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_root *root,
78 u64 bytenr, u64 num_bytes, int alloc);
79 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
80 struct btrfs_root *root,
81 u64 bytenr, u64 num_bytes, u64 parent,
82 u64 root_objectid, u64 owner_objectid,
83 u64 owner_offset, int refs_to_drop,
84 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->cached != BTRFS_CACHE_STARTED) {
319 spin_unlock(&cache->lock);
323 /* We're loading it the fast way, so we don't have a caching_ctl. */
324 if (!cache->caching_ctl) {
325 spin_unlock(&cache->lock);
329 ctl = cache->caching_ctl;
330 atomic_inc(&ctl->count);
331 spin_unlock(&cache->lock);
335 static void put_caching_control(struct btrfs_caching_control *ctl)
337 if (atomic_dec_and_test(&ctl->count))
342 * this is only called by cache_block_group, since we could have freed extents
343 * we need to check the pinned_extents for any extents that can't be used yet
344 * since their free space will be released as soon as the transaction commits.
346 static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
347 struct btrfs_fs_info *info, u64 start, u64 end)
349 u64 extent_start, extent_end, size, total_added = 0;
352 while (start < end) {
353 ret = find_first_extent_bit(info->pinned_extents, start,
354 &extent_start, &extent_end,
355 EXTENT_DIRTY | EXTENT_UPTODATE,
360 if (extent_start <= start) {
361 start = extent_end + 1;
362 } else if (extent_start > start && extent_start < end) {
363 size = extent_start - start;
365 ret = btrfs_add_free_space(block_group, start,
367 BUG_ON(ret); /* -ENOMEM or logic error */
368 start = extent_end + 1;
377 ret = btrfs_add_free_space(block_group, start, size);
378 BUG_ON(ret); /* -ENOMEM or logic error */
384 static noinline void caching_thread(struct btrfs_work *work)
386 struct btrfs_block_group_cache *block_group;
387 struct btrfs_fs_info *fs_info;
388 struct btrfs_caching_control *caching_ctl;
389 struct btrfs_root *extent_root;
390 struct btrfs_path *path;
391 struct extent_buffer *leaf;
392 struct btrfs_key key;
398 caching_ctl = container_of(work, struct btrfs_caching_control, work);
399 block_group = caching_ctl->block_group;
400 fs_info = block_group->fs_info;
401 extent_root = fs_info->extent_root;
403 path = btrfs_alloc_path();
407 last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
410 * We don't want to deadlock with somebody trying to allocate a new
411 * extent for the extent root while also trying to search the extent
412 * root to add free space. So we skip locking and search the commit
413 * root, since its read-only
415 path->skip_locking = 1;
416 path->search_commit_root = 1;
421 key.type = BTRFS_EXTENT_ITEM_KEY;
423 mutex_lock(&caching_ctl->mutex);
424 /* need to make sure the commit_root doesn't disappear */
425 down_read(&fs_info->commit_root_sem);
428 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
432 leaf = path->nodes[0];
433 nritems = btrfs_header_nritems(leaf);
436 if (btrfs_fs_closing(fs_info) > 1) {
441 if (path->slots[0] < nritems) {
442 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
444 ret = find_next_key(path, 0, &key);
448 if (need_resched() ||
449 rwsem_is_contended(&fs_info->commit_root_sem)) {
450 caching_ctl->progress = last;
451 btrfs_release_path(path);
452 up_read(&fs_info->commit_root_sem);
453 mutex_unlock(&caching_ctl->mutex);
458 ret = btrfs_next_leaf(extent_root, path);
463 leaf = path->nodes[0];
464 nritems = btrfs_header_nritems(leaf);
468 if (key.objectid < last) {
471 key.type = BTRFS_EXTENT_ITEM_KEY;
473 caching_ctl->progress = last;
474 btrfs_release_path(path);
478 if (key.objectid < block_group->key.objectid) {
483 if (key.objectid >= block_group->key.objectid +
484 block_group->key.offset)
487 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
488 key.type == BTRFS_METADATA_ITEM_KEY) {
489 total_found += add_new_free_space(block_group,
492 if (key.type == BTRFS_METADATA_ITEM_KEY)
493 last = key.objectid +
494 fs_info->tree_root->nodesize;
496 last = key.objectid + key.offset;
498 if (total_found > (1024 * 1024 * 2)) {
500 wake_up(&caching_ctl->wait);
507 total_found += add_new_free_space(block_group, fs_info, last,
508 block_group->key.objectid +
509 block_group->key.offset);
510 caching_ctl->progress = (u64)-1;
512 spin_lock(&block_group->lock);
513 block_group->caching_ctl = NULL;
514 block_group->cached = BTRFS_CACHE_FINISHED;
515 spin_unlock(&block_group->lock);
518 btrfs_free_path(path);
519 up_read(&fs_info->commit_root_sem);
521 free_excluded_extents(extent_root, block_group);
523 mutex_unlock(&caching_ctl->mutex);
526 spin_lock(&block_group->lock);
527 block_group->caching_ctl = NULL;
528 block_group->cached = BTRFS_CACHE_ERROR;
529 spin_unlock(&block_group->lock);
531 wake_up(&caching_ctl->wait);
533 put_caching_control(caching_ctl);
534 btrfs_put_block_group(block_group);
537 static int cache_block_group(struct btrfs_block_group_cache *cache,
541 struct btrfs_fs_info *fs_info = cache->fs_info;
542 struct btrfs_caching_control *caching_ctl;
545 caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
549 INIT_LIST_HEAD(&caching_ctl->list);
550 mutex_init(&caching_ctl->mutex);
551 init_waitqueue_head(&caching_ctl->wait);
552 caching_ctl->block_group = cache;
553 caching_ctl->progress = cache->key.objectid;
554 atomic_set(&caching_ctl->count, 1);
555 btrfs_init_work(&caching_ctl->work, btrfs_cache_helper,
556 caching_thread, NULL, NULL);
558 spin_lock(&cache->lock);
560 * This should be a rare occasion, but this could happen I think in the
561 * case where one thread starts to load the space cache info, and then
562 * some other thread starts a transaction commit which tries to do an
563 * allocation while the other thread is still loading the space cache
564 * info. The previous loop should have kept us from choosing this block
565 * group, but if we've moved to the state where we will wait on caching
566 * block groups we need to first check if we're doing a fast load here,
567 * so we can wait for it to finish, otherwise we could end up allocating
568 * from a block group who's cache gets evicted for one reason or
571 while (cache->cached == BTRFS_CACHE_FAST) {
572 struct btrfs_caching_control *ctl;
574 ctl = cache->caching_ctl;
575 atomic_inc(&ctl->count);
576 prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE);
577 spin_unlock(&cache->lock);
581 finish_wait(&ctl->wait, &wait);
582 put_caching_control(ctl);
583 spin_lock(&cache->lock);
586 if (cache->cached != BTRFS_CACHE_NO) {
587 spin_unlock(&cache->lock);
591 WARN_ON(cache->caching_ctl);
592 cache->caching_ctl = caching_ctl;
593 cache->cached = BTRFS_CACHE_FAST;
594 spin_unlock(&cache->lock);
596 if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
597 ret = load_free_space_cache(fs_info, cache);
599 spin_lock(&cache->lock);
601 cache->caching_ctl = NULL;
602 cache->cached = BTRFS_CACHE_FINISHED;
603 cache->last_byte_to_unpin = (u64)-1;
605 if (load_cache_only) {
606 cache->caching_ctl = NULL;
607 cache->cached = BTRFS_CACHE_NO;
609 cache->cached = BTRFS_CACHE_STARTED;
612 spin_unlock(&cache->lock);
613 wake_up(&caching_ctl->wait);
615 put_caching_control(caching_ctl);
616 free_excluded_extents(fs_info->extent_root, cache);
621 * We are not going to do the fast caching, set cached to the
622 * appropriate value and wakeup any waiters.
624 spin_lock(&cache->lock);
625 if (load_cache_only) {
626 cache->caching_ctl = NULL;
627 cache->cached = BTRFS_CACHE_NO;
629 cache->cached = BTRFS_CACHE_STARTED;
631 spin_unlock(&cache->lock);
632 wake_up(&caching_ctl->wait);
635 if (load_cache_only) {
636 put_caching_control(caching_ctl);
640 down_write(&fs_info->commit_root_sem);
641 atomic_inc(&caching_ctl->count);
642 list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups);
643 up_write(&fs_info->commit_root_sem);
645 btrfs_get_block_group(cache);
647 btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work);
653 * return the block group that starts at or after bytenr
655 static struct btrfs_block_group_cache *
656 btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr)
658 struct btrfs_block_group_cache *cache;
660 cache = block_group_cache_tree_search(info, bytenr, 0);
666 * return the block group that contains the given bytenr
668 struct btrfs_block_group_cache *btrfs_lookup_block_group(
669 struct btrfs_fs_info *info,
672 struct btrfs_block_group_cache *cache;
674 cache = block_group_cache_tree_search(info, bytenr, 1);
679 static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info,
682 struct list_head *head = &info->space_info;
683 struct btrfs_space_info *found;
685 flags &= BTRFS_BLOCK_GROUP_TYPE_MASK;
688 list_for_each_entry_rcu(found, head, list) {
689 if (found->flags & flags) {
699 * after adding space to the filesystem, we need to clear the full flags
700 * on all the space infos.
702 void btrfs_clear_space_info_full(struct btrfs_fs_info *info)
704 struct list_head *head = &info->space_info;
705 struct btrfs_space_info *found;
708 list_for_each_entry_rcu(found, head, list)
713 /* simple helper to search for an existing extent at a given offset */
714 int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len)
717 struct btrfs_key key;
718 struct btrfs_path *path;
720 path = btrfs_alloc_path();
724 key.objectid = start;
726 key.type = BTRFS_EXTENT_ITEM_KEY;
727 ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path,
730 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
731 if (key.objectid == start &&
732 key.type == BTRFS_METADATA_ITEM_KEY)
735 btrfs_free_path(path);
740 * helper function to lookup reference count and flags of a tree block.
742 * the head node for delayed ref is used to store the sum of all the
743 * reference count modifications queued up in the rbtree. the head
744 * node may also store the extent flags to set. This way you can check
745 * to see what the reference count and extent flags would be if all of
746 * the delayed refs are not processed.
748 int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans,
749 struct btrfs_root *root, u64 bytenr,
750 u64 offset, int metadata, u64 *refs, u64 *flags)
752 struct btrfs_delayed_ref_head *head;
753 struct btrfs_delayed_ref_root *delayed_refs;
754 struct btrfs_path *path;
755 struct btrfs_extent_item *ei;
756 struct extent_buffer *leaf;
757 struct btrfs_key key;
764 * If we don't have skinny metadata, don't bother doing anything
767 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) {
768 offset = root->nodesize;
772 path = btrfs_alloc_path();
777 path->skip_locking = 1;
778 path->search_commit_root = 1;
782 key.objectid = bytenr;
785 key.type = BTRFS_METADATA_ITEM_KEY;
787 key.type = BTRFS_EXTENT_ITEM_KEY;
790 ret = btrfs_search_slot(trans, root->fs_info->extent_root,
795 if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) {
796 if (path->slots[0]) {
798 btrfs_item_key_to_cpu(path->nodes[0], &key,
800 if (key.objectid == bytenr &&
801 key.type == BTRFS_EXTENT_ITEM_KEY &&
802 key.offset == root->nodesize)
806 key.objectid = bytenr;
807 key.type = BTRFS_EXTENT_ITEM_KEY;
808 key.offset = root->nodesize;
809 btrfs_release_path(path);
815 leaf = path->nodes[0];
816 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
817 if (item_size >= sizeof(*ei)) {
818 ei = btrfs_item_ptr(leaf, path->slots[0],
819 struct btrfs_extent_item);
820 num_refs = btrfs_extent_refs(leaf, ei);
821 extent_flags = btrfs_extent_flags(leaf, ei);
823 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
824 struct btrfs_extent_item_v0 *ei0;
825 BUG_ON(item_size != sizeof(*ei0));
826 ei0 = btrfs_item_ptr(leaf, path->slots[0],
827 struct btrfs_extent_item_v0);
828 num_refs = btrfs_extent_refs_v0(leaf, ei0);
829 /* FIXME: this isn't correct for data */
830 extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
835 BUG_ON(num_refs == 0);
845 delayed_refs = &trans->transaction->delayed_refs;
846 spin_lock(&delayed_refs->lock);
847 head = btrfs_find_delayed_ref_head(trans, bytenr);
849 if (!mutex_trylock(&head->mutex)) {
850 atomic_inc(&head->node.refs);
851 spin_unlock(&delayed_refs->lock);
853 btrfs_release_path(path);
856 * Mutex was contended, block until it's released and try
859 mutex_lock(&head->mutex);
860 mutex_unlock(&head->mutex);
861 btrfs_put_delayed_ref(&head->node);
864 spin_lock(&head->lock);
865 if (head->extent_op && head->extent_op->update_flags)
866 extent_flags |= head->extent_op->flags_to_set;
868 BUG_ON(num_refs == 0);
870 num_refs += head->node.ref_mod;
871 spin_unlock(&head->lock);
872 mutex_unlock(&head->mutex);
874 spin_unlock(&delayed_refs->lock);
876 WARN_ON(num_refs == 0);
880 *flags = extent_flags;
882 btrfs_free_path(path);
887 * Back reference rules. Back refs have three main goals:
889 * 1) differentiate between all holders of references to an extent so that
890 * when a reference is dropped we can make sure it was a valid reference
891 * before freeing the extent.
893 * 2) Provide enough information to quickly find the holders of an extent
894 * if we notice a given block is corrupted or bad.
896 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
897 * maintenance. This is actually the same as #2, but with a slightly
898 * different use case.
900 * There are two kinds of back refs. The implicit back refs is optimized
901 * for pointers in non-shared tree blocks. For a given pointer in a block,
902 * back refs of this kind provide information about the block's owner tree
903 * and the pointer's key. These information allow us to find the block by
904 * b-tree searching. The full back refs is for pointers in tree blocks not
905 * referenced by their owner trees. The location of tree block is recorded
906 * in the back refs. Actually the full back refs is generic, and can be
907 * used in all cases the implicit back refs is used. The major shortcoming
908 * of the full back refs is its overhead. Every time a tree block gets
909 * COWed, we have to update back refs entry for all pointers in it.
911 * For a newly allocated tree block, we use implicit back refs for
912 * pointers in it. This means most tree related operations only involve
913 * implicit back refs. For a tree block created in old transaction, the
914 * only way to drop a reference to it is COW it. So we can detect the
915 * event that tree block loses its owner tree's reference and do the
916 * back refs conversion.
918 * When a tree block is COW'd through a tree, there are four cases:
920 * The reference count of the block is one and the tree is the block's
921 * owner tree. Nothing to do in this case.
923 * The reference count of the block is one and the tree is not the
924 * block's owner tree. In this case, full back refs is used for pointers
925 * in the block. Remove these full back refs, add implicit back refs for
926 * every pointers in the new block.
928 * The reference count of the block is greater than one and the tree is
929 * the block's owner tree. In this case, implicit back refs is used for
930 * pointers in the block. Add full back refs for every pointers in the
931 * block, increase lower level extents' reference counts. The original
932 * implicit back refs are entailed to the new block.
934 * The reference count of the block is greater than one and the tree is
935 * not the block's owner tree. Add implicit back refs for every pointer in
936 * the new block, increase lower level extents' reference count.
938 * Back Reference Key composing:
940 * The key objectid corresponds to the first byte in the extent,
941 * The key type is used to differentiate between types of back refs.
942 * There are different meanings of the key offset for different types
945 * File extents can be referenced by:
947 * - multiple snapshots, subvolumes, or different generations in one subvol
948 * - different files inside a single subvolume
949 * - different offsets inside a file (bookend extents in file.c)
951 * The extent ref structure for the implicit back refs has fields for:
953 * - Objectid of the subvolume root
954 * - objectid of the file holding the reference
955 * - original offset in the file
956 * - how many bookend extents
958 * The key offset for the implicit back refs is hash of the first
961 * The extent ref structure for the full back refs has field for:
963 * - number of pointers in the tree leaf
965 * The key offset for the implicit back refs is the first byte of
968 * When a file extent is allocated, The implicit back refs is used.
969 * the fields are filled in:
971 * (root_key.objectid, inode objectid, offset in file, 1)
973 * When a file extent is removed file truncation, we find the
974 * corresponding implicit back refs and check the following fields:
976 * (btrfs_header_owner(leaf), inode objectid, offset in file)
978 * Btree extents can be referenced by:
980 * - Different subvolumes
982 * Both the implicit back refs and the full back refs for tree blocks
983 * only consist of key. The key offset for the implicit back refs is
984 * objectid of block's owner tree. The key offset for the full back refs
985 * is the first byte of parent block.
987 * When implicit back refs is used, information about the lowest key and
988 * level of the tree block are required. These information are stored in
989 * tree block info structure.
992 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
993 static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
994 struct btrfs_root *root,
995 struct btrfs_path *path,
996 u64 owner, u32 extra_size)
998 struct btrfs_extent_item *item;
999 struct btrfs_extent_item_v0 *ei0;
1000 struct btrfs_extent_ref_v0 *ref0;
1001 struct btrfs_tree_block_info *bi;
1002 struct extent_buffer *leaf;
1003 struct btrfs_key key;
1004 struct btrfs_key found_key;
1005 u32 new_size = sizeof(*item);
1009 leaf = path->nodes[0];
1010 BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0));
1012 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1013 ei0 = btrfs_item_ptr(leaf, path->slots[0],
1014 struct btrfs_extent_item_v0);
1015 refs = btrfs_extent_refs_v0(leaf, ei0);
1017 if (owner == (u64)-1) {
1019 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1020 ret = btrfs_next_leaf(root, path);
1023 BUG_ON(ret > 0); /* Corruption */
1024 leaf = path->nodes[0];
1026 btrfs_item_key_to_cpu(leaf, &found_key,
1028 BUG_ON(key.objectid != found_key.objectid);
1029 if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) {
1033 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1034 struct btrfs_extent_ref_v0);
1035 owner = btrfs_ref_objectid_v0(leaf, ref0);
1039 btrfs_release_path(path);
1041 if (owner < BTRFS_FIRST_FREE_OBJECTID)
1042 new_size += sizeof(*bi);
1044 new_size -= sizeof(*ei0);
1045 ret = btrfs_search_slot(trans, root, &key, path,
1046 new_size + extra_size, 1);
1049 BUG_ON(ret); /* Corruption */
1051 btrfs_extend_item(root, path, new_size);
1053 leaf = path->nodes[0];
1054 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1055 btrfs_set_extent_refs(leaf, item, refs);
1056 /* FIXME: get real generation */
1057 btrfs_set_extent_generation(leaf, item, 0);
1058 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1059 btrfs_set_extent_flags(leaf, item,
1060 BTRFS_EXTENT_FLAG_TREE_BLOCK |
1061 BTRFS_BLOCK_FLAG_FULL_BACKREF);
1062 bi = (struct btrfs_tree_block_info *)(item + 1);
1063 /* FIXME: get first key of the block */
1064 memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi));
1065 btrfs_set_tree_block_level(leaf, bi, (int)owner);
1067 btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA);
1069 btrfs_mark_buffer_dirty(leaf);
1074 static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset)
1076 u32 high_crc = ~(u32)0;
1077 u32 low_crc = ~(u32)0;
1080 lenum = cpu_to_le64(root_objectid);
1081 high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum));
1082 lenum = cpu_to_le64(owner);
1083 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1084 lenum = cpu_to_le64(offset);
1085 low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum));
1087 return ((u64)high_crc << 31) ^ (u64)low_crc;
1090 static u64 hash_extent_data_ref_item(struct extent_buffer *leaf,
1091 struct btrfs_extent_data_ref *ref)
1093 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref),
1094 btrfs_extent_data_ref_objectid(leaf, ref),
1095 btrfs_extent_data_ref_offset(leaf, ref));
1098 static int match_extent_data_ref(struct extent_buffer *leaf,
1099 struct btrfs_extent_data_ref *ref,
1100 u64 root_objectid, u64 owner, u64 offset)
1102 if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid ||
1103 btrfs_extent_data_ref_objectid(leaf, ref) != owner ||
1104 btrfs_extent_data_ref_offset(leaf, ref) != offset)
1109 static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans,
1110 struct btrfs_root *root,
1111 struct btrfs_path *path,
1112 u64 bytenr, u64 parent,
1114 u64 owner, u64 offset)
1116 struct btrfs_key key;
1117 struct btrfs_extent_data_ref *ref;
1118 struct extent_buffer *leaf;
1124 key.objectid = bytenr;
1126 key.type = BTRFS_SHARED_DATA_REF_KEY;
1127 key.offset = parent;
1129 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1130 key.offset = hash_extent_data_ref(root_objectid,
1135 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1144 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1145 key.type = BTRFS_EXTENT_REF_V0_KEY;
1146 btrfs_release_path(path);
1147 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1158 leaf = path->nodes[0];
1159 nritems = btrfs_header_nritems(leaf);
1161 if (path->slots[0] >= nritems) {
1162 ret = btrfs_next_leaf(root, path);
1168 leaf = path->nodes[0];
1169 nritems = btrfs_header_nritems(leaf);
1173 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1174 if (key.objectid != bytenr ||
1175 key.type != BTRFS_EXTENT_DATA_REF_KEY)
1178 ref = btrfs_item_ptr(leaf, path->slots[0],
1179 struct btrfs_extent_data_ref);
1181 if (match_extent_data_ref(leaf, ref, root_objectid,
1184 btrfs_release_path(path);
1196 static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans,
1197 struct btrfs_root *root,
1198 struct btrfs_path *path,
1199 u64 bytenr, u64 parent,
1200 u64 root_objectid, u64 owner,
1201 u64 offset, int refs_to_add)
1203 struct btrfs_key key;
1204 struct extent_buffer *leaf;
1209 key.objectid = bytenr;
1211 key.type = BTRFS_SHARED_DATA_REF_KEY;
1212 key.offset = parent;
1213 size = sizeof(struct btrfs_shared_data_ref);
1215 key.type = BTRFS_EXTENT_DATA_REF_KEY;
1216 key.offset = hash_extent_data_ref(root_objectid,
1218 size = sizeof(struct btrfs_extent_data_ref);
1221 ret = btrfs_insert_empty_item(trans, root, path, &key, size);
1222 if (ret && ret != -EEXIST)
1225 leaf = path->nodes[0];
1227 struct btrfs_shared_data_ref *ref;
1228 ref = btrfs_item_ptr(leaf, path->slots[0],
1229 struct btrfs_shared_data_ref);
1231 btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add);
1233 num_refs = btrfs_shared_data_ref_count(leaf, ref);
1234 num_refs += refs_to_add;
1235 btrfs_set_shared_data_ref_count(leaf, ref, num_refs);
1238 struct btrfs_extent_data_ref *ref;
1239 while (ret == -EEXIST) {
1240 ref = btrfs_item_ptr(leaf, path->slots[0],
1241 struct btrfs_extent_data_ref);
1242 if (match_extent_data_ref(leaf, ref, root_objectid,
1245 btrfs_release_path(path);
1247 ret = btrfs_insert_empty_item(trans, root, path, &key,
1249 if (ret && ret != -EEXIST)
1252 leaf = path->nodes[0];
1254 ref = btrfs_item_ptr(leaf, path->slots[0],
1255 struct btrfs_extent_data_ref);
1257 btrfs_set_extent_data_ref_root(leaf, ref,
1259 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
1260 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
1261 btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add);
1263 num_refs = btrfs_extent_data_ref_count(leaf, ref);
1264 num_refs += refs_to_add;
1265 btrfs_set_extent_data_ref_count(leaf, ref, num_refs);
1268 btrfs_mark_buffer_dirty(leaf);
1271 btrfs_release_path(path);
1275 static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans,
1276 struct btrfs_root *root,
1277 struct btrfs_path *path,
1278 int refs_to_drop, int *last_ref)
1280 struct btrfs_key key;
1281 struct btrfs_extent_data_ref *ref1 = NULL;
1282 struct btrfs_shared_data_ref *ref2 = NULL;
1283 struct extent_buffer *leaf;
1287 leaf = path->nodes[0];
1288 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1290 if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1291 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1292 struct btrfs_extent_data_ref);
1293 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1294 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1295 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1296 struct btrfs_shared_data_ref);
1297 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1298 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1299 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1300 struct btrfs_extent_ref_v0 *ref0;
1301 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1302 struct btrfs_extent_ref_v0);
1303 num_refs = btrfs_ref_count_v0(leaf, ref0);
1309 BUG_ON(num_refs < refs_to_drop);
1310 num_refs -= refs_to_drop;
1312 if (num_refs == 0) {
1313 ret = btrfs_del_item(trans, root, path);
1316 if (key.type == BTRFS_EXTENT_DATA_REF_KEY)
1317 btrfs_set_extent_data_ref_count(leaf, ref1, num_refs);
1318 else if (key.type == BTRFS_SHARED_DATA_REF_KEY)
1319 btrfs_set_shared_data_ref_count(leaf, ref2, num_refs);
1320 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1322 struct btrfs_extent_ref_v0 *ref0;
1323 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1324 struct btrfs_extent_ref_v0);
1325 btrfs_set_ref_count_v0(leaf, ref0, num_refs);
1328 btrfs_mark_buffer_dirty(leaf);
1333 static noinline u32 extent_data_ref_count(struct btrfs_root *root,
1334 struct btrfs_path *path,
1335 struct btrfs_extent_inline_ref *iref)
1337 struct btrfs_key key;
1338 struct extent_buffer *leaf;
1339 struct btrfs_extent_data_ref *ref1;
1340 struct btrfs_shared_data_ref *ref2;
1343 leaf = path->nodes[0];
1344 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1346 if (btrfs_extent_inline_ref_type(leaf, iref) ==
1347 BTRFS_EXTENT_DATA_REF_KEY) {
1348 ref1 = (struct btrfs_extent_data_ref *)(&iref->offset);
1349 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1351 ref2 = (struct btrfs_shared_data_ref *)(iref + 1);
1352 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1354 } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
1355 ref1 = btrfs_item_ptr(leaf, path->slots[0],
1356 struct btrfs_extent_data_ref);
1357 num_refs = btrfs_extent_data_ref_count(leaf, ref1);
1358 } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
1359 ref2 = btrfs_item_ptr(leaf, path->slots[0],
1360 struct btrfs_shared_data_ref);
1361 num_refs = btrfs_shared_data_ref_count(leaf, ref2);
1362 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1363 } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
1364 struct btrfs_extent_ref_v0 *ref0;
1365 ref0 = btrfs_item_ptr(leaf, path->slots[0],
1366 struct btrfs_extent_ref_v0);
1367 num_refs = btrfs_ref_count_v0(leaf, ref0);
1375 static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans,
1376 struct btrfs_root *root,
1377 struct btrfs_path *path,
1378 u64 bytenr, u64 parent,
1381 struct btrfs_key key;
1384 key.objectid = bytenr;
1386 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1387 key.offset = parent;
1389 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1390 key.offset = root_objectid;
1393 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1396 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1397 if (ret == -ENOENT && parent) {
1398 btrfs_release_path(path);
1399 key.type = BTRFS_EXTENT_REF_V0_KEY;
1400 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1408 static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans,
1409 struct btrfs_root *root,
1410 struct btrfs_path *path,
1411 u64 bytenr, u64 parent,
1414 struct btrfs_key key;
1417 key.objectid = bytenr;
1419 key.type = BTRFS_SHARED_BLOCK_REF_KEY;
1420 key.offset = parent;
1422 key.type = BTRFS_TREE_BLOCK_REF_KEY;
1423 key.offset = root_objectid;
1426 ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
1427 btrfs_release_path(path);
1431 static inline int extent_ref_type(u64 parent, u64 owner)
1434 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1436 type = BTRFS_SHARED_BLOCK_REF_KEY;
1438 type = BTRFS_TREE_BLOCK_REF_KEY;
1441 type = BTRFS_SHARED_DATA_REF_KEY;
1443 type = BTRFS_EXTENT_DATA_REF_KEY;
1448 static int find_next_key(struct btrfs_path *path, int level,
1449 struct btrfs_key *key)
1452 for (; level < BTRFS_MAX_LEVEL; level++) {
1453 if (!path->nodes[level])
1455 if (path->slots[level] + 1 >=
1456 btrfs_header_nritems(path->nodes[level]))
1459 btrfs_item_key_to_cpu(path->nodes[level], key,
1460 path->slots[level] + 1);
1462 btrfs_node_key_to_cpu(path->nodes[level], key,
1463 path->slots[level] + 1);
1470 * look for inline back ref. if back ref is found, *ref_ret is set
1471 * to the address of inline back ref, and 0 is returned.
1473 * if back ref isn't found, *ref_ret is set to the address where it
1474 * should be inserted, and -ENOENT is returned.
1476 * if insert is true and there are too many inline back refs, the path
1477 * points to the extent item, and -EAGAIN is returned.
1479 * NOTE: inline back refs are ordered in the same way that back ref
1480 * items in the tree are ordered.
1482 static noinline_for_stack
1483 int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
1484 struct btrfs_root *root,
1485 struct btrfs_path *path,
1486 struct btrfs_extent_inline_ref **ref_ret,
1487 u64 bytenr, u64 num_bytes,
1488 u64 parent, u64 root_objectid,
1489 u64 owner, u64 offset, int insert)
1491 struct btrfs_key key;
1492 struct extent_buffer *leaf;
1493 struct btrfs_extent_item *ei;
1494 struct btrfs_extent_inline_ref *iref;
1504 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
1507 key.objectid = bytenr;
1508 key.type = BTRFS_EXTENT_ITEM_KEY;
1509 key.offset = num_bytes;
1511 want = extent_ref_type(parent, owner);
1513 extra_size = btrfs_extent_inline_ref_size(want);
1514 path->keep_locks = 1;
1519 * Owner is our parent level, so we can just add one to get the level
1520 * for the block we are interested in.
1522 if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) {
1523 key.type = BTRFS_METADATA_ITEM_KEY;
1528 ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1);
1535 * We may be a newly converted file system which still has the old fat
1536 * extent entries for metadata, so try and see if we have one of those.
1538 if (ret > 0 && skinny_metadata) {
1539 skinny_metadata = false;
1540 if (path->slots[0]) {
1542 btrfs_item_key_to_cpu(path->nodes[0], &key,
1544 if (key.objectid == bytenr &&
1545 key.type == BTRFS_EXTENT_ITEM_KEY &&
1546 key.offset == num_bytes)
1550 key.objectid = bytenr;
1551 key.type = BTRFS_EXTENT_ITEM_KEY;
1552 key.offset = num_bytes;
1553 btrfs_release_path(path);
1558 if (ret && !insert) {
1561 } else if (WARN_ON(ret)) {
1566 leaf = path->nodes[0];
1567 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1568 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1569 if (item_size < sizeof(*ei)) {
1574 ret = convert_extent_item_v0(trans, root, path, owner,
1580 leaf = path->nodes[0];
1581 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1584 BUG_ON(item_size < sizeof(*ei));
1586 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1587 flags = btrfs_extent_flags(leaf, ei);
1589 ptr = (unsigned long)(ei + 1);
1590 end = (unsigned long)ei + item_size;
1592 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) {
1593 ptr += sizeof(struct btrfs_tree_block_info);
1603 iref = (struct btrfs_extent_inline_ref *)ptr;
1604 type = btrfs_extent_inline_ref_type(leaf, iref);
1608 ptr += btrfs_extent_inline_ref_size(type);
1612 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1613 struct btrfs_extent_data_ref *dref;
1614 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1615 if (match_extent_data_ref(leaf, dref, root_objectid,
1620 if (hash_extent_data_ref_item(leaf, dref) <
1621 hash_extent_data_ref(root_objectid, owner, offset))
1625 ref_offset = btrfs_extent_inline_ref_offset(leaf, iref);
1627 if (parent == ref_offset) {
1631 if (ref_offset < parent)
1634 if (root_objectid == ref_offset) {
1638 if (ref_offset < root_objectid)
1642 ptr += btrfs_extent_inline_ref_size(type);
1644 if (err == -ENOENT && insert) {
1645 if (item_size + extra_size >=
1646 BTRFS_MAX_EXTENT_ITEM_SIZE(root)) {
1651 * To add new inline back ref, we have to make sure
1652 * there is no corresponding back ref item.
1653 * For simplicity, we just do not add new inline back
1654 * ref if there is any kind of item for this block
1656 if (find_next_key(path, 0, &key) == 0 &&
1657 key.objectid == bytenr &&
1658 key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
1663 *ref_ret = (struct btrfs_extent_inline_ref *)ptr;
1666 path->keep_locks = 0;
1667 btrfs_unlock_up_safe(path, 1);
1673 * helper to add new inline back ref
1675 static noinline_for_stack
1676 void setup_inline_extent_backref(struct btrfs_root *root,
1677 struct btrfs_path *path,
1678 struct btrfs_extent_inline_ref *iref,
1679 u64 parent, u64 root_objectid,
1680 u64 owner, u64 offset, int refs_to_add,
1681 struct btrfs_delayed_extent_op *extent_op)
1683 struct extent_buffer *leaf;
1684 struct btrfs_extent_item *ei;
1687 unsigned long item_offset;
1692 leaf = path->nodes[0];
1693 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1694 item_offset = (unsigned long)iref - (unsigned long)ei;
1696 type = extent_ref_type(parent, owner);
1697 size = btrfs_extent_inline_ref_size(type);
1699 btrfs_extend_item(root, path, size);
1701 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1702 refs = btrfs_extent_refs(leaf, ei);
1703 refs += refs_to_add;
1704 btrfs_set_extent_refs(leaf, ei, refs);
1706 __run_delayed_extent_op(extent_op, leaf, ei);
1708 ptr = (unsigned long)ei + item_offset;
1709 end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]);
1710 if (ptr < end - size)
1711 memmove_extent_buffer(leaf, ptr + size, ptr,
1714 iref = (struct btrfs_extent_inline_ref *)ptr;
1715 btrfs_set_extent_inline_ref_type(leaf, iref, type);
1716 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1717 struct btrfs_extent_data_ref *dref;
1718 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1719 btrfs_set_extent_data_ref_root(leaf, dref, root_objectid);
1720 btrfs_set_extent_data_ref_objectid(leaf, dref, owner);
1721 btrfs_set_extent_data_ref_offset(leaf, dref, offset);
1722 btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add);
1723 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1724 struct btrfs_shared_data_ref *sref;
1725 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1726 btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add);
1727 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1728 } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) {
1729 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
1731 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
1733 btrfs_mark_buffer_dirty(leaf);
1736 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
1737 struct btrfs_root *root,
1738 struct btrfs_path *path,
1739 struct btrfs_extent_inline_ref **ref_ret,
1740 u64 bytenr, u64 num_bytes, u64 parent,
1741 u64 root_objectid, u64 owner, u64 offset)
1745 ret = lookup_inline_extent_backref(trans, root, path, ref_ret,
1746 bytenr, num_bytes, parent,
1747 root_objectid, owner, offset, 0);
1751 btrfs_release_path(path);
1754 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1755 ret = lookup_tree_block_ref(trans, root, path, bytenr, parent,
1758 ret = lookup_extent_data_ref(trans, root, path, bytenr, parent,
1759 root_objectid, owner, offset);
1765 * helper to update/remove inline back ref
1767 static noinline_for_stack
1768 void update_inline_extent_backref(struct btrfs_root *root,
1769 struct btrfs_path *path,
1770 struct btrfs_extent_inline_ref *iref,
1772 struct btrfs_delayed_extent_op *extent_op,
1775 struct extent_buffer *leaf;
1776 struct btrfs_extent_item *ei;
1777 struct btrfs_extent_data_ref *dref = NULL;
1778 struct btrfs_shared_data_ref *sref = NULL;
1786 leaf = path->nodes[0];
1787 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
1788 refs = btrfs_extent_refs(leaf, ei);
1789 WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0);
1790 refs += refs_to_mod;
1791 btrfs_set_extent_refs(leaf, ei, refs);
1793 __run_delayed_extent_op(extent_op, leaf, ei);
1795 type = btrfs_extent_inline_ref_type(leaf, iref);
1797 if (type == BTRFS_EXTENT_DATA_REF_KEY) {
1798 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1799 refs = btrfs_extent_data_ref_count(leaf, dref);
1800 } else if (type == BTRFS_SHARED_DATA_REF_KEY) {
1801 sref = (struct btrfs_shared_data_ref *)(iref + 1);
1802 refs = btrfs_shared_data_ref_count(leaf, sref);
1805 BUG_ON(refs_to_mod != -1);
1808 BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod);
1809 refs += refs_to_mod;
1812 if (type == BTRFS_EXTENT_DATA_REF_KEY)
1813 btrfs_set_extent_data_ref_count(leaf, dref, refs);
1815 btrfs_set_shared_data_ref_count(leaf, sref, refs);
1818 size = btrfs_extent_inline_ref_size(type);
1819 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1820 ptr = (unsigned long)iref;
1821 end = (unsigned long)ei + item_size;
1822 if (ptr + size < end)
1823 memmove_extent_buffer(leaf, ptr, ptr + size,
1826 btrfs_truncate_item(root, path, item_size, 1);
1828 btrfs_mark_buffer_dirty(leaf);
1831 static noinline_for_stack
1832 int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
1833 struct btrfs_root *root,
1834 struct btrfs_path *path,
1835 u64 bytenr, u64 num_bytes, u64 parent,
1836 u64 root_objectid, u64 owner,
1837 u64 offset, int refs_to_add,
1838 struct btrfs_delayed_extent_op *extent_op)
1840 struct btrfs_extent_inline_ref *iref;
1843 ret = lookup_inline_extent_backref(trans, root, path, &iref,
1844 bytenr, num_bytes, parent,
1845 root_objectid, owner, offset, 1);
1847 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
1848 update_inline_extent_backref(root, path, iref,
1849 refs_to_add, extent_op, NULL);
1850 } else if (ret == -ENOENT) {
1851 setup_inline_extent_backref(root, path, iref, parent,
1852 root_objectid, owner, offset,
1853 refs_to_add, extent_op);
1859 static int insert_extent_backref(struct btrfs_trans_handle *trans,
1860 struct btrfs_root *root,
1861 struct btrfs_path *path,
1862 u64 bytenr, u64 parent, u64 root_objectid,
1863 u64 owner, u64 offset, int refs_to_add)
1866 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1867 BUG_ON(refs_to_add != 1);
1868 ret = insert_tree_block_ref(trans, root, path, bytenr,
1869 parent, root_objectid);
1871 ret = insert_extent_data_ref(trans, root, path, bytenr,
1872 parent, root_objectid,
1873 owner, offset, refs_to_add);
1878 static int remove_extent_backref(struct btrfs_trans_handle *trans,
1879 struct btrfs_root *root,
1880 struct btrfs_path *path,
1881 struct btrfs_extent_inline_ref *iref,
1882 int refs_to_drop, int is_data, int *last_ref)
1886 BUG_ON(!is_data && refs_to_drop != 1);
1888 update_inline_extent_backref(root, path, iref,
1889 -refs_to_drop, NULL, last_ref);
1890 } else if (is_data) {
1891 ret = remove_extent_data_ref(trans, root, path, refs_to_drop,
1895 ret = btrfs_del_item(trans, root, path);
1900 static int btrfs_issue_discard(struct block_device *bdev,
1903 return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0);
1906 static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
1907 u64 num_bytes, u64 *actual_bytes)
1910 u64 discarded_bytes = 0;
1911 struct btrfs_bio *bbio = NULL;
1914 /* Tell the block device(s) that the sectors can be discarded */
1915 ret = btrfs_map_block(root->fs_info, REQ_DISCARD,
1916 bytenr, &num_bytes, &bbio, 0);
1917 /* Error condition is -ENOMEM */
1919 struct btrfs_bio_stripe *stripe = bbio->stripes;
1923 for (i = 0; i < bbio->num_stripes; i++, stripe++) {
1924 if (!stripe->dev->can_discard)
1927 ret = btrfs_issue_discard(stripe->dev->bdev,
1931 discarded_bytes += stripe->length;
1932 else if (ret != -EOPNOTSUPP)
1933 break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
1936 * Just in case we get back EOPNOTSUPP for some reason,
1937 * just ignore the return value so we don't screw up
1938 * people calling discard_extent.
1946 *actual_bytes = discarded_bytes;
1949 if (ret == -EOPNOTSUPP)
1954 /* Can return -ENOMEM */
1955 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1956 struct btrfs_root *root,
1957 u64 bytenr, u64 num_bytes, u64 parent,
1958 u64 root_objectid, u64 owner, u64 offset,
1962 struct btrfs_fs_info *fs_info = root->fs_info;
1964 BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
1965 root_objectid == BTRFS_TREE_LOG_OBJECTID);
1967 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
1968 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
1970 parent, root_objectid, (int)owner,
1971 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1973 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
1975 parent, root_objectid, owner, offset,
1976 BTRFS_ADD_DELAYED_REF, NULL, no_quota);
1981 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
1982 struct btrfs_root *root,
1983 u64 bytenr, u64 num_bytes,
1984 u64 parent, u64 root_objectid,
1985 u64 owner, u64 offset, int refs_to_add,
1987 struct btrfs_delayed_extent_op *extent_op)
1989 struct btrfs_fs_info *fs_info = root->fs_info;
1990 struct btrfs_path *path;
1991 struct extent_buffer *leaf;
1992 struct btrfs_extent_item *item;
1993 struct btrfs_key key;
1996 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL;
1998 path = btrfs_alloc_path();
2002 if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled)
2006 path->leave_spinning = 1;
2007 /* this will setup the path even if it fails to insert the back ref */
2008 ret = insert_inline_extent_backref(trans, fs_info->extent_root, path,
2009 bytenr, num_bytes, parent,
2010 root_objectid, owner, offset,
2011 refs_to_add, extent_op);
2012 if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota))
2015 * Ok we were able to insert an inline extent and it appears to be a new
2016 * reference, deal with the qgroup accounting.
2018 if (!ret && !no_quota) {
2019 ASSERT(root->fs_info->quota_enabled);
2020 leaf = path->nodes[0];
2021 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2022 item = btrfs_item_ptr(leaf, path->slots[0],
2023 struct btrfs_extent_item);
2024 if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add)
2025 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2026 btrfs_release_path(path);
2028 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2029 bytenr, num_bytes, type, 0);
2034 * Ok we had -EAGAIN which means we didn't have space to insert and
2035 * inline extent ref, so just update the reference count and add a
2038 leaf = path->nodes[0];
2039 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2040 item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2041 refs = btrfs_extent_refs(leaf, item);
2043 type = BTRFS_QGROUP_OPER_ADD_SHARED;
2044 btrfs_set_extent_refs(leaf, item, refs + refs_to_add);
2046 __run_delayed_extent_op(extent_op, leaf, item);
2048 btrfs_mark_buffer_dirty(leaf);
2049 btrfs_release_path(path);
2052 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
2053 bytenr, num_bytes, type, 0);
2059 path->leave_spinning = 1;
2060 /* now insert the actual backref */
2061 ret = insert_extent_backref(trans, root->fs_info->extent_root,
2062 path, bytenr, parent, root_objectid,
2063 owner, offset, refs_to_add);
2065 btrfs_abort_transaction(trans, root, ret);
2067 btrfs_free_path(path);
2071 static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
2072 struct btrfs_root *root,
2073 struct btrfs_delayed_ref_node *node,
2074 struct btrfs_delayed_extent_op *extent_op,
2075 int insert_reserved)
2078 struct btrfs_delayed_data_ref *ref;
2079 struct btrfs_key ins;
2084 ins.objectid = node->bytenr;
2085 ins.offset = node->num_bytes;
2086 ins.type = BTRFS_EXTENT_ITEM_KEY;
2088 ref = btrfs_delayed_node_to_data_ref(node);
2089 trace_run_delayed_data_ref(node, ref, node->action);
2091 if (node->type == BTRFS_SHARED_DATA_REF_KEY)
2092 parent = ref->parent;
2093 ref_root = ref->root;
2095 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2097 flags |= extent_op->flags_to_set;
2098 ret = alloc_reserved_file_extent(trans, root,
2099 parent, ref_root, flags,
2100 ref->objectid, ref->offset,
2101 &ins, node->ref_mod);
2102 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2103 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2104 node->num_bytes, parent,
2105 ref_root, ref->objectid,
2106 ref->offset, node->ref_mod,
2107 node->no_quota, extent_op);
2108 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2109 ret = __btrfs_free_extent(trans, root, node->bytenr,
2110 node->num_bytes, parent,
2111 ref_root, ref->objectid,
2112 ref->offset, node->ref_mod,
2113 extent_op, node->no_quota);
2120 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
2121 struct extent_buffer *leaf,
2122 struct btrfs_extent_item *ei)
2124 u64 flags = btrfs_extent_flags(leaf, ei);
2125 if (extent_op->update_flags) {
2126 flags |= extent_op->flags_to_set;
2127 btrfs_set_extent_flags(leaf, ei, flags);
2130 if (extent_op->update_key) {
2131 struct btrfs_tree_block_info *bi;
2132 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK));
2133 bi = (struct btrfs_tree_block_info *)(ei + 1);
2134 btrfs_set_tree_block_key(leaf, bi, &extent_op->key);
2138 static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
2139 struct btrfs_root *root,
2140 struct btrfs_delayed_ref_node *node,
2141 struct btrfs_delayed_extent_op *extent_op)
2143 struct btrfs_key key;
2144 struct btrfs_path *path;
2145 struct btrfs_extent_item *ei;
2146 struct extent_buffer *leaf;
2150 int metadata = !extent_op->is_data;
2155 if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2158 path = btrfs_alloc_path();
2162 key.objectid = node->bytenr;
2165 key.type = BTRFS_METADATA_ITEM_KEY;
2166 key.offset = extent_op->level;
2168 key.type = BTRFS_EXTENT_ITEM_KEY;
2169 key.offset = node->num_bytes;
2174 path->leave_spinning = 1;
2175 ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key,
2183 if (path->slots[0] > 0) {
2185 btrfs_item_key_to_cpu(path->nodes[0], &key,
2187 if (key.objectid == node->bytenr &&
2188 key.type == BTRFS_EXTENT_ITEM_KEY &&
2189 key.offset == node->num_bytes)
2193 btrfs_release_path(path);
2196 key.objectid = node->bytenr;
2197 key.offset = node->num_bytes;
2198 key.type = BTRFS_EXTENT_ITEM_KEY;
2207 leaf = path->nodes[0];
2208 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2209 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2210 if (item_size < sizeof(*ei)) {
2211 ret = convert_extent_item_v0(trans, root->fs_info->extent_root,
2217 leaf = path->nodes[0];
2218 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2221 BUG_ON(item_size < sizeof(*ei));
2222 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2223 __run_delayed_extent_op(extent_op, leaf, ei);
2225 btrfs_mark_buffer_dirty(leaf);
2227 btrfs_free_path(path);
2231 static int run_delayed_tree_ref(struct btrfs_trans_handle *trans,
2232 struct btrfs_root *root,
2233 struct btrfs_delayed_ref_node *node,
2234 struct btrfs_delayed_extent_op *extent_op,
2235 int insert_reserved)
2238 struct btrfs_delayed_tree_ref *ref;
2239 struct btrfs_key ins;
2242 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
2245 ref = btrfs_delayed_node_to_tree_ref(node);
2246 trace_run_delayed_tree_ref(node, ref, node->action);
2248 if (node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2249 parent = ref->parent;
2250 ref_root = ref->root;
2252 ins.objectid = node->bytenr;
2253 if (skinny_metadata) {
2254 ins.offset = ref->level;
2255 ins.type = BTRFS_METADATA_ITEM_KEY;
2257 ins.offset = node->num_bytes;
2258 ins.type = BTRFS_EXTENT_ITEM_KEY;
2261 BUG_ON(node->ref_mod != 1);
2262 if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) {
2263 BUG_ON(!extent_op || !extent_op->update_flags);
2264 ret = alloc_reserved_tree_block(trans, root,
2266 extent_op->flags_to_set,
2270 } else if (node->action == BTRFS_ADD_DELAYED_REF) {
2271 ret = __btrfs_inc_extent_ref(trans, root, node->bytenr,
2272 node->num_bytes, parent, ref_root,
2273 ref->level, 0, 1, node->no_quota,
2275 } else if (node->action == BTRFS_DROP_DELAYED_REF) {
2276 ret = __btrfs_free_extent(trans, root, node->bytenr,
2277 node->num_bytes, parent, ref_root,
2278 ref->level, 0, 1, extent_op,
2286 /* helper function to actually process a single delayed ref entry */
2287 static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
2288 struct btrfs_root *root,
2289 struct btrfs_delayed_ref_node *node,
2290 struct btrfs_delayed_extent_op *extent_op,
2291 int insert_reserved)
2295 if (trans->aborted) {
2296 if (insert_reserved)
2297 btrfs_pin_extent(root, node->bytenr,
2298 node->num_bytes, 1);
2302 if (btrfs_delayed_ref_is_head(node)) {
2303 struct btrfs_delayed_ref_head *head;
2305 * we've hit the end of the chain and we were supposed
2306 * to insert this extent into the tree. But, it got
2307 * deleted before we ever needed to insert it, so all
2308 * we have to do is clean up the accounting
2311 head = btrfs_delayed_node_to_head(node);
2312 trace_run_delayed_ref_head(node, head, node->action);
2314 if (insert_reserved) {
2315 btrfs_pin_extent(root, node->bytenr,
2316 node->num_bytes, 1);
2317 if (head->is_data) {
2318 ret = btrfs_del_csums(trans, root,
2326 if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
2327 node->type == BTRFS_SHARED_BLOCK_REF_KEY)
2328 ret = run_delayed_tree_ref(trans, root, node, extent_op,
2330 else if (node->type == BTRFS_EXTENT_DATA_REF_KEY ||
2331 node->type == BTRFS_SHARED_DATA_REF_KEY)
2332 ret = run_delayed_data_ref(trans, root, node, extent_op,
2339 static noinline struct btrfs_delayed_ref_node *
2340 select_delayed_ref(struct btrfs_delayed_ref_head *head)
2342 struct rb_node *node;
2343 struct btrfs_delayed_ref_node *ref, *last = NULL;;
2346 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2347 * this prevents ref count from going down to zero when
2348 * there still are pending delayed ref.
2350 node = rb_first(&head->ref_root);
2352 ref = rb_entry(node, struct btrfs_delayed_ref_node,
2354 if (ref->action == BTRFS_ADD_DELAYED_REF)
2356 else if (last == NULL)
2358 node = rb_next(node);
2364 * Returns 0 on success or if called with an already aborted transaction.
2365 * Returns -ENOMEM or -EIO on failure and will abort the transaction.
2367 static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2368 struct btrfs_root *root,
2371 struct btrfs_delayed_ref_root *delayed_refs;
2372 struct btrfs_delayed_ref_node *ref;
2373 struct btrfs_delayed_ref_head *locked_ref = NULL;
2374 struct btrfs_delayed_extent_op *extent_op;
2375 struct btrfs_fs_info *fs_info = root->fs_info;
2376 ktime_t start = ktime_get();
2378 unsigned long count = 0;
2379 unsigned long actual_count = 0;
2380 int must_insert_reserved = 0;
2382 delayed_refs = &trans->transaction->delayed_refs;
2388 spin_lock(&delayed_refs->lock);
2389 locked_ref = btrfs_select_ref_head(trans);
2391 spin_unlock(&delayed_refs->lock);
2395 /* grab the lock that says we are going to process
2396 * all the refs for this head */
2397 ret = btrfs_delayed_ref_lock(trans, locked_ref);
2398 spin_unlock(&delayed_refs->lock);
2400 * we may have dropped the spin lock to get the head
2401 * mutex lock, and that might have given someone else
2402 * time to free the head. If that's true, it has been
2403 * removed from our list and we can move on.
2405 if (ret == -EAGAIN) {
2413 * We need to try and merge add/drops of the same ref since we
2414 * can run into issues with relocate dropping the implicit ref
2415 * and then it being added back again before the drop can
2416 * finish. If we merged anything we need to re-loop so we can
2419 spin_lock(&locked_ref->lock);
2420 btrfs_merge_delayed_refs(trans, fs_info, delayed_refs,
2424 * locked_ref is the head node, so we have to go one
2425 * node back for any delayed ref updates
2427 ref = select_delayed_ref(locked_ref);
2429 if (ref && ref->seq &&
2430 btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) {
2431 spin_unlock(&locked_ref->lock);
2432 btrfs_delayed_ref_unlock(locked_ref);
2433 spin_lock(&delayed_refs->lock);
2434 locked_ref->processing = 0;
2435 delayed_refs->num_heads_ready++;
2436 spin_unlock(&delayed_refs->lock);
2444 * record the must insert reserved flag before we
2445 * drop the spin lock.
2447 must_insert_reserved = locked_ref->must_insert_reserved;
2448 locked_ref->must_insert_reserved = 0;
2450 extent_op = locked_ref->extent_op;
2451 locked_ref->extent_op = NULL;
2456 /* All delayed refs have been processed, Go ahead
2457 * and send the head node to run_one_delayed_ref,
2458 * so that any accounting fixes can happen
2460 ref = &locked_ref->node;
2462 if (extent_op && must_insert_reserved) {
2463 btrfs_free_delayed_extent_op(extent_op);
2468 spin_unlock(&locked_ref->lock);
2469 ret = run_delayed_extent_op(trans, root,
2471 btrfs_free_delayed_extent_op(extent_op);
2475 * Need to reset must_insert_reserved if
2476 * there was an error so the abort stuff
2477 * can cleanup the reserved space
2480 if (must_insert_reserved)
2481 locked_ref->must_insert_reserved = 1;
2482 locked_ref->processing = 0;
2483 btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret);
2484 btrfs_delayed_ref_unlock(locked_ref);
2491 * Need to drop our head ref lock and re-aqcuire the
2492 * delayed ref lock and then re-check to make sure
2495 spin_unlock(&locked_ref->lock);
2496 spin_lock(&delayed_refs->lock);
2497 spin_lock(&locked_ref->lock);
2498 if (rb_first(&locked_ref->ref_root) ||
2499 locked_ref->extent_op) {
2500 spin_unlock(&locked_ref->lock);
2501 spin_unlock(&delayed_refs->lock);
2505 delayed_refs->num_heads--;
2506 rb_erase(&locked_ref->href_node,
2507 &delayed_refs->href_root);
2508 spin_unlock(&delayed_refs->lock);
2512 rb_erase(&ref->rb_node, &locked_ref->ref_root);
2514 atomic_dec(&delayed_refs->num_entries);
2516 if (!btrfs_delayed_ref_is_head(ref)) {
2518 * when we play the delayed ref, also correct the
2521 switch (ref->action) {
2522 case BTRFS_ADD_DELAYED_REF:
2523 case BTRFS_ADD_DELAYED_EXTENT:
2524 locked_ref->node.ref_mod -= ref->ref_mod;
2526 case BTRFS_DROP_DELAYED_REF:
2527 locked_ref->node.ref_mod += ref->ref_mod;
2533 spin_unlock(&locked_ref->lock);
2535 ret = run_one_delayed_ref(trans, root, ref, extent_op,
2536 must_insert_reserved);
2538 btrfs_free_delayed_extent_op(extent_op);
2540 locked_ref->processing = 0;
2541 btrfs_delayed_ref_unlock(locked_ref);
2542 btrfs_put_delayed_ref(ref);
2543 btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret);
2548 * If this node is a head, that means all the refs in this head
2549 * have been dealt with, and we will pick the next head to deal
2550 * with, so we must unlock the head and drop it from the cluster
2551 * list before we release it.
2553 if (btrfs_delayed_ref_is_head(ref)) {
2554 btrfs_delayed_ref_unlock(locked_ref);
2557 btrfs_put_delayed_ref(ref);
2563 * We don't want to include ref heads since we can have empty ref heads
2564 * and those will drastically skew our runtime down since we just do
2565 * accounting, no actual extent tree updates.
2567 if (actual_count > 0) {
2568 u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start));
2572 * We weigh the current average higher than our current runtime
2573 * to avoid large swings in the average.
2575 spin_lock(&delayed_refs->lock);
2576 avg = fs_info->avg_delayed_ref_runtime * 3 + runtime;
2577 avg = div64_u64(avg, 4);
2578 fs_info->avg_delayed_ref_runtime = avg;
2579 spin_unlock(&delayed_refs->lock);
2584 #ifdef SCRAMBLE_DELAYED_REFS
2586 * Normally delayed refs get processed in ascending bytenr order. This
2587 * correlates in most cases to the order added. To expose dependencies on this
2588 * order, we start to process the tree in the middle instead of the beginning
2590 static u64 find_middle(struct rb_root *root)
2592 struct rb_node *n = root->rb_node;
2593 struct btrfs_delayed_ref_node *entry;
2596 u64 first = 0, last = 0;
2600 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2601 first = entry->bytenr;
2605 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2606 last = entry->bytenr;
2611 entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
2612 WARN_ON(!entry->in_tree);
2614 middle = entry->bytenr;
2627 static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads)
2631 num_bytes = heads * (sizeof(struct btrfs_extent_item) +
2632 sizeof(struct btrfs_extent_inline_ref));
2633 if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA))
2634 num_bytes += heads * sizeof(struct btrfs_tree_block_info);
2637 * We don't ever fill up leaves all the way so multiply by 2 just to be
2638 * closer to what we're really going to want to ouse.
2640 return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root));
2643 int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans,
2644 struct btrfs_root *root)
2646 struct btrfs_block_rsv *global_rsv;
2647 u64 num_heads = trans->transaction->delayed_refs.num_heads_ready;
2651 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
2652 num_heads = heads_to_leaves(root, num_heads);
2654 num_bytes += (num_heads - 1) * root->nodesize;
2656 global_rsv = &root->fs_info->global_block_rsv;
2659 * If we can't allocate any more chunks lets make sure we have _lots_ of
2660 * wiggle room since running delayed refs can create more delayed refs.
2662 if (global_rsv->space_info->full)
2665 spin_lock(&global_rsv->lock);
2666 if (global_rsv->reserved <= num_bytes)
2668 spin_unlock(&global_rsv->lock);
2672 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans,
2673 struct btrfs_root *root)
2675 struct btrfs_fs_info *fs_info = root->fs_info;
2677 atomic_read(&trans->transaction->delayed_refs.num_entries);
2682 avg_runtime = fs_info->avg_delayed_ref_runtime;
2683 val = num_entries * avg_runtime;
2684 if (num_entries * avg_runtime >= NSEC_PER_SEC)
2686 if (val >= NSEC_PER_SEC / 2)
2689 return btrfs_check_space_for_delayed_refs(trans, root);
2692 struct async_delayed_refs {
2693 struct btrfs_root *root;
2697 struct completion wait;
2698 struct btrfs_work work;
2701 static void delayed_ref_async_start(struct btrfs_work *work)
2703 struct async_delayed_refs *async;
2704 struct btrfs_trans_handle *trans;
2707 async = container_of(work, struct async_delayed_refs, work);
2709 trans = btrfs_join_transaction(async->root);
2710 if (IS_ERR(trans)) {
2711 async->error = PTR_ERR(trans);
2716 * trans->sync means that when we call end_transaciton, we won't
2717 * wait on delayed refs
2720 ret = btrfs_run_delayed_refs(trans, async->root, async->count);
2724 ret = btrfs_end_transaction(trans, async->root);
2725 if (ret && !async->error)
2729 complete(&async->wait);
2734 int btrfs_async_run_delayed_refs(struct btrfs_root *root,
2735 unsigned long count, int wait)
2737 struct async_delayed_refs *async;
2740 async = kmalloc(sizeof(*async), GFP_NOFS);
2744 async->root = root->fs_info->tree_root;
2745 async->count = count;
2751 init_completion(&async->wait);
2753 btrfs_init_work(&async->work, btrfs_extent_refs_helper,
2754 delayed_ref_async_start, NULL, NULL);
2756 btrfs_queue_work(root->fs_info->extent_workers, &async->work);
2759 wait_for_completion(&async->wait);
2768 * this starts processing the delayed reference count updates and
2769 * extent insertions we have queued up so far. count can be
2770 * 0, which means to process everything in the tree at the start
2771 * of the run (but not newly added entries), or it can be some target
2772 * number you'd like to process.
2774 * Returns 0 on success or if called with an aborted transaction
2775 * Returns <0 on error and aborts the transaction
2777 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
2778 struct btrfs_root *root, unsigned long count)
2780 struct rb_node *node;
2781 struct btrfs_delayed_ref_root *delayed_refs;
2782 struct btrfs_delayed_ref_head *head;
2784 int run_all = count == (unsigned long)-1;
2787 /* We'll clean this up in btrfs_cleanup_transaction */
2791 if (root == root->fs_info->extent_root)
2792 root = root->fs_info->tree_root;
2794 delayed_refs = &trans->transaction->delayed_refs;
2796 count = atomic_read(&delayed_refs->num_entries) * 2;
2801 #ifdef SCRAMBLE_DELAYED_REFS
2802 delayed_refs->run_delayed_start = find_middle(&delayed_refs->root);
2804 ret = __btrfs_run_delayed_refs(trans, root, count);
2806 btrfs_abort_transaction(trans, root, ret);
2811 if (!list_empty(&trans->new_bgs))
2812 btrfs_create_pending_block_groups(trans, root);
2814 spin_lock(&delayed_refs->lock);
2815 node = rb_first(&delayed_refs->href_root);
2817 spin_unlock(&delayed_refs->lock);
2820 count = (unsigned long)-1;
2823 head = rb_entry(node, struct btrfs_delayed_ref_head,
2825 if (btrfs_delayed_ref_is_head(&head->node)) {
2826 struct btrfs_delayed_ref_node *ref;
2829 atomic_inc(&ref->refs);
2831 spin_unlock(&delayed_refs->lock);
2833 * Mutex was contended, block until it's
2834 * released and try again
2836 mutex_lock(&head->mutex);
2837 mutex_unlock(&head->mutex);
2839 btrfs_put_delayed_ref(ref);
2845 node = rb_next(node);
2847 spin_unlock(&delayed_refs->lock);
2852 ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info);
2855 assert_qgroups_uptodate(trans);
2859 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
2860 struct btrfs_root *root,
2861 u64 bytenr, u64 num_bytes, u64 flags,
2862 int level, int is_data)
2864 struct btrfs_delayed_extent_op *extent_op;
2867 extent_op = btrfs_alloc_delayed_extent_op();
2871 extent_op->flags_to_set = flags;
2872 extent_op->update_flags = 1;
2873 extent_op->update_key = 0;
2874 extent_op->is_data = is_data ? 1 : 0;
2875 extent_op->level = level;
2877 ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
2878 num_bytes, extent_op);
2880 btrfs_free_delayed_extent_op(extent_op);
2884 static noinline int check_delayed_ref(struct btrfs_trans_handle *trans,
2885 struct btrfs_root *root,
2886 struct btrfs_path *path,
2887 u64 objectid, u64 offset, u64 bytenr)
2889 struct btrfs_delayed_ref_head *head;
2890 struct btrfs_delayed_ref_node *ref;
2891 struct btrfs_delayed_data_ref *data_ref;
2892 struct btrfs_delayed_ref_root *delayed_refs;
2893 struct rb_node *node;
2896 delayed_refs = &trans->transaction->delayed_refs;
2897 spin_lock(&delayed_refs->lock);
2898 head = btrfs_find_delayed_ref_head(trans, bytenr);
2900 spin_unlock(&delayed_refs->lock);
2904 if (!mutex_trylock(&head->mutex)) {
2905 atomic_inc(&head->node.refs);
2906 spin_unlock(&delayed_refs->lock);
2908 btrfs_release_path(path);
2911 * Mutex was contended, block until it's released and let
2914 mutex_lock(&head->mutex);
2915 mutex_unlock(&head->mutex);
2916 btrfs_put_delayed_ref(&head->node);
2919 spin_unlock(&delayed_refs->lock);
2921 spin_lock(&head->lock);
2922 node = rb_first(&head->ref_root);
2924 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2925 node = rb_next(node);
2927 /* If it's a shared ref we know a cross reference exists */
2928 if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) {
2933 data_ref = btrfs_delayed_node_to_data_ref(ref);
2936 * If our ref doesn't match the one we're currently looking at
2937 * then we have a cross reference.
2939 if (data_ref->root != root->root_key.objectid ||
2940 data_ref->objectid != objectid ||
2941 data_ref->offset != offset) {
2946 spin_unlock(&head->lock);
2947 mutex_unlock(&head->mutex);
2951 static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
2952 struct btrfs_root *root,
2953 struct btrfs_path *path,
2954 u64 objectid, u64 offset, u64 bytenr)
2956 struct btrfs_root *extent_root = root->fs_info->extent_root;
2957 struct extent_buffer *leaf;
2958 struct btrfs_extent_data_ref *ref;
2959 struct btrfs_extent_inline_ref *iref;
2960 struct btrfs_extent_item *ei;
2961 struct btrfs_key key;
2965 key.objectid = bytenr;
2966 key.offset = (u64)-1;
2967 key.type = BTRFS_EXTENT_ITEM_KEY;
2969 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2972 BUG_ON(ret == 0); /* Corruption */
2975 if (path->slots[0] == 0)
2979 leaf = path->nodes[0];
2980 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2982 if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY)
2986 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2987 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2988 if (item_size < sizeof(*ei)) {
2989 WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
2993 ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
2995 if (item_size != sizeof(*ei) +
2996 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY))
2999 if (btrfs_extent_generation(leaf, ei) <=
3000 btrfs_root_last_snapshot(&root->root_item))
3003 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
3004 if (btrfs_extent_inline_ref_type(leaf, iref) !=
3005 BTRFS_EXTENT_DATA_REF_KEY)
3008 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
3009 if (btrfs_extent_refs(leaf, ei) !=
3010 btrfs_extent_data_ref_count(leaf, ref) ||
3011 btrfs_extent_data_ref_root(leaf, ref) !=
3012 root->root_key.objectid ||
3013 btrfs_extent_data_ref_objectid(leaf, ref) != objectid ||
3014 btrfs_extent_data_ref_offset(leaf, ref) != offset)
3022 int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans,
3023 struct btrfs_root *root,
3024 u64 objectid, u64 offset, u64 bytenr)
3026 struct btrfs_path *path;
3030 path = btrfs_alloc_path();
3035 ret = check_committed_ref(trans, root, path, objectid,
3037 if (ret && ret != -ENOENT)
3040 ret2 = check_delayed_ref(trans, root, path, objectid,
3042 } while (ret2 == -EAGAIN);
3044 if (ret2 && ret2 != -ENOENT) {
3049 if (ret != -ENOENT || ret2 != -ENOENT)
3052 btrfs_free_path(path);
3053 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
3058 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
3059 struct btrfs_root *root,
3060 struct extent_buffer *buf,
3061 int full_backref, int inc)
3068 struct btrfs_key key;
3069 struct btrfs_file_extent_item *fi;
3073 int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
3074 u64, u64, u64, u64, u64, u64, int);
3076 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3077 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
3080 ref_root = btrfs_header_owner(buf);
3081 nritems = btrfs_header_nritems(buf);
3082 level = btrfs_header_level(buf);
3084 if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0)
3088 process_func = btrfs_inc_extent_ref;
3090 process_func = btrfs_free_extent;
3093 parent = buf->start;
3097 for (i = 0; i < nritems; i++) {
3099 btrfs_item_key_to_cpu(buf, &key, i);
3100 if (key.type != BTRFS_EXTENT_DATA_KEY)
3102 fi = btrfs_item_ptr(buf, i,
3103 struct btrfs_file_extent_item);
3104 if (btrfs_file_extent_type(buf, fi) ==
3105 BTRFS_FILE_EXTENT_INLINE)
3107 bytenr = btrfs_file_extent_disk_bytenr(buf, fi);
3111 num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi);
3112 key.offset -= btrfs_file_extent_offset(buf, fi);
3113 ret = process_func(trans, root, bytenr, num_bytes,
3114 parent, ref_root, key.objectid,
3119 bytenr = btrfs_node_blockptr(buf, i);
3120 num_bytes = root->nodesize;
3121 ret = process_func(trans, root, bytenr, num_bytes,
3122 parent, ref_root, level - 1, 0,
3133 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3134 struct extent_buffer *buf, int full_backref)
3136 return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
3139 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3140 struct extent_buffer *buf, int full_backref)
3142 return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
3145 static int write_one_cache_group(struct btrfs_trans_handle *trans,
3146 struct btrfs_root *root,
3147 struct btrfs_path *path,
3148 struct btrfs_block_group_cache *cache)
3151 struct btrfs_root *extent_root = root->fs_info->extent_root;
3153 struct extent_buffer *leaf;
3155 ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
3158 BUG_ON(ret); /* Corruption */
3160 leaf = path->nodes[0];
3161 bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
3162 write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item));
3163 btrfs_mark_buffer_dirty(leaf);
3164 btrfs_release_path(path);
3167 btrfs_abort_transaction(trans, root, ret);
3174 static struct btrfs_block_group_cache *
3175 next_block_group(struct btrfs_root *root,
3176 struct btrfs_block_group_cache *cache)
3178 struct rb_node *node;
3179 spin_lock(&root->fs_info->block_group_cache_lock);
3180 node = rb_next(&cache->cache_node);
3181 btrfs_put_block_group(cache);
3183 cache = rb_entry(node, struct btrfs_block_group_cache,
3185 btrfs_get_block_group(cache);
3188 spin_unlock(&root->fs_info->block_group_cache_lock);
3192 static int cache_save_setup(struct btrfs_block_group_cache *block_group,
3193 struct btrfs_trans_handle *trans,
3194 struct btrfs_path *path)
3196 struct btrfs_root *root = block_group->fs_info->tree_root;
3197 struct inode *inode = NULL;
3199 int dcs = BTRFS_DC_ERROR;
3205 * If this block group is smaller than 100 megs don't bother caching the
3208 if (block_group->key.offset < (100 * 1024 * 1024)) {
3209 spin_lock(&block_group->lock);
3210 block_group->disk_cache_state = BTRFS_DC_WRITTEN;
3211 spin_unlock(&block_group->lock);
3216 inode = lookup_free_space_inode(root, block_group, path);
3217 if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
3218 ret = PTR_ERR(inode);
3219 btrfs_release_path(path);
3223 if (IS_ERR(inode)) {
3227 if (block_group->ro)
3230 ret = create_free_space_inode(root, trans, block_group, path);
3236 /* We've already setup this transaction, go ahead and exit */
3237 if (block_group->cache_generation == trans->transid &&
3238 i_size_read(inode)) {
3239 dcs = BTRFS_DC_SETUP;
3244 * We want to set the generation to 0, that way if anything goes wrong
3245 * from here on out we know not to trust this cache when we load up next
3248 BTRFS_I(inode)->generation = 0;
3249 ret = btrfs_update_inode(trans, root, inode);
3252 if (i_size_read(inode) > 0) {
3253 ret = btrfs_check_trunc_cache_free_space(root,
3254 &root->fs_info->global_block_rsv);
3258 ret = btrfs_truncate_free_space_cache(root, trans, inode);
3263 spin_lock(&block_group->lock);
3264 if (block_group->cached != BTRFS_CACHE_FINISHED ||
3265 !btrfs_test_opt(root, SPACE_CACHE) ||
3266 block_group->delalloc_bytes) {
3268 * don't bother trying to write stuff out _if_
3269 * a) we're not cached,
3270 * b) we're with nospace_cache mount option.
3272 dcs = BTRFS_DC_WRITTEN;
3273 spin_unlock(&block_group->lock);
3276 spin_unlock(&block_group->lock);
3279 * Try to preallocate enough space based on how big the block group is.
3280 * Keep in mind this has to include any pinned space which could end up
3281 * taking up quite a bit since it's not folded into the other space
3284 num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024);
3289 num_pages *= PAGE_CACHE_SIZE;
3291 ret = btrfs_check_data_free_space(inode, num_pages);
3295 ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
3296 num_pages, num_pages,
3299 dcs = BTRFS_DC_SETUP;
3300 btrfs_free_reserved_data_space(inode, num_pages);
3305 btrfs_release_path(path);
3307 spin_lock(&block_group->lock);
3308 if (!ret && dcs == BTRFS_DC_SETUP)
3309 block_group->cache_generation = trans->transid;
3310 block_group->disk_cache_state = dcs;
3311 spin_unlock(&block_group->lock);
3316 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
3317 struct btrfs_root *root)
3319 struct btrfs_block_group_cache *cache;
3321 struct btrfs_path *path;
3324 path = btrfs_alloc_path();
3330 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3332 if (cache->disk_cache_state == BTRFS_DC_CLEAR)
3334 cache = next_block_group(root, cache);
3342 err = cache_save_setup(cache, trans, path);
3343 last = cache->key.objectid + cache->key.offset;
3344 btrfs_put_block_group(cache);
3349 err = btrfs_run_delayed_refs(trans, root,
3351 if (err) /* File system offline */
3355 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3357 if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
3358 btrfs_put_block_group(cache);
3364 cache = next_block_group(root, cache);
3373 if (cache->disk_cache_state == BTRFS_DC_SETUP)
3374 cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
3376 last = cache->key.objectid + cache->key.offset;
3378 err = write_one_cache_group(trans, root, path, cache);
3379 btrfs_put_block_group(cache);
3380 if (err) /* File system offline */
3386 * I don't think this is needed since we're just marking our
3387 * preallocated extent as written, but just in case it can't
3391 err = btrfs_run_delayed_refs(trans, root,
3393 if (err) /* File system offline */
3397 cache = btrfs_lookup_first_block_group(root->fs_info, last);
3400 * Really this shouldn't happen, but it could if we
3401 * couldn't write the entire preallocated extent and
3402 * splitting the extent resulted in a new block.
3405 btrfs_put_block_group(cache);
3408 if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3410 cache = next_block_group(root, cache);
3419 err = btrfs_write_out_cache(root, trans, cache, path);
3422 * If we didn't have an error then the cache state is still
3423 * NEED_WRITE, so we can set it to WRITTEN.
3425 if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
3426 cache->disk_cache_state = BTRFS_DC_WRITTEN;
3427 last = cache->key.objectid + cache->key.offset;
3428 btrfs_put_block_group(cache);
3432 btrfs_free_path(path);
3436 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
3438 struct btrfs_block_group_cache *block_group;
3441 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
3442 if (!block_group || block_group->ro)
3445 btrfs_put_block_group(block_group);
3449 static const char *alloc_name(u64 flags)
3452 case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA:
3454 case BTRFS_BLOCK_GROUP_METADATA:
3456 case BTRFS_BLOCK_GROUP_DATA:
3458 case BTRFS_BLOCK_GROUP_SYSTEM:
3462 return "invalid-combination";
3466 static int update_space_info(struct btrfs_fs_info *info, u64 flags,
3467 u64 total_bytes, u64 bytes_used,
3468 struct btrfs_space_info **space_info)
3470 struct btrfs_space_info *found;
3475 if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3476 BTRFS_BLOCK_GROUP_RAID10))
3481 found = __find_space_info(info, flags);
3483 spin_lock(&found->lock);
3484 found->total_bytes += total_bytes;
3485 found->disk_total += total_bytes * factor;
3486 found->bytes_used += bytes_used;
3487 found->disk_used += bytes_used * factor;
3489 spin_unlock(&found->lock);
3490 *space_info = found;
3493 found = kzalloc(sizeof(*found), GFP_NOFS);
3497 ret = percpu_counter_init(&found->total_bytes_pinned, 0);
3503 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
3504 INIT_LIST_HEAD(&found->block_groups[i]);
3505 init_rwsem(&found->groups_sem);
3506 spin_lock_init(&found->lock);
3507 found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK;
3508 found->total_bytes = total_bytes;
3509 found->disk_total = total_bytes * factor;
3510 found->bytes_used = bytes_used;
3511 found->disk_used = bytes_used * factor;
3512 found->bytes_pinned = 0;
3513 found->bytes_reserved = 0;
3514 found->bytes_readonly = 0;
3515 found->bytes_may_use = 0;
3517 found->force_alloc = CHUNK_ALLOC_NO_FORCE;
3518 found->chunk_alloc = 0;
3520 init_waitqueue_head(&found->wait);
3522 ret = kobject_init_and_add(&found->kobj, &space_info_ktype,
3523 info->space_info_kobj, "%s",
3524 alloc_name(found->flags));
3530 *space_info = found;
3531 list_add_rcu(&found->list, &info->space_info);
3532 if (flags & BTRFS_BLOCK_GROUP_DATA)
3533 info->data_sinfo = found;
3538 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
3540 u64 extra_flags = chunk_to_extended(flags) &
3541 BTRFS_EXTENDED_PROFILE_MASK;
3543 write_seqlock(&fs_info->profiles_lock);
3544 if (flags & BTRFS_BLOCK_GROUP_DATA)
3545 fs_info->avail_data_alloc_bits |= extra_flags;
3546 if (flags & BTRFS_BLOCK_GROUP_METADATA)
3547 fs_info->avail_metadata_alloc_bits |= extra_flags;
3548 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3549 fs_info->avail_system_alloc_bits |= extra_flags;
3550 write_sequnlock(&fs_info->profiles_lock);
3554 * returns target flags in extended format or 0 if restripe for this
3555 * chunk_type is not in progress
3557 * should be called with either volume_mutex or balance_lock held
3559 static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
3561 struct btrfs_balance_control *bctl = fs_info->balance_ctl;
3567 if (flags & BTRFS_BLOCK_GROUP_DATA &&
3568 bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3569 target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
3570 } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
3571 bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3572 target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
3573 } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
3574 bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3575 target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
3582 * @flags: available profiles in extended format (see ctree.h)
3584 * Returns reduced profile in chunk format. If profile changing is in
3585 * progress (either running or paused) picks the target profile (if it's
3586 * already available), otherwise falls back to plain reducing.
3588 static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
3590 u64 num_devices = root->fs_info->fs_devices->rw_devices;
3595 * see if restripe for this chunk_type is in progress, if so
3596 * try to reduce to the target profile
3598 spin_lock(&root->fs_info->balance_lock);
3599 target = get_restripe_target(root->fs_info, flags);
3601 /* pick target profile only if it's already available */
3602 if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
3603 spin_unlock(&root->fs_info->balance_lock);
3604 return extended_to_chunk(target);
3607 spin_unlock(&root->fs_info->balance_lock);
3609 /* First, mask out the RAID levels which aren't possible */
3610 if (num_devices == 1)
3611 flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 |
3612 BTRFS_BLOCK_GROUP_RAID5);
3613 if (num_devices < 3)
3614 flags &= ~BTRFS_BLOCK_GROUP_RAID6;
3615 if (num_devices < 4)
3616 flags &= ~BTRFS_BLOCK_GROUP_RAID10;
3618 tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3619 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 |
3620 BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10);
3623 if (tmp & BTRFS_BLOCK_GROUP_RAID6)
3624 tmp = BTRFS_BLOCK_GROUP_RAID6;
3625 else if (tmp & BTRFS_BLOCK_GROUP_RAID5)
3626 tmp = BTRFS_BLOCK_GROUP_RAID5;
3627 else if (tmp & BTRFS_BLOCK_GROUP_RAID10)
3628 tmp = BTRFS_BLOCK_GROUP_RAID10;
3629 else if (tmp & BTRFS_BLOCK_GROUP_RAID1)
3630 tmp = BTRFS_BLOCK_GROUP_RAID1;
3631 else if (tmp & BTRFS_BLOCK_GROUP_RAID0)
3632 tmp = BTRFS_BLOCK_GROUP_RAID0;
3634 return extended_to_chunk(flags | tmp);
3637 static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags)
3644 seq = read_seqbegin(&root->fs_info->profiles_lock);
3646 if (flags & BTRFS_BLOCK_GROUP_DATA)
3647 flags |= root->fs_info->avail_data_alloc_bits;
3648 else if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
3649 flags |= root->fs_info->avail_system_alloc_bits;
3650 else if (flags & BTRFS_BLOCK_GROUP_METADATA)
3651 flags |= root->fs_info->avail_metadata_alloc_bits;
3652 } while (read_seqretry(&root->fs_info->profiles_lock, seq));
3654 return btrfs_reduce_alloc_profile(root, flags);
3657 u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data)
3663 flags = BTRFS_BLOCK_GROUP_DATA;
3664 else if (root == root->fs_info->chunk_root)
3665 flags = BTRFS_BLOCK_GROUP_SYSTEM;
3667 flags = BTRFS_BLOCK_GROUP_METADATA;
3669 ret = get_alloc_profile(root, flags);
3674 * This will check the space that the inode allocates from to make sure we have
3675 * enough space for bytes.
3677 int btrfs_check_data_free_space(struct inode *inode, u64 bytes)
3679 struct btrfs_space_info *data_sinfo;
3680 struct btrfs_root *root = BTRFS_I(inode)->root;
3681 struct btrfs_fs_info *fs_info = root->fs_info;
3683 int ret = 0, committed = 0, alloc_chunk = 1;
3685 /* make sure bytes are sectorsize aligned */
3686 bytes = ALIGN(bytes, root->sectorsize);
3688 if (btrfs_is_free_space_inode(inode)) {
3690 ASSERT(current->journal_info);
3693 data_sinfo = fs_info->data_sinfo;
3698 /* make sure we have enough space to handle the data first */
3699 spin_lock(&data_sinfo->lock);
3700 used = data_sinfo->bytes_used + data_sinfo->bytes_reserved +
3701 data_sinfo->bytes_pinned + data_sinfo->bytes_readonly +
3702 data_sinfo->bytes_may_use;
3704 if (used + bytes > data_sinfo->total_bytes) {
3705 struct btrfs_trans_handle *trans;
3708 * if we don't have enough free bytes in this space then we need
3709 * to alloc a new chunk.
3711 if (!data_sinfo->full && alloc_chunk) {
3714 data_sinfo->force_alloc = CHUNK_ALLOC_FORCE;
3715 spin_unlock(&data_sinfo->lock);
3717 alloc_target = btrfs_get_alloc_profile(root, 1);
3719 * It is ugly that we don't call nolock join
3720 * transaction for the free space inode case here.
3721 * But it is safe because we only do the data space
3722 * reservation for the free space cache in the
3723 * transaction context, the common join transaction
3724 * just increase the counter of the current transaction
3725 * handler, doesn't try to acquire the trans_lock of
3728 trans = btrfs_join_transaction(root);
3730 return PTR_ERR(trans);
3732 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
3734 CHUNK_ALLOC_NO_FORCE);
3735 btrfs_end_transaction(trans, root);
3744 data_sinfo = fs_info->data_sinfo;
3750 * If we don't have enough pinned space to deal with this
3751 * allocation don't bother committing the transaction.
3753 if (percpu_counter_compare(&data_sinfo->total_bytes_pinned,
3756 spin_unlock(&data_sinfo->lock);
3758 /* commit the current transaction and try again */
3761 !atomic_read(&root->fs_info->open_ioctl_trans)) {
3764 trans = btrfs_join_transaction(root);
3766 return PTR_ERR(trans);
3767 ret = btrfs_commit_transaction(trans, root);
3773 trace_btrfs_space_reservation(root->fs_info,
3774 "space_info:enospc",
3775 data_sinfo->flags, bytes, 1);
3778 data_sinfo->bytes_may_use += bytes;
3779 trace_btrfs_space_reservation(root->fs_info, "space_info",
3780 data_sinfo->flags, bytes, 1);
3781 spin_unlock(&data_sinfo->lock);
3787 * Called if we need to clear a data reservation for this inode.
3789 void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
3791 struct btrfs_root *root = BTRFS_I(inode)->root;
3792 struct btrfs_space_info *data_sinfo;
3794 /* make sure bytes are sectorsize aligned */
3795 bytes = ALIGN(bytes, root->sectorsize);
3797 data_sinfo = root->fs_info->data_sinfo;
3798 spin_lock(&data_sinfo->lock);
3799 WARN_ON(data_sinfo->bytes_may_use < bytes);
3800 data_sinfo->bytes_may_use -= bytes;
3801 trace_btrfs_space_reservation(root->fs_info, "space_info",
3802 data_sinfo->flags, bytes, 0);
3803 spin_unlock(&data_sinfo->lock);
3806 static void force_metadata_allocation(struct btrfs_fs_info *info)
3808 struct list_head *head = &info->space_info;
3809 struct btrfs_space_info *found;
3812 list_for_each_entry_rcu(found, head, list) {
3813 if (found->flags & BTRFS_BLOCK_GROUP_METADATA)
3814 found->force_alloc = CHUNK_ALLOC_FORCE;
3819 static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global)
3821 return (global->size << 1);
3824 static int should_alloc_chunk(struct btrfs_root *root,
3825 struct btrfs_space_info *sinfo, int force)
3827 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
3828 u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly;
3829 u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved;
3832 if (force == CHUNK_ALLOC_FORCE)
3836 * We need to take into account the global rsv because for all intents
3837 * and purposes it's used space. Don't worry about locking the
3838 * global_rsv, it doesn't change except when the transaction commits.
3840 if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA)
3841 num_allocated += calc_global_rsv_need_space(global_rsv);
3844 * in limited mode, we want to have some free space up to
3845 * about 1% of the FS size.
3847 if (force == CHUNK_ALLOC_LIMITED) {
3848 thresh = btrfs_super_total_bytes(root->fs_info->super_copy);
3849 thresh = max_t(u64, 64 * 1024 * 1024,
3850 div_factor_fine(thresh, 1));
3852 if (num_bytes - num_allocated < thresh)
3856 if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8))
3861 static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
3865 if (type & (BTRFS_BLOCK_GROUP_RAID10 |
3866 BTRFS_BLOCK_GROUP_RAID0 |
3867 BTRFS_BLOCK_GROUP_RAID5 |
3868 BTRFS_BLOCK_GROUP_RAID6))
3869 num_dev = root->fs_info->fs_devices->rw_devices;
3870 else if (type & BTRFS_BLOCK_GROUP_RAID1)
3873 num_dev = 1; /* DUP or single */
3875 /* metadata for updaing devices and chunk tree */
3876 return btrfs_calc_trans_metadata_size(root, num_dev + 1);
3879 static void check_system_chunk(struct btrfs_trans_handle *trans,
3880 struct btrfs_root *root, u64 type)
3882 struct btrfs_space_info *info;
3886 info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
3887 spin_lock(&info->lock);
3888 left = info->total_bytes - info->bytes_used - info->bytes_pinned -
3889 info->bytes_reserved - info->bytes_readonly;
3890 spin_unlock(&info->lock);
3892 thresh = get_system_chunk_thresh(root, type);
3893 if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
3894 btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu",
3895 left, thresh, type);
3896 dump_space_info(info, 0, 0);
3899 if (left < thresh) {
3902 flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
3903 btrfs_alloc_chunk(trans, root, flags);
3907 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
3908 struct btrfs_root *extent_root, u64 flags, int force)
3910 struct btrfs_space_info *space_info;
3911 struct btrfs_fs_info *fs_info = extent_root->fs_info;
3912 int wait_for_alloc = 0;
3915 /* Don't re-enter if we're already allocating a chunk */
3916 if (trans->allocating_chunk)
3919 space_info = __find_space_info(extent_root->fs_info, flags);
3921 ret = update_space_info(extent_root->fs_info, flags,
3923 BUG_ON(ret); /* -ENOMEM */
3925 BUG_ON(!space_info); /* Logic error */
3928 spin_lock(&space_info->lock);
3929 if (force < space_info->force_alloc)
3930 force = space_info->force_alloc;
3931 if (space_info->full) {
3932 if (should_alloc_chunk(extent_root, space_info, force))
3936 spin_unlock(&space_info->lock);
3940 if (!should_alloc_chunk(extent_root, space_info, force)) {
3941 spin_unlock(&space_info->lock);
3943 } else if (space_info->chunk_alloc) {
3946 space_info->chunk_alloc = 1;
3949 spin_unlock(&space_info->lock);
3951 mutex_lock(&fs_info->chunk_mutex);
3954 * The chunk_mutex is held throughout the entirety of a chunk
3955 * allocation, so once we've acquired the chunk_mutex we know that the
3956 * other guy is done and we need to recheck and see if we should
3959 if (wait_for_alloc) {
3960 mutex_unlock(&fs_info->chunk_mutex);
3965 trans->allocating_chunk = true;
3968 * If we have mixed data/metadata chunks we want to make sure we keep
3969 * allocating mixed chunks instead of individual chunks.
3971 if (btrfs_mixed_space_info(space_info))
3972 flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA);
3975 * if we're doing a data chunk, go ahead and make sure that
3976 * we keep a reasonable number of metadata chunks allocated in the
3979 if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) {
3980 fs_info->data_chunk_allocations++;
3981 if (!(fs_info->data_chunk_allocations %
3982 fs_info->metadata_ratio))
3983 force_metadata_allocation(fs_info);
3987 * Check if we have enough space in SYSTEM chunk because we may need
3988 * to update devices.
3990 check_system_chunk(trans, extent_root, flags);
3992 ret = btrfs_alloc_chunk(trans, extent_root, flags);
3993 trans->allocating_chunk = false;
3995 spin_lock(&space_info->lock);
3996 if (ret < 0 && ret != -ENOSPC)
3999 space_info->full = 1;
4003 space_info->force_alloc = CHUNK_ALLOC_NO_FORCE;
4005 space_info->chunk_alloc = 0;
4006 spin_unlock(&space_info->lock);
4007 mutex_unlock(&fs_info->chunk_mutex);
4011 static int can_overcommit(struct btrfs_root *root,
4012 struct btrfs_space_info *space_info, u64 bytes,
4013 enum btrfs_reserve_flush_enum flush)
4015 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4016 u64 profile = btrfs_get_alloc_profile(root, 0);
4021 used = space_info->bytes_used + space_info->bytes_reserved +
4022 space_info->bytes_pinned + space_info->bytes_readonly;
4025 * We only want to allow over committing if we have lots of actual space
4026 * free, but if we don't have enough space to handle the global reserve
4027 * space then we could end up having a real enospc problem when trying
4028 * to allocate a chunk or some other such important allocation.
4030 spin_lock(&global_rsv->lock);
4031 space_size = calc_global_rsv_need_space(global_rsv);
4032 spin_unlock(&global_rsv->lock);
4033 if (used + space_size >= space_info->total_bytes)
4036 used += space_info->bytes_may_use;
4038 spin_lock(&root->fs_info->free_chunk_lock);
4039 avail = root->fs_info->free_chunk_space;
4040 spin_unlock(&root->fs_info->free_chunk_lock);
4043 * If we have dup, raid1 or raid10 then only half of the free
4044 * space is actually useable. For raid56, the space info used
4045 * doesn't include the parity drive, so we don't have to
4048 if (profile & (BTRFS_BLOCK_GROUP_DUP |
4049 BTRFS_BLOCK_GROUP_RAID1 |
4050 BTRFS_BLOCK_GROUP_RAID10))
4054 * If we aren't flushing all things, let us overcommit up to
4055 * 1/2th of the space. If we can flush, don't let us overcommit
4056 * too much, let it overcommit up to 1/8 of the space.
4058 if (flush == BTRFS_RESERVE_FLUSH_ALL)
4063 if (used + bytes < space_info->total_bytes + avail)
4068 static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root,
4069 unsigned long nr_pages, int nr_items)
4071 struct super_block *sb = root->fs_info->sb;
4073 if (down_read_trylock(&sb->s_umount)) {
4074 writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE);
4075 up_read(&sb->s_umount);
4078 * We needn't worry the filesystem going from r/w to r/o though
4079 * we don't acquire ->s_umount mutex, because the filesystem
4080 * should guarantee the delalloc inodes list be empty after
4081 * the filesystem is readonly(all dirty pages are written to
4084 btrfs_start_delalloc_roots(root->fs_info, 0, nr_items);
4085 if (!current->journal_info)
4086 btrfs_wait_ordered_roots(root->fs_info, nr_items);
4090 static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim)
4095 bytes = btrfs_calc_trans_metadata_size(root, 1);
4096 nr = (int)div64_u64(to_reclaim, bytes);
4102 #define EXTENT_SIZE_PER_ITEM (256 * 1024)
4105 * shrink metadata reservation for delalloc
4107 static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig,
4110 struct btrfs_block_rsv *block_rsv;
4111 struct btrfs_space_info *space_info;
4112 struct btrfs_trans_handle *trans;
4116 unsigned long nr_pages;
4119 enum btrfs_reserve_flush_enum flush;
4121 /* Calc the number of the pages we need flush for space reservation */
4122 items = calc_reclaim_items_nr(root, to_reclaim);
4123 to_reclaim = items * EXTENT_SIZE_PER_ITEM;
4125 trans = (struct btrfs_trans_handle *)current->journal_info;
4126 block_rsv = &root->fs_info->delalloc_block_rsv;
4127 space_info = block_rsv->space_info;
4129 delalloc_bytes = percpu_counter_sum_positive(
4130 &root->fs_info->delalloc_bytes);
4131 if (delalloc_bytes == 0) {
4135 btrfs_wait_ordered_roots(root->fs_info, items);
4140 while (delalloc_bytes && loops < 3) {
4141 max_reclaim = min(delalloc_bytes, to_reclaim);
4142 nr_pages = max_reclaim >> PAGE_CACHE_SHIFT;
4143 btrfs_writeback_inodes_sb_nr(root, nr_pages, items);
4145 * We need to wait for the async pages to actually start before
4148 max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages);
4152 if (max_reclaim <= nr_pages)
4155 max_reclaim -= nr_pages;
4157 wait_event(root->fs_info->async_submit_wait,
4158 atomic_read(&root->fs_info->async_delalloc_pages) <=
4162 flush = BTRFS_RESERVE_FLUSH_ALL;
4164 flush = BTRFS_RESERVE_NO_FLUSH;
4165 spin_lock(&space_info->lock);
4166 if (can_overcommit(root, space_info, orig, flush)) {
4167 spin_unlock(&space_info->lock);
4170 spin_unlock(&space_info->lock);
4173 if (wait_ordered && !trans) {
4174 btrfs_wait_ordered_roots(root->fs_info, items);
4176 time_left = schedule_timeout_killable(1);
4180 delalloc_bytes = percpu_counter_sum_positive(
4181 &root->fs_info->delalloc_bytes);
4186 * maybe_commit_transaction - possibly commit the transaction if its ok to
4187 * @root - the root we're allocating for
4188 * @bytes - the number of bytes we want to reserve
4189 * @force - force the commit
4191 * This will check to make sure that committing the transaction will actually
4192 * get us somewhere and then commit the transaction if it does. Otherwise it
4193 * will return -ENOSPC.
4195 static int may_commit_transaction(struct btrfs_root *root,
4196 struct btrfs_space_info *space_info,
4197 u64 bytes, int force)
4199 struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv;
4200 struct btrfs_trans_handle *trans;
4202 trans = (struct btrfs_trans_handle *)current->journal_info;
4209 /* See if there is enough pinned space to make this reservation */
4210 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4215 * See if there is some space in the delayed insertion reservation for
4218 if (space_info != delayed_rsv->space_info)
4221 spin_lock(&delayed_rsv->lock);
4222 if (percpu_counter_compare(&space_info->total_bytes_pinned,
4223 bytes - delayed_rsv->size) >= 0) {
4224 spin_unlock(&delayed_rsv->lock);
4227 spin_unlock(&delayed_rsv->lock);
4230 trans = btrfs_join_transaction(root);
4234 return btrfs_commit_transaction(trans, root);
4238 FLUSH_DELAYED_ITEMS_NR = 1,
4239 FLUSH_DELAYED_ITEMS = 2,
4241 FLUSH_DELALLOC_WAIT = 4,
4246 static int flush_space(struct btrfs_root *root,
4247 struct btrfs_space_info *space_info, u64 num_bytes,
4248 u64 orig_bytes, int state)
4250 struct btrfs_trans_handle *trans;
4255 case FLUSH_DELAYED_ITEMS_NR:
4256 case FLUSH_DELAYED_ITEMS:
4257 if (state == FLUSH_DELAYED_ITEMS_NR)
4258 nr = calc_reclaim_items_nr(root, num_bytes) * 2;
4262 trans = btrfs_join_transaction(root);
4263 if (IS_ERR(trans)) {
4264 ret = PTR_ERR(trans);
4267 ret = btrfs_run_delayed_items_nr(trans, root, nr);
4268 btrfs_end_transaction(trans, root);
4270 case FLUSH_DELALLOC:
4271 case FLUSH_DELALLOC_WAIT:
4272 shrink_delalloc(root, num_bytes * 2, orig_bytes,
4273 state == FLUSH_DELALLOC_WAIT);
4276 trans = btrfs_join_transaction(root);
4277 if (IS_ERR(trans)) {
4278 ret = PTR_ERR(trans);
4281 ret = do_chunk_alloc(trans, root->fs_info->extent_root,
4282 btrfs_get_alloc_profile(root, 0),
4283 CHUNK_ALLOC_NO_FORCE);
4284 btrfs_end_transaction(trans, root);
4289 ret = may_commit_transaction(root, space_info, orig_bytes, 0);
4300 btrfs_calc_reclaim_metadata_size(struct btrfs_root *root,
4301 struct btrfs_space_info *space_info)
4307 to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024,
4309 spin_lock(&space_info->lock);
4310 if (can_overcommit(root, space_info, to_reclaim,
4311 BTRFS_RESERVE_FLUSH_ALL)) {
4316 used = space_info->bytes_used + space_info->bytes_reserved +
4317 space_info->bytes_pinned + space_info->bytes_readonly +
4318 space_info->bytes_may_use;
4319 if (can_overcommit(root, space_info, 1024 * 1024,
4320 BTRFS_RESERVE_FLUSH_ALL))
4321 expected = div_factor_fine(space_info->total_bytes, 95);
4323 expected = div_factor_fine(space_info->total_bytes, 90);
4325 if (used > expected)
4326 to_reclaim = used - expected;
4329 to_reclaim = min(to_reclaim, space_info->bytes_may_use +
4330 space_info->bytes_reserved);
4332 spin_unlock(&space_info->lock);
4337 static inline int need_do_async_reclaim(struct btrfs_space_info *space_info,
4338 struct btrfs_fs_info *fs_info, u64 used)
4340 return (used >= div_factor_fine(space_info->total_bytes, 98) &&
4341 !btrfs_fs_closing(fs_info) &&
4342 !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state));
4345 static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info,
4346 struct btrfs_fs_info *fs_info,
4351 spin_lock(&space_info->lock);
4353 * We run out of space and have not got any free space via flush_space,
4354 * so don't bother doing async reclaim.
4356 if (flush_state > COMMIT_TRANS && space_info->full) {
4357 spin_unlock(&space_info->lock);
4361 used = space_info->bytes_used + space_info->bytes_reserved +
4362 space_info->bytes_pinned + space_info->bytes_readonly +
4363 space_info->bytes_may_use;
4364 if (need_do_async_reclaim(space_info, fs_info, used)) {
4365 spin_unlock(&space_info->lock);
4368 spin_unlock(&space_info->lock);
4373 static void btrfs_async_reclaim_metadata_space(struct work_struct *work)
4375 struct btrfs_fs_info *fs_info;
4376 struct btrfs_space_info *space_info;
4380 fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work);
4381 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4383 to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root,
4388 flush_state = FLUSH_DELAYED_ITEMS_NR;
4390 flush_space(fs_info->fs_root, space_info, to_reclaim,
4391 to_reclaim, flush_state);
4393 if (!btrfs_need_do_async_reclaim(space_info, fs_info,
4396 } while (flush_state <= COMMIT_TRANS);
4398 if (btrfs_need_do_async_reclaim(space_info, fs_info, flush_state))
4399 queue_work(system_unbound_wq, work);
4402 void btrfs_init_async_reclaim_work(struct work_struct *work)
4404 INIT_WORK(work, btrfs_async_reclaim_metadata_space);
4408 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
4409 * @root - the root we're allocating for
4410 * @block_rsv - the block_rsv we're allocating for
4411 * @orig_bytes - the number of bytes we want
4412 * @flush - whether or not we can flush to make our reservation
4414 * This will reserve orgi_bytes number of bytes from the space info associated
4415 * with the block_rsv. If there is not enough space it will make an attempt to
4416 * flush out space to make room. It will do this by flushing delalloc if
4417 * possible or committing the transaction. If flush is 0 then no attempts to
4418 * regain reservations will be made and this will fail if there is not enough
4421 static int reserve_metadata_bytes(struct btrfs_root *root,
4422 struct btrfs_block_rsv *block_rsv,
4424 enum btrfs_reserve_flush_enum flush)
4426 struct btrfs_space_info *space_info = block_rsv->space_info;
4428 u64 num_bytes = orig_bytes;
4429 int flush_state = FLUSH_DELAYED_ITEMS_NR;
4431 bool flushing = false;
4435 spin_lock(&space_info->lock);
4437 * We only want to wait if somebody other than us is flushing and we
4438 * are actually allowed to flush all things.
4440 while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing &&
4441 space_info->flush) {
4442 spin_unlock(&space_info->lock);
4444 * If we have a trans handle we can't wait because the flusher
4445 * may have to commit the transaction, which would mean we would
4446 * deadlock since we are waiting for the flusher to finish, but
4447 * hold the current transaction open.
4449 if (current->journal_info)
4451 ret = wait_event_killable(space_info->wait, !space_info->flush);
4452 /* Must have been killed, return */
4456 spin_lock(&space_info->lock);
4460 used = space_info->bytes_used + space_info->bytes_reserved +
4461 space_info->bytes_pinned + space_info->bytes_readonly +
4462 space_info->bytes_may_use;
4465 * The idea here is that we've not already over-reserved the block group
4466 * then we can go ahead and save our reservation first and then start
4467 * flushing if we need to. Otherwise if we've already overcommitted
4468 * lets start flushing stuff first and then come back and try to make
4471 if (used <= space_info->total_bytes) {
4472 if (used + orig_bytes <= space_info->total_bytes) {
4473 space_info->bytes_may_use += orig_bytes;
4474 trace_btrfs_space_reservation(root->fs_info,
4475 "space_info", space_info->flags, orig_bytes, 1);
4479 * Ok set num_bytes to orig_bytes since we aren't
4480 * overocmmitted, this way we only try and reclaim what
4483 num_bytes = orig_bytes;
4487 * Ok we're over committed, set num_bytes to the overcommitted
4488 * amount plus the amount of bytes that we need for this
4491 num_bytes = used - space_info->total_bytes +
4495 if (ret && can_overcommit(root, space_info, orig_bytes, flush)) {
4496 space_info->bytes_may_use += orig_bytes;
4497 trace_btrfs_space_reservation(root->fs_info, "space_info",
4498 space_info->flags, orig_bytes,
4504 * Couldn't make our reservation, save our place so while we're trying
4505 * to reclaim space we can actually use it instead of somebody else
4506 * stealing it from us.
4508 * We make the other tasks wait for the flush only when we can flush
4511 if (ret && flush != BTRFS_RESERVE_NO_FLUSH) {
4513 space_info->flush = 1;
4514 } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) {
4516 if (need_do_async_reclaim(space_info, root->fs_info, used) &&
4517 !work_busy(&root->fs_info->async_reclaim_work))
4518 queue_work(system_unbound_wq,
4519 &root->fs_info->async_reclaim_work);
4521 spin_unlock(&space_info->lock);
4523 if (!ret || flush == BTRFS_RESERVE_NO_FLUSH)
4526 ret = flush_space(root, space_info, num_bytes, orig_bytes,
4531 * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock
4532 * would happen. So skip delalloc flush.
4534 if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4535 (flush_state == FLUSH_DELALLOC ||
4536 flush_state == FLUSH_DELALLOC_WAIT))
4537 flush_state = ALLOC_CHUNK;
4541 else if (flush == BTRFS_RESERVE_FLUSH_LIMIT &&
4542 flush_state < COMMIT_TRANS)
4544 else if (flush == BTRFS_RESERVE_FLUSH_ALL &&
4545 flush_state <= COMMIT_TRANS)
4549 if (ret == -ENOSPC &&
4550 unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) {
4551 struct btrfs_block_rsv *global_rsv =
4552 &root->fs_info->global_block_rsv;
4554 if (block_rsv != global_rsv &&
4555 !block_rsv_use_bytes(global_rsv, orig_bytes))
4559 trace_btrfs_space_reservation(root->fs_info,
4560 "space_info:enospc",
4561 space_info->flags, orig_bytes, 1);
4563 spin_lock(&space_info->lock);
4564 space_info->flush = 0;
4565 wake_up_all(&space_info->wait);
4566 spin_unlock(&space_info->lock);
4571 static struct btrfs_block_rsv *get_block_rsv(
4572 const struct btrfs_trans_handle *trans,
4573 const struct btrfs_root *root)
4575 struct btrfs_block_rsv *block_rsv = NULL;
4577 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4578 block_rsv = trans->block_rsv;
4580 if (root == root->fs_info->csum_root && trans->adding_csums)
4581 block_rsv = trans->block_rsv;
4583 if (root == root->fs_info->uuid_root)
4584 block_rsv = trans->block_rsv;
4587 block_rsv = root->block_rsv;
4590 block_rsv = &root->fs_info->empty_block_rsv;
4595 static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv,
4599 spin_lock(&block_rsv->lock);
4600 if (block_rsv->reserved >= num_bytes) {
4601 block_rsv->reserved -= num_bytes;
4602 if (block_rsv->reserved < block_rsv->size)
4603 block_rsv->full = 0;
4606 spin_unlock(&block_rsv->lock);
4610 static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv,
4611 u64 num_bytes, int update_size)
4613 spin_lock(&block_rsv->lock);
4614 block_rsv->reserved += num_bytes;
4616 block_rsv->size += num_bytes;
4617 else if (block_rsv->reserved >= block_rsv->size)
4618 block_rsv->full = 1;
4619 spin_unlock(&block_rsv->lock);
4622 int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
4623 struct btrfs_block_rsv *dest, u64 num_bytes,
4626 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
4629 if (global_rsv->space_info != dest->space_info)
4632 spin_lock(&global_rsv->lock);
4633 min_bytes = div_factor(global_rsv->size, min_factor);
4634 if (global_rsv->reserved < min_bytes + num_bytes) {
4635 spin_unlock(&global_rsv->lock);
4638 global_rsv->reserved -= num_bytes;
4639 if (global_rsv->reserved < global_rsv->size)
4640 global_rsv->full = 0;
4641 spin_unlock(&global_rsv->lock);
4643 block_rsv_add_bytes(dest, num_bytes, 1);
4647 static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
4648 struct btrfs_block_rsv *block_rsv,
4649 struct btrfs_block_rsv *dest, u64 num_bytes)
4651 struct btrfs_space_info *space_info = block_rsv->space_info;
4653 spin_lock(&block_rsv->lock);
4654 if (num_bytes == (u64)-1)
4655 num_bytes = block_rsv->size;
4656 block_rsv->size -= num_bytes;
4657 if (block_rsv->reserved >= block_rsv->size) {
4658 num_bytes = block_rsv->reserved - block_rsv->size;
4659 block_rsv->reserved = block_rsv->size;
4660 block_rsv->full = 1;
4664 spin_unlock(&block_rsv->lock);
4666 if (num_bytes > 0) {
4668 spin_lock(&dest->lock);
4672 bytes_to_add = dest->size - dest->reserved;
4673 bytes_to_add = min(num_bytes, bytes_to_add);
4674 dest->reserved += bytes_to_add;
4675 if (dest->reserved >= dest->size)
4677 num_bytes -= bytes_to_add;
4679 spin_unlock(&dest->lock);
4682 spin_lock(&space_info->lock);
4683 space_info->bytes_may_use -= num_bytes;
4684 trace_btrfs_space_reservation(fs_info, "space_info",
4685 space_info->flags, num_bytes, 0);
4686 spin_unlock(&space_info->lock);
4691 static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src,
4692 struct btrfs_block_rsv *dst, u64 num_bytes)
4696 ret = block_rsv_use_bytes(src, num_bytes);
4700 block_rsv_add_bytes(dst, num_bytes, 1);
4704 void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type)
4706 memset(rsv, 0, sizeof(*rsv));
4707 spin_lock_init(&rsv->lock);
4711 struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root,
4712 unsigned short type)
4714 struct btrfs_block_rsv *block_rsv;
4715 struct btrfs_fs_info *fs_info = root->fs_info;
4717 block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS);
4721 btrfs_init_block_rsv(block_rsv, type);
4722 block_rsv->space_info = __find_space_info(fs_info,
4723 BTRFS_BLOCK_GROUP_METADATA);
4727 void btrfs_free_block_rsv(struct btrfs_root *root,
4728 struct btrfs_block_rsv *rsv)
4732 btrfs_block_rsv_release(root, rsv, (u64)-1);
4736 int btrfs_block_rsv_add(struct btrfs_root *root,
4737 struct btrfs_block_rsv *block_rsv, u64 num_bytes,
4738 enum btrfs_reserve_flush_enum flush)
4745 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4747 block_rsv_add_bytes(block_rsv, num_bytes, 1);
4754 int btrfs_block_rsv_check(struct btrfs_root *root,
4755 struct btrfs_block_rsv *block_rsv, int min_factor)
4763 spin_lock(&block_rsv->lock);
4764 num_bytes = div_factor(block_rsv->size, min_factor);
4765 if (block_rsv->reserved >= num_bytes)
4767 spin_unlock(&block_rsv->lock);
4772 int btrfs_block_rsv_refill(struct btrfs_root *root,
4773 struct btrfs_block_rsv *block_rsv, u64 min_reserved,
4774 enum btrfs_reserve_flush_enum flush)
4782 spin_lock(&block_rsv->lock);
4783 num_bytes = min_reserved;
4784 if (block_rsv->reserved >= num_bytes)
4787 num_bytes -= block_rsv->reserved;
4788 spin_unlock(&block_rsv->lock);
4793 ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush);
4795 block_rsv_add_bytes(block_rsv, num_bytes, 0);
4802 int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv,
4803 struct btrfs_block_rsv *dst_rsv,
4806 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4809 void btrfs_block_rsv_release(struct btrfs_root *root,
4810 struct btrfs_block_rsv *block_rsv,
4813 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
4814 if (global_rsv == block_rsv ||
4815 block_rsv->space_info != global_rsv->space_info)
4817 block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv,
4822 * helper to calculate size of global block reservation.
4823 * the desired value is sum of space used by extent tree,
4824 * checksum tree and root tree
4826 static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
4828 struct btrfs_space_info *sinfo;
4832 int csum_size = btrfs_super_csum_size(fs_info->super_copy);
4834 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA);
4835 spin_lock(&sinfo->lock);
4836 data_used = sinfo->bytes_used;
4837 spin_unlock(&sinfo->lock);
4839 sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4840 spin_lock(&sinfo->lock);
4841 if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA)
4843 meta_used = sinfo->bytes_used;
4844 spin_unlock(&sinfo->lock);
4846 num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) *
4848 num_bytes += div64_u64(data_used + meta_used, 50);
4850 if (num_bytes * 3 > meta_used)
4851 num_bytes = div64_u64(meta_used, 3);
4853 return ALIGN(num_bytes, fs_info->extent_root->nodesize << 10);
4856 static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
4858 struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv;
4859 struct btrfs_space_info *sinfo = block_rsv->space_info;
4862 num_bytes = calc_global_metadata_size(fs_info);
4864 spin_lock(&sinfo->lock);
4865 spin_lock(&block_rsv->lock);
4867 block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024);
4869 num_bytes = sinfo->bytes_used + sinfo->bytes_pinned +
4870 sinfo->bytes_reserved + sinfo->bytes_readonly +
4871 sinfo->bytes_may_use;
4873 if (sinfo->total_bytes > num_bytes) {
4874 num_bytes = sinfo->total_bytes - num_bytes;
4875 block_rsv->reserved += num_bytes;
4876 sinfo->bytes_may_use += num_bytes;
4877 trace_btrfs_space_reservation(fs_info, "space_info",
4878 sinfo->flags, num_bytes, 1);
4881 if (block_rsv->reserved >= block_rsv->size) {
4882 num_bytes = block_rsv->reserved - block_rsv->size;
4883 sinfo->bytes_may_use -= num_bytes;
4884 trace_btrfs_space_reservation(fs_info, "space_info",
4885 sinfo->flags, num_bytes, 0);
4886 block_rsv->reserved = block_rsv->size;
4887 block_rsv->full = 1;
4890 spin_unlock(&block_rsv->lock);
4891 spin_unlock(&sinfo->lock);
4894 static void init_global_block_rsv(struct btrfs_fs_info *fs_info)
4896 struct btrfs_space_info *space_info;
4898 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
4899 fs_info->chunk_block_rsv.space_info = space_info;
4901 space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA);
4902 fs_info->global_block_rsv.space_info = space_info;
4903 fs_info->delalloc_block_rsv.space_info = space_info;
4904 fs_info->trans_block_rsv.space_info = space_info;
4905 fs_info->empty_block_rsv.space_info = space_info;
4906 fs_info->delayed_block_rsv.space_info = space_info;
4908 fs_info->extent_root->block_rsv = &fs_info->global_block_rsv;
4909 fs_info->csum_root->block_rsv = &fs_info->global_block_rsv;
4910 fs_info->dev_root->block_rsv = &fs_info->global_block_rsv;
4911 fs_info->tree_root->block_rsv = &fs_info->global_block_rsv;
4912 if (fs_info->quota_root)
4913 fs_info->quota_root->block_rsv = &fs_info->global_block_rsv;
4914 fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv;
4916 update_global_block_rsv(fs_info);
4919 static void release_global_block_rsv(struct btrfs_fs_info *fs_info)
4921 block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL,
4923 WARN_ON(fs_info->delalloc_block_rsv.size > 0);
4924 WARN_ON(fs_info->delalloc_block_rsv.reserved > 0);
4925 WARN_ON(fs_info->trans_block_rsv.size > 0);
4926 WARN_ON(fs_info->trans_block_rsv.reserved > 0);
4927 WARN_ON(fs_info->chunk_block_rsv.size > 0);
4928 WARN_ON(fs_info->chunk_block_rsv.reserved > 0);
4929 WARN_ON(fs_info->delayed_block_rsv.size > 0);
4930 WARN_ON(fs_info->delayed_block_rsv.reserved > 0);
4933 void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
4934 struct btrfs_root *root)
4936 if (!trans->block_rsv)
4939 if (!trans->bytes_reserved)
4942 trace_btrfs_space_reservation(root->fs_info, "transaction",
4943 trans->transid, trans->bytes_reserved, 0);
4944 btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
4945 trans->bytes_reserved = 0;
4948 /* Can only return 0 or -ENOSPC */
4949 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
4950 struct inode *inode)
4952 struct btrfs_root *root = BTRFS_I(inode)->root;
4953 struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root);
4954 struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv;
4957 * We need to hold space in order to delete our orphan item once we've
4958 * added it, so this takes the reservation so we can release it later
4959 * when we are truly done with the orphan item.
4961 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4962 trace_btrfs_space_reservation(root->fs_info, "orphan",
4963 btrfs_ino(inode), num_bytes, 1);
4964 return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes);
4967 void btrfs_orphan_release_metadata(struct inode *inode)
4969 struct btrfs_root *root = BTRFS_I(inode)->root;
4970 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1);
4971 trace_btrfs_space_reservation(root->fs_info, "orphan",
4972 btrfs_ino(inode), num_bytes, 0);
4973 btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes);
4977 * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation
4978 * root: the root of the parent directory
4979 * rsv: block reservation
4980 * items: the number of items that we need do reservation
4981 * qgroup_reserved: used to return the reserved size in qgroup
4983 * This function is used to reserve the space for snapshot/subvolume
4984 * creation and deletion. Those operations are different with the
4985 * common file/directory operations, they change two fs/file trees
4986 * and root tree, the number of items that the qgroup reserves is
4987 * different with the free space reservation. So we can not use
4988 * the space reseravtion mechanism in start_transaction().
4990 int btrfs_subvolume_reserve_metadata(struct btrfs_root *root,
4991 struct btrfs_block_rsv *rsv,
4993 u64 *qgroup_reserved,
4994 bool use_global_rsv)
4998 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
5000 if (root->fs_info->quota_enabled) {
5001 /* One for parent inode, two for dir entries */
5002 num_bytes = 3 * root->nodesize;
5003 ret = btrfs_qgroup_reserve(root, num_bytes);
5010 *qgroup_reserved = num_bytes;
5012 num_bytes = btrfs_calc_trans_metadata_size(root, items);
5013 rsv->space_info = __find_space_info(root->fs_info,
5014 BTRFS_BLOCK_GROUP_METADATA);
5015 ret = btrfs_block_rsv_add(root, rsv, num_bytes,
5016 BTRFS_RESERVE_FLUSH_ALL);
5018 if (ret == -ENOSPC && use_global_rsv)
5019 ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes);
5022 if (*qgroup_reserved)
5023 btrfs_qgroup_free(root, *qgroup_reserved);
5029 void btrfs_subvolume_release_metadata(struct btrfs_root *root,
5030 struct btrfs_block_rsv *rsv,
5031 u64 qgroup_reserved)
5033 btrfs_block_rsv_release(root, rsv, (u64)-1);
5034 if (qgroup_reserved)
5035 btrfs_qgroup_free(root, qgroup_reserved);
5039 * drop_outstanding_extent - drop an outstanding extent
5040 * @inode: the inode we're dropping the extent for
5042 * This is called when we are freeing up an outstanding extent, either called
5043 * after an error or after an extent is written. This will return the number of
5044 * reserved extents that need to be freed. This must be called with
5045 * BTRFS_I(inode)->lock held.
5047 static unsigned drop_outstanding_extent(struct inode *inode)
5049 unsigned drop_inode_space = 0;
5050 unsigned dropped_extents = 0;
5052 BUG_ON(!BTRFS_I(inode)->outstanding_extents);
5053 BTRFS_I(inode)->outstanding_extents--;
5055 if (BTRFS_I(inode)->outstanding_extents == 0 &&
5056 test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5057 &BTRFS_I(inode)->runtime_flags))
5058 drop_inode_space = 1;
5061 * If we have more or the same amount of outsanding extents than we have
5062 * reserved then we need to leave the reserved extents count alone.
5064 if (BTRFS_I(inode)->outstanding_extents >=
5065 BTRFS_I(inode)->reserved_extents)
5066 return drop_inode_space;
5068 dropped_extents = BTRFS_I(inode)->reserved_extents -
5069 BTRFS_I(inode)->outstanding_extents;
5070 BTRFS_I(inode)->reserved_extents -= dropped_extents;
5071 return dropped_extents + drop_inode_space;
5075 * calc_csum_metadata_size - return the amount of metada space that must be
5076 * reserved/free'd for the given bytes.
5077 * @inode: the inode we're manipulating
5078 * @num_bytes: the number of bytes in question
5079 * @reserve: 1 if we are reserving space, 0 if we are freeing space
5081 * This adjusts the number of csum_bytes in the inode and then returns the
5082 * correct amount of metadata that must either be reserved or freed. We
5083 * calculate how many checksums we can fit into one leaf and then divide the
5084 * number of bytes that will need to be checksumed by this value to figure out
5085 * how many checksums will be required. If we are adding bytes then the number
5086 * may go up and we will return the number of additional bytes that must be
5087 * reserved. If it is going down we will return the number of bytes that must
5090 * This must be called with BTRFS_I(inode)->lock held.
5092 static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes,
5095 struct btrfs_root *root = BTRFS_I(inode)->root;
5097 int num_csums_per_leaf;
5101 if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM &&
5102 BTRFS_I(inode)->csum_bytes == 0)
5105 old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5107 BTRFS_I(inode)->csum_bytes += num_bytes;
5109 BTRFS_I(inode)->csum_bytes -= num_bytes;
5110 csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item);
5111 num_csums_per_leaf = (int)div64_u64(csum_size,
5112 sizeof(struct btrfs_csum_item) +
5113 sizeof(struct btrfs_disk_key));
5114 num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize);
5115 num_csums = num_csums + num_csums_per_leaf - 1;
5116 num_csums = num_csums / num_csums_per_leaf;
5118 old_csums = old_csums + num_csums_per_leaf - 1;
5119 old_csums = old_csums / num_csums_per_leaf;
5121 /* No change, no need to reserve more */
5122 if (old_csums == num_csums)
5126 return btrfs_calc_trans_metadata_size(root,
5127 num_csums - old_csums);
5129 return btrfs_calc_trans_metadata_size(root, old_csums - num_csums);
5132 int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes)
5134 struct btrfs_root *root = BTRFS_I(inode)->root;
5135 struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv;
5138 unsigned nr_extents = 0;
5139 int extra_reserve = 0;
5140 enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL;
5142 bool delalloc_lock = true;
5146 /* If we are a free space inode we need to not flush since we will be in
5147 * the middle of a transaction commit. We also don't need the delalloc
5148 * mutex since we won't race with anybody. We need this mostly to make
5149 * lockdep shut its filthy mouth.
5151 if (btrfs_is_free_space_inode(inode)) {
5152 flush = BTRFS_RESERVE_NO_FLUSH;
5153 delalloc_lock = false;
5156 if (flush != BTRFS_RESERVE_NO_FLUSH &&
5157 btrfs_transaction_in_commit(root->fs_info))
5158 schedule_timeout(1);
5161 mutex_lock(&BTRFS_I(inode)->delalloc_mutex);
5163 num_bytes = ALIGN(num_bytes, root->sectorsize);
5165 spin_lock(&BTRFS_I(inode)->lock);
5166 BTRFS_I(inode)->outstanding_extents++;
5168 if (BTRFS_I(inode)->outstanding_extents >
5169 BTRFS_I(inode)->reserved_extents)
5170 nr_extents = BTRFS_I(inode)->outstanding_extents -
5171 BTRFS_I(inode)->reserved_extents;
5174 * Add an item to reserve for updating the inode when we complete the
5177 if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5178 &BTRFS_I(inode)->runtime_flags)) {
5183 to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents);
5184 to_reserve += calc_csum_metadata_size(inode, num_bytes, 1);
5185 csum_bytes = BTRFS_I(inode)->csum_bytes;
5186 spin_unlock(&BTRFS_I(inode)->lock);
5188 if (root->fs_info->quota_enabled) {
5189 ret = btrfs_qgroup_reserve(root, num_bytes +
5190 nr_extents * root->nodesize);
5195 ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush);
5196 if (unlikely(ret)) {
5197 if (root->fs_info->quota_enabled)
5198 btrfs_qgroup_free(root, num_bytes +
5199 nr_extents * root->nodesize);
5203 spin_lock(&BTRFS_I(inode)->lock);
5204 if (extra_reserve) {
5205 set_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
5206 &BTRFS_I(inode)->runtime_flags);
5209 BTRFS_I(inode)->reserved_extents += nr_extents;
5210 spin_unlock(&BTRFS_I(inode)->lock);
5213 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5216 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5217 btrfs_ino(inode), to_reserve, 1);
5218 block_rsv_add_bytes(block_rsv, to_reserve, 1);
5223 spin_lock(&BTRFS_I(inode)->lock);
5224 dropped = drop_outstanding_extent(inode);
5226 * If the inodes csum_bytes is the same as the original
5227 * csum_bytes then we know we haven't raced with any free()ers
5228 * so we can just reduce our inodes csum bytes and carry on.
5230 if (BTRFS_I(inode)->csum_bytes == csum_bytes) {
5231 calc_csum_metadata_size(inode, num_bytes, 0);
5233 u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes;
5237 * This is tricky, but first we need to figure out how much we
5238 * free'd from any free-ers that occured during this
5239 * reservation, so we reset ->csum_bytes to the csum_bytes
5240 * before we dropped our lock, and then call the free for the
5241 * number of bytes that were freed while we were trying our
5244 bytes = csum_bytes - BTRFS_I(inode)->csum_bytes;
5245 BTRFS_I(inode)->csum_bytes = csum_bytes;
5246 to_free = calc_csum_metadata_size(inode, bytes, 0);
5250 * Now we need to see how much we would have freed had we not
5251 * been making this reservation and our ->csum_bytes were not
5252 * artificially inflated.
5254 BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes;
5255 bytes = csum_bytes - orig_csum_bytes;
5256 bytes = calc_csum_metadata_size(inode, bytes, 0);
5259 * Now reset ->csum_bytes to what it should be. If bytes is
5260 * more than to_free then we would have free'd more space had we
5261 * not had an artificially high ->csum_bytes, so we need to free
5262 * the remainder. If bytes is the same or less then we don't
5263 * need to do anything, the other free-ers did the correct
5266 BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes;
5267 if (bytes > to_free)
5268 to_free = bytes - to_free;
5272 spin_unlock(&BTRFS_I(inode)->lock);
5274 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5277 btrfs_block_rsv_release(root, block_rsv, to_free);
5278 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5279 btrfs_ino(inode), to_free, 0);
5282 mutex_unlock(&BTRFS_I(inode)->delalloc_mutex);
5287 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
5288 * @inode: the inode to release the reservation for
5289 * @num_bytes: the number of bytes we're releasing
5291 * This will release the metadata reservation for an inode. This can be called
5292 * once we complete IO for a given set of bytes to release their metadata
5295 void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes)
5297 struct btrfs_root *root = BTRFS_I(inode)->root;
5301 num_bytes = ALIGN(num_bytes, root->sectorsize);
5302 spin_lock(&BTRFS_I(inode)->lock);
5303 dropped = drop_outstanding_extent(inode);
5306 to_free = calc_csum_metadata_size(inode, num_bytes, 0);
5307 spin_unlock(&BTRFS_I(inode)->lock);
5309 to_free += btrfs_calc_trans_metadata_size(root, dropped);
5311 trace_btrfs_space_reservation(root->fs_info, "delalloc",
5312 btrfs_ino(inode), to_free, 0);
5313 if (root->fs_info->quota_enabled) {
5314 btrfs_qgroup_free(root, num_bytes +
5315 dropped * root->nodesize);
5318 btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv,
5323 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
5324 * @inode: inode we're writing to
5325 * @num_bytes: the number of bytes we want to allocate
5327 * This will do the following things
5329 * o reserve space in the data space info for num_bytes
5330 * o reserve space in the metadata space info based on number of outstanding
5331 * extents and how much csums will be needed
5332 * o add to the inodes ->delalloc_bytes
5333 * o add it to the fs_info's delalloc inodes list.
5335 * This will return 0 for success and -ENOSPC if there is no space left.
5337 int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes)
5341 ret = btrfs_check_data_free_space(inode, num_bytes);
5345 ret = btrfs_delalloc_reserve_metadata(inode, num_bytes);
5347 btrfs_free_reserved_data_space(inode, num_bytes);
5355 * btrfs_delalloc_release_space - release data and metadata space for delalloc
5356 * @inode: inode we're releasing space for
5357 * @num_bytes: the number of bytes we want to free up
5359 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
5360 * called in the case that we don't need the metadata AND data reservations
5361 * anymore. So if there is an error or we insert an inline extent.
5363 * This function will release the metadata space that was not used and will
5364 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
5365 * list if there are no delalloc bytes left.
5367 void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes)
5369 btrfs_delalloc_release_metadata(inode, num_bytes);
5370 btrfs_free_reserved_data_space(inode, num_bytes);
5373 static int update_block_group(struct btrfs_root *root,
5374 u64 bytenr, u64 num_bytes, int alloc)
5376 struct btrfs_block_group_cache *cache = NULL;
5377 struct btrfs_fs_info *info = root->fs_info;
5378 u64 total = num_bytes;
5383 /* block accounting for super block */
5384 spin_lock(&info->delalloc_root_lock);
5385 old_val = btrfs_super_bytes_used(info->super_copy);
5387 old_val += num_bytes;
5389 old_val -= num_bytes;
5390 btrfs_set_super_bytes_used(info->super_copy, old_val);
5391 spin_unlock(&info->delalloc_root_lock);
5394 cache = btrfs_lookup_block_group(info, bytenr);
5397 if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
5398 BTRFS_BLOCK_GROUP_RAID1 |
5399 BTRFS_BLOCK_GROUP_RAID10))
5404 * If this block group has free space cache written out, we
5405 * need to make sure to load it if we are removing space. This
5406 * is because we need the unpinning stage to actually add the
5407 * space back to the block group, otherwise we will leak space.
5409 if (!alloc && cache->cached == BTRFS_CACHE_NO)
5410 cache_block_group(cache, 1);
5412 byte_in_group = bytenr - cache->key.objectid;
5413 WARN_ON(byte_in_group > cache->key.offset);
5415 spin_lock(&cache->space_info->lock);
5416 spin_lock(&cache->lock);
5418 if (btrfs_test_opt(root, SPACE_CACHE) &&
5419 cache->disk_cache_state < BTRFS_DC_CLEAR)
5420 cache->disk_cache_state = BTRFS_DC_CLEAR;
5423 old_val = btrfs_block_group_used(&cache->item);
5424 num_bytes = min(total, cache->key.offset - byte_in_group);
5426 old_val += num_bytes;
5427 btrfs_set_block_group_used(&cache->item, old_val);
5428 cache->reserved -= num_bytes;
5429 cache->space_info->bytes_reserved -= num_bytes;
5430 cache->space_info->bytes_used += num_bytes;
5431 cache->space_info->disk_used += num_bytes * factor;
5432 spin_unlock(&cache->lock);
5433 spin_unlock(&cache->space_info->lock);
5435 old_val -= num_bytes;
5436 btrfs_set_block_group_used(&cache->item, old_val);
5437 cache->pinned += num_bytes;
5438 cache->space_info->bytes_pinned += num_bytes;
5439 cache->space_info->bytes_used -= num_bytes;
5440 cache->space_info->disk_used -= num_bytes * factor;
5441 spin_unlock(&cache->lock);
5442 spin_unlock(&cache->space_info->lock);
5444 set_extent_dirty(info->pinned_extents,
5445 bytenr, bytenr + num_bytes - 1,
5446 GFP_NOFS | __GFP_NOFAIL);
5448 btrfs_put_block_group(cache);
5450 bytenr += num_bytes;
5455 static u64 first_logical_byte(struct btrfs_root *root, u64 search_start)
5457 struct btrfs_block_group_cache *cache;
5460 spin_lock(&root->fs_info->block_group_cache_lock);
5461 bytenr = root->fs_info->first_logical_byte;
5462 spin_unlock(&root->fs_info->block_group_cache_lock);
5464 if (bytenr < (u64)-1)
5467 cache = btrfs_lookup_first_block_group(root->fs_info, search_start);
5471 bytenr = cache->key.objectid;
5472 btrfs_put_block_group(cache);
5477 static int pin_down_extent(struct btrfs_root *root,
5478 struct btrfs_block_group_cache *cache,
5479 u64 bytenr, u64 num_bytes, int reserved)
5481 spin_lock(&cache->space_info->lock);
5482 spin_lock(&cache->lock);
5483 cache->pinned += num_bytes;
5484 cache->space_info->bytes_pinned += num_bytes;
5486 cache->reserved -= num_bytes;
5487 cache->space_info->bytes_reserved -= num_bytes;
5489 spin_unlock(&cache->lock);
5490 spin_unlock(&cache->space_info->lock);
5492 set_extent_dirty(root->fs_info->pinned_extents, bytenr,
5493 bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL);
5495 trace_btrfs_reserved_extent_free(root, bytenr, num_bytes);
5500 * this function must be called within transaction
5502 int btrfs_pin_extent(struct btrfs_root *root,
5503 u64 bytenr, u64 num_bytes, int reserved)
5505 struct btrfs_block_group_cache *cache;
5507 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5508 BUG_ON(!cache); /* Logic error */
5510 pin_down_extent(root, cache, bytenr, num_bytes, reserved);
5512 btrfs_put_block_group(cache);
5517 * this function must be called within transaction
5519 int btrfs_pin_extent_for_log_replay(struct btrfs_root *root,
5520 u64 bytenr, u64 num_bytes)
5522 struct btrfs_block_group_cache *cache;
5525 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
5530 * pull in the free space cache (if any) so that our pin
5531 * removes the free space from the cache. We have load_only set
5532 * to one because the slow code to read in the free extents does check
5533 * the pinned extents.
5535 cache_block_group(cache, 1);
5537 pin_down_extent(root, cache, bytenr, num_bytes, 0);
5539 /* remove us from the free space cache (if we're there at all) */
5540 ret = btrfs_remove_free_space(cache, bytenr, num_bytes);
5541 btrfs_put_block_group(cache);
5545 static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes)
5548 struct btrfs_block_group_cache *block_group;
5549 struct btrfs_caching_control *caching_ctl;
5551 block_group = btrfs_lookup_block_group(root->fs_info, start);
5555 cache_block_group(block_group, 0);
5556 caching_ctl = get_caching_control(block_group);
5560 BUG_ON(!block_group_cache_done(block_group));
5561 ret = btrfs_remove_free_space(block_group, start, num_bytes);
5563 mutex_lock(&caching_ctl->mutex);
5565 if (start >= caching_ctl->progress) {
5566 ret = add_excluded_extent(root, start, num_bytes);
5567 } else if (start + num_bytes <= caching_ctl->progress) {
5568 ret = btrfs_remove_free_space(block_group,
5571 num_bytes = caching_ctl->progress - start;
5572 ret = btrfs_remove_free_space(block_group,
5577 num_bytes = (start + num_bytes) -
5578 caching_ctl->progress;
5579 start = caching_ctl->progress;
5580 ret = add_excluded_extent(root, start, num_bytes);
5583 mutex_unlock(&caching_ctl->mutex);
5584 put_caching_control(caching_ctl);
5586 btrfs_put_block_group(block_group);
5590 int btrfs_exclude_logged_extents(struct btrfs_root *log,
5591 struct extent_buffer *eb)
5593 struct btrfs_file_extent_item *item;
5594 struct btrfs_key key;
5598 if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS))
5601 for (i = 0; i < btrfs_header_nritems(eb); i++) {
5602 btrfs_item_key_to_cpu(eb, &key, i);
5603 if (key.type != BTRFS_EXTENT_DATA_KEY)
5605 item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
5606 found_type = btrfs_file_extent_type(eb, item);
5607 if (found_type == BTRFS_FILE_EXTENT_INLINE)
5609 if (btrfs_file_extent_disk_bytenr(eb, item) == 0)
5611 key.objectid = btrfs_file_extent_disk_bytenr(eb, item);
5612 key.offset = btrfs_file_extent_disk_num_bytes(eb, item);
5613 __exclude_logged_extent(log, key.objectid, key.offset);
5620 * btrfs_update_reserved_bytes - update the block_group and space info counters
5621 * @cache: The cache we are manipulating
5622 * @num_bytes: The number of bytes in question
5623 * @reserve: One of the reservation enums
5624 * @delalloc: The blocks are allocated for the delalloc write
5626 * This is called by the allocator when it reserves space, or by somebody who is
5627 * freeing space that was never actually used on disk. For example if you
5628 * reserve some space for a new leaf in transaction A and before transaction A
5629 * commits you free that leaf, you call this with reserve set to 0 in order to
5630 * clear the reservation.
5632 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
5633 * ENOSPC accounting. For data we handle the reservation through clearing the
5634 * delalloc bits in the io_tree. We have to do this since we could end up
5635 * allocating less disk space for the amount of data we have reserved in the
5636 * case of compression.
5638 * If this is a reservation and the block group has become read only we cannot
5639 * make the reservation and return -EAGAIN, otherwise this function always
5642 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
5643 u64 num_bytes, int reserve, int delalloc)
5645 struct btrfs_space_info *space_info = cache->space_info;
5648 spin_lock(&space_info->lock);
5649 spin_lock(&cache->lock);
5650 if (reserve != RESERVE_FREE) {
5654 cache->reserved += num_bytes;
5655 space_info->bytes_reserved += num_bytes;
5656 if (reserve == RESERVE_ALLOC) {
5657 trace_btrfs_space_reservation(cache->fs_info,
5658 "space_info", space_info->flags,
5660 space_info->bytes_may_use -= num_bytes;
5664 cache->delalloc_bytes += num_bytes;
5668 space_info->bytes_readonly += num_bytes;
5669 cache->reserved -= num_bytes;
5670 space_info->bytes_reserved -= num_bytes;
5673 cache->delalloc_bytes -= num_bytes;
5675 spin_unlock(&cache->lock);
5676 spin_unlock(&space_info->lock);
5680 void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
5681 struct btrfs_root *root)
5683 struct btrfs_fs_info *fs_info = root->fs_info;
5684 struct btrfs_caching_control *next;
5685 struct btrfs_caching_control *caching_ctl;
5686 struct btrfs_block_group_cache *cache;
5688 down_write(&fs_info->commit_root_sem);
5690 list_for_each_entry_safe(caching_ctl, next,
5691 &fs_info->caching_block_groups, list) {
5692 cache = caching_ctl->block_group;
5693 if (block_group_cache_done(cache)) {
5694 cache->last_byte_to_unpin = (u64)-1;
5695 list_del_init(&caching_ctl->list);
5696 put_caching_control(caching_ctl);
5698 cache->last_byte_to_unpin = caching_ctl->progress;
5702 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5703 fs_info->pinned_extents = &fs_info->freed_extents[1];
5705 fs_info->pinned_extents = &fs_info->freed_extents[0];
5707 up_write(&fs_info->commit_root_sem);
5709 update_global_block_rsv(fs_info);
5712 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
5714 struct btrfs_fs_info *fs_info = root->fs_info;
5715 struct btrfs_block_group_cache *cache = NULL;
5716 struct btrfs_space_info *space_info;
5717 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv;
5721 while (start <= end) {
5724 start >= cache->key.objectid + cache->key.offset) {
5726 btrfs_put_block_group(cache);
5727 cache = btrfs_lookup_block_group(fs_info, start);
5728 BUG_ON(!cache); /* Logic error */
5731 len = cache->key.objectid + cache->key.offset - start;
5732 len = min(len, end + 1 - start);
5734 if (start < cache->last_byte_to_unpin) {
5735 len = min(len, cache->last_byte_to_unpin - start);
5736 btrfs_add_free_space(cache, start, len);
5740 space_info = cache->space_info;
5742 spin_lock(&space_info->lock);
5743 spin_lock(&cache->lock);
5744 cache->pinned -= len;
5745 space_info->bytes_pinned -= len;
5746 percpu_counter_add(&space_info->total_bytes_pinned, -len);
5748 space_info->bytes_readonly += len;
5751 spin_unlock(&cache->lock);
5752 if (!readonly && global_rsv->space_info == space_info) {
5753 spin_lock(&global_rsv->lock);
5754 if (!global_rsv->full) {
5755 len = min(len, global_rsv->size -
5756 global_rsv->reserved);
5757 global_rsv->reserved += len;
5758 space_info->bytes_may_use += len;
5759 if (global_rsv->reserved >= global_rsv->size)
5760 global_rsv->full = 1;
5762 spin_unlock(&global_rsv->lock);
5764 spin_unlock(&space_info->lock);
5768 btrfs_put_block_group(cache);
5772 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
5773 struct btrfs_root *root)
5775 struct btrfs_fs_info *fs_info = root->fs_info;
5776 struct extent_io_tree *unpin;
5784 if (fs_info->pinned_extents == &fs_info->freed_extents[0])
5785 unpin = &fs_info->freed_extents[1];
5787 unpin = &fs_info->freed_extents[0];
5790 ret = find_first_extent_bit(unpin, 0, &start, &end,
5791 EXTENT_DIRTY, NULL);
5795 if (btrfs_test_opt(root, DISCARD))
5796 ret = btrfs_discard_extent(root, start,
5797 end + 1 - start, NULL);
5799 clear_extent_dirty(unpin, start, end, GFP_NOFS);
5800 unpin_extent_range(root, start, end);
5807 static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes,
5808 u64 owner, u64 root_objectid)
5810 struct btrfs_space_info *space_info;
5813 if (owner < BTRFS_FIRST_FREE_OBJECTID) {
5814 if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID)
5815 flags = BTRFS_BLOCK_GROUP_SYSTEM;
5817 flags = BTRFS_BLOCK_GROUP_METADATA;
5819 flags = BTRFS_BLOCK_GROUP_DATA;
5822 space_info = __find_space_info(fs_info, flags);
5823 BUG_ON(!space_info); /* Logic bug */
5824 percpu_counter_add(&space_info->total_bytes_pinned, num_bytes);
5828 static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
5829 struct btrfs_root *root,
5830 u64 bytenr, u64 num_bytes, u64 parent,
5831 u64 root_objectid, u64 owner_objectid,
5832 u64 owner_offset, int refs_to_drop,
5833 struct btrfs_delayed_extent_op *extent_op,
5836 struct btrfs_key key;
5837 struct btrfs_path *path;
5838 struct btrfs_fs_info *info = root->fs_info;
5839 struct btrfs_root *extent_root = info->extent_root;
5840 struct extent_buffer *leaf;
5841 struct btrfs_extent_item *ei;
5842 struct btrfs_extent_inline_ref *iref;
5845 int extent_slot = 0;
5846 int found_extent = 0;
5851 enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL;
5852 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
5855 if (!info->quota_enabled || !is_fstree(root_objectid))
5858 path = btrfs_alloc_path();
5863 path->leave_spinning = 1;
5865 is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID;
5866 BUG_ON(!is_data && refs_to_drop != 1);
5869 skinny_metadata = 0;
5871 ret = lookup_extent_backref(trans, extent_root, path, &iref,
5872 bytenr, num_bytes, parent,
5873 root_objectid, owner_objectid,
5876 extent_slot = path->slots[0];
5877 while (extent_slot >= 0) {
5878 btrfs_item_key_to_cpu(path->nodes[0], &key,
5880 if (key.objectid != bytenr)
5882 if (key.type == BTRFS_EXTENT_ITEM_KEY &&
5883 key.offset == num_bytes) {
5887 if (key.type == BTRFS_METADATA_ITEM_KEY &&
5888 key.offset == owner_objectid) {
5892 if (path->slots[0] - extent_slot > 5)
5896 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5897 item_size = btrfs_item_size_nr(path->nodes[0], extent_slot);
5898 if (found_extent && item_size < sizeof(*ei))
5901 if (!found_extent) {
5903 ret = remove_extent_backref(trans, extent_root, path,
5905 is_data, &last_ref);
5907 btrfs_abort_transaction(trans, extent_root, ret);
5910 btrfs_release_path(path);
5911 path->leave_spinning = 1;
5913 key.objectid = bytenr;
5914 key.type = BTRFS_EXTENT_ITEM_KEY;
5915 key.offset = num_bytes;
5917 if (!is_data && skinny_metadata) {
5918 key.type = BTRFS_METADATA_ITEM_KEY;
5919 key.offset = owner_objectid;
5922 ret = btrfs_search_slot(trans, extent_root,
5924 if (ret > 0 && skinny_metadata && path->slots[0]) {
5926 * Couldn't find our skinny metadata item,
5927 * see if we have ye olde extent item.
5930 btrfs_item_key_to_cpu(path->nodes[0], &key,
5932 if (key.objectid == bytenr &&
5933 key.type == BTRFS_EXTENT_ITEM_KEY &&
5934 key.offset == num_bytes)
5938 if (ret > 0 && skinny_metadata) {
5939 skinny_metadata = false;
5940 key.objectid = bytenr;
5941 key.type = BTRFS_EXTENT_ITEM_KEY;
5942 key.offset = num_bytes;
5943 btrfs_release_path(path);
5944 ret = btrfs_search_slot(trans, extent_root,
5949 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5952 btrfs_print_leaf(extent_root,
5956 btrfs_abort_transaction(trans, extent_root, ret);
5959 extent_slot = path->slots[0];
5961 } else if (WARN_ON(ret == -ENOENT)) {
5962 btrfs_print_leaf(extent_root, path->nodes[0]);
5964 "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu",
5965 bytenr, parent, root_objectid, owner_objectid,
5967 btrfs_abort_transaction(trans, extent_root, ret);
5970 btrfs_abort_transaction(trans, extent_root, ret);
5974 leaf = path->nodes[0];
5975 item_size = btrfs_item_size_nr(leaf, extent_slot);
5976 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
5977 if (item_size < sizeof(*ei)) {
5978 BUG_ON(found_extent || extent_slot != path->slots[0]);
5979 ret = convert_extent_item_v0(trans, extent_root, path,
5982 btrfs_abort_transaction(trans, extent_root, ret);
5986 btrfs_release_path(path);
5987 path->leave_spinning = 1;
5989 key.objectid = bytenr;
5990 key.type = BTRFS_EXTENT_ITEM_KEY;
5991 key.offset = num_bytes;
5993 ret = btrfs_search_slot(trans, extent_root, &key, path,
5996 btrfs_err(info, "umm, got %d back from search, was looking for %llu",
5998 btrfs_print_leaf(extent_root, path->nodes[0]);
6001 btrfs_abort_transaction(trans, extent_root, ret);
6005 extent_slot = path->slots[0];
6006 leaf = path->nodes[0];
6007 item_size = btrfs_item_size_nr(leaf, extent_slot);
6010 BUG_ON(item_size < sizeof(*ei));
6011 ei = btrfs_item_ptr(leaf, extent_slot,
6012 struct btrfs_extent_item);
6013 if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID &&
6014 key.type == BTRFS_EXTENT_ITEM_KEY) {
6015 struct btrfs_tree_block_info *bi;
6016 BUG_ON(item_size < sizeof(*ei) + sizeof(*bi));
6017 bi = (struct btrfs_tree_block_info *)(ei + 1);
6018 WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi));
6021 refs = btrfs_extent_refs(leaf, ei);
6022 if (refs < refs_to_drop) {
6023 btrfs_err(info, "trying to drop %d refs but we only have %Lu "
6024 "for bytenr %Lu", refs_to_drop, refs, bytenr);
6026 btrfs_abort_transaction(trans, extent_root, ret);
6029 refs -= refs_to_drop;
6032 type = BTRFS_QGROUP_OPER_SUB_SHARED;
6034 __run_delayed_extent_op(extent_op, leaf, ei);
6036 * In the case of inline back ref, reference count will
6037 * be updated by remove_extent_backref
6040 BUG_ON(!found_extent);
6042 btrfs_set_extent_refs(leaf, ei, refs);
6043 btrfs_mark_buffer_dirty(leaf);
6046 ret = remove_extent_backref(trans, extent_root, path,
6048 is_data, &last_ref);
6050 btrfs_abort_transaction(trans, extent_root, ret);
6054 add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid,
6058 BUG_ON(is_data && refs_to_drop !=
6059 extent_data_ref_count(root, path, iref));
6061 BUG_ON(path->slots[0] != extent_slot);
6063 BUG_ON(path->slots[0] != extent_slot + 1);
6064 path->slots[0] = extent_slot;
6070 ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
6073 btrfs_abort_transaction(trans, extent_root, ret);
6076 btrfs_release_path(path);
6079 ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
6081 btrfs_abort_transaction(trans, extent_root, ret);
6086 ret = update_block_group(root, bytenr, num_bytes, 0);
6088 btrfs_abort_transaction(trans, extent_root, ret);
6092 btrfs_release_path(path);
6094 /* Deal with the quota accounting */
6095 if (!ret && last_ref && !no_quota) {
6098 if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID &&
6099 type == BTRFS_QGROUP_OPER_SUB_SHARED)
6102 ret = btrfs_qgroup_record_ref(trans, info, root_objectid,
6103 bytenr, num_bytes, type,
6107 btrfs_free_path(path);
6112 * when we free an block, it is possible (and likely) that we free the last
6113 * delayed ref for that extent as well. This searches the delayed ref tree for
6114 * a given extent, and if there are no other delayed refs to be processed, it
6115 * removes it from the tree.
6117 static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
6118 struct btrfs_root *root, u64 bytenr)
6120 struct btrfs_delayed_ref_head *head;
6121 struct btrfs_delayed_ref_root *delayed_refs;
6124 delayed_refs = &trans->transaction->delayed_refs;
6125 spin_lock(&delayed_refs->lock);
6126 head = btrfs_find_delayed_ref_head(trans, bytenr);
6128 goto out_delayed_unlock;
6130 spin_lock(&head->lock);
6131 if (rb_first(&head->ref_root))
6134 if (head->extent_op) {
6135 if (!head->must_insert_reserved)
6137 btrfs_free_delayed_extent_op(head->extent_op);
6138 head->extent_op = NULL;
6142 * waiting for the lock here would deadlock. If someone else has it
6143 * locked they are already in the process of dropping it anyway
6145 if (!mutex_trylock(&head->mutex))
6149 * at this point we have a head with no other entries. Go
6150 * ahead and process it.
6152 head->node.in_tree = 0;
6153 rb_erase(&head->href_node, &delayed_refs->href_root);
6155 atomic_dec(&delayed_refs->num_entries);
6158 * we don't take a ref on the node because we're removing it from the
6159 * tree, so we just steal the ref the tree was holding.
6161 delayed_refs->num_heads--;
6162 if (head->processing == 0)
6163 delayed_refs->num_heads_ready--;
6164 head->processing = 0;
6165 spin_unlock(&head->lock);
6166 spin_unlock(&delayed_refs->lock);
6168 BUG_ON(head->extent_op);
6169 if (head->must_insert_reserved)
6172 mutex_unlock(&head->mutex);
6173 btrfs_put_delayed_ref(&head->node);
6176 spin_unlock(&head->lock);
6179 spin_unlock(&delayed_refs->lock);
6183 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
6184 struct btrfs_root *root,
6185 struct extent_buffer *buf,
6186 u64 parent, int last_ref)
6188 struct btrfs_block_group_cache *cache = NULL;
6192 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6193 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
6194 buf->start, buf->len,
6195 parent, root->root_key.objectid,
6196 btrfs_header_level(buf),
6197 BTRFS_DROP_DELAYED_REF, NULL, 0);
6198 BUG_ON(ret); /* -ENOMEM */
6204 cache = btrfs_lookup_block_group(root->fs_info, buf->start);
6206 if (btrfs_header_generation(buf) == trans->transid) {
6207 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
6208 ret = check_ref_cleanup(trans, root, buf->start);
6213 if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
6214 pin_down_extent(root, cache, buf->start, buf->len, 1);
6218 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags));
6220 btrfs_add_free_space(cache, buf->start, buf->len);
6221 btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE, 0);
6222 trace_btrfs_reserved_extent_free(root, buf->start, buf->len);
6227 add_pinned_bytes(root->fs_info, buf->len,
6228 btrfs_header_level(buf),
6229 root->root_key.objectid);
6232 * Deleting the buffer, clear the corrupt flag since it doesn't matter
6235 clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags);
6236 btrfs_put_block_group(cache);
6239 /* Can return -ENOMEM */
6240 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
6241 u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
6242 u64 owner, u64 offset, int no_quota)
6245 struct btrfs_fs_info *fs_info = root->fs_info;
6247 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
6248 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
6251 add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid);
6254 * tree log blocks never actually go into the extent allocation
6255 * tree, just update pinning info and exit early.
6257 if (root_objectid == BTRFS_TREE_LOG_OBJECTID) {
6258 WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID);
6259 /* unlocks the pinned mutex */
6260 btrfs_pin_extent(root, bytenr, num_bytes, 1);
6262 } else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
6263 ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
6265 parent, root_objectid, (int)owner,
6266 BTRFS_DROP_DELAYED_REF, NULL, no_quota);
6268 ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
6270 parent, root_objectid, owner,
6271 offset, BTRFS_DROP_DELAYED_REF,
6278 * when we wait for progress in the block group caching, its because
6279 * our allocation attempt failed at least once. So, we must sleep
6280 * and let some progress happen before we try again.
6282 * This function will sleep at least once waiting for new free space to
6283 * show up, and then it will check the block group free space numbers
6284 * for our min num_bytes. Another option is to have it go ahead
6285 * and look in the rbtree for a free extent of a given size, but this
6288 * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using
6289 * any of the information in this block group.
6291 static noinline void
6292 wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
6295 struct btrfs_caching_control *caching_ctl;
6297 caching_ctl = get_caching_control(cache);
6301 wait_event(caching_ctl->wait, block_group_cache_done(cache) ||
6302 (cache->free_space_ctl->free_space >= num_bytes));
6304 put_caching_control(caching_ctl);
6308 wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
6310 struct btrfs_caching_control *caching_ctl;
6313 caching_ctl = get_caching_control(cache);
6315 return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0;
6317 wait_event(caching_ctl->wait, block_group_cache_done(cache));
6318 if (cache->cached == BTRFS_CACHE_ERROR)
6320 put_caching_control(caching_ctl);
6324 int __get_raid_index(u64 flags)
6326 if (flags & BTRFS_BLOCK_GROUP_RAID10)
6327 return BTRFS_RAID_RAID10;
6328 else if (flags & BTRFS_BLOCK_GROUP_RAID1)
6329 return BTRFS_RAID_RAID1;
6330 else if (flags & BTRFS_BLOCK_GROUP_DUP)
6331 return BTRFS_RAID_DUP;
6332 else if (flags & BTRFS_BLOCK_GROUP_RAID0)
6333 return BTRFS_RAID_RAID0;
6334 else if (flags & BTRFS_BLOCK_GROUP_RAID5)
6335 return BTRFS_RAID_RAID5;
6336 else if (flags & BTRFS_BLOCK_GROUP_RAID6)
6337 return BTRFS_RAID_RAID6;
6339 return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */
6342 int get_block_group_index(struct btrfs_block_group_cache *cache)
6344 return __get_raid_index(cache->flags);
6347 static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = {
6348 [BTRFS_RAID_RAID10] = "raid10",
6349 [BTRFS_RAID_RAID1] = "raid1",
6350 [BTRFS_RAID_DUP] = "dup",
6351 [BTRFS_RAID_RAID0] = "raid0",
6352 [BTRFS_RAID_SINGLE] = "single",
6353 [BTRFS_RAID_RAID5] = "raid5",
6354 [BTRFS_RAID_RAID6] = "raid6",
6357 static const char *get_raid_name(enum btrfs_raid_types type)
6359 if (type >= BTRFS_NR_RAID_TYPES)
6362 return btrfs_raid_type_names[type];
6365 enum btrfs_loop_type {
6366 LOOP_CACHING_NOWAIT = 0,
6367 LOOP_CACHING_WAIT = 1,
6368 LOOP_ALLOC_CHUNK = 2,
6369 LOOP_NO_EMPTY_SIZE = 3,
6373 btrfs_lock_block_group(struct btrfs_block_group_cache *cache,
6377 down_read(&cache->data_rwsem);
6381 btrfs_grab_block_group(struct btrfs_block_group_cache *cache,
6384 btrfs_get_block_group(cache);
6386 down_read(&cache->data_rwsem);
6389 static struct btrfs_block_group_cache *
6390 btrfs_lock_cluster(struct btrfs_block_group_cache *block_group,
6391 struct btrfs_free_cluster *cluster,
6394 struct btrfs_block_group_cache *used_bg;
6395 bool locked = false;
6397 spin_lock(&cluster->refill_lock);
6399 if (used_bg == cluster->block_group)
6402 up_read(&used_bg->data_rwsem);
6403 btrfs_put_block_group(used_bg);
6406 used_bg = cluster->block_group;
6410 if (used_bg == block_group)
6413 btrfs_get_block_group(used_bg);
6418 if (down_read_trylock(&used_bg->data_rwsem))
6421 spin_unlock(&cluster->refill_lock);
6422 down_read(&used_bg->data_rwsem);
6428 btrfs_release_block_group(struct btrfs_block_group_cache *cache,
6432 up_read(&cache->data_rwsem);
6433 btrfs_put_block_group(cache);
6437 * walks the btree of allocated extents and find a hole of a given size.
6438 * The key ins is changed to record the hole:
6439 * ins->objectid == start position
6440 * ins->flags = BTRFS_EXTENT_ITEM_KEY
6441 * ins->offset == the size of the hole.
6442 * Any available blocks before search_start are skipped.
6444 * If there is no suitable free space, we will record the max size of
6445 * the free space extent currently.
6447 static noinline int find_free_extent(struct btrfs_root *orig_root,
6448 u64 num_bytes, u64 empty_size,
6449 u64 hint_byte, struct btrfs_key *ins,
6450 u64 flags, int delalloc)
6453 struct btrfs_root *root = orig_root->fs_info->extent_root;
6454 struct btrfs_free_cluster *last_ptr = NULL;
6455 struct btrfs_block_group_cache *block_group = NULL;
6456 u64 search_start = 0;
6457 u64 max_extent_size = 0;
6458 int empty_cluster = 2 * 1024 * 1024;
6459 struct btrfs_space_info *space_info;
6461 int index = __get_raid_index(flags);
6462 int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ?
6463 RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC;
6464 bool failed_cluster_refill = false;
6465 bool failed_alloc = false;
6466 bool use_cluster = true;
6467 bool have_caching_bg = false;
6469 WARN_ON(num_bytes < root->sectorsize);
6470 ins->type = BTRFS_EXTENT_ITEM_KEY;
6474 trace_find_free_extent(orig_root, num_bytes, empty_size, flags);
6476 space_info = __find_space_info(root->fs_info, flags);
6478 btrfs_err(root->fs_info, "No space info for %llu", flags);
6483 * If the space info is for both data and metadata it means we have a
6484 * small filesystem and we can't use the clustering stuff.
6486 if (btrfs_mixed_space_info(space_info))
6487 use_cluster = false;
6489 if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) {
6490 last_ptr = &root->fs_info->meta_alloc_cluster;
6491 if (!btrfs_test_opt(root, SSD))
6492 empty_cluster = 64 * 1024;
6495 if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster &&
6496 btrfs_test_opt(root, SSD)) {
6497 last_ptr = &root->fs_info->data_alloc_cluster;
6501 spin_lock(&last_ptr->lock);
6502 if (last_ptr->block_group)
6503 hint_byte = last_ptr->window_start;
6504 spin_unlock(&last_ptr->lock);
6507 search_start = max(search_start, first_logical_byte(root, 0));
6508 search_start = max(search_start, hint_byte);
6513 if (search_start == hint_byte) {
6514 block_group = btrfs_lookup_block_group(root->fs_info,
6517 * we don't want to use the block group if it doesn't match our
6518 * allocation bits, or if its not cached.
6520 * However if we are re-searching with an ideal block group
6521 * picked out then we don't care that the block group is cached.
6523 if (block_group && block_group_bits(block_group, flags) &&
6524 block_group->cached != BTRFS_CACHE_NO) {
6525 down_read(&space_info->groups_sem);
6526 if (list_empty(&block_group->list) ||
6529 * someone is removing this block group,
6530 * we can't jump into the have_block_group
6531 * target because our list pointers are not
6534 btrfs_put_block_group(block_group);
6535 up_read(&space_info->groups_sem);
6537 index = get_block_group_index(block_group);
6538 btrfs_lock_block_group(block_group, delalloc);
6539 goto have_block_group;
6541 } else if (block_group) {
6542 btrfs_put_block_group(block_group);
6546 have_caching_bg = false;
6547 down_read(&space_info->groups_sem);
6548 list_for_each_entry(block_group, &space_info->block_groups[index],
6553 btrfs_grab_block_group(block_group, delalloc);
6554 search_start = block_group->key.objectid;
6557 * this can happen if we end up cycling through all the
6558 * raid types, but we want to make sure we only allocate
6559 * for the proper type.
6561 if (!block_group_bits(block_group, flags)) {
6562 u64 extra = BTRFS_BLOCK_GROUP_DUP |
6563 BTRFS_BLOCK_GROUP_RAID1 |
6564 BTRFS_BLOCK_GROUP_RAID5 |
6565 BTRFS_BLOCK_GROUP_RAID6 |
6566 BTRFS_BLOCK_GROUP_RAID10;
6569 * if they asked for extra copies and this block group
6570 * doesn't provide them, bail. This does allow us to
6571 * fill raid0 from raid1.
6573 if ((flags & extra) && !(block_group->flags & extra))
6578 cached = block_group_cache_done(block_group);
6579 if (unlikely(!cached)) {
6580 ret = cache_block_group(block_group, 0);
6585 if (unlikely(block_group->cached == BTRFS_CACHE_ERROR))
6587 if (unlikely(block_group->ro))
6591 * Ok we want to try and use the cluster allocator, so
6595 struct btrfs_block_group_cache *used_block_group;
6596 unsigned long aligned_cluster;
6598 * the refill lock keeps out other
6599 * people trying to start a new cluster
6601 used_block_group = btrfs_lock_cluster(block_group,
6604 if (!used_block_group)
6605 goto refill_cluster;
6607 if (used_block_group != block_group &&
6608 (used_block_group->ro ||
6609 !block_group_bits(used_block_group, flags)))
6610 goto release_cluster;
6612 offset = btrfs_alloc_from_cluster(used_block_group,
6615 used_block_group->key.objectid,
6618 /* we have a block, we're done */
6619 spin_unlock(&last_ptr->refill_lock);
6620 trace_btrfs_reserve_extent_cluster(root,
6622 search_start, num_bytes);
6623 if (used_block_group != block_group) {
6624 btrfs_release_block_group(block_group,
6626 block_group = used_block_group;
6631 WARN_ON(last_ptr->block_group != used_block_group);
6633 /* If we are on LOOP_NO_EMPTY_SIZE, we can't
6634 * set up a new clusters, so lets just skip it
6635 * and let the allocator find whatever block
6636 * it can find. If we reach this point, we
6637 * will have tried the cluster allocator
6638 * plenty of times and not have found
6639 * anything, so we are likely way too
6640 * fragmented for the clustering stuff to find
6643 * However, if the cluster is taken from the
6644 * current block group, release the cluster
6645 * first, so that we stand a better chance of
6646 * succeeding in the unclustered
6648 if (loop >= LOOP_NO_EMPTY_SIZE &&
6649 used_block_group != block_group) {
6650 spin_unlock(&last_ptr->refill_lock);
6651 btrfs_release_block_group(used_block_group,
6653 goto unclustered_alloc;
6657 * this cluster didn't work out, free it and
6660 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6662 if (used_block_group != block_group)
6663 btrfs_release_block_group(used_block_group,
6666 if (loop >= LOOP_NO_EMPTY_SIZE) {
6667 spin_unlock(&last_ptr->refill_lock);
6668 goto unclustered_alloc;
6671 aligned_cluster = max_t(unsigned long,
6672 empty_cluster + empty_size,
6673 block_group->full_stripe_len);
6675 /* allocate a cluster in this block group */
6676 ret = btrfs_find_space_cluster(root, block_group,
6677 last_ptr, search_start,
6682 * now pull our allocation out of this
6685 offset = btrfs_alloc_from_cluster(block_group,
6691 /* we found one, proceed */
6692 spin_unlock(&last_ptr->refill_lock);
6693 trace_btrfs_reserve_extent_cluster(root,
6694 block_group, search_start,
6698 } else if (!cached && loop > LOOP_CACHING_NOWAIT
6699 && !failed_cluster_refill) {
6700 spin_unlock(&last_ptr->refill_lock);
6702 failed_cluster_refill = true;
6703 wait_block_group_cache_progress(block_group,
6704 num_bytes + empty_cluster + empty_size);
6705 goto have_block_group;
6709 * at this point we either didn't find a cluster
6710 * or we weren't able to allocate a block from our
6711 * cluster. Free the cluster we've been trying
6712 * to use, and go to the next block group
6714 btrfs_return_cluster_to_free_space(NULL, last_ptr);
6715 spin_unlock(&last_ptr->refill_lock);
6720 spin_lock(&block_group->free_space_ctl->tree_lock);
6722 block_group->free_space_ctl->free_space <
6723 num_bytes + empty_cluster + empty_size) {
6724 if (block_group->free_space_ctl->free_space >
6727 block_group->free_space_ctl->free_space;
6728 spin_unlock(&block_group->free_space_ctl->tree_lock);
6731 spin_unlock(&block_group->free_space_ctl->tree_lock);
6733 offset = btrfs_find_space_for_alloc(block_group, search_start,
6734 num_bytes, empty_size,
6737 * If we didn't find a chunk, and we haven't failed on this
6738 * block group before, and this block group is in the middle of
6739 * caching and we are ok with waiting, then go ahead and wait
6740 * for progress to be made, and set failed_alloc to true.
6742 * If failed_alloc is true then we've already waited on this
6743 * block group once and should move on to the next block group.
6745 if (!offset && !failed_alloc && !cached &&
6746 loop > LOOP_CACHING_NOWAIT) {
6747 wait_block_group_cache_progress(block_group,
6748 num_bytes + empty_size);
6749 failed_alloc = true;
6750 goto have_block_group;
6751 } else if (!offset) {
6753 have_caching_bg = true;
6757 search_start = ALIGN(offset, root->stripesize);
6759 /* move on to the next group */
6760 if (search_start + num_bytes >
6761 block_group->key.objectid + block_group->key.offset) {
6762 btrfs_add_free_space(block_group, offset, num_bytes);
6766 if (offset < search_start)
6767 btrfs_add_free_space(block_group, offset,
6768 search_start - offset);
6769 BUG_ON(offset > search_start);
6771 ret = btrfs_update_reserved_bytes(block_group, num_bytes,
6772 alloc_type, delalloc);
6773 if (ret == -EAGAIN) {
6774 btrfs_add_free_space(block_group, offset, num_bytes);
6778 /* we are all good, lets return */
6779 ins->objectid = search_start;
6780 ins->offset = num_bytes;
6782 trace_btrfs_reserve_extent(orig_root, block_group,
6783 search_start, num_bytes);
6784 btrfs_release_block_group(block_group, delalloc);
6787 failed_cluster_refill = false;
6788 failed_alloc = false;
6789 BUG_ON(index != get_block_group_index(block_group));
6790 btrfs_release_block_group(block_group, delalloc);
6792 up_read(&space_info->groups_sem);
6794 if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg)
6797 if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
6801 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
6802 * caching kthreads as we move along
6803 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
6804 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
6805 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
6808 if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
6811 if (loop == LOOP_ALLOC_CHUNK) {
6812 struct btrfs_trans_handle *trans;
6815 trans = current->journal_info;
6819 trans = btrfs_join_transaction(root);
6821 if (IS_ERR(trans)) {
6822 ret = PTR_ERR(trans);
6826 ret = do_chunk_alloc(trans, root, flags,
6829 * Do not bail out on ENOSPC since we
6830 * can do more things.
6832 if (ret < 0 && ret != -ENOSPC)
6833 btrfs_abort_transaction(trans,
6838 btrfs_end_transaction(trans, root);
6843 if (loop == LOOP_NO_EMPTY_SIZE) {
6849 } else if (!ins->objectid) {
6851 } else if (ins->objectid) {
6856 ins->offset = max_extent_size;
6860 static void dump_space_info(struct btrfs_space_info *info, u64 bytes,
6861 int dump_block_groups)
6863 struct btrfs_block_group_cache *cache;
6866 spin_lock(&info->lock);
6867 printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n",
6869 info->total_bytes - info->bytes_used - info->bytes_pinned -
6870 info->bytes_reserved - info->bytes_readonly,
6871 (info->full) ? "" : "not ");
6872 printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, "
6873 "reserved=%llu, may_use=%llu, readonly=%llu\n",
6874 info->total_bytes, info->bytes_used, info->bytes_pinned,
6875 info->bytes_reserved, info->bytes_may_use,
6876 info->bytes_readonly);
6877 spin_unlock(&info->lock);
6879 if (!dump_block_groups)
6882 down_read(&info->groups_sem);
6884 list_for_each_entry(cache, &info->block_groups[index], list) {
6885 spin_lock(&cache->lock);
6886 printk(KERN_INFO "BTRFS: "
6887 "block group %llu has %llu bytes, "
6888 "%llu used %llu pinned %llu reserved %s\n",
6889 cache->key.objectid, cache->key.offset,
6890 btrfs_block_group_used(&cache->item), cache->pinned,
6891 cache->reserved, cache->ro ? "[readonly]" : "");
6892 btrfs_dump_free_space(cache, bytes);
6893 spin_unlock(&cache->lock);
6895 if (++index < BTRFS_NR_RAID_TYPES)
6897 up_read(&info->groups_sem);
6900 int btrfs_reserve_extent(struct btrfs_root *root,
6901 u64 num_bytes, u64 min_alloc_size,
6902 u64 empty_size, u64 hint_byte,
6903 struct btrfs_key *ins, int is_data, int delalloc)
6905 bool final_tried = false;
6909 flags = btrfs_get_alloc_profile(root, is_data);
6911 WARN_ON(num_bytes < root->sectorsize);
6912 ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins,
6915 if (ret == -ENOSPC) {
6916 if (!final_tried && ins->offset) {
6917 num_bytes = min(num_bytes >> 1, ins->offset);
6918 num_bytes = round_down(num_bytes, root->sectorsize);
6919 num_bytes = max(num_bytes, min_alloc_size);
6920 if (num_bytes == min_alloc_size)
6923 } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
6924 struct btrfs_space_info *sinfo;
6926 sinfo = __find_space_info(root->fs_info, flags);
6927 btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu",
6930 dump_space_info(sinfo, num_bytes, 1);
6937 static int __btrfs_free_reserved_extent(struct btrfs_root *root,
6939 int pin, int delalloc)
6941 struct btrfs_block_group_cache *cache;
6944 cache = btrfs_lookup_block_group(root->fs_info, start);
6946 btrfs_err(root->fs_info, "Unable to find block group for %llu",
6951 if (btrfs_test_opt(root, DISCARD))
6952 ret = btrfs_discard_extent(root, start, len, NULL);
6955 pin_down_extent(root, cache, start, len, 1);
6957 btrfs_add_free_space(cache, start, len);
6958 btrfs_update_reserved_bytes(cache, len, RESERVE_FREE, delalloc);
6960 btrfs_put_block_group(cache);
6962 trace_btrfs_reserved_extent_free(root, start, len);
6967 int btrfs_free_reserved_extent(struct btrfs_root *root,
6968 u64 start, u64 len, int delalloc)
6970 return __btrfs_free_reserved_extent(root, start, len, 0, delalloc);
6973 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
6976 return __btrfs_free_reserved_extent(root, start, len, 1, 0);
6979 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
6980 struct btrfs_root *root,
6981 u64 parent, u64 root_objectid,
6982 u64 flags, u64 owner, u64 offset,
6983 struct btrfs_key *ins, int ref_mod)
6986 struct btrfs_fs_info *fs_info = root->fs_info;
6987 struct btrfs_extent_item *extent_item;
6988 struct btrfs_extent_inline_ref *iref;
6989 struct btrfs_path *path;
6990 struct extent_buffer *leaf;
6995 type = BTRFS_SHARED_DATA_REF_KEY;
6997 type = BTRFS_EXTENT_DATA_REF_KEY;
6999 size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type);
7001 path = btrfs_alloc_path();
7005 path->leave_spinning = 1;
7006 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7009 btrfs_free_path(path);
7013 leaf = path->nodes[0];
7014 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7015 struct btrfs_extent_item);
7016 btrfs_set_extent_refs(leaf, extent_item, ref_mod);
7017 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7018 btrfs_set_extent_flags(leaf, extent_item,
7019 flags | BTRFS_EXTENT_FLAG_DATA);
7021 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7022 btrfs_set_extent_inline_ref_type(leaf, iref, type);
7024 struct btrfs_shared_data_ref *ref;
7025 ref = (struct btrfs_shared_data_ref *)(iref + 1);
7026 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7027 btrfs_set_shared_data_ref_count(leaf, ref, ref_mod);
7029 struct btrfs_extent_data_ref *ref;
7030 ref = (struct btrfs_extent_data_ref *)(&iref->offset);
7031 btrfs_set_extent_data_ref_root(leaf, ref, root_objectid);
7032 btrfs_set_extent_data_ref_objectid(leaf, ref, owner);
7033 btrfs_set_extent_data_ref_offset(leaf, ref, offset);
7034 btrfs_set_extent_data_ref_count(leaf, ref, ref_mod);
7037 btrfs_mark_buffer_dirty(path->nodes[0]);
7038 btrfs_free_path(path);
7040 /* Always set parent to 0 here since its exclusive anyway. */
7041 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7042 ins->objectid, ins->offset,
7043 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7047 ret = update_block_group(root, ins->objectid, ins->offset, 1);
7048 if (ret) { /* -ENOENT, logic error */
7049 btrfs_err(fs_info, "update block group failed for %llu %llu",
7050 ins->objectid, ins->offset);
7053 trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset);
7057 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
7058 struct btrfs_root *root,
7059 u64 parent, u64 root_objectid,
7060 u64 flags, struct btrfs_disk_key *key,
7061 int level, struct btrfs_key *ins,
7065 struct btrfs_fs_info *fs_info = root->fs_info;
7066 struct btrfs_extent_item *extent_item;
7067 struct btrfs_tree_block_info *block_info;
7068 struct btrfs_extent_inline_ref *iref;
7069 struct btrfs_path *path;
7070 struct extent_buffer *leaf;
7071 u32 size = sizeof(*extent_item) + sizeof(*iref);
7072 u64 num_bytes = ins->offset;
7073 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7076 if (!skinny_metadata)
7077 size += sizeof(*block_info);
7079 path = btrfs_alloc_path();
7081 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7086 path->leave_spinning = 1;
7087 ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
7090 btrfs_free_and_pin_reserved_extent(root, ins->objectid,
7092 btrfs_free_path(path);
7096 leaf = path->nodes[0];
7097 extent_item = btrfs_item_ptr(leaf, path->slots[0],
7098 struct btrfs_extent_item);
7099 btrfs_set_extent_refs(leaf, extent_item, 1);
7100 btrfs_set_extent_generation(leaf, extent_item, trans->transid);
7101 btrfs_set_extent_flags(leaf, extent_item,
7102 flags | BTRFS_EXTENT_FLAG_TREE_BLOCK);
7104 if (skinny_metadata) {
7105 iref = (struct btrfs_extent_inline_ref *)(extent_item + 1);
7106 num_bytes = root->nodesize;
7108 block_info = (struct btrfs_tree_block_info *)(extent_item + 1);
7109 btrfs_set_tree_block_key(leaf, block_info, key);
7110 btrfs_set_tree_block_level(leaf, block_info, level);
7111 iref = (struct btrfs_extent_inline_ref *)(block_info + 1);
7115 BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
7116 btrfs_set_extent_inline_ref_type(leaf, iref,
7117 BTRFS_SHARED_BLOCK_REF_KEY);
7118 btrfs_set_extent_inline_ref_offset(leaf, iref, parent);
7120 btrfs_set_extent_inline_ref_type(leaf, iref,
7121 BTRFS_TREE_BLOCK_REF_KEY);
7122 btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
7125 btrfs_mark_buffer_dirty(leaf);
7126 btrfs_free_path(path);
7129 ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid,
7130 ins->objectid, num_bytes,
7131 BTRFS_QGROUP_OPER_ADD_EXCL, 0);
7136 ret = update_block_group(root, ins->objectid, root->nodesize, 1);
7137 if (ret) { /* -ENOENT, logic error */
7138 btrfs_err(fs_info, "update block group failed for %llu %llu",
7139 ins->objectid, ins->offset);
7143 trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->nodesize);
7147 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
7148 struct btrfs_root *root,
7149 u64 root_objectid, u64 owner,
7150 u64 offset, struct btrfs_key *ins)
7154 BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
7156 ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
7158 root_objectid, owner, offset,
7159 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
7164 * this is used by the tree logging recovery code. It records that
7165 * an extent has been allocated and makes sure to clear the free
7166 * space cache bits as well
7168 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
7169 struct btrfs_root *root,
7170 u64 root_objectid, u64 owner, u64 offset,
7171 struct btrfs_key *ins)
7174 struct btrfs_block_group_cache *block_group;
7177 * Mixed block groups will exclude before processing the log so we only
7178 * need to do the exlude dance if this fs isn't mixed.
7180 if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) {
7181 ret = __exclude_logged_extent(root, ins->objectid, ins->offset);
7186 block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
7190 ret = btrfs_update_reserved_bytes(block_group, ins->offset,
7191 RESERVE_ALLOC_NO_ACCOUNT, 0);
7192 BUG_ON(ret); /* logic error */
7193 ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
7194 0, owner, offset, ins, 1);
7195 btrfs_put_block_group(block_group);
7199 static struct extent_buffer *
7200 btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root,
7201 u64 bytenr, u32 blocksize, int level)
7203 struct extent_buffer *buf;
7205 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
7207 return ERR_PTR(-ENOMEM);
7208 btrfs_set_header_generation(buf, trans->transid);
7209 btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
7210 btrfs_tree_lock(buf);
7211 clean_tree_block(trans, root, buf);
7212 clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
7214 btrfs_set_lock_blocking(buf);
7215 btrfs_set_buffer_uptodate(buf);
7217 if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) {
7219 * we allow two log transactions at a time, use different
7220 * EXENT bit to differentiate dirty pages.
7222 if (root->log_transid % 2 == 0)
7223 set_extent_dirty(&root->dirty_log_pages, buf->start,
7224 buf->start + buf->len - 1, GFP_NOFS);
7226 set_extent_new(&root->dirty_log_pages, buf->start,
7227 buf->start + buf->len - 1, GFP_NOFS);
7229 set_extent_dirty(&trans->transaction->dirty_pages, buf->start,
7230 buf->start + buf->len - 1, GFP_NOFS);
7232 trans->blocks_used++;
7233 /* this returns a buffer locked for blocking */
7237 static struct btrfs_block_rsv *
7238 use_block_rsv(struct btrfs_trans_handle *trans,
7239 struct btrfs_root *root, u32 blocksize)
7241 struct btrfs_block_rsv *block_rsv;
7242 struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv;
7244 bool global_updated = false;
7246 block_rsv = get_block_rsv(trans, root);
7248 if (unlikely(block_rsv->size == 0))
7251 ret = block_rsv_use_bytes(block_rsv, blocksize);
7255 if (block_rsv->failfast)
7256 return ERR_PTR(ret);
7258 if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) {
7259 global_updated = true;
7260 update_global_block_rsv(root->fs_info);
7264 if (btrfs_test_opt(root, ENOSPC_DEBUG)) {
7265 static DEFINE_RATELIMIT_STATE(_rs,
7266 DEFAULT_RATELIMIT_INTERVAL * 10,
7267 /*DEFAULT_RATELIMIT_BURST*/ 1);
7268 if (__ratelimit(&_rs))
7270 "BTRFS: block rsv returned %d\n", ret);
7273 ret = reserve_metadata_bytes(root, block_rsv, blocksize,
7274 BTRFS_RESERVE_NO_FLUSH);
7278 * If we couldn't reserve metadata bytes try and use some from
7279 * the global reserve if its space type is the same as the global
7282 if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL &&
7283 block_rsv->space_info == global_rsv->space_info) {
7284 ret = block_rsv_use_bytes(global_rsv, blocksize);
7288 return ERR_PTR(ret);
7291 static void unuse_block_rsv(struct btrfs_fs_info *fs_info,
7292 struct btrfs_block_rsv *block_rsv, u32 blocksize)
7294 block_rsv_add_bytes(block_rsv, blocksize, 0);
7295 block_rsv_release_bytes(fs_info, block_rsv, NULL, 0);
7299 * finds a free extent and does all the dirty work required for allocation
7300 * returns the key for the extent through ins, and a tree buffer for
7301 * the first block of the extent through buf.
7303 * returns the tree buffer or NULL.
7305 struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
7306 struct btrfs_root *root, u32 blocksize,
7307 u64 parent, u64 root_objectid,
7308 struct btrfs_disk_key *key, int level,
7309 u64 hint, u64 empty_size)
7311 struct btrfs_key ins;
7312 struct btrfs_block_rsv *block_rsv;
7313 struct extent_buffer *buf;
7316 bool skinny_metadata = btrfs_fs_incompat(root->fs_info,
7319 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
7320 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state))) {
7321 buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr,
7324 root->alloc_bytenr += blocksize;
7328 block_rsv = use_block_rsv(trans, root, blocksize);
7329 if (IS_ERR(block_rsv))
7330 return ERR_CAST(block_rsv);
7332 ret = btrfs_reserve_extent(root, blocksize, blocksize,
7333 empty_size, hint, &ins, 0, 0);
7335 unuse_block_rsv(root->fs_info, block_rsv, blocksize);
7336 return ERR_PTR(ret);
7339 buf = btrfs_init_new_buffer(trans, root, ins.objectid,
7341 BUG_ON(IS_ERR(buf)); /* -ENOMEM */
7343 if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
7345 parent = ins.objectid;
7346 flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
7350 if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
7351 struct btrfs_delayed_extent_op *extent_op;
7352 extent_op = btrfs_alloc_delayed_extent_op();
7353 BUG_ON(!extent_op); /* -ENOMEM */
7355 memcpy(&extent_op->key, key, sizeof(extent_op->key));
7357 memset(&extent_op->key, 0, sizeof(extent_op->key));
7358 extent_op->flags_to_set = flags;
7359 if (skinny_metadata)
7360 extent_op->update_key = 0;
7362 extent_op->update_key = 1;
7363 extent_op->update_flags = 1;
7364 extent_op->is_data = 0;
7365 extent_op->level = level;
7367 ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
7369 ins.offset, parent, root_objectid,
7370 level, BTRFS_ADD_DELAYED_EXTENT,
7372 BUG_ON(ret); /* -ENOMEM */
7377 struct walk_control {
7378 u64 refs[BTRFS_MAX_LEVEL];
7379 u64 flags[BTRFS_MAX_LEVEL];
7380 struct btrfs_key update_progress;
7391 #define DROP_REFERENCE 1
7392 #define UPDATE_BACKREF 2
7394 static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
7395 struct btrfs_root *root,
7396 struct walk_control *wc,
7397 struct btrfs_path *path)
7405 struct btrfs_key key;
7406 struct extent_buffer *eb;
7411 if (path->slots[wc->level] < wc->reada_slot) {
7412 wc->reada_count = wc->reada_count * 2 / 3;
7413 wc->reada_count = max(wc->reada_count, 2);
7415 wc->reada_count = wc->reada_count * 3 / 2;
7416 wc->reada_count = min_t(int, wc->reada_count,
7417 BTRFS_NODEPTRS_PER_BLOCK(root));
7420 eb = path->nodes[wc->level];
7421 nritems = btrfs_header_nritems(eb);
7422 blocksize = root->nodesize;
7424 for (slot = path->slots[wc->level]; slot < nritems; slot++) {
7425 if (nread >= wc->reada_count)
7429 bytenr = btrfs_node_blockptr(eb, slot);
7430 generation = btrfs_node_ptr_generation(eb, slot);
7432 if (slot == path->slots[wc->level])
7435 if (wc->stage == UPDATE_BACKREF &&
7436 generation <= root->root_key.offset)
7439 /* We don't lock the tree block, it's OK to be racy here */
7440 ret = btrfs_lookup_extent_info(trans, root, bytenr,
7441 wc->level - 1, 1, &refs,
7443 /* We don't care about errors in readahead. */
7448 if (wc->stage == DROP_REFERENCE) {
7452 if (wc->level == 1 &&
7453 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7455 if (!wc->update_ref ||
7456 generation <= root->root_key.offset)
7458 btrfs_node_key_to_cpu(eb, &key, slot);
7459 ret = btrfs_comp_cpu_keys(&key,
7460 &wc->update_progress);
7464 if (wc->level == 1 &&
7465 (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7469 ret = readahead_tree_block(root, bytenr, blocksize,
7475 wc->reada_slot = slot;
7478 static int account_leaf_items(struct btrfs_trans_handle *trans,
7479 struct btrfs_root *root,
7480 struct extent_buffer *eb)
7482 int nr = btrfs_header_nritems(eb);
7483 int i, extent_type, ret;
7484 struct btrfs_key key;
7485 struct btrfs_file_extent_item *fi;
7486 u64 bytenr, num_bytes;
7488 for (i = 0; i < nr; i++) {
7489 btrfs_item_key_to_cpu(eb, &key, i);
7491 if (key.type != BTRFS_EXTENT_DATA_KEY)
7494 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
7495 /* filter out non qgroup-accountable extents */
7496 extent_type = btrfs_file_extent_type(eb, fi);
7498 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
7501 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
7505 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
7507 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7510 BTRFS_QGROUP_OPER_SUB_SUBTREE, 0);
7518 * Walk up the tree from the bottom, freeing leaves and any interior
7519 * nodes which have had all slots visited. If a node (leaf or
7520 * interior) is freed, the node above it will have it's slot
7521 * incremented. The root node will never be freed.
7523 * At the end of this function, we should have a path which has all
7524 * slots incremented to the next position for a search. If we need to
7525 * read a new node it will be NULL and the node above it will have the
7526 * correct slot selected for a later read.
7528 * If we increment the root nodes slot counter past the number of
7529 * elements, 1 is returned to signal completion of the search.
7531 static int adjust_slots_upwards(struct btrfs_root *root,
7532 struct btrfs_path *path, int root_level)
7536 struct extent_buffer *eb;
7538 if (root_level == 0)
7541 while (level <= root_level) {
7542 eb = path->nodes[level];
7543 nr = btrfs_header_nritems(eb);
7544 path->slots[level]++;
7545 slot = path->slots[level];
7546 if (slot >= nr || level == 0) {
7548 * Don't free the root - we will detect this
7549 * condition after our loop and return a
7550 * positive value for caller to stop walking the tree.
7552 if (level != root_level) {
7553 btrfs_tree_unlock_rw(eb, path->locks[level]);
7554 path->locks[level] = 0;
7556 free_extent_buffer(eb);
7557 path->nodes[level] = NULL;
7558 path->slots[level] = 0;
7562 * We have a valid slot to walk back down
7563 * from. Stop here so caller can process these
7572 eb = path->nodes[root_level];
7573 if (path->slots[root_level] >= btrfs_header_nritems(eb))
7580 * root_eb is the subtree root and is locked before this function is called.
7582 static int account_shared_subtree(struct btrfs_trans_handle *trans,
7583 struct btrfs_root *root,
7584 struct extent_buffer *root_eb,
7590 struct extent_buffer *eb = root_eb;
7591 struct btrfs_path *path = NULL;
7593 BUG_ON(root_level < 0 || root_level > BTRFS_MAX_LEVEL);
7594 BUG_ON(root_eb == NULL);
7596 if (!root->fs_info->quota_enabled)
7599 if (!extent_buffer_uptodate(root_eb)) {
7600 ret = btrfs_read_buffer(root_eb, root_gen);
7605 if (root_level == 0) {
7606 ret = account_leaf_items(trans, root, root_eb);
7610 path = btrfs_alloc_path();
7615 * Walk down the tree. Missing extent blocks are filled in as
7616 * we go. Metadata is accounted every time we read a new
7619 * When we reach a leaf, we account for file extent items in it,
7620 * walk back up the tree (adjusting slot pointers as we go)
7621 * and restart the search process.
7623 extent_buffer_get(root_eb); /* For path */
7624 path->nodes[root_level] = root_eb;
7625 path->slots[root_level] = 0;
7626 path->locks[root_level] = 0; /* so release_path doesn't try to unlock */
7629 while (level >= 0) {
7630 if (path->nodes[level] == NULL) {
7631 int child_bsize = root->nodesize;
7636 /* We need to get child blockptr/gen from
7637 * parent before we can read it. */
7638 eb = path->nodes[level + 1];
7639 parent_slot = path->slots[level + 1];
7640 child_bytenr = btrfs_node_blockptr(eb, parent_slot);
7641 child_gen = btrfs_node_ptr_generation(eb, parent_slot);
7643 eb = read_tree_block(root, child_bytenr, child_bsize,
7645 if (!eb || !extent_buffer_uptodate(eb)) {
7650 path->nodes[level] = eb;
7651 path->slots[level] = 0;
7653 btrfs_tree_read_lock(eb);
7654 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
7655 path->locks[level] = BTRFS_READ_LOCK_BLOCKING;
7657 ret = btrfs_qgroup_record_ref(trans, root->fs_info,
7661 BTRFS_QGROUP_OPER_SUB_SUBTREE,
7669 ret = account_leaf_items(trans, root, path->nodes[level]);
7673 /* Nonzero return here means we completed our search */
7674 ret = adjust_slots_upwards(root, path, root_level);
7678 /* Restart search with new slots */
7687 btrfs_free_path(path);
7693 * helper to process tree block while walking down the tree.
7695 * when wc->stage == UPDATE_BACKREF, this function updates
7696 * back refs for pointers in the block.
7698 * NOTE: return value 1 means we should stop walking down.
7700 static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
7701 struct btrfs_root *root,
7702 struct btrfs_path *path,
7703 struct walk_control *wc, int lookup_info)
7705 int level = wc->level;
7706 struct extent_buffer *eb = path->nodes[level];
7707 u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
7710 if (wc->stage == UPDATE_BACKREF &&
7711 btrfs_header_owner(eb) != root->root_key.objectid)
7715 * when reference count of tree block is 1, it won't increase
7716 * again. once full backref flag is set, we never clear it.
7719 ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
7720 (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) {
7721 BUG_ON(!path->locks[level]);
7722 ret = btrfs_lookup_extent_info(trans, root,
7723 eb->start, level, 1,
7726 BUG_ON(ret == -ENOMEM);
7729 BUG_ON(wc->refs[level] == 0);
7732 if (wc->stage == DROP_REFERENCE) {
7733 if (wc->refs[level] > 1)
7736 if (path->locks[level] && !wc->keep_locks) {
7737 btrfs_tree_unlock_rw(eb, path->locks[level]);
7738 path->locks[level] = 0;
7743 /* wc->stage == UPDATE_BACKREF */
7744 if (!(wc->flags[level] & flag)) {
7745 BUG_ON(!path->locks[level]);
7746 ret = btrfs_inc_ref(trans, root, eb, 1);
7747 BUG_ON(ret); /* -ENOMEM */
7748 ret = btrfs_dec_ref(trans, root, eb, 0);
7749 BUG_ON(ret); /* -ENOMEM */
7750 ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
7752 btrfs_header_level(eb), 0);
7753 BUG_ON(ret); /* -ENOMEM */
7754 wc->flags[level] |= flag;
7758 * the block is shared by multiple trees, so it's not good to
7759 * keep the tree lock
7761 if (path->locks[level] && level > 0) {
7762 btrfs_tree_unlock_rw(eb, path->locks[level]);
7763 path->locks[level] = 0;
7769 * helper to process tree block pointer.
7771 * when wc->stage == DROP_REFERENCE, this function checks
7772 * reference count of the block pointed to. if the block
7773 * is shared and we need update back refs for the subtree
7774 * rooted at the block, this function changes wc->stage to
7775 * UPDATE_BACKREF. if the block is shared and there is no
7776 * need to update back, this function drops the reference
7779 * NOTE: return value 1 means we should stop walking down.
7781 static noinline int do_walk_down(struct btrfs_trans_handle *trans,
7782 struct btrfs_root *root,
7783 struct btrfs_path *path,
7784 struct walk_control *wc, int *lookup_info)
7790 struct btrfs_key key;
7791 struct extent_buffer *next;
7792 int level = wc->level;
7795 bool need_account = false;
7797 generation = btrfs_node_ptr_generation(path->nodes[level],
7798 path->slots[level]);
7800 * if the lower level block was created before the snapshot
7801 * was created, we know there is no need to update back refs
7804 if (wc->stage == UPDATE_BACKREF &&
7805 generation <= root->root_key.offset) {
7810 bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]);
7811 blocksize = root->nodesize;
7813 next = btrfs_find_tree_block(root, bytenr, blocksize);
7815 next = btrfs_find_create_tree_block(root, bytenr, blocksize);
7818 btrfs_set_buffer_lockdep_class(root->root_key.objectid, next,
7822 btrfs_tree_lock(next);
7823 btrfs_set_lock_blocking(next);
7825 ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1,
7826 &wc->refs[level - 1],
7827 &wc->flags[level - 1]);
7829 btrfs_tree_unlock(next);
7833 if (unlikely(wc->refs[level - 1] == 0)) {
7834 btrfs_err(root->fs_info, "Missing references.");
7839 if (wc->stage == DROP_REFERENCE) {
7840 if (wc->refs[level - 1] > 1) {
7841 need_account = true;
7843 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7846 if (!wc->update_ref ||
7847 generation <= root->root_key.offset)
7850 btrfs_node_key_to_cpu(path->nodes[level], &key,
7851 path->slots[level]);
7852 ret = btrfs_comp_cpu_keys(&key, &wc->update_progress);
7856 wc->stage = UPDATE_BACKREF;
7857 wc->shared_level = level - 1;
7861 (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF))
7865 if (!btrfs_buffer_uptodate(next, generation, 0)) {
7866 btrfs_tree_unlock(next);
7867 free_extent_buffer(next);
7873 if (reada && level == 1)
7874 reada_walk_down(trans, root, wc, path);
7875 next = read_tree_block(root, bytenr, blocksize, generation);
7876 if (!next || !extent_buffer_uptodate(next)) {
7877 free_extent_buffer(next);
7880 btrfs_tree_lock(next);
7881 btrfs_set_lock_blocking(next);
7885 BUG_ON(level != btrfs_header_level(next));
7886 path->nodes[level] = next;
7887 path->slots[level] = 0;
7888 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7894 wc->refs[level - 1] = 0;
7895 wc->flags[level - 1] = 0;
7896 if (wc->stage == DROP_REFERENCE) {
7897 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
7898 parent = path->nodes[level]->start;
7900 BUG_ON(root->root_key.objectid !=
7901 btrfs_header_owner(path->nodes[level]));
7906 ret = account_shared_subtree(trans, root, next,
7907 generation, level - 1);
7909 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
7910 "%d accounting shared subtree. Quota "
7911 "is out of sync, rescan required.\n",
7912 root->fs_info->sb->s_id, ret);
7915 ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
7916 root->root_key.objectid, level - 1, 0, 0);
7917 BUG_ON(ret); /* -ENOMEM */
7919 btrfs_tree_unlock(next);
7920 free_extent_buffer(next);
7926 * helper to process tree block while walking up the tree.
7928 * when wc->stage == DROP_REFERENCE, this function drops
7929 * reference count on the block.
7931 * when wc->stage == UPDATE_BACKREF, this function changes
7932 * wc->stage back to DROP_REFERENCE if we changed wc->stage
7933 * to UPDATE_BACKREF previously while processing the block.
7935 * NOTE: return value 1 means we should stop walking up.
7937 static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
7938 struct btrfs_root *root,
7939 struct btrfs_path *path,
7940 struct walk_control *wc)
7943 int level = wc->level;
7944 struct extent_buffer *eb = path->nodes[level];
7947 if (wc->stage == UPDATE_BACKREF) {
7948 BUG_ON(wc->shared_level < level);
7949 if (level < wc->shared_level)
7952 ret = find_next_key(path, level + 1, &wc->update_progress);
7956 wc->stage = DROP_REFERENCE;
7957 wc->shared_level = -1;
7958 path->slots[level] = 0;
7961 * check reference count again if the block isn't locked.
7962 * we should start walking down the tree again if reference
7965 if (!path->locks[level]) {
7967 btrfs_tree_lock(eb);
7968 btrfs_set_lock_blocking(eb);
7969 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
7971 ret = btrfs_lookup_extent_info(trans, root,
7972 eb->start, level, 1,
7976 btrfs_tree_unlock_rw(eb, path->locks[level]);
7977 path->locks[level] = 0;
7980 BUG_ON(wc->refs[level] == 0);
7981 if (wc->refs[level] == 1) {
7982 btrfs_tree_unlock_rw(eb, path->locks[level]);
7983 path->locks[level] = 0;
7989 /* wc->stage == DROP_REFERENCE */
7990 BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
7992 if (wc->refs[level] == 1) {
7994 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
7995 ret = btrfs_dec_ref(trans, root, eb, 1);
7997 ret = btrfs_dec_ref(trans, root, eb, 0);
7998 BUG_ON(ret); /* -ENOMEM */
7999 ret = account_leaf_items(trans, root, eb);
8001 printk_ratelimited(KERN_ERR "BTRFS: %s Error "
8002 "%d accounting leaf items. Quota "
8003 "is out of sync, rescan required.\n",
8004 root->fs_info->sb->s_id, ret);
8007 /* make block locked assertion in clean_tree_block happy */
8008 if (!path->locks[level] &&
8009 btrfs_header_generation(eb) == trans->transid) {
8010 btrfs_tree_lock(eb);
8011 btrfs_set_lock_blocking(eb);
8012 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8014 clean_tree_block(trans, root, eb);
8017 if (eb == root->node) {
8018 if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8021 BUG_ON(root->root_key.objectid !=
8022 btrfs_header_owner(eb));
8024 if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
8025 parent = path->nodes[level + 1]->start;
8027 BUG_ON(root->root_key.objectid !=
8028 btrfs_header_owner(path->nodes[level + 1]));
8031 btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
8033 wc->refs[level] = 0;
8034 wc->flags[level] = 0;
8038 static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
8039 struct btrfs_root *root,
8040 struct btrfs_path *path,
8041 struct walk_control *wc)
8043 int level = wc->level;
8044 int lookup_info = 1;
8047 while (level >= 0) {
8048 ret = walk_down_proc(trans, root, path, wc, lookup_info);
8055 if (path->slots[level] >=
8056 btrfs_header_nritems(path->nodes[level]))
8059 ret = do_walk_down(trans, root, path, wc, &lookup_info);
8061 path->slots[level]++;
8070 static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
8071 struct btrfs_root *root,
8072 struct btrfs_path *path,
8073 struct walk_control *wc, int max_level)
8075 int level = wc->level;
8078 path->slots[level] = btrfs_header_nritems(path->nodes[level]);
8079 while (level < max_level && path->nodes[level]) {
8081 if (path->slots[level] + 1 <
8082 btrfs_header_nritems(path->nodes[level])) {
8083 path->slots[level]++;
8086 ret = walk_up_proc(trans, root, path, wc);
8090 if (path->locks[level]) {
8091 btrfs_tree_unlock_rw(path->nodes[level],
8092 path->locks[level]);
8093 path->locks[level] = 0;
8095 free_extent_buffer(path->nodes[level]);
8096 path->nodes[level] = NULL;
8104 * drop a subvolume tree.
8106 * this function traverses the tree freeing any blocks that only
8107 * referenced by the tree.
8109 * when a shared tree block is found. this function decreases its
8110 * reference count by one. if update_ref is true, this function
8111 * also make sure backrefs for the shared block and all lower level
8112 * blocks are properly updated.
8114 * If called with for_reloc == 0, may exit early with -EAGAIN
8116 int btrfs_drop_snapshot(struct btrfs_root *root,
8117 struct btrfs_block_rsv *block_rsv, int update_ref,
8120 struct btrfs_path *path;
8121 struct btrfs_trans_handle *trans;
8122 struct btrfs_root *tree_root = root->fs_info->tree_root;
8123 struct btrfs_root_item *root_item = &root->root_item;
8124 struct walk_control *wc;
8125 struct btrfs_key key;
8129 bool root_dropped = false;
8131 btrfs_debug(root->fs_info, "Drop subvolume %llu", root->objectid);
8133 path = btrfs_alloc_path();
8139 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8141 btrfs_free_path(path);
8146 trans = btrfs_start_transaction(tree_root, 0);
8147 if (IS_ERR(trans)) {
8148 err = PTR_ERR(trans);
8153 trans->block_rsv = block_rsv;
8155 if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
8156 level = btrfs_header_level(root->node);
8157 path->nodes[level] = btrfs_lock_root_node(root);
8158 btrfs_set_lock_blocking(path->nodes[level]);
8159 path->slots[level] = 0;
8160 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8161 memset(&wc->update_progress, 0,
8162 sizeof(wc->update_progress));
8164 btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
8165 memcpy(&wc->update_progress, &key,
8166 sizeof(wc->update_progress));
8168 level = root_item->drop_level;
8170 path->lowest_level = level;
8171 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
8172 path->lowest_level = 0;
8180 * unlock our path, this is safe because only this
8181 * function is allowed to delete this snapshot
8183 btrfs_unlock_up_safe(path, 0);
8185 level = btrfs_header_level(root->node);
8187 btrfs_tree_lock(path->nodes[level]);
8188 btrfs_set_lock_blocking(path->nodes[level]);
8189 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8191 ret = btrfs_lookup_extent_info(trans, root,
8192 path->nodes[level]->start,
8193 level, 1, &wc->refs[level],
8199 BUG_ON(wc->refs[level] == 0);
8201 if (level == root_item->drop_level)
8204 btrfs_tree_unlock(path->nodes[level]);
8205 path->locks[level] = 0;
8206 WARN_ON(wc->refs[level] != 1);
8212 wc->shared_level = -1;
8213 wc->stage = DROP_REFERENCE;
8214 wc->update_ref = update_ref;
8216 wc->for_reloc = for_reloc;
8217 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8221 ret = walk_down_tree(trans, root, path, wc);
8227 ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
8234 BUG_ON(wc->stage != DROP_REFERENCE);
8238 if (wc->stage == DROP_REFERENCE) {
8240 btrfs_node_key(path->nodes[level],
8241 &root_item->drop_progress,
8242 path->slots[level]);
8243 root_item->drop_level = level;
8246 BUG_ON(wc->level == 0);
8247 if (btrfs_should_end_transaction(trans, tree_root) ||
8248 (!for_reloc && btrfs_need_cleaner_sleep(root))) {
8249 ret = btrfs_update_root(trans, tree_root,
8253 btrfs_abort_transaction(trans, tree_root, ret);
8259 * Qgroup update accounting is run from
8260 * delayed ref handling. This usually works
8261 * out because delayed refs are normally the
8262 * only way qgroup updates are added. However,
8263 * we may have added updates during our tree
8264 * walk so run qgroups here to make sure we
8265 * don't lose any updates.
8267 ret = btrfs_delayed_qgroup_accounting(trans,
8270 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8271 "running qgroup updates "
8272 "during snapshot delete. "
8273 "Quota is out of sync, "
8274 "rescan required.\n", ret);
8276 btrfs_end_transaction_throttle(trans, tree_root);
8277 if (!for_reloc && btrfs_need_cleaner_sleep(root)) {
8278 pr_debug("BTRFS: drop snapshot early exit\n");
8283 trans = btrfs_start_transaction(tree_root, 0);
8284 if (IS_ERR(trans)) {
8285 err = PTR_ERR(trans);
8289 trans->block_rsv = block_rsv;
8292 btrfs_release_path(path);
8296 ret = btrfs_del_root(trans, tree_root, &root->root_key);
8298 btrfs_abort_transaction(trans, tree_root, ret);
8302 if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
8303 ret = btrfs_find_root(tree_root, &root->root_key, path,
8306 btrfs_abort_transaction(trans, tree_root, ret);
8309 } else if (ret > 0) {
8310 /* if we fail to delete the orphan item this time
8311 * around, it'll get picked up the next time.
8313 * The most common failure here is just -ENOENT.
8315 btrfs_del_orphan_item(trans, tree_root,
8316 root->root_key.objectid);
8320 if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) {
8321 btrfs_drop_and_free_fs_root(tree_root->fs_info, root);
8323 free_extent_buffer(root->node);
8324 free_extent_buffer(root->commit_root);
8325 btrfs_put_fs_root(root);
8327 root_dropped = true;
8329 ret = btrfs_delayed_qgroup_accounting(trans, tree_root->fs_info);
8331 printk_ratelimited(KERN_ERR "BTRFS: Failure %d "
8332 "running qgroup updates "
8333 "during snapshot delete. "
8334 "Quota is out of sync, "
8335 "rescan required.\n", ret);
8337 btrfs_end_transaction_throttle(trans, tree_root);
8340 btrfs_free_path(path);
8343 * So if we need to stop dropping the snapshot for whatever reason we
8344 * need to make sure to add it back to the dead root list so that we
8345 * keep trying to do the work later. This also cleans up roots if we
8346 * don't have it in the radix (like when we recover after a power fail
8347 * or unmount) so we don't leak memory.
8349 if (!for_reloc && root_dropped == false)
8350 btrfs_add_dead_root(root);
8351 if (err && err != -EAGAIN)
8352 btrfs_std_error(root->fs_info, err);
8357 * drop subtree rooted at tree block 'node'.
8359 * NOTE: this function will unlock and release tree block 'node'
8360 * only used by relocation code
8362 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
8363 struct btrfs_root *root,
8364 struct extent_buffer *node,
8365 struct extent_buffer *parent)
8367 struct btrfs_path *path;
8368 struct walk_control *wc;
8374 BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
8376 path = btrfs_alloc_path();
8380 wc = kzalloc(sizeof(*wc), GFP_NOFS);
8382 btrfs_free_path(path);
8386 btrfs_assert_tree_locked(parent);
8387 parent_level = btrfs_header_level(parent);
8388 extent_buffer_get(parent);
8389 path->nodes[parent_level] = parent;
8390 path->slots[parent_level] = btrfs_header_nritems(parent);
8392 btrfs_assert_tree_locked(node);
8393 level = btrfs_header_level(node);
8394 path->nodes[level] = node;
8395 path->slots[level] = 0;
8396 path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING;
8398 wc->refs[parent_level] = 1;
8399 wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
8401 wc->shared_level = -1;
8402 wc->stage = DROP_REFERENCE;
8406 wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
8409 wret = walk_down_tree(trans, root, path, wc);
8415 wret = walk_up_tree(trans, root, path, wc, parent_level);
8423 btrfs_free_path(path);
8427 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
8433 * if restripe for this chunk_type is on pick target profile and
8434 * return, otherwise do the usual balance
8436 stripped = get_restripe_target(root->fs_info, flags);
8438 return extended_to_chunk(stripped);
8440 num_devices = root->fs_info->fs_devices->rw_devices;
8442 stripped = BTRFS_BLOCK_GROUP_RAID0 |
8443 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
8444 BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
8446 if (num_devices == 1) {
8447 stripped |= BTRFS_BLOCK_GROUP_DUP;
8448 stripped = flags & ~stripped;
8450 /* turn raid0 into single device chunks */
8451 if (flags & BTRFS_BLOCK_GROUP_RAID0)
8454 /* turn mirroring into duplication */
8455 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
8456 BTRFS_BLOCK_GROUP_RAID10))
8457 return stripped | BTRFS_BLOCK_GROUP_DUP;
8459 /* they already had raid on here, just return */
8460 if (flags & stripped)
8463 stripped |= BTRFS_BLOCK_GROUP_DUP;
8464 stripped = flags & ~stripped;
8466 /* switch duplicated blocks with raid1 */
8467 if (flags & BTRFS_BLOCK_GROUP_DUP)
8468 return stripped | BTRFS_BLOCK_GROUP_RAID1;
8470 /* this is drive concat, leave it alone */
8476 static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force)
8478 struct btrfs_space_info *sinfo = cache->space_info;
8480 u64 min_allocable_bytes;
8485 * We need some metadata space and system metadata space for
8486 * allocating chunks in some corner cases until we force to set
8487 * it to be readonly.
8490 (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) &&
8492 min_allocable_bytes = 1 * 1024 * 1024;
8494 min_allocable_bytes = 0;
8496 spin_lock(&sinfo->lock);
8497 spin_lock(&cache->lock);
8504 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8505 cache->bytes_super - btrfs_block_group_used(&cache->item);
8507 if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned +
8508 sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes +
8509 min_allocable_bytes <= sinfo->total_bytes) {
8510 sinfo->bytes_readonly += num_bytes;
8515 spin_unlock(&cache->lock);
8516 spin_unlock(&sinfo->lock);
8520 int btrfs_set_block_group_ro(struct btrfs_root *root,
8521 struct btrfs_block_group_cache *cache)
8524 struct btrfs_trans_handle *trans;
8530 trans = btrfs_join_transaction(root);
8532 return PTR_ERR(trans);
8534 alloc_flags = update_block_group_flags(root, cache->flags);
8535 if (alloc_flags != cache->flags) {
8536 ret = do_chunk_alloc(trans, root, alloc_flags,
8542 ret = set_block_group_ro(cache, 0);
8545 alloc_flags = get_alloc_profile(root, cache->space_info->flags);
8546 ret = do_chunk_alloc(trans, root, alloc_flags,
8550 ret = set_block_group_ro(cache, 0);
8552 btrfs_end_transaction(trans, root);
8556 int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans,
8557 struct btrfs_root *root, u64 type)
8559 u64 alloc_flags = get_alloc_profile(root, type);
8560 return do_chunk_alloc(trans, root, alloc_flags,
8565 * helper to account the unused space of all the readonly block group in the
8566 * list. takes mirrors into account.
8568 static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list)
8570 struct btrfs_block_group_cache *block_group;
8574 list_for_each_entry(block_group, groups_list, list) {
8575 spin_lock(&block_group->lock);
8577 if (!block_group->ro) {
8578 spin_unlock(&block_group->lock);
8582 if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 |
8583 BTRFS_BLOCK_GROUP_RAID10 |
8584 BTRFS_BLOCK_GROUP_DUP))
8589 free_bytes += (block_group->key.offset -
8590 btrfs_block_group_used(&block_group->item)) *
8593 spin_unlock(&block_group->lock);
8600 * helper to account the unused space of all the readonly block group in the
8601 * space_info. takes mirrors into account.
8603 u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
8608 spin_lock(&sinfo->lock);
8610 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++)
8611 if (!list_empty(&sinfo->block_groups[i]))
8612 free_bytes += __btrfs_get_ro_block_group_free_space(
8613 &sinfo->block_groups[i]);
8615 spin_unlock(&sinfo->lock);
8620 void btrfs_set_block_group_rw(struct btrfs_root *root,
8621 struct btrfs_block_group_cache *cache)
8623 struct btrfs_space_info *sinfo = cache->space_info;
8628 spin_lock(&sinfo->lock);
8629 spin_lock(&cache->lock);
8630 num_bytes = cache->key.offset - cache->reserved - cache->pinned -
8631 cache->bytes_super - btrfs_block_group_used(&cache->item);
8632 sinfo->bytes_readonly -= num_bytes;
8634 spin_unlock(&cache->lock);
8635 spin_unlock(&sinfo->lock);
8639 * checks to see if its even possible to relocate this block group.
8641 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
8642 * ok to go ahead and try.
8644 int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
8646 struct btrfs_block_group_cache *block_group;
8647 struct btrfs_space_info *space_info;
8648 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
8649 struct btrfs_device *device;
8650 struct btrfs_trans_handle *trans;
8659 block_group = btrfs_lookup_block_group(root->fs_info, bytenr);
8661 /* odd, couldn't find the block group, leave it alone */
8665 min_free = btrfs_block_group_used(&block_group->item);
8667 /* no bytes used, we're good */
8671 space_info = block_group->space_info;
8672 spin_lock(&space_info->lock);
8674 full = space_info->full;
8677 * if this is the last block group we have in this space, we can't
8678 * relocate it unless we're able to allocate a new chunk below.
8680 * Otherwise, we need to make sure we have room in the space to handle
8681 * all of the extents from this block group. If we can, we're good
8683 if ((space_info->total_bytes != block_group->key.offset) &&
8684 (space_info->bytes_used + space_info->bytes_reserved +
8685 space_info->bytes_pinned + space_info->bytes_readonly +
8686 min_free < space_info->total_bytes)) {
8687 spin_unlock(&space_info->lock);
8690 spin_unlock(&space_info->lock);
8693 * ok we don't have enough space, but maybe we have free space on our
8694 * devices to allocate new chunks for relocation, so loop through our
8695 * alloc devices and guess if we have enough space. if this block
8696 * group is going to be restriped, run checks against the target
8697 * profile instead of the current one.
8709 target = get_restripe_target(root->fs_info, block_group->flags);
8711 index = __get_raid_index(extended_to_chunk(target));
8714 * this is just a balance, so if we were marked as full
8715 * we know there is no space for a new chunk
8720 index = get_block_group_index(block_group);
8723 if (index == BTRFS_RAID_RAID10) {
8727 } else if (index == BTRFS_RAID_RAID1) {
8729 } else if (index == BTRFS_RAID_DUP) {
8732 } else if (index == BTRFS_RAID_RAID0) {
8733 dev_min = fs_devices->rw_devices;
8734 do_div(min_free, dev_min);
8737 /* We need to do this so that we can look at pending chunks */
8738 trans = btrfs_join_transaction(root);
8739 if (IS_ERR(trans)) {
8740 ret = PTR_ERR(trans);
8744 mutex_lock(&root->fs_info->chunk_mutex);
8745 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
8749 * check to make sure we can actually find a chunk with enough
8750 * space to fit our block group in.
8752 if (device->total_bytes > device->bytes_used + min_free &&
8753 !device->is_tgtdev_for_dev_replace) {
8754 ret = find_free_dev_extent(trans, device, min_free,
8759 if (dev_nr >= dev_min)
8765 mutex_unlock(&root->fs_info->chunk_mutex);
8766 btrfs_end_transaction(trans, root);
8768 btrfs_put_block_group(block_group);
8772 static int find_first_block_group(struct btrfs_root *root,
8773 struct btrfs_path *path, struct btrfs_key *key)
8776 struct btrfs_key found_key;
8777 struct extent_buffer *leaf;
8780 ret = btrfs_search_slot(NULL, root, key, path, 0, 0);
8785 slot = path->slots[0];
8786 leaf = path->nodes[0];
8787 if (slot >= btrfs_header_nritems(leaf)) {
8788 ret = btrfs_next_leaf(root, path);
8795 btrfs_item_key_to_cpu(leaf, &found_key, slot);
8797 if (found_key.objectid >= key->objectid &&
8798 found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
8808 void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
8810 struct btrfs_block_group_cache *block_group;
8814 struct inode *inode;
8816 block_group = btrfs_lookup_first_block_group(info, last);
8817 while (block_group) {
8818 spin_lock(&block_group->lock);
8819 if (block_group->iref)
8821 spin_unlock(&block_group->lock);
8822 block_group = next_block_group(info->tree_root,
8832 inode = block_group->inode;
8833 block_group->iref = 0;
8834 block_group->inode = NULL;
8835 spin_unlock(&block_group->lock);
8837 last = block_group->key.objectid + block_group->key.offset;
8838 btrfs_put_block_group(block_group);
8842 int btrfs_free_block_groups(struct btrfs_fs_info *info)
8844 struct btrfs_block_group_cache *block_group;
8845 struct btrfs_space_info *space_info;
8846 struct btrfs_caching_control *caching_ctl;
8849 down_write(&info->commit_root_sem);
8850 while (!list_empty(&info->caching_block_groups)) {
8851 caching_ctl = list_entry(info->caching_block_groups.next,
8852 struct btrfs_caching_control, list);
8853 list_del(&caching_ctl->list);
8854 put_caching_control(caching_ctl);
8856 up_write(&info->commit_root_sem);
8858 spin_lock(&info->block_group_cache_lock);
8859 while ((n = rb_last(&info->block_group_cache_tree)) != NULL) {
8860 block_group = rb_entry(n, struct btrfs_block_group_cache,
8862 rb_erase(&block_group->cache_node,
8863 &info->block_group_cache_tree);
8864 spin_unlock(&info->block_group_cache_lock);
8866 down_write(&block_group->space_info->groups_sem);
8867 list_del(&block_group->list);
8868 up_write(&block_group->space_info->groups_sem);
8870 if (block_group->cached == BTRFS_CACHE_STARTED)
8871 wait_block_group_cache_done(block_group);
8874 * We haven't cached this block group, which means we could
8875 * possibly have excluded extents on this block group.
8877 if (block_group->cached == BTRFS_CACHE_NO ||
8878 block_group->cached == BTRFS_CACHE_ERROR)
8879 free_excluded_extents(info->extent_root, block_group);
8881 btrfs_remove_free_space_cache(block_group);
8882 btrfs_put_block_group(block_group);
8884 spin_lock(&info->block_group_cache_lock);
8886 spin_unlock(&info->block_group_cache_lock);
8888 /* now that all the block groups are freed, go through and
8889 * free all the space_info structs. This is only called during
8890 * the final stages of unmount, and so we know nobody is
8891 * using them. We call synchronize_rcu() once before we start,
8892 * just to be on the safe side.
8896 release_global_block_rsv(info);
8898 while (!list_empty(&info->space_info)) {
8901 space_info = list_entry(info->space_info.next,
8902 struct btrfs_space_info,
8904 if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) {
8905 if (WARN_ON(space_info->bytes_pinned > 0 ||
8906 space_info->bytes_reserved > 0 ||
8907 space_info->bytes_may_use > 0)) {
8908 dump_space_info(space_info, 0, 0);
8911 list_del(&space_info->list);
8912 for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) {
8913 struct kobject *kobj;
8914 kobj = space_info->block_group_kobjs[i];
8915 space_info->block_group_kobjs[i] = NULL;
8921 kobject_del(&space_info->kobj);
8922 kobject_put(&space_info->kobj);
8927 static void __link_block_group(struct btrfs_space_info *space_info,
8928 struct btrfs_block_group_cache *cache)
8930 int index = get_block_group_index(cache);
8933 down_write(&space_info->groups_sem);
8934 if (list_empty(&space_info->block_groups[index]))
8936 list_add_tail(&cache->list, &space_info->block_groups[index]);
8937 up_write(&space_info->groups_sem);
8940 struct raid_kobject *rkobj;
8943 rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS);
8946 rkobj->raid_type = index;
8947 kobject_init(&rkobj->kobj, &btrfs_raid_ktype);
8948 ret = kobject_add(&rkobj->kobj, &space_info->kobj,
8949 "%s", get_raid_name(index));
8951 kobject_put(&rkobj->kobj);
8954 space_info->block_group_kobjs[index] = &rkobj->kobj;
8959 pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n");
8962 static struct btrfs_block_group_cache *
8963 btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size)
8965 struct btrfs_block_group_cache *cache;
8967 cache = kzalloc(sizeof(*cache), GFP_NOFS);
8971 cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl),
8973 if (!cache->free_space_ctl) {
8978 cache->key.objectid = start;
8979 cache->key.offset = size;
8980 cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
8982 cache->sectorsize = root->sectorsize;
8983 cache->fs_info = root->fs_info;
8984 cache->full_stripe_len = btrfs_full_stripe_len(root,
8985 &root->fs_info->mapping_tree,
8987 atomic_set(&cache->count, 1);
8988 spin_lock_init(&cache->lock);
8989 init_rwsem(&cache->data_rwsem);
8990 INIT_LIST_HEAD(&cache->list);
8991 INIT_LIST_HEAD(&cache->cluster_list);
8992 INIT_LIST_HEAD(&cache->new_bg_list);
8993 btrfs_init_free_space_ctl(cache);
8998 int btrfs_read_block_groups(struct btrfs_root *root)
9000 struct btrfs_path *path;
9002 struct btrfs_block_group_cache *cache;
9003 struct btrfs_fs_info *info = root->fs_info;
9004 struct btrfs_space_info *space_info;
9005 struct btrfs_key key;
9006 struct btrfs_key found_key;
9007 struct extent_buffer *leaf;
9011 root = info->extent_root;
9014 key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
9015 path = btrfs_alloc_path();
9020 cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy);
9021 if (btrfs_test_opt(root, SPACE_CACHE) &&
9022 btrfs_super_generation(root->fs_info->super_copy) != cache_gen)
9024 if (btrfs_test_opt(root, CLEAR_CACHE))
9028 ret = find_first_block_group(root, path, &key);
9034 leaf = path->nodes[0];
9035 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
9037 cache = btrfs_create_block_group_cache(root, found_key.objectid,
9046 * When we mount with old space cache, we need to
9047 * set BTRFS_DC_CLEAR and set dirty flag.
9049 * a) Setting 'BTRFS_DC_CLEAR' makes sure that we
9050 * truncate the old free space cache inode and
9052 * b) Setting 'dirty flag' makes sure that we flush
9053 * the new space cache info onto disk.
9055 cache->disk_cache_state = BTRFS_DC_CLEAR;
9056 if (btrfs_test_opt(root, SPACE_CACHE))
9060 read_extent_buffer(leaf, &cache->item,
9061 btrfs_item_ptr_offset(leaf, path->slots[0]),
9062 sizeof(cache->item));
9063 cache->flags = btrfs_block_group_flags(&cache->item);
9065 key.objectid = found_key.objectid + found_key.offset;
9066 btrfs_release_path(path);
9069 * We need to exclude the super stripes now so that the space
9070 * info has super bytes accounted for, otherwise we'll think
9071 * we have more space than we actually do.
9073 ret = exclude_super_stripes(root, cache);
9076 * We may have excluded something, so call this just in
9079 free_excluded_extents(root, cache);
9080 btrfs_put_block_group(cache);
9085 * check for two cases, either we are full, and therefore
9086 * don't need to bother with the caching work since we won't
9087 * find any space, or we are empty, and we can just add all
9088 * the space in and be done with it. This saves us _alot_ of
9089 * time, particularly in the full case.
9091 if (found_key.offset == btrfs_block_group_used(&cache->item)) {
9092 cache->last_byte_to_unpin = (u64)-1;
9093 cache->cached = BTRFS_CACHE_FINISHED;
9094 free_excluded_extents(root, cache);
9095 } else if (btrfs_block_group_used(&cache->item) == 0) {
9096 cache->last_byte_to_unpin = (u64)-1;
9097 cache->cached = BTRFS_CACHE_FINISHED;
9098 add_new_free_space(cache, root->fs_info,
9100 found_key.objectid +
9102 free_excluded_extents(root, cache);
9105 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9107 btrfs_remove_free_space_cache(cache);
9108 btrfs_put_block_group(cache);
9112 ret = update_space_info(info, cache->flags, found_key.offset,
9113 btrfs_block_group_used(&cache->item),
9116 btrfs_remove_free_space_cache(cache);
9117 spin_lock(&info->block_group_cache_lock);
9118 rb_erase(&cache->cache_node,
9119 &info->block_group_cache_tree);
9120 spin_unlock(&info->block_group_cache_lock);
9121 btrfs_put_block_group(cache);
9125 cache->space_info = space_info;
9126 spin_lock(&cache->space_info->lock);
9127 cache->space_info->bytes_readonly += cache->bytes_super;
9128 spin_unlock(&cache->space_info->lock);
9130 __link_block_group(space_info, cache);
9132 set_avail_alloc_bits(root->fs_info, cache->flags);
9133 if (btrfs_chunk_readonly(root, cache->key.objectid))
9134 set_block_group_ro(cache, 1);
9137 list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) {
9138 if (!(get_alloc_profile(root, space_info->flags) &
9139 (BTRFS_BLOCK_GROUP_RAID10 |
9140 BTRFS_BLOCK_GROUP_RAID1 |
9141 BTRFS_BLOCK_GROUP_RAID5 |
9142 BTRFS_BLOCK_GROUP_RAID6 |
9143 BTRFS_BLOCK_GROUP_DUP)))
9146 * avoid allocating from un-mirrored block group if there are
9147 * mirrored block groups.
9149 list_for_each_entry(cache,
9150 &space_info->block_groups[BTRFS_RAID_RAID0],
9152 set_block_group_ro(cache, 1);
9153 list_for_each_entry(cache,
9154 &space_info->block_groups[BTRFS_RAID_SINGLE],
9156 set_block_group_ro(cache, 1);
9159 init_global_block_rsv(info);
9162 btrfs_free_path(path);
9166 void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans,
9167 struct btrfs_root *root)
9169 struct btrfs_block_group_cache *block_group, *tmp;
9170 struct btrfs_root *extent_root = root->fs_info->extent_root;
9171 struct btrfs_block_group_item item;
9172 struct btrfs_key key;
9175 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs,
9177 list_del_init(&block_group->new_bg_list);
9182 spin_lock(&block_group->lock);
9183 memcpy(&item, &block_group->item, sizeof(item));
9184 memcpy(&key, &block_group->key, sizeof(key));
9185 spin_unlock(&block_group->lock);
9187 ret = btrfs_insert_item(trans, extent_root, &key, &item,
9190 btrfs_abort_transaction(trans, extent_root, ret);
9191 ret = btrfs_finish_chunk_alloc(trans, extent_root,
9192 key.objectid, key.offset);
9194 btrfs_abort_transaction(trans, extent_root, ret);
9198 int btrfs_make_block_group(struct btrfs_trans_handle *trans,
9199 struct btrfs_root *root, u64 bytes_used,
9200 u64 type, u64 chunk_objectid, u64 chunk_offset,
9204 struct btrfs_root *extent_root;
9205 struct btrfs_block_group_cache *cache;
9207 extent_root = root->fs_info->extent_root;
9209 btrfs_set_log_full_commit(root->fs_info, trans);
9211 cache = btrfs_create_block_group_cache(root, chunk_offset, size);
9215 btrfs_set_block_group_used(&cache->item, bytes_used);
9216 btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid);
9217 btrfs_set_block_group_flags(&cache->item, type);
9219 cache->flags = type;
9220 cache->last_byte_to_unpin = (u64)-1;
9221 cache->cached = BTRFS_CACHE_FINISHED;
9222 ret = exclude_super_stripes(root, cache);
9225 * We may have excluded something, so call this just in
9228 free_excluded_extents(root, cache);
9229 btrfs_put_block_group(cache);
9233 add_new_free_space(cache, root->fs_info, chunk_offset,
9234 chunk_offset + size);
9236 free_excluded_extents(root, cache);
9238 ret = btrfs_add_block_group_cache(root->fs_info, cache);
9240 btrfs_remove_free_space_cache(cache);
9241 btrfs_put_block_group(cache);
9245 ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
9246 &cache->space_info);
9248 btrfs_remove_free_space_cache(cache);
9249 spin_lock(&root->fs_info->block_group_cache_lock);
9250 rb_erase(&cache->cache_node,
9251 &root->fs_info->block_group_cache_tree);
9252 spin_unlock(&root->fs_info->block_group_cache_lock);
9253 btrfs_put_block_group(cache);
9256 update_global_block_rsv(root->fs_info);
9258 spin_lock(&cache->space_info->lock);
9259 cache->space_info->bytes_readonly += cache->bytes_super;
9260 spin_unlock(&cache->space_info->lock);
9262 __link_block_group(cache->space_info, cache);
9264 list_add_tail(&cache->new_bg_list, &trans->new_bgs);
9266 set_avail_alloc_bits(extent_root->fs_info, type);
9271 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
9273 u64 extra_flags = chunk_to_extended(flags) &
9274 BTRFS_EXTENDED_PROFILE_MASK;
9276 write_seqlock(&fs_info->profiles_lock);
9277 if (flags & BTRFS_BLOCK_GROUP_DATA)
9278 fs_info->avail_data_alloc_bits &= ~extra_flags;
9279 if (flags & BTRFS_BLOCK_GROUP_METADATA)
9280 fs_info->avail_metadata_alloc_bits &= ~extra_flags;
9281 if (flags & BTRFS_BLOCK_GROUP_SYSTEM)
9282 fs_info->avail_system_alloc_bits &= ~extra_flags;
9283 write_sequnlock(&fs_info->profiles_lock);
9286 int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
9287 struct btrfs_root *root, u64 group_start)
9289 struct btrfs_path *path;
9290 struct btrfs_block_group_cache *block_group;
9291 struct btrfs_free_cluster *cluster;
9292 struct btrfs_root *tree_root = root->fs_info->tree_root;
9293 struct btrfs_key key;
9294 struct inode *inode;
9295 struct kobject *kobj = NULL;
9300 root = root->fs_info->extent_root;
9302 block_group = btrfs_lookup_block_group(root->fs_info, group_start);
9303 BUG_ON(!block_group);
9304 BUG_ON(!block_group->ro);
9307 * Free the reserved super bytes from this block group before
9310 free_excluded_extents(root, block_group);
9312 memcpy(&key, &block_group->key, sizeof(key));
9313 index = get_block_group_index(block_group);
9314 if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP |
9315 BTRFS_BLOCK_GROUP_RAID1 |
9316 BTRFS_BLOCK_GROUP_RAID10))
9321 /* make sure this block group isn't part of an allocation cluster */
9322 cluster = &root->fs_info->data_alloc_cluster;
9323 spin_lock(&cluster->refill_lock);
9324 btrfs_return_cluster_to_free_space(block_group, cluster);
9325 spin_unlock(&cluster->refill_lock);
9328 * make sure this block group isn't part of a metadata
9329 * allocation cluster
9331 cluster = &root->fs_info->meta_alloc_cluster;
9332 spin_lock(&cluster->refill_lock);
9333 btrfs_return_cluster_to_free_space(block_group, cluster);
9334 spin_unlock(&cluster->refill_lock);
9336 path = btrfs_alloc_path();
9342 inode = lookup_free_space_inode(tree_root, block_group, path);
9343 if (!IS_ERR(inode)) {
9344 ret = btrfs_orphan_add(trans, inode);
9346 btrfs_add_delayed_iput(inode);
9350 /* One for the block groups ref */
9351 spin_lock(&block_group->lock);
9352 if (block_group->iref) {
9353 block_group->iref = 0;
9354 block_group->inode = NULL;
9355 spin_unlock(&block_group->lock);
9358 spin_unlock(&block_group->lock);
9360 /* One for our lookup ref */
9361 btrfs_add_delayed_iput(inode);
9364 key.objectid = BTRFS_FREE_SPACE_OBJECTID;
9365 key.offset = block_group->key.objectid;
9368 ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
9372 btrfs_release_path(path);
9374 ret = btrfs_del_item(trans, tree_root, path);
9377 btrfs_release_path(path);
9380 spin_lock(&root->fs_info->block_group_cache_lock);
9381 rb_erase(&block_group->cache_node,
9382 &root->fs_info->block_group_cache_tree);
9384 if (root->fs_info->first_logical_byte == block_group->key.objectid)
9385 root->fs_info->first_logical_byte = (u64)-1;
9386 spin_unlock(&root->fs_info->block_group_cache_lock);
9388 down_write(&block_group->space_info->groups_sem);
9390 * we must use list_del_init so people can check to see if they
9391 * are still on the list after taking the semaphore
9393 list_del_init(&block_group->list);
9394 if (list_empty(&block_group->space_info->block_groups[index])) {
9395 kobj = block_group->space_info->block_group_kobjs[index];
9396 block_group->space_info->block_group_kobjs[index] = NULL;
9397 clear_avail_alloc_bits(root->fs_info, block_group->flags);
9399 up_write(&block_group->space_info->groups_sem);
9405 if (block_group->cached == BTRFS_CACHE_STARTED)
9406 wait_block_group_cache_done(block_group);
9408 btrfs_remove_free_space_cache(block_group);
9410 spin_lock(&block_group->space_info->lock);
9411 block_group->space_info->total_bytes -= block_group->key.offset;
9412 block_group->space_info->bytes_readonly -= block_group->key.offset;
9413 block_group->space_info->disk_total -= block_group->key.offset * factor;
9414 spin_unlock(&block_group->space_info->lock);
9416 memcpy(&key, &block_group->key, sizeof(key));
9418 btrfs_put_block_group(block_group);
9419 btrfs_put_block_group(block_group);
9421 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
9427 ret = btrfs_del_item(trans, root, path);
9429 btrfs_free_path(path);
9433 int btrfs_init_space_info(struct btrfs_fs_info *fs_info)
9435 struct btrfs_space_info *space_info;
9436 struct btrfs_super_block *disk_super;
9442 disk_super = fs_info->super_copy;
9443 if (!btrfs_super_root(disk_super))
9446 features = btrfs_super_incompat_flags(disk_super);
9447 if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
9450 flags = BTRFS_BLOCK_GROUP_SYSTEM;
9451 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9456 flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA;
9457 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9459 flags = BTRFS_BLOCK_GROUP_METADATA;
9460 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9464 flags = BTRFS_BLOCK_GROUP_DATA;
9465 ret = update_space_info(fs_info, flags, 0, 0, &space_info);
9471 int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
9473 return unpin_extent_range(root, start, end);
9476 int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr,
9477 u64 num_bytes, u64 *actual_bytes)
9479 return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes);
9482 int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range)
9484 struct btrfs_fs_info *fs_info = root->fs_info;
9485 struct btrfs_block_group_cache *cache = NULL;
9490 u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy);
9494 * try to trim all FS space, our block group may start from non-zero.
9496 if (range->len == total_bytes)
9497 cache = btrfs_lookup_first_block_group(fs_info, range->start);
9499 cache = btrfs_lookup_block_group(fs_info, range->start);
9502 if (cache->key.objectid >= (range->start + range->len)) {
9503 btrfs_put_block_group(cache);
9507 start = max(range->start, cache->key.objectid);
9508 end = min(range->start + range->len,
9509 cache->key.objectid + cache->key.offset);
9511 if (end - start >= range->minlen) {
9512 if (!block_group_cache_done(cache)) {
9513 ret = cache_block_group(cache, 0);
9515 btrfs_put_block_group(cache);
9518 ret = wait_block_group_cache_done(cache);
9520 btrfs_put_block_group(cache);
9524 ret = btrfs_trim_block_group(cache,
9530 trimmed += group_trimmed;
9532 btrfs_put_block_group(cache);
9537 cache = next_block_group(fs_info->tree_root, cache);
9540 range->len = trimmed;
9545 * btrfs_{start,end}_write() is similar to mnt_{want, drop}_write(),
9546 * they are used to prevent the some tasks writing data into the page cache
9547 * by nocow before the subvolume is snapshoted, but flush the data into
9548 * the disk after the snapshot creation.
9550 void btrfs_end_nocow_write(struct btrfs_root *root)
9552 percpu_counter_dec(&root->subv_writers->counter);
9554 * Make sure counter is updated before we wake up
9558 if (waitqueue_active(&root->subv_writers->wait))
9559 wake_up(&root->subv_writers->wait);
9562 int btrfs_start_nocow_write(struct btrfs_root *root)
9564 if (unlikely(atomic_read(&root->will_be_snapshoted)))
9567 percpu_counter_inc(&root->subv_writers->counter);
9569 * Make sure counter is updated before we check for snapshot creation.
9572 if (unlikely(atomic_read(&root->will_be_snapshoted))) {
9573 btrfs_end_nocow_write(root);