2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 #include <linux/sched.h>
19 #include <linux/pagemap.h>
20 #include <linux/writeback.h>
21 #include <linux/blkdev.h>
22 #include <linux/sort.h>
23 #include <linux/rcupdate.h>
24 #include <linux/kthread.h>
25 #include <linux/slab.h>
26 #include <linux/ratelimit.h>
27 #include <linux/percpu_counter.h>
31 #include "print-tree.h"
35 #include "free-space-cache.h"
40 #undef SCRAMBLE_DELAYED_REFS
43 * control flags for do_chunk_alloc's force field
44 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
45 * if we really need one.
47 * CHUNK_ALLOC_LIMITED means to only try and allocate one
48 * if we have very few chunks already allocated. This is
49 * used as part of the clustering code to help make sure
50 * we have a good pool of storage to cluster in, without
51 * filling the FS with empty chunks
53 * CHUNK_ALLOC_FORCE means it must try to allocate one
57 CHUNK_ALLOC_NO_FORCE = 0,
58 CHUNK_ALLOC_LIMITED = 1,
59 CHUNK_ALLOC_FORCE = 2,
63 * Control how reservations are dealt with.
65 * RESERVE_FREE - freeing a reservation.
66 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
68 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
69 * bytes_may_use as the ENOSPC accounting is done elsewhere
74 RESERVE_ALLOC_NO_ACCOUNT = 2,
77 static int update_block_group(struct btrfs_trans_handle *trans,
78 struct btrfs_root *root, u64 bytenr,
79 u64 num_bytes, int alloc);
80 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
81 struct btrfs_root *root,
82 struct btrfs_delayed_ref_node *node, u64 parent,
83 u64 root_objectid, u64 owner_objectid,
84 u64 owner_offset, int refs_to_drop,
85 struct btrfs_delayed_extent_op *extra_op);
86 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
87 struct extent_buffer *leaf,
88 struct btrfs_extent_item *ei);
89 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
90 struct btrfs_root *root,
91 u64 parent, u64 root_objectid,
92 u64 flags, u64 owner, u64 offset,
93 struct btrfs_key *ins, int ref_mod);
94 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
95 struct btrfs_root *root,
96 u64 parent, u64 root_objectid,
97 u64 flags, struct btrfs_disk_key *key,
98 int level, struct btrfs_key *ins,
100 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
101 struct btrfs_root *extent_root, u64 flags,
103 static int find_next_key(struct btrfs_path *path, int level,
104 struct btrfs_key *key);
105 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
106 int dump_block_groups);
107 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
108 u64 num_bytes, int reserve,
110 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
112 int btrfs_pin_extent(struct btrfs_root *root,
113 u64 bytenr, u64 num_bytes, int reserved);
116 block_group_cache_done(struct btrfs_block_group_cache *cache)
119 return cache->cached == BTRFS_CACHE_FINISHED ||
120 cache->cached == BTRFS_CACHE_ERROR;
123 static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
125 return (cache->flags & bits) == bits;
128 static void btrfs_get_block_group(struct btrfs_block_group_cache *cache)
130 atomic_inc(&cache->count);
133 void btrfs_put_block_group(struct btrfs_block_group_cache *cache)
135 if (atomic_dec_and_test(&cache->count)) {
136 WARN_ON(cache->pinned > 0);
137 WARN_ON(cache->reserved > 0);
138 kfree(cache->free_space_ctl);
144 * this adds the block group to the fs_info rb tree for the block group
147 static int btrfs_add_block_group_cache(struct btrfs_fs_info *info,
148 struct btrfs_block_group_cache *block_group)
151 struct rb_node *parent = NULL;
152 struct btrfs_block_group_cache *cache;
154 spin_lock(&info->block_group_cache_lock);
155 p = &info->block_group_cache_tree.rb_node;
159 cache = rb_entry(parent, struct btrfs_block_group_cache,
161 if (block_group->key.objectid < cache->key.objectid) {
163 } else if (block_group->key.objectid > cache->key.objectid) {
166 spin_unlock(&info->block_group_cache_lock);
171 rb_link_node(&block_group->cache_node, parent, p);
172 rb_insert_color(&block_group->cache_node,
173 &info->block_group_cache_tree);
175 if (info->first_logical_byte > block_group->key.objectid)
176 info->first_logical_byte = block_group->key.objectid;
178 spin_unlock(&info->block_group_cache_lock);
184 * This will return the block group at or after bytenr if contains is 0, else
185 * it will return the block group that contains the bytenr
187 static struct btrfs_block_group_cache *
188 block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
191 struct btrfs_block_group_cache *cache, *ret = NULL;
195 spin_lock(&info->block_group_cache_lock);
196 n = info->block_group_cache_tree.rb_node;
199 cache = rb_entry(n, struct btrfs_block_group_cache,
201 end = cache->key.objectid + cache->key.offset - 1;
202 start = cache->key.objectid;
204 if (bytenr < start) {
205 if (!contains && (!ret || start < ret->key.objectid))
208 } else if (bytenr > start) {
209 if (contains && bytenr <= end) {
220 btrfs_get_block_group(ret);
221 if (bytenr == 0 && info->first_logical_byte > ret->key.objectid)
222 info->first_logical_byte = ret->key.objectid;
224 spin_unlock(&info->block_group_cache_lock);
229 static int add_excluded_extent(struct btrfs_root *root,
230 u64 start, u64 num_bytes)
232 u64 end = start + num_bytes - 1;
233 set_extent_bits(&root->fs_info->freed_extents[0],
234 start, end, EXTENT_UPTODATE, GFP_NOFS);
235 set_extent_bits(&root->fs_info->freed_extents[1],
236 start, end, EXTENT_UPTODATE, GFP_NOFS);
240 static void free_excluded_extents(struct btrfs_root *root,
241 struct btrfs_block_group_cache *cache)
245 start = cache->key.objectid;
246 end = start + cache->key.offset - 1;
248 clear_extent_bits(&root->fs_info->freed_extents[0],
249 start, end, EXTENT_UPTODATE, GFP_NOFS);
250 clear_extent_bits(&root->fs_info->freed_extents[1],
251 start, end, EXTENT_UPTODATE, GFP_NOFS);
254 static int exclude_super_stripes(struct btrfs_root *root,
255 struct btrfs_block_group_cache *cache)
262 if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) {
263 stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid;
264 cache->bytes_super += stripe_len;
265 ret = add_excluded_extent(root, cache->key.objectid,
271 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
272 bytenr = btrfs_sb_offset(i);
273 ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
274 cache->key.objectid, bytenr,
275 0, &logical, &nr, &stripe_len);
282 if (logical[nr] > cache->key.objectid +
286 if (logical[nr] + stripe_len <= cache->key.objectid)
290 if (start < cache->key.objectid) {
291 start = cache->key.objectid;
292 len = (logical[nr] + stripe_len) - start;
294 len = min_t(u64, stripe_len,
295 cache->key.objectid +
296 cache->key.offset - start);
299 cache->bytes_super += len;
300 ret = add_excluded_extent(root, start, len);
312 static struct btrfs_caching_control *
313 get_caching_control(struct btrfs_block_group_cache *cache)
315 struct btrfs_caching_control *ctl;
317 spin_lock(&cache->lock);
318 if (!cache->caching_ctl) {
319 spin_unlock(&cache->lock);
323 ctl = cache->caching_ctl;
324 atomic_inc(&ctl->count);
325 spin_unlock(&cache->lock);
329 static void put_caching_control(struct btrfs_caching_control *ctl)
331 if (atomic_dec_and_test(&ctl->count))
335 #ifdef CONFIG_BTRFS_DEBUG
336 static void fragment_free_space(struct btrfs_root *root,
337 struct btrfs_block_group_cache *block_group)
339 u64 start = block_group->key.objectid;
340 u64 len = block_group->key.offset;
341 u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ?
342 root->nodesize : root->sectorsize;
343 u64 step = chunk << 1;
345 while (len > chunk) {
346 btrfs_remove_free_space(block_group, start, chunk);
357 * this is only called by cache_block_group, since we could have freed extents
358 * we need to check the pinned_extents for any extents that can't be used yet
359 * since their free space will be released as soon as the transaction commits.
361 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
362 struct btrfs_fs_info *info, u64 start, u64 end)
364 u64 extent_start, extent_end, size, total_added = 0;
367 while (start < end) {
368 ret = find_first_extent_bit(info->pinned_extents, start,
369 &extent_start, &extent_end,
370 EXTENT_DIRTY | EXTENT_UPTODATE,
375 if (extent_start <= start) {
376 start = extent_end + 1;
377 } else if (extent_start > start && extent_start < end) {
378 size = extent_start - start;
380 ret = btrfs_add_free_space(block_group, start,
382 BUG_ON(ret); /* -ENOMEM or logic error */
383 start = extent_end + 1;
392 ret = btrfs_add_free_space(block_group, start, size);
393 BUG_ON(ret); /* -ENOMEM or logic error */
399 static noinline void caching_thread(struct btrfs_work *work)
401 struct btrfs_block_group_cache *block_group;
402 struct btrfs_fs_info *fs_info;
403 struct btrfs_caching_control *caching_ctl;
404 struct btrfs_root *extent_root;
405 struct btrfs_path *path;
406 struct extent_buffer *leaf;
407 struct btrfs_key key;
414 caching_ctl = container_of(work, struct btrfs_caching_control, work);
415 block_group = caching_ctl->block_group;
416 fs_info = block_group->fs_info;
417 extent_root = fs_info->extent_root;
419 path = btrfs_alloc_path();
423 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
425 #ifdef CONFIG_BTRFS_DEBUG
427 * If we're fragmenting we don't want to make anybody think we can
428 * allocate from this block group until we've had a chance to fragment
431 if (btrfs_should_fragment_free_space(extent_root, block_group))
435 * We don't want to deadlock with somebody trying to allocate a new
436 * extent for the extent root while also trying to search the extent
437 * root to add free space. So we skip locking and search the commit
438 * root, since its read-only
440 path->skip_locking = 1;
441 path->search_commit_root = 1;
446 key.type = BTRFS_EXTENT_ITEM_KEY;
448 mutex_lock(&caching_ctl->mutex);
449 /* need to make sure the commit_root doesn't disappear */
450 down_read(&fs_info->commit_root_sem);
453 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
457 leaf = path->nodes[0];
458 nritems = btrfs_header_nritems(leaf);
461 if (btrfs_fs_closing(fs_info) > 1) {
466 if (path->slots[0] < nritems) {
467 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
469 ret = find_next_key(path, 0, &key);
473 if (need_resched() ||
474 rwsem_is_contended(&fs_info->commit_root_sem)) {
476 caching_ctl->progress = last;
477 btrfs_release_path(path);
478 up_read(&fs_info->commit_root_sem);
479 mutex_unlock(&caching_ctl->mutex);
484 ret = btrfs_next_leaf(extent_root, path);
489 leaf = path->nodes[0];
490 nritems = btrfs_header_nritems(leaf);
494 if (key.objectid < last) {
497 key.type = BTRFS_EXTENT_ITEM_KEY;
500 caching_ctl->progress = last;
501 btrfs_release_path(path);
505 if (key.objectid < block_group->key.objectid) {
510 if (key.objectid >= block_group->key.objectid +
511 block_group->key.offset)
514 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
515 key.type == BTRFS_METADATA_ITEM_KEY) {
516 total_found += add_new_free_space(block_group,
519 if (key.type == BTRFS_METADATA_ITEM_KEY)
520 last = key.objectid +
521 fs_info->tree_root->nodesize;
523 last = key.objectid + key.offset;
525 if (total_found > (1024 * 1024 * 2)) {
528 wake_up(&caching_ctl->wait);
535 total_found += add_new_free_space(block_group, fs_info, last,
536 block_group->key.objectid +
537 block_group->key.offset);
538 spin_lock(&block_group->lock);
539 block_group->caching_ctl = NULL;
540 block_group->cached = BTRFS_CACHE_FINISHED;
541 spin_unlock(&block_group->lock);
543 #ifdef CONFIG_BTRFS_DEBUG
544 if (btrfs_should_fragment_free_space(extent_root, block_group)) {
547 spin_lock(&block_group->space_info->lock);
548 spin_lock(&block_group->lock);
549 bytes_used = block_group->key.offset -
550 btrfs_block_group_used(&block_group->item);
551 block_group->space_info->bytes_used += bytes_used >> 1;
552 spin_unlock(&block_group->lock);
553 spin_unlock(&block_group->space_info->lock);
554 fragment_free_space(extent_root, block_group);
558 caching_ctl->progress = (u64)-1;
560 btrfs_free_path(path);
561 up_read(&fs_info->commit_root_sem);
563 free_excluded_extents(extent_root, block_group);
565 mutex_unlock(&caching_ctl->mutex);
568 spin_lock(&block_group->lock);
569 block_group->caching_ctl = NULL;
570 block_group->cached = BTRFS_CACHE_ERROR;
571 spin_unlock(&block_group->lock);
573 wake_up(&caching_ctl->wait);
575 put_caching_control(caching_ctl);
576 btrfs_put_block_group(block_group);
579 static int cache_block_group(struct btrfs_block_group_cache *cache,
583 struct btrfs_fs_info *fs_info = cache->fs_info;
584 struct btrfs_caching_control *caching_ctl;
587 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
591 INIT_LIST_HEAD(&caching_ctl->list);
592 mutex_init(&caching_ctl->mutex);
593 init_waitqueue_head(&caching_ctl->wait);
594 caching_ctl->block_group = cache;
595 caching_ctl->progress = cache->key.objectid;
596 atomic_set(&caching_ctl->count, 1);
597 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
598 caching_thread, NULL, NULL);
600 spin_lock(&cache->lock);
602 * This should be a rare occasion, but this could happen I think in the
603 * case where one thread starts to load the space cache info, and then
604 * some other thread starts a transaction commit which tries to do an
605 * allocation while the other thread is still loading the space cache
606 * info. The previous loop should have kept us from choosing this block
607 * group, but if we've moved to the state where we will wait on caching
608 * block groups we need to first check if we're doing a fast load here,
609 * so we can wait for it to finish, otherwise we could end up allocating
610 * from a block group who's cache gets evicted for one reason or
613 while (cache->cached == BTRFS_CACHE_FAST) {
614 struct btrfs_caching_control *ctl;
616 ctl = cache->caching_ctl;
617 atomic_inc(&ctl->count);
618 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
619 spin_unlock(&cache->lock);
623 finish_wait(&ctl->wait, &wait);
624 put_caching_control(ctl);
625 spin_lock(&cache->lock);
628 if (cache->cached != BTRFS_CACHE_NO) {
629 spin_unlock(&cache->lock);
633 WARN_ON(cache->caching_ctl);
634 cache->caching_ctl = caching_ctl;
635 cache->cached = BTRFS_CACHE_FAST;
636 spin_unlock(&cache->lock);
638 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
639 mutex_lock(&caching_ctl->mutex);
640 ret = load_free_space_cache(fs_info, cache);
642 spin_lock(&cache->lock);
644 cache->caching_ctl = NULL;
645 cache->cached = BTRFS_CACHE_FINISHED;
646 cache->last_byte_to_unpin = (u64)-1;
647 caching_ctl->progress = (u64)-1;
649 if (load_cache_only) {
650 cache->caching_ctl = NULL;
651 cache->cached = BTRFS_CACHE_NO;
653 cache->cached = BTRFS_CACHE_STARTED;
654 cache->has_caching_ctl = 1;
657 spin_unlock(&cache->lock);
658 #ifdef CONFIG_BTRFS_DEBUG
660 btrfs_should_fragment_free_space(fs_info->extent_root,
664 spin_lock(&cache->space_info->lock);
665 spin_lock(&cache->lock);
666 bytes_used = cache->key.offset -
667 btrfs_block_group_used(&cache->item);
668 cache->space_info->bytes_used += bytes_used >> 1;
669 spin_unlock(&cache->lock);
670 spin_unlock(&cache->space_info->lock);
671 fragment_free_space(fs_info->extent_root, cache);
674 mutex_unlock(&caching_ctl->mutex);
676 wake_up(&caching_ctl->wait);
678 put_caching_control(caching_ctl);
679 free_excluded_extents(fs_info->extent_root, cache);
684 * We are not going to do the fast caching, set cached to the
685 * appropriate value and wakeup any waiters.
687 spin_lock(&cache->lock);
688 if (load_cache_only) {
689 cache->caching_ctl = NULL;
690 cache->cached = BTRFS_CACHE_NO;
692 cache->cached = BTRFS_CACHE_STARTED;
693 cache->has_caching_ctl = 1;
695 spin_unlock(&cache->lock);
696 wake_up(&caching_ctl->wait);
699 if (load_cache_only) {
700 put_caching_control(caching_ctl);
704 down_write(&fs_info->commit_root_sem);
705 atomic_inc(&caching_ctl->count);
706 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
707 up_write(&fs_info->commit_root_sem);
709 btrfs_get_block_group(cache);
711 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
717 * return the block group that starts at or after bytenr
719 static struct btrfs_block_group_cache *
720 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
722 struct btrfs_block_group_cache *cache;
724 cache = block_group_cache_tree_search(info, bytenr, 0);
730 * return the block group that contains the given bytenr
732 struct btrfs_block_group_cache *btrfs_lookup_block_group(
733 struct btrfs_fs_info *info,
736 struct btrfs_block_group_cache *cache;
738 cache = block_group_cache_tree_search(info, bytenr, 1);
743 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
746 struct list_head *head = &info->space_info;
747 struct btrfs_space_info *found;
749 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
752 list_for_each_entry_rcu(found, head, list) {
753 if (found->flags & flags) {
763 * after adding space to the filesystem, we need to clear the full flags
764 * on all the space infos.
766 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
768 struct list_head *head = &info->space_info;
769 struct btrfs_space_info *found;
772 list_for_each_entry_rcu(found, head, list)
777 /* simple helper to search for an existing data extent at a given offset */
778 int btrfs_lookup_data_extent(struct btrfs_root *root, u64 start, u64 len)
781 struct btrfs_key key;
782 struct btrfs_path *path;
784 path = btrfs_alloc_path();
788 key.objectid = start;
790 key.type = BTRFS_EXTENT_ITEM_KEY;
791 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
793 btrfs_free_path(path);
798 * helper function to lookup reference count and flags of a tree block.
800 * the head node for delayed ref is used to store the sum of all the
801 * reference count modifications queued up in the rbtree. the head
802 * node may also store the extent flags to set. This way you can check
803 * to see what the reference count and extent flags would be if all of
804 * the delayed refs are not processed.
806 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
807 struct btrfs_root *root, u64 bytenr,
808 u64 offset, int metadata, u64 *refs, u64 *flags)
810 struct btrfs_delayed_ref_head *head;
811 struct btrfs_delayed_ref_root *delayed_refs;
812 struct btrfs_path *path;
813 struct btrfs_extent_item *ei;
814 struct extent_buffer *leaf;
815 struct btrfs_key key;
822 * If we don't have skinny metadata, don't bother doing anything
825 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
826 offset = root->nodesize;
830 path = btrfs_alloc_path();
835 path->skip_locking = 1;
836 path->search_commit_root = 1;
840 key.objectid = bytenr;
843 key.type = BTRFS_METADATA_ITEM_KEY;
845 key.type = BTRFS_EXTENT_ITEM_KEY;
847 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
852 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
853 if (path->slots[0]) {
855 btrfs_item_key_to_cpu(path->nodes[0], &key,
857 if (key.objectid == bytenr &&
858 key.type == BTRFS_EXTENT_ITEM_KEY &&
859 key.offset == root->nodesize)
865 leaf = path->nodes[0];
866 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
867 if (item_size >= sizeof(*ei)) {
868 ei = btrfs_item_ptr(leaf, path->slots[0],
869 struct btrfs_extent_item);
870 num_refs = btrfs_extent_refs(leaf, ei);
871 extent_flags = btrfs_extent_flags(leaf, ei);
873 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
874 struct btrfs_extent_item_v0 *ei0;
875 BUG_ON(item_size != sizeof(*ei0));
876 ei0 = btrfs_item_ptr(leaf, path->slots[0],
877 struct btrfs_extent_item_v0);
878 num_refs = btrfs_extent_refs_v0(leaf, ei0);
879 /* FIXME: this isn't correct for data */
880 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
885 BUG_ON(num_refs == 0);
895 delayed_refs = &trans->transaction->delayed_refs;
896 spin_lock(&delayed_refs->lock);
897 head = btrfs_find_delayed_ref_head(trans, bytenr);
899 if (!mutex_trylock(&head->mutex)) {
900 atomic_inc(&head->node.refs);
901 spin_unlock(&delayed_refs->lock);
903 btrfs_release_path(path);
906 * Mutex was contended, block until it's released and try
909 mutex_lock(&head->mutex);
910 mutex_unlock(&head->mutex);
911 btrfs_put_delayed_ref(&head->node);
914 spin_lock(&head->lock);
915 if (head->extent_op && head->extent_op->update_flags)
916 extent_flags |= head->extent_op->flags_to_set;
918 BUG_ON(num_refs == 0);
920 num_refs += head->node.ref_mod;
921 spin_unlock(&head->lock);
922 mutex_unlock(&head->mutex);
924 spin_unlock(&delayed_refs->lock);
926 WARN_ON(num_refs == 0);
930 *flags = extent_flags;
932 btrfs_free_path(path);
937 * Back reference rules. Back refs have three main goals:
939 * 1) differentiate between all holders of references to an extent so that
940 * when a reference is dropped we can make sure it was a valid reference
941 * before freeing the extent.
943 * 2) Provide enough information to quickly find the holders of an extent
944 * if we notice a given block is corrupted or bad.
946 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
947 * maintenance. This is actually the same as #2, but with a slightly
948 * different use case.
950 * There are two kinds of back refs. The implicit back refs is optimized
951 * for pointers in non-shared tree blocks. For a given pointer in a block,
952 * back refs of this kind provide information about the block's owner tree
953 * and the pointer's key. These information allow us to find the block by
954 * b-tree searching. The full back refs is for pointers in tree blocks not
955 * referenced by their owner trees. The location of tree block is recorded
956 * in the back refs. Actually the full back refs is generic, and can be
957 * used in all cases the implicit back refs is used. The major shortcoming
958 * of the full back refs is its overhead. Every time a tree block gets
959 * COWed, we have to update back refs entry for all pointers in it.
961 * For a newly allocated tree block, we use implicit back refs for
962 * pointers in it. This means most tree related operations only involve
963 * implicit back refs. For a tree block created in old transaction, the
964 * only way to drop a reference to it is COW it. So we can detect the
965 * event that tree block loses its owner tree's reference and do the
966 * back refs conversion.
968 * When a tree block is COW'd through a tree, there are four cases:
970 * The reference count of the block is one and the tree is the block's
971 * owner tree. Nothing to do in this case.
973 * The reference count of the block is one and the tree is not the
974 * block's owner tree. In this case, full back refs is used for pointers
975 * in the block. Remove these full back refs, add implicit back refs for
976 * every pointers in the new block.
978 * The reference count of the block is greater than one and the tree is
979 * the block's owner tree. In this case, implicit back refs is used for
980 * pointers in the block. Add full back refs for every pointers in the
981 * block, increase lower level extents' reference counts. The original
982 * implicit back refs are entailed to the new block.
984 * The reference count of the block is greater than one and the tree is
985 * not the block's owner tree. Add implicit back refs for every pointer in
986 * the new block, increase lower level extents' reference count.
988 * Back Reference Key composing:
990 * The key objectid corresponds to the first byte in the extent,
991 * The key type is used to differentiate between types of back refs.
992 * There are different meanings of the key offset for different types
995 * File extents can be referenced by:
997 * - multiple snapshots, subvolumes, or different generations in one subvol
998 * - different files inside a single subvolume
999 * - different offsets inside a file (bookend extents in file.c)
1001 * The extent ref structure for the implicit back refs has fields for:
1003 * - Objectid of the subvolume root
1004 * - objectid of the file holding the reference
1005 * - original offset in the file
1006 * - how many bookend extents
1008 * The key offset for the implicit back refs is hash of the first
1011 * The extent ref structure for the full back refs has field for:
1013 * - number of pointers in the tree leaf
1015 * The key offset for the implicit back refs is the first byte of
1018 * When a file extent is allocated, The implicit back refs is used.
1019 * the fields are filled in:
1021 * (root_key.objectid, inode objectid, offset in file, 1)
1023 * When a file extent is removed file truncation, we find the
1024 * corresponding implicit back refs and check the following fields:
1026 * (btrfs_header_owner(leaf), inode objectid, offset in file)
1028 * Btree extents can be referenced by:
1030 * - Different subvolumes
1032 * Both the implicit back refs and the full back refs for tree blocks
1033 * only consist of key. The key offset for the implicit back refs is
1034 * objectid of block's owner tree. The key offset for the full back refs
1035 * is the first byte of parent block.
1037 * When implicit back refs is used, information about the lowest key and
1038 * level of the tree block are required. These information are stored in
1039 * tree block info structure.
1042 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1043 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
1044 struct btrfs_root *root,
1045 struct btrfs_path *path,
1046 u64 owner, u32 extra_size)
1048 struct btrfs_extent_item *item;
1049 struct btrfs_extent_item_v0 *ei0;
1050 struct btrfs_extent_ref_v0 *ref0;
1051 struct btrfs_tree_block_info *bi;
1052 struct extent_buffer *leaf;
1053 struct btrfs_key key;
1054 struct btrfs_key found_key;
1055 u32 new_size = sizeof(*item);
1059 leaf = path->nodes[0];
1060 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1062 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1063 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1064 struct btrfs_extent_item_v0);
1065 refs = btrfs_extent_refs_v0(leaf, ei0);
1067 if (owner == (u64)-1) {
1069 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1070 ret = btrfs_next_leaf(root, path);
1073 BUG_ON(ret > 0); /* Corruption */
1074 leaf = path->nodes[0];
1076 btrfs_item_key_to_cpu(leaf, &found_key,
1078 BUG_ON(key.objectid != found_key.objectid);
1079 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1083 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1084 struct btrfs_extent_ref_v0);
1085 owner = btrfs_ref_objectid_v0(leaf, ref0);
1089 btrfs_release_path(path);
1091 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1092 new_size += sizeof(*bi);
1094 new_size -= sizeof(*ei0);
1095 ret = btrfs_search_slot(trans, root, &key, path,
1096 new_size + extra_size, 1);
1099 BUG_ON(ret); /* Corruption */
1101 btrfs_extend_item(root, path, new_size);
1103 leaf = path->nodes[0];
1104 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1105 btrfs_set_extent_refs(leaf, item, refs);
1106 /* FIXME: get real generation */
1107 btrfs_set_extent_generation(leaf, item, 0);
1108 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1109 btrfs_set_extent_flags(leaf, item,
1110 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1111 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1112 bi = (struct btrfs_tree_block_info *)(item + 1);
1113 /* FIXME: get first key of the block */
1114 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1115 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1117 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1119 btrfs_mark_buffer_dirty(leaf);
1124 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1126 u32 high_crc = ~(u32)0;
1127 u32 low_crc = ~(u32)0;
1130 lenum = cpu_to_le64(root_objectid);
1131 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1132 lenum = cpu_to_le64(owner);
1133 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1134 lenum = cpu_to_le64(offset);
1135 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1137 return ((u64)high_crc << 31) ^ (u64)low_crc;
1140 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1141 struct btrfs_extent_data_ref *ref)
1143 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1144 btrfs_extent_data_ref_objectid(leaf, ref),
1145 btrfs_extent_data_ref_offset(leaf, ref));
1148 static int match_extent_data_ref(struct extent_buffer *leaf,
1149 struct btrfs_extent_data_ref *ref,
1150 u64 root_objectid, u64 owner, u64 offset)
1152 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1153 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1154 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1159 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1160 struct btrfs_root *root,
1161 struct btrfs_path *path,
1162 u64 bytenr, u64 parent,
1164 u64 owner, u64 offset)
1166 struct btrfs_key key;
1167 struct btrfs_extent_data_ref *ref;
1168 struct extent_buffer *leaf;
1174 key.objectid = bytenr;
1176 key.type = BTRFS_SHARED_DATA_REF_KEY;
1177 key.offset = parent;
1179 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1180 key.offset = hash_extent_data_ref(root_objectid,
1185 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1194 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1195 key.type = BTRFS_EXTENT_REF_V0_KEY;
1196 btrfs_release_path(path);
1197 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1208 leaf = path->nodes[0];
1209 nritems = btrfs_header_nritems(leaf);
1211 if (path->slots[0] >= nritems) {
1212 ret = btrfs_next_leaf(root, path);
1218 leaf = path->nodes[0];
1219 nritems = btrfs_header_nritems(leaf);
1223 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1224 if (key.objectid != bytenr ||
1225 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1228 ref = btrfs_item_ptr(leaf, path->slots[0],
1229 struct btrfs_extent_data_ref);
1231 if (match_extent_data_ref(leaf, ref, root_objectid,
1234 btrfs_release_path(path);
1246 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1247 struct btrfs_root *root,
1248 struct btrfs_path *path,
1249 u64 bytenr, u64 parent,
1250 u64 root_objectid, u64 owner,
1251 u64 offset, int refs_to_add)
1253 struct btrfs_key key;
1254 struct extent_buffer *leaf;
1259 key.objectid = bytenr;
1261 key.type = BTRFS_SHARED_DATA_REF_KEY;
1262 key.offset = parent;
1263 size = sizeof(struct btrfs_shared_data_ref);
1265 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1266 key.offset = hash_extent_data_ref(root_objectid,
1268 size = sizeof(struct btrfs_extent_data_ref);
1271 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1272 if (ret && ret != -EEXIST)
1275 leaf = path->nodes[0];
1277 struct btrfs_shared_data_ref *ref;
1278 ref = btrfs_item_ptr(leaf, path->slots[0],
1279 struct btrfs_shared_data_ref);
1281 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1283 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1284 num_refs += refs_to_add;
1285 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1288 struct btrfs_extent_data_ref *ref;
1289 while (ret == -EEXIST) {
1290 ref = btrfs_item_ptr(leaf, path->slots[0],
1291 struct btrfs_extent_data_ref);
1292 if (match_extent_data_ref(leaf, ref, root_objectid,
1295 btrfs_release_path(path);
1297 ret = btrfs_insert_empty_item(trans, root, path, &key,
1299 if (ret && ret != -EEXIST)
1302 leaf = path->nodes[0];
1304 ref = btrfs_item_ptr(leaf, path->slots[0],
1305 struct btrfs_extent_data_ref);
1307 btrfs_set_extent_data_ref_root(leaf, ref,
1309 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1310 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1311 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1313 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1314 num_refs += refs_to_add;
1315 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1318 btrfs_mark_buffer_dirty(leaf);
1321 btrfs_release_path(path);
1325 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1326 struct btrfs_root *root,
1327 struct btrfs_path *path,
1328 int refs_to_drop, int *last_ref)
1330 struct btrfs_key key;
1331 struct btrfs_extent_data_ref *ref1 = NULL;
1332 struct btrfs_shared_data_ref *ref2 = NULL;
1333 struct extent_buffer *leaf;
1337 leaf = path->nodes[0];
1338 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1340 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1341 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1342 struct btrfs_extent_data_ref);
1343 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1344 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1345 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1346 struct btrfs_shared_data_ref);
1347 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1348 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1349 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1350 struct btrfs_extent_ref_v0 *ref0;
1351 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1352 struct btrfs_extent_ref_v0);
1353 num_refs = btrfs_ref_count_v0(leaf, ref0);
1359 BUG_ON(num_refs < refs_to_drop);
1360 num_refs -= refs_to_drop;
1362 if (num_refs == 0) {
1363 ret = btrfs_del_item(trans, root, path);
1366 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1367 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1368 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1369 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1370 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1372 struct btrfs_extent_ref_v0 *ref0;
1373 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1374 struct btrfs_extent_ref_v0);
1375 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1378 btrfs_mark_buffer_dirty(leaf);
1383 static noinline u32 extent_data_ref_count(struct btrfs_path *path,
1384 struct btrfs_extent_inline_ref *iref)
1386 struct btrfs_key key;
1387 struct extent_buffer *leaf;
1388 struct btrfs_extent_data_ref *ref1;
1389 struct btrfs_shared_data_ref *ref2;
1392 leaf = path->nodes[0];
1393 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1395 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1396 BTRFS_EXTENT_DATA_REF_KEY) {
1397 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1398 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1400 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1401 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1403 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1404 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1405 struct btrfs_extent_data_ref);
1406 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1407 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1408 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1409 struct btrfs_shared_data_ref);
1410 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1411 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1412 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1413 struct btrfs_extent_ref_v0 *ref0;
1414 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1415 struct btrfs_extent_ref_v0);
1416 num_refs = btrfs_ref_count_v0(leaf, ref0);
1424 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1425 struct btrfs_root *root,
1426 struct btrfs_path *path,
1427 u64 bytenr, u64 parent,
1430 struct btrfs_key key;
1433 key.objectid = bytenr;
1435 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1436 key.offset = parent;
1438 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1439 key.offset = root_objectid;
1442 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1445 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1446 if (ret == -ENOENT && parent) {
1447 btrfs_release_path(path);
1448 key.type = BTRFS_EXTENT_REF_V0_KEY;
1449 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1457 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1458 struct btrfs_root *root,
1459 struct btrfs_path *path,
1460 u64 bytenr, u64 parent,
1463 struct btrfs_key key;
1466 key.objectid = bytenr;
1468 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1469 key.offset = parent;
1471 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1472 key.offset = root_objectid;
1475 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1476 btrfs_release_path(path);
1480 static inline int extent_ref_type(u64 parent, u64 owner)
1483 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1485 type = BTRFS_SHARED_BLOCK_REF_KEY;
1487 type = BTRFS_TREE_BLOCK_REF_KEY;
1490 type = BTRFS_SHARED_DATA_REF_KEY;
1492 type = BTRFS_EXTENT_DATA_REF_KEY;
1497 static int find_next_key(struct btrfs_path *path, int level,
1498 struct btrfs_key *key)
1501 for (; level < BTRFS_MAX_LEVEL; level++) {
1502 if (!path->nodes[level])
1504 if (path->slots[level] + 1 >=
1505 btrfs_header_nritems(path->nodes[level]))
1508 btrfs_item_key_to_cpu(path->nodes[level], key,
1509 path->slots[level] + 1);
1511 btrfs_node_key_to_cpu(path->nodes[level], key,
1512 path->slots[level] + 1);
1519 * look for inline back ref. if back ref is found, *ref_ret is set
1520 * to the address of inline back ref, and 0 is returned.
1522 * if back ref isn't found, *ref_ret is set to the address where it
1523 * should be inserted, and -ENOENT is returned.
1525 * if insert is true and there are too many inline back refs, the path
1526 * points to the extent item, and -EAGAIN is returned.
1528 * NOTE: inline back refs are ordered in the same way that back ref
1529 * items in the tree are ordered.
1531 static noinline_for_stack
1532 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1533 struct btrfs_root *root,
1534 struct btrfs_path *path,
1535 struct btrfs_extent_inline_ref **ref_ret,
1536 u64 bytenr, u64 num_bytes,
1537 u64 parent, u64 root_objectid,
1538 u64 owner, u64 offset, int insert)
1540 struct btrfs_key key;
1541 struct extent_buffer *leaf;
1542 struct btrfs_extent_item *ei;
1543 struct btrfs_extent_inline_ref *iref;
1553 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1556 key.objectid = bytenr;
1557 key.type = BTRFS_EXTENT_ITEM_KEY;
1558 key.offset = num_bytes;
1560 want = extent_ref_type(parent, owner);
1562 extra_size = btrfs_extent_inline_ref_size(want);
1563 path->keep_locks = 1;
1568 * Owner is our parent level, so we can just add one to get the level
1569 * for the block we are interested in.
1571 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1572 key.type = BTRFS_METADATA_ITEM_KEY;
1577 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1584 * We may be a newly converted file system which still has the old fat
1585 * extent entries for metadata, so try and see if we have one of those.
1587 if (ret > 0 && skinny_metadata) {
1588 skinny_metadata = false;
1589 if (path->slots[0]) {
1591 btrfs_item_key_to_cpu(path->nodes[0], &key,
1593 if (key.objectid == bytenr &&
1594 key.type == BTRFS_EXTENT_ITEM_KEY &&
1595 key.offset == num_bytes)
1599 key.objectid = bytenr;
1600 key.type = BTRFS_EXTENT_ITEM_KEY;
1601 key.offset = num_bytes;
1602 btrfs_release_path(path);
1607 if (ret && !insert) {
1610 } else if (WARN_ON(ret)) {
1615 leaf = path->nodes[0];
1616 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1617 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1618 if (item_size < sizeof(*ei)) {
1623 ret = convert_extent_item_v0(trans, root, path, owner,
1629 leaf = path->nodes[0];
1630 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1633 BUG_ON(item_size < sizeof(*ei));
1635 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1636 flags = btrfs_extent_flags(leaf, ei);
1638 ptr = (unsigned long)(ei + 1);
1639 end = (unsigned long)ei + item_size;
1641 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1642 ptr += sizeof(struct btrfs_tree_block_info);
1652 iref = (struct btrfs_extent_inline_ref *)ptr;
1653 type = btrfs_extent_inline_ref_type(leaf, iref);
1657 ptr += btrfs_extent_inline_ref_size(type);
1661 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1662 struct btrfs_extent_data_ref *dref;
1663 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1664 if (match_extent_data_ref(leaf, dref, root_objectid,
1669 if (hash_extent_data_ref_item(leaf, dref) <
1670 hash_extent_data_ref(root_objectid, owner, offset))
1674 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1676 if (parent == ref_offset) {
1680 if (ref_offset < parent)
1683 if (root_objectid == ref_offset) {
1687 if (ref_offset < root_objectid)
1691 ptr += btrfs_extent_inline_ref_size(type);
1693 if (err == -ENOENT && insert) {
1694 if (item_size + extra_size >=
1695 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1700 * To add new inline back ref, we have to make sure
1701 * there is no corresponding back ref item.
1702 * For simplicity, we just do not add new inline back
1703 * ref if there is any kind of item for this block
1705 if (find_next_key(path, 0, &key) == 0 &&
1706 key.objectid == bytenr &&
1707 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1712 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1715 path->keep_locks = 0;
1716 btrfs_unlock_up_safe(path, 1);
1722 * helper to add new inline back ref
1724 static noinline_for_stack
1725 void setup_inline_extent_backref(struct btrfs_root *root,
1726 struct btrfs_path *path,
1727 struct btrfs_extent_inline_ref *iref,
1728 u64 parent, u64 root_objectid,
1729 u64 owner, u64 offset, int refs_to_add,
1730 struct btrfs_delayed_extent_op *extent_op)
1732 struct extent_buffer *leaf;
1733 struct btrfs_extent_item *ei;
1736 unsigned long item_offset;
1741 leaf = path->nodes[0];
1742 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1743 item_offset = (unsigned long)iref - (unsigned long)ei;
1745 type = extent_ref_type(parent, owner);
1746 size = btrfs_extent_inline_ref_size(type);
1748 btrfs_extend_item(root, path, size);
1750 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1751 refs = btrfs_extent_refs(leaf, ei);
1752 refs += refs_to_add;
1753 btrfs_set_extent_refs(leaf, ei, refs);
1755 __run_delayed_extent_op(extent_op, leaf, ei);
1757 ptr = (unsigned long)ei + item_offset;
1758 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1759 if (ptr < end - size)
1760 memmove_extent_buffer(leaf, ptr + size, ptr,
1763 iref = (struct btrfs_extent_inline_ref *)ptr;
1764 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1765 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1766 struct btrfs_extent_data_ref *dref;
1767 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1768 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1769 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1770 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1771 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1772 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1773 struct btrfs_shared_data_ref *sref;
1774 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1775 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1776 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1777 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1778 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1780 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1782 btrfs_mark_buffer_dirty(leaf);
1785 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1786 struct btrfs_root *root,
1787 struct btrfs_path *path,
1788 struct btrfs_extent_inline_ref **ref_ret,
1789 u64 bytenr, u64 num_bytes, u64 parent,
1790 u64 root_objectid, u64 owner, u64 offset)
1794 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1795 bytenr, num_bytes, parent,
1796 root_objectid, owner, offset, 0);
1800 btrfs_release_path(path);
1803 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1804 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1807 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1808 root_objectid, owner, offset);
1814 * helper to update/remove inline back ref
1816 static noinline_for_stack
1817 void update_inline_extent_backref(struct btrfs_root *root,
1818 struct btrfs_path *path,
1819 struct btrfs_extent_inline_ref *iref,
1821 struct btrfs_delayed_extent_op *extent_op,
1824 struct extent_buffer *leaf;
1825 struct btrfs_extent_item *ei;
1826 struct btrfs_extent_data_ref *dref = NULL;
1827 struct btrfs_shared_data_ref *sref = NULL;
1835 leaf = path->nodes[0];
1836 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1837 refs = btrfs_extent_refs(leaf, ei);
1838 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1839 refs += refs_to_mod;
1840 btrfs_set_extent_refs(leaf, ei, refs);
1842 __run_delayed_extent_op(extent_op, leaf, ei);
1844 type = btrfs_extent_inline_ref_type(leaf, iref);
1846 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1847 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1848 refs = btrfs_extent_data_ref_count(leaf, dref);
1849 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1850 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1851 refs = btrfs_shared_data_ref_count(leaf, sref);
1854 BUG_ON(refs_to_mod != -1);
1857 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1858 refs += refs_to_mod;
1861 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1862 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1864 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1867 size = btrfs_extent_inline_ref_size(type);
1868 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1869 ptr = (unsigned long)iref;
1870 end = (unsigned long)ei + item_size;
1871 if (ptr + size < end)
1872 memmove_extent_buffer(leaf, ptr, ptr + size,
1875 btrfs_truncate_item(root, path, item_size, 1);
1877 btrfs_mark_buffer_dirty(leaf);
1880 static noinline_for_stack
1881 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1882 struct btrfs_root *root,
1883 struct btrfs_path *path,
1884 u64 bytenr, u64 num_bytes, u64 parent,
1885 u64 root_objectid, u64 owner,
1886 u64 offset, int refs_to_add,
1887 struct btrfs_delayed_extent_op *extent_op)
1889 struct btrfs_extent_inline_ref *iref;
1892 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1893 bytenr, num_bytes, parent,
1894 root_objectid, owner, offset, 1);
1896 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1897 update_inline_extent_backref(root, path, iref,
1898 refs_to_add, extent_op, NULL);
1899 } else if (ret == -ENOENT) {
1900 setup_inline_extent_backref(root, path, iref, parent,
1901 root_objectid, owner, offset,
1902 refs_to_add, extent_op);
1908 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1909 struct btrfs_root *root,
1910 struct btrfs_path *path,
1911 u64 bytenr, u64 parent, u64 root_objectid,
1912 u64 owner, u64 offset, int refs_to_add)
1915 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1916 BUG_ON(refs_to_add != 1);
1917 ret = insert_tree_block_ref(trans, root, path, bytenr,
1918 parent, root_objectid);
1920 ret = insert_extent_data_ref(trans, root, path, bytenr,
1921 parent, root_objectid,
1922 owner, offset, refs_to_add);
1927 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1928 struct btrfs_root *root,
1929 struct btrfs_path *path,
1930 struct btrfs_extent_inline_ref *iref,
1931 int refs_to_drop, int is_data, int *last_ref)
1935 BUG_ON(!is_data && refs_to_drop != 1);
1937 update_inline_extent_backref(root, path, iref,
1938 -refs_to_drop, NULL, last_ref);
1939 } else if (is_data) {
1940 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1944 ret = btrfs_del_item(trans, root, path);
1949 #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len))
1950 static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len,
1951 u64 *discarded_bytes)
1954 u64 bytes_left, end;
1955 u64 aligned_start = ALIGN(start, 1 << 9);
1957 if (WARN_ON(start != aligned_start)) {
1958 len -= aligned_start - start;
1959 len = round_down(len, 1 << 9);
1960 start = aligned_start;
1963 *discarded_bytes = 0;
1971 /* Skip any superblocks on this device. */
1972 for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) {
1973 u64 sb_start = btrfs_sb_offset(j);
1974 u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE;
1975 u64 size = sb_start - start;
1977 if (!in_range(sb_start, start, bytes_left) &&
1978 !in_range(sb_end, start, bytes_left) &&
1979 !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE))
1983 * Superblock spans beginning of range. Adjust start and
1986 if (sb_start <= start) {
1987 start += sb_end - start;
1992 bytes_left = end - start;
1997 ret = blkdev_issue_discard(bdev, start >> 9, size >> 9,
2000 *discarded_bytes += size;
2001 else if (ret != -EOPNOTSUPP)
2010 bytes_left = end - start;
2014 ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9,
2017 *discarded_bytes += bytes_left;
2022 int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
2023 u64 num_bytes, u64 *actual_bytes)
2026 u64 discarded_bytes = 0;
2027 struct btrfs_bio *bbio = NULL;
2030 /* Tell the block device(s) that the sectors can be discarded */
2031 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
2032 bytenr, &num_bytes, &bbio, 0);
2033 /* Error condition is -ENOMEM */
2035 struct btrfs_bio_stripe *stripe = bbio->stripes;
2039 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
2041 if (!stripe->dev->can_discard)
2044 ret = btrfs_issue_discard(stripe->dev->bdev,
2049 discarded_bytes += bytes;
2050 else if (ret != -EOPNOTSUPP)
2051 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
2054 * Just in case we get back EOPNOTSUPP for some reason,
2055 * just ignore the return value so we don't screw up
2056 * people calling discard_extent.
2060 btrfs_put_bbio(bbio);
2064 *actual_bytes = discarded_bytes;
2067 if (ret == -EOPNOTSUPP)
2072 /* Can return -ENOMEM */
2073 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2074 struct btrfs_root *root,
2075 u64 bytenr, u64 num_bytes, u64 parent,
2076 u64 root_objectid, u64 owner, u64 offset,
2080 struct btrfs_fs_info *fs_info = root->fs_info;
2082 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
2083 root_objectid == BTRFS_TREE_LOG_OBJECTID);
2085 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
2086 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
2088 parent, root_objectid, (int)owner,
2089 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2091 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
2093 parent, root_objectid, owner, offset,
2094 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
2099 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
2100 struct btrfs_root *root,
2101 struct btrfs_delayed_ref_node *node,
2102 u64 parent, u64 root_objectid,
2103 u64 owner, u64 offset, int refs_to_add,
2104 struct btrfs_delayed_extent_op *extent_op)
2106 struct btrfs_fs_info *fs_info = root->fs_info;
2107 struct btrfs_path *path;
2108 struct extent_buffer *leaf;
2109 struct btrfs_extent_item *item;
2110 struct btrfs_key key;
2111 u64 bytenr = node->bytenr;
2112 u64 num_bytes = node->num_bytes;
2115 int no_quota = node->no_quota;
2117 path = btrfs_alloc_path();
2121 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2125 path->leave_spinning = 1;
2126 /* this will setup the path even if it fails to insert the back ref */
2127 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2128 bytenr, num_bytes, parent,
2129 root_objectid, owner, offset,
2130 refs_to_add, extent_op);
2131 if ((ret < 0 && ret != -EAGAIN) || !ret)
2135 * Ok we had -EAGAIN which means we didn't have space to insert and
2136 * inline extent ref, so just update the reference count and add a
2139 leaf = path->nodes[0];
2140 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2141 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2142 refs = btrfs_extent_refs(leaf, item);
2143 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2145 __run_delayed_extent_op(extent_op, leaf, item);
2147 btrfs_mark_buffer_dirty(leaf);
2148 btrfs_release_path(path);
2151 path->leave_spinning = 1;
2152 /* now insert the actual backref */
2153 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2154 path, bytenr, parent, root_objectid,
2155 owner, offset, refs_to_add);
2157 btrfs_abort_transaction(trans, root, ret);
2159 btrfs_free_path(path);
2163 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2164 struct btrfs_root *root,
2165 struct btrfs_delayed_ref_node *node,
2166 struct btrfs_delayed_extent_op *extent_op,
2167 int insert_reserved)
2170 struct btrfs_delayed_data_ref *ref;
2171 struct btrfs_key ins;
2176 ins.objectid = node->bytenr;
2177 ins.offset = node->num_bytes;
2178 ins.type = BTRFS_EXTENT_ITEM_KEY;
2180 ref = btrfs_delayed_node_to_data_ref(node);
2181 trace_run_delayed_data_ref(node, ref, node->action);
2183 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2184 parent = ref->parent;
2185 ref_root = ref->root;
2187 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2189 flags |= extent_op->flags_to_set;
2190 ret = alloc_reserved_file_extent(trans, root,
2191 parent, ref_root, flags,
2192 ref->objectid, ref->offset,
2193 &ins, node->ref_mod);
2194 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2195 ret = __btrfs_inc_extent_ref(trans, root, node, parent,
2196 ref_root, ref->objectid,
2197 ref->offset, node->ref_mod,
2199 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2200 ret = __btrfs_free_extent(trans, root, node, parent,
2201 ref_root, ref->objectid,
2202 ref->offset, node->ref_mod,
2210 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2211 struct extent_buffer *leaf,
2212 struct btrfs_extent_item *ei)
2214 u64 flags = btrfs_extent_flags(leaf, ei);
2215 if (extent_op->update_flags) {
2216 flags |= extent_op->flags_to_set;
2217 btrfs_set_extent_flags(leaf, ei, flags);
2220 if (extent_op->update_key) {
2221 struct btrfs_tree_block_info *bi;
2222 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2223 bi = (struct btrfs_tree_block_info *)(ei + 1);
2224 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2228 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2229 struct btrfs_root *root,
2230 struct btrfs_delayed_ref_node *node,
2231 struct btrfs_delayed_extent_op *extent_op)
2233 struct btrfs_key key;
2234 struct btrfs_path *path;
2235 struct btrfs_extent_item *ei;
2236 struct extent_buffer *leaf;
2240 int metadata = !extent_op->is_data;
2245 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2248 path = btrfs_alloc_path();
2252 key.objectid = node->bytenr;
2255 key.type = BTRFS_METADATA_ITEM_KEY;
2256 key.offset = extent_op->level;
2258 key.type = BTRFS_EXTENT_ITEM_KEY;
2259 key.offset = node->num_bytes;
2264 path->leave_spinning = 1;
2265 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2273 if (path->slots[0] > 0) {
2275 btrfs_item_key_to_cpu(path->nodes[0], &key,
2277 if (key.objectid == node->bytenr &&
2278 key.type == BTRFS_EXTENT_ITEM_KEY &&
2279 key.offset == node->num_bytes)
2283 btrfs_release_path(path);
2286 key.objectid = node->bytenr;
2287 key.offset = node->num_bytes;
2288 key.type = BTRFS_EXTENT_ITEM_KEY;
2297 leaf = path->nodes[0];
2298 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2299 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2300 if (item_size < sizeof(*ei)) {
2301 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2307 leaf = path->nodes[0];
2308 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2311 BUG_ON(item_size < sizeof(*ei));
2312 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2313 __run_delayed_extent_op(extent_op, leaf, ei);
2315 btrfs_mark_buffer_dirty(leaf);
2317 btrfs_free_path(path);
2321 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2322 struct btrfs_root *root,
2323 struct btrfs_delayed_ref_node *node,
2324 struct btrfs_delayed_extent_op *extent_op,
2325 int insert_reserved)
2328 struct btrfs_delayed_tree_ref *ref;
2329 struct btrfs_key ins;
2332 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2335 ref = btrfs_delayed_node_to_tree_ref(node);
2336 trace_run_delayed_tree_ref(node, ref, node->action);
2338 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2339 parent = ref->parent;
2340 ref_root = ref->root;
2342 ins.objectid = node->bytenr;
2343 if (skinny_metadata) {
2344 ins.offset = ref->level;
2345 ins.type = BTRFS_METADATA_ITEM_KEY;
2347 ins.offset = node->num_bytes;
2348 ins.type = BTRFS_EXTENT_ITEM_KEY;
2351 BUG_ON(node->ref_mod != 1);
2352 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2353 BUG_ON(!extent_op || !extent_op->update_flags);
2354 ret = alloc_reserved_tree_block(trans, root,
2356 extent_op->flags_to_set,
2360 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2361 ret = __btrfs_inc_extent_ref(trans, root, node,
2365 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2366 ret = __btrfs_free_extent(trans, root, node,
2368 ref->level, 0, 1, extent_op);
2375 /* helper function to actually process a single delayed ref entry */
2376 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2377 struct btrfs_root *root,
2378 struct btrfs_delayed_ref_node *node,
2379 struct btrfs_delayed_extent_op *extent_op,
2380 int insert_reserved)
2384 if (trans->aborted) {
2385 if (insert_reserved)
2386 btrfs_pin_extent(root, node->bytenr,
2387 node->num_bytes, 1);
2391 if (btrfs_delayed_ref_is_head(node)) {
2392 struct btrfs_delayed_ref_head *head;
2394 * we've hit the end of the chain and we were supposed
2395 * to insert this extent into the tree. But, it got
2396 * deleted before we ever needed to insert it, so all
2397 * we have to do is clean up the accounting
2400 head = btrfs_delayed_node_to_head(node);
2401 trace_run_delayed_ref_head(node, head, node->action);
2403 if (insert_reserved) {
2404 btrfs_pin_extent(root, node->bytenr,
2405 node->num_bytes, 1);
2406 if (head->is_data) {
2407 ret = btrfs_del_csums(trans, root,
2415 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2416 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2417 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2419 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2420 node->type == BTRFS_SHARED_DATA_REF_KEY)
2421 ret = run_delayed_data_ref(trans, root, node, extent_op,
2428 static inline struct btrfs_delayed_ref_node *
2429 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2431 struct btrfs_delayed_ref_node *ref;
2433 if (list_empty(&head->ref_list))
2437 * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first.
2438 * This is to prevent a ref count from going down to zero, which deletes
2439 * the extent item from the extent tree, when there still are references
2440 * to add, which would fail because they would not find the extent item.
2442 list_for_each_entry(ref, &head->ref_list, list) {
2443 if (ref->action == BTRFS_ADD_DELAYED_REF)
2447 return list_entry(head->ref_list.next, struct btrfs_delayed_ref_node,
2452 * Returns 0 on success or if called with an already aborted transaction.
2453 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2455 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2456 struct btrfs_root *root,
2459 struct btrfs_delayed_ref_root *delayed_refs;
2460 struct btrfs_delayed_ref_node *ref;
2461 struct btrfs_delayed_ref_head *locked_ref = NULL;
2462 struct btrfs_delayed_extent_op *extent_op;
2463 struct btrfs_fs_info *fs_info = root->fs_info;
2464 ktime_t start = ktime_get();
2466 unsigned long count = 0;
2467 unsigned long actual_count = 0;
2468 int must_insert_reserved = 0;
2470 delayed_refs = &trans->transaction->delayed_refs;
2476 spin_lock(&delayed_refs->lock);
2477 locked_ref = btrfs_select_ref_head(trans);
2479 spin_unlock(&delayed_refs->lock);
2483 /* grab the lock that says we are going to process
2484 * all the refs for this head */
2485 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2486 spin_unlock(&delayed_refs->lock);
2488 * we may have dropped the spin lock to get the head
2489 * mutex lock, and that might have given someone else
2490 * time to free the head. If that's true, it has been
2491 * removed from our list and we can move on.
2493 if (ret == -EAGAIN) {
2500 spin_lock(&locked_ref->lock);
2503 * locked_ref is the head node, so we have to go one
2504 * node back for any delayed ref updates
2506 ref = select_delayed_ref(locked_ref);
2508 if (ref && ref->seq &&
2509 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2510 spin_unlock(&locked_ref->lock);
2511 btrfs_delayed_ref_unlock(locked_ref);
2512 spin_lock(&delayed_refs->lock);
2513 locked_ref->processing = 0;
2514 delayed_refs->num_heads_ready++;
2515 spin_unlock(&delayed_refs->lock);
2523 * record the must insert reserved flag before we
2524 * drop the spin lock.
2526 must_insert_reserved = locked_ref->must_insert_reserved;
2527 locked_ref->must_insert_reserved = 0;
2529 extent_op = locked_ref->extent_op;
2530 locked_ref->extent_op = NULL;
2535 /* All delayed refs have been processed, Go ahead
2536 * and send the head node to run_one_delayed_ref,
2537 * so that any accounting fixes can happen
2539 ref = &locked_ref->node;
2541 if (extent_op && must_insert_reserved) {
2542 btrfs_free_delayed_extent_op(extent_op);
2547 spin_unlock(&locked_ref->lock);
2548 ret = run_delayed_extent_op(trans, root,
2550 btrfs_free_delayed_extent_op(extent_op);
2554 * Need to reset must_insert_reserved if
2555 * there was an error so the abort stuff
2556 * can cleanup the reserved space
2559 if (must_insert_reserved)
2560 locked_ref->must_insert_reserved = 1;
2561 locked_ref->processing = 0;
2562 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2563 btrfs_delayed_ref_unlock(locked_ref);
2570 * Need to drop our head ref lock and re-aqcuire the
2571 * delayed ref lock and then re-check to make sure
2574 spin_unlock(&locked_ref->lock);
2575 spin_lock(&delayed_refs->lock);
2576 spin_lock(&locked_ref->lock);
2577 if (!list_empty(&locked_ref->ref_list) ||
2578 locked_ref->extent_op) {
2579 spin_unlock(&locked_ref->lock);
2580 spin_unlock(&delayed_refs->lock);
2584 delayed_refs->num_heads--;
2585 rb_erase(&locked_ref->href_node,
2586 &delayed_refs->href_root);
2587 spin_unlock(&delayed_refs->lock);
2591 list_del(&ref->list);
2593 atomic_dec(&delayed_refs->num_entries);
2595 if (!btrfs_delayed_ref_is_head(ref)) {
2597 * when we play the delayed ref, also correct the
2600 switch (ref->action) {
2601 case BTRFS_ADD_DELAYED_REF:
2602 case BTRFS_ADD_DELAYED_EXTENT:
2603 locked_ref->node.ref_mod -= ref->ref_mod;
2605 case BTRFS_DROP_DELAYED_REF:
2606 locked_ref->node.ref_mod += ref->ref_mod;
2612 spin_unlock(&locked_ref->lock);
2614 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2615 must_insert_reserved);
2617 btrfs_free_delayed_extent_op(extent_op);
2619 locked_ref->processing = 0;
2620 btrfs_delayed_ref_unlock(locked_ref);
2621 btrfs_put_delayed_ref(ref);
2622 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2627 * If this node is a head, that means all the refs in this head
2628 * have been dealt with, and we will pick the next head to deal
2629 * with, so we must unlock the head and drop it from the cluster
2630 * list before we release it.
2632 if (btrfs_delayed_ref_is_head(ref)) {
2633 if (locked_ref->is_data &&
2634 locked_ref->total_ref_mod < 0) {
2635 spin_lock(&delayed_refs->lock);
2636 delayed_refs->pending_csums -= ref->num_bytes;
2637 spin_unlock(&delayed_refs->lock);
2639 btrfs_delayed_ref_unlock(locked_ref);
2642 btrfs_put_delayed_ref(ref);
2648 * We don't want to include ref heads since we can have empty ref heads
2649 * and those will drastically skew our runtime down since we just do
2650 * accounting, no actual extent tree updates.
2652 if (actual_count > 0) {
2653 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2657 * We weigh the current average higher than our current runtime
2658 * to avoid large swings in the average.
2660 spin_lock(&delayed_refs->lock);
2661 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2662 fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */
2663 spin_unlock(&delayed_refs->lock);
2668 #ifdef SCRAMBLE_DELAYED_REFS
2670 * Normally delayed refs get processed in ascending bytenr order. This
2671 * correlates in most cases to the order added. To expose dependencies on this
2672 * order, we start to process the tree in the middle instead of the beginning
2674 static u64 find_middle(struct rb_root *root)
2676 struct rb_node *n = root->rb_node;
2677 struct btrfs_delayed_ref_node *entry;
2680 u64 first = 0, last = 0;
2684 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2685 first = entry->bytenr;
2689 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2690 last = entry->bytenr;
2695 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2696 WARN_ON(!entry->in_tree);
2698 middle = entry->bytenr;
2711 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2715 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2716 sizeof(struct btrfs_extent_inline_ref));
2717 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2718 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2721 * We don't ever fill up leaves all the way so multiply by 2 just to be
2722 * closer to what we're really going to want to ouse.
2724 return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2728 * Takes the number of bytes to be csumm'ed and figures out how many leaves it
2729 * would require to store the csums for that many bytes.
2731 u64 btrfs_csum_bytes_to_leaves(struct btrfs_root *root, u64 csum_bytes)
2734 u64 num_csums_per_leaf;
2737 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
2738 num_csums_per_leaf = div64_u64(csum_size,
2739 (u64)btrfs_super_csum_size(root->fs_info->super_copy));
2740 num_csums = div64_u64(csum_bytes, root->sectorsize);
2741 num_csums += num_csums_per_leaf - 1;
2742 num_csums = div64_u64(num_csums, num_csums_per_leaf);
2746 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2747 struct btrfs_root *root)
2749 struct btrfs_block_rsv *global_rsv;
2750 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2751 u64 csum_bytes = trans->transaction->delayed_refs.pending_csums;
2752 u64 num_dirty_bgs = trans->transaction->num_dirty_bgs;
2753 u64 num_bytes, num_dirty_bgs_bytes;
2756 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2757 num_heads = heads_to_leaves(root, num_heads);
2759 num_bytes += (num_heads - 1) * root->nodesize;
2761 num_bytes += btrfs_csum_bytes_to_leaves(root, csum_bytes) * root->nodesize;
2762 num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(root,
2764 global_rsv = &root->fs_info->global_block_rsv;
2767 * If we can't allocate any more chunks lets make sure we have _lots_ of
2768 * wiggle room since running delayed refs can create more delayed refs.
2770 if (global_rsv->space_info->full) {
2771 num_dirty_bgs_bytes <<= 1;
2775 spin_lock(&global_rsv->lock);
2776 if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes)
2778 spin_unlock(&global_rsv->lock);
2782 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2783 struct btrfs_root *root)
2785 struct btrfs_fs_info *fs_info = root->fs_info;
2787 atomic_read(&trans->transaction->delayed_refs.num_entries);
2792 avg_runtime = fs_info->avg_delayed_ref_runtime;
2793 val = num_entries * avg_runtime;
2794 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2796 if (val >= NSEC_PER_SEC / 2)
2799 return btrfs_check_space_for_delayed_refs(trans, root);
2802 struct async_delayed_refs {
2803 struct btrfs_root *root;
2807 struct completion wait;
2808 struct btrfs_work work;
2811 static void delayed_ref_async_start(struct btrfs_work *work)
2813 struct async_delayed_refs *async;
2814 struct btrfs_trans_handle *trans;
2817 async = container_of(work, struct async_delayed_refs, work);
2819 trans = btrfs_join_transaction(async->root);
2820 if (IS_ERR(trans)) {
2821 async->error = PTR_ERR(trans);
2826 * trans->sync means that when we call end_transaciton, we won't
2827 * wait on delayed refs
2830 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2834 ret = btrfs_end_transaction(trans, async->root);
2835 if (ret && !async->error)
2839 complete(&async->wait);
2844 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2845 unsigned long count, int wait)
2847 struct async_delayed_refs *async;
2850 async = kmalloc(sizeof(*async), GFP_NOFS);
2854 async->root = root->fs_info->tree_root;
2855 async->count = count;
2861 init_completion(&async->wait);
2863 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2864 delayed_ref_async_start, NULL, NULL);
2866 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2869 wait_for_completion(&async->wait);
2878 * this starts processing the delayed reference count updates and
2879 * extent insertions we have queued up so far. count can be
2880 * 0, which means to process everything in the tree at the start
2881 * of the run (but not newly added entries), or it can be some target
2882 * number you'd like to process.
2884 * Returns 0 on success or if called with an aborted transaction
2885 * Returns <0 on error and aborts the transaction
2887 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2888 struct btrfs_root *root, unsigned long count)
2890 struct rb_node *node;
2891 struct btrfs_delayed_ref_root *delayed_refs;
2892 struct btrfs_delayed_ref_head *head;
2894 int run_all = count == (unsigned long)-1;
2895 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
2897 /* We'll clean this up in btrfs_cleanup_transaction */
2901 if (root == root->fs_info->extent_root)
2902 root = root->fs_info->tree_root;
2904 delayed_refs = &trans->transaction->delayed_refs;
2906 count = atomic_read(&delayed_refs->num_entries) * 2;
2909 #ifdef SCRAMBLE_DELAYED_REFS
2910 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2912 trans->can_flush_pending_bgs = false;
2913 ret = __btrfs_run_delayed_refs(trans, root, count);
2915 btrfs_abort_transaction(trans, root, ret);
2920 if (!list_empty(&trans->new_bgs))
2921 btrfs_create_pending_block_groups(trans, root);
2923 spin_lock(&delayed_refs->lock);
2924 node = rb_first(&delayed_refs->href_root);
2926 spin_unlock(&delayed_refs->lock);
2929 count = (unsigned long)-1;
2932 head = rb_entry(node, struct btrfs_delayed_ref_head,
2934 if (btrfs_delayed_ref_is_head(&head->node)) {
2935 struct btrfs_delayed_ref_node *ref;
2938 atomic_inc(&ref->refs);
2940 spin_unlock(&delayed_refs->lock);
2942 * Mutex was contended, block until it's
2943 * released and try again
2945 mutex_lock(&head->mutex);
2946 mutex_unlock(&head->mutex);
2948 btrfs_put_delayed_ref(ref);
2954 node = rb_next(node);
2956 spin_unlock(&delayed_refs->lock);
2961 assert_qgroups_uptodate(trans);
2962 trans->can_flush_pending_bgs = can_flush_pending_bgs;
2966 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2967 struct btrfs_root *root,
2968 u64 bytenr, u64 num_bytes, u64 flags,
2969 int level, int is_data)
2971 struct btrfs_delayed_extent_op *extent_op;
2974 extent_op = btrfs_alloc_delayed_extent_op();
2978 extent_op->flags_to_set = flags;
2979 extent_op->update_flags = 1;
2980 extent_op->update_key = 0;
2981 extent_op->is_data = is_data ? 1 : 0;
2982 extent_op->level = level;
2984 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2985 num_bytes, extent_op);
2987 btrfs_free_delayed_extent_op(extent_op);
2991 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2992 struct btrfs_root *root,
2993 struct btrfs_path *path,
2994 u64 objectid, u64 offset, u64 bytenr)
2996 struct btrfs_delayed_ref_head *head;
2997 struct btrfs_delayed_ref_node *ref;
2998 struct btrfs_delayed_data_ref *data_ref;
2999 struct btrfs_delayed_ref_root *delayed_refs;
3002 delayed_refs = &trans->transaction->delayed_refs;
3003 spin_lock(&delayed_refs->lock);
3004 head = btrfs_find_delayed_ref_head(trans, bytenr);
3006 spin_unlock(&delayed_refs->lock);
3010 if (!mutex_trylock(&head->mutex)) {
3011 atomic_inc(&head->node.refs);
3012 spin_unlock(&delayed_refs->lock);
3014 btrfs_release_path(path);
3017 * Mutex was contended, block until it's released and let
3020 mutex_lock(&head->mutex);
3021 mutex_unlock(&head->mutex);
3022 btrfs_put_delayed_ref(&head->node);
3025 spin_unlock(&delayed_refs->lock);
3027 spin_lock(&head->lock);
3028 list_for_each_entry(ref, &head->ref_list, list) {
3029 /* If it's a shared ref we know a cross reference exists */
3030 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
3035 data_ref = btrfs_delayed_node_to_data_ref(ref);
3038 * If our ref doesn't match the one we're currently looking at
3039 * then we have a cross reference.
3041 if (data_ref->root != root->root_key.objectid ||
3042 data_ref->objectid != objectid ||
3043 data_ref->offset != offset) {
3048 spin_unlock(&head->lock);
3049 mutex_unlock(&head->mutex);
3053 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
3054 struct btrfs_root *root,
3055 struct btrfs_path *path,
3056 u64 objectid, u64 offset, u64 bytenr)
3058 struct btrfs_root *extent_root = root->fs_info->extent_root;
3059 struct extent_buffer *leaf;
3060 struct btrfs_extent_data_ref *ref;
3061 struct btrfs_extent_inline_ref *iref;
3062 struct btrfs_extent_item *ei;
3063 struct btrfs_key key;
3067 key.objectid = bytenr;
3068 key.offset = (u64)-1;
3069 key.type = BTRFS_EXTENT_ITEM_KEY;
3071 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
3074 BUG_ON(ret == 0); /* Corruption */
3077 if (path->slots[0] == 0)
3081 leaf = path->nodes[0];
3082 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3084 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
3088 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3089 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3090 if (item_size < sizeof(*ei)) {
3091 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3095 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
3097 if (item_size != sizeof(*ei) +
3098 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
3101 if (btrfs_extent_generation(leaf, ei) <=
3102 btrfs_root_last_snapshot(&root->root_item))
3105 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3106 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3107 BTRFS_EXTENT_DATA_REF_KEY)
3110 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3111 if (btrfs_extent_refs(leaf, ei) !=
3112 btrfs_extent_data_ref_count(leaf, ref) ||
3113 btrfs_extent_data_ref_root(leaf, ref) !=
3114 root->root_key.objectid ||
3115 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3116 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3124 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3125 struct btrfs_root *root,
3126 u64 objectid, u64 offset, u64 bytenr)
3128 struct btrfs_path *path;
3132 path = btrfs_alloc_path();
3137 ret = check_committed_ref(trans, root, path, objectid,
3139 if (ret && ret != -ENOENT)
3142 ret2 = check_delayed_ref(trans, root, path, objectid,
3144 } while (ret2 == -EAGAIN);
3146 if (ret2 && ret2 != -ENOENT) {
3151 if (ret != -ENOENT || ret2 != -ENOENT)
3154 btrfs_free_path(path);
3155 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3160 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3161 struct btrfs_root *root,
3162 struct extent_buffer *buf,
3163 int full_backref, int inc)
3170 struct btrfs_key key;
3171 struct btrfs_file_extent_item *fi;
3175 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3176 u64, u64, u64, u64, u64, u64, int);
3179 if (btrfs_test_is_dummy_root(root))
3182 ref_root = btrfs_header_owner(buf);
3183 nritems = btrfs_header_nritems(buf);
3184 level = btrfs_header_level(buf);
3186 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3190 process_func = btrfs_inc_extent_ref;
3192 process_func = btrfs_free_extent;
3195 parent = buf->start;
3199 for (i = 0; i < nritems; i++) {
3201 btrfs_item_key_to_cpu(buf, &key, i);
3202 if (key.type != BTRFS_EXTENT_DATA_KEY)
3204 fi = btrfs_item_ptr(buf, i,
3205 struct btrfs_file_extent_item);
3206 if (btrfs_file_extent_type(buf, fi) ==
3207 BTRFS_FILE_EXTENT_INLINE)
3209 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3213 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3214 key.offset -= btrfs_file_extent_offset(buf, fi);
3215 ret = process_func(trans, root, bytenr, num_bytes,
3216 parent, ref_root, key.objectid,
3221 bytenr = btrfs_node_blockptr(buf, i);
3222 num_bytes = root->nodesize;
3223 ret = process_func(trans, root, bytenr, num_bytes,
3224 parent, ref_root, level - 1, 0,
3235 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3236 struct extent_buffer *buf, int full_backref)
3238 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3241 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3242 struct extent_buffer *buf, int full_backref)
3244 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3247 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3248 struct btrfs_root *root,
3249 struct btrfs_path *path,
3250 struct btrfs_block_group_cache *cache)
3253 struct btrfs_root *extent_root = root->fs_info->extent_root;
3255 struct extent_buffer *leaf;
3257 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3264 leaf = path->nodes[0];
3265 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3266 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3267 btrfs_mark_buffer_dirty(leaf);
3269 btrfs_release_path(path);
3274 static struct btrfs_block_group_cache *
3275 next_block_group(struct btrfs_root *root,
3276 struct btrfs_block_group_cache *cache)
3278 struct rb_node *node;
3280 spin_lock(&root->fs_info->block_group_cache_lock);
3282 /* If our block group was removed, we need a full search. */
3283 if (RB_EMPTY_NODE(&cache->cache_node)) {
3284 const u64 next_bytenr = cache->key.objectid + cache->key.offset;
3286 spin_unlock(&root->fs_info->block_group_cache_lock);
3287 btrfs_put_block_group(cache);
3288 cache = btrfs_lookup_first_block_group(root->fs_info,
3292 node = rb_next(&cache->cache_node);
3293 btrfs_put_block_group(cache);
3295 cache = rb_entry(node, struct btrfs_block_group_cache,
3297 btrfs_get_block_group(cache);
3300 spin_unlock(&root->fs_info->block_group_cache_lock);
3304 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3305 struct btrfs_trans_handle *trans,
3306 struct btrfs_path *path)
3308 struct btrfs_root *root = block_group->fs_info->tree_root;
3309 struct inode *inode = NULL;
3311 int dcs = BTRFS_DC_ERROR;
3317 * If this block group is smaller than 100 megs don't bother caching the
3320 if (block_group->key.offset < (100 * 1024 * 1024)) {
3321 spin_lock(&block_group->lock);
3322 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3323 spin_unlock(&block_group->lock);
3330 inode = lookup_free_space_inode(root, block_group, path);
3331 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3332 ret = PTR_ERR(inode);
3333 btrfs_release_path(path);
3337 if (IS_ERR(inode)) {
3341 if (block_group->ro)
3344 ret = create_free_space_inode(root, trans, block_group, path);
3350 /* We've already setup this transaction, go ahead and exit */
3351 if (block_group->cache_generation == trans->transid &&
3352 i_size_read(inode)) {
3353 dcs = BTRFS_DC_SETUP;
3358 * We want to set the generation to 0, that way if anything goes wrong
3359 * from here on out we know not to trust this cache when we load up next
3362 BTRFS_I(inode)->generation = 0;
3363 ret = btrfs_update_inode(trans, root, inode);
3366 * So theoretically we could recover from this, simply set the
3367 * super cache generation to 0 so we know to invalidate the
3368 * cache, but then we'd have to keep track of the block groups
3369 * that fail this way so we know we _have_ to reset this cache
3370 * before the next commit or risk reading stale cache. So to
3371 * limit our exposure to horrible edge cases lets just abort the
3372 * transaction, this only happens in really bad situations
3375 btrfs_abort_transaction(trans, root, ret);
3380 if (i_size_read(inode) > 0) {
3381 ret = btrfs_check_trunc_cache_free_space(root,
3382 &root->fs_info->global_block_rsv);
3386 ret = btrfs_truncate_free_space_cache(root, trans, NULL, inode);
3391 spin_lock(&block_group->lock);
3392 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3393 !btrfs_test_opt(root, SPACE_CACHE)) {
3395 * don't bother trying to write stuff out _if_
3396 * a) we're not cached,
3397 * b) we're with nospace_cache mount option.
3399 dcs = BTRFS_DC_WRITTEN;
3400 spin_unlock(&block_group->lock);
3403 spin_unlock(&block_group->lock);
3406 * Try to preallocate enough space based on how big the block group is.
3407 * Keep in mind this has to include any pinned space which could end up
3408 * taking up quite a bit since it's not folded into the other space
3411 num_pages = div_u64(block_group->key.offset, 256 * 1024 * 1024);
3416 num_pages *= PAGE_CACHE_SIZE;
3418 ret = btrfs_check_data_free_space(inode, num_pages, num_pages);
3422 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3423 num_pages, num_pages,
3426 dcs = BTRFS_DC_SETUP;
3427 btrfs_free_reserved_data_space(inode, num_pages);
3432 btrfs_release_path(path);
3434 spin_lock(&block_group->lock);
3435 if (!ret && dcs == BTRFS_DC_SETUP)
3436 block_group->cache_generation = trans->transid;
3437 block_group->disk_cache_state = dcs;
3438 spin_unlock(&block_group->lock);
3443 int btrfs_setup_space_cache(struct btrfs_trans_handle *trans,
3444 struct btrfs_root *root)
3446 struct btrfs_block_group_cache *cache, *tmp;
3447 struct btrfs_transaction *cur_trans = trans->transaction;
3448 struct btrfs_path *path;
3450 if (list_empty(&cur_trans->dirty_bgs) ||
3451 !btrfs_test_opt(root, SPACE_CACHE))
3454 path = btrfs_alloc_path();
3458 /* Could add new block groups, use _safe just in case */
3459 list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs,
3461 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3462 cache_save_setup(cache, trans, path);
3465 btrfs_free_path(path);
3470 * transaction commit does final block group cache writeback during a
3471 * critical section where nothing is allowed to change the FS. This is
3472 * required in order for the cache to actually match the block group,
3473 * but can introduce a lot of latency into the commit.
3475 * So, btrfs_start_dirty_block_groups is here to kick off block group
3476 * cache IO. There's a chance we'll have to redo some of it if the
3477 * block group changes again during the commit, but it greatly reduces
3478 * the commit latency by getting rid of the easy block groups while
3479 * we're still allowing others to join the commit.
3481 int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans,
3482 struct btrfs_root *root)
3484 struct btrfs_block_group_cache *cache;
3485 struct btrfs_transaction *cur_trans = trans->transaction;
3488 struct btrfs_path *path = NULL;
3490 struct list_head *io = &cur_trans->io_bgs;
3491 int num_started = 0;
3494 spin_lock(&cur_trans->dirty_bgs_lock);
3495 if (list_empty(&cur_trans->dirty_bgs)) {
3496 spin_unlock(&cur_trans->dirty_bgs_lock);
3499 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3500 spin_unlock(&cur_trans->dirty_bgs_lock);
3504 * make sure all the block groups on our dirty list actually
3507 btrfs_create_pending_block_groups(trans, root);
3510 path = btrfs_alloc_path();
3516 * cache_write_mutex is here only to save us from balance or automatic
3517 * removal of empty block groups deleting this block group while we are
3518 * writing out the cache
3520 mutex_lock(&trans->transaction->cache_write_mutex);
3521 while (!list_empty(&dirty)) {
3522 cache = list_first_entry(&dirty,
3523 struct btrfs_block_group_cache,
3526 * this can happen if something re-dirties a block
3527 * group that is already under IO. Just wait for it to
3528 * finish and then do it all again
3530 if (!list_empty(&cache->io_list)) {
3531 list_del_init(&cache->io_list);
3532 btrfs_wait_cache_io(root, trans, cache,
3533 &cache->io_ctl, path,
3534 cache->key.objectid);
3535 btrfs_put_block_group(cache);
3540 * btrfs_wait_cache_io uses the cache->dirty_list to decide
3541 * if it should update the cache_state. Don't delete
3542 * until after we wait.
3544 * Since we're not running in the commit critical section
3545 * we need the dirty_bgs_lock to protect from update_block_group
3547 spin_lock(&cur_trans->dirty_bgs_lock);
3548 list_del_init(&cache->dirty_list);
3549 spin_unlock(&cur_trans->dirty_bgs_lock);
3553 cache_save_setup(cache, trans, path);
3555 if (cache->disk_cache_state == BTRFS_DC_SETUP) {
3556 cache->io_ctl.inode = NULL;
3557 ret = btrfs_write_out_cache(root, trans, cache, path);
3558 if (ret == 0 && cache->io_ctl.inode) {
3563 * the cache_write_mutex is protecting
3566 list_add_tail(&cache->io_list, io);
3569 * if we failed to write the cache, the
3570 * generation will be bad and life goes on
3576 ret = write_one_cache_group(trans, root, path, cache);
3578 * Our block group might still be attached to the list
3579 * of new block groups in the transaction handle of some
3580 * other task (struct btrfs_trans_handle->new_bgs). This
3581 * means its block group item isn't yet in the extent
3582 * tree. If this happens ignore the error, as we will
3583 * try again later in the critical section of the
3584 * transaction commit.
3586 if (ret == -ENOENT) {
3588 spin_lock(&cur_trans->dirty_bgs_lock);
3589 if (list_empty(&cache->dirty_list)) {
3590 list_add_tail(&cache->dirty_list,
3591 &cur_trans->dirty_bgs);
3592 btrfs_get_block_group(cache);
3594 spin_unlock(&cur_trans->dirty_bgs_lock);
3596 btrfs_abort_transaction(trans, root, ret);
3600 /* if its not on the io list, we need to put the block group */
3602 btrfs_put_block_group(cache);
3608 * Avoid blocking other tasks for too long. It might even save
3609 * us from writing caches for block groups that are going to be
3612 mutex_unlock(&trans->transaction->cache_write_mutex);
3613 mutex_lock(&trans->transaction->cache_write_mutex);
3615 mutex_unlock(&trans->transaction->cache_write_mutex);
3618 * go through delayed refs for all the stuff we've just kicked off
3619 * and then loop back (just once)
3621 ret = btrfs_run_delayed_refs(trans, root, 0);
3622 if (!ret && loops == 0) {
3624 spin_lock(&cur_trans->dirty_bgs_lock);
3625 list_splice_init(&cur_trans->dirty_bgs, &dirty);
3627 * dirty_bgs_lock protects us from concurrent block group
3628 * deletes too (not just cache_write_mutex).
3630 if (!list_empty(&dirty)) {
3631 spin_unlock(&cur_trans->dirty_bgs_lock);
3634 spin_unlock(&cur_trans->dirty_bgs_lock);
3637 btrfs_free_path(path);
3641 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3642 struct btrfs_root *root)
3644 struct btrfs_block_group_cache *cache;
3645 struct btrfs_transaction *cur_trans = trans->transaction;
3648 struct btrfs_path *path;
3649 struct list_head *io = &cur_trans->io_bgs;
3650 int num_started = 0;
3652 path = btrfs_alloc_path();
3657 * We don't need the lock here since we are protected by the transaction
3658 * commit. We want to do the cache_save_setup first and then run the
3659 * delayed refs to make sure we have the best chance at doing this all
3662 while (!list_empty(&cur_trans->dirty_bgs)) {
3663 cache = list_first_entry(&cur_trans->dirty_bgs,
3664 struct btrfs_block_group_cache,
3668 * this can happen if cache_save_setup re-dirties a block
3669 * group that is already under IO. Just wait for it to
3670 * finish and then do it all again
3672 if (!list_empty(&cache->io_list)) {
3673 list_del_init(&cache->io_list);
3674 btrfs_wait_cache_io(root, trans, cache,
3675 &cache->io_ctl, path,
3676 cache->key.objectid);
3677 btrfs_put_block_group(cache);
3681 * don't remove from the dirty list until after we've waited
3684 list_del_init(&cache->dirty_list);
3687 cache_save_setup(cache, trans, path);
3690 ret = btrfs_run_delayed_refs(trans, root, (unsigned long) -1);
3692 if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) {
3693 cache->io_ctl.inode = NULL;
3694 ret = btrfs_write_out_cache(root, trans, cache, path);
3695 if (ret == 0 && cache->io_ctl.inode) {
3698 list_add_tail(&cache->io_list, io);
3701 * if we failed to write the cache, the
3702 * generation will be bad and life goes on
3708 ret = write_one_cache_group(trans, root, path, cache);
3710 btrfs_abort_transaction(trans, root, ret);
3713 /* if its not on the io list, we need to put the block group */
3715 btrfs_put_block_group(cache);
3718 while (!list_empty(io)) {
3719 cache = list_first_entry(io, struct btrfs_block_group_cache,
3721 list_del_init(&cache->io_list);
3722 btrfs_wait_cache_io(root, trans, cache,
3723 &cache->io_ctl, path, cache->key.objectid);
3724 btrfs_put_block_group(cache);
3727 btrfs_free_path(path);
3731 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3733 struct btrfs_block_group_cache *block_group;
3736 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3737 if (!block_group || block_group->ro)
3740 btrfs_put_block_group(block_group);
3744 static const char *alloc_name(u64 flags)
3747 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3749 case BTRFS_BLOCK_GROUP_METADATA:
3751 case BTRFS_BLOCK_GROUP_DATA:
3753 case BTRFS_BLOCK_GROUP_SYSTEM:
3757 return "invalid-combination";
3761 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3762 u64 total_bytes, u64 bytes_used,
3763 struct btrfs_space_info **space_info)
3765 struct btrfs_space_info *found;
3770 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3771 BTRFS_BLOCK_GROUP_RAID10))
3776 found = __find_space_info(info, flags);
3778 spin_lock(&found->lock);
3779 found->total_bytes += total_bytes;
3780 found->disk_total += total_bytes * factor;
3781 found->bytes_used += bytes_used;
3782 found->disk_used += bytes_used * factor;
3783 if (total_bytes > 0)
3785 spin_unlock(&found->lock);
3786 *space_info = found;
3789 found = kzalloc(sizeof(*found), GFP_NOFS);
3793 ret = percpu_counter_init(&found->total_bytes_pinned, 0, GFP_KERNEL);
3799 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3800 INIT_LIST_HEAD(&found->block_groups[i]);
3801 init_rwsem(&found->groups_sem);
3802 spin_lock_init(&found->lock);
3803 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3804 found->total_bytes = total_bytes;
3805 found->disk_total = total_bytes * factor;
3806 found->bytes_used = bytes_used;
3807 found->disk_used = bytes_used * factor;
3808 found->bytes_pinned = 0;
3809 found->bytes_reserved = 0;
3810 found->bytes_readonly = 0;
3811 found->bytes_may_use = 0;
3813 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3814 found->chunk_alloc = 0;
3816 init_waitqueue_head(&found->wait);
3817 INIT_LIST_HEAD(&found->ro_bgs);
3819 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3820 info->space_info_kobj, "%s",
3821 alloc_name(found->flags));
3827 *space_info = found;
3828 list_add_rcu(&found->list, &info->space_info);
3829 if (flags & BTRFS_BLOCK_GROUP_DATA)
3830 info->data_sinfo = found;
3835 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3837 u64 extra_flags = chunk_to_extended(flags) &
3838 BTRFS_EXTENDED_PROFILE_MASK;
3840 write_seqlock(&fs_info->profiles_lock);
3841 if (flags & BTRFS_BLOCK_GROUP_DATA)
3842 fs_info->avail_data_alloc_bits |= extra_flags;
3843 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3844 fs_info->avail_metadata_alloc_bits |= extra_flags;
3845 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3846 fs_info->avail_system_alloc_bits |= extra_flags;
3847 write_sequnlock(&fs_info->profiles_lock);
3851 * returns target flags in extended format or 0 if restripe for this
3852 * chunk_type is not in progress
3854 * should be called with either volume_mutex or balance_lock held
3856 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3858 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3864 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3865 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3866 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3867 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3868 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3869 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3870 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3871 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3872 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3879 * @flags: available profiles in extended format (see ctree.h)
3881 * Returns reduced profile in chunk format. If profile changing is in
3882 * progress (either running or paused) picks the target profile (if it's
3883 * already available), otherwise falls back to plain reducing.
3885 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3887 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3893 * see if restripe for this chunk_type is in progress, if so
3894 * try to reduce to the target profile
3896 spin_lock(&root->fs_info->balance_lock);
3897 target = get_restripe_target(root->fs_info, flags);
3899 /* pick target profile only if it's already available */
3900 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3901 spin_unlock(&root->fs_info->balance_lock);
3902 return extended_to_chunk(target);
3905 spin_unlock(&root->fs_info->balance_lock);
3907 /* First, mask out the RAID levels which aren't possible */
3908 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3909 if (num_devices >= btrfs_raid_array[raid_type].devs_min)
3910 allowed |= btrfs_raid_group[raid_type];
3914 if (allowed & BTRFS_BLOCK_GROUP_RAID6)
3915 allowed = BTRFS_BLOCK_GROUP_RAID6;
3916 else if (allowed & BTRFS_BLOCK_GROUP_RAID5)
3917 allowed = BTRFS_BLOCK_GROUP_RAID5;
3918 else if (allowed & BTRFS_BLOCK_GROUP_RAID10)
3919 allowed = BTRFS_BLOCK_GROUP_RAID10;
3920 else if (allowed & BTRFS_BLOCK_GROUP_RAID1)
3921 allowed = BTRFS_BLOCK_GROUP_RAID1;
3922 else if (allowed & BTRFS_BLOCK_GROUP_RAID0)
3923 allowed = BTRFS_BLOCK_GROUP_RAID0;
3925 flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK;
3927 return extended_to_chunk(flags | allowed);
3930 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3937 seq = read_seqbegin(&root->fs_info->profiles_lock);
3939 if (flags & BTRFS_BLOCK_GROUP_DATA)
3940 flags |= root->fs_info->avail_data_alloc_bits;
3941 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3942 flags |= root->fs_info->avail_system_alloc_bits;
3943 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3944 flags |= root->fs_info->avail_metadata_alloc_bits;
3945 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3947 return btrfs_reduce_alloc_profile(root, flags);
3950 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3956 flags = BTRFS_BLOCK_GROUP_DATA;
3957 else if (root == root->fs_info->chunk_root)
3958 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3960 flags = BTRFS_BLOCK_GROUP_METADATA;
3962 ret = get_alloc_profile(root, flags);
3967 * This will check the space that the inode allocates from to make sure we have
3968 * enough space for bytes.
3970 int btrfs_check_data_free_space(struct inode *inode, u64 bytes, u64 write_bytes)
3972 struct btrfs_space_info *data_sinfo;
3973 struct btrfs_root *root = BTRFS_I(inode)->root;
3974 struct btrfs_fs_info *fs_info = root->fs_info;
3977 int need_commit = 2;
3978 int have_pinned_space;
3980 /* make sure bytes are sectorsize aligned */
3981 bytes = ALIGN(bytes, root->sectorsize);
3983 if (btrfs_is_free_space_inode(inode)) {
3985 ASSERT(current->journal_info);
3988 data_sinfo = fs_info->data_sinfo;
3993 /* make sure we have enough space to handle the data first */
3994 spin_lock(&data_sinfo->lock);
3995 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3996 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3997 data_sinfo->bytes_may_use;
3999 if (used + bytes > data_sinfo->total_bytes) {
4000 struct btrfs_trans_handle *trans;
4003 * if we don't have enough free bytes in this space then we need
4004 * to alloc a new chunk.
4006 if (!data_sinfo->full) {
4009 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
4010 spin_unlock(&data_sinfo->lock);
4012 alloc_target = btrfs_get_alloc_profile(root, 1);
4014 * It is ugly that we don't call nolock join
4015 * transaction for the free space inode case here.
4016 * But it is safe because we only do the data space
4017 * reservation for the free space cache in the
4018 * transaction context, the common join transaction
4019 * just increase the counter of the current transaction
4020 * handler, doesn't try to acquire the trans_lock of
4023 trans = btrfs_join_transaction(root);
4025 return PTR_ERR(trans);
4027 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4029 CHUNK_ALLOC_NO_FORCE);
4030 btrfs_end_transaction(trans, root);
4035 have_pinned_space = 1;
4041 data_sinfo = fs_info->data_sinfo;
4047 * If we don't have enough pinned space to deal with this
4048 * allocation, and no removed chunk in current transaction,
4049 * don't bother committing the transaction.
4051 have_pinned_space = percpu_counter_compare(
4052 &data_sinfo->total_bytes_pinned,
4053 used + bytes - data_sinfo->total_bytes);
4054 spin_unlock(&data_sinfo->lock);
4056 /* commit the current transaction and try again */
4059 !atomic_read(&root->fs_info->open_ioctl_trans)) {
4062 if (need_commit > 0)
4063 btrfs_wait_ordered_roots(fs_info, -1);
4065 trans = btrfs_join_transaction(root);
4067 return PTR_ERR(trans);
4068 if (have_pinned_space >= 0 ||
4069 trans->transaction->have_free_bgs ||
4071 ret = btrfs_commit_transaction(trans, root);
4075 * make sure that all running delayed iput are
4078 down_write(&root->fs_info->delayed_iput_sem);
4079 up_write(&root->fs_info->delayed_iput_sem);
4082 btrfs_end_transaction(trans, root);
4086 trace_btrfs_space_reservation(root->fs_info,
4087 "space_info:enospc",
4088 data_sinfo->flags, bytes, 1);
4091 ret = btrfs_qgroup_reserve(root, write_bytes);
4094 data_sinfo->bytes_may_use += bytes;
4095 trace_btrfs_space_reservation(root->fs_info, "space_info",
4096 data_sinfo->flags, bytes, 1);
4098 spin_unlock(&data_sinfo->lock);
4104 * Called if we need to clear a data reservation for this inode.
4106 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
4108 struct btrfs_root *root = BTRFS_I(inode)->root;
4109 struct btrfs_space_info *data_sinfo;
4111 /* make sure bytes are sectorsize aligned */
4112 bytes = ALIGN(bytes, root->sectorsize);
4114 data_sinfo = root->fs_info->data_sinfo;
4115 spin_lock(&data_sinfo->lock);
4116 WARN_ON(data_sinfo->bytes_may_use < bytes);
4117 data_sinfo->bytes_may_use -= bytes;
4118 trace_btrfs_space_reservation(root->fs_info, "space_info",
4119 data_sinfo->flags, bytes, 0);
4120 spin_unlock(&data_sinfo->lock);
4123 static void force_metadata_allocation(struct btrfs_fs_info *info)
4125 struct list_head *head = &info->space_info;
4126 struct btrfs_space_info *found;
4129 list_for_each_entry_rcu(found, head, list) {
4130 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
4131 found->force_alloc = CHUNK_ALLOC_FORCE;
4136 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
4138 return (global->size << 1);
4141 static int should_alloc_chunk(struct btrfs_root *root,
4142 struct btrfs_space_info *sinfo, int force)
4144 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4145 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
4146 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
4149 if (force == CHUNK_ALLOC_FORCE)
4153 * We need to take into account the global rsv because for all intents
4154 * and purposes it's used space. Don't worry about locking the
4155 * global_rsv, it doesn't change except when the transaction commits.
4157 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
4158 num_allocated += calc_global_rsv_need_space(global_rsv);
4161 * in limited mode, we want to have some free space up to
4162 * about 1% of the FS size.
4164 if (force == CHUNK_ALLOC_LIMITED) {
4165 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
4166 thresh = max_t(u64, 64 * 1024 * 1024,
4167 div_factor_fine(thresh, 1));
4169 if (num_bytes - num_allocated < thresh)
4173 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
4178 static u64 get_profile_num_devs(struct btrfs_root *root, u64 type)
4182 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
4183 BTRFS_BLOCK_GROUP_RAID0 |
4184 BTRFS_BLOCK_GROUP_RAID5 |
4185 BTRFS_BLOCK_GROUP_RAID6))
4186 num_dev = root->fs_info->fs_devices->rw_devices;
4187 else if (type & BTRFS_BLOCK_GROUP_RAID1)
4190 num_dev = 1; /* DUP or single */
4196 * If @is_allocation is true, reserve space in the system space info necessary
4197 * for allocating a chunk, otherwise if it's false, reserve space necessary for
4200 void check_system_chunk(struct btrfs_trans_handle *trans,
4201 struct btrfs_root *root,
4204 struct btrfs_space_info *info;
4211 * Needed because we can end up allocating a system chunk and for an
4212 * atomic and race free space reservation in the chunk block reserve.
4214 ASSERT(mutex_is_locked(&root->fs_info->chunk_mutex));
4216 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4217 spin_lock(&info->lock);
4218 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
4219 info->bytes_reserved - info->bytes_readonly -
4220 info->bytes_may_use;
4221 spin_unlock(&info->lock);
4223 num_devs = get_profile_num_devs(root, type);
4225 /* num_devs device items to update and 1 chunk item to add or remove */
4226 thresh = btrfs_calc_trunc_metadata_size(root, num_devs) +
4227 btrfs_calc_trans_metadata_size(root, 1);
4229 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
4230 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
4231 left, thresh, type);
4232 dump_space_info(info, 0, 0);
4235 if (left < thresh) {
4238 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
4240 * Ignore failure to create system chunk. We might end up not
4241 * needing it, as we might not need to COW all nodes/leafs from
4242 * the paths we visit in the chunk tree (they were already COWed
4243 * or created in the current transaction for example).
4245 ret = btrfs_alloc_chunk(trans, root, flags);
4249 ret = btrfs_block_rsv_add(root->fs_info->chunk_root,
4250 &root->fs_info->chunk_block_rsv,
4251 thresh, BTRFS_RESERVE_NO_FLUSH);
4253 trans->chunk_bytes_reserved += thresh;
4257 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
4258 struct btrfs_root *extent_root, u64 flags, int force)
4260 struct btrfs_space_info *space_info;
4261 struct btrfs_fs_info *fs_info = extent_root->fs_info;
4262 int wait_for_alloc = 0;
4265 /* Don't re-enter if we're already allocating a chunk */
4266 if (trans->allocating_chunk)
4269 space_info = __find_space_info(extent_root->fs_info, flags);
4271 ret = update_space_info(extent_root->fs_info, flags,
4273 BUG_ON(ret); /* -ENOMEM */
4275 BUG_ON(!space_info); /* Logic error */
4278 spin_lock(&space_info->lock);
4279 if (force < space_info->force_alloc)
4280 force = space_info->force_alloc;
4281 if (space_info->full) {
4282 if (should_alloc_chunk(extent_root, space_info, force))
4286 spin_unlock(&space_info->lock);
4290 if (!should_alloc_chunk(extent_root, space_info, force)) {
4291 spin_unlock(&space_info->lock);
4293 } else if (space_info->chunk_alloc) {
4296 space_info->chunk_alloc = 1;
4299 spin_unlock(&space_info->lock);
4301 mutex_lock(&fs_info->chunk_mutex);
4304 * The chunk_mutex is held throughout the entirety of a chunk
4305 * allocation, so once we've acquired the chunk_mutex we know that the
4306 * other guy is done and we need to recheck and see if we should
4309 if (wait_for_alloc) {
4310 mutex_unlock(&fs_info->chunk_mutex);
4315 trans->allocating_chunk = true;
4318 * If we have mixed data/metadata chunks we want to make sure we keep
4319 * allocating mixed chunks instead of individual chunks.
4321 if (btrfs_mixed_space_info(space_info))
4322 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
4325 * if we're doing a data chunk, go ahead and make sure that
4326 * we keep a reasonable number of metadata chunks allocated in the
4329 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
4330 fs_info->data_chunk_allocations++;
4331 if (!(fs_info->data_chunk_allocations %
4332 fs_info->metadata_ratio))
4333 force_metadata_allocation(fs_info);
4337 * Check if we have enough space in SYSTEM chunk because we may need
4338 * to update devices.
4340 check_system_chunk(trans, extent_root, flags);
4342 ret = btrfs_alloc_chunk(trans, extent_root, flags);
4343 trans->allocating_chunk = false;
4345 spin_lock(&space_info->lock);
4346 if (ret < 0 && ret != -ENOSPC)
4349 space_info->full = 1;
4353 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4355 space_info->chunk_alloc = 0;
4356 spin_unlock(&space_info->lock);
4357 mutex_unlock(&fs_info->chunk_mutex);
4359 * When we allocate a new chunk we reserve space in the chunk block
4360 * reserve to make sure we can COW nodes/leafs in the chunk tree or
4361 * add new nodes/leafs to it if we end up needing to do it when
4362 * inserting the chunk item and updating device items as part of the
4363 * second phase of chunk allocation, performed by
4364 * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a
4365 * large number of new block groups to create in our transaction
4366 * handle's new_bgs list to avoid exhausting the chunk block reserve
4367 * in extreme cases - like having a single transaction create many new
4368 * block groups when starting to write out the free space caches of all
4369 * the block groups that were made dirty during the lifetime of the
4372 if (trans->can_flush_pending_bgs &&
4373 trans->chunk_bytes_reserved >= (2 * 1024 * 1024ull)) {
4374 btrfs_create_pending_block_groups(trans, trans->root);
4375 btrfs_trans_release_chunk_metadata(trans);
4380 static int can_overcommit(struct btrfs_root *root,
4381 struct btrfs_space_info *space_info, u64 bytes,
4382 enum btrfs_reserve_flush_enum flush)
4384 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4385 u64 profile = btrfs_get_alloc_profile(root, 0);
4390 used = space_info->bytes_used + space_info->bytes_reserved +
4391 space_info->bytes_pinned + space_info->bytes_readonly;
4394 * We only want to allow over committing if we have lots of actual space
4395 * free, but if we don't have enough space to handle the global reserve
4396 * space then we could end up having a real enospc problem when trying
4397 * to allocate a chunk or some other such important allocation.
4399 spin_lock(&global_rsv->lock);
4400 space_size = calc_global_rsv_need_space(global_rsv);
4401 spin_unlock(&global_rsv->lock);
4402 if (used + space_size >= space_info->total_bytes)
4405 used += space_info->bytes_may_use;
4407 spin_lock(&root->fs_info->free_chunk_lock);
4408 avail = root->fs_info->free_chunk_space;
4409 spin_unlock(&root->fs_info->free_chunk_lock);
4412 * If we have dup, raid1 or raid10 then only half of the free
4413 * space is actually useable. For raid56, the space info used
4414 * doesn't include the parity drive, so we don't have to
4417 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4418 BTRFS_BLOCK_GROUP_RAID1 |
4419 BTRFS_BLOCK_GROUP_RAID10))
4423 * If we aren't flushing all things, let us overcommit up to
4424 * 1/2th of the space. If we can flush, don't let us overcommit
4425 * too much, let it overcommit up to 1/8 of the space.
4427 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4432 if (used + bytes < space_info->total_bytes + avail)
4437 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4438 unsigned long nr_pages, int nr_items)
4440 struct super_block *sb = root->fs_info->sb;
4442 if (down_read_trylock(&sb->s_umount)) {
4443 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4444 up_read(&sb->s_umount);
4447 * We needn't worry the filesystem going from r/w to r/o though
4448 * we don't acquire ->s_umount mutex, because the filesystem
4449 * should guarantee the delalloc inodes list be empty after
4450 * the filesystem is readonly(all dirty pages are written to
4453 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4454 if (!current->journal_info)
4455 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4459 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4464 bytes = btrfs_calc_trans_metadata_size(root, 1);
4465 nr = (int)div64_u64(to_reclaim, bytes);
4471 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4474 * shrink metadata reservation for delalloc
4476 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4479 struct btrfs_block_rsv *block_rsv;
4480 struct btrfs_space_info *space_info;
4481 struct btrfs_trans_handle *trans;
4485 unsigned long nr_pages;
4488 enum btrfs_reserve_flush_enum flush;
4490 /* Calc the number of the pages we need flush for space reservation */
4491 items = calc_reclaim_items_nr(root, to_reclaim);
4492 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4494 trans = (struct btrfs_trans_handle *)current->journal_info;
4495 block_rsv = &root->fs_info->delalloc_block_rsv;
4496 space_info = block_rsv->space_info;
4498 delalloc_bytes = percpu_counter_sum_positive(
4499 &root->fs_info->delalloc_bytes);
4500 if (delalloc_bytes == 0) {
4504 btrfs_wait_ordered_roots(root->fs_info, items);
4509 while (delalloc_bytes && loops < 3) {
4510 max_reclaim = min(delalloc_bytes, to_reclaim);
4511 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4512 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4514 * We need to wait for the async pages to actually start before
4517 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4521 if (max_reclaim <= nr_pages)
4524 max_reclaim -= nr_pages;
4526 wait_event(root->fs_info->async_submit_wait,
4527 atomic_read(&root->fs_info->async_delalloc_pages) <=
4531 flush = BTRFS_RESERVE_FLUSH_ALL;
4533 flush = BTRFS_RESERVE_NO_FLUSH;
4534 spin_lock(&space_info->lock);
4535 if (can_overcommit(root, space_info, orig, flush)) {
4536 spin_unlock(&space_info->lock);
4539 spin_unlock(&space_info->lock);
4542 if (wait_ordered && !trans) {
4543 btrfs_wait_ordered_roots(root->fs_info, items);
4545 time_left = schedule_timeout_killable(1);
4549 delalloc_bytes = percpu_counter_sum_positive(
4550 &root->fs_info->delalloc_bytes);
4555 * maybe_commit_transaction - possibly commit the transaction if its ok to
4556 * @root - the root we're allocating for
4557 * @bytes - the number of bytes we want to reserve
4558 * @force - force the commit
4560 * This will check to make sure that committing the transaction will actually
4561 * get us somewhere and then commit the transaction if it does. Otherwise it
4562 * will return -ENOSPC.
4564 static int may_commit_transaction(struct btrfs_root *root,
4565 struct btrfs_space_info *space_info,
4566 u64 bytes, int force)
4568 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4569 struct btrfs_trans_handle *trans;
4571 trans = (struct btrfs_trans_handle *)current->journal_info;
4578 /* See if there is enough pinned space to make this reservation */
4579 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4584 * See if there is some space in the delayed insertion reservation for
4587 if (space_info != delayed_rsv->space_info)
4590 spin_lock(&delayed_rsv->lock);
4591 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4592 bytes - delayed_rsv->size) >= 0) {
4593 spin_unlock(&delayed_rsv->lock);
4596 spin_unlock(&delayed_rsv->lock);
4599 trans = btrfs_join_transaction(root);
4603 return btrfs_commit_transaction(trans, root);
4607 FLUSH_DELAYED_ITEMS_NR = 1,
4608 FLUSH_DELAYED_ITEMS = 2,
4610 FLUSH_DELALLOC_WAIT = 4,
4615 static int flush_space(struct btrfs_root *root,
4616 struct btrfs_space_info *space_info, u64 num_bytes,
4617 u64 orig_bytes, int state)
4619 struct btrfs_trans_handle *trans;
4624 case FLUSH_DELAYED_ITEMS_NR:
4625 case FLUSH_DELAYED_ITEMS:
4626 if (state == FLUSH_DELAYED_ITEMS_NR)
4627 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4631 trans = btrfs_join_transaction(root);
4632 if (IS_ERR(trans)) {
4633 ret = PTR_ERR(trans);
4636 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4637 btrfs_end_transaction(trans, root);
4639 case FLUSH_DELALLOC:
4640 case FLUSH_DELALLOC_WAIT:
4641 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4642 state == FLUSH_DELALLOC_WAIT);
4645 trans = btrfs_join_transaction(root);
4646 if (IS_ERR(trans)) {
4647 ret = PTR_ERR(trans);
4650 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4651 btrfs_get_alloc_profile(root, 0),
4652 CHUNK_ALLOC_NO_FORCE);
4653 btrfs_end_transaction(trans, root);
4658 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4669 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4670 struct btrfs_space_info *space_info)
4676 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4678 spin_lock(&space_info->lock);
4679 if (can_overcommit(root, space_info, to_reclaim,
4680 BTRFS_RESERVE_FLUSH_ALL)) {
4685 used = space_info->bytes_used + space_info->bytes_reserved +
4686 space_info->bytes_pinned + space_info->bytes_readonly +
4687 space_info->bytes_may_use;
4688 if (can_overcommit(root, space_info, 1024 * 1024,
4689 BTRFS_RESERVE_FLUSH_ALL))
4690 expected = div_factor_fine(space_info->total_bytes, 95);
4692 expected = div_factor_fine(space_info->total_bytes, 90);
4694 if (used > expected)
4695 to_reclaim = used - expected;
4698 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4699 space_info->bytes_reserved);
4701 spin_unlock(&space_info->lock);
4706 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4707 struct btrfs_fs_info *fs_info, u64 used)
4709 u64 thresh = div_factor_fine(space_info->total_bytes, 98);
4711 /* If we're just plain full then async reclaim just slows us down. */
4712 if (space_info->bytes_used >= thresh)
4715 return (used >= thresh && !btrfs_fs_closing(fs_info) &&
4716 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4719 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4720 struct btrfs_fs_info *fs_info,
4725 spin_lock(&space_info->lock);
4727 * We run out of space and have not got any free space via flush_space,
4728 * so don't bother doing async reclaim.
4730 if (flush_state > COMMIT_TRANS && space_info->full) {
4731 spin_unlock(&space_info->lock);
4735 used = space_info->bytes_used + space_info->bytes_reserved +
4736 space_info->bytes_pinned + space_info->bytes_readonly +
4737 space_info->bytes_may_use;
4738 if (need_do_async_reclaim(space_info, fs_info, used)) {
4739 spin_unlock(&space_info->lock);
4742 spin_unlock(&space_info->lock);
4747 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4749 struct btrfs_fs_info *fs_info;
4750 struct btrfs_space_info *space_info;
4754 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4755 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4757 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4762 flush_state = FLUSH_DELAYED_ITEMS_NR;
4764 flush_space(fs_info->fs_root, space_info, to_reclaim,
4765 to_reclaim, flush_state);
4767 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4770 } while (flush_state < COMMIT_TRANS);
4773 void btrfs_init_async_reclaim_work(struct work_struct *work)
4775 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4779 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4780 * @root - the root we're allocating for
4781 * @block_rsv - the block_rsv we're allocating for
4782 * @orig_bytes - the number of bytes we want
4783 * @flush - whether or not we can flush to make our reservation
4785 * This will reserve orgi_bytes number of bytes from the space info associated
4786 * with the block_rsv. If there is not enough space it will make an attempt to
4787 * flush out space to make room. It will do this by flushing delalloc if
4788 * possible or committing the transaction. If flush is 0 then no attempts to
4789 * regain reservations will be made and this will fail if there is not enough
4792 static int reserve_metadata_bytes(struct btrfs_root *root,
4793 struct btrfs_block_rsv *block_rsv,
4795 enum btrfs_reserve_flush_enum flush)
4797 struct btrfs_space_info *space_info = block_rsv->space_info;
4799 u64 num_bytes = orig_bytes;
4800 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4802 bool flushing = false;
4806 spin_lock(&space_info->lock);
4808 * We only want to wait if somebody other than us is flushing and we
4809 * are actually allowed to flush all things.
4811 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4812 space_info->flush) {
4813 spin_unlock(&space_info->lock);
4815 * If we have a trans handle we can't wait because the flusher
4816 * may have to commit the transaction, which would mean we would
4817 * deadlock since we are waiting for the flusher to finish, but
4818 * hold the current transaction open.
4820 if (current->journal_info)
4822 ret = wait_event_killable(space_info->wait, !space_info->flush);
4823 /* Must have been killed, return */
4827 spin_lock(&space_info->lock);
4831 used = space_info->bytes_used + space_info->bytes_reserved +
4832 space_info->bytes_pinned + space_info->bytes_readonly +
4833 space_info->bytes_may_use;
4836 * The idea here is that we've not already over-reserved the block group
4837 * then we can go ahead and save our reservation first and then start
4838 * flushing if we need to. Otherwise if we've already overcommitted
4839 * lets start flushing stuff first and then come back and try to make
4842 if (used <= space_info->total_bytes) {
4843 if (used + orig_bytes <= space_info->total_bytes) {
4844 space_info->bytes_may_use += orig_bytes;
4845 trace_btrfs_space_reservation(root->fs_info,
4846 "space_info", space_info->flags, orig_bytes, 1);
4850 * Ok set num_bytes to orig_bytes since we aren't
4851 * overocmmitted, this way we only try and reclaim what
4854 num_bytes = orig_bytes;
4858 * Ok we're over committed, set num_bytes to the overcommitted
4859 * amount plus the amount of bytes that we need for this
4862 num_bytes = used - space_info->total_bytes +
4866 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4867 space_info->bytes_may_use += orig_bytes;
4868 trace_btrfs_space_reservation(root->fs_info, "space_info",
4869 space_info->flags, orig_bytes,
4875 * Couldn't make our reservation, save our place so while we're trying
4876 * to reclaim space we can actually use it instead of somebody else
4877 * stealing it from us.
4879 * We make the other tasks wait for the flush only when we can flush
4882 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4884 space_info->flush = 1;
4885 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4888 * We will do the space reservation dance during log replay,
4889 * which means we won't have fs_info->fs_root set, so don't do
4890 * the async reclaim as we will panic.
4892 if (!root->fs_info->log_root_recovering &&
4893 need_do_async_reclaim(space_info, root->fs_info, used) &&
4894 !work_busy(&root->fs_info->async_reclaim_work))
4895 queue_work(system_unbound_wq,
4896 &root->fs_info->async_reclaim_work);
4898 spin_unlock(&space_info->lock);
4900 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4903 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4908 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4909 * would happen. So skip delalloc flush.
4911 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4912 (flush_state == FLUSH_DELALLOC ||
4913 flush_state == FLUSH_DELALLOC_WAIT))
4914 flush_state = ALLOC_CHUNK;
4918 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4919 flush_state < COMMIT_TRANS)
4921 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4922 flush_state <= COMMIT_TRANS)
4926 if (ret == -ENOSPC &&
4927 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4928 struct btrfs_block_rsv *global_rsv =
4929 &root->fs_info->global_block_rsv;
4931 if (block_rsv != global_rsv &&
4932 !block_rsv_use_bytes(global_rsv, orig_bytes))
4936 trace_btrfs_space_reservation(root->fs_info,
4937 "space_info:enospc",
4938 space_info->flags, orig_bytes, 1);
4940 spin_lock(&space_info->lock);
4941 space_info->flush = 0;
4942 wake_up_all(&space_info->wait);
4943 spin_unlock(&space_info->lock);
4948 static struct btrfs_block_rsv *get_block_rsv(
4949 const struct btrfs_trans_handle *trans,
4950 const struct btrfs_root *root)
4952 struct btrfs_block_rsv *block_rsv = NULL;
4954 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4955 (root == root->fs_info->csum_root && trans->adding_csums) ||
4956 (root == root->fs_info->uuid_root))
4957 block_rsv = trans->block_rsv;
4960 block_rsv = root->block_rsv;
4963 block_rsv = &root->fs_info->empty_block_rsv;
4968 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4972 spin_lock(&block_rsv->lock);
4973 if (block_rsv->reserved >= num_bytes) {
4974 block_rsv->reserved -= num_bytes;
4975 if (block_rsv->reserved < block_rsv->size)
4976 block_rsv->full = 0;
4979 spin_unlock(&block_rsv->lock);
4983 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4984 u64 num_bytes, int update_size)
4986 spin_lock(&block_rsv->lock);
4987 block_rsv->reserved += num_bytes;
4989 block_rsv->size += num_bytes;
4990 else if (block_rsv->reserved >= block_rsv->size)
4991 block_rsv->full = 1;
4992 spin_unlock(&block_rsv->lock);
4995 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4996 struct btrfs_block_rsv *dest, u64 num_bytes,
4999 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5002 if (global_rsv->space_info != dest->space_info)
5005 spin_lock(&global_rsv->lock);
5006 min_bytes = div_factor(global_rsv->size, min_factor);
5007 if (global_rsv->reserved < min_bytes + num_bytes) {
5008 spin_unlock(&global_rsv->lock);
5011 global_rsv->reserved -= num_bytes;
5012 if (global_rsv->reserved < global_rsv->size)
5013 global_rsv->full = 0;
5014 spin_unlock(&global_rsv->lock);
5016 block_rsv_add_bytes(dest, num_bytes, 1);
5020 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
5021 struct btrfs_block_rsv *block_rsv,
5022 struct btrfs_block_rsv *dest, u64 num_bytes)
5024 struct btrfs_space_info *space_info = block_rsv->space_info;
5026 spin_lock(&block_rsv->lock);
5027 if (num_bytes == (u64)-1)
5028 num_bytes = block_rsv->size;
5029 block_rsv->size -= num_bytes;
5030 if (block_rsv->reserved >= block_rsv->size) {
5031 num_bytes = block_rsv->reserved - block_rsv->size;
5032 block_rsv->reserved = block_rsv->size;
5033 block_rsv->full = 1;
5037 spin_unlock(&block_rsv->lock);
5039 if (num_bytes > 0) {
5041 spin_lock(&dest->lock);
5045 bytes_to_add = dest->size - dest->reserved;
5046 bytes_to_add = min(num_bytes, bytes_to_add);
5047 dest->reserved += bytes_to_add;
5048 if (dest->reserved >= dest->size)
5050 num_bytes -= bytes_to_add;
5052 spin_unlock(&dest->lock);
5055 spin_lock(&space_info->lock);
5056 space_info->bytes_may_use -= num_bytes;
5057 trace_btrfs_space_reservation(fs_info, "space_info",
5058 space_info->flags, num_bytes, 0);
5059 spin_unlock(&space_info->lock);
5064 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
5065 struct btrfs_block_rsv *dst, u64 num_bytes)
5069 ret = block_rsv_use_bytes(src, num_bytes);
5073 block_rsv_add_bytes(dst, num_bytes, 1);
5077 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
5079 memset(rsv, 0, sizeof(*rsv));
5080 spin_lock_init(&rsv->lock);
5084 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
5085 unsigned short type)
5087 struct btrfs_block_rsv *block_rsv;
5088 struct btrfs_fs_info *fs_info = root->fs_info;
5090 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
5094 btrfs_init_block_rsv(block_rsv, type);
5095 block_rsv->space_info = __find_space_info(fs_info,
5096 BTRFS_BLOCK_GROUP_METADATA);
5100 void btrfs_free_block_rsv(struct btrfs_root *root,
5101 struct btrfs_block_rsv *rsv)
5105 btrfs_block_rsv_release(root, rsv, (u64)-1);
5109 void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv)
5114 int btrfs_block_rsv_add(struct btrfs_root *root,
5115 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
5116 enum btrfs_reserve_flush_enum flush)
5123 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5125 block_rsv_add_bytes(block_rsv, num_bytes, 1);
5132 int btrfs_block_rsv_check(struct btrfs_root *root,
5133 struct btrfs_block_rsv *block_rsv, int min_factor)
5141 spin_lock(&block_rsv->lock);
5142 num_bytes = div_factor(block_rsv->size, min_factor);
5143 if (block_rsv->reserved >= num_bytes)
5145 spin_unlock(&block_rsv->lock);
5150 int btrfs_block_rsv_refill(struct btrfs_root *root,
5151 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
5152 enum btrfs_reserve_flush_enum flush)
5160 spin_lock(&block_rsv->lock);
5161 num_bytes = min_reserved;
5162 if (block_rsv->reserved >= num_bytes)
5165 num_bytes -= block_rsv->reserved;
5166 spin_unlock(&block_rsv->lock);
5171 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
5173 block_rsv_add_bytes(block_rsv, num_bytes, 0);
5180 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
5181 struct btrfs_block_rsv *dst_rsv,
5184 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5187 void btrfs_block_rsv_release(struct btrfs_root *root,
5188 struct btrfs_block_rsv *block_rsv,
5191 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5192 if (global_rsv == block_rsv ||
5193 block_rsv->space_info != global_rsv->space_info)
5195 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
5200 * helper to calculate size of global block reservation.
5201 * the desired value is sum of space used by extent tree,
5202 * checksum tree and root tree
5204 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
5206 struct btrfs_space_info *sinfo;
5210 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
5212 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
5213 spin_lock(&sinfo->lock);
5214 data_used = sinfo->bytes_used;
5215 spin_unlock(&sinfo->lock);
5217 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5218 spin_lock(&sinfo->lock);
5219 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
5221 meta_used = sinfo->bytes_used;
5222 spin_unlock(&sinfo->lock);
5224 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
5226 num_bytes += div_u64(data_used + meta_used, 50);
5228 if (num_bytes * 3 > meta_used)
5229 num_bytes = div_u64(meta_used, 3);
5231 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
5234 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
5236 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
5237 struct btrfs_space_info *sinfo = block_rsv->space_info;
5240 num_bytes = calc_global_metadata_size(fs_info);
5242 spin_lock(&sinfo->lock);
5243 spin_lock(&block_rsv->lock);
5245 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
5247 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
5248 sinfo->bytes_reserved + sinfo->bytes_readonly +
5249 sinfo->bytes_may_use;
5251 if (sinfo->total_bytes > num_bytes) {
5252 num_bytes = sinfo->total_bytes - num_bytes;
5253 block_rsv->reserved += num_bytes;
5254 sinfo->bytes_may_use += num_bytes;
5255 trace_btrfs_space_reservation(fs_info, "space_info",
5256 sinfo->flags, num_bytes, 1);
5259 if (block_rsv->reserved >= block_rsv->size) {
5260 num_bytes = block_rsv->reserved - block_rsv->size;
5261 sinfo->bytes_may_use -= num_bytes;
5262 trace_btrfs_space_reservation(fs_info, "space_info",
5263 sinfo->flags, num_bytes, 0);
5264 block_rsv->reserved = block_rsv->size;
5265 block_rsv->full = 1;
5268 spin_unlock(&block_rsv->lock);
5269 spin_unlock(&sinfo->lock);
5272 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
5274 struct btrfs_space_info *space_info;
5276 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
5277 fs_info->chunk_block_rsv.space_info = space_info;
5279 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
5280 fs_info->global_block_rsv.space_info = space_info;
5281 fs_info->delalloc_block_rsv.space_info = space_info;
5282 fs_info->trans_block_rsv.space_info = space_info;
5283 fs_info->empty_block_rsv.space_info = space_info;
5284 fs_info->delayed_block_rsv.space_info = space_info;
5286 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
5287 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
5288 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
5289 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
5290 if (fs_info->quota_root)
5291 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
5292 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
5294 update_global_block_rsv(fs_info);
5297 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
5299 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
5301 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
5302 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
5303 WARN_ON(fs_info->trans_block_rsv.size > 0);
5304 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
5305 WARN_ON(fs_info->chunk_block_rsv.size > 0);
5306 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
5307 WARN_ON(fs_info->delayed_block_rsv.size > 0);
5308 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
5311 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
5312 struct btrfs_root *root)
5314 if (!trans->block_rsv)
5317 if (!trans->bytes_reserved)
5320 trace_btrfs_space_reservation(root->fs_info, "transaction",
5321 trans->transid, trans->bytes_reserved, 0);
5322 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
5323 trans->bytes_reserved = 0;
5327 * To be called after all the new block groups attached to the transaction
5328 * handle have been created (btrfs_create_pending_block_groups()).
5330 void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans)
5332 struct btrfs_fs_info *fs_info = trans->root->fs_info;
5334 if (!trans->chunk_bytes_reserved)
5337 WARN_ON_ONCE(!list_empty(&trans->new_bgs));
5339 block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL,
5340 trans->chunk_bytes_reserved);
5341 trans->chunk_bytes_reserved = 0;
5344 /* Can only return 0 or -ENOSPC */
5345 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
5346 struct inode *inode)
5348 struct btrfs_root *root = BTRFS_I(inode)->root;
5349 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
5350 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
5353 * We need to hold space in order to delete our orphan item once we've
5354 * added it, so this takes the reservation so we can release it later
5355 * when we are truly done with the orphan item.
5357 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5358 trace_btrfs_space_reservation(root->fs_info, "orphan",
5359 btrfs_ino(inode), num_bytes, 1);
5360 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
5363 void btrfs_orphan_release_metadata(struct inode *inode)
5365 struct btrfs_root *root = BTRFS_I(inode)->root;
5366 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
5367 trace_btrfs_space_reservation(root->fs_info, "orphan",
5368 btrfs_ino(inode), num_bytes, 0);
5369 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
5373 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
5374 * root: the root of the parent directory
5375 * rsv: block reservation
5376 * items: the number of items that we need do reservation
5377 * qgroup_reserved: used to return the reserved size in qgroup
5379 * This function is used to reserve the space for snapshot/subvolume
5380 * creation and deletion. Those operations are different with the
5381 * common file/directory operations, they change two fs/file trees
5382 * and root tree, the number of items that the qgroup reserves is
5383 * different with the free space reservation. So we can not use
5384 * the space reseravtion mechanism in start_transaction().
5386 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
5387 struct btrfs_block_rsv *rsv,
5389 u64 *qgroup_reserved,
5390 bool use_global_rsv)
5394 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5396 if (root->fs_info->quota_enabled) {
5397 /* One for parent inode, two for dir entries */
5398 num_bytes = 3 * root->nodesize;
5399 ret = btrfs_qgroup_reserve(root, num_bytes);
5406 *qgroup_reserved = num_bytes;
5408 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5409 rsv->space_info = __find_space_info(root->fs_info,
5410 BTRFS_BLOCK_GROUP_METADATA);
5411 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5412 BTRFS_RESERVE_FLUSH_ALL);
5414 if (ret == -ENOSPC && use_global_rsv)
5415 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5418 if (*qgroup_reserved)
5419 btrfs_qgroup_free(root, *qgroup_reserved);
5425 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5426 struct btrfs_block_rsv *rsv,
5427 u64 qgroup_reserved)
5429 btrfs_block_rsv_release(root, rsv, (u64)-1);
5433 * drop_outstanding_extent - drop an outstanding extent
5434 * @inode: the inode we're dropping the extent for
5435 * @num_bytes: the number of bytes we're relaseing.
5437 * This is called when we are freeing up an outstanding extent, either called
5438 * after an error or after an extent is written. This will return the number of
5439 * reserved extents that need to be freed. This must be called with
5440 * BTRFS_I(inode)->lock held.
5442 static unsigned drop_outstanding_extent(struct inode *inode, u64 num_bytes)
5444 unsigned drop_inode_space = 0;
5445 unsigned dropped_extents = 0;
5446 unsigned num_extents = 0;
5448 num_extents = (unsigned)div64_u64(num_bytes +
5449 BTRFS_MAX_EXTENT_SIZE - 1,
5450 BTRFS_MAX_EXTENT_SIZE);
5451 ASSERT(num_extents);
5452 ASSERT(BTRFS_I(inode)->outstanding_extents >= num_extents);
5453 BTRFS_I(inode)->outstanding_extents -= num_extents;
5455 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5456 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5457 &BTRFS_I(inode)->runtime_flags))
5458 drop_inode_space = 1;
5461 * If we have more or the same amount of outsanding extents than we have
5462 * reserved then we need to leave the reserved extents count alone.
5464 if (BTRFS_I(inode)->outstanding_extents >=
5465 BTRFS_I(inode)->reserved_extents)
5466 return drop_inode_space;
5468 dropped_extents = BTRFS_I(inode)->reserved_extents -
5469 BTRFS_I(inode)->outstanding_extents;
5470 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5471 return dropped_extents + drop_inode_space;
5475 * calc_csum_metadata_size - return the amount of metada space that must be
5476 * reserved/free'd for the given bytes.
5477 * @inode: the inode we're manipulating
5478 * @num_bytes: the number of bytes in question
5479 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5481 * This adjusts the number of csum_bytes in the inode and then returns the
5482 * correct amount of metadata that must either be reserved or freed. We
5483 * calculate how many checksums we can fit into one leaf and then divide the
5484 * number of bytes that will need to be checksumed by this value to figure out
5485 * how many checksums will be required. If we are adding bytes then the number
5486 * may go up and we will return the number of additional bytes that must be
5487 * reserved. If it is going down we will return the number of bytes that must
5490 * This must be called with BTRFS_I(inode)->lock held.
5492 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5495 struct btrfs_root *root = BTRFS_I(inode)->root;
5496 u64 old_csums, num_csums;
5498 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5499 BTRFS_I(inode)->csum_bytes == 0)
5502 old_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5504 BTRFS_I(inode)->csum_bytes += num_bytes;
5506 BTRFS_I(inode)->csum_bytes -= num_bytes;
5507 num_csums = btrfs_csum_bytes_to_leaves(root, BTRFS_I(inode)->csum_bytes);
5509 /* No change, no need to reserve more */
5510 if (old_csums == num_csums)
5514 return btrfs_calc_trans_metadata_size(root,
5515 num_csums - old_csums);
5517 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5520 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5522 struct btrfs_root *root = BTRFS_I(inode)->root;
5523 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5526 unsigned nr_extents = 0;
5527 int extra_reserve = 0;
5528 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5530 bool delalloc_lock = true;
5534 /* If we are a free space inode we need to not flush since we will be in
5535 * the middle of a transaction commit. We also don't need the delalloc
5536 * mutex since we won't race with anybody. We need this mostly to make
5537 * lockdep shut its filthy mouth.
5539 if (btrfs_is_free_space_inode(inode)) {
5540 flush = BTRFS_RESERVE_NO_FLUSH;
5541 delalloc_lock = false;
5544 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5545 btrfs_transaction_in_commit(root->fs_info))
5546 schedule_timeout(1);
5549 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5551 num_bytes = ALIGN(num_bytes, root->sectorsize);
5553 spin_lock(&BTRFS_I(inode)->lock);
5554 nr_extents = (unsigned)div64_u64(num_bytes +
5555 BTRFS_MAX_EXTENT_SIZE - 1,
5556 BTRFS_MAX_EXTENT_SIZE);
5557 BTRFS_I(inode)->outstanding_extents += nr_extents;
5560 if (BTRFS_I(inode)->outstanding_extents >
5561 BTRFS_I(inode)->reserved_extents)
5562 nr_extents = BTRFS_I(inode)->outstanding_extents -
5563 BTRFS_I(inode)->reserved_extents;
5566 * Add an item to reserve for updating the inode when we complete the
5569 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5570 &BTRFS_I(inode)->runtime_flags)) {
5575 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5576 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5577 csum_bytes = BTRFS_I(inode)->csum_bytes;
5578 spin_unlock(&BTRFS_I(inode)->lock);
5580 if (root->fs_info->quota_enabled) {
5581 ret = btrfs_qgroup_reserve(root, nr_extents * root->nodesize);
5586 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5587 if (unlikely(ret)) {
5588 if (root->fs_info->quota_enabled)
5589 btrfs_qgroup_free(root, nr_extents * root->nodesize);
5593 spin_lock(&BTRFS_I(inode)->lock);
5594 if (extra_reserve) {
5595 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5596 &BTRFS_I(inode)->runtime_flags);
5599 BTRFS_I(inode)->reserved_extents += nr_extents;
5600 spin_unlock(&BTRFS_I(inode)->lock);
5603 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5606 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5607 btrfs_ino(inode), to_reserve, 1);
5608 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5613 spin_lock(&BTRFS_I(inode)->lock);
5614 dropped = drop_outstanding_extent(inode, num_bytes);
5616 * If the inodes csum_bytes is the same as the original
5617 * csum_bytes then we know we haven't raced with any free()ers
5618 * so we can just reduce our inodes csum bytes and carry on.
5620 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5621 calc_csum_metadata_size(inode, num_bytes, 0);
5623 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5627 * This is tricky, but first we need to figure out how much we
5628 * free'd from any free-ers that occured during this
5629 * reservation, so we reset ->csum_bytes to the csum_bytes
5630 * before we dropped our lock, and then call the free for the
5631 * number of bytes that were freed while we were trying our
5634 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5635 BTRFS_I(inode)->csum_bytes = csum_bytes;
5636 to_free = calc_csum_metadata_size(inode, bytes, 0);
5640 * Now we need to see how much we would have freed had we not
5641 * been making this reservation and our ->csum_bytes were not
5642 * artificially inflated.
5644 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5645 bytes = csum_bytes - orig_csum_bytes;
5646 bytes = calc_csum_metadata_size(inode, bytes, 0);
5649 * Now reset ->csum_bytes to what it should be. If bytes is
5650 * more than to_free then we would have free'd more space had we
5651 * not had an artificially high ->csum_bytes, so we need to free
5652 * the remainder. If bytes is the same or less then we don't
5653 * need to do anything, the other free-ers did the correct
5656 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5657 if (bytes > to_free)
5658 to_free = bytes - to_free;
5662 spin_unlock(&BTRFS_I(inode)->lock);
5664 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5667 btrfs_block_rsv_release(root, block_rsv, to_free);
5668 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5669 btrfs_ino(inode), to_free, 0);
5672 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5677 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5678 * @inode: the inode to release the reservation for
5679 * @num_bytes: the number of bytes we're releasing
5681 * This will release the metadata reservation for an inode. This can be called
5682 * once we complete IO for a given set of bytes to release their metadata
5685 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5687 struct btrfs_root *root = BTRFS_I(inode)->root;
5691 num_bytes = ALIGN(num_bytes, root->sectorsize);
5692 spin_lock(&BTRFS_I(inode)->lock);
5693 dropped = drop_outstanding_extent(inode, num_bytes);
5696 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5697 spin_unlock(&BTRFS_I(inode)->lock);
5699 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5701 if (btrfs_test_is_dummy_root(root))
5704 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5705 btrfs_ino(inode), to_free, 0);
5707 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5712 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5713 * @inode: inode we're writing to
5714 * @num_bytes: the number of bytes we want to allocate
5716 * This will do the following things
5718 * o reserve space in the data space info for num_bytes
5719 * o reserve space in the metadata space info based on number of outstanding
5720 * extents and how much csums will be needed
5721 * o add to the inodes ->delalloc_bytes
5722 * o add it to the fs_info's delalloc inodes list.
5724 * This will return 0 for success and -ENOSPC if there is no space left.
5726 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5730 ret = btrfs_check_data_free_space(inode, num_bytes, num_bytes);
5734 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5736 btrfs_free_reserved_data_space(inode, num_bytes);
5744 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5745 * @inode: inode we're releasing space for
5746 * @num_bytes: the number of bytes we want to free up
5748 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5749 * called in the case that we don't need the metadata AND data reservations
5750 * anymore. So if there is an error or we insert an inline extent.
5752 * This function will release the metadata space that was not used and will
5753 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5754 * list if there are no delalloc bytes left.
5756 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5758 btrfs_delalloc_release_metadata(inode, num_bytes);
5759 btrfs_free_reserved_data_space(inode, num_bytes);
5762 static int update_block_group(struct btrfs_trans_handle *trans,
5763 struct btrfs_root *root, u64 bytenr,
5764 u64 num_bytes, int alloc)
5766 struct btrfs_block_group_cache *cache = NULL;
5767 struct btrfs_fs_info *info = root->fs_info;
5768 u64 total = num_bytes;
5773 /* block accounting for super block */
5774 spin_lock(&info->delalloc_root_lock);
5775 old_val = btrfs_super_bytes_used(info->super_copy);
5777 old_val += num_bytes;
5779 old_val -= num_bytes;
5780 btrfs_set_super_bytes_used(info->super_copy, old_val);
5781 spin_unlock(&info->delalloc_root_lock);
5784 cache = btrfs_lookup_block_group(info, bytenr);
5787 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5788 BTRFS_BLOCK_GROUP_RAID1 |
5789 BTRFS_BLOCK_GROUP_RAID10))
5794 * If this block group has free space cache written out, we
5795 * need to make sure to load it if we are removing space. This
5796 * is because we need the unpinning stage to actually add the
5797 * space back to the block group, otherwise we will leak space.
5799 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5800 cache_block_group(cache, 1);
5802 byte_in_group = bytenr - cache->key.objectid;
5803 WARN_ON(byte_in_group > cache->key.offset);
5805 spin_lock(&cache->space_info->lock);
5806 spin_lock(&cache->lock);
5808 if (btrfs_test_opt(root, SPACE_CACHE) &&
5809 cache->disk_cache_state < BTRFS_DC_CLEAR)
5810 cache->disk_cache_state = BTRFS_DC_CLEAR;
5812 old_val = btrfs_block_group_used(&cache->item);
5813 num_bytes = min(total, cache->key.offset - byte_in_group);
5815 old_val += num_bytes;
5816 btrfs_set_block_group_used(&cache->item, old_val);
5817 cache->reserved -= num_bytes;
5818 cache->space_info->bytes_reserved -= num_bytes;
5819 cache->space_info->bytes_used += num_bytes;
5820 cache->space_info->disk_used += num_bytes * factor;
5821 spin_unlock(&cache->lock);
5822 spin_unlock(&cache->space_info->lock);
5824 old_val -= num_bytes;
5825 btrfs_set_block_group_used(&cache->item, old_val);
5826 cache->pinned += num_bytes;
5827 cache->space_info->bytes_pinned += num_bytes;
5828 cache->space_info->bytes_used -= num_bytes;
5829 cache->space_info->disk_used -= num_bytes * factor;
5830 spin_unlock(&cache->lock);
5831 spin_unlock(&cache->space_info->lock);
5833 set_extent_dirty(info->pinned_extents,
5834 bytenr, bytenr + num_bytes - 1,
5835 GFP_NOFS | __GFP_NOFAIL);
5837 * No longer have used bytes in this block group, queue
5841 spin_lock(&info->unused_bgs_lock);
5842 if (list_empty(&cache->bg_list)) {
5843 btrfs_get_block_group(cache);
5844 list_add_tail(&cache->bg_list,
5847 spin_unlock(&info->unused_bgs_lock);
5851 spin_lock(&trans->transaction->dirty_bgs_lock);
5852 if (list_empty(&cache->dirty_list)) {
5853 list_add_tail(&cache->dirty_list,
5854 &trans->transaction->dirty_bgs);
5855 trans->transaction->num_dirty_bgs++;
5856 btrfs_get_block_group(cache);
5858 spin_unlock(&trans->transaction->dirty_bgs_lock);
5860 btrfs_put_block_group(cache);
5862 bytenr += num_bytes;
5867 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5869 struct btrfs_block_group_cache *cache;
5872 spin_lock(&root->fs_info->block_group_cache_lock);
5873 bytenr = root->fs_info->first_logical_byte;
5874 spin_unlock(&root->fs_info->block_group_cache_lock);
5876 if (bytenr < (u64)-1)
5879 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5883 bytenr = cache->key.objectid;
5884 btrfs_put_block_group(cache);
5889 static int pin_down_extent(struct btrfs_root *root,
5890 struct btrfs_block_group_cache *cache,
5891 u64 bytenr, u64 num_bytes, int reserved)
5893 spin_lock(&cache->space_info->lock);
5894 spin_lock(&cache->lock);
5895 cache->pinned += num_bytes;
5896 cache->space_info->bytes_pinned += num_bytes;
5898 cache->reserved -= num_bytes;
5899 cache->space_info->bytes_reserved -= num_bytes;
5901 spin_unlock(&cache->lock);
5902 spin_unlock(&cache->space_info->lock);
5904 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5905 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5907 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5912 * this function must be called within transaction
5914 int btrfs_pin_extent(struct btrfs_root *root,
5915 u64 bytenr, u64 num_bytes, int reserved)
5917 struct btrfs_block_group_cache *cache;
5919 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5920 BUG_ON(!cache); /* Logic error */
5922 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5924 btrfs_put_block_group(cache);
5929 * this function must be called within transaction
5931 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5932 u64 bytenr, u64 num_bytes)
5934 struct btrfs_block_group_cache *cache;
5937 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5942 * pull in the free space cache (if any) so that our pin
5943 * removes the free space from the cache. We have load_only set
5944 * to one because the slow code to read in the free extents does check
5945 * the pinned extents.
5947 cache_block_group(cache, 1);
5949 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5951 /* remove us from the free space cache (if we're there at all) */
5952 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5953 btrfs_put_block_group(cache);
5957 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5960 struct btrfs_block_group_cache *block_group;
5961 struct btrfs_caching_control *caching_ctl;
5963 block_group = btrfs_lookup_block_group(root->fs_info, start);
5967 cache_block_group(block_group, 0);
5968 caching_ctl = get_caching_control(block_group);
5972 BUG_ON(!block_group_cache_done(block_group));
5973 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5975 mutex_lock(&caching_ctl->mutex);
5977 if (start >= caching_ctl->progress) {
5978 ret = add_excluded_extent(root, start, num_bytes);
5979 } else if (start + num_bytes <= caching_ctl->progress) {
5980 ret = btrfs_remove_free_space(block_group,
5983 num_bytes = caching_ctl->progress - start;
5984 ret = btrfs_remove_free_space(block_group,
5989 num_bytes = (start + num_bytes) -
5990 caching_ctl->progress;
5991 start = caching_ctl->progress;
5992 ret = add_excluded_extent(root, start, num_bytes);
5995 mutex_unlock(&caching_ctl->mutex);
5996 put_caching_control(caching_ctl);
5998 btrfs_put_block_group(block_group);
6002 int btrfs_exclude_logged_extents(struct btrfs_root *log,
6003 struct extent_buffer *eb)
6005 struct btrfs_file_extent_item *item;
6006 struct btrfs_key key;
6010 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
6013 for (i = 0; i < btrfs_header_nritems(eb); i++) {
6014 btrfs_item_key_to_cpu(eb, &key, i);
6015 if (key.type != BTRFS_EXTENT_DATA_KEY)
6017 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
6018 found_type = btrfs_file_extent_type(eb, item);
6019 if (found_type == BTRFS_FILE_EXTENT_INLINE)
6021 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
6023 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
6024 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
6025 __exclude_logged_extent(log, key.objectid, key.offset);
6032 * btrfs_update_reserved_bytes - update the block_group and space info counters
6033 * @cache: The cache we are manipulating
6034 * @num_bytes: The number of bytes in question
6035 * @reserve: One of the reservation enums
6036 * @delalloc: The blocks are allocated for the delalloc write
6038 * This is called by the allocator when it reserves space, or by somebody who is
6039 * freeing space that was never actually used on disk. For example if you
6040 * reserve some space for a new leaf in transaction A and before transaction A
6041 * commits you free that leaf, you call this with reserve set to 0 in order to
6042 * clear the reservation.
6044 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
6045 * ENOSPC accounting. For data we handle the reservation through clearing the
6046 * delalloc bits in the io_tree. We have to do this since we could end up
6047 * allocating less disk space for the amount of data we have reserved in the
6048 * case of compression.
6050 * If this is a reservation and the block group has become read only we cannot
6051 * make the reservation and return -EAGAIN, otherwise this function always
6054 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
6055 u64 num_bytes, int reserve, int delalloc)
6057 struct btrfs_space_info *space_info = cache->space_info;
6060 spin_lock(&space_info->lock);
6061 spin_lock(&cache->lock);
6062 if (reserve != RESERVE_FREE) {
6066 cache->reserved += num_bytes;
6067 space_info->bytes_reserved += num_bytes;
6068 if (reserve == RESERVE_ALLOC) {
6069 trace_btrfs_space_reservation(cache->fs_info,
6070 "space_info", space_info->flags,
6072 space_info->bytes_may_use -= num_bytes;
6076 cache->delalloc_bytes += num_bytes;
6080 space_info->bytes_readonly += num_bytes;
6081 cache->reserved -= num_bytes;
6082 space_info->bytes_reserved -= num_bytes;
6085 cache->delalloc_bytes -= num_bytes;
6087 spin_unlock(&cache->lock);
6088 spin_unlock(&space_info->lock);
6092 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
6093 struct btrfs_root *root)
6095 struct btrfs_fs_info *fs_info = root->fs_info;
6096 struct btrfs_caching_control *next;
6097 struct btrfs_caching_control *caching_ctl;
6098 struct btrfs_block_group_cache *cache;
6100 down_write(&fs_info->commit_root_sem);
6102 list_for_each_entry_safe(caching_ctl, next,
6103 &fs_info->caching_block_groups, list) {
6104 cache = caching_ctl->block_group;
6105 if (block_group_cache_done(cache)) {
6106 cache->last_byte_to_unpin = (u64)-1;
6107 list_del_init(&caching_ctl->list);
6108 put_caching_control(caching_ctl);
6110 cache->last_byte_to_unpin = caching_ctl->progress;
6114 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6115 fs_info->pinned_extents = &fs_info->freed_extents[1];
6117 fs_info->pinned_extents = &fs_info->freed_extents[0];
6119 up_write(&fs_info->commit_root_sem);
6121 update_global_block_rsv(fs_info);
6124 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end,
6125 const bool return_free_space)
6127 struct btrfs_fs_info *fs_info = root->fs_info;
6128 struct btrfs_block_group_cache *cache = NULL;
6129 struct btrfs_space_info *space_info;
6130 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
6134 while (start <= end) {
6137 start >= cache->key.objectid + cache->key.offset) {
6139 btrfs_put_block_group(cache);
6140 cache = btrfs_lookup_block_group(fs_info, start);
6141 BUG_ON(!cache); /* Logic error */
6144 len = cache->key.objectid + cache->key.offset - start;
6145 len = min(len, end + 1 - start);
6147 if (start < cache->last_byte_to_unpin) {
6148 len = min(len, cache->last_byte_to_unpin - start);
6149 if (return_free_space)
6150 btrfs_add_free_space(cache, start, len);
6154 space_info = cache->space_info;
6156 spin_lock(&space_info->lock);
6157 spin_lock(&cache->lock);
6158 cache->pinned -= len;
6159 space_info->bytes_pinned -= len;
6160 percpu_counter_add(&space_info->total_bytes_pinned, -len);
6162 space_info->bytes_readonly += len;
6165 spin_unlock(&cache->lock);
6166 if (!readonly && global_rsv->space_info == space_info) {
6167 spin_lock(&global_rsv->lock);
6168 if (!global_rsv->full) {
6169 len = min(len, global_rsv->size -
6170 global_rsv->reserved);
6171 global_rsv->reserved += len;
6172 space_info->bytes_may_use += len;
6173 if (global_rsv->reserved >= global_rsv->size)
6174 global_rsv->full = 1;
6176 spin_unlock(&global_rsv->lock);
6178 spin_unlock(&space_info->lock);
6182 btrfs_put_block_group(cache);
6186 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
6187 struct btrfs_root *root)
6189 struct btrfs_fs_info *fs_info = root->fs_info;
6190 struct btrfs_block_group_cache *block_group, *tmp;
6191 struct list_head *deleted_bgs;
6192 struct extent_io_tree *unpin;
6197 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
6198 unpin = &fs_info->freed_extents[1];
6200 unpin = &fs_info->freed_extents[0];
6202 while (!trans->aborted) {
6203 mutex_lock(&fs_info->unused_bg_unpin_mutex);
6204 ret = find_first_extent_bit(unpin, 0, &start, &end,
6205 EXTENT_DIRTY, NULL);
6207 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6211 if (btrfs_test_opt(root, DISCARD))
6212 ret = btrfs_discard_extent(root, start,
6213 end + 1 - start, NULL);
6215 clear_extent_dirty(unpin, start, end, GFP_NOFS);
6216 unpin_extent_range(root, start, end, true);
6217 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
6222 * Transaction is finished. We don't need the lock anymore. We
6223 * do need to clean up the block groups in case of a transaction
6226 deleted_bgs = &trans->transaction->deleted_bgs;
6227 list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) {
6231 if (!trans->aborted)
6232 ret = btrfs_discard_extent(root,
6233 block_group->key.objectid,
6234 block_group->key.offset,
6237 list_del_init(&block_group->bg_list);
6238 btrfs_put_block_group_trimming(block_group);
6239 btrfs_put_block_group(block_group);
6242 const char *errstr = btrfs_decode_error(ret);
6244 "Discard failed while removing blockgroup: errno=%d %s\n",
6252 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
6253 u64 owner, u64 root_objectid)
6255 struct btrfs_space_info *space_info;
6258 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6259 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
6260 flags = BTRFS_BLOCK_GROUP_SYSTEM;
6262 flags = BTRFS_BLOCK_GROUP_METADATA;
6264 flags = BTRFS_BLOCK_GROUP_DATA;
6267 space_info = __find_space_info(fs_info, flags);
6268 BUG_ON(!space_info); /* Logic bug */
6269 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
6273 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
6274 struct btrfs_root *root,
6275 struct btrfs_delayed_ref_node *node, u64 parent,
6276 u64 root_objectid, u64 owner_objectid,
6277 u64 owner_offset, int refs_to_drop,
6278 struct btrfs_delayed_extent_op *extent_op)
6280 struct btrfs_key key;
6281 struct btrfs_path *path;
6282 struct btrfs_fs_info *info = root->fs_info;
6283 struct btrfs_root *extent_root = info->extent_root;
6284 struct extent_buffer *leaf;
6285 struct btrfs_extent_item *ei;
6286 struct btrfs_extent_inline_ref *iref;
6289 int extent_slot = 0;
6290 int found_extent = 0;
6292 int no_quota = node->no_quota;
6295 u64 bytenr = node->bytenr;
6296 u64 num_bytes = node->num_bytes;
6298 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
6301 if (!info->quota_enabled || !is_fstree(root_objectid))
6304 path = btrfs_alloc_path();
6309 path->leave_spinning = 1;
6311 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
6312 BUG_ON(!is_data && refs_to_drop != 1);
6315 skinny_metadata = 0;
6317 ret = lookup_extent_backref(trans, extent_root, path, &iref,
6318 bytenr, num_bytes, parent,
6319 root_objectid, owner_objectid,
6322 extent_slot = path->slots[0];
6323 while (extent_slot >= 0) {
6324 btrfs_item_key_to_cpu(path->nodes[0], &key,
6326 if (key.objectid != bytenr)
6328 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
6329 key.offset == num_bytes) {
6333 if (key.type == BTRFS_METADATA_ITEM_KEY &&
6334 key.offset == owner_objectid) {
6338 if (path->slots[0] - extent_slot > 5)
6342 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6343 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
6344 if (found_extent && item_size < sizeof(*ei))
6347 if (!found_extent) {
6349 ret = remove_extent_backref(trans, extent_root, path,
6351 is_data, &last_ref);
6353 btrfs_abort_transaction(trans, extent_root, ret);
6356 btrfs_release_path(path);
6357 path->leave_spinning = 1;
6359 key.objectid = bytenr;
6360 key.type = BTRFS_EXTENT_ITEM_KEY;
6361 key.offset = num_bytes;
6363 if (!is_data && skinny_metadata) {
6364 key.type = BTRFS_METADATA_ITEM_KEY;
6365 key.offset = owner_objectid;
6368 ret = btrfs_search_slot(trans, extent_root,
6370 if (ret > 0 && skinny_metadata && path->slots[0]) {
6372 * Couldn't find our skinny metadata item,
6373 * see if we have ye olde extent item.
6376 btrfs_item_key_to_cpu(path->nodes[0], &key,
6378 if (key.objectid == bytenr &&
6379 key.type == BTRFS_EXTENT_ITEM_KEY &&
6380 key.offset == num_bytes)
6384 if (ret > 0 && skinny_metadata) {
6385 skinny_metadata = false;
6386 key.objectid = bytenr;
6387 key.type = BTRFS_EXTENT_ITEM_KEY;
6388 key.offset = num_bytes;
6389 btrfs_release_path(path);
6390 ret = btrfs_search_slot(trans, extent_root,
6395 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6398 btrfs_print_leaf(extent_root,
6402 btrfs_abort_transaction(trans, extent_root, ret);
6405 extent_slot = path->slots[0];
6407 } else if (WARN_ON(ret == -ENOENT)) {
6408 btrfs_print_leaf(extent_root, path->nodes[0]);
6410 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
6411 bytenr, parent, root_objectid, owner_objectid,
6413 btrfs_abort_transaction(trans, extent_root, ret);
6416 btrfs_abort_transaction(trans, extent_root, ret);
6420 leaf = path->nodes[0];
6421 item_size = btrfs_item_size_nr(leaf, extent_slot);
6422 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
6423 if (item_size < sizeof(*ei)) {
6424 BUG_ON(found_extent || extent_slot != path->slots[0]);
6425 ret = convert_extent_item_v0(trans, extent_root, path,
6428 btrfs_abort_transaction(trans, extent_root, ret);
6432 btrfs_release_path(path);
6433 path->leave_spinning = 1;
6435 key.objectid = bytenr;
6436 key.type = BTRFS_EXTENT_ITEM_KEY;
6437 key.offset = num_bytes;
6439 ret = btrfs_search_slot(trans, extent_root, &key, path,
6442 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
6444 btrfs_print_leaf(extent_root, path->nodes[0]);
6447 btrfs_abort_transaction(trans, extent_root, ret);
6451 extent_slot = path->slots[0];
6452 leaf = path->nodes[0];
6453 item_size = btrfs_item_size_nr(leaf, extent_slot);
6456 BUG_ON(item_size < sizeof(*ei));
6457 ei = btrfs_item_ptr(leaf, extent_slot,
6458 struct btrfs_extent_item);
6459 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6460 key.type == BTRFS_EXTENT_ITEM_KEY) {
6461 struct btrfs_tree_block_info *bi;
6462 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6463 bi = (struct btrfs_tree_block_info *)(ei + 1);
6464 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6467 refs = btrfs_extent_refs(leaf, ei);
6468 if (refs < refs_to_drop) {
6469 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6470 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6472 btrfs_abort_transaction(trans, extent_root, ret);
6475 refs -= refs_to_drop;
6479 __run_delayed_extent_op(extent_op, leaf, ei);
6481 * In the case of inline back ref, reference count will
6482 * be updated by remove_extent_backref
6485 BUG_ON(!found_extent);
6487 btrfs_set_extent_refs(leaf, ei, refs);
6488 btrfs_mark_buffer_dirty(leaf);
6491 ret = remove_extent_backref(trans, extent_root, path,
6493 is_data, &last_ref);
6495 btrfs_abort_transaction(trans, extent_root, ret);
6499 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6503 BUG_ON(is_data && refs_to_drop !=
6504 extent_data_ref_count(path, iref));
6506 BUG_ON(path->slots[0] != extent_slot);
6508 BUG_ON(path->slots[0] != extent_slot + 1);
6509 path->slots[0] = extent_slot;
6515 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6518 btrfs_abort_transaction(trans, extent_root, ret);
6521 btrfs_release_path(path);
6524 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6526 btrfs_abort_transaction(trans, extent_root, ret);
6531 ret = update_block_group(trans, root, bytenr, num_bytes, 0);
6533 btrfs_abort_transaction(trans, extent_root, ret);
6537 btrfs_release_path(path);
6540 btrfs_free_path(path);
6545 * when we free an block, it is possible (and likely) that we free the last
6546 * delayed ref for that extent as well. This searches the delayed ref tree for
6547 * a given extent, and if there are no other delayed refs to be processed, it
6548 * removes it from the tree.
6550 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6551 struct btrfs_root *root, u64 bytenr)
6553 struct btrfs_delayed_ref_head *head;
6554 struct btrfs_delayed_ref_root *delayed_refs;
6557 delayed_refs = &trans->transaction->delayed_refs;
6558 spin_lock(&delayed_refs->lock);
6559 head = btrfs_find_delayed_ref_head(trans, bytenr);
6561 goto out_delayed_unlock;
6563 spin_lock(&head->lock);
6564 if (!list_empty(&head->ref_list))
6567 if (head->extent_op) {
6568 if (!head->must_insert_reserved)
6570 btrfs_free_delayed_extent_op(head->extent_op);
6571 head->extent_op = NULL;
6575 * waiting for the lock here would deadlock. If someone else has it
6576 * locked they are already in the process of dropping it anyway
6578 if (!mutex_trylock(&head->mutex))
6582 * at this point we have a head with no other entries. Go
6583 * ahead and process it.
6585 head->node.in_tree = 0;
6586 rb_erase(&head->href_node, &delayed_refs->href_root);
6588 atomic_dec(&delayed_refs->num_entries);
6591 * we don't take a ref on the node because we're removing it from the
6592 * tree, so we just steal the ref the tree was holding.
6594 delayed_refs->num_heads--;
6595 if (head->processing == 0)
6596 delayed_refs->num_heads_ready--;
6597 head->processing = 0;
6598 spin_unlock(&head->lock);
6599 spin_unlock(&delayed_refs->lock);
6601 BUG_ON(head->extent_op);
6602 if (head->must_insert_reserved)
6605 mutex_unlock(&head->mutex);
6606 btrfs_put_delayed_ref(&head->node);
6609 spin_unlock(&head->lock);
6612 spin_unlock(&delayed_refs->lock);
6616 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6617 struct btrfs_root *root,
6618 struct extent_buffer *buf,
6619 u64 parent, int last_ref)
6624 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6625 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6626 buf->start, buf->len,
6627 parent, root->root_key.objectid,
6628 btrfs_header_level(buf),
6629 BTRFS_DROP_DELAYED_REF, NULL, 0);
6630 BUG_ON(ret); /* -ENOMEM */
6636 if (btrfs_header_generation(buf) == trans->transid) {
6637 struct btrfs_block_group_cache *cache;
6639 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6640 ret = check_ref_cleanup(trans, root, buf->start);
6645 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6647 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6648 pin_down_extent(root, cache, buf->start, buf->len, 1);
6649 btrfs_put_block_group(cache);
6653 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6655 btrfs_add_free_space(cache, buf->start, buf->len);
6656 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6657 btrfs_put_block_group(cache);
6658 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6663 add_pinned_bytes(root->fs_info, buf->len,
6664 btrfs_header_level(buf),
6665 root->root_key.objectid);
6668 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6671 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6674 /* Can return -ENOMEM */
6675 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6676 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6677 u64 owner, u64 offset, int no_quota)
6680 struct btrfs_fs_info *fs_info = root->fs_info;
6682 if (btrfs_test_is_dummy_root(root))
6685 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6688 * tree log blocks never actually go into the extent allocation
6689 * tree, just update pinning info and exit early.
6691 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6692 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6693 /* unlocks the pinned mutex */
6694 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6696 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6697 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6699 parent, root_objectid, (int)owner,
6700 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6702 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6704 parent, root_objectid, owner,
6705 offset, BTRFS_DROP_DELAYED_REF,
6712 * when we wait for progress in the block group caching, its because
6713 * our allocation attempt failed at least once. So, we must sleep
6714 * and let some progress happen before we try again.
6716 * This function will sleep at least once waiting for new free space to
6717 * show up, and then it will check the block group free space numbers
6718 * for our min num_bytes. Another option is to have it go ahead
6719 * and look in the rbtree for a free extent of a given size, but this
6722 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6723 * any of the information in this block group.
6725 static noinline void
6726 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6729 struct btrfs_caching_control *caching_ctl;
6731 caching_ctl = get_caching_control(cache);
6735 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6736 (cache->free_space_ctl->free_space >= num_bytes));
6738 put_caching_control(caching_ctl);
6742 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6744 struct btrfs_caching_control *caching_ctl;
6747 caching_ctl = get_caching_control(cache);
6749 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6751 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6752 if (cache->cached == BTRFS_CACHE_ERROR)
6754 put_caching_control(caching_ctl);
6758 int __get_raid_index(u64 flags)
6760 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6761 return BTRFS_RAID_RAID10;
6762 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6763 return BTRFS_RAID_RAID1;
6764 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6765 return BTRFS_RAID_DUP;
6766 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6767 return BTRFS_RAID_RAID0;
6768 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6769 return BTRFS_RAID_RAID5;
6770 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6771 return BTRFS_RAID_RAID6;
6773 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6776 int get_block_group_index(struct btrfs_block_group_cache *cache)
6778 return __get_raid_index(cache->flags);
6781 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6782 [BTRFS_RAID_RAID10] = "raid10",
6783 [BTRFS_RAID_RAID1] = "raid1",
6784 [BTRFS_RAID_DUP] = "dup",
6785 [BTRFS_RAID_RAID0] = "raid0",
6786 [BTRFS_RAID_SINGLE] = "single",
6787 [BTRFS_RAID_RAID5] = "raid5",
6788 [BTRFS_RAID_RAID6] = "raid6",
6791 static const char *get_raid_name(enum btrfs_raid_types type)
6793 if (type >= BTRFS_NR_RAID_TYPES)
6796 return btrfs_raid_type_names[type];
6799 enum btrfs_loop_type {
6800 LOOP_CACHING_NOWAIT = 0,
6801 LOOP_CACHING_WAIT = 1,
6802 LOOP_ALLOC_CHUNK = 2,
6803 LOOP_NO_EMPTY_SIZE = 3,
6807 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6811 down_read(&cache->data_rwsem);
6815 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6818 btrfs_get_block_group(cache);
6820 down_read(&cache->data_rwsem);
6823 static struct btrfs_block_group_cache *
6824 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6825 struct btrfs_free_cluster *cluster,
6828 struct btrfs_block_group_cache *used_bg;
6829 bool locked = false;
6831 spin_lock(&cluster->refill_lock);
6833 if (used_bg == cluster->block_group)
6836 up_read(&used_bg->data_rwsem);
6837 btrfs_put_block_group(used_bg);
6840 used_bg = cluster->block_group;
6844 if (used_bg == block_group)
6847 btrfs_get_block_group(used_bg);
6852 if (down_read_trylock(&used_bg->data_rwsem))
6855 spin_unlock(&cluster->refill_lock);
6856 down_read(&used_bg->data_rwsem);
6862 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6866 up_read(&cache->data_rwsem);
6867 btrfs_put_block_group(cache);
6871 * walks the btree of allocated extents and find a hole of a given size.
6872 * The key ins is changed to record the hole:
6873 * ins->objectid == start position
6874 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6875 * ins->offset == the size of the hole.
6876 * Any available blocks before search_start are skipped.
6878 * If there is no suitable free space, we will record the max size of
6879 * the free space extent currently.
6881 static noinline int find_free_extent(struct btrfs_root *orig_root,
6882 u64 num_bytes, u64 empty_size,
6883 u64 hint_byte, struct btrfs_key *ins,
6884 u64 flags, int delalloc)
6887 struct btrfs_root *root = orig_root->fs_info->extent_root;
6888 struct btrfs_free_cluster *last_ptr = NULL;
6889 struct btrfs_block_group_cache *block_group = NULL;
6890 u64 search_start = 0;
6891 u64 max_extent_size = 0;
6892 int empty_cluster = 2 * 1024 * 1024;
6893 struct btrfs_space_info *space_info;
6895 int index = __get_raid_index(flags);
6896 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6897 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6898 bool failed_cluster_refill = false;
6899 bool failed_alloc = false;
6900 bool use_cluster = true;
6901 bool have_caching_bg = false;
6903 WARN_ON(num_bytes < root->sectorsize);
6904 ins->type = BTRFS_EXTENT_ITEM_KEY;
6908 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6910 space_info = __find_space_info(root->fs_info, flags);
6912 btrfs_err(root->fs_info, "No space info for %llu", flags);
6917 * If the space info is for both data and metadata it means we have a
6918 * small filesystem and we can't use the clustering stuff.
6920 if (btrfs_mixed_space_info(space_info))
6921 use_cluster = false;
6923 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6924 last_ptr = &root->fs_info->meta_alloc_cluster;
6925 if (!btrfs_test_opt(root, SSD))
6926 empty_cluster = 64 * 1024;
6929 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6930 btrfs_test_opt(root, SSD)) {
6931 last_ptr = &root->fs_info->data_alloc_cluster;
6935 spin_lock(&last_ptr->lock);
6936 if (last_ptr->block_group)
6937 hint_byte = last_ptr->window_start;
6938 spin_unlock(&last_ptr->lock);
6941 search_start = max(search_start, first_logical_byte(root, 0));
6942 search_start = max(search_start, hint_byte);
6947 if (search_start == hint_byte) {
6948 block_group = btrfs_lookup_block_group(root->fs_info,
6951 * we don't want to use the block group if it doesn't match our
6952 * allocation bits, or if its not cached.
6954 * However if we are re-searching with an ideal block group
6955 * picked out then we don't care that the block group is cached.
6957 if (block_group && block_group_bits(block_group, flags) &&
6958 block_group->cached != BTRFS_CACHE_NO) {
6959 down_read(&space_info->groups_sem);
6960 if (list_empty(&block_group->list) ||
6963 * someone is removing this block group,
6964 * we can't jump into the have_block_group
6965 * target because our list pointers are not
6968 btrfs_put_block_group(block_group);
6969 up_read(&space_info->groups_sem);
6971 index = get_block_group_index(block_group);
6972 btrfs_lock_block_group(block_group, delalloc);
6973 goto have_block_group;
6975 } else if (block_group) {
6976 btrfs_put_block_group(block_group);
6980 have_caching_bg = false;
6981 down_read(&space_info->groups_sem);
6982 list_for_each_entry(block_group, &space_info->block_groups[index],
6987 btrfs_grab_block_group(block_group, delalloc);
6988 search_start = block_group->key.objectid;
6991 * this can happen if we end up cycling through all the
6992 * raid types, but we want to make sure we only allocate
6993 * for the proper type.
6995 if (!block_group_bits(block_group, flags)) {
6996 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6997 BTRFS_BLOCK_GROUP_RAID1 |
6998 BTRFS_BLOCK_GROUP_RAID5 |
6999 BTRFS_BLOCK_GROUP_RAID6 |
7000 BTRFS_BLOCK_GROUP_RAID10;
7003 * if they asked for extra copies and this block group
7004 * doesn't provide them, bail. This does allow us to
7005 * fill raid0 from raid1.
7007 if ((flags & extra) && !(block_group->flags & extra))
7012 cached = block_group_cache_done(block_group);
7013 if (unlikely(!cached)) {
7014 ret = cache_block_group(block_group, 0);
7019 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
7021 if (unlikely(block_group->ro))
7025 * Ok we want to try and use the cluster allocator, so
7029 struct btrfs_block_group_cache *used_block_group;
7030 unsigned long aligned_cluster;
7032 * the refill lock keeps out other
7033 * people trying to start a new cluster
7035 used_block_group = btrfs_lock_cluster(block_group,
7038 if (!used_block_group)
7039 goto refill_cluster;
7041 if (used_block_group != block_group &&
7042 (used_block_group->ro ||
7043 !block_group_bits(used_block_group, flags)))
7044 goto release_cluster;
7046 offset = btrfs_alloc_from_cluster(used_block_group,
7049 used_block_group->key.objectid,
7052 /* we have a block, we're done */
7053 spin_unlock(&last_ptr->refill_lock);
7054 trace_btrfs_reserve_extent_cluster(root,
7056 search_start, num_bytes);
7057 if (used_block_group != block_group) {
7058 btrfs_release_block_group(block_group,
7060 block_group = used_block_group;
7065 WARN_ON(last_ptr->block_group != used_block_group);
7067 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
7068 * set up a new clusters, so lets just skip it
7069 * and let the allocator find whatever block
7070 * it can find. If we reach this point, we
7071 * will have tried the cluster allocator
7072 * plenty of times and not have found
7073 * anything, so we are likely way too
7074 * fragmented for the clustering stuff to find
7077 * However, if the cluster is taken from the
7078 * current block group, release the cluster
7079 * first, so that we stand a better chance of
7080 * succeeding in the unclustered
7082 if (loop >= LOOP_NO_EMPTY_SIZE &&
7083 used_block_group != block_group) {
7084 spin_unlock(&last_ptr->refill_lock);
7085 btrfs_release_block_group(used_block_group,
7087 goto unclustered_alloc;
7091 * this cluster didn't work out, free it and
7094 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7096 if (used_block_group != block_group)
7097 btrfs_release_block_group(used_block_group,
7100 if (loop >= LOOP_NO_EMPTY_SIZE) {
7101 spin_unlock(&last_ptr->refill_lock);
7102 goto unclustered_alloc;
7105 aligned_cluster = max_t(unsigned long,
7106 empty_cluster + empty_size,
7107 block_group->full_stripe_len);
7109 /* allocate a cluster in this block group */
7110 ret = btrfs_find_space_cluster(root, block_group,
7111 last_ptr, search_start,
7116 * now pull our allocation out of this
7119 offset = btrfs_alloc_from_cluster(block_group,
7125 /* we found one, proceed */
7126 spin_unlock(&last_ptr->refill_lock);
7127 trace_btrfs_reserve_extent_cluster(root,
7128 block_group, search_start,
7132 } else if (!cached && loop > LOOP_CACHING_NOWAIT
7133 && !failed_cluster_refill) {
7134 spin_unlock(&last_ptr->refill_lock);
7136 failed_cluster_refill = true;
7137 wait_block_group_cache_progress(block_group,
7138 num_bytes + empty_cluster + empty_size);
7139 goto have_block_group;
7143 * at this point we either didn't find a cluster
7144 * or we weren't able to allocate a block from our
7145 * cluster. Free the cluster we've been trying
7146 * to use, and go to the next block group
7148 btrfs_return_cluster_to_free_space(NULL, last_ptr);
7149 spin_unlock(&last_ptr->refill_lock);
7154 spin_lock(&block_group->free_space_ctl->tree_lock);
7156 block_group->free_space_ctl->free_space <
7157 num_bytes + empty_cluster + empty_size) {
7158 if (block_group->free_space_ctl->free_space >
7161 block_group->free_space_ctl->free_space;
7162 spin_unlock(&block_group->free_space_ctl->tree_lock);
7165 spin_unlock(&block_group->free_space_ctl->tree_lock);
7167 offset = btrfs_find_space_for_alloc(block_group, search_start,
7168 num_bytes, empty_size,
7171 * If we didn't find a chunk, and we haven't failed on this
7172 * block group before, and this block group is in the middle of
7173 * caching and we are ok with waiting, then go ahead and wait
7174 * for progress to be made, and set failed_alloc to true.
7176 * If failed_alloc is true then we've already waited on this
7177 * block group once and should move on to the next block group.
7179 if (!offset && !failed_alloc && !cached &&
7180 loop > LOOP_CACHING_NOWAIT) {
7181 wait_block_group_cache_progress(block_group,
7182 num_bytes + empty_size);
7183 failed_alloc = true;
7184 goto have_block_group;
7185 } else if (!offset) {
7187 have_caching_bg = true;
7191 search_start = ALIGN(offset, root->stripesize);
7193 /* move on to the next group */
7194 if (search_start + num_bytes >
7195 block_group->key.objectid + block_group->key.offset) {
7196 btrfs_add_free_space(block_group, offset, num_bytes);
7200 if (offset < search_start)
7201 btrfs_add_free_space(block_group, offset,
7202 search_start - offset);
7203 BUG_ON(offset > search_start);
7205 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
7206 alloc_type, delalloc);
7207 if (ret == -EAGAIN) {
7208 btrfs_add_free_space(block_group, offset, num_bytes);
7212 /* we are all good, lets return */
7213 ins->objectid = search_start;
7214 ins->offset = num_bytes;
7216 trace_btrfs_reserve_extent(orig_root, block_group,
7217 search_start, num_bytes);
7218 btrfs_release_block_group(block_group, delalloc);
7221 failed_cluster_refill = false;
7222 failed_alloc = false;
7223 BUG_ON(index != get_block_group_index(block_group));
7224 btrfs_release_block_group(block_group, delalloc);
7226 up_read(&space_info->groups_sem);
7228 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
7231 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
7235 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
7236 * caching kthreads as we move along
7237 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
7238 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
7239 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
7242 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
7245 if (loop == LOOP_ALLOC_CHUNK) {
7246 struct btrfs_trans_handle *trans;
7249 trans = current->journal_info;
7253 trans = btrfs_join_transaction(root);
7255 if (IS_ERR(trans)) {
7256 ret = PTR_ERR(trans);
7260 ret = do_chunk_alloc(trans, root, flags,
7263 * Do not bail out on ENOSPC since we
7264 * can do more things.
7266 if (ret < 0 && ret != -ENOSPC)
7267 btrfs_abort_transaction(trans,
7272 btrfs_end_transaction(trans, root);
7277 if (loop == LOOP_NO_EMPTY_SIZE) {
7283 } else if (!ins->objectid) {
7285 } else if (ins->objectid) {
7290 ins->offset = max_extent_size;
7294 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
7295 int dump_block_groups)
7297 struct btrfs_block_group_cache *cache;
7300 spin_lock(&info->lock);
7301 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
7303 info->total_bytes - info->bytes_used - info->bytes_pinned -
7304 info->bytes_reserved - info->bytes_readonly,
7305 (info->full) ? "" : "not ");
7306 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
7307 "reserved=%llu, may_use=%llu, readonly=%llu\n",
7308 info->total_bytes, info->bytes_used, info->bytes_pinned,
7309 info->bytes_reserved, info->bytes_may_use,
7310 info->bytes_readonly);
7311 spin_unlock(&info->lock);
7313 if (!dump_block_groups)
7316 down_read(&info->groups_sem);
7318 list_for_each_entry(cache, &info->block_groups[index], list) {
7319 spin_lock(&cache->lock);
7320 printk(KERN_INFO "BTRFS: "
7321 "block group %llu has %llu bytes, "
7322 "%llu used %llu pinned %llu reserved %s\n",
7323 cache->key.objectid, cache->key.offset,
7324 btrfs_block_group_used(&cache->item), cache->pinned,
7325 cache->reserved, cache->ro ? "[readonly]" : "");
7326 btrfs_dump_free_space(cache, bytes);
7327 spin_unlock(&cache->lock);
7329 if (++index < BTRFS_NR_RAID_TYPES)
7331 up_read(&info->groups_sem);
7334 int btrfs_reserve_extent(struct btrfs_root *root,
7335 u64 num_bytes, u64 min_alloc_size,
7336 u64 empty_size, u64 hint_byte,
7337 struct btrfs_key *ins, int is_data, int delalloc)
7339 bool final_tried = false;
7343 flags = btrfs_get_alloc_profile(root, is_data);
7345 WARN_ON(num_bytes < root->sectorsize);
7346 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
7349 if (ret == -ENOSPC) {
7350 if (!final_tried && ins->offset) {
7351 num_bytes = min(num_bytes >> 1, ins->offset);
7352 num_bytes = round_down(num_bytes, root->sectorsize);
7353 num_bytes = max(num_bytes, min_alloc_size);
7354 if (num_bytes == min_alloc_size)
7357 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7358 struct btrfs_space_info *sinfo;
7360 sinfo = __find_space_info(root->fs_info, flags);
7361 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
7364 dump_space_info(sinfo, num_bytes, 1);
7371 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
7373 int pin, int delalloc)
7375 struct btrfs_block_group_cache *cache;
7378 cache = btrfs_lookup_block_group(root->fs_info, start);
7380 btrfs_err(root->fs_info, "Unable to find block group for %llu",
7386 pin_down_extent(root, cache, start, len, 1);
7388 if (btrfs_test_opt(root, DISCARD))
7389 ret = btrfs_discard_extent(root, start, len, NULL);
7390 btrfs_add_free_space(cache, start, len);
7391 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
7394 btrfs_put_block_group(cache);
7396 trace_btrfs_reserved_extent_free(root, start, len);
7401 int btrfs_free_reserved_extent(struct btrfs_root *root,
7402 u64 start, u64 len, int delalloc)
7404 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
7407 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
7410 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
7413 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7414 struct btrfs_root *root,
7415 u64 parent, u64 root_objectid,
7416 u64 flags, u64 owner, u64 offset,
7417 struct btrfs_key *ins, int ref_mod)
7420 struct btrfs_fs_info *fs_info = root->fs_info;
7421 struct btrfs_extent_item *extent_item;
7422 struct btrfs_extent_inline_ref *iref;
7423 struct btrfs_path *path;
7424 struct extent_buffer *leaf;
7429 type = BTRFS_SHARED_DATA_REF_KEY;
7431 type = BTRFS_EXTENT_DATA_REF_KEY;
7433 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7435 path = btrfs_alloc_path();
7439 path->leave_spinning = 1;
7440 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7443 btrfs_free_path(path);
7447 leaf = path->nodes[0];
7448 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7449 struct btrfs_extent_item);
7450 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7451 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7452 btrfs_set_extent_flags(leaf, extent_item,
7453 flags | BTRFS_EXTENT_FLAG_DATA);
7455 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7456 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7458 struct btrfs_shared_data_ref *ref;
7459 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7460 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7461 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7463 struct btrfs_extent_data_ref *ref;
7464 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7465 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7466 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7467 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7468 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7471 btrfs_mark_buffer_dirty(path->nodes[0]);
7472 btrfs_free_path(path);
7474 ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
7475 if (ret) { /* -ENOENT, logic error */
7476 btrfs_err(fs_info, "update block group failed for %llu %llu",
7477 ins->objectid, ins->offset);
7480 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7484 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7485 struct btrfs_root *root,
7486 u64 parent, u64 root_objectid,
7487 u64 flags, struct btrfs_disk_key *key,
7488 int level, struct btrfs_key *ins,
7492 struct btrfs_fs_info *fs_info = root->fs_info;
7493 struct btrfs_extent_item *extent_item;
7494 struct btrfs_tree_block_info *block_info;
7495 struct btrfs_extent_inline_ref *iref;
7496 struct btrfs_path *path;
7497 struct extent_buffer *leaf;
7498 u32 size = sizeof(*extent_item) + sizeof(*iref);
7499 u64 num_bytes = ins->offset;
7500 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7503 if (!skinny_metadata)
7504 size += sizeof(*block_info);
7506 path = btrfs_alloc_path();
7508 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7513 path->leave_spinning = 1;
7514 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7517 btrfs_free_path(path);
7518 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7523 leaf = path->nodes[0];
7524 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7525 struct btrfs_extent_item);
7526 btrfs_set_extent_refs(leaf, extent_item, 1);
7527 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7528 btrfs_set_extent_flags(leaf, extent_item,
7529 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7531 if (skinny_metadata) {
7532 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7533 num_bytes = root->nodesize;
7535 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7536 btrfs_set_tree_block_key(leaf, block_info, key);
7537 btrfs_set_tree_block_level(leaf, block_info, level);
7538 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7542 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7543 btrfs_set_extent_inline_ref_type(leaf, iref,
7544 BTRFS_SHARED_BLOCK_REF_KEY);
7545 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7547 btrfs_set_extent_inline_ref_type(leaf, iref,
7548 BTRFS_TREE_BLOCK_REF_KEY);
7549 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7552 btrfs_mark_buffer_dirty(leaf);
7553 btrfs_free_path(path);
7555 ret = update_block_group(trans, root, ins->objectid, root->nodesize,
7557 if (ret) { /* -ENOENT, logic error */
7558 btrfs_err(fs_info, "update block group failed for %llu %llu",
7559 ins->objectid, ins->offset);
7563 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7567 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7568 struct btrfs_root *root,
7569 u64 root_objectid, u64 owner,
7570 u64 offset, struct btrfs_key *ins)
7574 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7576 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7578 root_objectid, owner, offset,
7579 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7584 * this is used by the tree logging recovery code. It records that
7585 * an extent has been allocated and makes sure to clear the free
7586 * space cache bits as well
7588 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7589 struct btrfs_root *root,
7590 u64 root_objectid, u64 owner, u64 offset,
7591 struct btrfs_key *ins)
7594 struct btrfs_block_group_cache *block_group;
7597 * Mixed block groups will exclude before processing the log so we only
7598 * need to do the exlude dance if this fs isn't mixed.
7600 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7601 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7606 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7610 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7611 RESERVE_ALLOC_NO_ACCOUNT, 0);
7612 BUG_ON(ret); /* logic error */
7613 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7614 0, owner, offset, ins, 1);
7615 btrfs_put_block_group(block_group);
7619 static struct extent_buffer *
7620 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7621 u64 bytenr, int level)
7623 struct extent_buffer *buf;
7625 buf = btrfs_find_create_tree_block(root, bytenr);
7627 return ERR_PTR(-ENOMEM);
7628 btrfs_set_header_generation(buf, trans->transid);
7629 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7630 btrfs_tree_lock(buf);
7631 clean_tree_block(trans, root->fs_info, buf);
7632 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7634 btrfs_set_lock_blocking(buf);
7635 btrfs_set_buffer_uptodate(buf);
7637 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7638 buf->log_index = root->log_transid % 2;
7640 * we allow two log transactions at a time, use different
7641 * EXENT bit to differentiate dirty pages.
7643 if (buf->log_index == 0)
7644 set_extent_dirty(&root->dirty_log_pages, buf->start,
7645 buf->start + buf->len - 1, GFP_NOFS);
7647 set_extent_new(&root->dirty_log_pages, buf->start,
7648 buf->start + buf->len - 1, GFP_NOFS);
7650 buf->log_index = -1;
7651 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7652 buf->start + buf->len - 1, GFP_NOFS);
7654 trans->blocks_used++;
7655 /* this returns a buffer locked for blocking */
7659 static struct btrfs_block_rsv *
7660 use_block_rsv(struct btrfs_trans_handle *trans,
7661 struct btrfs_root *root, u32 blocksize)
7663 struct btrfs_block_rsv *block_rsv;
7664 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7666 bool global_updated = false;
7668 block_rsv = get_block_rsv(trans, root);
7670 if (unlikely(block_rsv->size == 0))
7673 ret = block_rsv_use_bytes(block_rsv, blocksize);
7677 if (block_rsv->failfast)
7678 return ERR_PTR(ret);
7680 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7681 global_updated = true;
7682 update_global_block_rsv(root->fs_info);
7686 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7687 static DEFINE_RATELIMIT_STATE(_rs,
7688 DEFAULT_RATELIMIT_INTERVAL * 10,
7689 /*DEFAULT_RATELIMIT_BURST*/ 1);
7690 if (__ratelimit(&_rs))
7692 "BTRFS: block rsv returned %d\n", ret);
7695 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7696 BTRFS_RESERVE_NO_FLUSH);
7700 * If we couldn't reserve metadata bytes try and use some from
7701 * the global reserve if its space type is the same as the global
7704 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7705 block_rsv->space_info == global_rsv->space_info) {
7706 ret = block_rsv_use_bytes(global_rsv, blocksize);
7710 return ERR_PTR(ret);
7713 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7714 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7716 block_rsv_add_bytes(block_rsv, blocksize, 0);
7717 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7721 * finds a free extent and does all the dirty work required for allocation
7722 * returns the tree buffer or an ERR_PTR on error.
7724 struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans,
7725 struct btrfs_root *root,
7726 u64 parent, u64 root_objectid,
7727 struct btrfs_disk_key *key, int level,
7728 u64 hint, u64 empty_size)
7730 struct btrfs_key ins;
7731 struct btrfs_block_rsv *block_rsv;
7732 struct extent_buffer *buf;
7733 struct btrfs_delayed_extent_op *extent_op;
7736 u32 blocksize = root->nodesize;
7737 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7740 if (btrfs_test_is_dummy_root(root)) {
7741 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7744 root->alloc_bytenr += blocksize;
7748 block_rsv = use_block_rsv(trans, root, blocksize);
7749 if (IS_ERR(block_rsv))
7750 return ERR_CAST(block_rsv);
7752 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7753 empty_size, hint, &ins, 0, 0);
7757 buf = btrfs_init_new_buffer(trans, root, ins.objectid, level);
7760 goto out_free_reserved;
7763 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7765 parent = ins.objectid;
7766 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7770 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7771 extent_op = btrfs_alloc_delayed_extent_op();
7777 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7779 memset(&extent_op->key, 0, sizeof(extent_op->key));
7780 extent_op->flags_to_set = flags;
7781 if (skinny_metadata)
7782 extent_op->update_key = 0;
7784 extent_op->update_key = 1;
7785 extent_op->update_flags = 1;
7786 extent_op->is_data = 0;
7787 extent_op->level = level;
7789 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7790 ins.objectid, ins.offset,
7791 parent, root_objectid, level,
7792 BTRFS_ADD_DELAYED_EXTENT,
7795 goto out_free_delayed;
7800 btrfs_free_delayed_extent_op(extent_op);
7802 free_extent_buffer(buf);
7804 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 0);
7806 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7807 return ERR_PTR(ret);
7810 struct walk_control {
7811 u64 refs[BTRFS_MAX_LEVEL];
7812 u64 flags[BTRFS_MAX_LEVEL];
7813 struct btrfs_key update_progress;
7824 #define DROP_REFERENCE 1
7825 #define UPDATE_BACKREF 2
7827 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7828 struct btrfs_root *root,
7829 struct walk_control *wc,
7830 struct btrfs_path *path)
7838 struct btrfs_key key;
7839 struct extent_buffer *eb;
7844 if (path->slots[wc->level] < wc->reada_slot) {
7845 wc->reada_count = wc->reada_count * 2 / 3;
7846 wc->reada_count = max(wc->reada_count, 2);
7848 wc->reada_count = wc->reada_count * 3 / 2;
7849 wc->reada_count = min_t(int, wc->reada_count,
7850 BTRFS_NODEPTRS_PER_BLOCK(root));
7853 eb = path->nodes[wc->level];
7854 nritems = btrfs_header_nritems(eb);
7855 blocksize = root->nodesize;
7857 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7858 if (nread >= wc->reada_count)
7862 bytenr = btrfs_node_blockptr(eb, slot);
7863 generation = btrfs_node_ptr_generation(eb, slot);
7865 if (slot == path->slots[wc->level])
7868 if (wc->stage == UPDATE_BACKREF &&
7869 generation <= root->root_key.offset)
7872 /* We don't lock the tree block, it's OK to be racy here */
7873 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7874 wc->level - 1, 1, &refs,
7876 /* We don't care about errors in readahead. */
7881 if (wc->stage == DROP_REFERENCE) {
7885 if (wc->level == 1 &&
7886 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7888 if (!wc->update_ref ||
7889 generation <= root->root_key.offset)
7891 btrfs_node_key_to_cpu(eb, &key, slot);
7892 ret = btrfs_comp_cpu_keys(&key,
7893 &wc->update_progress);
7897 if (wc->level == 1 &&
7898 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7902 readahead_tree_block(root, bytenr);
7905 wc->reada_slot = slot;
7909 * TODO: Modify related function to add related node/leaf to dirty_extent_root,
7910 * for later qgroup accounting.
7912 * Current, this function does nothing.
7914 static int account_leaf_items(struct btrfs_trans_handle *trans,
7915 struct btrfs_root *root,
7916 struct extent_buffer *eb)
7918 int nr = btrfs_header_nritems(eb);
7920 struct btrfs_key key;
7921 struct btrfs_file_extent_item *fi;
7922 u64 bytenr, num_bytes;
7924 for (i = 0; i < nr; i++) {
7925 btrfs_item_key_to_cpu(eb, &key, i);
7927 if (key.type != BTRFS_EXTENT_DATA_KEY)
7930 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7931 /* filter out non qgroup-accountable extents */
7932 extent_type = btrfs_file_extent_type(eb, fi);
7934 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7937 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7941 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7947 * Walk up the tree from the bottom, freeing leaves and any interior
7948 * nodes which have had all slots visited. If a node (leaf or
7949 * interior) is freed, the node above it will have it's slot
7950 * incremented. The root node will never be freed.
7952 * At the end of this function, we should have a path which has all
7953 * slots incremented to the next position for a search. If we need to
7954 * read a new node it will be NULL and the node above it will have the
7955 * correct slot selected for a later read.
7957 * If we increment the root nodes slot counter past the number of
7958 * elements, 1 is returned to signal completion of the search.
7960 static int adjust_slots_upwards(struct btrfs_root *root,
7961 struct btrfs_path *path, int root_level)
7965 struct extent_buffer *eb;
7967 if (root_level == 0)
7970 while (level <= root_level) {
7971 eb = path->nodes[level];
7972 nr = btrfs_header_nritems(eb);
7973 path->slots[level]++;
7974 slot = path->slots[level];
7975 if (slot >= nr || level == 0) {
7977 * Don't free the root - we will detect this
7978 * condition after our loop and return a
7979 * positive value for caller to stop walking the tree.
7981 if (level != root_level) {
7982 btrfs_tree_unlock_rw(eb, path->locks[level]);
7983 path->locks[level] = 0;
7985 free_extent_buffer(eb);
7986 path->nodes[level] = NULL;
7987 path->slots[level] = 0;
7991 * We have a valid slot to walk back down
7992 * from. Stop here so caller can process these
8001 eb = path->nodes[root_level];
8002 if (path->slots[root_level] >= btrfs_header_nritems(eb))
8009 * root_eb is the subtree root and is locked before this function is called.
8010 * TODO: Modify this function to mark all (including complete shared node)
8011 * to dirty_extent_root to allow it get accounted in qgroup.
8013 static int account_shared_subtree(struct btrfs_trans_handle *trans,
8014 struct btrfs_root *root,
8015 struct extent_buffer *root_eb,
8021 struct extent_buffer *eb = root_eb;
8022 struct btrfs_path *path = NULL;
8024 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
8025 BUG_ON(root_eb == NULL);
8027 if (!root->fs_info->quota_enabled)
8030 if (!extent_buffer_uptodate(root_eb)) {
8031 ret = btrfs_read_buffer(root_eb, root_gen);
8036 if (root_level == 0) {
8037 ret = account_leaf_items(trans, root, root_eb);
8041 path = btrfs_alloc_path();
8046 * Walk down the tree. Missing extent blocks are filled in as
8047 * we go. Metadata is accounted every time we read a new
8050 * When we reach a leaf, we account for file extent items in it,
8051 * walk back up the tree (adjusting slot pointers as we go)
8052 * and restart the search process.
8054 extent_buffer_get(root_eb); /* For path */
8055 path->nodes[root_level] = root_eb;
8056 path->slots[root_level] = 0;
8057 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
8060 while (level >= 0) {
8061 if (path->nodes[level] == NULL) {
8066 /* We need to get child blockptr/gen from
8067 * parent before we can read it. */
8068 eb = path->nodes[level + 1];
8069 parent_slot = path->slots[level + 1];
8070 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
8071 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
8073 eb = read_tree_block(root, child_bytenr, child_gen);
8077 } else if (!extent_buffer_uptodate(eb)) {
8078 free_extent_buffer(eb);
8083 path->nodes[level] = eb;
8084 path->slots[level] = 0;
8086 btrfs_tree_read_lock(eb);
8087 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
8088 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
8092 ret = account_leaf_items(trans, root, path->nodes[level]);
8096 /* Nonzero return here means we completed our search */
8097 ret = adjust_slots_upwards(root, path, root_level);
8101 /* Restart search with new slots */
8110 btrfs_free_path(path);
8116 * helper to process tree block while walking down the tree.
8118 * when wc->stage == UPDATE_BACKREF, this function updates
8119 * back refs for pointers in the block.
8121 * NOTE: return value 1 means we should stop walking down.
8123 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
8124 struct btrfs_root *root,
8125 struct btrfs_path *path,
8126 struct walk_control *wc, int lookup_info)
8128 int level = wc->level;
8129 struct extent_buffer *eb = path->nodes[level];
8130 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8133 if (wc->stage == UPDATE_BACKREF &&
8134 btrfs_header_owner(eb) != root->root_key.objectid)
8138 * when reference count of tree block is 1, it won't increase
8139 * again. once full backref flag is set, we never clear it.
8142 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
8143 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
8144 BUG_ON(!path->locks[level]);
8145 ret = btrfs_lookup_extent_info(trans, root,
8146 eb->start, level, 1,
8149 BUG_ON(ret == -ENOMEM);
8152 BUG_ON(wc->refs[level] == 0);
8155 if (wc->stage == DROP_REFERENCE) {
8156 if (wc->refs[level] > 1)
8159 if (path->locks[level] && !wc->keep_locks) {
8160 btrfs_tree_unlock_rw(eb, path->locks[level]);
8161 path->locks[level] = 0;
8166 /* wc->stage == UPDATE_BACKREF */
8167 if (!(wc->flags[level] & flag)) {
8168 BUG_ON(!path->locks[level]);
8169 ret = btrfs_inc_ref(trans, root, eb, 1);
8170 BUG_ON(ret); /* -ENOMEM */
8171 ret = btrfs_dec_ref(trans, root, eb, 0);
8172 BUG_ON(ret); /* -ENOMEM */
8173 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
8175 btrfs_header_level(eb), 0);
8176 BUG_ON(ret); /* -ENOMEM */
8177 wc->flags[level] |= flag;
8181 * the block is shared by multiple trees, so it's not good to
8182 * keep the tree lock
8184 if (path->locks[level] && level > 0) {
8185 btrfs_tree_unlock_rw(eb, path->locks[level]);
8186 path->locks[level] = 0;
8192 * helper to process tree block pointer.
8194 * when wc->stage == DROP_REFERENCE, this function checks
8195 * reference count of the block pointed to. if the block
8196 * is shared and we need update back refs for the subtree
8197 * rooted at the block, this function changes wc->stage to
8198 * UPDATE_BACKREF. if the block is shared and there is no
8199 * need to update back, this function drops the reference
8202 * NOTE: return value 1 means we should stop walking down.
8204 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
8205 struct btrfs_root *root,
8206 struct btrfs_path *path,
8207 struct walk_control *wc, int *lookup_info)
8213 struct btrfs_key key;
8214 struct extent_buffer *next;
8215 int level = wc->level;
8218 bool need_account = false;
8220 generation = btrfs_node_ptr_generation(path->nodes[level],
8221 path->slots[level]);
8223 * if the lower level block was created before the snapshot
8224 * was created, we know there is no need to update back refs
8227 if (wc->stage == UPDATE_BACKREF &&
8228 generation <= root->root_key.offset) {
8233 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
8234 blocksize = root->nodesize;
8236 next = btrfs_find_tree_block(root->fs_info, bytenr);
8238 next = btrfs_find_create_tree_block(root, bytenr);
8241 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
8245 btrfs_tree_lock(next);
8246 btrfs_set_lock_blocking(next);
8248 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
8249 &wc->refs[level - 1],
8250 &wc->flags[level - 1]);
8252 btrfs_tree_unlock(next);
8256 if (unlikely(wc->refs[level - 1] == 0)) {
8257 btrfs_err(root->fs_info, "Missing references.");
8262 if (wc->stage == DROP_REFERENCE) {
8263 if (wc->refs[level - 1] > 1) {
8264 need_account = true;
8266 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8269 if (!wc->update_ref ||
8270 generation <= root->root_key.offset)
8273 btrfs_node_key_to_cpu(path->nodes[level], &key,
8274 path->slots[level]);
8275 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
8279 wc->stage = UPDATE_BACKREF;
8280 wc->shared_level = level - 1;
8284 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
8288 if (!btrfs_buffer_uptodate(next, generation, 0)) {
8289 btrfs_tree_unlock(next);
8290 free_extent_buffer(next);
8296 if (reada && level == 1)
8297 reada_walk_down(trans, root, wc, path);
8298 next = read_tree_block(root, bytenr, generation);
8300 return PTR_ERR(next);
8301 } else if (!extent_buffer_uptodate(next)) {
8302 free_extent_buffer(next);
8305 btrfs_tree_lock(next);
8306 btrfs_set_lock_blocking(next);
8310 BUG_ON(level != btrfs_header_level(next));
8311 path->nodes[level] = next;
8312 path->slots[level] = 0;
8313 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8319 wc->refs[level - 1] = 0;
8320 wc->flags[level - 1] = 0;
8321 if (wc->stage == DROP_REFERENCE) {
8322 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
8323 parent = path->nodes[level]->start;
8325 BUG_ON(root->root_key.objectid !=
8326 btrfs_header_owner(path->nodes[level]));
8331 ret = account_shared_subtree(trans, root, next,
8332 generation, level - 1);
8334 btrfs_err_rl(root->fs_info,
8336 "%d accounting shared subtree. Quota "
8337 "is out of sync, rescan required.",
8341 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
8342 root->root_key.objectid, level - 1, 0, 0);
8343 BUG_ON(ret); /* -ENOMEM */
8345 btrfs_tree_unlock(next);
8346 free_extent_buffer(next);
8352 * helper to process tree block while walking up the tree.
8354 * when wc->stage == DROP_REFERENCE, this function drops
8355 * reference count on the block.
8357 * when wc->stage == UPDATE_BACKREF, this function changes
8358 * wc->stage back to DROP_REFERENCE if we changed wc->stage
8359 * to UPDATE_BACKREF previously while processing the block.
8361 * NOTE: return value 1 means we should stop walking up.
8363 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
8364 struct btrfs_root *root,
8365 struct btrfs_path *path,
8366 struct walk_control *wc)
8369 int level = wc->level;
8370 struct extent_buffer *eb = path->nodes[level];
8373 if (wc->stage == UPDATE_BACKREF) {
8374 BUG_ON(wc->shared_level < level);
8375 if (level < wc->shared_level)
8378 ret = find_next_key(path, level + 1, &wc->update_progress);
8382 wc->stage = DROP_REFERENCE;
8383 wc->shared_level = -1;
8384 path->slots[level] = 0;
8387 * check reference count again if the block isn't locked.
8388 * we should start walking down the tree again if reference
8391 if (!path->locks[level]) {
8393 btrfs_tree_lock(eb);
8394 btrfs_set_lock_blocking(eb);
8395 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8397 ret = btrfs_lookup_extent_info(trans, root,
8398 eb->start, level, 1,
8402 btrfs_tree_unlock_rw(eb, path->locks[level]);
8403 path->locks[level] = 0;
8406 BUG_ON(wc->refs[level] == 0);
8407 if (wc->refs[level] == 1) {
8408 btrfs_tree_unlock_rw(eb, path->locks[level]);
8409 path->locks[level] = 0;
8415 /* wc->stage == DROP_REFERENCE */
8416 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
8418 if (wc->refs[level] == 1) {
8420 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8421 ret = btrfs_dec_ref(trans, root, eb, 1);
8423 ret = btrfs_dec_ref(trans, root, eb, 0);
8424 BUG_ON(ret); /* -ENOMEM */
8425 ret = account_leaf_items(trans, root, eb);
8427 btrfs_err_rl(root->fs_info,
8429 "%d accounting leaf items. Quota "
8430 "is out of sync, rescan required.",
8434 /* make block locked assertion in clean_tree_block happy */
8435 if (!path->locks[level] &&
8436 btrfs_header_generation(eb) == trans->transid) {
8437 btrfs_tree_lock(eb);
8438 btrfs_set_lock_blocking(eb);
8439 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8441 clean_tree_block(trans, root->fs_info, eb);
8444 if (eb == root->node) {
8445 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8448 BUG_ON(root->root_key.objectid !=
8449 btrfs_header_owner(eb));
8451 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8452 parent = path->nodes[level + 1]->start;
8454 BUG_ON(root->root_key.objectid !=
8455 btrfs_header_owner(path->nodes[level + 1]));
8458 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8460 wc->refs[level] = 0;
8461 wc->flags[level] = 0;
8465 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8466 struct btrfs_root *root,
8467 struct btrfs_path *path,
8468 struct walk_control *wc)
8470 int level = wc->level;
8471 int lookup_info = 1;
8474 while (level >= 0) {
8475 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8482 if (path->slots[level] >=
8483 btrfs_header_nritems(path->nodes[level]))
8486 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8488 path->slots[level]++;
8497 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8498 struct btrfs_root *root,
8499 struct btrfs_path *path,
8500 struct walk_control *wc, int max_level)
8502 int level = wc->level;
8505 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8506 while (level < max_level && path->nodes[level]) {
8508 if (path->slots[level] + 1 <
8509 btrfs_header_nritems(path->nodes[level])) {
8510 path->slots[level]++;
8513 ret = walk_up_proc(trans, root, path, wc);
8517 if (path->locks[level]) {
8518 btrfs_tree_unlock_rw(path->nodes[level],
8519 path->locks[level]);
8520 path->locks[level] = 0;
8522 free_extent_buffer(path->nodes[level]);
8523 path->nodes[level] = NULL;
8531 * drop a subvolume tree.
8533 * this function traverses the tree freeing any blocks that only
8534 * referenced by the tree.
8536 * when a shared tree block is found. this function decreases its
8537 * reference count by one. if update_ref is true, this function
8538 * also make sure backrefs for the shared block and all lower level
8539 * blocks are properly updated.
8541 * If called with for_reloc == 0, may exit early with -EAGAIN
8543 int btrfs_drop_snapshot(struct btrfs_root *root,
8544 struct btrfs_block_rsv *block_rsv, int update_ref,
8547 struct btrfs_path *path;
8548 struct btrfs_trans_handle *trans;
8549 struct btrfs_root *tree_root = root->fs_info->tree_root;
8550 struct btrfs_root_item *root_item = &root->root_item;
8551 struct walk_control *wc;
8552 struct btrfs_key key;
8556 bool root_dropped = false;
8558 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8560 path = btrfs_alloc_path();
8566 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8568 btrfs_free_path(path);
8573 trans = btrfs_start_transaction(tree_root, 0);
8574 if (IS_ERR(trans)) {
8575 err = PTR_ERR(trans);
8580 trans->block_rsv = block_rsv;
8582 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8583 level = btrfs_header_level(root->node);
8584 path->nodes[level] = btrfs_lock_root_node(root);
8585 btrfs_set_lock_blocking(path->nodes[level]);
8586 path->slots[level] = 0;
8587 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8588 memset(&wc->update_progress, 0,
8589 sizeof(wc->update_progress));
8591 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8592 memcpy(&wc->update_progress, &key,
8593 sizeof(wc->update_progress));
8595 level = root_item->drop_level;
8597 path->lowest_level = level;
8598 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8599 path->lowest_level = 0;
8607 * unlock our path, this is safe because only this
8608 * function is allowed to delete this snapshot
8610 btrfs_unlock_up_safe(path, 0);
8612 level = btrfs_header_level(root->node);
8614 btrfs_tree_lock(path->nodes[level]);
8615 btrfs_set_lock_blocking(path->nodes[level]);
8616 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8618 ret = btrfs_lookup_extent_info(trans, root,
8619 path->nodes[level]->start,
8620 level, 1, &wc->refs[level],
8626 BUG_ON(wc->refs[level] == 0);
8628 if (level == root_item->drop_level)
8631 btrfs_tree_unlock(path->nodes[level]);
8632 path->locks[level] = 0;
8633 WARN_ON(wc->refs[level] != 1);
8639 wc->shared_level = -1;
8640 wc->stage = DROP_REFERENCE;
8641 wc->update_ref = update_ref;
8643 wc->for_reloc = for_reloc;
8644 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8648 ret = walk_down_tree(trans, root, path, wc);
8654 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8661 BUG_ON(wc->stage != DROP_REFERENCE);
8665 if (wc->stage == DROP_REFERENCE) {
8667 btrfs_node_key(path->nodes[level],
8668 &root_item->drop_progress,
8669 path->slots[level]);
8670 root_item->drop_level = level;
8673 BUG_ON(wc->level == 0);
8674 if (btrfs_should_end_transaction(trans, tree_root) ||
8675 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8676 ret = btrfs_update_root(trans, tree_root,
8680 btrfs_abort_transaction(trans, tree_root, ret);
8685 btrfs_end_transaction_throttle(trans, tree_root);
8686 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8687 pr_debug("BTRFS: drop snapshot early exit\n");
8692 trans = btrfs_start_transaction(tree_root, 0);
8693 if (IS_ERR(trans)) {
8694 err = PTR_ERR(trans);
8698 trans->block_rsv = block_rsv;
8701 btrfs_release_path(path);
8705 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8707 btrfs_abort_transaction(trans, tree_root, ret);
8711 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8712 ret = btrfs_find_root(tree_root, &root->root_key, path,
8715 btrfs_abort_transaction(trans, tree_root, ret);
8718 } else if (ret > 0) {
8719 /* if we fail to delete the orphan item this time
8720 * around, it'll get picked up the next time.
8722 * The most common failure here is just -ENOENT.
8724 btrfs_del_orphan_item(trans, tree_root,
8725 root->root_key.objectid);
8729 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8730 btrfs_add_dropped_root(trans, root);
8732 free_extent_buffer(root->node);
8733 free_extent_buffer(root->commit_root);
8734 btrfs_put_fs_root(root);
8736 root_dropped = true;
8738 btrfs_end_transaction_throttle(trans, tree_root);
8741 btrfs_free_path(path);
8744 * So if we need to stop dropping the snapshot for whatever reason we
8745 * need to make sure to add it back to the dead root list so that we
8746 * keep trying to do the work later. This also cleans up roots if we
8747 * don't have it in the radix (like when we recover after a power fail
8748 * or unmount) so we don't leak memory.
8750 if (!for_reloc && root_dropped == false)
8751 btrfs_add_dead_root(root);
8752 if (err && err != -EAGAIN)
8753 btrfs_std_error(root->fs_info, err, NULL);
8758 * drop subtree rooted at tree block 'node'.
8760 * NOTE: this function will unlock and release tree block 'node'
8761 * only used by relocation code
8763 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8764 struct btrfs_root *root,
8765 struct extent_buffer *node,
8766 struct extent_buffer *parent)
8768 struct btrfs_path *path;
8769 struct walk_control *wc;
8775 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8777 path = btrfs_alloc_path();
8781 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8783 btrfs_free_path(path);
8787 btrfs_assert_tree_locked(parent);
8788 parent_level = btrfs_header_level(parent);
8789 extent_buffer_get(parent);
8790 path->nodes[parent_level] = parent;
8791 path->slots[parent_level] = btrfs_header_nritems(parent);
8793 btrfs_assert_tree_locked(node);
8794 level = btrfs_header_level(node);
8795 path->nodes[level] = node;
8796 path->slots[level] = 0;
8797 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8799 wc->refs[parent_level] = 1;
8800 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8802 wc->shared_level = -1;
8803 wc->stage = DROP_REFERENCE;
8807 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8810 wret = walk_down_tree(trans, root, path, wc);
8816 wret = walk_up_tree(trans, root, path, wc, parent_level);
8824 btrfs_free_path(path);
8828 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8834 * if restripe for this chunk_type is on pick target profile and
8835 * return, otherwise do the usual balance
8837 stripped = get_restripe_target(root->fs_info, flags);
8839 return extended_to_chunk(stripped);
8841 num_devices = root->fs_info->fs_devices->rw_devices;
8843 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8844 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8845 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8847 if (num_devices == 1) {
8848 stripped |= BTRFS_BLOCK_GROUP_DUP;
8849 stripped = flags & ~stripped;
8851 /* turn raid0 into single device chunks */
8852 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8855 /* turn mirroring into duplication */
8856 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8857 BTRFS_BLOCK_GROUP_RAID10))
8858 return stripped | BTRFS_BLOCK_GROUP_DUP;
8860 /* they already had raid on here, just return */
8861 if (flags & stripped)
8864 stripped |= BTRFS_BLOCK_GROUP_DUP;
8865 stripped = flags & ~stripped;
8867 /* switch duplicated blocks with raid1 */
8868 if (flags & BTRFS_BLOCK_GROUP_DUP)
8869 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8871 /* this is drive concat, leave it alone */
8877 static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8879 struct btrfs_space_info *sinfo = cache->space_info;
8881 u64 min_allocable_bytes;
8885 * We need some metadata space and system metadata space for
8886 * allocating chunks in some corner cases until we force to set
8887 * it to be readonly.
8890 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8892 min_allocable_bytes = 1 * 1024 * 1024;
8894 min_allocable_bytes = 0;
8896 spin_lock(&sinfo->lock);
8897 spin_lock(&cache->lock);
8905 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8906 cache->bytes_super - btrfs_block_group_used(&cache->item);
8908 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8909 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8910 min_allocable_bytes <= sinfo->total_bytes) {
8911 sinfo->bytes_readonly += num_bytes;
8913 list_add_tail(&cache->ro_list, &sinfo->ro_bgs);
8917 spin_unlock(&cache->lock);
8918 spin_unlock(&sinfo->lock);
8922 int btrfs_inc_block_group_ro(struct btrfs_root *root,
8923 struct btrfs_block_group_cache *cache)
8926 struct btrfs_trans_handle *trans;
8931 trans = btrfs_join_transaction(root);
8933 return PTR_ERR(trans);
8936 * we're not allowed to set block groups readonly after the dirty
8937 * block groups cache has started writing. If it already started,
8938 * back off and let this transaction commit
8940 mutex_lock(&root->fs_info->ro_block_group_mutex);
8941 if (trans->transaction->dirty_bg_run) {
8942 u64 transid = trans->transid;
8944 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8945 btrfs_end_transaction(trans, root);
8947 ret = btrfs_wait_for_commit(root, transid);
8954 * if we are changing raid levels, try to allocate a corresponding
8955 * block group with the new raid level.
8957 alloc_flags = update_block_group_flags(root, cache->flags);
8958 if (alloc_flags != cache->flags) {
8959 ret = do_chunk_alloc(trans, root, alloc_flags,
8962 * ENOSPC is allowed here, we may have enough space
8963 * already allocated at the new raid level to
8972 ret = inc_block_group_ro(cache, 0);
8975 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8976 ret = do_chunk_alloc(trans, root, alloc_flags,
8980 ret = inc_block_group_ro(cache, 0);
8982 if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) {
8983 alloc_flags = update_block_group_flags(root, cache->flags);
8984 lock_chunks(root->fs_info->chunk_root);
8985 check_system_chunk(trans, root, alloc_flags);
8986 unlock_chunks(root->fs_info->chunk_root);
8988 mutex_unlock(&root->fs_info->ro_block_group_mutex);
8990 btrfs_end_transaction(trans, root);
8994 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8995 struct btrfs_root *root, u64 type)
8997 u64 alloc_flags = get_alloc_profile(root, type);
8998 return do_chunk_alloc(trans, root, alloc_flags,
9003 * helper to account the unused space of all the readonly block group in the
9004 * space_info. takes mirrors into account.
9006 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
9008 struct btrfs_block_group_cache *block_group;
9012 /* It's df, we don't care if it's racey */
9013 if (list_empty(&sinfo->ro_bgs))
9016 spin_lock(&sinfo->lock);
9017 list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) {
9018 spin_lock(&block_group->lock);
9020 if (!block_group->ro) {
9021 spin_unlock(&block_group->lock);
9025 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
9026 BTRFS_BLOCK_GROUP_RAID10 |
9027 BTRFS_BLOCK_GROUP_DUP))
9032 free_bytes += (block_group->key.offset -
9033 btrfs_block_group_used(&block_group->item)) *
9036 spin_unlock(&block_group->lock);
9038 spin_unlock(&sinfo->lock);
9043 void btrfs_dec_block_group_ro(struct btrfs_root *root,
9044 struct btrfs_block_group_cache *cache)
9046 struct btrfs_space_info *sinfo = cache->space_info;
9051 spin_lock(&sinfo->lock);
9052 spin_lock(&cache->lock);
9054 num_bytes = cache->key.offset - cache->reserved -
9055 cache->pinned - cache->bytes_super -
9056 btrfs_block_group_used(&cache->item);
9057 sinfo->bytes_readonly -= num_bytes;
9058 list_del_init(&cache->ro_list);
9060 spin_unlock(&cache->lock);
9061 spin_unlock(&sinfo->lock);
9065 * checks to see if its even possible to relocate this block group.
9067 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
9068 * ok to go ahead and try.
9070 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
9072 struct btrfs_block_group_cache *block_group;
9073 struct btrfs_space_info *space_info;
9074 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
9075 struct btrfs_device *device;
9076 struct btrfs_trans_handle *trans;
9085 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
9087 /* odd, couldn't find the block group, leave it alone */
9091 min_free = btrfs_block_group_used(&block_group->item);
9093 /* no bytes used, we're good */
9097 space_info = block_group->space_info;
9098 spin_lock(&space_info->lock);
9100 full = space_info->full;
9103 * if this is the last block group we have in this space, we can't
9104 * relocate it unless we're able to allocate a new chunk below.
9106 * Otherwise, we need to make sure we have room in the space to handle
9107 * all of the extents from this block group. If we can, we're good
9109 if ((space_info->total_bytes != block_group->key.offset) &&
9110 (space_info->bytes_used + space_info->bytes_reserved +
9111 space_info->bytes_pinned + space_info->bytes_readonly +
9112 min_free < space_info->total_bytes)) {
9113 spin_unlock(&space_info->lock);
9116 spin_unlock(&space_info->lock);
9119 * ok we don't have enough space, but maybe we have free space on our
9120 * devices to allocate new chunks for relocation, so loop through our
9121 * alloc devices and guess if we have enough space. if this block
9122 * group is going to be restriped, run checks against the target
9123 * profile instead of the current one.
9135 target = get_restripe_target(root->fs_info, block_group->flags);
9137 index = __get_raid_index(extended_to_chunk(target));
9140 * this is just a balance, so if we were marked as full
9141 * we know there is no space for a new chunk
9146 index = get_block_group_index(block_group);
9149 if (index == BTRFS_RAID_RAID10) {
9153 } else if (index == BTRFS_RAID_RAID1) {
9155 } else if (index == BTRFS_RAID_DUP) {
9158 } else if (index == BTRFS_RAID_RAID0) {
9159 dev_min = fs_devices->rw_devices;
9160 min_free = div64_u64(min_free, dev_min);
9163 /* We need to do this so that we can look at pending chunks */
9164 trans = btrfs_join_transaction(root);
9165 if (IS_ERR(trans)) {
9166 ret = PTR_ERR(trans);
9170 mutex_lock(&root->fs_info->chunk_mutex);
9171 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
9175 * check to make sure we can actually find a chunk with enough
9176 * space to fit our block group in.
9178 if (device->total_bytes > device->bytes_used + min_free &&
9179 !device->is_tgtdev_for_dev_replace) {
9180 ret = find_free_dev_extent(trans, device, min_free,
9185 if (dev_nr >= dev_min)
9191 mutex_unlock(&root->fs_info->chunk_mutex);
9192 btrfs_end_transaction(trans, root);
9194 btrfs_put_block_group(block_group);
9198 static int find_first_block_group(struct btrfs_root *root,
9199 struct btrfs_path *path, struct btrfs_key *key)
9202 struct btrfs_key found_key;
9203 struct extent_buffer *leaf;
9206 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
9211 slot = path->slots[0];
9212 leaf = path->nodes[0];
9213 if (slot >= btrfs_header_nritems(leaf)) {
9214 ret = btrfs_next_leaf(root, path);
9221 btrfs_item_key_to_cpu(leaf, &found_key, slot);
9223 if (found_key.objectid >= key->objectid &&
9224 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
9234 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
9236 struct btrfs_block_group_cache *block_group;
9240 struct inode *inode;
9242 block_group = btrfs_lookup_first_block_group(info, last);
9243 while (block_group) {
9244 spin_lock(&block_group->lock);
9245 if (block_group->iref)
9247 spin_unlock(&block_group->lock);
9248 block_group = next_block_group(info->tree_root,
9258 inode = block_group->inode;
9259 block_group->iref = 0;
9260 block_group->inode = NULL;
9261 spin_unlock(&block_group->lock);
9263 last = block_group->key.objectid + block_group->key.offset;
9264 btrfs_put_block_group(block_group);
9268 int btrfs_free_block_groups(struct btrfs_fs_info *info)
9270 struct btrfs_block_group_cache *block_group;
9271 struct btrfs_space_info *space_info;
9272 struct btrfs_caching_control *caching_ctl;
9275 down_write(&info->commit_root_sem);
9276 while (!list_empty(&info->caching_block_groups)) {
9277 caching_ctl = list_entry(info->caching_block_groups.next,
9278 struct btrfs_caching_control, list);
9279 list_del(&caching_ctl->list);
9280 put_caching_control(caching_ctl);
9282 up_write(&info->commit_root_sem);
9284 spin_lock(&info->unused_bgs_lock);
9285 while (!list_empty(&info->unused_bgs)) {
9286 block_group = list_first_entry(&info->unused_bgs,
9287 struct btrfs_block_group_cache,
9289 list_del_init(&block_group->bg_list);
9290 btrfs_put_block_group(block_group);
9292 spin_unlock(&info->unused_bgs_lock);
9294 spin_lock(&info->block_group_cache_lock);
9295 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
9296 block_group = rb_entry(n, struct btrfs_block_group_cache,
9298 rb_erase(&block_group->cache_node,
9299 &info->block_group_cache_tree);
9300 RB_CLEAR_NODE(&block_group->cache_node);
9301 spin_unlock(&info->block_group_cache_lock);
9303 down_write(&block_group->space_info->groups_sem);
9304 list_del(&block_group->list);
9305 up_write(&block_group->space_info->groups_sem);
9307 if (block_group->cached == BTRFS_CACHE_STARTED)
9308 wait_block_group_cache_done(block_group);
9311 * We haven't cached this block group, which means we could
9312 * possibly have excluded extents on this block group.
9314 if (block_group->cached == BTRFS_CACHE_NO ||
9315 block_group->cached == BTRFS_CACHE_ERROR)
9316 free_excluded_extents(info->extent_root, block_group);
9318 btrfs_remove_free_space_cache(block_group);
9319 btrfs_put_block_group(block_group);
9321 spin_lock(&info->block_group_cache_lock);
9323 spin_unlock(&info->block_group_cache_lock);
9325 /* now that all the block groups are freed, go through and
9326 * free all the space_info structs. This is only called during
9327 * the final stages of unmount, and so we know nobody is
9328 * using them. We call synchronize_rcu() once before we start,
9329 * just to be on the safe side.
9333 release_global_block_rsv(info);
9335 while (!list_empty(&info->space_info)) {
9338 space_info = list_entry(info->space_info.next,
9339 struct btrfs_space_info,
9341 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
9342 if (WARN_ON(space_info->bytes_pinned > 0 ||
9343 space_info->bytes_reserved > 0 ||
9344 space_info->bytes_may_use > 0)) {
9345 dump_space_info(space_info, 0, 0);
9348 list_del(&space_info->list);
9349 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
9350 struct kobject *kobj;
9351 kobj = space_info->block_group_kobjs[i];
9352 space_info->block_group_kobjs[i] = NULL;
9358 kobject_del(&space_info->kobj);
9359 kobject_put(&space_info->kobj);
9364 static void __link_block_group(struct btrfs_space_info *space_info,
9365 struct btrfs_block_group_cache *cache)
9367 int index = get_block_group_index(cache);
9370 down_write(&space_info->groups_sem);
9371 if (list_empty(&space_info->block_groups[index]))
9373 list_add_tail(&cache->list, &space_info->block_groups[index]);
9374 up_write(&space_info->groups_sem);
9377 struct raid_kobject *rkobj;
9380 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
9383 rkobj->raid_type = index;
9384 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
9385 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
9386 "%s", get_raid_name(index));
9388 kobject_put(&rkobj->kobj);
9391 space_info->block_group_kobjs[index] = &rkobj->kobj;
9396 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
9399 static struct btrfs_block_group_cache *
9400 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
9402 struct btrfs_block_group_cache *cache;
9404 cache = kzalloc(sizeof(*cache), GFP_NOFS);
9408 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
9410 if (!cache->free_space_ctl) {
9415 cache->key.objectid = start;
9416 cache->key.offset = size;
9417 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9419 cache->sectorsize = root->sectorsize;
9420 cache->fs_info = root->fs_info;
9421 cache->full_stripe_len = btrfs_full_stripe_len(root,
9422 &root->fs_info->mapping_tree,
9424 atomic_set(&cache->count, 1);
9425 spin_lock_init(&cache->lock);
9426 init_rwsem(&cache->data_rwsem);
9427 INIT_LIST_HEAD(&cache->list);
9428 INIT_LIST_HEAD(&cache->cluster_list);
9429 INIT_LIST_HEAD(&cache->bg_list);
9430 INIT_LIST_HEAD(&cache->ro_list);
9431 INIT_LIST_HEAD(&cache->dirty_list);
9432 INIT_LIST_HEAD(&cache->io_list);
9433 btrfs_init_free_space_ctl(cache);
9434 atomic_set(&cache->trimming, 0);
9439 int btrfs_read_block_groups(struct btrfs_root *root)
9441 struct btrfs_path *path;
9443 struct btrfs_block_group_cache *cache;
9444 struct btrfs_fs_info *info = root->fs_info;
9445 struct btrfs_space_info *space_info;
9446 struct btrfs_key key;
9447 struct btrfs_key found_key;
9448 struct extent_buffer *leaf;
9452 root = info->extent_root;
9455 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9456 path = btrfs_alloc_path();
9461 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9462 if (btrfs_test_opt(root, SPACE_CACHE) &&
9463 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9465 if (btrfs_test_opt(root, CLEAR_CACHE))
9469 ret = find_first_block_group(root, path, &key);
9475 leaf = path->nodes[0];
9476 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9478 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9487 * When we mount with old space cache, we need to
9488 * set BTRFS_DC_CLEAR and set dirty flag.
9490 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9491 * truncate the old free space cache inode and
9493 * b) Setting 'dirty flag' makes sure that we flush
9494 * the new space cache info onto disk.
9496 if (btrfs_test_opt(root, SPACE_CACHE))
9497 cache->disk_cache_state = BTRFS_DC_CLEAR;
9500 read_extent_buffer(leaf, &cache->item,
9501 btrfs_item_ptr_offset(leaf, path->slots[0]),
9502 sizeof(cache->item));
9503 cache->flags = btrfs_block_group_flags(&cache->item);
9505 key.objectid = found_key.objectid + found_key.offset;
9506 btrfs_release_path(path);
9509 * We need to exclude the super stripes now so that the space
9510 * info has super bytes accounted for, otherwise we'll think
9511 * we have more space than we actually do.
9513 ret = exclude_super_stripes(root, cache);
9516 * We may have excluded something, so call this just in
9519 free_excluded_extents(root, cache);
9520 btrfs_put_block_group(cache);
9525 * check for two cases, either we are full, and therefore
9526 * don't need to bother with the caching work since we won't
9527 * find any space, or we are empty, and we can just add all
9528 * the space in and be done with it. This saves us _alot_ of
9529 * time, particularly in the full case.
9531 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9532 cache->last_byte_to_unpin = (u64)-1;
9533 cache->cached = BTRFS_CACHE_FINISHED;
9534 free_excluded_extents(root, cache);
9535 } else if (btrfs_block_group_used(&cache->item) == 0) {
9536 cache->last_byte_to_unpin = (u64)-1;
9537 cache->cached = BTRFS_CACHE_FINISHED;
9538 add_new_free_space(cache, root->fs_info,
9540 found_key.objectid +
9542 free_excluded_extents(root, cache);
9545 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9547 btrfs_remove_free_space_cache(cache);
9548 btrfs_put_block_group(cache);
9552 ret = update_space_info(info, cache->flags, found_key.offset,
9553 btrfs_block_group_used(&cache->item),
9556 btrfs_remove_free_space_cache(cache);
9557 spin_lock(&info->block_group_cache_lock);
9558 rb_erase(&cache->cache_node,
9559 &info->block_group_cache_tree);
9560 RB_CLEAR_NODE(&cache->cache_node);
9561 spin_unlock(&info->block_group_cache_lock);
9562 btrfs_put_block_group(cache);
9566 cache->space_info = space_info;
9567 spin_lock(&cache->space_info->lock);
9568 cache->space_info->bytes_readonly += cache->bytes_super;
9569 spin_unlock(&cache->space_info->lock);
9571 __link_block_group(space_info, cache);
9573 set_avail_alloc_bits(root->fs_info, cache->flags);
9574 if (btrfs_chunk_readonly(root, cache->key.objectid)) {
9575 inc_block_group_ro(cache, 1);
9576 } else if (btrfs_block_group_used(&cache->item) == 0) {
9577 spin_lock(&info->unused_bgs_lock);
9578 /* Should always be true but just in case. */
9579 if (list_empty(&cache->bg_list)) {
9580 btrfs_get_block_group(cache);
9581 list_add_tail(&cache->bg_list,
9584 spin_unlock(&info->unused_bgs_lock);
9588 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9589 if (!(get_alloc_profile(root, space_info->flags) &
9590 (BTRFS_BLOCK_GROUP_RAID10 |
9591 BTRFS_BLOCK_GROUP_RAID1 |
9592 BTRFS_BLOCK_GROUP_RAID5 |
9593 BTRFS_BLOCK_GROUP_RAID6 |
9594 BTRFS_BLOCK_GROUP_DUP)))
9597 * avoid allocating from un-mirrored block group if there are
9598 * mirrored block groups.
9600 list_for_each_entry(cache,
9601 &space_info->block_groups[BTRFS_RAID_RAID0],
9603 inc_block_group_ro(cache, 1);
9604 list_for_each_entry(cache,
9605 &space_info->block_groups[BTRFS_RAID_SINGLE],
9607 inc_block_group_ro(cache, 1);
9610 init_global_block_rsv(info);
9613 btrfs_free_path(path);
9617 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9618 struct btrfs_root *root)
9620 struct btrfs_block_group_cache *block_group, *tmp;
9621 struct btrfs_root *extent_root = root->fs_info->extent_root;
9622 struct btrfs_block_group_item item;
9623 struct btrfs_key key;
9625 bool can_flush_pending_bgs = trans->can_flush_pending_bgs;
9627 trans->can_flush_pending_bgs = false;
9628 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) {
9632 spin_lock(&block_group->lock);
9633 memcpy(&item, &block_group->item, sizeof(item));
9634 memcpy(&key, &block_group->key, sizeof(key));
9635 spin_unlock(&block_group->lock);
9637 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9640 btrfs_abort_transaction(trans, extent_root, ret);
9641 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9642 key.objectid, key.offset);
9644 btrfs_abort_transaction(trans, extent_root, ret);
9646 list_del_init(&block_group->bg_list);
9648 trans->can_flush_pending_bgs = can_flush_pending_bgs;
9651 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9652 struct btrfs_root *root, u64 bytes_used,
9653 u64 type, u64 chunk_objectid, u64 chunk_offset,
9657 struct btrfs_root *extent_root;
9658 struct btrfs_block_group_cache *cache;
9660 extent_root = root->fs_info->extent_root;
9662 btrfs_set_log_full_commit(root->fs_info, trans);
9664 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9668 btrfs_set_block_group_used(&cache->item, bytes_used);
9669 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9670 btrfs_set_block_group_flags(&cache->item, type);
9672 cache->flags = type;
9673 cache->last_byte_to_unpin = (u64)-1;
9674 cache->cached = BTRFS_CACHE_FINISHED;
9675 ret = exclude_super_stripes(root, cache);
9678 * We may have excluded something, so call this just in
9681 free_excluded_extents(root, cache);
9682 btrfs_put_block_group(cache);
9686 add_new_free_space(cache, root->fs_info, chunk_offset,
9687 chunk_offset + size);
9689 free_excluded_extents(root, cache);
9691 #ifdef CONFIG_BTRFS_DEBUG
9692 if (btrfs_should_fragment_free_space(root, cache)) {
9693 u64 new_bytes_used = size - bytes_used;
9695 bytes_used += new_bytes_used >> 1;
9696 fragment_free_space(root, cache);
9700 * Call to ensure the corresponding space_info object is created and
9701 * assigned to our block group, but don't update its counters just yet.
9702 * We want our bg to be added to the rbtree with its ->space_info set.
9704 ret = update_space_info(root->fs_info, cache->flags, 0, 0,
9705 &cache->space_info);
9707 btrfs_remove_free_space_cache(cache);
9708 btrfs_put_block_group(cache);
9712 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9714 btrfs_remove_free_space_cache(cache);
9715 btrfs_put_block_group(cache);
9720 * Now that our block group has its ->space_info set and is inserted in
9721 * the rbtree, update the space info's counters.
9723 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9724 &cache->space_info);
9726 btrfs_remove_free_space_cache(cache);
9727 spin_lock(&root->fs_info->block_group_cache_lock);
9728 rb_erase(&cache->cache_node,
9729 &root->fs_info->block_group_cache_tree);
9730 RB_CLEAR_NODE(&cache->cache_node);
9731 spin_unlock(&root->fs_info->block_group_cache_lock);
9732 btrfs_put_block_group(cache);
9735 update_global_block_rsv(root->fs_info);
9737 spin_lock(&cache->space_info->lock);
9738 cache->space_info->bytes_readonly += cache->bytes_super;
9739 spin_unlock(&cache->space_info->lock);
9741 __link_block_group(cache->space_info, cache);
9743 list_add_tail(&cache->bg_list, &trans->new_bgs);
9745 set_avail_alloc_bits(extent_root->fs_info, type);
9750 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9752 u64 extra_flags = chunk_to_extended(flags) &
9753 BTRFS_EXTENDED_PROFILE_MASK;
9755 write_seqlock(&fs_info->profiles_lock);
9756 if (flags & BTRFS_BLOCK_GROUP_DATA)
9757 fs_info->avail_data_alloc_bits &= ~extra_flags;
9758 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9759 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9760 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9761 fs_info->avail_system_alloc_bits &= ~extra_flags;
9762 write_sequnlock(&fs_info->profiles_lock);
9765 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9766 struct btrfs_root *root, u64 group_start,
9767 struct extent_map *em)
9769 struct btrfs_path *path;
9770 struct btrfs_block_group_cache *block_group;
9771 struct btrfs_free_cluster *cluster;
9772 struct btrfs_root *tree_root = root->fs_info->tree_root;
9773 struct btrfs_key key;
9774 struct inode *inode;
9775 struct kobject *kobj = NULL;
9779 struct btrfs_caching_control *caching_ctl = NULL;
9782 root = root->fs_info->extent_root;
9784 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9785 BUG_ON(!block_group);
9786 BUG_ON(!block_group->ro);
9789 * Free the reserved super bytes from this block group before
9792 free_excluded_extents(root, block_group);
9794 memcpy(&key, &block_group->key, sizeof(key));
9795 index = get_block_group_index(block_group);
9796 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9797 BTRFS_BLOCK_GROUP_RAID1 |
9798 BTRFS_BLOCK_GROUP_RAID10))
9803 /* make sure this block group isn't part of an allocation cluster */
9804 cluster = &root->fs_info->data_alloc_cluster;
9805 spin_lock(&cluster->refill_lock);
9806 btrfs_return_cluster_to_free_space(block_group, cluster);
9807 spin_unlock(&cluster->refill_lock);
9810 * make sure this block group isn't part of a metadata
9811 * allocation cluster
9813 cluster = &root->fs_info->meta_alloc_cluster;
9814 spin_lock(&cluster->refill_lock);
9815 btrfs_return_cluster_to_free_space(block_group, cluster);
9816 spin_unlock(&cluster->refill_lock);
9818 path = btrfs_alloc_path();
9825 * get the inode first so any iput calls done for the io_list
9826 * aren't the final iput (no unlinks allowed now)
9828 inode = lookup_free_space_inode(tree_root, block_group, path);
9830 mutex_lock(&trans->transaction->cache_write_mutex);
9832 * make sure our free spache cache IO is done before remove the
9835 spin_lock(&trans->transaction->dirty_bgs_lock);
9836 if (!list_empty(&block_group->io_list)) {
9837 list_del_init(&block_group->io_list);
9839 WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode);
9841 spin_unlock(&trans->transaction->dirty_bgs_lock);
9842 btrfs_wait_cache_io(root, trans, block_group,
9843 &block_group->io_ctl, path,
9844 block_group->key.objectid);
9845 btrfs_put_block_group(block_group);
9846 spin_lock(&trans->transaction->dirty_bgs_lock);
9849 if (!list_empty(&block_group->dirty_list)) {
9850 list_del_init(&block_group->dirty_list);
9851 btrfs_put_block_group(block_group);
9853 spin_unlock(&trans->transaction->dirty_bgs_lock);
9854 mutex_unlock(&trans->transaction->cache_write_mutex);
9856 if (!IS_ERR(inode)) {
9857 ret = btrfs_orphan_add(trans, inode);
9859 btrfs_add_delayed_iput(inode);
9863 /* One for the block groups ref */
9864 spin_lock(&block_group->lock);
9865 if (block_group->iref) {
9866 block_group->iref = 0;
9867 block_group->inode = NULL;
9868 spin_unlock(&block_group->lock);
9871 spin_unlock(&block_group->lock);
9873 /* One for our lookup ref */
9874 btrfs_add_delayed_iput(inode);
9877 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9878 key.offset = block_group->key.objectid;
9881 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9885 btrfs_release_path(path);
9887 ret = btrfs_del_item(trans, tree_root, path);
9890 btrfs_release_path(path);
9893 spin_lock(&root->fs_info->block_group_cache_lock);
9894 rb_erase(&block_group->cache_node,
9895 &root->fs_info->block_group_cache_tree);
9896 RB_CLEAR_NODE(&block_group->cache_node);
9898 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9899 root->fs_info->first_logical_byte = (u64)-1;
9900 spin_unlock(&root->fs_info->block_group_cache_lock);
9902 down_write(&block_group->space_info->groups_sem);
9904 * we must use list_del_init so people can check to see if they
9905 * are still on the list after taking the semaphore
9907 list_del_init(&block_group->list);
9908 if (list_empty(&block_group->space_info->block_groups[index])) {
9909 kobj = block_group->space_info->block_group_kobjs[index];
9910 block_group->space_info->block_group_kobjs[index] = NULL;
9911 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9913 up_write(&block_group->space_info->groups_sem);
9919 if (block_group->has_caching_ctl)
9920 caching_ctl = get_caching_control(block_group);
9921 if (block_group->cached == BTRFS_CACHE_STARTED)
9922 wait_block_group_cache_done(block_group);
9923 if (block_group->has_caching_ctl) {
9924 down_write(&root->fs_info->commit_root_sem);
9926 struct btrfs_caching_control *ctl;
9928 list_for_each_entry(ctl,
9929 &root->fs_info->caching_block_groups, list)
9930 if (ctl->block_group == block_group) {
9932 atomic_inc(&caching_ctl->count);
9937 list_del_init(&caching_ctl->list);
9938 up_write(&root->fs_info->commit_root_sem);
9940 /* Once for the caching bgs list and once for us. */
9941 put_caching_control(caching_ctl);
9942 put_caching_control(caching_ctl);
9946 spin_lock(&trans->transaction->dirty_bgs_lock);
9947 if (!list_empty(&block_group->dirty_list)) {
9950 if (!list_empty(&block_group->io_list)) {
9953 spin_unlock(&trans->transaction->dirty_bgs_lock);
9954 btrfs_remove_free_space_cache(block_group);
9956 spin_lock(&block_group->space_info->lock);
9957 list_del_init(&block_group->ro_list);
9959 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
9960 WARN_ON(block_group->space_info->total_bytes
9961 < block_group->key.offset);
9962 WARN_ON(block_group->space_info->bytes_readonly
9963 < block_group->key.offset);
9964 WARN_ON(block_group->space_info->disk_total
9965 < block_group->key.offset * factor);
9967 block_group->space_info->total_bytes -= block_group->key.offset;
9968 block_group->space_info->bytes_readonly -= block_group->key.offset;
9969 block_group->space_info->disk_total -= block_group->key.offset * factor;
9971 spin_unlock(&block_group->space_info->lock);
9973 memcpy(&key, &block_group->key, sizeof(key));
9976 if (!list_empty(&em->list)) {
9977 /* We're in the transaction->pending_chunks list. */
9978 free_extent_map(em);
9980 spin_lock(&block_group->lock);
9981 block_group->removed = 1;
9983 * At this point trimming can't start on this block group, because we
9984 * removed the block group from the tree fs_info->block_group_cache_tree
9985 * so no one can't find it anymore and even if someone already got this
9986 * block group before we removed it from the rbtree, they have already
9987 * incremented block_group->trimming - if they didn't, they won't find
9988 * any free space entries because we already removed them all when we
9989 * called btrfs_remove_free_space_cache().
9991 * And we must not remove the extent map from the fs_info->mapping_tree
9992 * to prevent the same logical address range and physical device space
9993 * ranges from being reused for a new block group. This is because our
9994 * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is
9995 * completely transactionless, so while it is trimming a range the
9996 * currently running transaction might finish and a new one start,
9997 * allowing for new block groups to be created that can reuse the same
9998 * physical device locations unless we take this special care.
10000 * There may also be an implicit trim operation if the file system
10001 * is mounted with -odiscard. The same protections must remain
10002 * in place until the extents have been discarded completely when
10003 * the transaction commit has completed.
10005 remove_em = (atomic_read(&block_group->trimming) == 0);
10007 * Make sure a trimmer task always sees the em in the pinned_chunks list
10008 * if it sees block_group->removed == 1 (needs to lock block_group->lock
10009 * before checking block_group->removed).
10013 * Our em might be in trans->transaction->pending_chunks which
10014 * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks),
10015 * and so is the fs_info->pinned_chunks list.
10017 * So at this point we must be holding the chunk_mutex to avoid
10018 * any races with chunk allocation (more specifically at
10019 * volumes.c:contains_pending_extent()), to ensure it always
10020 * sees the em, either in the pending_chunks list or in the
10021 * pinned_chunks list.
10023 list_move_tail(&em->list, &root->fs_info->pinned_chunks);
10025 spin_unlock(&block_group->lock);
10028 struct extent_map_tree *em_tree;
10030 em_tree = &root->fs_info->mapping_tree.map_tree;
10031 write_lock(&em_tree->lock);
10033 * The em might be in the pending_chunks list, so make sure the
10034 * chunk mutex is locked, since remove_extent_mapping() will
10035 * delete us from that list.
10037 remove_extent_mapping(em_tree, em);
10038 write_unlock(&em_tree->lock);
10039 /* once for the tree */
10040 free_extent_map(em);
10043 unlock_chunks(root);
10045 btrfs_put_block_group(block_group);
10046 btrfs_put_block_group(block_group);
10048 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
10054 ret = btrfs_del_item(trans, root, path);
10056 btrfs_free_path(path);
10061 * Process the unused_bgs list and remove any that don't have any allocated
10062 * space inside of them.
10064 void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info)
10066 struct btrfs_block_group_cache *block_group;
10067 struct btrfs_space_info *space_info;
10068 struct btrfs_root *root = fs_info->extent_root;
10069 struct btrfs_trans_handle *trans;
10072 if (!fs_info->open)
10075 spin_lock(&fs_info->unused_bgs_lock);
10076 while (!list_empty(&fs_info->unused_bgs)) {
10080 block_group = list_first_entry(&fs_info->unused_bgs,
10081 struct btrfs_block_group_cache,
10083 space_info = block_group->space_info;
10084 list_del_init(&block_group->bg_list);
10085 if (ret || btrfs_mixed_space_info(space_info)) {
10086 btrfs_put_block_group(block_group);
10089 spin_unlock(&fs_info->unused_bgs_lock);
10091 mutex_lock(&root->fs_info->delete_unused_bgs_mutex);
10093 /* Don't want to race with allocators so take the groups_sem */
10094 down_write(&space_info->groups_sem);
10095 spin_lock(&block_group->lock);
10096 if (block_group->reserved ||
10097 btrfs_block_group_used(&block_group->item) ||
10100 * We want to bail if we made new allocations or have
10101 * outstanding allocations in this block group. We do
10102 * the ro check in case balance is currently acting on
10103 * this block group.
10105 spin_unlock(&block_group->lock);
10106 up_write(&space_info->groups_sem);
10109 spin_unlock(&block_group->lock);
10111 /* We don't want to force the issue, only flip if it's ok. */
10112 ret = inc_block_group_ro(block_group, 0);
10113 up_write(&space_info->groups_sem);
10120 * Want to do this before we do anything else so we can recover
10121 * properly if we fail to join the transaction.
10123 /* 1 for btrfs_orphan_reserve_metadata() */
10124 trans = btrfs_start_transaction(root, 1);
10125 if (IS_ERR(trans)) {
10126 btrfs_dec_block_group_ro(root, block_group);
10127 ret = PTR_ERR(trans);
10132 * We could have pending pinned extents for this block group,
10133 * just delete them, we don't care about them anymore.
10135 start = block_group->key.objectid;
10136 end = start + block_group->key.offset - 1;
10138 * Hold the unused_bg_unpin_mutex lock to avoid racing with
10139 * btrfs_finish_extent_commit(). If we are at transaction N,
10140 * another task might be running finish_extent_commit() for the
10141 * previous transaction N - 1, and have seen a range belonging
10142 * to the block group in freed_extents[] before we were able to
10143 * clear the whole block group range from freed_extents[]. This
10144 * means that task can lookup for the block group after we
10145 * unpinned it from freed_extents[] and removed it, leading to
10146 * a BUG_ON() at btrfs_unpin_extent_range().
10148 mutex_lock(&fs_info->unused_bg_unpin_mutex);
10149 ret = clear_extent_bits(&fs_info->freed_extents[0], start, end,
10150 EXTENT_DIRTY, GFP_NOFS);
10152 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10153 btrfs_dec_block_group_ro(root, block_group);
10156 ret = clear_extent_bits(&fs_info->freed_extents[1], start, end,
10157 EXTENT_DIRTY, GFP_NOFS);
10159 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10160 btrfs_dec_block_group_ro(root, block_group);
10163 mutex_unlock(&fs_info->unused_bg_unpin_mutex);
10165 /* Reset pinned so btrfs_put_block_group doesn't complain */
10166 spin_lock(&space_info->lock);
10167 spin_lock(&block_group->lock);
10169 space_info->bytes_pinned -= block_group->pinned;
10170 space_info->bytes_readonly += block_group->pinned;
10171 percpu_counter_add(&space_info->total_bytes_pinned,
10172 -block_group->pinned);
10173 block_group->pinned = 0;
10175 spin_unlock(&block_group->lock);
10176 spin_unlock(&space_info->lock);
10178 /* DISCARD can flip during remount */
10179 trimming = btrfs_test_opt(root, DISCARD);
10181 /* Implicit trim during transaction commit. */
10183 btrfs_get_block_group_trimming(block_group);
10186 * Btrfs_remove_chunk will abort the transaction if things go
10189 ret = btrfs_remove_chunk(trans, root,
10190 block_group->key.objectid);
10194 btrfs_put_block_group_trimming(block_group);
10199 * If we're not mounted with -odiscard, we can just forget
10200 * about this block group. Otherwise we'll need to wait
10201 * until transaction commit to do the actual discard.
10204 WARN_ON(!list_empty(&block_group->bg_list));
10205 spin_lock(&trans->transaction->deleted_bgs_lock);
10206 list_move(&block_group->bg_list,
10207 &trans->transaction->deleted_bgs);
10208 spin_unlock(&trans->transaction->deleted_bgs_lock);
10209 btrfs_get_block_group(block_group);
10212 btrfs_end_transaction(trans, root);
10214 mutex_unlock(&root->fs_info->delete_unused_bgs_mutex);
10215 btrfs_put_block_group(block_group);
10216 spin_lock(&fs_info->unused_bgs_lock);
10218 spin_unlock(&fs_info->unused_bgs_lock);
10221 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
10223 struct btrfs_space_info *space_info;
10224 struct btrfs_super_block *disk_super;
10230 disk_super = fs_info->super_copy;
10231 if (!btrfs_super_root(disk_super))
10234 features = btrfs_super_incompat_flags(disk_super);
10235 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
10238 flags = BTRFS_BLOCK_GROUP_SYSTEM;
10239 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10244 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
10245 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10247 flags = BTRFS_BLOCK_GROUP_METADATA;
10248 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10252 flags = BTRFS_BLOCK_GROUP_DATA;
10253 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
10259 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
10261 return unpin_extent_range(root, start, end, false);
10265 * It used to be that old block groups would be left around forever.
10266 * Iterating over them would be enough to trim unused space. Since we
10267 * now automatically remove them, we also need to iterate over unallocated
10270 * We don't want a transaction for this since the discard may take a
10271 * substantial amount of time. We don't require that a transaction be
10272 * running, but we do need to take a running transaction into account
10273 * to ensure that we're not discarding chunks that were released in
10274 * the current transaction.
10276 * Holding the chunks lock will prevent other threads from allocating
10277 * or releasing chunks, but it won't prevent a running transaction
10278 * from committing and releasing the memory that the pending chunks
10279 * list head uses. For that, we need to take a reference to the
10282 static int btrfs_trim_free_extents(struct btrfs_device *device,
10283 u64 minlen, u64 *trimmed)
10285 u64 start = 0, len = 0;
10290 /* Not writeable = nothing to do. */
10291 if (!device->writeable)
10294 /* No free space = nothing to do. */
10295 if (device->total_bytes <= device->bytes_used)
10301 struct btrfs_fs_info *fs_info = device->dev_root->fs_info;
10302 struct btrfs_transaction *trans;
10305 ret = mutex_lock_interruptible(&fs_info->chunk_mutex);
10309 down_read(&fs_info->commit_root_sem);
10311 spin_lock(&fs_info->trans_lock);
10312 trans = fs_info->running_transaction;
10314 atomic_inc(&trans->use_count);
10315 spin_unlock(&fs_info->trans_lock);
10317 ret = find_free_dev_extent_start(trans, device, minlen, start,
10320 btrfs_put_transaction(trans);
10323 up_read(&fs_info->commit_root_sem);
10324 mutex_unlock(&fs_info->chunk_mutex);
10325 if (ret == -ENOSPC)
10330 ret = btrfs_issue_discard(device->bdev, start, len, &bytes);
10331 up_read(&fs_info->commit_root_sem);
10332 mutex_unlock(&fs_info->chunk_mutex);
10340 if (fatal_signal_pending(current)) {
10341 ret = -ERESTARTSYS;
10351 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
10353 struct btrfs_fs_info *fs_info = root->fs_info;
10354 struct btrfs_block_group_cache *cache = NULL;
10355 struct btrfs_device *device;
10356 struct list_head *devices;
10361 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
10365 * try to trim all FS space, our block group may start from non-zero.
10367 if (range->len == total_bytes)
10368 cache = btrfs_lookup_first_block_group(fs_info, range->start);
10370 cache = btrfs_lookup_block_group(fs_info, range->start);
10373 if (cache->key.objectid >= (range->start + range->len)) {
10374 btrfs_put_block_group(cache);
10378 start = max(range->start, cache->key.objectid);
10379 end = min(range->start + range->len,
10380 cache->key.objectid + cache->key.offset);
10382 if (end - start >= range->minlen) {
10383 if (!block_group_cache_done(cache)) {
10384 ret = cache_block_group(cache, 0);
10386 btrfs_put_block_group(cache);
10389 ret = wait_block_group_cache_done(cache);
10391 btrfs_put_block_group(cache);
10395 ret = btrfs_trim_block_group(cache,
10401 trimmed += group_trimmed;
10403 btrfs_put_block_group(cache);
10408 cache = next_block_group(fs_info->tree_root, cache);
10411 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
10412 devices = &root->fs_info->fs_devices->alloc_list;
10413 list_for_each_entry(device, devices, dev_alloc_list) {
10414 ret = btrfs_trim_free_extents(device, range->minlen,
10419 trimmed += group_trimmed;
10421 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
10423 range->len = trimmed;
10428 * btrfs_{start,end}_write_no_snapshoting() are similar to
10429 * mnt_{want,drop}_write(), they are used to prevent some tasks from writing
10430 * data into the page cache through nocow before the subvolume is snapshoted,
10431 * but flush the data into disk after the snapshot creation, or to prevent
10432 * operations while snapshoting is ongoing and that cause the snapshot to be
10433 * inconsistent (writes followed by expanding truncates for example).
10435 void btrfs_end_write_no_snapshoting(struct btrfs_root *root)
10437 percpu_counter_dec(&root->subv_writers->counter);
10439 * Make sure counter is updated before we wake up waiters.
10442 if (waitqueue_active(&root->subv_writers->wait))
10443 wake_up(&root->subv_writers->wait);
10446 int btrfs_start_write_no_snapshoting(struct btrfs_root *root)
10448 if (atomic_read(&root->will_be_snapshoted))
10451 percpu_counter_inc(&root->subv_writers->counter);
10453 * Make sure counter is updated before we check for snapshot creation.
10456 if (atomic_read(&root->will_be_snapshoted)) {
10457 btrfs_end_write_no_snapshoting(root);