(block_group);
10421 spin_lock(&fs_info->unused_bgs_lock);
10423 spin_unlock(&fs_info->unused_bgs_lock);
10426 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10428 struct btrfs_space_info *space_info;
10429 struct btrfs_super_block *disk_super;
10435 disk_super = fs_info->super_copy;
10436 if (!btrfs_super_root(disk_super))
10439 features = btrfs_super_incompat_flags(disk_super);
10440 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10443 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10444 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10449 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10450 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10452 flags = BTRFS_BLOCK_GROUP_METADATA;
10453 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10457 flags = BTRFS_BLOCK_GROUP_DATA;
10458 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10464 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10466 return unpin_extent_range(root, start, end, false);
10470 * It used to be that old block groups would be left around forever.
10471 * Iterating over them would be enough to trim unused space. Since we
10472 * now automatically remove them, we also need to iterate over unallocated
10475 * We don't want a transaction for this since the discard may take a
10476 * substantial amount of time. We don't require that a transaction be
10477 * running, but we do need to take a running transaction into account
10478 * to ensure that we're not discarding chunks that were released in
10479 * the current transaction.
10481 * Holding the chunks lock will prevent other threads from allocating
10482 * or releasing chunks, but it won't prevent a running transaction
10483 * from committing and releasing the memory that the pending chunks
10484 * list head uses. For that, we need to take a reference to the
10487 static int btrfs_trim_free_extents(struct btrfs_device *device,
10488 u64 minlen, u64 *trimmed)
10490 u64 start = 0, len = 0;
10495 /* Not writeable = nothing to do. */
10496 if (!device->writeable)
10499 /* No free space = nothing to do. */
10500 if (device->total_bytes <= device->bytes_used)
10506 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10507 struct btrfs_transaction *trans;
10510 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10514 down_read(&fs_info->commit_root_sem);
10516 spin_lock(&fs_info->trans_lock);
10517 trans = fs_info->running_transaction;
10519 atomic_inc(&trans->use_count);
10520 spin_unlock(&fs_info->trans_lock);
10522 ret = find_free_dev_extent_start(trans, device, minlen, start,
10525 btrfs_put_transaction(trans);
10528 up_read(&fs_info->commit_root_sem);
10529 mutex_unlock(&fs_info->chunk_mutex);
10530 if (ret == -ENOSPC)
10535 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10536 up_read(&fs_info->commit_root_sem);
10537 mutex_unlock(&fs_info->chunk_mutex);
10545 if (fatal_signal_pending(current)) {
10546 ret = -ERESTARTSYS;
10556 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10558 struct btrfs_fs_info *fs_info = root->fs_info;
10559 struct btrfs_block_group_cache *cache = NULL;
10560 struct btrfs_device *device;
10561 struct list_head *devices;
10566 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10570 * try to trim all FS space, our block group may start from non-zero.
10572 if (range->len == total_bytes)
10573 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10575 cache = btrfs_lookup_block_group(fs_info, range->start);
10578 if (cache->key.objectid >= (range->start + range->len)) {
10579 btrfs_put_block_group(cache);
10583 start = max(range->start, cache->key.objectid);
10584 end = min(range->start + range->len,
10585 cache->key.objectid + cache->key.offset);
10587 if (end - start >= range->minlen) {
10588 if (!block_group_cache_done(cache)) {
10589 ret = cache_block_group(cache, 0);
10591 btrfs_put_block_group(cache);
10594 ret = wait_block_group_cache_done(cache);
10596 btrfs_put_block_group(cache);
10600 ret = btrfs_trim_block_group(cache,
10606 trimmed += group_trimmed;
10608 btrfs_put_block_group(cache);
10613 cache = next_block_group(fs_info->tree_root, cache);
10616 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10617 devices = &root->fs_info->fs_devices->alloc_list;
10618 list_for_each_entry(device, devices, dev_alloc_list) {
10619 ret = btrfs_trim_free_extents(device, range->minlen,
10624 trimmed += group_trimmed;
10626 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10628 range->len = trimmed;
10633 * btrfs_{start,end}_write_no_snapshoting() are similar to
10634 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10635 * data into the page cache through nocow before the subvolume is snapshoted,
10636 * but flush the data into disk after the snapshot creation, or to prevent
10637 * operations while snapshoting is ongoing and that cause the snapshot to be
10638 * inconsistent (writes followed by expanding truncates for example).
10640 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10642 percpu_counter_dec(&root->subv_writers->counter);
10644 * Make sure counter is updated before we wake up waiters.
10647 if (waitqueue_active(&root->subv_writers->wait))
10648 wake_up(&root->subv_writers->wait);
10651 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10653 if (atomic_read(&root->will_be_snapshoted))
10656 percpu_counter_inc(&root->subv_writers->counter);
10658 * Make sure counter is updated before we check for snapshot creation.
10661 if (atomic_read(&root->will_be_snapshoted)) {
10662 btrfs_end_write_no_snapshoting(root);