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>
30 #include "print-tree.h"
31 #include "transaction.h"
34 #include "free-space-cache.h"
36 /* control flags for do_chunk_alloc's force field
37 * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk
38 * if we really need one.
40 * CHUNK_ALLOC_FORCE means it must try to allocate one
42 * CHUNK_ALLOC_LIMITED means to only try and allocate one
43 * if we have very few chunks already allocated. This is
44 * used as part of the clustering code to help make sure
45 * we have a good pool of storage to cluster in, without
46 * filling the FS with empty chunks
50 CHUNK_ALLOC_NO_FORCE
= 0,
51 CHUNK_ALLOC_FORCE
= 1,
52 CHUNK_ALLOC_LIMITED
= 2,
56 * Control how reservations are dealt with.
58 * RESERVE_FREE - freeing a reservation.
59 * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for
61 * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update
62 * bytes_may_use as the ENOSPC accounting is done elsewhere
67 RESERVE_ALLOC_NO_ACCOUNT
= 2,
70 static int update_block_group(struct btrfs_trans_handle
*trans
,
71 struct btrfs_root
*root
,
72 u64 bytenr
, u64 num_bytes
, int alloc
);
73 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
74 struct btrfs_root
*root
,
75 u64 bytenr
, u64 num_bytes
, u64 parent
,
76 u64 root_objectid
, u64 owner_objectid
,
77 u64 owner_offset
, int refs_to_drop
,
78 struct btrfs_delayed_extent_op
*extra_op
);
79 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
80 struct extent_buffer
*leaf
,
81 struct btrfs_extent_item
*ei
);
82 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
83 struct btrfs_root
*root
,
84 u64 parent
, u64 root_objectid
,
85 u64 flags
, u64 owner
, u64 offset
,
86 struct btrfs_key
*ins
, int ref_mod
);
87 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
88 struct btrfs_root
*root
,
89 u64 parent
, u64 root_objectid
,
90 u64 flags
, struct btrfs_disk_key
*key
,
91 int level
, struct btrfs_key
*ins
);
92 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
93 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
94 u64 flags
, int force
);
95 static int find_next_key(struct btrfs_path
*path
, int level
,
96 struct btrfs_key
*key
);
97 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
98 int dump_block_groups
);
99 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
100 u64 num_bytes
, int reserve
);
103 block_group_cache_done(struct btrfs_block_group_cache
*cache
)
106 return cache
->cached
== BTRFS_CACHE_FINISHED
;
109 static int block_group_bits(struct btrfs_block_group_cache
*cache
, u64 bits
)
111 return (cache
->flags
& bits
) == bits
;
114 static void btrfs_get_block_group(struct btrfs_block_group_cache
*cache
)
116 atomic_inc(&cache
->count
);
119 void btrfs_put_block_group(struct btrfs_block_group_cache
*cache
)
121 if (atomic_dec_and_test(&cache
->count
)) {
122 WARN_ON(cache
->pinned
> 0);
123 WARN_ON(cache
->reserved
> 0);
124 kfree(cache
->free_space_ctl
);
130 * this adds the block group to the fs_info rb tree for the block group
133 static int btrfs_add_block_group_cache(struct btrfs_fs_info
*info
,
134 struct btrfs_block_group_cache
*block_group
)
137 struct rb_node
*parent
= NULL
;
138 struct btrfs_block_group_cache
*cache
;
140 spin_lock(&info
->block_group_cache_lock
);
141 p
= &info
->block_group_cache_tree
.rb_node
;
145 cache
= rb_entry(parent
, struct btrfs_block_group_cache
,
147 if (block_group
->key
.objectid
< cache
->key
.objectid
) {
149 } else if (block_group
->key
.objectid
> cache
->key
.objectid
) {
152 spin_unlock(&info
->block_group_cache_lock
);
157 rb_link_node(&block_group
->cache_node
, parent
, p
);
158 rb_insert_color(&block_group
->cache_node
,
159 &info
->block_group_cache_tree
);
160 spin_unlock(&info
->block_group_cache_lock
);
166 * This will return the block group at or after bytenr if contains is 0, else
167 * it will return the block group that contains the bytenr
169 static struct btrfs_block_group_cache
*
170 block_group_cache_tree_search(struct btrfs_fs_info
*info
, u64 bytenr
,
173 struct btrfs_block_group_cache
*cache
, *ret
= NULL
;
177 spin_lock(&info
->block_group_cache_lock
);
178 n
= info
->block_group_cache_tree
.rb_node
;
181 cache
= rb_entry(n
, struct btrfs_block_group_cache
,
183 end
= cache
->key
.objectid
+ cache
->key
.offset
- 1;
184 start
= cache
->key
.objectid
;
186 if (bytenr
< start
) {
187 if (!contains
&& (!ret
|| start
< ret
->key
.objectid
))
190 } else if (bytenr
> start
) {
191 if (contains
&& bytenr
<= end
) {
202 btrfs_get_block_group(ret
);
203 spin_unlock(&info
->block_group_cache_lock
);
208 static int add_excluded_extent(struct btrfs_root
*root
,
209 u64 start
, u64 num_bytes
)
211 u64 end
= start
+ num_bytes
- 1;
212 set_extent_bits(&root
->fs_info
->freed_extents
[0],
213 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
214 set_extent_bits(&root
->fs_info
->freed_extents
[1],
215 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
219 static void free_excluded_extents(struct btrfs_root
*root
,
220 struct btrfs_block_group_cache
*cache
)
224 start
= cache
->key
.objectid
;
225 end
= start
+ cache
->key
.offset
- 1;
227 clear_extent_bits(&root
->fs_info
->freed_extents
[0],
228 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
229 clear_extent_bits(&root
->fs_info
->freed_extents
[1],
230 start
, end
, EXTENT_UPTODATE
, GFP_NOFS
);
233 static int exclude_super_stripes(struct btrfs_root
*root
,
234 struct btrfs_block_group_cache
*cache
)
241 if (cache
->key
.objectid
< BTRFS_SUPER_INFO_OFFSET
) {
242 stripe_len
= BTRFS_SUPER_INFO_OFFSET
- cache
->key
.objectid
;
243 cache
->bytes_super
+= stripe_len
;
244 ret
= add_excluded_extent(root
, cache
->key
.objectid
,
249 for (i
= 0; i
< BTRFS_SUPER_MIRROR_MAX
; i
++) {
250 bytenr
= btrfs_sb_offset(i
);
251 ret
= btrfs_rmap_block(&root
->fs_info
->mapping_tree
,
252 cache
->key
.objectid
, bytenr
,
253 0, &logical
, &nr
, &stripe_len
);
257 cache
->bytes_super
+= stripe_len
;
258 ret
= add_excluded_extent(root
, logical
[nr
],
268 static struct btrfs_caching_control
*
269 get_caching_control(struct btrfs_block_group_cache
*cache
)
271 struct btrfs_caching_control
*ctl
;
273 spin_lock(&cache
->lock
);
274 if (cache
->cached
!= BTRFS_CACHE_STARTED
) {
275 spin_unlock(&cache
->lock
);
279 /* We're loading it the fast way, so we don't have a caching_ctl. */
280 if (!cache
->caching_ctl
) {
281 spin_unlock(&cache
->lock
);
285 ctl
= cache
->caching_ctl
;
286 atomic_inc(&ctl
->count
);
287 spin_unlock(&cache
->lock
);
291 static void put_caching_control(struct btrfs_caching_control
*ctl
)
293 if (atomic_dec_and_test(&ctl
->count
))
298 * this is only called by cache_block_group, since we could have freed extents
299 * we need to check the pinned_extents for any extents that can't be used yet
300 * since their free space will be released as soon as the transaction commits.
302 static u64
add_new_free_space(struct btrfs_block_group_cache
*block_group
,
303 struct btrfs_fs_info
*info
, u64 start
, u64 end
)
305 u64 extent_start
, extent_end
, size
, total_added
= 0;
308 while (start
< end
) {
309 ret
= find_first_extent_bit(info
->pinned_extents
, start
,
310 &extent_start
, &extent_end
,
311 EXTENT_DIRTY
| EXTENT_UPTODATE
);
315 if (extent_start
<= start
) {
316 start
= extent_end
+ 1;
317 } else if (extent_start
> start
&& extent_start
< end
) {
318 size
= extent_start
- start
;
320 ret
= btrfs_add_free_space(block_group
, start
,
323 start
= extent_end
+ 1;
332 ret
= btrfs_add_free_space(block_group
, start
, size
);
339 static noinline
void caching_thread(struct btrfs_work
*work
)
341 struct btrfs_block_group_cache
*block_group
;
342 struct btrfs_fs_info
*fs_info
;
343 struct btrfs_caching_control
*caching_ctl
;
344 struct btrfs_root
*extent_root
;
345 struct btrfs_path
*path
;
346 struct extent_buffer
*leaf
;
347 struct btrfs_key key
;
353 caching_ctl
= container_of(work
, struct btrfs_caching_control
, work
);
354 block_group
= caching_ctl
->block_group
;
355 fs_info
= block_group
->fs_info
;
356 extent_root
= fs_info
->extent_root
;
358 path
= btrfs_alloc_path();
362 last
= max_t(u64
, block_group
->key
.objectid
, BTRFS_SUPER_INFO_OFFSET
);
365 * We don't want to deadlock with somebody trying to allocate a new
366 * extent for the extent root while also trying to search the extent
367 * root to add free space. So we skip locking and search the commit
368 * root, since its read-only
370 path
->skip_locking
= 1;
371 path
->search_commit_root
= 1;
376 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
378 mutex_lock(&caching_ctl
->mutex
);
379 /* need to make sure the commit_root doesn't disappear */
380 down_read(&fs_info
->extent_commit_sem
);
382 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
386 leaf
= path
->nodes
[0];
387 nritems
= btrfs_header_nritems(leaf
);
390 if (btrfs_fs_closing(fs_info
) > 1) {
395 if (path
->slots
[0] < nritems
) {
396 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
398 ret
= find_next_key(path
, 0, &key
);
402 if (need_resched() ||
403 btrfs_next_leaf(extent_root
, path
)) {
404 caching_ctl
->progress
= last
;
405 btrfs_release_path(path
);
406 up_read(&fs_info
->extent_commit_sem
);
407 mutex_unlock(&caching_ctl
->mutex
);
411 leaf
= path
->nodes
[0];
412 nritems
= btrfs_header_nritems(leaf
);
416 if (key
.objectid
< block_group
->key
.objectid
) {
421 if (key
.objectid
>= block_group
->key
.objectid
+
422 block_group
->key
.offset
)
425 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
) {
426 total_found
+= add_new_free_space(block_group
,
429 last
= key
.objectid
+ key
.offset
;
431 if (total_found
> (1024 * 1024 * 2)) {
433 wake_up(&caching_ctl
->wait
);
440 total_found
+= add_new_free_space(block_group
, fs_info
, last
,
441 block_group
->key
.objectid
+
442 block_group
->key
.offset
);
443 caching_ctl
->progress
= (u64
)-1;
445 spin_lock(&block_group
->lock
);
446 block_group
->caching_ctl
= NULL
;
447 block_group
->cached
= BTRFS_CACHE_FINISHED
;
448 spin_unlock(&block_group
->lock
);
451 btrfs_free_path(path
);
452 up_read(&fs_info
->extent_commit_sem
);
454 free_excluded_extents(extent_root
, block_group
);
456 mutex_unlock(&caching_ctl
->mutex
);
458 wake_up(&caching_ctl
->wait
);
460 put_caching_control(caching_ctl
);
461 btrfs_put_block_group(block_group
);
464 static int cache_block_group(struct btrfs_block_group_cache
*cache
,
465 struct btrfs_trans_handle
*trans
,
466 struct btrfs_root
*root
,
469 struct btrfs_fs_info
*fs_info
= cache
->fs_info
;
470 struct btrfs_caching_control
*caching_ctl
;
474 if (cache
->cached
!= BTRFS_CACHE_NO
)
478 * We can't do the read from on-disk cache during a commit since we need
479 * to have the normal tree locking. Also if we are currently trying to
480 * allocate blocks for the tree root we can't do the fast caching since
481 * we likely hold important locks.
483 if (trans
&& (!trans
->transaction
->in_commit
) &&
484 (root
&& root
!= root
->fs_info
->tree_root
) &&
485 btrfs_test_opt(root
, SPACE_CACHE
)) {
486 spin_lock(&cache
->lock
);
487 if (cache
->cached
!= BTRFS_CACHE_NO
) {
488 spin_unlock(&cache
->lock
);
491 cache
->cached
= BTRFS_CACHE_STARTED
;
492 spin_unlock(&cache
->lock
);
494 ret
= load_free_space_cache(fs_info
, cache
);
496 spin_lock(&cache
->lock
);
498 cache
->cached
= BTRFS_CACHE_FINISHED
;
499 cache
->last_byte_to_unpin
= (u64
)-1;
501 cache
->cached
= BTRFS_CACHE_NO
;
503 spin_unlock(&cache
->lock
);
505 free_excluded_extents(fs_info
->extent_root
, cache
);
513 caching_ctl
= kzalloc(sizeof(*caching_ctl
), GFP_NOFS
);
514 BUG_ON(!caching_ctl
);
516 INIT_LIST_HEAD(&caching_ctl
->list
);
517 mutex_init(&caching_ctl
->mutex
);
518 init_waitqueue_head(&caching_ctl
->wait
);
519 caching_ctl
->block_group
= cache
;
520 caching_ctl
->progress
= cache
->key
.objectid
;
521 /* one for caching kthread, one for caching block group list */
522 atomic_set(&caching_ctl
->count
, 2);
523 caching_ctl
->work
.func
= caching_thread
;
525 spin_lock(&cache
->lock
);
526 if (cache
->cached
!= BTRFS_CACHE_NO
) {
527 spin_unlock(&cache
->lock
);
531 cache
->caching_ctl
= caching_ctl
;
532 cache
->cached
= BTRFS_CACHE_STARTED
;
533 spin_unlock(&cache
->lock
);
535 down_write(&fs_info
->extent_commit_sem
);
536 list_add_tail(&caching_ctl
->list
, &fs_info
->caching_block_groups
);
537 up_write(&fs_info
->extent_commit_sem
);
539 btrfs_get_block_group(cache
);
541 btrfs_queue_worker(&fs_info
->caching_workers
, &caching_ctl
->work
);
547 * return the block group that starts at or after bytenr
549 static struct btrfs_block_group_cache
*
550 btrfs_lookup_first_block_group(struct btrfs_fs_info
*info
, u64 bytenr
)
552 struct btrfs_block_group_cache
*cache
;
554 cache
= block_group_cache_tree_search(info
, bytenr
, 0);
560 * return the block group that contains the given bytenr
562 struct btrfs_block_group_cache
*btrfs_lookup_block_group(
563 struct btrfs_fs_info
*info
,
566 struct btrfs_block_group_cache
*cache
;
568 cache
= block_group_cache_tree_search(info
, bytenr
, 1);
573 static struct btrfs_space_info
*__find_space_info(struct btrfs_fs_info
*info
,
576 struct list_head
*head
= &info
->space_info
;
577 struct btrfs_space_info
*found
;
579 flags
&= BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_SYSTEM
|
580 BTRFS_BLOCK_GROUP_METADATA
;
583 list_for_each_entry_rcu(found
, head
, list
) {
584 if (found
->flags
& flags
) {
594 * after adding space to the filesystem, we need to clear the full flags
595 * on all the space infos.
597 void btrfs_clear_space_info_full(struct btrfs_fs_info
*info
)
599 struct list_head
*head
= &info
->space_info
;
600 struct btrfs_space_info
*found
;
603 list_for_each_entry_rcu(found
, head
, list
)
608 static u64
div_factor(u64 num
, int factor
)
617 static u64
div_factor_fine(u64 num
, int factor
)
626 u64
btrfs_find_block_group(struct btrfs_root
*root
,
627 u64 search_start
, u64 search_hint
, int owner
)
629 struct btrfs_block_group_cache
*cache
;
631 u64 last
= max(search_hint
, search_start
);
638 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
642 spin_lock(&cache
->lock
);
643 last
= cache
->key
.objectid
+ cache
->key
.offset
;
644 used
= btrfs_block_group_used(&cache
->item
);
646 if ((full_search
|| !cache
->ro
) &&
647 block_group_bits(cache
, BTRFS_BLOCK_GROUP_METADATA
)) {
648 if (used
+ cache
->pinned
+ cache
->reserved
<
649 div_factor(cache
->key
.offset
, factor
)) {
650 group_start
= cache
->key
.objectid
;
651 spin_unlock(&cache
->lock
);
652 btrfs_put_block_group(cache
);
656 spin_unlock(&cache
->lock
);
657 btrfs_put_block_group(cache
);
665 if (!full_search
&& factor
< 10) {
675 /* simple helper to search for an existing extent at a given offset */
676 int btrfs_lookup_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
679 struct btrfs_key key
;
680 struct btrfs_path
*path
;
682 path
= btrfs_alloc_path();
686 key
.objectid
= start
;
688 btrfs_set_key_type(&key
, BTRFS_EXTENT_ITEM_KEY
);
689 ret
= btrfs_search_slot(NULL
, root
->fs_info
->extent_root
, &key
, path
,
691 btrfs_free_path(path
);
696 * helper function to lookup reference count and flags of extent.
698 * the head node for delayed ref is used to store the sum of all the
699 * reference count modifications queued up in the rbtree. the head
700 * node may also store the extent flags to set. This way you can check
701 * to see what the reference count and extent flags would be if all of
702 * the delayed refs are not processed.
704 int btrfs_lookup_extent_info(struct btrfs_trans_handle
*trans
,
705 struct btrfs_root
*root
, u64 bytenr
,
706 u64 num_bytes
, u64
*refs
, u64
*flags
)
708 struct btrfs_delayed_ref_head
*head
;
709 struct btrfs_delayed_ref_root
*delayed_refs
;
710 struct btrfs_path
*path
;
711 struct btrfs_extent_item
*ei
;
712 struct extent_buffer
*leaf
;
713 struct btrfs_key key
;
719 path
= btrfs_alloc_path();
723 key
.objectid
= bytenr
;
724 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
725 key
.offset
= num_bytes
;
727 path
->skip_locking
= 1;
728 path
->search_commit_root
= 1;
731 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
,
737 leaf
= path
->nodes
[0];
738 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
739 if (item_size
>= sizeof(*ei
)) {
740 ei
= btrfs_item_ptr(leaf
, path
->slots
[0],
741 struct btrfs_extent_item
);
742 num_refs
= btrfs_extent_refs(leaf
, ei
);
743 extent_flags
= btrfs_extent_flags(leaf
, ei
);
745 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
746 struct btrfs_extent_item_v0
*ei0
;
747 BUG_ON(item_size
!= sizeof(*ei0
));
748 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
749 struct btrfs_extent_item_v0
);
750 num_refs
= btrfs_extent_refs_v0(leaf
, ei0
);
751 /* FIXME: this isn't correct for data */
752 extent_flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
757 BUG_ON(num_refs
== 0);
767 delayed_refs
= &trans
->transaction
->delayed_refs
;
768 spin_lock(&delayed_refs
->lock
);
769 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
771 if (!mutex_trylock(&head
->mutex
)) {
772 atomic_inc(&head
->node
.refs
);
773 spin_unlock(&delayed_refs
->lock
);
775 btrfs_release_path(path
);
778 * Mutex was contended, block until it's released and try
781 mutex_lock(&head
->mutex
);
782 mutex_unlock(&head
->mutex
);
783 btrfs_put_delayed_ref(&head
->node
);
786 if (head
->extent_op
&& head
->extent_op
->update_flags
)
787 extent_flags
|= head
->extent_op
->flags_to_set
;
789 BUG_ON(num_refs
== 0);
791 num_refs
+= head
->node
.ref_mod
;
792 mutex_unlock(&head
->mutex
);
794 spin_unlock(&delayed_refs
->lock
);
796 WARN_ON(num_refs
== 0);
800 *flags
= extent_flags
;
802 btrfs_free_path(path
);
807 * Back reference rules. Back refs have three main goals:
809 * 1) differentiate between all holders of references to an extent so that
810 * when a reference is dropped we can make sure it was a valid reference
811 * before freeing the extent.
813 * 2) Provide enough information to quickly find the holders of an extent
814 * if we notice a given block is corrupted or bad.
816 * 3) Make it easy to migrate blocks for FS shrinking or storage pool
817 * maintenance. This is actually the same as #2, but with a slightly
818 * different use case.
820 * There are two kinds of back refs. The implicit back refs is optimized
821 * for pointers in non-shared tree blocks. For a given pointer in a block,
822 * back refs of this kind provide information about the block's owner tree
823 * and the pointer's key. These information allow us to find the block by
824 * b-tree searching. The full back refs is for pointers in tree blocks not
825 * referenced by their owner trees. The location of tree block is recorded
826 * in the back refs. Actually the full back refs is generic, and can be
827 * used in all cases the implicit back refs is used. The major shortcoming
828 * of the full back refs is its overhead. Every time a tree block gets
829 * COWed, we have to update back refs entry for all pointers in it.
831 * For a newly allocated tree block, we use implicit back refs for
832 * pointers in it. This means most tree related operations only involve
833 * implicit back refs. For a tree block created in old transaction, the
834 * only way to drop a reference to it is COW it. So we can detect the
835 * event that tree block loses its owner tree's reference and do the
836 * back refs conversion.
838 * When a tree block is COW'd through a tree, there are four cases:
840 * The reference count of the block is one and the tree is the block's
841 * owner tree. Nothing to do in this case.
843 * The reference count of the block is one and the tree is not the
844 * block's owner tree. In this case, full back refs is used for pointers
845 * in the block. Remove these full back refs, add implicit back refs for
846 * every pointers in the new block.
848 * The reference count of the block is greater than one and the tree is
849 * the block's owner tree. In this case, implicit back refs is used for
850 * pointers in the block. Add full back refs for every pointers in the
851 * block, increase lower level extents' reference counts. The original
852 * implicit back refs are entailed to the new block.
854 * The reference count of the block is greater than one and the tree is
855 * not the block's owner tree. Add implicit back refs for every pointer in
856 * the new block, increase lower level extents' reference count.
858 * Back Reference Key composing:
860 * The key objectid corresponds to the first byte in the extent,
861 * The key type is used to differentiate between types of back refs.
862 * There are different meanings of the key offset for different types
865 * File extents can be referenced by:
867 * - multiple snapshots, subvolumes, or different generations in one subvol
868 * - different files inside a single subvolume
869 * - different offsets inside a file (bookend extents in file.c)
871 * The extent ref structure for the implicit back refs has fields for:
873 * - Objectid of the subvolume root
874 * - objectid of the file holding the reference
875 * - original offset in the file
876 * - how many bookend extents
878 * The key offset for the implicit back refs is hash of the first
881 * The extent ref structure for the full back refs has field for:
883 * - number of pointers in the tree leaf
885 * The key offset for the implicit back refs is the first byte of
888 * When a file extent is allocated, The implicit back refs is used.
889 * the fields are filled in:
891 * (root_key.objectid, inode objectid, offset in file, 1)
893 * When a file extent is removed file truncation, we find the
894 * corresponding implicit back refs and check the following fields:
896 * (btrfs_header_owner(leaf), inode objectid, offset in file)
898 * Btree extents can be referenced by:
900 * - Different subvolumes
902 * Both the implicit back refs and the full back refs for tree blocks
903 * only consist of key. The key offset for the implicit back refs is
904 * objectid of block's owner tree. The key offset for the full back refs
905 * is the first byte of parent block.
907 * When implicit back refs is used, information about the lowest key and
908 * level of the tree block are required. These information are stored in
909 * tree block info structure.
912 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
913 static int convert_extent_item_v0(struct btrfs_trans_handle
*trans
,
914 struct btrfs_root
*root
,
915 struct btrfs_path
*path
,
916 u64 owner
, u32 extra_size
)
918 struct btrfs_extent_item
*item
;
919 struct btrfs_extent_item_v0
*ei0
;
920 struct btrfs_extent_ref_v0
*ref0
;
921 struct btrfs_tree_block_info
*bi
;
922 struct extent_buffer
*leaf
;
923 struct btrfs_key key
;
924 struct btrfs_key found_key
;
925 u32 new_size
= sizeof(*item
);
929 leaf
= path
->nodes
[0];
930 BUG_ON(btrfs_item_size_nr(leaf
, path
->slots
[0]) != sizeof(*ei0
));
932 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
933 ei0
= btrfs_item_ptr(leaf
, path
->slots
[0],
934 struct btrfs_extent_item_v0
);
935 refs
= btrfs_extent_refs_v0(leaf
, ei0
);
937 if (owner
== (u64
)-1) {
939 if (path
->slots
[0] >= btrfs_header_nritems(leaf
)) {
940 ret
= btrfs_next_leaf(root
, path
);
944 leaf
= path
->nodes
[0];
946 btrfs_item_key_to_cpu(leaf
, &found_key
,
948 BUG_ON(key
.objectid
!= found_key
.objectid
);
949 if (found_key
.type
!= BTRFS_EXTENT_REF_V0_KEY
) {
953 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
954 struct btrfs_extent_ref_v0
);
955 owner
= btrfs_ref_objectid_v0(leaf
, ref0
);
959 btrfs_release_path(path
);
961 if (owner
< BTRFS_FIRST_FREE_OBJECTID
)
962 new_size
+= sizeof(*bi
);
964 new_size
-= sizeof(*ei0
);
965 ret
= btrfs_search_slot(trans
, root
, &key
, path
,
966 new_size
+ extra_size
, 1);
971 ret
= btrfs_extend_item(trans
, root
, path
, new_size
);
973 leaf
= path
->nodes
[0];
974 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
975 btrfs_set_extent_refs(leaf
, item
, refs
);
976 /* FIXME: get real generation */
977 btrfs_set_extent_generation(leaf
, item
, 0);
978 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
979 btrfs_set_extent_flags(leaf
, item
,
980 BTRFS_EXTENT_FLAG_TREE_BLOCK
|
981 BTRFS_BLOCK_FLAG_FULL_BACKREF
);
982 bi
= (struct btrfs_tree_block_info
*)(item
+ 1);
983 /* FIXME: get first key of the block */
984 memset_extent_buffer(leaf
, 0, (unsigned long)bi
, sizeof(*bi
));
985 btrfs_set_tree_block_level(leaf
, bi
, (int)owner
);
987 btrfs_set_extent_flags(leaf
, item
, BTRFS_EXTENT_FLAG_DATA
);
989 btrfs_mark_buffer_dirty(leaf
);
994 static u64
hash_extent_data_ref(u64 root_objectid
, u64 owner
, u64 offset
)
996 u32 high_crc
= ~(u32
)0;
997 u32 low_crc
= ~(u32
)0;
1000 lenum
= cpu_to_le64(root_objectid
);
1001 high_crc
= crc32c(high_crc
, &lenum
, sizeof(lenum
));
1002 lenum
= cpu_to_le64(owner
);
1003 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1004 lenum
= cpu_to_le64(offset
);
1005 low_crc
= crc32c(low_crc
, &lenum
, sizeof(lenum
));
1007 return ((u64
)high_crc
<< 31) ^ (u64
)low_crc
;
1010 static u64
hash_extent_data_ref_item(struct extent_buffer
*leaf
,
1011 struct btrfs_extent_data_ref
*ref
)
1013 return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf
, ref
),
1014 btrfs_extent_data_ref_objectid(leaf
, ref
),
1015 btrfs_extent_data_ref_offset(leaf
, ref
));
1018 static int match_extent_data_ref(struct extent_buffer
*leaf
,
1019 struct btrfs_extent_data_ref
*ref
,
1020 u64 root_objectid
, u64 owner
, u64 offset
)
1022 if (btrfs_extent_data_ref_root(leaf
, ref
) != root_objectid
||
1023 btrfs_extent_data_ref_objectid(leaf
, ref
) != owner
||
1024 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
1029 static noinline
int lookup_extent_data_ref(struct btrfs_trans_handle
*trans
,
1030 struct btrfs_root
*root
,
1031 struct btrfs_path
*path
,
1032 u64 bytenr
, u64 parent
,
1034 u64 owner
, u64 offset
)
1036 struct btrfs_key key
;
1037 struct btrfs_extent_data_ref
*ref
;
1038 struct extent_buffer
*leaf
;
1044 key
.objectid
= bytenr
;
1046 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1047 key
.offset
= parent
;
1049 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1050 key
.offset
= hash_extent_data_ref(root_objectid
,
1055 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1064 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1065 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1066 btrfs_release_path(path
);
1067 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1078 leaf
= path
->nodes
[0];
1079 nritems
= btrfs_header_nritems(leaf
);
1081 if (path
->slots
[0] >= nritems
) {
1082 ret
= btrfs_next_leaf(root
, path
);
1088 leaf
= path
->nodes
[0];
1089 nritems
= btrfs_header_nritems(leaf
);
1093 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1094 if (key
.objectid
!= bytenr
||
1095 key
.type
!= BTRFS_EXTENT_DATA_REF_KEY
)
1098 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1099 struct btrfs_extent_data_ref
);
1101 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1104 btrfs_release_path(path
);
1116 static noinline
int insert_extent_data_ref(struct btrfs_trans_handle
*trans
,
1117 struct btrfs_root
*root
,
1118 struct btrfs_path
*path
,
1119 u64 bytenr
, u64 parent
,
1120 u64 root_objectid
, u64 owner
,
1121 u64 offset
, int refs_to_add
)
1123 struct btrfs_key key
;
1124 struct extent_buffer
*leaf
;
1129 key
.objectid
= bytenr
;
1131 key
.type
= BTRFS_SHARED_DATA_REF_KEY
;
1132 key
.offset
= parent
;
1133 size
= sizeof(struct btrfs_shared_data_ref
);
1135 key
.type
= BTRFS_EXTENT_DATA_REF_KEY
;
1136 key
.offset
= hash_extent_data_ref(root_objectid
,
1138 size
= sizeof(struct btrfs_extent_data_ref
);
1141 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, size
);
1142 if (ret
&& ret
!= -EEXIST
)
1145 leaf
= path
->nodes
[0];
1147 struct btrfs_shared_data_ref
*ref
;
1148 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1149 struct btrfs_shared_data_ref
);
1151 btrfs_set_shared_data_ref_count(leaf
, ref
, refs_to_add
);
1153 num_refs
= btrfs_shared_data_ref_count(leaf
, ref
);
1154 num_refs
+= refs_to_add
;
1155 btrfs_set_shared_data_ref_count(leaf
, ref
, num_refs
);
1158 struct btrfs_extent_data_ref
*ref
;
1159 while (ret
== -EEXIST
) {
1160 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1161 struct btrfs_extent_data_ref
);
1162 if (match_extent_data_ref(leaf
, ref
, root_objectid
,
1165 btrfs_release_path(path
);
1167 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
,
1169 if (ret
&& ret
!= -EEXIST
)
1172 leaf
= path
->nodes
[0];
1174 ref
= btrfs_item_ptr(leaf
, path
->slots
[0],
1175 struct btrfs_extent_data_ref
);
1177 btrfs_set_extent_data_ref_root(leaf
, ref
,
1179 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
1180 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
1181 btrfs_set_extent_data_ref_count(leaf
, ref
, refs_to_add
);
1183 num_refs
= btrfs_extent_data_ref_count(leaf
, ref
);
1184 num_refs
+= refs_to_add
;
1185 btrfs_set_extent_data_ref_count(leaf
, ref
, num_refs
);
1188 btrfs_mark_buffer_dirty(leaf
);
1191 btrfs_release_path(path
);
1195 static noinline
int remove_extent_data_ref(struct btrfs_trans_handle
*trans
,
1196 struct btrfs_root
*root
,
1197 struct btrfs_path
*path
,
1200 struct btrfs_key key
;
1201 struct btrfs_extent_data_ref
*ref1
= NULL
;
1202 struct btrfs_shared_data_ref
*ref2
= NULL
;
1203 struct extent_buffer
*leaf
;
1207 leaf
= path
->nodes
[0];
1208 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1210 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1211 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1212 struct btrfs_extent_data_ref
);
1213 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1214 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1215 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1216 struct btrfs_shared_data_ref
);
1217 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1218 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1219 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1220 struct btrfs_extent_ref_v0
*ref0
;
1221 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1222 struct btrfs_extent_ref_v0
);
1223 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1229 BUG_ON(num_refs
< refs_to_drop
);
1230 num_refs
-= refs_to_drop
;
1232 if (num_refs
== 0) {
1233 ret
= btrfs_del_item(trans
, root
, path
);
1235 if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
)
1236 btrfs_set_extent_data_ref_count(leaf
, ref1
, num_refs
);
1237 else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
)
1238 btrfs_set_shared_data_ref_count(leaf
, ref2
, num_refs
);
1239 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1241 struct btrfs_extent_ref_v0
*ref0
;
1242 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1243 struct btrfs_extent_ref_v0
);
1244 btrfs_set_ref_count_v0(leaf
, ref0
, num_refs
);
1247 btrfs_mark_buffer_dirty(leaf
);
1252 static noinline u32
extent_data_ref_count(struct btrfs_root
*root
,
1253 struct btrfs_path
*path
,
1254 struct btrfs_extent_inline_ref
*iref
)
1256 struct btrfs_key key
;
1257 struct extent_buffer
*leaf
;
1258 struct btrfs_extent_data_ref
*ref1
;
1259 struct btrfs_shared_data_ref
*ref2
;
1262 leaf
= path
->nodes
[0];
1263 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
1265 if (btrfs_extent_inline_ref_type(leaf
, iref
) ==
1266 BTRFS_EXTENT_DATA_REF_KEY
) {
1267 ref1
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1268 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1270 ref2
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1271 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1273 } else if (key
.type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1274 ref1
= btrfs_item_ptr(leaf
, path
->slots
[0],
1275 struct btrfs_extent_data_ref
);
1276 num_refs
= btrfs_extent_data_ref_count(leaf
, ref1
);
1277 } else if (key
.type
== BTRFS_SHARED_DATA_REF_KEY
) {
1278 ref2
= btrfs_item_ptr(leaf
, path
->slots
[0],
1279 struct btrfs_shared_data_ref
);
1280 num_refs
= btrfs_shared_data_ref_count(leaf
, ref2
);
1281 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1282 } else if (key
.type
== BTRFS_EXTENT_REF_V0_KEY
) {
1283 struct btrfs_extent_ref_v0
*ref0
;
1284 ref0
= btrfs_item_ptr(leaf
, path
->slots
[0],
1285 struct btrfs_extent_ref_v0
);
1286 num_refs
= btrfs_ref_count_v0(leaf
, ref0
);
1294 static noinline
int lookup_tree_block_ref(struct btrfs_trans_handle
*trans
,
1295 struct btrfs_root
*root
,
1296 struct btrfs_path
*path
,
1297 u64 bytenr
, u64 parent
,
1300 struct btrfs_key key
;
1303 key
.objectid
= bytenr
;
1305 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1306 key
.offset
= parent
;
1308 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1309 key
.offset
= root_objectid
;
1312 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1315 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1316 if (ret
== -ENOENT
&& parent
) {
1317 btrfs_release_path(path
);
1318 key
.type
= BTRFS_EXTENT_REF_V0_KEY
;
1319 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
1327 static noinline
int insert_tree_block_ref(struct btrfs_trans_handle
*trans
,
1328 struct btrfs_root
*root
,
1329 struct btrfs_path
*path
,
1330 u64 bytenr
, u64 parent
,
1333 struct btrfs_key key
;
1336 key
.objectid
= bytenr
;
1338 key
.type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1339 key
.offset
= parent
;
1341 key
.type
= BTRFS_TREE_BLOCK_REF_KEY
;
1342 key
.offset
= root_objectid
;
1345 ret
= btrfs_insert_empty_item(trans
, root
, path
, &key
, 0);
1346 btrfs_release_path(path
);
1350 static inline int extent_ref_type(u64 parent
, u64 owner
)
1353 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1355 type
= BTRFS_SHARED_BLOCK_REF_KEY
;
1357 type
= BTRFS_TREE_BLOCK_REF_KEY
;
1360 type
= BTRFS_SHARED_DATA_REF_KEY
;
1362 type
= BTRFS_EXTENT_DATA_REF_KEY
;
1367 static int find_next_key(struct btrfs_path
*path
, int level
,
1368 struct btrfs_key
*key
)
1371 for (; level
< BTRFS_MAX_LEVEL
; level
++) {
1372 if (!path
->nodes
[level
])
1374 if (path
->slots
[level
] + 1 >=
1375 btrfs_header_nritems(path
->nodes
[level
]))
1378 btrfs_item_key_to_cpu(path
->nodes
[level
], key
,
1379 path
->slots
[level
] + 1);
1381 btrfs_node_key_to_cpu(path
->nodes
[level
], key
,
1382 path
->slots
[level
] + 1);
1389 * look for inline back ref. if back ref is found, *ref_ret is set
1390 * to the address of inline back ref, and 0 is returned.
1392 * if back ref isn't found, *ref_ret is set to the address where it
1393 * should be inserted, and -ENOENT is returned.
1395 * if insert is true and there are too many inline back refs, the path
1396 * points to the extent item, and -EAGAIN is returned.
1398 * NOTE: inline back refs are ordered in the same way that back ref
1399 * items in the tree are ordered.
1401 static noinline_for_stack
1402 int lookup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1403 struct btrfs_root
*root
,
1404 struct btrfs_path
*path
,
1405 struct btrfs_extent_inline_ref
**ref_ret
,
1406 u64 bytenr
, u64 num_bytes
,
1407 u64 parent
, u64 root_objectid
,
1408 u64 owner
, u64 offset
, int insert
)
1410 struct btrfs_key key
;
1411 struct extent_buffer
*leaf
;
1412 struct btrfs_extent_item
*ei
;
1413 struct btrfs_extent_inline_ref
*iref
;
1424 key
.objectid
= bytenr
;
1425 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1426 key
.offset
= num_bytes
;
1428 want
= extent_ref_type(parent
, owner
);
1430 extra_size
= btrfs_extent_inline_ref_size(want
);
1431 path
->keep_locks
= 1;
1434 ret
= btrfs_search_slot(trans
, root
, &key
, path
, extra_size
, 1);
1441 leaf
= path
->nodes
[0];
1442 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1443 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1444 if (item_size
< sizeof(*ei
)) {
1449 ret
= convert_extent_item_v0(trans
, root
, path
, owner
,
1455 leaf
= path
->nodes
[0];
1456 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1459 BUG_ON(item_size
< sizeof(*ei
));
1461 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1462 flags
= btrfs_extent_flags(leaf
, ei
);
1464 ptr
= (unsigned long)(ei
+ 1);
1465 end
= (unsigned long)ei
+ item_size
;
1467 if (flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
) {
1468 ptr
+= sizeof(struct btrfs_tree_block_info
);
1471 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_DATA
));
1480 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1481 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1485 ptr
+= btrfs_extent_inline_ref_size(type
);
1489 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1490 struct btrfs_extent_data_ref
*dref
;
1491 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1492 if (match_extent_data_ref(leaf
, dref
, root_objectid
,
1497 if (hash_extent_data_ref_item(leaf
, dref
) <
1498 hash_extent_data_ref(root_objectid
, owner
, offset
))
1502 ref_offset
= btrfs_extent_inline_ref_offset(leaf
, iref
);
1504 if (parent
== ref_offset
) {
1508 if (ref_offset
< parent
)
1511 if (root_objectid
== ref_offset
) {
1515 if (ref_offset
< root_objectid
)
1519 ptr
+= btrfs_extent_inline_ref_size(type
);
1521 if (err
== -ENOENT
&& insert
) {
1522 if (item_size
+ extra_size
>=
1523 BTRFS_MAX_EXTENT_ITEM_SIZE(root
)) {
1528 * To add new inline back ref, we have to make sure
1529 * there is no corresponding back ref item.
1530 * For simplicity, we just do not add new inline back
1531 * ref if there is any kind of item for this block
1533 if (find_next_key(path
, 0, &key
) == 0 &&
1534 key
.objectid
== bytenr
&&
1535 key
.type
< BTRFS_BLOCK_GROUP_ITEM_KEY
) {
1540 *ref_ret
= (struct btrfs_extent_inline_ref
*)ptr
;
1543 path
->keep_locks
= 0;
1544 btrfs_unlock_up_safe(path
, 1);
1550 * helper to add new inline back ref
1552 static noinline_for_stack
1553 int setup_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1554 struct btrfs_root
*root
,
1555 struct btrfs_path
*path
,
1556 struct btrfs_extent_inline_ref
*iref
,
1557 u64 parent
, u64 root_objectid
,
1558 u64 owner
, u64 offset
, int refs_to_add
,
1559 struct btrfs_delayed_extent_op
*extent_op
)
1561 struct extent_buffer
*leaf
;
1562 struct btrfs_extent_item
*ei
;
1565 unsigned long item_offset
;
1571 leaf
= path
->nodes
[0];
1572 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1573 item_offset
= (unsigned long)iref
- (unsigned long)ei
;
1575 type
= extent_ref_type(parent
, owner
);
1576 size
= btrfs_extent_inline_ref_size(type
);
1578 ret
= btrfs_extend_item(trans
, root
, path
, size
);
1580 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1581 refs
= btrfs_extent_refs(leaf
, ei
);
1582 refs
+= refs_to_add
;
1583 btrfs_set_extent_refs(leaf
, ei
, refs
);
1585 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1587 ptr
= (unsigned long)ei
+ item_offset
;
1588 end
= (unsigned long)ei
+ btrfs_item_size_nr(leaf
, path
->slots
[0]);
1589 if (ptr
< end
- size
)
1590 memmove_extent_buffer(leaf
, ptr
+ size
, ptr
,
1593 iref
= (struct btrfs_extent_inline_ref
*)ptr
;
1594 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
1595 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1596 struct btrfs_extent_data_ref
*dref
;
1597 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1598 btrfs_set_extent_data_ref_root(leaf
, dref
, root_objectid
);
1599 btrfs_set_extent_data_ref_objectid(leaf
, dref
, owner
);
1600 btrfs_set_extent_data_ref_offset(leaf
, dref
, offset
);
1601 btrfs_set_extent_data_ref_count(leaf
, dref
, refs_to_add
);
1602 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1603 struct btrfs_shared_data_ref
*sref
;
1604 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1605 btrfs_set_shared_data_ref_count(leaf
, sref
, refs_to_add
);
1606 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1607 } else if (type
== BTRFS_SHARED_BLOCK_REF_KEY
) {
1608 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
1610 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
1612 btrfs_mark_buffer_dirty(leaf
);
1616 static int lookup_extent_backref(struct btrfs_trans_handle
*trans
,
1617 struct btrfs_root
*root
,
1618 struct btrfs_path
*path
,
1619 struct btrfs_extent_inline_ref
**ref_ret
,
1620 u64 bytenr
, u64 num_bytes
, u64 parent
,
1621 u64 root_objectid
, u64 owner
, u64 offset
)
1625 ret
= lookup_inline_extent_backref(trans
, root
, path
, ref_ret
,
1626 bytenr
, num_bytes
, parent
,
1627 root_objectid
, owner
, offset
, 0);
1631 btrfs_release_path(path
);
1634 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1635 ret
= lookup_tree_block_ref(trans
, root
, path
, bytenr
, parent
,
1638 ret
= lookup_extent_data_ref(trans
, root
, path
, bytenr
, parent
,
1639 root_objectid
, owner
, offset
);
1645 * helper to update/remove inline back ref
1647 static noinline_for_stack
1648 int update_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1649 struct btrfs_root
*root
,
1650 struct btrfs_path
*path
,
1651 struct btrfs_extent_inline_ref
*iref
,
1653 struct btrfs_delayed_extent_op
*extent_op
)
1655 struct extent_buffer
*leaf
;
1656 struct btrfs_extent_item
*ei
;
1657 struct btrfs_extent_data_ref
*dref
= NULL
;
1658 struct btrfs_shared_data_ref
*sref
= NULL
;
1667 leaf
= path
->nodes
[0];
1668 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1669 refs
= btrfs_extent_refs(leaf
, ei
);
1670 WARN_ON(refs_to_mod
< 0 && refs
+ refs_to_mod
<= 0);
1671 refs
+= refs_to_mod
;
1672 btrfs_set_extent_refs(leaf
, ei
, refs
);
1674 __run_delayed_extent_op(extent_op
, leaf
, ei
);
1676 type
= btrfs_extent_inline_ref_type(leaf
, iref
);
1678 if (type
== BTRFS_EXTENT_DATA_REF_KEY
) {
1679 dref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
1680 refs
= btrfs_extent_data_ref_count(leaf
, dref
);
1681 } else if (type
== BTRFS_SHARED_DATA_REF_KEY
) {
1682 sref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
1683 refs
= btrfs_shared_data_ref_count(leaf
, sref
);
1686 BUG_ON(refs_to_mod
!= -1);
1689 BUG_ON(refs_to_mod
< 0 && refs
< -refs_to_mod
);
1690 refs
+= refs_to_mod
;
1693 if (type
== BTRFS_EXTENT_DATA_REF_KEY
)
1694 btrfs_set_extent_data_ref_count(leaf
, dref
, refs
);
1696 btrfs_set_shared_data_ref_count(leaf
, sref
, refs
);
1698 size
= btrfs_extent_inline_ref_size(type
);
1699 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
1700 ptr
= (unsigned long)iref
;
1701 end
= (unsigned long)ei
+ item_size
;
1702 if (ptr
+ size
< end
)
1703 memmove_extent_buffer(leaf
, ptr
, ptr
+ size
,
1706 ret
= btrfs_truncate_item(trans
, root
, path
, item_size
, 1);
1708 btrfs_mark_buffer_dirty(leaf
);
1712 static noinline_for_stack
1713 int insert_inline_extent_backref(struct btrfs_trans_handle
*trans
,
1714 struct btrfs_root
*root
,
1715 struct btrfs_path
*path
,
1716 u64 bytenr
, u64 num_bytes
, u64 parent
,
1717 u64 root_objectid
, u64 owner
,
1718 u64 offset
, int refs_to_add
,
1719 struct btrfs_delayed_extent_op
*extent_op
)
1721 struct btrfs_extent_inline_ref
*iref
;
1724 ret
= lookup_inline_extent_backref(trans
, root
, path
, &iref
,
1725 bytenr
, num_bytes
, parent
,
1726 root_objectid
, owner
, offset
, 1);
1728 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
);
1729 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1730 refs_to_add
, extent_op
);
1731 } else if (ret
== -ENOENT
) {
1732 ret
= setup_inline_extent_backref(trans
, root
, path
, iref
,
1733 parent
, root_objectid
,
1734 owner
, offset
, refs_to_add
,
1740 static int insert_extent_backref(struct btrfs_trans_handle
*trans
,
1741 struct btrfs_root
*root
,
1742 struct btrfs_path
*path
,
1743 u64 bytenr
, u64 parent
, u64 root_objectid
,
1744 u64 owner
, u64 offset
, int refs_to_add
)
1747 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1748 BUG_ON(refs_to_add
!= 1);
1749 ret
= insert_tree_block_ref(trans
, root
, path
, bytenr
,
1750 parent
, root_objectid
);
1752 ret
= insert_extent_data_ref(trans
, root
, path
, bytenr
,
1753 parent
, root_objectid
,
1754 owner
, offset
, refs_to_add
);
1759 static int remove_extent_backref(struct btrfs_trans_handle
*trans
,
1760 struct btrfs_root
*root
,
1761 struct btrfs_path
*path
,
1762 struct btrfs_extent_inline_ref
*iref
,
1763 int refs_to_drop
, int is_data
)
1767 BUG_ON(!is_data
&& refs_to_drop
!= 1);
1769 ret
= update_inline_extent_backref(trans
, root
, path
, iref
,
1770 -refs_to_drop
, NULL
);
1771 } else if (is_data
) {
1772 ret
= remove_extent_data_ref(trans
, root
, path
, refs_to_drop
);
1774 ret
= btrfs_del_item(trans
, root
, path
);
1779 static int btrfs_issue_discard(struct block_device
*bdev
,
1782 return blkdev_issue_discard(bdev
, start
>> 9, len
>> 9, GFP_NOFS
, 0);
1785 static int btrfs_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
1786 u64 num_bytes
, u64
*actual_bytes
)
1789 u64 discarded_bytes
= 0;
1790 struct btrfs_multi_bio
*multi
= NULL
;
1793 /* Tell the block device(s) that the sectors can be discarded */
1794 ret
= btrfs_map_block(&root
->fs_info
->mapping_tree
, REQ_DISCARD
,
1795 bytenr
, &num_bytes
, &multi
, 0);
1797 struct btrfs_bio_stripe
*stripe
= multi
->stripes
;
1801 for (i
= 0; i
< multi
->num_stripes
; i
++, stripe
++) {
1802 if (!stripe
->dev
->can_discard
)
1805 ret
= btrfs_issue_discard(stripe
->dev
->bdev
,
1809 discarded_bytes
+= stripe
->length
;
1810 else if (ret
!= -EOPNOTSUPP
)
1814 * Just in case we get back EOPNOTSUPP for some reason,
1815 * just ignore the return value so we don't screw up
1816 * people calling discard_extent.
1824 *actual_bytes
= discarded_bytes
;
1830 int btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1831 struct btrfs_root
*root
,
1832 u64 bytenr
, u64 num_bytes
, u64 parent
,
1833 u64 root_objectid
, u64 owner
, u64 offset
)
1836 BUG_ON(owner
< BTRFS_FIRST_FREE_OBJECTID
&&
1837 root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
1839 if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
1840 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
1841 parent
, root_objectid
, (int)owner
,
1842 BTRFS_ADD_DELAYED_REF
, NULL
);
1844 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
1845 parent
, root_objectid
, owner
, offset
,
1846 BTRFS_ADD_DELAYED_REF
, NULL
);
1851 static int __btrfs_inc_extent_ref(struct btrfs_trans_handle
*trans
,
1852 struct btrfs_root
*root
,
1853 u64 bytenr
, u64 num_bytes
,
1854 u64 parent
, u64 root_objectid
,
1855 u64 owner
, u64 offset
, int refs_to_add
,
1856 struct btrfs_delayed_extent_op
*extent_op
)
1858 struct btrfs_path
*path
;
1859 struct extent_buffer
*leaf
;
1860 struct btrfs_extent_item
*item
;
1865 path
= btrfs_alloc_path();
1870 path
->leave_spinning
= 1;
1871 /* this will setup the path even if it fails to insert the back ref */
1872 ret
= insert_inline_extent_backref(trans
, root
->fs_info
->extent_root
,
1873 path
, bytenr
, num_bytes
, parent
,
1874 root_objectid
, owner
, offset
,
1875 refs_to_add
, extent_op
);
1879 if (ret
!= -EAGAIN
) {
1884 leaf
= path
->nodes
[0];
1885 item
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
1886 refs
= btrfs_extent_refs(leaf
, item
);
1887 btrfs_set_extent_refs(leaf
, item
, refs
+ refs_to_add
);
1889 __run_delayed_extent_op(extent_op
, leaf
, item
);
1891 btrfs_mark_buffer_dirty(leaf
);
1892 btrfs_release_path(path
);
1895 path
->leave_spinning
= 1;
1897 /* now insert the actual backref */
1898 ret
= insert_extent_backref(trans
, root
->fs_info
->extent_root
,
1899 path
, bytenr
, parent
, root_objectid
,
1900 owner
, offset
, refs_to_add
);
1903 btrfs_free_path(path
);
1907 static int run_delayed_data_ref(struct btrfs_trans_handle
*trans
,
1908 struct btrfs_root
*root
,
1909 struct btrfs_delayed_ref_node
*node
,
1910 struct btrfs_delayed_extent_op
*extent_op
,
1911 int insert_reserved
)
1914 struct btrfs_delayed_data_ref
*ref
;
1915 struct btrfs_key ins
;
1920 ins
.objectid
= node
->bytenr
;
1921 ins
.offset
= node
->num_bytes
;
1922 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
1924 ref
= btrfs_delayed_node_to_data_ref(node
);
1925 if (node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
1926 parent
= ref
->parent
;
1928 ref_root
= ref
->root
;
1930 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
1932 BUG_ON(extent_op
->update_key
);
1933 flags
|= extent_op
->flags_to_set
;
1935 ret
= alloc_reserved_file_extent(trans
, root
,
1936 parent
, ref_root
, flags
,
1937 ref
->objectid
, ref
->offset
,
1938 &ins
, node
->ref_mod
);
1939 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
1940 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
1941 node
->num_bytes
, parent
,
1942 ref_root
, ref
->objectid
,
1943 ref
->offset
, node
->ref_mod
,
1945 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
1946 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
1947 node
->num_bytes
, parent
,
1948 ref_root
, ref
->objectid
,
1949 ref
->offset
, node
->ref_mod
,
1957 static void __run_delayed_extent_op(struct btrfs_delayed_extent_op
*extent_op
,
1958 struct extent_buffer
*leaf
,
1959 struct btrfs_extent_item
*ei
)
1961 u64 flags
= btrfs_extent_flags(leaf
, ei
);
1962 if (extent_op
->update_flags
) {
1963 flags
|= extent_op
->flags_to_set
;
1964 btrfs_set_extent_flags(leaf
, ei
, flags
);
1967 if (extent_op
->update_key
) {
1968 struct btrfs_tree_block_info
*bi
;
1969 BUG_ON(!(flags
& BTRFS_EXTENT_FLAG_TREE_BLOCK
));
1970 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
1971 btrfs_set_tree_block_key(leaf
, bi
, &extent_op
->key
);
1975 static int run_delayed_extent_op(struct btrfs_trans_handle
*trans
,
1976 struct btrfs_root
*root
,
1977 struct btrfs_delayed_ref_node
*node
,
1978 struct btrfs_delayed_extent_op
*extent_op
)
1980 struct btrfs_key key
;
1981 struct btrfs_path
*path
;
1982 struct btrfs_extent_item
*ei
;
1983 struct extent_buffer
*leaf
;
1988 path
= btrfs_alloc_path();
1992 key
.objectid
= node
->bytenr
;
1993 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
1994 key
.offset
= node
->num_bytes
;
1997 path
->leave_spinning
= 1;
1998 ret
= btrfs_search_slot(trans
, root
->fs_info
->extent_root
, &key
,
2009 leaf
= path
->nodes
[0];
2010 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2011 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2012 if (item_size
< sizeof(*ei
)) {
2013 ret
= convert_extent_item_v0(trans
, root
->fs_info
->extent_root
,
2019 leaf
= path
->nodes
[0];
2020 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2023 BUG_ON(item_size
< sizeof(*ei
));
2024 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2025 __run_delayed_extent_op(extent_op
, leaf
, ei
);
2027 btrfs_mark_buffer_dirty(leaf
);
2029 btrfs_free_path(path
);
2033 static int run_delayed_tree_ref(struct btrfs_trans_handle
*trans
,
2034 struct btrfs_root
*root
,
2035 struct btrfs_delayed_ref_node
*node
,
2036 struct btrfs_delayed_extent_op
*extent_op
,
2037 int insert_reserved
)
2040 struct btrfs_delayed_tree_ref
*ref
;
2041 struct btrfs_key ins
;
2045 ins
.objectid
= node
->bytenr
;
2046 ins
.offset
= node
->num_bytes
;
2047 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
2049 ref
= btrfs_delayed_node_to_tree_ref(node
);
2050 if (node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2051 parent
= ref
->parent
;
2053 ref_root
= ref
->root
;
2055 BUG_ON(node
->ref_mod
!= 1);
2056 if (node
->action
== BTRFS_ADD_DELAYED_REF
&& insert_reserved
) {
2057 BUG_ON(!extent_op
|| !extent_op
->update_flags
||
2058 !extent_op
->update_key
);
2059 ret
= alloc_reserved_tree_block(trans
, root
,
2061 extent_op
->flags_to_set
,
2064 } else if (node
->action
== BTRFS_ADD_DELAYED_REF
) {
2065 ret
= __btrfs_inc_extent_ref(trans
, root
, node
->bytenr
,
2066 node
->num_bytes
, parent
, ref_root
,
2067 ref
->level
, 0, 1, extent_op
);
2068 } else if (node
->action
== BTRFS_DROP_DELAYED_REF
) {
2069 ret
= __btrfs_free_extent(trans
, root
, node
->bytenr
,
2070 node
->num_bytes
, parent
, ref_root
,
2071 ref
->level
, 0, 1, extent_op
);
2078 /* helper function to actually process a single delayed ref entry */
2079 static int run_one_delayed_ref(struct btrfs_trans_handle
*trans
,
2080 struct btrfs_root
*root
,
2081 struct btrfs_delayed_ref_node
*node
,
2082 struct btrfs_delayed_extent_op
*extent_op
,
2083 int insert_reserved
)
2086 if (btrfs_delayed_ref_is_head(node
)) {
2087 struct btrfs_delayed_ref_head
*head
;
2089 * we've hit the end of the chain and we were supposed
2090 * to insert this extent into the tree. But, it got
2091 * deleted before we ever needed to insert it, so all
2092 * we have to do is clean up the accounting
2095 head
= btrfs_delayed_node_to_head(node
);
2096 if (insert_reserved
) {
2097 btrfs_pin_extent(root
, node
->bytenr
,
2098 node
->num_bytes
, 1);
2099 if (head
->is_data
) {
2100 ret
= btrfs_del_csums(trans
, root
,
2106 mutex_unlock(&head
->mutex
);
2110 if (node
->type
== BTRFS_TREE_BLOCK_REF_KEY
||
2111 node
->type
== BTRFS_SHARED_BLOCK_REF_KEY
)
2112 ret
= run_delayed_tree_ref(trans
, root
, node
, extent_op
,
2114 else if (node
->type
== BTRFS_EXTENT_DATA_REF_KEY
||
2115 node
->type
== BTRFS_SHARED_DATA_REF_KEY
)
2116 ret
= run_delayed_data_ref(trans
, root
, node
, extent_op
,
2123 static noinline
struct btrfs_delayed_ref_node
*
2124 select_delayed_ref(struct btrfs_delayed_ref_head
*head
)
2126 struct rb_node
*node
;
2127 struct btrfs_delayed_ref_node
*ref
;
2128 int action
= BTRFS_ADD_DELAYED_REF
;
2131 * select delayed ref of type BTRFS_ADD_DELAYED_REF first.
2132 * this prevents ref count from going down to zero when
2133 * there still are pending delayed ref.
2135 node
= rb_prev(&head
->node
.rb_node
);
2139 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2141 if (ref
->bytenr
!= head
->node
.bytenr
)
2143 if (ref
->action
== action
)
2145 node
= rb_prev(node
);
2147 if (action
== BTRFS_ADD_DELAYED_REF
) {
2148 action
= BTRFS_DROP_DELAYED_REF
;
2154 static noinline
int run_clustered_refs(struct btrfs_trans_handle
*trans
,
2155 struct btrfs_root
*root
,
2156 struct list_head
*cluster
)
2158 struct btrfs_delayed_ref_root
*delayed_refs
;
2159 struct btrfs_delayed_ref_node
*ref
;
2160 struct btrfs_delayed_ref_head
*locked_ref
= NULL
;
2161 struct btrfs_delayed_extent_op
*extent_op
;
2164 int must_insert_reserved
= 0;
2166 delayed_refs
= &trans
->transaction
->delayed_refs
;
2169 /* pick a new head ref from the cluster list */
2170 if (list_empty(cluster
))
2173 locked_ref
= list_entry(cluster
->next
,
2174 struct btrfs_delayed_ref_head
, cluster
);
2176 /* grab the lock that says we are going to process
2177 * all the refs for this head */
2178 ret
= btrfs_delayed_ref_lock(trans
, locked_ref
);
2181 * we may have dropped the spin lock to get the head
2182 * mutex lock, and that might have given someone else
2183 * time to free the head. If that's true, it has been
2184 * removed from our list and we can move on.
2186 if (ret
== -EAGAIN
) {
2194 * record the must insert reserved flag before we
2195 * drop the spin lock.
2197 must_insert_reserved
= locked_ref
->must_insert_reserved
;
2198 locked_ref
->must_insert_reserved
= 0;
2200 extent_op
= locked_ref
->extent_op
;
2201 locked_ref
->extent_op
= NULL
;
2204 * locked_ref is the head node, so we have to go one
2205 * node back for any delayed ref updates
2207 ref
= select_delayed_ref(locked_ref
);
2209 /* All delayed refs have been processed, Go ahead
2210 * and send the head node to run_one_delayed_ref,
2211 * so that any accounting fixes can happen
2213 ref
= &locked_ref
->node
;
2215 if (extent_op
&& must_insert_reserved
) {
2221 spin_unlock(&delayed_refs
->lock
);
2223 ret
= run_delayed_extent_op(trans
, root
,
2229 spin_lock(&delayed_refs
->lock
);
2233 list_del_init(&locked_ref
->cluster
);
2238 rb_erase(&ref
->rb_node
, &delayed_refs
->root
);
2239 delayed_refs
->num_entries
--;
2241 spin_unlock(&delayed_refs
->lock
);
2243 ret
= run_one_delayed_ref(trans
, root
, ref
, extent_op
,
2244 must_insert_reserved
);
2247 btrfs_put_delayed_ref(ref
);
2252 spin_lock(&delayed_refs
->lock
);
2258 * this starts processing the delayed reference count updates and
2259 * extent insertions we have queued up so far. count can be
2260 * 0, which means to process everything in the tree at the start
2261 * of the run (but not newly added entries), or it can be some target
2262 * number you'd like to process.
2264 int btrfs_run_delayed_refs(struct btrfs_trans_handle
*trans
,
2265 struct btrfs_root
*root
, unsigned long count
)
2267 struct rb_node
*node
;
2268 struct btrfs_delayed_ref_root
*delayed_refs
;
2269 struct btrfs_delayed_ref_node
*ref
;
2270 struct list_head cluster
;
2272 int run_all
= count
== (unsigned long)-1;
2275 if (root
== root
->fs_info
->extent_root
)
2276 root
= root
->fs_info
->tree_root
;
2278 delayed_refs
= &trans
->transaction
->delayed_refs
;
2279 INIT_LIST_HEAD(&cluster
);
2281 spin_lock(&delayed_refs
->lock
);
2283 count
= delayed_refs
->num_entries
* 2;
2287 if (!(run_all
|| run_most
) &&
2288 delayed_refs
->num_heads_ready
< 64)
2292 * go find something we can process in the rbtree. We start at
2293 * the beginning of the tree, and then build a cluster
2294 * of refs to process starting at the first one we are able to
2297 ret
= btrfs_find_ref_cluster(trans
, &cluster
,
2298 delayed_refs
->run_delayed_start
);
2302 ret
= run_clustered_refs(trans
, root
, &cluster
);
2305 count
-= min_t(unsigned long, ret
, count
);
2312 node
= rb_first(&delayed_refs
->root
);
2315 count
= (unsigned long)-1;
2318 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
,
2320 if (btrfs_delayed_ref_is_head(ref
)) {
2321 struct btrfs_delayed_ref_head
*head
;
2323 head
= btrfs_delayed_node_to_head(ref
);
2324 atomic_inc(&ref
->refs
);
2326 spin_unlock(&delayed_refs
->lock
);
2328 * Mutex was contended, block until it's
2329 * released and try again
2331 mutex_lock(&head
->mutex
);
2332 mutex_unlock(&head
->mutex
);
2334 btrfs_put_delayed_ref(ref
);
2338 node
= rb_next(node
);
2340 spin_unlock(&delayed_refs
->lock
);
2341 schedule_timeout(1);
2345 spin_unlock(&delayed_refs
->lock
);
2349 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle
*trans
,
2350 struct btrfs_root
*root
,
2351 u64 bytenr
, u64 num_bytes
, u64 flags
,
2354 struct btrfs_delayed_extent_op
*extent_op
;
2357 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
2361 extent_op
->flags_to_set
= flags
;
2362 extent_op
->update_flags
= 1;
2363 extent_op
->update_key
= 0;
2364 extent_op
->is_data
= is_data
? 1 : 0;
2366 ret
= btrfs_add_delayed_extent_op(trans
, bytenr
, num_bytes
, extent_op
);
2372 static noinline
int check_delayed_ref(struct btrfs_trans_handle
*trans
,
2373 struct btrfs_root
*root
,
2374 struct btrfs_path
*path
,
2375 u64 objectid
, u64 offset
, u64 bytenr
)
2377 struct btrfs_delayed_ref_head
*head
;
2378 struct btrfs_delayed_ref_node
*ref
;
2379 struct btrfs_delayed_data_ref
*data_ref
;
2380 struct btrfs_delayed_ref_root
*delayed_refs
;
2381 struct rb_node
*node
;
2385 delayed_refs
= &trans
->transaction
->delayed_refs
;
2386 spin_lock(&delayed_refs
->lock
);
2387 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
2391 if (!mutex_trylock(&head
->mutex
)) {
2392 atomic_inc(&head
->node
.refs
);
2393 spin_unlock(&delayed_refs
->lock
);
2395 btrfs_release_path(path
);
2398 * Mutex was contended, block until it's released and let
2401 mutex_lock(&head
->mutex
);
2402 mutex_unlock(&head
->mutex
);
2403 btrfs_put_delayed_ref(&head
->node
);
2407 node
= rb_prev(&head
->node
.rb_node
);
2411 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2413 if (ref
->bytenr
!= bytenr
)
2417 if (ref
->type
!= BTRFS_EXTENT_DATA_REF_KEY
)
2420 data_ref
= btrfs_delayed_node_to_data_ref(ref
);
2422 node
= rb_prev(node
);
2424 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
2425 if (ref
->bytenr
== bytenr
)
2429 if (data_ref
->root
!= root
->root_key
.objectid
||
2430 data_ref
->objectid
!= objectid
|| data_ref
->offset
!= offset
)
2435 mutex_unlock(&head
->mutex
);
2437 spin_unlock(&delayed_refs
->lock
);
2441 static noinline
int check_committed_ref(struct btrfs_trans_handle
*trans
,
2442 struct btrfs_root
*root
,
2443 struct btrfs_path
*path
,
2444 u64 objectid
, u64 offset
, u64 bytenr
)
2446 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2447 struct extent_buffer
*leaf
;
2448 struct btrfs_extent_data_ref
*ref
;
2449 struct btrfs_extent_inline_ref
*iref
;
2450 struct btrfs_extent_item
*ei
;
2451 struct btrfs_key key
;
2455 key
.objectid
= bytenr
;
2456 key
.offset
= (u64
)-1;
2457 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
2459 ret
= btrfs_search_slot(NULL
, extent_root
, &key
, path
, 0, 0);
2465 if (path
->slots
[0] == 0)
2469 leaf
= path
->nodes
[0];
2470 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
2472 if (key
.objectid
!= bytenr
|| key
.type
!= BTRFS_EXTENT_ITEM_KEY
)
2476 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2477 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
2478 if (item_size
< sizeof(*ei
)) {
2479 WARN_ON(item_size
!= sizeof(struct btrfs_extent_item_v0
));
2483 ei
= btrfs_item_ptr(leaf
, path
->slots
[0], struct btrfs_extent_item
);
2485 if (item_size
!= sizeof(*ei
) +
2486 btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY
))
2489 if (btrfs_extent_generation(leaf
, ei
) <=
2490 btrfs_root_last_snapshot(&root
->root_item
))
2493 iref
= (struct btrfs_extent_inline_ref
*)(ei
+ 1);
2494 if (btrfs_extent_inline_ref_type(leaf
, iref
) !=
2495 BTRFS_EXTENT_DATA_REF_KEY
)
2498 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
2499 if (btrfs_extent_refs(leaf
, ei
) !=
2500 btrfs_extent_data_ref_count(leaf
, ref
) ||
2501 btrfs_extent_data_ref_root(leaf
, ref
) !=
2502 root
->root_key
.objectid
||
2503 btrfs_extent_data_ref_objectid(leaf
, ref
) != objectid
||
2504 btrfs_extent_data_ref_offset(leaf
, ref
) != offset
)
2512 int btrfs_cross_ref_exist(struct btrfs_trans_handle
*trans
,
2513 struct btrfs_root
*root
,
2514 u64 objectid
, u64 offset
, u64 bytenr
)
2516 struct btrfs_path
*path
;
2520 path
= btrfs_alloc_path();
2525 ret
= check_committed_ref(trans
, root
, path
, objectid
,
2527 if (ret
&& ret
!= -ENOENT
)
2530 ret2
= check_delayed_ref(trans
, root
, path
, objectid
,
2532 } while (ret2
== -EAGAIN
);
2534 if (ret2
&& ret2
!= -ENOENT
) {
2539 if (ret
!= -ENOENT
|| ret2
!= -ENOENT
)
2542 btrfs_free_path(path
);
2543 if (root
->root_key
.objectid
== BTRFS_DATA_RELOC_TREE_OBJECTID
)
2548 static int __btrfs_mod_ref(struct btrfs_trans_handle
*trans
,
2549 struct btrfs_root
*root
,
2550 struct extent_buffer
*buf
,
2551 int full_backref
, int inc
)
2558 struct btrfs_key key
;
2559 struct btrfs_file_extent_item
*fi
;
2563 int (*process_func
)(struct btrfs_trans_handle
*, struct btrfs_root
*,
2564 u64
, u64
, u64
, u64
, u64
, u64
);
2566 ref_root
= btrfs_header_owner(buf
);
2567 nritems
= btrfs_header_nritems(buf
);
2568 level
= btrfs_header_level(buf
);
2570 if (!root
->ref_cows
&& level
== 0)
2574 process_func
= btrfs_inc_extent_ref
;
2576 process_func
= btrfs_free_extent
;
2579 parent
= buf
->start
;
2583 for (i
= 0; i
< nritems
; i
++) {
2585 btrfs_item_key_to_cpu(buf
, &key
, i
);
2586 if (btrfs_key_type(&key
) != BTRFS_EXTENT_DATA_KEY
)
2588 fi
= btrfs_item_ptr(buf
, i
,
2589 struct btrfs_file_extent_item
);
2590 if (btrfs_file_extent_type(buf
, fi
) ==
2591 BTRFS_FILE_EXTENT_INLINE
)
2593 bytenr
= btrfs_file_extent_disk_bytenr(buf
, fi
);
2597 num_bytes
= btrfs_file_extent_disk_num_bytes(buf
, fi
);
2598 key
.offset
-= btrfs_file_extent_offset(buf
, fi
);
2599 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2600 parent
, ref_root
, key
.objectid
,
2605 bytenr
= btrfs_node_blockptr(buf
, i
);
2606 num_bytes
= btrfs_level_size(root
, level
- 1);
2607 ret
= process_func(trans
, root
, bytenr
, num_bytes
,
2608 parent
, ref_root
, level
- 1, 0);
2619 int btrfs_inc_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2620 struct extent_buffer
*buf
, int full_backref
)
2622 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 1);
2625 int btrfs_dec_ref(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2626 struct extent_buffer
*buf
, int full_backref
)
2628 return __btrfs_mod_ref(trans
, root
, buf
, full_backref
, 0);
2631 static int write_one_cache_group(struct btrfs_trans_handle
*trans
,
2632 struct btrfs_root
*root
,
2633 struct btrfs_path
*path
,
2634 struct btrfs_block_group_cache
*cache
)
2637 struct btrfs_root
*extent_root
= root
->fs_info
->extent_root
;
2639 struct extent_buffer
*leaf
;
2641 ret
= btrfs_search_slot(trans
, extent_root
, &cache
->key
, path
, 0, 1);
2646 leaf
= path
->nodes
[0];
2647 bi
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2648 write_extent_buffer(leaf
, &cache
->item
, bi
, sizeof(cache
->item
));
2649 btrfs_mark_buffer_dirty(leaf
);
2650 btrfs_release_path(path
);
2658 static struct btrfs_block_group_cache
*
2659 next_block_group(struct btrfs_root
*root
,
2660 struct btrfs_block_group_cache
*cache
)
2662 struct rb_node
*node
;
2663 spin_lock(&root
->fs_info
->block_group_cache_lock
);
2664 node
= rb_next(&cache
->cache_node
);
2665 btrfs_put_block_group(cache
);
2667 cache
= rb_entry(node
, struct btrfs_block_group_cache
,
2669 btrfs_get_block_group(cache
);
2672 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
2676 static int cache_save_setup(struct btrfs_block_group_cache
*block_group
,
2677 struct btrfs_trans_handle
*trans
,
2678 struct btrfs_path
*path
)
2680 struct btrfs_root
*root
= block_group
->fs_info
->tree_root
;
2681 struct inode
*inode
= NULL
;
2683 int dcs
= BTRFS_DC_ERROR
;
2689 * If this block group is smaller than 100 megs don't bother caching the
2692 if (block_group
->key
.offset
< (100 * 1024 * 1024)) {
2693 spin_lock(&block_group
->lock
);
2694 block_group
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2695 spin_unlock(&block_group
->lock
);
2700 inode
= lookup_free_space_inode(root
, block_group
, path
);
2701 if (IS_ERR(inode
) && PTR_ERR(inode
) != -ENOENT
) {
2702 ret
= PTR_ERR(inode
);
2703 btrfs_release_path(path
);
2707 if (IS_ERR(inode
)) {
2711 if (block_group
->ro
)
2714 ret
= create_free_space_inode(root
, trans
, block_group
, path
);
2720 /* We've already setup this transaction, go ahead and exit */
2721 if (block_group
->cache_generation
== trans
->transid
&&
2722 i_size_read(inode
)) {
2723 dcs
= BTRFS_DC_SETUP
;
2728 * We want to set the generation to 0, that way if anything goes wrong
2729 * from here on out we know not to trust this cache when we load up next
2732 BTRFS_I(inode
)->generation
= 0;
2733 ret
= btrfs_update_inode(trans
, root
, inode
);
2736 if (i_size_read(inode
) > 0) {
2737 ret
= btrfs_truncate_free_space_cache(root
, trans
, path
,
2743 spin_lock(&block_group
->lock
);
2744 if (block_group
->cached
!= BTRFS_CACHE_FINISHED
) {
2745 /* We're not cached, don't bother trying to write stuff out */
2746 dcs
= BTRFS_DC_WRITTEN
;
2747 spin_unlock(&block_group
->lock
);
2750 spin_unlock(&block_group
->lock
);
2752 num_pages
= (int)div64_u64(block_group
->key
.offset
, 1024 * 1024 * 1024);
2757 * Just to make absolutely sure we have enough space, we're going to
2758 * preallocate 12 pages worth of space for each block group. In
2759 * practice we ought to use at most 8, but we need extra space so we can
2760 * add our header and have a terminator between the extents and the
2764 num_pages
*= PAGE_CACHE_SIZE
;
2766 ret
= btrfs_check_data_free_space(inode
, num_pages
);
2770 ret
= btrfs_prealloc_file_range_trans(inode
, trans
, 0, 0, num_pages
,
2771 num_pages
, num_pages
,
2774 dcs
= BTRFS_DC_SETUP
;
2775 btrfs_free_reserved_data_space(inode
, num_pages
);
2780 btrfs_release_path(path
);
2782 spin_lock(&block_group
->lock
);
2784 block_group
->cache_generation
= trans
->transid
;
2785 block_group
->disk_cache_state
= dcs
;
2786 spin_unlock(&block_group
->lock
);
2791 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle
*trans
,
2792 struct btrfs_root
*root
)
2794 struct btrfs_block_group_cache
*cache
;
2796 struct btrfs_path
*path
;
2799 path
= btrfs_alloc_path();
2805 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2807 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
)
2809 cache
= next_block_group(root
, cache
);
2817 err
= cache_save_setup(cache
, trans
, path
);
2818 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2819 btrfs_put_block_group(cache
);
2824 err
= btrfs_run_delayed_refs(trans
, root
,
2829 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2831 if (cache
->disk_cache_state
== BTRFS_DC_CLEAR
) {
2832 btrfs_put_block_group(cache
);
2838 cache
= next_block_group(root
, cache
);
2847 if (cache
->disk_cache_state
== BTRFS_DC_SETUP
)
2848 cache
->disk_cache_state
= BTRFS_DC_NEED_WRITE
;
2850 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2852 err
= write_one_cache_group(trans
, root
, path
, cache
);
2854 btrfs_put_block_group(cache
);
2859 * I don't think this is needed since we're just marking our
2860 * preallocated extent as written, but just in case it can't
2864 err
= btrfs_run_delayed_refs(trans
, root
,
2869 cache
= btrfs_lookup_first_block_group(root
->fs_info
, last
);
2872 * Really this shouldn't happen, but it could if we
2873 * couldn't write the entire preallocated extent and
2874 * splitting the extent resulted in a new block.
2877 btrfs_put_block_group(cache
);
2880 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2882 cache
= next_block_group(root
, cache
);
2891 btrfs_write_out_cache(root
, trans
, cache
, path
);
2894 * If we didn't have an error then the cache state is still
2895 * NEED_WRITE, so we can set it to WRITTEN.
2897 if (cache
->disk_cache_state
== BTRFS_DC_NEED_WRITE
)
2898 cache
->disk_cache_state
= BTRFS_DC_WRITTEN
;
2899 last
= cache
->key
.objectid
+ cache
->key
.offset
;
2900 btrfs_put_block_group(cache
);
2903 btrfs_free_path(path
);
2907 int btrfs_extent_readonly(struct btrfs_root
*root
, u64 bytenr
)
2909 struct btrfs_block_group_cache
*block_group
;
2912 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
2913 if (!block_group
|| block_group
->ro
)
2916 btrfs_put_block_group(block_group
);
2920 static int update_space_info(struct btrfs_fs_info
*info
, u64 flags
,
2921 u64 total_bytes
, u64 bytes_used
,
2922 struct btrfs_space_info
**space_info
)
2924 struct btrfs_space_info
*found
;
2928 if (flags
& (BTRFS_BLOCK_GROUP_DUP
| BTRFS_BLOCK_GROUP_RAID1
|
2929 BTRFS_BLOCK_GROUP_RAID10
))
2934 found
= __find_space_info(info
, flags
);
2936 spin_lock(&found
->lock
);
2937 found
->total_bytes
+= total_bytes
;
2938 found
->disk_total
+= total_bytes
* factor
;
2939 found
->bytes_used
+= bytes_used
;
2940 found
->disk_used
+= bytes_used
* factor
;
2942 spin_unlock(&found
->lock
);
2943 *space_info
= found
;
2946 found
= kzalloc(sizeof(*found
), GFP_NOFS
);
2950 for (i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
2951 INIT_LIST_HEAD(&found
->block_groups
[i
]);
2952 init_rwsem(&found
->groups_sem
);
2953 spin_lock_init(&found
->lock
);
2954 found
->flags
= flags
& (BTRFS_BLOCK_GROUP_DATA
|
2955 BTRFS_BLOCK_GROUP_SYSTEM
|
2956 BTRFS_BLOCK_GROUP_METADATA
);
2957 found
->total_bytes
= total_bytes
;
2958 found
->disk_total
= total_bytes
* factor
;
2959 found
->bytes_used
= bytes_used
;
2960 found
->disk_used
= bytes_used
* factor
;
2961 found
->bytes_pinned
= 0;
2962 found
->bytes_reserved
= 0;
2963 found
->bytes_readonly
= 0;
2964 found
->bytes_may_use
= 0;
2966 found
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
2967 found
->chunk_alloc
= 0;
2969 init_waitqueue_head(&found
->wait
);
2970 *space_info
= found
;
2971 list_add_rcu(&found
->list
, &info
->space_info
);
2975 static void set_avail_alloc_bits(struct btrfs_fs_info
*fs_info
, u64 flags
)
2977 u64 extra_flags
= flags
& (BTRFS_BLOCK_GROUP_RAID0
|
2978 BTRFS_BLOCK_GROUP_RAID1
|
2979 BTRFS_BLOCK_GROUP_RAID10
|
2980 BTRFS_BLOCK_GROUP_DUP
);
2982 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
2983 fs_info
->avail_data_alloc_bits
|= extra_flags
;
2984 if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
2985 fs_info
->avail_metadata_alloc_bits
|= extra_flags
;
2986 if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
2987 fs_info
->avail_system_alloc_bits
|= extra_flags
;
2991 u64
btrfs_reduce_alloc_profile(struct btrfs_root
*root
, u64 flags
)
2994 * we add in the count of missing devices because we want
2995 * to make sure that any RAID levels on a degraded FS
2996 * continue to be honored.
2998 u64 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
2999 root
->fs_info
->fs_devices
->missing_devices
;
3001 if (num_devices
== 1)
3002 flags
&= ~(BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID0
);
3003 if (num_devices
< 4)
3004 flags
&= ~BTRFS_BLOCK_GROUP_RAID10
;
3006 if ((flags
& BTRFS_BLOCK_GROUP_DUP
) &&
3007 (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
3008 BTRFS_BLOCK_GROUP_RAID10
))) {
3009 flags
&= ~BTRFS_BLOCK_GROUP_DUP
;
3012 if ((flags
& BTRFS_BLOCK_GROUP_RAID1
) &&
3013 (flags
& BTRFS_BLOCK_GROUP_RAID10
)) {
3014 flags
&= ~BTRFS_BLOCK_GROUP_RAID1
;
3017 if ((flags
& BTRFS_BLOCK_GROUP_RAID0
) &&
3018 ((flags
& BTRFS_BLOCK_GROUP_RAID1
) |
3019 (flags
& BTRFS_BLOCK_GROUP_RAID10
) |
3020 (flags
& BTRFS_BLOCK_GROUP_DUP
)))
3021 flags
&= ~BTRFS_BLOCK_GROUP_RAID0
;
3025 static u64
get_alloc_profile(struct btrfs_root
*root
, u64 flags
)
3027 if (flags
& BTRFS_BLOCK_GROUP_DATA
)
3028 flags
|= root
->fs_info
->avail_data_alloc_bits
&
3029 root
->fs_info
->data_alloc_profile
;
3030 else if (flags
& BTRFS_BLOCK_GROUP_SYSTEM
)
3031 flags
|= root
->fs_info
->avail_system_alloc_bits
&
3032 root
->fs_info
->system_alloc_profile
;
3033 else if (flags
& BTRFS_BLOCK_GROUP_METADATA
)
3034 flags
|= root
->fs_info
->avail_metadata_alloc_bits
&
3035 root
->fs_info
->metadata_alloc_profile
;
3036 return btrfs_reduce_alloc_profile(root
, flags
);
3039 u64
btrfs_get_alloc_profile(struct btrfs_root
*root
, int data
)
3044 flags
= BTRFS_BLOCK_GROUP_DATA
;
3045 else if (root
== root
->fs_info
->chunk_root
)
3046 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
3048 flags
= BTRFS_BLOCK_GROUP_METADATA
;
3050 return get_alloc_profile(root
, flags
);
3053 void btrfs_set_inode_space_info(struct btrfs_root
*root
, struct inode
*inode
)
3055 BTRFS_I(inode
)->space_info
= __find_space_info(root
->fs_info
,
3056 BTRFS_BLOCK_GROUP_DATA
);
3060 * This will check the space that the inode allocates from to make sure we have
3061 * enough space for bytes.
3063 int btrfs_check_data_free_space(struct inode
*inode
, u64 bytes
)
3065 struct btrfs_space_info
*data_sinfo
;
3066 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3068 int ret
= 0, committed
= 0, alloc_chunk
= 1;
3070 /* make sure bytes are sectorsize aligned */
3071 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3073 if (root
== root
->fs_info
->tree_root
||
3074 BTRFS_I(inode
)->location
.objectid
== BTRFS_FREE_INO_OBJECTID
) {
3079 data_sinfo
= BTRFS_I(inode
)->space_info
;
3084 /* make sure we have enough space to handle the data first */
3085 spin_lock(&data_sinfo
->lock
);
3086 used
= data_sinfo
->bytes_used
+ data_sinfo
->bytes_reserved
+
3087 data_sinfo
->bytes_pinned
+ data_sinfo
->bytes_readonly
+
3088 data_sinfo
->bytes_may_use
;
3090 if (used
+ bytes
> data_sinfo
->total_bytes
) {
3091 struct btrfs_trans_handle
*trans
;
3094 * if we don't have enough free bytes in this space then we need
3095 * to alloc a new chunk.
3097 if (!data_sinfo
->full
&& alloc_chunk
) {
3100 data_sinfo
->force_alloc
= CHUNK_ALLOC_FORCE
;
3101 spin_unlock(&data_sinfo
->lock
);
3103 alloc_target
= btrfs_get_alloc_profile(root
, 1);
3104 trans
= btrfs_join_transaction(root
);
3106 return PTR_ERR(trans
);
3108 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
3109 bytes
+ 2 * 1024 * 1024,
3111 CHUNK_ALLOC_NO_FORCE
);
3112 btrfs_end_transaction(trans
, root
);
3121 btrfs_set_inode_space_info(root
, inode
);
3122 data_sinfo
= BTRFS_I(inode
)->space_info
;
3128 * If we have less pinned bytes than we want to allocate then
3129 * don't bother committing the transaction, it won't help us.
3131 if (data_sinfo
->bytes_pinned
< bytes
)
3133 spin_unlock(&data_sinfo
->lock
);
3135 /* commit the current transaction and try again */
3138 !atomic_read(&root
->fs_info
->open_ioctl_trans
)) {
3140 trans
= btrfs_join_transaction(root
);
3142 return PTR_ERR(trans
);
3143 ret
= btrfs_commit_transaction(trans
, root
);
3151 data_sinfo
->bytes_may_use
+= bytes
;
3152 spin_unlock(&data_sinfo
->lock
);
3158 * Called if we need to clear a data reservation for this inode.
3160 void btrfs_free_reserved_data_space(struct inode
*inode
, u64 bytes
)
3162 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3163 struct btrfs_space_info
*data_sinfo
;
3165 /* make sure bytes are sectorsize aligned */
3166 bytes
= (bytes
+ root
->sectorsize
- 1) & ~((u64
)root
->sectorsize
- 1);
3168 data_sinfo
= BTRFS_I(inode
)->space_info
;
3169 spin_lock(&data_sinfo
->lock
);
3170 data_sinfo
->bytes_may_use
-= bytes
;
3171 spin_unlock(&data_sinfo
->lock
);
3174 static void force_metadata_allocation(struct btrfs_fs_info
*info
)
3176 struct list_head
*head
= &info
->space_info
;
3177 struct btrfs_space_info
*found
;
3180 list_for_each_entry_rcu(found
, head
, list
) {
3181 if (found
->flags
& BTRFS_BLOCK_GROUP_METADATA
)
3182 found
->force_alloc
= CHUNK_ALLOC_FORCE
;
3187 static int should_alloc_chunk(struct btrfs_root
*root
,
3188 struct btrfs_space_info
*sinfo
, u64 alloc_bytes
,
3191 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3192 u64 num_bytes
= sinfo
->total_bytes
- sinfo
->bytes_readonly
;
3193 u64 num_allocated
= sinfo
->bytes_used
+ sinfo
->bytes_reserved
;
3196 if (force
== CHUNK_ALLOC_FORCE
)
3200 * We need to take into account the global rsv because for all intents
3201 * and purposes it's used space. Don't worry about locking the
3202 * global_rsv, it doesn't change except when the transaction commits.
3204 num_allocated
+= global_rsv
->size
;
3207 * in limited mode, we want to have some free space up to
3208 * about 1% of the FS size.
3210 if (force
== CHUNK_ALLOC_LIMITED
) {
3211 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3212 thresh
= max_t(u64
, 64 * 1024 * 1024,
3213 div_factor_fine(thresh
, 1));
3215 if (num_bytes
- num_allocated
< thresh
)
3220 * we have two similar checks here, one based on percentage
3221 * and once based on a hard number of 256MB. The idea
3222 * is that if we have a good amount of free
3223 * room, don't allocate a chunk. A good mount is
3224 * less than 80% utilized of the chunks we have allocated,
3225 * or more than 256MB free
3227 if (num_allocated
+ alloc_bytes
+ 256 * 1024 * 1024 < num_bytes
)
3230 if (num_allocated
+ alloc_bytes
< div_factor(num_bytes
, 8))
3233 thresh
= btrfs_super_total_bytes(&root
->fs_info
->super_copy
);
3235 /* 256MB or 5% of the FS */
3236 thresh
= max_t(u64
, 256 * 1024 * 1024, div_factor_fine(thresh
, 5));
3238 if (num_bytes
> thresh
&& sinfo
->bytes_used
< div_factor(num_bytes
, 3))
3243 static int do_chunk_alloc(struct btrfs_trans_handle
*trans
,
3244 struct btrfs_root
*extent_root
, u64 alloc_bytes
,
3245 u64 flags
, int force
)
3247 struct btrfs_space_info
*space_info
;
3248 struct btrfs_fs_info
*fs_info
= extent_root
->fs_info
;
3249 int wait_for_alloc
= 0;
3252 flags
= btrfs_reduce_alloc_profile(extent_root
, flags
);
3254 space_info
= __find_space_info(extent_root
->fs_info
, flags
);
3256 ret
= update_space_info(extent_root
->fs_info
, flags
,
3260 BUG_ON(!space_info
);
3263 spin_lock(&space_info
->lock
);
3264 if (space_info
->force_alloc
)
3265 force
= space_info
->force_alloc
;
3266 if (space_info
->full
) {
3267 spin_unlock(&space_info
->lock
);
3271 if (!should_alloc_chunk(extent_root
, space_info
, alloc_bytes
, force
)) {
3272 spin_unlock(&space_info
->lock
);
3274 } else if (space_info
->chunk_alloc
) {
3277 space_info
->chunk_alloc
= 1;
3280 spin_unlock(&space_info
->lock
);
3282 mutex_lock(&fs_info
->chunk_mutex
);
3285 * The chunk_mutex is held throughout the entirety of a chunk
3286 * allocation, so once we've acquired the chunk_mutex we know that the
3287 * other guy is done and we need to recheck and see if we should
3290 if (wait_for_alloc
) {
3291 mutex_unlock(&fs_info
->chunk_mutex
);
3297 * If we have mixed data/metadata chunks we want to make sure we keep
3298 * allocating mixed chunks instead of individual chunks.
3300 if (btrfs_mixed_space_info(space_info
))
3301 flags
|= (BTRFS_BLOCK_GROUP_DATA
| BTRFS_BLOCK_GROUP_METADATA
);
3304 * if we're doing a data chunk, go ahead and make sure that
3305 * we keep a reasonable number of metadata chunks allocated in the
3308 if (flags
& BTRFS_BLOCK_GROUP_DATA
&& fs_info
->metadata_ratio
) {
3309 fs_info
->data_chunk_allocations
++;
3310 if (!(fs_info
->data_chunk_allocations
%
3311 fs_info
->metadata_ratio
))
3312 force_metadata_allocation(fs_info
);
3315 ret
= btrfs_alloc_chunk(trans
, extent_root
, flags
);
3316 if (ret
< 0 && ret
!= -ENOSPC
)
3319 spin_lock(&space_info
->lock
);
3321 space_info
->full
= 1;
3325 space_info
->force_alloc
= CHUNK_ALLOC_NO_FORCE
;
3326 space_info
->chunk_alloc
= 0;
3327 spin_unlock(&space_info
->lock
);
3329 mutex_unlock(&extent_root
->fs_info
->chunk_mutex
);
3334 * shrink metadata reservation for delalloc
3336 static int shrink_delalloc(struct btrfs_trans_handle
*trans
,
3337 struct btrfs_root
*root
, u64 to_reclaim
, int sync
)
3339 struct btrfs_block_rsv
*block_rsv
;
3340 struct btrfs_space_info
*space_info
;
3345 int nr_pages
= (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT
;
3347 unsigned long progress
;
3349 block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
3350 space_info
= block_rsv
->space_info
;
3353 reserved
= space_info
->bytes_may_use
;
3354 progress
= space_info
->reservation_progress
;
3360 if (root
->fs_info
->delalloc_bytes
== 0) {
3363 btrfs_wait_ordered_extents(root
, 0, 0);
3367 max_reclaim
= min(reserved
, to_reclaim
);
3369 while (loops
< 1024) {
3370 /* have the flusher threads jump in and do some IO */
3372 nr_pages
= min_t(unsigned long, nr_pages
,
3373 root
->fs_info
->delalloc_bytes
>> PAGE_CACHE_SHIFT
);
3374 writeback_inodes_sb_nr_if_idle(root
->fs_info
->sb
, nr_pages
);
3376 spin_lock(&space_info
->lock
);
3377 if (reserved
> space_info
->bytes_may_use
)
3378 reclaimed
+= reserved
- space_info
->bytes_may_use
;
3379 reserved
= space_info
->bytes_may_use
;
3380 spin_unlock(&space_info
->lock
);
3384 if (reserved
== 0 || reclaimed
>= max_reclaim
)
3387 if (trans
&& trans
->transaction
->blocked
)
3390 time_left
= schedule_timeout_interruptible(1);
3392 /* We were interrupted, exit */
3396 /* we've kicked the IO a few times, if anything has been freed,
3397 * exit. There is no sense in looping here for a long time
3398 * when we really need to commit the transaction, or there are
3399 * just too many writers without enough free space
3404 if (progress
!= space_info
->reservation_progress
)
3409 if (reclaimed
< to_reclaim
&& !trans
)
3410 btrfs_wait_ordered_extents(root
, 0, 0);
3411 return reclaimed
>= to_reclaim
;
3415 * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space
3416 * @root - the root we're allocating for
3417 * @block_rsv - the block_rsv we're allocating for
3418 * @orig_bytes - the number of bytes we want
3419 * @flush - wether or not we can flush to make our reservation
3420 * @check - wether this is just to check if we have enough space or not
3422 * This will reserve orgi_bytes number of bytes from the space info associated
3423 * with the block_rsv. If there is not enough space it will make an attempt to
3424 * flush out space to make room. It will do this by flushing delalloc if
3425 * possible or committing the transaction. If flush is 0 then no attempts to
3426 * regain reservations will be made and this will fail if there is not enough
3429 static int reserve_metadata_bytes(struct btrfs_root
*root
,
3430 struct btrfs_block_rsv
*block_rsv
,
3431 u64 orig_bytes
, int flush
, int check
)
3433 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3434 struct btrfs_trans_handle
*trans
;
3436 u64 num_bytes
= orig_bytes
;
3439 bool committed
= false;
3440 bool flushing
= false;
3442 trans
= (struct btrfs_trans_handle
*)current
->journal_info
;
3445 spin_lock(&space_info
->lock
);
3447 * We only want to wait if somebody other than us is flushing and we are
3448 * actually alloed to flush.
3450 while (flush
&& !flushing
&& space_info
->flush
) {
3451 spin_unlock(&space_info
->lock
);
3453 * If we have a trans handle we can't wait because the flusher
3454 * may have to commit the transaction, which would mean we would
3455 * deadlock since we are waiting for the flusher to finish, but
3456 * hold the current transaction open.
3460 ret
= wait_event_interruptible(space_info
->wait
,
3461 !space_info
->flush
);
3462 /* Must have been interrupted, return */
3466 spin_lock(&space_info
->lock
);
3470 used
= space_info
->bytes_used
+ space_info
->bytes_reserved
+
3471 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
3472 space_info
->bytes_may_use
;
3475 * The idea here is that we've not already over-reserved the block group
3476 * then we can go ahead and save our reservation first and then start
3477 * flushing if we need to. Otherwise if we've already overcommitted
3478 * lets start flushing stuff first and then come back and try to make
3481 if (used
<= space_info
->total_bytes
) {
3482 if (used
+ orig_bytes
<= space_info
->total_bytes
) {
3483 space_info
->bytes_may_use
+= orig_bytes
;
3487 * Ok set num_bytes to orig_bytes since we aren't
3488 * overocmmitted, this way we only try and reclaim what
3491 num_bytes
= orig_bytes
;
3495 * Ok we're over committed, set num_bytes to the overcommitted
3496 * amount plus the amount of bytes that we need for this
3499 num_bytes
= used
- space_info
->total_bytes
+
3500 (orig_bytes
* (retries
+ 1));
3503 if (ret
&& !check
) {
3504 u64 profile
= btrfs_get_alloc_profile(root
, 0);
3507 spin_lock(&root
->fs_info
->free_chunk_lock
);
3508 avail
= root
->fs_info
->free_chunk_space
;
3511 * If we have dup, raid1 or raid10 then only half of the free
3512 * space is actually useable.
3514 if (profile
& (BTRFS_BLOCK_GROUP_DUP
|
3515 BTRFS_BLOCK_GROUP_RAID1
|
3516 BTRFS_BLOCK_GROUP_RAID10
))
3520 * If we aren't flushing don't let us overcommit too much, say
3521 * 1/8th of the space. If we can flush, let it overcommit up to
3528 spin_unlock(&root
->fs_info
->free_chunk_lock
);
3530 if (used
+ num_bytes
< space_info
->total_bytes
+ avail
) {
3531 space_info
->bytes_may_use
+= orig_bytes
;
3537 * Couldn't make our reservation, save our place so while we're trying
3538 * to reclaim space we can actually use it instead of somebody else
3539 * stealing it from us.
3543 space_info
->flush
= 1;
3546 spin_unlock(&space_info
->lock
);
3552 * We do synchronous shrinking since we don't actually unreserve
3553 * metadata until after the IO is completed.
3555 ret
= shrink_delalloc(trans
, root
, num_bytes
, 1);
3562 * So if we were overcommitted it's possible that somebody else flushed
3563 * out enough space and we simply didn't have enough space to reclaim,
3564 * so go back around and try again.
3579 trans
= btrfs_join_transaction(root
);
3582 ret
= btrfs_commit_transaction(trans
, root
);
3591 spin_lock(&space_info
->lock
);
3592 space_info
->flush
= 0;
3593 wake_up_all(&space_info
->wait
);
3594 spin_unlock(&space_info
->lock
);
3599 static struct btrfs_block_rsv
*get_block_rsv(struct btrfs_trans_handle
*trans
,
3600 struct btrfs_root
*root
)
3602 struct btrfs_block_rsv
*block_rsv
= NULL
;
3604 if (root
->ref_cows
|| root
== root
->fs_info
->csum_root
)
3605 block_rsv
= trans
->block_rsv
;
3608 block_rsv
= root
->block_rsv
;
3611 block_rsv
= &root
->fs_info
->empty_block_rsv
;
3616 static int block_rsv_use_bytes(struct btrfs_block_rsv
*block_rsv
,
3620 spin_lock(&block_rsv
->lock
);
3621 if (block_rsv
->reserved
>= num_bytes
) {
3622 block_rsv
->reserved
-= num_bytes
;
3623 if (block_rsv
->reserved
< block_rsv
->size
)
3624 block_rsv
->full
= 0;
3627 spin_unlock(&block_rsv
->lock
);
3631 static void block_rsv_add_bytes(struct btrfs_block_rsv
*block_rsv
,
3632 u64 num_bytes
, int update_size
)
3634 spin_lock(&block_rsv
->lock
);
3635 block_rsv
->reserved
+= num_bytes
;
3637 block_rsv
->size
+= num_bytes
;
3638 else if (block_rsv
->reserved
>= block_rsv
->size
)
3639 block_rsv
->full
= 1;
3640 spin_unlock(&block_rsv
->lock
);
3643 static void block_rsv_release_bytes(struct btrfs_block_rsv
*block_rsv
,
3644 struct btrfs_block_rsv
*dest
, u64 num_bytes
)
3646 struct btrfs_space_info
*space_info
= block_rsv
->space_info
;
3648 spin_lock(&block_rsv
->lock
);
3649 if (num_bytes
== (u64
)-1)
3650 num_bytes
= block_rsv
->size
;
3651 block_rsv
->size
-= num_bytes
;
3652 if (block_rsv
->reserved
>= block_rsv
->size
) {
3653 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3654 block_rsv
->reserved
= block_rsv
->size
;
3655 block_rsv
->full
= 1;
3659 spin_unlock(&block_rsv
->lock
);
3661 if (num_bytes
> 0) {
3663 spin_lock(&dest
->lock
);
3667 bytes_to_add
= dest
->size
- dest
->reserved
;
3668 bytes_to_add
= min(num_bytes
, bytes_to_add
);
3669 dest
->reserved
+= bytes_to_add
;
3670 if (dest
->reserved
>= dest
->size
)
3672 num_bytes
-= bytes_to_add
;
3674 spin_unlock(&dest
->lock
);
3677 spin_lock(&space_info
->lock
);
3678 space_info
->bytes_may_use
-= num_bytes
;
3679 space_info
->reservation_progress
++;
3680 spin_unlock(&space_info
->lock
);
3685 static int block_rsv_migrate_bytes(struct btrfs_block_rsv
*src
,
3686 struct btrfs_block_rsv
*dst
, u64 num_bytes
)
3690 ret
= block_rsv_use_bytes(src
, num_bytes
);
3694 block_rsv_add_bytes(dst
, num_bytes
, 1);
3698 void btrfs_init_block_rsv(struct btrfs_block_rsv
*rsv
)
3700 memset(rsv
, 0, sizeof(*rsv
));
3701 spin_lock_init(&rsv
->lock
);
3704 struct btrfs_block_rsv
*btrfs_alloc_block_rsv(struct btrfs_root
*root
)
3706 struct btrfs_block_rsv
*block_rsv
;
3707 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3709 block_rsv
= kmalloc(sizeof(*block_rsv
), GFP_NOFS
);
3713 btrfs_init_block_rsv(block_rsv
);
3714 block_rsv
->space_info
= __find_space_info(fs_info
,
3715 BTRFS_BLOCK_GROUP_METADATA
);
3719 void btrfs_free_block_rsv(struct btrfs_root
*root
,
3720 struct btrfs_block_rsv
*rsv
)
3722 btrfs_block_rsv_release(root
, rsv
, (u64
)-1);
3726 int btrfs_block_rsv_add(struct btrfs_root
*root
,
3727 struct btrfs_block_rsv
*block_rsv
,
3735 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, 1, 0);
3737 block_rsv_add_bytes(block_rsv
, num_bytes
, 1);
3744 int btrfs_block_rsv_check(struct btrfs_root
*root
,
3745 struct btrfs_block_rsv
*block_rsv
,
3746 u64 min_reserved
, int min_factor
, int flush
)
3754 spin_lock(&block_rsv
->lock
);
3756 num_bytes
= div_factor(block_rsv
->size
, min_factor
);
3757 if (min_reserved
> num_bytes
)
3758 num_bytes
= min_reserved
;
3760 if (block_rsv
->reserved
>= num_bytes
)
3763 num_bytes
-= block_rsv
->reserved
;
3764 spin_unlock(&block_rsv
->lock
);
3769 ret
= reserve_metadata_bytes(root
, block_rsv
, num_bytes
, flush
, !flush
);
3771 block_rsv_add_bytes(block_rsv
, num_bytes
, 0);
3778 int btrfs_block_rsv_migrate(struct btrfs_block_rsv
*src_rsv
,
3779 struct btrfs_block_rsv
*dst_rsv
,
3782 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3785 void btrfs_block_rsv_release(struct btrfs_root
*root
,
3786 struct btrfs_block_rsv
*block_rsv
,
3789 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
3790 if (global_rsv
->full
|| global_rsv
== block_rsv
||
3791 block_rsv
->space_info
!= global_rsv
->space_info
)
3793 block_rsv_release_bytes(block_rsv
, global_rsv
, num_bytes
);
3797 * helper to calculate size of global block reservation.
3798 * the desired value is sum of space used by extent tree,
3799 * checksum tree and root tree
3801 static u64
calc_global_metadata_size(struct btrfs_fs_info
*fs_info
)
3803 struct btrfs_space_info
*sinfo
;
3807 int csum_size
= btrfs_super_csum_size(&fs_info
->super_copy
);
3809 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_DATA
);
3810 spin_lock(&sinfo
->lock
);
3811 data_used
= sinfo
->bytes_used
;
3812 spin_unlock(&sinfo
->lock
);
3814 sinfo
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3815 spin_lock(&sinfo
->lock
);
3816 if (sinfo
->flags
& BTRFS_BLOCK_GROUP_DATA
)
3818 meta_used
= sinfo
->bytes_used
;
3819 spin_unlock(&sinfo
->lock
);
3821 num_bytes
= (data_used
>> fs_info
->sb
->s_blocksize_bits
) *
3823 num_bytes
+= div64_u64(data_used
+ meta_used
, 50);
3825 if (num_bytes
* 3 > meta_used
)
3826 num_bytes
= div64_u64(meta_used
, 3);
3828 return ALIGN(num_bytes
, fs_info
->extent_root
->leafsize
<< 10);
3831 static void update_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3833 struct btrfs_block_rsv
*block_rsv
= &fs_info
->global_block_rsv
;
3834 struct btrfs_space_info
*sinfo
= block_rsv
->space_info
;
3837 num_bytes
= calc_global_metadata_size(fs_info
);
3839 spin_lock(&block_rsv
->lock
);
3840 spin_lock(&sinfo
->lock
);
3842 block_rsv
->size
= num_bytes
;
3844 num_bytes
= sinfo
->bytes_used
+ sinfo
->bytes_pinned
+
3845 sinfo
->bytes_reserved
+ sinfo
->bytes_readonly
+
3846 sinfo
->bytes_may_use
;
3848 if (sinfo
->total_bytes
> num_bytes
) {
3849 num_bytes
= sinfo
->total_bytes
- num_bytes
;
3850 block_rsv
->reserved
+= num_bytes
;
3851 sinfo
->bytes_may_use
+= num_bytes
;
3854 if (block_rsv
->reserved
>= block_rsv
->size
) {
3855 num_bytes
= block_rsv
->reserved
- block_rsv
->size
;
3856 sinfo
->bytes_may_use
-= num_bytes
;
3857 sinfo
->reservation_progress
++;
3858 block_rsv
->reserved
= block_rsv
->size
;
3859 block_rsv
->full
= 1;
3862 spin_unlock(&sinfo
->lock
);
3863 spin_unlock(&block_rsv
->lock
);
3866 static void init_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3868 struct btrfs_space_info
*space_info
;
3870 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_SYSTEM
);
3871 fs_info
->chunk_block_rsv
.space_info
= space_info
;
3873 space_info
= __find_space_info(fs_info
, BTRFS_BLOCK_GROUP_METADATA
);
3874 fs_info
->global_block_rsv
.space_info
= space_info
;
3875 fs_info
->delalloc_block_rsv
.space_info
= space_info
;
3876 fs_info
->trans_block_rsv
.space_info
= space_info
;
3877 fs_info
->empty_block_rsv
.space_info
= space_info
;
3879 fs_info
->extent_root
->block_rsv
= &fs_info
->global_block_rsv
;
3880 fs_info
->csum_root
->block_rsv
= &fs_info
->global_block_rsv
;
3881 fs_info
->dev_root
->block_rsv
= &fs_info
->global_block_rsv
;
3882 fs_info
->tree_root
->block_rsv
= &fs_info
->global_block_rsv
;
3883 fs_info
->chunk_root
->block_rsv
= &fs_info
->chunk_block_rsv
;
3885 update_global_block_rsv(fs_info
);
3888 static void release_global_block_rsv(struct btrfs_fs_info
*fs_info
)
3890 block_rsv_release_bytes(&fs_info
->global_block_rsv
, NULL
, (u64
)-1);
3891 WARN_ON(fs_info
->delalloc_block_rsv
.size
> 0);
3892 WARN_ON(fs_info
->delalloc_block_rsv
.reserved
> 0);
3893 WARN_ON(fs_info
->trans_block_rsv
.size
> 0);
3894 WARN_ON(fs_info
->trans_block_rsv
.reserved
> 0);
3895 WARN_ON(fs_info
->chunk_block_rsv
.size
> 0);
3896 WARN_ON(fs_info
->chunk_block_rsv
.reserved
> 0);
3899 void btrfs_trans_release_metadata(struct btrfs_trans_handle
*trans
,
3900 struct btrfs_root
*root
)
3902 struct btrfs_block_rsv
*block_rsv
;
3904 if (!trans
->bytes_reserved
)
3907 block_rsv
= &root
->fs_info
->trans_block_rsv
;
3908 btrfs_block_rsv_release(root
, block_rsv
, trans
->bytes_reserved
);
3909 trans
->bytes_reserved
= 0;
3912 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle
*trans
,
3913 struct inode
*inode
)
3915 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3916 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3917 struct btrfs_block_rsv
*dst_rsv
= root
->orphan_block_rsv
;
3920 * We need to hold space in order to delete our orphan item once we've
3921 * added it, so this takes the reservation so we can release it later
3922 * when we are truly done with the orphan item.
3924 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
3925 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3928 void btrfs_orphan_release_metadata(struct inode
*inode
)
3930 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
3931 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 1);
3932 btrfs_block_rsv_release(root
, root
->orphan_block_rsv
, num_bytes
);
3935 int btrfs_snap_reserve_metadata(struct btrfs_trans_handle
*trans
,
3936 struct btrfs_pending_snapshot
*pending
)
3938 struct btrfs_root
*root
= pending
->root
;
3939 struct btrfs_block_rsv
*src_rsv
= get_block_rsv(trans
, root
);
3940 struct btrfs_block_rsv
*dst_rsv
= &pending
->block_rsv
;
3942 * two for root back/forward refs, two for directory entries
3943 * and one for root of the snapshot.
3945 u64 num_bytes
= btrfs_calc_trans_metadata_size(root
, 5);
3946 dst_rsv
->space_info
= src_rsv
->space_info
;
3947 return block_rsv_migrate_bytes(src_rsv
, dst_rsv
, num_bytes
);
3951 * drop_outstanding_extent - drop an outstanding extent
3952 * @inode: the inode we're dropping the extent for
3954 * This is called when we are freeing up an outstanding extent, either called
3955 * after an error or after an extent is written. This will return the number of
3956 * reserved extents that need to be freed. This must be called with
3957 * BTRFS_I(inode)->lock held.
3959 static unsigned drop_outstanding_extent(struct inode
*inode
)
3961 unsigned dropped_extents
= 0;
3963 BUG_ON(!BTRFS_I(inode
)->outstanding_extents
);
3964 BTRFS_I(inode
)->outstanding_extents
--;
3967 * If we have more or the same amount of outsanding extents than we have
3968 * reserved then we need to leave the reserved extents count alone.
3970 if (BTRFS_I(inode
)->outstanding_extents
>=
3971 BTRFS_I(inode
)->reserved_extents
)
3974 dropped_extents
= BTRFS_I(inode
)->reserved_extents
-
3975 BTRFS_I(inode
)->outstanding_extents
;
3976 BTRFS_I(inode
)->reserved_extents
-= dropped_extents
;
3977 return dropped_extents
;
3981 * calc_csum_metadata_size - return the amount of metada space that must be
3982 * reserved/free'd for the given bytes.
3983 * @inode: the inode we're manipulating
3984 * @num_bytes: the number of bytes in question
3985 * @reserve: 1 if we are reserving space, 0 if we are freeing space
3987 * This adjusts the number of csum_bytes in the inode and then returns the
3988 * correct amount of metadata that must either be reserved or freed. We
3989 * calculate how many checksums we can fit into one leaf and then divide the
3990 * number of bytes that will need to be checksumed by this value to figure out
3991 * how many checksums will be required. If we are adding bytes then the number
3992 * may go up and we will return the number of additional bytes that must be
3993 * reserved. If it is going down we will return the number of bytes that must
3996 * This must be called with BTRFS_I(inode)->lock held.
3998 static u64
calc_csum_metadata_size(struct inode
*inode
, u64 num_bytes
,
4001 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4003 int num_csums_per_leaf
;
4007 if (BTRFS_I(inode
)->flags
& BTRFS_INODE_NODATASUM
&&
4008 BTRFS_I(inode
)->csum_bytes
== 0)
4011 old_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4013 BTRFS_I(inode
)->csum_bytes
+= num_bytes
;
4015 BTRFS_I(inode
)->csum_bytes
-= num_bytes
;
4016 csum_size
= BTRFS_LEAF_DATA_SIZE(root
) - sizeof(struct btrfs_item
);
4017 num_csums_per_leaf
= (int)div64_u64(csum_size
,
4018 sizeof(struct btrfs_csum_item
) +
4019 sizeof(struct btrfs_disk_key
));
4020 num_csums
= (int)div64_u64(BTRFS_I(inode
)->csum_bytes
, root
->sectorsize
);
4021 num_csums
= num_csums
+ num_csums_per_leaf
- 1;
4022 num_csums
= num_csums
/ num_csums_per_leaf
;
4024 old_csums
= old_csums
+ num_csums_per_leaf
- 1;
4025 old_csums
= old_csums
/ num_csums_per_leaf
;
4027 /* No change, no need to reserve more */
4028 if (old_csums
== num_csums
)
4032 return btrfs_calc_trans_metadata_size(root
,
4033 num_csums
- old_csums
);
4035 return btrfs_calc_trans_metadata_size(root
, old_csums
- num_csums
);
4038 int btrfs_delalloc_reserve_metadata(struct inode
*inode
, u64 num_bytes
)
4040 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4041 struct btrfs_block_rsv
*block_rsv
= &root
->fs_info
->delalloc_block_rsv
;
4043 unsigned nr_extents
= 0;
4047 if (btrfs_is_free_space_inode(root
, inode
))
4050 if (flush
&& btrfs_transaction_in_commit(root
->fs_info
))
4051 schedule_timeout(1);
4053 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4055 spin_lock(&BTRFS_I(inode
)->lock
);
4056 BTRFS_I(inode
)->outstanding_extents
++;
4058 if (BTRFS_I(inode
)->outstanding_extents
>
4059 BTRFS_I(inode
)->reserved_extents
) {
4060 nr_extents
= BTRFS_I(inode
)->outstanding_extents
-
4061 BTRFS_I(inode
)->reserved_extents
;
4062 BTRFS_I(inode
)->reserved_extents
+= nr_extents
;
4064 to_reserve
= btrfs_calc_trans_metadata_size(root
, nr_extents
);
4066 to_reserve
+= calc_csum_metadata_size(inode
, num_bytes
, 1);
4067 spin_unlock(&BTRFS_I(inode
)->lock
);
4069 ret
= reserve_metadata_bytes(root
, block_rsv
, to_reserve
, flush
, 0);
4074 spin_lock(&BTRFS_I(inode
)->lock
);
4075 dropped
= drop_outstanding_extent(inode
);
4076 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4077 spin_unlock(&BTRFS_I(inode
)->lock
);
4078 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4081 * Somebody could have come in and twiddled with the
4082 * reservation, so if we have to free more than we would have
4083 * reserved from this reservation go ahead and release those
4086 to_free
-= to_reserve
;
4088 btrfs_block_rsv_release(root
, block_rsv
, to_free
);
4092 block_rsv_add_bytes(block_rsv
, to_reserve
, 1);
4098 * btrfs_delalloc_release_metadata - release a metadata reservation for an inode
4099 * @inode: the inode to release the reservation for
4100 * @num_bytes: the number of bytes we're releasing
4102 * This will release the metadata reservation for an inode. This can be called
4103 * once we complete IO for a given set of bytes to release their metadata
4106 void btrfs_delalloc_release_metadata(struct inode
*inode
, u64 num_bytes
)
4108 struct btrfs_root
*root
= BTRFS_I(inode
)->root
;
4112 num_bytes
= ALIGN(num_bytes
, root
->sectorsize
);
4113 spin_lock(&BTRFS_I(inode
)->lock
);
4114 dropped
= drop_outstanding_extent(inode
);
4116 to_free
= calc_csum_metadata_size(inode
, num_bytes
, 0);
4117 spin_unlock(&BTRFS_I(inode
)->lock
);
4119 to_free
+= btrfs_calc_trans_metadata_size(root
, dropped
);
4121 btrfs_block_rsv_release(root
, &root
->fs_info
->delalloc_block_rsv
,
4126 * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc
4127 * @inode: inode we're writing to
4128 * @num_bytes: the number of bytes we want to allocate
4130 * This will do the following things
4132 * o reserve space in the data space info for num_bytes
4133 * o reserve space in the metadata space info based on number of outstanding
4134 * extents and how much csums will be needed
4135 * o add to the inodes ->delalloc_bytes
4136 * o add it to the fs_info's delalloc inodes list.
4138 * This will return 0 for success and -ENOSPC if there is no space left.
4140 int btrfs_delalloc_reserve_space(struct inode
*inode
, u64 num_bytes
)
4144 ret
= btrfs_check_data_free_space(inode
, num_bytes
);
4148 ret
= btrfs_delalloc_reserve_metadata(inode
, num_bytes
);
4150 btrfs_free_reserved_data_space(inode
, num_bytes
);
4158 * btrfs_delalloc_release_space - release data and metadata space for delalloc
4159 * @inode: inode we're releasing space for
4160 * @num_bytes: the number of bytes we want to free up
4162 * This must be matched with a call to btrfs_delalloc_reserve_space. This is
4163 * called in the case that we don't need the metadata AND data reservations
4164 * anymore. So if there is an error or we insert an inline extent.
4166 * This function will release the metadata space that was not used and will
4167 * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes
4168 * list if there are no delalloc bytes left.
4170 void btrfs_delalloc_release_space(struct inode
*inode
, u64 num_bytes
)
4172 btrfs_delalloc_release_metadata(inode
, num_bytes
);
4173 btrfs_free_reserved_data_space(inode
, num_bytes
);
4176 static int update_block_group(struct btrfs_trans_handle
*trans
,
4177 struct btrfs_root
*root
,
4178 u64 bytenr
, u64 num_bytes
, int alloc
)
4180 struct btrfs_block_group_cache
*cache
= NULL
;
4181 struct btrfs_fs_info
*info
= root
->fs_info
;
4182 u64 total
= num_bytes
;
4187 /* block accounting for super block */
4188 spin_lock(&info
->delalloc_lock
);
4189 old_val
= btrfs_super_bytes_used(&info
->super_copy
);
4191 old_val
+= num_bytes
;
4193 old_val
-= num_bytes
;
4194 btrfs_set_super_bytes_used(&info
->super_copy
, old_val
);
4195 spin_unlock(&info
->delalloc_lock
);
4198 cache
= btrfs_lookup_block_group(info
, bytenr
);
4201 if (cache
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
4202 BTRFS_BLOCK_GROUP_RAID1
|
4203 BTRFS_BLOCK_GROUP_RAID10
))
4208 * If this block group has free space cache written out, we
4209 * need to make sure to load it if we are removing space. This
4210 * is because we need the unpinning stage to actually add the
4211 * space back to the block group, otherwise we will leak space.
4213 if (!alloc
&& cache
->cached
== BTRFS_CACHE_NO
)
4214 cache_block_group(cache
, trans
, NULL
, 1);
4216 byte_in_group
= bytenr
- cache
->key
.objectid
;
4217 WARN_ON(byte_in_group
> cache
->key
.offset
);
4219 spin_lock(&cache
->space_info
->lock
);
4220 spin_lock(&cache
->lock
);
4222 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
4223 cache
->disk_cache_state
< BTRFS_DC_CLEAR
)
4224 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
4227 old_val
= btrfs_block_group_used(&cache
->item
);
4228 num_bytes
= min(total
, cache
->key
.offset
- byte_in_group
);
4230 old_val
+= num_bytes
;
4231 btrfs_set_block_group_used(&cache
->item
, old_val
);
4232 cache
->reserved
-= num_bytes
;
4233 cache
->space_info
->bytes_reserved
-= num_bytes
;
4234 cache
->space_info
->bytes_used
+= num_bytes
;
4235 cache
->space_info
->disk_used
+= num_bytes
* factor
;
4236 spin_unlock(&cache
->lock
);
4237 spin_unlock(&cache
->space_info
->lock
);
4239 old_val
-= num_bytes
;
4240 btrfs_set_block_group_used(&cache
->item
, old_val
);
4241 cache
->pinned
+= num_bytes
;
4242 cache
->space_info
->bytes_pinned
+= num_bytes
;
4243 cache
->space_info
->bytes_used
-= num_bytes
;
4244 cache
->space_info
->disk_used
-= num_bytes
* factor
;
4245 spin_unlock(&cache
->lock
);
4246 spin_unlock(&cache
->space_info
->lock
);
4248 set_extent_dirty(info
->pinned_extents
,
4249 bytenr
, bytenr
+ num_bytes
- 1,
4250 GFP_NOFS
| __GFP_NOFAIL
);
4252 btrfs_put_block_group(cache
);
4254 bytenr
+= num_bytes
;
4259 static u64
first_logical_byte(struct btrfs_root
*root
, u64 search_start
)
4261 struct btrfs_block_group_cache
*cache
;
4264 cache
= btrfs_lookup_first_block_group(root
->fs_info
, search_start
);
4268 bytenr
= cache
->key
.objectid
;
4269 btrfs_put_block_group(cache
);
4274 static int pin_down_extent(struct btrfs_root
*root
,
4275 struct btrfs_block_group_cache
*cache
,
4276 u64 bytenr
, u64 num_bytes
, int reserved
)
4278 spin_lock(&cache
->space_info
->lock
);
4279 spin_lock(&cache
->lock
);
4280 cache
->pinned
+= num_bytes
;
4281 cache
->space_info
->bytes_pinned
+= num_bytes
;
4283 cache
->reserved
-= num_bytes
;
4284 cache
->space_info
->bytes_reserved
-= num_bytes
;
4286 spin_unlock(&cache
->lock
);
4287 spin_unlock(&cache
->space_info
->lock
);
4289 set_extent_dirty(root
->fs_info
->pinned_extents
, bytenr
,
4290 bytenr
+ num_bytes
- 1, GFP_NOFS
| __GFP_NOFAIL
);
4295 * this function must be called within transaction
4297 int btrfs_pin_extent(struct btrfs_root
*root
,
4298 u64 bytenr
, u64 num_bytes
, int reserved
)
4300 struct btrfs_block_group_cache
*cache
;
4302 cache
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
4305 pin_down_extent(root
, cache
, bytenr
, num_bytes
, reserved
);
4307 btrfs_put_block_group(cache
);
4312 * btrfs_update_reserved_bytes - update the block_group and space info counters
4313 * @cache: The cache we are manipulating
4314 * @num_bytes: The number of bytes in question
4315 * @reserve: One of the reservation enums
4317 * This is called by the allocator when it reserves space, or by somebody who is
4318 * freeing space that was never actually used on disk. For example if you
4319 * reserve some space for a new leaf in transaction A and before transaction A
4320 * commits you free that leaf, you call this with reserve set to 0 in order to
4321 * clear the reservation.
4323 * Metadata reservations should be called with RESERVE_ALLOC so we do the proper
4324 * ENOSPC accounting. For data we handle the reservation through clearing the
4325 * delalloc bits in the io_tree. We have to do this since we could end up
4326 * allocating less disk space for the amount of data we have reserved in the
4327 * case of compression.
4329 * If this is a reservation and the block group has become read only we cannot
4330 * make the reservation and return -EAGAIN, otherwise this function always
4333 static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache
*cache
,
4334 u64 num_bytes
, int reserve
)
4336 struct btrfs_space_info
*space_info
= cache
->space_info
;
4338 spin_lock(&space_info
->lock
);
4339 spin_lock(&cache
->lock
);
4340 if (reserve
!= RESERVE_FREE
) {
4344 cache
->reserved
+= num_bytes
;
4345 space_info
->bytes_reserved
+= num_bytes
;
4346 if (reserve
== RESERVE_ALLOC
) {
4347 BUG_ON(space_info
->bytes_may_use
< num_bytes
);
4348 space_info
->bytes_may_use
-= num_bytes
;
4353 space_info
->bytes_readonly
+= num_bytes
;
4354 cache
->reserved
-= num_bytes
;
4355 space_info
->bytes_reserved
-= num_bytes
;
4356 space_info
->reservation_progress
++;
4358 spin_unlock(&cache
->lock
);
4359 spin_unlock(&space_info
->lock
);
4363 int btrfs_prepare_extent_commit(struct btrfs_trans_handle
*trans
,
4364 struct btrfs_root
*root
)
4366 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4367 struct btrfs_caching_control
*next
;
4368 struct btrfs_caching_control
*caching_ctl
;
4369 struct btrfs_block_group_cache
*cache
;
4371 down_write(&fs_info
->extent_commit_sem
);
4373 list_for_each_entry_safe(caching_ctl
, next
,
4374 &fs_info
->caching_block_groups
, list
) {
4375 cache
= caching_ctl
->block_group
;
4376 if (block_group_cache_done(cache
)) {
4377 cache
->last_byte_to_unpin
= (u64
)-1;
4378 list_del_init(&caching_ctl
->list
);
4379 put_caching_control(caching_ctl
);
4381 cache
->last_byte_to_unpin
= caching_ctl
->progress
;
4385 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4386 fs_info
->pinned_extents
= &fs_info
->freed_extents
[1];
4388 fs_info
->pinned_extents
= &fs_info
->freed_extents
[0];
4390 up_write(&fs_info
->extent_commit_sem
);
4392 update_global_block_rsv(fs_info
);
4396 static int unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
4398 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4399 struct btrfs_block_group_cache
*cache
= NULL
;
4402 while (start
<= end
) {
4404 start
>= cache
->key
.objectid
+ cache
->key
.offset
) {
4406 btrfs_put_block_group(cache
);
4407 cache
= btrfs_lookup_block_group(fs_info
, start
);
4411 len
= cache
->key
.objectid
+ cache
->key
.offset
- start
;
4412 len
= min(len
, end
+ 1 - start
);
4414 if (start
< cache
->last_byte_to_unpin
) {
4415 len
= min(len
, cache
->last_byte_to_unpin
- start
);
4416 btrfs_add_free_space(cache
, start
, len
);
4421 spin_lock(&cache
->space_info
->lock
);
4422 spin_lock(&cache
->lock
);
4423 cache
->pinned
-= len
;
4424 cache
->space_info
->bytes_pinned
-= len
;
4426 cache
->space_info
->bytes_readonly
+= len
;
4427 spin_unlock(&cache
->lock
);
4428 spin_unlock(&cache
->space_info
->lock
);
4432 btrfs_put_block_group(cache
);
4436 int btrfs_finish_extent_commit(struct btrfs_trans_handle
*trans
,
4437 struct btrfs_root
*root
)
4439 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4440 struct extent_io_tree
*unpin
;
4445 if (fs_info
->pinned_extents
== &fs_info
->freed_extents
[0])
4446 unpin
= &fs_info
->freed_extents
[1];
4448 unpin
= &fs_info
->freed_extents
[0];
4451 ret
= find_first_extent_bit(unpin
, 0, &start
, &end
,
4456 if (btrfs_test_opt(root
, DISCARD
))
4457 ret
= btrfs_discard_extent(root
, start
,
4458 end
+ 1 - start
, NULL
);
4460 clear_extent_dirty(unpin
, start
, end
, GFP_NOFS
);
4461 unpin_extent_range(root
, start
, end
);
4468 static int __btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4469 struct btrfs_root
*root
,
4470 u64 bytenr
, u64 num_bytes
, u64 parent
,
4471 u64 root_objectid
, u64 owner_objectid
,
4472 u64 owner_offset
, int refs_to_drop
,
4473 struct btrfs_delayed_extent_op
*extent_op
)
4475 struct btrfs_key key
;
4476 struct btrfs_path
*path
;
4477 struct btrfs_fs_info
*info
= root
->fs_info
;
4478 struct btrfs_root
*extent_root
= info
->extent_root
;
4479 struct extent_buffer
*leaf
;
4480 struct btrfs_extent_item
*ei
;
4481 struct btrfs_extent_inline_ref
*iref
;
4484 int extent_slot
= 0;
4485 int found_extent
= 0;
4490 path
= btrfs_alloc_path();
4495 path
->leave_spinning
= 1;
4497 is_data
= owner_objectid
>= BTRFS_FIRST_FREE_OBJECTID
;
4498 BUG_ON(!is_data
&& refs_to_drop
!= 1);
4500 ret
= lookup_extent_backref(trans
, extent_root
, path
, &iref
,
4501 bytenr
, num_bytes
, parent
,
4502 root_objectid
, owner_objectid
,
4505 extent_slot
= path
->slots
[0];
4506 while (extent_slot
>= 0) {
4507 btrfs_item_key_to_cpu(path
->nodes
[0], &key
,
4509 if (key
.objectid
!= bytenr
)
4511 if (key
.type
== BTRFS_EXTENT_ITEM_KEY
&&
4512 key
.offset
== num_bytes
) {
4516 if (path
->slots
[0] - extent_slot
> 5)
4520 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4521 item_size
= btrfs_item_size_nr(path
->nodes
[0], extent_slot
);
4522 if (found_extent
&& item_size
< sizeof(*ei
))
4525 if (!found_extent
) {
4527 ret
= remove_extent_backref(trans
, extent_root
, path
,
4531 btrfs_release_path(path
);
4532 path
->leave_spinning
= 1;
4534 key
.objectid
= bytenr
;
4535 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4536 key
.offset
= num_bytes
;
4538 ret
= btrfs_search_slot(trans
, extent_root
,
4541 printk(KERN_ERR
"umm, got %d back from search"
4542 ", was looking for %llu\n", ret
,
4543 (unsigned long long)bytenr
);
4545 btrfs_print_leaf(extent_root
,
4549 extent_slot
= path
->slots
[0];
4552 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4554 printk(KERN_ERR
"btrfs unable to find ref byte nr %llu "
4555 "parent %llu root %llu owner %llu offset %llu\n",
4556 (unsigned long long)bytenr
,
4557 (unsigned long long)parent
,
4558 (unsigned long long)root_objectid
,
4559 (unsigned long long)owner_objectid
,
4560 (unsigned long long)owner_offset
);
4563 leaf
= path
->nodes
[0];
4564 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4565 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4566 if (item_size
< sizeof(*ei
)) {
4567 BUG_ON(found_extent
|| extent_slot
!= path
->slots
[0]);
4568 ret
= convert_extent_item_v0(trans
, extent_root
, path
,
4572 btrfs_release_path(path
);
4573 path
->leave_spinning
= 1;
4575 key
.objectid
= bytenr
;
4576 key
.type
= BTRFS_EXTENT_ITEM_KEY
;
4577 key
.offset
= num_bytes
;
4579 ret
= btrfs_search_slot(trans
, extent_root
, &key
, path
,
4582 printk(KERN_ERR
"umm, got %d back from search"
4583 ", was looking for %llu\n", ret
,
4584 (unsigned long long)bytenr
);
4585 btrfs_print_leaf(extent_root
, path
->nodes
[0]);
4588 extent_slot
= path
->slots
[0];
4589 leaf
= path
->nodes
[0];
4590 item_size
= btrfs_item_size_nr(leaf
, extent_slot
);
4593 BUG_ON(item_size
< sizeof(*ei
));
4594 ei
= btrfs_item_ptr(leaf
, extent_slot
,
4595 struct btrfs_extent_item
);
4596 if (owner_objectid
< BTRFS_FIRST_FREE_OBJECTID
) {
4597 struct btrfs_tree_block_info
*bi
;
4598 BUG_ON(item_size
< sizeof(*ei
) + sizeof(*bi
));
4599 bi
= (struct btrfs_tree_block_info
*)(ei
+ 1);
4600 WARN_ON(owner_objectid
!= btrfs_tree_block_level(leaf
, bi
));
4603 refs
= btrfs_extent_refs(leaf
, ei
);
4604 BUG_ON(refs
< refs_to_drop
);
4605 refs
-= refs_to_drop
;
4609 __run_delayed_extent_op(extent_op
, leaf
, ei
);
4611 * In the case of inline back ref, reference count will
4612 * be updated by remove_extent_backref
4615 BUG_ON(!found_extent
);
4617 btrfs_set_extent_refs(leaf
, ei
, refs
);
4618 btrfs_mark_buffer_dirty(leaf
);
4621 ret
= remove_extent_backref(trans
, extent_root
, path
,
4628 BUG_ON(is_data
&& refs_to_drop
!=
4629 extent_data_ref_count(root
, path
, iref
));
4631 BUG_ON(path
->slots
[0] != extent_slot
);
4633 BUG_ON(path
->slots
[0] != extent_slot
+ 1);
4634 path
->slots
[0] = extent_slot
;
4639 ret
= btrfs_del_items(trans
, extent_root
, path
, path
->slots
[0],
4642 btrfs_release_path(path
);
4645 ret
= btrfs_del_csums(trans
, root
, bytenr
, num_bytes
);
4648 invalidate_mapping_pages(info
->btree_inode
->i_mapping
,
4649 bytenr
>> PAGE_CACHE_SHIFT
,
4650 (bytenr
+ num_bytes
- 1) >> PAGE_CACHE_SHIFT
);
4653 ret
= update_block_group(trans
, root
, bytenr
, num_bytes
, 0);
4656 btrfs_free_path(path
);
4661 * when we free an block, it is possible (and likely) that we free the last
4662 * delayed ref for that extent as well. This searches the delayed ref tree for
4663 * a given extent, and if there are no other delayed refs to be processed, it
4664 * removes it from the tree.
4666 static noinline
int check_ref_cleanup(struct btrfs_trans_handle
*trans
,
4667 struct btrfs_root
*root
, u64 bytenr
)
4669 struct btrfs_delayed_ref_head
*head
;
4670 struct btrfs_delayed_ref_root
*delayed_refs
;
4671 struct btrfs_delayed_ref_node
*ref
;
4672 struct rb_node
*node
;
4675 delayed_refs
= &trans
->transaction
->delayed_refs
;
4676 spin_lock(&delayed_refs
->lock
);
4677 head
= btrfs_find_delayed_ref_head(trans
, bytenr
);
4681 node
= rb_prev(&head
->node
.rb_node
);
4685 ref
= rb_entry(node
, struct btrfs_delayed_ref_node
, rb_node
);
4687 /* there are still entries for this ref, we can't drop it */
4688 if (ref
->bytenr
== bytenr
)
4691 if (head
->extent_op
) {
4692 if (!head
->must_insert_reserved
)
4694 kfree(head
->extent_op
);
4695 head
->extent_op
= NULL
;
4699 * waiting for the lock here would deadlock. If someone else has it
4700 * locked they are already in the process of dropping it anyway
4702 if (!mutex_trylock(&head
->mutex
))
4706 * at this point we have a head with no other entries. Go
4707 * ahead and process it.
4709 head
->node
.in_tree
= 0;
4710 rb_erase(&head
->node
.rb_node
, &delayed_refs
->root
);
4712 delayed_refs
->num_entries
--;
4715 * we don't take a ref on the node because we're removing it from the
4716 * tree, so we just steal the ref the tree was holding.
4718 delayed_refs
->num_heads
--;
4719 if (list_empty(&head
->cluster
))
4720 delayed_refs
->num_heads_ready
--;
4722 list_del_init(&head
->cluster
);
4723 spin_unlock(&delayed_refs
->lock
);
4725 BUG_ON(head
->extent_op
);
4726 if (head
->must_insert_reserved
)
4729 mutex_unlock(&head
->mutex
);
4730 btrfs_put_delayed_ref(&head
->node
);
4733 spin_unlock(&delayed_refs
->lock
);
4737 void btrfs_free_tree_block(struct btrfs_trans_handle
*trans
,
4738 struct btrfs_root
*root
,
4739 struct extent_buffer
*buf
,
4740 u64 parent
, int last_ref
)
4742 struct btrfs_block_group_cache
*cache
= NULL
;
4745 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4746 ret
= btrfs_add_delayed_tree_ref(trans
, buf
->start
, buf
->len
,
4747 parent
, root
->root_key
.objectid
,
4748 btrfs_header_level(buf
),
4749 BTRFS_DROP_DELAYED_REF
, NULL
);
4756 cache
= btrfs_lookup_block_group(root
->fs_info
, buf
->start
);
4758 if (btrfs_header_generation(buf
) == trans
->transid
) {
4759 if (root
->root_key
.objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
4760 ret
= check_ref_cleanup(trans
, root
, buf
->start
);
4765 if (btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
4766 pin_down_extent(root
, cache
, buf
->start
, buf
->len
, 1);
4770 WARN_ON(test_bit(EXTENT_BUFFER_DIRTY
, &buf
->bflags
));
4772 btrfs_add_free_space(cache
, buf
->start
, buf
->len
);
4773 btrfs_update_reserved_bytes(cache
, buf
->len
, RESERVE_FREE
);
4777 * Deleting the buffer, clear the corrupt flag since it doesn't matter
4780 clear_bit(EXTENT_BUFFER_CORRUPT
, &buf
->bflags
);
4781 btrfs_put_block_group(cache
);
4784 int btrfs_free_extent(struct btrfs_trans_handle
*trans
,
4785 struct btrfs_root
*root
,
4786 u64 bytenr
, u64 num_bytes
, u64 parent
,
4787 u64 root_objectid
, u64 owner
, u64 offset
)
4792 * tree log blocks never actually go into the extent allocation
4793 * tree, just update pinning info and exit early.
4795 if (root_objectid
== BTRFS_TREE_LOG_OBJECTID
) {
4796 WARN_ON(owner
>= BTRFS_FIRST_FREE_OBJECTID
);
4797 /* unlocks the pinned mutex */
4798 btrfs_pin_extent(root
, bytenr
, num_bytes
, 1);
4800 } else if (owner
< BTRFS_FIRST_FREE_OBJECTID
) {
4801 ret
= btrfs_add_delayed_tree_ref(trans
, bytenr
, num_bytes
,
4802 parent
, root_objectid
, (int)owner
,
4803 BTRFS_DROP_DELAYED_REF
, NULL
);
4806 ret
= btrfs_add_delayed_data_ref(trans
, bytenr
, num_bytes
,
4807 parent
, root_objectid
, owner
,
4808 offset
, BTRFS_DROP_DELAYED_REF
, NULL
);
4814 static u64
stripe_align(struct btrfs_root
*root
, u64 val
)
4816 u64 mask
= ((u64
)root
->stripesize
- 1);
4817 u64 ret
= (val
+ mask
) & ~mask
;
4822 * when we wait for progress in the block group caching, its because
4823 * our allocation attempt failed at least once. So, we must sleep
4824 * and let some progress happen before we try again.
4826 * This function will sleep at least once waiting for new free space to
4827 * show up, and then it will check the block group free space numbers
4828 * for our min num_bytes. Another option is to have it go ahead
4829 * and look in the rbtree for a free extent of a given size, but this
4833 wait_block_group_cache_progress(struct btrfs_block_group_cache
*cache
,
4836 struct btrfs_caching_control
*caching_ctl
;
4839 caching_ctl
= get_caching_control(cache
);
4843 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
) ||
4844 (cache
->free_space_ctl
->free_space
>= num_bytes
));
4846 put_caching_control(caching_ctl
);
4851 wait_block_group_cache_done(struct btrfs_block_group_cache
*cache
)
4853 struct btrfs_caching_control
*caching_ctl
;
4856 caching_ctl
= get_caching_control(cache
);
4860 wait_event(caching_ctl
->wait
, block_group_cache_done(cache
));
4862 put_caching_control(caching_ctl
);
4866 static int get_block_group_index(struct btrfs_block_group_cache
*cache
)
4869 if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID10
)
4871 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID1
)
4873 else if (cache
->flags
& BTRFS_BLOCK_GROUP_DUP
)
4875 else if (cache
->flags
& BTRFS_BLOCK_GROUP_RAID0
)
4882 enum btrfs_loop_type
{
4883 LOOP_FIND_IDEAL
= 0,
4884 LOOP_CACHING_NOWAIT
= 1,
4885 LOOP_CACHING_WAIT
= 2,
4886 LOOP_ALLOC_CHUNK
= 3,
4887 LOOP_NO_EMPTY_SIZE
= 4,
4891 * walks the btree of allocated extents and find a hole of a given size.
4892 * The key ins is changed to record the hole:
4893 * ins->objectid == block start
4894 * ins->flags = BTRFS_EXTENT_ITEM_KEY
4895 * ins->offset == number of blocks
4896 * Any available blocks before search_start are skipped.
4898 static noinline
int find_free_extent(struct btrfs_trans_handle
*trans
,
4899 struct btrfs_root
*orig_root
,
4900 u64 num_bytes
, u64 empty_size
,
4901 u64 search_start
, u64 search_end
,
4902 u64 hint_byte
, struct btrfs_key
*ins
,
4906 struct btrfs_root
*root
= orig_root
->fs_info
->extent_root
;
4907 struct btrfs_free_cluster
*last_ptr
= NULL
;
4908 struct btrfs_block_group_cache
*block_group
= NULL
;
4909 int empty_cluster
= 2 * 1024 * 1024;
4910 int allowed_chunk_alloc
= 0;
4911 int done_chunk_alloc
= 0;
4912 struct btrfs_space_info
*space_info
;
4913 int last_ptr_loop
= 0;
4916 int alloc_type
= (data
& BTRFS_BLOCK_GROUP_DATA
) ?
4917 RESERVE_ALLOC_NO_ACCOUNT
: RESERVE_ALLOC
;
4918 bool found_uncached_bg
= false;
4919 bool failed_cluster_refill
= false;
4920 bool failed_alloc
= false;
4921 bool use_cluster
= true;
4922 u64 ideal_cache_percent
= 0;
4923 u64 ideal_cache_offset
= 0;
4925 WARN_ON(num_bytes
< root
->sectorsize
);
4926 btrfs_set_key_type(ins
, BTRFS_EXTENT_ITEM_KEY
);
4930 space_info
= __find_space_info(root
->fs_info
, data
);
4932 printk(KERN_ERR
"No space info for %llu\n", data
);
4937 * If the space info is for both data and metadata it means we have a
4938 * small filesystem and we can't use the clustering stuff.
4940 if (btrfs_mixed_space_info(space_info
))
4941 use_cluster
= false;
4943 if (orig_root
->ref_cows
|| empty_size
)
4944 allowed_chunk_alloc
= 1;
4946 if (data
& BTRFS_BLOCK_GROUP_METADATA
&& use_cluster
) {
4947 last_ptr
= &root
->fs_info
->meta_alloc_cluster
;
4948 if (!btrfs_test_opt(root
, SSD
))
4949 empty_cluster
= 64 * 1024;
4952 if ((data
& BTRFS_BLOCK_GROUP_DATA
) && use_cluster
&&
4953 btrfs_test_opt(root
, SSD
)) {
4954 last_ptr
= &root
->fs_info
->data_alloc_cluster
;
4958 spin_lock(&last_ptr
->lock
);
4959 if (last_ptr
->block_group
)
4960 hint_byte
= last_ptr
->window_start
;
4961 spin_unlock(&last_ptr
->lock
);
4964 search_start
= max(search_start
, first_logical_byte(root
, 0));
4965 search_start
= max(search_start
, hint_byte
);
4970 if (search_start
== hint_byte
) {
4972 block_group
= btrfs_lookup_block_group(root
->fs_info
,
4975 * we don't want to use the block group if it doesn't match our
4976 * allocation bits, or if its not cached.
4978 * However if we are re-searching with an ideal block group
4979 * picked out then we don't care that the block group is cached.
4981 if (block_group
&& block_group_bits(block_group
, data
) &&
4982 (block_group
->cached
!= BTRFS_CACHE_NO
||
4983 search_start
== ideal_cache_offset
)) {
4984 down_read(&space_info
->groups_sem
);
4985 if (list_empty(&block_group
->list
) ||
4988 * someone is removing this block group,
4989 * we can't jump into the have_block_group
4990 * target because our list pointers are not
4993 btrfs_put_block_group(block_group
);
4994 up_read(&space_info
->groups_sem
);
4996 index
= get_block_group_index(block_group
);
4997 goto have_block_group
;
4999 } else if (block_group
) {
5000 btrfs_put_block_group(block_group
);
5004 down_read(&space_info
->groups_sem
);
5005 list_for_each_entry(block_group
, &space_info
->block_groups
[index
],
5010 btrfs_get_block_group(block_group
);
5011 search_start
= block_group
->key
.objectid
;
5014 * this can happen if we end up cycling through all the
5015 * raid types, but we want to make sure we only allocate
5016 * for the proper type.
5018 if (!block_group_bits(block_group
, data
)) {
5019 u64 extra
= BTRFS_BLOCK_GROUP_DUP
|
5020 BTRFS_BLOCK_GROUP_RAID1
|
5021 BTRFS_BLOCK_GROUP_RAID10
;
5024 * if they asked for extra copies and this block group
5025 * doesn't provide them, bail. This does allow us to
5026 * fill raid0 from raid1.
5028 if ((data
& extra
) && !(block_group
->flags
& extra
))
5033 if (unlikely(block_group
->cached
== BTRFS_CACHE_NO
)) {
5036 ret
= cache_block_group(block_group
, trans
,
5038 if (block_group
->cached
== BTRFS_CACHE_FINISHED
)
5039 goto have_block_group
;
5041 free_percent
= btrfs_block_group_used(&block_group
->item
);
5042 free_percent
*= 100;
5043 free_percent
= div64_u64(free_percent
,
5044 block_group
->key
.offset
);
5045 free_percent
= 100 - free_percent
;
5046 if (free_percent
> ideal_cache_percent
&&
5047 likely(!block_group
->ro
)) {
5048 ideal_cache_offset
= block_group
->key
.objectid
;
5049 ideal_cache_percent
= free_percent
;
5053 * The caching workers are limited to 2 threads, so we
5054 * can queue as much work as we care to.
5056 if (loop
> LOOP_FIND_IDEAL
) {
5057 ret
= cache_block_group(block_group
, trans
,
5061 found_uncached_bg
= true;
5064 * If loop is set for cached only, try the next block
5067 if (loop
== LOOP_FIND_IDEAL
)
5071 cached
= block_group_cache_done(block_group
);
5072 if (unlikely(!cached
))
5073 found_uncached_bg
= true;
5075 if (unlikely(block_group
->ro
))
5078 spin_lock(&block_group
->free_space_ctl
->tree_lock
);
5080 block_group
->free_space_ctl
->free_space
<
5081 num_bytes
+ empty_size
) {
5082 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5085 spin_unlock(&block_group
->free_space_ctl
->tree_lock
);
5088 * Ok we want to try and use the cluster allocator, so lets look
5089 * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will
5090 * have tried the cluster allocator plenty of times at this
5091 * point and not have found anything, so we are likely way too
5092 * fragmented for the clustering stuff to find anything, so lets
5093 * just skip it and let the allocator find whatever block it can
5096 if (last_ptr
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5098 * the refill lock keeps out other
5099 * people trying to start a new cluster
5101 spin_lock(&last_ptr
->refill_lock
);
5102 if (last_ptr
->block_group
&&
5103 (last_ptr
->block_group
->ro
||
5104 !block_group_bits(last_ptr
->block_group
, data
))) {
5106 goto refill_cluster
;
5109 offset
= btrfs_alloc_from_cluster(block_group
, last_ptr
,
5110 num_bytes
, search_start
);
5112 /* we have a block, we're done */
5113 spin_unlock(&last_ptr
->refill_lock
);
5117 spin_lock(&last_ptr
->lock
);
5119 * whoops, this cluster doesn't actually point to
5120 * this block group. Get a ref on the block
5121 * group is does point to and try again
5123 if (!last_ptr_loop
&& last_ptr
->block_group
&&
5124 last_ptr
->block_group
!= block_group
&&
5126 get_block_group_index(last_ptr
->block_group
)) {
5128 btrfs_put_block_group(block_group
);
5129 block_group
= last_ptr
->block_group
;
5130 btrfs_get_block_group(block_group
);
5131 spin_unlock(&last_ptr
->lock
);
5132 spin_unlock(&last_ptr
->refill_lock
);
5135 search_start
= block_group
->key
.objectid
;
5137 * we know this block group is properly
5138 * in the list because
5139 * btrfs_remove_block_group, drops the
5140 * cluster before it removes the block
5141 * group from the list
5143 goto have_block_group
;
5145 spin_unlock(&last_ptr
->lock
);
5148 * this cluster didn't work out, free it and
5151 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5155 /* allocate a cluster in this block group */
5156 ret
= btrfs_find_space_cluster(trans
, root
,
5157 block_group
, last_ptr
,
5159 empty_cluster
+ empty_size
);
5162 * now pull our allocation out of this
5165 offset
= btrfs_alloc_from_cluster(block_group
,
5166 last_ptr
, num_bytes
,
5169 /* we found one, proceed */
5170 spin_unlock(&last_ptr
->refill_lock
);
5173 } else if (!cached
&& loop
> LOOP_CACHING_NOWAIT
5174 && !failed_cluster_refill
) {
5175 spin_unlock(&last_ptr
->refill_lock
);
5177 failed_cluster_refill
= true;
5178 wait_block_group_cache_progress(block_group
,
5179 num_bytes
+ empty_cluster
+ empty_size
);
5180 goto have_block_group
;
5184 * at this point we either didn't find a cluster
5185 * or we weren't able to allocate a block from our
5186 * cluster. Free the cluster we've been trying
5187 * to use, and go to the next block group
5189 btrfs_return_cluster_to_free_space(NULL
, last_ptr
);
5190 spin_unlock(&last_ptr
->refill_lock
);
5194 offset
= btrfs_find_space_for_alloc(block_group
, search_start
,
5195 num_bytes
, empty_size
);
5197 * If we didn't find a chunk, and we haven't failed on this
5198 * block group before, and this block group is in the middle of
5199 * caching and we are ok with waiting, then go ahead and wait
5200 * for progress to be made, and set failed_alloc to true.
5202 * If failed_alloc is true then we've already waited on this
5203 * block group once and should move on to the next block group.
5205 if (!offset
&& !failed_alloc
&& !cached
&&
5206 loop
> LOOP_CACHING_NOWAIT
) {
5207 wait_block_group_cache_progress(block_group
,
5208 num_bytes
+ empty_size
);
5209 failed_alloc
= true;
5210 goto have_block_group
;
5211 } else if (!offset
) {
5215 search_start
= stripe_align(root
, offset
);
5216 /* move on to the next group */
5217 if (search_start
+ num_bytes
>= search_end
) {
5218 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5222 /* move on to the next group */
5223 if (search_start
+ num_bytes
>
5224 block_group
->key
.objectid
+ block_group
->key
.offset
) {
5225 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5229 ins
->objectid
= search_start
;
5230 ins
->offset
= num_bytes
;
5232 if (offset
< search_start
)
5233 btrfs_add_free_space(block_group
, offset
,
5234 search_start
- offset
);
5235 BUG_ON(offset
> search_start
);
5237 ret
= btrfs_update_reserved_bytes(block_group
, num_bytes
,
5239 if (ret
== -EAGAIN
) {
5240 btrfs_add_free_space(block_group
, offset
, num_bytes
);
5244 /* we are all good, lets return */
5245 ins
->objectid
= search_start
;
5246 ins
->offset
= num_bytes
;
5248 if (offset
< search_start
)
5249 btrfs_add_free_space(block_group
, offset
,
5250 search_start
- offset
);
5251 BUG_ON(offset
> search_start
);
5252 btrfs_put_block_group(block_group
);
5255 failed_cluster_refill
= false;
5256 failed_alloc
= false;
5257 BUG_ON(index
!= get_block_group_index(block_group
));
5258 btrfs_put_block_group(block_group
);
5260 up_read(&space_info
->groups_sem
);
5262 if (!ins
->objectid
&& ++index
< BTRFS_NR_RAID_TYPES
)
5265 /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
5266 * for them to make caching progress. Also
5267 * determine the best possible bg to cache
5268 * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
5269 * caching kthreads as we move along
5270 * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
5271 * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
5272 * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
5275 if (!ins
->objectid
&& loop
< LOOP_NO_EMPTY_SIZE
) {
5277 if (loop
== LOOP_FIND_IDEAL
&& found_uncached_bg
) {
5278 found_uncached_bg
= false;
5280 if (!ideal_cache_percent
)
5284 * 1 of the following 2 things have happened so far
5286 * 1) We found an ideal block group for caching that
5287 * is mostly full and will cache quickly, so we might
5288 * as well wait for it.
5290 * 2) We searched for cached only and we didn't find
5291 * anything, and we didn't start any caching kthreads
5292 * either, so chances are we will loop through and
5293 * start a couple caching kthreads, and then come back
5294 * around and just wait for them. This will be slower
5295 * because we will have 2 caching kthreads reading at
5296 * the same time when we could have just started one
5297 * and waited for it to get far enough to give us an
5298 * allocation, so go ahead and go to the wait caching
5301 loop
= LOOP_CACHING_WAIT
;
5302 search_start
= ideal_cache_offset
;
5303 ideal_cache_percent
= 0;
5305 } else if (loop
== LOOP_FIND_IDEAL
) {
5307 * Didn't find a uncached bg, wait on anything we find
5310 loop
= LOOP_CACHING_WAIT
;
5316 if (loop
== LOOP_ALLOC_CHUNK
) {
5317 if (allowed_chunk_alloc
) {
5318 ret
= do_chunk_alloc(trans
, root
, num_bytes
+
5319 2 * 1024 * 1024, data
,
5320 CHUNK_ALLOC_LIMITED
);
5321 allowed_chunk_alloc
= 0;
5323 done_chunk_alloc
= 1;
5324 } else if (!done_chunk_alloc
&&
5325 space_info
->force_alloc
==
5326 CHUNK_ALLOC_NO_FORCE
) {
5327 space_info
->force_alloc
= CHUNK_ALLOC_LIMITED
;
5331 * We didn't allocate a chunk, go ahead and drop the
5332 * empty size and loop again.
5334 if (!done_chunk_alloc
)
5335 loop
= LOOP_NO_EMPTY_SIZE
;
5338 if (loop
== LOOP_NO_EMPTY_SIZE
) {
5344 } else if (!ins
->objectid
) {
5346 } else if (ins
->objectid
) {
5353 static void dump_space_info(struct btrfs_space_info
*info
, u64 bytes
,
5354 int dump_block_groups
)
5356 struct btrfs_block_group_cache
*cache
;
5359 spin_lock(&info
->lock
);
5360 printk(KERN_INFO
"space_info %llu has %llu free, is %sfull\n",
5361 (unsigned long long)info
->flags
,
5362 (unsigned long long)(info
->total_bytes
- info
->bytes_used
-
5363 info
->bytes_pinned
- info
->bytes_reserved
-
5364 info
->bytes_readonly
),
5365 (info
->full
) ? "" : "not ");
5366 printk(KERN_INFO
"space_info total=%llu, used=%llu, pinned=%llu, "
5367 "reserved=%llu, may_use=%llu, readonly=%llu\n",
5368 (unsigned long long)info
->total_bytes
,
5369 (unsigned long long)info
->bytes_used
,
5370 (unsigned long long)info
->bytes_pinned
,
5371 (unsigned long long)info
->bytes_reserved
,
5372 (unsigned long long)info
->bytes_may_use
,
5373 (unsigned long long)info
->bytes_readonly
);
5374 spin_unlock(&info
->lock
);
5376 if (!dump_block_groups
)
5379 down_read(&info
->groups_sem
);
5381 list_for_each_entry(cache
, &info
->block_groups
[index
], list
) {
5382 spin_lock(&cache
->lock
);
5383 printk(KERN_INFO
"block group %llu has %llu bytes, %llu used "
5384 "%llu pinned %llu reserved\n",
5385 (unsigned long long)cache
->key
.objectid
,
5386 (unsigned long long)cache
->key
.offset
,
5387 (unsigned long long)btrfs_block_group_used(&cache
->item
),
5388 (unsigned long long)cache
->pinned
,
5389 (unsigned long long)cache
->reserved
);
5390 btrfs_dump_free_space(cache
, bytes
);
5391 spin_unlock(&cache
->lock
);
5393 if (++index
< BTRFS_NR_RAID_TYPES
)
5395 up_read(&info
->groups_sem
);
5398 int btrfs_reserve_extent(struct btrfs_trans_handle
*trans
,
5399 struct btrfs_root
*root
,
5400 u64 num_bytes
, u64 min_alloc_size
,
5401 u64 empty_size
, u64 hint_byte
,
5402 u64 search_end
, struct btrfs_key
*ins
,
5406 u64 search_start
= 0;
5408 data
= btrfs_get_alloc_profile(root
, data
);
5411 * the only place that sets empty_size is btrfs_realloc_node, which
5412 * is not called recursively on allocations
5414 if (empty_size
|| root
->ref_cows
)
5415 ret
= do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5416 num_bytes
+ 2 * 1024 * 1024, data
,
5417 CHUNK_ALLOC_NO_FORCE
);
5419 WARN_ON(num_bytes
< root
->sectorsize
);
5420 ret
= find_free_extent(trans
, root
, num_bytes
, empty_size
,
5421 search_start
, search_end
, hint_byte
,
5424 if (ret
== -ENOSPC
&& num_bytes
> min_alloc_size
) {
5425 num_bytes
= num_bytes
>> 1;
5426 num_bytes
= num_bytes
& ~(root
->sectorsize
- 1);
5427 num_bytes
= max(num_bytes
, min_alloc_size
);
5428 do_chunk_alloc(trans
, root
->fs_info
->extent_root
,
5429 num_bytes
, data
, CHUNK_ALLOC_FORCE
);
5432 if (ret
== -ENOSPC
&& btrfs_test_opt(root
, ENOSPC_DEBUG
)) {
5433 struct btrfs_space_info
*sinfo
;
5435 sinfo
= __find_space_info(root
->fs_info
, data
);
5436 printk(KERN_ERR
"btrfs allocation failed flags %llu, "
5437 "wanted %llu\n", (unsigned long long)data
,
5438 (unsigned long long)num_bytes
);
5439 dump_space_info(sinfo
, num_bytes
, 1);
5442 trace_btrfs_reserved_extent_alloc(root
, ins
->objectid
, ins
->offset
);
5447 int btrfs_free_reserved_extent(struct btrfs_root
*root
, u64 start
, u64 len
)
5449 struct btrfs_block_group_cache
*cache
;
5452 cache
= btrfs_lookup_block_group(root
->fs_info
, start
);
5454 printk(KERN_ERR
"Unable to find block group for %llu\n",
5455 (unsigned long long)start
);
5459 if (btrfs_test_opt(root
, DISCARD
))
5460 ret
= btrfs_discard_extent(root
, start
, len
, NULL
);
5462 btrfs_add_free_space(cache
, start
, len
);
5463 btrfs_update_reserved_bytes(cache
, len
, RESERVE_FREE
);
5464 btrfs_put_block_group(cache
);
5466 trace_btrfs_reserved_extent_free(root
, start
, len
);
5471 static int alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5472 struct btrfs_root
*root
,
5473 u64 parent
, u64 root_objectid
,
5474 u64 flags
, u64 owner
, u64 offset
,
5475 struct btrfs_key
*ins
, int ref_mod
)
5478 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5479 struct btrfs_extent_item
*extent_item
;
5480 struct btrfs_extent_inline_ref
*iref
;
5481 struct btrfs_path
*path
;
5482 struct extent_buffer
*leaf
;
5487 type
= BTRFS_SHARED_DATA_REF_KEY
;
5489 type
= BTRFS_EXTENT_DATA_REF_KEY
;
5491 size
= sizeof(*extent_item
) + btrfs_extent_inline_ref_size(type
);
5493 path
= btrfs_alloc_path();
5497 path
->leave_spinning
= 1;
5498 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5502 leaf
= path
->nodes
[0];
5503 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5504 struct btrfs_extent_item
);
5505 btrfs_set_extent_refs(leaf
, extent_item
, ref_mod
);
5506 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5507 btrfs_set_extent_flags(leaf
, extent_item
,
5508 flags
| BTRFS_EXTENT_FLAG_DATA
);
5510 iref
= (struct btrfs_extent_inline_ref
*)(extent_item
+ 1);
5511 btrfs_set_extent_inline_ref_type(leaf
, iref
, type
);
5513 struct btrfs_shared_data_ref
*ref
;
5514 ref
= (struct btrfs_shared_data_ref
*)(iref
+ 1);
5515 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5516 btrfs_set_shared_data_ref_count(leaf
, ref
, ref_mod
);
5518 struct btrfs_extent_data_ref
*ref
;
5519 ref
= (struct btrfs_extent_data_ref
*)(&iref
->offset
);
5520 btrfs_set_extent_data_ref_root(leaf
, ref
, root_objectid
);
5521 btrfs_set_extent_data_ref_objectid(leaf
, ref
, owner
);
5522 btrfs_set_extent_data_ref_offset(leaf
, ref
, offset
);
5523 btrfs_set_extent_data_ref_count(leaf
, ref
, ref_mod
);
5526 btrfs_mark_buffer_dirty(path
->nodes
[0]);
5527 btrfs_free_path(path
);
5529 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5531 printk(KERN_ERR
"btrfs update block group failed for %llu "
5532 "%llu\n", (unsigned long long)ins
->objectid
,
5533 (unsigned long long)ins
->offset
);
5539 static int alloc_reserved_tree_block(struct btrfs_trans_handle
*trans
,
5540 struct btrfs_root
*root
,
5541 u64 parent
, u64 root_objectid
,
5542 u64 flags
, struct btrfs_disk_key
*key
,
5543 int level
, struct btrfs_key
*ins
)
5546 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
5547 struct btrfs_extent_item
*extent_item
;
5548 struct btrfs_tree_block_info
*block_info
;
5549 struct btrfs_extent_inline_ref
*iref
;
5550 struct btrfs_path
*path
;
5551 struct extent_buffer
*leaf
;
5552 u32 size
= sizeof(*extent_item
) + sizeof(*block_info
) + sizeof(*iref
);
5554 path
= btrfs_alloc_path();
5558 path
->leave_spinning
= 1;
5559 ret
= btrfs_insert_empty_item(trans
, fs_info
->extent_root
, path
,
5563 leaf
= path
->nodes
[0];
5564 extent_item
= btrfs_item_ptr(leaf
, path
->slots
[0],
5565 struct btrfs_extent_item
);
5566 btrfs_set_extent_refs(leaf
, extent_item
, 1);
5567 btrfs_set_extent_generation(leaf
, extent_item
, trans
->transid
);
5568 btrfs_set_extent_flags(leaf
, extent_item
,
5569 flags
| BTRFS_EXTENT_FLAG_TREE_BLOCK
);
5570 block_info
= (struct btrfs_tree_block_info
*)(extent_item
+ 1);
5572 btrfs_set_tree_block_key(leaf
, block_info
, key
);
5573 btrfs_set_tree_block_level(leaf
, block_info
, level
);
5575 iref
= (struct btrfs_extent_inline_ref
*)(block_info
+ 1);
5577 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
5578 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5579 BTRFS_SHARED_BLOCK_REF_KEY
);
5580 btrfs_set_extent_inline_ref_offset(leaf
, iref
, parent
);
5582 btrfs_set_extent_inline_ref_type(leaf
, iref
,
5583 BTRFS_TREE_BLOCK_REF_KEY
);
5584 btrfs_set_extent_inline_ref_offset(leaf
, iref
, root_objectid
);
5587 btrfs_mark_buffer_dirty(leaf
);
5588 btrfs_free_path(path
);
5590 ret
= update_block_group(trans
, root
, ins
->objectid
, ins
->offset
, 1);
5592 printk(KERN_ERR
"btrfs update block group failed for %llu "
5593 "%llu\n", (unsigned long long)ins
->objectid
,
5594 (unsigned long long)ins
->offset
);
5600 int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle
*trans
,
5601 struct btrfs_root
*root
,
5602 u64 root_objectid
, u64 owner
,
5603 u64 offset
, struct btrfs_key
*ins
)
5607 BUG_ON(root_objectid
== BTRFS_TREE_LOG_OBJECTID
);
5609 ret
= btrfs_add_delayed_data_ref(trans
, ins
->objectid
, ins
->offset
,
5610 0, root_objectid
, owner
, offset
,
5611 BTRFS_ADD_DELAYED_EXTENT
, NULL
);
5616 * this is used by the tree logging recovery code. It records that
5617 * an extent has been allocated and makes sure to clear the free
5618 * space cache bits as well
5620 int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle
*trans
,
5621 struct btrfs_root
*root
,
5622 u64 root_objectid
, u64 owner
, u64 offset
,
5623 struct btrfs_key
*ins
)
5626 struct btrfs_block_group_cache
*block_group
;
5627 struct btrfs_caching_control
*caching_ctl
;
5628 u64 start
= ins
->objectid
;
5629 u64 num_bytes
= ins
->offset
;
5631 block_group
= btrfs_lookup_block_group(root
->fs_info
, ins
->objectid
);
5632 cache_block_group(block_group
, trans
, NULL
, 0);
5633 caching_ctl
= get_caching_control(block_group
);
5636 BUG_ON(!block_group_cache_done(block_group
));
5637 ret
= btrfs_remove_free_space(block_group
, start
, num_bytes
);
5640 mutex_lock(&caching_ctl
->mutex
);
5642 if (start
>= caching_ctl
->progress
) {
5643 ret
= add_excluded_extent(root
, start
, num_bytes
);
5645 } else if (start
+ num_bytes
<= caching_ctl
->progress
) {
5646 ret
= btrfs_remove_free_space(block_group
,
5650 num_bytes
= caching_ctl
->progress
- start
;
5651 ret
= btrfs_remove_free_space(block_group
,
5655 start
= caching_ctl
->progress
;
5656 num_bytes
= ins
->objectid
+ ins
->offset
-
5657 caching_ctl
->progress
;
5658 ret
= add_excluded_extent(root
, start
, num_bytes
);
5662 mutex_unlock(&caching_ctl
->mutex
);
5663 put_caching_control(caching_ctl
);
5666 ret
= btrfs_update_reserved_bytes(block_group
, ins
->offset
,
5667 RESERVE_ALLOC_NO_ACCOUNT
);
5669 btrfs_put_block_group(block_group
);
5670 ret
= alloc_reserved_file_extent(trans
, root
, 0, root_objectid
,
5671 0, owner
, offset
, ins
, 1);
5675 struct extent_buffer
*btrfs_init_new_buffer(struct btrfs_trans_handle
*trans
,
5676 struct btrfs_root
*root
,
5677 u64 bytenr
, u32 blocksize
,
5680 struct extent_buffer
*buf
;
5682 buf
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
5684 return ERR_PTR(-ENOMEM
);
5685 btrfs_set_header_generation(buf
, trans
->transid
);
5686 btrfs_set_buffer_lockdep_class(root
->root_key
.objectid
, buf
, level
);
5687 btrfs_tree_lock(buf
);
5688 clean_tree_block(trans
, root
, buf
);
5690 btrfs_set_lock_blocking(buf
);
5691 btrfs_set_buffer_uptodate(buf
);
5693 if (root
->root_key
.objectid
== BTRFS_TREE_LOG_OBJECTID
) {
5695 * we allow two log transactions at a time, use different
5696 * EXENT bit to differentiate dirty pages.
5698 if (root
->log_transid
% 2 == 0)
5699 set_extent_dirty(&root
->dirty_log_pages
, buf
->start
,
5700 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5702 set_extent_new(&root
->dirty_log_pages
, buf
->start
,
5703 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5705 set_extent_dirty(&trans
->transaction
->dirty_pages
, buf
->start
,
5706 buf
->start
+ buf
->len
- 1, GFP_NOFS
);
5708 trans
->blocks_used
++;
5709 /* this returns a buffer locked for blocking */
5713 static struct btrfs_block_rsv
*
5714 use_block_rsv(struct btrfs_trans_handle
*trans
,
5715 struct btrfs_root
*root
, u32 blocksize
)
5717 struct btrfs_block_rsv
*block_rsv
;
5718 struct btrfs_block_rsv
*global_rsv
= &root
->fs_info
->global_block_rsv
;
5721 block_rsv
= get_block_rsv(trans
, root
);
5723 if (block_rsv
->size
== 0) {
5724 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
, 0, 0);
5726 * If we couldn't reserve metadata bytes try and use some from
5727 * the global reserve.
5729 if (ret
&& block_rsv
!= global_rsv
) {
5730 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5733 return ERR_PTR(ret
);
5735 return ERR_PTR(ret
);
5740 ret
= block_rsv_use_bytes(block_rsv
, blocksize
);
5745 ret
= reserve_metadata_bytes(root
, block_rsv
, blocksize
, 0, 0);
5748 } else if (ret
&& block_rsv
!= global_rsv
) {
5749 ret
= block_rsv_use_bytes(global_rsv
, blocksize
);
5755 return ERR_PTR(-ENOSPC
);
5758 static void unuse_block_rsv(struct btrfs_block_rsv
*block_rsv
, u32 blocksize
)
5760 block_rsv_add_bytes(block_rsv
, blocksize
, 0);
5761 block_rsv_release_bytes(block_rsv
, NULL
, 0);
5765 * finds a free extent and does all the dirty work required for allocation
5766 * returns the key for the extent through ins, and a tree buffer for
5767 * the first block of the extent through buf.
5769 * returns the tree buffer or NULL.
5771 struct extent_buffer
*btrfs_alloc_free_block(struct btrfs_trans_handle
*trans
,
5772 struct btrfs_root
*root
, u32 blocksize
,
5773 u64 parent
, u64 root_objectid
,
5774 struct btrfs_disk_key
*key
, int level
,
5775 u64 hint
, u64 empty_size
)
5777 struct btrfs_key ins
;
5778 struct btrfs_block_rsv
*block_rsv
;
5779 struct extent_buffer
*buf
;
5784 block_rsv
= use_block_rsv(trans
, root
, blocksize
);
5785 if (IS_ERR(block_rsv
))
5786 return ERR_CAST(block_rsv
);
5788 ret
= btrfs_reserve_extent(trans
, root
, blocksize
, blocksize
,
5789 empty_size
, hint
, (u64
)-1, &ins
, 0);
5791 unuse_block_rsv(block_rsv
, blocksize
);
5792 return ERR_PTR(ret
);
5795 buf
= btrfs_init_new_buffer(trans
, root
, ins
.objectid
,
5797 BUG_ON(IS_ERR(buf
));
5799 if (root_objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
5801 parent
= ins
.objectid
;
5802 flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5806 if (root_objectid
!= BTRFS_TREE_LOG_OBJECTID
) {
5807 struct btrfs_delayed_extent_op
*extent_op
;
5808 extent_op
= kmalloc(sizeof(*extent_op
), GFP_NOFS
);
5811 memcpy(&extent_op
->key
, key
, sizeof(extent_op
->key
));
5813 memset(&extent_op
->key
, 0, sizeof(extent_op
->key
));
5814 extent_op
->flags_to_set
= flags
;
5815 extent_op
->update_key
= 1;
5816 extent_op
->update_flags
= 1;
5817 extent_op
->is_data
= 0;
5819 ret
= btrfs_add_delayed_tree_ref(trans
, ins
.objectid
,
5820 ins
.offset
, parent
, root_objectid
,
5821 level
, BTRFS_ADD_DELAYED_EXTENT
,
5828 struct walk_control
{
5829 u64 refs
[BTRFS_MAX_LEVEL
];
5830 u64 flags
[BTRFS_MAX_LEVEL
];
5831 struct btrfs_key update_progress
;
5841 #define DROP_REFERENCE 1
5842 #define UPDATE_BACKREF 2
5844 static noinline
void reada_walk_down(struct btrfs_trans_handle
*trans
,
5845 struct btrfs_root
*root
,
5846 struct walk_control
*wc
,
5847 struct btrfs_path
*path
)
5855 struct btrfs_key key
;
5856 struct extent_buffer
*eb
;
5861 if (path
->slots
[wc
->level
] < wc
->reada_slot
) {
5862 wc
->reada_count
= wc
->reada_count
* 2 / 3;
5863 wc
->reada_count
= max(wc
->reada_count
, 2);
5865 wc
->reada_count
= wc
->reada_count
* 3 / 2;
5866 wc
->reada_count
= min_t(int, wc
->reada_count
,
5867 BTRFS_NODEPTRS_PER_BLOCK(root
));
5870 eb
= path
->nodes
[wc
->level
];
5871 nritems
= btrfs_header_nritems(eb
);
5872 blocksize
= btrfs_level_size(root
, wc
->level
- 1);
5874 for (slot
= path
->slots
[wc
->level
]; slot
< nritems
; slot
++) {
5875 if (nread
>= wc
->reada_count
)
5879 bytenr
= btrfs_node_blockptr(eb
, slot
);
5880 generation
= btrfs_node_ptr_generation(eb
, slot
);
5882 if (slot
== path
->slots
[wc
->level
])
5885 if (wc
->stage
== UPDATE_BACKREF
&&
5886 generation
<= root
->root_key
.offset
)
5889 /* We don't lock the tree block, it's OK to be racy here */
5890 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
5895 if (wc
->stage
== DROP_REFERENCE
) {
5899 if (wc
->level
== 1 &&
5900 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5902 if (!wc
->update_ref
||
5903 generation
<= root
->root_key
.offset
)
5905 btrfs_node_key_to_cpu(eb
, &key
, slot
);
5906 ret
= btrfs_comp_cpu_keys(&key
,
5907 &wc
->update_progress
);
5911 if (wc
->level
== 1 &&
5912 (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
))
5916 ret
= readahead_tree_block(root
, bytenr
, blocksize
,
5922 wc
->reada_slot
= slot
;
5926 * hepler to process tree block while walking down the tree.
5928 * when wc->stage == UPDATE_BACKREF, this function updates
5929 * back refs for pointers in the block.
5931 * NOTE: return value 1 means we should stop walking down.
5933 static noinline
int walk_down_proc(struct btrfs_trans_handle
*trans
,
5934 struct btrfs_root
*root
,
5935 struct btrfs_path
*path
,
5936 struct walk_control
*wc
, int lookup_info
)
5938 int level
= wc
->level
;
5939 struct extent_buffer
*eb
= path
->nodes
[level
];
5940 u64 flag
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
5943 if (wc
->stage
== UPDATE_BACKREF
&&
5944 btrfs_header_owner(eb
) != root
->root_key
.objectid
)
5948 * when reference count of tree block is 1, it won't increase
5949 * again. once full backref flag is set, we never clear it.
5952 ((wc
->stage
== DROP_REFERENCE
&& wc
->refs
[level
] != 1) ||
5953 (wc
->stage
== UPDATE_BACKREF
&& !(wc
->flags
[level
] & flag
)))) {
5954 BUG_ON(!path
->locks
[level
]);
5955 ret
= btrfs_lookup_extent_info(trans
, root
,
5960 BUG_ON(wc
->refs
[level
] == 0);
5963 if (wc
->stage
== DROP_REFERENCE
) {
5964 if (wc
->refs
[level
] > 1)
5967 if (path
->locks
[level
] && !wc
->keep_locks
) {
5968 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5969 path
->locks
[level
] = 0;
5974 /* wc->stage == UPDATE_BACKREF */
5975 if (!(wc
->flags
[level
] & flag
)) {
5976 BUG_ON(!path
->locks
[level
]);
5977 ret
= btrfs_inc_ref(trans
, root
, eb
, 1);
5979 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
5981 ret
= btrfs_set_disk_extent_flags(trans
, root
, eb
->start
,
5984 wc
->flags
[level
] |= flag
;
5988 * the block is shared by multiple trees, so it's not good to
5989 * keep the tree lock
5991 if (path
->locks
[level
] && level
> 0) {
5992 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
5993 path
->locks
[level
] = 0;
5999 * hepler to process tree block pointer.
6001 * when wc->stage == DROP_REFERENCE, this function checks
6002 * reference count of the block pointed to. if the block
6003 * is shared and we need update back refs for the subtree
6004 * rooted at the block, this function changes wc->stage to
6005 * UPDATE_BACKREF. if the block is shared and there is no
6006 * need to update back, this function drops the reference
6009 * NOTE: return value 1 means we should stop walking down.
6011 static noinline
int do_walk_down(struct btrfs_trans_handle
*trans
,
6012 struct btrfs_root
*root
,
6013 struct btrfs_path
*path
,
6014 struct walk_control
*wc
, int *lookup_info
)
6020 struct btrfs_key key
;
6021 struct extent_buffer
*next
;
6022 int level
= wc
->level
;
6026 generation
= btrfs_node_ptr_generation(path
->nodes
[level
],
6027 path
->slots
[level
]);
6029 * if the lower level block was created before the snapshot
6030 * was created, we know there is no need to update back refs
6033 if (wc
->stage
== UPDATE_BACKREF
&&
6034 generation
<= root
->root_key
.offset
) {
6039 bytenr
= btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]);
6040 blocksize
= btrfs_level_size(root
, level
- 1);
6042 next
= btrfs_find_tree_block(root
, bytenr
, blocksize
);
6044 next
= btrfs_find_create_tree_block(root
, bytenr
, blocksize
);
6049 btrfs_tree_lock(next
);
6050 btrfs_set_lock_blocking(next
);
6052 ret
= btrfs_lookup_extent_info(trans
, root
, bytenr
, blocksize
,
6053 &wc
->refs
[level
- 1],
6054 &wc
->flags
[level
- 1]);
6056 BUG_ON(wc
->refs
[level
- 1] == 0);
6059 if (wc
->stage
== DROP_REFERENCE
) {
6060 if (wc
->refs
[level
- 1] > 1) {
6062 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6065 if (!wc
->update_ref
||
6066 generation
<= root
->root_key
.offset
)
6069 btrfs_node_key_to_cpu(path
->nodes
[level
], &key
,
6070 path
->slots
[level
]);
6071 ret
= btrfs_comp_cpu_keys(&key
, &wc
->update_progress
);
6075 wc
->stage
= UPDATE_BACKREF
;
6076 wc
->shared_level
= level
- 1;
6080 (wc
->flags
[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF
))
6084 if (!btrfs_buffer_uptodate(next
, generation
)) {
6085 btrfs_tree_unlock(next
);
6086 free_extent_buffer(next
);
6092 if (reada
&& level
== 1)
6093 reada_walk_down(trans
, root
, wc
, path
);
6094 next
= read_tree_block(root
, bytenr
, blocksize
, generation
);
6097 btrfs_tree_lock(next
);
6098 btrfs_set_lock_blocking(next
);
6102 BUG_ON(level
!= btrfs_header_level(next
));
6103 path
->nodes
[level
] = next
;
6104 path
->slots
[level
] = 0;
6105 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6111 wc
->refs
[level
- 1] = 0;
6112 wc
->flags
[level
- 1] = 0;
6113 if (wc
->stage
== DROP_REFERENCE
) {
6114 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
6115 parent
= path
->nodes
[level
]->start
;
6117 BUG_ON(root
->root_key
.objectid
!=
6118 btrfs_header_owner(path
->nodes
[level
]));
6122 ret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
, parent
,
6123 root
->root_key
.objectid
, level
- 1, 0);
6126 btrfs_tree_unlock(next
);
6127 free_extent_buffer(next
);
6133 * hepler to process tree block while walking up the tree.
6135 * when wc->stage == DROP_REFERENCE, this function drops
6136 * reference count on the block.
6138 * when wc->stage == UPDATE_BACKREF, this function changes
6139 * wc->stage back to DROP_REFERENCE if we changed wc->stage
6140 * to UPDATE_BACKREF previously while processing the block.
6142 * NOTE: return value 1 means we should stop walking up.
6144 static noinline
int walk_up_proc(struct btrfs_trans_handle
*trans
,
6145 struct btrfs_root
*root
,
6146 struct btrfs_path
*path
,
6147 struct walk_control
*wc
)
6150 int level
= wc
->level
;
6151 struct extent_buffer
*eb
= path
->nodes
[level
];
6154 if (wc
->stage
== UPDATE_BACKREF
) {
6155 BUG_ON(wc
->shared_level
< level
);
6156 if (level
< wc
->shared_level
)
6159 ret
= find_next_key(path
, level
+ 1, &wc
->update_progress
);
6163 wc
->stage
= DROP_REFERENCE
;
6164 wc
->shared_level
= -1;
6165 path
->slots
[level
] = 0;
6168 * check reference count again if the block isn't locked.
6169 * we should start walking down the tree again if reference
6172 if (!path
->locks
[level
]) {
6174 btrfs_tree_lock(eb
);
6175 btrfs_set_lock_blocking(eb
);
6176 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6178 ret
= btrfs_lookup_extent_info(trans
, root
,
6183 BUG_ON(wc
->refs
[level
] == 0);
6184 if (wc
->refs
[level
] == 1) {
6185 btrfs_tree_unlock_rw(eb
, path
->locks
[level
]);
6191 /* wc->stage == DROP_REFERENCE */
6192 BUG_ON(wc
->refs
[level
] > 1 && !path
->locks
[level
]);
6194 if (wc
->refs
[level
] == 1) {
6196 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6197 ret
= btrfs_dec_ref(trans
, root
, eb
, 1);
6199 ret
= btrfs_dec_ref(trans
, root
, eb
, 0);
6202 /* make block locked assertion in clean_tree_block happy */
6203 if (!path
->locks
[level
] &&
6204 btrfs_header_generation(eb
) == trans
->transid
) {
6205 btrfs_tree_lock(eb
);
6206 btrfs_set_lock_blocking(eb
);
6207 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6209 clean_tree_block(trans
, root
, eb
);
6212 if (eb
== root
->node
) {
6213 if (wc
->flags
[level
] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6216 BUG_ON(root
->root_key
.objectid
!=
6217 btrfs_header_owner(eb
));
6219 if (wc
->flags
[level
+ 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF
)
6220 parent
= path
->nodes
[level
+ 1]->start
;
6222 BUG_ON(root
->root_key
.objectid
!=
6223 btrfs_header_owner(path
->nodes
[level
+ 1]));
6226 btrfs_free_tree_block(trans
, root
, eb
, parent
, wc
->refs
[level
] == 1);
6228 wc
->refs
[level
] = 0;
6229 wc
->flags
[level
] = 0;
6233 static noinline
int walk_down_tree(struct btrfs_trans_handle
*trans
,
6234 struct btrfs_root
*root
,
6235 struct btrfs_path
*path
,
6236 struct walk_control
*wc
)
6238 int level
= wc
->level
;
6239 int lookup_info
= 1;
6242 while (level
>= 0) {
6243 ret
= walk_down_proc(trans
, root
, path
, wc
, lookup_info
);
6250 if (path
->slots
[level
] >=
6251 btrfs_header_nritems(path
->nodes
[level
]))
6254 ret
= do_walk_down(trans
, root
, path
, wc
, &lookup_info
);
6256 path
->slots
[level
]++;
6265 static noinline
int walk_up_tree(struct btrfs_trans_handle
*trans
,
6266 struct btrfs_root
*root
,
6267 struct btrfs_path
*path
,
6268 struct walk_control
*wc
, int max_level
)
6270 int level
= wc
->level
;
6273 path
->slots
[level
] = btrfs_header_nritems(path
->nodes
[level
]);
6274 while (level
< max_level
&& path
->nodes
[level
]) {
6276 if (path
->slots
[level
] + 1 <
6277 btrfs_header_nritems(path
->nodes
[level
])) {
6278 path
->slots
[level
]++;
6281 ret
= walk_up_proc(trans
, root
, path
, wc
);
6285 if (path
->locks
[level
]) {
6286 btrfs_tree_unlock_rw(path
->nodes
[level
],
6287 path
->locks
[level
]);
6288 path
->locks
[level
] = 0;
6290 free_extent_buffer(path
->nodes
[level
]);
6291 path
->nodes
[level
] = NULL
;
6299 * drop a subvolume tree.
6301 * this function traverses the tree freeing any blocks that only
6302 * referenced by the tree.
6304 * when a shared tree block is found. this function decreases its
6305 * reference count by one. if update_ref is true, this function
6306 * also make sure backrefs for the shared block and all lower level
6307 * blocks are properly updated.
6309 void btrfs_drop_snapshot(struct btrfs_root
*root
,
6310 struct btrfs_block_rsv
*block_rsv
, int update_ref
)
6312 struct btrfs_path
*path
;
6313 struct btrfs_trans_handle
*trans
;
6314 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
6315 struct btrfs_root_item
*root_item
= &root
->root_item
;
6316 struct walk_control
*wc
;
6317 struct btrfs_key key
;
6322 path
= btrfs_alloc_path();
6328 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6330 btrfs_free_path(path
);
6335 trans
= btrfs_start_transaction(tree_root
, 0);
6336 BUG_ON(IS_ERR(trans
));
6339 trans
->block_rsv
= block_rsv
;
6341 if (btrfs_disk_key_objectid(&root_item
->drop_progress
) == 0) {
6342 level
= btrfs_header_level(root
->node
);
6343 path
->nodes
[level
] = btrfs_lock_root_node(root
);
6344 btrfs_set_lock_blocking(path
->nodes
[level
]);
6345 path
->slots
[level
] = 0;
6346 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6347 memset(&wc
->update_progress
, 0,
6348 sizeof(wc
->update_progress
));
6350 btrfs_disk_key_to_cpu(&key
, &root_item
->drop_progress
);
6351 memcpy(&wc
->update_progress
, &key
,
6352 sizeof(wc
->update_progress
));
6354 level
= root_item
->drop_level
;
6356 path
->lowest_level
= level
;
6357 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
6358 path
->lowest_level
= 0;
6366 * unlock our path, this is safe because only this
6367 * function is allowed to delete this snapshot
6369 btrfs_unlock_up_safe(path
, 0);
6371 level
= btrfs_header_level(root
->node
);
6373 btrfs_tree_lock(path
->nodes
[level
]);
6374 btrfs_set_lock_blocking(path
->nodes
[level
]);
6376 ret
= btrfs_lookup_extent_info(trans
, root
,
6377 path
->nodes
[level
]->start
,
6378 path
->nodes
[level
]->len
,
6382 BUG_ON(wc
->refs
[level
] == 0);
6384 if (level
== root_item
->drop_level
)
6387 btrfs_tree_unlock(path
->nodes
[level
]);
6388 WARN_ON(wc
->refs
[level
] != 1);
6394 wc
->shared_level
= -1;
6395 wc
->stage
= DROP_REFERENCE
;
6396 wc
->update_ref
= update_ref
;
6398 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6401 ret
= walk_down_tree(trans
, root
, path
, wc
);
6407 ret
= walk_up_tree(trans
, root
, path
, wc
, BTRFS_MAX_LEVEL
);
6414 BUG_ON(wc
->stage
!= DROP_REFERENCE
);
6418 if (wc
->stage
== DROP_REFERENCE
) {
6420 btrfs_node_key(path
->nodes
[level
],
6421 &root_item
->drop_progress
,
6422 path
->slots
[level
]);
6423 root_item
->drop_level
= level
;
6426 BUG_ON(wc
->level
== 0);
6427 if (btrfs_should_end_transaction(trans
, tree_root
)) {
6428 ret
= btrfs_update_root(trans
, tree_root
,
6433 btrfs_end_transaction_throttle(trans
, tree_root
);
6434 trans
= btrfs_start_transaction(tree_root
, 0);
6435 BUG_ON(IS_ERR(trans
));
6437 trans
->block_rsv
= block_rsv
;
6440 btrfs_release_path(path
);
6443 ret
= btrfs_del_root(trans
, tree_root
, &root
->root_key
);
6446 if (root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
) {
6447 ret
= btrfs_find_last_root(tree_root
, root
->root_key
.objectid
,
6451 /* if we fail to delete the orphan item this time
6452 * around, it'll get picked up the next time.
6454 * The most common failure here is just -ENOENT.
6456 btrfs_del_orphan_item(trans
, tree_root
,
6457 root
->root_key
.objectid
);
6461 if (root
->in_radix
) {
6462 btrfs_free_fs_root(tree_root
->fs_info
, root
);
6464 free_extent_buffer(root
->node
);
6465 free_extent_buffer(root
->commit_root
);
6469 btrfs_end_transaction_throttle(trans
, tree_root
);
6471 btrfs_free_path(path
);
6474 btrfs_std_error(root
->fs_info
, err
);
6479 * drop subtree rooted at tree block 'node'.
6481 * NOTE: this function will unlock and release tree block 'node'
6483 int btrfs_drop_subtree(struct btrfs_trans_handle
*trans
,
6484 struct btrfs_root
*root
,
6485 struct extent_buffer
*node
,
6486 struct extent_buffer
*parent
)
6488 struct btrfs_path
*path
;
6489 struct walk_control
*wc
;
6495 BUG_ON(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
);
6497 path
= btrfs_alloc_path();
6501 wc
= kzalloc(sizeof(*wc
), GFP_NOFS
);
6503 btrfs_free_path(path
);
6507 btrfs_assert_tree_locked(parent
);
6508 parent_level
= btrfs_header_level(parent
);
6509 extent_buffer_get(parent
);
6510 path
->nodes
[parent_level
] = parent
;
6511 path
->slots
[parent_level
] = btrfs_header_nritems(parent
);
6513 btrfs_assert_tree_locked(node
);
6514 level
= btrfs_header_level(node
);
6515 path
->nodes
[level
] = node
;
6516 path
->slots
[level
] = 0;
6517 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
6519 wc
->refs
[parent_level
] = 1;
6520 wc
->flags
[parent_level
] = BTRFS_BLOCK_FLAG_FULL_BACKREF
;
6522 wc
->shared_level
= -1;
6523 wc
->stage
= DROP_REFERENCE
;
6526 wc
->reada_count
= BTRFS_NODEPTRS_PER_BLOCK(root
);
6529 wret
= walk_down_tree(trans
, root
, path
, wc
);
6535 wret
= walk_up_tree(trans
, root
, path
, wc
, parent_level
);
6543 btrfs_free_path(path
);
6547 static u64
update_block_group_flags(struct btrfs_root
*root
, u64 flags
)
6550 u64 stripped
= BTRFS_BLOCK_GROUP_RAID0
|
6551 BTRFS_BLOCK_GROUP_RAID1
| BTRFS_BLOCK_GROUP_RAID10
;
6554 * we add in the count of missing devices because we want
6555 * to make sure that any RAID levels on a degraded FS
6556 * continue to be honored.
6558 num_devices
= root
->fs_info
->fs_devices
->rw_devices
+
6559 root
->fs_info
->fs_devices
->missing_devices
;
6561 if (num_devices
== 1) {
6562 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6563 stripped
= flags
& ~stripped
;
6565 /* turn raid0 into single device chunks */
6566 if (flags
& BTRFS_BLOCK_GROUP_RAID0
)
6569 /* turn mirroring into duplication */
6570 if (flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6571 BTRFS_BLOCK_GROUP_RAID10
))
6572 return stripped
| BTRFS_BLOCK_GROUP_DUP
;
6575 /* they already had raid on here, just return */
6576 if (flags
& stripped
)
6579 stripped
|= BTRFS_BLOCK_GROUP_DUP
;
6580 stripped
= flags
& ~stripped
;
6582 /* switch duplicated blocks with raid1 */
6583 if (flags
& BTRFS_BLOCK_GROUP_DUP
)
6584 return stripped
| BTRFS_BLOCK_GROUP_RAID1
;
6586 /* turn single device chunks into raid0 */
6587 return stripped
| BTRFS_BLOCK_GROUP_RAID0
;
6592 static int set_block_group_ro(struct btrfs_block_group_cache
*cache
, int force
)
6594 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6596 u64 min_allocable_bytes
;
6601 * We need some metadata space and system metadata space for
6602 * allocating chunks in some corner cases until we force to set
6603 * it to be readonly.
6606 (BTRFS_BLOCK_GROUP_SYSTEM
| BTRFS_BLOCK_GROUP_METADATA
)) &&
6608 min_allocable_bytes
= 1 * 1024 * 1024;
6610 min_allocable_bytes
= 0;
6612 spin_lock(&sinfo
->lock
);
6613 spin_lock(&cache
->lock
);
6620 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6621 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6623 if (sinfo
->bytes_used
+ sinfo
->bytes_reserved
+ sinfo
->bytes_pinned
+
6624 sinfo
->bytes_may_use
+ sinfo
->bytes_readonly
+ num_bytes
+
6625 min_allocable_bytes
<= sinfo
->total_bytes
) {
6626 sinfo
->bytes_readonly
+= num_bytes
;
6631 spin_unlock(&cache
->lock
);
6632 spin_unlock(&sinfo
->lock
);
6636 int btrfs_set_block_group_ro(struct btrfs_root
*root
,
6637 struct btrfs_block_group_cache
*cache
)
6640 struct btrfs_trans_handle
*trans
;
6646 trans
= btrfs_join_transaction(root
);
6647 BUG_ON(IS_ERR(trans
));
6649 alloc_flags
= update_block_group_flags(root
, cache
->flags
);
6650 if (alloc_flags
!= cache
->flags
)
6651 do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6654 ret
= set_block_group_ro(cache
, 0);
6657 alloc_flags
= get_alloc_profile(root
, cache
->space_info
->flags
);
6658 ret
= do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6662 ret
= set_block_group_ro(cache
, 0);
6664 btrfs_end_transaction(trans
, root
);
6668 int btrfs_force_chunk_alloc(struct btrfs_trans_handle
*trans
,
6669 struct btrfs_root
*root
, u64 type
)
6671 u64 alloc_flags
= get_alloc_profile(root
, type
);
6672 return do_chunk_alloc(trans
, root
, 2 * 1024 * 1024, alloc_flags
,
6677 * helper to account the unused space of all the readonly block group in the
6678 * list. takes mirrors into account.
6680 static u64
__btrfs_get_ro_block_group_free_space(struct list_head
*groups_list
)
6682 struct btrfs_block_group_cache
*block_group
;
6686 list_for_each_entry(block_group
, groups_list
, list
) {
6687 spin_lock(&block_group
->lock
);
6689 if (!block_group
->ro
) {
6690 spin_unlock(&block_group
->lock
);
6694 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_RAID1
|
6695 BTRFS_BLOCK_GROUP_RAID10
|
6696 BTRFS_BLOCK_GROUP_DUP
))
6701 free_bytes
+= (block_group
->key
.offset
-
6702 btrfs_block_group_used(&block_group
->item
)) *
6705 spin_unlock(&block_group
->lock
);
6712 * helper to account the unused space of all the readonly block group in the
6713 * space_info. takes mirrors into account.
6715 u64
btrfs_account_ro_block_groups_free_space(struct btrfs_space_info
*sinfo
)
6720 spin_lock(&sinfo
->lock
);
6722 for(i
= 0; i
< BTRFS_NR_RAID_TYPES
; i
++)
6723 if (!list_empty(&sinfo
->block_groups
[i
]))
6724 free_bytes
+= __btrfs_get_ro_block_group_free_space(
6725 &sinfo
->block_groups
[i
]);
6727 spin_unlock(&sinfo
->lock
);
6732 int btrfs_set_block_group_rw(struct btrfs_root
*root
,
6733 struct btrfs_block_group_cache
*cache
)
6735 struct btrfs_space_info
*sinfo
= cache
->space_info
;
6740 spin_lock(&sinfo
->lock
);
6741 spin_lock(&cache
->lock
);
6742 num_bytes
= cache
->key
.offset
- cache
->reserved
- cache
->pinned
-
6743 cache
->bytes_super
- btrfs_block_group_used(&cache
->item
);
6744 sinfo
->bytes_readonly
-= num_bytes
;
6746 spin_unlock(&cache
->lock
);
6747 spin_unlock(&sinfo
->lock
);
6752 * checks to see if its even possible to relocate this block group.
6754 * @return - -1 if it's not a good idea to relocate this block group, 0 if its
6755 * ok to go ahead and try.
6757 int btrfs_can_relocate(struct btrfs_root
*root
, u64 bytenr
)
6759 struct btrfs_block_group_cache
*block_group
;
6760 struct btrfs_space_info
*space_info
;
6761 struct btrfs_fs_devices
*fs_devices
= root
->fs_info
->fs_devices
;
6762 struct btrfs_device
*device
;
6770 block_group
= btrfs_lookup_block_group(root
->fs_info
, bytenr
);
6772 /* odd, couldn't find the block group, leave it alone */
6776 min_free
= btrfs_block_group_used(&block_group
->item
);
6778 /* no bytes used, we're good */
6782 space_info
= block_group
->space_info
;
6783 spin_lock(&space_info
->lock
);
6785 full
= space_info
->full
;
6788 * if this is the last block group we have in this space, we can't
6789 * relocate it unless we're able to allocate a new chunk below.
6791 * Otherwise, we need to make sure we have room in the space to handle
6792 * all of the extents from this block group. If we can, we're good
6794 if ((space_info
->total_bytes
!= block_group
->key
.offset
) &&
6795 (space_info
->bytes_used
+ space_info
->bytes_reserved
+
6796 space_info
->bytes_pinned
+ space_info
->bytes_readonly
+
6797 min_free
< space_info
->total_bytes
)) {
6798 spin_unlock(&space_info
->lock
);
6801 spin_unlock(&space_info
->lock
);
6804 * ok we don't have enough space, but maybe we have free space on our
6805 * devices to allocate new chunks for relocation, so loop through our
6806 * alloc devices and guess if we have enough space. However, if we
6807 * were marked as full, then we know there aren't enough chunks, and we
6822 index
= get_block_group_index(block_group
);
6827 } else if (index
== 1) {
6829 } else if (index
== 2) {
6832 } else if (index
== 3) {
6833 dev_min
= fs_devices
->rw_devices
;
6834 do_div(min_free
, dev_min
);
6837 mutex_lock(&root
->fs_info
->chunk_mutex
);
6838 list_for_each_entry(device
, &fs_devices
->alloc_list
, dev_alloc_list
) {
6842 * check to make sure we can actually find a chunk with enough
6843 * space to fit our block group in.
6845 if (device
->total_bytes
> device
->bytes_used
+ min_free
) {
6846 ret
= find_free_dev_extent(NULL
, device
, min_free
,
6851 if (dev_nr
>= dev_min
)
6857 mutex_unlock(&root
->fs_info
->chunk_mutex
);
6859 btrfs_put_block_group(block_group
);
6863 static int find_first_block_group(struct btrfs_root
*root
,
6864 struct btrfs_path
*path
, struct btrfs_key
*key
)
6867 struct btrfs_key found_key
;
6868 struct extent_buffer
*leaf
;
6871 ret
= btrfs_search_slot(NULL
, root
, key
, path
, 0, 0);
6876 slot
= path
->slots
[0];
6877 leaf
= path
->nodes
[0];
6878 if (slot
>= btrfs_header_nritems(leaf
)) {
6879 ret
= btrfs_next_leaf(root
, path
);
6886 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
6888 if (found_key
.objectid
>= key
->objectid
&&
6889 found_key
.type
== BTRFS_BLOCK_GROUP_ITEM_KEY
) {
6899 void btrfs_put_block_group_cache(struct btrfs_fs_info
*info
)
6901 struct btrfs_block_group_cache
*block_group
;
6905 struct inode
*inode
;
6907 block_group
= btrfs_lookup_first_block_group(info
, last
);
6908 while (block_group
) {
6909 spin_lock(&block_group
->lock
);
6910 if (block_group
->iref
)
6912 spin_unlock(&block_group
->lock
);
6913 block_group
= next_block_group(info
->tree_root
,
6923 inode
= block_group
->inode
;
6924 block_group
->iref
= 0;
6925 block_group
->inode
= NULL
;
6926 spin_unlock(&block_group
->lock
);
6928 last
= block_group
->key
.objectid
+ block_group
->key
.offset
;
6929 btrfs_put_block_group(block_group
);
6933 int btrfs_free_block_groups(struct btrfs_fs_info
*info
)
6935 struct btrfs_block_group_cache
*block_group
;
6936 struct btrfs_space_info
*space_info
;
6937 struct btrfs_caching_control
*caching_ctl
;
6940 down_write(&info
->extent_commit_sem
);
6941 while (!list_empty(&info
->caching_block_groups
)) {
6942 caching_ctl
= list_entry(info
->caching_block_groups
.next
,
6943 struct btrfs_caching_control
, list
);
6944 list_del(&caching_ctl
->list
);
6945 put_caching_control(caching_ctl
);
6947 up_write(&info
->extent_commit_sem
);
6949 spin_lock(&info
->block_group_cache_lock
);
6950 while ((n
= rb_last(&info
->block_group_cache_tree
)) != NULL
) {
6951 block_group
= rb_entry(n
, struct btrfs_block_group_cache
,
6953 rb_erase(&block_group
->cache_node
,
6954 &info
->block_group_cache_tree
);
6955 spin_unlock(&info
->block_group_cache_lock
);
6957 down_write(&block_group
->space_info
->groups_sem
);
6958 list_del(&block_group
->list
);
6959 up_write(&block_group
->space_info
->groups_sem
);
6961 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
6962 wait_block_group_cache_done(block_group
);
6965 * We haven't cached this block group, which means we could
6966 * possibly have excluded extents on this block group.
6968 if (block_group
->cached
== BTRFS_CACHE_NO
)
6969 free_excluded_extents(info
->extent_root
, block_group
);
6971 btrfs_remove_free_space_cache(block_group
);
6972 btrfs_put_block_group(block_group
);
6974 spin_lock(&info
->block_group_cache_lock
);
6976 spin_unlock(&info
->block_group_cache_lock
);
6978 /* now that all the block groups are freed, go through and
6979 * free all the space_info structs. This is only called during
6980 * the final stages of unmount, and so we know nobody is
6981 * using them. We call synchronize_rcu() once before we start,
6982 * just to be on the safe side.
6986 release_global_block_rsv(info
);
6988 while(!list_empty(&info
->space_info
)) {
6989 space_info
= list_entry(info
->space_info
.next
,
6990 struct btrfs_space_info
,
6992 if (space_info
->bytes_pinned
> 0 ||
6993 space_info
->bytes_reserved
> 0 ||
6994 space_info
->bytes_may_use
> 0) {
6996 dump_space_info(space_info
, 0, 0);
6998 list_del(&space_info
->list
);
7004 static void __link_block_group(struct btrfs_space_info
*space_info
,
7005 struct btrfs_block_group_cache
*cache
)
7007 int index
= get_block_group_index(cache
);
7009 down_write(&space_info
->groups_sem
);
7010 list_add_tail(&cache
->list
, &space_info
->block_groups
[index
]);
7011 up_write(&space_info
->groups_sem
);
7014 int btrfs_read_block_groups(struct btrfs_root
*root
)
7016 struct btrfs_path
*path
;
7018 struct btrfs_block_group_cache
*cache
;
7019 struct btrfs_fs_info
*info
= root
->fs_info
;
7020 struct btrfs_space_info
*space_info
;
7021 struct btrfs_key key
;
7022 struct btrfs_key found_key
;
7023 struct extent_buffer
*leaf
;
7027 root
= info
->extent_root
;
7030 btrfs_set_key_type(&key
, BTRFS_BLOCK_GROUP_ITEM_KEY
);
7031 path
= btrfs_alloc_path();
7036 cache_gen
= btrfs_super_cache_generation(&root
->fs_info
->super_copy
);
7037 if (btrfs_test_opt(root
, SPACE_CACHE
) &&
7038 btrfs_super_generation(&root
->fs_info
->super_copy
) != cache_gen
)
7040 if (btrfs_test_opt(root
, CLEAR_CACHE
))
7044 ret
= find_first_block_group(root
, path
, &key
);
7049 leaf
= path
->nodes
[0];
7050 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
7051 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7056 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7058 if (!cache
->free_space_ctl
) {
7064 atomic_set(&cache
->count
, 1);
7065 spin_lock_init(&cache
->lock
);
7066 cache
->fs_info
= info
;
7067 INIT_LIST_HEAD(&cache
->list
);
7068 INIT_LIST_HEAD(&cache
->cluster_list
);
7071 cache
->disk_cache_state
= BTRFS_DC_CLEAR
;
7073 read_extent_buffer(leaf
, &cache
->item
,
7074 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
7075 sizeof(cache
->item
));
7076 memcpy(&cache
->key
, &found_key
, sizeof(found_key
));
7078 key
.objectid
= found_key
.objectid
+ found_key
.offset
;
7079 btrfs_release_path(path
);
7080 cache
->flags
= btrfs_block_group_flags(&cache
->item
);
7081 cache
->sectorsize
= root
->sectorsize
;
7083 btrfs_init_free_space_ctl(cache
);
7086 * We need to exclude the super stripes now so that the space
7087 * info has super bytes accounted for, otherwise we'll think
7088 * we have more space than we actually do.
7090 exclude_super_stripes(root
, cache
);
7093 * check for two cases, either we are full, and therefore
7094 * don't need to bother with the caching work since we won't
7095 * find any space, or we are empty, and we can just add all
7096 * the space in and be done with it. This saves us _alot_ of
7097 * time, particularly in the full case.
7099 if (found_key
.offset
== btrfs_block_group_used(&cache
->item
)) {
7100 cache
->last_byte_to_unpin
= (u64
)-1;
7101 cache
->cached
= BTRFS_CACHE_FINISHED
;
7102 free_excluded_extents(root
, cache
);
7103 } else if (btrfs_block_group_used(&cache
->item
) == 0) {
7104 cache
->last_byte_to_unpin
= (u64
)-1;
7105 cache
->cached
= BTRFS_CACHE_FINISHED
;
7106 add_new_free_space(cache
, root
->fs_info
,
7108 found_key
.objectid
+
7110 free_excluded_extents(root
, cache
);
7113 ret
= update_space_info(info
, cache
->flags
, found_key
.offset
,
7114 btrfs_block_group_used(&cache
->item
),
7117 cache
->space_info
= space_info
;
7118 spin_lock(&cache
->space_info
->lock
);
7119 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7120 spin_unlock(&cache
->space_info
->lock
);
7122 __link_block_group(space_info
, cache
);
7124 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7127 set_avail_alloc_bits(root
->fs_info
, cache
->flags
);
7128 if (btrfs_chunk_readonly(root
, cache
->key
.objectid
))
7129 set_block_group_ro(cache
, 1);
7132 list_for_each_entry_rcu(space_info
, &root
->fs_info
->space_info
, list
) {
7133 if (!(get_alloc_profile(root
, space_info
->flags
) &
7134 (BTRFS_BLOCK_GROUP_RAID10
|
7135 BTRFS_BLOCK_GROUP_RAID1
|
7136 BTRFS_BLOCK_GROUP_DUP
)))
7139 * avoid allocating from un-mirrored block group if there are
7140 * mirrored block groups.
7142 list_for_each_entry(cache
, &space_info
->block_groups
[3], list
)
7143 set_block_group_ro(cache
, 1);
7144 list_for_each_entry(cache
, &space_info
->block_groups
[4], list
)
7145 set_block_group_ro(cache
, 1);
7148 init_global_block_rsv(info
);
7151 btrfs_free_path(path
);
7155 int btrfs_make_block_group(struct btrfs_trans_handle
*trans
,
7156 struct btrfs_root
*root
, u64 bytes_used
,
7157 u64 type
, u64 chunk_objectid
, u64 chunk_offset
,
7161 struct btrfs_root
*extent_root
;
7162 struct btrfs_block_group_cache
*cache
;
7164 extent_root
= root
->fs_info
->extent_root
;
7166 root
->fs_info
->last_trans_log_full_commit
= trans
->transid
;
7168 cache
= kzalloc(sizeof(*cache
), GFP_NOFS
);
7171 cache
->free_space_ctl
= kzalloc(sizeof(*cache
->free_space_ctl
),
7173 if (!cache
->free_space_ctl
) {
7178 cache
->key
.objectid
= chunk_offset
;
7179 cache
->key
.offset
= size
;
7180 cache
->key
.type
= BTRFS_BLOCK_GROUP_ITEM_KEY
;
7181 cache
->sectorsize
= root
->sectorsize
;
7182 cache
->fs_info
= root
->fs_info
;
7184 atomic_set(&cache
->count
, 1);
7185 spin_lock_init(&cache
->lock
);
7186 INIT_LIST_HEAD(&cache
->list
);
7187 INIT_LIST_HEAD(&cache
->cluster_list
);
7189 btrfs_init_free_space_ctl(cache
);
7191 btrfs_set_block_group_used(&cache
->item
, bytes_used
);
7192 btrfs_set_block_group_chunk_objectid(&cache
->item
, chunk_objectid
);
7193 cache
->flags
= type
;
7194 btrfs_set_block_group_flags(&cache
->item
, type
);
7196 cache
->last_byte_to_unpin
= (u64
)-1;
7197 cache
->cached
= BTRFS_CACHE_FINISHED
;
7198 exclude_super_stripes(root
, cache
);
7200 add_new_free_space(cache
, root
->fs_info
, chunk_offset
,
7201 chunk_offset
+ size
);
7203 free_excluded_extents(root
, cache
);
7205 ret
= update_space_info(root
->fs_info
, cache
->flags
, size
, bytes_used
,
7206 &cache
->space_info
);
7209 spin_lock(&cache
->space_info
->lock
);
7210 cache
->space_info
->bytes_readonly
+= cache
->bytes_super
;
7211 spin_unlock(&cache
->space_info
->lock
);
7213 __link_block_group(cache
->space_info
, cache
);
7215 ret
= btrfs_add_block_group_cache(root
->fs_info
, cache
);
7218 ret
= btrfs_insert_item(trans
, extent_root
, &cache
->key
, &cache
->item
,
7219 sizeof(cache
->item
));
7222 set_avail_alloc_bits(extent_root
->fs_info
, type
);
7227 int btrfs_remove_block_group(struct btrfs_trans_handle
*trans
,
7228 struct btrfs_root
*root
, u64 group_start
)
7230 struct btrfs_path
*path
;
7231 struct btrfs_block_group_cache
*block_group
;
7232 struct btrfs_free_cluster
*cluster
;
7233 struct btrfs_root
*tree_root
= root
->fs_info
->tree_root
;
7234 struct btrfs_key key
;
7235 struct inode
*inode
;
7239 root
= root
->fs_info
->extent_root
;
7241 block_group
= btrfs_lookup_block_group(root
->fs_info
, group_start
);
7242 BUG_ON(!block_group
);
7243 BUG_ON(!block_group
->ro
);
7246 * Free the reserved super bytes from this block group before
7249 free_excluded_extents(root
, block_group
);
7251 memcpy(&key
, &block_group
->key
, sizeof(key
));
7252 if (block_group
->flags
& (BTRFS_BLOCK_GROUP_DUP
|
7253 BTRFS_BLOCK_GROUP_RAID1
|
7254 BTRFS_BLOCK_GROUP_RAID10
))
7259 /* make sure this block group isn't part of an allocation cluster */
7260 cluster
= &root
->fs_info
->data_alloc_cluster
;
7261 spin_lock(&cluster
->refill_lock
);
7262 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7263 spin_unlock(&cluster
->refill_lock
);
7266 * make sure this block group isn't part of a metadata
7267 * allocation cluster
7269 cluster
= &root
->fs_info
->meta_alloc_cluster
;
7270 spin_lock(&cluster
->refill_lock
);
7271 btrfs_return_cluster_to_free_space(block_group
, cluster
);
7272 spin_unlock(&cluster
->refill_lock
);
7274 path
= btrfs_alloc_path();
7280 inode
= lookup_free_space_inode(root
, block_group
, path
);
7281 if (!IS_ERR(inode
)) {
7282 ret
= btrfs_orphan_add(trans
, inode
);
7285 /* One for the block groups ref */
7286 spin_lock(&block_group
->lock
);
7287 if (block_group
->iref
) {
7288 block_group
->iref
= 0;
7289 block_group
->inode
= NULL
;
7290 spin_unlock(&block_group
->lock
);
7293 spin_unlock(&block_group
->lock
);
7295 /* One for our lookup ref */
7296 btrfs_add_delayed_iput(inode
);
7299 key
.objectid
= BTRFS_FREE_SPACE_OBJECTID
;
7300 key
.offset
= block_group
->key
.objectid
;
7303 ret
= btrfs_search_slot(trans
, tree_root
, &key
, path
, -1, 1);
7307 btrfs_release_path(path
);
7309 ret
= btrfs_del_item(trans
, tree_root
, path
);
7312 btrfs_release_path(path
);
7315 spin_lock(&root
->fs_info
->block_group_cache_lock
);
7316 rb_erase(&block_group
->cache_node
,
7317 &root
->fs_info
->block_group_cache_tree
);
7318 spin_unlock(&root
->fs_info
->block_group_cache_lock
);
7320 down_write(&block_group
->space_info
->groups_sem
);
7322 * we must use list_del_init so people can check to see if they
7323 * are still on the list after taking the semaphore
7325 list_del_init(&block_group
->list
);
7326 up_write(&block_group
->space_info
->groups_sem
);
7328 if (block_group
->cached
== BTRFS_CACHE_STARTED
)
7329 wait_block_group_cache_done(block_group
);
7331 btrfs_remove_free_space_cache(block_group
);
7333 spin_lock(&block_group
->space_info
->lock
);
7334 block_group
->space_info
->total_bytes
-= block_group
->key
.offset
;
7335 block_group
->space_info
->bytes_readonly
-= block_group
->key
.offset
;
7336 block_group
->space_info
->disk_total
-= block_group
->key
.offset
* factor
;
7337 spin_unlock(&block_group
->space_info
->lock
);
7339 memcpy(&key
, &block_group
->key
, sizeof(key
));
7341 btrfs_clear_space_info_full(root
->fs_info
);
7343 btrfs_put_block_group(block_group
);
7344 btrfs_put_block_group(block_group
);
7346 ret
= btrfs_search_slot(trans
, root
, &key
, path
, -1, 1);
7352 ret
= btrfs_del_item(trans
, root
, path
);
7354 btrfs_free_path(path
);
7358 int btrfs_init_space_info(struct btrfs_fs_info
*fs_info
)
7360 struct btrfs_space_info
*space_info
;
7361 struct btrfs_super_block
*disk_super
;
7367 disk_super
= &fs_info
->super_copy
;
7368 if (!btrfs_super_root(disk_super
))
7371 features
= btrfs_super_incompat_flags(disk_super
);
7372 if (features
& BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS
)
7375 flags
= BTRFS_BLOCK_GROUP_SYSTEM
;
7376 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7381 flags
= BTRFS_BLOCK_GROUP_METADATA
| BTRFS_BLOCK_GROUP_DATA
;
7382 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7384 flags
= BTRFS_BLOCK_GROUP_METADATA
;
7385 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7389 flags
= BTRFS_BLOCK_GROUP_DATA
;
7390 ret
= update_space_info(fs_info
, flags
, 0, 0, &space_info
);
7396 int btrfs_error_unpin_extent_range(struct btrfs_root
*root
, u64 start
, u64 end
)
7398 return unpin_extent_range(root
, start
, end
);
7401 int btrfs_error_discard_extent(struct btrfs_root
*root
, u64 bytenr
,
7402 u64 num_bytes
, u64
*actual_bytes
)
7404 return btrfs_discard_extent(root
, bytenr
, num_bytes
, actual_bytes
);
7407 int btrfs_trim_fs(struct btrfs_root
*root
, struct fstrim_range
*range
)
7409 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
7410 struct btrfs_block_group_cache
*cache
= NULL
;
7417 cache
= btrfs_lookup_block_group(fs_info
, range
->start
);
7420 if (cache
->key
.objectid
>= (range
->start
+ range
->len
)) {
7421 btrfs_put_block_group(cache
);
7425 start
= max(range
->start
, cache
->key
.objectid
);
7426 end
= min(range
->start
+ range
->len
,
7427 cache
->key
.objectid
+ cache
->key
.offset
);
7429 if (end
- start
>= range
->minlen
) {
7430 if (!block_group_cache_done(cache
)) {
7431 ret
= cache_block_group(cache
, NULL
, root
, 0);
7433 wait_block_group_cache_done(cache
);
7435 ret
= btrfs_trim_block_group(cache
,
7441 trimmed
+= group_trimmed
;
7443 btrfs_put_block_group(cache
);
7448 cache
= next_block_group(fs_info
->tree_root
, cache
);
7451 range
->len
= trimmed
;